xref: /openbmc/linux/block/blk-core.c (revision b7b3c35e)
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 void blk_set_queue_dying(struct request_queue *q)
288 {
289 	blk_queue_flag_set(QUEUE_FLAG_DYING, q);
290 	blk_queue_start_drain(q);
291 }
292 EXPORT_SYMBOL_GPL(blk_set_queue_dying);
293 
294 /**
295  * blk_cleanup_queue - shutdown a request queue
296  * @q: request queue to shutdown
297  *
298  * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
299  * put it.  All future requests will be failed immediately with -ENODEV.
300  *
301  * Context: can sleep
302  */
303 void blk_cleanup_queue(struct request_queue *q)
304 {
305 	/* cannot be called from atomic context */
306 	might_sleep();
307 
308 	WARN_ON_ONCE(blk_queue_registered(q));
309 
310 	/* mark @q DYING, no new request or merges will be allowed afterwards */
311 	blk_set_queue_dying(q);
312 
313 	blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
314 	blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
315 
316 	/*
317 	 * Drain all requests queued before DYING marking. Set DEAD flag to
318 	 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
319 	 * after draining finished.
320 	 */
321 	blk_freeze_queue(q);
322 
323 	blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
324 
325 	blk_sync_queue(q);
326 	if (queue_is_mq(q)) {
327 		blk_mq_cancel_work_sync(q);
328 		blk_mq_exit_queue(q);
329 	}
330 
331 	/*
332 	 * In theory, request pool of sched_tags belongs to request queue.
333 	 * However, the current implementation requires tag_set for freeing
334 	 * requests, so free the pool now.
335 	 *
336 	 * Queue has become frozen, there can't be any in-queue requests, so
337 	 * it is safe to free requests now.
338 	 */
339 	mutex_lock(&q->sysfs_lock);
340 	if (q->elevator)
341 		blk_mq_sched_free_rqs(q);
342 	mutex_unlock(&q->sysfs_lock);
343 
344 	percpu_ref_exit(&q->q_usage_counter);
345 
346 	/* @q is and will stay empty, shutdown and put */
347 	blk_put_queue(q);
348 }
349 EXPORT_SYMBOL(blk_cleanup_queue);
350 
351 /**
352  * blk_queue_enter() - try to increase q->q_usage_counter
353  * @q: request queue pointer
354  * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
355  */
356 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
357 {
358 	const bool pm = flags & BLK_MQ_REQ_PM;
359 
360 	while (!blk_try_enter_queue(q, pm)) {
361 		if (flags & BLK_MQ_REQ_NOWAIT)
362 			return -EBUSY;
363 
364 		/*
365 		 * read pair of barrier in blk_freeze_queue_start(), we need to
366 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
367 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
368 		 * following wait may never return if the two reads are
369 		 * reordered.
370 		 */
371 		smp_rmb();
372 		wait_event(q->mq_freeze_wq,
373 			   (!q->mq_freeze_depth &&
374 			    blk_pm_resume_queue(pm, q)) ||
375 			   blk_queue_dying(q));
376 		if (blk_queue_dying(q))
377 			return -ENODEV;
378 	}
379 
380 	return 0;
381 }
382 
383 int __bio_queue_enter(struct request_queue *q, struct bio *bio)
384 {
385 	while (!blk_try_enter_queue(q, false)) {
386 		struct gendisk *disk = bio->bi_bdev->bd_disk;
387 
388 		if (bio->bi_opf & REQ_NOWAIT) {
389 			if (test_bit(GD_DEAD, &disk->state))
390 				goto dead;
391 			bio_wouldblock_error(bio);
392 			return -EBUSY;
393 		}
394 
395 		/*
396 		 * read pair of barrier in blk_freeze_queue_start(), we need to
397 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
398 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
399 		 * following wait may never return if the two reads are
400 		 * reordered.
401 		 */
402 		smp_rmb();
403 		wait_event(q->mq_freeze_wq,
404 			   (!q->mq_freeze_depth &&
405 			    blk_pm_resume_queue(false, q)) ||
406 			   test_bit(GD_DEAD, &disk->state));
407 		if (test_bit(GD_DEAD, &disk->state))
408 			goto dead;
409 	}
410 
411 	return 0;
412 dead:
413 	bio_io_error(bio);
414 	return -ENODEV;
415 }
416 
417 void blk_queue_exit(struct request_queue *q)
418 {
419 	percpu_ref_put(&q->q_usage_counter);
420 }
421 
422 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
423 {
424 	struct request_queue *q =
425 		container_of(ref, struct request_queue, q_usage_counter);
426 
427 	wake_up_all(&q->mq_freeze_wq);
428 }
429 
430 static void blk_rq_timed_out_timer(struct timer_list *t)
431 {
432 	struct request_queue *q = from_timer(q, t, timeout);
433 
434 	kblockd_schedule_work(&q->timeout_work);
435 }
436 
437 static void blk_timeout_work(struct work_struct *work)
438 {
439 }
440 
441 struct request_queue *blk_alloc_queue(int node_id, bool alloc_srcu)
442 {
443 	struct request_queue *q;
444 	int ret;
445 
446 	q = kmem_cache_alloc_node(blk_get_queue_kmem_cache(alloc_srcu),
447 			GFP_KERNEL | __GFP_ZERO, node_id);
448 	if (!q)
449 		return NULL;
450 
451 	if (alloc_srcu) {
452 		blk_queue_flag_set(QUEUE_FLAG_HAS_SRCU, q);
453 		if (init_srcu_struct(q->srcu) != 0)
454 			goto fail_q;
455 	}
456 
457 	q->last_merge = NULL;
458 
459 	q->id = ida_simple_get(&blk_queue_ida, 0, 0, GFP_KERNEL);
460 	if (q->id < 0)
461 		goto fail_srcu;
462 
463 	ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, 0);
464 	if (ret)
465 		goto fail_id;
466 
467 	q->stats = blk_alloc_queue_stats();
468 	if (!q->stats)
469 		goto fail_split;
470 
471 	q->node = node_id;
472 
473 	atomic_set(&q->nr_active_requests_shared_tags, 0);
474 
475 	timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
476 	INIT_WORK(&q->timeout_work, blk_timeout_work);
477 	INIT_LIST_HEAD(&q->icq_list);
478 #ifdef CONFIG_BLK_CGROUP
479 	INIT_LIST_HEAD(&q->blkg_list);
480 #endif
481 
482 	kobject_init(&q->kobj, &blk_queue_ktype);
483 
484 	mutex_init(&q->debugfs_mutex);
485 	mutex_init(&q->sysfs_lock);
486 	mutex_init(&q->sysfs_dir_lock);
487 	spin_lock_init(&q->queue_lock);
488 
489 	init_waitqueue_head(&q->mq_freeze_wq);
490 	mutex_init(&q->mq_freeze_lock);
491 
492 	/*
493 	 * Init percpu_ref in atomic mode so that it's faster to shutdown.
494 	 * See blk_register_queue() for details.
495 	 */
496 	if (percpu_ref_init(&q->q_usage_counter,
497 				blk_queue_usage_counter_release,
498 				PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
499 		goto fail_stats;
500 
501 	if (blkcg_init_queue(q))
502 		goto fail_ref;
503 
504 	blk_queue_dma_alignment(q, 511);
505 	blk_set_default_limits(&q->limits);
506 	q->nr_requests = BLKDEV_DEFAULT_RQ;
507 
508 	return q;
509 
510 fail_ref:
511 	percpu_ref_exit(&q->q_usage_counter);
512 fail_stats:
513 	blk_free_queue_stats(q->stats);
514 fail_split:
515 	bioset_exit(&q->bio_split);
516 fail_id:
517 	ida_simple_remove(&blk_queue_ida, q->id);
518 fail_srcu:
519 	if (alloc_srcu)
520 		cleanup_srcu_struct(q->srcu);
521 fail_q:
522 	kmem_cache_free(blk_get_queue_kmem_cache(alloc_srcu), q);
523 	return NULL;
524 }
525 
526 /**
527  * blk_get_queue - increment the request_queue refcount
528  * @q: the request_queue structure to increment the refcount for
529  *
530  * Increment the refcount of the request_queue kobject.
531  *
532  * Context: Any context.
533  */
534 bool blk_get_queue(struct request_queue *q)
535 {
536 	if (likely(!blk_queue_dying(q))) {
537 		__blk_get_queue(q);
538 		return true;
539 	}
540 
541 	return false;
542 }
543 EXPORT_SYMBOL(blk_get_queue);
544 
545 static void handle_bad_sector(struct bio *bio, sector_t maxsector)
546 {
547 	char b[BDEVNAME_SIZE];
548 
549 	pr_info_ratelimited("%s: attempt to access beyond end of device\n"
550 			    "%s: rw=%d, want=%llu, limit=%llu\n",
551 			    current->comm,
552 			    bio_devname(bio, b), bio->bi_opf,
553 			    bio_end_sector(bio), maxsector);
554 }
555 
556 #ifdef CONFIG_FAIL_MAKE_REQUEST
557 
558 static DECLARE_FAULT_ATTR(fail_make_request);
559 
560 static int __init setup_fail_make_request(char *str)
561 {
562 	return setup_fault_attr(&fail_make_request, str);
563 }
564 __setup("fail_make_request=", setup_fail_make_request);
565 
566 bool should_fail_request(struct block_device *part, unsigned int bytes)
567 {
568 	return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
569 }
570 
571 static int __init fail_make_request_debugfs(void)
572 {
573 	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
574 						NULL, &fail_make_request);
575 
576 	return PTR_ERR_OR_ZERO(dir);
577 }
578 
579 late_initcall(fail_make_request_debugfs);
580 #endif /* CONFIG_FAIL_MAKE_REQUEST */
581 
582 static inline bool bio_check_ro(struct bio *bio)
583 {
584 	if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
585 		char b[BDEVNAME_SIZE];
586 
587 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
588 			return false;
589 
590 		WARN_ONCE(1,
591 		       "Trying to write to read-only block-device %s (partno %d)\n",
592 			bio_devname(bio, b), bio->bi_bdev->bd_partno);
593 		/* Older lvm-tools actually trigger this */
594 		return false;
595 	}
596 
597 	return false;
598 }
599 
600 static noinline int should_fail_bio(struct bio *bio)
601 {
602 	if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
603 		return -EIO;
604 	return 0;
605 }
606 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
607 
608 /*
609  * Check whether this bio extends beyond the end of the device or partition.
610  * This may well happen - the kernel calls bread() without checking the size of
611  * the device, e.g., when mounting a file system.
612  */
613 static inline int bio_check_eod(struct bio *bio)
614 {
615 	sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
616 	unsigned int nr_sectors = bio_sectors(bio);
617 
618 	if (nr_sectors && maxsector &&
619 	    (nr_sectors > maxsector ||
620 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
621 		handle_bad_sector(bio, maxsector);
622 		return -EIO;
623 	}
624 	return 0;
625 }
626 
627 /*
628  * Remap block n of partition p to block n+start(p) of the disk.
629  */
630 static int blk_partition_remap(struct bio *bio)
631 {
632 	struct block_device *p = bio->bi_bdev;
633 
634 	if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
635 		return -EIO;
636 	if (bio_sectors(bio)) {
637 		bio->bi_iter.bi_sector += p->bd_start_sect;
638 		trace_block_bio_remap(bio, p->bd_dev,
639 				      bio->bi_iter.bi_sector -
640 				      p->bd_start_sect);
641 	}
642 	bio_set_flag(bio, BIO_REMAPPED);
643 	return 0;
644 }
645 
646 /*
647  * Check write append to a zoned block device.
648  */
649 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
650 						 struct bio *bio)
651 {
652 	sector_t pos = bio->bi_iter.bi_sector;
653 	int nr_sectors = bio_sectors(bio);
654 
655 	/* Only applicable to zoned block devices */
656 	if (!blk_queue_is_zoned(q))
657 		return BLK_STS_NOTSUPP;
658 
659 	/* The bio sector must point to the start of a sequential zone */
660 	if (pos & (blk_queue_zone_sectors(q) - 1) ||
661 	    !blk_queue_zone_is_seq(q, pos))
662 		return BLK_STS_IOERR;
663 
664 	/*
665 	 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
666 	 * split and could result in non-contiguous sectors being written in
667 	 * different zones.
668 	 */
669 	if (nr_sectors > q->limits.chunk_sectors)
670 		return BLK_STS_IOERR;
671 
672 	/* Make sure the BIO is small enough and will not get split */
673 	if (nr_sectors > q->limits.max_zone_append_sectors)
674 		return BLK_STS_IOERR;
675 
676 	bio->bi_opf |= REQ_NOMERGE;
677 
678 	return BLK_STS_OK;
679 }
680 
681 noinline_for_stack bool submit_bio_checks(struct bio *bio)
682 {
683 	struct block_device *bdev = bio->bi_bdev;
684 	struct request_queue *q = bdev_get_queue(bdev);
685 	blk_status_t status = BLK_STS_IOERR;
686 	struct blk_plug *plug;
687 
688 	might_sleep();
689 
690 	plug = blk_mq_plug(q, bio);
691 	if (plug && plug->nowait)
692 		bio->bi_opf |= REQ_NOWAIT;
693 
694 	/*
695 	 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
696 	 * if queue does not support NOWAIT.
697 	 */
698 	if ((bio->bi_opf & REQ_NOWAIT) && !blk_queue_nowait(q))
699 		goto not_supported;
700 
701 	if (should_fail_bio(bio))
702 		goto end_io;
703 	if (unlikely(bio_check_ro(bio)))
704 		goto end_io;
705 	if (!bio_flagged(bio, BIO_REMAPPED)) {
706 		if (unlikely(bio_check_eod(bio)))
707 			goto end_io;
708 		if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
709 			goto end_io;
710 	}
711 
712 	/*
713 	 * Filter flush bio's early so that bio based drivers without flush
714 	 * support don't have to worry about them.
715 	 */
716 	if (op_is_flush(bio->bi_opf) &&
717 	    !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
718 		bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
719 		if (!bio_sectors(bio)) {
720 			status = BLK_STS_OK;
721 			goto end_io;
722 		}
723 	}
724 
725 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
726 		bio_clear_polled(bio);
727 
728 	switch (bio_op(bio)) {
729 	case REQ_OP_DISCARD:
730 		if (!blk_queue_discard(q))
731 			goto not_supported;
732 		break;
733 	case REQ_OP_SECURE_ERASE:
734 		if (!blk_queue_secure_erase(q))
735 			goto not_supported;
736 		break;
737 	case REQ_OP_WRITE_SAME:
738 		if (!q->limits.max_write_same_sectors)
739 			goto not_supported;
740 		break;
741 	case REQ_OP_ZONE_APPEND:
742 		status = blk_check_zone_append(q, bio);
743 		if (status != BLK_STS_OK)
744 			goto end_io;
745 		break;
746 	case REQ_OP_ZONE_RESET:
747 	case REQ_OP_ZONE_OPEN:
748 	case REQ_OP_ZONE_CLOSE:
749 	case REQ_OP_ZONE_FINISH:
750 		if (!blk_queue_is_zoned(q))
751 			goto not_supported;
752 		break;
753 	case REQ_OP_ZONE_RESET_ALL:
754 		if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
755 			goto not_supported;
756 		break;
757 	case REQ_OP_WRITE_ZEROES:
758 		if (!q->limits.max_write_zeroes_sectors)
759 			goto not_supported;
760 		break;
761 	default:
762 		break;
763 	}
764 
765 	if (blk_throtl_bio(bio))
766 		return false;
767 
768 	blk_cgroup_bio_start(bio);
769 	blkcg_bio_issue_init(bio);
770 
771 	if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
772 		trace_block_bio_queue(bio);
773 		/* Now that enqueuing has been traced, we need to trace
774 		 * completion as well.
775 		 */
776 		bio_set_flag(bio, BIO_TRACE_COMPLETION);
777 	}
778 	return true;
779 
780 not_supported:
781 	status = BLK_STS_NOTSUPP;
782 end_io:
783 	bio->bi_status = status;
784 	bio_endio(bio);
785 	return false;
786 }
787 
788 static void __submit_bio_fops(struct gendisk *disk, struct bio *bio)
789 {
790 	if (blk_crypto_bio_prep(&bio)) {
791 		if (likely(bio_queue_enter(bio) == 0)) {
792 			disk->fops->submit_bio(bio);
793 			blk_queue_exit(disk->queue);
794 		}
795 	}
796 }
797 
798 static void __submit_bio(struct bio *bio)
799 {
800 	struct gendisk *disk = bio->bi_bdev->bd_disk;
801 
802 	if (unlikely(!submit_bio_checks(bio)))
803 		return;
804 
805 	if (!disk->fops->submit_bio)
806 		blk_mq_submit_bio(bio);
807 	else
808 		__submit_bio_fops(disk, bio);
809 }
810 
811 /*
812  * The loop in this function may be a bit non-obvious, and so deserves some
813  * explanation:
814  *
815  *  - Before entering the loop, bio->bi_next is NULL (as all callers ensure
816  *    that), so we have a list with a single bio.
817  *  - We pretend that we have just taken it off a longer list, so we assign
818  *    bio_list to a pointer to the bio_list_on_stack, thus initialising the
819  *    bio_list of new bios to be added.  ->submit_bio() may indeed add some more
820  *    bios through a recursive call to submit_bio_noacct.  If it did, we find a
821  *    non-NULL value in bio_list and re-enter the loop from the top.
822  *  - In this case we really did just take the bio of the top of the list (no
823  *    pretending) and so remove it from bio_list, and call into ->submit_bio()
824  *    again.
825  *
826  * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
827  * bio_list_on_stack[1] contains bios that were submitted before the current
828  *	->submit_bio_bio, but that haven't been processed yet.
829  */
830 static void __submit_bio_noacct(struct bio *bio)
831 {
832 	struct bio_list bio_list_on_stack[2];
833 
834 	BUG_ON(bio->bi_next);
835 
836 	bio_list_init(&bio_list_on_stack[0]);
837 	current->bio_list = bio_list_on_stack;
838 
839 	do {
840 		struct request_queue *q = bdev_get_queue(bio->bi_bdev);
841 		struct bio_list lower, same;
842 
843 		/*
844 		 * Create a fresh bio_list for all subordinate requests.
845 		 */
846 		bio_list_on_stack[1] = bio_list_on_stack[0];
847 		bio_list_init(&bio_list_on_stack[0]);
848 
849 		__submit_bio(bio);
850 
851 		/*
852 		 * Sort new bios into those for a lower level and those for the
853 		 * same level.
854 		 */
855 		bio_list_init(&lower);
856 		bio_list_init(&same);
857 		while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
858 			if (q == bdev_get_queue(bio->bi_bdev))
859 				bio_list_add(&same, bio);
860 			else
861 				bio_list_add(&lower, bio);
862 
863 		/*
864 		 * Now assemble so we handle the lowest level first.
865 		 */
866 		bio_list_merge(&bio_list_on_stack[0], &lower);
867 		bio_list_merge(&bio_list_on_stack[0], &same);
868 		bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
869 	} while ((bio = bio_list_pop(&bio_list_on_stack[0])));
870 
871 	current->bio_list = NULL;
872 }
873 
874 static void __submit_bio_noacct_mq(struct bio *bio)
875 {
876 	struct bio_list bio_list[2] = { };
877 
878 	current->bio_list = bio_list;
879 
880 	do {
881 		__submit_bio(bio);
882 	} while ((bio = bio_list_pop(&bio_list[0])));
883 
884 	current->bio_list = NULL;
885 }
886 
887 /**
888  * submit_bio_noacct - re-submit a bio to the block device layer for I/O
889  * @bio:  The bio describing the location in memory and on the device.
890  *
891  * This is a version of submit_bio() that shall only be used for I/O that is
892  * resubmitted to lower level drivers by stacking block drivers.  All file
893  * systems and other upper level users of the block layer should use
894  * submit_bio() instead.
895  */
896 void submit_bio_noacct(struct bio *bio)
897 {
898 	/*
899 	 * We only want one ->submit_bio to be active at a time, else stack
900 	 * usage with stacked devices could be a problem.  Use current->bio_list
901 	 * to collect a list of requests submited by a ->submit_bio method while
902 	 * it is active, and then process them after it returned.
903 	 */
904 	if (current->bio_list)
905 		bio_list_add(&current->bio_list[0], bio);
906 	else if (!bio->bi_bdev->bd_disk->fops->submit_bio)
907 		__submit_bio_noacct_mq(bio);
908 	else
909 		__submit_bio_noacct(bio);
910 }
911 EXPORT_SYMBOL(submit_bio_noacct);
912 
913 /**
914  * submit_bio - submit a bio to the block device layer for I/O
915  * @bio: The &struct bio which describes the I/O
916  *
917  * submit_bio() is used to submit I/O requests to block devices.  It is passed a
918  * fully set up &struct bio that describes the I/O that needs to be done.  The
919  * bio will be send to the device described by the bi_bdev field.
920  *
921  * The success/failure status of the request, along with notification of
922  * completion, is delivered asynchronously through the ->bi_end_io() callback
923  * in @bio.  The bio must NOT be touched by thecaller until ->bi_end_io() has
924  * been called.
925  */
926 void submit_bio(struct bio *bio)
927 {
928 	if (blkcg_punt_bio_submit(bio))
929 		return;
930 
931 	/*
932 	 * If it's a regular read/write or a barrier with data attached,
933 	 * go through the normal accounting stuff before submission.
934 	 */
935 	if (bio_has_data(bio)) {
936 		unsigned int count;
937 
938 		if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
939 			count = queue_logical_block_size(
940 					bdev_get_queue(bio->bi_bdev)) >> 9;
941 		else
942 			count = bio_sectors(bio);
943 
944 		if (op_is_write(bio_op(bio))) {
945 			count_vm_events(PGPGOUT, count);
946 		} else {
947 			task_io_account_read(bio->bi_iter.bi_size);
948 			count_vm_events(PGPGIN, count);
949 		}
950 	}
951 
952 	/*
953 	 * If we're reading data that is part of the userspace workingset, count
954 	 * submission time as memory stall.  When the device is congested, or
955 	 * the submitting cgroup IO-throttled, submission can be a significant
956 	 * part of overall IO time.
957 	 */
958 	if (unlikely(bio_op(bio) == REQ_OP_READ &&
959 	    bio_flagged(bio, BIO_WORKINGSET))) {
960 		unsigned long pflags;
961 
962 		psi_memstall_enter(&pflags);
963 		submit_bio_noacct(bio);
964 		psi_memstall_leave(&pflags);
965 		return;
966 	}
967 
968 	submit_bio_noacct(bio);
969 }
970 EXPORT_SYMBOL(submit_bio);
971 
972 /**
973  * bio_poll - poll for BIO completions
974  * @bio: bio to poll for
975  * @iob: batches of IO
976  * @flags: BLK_POLL_* flags that control the behavior
977  *
978  * Poll for completions on queue associated with the bio. Returns number of
979  * completed entries found.
980  *
981  * Note: the caller must either be the context that submitted @bio, or
982  * be in a RCU critical section to prevent freeing of @bio.
983  */
984 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
985 {
986 	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
987 	blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
988 	int ret;
989 
990 	if (cookie == BLK_QC_T_NONE ||
991 	    !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
992 		return 0;
993 
994 	if (current->plug)
995 		blk_flush_plug(current->plug, false);
996 
997 	if (blk_queue_enter(q, BLK_MQ_REQ_NOWAIT))
998 		return 0;
999 	if (WARN_ON_ONCE(!queue_is_mq(q)))
1000 		ret = 0;	/* not yet implemented, should not happen */
1001 	else
1002 		ret = blk_mq_poll(q, cookie, iob, flags);
1003 	blk_queue_exit(q);
1004 	return ret;
1005 }
1006 EXPORT_SYMBOL_GPL(bio_poll);
1007 
1008 /*
1009  * Helper to implement file_operations.iopoll.  Requires the bio to be stored
1010  * in iocb->private, and cleared before freeing the bio.
1011  */
1012 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
1013 		    unsigned int flags)
1014 {
1015 	struct bio *bio;
1016 	int ret = 0;
1017 
1018 	/*
1019 	 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
1020 	 * point to a freshly allocated bio at this point.  If that happens
1021 	 * we have a few cases to consider:
1022 	 *
1023 	 *  1) the bio is beeing initialized and bi_bdev is NULL.  We can just
1024 	 *     simply nothing in this case
1025 	 *  2) the bio points to a not poll enabled device.  bio_poll will catch
1026 	 *     this and return 0
1027 	 *  3) the bio points to a poll capable device, including but not
1028 	 *     limited to the one that the original bio pointed to.  In this
1029 	 *     case we will call into the actual poll method and poll for I/O,
1030 	 *     even if we don't need to, but it won't cause harm either.
1031 	 *
1032 	 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
1033 	 * is still allocated. Because partitions hold a reference to the whole
1034 	 * device bdev and thus disk, the disk is also still valid.  Grabbing
1035 	 * a reference to the queue in bio_poll() ensures the hctxs and requests
1036 	 * are still valid as well.
1037 	 */
1038 	rcu_read_lock();
1039 	bio = READ_ONCE(kiocb->private);
1040 	if (bio && bio->bi_bdev)
1041 		ret = bio_poll(bio, iob, flags);
1042 	rcu_read_unlock();
1043 
1044 	return ret;
1045 }
1046 EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
1047 
1048 void update_io_ticks(struct block_device *part, unsigned long now, bool end)
1049 {
1050 	unsigned long stamp;
1051 again:
1052 	stamp = READ_ONCE(part->bd_stamp);
1053 	if (unlikely(time_after(now, stamp))) {
1054 		if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp))
1055 			__part_stat_add(part, io_ticks, end ? now - stamp : 1);
1056 	}
1057 	if (part->bd_partno) {
1058 		part = bdev_whole(part);
1059 		goto again;
1060 	}
1061 }
1062 
1063 static unsigned long __part_start_io_acct(struct block_device *part,
1064 					  unsigned int sectors, unsigned int op,
1065 					  unsigned long start_time)
1066 {
1067 	const int sgrp = op_stat_group(op);
1068 
1069 	part_stat_lock();
1070 	update_io_ticks(part, start_time, false);
1071 	part_stat_inc(part, ios[sgrp]);
1072 	part_stat_add(part, sectors[sgrp], sectors);
1073 	part_stat_local_inc(part, in_flight[op_is_write(op)]);
1074 	part_stat_unlock();
1075 
1076 	return start_time;
1077 }
1078 
1079 /**
1080  * bio_start_io_acct_time - start I/O accounting for bio based drivers
1081  * @bio:	bio to start account for
1082  * @start_time:	start time that should be passed back to bio_end_io_acct().
1083  */
1084 void bio_start_io_acct_time(struct bio *bio, unsigned long start_time)
1085 {
1086 	__part_start_io_acct(bio->bi_bdev, bio_sectors(bio),
1087 			     bio_op(bio), start_time);
1088 }
1089 EXPORT_SYMBOL_GPL(bio_start_io_acct_time);
1090 
1091 /**
1092  * bio_start_io_acct - start I/O accounting for bio based drivers
1093  * @bio:	bio to start account for
1094  *
1095  * Returns the start time that should be passed back to bio_end_io_acct().
1096  */
1097 unsigned long bio_start_io_acct(struct bio *bio)
1098 {
1099 	return __part_start_io_acct(bio->bi_bdev, bio_sectors(bio),
1100 				    bio_op(bio), jiffies);
1101 }
1102 EXPORT_SYMBOL_GPL(bio_start_io_acct);
1103 
1104 unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
1105 				 unsigned int op)
1106 {
1107 	return __part_start_io_acct(disk->part0, sectors, op, jiffies);
1108 }
1109 EXPORT_SYMBOL(disk_start_io_acct);
1110 
1111 static void __part_end_io_acct(struct block_device *part, unsigned int op,
1112 			       unsigned long start_time)
1113 {
1114 	const int sgrp = op_stat_group(op);
1115 	unsigned long now = READ_ONCE(jiffies);
1116 	unsigned long duration = now - start_time;
1117 
1118 	part_stat_lock();
1119 	update_io_ticks(part, now, true);
1120 	part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration));
1121 	part_stat_local_dec(part, in_flight[op_is_write(op)]);
1122 	part_stat_unlock();
1123 }
1124 
1125 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
1126 		struct block_device *orig_bdev)
1127 {
1128 	__part_end_io_acct(orig_bdev, bio_op(bio), start_time);
1129 }
1130 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1131 
1132 void disk_end_io_acct(struct gendisk *disk, unsigned int op,
1133 		      unsigned long start_time)
1134 {
1135 	__part_end_io_acct(disk->part0, op, start_time);
1136 }
1137 EXPORT_SYMBOL(disk_end_io_acct);
1138 
1139 /**
1140  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1141  * @q : the queue of the device being checked
1142  *
1143  * Description:
1144  *    Check if underlying low-level drivers of a device are busy.
1145  *    If the drivers want to export their busy state, they must set own
1146  *    exporting function using blk_queue_lld_busy() first.
1147  *
1148  *    Basically, this function is used only by request stacking drivers
1149  *    to stop dispatching requests to underlying devices when underlying
1150  *    devices are busy.  This behavior helps more I/O merging on the queue
1151  *    of the request stacking driver and prevents I/O throughput regression
1152  *    on burst I/O load.
1153  *
1154  * Return:
1155  *    0 - Not busy (The request stacking driver should dispatch request)
1156  *    1 - Busy (The request stacking driver should stop dispatching request)
1157  */
1158 int blk_lld_busy(struct request_queue *q)
1159 {
1160 	if (queue_is_mq(q) && q->mq_ops->busy)
1161 		return q->mq_ops->busy(q);
1162 
1163 	return 0;
1164 }
1165 EXPORT_SYMBOL_GPL(blk_lld_busy);
1166 
1167 int kblockd_schedule_work(struct work_struct *work)
1168 {
1169 	return queue_work(kblockd_workqueue, work);
1170 }
1171 EXPORT_SYMBOL(kblockd_schedule_work);
1172 
1173 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1174 				unsigned long delay)
1175 {
1176 	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1177 }
1178 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1179 
1180 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1181 {
1182 	struct task_struct *tsk = current;
1183 
1184 	/*
1185 	 * If this is a nested plug, don't actually assign it.
1186 	 */
1187 	if (tsk->plug)
1188 		return;
1189 
1190 	plug->mq_list = NULL;
1191 	plug->cached_rq = NULL;
1192 	plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1193 	plug->rq_count = 0;
1194 	plug->multiple_queues = false;
1195 	plug->has_elevator = false;
1196 	plug->nowait = false;
1197 	INIT_LIST_HEAD(&plug->cb_list);
1198 
1199 	/*
1200 	 * Store ordering should not be needed here, since a potential
1201 	 * preempt will imply a full memory barrier
1202 	 */
1203 	tsk->plug = plug;
1204 }
1205 
1206 /**
1207  * blk_start_plug - initialize blk_plug and track it inside the task_struct
1208  * @plug:	The &struct blk_plug that needs to be initialized
1209  *
1210  * Description:
1211  *   blk_start_plug() indicates to the block layer an intent by the caller
1212  *   to submit multiple I/O requests in a batch.  The block layer may use
1213  *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1214  *   is called.  However, the block layer may choose to submit requests
1215  *   before a call to blk_finish_plug() if the number of queued I/Os
1216  *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1217  *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1218  *   the task schedules (see below).
1219  *
1220  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1221  *   pending I/O should the task end up blocking between blk_start_plug() and
1222  *   blk_finish_plug(). This is important from a performance perspective, but
1223  *   also ensures that we don't deadlock. For instance, if the task is blocking
1224  *   for a memory allocation, memory reclaim could end up wanting to free a
1225  *   page belonging to that request that is currently residing in our private
1226  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1227  *   this kind of deadlock.
1228  */
1229 void blk_start_plug(struct blk_plug *plug)
1230 {
1231 	blk_start_plug_nr_ios(plug, 1);
1232 }
1233 EXPORT_SYMBOL(blk_start_plug);
1234 
1235 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1236 {
1237 	LIST_HEAD(callbacks);
1238 
1239 	while (!list_empty(&plug->cb_list)) {
1240 		list_splice_init(&plug->cb_list, &callbacks);
1241 
1242 		while (!list_empty(&callbacks)) {
1243 			struct blk_plug_cb *cb = list_first_entry(&callbacks,
1244 							  struct blk_plug_cb,
1245 							  list);
1246 			list_del(&cb->list);
1247 			cb->callback(cb, from_schedule);
1248 		}
1249 	}
1250 }
1251 
1252 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1253 				      int size)
1254 {
1255 	struct blk_plug *plug = current->plug;
1256 	struct blk_plug_cb *cb;
1257 
1258 	if (!plug)
1259 		return NULL;
1260 
1261 	list_for_each_entry(cb, &plug->cb_list, list)
1262 		if (cb->callback == unplug && cb->data == data)
1263 			return cb;
1264 
1265 	/* Not currently on the callback list */
1266 	BUG_ON(size < sizeof(*cb));
1267 	cb = kzalloc(size, GFP_ATOMIC);
1268 	if (cb) {
1269 		cb->data = data;
1270 		cb->callback = unplug;
1271 		list_add(&cb->list, &plug->cb_list);
1272 	}
1273 	return cb;
1274 }
1275 EXPORT_SYMBOL(blk_check_plugged);
1276 
1277 void blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1278 {
1279 	if (!list_empty(&plug->cb_list))
1280 		flush_plug_callbacks(plug, from_schedule);
1281 	if (!rq_list_empty(plug->mq_list))
1282 		blk_mq_flush_plug_list(plug, from_schedule);
1283 	/*
1284 	 * Unconditionally flush out cached requests, even if the unplug
1285 	 * event came from schedule. Since we know hold references to the
1286 	 * queue for cached requests, we don't want a blocked task holding
1287 	 * up a queue freeze/quiesce event.
1288 	 */
1289 	if (unlikely(!rq_list_empty(plug->cached_rq)))
1290 		blk_mq_free_plug_rqs(plug);
1291 }
1292 
1293 /**
1294  * blk_finish_plug - mark the end of a batch of submitted I/O
1295  * @plug:	The &struct blk_plug passed to blk_start_plug()
1296  *
1297  * Description:
1298  * Indicate that a batch of I/O submissions is complete.  This function
1299  * must be paired with an initial call to blk_start_plug().  The intent
1300  * is to allow the block layer to optimize I/O submission.  See the
1301  * documentation for blk_start_plug() for more information.
1302  */
1303 void blk_finish_plug(struct blk_plug *plug)
1304 {
1305 	if (plug == current->plug) {
1306 		blk_flush_plug(plug, false);
1307 		current->plug = NULL;
1308 	}
1309 }
1310 EXPORT_SYMBOL(blk_finish_plug);
1311 
1312 void blk_io_schedule(void)
1313 {
1314 	/* Prevent hang_check timer from firing at us during very long I/O */
1315 	unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1316 
1317 	if (timeout)
1318 		io_schedule_timeout(timeout);
1319 	else
1320 		io_schedule();
1321 }
1322 EXPORT_SYMBOL_GPL(blk_io_schedule);
1323 
1324 int __init blk_dev_init(void)
1325 {
1326 	BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1327 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1328 			sizeof_field(struct request, cmd_flags));
1329 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1330 			sizeof_field(struct bio, bi_opf));
1331 	BUILD_BUG_ON(ALIGN(offsetof(struct request_queue, srcu),
1332 			   __alignof__(struct request_queue)) !=
1333 		     sizeof(struct request_queue));
1334 
1335 	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
1336 	kblockd_workqueue = alloc_workqueue("kblockd",
1337 					    WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1338 	if (!kblockd_workqueue)
1339 		panic("Failed to create kblockd\n");
1340 
1341 	blk_requestq_cachep = kmem_cache_create("request_queue",
1342 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1343 
1344 	blk_requestq_srcu_cachep = kmem_cache_create("request_queue_srcu",
1345 			sizeof(struct request_queue) +
1346 			sizeof(struct srcu_struct), 0, SLAB_PANIC, NULL);
1347 
1348 	blk_debugfs_root = debugfs_create_dir("block", NULL);
1349 
1350 	return 0;
1351 }
1352