xref: /openbmc/linux/block/blk-core.c (revision 56b5b1c7)
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/blk-cgroup.h>
38 #include <linux/t10-pi.h>
39 #include <linux/debugfs.h>
40 #include <linux/bpf.h>
41 #include <linux/psi.h>
42 #include <linux/part_stat.h>
43 #include <linux/sched/sysctl.h>
44 #include <linux/blk-crypto.h>
45 
46 #define CREATE_TRACE_POINTS
47 #include <trace/events/block.h>
48 
49 #include "blk.h"
50 #include "blk-mq-sched.h"
51 #include "blk-pm.h"
52 #include "blk-throttle.h"
53 
54 struct dentry *blk_debugfs_root;
55 
56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
61 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert);
62 
63 DEFINE_IDA(blk_queue_ida);
64 
65 /*
66  * For queue allocation
67  */
68 struct kmem_cache *blk_requestq_cachep;
69 struct kmem_cache *blk_requestq_srcu_cachep;
70 
71 /*
72  * Controlling structure to kblockd
73  */
74 static struct workqueue_struct *kblockd_workqueue;
75 
76 /**
77  * blk_queue_flag_set - atomically set a queue flag
78  * @flag: flag to be set
79  * @q: request queue
80  */
81 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
82 {
83 	set_bit(flag, &q->queue_flags);
84 }
85 EXPORT_SYMBOL(blk_queue_flag_set);
86 
87 /**
88  * blk_queue_flag_clear - atomically clear a queue flag
89  * @flag: flag to be cleared
90  * @q: request queue
91  */
92 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
93 {
94 	clear_bit(flag, &q->queue_flags);
95 }
96 EXPORT_SYMBOL(blk_queue_flag_clear);
97 
98 /**
99  * blk_queue_flag_test_and_set - atomically test and set a queue flag
100  * @flag: flag to be set
101  * @q: request queue
102  *
103  * Returns the previous value of @flag - 0 if the flag was not set and 1 if
104  * the flag was already set.
105  */
106 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
107 {
108 	return test_and_set_bit(flag, &q->queue_flags);
109 }
110 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
111 
112 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
113 static const char *const blk_op_name[] = {
114 	REQ_OP_NAME(READ),
115 	REQ_OP_NAME(WRITE),
116 	REQ_OP_NAME(FLUSH),
117 	REQ_OP_NAME(DISCARD),
118 	REQ_OP_NAME(SECURE_ERASE),
119 	REQ_OP_NAME(ZONE_RESET),
120 	REQ_OP_NAME(ZONE_RESET_ALL),
121 	REQ_OP_NAME(ZONE_OPEN),
122 	REQ_OP_NAME(ZONE_CLOSE),
123 	REQ_OP_NAME(ZONE_FINISH),
124 	REQ_OP_NAME(ZONE_APPEND),
125 	REQ_OP_NAME(WRITE_SAME),
126 	REQ_OP_NAME(WRITE_ZEROES),
127 	REQ_OP_NAME(DRV_IN),
128 	REQ_OP_NAME(DRV_OUT),
129 };
130 #undef REQ_OP_NAME
131 
132 /**
133  * blk_op_str - Return string XXX in the REQ_OP_XXX.
134  * @op: REQ_OP_XXX.
135  *
136  * Description: Centralize block layer function to convert REQ_OP_XXX into
137  * string format. Useful in the debugging and tracing bio or request. For
138  * invalid REQ_OP_XXX it returns string "UNKNOWN".
139  */
140 inline const char *blk_op_str(unsigned int op)
141 {
142 	const char *op_str = "UNKNOWN";
143 
144 	if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
145 		op_str = blk_op_name[op];
146 
147 	return op_str;
148 }
149 EXPORT_SYMBOL_GPL(blk_op_str);
150 
151 static const struct {
152 	int		errno;
153 	const char	*name;
154 } blk_errors[] = {
155 	[BLK_STS_OK]		= { 0,		"" },
156 	[BLK_STS_NOTSUPP]	= { -EOPNOTSUPP, "operation not supported" },
157 	[BLK_STS_TIMEOUT]	= { -ETIMEDOUT,	"timeout" },
158 	[BLK_STS_NOSPC]		= { -ENOSPC,	"critical space allocation" },
159 	[BLK_STS_TRANSPORT]	= { -ENOLINK,	"recoverable transport" },
160 	[BLK_STS_TARGET]	= { -EREMOTEIO,	"critical target" },
161 	[BLK_STS_NEXUS]		= { -EBADE,	"critical nexus" },
162 	[BLK_STS_MEDIUM]	= { -ENODATA,	"critical medium" },
163 	[BLK_STS_PROTECTION]	= { -EILSEQ,	"protection" },
164 	[BLK_STS_RESOURCE]	= { -ENOMEM,	"kernel resource" },
165 	[BLK_STS_DEV_RESOURCE]	= { -EBUSY,	"device resource" },
166 	[BLK_STS_AGAIN]		= { -EAGAIN,	"nonblocking retry" },
167 
168 	/* device mapper special case, should not leak out: */
169 	[BLK_STS_DM_REQUEUE]	= { -EREMCHG, "dm internal retry" },
170 
171 	/* zone device specific errors */
172 	[BLK_STS_ZONE_OPEN_RESOURCE]	= { -ETOOMANYREFS, "open zones exceeded" },
173 	[BLK_STS_ZONE_ACTIVE_RESOURCE]	= { -EOVERFLOW, "active zones exceeded" },
174 
175 	/* everything else not covered above: */
176 	[BLK_STS_IOERR]		= { -EIO,	"I/O" },
177 };
178 
179 blk_status_t errno_to_blk_status(int errno)
180 {
181 	int i;
182 
183 	for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
184 		if (blk_errors[i].errno == errno)
185 			return (__force blk_status_t)i;
186 	}
187 
188 	return BLK_STS_IOERR;
189 }
190 EXPORT_SYMBOL_GPL(errno_to_blk_status);
191 
192 int blk_status_to_errno(blk_status_t status)
193 {
194 	int idx = (__force int)status;
195 
196 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
197 		return -EIO;
198 	return blk_errors[idx].errno;
199 }
200 EXPORT_SYMBOL_GPL(blk_status_to_errno);
201 
202 const char *blk_status_to_str(blk_status_t status)
203 {
204 	int idx = (__force int)status;
205 
206 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
207 		return "<null>";
208 	return blk_errors[idx].name;
209 }
210 
211 /**
212  * blk_sync_queue - cancel any pending callbacks on a queue
213  * @q: the queue
214  *
215  * Description:
216  *     The block layer may perform asynchronous callback activity
217  *     on a queue, such as calling the unplug function after a timeout.
218  *     A block device may call blk_sync_queue to ensure that any
219  *     such activity is cancelled, thus allowing it to release resources
220  *     that the callbacks might use. The caller must already have made sure
221  *     that its ->submit_bio will not re-add plugging prior to calling
222  *     this function.
223  *
224  *     This function does not cancel any asynchronous activity arising
225  *     out of elevator or throttling code. That would require elevator_exit()
226  *     and blkcg_exit_queue() to be called with queue lock initialized.
227  *
228  */
229 void blk_sync_queue(struct request_queue *q)
230 {
231 	del_timer_sync(&q->timeout);
232 	cancel_work_sync(&q->timeout_work);
233 }
234 EXPORT_SYMBOL(blk_sync_queue);
235 
236 /**
237  * blk_set_pm_only - increment pm_only counter
238  * @q: request queue pointer
239  */
240 void blk_set_pm_only(struct request_queue *q)
241 {
242 	atomic_inc(&q->pm_only);
243 }
244 EXPORT_SYMBOL_GPL(blk_set_pm_only);
245 
246 void blk_clear_pm_only(struct request_queue *q)
247 {
248 	int pm_only;
249 
250 	pm_only = atomic_dec_return(&q->pm_only);
251 	WARN_ON_ONCE(pm_only < 0);
252 	if (pm_only == 0)
253 		wake_up_all(&q->mq_freeze_wq);
254 }
255 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
256 
257 /**
258  * blk_put_queue - decrement the request_queue refcount
259  * @q: the request_queue structure to decrement the refcount for
260  *
261  * Decrements the refcount of the request_queue kobject. When this reaches 0
262  * we'll have blk_release_queue() called.
263  *
264  * Context: Any context, but the last reference must not be dropped from
265  *          atomic context.
266  */
267 void blk_put_queue(struct request_queue *q)
268 {
269 	kobject_put(&q->kobj);
270 }
271 EXPORT_SYMBOL(blk_put_queue);
272 
273 void blk_queue_start_drain(struct request_queue *q)
274 {
275 	/*
276 	 * When queue DYING flag is set, we need to block new req
277 	 * entering queue, so we call blk_freeze_queue_start() to
278 	 * prevent I/O from crossing blk_queue_enter().
279 	 */
280 	blk_freeze_queue_start(q);
281 	if (queue_is_mq(q))
282 		blk_mq_wake_waiters(q);
283 	/* Make blk_queue_enter() reexamine the DYING flag. */
284 	wake_up_all(&q->mq_freeze_wq);
285 }
286 
287 /**
288  * blk_cleanup_queue - shutdown a request queue
289  * @q: request queue to shutdown
290  *
291  * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
292  * put it.  All future requests will be failed immediately with -ENODEV.
293  *
294  * Context: can sleep
295  */
296 void blk_cleanup_queue(struct request_queue *q)
297 {
298 	/* cannot be called from atomic context */
299 	might_sleep();
300 
301 	WARN_ON_ONCE(blk_queue_registered(q));
302 
303 	/* mark @q DYING, no new request or merges will be allowed afterwards */
304 	blk_queue_flag_set(QUEUE_FLAG_DYING, q);
305 	blk_queue_start_drain(q);
306 
307 	blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
308 	blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
309 
310 	/*
311 	 * Drain all requests queued before DYING marking. Set DEAD flag to
312 	 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
313 	 * after draining finished.
314 	 */
315 	blk_freeze_queue(q);
316 
317 	blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
318 
319 	blk_sync_queue(q);
320 	if (queue_is_mq(q)) {
321 		blk_mq_cancel_work_sync(q);
322 		blk_mq_exit_queue(q);
323 	}
324 
325 	/*
326 	 * In theory, request pool of sched_tags belongs to request queue.
327 	 * However, the current implementation requires tag_set for freeing
328 	 * requests, so free the pool now.
329 	 *
330 	 * Queue has become frozen, there can't be any in-queue requests, so
331 	 * it is safe to free requests now.
332 	 */
333 	mutex_lock(&q->sysfs_lock);
334 	if (q->elevator)
335 		blk_mq_sched_free_rqs(q);
336 	mutex_unlock(&q->sysfs_lock);
337 
338 	percpu_ref_exit(&q->q_usage_counter);
339 
340 	/* @q is and will stay empty, shutdown and put */
341 	blk_put_queue(q);
342 }
343 EXPORT_SYMBOL(blk_cleanup_queue);
344 
345 /**
346  * blk_queue_enter() - try to increase q->q_usage_counter
347  * @q: request queue pointer
348  * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
349  */
350 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
351 {
352 	const bool pm = flags & BLK_MQ_REQ_PM;
353 
354 	while (!blk_try_enter_queue(q, pm)) {
355 		if (flags & BLK_MQ_REQ_NOWAIT)
356 			return -EBUSY;
357 
358 		/*
359 		 * read pair of barrier in blk_freeze_queue_start(), we need to
360 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
361 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
362 		 * following wait may never return if the two reads are
363 		 * reordered.
364 		 */
365 		smp_rmb();
366 		wait_event(q->mq_freeze_wq,
367 			   (!q->mq_freeze_depth &&
368 			    blk_pm_resume_queue(pm, q)) ||
369 			   blk_queue_dying(q));
370 		if (blk_queue_dying(q))
371 			return -ENODEV;
372 	}
373 
374 	return 0;
375 }
376 
377 int __bio_queue_enter(struct request_queue *q, struct bio *bio)
378 {
379 	while (!blk_try_enter_queue(q, false)) {
380 		struct gendisk *disk = bio->bi_bdev->bd_disk;
381 
382 		if (bio->bi_opf & REQ_NOWAIT) {
383 			if (test_bit(GD_DEAD, &disk->state))
384 				goto dead;
385 			bio_wouldblock_error(bio);
386 			return -EBUSY;
387 		}
388 
389 		/*
390 		 * read pair of barrier in blk_freeze_queue_start(), we need to
391 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
392 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
393 		 * following wait may never return if the two reads are
394 		 * reordered.
395 		 */
396 		smp_rmb();
397 		wait_event(q->mq_freeze_wq,
398 			   (!q->mq_freeze_depth &&
399 			    blk_pm_resume_queue(false, q)) ||
400 			   test_bit(GD_DEAD, &disk->state));
401 		if (test_bit(GD_DEAD, &disk->state))
402 			goto dead;
403 	}
404 
405 	return 0;
406 dead:
407 	bio_io_error(bio);
408 	return -ENODEV;
409 }
410 
411 void blk_queue_exit(struct request_queue *q)
412 {
413 	percpu_ref_put(&q->q_usage_counter);
414 }
415 
416 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
417 {
418 	struct request_queue *q =
419 		container_of(ref, struct request_queue, q_usage_counter);
420 
421 	wake_up_all(&q->mq_freeze_wq);
422 }
423 
424 static void blk_rq_timed_out_timer(struct timer_list *t)
425 {
426 	struct request_queue *q = from_timer(q, t, timeout);
427 
428 	kblockd_schedule_work(&q->timeout_work);
429 }
430 
431 static void blk_timeout_work(struct work_struct *work)
432 {
433 }
434 
435 struct request_queue *blk_alloc_queue(int node_id, bool alloc_srcu)
436 {
437 	struct request_queue *q;
438 	int ret;
439 
440 	q = kmem_cache_alloc_node(blk_get_queue_kmem_cache(alloc_srcu),
441 			GFP_KERNEL | __GFP_ZERO, node_id);
442 	if (!q)
443 		return NULL;
444 
445 	if (alloc_srcu) {
446 		blk_queue_flag_set(QUEUE_FLAG_HAS_SRCU, q);
447 		if (init_srcu_struct(q->srcu) != 0)
448 			goto fail_q;
449 	}
450 
451 	q->last_merge = NULL;
452 
453 	q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
454 	if (q->id < 0)
455 		goto fail_srcu;
456 
457 	ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, 0);
458 	if (ret)
459 		goto fail_id;
460 
461 	q->stats = blk_alloc_queue_stats();
462 	if (!q->stats)
463 		goto fail_split;
464 
465 	q->node = node_id;
466 
467 	atomic_set(&q->nr_active_requests_shared_tags, 0);
468 
469 	timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
470 	INIT_WORK(&q->timeout_work, blk_timeout_work);
471 	INIT_LIST_HEAD(&q->icq_list);
472 #ifdef CONFIG_BLK_CGROUP
473 	INIT_LIST_HEAD(&q->blkg_list);
474 #endif
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 	if (blkcg_init_queue(q))
496 		goto fail_ref;
497 
498 	blk_queue_dma_alignment(q, 511);
499 	blk_set_default_limits(&q->limits);
500 	q->nr_requests = BLKDEV_DEFAULT_RQ;
501 
502 	return q;
503 
504 fail_ref:
505 	percpu_ref_exit(&q->q_usage_counter);
506 fail_stats:
507 	blk_free_queue_stats(q->stats);
508 fail_split:
509 	bioset_exit(&q->bio_split);
510 fail_id:
511 	ida_simple_remove(&blk_queue_ida, q->id);
512 fail_srcu:
513 	if (alloc_srcu)
514 		cleanup_srcu_struct(q->srcu);
515 fail_q:
516 	kmem_cache_free(blk_get_queue_kmem_cache(alloc_srcu), q);
517 	return NULL;
518 }
519 
520 /**
521  * blk_get_queue - increment the request_queue refcount
522  * @q: the request_queue structure to increment the refcount for
523  *
524  * Increment the refcount of the request_queue kobject.
525  *
526  * Context: Any context.
527  */
528 bool blk_get_queue(struct request_queue *q)
529 {
530 	if (likely(!blk_queue_dying(q))) {
531 		__blk_get_queue(q);
532 		return true;
533 	}
534 
535 	return false;
536 }
537 EXPORT_SYMBOL(blk_get_queue);
538 
539 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
540 {
541 	char b[BDEVNAME_SIZE];
542 
543 	pr_info_ratelimited("%s: attempt to access beyond end of device\n"
544 			    "%s: rw=%d, want=%llu, limit=%llu\n",
545 			    current->comm,
546 			    bio_devname(bio, b), bio->bi_opf,
547 			    bio_end_sector(bio), maxsector);
548 }
549 
550 #ifdef CONFIG_FAIL_MAKE_REQUEST
551 
552 static DECLARE_FAULT_ATTR(fail_make_request);
553 
554 static int __init setup_fail_make_request(char *str)
555 {
556 	return setup_fault_attr(&fail_make_request, str);
557 }
558 __setup("fail_make_request=", setup_fail_make_request);
559 
560 bool should_fail_request(struct block_device *part, unsigned int bytes)
561 {
562 	return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
563 }
564 
565 static int __init fail_make_request_debugfs(void)
566 {
567 	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
568 						NULL, &fail_make_request);
569 
570 	return PTR_ERR_OR_ZERO(dir);
571 }
572 
573 late_initcall(fail_make_request_debugfs);
574 #endif /* CONFIG_FAIL_MAKE_REQUEST */
575 
576 static inline bool bio_check_ro(struct bio *bio)
577 {
578 	if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
579 		char b[BDEVNAME_SIZE];
580 
581 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
582 			return false;
583 
584 		WARN_ONCE(1,
585 		       "Trying to write to read-only block-device %s (partno %d)\n",
586 			bio_devname(bio, b), bio->bi_bdev->bd_partno);
587 		/* Older lvm-tools actually trigger this */
588 		return false;
589 	}
590 
591 	return false;
592 }
593 
594 static noinline int should_fail_bio(struct bio *bio)
595 {
596 	if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
597 		return -EIO;
598 	return 0;
599 }
600 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
601 
602 /*
603  * Check whether this bio extends beyond the end of the device or partition.
604  * This may well happen - the kernel calls bread() without checking the size of
605  * the device, e.g., when mounting a file system.
606  */
607 static inline int bio_check_eod(struct bio *bio)
608 {
609 	sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
610 	unsigned int nr_sectors = bio_sectors(bio);
611 
612 	if (nr_sectors && maxsector &&
613 	    (nr_sectors > maxsector ||
614 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
615 		handle_bad_sector(bio, maxsector);
616 		return -EIO;
617 	}
618 	return 0;
619 }
620 
621 /*
622  * Remap block n of partition p to block n+start(p) of the disk.
623  */
624 static int blk_partition_remap(struct bio *bio)
625 {
626 	struct block_device *p = bio->bi_bdev;
627 
628 	if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
629 		return -EIO;
630 	if (bio_sectors(bio)) {
631 		bio->bi_iter.bi_sector += p->bd_start_sect;
632 		trace_block_bio_remap(bio, p->bd_dev,
633 				      bio->bi_iter.bi_sector -
634 				      p->bd_start_sect);
635 	}
636 	bio_set_flag(bio, BIO_REMAPPED);
637 	return 0;
638 }
639 
640 /*
641  * Check write append to a zoned block device.
642  */
643 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
644 						 struct bio *bio)
645 {
646 	sector_t pos = bio->bi_iter.bi_sector;
647 	int nr_sectors = bio_sectors(bio);
648 
649 	/* Only applicable to zoned block devices */
650 	if (!blk_queue_is_zoned(q))
651 		return BLK_STS_NOTSUPP;
652 
653 	/* The bio sector must point to the start of a sequential zone */
654 	if (pos & (blk_queue_zone_sectors(q) - 1) ||
655 	    !blk_queue_zone_is_seq(q, pos))
656 		return BLK_STS_IOERR;
657 
658 	/*
659 	 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
660 	 * split and could result in non-contiguous sectors being written in
661 	 * different zones.
662 	 */
663 	if (nr_sectors > q->limits.chunk_sectors)
664 		return BLK_STS_IOERR;
665 
666 	/* Make sure the BIO is small enough and will not get split */
667 	if (nr_sectors > q->limits.max_zone_append_sectors)
668 		return BLK_STS_IOERR;
669 
670 	bio->bi_opf |= REQ_NOMERGE;
671 
672 	return BLK_STS_OK;
673 }
674 
675 noinline_for_stack bool submit_bio_checks(struct bio *bio)
676 {
677 	struct block_device *bdev = bio->bi_bdev;
678 	struct request_queue *q = bdev_get_queue(bdev);
679 	blk_status_t status = BLK_STS_IOERR;
680 	struct blk_plug *plug;
681 
682 	might_sleep();
683 
684 	plug = blk_mq_plug(q, bio);
685 	if (plug && plug->nowait)
686 		bio->bi_opf |= REQ_NOWAIT;
687 
688 	/*
689 	 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
690 	 * if queue does not support NOWAIT.
691 	 */
692 	if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q))
693 		goto not_supported;
694 
695 	if (should_fail_bio(bio))
696 		goto end_io;
697 	if (unlikely(bio_check_ro(bio)))
698 		goto end_io;
699 	if (!bio_flagged(bio, BIO_REMAPPED)) {
700 		if (unlikely(bio_check_eod(bio)))
701 			goto end_io;
702 		if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
703 			goto end_io;
704 	}
705 
706 	/*
707 	 * Filter flush bio's early so that bio based drivers without flush
708 	 * support don't have to worry about them.
709 	 */
710 	if (op_is_flush(bio->bi_opf) &&
711 	    !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
712 		bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
713 		if (!bio_sectors(bio)) {
714 			status = BLK_STS_OK;
715 			goto end_io;
716 		}
717 	}
718 
719 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
720 		bio_clear_polled(bio);
721 
722 	switch (bio_op(bio)) {
723 	case REQ_OP_DISCARD:
724 		if (!blk_queue_discard(q))
725 			goto not_supported;
726 		break;
727 	case REQ_OP_SECURE_ERASE:
728 		if (!blk_queue_secure_erase(q))
729 			goto not_supported;
730 		break;
731 	case REQ_OP_WRITE_SAME:
732 		if (!q->limits.max_write_same_sectors)
733 			goto not_supported;
734 		break;
735 	case REQ_OP_ZONE_APPEND:
736 		status = blk_check_zone_append(q, bio);
737 		if (status != BLK_STS_OK)
738 			goto end_io;
739 		break;
740 	case REQ_OP_ZONE_RESET:
741 	case REQ_OP_ZONE_OPEN:
742 	case REQ_OP_ZONE_CLOSE:
743 	case REQ_OP_ZONE_FINISH:
744 		if (!blk_queue_is_zoned(q))
745 			goto not_supported;
746 		break;
747 	case REQ_OP_ZONE_RESET_ALL:
748 		if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
749 			goto not_supported;
750 		break;
751 	case REQ_OP_WRITE_ZEROES:
752 		if (!q->limits.max_write_zeroes_sectors)
753 			goto not_supported;
754 		break;
755 	default:
756 		break;
757 	}
758 
759 	if (blk_throtl_bio(bio))
760 		return false;
761 
762 	blk_cgroup_bio_start(bio);
763 	blkcg_bio_issue_init(bio);
764 
765 	if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
766 		trace_block_bio_queue(bio);
767 		/* Now that enqueuing has been traced, we need to trace
768 		 * completion as well.
769 		 */
770 		bio_set_flag(bio, BIO_TRACE_COMPLETION);
771 	}
772 	return true;
773 
774 not_supported:
775 	status = BLK_STS_NOTSUPP;
776 end_io:
777 	bio->bi_status = status;
778 	bio_endio(bio);
779 	return false;
780 }
781 
782 static void __submit_bio_fops(struct gendisk *disk, struct bio *bio)
783 {
784 	if (blk_crypto_bio_prep(&bio)) {
785 		if (likely(bio_queue_enter(bio) == 0)) {
786 			disk->fops->submit_bio(bio);
787 			blk_queue_exit(disk->queue);
788 		}
789 	}
790 }
791 
792 static void __submit_bio(struct bio *bio)
793 {
794 	struct gendisk *disk = bio->bi_bdev->bd_disk;
795 
796 	if (unlikely(!submit_bio_checks(bio)))
797 		return;
798 
799 	if (!disk->fops->submit_bio)
800 		blk_mq_submit_bio(bio);
801 	else
802 		__submit_bio_fops(disk, bio);
803 }
804 
805 /*
806  * The loop in this function may be a bit non-obvious, and so deserves some
807  * explanation:
808  *
809  *  - Before entering the loop, bio->bi_next is NULL (as all callers ensure
810  *    that), so we have a list with a single bio.
811  *  - We pretend that we have just taken it off a longer list, so we assign
812  *    bio_list to a pointer to the bio_list_on_stack, thus initialising the
813  *    bio_list of new bios to be added.  ->submit_bio() may indeed add some more
814  *    bios through a recursive call to submit_bio_noacct.  If it did, we find a
815  *    non-NULL value in bio_list and re-enter the loop from the top.
816  *  - In this case we really did just take the bio of the top of the list (no
817  *    pretending) and so remove it from bio_list, and call into ->submit_bio()
818  *    again.
819  *
820  * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
821  * bio_list_on_stack[1] contains bios that were submitted before the current
822  *	->submit_bio_bio, but that haven't been processed yet.
823  */
824 static void __submit_bio_noacct(struct bio *bio)
825 {
826 	struct bio_list bio_list_on_stack[2];
827 
828 	BUG_ON(bio->bi_next);
829 
830 	bio_list_init(&bio_list_on_stack[0]);
831 	current->bio_list = bio_list_on_stack;
832 
833 	do {
834 		struct request_queue *q = bdev_get_queue(bio->bi_bdev);
835 		struct bio_list lower, same;
836 
837 		/*
838 		 * Create a fresh bio_list for all subordinate requests.
839 		 */
840 		bio_list_on_stack[1] = bio_list_on_stack[0];
841 		bio_list_init(&bio_list_on_stack[0]);
842 
843 		__submit_bio(bio);
844 
845 		/*
846 		 * Sort new bios into those for a lower level and those for the
847 		 * same level.
848 		 */
849 		bio_list_init(&lower);
850 		bio_list_init(&same);
851 		while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
852 			if (q == bdev_get_queue(bio->bi_bdev))
853 				bio_list_add(&same, bio);
854 			else
855 				bio_list_add(&lower, bio);
856 
857 		/*
858 		 * Now assemble so we handle the lowest level first.
859 		 */
860 		bio_list_merge(&bio_list_on_stack[0], &lower);
861 		bio_list_merge(&bio_list_on_stack[0], &same);
862 		bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
863 	} while ((bio = bio_list_pop(&bio_list_on_stack[0])));
864 
865 	current->bio_list = NULL;
866 }
867 
868 static void __submit_bio_noacct_mq(struct bio *bio)
869 {
870 	struct bio_list bio_list[2] = { };
871 
872 	current->bio_list = bio_list;
873 
874 	do {
875 		__submit_bio(bio);
876 	} while ((bio = bio_list_pop(&bio_list[0])));
877 
878 	current->bio_list = NULL;
879 }
880 
881 /**
882  * submit_bio_noacct - re-submit a bio to the block device layer for I/O
883  * @bio:  The bio describing the location in memory and on the device.
884  *
885  * This is a version of submit_bio() that shall only be used for I/O that is
886  * resubmitted to lower level drivers by stacking block drivers.  All file
887  * systems and other upper level users of the block layer should use
888  * submit_bio() instead.
889  */
890 void submit_bio_noacct(struct bio *bio)
891 {
892 	/*
893 	 * We only want one ->submit_bio to be active at a time, else stack
894 	 * usage with stacked devices could be a problem.  Use current->bio_list
895 	 * to collect a list of requests submited by a ->submit_bio method while
896 	 * it is active, and then process them after it returned.
897 	 */
898 	if (current->bio_list)
899 		bio_list_add(&current->bio_list[0], bio);
900 	else if (!bio->bi_bdev->bd_disk->fops->submit_bio)
901 		__submit_bio_noacct_mq(bio);
902 	else
903 		__submit_bio_noacct(bio);
904 }
905 EXPORT_SYMBOL(submit_bio_noacct);
906 
907 /**
908  * submit_bio - submit a bio to the block device layer for I/O
909  * @bio: The &struct bio which describes the I/O
910  *
911  * submit_bio() is used to submit I/O requests to block devices.  It is passed a
912  * fully set up &struct bio that describes the I/O that needs to be done.  The
913  * bio will be send to the device described by the bi_bdev field.
914  *
915  * The success/failure status of the request, along with notification of
916  * completion, is delivered asynchronously through the ->bi_end_io() callback
917  * in @bio.  The bio must NOT be touched by thecaller until ->bi_end_io() has
918  * been called.
919  */
920 void submit_bio(struct bio *bio)
921 {
922 	if (blkcg_punt_bio_submit(bio))
923 		return;
924 
925 	/*
926 	 * If it's a regular read/write or a barrier with data attached,
927 	 * go through the normal accounting stuff before submission.
928 	 */
929 	if (bio_has_data(bio)) {
930 		unsigned int count;
931 
932 		if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
933 			count = queue_logical_block_size(
934 					bdev_get_queue(bio->bi_bdev)) >> 9;
935 		else
936 			count = bio_sectors(bio);
937 
938 		if (op_is_write(bio_op(bio))) {
939 			count_vm_events(PGPGOUT, count);
940 		} else {
941 			task_io_account_read(bio->bi_iter.bi_size);
942 			count_vm_events(PGPGIN, count);
943 		}
944 	}
945 
946 	/*
947 	 * If we're reading data that is part of the userspace workingset, count
948 	 * submission time as memory stall.  When the device is congested, or
949 	 * the submitting cgroup IO-throttled, submission can be a significant
950 	 * part of overall IO time.
951 	 */
952 	if (unlikely(bio_op(bio) == REQ_OP_READ &&
953 	    bio_flagged(bio, BIO_WORKINGSET))) {
954 		unsigned long pflags;
955 
956 		psi_memstall_enter(&pflags);
957 		submit_bio_noacct(bio);
958 		psi_memstall_leave(&pflags);
959 		return;
960 	}
961 
962 	submit_bio_noacct(bio);
963 }
964 EXPORT_SYMBOL(submit_bio);
965 
966 /**
967  * bio_poll - poll for BIO completions
968  * @bio: bio to poll for
969  * @iob: batches of IO
970  * @flags: BLK_POLL_* flags that control the behavior
971  *
972  * Poll for completions on queue associated with the bio. Returns number of
973  * completed entries found.
974  *
975  * Note: the caller must either be the context that submitted @bio, or
976  * be in a RCU critical section to prevent freeing of @bio.
977  */
978 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
979 {
980 	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
981 	blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
982 	int ret;
983 
984 	if (cookie == BLK_QC_T_NONE ||
985 	    !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
986 		return 0;
987 
988 	if (current->plug)
989 		blk_flush_plug(current->plug, false);
990 
991 	if (blk_queue_enter(q, BLK_MQ_REQ_NOWAIT))
992 		return 0;
993 	if (WARN_ON_ONCE(!queue_is_mq(q)))
994 		ret = 0;	/* not yet implemented, should not happen */
995 	else
996 		ret = blk_mq_poll(q, cookie, iob, flags);
997 	blk_queue_exit(q);
998 	return ret;
999 }
1000 EXPORT_SYMBOL_GPL(bio_poll);
1001 
1002 /*
1003  * Helper to implement file_operations.iopoll.  Requires the bio to be stored
1004  * in iocb->private, and cleared before freeing the bio.
1005  */
1006 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
1007 		    unsigned int flags)
1008 {
1009 	struct bio *bio;
1010 	int ret = 0;
1011 
1012 	/*
1013 	 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
1014 	 * point to a freshly allocated bio at this point.  If that happens
1015 	 * we have a few cases to consider:
1016 	 *
1017 	 *  1) the bio is beeing initialized and bi_bdev is NULL.  We can just
1018 	 *     simply nothing in this case
1019 	 *  2) the bio points to a not poll enabled device.  bio_poll will catch
1020 	 *     this and return 0
1021 	 *  3) the bio points to a poll capable device, including but not
1022 	 *     limited to the one that the original bio pointed to.  In this
1023 	 *     case we will call into the actual poll method and poll for I/O,
1024 	 *     even if we don't need to, but it won't cause harm either.
1025 	 *
1026 	 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
1027 	 * is still allocated. Because partitions hold a reference to the whole
1028 	 * device bdev and thus disk, the disk is also still valid.  Grabbing
1029 	 * a reference to the queue in bio_poll() ensures the hctxs and requests
1030 	 * are still valid as well.
1031 	 */
1032 	rcu_read_lock();
1033 	bio = READ_ONCE(kiocb->private);
1034 	if (bio && bio->bi_bdev)
1035 		ret = bio_poll(bio, iob, flags);
1036 	rcu_read_unlock();
1037 
1038 	return ret;
1039 }
1040 EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
1041 
1042 void update_io_ticks(struct block_device *part, unsigned long now, bool end)
1043 {
1044 	unsigned long stamp;
1045 again:
1046 	stamp = READ_ONCE(part->bd_stamp);
1047 	if (unlikely(time_after(now, stamp))) {
1048 		if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp))
1049 			__part_stat_add(part, io_ticks, end ? now - stamp : 1);
1050 	}
1051 	if (part->bd_partno) {
1052 		part = bdev_whole(part);
1053 		goto again;
1054 	}
1055 }
1056 
1057 static unsigned long __part_start_io_acct(struct block_device *part,
1058 					  unsigned int sectors, unsigned int op,
1059 					  unsigned long start_time)
1060 {
1061 	const int sgrp = op_stat_group(op);
1062 
1063 	part_stat_lock();
1064 	update_io_ticks(part, start_time, false);
1065 	part_stat_inc(part, ios[sgrp]);
1066 	part_stat_add(part, sectors[sgrp], sectors);
1067 	part_stat_local_inc(part, in_flight[op_is_write(op)]);
1068 	part_stat_unlock();
1069 
1070 	return start_time;
1071 }
1072 
1073 /**
1074  * bio_start_io_acct_time - start I/O accounting for bio based drivers
1075  * @bio:	bio to start account for
1076  * @start_time:	start time that should be passed back to bio_end_io_acct().
1077  */
1078 void bio_start_io_acct_time(struct bio *bio, unsigned long start_time)
1079 {
1080 	__part_start_io_acct(bio->bi_bdev, bio_sectors(bio),
1081 			     bio_op(bio), start_time);
1082 }
1083 EXPORT_SYMBOL_GPL(bio_start_io_acct_time);
1084 
1085 /**
1086  * bio_start_io_acct - start I/O accounting for bio based drivers
1087  * @bio:	bio to start account for
1088  *
1089  * Returns the start time that should be passed back to bio_end_io_acct().
1090  */
1091 unsigned long bio_start_io_acct(struct bio *bio)
1092 {
1093 	return __part_start_io_acct(bio->bi_bdev, bio_sectors(bio),
1094 				    bio_op(bio), jiffies);
1095 }
1096 EXPORT_SYMBOL_GPL(bio_start_io_acct);
1097 
1098 unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
1099 				 unsigned int op)
1100 {
1101 	return __part_start_io_acct(disk->part0, sectors, op, jiffies);
1102 }
1103 EXPORT_SYMBOL(disk_start_io_acct);
1104 
1105 static void __part_end_io_acct(struct block_device *part, unsigned int op,
1106 			       unsigned long start_time)
1107 {
1108 	const int sgrp = op_stat_group(op);
1109 	unsigned long now = READ_ONCE(jiffies);
1110 	unsigned long duration = now - start_time;
1111 
1112 	part_stat_lock();
1113 	update_io_ticks(part, now, true);
1114 	part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1115 	part_stat_local_dec(part, in_flight[op_is_write(op)]);
1116 	part_stat_unlock();
1117 }
1118 
1119 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
1120 		struct block_device *orig_bdev)
1121 {
1122 	__part_end_io_acct(orig_bdev, bio_op(bio), start_time);
1123 }
1124 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1125 
1126 void disk_end_io_acct(struct gendisk *disk, unsigned int op,
1127 		      unsigned long start_time)
1128 {
1129 	__part_end_io_acct(disk->part0, op, start_time);
1130 }
1131 EXPORT_SYMBOL(disk_end_io_acct);
1132 
1133 /**
1134  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1135  * @q : the queue of the device being checked
1136  *
1137  * Description:
1138  *    Check if underlying low-level drivers of a device are busy.
1139  *    If the drivers want to export their busy state, they must set own
1140  *    exporting function using blk_queue_lld_busy() first.
1141  *
1142  *    Basically, this function is used only by request stacking drivers
1143  *    to stop dispatching requests to underlying devices when underlying
1144  *    devices are busy.  This behavior helps more I/O merging on the queue
1145  *    of the request stacking driver and prevents I/O throughput regression
1146  *    on burst I/O load.
1147  *
1148  * Return:
1149  *    0 - Not busy (The request stacking driver should dispatch request)
1150  *    1 - Busy (The request stacking driver should stop dispatching request)
1151  */
1152 int blk_lld_busy(struct request_queue *q)
1153 {
1154 	if (queue_is_mq(q) && q->mq_ops->busy)
1155 		return q->mq_ops->busy(q);
1156 
1157 	return 0;
1158 }
1159 EXPORT_SYMBOL_GPL(blk_lld_busy);
1160 
1161 int kblockd_schedule_work(struct work_struct *work)
1162 {
1163 	return queue_work(kblockd_workqueue, work);
1164 }
1165 EXPORT_SYMBOL(kblockd_schedule_work);
1166 
1167 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1168 				unsigned long delay)
1169 {
1170 	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1171 }
1172 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1173 
1174 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1175 {
1176 	struct task_struct *tsk = current;
1177 
1178 	/*
1179 	 * If this is a nested plug, don't actually assign it.
1180 	 */
1181 	if (tsk->plug)
1182 		return;
1183 
1184 	plug->mq_list = NULL;
1185 	plug->cached_rq = NULL;
1186 	plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1187 	plug->rq_count = 0;
1188 	plug->multiple_queues = false;
1189 	plug->has_elevator = false;
1190 	plug->nowait = false;
1191 	INIT_LIST_HEAD(&plug->cb_list);
1192 
1193 	/*
1194 	 * Store ordering should not be needed here, since a potential
1195 	 * preempt will imply a full memory barrier
1196 	 */
1197 	tsk->plug = plug;
1198 }
1199 
1200 /**
1201  * blk_start_plug - initialize blk_plug and track it inside the task_struct
1202  * @plug:	The &struct blk_plug that needs to be initialized
1203  *
1204  * Description:
1205  *   blk_start_plug() indicates to the block layer an intent by the caller
1206  *   to submit multiple I/O requests in a batch.  The block layer may use
1207  *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1208  *   is called.  However, the block layer may choose to submit requests
1209  *   before a call to blk_finish_plug() if the number of queued I/Os
1210  *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1211  *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1212  *   the task schedules (see below).
1213  *
1214  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1215  *   pending I/O should the task end up blocking between blk_start_plug() and
1216  *   blk_finish_plug(). This is important from a performance perspective, but
1217  *   also ensures that we don't deadlock. For instance, if the task is blocking
1218  *   for a memory allocation, memory reclaim could end up wanting to free a
1219  *   page belonging to that request that is currently residing in our private
1220  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1221  *   this kind of deadlock.
1222  */
1223 void blk_start_plug(struct blk_plug *plug)
1224 {
1225 	blk_start_plug_nr_ios(plug, 1);
1226 }
1227 EXPORT_SYMBOL(blk_start_plug);
1228 
1229 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1230 {
1231 	LIST_HEAD(callbacks);
1232 
1233 	while (!list_empty(&plug->cb_list)) {
1234 		list_splice_init(&plug->cb_list, &callbacks);
1235 
1236 		while (!list_empty(&callbacks)) {
1237 			struct blk_plug_cb *cb = list_first_entry(&callbacks,
1238 							  struct blk_plug_cb,
1239 							  list);
1240 			list_del(&cb->list);
1241 			cb->callback(cb, from_schedule);
1242 		}
1243 	}
1244 }
1245 
1246 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1247 				      int size)
1248 {
1249 	struct blk_plug *plug = current->plug;
1250 	struct blk_plug_cb *cb;
1251 
1252 	if (!plug)
1253 		return NULL;
1254 
1255 	list_for_each_entry(cb, &plug->cb_list, list)
1256 		if (cb->callback == unplug && cb->data == data)
1257 			return cb;
1258 
1259 	/* Not currently on the callback list */
1260 	BUG_ON(size < sizeof(*cb));
1261 	cb = kzalloc(size, GFP_ATOMIC);
1262 	if (cb) {
1263 		cb->data = data;
1264 		cb->callback = unplug;
1265 		list_add(&cb->list, &plug->cb_list);
1266 	}
1267 	return cb;
1268 }
1269 EXPORT_SYMBOL(blk_check_plugged);
1270 
1271 void blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1272 {
1273 	if (!list_empty(&plug->cb_list))
1274 		flush_plug_callbacks(plug, from_schedule);
1275 	if (!rq_list_empty(plug->mq_list))
1276 		blk_mq_flush_plug_list(plug, from_schedule);
1277 	/*
1278 	 * Unconditionally flush out cached requests, even if the unplug
1279 	 * event came from schedule. Since we know hold references to the
1280 	 * queue for cached requests, we don't want a blocked task holding
1281 	 * up a queue freeze/quiesce event.
1282 	 */
1283 	if (unlikely(!rq_list_empty(plug->cached_rq)))
1284 		blk_mq_free_plug_rqs(plug);
1285 }
1286 
1287 /**
1288  * blk_finish_plug - mark the end of a batch of submitted I/O
1289  * @plug:	The &struct blk_plug passed to blk_start_plug()
1290  *
1291  * Description:
1292  * Indicate that a batch of I/O submissions is complete.  This function
1293  * must be paired with an initial call to blk_start_plug().  The intent
1294  * is to allow the block layer to optimize I/O submission.  See the
1295  * documentation for blk_start_plug() for more information.
1296  */
1297 void blk_finish_plug(struct blk_plug *plug)
1298 {
1299 	if (plug == current->plug) {
1300 		blk_flush_plug(plug, false);
1301 		current->plug = NULL;
1302 	}
1303 }
1304 EXPORT_SYMBOL(blk_finish_plug);
1305 
1306 void blk_io_schedule(void)
1307 {
1308 	/* Prevent hang_check timer from firing at us during very long I/O */
1309 	unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1310 
1311 	if (timeout)
1312 		io_schedule_timeout(timeout);
1313 	else
1314 		io_schedule();
1315 }
1316 EXPORT_SYMBOL_GPL(blk_io_schedule);
1317 
1318 int __init blk_dev_init(void)
1319 {
1320 	BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1321 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1322 			sizeof_field(struct request, cmd_flags));
1323 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1324 			sizeof_field(struct bio, bi_opf));
1325 	BUILD_BUG_ON(ALIGN(offsetof(struct request_queue, srcu),
1326 			   __alignof__(struct request_queue)) !=
1327 		     sizeof(struct request_queue));
1328 
1329 	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
1330 	kblockd_workqueue = alloc_workqueue("kblockd",
1331 					    WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1332 	if (!kblockd_workqueue)
1333 		panic("Failed to create kblockd\n");
1334 
1335 	blk_requestq_cachep = kmem_cache_create("request_queue",
1336 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1337 
1338 	blk_requestq_srcu_cachep = kmem_cache_create("request_queue_srcu",
1339 			sizeof(struct request_queue) +
1340 			sizeof(struct srcu_struct), 0, SLAB_PANIC, NULL);
1341 
1342 	blk_debugfs_root = debugfs_create_dir("block", NULL);
1343 
1344 	return 0;
1345 }
1346