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 queue_for_each_hw_ctx(q
, hctx
, i
)
343 cancel_delayed_work_sync(&hctx
->run_work
);
345 cancel_delayed_work_sync(&q
->delay_work
);
348 EXPORT_SYMBOL(blk_sync_queue
);
351 * blk_set_preempt_only - set QUEUE_FLAG_PREEMPT_ONLY
352 * @q: request queue pointer
354 * Returns the previous value of the PREEMPT_ONLY flag - 0 if the flag was not
355 * set and 1 if the flag was already set.
357 int blk_set_preempt_only(struct request_queue
*q
)
362 spin_lock_irqsave(q
->queue_lock
, flags
);
363 res
= queue_flag_test_and_set(QUEUE_FLAG_PREEMPT_ONLY
, q
);
364 spin_unlock_irqrestore(q
->queue_lock
, flags
);
368 EXPORT_SYMBOL_GPL(blk_set_preempt_only
);
370 void blk_clear_preempt_only(struct request_queue
*q
)
374 spin_lock_irqsave(q
->queue_lock
, flags
);
375 queue_flag_clear(QUEUE_FLAG_PREEMPT_ONLY
, q
);
376 wake_up_all(&q
->mq_freeze_wq
);
377 spin_unlock_irqrestore(q
->queue_lock
, flags
);
379 EXPORT_SYMBOL_GPL(blk_clear_preempt_only
);
382 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
383 * @q: The queue to run
386 * Invoke request handling on a queue if there are any pending requests.
387 * May be used to restart request handling after a request has completed.
388 * This variant runs the queue whether or not the queue has been
389 * stopped. Must be called with the queue lock held and interrupts
390 * disabled. See also @blk_run_queue.
392 inline void __blk_run_queue_uncond(struct request_queue
*q
)
394 lockdep_assert_held(q
->queue_lock
);
395 WARN_ON_ONCE(q
->mq_ops
);
397 if (unlikely(blk_queue_dead(q
)))
401 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
402 * the queue lock internally. As a result multiple threads may be
403 * running such a request function concurrently. Keep track of the
404 * number of active request_fn invocations such that blk_drain_queue()
405 * can wait until all these request_fn calls have finished.
407 q
->request_fn_active
++;
409 q
->request_fn_active
--;
411 EXPORT_SYMBOL_GPL(__blk_run_queue_uncond
);
414 * __blk_run_queue - run a single device queue
415 * @q: The queue to run
418 * See @blk_run_queue.
420 void __blk_run_queue(struct request_queue
*q
)
422 lockdep_assert_held(q
->queue_lock
);
423 WARN_ON_ONCE(q
->mq_ops
);
425 if (unlikely(blk_queue_stopped(q
)))
428 __blk_run_queue_uncond(q
);
430 EXPORT_SYMBOL(__blk_run_queue
);
433 * blk_run_queue_async - run a single device queue in workqueue context
434 * @q: The queue to run
437 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
441 * Since it is not allowed to run q->delay_work after blk_cleanup_queue()
442 * has canceled q->delay_work, callers must hold the queue lock to avoid
443 * race conditions between blk_cleanup_queue() and blk_run_queue_async().
445 void blk_run_queue_async(struct request_queue
*q
)
447 lockdep_assert_held(q
->queue_lock
);
448 WARN_ON_ONCE(q
->mq_ops
);
450 if (likely(!blk_queue_stopped(q
) && !blk_queue_dead(q
)))
451 mod_delayed_work(kblockd_workqueue
, &q
->delay_work
, 0);
453 EXPORT_SYMBOL(blk_run_queue_async
);
456 * blk_run_queue - run a single device queue
457 * @q: The queue to run
460 * Invoke request handling on this queue, if it has pending work to do.
461 * May be used to restart queueing when a request has completed.
463 void blk_run_queue(struct request_queue
*q
)
467 WARN_ON_ONCE(q
->mq_ops
);
469 spin_lock_irqsave(q
->queue_lock
, flags
);
471 spin_unlock_irqrestore(q
->queue_lock
, flags
);
473 EXPORT_SYMBOL(blk_run_queue
);
475 void blk_put_queue(struct request_queue
*q
)
477 kobject_put(&q
->kobj
);
479 EXPORT_SYMBOL(blk_put_queue
);
482 * __blk_drain_queue - drain requests from request_queue
484 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
486 * Drain requests from @q. If @drain_all is set, all requests are drained.
487 * If not, only ELVPRIV requests are drained. The caller is responsible
488 * for ensuring that no new requests which need to be drained are queued.
490 static void __blk_drain_queue(struct request_queue
*q
, bool drain_all
)
491 __releases(q
->queue_lock
)
492 __acquires(q
->queue_lock
)
496 lockdep_assert_held(q
->queue_lock
);
497 WARN_ON_ONCE(q
->mq_ops
);
503 * The caller might be trying to drain @q before its
504 * elevator is initialized.
507 elv_drain_elevator(q
);
509 blkcg_drain_queue(q
);
512 * This function might be called on a queue which failed
513 * driver init after queue creation or is not yet fully
514 * active yet. Some drivers (e.g. fd and loop) get unhappy
515 * in such cases. Kick queue iff dispatch queue has
516 * something on it and @q has request_fn set.
518 if (!list_empty(&q
->queue_head
) && q
->request_fn
)
521 drain
|= q
->nr_rqs_elvpriv
;
522 drain
|= q
->request_fn_active
;
525 * Unfortunately, requests are queued at and tracked from
526 * multiple places and there's no single counter which can
527 * be drained. Check all the queues and counters.
530 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
531 drain
|= !list_empty(&q
->queue_head
);
532 for (i
= 0; i
< 2; i
++) {
533 drain
|= q
->nr_rqs
[i
];
534 drain
|= q
->in_flight
[i
];
536 drain
|= !list_empty(&fq
->flush_queue
[i
]);
543 spin_unlock_irq(q
->queue_lock
);
547 spin_lock_irq(q
->queue_lock
);
551 * With queue marked dead, any woken up waiter will fail the
552 * allocation path, so the wakeup chaining is lost and we're
553 * left with hung waiters. We need to wake up those waiters.
556 struct request_list
*rl
;
558 blk_queue_for_each_rl(rl
, q
)
559 for (i
= 0; i
< ARRAY_SIZE(rl
->wait
); i
++)
560 wake_up_all(&rl
->wait
[i
]);
564 void blk_drain_queue(struct request_queue
*q
)
566 spin_lock_irq(q
->queue_lock
);
567 __blk_drain_queue(q
, true);
568 spin_unlock_irq(q
->queue_lock
);
572 * blk_queue_bypass_start - enter queue bypass mode
573 * @q: queue of interest
575 * In bypass mode, only the dispatch FIFO queue of @q is used. This
576 * function makes @q enter bypass mode and drains all requests which were
577 * throttled or issued before. On return, it's guaranteed that no request
578 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
579 * inside queue or RCU read lock.
581 void blk_queue_bypass_start(struct request_queue
*q
)
583 WARN_ON_ONCE(q
->mq_ops
);
585 spin_lock_irq(q
->queue_lock
);
587 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
588 spin_unlock_irq(q
->queue_lock
);
591 * Queues start drained. Skip actual draining till init is
592 * complete. This avoids lenghty delays during queue init which
593 * can happen many times during boot.
595 if (blk_queue_init_done(q
)) {
596 spin_lock_irq(q
->queue_lock
);
597 __blk_drain_queue(q
, false);
598 spin_unlock_irq(q
->queue_lock
);
600 /* ensure blk_queue_bypass() is %true inside RCU read lock */
604 EXPORT_SYMBOL_GPL(blk_queue_bypass_start
);
607 * blk_queue_bypass_end - leave queue bypass mode
608 * @q: queue of interest
610 * Leave bypass mode and restore the normal queueing behavior.
612 * Note: although blk_queue_bypass_start() is only called for blk-sq queues,
613 * this function is called for both blk-sq and blk-mq queues.
615 void blk_queue_bypass_end(struct request_queue
*q
)
617 spin_lock_irq(q
->queue_lock
);
618 if (!--q
->bypass_depth
)
619 queue_flag_clear(QUEUE_FLAG_BYPASS
, q
);
620 WARN_ON_ONCE(q
->bypass_depth
< 0);
621 spin_unlock_irq(q
->queue_lock
);
623 EXPORT_SYMBOL_GPL(blk_queue_bypass_end
);
625 void blk_set_queue_dying(struct request_queue
*q
)
627 spin_lock_irq(q
->queue_lock
);
628 queue_flag_set(QUEUE_FLAG_DYING
, q
);
629 spin_unlock_irq(q
->queue_lock
);
632 * When queue DYING flag is set, we need to block new req
633 * entering queue, so we call blk_freeze_queue_start() to
634 * prevent I/O from crossing blk_queue_enter().
636 blk_freeze_queue_start(q
);
639 blk_mq_wake_waiters(q
);
641 struct request_list
*rl
;
643 spin_lock_irq(q
->queue_lock
);
644 blk_queue_for_each_rl(rl
, q
) {
646 wake_up_all(&rl
->wait
[BLK_RW_SYNC
]);
647 wake_up_all(&rl
->wait
[BLK_RW_ASYNC
]);
650 spin_unlock_irq(q
->queue_lock
);
653 /* Make blk_queue_enter() reexamine the DYING flag. */
654 wake_up_all(&q
->mq_freeze_wq
);
656 EXPORT_SYMBOL_GPL(blk_set_queue_dying
);
659 * blk_cleanup_queue - shutdown a request queue
660 * @q: request queue to shutdown
662 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
663 * put it. All future requests will be failed immediately with -ENODEV.
665 void blk_cleanup_queue(struct request_queue
*q
)
667 spinlock_t
*lock
= q
->queue_lock
;
669 /* mark @q DYING, no new request or merges will be allowed afterwards */
670 mutex_lock(&q
->sysfs_lock
);
671 blk_set_queue_dying(q
);
675 * A dying queue is permanently in bypass mode till released. Note
676 * that, unlike blk_queue_bypass_start(), we aren't performing
677 * synchronize_rcu() after entering bypass mode to avoid the delay
678 * as some drivers create and destroy a lot of queues while
679 * probing. This is still safe because blk_release_queue() will be
680 * called only after the queue refcnt drops to zero and nothing,
681 * RCU or not, would be traversing the queue by then.
684 queue_flag_set(QUEUE_FLAG_BYPASS
, q
);
686 queue_flag_set(QUEUE_FLAG_NOMERGES
, q
);
687 queue_flag_set(QUEUE_FLAG_NOXMERGES
, q
);
688 queue_flag_set(QUEUE_FLAG_DYING
, q
);
689 spin_unlock_irq(lock
);
690 mutex_unlock(&q
->sysfs_lock
);
693 * Drain all requests queued before DYING marking. Set DEAD flag to
694 * prevent that q->request_fn() gets invoked after draining finished.
698 queue_flag_set(QUEUE_FLAG_DEAD
, q
);
699 spin_unlock_irq(lock
);
702 * make sure all in-progress dispatch are completed because
703 * blk_freeze_queue() can only complete all requests, and
704 * dispatch may still be in-progress since we dispatch requests
705 * from more than one contexts.
707 * We rely on driver to deal with the race in case that queue
708 * initialization isn't done.
710 if (q
->mq_ops
&& blk_queue_init_done(q
))
711 blk_mq_quiesce_queue(q
);
713 /* for synchronous bio-based driver finish in-flight integrity i/o */
714 blk_flush_integrity();
716 /* @q won't process any more request, flush async actions */
717 del_timer_sync(&q
->backing_dev_info
->laptop_mode_wb_timer
);
721 blk_mq_exit_queue(q
);
723 percpu_ref_exit(&q
->q_usage_counter
);
726 if (q
->queue_lock
!= &q
->__queue_lock
)
727 q
->queue_lock
= &q
->__queue_lock
;
728 spin_unlock_irq(lock
);
730 /* @q is and will stay empty, shutdown and put */
733 EXPORT_SYMBOL(blk_cleanup_queue
);
735 /* Allocate memory local to the request queue */
736 static void *alloc_request_simple(gfp_t gfp_mask
, void *data
)
738 struct request_queue
*q
= data
;
740 return kmem_cache_alloc_node(request_cachep
, gfp_mask
, q
->node
);
743 static void free_request_simple(void *element
, void *data
)
745 kmem_cache_free(request_cachep
, element
);
748 static void *alloc_request_size(gfp_t gfp_mask
, void *data
)
750 struct request_queue
*q
= data
;
753 rq
= kmalloc_node(sizeof(struct request
) + q
->cmd_size
, gfp_mask
,
755 if (rq
&& q
->init_rq_fn
&& q
->init_rq_fn(q
, rq
, gfp_mask
) < 0) {
762 static void free_request_size(void *element
, void *data
)
764 struct request_queue
*q
= data
;
767 q
->exit_rq_fn(q
, element
);
771 int blk_init_rl(struct request_list
*rl
, struct request_queue
*q
,
774 if (unlikely(rl
->rq_pool
) || q
->mq_ops
)
778 rl
->count
[BLK_RW_SYNC
] = rl
->count
[BLK_RW_ASYNC
] = 0;
779 rl
->starved
[BLK_RW_SYNC
] = rl
->starved
[BLK_RW_ASYNC
] = 0;
780 init_waitqueue_head(&rl
->wait
[BLK_RW_SYNC
]);
781 init_waitqueue_head(&rl
->wait
[BLK_RW_ASYNC
]);
784 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
785 alloc_request_size
, free_request_size
,
786 q
, gfp_mask
, q
->node
);
788 rl
->rq_pool
= mempool_create_node(BLKDEV_MIN_RQ
,
789 alloc_request_simple
, free_request_simple
,
790 q
, gfp_mask
, q
->node
);
795 if (rl
!= &q
->root_rl
)
796 WARN_ON_ONCE(!blk_get_queue(q
));
801 void blk_exit_rl(struct request_queue
*q
, struct request_list
*rl
)
804 mempool_destroy(rl
->rq_pool
);
805 if (rl
!= &q
->root_rl
)
810 struct request_queue
*blk_alloc_queue(gfp_t gfp_mask
)
812 return blk_alloc_queue_node(gfp_mask
, NUMA_NO_NODE
);
814 EXPORT_SYMBOL(blk_alloc_queue
);
817 * blk_queue_enter() - try to increase q->q_usage_counter
818 * @q: request queue pointer
819 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
821 int blk_queue_enter(struct request_queue
*q
, blk_mq_req_flags_t flags
)
823 const bool preempt
= flags
& BLK_MQ_REQ_PREEMPT
;
826 bool success
= false;
829 if (percpu_ref_tryget_live(&q
->q_usage_counter
)) {
831 * The code that sets the PREEMPT_ONLY flag is
832 * responsible for ensuring that that flag is globally
833 * visible before the queue is unfrozen.
835 if (preempt
|| !blk_queue_preempt_only(q
)) {
838 percpu_ref_put(&q
->q_usage_counter
);
846 if (flags
& BLK_MQ_REQ_NOWAIT
)
850 * read pair of barrier in blk_freeze_queue_start(),
851 * we need to order reading __PERCPU_REF_DEAD flag of
852 * .q_usage_counter and reading .mq_freeze_depth or
853 * queue dying flag, otherwise the following wait may
854 * never return if the two reads are reordered.
858 wait_event(q
->mq_freeze_wq
,
859 (atomic_read(&q
->mq_freeze_depth
) == 0 &&
860 (preempt
|| !blk_queue_preempt_only(q
))) ||
862 if (blk_queue_dying(q
))
867 void blk_queue_exit(struct request_queue
*q
)
869 percpu_ref_put(&q
->q_usage_counter
);
872 static void blk_queue_usage_counter_release(struct percpu_ref
*ref
)
874 struct request_queue
*q
=
875 container_of(ref
, struct request_queue
, q_usage_counter
);
877 wake_up_all(&q
->mq_freeze_wq
);
880 static void blk_rq_timed_out_timer(struct timer_list
*t
)
882 struct request_queue
*q
= from_timer(q
, t
, timeout
);
884 kblockd_schedule_work(&q
->timeout_work
);
887 static void blk_timeout_work_dummy(struct work_struct
*work
)
891 struct request_queue
*blk_alloc_queue_node(gfp_t gfp_mask
, int node_id
)
893 struct request_queue
*q
;
895 q
= kmem_cache_alloc_node(blk_requestq_cachep
,
896 gfp_mask
| __GFP_ZERO
, node_id
);
900 q
->id
= ida_simple_get(&blk_queue_ida
, 0, 0, gfp_mask
);
904 q
->bio_split
= bioset_create(BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
);
908 q
->backing_dev_info
= bdi_alloc_node(gfp_mask
, node_id
);
909 if (!q
->backing_dev_info
)
912 q
->stats
= blk_alloc_queue_stats();
916 q
->backing_dev_info
->ra_pages
=
917 (VM_MAX_READAHEAD
* 1024) / PAGE_SIZE
;
918 q
->backing_dev_info
->capabilities
= BDI_CAP_CGROUP_WRITEBACK
;
919 q
->backing_dev_info
->name
= "block";
922 timer_setup(&q
->backing_dev_info
->laptop_mode_wb_timer
,
923 laptop_mode_timer_fn
, 0);
924 timer_setup(&q
->timeout
, blk_rq_timed_out_timer
, 0);
925 INIT_WORK(&q
->timeout_work
, blk_timeout_work_dummy
);
926 INIT_LIST_HEAD(&q
->queue_head
);
927 INIT_LIST_HEAD(&q
->timeout_list
);
928 INIT_LIST_HEAD(&q
->icq_list
);
929 #ifdef CONFIG_BLK_CGROUP
930 INIT_LIST_HEAD(&q
->blkg_list
);
932 INIT_DELAYED_WORK(&q
->delay_work
, blk_delay_work
);
934 kobject_init(&q
->kobj
, &blk_queue_ktype
);
936 #ifdef CONFIG_BLK_DEV_IO_TRACE
937 mutex_init(&q
->blk_trace_mutex
);
939 mutex_init(&q
->sysfs_lock
);
940 spin_lock_init(&q
->__queue_lock
);
943 * By default initialize queue_lock to internal lock and driver can
944 * override it later if need be.
946 q
->queue_lock
= &q
->__queue_lock
;
949 * A queue starts its life with bypass turned on to avoid
950 * unnecessary bypass on/off overhead and nasty surprises during
951 * init. The initial bypass will be finished when the queue is
952 * registered by blk_register_queue().
955 __set_bit(QUEUE_FLAG_BYPASS
, &q
->queue_flags
);
957 init_waitqueue_head(&q
->mq_freeze_wq
);
960 * Init percpu_ref in atomic mode so that it's faster to shutdown.
961 * See blk_register_queue() for details.
963 if (percpu_ref_init(&q
->q_usage_counter
,
964 blk_queue_usage_counter_release
,
965 PERCPU_REF_INIT_ATOMIC
, GFP_KERNEL
))
968 if (blkcg_init_queue(q
))
974 percpu_ref_exit(&q
->q_usage_counter
);
976 blk_free_queue_stats(q
->stats
);
978 bdi_put(q
->backing_dev_info
);
980 bioset_free(q
->bio_split
);
982 ida_simple_remove(&blk_queue_ida
, q
->id
);
984 kmem_cache_free(blk_requestq_cachep
, q
);
987 EXPORT_SYMBOL(blk_alloc_queue_node
);
990 * blk_init_queue - prepare a request queue for use with a block device
991 * @rfn: The function to be called to process requests that have been
992 * placed on the queue.
993 * @lock: Request queue spin lock
996 * If a block device wishes to use the standard request handling procedures,
997 * which sorts requests and coalesces adjacent requests, then it must
998 * call blk_init_queue(). The function @rfn will be called when there
999 * are requests on the queue that need to be processed. If the device
1000 * supports plugging, then @rfn may not be called immediately when requests
1001 * are available on the queue, but may be called at some time later instead.
1002 * Plugged queues are generally unplugged when a buffer belonging to one
1003 * of the requests on the queue is needed, or due to memory pressure.
1005 * @rfn is not required, or even expected, to remove all requests off the
1006 * queue, but only as many as it can handle at a time. If it does leave
1007 * requests on the queue, it is responsible for arranging that the requests
1008 * get dealt with eventually.
1010 * The queue spin lock must be held while manipulating the requests on the
1011 * request queue; this lock will be taken also from interrupt context, so irq
1012 * disabling is needed for it.
1014 * Function returns a pointer to the initialized request queue, or %NULL if
1015 * it didn't succeed.
1018 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1019 * when the block device is deactivated (such as at module unload).
1022 struct request_queue
*blk_init_queue(request_fn_proc
*rfn
, spinlock_t
*lock
)
1024 return blk_init_queue_node(rfn
, lock
, NUMA_NO_NODE
);
1026 EXPORT_SYMBOL(blk_init_queue
);
1028 struct request_queue
*
1029 blk_init_queue_node(request_fn_proc
*rfn
, spinlock_t
*lock
, int node_id
)
1031 struct request_queue
*q
;
1033 q
= blk_alloc_queue_node(GFP_KERNEL
, node_id
);
1037 q
->request_fn
= rfn
;
1039 q
->queue_lock
= lock
;
1040 if (blk_init_allocated_queue(q
) < 0) {
1041 blk_cleanup_queue(q
);
1047 EXPORT_SYMBOL(blk_init_queue_node
);
1049 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
);
1052 int blk_init_allocated_queue(struct request_queue
*q
)
1054 WARN_ON_ONCE(q
->mq_ops
);
1056 q
->fq
= blk_alloc_flush_queue(q
, NUMA_NO_NODE
, q
->cmd_size
, GFP_KERNEL
);
1060 if (q
->init_rq_fn
&& q
->init_rq_fn(q
, q
->fq
->flush_rq
, GFP_KERNEL
))
1061 goto out_free_flush_queue
;
1063 if (blk_init_rl(&q
->root_rl
, q
, GFP_KERNEL
))
1064 goto out_exit_flush_rq
;
1066 INIT_WORK(&q
->timeout_work
, blk_timeout_work
);
1067 q
->queue_flags
|= QUEUE_FLAG_DEFAULT
;
1070 * This also sets hw/phys segments, boundary and size
1072 blk_queue_make_request(q
, blk_queue_bio
);
1074 q
->sg_reserved_size
= INT_MAX
;
1076 /* Protect q->elevator from elevator_change */
1077 mutex_lock(&q
->sysfs_lock
);
1080 if (elevator_init(q
, NULL
)) {
1081 mutex_unlock(&q
->sysfs_lock
);
1082 goto out_exit_flush_rq
;
1085 mutex_unlock(&q
->sysfs_lock
);
1090 q
->exit_rq_fn(q
, q
->fq
->flush_rq
);
1091 out_free_flush_queue
:
1092 blk_free_flush_queue(q
->fq
);
1096 EXPORT_SYMBOL(blk_init_allocated_queue
);
1098 bool blk_get_queue(struct request_queue
*q
)
1100 if (likely(!blk_queue_dying(q
))) {
1107 EXPORT_SYMBOL(blk_get_queue
);
1109 static inline void blk_free_request(struct request_list
*rl
, struct request
*rq
)
1111 if (rq
->rq_flags
& RQF_ELVPRIV
) {
1112 elv_put_request(rl
->q
, rq
);
1114 put_io_context(rq
->elv
.icq
->ioc
);
1117 mempool_free(rq
, rl
->rq_pool
);
1121 * ioc_batching returns true if the ioc is a valid batching request and
1122 * should be given priority access to a request.
1124 static inline int ioc_batching(struct request_queue
*q
, struct io_context
*ioc
)
1130 * Make sure the process is able to allocate at least 1 request
1131 * even if the batch times out, otherwise we could theoretically
1134 return ioc
->nr_batch_requests
== q
->nr_batching
||
1135 (ioc
->nr_batch_requests
> 0
1136 && time_before(jiffies
, ioc
->last_waited
+ BLK_BATCH_TIME
));
1140 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1141 * will cause the process to be a "batcher" on all queues in the system. This
1142 * is the behaviour we want though - once it gets a wakeup it should be given
1145 static void ioc_set_batching(struct request_queue
*q
, struct io_context
*ioc
)
1147 if (!ioc
|| ioc_batching(q
, ioc
))
1150 ioc
->nr_batch_requests
= q
->nr_batching
;
1151 ioc
->last_waited
= jiffies
;
1154 static void __freed_request(struct request_list
*rl
, int sync
)
1156 struct request_queue
*q
= rl
->q
;
1158 if (rl
->count
[sync
] < queue_congestion_off_threshold(q
))
1159 blk_clear_congested(rl
, sync
);
1161 if (rl
->count
[sync
] + 1 <= q
->nr_requests
) {
1162 if (waitqueue_active(&rl
->wait
[sync
]))
1163 wake_up(&rl
->wait
[sync
]);
1165 blk_clear_rl_full(rl
, sync
);
1170 * A request has just been released. Account for it, update the full and
1171 * congestion status, wake up any waiters. Called under q->queue_lock.
1173 static void freed_request(struct request_list
*rl
, bool sync
,
1174 req_flags_t rq_flags
)
1176 struct request_queue
*q
= rl
->q
;
1180 if (rq_flags
& RQF_ELVPRIV
)
1181 q
->nr_rqs_elvpriv
--;
1183 __freed_request(rl
, sync
);
1185 if (unlikely(rl
->starved
[sync
^ 1]))
1186 __freed_request(rl
, sync
^ 1);
1189 int blk_update_nr_requests(struct request_queue
*q
, unsigned int nr
)
1191 struct request_list
*rl
;
1192 int on_thresh
, off_thresh
;
1194 WARN_ON_ONCE(q
->mq_ops
);
1196 spin_lock_irq(q
->queue_lock
);
1197 q
->nr_requests
= nr
;
1198 blk_queue_congestion_threshold(q
);
1199 on_thresh
= queue_congestion_on_threshold(q
);
1200 off_thresh
= queue_congestion_off_threshold(q
);
1202 blk_queue_for_each_rl(rl
, q
) {
1203 if (rl
->count
[BLK_RW_SYNC
] >= on_thresh
)
1204 blk_set_congested(rl
, BLK_RW_SYNC
);
1205 else if (rl
->count
[BLK_RW_SYNC
] < off_thresh
)
1206 blk_clear_congested(rl
, BLK_RW_SYNC
);
1208 if (rl
->count
[BLK_RW_ASYNC
] >= on_thresh
)
1209 blk_set_congested(rl
, BLK_RW_ASYNC
);
1210 else if (rl
->count
[BLK_RW_ASYNC
] < off_thresh
)
1211 blk_clear_congested(rl
, BLK_RW_ASYNC
);
1213 if (rl
->count
[BLK_RW_SYNC
] >= q
->nr_requests
) {
1214 blk_set_rl_full(rl
, BLK_RW_SYNC
);
1216 blk_clear_rl_full(rl
, BLK_RW_SYNC
);
1217 wake_up(&rl
->wait
[BLK_RW_SYNC
]);
1220 if (rl
->count
[BLK_RW_ASYNC
] >= q
->nr_requests
) {
1221 blk_set_rl_full(rl
, BLK_RW_ASYNC
);
1223 blk_clear_rl_full(rl
, BLK_RW_ASYNC
);
1224 wake_up(&rl
->wait
[BLK_RW_ASYNC
]);
1228 spin_unlock_irq(q
->queue_lock
);
1233 * __get_request - get a free request
1234 * @rl: request list to allocate from
1235 * @op: operation and flags
1236 * @bio: bio to allocate request for (can be %NULL)
1237 * @flags: BLQ_MQ_REQ_* flags
1239 * Get a free request from @q. This function may fail under memory
1240 * pressure or if @q is dead.
1242 * Must be called with @q->queue_lock held and,
1243 * Returns ERR_PTR on failure, with @q->queue_lock held.
1244 * Returns request pointer on success, with @q->queue_lock *not held*.
1246 static struct request
*__get_request(struct request_list
*rl
, unsigned int op
,
1247 struct bio
*bio
, blk_mq_req_flags_t flags
)
1249 struct request_queue
*q
= rl
->q
;
1251 struct elevator_type
*et
= q
->elevator
->type
;
1252 struct io_context
*ioc
= rq_ioc(bio
);
1253 struct io_cq
*icq
= NULL
;
1254 const bool is_sync
= op_is_sync(op
);
1256 gfp_t gfp_mask
= flags
& BLK_MQ_REQ_NOWAIT
? GFP_ATOMIC
:
1257 __GFP_DIRECT_RECLAIM
;
1258 req_flags_t rq_flags
= RQF_ALLOCED
;
1260 lockdep_assert_held(q
->queue_lock
);
1262 if (unlikely(blk_queue_dying(q
)))
1263 return ERR_PTR(-ENODEV
);
1265 may_queue
= elv_may_queue(q
, op
);
1266 if (may_queue
== ELV_MQUEUE_NO
)
1269 if (rl
->count
[is_sync
]+1 >= queue_congestion_on_threshold(q
)) {
1270 if (rl
->count
[is_sync
]+1 >= q
->nr_requests
) {
1272 * The queue will fill after this allocation, so set
1273 * it as full, and mark this process as "batching".
1274 * This process will be allowed to complete a batch of
1275 * requests, others will be blocked.
1277 if (!blk_rl_full(rl
, is_sync
)) {
1278 ioc_set_batching(q
, ioc
);
1279 blk_set_rl_full(rl
, is_sync
);
1281 if (may_queue
!= ELV_MQUEUE_MUST
1282 && !ioc_batching(q
, ioc
)) {
1284 * The queue is full and the allocating
1285 * process is not a "batcher", and not
1286 * exempted by the IO scheduler
1288 return ERR_PTR(-ENOMEM
);
1292 blk_set_congested(rl
, is_sync
);
1296 * Only allow batching queuers to allocate up to 50% over the defined
1297 * limit of requests, otherwise we could have thousands of requests
1298 * allocated with any setting of ->nr_requests
1300 if (rl
->count
[is_sync
] >= (3 * q
->nr_requests
/ 2))
1301 return ERR_PTR(-ENOMEM
);
1303 q
->nr_rqs
[is_sync
]++;
1304 rl
->count
[is_sync
]++;
1305 rl
->starved
[is_sync
] = 0;
1308 * Decide whether the new request will be managed by elevator. If
1309 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1310 * prevent the current elevator from being destroyed until the new
1311 * request is freed. This guarantees icq's won't be destroyed and
1312 * makes creating new ones safe.
1314 * Flush requests do not use the elevator so skip initialization.
1315 * This allows a request to share the flush and elevator data.
1317 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1318 * it will be created after releasing queue_lock.
1320 if (!op_is_flush(op
) && !blk_queue_bypass(q
)) {
1321 rq_flags
|= RQF_ELVPRIV
;
1322 q
->nr_rqs_elvpriv
++;
1323 if (et
->icq_cache
&& ioc
)
1324 icq
= ioc_lookup_icq(ioc
, q
);
1327 if (blk_queue_io_stat(q
))
1328 rq_flags
|= RQF_IO_STAT
;
1329 spin_unlock_irq(q
->queue_lock
);
1331 /* allocate and init request */
1332 rq
= mempool_alloc(rl
->rq_pool
, gfp_mask
);
1337 blk_rq_set_rl(rq
, rl
);
1339 rq
->rq_flags
= rq_flags
;
1340 if (flags
& BLK_MQ_REQ_PREEMPT
)
1341 rq
->rq_flags
|= RQF_PREEMPT
;
1344 if (rq_flags
& RQF_ELVPRIV
) {
1345 if (unlikely(et
->icq_cache
&& !icq
)) {
1347 icq
= ioc_create_icq(ioc
, q
, gfp_mask
);
1353 if (unlikely(elv_set_request(q
, rq
, bio
, gfp_mask
)))
1356 /* @rq->elv.icq holds io_context until @rq is freed */
1358 get_io_context(icq
->ioc
);
1362 * ioc may be NULL here, and ioc_batching will be false. That's
1363 * OK, if the queue is under the request limit then requests need
1364 * not count toward the nr_batch_requests limit. There will always
1365 * be some limit enforced by BLK_BATCH_TIME.
1367 if (ioc_batching(q
, ioc
))
1368 ioc
->nr_batch_requests
--;
1370 trace_block_getrq(q
, bio
, op
);
1375 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1376 * and may fail indefinitely under memory pressure and thus
1377 * shouldn't stall IO. Treat this request as !elvpriv. This will
1378 * disturb iosched and blkcg but weird is bettern than dead.
1380 printk_ratelimited(KERN_WARNING
"%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1381 __func__
, dev_name(q
->backing_dev_info
->dev
));
1383 rq
->rq_flags
&= ~RQF_ELVPRIV
;
1386 spin_lock_irq(q
->queue_lock
);
1387 q
->nr_rqs_elvpriv
--;
1388 spin_unlock_irq(q
->queue_lock
);
1393 * Allocation failed presumably due to memory. Undo anything we
1394 * might have messed up.
1396 * Allocating task should really be put onto the front of the wait
1397 * queue, but this is pretty rare.
1399 spin_lock_irq(q
->queue_lock
);
1400 freed_request(rl
, is_sync
, rq_flags
);
1403 * in the very unlikely event that allocation failed and no
1404 * requests for this direction was pending, mark us starved so that
1405 * freeing of a request in the other direction will notice
1406 * us. another possible fix would be to split the rq mempool into
1410 if (unlikely(rl
->count
[is_sync
] == 0))
1411 rl
->starved
[is_sync
] = 1;
1412 return ERR_PTR(-ENOMEM
);
1416 * get_request - get a free request
1417 * @q: request_queue to allocate request from
1418 * @op: operation and flags
1419 * @bio: bio to allocate request for (can be %NULL)
1420 * @flags: BLK_MQ_REQ_* flags.
1422 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1423 * this function keeps retrying under memory pressure and fails iff @q is dead.
1425 * Must be called with @q->queue_lock held and,
1426 * Returns ERR_PTR on failure, with @q->queue_lock held.
1427 * Returns request pointer on success, with @q->queue_lock *not held*.
1429 static struct request
*get_request(struct request_queue
*q
, unsigned int op
,
1430 struct bio
*bio
, blk_mq_req_flags_t flags
)
1432 const bool is_sync
= op_is_sync(op
);
1434 struct request_list
*rl
;
1437 lockdep_assert_held(q
->queue_lock
);
1438 WARN_ON_ONCE(q
->mq_ops
);
1440 rl
= blk_get_rl(q
, bio
); /* transferred to @rq on success */
1442 rq
= __get_request(rl
, op
, bio
, flags
);
1446 if (op
& REQ_NOWAIT
) {
1448 return ERR_PTR(-EAGAIN
);
1451 if ((flags
& BLK_MQ_REQ_NOWAIT
) || unlikely(blk_queue_dying(q
))) {
1456 /* wait on @rl and retry */
1457 prepare_to_wait_exclusive(&rl
->wait
[is_sync
], &wait
,
1458 TASK_UNINTERRUPTIBLE
);
1460 trace_block_sleeprq(q
, bio
, op
);
1462 spin_unlock_irq(q
->queue_lock
);
1466 * After sleeping, we become a "batching" process and will be able
1467 * to allocate at least one request, and up to a big batch of them
1468 * for a small period time. See ioc_batching, ioc_set_batching
1470 ioc_set_batching(q
, current
->io_context
);
1472 spin_lock_irq(q
->queue_lock
);
1473 finish_wait(&rl
->wait
[is_sync
], &wait
);
1478 /* flags: BLK_MQ_REQ_PREEMPT and/or BLK_MQ_REQ_NOWAIT. */
1479 static struct request
*blk_old_get_request(struct request_queue
*q
,
1480 unsigned int op
, blk_mq_req_flags_t flags
)
1483 gfp_t gfp_mask
= flags
& BLK_MQ_REQ_NOWAIT
? GFP_ATOMIC
:
1484 __GFP_DIRECT_RECLAIM
;
1487 WARN_ON_ONCE(q
->mq_ops
);
1489 /* create ioc upfront */
1490 create_io_context(gfp_mask
, q
->node
);
1492 ret
= blk_queue_enter(q
, flags
);
1494 return ERR_PTR(ret
);
1495 spin_lock_irq(q
->queue_lock
);
1496 rq
= get_request(q
, op
, NULL
, flags
);
1498 spin_unlock_irq(q
->queue_lock
);
1503 /* q->queue_lock is unlocked at this point */
1505 rq
->__sector
= (sector_t
) -1;
1506 rq
->bio
= rq
->biotail
= NULL
;
1511 * blk_get_request_flags - allocate a request
1512 * @q: request queue to allocate a request for
1513 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
1514 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
1516 struct request
*blk_get_request_flags(struct request_queue
*q
, unsigned int op
,
1517 blk_mq_req_flags_t flags
)
1519 struct request
*req
;
1521 WARN_ON_ONCE(op
& REQ_NOWAIT
);
1522 WARN_ON_ONCE(flags
& ~(BLK_MQ_REQ_NOWAIT
| BLK_MQ_REQ_PREEMPT
));
1525 req
= blk_mq_alloc_request(q
, op
, flags
);
1526 if (!IS_ERR(req
) && q
->mq_ops
->initialize_rq_fn
)
1527 q
->mq_ops
->initialize_rq_fn(req
);
1529 req
= blk_old_get_request(q
, op
, flags
);
1530 if (!IS_ERR(req
) && q
->initialize_rq_fn
)
1531 q
->initialize_rq_fn(req
);
1536 EXPORT_SYMBOL(blk_get_request_flags
);
1538 struct request
*blk_get_request(struct request_queue
*q
, unsigned int op
,
1541 return blk_get_request_flags(q
, op
, gfp_mask
& __GFP_DIRECT_RECLAIM
?
1542 0 : BLK_MQ_REQ_NOWAIT
);
1544 EXPORT_SYMBOL(blk_get_request
);
1547 * blk_requeue_request - put a request back on queue
1548 * @q: request queue where request should be inserted
1549 * @rq: request to be inserted
1552 * Drivers often keep queueing requests until the hardware cannot accept
1553 * more, when that condition happens we need to put the request back
1554 * on the queue. Must be called with queue lock held.
1556 void blk_requeue_request(struct request_queue
*q
, struct request
*rq
)
1558 lockdep_assert_held(q
->queue_lock
);
1559 WARN_ON_ONCE(q
->mq_ops
);
1561 blk_delete_timer(rq
);
1562 blk_clear_rq_complete(rq
);
1563 trace_block_rq_requeue(q
, rq
);
1564 wbt_requeue(q
->rq_wb
, &rq
->issue_stat
);
1566 if (rq
->rq_flags
& RQF_QUEUED
)
1567 blk_queue_end_tag(q
, rq
);
1569 BUG_ON(blk_queued_rq(rq
));
1571 elv_requeue_request(q
, rq
);
1573 EXPORT_SYMBOL(blk_requeue_request
);
1575 static void add_acct_request(struct request_queue
*q
, struct request
*rq
,
1578 blk_account_io_start(rq
, true);
1579 __elv_add_request(q
, rq
, where
);
1582 static void part_round_stats_single(struct request_queue
*q
, int cpu
,
1583 struct hd_struct
*part
, unsigned long now
,
1584 unsigned int inflight
)
1587 __part_stat_add(cpu
, part
, time_in_queue
,
1588 inflight
* (now
- part
->stamp
));
1589 __part_stat_add(cpu
, part
, io_ticks
, (now
- part
->stamp
));
1595 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1596 * @q: target block queue
1597 * @cpu: cpu number for stats access
1598 * @part: target partition
1600 * The average IO queue length and utilisation statistics are maintained
1601 * by observing the current state of the queue length and the amount of
1602 * time it has been in this state for.
1604 * Normally, that accounting is done on IO completion, but that can result
1605 * in more than a second's worth of IO being accounted for within any one
1606 * second, leading to >100% utilisation. To deal with that, we call this
1607 * function to do a round-off before returning the results when reading
1608 * /proc/diskstats. This accounts immediately for all queue usage up to
1609 * the current jiffies and restarts the counters again.
1611 void part_round_stats(struct request_queue
*q
, int cpu
, struct hd_struct
*part
)
1613 struct hd_struct
*part2
= NULL
;
1614 unsigned long now
= jiffies
;
1615 unsigned int inflight
[2];
1618 if (part
->stamp
!= now
)
1622 part2
= &part_to_disk(part
)->part0
;
1623 if (part2
->stamp
!= now
)
1630 part_in_flight(q
, part
, inflight
);
1633 part_round_stats_single(q
, cpu
, part2
, now
, inflight
[1]);
1635 part_round_stats_single(q
, cpu
, part
, now
, inflight
[0]);
1637 EXPORT_SYMBOL_GPL(part_round_stats
);
1640 static void blk_pm_put_request(struct request
*rq
)
1642 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && !--rq
->q
->nr_pending
)
1643 pm_runtime_mark_last_busy(rq
->q
->dev
);
1646 static inline void blk_pm_put_request(struct request
*rq
) {}
1649 void __blk_put_request(struct request_queue
*q
, struct request
*req
)
1651 req_flags_t rq_flags
= req
->rq_flags
;
1657 blk_mq_free_request(req
);
1661 lockdep_assert_held(q
->queue_lock
);
1663 blk_pm_put_request(req
);
1665 elv_completed_request(q
, req
);
1667 /* this is a bio leak */
1668 WARN_ON(req
->bio
!= NULL
);
1670 wbt_done(q
->rq_wb
, &req
->issue_stat
);
1673 * Request may not have originated from ll_rw_blk. if not,
1674 * it didn't come out of our reserved rq pools
1676 if (rq_flags
& RQF_ALLOCED
) {
1677 struct request_list
*rl
= blk_rq_rl(req
);
1678 bool sync
= op_is_sync(req
->cmd_flags
);
1680 BUG_ON(!list_empty(&req
->queuelist
));
1681 BUG_ON(ELV_ON_HASH(req
));
1683 blk_free_request(rl
, req
);
1684 freed_request(rl
, sync
, rq_flags
);
1689 EXPORT_SYMBOL_GPL(__blk_put_request
);
1691 void blk_put_request(struct request
*req
)
1693 struct request_queue
*q
= req
->q
;
1696 blk_mq_free_request(req
);
1698 unsigned long flags
;
1700 spin_lock_irqsave(q
->queue_lock
, flags
);
1701 __blk_put_request(q
, req
);
1702 spin_unlock_irqrestore(q
->queue_lock
, flags
);
1705 EXPORT_SYMBOL(blk_put_request
);
1707 bool bio_attempt_back_merge(struct request_queue
*q
, struct request
*req
,
1710 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1712 if (!ll_back_merge_fn(q
, req
, bio
))
1715 trace_block_bio_backmerge(q
, req
, bio
);
1717 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1718 blk_rq_set_mixed_merge(req
);
1720 req
->biotail
->bi_next
= bio
;
1722 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1723 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1725 blk_account_io_start(req
, false);
1729 bool bio_attempt_front_merge(struct request_queue
*q
, struct request
*req
,
1732 const int ff
= bio
->bi_opf
& REQ_FAILFAST_MASK
;
1734 if (!ll_front_merge_fn(q
, req
, bio
))
1737 trace_block_bio_frontmerge(q
, req
, bio
);
1739 if ((req
->cmd_flags
& REQ_FAILFAST_MASK
) != ff
)
1740 blk_rq_set_mixed_merge(req
);
1742 bio
->bi_next
= req
->bio
;
1745 req
->__sector
= bio
->bi_iter
.bi_sector
;
1746 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1747 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1749 blk_account_io_start(req
, false);
1753 bool bio_attempt_discard_merge(struct request_queue
*q
, struct request
*req
,
1756 unsigned short segments
= blk_rq_nr_discard_segments(req
);
1758 if (segments
>= queue_max_discard_segments(q
))
1760 if (blk_rq_sectors(req
) + bio_sectors(bio
) >
1761 blk_rq_get_max_sectors(req
, blk_rq_pos(req
)))
1764 req
->biotail
->bi_next
= bio
;
1766 req
->__data_len
+= bio
->bi_iter
.bi_size
;
1767 req
->ioprio
= ioprio_best(req
->ioprio
, bio_prio(bio
));
1768 req
->nr_phys_segments
= segments
+ 1;
1770 blk_account_io_start(req
, false);
1773 req_set_nomerge(q
, req
);
1778 * blk_attempt_plug_merge - try to merge with %current's plugged list
1779 * @q: request_queue new bio is being queued at
1780 * @bio: new bio being queued
1781 * @request_count: out parameter for number of traversed plugged requests
1782 * @same_queue_rq: pointer to &struct request that gets filled in when
1783 * another request associated with @q is found on the plug list
1784 * (optional, may be %NULL)
1786 * Determine whether @bio being queued on @q can be merged with a request
1787 * on %current's plugged list. Returns %true if merge was successful,
1790 * Plugging coalesces IOs from the same issuer for the same purpose without
1791 * going through @q->queue_lock. As such it's more of an issuing mechanism
1792 * than scheduling, and the request, while may have elvpriv data, is not
1793 * added on the elevator at this point. In addition, we don't have
1794 * reliable access to the elevator outside queue lock. Only check basic
1795 * merging parameters without querying the elevator.
1797 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1799 bool blk_attempt_plug_merge(struct request_queue
*q
, struct bio
*bio
,
1800 unsigned int *request_count
,
1801 struct request
**same_queue_rq
)
1803 struct blk_plug
*plug
;
1805 struct list_head
*plug_list
;
1807 plug
= current
->plug
;
1813 plug_list
= &plug
->mq_list
;
1815 plug_list
= &plug
->list
;
1817 list_for_each_entry_reverse(rq
, plug_list
, queuelist
) {
1818 bool merged
= false;
1823 * Only blk-mq multiple hardware queues case checks the
1824 * rq in the same queue, there should be only one such
1828 *same_queue_rq
= rq
;
1831 if (rq
->q
!= q
|| !blk_rq_merge_ok(rq
, bio
))
1834 switch (blk_try_merge(rq
, bio
)) {
1835 case ELEVATOR_BACK_MERGE
:
1836 merged
= bio_attempt_back_merge(q
, rq
, bio
);
1838 case ELEVATOR_FRONT_MERGE
:
1839 merged
= bio_attempt_front_merge(q
, rq
, bio
);
1841 case ELEVATOR_DISCARD_MERGE
:
1842 merged
= bio_attempt_discard_merge(q
, rq
, bio
);
1855 unsigned int blk_plug_queued_count(struct request_queue
*q
)
1857 struct blk_plug
*plug
;
1859 struct list_head
*plug_list
;
1860 unsigned int ret
= 0;
1862 plug
= current
->plug
;
1867 plug_list
= &plug
->mq_list
;
1869 plug_list
= &plug
->list
;
1871 list_for_each_entry(rq
, plug_list
, queuelist
) {
1879 void blk_init_request_from_bio(struct request
*req
, struct bio
*bio
)
1881 struct io_context
*ioc
= rq_ioc(bio
);
1883 if (bio
->bi_opf
& REQ_RAHEAD
)
1884 req
->cmd_flags
|= REQ_FAILFAST_MASK
;
1886 req
->__sector
= bio
->bi_iter
.bi_sector
;
1887 if (ioprio_valid(bio_prio(bio
)))
1888 req
->ioprio
= bio_prio(bio
);
1890 req
->ioprio
= ioc
->ioprio
;
1892 req
->ioprio
= IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE
, 0);
1893 req
->write_hint
= bio
->bi_write_hint
;
1894 blk_rq_bio_prep(req
->q
, req
, bio
);
1896 EXPORT_SYMBOL_GPL(blk_init_request_from_bio
);
1898 static blk_qc_t
blk_queue_bio(struct request_queue
*q
, struct bio
*bio
)
1900 struct blk_plug
*plug
;
1901 int where
= ELEVATOR_INSERT_SORT
;
1902 struct request
*req
, *free
;
1903 unsigned int request_count
= 0;
1904 unsigned int wb_acct
;
1907 * low level driver can indicate that it wants pages above a
1908 * certain limit bounced to low memory (ie for highmem, or even
1909 * ISA dma in theory)
1911 blk_queue_bounce(q
, &bio
);
1913 blk_queue_split(q
, &bio
);
1915 if (!bio_integrity_prep(bio
))
1916 return BLK_QC_T_NONE
;
1918 if (op_is_flush(bio
->bi_opf
)) {
1919 spin_lock_irq(q
->queue_lock
);
1920 where
= ELEVATOR_INSERT_FLUSH
;
1925 * Check if we can merge with the plugged list before grabbing
1928 if (!blk_queue_nomerges(q
)) {
1929 if (blk_attempt_plug_merge(q
, bio
, &request_count
, NULL
))
1930 return BLK_QC_T_NONE
;
1932 request_count
= blk_plug_queued_count(q
);
1934 spin_lock_irq(q
->queue_lock
);
1936 switch (elv_merge(q
, &req
, bio
)) {
1937 case ELEVATOR_BACK_MERGE
:
1938 if (!bio_attempt_back_merge(q
, req
, bio
))
1940 elv_bio_merged(q
, req
, bio
);
1941 free
= attempt_back_merge(q
, req
);
1943 __blk_put_request(q
, free
);
1945 elv_merged_request(q
, req
, ELEVATOR_BACK_MERGE
);
1947 case ELEVATOR_FRONT_MERGE
:
1948 if (!bio_attempt_front_merge(q
, req
, bio
))
1950 elv_bio_merged(q
, req
, bio
);
1951 free
= attempt_front_merge(q
, req
);
1953 __blk_put_request(q
, free
);
1955 elv_merged_request(q
, req
, ELEVATOR_FRONT_MERGE
);
1962 wb_acct
= wbt_wait(q
->rq_wb
, bio
, q
->queue_lock
);
1965 * Grab a free request. This is might sleep but can not fail.
1966 * Returns with the queue unlocked.
1968 blk_queue_enter_live(q
);
1969 req
= get_request(q
, bio
->bi_opf
, bio
, 0);
1972 __wbt_done(q
->rq_wb
, wb_acct
);
1973 if (PTR_ERR(req
) == -ENOMEM
)
1974 bio
->bi_status
= BLK_STS_RESOURCE
;
1976 bio
->bi_status
= BLK_STS_IOERR
;
1981 wbt_track(&req
->issue_stat
, wb_acct
);
1984 * After dropping the lock and possibly sleeping here, our request
1985 * may now be mergeable after it had proven unmergeable (above).
1986 * We don't worry about that case for efficiency. It won't happen
1987 * often, and the elevators are able to handle it.
1989 blk_init_request_from_bio(req
, bio
);
1991 if (test_bit(QUEUE_FLAG_SAME_COMP
, &q
->queue_flags
))
1992 req
->cpu
= raw_smp_processor_id();
1994 plug
= current
->plug
;
1997 * If this is the first request added after a plug, fire
2000 * @request_count may become stale because of schedule
2001 * out, so check plug list again.
2003 if (!request_count
|| list_empty(&plug
->list
))
2004 trace_block_plug(q
);
2006 struct request
*last
= list_entry_rq(plug
->list
.prev
);
2007 if (request_count
>= BLK_MAX_REQUEST_COUNT
||
2008 blk_rq_bytes(last
) >= BLK_PLUG_FLUSH_SIZE
) {
2009 blk_flush_plug_list(plug
, false);
2010 trace_block_plug(q
);
2013 list_add_tail(&req
->queuelist
, &plug
->list
);
2014 blk_account_io_start(req
, true);
2016 spin_lock_irq(q
->queue_lock
);
2017 add_acct_request(q
, req
, where
);
2020 spin_unlock_irq(q
->queue_lock
);
2023 return BLK_QC_T_NONE
;
2026 static void handle_bad_sector(struct bio
*bio
)
2028 char b
[BDEVNAME_SIZE
];
2030 printk(KERN_INFO
"attempt to access beyond end of device\n");
2031 printk(KERN_INFO
"%s: rw=%d, want=%Lu, limit=%Lu\n",
2032 bio_devname(bio
, b
), bio
->bi_opf
,
2033 (unsigned long long)bio_end_sector(bio
),
2034 (long long)get_capacity(bio
->bi_disk
));
2037 #ifdef CONFIG_FAIL_MAKE_REQUEST
2039 static DECLARE_FAULT_ATTR(fail_make_request
);
2041 static int __init
setup_fail_make_request(char *str
)
2043 return setup_fault_attr(&fail_make_request
, str
);
2045 __setup("fail_make_request=", setup_fail_make_request
);
2047 static bool should_fail_request(struct hd_struct
*part
, unsigned int bytes
)
2049 return part
->make_it_fail
&& should_fail(&fail_make_request
, bytes
);
2052 static int __init
fail_make_request_debugfs(void)
2054 struct dentry
*dir
= fault_create_debugfs_attr("fail_make_request",
2055 NULL
, &fail_make_request
);
2057 return PTR_ERR_OR_ZERO(dir
);
2060 late_initcall(fail_make_request_debugfs
);
2062 #else /* CONFIG_FAIL_MAKE_REQUEST */
2064 static inline bool should_fail_request(struct hd_struct
*part
,
2070 #endif /* CONFIG_FAIL_MAKE_REQUEST */
2073 * Remap block n of partition p to block n+start(p) of the disk.
2075 static inline int blk_partition_remap(struct bio
*bio
)
2077 struct hd_struct
*p
;
2081 * Zone reset does not include bi_size so bio_sectors() is always 0.
2082 * Include a test for the reset op code and perform the remap if needed.
2084 if (!bio
->bi_partno
||
2085 (!bio_sectors(bio
) && bio_op(bio
) != REQ_OP_ZONE_RESET
))
2089 p
= __disk_get_part(bio
->bi_disk
, bio
->bi_partno
);
2090 if (likely(p
&& !should_fail_request(p
, bio
->bi_iter
.bi_size
))) {
2091 bio
->bi_iter
.bi_sector
+= p
->start_sect
;
2093 trace_block_bio_remap(bio
->bi_disk
->queue
, bio
, part_devt(p
),
2094 bio
->bi_iter
.bi_sector
- p
->start_sect
);
2096 printk("%s: fail for partition %d\n", __func__
, bio
->bi_partno
);
2105 * Check whether this bio extends beyond the end of the device.
2107 static inline int bio_check_eod(struct bio
*bio
, unsigned int nr_sectors
)
2114 /* Test device or partition size, when known. */
2115 maxsector
= get_capacity(bio
->bi_disk
);
2117 sector_t sector
= bio
->bi_iter
.bi_sector
;
2119 if (maxsector
< nr_sectors
|| maxsector
- nr_sectors
< sector
) {
2121 * This may well happen - the kernel calls bread()
2122 * without checking the size of the device, e.g., when
2123 * mounting a device.
2125 handle_bad_sector(bio
);
2133 static noinline_for_stack
bool
2134 generic_make_request_checks(struct bio
*bio
)
2136 struct request_queue
*q
;
2137 int nr_sectors
= bio_sectors(bio
);
2138 blk_status_t status
= BLK_STS_IOERR
;
2139 char b
[BDEVNAME_SIZE
];
2143 if (bio_check_eod(bio
, nr_sectors
))
2146 q
= bio
->bi_disk
->queue
;
2149 "generic_make_request: Trying to access "
2150 "nonexistent block-device %s (%Lu)\n",
2151 bio_devname(bio
, b
), (long long)bio
->bi_iter
.bi_sector
);
2156 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2157 * if queue is not a request based queue.
2160 if ((bio
->bi_opf
& REQ_NOWAIT
) && !queue_is_rq_based(q
))
2163 if (should_fail_request(&bio
->bi_disk
->part0
, bio
->bi_iter
.bi_size
))
2166 if (blk_partition_remap(bio
))
2169 if (bio_check_eod(bio
, nr_sectors
))
2173 * Filter flush bio's early so that make_request based
2174 * drivers without flush support don't have to worry
2177 if (op_is_flush(bio
->bi_opf
) &&
2178 !test_bit(QUEUE_FLAG_WC
, &q
->queue_flags
)) {
2179 bio
->bi_opf
&= ~(REQ_PREFLUSH
| REQ_FUA
);
2181 status
= BLK_STS_OK
;
2186 switch (bio_op(bio
)) {
2187 case REQ_OP_DISCARD
:
2188 if (!blk_queue_discard(q
))
2191 case REQ_OP_SECURE_ERASE
:
2192 if (!blk_queue_secure_erase(q
))
2195 case REQ_OP_WRITE_SAME
:
2196 if (!q
->limits
.max_write_same_sectors
)
2199 case REQ_OP_ZONE_REPORT
:
2200 case REQ_OP_ZONE_RESET
:
2201 if (!blk_queue_is_zoned(q
))
2204 case REQ_OP_WRITE_ZEROES
:
2205 if (!q
->limits
.max_write_zeroes_sectors
)
2213 * Various block parts want %current->io_context and lazy ioc
2214 * allocation ends up trading a lot of pain for a small amount of
2215 * memory. Just allocate it upfront. This may fail and block
2216 * layer knows how to live with it.
2218 create_io_context(GFP_ATOMIC
, q
->node
);
2220 if (!blkcg_bio_issue_check(q
, bio
))
2223 if (!bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
2224 trace_block_bio_queue(q
, bio
);
2225 /* Now that enqueuing has been traced, we need to trace
2226 * completion as well.
2228 bio_set_flag(bio
, BIO_TRACE_COMPLETION
);
2233 status
= BLK_STS_NOTSUPP
;
2235 bio
->bi_status
= status
;
2241 * generic_make_request - hand a buffer to its device driver for I/O
2242 * @bio: The bio describing the location in memory and on the device.
2244 * generic_make_request() is used to make I/O requests of block
2245 * devices. It is passed a &struct bio, which describes the I/O that needs
2248 * generic_make_request() does not return any status. The
2249 * success/failure status of the request, along with notification of
2250 * completion, is delivered asynchronously through the bio->bi_end_io
2251 * function described (one day) else where.
2253 * The caller of generic_make_request must make sure that bi_io_vec
2254 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2255 * set to describe the device address, and the
2256 * bi_end_io and optionally bi_private are set to describe how
2257 * completion notification should be signaled.
2259 * generic_make_request and the drivers it calls may use bi_next if this
2260 * bio happens to be merged with someone else, and may resubmit the bio to
2261 * a lower device by calling into generic_make_request recursively, which
2262 * means the bio should NOT be touched after the call to ->make_request_fn.
2264 blk_qc_t
generic_make_request(struct bio
*bio
)
2267 * bio_list_on_stack[0] contains bios submitted by the current
2269 * bio_list_on_stack[1] contains bios that were submitted before
2270 * the current make_request_fn, but that haven't been processed
2273 struct bio_list bio_list_on_stack
[2];
2274 blk_qc_t ret
= BLK_QC_T_NONE
;
2276 if (!generic_make_request_checks(bio
))
2280 * We only want one ->make_request_fn to be active at a time, else
2281 * stack usage with stacked devices could be a problem. So use
2282 * current->bio_list to keep a list of requests submited by a
2283 * make_request_fn function. current->bio_list is also used as a
2284 * flag to say if generic_make_request is currently active in this
2285 * task or not. If it is NULL, then no make_request is active. If
2286 * it is non-NULL, then a make_request is active, and new requests
2287 * should be added at the tail
2289 if (current
->bio_list
) {
2290 bio_list_add(¤t
->bio_list
[0], bio
);
2294 /* following loop may be a bit non-obvious, and so deserves some
2296 * Before entering the loop, bio->bi_next is NULL (as all callers
2297 * ensure that) so we have a list with a single bio.
2298 * We pretend that we have just taken it off a longer list, so
2299 * we assign bio_list to a pointer to the bio_list_on_stack,
2300 * thus initialising the bio_list of new bios to be
2301 * added. ->make_request() may indeed add some more bios
2302 * through a recursive call to generic_make_request. If it
2303 * did, we find a non-NULL value in bio_list and re-enter the loop
2304 * from the top. In this case we really did just take the bio
2305 * of the top of the list (no pretending) and so remove it from
2306 * bio_list, and call into ->make_request() again.
2308 BUG_ON(bio
->bi_next
);
2309 bio_list_init(&bio_list_on_stack
[0]);
2310 current
->bio_list
= bio_list_on_stack
;
2312 struct request_queue
*q
= bio
->bi_disk
->queue
;
2313 blk_mq_req_flags_t flags
= bio
->bi_opf
& REQ_NOWAIT
?
2314 BLK_MQ_REQ_NOWAIT
: 0;
2316 if (likely(blk_queue_enter(q
, flags
) == 0)) {
2317 struct bio_list lower
, same
;
2319 /* Create a fresh bio_list for all subordinate requests */
2320 bio_list_on_stack
[1] = bio_list_on_stack
[0];
2321 bio_list_init(&bio_list_on_stack
[0]);
2322 ret
= q
->make_request_fn(q
, bio
);
2326 /* sort new bios into those for a lower level
2327 * and those for the same level
2329 bio_list_init(&lower
);
2330 bio_list_init(&same
);
2331 while ((bio
= bio_list_pop(&bio_list_on_stack
[0])) != NULL
)
2332 if (q
== bio
->bi_disk
->queue
)
2333 bio_list_add(&same
, bio
);
2335 bio_list_add(&lower
, bio
);
2336 /* now assemble so we handle the lowest level first */
2337 bio_list_merge(&bio_list_on_stack
[0], &lower
);
2338 bio_list_merge(&bio_list_on_stack
[0], &same
);
2339 bio_list_merge(&bio_list_on_stack
[0], &bio_list_on_stack
[1]);
2341 if (unlikely(!blk_queue_dying(q
) &&
2342 (bio
->bi_opf
& REQ_NOWAIT
)))
2343 bio_wouldblock_error(bio
);
2347 bio
= bio_list_pop(&bio_list_on_stack
[0]);
2349 current
->bio_list
= NULL
; /* deactivate */
2354 EXPORT_SYMBOL(generic_make_request
);
2357 * direct_make_request - hand a buffer directly to its device driver for I/O
2358 * @bio: The bio describing the location in memory and on the device.
2360 * This function behaves like generic_make_request(), but does not protect
2361 * against recursion. Must only be used if the called driver is known
2362 * to not call generic_make_request (or direct_make_request) again from
2363 * its make_request function. (Calling direct_make_request again from
2364 * a workqueue is perfectly fine as that doesn't recurse).
2366 blk_qc_t
direct_make_request(struct bio
*bio
)
2368 struct request_queue
*q
= bio
->bi_disk
->queue
;
2369 bool nowait
= bio
->bi_opf
& REQ_NOWAIT
;
2372 if (!generic_make_request_checks(bio
))
2373 return BLK_QC_T_NONE
;
2375 if (unlikely(blk_queue_enter(q
, nowait
? BLK_MQ_REQ_NOWAIT
: 0))) {
2376 if (nowait
&& !blk_queue_dying(q
))
2377 bio
->bi_status
= BLK_STS_AGAIN
;
2379 bio
->bi_status
= BLK_STS_IOERR
;
2381 return BLK_QC_T_NONE
;
2384 ret
= q
->make_request_fn(q
, bio
);
2388 EXPORT_SYMBOL_GPL(direct_make_request
);
2391 * submit_bio - submit a bio to the block device layer for I/O
2392 * @bio: The &struct bio which describes the I/O
2394 * submit_bio() is very similar in purpose to generic_make_request(), and
2395 * uses that function to do most of the work. Both are fairly rough
2396 * interfaces; @bio must be presetup and ready for I/O.
2399 blk_qc_t
submit_bio(struct bio
*bio
)
2402 * If it's a regular read/write or a barrier with data attached,
2403 * go through the normal accounting stuff before submission.
2405 if (bio_has_data(bio
)) {
2408 if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
2409 count
= queue_logical_block_size(bio
->bi_disk
->queue
) >> 9;
2411 count
= bio_sectors(bio
);
2413 if (op_is_write(bio_op(bio
))) {
2414 count_vm_events(PGPGOUT
, count
);
2416 task_io_account_read(bio
->bi_iter
.bi_size
);
2417 count_vm_events(PGPGIN
, count
);
2420 if (unlikely(block_dump
)) {
2421 char b
[BDEVNAME_SIZE
];
2422 printk(KERN_DEBUG
"%s(%d): %s block %Lu on %s (%u sectors)\n",
2423 current
->comm
, task_pid_nr(current
),
2424 op_is_write(bio_op(bio
)) ? "WRITE" : "READ",
2425 (unsigned long long)bio
->bi_iter
.bi_sector
,
2426 bio_devname(bio
, b
), count
);
2430 return generic_make_request(bio
);
2432 EXPORT_SYMBOL(submit_bio
);
2434 bool blk_poll(struct request_queue
*q
, blk_qc_t cookie
)
2436 if (!q
->poll_fn
|| !blk_qc_t_valid(cookie
))
2440 blk_flush_plug_list(current
->plug
, false);
2441 return q
->poll_fn(q
, cookie
);
2443 EXPORT_SYMBOL_GPL(blk_poll
);
2446 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2447 * for new the queue limits
2449 * @rq: the request being checked
2452 * @rq may have been made based on weaker limitations of upper-level queues
2453 * in request stacking drivers, and it may violate the limitation of @q.
2454 * Since the block layer and the underlying device driver trust @rq
2455 * after it is inserted to @q, it should be checked against @q before
2456 * the insertion using this generic function.
2458 * Request stacking drivers like request-based dm may change the queue
2459 * limits when retrying requests on other queues. Those requests need
2460 * to be checked against the new queue limits again during dispatch.
2462 static int blk_cloned_rq_check_limits(struct request_queue
*q
,
2465 if (blk_rq_sectors(rq
) > blk_queue_get_max_sectors(q
, req_op(rq
))) {
2466 printk(KERN_ERR
"%s: over max size limit.\n", __func__
);
2471 * queue's settings related to segment counting like q->bounce_pfn
2472 * may differ from that of other stacking queues.
2473 * Recalculate it to check the request correctly on this queue's
2476 blk_recalc_rq_segments(rq
);
2477 if (rq
->nr_phys_segments
> queue_max_segments(q
)) {
2478 printk(KERN_ERR
"%s: over max segments limit.\n", __func__
);
2486 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2487 * @q: the queue to submit the request
2488 * @rq: the request being queued
2490 blk_status_t
blk_insert_cloned_request(struct request_queue
*q
, struct request
*rq
)
2492 unsigned long flags
;
2493 int where
= ELEVATOR_INSERT_BACK
;
2495 if (blk_cloned_rq_check_limits(q
, rq
))
2496 return BLK_STS_IOERR
;
2499 should_fail_request(&rq
->rq_disk
->part0
, blk_rq_bytes(rq
)))
2500 return BLK_STS_IOERR
;
2503 if (blk_queue_io_stat(q
))
2504 blk_account_io_start(rq
, true);
2506 * Since we have a scheduler attached on the top device,
2507 * bypass a potential scheduler on the bottom device for
2510 blk_mq_request_bypass_insert(rq
, true);
2514 spin_lock_irqsave(q
->queue_lock
, flags
);
2515 if (unlikely(blk_queue_dying(q
))) {
2516 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2517 return BLK_STS_IOERR
;
2521 * Submitting request must be dequeued before calling this function
2522 * because it will be linked to another request_queue
2524 BUG_ON(blk_queued_rq(rq
));
2526 if (op_is_flush(rq
->cmd_flags
))
2527 where
= ELEVATOR_INSERT_FLUSH
;
2529 add_acct_request(q
, rq
, where
);
2530 if (where
== ELEVATOR_INSERT_FLUSH
)
2532 spin_unlock_irqrestore(q
->queue_lock
, flags
);
2536 EXPORT_SYMBOL_GPL(blk_insert_cloned_request
);
2539 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2540 * @rq: request to examine
2543 * A request could be merge of IOs which require different failure
2544 * handling. This function determines the number of bytes which
2545 * can be failed from the beginning of the request without
2546 * crossing into area which need to be retried further.
2549 * The number of bytes to fail.
2551 unsigned int blk_rq_err_bytes(const struct request
*rq
)
2553 unsigned int ff
= rq
->cmd_flags
& REQ_FAILFAST_MASK
;
2554 unsigned int bytes
= 0;
2557 if (!(rq
->rq_flags
& RQF_MIXED_MERGE
))
2558 return blk_rq_bytes(rq
);
2561 * Currently the only 'mixing' which can happen is between
2562 * different fastfail types. We can safely fail portions
2563 * which have all the failfast bits that the first one has -
2564 * the ones which are at least as eager to fail as the first
2567 for (bio
= rq
->bio
; bio
; bio
= bio
->bi_next
) {
2568 if ((bio
->bi_opf
& ff
) != ff
)
2570 bytes
+= bio
->bi_iter
.bi_size
;
2573 /* this could lead to infinite loop */
2574 BUG_ON(blk_rq_bytes(rq
) && !bytes
);
2577 EXPORT_SYMBOL_GPL(blk_rq_err_bytes
);
2579 void blk_account_io_completion(struct request
*req
, unsigned int bytes
)
2581 if (blk_do_io_stat(req
)) {
2582 const int rw
= rq_data_dir(req
);
2583 struct hd_struct
*part
;
2586 cpu
= part_stat_lock();
2588 part_stat_add(cpu
, part
, sectors
[rw
], bytes
>> 9);
2593 void blk_account_io_done(struct request
*req
)
2596 * Account IO completion. flush_rq isn't accounted as a
2597 * normal IO on queueing nor completion. Accounting the
2598 * containing request is enough.
2600 if (blk_do_io_stat(req
) && !(req
->rq_flags
& RQF_FLUSH_SEQ
)) {
2601 unsigned long duration
= jiffies
- req
->start_time
;
2602 const int rw
= rq_data_dir(req
);
2603 struct hd_struct
*part
;
2606 cpu
= part_stat_lock();
2609 part_stat_inc(cpu
, part
, ios
[rw
]);
2610 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
2611 part_round_stats(req
->q
, cpu
, part
);
2612 part_dec_in_flight(req
->q
, part
, rw
);
2614 hd_struct_put(part
);
2621 * Don't process normal requests when queue is suspended
2622 * or in the process of suspending/resuming
2624 static bool blk_pm_allow_request(struct request
*rq
)
2626 switch (rq
->q
->rpm_status
) {
2628 case RPM_SUSPENDING
:
2629 return rq
->rq_flags
& RQF_PM
;
2637 static bool blk_pm_allow_request(struct request
*rq
)
2643 void blk_account_io_start(struct request
*rq
, bool new_io
)
2645 struct hd_struct
*part
;
2646 int rw
= rq_data_dir(rq
);
2649 if (!blk_do_io_stat(rq
))
2652 cpu
= part_stat_lock();
2656 part_stat_inc(cpu
, part
, merges
[rw
]);
2658 part
= disk_map_sector_rcu(rq
->rq_disk
, blk_rq_pos(rq
));
2659 if (!hd_struct_try_get(part
)) {
2661 * The partition is already being removed,
2662 * the request will be accounted on the disk only
2664 * We take a reference on disk->part0 although that
2665 * partition will never be deleted, so we can treat
2666 * it as any other partition.
2668 part
= &rq
->rq_disk
->part0
;
2669 hd_struct_get(part
);
2671 part_round_stats(rq
->q
, cpu
, part
);
2672 part_inc_in_flight(rq
->q
, part
, rw
);
2679 static struct request
*elv_next_request(struct request_queue
*q
)
2682 struct blk_flush_queue
*fq
= blk_get_flush_queue(q
, NULL
);
2684 WARN_ON_ONCE(q
->mq_ops
);
2687 list_for_each_entry(rq
, &q
->queue_head
, queuelist
) {
2688 if (blk_pm_allow_request(rq
))
2691 if (rq
->rq_flags
& RQF_SOFTBARRIER
)
2696 * Flush request is running and flush request isn't queueable
2697 * in the drive, we can hold the queue till flush request is
2698 * finished. Even we don't do this, driver can't dispatch next
2699 * requests and will requeue them. And this can improve
2700 * throughput too. For example, we have request flush1, write1,
2701 * flush 2. flush1 is dispatched, then queue is hold, write1
2702 * isn't inserted to queue. After flush1 is finished, flush2
2703 * will be dispatched. Since disk cache is already clean,
2704 * flush2 will be finished very soon, so looks like flush2 is
2706 * Since the queue is hold, a flag is set to indicate the queue
2707 * should be restarted later. Please see flush_end_io() for
2710 if (fq
->flush_pending_idx
!= fq
->flush_running_idx
&&
2711 !queue_flush_queueable(q
)) {
2712 fq
->flush_queue_delayed
= 1;
2715 if (unlikely(blk_queue_bypass(q
)) ||
2716 !q
->elevator
->type
->ops
.sq
.elevator_dispatch_fn(q
, 0))
2722 * blk_peek_request - peek at the top of a request queue
2723 * @q: request queue to peek at
2726 * Return the request at the top of @q. The returned request
2727 * should be started using blk_start_request() before LLD starts
2731 * Pointer to the request at the top of @q if available. Null
2734 struct request
*blk_peek_request(struct request_queue
*q
)
2739 lockdep_assert_held(q
->queue_lock
);
2740 WARN_ON_ONCE(q
->mq_ops
);
2742 while ((rq
= elv_next_request(q
)) != NULL
) {
2743 if (!(rq
->rq_flags
& RQF_STARTED
)) {
2745 * This is the first time the device driver
2746 * sees this request (possibly after
2747 * requeueing). Notify IO scheduler.
2749 if (rq
->rq_flags
& RQF_SORTED
)
2750 elv_activate_rq(q
, rq
);
2753 * just mark as started even if we don't start
2754 * it, a request that has been delayed should
2755 * not be passed by new incoming requests
2757 rq
->rq_flags
|= RQF_STARTED
;
2758 trace_block_rq_issue(q
, rq
);
2761 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
2762 q
->end_sector
= rq_end_sector(rq
);
2763 q
->boundary_rq
= NULL
;
2766 if (rq
->rq_flags
& RQF_DONTPREP
)
2769 if (q
->dma_drain_size
&& blk_rq_bytes(rq
)) {
2771 * make sure space for the drain appears we
2772 * know we can do this because max_hw_segments
2773 * has been adjusted to be one fewer than the
2776 rq
->nr_phys_segments
++;
2782 ret
= q
->prep_rq_fn(q
, rq
);
2783 if (ret
== BLKPREP_OK
) {
2785 } else if (ret
== BLKPREP_DEFER
) {
2787 * the request may have been (partially) prepped.
2788 * we need to keep this request in the front to
2789 * avoid resource deadlock. RQF_STARTED will
2790 * prevent other fs requests from passing this one.
2792 if (q
->dma_drain_size
&& blk_rq_bytes(rq
) &&
2793 !(rq
->rq_flags
& RQF_DONTPREP
)) {
2795 * remove the space for the drain we added
2796 * so that we don't add it again
2798 --rq
->nr_phys_segments
;
2803 } else if (ret
== BLKPREP_KILL
|| ret
== BLKPREP_INVALID
) {
2804 rq
->rq_flags
|= RQF_QUIET
;
2806 * Mark this request as started so we don't trigger
2807 * any debug logic in the end I/O path.
2809 blk_start_request(rq
);
2810 __blk_end_request_all(rq
, ret
== BLKPREP_INVALID
?
2811 BLK_STS_TARGET
: BLK_STS_IOERR
);
2813 printk(KERN_ERR
"%s: bad return=%d\n", __func__
, ret
);
2820 EXPORT_SYMBOL(blk_peek_request
);
2822 static void blk_dequeue_request(struct request
*rq
)
2824 struct request_queue
*q
= rq
->q
;
2826 BUG_ON(list_empty(&rq
->queuelist
));
2827 BUG_ON(ELV_ON_HASH(rq
));
2829 list_del_init(&rq
->queuelist
);
2832 * the time frame between a request being removed from the lists
2833 * and to it is freed is accounted as io that is in progress at
2836 if (blk_account_rq(rq
)) {
2837 q
->in_flight
[rq_is_sync(rq
)]++;
2838 set_io_start_time_ns(rq
);
2843 * blk_start_request - start request processing on the driver
2844 * @req: request to dequeue
2847 * Dequeue @req and start timeout timer on it. This hands off the
2848 * request to the driver.
2850 void blk_start_request(struct request
*req
)
2852 lockdep_assert_held(req
->q
->queue_lock
);
2853 WARN_ON_ONCE(req
->q
->mq_ops
);
2855 blk_dequeue_request(req
);
2857 if (test_bit(QUEUE_FLAG_STATS
, &req
->q
->queue_flags
)) {
2858 blk_stat_set_issue(&req
->issue_stat
, blk_rq_sectors(req
));
2859 req
->rq_flags
|= RQF_STATS
;
2860 wbt_issue(req
->q
->rq_wb
, &req
->issue_stat
);
2863 BUG_ON(test_bit(REQ_ATOM_COMPLETE
, &req
->atomic_flags
));
2866 EXPORT_SYMBOL(blk_start_request
);
2869 * blk_fetch_request - fetch a request from a request queue
2870 * @q: request queue to fetch a request from
2873 * Return the request at the top of @q. The request is started on
2874 * return and LLD can start processing it immediately.
2877 * Pointer to the request at the top of @q if available. Null
2880 struct request
*blk_fetch_request(struct request_queue
*q
)
2884 lockdep_assert_held(q
->queue_lock
);
2885 WARN_ON_ONCE(q
->mq_ops
);
2887 rq
= blk_peek_request(q
);
2889 blk_start_request(rq
);
2892 EXPORT_SYMBOL(blk_fetch_request
);
2895 * Steal bios from a request and add them to a bio list.
2896 * The request must not have been partially completed before.
2898 void blk_steal_bios(struct bio_list
*list
, struct request
*rq
)
2902 list
->tail
->bi_next
= rq
->bio
;
2904 list
->head
= rq
->bio
;
2905 list
->tail
= rq
->biotail
;
2913 EXPORT_SYMBOL_GPL(blk_steal_bios
);
2916 * blk_update_request - Special helper function for request stacking drivers
2917 * @req: the request being processed
2918 * @error: block status code
2919 * @nr_bytes: number of bytes to complete @req
2922 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2923 * the request structure even if @req doesn't have leftover.
2924 * If @req has leftover, sets it up for the next range of segments.
2926 * This special helper function is only for request stacking drivers
2927 * (e.g. request-based dm) so that they can handle partial completion.
2928 * Actual device drivers should use blk_end_request instead.
2930 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2931 * %false return from this function.
2934 * %false - this request doesn't have any more data
2935 * %true - this request has more data
2937 bool blk_update_request(struct request
*req
, blk_status_t error
,
2938 unsigned int nr_bytes
)
2942 trace_block_rq_complete(req
, blk_status_to_errno(error
), nr_bytes
);
2947 if (unlikely(error
&& !blk_rq_is_passthrough(req
) &&
2948 !(req
->rq_flags
& RQF_QUIET
)))
2949 print_req_error(req
, error
);
2951 blk_account_io_completion(req
, nr_bytes
);
2955 struct bio
*bio
= req
->bio
;
2956 unsigned bio_bytes
= min(bio
->bi_iter
.bi_size
, nr_bytes
);
2958 if (bio_bytes
== bio
->bi_iter
.bi_size
)
2959 req
->bio
= bio
->bi_next
;
2961 /* Completion has already been traced */
2962 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
2963 req_bio_endio(req
, bio
, bio_bytes
, error
);
2965 total_bytes
+= bio_bytes
;
2966 nr_bytes
-= bio_bytes
;
2977 * Reset counters so that the request stacking driver
2978 * can find how many bytes remain in the request
2981 req
->__data_len
= 0;
2985 req
->__data_len
-= total_bytes
;
2987 /* update sector only for requests with clear definition of sector */
2988 if (!blk_rq_is_passthrough(req
))
2989 req
->__sector
+= total_bytes
>> 9;
2991 /* mixed attributes always follow the first bio */
2992 if (req
->rq_flags
& RQF_MIXED_MERGE
) {
2993 req
->cmd_flags
&= ~REQ_FAILFAST_MASK
;
2994 req
->cmd_flags
|= req
->bio
->bi_opf
& REQ_FAILFAST_MASK
;
2997 if (!(req
->rq_flags
& RQF_SPECIAL_PAYLOAD
)) {
2999 * If total number of sectors is less than the first segment
3000 * size, something has gone terribly wrong.
3002 if (blk_rq_bytes(req
) < blk_rq_cur_bytes(req
)) {
3003 blk_dump_rq_flags(req
, "request botched");
3004 req
->__data_len
= blk_rq_cur_bytes(req
);
3007 /* recalculate the number of segments */
3008 blk_recalc_rq_segments(req
);
3013 EXPORT_SYMBOL_GPL(blk_update_request
);
3015 static bool blk_update_bidi_request(struct request
*rq
, blk_status_t error
,
3016 unsigned int nr_bytes
,
3017 unsigned int bidi_bytes
)
3019 if (blk_update_request(rq
, error
, nr_bytes
))
3022 /* Bidi request must be completed as a whole */
3023 if (unlikely(blk_bidi_rq(rq
)) &&
3024 blk_update_request(rq
->next_rq
, error
, bidi_bytes
))
3027 if (blk_queue_add_random(rq
->q
))
3028 add_disk_randomness(rq
->rq_disk
);
3034 * blk_unprep_request - unprepare a request
3037 * This function makes a request ready for complete resubmission (or
3038 * completion). It happens only after all error handling is complete,
3039 * so represents the appropriate moment to deallocate any resources
3040 * that were allocated to the request in the prep_rq_fn. The queue
3041 * lock is held when calling this.
3043 void blk_unprep_request(struct request
*req
)
3045 struct request_queue
*q
= req
->q
;
3047 req
->rq_flags
&= ~RQF_DONTPREP
;
3048 if (q
->unprep_rq_fn
)
3049 q
->unprep_rq_fn(q
, req
);
3051 EXPORT_SYMBOL_GPL(blk_unprep_request
);
3053 void blk_finish_request(struct request
*req
, blk_status_t error
)
3055 struct request_queue
*q
= req
->q
;
3057 lockdep_assert_held(req
->q
->queue_lock
);
3058 WARN_ON_ONCE(q
->mq_ops
);
3060 if (req
->rq_flags
& RQF_STATS
)
3063 if (req
->rq_flags
& RQF_QUEUED
)
3064 blk_queue_end_tag(q
, req
);
3066 BUG_ON(blk_queued_rq(req
));
3068 if (unlikely(laptop_mode
) && !blk_rq_is_passthrough(req
))
3069 laptop_io_completion(req
->q
->backing_dev_info
);
3071 blk_delete_timer(req
);
3073 if (req
->rq_flags
& RQF_DONTPREP
)
3074 blk_unprep_request(req
);
3076 blk_account_io_done(req
);
3079 wbt_done(req
->q
->rq_wb
, &req
->issue_stat
);
3080 req
->end_io(req
, error
);
3082 if (blk_bidi_rq(req
))
3083 __blk_put_request(req
->next_rq
->q
, req
->next_rq
);
3085 __blk_put_request(q
, req
);
3088 EXPORT_SYMBOL(blk_finish_request
);
3091 * blk_end_bidi_request - Complete a bidi request
3092 * @rq: the request to complete
3093 * @error: block status code
3094 * @nr_bytes: number of bytes to complete @rq
3095 * @bidi_bytes: number of bytes to complete @rq->next_rq
3098 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
3099 * Drivers that supports bidi can safely call this member for any
3100 * type of request, bidi or uni. In the later case @bidi_bytes is
3104 * %false - we are done with this request
3105 * %true - still buffers pending for this request
3107 static bool blk_end_bidi_request(struct request
*rq
, blk_status_t error
,
3108 unsigned int nr_bytes
, unsigned int bidi_bytes
)
3110 struct request_queue
*q
= rq
->q
;
3111 unsigned long flags
;
3113 WARN_ON_ONCE(q
->mq_ops
);
3115 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
3118 spin_lock_irqsave(q
->queue_lock
, flags
);
3119 blk_finish_request(rq
, error
);
3120 spin_unlock_irqrestore(q
->queue_lock
, flags
);
3126 * __blk_end_bidi_request - Complete a bidi request with queue lock held
3127 * @rq: the request to complete
3128 * @error: block status code
3129 * @nr_bytes: number of bytes to complete @rq
3130 * @bidi_bytes: number of bytes to complete @rq->next_rq
3133 * Identical to blk_end_bidi_request() except that queue lock is
3134 * assumed to be locked on entry and remains so on return.
3137 * %false - we are done with this request
3138 * %true - still buffers pending for this request
3140 static bool __blk_end_bidi_request(struct request
*rq
, blk_status_t error
,
3141 unsigned int nr_bytes
, unsigned int bidi_bytes
)
3143 lockdep_assert_held(rq
->q
->queue_lock
);
3144 WARN_ON_ONCE(rq
->q
->mq_ops
);
3146 if (blk_update_bidi_request(rq
, error
, nr_bytes
, bidi_bytes
))
3149 blk_finish_request(rq
, error
);
3155 * blk_end_request - Helper function for drivers to complete the request.
3156 * @rq: the request being processed
3157 * @error: block status code
3158 * @nr_bytes: number of bytes to complete
3161 * Ends I/O on a number of bytes attached to @rq.
3162 * If @rq has leftover, sets it up for the next range of segments.
3165 * %false - we are done with this request
3166 * %true - still buffers pending for this request
3168 bool blk_end_request(struct request
*rq
, blk_status_t error
,
3169 unsigned int nr_bytes
)
3171 WARN_ON_ONCE(rq
->q
->mq_ops
);
3172 return blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
3174 EXPORT_SYMBOL(blk_end_request
);
3177 * blk_end_request_all - Helper function for drives to finish the request.
3178 * @rq: the request to finish
3179 * @error: block status code
3182 * Completely finish @rq.
3184 void blk_end_request_all(struct request
*rq
, blk_status_t error
)
3187 unsigned int bidi_bytes
= 0;
3189 if (unlikely(blk_bidi_rq(rq
)))
3190 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
3192 pending
= blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
3195 EXPORT_SYMBOL(blk_end_request_all
);
3198 * __blk_end_request - Helper function for drivers to complete the request.
3199 * @rq: the request being processed
3200 * @error: block status code
3201 * @nr_bytes: number of bytes to complete
3204 * Must be called with queue lock held unlike blk_end_request().
3207 * %false - we are done with this request
3208 * %true - still buffers pending for this request
3210 bool __blk_end_request(struct request
*rq
, blk_status_t error
,
3211 unsigned int nr_bytes
)
3213 lockdep_assert_held(rq
->q
->queue_lock
);
3214 WARN_ON_ONCE(rq
->q
->mq_ops
);
3216 return __blk_end_bidi_request(rq
, error
, nr_bytes
, 0);
3218 EXPORT_SYMBOL(__blk_end_request
);
3221 * __blk_end_request_all - Helper function for drives to finish the request.
3222 * @rq: the request to finish
3223 * @error: block status code
3226 * Completely finish @rq. Must be called with queue lock held.
3228 void __blk_end_request_all(struct request
*rq
, blk_status_t error
)
3231 unsigned int bidi_bytes
= 0;
3233 lockdep_assert_held(rq
->q
->queue_lock
);
3234 WARN_ON_ONCE(rq
->q
->mq_ops
);
3236 if (unlikely(blk_bidi_rq(rq
)))
3237 bidi_bytes
= blk_rq_bytes(rq
->next_rq
);
3239 pending
= __blk_end_bidi_request(rq
, error
, blk_rq_bytes(rq
), bidi_bytes
);
3242 EXPORT_SYMBOL(__blk_end_request_all
);
3245 * __blk_end_request_cur - Helper function to finish the current request chunk.
3246 * @rq: the request to finish the current chunk for
3247 * @error: block status code
3250 * Complete the current consecutively mapped chunk from @rq. Must
3251 * be called with queue lock held.
3254 * %false - we are done with this request
3255 * %true - still buffers pending for this request
3257 bool __blk_end_request_cur(struct request
*rq
, blk_status_t error
)
3259 return __blk_end_request(rq
, error
, blk_rq_cur_bytes(rq
));
3261 EXPORT_SYMBOL(__blk_end_request_cur
);
3263 void blk_rq_bio_prep(struct request_queue
*q
, struct request
*rq
,
3266 if (bio_has_data(bio
))
3267 rq
->nr_phys_segments
= bio_phys_segments(q
, bio
);
3268 else if (bio_op(bio
) == REQ_OP_DISCARD
)
3269 rq
->nr_phys_segments
= 1;
3271 rq
->__data_len
= bio
->bi_iter
.bi_size
;
3272 rq
->bio
= rq
->biotail
= bio
;
3275 rq
->rq_disk
= bio
->bi_disk
;
3278 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3280 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3281 * @rq: the request to be flushed
3284 * Flush all pages in @rq.
3286 void rq_flush_dcache_pages(struct request
*rq
)
3288 struct req_iterator iter
;
3289 struct bio_vec bvec
;
3291 rq_for_each_segment(bvec
, rq
, iter
)
3292 flush_dcache_page(bvec
.bv_page
);
3294 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages
);
3298 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3299 * @q : the queue of the device being checked
3302 * Check if underlying low-level drivers of a device are busy.
3303 * If the drivers want to export their busy state, they must set own
3304 * exporting function using blk_queue_lld_busy() first.
3306 * Basically, this function is used only by request stacking drivers
3307 * to stop dispatching requests to underlying devices when underlying
3308 * devices are busy. This behavior helps more I/O merging on the queue
3309 * of the request stacking driver and prevents I/O throughput regression
3310 * on burst I/O load.
3313 * 0 - Not busy (The request stacking driver should dispatch request)
3314 * 1 - Busy (The request stacking driver should stop dispatching request)
3316 int blk_lld_busy(struct request_queue
*q
)
3319 return q
->lld_busy_fn(q
);
3323 EXPORT_SYMBOL_GPL(blk_lld_busy
);
3326 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3327 * @rq: the clone request to be cleaned up
3330 * Free all bios in @rq for a cloned request.
3332 void blk_rq_unprep_clone(struct request
*rq
)
3336 while ((bio
= rq
->bio
) != NULL
) {
3337 rq
->bio
= bio
->bi_next
;
3342 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone
);
3345 * Copy attributes of the original request to the clone request.
3346 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3348 static void __blk_rq_prep_clone(struct request
*dst
, struct request
*src
)
3350 dst
->cpu
= src
->cpu
;
3351 dst
->__sector
= blk_rq_pos(src
);
3352 dst
->__data_len
= blk_rq_bytes(src
);
3353 if (src
->rq_flags
& RQF_SPECIAL_PAYLOAD
) {
3354 dst
->rq_flags
|= RQF_SPECIAL_PAYLOAD
;
3355 dst
->special_vec
= src
->special_vec
;
3357 dst
->nr_phys_segments
= src
->nr_phys_segments
;
3358 dst
->ioprio
= src
->ioprio
;
3359 dst
->extra_len
= src
->extra_len
;
3363 * blk_rq_prep_clone - Helper function to setup clone request
3364 * @rq: the request to be setup
3365 * @rq_src: original request to be cloned
3366 * @bs: bio_set that bios for clone are allocated from
3367 * @gfp_mask: memory allocation mask for bio
3368 * @bio_ctr: setup function to be called for each clone bio.
3369 * Returns %0 for success, non %0 for failure.
3370 * @data: private data to be passed to @bio_ctr
3373 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3374 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3375 * are not copied, and copying such parts is the caller's responsibility.
3376 * Also, pages which the original bios are pointing to are not copied
3377 * and the cloned bios just point same pages.
3378 * So cloned bios must be completed before original bios, which means
3379 * the caller must complete @rq before @rq_src.
3381 int blk_rq_prep_clone(struct request
*rq
, struct request
*rq_src
,
3382 struct bio_set
*bs
, gfp_t gfp_mask
,
3383 int (*bio_ctr
)(struct bio
*, struct bio
*, void *),
3386 struct bio
*bio
, *bio_src
;
3391 __rq_for_each_bio(bio_src
, rq_src
) {
3392 bio
= bio_clone_fast(bio_src
, gfp_mask
, bs
);
3396 if (bio_ctr
&& bio_ctr(bio
, bio_src
, data
))
3400 rq
->biotail
->bi_next
= bio
;
3403 rq
->bio
= rq
->biotail
= bio
;
3406 __blk_rq_prep_clone(rq
, rq_src
);
3413 blk_rq_unprep_clone(rq
);
3417 EXPORT_SYMBOL_GPL(blk_rq_prep_clone
);
3419 int kblockd_schedule_work(struct work_struct
*work
)
3421 return queue_work(kblockd_workqueue
, work
);
3423 EXPORT_SYMBOL(kblockd_schedule_work
);
3425 int kblockd_schedule_work_on(int cpu
, struct work_struct
*work
)
3427 return queue_work_on(cpu
, kblockd_workqueue
, work
);
3429 EXPORT_SYMBOL(kblockd_schedule_work_on
);
3431 int kblockd_mod_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3432 unsigned long delay
)
3434 return mod_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3436 EXPORT_SYMBOL(kblockd_mod_delayed_work_on
);
3438 int kblockd_schedule_delayed_work(struct delayed_work
*dwork
,
3439 unsigned long delay
)
3441 return queue_delayed_work(kblockd_workqueue
, dwork
, delay
);
3443 EXPORT_SYMBOL(kblockd_schedule_delayed_work
);
3445 int kblockd_schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3446 unsigned long delay
)
3448 return queue_delayed_work_on(cpu
, kblockd_workqueue
, dwork
, delay
);
3450 EXPORT_SYMBOL(kblockd_schedule_delayed_work_on
);
3453 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3454 * @plug: The &struct blk_plug that needs to be initialized
3457 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3458 * pending I/O should the task end up blocking between blk_start_plug() and
3459 * blk_finish_plug(). This is important from a performance perspective, but
3460 * also ensures that we don't deadlock. For instance, if the task is blocking
3461 * for a memory allocation, memory reclaim could end up wanting to free a
3462 * page belonging to that request that is currently residing in our private
3463 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3464 * this kind of deadlock.
3466 void blk_start_plug(struct blk_plug
*plug
)
3468 struct task_struct
*tsk
= current
;
3471 * If this is a nested plug, don't actually assign it.
3476 INIT_LIST_HEAD(&plug
->list
);
3477 INIT_LIST_HEAD(&plug
->mq_list
);
3478 INIT_LIST_HEAD(&plug
->cb_list
);
3480 * Store ordering should not be needed here, since a potential
3481 * preempt will imply a full memory barrier
3485 EXPORT_SYMBOL(blk_start_plug
);
3487 static int plug_rq_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3489 struct request
*rqa
= container_of(a
, struct request
, queuelist
);
3490 struct request
*rqb
= container_of(b
, struct request
, queuelist
);
3492 return !(rqa
->q
< rqb
->q
||
3493 (rqa
->q
== rqb
->q
&& blk_rq_pos(rqa
) < blk_rq_pos(rqb
)));
3497 * If 'from_schedule' is true, then postpone the dispatch of requests
3498 * until a safe kblockd context. We due this to avoid accidental big
3499 * additional stack usage in driver dispatch, in places where the originally
3500 * plugger did not intend it.
3502 static void queue_unplugged(struct request_queue
*q
, unsigned int depth
,
3504 __releases(q
->queue_lock
)
3506 lockdep_assert_held(q
->queue_lock
);
3508 trace_block_unplug(q
, depth
, !from_schedule
);
3511 blk_run_queue_async(q
);
3514 spin_unlock(q
->queue_lock
);
3517 static void flush_plug_callbacks(struct blk_plug
*plug
, bool from_schedule
)
3519 LIST_HEAD(callbacks
);
3521 while (!list_empty(&plug
->cb_list
)) {
3522 list_splice_init(&plug
->cb_list
, &callbacks
);
3524 while (!list_empty(&callbacks
)) {
3525 struct blk_plug_cb
*cb
= list_first_entry(&callbacks
,
3528 list_del(&cb
->list
);
3529 cb
->callback(cb
, from_schedule
);
3534 struct blk_plug_cb
*blk_check_plugged(blk_plug_cb_fn unplug
, void *data
,
3537 struct blk_plug
*plug
= current
->plug
;
3538 struct blk_plug_cb
*cb
;
3543 list_for_each_entry(cb
, &plug
->cb_list
, list
)
3544 if (cb
->callback
== unplug
&& cb
->data
== data
)
3547 /* Not currently on the callback list */
3548 BUG_ON(size
< sizeof(*cb
));
3549 cb
= kzalloc(size
, GFP_ATOMIC
);
3552 cb
->callback
= unplug
;
3553 list_add(&cb
->list
, &plug
->cb_list
);
3557 EXPORT_SYMBOL(blk_check_plugged
);
3559 void blk_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
)
3561 struct request_queue
*q
;
3562 unsigned long flags
;
3567 flush_plug_callbacks(plug
, from_schedule
);
3569 if (!list_empty(&plug
->mq_list
))
3570 blk_mq_flush_plug_list(plug
, from_schedule
);
3572 if (list_empty(&plug
->list
))
3575 list_splice_init(&plug
->list
, &list
);
3577 list_sort(NULL
, &list
, plug_rq_cmp
);
3583 * Save and disable interrupts here, to avoid doing it for every
3584 * queue lock we have to take.
3586 local_irq_save(flags
);
3587 while (!list_empty(&list
)) {
3588 rq
= list_entry_rq(list
.next
);
3589 list_del_init(&rq
->queuelist
);
3593 * This drops the queue lock
3596 queue_unplugged(q
, depth
, from_schedule
);
3599 spin_lock(q
->queue_lock
);
3603 * Short-circuit if @q is dead
3605 if (unlikely(blk_queue_dying(q
))) {
3606 __blk_end_request_all(rq
, BLK_STS_IOERR
);
3611 * rq is already accounted, so use raw insert
3613 if (op_is_flush(rq
->cmd_flags
))
3614 __elv_add_request(q
, rq
, ELEVATOR_INSERT_FLUSH
);
3616 __elv_add_request(q
, rq
, ELEVATOR_INSERT_SORT_MERGE
);
3622 * This drops the queue lock
3625 queue_unplugged(q
, depth
, from_schedule
);
3627 local_irq_restore(flags
);
3630 void blk_finish_plug(struct blk_plug
*plug
)
3632 if (plug
!= current
->plug
)
3634 blk_flush_plug_list(plug
, false);
3636 current
->plug
= NULL
;
3638 EXPORT_SYMBOL(blk_finish_plug
);
3642 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3643 * @q: the queue of the device
3644 * @dev: the device the queue belongs to
3647 * Initialize runtime-PM-related fields for @q and start auto suspend for
3648 * @dev. Drivers that want to take advantage of request-based runtime PM
3649 * should call this function after @dev has been initialized, and its
3650 * request queue @q has been allocated, and runtime PM for it can not happen
3651 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3652 * cases, driver should call this function before any I/O has taken place.
3654 * This function takes care of setting up using auto suspend for the device,
3655 * the autosuspend delay is set to -1 to make runtime suspend impossible
3656 * until an updated value is either set by user or by driver. Drivers do
3657 * not need to touch other autosuspend settings.
3659 * The block layer runtime PM is request based, so only works for drivers
3660 * that use request as their IO unit instead of those directly use bio's.
3662 void blk_pm_runtime_init(struct request_queue
*q
, struct device
*dev
)
3664 /* Don't enable runtime PM for blk-mq until it is ready */
3666 pm_runtime_disable(dev
);
3671 q
->rpm_status
= RPM_ACTIVE
;
3672 pm_runtime_set_autosuspend_delay(q
->dev
, -1);
3673 pm_runtime_use_autosuspend(q
->dev
);
3675 EXPORT_SYMBOL(blk_pm_runtime_init
);
3678 * blk_pre_runtime_suspend - Pre runtime suspend check
3679 * @q: the queue of the device
3682 * This function will check if runtime suspend is allowed for the device
3683 * by examining if there are any requests pending in the queue. If there
3684 * are requests pending, the device can not be runtime suspended; otherwise,
3685 * the queue's status will be updated to SUSPENDING and the driver can
3686 * proceed to suspend the device.
3688 * For the not allowed case, we mark last busy for the device so that
3689 * runtime PM core will try to autosuspend it some time later.
3691 * This function should be called near the start of the device's
3692 * runtime_suspend callback.
3695 * 0 - OK to runtime suspend the device
3696 * -EBUSY - Device should not be runtime suspended
3698 int blk_pre_runtime_suspend(struct request_queue
*q
)
3705 spin_lock_irq(q
->queue_lock
);
3706 if (q
->nr_pending
) {
3708 pm_runtime_mark_last_busy(q
->dev
);
3710 q
->rpm_status
= RPM_SUSPENDING
;
3712 spin_unlock_irq(q
->queue_lock
);
3715 EXPORT_SYMBOL(blk_pre_runtime_suspend
);
3718 * blk_post_runtime_suspend - Post runtime suspend processing
3719 * @q: the queue of the device
3720 * @err: return value of the device's runtime_suspend function
3723 * Update the queue's runtime status according to the return value of the
3724 * device's runtime suspend function and mark last busy for the device so
3725 * that PM core will try to auto suspend the device at a later time.
3727 * This function should be called near the end of the device's
3728 * runtime_suspend callback.
3730 void blk_post_runtime_suspend(struct request_queue
*q
, int err
)
3735 spin_lock_irq(q
->queue_lock
);
3737 q
->rpm_status
= RPM_SUSPENDED
;
3739 q
->rpm_status
= RPM_ACTIVE
;
3740 pm_runtime_mark_last_busy(q
->dev
);
3742 spin_unlock_irq(q
->queue_lock
);
3744 EXPORT_SYMBOL(blk_post_runtime_suspend
);
3747 * blk_pre_runtime_resume - Pre runtime resume processing
3748 * @q: the queue of the device
3751 * Update the queue's runtime status to RESUMING in preparation for the
3752 * runtime resume of the device.
3754 * This function should be called near the start of the device's
3755 * runtime_resume callback.
3757 void blk_pre_runtime_resume(struct request_queue
*q
)
3762 spin_lock_irq(q
->queue_lock
);
3763 q
->rpm_status
= RPM_RESUMING
;
3764 spin_unlock_irq(q
->queue_lock
);
3766 EXPORT_SYMBOL(blk_pre_runtime_resume
);
3769 * blk_post_runtime_resume - Post runtime resume processing
3770 * @q: the queue of the device
3771 * @err: return value of the device's runtime_resume function
3774 * Update the queue's runtime status according to the return value of the
3775 * device's runtime_resume function. If it is successfully resumed, process
3776 * the requests that are queued into the device's queue when it is resuming
3777 * and then mark last busy and initiate autosuspend for it.
3779 * This function should be called near the end of the device's
3780 * runtime_resume callback.
3782 void blk_post_runtime_resume(struct request_queue
*q
, int err
)
3787 spin_lock_irq(q
->queue_lock
);
3789 q
->rpm_status
= RPM_ACTIVE
;
3791 pm_runtime_mark_last_busy(q
->dev
);
3792 pm_request_autosuspend(q
->dev
);
3794 q
->rpm_status
= RPM_SUSPENDED
;
3796 spin_unlock_irq(q
->queue_lock
);
3798 EXPORT_SYMBOL(blk_post_runtime_resume
);
3801 * blk_set_runtime_active - Force runtime status of the queue to be active
3802 * @q: the queue of the device
3804 * If the device is left runtime suspended during system suspend the resume
3805 * hook typically resumes the device and corrects runtime status
3806 * accordingly. However, that does not affect the queue runtime PM status
3807 * which is still "suspended". This prevents processing requests from the
3810 * This function can be used in driver's resume hook to correct queue
3811 * runtime PM status and re-enable peeking requests from the queue. It
3812 * should be called before first request is added to the queue.
3814 void blk_set_runtime_active(struct request_queue
*q
)
3816 spin_lock_irq(q
->queue_lock
);
3817 q
->rpm_status
= RPM_ACTIVE
;
3818 pm_runtime_mark_last_busy(q
->dev
);
3819 pm_request_autosuspend(q
->dev
);
3820 spin_unlock_irq(q
->queue_lock
);
3822 EXPORT_SYMBOL(blk_set_runtime_active
);
3825 int __init
blk_dev_init(void)
3827 BUILD_BUG_ON(REQ_OP_LAST
>= (1 << REQ_OP_BITS
));
3828 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3829 FIELD_SIZEOF(struct request
, cmd_flags
));
3830 BUILD_BUG_ON(REQ_OP_BITS
+ REQ_FLAG_BITS
> 8 *
3831 FIELD_SIZEOF(struct bio
, bi_opf
));
3833 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3834 kblockd_workqueue
= alloc_workqueue("kblockd",
3835 WQ_MEM_RECLAIM
| WQ_HIGHPRI
, 0);
3836 if (!kblockd_workqueue
)
3837 panic("Failed to create kblockd\n");
3839 request_cachep
= kmem_cache_create("blkdev_requests",
3840 sizeof(struct request
), 0, SLAB_PANIC
, NULL
);
3842 blk_requestq_cachep
= kmem_cache_create("request_queue",
3843 sizeof(struct request_queue
), 0, SLAB_PANIC
, NULL
);
3845 #ifdef CONFIG_DEBUG_FS
3846 blk_debugfs_root
= debugfs_create_dir("block", NULL
);