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>
44 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap
);
45 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap
);
46 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete
);
47 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split
);
48 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug
);
50 DEFINE_IDA(blk_queue_ida
);
53 * For the allocated request tables
55 struct kmem_cache
*request_cachep
;
58 * For queue allocation
60 struct kmem_cache
*blk_requestq_cachep
;
63 * Controlling structure to kblockd
65 static struct workqueue_struct
*kblockd_workqueue
;
67 static void blk_clear_congested(struct request_list
*rl
, int sync
)
69 #ifdef CONFIG_CGROUP_WRITEBACK
70 clear_wb_congested(rl
->blkg
->wb_congested
, sync
);
73 * If !CGROUP_WRITEBACK, all blkg's map to bdi->wb and we shouldn't
74 * flip its congestion state for events on other blkcgs.
76 if (rl
== &rl
->q
->root_rl
)
77 clear_wb_congested(rl
->q
->backing_dev_info
.wb
.congested
, sync
);
81 static void blk_set_congested(struct request_list
*rl
, int sync
)
83 #ifdef CONFIG_CGROUP_WRITEBACK
84 set_wb_congested(rl
->blkg
->wb_congested
, sync
);
86 /* see blk_clear_congested() */
87 if (rl
== &rl
->q
->root_rl
)
88 set_wb_congested(rl
->q
->backing_dev_info
.wb
.congested
, sync
);
92 void blk_queue_congestion_threshold(struct request_queue
*q
)
96 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) + 1;
97 if (nr
> q
->nr_requests
)
99 q
->nr_congestion_on
= nr
;
101 nr
= q
->nr_requests
- (q
->nr_requests
/ 8) - (q
->nr_requests
/ 16) - 1;
104 q
->nr_congestion_off
= nr
;
108 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
111 * Locates the passed device's request queue and returns the address of its
112 * backing_dev_info. This function can only be called if @bdev is opened
113 * and the return value is never NULL.
115 struct backing_dev_info
*blk_get_backing_dev_info(struct block_device
*bdev
)
117 struct request_queue
*q
= bdev_get_queue(bdev
);
119 return &q
->backing_dev_info
;
121 EXPORT_SYMBOL(blk_get_backing_dev_info
);
123 void blk_rq_init(struct request_queue
*q
, struct request
*rq
)
125 memset(rq
, 0, sizeof(*rq
));
127 INIT_LIST_HEAD(&rq
->queuelist
);
128 INIT_LIST_HEAD(&rq
->timeout_list
);
131 rq
->__sector
= (sector_t
) -1;
132 INIT_HLIST_NODE(&rq
->hash
);
133 RB_CLEAR_NODE(&rq
->rb_node
);
135 rq
->cmd_len
= BLK_MAX_CDB
;
137 rq
->start_time
= jiffies
;
138 set_start_time_ns(rq
);
141 EXPORT_SYMBOL(blk_rq_init
);
143 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
144 unsigned int nbytes
, int error
)
147 bio
->bi_error
= error
;
149 if (unlikely(rq
->rq_flags
& RQF_QUIET
))
150 bio_set_flag(bio
, BIO_QUIET
);
152 bio_advance(bio
, nbytes
);
154 /* don't actually finish bio if it's part of flush sequence */
155 if (bio
->bi_iter
.bi_size
== 0 && !(rq
->rq_flags
& RQF_FLUSH_SEQ
))
159 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
163 printk(KERN_INFO
"%s: dev %s: type=%x, flags=%llx\n", msg
,
164 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?", rq
->cmd_type
,
165 (unsigned long long) rq
->cmd_flags
);
167 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
168 (unsigned long long)blk_rq_pos(rq
),
169 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
170 printk(KERN_INFO
" bio %p, biotail %p, len %u\n",
171 rq
->bio
, rq
->biotail
, blk_rq_bytes(rq
));
173 if (rq
->cmd_type
== REQ_TYPE_BLOCK_PC
) {
174 printk(KERN_INFO
" cdb: ");
175 for (bit
= 0; bit
< BLK_MAX_CDB
; bit
++)
176 printk("%02x ", rq
->cmd
[bit
]);
180 EXPORT_SYMBOL(blk_dump_rq_flags
);
182 static void blk_delay_work(struct work_struct
*work
)
184 struct request_queue
*q
;
186 q
= container_of(work
, struct request_queue
, delay_work
.work
);
187 spin_lock_irq(q
->queue_lock
);
189 spin_unlock_irq(q
->queue_lock
);
193 * blk_delay_queue - restart queueing after defined interval
194 * @q: The &struct request_queue in question
195 * @msecs: Delay in msecs
198 * Sometimes queueing needs to be postponed for a little while, to allow
199 * resources to come back. This function will make sure that queueing is
200 * restarted around the specified time. Queue lock must be held.
202 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
204 if (likely(!blk_queue_dead(q
)))
205 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
206 msecs_to_jiffies(msecs
));
208 EXPORT_SYMBOL(blk_delay_queue
);
211 * blk_start_queue_async - asynchronously restart a previously stopped queue
212 * @q: The &struct request_queue in question
215 * blk_start_queue_async() will clear the stop flag on the queue, and
216 * ensure that the request_fn for the queue is run from an async
219 void blk_start_queue_async(struct request_queue
*q
)
221 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
222 blk_run_queue_async(q
);
224 EXPORT_SYMBOL(blk_start_queue_async
);
227 * blk_start_queue - restart a previously stopped queue
228 * @q: The &struct request_queue in question
231 * blk_start_queue() will clear the stop flag on the queue, and call
232 * the request_fn for the queue if it was in a stopped state when
233 * entered. Also see blk_stop_queue(). Queue lock must be held.
235 void blk_start_queue(struct request_queue
*q
)
237 WARN_ON(!irqs_disabled());
239 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
242 EXPORT_SYMBOL(blk_start_queue
);
245 * blk_stop_queue - stop a queue
246 * @q: The &struct request_queue in question
249 * The Linux block layer assumes that a block driver will consume all
250 * entries on the request queue when the request_fn strategy is called.
251 * Often this will not happen, because of hardware limitations (queue
252 * depth settings). If a device driver gets a 'queue full' response,
253 * or if it simply chooses not to queue more I/O at one point, it can
254 * call this function to prevent the request_fn from being called until
255 * the driver has signalled it's ready to go again. This happens by calling
256 * blk_start_queue() to restart queue operations. Queue lock must be held.
258 void blk_stop_queue(struct request_queue
*q
)
260 cancel_delayed_work(&q
->delay_work
);
261 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
263 EXPORT_SYMBOL(blk_stop_queue
);
266 * blk_sync_queue - cancel any pending callbacks on a queue
270 * The block layer may perform asynchronous callback activity
271 * on a queue, such as calling the unplug function after a timeout.
272 * A block device may call blk_sync_queue to ensure that any
273 * such activity is cancelled, thus allowing it to release resources
274 * that the callbacks might use. The caller must already have made sure
275 * that its ->make_request_fn will not re-add plugging prior to calling
278 * This function does not cancel any asynchronous activity arising
279 * out of elevator or throttling code. That would require elevator_exit()
280 * and blkcg_exit_queue() to be called with queue lock initialized.
283 void blk_sync_queue(struct request_queue
*q
)
285 del_timer_sync(&q
->timeout
);
288 struct blk_mq_hw_ctx
*hctx
;
291 queue_for_each_hw_ctx(q
, hctx
, i
) {
292 cancel_work_sync(&hctx
->run_work
);
293 cancel_delayed_work_sync(&hctx
->delay_work
);
296 cancel_delayed_work_sync(&q
->delay_work
);
299 EXPORT_SYMBOL(blk_sync_queue
);
302 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
303 * @q: The queue to run
306 * Invoke request handling on a queue if there are any pending requests.
307 * May be used to restart request handling after a request has completed.
308 * This variant runs the queue whether or not the queue has been
309 * stopped. Must be called with the queue lock held and interrupts
310 * disabled. See also @blk_run_queue.
312 inline void __blk_run_queue_uncond(struct request_queue
*q
)
314 if (unlikely(blk_queue_dead(q
)))
318 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
319 * the queue lock internally. As a result multiple threads may be
320 * running such a request function concurrently. Keep track of the
321 * number of active request_fn invocations such that blk_drain_queue()
322 * can wait until all these request_fn calls have finished.
324 q
->request_fn_active
++;
326 q
->request_fn_active
--;
328 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond
);
331 * __blk_run_queue - run a single device queue
332 * @q: The queue to run
335 * See @blk_run_queue. This variant must be called with the queue lock
336 * held and interrupts disabled.
338 void __blk_run_queue(struct request_queue
*q
)
340 if (unlikely(blk_queue_stopped(q
)))
343 __blk_run_queue_uncond(q
);
345 EXPORT_SYMBOL(__blk_run_queue
);
348 * blk_run_queue_async - run a single device queue in workqueue context
349 * @q: The queue to run
352 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
353 * of us. The caller must hold the queue lock.
355 void blk_run_queue_async(struct request_queue
*q
)
357 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
358 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
360 EXPORT_SYMBOL(blk_run_queue_async
);
363 * blk_run_queue - run a single device queue
364 * @q: The queue to run
367 * Invoke request handling on this queue, if it has pending work to do.
368 * May be used to restart queueing when a request has completed.
370 void blk_run_queue(struct request_queue
*q
)
374 spin_lock_irqsave(q
->queue_lock
, flags
);
376 spin_unlock_irqrestore(q
->queue_lock
, flags
);
378 EXPORT_SYMBOL(blk_run_queue
);
380 void blk_put_queue(struct request_queue
*q
)
382 kobject_put(&q
->kobj
);
384 EXPORT_SYMBOL(blk_put_queue
);
387 * __blk_drain_queue - drain requests from request_queue
389 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
391 * Drain requests from @q. If @drain_all is set, all requests are drained.
392 * If not, only ELVPRIV requests are drained. The caller is responsible
393 * for ensuring that no new requests which need to be drained are queued.
395 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
396 __releases(q
->queue_lock
)
397 __acquires(q
->queue_lock
)
401 lockdep_assert_held(q
->queue_lock
);
407 * The caller might be trying to drain @q before its
408 * elevator is initialized.
411 elv_drain_elevator(q
);
413 blkcg_drain_queue(q
);
416 * This function might be called on a queue which failed
417 * driver init after queue creation or is not yet fully
418 * active yet. Some drivers (e.g. fd and loop) get unhappy
419 * in such cases. Kick queue iff dispatch queue has
420 * something on it and @q has request_fn set.
422 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
425 drain
|= q
->nr_rqs_elvpriv
;
426 drain
|= q
->request_fn_active
;
429 * Unfortunately, requests are queued at and tracked from
430 * multiple places and there's no single counter which can
431 * be drained. Check all the queues and counters.
434 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
435 drain
|= !list_empty(&q
->queue_head
);
436 for (i
= 0; i
< 2; i
++) {
437 drain
|= q
->nr_rqs
[i
];
438 drain
|= q
->in_flight
[i
];
440 drain
|= !list_empty(&fq
->flush_queue
[i
]);
447 spin_unlock_irq(q
->queue_lock
);
451 spin_lock_irq(q
->queue_lock
);
455 * With queue marked dead, any woken up waiter will fail the
456 * allocation path, so the wakeup chaining is lost and we're
457 * left with hung waiters. We need to wake up those waiters.
460 struct request_list
*rl
;
462 blk_queue_for_each_rl(rl
, q
)
463 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
464 wake_up_all(&rl
->wait
[i
]);
469 * blk_queue_bypass_start - enter queue bypass mode
470 * @q: queue of interest
472 * In bypass mode, only the dispatch FIFO queue of @q is used. This
473 * function makes @q enter bypass mode and drains all requests which were
474 * throttled or issued before. On return, it's guaranteed that no request
475 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
476 * inside queue or RCU read lock.
478 void blk_queue_bypass_start(struct request_queue
*q
)
480 spin_lock_irq(q
->queue_lock
);
482 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
483 spin_unlock_irq(q
->queue_lock
);
486 * Queues start drained. Skip actual draining till init is
487 * complete. This avoids lenghty delays during queue init which
488 * can happen many times during boot.
490 if (blk_queue_init_done(q
)) {
491 spin_lock_irq(q
->queue_lock
);
492 __blk_drain_queue(q
, false);
493 spin_unlock_irq(q
->queue_lock
);
495 /* ensure blk_queue_bypass() is %true inside RCU read lock */
499 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
502 * blk_queue_bypass_end - leave queue bypass mode
503 * @q: queue of interest
505 * Leave bypass mode and restore the normal queueing behavior.
507 void blk_queue_bypass_end(struct request_queue
*q
)
509 spin_lock_irq(q
->queue_lock
);
510 if (!--q
->bypass_depth
)
511 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
512 WARN_ON_ONCE(q
->bypass_depth
< 0);
513 spin_unlock_irq(q
->queue_lock
);
515 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
517 void blk_set_queue_dying(struct request_queue
*q
)
519 spin_lock_irq(q
->queue_lock
);
520 queue_flag_set(QUEUE_FLAG_DYING
, q
);
521 spin_unlock_irq(q
->queue_lock
);
524 blk_mq_wake_waiters(q
);
526 struct request_list
*rl
;
528 blk_queue_for_each_rl(rl
, q
) {
530 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
531 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
536 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
539 * blk_cleanup_queue - shutdown a request queue
540 * @q: request queue to shutdown
542 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
543 * put it. All future requests will be failed immediately with -ENODEV.
545 void blk_cleanup_queue(struct request_queue
*q
)
547 spinlock_t
*lock
= q
->queue_lock
;
549 /* mark @q DYING, no new request or merges will be allowed afterwards */
550 mutex_lock(&q
->sysfs_lock
);
551 blk_set_queue_dying(q
);
555 * A dying queue is permanently in bypass mode till released. Note
556 * that, unlike blk_queue_bypass_start(), we aren't performing
557 * synchronize_rcu() after entering bypass mode to avoid the delay
558 * as some drivers create and destroy a lot of queues while
559 * probing. This is still safe because blk_release_queue() will be
560 * called only after the queue refcnt drops to zero and nothing,
561 * RCU or not, would be traversing the queue by then.
564 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
566 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
567 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
568 queue_flag_set(QUEUE_FLAG_DYING
, q
);
569 spin_unlock_irq(lock
);
570 mutex_unlock(&q
->sysfs_lock
);
573 * Drain all requests queued before DYING marking. Set DEAD flag to
574 * prevent that q->request_fn() gets invoked after draining finished.
579 __blk_drain_queue(q
, true);
580 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
581 spin_unlock_irq(lock
);
583 /* for synchronous bio-based driver finish in-flight integrity i/o */
584 blk_flush_integrity();
586 /* @q won't process any more request, flush async actions */
587 del_timer_sync(&q
->backing_dev_info
.laptop_mode_wb_timer
);
591 blk_mq_free_queue(q
);
592 percpu_ref_exit(&q
->q_usage_counter
);
595 if (q
->queue_lock
!= &q
->__queue_lock
)
596 q
->queue_lock
= &q
->__queue_lock
;
597 spin_unlock_irq(lock
);
599 bdi_unregister(&q
->backing_dev_info
);
601 /* @q is and will stay empty, shutdown and put */
604 EXPORT_SYMBOL(blk_cleanup_queue
);
606 /* Allocate memory local to the request queue */
607 static void *alloc_request_struct(gfp_t gfp_mask
, void *data
)
609 int nid
= (int)(long)data
;
610 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, nid
);
613 static void free_request_struct(void *element
, void *unused
)
615 kmem_cache_free(request_cachep
, element
);
618 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
621 if (unlikely(rl
->rq_pool
))
625 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
626 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
627 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
628 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
630 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
, alloc_request_struct
,
632 (void *)(long)q
->node
, gfp_mask
,
640 void blk_exit_rl(struct request_list
*rl
)
643 mempool_destroy(rl
->rq_pool
);
646 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
648 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
650 EXPORT_SYMBOL(blk_alloc_queue
);
652 int blk_queue_enter(struct request_queue
*q
, bool nowait
)
657 if (percpu_ref_tryget_live(&q
->q_usage_counter
))
663 ret
= wait_event_interruptible(q
->mq_freeze_wq
,
664 !atomic_read(&q
->mq_freeze_depth
) ||
666 if (blk_queue_dying(q
))
673 void blk_queue_exit(struct request_queue
*q
)
675 percpu_ref_put(&q
->q_usage_counter
);
678 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
680 struct request_queue
*q
=
681 container_of(ref
, struct request_queue
, q_usage_counter
);
683 wake_up_all(&q
->mq_freeze_wq
);
686 static void blk_rq_timed_out_timer(unsigned long data
)
688 struct request_queue
*q
= (struct request_queue
*)data
;
690 kblockd_schedule_work(&q
->timeout_work
);
693 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
695 struct request_queue
*q
;
698 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
699 gfp_mask
| __GFP_ZERO
, node_id
);
703 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
707 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0);
711 q
->backing_dev_info
.ra_pages
=
712 (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
713 q
->backing_dev_info
.capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
714 q
->backing_dev_info
.name
= "block";
717 err
= bdi_init(&q
->backing_dev_info
);
721 setup_timer(&q
->backing_dev_info
.laptop_mode_wb_timer
,
722 laptop_mode_timer_fn
, (unsigned long) q
);
723 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
724 INIT_LIST_HEAD(&q
->queue_head
);
725 INIT_LIST_HEAD(&q
->timeout_list
);
726 INIT_LIST_HEAD(&q
->icq_list
);
727 #ifdef CONFIG_BLK_CGROUP
728 INIT_LIST_HEAD(&q
->blkg_list
);
730 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
732 kobject_init(&q
->kobj
, &blk_queue_ktype
);
734 mutex_init(&q
->sysfs_lock
);
735 spin_lock_init(&q
->__queue_lock
);
738 * By default initialize queue_lock to internal lock and driver can
739 * override it later if need be.
741 q
->queue_lock
= &q
->__queue_lock
;
744 * A queue starts its life with bypass turned on to avoid
745 * unnecessary bypass on/off overhead and nasty surprises during
746 * init. The initial bypass will be finished when the queue is
747 * registered by blk_register_queue().
750 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
752 init_waitqueue_head(&q
->mq_freeze_wq
);
755 * Init percpu_ref in atomic mode so that it's faster to shutdown.
756 * See blk_register_queue() for details.
758 if (percpu_ref_init(&q
->q_usage_counter
,
759 blk_queue_usage_counter_release
,
760 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
763 if (blkcg_init_queue(q
))
769 percpu_ref_exit(&q
->q_usage_counter
);
771 bdi_destroy(&q
->backing_dev_info
);
773 bioset_free(q
->bio_split
);
775 ida_simple_remove(&blk_queue_ida
, q
->id
);
777 kmem_cache_free(blk_requestq_cachep
, q
);
780 EXPORT_SYMBOL(blk_alloc_queue_node
);
783 * blk_init_queue - prepare a request queue for use with a block device
784 * @rfn: The function to be called to process requests that have been
785 * placed on the queue.
786 * @lock: Request queue spin lock
789 * If a block device wishes to use the standard request handling procedures,
790 * which sorts requests and coalesces adjacent requests, then it must
791 * call blk_init_queue(). The function @rfn will be called when there
792 * are requests on the queue that need to be processed. If the device
793 * supports plugging, then @rfn may not be called immediately when requests
794 * are available on the queue, but may be called at some time later instead.
795 * Plugged queues are generally unplugged when a buffer belonging to one
796 * of the requests on the queue is needed, or due to memory pressure.
798 * @rfn is not required, or even expected, to remove all requests off the
799 * queue, but only as many as it can handle at a time. If it does leave
800 * requests on the queue, it is responsible for arranging that the requests
801 * get dealt with eventually.
803 * The queue spin lock must be held while manipulating the requests on the
804 * request queue; this lock will be taken also from interrupt context, so irq
805 * disabling is needed for it.
807 * Function returns a pointer to the initialized request queue, or %NULL if
811 * blk_init_queue() must be paired with a blk_cleanup_queue() call
812 * when the block device is deactivated (such as at module unload).
815 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
817 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
819 EXPORT_SYMBOL(blk_init_queue
);
821 struct request_queue
*
822 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
824 struct request_queue
*uninit_q
, *q
;
826 uninit_q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
830 q
= blk_init_allocated_queue(uninit_q
, rfn
, lock
);
832 blk_cleanup_queue(uninit_q
);
836 EXPORT_SYMBOL(blk_init_queue_node
);
838 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
840 struct request_queue
*
841 blk_init_allocated_queue(struct request_queue
*q
, request_fn_proc
*rfn
,
847 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, 0);
851 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
854 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
856 q
->prep_rq_fn
= NULL
;
857 q
->unprep_rq_fn
= NULL
;
858 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
860 /* Override internal queue lock with supplied lock pointer */
862 q
->queue_lock
= lock
;
865 * This also sets hw/phys segments, boundary and size
867 blk_queue_make_request(q
, blk_queue_bio
);
869 q
->sg_reserved_size
= INT_MAX
;
871 /* Protect q->elevator from elevator_change */
872 mutex_lock(&q
->sysfs_lock
);
875 if (elevator_init(q
, NULL
)) {
876 mutex_unlock(&q
->sysfs_lock
);
880 mutex_unlock(&q
->sysfs_lock
);
885 blk_free_flush_queue(q
->fq
);
889 EXPORT_SYMBOL(blk_init_allocated_queue
);
891 bool blk_get_queue(struct request_queue
*q
)
893 if (likely(!blk_queue_dying(q
))) {
900 EXPORT_SYMBOL(blk_get_queue
);
902 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
904 if (rq
->rq_flags
& RQF_ELVPRIV
) {
905 elv_put_request(rl
->q
, rq
);
907 put_io_context(rq
->elv
.icq
->ioc
);
910 mempool_free(rq
, rl
->rq_pool
);
914 * ioc_batching returns true if the ioc is a valid batching request and
915 * should be given priority access to a request.
917 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
923 * Make sure the process is able to allocate at least 1 request
924 * even if the batch times out, otherwise we could theoretically
927 return ioc
->nr_batch_requests
== q
->nr_batching
||
928 (ioc
->nr_batch_requests
> 0
929 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
933 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
934 * will cause the process to be a "batcher" on all queues in the system. This
935 * is the behaviour we want though - once it gets a wakeup it should be given
938 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
940 if (!ioc
|| ioc_batching(q
, ioc
))
943 ioc
->nr_batch_requests
= q
->nr_batching
;
944 ioc
->last_waited
= jiffies
;
947 static void __freed_request(struct request_list
*rl
, int sync
)
949 struct request_queue
*q
= rl
->q
;
951 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
952 blk_clear_congested(rl
, sync
);
954 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
955 if (waitqueue_active(&rl
->wait
[sync
]))
956 wake_up(&rl
->wait
[sync
]);
958 blk_clear_rl_full(rl
, sync
);
963 * A request has just been released. Account for it, update the full and
964 * congestion status, wake up any waiters. Called under q->queue_lock.
966 static void freed_request(struct request_list
*rl
, bool sync
,
967 req_flags_t rq_flags
)
969 struct request_queue
*q
= rl
->q
;
973 if (rq_flags
& RQF_ELVPRIV
)
976 __freed_request(rl
, sync
);
978 if (unlikely(rl
->starved
[sync
^ 1]))
979 __freed_request(rl
, sync
^ 1);
982 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
984 struct request_list
*rl
;
985 int on_thresh
, off_thresh
;
987 spin_lock_irq(q
->queue_lock
);
989 blk_queue_congestion_threshold(q
);
990 on_thresh
= queue_congestion_on_threshold(q
);
991 off_thresh
= queue_congestion_off_threshold(q
);
993 blk_queue_for_each_rl(rl
, q
) {
994 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
995 blk_set_congested(rl
, BLK_RW_SYNC
);
996 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
997 blk_clear_congested(rl
, BLK_RW_SYNC
);
999 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
1000 blk_set_congested(rl
, BLK_RW_ASYNC
);
1001 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
1002 blk_clear_congested(rl
, BLK_RW_ASYNC
);
1004 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
1005 blk_set_rl_full(rl
, BLK_RW_SYNC
);
1007 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1008 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1011 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1012 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1014 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1015 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1019 spin_unlock_irq(q
->queue_lock
);
1024 * Determine if elevator data should be initialized when allocating the
1025 * request associated with @bio.
1027 static bool blk_rq_should_init_elevator(struct bio
*bio
)
1033 * Flush requests do not use the elevator so skip initialization.
1034 * This allows a request to share the flush and elevator data.
1036 if (bio
->bi_opf
& (REQ_PREFLUSH
| REQ_FUA
))
1043 * rq_ioc - determine io_context for request allocation
1044 * @bio: request being allocated is for this bio (can be %NULL)
1046 * Determine io_context to use for request allocation for @bio. May return
1047 * %NULL if %current->io_context doesn't exist.
1049 static struct io_context
*rq_ioc(struct bio
*bio
)
1051 #ifdef CONFIG_BLK_CGROUP
1052 if (bio
&& bio
->bi_ioc
)
1055 return current
->io_context
;
1059 * __get_request - get a free request
1060 * @rl: request list to allocate from
1061 * @op: operation and flags
1062 * @bio: bio to allocate request for (can be %NULL)
1063 * @gfp_mask: allocation mask
1065 * Get a free request from @q. This function may fail under memory
1066 * pressure or if @q is dead.
1068 * Must be called with @q->queue_lock held and,
1069 * Returns ERR_PTR on failure, with @q->queue_lock held.
1070 * Returns request pointer on success, with @q->queue_lock *not held*.
1072 static struct request
*__get_request(struct request_list
*rl
, unsigned int op
,
1073 struct bio
*bio
, gfp_t gfp_mask
)
1075 struct request_queue
*q
= rl
->q
;
1077 struct elevator_type
*et
= q
->elevator
->type
;
1078 struct io_context
*ioc
= rq_ioc(bio
);
1079 struct io_cq
*icq
= NULL
;
1080 const bool is_sync
= op_is_sync(op
);
1082 req_flags_t rq_flags
= RQF_ALLOCED
;
1084 if (unlikely(blk_queue_dying(q
)))
1085 return ERR_PTR(-ENODEV
);
1087 may_queue
= elv_may_queue(q
, op
);
1088 if (may_queue
== ELV_MQUEUE_NO
)
1091 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1092 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1094 * The queue will fill after this allocation, so set
1095 * it as full, and mark this process as "batching".
1096 * This process will be allowed to complete a batch of
1097 * requests, others will be blocked.
1099 if (!blk_rl_full(rl
, is_sync
)) {
1100 ioc_set_batching(q
, ioc
);
1101 blk_set_rl_full(rl
, is_sync
);
1103 if (may_queue
!= ELV_MQUEUE_MUST
1104 && !ioc_batching(q
, ioc
)) {
1106 * The queue is full and the allocating
1107 * process is not a "batcher", and not
1108 * exempted by the IO scheduler
1110 return ERR_PTR(-ENOMEM
);
1114 blk_set_congested(rl
, is_sync
);
1118 * Only allow batching queuers to allocate up to 50% over the defined
1119 * limit of requests, otherwise we could have thousands of requests
1120 * allocated with any setting of ->nr_requests
1122 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1123 return ERR_PTR(-ENOMEM
);
1125 q
->nr_rqs
[is_sync
]++;
1126 rl
->count
[is_sync
]++;
1127 rl
->starved
[is_sync
] = 0;
1130 * Decide whether the new request will be managed by elevator. If
1131 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1132 * prevent the current elevator from being destroyed until the new
1133 * request is freed. This guarantees icq's won't be destroyed and
1134 * makes creating new ones safe.
1136 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1137 * it will be created after releasing queue_lock.
1139 if (blk_rq_should_init_elevator(bio
) && !blk_queue_bypass(q
)) {
1140 rq_flags
|= RQF_ELVPRIV
;
1141 q
->nr_rqs_elvpriv
++;
1142 if (et
->icq_cache
&& ioc
)
1143 icq
= ioc_lookup_icq(ioc
, q
);
1146 if (blk_queue_io_stat(q
))
1147 rq_flags
|= RQF_IO_STAT
;
1148 spin_unlock_irq(q
->queue_lock
);
1150 /* allocate and init request */
1151 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1156 blk_rq_set_rl(rq
, rl
);
1158 rq
->rq_flags
= rq_flags
;
1161 if (rq_flags
& RQF_ELVPRIV
) {
1162 if (unlikely(et
->icq_cache
&& !icq
)) {
1164 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1170 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1173 /* @rq->elv.icq holds io_context until @rq is freed */
1175 get_io_context(icq
->ioc
);
1179 * ioc may be NULL here, and ioc_batching will be false. That's
1180 * OK, if the queue is under the request limit then requests need
1181 * not count toward the nr_batch_requests limit. There will always
1182 * be some limit enforced by BLK_BATCH_TIME.
1184 if (ioc_batching(q
, ioc
))
1185 ioc
->nr_batch_requests
--;
1187 trace_block_getrq(q
, bio
, op
);
1192 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1193 * and may fail indefinitely under memory pressure and thus
1194 * shouldn't stall IO. Treat this request as !elvpriv. This will
1195 * disturb iosched and blkcg but weird is bettern than dead.
1197 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1198 __func__
, dev_name(q
->backing_dev_info
.dev
));
1200 rq
->rq_flags
&= ~RQF_ELVPRIV
;
1203 spin_lock_irq(q
->queue_lock
);
1204 q
->nr_rqs_elvpriv
--;
1205 spin_unlock_irq(q
->queue_lock
);
1210 * Allocation failed presumably due to memory. Undo anything we
1211 * might have messed up.
1213 * Allocating task should really be put onto the front of the wait
1214 * queue, but this is pretty rare.
1216 spin_lock_irq(q
->queue_lock
);
1217 freed_request(rl
, is_sync
, rq_flags
);
1220 * in the very unlikely event that allocation failed and no
1221 * requests for this direction was pending, mark us starved so that
1222 * freeing of a request in the other direction will notice
1223 * us. another possible fix would be to split the rq mempool into
1227 if (unlikely(rl
->count
[is_sync
] == 0))
1228 rl
->starved
[is_sync
] = 1;
1229 return ERR_PTR(-ENOMEM
);
1233 * get_request - get a free request
1234 * @q: request_queue to allocate request from
1235 * @op: operation and flags
1236 * @bio: bio to allocate request for (can be %NULL)
1237 * @gfp_mask: allocation mask
1239 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1240 * this function keeps retrying under memory pressure and fails iff @q is dead.
1242 * Must be called with @q->queue_lock held and,
1243 * Returns ERR_PTR on failure, with @q->queue_lock held.
1244 * Returns request pointer on success, with @q->queue_lock *not held*.
1246 static struct request
*get_request(struct request_queue
*q
, unsigned int op
,
1247 struct bio
*bio
, gfp_t gfp_mask
)
1249 const bool is_sync
= op_is_sync(op
);
1251 struct request_list
*rl
;
1254 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1256 rq
= __get_request(rl
, op
, bio
, gfp_mask
);
1260 if (!gfpflags_allow_blocking(gfp_mask
) || unlikely(blk_queue_dying(q
))) {
1265 /* wait on @rl and retry */
1266 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1267 TASK_UNINTERRUPTIBLE
);
1269 trace_block_sleeprq(q
, bio
, op
);
1271 spin_unlock_irq(q
->queue_lock
);
1275 * After sleeping, we become a "batching" process and will be able
1276 * to allocate at least one request, and up to a big batch of them
1277 * for a small period time. See ioc_batching, ioc_set_batching
1279 ioc_set_batching(q
, current
->io_context
);
1281 spin_lock_irq(q
->queue_lock
);
1282 finish_wait(&rl
->wait
[is_sync
], &wait
);
1287 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1292 BUG_ON(rw
!= READ
&& rw
!= WRITE
);
1294 /* create ioc upfront */
1295 create_io_context(gfp_mask
, q
->node
);
1297 spin_lock_irq(q
->queue_lock
);
1298 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1300 spin_unlock_irq(q
->queue_lock
);
1304 /* q->queue_lock is unlocked at this point */
1306 rq
->__sector
= (sector_t
) -1;
1307 rq
->bio
= rq
->biotail
= NULL
;
1311 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1314 return blk_mq_alloc_request(q
, rw
,
1315 (gfp_mask
& __GFP_DIRECT_RECLAIM
) ?
1316 0 : BLK_MQ_REQ_NOWAIT
);
1318 return blk_old_get_request(q
, rw
, gfp_mask
);
1320 EXPORT_SYMBOL(blk_get_request
);
1323 * blk_rq_set_block_pc - initialize a request to type BLOCK_PC
1324 * @rq: request to be initialized
1327 void blk_rq_set_block_pc(struct request
*rq
)
1329 rq
->cmd_type
= REQ_TYPE_BLOCK_PC
;
1330 memset(rq
->__cmd
, 0, sizeof(rq
->__cmd
));
1332 EXPORT_SYMBOL(blk_rq_set_block_pc
);
1335 * blk_requeue_request - put a request back on queue
1336 * @q: request queue where request should be inserted
1337 * @rq: request to be inserted
1340 * Drivers often keep queueing requests until the hardware cannot accept
1341 * more, when that condition happens we need to put the request back
1342 * on the queue. Must be called with queue lock held.
1344 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1346 blk_delete_timer(rq
);
1347 blk_clear_rq_complete(rq
);
1348 trace_block_rq_requeue(q
, rq
);
1349 wbt_requeue(q
->rq_wb
, &rq
->issue_stat
);
1351 if (rq
->rq_flags
& RQF_QUEUED
)
1352 blk_queue_end_tag(q
, rq
);
1354 BUG_ON(blk_queued_rq(rq
));
1356 elv_requeue_request(q
, rq
);
1358 EXPORT_SYMBOL(blk_requeue_request
);
1360 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1363 blk_account_io_start(rq
, true);
1364 __elv_add_request(q
, rq
, where
);
1367 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1372 if (now
== part
->stamp
)
1375 inflight
= part_in_flight(part
);
1377 __part_stat_add(cpu
, part
, time_in_queue
,
1378 inflight
* (now
- part
->stamp
));
1379 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1385 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1386 * @cpu: cpu number for stats access
1387 * @part: target partition
1389 * The average IO queue length and utilisation statistics are maintained
1390 * by observing the current state of the queue length and the amount of
1391 * time it has been in this state for.
1393 * Normally, that accounting is done on IO completion, but that can result
1394 * in more than a second's worth of IO being accounted for within any one
1395 * second, leading to >100% utilisation. To deal with that, we call this
1396 * function to do a round-off before returning the results when reading
1397 * /proc/diskstats. This accounts immediately for all queue usage up to
1398 * the current jiffies and restarts the counters again.
1400 void part_round_stats(int cpu
, struct hd_struct
*part
)
1402 unsigned long now
= jiffies
;
1405 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1406 part_round_stats_single(cpu
, part
, now
);
1408 EXPORT_SYMBOL_GPL(part_round_stats
);
1411 static void blk_pm_put_request(struct request
*rq
)
1413 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1414 pm_runtime_mark_last_busy(rq
->q
->dev
);
1417 static inline void blk_pm_put_request(struct request
*rq
) {}
1421 * queue lock must be held
1423 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1425 req_flags_t rq_flags
= req
->rq_flags
;
1431 blk_mq_free_request(req
);
1435 blk_pm_put_request(req
);
1437 elv_completed_request(q
, req
);
1439 /* this is a bio leak */
1440 WARN_ON(req
->bio
!= NULL
);
1442 wbt_done(q
->rq_wb
, &req
->issue_stat
);
1445 * Request may not have originated from ll_rw_blk. if not,
1446 * it didn't come out of our reserved rq pools
1448 if (rq_flags
& RQF_ALLOCED
) {
1449 struct request_list
*rl
= blk_rq_rl(req
);
1450 bool sync
= op_is_sync(req
->cmd_flags
);
1452 BUG_ON(!list_empty(&req
->queuelist
));
1453 BUG_ON(ELV_ON_HASH(req
));
1455 blk_free_request(rl
, req
);
1456 freed_request(rl
, sync
, rq_flags
);
1460 EXPORT_SYMBOL_GPL(__blk_put_request
);
1462 void blk_put_request(struct request
*req
)
1464 struct request_queue
*q
= req
->q
;
1467 blk_mq_free_request(req
);
1469 unsigned long flags
;
1471 spin_lock_irqsave(q
->queue_lock
, flags
);
1472 __blk_put_request(q
, req
);
1473 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1476 EXPORT_SYMBOL(blk_put_request
);
1479 * blk_add_request_payload - add a payload to a request
1480 * @rq: request to update
1481 * @page: page backing the payload
1482 * @offset: offset in page
1483 * @len: length of the payload.
1485 * This allows to later add a payload to an already submitted request by
1486 * a block driver. The driver needs to take care of freeing the payload
1489 * Note that this is a quite horrible hack and nothing but handling of
1490 * discard requests should ever use it.
1492 void blk_add_request_payload(struct request
*rq
, struct page
*page
,
1493 int offset
, unsigned int len
)
1495 struct bio
*bio
= rq
->bio
;
1497 bio
->bi_io_vec
->bv_page
= page
;
1498 bio
->bi_io_vec
->bv_offset
= offset
;
1499 bio
->bi_io_vec
->bv_len
= len
;
1501 bio
->bi_iter
.bi_size
= len
;
1503 bio
->bi_phys_segments
= 1;
1505 rq
->__data_len
= rq
->resid_len
= len
;
1506 rq
->nr_phys_segments
= 1;
1508 EXPORT_SYMBOL_GPL(blk_add_request_payload
);
1510 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1513 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1515 if (!ll_back_merge_fn(q
, req
, bio
))
1518 trace_block_bio_backmerge(q
, req
, bio
);
1520 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1521 blk_rq_set_mixed_merge(req
);
1523 req
->biotail
->bi_next
= bio
;
1525 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1526 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1528 blk_account_io_start(req
, false);
1532 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1535 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1537 if (!ll_front_merge_fn(q
, req
, bio
))
1540 trace_block_bio_frontmerge(q
, req
, bio
);
1542 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1543 blk_rq_set_mixed_merge(req
);
1545 bio
->bi_next
= req
->bio
;
1548 req
->__sector
= bio
->bi_iter
.bi_sector
;
1549 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1550 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1552 blk_account_io_start(req
, false);
1557 * blk_attempt_plug_merge - try to merge with %current's plugged list
1558 * @q: request_queue new bio is being queued at
1559 * @bio: new bio being queued
1560 * @request_count: out parameter for number of traversed plugged requests
1561 * @same_queue_rq: pointer to &struct request that gets filled in when
1562 * another request associated with @q is found on the plug list
1563 * (optional, may be %NULL)
1565 * Determine whether @bio being queued on @q can be merged with a request
1566 * on %current's plugged list. Returns %true if merge was successful,
1569 * Plugging coalesces IOs from the same issuer for the same purpose without
1570 * going through @q->queue_lock. As such it's more of an issuing mechanism
1571 * than scheduling, and the request, while may have elvpriv data, is not
1572 * added on the elevator at this point. In addition, we don't have
1573 * reliable access to the elevator outside queue lock. Only check basic
1574 * merging parameters without querying the elevator.
1576 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1578 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1579 unsigned int *request_count
,
1580 struct request
**same_queue_rq
)
1582 struct blk_plug
*plug
;
1585 struct list_head
*plug_list
;
1587 plug
= current
->plug
;
1593 plug_list
= &plug
->mq_list
;
1595 plug_list
= &plug
->list
;
1597 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1603 * Only blk-mq multiple hardware queues case checks the
1604 * rq in the same queue, there should be only one such
1608 *same_queue_rq
= rq
;
1611 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1614 el_ret
= blk_try_merge(rq
, bio
);
1615 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1616 ret
= bio_attempt_back_merge(q
, rq
, bio
);
1619 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1620 ret
= bio_attempt_front_merge(q
, rq
, bio
);
1629 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1631 struct blk_plug
*plug
;
1633 struct list_head
*plug_list
;
1634 unsigned int ret
= 0;
1636 plug
= current
->plug
;
1641 plug_list
= &plug
->mq_list
;
1643 plug_list
= &plug
->list
;
1645 list_for_each_entry(rq
, plug_list
, queuelist
) {
1653 void init_request_from_bio(struct request
*req
, struct bio
*bio
)
1655 req
->cmd_type
= REQ_TYPE_FS
;
1656 if (bio
->bi_opf
& REQ_RAHEAD
)
1657 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1660 req
->__sector
= bio
->bi_iter
.bi_sector
;
1661 req
->ioprio
= bio_prio(bio
);
1662 blk_rq_bio_prep(req
->q
, req
, bio
);
1665 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1667 struct blk_plug
*plug
;
1668 int el_ret
, where
= ELEVATOR_INSERT_SORT
;
1669 struct request
*req
;
1670 unsigned int request_count
= 0;
1671 unsigned int wb_acct
;
1674 * low level driver can indicate that it wants pages above a
1675 * certain limit bounced to low memory (ie for highmem, or even
1676 * ISA dma in theory)
1678 blk_queue_bounce(q
, &bio
);
1680 blk_queue_split(q
, &bio
, q
->bio_split
);
1682 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1683 bio
->bi_error
= -EIO
;
1685 return BLK_QC_T_NONE
;
1688 if (bio
->bi_opf
& (REQ_PREFLUSH
| REQ_FUA
)) {
1689 spin_lock_irq(q
->queue_lock
);
1690 where
= ELEVATOR_INSERT_FLUSH
;
1695 * Check if we can merge with the plugged list before grabbing
1698 if (!blk_queue_nomerges(q
)) {
1699 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1700 return BLK_QC_T_NONE
;
1702 request_count
= blk_plug_queued_count(q
);
1704 spin_lock_irq(q
->queue_lock
);
1706 el_ret
= elv_merge(q
, &req
, bio
);
1707 if (el_ret
== ELEVATOR_BACK_MERGE
) {
1708 if (bio_attempt_back_merge(q
, req
, bio
)) {
1709 elv_bio_merged(q
, req
, bio
);
1710 if (!attempt_back_merge(q
, req
))
1711 elv_merged_request(q
, req
, el_ret
);
1714 } else if (el_ret
== ELEVATOR_FRONT_MERGE
) {
1715 if (bio_attempt_front_merge(q
, req
, bio
)) {
1716 elv_bio_merged(q
, req
, bio
);
1717 if (!attempt_front_merge(q
, req
))
1718 elv_merged_request(q
, req
, el_ret
);
1724 wb_acct
= wbt_wait(q
->rq_wb
, bio
, q
->queue_lock
);
1727 * Grab a free request. This is might sleep but can not fail.
1728 * Returns with the queue unlocked.
1730 req
= get_request(q
, bio
->bi_opf
, bio
, GFP_NOIO
);
1732 __wbt_done(q
->rq_wb
, wb_acct
);
1733 bio
->bi_error
= PTR_ERR(req
);
1738 wbt_track(&req
->issue_stat
, wb_acct
);
1741 * After dropping the lock and possibly sleeping here, our request
1742 * may now be mergeable after it had proven unmergeable (above).
1743 * We don't worry about that case for efficiency. It won't happen
1744 * often, and the elevators are able to handle it.
1746 init_request_from_bio(req
, bio
);
1748 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1749 req
->cpu
= raw_smp_processor_id();
1751 plug
= current
->plug
;
1754 * If this is the first request added after a plug, fire
1757 * @request_count may become stale because of schedule
1758 * out, so check plug list again.
1760 if (!request_count
|| list_empty(&plug
->list
))
1761 trace_block_plug(q
);
1763 struct request
*last
= list_entry_rq(plug
->list
.prev
);
1764 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
1765 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
1766 blk_flush_plug_list(plug
, false);
1767 trace_block_plug(q
);
1770 list_add_tail(&req
->queuelist
, &plug
->list
);
1771 blk_account_io_start(req
, true);
1773 spin_lock_irq(q
->queue_lock
);
1774 add_acct_request(q
, req
, where
);
1777 spin_unlock_irq(q
->queue_lock
);
1780 return BLK_QC_T_NONE
;
1784 * If bio->bi_dev is a partition, remap the location
1786 static inline void blk_partition_remap(struct bio
*bio
)
1788 struct block_device
*bdev
= bio
->bi_bdev
;
1790 if (bio_sectors(bio
) && bdev
!= bdev
->bd_contains
) {
1791 struct hd_struct
*p
= bdev
->bd_part
;
1793 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1794 bio
->bi_bdev
= bdev
->bd_contains
;
1796 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1798 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1802 static void handle_bad_sector(struct bio
*bio
)
1804 char b
[BDEVNAME_SIZE
];
1806 printk(KERN_INFO
"attempt to access beyond end of device\n");
1807 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
1808 bdevname(bio
->bi_bdev
, b
),
1810 (unsigned long long)bio_end_sector(bio
),
1811 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1814 #ifdef CONFIG_FAIL_MAKE_REQUEST
1816 static DECLARE_FAULT_ATTR(fail_make_request
);
1818 static int __init
setup_fail_make_request(char *str
)
1820 return setup_fault_attr(&fail_make_request
, str
);
1822 __setup("fail_make_request=", setup_fail_make_request
);
1824 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1826 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1829 static int __init
fail_make_request_debugfs(void)
1831 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1832 NULL
, &fail_make_request
);
1834 return PTR_ERR_OR_ZERO(dir
);
1837 late_initcall(fail_make_request_debugfs
);
1839 #else /* CONFIG_FAIL_MAKE_REQUEST */
1841 static inline bool should_fail_request(struct hd_struct
*part
,
1847 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1850 * Check whether this bio extends beyond the end of the device.
1852 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1859 /* Test device or partition size, when known. */
1860 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1862 sector_t sector
= bio
->bi_iter
.bi_sector
;
1864 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1866 * This may well happen - the kernel calls bread()
1867 * without checking the size of the device, e.g., when
1868 * mounting a device.
1870 handle_bad_sector(bio
);
1878 static noinline_for_stack
bool
1879 generic_make_request_checks(struct bio
*bio
)
1881 struct request_queue
*q
;
1882 int nr_sectors
= bio_sectors(bio
);
1884 char b
[BDEVNAME_SIZE
];
1885 struct hd_struct
*part
;
1889 if (bio_check_eod(bio
, nr_sectors
))
1892 q
= bdev_get_queue(bio
->bi_bdev
);
1895 "generic_make_request: Trying to access "
1896 "nonexistent block-device %s (%Lu)\n",
1897 bdevname(bio
->bi_bdev
, b
),
1898 (long long) bio
->bi_iter
.bi_sector
);
1902 part
= bio
->bi_bdev
->bd_part
;
1903 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1904 should_fail_request(&part_to_disk(part
)->part0
,
1905 bio
->bi_iter
.bi_size
))
1909 * If this device has partitions, remap block n
1910 * of partition p to block n+start(p) of the disk.
1912 blk_partition_remap(bio
);
1914 if (bio_check_eod(bio
, nr_sectors
))
1918 * Filter flush bio's early so that make_request based
1919 * drivers without flush support don't have to worry
1922 if ((bio
->bi_opf
& (REQ_PREFLUSH
| REQ_FUA
)) &&
1923 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
1924 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
1931 switch (bio_op(bio
)) {
1932 case REQ_OP_DISCARD
:
1933 if (!blk_queue_discard(q
))
1936 case REQ_OP_SECURE_ERASE
:
1937 if (!blk_queue_secure_erase(q
))
1940 case REQ_OP_WRITE_SAME
:
1941 if (!bdev_write_same(bio
->bi_bdev
))
1943 case REQ_OP_ZONE_REPORT
:
1944 case REQ_OP_ZONE_RESET
:
1945 if (!bdev_is_zoned(bio
->bi_bdev
))
1953 * Various block parts want %current->io_context and lazy ioc
1954 * allocation ends up trading a lot of pain for a small amount of
1955 * memory. Just allocate it upfront. This may fail and block
1956 * layer knows how to live with it.
1958 create_io_context(GFP_ATOMIC
, q
->node
);
1960 if (!blkcg_bio_issue_check(q
, bio
))
1963 trace_block_bio_queue(q
, bio
);
1969 bio
->bi_error
= err
;
1975 * generic_make_request - hand a buffer to its device driver for I/O
1976 * @bio: The bio describing the location in memory and on the device.
1978 * generic_make_request() is used to make I/O requests of block
1979 * devices. It is passed a &struct bio, which describes the I/O that needs
1982 * generic_make_request() does not return any status. The
1983 * success/failure status of the request, along with notification of
1984 * completion, is delivered asynchronously through the bio->bi_end_io
1985 * function described (one day) else where.
1987 * The caller of generic_make_request must make sure that bi_io_vec
1988 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1989 * set to describe the device address, and the
1990 * bi_end_io and optionally bi_private are set to describe how
1991 * completion notification should be signaled.
1993 * generic_make_request and the drivers it calls may use bi_next if this
1994 * bio happens to be merged with someone else, and may resubmit the bio to
1995 * a lower device by calling into generic_make_request recursively, which
1996 * means the bio should NOT be touched after the call to ->make_request_fn.
1998 blk_qc_t
generic_make_request(struct bio
*bio
)
2000 struct bio_list bio_list_on_stack
;
2001 blk_qc_t ret
= BLK_QC_T_NONE
;
2003 if (!generic_make_request_checks(bio
))
2007 * We only want one ->make_request_fn to be active at a time, else
2008 * stack usage with stacked devices could be a problem. So use
2009 * current->bio_list to keep a list of requests submited by a
2010 * make_request_fn function. current->bio_list is also used as a
2011 * flag to say if generic_make_request is currently active in this
2012 * task or not. If it is NULL, then no make_request is active. If
2013 * it is non-NULL, then a make_request is active, and new requests
2014 * should be added at the tail
2016 if (current
->bio_list
) {
2017 bio_list_add(current
->bio_list
, bio
);
2021 /* following loop may be a bit non-obvious, and so deserves some
2023 * Before entering the loop, bio->bi_next is NULL (as all callers
2024 * ensure that) so we have a list with a single bio.
2025 * We pretend that we have just taken it off a longer list, so
2026 * we assign bio_list to a pointer to the bio_list_on_stack,
2027 * thus initialising the bio_list of new bios to be
2028 * added. ->make_request() may indeed add some more bios
2029 * through a recursive call to generic_make_request. If it
2030 * did, we find a non-NULL value in bio_list and re-enter the loop
2031 * from the top. In this case we really did just take the bio
2032 * of the top of the list (no pretending) and so remove it from
2033 * bio_list, and call into ->make_request() again.
2035 BUG_ON(bio
->bi_next
);
2036 bio_list_init(&bio_list_on_stack
);
2037 current
->bio_list
= &bio_list_on_stack
;
2039 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
2041 if (likely(blk_queue_enter(q
, false) == 0)) {
2042 ret
= q
->make_request_fn(q
, bio
);
2046 bio
= bio_list_pop(current
->bio_list
);
2048 struct bio
*bio_next
= bio_list_pop(current
->bio_list
);
2054 current
->bio_list
= NULL
; /* deactivate */
2059 EXPORT_SYMBOL(generic_make_request
);
2062 * submit_bio - submit a bio to the block device layer for I/O
2063 * @bio: The &struct bio which describes the I/O
2065 * submit_bio() is very similar in purpose to generic_make_request(), and
2066 * uses that function to do most of the work. Both are fairly rough
2067 * interfaces; @bio must be presetup and ready for I/O.
2070 blk_qc_t
submit_bio(struct bio
*bio
)
2073 * If it's a regular read/write or a barrier with data attached,
2074 * go through the normal accounting stuff before submission.
2076 if (bio_has_data(bio
)) {
2079 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2080 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
2082 count
= bio_sectors(bio
);
2084 if (op_is_write(bio_op(bio
))) {
2085 count_vm_events(PGPGOUT
, count
);
2087 task_io_account_read(bio
->bi_iter
.bi_size
);
2088 count_vm_events(PGPGIN
, count
);
2091 if (unlikely(block_dump
)) {
2092 char b
[BDEVNAME_SIZE
];
2093 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2094 current
->comm
, task_pid_nr(current
),
2095 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2096 (unsigned long long)bio
->bi_iter
.bi_sector
,
2097 bdevname(bio
->bi_bdev
, b
),
2102 return generic_make_request(bio
);
2104 EXPORT_SYMBOL(submit_bio
);
2107 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2108 * for new the queue limits
2110 * @rq: the request being checked
2113 * @rq may have been made based on weaker limitations of upper-level queues
2114 * in request stacking drivers, and it may violate the limitation of @q.
2115 * Since the block layer and the underlying device driver trust @rq
2116 * after it is inserted to @q, it should be checked against @q before
2117 * the insertion using this generic function.
2119 * Request stacking drivers like request-based dm may change the queue
2120 * limits when retrying requests on other queues. Those requests need
2121 * to be checked against the new queue limits again during dispatch.
2123 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2126 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2127 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2132 * queue's settings related to segment counting like q->bounce_pfn
2133 * may differ from that of other stacking queues.
2134 * Recalculate it to check the request correctly on this queue's
2137 blk_recalc_rq_segments(rq
);
2138 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2139 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2147 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2148 * @q: the queue to submit the request
2149 * @rq: the request being queued
2151 int blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2153 unsigned long flags
;
2154 int where
= ELEVATOR_INSERT_BACK
;
2156 if (blk_cloned_rq_check_limits(q
, rq
))
2160 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2164 if (blk_queue_io_stat(q
))
2165 blk_account_io_start(rq
, true);
2166 blk_mq_insert_request(rq
, false, true, false);
2170 spin_lock_irqsave(q
->queue_lock
, flags
);
2171 if (unlikely(blk_queue_dying(q
))) {
2172 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2177 * Submitting request must be dequeued before calling this function
2178 * because it will be linked to another request_queue
2180 BUG_ON(blk_queued_rq(rq
));
2182 if (rq
->cmd_flags
& (REQ_PREFLUSH
| REQ_FUA
))
2183 where
= ELEVATOR_INSERT_FLUSH
;
2185 add_acct_request(q
, rq
, where
);
2186 if (where
== ELEVATOR_INSERT_FLUSH
)
2188 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2192 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2195 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2196 * @rq: request to examine
2199 * A request could be merge of IOs which require different failure
2200 * handling. This function determines the number of bytes which
2201 * can be failed from the beginning of the request without
2202 * crossing into area which need to be retried further.
2205 * The number of bytes to fail.
2208 * queue_lock must be held.
2210 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2212 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2213 unsigned int bytes
= 0;
2216 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2217 return blk_rq_bytes(rq
);
2220 * Currently the only 'mixing' which can happen is between
2221 * different fastfail types. We can safely fail portions
2222 * which have all the failfast bits that the first one has -
2223 * the ones which are at least as eager to fail as the first
2226 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2227 if ((bio
->bi_opf
& ff
) != ff
)
2229 bytes
+= bio
->bi_iter
.bi_size
;
2232 /* this could lead to infinite loop */
2233 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2236 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2238 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2240 if (blk_do_io_stat(req
)) {
2241 const int rw
= rq_data_dir(req
);
2242 struct hd_struct
*part
;
2245 cpu
= part_stat_lock();
2247 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2252 void blk_account_io_done(struct request
*req
)
2255 * Account IO completion. flush_rq isn't accounted as a
2256 * normal IO on queueing nor completion. Accounting the
2257 * containing request is enough.
2259 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2260 unsigned long duration
= jiffies
- req
->start_time
;
2261 const int rw
= rq_data_dir(req
);
2262 struct hd_struct
*part
;
2265 cpu
= part_stat_lock();
2268 part_stat_inc(cpu
, part
, ios
[rw
]);
2269 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2270 part_round_stats(cpu
, part
);
2271 part_dec_in_flight(part
, rw
);
2273 hd_struct_put(part
);
2280 * Don't process normal requests when queue is suspended
2281 * or in the process of suspending/resuming
2283 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2286 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2287 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->rq_flags
& RQF_PM
))))
2293 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2300 void blk_account_io_start(struct request
*rq
, bool new_io
)
2302 struct hd_struct
*part
;
2303 int rw
= rq_data_dir(rq
);
2306 if (!blk_do_io_stat(rq
))
2309 cpu
= part_stat_lock();
2313 part_stat_inc(cpu
, part
, merges
[rw
]);
2315 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2316 if (!hd_struct_try_get(part
)) {
2318 * The partition is already being removed,
2319 * the request will be accounted on the disk only
2321 * We take a reference on disk->part0 although that
2322 * partition will never be deleted, so we can treat
2323 * it as any other partition.
2325 part
= &rq
->rq_disk
->part0
;
2326 hd_struct_get(part
);
2328 part_round_stats(cpu
, part
);
2329 part_inc_in_flight(part
, rw
);
2337 * blk_peek_request - peek at the top of a request queue
2338 * @q: request queue to peek at
2341 * Return the request at the top of @q. The returned request
2342 * should be started using blk_start_request() before LLD starts
2346 * Pointer to the request at the top of @q if available. Null
2350 * queue_lock must be held.
2352 struct request
*blk_peek_request(struct request_queue
*q
)
2357 while ((rq
= __elv_next_request(q
)) != NULL
) {
2359 rq
= blk_pm_peek_request(q
, rq
);
2363 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2365 * This is the first time the device driver
2366 * sees this request (possibly after
2367 * requeueing). Notify IO scheduler.
2369 if (rq
->rq_flags
& RQF_SORTED
)
2370 elv_activate_rq(q
, rq
);
2373 * just mark as started even if we don't start
2374 * it, a request that has been delayed should
2375 * not be passed by new incoming requests
2377 rq
->rq_flags
|= RQF_STARTED
;
2378 trace_block_rq_issue(q
, rq
);
2381 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2382 q
->end_sector
= rq_end_sector(rq
);
2383 q
->boundary_rq
= NULL
;
2386 if (rq
->rq_flags
& RQF_DONTPREP
)
2389 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2391 * make sure space for the drain appears we
2392 * know we can do this because max_hw_segments
2393 * has been adjusted to be one fewer than the
2396 rq
->nr_phys_segments
++;
2402 ret
= q
->prep_rq_fn(q
, rq
);
2403 if (ret
== BLKPREP_OK
) {
2405 } else if (ret
== BLKPREP_DEFER
) {
2407 * the request may have been (partially) prepped.
2408 * we need to keep this request in the front to
2409 * avoid resource deadlock. RQF_STARTED will
2410 * prevent other fs requests from passing this one.
2412 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2413 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2415 * remove the space for the drain we added
2416 * so that we don't add it again
2418 --rq
->nr_phys_segments
;
2423 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2424 int err
= (ret
== BLKPREP_INVALID
) ? -EREMOTEIO
: -EIO
;
2426 rq
->rq_flags
|= RQF_QUIET
;
2428 * Mark this request as started so we don't trigger
2429 * any debug logic in the end I/O path.
2431 blk_start_request(rq
);
2432 __blk_end_request_all(rq
, err
);
2434 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2441 EXPORT_SYMBOL(blk_peek_request
);
2443 void blk_dequeue_request(struct request
*rq
)
2445 struct request_queue
*q
= rq
->q
;
2447 BUG_ON(list_empty(&rq
->queuelist
));
2448 BUG_ON(ELV_ON_HASH(rq
));
2450 list_del_init(&rq
->queuelist
);
2453 * the time frame between a request being removed from the lists
2454 * and to it is freed is accounted as io that is in progress at
2457 if (blk_account_rq(rq
)) {
2458 q
->in_flight
[rq_is_sync(rq
)]++;
2459 set_io_start_time_ns(rq
);
2464 * blk_start_request - start request processing on the driver
2465 * @req: request to dequeue
2468 * Dequeue @req and start timeout timer on it. This hands off the
2469 * request to the driver.
2471 * Block internal functions which don't want to start timer should
2472 * call blk_dequeue_request().
2475 * queue_lock must be held.
2477 void blk_start_request(struct request
*req
)
2479 blk_dequeue_request(req
);
2481 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2482 blk_stat_set_issue_time(&req
->issue_stat
);
2483 req
->rq_flags
|= RQF_STATS
;
2484 wbt_issue(req
->q
->rq_wb
, &req
->issue_stat
);
2488 * We are now handing the request to the hardware, initialize
2489 * resid_len to full count and add the timeout handler.
2491 req
->resid_len
= blk_rq_bytes(req
);
2492 if (unlikely(blk_bidi_rq(req
)))
2493 req
->next_rq
->resid_len
= blk_rq_bytes(req
->next_rq
);
2495 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2498 EXPORT_SYMBOL(blk_start_request
);
2501 * blk_fetch_request - fetch a request from a request queue
2502 * @q: request queue to fetch a request from
2505 * Return the request at the top of @q. The request is started on
2506 * return and LLD can start processing it immediately.
2509 * Pointer to the request at the top of @q if available. Null
2513 * queue_lock must be held.
2515 struct request
*blk_fetch_request(struct request_queue
*q
)
2519 rq
= blk_peek_request(q
);
2521 blk_start_request(rq
);
2524 EXPORT_SYMBOL(blk_fetch_request
);
2527 * blk_update_request - Special helper function for request stacking drivers
2528 * @req: the request being processed
2529 * @error: %0 for success, < %0 for error
2530 * @nr_bytes: number of bytes to complete @req
2533 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2534 * the request structure even if @req doesn't have leftover.
2535 * If @req has leftover, sets it up for the next range of segments.
2537 * This special helper function is only for request stacking drivers
2538 * (e.g. request-based dm) so that they can handle partial completion.
2539 * Actual device drivers should use blk_end_request instead.
2541 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2542 * %false return from this function.
2545 * %false - this request doesn't have any more data
2546 * %true - this request has more data
2548 bool blk_update_request(struct request
*req
, int error
, unsigned int nr_bytes
)
2552 trace_block_rq_complete(req
->q
, req
, nr_bytes
);
2558 * For fs requests, rq is just carrier of independent bio's
2559 * and each partial completion should be handled separately.
2560 * Reset per-request error on each partial completion.
2562 * TODO: tj: This is too subtle. It would be better to let
2563 * low level drivers do what they see fit.
2565 if (req
->cmd_type
== REQ_TYPE_FS
)
2568 if (error
&& req
->cmd_type
== REQ_TYPE_FS
&&
2569 !(req
->rq_flags
& RQF_QUIET
)) {
2574 error_type
= "recoverable transport";
2577 error_type
= "critical target";
2580 error_type
= "critical nexus";
2583 error_type
= "timeout";
2586 error_type
= "critical space allocation";
2589 error_type
= "critical medium";
2596 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
2597 __func__
, error_type
, req
->rq_disk
?
2598 req
->rq_disk
->disk_name
: "?",
2599 (unsigned long long)blk_rq_pos(req
));
2603 blk_account_io_completion(req
, nr_bytes
);
2607 struct bio
*bio
= req
->bio
;
2608 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2610 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2611 req
->bio
= bio
->bi_next
;
2613 req_bio_endio(req
, bio
, bio_bytes
, error
);
2615 total_bytes
+= bio_bytes
;
2616 nr_bytes
-= bio_bytes
;
2627 * Reset counters so that the request stacking driver
2628 * can find how many bytes remain in the request
2631 req
->__data_len
= 0;
2635 req
->__data_len
-= total_bytes
;
2637 /* update sector only for requests with clear definition of sector */
2638 if (req
->cmd_type
== REQ_TYPE_FS
)
2639 req
->__sector
+= total_bytes
>> 9;
2641 /* mixed attributes always follow the first bio */
2642 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2643 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2644 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2648 * If total number of sectors is less than the first segment
2649 * size, something has gone terribly wrong.
2651 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2652 blk_dump_rq_flags(req
, "request botched");
2653 req
->__data_len
= blk_rq_cur_bytes(req
);
2656 /* recalculate the number of segments */
2657 blk_recalc_rq_segments(req
);
2661 EXPORT_SYMBOL_GPL(blk_update_request
);
2663 static bool blk_update_bidi_request(struct request
*rq
, int error
,
2664 unsigned int nr_bytes
,
2665 unsigned int bidi_bytes
)
2667 if (blk_update_request(rq
, error
, nr_bytes
))
2670 /* Bidi request must be completed as a whole */
2671 if (unlikely(blk_bidi_rq(rq
)) &&
2672 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2675 if (blk_queue_add_random(rq
->q
))
2676 add_disk_randomness(rq
->rq_disk
);
2682 * blk_unprep_request - unprepare a request
2685 * This function makes a request ready for complete resubmission (or
2686 * completion). It happens only after all error handling is complete,
2687 * so represents the appropriate moment to deallocate any resources
2688 * that were allocated to the request in the prep_rq_fn. The queue
2689 * lock is held when calling this.
2691 void blk_unprep_request(struct request
*req
)
2693 struct request_queue
*q
= req
->q
;
2695 req
->rq_flags
&= ~RQF_DONTPREP
;
2696 if (q
->unprep_rq_fn
)
2697 q
->unprep_rq_fn(q
, req
);
2699 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2702 * queue lock must be held
2704 void blk_finish_request(struct request
*req
, int error
)
2706 struct request_queue
*q
= req
->q
;
2708 if (req
->rq_flags
& RQF_STATS
)
2709 blk_stat_add(&q
->rq_stats
[rq_data_dir(req
)], req
);
2711 if (req
->rq_flags
& RQF_QUEUED
)
2712 blk_queue_end_tag(q
, req
);
2714 BUG_ON(blk_queued_rq(req
));
2716 if (unlikely(laptop_mode
) && req
->cmd_type
== REQ_TYPE_FS
)
2717 laptop_io_completion(&req
->q
->backing_dev_info
);
2719 blk_delete_timer(req
);
2721 if (req
->rq_flags
& RQF_DONTPREP
)
2722 blk_unprep_request(req
);
2724 blk_account_io_done(req
);
2727 wbt_done(req
->q
->rq_wb
, &req
->issue_stat
);
2728 req
->end_io(req
, error
);
2730 if (blk_bidi_rq(req
))
2731 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2733 __blk_put_request(q
, req
);
2736 EXPORT_SYMBOL(blk_finish_request
);
2739 * blk_end_bidi_request - Complete a bidi request
2740 * @rq: the request to complete
2741 * @error: %0 for success, < %0 for error
2742 * @nr_bytes: number of bytes to complete @rq
2743 * @bidi_bytes: number of bytes to complete @rq->next_rq
2746 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2747 * Drivers that supports bidi can safely call this member for any
2748 * type of request, bidi or uni. In the later case @bidi_bytes is
2752 * %false - we are done with this request
2753 * %true - still buffers pending for this request
2755 static bool blk_end_bidi_request(struct request
*rq
, int error
,
2756 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2758 struct request_queue
*q
= rq
->q
;
2759 unsigned long flags
;
2761 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2764 spin_lock_irqsave(q
->queue_lock
, flags
);
2765 blk_finish_request(rq
, error
);
2766 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2772 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2773 * @rq: the request to complete
2774 * @error: %0 for success, < %0 for error
2775 * @nr_bytes: number of bytes to complete @rq
2776 * @bidi_bytes: number of bytes to complete @rq->next_rq
2779 * Identical to blk_end_bidi_request() except that queue lock is
2780 * assumed to be locked on entry and remains so on return.
2783 * %false - we are done with this request
2784 * %true - still buffers pending for this request
2786 bool __blk_end_bidi_request(struct request
*rq
, int error
,
2787 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2789 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2792 blk_finish_request(rq
, error
);
2798 * blk_end_request - Helper function for drivers to complete the request.
2799 * @rq: the request being processed
2800 * @error: %0 for success, < %0 for error
2801 * @nr_bytes: number of bytes to complete
2804 * Ends I/O on a number of bytes attached to @rq.
2805 * If @rq has leftover, sets it up for the next range of segments.
2808 * %false - we are done with this request
2809 * %true - still buffers pending for this request
2811 bool blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2813 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2815 EXPORT_SYMBOL(blk_end_request
);
2818 * blk_end_request_all - Helper function for drives to finish the request.
2819 * @rq: the request to finish
2820 * @error: %0 for success, < %0 for error
2823 * Completely finish @rq.
2825 void blk_end_request_all(struct request
*rq
, int error
)
2828 unsigned int bidi_bytes
= 0;
2830 if (unlikely(blk_bidi_rq(rq
)))
2831 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2833 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2836 EXPORT_SYMBOL(blk_end_request_all
);
2839 * blk_end_request_cur - Helper function to finish the current request chunk.
2840 * @rq: the request to finish the current chunk for
2841 * @error: %0 for success, < %0 for error
2844 * Complete the current consecutively mapped chunk from @rq.
2847 * %false - we are done with this request
2848 * %true - still buffers pending for this request
2850 bool blk_end_request_cur(struct request
*rq
, int error
)
2852 return blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2854 EXPORT_SYMBOL(blk_end_request_cur
);
2857 * blk_end_request_err - Finish a request till the next failure boundary.
2858 * @rq: the request to finish till the next failure boundary for
2859 * @error: must be negative errno
2862 * Complete @rq till the next failure boundary.
2865 * %false - we are done with this request
2866 * %true - still buffers pending for this request
2868 bool blk_end_request_err(struct request
*rq
, int error
)
2870 WARN_ON(error
>= 0);
2871 return blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2873 EXPORT_SYMBOL_GPL(blk_end_request_err
);
2876 * __blk_end_request - Helper function for drivers to complete the request.
2877 * @rq: the request being processed
2878 * @error: %0 for success, < %0 for error
2879 * @nr_bytes: number of bytes to complete
2882 * Must be called with queue lock held unlike blk_end_request().
2885 * %false - we are done with this request
2886 * %true - still buffers pending for this request
2888 bool __blk_end_request(struct request
*rq
, int error
, unsigned int nr_bytes
)
2890 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2892 EXPORT_SYMBOL(__blk_end_request
);
2895 * __blk_end_request_all - Helper function for drives to finish the request.
2896 * @rq: the request to finish
2897 * @error: %0 for success, < %0 for error
2900 * Completely finish @rq. Must be called with queue lock held.
2902 void __blk_end_request_all(struct request
*rq
, int error
)
2905 unsigned int bidi_bytes
= 0;
2907 if (unlikely(blk_bidi_rq(rq
)))
2908 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2910 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2913 EXPORT_SYMBOL(__blk_end_request_all
);
2916 * __blk_end_request_cur - Helper function to finish the current request chunk.
2917 * @rq: the request to finish the current chunk for
2918 * @error: %0 for success, < %0 for error
2921 * Complete the current consecutively mapped chunk from @rq. Must
2922 * be called with queue lock held.
2925 * %false - we are done with this request
2926 * %true - still buffers pending for this request
2928 bool __blk_end_request_cur(struct request
*rq
, int error
)
2930 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2932 EXPORT_SYMBOL(__blk_end_request_cur
);
2935 * __blk_end_request_err - Finish a request till the next failure boundary.
2936 * @rq: the request to finish till the next failure boundary for
2937 * @error: must be negative errno
2940 * Complete @rq till the next failure boundary. Must be called
2941 * with queue lock held.
2944 * %false - we are done with this request
2945 * %true - still buffers pending for this request
2947 bool __blk_end_request_err(struct request
*rq
, int error
)
2949 WARN_ON(error
>= 0);
2950 return __blk_end_request(rq
, error
, blk_rq_err_bytes(rq
));
2952 EXPORT_SYMBOL_GPL(__blk_end_request_err
);
2954 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2957 if (bio_has_data(bio
))
2958 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2960 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2961 rq
->bio
= rq
->biotail
= bio
;
2964 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2967 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2969 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2970 * @rq: the request to be flushed
2973 * Flush all pages in @rq.
2975 void rq_flush_dcache_pages(struct request
*rq
)
2977 struct req_iterator iter
;
2978 struct bio_vec bvec
;
2980 rq_for_each_segment(bvec
, rq
, iter
)
2981 flush_dcache_page(bvec
.bv_page
);
2983 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2987 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2988 * @q : the queue of the device being checked
2991 * Check if underlying low-level drivers of a device are busy.
2992 * If the drivers want to export their busy state, they must set own
2993 * exporting function using blk_queue_lld_busy() first.
2995 * Basically, this function is used only by request stacking drivers
2996 * to stop dispatching requests to underlying devices when underlying
2997 * devices are busy. This behavior helps more I/O merging on the queue
2998 * of the request stacking driver and prevents I/O throughput regression
2999 * on burst I/O load.
3002 * 0 - Not busy (The request stacking driver should dispatch request)
3003 * 1 - Busy (The request stacking driver should stop dispatching request)
3005 int blk_lld_busy(struct request_queue
*q
)
3008 return q
->lld_busy_fn(q
);
3012 EXPORT_SYMBOL_GPL(blk_lld_busy
);
3015 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3016 * @rq: the clone request to be cleaned up
3019 * Free all bios in @rq for a cloned request.
3021 void blk_rq_unprep_clone(struct request
*rq
)
3025 while ((bio
= rq
->bio
) != NULL
) {
3026 rq
->bio
= bio
->bi_next
;
3031 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3034 * Copy attributes of the original request to the clone request.
3035 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3037 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3039 dst
->cpu
= src
->cpu
;
3040 dst
->cmd_flags
= src
->cmd_flags
| REQ_NOMERGE
;
3041 dst
->cmd_type
= src
->cmd_type
;
3042 dst
->__sector
= blk_rq_pos(src
);
3043 dst
->__data_len
= blk_rq_bytes(src
);
3044 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3045 dst
->ioprio
= src
->ioprio
;
3046 dst
->extra_len
= src
->extra_len
;
3050 * blk_rq_prep_clone - Helper function to setup clone request
3051 * @rq: the request to be setup
3052 * @rq_src: original request to be cloned
3053 * @bs: bio_set that bios for clone are allocated from
3054 * @gfp_mask: memory allocation mask for bio
3055 * @bio_ctr: setup function to be called for each clone bio.
3056 * Returns %0 for success, non %0 for failure.
3057 * @data: private data to be passed to @bio_ctr
3060 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3061 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3062 * are not copied, and copying such parts is the caller's responsibility.
3063 * Also, pages which the original bios are pointing to are not copied
3064 * and the cloned bios just point same pages.
3065 * So cloned bios must be completed before original bios, which means
3066 * the caller must complete @rq before @rq_src.
3068 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3069 struct bio_set
*bs
, gfp_t gfp_mask
,
3070 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3073 struct bio
*bio
, *bio_src
;
3078 __rq_for_each_bio(bio_src
, rq_src
) {
3079 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3083 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3087 rq
->biotail
->bi_next
= bio
;
3090 rq
->bio
= rq
->biotail
= bio
;
3093 __blk_rq_prep_clone(rq
, rq_src
);
3100 blk_rq_unprep_clone(rq
);
3104 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3106 int kblockd_schedule_work(struct work_struct
*work
)
3108 return queue_work(kblockd_workqueue
, work
);
3110 EXPORT_SYMBOL(kblockd_schedule_work
);
3112 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3114 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3116 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3118 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3119 unsigned long delay
)
3121 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3123 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3125 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3126 unsigned long delay
)
3128 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3130 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3133 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3134 * @plug: The &struct blk_plug that needs to be initialized
3137 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3138 * pending I/O should the task end up blocking between blk_start_plug() and
3139 * blk_finish_plug(). This is important from a performance perspective, but
3140 * also ensures that we don't deadlock. For instance, if the task is blocking
3141 * for a memory allocation, memory reclaim could end up wanting to free a
3142 * page belonging to that request that is currently residing in our private
3143 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3144 * this kind of deadlock.
3146 void blk_start_plug(struct blk_plug
*plug
)
3148 struct task_struct
*tsk
= current
;
3151 * If this is a nested plug, don't actually assign it.
3156 INIT_LIST_HEAD(&plug
->list
);
3157 INIT_LIST_HEAD(&plug
->mq_list
);
3158 INIT_LIST_HEAD(&plug
->cb_list
);
3160 * Store ordering should not be needed here, since a potential
3161 * preempt will imply a full memory barrier
3165 EXPORT_SYMBOL(blk_start_plug
);
3167 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3169 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3170 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3172 return !(rqa
->q
< rqb
->q
||
3173 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3177 * If 'from_schedule' is true, then postpone the dispatch of requests
3178 * until a safe kblockd context. We due this to avoid accidental big
3179 * additional stack usage in driver dispatch, in places where the originally
3180 * plugger did not intend it.
3182 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3184 __releases(q
->queue_lock
)
3186 trace_block_unplug(q
, depth
, !from_schedule
);
3189 blk_run_queue_async(q
);
3192 spin_unlock(q
->queue_lock
);
3195 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3197 LIST_HEAD(callbacks
);
3199 while (!list_empty(&plug
->cb_list
)) {
3200 list_splice_init(&plug
->cb_list
, &callbacks
);
3202 while (!list_empty(&callbacks
)) {
3203 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3206 list_del(&cb
->list
);
3207 cb
->callback(cb
, from_schedule
);
3212 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3215 struct blk_plug
*plug
= current
->plug
;
3216 struct blk_plug_cb
*cb
;
3221 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3222 if (cb
->callback
== unplug
&& cb
->data
== data
)
3225 /* Not currently on the callback list */
3226 BUG_ON(size
< sizeof(*cb
));
3227 cb
= kzalloc(size
, GFP_ATOMIC
);
3230 cb
->callback
= unplug
;
3231 list_add(&cb
->list
, &plug
->cb_list
);
3235 EXPORT_SYMBOL(blk_check_plugged
);
3237 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3239 struct request_queue
*q
;
3240 unsigned long flags
;
3245 flush_plug_callbacks(plug
, from_schedule
);
3247 if (!list_empty(&plug
->mq_list
))
3248 blk_mq_flush_plug_list(plug
, from_schedule
);
3250 if (list_empty(&plug
->list
))
3253 list_splice_init(&plug
->list
, &list
);
3255 list_sort(NULL
, &list
, plug_rq_cmp
);
3261 * Save and disable interrupts here, to avoid doing it for every
3262 * queue lock we have to take.
3264 local_irq_save(flags
);
3265 while (!list_empty(&list
)) {
3266 rq
= list_entry_rq(list
.next
);
3267 list_del_init(&rq
->queuelist
);
3271 * This drops the queue lock
3274 queue_unplugged(q
, depth
, from_schedule
);
3277 spin_lock(q
->queue_lock
);
3281 * Short-circuit if @q is dead
3283 if (unlikely(blk_queue_dying(q
))) {
3284 __blk_end_request_all(rq
, -ENODEV
);
3289 * rq is already accounted, so use raw insert
3291 if (rq
->cmd_flags
& (REQ_PREFLUSH
| REQ_FUA
))
3292 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3294 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3300 * This drops the queue lock
3303 queue_unplugged(q
, depth
, from_schedule
);
3305 local_irq_restore(flags
);
3308 void blk_finish_plug(struct blk_plug
*plug
)
3310 if (plug
!= current
->plug
)
3312 blk_flush_plug_list(plug
, false);
3314 current
->plug
= NULL
;
3316 EXPORT_SYMBOL(blk_finish_plug
);
3320 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3321 * @q: the queue of the device
3322 * @dev: the device the queue belongs to
3325 * Initialize runtime-PM-related fields for @q and start auto suspend for
3326 * @dev. Drivers that want to take advantage of request-based runtime PM
3327 * should call this function after @dev has been initialized, and its
3328 * request queue @q has been allocated, and runtime PM for it can not happen
3329 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3330 * cases, driver should call this function before any I/O has taken place.
3332 * This function takes care of setting up using auto suspend for the device,
3333 * the autosuspend delay is set to -1 to make runtime suspend impossible
3334 * until an updated value is either set by user or by driver. Drivers do
3335 * not need to touch other autosuspend settings.
3337 * The block layer runtime PM is request based, so only works for drivers
3338 * that use request as their IO unit instead of those directly use bio's.
3340 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3343 q
->rpm_status
= RPM_ACTIVE
;
3344 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3345 pm_runtime_use_autosuspend(q
->dev
);
3347 EXPORT_SYMBOL(blk_pm_runtime_init
);
3350 * blk_pre_runtime_suspend - Pre runtime suspend check
3351 * @q: the queue of the device
3354 * This function will check if runtime suspend is allowed for the device
3355 * by examining if there are any requests pending in the queue. If there
3356 * are requests pending, the device can not be runtime suspended; otherwise,
3357 * the queue's status will be updated to SUSPENDING and the driver can
3358 * proceed to suspend the device.
3360 * For the not allowed case, we mark last busy for the device so that
3361 * runtime PM core will try to autosuspend it some time later.
3363 * This function should be called near the start of the device's
3364 * runtime_suspend callback.
3367 * 0 - OK to runtime suspend the device
3368 * -EBUSY - Device should not be runtime suspended
3370 int blk_pre_runtime_suspend(struct request_queue
*q
)
3377 spin_lock_irq(q
->queue_lock
);
3378 if (q
->nr_pending
) {
3380 pm_runtime_mark_last_busy(q
->dev
);
3382 q
->rpm_status
= RPM_SUSPENDING
;
3384 spin_unlock_irq(q
->queue_lock
);
3387 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3390 * blk_post_runtime_suspend - Post runtime suspend processing
3391 * @q: the queue of the device
3392 * @err: return value of the device's runtime_suspend function
3395 * Update the queue's runtime status according to the return value of the
3396 * device's runtime suspend function and mark last busy for the device so
3397 * that PM core will try to auto suspend the device at a later time.
3399 * This function should be called near the end of the device's
3400 * runtime_suspend callback.
3402 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3407 spin_lock_irq(q
->queue_lock
);
3409 q
->rpm_status
= RPM_SUSPENDED
;
3411 q
->rpm_status
= RPM_ACTIVE
;
3412 pm_runtime_mark_last_busy(q
->dev
);
3414 spin_unlock_irq(q
->queue_lock
);
3416 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3419 * blk_pre_runtime_resume - Pre runtime resume processing
3420 * @q: the queue of the device
3423 * Update the queue's runtime status to RESUMING in preparation for the
3424 * runtime resume of the device.
3426 * This function should be called near the start of the device's
3427 * runtime_resume callback.
3429 void blk_pre_runtime_resume(struct request_queue
*q
)
3434 spin_lock_irq(q
->queue_lock
);
3435 q
->rpm_status
= RPM_RESUMING
;
3436 spin_unlock_irq(q
->queue_lock
);
3438 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3441 * blk_post_runtime_resume - Post runtime resume processing
3442 * @q: the queue of the device
3443 * @err: return value of the device's runtime_resume function
3446 * Update the queue's runtime status according to the return value of the
3447 * device's runtime_resume function. If it is successfully resumed, process
3448 * the requests that are queued into the device's queue when it is resuming
3449 * and then mark last busy and initiate autosuspend for it.
3451 * This function should be called near the end of the device's
3452 * runtime_resume callback.
3454 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3459 spin_lock_irq(q
->queue_lock
);
3461 q
->rpm_status
= RPM_ACTIVE
;
3463 pm_runtime_mark_last_busy(q
->dev
);
3464 pm_request_autosuspend(q
->dev
);
3466 q
->rpm_status
= RPM_SUSPENDED
;
3468 spin_unlock_irq(q
->queue_lock
);
3470 EXPORT_SYMBOL(blk_post_runtime_resume
);
3473 * blk_set_runtime_active - Force runtime status of the queue to be active
3474 * @q: the queue of the device
3476 * If the device is left runtime suspended during system suspend the resume
3477 * hook typically resumes the device and corrects runtime status
3478 * accordingly. However, that does not affect the queue runtime PM status
3479 * which is still "suspended". This prevents processing requests from the
3482 * This function can be used in driver's resume hook to correct queue
3483 * runtime PM status and re-enable peeking requests from the queue. It
3484 * should be called before first request is added to the queue.
3486 void blk_set_runtime_active(struct request_queue
*q
)
3488 spin_lock_irq(q
->queue_lock
);
3489 q
->rpm_status
= RPM_ACTIVE
;
3490 pm_runtime_mark_last_busy(q
->dev
);
3491 pm_request_autosuspend(q
->dev
);
3492 spin_unlock_irq(q
->queue_lock
);
3494 EXPORT_SYMBOL(blk_set_runtime_active
);
3497 int __init
blk_dev_init(void)
3499 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
3500 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3501 FIELD_SIZEOF(struct request
, cmd_flags
));
3502 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3503 FIELD_SIZEOF(struct bio
, bi_opf
));
3505 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3506 kblockd_workqueue
= alloc_workqueue("kblockd",
3507 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3508 if (!kblockd_workqueue
)
3509 panic("Failed to create kblockd\n");
3511 request_cachep
= kmem_cache_create("blkdev_requests",
3512 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3514 blk_requestq_cachep
= kmem_cache_create("request_queue",
3515 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);