xref: /linux/block/blk-core.c (revision a0dd82d6d863f03859ccab6c71e84e917f4d7082)
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 #include "blk-ioprio.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 static DEFINE_IDA(blk_queue_ida);
64 
65 /*
66  * For queue allocation
67  */
68 static struct kmem_cache *blk_requestq_cachep;
69 
70 /*
71  * Controlling structure to kblockd
72  */
73 static struct workqueue_struct *kblockd_workqueue;
74 
75 /**
76  * blk_queue_flag_set - atomically set a queue flag
77  * @flag: flag to be set
78  * @q: request queue
79  */
80 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
81 {
82 	set_bit(flag, &q->queue_flags);
83 }
84 EXPORT_SYMBOL(blk_queue_flag_set);
85 
86 /**
87  * blk_queue_flag_clear - atomically clear a queue flag
88  * @flag: flag to be cleared
89  * @q: request queue
90  */
91 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
92 {
93 	clear_bit(flag, &q->queue_flags);
94 }
95 EXPORT_SYMBOL(blk_queue_flag_clear);
96 
97 /**
98  * blk_queue_flag_test_and_set - atomically test and set a queue flag
99  * @flag: flag to be set
100  * @q: request queue
101  *
102  * Returns the previous value of @flag - 0 if the flag was not set and 1 if
103  * the flag was already set.
104  */
105 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
106 {
107 	return test_and_set_bit(flag, &q->queue_flags);
108 }
109 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
110 
111 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
112 static const char *const blk_op_name[] = {
113 	REQ_OP_NAME(READ),
114 	REQ_OP_NAME(WRITE),
115 	REQ_OP_NAME(FLUSH),
116 	REQ_OP_NAME(DISCARD),
117 	REQ_OP_NAME(SECURE_ERASE),
118 	REQ_OP_NAME(ZONE_RESET),
119 	REQ_OP_NAME(ZONE_RESET_ALL),
120 	REQ_OP_NAME(ZONE_OPEN),
121 	REQ_OP_NAME(ZONE_CLOSE),
122 	REQ_OP_NAME(ZONE_FINISH),
123 	REQ_OP_NAME(ZONE_APPEND),
124 	REQ_OP_NAME(WRITE_ZEROES),
125 	REQ_OP_NAME(DRV_IN),
126 	REQ_OP_NAME(DRV_OUT),
127 };
128 #undef REQ_OP_NAME
129 
130 /**
131  * blk_op_str - Return string XXX in the REQ_OP_XXX.
132  * @op: REQ_OP_XXX.
133  *
134  * Description: Centralize block layer function to convert REQ_OP_XXX into
135  * string format. Useful in the debugging and tracing bio or request. For
136  * invalid REQ_OP_XXX it returns string "UNKNOWN".
137  */
138 inline const char *blk_op_str(enum req_op op)
139 {
140 	const char *op_str = "UNKNOWN";
141 
142 	if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
143 		op_str = blk_op_name[op];
144 
145 	return op_str;
146 }
147 EXPORT_SYMBOL_GPL(blk_op_str);
148 
149 static const struct {
150 	int		errno;
151 	const char	*name;
152 } blk_errors[] = {
153 	[BLK_STS_OK]		= { 0,		"" },
154 	[BLK_STS_NOTSUPP]	= { -EOPNOTSUPP, "operation not supported" },
155 	[BLK_STS_TIMEOUT]	= { -ETIMEDOUT,	"timeout" },
156 	[BLK_STS_NOSPC]		= { -ENOSPC,	"critical space allocation" },
157 	[BLK_STS_TRANSPORT]	= { -ENOLINK,	"recoverable transport" },
158 	[BLK_STS_TARGET]	= { -EREMOTEIO,	"critical target" },
159 	[BLK_STS_RESV_CONFLICT]	= { -EBADE,	"reservation conflict" },
160 	[BLK_STS_MEDIUM]	= { -ENODATA,	"critical medium" },
161 	[BLK_STS_PROTECTION]	= { -EILSEQ,	"protection" },
162 	[BLK_STS_RESOURCE]	= { -ENOMEM,	"kernel resource" },
163 	[BLK_STS_DEV_RESOURCE]	= { -EBUSY,	"device resource" },
164 	[BLK_STS_AGAIN]		= { -EAGAIN,	"nonblocking retry" },
165 	[BLK_STS_OFFLINE]	= { -ENODEV,	"device offline" },
166 
167 	/* device mapper special case, should not leak out: */
168 	[BLK_STS_DM_REQUEUE]	= { -EREMCHG, "dm internal retry" },
169 
170 	/* zone device specific errors */
171 	[BLK_STS_ZONE_OPEN_RESOURCE]	= { -ETOOMANYREFS, "open zones exceeded" },
172 	[BLK_STS_ZONE_ACTIVE_RESOURCE]	= { -EOVERFLOW, "active zones exceeded" },
173 
174 	/* Command duration limit device-side timeout */
175 	[BLK_STS_DURATION_LIMIT]	= { -ETIME, "duration limit exceeded" },
176 
177 	/* everything else not covered above: */
178 	[BLK_STS_IOERR]		= { -EIO,	"I/O" },
179 };
180 
181 blk_status_t errno_to_blk_status(int errno)
182 {
183 	int i;
184 
185 	for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
186 		if (blk_errors[i].errno == errno)
187 			return (__force blk_status_t)i;
188 	}
189 
190 	return BLK_STS_IOERR;
191 }
192 EXPORT_SYMBOL_GPL(errno_to_blk_status);
193 
194 int blk_status_to_errno(blk_status_t status)
195 {
196 	int idx = (__force int)status;
197 
198 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
199 		return -EIO;
200 	return blk_errors[idx].errno;
201 }
202 EXPORT_SYMBOL_GPL(blk_status_to_errno);
203 
204 const char *blk_status_to_str(blk_status_t status)
205 {
206 	int idx = (__force int)status;
207 
208 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
209 		return "<null>";
210 	return blk_errors[idx].name;
211 }
212 EXPORT_SYMBOL_GPL(blk_status_to_str);
213 
214 /**
215  * blk_sync_queue - cancel any pending callbacks on a queue
216  * @q: the queue
217  *
218  * Description:
219  *     The block layer may perform asynchronous callback activity
220  *     on a queue, such as calling the unplug function after a timeout.
221  *     A block device may call blk_sync_queue to ensure that any
222  *     such activity is cancelled, thus allowing it to release resources
223  *     that the callbacks might use. The caller must already have made sure
224  *     that its ->submit_bio will not re-add plugging prior to calling
225  *     this function.
226  *
227  *     This function does not cancel any asynchronous activity arising
228  *     out of elevator or throttling code. That would require elevator_exit()
229  *     and blkcg_exit_queue() to be called with queue lock initialized.
230  *
231  */
232 void blk_sync_queue(struct request_queue *q)
233 {
234 	del_timer_sync(&q->timeout);
235 	cancel_work_sync(&q->timeout_work);
236 }
237 EXPORT_SYMBOL(blk_sync_queue);
238 
239 /**
240  * blk_set_pm_only - increment pm_only counter
241  * @q: request queue pointer
242  */
243 void blk_set_pm_only(struct request_queue *q)
244 {
245 	atomic_inc(&q->pm_only);
246 }
247 EXPORT_SYMBOL_GPL(blk_set_pm_only);
248 
249 void blk_clear_pm_only(struct request_queue *q)
250 {
251 	int pm_only;
252 
253 	pm_only = atomic_dec_return(&q->pm_only);
254 	WARN_ON_ONCE(pm_only < 0);
255 	if (pm_only == 0)
256 		wake_up_all(&q->mq_freeze_wq);
257 }
258 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
259 
260 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
261 {
262 	struct request_queue *q = container_of(rcu_head,
263 			struct request_queue, rcu_head);
264 
265 	percpu_ref_exit(&q->q_usage_counter);
266 	kmem_cache_free(blk_requestq_cachep, q);
267 }
268 
269 static void blk_free_queue(struct request_queue *q)
270 {
271 	blk_free_queue_stats(q->stats);
272 	if (queue_is_mq(q))
273 		blk_mq_release(q);
274 
275 	ida_free(&blk_queue_ida, q->id);
276 	call_rcu(&q->rcu_head, blk_free_queue_rcu);
277 }
278 
279 /**
280  * blk_put_queue - decrement the request_queue refcount
281  * @q: the request_queue structure to decrement the refcount for
282  *
283  * Decrements the refcount of the request_queue and free it when the refcount
284  * reaches 0.
285  */
286 void blk_put_queue(struct request_queue *q)
287 {
288 	if (refcount_dec_and_test(&q->refs))
289 		blk_free_queue(q);
290 }
291 EXPORT_SYMBOL(blk_put_queue);
292 
293 void blk_queue_start_drain(struct request_queue *q)
294 {
295 	/*
296 	 * When queue DYING flag is set, we need to block new req
297 	 * entering queue, so we call blk_freeze_queue_start() to
298 	 * prevent I/O from crossing blk_queue_enter().
299 	 */
300 	blk_freeze_queue_start(q);
301 	if (queue_is_mq(q))
302 		blk_mq_wake_waiters(q);
303 	/* Make blk_queue_enter() reexamine the DYING flag. */
304 	wake_up_all(&q->mq_freeze_wq);
305 }
306 
307 /**
308  * blk_queue_enter() - try to increase q->q_usage_counter
309  * @q: request queue pointer
310  * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
311  */
312 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
313 {
314 	const bool pm = flags & BLK_MQ_REQ_PM;
315 
316 	while (!blk_try_enter_queue(q, pm)) {
317 		if (flags & BLK_MQ_REQ_NOWAIT)
318 			return -EAGAIN;
319 
320 		/*
321 		 * read pair of barrier in blk_freeze_queue_start(), we need to
322 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
323 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
324 		 * following wait may never return if the two reads are
325 		 * reordered.
326 		 */
327 		smp_rmb();
328 		wait_event(q->mq_freeze_wq,
329 			   (!q->mq_freeze_depth &&
330 			    blk_pm_resume_queue(pm, q)) ||
331 			   blk_queue_dying(q));
332 		if (blk_queue_dying(q))
333 			return -ENODEV;
334 	}
335 
336 	return 0;
337 }
338 
339 int __bio_queue_enter(struct request_queue *q, struct bio *bio)
340 {
341 	while (!blk_try_enter_queue(q, false)) {
342 		struct gendisk *disk = bio->bi_bdev->bd_disk;
343 
344 		if (bio->bi_opf & REQ_NOWAIT) {
345 			if (test_bit(GD_DEAD, &disk->state))
346 				goto dead;
347 			bio_wouldblock_error(bio);
348 			return -EAGAIN;
349 		}
350 
351 		/*
352 		 * read pair of barrier in blk_freeze_queue_start(), we need to
353 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
354 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
355 		 * following wait may never return if the two reads are
356 		 * reordered.
357 		 */
358 		smp_rmb();
359 		wait_event(q->mq_freeze_wq,
360 			   (!q->mq_freeze_depth &&
361 			    blk_pm_resume_queue(false, q)) ||
362 			   test_bit(GD_DEAD, &disk->state));
363 		if (test_bit(GD_DEAD, &disk->state))
364 			goto dead;
365 	}
366 
367 	return 0;
368 dead:
369 	bio_io_error(bio);
370 	return -ENODEV;
371 }
372 
373 void blk_queue_exit(struct request_queue *q)
374 {
375 	percpu_ref_put(&q->q_usage_counter);
376 }
377 
378 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
379 {
380 	struct request_queue *q =
381 		container_of(ref, struct request_queue, q_usage_counter);
382 
383 	wake_up_all(&q->mq_freeze_wq);
384 }
385 
386 static void blk_rq_timed_out_timer(struct timer_list *t)
387 {
388 	struct request_queue *q = from_timer(q, t, timeout);
389 
390 	kblockd_schedule_work(&q->timeout_work);
391 }
392 
393 static void blk_timeout_work(struct work_struct *work)
394 {
395 }
396 
397 struct request_queue *blk_alloc_queue(int node_id)
398 {
399 	struct request_queue *q;
400 
401 	q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO,
402 				  node_id);
403 	if (!q)
404 		return NULL;
405 
406 	q->last_merge = NULL;
407 
408 	q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL);
409 	if (q->id < 0)
410 		goto fail_q;
411 
412 	q->stats = blk_alloc_queue_stats();
413 	if (!q->stats)
414 		goto fail_id;
415 
416 	q->node = node_id;
417 
418 	atomic_set(&q->nr_active_requests_shared_tags, 0);
419 
420 	timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
421 	INIT_WORK(&q->timeout_work, blk_timeout_work);
422 	INIT_LIST_HEAD(&q->icq_list);
423 
424 	refcount_set(&q->refs, 1);
425 	mutex_init(&q->debugfs_mutex);
426 	mutex_init(&q->sysfs_lock);
427 	mutex_init(&q->sysfs_dir_lock);
428 	mutex_init(&q->rq_qos_mutex);
429 	spin_lock_init(&q->queue_lock);
430 
431 	init_waitqueue_head(&q->mq_freeze_wq);
432 	mutex_init(&q->mq_freeze_lock);
433 
434 	/*
435 	 * Init percpu_ref in atomic mode so that it's faster to shutdown.
436 	 * See blk_register_queue() for details.
437 	 */
438 	if (percpu_ref_init(&q->q_usage_counter,
439 				blk_queue_usage_counter_release,
440 				PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
441 		goto fail_stats;
442 
443 	blk_set_default_limits(&q->limits);
444 	q->nr_requests = BLKDEV_DEFAULT_RQ;
445 
446 	return q;
447 
448 fail_stats:
449 	blk_free_queue_stats(q->stats);
450 fail_id:
451 	ida_free(&blk_queue_ida, q->id);
452 fail_q:
453 	kmem_cache_free(blk_requestq_cachep, q);
454 	return NULL;
455 }
456 
457 /**
458  * blk_get_queue - increment the request_queue refcount
459  * @q: the request_queue structure to increment the refcount for
460  *
461  * Increment the refcount of the request_queue kobject.
462  *
463  * Context: Any context.
464  */
465 bool blk_get_queue(struct request_queue *q)
466 {
467 	if (unlikely(blk_queue_dying(q)))
468 		return false;
469 	refcount_inc(&q->refs);
470 	return true;
471 }
472 EXPORT_SYMBOL(blk_get_queue);
473 
474 #ifdef CONFIG_FAIL_MAKE_REQUEST
475 
476 static DECLARE_FAULT_ATTR(fail_make_request);
477 
478 static int __init setup_fail_make_request(char *str)
479 {
480 	return setup_fault_attr(&fail_make_request, str);
481 }
482 __setup("fail_make_request=", setup_fail_make_request);
483 
484 bool should_fail_request(struct block_device *part, unsigned int bytes)
485 {
486 	return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
487 }
488 
489 static int __init fail_make_request_debugfs(void)
490 {
491 	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
492 						NULL, &fail_make_request);
493 
494 	return PTR_ERR_OR_ZERO(dir);
495 }
496 
497 late_initcall(fail_make_request_debugfs);
498 #endif /* CONFIG_FAIL_MAKE_REQUEST */
499 
500 static inline void bio_check_ro(struct bio *bio)
501 {
502 	if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
503 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
504 			return;
505 
506 		if (bio->bi_bdev->bd_ro_warned)
507 			return;
508 
509 		bio->bi_bdev->bd_ro_warned = true;
510 		/*
511 		 * Use ioctl to set underlying disk of raid/dm to read-only
512 		 * will trigger this.
513 		 */
514 		pr_warn("Trying to write to read-only block-device %pg\n",
515 			bio->bi_bdev);
516 	}
517 }
518 
519 static noinline int should_fail_bio(struct bio *bio)
520 {
521 	if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
522 		return -EIO;
523 	return 0;
524 }
525 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
526 
527 /*
528  * Check whether this bio extends beyond the end of the device or partition.
529  * This may well happen - the kernel calls bread() without checking the size of
530  * the device, e.g., when mounting a file system.
531  */
532 static inline int bio_check_eod(struct bio *bio)
533 {
534 	sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
535 	unsigned int nr_sectors = bio_sectors(bio);
536 
537 	if (nr_sectors &&
538 	    (nr_sectors > maxsector ||
539 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
540 		pr_info_ratelimited("%s: attempt to access beyond end of device\n"
541 				    "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
542 				    current->comm, bio->bi_bdev, bio->bi_opf,
543 				    bio->bi_iter.bi_sector, nr_sectors, maxsector);
544 		return -EIO;
545 	}
546 	return 0;
547 }
548 
549 /*
550  * Remap block n of partition p to block n+start(p) of the disk.
551  */
552 static int blk_partition_remap(struct bio *bio)
553 {
554 	struct block_device *p = bio->bi_bdev;
555 
556 	if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
557 		return -EIO;
558 	if (bio_sectors(bio)) {
559 		bio->bi_iter.bi_sector += p->bd_start_sect;
560 		trace_block_bio_remap(bio, p->bd_dev,
561 				      bio->bi_iter.bi_sector -
562 				      p->bd_start_sect);
563 	}
564 	bio_set_flag(bio, BIO_REMAPPED);
565 	return 0;
566 }
567 
568 /*
569  * Check write append to a zoned block device.
570  */
571 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
572 						 struct bio *bio)
573 {
574 	int nr_sectors = bio_sectors(bio);
575 
576 	/* Only applicable to zoned block devices */
577 	if (!bdev_is_zoned(bio->bi_bdev))
578 		return BLK_STS_NOTSUPP;
579 
580 	/* The bio sector must point to the start of a sequential zone */
581 	if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector) ||
582 	    !bio_zone_is_seq(bio))
583 		return BLK_STS_IOERR;
584 
585 	/*
586 	 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
587 	 * split and could result in non-contiguous sectors being written in
588 	 * different zones.
589 	 */
590 	if (nr_sectors > q->limits.chunk_sectors)
591 		return BLK_STS_IOERR;
592 
593 	/* Make sure the BIO is small enough and will not get split */
594 	if (nr_sectors > q->limits.max_zone_append_sectors)
595 		return BLK_STS_IOERR;
596 
597 	bio->bi_opf |= REQ_NOMERGE;
598 
599 	return BLK_STS_OK;
600 }
601 
602 static void __submit_bio(struct bio *bio)
603 {
604 	if (unlikely(!blk_crypto_bio_prep(&bio)))
605 		return;
606 
607 	if (!bio->bi_bdev->bd_has_submit_bio) {
608 		blk_mq_submit_bio(bio);
609 	} else if (likely(bio_queue_enter(bio) == 0)) {
610 		struct gendisk *disk = bio->bi_bdev->bd_disk;
611 
612 		disk->fops->submit_bio(bio);
613 		blk_queue_exit(disk->queue);
614 	}
615 }
616 
617 /*
618  * The loop in this function may be a bit non-obvious, and so deserves some
619  * explanation:
620  *
621  *  - Before entering the loop, bio->bi_next is NULL (as all callers ensure
622  *    that), so we have a list with a single bio.
623  *  - We pretend that we have just taken it off a longer list, so we assign
624  *    bio_list to a pointer to the bio_list_on_stack, thus initialising the
625  *    bio_list of new bios to be added.  ->submit_bio() may indeed add some more
626  *    bios through a recursive call to submit_bio_noacct.  If it did, we find a
627  *    non-NULL value in bio_list and re-enter the loop from the top.
628  *  - In this case we really did just take the bio of the top of the list (no
629  *    pretending) and so remove it from bio_list, and call into ->submit_bio()
630  *    again.
631  *
632  * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
633  * bio_list_on_stack[1] contains bios that were submitted before the current
634  *	->submit_bio, but that haven't been processed yet.
635  */
636 static void __submit_bio_noacct(struct bio *bio)
637 {
638 	struct bio_list bio_list_on_stack[2];
639 
640 	BUG_ON(bio->bi_next);
641 
642 	bio_list_init(&bio_list_on_stack[0]);
643 	current->bio_list = bio_list_on_stack;
644 
645 	do {
646 		struct request_queue *q = bdev_get_queue(bio->bi_bdev);
647 		struct bio_list lower, same;
648 
649 		/*
650 		 * Create a fresh bio_list for all subordinate requests.
651 		 */
652 		bio_list_on_stack[1] = bio_list_on_stack[0];
653 		bio_list_init(&bio_list_on_stack[0]);
654 
655 		__submit_bio(bio);
656 
657 		/*
658 		 * Sort new bios into those for a lower level and those for the
659 		 * same level.
660 		 */
661 		bio_list_init(&lower);
662 		bio_list_init(&same);
663 		while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
664 			if (q == bdev_get_queue(bio->bi_bdev))
665 				bio_list_add(&same, bio);
666 			else
667 				bio_list_add(&lower, bio);
668 
669 		/*
670 		 * Now assemble so we handle the lowest level first.
671 		 */
672 		bio_list_merge(&bio_list_on_stack[0], &lower);
673 		bio_list_merge(&bio_list_on_stack[0], &same);
674 		bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
675 	} while ((bio = bio_list_pop(&bio_list_on_stack[0])));
676 
677 	current->bio_list = NULL;
678 }
679 
680 static void __submit_bio_noacct_mq(struct bio *bio)
681 {
682 	struct bio_list bio_list[2] = { };
683 
684 	current->bio_list = bio_list;
685 
686 	do {
687 		__submit_bio(bio);
688 	} while ((bio = bio_list_pop(&bio_list[0])));
689 
690 	current->bio_list = NULL;
691 }
692 
693 void submit_bio_noacct_nocheck(struct bio *bio)
694 {
695 	blk_cgroup_bio_start(bio);
696 	blkcg_bio_issue_init(bio);
697 
698 	if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
699 		trace_block_bio_queue(bio);
700 		/*
701 		 * Now that enqueuing has been traced, we need to trace
702 		 * completion as well.
703 		 */
704 		bio_set_flag(bio, BIO_TRACE_COMPLETION);
705 	}
706 
707 	/*
708 	 * We only want one ->submit_bio to be active at a time, else stack
709 	 * usage with stacked devices could be a problem.  Use current->bio_list
710 	 * to collect a list of requests submited by a ->submit_bio method while
711 	 * it is active, and then process them after it returned.
712 	 */
713 	if (current->bio_list)
714 		bio_list_add(&current->bio_list[0], bio);
715 	else if (!bio->bi_bdev->bd_has_submit_bio)
716 		__submit_bio_noacct_mq(bio);
717 	else
718 		__submit_bio_noacct(bio);
719 }
720 
721 /**
722  * submit_bio_noacct - re-submit a bio to the block device layer for I/O
723  * @bio:  The bio describing the location in memory and on the device.
724  *
725  * This is a version of submit_bio() that shall only be used for I/O that is
726  * resubmitted to lower level drivers by stacking block drivers.  All file
727  * systems and other upper level users of the block layer should use
728  * submit_bio() instead.
729  */
730 void submit_bio_noacct(struct bio *bio)
731 {
732 	struct block_device *bdev = bio->bi_bdev;
733 	struct request_queue *q = bdev_get_queue(bdev);
734 	blk_status_t status = BLK_STS_IOERR;
735 
736 	might_sleep();
737 
738 	/*
739 	 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
740 	 * if queue does not support NOWAIT.
741 	 */
742 	if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev))
743 		goto not_supported;
744 
745 	if (should_fail_bio(bio))
746 		goto end_io;
747 	bio_check_ro(bio);
748 	if (!bio_flagged(bio, BIO_REMAPPED)) {
749 		if (unlikely(bio_check_eod(bio)))
750 			goto end_io;
751 		if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
752 			goto end_io;
753 	}
754 
755 	/*
756 	 * Filter flush bio's early so that bio based drivers without flush
757 	 * support don't have to worry about them.
758 	 */
759 	if (op_is_flush(bio->bi_opf)) {
760 		if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE &&
761 				 bio_op(bio) != REQ_OP_ZONE_APPEND))
762 			goto end_io;
763 		if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
764 			bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
765 			if (!bio_sectors(bio)) {
766 				status = BLK_STS_OK;
767 				goto end_io;
768 			}
769 		}
770 	}
771 
772 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
773 		bio_clear_polled(bio);
774 
775 	switch (bio_op(bio)) {
776 	case REQ_OP_READ:
777 	case REQ_OP_WRITE:
778 		break;
779 	case REQ_OP_FLUSH:
780 		/*
781 		 * REQ_OP_FLUSH can't be submitted through bios, it is only
782 		 * synthetized in struct request by the flush state machine.
783 		 */
784 		goto not_supported;
785 	case REQ_OP_DISCARD:
786 		if (!bdev_max_discard_sectors(bdev))
787 			goto not_supported;
788 		break;
789 	case REQ_OP_SECURE_ERASE:
790 		if (!bdev_max_secure_erase_sectors(bdev))
791 			goto not_supported;
792 		break;
793 	case REQ_OP_ZONE_APPEND:
794 		status = blk_check_zone_append(q, bio);
795 		if (status != BLK_STS_OK)
796 			goto end_io;
797 		break;
798 	case REQ_OP_WRITE_ZEROES:
799 		if (!q->limits.max_write_zeroes_sectors)
800 			goto not_supported;
801 		break;
802 	case REQ_OP_ZONE_RESET:
803 	case REQ_OP_ZONE_OPEN:
804 	case REQ_OP_ZONE_CLOSE:
805 	case REQ_OP_ZONE_FINISH:
806 		if (!bdev_is_zoned(bio->bi_bdev))
807 			goto not_supported;
808 		break;
809 	case REQ_OP_ZONE_RESET_ALL:
810 		if (!bdev_is_zoned(bio->bi_bdev) || !blk_queue_zone_resetall(q))
811 			goto not_supported;
812 		break;
813 	case REQ_OP_DRV_IN:
814 	case REQ_OP_DRV_OUT:
815 		/*
816 		 * Driver private operations are only used with passthrough
817 		 * requests.
818 		 */
819 		fallthrough;
820 	default:
821 		goto not_supported;
822 	}
823 
824 	if (blk_throtl_bio(bio))
825 		return;
826 	submit_bio_noacct_nocheck(bio);
827 	return;
828 
829 not_supported:
830 	status = BLK_STS_NOTSUPP;
831 end_io:
832 	bio->bi_status = status;
833 	bio_endio(bio);
834 }
835 EXPORT_SYMBOL(submit_bio_noacct);
836 
837 static void bio_set_ioprio(struct bio *bio)
838 {
839 	/* Nobody set ioprio so far? Initialize it based on task's nice value */
840 	if (IOPRIO_PRIO_CLASS(bio->bi_ioprio) == IOPRIO_CLASS_NONE)
841 		bio->bi_ioprio = get_current_ioprio();
842 	blkcg_set_ioprio(bio);
843 }
844 
845 /**
846  * submit_bio - submit a bio to the block device layer for I/O
847  * @bio: The &struct bio which describes the I/O
848  *
849  * submit_bio() is used to submit I/O requests to block devices.  It is passed a
850  * fully set up &struct bio that describes the I/O that needs to be done.  The
851  * bio will be send to the device described by the bi_bdev field.
852  *
853  * The success/failure status of the request, along with notification of
854  * completion, is delivered asynchronously through the ->bi_end_io() callback
855  * in @bio.  The bio must NOT be touched by the caller until ->bi_end_io() has
856  * been called.
857  */
858 void submit_bio(struct bio *bio)
859 {
860 	if (bio_op(bio) == REQ_OP_READ) {
861 		task_io_account_read(bio->bi_iter.bi_size);
862 		count_vm_events(PGPGIN, bio_sectors(bio));
863 	} else if (bio_op(bio) == REQ_OP_WRITE) {
864 		count_vm_events(PGPGOUT, bio_sectors(bio));
865 	}
866 
867 	bio_set_ioprio(bio);
868 	submit_bio_noacct(bio);
869 }
870 EXPORT_SYMBOL(submit_bio);
871 
872 /**
873  * bio_poll - poll for BIO completions
874  * @bio: bio to poll for
875  * @iob: batches of IO
876  * @flags: BLK_POLL_* flags that control the behavior
877  *
878  * Poll for completions on queue associated with the bio. Returns number of
879  * completed entries found.
880  *
881  * Note: the caller must either be the context that submitted @bio, or
882  * be in a RCU critical section to prevent freeing of @bio.
883  */
884 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
885 {
886 	blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
887 	struct block_device *bdev;
888 	struct request_queue *q;
889 	int ret = 0;
890 
891 	bdev = READ_ONCE(bio->bi_bdev);
892 	if (!bdev)
893 		return 0;
894 
895 	q = bdev_get_queue(bdev);
896 	if (cookie == BLK_QC_T_NONE ||
897 	    !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
898 		return 0;
899 
900 	/*
901 	 * As the requests that require a zone lock are not plugged in the
902 	 * first place, directly accessing the plug instead of using
903 	 * blk_mq_plug() should not have any consequences during flushing for
904 	 * zoned devices.
905 	 */
906 	blk_flush_plug(current->plug, false);
907 
908 	/*
909 	 * We need to be able to enter a frozen queue, similar to how
910 	 * timeouts also need to do that. If that is blocked, then we can
911 	 * have pending IO when a queue freeze is started, and then the
912 	 * wait for the freeze to finish will wait for polled requests to
913 	 * timeout as the poller is preventer from entering the queue and
914 	 * completing them. As long as we prevent new IO from being queued,
915 	 * that should be all that matters.
916 	 */
917 	if (!percpu_ref_tryget(&q->q_usage_counter))
918 		return 0;
919 	if (queue_is_mq(q)) {
920 		ret = blk_mq_poll(q, cookie, iob, flags);
921 	} else {
922 		struct gendisk *disk = q->disk;
923 
924 		if (disk && disk->fops->poll_bio)
925 			ret = disk->fops->poll_bio(bio, iob, flags);
926 	}
927 	blk_queue_exit(q);
928 	return ret;
929 }
930 EXPORT_SYMBOL_GPL(bio_poll);
931 
932 /*
933  * Helper to implement file_operations.iopoll.  Requires the bio to be stored
934  * in iocb->private, and cleared before freeing the bio.
935  */
936 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
937 		    unsigned int flags)
938 {
939 	struct bio *bio;
940 	int ret = 0;
941 
942 	/*
943 	 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
944 	 * point to a freshly allocated bio at this point.  If that happens
945 	 * we have a few cases to consider:
946 	 *
947 	 *  1) the bio is beeing initialized and bi_bdev is NULL.  We can just
948 	 *     simply nothing in this case
949 	 *  2) the bio points to a not poll enabled device.  bio_poll will catch
950 	 *     this and return 0
951 	 *  3) the bio points to a poll capable device, including but not
952 	 *     limited to the one that the original bio pointed to.  In this
953 	 *     case we will call into the actual poll method and poll for I/O,
954 	 *     even if we don't need to, but it won't cause harm either.
955 	 *
956 	 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
957 	 * is still allocated. Because partitions hold a reference to the whole
958 	 * device bdev and thus disk, the disk is also still valid.  Grabbing
959 	 * a reference to the queue in bio_poll() ensures the hctxs and requests
960 	 * are still valid as well.
961 	 */
962 	rcu_read_lock();
963 	bio = READ_ONCE(kiocb->private);
964 	if (bio)
965 		ret = bio_poll(bio, iob, flags);
966 	rcu_read_unlock();
967 
968 	return ret;
969 }
970 EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
971 
972 void update_io_ticks(struct block_device *part, unsigned long now, bool end)
973 {
974 	unsigned long stamp;
975 again:
976 	stamp = READ_ONCE(part->bd_stamp);
977 	if (unlikely(time_after(now, stamp))) {
978 		if (likely(try_cmpxchg(&part->bd_stamp, &stamp, now)))
979 			__part_stat_add(part, io_ticks, end ? now - stamp : 1);
980 	}
981 	if (part->bd_partno) {
982 		part = bdev_whole(part);
983 		goto again;
984 	}
985 }
986 
987 unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op,
988 				 unsigned long start_time)
989 {
990 	part_stat_lock();
991 	update_io_ticks(bdev, start_time, false);
992 	part_stat_local_inc(bdev, in_flight[op_is_write(op)]);
993 	part_stat_unlock();
994 
995 	return start_time;
996 }
997 EXPORT_SYMBOL(bdev_start_io_acct);
998 
999 /**
1000  * bio_start_io_acct - start I/O accounting for bio based drivers
1001  * @bio:	bio to start account for
1002  *
1003  * Returns the start time that should be passed back to bio_end_io_acct().
1004  */
1005 unsigned long bio_start_io_acct(struct bio *bio)
1006 {
1007 	return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies);
1008 }
1009 EXPORT_SYMBOL_GPL(bio_start_io_acct);
1010 
1011 void bdev_end_io_acct(struct block_device *bdev, enum req_op op,
1012 		      unsigned int sectors, unsigned long start_time)
1013 {
1014 	const int sgrp = op_stat_group(op);
1015 	unsigned long now = READ_ONCE(jiffies);
1016 	unsigned long duration = now - start_time;
1017 
1018 	part_stat_lock();
1019 	update_io_ticks(bdev, now, true);
1020 	part_stat_inc(bdev, ios[sgrp]);
1021 	part_stat_add(bdev, sectors[sgrp], sectors);
1022 	part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration));
1023 	part_stat_local_dec(bdev, in_flight[op_is_write(op)]);
1024 	part_stat_unlock();
1025 }
1026 EXPORT_SYMBOL(bdev_end_io_acct);
1027 
1028 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
1029 			      struct block_device *orig_bdev)
1030 {
1031 	bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time);
1032 }
1033 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1034 
1035 /**
1036  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1037  * @q : the queue of the device being checked
1038  *
1039  * Description:
1040  *    Check if underlying low-level drivers of a device are busy.
1041  *    If the drivers want to export their busy state, they must set own
1042  *    exporting function using blk_queue_lld_busy() first.
1043  *
1044  *    Basically, this function is used only by request stacking drivers
1045  *    to stop dispatching requests to underlying devices when underlying
1046  *    devices are busy.  This behavior helps more I/O merging on the queue
1047  *    of the request stacking driver and prevents I/O throughput regression
1048  *    on burst I/O load.
1049  *
1050  * Return:
1051  *    0 - Not busy (The request stacking driver should dispatch request)
1052  *    1 - Busy (The request stacking driver should stop dispatching request)
1053  */
1054 int blk_lld_busy(struct request_queue *q)
1055 {
1056 	if (queue_is_mq(q) && q->mq_ops->busy)
1057 		return q->mq_ops->busy(q);
1058 
1059 	return 0;
1060 }
1061 EXPORT_SYMBOL_GPL(blk_lld_busy);
1062 
1063 int kblockd_schedule_work(struct work_struct *work)
1064 {
1065 	return queue_work(kblockd_workqueue, work);
1066 }
1067 EXPORT_SYMBOL(kblockd_schedule_work);
1068 
1069 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1070 				unsigned long delay)
1071 {
1072 	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1073 }
1074 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1075 
1076 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1077 {
1078 	struct task_struct *tsk = current;
1079 
1080 	/*
1081 	 * If this is a nested plug, don't actually assign it.
1082 	 */
1083 	if (tsk->plug)
1084 		return;
1085 
1086 	plug->mq_list = NULL;
1087 	plug->cached_rq = NULL;
1088 	plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1089 	plug->rq_count = 0;
1090 	plug->multiple_queues = false;
1091 	plug->has_elevator = false;
1092 	INIT_LIST_HEAD(&plug->cb_list);
1093 
1094 	/*
1095 	 * Store ordering should not be needed here, since a potential
1096 	 * preempt will imply a full memory barrier
1097 	 */
1098 	tsk->plug = plug;
1099 }
1100 
1101 /**
1102  * blk_start_plug - initialize blk_plug and track it inside the task_struct
1103  * @plug:	The &struct blk_plug that needs to be initialized
1104  *
1105  * Description:
1106  *   blk_start_plug() indicates to the block layer an intent by the caller
1107  *   to submit multiple I/O requests in a batch.  The block layer may use
1108  *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1109  *   is called.  However, the block layer may choose to submit requests
1110  *   before a call to blk_finish_plug() if the number of queued I/Os
1111  *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1112  *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1113  *   the task schedules (see below).
1114  *
1115  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1116  *   pending I/O should the task end up blocking between blk_start_plug() and
1117  *   blk_finish_plug(). This is important from a performance perspective, but
1118  *   also ensures that we don't deadlock. For instance, if the task is blocking
1119  *   for a memory allocation, memory reclaim could end up wanting to free a
1120  *   page belonging to that request that is currently residing in our private
1121  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1122  *   this kind of deadlock.
1123  */
1124 void blk_start_plug(struct blk_plug *plug)
1125 {
1126 	blk_start_plug_nr_ios(plug, 1);
1127 }
1128 EXPORT_SYMBOL(blk_start_plug);
1129 
1130 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1131 {
1132 	LIST_HEAD(callbacks);
1133 
1134 	while (!list_empty(&plug->cb_list)) {
1135 		list_splice_init(&plug->cb_list, &callbacks);
1136 
1137 		while (!list_empty(&callbacks)) {
1138 			struct blk_plug_cb *cb = list_first_entry(&callbacks,
1139 							  struct blk_plug_cb,
1140 							  list);
1141 			list_del(&cb->list);
1142 			cb->callback(cb, from_schedule);
1143 		}
1144 	}
1145 }
1146 
1147 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1148 				      int size)
1149 {
1150 	struct blk_plug *plug = current->plug;
1151 	struct blk_plug_cb *cb;
1152 
1153 	if (!plug)
1154 		return NULL;
1155 
1156 	list_for_each_entry(cb, &plug->cb_list, list)
1157 		if (cb->callback == unplug && cb->data == data)
1158 			return cb;
1159 
1160 	/* Not currently on the callback list */
1161 	BUG_ON(size < sizeof(*cb));
1162 	cb = kzalloc(size, GFP_ATOMIC);
1163 	if (cb) {
1164 		cb->data = data;
1165 		cb->callback = unplug;
1166 		list_add(&cb->list, &plug->cb_list);
1167 	}
1168 	return cb;
1169 }
1170 EXPORT_SYMBOL(blk_check_plugged);
1171 
1172 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1173 {
1174 	if (!list_empty(&plug->cb_list))
1175 		flush_plug_callbacks(plug, from_schedule);
1176 	blk_mq_flush_plug_list(plug, from_schedule);
1177 	/*
1178 	 * Unconditionally flush out cached requests, even if the unplug
1179 	 * event came from schedule. Since we know hold references to the
1180 	 * queue for cached requests, we don't want a blocked task holding
1181 	 * up a queue freeze/quiesce event.
1182 	 */
1183 	if (unlikely(!rq_list_empty(plug->cached_rq)))
1184 		blk_mq_free_plug_rqs(plug);
1185 }
1186 
1187 /**
1188  * blk_finish_plug - mark the end of a batch of submitted I/O
1189  * @plug:	The &struct blk_plug passed to blk_start_plug()
1190  *
1191  * Description:
1192  * Indicate that a batch of I/O submissions is complete.  This function
1193  * must be paired with an initial call to blk_start_plug().  The intent
1194  * is to allow the block layer to optimize I/O submission.  See the
1195  * documentation for blk_start_plug() for more information.
1196  */
1197 void blk_finish_plug(struct blk_plug *plug)
1198 {
1199 	if (plug == current->plug) {
1200 		__blk_flush_plug(plug, false);
1201 		current->plug = NULL;
1202 	}
1203 }
1204 EXPORT_SYMBOL(blk_finish_plug);
1205 
1206 void blk_io_schedule(void)
1207 {
1208 	/* Prevent hang_check timer from firing at us during very long I/O */
1209 	unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1210 
1211 	if (timeout)
1212 		io_schedule_timeout(timeout);
1213 	else
1214 		io_schedule();
1215 }
1216 EXPORT_SYMBOL_GPL(blk_io_schedule);
1217 
1218 int __init blk_dev_init(void)
1219 {
1220 	BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS));
1221 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1222 			sizeof_field(struct request, cmd_flags));
1223 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1224 			sizeof_field(struct bio, bi_opf));
1225 
1226 	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
1227 	kblockd_workqueue = alloc_workqueue("kblockd",
1228 					    WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1229 	if (!kblockd_workqueue)
1230 		panic("Failed to create kblockd\n");
1231 
1232 	blk_requestq_cachep = kmem_cache_create("request_queue",
1233 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1234 
1235 	blk_debugfs_root = debugfs_create_dir("block", NULL);
1236 
1237 	return 0;
1238 }
1239