xref: /linux/block/blk-core.c (revision 5dfe7a7e52ccdf60dfd11ccbe509e4365ea721ca)
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_NEXUS]		= { -EBADE,	"critical nexus" },
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 	/* everything else not covered above: */
174 	[BLK_STS_IOERR]		= { -EIO,	"I/O" },
175 };
176 
177 blk_status_t errno_to_blk_status(int errno)
178 {
179 	int i;
180 
181 	for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
182 		if (blk_errors[i].errno == errno)
183 			return (__force blk_status_t)i;
184 	}
185 
186 	return BLK_STS_IOERR;
187 }
188 EXPORT_SYMBOL_GPL(errno_to_blk_status);
189 
190 int blk_status_to_errno(blk_status_t status)
191 {
192 	int idx = (__force int)status;
193 
194 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
195 		return -EIO;
196 	return blk_errors[idx].errno;
197 }
198 EXPORT_SYMBOL_GPL(blk_status_to_errno);
199 
200 const char *blk_status_to_str(blk_status_t status)
201 {
202 	int idx = (__force int)status;
203 
204 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
205 		return "<null>";
206 	return blk_errors[idx].name;
207 }
208 
209 /**
210  * blk_sync_queue - cancel any pending callbacks on a queue
211  * @q: the queue
212  *
213  * Description:
214  *     The block layer may perform asynchronous callback activity
215  *     on a queue, such as calling the unplug function after a timeout.
216  *     A block device may call blk_sync_queue to ensure that any
217  *     such activity is cancelled, thus allowing it to release resources
218  *     that the callbacks might use. The caller must already have made sure
219  *     that its ->submit_bio will not re-add plugging prior to calling
220  *     this function.
221  *
222  *     This function does not cancel any asynchronous activity arising
223  *     out of elevator or throttling code. That would require elevator_exit()
224  *     and blkcg_exit_queue() to be called with queue lock initialized.
225  *
226  */
227 void blk_sync_queue(struct request_queue *q)
228 {
229 	del_timer_sync(&q->timeout);
230 	cancel_work_sync(&q->timeout_work);
231 }
232 EXPORT_SYMBOL(blk_sync_queue);
233 
234 /**
235  * blk_set_pm_only - increment pm_only counter
236  * @q: request queue pointer
237  */
238 void blk_set_pm_only(struct request_queue *q)
239 {
240 	atomic_inc(&q->pm_only);
241 }
242 EXPORT_SYMBOL_GPL(blk_set_pm_only);
243 
244 void blk_clear_pm_only(struct request_queue *q)
245 {
246 	int pm_only;
247 
248 	pm_only = atomic_dec_return(&q->pm_only);
249 	WARN_ON_ONCE(pm_only < 0);
250 	if (pm_only == 0)
251 		wake_up_all(&q->mq_freeze_wq);
252 }
253 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
254 
255 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
256 {
257 	struct request_queue *q = container_of(rcu_head,
258 			struct request_queue, rcu_head);
259 
260 	percpu_ref_exit(&q->q_usage_counter);
261 	kmem_cache_free(blk_requestq_cachep, q);
262 }
263 
264 static void blk_free_queue(struct request_queue *q)
265 {
266 	blk_free_queue_stats(q->stats);
267 	if (queue_is_mq(q))
268 		blk_mq_release(q);
269 
270 	ida_free(&blk_queue_ida, q->id);
271 	call_rcu(&q->rcu_head, blk_free_queue_rcu);
272 }
273 
274 /**
275  * blk_put_queue - decrement the request_queue refcount
276  * @q: the request_queue structure to decrement the refcount for
277  *
278  * Decrements the refcount of the request_queue and free it when the refcount
279  * reaches 0.
280  */
281 void blk_put_queue(struct request_queue *q)
282 {
283 	if (refcount_dec_and_test(&q->refs))
284 		blk_free_queue(q);
285 }
286 EXPORT_SYMBOL(blk_put_queue);
287 
288 void blk_queue_start_drain(struct request_queue *q)
289 {
290 	/*
291 	 * When queue DYING flag is set, we need to block new req
292 	 * entering queue, so we call blk_freeze_queue_start() to
293 	 * prevent I/O from crossing blk_queue_enter().
294 	 */
295 	blk_freeze_queue_start(q);
296 	if (queue_is_mq(q))
297 		blk_mq_wake_waiters(q);
298 	/* Make blk_queue_enter() reexamine the DYING flag. */
299 	wake_up_all(&q->mq_freeze_wq);
300 }
301 
302 /**
303  * blk_queue_enter() - try to increase q->q_usage_counter
304  * @q: request queue pointer
305  * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
306  */
307 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
308 {
309 	const bool pm = flags & BLK_MQ_REQ_PM;
310 
311 	while (!blk_try_enter_queue(q, pm)) {
312 		if (flags & BLK_MQ_REQ_NOWAIT)
313 			return -EAGAIN;
314 
315 		/*
316 		 * read pair of barrier in blk_freeze_queue_start(), we need to
317 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
318 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
319 		 * following wait may never return if the two reads are
320 		 * reordered.
321 		 */
322 		smp_rmb();
323 		wait_event(q->mq_freeze_wq,
324 			   (!q->mq_freeze_depth &&
325 			    blk_pm_resume_queue(pm, q)) ||
326 			   blk_queue_dying(q));
327 		if (blk_queue_dying(q))
328 			return -ENODEV;
329 	}
330 
331 	return 0;
332 }
333 
334 int __bio_queue_enter(struct request_queue *q, struct bio *bio)
335 {
336 	while (!blk_try_enter_queue(q, false)) {
337 		struct gendisk *disk = bio->bi_bdev->bd_disk;
338 
339 		if (bio->bi_opf & REQ_NOWAIT) {
340 			if (test_bit(GD_DEAD, &disk->state))
341 				goto dead;
342 			bio_wouldblock_error(bio);
343 			return -EAGAIN;
344 		}
345 
346 		/*
347 		 * read pair of barrier in blk_freeze_queue_start(), we need to
348 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
349 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
350 		 * following wait may never return if the two reads are
351 		 * reordered.
352 		 */
353 		smp_rmb();
354 		wait_event(q->mq_freeze_wq,
355 			   (!q->mq_freeze_depth &&
356 			    blk_pm_resume_queue(false, q)) ||
357 			   test_bit(GD_DEAD, &disk->state));
358 		if (test_bit(GD_DEAD, &disk->state))
359 			goto dead;
360 	}
361 
362 	return 0;
363 dead:
364 	bio_io_error(bio);
365 	return -ENODEV;
366 }
367 
368 void blk_queue_exit(struct request_queue *q)
369 {
370 	percpu_ref_put(&q->q_usage_counter);
371 }
372 
373 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
374 {
375 	struct request_queue *q =
376 		container_of(ref, struct request_queue, q_usage_counter);
377 
378 	wake_up_all(&q->mq_freeze_wq);
379 }
380 
381 static void blk_rq_timed_out_timer(struct timer_list *t)
382 {
383 	struct request_queue *q = from_timer(q, t, timeout);
384 
385 	kblockd_schedule_work(&q->timeout_work);
386 }
387 
388 static void blk_timeout_work(struct work_struct *work)
389 {
390 }
391 
392 struct request_queue *blk_alloc_queue(int node_id)
393 {
394 	struct request_queue *q;
395 
396 	q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO,
397 				  node_id);
398 	if (!q)
399 		return NULL;
400 
401 	q->last_merge = NULL;
402 
403 	q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL);
404 	if (q->id < 0)
405 		goto fail_q;
406 
407 	q->stats = blk_alloc_queue_stats();
408 	if (!q->stats)
409 		goto fail_id;
410 
411 	q->node = node_id;
412 
413 	atomic_set(&q->nr_active_requests_shared_tags, 0);
414 
415 	timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
416 	INIT_WORK(&q->timeout_work, blk_timeout_work);
417 	INIT_LIST_HEAD(&q->icq_list);
418 
419 	refcount_set(&q->refs, 1);
420 	mutex_init(&q->debugfs_mutex);
421 	mutex_init(&q->sysfs_lock);
422 	mutex_init(&q->sysfs_dir_lock);
423 	mutex_init(&q->rq_qos_mutex);
424 	spin_lock_init(&q->queue_lock);
425 
426 	init_waitqueue_head(&q->mq_freeze_wq);
427 	mutex_init(&q->mq_freeze_lock);
428 
429 	/*
430 	 * Init percpu_ref in atomic mode so that it's faster to shutdown.
431 	 * See blk_register_queue() for details.
432 	 */
433 	if (percpu_ref_init(&q->q_usage_counter,
434 				blk_queue_usage_counter_release,
435 				PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
436 		goto fail_stats;
437 
438 	blk_set_default_limits(&q->limits);
439 	q->nr_requests = BLKDEV_DEFAULT_RQ;
440 
441 	return q;
442 
443 fail_stats:
444 	blk_free_queue_stats(q->stats);
445 fail_id:
446 	ida_free(&blk_queue_ida, q->id);
447 fail_q:
448 	kmem_cache_free(blk_requestq_cachep, q);
449 	return NULL;
450 }
451 
452 /**
453  * blk_get_queue - increment the request_queue refcount
454  * @q: the request_queue structure to increment the refcount for
455  *
456  * Increment the refcount of the request_queue kobject.
457  *
458  * Context: Any context.
459  */
460 bool blk_get_queue(struct request_queue *q)
461 {
462 	if (unlikely(blk_queue_dying(q)))
463 		return false;
464 	refcount_inc(&q->refs);
465 	return true;
466 }
467 EXPORT_SYMBOL(blk_get_queue);
468 
469 #ifdef CONFIG_FAIL_MAKE_REQUEST
470 
471 static DECLARE_FAULT_ATTR(fail_make_request);
472 
473 static int __init setup_fail_make_request(char *str)
474 {
475 	return setup_fault_attr(&fail_make_request, str);
476 }
477 __setup("fail_make_request=", setup_fail_make_request);
478 
479 bool should_fail_request(struct block_device *part, unsigned int bytes)
480 {
481 	return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
482 }
483 
484 static int __init fail_make_request_debugfs(void)
485 {
486 	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
487 						NULL, &fail_make_request);
488 
489 	return PTR_ERR_OR_ZERO(dir);
490 }
491 
492 late_initcall(fail_make_request_debugfs);
493 #endif /* CONFIG_FAIL_MAKE_REQUEST */
494 
495 static inline void bio_check_ro(struct bio *bio)
496 {
497 	if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
498 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
499 			return;
500 		pr_warn("Trying to write to read-only block-device %pg\n",
501 			bio->bi_bdev);
502 		/* Older lvm-tools actually trigger this */
503 	}
504 }
505 
506 static noinline int should_fail_bio(struct bio *bio)
507 {
508 	if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
509 		return -EIO;
510 	return 0;
511 }
512 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
513 
514 /*
515  * Check whether this bio extends beyond the end of the device or partition.
516  * This may well happen - the kernel calls bread() without checking the size of
517  * the device, e.g., when mounting a file system.
518  */
519 static inline int bio_check_eod(struct bio *bio)
520 {
521 	sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
522 	unsigned int nr_sectors = bio_sectors(bio);
523 
524 	if (nr_sectors &&
525 	    (nr_sectors > maxsector ||
526 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
527 		pr_info_ratelimited("%s: attempt to access beyond end of device\n"
528 				    "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
529 				    current->comm, bio->bi_bdev, bio->bi_opf,
530 				    bio->bi_iter.bi_sector, nr_sectors, maxsector);
531 		return -EIO;
532 	}
533 	return 0;
534 }
535 
536 /*
537  * Remap block n of partition p to block n+start(p) of the disk.
538  */
539 static int blk_partition_remap(struct bio *bio)
540 {
541 	struct block_device *p = bio->bi_bdev;
542 
543 	if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
544 		return -EIO;
545 	if (bio_sectors(bio)) {
546 		bio->bi_iter.bi_sector += p->bd_start_sect;
547 		trace_block_bio_remap(bio, p->bd_dev,
548 				      bio->bi_iter.bi_sector -
549 				      p->bd_start_sect);
550 	}
551 	bio_set_flag(bio, BIO_REMAPPED);
552 	return 0;
553 }
554 
555 /*
556  * Check write append to a zoned block device.
557  */
558 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
559 						 struct bio *bio)
560 {
561 	int nr_sectors = bio_sectors(bio);
562 
563 	/* Only applicable to zoned block devices */
564 	if (!bdev_is_zoned(bio->bi_bdev))
565 		return BLK_STS_NOTSUPP;
566 
567 	/* The bio sector must point to the start of a sequential zone */
568 	if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector) ||
569 	    !bio_zone_is_seq(bio))
570 		return BLK_STS_IOERR;
571 
572 	/*
573 	 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
574 	 * split and could result in non-contiguous sectors being written in
575 	 * different zones.
576 	 */
577 	if (nr_sectors > q->limits.chunk_sectors)
578 		return BLK_STS_IOERR;
579 
580 	/* Make sure the BIO is small enough and will not get split */
581 	if (nr_sectors > q->limits.max_zone_append_sectors)
582 		return BLK_STS_IOERR;
583 
584 	bio->bi_opf |= REQ_NOMERGE;
585 
586 	return BLK_STS_OK;
587 }
588 
589 static void __submit_bio(struct bio *bio)
590 {
591 	if (unlikely(!blk_crypto_bio_prep(&bio)))
592 		return;
593 
594 	if (!bio->bi_bdev->bd_has_submit_bio) {
595 		blk_mq_submit_bio(bio);
596 	} else if (likely(bio_queue_enter(bio) == 0)) {
597 		struct gendisk *disk = bio->bi_bdev->bd_disk;
598 
599 		disk->fops->submit_bio(bio);
600 		blk_queue_exit(disk->queue);
601 	}
602 }
603 
604 /*
605  * The loop in this function may be a bit non-obvious, and so deserves some
606  * explanation:
607  *
608  *  - Before entering the loop, bio->bi_next is NULL (as all callers ensure
609  *    that), so we have a list with a single bio.
610  *  - We pretend that we have just taken it off a longer list, so we assign
611  *    bio_list to a pointer to the bio_list_on_stack, thus initialising the
612  *    bio_list of new bios to be added.  ->submit_bio() may indeed add some more
613  *    bios through a recursive call to submit_bio_noacct.  If it did, we find a
614  *    non-NULL value in bio_list and re-enter the loop from the top.
615  *  - In this case we really did just take the bio of the top of the list (no
616  *    pretending) and so remove it from bio_list, and call into ->submit_bio()
617  *    again.
618  *
619  * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
620  * bio_list_on_stack[1] contains bios that were submitted before the current
621  *	->submit_bio, but that haven't been processed yet.
622  */
623 static void __submit_bio_noacct(struct bio *bio)
624 {
625 	struct bio_list bio_list_on_stack[2];
626 
627 	BUG_ON(bio->bi_next);
628 
629 	bio_list_init(&bio_list_on_stack[0]);
630 	current->bio_list = bio_list_on_stack;
631 
632 	do {
633 		struct request_queue *q = bdev_get_queue(bio->bi_bdev);
634 		struct bio_list lower, same;
635 
636 		/*
637 		 * Create a fresh bio_list for all subordinate requests.
638 		 */
639 		bio_list_on_stack[1] = bio_list_on_stack[0];
640 		bio_list_init(&bio_list_on_stack[0]);
641 
642 		__submit_bio(bio);
643 
644 		/*
645 		 * Sort new bios into those for a lower level and those for the
646 		 * same level.
647 		 */
648 		bio_list_init(&lower);
649 		bio_list_init(&same);
650 		while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
651 			if (q == bdev_get_queue(bio->bi_bdev))
652 				bio_list_add(&same, bio);
653 			else
654 				bio_list_add(&lower, bio);
655 
656 		/*
657 		 * Now assemble so we handle the lowest level first.
658 		 */
659 		bio_list_merge(&bio_list_on_stack[0], &lower);
660 		bio_list_merge(&bio_list_on_stack[0], &same);
661 		bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
662 	} while ((bio = bio_list_pop(&bio_list_on_stack[0])));
663 
664 	current->bio_list = NULL;
665 }
666 
667 static void __submit_bio_noacct_mq(struct bio *bio)
668 {
669 	struct bio_list bio_list[2] = { };
670 
671 	current->bio_list = bio_list;
672 
673 	do {
674 		__submit_bio(bio);
675 	} while ((bio = bio_list_pop(&bio_list[0])));
676 
677 	current->bio_list = NULL;
678 }
679 
680 void submit_bio_noacct_nocheck(struct bio *bio)
681 {
682 	blk_cgroup_bio_start(bio);
683 	blkcg_bio_issue_init(bio);
684 
685 	if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
686 		trace_block_bio_queue(bio);
687 		/*
688 		 * Now that enqueuing has been traced, we need to trace
689 		 * completion as well.
690 		 */
691 		bio_set_flag(bio, BIO_TRACE_COMPLETION);
692 	}
693 
694 	/*
695 	 * We only want one ->submit_bio to be active at a time, else stack
696 	 * usage with stacked devices could be a problem.  Use current->bio_list
697 	 * to collect a list of requests submited by a ->submit_bio method while
698 	 * it is active, and then process them after it returned.
699 	 */
700 	if (current->bio_list)
701 		bio_list_add(&current->bio_list[0], bio);
702 	else if (!bio->bi_bdev->bd_has_submit_bio)
703 		__submit_bio_noacct_mq(bio);
704 	else
705 		__submit_bio_noacct(bio);
706 }
707 
708 /**
709  * submit_bio_noacct - re-submit a bio to the block device layer for I/O
710  * @bio:  The bio describing the location in memory and on the device.
711  *
712  * This is a version of submit_bio() that shall only be used for I/O that is
713  * resubmitted to lower level drivers by stacking block drivers.  All file
714  * systems and other upper level users of the block layer should use
715  * submit_bio() instead.
716  */
717 void submit_bio_noacct(struct bio *bio)
718 {
719 	struct block_device *bdev = bio->bi_bdev;
720 	struct request_queue *q = bdev_get_queue(bdev);
721 	blk_status_t status = BLK_STS_IOERR;
722 	struct blk_plug *plug;
723 
724 	might_sleep();
725 
726 	plug = blk_mq_plug(bio);
727 	if (plug && plug->nowait)
728 		bio->bi_opf |= REQ_NOWAIT;
729 
730 	/*
731 	 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
732 	 * if queue does not support NOWAIT.
733 	 */
734 	if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev))
735 		goto not_supported;
736 
737 	if (should_fail_bio(bio))
738 		goto end_io;
739 	bio_check_ro(bio);
740 	if (!bio_flagged(bio, BIO_REMAPPED)) {
741 		if (unlikely(bio_check_eod(bio)))
742 			goto end_io;
743 		if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
744 			goto end_io;
745 	}
746 
747 	/*
748 	 * Filter flush bio's early so that bio based drivers without flush
749 	 * support don't have to worry about them.
750 	 */
751 	if (op_is_flush(bio->bi_opf)) {
752 		if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE &&
753 				 bio_op(bio) != REQ_OP_ZONE_APPEND))
754 			goto end_io;
755 		if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
756 			bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
757 			if (!bio_sectors(bio)) {
758 				status = BLK_STS_OK;
759 				goto end_io;
760 			}
761 		}
762 	}
763 
764 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
765 		bio_clear_polled(bio);
766 
767 	switch (bio_op(bio)) {
768 	case REQ_OP_DISCARD:
769 		if (!bdev_max_discard_sectors(bdev))
770 			goto not_supported;
771 		break;
772 	case REQ_OP_SECURE_ERASE:
773 		if (!bdev_max_secure_erase_sectors(bdev))
774 			goto not_supported;
775 		break;
776 	case REQ_OP_ZONE_APPEND:
777 		status = blk_check_zone_append(q, bio);
778 		if (status != BLK_STS_OK)
779 			goto end_io;
780 		break;
781 	case REQ_OP_ZONE_RESET:
782 	case REQ_OP_ZONE_OPEN:
783 	case REQ_OP_ZONE_CLOSE:
784 	case REQ_OP_ZONE_FINISH:
785 		if (!bdev_is_zoned(bio->bi_bdev))
786 			goto not_supported;
787 		break;
788 	case REQ_OP_ZONE_RESET_ALL:
789 		if (!bdev_is_zoned(bio->bi_bdev) || !blk_queue_zone_resetall(q))
790 			goto not_supported;
791 		break;
792 	case REQ_OP_WRITE_ZEROES:
793 		if (!q->limits.max_write_zeroes_sectors)
794 			goto not_supported;
795 		break;
796 	default:
797 		break;
798 	}
799 
800 	if (blk_throtl_bio(bio))
801 		return;
802 	submit_bio_noacct_nocheck(bio);
803 	return;
804 
805 not_supported:
806 	status = BLK_STS_NOTSUPP;
807 end_io:
808 	bio->bi_status = status;
809 	bio_endio(bio);
810 }
811 EXPORT_SYMBOL(submit_bio_noacct);
812 
813 /**
814  * submit_bio - submit a bio to the block device layer for I/O
815  * @bio: The &struct bio which describes the I/O
816  *
817  * submit_bio() is used to submit I/O requests to block devices.  It is passed a
818  * fully set up &struct bio that describes the I/O that needs to be done.  The
819  * bio will be send to the device described by the bi_bdev field.
820  *
821  * The success/failure status of the request, along with notification of
822  * completion, is delivered asynchronously through the ->bi_end_io() callback
823  * in @bio.  The bio must NOT be touched by the caller until ->bi_end_io() has
824  * been called.
825  */
826 void submit_bio(struct bio *bio)
827 {
828 	if (bio_op(bio) == REQ_OP_READ) {
829 		task_io_account_read(bio->bi_iter.bi_size);
830 		count_vm_events(PGPGIN, bio_sectors(bio));
831 	} else if (bio_op(bio) == REQ_OP_WRITE) {
832 		count_vm_events(PGPGOUT, bio_sectors(bio));
833 	}
834 
835 	submit_bio_noacct(bio);
836 }
837 EXPORT_SYMBOL(submit_bio);
838 
839 /**
840  * bio_poll - poll for BIO completions
841  * @bio: bio to poll for
842  * @iob: batches of IO
843  * @flags: BLK_POLL_* flags that control the behavior
844  *
845  * Poll for completions on queue associated with the bio. Returns number of
846  * completed entries found.
847  *
848  * Note: the caller must either be the context that submitted @bio, or
849  * be in a RCU critical section to prevent freeing of @bio.
850  */
851 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
852 {
853 	blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
854 	struct block_device *bdev;
855 	struct request_queue *q;
856 	int ret = 0;
857 
858 	bdev = READ_ONCE(bio->bi_bdev);
859 	if (!bdev)
860 		return 0;
861 
862 	q = bdev_get_queue(bdev);
863 	if (cookie == BLK_QC_T_NONE ||
864 	    !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
865 		return 0;
866 
867 	/*
868 	 * As the requests that require a zone lock are not plugged in the
869 	 * first place, directly accessing the plug instead of using
870 	 * blk_mq_plug() should not have any consequences during flushing for
871 	 * zoned devices.
872 	 */
873 	blk_flush_plug(current->plug, false);
874 
875 	/*
876 	 * We need to be able to enter a frozen queue, similar to how
877 	 * timeouts also need to do that. If that is blocked, then we can
878 	 * have pending IO when a queue freeze is started, and then the
879 	 * wait for the freeze to finish will wait for polled requests to
880 	 * timeout as the poller is preventer from entering the queue and
881 	 * completing them. As long as we prevent new IO from being queued,
882 	 * that should be all that matters.
883 	 */
884 	if (!percpu_ref_tryget(&q->q_usage_counter))
885 		return 0;
886 	if (queue_is_mq(q)) {
887 		ret = blk_mq_poll(q, cookie, iob, flags);
888 	} else {
889 		struct gendisk *disk = q->disk;
890 
891 		if (disk && disk->fops->poll_bio)
892 			ret = disk->fops->poll_bio(bio, iob, flags);
893 	}
894 	blk_queue_exit(q);
895 	return ret;
896 }
897 EXPORT_SYMBOL_GPL(bio_poll);
898 
899 /*
900  * Helper to implement file_operations.iopoll.  Requires the bio to be stored
901  * in iocb->private, and cleared before freeing the bio.
902  */
903 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
904 		    unsigned int flags)
905 {
906 	struct bio *bio;
907 	int ret = 0;
908 
909 	/*
910 	 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
911 	 * point to a freshly allocated bio at this point.  If that happens
912 	 * we have a few cases to consider:
913 	 *
914 	 *  1) the bio is beeing initialized and bi_bdev is NULL.  We can just
915 	 *     simply nothing in this case
916 	 *  2) the bio points to a not poll enabled device.  bio_poll will catch
917 	 *     this and return 0
918 	 *  3) the bio points to a poll capable device, including but not
919 	 *     limited to the one that the original bio pointed to.  In this
920 	 *     case we will call into the actual poll method and poll for I/O,
921 	 *     even if we don't need to, but it won't cause harm either.
922 	 *
923 	 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
924 	 * is still allocated. Because partitions hold a reference to the whole
925 	 * device bdev and thus disk, the disk is also still valid.  Grabbing
926 	 * a reference to the queue in bio_poll() ensures the hctxs and requests
927 	 * are still valid as well.
928 	 */
929 	rcu_read_lock();
930 	bio = READ_ONCE(kiocb->private);
931 	if (bio)
932 		ret = bio_poll(bio, iob, flags);
933 	rcu_read_unlock();
934 
935 	return ret;
936 }
937 EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
938 
939 void update_io_ticks(struct block_device *part, unsigned long now, bool end)
940 {
941 	unsigned long stamp;
942 again:
943 	stamp = READ_ONCE(part->bd_stamp);
944 	if (unlikely(time_after(now, stamp))) {
945 		if (likely(try_cmpxchg(&part->bd_stamp, &stamp, now)))
946 			__part_stat_add(part, io_ticks, end ? now - stamp : 1);
947 	}
948 	if (part->bd_partno) {
949 		part = bdev_whole(part);
950 		goto again;
951 	}
952 }
953 
954 unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op,
955 				 unsigned long start_time)
956 {
957 	part_stat_lock();
958 	update_io_ticks(bdev, start_time, false);
959 	part_stat_local_inc(bdev, in_flight[op_is_write(op)]);
960 	part_stat_unlock();
961 
962 	return start_time;
963 }
964 EXPORT_SYMBOL(bdev_start_io_acct);
965 
966 /**
967  * bio_start_io_acct - start I/O accounting for bio based drivers
968  * @bio:	bio to start account for
969  *
970  * Returns the start time that should be passed back to bio_end_io_acct().
971  */
972 unsigned long bio_start_io_acct(struct bio *bio)
973 {
974 	return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies);
975 }
976 EXPORT_SYMBOL_GPL(bio_start_io_acct);
977 
978 void bdev_end_io_acct(struct block_device *bdev, enum req_op op,
979 		      unsigned int sectors, unsigned long start_time)
980 {
981 	const int sgrp = op_stat_group(op);
982 	unsigned long now = READ_ONCE(jiffies);
983 	unsigned long duration = now - start_time;
984 
985 	part_stat_lock();
986 	update_io_ticks(bdev, now, true);
987 	part_stat_inc(bdev, ios[sgrp]);
988 	part_stat_add(bdev, sectors[sgrp], sectors);
989 	part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration));
990 	part_stat_local_dec(bdev, in_flight[op_is_write(op)]);
991 	part_stat_unlock();
992 }
993 EXPORT_SYMBOL(bdev_end_io_acct);
994 
995 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
996 			      struct block_device *orig_bdev)
997 {
998 	bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time);
999 }
1000 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1001 
1002 /**
1003  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1004  * @q : the queue of the device being checked
1005  *
1006  * Description:
1007  *    Check if underlying low-level drivers of a device are busy.
1008  *    If the drivers want to export their busy state, they must set own
1009  *    exporting function using blk_queue_lld_busy() first.
1010  *
1011  *    Basically, this function is used only by request stacking drivers
1012  *    to stop dispatching requests to underlying devices when underlying
1013  *    devices are busy.  This behavior helps more I/O merging on the queue
1014  *    of the request stacking driver and prevents I/O throughput regression
1015  *    on burst I/O load.
1016  *
1017  * Return:
1018  *    0 - Not busy (The request stacking driver should dispatch request)
1019  *    1 - Busy (The request stacking driver should stop dispatching request)
1020  */
1021 int blk_lld_busy(struct request_queue *q)
1022 {
1023 	if (queue_is_mq(q) && q->mq_ops->busy)
1024 		return q->mq_ops->busy(q);
1025 
1026 	return 0;
1027 }
1028 EXPORT_SYMBOL_GPL(blk_lld_busy);
1029 
1030 int kblockd_schedule_work(struct work_struct *work)
1031 {
1032 	return queue_work(kblockd_workqueue, work);
1033 }
1034 EXPORT_SYMBOL(kblockd_schedule_work);
1035 
1036 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1037 				unsigned long delay)
1038 {
1039 	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1040 }
1041 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1042 
1043 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1044 {
1045 	struct task_struct *tsk = current;
1046 
1047 	/*
1048 	 * If this is a nested plug, don't actually assign it.
1049 	 */
1050 	if (tsk->plug)
1051 		return;
1052 
1053 	plug->mq_list = NULL;
1054 	plug->cached_rq = NULL;
1055 	plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1056 	plug->rq_count = 0;
1057 	plug->multiple_queues = false;
1058 	plug->has_elevator = false;
1059 	plug->nowait = false;
1060 	INIT_LIST_HEAD(&plug->cb_list);
1061 
1062 	/*
1063 	 * Store ordering should not be needed here, since a potential
1064 	 * preempt will imply a full memory barrier
1065 	 */
1066 	tsk->plug = plug;
1067 }
1068 
1069 /**
1070  * blk_start_plug - initialize blk_plug and track it inside the task_struct
1071  * @plug:	The &struct blk_plug that needs to be initialized
1072  *
1073  * Description:
1074  *   blk_start_plug() indicates to the block layer an intent by the caller
1075  *   to submit multiple I/O requests in a batch.  The block layer may use
1076  *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1077  *   is called.  However, the block layer may choose to submit requests
1078  *   before a call to blk_finish_plug() if the number of queued I/Os
1079  *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1080  *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1081  *   the task schedules (see below).
1082  *
1083  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1084  *   pending I/O should the task end up blocking between blk_start_plug() and
1085  *   blk_finish_plug(). This is important from a performance perspective, but
1086  *   also ensures that we don't deadlock. For instance, if the task is blocking
1087  *   for a memory allocation, memory reclaim could end up wanting to free a
1088  *   page belonging to that request that is currently residing in our private
1089  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1090  *   this kind of deadlock.
1091  */
1092 void blk_start_plug(struct blk_plug *plug)
1093 {
1094 	blk_start_plug_nr_ios(plug, 1);
1095 }
1096 EXPORT_SYMBOL(blk_start_plug);
1097 
1098 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1099 {
1100 	LIST_HEAD(callbacks);
1101 
1102 	while (!list_empty(&plug->cb_list)) {
1103 		list_splice_init(&plug->cb_list, &callbacks);
1104 
1105 		while (!list_empty(&callbacks)) {
1106 			struct blk_plug_cb *cb = list_first_entry(&callbacks,
1107 							  struct blk_plug_cb,
1108 							  list);
1109 			list_del(&cb->list);
1110 			cb->callback(cb, from_schedule);
1111 		}
1112 	}
1113 }
1114 
1115 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1116 				      int size)
1117 {
1118 	struct blk_plug *plug = current->plug;
1119 	struct blk_plug_cb *cb;
1120 
1121 	if (!plug)
1122 		return NULL;
1123 
1124 	list_for_each_entry(cb, &plug->cb_list, list)
1125 		if (cb->callback == unplug && cb->data == data)
1126 			return cb;
1127 
1128 	/* Not currently on the callback list */
1129 	BUG_ON(size < sizeof(*cb));
1130 	cb = kzalloc(size, GFP_ATOMIC);
1131 	if (cb) {
1132 		cb->data = data;
1133 		cb->callback = unplug;
1134 		list_add(&cb->list, &plug->cb_list);
1135 	}
1136 	return cb;
1137 }
1138 EXPORT_SYMBOL(blk_check_plugged);
1139 
1140 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1141 {
1142 	if (!list_empty(&plug->cb_list))
1143 		flush_plug_callbacks(plug, from_schedule);
1144 	if (!rq_list_empty(plug->mq_list))
1145 		blk_mq_flush_plug_list(plug, from_schedule);
1146 	/*
1147 	 * Unconditionally flush out cached requests, even if the unplug
1148 	 * event came from schedule. Since we know hold references to the
1149 	 * queue for cached requests, we don't want a blocked task holding
1150 	 * up a queue freeze/quiesce event.
1151 	 */
1152 	if (unlikely(!rq_list_empty(plug->cached_rq)))
1153 		blk_mq_free_plug_rqs(plug);
1154 }
1155 
1156 /**
1157  * blk_finish_plug - mark the end of a batch of submitted I/O
1158  * @plug:	The &struct blk_plug passed to blk_start_plug()
1159  *
1160  * Description:
1161  * Indicate that a batch of I/O submissions is complete.  This function
1162  * must be paired with an initial call to blk_start_plug().  The intent
1163  * is to allow the block layer to optimize I/O submission.  See the
1164  * documentation for blk_start_plug() for more information.
1165  */
1166 void blk_finish_plug(struct blk_plug *plug)
1167 {
1168 	if (plug == current->plug) {
1169 		__blk_flush_plug(plug, false);
1170 		current->plug = NULL;
1171 	}
1172 }
1173 EXPORT_SYMBOL(blk_finish_plug);
1174 
1175 void blk_io_schedule(void)
1176 {
1177 	/* Prevent hang_check timer from firing at us during very long I/O */
1178 	unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1179 
1180 	if (timeout)
1181 		io_schedule_timeout(timeout);
1182 	else
1183 		io_schedule();
1184 }
1185 EXPORT_SYMBOL_GPL(blk_io_schedule);
1186 
1187 int __init blk_dev_init(void)
1188 {
1189 	BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS));
1190 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1191 			sizeof_field(struct request, cmd_flags));
1192 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1193 			sizeof_field(struct bio, bi_opf));
1194 
1195 	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
1196 	kblockd_workqueue = alloc_workqueue("kblockd",
1197 					    WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1198 	if (!kblockd_workqueue)
1199 		panic("Failed to create kblockd\n");
1200 
1201 	blk_requestq_cachep = kmem_cache_create("request_queue",
1202 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1203 
1204 	blk_debugfs_root = debugfs_create_dir("block", NULL);
1205 
1206 	return 0;
1207 }
1208