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" },
146 [BLK_STS_AGAIN
] = { -EAGAIN
, "nonblocking retry" },
148 /* device mapper special case, should not leak out: */
149 [BLK_STS_DM_REQUEUE
] = { -EREMCHG
, "dm internal retry" },
151 /* everything else not covered above: */
152 [BLK_STS_IOERR
] = { -EIO
, "I/O" },
155 blk_status_t
errno_to_blk_status(int errno
)
159 for (i
= 0; i
< ARRAY_SIZE(blk_errors
); i
++) {
160 if (blk_errors
[i
].errno
== errno
)
161 return (__force blk_status_t
)i
;
164 return BLK_STS_IOERR
;
166 EXPORT_SYMBOL_GPL(errno_to_blk_status
);
168 int blk_status_to_errno(blk_status_t status
)
170 int idx
= (__force
int)status
;
172 if (WARN_ON_ONCE(idx
>= ARRAY_SIZE(blk_errors
)))
174 return blk_errors
[idx
].errno
;
176 EXPORT_SYMBOL_GPL(blk_status_to_errno
);
178 static void print_req_error(struct request
*req
, blk_status_t status
)
180 int idx
= (__force
int)status
;
182 if (WARN_ON_ONCE(idx
>= ARRAY_SIZE(blk_errors
)))
185 printk_ratelimited(KERN_ERR
"%s: %s error, dev %s, sector %llu\n",
186 __func__
, blk_errors
[idx
].name
, req
->rq_disk
?
187 req
->rq_disk
->disk_name
: "?",
188 (unsigned long long)blk_rq_pos(req
));
191 static void req_bio_endio(struct request
*rq
, struct bio
*bio
,
192 unsigned int nbytes
, blk_status_t error
)
195 bio
->bi_status
= error
;
197 if (unlikely(rq
->rq_flags
& RQF_QUIET
))
198 bio_set_flag(bio
, BIO_QUIET
);
200 bio_advance(bio
, nbytes
);
202 /* don't actually finish bio if it's part of flush sequence */
203 if (bio
->bi_iter
.bi_size
== 0 && !(rq
->rq_flags
& RQF_FLUSH_SEQ
))
207 void blk_dump_rq_flags(struct request
*rq
, char *msg
)
209 printk(KERN_INFO
"%s: dev %s: flags=%llx\n", msg
,
210 rq
->rq_disk
? rq
->rq_disk
->disk_name
: "?",
211 (unsigned long long) rq
->cmd_flags
);
213 printk(KERN_INFO
" sector %llu, nr/cnr %u/%u\n",
214 (unsigned long long)blk_rq_pos(rq
),
215 blk_rq_sectors(rq
), blk_rq_cur_sectors(rq
));
216 printk(KERN_INFO
" bio %p, biotail %p, len %u\n",
217 rq
->bio
, rq
->biotail
, blk_rq_bytes(rq
));
219 EXPORT_SYMBOL(blk_dump_rq_flags
);
221 static void blk_delay_work(struct work_struct
*work
)
223 struct request_queue
*q
;
225 q
= container_of(work
, struct request_queue
, delay_work
.work
);
226 spin_lock_irq(q
->queue_lock
);
228 spin_unlock_irq(q
->queue_lock
);
232 * blk_delay_queue - restart queueing after defined interval
233 * @q: The &struct request_queue in question
234 * @msecs: Delay in msecs
237 * Sometimes queueing needs to be postponed for a little while, to allow
238 * resources to come back. This function will make sure that queueing is
239 * restarted around the specified time.
241 void blk_delay_queue(struct request_queue
*q
, unsigned long msecs
)
243 lockdep_assert_held(q
->queue_lock
);
244 WARN_ON_ONCE(q
->mq_ops
);
246 if (likely(!blk_queue_dead(q
)))
247 queue_delayed_work(kblockd_workqueue
, &q
->delay_work
,
248 msecs_to_jiffies(msecs
));
250 EXPORT_SYMBOL(blk_delay_queue
);
253 * blk_start_queue_async - asynchronously restart a previously stopped queue
254 * @q: The &struct request_queue in question
257 * blk_start_queue_async() will clear the stop flag on the queue, and
258 * ensure that the request_fn for the queue is run from an async
261 void blk_start_queue_async(struct request_queue
*q
)
263 lockdep_assert_held(q
->queue_lock
);
264 WARN_ON_ONCE(q
->mq_ops
);
266 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
267 blk_run_queue_async(q
);
269 EXPORT_SYMBOL(blk_start_queue_async
);
272 * blk_start_queue - restart a previously stopped queue
273 * @q: The &struct request_queue in question
276 * blk_start_queue() will clear the stop flag on the queue, and call
277 * the request_fn for the queue if it was in a stopped state when
278 * entered. Also see blk_stop_queue().
280 void blk_start_queue(struct request_queue
*q
)
282 lockdep_assert_held(q
->queue_lock
);
283 WARN_ON(!in_interrupt() && !irqs_disabled());
284 WARN_ON_ONCE(q
->mq_ops
);
286 queue_flag_clear(QUEUE_FLAG_STOPPED
, q
);
289 EXPORT_SYMBOL(blk_start_queue
);
292 * blk_stop_queue - stop a queue
293 * @q: The &struct request_queue in question
296 * The Linux block layer assumes that a block driver will consume all
297 * entries on the request queue when the request_fn strategy is called.
298 * Often this will not happen, because of hardware limitations (queue
299 * depth settings). If a device driver gets a 'queue full' response,
300 * or if it simply chooses not to queue more I/O at one point, it can
301 * call this function to prevent the request_fn from being called until
302 * the driver has signalled it's ready to go again. This happens by calling
303 * blk_start_queue() to restart queue operations.
305 void blk_stop_queue(struct request_queue
*q
)
307 lockdep_assert_held(q
->queue_lock
);
308 WARN_ON_ONCE(q
->mq_ops
);
310 cancel_delayed_work(&q
->delay_work
);
311 queue_flag_set(QUEUE_FLAG_STOPPED
, q
);
313 EXPORT_SYMBOL(blk_stop_queue
);
316 * blk_sync_queue - cancel any pending callbacks on a queue
320 * The block layer may perform asynchronous callback activity
321 * on a queue, such as calling the unplug function after a timeout.
322 * A block device may call blk_sync_queue to ensure that any
323 * such activity is cancelled, thus allowing it to release resources
324 * that the callbacks might use. The caller must already have made sure
325 * that its ->make_request_fn will not re-add plugging prior to calling
328 * This function does not cancel any asynchronous activity arising
329 * out of elevator or throttling code. That would require elevator_exit()
330 * and blkcg_exit_queue() to be called with queue lock initialized.
333 void blk_sync_queue(struct request_queue
*q
)
335 del_timer_sync(&q
->timeout
);
336 cancel_work_sync(&q
->timeout_work
);
339 struct blk_mq_hw_ctx
*hctx
;
342 cancel_delayed_work_sync(&q
->requeue_work
);
343 queue_for_each_hw_ctx(q
, hctx
, i
)
344 cancel_delayed_work_sync(&hctx
->run_work
);
346 cancel_delayed_work_sync(&q
->delay_work
);
349 EXPORT_SYMBOL(blk_sync_queue
);
352 * blk_set_preempt_only - set QUEUE_FLAG_PREEMPT_ONLY
353 * @q: request queue pointer
355 * Returns the previous value of the PREEMPT_ONLY flag - 0 if the flag was not
356 * set and 1 if the flag was already set.
358 int blk_set_preempt_only(struct request_queue
*q
)
363 spin_lock_irqsave(q
->queue_lock
, flags
);
364 res
= queue_flag_test_and_set(QUEUE_FLAG_PREEMPT_ONLY
, q
);
365 spin_unlock_irqrestore(q
->queue_lock
, flags
);
369 EXPORT_SYMBOL_GPL(blk_set_preempt_only
);
371 void blk_clear_preempt_only(struct request_queue
*q
)
375 spin_lock_irqsave(q
->queue_lock
, flags
);
376 queue_flag_clear(QUEUE_FLAG_PREEMPT_ONLY
, q
);
377 wake_up_all(&q
->mq_freeze_wq
);
378 spin_unlock_irqrestore(q
->queue_lock
, flags
);
380 EXPORT_SYMBOL_GPL(blk_clear_preempt_only
);
383 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
384 * @q: The queue to run
387 * Invoke request handling on a queue if there are any pending requests.
388 * May be used to restart request handling after a request has completed.
389 * This variant runs the queue whether or not the queue has been
390 * stopped. Must be called with the queue lock held and interrupts
391 * disabled. See also @blk_run_queue.
393 inline void __blk_run_queue_uncond(struct request_queue
*q
)
395 lockdep_assert_held(q
->queue_lock
);
396 WARN_ON_ONCE(q
->mq_ops
);
398 if (unlikely(blk_queue_dead(q
)))
402 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
403 * the queue lock internally. As a result multiple threads may be
404 * running such a request function concurrently. Keep track of the
405 * number of active request_fn invocations such that blk_drain_queue()
406 * can wait until all these request_fn calls have finished.
408 q
->request_fn_active
++;
410 q
->request_fn_active
--;
412 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond
);
415 * __blk_run_queue - run a single device queue
416 * @q: The queue to run
419 * See @blk_run_queue.
421 void __blk_run_queue(struct request_queue
*q
)
423 lockdep_assert_held(q
->queue_lock
);
424 WARN_ON_ONCE(q
->mq_ops
);
426 if (unlikely(blk_queue_stopped(q
)))
429 __blk_run_queue_uncond(q
);
431 EXPORT_SYMBOL(__blk_run_queue
);
434 * blk_run_queue_async - run a single device queue in workqueue context
435 * @q: The queue to run
438 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
442 * Since it is not allowed to run q->delay_work after blk_cleanup_queue()
443 * has canceled q->delay_work, callers must hold the queue lock to avoid
444 * race conditions between blk_cleanup_queue() and blk_run_queue_async().
446 void blk_run_queue_async(struct request_queue
*q
)
448 lockdep_assert_held(q
->queue_lock
);
449 WARN_ON_ONCE(q
->mq_ops
);
451 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
452 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
454 EXPORT_SYMBOL(blk_run_queue_async
);
457 * blk_run_queue - run a single device queue
458 * @q: The queue to run
461 * Invoke request handling on this queue, if it has pending work to do.
462 * May be used to restart queueing when a request has completed.
464 void blk_run_queue(struct request_queue
*q
)
468 WARN_ON_ONCE(q
->mq_ops
);
470 spin_lock_irqsave(q
->queue_lock
, flags
);
472 spin_unlock_irqrestore(q
->queue_lock
, flags
);
474 EXPORT_SYMBOL(blk_run_queue
);
476 void blk_put_queue(struct request_queue
*q
)
478 kobject_put(&q
->kobj
);
480 EXPORT_SYMBOL(blk_put_queue
);
483 * __blk_drain_queue - drain requests from request_queue
485 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
487 * Drain requests from @q. If @drain_all is set, all requests are drained.
488 * If not, only ELVPRIV requests are drained. The caller is responsible
489 * for ensuring that no new requests which need to be drained are queued.
491 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
492 __releases(q
->queue_lock
)
493 __acquires(q
->queue_lock
)
497 lockdep_assert_held(q
->queue_lock
);
498 WARN_ON_ONCE(q
->mq_ops
);
504 * The caller might be trying to drain @q before its
505 * elevator is initialized.
508 elv_drain_elevator(q
);
510 blkcg_drain_queue(q
);
513 * This function might be called on a queue which failed
514 * driver init after queue creation or is not yet fully
515 * active yet. Some drivers (e.g. fd and loop) get unhappy
516 * in such cases. Kick queue iff dispatch queue has
517 * something on it and @q has request_fn set.
519 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
522 drain
|= q
->nr_rqs_elvpriv
;
523 drain
|= q
->request_fn_active
;
526 * Unfortunately, requests are queued at and tracked from
527 * multiple places and there's no single counter which can
528 * be drained. Check all the queues and counters.
531 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
532 drain
|= !list_empty(&q
->queue_head
);
533 for (i
= 0; i
< 2; i
++) {
534 drain
|= q
->nr_rqs
[i
];
535 drain
|= q
->in_flight
[i
];
537 drain
|= !list_empty(&fq
->flush_queue
[i
]);
544 spin_unlock_irq(q
->queue_lock
);
548 spin_lock_irq(q
->queue_lock
);
552 * With queue marked dead, any woken up waiter will fail the
553 * allocation path, so the wakeup chaining is lost and we're
554 * left with hung waiters. We need to wake up those waiters.
557 struct request_list
*rl
;
559 blk_queue_for_each_rl(rl
, q
)
560 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
561 wake_up_all(&rl
->wait
[i
]);
565 void blk_drain_queue(struct request_queue
*q
)
567 spin_lock_irq(q
->queue_lock
);
568 __blk_drain_queue(q
, true);
569 spin_unlock_irq(q
->queue_lock
);
573 * blk_queue_bypass_start - enter queue bypass mode
574 * @q: queue of interest
576 * In bypass mode, only the dispatch FIFO queue of @q is used. This
577 * function makes @q enter bypass mode and drains all requests which were
578 * throttled or issued before. On return, it's guaranteed that no request
579 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
580 * inside queue or RCU read lock.
582 void blk_queue_bypass_start(struct request_queue
*q
)
584 WARN_ON_ONCE(q
->mq_ops
);
586 spin_lock_irq(q
->queue_lock
);
588 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
589 spin_unlock_irq(q
->queue_lock
);
592 * Queues start drained. Skip actual draining till init is
593 * complete. This avoids lenghty delays during queue init which
594 * can happen many times during boot.
596 if (blk_queue_init_done(q
)) {
597 spin_lock_irq(q
->queue_lock
);
598 __blk_drain_queue(q
, false);
599 spin_unlock_irq(q
->queue_lock
);
601 /* ensure blk_queue_bypass() is %true inside RCU read lock */
605 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
608 * blk_queue_bypass_end - leave queue bypass mode
609 * @q: queue of interest
611 * Leave bypass mode and restore the normal queueing behavior.
613 * Note: although blk_queue_bypass_start() is only called for blk-sq queues,
614 * this function is called for both blk-sq and blk-mq queues.
616 void blk_queue_bypass_end(struct request_queue
*q
)
618 spin_lock_irq(q
->queue_lock
);
619 if (!--q
->bypass_depth
)
620 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
621 WARN_ON_ONCE(q
->bypass_depth
< 0);
622 spin_unlock_irq(q
->queue_lock
);
624 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
626 void blk_set_queue_dying(struct request_queue
*q
)
628 spin_lock_irq(q
->queue_lock
);
629 queue_flag_set(QUEUE_FLAG_DYING
, q
);
630 spin_unlock_irq(q
->queue_lock
);
633 * When queue DYING flag is set, we need to block new req
634 * entering queue, so we call blk_freeze_queue_start() to
635 * prevent I/O from crossing blk_queue_enter().
637 blk_freeze_queue_start(q
);
640 blk_mq_wake_waiters(q
);
642 struct request_list
*rl
;
644 spin_lock_irq(q
->queue_lock
);
645 blk_queue_for_each_rl(rl
, q
) {
647 wake_up_all(&rl
->wait
[BLK_RW_SYNC
]);
648 wake_up_all(&rl
->wait
[BLK_RW_ASYNC
]);
651 spin_unlock_irq(q
->queue_lock
);
654 /* Make blk_queue_enter() reexamine the DYING flag. */
655 wake_up_all(&q
->mq_freeze_wq
);
657 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
660 * blk_cleanup_queue - shutdown a request queue
661 * @q: request queue to shutdown
663 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
664 * put it. All future requests will be failed immediately with -ENODEV.
666 void blk_cleanup_queue(struct request_queue
*q
)
668 spinlock_t
*lock
= q
->queue_lock
;
670 /* mark @q DYING, no new request or merges will be allowed afterwards */
671 mutex_lock(&q
->sysfs_lock
);
672 blk_set_queue_dying(q
);
676 * A dying queue is permanently in bypass mode till released. Note
677 * that, unlike blk_queue_bypass_start(), we aren't performing
678 * synchronize_rcu() after entering bypass mode to avoid the delay
679 * as some drivers create and destroy a lot of queues while
680 * probing. This is still safe because blk_release_queue() will be
681 * called only after the queue refcnt drops to zero and nothing,
682 * RCU or not, would be traversing the queue by then.
685 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
687 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
688 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
689 queue_flag_set(QUEUE_FLAG_DYING
, q
);
690 spin_unlock_irq(lock
);
691 mutex_unlock(&q
->sysfs_lock
);
694 * Drain all requests queued before DYING marking. Set DEAD flag to
695 * prevent that q->request_fn() gets invoked after draining finished.
699 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
700 spin_unlock_irq(lock
);
703 * make sure all in-progress dispatch are completed because
704 * blk_freeze_queue() can only complete all requests, and
705 * dispatch may still be in-progress since we dispatch requests
706 * from more than one contexts
709 blk_mq_quiesce_queue(q
);
711 /* for synchronous bio-based driver finish in-flight integrity i/o */
712 blk_flush_integrity();
714 /* @q won't process any more request, flush async actions */
715 del_timer_sync(&q
->backing_dev_info
->laptop_mode_wb_timer
);
719 blk_mq_free_queue(q
);
720 percpu_ref_exit(&q
->q_usage_counter
);
723 if (q
->queue_lock
!= &q
->__queue_lock
)
724 q
->queue_lock
= &q
->__queue_lock
;
725 spin_unlock_irq(lock
);
727 /* @q is and will stay empty, shutdown and put */
730 EXPORT_SYMBOL(blk_cleanup_queue
);
732 /* Allocate memory local to the request queue */
733 static void *alloc_request_simple(gfp_t gfp_mask
, void *data
)
735 struct request_queue
*q
= data
;
737 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, q
->node
);
740 static void free_request_simple(void *element
, void *data
)
742 kmem_cache_free(request_cachep
, element
);
745 static void *alloc_request_size(gfp_t gfp_mask
, void *data
)
747 struct request_queue
*q
= data
;
750 rq
= kmalloc_node(sizeof(struct request
) + q
->cmd_size
, gfp_mask
,
752 if (rq
&& q
->init_rq_fn
&& q
->init_rq_fn(q
, rq
, gfp_mask
) < 0) {
759 static void free_request_size(void *element
, void *data
)
761 struct request_queue
*q
= data
;
764 q
->exit_rq_fn(q
, element
);
768 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
771 if (unlikely(rl
->rq_pool
) || q
->mq_ops
)
775 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
776 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
777 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
778 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
781 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
782 alloc_request_size
, free_request_size
,
783 q
, gfp_mask
, q
->node
);
785 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
786 alloc_request_simple
, free_request_simple
,
787 q
, gfp_mask
, q
->node
);
792 if (rl
!= &q
->root_rl
)
793 WARN_ON_ONCE(!blk_get_queue(q
));
798 void blk_exit_rl(struct request_queue
*q
, struct request_list
*rl
)
801 mempool_destroy(rl
->rq_pool
);
802 if (rl
!= &q
->root_rl
)
807 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
809 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
811 EXPORT_SYMBOL(blk_alloc_queue
);
814 * blk_queue_enter() - try to increase q->q_usage_counter
815 * @q: request queue pointer
816 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
818 int blk_queue_enter(struct request_queue
*q
, blk_mq_req_flags_t flags
)
820 const bool preempt
= flags
& BLK_MQ_REQ_PREEMPT
;
823 bool success
= false;
826 if (percpu_ref_tryget_live(&q
->q_usage_counter
)) {
828 * The code that sets the PREEMPT_ONLY flag is
829 * responsible for ensuring that that flag is globally
830 * visible before the queue is unfrozen.
832 if (preempt
|| !blk_queue_preempt_only(q
)) {
835 percpu_ref_put(&q
->q_usage_counter
);
843 if (flags
& BLK_MQ_REQ_NOWAIT
)
847 * read pair of barrier in blk_freeze_queue_start(),
848 * we need to order reading __PERCPU_REF_DEAD flag of
849 * .q_usage_counter and reading .mq_freeze_depth or
850 * queue dying flag, otherwise the following wait may
851 * never return if the two reads are reordered.
855 wait_event(q
->mq_freeze_wq
,
856 (atomic_read(&q
->mq_freeze_depth
) == 0 &&
857 (preempt
|| !blk_queue_preempt_only(q
))) ||
859 if (blk_queue_dying(q
))
864 void blk_queue_exit(struct request_queue
*q
)
866 percpu_ref_put(&q
->q_usage_counter
);
869 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
871 struct request_queue
*q
=
872 container_of(ref
, struct request_queue
, q_usage_counter
);
874 wake_up_all(&q
->mq_freeze_wq
);
877 static void blk_rq_timed_out_timer(struct timer_list
*t
)
879 struct request_queue
*q
= from_timer(q
, t
, timeout
);
881 kblockd_schedule_work(&q
->timeout_work
);
884 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
886 struct request_queue
*q
;
888 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
889 gfp_mask
| __GFP_ZERO
, node_id
);
893 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
897 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
);
901 q
->backing_dev_info
= bdi_alloc_node(gfp_mask
, node_id
);
902 if (!q
->backing_dev_info
)
905 q
->stats
= blk_alloc_queue_stats();
909 q
->backing_dev_info
->ra_pages
=
910 (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
911 q
->backing_dev_info
->capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
912 q
->backing_dev_info
->name
= "block";
915 timer_setup(&q
->backing_dev_info
->laptop_mode_wb_timer
,
916 laptop_mode_timer_fn
, 0);
917 timer_setup(&q
->timeout
, blk_rq_timed_out_timer
, 0);
918 INIT_WORK(&q
->timeout_work
, NULL
);
919 INIT_LIST_HEAD(&q
->queue_head
);
920 INIT_LIST_HEAD(&q
->timeout_list
);
921 INIT_LIST_HEAD(&q
->icq_list
);
922 #ifdef CONFIG_BLK_CGROUP
923 INIT_LIST_HEAD(&q
->blkg_list
);
925 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
927 kobject_init(&q
->kobj
, &blk_queue_ktype
);
929 #ifdef CONFIG_BLK_DEV_IO_TRACE
930 mutex_init(&q
->blk_trace_mutex
);
932 mutex_init(&q
->sysfs_lock
);
933 spin_lock_init(&q
->__queue_lock
);
936 * By default initialize queue_lock to internal lock and driver can
937 * override it later if need be.
939 q
->queue_lock
= &q
->__queue_lock
;
942 * A queue starts its life with bypass turned on to avoid
943 * unnecessary bypass on/off overhead and nasty surprises during
944 * init. The initial bypass will be finished when the queue is
945 * registered by blk_register_queue().
948 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
950 init_waitqueue_head(&q
->mq_freeze_wq
);
953 * Init percpu_ref in atomic mode so that it's faster to shutdown.
954 * See blk_register_queue() for details.
956 if (percpu_ref_init(&q
->q_usage_counter
,
957 blk_queue_usage_counter_release
,
958 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
961 if (blkcg_init_queue(q
))
967 percpu_ref_exit(&q
->q_usage_counter
);
969 blk_free_queue_stats(q
->stats
);
971 bdi_put(q
->backing_dev_info
);
973 bioset_free(q
->bio_split
);
975 ida_simple_remove(&blk_queue_ida
, q
->id
);
977 kmem_cache_free(blk_requestq_cachep
, q
);
980 EXPORT_SYMBOL(blk_alloc_queue_node
);
983 * blk_init_queue - prepare a request queue for use with a block device
984 * @rfn: The function to be called to process requests that have been
985 * placed on the queue.
986 * @lock: Request queue spin lock
989 * If a block device wishes to use the standard request handling procedures,
990 * which sorts requests and coalesces adjacent requests, then it must
991 * call blk_init_queue(). The function @rfn will be called when there
992 * are requests on the queue that need to be processed. If the device
993 * supports plugging, then @rfn may not be called immediately when requests
994 * are available on the queue, but may be called at some time later instead.
995 * Plugged queues are generally unplugged when a buffer belonging to one
996 * of the requests on the queue is needed, or due to memory pressure.
998 * @rfn is not required, or even expected, to remove all requests off the
999 * queue, but only as many as it can handle at a time. If it does leave
1000 * requests on the queue, it is responsible for arranging that the requests
1001 * get dealt with eventually.
1003 * The queue spin lock must be held while manipulating the requests on the
1004 * request queue; this lock will be taken also from interrupt context, so irq
1005 * disabling is needed for it.
1007 * Function returns a pointer to the initialized request queue, or %NULL if
1008 * it didn't succeed.
1011 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1012 * when the block device is deactivated (such as at module unload).
1015 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1017 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
1019 EXPORT_SYMBOL(blk_init_queue
);
1021 struct request_queue
*
1022 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1024 struct request_queue
*q
;
1026 q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1030 q
->request_fn
= rfn
;
1032 q
->queue_lock
= lock
;
1033 if (blk_init_allocated_queue(q
) < 0) {
1034 blk_cleanup_queue(q
);
1040 EXPORT_SYMBOL(blk_init_queue_node
);
1042 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
1045 int blk_init_allocated_queue(struct request_queue
*q
)
1047 WARN_ON_ONCE(q
->mq_ops
);
1049 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, q
->cmd_size
);
1053 if (q
->init_rq_fn
&& q
->init_rq_fn(q
, q
->fq
->flush_rq
, GFP_KERNEL
))
1054 goto out_free_flush_queue
;
1056 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
1057 goto out_exit_flush_rq
;
1059 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
1060 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
1063 * This also sets hw/phys segments, boundary and size
1065 blk_queue_make_request(q
, blk_queue_bio
);
1067 q
->sg_reserved_size
= INT_MAX
;
1069 /* Protect q->elevator from elevator_change */
1070 mutex_lock(&q
->sysfs_lock
);
1073 if (elevator_init(q
, NULL
)) {
1074 mutex_unlock(&q
->sysfs_lock
);
1075 goto out_exit_flush_rq
;
1078 mutex_unlock(&q
->sysfs_lock
);
1083 q
->exit_rq_fn(q
, q
->fq
->flush_rq
);
1084 out_free_flush_queue
:
1085 blk_free_flush_queue(q
->fq
);
1088 EXPORT_SYMBOL(blk_init_allocated_queue
);
1090 bool blk_get_queue(struct request_queue
*q
)
1092 if (likely(!blk_queue_dying(q
))) {
1099 EXPORT_SYMBOL(blk_get_queue
);
1101 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
1103 if (rq
->rq_flags
& RQF_ELVPRIV
) {
1104 elv_put_request(rl
->q
, rq
);
1106 put_io_context(rq
->elv
.icq
->ioc
);
1109 mempool_free(rq
, rl
->rq_pool
);
1113 * ioc_batching returns true if the ioc is a valid batching request and
1114 * should be given priority access to a request.
1116 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
1122 * Make sure the process is able to allocate at least 1 request
1123 * even if the batch times out, otherwise we could theoretically
1126 return ioc
->nr_batch_requests
== q
->nr_batching
||
1127 (ioc
->nr_batch_requests
> 0
1128 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
1132 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1133 * will cause the process to be a "batcher" on all queues in the system. This
1134 * is the behaviour we want though - once it gets a wakeup it should be given
1137 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
1139 if (!ioc
|| ioc_batching(q
, ioc
))
1142 ioc
->nr_batch_requests
= q
->nr_batching
;
1143 ioc
->last_waited
= jiffies
;
1146 static void __freed_request(struct request_list
*rl
, int sync
)
1148 struct request_queue
*q
= rl
->q
;
1150 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
1151 blk_clear_congested(rl
, sync
);
1153 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
1154 if (waitqueue_active(&rl
->wait
[sync
]))
1155 wake_up(&rl
->wait
[sync
]);
1157 blk_clear_rl_full(rl
, sync
);
1162 * A request has just been released. Account for it, update the full and
1163 * congestion status, wake up any waiters. Called under q->queue_lock.
1165 static void freed_request(struct request_list
*rl
, bool sync
,
1166 req_flags_t rq_flags
)
1168 struct request_queue
*q
= rl
->q
;
1172 if (rq_flags
& RQF_ELVPRIV
)
1173 q
->nr_rqs_elvpriv
--;
1175 __freed_request(rl
, sync
);
1177 if (unlikely(rl
->starved
[sync
^ 1]))
1178 __freed_request(rl
, sync
^ 1);
1181 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
1183 struct request_list
*rl
;
1184 int on_thresh
, off_thresh
;
1186 WARN_ON_ONCE(q
->mq_ops
);
1188 spin_lock_irq(q
->queue_lock
);
1189 q
->nr_requests
= nr
;
1190 blk_queue_congestion_threshold(q
);
1191 on_thresh
= queue_congestion_on_threshold(q
);
1192 off_thresh
= queue_congestion_off_threshold(q
);
1194 blk_queue_for_each_rl(rl
, q
) {
1195 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
1196 blk_set_congested(rl
, BLK_RW_SYNC
);
1197 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
1198 blk_clear_congested(rl
, BLK_RW_SYNC
);
1200 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
1201 blk_set_congested(rl
, BLK_RW_ASYNC
);
1202 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
1203 blk_clear_congested(rl
, BLK_RW_ASYNC
);
1205 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
1206 blk_set_rl_full(rl
, BLK_RW_SYNC
);
1208 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1209 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1212 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1213 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1215 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1216 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1220 spin_unlock_irq(q
->queue_lock
);
1225 * __get_request - get a free request
1226 * @rl: request list to allocate from
1227 * @op: operation and flags
1228 * @bio: bio to allocate request for (can be %NULL)
1229 * @flags: BLQ_MQ_REQ_* flags
1231 * Get a free request from @q. This function may fail under memory
1232 * pressure or if @q is dead.
1234 * Must be called with @q->queue_lock held and,
1235 * Returns ERR_PTR on failure, with @q->queue_lock held.
1236 * Returns request pointer on success, with @q->queue_lock *not held*.
1238 static struct request
*__get_request(struct request_list
*rl
, unsigned int op
,
1239 struct bio
*bio
, blk_mq_req_flags_t flags
)
1241 struct request_queue
*q
= rl
->q
;
1243 struct elevator_type
*et
= q
->elevator
->type
;
1244 struct io_context
*ioc
= rq_ioc(bio
);
1245 struct io_cq
*icq
= NULL
;
1246 const bool is_sync
= op_is_sync(op
);
1248 gfp_t gfp_mask
= flags
& BLK_MQ_REQ_NOWAIT
? GFP_ATOMIC
:
1249 __GFP_DIRECT_RECLAIM
;
1250 req_flags_t rq_flags
= RQF_ALLOCED
;
1252 lockdep_assert_held(q
->queue_lock
);
1254 if (unlikely(blk_queue_dying(q
)))
1255 return ERR_PTR(-ENODEV
);
1257 may_queue
= elv_may_queue(q
, op
);
1258 if (may_queue
== ELV_MQUEUE_NO
)
1261 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1262 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1264 * The queue will fill after this allocation, so set
1265 * it as full, and mark this process as "batching".
1266 * This process will be allowed to complete a batch of
1267 * requests, others will be blocked.
1269 if (!blk_rl_full(rl
, is_sync
)) {
1270 ioc_set_batching(q
, ioc
);
1271 blk_set_rl_full(rl
, is_sync
);
1273 if (may_queue
!= ELV_MQUEUE_MUST
1274 && !ioc_batching(q
, ioc
)) {
1276 * The queue is full and the allocating
1277 * process is not a "batcher", and not
1278 * exempted by the IO scheduler
1280 return ERR_PTR(-ENOMEM
);
1284 blk_set_congested(rl
, is_sync
);
1288 * Only allow batching queuers to allocate up to 50% over the defined
1289 * limit of requests, otherwise we could have thousands of requests
1290 * allocated with any setting of ->nr_requests
1292 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1293 return ERR_PTR(-ENOMEM
);
1295 q
->nr_rqs
[is_sync
]++;
1296 rl
->count
[is_sync
]++;
1297 rl
->starved
[is_sync
] = 0;
1300 * Decide whether the new request will be managed by elevator. If
1301 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1302 * prevent the current elevator from being destroyed until the new
1303 * request is freed. This guarantees icq's won't be destroyed and
1304 * makes creating new ones safe.
1306 * Flush requests do not use the elevator so skip initialization.
1307 * This allows a request to share the flush and elevator data.
1309 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1310 * it will be created after releasing queue_lock.
1312 if (!op_is_flush(op
) && !blk_queue_bypass(q
)) {
1313 rq_flags
|= RQF_ELVPRIV
;
1314 q
->nr_rqs_elvpriv
++;
1315 if (et
->icq_cache
&& ioc
)
1316 icq
= ioc_lookup_icq(ioc
, q
);
1319 if (blk_queue_io_stat(q
))
1320 rq_flags
|= RQF_IO_STAT
;
1321 spin_unlock_irq(q
->queue_lock
);
1323 /* allocate and init request */
1324 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1329 blk_rq_set_rl(rq
, rl
);
1331 rq
->rq_flags
= rq_flags
;
1332 if (flags
& BLK_MQ_REQ_PREEMPT
)
1333 rq
->rq_flags
|= RQF_PREEMPT
;
1336 if (rq_flags
& RQF_ELVPRIV
) {
1337 if (unlikely(et
->icq_cache
&& !icq
)) {
1339 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1345 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1348 /* @rq->elv.icq holds io_context until @rq is freed */
1350 get_io_context(icq
->ioc
);
1354 * ioc may be NULL here, and ioc_batching will be false. That's
1355 * OK, if the queue is under the request limit then requests need
1356 * not count toward the nr_batch_requests limit. There will always
1357 * be some limit enforced by BLK_BATCH_TIME.
1359 if (ioc_batching(q
, ioc
))
1360 ioc
->nr_batch_requests
--;
1362 trace_block_getrq(q
, bio
, op
);
1367 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1368 * and may fail indefinitely under memory pressure and thus
1369 * shouldn't stall IO. Treat this request as !elvpriv. This will
1370 * disturb iosched and blkcg but weird is bettern than dead.
1372 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1373 __func__
, dev_name(q
->backing_dev_info
->dev
));
1375 rq
->rq_flags
&= ~RQF_ELVPRIV
;
1378 spin_lock_irq(q
->queue_lock
);
1379 q
->nr_rqs_elvpriv
--;
1380 spin_unlock_irq(q
->queue_lock
);
1385 * Allocation failed presumably due to memory. Undo anything we
1386 * might have messed up.
1388 * Allocating task should really be put onto the front of the wait
1389 * queue, but this is pretty rare.
1391 spin_lock_irq(q
->queue_lock
);
1392 freed_request(rl
, is_sync
, rq_flags
);
1395 * in the very unlikely event that allocation failed and no
1396 * requests for this direction was pending, mark us starved so that
1397 * freeing of a request in the other direction will notice
1398 * us. another possible fix would be to split the rq mempool into
1402 if (unlikely(rl
->count
[is_sync
] == 0))
1403 rl
->starved
[is_sync
] = 1;
1404 return ERR_PTR(-ENOMEM
);
1408 * get_request - get a free request
1409 * @q: request_queue to allocate request from
1410 * @op: operation and flags
1411 * @bio: bio to allocate request for (can be %NULL)
1412 * @flags: BLK_MQ_REQ_* flags.
1414 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1415 * this function keeps retrying under memory pressure and fails iff @q is dead.
1417 * Must be called with @q->queue_lock held and,
1418 * Returns ERR_PTR on failure, with @q->queue_lock held.
1419 * Returns request pointer on success, with @q->queue_lock *not held*.
1421 static struct request
*get_request(struct request_queue
*q
, unsigned int op
,
1422 struct bio
*bio
, blk_mq_req_flags_t flags
)
1424 const bool is_sync
= op_is_sync(op
);
1426 struct request_list
*rl
;
1429 lockdep_assert_held(q
->queue_lock
);
1430 WARN_ON_ONCE(q
->mq_ops
);
1432 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1434 rq
= __get_request(rl
, op
, bio
, flags
);
1438 if (op
& REQ_NOWAIT
) {
1440 return ERR_PTR(-EAGAIN
);
1443 if ((flags
& BLK_MQ_REQ_NOWAIT
) || unlikely(blk_queue_dying(q
))) {
1448 /* wait on @rl and retry */
1449 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1450 TASK_UNINTERRUPTIBLE
);
1452 trace_block_sleeprq(q
, bio
, op
);
1454 spin_unlock_irq(q
->queue_lock
);
1458 * After sleeping, we become a "batching" process and will be able
1459 * to allocate at least one request, and up to a big batch of them
1460 * for a small period time. See ioc_batching, ioc_set_batching
1462 ioc_set_batching(q
, current
->io_context
);
1464 spin_lock_irq(q
->queue_lock
);
1465 finish_wait(&rl
->wait
[is_sync
], &wait
);
1470 /* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */
1471 static struct request
*blk_old_get_request(struct request_queue
*q
,
1472 unsigned int op
, blk_mq_req_flags_t flags
)
1475 gfp_t gfp_mask
= flags
& BLK_MQ_REQ_NOWAIT
? GFP_ATOMIC
:
1476 __GFP_DIRECT_RECLAIM
;
1479 WARN_ON_ONCE(q
->mq_ops
);
1481 /* create ioc upfront */
1482 create_io_context(gfp_mask
, q
->node
);
1484 ret
= blk_queue_enter(q
, flags
);
1486 return ERR_PTR(ret
);
1487 spin_lock_irq(q
->queue_lock
);
1488 rq
= get_request(q
, op
, NULL
, flags
);
1490 spin_unlock_irq(q
->queue_lock
);
1495 /* q->queue_lock is unlocked at this point */
1497 rq
->__sector
= (sector_t
) -1;
1498 rq
->bio
= rq
->biotail
= NULL
;
1503 * blk_get_request_flags - allocate a request
1504 * @q: request queue to allocate a request for
1505 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
1506 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
1508 struct request
*blk_get_request_flags(struct request_queue
*q
, unsigned int op
,
1509 blk_mq_req_flags_t flags
)
1511 struct request
*req
;
1513 WARN_ON_ONCE(op
& REQ_NOWAIT
);
1514 WARN_ON_ONCE(flags
& ~(BLK_MQ_REQ_NOWAIT
| BLK_MQ_REQ_PREEMPT
));
1517 req
= blk_mq_alloc_request(q
, op
, flags
);
1518 if (!IS_ERR(req
) && q
->mq_ops
->initialize_rq_fn
)
1519 q
->mq_ops
->initialize_rq_fn(req
);
1521 req
= blk_old_get_request(q
, op
, flags
);
1522 if (!IS_ERR(req
) && q
->initialize_rq_fn
)
1523 q
->initialize_rq_fn(req
);
1528 EXPORT_SYMBOL(blk_get_request_flags
);
1530 struct request
*blk_get_request(struct request_queue
*q
, unsigned int op
,
1533 return blk_get_request_flags(q
, op
, gfp_mask
& __GFP_DIRECT_RECLAIM
?
1534 0 : BLK_MQ_REQ_NOWAIT
);
1536 EXPORT_SYMBOL(blk_get_request
);
1539 * blk_requeue_request - put a request back on queue
1540 * @q: request queue where request should be inserted
1541 * @rq: request to be inserted
1544 * Drivers often keep queueing requests until the hardware cannot accept
1545 * more, when that condition happens we need to put the request back
1546 * on the queue. Must be called with queue lock held.
1548 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1550 lockdep_assert_held(q
->queue_lock
);
1551 WARN_ON_ONCE(q
->mq_ops
);
1553 blk_delete_timer(rq
);
1554 blk_clear_rq_complete(rq
);
1555 trace_block_rq_requeue(q
, rq
);
1556 wbt_requeue(q
->rq_wb
, &rq
->issue_stat
);
1558 if (rq
->rq_flags
& RQF_QUEUED
)
1559 blk_queue_end_tag(q
, rq
);
1561 BUG_ON(blk_queued_rq(rq
));
1563 elv_requeue_request(q
, rq
);
1565 EXPORT_SYMBOL(blk_requeue_request
);
1567 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1570 blk_account_io_start(rq
, true);
1571 __elv_add_request(q
, rq
, where
);
1574 static void part_round_stats_single(struct request_queue
*q
, int cpu
,
1575 struct hd_struct
*part
, unsigned long now
,
1576 unsigned int inflight
)
1579 __part_stat_add(cpu
, part
, time_in_queue
,
1580 inflight
* (now
- part
->stamp
));
1581 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1587 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1588 * @q: target block queue
1589 * @cpu: cpu number for stats access
1590 * @part: target partition
1592 * The average IO queue length and utilisation statistics are maintained
1593 * by observing the current state of the queue length and the amount of
1594 * time it has been in this state for.
1596 * Normally, that accounting is done on IO completion, but that can result
1597 * in more than a second's worth of IO being accounted for within any one
1598 * second, leading to >100% utilisation. To deal with that, we call this
1599 * function to do a round-off before returning the results when reading
1600 * /proc/diskstats. This accounts immediately for all queue usage up to
1601 * the current jiffies and restarts the counters again.
1603 void part_round_stats(struct request_queue
*q
, int cpu
, struct hd_struct
*part
)
1605 struct hd_struct
*part2
= NULL
;
1606 unsigned long now
= jiffies
;
1607 unsigned int inflight
[2];
1610 if (part
->stamp
!= now
)
1614 part2
= &part_to_disk(part
)->part0
;
1615 if (part2
->stamp
!= now
)
1622 part_in_flight(q
, part
, inflight
);
1625 part_round_stats_single(q
, cpu
, part2
, now
, inflight
[1]);
1627 part_round_stats_single(q
, cpu
, part
, now
, inflight
[0]);
1629 EXPORT_SYMBOL_GPL(part_round_stats
);
1632 static void blk_pm_put_request(struct request
*rq
)
1634 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1635 pm_runtime_mark_last_busy(rq
->q
->dev
);
1638 static inline void blk_pm_put_request(struct request
*rq
) {}
1641 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1643 req_flags_t rq_flags
= req
->rq_flags
;
1649 blk_mq_free_request(req
);
1653 lockdep_assert_held(q
->queue_lock
);
1655 blk_pm_put_request(req
);
1657 elv_completed_request(q
, req
);
1659 /* this is a bio leak */
1660 WARN_ON(req
->bio
!= NULL
);
1662 wbt_done(q
->rq_wb
, &req
->issue_stat
);
1665 * Request may not have originated from ll_rw_blk. if not,
1666 * it didn't come out of our reserved rq pools
1668 if (rq_flags
& RQF_ALLOCED
) {
1669 struct request_list
*rl
= blk_rq_rl(req
);
1670 bool sync
= op_is_sync(req
->cmd_flags
);
1672 BUG_ON(!list_empty(&req
->queuelist
));
1673 BUG_ON(ELV_ON_HASH(req
));
1675 blk_free_request(rl
, req
);
1676 freed_request(rl
, sync
, rq_flags
);
1681 EXPORT_SYMBOL_GPL(__blk_put_request
);
1683 void blk_put_request(struct request
*req
)
1685 struct request_queue
*q
= req
->q
;
1688 blk_mq_free_request(req
);
1690 unsigned long flags
;
1692 spin_lock_irqsave(q
->queue_lock
, flags
);
1693 __blk_put_request(q
, req
);
1694 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1697 EXPORT_SYMBOL(blk_put_request
);
1699 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1702 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1704 if (!ll_back_merge_fn(q
, req
, bio
))
1707 trace_block_bio_backmerge(q
, req
, bio
);
1709 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1710 blk_rq_set_mixed_merge(req
);
1712 req
->biotail
->bi_next
= bio
;
1714 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1715 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1717 blk_account_io_start(req
, false);
1721 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1724 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1726 if (!ll_front_merge_fn(q
, req
, bio
))
1729 trace_block_bio_frontmerge(q
, req
, bio
);
1731 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1732 blk_rq_set_mixed_merge(req
);
1734 bio
->bi_next
= req
->bio
;
1737 req
->__sector
= bio
->bi_iter
.bi_sector
;
1738 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1739 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1741 blk_account_io_start(req
, false);
1745 bool bio_attempt_discard_merge(struct request_queue
*q
, struct request
*req
,
1748 unsigned short segments
= blk_rq_nr_discard_segments(req
);
1750 if (segments
>= queue_max_discard_segments(q
))
1752 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
1753 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
1756 req
->biotail
->bi_next
= bio
;
1758 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1759 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1760 req
->nr_phys_segments
= segments
+ 1;
1762 blk_account_io_start(req
, false);
1765 req_set_nomerge(q
, req
);
1770 * blk_attempt_plug_merge - try to merge with %current's plugged list
1771 * @q: request_queue new bio is being queued at
1772 * @bio: new bio being queued
1773 * @request_count: out parameter for number of traversed plugged requests
1774 * @same_queue_rq: pointer to &struct request that gets filled in when
1775 * another request associated with @q is found on the plug list
1776 * (optional, may be %NULL)
1778 * Determine whether @bio being queued on @q can be merged with a request
1779 * on %current's plugged list. Returns %true if merge was successful,
1782 * Plugging coalesces IOs from the same issuer for the same purpose without
1783 * going through @q->queue_lock. As such it's more of an issuing mechanism
1784 * than scheduling, and the request, while may have elvpriv data, is not
1785 * added on the elevator at this point. In addition, we don't have
1786 * reliable access to the elevator outside queue lock. Only check basic
1787 * merging parameters without querying the elevator.
1789 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1791 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1792 unsigned int *request_count
,
1793 struct request
**same_queue_rq
)
1795 struct blk_plug
*plug
;
1797 struct list_head
*plug_list
;
1799 plug
= current
->plug
;
1805 plug_list
= &plug
->mq_list
;
1807 plug_list
= &plug
->list
;
1809 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1810 bool merged
= false;
1815 * Only blk-mq multiple hardware queues case checks the
1816 * rq in the same queue, there should be only one such
1820 *same_queue_rq
= rq
;
1823 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1826 switch (blk_try_merge(rq
, bio
)) {
1827 case ELEVATOR_BACK_MERGE
:
1828 merged
= bio_attempt_back_merge(q
, rq
, bio
);
1830 case ELEVATOR_FRONT_MERGE
:
1831 merged
= bio_attempt_front_merge(q
, rq
, bio
);
1833 case ELEVATOR_DISCARD_MERGE
:
1834 merged
= bio_attempt_discard_merge(q
, rq
, bio
);
1847 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1849 struct blk_plug
*plug
;
1851 struct list_head
*plug_list
;
1852 unsigned int ret
= 0;
1854 plug
= current
->plug
;
1859 plug_list
= &plug
->mq_list
;
1861 plug_list
= &plug
->list
;
1863 list_for_each_entry(rq
, plug_list
, queuelist
) {
1871 void blk_init_request_from_bio(struct request
*req
, struct bio
*bio
)
1873 struct io_context
*ioc
= rq_ioc(bio
);
1875 if (bio
->bi_opf
& REQ_RAHEAD
)
1876 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1878 req
->__sector
= bio
->bi_iter
.bi_sector
;
1879 if (ioprio_valid(bio_prio(bio
)))
1880 req
->ioprio
= bio_prio(bio
);
1882 req
->ioprio
= ioc
->ioprio
;
1884 req
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE
, 0);
1885 req
->write_hint
= bio
->bi_write_hint
;
1886 blk_rq_bio_prep(req
->q
, req
, bio
);
1888 EXPORT_SYMBOL_GPL(blk_init_request_from_bio
);
1890 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1892 struct blk_plug
*plug
;
1893 int where
= ELEVATOR_INSERT_SORT
;
1894 struct request
*req
, *free
;
1895 unsigned int request_count
= 0;
1896 unsigned int wb_acct
;
1899 * low level driver can indicate that it wants pages above a
1900 * certain limit bounced to low memory (ie for highmem, or even
1901 * ISA dma in theory)
1903 blk_queue_bounce(q
, &bio
);
1905 blk_queue_split(q
, &bio
);
1907 if (!bio_integrity_prep(bio
))
1908 return BLK_QC_T_NONE
;
1910 if (op_is_flush(bio
->bi_opf
)) {
1911 spin_lock_irq(q
->queue_lock
);
1912 where
= ELEVATOR_INSERT_FLUSH
;
1917 * Check if we can merge with the plugged list before grabbing
1920 if (!blk_queue_nomerges(q
)) {
1921 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1922 return BLK_QC_T_NONE
;
1924 request_count
= blk_plug_queued_count(q
);
1926 spin_lock_irq(q
->queue_lock
);
1928 switch (elv_merge(q
, &req
, bio
)) {
1929 case ELEVATOR_BACK_MERGE
:
1930 if (!bio_attempt_back_merge(q
, req
, bio
))
1932 elv_bio_merged(q
, req
, bio
);
1933 free
= attempt_back_merge(q
, req
);
1935 __blk_put_request(q
, free
);
1937 elv_merged_request(q
, req
, ELEVATOR_BACK_MERGE
);
1939 case ELEVATOR_FRONT_MERGE
:
1940 if (!bio_attempt_front_merge(q
, req
, bio
))
1942 elv_bio_merged(q
, req
, bio
);
1943 free
= attempt_front_merge(q
, req
);
1945 __blk_put_request(q
, free
);
1947 elv_merged_request(q
, req
, ELEVATOR_FRONT_MERGE
);
1954 wb_acct
= wbt_wait(q
->rq_wb
, bio
, q
->queue_lock
);
1957 * Grab a free request. This is might sleep but can not fail.
1958 * Returns with the queue unlocked.
1960 blk_queue_enter_live(q
);
1961 req
= get_request(q
, bio
->bi_opf
, bio
, 0);
1964 __wbt_done(q
->rq_wb
, wb_acct
);
1965 if (PTR_ERR(req
) == -ENOMEM
)
1966 bio
->bi_status
= BLK_STS_RESOURCE
;
1968 bio
->bi_status
= BLK_STS_IOERR
;
1973 wbt_track(&req
->issue_stat
, wb_acct
);
1976 * After dropping the lock and possibly sleeping here, our request
1977 * may now be mergeable after it had proven unmergeable (above).
1978 * We don't worry about that case for efficiency. It won't happen
1979 * often, and the elevators are able to handle it.
1981 blk_init_request_from_bio(req
, bio
);
1983 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1984 req
->cpu
= raw_smp_processor_id();
1986 plug
= current
->plug
;
1989 * If this is the first request added after a plug, fire
1992 * @request_count may become stale because of schedule
1993 * out, so check plug list again.
1995 if (!request_count
|| list_empty(&plug
->list
))
1996 trace_block_plug(q
);
1998 struct request
*last
= list_entry_rq(plug
->list
.prev
);
1999 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
2000 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
2001 blk_flush_plug_list(plug
, false);
2002 trace_block_plug(q
);
2005 list_add_tail(&req
->queuelist
, &plug
->list
);
2006 blk_account_io_start(req
, true);
2008 spin_lock_irq(q
->queue_lock
);
2009 add_acct_request(q
, req
, where
);
2012 spin_unlock_irq(q
->queue_lock
);
2015 return BLK_QC_T_NONE
;
2018 static void handle_bad_sector(struct bio
*bio
)
2020 char b
[BDEVNAME_SIZE
];
2022 printk(KERN_INFO
"attempt to access beyond end of device\n");
2023 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
2024 bio_devname(bio
, b
), bio
->bi_opf
,
2025 (unsigned long long)bio_end_sector(bio
),
2026 (long long)get_capacity(bio
->bi_disk
));
2029 #ifdef CONFIG_FAIL_MAKE_REQUEST
2031 static DECLARE_FAULT_ATTR(fail_make_request
);
2033 static int __init
setup_fail_make_request(char *str
)
2035 return setup_fault_attr(&fail_make_request
, str
);
2037 __setup("fail_make_request=", setup_fail_make_request
);
2039 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
2041 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
2044 static int __init
fail_make_request_debugfs(void)
2046 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
2047 NULL
, &fail_make_request
);
2049 return PTR_ERR_OR_ZERO(dir
);
2052 late_initcall(fail_make_request_debugfs
);
2054 #else /* CONFIG_FAIL_MAKE_REQUEST */
2056 static inline bool should_fail_request(struct hd_struct
*part
,
2062 #endif /* CONFIG_FAIL_MAKE_REQUEST */
2065 * Remap block n of partition p to block n+start(p) of the disk.
2067 static inline int blk_partition_remap(struct bio
*bio
)
2069 struct hd_struct
*p
;
2073 * Zone reset does not include bi_size so bio_sectors() is always 0.
2074 * Include a test for the reset op code and perform the remap if needed.
2076 if (!bio
->bi_partno
||
2077 (!bio_sectors(bio
) && bio_op(bio
) != REQ_OP_ZONE_RESET
))
2081 p
= __disk_get_part(bio
->bi_disk
, bio
->bi_partno
);
2082 if (likely(p
&& !should_fail_request(p
, bio
->bi_iter
.bi_size
))) {
2083 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
2085 trace_block_bio_remap(bio
->bi_disk
->queue
, bio
, part_devt(p
),
2086 bio
->bi_iter
.bi_sector
- p
->start_sect
);
2088 printk("%s: fail for partition %d\n", __func__
, bio
->bi_partno
);
2097 * Check whether this bio extends beyond the end of the device.
2099 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
2106 /* Test device or partition size, when known. */
2107 maxsector
= get_capacity(bio
->bi_disk
);
2109 sector_t sector
= bio
->bi_iter
.bi_sector
;
2111 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
2113 * This may well happen - the kernel calls bread()
2114 * without checking the size of the device, e.g., when
2115 * mounting a device.
2117 handle_bad_sector(bio
);
2125 static noinline_for_stack
bool
2126 generic_make_request_checks(struct bio
*bio
)
2128 struct request_queue
*q
;
2129 int nr_sectors
= bio_sectors(bio
);
2130 blk_status_t status
= BLK_STS_IOERR
;
2131 char b
[BDEVNAME_SIZE
];
2135 if (bio_check_eod(bio
, nr_sectors
))
2138 q
= bio
->bi_disk
->queue
;
2141 "generic_make_request: Trying to access "
2142 "nonexistent block-device %s (%Lu)\n",
2143 bio_devname(bio
, b
), (long long)bio
->bi_iter
.bi_sector
);
2148 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2149 * if queue is not a request based queue.
2152 if ((bio
->bi_opf
& REQ_NOWAIT
) && !queue_is_rq_based(q
))
2155 if (should_fail_request(&bio
->bi_disk
->part0
, bio
->bi_iter
.bi_size
))
2158 if (blk_partition_remap(bio
))
2161 if (bio_check_eod(bio
, nr_sectors
))
2165 * Filter flush bio's early so that make_request based
2166 * drivers without flush support don't have to worry
2169 if (op_is_flush(bio
->bi_opf
) &&
2170 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
2171 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
2173 status
= BLK_STS_OK
;
2178 switch (bio_op(bio
)) {
2179 case REQ_OP_DISCARD
:
2180 if (!blk_queue_discard(q
))
2183 case REQ_OP_SECURE_ERASE
:
2184 if (!blk_queue_secure_erase(q
))
2187 case REQ_OP_WRITE_SAME
:
2188 if (!q
->limits
.max_write_same_sectors
)
2191 case REQ_OP_ZONE_REPORT
:
2192 case REQ_OP_ZONE_RESET
:
2193 if (!blk_queue_is_zoned(q
))
2196 case REQ_OP_WRITE_ZEROES
:
2197 if (!q
->limits
.max_write_zeroes_sectors
)
2205 * Various block parts want %current->io_context and lazy ioc
2206 * allocation ends up trading a lot of pain for a small amount of
2207 * memory. Just allocate it upfront. This may fail and block
2208 * layer knows how to live with it.
2210 create_io_context(GFP_ATOMIC
, q
->node
);
2212 if (!blkcg_bio_issue_check(q
, bio
))
2215 if (!bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
2216 trace_block_bio_queue(q
, bio
);
2217 /* Now that enqueuing has been traced, we need to trace
2218 * completion as well.
2220 bio_set_flag(bio
, BIO_TRACE_COMPLETION
);
2225 status
= BLK_STS_NOTSUPP
;
2227 bio
->bi_status
= status
;
2233 * generic_make_request - hand a buffer to its device driver for I/O
2234 * @bio: The bio describing the location in memory and on the device.
2236 * generic_make_request() is used to make I/O requests of block
2237 * devices. It is passed a &struct bio, which describes the I/O that needs
2240 * generic_make_request() does not return any status. The
2241 * success/failure status of the request, along with notification of
2242 * completion, is delivered asynchronously through the bio->bi_end_io
2243 * function described (one day) else where.
2245 * The caller of generic_make_request must make sure that bi_io_vec
2246 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2247 * set to describe the device address, and the
2248 * bi_end_io and optionally bi_private are set to describe how
2249 * completion notification should be signaled.
2251 * generic_make_request and the drivers it calls may use bi_next if this
2252 * bio happens to be merged with someone else, and may resubmit the bio to
2253 * a lower device by calling into generic_make_request recursively, which
2254 * means the bio should NOT be touched after the call to ->make_request_fn.
2256 blk_qc_t
generic_make_request(struct bio
*bio
)
2259 * bio_list_on_stack[0] contains bios submitted by the current
2261 * bio_list_on_stack[1] contains bios that were submitted before
2262 * the current make_request_fn, but that haven't been processed
2265 struct bio_list bio_list_on_stack
[2];
2266 blk_qc_t ret
= BLK_QC_T_NONE
;
2268 if (!generic_make_request_checks(bio
))
2272 * We only want one ->make_request_fn to be active at a time, else
2273 * stack usage with stacked devices could be a problem. So use
2274 * current->bio_list to keep a list of requests submited by a
2275 * make_request_fn function. current->bio_list is also used as a
2276 * flag to say if generic_make_request is currently active in this
2277 * task or not. If it is NULL, then no make_request is active. If
2278 * it is non-NULL, then a make_request is active, and new requests
2279 * should be added at the tail
2281 if (current
->bio_list
) {
2282 bio_list_add(¤t
->bio_list
[0], bio
);
2286 /* following loop may be a bit non-obvious, and so deserves some
2288 * Before entering the loop, bio->bi_next is NULL (as all callers
2289 * ensure that) so we have a list with a single bio.
2290 * We pretend that we have just taken it off a longer list, so
2291 * we assign bio_list to a pointer to the bio_list_on_stack,
2292 * thus initialising the bio_list of new bios to be
2293 * added. ->make_request() may indeed add some more bios
2294 * through a recursive call to generic_make_request. If it
2295 * did, we find a non-NULL value in bio_list and re-enter the loop
2296 * from the top. In this case we really did just take the bio
2297 * of the top of the list (no pretending) and so remove it from
2298 * bio_list, and call into ->make_request() again.
2300 BUG_ON(bio
->bi_next
);
2301 bio_list_init(&bio_list_on_stack
[0]);
2302 current
->bio_list
= bio_list_on_stack
;
2304 struct request_queue
*q
= bio
->bi_disk
->queue
;
2305 blk_mq_req_flags_t flags
= bio
->bi_opf
& REQ_NOWAIT
?
2306 BLK_MQ_REQ_NOWAIT
: 0;
2308 if (likely(blk_queue_enter(q
, flags
) == 0)) {
2309 struct bio_list lower
, same
;
2311 /* Create a fresh bio_list for all subordinate requests */
2312 bio_list_on_stack
[1] = bio_list_on_stack
[0];
2313 bio_list_init(&bio_list_on_stack
[0]);
2314 ret
= q
->make_request_fn(q
, bio
);
2318 /* sort new bios into those for a lower level
2319 * and those for the same level
2321 bio_list_init(&lower
);
2322 bio_list_init(&same
);
2323 while ((bio
= bio_list_pop(&bio_list_on_stack
[0])) != NULL
)
2324 if (q
== bio
->bi_disk
->queue
)
2325 bio_list_add(&same
, bio
);
2327 bio_list_add(&lower
, bio
);
2328 /* now assemble so we handle the lowest level first */
2329 bio_list_merge(&bio_list_on_stack
[0], &lower
);
2330 bio_list_merge(&bio_list_on_stack
[0], &same
);
2331 bio_list_merge(&bio_list_on_stack
[0], &bio_list_on_stack
[1]);
2333 if (unlikely(!blk_queue_dying(q
) &&
2334 (bio
->bi_opf
& REQ_NOWAIT
)))
2335 bio_wouldblock_error(bio
);
2339 bio
= bio_list_pop(&bio_list_on_stack
[0]);
2341 current
->bio_list
= NULL
; /* deactivate */
2346 EXPORT_SYMBOL(generic_make_request
);
2349 * direct_make_request - hand a buffer directly to its device driver for I/O
2350 * @bio: The bio describing the location in memory and on the device.
2352 * This function behaves like generic_make_request(), but does not protect
2353 * against recursion. Must only be used if the called driver is known
2354 * to not call generic_make_request (or direct_make_request) again from
2355 * its make_request function. (Calling direct_make_request again from
2356 * a workqueue is perfectly fine as that doesn't recurse).
2358 blk_qc_t
direct_make_request(struct bio
*bio
)
2360 struct request_queue
*q
= bio
->bi_disk
->queue
;
2361 bool nowait
= bio
->bi_opf
& REQ_NOWAIT
;
2364 if (!generic_make_request_checks(bio
))
2365 return BLK_QC_T_NONE
;
2367 if (unlikely(blk_queue_enter(q
, nowait
? BLK_MQ_REQ_NOWAIT
: 0))) {
2368 if (nowait
&& !blk_queue_dying(q
))
2369 bio
->bi_status
= BLK_STS_AGAIN
;
2371 bio
->bi_status
= BLK_STS_IOERR
;
2373 return BLK_QC_T_NONE
;
2376 ret
= q
->make_request_fn(q
, bio
);
2380 EXPORT_SYMBOL_GPL(direct_make_request
);
2383 * submit_bio - submit a bio to the block device layer for I/O
2384 * @bio: The &struct bio which describes the I/O
2386 * submit_bio() is very similar in purpose to generic_make_request(), and
2387 * uses that function to do most of the work. Both are fairly rough
2388 * interfaces; @bio must be presetup and ready for I/O.
2391 blk_qc_t
submit_bio(struct bio
*bio
)
2394 * If it's a regular read/write or a barrier with data attached,
2395 * go through the normal accounting stuff before submission.
2397 if (bio_has_data(bio
)) {
2400 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2401 count
= queue_logical_block_size(bio
->bi_disk
->queue
) >> 9;
2403 count
= bio_sectors(bio
);
2405 if (op_is_write(bio_op(bio
))) {
2406 count_vm_events(PGPGOUT
, count
);
2408 task_io_account_read(bio
->bi_iter
.bi_size
);
2409 count_vm_events(PGPGIN
, count
);
2412 if (unlikely(block_dump
)) {
2413 char b
[BDEVNAME_SIZE
];
2414 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2415 current
->comm
, task_pid_nr(current
),
2416 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2417 (unsigned long long)bio
->bi_iter
.bi_sector
,
2418 bio_devname(bio
, b
), count
);
2422 return generic_make_request(bio
);
2424 EXPORT_SYMBOL(submit_bio
);
2426 bool blk_poll(struct request_queue
*q
, blk_qc_t cookie
)
2428 if (!q
->poll_fn
|| !blk_qc_t_valid(cookie
))
2432 blk_flush_plug_list(current
->plug
, false);
2433 return q
->poll_fn(q
, cookie
);
2435 EXPORT_SYMBOL_GPL(blk_poll
);
2438 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2439 * for new the queue limits
2441 * @rq: the request being checked
2444 * @rq may have been made based on weaker limitations of upper-level queues
2445 * in request stacking drivers, and it may violate the limitation of @q.
2446 * Since the block layer and the underlying device driver trust @rq
2447 * after it is inserted to @q, it should be checked against @q before
2448 * the insertion using this generic function.
2450 * Request stacking drivers like request-based dm may change the queue
2451 * limits when retrying requests on other queues. Those requests need
2452 * to be checked against the new queue limits again during dispatch.
2454 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2457 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2458 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2463 * queue's settings related to segment counting like q->bounce_pfn
2464 * may differ from that of other stacking queues.
2465 * Recalculate it to check the request correctly on this queue's
2468 blk_recalc_rq_segments(rq
);
2469 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2470 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2478 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2479 * @q: the queue to submit the request
2480 * @rq: the request being queued
2482 blk_status_t
blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2484 unsigned long flags
;
2485 int where
= ELEVATOR_INSERT_BACK
;
2487 if (blk_cloned_rq_check_limits(q
, rq
))
2488 return BLK_STS_IOERR
;
2491 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2492 return BLK_STS_IOERR
;
2495 if (blk_queue_io_stat(q
))
2496 blk_account_io_start(rq
, true);
2498 * Since we have a scheduler attached on the top device,
2499 * bypass a potential scheduler on the bottom device for
2502 blk_mq_request_bypass_insert(rq
, true);
2506 spin_lock_irqsave(q
->queue_lock
, flags
);
2507 if (unlikely(blk_queue_dying(q
))) {
2508 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2509 return BLK_STS_IOERR
;
2513 * Submitting request must be dequeued before calling this function
2514 * because it will be linked to another request_queue
2516 BUG_ON(blk_queued_rq(rq
));
2518 if (op_is_flush(rq
->cmd_flags
))
2519 where
= ELEVATOR_INSERT_FLUSH
;
2521 add_acct_request(q
, rq
, where
);
2522 if (where
== ELEVATOR_INSERT_FLUSH
)
2524 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2528 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2531 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2532 * @rq: request to examine
2535 * A request could be merge of IOs which require different failure
2536 * handling. This function determines the number of bytes which
2537 * can be failed from the beginning of the request without
2538 * crossing into area which need to be retried further.
2541 * The number of bytes to fail.
2543 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2545 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2546 unsigned int bytes
= 0;
2549 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2550 return blk_rq_bytes(rq
);
2553 * Currently the only 'mixing' which can happen is between
2554 * different fastfail types. We can safely fail portions
2555 * which have all the failfast bits that the first one has -
2556 * the ones which are at least as eager to fail as the first
2559 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2560 if ((bio
->bi_opf
& ff
) != ff
)
2562 bytes
+= bio
->bi_iter
.bi_size
;
2565 /* this could lead to infinite loop */
2566 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2569 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2571 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2573 if (blk_do_io_stat(req
)) {
2574 const int rw
= rq_data_dir(req
);
2575 struct hd_struct
*part
;
2578 cpu
= part_stat_lock();
2580 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2585 void blk_account_io_done(struct request
*req
)
2588 * Account IO completion. flush_rq isn't accounted as a
2589 * normal IO on queueing nor completion. Accounting the
2590 * containing request is enough.
2592 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2593 unsigned long duration
= jiffies
- req
->start_time
;
2594 const int rw
= rq_data_dir(req
);
2595 struct hd_struct
*part
;
2598 cpu
= part_stat_lock();
2601 part_stat_inc(cpu
, part
, ios
[rw
]);
2602 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2603 part_round_stats(req
->q
, cpu
, part
);
2604 part_dec_in_flight(req
->q
, part
, rw
);
2606 hd_struct_put(part
);
2613 * Don't process normal requests when queue is suspended
2614 * or in the process of suspending/resuming
2616 static bool blk_pm_allow_request(struct request
*rq
)
2618 switch (rq
->q
->rpm_status
) {
2620 case RPM_SUSPENDING
:
2621 return rq
->rq_flags
& RQF_PM
;
2629 static bool blk_pm_allow_request(struct request
*rq
)
2635 void blk_account_io_start(struct request
*rq
, bool new_io
)
2637 struct hd_struct
*part
;
2638 int rw
= rq_data_dir(rq
);
2641 if (!blk_do_io_stat(rq
))
2644 cpu
= part_stat_lock();
2648 part_stat_inc(cpu
, part
, merges
[rw
]);
2650 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2651 if (!hd_struct_try_get(part
)) {
2653 * The partition is already being removed,
2654 * the request will be accounted on the disk only
2656 * We take a reference on disk->part0 although that
2657 * partition will never be deleted, so we can treat
2658 * it as any other partition.
2660 part
= &rq
->rq_disk
->part0
;
2661 hd_struct_get(part
);
2663 part_round_stats(rq
->q
, cpu
, part
);
2664 part_inc_in_flight(rq
->q
, part
, rw
);
2671 static struct request
*elv_next_request(struct request_queue
*q
)
2674 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
2676 WARN_ON_ONCE(q
->mq_ops
);
2679 list_for_each_entry(rq
, &q
->queue_head
, queuelist
) {
2680 if (blk_pm_allow_request(rq
))
2683 if (rq
->rq_flags
& RQF_SOFTBARRIER
)
2688 * Flush request is running and flush request isn't queueable
2689 * in the drive, we can hold the queue till flush request is
2690 * finished. Even we don't do this, driver can't dispatch next
2691 * requests and will requeue them. And this can improve
2692 * throughput too. For example, we have request flush1, write1,
2693 * flush 2. flush1 is dispatched, then queue is hold, write1
2694 * isn't inserted to queue. After flush1 is finished, flush2
2695 * will be dispatched. Since disk cache is already clean,
2696 * flush2 will be finished very soon, so looks like flush2 is
2698 * Since the queue is hold, a flag is set to indicate the queue
2699 * should be restarted later. Please see flush_end_io() for
2702 if (fq
->flush_pending_idx
!= fq
->flush_running_idx
&&
2703 !queue_flush_queueable(q
)) {
2704 fq
->flush_queue_delayed
= 1;
2707 if (unlikely(blk_queue_bypass(q
)) ||
2708 !q
->elevator
->type
->ops
.sq
.elevator_dispatch_fn(q
, 0))
2714 * blk_peek_request - peek at the top of a request queue
2715 * @q: request queue to peek at
2718 * Return the request at the top of @q. The returned request
2719 * should be started using blk_start_request() before LLD starts
2723 * Pointer to the request at the top of @q if available. Null
2726 struct request
*blk_peek_request(struct request_queue
*q
)
2731 lockdep_assert_held(q
->queue_lock
);
2732 WARN_ON_ONCE(q
->mq_ops
);
2734 while ((rq
= elv_next_request(q
)) != NULL
) {
2735 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2737 * This is the first time the device driver
2738 * sees this request (possibly after
2739 * requeueing). Notify IO scheduler.
2741 if (rq
->rq_flags
& RQF_SORTED
)
2742 elv_activate_rq(q
, rq
);
2745 * just mark as started even if we don't start
2746 * it, a request that has been delayed should
2747 * not be passed by new incoming requests
2749 rq
->rq_flags
|= RQF_STARTED
;
2750 trace_block_rq_issue(q
, rq
);
2753 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2754 q
->end_sector
= rq_end_sector(rq
);
2755 q
->boundary_rq
= NULL
;
2758 if (rq
->rq_flags
& RQF_DONTPREP
)
2761 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2763 * make sure space for the drain appears we
2764 * know we can do this because max_hw_segments
2765 * has been adjusted to be one fewer than the
2768 rq
->nr_phys_segments
++;
2774 ret
= q
->prep_rq_fn(q
, rq
);
2775 if (ret
== BLKPREP_OK
) {
2777 } else if (ret
== BLKPREP_DEFER
) {
2779 * the request may have been (partially) prepped.
2780 * we need to keep this request in the front to
2781 * avoid resource deadlock. RQF_STARTED will
2782 * prevent other fs requests from passing this one.
2784 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2785 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2787 * remove the space for the drain we added
2788 * so that we don't add it again
2790 --rq
->nr_phys_segments
;
2795 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2796 rq
->rq_flags
|= RQF_QUIET
;
2798 * Mark this request as started so we don't trigger
2799 * any debug logic in the end I/O path.
2801 blk_start_request(rq
);
2802 __blk_end_request_all(rq
, ret
== BLKPREP_INVALID
?
2803 BLK_STS_TARGET
: BLK_STS_IOERR
);
2805 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2812 EXPORT_SYMBOL(blk_peek_request
);
2814 static void blk_dequeue_request(struct request
*rq
)
2816 struct request_queue
*q
= rq
->q
;
2818 BUG_ON(list_empty(&rq
->queuelist
));
2819 BUG_ON(ELV_ON_HASH(rq
));
2821 list_del_init(&rq
->queuelist
);
2824 * the time frame between a request being removed from the lists
2825 * and to it is freed is accounted as io that is in progress at
2828 if (blk_account_rq(rq
)) {
2829 q
->in_flight
[rq_is_sync(rq
)]++;
2830 set_io_start_time_ns(rq
);
2835 * blk_start_request - start request processing on the driver
2836 * @req: request to dequeue
2839 * Dequeue @req and start timeout timer on it. This hands off the
2840 * request to the driver.
2842 void blk_start_request(struct request
*req
)
2844 lockdep_assert_held(req
->q
->queue_lock
);
2845 WARN_ON_ONCE(req
->q
->mq_ops
);
2847 blk_dequeue_request(req
);
2849 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2850 blk_stat_set_issue(&req
->issue_stat
, blk_rq_sectors(req
));
2851 req
->rq_flags
|= RQF_STATS
;
2852 wbt_issue(req
->q
->rq_wb
, &req
->issue_stat
);
2855 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2858 EXPORT_SYMBOL(blk_start_request
);
2861 * blk_fetch_request - fetch a request from a request queue
2862 * @q: request queue to fetch a request from
2865 * Return the request at the top of @q. The request is started on
2866 * return and LLD can start processing it immediately.
2869 * Pointer to the request at the top of @q if available. Null
2872 struct request
*blk_fetch_request(struct request_queue
*q
)
2876 lockdep_assert_held(q
->queue_lock
);
2877 WARN_ON_ONCE(q
->mq_ops
);
2879 rq
= blk_peek_request(q
);
2881 blk_start_request(rq
);
2884 EXPORT_SYMBOL(blk_fetch_request
);
2887 * Steal bios from a request and add them to a bio list.
2888 * The request must not have been partially completed before.
2890 void blk_steal_bios(struct bio_list
*list
, struct request
*rq
)
2894 list
->tail
->bi_next
= rq
->bio
;
2896 list
->head
= rq
->bio
;
2897 list
->tail
= rq
->biotail
;
2905 EXPORT_SYMBOL_GPL(blk_steal_bios
);
2908 * blk_update_request - Special helper function for request stacking drivers
2909 * @req: the request being processed
2910 * @error: block status code
2911 * @nr_bytes: number of bytes to complete @req
2914 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2915 * the request structure even if @req doesn't have leftover.
2916 * If @req has leftover, sets it up for the next range of segments.
2918 * This special helper function is only for request stacking drivers
2919 * (e.g. request-based dm) so that they can handle partial completion.
2920 * Actual device drivers should use blk_end_request instead.
2922 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2923 * %false return from this function.
2926 * %false - this request doesn't have any more data
2927 * %true - this request has more data
2929 bool blk_update_request(struct request
*req
, blk_status_t error
,
2930 unsigned int nr_bytes
)
2934 trace_block_rq_complete(req
, blk_status_to_errno(error
), nr_bytes
);
2939 if (unlikely(error
&& !blk_rq_is_passthrough(req
) &&
2940 !(req
->rq_flags
& RQF_QUIET
)))
2941 print_req_error(req
, error
);
2943 blk_account_io_completion(req
, nr_bytes
);
2947 struct bio
*bio
= req
->bio
;
2948 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2950 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2951 req
->bio
= bio
->bi_next
;
2953 /* Completion has already been traced */
2954 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
2955 req_bio_endio(req
, bio
, bio_bytes
, error
);
2957 total_bytes
+= bio_bytes
;
2958 nr_bytes
-= bio_bytes
;
2969 * Reset counters so that the request stacking driver
2970 * can find how many bytes remain in the request
2973 req
->__data_len
= 0;
2977 req
->__data_len
-= total_bytes
;
2979 /* update sector only for requests with clear definition of sector */
2980 if (!blk_rq_is_passthrough(req
))
2981 req
->__sector
+= total_bytes
>> 9;
2983 /* mixed attributes always follow the first bio */
2984 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2985 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2986 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2989 if (!(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
)) {
2991 * If total number of sectors is less than the first segment
2992 * size, something has gone terribly wrong.
2994 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
2995 blk_dump_rq_flags(req
, "request botched");
2996 req
->__data_len
= blk_rq_cur_bytes(req
);
2999 /* recalculate the number of segments */
3000 blk_recalc_rq_segments(req
);
3005 EXPORT_SYMBOL_GPL(blk_update_request
);
3007 static bool blk_update_bidi_request(struct request
*rq
, blk_status_t error
,
3008 unsigned int nr_bytes
,
3009 unsigned int bidi_bytes
)
3011 if (blk_update_request(rq
, error
, nr_bytes
))
3014 /* Bidi request must be completed as a whole */
3015 if (unlikely(blk_bidi_rq(rq
)) &&
3016 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
3019 if (blk_queue_add_random(rq
->q
))
3020 add_disk_randomness(rq
->rq_disk
);
3026 * blk_unprep_request - unprepare a request
3029 * This function makes a request ready for complete resubmission (or
3030 * completion). It happens only after all error handling is complete,
3031 * so represents the appropriate moment to deallocate any resources
3032 * that were allocated to the request in the prep_rq_fn. The queue
3033 * lock is held when calling this.
3035 void blk_unprep_request(struct request
*req
)
3037 struct request_queue
*q
= req
->q
;
3039 req
->rq_flags
&= ~RQF_DONTPREP
;
3040 if (q
->unprep_rq_fn
)
3041 q
->unprep_rq_fn(q
, req
);
3043 EXPORT_SYMBOL_GPL(blk_unprep_request
);
3045 void blk_finish_request(struct request
*req
, blk_status_t error
)
3047 struct request_queue
*q
= req
->q
;
3049 lockdep_assert_held(req
->q
->queue_lock
);
3050 WARN_ON_ONCE(q
->mq_ops
);
3052 if (req
->rq_flags
& RQF_STATS
)
3055 if (req
->rq_flags
& RQF_QUEUED
)
3056 blk_queue_end_tag(q
, req
);
3058 BUG_ON(blk_queued_rq(req
));
3060 if (unlikely(laptop_mode
) && !blk_rq_is_passthrough(req
))
3061 laptop_io_completion(req
->q
->backing_dev_info
);
3063 blk_delete_timer(req
);
3065 if (req
->rq_flags
& RQF_DONTPREP
)
3066 blk_unprep_request(req
);
3068 blk_account_io_done(req
);
3071 wbt_done(req
->q
->rq_wb
, &req
->issue_stat
);
3072 req
->end_io(req
, error
);
3074 if (blk_bidi_rq(req
))
3075 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
3077 __blk_put_request(q
, req
);
3080 EXPORT_SYMBOL(blk_finish_request
);
3083 * blk_end_bidi_request - Complete a bidi request
3084 * @rq: the request to complete
3085 * @error: block status code
3086 * @nr_bytes: number of bytes to complete @rq
3087 * @bidi_bytes: number of bytes to complete @rq->next_rq
3090 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
3091 * Drivers that supports bidi can safely call this member for any
3092 * type of request, bidi or uni. In the later case @bidi_bytes is
3096 * %false - we are done with this request
3097 * %true - still buffers pending for this request
3099 static bool blk_end_bidi_request(struct request
*rq
, blk_status_t error
,
3100 unsigned int nr_bytes
, unsigned int bidi_bytes
)
3102 struct request_queue
*q
= rq
->q
;
3103 unsigned long flags
;
3105 WARN_ON_ONCE(q
->mq_ops
);
3107 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
3110 spin_lock_irqsave(q
->queue_lock
, flags
);
3111 blk_finish_request(rq
, error
);
3112 spin_unlock_irqrestore(q
->queue_lock
, flags
);
3118 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3119 * @rq: the request to complete
3120 * @error: block status code
3121 * @nr_bytes: number of bytes to complete @rq
3122 * @bidi_bytes: number of bytes to complete @rq->next_rq
3125 * Identical to blk_end_bidi_request() except that queue lock is
3126 * assumed to be locked on entry and remains so on return.
3129 * %false - we are done with this request
3130 * %true - still buffers pending for this request
3132 static bool __blk_end_bidi_request(struct request
*rq
, blk_status_t error
,
3133 unsigned int nr_bytes
, unsigned int bidi_bytes
)
3135 lockdep_assert_held(rq
->q
->queue_lock
);
3136 WARN_ON_ONCE(rq
->q
->mq_ops
);
3138 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
3141 blk_finish_request(rq
, error
);
3147 * blk_end_request - Helper function for drivers to complete the request.
3148 * @rq: the request being processed
3149 * @error: block status code
3150 * @nr_bytes: number of bytes to complete
3153 * Ends I/O on a number of bytes attached to @rq.
3154 * If @rq has leftover, sets it up for the next range of segments.
3157 * %false - we are done with this request
3158 * %true - still buffers pending for this request
3160 bool blk_end_request(struct request
*rq
, blk_status_t error
,
3161 unsigned int nr_bytes
)
3163 WARN_ON_ONCE(rq
->q
->mq_ops
);
3164 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
3166 EXPORT_SYMBOL(blk_end_request
);
3169 * blk_end_request_all - Helper function for drives to finish the request.
3170 * @rq: the request to finish
3171 * @error: block status code
3174 * Completely finish @rq.
3176 void blk_end_request_all(struct request
*rq
, blk_status_t error
)
3179 unsigned int bidi_bytes
= 0;
3181 if (unlikely(blk_bidi_rq(rq
)))
3182 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
3184 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
3187 EXPORT_SYMBOL(blk_end_request_all
);
3190 * __blk_end_request - Helper function for drivers to complete the request.
3191 * @rq: the request being processed
3192 * @error: block status code
3193 * @nr_bytes: number of bytes to complete
3196 * Must be called with queue lock held unlike blk_end_request().
3199 * %false - we are done with this request
3200 * %true - still buffers pending for this request
3202 bool __blk_end_request(struct request
*rq
, blk_status_t error
,
3203 unsigned int nr_bytes
)
3205 lockdep_assert_held(rq
->q
->queue_lock
);
3206 WARN_ON_ONCE(rq
->q
->mq_ops
);
3208 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
3210 EXPORT_SYMBOL(__blk_end_request
);
3213 * __blk_end_request_all - Helper function for drives to finish the request.
3214 * @rq: the request to finish
3215 * @error: block status code
3218 * Completely finish @rq. Must be called with queue lock held.
3220 void __blk_end_request_all(struct request
*rq
, blk_status_t error
)
3223 unsigned int bidi_bytes
= 0;
3225 lockdep_assert_held(rq
->q
->queue_lock
);
3226 WARN_ON_ONCE(rq
->q
->mq_ops
);
3228 if (unlikely(blk_bidi_rq(rq
)))
3229 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
3231 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
3234 EXPORT_SYMBOL(__blk_end_request_all
);
3237 * __blk_end_request_cur - Helper function to finish the current request chunk.
3238 * @rq: the request to finish the current chunk for
3239 * @error: block status code
3242 * Complete the current consecutively mapped chunk from @rq. Must
3243 * be called with queue lock held.
3246 * %false - we are done with this request
3247 * %true - still buffers pending for this request
3249 bool __blk_end_request_cur(struct request
*rq
, blk_status_t error
)
3251 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
3253 EXPORT_SYMBOL(__blk_end_request_cur
);
3255 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
3258 if (bio_has_data(bio
))
3259 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3260 else if (bio_op(bio
) == REQ_OP_DISCARD
)
3261 rq
->nr_phys_segments
= 1;
3263 rq
->__data_len
= bio
->bi_iter
.bi_size
;
3264 rq
->bio
= rq
->biotail
= bio
;
3267 rq
->rq_disk
= bio
->bi_disk
;
3270 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3272 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3273 * @rq: the request to be flushed
3276 * Flush all pages in @rq.
3278 void rq_flush_dcache_pages(struct request
*rq
)
3280 struct req_iterator iter
;
3281 struct bio_vec bvec
;
3283 rq_for_each_segment(bvec
, rq
, iter
)
3284 flush_dcache_page(bvec
.bv_page
);
3286 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
3290 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3291 * @q : the queue of the device being checked
3294 * Check if underlying low-level drivers of a device are busy.
3295 * If the drivers want to export their busy state, they must set own
3296 * exporting function using blk_queue_lld_busy() first.
3298 * Basically, this function is used only by request stacking drivers
3299 * to stop dispatching requests to underlying devices when underlying
3300 * devices are busy. This behavior helps more I/O merging on the queue
3301 * of the request stacking driver and prevents I/O throughput regression
3302 * on burst I/O load.
3305 * 0 - Not busy (The request stacking driver should dispatch request)
3306 * 1 - Busy (The request stacking driver should stop dispatching request)
3308 int blk_lld_busy(struct request_queue
*q
)
3311 return q
->lld_busy_fn(q
);
3315 EXPORT_SYMBOL_GPL(blk_lld_busy
);
3318 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3319 * @rq: the clone request to be cleaned up
3322 * Free all bios in @rq for a cloned request.
3324 void blk_rq_unprep_clone(struct request
*rq
)
3328 while ((bio
= rq
->bio
) != NULL
) {
3329 rq
->bio
= bio
->bi_next
;
3334 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3337 * Copy attributes of the original request to the clone request.
3338 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3340 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3342 dst
->cpu
= src
->cpu
;
3343 dst
->__sector
= blk_rq_pos(src
);
3344 dst
->__data_len
= blk_rq_bytes(src
);
3345 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3346 dst
->ioprio
= src
->ioprio
;
3347 dst
->extra_len
= src
->extra_len
;
3351 * blk_rq_prep_clone - Helper function to setup clone request
3352 * @rq: the request to be setup
3353 * @rq_src: original request to be cloned
3354 * @bs: bio_set that bios for clone are allocated from
3355 * @gfp_mask: memory allocation mask for bio
3356 * @bio_ctr: setup function to be called for each clone bio.
3357 * Returns %0 for success, non %0 for failure.
3358 * @data: private data to be passed to @bio_ctr
3361 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3362 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3363 * are not copied, and copying such parts is the caller's responsibility.
3364 * Also, pages which the original bios are pointing to are not copied
3365 * and the cloned bios just point same pages.
3366 * So cloned bios must be completed before original bios, which means
3367 * the caller must complete @rq before @rq_src.
3369 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3370 struct bio_set
*bs
, gfp_t gfp_mask
,
3371 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3374 struct bio
*bio
, *bio_src
;
3379 __rq_for_each_bio(bio_src
, rq_src
) {
3380 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3384 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3388 rq
->biotail
->bi_next
= bio
;
3391 rq
->bio
= rq
->biotail
= bio
;
3394 __blk_rq_prep_clone(rq
, rq_src
);
3401 blk_rq_unprep_clone(rq
);
3405 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3407 int kblockd_schedule_work(struct work_struct
*work
)
3409 return queue_work(kblockd_workqueue
, work
);
3411 EXPORT_SYMBOL(kblockd_schedule_work
);
3413 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3415 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3417 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3419 int kblockd_mod_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3420 unsigned long delay
)
3422 return mod_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3424 EXPORT_SYMBOL(kblockd_mod_delayed_work_on
);
3426 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3427 unsigned long delay
)
3429 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3431 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3433 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3434 unsigned long delay
)
3436 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3438 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3441 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3442 * @plug: The &struct blk_plug that needs to be initialized
3445 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3446 * pending I/O should the task end up blocking between blk_start_plug() and
3447 * blk_finish_plug(). This is important from a performance perspective, but
3448 * also ensures that we don't deadlock. For instance, if the task is blocking
3449 * for a memory allocation, memory reclaim could end up wanting to free a
3450 * page belonging to that request that is currently residing in our private
3451 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3452 * this kind of deadlock.
3454 void blk_start_plug(struct blk_plug
*plug
)
3456 struct task_struct
*tsk
= current
;
3459 * If this is a nested plug, don't actually assign it.
3464 INIT_LIST_HEAD(&plug
->list
);
3465 INIT_LIST_HEAD(&plug
->mq_list
);
3466 INIT_LIST_HEAD(&plug
->cb_list
);
3468 * Store ordering should not be needed here, since a potential
3469 * preempt will imply a full memory barrier
3473 EXPORT_SYMBOL(blk_start_plug
);
3475 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3477 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3478 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3480 return !(rqa
->q
< rqb
->q
||
3481 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3485 * If 'from_schedule' is true, then postpone the dispatch of requests
3486 * until a safe kblockd context. We due this to avoid accidental big
3487 * additional stack usage in driver dispatch, in places where the originally
3488 * plugger did not intend it.
3490 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3492 __releases(q
->queue_lock
)
3494 lockdep_assert_held(q
->queue_lock
);
3496 trace_block_unplug(q
, depth
, !from_schedule
);
3499 blk_run_queue_async(q
);
3502 spin_unlock(q
->queue_lock
);
3505 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3507 LIST_HEAD(callbacks
);
3509 while (!list_empty(&plug
->cb_list
)) {
3510 list_splice_init(&plug
->cb_list
, &callbacks
);
3512 while (!list_empty(&callbacks
)) {
3513 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3516 list_del(&cb
->list
);
3517 cb
->callback(cb
, from_schedule
);
3522 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3525 struct blk_plug
*plug
= current
->plug
;
3526 struct blk_plug_cb
*cb
;
3531 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3532 if (cb
->callback
== unplug
&& cb
->data
== data
)
3535 /* Not currently on the callback list */
3536 BUG_ON(size
< sizeof(*cb
));
3537 cb
= kzalloc(size
, GFP_ATOMIC
);
3540 cb
->callback
= unplug
;
3541 list_add(&cb
->list
, &plug
->cb_list
);
3545 EXPORT_SYMBOL(blk_check_plugged
);
3547 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3549 struct request_queue
*q
;
3550 unsigned long flags
;
3555 flush_plug_callbacks(plug
, from_schedule
);
3557 if (!list_empty(&plug
->mq_list
))
3558 blk_mq_flush_plug_list(plug
, from_schedule
);
3560 if (list_empty(&plug
->list
))
3563 list_splice_init(&plug
->list
, &list
);
3565 list_sort(NULL
, &list
, plug_rq_cmp
);
3571 * Save and disable interrupts here, to avoid doing it for every
3572 * queue lock we have to take.
3574 local_irq_save(flags
);
3575 while (!list_empty(&list
)) {
3576 rq
= list_entry_rq(list
.next
);
3577 list_del_init(&rq
->queuelist
);
3581 * This drops the queue lock
3584 queue_unplugged(q
, depth
, from_schedule
);
3587 spin_lock(q
->queue_lock
);
3591 * Short-circuit if @q is dead
3593 if (unlikely(blk_queue_dying(q
))) {
3594 __blk_end_request_all(rq
, BLK_STS_IOERR
);
3599 * rq is already accounted, so use raw insert
3601 if (op_is_flush(rq
->cmd_flags
))
3602 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3604 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3610 * This drops the queue lock
3613 queue_unplugged(q
, depth
, from_schedule
);
3615 local_irq_restore(flags
);
3618 void blk_finish_plug(struct blk_plug
*plug
)
3620 if (plug
!= current
->plug
)
3622 blk_flush_plug_list(plug
, false);
3624 current
->plug
= NULL
;
3626 EXPORT_SYMBOL(blk_finish_plug
);
3630 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3631 * @q: the queue of the device
3632 * @dev: the device the queue belongs to
3635 * Initialize runtime-PM-related fields for @q and start auto suspend for
3636 * @dev. Drivers that want to take advantage of request-based runtime PM
3637 * should call this function after @dev has been initialized, and its
3638 * request queue @q has been allocated, and runtime PM for it can not happen
3639 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3640 * cases, driver should call this function before any I/O has taken place.
3642 * This function takes care of setting up using auto suspend for the device,
3643 * the autosuspend delay is set to -1 to make runtime suspend impossible
3644 * until an updated value is either set by user or by driver. Drivers do
3645 * not need to touch other autosuspend settings.
3647 * The block layer runtime PM is request based, so only works for drivers
3648 * that use request as their IO unit instead of those directly use bio's.
3650 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3652 /* not support for RQF_PM and ->rpm_status in blk-mq yet */
3657 q
->rpm_status
= RPM_ACTIVE
;
3658 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3659 pm_runtime_use_autosuspend(q
->dev
);
3661 EXPORT_SYMBOL(blk_pm_runtime_init
);
3664 * blk_pre_runtime_suspend - Pre runtime suspend check
3665 * @q: the queue of the device
3668 * This function will check if runtime suspend is allowed for the device
3669 * by examining if there are any requests pending in the queue. If there
3670 * are requests pending, the device can not be runtime suspended; otherwise,
3671 * the queue's status will be updated to SUSPENDING and the driver can
3672 * proceed to suspend the device.
3674 * For the not allowed case, we mark last busy for the device so that
3675 * runtime PM core will try to autosuspend it some time later.
3677 * This function should be called near the start of the device's
3678 * runtime_suspend callback.
3681 * 0 - OK to runtime suspend the device
3682 * -EBUSY - Device should not be runtime suspended
3684 int blk_pre_runtime_suspend(struct request_queue
*q
)
3691 spin_lock_irq(q
->queue_lock
);
3692 if (q
->nr_pending
) {
3694 pm_runtime_mark_last_busy(q
->dev
);
3696 q
->rpm_status
= RPM_SUSPENDING
;
3698 spin_unlock_irq(q
->queue_lock
);
3701 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3704 * blk_post_runtime_suspend - Post runtime suspend processing
3705 * @q: the queue of the device
3706 * @err: return value of the device's runtime_suspend function
3709 * Update the queue's runtime status according to the return value of the
3710 * device's runtime suspend function and mark last busy for the device so
3711 * that PM core will try to auto suspend the device at a later time.
3713 * This function should be called near the end of the device's
3714 * runtime_suspend callback.
3716 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3721 spin_lock_irq(q
->queue_lock
);
3723 q
->rpm_status
= RPM_SUSPENDED
;
3725 q
->rpm_status
= RPM_ACTIVE
;
3726 pm_runtime_mark_last_busy(q
->dev
);
3728 spin_unlock_irq(q
->queue_lock
);
3730 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3733 * blk_pre_runtime_resume - Pre runtime resume processing
3734 * @q: the queue of the device
3737 * Update the queue's runtime status to RESUMING in preparation for the
3738 * runtime resume of the device.
3740 * This function should be called near the start of the device's
3741 * runtime_resume callback.
3743 void blk_pre_runtime_resume(struct request_queue
*q
)
3748 spin_lock_irq(q
->queue_lock
);
3749 q
->rpm_status
= RPM_RESUMING
;
3750 spin_unlock_irq(q
->queue_lock
);
3752 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3755 * blk_post_runtime_resume - Post runtime resume processing
3756 * @q: the queue of the device
3757 * @err: return value of the device's runtime_resume function
3760 * Update the queue's runtime status according to the return value of the
3761 * device's runtime_resume function. If it is successfully resumed, process
3762 * the requests that are queued into the device's queue when it is resuming
3763 * and then mark last busy and initiate autosuspend for it.
3765 * This function should be called near the end of the device's
3766 * runtime_resume callback.
3768 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3773 spin_lock_irq(q
->queue_lock
);
3775 q
->rpm_status
= RPM_ACTIVE
;
3777 pm_runtime_mark_last_busy(q
->dev
);
3778 pm_request_autosuspend(q
->dev
);
3780 q
->rpm_status
= RPM_SUSPENDED
;
3782 spin_unlock_irq(q
->queue_lock
);
3784 EXPORT_SYMBOL(blk_post_runtime_resume
);
3787 * blk_set_runtime_active - Force runtime status of the queue to be active
3788 * @q: the queue of the device
3790 * If the device is left runtime suspended during system suspend the resume
3791 * hook typically resumes the device and corrects runtime status
3792 * accordingly. However, that does not affect the queue runtime PM status
3793 * which is still "suspended". This prevents processing requests from the
3796 * This function can be used in driver's resume hook to correct queue
3797 * runtime PM status and re-enable peeking requests from the queue. It
3798 * should be called before first request is added to the queue.
3800 void blk_set_runtime_active(struct request_queue
*q
)
3802 spin_lock_irq(q
->queue_lock
);
3803 q
->rpm_status
= RPM_ACTIVE
;
3804 pm_runtime_mark_last_busy(q
->dev
);
3805 pm_request_autosuspend(q
->dev
);
3806 spin_unlock_irq(q
->queue_lock
);
3808 EXPORT_SYMBOL(blk_set_runtime_active
);
3811 int __init
blk_dev_init(void)
3813 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
3814 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3815 FIELD_SIZEOF(struct request
, cmd_flags
));
3816 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3817 FIELD_SIZEOF(struct bio
, bi_opf
));
3819 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3820 kblockd_workqueue
= alloc_workqueue("kblockd",
3821 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3822 if (!kblockd_workqueue
)
3823 panic("Failed to create kblockd\n");
3825 request_cachep
= kmem_cache_create("blkdev_requests",
3826 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3828 blk_requestq_cachep
= kmem_cache_create("request_queue",
3829 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
3831 #ifdef CONFIG_DEBUG_FS
3832 blk_debugfs_root
= debugfs_create_dir("block", NULL
);