xref: /linux/block/blk-core.c (revision baaa68a9796ef2cadfe5caaf4c730412eda0f31c)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 1991, 1992 Linus Torvalds
4  * Copyright (C) 1994,      Karl Keyte: Added support for disk statistics
5  * Elevator latency, (C) 2000  Andrea Arcangeli <andrea@suse.de> SuSE
6  * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
7  * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8  *	-  July2000
9  * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
10  */
11 
12 /*
13  * This handles all read/write requests to block devices
14  */
15 #include <linux/kernel.h>
16 #include <linux/module.h>
17 #include <linux/bio.h>
18 #include <linux/blkdev.h>
19 #include <linux/blk-pm.h>
20 #include <linux/blk-integrity.h>
21 #include <linux/highmem.h>
22 #include <linux/mm.h>
23 #include <linux/pagemap.h>
24 #include <linux/kernel_stat.h>
25 #include <linux/string.h>
26 #include <linux/init.h>
27 #include <linux/completion.h>
28 #include <linux/slab.h>
29 #include <linux/swap.h>
30 #include <linux/writeback.h>
31 #include <linux/task_io_accounting_ops.h>
32 #include <linux/fault-inject.h>
33 #include <linux/list_sort.h>
34 #include <linux/delay.h>
35 #include <linux/ratelimit.h>
36 #include <linux/pm_runtime.h>
37 #include <linux/t10-pi.h>
38 #include <linux/debugfs.h>
39 #include <linux/bpf.h>
40 #include <linux/psi.h>
41 #include <linux/part_stat.h>
42 #include <linux/sched/sysctl.h>
43 #include <linux/blk-crypto.h>
44 
45 #define CREATE_TRACE_POINTS
46 #include <trace/events/block.h>
47 
48 #include "blk.h"
49 #include "blk-mq-sched.h"
50 #include "blk-pm.h"
51 #include "blk-cgroup.h"
52 #include "blk-throttle.h"
53 #include "blk-rq-qos.h"
54 
55 struct dentry *blk_debugfs_root;
56 
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
61 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
62 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert);
63 
64 DEFINE_IDA(blk_queue_ida);
65 
66 /*
67  * For queue allocation
68  */
69 struct kmem_cache *blk_requestq_cachep;
70 struct kmem_cache *blk_requestq_srcu_cachep;
71 
72 /*
73  * Controlling structure to kblockd
74  */
75 static struct workqueue_struct *kblockd_workqueue;
76 
77 /**
78  * blk_queue_flag_set - atomically set a queue flag
79  * @flag: flag to be set
80  * @q: request queue
81  */
82 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
83 {
84 	set_bit(flag, &q->queue_flags);
85 }
86 EXPORT_SYMBOL(blk_queue_flag_set);
87 
88 /**
89  * blk_queue_flag_clear - atomically clear a queue flag
90  * @flag: flag to be cleared
91  * @q: request queue
92  */
93 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
94 {
95 	clear_bit(flag, &q->queue_flags);
96 }
97 EXPORT_SYMBOL(blk_queue_flag_clear);
98 
99 /**
100  * blk_queue_flag_test_and_set - atomically test and set a queue flag
101  * @flag: flag to be set
102  * @q: request queue
103  *
104  * Returns the previous value of @flag - 0 if the flag was not set and 1 if
105  * the flag was already set.
106  */
107 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
108 {
109 	return test_and_set_bit(flag, &q->queue_flags);
110 }
111 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
112 
113 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
114 static const char *const blk_op_name[] = {
115 	REQ_OP_NAME(READ),
116 	REQ_OP_NAME(WRITE),
117 	REQ_OP_NAME(FLUSH),
118 	REQ_OP_NAME(DISCARD),
119 	REQ_OP_NAME(SECURE_ERASE),
120 	REQ_OP_NAME(ZONE_RESET),
121 	REQ_OP_NAME(ZONE_RESET_ALL),
122 	REQ_OP_NAME(ZONE_OPEN),
123 	REQ_OP_NAME(ZONE_CLOSE),
124 	REQ_OP_NAME(ZONE_FINISH),
125 	REQ_OP_NAME(ZONE_APPEND),
126 	REQ_OP_NAME(WRITE_SAME),
127 	REQ_OP_NAME(WRITE_ZEROES),
128 	REQ_OP_NAME(DRV_IN),
129 	REQ_OP_NAME(DRV_OUT),
130 };
131 #undef REQ_OP_NAME
132 
133 /**
134  * blk_op_str - Return string XXX in the REQ_OP_XXX.
135  * @op: REQ_OP_XXX.
136  *
137  * Description: Centralize block layer function to convert REQ_OP_XXX into
138  * string format. Useful in the debugging and tracing bio or request. For
139  * invalid REQ_OP_XXX it returns string "UNKNOWN".
140  */
141 inline const char *blk_op_str(unsigned int op)
142 {
143 	const char *op_str = "UNKNOWN";
144 
145 	if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
146 		op_str = blk_op_name[op];
147 
148 	return op_str;
149 }
150 EXPORT_SYMBOL_GPL(blk_op_str);
151 
152 static const struct {
153 	int		errno;
154 	const char	*name;
155 } blk_errors[] = {
156 	[BLK_STS_OK]		= { 0,		"" },
157 	[BLK_STS_NOTSUPP]	= { -EOPNOTSUPP, "operation not supported" },
158 	[BLK_STS_TIMEOUT]	= { -ETIMEDOUT,	"timeout" },
159 	[BLK_STS_NOSPC]		= { -ENOSPC,	"critical space allocation" },
160 	[BLK_STS_TRANSPORT]	= { -ENOLINK,	"recoverable transport" },
161 	[BLK_STS_TARGET]	= { -EREMOTEIO,	"critical target" },
162 	[BLK_STS_NEXUS]		= { -EBADE,	"critical nexus" },
163 	[BLK_STS_MEDIUM]	= { -ENODATA,	"critical medium" },
164 	[BLK_STS_PROTECTION]	= { -EILSEQ,	"protection" },
165 	[BLK_STS_RESOURCE]	= { -ENOMEM,	"kernel resource" },
166 	[BLK_STS_DEV_RESOURCE]	= { -EBUSY,	"device resource" },
167 	[BLK_STS_AGAIN]		= { -EAGAIN,	"nonblocking retry" },
168 	[BLK_STS_OFFLINE]	= { -ENODEV,	"device offline" },
169 
170 	/* device mapper special case, should not leak out: */
171 	[BLK_STS_DM_REQUEUE]	= { -EREMCHG, "dm internal retry" },
172 
173 	/* zone device specific errors */
174 	[BLK_STS_ZONE_OPEN_RESOURCE]	= { -ETOOMANYREFS, "open zones exceeded" },
175 	[BLK_STS_ZONE_ACTIVE_RESOURCE]	= { -EOVERFLOW, "active zones exceeded" },
176 
177 	/* everything else not covered above: */
178 	[BLK_STS_IOERR]		= { -EIO,	"I/O" },
179 };
180 
181 blk_status_t errno_to_blk_status(int errno)
182 {
183 	int i;
184 
185 	for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
186 		if (blk_errors[i].errno == errno)
187 			return (__force blk_status_t)i;
188 	}
189 
190 	return BLK_STS_IOERR;
191 }
192 EXPORT_SYMBOL_GPL(errno_to_blk_status);
193 
194 int blk_status_to_errno(blk_status_t status)
195 {
196 	int idx = (__force int)status;
197 
198 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
199 		return -EIO;
200 	return blk_errors[idx].errno;
201 }
202 EXPORT_SYMBOL_GPL(blk_status_to_errno);
203 
204 const char *blk_status_to_str(blk_status_t status)
205 {
206 	int idx = (__force int)status;
207 
208 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
209 		return "<null>";
210 	return blk_errors[idx].name;
211 }
212 
213 /**
214  * blk_sync_queue - cancel any pending callbacks on a queue
215  * @q: the queue
216  *
217  * Description:
218  *     The block layer may perform asynchronous callback activity
219  *     on a queue, such as calling the unplug function after a timeout.
220  *     A block device may call blk_sync_queue to ensure that any
221  *     such activity is cancelled, thus allowing it to release resources
222  *     that the callbacks might use. The caller must already have made sure
223  *     that its ->submit_bio will not re-add plugging prior to calling
224  *     this function.
225  *
226  *     This function does not cancel any asynchronous activity arising
227  *     out of elevator or throttling code. That would require elevator_exit()
228  *     and blkcg_exit_queue() to be called with queue lock initialized.
229  *
230  */
231 void blk_sync_queue(struct request_queue *q)
232 {
233 	del_timer_sync(&q->timeout);
234 	cancel_work_sync(&q->timeout_work);
235 }
236 EXPORT_SYMBOL(blk_sync_queue);
237 
238 /**
239  * blk_set_pm_only - increment pm_only counter
240  * @q: request queue pointer
241  */
242 void blk_set_pm_only(struct request_queue *q)
243 {
244 	atomic_inc(&q->pm_only);
245 }
246 EXPORT_SYMBOL_GPL(blk_set_pm_only);
247 
248 void blk_clear_pm_only(struct request_queue *q)
249 {
250 	int pm_only;
251 
252 	pm_only = atomic_dec_return(&q->pm_only);
253 	WARN_ON_ONCE(pm_only < 0);
254 	if (pm_only == 0)
255 		wake_up_all(&q->mq_freeze_wq);
256 }
257 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
258 
259 /**
260  * blk_put_queue - decrement the request_queue refcount
261  * @q: the request_queue structure to decrement the refcount for
262  *
263  * Decrements the refcount of the request_queue kobject. When this reaches 0
264  * we'll have blk_release_queue() called.
265  *
266  * Context: Any context, but the last reference must not be dropped from
267  *          atomic context.
268  */
269 void blk_put_queue(struct request_queue *q)
270 {
271 	kobject_put(&q->kobj);
272 }
273 EXPORT_SYMBOL(blk_put_queue);
274 
275 void blk_queue_start_drain(struct request_queue *q)
276 {
277 	/*
278 	 * When queue DYING flag is set, we need to block new req
279 	 * entering queue, so we call blk_freeze_queue_start() to
280 	 * prevent I/O from crossing blk_queue_enter().
281 	 */
282 	blk_freeze_queue_start(q);
283 	if (queue_is_mq(q))
284 		blk_mq_wake_waiters(q);
285 	/* Make blk_queue_enter() reexamine the DYING flag. */
286 	wake_up_all(&q->mq_freeze_wq);
287 }
288 
289 /**
290  * blk_cleanup_queue - shutdown a request queue
291  * @q: request queue to shutdown
292  *
293  * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
294  * put it.  All future requests will be failed immediately with -ENODEV.
295  *
296  * Context: can sleep
297  */
298 void blk_cleanup_queue(struct request_queue *q)
299 {
300 	/* cannot be called from atomic context */
301 	might_sleep();
302 
303 	WARN_ON_ONCE(blk_queue_registered(q));
304 
305 	/* mark @q DYING, no new request or merges will be allowed afterwards */
306 	blk_queue_flag_set(QUEUE_FLAG_DYING, q);
307 	blk_queue_start_drain(q);
308 
309 	blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
310 	blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
311 
312 	/*
313 	 * Drain all requests queued before DYING marking. Set DEAD flag to
314 	 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
315 	 * after draining finished.
316 	 */
317 	blk_freeze_queue(q);
318 
319 	/* cleanup rq qos structures for queue without disk */
320 	rq_qos_exit(q);
321 
322 	blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
323 
324 	blk_sync_queue(q);
325 	if (queue_is_mq(q)) {
326 		blk_mq_cancel_work_sync(q);
327 		blk_mq_exit_queue(q);
328 	}
329 
330 	/*
331 	 * In theory, request pool of sched_tags belongs to request queue.
332 	 * However, the current implementation requires tag_set for freeing
333 	 * requests, so free the pool now.
334 	 *
335 	 * Queue has become frozen, there can't be any in-queue requests, so
336 	 * it is safe to free requests now.
337 	 */
338 	mutex_lock(&q->sysfs_lock);
339 	if (q->elevator)
340 		blk_mq_sched_free_rqs(q);
341 	mutex_unlock(&q->sysfs_lock);
342 
343 	/* @q is and will stay empty, shutdown and put */
344 	blk_put_queue(q);
345 }
346 EXPORT_SYMBOL(blk_cleanup_queue);
347 
348 /**
349  * blk_queue_enter() - try to increase q->q_usage_counter
350  * @q: request queue pointer
351  * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
352  */
353 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
354 {
355 	const bool pm = flags & BLK_MQ_REQ_PM;
356 
357 	while (!blk_try_enter_queue(q, pm)) {
358 		if (flags & BLK_MQ_REQ_NOWAIT)
359 			return -EBUSY;
360 
361 		/*
362 		 * read pair of barrier in blk_freeze_queue_start(), we need to
363 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
364 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
365 		 * following wait may never return if the two reads are
366 		 * reordered.
367 		 */
368 		smp_rmb();
369 		wait_event(q->mq_freeze_wq,
370 			   (!q->mq_freeze_depth &&
371 			    blk_pm_resume_queue(pm, q)) ||
372 			   blk_queue_dying(q));
373 		if (blk_queue_dying(q))
374 			return -ENODEV;
375 	}
376 
377 	return 0;
378 }
379 
380 int __bio_queue_enter(struct request_queue *q, struct bio *bio)
381 {
382 	while (!blk_try_enter_queue(q, false)) {
383 		struct gendisk *disk = bio->bi_bdev->bd_disk;
384 
385 		if (bio->bi_opf & REQ_NOWAIT) {
386 			if (test_bit(GD_DEAD, &disk->state))
387 				goto dead;
388 			bio_wouldblock_error(bio);
389 			return -EBUSY;
390 		}
391 
392 		/*
393 		 * read pair of barrier in blk_freeze_queue_start(), we need to
394 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
395 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
396 		 * following wait may never return if the two reads are
397 		 * reordered.
398 		 */
399 		smp_rmb();
400 		wait_event(q->mq_freeze_wq,
401 			   (!q->mq_freeze_depth &&
402 			    blk_pm_resume_queue(false, q)) ||
403 			   test_bit(GD_DEAD, &disk->state));
404 		if (test_bit(GD_DEAD, &disk->state))
405 			goto dead;
406 	}
407 
408 	return 0;
409 dead:
410 	bio_io_error(bio);
411 	return -ENODEV;
412 }
413 
414 void blk_queue_exit(struct request_queue *q)
415 {
416 	percpu_ref_put(&q->q_usage_counter);
417 }
418 
419 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
420 {
421 	struct request_queue *q =
422 		container_of(ref, struct request_queue, q_usage_counter);
423 
424 	wake_up_all(&q->mq_freeze_wq);
425 }
426 
427 static void blk_rq_timed_out_timer(struct timer_list *t)
428 {
429 	struct request_queue *q = from_timer(q, t, timeout);
430 
431 	kblockd_schedule_work(&q->timeout_work);
432 }
433 
434 static void blk_timeout_work(struct work_struct *work)
435 {
436 }
437 
438 struct request_queue *blk_alloc_queue(int node_id, bool alloc_srcu)
439 {
440 	struct request_queue *q;
441 	int ret;
442 
443 	q = kmem_cache_alloc_node(blk_get_queue_kmem_cache(alloc_srcu),
444 			GFP_KERNEL | __GFP_ZERO, node_id);
445 	if (!q)
446 		return NULL;
447 
448 	if (alloc_srcu) {
449 		blk_queue_flag_set(QUEUE_FLAG_HAS_SRCU, q);
450 		if (init_srcu_struct(q->srcu) != 0)
451 			goto fail_q;
452 	}
453 
454 	q->last_merge = NULL;
455 
456 	q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
457 	if (q->id < 0)
458 		goto fail_srcu;
459 
460 	ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, 0);
461 	if (ret)
462 		goto fail_id;
463 
464 	q->stats = blk_alloc_queue_stats();
465 	if (!q->stats)
466 		goto fail_split;
467 
468 	q->node = node_id;
469 
470 	atomic_set(&q->nr_active_requests_shared_tags, 0);
471 
472 	timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
473 	INIT_WORK(&q->timeout_work, blk_timeout_work);
474 	INIT_LIST_HEAD(&q->icq_list);
475 
476 	kobject_init(&q->kobj, &blk_queue_ktype);
477 
478 	mutex_init(&q->debugfs_mutex);
479 	mutex_init(&q->sysfs_lock);
480 	mutex_init(&q->sysfs_dir_lock);
481 	spin_lock_init(&q->queue_lock);
482 
483 	init_waitqueue_head(&q->mq_freeze_wq);
484 	mutex_init(&q->mq_freeze_lock);
485 
486 	/*
487 	 * Init percpu_ref in atomic mode so that it's faster to shutdown.
488 	 * See blk_register_queue() for details.
489 	 */
490 	if (percpu_ref_init(&q->q_usage_counter,
491 				blk_queue_usage_counter_release,
492 				PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
493 		goto fail_stats;
494 
495 	blk_queue_dma_alignment(q, 511);
496 	blk_set_default_limits(&q->limits);
497 	q->nr_requests = BLKDEV_DEFAULT_RQ;
498 
499 	return q;
500 
501 fail_stats:
502 	blk_free_queue_stats(q->stats);
503 fail_split:
504 	bioset_exit(&q->bio_split);
505 fail_id:
506 	ida_simple_remove(&blk_queue_ida, q->id);
507 fail_srcu:
508 	if (alloc_srcu)
509 		cleanup_srcu_struct(q->srcu);
510 fail_q:
511 	kmem_cache_free(blk_get_queue_kmem_cache(alloc_srcu), q);
512 	return NULL;
513 }
514 
515 /**
516  * blk_get_queue - increment the request_queue refcount
517  * @q: the request_queue structure to increment the refcount for
518  *
519  * Increment the refcount of the request_queue kobject.
520  *
521  * Context: Any context.
522  */
523 bool blk_get_queue(struct request_queue *q)
524 {
525 	if (likely(!blk_queue_dying(q))) {
526 		__blk_get_queue(q);
527 		return true;
528 	}
529 
530 	return false;
531 }
532 EXPORT_SYMBOL(blk_get_queue);
533 
534 #ifdef CONFIG_FAIL_MAKE_REQUEST
535 
536 static DECLARE_FAULT_ATTR(fail_make_request);
537 
538 static int __init setup_fail_make_request(char *str)
539 {
540 	return setup_fault_attr(&fail_make_request, str);
541 }
542 __setup("fail_make_request=", setup_fail_make_request);
543 
544 bool should_fail_request(struct block_device *part, unsigned int bytes)
545 {
546 	return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
547 }
548 
549 static int __init fail_make_request_debugfs(void)
550 {
551 	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
552 						NULL, &fail_make_request);
553 
554 	return PTR_ERR_OR_ZERO(dir);
555 }
556 
557 late_initcall(fail_make_request_debugfs);
558 #endif /* CONFIG_FAIL_MAKE_REQUEST */
559 
560 static inline bool bio_check_ro(struct bio *bio)
561 {
562 	if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
563 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
564 			return false;
565 		pr_warn("Trying to write to read-only block-device %pg\n",
566 			bio->bi_bdev);
567 		/* Older lvm-tools actually trigger this */
568 		return false;
569 	}
570 
571 	return false;
572 }
573 
574 static noinline int should_fail_bio(struct bio *bio)
575 {
576 	if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
577 		return -EIO;
578 	return 0;
579 }
580 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
581 
582 /*
583  * Check whether this bio extends beyond the end of the device or partition.
584  * This may well happen - the kernel calls bread() without checking the size of
585  * the device, e.g., when mounting a file system.
586  */
587 static inline int bio_check_eod(struct bio *bio)
588 {
589 	sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
590 	unsigned int nr_sectors = bio_sectors(bio);
591 
592 	if (nr_sectors && maxsector &&
593 	    (nr_sectors > maxsector ||
594 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
595 		pr_info_ratelimited("%s: attempt to access beyond end of device\n"
596 				    "%pg: rw=%d, want=%llu, limit=%llu\n",
597 				    current->comm,
598 				    bio->bi_bdev, bio->bi_opf,
599 				    bio_end_sector(bio), maxsector);
600 		return -EIO;
601 	}
602 	return 0;
603 }
604 
605 /*
606  * Remap block n of partition p to block n+start(p) of the disk.
607  */
608 static int blk_partition_remap(struct bio *bio)
609 {
610 	struct block_device *p = bio->bi_bdev;
611 
612 	if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
613 		return -EIO;
614 	if (bio_sectors(bio)) {
615 		bio->bi_iter.bi_sector += p->bd_start_sect;
616 		trace_block_bio_remap(bio, p->bd_dev,
617 				      bio->bi_iter.bi_sector -
618 				      p->bd_start_sect);
619 	}
620 	bio_set_flag(bio, BIO_REMAPPED);
621 	return 0;
622 }
623 
624 /*
625  * Check write append to a zoned block device.
626  */
627 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
628 						 struct bio *bio)
629 {
630 	sector_t pos = bio->bi_iter.bi_sector;
631 	int nr_sectors = bio_sectors(bio);
632 
633 	/* Only applicable to zoned block devices */
634 	if (!blk_queue_is_zoned(q))
635 		return BLK_STS_NOTSUPP;
636 
637 	/* The bio sector must point to the start of a sequential zone */
638 	if (pos & (blk_queue_zone_sectors(q) - 1) ||
639 	    !blk_queue_zone_is_seq(q, pos))
640 		return BLK_STS_IOERR;
641 
642 	/*
643 	 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
644 	 * split and could result in non-contiguous sectors being written in
645 	 * different zones.
646 	 */
647 	if (nr_sectors > q->limits.chunk_sectors)
648 		return BLK_STS_IOERR;
649 
650 	/* Make sure the BIO is small enough and will not get split */
651 	if (nr_sectors > q->limits.max_zone_append_sectors)
652 		return BLK_STS_IOERR;
653 
654 	bio->bi_opf |= REQ_NOMERGE;
655 
656 	return BLK_STS_OK;
657 }
658 
659 static void __submit_bio(struct bio *bio)
660 {
661 	struct gendisk *disk = bio->bi_bdev->bd_disk;
662 
663 	if (unlikely(!blk_crypto_bio_prep(&bio)))
664 		return;
665 
666 	if (!disk->fops->submit_bio) {
667 		blk_mq_submit_bio(bio);
668 	} else if (likely(bio_queue_enter(bio) == 0)) {
669 		disk->fops->submit_bio(bio);
670 		blk_queue_exit(disk->queue);
671 	}
672 }
673 
674 /*
675  * The loop in this function may be a bit non-obvious, and so deserves some
676  * explanation:
677  *
678  *  - Before entering the loop, bio->bi_next is NULL (as all callers ensure
679  *    that), so we have a list with a single bio.
680  *  - We pretend that we have just taken it off a longer list, so we assign
681  *    bio_list to a pointer to the bio_list_on_stack, thus initialising the
682  *    bio_list of new bios to be added.  ->submit_bio() may indeed add some more
683  *    bios through a recursive call to submit_bio_noacct.  If it did, we find a
684  *    non-NULL value in bio_list and re-enter the loop from the top.
685  *  - In this case we really did just take the bio of the top of the list (no
686  *    pretending) and so remove it from bio_list, and call into ->submit_bio()
687  *    again.
688  *
689  * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
690  * bio_list_on_stack[1] contains bios that were submitted before the current
691  *	->submit_bio_bio, but that haven't been processed yet.
692  */
693 static void __submit_bio_noacct(struct bio *bio)
694 {
695 	struct bio_list bio_list_on_stack[2];
696 
697 	BUG_ON(bio->bi_next);
698 
699 	bio_list_init(&bio_list_on_stack[0]);
700 	current->bio_list = bio_list_on_stack;
701 
702 	do {
703 		struct request_queue *q = bdev_get_queue(bio->bi_bdev);
704 		struct bio_list lower, same;
705 
706 		/*
707 		 * Create a fresh bio_list for all subordinate requests.
708 		 */
709 		bio_list_on_stack[1] = bio_list_on_stack[0];
710 		bio_list_init(&bio_list_on_stack[0]);
711 
712 		__submit_bio(bio);
713 
714 		/*
715 		 * Sort new bios into those for a lower level and those for the
716 		 * same level.
717 		 */
718 		bio_list_init(&lower);
719 		bio_list_init(&same);
720 		while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
721 			if (q == bdev_get_queue(bio->bi_bdev))
722 				bio_list_add(&same, bio);
723 			else
724 				bio_list_add(&lower, bio);
725 
726 		/*
727 		 * Now assemble so we handle the lowest level first.
728 		 */
729 		bio_list_merge(&bio_list_on_stack[0], &lower);
730 		bio_list_merge(&bio_list_on_stack[0], &same);
731 		bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
732 	} while ((bio = bio_list_pop(&bio_list_on_stack[0])));
733 
734 	current->bio_list = NULL;
735 }
736 
737 static void __submit_bio_noacct_mq(struct bio *bio)
738 {
739 	struct bio_list bio_list[2] = { };
740 
741 	current->bio_list = bio_list;
742 
743 	do {
744 		__submit_bio(bio);
745 	} while ((bio = bio_list_pop(&bio_list[0])));
746 
747 	current->bio_list = NULL;
748 }
749 
750 void submit_bio_noacct_nocheck(struct bio *bio)
751 {
752 	/*
753 	 * We only want one ->submit_bio to be active at a time, else stack
754 	 * usage with stacked devices could be a problem.  Use current->bio_list
755 	 * to collect a list of requests submited by a ->submit_bio method while
756 	 * it is active, and then process them after it returned.
757 	 */
758 	if (current->bio_list)
759 		bio_list_add(&current->bio_list[0], bio);
760 	else if (!bio->bi_bdev->bd_disk->fops->submit_bio)
761 		__submit_bio_noacct_mq(bio);
762 	else
763 		__submit_bio_noacct(bio);
764 }
765 
766 /**
767  * submit_bio_noacct - re-submit a bio to the block device layer for I/O
768  * @bio:  The bio describing the location in memory and on the device.
769  *
770  * This is a version of submit_bio() that shall only be used for I/O that is
771  * resubmitted to lower level drivers by stacking block drivers.  All file
772  * systems and other upper level users of the block layer should use
773  * submit_bio() instead.
774  */
775 void submit_bio_noacct(struct bio *bio)
776 {
777 	struct block_device *bdev = bio->bi_bdev;
778 	struct request_queue *q = bdev_get_queue(bdev);
779 	blk_status_t status = BLK_STS_IOERR;
780 	struct blk_plug *plug;
781 
782 	might_sleep();
783 
784 	plug = blk_mq_plug(q, bio);
785 	if (plug && plug->nowait)
786 		bio->bi_opf |= REQ_NOWAIT;
787 
788 	/*
789 	 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
790 	 * if queue does not support NOWAIT.
791 	 */
792 	if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q))
793 		goto not_supported;
794 
795 	if (should_fail_bio(bio))
796 		goto end_io;
797 	if (unlikely(bio_check_ro(bio)))
798 		goto end_io;
799 	if (!bio_flagged(bio, BIO_REMAPPED)) {
800 		if (unlikely(bio_check_eod(bio)))
801 			goto end_io;
802 		if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
803 			goto end_io;
804 	}
805 
806 	/*
807 	 * Filter flush bio's early so that bio based drivers without flush
808 	 * support don't have to worry about them.
809 	 */
810 	if (op_is_flush(bio->bi_opf) &&
811 	    !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
812 		bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
813 		if (!bio_sectors(bio)) {
814 			status = BLK_STS_OK;
815 			goto end_io;
816 		}
817 	}
818 
819 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
820 		bio_clear_polled(bio);
821 
822 	switch (bio_op(bio)) {
823 	case REQ_OP_DISCARD:
824 		if (!blk_queue_discard(q))
825 			goto not_supported;
826 		break;
827 	case REQ_OP_SECURE_ERASE:
828 		if (!blk_queue_secure_erase(q))
829 			goto not_supported;
830 		break;
831 	case REQ_OP_WRITE_SAME:
832 		if (!q->limits.max_write_same_sectors)
833 			goto not_supported;
834 		break;
835 	case REQ_OP_ZONE_APPEND:
836 		status = blk_check_zone_append(q, bio);
837 		if (status != BLK_STS_OK)
838 			goto end_io;
839 		break;
840 	case REQ_OP_ZONE_RESET:
841 	case REQ_OP_ZONE_OPEN:
842 	case REQ_OP_ZONE_CLOSE:
843 	case REQ_OP_ZONE_FINISH:
844 		if (!blk_queue_is_zoned(q))
845 			goto not_supported;
846 		break;
847 	case REQ_OP_ZONE_RESET_ALL:
848 		if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
849 			goto not_supported;
850 		break;
851 	case REQ_OP_WRITE_ZEROES:
852 		if (!q->limits.max_write_zeroes_sectors)
853 			goto not_supported;
854 		break;
855 	default:
856 		break;
857 	}
858 
859 	if (blk_throtl_bio(bio))
860 		return;
861 
862 	blk_cgroup_bio_start(bio);
863 	blkcg_bio_issue_init(bio);
864 
865 	if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
866 		trace_block_bio_queue(bio);
867 		/* Now that enqueuing has been traced, we need to trace
868 		 * completion as well.
869 		 */
870 		bio_set_flag(bio, BIO_TRACE_COMPLETION);
871 	}
872 	submit_bio_noacct_nocheck(bio);
873 	return;
874 
875 not_supported:
876 	status = BLK_STS_NOTSUPP;
877 end_io:
878 	bio->bi_status = status;
879 	bio_endio(bio);
880 }
881 EXPORT_SYMBOL(submit_bio_noacct);
882 
883 /**
884  * submit_bio - submit a bio to the block device layer for I/O
885  * @bio: The &struct bio which describes the I/O
886  *
887  * submit_bio() is used to submit I/O requests to block devices.  It is passed a
888  * fully set up &struct bio that describes the I/O that needs to be done.  The
889  * bio will be send to the device described by the bi_bdev field.
890  *
891  * The success/failure status of the request, along with notification of
892  * completion, is delivered asynchronously through the ->bi_end_io() callback
893  * in @bio.  The bio must NOT be touched by thecaller until ->bi_end_io() has
894  * been called.
895  */
896 void submit_bio(struct bio *bio)
897 {
898 	if (blkcg_punt_bio_submit(bio))
899 		return;
900 
901 	/*
902 	 * If it's a regular read/write or a barrier with data attached,
903 	 * go through the normal accounting stuff before submission.
904 	 */
905 	if (bio_has_data(bio)) {
906 		unsigned int count;
907 
908 		if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
909 			count = queue_logical_block_size(
910 					bdev_get_queue(bio->bi_bdev)) >> 9;
911 		else
912 			count = bio_sectors(bio);
913 
914 		if (op_is_write(bio_op(bio))) {
915 			count_vm_events(PGPGOUT, count);
916 		} else {
917 			task_io_account_read(bio->bi_iter.bi_size);
918 			count_vm_events(PGPGIN, count);
919 		}
920 	}
921 
922 	/*
923 	 * If we're reading data that is part of the userspace workingset, count
924 	 * submission time as memory stall.  When the device is congested, or
925 	 * the submitting cgroup IO-throttled, submission can be a significant
926 	 * part of overall IO time.
927 	 */
928 	if (unlikely(bio_op(bio) == REQ_OP_READ &&
929 	    bio_flagged(bio, BIO_WORKINGSET))) {
930 		unsigned long pflags;
931 
932 		psi_memstall_enter(&pflags);
933 		submit_bio_noacct(bio);
934 		psi_memstall_leave(&pflags);
935 		return;
936 	}
937 
938 	submit_bio_noacct(bio);
939 }
940 EXPORT_SYMBOL(submit_bio);
941 
942 /**
943  * bio_poll - poll for BIO completions
944  * @bio: bio to poll for
945  * @iob: batches of IO
946  * @flags: BLK_POLL_* flags that control the behavior
947  *
948  * Poll for completions on queue associated with the bio. Returns number of
949  * completed entries found.
950  *
951  * Note: the caller must either be the context that submitted @bio, or
952  * be in a RCU critical section to prevent freeing of @bio.
953  */
954 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
955 {
956 	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
957 	blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
958 	int ret;
959 
960 	if (cookie == BLK_QC_T_NONE ||
961 	    !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
962 		return 0;
963 
964 	blk_flush_plug(current->plug, false);
965 
966 	if (blk_queue_enter(q, BLK_MQ_REQ_NOWAIT))
967 		return 0;
968 	if (WARN_ON_ONCE(!queue_is_mq(q)))
969 		ret = 0;	/* not yet implemented, should not happen */
970 	else
971 		ret = blk_mq_poll(q, cookie, iob, flags);
972 	blk_queue_exit(q);
973 	return ret;
974 }
975 EXPORT_SYMBOL_GPL(bio_poll);
976 
977 /*
978  * Helper to implement file_operations.iopoll.  Requires the bio to be stored
979  * in iocb->private, and cleared before freeing the bio.
980  */
981 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
982 		    unsigned int flags)
983 {
984 	struct bio *bio;
985 	int ret = 0;
986 
987 	/*
988 	 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
989 	 * point to a freshly allocated bio at this point.  If that happens
990 	 * we have a few cases to consider:
991 	 *
992 	 *  1) the bio is beeing initialized and bi_bdev is NULL.  We can just
993 	 *     simply nothing in this case
994 	 *  2) the bio points to a not poll enabled device.  bio_poll will catch
995 	 *     this and return 0
996 	 *  3) the bio points to a poll capable device, including but not
997 	 *     limited to the one that the original bio pointed to.  In this
998 	 *     case we will call into the actual poll method and poll for I/O,
999 	 *     even if we don't need to, but it won't cause harm either.
1000 	 *
1001 	 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
1002 	 * is still allocated. Because partitions hold a reference to the whole
1003 	 * device bdev and thus disk, the disk is also still valid.  Grabbing
1004 	 * a reference to the queue in bio_poll() ensures the hctxs and requests
1005 	 * are still valid as well.
1006 	 */
1007 	rcu_read_lock();
1008 	bio = READ_ONCE(kiocb->private);
1009 	if (bio && bio->bi_bdev)
1010 		ret = bio_poll(bio, iob, flags);
1011 	rcu_read_unlock();
1012 
1013 	return ret;
1014 }
1015 EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
1016 
1017 void update_io_ticks(struct block_device *part, unsigned long now, bool end)
1018 {
1019 	unsigned long stamp;
1020 again:
1021 	stamp = READ_ONCE(part->bd_stamp);
1022 	if (unlikely(time_after(now, stamp))) {
1023 		if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp))
1024 			__part_stat_add(part, io_ticks, end ? now - stamp : 1);
1025 	}
1026 	if (part->bd_partno) {
1027 		part = bdev_whole(part);
1028 		goto again;
1029 	}
1030 }
1031 
1032 static unsigned long __part_start_io_acct(struct block_device *part,
1033 					  unsigned int sectors, unsigned int op,
1034 					  unsigned long start_time)
1035 {
1036 	const int sgrp = op_stat_group(op);
1037 
1038 	part_stat_lock();
1039 	update_io_ticks(part, start_time, false);
1040 	part_stat_inc(part, ios[sgrp]);
1041 	part_stat_add(part, sectors[sgrp], sectors);
1042 	part_stat_local_inc(part, in_flight[op_is_write(op)]);
1043 	part_stat_unlock();
1044 
1045 	return start_time;
1046 }
1047 
1048 /**
1049  * bio_start_io_acct_time - start I/O accounting for bio based drivers
1050  * @bio:	bio to start account for
1051  * @start_time:	start time that should be passed back to bio_end_io_acct().
1052  */
1053 void bio_start_io_acct_time(struct bio *bio, unsigned long start_time)
1054 {
1055 	__part_start_io_acct(bio->bi_bdev, bio_sectors(bio),
1056 			     bio_op(bio), start_time);
1057 }
1058 EXPORT_SYMBOL_GPL(bio_start_io_acct_time);
1059 
1060 /**
1061  * bio_start_io_acct - start I/O accounting for bio based drivers
1062  * @bio:	bio to start account for
1063  *
1064  * Returns the start time that should be passed back to bio_end_io_acct().
1065  */
1066 unsigned long bio_start_io_acct(struct bio *bio)
1067 {
1068 	return __part_start_io_acct(bio->bi_bdev, bio_sectors(bio),
1069 				    bio_op(bio), jiffies);
1070 }
1071 EXPORT_SYMBOL_GPL(bio_start_io_acct);
1072 
1073 unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
1074 				 unsigned int op)
1075 {
1076 	return __part_start_io_acct(disk->part0, sectors, op, jiffies);
1077 }
1078 EXPORT_SYMBOL(disk_start_io_acct);
1079 
1080 static void __part_end_io_acct(struct block_device *part, unsigned int op,
1081 			       unsigned long start_time)
1082 {
1083 	const int sgrp = op_stat_group(op);
1084 	unsigned long now = READ_ONCE(jiffies);
1085 	unsigned long duration = now - start_time;
1086 
1087 	part_stat_lock();
1088 	update_io_ticks(part, now, true);
1089 	part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1090 	part_stat_local_dec(part, in_flight[op_is_write(op)]);
1091 	part_stat_unlock();
1092 }
1093 
1094 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
1095 		struct block_device *orig_bdev)
1096 {
1097 	__part_end_io_acct(orig_bdev, bio_op(bio), start_time);
1098 }
1099 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1100 
1101 void disk_end_io_acct(struct gendisk *disk, unsigned int op,
1102 		      unsigned long start_time)
1103 {
1104 	__part_end_io_acct(disk->part0, op, start_time);
1105 }
1106 EXPORT_SYMBOL(disk_end_io_acct);
1107 
1108 /**
1109  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1110  * @q : the queue of the device being checked
1111  *
1112  * Description:
1113  *    Check if underlying low-level drivers of a device are busy.
1114  *    If the drivers want to export their busy state, they must set own
1115  *    exporting function using blk_queue_lld_busy() first.
1116  *
1117  *    Basically, this function is used only by request stacking drivers
1118  *    to stop dispatching requests to underlying devices when underlying
1119  *    devices are busy.  This behavior helps more I/O merging on the queue
1120  *    of the request stacking driver and prevents I/O throughput regression
1121  *    on burst I/O load.
1122  *
1123  * Return:
1124  *    0 - Not busy (The request stacking driver should dispatch request)
1125  *    1 - Busy (The request stacking driver should stop dispatching request)
1126  */
1127 int blk_lld_busy(struct request_queue *q)
1128 {
1129 	if (queue_is_mq(q) && q->mq_ops->busy)
1130 		return q->mq_ops->busy(q);
1131 
1132 	return 0;
1133 }
1134 EXPORT_SYMBOL_GPL(blk_lld_busy);
1135 
1136 int kblockd_schedule_work(struct work_struct *work)
1137 {
1138 	return queue_work(kblockd_workqueue, work);
1139 }
1140 EXPORT_SYMBOL(kblockd_schedule_work);
1141 
1142 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1143 				unsigned long delay)
1144 {
1145 	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1146 }
1147 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1148 
1149 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1150 {
1151 	struct task_struct *tsk = current;
1152 
1153 	/*
1154 	 * If this is a nested plug, don't actually assign it.
1155 	 */
1156 	if (tsk->plug)
1157 		return;
1158 
1159 	plug->mq_list = NULL;
1160 	plug->cached_rq = NULL;
1161 	plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1162 	plug->rq_count = 0;
1163 	plug->multiple_queues = false;
1164 	plug->has_elevator = false;
1165 	plug->nowait = false;
1166 	INIT_LIST_HEAD(&plug->cb_list);
1167 
1168 	/*
1169 	 * Store ordering should not be needed here, since a potential
1170 	 * preempt will imply a full memory barrier
1171 	 */
1172 	tsk->plug = plug;
1173 }
1174 
1175 /**
1176  * blk_start_plug - initialize blk_plug and track it inside the task_struct
1177  * @plug:	The &struct blk_plug that needs to be initialized
1178  *
1179  * Description:
1180  *   blk_start_plug() indicates to the block layer an intent by the caller
1181  *   to submit multiple I/O requests in a batch.  The block layer may use
1182  *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1183  *   is called.  However, the block layer may choose to submit requests
1184  *   before a call to blk_finish_plug() if the number of queued I/Os
1185  *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1186  *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1187  *   the task schedules (see below).
1188  *
1189  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1190  *   pending I/O should the task end up blocking between blk_start_plug() and
1191  *   blk_finish_plug(). This is important from a performance perspective, but
1192  *   also ensures that we don't deadlock. For instance, if the task is blocking
1193  *   for a memory allocation, memory reclaim could end up wanting to free a
1194  *   page belonging to that request that is currently residing in our private
1195  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1196  *   this kind of deadlock.
1197  */
1198 void blk_start_plug(struct blk_plug *plug)
1199 {
1200 	blk_start_plug_nr_ios(plug, 1);
1201 }
1202 EXPORT_SYMBOL(blk_start_plug);
1203 
1204 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1205 {
1206 	LIST_HEAD(callbacks);
1207 
1208 	while (!list_empty(&plug->cb_list)) {
1209 		list_splice_init(&plug->cb_list, &callbacks);
1210 
1211 		while (!list_empty(&callbacks)) {
1212 			struct blk_plug_cb *cb = list_first_entry(&callbacks,
1213 							  struct blk_plug_cb,
1214 							  list);
1215 			list_del(&cb->list);
1216 			cb->callback(cb, from_schedule);
1217 		}
1218 	}
1219 }
1220 
1221 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1222 				      int size)
1223 {
1224 	struct blk_plug *plug = current->plug;
1225 	struct blk_plug_cb *cb;
1226 
1227 	if (!plug)
1228 		return NULL;
1229 
1230 	list_for_each_entry(cb, &plug->cb_list, list)
1231 		if (cb->callback == unplug && cb->data == data)
1232 			return cb;
1233 
1234 	/* Not currently on the callback list */
1235 	BUG_ON(size < sizeof(*cb));
1236 	cb = kzalloc(size, GFP_ATOMIC);
1237 	if (cb) {
1238 		cb->data = data;
1239 		cb->callback = unplug;
1240 		list_add(&cb->list, &plug->cb_list);
1241 	}
1242 	return cb;
1243 }
1244 EXPORT_SYMBOL(blk_check_plugged);
1245 
1246 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1247 {
1248 	if (!list_empty(&plug->cb_list))
1249 		flush_plug_callbacks(plug, from_schedule);
1250 	if (!rq_list_empty(plug->mq_list))
1251 		blk_mq_flush_plug_list(plug, from_schedule);
1252 	/*
1253 	 * Unconditionally flush out cached requests, even if the unplug
1254 	 * event came from schedule. Since we know hold references to the
1255 	 * queue for cached requests, we don't want a blocked task holding
1256 	 * up a queue freeze/quiesce event.
1257 	 */
1258 	if (unlikely(!rq_list_empty(plug->cached_rq)))
1259 		blk_mq_free_plug_rqs(plug);
1260 }
1261 
1262 /**
1263  * blk_finish_plug - mark the end of a batch of submitted I/O
1264  * @plug:	The &struct blk_plug passed to blk_start_plug()
1265  *
1266  * Description:
1267  * Indicate that a batch of I/O submissions is complete.  This function
1268  * must be paired with an initial call to blk_start_plug().  The intent
1269  * is to allow the block layer to optimize I/O submission.  See the
1270  * documentation for blk_start_plug() for more information.
1271  */
1272 void blk_finish_plug(struct blk_plug *plug)
1273 {
1274 	if (plug == current->plug) {
1275 		__blk_flush_plug(plug, false);
1276 		current->plug = NULL;
1277 	}
1278 }
1279 EXPORT_SYMBOL(blk_finish_plug);
1280 
1281 void blk_io_schedule(void)
1282 {
1283 	/* Prevent hang_check timer from firing at us during very long I/O */
1284 	unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1285 
1286 	if (timeout)
1287 		io_schedule_timeout(timeout);
1288 	else
1289 		io_schedule();
1290 }
1291 EXPORT_SYMBOL_GPL(blk_io_schedule);
1292 
1293 int __init blk_dev_init(void)
1294 {
1295 	BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1296 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1297 			sizeof_field(struct request, cmd_flags));
1298 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1299 			sizeof_field(struct bio, bi_opf));
1300 	BUILD_BUG_ON(ALIGN(offsetof(struct request_queue, srcu),
1301 			   __alignof__(struct request_queue)) !=
1302 		     sizeof(struct request_queue));
1303 
1304 	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
1305 	kblockd_workqueue = alloc_workqueue("kblockd",
1306 					    WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1307 	if (!kblockd_workqueue)
1308 		panic("Failed to create kblockd\n");
1309 
1310 	blk_requestq_cachep = kmem_cache_create("request_queue",
1311 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1312 
1313 	blk_requestq_srcu_cachep = kmem_cache_create("request_queue_srcu",
1314 			sizeof(struct request_queue) +
1315 			sizeof(struct srcu_struct), 0, SLAB_PANIC, NULL);
1316 
1317 	blk_debugfs_root = debugfs_create_dir("block", NULL);
1318 
1319 	return 0;
1320 }
1321