]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame_incremental - block/blk-core.c
xen-blkfront: Avoid that gcc 7 warns about fall-through when building with W=1
[mirror_ubuntu-bionic-kernel.git] / block / blk-core.c
... / ...
CommitLineData
1/*
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>
7 * - July2000
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */
10
11/*
12 * This handles all read/write requests to block devices
13 */
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>
21#include <linux/mm.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>
37
38#define CREATE_TRACE_POINTS
39#include <trace/events/block.h>
40
41#include "blk.h"
42#include "blk-mq.h"
43#include "blk-mq-sched.h"
44#include "blk-wbt.h"
45
46#ifdef CONFIG_DEBUG_FS
47struct dentry *blk_debugfs_root;
48#endif
49
50EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
51EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
52EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
53EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
54EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
55
56DEFINE_IDA(blk_queue_ida);
57
58/*
59 * For the allocated request tables
60 */
61struct kmem_cache *request_cachep;
62
63/*
64 * For queue allocation
65 */
66struct kmem_cache *blk_requestq_cachep;
67
68/*
69 * Controlling structure to kblockd
70 */
71static struct workqueue_struct *kblockd_workqueue;
72
73static void blk_clear_congested(struct request_list *rl, int sync)
74{
75#ifdef CONFIG_CGROUP_WRITEBACK
76 clear_wb_congested(rl->blkg->wb_congested, sync);
77#else
78 /*
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.
81 */
82 if (rl == &rl->q->root_rl)
83 clear_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
84#endif
85}
86
87static void blk_set_congested(struct request_list *rl, int sync)
88{
89#ifdef CONFIG_CGROUP_WRITEBACK
90 set_wb_congested(rl->blkg->wb_congested, sync);
91#else
92 /* see blk_clear_congested() */
93 if (rl == &rl->q->root_rl)
94 set_wb_congested(rl->q->backing_dev_info->wb.congested, sync);
95#endif
96}
97
98void blk_queue_congestion_threshold(struct request_queue *q)
99{
100 int nr;
101
102 nr = q->nr_requests - (q->nr_requests / 8) + 1;
103 if (nr > q->nr_requests)
104 nr = q->nr_requests;
105 q->nr_congestion_on = nr;
106
107 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
108 if (nr < 1)
109 nr = 1;
110 q->nr_congestion_off = nr;
111}
112
113void blk_rq_init(struct request_queue *q, struct request *rq)
114{
115 memset(rq, 0, sizeof(*rq));
116
117 INIT_LIST_HEAD(&rq->queuelist);
118 INIT_LIST_HEAD(&rq->timeout_list);
119 rq->cpu = -1;
120 rq->q = q;
121 rq->__sector = (sector_t) -1;
122 INIT_HLIST_NODE(&rq->hash);
123 RB_CLEAR_NODE(&rq->rb_node);
124 rq->tag = -1;
125 rq->internal_tag = -1;
126 rq->start_time = jiffies;
127 set_start_time_ns(rq);
128 rq->part = NULL;
129}
130EXPORT_SYMBOL(blk_rq_init);
131
132static const struct {
133 int errno;
134 const char *name;
135} blk_errors[] = {
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" },
147
148 /* device mapper special case, should not leak out: */
149 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
150
151 /* everything else not covered above: */
152 [BLK_STS_IOERR] = { -EIO, "I/O" },
153};
154
155blk_status_t errno_to_blk_status(int errno)
156{
157 int i;
158
159 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
160 if (blk_errors[i].errno == errno)
161 return (__force blk_status_t)i;
162 }
163
164 return BLK_STS_IOERR;
165}
166EXPORT_SYMBOL_GPL(errno_to_blk_status);
167
168int blk_status_to_errno(blk_status_t status)
169{
170 int idx = (__force int)status;
171
172 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
173 return -EIO;
174 return blk_errors[idx].errno;
175}
176EXPORT_SYMBOL_GPL(blk_status_to_errno);
177
178static void print_req_error(struct request *req, blk_status_t status)
179{
180 int idx = (__force int)status;
181
182 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
183 return;
184
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));
189}
190
191static void req_bio_endio(struct request *rq, struct bio *bio,
192 unsigned int nbytes, blk_status_t error)
193{
194 if (error)
195 bio->bi_status = error;
196
197 if (unlikely(rq->rq_flags & RQF_QUIET))
198 bio_set_flag(bio, BIO_QUIET);
199
200 bio_advance(bio, nbytes);
201
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))
204 bio_endio(bio);
205}
206
207void blk_dump_rq_flags(struct request *rq, char *msg)
208{
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);
212
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));
218}
219EXPORT_SYMBOL(blk_dump_rq_flags);
220
221static void blk_delay_work(struct work_struct *work)
222{
223 struct request_queue *q;
224
225 q = container_of(work, struct request_queue, delay_work.work);
226 spin_lock_irq(q->queue_lock);
227 __blk_run_queue(q);
228 spin_unlock_irq(q->queue_lock);
229}
230
231/**
232 * blk_delay_queue - restart queueing after defined interval
233 * @q: The &struct request_queue in question
234 * @msecs: Delay in msecs
235 *
236 * Description:
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.
240 */
241void blk_delay_queue(struct request_queue *q, unsigned long msecs)
242{
243 lockdep_assert_held(q->queue_lock);
244 WARN_ON_ONCE(q->mq_ops);
245
246 if (likely(!blk_queue_dead(q)))
247 queue_delayed_work(kblockd_workqueue, &q->delay_work,
248 msecs_to_jiffies(msecs));
249}
250EXPORT_SYMBOL(blk_delay_queue);
251
252/**
253 * blk_start_queue_async - asynchronously restart a previously stopped queue
254 * @q: The &struct request_queue in question
255 *
256 * Description:
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
259 * context.
260 **/
261void blk_start_queue_async(struct request_queue *q)
262{
263 lockdep_assert_held(q->queue_lock);
264 WARN_ON_ONCE(q->mq_ops);
265
266 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
267 blk_run_queue_async(q);
268}
269EXPORT_SYMBOL(blk_start_queue_async);
270
271/**
272 * blk_start_queue - restart a previously stopped queue
273 * @q: The &struct request_queue in question
274 *
275 * Description:
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().
279 **/
280void blk_start_queue(struct request_queue *q)
281{
282 lockdep_assert_held(q->queue_lock);
283 WARN_ON(!irqs_disabled());
284 WARN_ON_ONCE(q->mq_ops);
285
286 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
287 __blk_run_queue(q);
288}
289EXPORT_SYMBOL(blk_start_queue);
290
291/**
292 * blk_stop_queue - stop a queue
293 * @q: The &struct request_queue in question
294 *
295 * Description:
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.
304 **/
305void blk_stop_queue(struct request_queue *q)
306{
307 lockdep_assert_held(q->queue_lock);
308 WARN_ON_ONCE(q->mq_ops);
309
310 cancel_delayed_work(&q->delay_work);
311 queue_flag_set(QUEUE_FLAG_STOPPED, q);
312}
313EXPORT_SYMBOL(blk_stop_queue);
314
315/**
316 * blk_sync_queue - cancel any pending callbacks on a queue
317 * @q: the queue
318 *
319 * Description:
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
326 * this function.
327 *
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.
331 *
332 */
333void blk_sync_queue(struct request_queue *q)
334{
335 del_timer_sync(&q->timeout);
336
337 if (q->mq_ops) {
338 struct blk_mq_hw_ctx *hctx;
339 int i;
340
341 queue_for_each_hw_ctx(q, hctx, i)
342 cancel_delayed_work_sync(&hctx->run_work);
343 } else {
344 cancel_delayed_work_sync(&q->delay_work);
345 }
346}
347EXPORT_SYMBOL(blk_sync_queue);
348
349/**
350 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
351 * @q: The queue to run
352 *
353 * Description:
354 * Invoke request handling on a queue if there are any pending requests.
355 * May be used to restart request handling after a request has completed.
356 * This variant runs the queue whether or not the queue has been
357 * stopped. Must be called with the queue lock held and interrupts
358 * disabled. See also @blk_run_queue.
359 */
360inline void __blk_run_queue_uncond(struct request_queue *q)
361{
362 lockdep_assert_held(q->queue_lock);
363 WARN_ON_ONCE(q->mq_ops);
364
365 if (unlikely(blk_queue_dead(q)))
366 return;
367
368 /*
369 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
370 * the queue lock internally. As a result multiple threads may be
371 * running such a request function concurrently. Keep track of the
372 * number of active request_fn invocations such that blk_drain_queue()
373 * can wait until all these request_fn calls have finished.
374 */
375 q->request_fn_active++;
376 q->request_fn(q);
377 q->request_fn_active--;
378}
379EXPORT_SYMBOL_GPL(__blk_run_queue_uncond);
380
381/**
382 * __blk_run_queue - run a single device queue
383 * @q: The queue to run
384 *
385 * Description:
386 * See @blk_run_queue.
387 */
388void __blk_run_queue(struct request_queue *q)
389{
390 lockdep_assert_held(q->queue_lock);
391 WARN_ON_ONCE(q->mq_ops);
392
393 if (unlikely(blk_queue_stopped(q)))
394 return;
395
396 __blk_run_queue_uncond(q);
397}
398EXPORT_SYMBOL(__blk_run_queue);
399
400/**
401 * blk_run_queue_async - run a single device queue in workqueue context
402 * @q: The queue to run
403 *
404 * Description:
405 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
406 * of us.
407 *
408 * Note:
409 * Since it is not allowed to run q->delay_work after blk_cleanup_queue()
410 * has canceled q->delay_work, callers must hold the queue lock to avoid
411 * race conditions between blk_cleanup_queue() and blk_run_queue_async().
412 */
413void blk_run_queue_async(struct request_queue *q)
414{
415 lockdep_assert_held(q->queue_lock);
416 WARN_ON_ONCE(q->mq_ops);
417
418 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
419 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
420}
421EXPORT_SYMBOL(blk_run_queue_async);
422
423/**
424 * blk_run_queue - run a single device queue
425 * @q: The queue to run
426 *
427 * Description:
428 * Invoke request handling on this queue, if it has pending work to do.
429 * May be used to restart queueing when a request has completed.
430 */
431void blk_run_queue(struct request_queue *q)
432{
433 unsigned long flags;
434
435 WARN_ON_ONCE(q->mq_ops);
436
437 spin_lock_irqsave(q->queue_lock, flags);
438 __blk_run_queue(q);
439 spin_unlock_irqrestore(q->queue_lock, flags);
440}
441EXPORT_SYMBOL(blk_run_queue);
442
443void blk_put_queue(struct request_queue *q)
444{
445 kobject_put(&q->kobj);
446}
447EXPORT_SYMBOL(blk_put_queue);
448
449/**
450 * __blk_drain_queue - drain requests from request_queue
451 * @q: queue to drain
452 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
453 *
454 * Drain requests from @q. If @drain_all is set, all requests are drained.
455 * If not, only ELVPRIV requests are drained. The caller is responsible
456 * for ensuring that no new requests which need to be drained are queued.
457 */
458static void __blk_drain_queue(struct request_queue *q, bool drain_all)
459 __releases(q->queue_lock)
460 __acquires(q->queue_lock)
461{
462 int i;
463
464 lockdep_assert_held(q->queue_lock);
465 WARN_ON_ONCE(q->mq_ops);
466
467 while (true) {
468 bool drain = false;
469
470 /*
471 * The caller might be trying to drain @q before its
472 * elevator is initialized.
473 */
474 if (q->elevator)
475 elv_drain_elevator(q);
476
477 blkcg_drain_queue(q);
478
479 /*
480 * This function might be called on a queue which failed
481 * driver init after queue creation or is not yet fully
482 * active yet. Some drivers (e.g. fd and loop) get unhappy
483 * in such cases. Kick queue iff dispatch queue has
484 * something on it and @q has request_fn set.
485 */
486 if (!list_empty(&q->queue_head) && q->request_fn)
487 __blk_run_queue(q);
488
489 drain |= q->nr_rqs_elvpriv;
490 drain |= q->request_fn_active;
491
492 /*
493 * Unfortunately, requests are queued at and tracked from
494 * multiple places and there's no single counter which can
495 * be drained. Check all the queues and counters.
496 */
497 if (drain_all) {
498 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
499 drain |= !list_empty(&q->queue_head);
500 for (i = 0; i < 2; i++) {
501 drain |= q->nr_rqs[i];
502 drain |= q->in_flight[i];
503 if (fq)
504 drain |= !list_empty(&fq->flush_queue[i]);
505 }
506 }
507
508 if (!drain)
509 break;
510
511 spin_unlock_irq(q->queue_lock);
512
513 msleep(10);
514
515 spin_lock_irq(q->queue_lock);
516 }
517
518 /*
519 * With queue marked dead, any woken up waiter will fail the
520 * allocation path, so the wakeup chaining is lost and we're
521 * left with hung waiters. We need to wake up those waiters.
522 */
523 if (q->request_fn) {
524 struct request_list *rl;
525
526 blk_queue_for_each_rl(rl, q)
527 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
528 wake_up_all(&rl->wait[i]);
529 }
530}
531
532/**
533 * blk_queue_bypass_start - enter queue bypass mode
534 * @q: queue of interest
535 *
536 * In bypass mode, only the dispatch FIFO queue of @q is used. This
537 * function makes @q enter bypass mode and drains all requests which were
538 * throttled or issued before. On return, it's guaranteed that no request
539 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
540 * inside queue or RCU read lock.
541 */
542void blk_queue_bypass_start(struct request_queue *q)
543{
544 WARN_ON_ONCE(q->mq_ops);
545
546 spin_lock_irq(q->queue_lock);
547 q->bypass_depth++;
548 queue_flag_set(QUEUE_FLAG_BYPASS, q);
549 spin_unlock_irq(q->queue_lock);
550
551 /*
552 * Queues start drained. Skip actual draining till init is
553 * complete. This avoids lenghty delays during queue init which
554 * can happen many times during boot.
555 */
556 if (blk_queue_init_done(q)) {
557 spin_lock_irq(q->queue_lock);
558 __blk_drain_queue(q, false);
559 spin_unlock_irq(q->queue_lock);
560
561 /* ensure blk_queue_bypass() is %true inside RCU read lock */
562 synchronize_rcu();
563 }
564}
565EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
566
567/**
568 * blk_queue_bypass_end - leave queue bypass mode
569 * @q: queue of interest
570 *
571 * Leave bypass mode and restore the normal queueing behavior.
572 *
573 * Note: although blk_queue_bypass_start() is only called for blk-sq queues,
574 * this function is called for both blk-sq and blk-mq queues.
575 */
576void blk_queue_bypass_end(struct request_queue *q)
577{
578 spin_lock_irq(q->queue_lock);
579 if (!--q->bypass_depth)
580 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
581 WARN_ON_ONCE(q->bypass_depth < 0);
582 spin_unlock_irq(q->queue_lock);
583}
584EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
585
586void blk_set_queue_dying(struct request_queue *q)
587{
588 spin_lock_irq(q->queue_lock);
589 queue_flag_set(QUEUE_FLAG_DYING, q);
590 spin_unlock_irq(q->queue_lock);
591
592 /*
593 * When queue DYING flag is set, we need to block new req
594 * entering queue, so we call blk_freeze_queue_start() to
595 * prevent I/O from crossing blk_queue_enter().
596 */
597 blk_freeze_queue_start(q);
598
599 if (q->mq_ops)
600 blk_mq_wake_waiters(q);
601 else {
602 struct request_list *rl;
603
604 spin_lock_irq(q->queue_lock);
605 blk_queue_for_each_rl(rl, q) {
606 if (rl->rq_pool) {
607 wake_up(&rl->wait[BLK_RW_SYNC]);
608 wake_up(&rl->wait[BLK_RW_ASYNC]);
609 }
610 }
611 spin_unlock_irq(q->queue_lock);
612 }
613}
614EXPORT_SYMBOL_GPL(blk_set_queue_dying);
615
616/**
617 * blk_cleanup_queue - shutdown a request queue
618 * @q: request queue to shutdown
619 *
620 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
621 * put it. All future requests will be failed immediately with -ENODEV.
622 */
623void blk_cleanup_queue(struct request_queue *q)
624{
625 spinlock_t *lock = q->queue_lock;
626
627 /* mark @q DYING, no new request or merges will be allowed afterwards */
628 mutex_lock(&q->sysfs_lock);
629 blk_set_queue_dying(q);
630 spin_lock_irq(lock);
631
632 /*
633 * A dying queue is permanently in bypass mode till released. Note
634 * that, unlike blk_queue_bypass_start(), we aren't performing
635 * synchronize_rcu() after entering bypass mode to avoid the delay
636 * as some drivers create and destroy a lot of queues while
637 * probing. This is still safe because blk_release_queue() will be
638 * called only after the queue refcnt drops to zero and nothing,
639 * RCU or not, would be traversing the queue by then.
640 */
641 q->bypass_depth++;
642 queue_flag_set(QUEUE_FLAG_BYPASS, q);
643
644 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
645 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
646 queue_flag_set(QUEUE_FLAG_DYING, q);
647 spin_unlock_irq(lock);
648 mutex_unlock(&q->sysfs_lock);
649
650 /*
651 * Drain all requests queued before DYING marking. Set DEAD flag to
652 * prevent that q->request_fn() gets invoked after draining finished.
653 */
654 blk_freeze_queue(q);
655 spin_lock_irq(lock);
656 if (!q->mq_ops)
657 __blk_drain_queue(q, true);
658 queue_flag_set(QUEUE_FLAG_DEAD, q);
659 spin_unlock_irq(lock);
660
661 /* for synchronous bio-based driver finish in-flight integrity i/o */
662 blk_flush_integrity();
663
664 /* @q won't process any more request, flush async actions */
665 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
666 blk_sync_queue(q);
667
668 if (q->mq_ops)
669 blk_mq_free_queue(q);
670 percpu_ref_exit(&q->q_usage_counter);
671
672 spin_lock_irq(lock);
673 if (q->queue_lock != &q->__queue_lock)
674 q->queue_lock = &q->__queue_lock;
675 spin_unlock_irq(lock);
676
677 /* @q is and will stay empty, shutdown and put */
678 blk_put_queue(q);
679}
680EXPORT_SYMBOL(blk_cleanup_queue);
681
682/* Allocate memory local to the request queue */
683static void *alloc_request_simple(gfp_t gfp_mask, void *data)
684{
685 struct request_queue *q = data;
686
687 return kmem_cache_alloc_node(request_cachep, gfp_mask, q->node);
688}
689
690static void free_request_simple(void *element, void *data)
691{
692 kmem_cache_free(request_cachep, element);
693}
694
695static void *alloc_request_size(gfp_t gfp_mask, void *data)
696{
697 struct request_queue *q = data;
698 struct request *rq;
699
700 rq = kmalloc_node(sizeof(struct request) + q->cmd_size, gfp_mask,
701 q->node);
702 if (rq && q->init_rq_fn && q->init_rq_fn(q, rq, gfp_mask) < 0) {
703 kfree(rq);
704 rq = NULL;
705 }
706 return rq;
707}
708
709static void free_request_size(void *element, void *data)
710{
711 struct request_queue *q = data;
712
713 if (q->exit_rq_fn)
714 q->exit_rq_fn(q, element);
715 kfree(element);
716}
717
718int blk_init_rl(struct request_list *rl, struct request_queue *q,
719 gfp_t gfp_mask)
720{
721 if (unlikely(rl->rq_pool))
722 return 0;
723
724 rl->q = q;
725 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
726 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
727 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
728 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
729
730 if (q->cmd_size) {
731 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
732 alloc_request_size, free_request_size,
733 q, gfp_mask, q->node);
734 } else {
735 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ,
736 alloc_request_simple, free_request_simple,
737 q, gfp_mask, q->node);
738 }
739 if (!rl->rq_pool)
740 return -ENOMEM;
741
742 if (rl != &q->root_rl)
743 WARN_ON_ONCE(!blk_get_queue(q));
744
745 return 0;
746}
747
748void blk_exit_rl(struct request_queue *q, struct request_list *rl)
749{
750 if (rl->rq_pool) {
751 mempool_destroy(rl->rq_pool);
752 if (rl != &q->root_rl)
753 blk_put_queue(q);
754 }
755}
756
757struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
758{
759 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
760}
761EXPORT_SYMBOL(blk_alloc_queue);
762
763int blk_queue_enter(struct request_queue *q, bool nowait)
764{
765 while (true) {
766 int ret;
767
768 if (percpu_ref_tryget_live(&q->q_usage_counter))
769 return 0;
770
771 if (nowait)
772 return -EBUSY;
773
774 /*
775 * read pair of barrier in blk_freeze_queue_start(),
776 * we need to order reading __PERCPU_REF_DEAD flag of
777 * .q_usage_counter and reading .mq_freeze_depth or
778 * queue dying flag, otherwise the following wait may
779 * never return if the two reads are reordered.
780 */
781 smp_rmb();
782
783 ret = wait_event_interruptible(q->mq_freeze_wq,
784 !atomic_read(&q->mq_freeze_depth) ||
785 blk_queue_dying(q));
786 if (blk_queue_dying(q))
787 return -ENODEV;
788 if (ret)
789 return ret;
790 }
791}
792
793void blk_queue_exit(struct request_queue *q)
794{
795 percpu_ref_put(&q->q_usage_counter);
796}
797
798static void blk_queue_usage_counter_release(struct percpu_ref *ref)
799{
800 struct request_queue *q =
801 container_of(ref, struct request_queue, q_usage_counter);
802
803 wake_up_all(&q->mq_freeze_wq);
804}
805
806static void blk_rq_timed_out_timer(unsigned long data)
807{
808 struct request_queue *q = (struct request_queue *)data;
809
810 kblockd_schedule_work(&q->timeout_work);
811}
812
813struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
814{
815 struct request_queue *q;
816
817 q = kmem_cache_alloc_node(blk_requestq_cachep,
818 gfp_mask | __GFP_ZERO, node_id);
819 if (!q)
820 return NULL;
821
822 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
823 if (q->id < 0)
824 goto fail_q;
825
826 q->bio_split = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
827 if (!q->bio_split)
828 goto fail_id;
829
830 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
831 if (!q->backing_dev_info)
832 goto fail_split;
833
834 q->stats = blk_alloc_queue_stats();
835 if (!q->stats)
836 goto fail_stats;
837
838 q->backing_dev_info->ra_pages =
839 (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
840 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
841 q->backing_dev_info->name = "block";
842 q->node = node_id;
843
844 setup_timer(&q->backing_dev_info->laptop_mode_wb_timer,
845 laptop_mode_timer_fn, (unsigned long) q);
846 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
847 INIT_LIST_HEAD(&q->queue_head);
848 INIT_LIST_HEAD(&q->timeout_list);
849 INIT_LIST_HEAD(&q->icq_list);
850#ifdef CONFIG_BLK_CGROUP
851 INIT_LIST_HEAD(&q->blkg_list);
852#endif
853 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
854
855 kobject_init(&q->kobj, &blk_queue_ktype);
856
857 mutex_init(&q->sysfs_lock);
858 spin_lock_init(&q->__queue_lock);
859
860 /*
861 * By default initialize queue_lock to internal lock and driver can
862 * override it later if need be.
863 */
864 q->queue_lock = &q->__queue_lock;
865
866 /*
867 * A queue starts its life with bypass turned on to avoid
868 * unnecessary bypass on/off overhead and nasty surprises during
869 * init. The initial bypass will be finished when the queue is
870 * registered by blk_register_queue().
871 */
872 q->bypass_depth = 1;
873 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
874
875 init_waitqueue_head(&q->mq_freeze_wq);
876
877 /*
878 * Init percpu_ref in atomic mode so that it's faster to shutdown.
879 * See blk_register_queue() for details.
880 */
881 if (percpu_ref_init(&q->q_usage_counter,
882 blk_queue_usage_counter_release,
883 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
884 goto fail_bdi;
885
886 if (blkcg_init_queue(q))
887 goto fail_ref;
888
889 return q;
890
891fail_ref:
892 percpu_ref_exit(&q->q_usage_counter);
893fail_bdi:
894 blk_free_queue_stats(q->stats);
895fail_stats:
896 bdi_put(q->backing_dev_info);
897fail_split:
898 bioset_free(q->bio_split);
899fail_id:
900 ida_simple_remove(&blk_queue_ida, q->id);
901fail_q:
902 kmem_cache_free(blk_requestq_cachep, q);
903 return NULL;
904}
905EXPORT_SYMBOL(blk_alloc_queue_node);
906
907/**
908 * blk_init_queue - prepare a request queue for use with a block device
909 * @rfn: The function to be called to process requests that have been
910 * placed on the queue.
911 * @lock: Request queue spin lock
912 *
913 * Description:
914 * If a block device wishes to use the standard request handling procedures,
915 * which sorts requests and coalesces adjacent requests, then it must
916 * call blk_init_queue(). The function @rfn will be called when there
917 * are requests on the queue that need to be processed. If the device
918 * supports plugging, then @rfn may not be called immediately when requests
919 * are available on the queue, but may be called at some time later instead.
920 * Plugged queues are generally unplugged when a buffer belonging to one
921 * of the requests on the queue is needed, or due to memory pressure.
922 *
923 * @rfn is not required, or even expected, to remove all requests off the
924 * queue, but only as many as it can handle at a time. If it does leave
925 * requests on the queue, it is responsible for arranging that the requests
926 * get dealt with eventually.
927 *
928 * The queue spin lock must be held while manipulating the requests on the
929 * request queue; this lock will be taken also from interrupt context, so irq
930 * disabling is needed for it.
931 *
932 * Function returns a pointer to the initialized request queue, or %NULL if
933 * it didn't succeed.
934 *
935 * Note:
936 * blk_init_queue() must be paired with a blk_cleanup_queue() call
937 * when the block device is deactivated (such as at module unload).
938 **/
939
940struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
941{
942 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
943}
944EXPORT_SYMBOL(blk_init_queue);
945
946struct request_queue *
947blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
948{
949 struct request_queue *q;
950
951 q = blk_alloc_queue_node(GFP_KERNEL, node_id);
952 if (!q)
953 return NULL;
954
955 q->request_fn = rfn;
956 if (lock)
957 q->queue_lock = lock;
958 if (blk_init_allocated_queue(q) < 0) {
959 blk_cleanup_queue(q);
960 return NULL;
961 }
962
963 return q;
964}
965EXPORT_SYMBOL(blk_init_queue_node);
966
967static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio);
968
969
970int blk_init_allocated_queue(struct request_queue *q)
971{
972 WARN_ON_ONCE(q->mq_ops);
973
974 q->fq = blk_alloc_flush_queue(q, NUMA_NO_NODE, q->cmd_size);
975 if (!q->fq)
976 return -ENOMEM;
977
978 if (q->init_rq_fn && q->init_rq_fn(q, q->fq->flush_rq, GFP_KERNEL))
979 goto out_free_flush_queue;
980
981 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
982 goto out_exit_flush_rq;
983
984 INIT_WORK(&q->timeout_work, blk_timeout_work);
985 q->queue_flags |= QUEUE_FLAG_DEFAULT;
986
987 /*
988 * This also sets hw/phys segments, boundary and size
989 */
990 blk_queue_make_request(q, blk_queue_bio);
991
992 q->sg_reserved_size = INT_MAX;
993
994 /* Protect q->elevator from elevator_change */
995 mutex_lock(&q->sysfs_lock);
996
997 /* init elevator */
998 if (elevator_init(q, NULL)) {
999 mutex_unlock(&q->sysfs_lock);
1000 goto out_exit_flush_rq;
1001 }
1002
1003 mutex_unlock(&q->sysfs_lock);
1004 return 0;
1005
1006out_exit_flush_rq:
1007 if (q->exit_rq_fn)
1008 q->exit_rq_fn(q, q->fq->flush_rq);
1009out_free_flush_queue:
1010 blk_free_flush_queue(q->fq);
1011 return -ENOMEM;
1012}
1013EXPORT_SYMBOL(blk_init_allocated_queue);
1014
1015bool blk_get_queue(struct request_queue *q)
1016{
1017 if (likely(!blk_queue_dying(q))) {
1018 __blk_get_queue(q);
1019 return true;
1020 }
1021
1022 return false;
1023}
1024EXPORT_SYMBOL(blk_get_queue);
1025
1026static inline void blk_free_request(struct request_list *rl, struct request *rq)
1027{
1028 if (rq->rq_flags & RQF_ELVPRIV) {
1029 elv_put_request(rl->q, rq);
1030 if (rq->elv.icq)
1031 put_io_context(rq->elv.icq->ioc);
1032 }
1033
1034 mempool_free(rq, rl->rq_pool);
1035}
1036
1037/*
1038 * ioc_batching returns true if the ioc is a valid batching request and
1039 * should be given priority access to a request.
1040 */
1041static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
1042{
1043 if (!ioc)
1044 return 0;
1045
1046 /*
1047 * Make sure the process is able to allocate at least 1 request
1048 * even if the batch times out, otherwise we could theoretically
1049 * lose wakeups.
1050 */
1051 return ioc->nr_batch_requests == q->nr_batching ||
1052 (ioc->nr_batch_requests > 0
1053 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
1054}
1055
1056/*
1057 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
1058 * will cause the process to be a "batcher" on all queues in the system. This
1059 * is the behaviour we want though - once it gets a wakeup it should be given
1060 * a nice run.
1061 */
1062static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
1063{
1064 if (!ioc || ioc_batching(q, ioc))
1065 return;
1066
1067 ioc->nr_batch_requests = q->nr_batching;
1068 ioc->last_waited = jiffies;
1069}
1070
1071static void __freed_request(struct request_list *rl, int sync)
1072{
1073 struct request_queue *q = rl->q;
1074
1075 if (rl->count[sync] < queue_congestion_off_threshold(q))
1076 blk_clear_congested(rl, sync);
1077
1078 if (rl->count[sync] + 1 <= q->nr_requests) {
1079 if (waitqueue_active(&rl->wait[sync]))
1080 wake_up(&rl->wait[sync]);
1081
1082 blk_clear_rl_full(rl, sync);
1083 }
1084}
1085
1086/*
1087 * A request has just been released. Account for it, update the full and
1088 * congestion status, wake up any waiters. Called under q->queue_lock.
1089 */
1090static void freed_request(struct request_list *rl, bool sync,
1091 req_flags_t rq_flags)
1092{
1093 struct request_queue *q = rl->q;
1094
1095 q->nr_rqs[sync]--;
1096 rl->count[sync]--;
1097 if (rq_flags & RQF_ELVPRIV)
1098 q->nr_rqs_elvpriv--;
1099
1100 __freed_request(rl, sync);
1101
1102 if (unlikely(rl->starved[sync ^ 1]))
1103 __freed_request(rl, sync ^ 1);
1104}
1105
1106int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
1107{
1108 struct request_list *rl;
1109 int on_thresh, off_thresh;
1110
1111 WARN_ON_ONCE(q->mq_ops);
1112
1113 spin_lock_irq(q->queue_lock);
1114 q->nr_requests = nr;
1115 blk_queue_congestion_threshold(q);
1116 on_thresh = queue_congestion_on_threshold(q);
1117 off_thresh = queue_congestion_off_threshold(q);
1118
1119 blk_queue_for_each_rl(rl, q) {
1120 if (rl->count[BLK_RW_SYNC] >= on_thresh)
1121 blk_set_congested(rl, BLK_RW_SYNC);
1122 else if (rl->count[BLK_RW_SYNC] < off_thresh)
1123 blk_clear_congested(rl, BLK_RW_SYNC);
1124
1125 if (rl->count[BLK_RW_ASYNC] >= on_thresh)
1126 blk_set_congested(rl, BLK_RW_ASYNC);
1127 else if (rl->count[BLK_RW_ASYNC] < off_thresh)
1128 blk_clear_congested(rl, BLK_RW_ASYNC);
1129
1130 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
1131 blk_set_rl_full(rl, BLK_RW_SYNC);
1132 } else {
1133 blk_clear_rl_full(rl, BLK_RW_SYNC);
1134 wake_up(&rl->wait[BLK_RW_SYNC]);
1135 }
1136
1137 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
1138 blk_set_rl_full(rl, BLK_RW_ASYNC);
1139 } else {
1140 blk_clear_rl_full(rl, BLK_RW_ASYNC);
1141 wake_up(&rl->wait[BLK_RW_ASYNC]);
1142 }
1143 }
1144
1145 spin_unlock_irq(q->queue_lock);
1146 return 0;
1147}
1148
1149/**
1150 * __get_request - get a free request
1151 * @rl: request list to allocate from
1152 * @op: operation and flags
1153 * @bio: bio to allocate request for (can be %NULL)
1154 * @gfp_mask: allocation mask
1155 *
1156 * Get a free request from @q. This function may fail under memory
1157 * pressure or if @q is dead.
1158 *
1159 * Must be called with @q->queue_lock held and,
1160 * Returns ERR_PTR on failure, with @q->queue_lock held.
1161 * Returns request pointer on success, with @q->queue_lock *not held*.
1162 */
1163static struct request *__get_request(struct request_list *rl, unsigned int op,
1164 struct bio *bio, gfp_t gfp_mask)
1165{
1166 struct request_queue *q = rl->q;
1167 struct request *rq;
1168 struct elevator_type *et = q->elevator->type;
1169 struct io_context *ioc = rq_ioc(bio);
1170 struct io_cq *icq = NULL;
1171 const bool is_sync = op_is_sync(op);
1172 int may_queue;
1173 req_flags_t rq_flags = RQF_ALLOCED;
1174
1175 lockdep_assert_held(q->queue_lock);
1176
1177 if (unlikely(blk_queue_dying(q)))
1178 return ERR_PTR(-ENODEV);
1179
1180 may_queue = elv_may_queue(q, op);
1181 if (may_queue == ELV_MQUEUE_NO)
1182 goto rq_starved;
1183
1184 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
1185 if (rl->count[is_sync]+1 >= q->nr_requests) {
1186 /*
1187 * The queue will fill after this allocation, so set
1188 * it as full, and mark this process as "batching".
1189 * This process will be allowed to complete a batch of
1190 * requests, others will be blocked.
1191 */
1192 if (!blk_rl_full(rl, is_sync)) {
1193 ioc_set_batching(q, ioc);
1194 blk_set_rl_full(rl, is_sync);
1195 } else {
1196 if (may_queue != ELV_MQUEUE_MUST
1197 && !ioc_batching(q, ioc)) {
1198 /*
1199 * The queue is full and the allocating
1200 * process is not a "batcher", and not
1201 * exempted by the IO scheduler
1202 */
1203 return ERR_PTR(-ENOMEM);
1204 }
1205 }
1206 }
1207 blk_set_congested(rl, is_sync);
1208 }
1209
1210 /*
1211 * Only allow batching queuers to allocate up to 50% over the defined
1212 * limit of requests, otherwise we could have thousands of requests
1213 * allocated with any setting of ->nr_requests
1214 */
1215 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
1216 return ERR_PTR(-ENOMEM);
1217
1218 q->nr_rqs[is_sync]++;
1219 rl->count[is_sync]++;
1220 rl->starved[is_sync] = 0;
1221
1222 /*
1223 * Decide whether the new request will be managed by elevator. If
1224 * so, mark @rq_flags and increment elvpriv. Non-zero elvpriv will
1225 * prevent the current elevator from being destroyed until the new
1226 * request is freed. This guarantees icq's won't be destroyed and
1227 * makes creating new ones safe.
1228 *
1229 * Flush requests do not use the elevator so skip initialization.
1230 * This allows a request to share the flush and elevator data.
1231 *
1232 * Also, lookup icq while holding queue_lock. If it doesn't exist,
1233 * it will be created after releasing queue_lock.
1234 */
1235 if (!op_is_flush(op) && !blk_queue_bypass(q)) {
1236 rq_flags |= RQF_ELVPRIV;
1237 q->nr_rqs_elvpriv++;
1238 if (et->icq_cache && ioc)
1239 icq = ioc_lookup_icq(ioc, q);
1240 }
1241
1242 if (blk_queue_io_stat(q))
1243 rq_flags |= RQF_IO_STAT;
1244 spin_unlock_irq(q->queue_lock);
1245
1246 /* allocate and init request */
1247 rq = mempool_alloc(rl->rq_pool, gfp_mask);
1248 if (!rq)
1249 goto fail_alloc;
1250
1251 blk_rq_init(q, rq);
1252 blk_rq_set_rl(rq, rl);
1253 rq->cmd_flags = op;
1254 rq->rq_flags = rq_flags;
1255
1256 /* init elvpriv */
1257 if (rq_flags & RQF_ELVPRIV) {
1258 if (unlikely(et->icq_cache && !icq)) {
1259 if (ioc)
1260 icq = ioc_create_icq(ioc, q, gfp_mask);
1261 if (!icq)
1262 goto fail_elvpriv;
1263 }
1264
1265 rq->elv.icq = icq;
1266 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1267 goto fail_elvpriv;
1268
1269 /* @rq->elv.icq holds io_context until @rq is freed */
1270 if (icq)
1271 get_io_context(icq->ioc);
1272 }
1273out:
1274 /*
1275 * ioc may be NULL here, and ioc_batching will be false. That's
1276 * OK, if the queue is under the request limit then requests need
1277 * not count toward the nr_batch_requests limit. There will always
1278 * be some limit enforced by BLK_BATCH_TIME.
1279 */
1280 if (ioc_batching(q, ioc))
1281 ioc->nr_batch_requests--;
1282
1283 trace_block_getrq(q, bio, op);
1284 return rq;
1285
1286fail_elvpriv:
1287 /*
1288 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1289 * and may fail indefinitely under memory pressure and thus
1290 * shouldn't stall IO. Treat this request as !elvpriv. This will
1291 * disturb iosched and blkcg but weird is bettern than dead.
1292 */
1293 printk_ratelimited(KERN_WARNING "%s: dev %s: request aux data allocation failed, iosched may be disturbed\n",
1294 __func__, dev_name(q->backing_dev_info->dev));
1295
1296 rq->rq_flags &= ~RQF_ELVPRIV;
1297 rq->elv.icq = NULL;
1298
1299 spin_lock_irq(q->queue_lock);
1300 q->nr_rqs_elvpriv--;
1301 spin_unlock_irq(q->queue_lock);
1302 goto out;
1303
1304fail_alloc:
1305 /*
1306 * Allocation failed presumably due to memory. Undo anything we
1307 * might have messed up.
1308 *
1309 * Allocating task should really be put onto the front of the wait
1310 * queue, but this is pretty rare.
1311 */
1312 spin_lock_irq(q->queue_lock);
1313 freed_request(rl, is_sync, rq_flags);
1314
1315 /*
1316 * in the very unlikely event that allocation failed and no
1317 * requests for this direction was pending, mark us starved so that
1318 * freeing of a request in the other direction will notice
1319 * us. another possible fix would be to split the rq mempool into
1320 * READ and WRITE
1321 */
1322rq_starved:
1323 if (unlikely(rl->count[is_sync] == 0))
1324 rl->starved[is_sync] = 1;
1325 return ERR_PTR(-ENOMEM);
1326}
1327
1328/**
1329 * get_request - get a free request
1330 * @q: request_queue to allocate request from
1331 * @op: operation and flags
1332 * @bio: bio to allocate request for (can be %NULL)
1333 * @gfp_mask: allocation mask
1334 *
1335 * Get a free request from @q. If %__GFP_DIRECT_RECLAIM is set in @gfp_mask,
1336 * this function keeps retrying under memory pressure and fails iff @q is dead.
1337 *
1338 * Must be called with @q->queue_lock held and,
1339 * Returns ERR_PTR on failure, with @q->queue_lock held.
1340 * Returns request pointer on success, with @q->queue_lock *not held*.
1341 */
1342static struct request *get_request(struct request_queue *q, unsigned int op,
1343 struct bio *bio, gfp_t gfp_mask)
1344{
1345 const bool is_sync = op_is_sync(op);
1346 DEFINE_WAIT(wait);
1347 struct request_list *rl;
1348 struct request *rq;
1349
1350 lockdep_assert_held(q->queue_lock);
1351 WARN_ON_ONCE(q->mq_ops);
1352
1353 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1354retry:
1355 rq = __get_request(rl, op, bio, gfp_mask);
1356 if (!IS_ERR(rq))
1357 return rq;
1358
1359 if (op & REQ_NOWAIT) {
1360 blk_put_rl(rl);
1361 return ERR_PTR(-EAGAIN);
1362 }
1363
1364 if (!gfpflags_allow_blocking(gfp_mask) || unlikely(blk_queue_dying(q))) {
1365 blk_put_rl(rl);
1366 return rq;
1367 }
1368
1369 /* wait on @rl and retry */
1370 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1371 TASK_UNINTERRUPTIBLE);
1372
1373 trace_block_sleeprq(q, bio, op);
1374
1375 spin_unlock_irq(q->queue_lock);
1376 io_schedule();
1377
1378 /*
1379 * After sleeping, we become a "batching" process and will be able
1380 * to allocate at least one request, and up to a big batch of them
1381 * for a small period time. See ioc_batching, ioc_set_batching
1382 */
1383 ioc_set_batching(q, current->io_context);
1384
1385 spin_lock_irq(q->queue_lock);
1386 finish_wait(&rl->wait[is_sync], &wait);
1387
1388 goto retry;
1389}
1390
1391static struct request *blk_old_get_request(struct request_queue *q,
1392 unsigned int op, gfp_t gfp_mask)
1393{
1394 struct request *rq;
1395
1396 WARN_ON_ONCE(q->mq_ops);
1397
1398 /* create ioc upfront */
1399 create_io_context(gfp_mask, q->node);
1400
1401 spin_lock_irq(q->queue_lock);
1402 rq = get_request(q, op, NULL, gfp_mask);
1403 if (IS_ERR(rq)) {
1404 spin_unlock_irq(q->queue_lock);
1405 return rq;
1406 }
1407
1408 /* q->queue_lock is unlocked at this point */
1409 rq->__data_len = 0;
1410 rq->__sector = (sector_t) -1;
1411 rq->bio = rq->biotail = NULL;
1412 return rq;
1413}
1414
1415struct request *blk_get_request(struct request_queue *q, unsigned int op,
1416 gfp_t gfp_mask)
1417{
1418 struct request *req;
1419
1420 if (q->mq_ops) {
1421 req = blk_mq_alloc_request(q, op,
1422 (gfp_mask & __GFP_DIRECT_RECLAIM) ?
1423 0 : BLK_MQ_REQ_NOWAIT);
1424 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
1425 q->mq_ops->initialize_rq_fn(req);
1426 } else {
1427 req = blk_old_get_request(q, op, gfp_mask);
1428 if (!IS_ERR(req) && q->initialize_rq_fn)
1429 q->initialize_rq_fn(req);
1430 }
1431
1432 return req;
1433}
1434EXPORT_SYMBOL(blk_get_request);
1435
1436/**
1437 * blk_requeue_request - put a request back on queue
1438 * @q: request queue where request should be inserted
1439 * @rq: request to be inserted
1440 *
1441 * Description:
1442 * Drivers often keep queueing requests until the hardware cannot accept
1443 * more, when that condition happens we need to put the request back
1444 * on the queue. Must be called with queue lock held.
1445 */
1446void blk_requeue_request(struct request_queue *q, struct request *rq)
1447{
1448 lockdep_assert_held(q->queue_lock);
1449 WARN_ON_ONCE(q->mq_ops);
1450
1451 blk_delete_timer(rq);
1452 blk_clear_rq_complete(rq);
1453 trace_block_rq_requeue(q, rq);
1454 wbt_requeue(q->rq_wb, &rq->issue_stat);
1455
1456 if (rq->rq_flags & RQF_QUEUED)
1457 blk_queue_end_tag(q, rq);
1458
1459 BUG_ON(blk_queued_rq(rq));
1460
1461 elv_requeue_request(q, rq);
1462}
1463EXPORT_SYMBOL(blk_requeue_request);
1464
1465static void add_acct_request(struct request_queue *q, struct request *rq,
1466 int where)
1467{
1468 blk_account_io_start(rq, true);
1469 __elv_add_request(q, rq, where);
1470}
1471
1472static void part_round_stats_single(struct request_queue *q, int cpu,
1473 struct hd_struct *part, unsigned long now,
1474 unsigned int inflight)
1475{
1476 if (inflight) {
1477 __part_stat_add(cpu, part, time_in_queue,
1478 inflight * (now - part->stamp));
1479 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1480 }
1481 part->stamp = now;
1482}
1483
1484/**
1485 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1486 * @q: target block queue
1487 * @cpu: cpu number for stats access
1488 * @part: target partition
1489 *
1490 * The average IO queue length and utilisation statistics are maintained
1491 * by observing the current state of the queue length and the amount of
1492 * time it has been in this state for.
1493 *
1494 * Normally, that accounting is done on IO completion, but that can result
1495 * in more than a second's worth of IO being accounted for within any one
1496 * second, leading to >100% utilisation. To deal with that, we call this
1497 * function to do a round-off before returning the results when reading
1498 * /proc/diskstats. This accounts immediately for all queue usage up to
1499 * the current jiffies and restarts the counters again.
1500 */
1501void part_round_stats(struct request_queue *q, int cpu, struct hd_struct *part)
1502{
1503 struct hd_struct *part2 = NULL;
1504 unsigned long now = jiffies;
1505 unsigned int inflight[2];
1506 int stats = 0;
1507
1508 if (part->stamp != now)
1509 stats |= 1;
1510
1511 if (part->partno) {
1512 part2 = &part_to_disk(part)->part0;
1513 if (part2->stamp != now)
1514 stats |= 2;
1515 }
1516
1517 if (!stats)
1518 return;
1519
1520 part_in_flight(q, part, inflight);
1521
1522 if (stats & 2)
1523 part_round_stats_single(q, cpu, part2, now, inflight[1]);
1524 if (stats & 1)
1525 part_round_stats_single(q, cpu, part, now, inflight[0]);
1526}
1527EXPORT_SYMBOL_GPL(part_round_stats);
1528
1529#ifdef CONFIG_PM
1530static void blk_pm_put_request(struct request *rq)
1531{
1532 if (rq->q->dev && !(rq->rq_flags & RQF_PM) && !--rq->q->nr_pending)
1533 pm_runtime_mark_last_busy(rq->q->dev);
1534}
1535#else
1536static inline void blk_pm_put_request(struct request *rq) {}
1537#endif
1538
1539void __blk_put_request(struct request_queue *q, struct request *req)
1540{
1541 req_flags_t rq_flags = req->rq_flags;
1542
1543 if (unlikely(!q))
1544 return;
1545
1546 if (q->mq_ops) {
1547 blk_mq_free_request(req);
1548 return;
1549 }
1550
1551 lockdep_assert_held(q->queue_lock);
1552
1553 blk_pm_put_request(req);
1554
1555 elv_completed_request(q, req);
1556
1557 /* this is a bio leak */
1558 WARN_ON(req->bio != NULL);
1559
1560 wbt_done(q->rq_wb, &req->issue_stat);
1561
1562 /*
1563 * Request may not have originated from ll_rw_blk. if not,
1564 * it didn't come out of our reserved rq pools
1565 */
1566 if (rq_flags & RQF_ALLOCED) {
1567 struct request_list *rl = blk_rq_rl(req);
1568 bool sync = op_is_sync(req->cmd_flags);
1569
1570 BUG_ON(!list_empty(&req->queuelist));
1571 BUG_ON(ELV_ON_HASH(req));
1572
1573 blk_free_request(rl, req);
1574 freed_request(rl, sync, rq_flags);
1575 blk_put_rl(rl);
1576 }
1577}
1578EXPORT_SYMBOL_GPL(__blk_put_request);
1579
1580void blk_put_request(struct request *req)
1581{
1582 struct request_queue *q = req->q;
1583
1584 if (q->mq_ops)
1585 blk_mq_free_request(req);
1586 else {
1587 unsigned long flags;
1588
1589 spin_lock_irqsave(q->queue_lock, flags);
1590 __blk_put_request(q, req);
1591 spin_unlock_irqrestore(q->queue_lock, flags);
1592 }
1593}
1594EXPORT_SYMBOL(blk_put_request);
1595
1596bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1597 struct bio *bio)
1598{
1599 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1600
1601 if (!ll_back_merge_fn(q, req, bio))
1602 return false;
1603
1604 trace_block_bio_backmerge(q, req, bio);
1605
1606 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1607 blk_rq_set_mixed_merge(req);
1608
1609 req->biotail->bi_next = bio;
1610 req->biotail = bio;
1611 req->__data_len += bio->bi_iter.bi_size;
1612 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1613
1614 blk_account_io_start(req, false);
1615 return true;
1616}
1617
1618bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
1619 struct bio *bio)
1620{
1621 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
1622
1623 if (!ll_front_merge_fn(q, req, bio))
1624 return false;
1625
1626 trace_block_bio_frontmerge(q, req, bio);
1627
1628 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1629 blk_rq_set_mixed_merge(req);
1630
1631 bio->bi_next = req->bio;
1632 req->bio = bio;
1633
1634 req->__sector = bio->bi_iter.bi_sector;
1635 req->__data_len += bio->bi_iter.bi_size;
1636 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1637
1638 blk_account_io_start(req, false);
1639 return true;
1640}
1641
1642bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
1643 struct bio *bio)
1644{
1645 unsigned short segments = blk_rq_nr_discard_segments(req);
1646
1647 if (segments >= queue_max_discard_segments(q))
1648 goto no_merge;
1649 if (blk_rq_sectors(req) + bio_sectors(bio) >
1650 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
1651 goto no_merge;
1652
1653 req->biotail->bi_next = bio;
1654 req->biotail = bio;
1655 req->__data_len += bio->bi_iter.bi_size;
1656 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1657 req->nr_phys_segments = segments + 1;
1658
1659 blk_account_io_start(req, false);
1660 return true;
1661no_merge:
1662 req_set_nomerge(q, req);
1663 return false;
1664}
1665
1666/**
1667 * blk_attempt_plug_merge - try to merge with %current's plugged list
1668 * @q: request_queue new bio is being queued at
1669 * @bio: new bio being queued
1670 * @request_count: out parameter for number of traversed plugged requests
1671 * @same_queue_rq: pointer to &struct request that gets filled in when
1672 * another request associated with @q is found on the plug list
1673 * (optional, may be %NULL)
1674 *
1675 * Determine whether @bio being queued on @q can be merged with a request
1676 * on %current's plugged list. Returns %true if merge was successful,
1677 * otherwise %false.
1678 *
1679 * Plugging coalesces IOs from the same issuer for the same purpose without
1680 * going through @q->queue_lock. As such it's more of an issuing mechanism
1681 * than scheduling, and the request, while may have elvpriv data, is not
1682 * added on the elevator at this point. In addition, we don't have
1683 * reliable access to the elevator outside queue lock. Only check basic
1684 * merging parameters without querying the elevator.
1685 *
1686 * Caller must ensure !blk_queue_nomerges(q) beforehand.
1687 */
1688bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
1689 unsigned int *request_count,
1690 struct request **same_queue_rq)
1691{
1692 struct blk_plug *plug;
1693 struct request *rq;
1694 struct list_head *plug_list;
1695
1696 plug = current->plug;
1697 if (!plug)
1698 return false;
1699 *request_count = 0;
1700
1701 if (q->mq_ops)
1702 plug_list = &plug->mq_list;
1703 else
1704 plug_list = &plug->list;
1705
1706 list_for_each_entry_reverse(rq, plug_list, queuelist) {
1707 bool merged = false;
1708
1709 if (rq->q == q) {
1710 (*request_count)++;
1711 /*
1712 * Only blk-mq multiple hardware queues case checks the
1713 * rq in the same queue, there should be only one such
1714 * rq in a queue
1715 **/
1716 if (same_queue_rq)
1717 *same_queue_rq = rq;
1718 }
1719
1720 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1721 continue;
1722
1723 switch (blk_try_merge(rq, bio)) {
1724 case ELEVATOR_BACK_MERGE:
1725 merged = bio_attempt_back_merge(q, rq, bio);
1726 break;
1727 case ELEVATOR_FRONT_MERGE:
1728 merged = bio_attempt_front_merge(q, rq, bio);
1729 break;
1730 case ELEVATOR_DISCARD_MERGE:
1731 merged = bio_attempt_discard_merge(q, rq, bio);
1732 break;
1733 default:
1734 break;
1735 }
1736
1737 if (merged)
1738 return true;
1739 }
1740
1741 return false;
1742}
1743
1744unsigned int blk_plug_queued_count(struct request_queue *q)
1745{
1746 struct blk_plug *plug;
1747 struct request *rq;
1748 struct list_head *plug_list;
1749 unsigned int ret = 0;
1750
1751 plug = current->plug;
1752 if (!plug)
1753 goto out;
1754
1755 if (q->mq_ops)
1756 plug_list = &plug->mq_list;
1757 else
1758 plug_list = &plug->list;
1759
1760 list_for_each_entry(rq, plug_list, queuelist) {
1761 if (rq->q == q)
1762 ret++;
1763 }
1764out:
1765 return ret;
1766}
1767
1768void blk_init_request_from_bio(struct request *req, struct bio *bio)
1769{
1770 struct io_context *ioc = rq_ioc(bio);
1771
1772 if (bio->bi_opf & REQ_RAHEAD)
1773 req->cmd_flags |= REQ_FAILFAST_MASK;
1774
1775 req->__sector = bio->bi_iter.bi_sector;
1776 if (ioprio_valid(bio_prio(bio)))
1777 req->ioprio = bio_prio(bio);
1778 else if (ioc)
1779 req->ioprio = ioc->ioprio;
1780 else
1781 req->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
1782 req->write_hint = bio->bi_write_hint;
1783 blk_rq_bio_prep(req->q, req, bio);
1784}
1785EXPORT_SYMBOL_GPL(blk_init_request_from_bio);
1786
1787static blk_qc_t blk_queue_bio(struct request_queue *q, struct bio *bio)
1788{
1789 struct blk_plug *plug;
1790 int where = ELEVATOR_INSERT_SORT;
1791 struct request *req, *free;
1792 unsigned int request_count = 0;
1793 unsigned int wb_acct;
1794
1795 /*
1796 * low level driver can indicate that it wants pages above a
1797 * certain limit bounced to low memory (ie for highmem, or even
1798 * ISA dma in theory)
1799 */
1800 blk_queue_bounce(q, &bio);
1801
1802 blk_queue_split(q, &bio);
1803
1804 if (!bio_integrity_prep(bio))
1805 return BLK_QC_T_NONE;
1806
1807 if (op_is_flush(bio->bi_opf)) {
1808 spin_lock_irq(q->queue_lock);
1809 where = ELEVATOR_INSERT_FLUSH;
1810 goto get_rq;
1811 }
1812
1813 /*
1814 * Check if we can merge with the plugged list before grabbing
1815 * any locks.
1816 */
1817 if (!blk_queue_nomerges(q)) {
1818 if (blk_attempt_plug_merge(q, bio, &request_count, NULL))
1819 return BLK_QC_T_NONE;
1820 } else
1821 request_count = blk_plug_queued_count(q);
1822
1823 spin_lock_irq(q->queue_lock);
1824
1825 switch (elv_merge(q, &req, bio)) {
1826 case ELEVATOR_BACK_MERGE:
1827 if (!bio_attempt_back_merge(q, req, bio))
1828 break;
1829 elv_bio_merged(q, req, bio);
1830 free = attempt_back_merge(q, req);
1831 if (free)
1832 __blk_put_request(q, free);
1833 else
1834 elv_merged_request(q, req, ELEVATOR_BACK_MERGE);
1835 goto out_unlock;
1836 case ELEVATOR_FRONT_MERGE:
1837 if (!bio_attempt_front_merge(q, req, bio))
1838 break;
1839 elv_bio_merged(q, req, bio);
1840 free = attempt_front_merge(q, req);
1841 if (free)
1842 __blk_put_request(q, free);
1843 else
1844 elv_merged_request(q, req, ELEVATOR_FRONT_MERGE);
1845 goto out_unlock;
1846 default:
1847 break;
1848 }
1849
1850get_rq:
1851 wb_acct = wbt_wait(q->rq_wb, bio, q->queue_lock);
1852
1853 /*
1854 * Grab a free request. This is might sleep but can not fail.
1855 * Returns with the queue unlocked.
1856 */
1857 req = get_request(q, bio->bi_opf, bio, GFP_NOIO);
1858 if (IS_ERR(req)) {
1859 __wbt_done(q->rq_wb, wb_acct);
1860 if (PTR_ERR(req) == -ENOMEM)
1861 bio->bi_status = BLK_STS_RESOURCE;
1862 else
1863 bio->bi_status = BLK_STS_IOERR;
1864 bio_endio(bio);
1865 goto out_unlock;
1866 }
1867
1868 wbt_track(&req->issue_stat, wb_acct);
1869
1870 /*
1871 * After dropping the lock and possibly sleeping here, our request
1872 * may now be mergeable after it had proven unmergeable (above).
1873 * We don't worry about that case for efficiency. It won't happen
1874 * often, and the elevators are able to handle it.
1875 */
1876 blk_init_request_from_bio(req, bio);
1877
1878 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1879 req->cpu = raw_smp_processor_id();
1880
1881 plug = current->plug;
1882 if (plug) {
1883 /*
1884 * If this is the first request added after a plug, fire
1885 * of a plug trace.
1886 *
1887 * @request_count may become stale because of schedule
1888 * out, so check plug list again.
1889 */
1890 if (!request_count || list_empty(&plug->list))
1891 trace_block_plug(q);
1892 else {
1893 struct request *last = list_entry_rq(plug->list.prev);
1894 if (request_count >= BLK_MAX_REQUEST_COUNT ||
1895 blk_rq_bytes(last) >= BLK_PLUG_FLUSH_SIZE) {
1896 blk_flush_plug_list(plug, false);
1897 trace_block_plug(q);
1898 }
1899 }
1900 list_add_tail(&req->queuelist, &plug->list);
1901 blk_account_io_start(req, true);
1902 } else {
1903 spin_lock_irq(q->queue_lock);
1904 add_acct_request(q, req, where);
1905 __blk_run_queue(q);
1906out_unlock:
1907 spin_unlock_irq(q->queue_lock);
1908 }
1909
1910 return BLK_QC_T_NONE;
1911}
1912
1913/*
1914 * If bio->bi_dev is a partition, remap the location
1915 */
1916static inline void blk_partition_remap(struct bio *bio)
1917{
1918 struct block_device *bdev = bio->bi_bdev;
1919
1920 /*
1921 * Zone reset does not include bi_size so bio_sectors() is always 0.
1922 * Include a test for the reset op code and perform the remap if needed.
1923 */
1924 if (bdev != bdev->bd_contains &&
1925 (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET)) {
1926 struct hd_struct *p = bdev->bd_part;
1927
1928 bio->bi_iter.bi_sector += p->start_sect;
1929 bio->bi_bdev = bdev->bd_contains;
1930
1931 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1932 bdev->bd_dev,
1933 bio->bi_iter.bi_sector - p->start_sect);
1934 }
1935}
1936
1937static void handle_bad_sector(struct bio *bio)
1938{
1939 char b[BDEVNAME_SIZE];
1940
1941 printk(KERN_INFO "attempt to access beyond end of device\n");
1942 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
1943 bdevname(bio->bi_bdev, b),
1944 bio->bi_opf,
1945 (unsigned long long)bio_end_sector(bio),
1946 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1947}
1948
1949#ifdef CONFIG_FAIL_MAKE_REQUEST
1950
1951static DECLARE_FAULT_ATTR(fail_make_request);
1952
1953static int __init setup_fail_make_request(char *str)
1954{
1955 return setup_fault_attr(&fail_make_request, str);
1956}
1957__setup("fail_make_request=", setup_fail_make_request);
1958
1959static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1960{
1961 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1962}
1963
1964static int __init fail_make_request_debugfs(void)
1965{
1966 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1967 NULL, &fail_make_request);
1968
1969 return PTR_ERR_OR_ZERO(dir);
1970}
1971
1972late_initcall(fail_make_request_debugfs);
1973
1974#else /* CONFIG_FAIL_MAKE_REQUEST */
1975
1976static inline bool should_fail_request(struct hd_struct *part,
1977 unsigned int bytes)
1978{
1979 return false;
1980}
1981
1982#endif /* CONFIG_FAIL_MAKE_REQUEST */
1983
1984/*
1985 * Check whether this bio extends beyond the end of the device.
1986 */
1987static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1988{
1989 sector_t maxsector;
1990
1991 if (!nr_sectors)
1992 return 0;
1993
1994 /* Test device or partition size, when known. */
1995 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1996 if (maxsector) {
1997 sector_t sector = bio->bi_iter.bi_sector;
1998
1999 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
2000 /*
2001 * This may well happen - the kernel calls bread()
2002 * without checking the size of the device, e.g., when
2003 * mounting a device.
2004 */
2005 handle_bad_sector(bio);
2006 return 1;
2007 }
2008 }
2009
2010 return 0;
2011}
2012
2013static noinline_for_stack bool
2014generic_make_request_checks(struct bio *bio)
2015{
2016 struct request_queue *q;
2017 int nr_sectors = bio_sectors(bio);
2018 blk_status_t status = BLK_STS_IOERR;
2019 char b[BDEVNAME_SIZE];
2020 struct hd_struct *part;
2021
2022 might_sleep();
2023
2024 if (bio_check_eod(bio, nr_sectors))
2025 goto end_io;
2026
2027 q = bdev_get_queue(bio->bi_bdev);
2028 if (unlikely(!q)) {
2029 printk(KERN_ERR
2030 "generic_make_request: Trying to access "
2031 "nonexistent block-device %s (%Lu)\n",
2032 bdevname(bio->bi_bdev, b),
2033 (long long) bio->bi_iter.bi_sector);
2034 goto end_io;
2035 }
2036
2037 /*
2038 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
2039 * if queue is not a request based queue.
2040 */
2041
2042 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_rq_based(q))
2043 goto not_supported;
2044
2045 part = bio->bi_bdev->bd_part;
2046 if (should_fail_request(part, bio->bi_iter.bi_size) ||
2047 should_fail_request(&part_to_disk(part)->part0,
2048 bio->bi_iter.bi_size))
2049 goto end_io;
2050
2051 /*
2052 * If this device has partitions, remap block n
2053 * of partition p to block n+start(p) of the disk.
2054 */
2055 blk_partition_remap(bio);
2056
2057 if (bio_check_eod(bio, nr_sectors))
2058 goto end_io;
2059
2060 /*
2061 * Filter flush bio's early so that make_request based
2062 * drivers without flush support don't have to worry
2063 * about them.
2064 */
2065 if (op_is_flush(bio->bi_opf) &&
2066 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
2067 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
2068 if (!nr_sectors) {
2069 status = BLK_STS_OK;
2070 goto end_io;
2071 }
2072 }
2073
2074 switch (bio_op(bio)) {
2075 case REQ_OP_DISCARD:
2076 if (!blk_queue_discard(q))
2077 goto not_supported;
2078 break;
2079 case REQ_OP_SECURE_ERASE:
2080 if (!blk_queue_secure_erase(q))
2081 goto not_supported;
2082 break;
2083 case REQ_OP_WRITE_SAME:
2084 if (!bdev_write_same(bio->bi_bdev))
2085 goto not_supported;
2086 break;
2087 case REQ_OP_ZONE_REPORT:
2088 case REQ_OP_ZONE_RESET:
2089 if (!bdev_is_zoned(bio->bi_bdev))
2090 goto not_supported;
2091 break;
2092 case REQ_OP_WRITE_ZEROES:
2093 if (!bdev_write_zeroes_sectors(bio->bi_bdev))
2094 goto not_supported;
2095 break;
2096 default:
2097 break;
2098 }
2099
2100 /*
2101 * Various block parts want %current->io_context and lazy ioc
2102 * allocation ends up trading a lot of pain for a small amount of
2103 * memory. Just allocate it upfront. This may fail and block
2104 * layer knows how to live with it.
2105 */
2106 create_io_context(GFP_ATOMIC, q->node);
2107
2108 if (!blkcg_bio_issue_check(q, bio))
2109 return false;
2110
2111 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
2112 trace_block_bio_queue(q, bio);
2113 /* Now that enqueuing has been traced, we need to trace
2114 * completion as well.
2115 */
2116 bio_set_flag(bio, BIO_TRACE_COMPLETION);
2117 }
2118 return true;
2119
2120not_supported:
2121 status = BLK_STS_NOTSUPP;
2122end_io:
2123 bio->bi_status = status;
2124 bio_endio(bio);
2125 return false;
2126}
2127
2128/**
2129 * generic_make_request - hand a buffer to its device driver for I/O
2130 * @bio: The bio describing the location in memory and on the device.
2131 *
2132 * generic_make_request() is used to make I/O requests of block
2133 * devices. It is passed a &struct bio, which describes the I/O that needs
2134 * to be done.
2135 *
2136 * generic_make_request() does not return any status. The
2137 * success/failure status of the request, along with notification of
2138 * completion, is delivered asynchronously through the bio->bi_end_io
2139 * function described (one day) else where.
2140 *
2141 * The caller of generic_make_request must make sure that bi_io_vec
2142 * are set to describe the memory buffer, and that bi_dev and bi_sector are
2143 * set to describe the device address, and the
2144 * bi_end_io and optionally bi_private are set to describe how
2145 * completion notification should be signaled.
2146 *
2147 * generic_make_request and the drivers it calls may use bi_next if this
2148 * bio happens to be merged with someone else, and may resubmit the bio to
2149 * a lower device by calling into generic_make_request recursively, which
2150 * means the bio should NOT be touched after the call to ->make_request_fn.
2151 */
2152blk_qc_t generic_make_request(struct bio *bio)
2153{
2154 /*
2155 * bio_list_on_stack[0] contains bios submitted by the current
2156 * make_request_fn.
2157 * bio_list_on_stack[1] contains bios that were submitted before
2158 * the current make_request_fn, but that haven't been processed
2159 * yet.
2160 */
2161 struct bio_list bio_list_on_stack[2];
2162 blk_qc_t ret = BLK_QC_T_NONE;
2163
2164 if (!generic_make_request_checks(bio))
2165 goto out;
2166
2167 /*
2168 * We only want one ->make_request_fn to be active at a time, else
2169 * stack usage with stacked devices could be a problem. So use
2170 * current->bio_list to keep a list of requests submited by a
2171 * make_request_fn function. current->bio_list is also used as a
2172 * flag to say if generic_make_request is currently active in this
2173 * task or not. If it is NULL, then no make_request is active. If
2174 * it is non-NULL, then a make_request is active, and new requests
2175 * should be added at the tail
2176 */
2177 if (current->bio_list) {
2178 bio_list_add(&current->bio_list[0], bio);
2179 goto out;
2180 }
2181
2182 /* following loop may be a bit non-obvious, and so deserves some
2183 * explanation.
2184 * Before entering the loop, bio->bi_next is NULL (as all callers
2185 * ensure that) so we have a list with a single bio.
2186 * We pretend that we have just taken it off a longer list, so
2187 * we assign bio_list to a pointer to the bio_list_on_stack,
2188 * thus initialising the bio_list of new bios to be
2189 * added. ->make_request() may indeed add some more bios
2190 * through a recursive call to generic_make_request. If it
2191 * did, we find a non-NULL value in bio_list and re-enter the loop
2192 * from the top. In this case we really did just take the bio
2193 * of the top of the list (no pretending) and so remove it from
2194 * bio_list, and call into ->make_request() again.
2195 */
2196 BUG_ON(bio->bi_next);
2197 bio_list_init(&bio_list_on_stack[0]);
2198 current->bio_list = bio_list_on_stack;
2199 do {
2200 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
2201
2202 if (likely(blk_queue_enter(q, bio->bi_opf & REQ_NOWAIT) == 0)) {
2203 struct bio_list lower, same;
2204
2205 /* Create a fresh bio_list for all subordinate requests */
2206 bio_list_on_stack[1] = bio_list_on_stack[0];
2207 bio_list_init(&bio_list_on_stack[0]);
2208 ret = q->make_request_fn(q, bio);
2209
2210 blk_queue_exit(q);
2211
2212 /* sort new bios into those for a lower level
2213 * and those for the same level
2214 */
2215 bio_list_init(&lower);
2216 bio_list_init(&same);
2217 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
2218 if (q == bdev_get_queue(bio->bi_bdev))
2219 bio_list_add(&same, bio);
2220 else
2221 bio_list_add(&lower, bio);
2222 /* now assemble so we handle the lowest level first */
2223 bio_list_merge(&bio_list_on_stack[0], &lower);
2224 bio_list_merge(&bio_list_on_stack[0], &same);
2225 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
2226 } else {
2227 if (unlikely(!blk_queue_dying(q) &&
2228 (bio->bi_opf & REQ_NOWAIT)))
2229 bio_wouldblock_error(bio);
2230 else
2231 bio_io_error(bio);
2232 }
2233 bio = bio_list_pop(&bio_list_on_stack[0]);
2234 } while (bio);
2235 current->bio_list = NULL; /* deactivate */
2236
2237out:
2238 return ret;
2239}
2240EXPORT_SYMBOL(generic_make_request);
2241
2242/**
2243 * submit_bio - submit a bio to the block device layer for I/O
2244 * @bio: The &struct bio which describes the I/O
2245 *
2246 * submit_bio() is very similar in purpose to generic_make_request(), and
2247 * uses that function to do most of the work. Both are fairly rough
2248 * interfaces; @bio must be presetup and ready for I/O.
2249 *
2250 */
2251blk_qc_t submit_bio(struct bio *bio)
2252{
2253 /*
2254 * If it's a regular read/write or a barrier with data attached,
2255 * go through the normal accounting stuff before submission.
2256 */
2257 if (bio_has_data(bio)) {
2258 unsigned int count;
2259
2260 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
2261 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
2262 else
2263 count = bio_sectors(bio);
2264
2265 if (op_is_write(bio_op(bio))) {
2266 count_vm_events(PGPGOUT, count);
2267 } else {
2268 task_io_account_read(bio->bi_iter.bi_size);
2269 count_vm_events(PGPGIN, count);
2270 }
2271
2272 if (unlikely(block_dump)) {
2273 char b[BDEVNAME_SIZE];
2274 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
2275 current->comm, task_pid_nr(current),
2276 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
2277 (unsigned long long)bio->bi_iter.bi_sector,
2278 bdevname(bio->bi_bdev, b),
2279 count);
2280 }
2281 }
2282
2283 return generic_make_request(bio);
2284}
2285EXPORT_SYMBOL(submit_bio);
2286
2287/**
2288 * blk_cloned_rq_check_limits - Helper function to check a cloned request
2289 * for new the queue limits
2290 * @q: the queue
2291 * @rq: the request being checked
2292 *
2293 * Description:
2294 * @rq may have been made based on weaker limitations of upper-level queues
2295 * in request stacking drivers, and it may violate the limitation of @q.
2296 * Since the block layer and the underlying device driver trust @rq
2297 * after it is inserted to @q, it should be checked against @q before
2298 * the insertion using this generic function.
2299 *
2300 * Request stacking drivers like request-based dm may change the queue
2301 * limits when retrying requests on other queues. Those requests need
2302 * to be checked against the new queue limits again during dispatch.
2303 */
2304static int blk_cloned_rq_check_limits(struct request_queue *q,
2305 struct request *rq)
2306{
2307 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
2308 printk(KERN_ERR "%s: over max size limit.\n", __func__);
2309 return -EIO;
2310 }
2311
2312 /*
2313 * queue's settings related to segment counting like q->bounce_pfn
2314 * may differ from that of other stacking queues.
2315 * Recalculate it to check the request correctly on this queue's
2316 * limitation.
2317 */
2318 blk_recalc_rq_segments(rq);
2319 if (rq->nr_phys_segments > queue_max_segments(q)) {
2320 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
2321 return -EIO;
2322 }
2323
2324 return 0;
2325}
2326
2327/**
2328 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
2329 * @q: the queue to submit the request
2330 * @rq: the request being queued
2331 */
2332blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
2333{
2334 unsigned long flags;
2335 int where = ELEVATOR_INSERT_BACK;
2336
2337 if (blk_cloned_rq_check_limits(q, rq))
2338 return BLK_STS_IOERR;
2339
2340 if (rq->rq_disk &&
2341 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
2342 return BLK_STS_IOERR;
2343
2344 if (q->mq_ops) {
2345 if (blk_queue_io_stat(q))
2346 blk_account_io_start(rq, true);
2347 blk_mq_sched_insert_request(rq, false, true, false, false);
2348 return BLK_STS_OK;
2349 }
2350
2351 spin_lock_irqsave(q->queue_lock, flags);
2352 if (unlikely(blk_queue_dying(q))) {
2353 spin_unlock_irqrestore(q->queue_lock, flags);
2354 return BLK_STS_IOERR;
2355 }
2356
2357 /*
2358 * Submitting request must be dequeued before calling this function
2359 * because it will be linked to another request_queue
2360 */
2361 BUG_ON(blk_queued_rq(rq));
2362
2363 if (op_is_flush(rq->cmd_flags))
2364 where = ELEVATOR_INSERT_FLUSH;
2365
2366 add_acct_request(q, rq, where);
2367 if (where == ELEVATOR_INSERT_FLUSH)
2368 __blk_run_queue(q);
2369 spin_unlock_irqrestore(q->queue_lock, flags);
2370
2371 return BLK_STS_OK;
2372}
2373EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
2374
2375/**
2376 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
2377 * @rq: request to examine
2378 *
2379 * Description:
2380 * A request could be merge of IOs which require different failure
2381 * handling. This function determines the number of bytes which
2382 * can be failed from the beginning of the request without
2383 * crossing into area which need to be retried further.
2384 *
2385 * Return:
2386 * The number of bytes to fail.
2387 */
2388unsigned int blk_rq_err_bytes(const struct request *rq)
2389{
2390 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
2391 unsigned int bytes = 0;
2392 struct bio *bio;
2393
2394 if (!(rq->rq_flags & RQF_MIXED_MERGE))
2395 return blk_rq_bytes(rq);
2396
2397 /*
2398 * Currently the only 'mixing' which can happen is between
2399 * different fastfail types. We can safely fail portions
2400 * which have all the failfast bits that the first one has -
2401 * the ones which are at least as eager to fail as the first
2402 * one.
2403 */
2404 for (bio = rq->bio; bio; bio = bio->bi_next) {
2405 if ((bio->bi_opf & ff) != ff)
2406 break;
2407 bytes += bio->bi_iter.bi_size;
2408 }
2409
2410 /* this could lead to infinite loop */
2411 BUG_ON(blk_rq_bytes(rq) && !bytes);
2412 return bytes;
2413}
2414EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2415
2416void blk_account_io_completion(struct request *req, unsigned int bytes)
2417{
2418 if (blk_do_io_stat(req)) {
2419 const int rw = rq_data_dir(req);
2420 struct hd_struct *part;
2421 int cpu;
2422
2423 cpu = part_stat_lock();
2424 part = req->part;
2425 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2426 part_stat_unlock();
2427 }
2428}
2429
2430void blk_account_io_done(struct request *req)
2431{
2432 /*
2433 * Account IO completion. flush_rq isn't accounted as a
2434 * normal IO on queueing nor completion. Accounting the
2435 * containing request is enough.
2436 */
2437 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
2438 unsigned long duration = jiffies - req->start_time;
2439 const int rw = rq_data_dir(req);
2440 struct hd_struct *part;
2441 int cpu;
2442
2443 cpu = part_stat_lock();
2444 part = req->part;
2445
2446 part_stat_inc(cpu, part, ios[rw]);
2447 part_stat_add(cpu, part, ticks[rw], duration);
2448 part_round_stats(req->q, cpu, part);
2449 part_dec_in_flight(req->q, part, rw);
2450
2451 hd_struct_put(part);
2452 part_stat_unlock();
2453 }
2454}
2455
2456#ifdef CONFIG_PM
2457/*
2458 * Don't process normal requests when queue is suspended
2459 * or in the process of suspending/resuming
2460 */
2461static struct request *blk_pm_peek_request(struct request_queue *q,
2462 struct request *rq)
2463{
2464 if (q->dev && (q->rpm_status == RPM_SUSPENDED ||
2465 (q->rpm_status != RPM_ACTIVE && !(rq->rq_flags & RQF_PM))))
2466 return NULL;
2467 else
2468 return rq;
2469}
2470#else
2471static inline struct request *blk_pm_peek_request(struct request_queue *q,
2472 struct request *rq)
2473{
2474 return rq;
2475}
2476#endif
2477
2478void blk_account_io_start(struct request *rq, bool new_io)
2479{
2480 struct hd_struct *part;
2481 int rw = rq_data_dir(rq);
2482 int cpu;
2483
2484 if (!blk_do_io_stat(rq))
2485 return;
2486
2487 cpu = part_stat_lock();
2488
2489 if (!new_io) {
2490 part = rq->part;
2491 part_stat_inc(cpu, part, merges[rw]);
2492 } else {
2493 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
2494 if (!hd_struct_try_get(part)) {
2495 /*
2496 * The partition is already being removed,
2497 * the request will be accounted on the disk only
2498 *
2499 * We take a reference on disk->part0 although that
2500 * partition will never be deleted, so we can treat
2501 * it as any other partition.
2502 */
2503 part = &rq->rq_disk->part0;
2504 hd_struct_get(part);
2505 }
2506 part_round_stats(rq->q, cpu, part);
2507 part_inc_in_flight(rq->q, part, rw);
2508 rq->part = part;
2509 }
2510
2511 part_stat_unlock();
2512}
2513
2514/**
2515 * blk_peek_request - peek at the top of a request queue
2516 * @q: request queue to peek at
2517 *
2518 * Description:
2519 * Return the request at the top of @q. The returned request
2520 * should be started using blk_start_request() before LLD starts
2521 * processing it.
2522 *
2523 * Return:
2524 * Pointer to the request at the top of @q if available. Null
2525 * otherwise.
2526 */
2527struct request *blk_peek_request(struct request_queue *q)
2528{
2529 struct request *rq;
2530 int ret;
2531
2532 lockdep_assert_held(q->queue_lock);
2533 WARN_ON_ONCE(q->mq_ops);
2534
2535 while ((rq = __elv_next_request(q)) != NULL) {
2536
2537 rq = blk_pm_peek_request(q, rq);
2538 if (!rq)
2539 break;
2540
2541 if (!(rq->rq_flags & RQF_STARTED)) {
2542 /*
2543 * This is the first time the device driver
2544 * sees this request (possibly after
2545 * requeueing). Notify IO scheduler.
2546 */
2547 if (rq->rq_flags & RQF_SORTED)
2548 elv_activate_rq(q, rq);
2549
2550 /*
2551 * just mark as started even if we don't start
2552 * it, a request that has been delayed should
2553 * not be passed by new incoming requests
2554 */
2555 rq->rq_flags |= RQF_STARTED;
2556 trace_block_rq_issue(q, rq);
2557 }
2558
2559 if (!q->boundary_rq || q->boundary_rq == rq) {
2560 q->end_sector = rq_end_sector(rq);
2561 q->boundary_rq = NULL;
2562 }
2563
2564 if (rq->rq_flags & RQF_DONTPREP)
2565 break;
2566
2567 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2568 /*
2569 * make sure space for the drain appears we
2570 * know we can do this because max_hw_segments
2571 * has been adjusted to be one fewer than the
2572 * device can handle
2573 */
2574 rq->nr_phys_segments++;
2575 }
2576
2577 if (!q->prep_rq_fn)
2578 break;
2579
2580 ret = q->prep_rq_fn(q, rq);
2581 if (ret == BLKPREP_OK) {
2582 break;
2583 } else if (ret == BLKPREP_DEFER) {
2584 /*
2585 * the request may have been (partially) prepped.
2586 * we need to keep this request in the front to
2587 * avoid resource deadlock. RQF_STARTED will
2588 * prevent other fs requests from passing this one.
2589 */
2590 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2591 !(rq->rq_flags & RQF_DONTPREP)) {
2592 /*
2593 * remove the space for the drain we added
2594 * so that we don't add it again
2595 */
2596 --rq->nr_phys_segments;
2597 }
2598
2599 rq = NULL;
2600 break;
2601 } else if (ret == BLKPREP_KILL || ret == BLKPREP_INVALID) {
2602 rq->rq_flags |= RQF_QUIET;
2603 /*
2604 * Mark this request as started so we don't trigger
2605 * any debug logic in the end I/O path.
2606 */
2607 blk_start_request(rq);
2608 __blk_end_request_all(rq, ret == BLKPREP_INVALID ?
2609 BLK_STS_TARGET : BLK_STS_IOERR);
2610 } else {
2611 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2612 break;
2613 }
2614 }
2615
2616 return rq;
2617}
2618EXPORT_SYMBOL(blk_peek_request);
2619
2620void blk_dequeue_request(struct request *rq)
2621{
2622 struct request_queue *q = rq->q;
2623
2624 BUG_ON(list_empty(&rq->queuelist));
2625 BUG_ON(ELV_ON_HASH(rq));
2626
2627 list_del_init(&rq->queuelist);
2628
2629 /*
2630 * the time frame between a request being removed from the lists
2631 * and to it is freed is accounted as io that is in progress at
2632 * the driver side.
2633 */
2634 if (blk_account_rq(rq)) {
2635 q->in_flight[rq_is_sync(rq)]++;
2636 set_io_start_time_ns(rq);
2637 }
2638}
2639
2640/**
2641 * blk_start_request - start request processing on the driver
2642 * @req: request to dequeue
2643 *
2644 * Description:
2645 * Dequeue @req and start timeout timer on it. This hands off the
2646 * request to the driver.
2647 *
2648 * Block internal functions which don't want to start timer should
2649 * call blk_dequeue_request().
2650 */
2651void blk_start_request(struct request *req)
2652{
2653 lockdep_assert_held(req->q->queue_lock);
2654 WARN_ON_ONCE(req->q->mq_ops);
2655
2656 blk_dequeue_request(req);
2657
2658 if (test_bit(QUEUE_FLAG_STATS, &req->q->queue_flags)) {
2659 blk_stat_set_issue(&req->issue_stat, blk_rq_sectors(req));
2660 req->rq_flags |= RQF_STATS;
2661 wbt_issue(req->q->rq_wb, &req->issue_stat);
2662 }
2663
2664 BUG_ON(test_bit(REQ_ATOM_COMPLETE, &req->atomic_flags));
2665 blk_add_timer(req);
2666}
2667EXPORT_SYMBOL(blk_start_request);
2668
2669/**
2670 * blk_fetch_request - fetch a request from a request queue
2671 * @q: request queue to fetch a request from
2672 *
2673 * Description:
2674 * Return the request at the top of @q. The request is started on
2675 * return and LLD can start processing it immediately.
2676 *
2677 * Return:
2678 * Pointer to the request at the top of @q if available. Null
2679 * otherwise.
2680 */
2681struct request *blk_fetch_request(struct request_queue *q)
2682{
2683 struct request *rq;
2684
2685 lockdep_assert_held(q->queue_lock);
2686 WARN_ON_ONCE(q->mq_ops);
2687
2688 rq = blk_peek_request(q);
2689 if (rq)
2690 blk_start_request(rq);
2691 return rq;
2692}
2693EXPORT_SYMBOL(blk_fetch_request);
2694
2695/**
2696 * blk_update_request - Special helper function for request stacking drivers
2697 * @req: the request being processed
2698 * @error: block status code
2699 * @nr_bytes: number of bytes to complete @req
2700 *
2701 * Description:
2702 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2703 * the request structure even if @req doesn't have leftover.
2704 * If @req has leftover, sets it up for the next range of segments.
2705 *
2706 * This special helper function is only for request stacking drivers
2707 * (e.g. request-based dm) so that they can handle partial completion.
2708 * Actual device drivers should use blk_end_request instead.
2709 *
2710 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2711 * %false return from this function.
2712 *
2713 * Return:
2714 * %false - this request doesn't have any more data
2715 * %true - this request has more data
2716 **/
2717bool blk_update_request(struct request *req, blk_status_t error,
2718 unsigned int nr_bytes)
2719{
2720 int total_bytes;
2721
2722 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
2723
2724 if (!req->bio)
2725 return false;
2726
2727 if (unlikely(error && !blk_rq_is_passthrough(req) &&
2728 !(req->rq_flags & RQF_QUIET)))
2729 print_req_error(req, error);
2730
2731 blk_account_io_completion(req, nr_bytes);
2732
2733 total_bytes = 0;
2734 while (req->bio) {
2735 struct bio *bio = req->bio;
2736 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
2737
2738 if (bio_bytes == bio->bi_iter.bi_size)
2739 req->bio = bio->bi_next;
2740
2741 /* Completion has already been traced */
2742 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
2743 req_bio_endio(req, bio, bio_bytes, error);
2744
2745 total_bytes += bio_bytes;
2746 nr_bytes -= bio_bytes;
2747
2748 if (!nr_bytes)
2749 break;
2750 }
2751
2752 /*
2753 * completely done
2754 */
2755 if (!req->bio) {
2756 /*
2757 * Reset counters so that the request stacking driver
2758 * can find how many bytes remain in the request
2759 * later.
2760 */
2761 req->__data_len = 0;
2762 return false;
2763 }
2764
2765 req->__data_len -= total_bytes;
2766
2767 /* update sector only for requests with clear definition of sector */
2768 if (!blk_rq_is_passthrough(req))
2769 req->__sector += total_bytes >> 9;
2770
2771 /* mixed attributes always follow the first bio */
2772 if (req->rq_flags & RQF_MIXED_MERGE) {
2773 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2774 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
2775 }
2776
2777 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
2778 /*
2779 * If total number of sectors is less than the first segment
2780 * size, something has gone terribly wrong.
2781 */
2782 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2783 blk_dump_rq_flags(req, "request botched");
2784 req->__data_len = blk_rq_cur_bytes(req);
2785 }
2786
2787 /* recalculate the number of segments */
2788 blk_recalc_rq_segments(req);
2789 }
2790
2791 return true;
2792}
2793EXPORT_SYMBOL_GPL(blk_update_request);
2794
2795static bool blk_update_bidi_request(struct request *rq, blk_status_t error,
2796 unsigned int nr_bytes,
2797 unsigned int bidi_bytes)
2798{
2799 if (blk_update_request(rq, error, nr_bytes))
2800 return true;
2801
2802 /* Bidi request must be completed as a whole */
2803 if (unlikely(blk_bidi_rq(rq)) &&
2804 blk_update_request(rq->next_rq, error, bidi_bytes))
2805 return true;
2806
2807 if (blk_queue_add_random(rq->q))
2808 add_disk_randomness(rq->rq_disk);
2809
2810 return false;
2811}
2812
2813/**
2814 * blk_unprep_request - unprepare a request
2815 * @req: the request
2816 *
2817 * This function makes a request ready for complete resubmission (or
2818 * completion). It happens only after all error handling is complete,
2819 * so represents the appropriate moment to deallocate any resources
2820 * that were allocated to the request in the prep_rq_fn. The queue
2821 * lock is held when calling this.
2822 */
2823void blk_unprep_request(struct request *req)
2824{
2825 struct request_queue *q = req->q;
2826
2827 req->rq_flags &= ~RQF_DONTPREP;
2828 if (q->unprep_rq_fn)
2829 q->unprep_rq_fn(q, req);
2830}
2831EXPORT_SYMBOL_GPL(blk_unprep_request);
2832
2833void blk_finish_request(struct request *req, blk_status_t error)
2834{
2835 struct request_queue *q = req->q;
2836
2837 lockdep_assert_held(req->q->queue_lock);
2838 WARN_ON_ONCE(q->mq_ops);
2839
2840 if (req->rq_flags & RQF_STATS)
2841 blk_stat_add(req);
2842
2843 if (req->rq_flags & RQF_QUEUED)
2844 blk_queue_end_tag(q, req);
2845
2846 BUG_ON(blk_queued_rq(req));
2847
2848 if (unlikely(laptop_mode) && !blk_rq_is_passthrough(req))
2849 laptop_io_completion(req->q->backing_dev_info);
2850
2851 blk_delete_timer(req);
2852
2853 if (req->rq_flags & RQF_DONTPREP)
2854 blk_unprep_request(req);
2855
2856 blk_account_io_done(req);
2857
2858 if (req->end_io) {
2859 wbt_done(req->q->rq_wb, &req->issue_stat);
2860 req->end_io(req, error);
2861 } else {
2862 if (blk_bidi_rq(req))
2863 __blk_put_request(req->next_rq->q, req->next_rq);
2864
2865 __blk_put_request(q, req);
2866 }
2867}
2868EXPORT_SYMBOL(blk_finish_request);
2869
2870/**
2871 * blk_end_bidi_request - Complete a bidi request
2872 * @rq: the request to complete
2873 * @error: block status code
2874 * @nr_bytes: number of bytes to complete @rq
2875 * @bidi_bytes: number of bytes to complete @rq->next_rq
2876 *
2877 * Description:
2878 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2879 * Drivers that supports bidi can safely call this member for any
2880 * type of request, bidi or uni. In the later case @bidi_bytes is
2881 * just ignored.
2882 *
2883 * Return:
2884 * %false - we are done with this request
2885 * %true - still buffers pending for this request
2886 **/
2887static bool blk_end_bidi_request(struct request *rq, blk_status_t error,
2888 unsigned int nr_bytes, unsigned int bidi_bytes)
2889{
2890 struct request_queue *q = rq->q;
2891 unsigned long flags;
2892
2893 WARN_ON_ONCE(q->mq_ops);
2894
2895 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2896 return true;
2897
2898 spin_lock_irqsave(q->queue_lock, flags);
2899 blk_finish_request(rq, error);
2900 spin_unlock_irqrestore(q->queue_lock, flags);
2901
2902 return false;
2903}
2904
2905/**
2906 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2907 * @rq: the request to complete
2908 * @error: block status code
2909 * @nr_bytes: number of bytes to complete @rq
2910 * @bidi_bytes: number of bytes to complete @rq->next_rq
2911 *
2912 * Description:
2913 * Identical to blk_end_bidi_request() except that queue lock is
2914 * assumed to be locked on entry and remains so on return.
2915 *
2916 * Return:
2917 * %false - we are done with this request
2918 * %true - still buffers pending for this request
2919 **/
2920static bool __blk_end_bidi_request(struct request *rq, blk_status_t error,
2921 unsigned int nr_bytes, unsigned int bidi_bytes)
2922{
2923 lockdep_assert_held(rq->q->queue_lock);
2924 WARN_ON_ONCE(rq->q->mq_ops);
2925
2926 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2927 return true;
2928
2929 blk_finish_request(rq, error);
2930
2931 return false;
2932}
2933
2934/**
2935 * blk_end_request - Helper function for drivers to complete the request.
2936 * @rq: the request being processed
2937 * @error: block status code
2938 * @nr_bytes: number of bytes to complete
2939 *
2940 * Description:
2941 * Ends I/O on a number of bytes attached to @rq.
2942 * If @rq has leftover, sets it up for the next range of segments.
2943 *
2944 * Return:
2945 * %false - we are done with this request
2946 * %true - still buffers pending for this request
2947 **/
2948bool blk_end_request(struct request *rq, blk_status_t error,
2949 unsigned int nr_bytes)
2950{
2951 WARN_ON_ONCE(rq->q->mq_ops);
2952 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2953}
2954EXPORT_SYMBOL(blk_end_request);
2955
2956/**
2957 * blk_end_request_all - Helper function for drives to finish the request.
2958 * @rq: the request to finish
2959 * @error: block status code
2960 *
2961 * Description:
2962 * Completely finish @rq.
2963 */
2964void blk_end_request_all(struct request *rq, blk_status_t error)
2965{
2966 bool pending;
2967 unsigned int bidi_bytes = 0;
2968
2969 if (unlikely(blk_bidi_rq(rq)))
2970 bidi_bytes = blk_rq_bytes(rq->next_rq);
2971
2972 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2973 BUG_ON(pending);
2974}
2975EXPORT_SYMBOL(blk_end_request_all);
2976
2977/**
2978 * __blk_end_request - Helper function for drivers to complete the request.
2979 * @rq: the request being processed
2980 * @error: block status code
2981 * @nr_bytes: number of bytes to complete
2982 *
2983 * Description:
2984 * Must be called with queue lock held unlike blk_end_request().
2985 *
2986 * Return:
2987 * %false - we are done with this request
2988 * %true - still buffers pending for this request
2989 **/
2990bool __blk_end_request(struct request *rq, blk_status_t error,
2991 unsigned int nr_bytes)
2992{
2993 lockdep_assert_held(rq->q->queue_lock);
2994 WARN_ON_ONCE(rq->q->mq_ops);
2995
2996 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2997}
2998EXPORT_SYMBOL(__blk_end_request);
2999
3000/**
3001 * __blk_end_request_all - Helper function for drives to finish the request.
3002 * @rq: the request to finish
3003 * @error: block status code
3004 *
3005 * Description:
3006 * Completely finish @rq. Must be called with queue lock held.
3007 */
3008void __blk_end_request_all(struct request *rq, blk_status_t error)
3009{
3010 bool pending;
3011 unsigned int bidi_bytes = 0;
3012
3013 lockdep_assert_held(rq->q->queue_lock);
3014 WARN_ON_ONCE(rq->q->mq_ops);
3015
3016 if (unlikely(blk_bidi_rq(rq)))
3017 bidi_bytes = blk_rq_bytes(rq->next_rq);
3018
3019 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
3020 BUG_ON(pending);
3021}
3022EXPORT_SYMBOL(__blk_end_request_all);
3023
3024/**
3025 * __blk_end_request_cur - Helper function to finish the current request chunk.
3026 * @rq: the request to finish the current chunk for
3027 * @error: block status code
3028 *
3029 * Description:
3030 * Complete the current consecutively mapped chunk from @rq. Must
3031 * be called with queue lock held.
3032 *
3033 * Return:
3034 * %false - we are done with this request
3035 * %true - still buffers pending for this request
3036 */
3037bool __blk_end_request_cur(struct request *rq, blk_status_t error)
3038{
3039 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
3040}
3041EXPORT_SYMBOL(__blk_end_request_cur);
3042
3043void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
3044 struct bio *bio)
3045{
3046 if (bio_has_data(bio))
3047 rq->nr_phys_segments = bio_phys_segments(q, bio);
3048
3049 rq->__data_len = bio->bi_iter.bi_size;
3050 rq->bio = rq->biotail = bio;
3051
3052 if (bio->bi_bdev)
3053 rq->rq_disk = bio->bi_bdev->bd_disk;
3054}
3055
3056#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
3057/**
3058 * rq_flush_dcache_pages - Helper function to flush all pages in a request
3059 * @rq: the request to be flushed
3060 *
3061 * Description:
3062 * Flush all pages in @rq.
3063 */
3064void rq_flush_dcache_pages(struct request *rq)
3065{
3066 struct req_iterator iter;
3067 struct bio_vec bvec;
3068
3069 rq_for_each_segment(bvec, rq, iter)
3070 flush_dcache_page(bvec.bv_page);
3071}
3072EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
3073#endif
3074
3075/**
3076 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
3077 * @q : the queue of the device being checked
3078 *
3079 * Description:
3080 * Check if underlying low-level drivers of a device are busy.
3081 * If the drivers want to export their busy state, they must set own
3082 * exporting function using blk_queue_lld_busy() first.
3083 *
3084 * Basically, this function is used only by request stacking drivers
3085 * to stop dispatching requests to underlying devices when underlying
3086 * devices are busy. This behavior helps more I/O merging on the queue
3087 * of the request stacking driver and prevents I/O throughput regression
3088 * on burst I/O load.
3089 *
3090 * Return:
3091 * 0 - Not busy (The request stacking driver should dispatch request)
3092 * 1 - Busy (The request stacking driver should stop dispatching request)
3093 */
3094int blk_lld_busy(struct request_queue *q)
3095{
3096 if (q->lld_busy_fn)
3097 return q->lld_busy_fn(q);
3098
3099 return 0;
3100}
3101EXPORT_SYMBOL_GPL(blk_lld_busy);
3102
3103/**
3104 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
3105 * @rq: the clone request to be cleaned up
3106 *
3107 * Description:
3108 * Free all bios in @rq for a cloned request.
3109 */
3110void blk_rq_unprep_clone(struct request *rq)
3111{
3112 struct bio *bio;
3113
3114 while ((bio = rq->bio) != NULL) {
3115 rq->bio = bio->bi_next;
3116
3117 bio_put(bio);
3118 }
3119}
3120EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
3121
3122/*
3123 * Copy attributes of the original request to the clone request.
3124 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
3125 */
3126static void __blk_rq_prep_clone(struct request *dst, struct request *src)
3127{
3128 dst->cpu = src->cpu;
3129 dst->__sector = blk_rq_pos(src);
3130 dst->__data_len = blk_rq_bytes(src);
3131 dst->nr_phys_segments = src->nr_phys_segments;
3132 dst->ioprio = src->ioprio;
3133 dst->extra_len = src->extra_len;
3134}
3135
3136/**
3137 * blk_rq_prep_clone - Helper function to setup clone request
3138 * @rq: the request to be setup
3139 * @rq_src: original request to be cloned
3140 * @bs: bio_set that bios for clone are allocated from
3141 * @gfp_mask: memory allocation mask for bio
3142 * @bio_ctr: setup function to be called for each clone bio.
3143 * Returns %0 for success, non %0 for failure.
3144 * @data: private data to be passed to @bio_ctr
3145 *
3146 * Description:
3147 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
3148 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
3149 * are not copied, and copying such parts is the caller's responsibility.
3150 * Also, pages which the original bios are pointing to are not copied
3151 * and the cloned bios just point same pages.
3152 * So cloned bios must be completed before original bios, which means
3153 * the caller must complete @rq before @rq_src.
3154 */
3155int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
3156 struct bio_set *bs, gfp_t gfp_mask,
3157 int (*bio_ctr)(struct bio *, struct bio *, void *),
3158 void *data)
3159{
3160 struct bio *bio, *bio_src;
3161
3162 if (!bs)
3163 bs = fs_bio_set;
3164
3165 __rq_for_each_bio(bio_src, rq_src) {
3166 bio = bio_clone_fast(bio_src, gfp_mask, bs);
3167 if (!bio)
3168 goto free_and_out;
3169
3170 if (bio_ctr && bio_ctr(bio, bio_src, data))
3171 goto free_and_out;
3172
3173 if (rq->bio) {
3174 rq->biotail->bi_next = bio;
3175 rq->biotail = bio;
3176 } else
3177 rq->bio = rq->biotail = bio;
3178 }
3179
3180 __blk_rq_prep_clone(rq, rq_src);
3181
3182 return 0;
3183
3184free_and_out:
3185 if (bio)
3186 bio_put(bio);
3187 blk_rq_unprep_clone(rq);
3188
3189 return -ENOMEM;
3190}
3191EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
3192
3193int kblockd_schedule_work(struct work_struct *work)
3194{
3195 return queue_work(kblockd_workqueue, work);
3196}
3197EXPORT_SYMBOL(kblockd_schedule_work);
3198
3199int kblockd_schedule_work_on(int cpu, struct work_struct *work)
3200{
3201 return queue_work_on(cpu, kblockd_workqueue, work);
3202}
3203EXPORT_SYMBOL(kblockd_schedule_work_on);
3204
3205int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
3206 unsigned long delay)
3207{
3208 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3209}
3210EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
3211
3212int kblockd_schedule_delayed_work(struct delayed_work *dwork,
3213 unsigned long delay)
3214{
3215 return queue_delayed_work(kblockd_workqueue, dwork, delay);
3216}
3217EXPORT_SYMBOL(kblockd_schedule_delayed_work);
3218
3219int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3220 unsigned long delay)
3221{
3222 return queue_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
3223}
3224EXPORT_SYMBOL(kblockd_schedule_delayed_work_on);
3225
3226/**
3227 * blk_start_plug - initialize blk_plug and track it inside the task_struct
3228 * @plug: The &struct blk_plug that needs to be initialized
3229 *
3230 * Description:
3231 * Tracking blk_plug inside the task_struct will help with auto-flushing the
3232 * pending I/O should the task end up blocking between blk_start_plug() and
3233 * blk_finish_plug(). This is important from a performance perspective, but
3234 * also ensures that we don't deadlock. For instance, if the task is blocking
3235 * for a memory allocation, memory reclaim could end up wanting to free a
3236 * page belonging to that request that is currently residing in our private
3237 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
3238 * this kind of deadlock.
3239 */
3240void blk_start_plug(struct blk_plug *plug)
3241{
3242 struct task_struct *tsk = current;
3243
3244 /*
3245 * If this is a nested plug, don't actually assign it.
3246 */
3247 if (tsk->plug)
3248 return;
3249
3250 INIT_LIST_HEAD(&plug->list);
3251 INIT_LIST_HEAD(&plug->mq_list);
3252 INIT_LIST_HEAD(&plug->cb_list);
3253 /*
3254 * Store ordering should not be needed here, since a potential
3255 * preempt will imply a full memory barrier
3256 */
3257 tsk->plug = plug;
3258}
3259EXPORT_SYMBOL(blk_start_plug);
3260
3261static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
3262{
3263 struct request *rqa = container_of(a, struct request, queuelist);
3264 struct request *rqb = container_of(b, struct request, queuelist);
3265
3266 return !(rqa->q < rqb->q ||
3267 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
3268}
3269
3270/*
3271 * If 'from_schedule' is true, then postpone the dispatch of requests
3272 * until a safe kblockd context. We due this to avoid accidental big
3273 * additional stack usage in driver dispatch, in places where the originally
3274 * plugger did not intend it.
3275 */
3276static void queue_unplugged(struct request_queue *q, unsigned int depth,
3277 bool from_schedule)
3278 __releases(q->queue_lock)
3279{
3280 lockdep_assert_held(q->queue_lock);
3281
3282 trace_block_unplug(q, depth, !from_schedule);
3283
3284 if (from_schedule)
3285 blk_run_queue_async(q);
3286 else
3287 __blk_run_queue(q);
3288 spin_unlock(q->queue_lock);
3289}
3290
3291static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
3292{
3293 LIST_HEAD(callbacks);
3294
3295 while (!list_empty(&plug->cb_list)) {
3296 list_splice_init(&plug->cb_list, &callbacks);
3297
3298 while (!list_empty(&callbacks)) {
3299 struct blk_plug_cb *cb = list_first_entry(&callbacks,
3300 struct blk_plug_cb,
3301 list);
3302 list_del(&cb->list);
3303 cb->callback(cb, from_schedule);
3304 }
3305 }
3306}
3307
3308struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
3309 int size)
3310{
3311 struct blk_plug *plug = current->plug;
3312 struct blk_plug_cb *cb;
3313
3314 if (!plug)
3315 return NULL;
3316
3317 list_for_each_entry(cb, &plug->cb_list, list)
3318 if (cb->callback == unplug && cb->data == data)
3319 return cb;
3320
3321 /* Not currently on the callback list */
3322 BUG_ON(size < sizeof(*cb));
3323 cb = kzalloc(size, GFP_ATOMIC);
3324 if (cb) {
3325 cb->data = data;
3326 cb->callback = unplug;
3327 list_add(&cb->list, &plug->cb_list);
3328 }
3329 return cb;
3330}
3331EXPORT_SYMBOL(blk_check_plugged);
3332
3333void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
3334{
3335 struct request_queue *q;
3336 unsigned long flags;
3337 struct request *rq;
3338 LIST_HEAD(list);
3339 unsigned int depth;
3340
3341 flush_plug_callbacks(plug, from_schedule);
3342
3343 if (!list_empty(&plug->mq_list))
3344 blk_mq_flush_plug_list(plug, from_schedule);
3345
3346 if (list_empty(&plug->list))
3347 return;
3348
3349 list_splice_init(&plug->list, &list);
3350
3351 list_sort(NULL, &list, plug_rq_cmp);
3352
3353 q = NULL;
3354 depth = 0;
3355
3356 /*
3357 * Save and disable interrupts here, to avoid doing it for every
3358 * queue lock we have to take.
3359 */
3360 local_irq_save(flags);
3361 while (!list_empty(&list)) {
3362 rq = list_entry_rq(list.next);
3363 list_del_init(&rq->queuelist);
3364 BUG_ON(!rq->q);
3365 if (rq->q != q) {
3366 /*
3367 * This drops the queue lock
3368 */
3369 if (q)
3370 queue_unplugged(q, depth, from_schedule);
3371 q = rq->q;
3372 depth = 0;
3373 spin_lock(q->queue_lock);
3374 }
3375
3376 /*
3377 * Short-circuit if @q is dead
3378 */
3379 if (unlikely(blk_queue_dying(q))) {
3380 __blk_end_request_all(rq, BLK_STS_IOERR);
3381 continue;
3382 }
3383
3384 /*
3385 * rq is already accounted, so use raw insert
3386 */
3387 if (op_is_flush(rq->cmd_flags))
3388 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3389 else
3390 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3391
3392 depth++;
3393 }
3394
3395 /*
3396 * This drops the queue lock
3397 */
3398 if (q)
3399 queue_unplugged(q, depth, from_schedule);
3400
3401 local_irq_restore(flags);
3402}
3403
3404void blk_finish_plug(struct blk_plug *plug)
3405{
3406 if (plug != current->plug)
3407 return;
3408 blk_flush_plug_list(plug, false);
3409
3410 current->plug = NULL;
3411}
3412EXPORT_SYMBOL(blk_finish_plug);
3413
3414#ifdef CONFIG_PM
3415/**
3416 * blk_pm_runtime_init - Block layer runtime PM initialization routine
3417 * @q: the queue of the device
3418 * @dev: the device the queue belongs to
3419 *
3420 * Description:
3421 * Initialize runtime-PM-related fields for @q and start auto suspend for
3422 * @dev. Drivers that want to take advantage of request-based runtime PM
3423 * should call this function after @dev has been initialized, and its
3424 * request queue @q has been allocated, and runtime PM for it can not happen
3425 * yet(either due to disabled/forbidden or its usage_count > 0). In most
3426 * cases, driver should call this function before any I/O has taken place.
3427 *
3428 * This function takes care of setting up using auto suspend for the device,
3429 * the autosuspend delay is set to -1 to make runtime suspend impossible
3430 * until an updated value is either set by user or by driver. Drivers do
3431 * not need to touch other autosuspend settings.
3432 *
3433 * The block layer runtime PM is request based, so only works for drivers
3434 * that use request as their IO unit instead of those directly use bio's.
3435 */
3436void blk_pm_runtime_init(struct request_queue *q, struct device *dev)
3437{
3438 /* not support for RQF_PM and ->rpm_status in blk-mq yet */
3439 if (q->mq_ops)
3440 return;
3441
3442 q->dev = dev;
3443 q->rpm_status = RPM_ACTIVE;
3444 pm_runtime_set_autosuspend_delay(q->dev, -1);
3445 pm_runtime_use_autosuspend(q->dev);
3446}
3447EXPORT_SYMBOL(blk_pm_runtime_init);
3448
3449/**
3450 * blk_pre_runtime_suspend - Pre runtime suspend check
3451 * @q: the queue of the device
3452 *
3453 * Description:
3454 * This function will check if runtime suspend is allowed for the device
3455 * by examining if there are any requests pending in the queue. If there
3456 * are requests pending, the device can not be runtime suspended; otherwise,
3457 * the queue's status will be updated to SUSPENDING and the driver can
3458 * proceed to suspend the device.
3459 *
3460 * For the not allowed case, we mark last busy for the device so that
3461 * runtime PM core will try to autosuspend it some time later.
3462 *
3463 * This function should be called near the start of the device's
3464 * runtime_suspend callback.
3465 *
3466 * Return:
3467 * 0 - OK to runtime suspend the device
3468 * -EBUSY - Device should not be runtime suspended
3469 */
3470int blk_pre_runtime_suspend(struct request_queue *q)
3471{
3472 int ret = 0;
3473
3474 if (!q->dev)
3475 return ret;
3476
3477 spin_lock_irq(q->queue_lock);
3478 if (q->nr_pending) {
3479 ret = -EBUSY;
3480 pm_runtime_mark_last_busy(q->dev);
3481 } else {
3482 q->rpm_status = RPM_SUSPENDING;
3483 }
3484 spin_unlock_irq(q->queue_lock);
3485 return ret;
3486}
3487EXPORT_SYMBOL(blk_pre_runtime_suspend);
3488
3489/**
3490 * blk_post_runtime_suspend - Post runtime suspend processing
3491 * @q: the queue of the device
3492 * @err: return value of the device's runtime_suspend function
3493 *
3494 * Description:
3495 * Update the queue's runtime status according to the return value of the
3496 * device's runtime suspend function and mark last busy for the device so
3497 * that PM core will try to auto suspend the device at a later time.
3498 *
3499 * This function should be called near the end of the device's
3500 * runtime_suspend callback.
3501 */
3502void blk_post_runtime_suspend(struct request_queue *q, int err)
3503{
3504 if (!q->dev)
3505 return;
3506
3507 spin_lock_irq(q->queue_lock);
3508 if (!err) {
3509 q->rpm_status = RPM_SUSPENDED;
3510 } else {
3511 q->rpm_status = RPM_ACTIVE;
3512 pm_runtime_mark_last_busy(q->dev);
3513 }
3514 spin_unlock_irq(q->queue_lock);
3515}
3516EXPORT_SYMBOL(blk_post_runtime_suspend);
3517
3518/**
3519 * blk_pre_runtime_resume - Pre runtime resume processing
3520 * @q: the queue of the device
3521 *
3522 * Description:
3523 * Update the queue's runtime status to RESUMING in preparation for the
3524 * runtime resume of the device.
3525 *
3526 * This function should be called near the start of the device's
3527 * runtime_resume callback.
3528 */
3529void blk_pre_runtime_resume(struct request_queue *q)
3530{
3531 if (!q->dev)
3532 return;
3533
3534 spin_lock_irq(q->queue_lock);
3535 q->rpm_status = RPM_RESUMING;
3536 spin_unlock_irq(q->queue_lock);
3537}
3538EXPORT_SYMBOL(blk_pre_runtime_resume);
3539
3540/**
3541 * blk_post_runtime_resume - Post runtime resume processing
3542 * @q: the queue of the device
3543 * @err: return value of the device's runtime_resume function
3544 *
3545 * Description:
3546 * Update the queue's runtime status according to the return value of the
3547 * device's runtime_resume function. If it is successfully resumed, process
3548 * the requests that are queued into the device's queue when it is resuming
3549 * and then mark last busy and initiate autosuspend for it.
3550 *
3551 * This function should be called near the end of the device's
3552 * runtime_resume callback.
3553 */
3554void blk_post_runtime_resume(struct request_queue *q, int err)
3555{
3556 if (!q->dev)
3557 return;
3558
3559 spin_lock_irq(q->queue_lock);
3560 if (!err) {
3561 q->rpm_status = RPM_ACTIVE;
3562 __blk_run_queue(q);
3563 pm_runtime_mark_last_busy(q->dev);
3564 pm_request_autosuspend(q->dev);
3565 } else {
3566 q->rpm_status = RPM_SUSPENDED;
3567 }
3568 spin_unlock_irq(q->queue_lock);
3569}
3570EXPORT_SYMBOL(blk_post_runtime_resume);
3571
3572/**
3573 * blk_set_runtime_active - Force runtime status of the queue to be active
3574 * @q: the queue of the device
3575 *
3576 * If the device is left runtime suspended during system suspend the resume
3577 * hook typically resumes the device and corrects runtime status
3578 * accordingly. However, that does not affect the queue runtime PM status
3579 * which is still "suspended". This prevents processing requests from the
3580 * queue.
3581 *
3582 * This function can be used in driver's resume hook to correct queue
3583 * runtime PM status and re-enable peeking requests from the queue. It
3584 * should be called before first request is added to the queue.
3585 */
3586void blk_set_runtime_active(struct request_queue *q)
3587{
3588 spin_lock_irq(q->queue_lock);
3589 q->rpm_status = RPM_ACTIVE;
3590 pm_runtime_mark_last_busy(q->dev);
3591 pm_request_autosuspend(q->dev);
3592 spin_unlock_irq(q->queue_lock);
3593}
3594EXPORT_SYMBOL(blk_set_runtime_active);
3595#endif
3596
3597int __init blk_dev_init(void)
3598{
3599 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
3600 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3601 FIELD_SIZEOF(struct request, cmd_flags));
3602 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
3603 FIELD_SIZEOF(struct bio, bi_opf));
3604
3605 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3606 kblockd_workqueue = alloc_workqueue("kblockd",
3607 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3608 if (!kblockd_workqueue)
3609 panic("Failed to create kblockd\n");
3610
3611 request_cachep = kmem_cache_create("blkdev_requests",
3612 sizeof(struct request), 0, SLAB_PANIC, NULL);
3613
3614 blk_requestq_cachep = kmem_cache_create("request_queue",
3615 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3616
3617#ifdef CONFIG_DEBUG_FS
3618 blk_debugfs_root = debugfs_create_dir("block", NULL);
3619#endif
3620
3621 return 0;
3622}