xref: /linux/block/blk-core.c (revision a4ff64edf9edc8f05e2183610dc8306d3279c6ac)
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/part_stat.h>
41 #include <linux/sched/sysctl.h>
42 #include <linux/blk-crypto.h>
43 
44 #define CREATE_TRACE_POINTS
45 #include <trace/events/block.h>
46 
47 #include "blk.h"
48 #include "blk-mq-sched.h"
49 #include "blk-pm.h"
50 #include "blk-cgroup.h"
51 #include "blk-throttle.h"
52 
53 struct dentry *blk_debugfs_root;
54 
55 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert);
61 
62 static DEFINE_IDA(blk_queue_ida);
63 
64 /*
65  * For queue allocation
66  */
67 static struct kmem_cache *blk_requestq_cachep;
68 
69 /*
70  * Controlling structure to kblockd
71  */
72 static struct workqueue_struct *kblockd_workqueue;
73 
74 /**
75  * blk_queue_flag_set - atomically set a queue flag
76  * @flag: flag to be set
77  * @q: request queue
78  */
79 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
80 {
81 	set_bit(flag, &q->queue_flags);
82 }
83 EXPORT_SYMBOL(blk_queue_flag_set);
84 
85 /**
86  * blk_queue_flag_clear - atomically clear a queue flag
87  * @flag: flag to be cleared
88  * @q: request queue
89  */
90 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
91 {
92 	clear_bit(flag, &q->queue_flags);
93 }
94 EXPORT_SYMBOL(blk_queue_flag_clear);
95 
96 /**
97  * blk_queue_flag_test_and_set - atomically test and set a queue flag
98  * @flag: flag to be set
99  * @q: request queue
100  *
101  * Returns the previous value of @flag - 0 if the flag was not set and 1 if
102  * the flag was already set.
103  */
104 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
105 {
106 	return test_and_set_bit(flag, &q->queue_flags);
107 }
108 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
109 
110 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
111 static const char *const blk_op_name[] = {
112 	REQ_OP_NAME(READ),
113 	REQ_OP_NAME(WRITE),
114 	REQ_OP_NAME(FLUSH),
115 	REQ_OP_NAME(DISCARD),
116 	REQ_OP_NAME(SECURE_ERASE),
117 	REQ_OP_NAME(ZONE_RESET),
118 	REQ_OP_NAME(ZONE_RESET_ALL),
119 	REQ_OP_NAME(ZONE_OPEN),
120 	REQ_OP_NAME(ZONE_CLOSE),
121 	REQ_OP_NAME(ZONE_FINISH),
122 	REQ_OP_NAME(ZONE_APPEND),
123 	REQ_OP_NAME(WRITE_ZEROES),
124 	REQ_OP_NAME(DRV_IN),
125 	REQ_OP_NAME(DRV_OUT),
126 };
127 #undef REQ_OP_NAME
128 
129 /**
130  * blk_op_str - Return string XXX in the REQ_OP_XXX.
131  * @op: REQ_OP_XXX.
132  *
133  * Description: Centralize block layer function to convert REQ_OP_XXX into
134  * string format. Useful in the debugging and tracing bio or request. For
135  * invalid REQ_OP_XXX it returns string "UNKNOWN".
136  */
137 inline const char *blk_op_str(enum req_op op)
138 {
139 	const char *op_str = "UNKNOWN";
140 
141 	if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
142 		op_str = blk_op_name[op];
143 
144 	return op_str;
145 }
146 EXPORT_SYMBOL_GPL(blk_op_str);
147 
148 static const struct {
149 	int		errno;
150 	const char	*name;
151 } blk_errors[] = {
152 	[BLK_STS_OK]		= { 0,		"" },
153 	[BLK_STS_NOTSUPP]	= { -EOPNOTSUPP, "operation not supported" },
154 	[BLK_STS_TIMEOUT]	= { -ETIMEDOUT,	"timeout" },
155 	[BLK_STS_NOSPC]		= { -ENOSPC,	"critical space allocation" },
156 	[BLK_STS_TRANSPORT]	= { -ENOLINK,	"recoverable transport" },
157 	[BLK_STS_TARGET]	= { -EREMOTEIO,	"critical target" },
158 	[BLK_STS_RESV_CONFLICT]	= { -EBADE,	"reservation conflict" },
159 	[BLK_STS_MEDIUM]	= { -ENODATA,	"critical medium" },
160 	[BLK_STS_PROTECTION]	= { -EILSEQ,	"protection" },
161 	[BLK_STS_RESOURCE]	= { -ENOMEM,	"kernel resource" },
162 	[BLK_STS_DEV_RESOURCE]	= { -EBUSY,	"device resource" },
163 	[BLK_STS_AGAIN]		= { -EAGAIN,	"nonblocking retry" },
164 	[BLK_STS_OFFLINE]	= { -ENODEV,	"device offline" },
165 
166 	/* device mapper special case, should not leak out: */
167 	[BLK_STS_DM_REQUEUE]	= { -EREMCHG, "dm internal retry" },
168 
169 	/* zone device specific errors */
170 	[BLK_STS_ZONE_OPEN_RESOURCE]	= { -ETOOMANYREFS, "open zones exceeded" },
171 	[BLK_STS_ZONE_ACTIVE_RESOURCE]	= { -EOVERFLOW, "active zones exceeded" },
172 
173 	/* Command duration limit device-side timeout */
174 	[BLK_STS_DURATION_LIMIT]	= { -ETIME, "duration limit exceeded" },
175 
176 	/* everything else not covered above: */
177 	[BLK_STS_IOERR]		= { -EIO,	"I/O" },
178 };
179 
180 blk_status_t errno_to_blk_status(int errno)
181 {
182 	int i;
183 
184 	for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
185 		if (blk_errors[i].errno == errno)
186 			return (__force blk_status_t)i;
187 	}
188 
189 	return BLK_STS_IOERR;
190 }
191 EXPORT_SYMBOL_GPL(errno_to_blk_status);
192 
193 int blk_status_to_errno(blk_status_t status)
194 {
195 	int idx = (__force int)status;
196 
197 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
198 		return -EIO;
199 	return blk_errors[idx].errno;
200 }
201 EXPORT_SYMBOL_GPL(blk_status_to_errno);
202 
203 const char *blk_status_to_str(blk_status_t status)
204 {
205 	int idx = (__force int)status;
206 
207 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
208 		return "<null>";
209 	return blk_errors[idx].name;
210 }
211 EXPORT_SYMBOL_GPL(blk_status_to_str);
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 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
260 {
261 	struct request_queue *q = container_of(rcu_head,
262 			struct request_queue, rcu_head);
263 
264 	percpu_ref_exit(&q->q_usage_counter);
265 	kmem_cache_free(blk_requestq_cachep, q);
266 }
267 
268 static void blk_free_queue(struct request_queue *q)
269 {
270 	blk_free_queue_stats(q->stats);
271 	if (queue_is_mq(q))
272 		blk_mq_release(q);
273 
274 	ida_free(&blk_queue_ida, q->id);
275 	call_rcu(&q->rcu_head, blk_free_queue_rcu);
276 }
277 
278 /**
279  * blk_put_queue - decrement the request_queue refcount
280  * @q: the request_queue structure to decrement the refcount for
281  *
282  * Decrements the refcount of the request_queue and free it when the refcount
283  * reaches 0.
284  */
285 void blk_put_queue(struct request_queue *q)
286 {
287 	if (refcount_dec_and_test(&q->refs))
288 		blk_free_queue(q);
289 }
290 EXPORT_SYMBOL(blk_put_queue);
291 
292 void blk_queue_start_drain(struct request_queue *q)
293 {
294 	/*
295 	 * When queue DYING flag is set, we need to block new req
296 	 * entering queue, so we call blk_freeze_queue_start() to
297 	 * prevent I/O from crossing blk_queue_enter().
298 	 */
299 	blk_freeze_queue_start(q);
300 	if (queue_is_mq(q))
301 		blk_mq_wake_waiters(q);
302 	/* Make blk_queue_enter() reexamine the DYING flag. */
303 	wake_up_all(&q->mq_freeze_wq);
304 }
305 
306 /**
307  * blk_queue_enter() - try to increase q->q_usage_counter
308  * @q: request queue pointer
309  * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
310  */
311 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
312 {
313 	const bool pm = flags & BLK_MQ_REQ_PM;
314 
315 	while (!blk_try_enter_queue(q, pm)) {
316 		if (flags & BLK_MQ_REQ_NOWAIT)
317 			return -EAGAIN;
318 
319 		/*
320 		 * read pair of barrier in blk_freeze_queue_start(), we need to
321 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
322 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
323 		 * following wait may never return if the two reads are
324 		 * reordered.
325 		 */
326 		smp_rmb();
327 		wait_event(q->mq_freeze_wq,
328 			   (!q->mq_freeze_depth &&
329 			    blk_pm_resume_queue(pm, q)) ||
330 			   blk_queue_dying(q));
331 		if (blk_queue_dying(q))
332 			return -ENODEV;
333 	}
334 
335 	return 0;
336 }
337 
338 int __bio_queue_enter(struct request_queue *q, struct bio *bio)
339 {
340 	while (!blk_try_enter_queue(q, false)) {
341 		struct gendisk *disk = bio->bi_bdev->bd_disk;
342 
343 		if (bio->bi_opf & REQ_NOWAIT) {
344 			if (test_bit(GD_DEAD, &disk->state))
345 				goto dead;
346 			bio_wouldblock_error(bio);
347 			return -EAGAIN;
348 		}
349 
350 		/*
351 		 * read pair of barrier in blk_freeze_queue_start(), we need to
352 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
353 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
354 		 * following wait may never return if the two reads are
355 		 * reordered.
356 		 */
357 		smp_rmb();
358 		wait_event(q->mq_freeze_wq,
359 			   (!q->mq_freeze_depth &&
360 			    blk_pm_resume_queue(false, q)) ||
361 			   test_bit(GD_DEAD, &disk->state));
362 		if (test_bit(GD_DEAD, &disk->state))
363 			goto dead;
364 	}
365 
366 	return 0;
367 dead:
368 	bio_io_error(bio);
369 	return -ENODEV;
370 }
371 
372 void blk_queue_exit(struct request_queue *q)
373 {
374 	percpu_ref_put(&q->q_usage_counter);
375 }
376 
377 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
378 {
379 	struct request_queue *q =
380 		container_of(ref, struct request_queue, q_usage_counter);
381 
382 	wake_up_all(&q->mq_freeze_wq);
383 }
384 
385 static void blk_rq_timed_out_timer(struct timer_list *t)
386 {
387 	struct request_queue *q = from_timer(q, t, timeout);
388 
389 	kblockd_schedule_work(&q->timeout_work);
390 }
391 
392 static void blk_timeout_work(struct work_struct *work)
393 {
394 }
395 
396 struct request_queue *blk_alloc_queue(int node_id)
397 {
398 	struct request_queue *q;
399 
400 	q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO,
401 				  node_id);
402 	if (!q)
403 		return NULL;
404 
405 	q->last_merge = NULL;
406 
407 	q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL);
408 	if (q->id < 0)
409 		goto fail_q;
410 
411 	q->stats = blk_alloc_queue_stats();
412 	if (!q->stats)
413 		goto fail_id;
414 
415 	q->node = node_id;
416 
417 	atomic_set(&q->nr_active_requests_shared_tags, 0);
418 
419 	timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
420 	INIT_WORK(&q->timeout_work, blk_timeout_work);
421 	INIT_LIST_HEAD(&q->icq_list);
422 
423 	refcount_set(&q->refs, 1);
424 	mutex_init(&q->debugfs_mutex);
425 	mutex_init(&q->sysfs_lock);
426 	mutex_init(&q->sysfs_dir_lock);
427 	mutex_init(&q->rq_qos_mutex);
428 	spin_lock_init(&q->queue_lock);
429 
430 	init_waitqueue_head(&q->mq_freeze_wq);
431 	mutex_init(&q->mq_freeze_lock);
432 
433 	/*
434 	 * Init percpu_ref in atomic mode so that it's faster to shutdown.
435 	 * See blk_register_queue() for details.
436 	 */
437 	if (percpu_ref_init(&q->q_usage_counter,
438 				blk_queue_usage_counter_release,
439 				PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
440 		goto fail_stats;
441 
442 	blk_set_default_limits(&q->limits);
443 	q->nr_requests = BLKDEV_DEFAULT_RQ;
444 
445 	return q;
446 
447 fail_stats:
448 	blk_free_queue_stats(q->stats);
449 fail_id:
450 	ida_free(&blk_queue_ida, q->id);
451 fail_q:
452 	kmem_cache_free(blk_requestq_cachep, q);
453 	return NULL;
454 }
455 
456 /**
457  * blk_get_queue - increment the request_queue refcount
458  * @q: the request_queue structure to increment the refcount for
459  *
460  * Increment the refcount of the request_queue kobject.
461  *
462  * Context: Any context.
463  */
464 bool blk_get_queue(struct request_queue *q)
465 {
466 	if (unlikely(blk_queue_dying(q)))
467 		return false;
468 	refcount_inc(&q->refs);
469 	return true;
470 }
471 EXPORT_SYMBOL(blk_get_queue);
472 
473 #ifdef CONFIG_FAIL_MAKE_REQUEST
474 
475 static DECLARE_FAULT_ATTR(fail_make_request);
476 
477 static int __init setup_fail_make_request(char *str)
478 {
479 	return setup_fault_attr(&fail_make_request, str);
480 }
481 __setup("fail_make_request=", setup_fail_make_request);
482 
483 bool should_fail_request(struct block_device *part, unsigned int bytes)
484 {
485 	return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
486 }
487 
488 static int __init fail_make_request_debugfs(void)
489 {
490 	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
491 						NULL, &fail_make_request);
492 
493 	return PTR_ERR_OR_ZERO(dir);
494 }
495 
496 late_initcall(fail_make_request_debugfs);
497 #endif /* CONFIG_FAIL_MAKE_REQUEST */
498 
499 static inline void bio_check_ro(struct bio *bio)
500 {
501 	if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
502 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
503 			return;
504 		pr_warn_ratelimited("Trying to write to read-only block-device %pg\n",
505 				    bio->bi_bdev);
506 		/* Older lvm-tools actually trigger this */
507 	}
508 }
509 
510 static noinline int should_fail_bio(struct bio *bio)
511 {
512 	if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
513 		return -EIO;
514 	return 0;
515 }
516 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
517 
518 /*
519  * Check whether this bio extends beyond the end of the device or partition.
520  * This may well happen - the kernel calls bread() without checking the size of
521  * the device, e.g., when mounting a file system.
522  */
523 static inline int bio_check_eod(struct bio *bio)
524 {
525 	sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
526 	unsigned int nr_sectors = bio_sectors(bio);
527 
528 	if (nr_sectors &&
529 	    (nr_sectors > maxsector ||
530 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
531 		pr_info_ratelimited("%s: attempt to access beyond end of device\n"
532 				    "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
533 				    current->comm, bio->bi_bdev, bio->bi_opf,
534 				    bio->bi_iter.bi_sector, nr_sectors, maxsector);
535 		return -EIO;
536 	}
537 	return 0;
538 }
539 
540 /*
541  * Remap block n of partition p to block n+start(p) of the disk.
542  */
543 static int blk_partition_remap(struct bio *bio)
544 {
545 	struct block_device *p = bio->bi_bdev;
546 
547 	if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
548 		return -EIO;
549 	if (bio_sectors(bio)) {
550 		bio->bi_iter.bi_sector += p->bd_start_sect;
551 		trace_block_bio_remap(bio, p->bd_dev,
552 				      bio->bi_iter.bi_sector -
553 				      p->bd_start_sect);
554 	}
555 	bio_set_flag(bio, BIO_REMAPPED);
556 	return 0;
557 }
558 
559 /*
560  * Check write append to a zoned block device.
561  */
562 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
563 						 struct bio *bio)
564 {
565 	int nr_sectors = bio_sectors(bio);
566 
567 	/* Only applicable to zoned block devices */
568 	if (!bdev_is_zoned(bio->bi_bdev))
569 		return BLK_STS_NOTSUPP;
570 
571 	/* The bio sector must point to the start of a sequential zone */
572 	if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector) ||
573 	    !bio_zone_is_seq(bio))
574 		return BLK_STS_IOERR;
575 
576 	/*
577 	 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
578 	 * split and could result in non-contiguous sectors being written in
579 	 * different zones.
580 	 */
581 	if (nr_sectors > q->limits.chunk_sectors)
582 		return BLK_STS_IOERR;
583 
584 	/* Make sure the BIO is small enough and will not get split */
585 	if (nr_sectors > q->limits.max_zone_append_sectors)
586 		return BLK_STS_IOERR;
587 
588 	bio->bi_opf |= REQ_NOMERGE;
589 
590 	return BLK_STS_OK;
591 }
592 
593 static void __submit_bio(struct bio *bio)
594 {
595 	if (unlikely(!blk_crypto_bio_prep(&bio)))
596 		return;
597 
598 	if (!bio->bi_bdev->bd_has_submit_bio) {
599 		blk_mq_submit_bio(bio);
600 	} else if (likely(bio_queue_enter(bio) == 0)) {
601 		struct gendisk *disk = bio->bi_bdev->bd_disk;
602 
603 		disk->fops->submit_bio(bio);
604 		blk_queue_exit(disk->queue);
605 	}
606 }
607 
608 /*
609  * The loop in this function may be a bit non-obvious, and so deserves some
610  * explanation:
611  *
612  *  - Before entering the loop, bio->bi_next is NULL (as all callers ensure
613  *    that), so we have a list with a single bio.
614  *  - We pretend that we have just taken it off a longer list, so we assign
615  *    bio_list to a pointer to the bio_list_on_stack, thus initialising the
616  *    bio_list of new bios to be added.  ->submit_bio() may indeed add some more
617  *    bios through a recursive call to submit_bio_noacct.  If it did, we find a
618  *    non-NULL value in bio_list and re-enter the loop from the top.
619  *  - In this case we really did just take the bio of the top of the list (no
620  *    pretending) and so remove it from bio_list, and call into ->submit_bio()
621  *    again.
622  *
623  * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
624  * bio_list_on_stack[1] contains bios that were submitted before the current
625  *	->submit_bio, but that haven't been processed yet.
626  */
627 static void __submit_bio_noacct(struct bio *bio)
628 {
629 	struct bio_list bio_list_on_stack[2];
630 
631 	BUG_ON(bio->bi_next);
632 
633 	bio_list_init(&bio_list_on_stack[0]);
634 	current->bio_list = bio_list_on_stack;
635 
636 	do {
637 		struct request_queue *q = bdev_get_queue(bio->bi_bdev);
638 		struct bio_list lower, same;
639 
640 		/*
641 		 * Create a fresh bio_list for all subordinate requests.
642 		 */
643 		bio_list_on_stack[1] = bio_list_on_stack[0];
644 		bio_list_init(&bio_list_on_stack[0]);
645 
646 		__submit_bio(bio);
647 
648 		/*
649 		 * Sort new bios into those for a lower level and those for the
650 		 * same level.
651 		 */
652 		bio_list_init(&lower);
653 		bio_list_init(&same);
654 		while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
655 			if (q == bdev_get_queue(bio->bi_bdev))
656 				bio_list_add(&same, bio);
657 			else
658 				bio_list_add(&lower, bio);
659 
660 		/*
661 		 * Now assemble so we handle the lowest level first.
662 		 */
663 		bio_list_merge(&bio_list_on_stack[0], &lower);
664 		bio_list_merge(&bio_list_on_stack[0], &same);
665 		bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
666 	} while ((bio = bio_list_pop(&bio_list_on_stack[0])));
667 
668 	current->bio_list = NULL;
669 }
670 
671 static void __submit_bio_noacct_mq(struct bio *bio)
672 {
673 	struct bio_list bio_list[2] = { };
674 
675 	current->bio_list = bio_list;
676 
677 	do {
678 		__submit_bio(bio);
679 	} while ((bio = bio_list_pop(&bio_list[0])));
680 
681 	current->bio_list = NULL;
682 }
683 
684 void submit_bio_noacct_nocheck(struct bio *bio)
685 {
686 	blk_cgroup_bio_start(bio);
687 	blkcg_bio_issue_init(bio);
688 
689 	if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
690 		trace_block_bio_queue(bio);
691 		/*
692 		 * Now that enqueuing has been traced, we need to trace
693 		 * completion as well.
694 		 */
695 		bio_set_flag(bio, BIO_TRACE_COMPLETION);
696 	}
697 
698 	/*
699 	 * We only want one ->submit_bio to be active at a time, else stack
700 	 * usage with stacked devices could be a problem.  Use current->bio_list
701 	 * to collect a list of requests submited by a ->submit_bio method while
702 	 * it is active, and then process them after it returned.
703 	 */
704 	if (current->bio_list)
705 		bio_list_add(&current->bio_list[0], bio);
706 	else if (!bio->bi_bdev->bd_has_submit_bio)
707 		__submit_bio_noacct_mq(bio);
708 	else
709 		__submit_bio_noacct(bio);
710 }
711 
712 /**
713  * submit_bio_noacct - re-submit a bio to the block device layer for I/O
714  * @bio:  The bio describing the location in memory and on the device.
715  *
716  * This is a version of submit_bio() that shall only be used for I/O that is
717  * resubmitted to lower level drivers by stacking block drivers.  All file
718  * systems and other upper level users of the block layer should use
719  * submit_bio() instead.
720  */
721 void submit_bio_noacct(struct bio *bio)
722 {
723 	struct block_device *bdev = bio->bi_bdev;
724 	struct request_queue *q = bdev_get_queue(bdev);
725 	blk_status_t status = BLK_STS_IOERR;
726 
727 	might_sleep();
728 
729 	/*
730 	 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
731 	 * if queue does not support NOWAIT.
732 	 */
733 	if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev))
734 		goto not_supported;
735 
736 	if (should_fail_bio(bio))
737 		goto end_io;
738 	bio_check_ro(bio);
739 	if (!bio_flagged(bio, BIO_REMAPPED)) {
740 		if (unlikely(bio_check_eod(bio)))
741 			goto end_io;
742 		if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
743 			goto end_io;
744 	}
745 
746 	/*
747 	 * Filter flush bio's early so that bio based drivers without flush
748 	 * support don't have to worry about them.
749 	 */
750 	if (op_is_flush(bio->bi_opf)) {
751 		if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE &&
752 				 bio_op(bio) != REQ_OP_ZONE_APPEND))
753 			goto end_io;
754 		if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
755 			bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
756 			if (!bio_sectors(bio)) {
757 				status = BLK_STS_OK;
758 				goto end_io;
759 			}
760 		}
761 	}
762 
763 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
764 		bio_clear_polled(bio);
765 
766 	switch (bio_op(bio)) {
767 	case REQ_OP_DISCARD:
768 		if (!bdev_max_discard_sectors(bdev))
769 			goto not_supported;
770 		break;
771 	case REQ_OP_SECURE_ERASE:
772 		if (!bdev_max_secure_erase_sectors(bdev))
773 			goto not_supported;
774 		break;
775 	case REQ_OP_ZONE_APPEND:
776 		status = blk_check_zone_append(q, bio);
777 		if (status != BLK_STS_OK)
778 			goto end_io;
779 		break;
780 	case REQ_OP_ZONE_RESET:
781 	case REQ_OP_ZONE_OPEN:
782 	case REQ_OP_ZONE_CLOSE:
783 	case REQ_OP_ZONE_FINISH:
784 		if (!bdev_is_zoned(bio->bi_bdev))
785 			goto not_supported;
786 		break;
787 	case REQ_OP_ZONE_RESET_ALL:
788 		if (!bdev_is_zoned(bio->bi_bdev) || !blk_queue_zone_resetall(q))
789 			goto not_supported;
790 		break;
791 	case REQ_OP_WRITE_ZEROES:
792 		if (!q->limits.max_write_zeroes_sectors)
793 			goto not_supported;
794 		break;
795 	default:
796 		break;
797 	}
798 
799 	if (blk_throtl_bio(bio))
800 		return;
801 	submit_bio_noacct_nocheck(bio);
802 	return;
803 
804 not_supported:
805 	status = BLK_STS_NOTSUPP;
806 end_io:
807 	bio->bi_status = status;
808 	bio_endio(bio);
809 }
810 EXPORT_SYMBOL(submit_bio_noacct);
811 
812 /**
813  * submit_bio - submit a bio to the block device layer for I/O
814  * @bio: The &struct bio which describes the I/O
815  *
816  * submit_bio() is used to submit I/O requests to block devices.  It is passed a
817  * fully set up &struct bio that describes the I/O that needs to be done.  The
818  * bio will be send to the device described by the bi_bdev field.
819  *
820  * The success/failure status of the request, along with notification of
821  * completion, is delivered asynchronously through the ->bi_end_io() callback
822  * in @bio.  The bio must NOT be touched by the caller until ->bi_end_io() has
823  * been called.
824  */
825 void submit_bio(struct bio *bio)
826 {
827 	if (bio_op(bio) == REQ_OP_READ) {
828 		task_io_account_read(bio->bi_iter.bi_size);
829 		count_vm_events(PGPGIN, bio_sectors(bio));
830 	} else if (bio_op(bio) == REQ_OP_WRITE) {
831 		count_vm_events(PGPGOUT, bio_sectors(bio));
832 	}
833 
834 	submit_bio_noacct(bio);
835 }
836 EXPORT_SYMBOL(submit_bio);
837 
838 /**
839  * bio_poll - poll for BIO completions
840  * @bio: bio to poll for
841  * @iob: batches of IO
842  * @flags: BLK_POLL_* flags that control the behavior
843  *
844  * Poll for completions on queue associated with the bio. Returns number of
845  * completed entries found.
846  *
847  * Note: the caller must either be the context that submitted @bio, or
848  * be in a RCU critical section to prevent freeing of @bio.
849  */
850 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
851 {
852 	blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
853 	struct block_device *bdev;
854 	struct request_queue *q;
855 	int ret = 0;
856 
857 	bdev = READ_ONCE(bio->bi_bdev);
858 	if (!bdev)
859 		return 0;
860 
861 	q = bdev_get_queue(bdev);
862 	if (cookie == BLK_QC_T_NONE ||
863 	    !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
864 		return 0;
865 
866 	/*
867 	 * As the requests that require a zone lock are not plugged in the
868 	 * first place, directly accessing the plug instead of using
869 	 * blk_mq_plug() should not have any consequences during flushing for
870 	 * zoned devices.
871 	 */
872 	blk_flush_plug(current->plug, false);
873 
874 	/*
875 	 * We need to be able to enter a frozen queue, similar to how
876 	 * timeouts also need to do that. If that is blocked, then we can
877 	 * have pending IO when a queue freeze is started, and then the
878 	 * wait for the freeze to finish will wait for polled requests to
879 	 * timeout as the poller is preventer from entering the queue and
880 	 * completing them. As long as we prevent new IO from being queued,
881 	 * that should be all that matters.
882 	 */
883 	if (!percpu_ref_tryget(&q->q_usage_counter))
884 		return 0;
885 	if (queue_is_mq(q)) {
886 		ret = blk_mq_poll(q, cookie, iob, flags);
887 	} else {
888 		struct gendisk *disk = q->disk;
889 
890 		if (disk && disk->fops->poll_bio)
891 			ret = disk->fops->poll_bio(bio, iob, flags);
892 	}
893 	blk_queue_exit(q);
894 	return ret;
895 }
896 EXPORT_SYMBOL_GPL(bio_poll);
897 
898 /*
899  * Helper to implement file_operations.iopoll.  Requires the bio to be stored
900  * in iocb->private, and cleared before freeing the bio.
901  */
902 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
903 		    unsigned int flags)
904 {
905 	struct bio *bio;
906 	int ret = 0;
907 
908 	/*
909 	 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
910 	 * point to a freshly allocated bio at this point.  If that happens
911 	 * we have a few cases to consider:
912 	 *
913 	 *  1) the bio is beeing initialized and bi_bdev is NULL.  We can just
914 	 *     simply nothing in this case
915 	 *  2) the bio points to a not poll enabled device.  bio_poll will catch
916 	 *     this and return 0
917 	 *  3) the bio points to a poll capable device, including but not
918 	 *     limited to the one that the original bio pointed to.  In this
919 	 *     case we will call into the actual poll method and poll for I/O,
920 	 *     even if we don't need to, but it won't cause harm either.
921 	 *
922 	 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
923 	 * is still allocated. Because partitions hold a reference to the whole
924 	 * device bdev and thus disk, the disk is also still valid.  Grabbing
925 	 * a reference to the queue in bio_poll() ensures the hctxs and requests
926 	 * are still valid as well.
927 	 */
928 	rcu_read_lock();
929 	bio = READ_ONCE(kiocb->private);
930 	if (bio)
931 		ret = bio_poll(bio, iob, flags);
932 	rcu_read_unlock();
933 
934 	return ret;
935 }
936 EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
937 
938 void update_io_ticks(struct block_device *part, unsigned long now, bool end)
939 {
940 	unsigned long stamp;
941 again:
942 	stamp = READ_ONCE(part->bd_stamp);
943 	if (unlikely(time_after(now, stamp))) {
944 		if (likely(try_cmpxchg(&part->bd_stamp, &stamp, now)))
945 			__part_stat_add(part, io_ticks, end ? now - stamp : 1);
946 	}
947 	if (part->bd_partno) {
948 		part = bdev_whole(part);
949 		goto again;
950 	}
951 }
952 
953 unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op,
954 				 unsigned long start_time)
955 {
956 	part_stat_lock();
957 	update_io_ticks(bdev, start_time, false);
958 	part_stat_local_inc(bdev, in_flight[op_is_write(op)]);
959 	part_stat_unlock();
960 
961 	return start_time;
962 }
963 EXPORT_SYMBOL(bdev_start_io_acct);
964 
965 /**
966  * bio_start_io_acct - start I/O accounting for bio based drivers
967  * @bio:	bio to start account for
968  *
969  * Returns the start time that should be passed back to bio_end_io_acct().
970  */
971 unsigned long bio_start_io_acct(struct bio *bio)
972 {
973 	return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies);
974 }
975 EXPORT_SYMBOL_GPL(bio_start_io_acct);
976 
977 void bdev_end_io_acct(struct block_device *bdev, enum req_op op,
978 		      unsigned int sectors, unsigned long start_time)
979 {
980 	const int sgrp = op_stat_group(op);
981 	unsigned long now = READ_ONCE(jiffies);
982 	unsigned long duration = now - start_time;
983 
984 	part_stat_lock();
985 	update_io_ticks(bdev, now, true);
986 	part_stat_inc(bdev, ios[sgrp]);
987 	part_stat_add(bdev, sectors[sgrp], sectors);
988 	part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration));
989 	part_stat_local_dec(bdev, in_flight[op_is_write(op)]);
990 	part_stat_unlock();
991 }
992 EXPORT_SYMBOL(bdev_end_io_acct);
993 
994 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
995 			      struct block_device *orig_bdev)
996 {
997 	bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time);
998 }
999 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1000 
1001 /**
1002  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1003  * @q : the queue of the device being checked
1004  *
1005  * Description:
1006  *    Check if underlying low-level drivers of a device are busy.
1007  *    If the drivers want to export their busy state, they must set own
1008  *    exporting function using blk_queue_lld_busy() first.
1009  *
1010  *    Basically, this function is used only by request stacking drivers
1011  *    to stop dispatching requests to underlying devices when underlying
1012  *    devices are busy.  This behavior helps more I/O merging on the queue
1013  *    of the request stacking driver and prevents I/O throughput regression
1014  *    on burst I/O load.
1015  *
1016  * Return:
1017  *    0 - Not busy (The request stacking driver should dispatch request)
1018  *    1 - Busy (The request stacking driver should stop dispatching request)
1019  */
1020 int blk_lld_busy(struct request_queue *q)
1021 {
1022 	if (queue_is_mq(q) && q->mq_ops->busy)
1023 		return q->mq_ops->busy(q);
1024 
1025 	return 0;
1026 }
1027 EXPORT_SYMBOL_GPL(blk_lld_busy);
1028 
1029 int kblockd_schedule_work(struct work_struct *work)
1030 {
1031 	return queue_work(kblockd_workqueue, work);
1032 }
1033 EXPORT_SYMBOL(kblockd_schedule_work);
1034 
1035 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1036 				unsigned long delay)
1037 {
1038 	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1039 }
1040 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1041 
1042 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1043 {
1044 	struct task_struct *tsk = current;
1045 
1046 	/*
1047 	 * If this is a nested plug, don't actually assign it.
1048 	 */
1049 	if (tsk->plug)
1050 		return;
1051 
1052 	plug->mq_list = NULL;
1053 	plug->cached_rq = NULL;
1054 	plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1055 	plug->rq_count = 0;
1056 	plug->multiple_queues = false;
1057 	plug->has_elevator = false;
1058 	INIT_LIST_HEAD(&plug->cb_list);
1059 
1060 	/*
1061 	 * Store ordering should not be needed here, since a potential
1062 	 * preempt will imply a full memory barrier
1063 	 */
1064 	tsk->plug = plug;
1065 }
1066 
1067 /**
1068  * blk_start_plug - initialize blk_plug and track it inside the task_struct
1069  * @plug:	The &struct blk_plug that needs to be initialized
1070  *
1071  * Description:
1072  *   blk_start_plug() indicates to the block layer an intent by the caller
1073  *   to submit multiple I/O requests in a batch.  The block layer may use
1074  *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1075  *   is called.  However, the block layer may choose to submit requests
1076  *   before a call to blk_finish_plug() if the number of queued I/Os
1077  *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1078  *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1079  *   the task schedules (see below).
1080  *
1081  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1082  *   pending I/O should the task end up blocking between blk_start_plug() and
1083  *   blk_finish_plug(). This is important from a performance perspective, but
1084  *   also ensures that we don't deadlock. For instance, if the task is blocking
1085  *   for a memory allocation, memory reclaim could end up wanting to free a
1086  *   page belonging to that request that is currently residing in our private
1087  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1088  *   this kind of deadlock.
1089  */
1090 void blk_start_plug(struct blk_plug *plug)
1091 {
1092 	blk_start_plug_nr_ios(plug, 1);
1093 }
1094 EXPORT_SYMBOL(blk_start_plug);
1095 
1096 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1097 {
1098 	LIST_HEAD(callbacks);
1099 
1100 	while (!list_empty(&plug->cb_list)) {
1101 		list_splice_init(&plug->cb_list, &callbacks);
1102 
1103 		while (!list_empty(&callbacks)) {
1104 			struct blk_plug_cb *cb = list_first_entry(&callbacks,
1105 							  struct blk_plug_cb,
1106 							  list);
1107 			list_del(&cb->list);
1108 			cb->callback(cb, from_schedule);
1109 		}
1110 	}
1111 }
1112 
1113 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1114 				      int size)
1115 {
1116 	struct blk_plug *plug = current->plug;
1117 	struct blk_plug_cb *cb;
1118 
1119 	if (!plug)
1120 		return NULL;
1121 
1122 	list_for_each_entry(cb, &plug->cb_list, list)
1123 		if (cb->callback == unplug && cb->data == data)
1124 			return cb;
1125 
1126 	/* Not currently on the callback list */
1127 	BUG_ON(size < sizeof(*cb));
1128 	cb = kzalloc(size, GFP_ATOMIC);
1129 	if (cb) {
1130 		cb->data = data;
1131 		cb->callback = unplug;
1132 		list_add(&cb->list, &plug->cb_list);
1133 	}
1134 	return cb;
1135 }
1136 EXPORT_SYMBOL(blk_check_plugged);
1137 
1138 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1139 {
1140 	if (!list_empty(&plug->cb_list))
1141 		flush_plug_callbacks(plug, from_schedule);
1142 	blk_mq_flush_plug_list(plug, from_schedule);
1143 	/*
1144 	 * Unconditionally flush out cached requests, even if the unplug
1145 	 * event came from schedule. Since we know hold references to the
1146 	 * queue for cached requests, we don't want a blocked task holding
1147 	 * up a queue freeze/quiesce event.
1148 	 */
1149 	if (unlikely(!rq_list_empty(plug->cached_rq)))
1150 		blk_mq_free_plug_rqs(plug);
1151 }
1152 
1153 /**
1154  * blk_finish_plug - mark the end of a batch of submitted I/O
1155  * @plug:	The &struct blk_plug passed to blk_start_plug()
1156  *
1157  * Description:
1158  * Indicate that a batch of I/O submissions is complete.  This function
1159  * must be paired with an initial call to blk_start_plug().  The intent
1160  * is to allow the block layer to optimize I/O submission.  See the
1161  * documentation for blk_start_plug() for more information.
1162  */
1163 void blk_finish_plug(struct blk_plug *plug)
1164 {
1165 	if (plug == current->plug) {
1166 		__blk_flush_plug(plug, false);
1167 		current->plug = NULL;
1168 	}
1169 }
1170 EXPORT_SYMBOL(blk_finish_plug);
1171 
1172 void blk_io_schedule(void)
1173 {
1174 	/* Prevent hang_check timer from firing at us during very long I/O */
1175 	unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1176 
1177 	if (timeout)
1178 		io_schedule_timeout(timeout);
1179 	else
1180 		io_schedule();
1181 }
1182 EXPORT_SYMBOL_GPL(blk_io_schedule);
1183 
1184 int __init blk_dev_init(void)
1185 {
1186 	BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS));
1187 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1188 			sizeof_field(struct request, cmd_flags));
1189 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1190 			sizeof_field(struct bio, bi_opf));
1191 
1192 	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
1193 	kblockd_workqueue = alloc_workqueue("kblockd",
1194 					    WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1195 	if (!kblockd_workqueue)
1196 		panic("Failed to create kblockd\n");
1197 
1198 	blk_requestq_cachep = kmem_cache_create("request_queue",
1199 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1200 
1201 	blk_debugfs_root = debugfs_create_dir("block", NULL);
1202 
1203 	return 0;
1204 }
1205