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 const struct {
136 [BLK_STS_OK
] = { 0, "" },
137 [BLK_STS_NOTSUPP
] = { -EOPNOTSUPP
, "operation not supported" },
138 [BLK_STS_TIMEOUT
] = { -ETIMEDOUT
, "timeout" },
139 [BLK_STS_NOSPC
] = { -ENOSPC
, "critical space allocation" },
140 [BLK_STS_TRANSPORT
] = { -ENOLINK
, "recoverable transport" },
141 [BLK_STS_TARGET
] = { -EREMOTEIO
, "critical target" },
142 [BLK_STS_NEXUS
] = { -EBADE
, "critical nexus" },
143 [BLK_STS_MEDIUM
] = { -ENODATA
, "critical medium" },
144 [BLK_STS_PROTECTION
] = { -EILSEQ
, "protection" },
145 [BLK_STS_RESOURCE
] = { -ENOMEM
, "kernel resource" },
147 /* device mapper special case, should not leak out: */
148 [BLK_STS_DM_REQUEUE
] = { -EREMCHG
, "dm internal retry" },
150 /* everything else not covered above: */
151 [BLK_STS_IOERR
] = { -EIO
, "I/O" },
154 blk_status_t
errno_to_blk_status(int errno
)
158 for (i
= 0; i
< ARRAY_SIZE(blk_errors
); i
++) {
159 if (blk_errors
[i
].errno
== errno
)
160 return (__force blk_status_t
)i
;
163 return BLK_STS_IOERR
;
165 EXPORT_SYMBOL_GPL(errno_to_blk_status
);
167 int blk_status_to_errno(blk_status_t status
)
169 int idx
= (__force
int)status
;
171 if (WARN_ON_ONCE(idx
> ARRAY_SIZE(blk_errors
)))
173 return blk_errors
[idx
].errno
;
175 EXPORT_SYMBOL_GPL(blk_status_to_errno
);
177 static void print_req_error(struct request
*req
, blk_status_t status
)
179 int idx
= (__force
int)status
;
181 if (WARN_ON_ONCE(idx
> ARRAY_SIZE(blk_errors
)))
184 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
185 __func__
, blk_errors
[idx
].name
, req
->rq_disk
?
186 req
->rq_disk
->disk_name
: "?",
187 (unsigned long long)blk_rq_pos(req
));
190 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
191 unsigned int nbytes
, blk_status_t error
)
194 bio
->bi_status
= error
;
196 if (unlikely(rq
->rq_flags
& RQF_QUIET
))
197 bio_set_flag(bio
, BIO_QUIET
);
199 bio_advance(bio
, nbytes
);
201 /* don't actually finish bio if it's part of flush sequence */
202 if (bio
->bi_iter
.bi_size
== 0 && !(rq
->rq_flags
& RQF_FLUSH_SEQ
))
206 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
208 printk(KERN_INFO
"%s: dev %s: flags=%llx\n", msg
,
209 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?",
210 (unsigned long long) rq
->cmd_flags
);
212 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
213 (unsigned long long)blk_rq_pos(rq
),
214 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
215 printk(KERN_INFO
" bio %p, biotail %p, len %u\n",
216 rq
->bio
, rq
->biotail
, blk_rq_bytes(rq
));
218 EXPORT_SYMBOL(blk_dump_rq_flags
);
220 static void blk_delay_work(struct work_struct
*work
)
222 struct request_queue
*q
;
224 q
= container_of(work
, struct request_queue
, delay_work
.work
);
225 spin_lock_irq(q
->queue_lock
);
227 spin_unlock_irq(q
->queue_lock
);
231 * blk_delay_queue - restart queueing after defined interval
232 * @q: The &struct request_queue in question
233 * @msecs: Delay in msecs
236 * Sometimes queueing needs to be postponed for a little while, to allow
237 * resources to come back. This function will make sure that queueing is
238 * restarted around the specified time. Queue lock must be held.
240 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
242 if (likely(!blk_queue_dead(q
)))
243 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
244 msecs_to_jiffies(msecs
));
246 EXPORT_SYMBOL(blk_delay_queue
);
249 * blk_start_queue_async - asynchronously restart a previously stopped queue
250 * @q: The &struct request_queue in question
253 * blk_start_queue_async() will clear the stop flag on the queue, and
254 * ensure that the request_fn for the queue is run from an async
257 void blk_start_queue_async(struct request_queue
*q
)
259 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
260 blk_run_queue_async(q
);
262 EXPORT_SYMBOL(blk_start_queue_async
);
265 * blk_start_queue - restart a previously stopped queue
266 * @q: The &struct request_queue in question
269 * blk_start_queue() will clear the stop flag on the queue, and call
270 * the request_fn for the queue if it was in a stopped state when
271 * entered. Also see blk_stop_queue(). Queue lock must be held.
273 void blk_start_queue(struct request_queue
*q
)
275 WARN_ON(!irqs_disabled());
277 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
280 EXPORT_SYMBOL(blk_start_queue
);
283 * blk_stop_queue - stop a queue
284 * @q: The &struct request_queue in question
287 * The Linux block layer assumes that a block driver will consume all
288 * entries on the request queue when the request_fn strategy is called.
289 * Often this will not happen, because of hardware limitations (queue
290 * depth settings). If a device driver gets a 'queue full' response,
291 * or if it simply chooses not to queue more I/O at one point, it can
292 * call this function to prevent the request_fn from being called until
293 * the driver has signalled it's ready to go again. This happens by calling
294 * blk_start_queue() to restart queue operations. Queue lock must be held.
296 void blk_stop_queue(struct request_queue
*q
)
298 cancel_delayed_work(&q
->delay_work
);
299 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
301 EXPORT_SYMBOL(blk_stop_queue
);
304 * blk_sync_queue - cancel any pending callbacks on a queue
308 * The block layer may perform asynchronous callback activity
309 * on a queue, such as calling the unplug function after a timeout.
310 * A block device may call blk_sync_queue to ensure that any
311 * such activity is cancelled, thus allowing it to release resources
312 * that the callbacks might use. The caller must already have made sure
313 * that its ->make_request_fn will not re-add plugging prior to calling
316 * This function does not cancel any asynchronous activity arising
317 * out of elevator or throttling code. That would require elevator_exit()
318 * and blkcg_exit_queue() to be called with queue lock initialized.
321 void blk_sync_queue(struct request_queue
*q
)
323 del_timer_sync(&q
->timeout
);
326 struct blk_mq_hw_ctx
*hctx
;
329 queue_for_each_hw_ctx(q
, hctx
, i
)
330 cancel_delayed_work_sync(&hctx
->run_work
);
332 cancel_delayed_work_sync(&q
->delay_work
);
335 EXPORT_SYMBOL(blk_sync_queue
);
338 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
339 * @q: The queue to run
342 * Invoke request handling on a queue if there are any pending requests.
343 * May be used to restart request handling after a request has completed.
344 * This variant runs the queue whether or not the queue has been
345 * stopped. Must be called with the queue lock held and interrupts
346 * disabled. See also @blk_run_queue.
348 inline void __blk_run_queue_uncond(struct request_queue
*q
)
350 if (unlikely(blk_queue_dead(q
)))
354 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
355 * the queue lock internally. As a result multiple threads may be
356 * running such a request function concurrently. Keep track of the
357 * number of active request_fn invocations such that blk_drain_queue()
358 * can wait until all these request_fn calls have finished.
360 q
->request_fn_active
++;
362 q
->request_fn_active
--;
364 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond
);
367 * __blk_run_queue - run a single device queue
368 * @q: The queue to run
371 * See @blk_run_queue. This variant must be called with the queue lock
372 * held and interrupts disabled.
374 void __blk_run_queue(struct request_queue
*q
)
376 if (unlikely(blk_queue_stopped(q
)))
379 __blk_run_queue_uncond(q
);
381 EXPORT_SYMBOL(__blk_run_queue
);
384 * blk_run_queue_async - run a single device queue in workqueue context
385 * @q: The queue to run
388 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
389 * of us. The caller must hold the queue lock.
391 void blk_run_queue_async(struct request_queue
*q
)
393 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
394 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
396 EXPORT_SYMBOL(blk_run_queue_async
);
399 * blk_run_queue - run a single device queue
400 * @q: The queue to run
403 * Invoke request handling on this queue, if it has pending work to do.
404 * May be used to restart queueing when a request has completed.
406 void blk_run_queue(struct request_queue
*q
)
410 spin_lock_irqsave(q
->queue_lock
, flags
);
412 spin_unlock_irqrestore(q
->queue_lock
, flags
);
414 EXPORT_SYMBOL(blk_run_queue
);
416 void blk_put_queue(struct request_queue
*q
)
418 kobject_put(&q
->kobj
);
420 EXPORT_SYMBOL(blk_put_queue
);
423 * __blk_drain_queue - drain requests from request_queue
425 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
427 * Drain requests from @q. If @drain_all is set, all requests are drained.
428 * If not, only ELVPRIV requests are drained. The caller is responsible
429 * for ensuring that no new requests which need to be drained are queued.
431 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
432 __releases(q
->queue_lock
)
433 __acquires(q
->queue_lock
)
437 lockdep_assert_held(q
->queue_lock
);
443 * The caller might be trying to drain @q before its
444 * elevator is initialized.
447 elv_drain_elevator(q
);
449 blkcg_drain_queue(q
);
452 * This function might be called on a queue which failed
453 * driver init after queue creation or is not yet fully
454 * active yet. Some drivers (e.g. fd and loop) get unhappy
455 * in such cases. Kick queue iff dispatch queue has
456 * something on it and @q has request_fn set.
458 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
461 drain
|= q
->nr_rqs_elvpriv
;
462 drain
|= q
->request_fn_active
;
465 * Unfortunately, requests are queued at and tracked from
466 * multiple places and there's no single counter which can
467 * be drained. Check all the queues and counters.
470 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
471 drain
|= !list_empty(&q
->queue_head
);
472 for (i
= 0; i
< 2; i
++) {
473 drain
|= q
->nr_rqs
[i
];
474 drain
|= q
->in_flight
[i
];
476 drain
|= !list_empty(&fq
->flush_queue
[i
]);
483 spin_unlock_irq(q
->queue_lock
);
487 spin_lock_irq(q
->queue_lock
);
491 * With queue marked dead, any woken up waiter will fail the
492 * allocation path, so the wakeup chaining is lost and we're
493 * left with hung waiters. We need to wake up those waiters.
496 struct request_list
*rl
;
498 blk_queue_for_each_rl(rl
, q
)
499 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
500 wake_up_all(&rl
->wait
[i
]);
505 * blk_queue_bypass_start - enter queue bypass mode
506 * @q: queue of interest
508 * In bypass mode, only the dispatch FIFO queue of @q is used. This
509 * function makes @q enter bypass mode and drains all requests which were
510 * throttled or issued before. On return, it's guaranteed that no request
511 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
512 * inside queue or RCU read lock.
514 void blk_queue_bypass_start(struct request_queue
*q
)
516 spin_lock_irq(q
->queue_lock
);
518 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
519 spin_unlock_irq(q
->queue_lock
);
522 * Queues start drained. Skip actual draining till init is
523 * complete. This avoids lenghty delays during queue init which
524 * can happen many times during boot.
526 if (blk_queue_init_done(q
)) {
527 spin_lock_irq(q
->queue_lock
);
528 __blk_drain_queue(q
, false);
529 spin_unlock_irq(q
->queue_lock
);
531 /* ensure blk_queue_bypass() is %true inside RCU read lock */
535 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
538 * blk_queue_bypass_end - leave queue bypass mode
539 * @q: queue of interest
541 * Leave bypass mode and restore the normal queueing behavior.
543 void blk_queue_bypass_end(struct request_queue
*q
)
545 spin_lock_irq(q
->queue_lock
);
546 if (!--q
->bypass_depth
)
547 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
548 WARN_ON_ONCE(q
->bypass_depth
< 0);
549 spin_unlock_irq(q
->queue_lock
);
551 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
553 void blk_set_queue_dying(struct request_queue
*q
)
555 spin_lock_irq(q
->queue_lock
);
556 queue_flag_set(QUEUE_FLAG_DYING
, q
);
557 spin_unlock_irq(q
->queue_lock
);
560 * When queue DYING flag is set, we need to block new req
561 * entering queue, so we call blk_freeze_queue_start() to
562 * prevent I/O from crossing blk_queue_enter().
564 blk_freeze_queue_start(q
);
567 blk_mq_wake_waiters(q
);
569 struct request_list
*rl
;
571 spin_lock_irq(q
->queue_lock
);
572 blk_queue_for_each_rl(rl
, q
) {
574 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
575 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
578 spin_unlock_irq(q
->queue_lock
);
581 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
584 * blk_cleanup_queue - shutdown a request queue
585 * @q: request queue to shutdown
587 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
588 * put it. All future requests will be failed immediately with -ENODEV.
590 void blk_cleanup_queue(struct request_queue
*q
)
592 spinlock_t
*lock
= q
->queue_lock
;
594 /* mark @q DYING, no new request or merges will be allowed afterwards */
595 mutex_lock(&q
->sysfs_lock
);
596 blk_set_queue_dying(q
);
600 * A dying queue is permanently in bypass mode till released. Note
601 * that, unlike blk_queue_bypass_start(), we aren't performing
602 * synchronize_rcu() after entering bypass mode to avoid the delay
603 * as some drivers create and destroy a lot of queues while
604 * probing. This is still safe because blk_release_queue() will be
605 * called only after the queue refcnt drops to zero and nothing,
606 * RCU or not, would be traversing the queue by then.
609 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
611 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
612 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
613 queue_flag_set(QUEUE_FLAG_DYING
, q
);
614 spin_unlock_irq(lock
);
615 mutex_unlock(&q
->sysfs_lock
);
618 * Drain all requests queued before DYING marking. Set DEAD flag to
619 * prevent that q->request_fn() gets invoked after draining finished.
624 __blk_drain_queue(q
, true);
625 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
626 spin_unlock_irq(lock
);
628 /* for synchronous bio-based driver finish in-flight integrity i/o */
629 blk_flush_integrity();
631 /* @q won't process any more request, flush async actions */
632 del_timer_sync(&q
->backing_dev_info
->laptop_mode_wb_timer
);
636 blk_mq_free_queue(q
);
637 percpu_ref_exit(&q
->q_usage_counter
);
640 if (q
->queue_lock
!= &q
->__queue_lock
)
641 q
->queue_lock
= &q
->__queue_lock
;
642 spin_unlock_irq(lock
);
644 /* @q is and will stay empty, shutdown and put */
647 EXPORT_SYMBOL(blk_cleanup_queue
);
649 /* Allocate memory local to the request queue */
650 static void *alloc_request_simple(gfp_t gfp_mask
, void *data
)
652 struct request_queue
*q
= data
;
654 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, q
->node
);
657 static void free_request_simple(void *element
, void *data
)
659 kmem_cache_free(request_cachep
, element
);
662 static void *alloc_request_size(gfp_t gfp_mask
, void *data
)
664 struct request_queue
*q
= data
;
667 rq
= kmalloc_node(sizeof(struct request
) + q
->cmd_size
, gfp_mask
,
669 if (rq
&& q
->init_rq_fn
&& q
->init_rq_fn(q
, rq
, gfp_mask
) < 0) {
676 static void free_request_size(void *element
, void *data
)
678 struct request_queue
*q
= data
;
681 q
->exit_rq_fn(q
, element
);
685 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
688 if (unlikely(rl
->rq_pool
))
692 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
693 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
694 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
695 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
698 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
699 alloc_request_size
, free_request_size
,
700 q
, gfp_mask
, q
->node
);
702 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
703 alloc_request_simple
, free_request_simple
,
704 q
, gfp_mask
, q
->node
);
709 if (rl
!= &q
->root_rl
)
710 WARN_ON_ONCE(!blk_get_queue(q
));
715 void blk_exit_rl(struct request_queue
*q
, struct request_list
*rl
)
718 mempool_destroy(rl
->rq_pool
);
719 if (rl
!= &q
->root_rl
)
724 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
726 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
728 EXPORT_SYMBOL(blk_alloc_queue
);
730 int blk_queue_enter(struct request_queue
*q
, bool nowait
)
735 if (percpu_ref_tryget_live(&q
->q_usage_counter
))
742 * read pair of barrier in blk_freeze_queue_start(),
743 * we need to order reading __PERCPU_REF_DEAD flag of
744 * .q_usage_counter and reading .mq_freeze_depth or
745 * queue dying flag, otherwise the following wait may
746 * never return if the two reads are reordered.
750 ret
= wait_event_interruptible(q
->mq_freeze_wq
,
751 !atomic_read(&q
->mq_freeze_depth
) ||
753 if (blk_queue_dying(q
))
760 void blk_queue_exit(struct request_queue
*q
)
762 percpu_ref_put(&q
->q_usage_counter
);
765 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
767 struct request_queue
*q
=
768 container_of(ref
, struct request_queue
, q_usage_counter
);
770 wake_up_all(&q
->mq_freeze_wq
);
773 static void blk_rq_timed_out_timer(unsigned long data
)
775 struct request_queue
*q
= (struct request_queue
*)data
;
777 kblockd_schedule_work(&q
->timeout_work
);
780 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
782 struct request_queue
*q
;
784 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
785 gfp_mask
| __GFP_ZERO
, node_id
);
789 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
793 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0);
797 q
->backing_dev_info
= bdi_alloc_node(gfp_mask
, node_id
);
798 if (!q
->backing_dev_info
)
801 q
->stats
= blk_alloc_queue_stats();
805 q
->backing_dev_info
->ra_pages
=
806 (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
807 q
->backing_dev_info
->capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
808 q
->backing_dev_info
->name
= "block";
811 setup_timer(&q
->backing_dev_info
->laptop_mode_wb_timer
,
812 laptop_mode_timer_fn
, (unsigned long) q
);
813 setup_timer(&q
->timeout
, blk_rq_timed_out_timer
, (unsigned long) q
);
814 INIT_LIST_HEAD(&q
->queue_head
);
815 INIT_LIST_HEAD(&q
->timeout_list
);
816 INIT_LIST_HEAD(&q
->icq_list
);
817 #ifdef CONFIG_BLK_CGROUP
818 INIT_LIST_HEAD(&q
->blkg_list
);
820 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
822 kobject_init(&q
->kobj
, &blk_queue_ktype
);
824 mutex_init(&q
->sysfs_lock
);
825 spin_lock_init(&q
->__queue_lock
);
828 * By default initialize queue_lock to internal lock and driver can
829 * override it later if need be.
831 q
->queue_lock
= &q
->__queue_lock
;
834 * A queue starts its life with bypass turned on to avoid
835 * unnecessary bypass on/off overhead and nasty surprises during
836 * init. The initial bypass will be finished when the queue is
837 * registered by blk_register_queue().
840 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
842 init_waitqueue_head(&q
->mq_freeze_wq
);
845 * Init percpu_ref in atomic mode so that it's faster to shutdown.
846 * See blk_register_queue() for details.
848 if (percpu_ref_init(&q
->q_usage_counter
,
849 blk_queue_usage_counter_release
,
850 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
853 if (blkcg_init_queue(q
))
859 percpu_ref_exit(&q
->q_usage_counter
);
861 blk_free_queue_stats(q
->stats
);
863 bdi_put(q
->backing_dev_info
);
865 bioset_free(q
->bio_split
);
867 ida_simple_remove(&blk_queue_ida
, q
->id
);
869 kmem_cache_free(blk_requestq_cachep
, q
);
872 EXPORT_SYMBOL(blk_alloc_queue_node
);
875 * blk_init_queue - prepare a request queue for use with a block device
876 * @rfn: The function to be called to process requests that have been
877 * placed on the queue.
878 * @lock: Request queue spin lock
881 * If a block device wishes to use the standard request handling procedures,
882 * which sorts requests and coalesces adjacent requests, then it must
883 * call blk_init_queue(). The function @rfn will be called when there
884 * are requests on the queue that need to be processed. If the device
885 * supports plugging, then @rfn may not be called immediately when requests
886 * are available on the queue, but may be called at some time later instead.
887 * Plugged queues are generally unplugged when a buffer belonging to one
888 * of the requests on the queue is needed, or due to memory pressure.
890 * @rfn is not required, or even expected, to remove all requests off the
891 * queue, but only as many as it can handle at a time. If it does leave
892 * requests on the queue, it is responsible for arranging that the requests
893 * get dealt with eventually.
895 * The queue spin lock must be held while manipulating the requests on the
896 * request queue; this lock will be taken also from interrupt context, so irq
897 * disabling is needed for it.
899 * Function returns a pointer to the initialized request queue, or %NULL if
903 * blk_init_queue() must be paired with a blk_cleanup_queue() call
904 * when the block device is deactivated (such as at module unload).
907 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
909 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
911 EXPORT_SYMBOL(blk_init_queue
);
913 struct request_queue
*
914 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
916 struct request_queue
*q
;
918 q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
924 q
->queue_lock
= lock
;
925 if (blk_init_allocated_queue(q
) < 0) {
926 blk_cleanup_queue(q
);
932 EXPORT_SYMBOL(blk_init_queue_node
);
934 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
937 int blk_init_allocated_queue(struct request_queue
*q
)
939 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, q
->cmd_size
);
943 if (q
->init_rq_fn
&& q
->init_rq_fn(q
, q
->fq
->flush_rq
, GFP_KERNEL
))
944 goto out_free_flush_queue
;
946 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
947 goto out_exit_flush_rq
;
949 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
950 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
953 * This also sets hw/phys segments, boundary and size
955 blk_queue_make_request(q
, blk_queue_bio
);
957 q
->sg_reserved_size
= INT_MAX
;
959 /* Protect q->elevator from elevator_change */
960 mutex_lock(&q
->sysfs_lock
);
963 if (elevator_init(q
, NULL
)) {
964 mutex_unlock(&q
->sysfs_lock
);
965 goto out_exit_flush_rq
;
968 mutex_unlock(&q
->sysfs_lock
);
973 q
->exit_rq_fn(q
, q
->fq
->flush_rq
);
974 out_free_flush_queue
:
975 blk_free_flush_queue(q
->fq
);
978 EXPORT_SYMBOL(blk_init_allocated_queue
);
980 bool blk_get_queue(struct request_queue
*q
)
982 if (likely(!blk_queue_dying(q
))) {
989 EXPORT_SYMBOL(blk_get_queue
);
991 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
993 if (rq
->rq_flags
& RQF_ELVPRIV
) {
994 elv_put_request(rl
->q
, rq
);
996 put_io_context(rq
->elv
.icq
->ioc
);
999 mempool_free(rq
, rl
->rq_pool
);
1003 * ioc_batching returns true if the ioc is a valid batching request and
1004 * should be given priority access to a request.
1006 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
1012 * Make sure the process is able to allocate at least 1 request
1013 * even if the batch times out, otherwise we could theoretically
1016 return ioc
->nr_batch_requests
== q
->nr_batching
||
1017 (ioc
->nr_batch_requests
> 0
1018 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
1022 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1023 * will cause the process to be a "batcher" on all queues in the system. This
1024 * is the behaviour we want though - once it gets a wakeup it should be given
1027 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
1029 if (!ioc
|| ioc_batching(q
, ioc
))
1032 ioc
->nr_batch_requests
= q
->nr_batching
;
1033 ioc
->last_waited
= jiffies
;
1036 static void __freed_request(struct request_list
*rl
, int sync
)
1038 struct request_queue
*q
= rl
->q
;
1040 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
1041 blk_clear_congested(rl
, sync
);
1043 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
1044 if (waitqueue_active(&rl
->wait
[sync
]))
1045 wake_up(&rl
->wait
[sync
]);
1047 blk_clear_rl_full(rl
, sync
);
1052 * A request has just been released. Account for it, update the full and
1053 * congestion status, wake up any waiters. Called under q->queue_lock.
1055 static void freed_request(struct request_list
*rl
, bool sync
,
1056 req_flags_t rq_flags
)
1058 struct request_queue
*q
= rl
->q
;
1062 if (rq_flags
& RQF_ELVPRIV
)
1063 q
->nr_rqs_elvpriv
--;
1065 __freed_request(rl
, sync
);
1067 if (unlikely(rl
->starved
[sync
^ 1]))
1068 __freed_request(rl
, sync
^ 1);
1071 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
1073 struct request_list
*rl
;
1074 int on_thresh
, off_thresh
;
1076 spin_lock_irq(q
->queue_lock
);
1077 q
->nr_requests
= nr
;
1078 blk_queue_congestion_threshold(q
);
1079 on_thresh
= queue_congestion_on_threshold(q
);
1080 off_thresh
= queue_congestion_off_threshold(q
);
1082 blk_queue_for_each_rl(rl
, q
) {
1083 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
1084 blk_set_congested(rl
, BLK_RW_SYNC
);
1085 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
1086 blk_clear_congested(rl
, BLK_RW_SYNC
);
1088 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
1089 blk_set_congested(rl
, BLK_RW_ASYNC
);
1090 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
1091 blk_clear_congested(rl
, BLK_RW_ASYNC
);
1093 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
1094 blk_set_rl_full(rl
, BLK_RW_SYNC
);
1096 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1097 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1100 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1101 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1103 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1104 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1108 spin_unlock_irq(q
->queue_lock
);
1113 * __get_request - get a free request
1114 * @rl: request list to allocate from
1115 * @op: operation and flags
1116 * @bio: bio to allocate request for (can be %NULL)
1117 * @gfp_mask: allocation mask
1119 * Get a free request from @q. This function may fail under memory
1120 * pressure or if @q is dead.
1122 * Must be called with @q->queue_lock held and,
1123 * Returns ERR_PTR on failure, with @q->queue_lock held.
1124 * Returns request pointer on success, with @q->queue_lock *not held*.
1126 static struct request
*__get_request(struct request_list
*rl
, unsigned int op
,
1127 struct bio
*bio
, gfp_t gfp_mask
)
1129 struct request_queue
*q
= rl
->q
;
1131 struct elevator_type
*et
= q
->elevator
->type
;
1132 struct io_context
*ioc
= rq_ioc(bio
);
1133 struct io_cq
*icq
= NULL
;
1134 const bool is_sync
= op_is_sync(op
);
1136 req_flags_t rq_flags
= RQF_ALLOCED
;
1138 if (unlikely(blk_queue_dying(q
)))
1139 return ERR_PTR(-ENODEV
);
1141 may_queue
= elv_may_queue(q
, op
);
1142 if (may_queue
== ELV_MQUEUE_NO
)
1145 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1146 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1148 * The queue will fill after this allocation, so set
1149 * it as full, and mark this process as "batching".
1150 * This process will be allowed to complete a batch of
1151 * requests, others will be blocked.
1153 if (!blk_rl_full(rl
, is_sync
)) {
1154 ioc_set_batching(q
, ioc
);
1155 blk_set_rl_full(rl
, is_sync
);
1157 if (may_queue
!= ELV_MQUEUE_MUST
1158 && !ioc_batching(q
, ioc
)) {
1160 * The queue is full and the allocating
1161 * process is not a "batcher", and not
1162 * exempted by the IO scheduler
1164 return ERR_PTR(-ENOMEM
);
1168 blk_set_congested(rl
, is_sync
);
1172 * Only allow batching queuers to allocate up to 50% over the defined
1173 * limit of requests, otherwise we could have thousands of requests
1174 * allocated with any setting of ->nr_requests
1176 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1177 return ERR_PTR(-ENOMEM
);
1179 q
->nr_rqs
[is_sync
]++;
1180 rl
->count
[is_sync
]++;
1181 rl
->starved
[is_sync
] = 0;
1184 * Decide whether the new request will be managed by elevator. If
1185 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1186 * prevent the current elevator from being destroyed until the new
1187 * request is freed. This guarantees icq's won't be destroyed and
1188 * makes creating new ones safe.
1190 * Flush requests do not use the elevator so skip initialization.
1191 * This allows a request to share the flush and elevator data.
1193 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1194 * it will be created after releasing queue_lock.
1196 if (!op_is_flush(op
) && !blk_queue_bypass(q
)) {
1197 rq_flags
|= RQF_ELVPRIV
;
1198 q
->nr_rqs_elvpriv
++;
1199 if (et
->icq_cache
&& ioc
)
1200 icq
= ioc_lookup_icq(ioc
, q
);
1203 if (blk_queue_io_stat(q
))
1204 rq_flags
|= RQF_IO_STAT
;
1205 spin_unlock_irq(q
->queue_lock
);
1207 /* allocate and init request */
1208 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1213 blk_rq_set_rl(rq
, rl
);
1215 rq
->rq_flags
= rq_flags
;
1218 if (rq_flags
& RQF_ELVPRIV
) {
1219 if (unlikely(et
->icq_cache
&& !icq
)) {
1221 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1227 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1230 /* @rq->elv.icq holds io_context until @rq is freed */
1232 get_io_context(icq
->ioc
);
1236 * ioc may be NULL here, and ioc_batching will be false. That's
1237 * OK, if the queue is under the request limit then requests need
1238 * not count toward the nr_batch_requests limit. There will always
1239 * be some limit enforced by BLK_BATCH_TIME.
1241 if (ioc_batching(q
, ioc
))
1242 ioc
->nr_batch_requests
--;
1244 trace_block_getrq(q
, bio
, op
);
1249 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1250 * and may fail indefinitely under memory pressure and thus
1251 * shouldn't stall IO. Treat this request as !elvpriv. This will
1252 * disturb iosched and blkcg but weird is bettern than dead.
1254 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1255 __func__
, dev_name(q
->backing_dev_info
->dev
));
1257 rq
->rq_flags
&= ~RQF_ELVPRIV
;
1260 spin_lock_irq(q
->queue_lock
);
1261 q
->nr_rqs_elvpriv
--;
1262 spin_unlock_irq(q
->queue_lock
);
1267 * Allocation failed presumably due to memory. Undo anything we
1268 * might have messed up.
1270 * Allocating task should really be put onto the front of the wait
1271 * queue, but this is pretty rare.
1273 spin_lock_irq(q
->queue_lock
);
1274 freed_request(rl
, is_sync
, rq_flags
);
1277 * in the very unlikely event that allocation failed and no
1278 * requests for this direction was pending, mark us starved so that
1279 * freeing of a request in the other direction will notice
1280 * us. another possible fix would be to split the rq mempool into
1284 if (unlikely(rl
->count
[is_sync
] == 0))
1285 rl
->starved
[is_sync
] = 1;
1286 return ERR_PTR(-ENOMEM
);
1290 * get_request - get a free request
1291 * @q: request_queue to allocate request from
1292 * @op: operation and flags
1293 * @bio: bio to allocate request for (can be %NULL)
1294 * @gfp_mask: allocation mask
1296 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1297 * this function keeps retrying under memory pressure and fails iff @q is dead.
1299 * Must be called with @q->queue_lock held and,
1300 * Returns ERR_PTR on failure, with @q->queue_lock held.
1301 * Returns request pointer on success, with @q->queue_lock *not held*.
1303 static struct request
*get_request(struct request_queue
*q
, unsigned int op
,
1304 struct bio
*bio
, gfp_t gfp_mask
)
1306 const bool is_sync
= op_is_sync(op
);
1308 struct request_list
*rl
;
1311 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1313 rq
= __get_request(rl
, op
, bio
, gfp_mask
);
1317 if (!gfpflags_allow_blocking(gfp_mask
) || unlikely(blk_queue_dying(q
))) {
1322 /* wait on @rl and retry */
1323 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1324 TASK_UNINTERRUPTIBLE
);
1326 trace_block_sleeprq(q
, bio
, op
);
1328 spin_unlock_irq(q
->queue_lock
);
1332 * After sleeping, we become a "batching" process and will be able
1333 * to allocate at least one request, and up to a big batch of them
1334 * for a small period time. See ioc_batching, ioc_set_batching
1336 ioc_set_batching(q
, current
->io_context
);
1338 spin_lock_irq(q
->queue_lock
);
1339 finish_wait(&rl
->wait
[is_sync
], &wait
);
1344 static struct request
*blk_old_get_request(struct request_queue
*q
, int rw
,
1349 /* create ioc upfront */
1350 create_io_context(gfp_mask
, q
->node
);
1352 spin_lock_irq(q
->queue_lock
);
1353 rq
= get_request(q
, rw
, NULL
, gfp_mask
);
1355 spin_unlock_irq(q
->queue_lock
);
1359 /* q->queue_lock is unlocked at this point */
1361 rq
->__sector
= (sector_t
) -1;
1362 rq
->bio
= rq
->biotail
= NULL
;
1366 struct request
*blk_get_request(struct request_queue
*q
, int rw
, gfp_t gfp_mask
)
1369 return blk_mq_alloc_request(q
, rw
,
1370 (gfp_mask
& __GFP_DIRECT_RECLAIM
) ?
1371 0 : BLK_MQ_REQ_NOWAIT
);
1373 return blk_old_get_request(q
, rw
, gfp_mask
);
1375 EXPORT_SYMBOL(blk_get_request
);
1378 * blk_requeue_request - put a request back on queue
1379 * @q: request queue where request should be inserted
1380 * @rq: request to be inserted
1383 * Drivers often keep queueing requests until the hardware cannot accept
1384 * more, when that condition happens we need to put the request back
1385 * on the queue. Must be called with queue lock held.
1387 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1389 blk_delete_timer(rq
);
1390 blk_clear_rq_complete(rq
);
1391 trace_block_rq_requeue(q
, rq
);
1392 wbt_requeue(q
->rq_wb
, &rq
->issue_stat
);
1394 if (rq
->rq_flags
& RQF_QUEUED
)
1395 blk_queue_end_tag(q
, rq
);
1397 BUG_ON(blk_queued_rq(rq
));
1399 elv_requeue_request(q
, rq
);
1401 EXPORT_SYMBOL(blk_requeue_request
);
1403 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1406 blk_account_io_start(rq
, true);
1407 __elv_add_request(q
, rq
, where
);
1410 static void part_round_stats_single(int cpu
, struct hd_struct
*part
,
1415 if (now
== part
->stamp
)
1418 inflight
= part_in_flight(part
);
1420 __part_stat_add(cpu
, part
, time_in_queue
,
1421 inflight
* (now
- part
->stamp
));
1422 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1428 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1429 * @cpu: cpu number for stats access
1430 * @part: target partition
1432 * The average IO queue length and utilisation statistics are maintained
1433 * by observing the current state of the queue length and the amount of
1434 * time it has been in this state for.
1436 * Normally, that accounting is done on IO completion, but that can result
1437 * in more than a second's worth of IO being accounted for within any one
1438 * second, leading to >100% utilisation. To deal with that, we call this
1439 * function to do a round-off before returning the results when reading
1440 * /proc/diskstats. This accounts immediately for all queue usage up to
1441 * the current jiffies and restarts the counters again.
1443 void part_round_stats(int cpu
, struct hd_struct
*part
)
1445 unsigned long now
= jiffies
;
1448 part_round_stats_single(cpu
, &part_to_disk(part
)->part0
, now
);
1449 part_round_stats_single(cpu
, part
, now
);
1451 EXPORT_SYMBOL_GPL(part_round_stats
);
1454 static void blk_pm_put_request(struct request
*rq
)
1456 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1457 pm_runtime_mark_last_busy(rq
->q
->dev
);
1460 static inline void blk_pm_put_request(struct request
*rq
) {}
1464 * queue lock must be held
1466 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1468 req_flags_t rq_flags
= req
->rq_flags
;
1474 blk_mq_free_request(req
);
1478 blk_pm_put_request(req
);
1480 elv_completed_request(q
, req
);
1482 /* this is a bio leak */
1483 WARN_ON(req
->bio
!= NULL
);
1485 wbt_done(q
->rq_wb
, &req
->issue_stat
);
1488 * Request may not have originated from ll_rw_blk. if not,
1489 * it didn't come out of our reserved rq pools
1491 if (rq_flags
& RQF_ALLOCED
) {
1492 struct request_list
*rl
= blk_rq_rl(req
);
1493 bool sync
= op_is_sync(req
->cmd_flags
);
1495 BUG_ON(!list_empty(&req
->queuelist
));
1496 BUG_ON(ELV_ON_HASH(req
));
1498 blk_free_request(rl
, req
);
1499 freed_request(rl
, sync
, rq_flags
);
1503 EXPORT_SYMBOL_GPL(__blk_put_request
);
1505 void blk_put_request(struct request
*req
)
1507 struct request_queue
*q
= req
->q
;
1510 blk_mq_free_request(req
);
1512 unsigned long flags
;
1514 spin_lock_irqsave(q
->queue_lock
, flags
);
1515 __blk_put_request(q
, req
);
1516 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1519 EXPORT_SYMBOL(blk_put_request
);
1521 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1524 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1526 if (!ll_back_merge_fn(q
, req
, bio
))
1529 trace_block_bio_backmerge(q
, req
, bio
);
1531 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1532 blk_rq_set_mixed_merge(req
);
1534 req
->biotail
->bi_next
= bio
;
1536 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1537 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1539 blk_account_io_start(req
, false);
1543 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1546 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1548 if (!ll_front_merge_fn(q
, req
, bio
))
1551 trace_block_bio_frontmerge(q
, req
, bio
);
1553 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1554 blk_rq_set_mixed_merge(req
);
1556 bio
->bi_next
= req
->bio
;
1559 req
->__sector
= bio
->bi_iter
.bi_sector
;
1560 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1561 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1563 blk_account_io_start(req
, false);
1567 bool bio_attempt_discard_merge(struct request_queue
*q
, struct request
*req
,
1570 unsigned short segments
= blk_rq_nr_discard_segments(req
);
1572 if (segments
>= queue_max_discard_segments(q
))
1574 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
1575 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
1578 req
->biotail
->bi_next
= bio
;
1580 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1581 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1582 req
->nr_phys_segments
= segments
+ 1;
1584 blk_account_io_start(req
, false);
1587 req_set_nomerge(q
, req
);
1592 * blk_attempt_plug_merge - try to merge with %current's plugged list
1593 * @q: request_queue new bio is being queued at
1594 * @bio: new bio being queued
1595 * @request_count: out parameter for number of traversed plugged requests
1596 * @same_queue_rq: pointer to &struct request that gets filled in when
1597 * another request associated with @q is found on the plug list
1598 * (optional, may be %NULL)
1600 * Determine whether @bio being queued on @q can be merged with a request
1601 * on %current's plugged list. Returns %true if merge was successful,
1604 * Plugging coalesces IOs from the same issuer for the same purpose without
1605 * going through @q->queue_lock. As such it's more of an issuing mechanism
1606 * than scheduling, and the request, while may have elvpriv data, is not
1607 * added on the elevator at this point. In addition, we don't have
1608 * reliable access to the elevator outside queue lock. Only check basic
1609 * merging parameters without querying the elevator.
1611 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1613 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1614 unsigned int *request_count
,
1615 struct request
**same_queue_rq
)
1617 struct blk_plug
*plug
;
1619 struct list_head
*plug_list
;
1621 plug
= current
->plug
;
1627 plug_list
= &plug
->mq_list
;
1629 plug_list
= &plug
->list
;
1631 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1632 bool merged
= false;
1637 * Only blk-mq multiple hardware queues case checks the
1638 * rq in the same queue, there should be only one such
1642 *same_queue_rq
= rq
;
1645 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1648 switch (blk_try_merge(rq
, bio
)) {
1649 case ELEVATOR_BACK_MERGE
:
1650 merged
= bio_attempt_back_merge(q
, rq
, bio
);
1652 case ELEVATOR_FRONT_MERGE
:
1653 merged
= bio_attempt_front_merge(q
, rq
, bio
);
1655 case ELEVATOR_DISCARD_MERGE
:
1656 merged
= bio_attempt_discard_merge(q
, rq
, bio
);
1669 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1671 struct blk_plug
*plug
;
1673 struct list_head
*plug_list
;
1674 unsigned int ret
= 0;
1676 plug
= current
->plug
;
1681 plug_list
= &plug
->mq_list
;
1683 plug_list
= &plug
->list
;
1685 list_for_each_entry(rq
, plug_list
, queuelist
) {
1693 void blk_init_request_from_bio(struct request
*req
, struct bio
*bio
)
1695 struct io_context
*ioc
= rq_ioc(bio
);
1697 if (bio
->bi_opf
& REQ_RAHEAD
)
1698 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1700 req
->__sector
= bio
->bi_iter
.bi_sector
;
1701 if (ioprio_valid(bio_prio(bio
)))
1702 req
->ioprio
= bio_prio(bio
);
1704 req
->ioprio
= ioc
->ioprio
;
1706 req
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE
, 0);
1707 blk_rq_bio_prep(req
->q
, req
, bio
);
1709 EXPORT_SYMBOL_GPL(blk_init_request_from_bio
);
1711 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1713 struct blk_plug
*plug
;
1714 int where
= ELEVATOR_INSERT_SORT
;
1715 struct request
*req
, *free
;
1716 unsigned int request_count
= 0;
1717 unsigned int wb_acct
;
1720 * low level driver can indicate that it wants pages above a
1721 * certain limit bounced to low memory (ie for highmem, or even
1722 * ISA dma in theory)
1724 blk_queue_bounce(q
, &bio
);
1726 blk_queue_split(q
, &bio
);
1728 if (bio_integrity_enabled(bio
) && bio_integrity_prep(bio
)) {
1729 bio
->bi_status
= BLK_STS_IOERR
;
1731 return BLK_QC_T_NONE
;
1734 if (op_is_flush(bio
->bi_opf
)) {
1735 spin_lock_irq(q
->queue_lock
);
1736 where
= ELEVATOR_INSERT_FLUSH
;
1741 * Check if we can merge with the plugged list before grabbing
1744 if (!blk_queue_nomerges(q
)) {
1745 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1746 return BLK_QC_T_NONE
;
1748 request_count
= blk_plug_queued_count(q
);
1750 spin_lock_irq(q
->queue_lock
);
1752 switch (elv_merge(q
, &req
, bio
)) {
1753 case ELEVATOR_BACK_MERGE
:
1754 if (!bio_attempt_back_merge(q
, req
, bio
))
1756 elv_bio_merged(q
, req
, bio
);
1757 free
= attempt_back_merge(q
, req
);
1759 __blk_put_request(q
, free
);
1761 elv_merged_request(q
, req
, ELEVATOR_BACK_MERGE
);
1763 case ELEVATOR_FRONT_MERGE
:
1764 if (!bio_attempt_front_merge(q
, req
, bio
))
1766 elv_bio_merged(q
, req
, bio
);
1767 free
= attempt_front_merge(q
, req
);
1769 __blk_put_request(q
, free
);
1771 elv_merged_request(q
, req
, ELEVATOR_FRONT_MERGE
);
1778 wb_acct
= wbt_wait(q
->rq_wb
, bio
, q
->queue_lock
);
1781 * Grab a free request. This is might sleep but can not fail.
1782 * Returns with the queue unlocked.
1784 req
= get_request(q
, bio
->bi_opf
, bio
, GFP_NOIO
);
1786 __wbt_done(q
->rq_wb
, wb_acct
);
1787 if (PTR_ERR(req
) == -ENOMEM
)
1788 bio
->bi_status
= BLK_STS_RESOURCE
;
1790 bio
->bi_status
= BLK_STS_IOERR
;
1795 wbt_track(&req
->issue_stat
, wb_acct
);
1798 * After dropping the lock and possibly sleeping here, our request
1799 * may now be mergeable after it had proven unmergeable (above).
1800 * We don't worry about that case for efficiency. It won't happen
1801 * often, and the elevators are able to handle it.
1803 blk_init_request_from_bio(req
, bio
);
1805 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1806 req
->cpu
= raw_smp_processor_id();
1808 plug
= current
->plug
;
1811 * If this is the first request added after a plug, fire
1814 * @request_count may become stale because of schedule
1815 * out, so check plug list again.
1817 if (!request_count
|| list_empty(&plug
->list
))
1818 trace_block_plug(q
);
1820 struct request
*last
= list_entry_rq(plug
->list
.prev
);
1821 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
1822 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
1823 blk_flush_plug_list(plug
, false);
1824 trace_block_plug(q
);
1827 list_add_tail(&req
->queuelist
, &plug
->list
);
1828 blk_account_io_start(req
, true);
1830 spin_lock_irq(q
->queue_lock
);
1831 add_acct_request(q
, req
, where
);
1834 spin_unlock_irq(q
->queue_lock
);
1837 return BLK_QC_T_NONE
;
1841 * If bio->bi_dev is a partition, remap the location
1843 static inline void blk_partition_remap(struct bio
*bio
)
1845 struct block_device
*bdev
= bio
->bi_bdev
;
1848 * Zone reset does not include bi_size so bio_sectors() is always 0.
1849 * Include a test for the reset op code and perform the remap if needed.
1851 if (bdev
!= bdev
->bd_contains
&&
1852 (bio_sectors(bio
) || bio_op(bio
) == REQ_OP_ZONE_RESET
)) {
1853 struct hd_struct
*p
= bdev
->bd_part
;
1855 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
1856 bio
->bi_bdev
= bdev
->bd_contains
;
1858 trace_block_bio_remap(bdev_get_queue(bio
->bi_bdev
), bio
,
1860 bio
->bi_iter
.bi_sector
- p
->start_sect
);
1864 static void handle_bad_sector(struct bio
*bio
)
1866 char b
[BDEVNAME_SIZE
];
1868 printk(KERN_INFO
"attempt to access beyond end of device\n");
1869 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
1870 bdevname(bio
->bi_bdev
, b
),
1872 (unsigned long long)bio_end_sector(bio
),
1873 (long long)(i_size_read(bio
->bi_bdev
->bd_inode
) >> 9));
1876 #ifdef CONFIG_FAIL_MAKE_REQUEST
1878 static DECLARE_FAULT_ATTR(fail_make_request
);
1880 static int __init
setup_fail_make_request(char *str
)
1882 return setup_fault_attr(&fail_make_request
, str
);
1884 __setup("fail_make_request=", setup_fail_make_request
);
1886 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
1888 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
1891 static int __init
fail_make_request_debugfs(void)
1893 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
1894 NULL
, &fail_make_request
);
1896 return PTR_ERR_OR_ZERO(dir
);
1899 late_initcall(fail_make_request_debugfs
);
1901 #else /* CONFIG_FAIL_MAKE_REQUEST */
1903 static inline bool should_fail_request(struct hd_struct
*part
,
1909 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1912 * Check whether this bio extends beyond the end of the device.
1914 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
1921 /* Test device or partition size, when known. */
1922 maxsector
= i_size_read(bio
->bi_bdev
->bd_inode
) >> 9;
1924 sector_t sector
= bio
->bi_iter
.bi_sector
;
1926 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
1928 * This may well happen - the kernel calls bread()
1929 * without checking the size of the device, e.g., when
1930 * mounting a device.
1932 handle_bad_sector(bio
);
1940 static noinline_for_stack
bool
1941 generic_make_request_checks(struct bio
*bio
)
1943 struct request_queue
*q
;
1944 int nr_sectors
= bio_sectors(bio
);
1945 blk_status_t status
= BLK_STS_IOERR
;
1946 char b
[BDEVNAME_SIZE
];
1947 struct hd_struct
*part
;
1951 if (bio_check_eod(bio
, nr_sectors
))
1954 q
= bdev_get_queue(bio
->bi_bdev
);
1957 "generic_make_request: Trying to access "
1958 "nonexistent block-device %s (%Lu)\n",
1959 bdevname(bio
->bi_bdev
, b
),
1960 (long long) bio
->bi_iter
.bi_sector
);
1964 part
= bio
->bi_bdev
->bd_part
;
1965 if (should_fail_request(part
, bio
->bi_iter
.bi_size
) ||
1966 should_fail_request(&part_to_disk(part
)->part0
,
1967 bio
->bi_iter
.bi_size
))
1971 * If this device has partitions, remap block n
1972 * of partition p to block n+start(p) of the disk.
1974 blk_partition_remap(bio
);
1976 if (bio_check_eod(bio
, nr_sectors
))
1980 * Filter flush bio's early so that make_request based
1981 * drivers without flush support don't have to worry
1984 if (op_is_flush(bio
->bi_opf
) &&
1985 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
1986 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
1988 status
= BLK_STS_OK
;
1993 switch (bio_op(bio
)) {
1994 case REQ_OP_DISCARD
:
1995 if (!blk_queue_discard(q
))
1998 case REQ_OP_SECURE_ERASE
:
1999 if (!blk_queue_secure_erase(q
))
2002 case REQ_OP_WRITE_SAME
:
2003 if (!bdev_write_same(bio
->bi_bdev
))
2006 case REQ_OP_ZONE_REPORT
:
2007 case REQ_OP_ZONE_RESET
:
2008 if (!bdev_is_zoned(bio
->bi_bdev
))
2011 case REQ_OP_WRITE_ZEROES
:
2012 if (!bdev_write_zeroes_sectors(bio
->bi_bdev
))
2020 * Various block parts want %current->io_context and lazy ioc
2021 * allocation ends up trading a lot of pain for a small amount of
2022 * memory. Just allocate it upfront. This may fail and block
2023 * layer knows how to live with it.
2025 create_io_context(GFP_ATOMIC
, q
->node
);
2027 if (!blkcg_bio_issue_check(q
, bio
))
2030 if (!bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
2031 trace_block_bio_queue(q
, bio
);
2032 /* Now that enqueuing has been traced, we need to trace
2033 * completion as well.
2035 bio_set_flag(bio
, BIO_TRACE_COMPLETION
);
2040 status
= BLK_STS_NOTSUPP
;
2042 bio
->bi_status
= status
;
2048 * generic_make_request - hand a buffer to its device driver for I/O
2049 * @bio: The bio describing the location in memory and on the device.
2051 * generic_make_request() is used to make I/O requests of block
2052 * devices. It is passed a &struct bio, which describes the I/O that needs
2055 * generic_make_request() does not return any status. The
2056 * success/failure status of the request, along with notification of
2057 * completion, is delivered asynchronously through the bio->bi_end_io
2058 * function described (one day) else where.
2060 * The caller of generic_make_request must make sure that bi_io_vec
2061 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2062 * set to describe the device address, and the
2063 * bi_end_io and optionally bi_private are set to describe how
2064 * completion notification should be signaled.
2066 * generic_make_request and the drivers it calls may use bi_next if this
2067 * bio happens to be merged with someone else, and may resubmit the bio to
2068 * a lower device by calling into generic_make_request recursively, which
2069 * means the bio should NOT be touched after the call to ->make_request_fn.
2071 blk_qc_t
generic_make_request(struct bio
*bio
)
2074 * bio_list_on_stack[0] contains bios submitted by the current
2076 * bio_list_on_stack[1] contains bios that were submitted before
2077 * the current make_request_fn, but that haven't been processed
2080 struct bio_list bio_list_on_stack
[2];
2081 blk_qc_t ret
= BLK_QC_T_NONE
;
2083 if (!generic_make_request_checks(bio
))
2087 * We only want one ->make_request_fn to be active at a time, else
2088 * stack usage with stacked devices could be a problem. So use
2089 * current->bio_list to keep a list of requests submited by a
2090 * make_request_fn function. current->bio_list is also used as a
2091 * flag to say if generic_make_request is currently active in this
2092 * task or not. If it is NULL, then no make_request is active. If
2093 * it is non-NULL, then a make_request is active, and new requests
2094 * should be added at the tail
2096 if (current
->bio_list
) {
2097 bio_list_add(¤t
->bio_list
[0], bio
);
2101 /* following loop may be a bit non-obvious, and so deserves some
2103 * Before entering the loop, bio->bi_next is NULL (as all callers
2104 * ensure that) so we have a list with a single bio.
2105 * We pretend that we have just taken it off a longer list, so
2106 * we assign bio_list to a pointer to the bio_list_on_stack,
2107 * thus initialising the bio_list of new bios to be
2108 * added. ->make_request() may indeed add some more bios
2109 * through a recursive call to generic_make_request. If it
2110 * did, we find a non-NULL value in bio_list and re-enter the loop
2111 * from the top. In this case we really did just take the bio
2112 * of the top of the list (no pretending) and so remove it from
2113 * bio_list, and call into ->make_request() again.
2115 BUG_ON(bio
->bi_next
);
2116 bio_list_init(&bio_list_on_stack
[0]);
2117 current
->bio_list
= bio_list_on_stack
;
2119 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
2121 if (likely(blk_queue_enter(q
, false) == 0)) {
2122 struct bio_list lower
, same
;
2124 /* Create a fresh bio_list for all subordinate requests */
2125 bio_list_on_stack
[1] = bio_list_on_stack
[0];
2126 bio_list_init(&bio_list_on_stack
[0]);
2127 ret
= q
->make_request_fn(q
, bio
);
2131 /* sort new bios into those for a lower level
2132 * and those for the same level
2134 bio_list_init(&lower
);
2135 bio_list_init(&same
);
2136 while ((bio
= bio_list_pop(&bio_list_on_stack
[0])) != NULL
)
2137 if (q
== bdev_get_queue(bio
->bi_bdev
))
2138 bio_list_add(&same
, bio
);
2140 bio_list_add(&lower
, bio
);
2141 /* now assemble so we handle the lowest level first */
2142 bio_list_merge(&bio_list_on_stack
[0], &lower
);
2143 bio_list_merge(&bio_list_on_stack
[0], &same
);
2144 bio_list_merge(&bio_list_on_stack
[0], &bio_list_on_stack
[1]);
2148 bio
= bio_list_pop(&bio_list_on_stack
[0]);
2150 current
->bio_list
= NULL
; /* deactivate */
2155 EXPORT_SYMBOL(generic_make_request
);
2158 * submit_bio - submit a bio to the block device layer for I/O
2159 * @bio: The &struct bio which describes the I/O
2161 * submit_bio() is very similar in purpose to generic_make_request(), and
2162 * uses that function to do most of the work. Both are fairly rough
2163 * interfaces; @bio must be presetup and ready for I/O.
2166 blk_qc_t
submit_bio(struct bio
*bio
)
2169 * If it's a regular read/write or a barrier with data attached,
2170 * go through the normal accounting stuff before submission.
2172 if (bio_has_data(bio
)) {
2175 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2176 count
= bdev_logical_block_size(bio
->bi_bdev
) >> 9;
2178 count
= bio_sectors(bio
);
2180 if (op_is_write(bio_op(bio
))) {
2181 count_vm_events(PGPGOUT
, count
);
2183 task_io_account_read(bio
->bi_iter
.bi_size
);
2184 count_vm_events(PGPGIN
, count
);
2187 if (unlikely(block_dump
)) {
2188 char b
[BDEVNAME_SIZE
];
2189 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2190 current
->comm
, task_pid_nr(current
),
2191 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2192 (unsigned long long)bio
->bi_iter
.bi_sector
,
2193 bdevname(bio
->bi_bdev
, b
),
2198 return generic_make_request(bio
);
2200 EXPORT_SYMBOL(submit_bio
);
2203 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2204 * for new the queue limits
2206 * @rq: the request being checked
2209 * @rq may have been made based on weaker limitations of upper-level queues
2210 * in request stacking drivers, and it may violate the limitation of @q.
2211 * Since the block layer and the underlying device driver trust @rq
2212 * after it is inserted to @q, it should be checked against @q before
2213 * the insertion using this generic function.
2215 * Request stacking drivers like request-based dm may change the queue
2216 * limits when retrying requests on other queues. Those requests need
2217 * to be checked against the new queue limits again during dispatch.
2219 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2222 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2223 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2228 * queue's settings related to segment counting like q->bounce_pfn
2229 * may differ from that of other stacking queues.
2230 * Recalculate it to check the request correctly on this queue's
2233 blk_recalc_rq_segments(rq
);
2234 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2235 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2243 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2244 * @q: the queue to submit the request
2245 * @rq: the request being queued
2247 blk_status_t
blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2249 unsigned long flags
;
2250 int where
= ELEVATOR_INSERT_BACK
;
2252 if (blk_cloned_rq_check_limits(q
, rq
))
2253 return BLK_STS_IOERR
;
2256 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2257 return BLK_STS_IOERR
;
2260 if (blk_queue_io_stat(q
))
2261 blk_account_io_start(rq
, true);
2262 blk_mq_sched_insert_request(rq
, false, true, false, false);
2266 spin_lock_irqsave(q
->queue_lock
, flags
);
2267 if (unlikely(blk_queue_dying(q
))) {
2268 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2269 return BLK_STS_IOERR
;
2273 * Submitting request must be dequeued before calling this function
2274 * because it will be linked to another request_queue
2276 BUG_ON(blk_queued_rq(rq
));
2278 if (op_is_flush(rq
->cmd_flags
))
2279 where
= ELEVATOR_INSERT_FLUSH
;
2281 add_acct_request(q
, rq
, where
);
2282 if (where
== ELEVATOR_INSERT_FLUSH
)
2284 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2288 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2291 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2292 * @rq: request to examine
2295 * A request could be merge of IOs which require different failure
2296 * handling. This function determines the number of bytes which
2297 * can be failed from the beginning of the request without
2298 * crossing into area which need to be retried further.
2301 * The number of bytes to fail.
2304 * queue_lock must be held.
2306 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2308 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2309 unsigned int bytes
= 0;
2312 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2313 return blk_rq_bytes(rq
);
2316 * Currently the only 'mixing' which can happen is between
2317 * different fastfail types. We can safely fail portions
2318 * which have all the failfast bits that the first one has -
2319 * the ones which are at least as eager to fail as the first
2322 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2323 if ((bio
->bi_opf
& ff
) != ff
)
2325 bytes
+= bio
->bi_iter
.bi_size
;
2328 /* this could lead to infinite loop */
2329 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2332 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2334 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2336 if (blk_do_io_stat(req
)) {
2337 const int rw
= rq_data_dir(req
);
2338 struct hd_struct
*part
;
2341 cpu
= part_stat_lock();
2343 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2348 void blk_account_io_done(struct request
*req
)
2351 * Account IO completion. flush_rq isn't accounted as a
2352 * normal IO on queueing nor completion. Accounting the
2353 * containing request is enough.
2355 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2356 unsigned long duration
= jiffies
- req
->start_time
;
2357 const int rw
= rq_data_dir(req
);
2358 struct hd_struct
*part
;
2361 cpu
= part_stat_lock();
2364 part_stat_inc(cpu
, part
, ios
[rw
]);
2365 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2366 part_round_stats(cpu
, part
);
2367 part_dec_in_flight(part
, rw
);
2369 hd_struct_put(part
);
2376 * Don't process normal requests when queue is suspended
2377 * or in the process of suspending/resuming
2379 static struct request
*blk_pm_peek_request(struct request_queue
*q
,
2382 if (q
->dev
&& (q
->rpm_status
== RPM_SUSPENDED
||
2383 (q
->rpm_status
!= RPM_ACTIVE
&& !(rq
->rq_flags
& RQF_PM
))))
2389 static inline struct request
*blk_pm_peek_request(struct request_queue
*q
,
2396 void blk_account_io_start(struct request
*rq
, bool new_io
)
2398 struct hd_struct
*part
;
2399 int rw
= rq_data_dir(rq
);
2402 if (!blk_do_io_stat(rq
))
2405 cpu
= part_stat_lock();
2409 part_stat_inc(cpu
, part
, merges
[rw
]);
2411 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2412 if (!hd_struct_try_get(part
)) {
2414 * The partition is already being removed,
2415 * the request will be accounted on the disk only
2417 * We take a reference on disk->part0 although that
2418 * partition will never be deleted, so we can treat
2419 * it as any other partition.
2421 part
= &rq
->rq_disk
->part0
;
2422 hd_struct_get(part
);
2424 part_round_stats(cpu
, part
);
2425 part_inc_in_flight(part
, rw
);
2433 * blk_peek_request - peek at the top of a request queue
2434 * @q: request queue to peek at
2437 * Return the request at the top of @q. The returned request
2438 * should be started using blk_start_request() before LLD starts
2442 * Pointer to the request at the top of @q if available. Null
2446 * queue_lock must be held.
2448 struct request
*blk_peek_request(struct request_queue
*q
)
2453 while ((rq
= __elv_next_request(q
)) != NULL
) {
2455 rq
= blk_pm_peek_request(q
, rq
);
2459 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2461 * This is the first time the device driver
2462 * sees this request (possibly after
2463 * requeueing). Notify IO scheduler.
2465 if (rq
->rq_flags
& RQF_SORTED
)
2466 elv_activate_rq(q
, rq
);
2469 * just mark as started even if we don't start
2470 * it, a request that has been delayed should
2471 * not be passed by new incoming requests
2473 rq
->rq_flags
|= RQF_STARTED
;
2474 trace_block_rq_issue(q
, rq
);
2477 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2478 q
->end_sector
= rq_end_sector(rq
);
2479 q
->boundary_rq
= NULL
;
2482 if (rq
->rq_flags
& RQF_DONTPREP
)
2485 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2487 * make sure space for the drain appears we
2488 * know we can do this because max_hw_segments
2489 * has been adjusted to be one fewer than the
2492 rq
->nr_phys_segments
++;
2498 ret
= q
->prep_rq_fn(q
, rq
);
2499 if (ret
== BLKPREP_OK
) {
2501 } else if (ret
== BLKPREP_DEFER
) {
2503 * the request may have been (partially) prepped.
2504 * we need to keep this request in the front to
2505 * avoid resource deadlock. RQF_STARTED will
2506 * prevent other fs requests from passing this one.
2508 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2509 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2511 * remove the space for the drain we added
2512 * so that we don't add it again
2514 --rq
->nr_phys_segments
;
2519 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2520 rq
->rq_flags
|= RQF_QUIET
;
2522 * Mark this request as started so we don't trigger
2523 * any debug logic in the end I/O path.
2525 blk_start_request(rq
);
2526 __blk_end_request_all(rq
, ret
== BLKPREP_INVALID
?
2527 BLK_STS_TARGET
: BLK_STS_IOERR
);
2529 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2536 EXPORT_SYMBOL(blk_peek_request
);
2538 void blk_dequeue_request(struct request
*rq
)
2540 struct request_queue
*q
= rq
->q
;
2542 BUG_ON(list_empty(&rq
->queuelist
));
2543 BUG_ON(ELV_ON_HASH(rq
));
2545 list_del_init(&rq
->queuelist
);
2548 * the time frame between a request being removed from the lists
2549 * and to it is freed is accounted as io that is in progress at
2552 if (blk_account_rq(rq
)) {
2553 q
->in_flight
[rq_is_sync(rq
)]++;
2554 set_io_start_time_ns(rq
);
2559 * blk_start_request - start request processing on the driver
2560 * @req: request to dequeue
2563 * Dequeue @req and start timeout timer on it. This hands off the
2564 * request to the driver.
2566 * Block internal functions which don't want to start timer should
2567 * call blk_dequeue_request().
2570 * queue_lock must be held.
2572 void blk_start_request(struct request
*req
)
2574 blk_dequeue_request(req
);
2576 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2577 blk_stat_set_issue(&req
->issue_stat
, blk_rq_sectors(req
));
2578 req
->rq_flags
|= RQF_STATS
;
2579 wbt_issue(req
->q
->rq_wb
, &req
->issue_stat
);
2582 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2585 EXPORT_SYMBOL(blk_start_request
);
2588 * blk_fetch_request - fetch a request from a request queue
2589 * @q: request queue to fetch a request from
2592 * Return the request at the top of @q. The request is started on
2593 * return and LLD can start processing it immediately.
2596 * Pointer to the request at the top of @q if available. Null
2600 * queue_lock must be held.
2602 struct request
*blk_fetch_request(struct request_queue
*q
)
2606 rq
= blk_peek_request(q
);
2608 blk_start_request(rq
);
2611 EXPORT_SYMBOL(blk_fetch_request
);
2614 * blk_update_request - Special helper function for request stacking drivers
2615 * @req: the request being processed
2616 * @error: block status code
2617 * @nr_bytes: number of bytes to complete @req
2620 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2621 * the request structure even if @req doesn't have leftover.
2622 * If @req has leftover, sets it up for the next range of segments.
2624 * This special helper function is only for request stacking drivers
2625 * (e.g. request-based dm) so that they can handle partial completion.
2626 * Actual device drivers should use blk_end_request instead.
2628 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2629 * %false return from this function.
2632 * %false - this request doesn't have any more data
2633 * %true - this request has more data
2635 bool blk_update_request(struct request
*req
, blk_status_t error
,
2636 unsigned int nr_bytes
)
2640 trace_block_rq_complete(req
, blk_status_to_errno(error
), nr_bytes
);
2645 if (unlikely(error
&& !blk_rq_is_passthrough(req
) &&
2646 !(req
->rq_flags
& RQF_QUIET
)))
2647 print_req_error(req
, error
);
2649 blk_account_io_completion(req
, nr_bytes
);
2653 struct bio
*bio
= req
->bio
;
2654 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2656 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2657 req
->bio
= bio
->bi_next
;
2659 /* Completion has already been traced */
2660 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
2661 req_bio_endio(req
, bio
, bio_bytes
, error
);
2663 total_bytes
+= bio_bytes
;
2664 nr_bytes
-= bio_bytes
;
2675 * Reset counters so that the request stacking driver
2676 * can find how many bytes remain in the request
2679 req
->__data_len
= 0;
2683 req
->__data_len
-= total_bytes
;
2685 /* update sector only for requests with clear definition of sector */
2686 if (!blk_rq_is_passthrough(req
))
2687 req
->__sector
+= total_bytes
>> 9;
2689 /* mixed attributes always follow the first bio */
2690 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2691 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2692 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2695 if (!(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
)) {
2697 * If total number of sectors is less than the first segment
2698 * size, something has gone terribly wrong.
2700 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2701 blk_dump_rq_flags(req
, "request botched");
2702 req
->__data_len
= blk_rq_cur_bytes(req
);
2705 /* recalculate the number of segments */
2706 blk_recalc_rq_segments(req
);
2711 EXPORT_SYMBOL_GPL(blk_update_request
);
2713 static bool blk_update_bidi_request(struct request
*rq
, blk_status_t error
,
2714 unsigned int nr_bytes
,
2715 unsigned int bidi_bytes
)
2717 if (blk_update_request(rq
, error
, nr_bytes
))
2720 /* Bidi request must be completed as a whole */
2721 if (unlikely(blk_bidi_rq(rq
)) &&
2722 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
2725 if (blk_queue_add_random(rq
->q
))
2726 add_disk_randomness(rq
->rq_disk
);
2732 * blk_unprep_request - unprepare a request
2735 * This function makes a request ready for complete resubmission (or
2736 * completion). It happens only after all error handling is complete,
2737 * so represents the appropriate moment to deallocate any resources
2738 * that were allocated to the request in the prep_rq_fn. The queue
2739 * lock is held when calling this.
2741 void blk_unprep_request(struct request
*req
)
2743 struct request_queue
*q
= req
->q
;
2745 req
->rq_flags
&= ~RQF_DONTPREP
;
2746 if (q
->unprep_rq_fn
)
2747 q
->unprep_rq_fn(q
, req
);
2749 EXPORT_SYMBOL_GPL(blk_unprep_request
);
2752 * queue lock must be held
2754 void blk_finish_request(struct request
*req
, blk_status_t error
)
2756 struct request_queue
*q
= req
->q
;
2758 if (req
->rq_flags
& RQF_STATS
)
2761 if (req
->rq_flags
& RQF_QUEUED
)
2762 blk_queue_end_tag(q
, req
);
2764 BUG_ON(blk_queued_rq(req
));
2766 if (unlikely(laptop_mode
) && !blk_rq_is_passthrough(req
))
2767 laptop_io_completion(req
->q
->backing_dev_info
);
2769 blk_delete_timer(req
);
2771 if (req
->rq_flags
& RQF_DONTPREP
)
2772 blk_unprep_request(req
);
2774 blk_account_io_done(req
);
2777 wbt_done(req
->q
->rq_wb
, &req
->issue_stat
);
2778 req
->end_io(req
, error
);
2780 if (blk_bidi_rq(req
))
2781 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
2783 __blk_put_request(q
, req
);
2786 EXPORT_SYMBOL(blk_finish_request
);
2789 * blk_end_bidi_request - Complete a bidi request
2790 * @rq: the request to complete
2791 * @error: block status code
2792 * @nr_bytes: number of bytes to complete @rq
2793 * @bidi_bytes: number of bytes to complete @rq->next_rq
2796 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2797 * Drivers that supports bidi can safely call this member for any
2798 * type of request, bidi or uni. In the later case @bidi_bytes is
2802 * %false - we are done with this request
2803 * %true - still buffers pending for this request
2805 static bool blk_end_bidi_request(struct request
*rq
, blk_status_t error
,
2806 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2808 struct request_queue
*q
= rq
->q
;
2809 unsigned long flags
;
2811 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2814 spin_lock_irqsave(q
->queue_lock
, flags
);
2815 blk_finish_request(rq
, error
);
2816 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2822 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2823 * @rq: the request to complete
2824 * @error: block status code
2825 * @nr_bytes: number of bytes to complete @rq
2826 * @bidi_bytes: number of bytes to complete @rq->next_rq
2829 * Identical to blk_end_bidi_request() except that queue lock is
2830 * assumed to be locked on entry and remains so on return.
2833 * %false - we are done with this request
2834 * %true - still buffers pending for this request
2836 static bool __blk_end_bidi_request(struct request
*rq
, blk_status_t error
,
2837 unsigned int nr_bytes
, unsigned int bidi_bytes
)
2839 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
2842 blk_finish_request(rq
, error
);
2848 * blk_end_request - Helper function for drivers to complete the request.
2849 * @rq: the request being processed
2850 * @error: block status code
2851 * @nr_bytes: number of bytes to complete
2854 * Ends I/O on a number of bytes attached to @rq.
2855 * If @rq has leftover, sets it up for the next range of segments.
2858 * %false - we are done with this request
2859 * %true - still buffers pending for this request
2861 bool blk_end_request(struct request
*rq
, blk_status_t error
,
2862 unsigned int nr_bytes
)
2864 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2866 EXPORT_SYMBOL(blk_end_request
);
2869 * blk_end_request_all - Helper function for drives to finish the request.
2870 * @rq: the request to finish
2871 * @error: block status code
2874 * Completely finish @rq.
2876 void blk_end_request_all(struct request
*rq
, blk_status_t error
)
2879 unsigned int bidi_bytes
= 0;
2881 if (unlikely(blk_bidi_rq(rq
)))
2882 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2884 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2887 EXPORT_SYMBOL(blk_end_request_all
);
2890 * __blk_end_request - Helper function for drivers to complete the request.
2891 * @rq: the request being processed
2892 * @error: block status code
2893 * @nr_bytes: number of bytes to complete
2896 * Must be called with queue lock held unlike blk_end_request().
2899 * %false - we are done with this request
2900 * %true - still buffers pending for this request
2902 bool __blk_end_request(struct request
*rq
, blk_status_t error
,
2903 unsigned int nr_bytes
)
2905 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
2907 EXPORT_SYMBOL(__blk_end_request
);
2910 * __blk_end_request_all - Helper function for drives to finish the request.
2911 * @rq: the request to finish
2912 * @error: block status code
2915 * Completely finish @rq. Must be called with queue lock held.
2917 void __blk_end_request_all(struct request
*rq
, blk_status_t error
)
2920 unsigned int bidi_bytes
= 0;
2922 if (unlikely(blk_bidi_rq(rq
)))
2923 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
2925 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
2928 EXPORT_SYMBOL(__blk_end_request_all
);
2931 * __blk_end_request_cur - Helper function to finish the current request chunk.
2932 * @rq: the request to finish the current chunk for
2933 * @error: block status code
2936 * Complete the current consecutively mapped chunk from @rq. Must
2937 * be called with queue lock held.
2940 * %false - we are done with this request
2941 * %true - still buffers pending for this request
2943 bool __blk_end_request_cur(struct request
*rq
, blk_status_t error
)
2945 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
2947 EXPORT_SYMBOL(__blk_end_request_cur
);
2949 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
2952 if (bio_has_data(bio
))
2953 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
2955 rq
->__data_len
= bio
->bi_iter
.bi_size
;
2956 rq
->bio
= rq
->biotail
= bio
;
2959 rq
->rq_disk
= bio
->bi_bdev
->bd_disk
;
2962 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2964 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2965 * @rq: the request to be flushed
2968 * Flush all pages in @rq.
2970 void rq_flush_dcache_pages(struct request
*rq
)
2972 struct req_iterator iter
;
2973 struct bio_vec bvec
;
2975 rq_for_each_segment(bvec
, rq
, iter
)
2976 flush_dcache_page(bvec
.bv_page
);
2978 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
2982 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2983 * @q : the queue of the device being checked
2986 * Check if underlying low-level drivers of a device are busy.
2987 * If the drivers want to export their busy state, they must set own
2988 * exporting function using blk_queue_lld_busy() first.
2990 * Basically, this function is used only by request stacking drivers
2991 * to stop dispatching requests to underlying devices when underlying
2992 * devices are busy. This behavior helps more I/O merging on the queue
2993 * of the request stacking driver and prevents I/O throughput regression
2994 * on burst I/O load.
2997 * 0 - Not busy (The request stacking driver should dispatch request)
2998 * 1 - Busy (The request stacking driver should stop dispatching request)
3000 int blk_lld_busy(struct request_queue
*q
)
3003 return q
->lld_busy_fn(q
);
3007 EXPORT_SYMBOL_GPL(blk_lld_busy
);
3010 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3011 * @rq: the clone request to be cleaned up
3014 * Free all bios in @rq for a cloned request.
3016 void blk_rq_unprep_clone(struct request
*rq
)
3020 while ((bio
= rq
->bio
) != NULL
) {
3021 rq
->bio
= bio
->bi_next
;
3026 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3029 * Copy attributes of the original request to the clone request.
3030 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3032 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3034 dst
->cpu
= src
->cpu
;
3035 dst
->__sector
= blk_rq_pos(src
);
3036 dst
->__data_len
= blk_rq_bytes(src
);
3037 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3038 dst
->ioprio
= src
->ioprio
;
3039 dst
->extra_len
= src
->extra_len
;
3043 * blk_rq_prep_clone - Helper function to setup clone request
3044 * @rq: the request to be setup
3045 * @rq_src: original request to be cloned
3046 * @bs: bio_set that bios for clone are allocated from
3047 * @gfp_mask: memory allocation mask for bio
3048 * @bio_ctr: setup function to be called for each clone bio.
3049 * Returns %0 for success, non %0 for failure.
3050 * @data: private data to be passed to @bio_ctr
3053 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3054 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3055 * are not copied, and copying such parts is the caller's responsibility.
3056 * Also, pages which the original bios are pointing to are not copied
3057 * and the cloned bios just point same pages.
3058 * So cloned bios must be completed before original bios, which means
3059 * the caller must complete @rq before @rq_src.
3061 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3062 struct bio_set
*bs
, gfp_t gfp_mask
,
3063 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3066 struct bio
*bio
, *bio_src
;
3071 __rq_for_each_bio(bio_src
, rq_src
) {
3072 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3076 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3080 rq
->biotail
->bi_next
= bio
;
3083 rq
->bio
= rq
->biotail
= bio
;
3086 __blk_rq_prep_clone(rq
, rq_src
);
3093 blk_rq_unprep_clone(rq
);
3097 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3099 int kblockd_schedule_work(struct work_struct
*work
)
3101 return queue_work(kblockd_workqueue
, work
);
3103 EXPORT_SYMBOL(kblockd_schedule_work
);
3105 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3107 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3109 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3111 int kblockd_mod_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3112 unsigned long delay
)
3114 return mod_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3116 EXPORT_SYMBOL(kblockd_mod_delayed_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
, BLK_STS_IOERR
);
3289 * rq is already accounted, so use raw insert
3291 if (op_is_flush(rq
->cmd_flags
))
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
);
3517 #ifdef CONFIG_DEBUG_FS
3518 blk_debugfs_root
= debugfs_create_dir("block", NULL
);