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.
708 * No need to quiesce queue if it isn't initialized yet since
709 * blk_freeze_queue() should be enough for cases of passthrough
712 if (q
->mq_ops
&& blk_queue_init_done(q
))
713 blk_mq_quiesce_queue(q
);
715 /* for synchronous bio-based driver finish in-flight integrity i/o */
716 blk_flush_integrity();
718 /* @q won't process any more request, flush async actions */
719 del_timer_sync(&q
->backing_dev_info
->laptop_mode_wb_timer
);
723 blk_mq_free_queue(q
);
724 percpu_ref_exit(&q
->q_usage_counter
);
727 if (q
->queue_lock
!= &q
->__queue_lock
)
728 q
->queue_lock
= &q
->__queue_lock
;
729 spin_unlock_irq(lock
);
731 /* @q is and will stay empty, shutdown and put */
734 EXPORT_SYMBOL(blk_cleanup_queue
);
736 /* Allocate memory local to the request queue */
737 static void *alloc_request_simple(gfp_t gfp_mask
, void *data
)
739 struct request_queue
*q
= data
;
741 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, q
->node
);
744 static void free_request_simple(void *element
, void *data
)
746 kmem_cache_free(request_cachep
, element
);
749 static void *alloc_request_size(gfp_t gfp_mask
, void *data
)
751 struct request_queue
*q
= data
;
754 rq
= kmalloc_node(sizeof(struct request
) + q
->cmd_size
, gfp_mask
,
756 if (rq
&& q
->init_rq_fn
&& q
->init_rq_fn(q
, rq
, gfp_mask
) < 0) {
763 static void free_request_size(void *element
, void *data
)
765 struct request_queue
*q
= data
;
768 q
->exit_rq_fn(q
, element
);
772 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
775 if (unlikely(rl
->rq_pool
) || q
->mq_ops
)
779 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
780 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
781 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
782 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
785 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
786 alloc_request_size
, free_request_size
,
787 q
, gfp_mask
, q
->node
);
789 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
790 alloc_request_simple
, free_request_simple
,
791 q
, gfp_mask
, q
->node
);
796 if (rl
!= &q
->root_rl
)
797 WARN_ON_ONCE(!blk_get_queue(q
));
802 void blk_exit_rl(struct request_queue
*q
, struct request_list
*rl
)
805 mempool_destroy(rl
->rq_pool
);
806 if (rl
!= &q
->root_rl
)
811 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
813 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
815 EXPORT_SYMBOL(blk_alloc_queue
);
818 * blk_queue_enter() - try to increase q->q_usage_counter
819 * @q: request queue pointer
820 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
822 int blk_queue_enter(struct request_queue
*q
, blk_mq_req_flags_t flags
)
824 const bool preempt
= flags
& BLK_MQ_REQ_PREEMPT
;
827 bool success
= false;
830 if (percpu_ref_tryget_live(&q
->q_usage_counter
)) {
832 * The code that sets the PREEMPT_ONLY flag is
833 * responsible for ensuring that that flag is globally
834 * visible before the queue is unfrozen.
836 if (preempt
|| !blk_queue_preempt_only(q
)) {
839 percpu_ref_put(&q
->q_usage_counter
);
847 if (flags
& BLK_MQ_REQ_NOWAIT
)
851 * read pair of barrier in blk_freeze_queue_start(),
852 * we need to order reading __PERCPU_REF_DEAD flag of
853 * .q_usage_counter and reading .mq_freeze_depth or
854 * queue dying flag, otherwise the following wait may
855 * never return if the two reads are reordered.
859 wait_event(q
->mq_freeze_wq
,
860 (atomic_read(&q
->mq_freeze_depth
) == 0 &&
861 (preempt
|| !blk_queue_preempt_only(q
))) ||
863 if (blk_queue_dying(q
))
868 void blk_queue_exit(struct request_queue
*q
)
870 percpu_ref_put(&q
->q_usage_counter
);
873 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
875 struct request_queue
*q
=
876 container_of(ref
, struct request_queue
, q_usage_counter
);
878 wake_up_all(&q
->mq_freeze_wq
);
881 static void blk_rq_timed_out_timer(struct timer_list
*t
)
883 struct request_queue
*q
= from_timer(q
, t
, timeout
);
885 kblockd_schedule_work(&q
->timeout_work
);
888 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
890 struct request_queue
*q
;
892 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
893 gfp_mask
| __GFP_ZERO
, node_id
);
897 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
901 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
);
905 q
->backing_dev_info
= bdi_alloc_node(gfp_mask
, node_id
);
906 if (!q
->backing_dev_info
)
909 q
->stats
= blk_alloc_queue_stats();
913 q
->backing_dev_info
->ra_pages
=
914 (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
915 q
->backing_dev_info
->capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
916 q
->backing_dev_info
->name
= "block";
919 timer_setup(&q
->backing_dev_info
->laptop_mode_wb_timer
,
920 laptop_mode_timer_fn
, 0);
921 timer_setup(&q
->timeout
, blk_rq_timed_out_timer
, 0);
922 INIT_WORK(&q
->timeout_work
, NULL
);
923 INIT_LIST_HEAD(&q
->queue_head
);
924 INIT_LIST_HEAD(&q
->timeout_list
);
925 INIT_LIST_HEAD(&q
->icq_list
);
926 #ifdef CONFIG_BLK_CGROUP
927 INIT_LIST_HEAD(&q
->blkg_list
);
929 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
931 kobject_init(&q
->kobj
, &blk_queue_ktype
);
933 #ifdef CONFIG_BLK_DEV_IO_TRACE
934 mutex_init(&q
->blk_trace_mutex
);
936 mutex_init(&q
->sysfs_lock
);
937 spin_lock_init(&q
->__queue_lock
);
940 * By default initialize queue_lock to internal lock and driver can
941 * override it later if need be.
943 q
->queue_lock
= &q
->__queue_lock
;
946 * A queue starts its life with bypass turned on to avoid
947 * unnecessary bypass on/off overhead and nasty surprises during
948 * init. The initial bypass will be finished when the queue is
949 * registered by blk_register_queue().
952 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
954 init_waitqueue_head(&q
->mq_freeze_wq
);
957 * Init percpu_ref in atomic mode so that it's faster to shutdown.
958 * See blk_register_queue() for details.
960 if (percpu_ref_init(&q
->q_usage_counter
,
961 blk_queue_usage_counter_release
,
962 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
965 if (blkcg_init_queue(q
))
971 percpu_ref_exit(&q
->q_usage_counter
);
973 blk_free_queue_stats(q
->stats
);
975 bdi_put(q
->backing_dev_info
);
977 bioset_free(q
->bio_split
);
979 ida_simple_remove(&blk_queue_ida
, q
->id
);
981 kmem_cache_free(blk_requestq_cachep
, q
);
984 EXPORT_SYMBOL(blk_alloc_queue_node
);
987 * blk_init_queue - prepare a request queue for use with a block device
988 * @rfn: The function to be called to process requests that have been
989 * placed on the queue.
990 * @lock: Request queue spin lock
993 * If a block device wishes to use the standard request handling procedures,
994 * which sorts requests and coalesces adjacent requests, then it must
995 * call blk_init_queue(). The function @rfn will be called when there
996 * are requests on the queue that need to be processed. If the device
997 * supports plugging, then @rfn may not be called immediately when requests
998 * are available on the queue, but may be called at some time later instead.
999 * Plugged queues are generally unplugged when a buffer belonging to one
1000 * of the requests on the queue is needed, or due to memory pressure.
1002 * @rfn is not required, or even expected, to remove all requests off the
1003 * queue, but only as many as it can handle at a time. If it does leave
1004 * requests on the queue, it is responsible for arranging that the requests
1005 * get dealt with eventually.
1007 * The queue spin lock must be held while manipulating the requests on the
1008 * request queue; this lock will be taken also from interrupt context, so irq
1009 * disabling is needed for it.
1011 * Function returns a pointer to the initialized request queue, or %NULL if
1012 * it didn't succeed.
1015 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1016 * when the block device is deactivated (such as at module unload).
1019 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1021 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
1023 EXPORT_SYMBOL(blk_init_queue
);
1025 struct request_queue
*
1026 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1028 struct request_queue
*q
;
1030 q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1034 q
->request_fn
= rfn
;
1036 q
->queue_lock
= lock
;
1037 if (blk_init_allocated_queue(q
) < 0) {
1038 blk_cleanup_queue(q
);
1044 EXPORT_SYMBOL(blk_init_queue_node
);
1046 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
1049 int blk_init_allocated_queue(struct request_queue
*q
)
1051 WARN_ON_ONCE(q
->mq_ops
);
1053 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, q
->cmd_size
);
1057 if (q
->init_rq_fn
&& q
->init_rq_fn(q
, q
->fq
->flush_rq
, GFP_KERNEL
))
1058 goto out_free_flush_queue
;
1060 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
1061 goto out_exit_flush_rq
;
1063 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
1064 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
1067 * This also sets hw/phys segments, boundary and size
1069 blk_queue_make_request(q
, blk_queue_bio
);
1071 q
->sg_reserved_size
= INT_MAX
;
1073 /* Protect q->elevator from elevator_change */
1074 mutex_lock(&q
->sysfs_lock
);
1077 if (elevator_init(q
, NULL
)) {
1078 mutex_unlock(&q
->sysfs_lock
);
1079 goto out_exit_flush_rq
;
1082 mutex_unlock(&q
->sysfs_lock
);
1087 q
->exit_rq_fn(q
, q
->fq
->flush_rq
);
1088 out_free_flush_queue
:
1089 blk_free_flush_queue(q
->fq
);
1093 EXPORT_SYMBOL(blk_init_allocated_queue
);
1095 bool blk_get_queue(struct request_queue
*q
)
1097 if (likely(!blk_queue_dying(q
))) {
1104 EXPORT_SYMBOL(blk_get_queue
);
1106 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
1108 if (rq
->rq_flags
& RQF_ELVPRIV
) {
1109 elv_put_request(rl
->q
, rq
);
1111 put_io_context(rq
->elv
.icq
->ioc
);
1114 mempool_free(rq
, rl
->rq_pool
);
1118 * ioc_batching returns true if the ioc is a valid batching request and
1119 * should be given priority access to a request.
1121 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
1127 * Make sure the process is able to allocate at least 1 request
1128 * even if the batch times out, otherwise we could theoretically
1131 return ioc
->nr_batch_requests
== q
->nr_batching
||
1132 (ioc
->nr_batch_requests
> 0
1133 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
1137 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1138 * will cause the process to be a "batcher" on all queues in the system. This
1139 * is the behaviour we want though - once it gets a wakeup it should be given
1142 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
1144 if (!ioc
|| ioc_batching(q
, ioc
))
1147 ioc
->nr_batch_requests
= q
->nr_batching
;
1148 ioc
->last_waited
= jiffies
;
1151 static void __freed_request(struct request_list
*rl
, int sync
)
1153 struct request_queue
*q
= rl
->q
;
1155 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
1156 blk_clear_congested(rl
, sync
);
1158 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
1159 if (waitqueue_active(&rl
->wait
[sync
]))
1160 wake_up(&rl
->wait
[sync
]);
1162 blk_clear_rl_full(rl
, sync
);
1167 * A request has just been released. Account for it, update the full and
1168 * congestion status, wake up any waiters. Called under q->queue_lock.
1170 static void freed_request(struct request_list
*rl
, bool sync
,
1171 req_flags_t rq_flags
)
1173 struct request_queue
*q
= rl
->q
;
1177 if (rq_flags
& RQF_ELVPRIV
)
1178 q
->nr_rqs_elvpriv
--;
1180 __freed_request(rl
, sync
);
1182 if (unlikely(rl
->starved
[sync
^ 1]))
1183 __freed_request(rl
, sync
^ 1);
1186 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
1188 struct request_list
*rl
;
1189 int on_thresh
, off_thresh
;
1191 WARN_ON_ONCE(q
->mq_ops
);
1193 spin_lock_irq(q
->queue_lock
);
1194 q
->nr_requests
= nr
;
1195 blk_queue_congestion_threshold(q
);
1196 on_thresh
= queue_congestion_on_threshold(q
);
1197 off_thresh
= queue_congestion_off_threshold(q
);
1199 blk_queue_for_each_rl(rl
, q
) {
1200 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
1201 blk_set_congested(rl
, BLK_RW_SYNC
);
1202 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
1203 blk_clear_congested(rl
, BLK_RW_SYNC
);
1205 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
1206 blk_set_congested(rl
, BLK_RW_ASYNC
);
1207 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
1208 blk_clear_congested(rl
, BLK_RW_ASYNC
);
1210 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
1211 blk_set_rl_full(rl
, BLK_RW_SYNC
);
1213 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1214 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1217 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1218 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1220 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1221 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1225 spin_unlock_irq(q
->queue_lock
);
1230 * __get_request - get a free request
1231 * @rl: request list to allocate from
1232 * @op: operation and flags
1233 * @bio: bio to allocate request for (can be %NULL)
1234 * @flags: BLQ_MQ_REQ_* flags
1236 * Get a free request from @q. This function may fail under memory
1237 * pressure or if @q is dead.
1239 * Must be called with @q->queue_lock held and,
1240 * Returns ERR_PTR on failure, with @q->queue_lock held.
1241 * Returns request pointer on success, with @q->queue_lock *not held*.
1243 static struct request
*__get_request(struct request_list
*rl
, unsigned int op
,
1244 struct bio
*bio
, blk_mq_req_flags_t flags
)
1246 struct request_queue
*q
= rl
->q
;
1248 struct elevator_type
*et
= q
->elevator
->type
;
1249 struct io_context
*ioc
= rq_ioc(bio
);
1250 struct io_cq
*icq
= NULL
;
1251 const bool is_sync
= op_is_sync(op
);
1253 gfp_t gfp_mask
= flags
& BLK_MQ_REQ_NOWAIT
? GFP_ATOMIC
:
1254 __GFP_DIRECT_RECLAIM
;
1255 req_flags_t rq_flags
= RQF_ALLOCED
;
1257 lockdep_assert_held(q
->queue_lock
);
1259 if (unlikely(blk_queue_dying(q
)))
1260 return ERR_PTR(-ENODEV
);
1262 may_queue
= elv_may_queue(q
, op
);
1263 if (may_queue
== ELV_MQUEUE_NO
)
1266 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1267 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1269 * The queue will fill after this allocation, so set
1270 * it as full, and mark this process as "batching".
1271 * This process will be allowed to complete a batch of
1272 * requests, others will be blocked.
1274 if (!blk_rl_full(rl
, is_sync
)) {
1275 ioc_set_batching(q
, ioc
);
1276 blk_set_rl_full(rl
, is_sync
);
1278 if (may_queue
!= ELV_MQUEUE_MUST
1279 && !ioc_batching(q
, ioc
)) {
1281 * The queue is full and the allocating
1282 * process is not a "batcher", and not
1283 * exempted by the IO scheduler
1285 return ERR_PTR(-ENOMEM
);
1289 blk_set_congested(rl
, is_sync
);
1293 * Only allow batching queuers to allocate up to 50% over the defined
1294 * limit of requests, otherwise we could have thousands of requests
1295 * allocated with any setting of ->nr_requests
1297 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1298 return ERR_PTR(-ENOMEM
);
1300 q
->nr_rqs
[is_sync
]++;
1301 rl
->count
[is_sync
]++;
1302 rl
->starved
[is_sync
] = 0;
1305 * Decide whether the new request will be managed by elevator. If
1306 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1307 * prevent the current elevator from being destroyed until the new
1308 * request is freed. This guarantees icq's won't be destroyed and
1309 * makes creating new ones safe.
1311 * Flush requests do not use the elevator so skip initialization.
1312 * This allows a request to share the flush and elevator data.
1314 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1315 * it will be created after releasing queue_lock.
1317 if (!op_is_flush(op
) && !blk_queue_bypass(q
)) {
1318 rq_flags
|= RQF_ELVPRIV
;
1319 q
->nr_rqs_elvpriv
++;
1320 if (et
->icq_cache
&& ioc
)
1321 icq
= ioc_lookup_icq(ioc
, q
);
1324 if (blk_queue_io_stat(q
))
1325 rq_flags
|= RQF_IO_STAT
;
1326 spin_unlock_irq(q
->queue_lock
);
1328 /* allocate and init request */
1329 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1334 blk_rq_set_rl(rq
, rl
);
1336 rq
->rq_flags
= rq_flags
;
1337 if (flags
& BLK_MQ_REQ_PREEMPT
)
1338 rq
->rq_flags
|= RQF_PREEMPT
;
1341 if (rq_flags
& RQF_ELVPRIV
) {
1342 if (unlikely(et
->icq_cache
&& !icq
)) {
1344 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1350 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1353 /* @rq->elv.icq holds io_context until @rq is freed */
1355 get_io_context(icq
->ioc
);
1359 * ioc may be NULL here, and ioc_batching will be false. That's
1360 * OK, if the queue is under the request limit then requests need
1361 * not count toward the nr_batch_requests limit. There will always
1362 * be some limit enforced by BLK_BATCH_TIME.
1364 if (ioc_batching(q
, ioc
))
1365 ioc
->nr_batch_requests
--;
1367 trace_block_getrq(q
, bio
, op
);
1372 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1373 * and may fail indefinitely under memory pressure and thus
1374 * shouldn't stall IO. Treat this request as !elvpriv. This will
1375 * disturb iosched and blkcg but weird is bettern than dead.
1377 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1378 __func__
, dev_name(q
->backing_dev_info
->dev
));
1380 rq
->rq_flags
&= ~RQF_ELVPRIV
;
1383 spin_lock_irq(q
->queue_lock
);
1384 q
->nr_rqs_elvpriv
--;
1385 spin_unlock_irq(q
->queue_lock
);
1390 * Allocation failed presumably due to memory. Undo anything we
1391 * might have messed up.
1393 * Allocating task should really be put onto the front of the wait
1394 * queue, but this is pretty rare.
1396 spin_lock_irq(q
->queue_lock
);
1397 freed_request(rl
, is_sync
, rq_flags
);
1400 * in the very unlikely event that allocation failed and no
1401 * requests for this direction was pending, mark us starved so that
1402 * freeing of a request in the other direction will notice
1403 * us. another possible fix would be to split the rq mempool into
1407 if (unlikely(rl
->count
[is_sync
] == 0))
1408 rl
->starved
[is_sync
] = 1;
1409 return ERR_PTR(-ENOMEM
);
1413 * get_request - get a free request
1414 * @q: request_queue to allocate request from
1415 * @op: operation and flags
1416 * @bio: bio to allocate request for (can be %NULL)
1417 * @flags: BLK_MQ_REQ_* flags.
1419 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1420 * this function keeps retrying under memory pressure and fails iff @q is dead.
1422 * Must be called with @q->queue_lock held and,
1423 * Returns ERR_PTR on failure, with @q->queue_lock held.
1424 * Returns request pointer on success, with @q->queue_lock *not held*.
1426 static struct request
*get_request(struct request_queue
*q
, unsigned int op
,
1427 struct bio
*bio
, blk_mq_req_flags_t flags
)
1429 const bool is_sync
= op_is_sync(op
);
1431 struct request_list
*rl
;
1434 lockdep_assert_held(q
->queue_lock
);
1435 WARN_ON_ONCE(q
->mq_ops
);
1437 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1439 rq
= __get_request(rl
, op
, bio
, flags
);
1443 if (op
& REQ_NOWAIT
) {
1445 return ERR_PTR(-EAGAIN
);
1448 if ((flags
& BLK_MQ_REQ_NOWAIT
) || unlikely(blk_queue_dying(q
))) {
1453 /* wait on @rl and retry */
1454 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1455 TASK_UNINTERRUPTIBLE
);
1457 trace_block_sleeprq(q
, bio
, op
);
1459 spin_unlock_irq(q
->queue_lock
);
1463 * After sleeping, we become a "batching" process and will be able
1464 * to allocate at least one request, and up to a big batch of them
1465 * for a small period time. See ioc_batching, ioc_set_batching
1467 ioc_set_batching(q
, current
->io_context
);
1469 spin_lock_irq(q
->queue_lock
);
1470 finish_wait(&rl
->wait
[is_sync
], &wait
);
1475 /* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */
1476 static struct request
*blk_old_get_request(struct request_queue
*q
,
1477 unsigned int op
, blk_mq_req_flags_t flags
)
1480 gfp_t gfp_mask
= flags
& BLK_MQ_REQ_NOWAIT
? GFP_ATOMIC
:
1481 __GFP_DIRECT_RECLAIM
;
1484 WARN_ON_ONCE(q
->mq_ops
);
1486 /* create ioc upfront */
1487 create_io_context(gfp_mask
, q
->node
);
1489 ret
= blk_queue_enter(q
, flags
);
1491 return ERR_PTR(ret
);
1492 spin_lock_irq(q
->queue_lock
);
1493 rq
= get_request(q
, op
, NULL
, flags
);
1495 spin_unlock_irq(q
->queue_lock
);
1500 /* q->queue_lock is unlocked at this point */
1502 rq
->__sector
= (sector_t
) -1;
1503 rq
->bio
= rq
->biotail
= NULL
;
1508 * blk_get_request_flags - allocate a request
1509 * @q: request queue to allocate a request for
1510 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
1511 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
1513 struct request
*blk_get_request_flags(struct request_queue
*q
, unsigned int op
,
1514 blk_mq_req_flags_t flags
)
1516 struct request
*req
;
1518 WARN_ON_ONCE(op
& REQ_NOWAIT
);
1519 WARN_ON_ONCE(flags
& ~(BLK_MQ_REQ_NOWAIT
| BLK_MQ_REQ_PREEMPT
));
1522 req
= blk_mq_alloc_request(q
, op
, flags
);
1523 if (!IS_ERR(req
) && q
->mq_ops
->initialize_rq_fn
)
1524 q
->mq_ops
->initialize_rq_fn(req
);
1526 req
= blk_old_get_request(q
, op
, flags
);
1527 if (!IS_ERR(req
) && q
->initialize_rq_fn
)
1528 q
->initialize_rq_fn(req
);
1533 EXPORT_SYMBOL(blk_get_request_flags
);
1535 struct request
*blk_get_request(struct request_queue
*q
, unsigned int op
,
1538 return blk_get_request_flags(q
, op
, gfp_mask
& __GFP_DIRECT_RECLAIM
?
1539 0 : BLK_MQ_REQ_NOWAIT
);
1541 EXPORT_SYMBOL(blk_get_request
);
1544 * blk_requeue_request - put a request back on queue
1545 * @q: request queue where request should be inserted
1546 * @rq: request to be inserted
1549 * Drivers often keep queueing requests until the hardware cannot accept
1550 * more, when that condition happens we need to put the request back
1551 * on the queue. Must be called with queue lock held.
1553 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1555 lockdep_assert_held(q
->queue_lock
);
1556 WARN_ON_ONCE(q
->mq_ops
);
1558 blk_delete_timer(rq
);
1559 blk_clear_rq_complete(rq
);
1560 trace_block_rq_requeue(q
, rq
);
1561 wbt_requeue(q
->rq_wb
, &rq
->issue_stat
);
1563 if (rq
->rq_flags
& RQF_QUEUED
)
1564 blk_queue_end_tag(q
, rq
);
1566 BUG_ON(blk_queued_rq(rq
));
1568 elv_requeue_request(q
, rq
);
1570 EXPORT_SYMBOL(blk_requeue_request
);
1572 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1575 blk_account_io_start(rq
, true);
1576 __elv_add_request(q
, rq
, where
);
1579 static void part_round_stats_single(struct request_queue
*q
, int cpu
,
1580 struct hd_struct
*part
, unsigned long now
,
1581 unsigned int inflight
)
1584 __part_stat_add(cpu
, part
, time_in_queue
,
1585 inflight
* (now
- part
->stamp
));
1586 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1592 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1593 * @q: target block queue
1594 * @cpu: cpu number for stats access
1595 * @part: target partition
1597 * The average IO queue length and utilisation statistics are maintained
1598 * by observing the current state of the queue length and the amount of
1599 * time it has been in this state for.
1601 * Normally, that accounting is done on IO completion, but that can result
1602 * in more than a second's worth of IO being accounted for within any one
1603 * second, leading to >100% utilisation. To deal with that, we call this
1604 * function to do a round-off before returning the results when reading
1605 * /proc/diskstats. This accounts immediately for all queue usage up to
1606 * the current jiffies and restarts the counters again.
1608 void part_round_stats(struct request_queue
*q
, int cpu
, struct hd_struct
*part
)
1610 struct hd_struct
*part2
= NULL
;
1611 unsigned long now
= jiffies
;
1612 unsigned int inflight
[2];
1615 if (part
->stamp
!= now
)
1619 part2
= &part_to_disk(part
)->part0
;
1620 if (part2
->stamp
!= now
)
1627 part_in_flight(q
, part
, inflight
);
1630 part_round_stats_single(q
, cpu
, part2
, now
, inflight
[1]);
1632 part_round_stats_single(q
, cpu
, part
, now
, inflight
[0]);
1634 EXPORT_SYMBOL_GPL(part_round_stats
);
1637 static void blk_pm_put_request(struct request
*rq
)
1639 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1640 pm_runtime_mark_last_busy(rq
->q
->dev
);
1643 static inline void blk_pm_put_request(struct request
*rq
) {}
1646 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1648 req_flags_t rq_flags
= req
->rq_flags
;
1654 blk_mq_free_request(req
);
1658 lockdep_assert_held(q
->queue_lock
);
1660 blk_pm_put_request(req
);
1662 elv_completed_request(q
, req
);
1664 /* this is a bio leak */
1665 WARN_ON(req
->bio
!= NULL
);
1667 wbt_done(q
->rq_wb
, &req
->issue_stat
);
1670 * Request may not have originated from ll_rw_blk. if not,
1671 * it didn't come out of our reserved rq pools
1673 if (rq_flags
& RQF_ALLOCED
) {
1674 struct request_list
*rl
= blk_rq_rl(req
);
1675 bool sync
= op_is_sync(req
->cmd_flags
);
1677 BUG_ON(!list_empty(&req
->queuelist
));
1678 BUG_ON(ELV_ON_HASH(req
));
1680 blk_free_request(rl
, req
);
1681 freed_request(rl
, sync
, rq_flags
);
1686 EXPORT_SYMBOL_GPL(__blk_put_request
);
1688 void blk_put_request(struct request
*req
)
1690 struct request_queue
*q
= req
->q
;
1693 blk_mq_free_request(req
);
1695 unsigned long flags
;
1697 spin_lock_irqsave(q
->queue_lock
, flags
);
1698 __blk_put_request(q
, req
);
1699 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1702 EXPORT_SYMBOL(blk_put_request
);
1704 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1707 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1709 if (!ll_back_merge_fn(q
, req
, bio
))
1712 trace_block_bio_backmerge(q
, req
, bio
);
1714 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1715 blk_rq_set_mixed_merge(req
);
1717 req
->biotail
->bi_next
= bio
;
1719 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1720 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1722 blk_account_io_start(req
, false);
1726 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1729 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1731 if (!ll_front_merge_fn(q
, req
, bio
))
1734 trace_block_bio_frontmerge(q
, req
, bio
);
1736 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1737 blk_rq_set_mixed_merge(req
);
1739 bio
->bi_next
= req
->bio
;
1742 req
->__sector
= bio
->bi_iter
.bi_sector
;
1743 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1744 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1746 blk_account_io_start(req
, false);
1750 bool bio_attempt_discard_merge(struct request_queue
*q
, struct request
*req
,
1753 unsigned short segments
= blk_rq_nr_discard_segments(req
);
1755 if (segments
>= queue_max_discard_segments(q
))
1757 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
1758 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
1761 req
->biotail
->bi_next
= bio
;
1763 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1764 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1765 req
->nr_phys_segments
= segments
+ 1;
1767 blk_account_io_start(req
, false);
1770 req_set_nomerge(q
, req
);
1775 * blk_attempt_plug_merge - try to merge with %current's plugged list
1776 * @q: request_queue new bio is being queued at
1777 * @bio: new bio being queued
1778 * @request_count: out parameter for number of traversed plugged requests
1779 * @same_queue_rq: pointer to &struct request that gets filled in when
1780 * another request associated with @q is found on the plug list
1781 * (optional, may be %NULL)
1783 * Determine whether @bio being queued on @q can be merged with a request
1784 * on %current's plugged list. Returns %true if merge was successful,
1787 * Plugging coalesces IOs from the same issuer for the same purpose without
1788 * going through @q->queue_lock. As such it's more of an issuing mechanism
1789 * than scheduling, and the request, while may have elvpriv data, is not
1790 * added on the elevator at this point. In addition, we don't have
1791 * reliable access to the elevator outside queue lock. Only check basic
1792 * merging parameters without querying the elevator.
1794 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1796 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1797 unsigned int *request_count
,
1798 struct request
**same_queue_rq
)
1800 struct blk_plug
*plug
;
1802 struct list_head
*plug_list
;
1804 plug
= current
->plug
;
1810 plug_list
= &plug
->mq_list
;
1812 plug_list
= &plug
->list
;
1814 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1815 bool merged
= false;
1820 * Only blk-mq multiple hardware queues case checks the
1821 * rq in the same queue, there should be only one such
1825 *same_queue_rq
= rq
;
1828 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1831 switch (blk_try_merge(rq
, bio
)) {
1832 case ELEVATOR_BACK_MERGE
:
1833 merged
= bio_attempt_back_merge(q
, rq
, bio
);
1835 case ELEVATOR_FRONT_MERGE
:
1836 merged
= bio_attempt_front_merge(q
, rq
, bio
);
1838 case ELEVATOR_DISCARD_MERGE
:
1839 merged
= bio_attempt_discard_merge(q
, rq
, bio
);
1852 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1854 struct blk_plug
*plug
;
1856 struct list_head
*plug_list
;
1857 unsigned int ret
= 0;
1859 plug
= current
->plug
;
1864 plug_list
= &plug
->mq_list
;
1866 plug_list
= &plug
->list
;
1868 list_for_each_entry(rq
, plug_list
, queuelist
) {
1876 void blk_init_request_from_bio(struct request
*req
, struct bio
*bio
)
1878 struct io_context
*ioc
= rq_ioc(bio
);
1880 if (bio
->bi_opf
& REQ_RAHEAD
)
1881 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1883 req
->__sector
= bio
->bi_iter
.bi_sector
;
1884 if (ioprio_valid(bio_prio(bio
)))
1885 req
->ioprio
= bio_prio(bio
);
1887 req
->ioprio
= ioc
->ioprio
;
1889 req
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE
, 0);
1890 req
->write_hint
= bio
->bi_write_hint
;
1891 blk_rq_bio_prep(req
->q
, req
, bio
);
1893 EXPORT_SYMBOL_GPL(blk_init_request_from_bio
);
1895 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1897 struct blk_plug
*plug
;
1898 int where
= ELEVATOR_INSERT_SORT
;
1899 struct request
*req
, *free
;
1900 unsigned int request_count
= 0;
1901 unsigned int wb_acct
;
1904 * low level driver can indicate that it wants pages above a
1905 * certain limit bounced to low memory (ie for highmem, or even
1906 * ISA dma in theory)
1908 blk_queue_bounce(q
, &bio
);
1910 blk_queue_split(q
, &bio
);
1912 if (!bio_integrity_prep(bio
))
1913 return BLK_QC_T_NONE
;
1915 if (op_is_flush(bio
->bi_opf
)) {
1916 spin_lock_irq(q
->queue_lock
);
1917 where
= ELEVATOR_INSERT_FLUSH
;
1922 * Check if we can merge with the plugged list before grabbing
1925 if (!blk_queue_nomerges(q
)) {
1926 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1927 return BLK_QC_T_NONE
;
1929 request_count
= blk_plug_queued_count(q
);
1931 spin_lock_irq(q
->queue_lock
);
1933 switch (elv_merge(q
, &req
, bio
)) {
1934 case ELEVATOR_BACK_MERGE
:
1935 if (!bio_attempt_back_merge(q
, req
, bio
))
1937 elv_bio_merged(q
, req
, bio
);
1938 free
= attempt_back_merge(q
, req
);
1940 __blk_put_request(q
, free
);
1942 elv_merged_request(q
, req
, ELEVATOR_BACK_MERGE
);
1944 case ELEVATOR_FRONT_MERGE
:
1945 if (!bio_attempt_front_merge(q
, req
, bio
))
1947 elv_bio_merged(q
, req
, bio
);
1948 free
= attempt_front_merge(q
, req
);
1950 __blk_put_request(q
, free
);
1952 elv_merged_request(q
, req
, ELEVATOR_FRONT_MERGE
);
1959 wb_acct
= wbt_wait(q
->rq_wb
, bio
, q
->queue_lock
);
1962 * Grab a free request. This is might sleep but can not fail.
1963 * Returns with the queue unlocked.
1965 blk_queue_enter_live(q
);
1966 req
= get_request(q
, bio
->bi_opf
, bio
, 0);
1969 __wbt_done(q
->rq_wb
, wb_acct
);
1970 if (PTR_ERR(req
) == -ENOMEM
)
1971 bio
->bi_status
= BLK_STS_RESOURCE
;
1973 bio
->bi_status
= BLK_STS_IOERR
;
1978 wbt_track(&req
->issue_stat
, wb_acct
);
1981 * After dropping the lock and possibly sleeping here, our request
1982 * may now be mergeable after it had proven unmergeable (above).
1983 * We don't worry about that case for efficiency. It won't happen
1984 * often, and the elevators are able to handle it.
1986 blk_init_request_from_bio(req
, bio
);
1988 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1989 req
->cpu
= raw_smp_processor_id();
1991 plug
= current
->plug
;
1994 * If this is the first request added after a plug, fire
1997 * @request_count may become stale because of schedule
1998 * out, so check plug list again.
2000 if (!request_count
|| list_empty(&plug
->list
))
2001 trace_block_plug(q
);
2003 struct request
*last
= list_entry_rq(plug
->list
.prev
);
2004 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
2005 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
2006 blk_flush_plug_list(plug
, false);
2007 trace_block_plug(q
);
2010 list_add_tail(&req
->queuelist
, &plug
->list
);
2011 blk_account_io_start(req
, true);
2013 spin_lock_irq(q
->queue_lock
);
2014 add_acct_request(q
, req
, where
);
2017 spin_unlock_irq(q
->queue_lock
);
2020 return BLK_QC_T_NONE
;
2023 static void handle_bad_sector(struct bio
*bio
)
2025 char b
[BDEVNAME_SIZE
];
2027 printk(KERN_INFO
"attempt to access beyond end of device\n");
2028 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
2029 bio_devname(bio
, b
), bio
->bi_opf
,
2030 (unsigned long long)bio_end_sector(bio
),
2031 (long long)get_capacity(bio
->bi_disk
));
2034 #ifdef CONFIG_FAIL_MAKE_REQUEST
2036 static DECLARE_FAULT_ATTR(fail_make_request
);
2038 static int __init
setup_fail_make_request(char *str
)
2040 return setup_fault_attr(&fail_make_request
, str
);
2042 __setup("fail_make_request=", setup_fail_make_request
);
2044 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
2046 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
2049 static int __init
fail_make_request_debugfs(void)
2051 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
2052 NULL
, &fail_make_request
);
2054 return PTR_ERR_OR_ZERO(dir
);
2057 late_initcall(fail_make_request_debugfs
);
2059 #else /* CONFIG_FAIL_MAKE_REQUEST */
2061 static inline bool should_fail_request(struct hd_struct
*part
,
2067 #endif /* CONFIG_FAIL_MAKE_REQUEST */
2070 * Remap block n of partition p to block n+start(p) of the disk.
2072 static inline int blk_partition_remap(struct bio
*bio
)
2074 struct hd_struct
*p
;
2078 * Zone reset does not include bi_size so bio_sectors() is always 0.
2079 * Include a test for the reset op code and perform the remap if needed.
2081 if (!bio
->bi_partno
||
2082 (!bio_sectors(bio
) && bio_op(bio
) != REQ_OP_ZONE_RESET
))
2086 p
= __disk_get_part(bio
->bi_disk
, bio
->bi_partno
);
2087 if (likely(p
&& !should_fail_request(p
, bio
->bi_iter
.bi_size
))) {
2088 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
2090 trace_block_bio_remap(bio
->bi_disk
->queue
, bio
, part_devt(p
),
2091 bio
->bi_iter
.bi_sector
- p
->start_sect
);
2093 printk("%s: fail for partition %d\n", __func__
, bio
->bi_partno
);
2102 * Check whether this bio extends beyond the end of the device.
2104 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
2111 /* Test device or partition size, when known. */
2112 maxsector
= get_capacity(bio
->bi_disk
);
2114 sector_t sector
= bio
->bi_iter
.bi_sector
;
2116 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
2118 * This may well happen - the kernel calls bread()
2119 * without checking the size of the device, e.g., when
2120 * mounting a device.
2122 handle_bad_sector(bio
);
2130 static noinline_for_stack
bool
2131 generic_make_request_checks(struct bio
*bio
)
2133 struct request_queue
*q
;
2134 int nr_sectors
= bio_sectors(bio
);
2135 blk_status_t status
= BLK_STS_IOERR
;
2136 char b
[BDEVNAME_SIZE
];
2140 if (bio_check_eod(bio
, nr_sectors
))
2143 q
= bio
->bi_disk
->queue
;
2146 "generic_make_request: Trying to access "
2147 "nonexistent block-device %s (%Lu)\n",
2148 bio_devname(bio
, b
), (long long)bio
->bi_iter
.bi_sector
);
2153 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2154 * if queue is not a request based queue.
2157 if ((bio
->bi_opf
& REQ_NOWAIT
) && !queue_is_rq_based(q
))
2160 if (should_fail_request(&bio
->bi_disk
->part0
, bio
->bi_iter
.bi_size
))
2163 if (blk_partition_remap(bio
))
2166 if (bio_check_eod(bio
, nr_sectors
))
2170 * Filter flush bio's early so that make_request based
2171 * drivers without flush support don't have to worry
2174 if (op_is_flush(bio
->bi_opf
) &&
2175 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
2176 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
2178 status
= BLK_STS_OK
;
2183 switch (bio_op(bio
)) {
2184 case REQ_OP_DISCARD
:
2185 if (!blk_queue_discard(q
))
2188 case REQ_OP_SECURE_ERASE
:
2189 if (!blk_queue_secure_erase(q
))
2192 case REQ_OP_WRITE_SAME
:
2193 if (!q
->limits
.max_write_same_sectors
)
2196 case REQ_OP_ZONE_REPORT
:
2197 case REQ_OP_ZONE_RESET
:
2198 if (!blk_queue_is_zoned(q
))
2201 case REQ_OP_WRITE_ZEROES
:
2202 if (!q
->limits
.max_write_zeroes_sectors
)
2210 * Various block parts want %current->io_context and lazy ioc
2211 * allocation ends up trading a lot of pain for a small amount of
2212 * memory. Just allocate it upfront. This may fail and block
2213 * layer knows how to live with it.
2215 create_io_context(GFP_ATOMIC
, q
->node
);
2217 if (!blkcg_bio_issue_check(q
, bio
))
2220 if (!bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
2221 trace_block_bio_queue(q
, bio
);
2222 /* Now that enqueuing has been traced, we need to trace
2223 * completion as well.
2225 bio_set_flag(bio
, BIO_TRACE_COMPLETION
);
2230 status
= BLK_STS_NOTSUPP
;
2232 bio
->bi_status
= status
;
2238 * generic_make_request - hand a buffer to its device driver for I/O
2239 * @bio: The bio describing the location in memory and on the device.
2241 * generic_make_request() is used to make I/O requests of block
2242 * devices. It is passed a &struct bio, which describes the I/O that needs
2245 * generic_make_request() does not return any status. The
2246 * success/failure status of the request, along with notification of
2247 * completion, is delivered asynchronously through the bio->bi_end_io
2248 * function described (one day) else where.
2250 * The caller of generic_make_request must make sure that bi_io_vec
2251 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2252 * set to describe the device address, and the
2253 * bi_end_io and optionally bi_private are set to describe how
2254 * completion notification should be signaled.
2256 * generic_make_request and the drivers it calls may use bi_next if this
2257 * bio happens to be merged with someone else, and may resubmit the bio to
2258 * a lower device by calling into generic_make_request recursively, which
2259 * means the bio should NOT be touched after the call to ->make_request_fn.
2261 blk_qc_t
generic_make_request(struct bio
*bio
)
2264 * bio_list_on_stack[0] contains bios submitted by the current
2266 * bio_list_on_stack[1] contains bios that were submitted before
2267 * the current make_request_fn, but that haven't been processed
2270 struct bio_list bio_list_on_stack
[2];
2271 blk_qc_t ret
= BLK_QC_T_NONE
;
2273 if (!generic_make_request_checks(bio
))
2277 * We only want one ->make_request_fn to be active at a time, else
2278 * stack usage with stacked devices could be a problem. So use
2279 * current->bio_list to keep a list of requests submited by a
2280 * make_request_fn function. current->bio_list is also used as a
2281 * flag to say if generic_make_request is currently active in this
2282 * task or not. If it is NULL, then no make_request is active. If
2283 * it is non-NULL, then a make_request is active, and new requests
2284 * should be added at the tail
2286 if (current
->bio_list
) {
2287 bio_list_add(¤t
->bio_list
[0], bio
);
2291 /* following loop may be a bit non-obvious, and so deserves some
2293 * Before entering the loop, bio->bi_next is NULL (as all callers
2294 * ensure that) so we have a list with a single bio.
2295 * We pretend that we have just taken it off a longer list, so
2296 * we assign bio_list to a pointer to the bio_list_on_stack,
2297 * thus initialising the bio_list of new bios to be
2298 * added. ->make_request() may indeed add some more bios
2299 * through a recursive call to generic_make_request. If it
2300 * did, we find a non-NULL value in bio_list and re-enter the loop
2301 * from the top. In this case we really did just take the bio
2302 * of the top of the list (no pretending) and so remove it from
2303 * bio_list, and call into ->make_request() again.
2305 BUG_ON(bio
->bi_next
);
2306 bio_list_init(&bio_list_on_stack
[0]);
2307 current
->bio_list
= bio_list_on_stack
;
2309 struct request_queue
*q
= bio
->bi_disk
->queue
;
2310 blk_mq_req_flags_t flags
= bio
->bi_opf
& REQ_NOWAIT
?
2311 BLK_MQ_REQ_NOWAIT
: 0;
2313 if (likely(blk_queue_enter(q
, flags
) == 0)) {
2314 struct bio_list lower
, same
;
2316 /* Create a fresh bio_list for all subordinate requests */
2317 bio_list_on_stack
[1] = bio_list_on_stack
[0];
2318 bio_list_init(&bio_list_on_stack
[0]);
2319 ret
= q
->make_request_fn(q
, bio
);
2323 /* sort new bios into those for a lower level
2324 * and those for the same level
2326 bio_list_init(&lower
);
2327 bio_list_init(&same
);
2328 while ((bio
= bio_list_pop(&bio_list_on_stack
[0])) != NULL
)
2329 if (q
== bio
->bi_disk
->queue
)
2330 bio_list_add(&same
, bio
);
2332 bio_list_add(&lower
, bio
);
2333 /* now assemble so we handle the lowest level first */
2334 bio_list_merge(&bio_list_on_stack
[0], &lower
);
2335 bio_list_merge(&bio_list_on_stack
[0], &same
);
2336 bio_list_merge(&bio_list_on_stack
[0], &bio_list_on_stack
[1]);
2338 if (unlikely(!blk_queue_dying(q
) &&
2339 (bio
->bi_opf
& REQ_NOWAIT
)))
2340 bio_wouldblock_error(bio
);
2344 bio
= bio_list_pop(&bio_list_on_stack
[0]);
2346 current
->bio_list
= NULL
; /* deactivate */
2351 EXPORT_SYMBOL(generic_make_request
);
2354 * direct_make_request - hand a buffer directly to its device driver for I/O
2355 * @bio: The bio describing the location in memory and on the device.
2357 * This function behaves like generic_make_request(), but does not protect
2358 * against recursion. Must only be used if the called driver is known
2359 * to not call generic_make_request (or direct_make_request) again from
2360 * its make_request function. (Calling direct_make_request again from
2361 * a workqueue is perfectly fine as that doesn't recurse).
2363 blk_qc_t
direct_make_request(struct bio
*bio
)
2365 struct request_queue
*q
= bio
->bi_disk
->queue
;
2366 bool nowait
= bio
->bi_opf
& REQ_NOWAIT
;
2369 if (!generic_make_request_checks(bio
))
2370 return BLK_QC_T_NONE
;
2372 if (unlikely(blk_queue_enter(q
, nowait
? BLK_MQ_REQ_NOWAIT
: 0))) {
2373 if (nowait
&& !blk_queue_dying(q
))
2374 bio
->bi_status
= BLK_STS_AGAIN
;
2376 bio
->bi_status
= BLK_STS_IOERR
;
2378 return BLK_QC_T_NONE
;
2381 ret
= q
->make_request_fn(q
, bio
);
2385 EXPORT_SYMBOL_GPL(direct_make_request
);
2388 * submit_bio - submit a bio to the block device layer for I/O
2389 * @bio: The &struct bio which describes the I/O
2391 * submit_bio() is very similar in purpose to generic_make_request(), and
2392 * uses that function to do most of the work. Both are fairly rough
2393 * interfaces; @bio must be presetup and ready for I/O.
2396 blk_qc_t
submit_bio(struct bio
*bio
)
2399 * If it's a regular read/write or a barrier with data attached,
2400 * go through the normal accounting stuff before submission.
2402 if (bio_has_data(bio
)) {
2405 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2406 count
= queue_logical_block_size(bio
->bi_disk
->queue
) >> 9;
2408 count
= bio_sectors(bio
);
2410 if (op_is_write(bio_op(bio
))) {
2411 count_vm_events(PGPGOUT
, count
);
2413 task_io_account_read(bio
->bi_iter
.bi_size
);
2414 count_vm_events(PGPGIN
, count
);
2417 if (unlikely(block_dump
)) {
2418 char b
[BDEVNAME_SIZE
];
2419 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2420 current
->comm
, task_pid_nr(current
),
2421 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2422 (unsigned long long)bio
->bi_iter
.bi_sector
,
2423 bio_devname(bio
, b
), count
);
2427 return generic_make_request(bio
);
2429 EXPORT_SYMBOL(submit_bio
);
2431 bool blk_poll(struct request_queue
*q
, blk_qc_t cookie
)
2433 if (!q
->poll_fn
|| !blk_qc_t_valid(cookie
))
2437 blk_flush_plug_list(current
->plug
, false);
2438 return q
->poll_fn(q
, cookie
);
2440 EXPORT_SYMBOL_GPL(blk_poll
);
2443 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2444 * for new the queue limits
2446 * @rq: the request being checked
2449 * @rq may have been made based on weaker limitations of upper-level queues
2450 * in request stacking drivers, and it may violate the limitation of @q.
2451 * Since the block layer and the underlying device driver trust @rq
2452 * after it is inserted to @q, it should be checked against @q before
2453 * the insertion using this generic function.
2455 * Request stacking drivers like request-based dm may change the queue
2456 * limits when retrying requests on other queues. Those requests need
2457 * to be checked against the new queue limits again during dispatch.
2459 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2462 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2463 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2468 * queue's settings related to segment counting like q->bounce_pfn
2469 * may differ from that of other stacking queues.
2470 * Recalculate it to check the request correctly on this queue's
2473 blk_recalc_rq_segments(rq
);
2474 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2475 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2483 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2484 * @q: the queue to submit the request
2485 * @rq: the request being queued
2487 blk_status_t
blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2489 unsigned long flags
;
2490 int where
= ELEVATOR_INSERT_BACK
;
2492 if (blk_cloned_rq_check_limits(q
, rq
))
2493 return BLK_STS_IOERR
;
2496 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2497 return BLK_STS_IOERR
;
2500 if (blk_queue_io_stat(q
))
2501 blk_account_io_start(rq
, true);
2503 * Since we have a scheduler attached on the top device,
2504 * bypass a potential scheduler on the bottom device for
2507 blk_mq_request_bypass_insert(rq
, true);
2511 spin_lock_irqsave(q
->queue_lock
, flags
);
2512 if (unlikely(blk_queue_dying(q
))) {
2513 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2514 return BLK_STS_IOERR
;
2518 * Submitting request must be dequeued before calling this function
2519 * because it will be linked to another request_queue
2521 BUG_ON(blk_queued_rq(rq
));
2523 if (op_is_flush(rq
->cmd_flags
))
2524 where
= ELEVATOR_INSERT_FLUSH
;
2526 add_acct_request(q
, rq
, where
);
2527 if (where
== ELEVATOR_INSERT_FLUSH
)
2529 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2533 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2536 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2537 * @rq: request to examine
2540 * A request could be merge of IOs which require different failure
2541 * handling. This function determines the number of bytes which
2542 * can be failed from the beginning of the request without
2543 * crossing into area which need to be retried further.
2546 * The number of bytes to fail.
2548 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2550 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2551 unsigned int bytes
= 0;
2554 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2555 return blk_rq_bytes(rq
);
2558 * Currently the only 'mixing' which can happen is between
2559 * different fastfail types. We can safely fail portions
2560 * which have all the failfast bits that the first one has -
2561 * the ones which are at least as eager to fail as the first
2564 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2565 if ((bio
->bi_opf
& ff
) != ff
)
2567 bytes
+= bio
->bi_iter
.bi_size
;
2570 /* this could lead to infinite loop */
2571 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2574 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2576 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2578 if (blk_do_io_stat(req
)) {
2579 const int rw
= rq_data_dir(req
);
2580 struct hd_struct
*part
;
2583 cpu
= part_stat_lock();
2585 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2590 void blk_account_io_done(struct request
*req
)
2593 * Account IO completion. flush_rq isn't accounted as a
2594 * normal IO on queueing nor completion. Accounting the
2595 * containing request is enough.
2597 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2598 unsigned long duration
= jiffies
- req
->start_time
;
2599 const int rw
= rq_data_dir(req
);
2600 struct hd_struct
*part
;
2603 cpu
= part_stat_lock();
2606 part_stat_inc(cpu
, part
, ios
[rw
]);
2607 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2608 part_round_stats(req
->q
, cpu
, part
);
2609 part_dec_in_flight(req
->q
, part
, rw
);
2611 hd_struct_put(part
);
2618 * Don't process normal requests when queue is suspended
2619 * or in the process of suspending/resuming
2621 static bool blk_pm_allow_request(struct request
*rq
)
2623 switch (rq
->q
->rpm_status
) {
2625 case RPM_SUSPENDING
:
2626 return rq
->rq_flags
& RQF_PM
;
2634 static bool blk_pm_allow_request(struct request
*rq
)
2640 void blk_account_io_start(struct request
*rq
, bool new_io
)
2642 struct hd_struct
*part
;
2643 int rw
= rq_data_dir(rq
);
2646 if (!blk_do_io_stat(rq
))
2649 cpu
= part_stat_lock();
2653 part_stat_inc(cpu
, part
, merges
[rw
]);
2655 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2656 if (!hd_struct_try_get(part
)) {
2658 * The partition is already being removed,
2659 * the request will be accounted on the disk only
2661 * We take a reference on disk->part0 although that
2662 * partition will never be deleted, so we can treat
2663 * it as any other partition.
2665 part
= &rq
->rq_disk
->part0
;
2666 hd_struct_get(part
);
2668 part_round_stats(rq
->q
, cpu
, part
);
2669 part_inc_in_flight(rq
->q
, part
, rw
);
2676 static struct request
*elv_next_request(struct request_queue
*q
)
2679 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
2681 WARN_ON_ONCE(q
->mq_ops
);
2684 list_for_each_entry(rq
, &q
->queue_head
, queuelist
) {
2685 if (blk_pm_allow_request(rq
))
2688 if (rq
->rq_flags
& RQF_SOFTBARRIER
)
2693 * Flush request is running and flush request isn't queueable
2694 * in the drive, we can hold the queue till flush request is
2695 * finished. Even we don't do this, driver can't dispatch next
2696 * requests and will requeue them. And this can improve
2697 * throughput too. For example, we have request flush1, write1,
2698 * flush 2. flush1 is dispatched, then queue is hold, write1
2699 * isn't inserted to queue. After flush1 is finished, flush2
2700 * will be dispatched. Since disk cache is already clean,
2701 * flush2 will be finished very soon, so looks like flush2 is
2703 * Since the queue is hold, a flag is set to indicate the queue
2704 * should be restarted later. Please see flush_end_io() for
2707 if (fq
->flush_pending_idx
!= fq
->flush_running_idx
&&
2708 !queue_flush_queueable(q
)) {
2709 fq
->flush_queue_delayed
= 1;
2712 if (unlikely(blk_queue_bypass(q
)) ||
2713 !q
->elevator
->type
->ops
.sq
.elevator_dispatch_fn(q
, 0))
2719 * blk_peek_request - peek at the top of a request queue
2720 * @q: request queue to peek at
2723 * Return the request at the top of @q. The returned request
2724 * should be started using blk_start_request() before LLD starts
2728 * Pointer to the request at the top of @q if available. Null
2731 struct request
*blk_peek_request(struct request_queue
*q
)
2736 lockdep_assert_held(q
->queue_lock
);
2737 WARN_ON_ONCE(q
->mq_ops
);
2739 while ((rq
= elv_next_request(q
)) != NULL
) {
2740 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2742 * This is the first time the device driver
2743 * sees this request (possibly after
2744 * requeueing). Notify IO scheduler.
2746 if (rq
->rq_flags
& RQF_SORTED
)
2747 elv_activate_rq(q
, rq
);
2750 * just mark as started even if we don't start
2751 * it, a request that has been delayed should
2752 * not be passed by new incoming requests
2754 rq
->rq_flags
|= RQF_STARTED
;
2755 trace_block_rq_issue(q
, rq
);
2758 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2759 q
->end_sector
= rq_end_sector(rq
);
2760 q
->boundary_rq
= NULL
;
2763 if (rq
->rq_flags
& RQF_DONTPREP
)
2766 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2768 * make sure space for the drain appears we
2769 * know we can do this because max_hw_segments
2770 * has been adjusted to be one fewer than the
2773 rq
->nr_phys_segments
++;
2779 ret
= q
->prep_rq_fn(q
, rq
);
2780 if (ret
== BLKPREP_OK
) {
2782 } else if (ret
== BLKPREP_DEFER
) {
2784 * the request may have been (partially) prepped.
2785 * we need to keep this request in the front to
2786 * avoid resource deadlock. RQF_STARTED will
2787 * prevent other fs requests from passing this one.
2789 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2790 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2792 * remove the space for the drain we added
2793 * so that we don't add it again
2795 --rq
->nr_phys_segments
;
2800 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2801 rq
->rq_flags
|= RQF_QUIET
;
2803 * Mark this request as started so we don't trigger
2804 * any debug logic in the end I/O path.
2806 blk_start_request(rq
);
2807 __blk_end_request_all(rq
, ret
== BLKPREP_INVALID
?
2808 BLK_STS_TARGET
: BLK_STS_IOERR
);
2810 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2817 EXPORT_SYMBOL(blk_peek_request
);
2819 static void blk_dequeue_request(struct request
*rq
)
2821 struct request_queue
*q
= rq
->q
;
2823 BUG_ON(list_empty(&rq
->queuelist
));
2824 BUG_ON(ELV_ON_HASH(rq
));
2826 list_del_init(&rq
->queuelist
);
2829 * the time frame between a request being removed from the lists
2830 * and to it is freed is accounted as io that is in progress at
2833 if (blk_account_rq(rq
)) {
2834 q
->in_flight
[rq_is_sync(rq
)]++;
2835 set_io_start_time_ns(rq
);
2840 * blk_start_request - start request processing on the driver
2841 * @req: request to dequeue
2844 * Dequeue @req and start timeout timer on it. This hands off the
2845 * request to the driver.
2847 void blk_start_request(struct request
*req
)
2849 lockdep_assert_held(req
->q
->queue_lock
);
2850 WARN_ON_ONCE(req
->q
->mq_ops
);
2852 blk_dequeue_request(req
);
2854 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2855 blk_stat_set_issue(&req
->issue_stat
, blk_rq_sectors(req
));
2856 req
->rq_flags
|= RQF_STATS
;
2857 wbt_issue(req
->q
->rq_wb
, &req
->issue_stat
);
2860 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2863 EXPORT_SYMBOL(blk_start_request
);
2866 * blk_fetch_request - fetch a request from a request queue
2867 * @q: request queue to fetch a request from
2870 * Return the request at the top of @q. The request is started on
2871 * return and LLD can start processing it immediately.
2874 * Pointer to the request at the top of @q if available. Null
2877 struct request
*blk_fetch_request(struct request_queue
*q
)
2881 lockdep_assert_held(q
->queue_lock
);
2882 WARN_ON_ONCE(q
->mq_ops
);
2884 rq
= blk_peek_request(q
);
2886 blk_start_request(rq
);
2889 EXPORT_SYMBOL(blk_fetch_request
);
2892 * Steal bios from a request and add them to a bio list.
2893 * The request must not have been partially completed before.
2895 void blk_steal_bios(struct bio_list
*list
, struct request
*rq
)
2899 list
->tail
->bi_next
= rq
->bio
;
2901 list
->head
= rq
->bio
;
2902 list
->tail
= rq
->biotail
;
2910 EXPORT_SYMBOL_GPL(blk_steal_bios
);
2913 * blk_update_request - Special helper function for request stacking drivers
2914 * @req: the request being processed
2915 * @error: block status code
2916 * @nr_bytes: number of bytes to complete @req
2919 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2920 * the request structure even if @req doesn't have leftover.
2921 * If @req has leftover, sets it up for the next range of segments.
2923 * This special helper function is only for request stacking drivers
2924 * (e.g. request-based dm) so that they can handle partial completion.
2925 * Actual device drivers should use blk_end_request instead.
2927 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2928 * %false return from this function.
2931 * %false - this request doesn't have any more data
2932 * %true - this request has more data
2934 bool blk_update_request(struct request
*req
, blk_status_t error
,
2935 unsigned int nr_bytes
)
2939 trace_block_rq_complete(req
, blk_status_to_errno(error
), nr_bytes
);
2944 if (unlikely(error
&& !blk_rq_is_passthrough(req
) &&
2945 !(req
->rq_flags
& RQF_QUIET
)))
2946 print_req_error(req
, error
);
2948 blk_account_io_completion(req
, nr_bytes
);
2952 struct bio
*bio
= req
->bio
;
2953 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2955 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2956 req
->bio
= bio
->bi_next
;
2958 /* Completion has already been traced */
2959 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
2960 req_bio_endio(req
, bio
, bio_bytes
, error
);
2962 total_bytes
+= bio_bytes
;
2963 nr_bytes
-= bio_bytes
;
2974 * Reset counters so that the request stacking driver
2975 * can find how many bytes remain in the request
2978 req
->__data_len
= 0;
2982 req
->__data_len
-= total_bytes
;
2984 /* update sector only for requests with clear definition of sector */
2985 if (!blk_rq_is_passthrough(req
))
2986 req
->__sector
+= total_bytes
>> 9;
2988 /* mixed attributes always follow the first bio */
2989 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2990 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2991 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2994 if (!(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
)) {
2996 * If total number of sectors is less than the first segment
2997 * size, something has gone terribly wrong.
2999 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
3000 blk_dump_rq_flags(req
, "request botched");
3001 req
->__data_len
= blk_rq_cur_bytes(req
);
3004 /* recalculate the number of segments */
3005 blk_recalc_rq_segments(req
);
3010 EXPORT_SYMBOL_GPL(blk_update_request
);
3012 static bool blk_update_bidi_request(struct request
*rq
, blk_status_t error
,
3013 unsigned int nr_bytes
,
3014 unsigned int bidi_bytes
)
3016 if (blk_update_request(rq
, error
, nr_bytes
))
3019 /* Bidi request must be completed as a whole */
3020 if (unlikely(blk_bidi_rq(rq
)) &&
3021 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
3024 if (blk_queue_add_random(rq
->q
))
3025 add_disk_randomness(rq
->rq_disk
);
3031 * blk_unprep_request - unprepare a request
3034 * This function makes a request ready for complete resubmission (or
3035 * completion). It happens only after all error handling is complete,
3036 * so represents the appropriate moment to deallocate any resources
3037 * that were allocated to the request in the prep_rq_fn. The queue
3038 * lock is held when calling this.
3040 void blk_unprep_request(struct request
*req
)
3042 struct request_queue
*q
= req
->q
;
3044 req
->rq_flags
&= ~RQF_DONTPREP
;
3045 if (q
->unprep_rq_fn
)
3046 q
->unprep_rq_fn(q
, req
);
3048 EXPORT_SYMBOL_GPL(blk_unprep_request
);
3050 void blk_finish_request(struct request
*req
, blk_status_t error
)
3052 struct request_queue
*q
= req
->q
;
3054 lockdep_assert_held(req
->q
->queue_lock
);
3055 WARN_ON_ONCE(q
->mq_ops
);
3057 if (req
->rq_flags
& RQF_STATS
)
3060 if (req
->rq_flags
& RQF_QUEUED
)
3061 blk_queue_end_tag(q
, req
);
3063 BUG_ON(blk_queued_rq(req
));
3065 if (unlikely(laptop_mode
) && !blk_rq_is_passthrough(req
))
3066 laptop_io_completion(req
->q
->backing_dev_info
);
3068 blk_delete_timer(req
);
3070 if (req
->rq_flags
& RQF_DONTPREP
)
3071 blk_unprep_request(req
);
3073 blk_account_io_done(req
);
3076 wbt_done(req
->q
->rq_wb
, &req
->issue_stat
);
3077 req
->end_io(req
, error
);
3079 if (blk_bidi_rq(req
))
3080 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
3082 __blk_put_request(q
, req
);
3085 EXPORT_SYMBOL(blk_finish_request
);
3088 * blk_end_bidi_request - Complete a bidi request
3089 * @rq: the request to complete
3090 * @error: block status code
3091 * @nr_bytes: number of bytes to complete @rq
3092 * @bidi_bytes: number of bytes to complete @rq->next_rq
3095 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
3096 * Drivers that supports bidi can safely call this member for any
3097 * type of request, bidi or uni. In the later case @bidi_bytes is
3101 * %false - we are done with this request
3102 * %true - still buffers pending for this request
3104 static bool blk_end_bidi_request(struct request
*rq
, blk_status_t error
,
3105 unsigned int nr_bytes
, unsigned int bidi_bytes
)
3107 struct request_queue
*q
= rq
->q
;
3108 unsigned long flags
;
3110 WARN_ON_ONCE(q
->mq_ops
);
3112 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
3115 spin_lock_irqsave(q
->queue_lock
, flags
);
3116 blk_finish_request(rq
, error
);
3117 spin_unlock_irqrestore(q
->queue_lock
, flags
);
3123 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3124 * @rq: the request to complete
3125 * @error: block status code
3126 * @nr_bytes: number of bytes to complete @rq
3127 * @bidi_bytes: number of bytes to complete @rq->next_rq
3130 * Identical to blk_end_bidi_request() except that queue lock is
3131 * assumed to be locked on entry and remains so on return.
3134 * %false - we are done with this request
3135 * %true - still buffers pending for this request
3137 static bool __blk_end_bidi_request(struct request
*rq
, blk_status_t error
,
3138 unsigned int nr_bytes
, unsigned int bidi_bytes
)
3140 lockdep_assert_held(rq
->q
->queue_lock
);
3141 WARN_ON_ONCE(rq
->q
->mq_ops
);
3143 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
3146 blk_finish_request(rq
, error
);
3152 * blk_end_request - Helper function for drivers to complete the request.
3153 * @rq: the request being processed
3154 * @error: block status code
3155 * @nr_bytes: number of bytes to complete
3158 * Ends I/O on a number of bytes attached to @rq.
3159 * If @rq has leftover, sets it up for the next range of segments.
3162 * %false - we are done with this request
3163 * %true - still buffers pending for this request
3165 bool blk_end_request(struct request
*rq
, blk_status_t error
,
3166 unsigned int nr_bytes
)
3168 WARN_ON_ONCE(rq
->q
->mq_ops
);
3169 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
3171 EXPORT_SYMBOL(blk_end_request
);
3174 * blk_end_request_all - Helper function for drives to finish the request.
3175 * @rq: the request to finish
3176 * @error: block status code
3179 * Completely finish @rq.
3181 void blk_end_request_all(struct request
*rq
, blk_status_t error
)
3184 unsigned int bidi_bytes
= 0;
3186 if (unlikely(blk_bidi_rq(rq
)))
3187 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
3189 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
3192 EXPORT_SYMBOL(blk_end_request_all
);
3195 * __blk_end_request - Helper function for drivers to complete the request.
3196 * @rq: the request being processed
3197 * @error: block status code
3198 * @nr_bytes: number of bytes to complete
3201 * Must be called with queue lock held unlike blk_end_request().
3204 * %false - we are done with this request
3205 * %true - still buffers pending for this request
3207 bool __blk_end_request(struct request
*rq
, blk_status_t error
,
3208 unsigned int nr_bytes
)
3210 lockdep_assert_held(rq
->q
->queue_lock
);
3211 WARN_ON_ONCE(rq
->q
->mq_ops
);
3213 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
3215 EXPORT_SYMBOL(__blk_end_request
);
3218 * __blk_end_request_all - Helper function for drives to finish the request.
3219 * @rq: the request to finish
3220 * @error: block status code
3223 * Completely finish @rq. Must be called with queue lock held.
3225 void __blk_end_request_all(struct request
*rq
, blk_status_t error
)
3228 unsigned int bidi_bytes
= 0;
3230 lockdep_assert_held(rq
->q
->queue_lock
);
3231 WARN_ON_ONCE(rq
->q
->mq_ops
);
3233 if (unlikely(blk_bidi_rq(rq
)))
3234 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
3236 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
3239 EXPORT_SYMBOL(__blk_end_request_all
);
3242 * __blk_end_request_cur - Helper function to finish the current request chunk.
3243 * @rq: the request to finish the current chunk for
3244 * @error: block status code
3247 * Complete the current consecutively mapped chunk from @rq. Must
3248 * be called with queue lock held.
3251 * %false - we are done with this request
3252 * %true - still buffers pending for this request
3254 bool __blk_end_request_cur(struct request
*rq
, blk_status_t error
)
3256 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
3258 EXPORT_SYMBOL(__blk_end_request_cur
);
3260 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
3263 if (bio_has_data(bio
))
3264 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3265 else if (bio_op(bio
) == REQ_OP_DISCARD
)
3266 rq
->nr_phys_segments
= 1;
3268 rq
->__data_len
= bio
->bi_iter
.bi_size
;
3269 rq
->bio
= rq
->biotail
= bio
;
3272 rq
->rq_disk
= bio
->bi_disk
;
3275 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3277 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3278 * @rq: the request to be flushed
3281 * Flush all pages in @rq.
3283 void rq_flush_dcache_pages(struct request
*rq
)
3285 struct req_iterator iter
;
3286 struct bio_vec bvec
;
3288 rq_for_each_segment(bvec
, rq
, iter
)
3289 flush_dcache_page(bvec
.bv_page
);
3291 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
3295 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3296 * @q : the queue of the device being checked
3299 * Check if underlying low-level drivers of a device are busy.
3300 * If the drivers want to export their busy state, they must set own
3301 * exporting function using blk_queue_lld_busy() first.
3303 * Basically, this function is used only by request stacking drivers
3304 * to stop dispatching requests to underlying devices when underlying
3305 * devices are busy. This behavior helps more I/O merging on the queue
3306 * of the request stacking driver and prevents I/O throughput regression
3307 * on burst I/O load.
3310 * 0 - Not busy (The request stacking driver should dispatch request)
3311 * 1 - Busy (The request stacking driver should stop dispatching request)
3313 int blk_lld_busy(struct request_queue
*q
)
3316 return q
->lld_busy_fn(q
);
3320 EXPORT_SYMBOL_GPL(blk_lld_busy
);
3323 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3324 * @rq: the clone request to be cleaned up
3327 * Free all bios in @rq for a cloned request.
3329 void blk_rq_unprep_clone(struct request
*rq
)
3333 while ((bio
= rq
->bio
) != NULL
) {
3334 rq
->bio
= bio
->bi_next
;
3339 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3342 * Copy attributes of the original request to the clone request.
3343 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3345 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3347 dst
->cpu
= src
->cpu
;
3348 dst
->__sector
= blk_rq_pos(src
);
3349 dst
->__data_len
= blk_rq_bytes(src
);
3350 if (src
->rq_flags
& RQF_SPECIAL_PAYLOAD
) {
3351 dst
->rq_flags
|= RQF_SPECIAL_PAYLOAD
;
3352 dst
->special_vec
= src
->special_vec
;
3354 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3355 dst
->ioprio
= src
->ioprio
;
3356 dst
->extra_len
= src
->extra_len
;
3360 * blk_rq_prep_clone - Helper function to setup clone request
3361 * @rq: the request to be setup
3362 * @rq_src: original request to be cloned
3363 * @bs: bio_set that bios for clone are allocated from
3364 * @gfp_mask: memory allocation mask for bio
3365 * @bio_ctr: setup function to be called for each clone bio.
3366 * Returns %0 for success, non %0 for failure.
3367 * @data: private data to be passed to @bio_ctr
3370 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3371 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3372 * are not copied, and copying such parts is the caller's responsibility.
3373 * Also, pages which the original bios are pointing to are not copied
3374 * and the cloned bios just point same pages.
3375 * So cloned bios must be completed before original bios, which means
3376 * the caller must complete @rq before @rq_src.
3378 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3379 struct bio_set
*bs
, gfp_t gfp_mask
,
3380 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3383 struct bio
*bio
, *bio_src
;
3388 __rq_for_each_bio(bio_src
, rq_src
) {
3389 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3393 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3397 rq
->biotail
->bi_next
= bio
;
3400 rq
->bio
= rq
->biotail
= bio
;
3403 __blk_rq_prep_clone(rq
, rq_src
);
3410 blk_rq_unprep_clone(rq
);
3414 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3416 int kblockd_schedule_work(struct work_struct
*work
)
3418 return queue_work(kblockd_workqueue
, work
);
3420 EXPORT_SYMBOL(kblockd_schedule_work
);
3422 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3424 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3426 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3428 int kblockd_mod_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3429 unsigned long delay
)
3431 return mod_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3433 EXPORT_SYMBOL(kblockd_mod_delayed_work_on
);
3435 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3436 unsigned long delay
)
3438 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3440 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3442 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3443 unsigned long delay
)
3445 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3447 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3450 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3451 * @plug: The &struct blk_plug that needs to be initialized
3454 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3455 * pending I/O should the task end up blocking between blk_start_plug() and
3456 * blk_finish_plug(). This is important from a performance perspective, but
3457 * also ensures that we don't deadlock. For instance, if the task is blocking
3458 * for a memory allocation, memory reclaim could end up wanting to free a
3459 * page belonging to that request that is currently residing in our private
3460 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3461 * this kind of deadlock.
3463 void blk_start_plug(struct blk_plug
*plug
)
3465 struct task_struct
*tsk
= current
;
3468 * If this is a nested plug, don't actually assign it.
3473 INIT_LIST_HEAD(&plug
->list
);
3474 INIT_LIST_HEAD(&plug
->mq_list
);
3475 INIT_LIST_HEAD(&plug
->cb_list
);
3477 * Store ordering should not be needed here, since a potential
3478 * preempt will imply a full memory barrier
3482 EXPORT_SYMBOL(blk_start_plug
);
3484 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3486 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3487 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3489 return !(rqa
->q
< rqb
->q
||
3490 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3494 * If 'from_schedule' is true, then postpone the dispatch of requests
3495 * until a safe kblockd context. We due this to avoid accidental big
3496 * additional stack usage in driver dispatch, in places where the originally
3497 * plugger did not intend it.
3499 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3501 __releases(q
->queue_lock
)
3503 lockdep_assert_held(q
->queue_lock
);
3505 trace_block_unplug(q
, depth
, !from_schedule
);
3508 blk_run_queue_async(q
);
3511 spin_unlock(q
->queue_lock
);
3514 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3516 LIST_HEAD(callbacks
);
3518 while (!list_empty(&plug
->cb_list
)) {
3519 list_splice_init(&plug
->cb_list
, &callbacks
);
3521 while (!list_empty(&callbacks
)) {
3522 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3525 list_del(&cb
->list
);
3526 cb
->callback(cb
, from_schedule
);
3531 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3534 struct blk_plug
*plug
= current
->plug
;
3535 struct blk_plug_cb
*cb
;
3540 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3541 if (cb
->callback
== unplug
&& cb
->data
== data
)
3544 /* Not currently on the callback list */
3545 BUG_ON(size
< sizeof(*cb
));
3546 cb
= kzalloc(size
, GFP_ATOMIC
);
3549 cb
->callback
= unplug
;
3550 list_add(&cb
->list
, &plug
->cb_list
);
3554 EXPORT_SYMBOL(blk_check_plugged
);
3556 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3558 struct request_queue
*q
;
3559 unsigned long flags
;
3564 flush_plug_callbacks(plug
, from_schedule
);
3566 if (!list_empty(&plug
->mq_list
))
3567 blk_mq_flush_plug_list(plug
, from_schedule
);
3569 if (list_empty(&plug
->list
))
3572 list_splice_init(&plug
->list
, &list
);
3574 list_sort(NULL
, &list
, plug_rq_cmp
);
3580 * Save and disable interrupts here, to avoid doing it for every
3581 * queue lock we have to take.
3583 local_irq_save(flags
);
3584 while (!list_empty(&list
)) {
3585 rq
= list_entry_rq(list
.next
);
3586 list_del_init(&rq
->queuelist
);
3590 * This drops the queue lock
3593 queue_unplugged(q
, depth
, from_schedule
);
3596 spin_lock(q
->queue_lock
);
3600 * Short-circuit if @q is dead
3602 if (unlikely(blk_queue_dying(q
))) {
3603 __blk_end_request_all(rq
, BLK_STS_IOERR
);
3608 * rq is already accounted, so use raw insert
3610 if (op_is_flush(rq
->cmd_flags
))
3611 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3613 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3619 * This drops the queue lock
3622 queue_unplugged(q
, depth
, from_schedule
);
3624 local_irq_restore(flags
);
3627 void blk_finish_plug(struct blk_plug
*plug
)
3629 if (plug
!= current
->plug
)
3631 blk_flush_plug_list(plug
, false);
3633 current
->plug
= NULL
;
3635 EXPORT_SYMBOL(blk_finish_plug
);
3639 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3640 * @q: the queue of the device
3641 * @dev: the device the queue belongs to
3644 * Initialize runtime-PM-related fields for @q and start auto suspend for
3645 * @dev. Drivers that want to take advantage of request-based runtime PM
3646 * should call this function after @dev has been initialized, and its
3647 * request queue @q has been allocated, and runtime PM for it can not happen
3648 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3649 * cases, driver should call this function before any I/O has taken place.
3651 * This function takes care of setting up using auto suspend for the device,
3652 * the autosuspend delay is set to -1 to make runtime suspend impossible
3653 * until an updated value is either set by user or by driver. Drivers do
3654 * not need to touch other autosuspend settings.
3656 * The block layer runtime PM is request based, so only works for drivers
3657 * that use request as their IO unit instead of those directly use bio's.
3659 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3661 /* Don't enable runtime PM for blk-mq until it is ready */
3663 pm_runtime_disable(dev
);
3668 q
->rpm_status
= RPM_ACTIVE
;
3669 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3670 pm_runtime_use_autosuspend(q
->dev
);
3672 EXPORT_SYMBOL(blk_pm_runtime_init
);
3675 * blk_pre_runtime_suspend - Pre runtime suspend check
3676 * @q: the queue of the device
3679 * This function will check if runtime suspend is allowed for the device
3680 * by examining if there are any requests pending in the queue. If there
3681 * are requests pending, the device can not be runtime suspended; otherwise,
3682 * the queue's status will be updated to SUSPENDING and the driver can
3683 * proceed to suspend the device.
3685 * For the not allowed case, we mark last busy for the device so that
3686 * runtime PM core will try to autosuspend it some time later.
3688 * This function should be called near the start of the device's
3689 * runtime_suspend callback.
3692 * 0 - OK to runtime suspend the device
3693 * -EBUSY - Device should not be runtime suspended
3695 int blk_pre_runtime_suspend(struct request_queue
*q
)
3702 spin_lock_irq(q
->queue_lock
);
3703 if (q
->nr_pending
) {
3705 pm_runtime_mark_last_busy(q
->dev
);
3707 q
->rpm_status
= RPM_SUSPENDING
;
3709 spin_unlock_irq(q
->queue_lock
);
3712 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3715 * blk_post_runtime_suspend - Post runtime suspend processing
3716 * @q: the queue of the device
3717 * @err: return value of the device's runtime_suspend function
3720 * Update the queue's runtime status according to the return value of the
3721 * device's runtime suspend function and mark last busy for the device so
3722 * that PM core will try to auto suspend the device at a later time.
3724 * This function should be called near the end of the device's
3725 * runtime_suspend callback.
3727 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3732 spin_lock_irq(q
->queue_lock
);
3734 q
->rpm_status
= RPM_SUSPENDED
;
3736 q
->rpm_status
= RPM_ACTIVE
;
3737 pm_runtime_mark_last_busy(q
->dev
);
3739 spin_unlock_irq(q
->queue_lock
);
3741 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3744 * blk_pre_runtime_resume - Pre runtime resume processing
3745 * @q: the queue of the device
3748 * Update the queue's runtime status to RESUMING in preparation for the
3749 * runtime resume of the device.
3751 * This function should be called near the start of the device's
3752 * runtime_resume callback.
3754 void blk_pre_runtime_resume(struct request_queue
*q
)
3759 spin_lock_irq(q
->queue_lock
);
3760 q
->rpm_status
= RPM_RESUMING
;
3761 spin_unlock_irq(q
->queue_lock
);
3763 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3766 * blk_post_runtime_resume - Post runtime resume processing
3767 * @q: the queue of the device
3768 * @err: return value of the device's runtime_resume function
3771 * Update the queue's runtime status according to the return value of the
3772 * device's runtime_resume function. If it is successfully resumed, process
3773 * the requests that are queued into the device's queue when it is resuming
3774 * and then mark last busy and initiate autosuspend for it.
3776 * This function should be called near the end of the device's
3777 * runtime_resume callback.
3779 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3784 spin_lock_irq(q
->queue_lock
);
3786 q
->rpm_status
= RPM_ACTIVE
;
3788 pm_runtime_mark_last_busy(q
->dev
);
3789 pm_request_autosuspend(q
->dev
);
3791 q
->rpm_status
= RPM_SUSPENDED
;
3793 spin_unlock_irq(q
->queue_lock
);
3795 EXPORT_SYMBOL(blk_post_runtime_resume
);
3798 * blk_set_runtime_active - Force runtime status of the queue to be active
3799 * @q: the queue of the device
3801 * If the device is left runtime suspended during system suspend the resume
3802 * hook typically resumes the device and corrects runtime status
3803 * accordingly. However, that does not affect the queue runtime PM status
3804 * which is still "suspended". This prevents processing requests from the
3807 * This function can be used in driver's resume hook to correct queue
3808 * runtime PM status and re-enable peeking requests from the queue. It
3809 * should be called before first request is added to the queue.
3811 void blk_set_runtime_active(struct request_queue
*q
)
3813 spin_lock_irq(q
->queue_lock
);
3814 q
->rpm_status
= RPM_ACTIVE
;
3815 pm_runtime_mark_last_busy(q
->dev
);
3816 pm_request_autosuspend(q
->dev
);
3817 spin_unlock_irq(q
->queue_lock
);
3819 EXPORT_SYMBOL(blk_set_runtime_active
);
3822 int __init
blk_dev_init(void)
3824 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
3825 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3826 FIELD_SIZEOF(struct request
, cmd_flags
));
3827 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3828 FIELD_SIZEOF(struct bio
, bi_opf
));
3830 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3831 kblockd_workqueue
= alloc_workqueue("kblockd",
3832 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3833 if (!kblockd_workqueue
)
3834 panic("Failed to create kblockd\n");
3836 request_cachep
= kmem_cache_create("blkdev_requests",
3837 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3839 blk_requestq_cachep
= kmem_cache_create("request_queue",
3840 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
3842 #ifdef CONFIG_DEBUG_FS
3843 blk_debugfs_root
= debugfs_create_dir("block", NULL
);