xref: /openbmc/linux/block/blk-core.c (revision ac3cb6de)
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 
212 /**
213  * blk_sync_queue - cancel any pending callbacks on a queue
214  * @q: the queue
215  *
216  * Description:
217  *     The block layer may perform asynchronous callback activity
218  *     on a queue, such as calling the unplug function after a timeout.
219  *     A block device may call blk_sync_queue to ensure that any
220  *     such activity is cancelled, thus allowing it to release resources
221  *     that the callbacks might use. The caller must already have made sure
222  *     that its ->submit_bio will not re-add plugging prior to calling
223  *     this function.
224  *
225  *     This function does not cancel any asynchronous activity arising
226  *     out of elevator or throttling code. That would require elevator_exit()
227  *     and blkcg_exit_queue() to be called with queue lock initialized.
228  *
229  */
230 void blk_sync_queue(struct request_queue *q)
231 {
232 	del_timer_sync(&q->timeout);
233 	cancel_work_sync(&q->timeout_work);
234 }
235 EXPORT_SYMBOL(blk_sync_queue);
236 
237 /**
238  * blk_set_pm_only - increment pm_only counter
239  * @q: request queue pointer
240  */
241 void blk_set_pm_only(struct request_queue *q)
242 {
243 	atomic_inc(&q->pm_only);
244 }
245 EXPORT_SYMBOL_GPL(blk_set_pm_only);
246 
247 void blk_clear_pm_only(struct request_queue *q)
248 {
249 	int pm_only;
250 
251 	pm_only = atomic_dec_return(&q->pm_only);
252 	WARN_ON_ONCE(pm_only < 0);
253 	if (pm_only == 0)
254 		wake_up_all(&q->mq_freeze_wq);
255 }
256 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
257 
258 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
259 {
260 	struct request_queue *q = container_of(rcu_head,
261 			struct request_queue, rcu_head);
262 
263 	percpu_ref_exit(&q->q_usage_counter);
264 	kmem_cache_free(blk_requestq_cachep, q);
265 }
266 
267 static void blk_free_queue(struct request_queue *q)
268 {
269 	blk_free_queue_stats(q->stats);
270 	if (queue_is_mq(q))
271 		blk_mq_release(q);
272 
273 	ida_free(&blk_queue_ida, q->id);
274 	call_rcu(&q->rcu_head, blk_free_queue_rcu);
275 }
276 
277 /**
278  * blk_put_queue - decrement the request_queue refcount
279  * @q: the request_queue structure to decrement the refcount for
280  *
281  * Decrements the refcount of the request_queue and free it when the refcount
282  * reaches 0.
283  */
284 void blk_put_queue(struct request_queue *q)
285 {
286 	if (refcount_dec_and_test(&q->refs))
287 		blk_free_queue(q);
288 }
289 EXPORT_SYMBOL(blk_put_queue);
290 
291 void blk_queue_start_drain(struct request_queue *q)
292 {
293 	/*
294 	 * When queue DYING flag is set, we need to block new req
295 	 * entering queue, so we call blk_freeze_queue_start() to
296 	 * prevent I/O from crossing blk_queue_enter().
297 	 */
298 	blk_freeze_queue_start(q);
299 	if (queue_is_mq(q))
300 		blk_mq_wake_waiters(q);
301 	/* Make blk_queue_enter() reexamine the DYING flag. */
302 	wake_up_all(&q->mq_freeze_wq);
303 }
304 
305 /**
306  * blk_queue_enter() - try to increase q->q_usage_counter
307  * @q: request queue pointer
308  * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
309  */
310 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
311 {
312 	const bool pm = flags & BLK_MQ_REQ_PM;
313 
314 	while (!blk_try_enter_queue(q, pm)) {
315 		if (flags & BLK_MQ_REQ_NOWAIT)
316 			return -EAGAIN;
317 
318 		/*
319 		 * read pair of barrier in blk_freeze_queue_start(), we need to
320 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
321 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
322 		 * following wait may never return if the two reads are
323 		 * reordered.
324 		 */
325 		smp_rmb();
326 		wait_event(q->mq_freeze_wq,
327 			   (!q->mq_freeze_depth &&
328 			    blk_pm_resume_queue(pm, q)) ||
329 			   blk_queue_dying(q));
330 		if (blk_queue_dying(q))
331 			return -ENODEV;
332 	}
333 
334 	return 0;
335 }
336 
337 int __bio_queue_enter(struct request_queue *q, struct bio *bio)
338 {
339 	while (!blk_try_enter_queue(q, false)) {
340 		struct gendisk *disk = bio->bi_bdev->bd_disk;
341 
342 		if (bio->bi_opf & REQ_NOWAIT) {
343 			if (test_bit(GD_DEAD, &disk->state))
344 				goto dead;
345 			bio_wouldblock_error(bio);
346 			return -EAGAIN;
347 		}
348 
349 		/*
350 		 * read pair of barrier in blk_freeze_queue_start(), we need to
351 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
352 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
353 		 * following wait may never return if the two reads are
354 		 * reordered.
355 		 */
356 		smp_rmb();
357 		wait_event(q->mq_freeze_wq,
358 			   (!q->mq_freeze_depth &&
359 			    blk_pm_resume_queue(false, q)) ||
360 			   test_bit(GD_DEAD, &disk->state));
361 		if (test_bit(GD_DEAD, &disk->state))
362 			goto dead;
363 	}
364 
365 	return 0;
366 dead:
367 	bio_io_error(bio);
368 	return -ENODEV;
369 }
370 
371 void blk_queue_exit(struct request_queue *q)
372 {
373 	percpu_ref_put(&q->q_usage_counter);
374 }
375 
376 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
377 {
378 	struct request_queue *q =
379 		container_of(ref, struct request_queue, q_usage_counter);
380 
381 	wake_up_all(&q->mq_freeze_wq);
382 }
383 
384 static void blk_rq_timed_out_timer(struct timer_list *t)
385 {
386 	struct request_queue *q = from_timer(q, t, timeout);
387 
388 	kblockd_schedule_work(&q->timeout_work);
389 }
390 
391 static void blk_timeout_work(struct work_struct *work)
392 {
393 }
394 
395 struct request_queue *blk_alloc_queue(int node_id)
396 {
397 	struct request_queue *q;
398 
399 	q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO,
400 				  node_id);
401 	if (!q)
402 		return NULL;
403 
404 	q->last_merge = NULL;
405 
406 	q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL);
407 	if (q->id < 0)
408 		goto fail_q;
409 
410 	q->stats = blk_alloc_queue_stats();
411 	if (!q->stats)
412 		goto fail_id;
413 
414 	q->node = node_id;
415 
416 	atomic_set(&q->nr_active_requests_shared_tags, 0);
417 
418 	timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
419 	INIT_WORK(&q->timeout_work, blk_timeout_work);
420 	INIT_LIST_HEAD(&q->icq_list);
421 
422 	refcount_set(&q->refs, 1);
423 	mutex_init(&q->debugfs_mutex);
424 	mutex_init(&q->sysfs_lock);
425 	mutex_init(&q->sysfs_dir_lock);
426 	mutex_init(&q->rq_qos_mutex);
427 	spin_lock_init(&q->queue_lock);
428 
429 	init_waitqueue_head(&q->mq_freeze_wq);
430 	mutex_init(&q->mq_freeze_lock);
431 
432 	/*
433 	 * Init percpu_ref in atomic mode so that it's faster to shutdown.
434 	 * See blk_register_queue() for details.
435 	 */
436 	if (percpu_ref_init(&q->q_usage_counter,
437 				blk_queue_usage_counter_release,
438 				PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
439 		goto fail_stats;
440 
441 	blk_set_default_limits(&q->limits);
442 	q->nr_requests = BLKDEV_DEFAULT_RQ;
443 
444 	return q;
445 
446 fail_stats:
447 	blk_free_queue_stats(q->stats);
448 fail_id:
449 	ida_free(&blk_queue_ida, q->id);
450 fail_q:
451 	kmem_cache_free(blk_requestq_cachep, q);
452 	return NULL;
453 }
454 
455 /**
456  * blk_get_queue - increment the request_queue refcount
457  * @q: the request_queue structure to increment the refcount for
458  *
459  * Increment the refcount of the request_queue kobject.
460  *
461  * Context: Any context.
462  */
463 bool blk_get_queue(struct request_queue *q)
464 {
465 	if (unlikely(blk_queue_dying(q)))
466 		return false;
467 	refcount_inc(&q->refs);
468 	return true;
469 }
470 EXPORT_SYMBOL(blk_get_queue);
471 
472 #ifdef CONFIG_FAIL_MAKE_REQUEST
473 
474 static DECLARE_FAULT_ATTR(fail_make_request);
475 
476 static int __init setup_fail_make_request(char *str)
477 {
478 	return setup_fault_attr(&fail_make_request, str);
479 }
480 __setup("fail_make_request=", setup_fail_make_request);
481 
482 bool should_fail_request(struct block_device *part, unsigned int bytes)
483 {
484 	return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
485 }
486 
487 static int __init fail_make_request_debugfs(void)
488 {
489 	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
490 						NULL, &fail_make_request);
491 
492 	return PTR_ERR_OR_ZERO(dir);
493 }
494 
495 late_initcall(fail_make_request_debugfs);
496 #endif /* CONFIG_FAIL_MAKE_REQUEST */
497 
498 static inline void bio_check_ro(struct bio *bio)
499 {
500 	if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
501 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
502 			return;
503 		pr_warn("Trying to write to read-only block-device %pg\n",
504 			bio->bi_bdev);
505 		/* Older lvm-tools actually trigger this */
506 	}
507 }
508 
509 static noinline int should_fail_bio(struct bio *bio)
510 {
511 	if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
512 		return -EIO;
513 	return 0;
514 }
515 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
516 
517 /*
518  * Check whether this bio extends beyond the end of the device or partition.
519  * This may well happen - the kernel calls bread() without checking the size of
520  * the device, e.g., when mounting a file system.
521  */
522 static inline int bio_check_eod(struct bio *bio)
523 {
524 	sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
525 	unsigned int nr_sectors = bio_sectors(bio);
526 
527 	if (nr_sectors &&
528 	    (nr_sectors > maxsector ||
529 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
530 		pr_info_ratelimited("%s: attempt to access beyond end of device\n"
531 				    "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
532 				    current->comm, bio->bi_bdev, bio->bi_opf,
533 				    bio->bi_iter.bi_sector, nr_sectors, maxsector);
534 		return -EIO;
535 	}
536 	return 0;
537 }
538 
539 /*
540  * Remap block n of partition p to block n+start(p) of the disk.
541  */
542 static int blk_partition_remap(struct bio *bio)
543 {
544 	struct block_device *p = bio->bi_bdev;
545 
546 	if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
547 		return -EIO;
548 	if (bio_sectors(bio)) {
549 		bio->bi_iter.bi_sector += p->bd_start_sect;
550 		trace_block_bio_remap(bio, p->bd_dev,
551 				      bio->bi_iter.bi_sector -
552 				      p->bd_start_sect);
553 	}
554 	bio_set_flag(bio, BIO_REMAPPED);
555 	return 0;
556 }
557 
558 /*
559  * Check write append to a zoned block device.
560  */
561 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
562 						 struct bio *bio)
563 {
564 	int nr_sectors = bio_sectors(bio);
565 
566 	/* Only applicable to zoned block devices */
567 	if (!bdev_is_zoned(bio->bi_bdev))
568 		return BLK_STS_NOTSUPP;
569 
570 	/* The bio sector must point to the start of a sequential zone */
571 	if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector) ||
572 	    !bio_zone_is_seq(bio))
573 		return BLK_STS_IOERR;
574 
575 	/*
576 	 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
577 	 * split and could result in non-contiguous sectors being written in
578 	 * different zones.
579 	 */
580 	if (nr_sectors > q->limits.chunk_sectors)
581 		return BLK_STS_IOERR;
582 
583 	/* Make sure the BIO is small enough and will not get split */
584 	if (nr_sectors > q->limits.max_zone_append_sectors)
585 		return BLK_STS_IOERR;
586 
587 	bio->bi_opf |= REQ_NOMERGE;
588 
589 	return BLK_STS_OK;
590 }
591 
592 static void __submit_bio(struct bio *bio)
593 {
594 	if (unlikely(!blk_crypto_bio_prep(&bio)))
595 		return;
596 
597 	if (!bio->bi_bdev->bd_has_submit_bio) {
598 		blk_mq_submit_bio(bio);
599 	} else if (likely(bio_queue_enter(bio) == 0)) {
600 		struct gendisk *disk = bio->bi_bdev->bd_disk;
601 
602 		disk->fops->submit_bio(bio);
603 		blk_queue_exit(disk->queue);
604 	}
605 }
606 
607 /*
608  * The loop in this function may be a bit non-obvious, and so deserves some
609  * explanation:
610  *
611  *  - Before entering the loop, bio->bi_next is NULL (as all callers ensure
612  *    that), so we have a list with a single bio.
613  *  - We pretend that we have just taken it off a longer list, so we assign
614  *    bio_list to a pointer to the bio_list_on_stack, thus initialising the
615  *    bio_list of new bios to be added.  ->submit_bio() may indeed add some more
616  *    bios through a recursive call to submit_bio_noacct.  If it did, we find a
617  *    non-NULL value in bio_list and re-enter the loop from the top.
618  *  - In this case we really did just take the bio of the top of the list (no
619  *    pretending) and so remove it from bio_list, and call into ->submit_bio()
620  *    again.
621  *
622  * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
623  * bio_list_on_stack[1] contains bios that were submitted before the current
624  *	->submit_bio, but that haven't been processed yet.
625  */
626 static void __submit_bio_noacct(struct bio *bio)
627 {
628 	struct bio_list bio_list_on_stack[2];
629 
630 	BUG_ON(bio->bi_next);
631 
632 	bio_list_init(&bio_list_on_stack[0]);
633 	current->bio_list = bio_list_on_stack;
634 
635 	do {
636 		struct request_queue *q = bdev_get_queue(bio->bi_bdev);
637 		struct bio_list lower, same;
638 
639 		/*
640 		 * Create a fresh bio_list for all subordinate requests.
641 		 */
642 		bio_list_on_stack[1] = bio_list_on_stack[0];
643 		bio_list_init(&bio_list_on_stack[0]);
644 
645 		__submit_bio(bio);
646 
647 		/*
648 		 * Sort new bios into those for a lower level and those for the
649 		 * same level.
650 		 */
651 		bio_list_init(&lower);
652 		bio_list_init(&same);
653 		while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
654 			if (q == bdev_get_queue(bio->bi_bdev))
655 				bio_list_add(&same, bio);
656 			else
657 				bio_list_add(&lower, bio);
658 
659 		/*
660 		 * Now assemble so we handle the lowest level first.
661 		 */
662 		bio_list_merge(&bio_list_on_stack[0], &lower);
663 		bio_list_merge(&bio_list_on_stack[0], &same);
664 		bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
665 	} while ((bio = bio_list_pop(&bio_list_on_stack[0])));
666 
667 	current->bio_list = NULL;
668 }
669 
670 static void __submit_bio_noacct_mq(struct bio *bio)
671 {
672 	struct bio_list bio_list[2] = { };
673 
674 	current->bio_list = bio_list;
675 
676 	do {
677 		__submit_bio(bio);
678 	} while ((bio = bio_list_pop(&bio_list[0])));
679 
680 	current->bio_list = NULL;
681 }
682 
683 void submit_bio_noacct_nocheck(struct bio *bio)
684 {
685 	blk_cgroup_bio_start(bio);
686 	blkcg_bio_issue_init(bio);
687 
688 	if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
689 		trace_block_bio_queue(bio);
690 		/*
691 		 * Now that enqueuing has been traced, we need to trace
692 		 * completion as well.
693 		 */
694 		bio_set_flag(bio, BIO_TRACE_COMPLETION);
695 	}
696 
697 	/*
698 	 * We only want one ->submit_bio to be active at a time, else stack
699 	 * usage with stacked devices could be a problem.  Use current->bio_list
700 	 * to collect a list of requests submited by a ->submit_bio method while
701 	 * it is active, and then process them after it returned.
702 	 */
703 	if (current->bio_list)
704 		bio_list_add(&current->bio_list[0], bio);
705 	else if (!bio->bi_bdev->bd_has_submit_bio)
706 		__submit_bio_noacct_mq(bio);
707 	else
708 		__submit_bio_noacct(bio);
709 }
710 
711 /**
712  * submit_bio_noacct - re-submit a bio to the block device layer for I/O
713  * @bio:  The bio describing the location in memory and on the device.
714  *
715  * This is a version of submit_bio() that shall only be used for I/O that is
716  * resubmitted to lower level drivers by stacking block drivers.  All file
717  * systems and other upper level users of the block layer should use
718  * submit_bio() instead.
719  */
720 void submit_bio_noacct(struct bio *bio)
721 {
722 	struct block_device *bdev = bio->bi_bdev;
723 	struct request_queue *q = bdev_get_queue(bdev);
724 	blk_status_t status = BLK_STS_IOERR;
725 	struct blk_plug *plug;
726 
727 	might_sleep();
728 
729 	plug = blk_mq_plug(bio);
730 	if (plug && plug->nowait)
731 		bio->bi_opf |= REQ_NOWAIT;
732 
733 	/*
734 	 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
735 	 * if queue does not support NOWAIT.
736 	 */
737 	if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev))
738 		goto not_supported;
739 
740 	if (should_fail_bio(bio))
741 		goto end_io;
742 	bio_check_ro(bio);
743 	if (!bio_flagged(bio, BIO_REMAPPED)) {
744 		if (unlikely(bio_check_eod(bio)))
745 			goto end_io;
746 		if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
747 			goto end_io;
748 	}
749 
750 	/*
751 	 * Filter flush bio's early so that bio based drivers without flush
752 	 * support don't have to worry about them.
753 	 */
754 	if (op_is_flush(bio->bi_opf)) {
755 		if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE &&
756 				 bio_op(bio) != REQ_OP_ZONE_APPEND))
757 			goto end_io;
758 		if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
759 			bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
760 			if (!bio_sectors(bio)) {
761 				status = BLK_STS_OK;
762 				goto end_io;
763 			}
764 		}
765 	}
766 
767 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
768 		bio_clear_polled(bio);
769 
770 	switch (bio_op(bio)) {
771 	case REQ_OP_DISCARD:
772 		if (!bdev_max_discard_sectors(bdev))
773 			goto not_supported;
774 		break;
775 	case REQ_OP_SECURE_ERASE:
776 		if (!bdev_max_secure_erase_sectors(bdev))
777 			goto not_supported;
778 		break;
779 	case REQ_OP_ZONE_APPEND:
780 		status = blk_check_zone_append(q, bio);
781 		if (status != BLK_STS_OK)
782 			goto end_io;
783 		break;
784 	case REQ_OP_ZONE_RESET:
785 	case REQ_OP_ZONE_OPEN:
786 	case REQ_OP_ZONE_CLOSE:
787 	case REQ_OP_ZONE_FINISH:
788 		if (!bdev_is_zoned(bio->bi_bdev))
789 			goto not_supported;
790 		break;
791 	case REQ_OP_ZONE_RESET_ALL:
792 		if (!bdev_is_zoned(bio->bi_bdev) || !blk_queue_zone_resetall(q))
793 			goto not_supported;
794 		break;
795 	case REQ_OP_WRITE_ZEROES:
796 		if (!q->limits.max_write_zeroes_sectors)
797 			goto not_supported;
798 		break;
799 	default:
800 		break;
801 	}
802 
803 	if (blk_throtl_bio(bio))
804 		return;
805 	submit_bio_noacct_nocheck(bio);
806 	return;
807 
808 not_supported:
809 	status = BLK_STS_NOTSUPP;
810 end_io:
811 	bio->bi_status = status;
812 	bio_endio(bio);
813 }
814 EXPORT_SYMBOL(submit_bio_noacct);
815 
816 /**
817  * submit_bio - submit a bio to the block device layer for I/O
818  * @bio: The &struct bio which describes the I/O
819  *
820  * submit_bio() is used to submit I/O requests to block devices.  It is passed a
821  * fully set up &struct bio that describes the I/O that needs to be done.  The
822  * bio will be send to the device described by the bi_bdev field.
823  *
824  * The success/failure status of the request, along with notification of
825  * completion, is delivered asynchronously through the ->bi_end_io() callback
826  * in @bio.  The bio must NOT be touched by the caller until ->bi_end_io() has
827  * been called.
828  */
829 void submit_bio(struct bio *bio)
830 {
831 	if (bio_op(bio) == REQ_OP_READ) {
832 		task_io_account_read(bio->bi_iter.bi_size);
833 		count_vm_events(PGPGIN, bio_sectors(bio));
834 	} else if (bio_op(bio) == REQ_OP_WRITE) {
835 		count_vm_events(PGPGOUT, bio_sectors(bio));
836 	}
837 
838 	submit_bio_noacct(bio);
839 }
840 EXPORT_SYMBOL(submit_bio);
841 
842 /**
843  * bio_poll - poll for BIO completions
844  * @bio: bio to poll for
845  * @iob: batches of IO
846  * @flags: BLK_POLL_* flags that control the behavior
847  *
848  * Poll for completions on queue associated with the bio. Returns number of
849  * completed entries found.
850  *
851  * Note: the caller must either be the context that submitted @bio, or
852  * be in a RCU critical section to prevent freeing of @bio.
853  */
854 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
855 {
856 	blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
857 	struct block_device *bdev;
858 	struct request_queue *q;
859 	int ret = 0;
860 
861 	bdev = READ_ONCE(bio->bi_bdev);
862 	if (!bdev)
863 		return 0;
864 
865 	q = bdev_get_queue(bdev);
866 	if (cookie == BLK_QC_T_NONE ||
867 	    !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
868 		return 0;
869 
870 	/*
871 	 * As the requests that require a zone lock are not plugged in the
872 	 * first place, directly accessing the plug instead of using
873 	 * blk_mq_plug() should not have any consequences during flushing for
874 	 * zoned devices.
875 	 */
876 	blk_flush_plug(current->plug, false);
877 
878 	/*
879 	 * We need to be able to enter a frozen queue, similar to how
880 	 * timeouts also need to do that. If that is blocked, then we can
881 	 * have pending IO when a queue freeze is started, and then the
882 	 * wait for the freeze to finish will wait for polled requests to
883 	 * timeout as the poller is preventer from entering the queue and
884 	 * completing them. As long as we prevent new IO from being queued,
885 	 * that should be all that matters.
886 	 */
887 	if (!percpu_ref_tryget(&q->q_usage_counter))
888 		return 0;
889 	if (queue_is_mq(q)) {
890 		ret = blk_mq_poll(q, cookie, iob, flags);
891 	} else {
892 		struct gendisk *disk = q->disk;
893 
894 		if (disk && disk->fops->poll_bio)
895 			ret = disk->fops->poll_bio(bio, iob, flags);
896 	}
897 	blk_queue_exit(q);
898 	return ret;
899 }
900 EXPORT_SYMBOL_GPL(bio_poll);
901 
902 /*
903  * Helper to implement file_operations.iopoll.  Requires the bio to be stored
904  * in iocb->private, and cleared before freeing the bio.
905  */
906 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
907 		    unsigned int flags)
908 {
909 	struct bio *bio;
910 	int ret = 0;
911 
912 	/*
913 	 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
914 	 * point to a freshly allocated bio at this point.  If that happens
915 	 * we have a few cases to consider:
916 	 *
917 	 *  1) the bio is beeing initialized and bi_bdev is NULL.  We can just
918 	 *     simply nothing in this case
919 	 *  2) the bio points to a not poll enabled device.  bio_poll will catch
920 	 *     this and return 0
921 	 *  3) the bio points to a poll capable device, including but not
922 	 *     limited to the one that the original bio pointed to.  In this
923 	 *     case we will call into the actual poll method and poll for I/O,
924 	 *     even if we don't need to, but it won't cause harm either.
925 	 *
926 	 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
927 	 * is still allocated. Because partitions hold a reference to the whole
928 	 * device bdev and thus disk, the disk is also still valid.  Grabbing
929 	 * a reference to the queue in bio_poll() ensures the hctxs and requests
930 	 * are still valid as well.
931 	 */
932 	rcu_read_lock();
933 	bio = READ_ONCE(kiocb->private);
934 	if (bio)
935 		ret = bio_poll(bio, iob, flags);
936 	rcu_read_unlock();
937 
938 	return ret;
939 }
940 EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
941 
942 void update_io_ticks(struct block_device *part, unsigned long now, bool end)
943 {
944 	unsigned long stamp;
945 again:
946 	stamp = READ_ONCE(part->bd_stamp);
947 	if (unlikely(time_after(now, stamp))) {
948 		if (likely(try_cmpxchg(&part->bd_stamp, &stamp, now)))
949 			__part_stat_add(part, io_ticks, end ? now - stamp : 1);
950 	}
951 	if (part->bd_partno) {
952 		part = bdev_whole(part);
953 		goto again;
954 	}
955 }
956 
957 unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op,
958 				 unsigned long start_time)
959 {
960 	part_stat_lock();
961 	update_io_ticks(bdev, start_time, false);
962 	part_stat_local_inc(bdev, in_flight[op_is_write(op)]);
963 	part_stat_unlock();
964 
965 	return start_time;
966 }
967 EXPORT_SYMBOL(bdev_start_io_acct);
968 
969 /**
970  * bio_start_io_acct - start I/O accounting for bio based drivers
971  * @bio:	bio to start account for
972  *
973  * Returns the start time that should be passed back to bio_end_io_acct().
974  */
975 unsigned long bio_start_io_acct(struct bio *bio)
976 {
977 	return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies);
978 }
979 EXPORT_SYMBOL_GPL(bio_start_io_acct);
980 
981 void bdev_end_io_acct(struct block_device *bdev, enum req_op op,
982 		      unsigned int sectors, unsigned long start_time)
983 {
984 	const int sgrp = op_stat_group(op);
985 	unsigned long now = READ_ONCE(jiffies);
986 	unsigned long duration = now - start_time;
987 
988 	part_stat_lock();
989 	update_io_ticks(bdev, now, true);
990 	part_stat_inc(bdev, ios[sgrp]);
991 	part_stat_add(bdev, sectors[sgrp], sectors);
992 	part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration));
993 	part_stat_local_dec(bdev, in_flight[op_is_write(op)]);
994 	part_stat_unlock();
995 }
996 EXPORT_SYMBOL(bdev_end_io_acct);
997 
998 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
999 			      struct block_device *orig_bdev)
1000 {
1001 	bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time);
1002 }
1003 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1004 
1005 /**
1006  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1007  * @q : the queue of the device being checked
1008  *
1009  * Description:
1010  *    Check if underlying low-level drivers of a device are busy.
1011  *    If the drivers want to export their busy state, they must set own
1012  *    exporting function using blk_queue_lld_busy() first.
1013  *
1014  *    Basically, this function is used only by request stacking drivers
1015  *    to stop dispatching requests to underlying devices when underlying
1016  *    devices are busy.  This behavior helps more I/O merging on the queue
1017  *    of the request stacking driver and prevents I/O throughput regression
1018  *    on burst I/O load.
1019  *
1020  * Return:
1021  *    0 - Not busy (The request stacking driver should dispatch request)
1022  *    1 - Busy (The request stacking driver should stop dispatching request)
1023  */
1024 int blk_lld_busy(struct request_queue *q)
1025 {
1026 	if (queue_is_mq(q) && q->mq_ops->busy)
1027 		return q->mq_ops->busy(q);
1028 
1029 	return 0;
1030 }
1031 EXPORT_SYMBOL_GPL(blk_lld_busy);
1032 
1033 int kblockd_schedule_work(struct work_struct *work)
1034 {
1035 	return queue_work(kblockd_workqueue, work);
1036 }
1037 EXPORT_SYMBOL(kblockd_schedule_work);
1038 
1039 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1040 				unsigned long delay)
1041 {
1042 	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1043 }
1044 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1045 
1046 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1047 {
1048 	struct task_struct *tsk = current;
1049 
1050 	/*
1051 	 * If this is a nested plug, don't actually assign it.
1052 	 */
1053 	if (tsk->plug)
1054 		return;
1055 
1056 	plug->mq_list = NULL;
1057 	plug->cached_rq = NULL;
1058 	plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1059 	plug->rq_count = 0;
1060 	plug->multiple_queues = false;
1061 	plug->has_elevator = false;
1062 	plug->nowait = false;
1063 	INIT_LIST_HEAD(&plug->cb_list);
1064 
1065 	/*
1066 	 * Store ordering should not be needed here, since a potential
1067 	 * preempt will imply a full memory barrier
1068 	 */
1069 	tsk->plug = plug;
1070 }
1071 
1072 /**
1073  * blk_start_plug - initialize blk_plug and track it inside the task_struct
1074  * @plug:	The &struct blk_plug that needs to be initialized
1075  *
1076  * Description:
1077  *   blk_start_plug() indicates to the block layer an intent by the caller
1078  *   to submit multiple I/O requests in a batch.  The block layer may use
1079  *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1080  *   is called.  However, the block layer may choose to submit requests
1081  *   before a call to blk_finish_plug() if the number of queued I/Os
1082  *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1083  *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1084  *   the task schedules (see below).
1085  *
1086  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1087  *   pending I/O should the task end up blocking between blk_start_plug() and
1088  *   blk_finish_plug(). This is important from a performance perspective, but
1089  *   also ensures that we don't deadlock. For instance, if the task is blocking
1090  *   for a memory allocation, memory reclaim could end up wanting to free a
1091  *   page belonging to that request that is currently residing in our private
1092  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1093  *   this kind of deadlock.
1094  */
1095 void blk_start_plug(struct blk_plug *plug)
1096 {
1097 	blk_start_plug_nr_ios(plug, 1);
1098 }
1099 EXPORT_SYMBOL(blk_start_plug);
1100 
1101 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1102 {
1103 	LIST_HEAD(callbacks);
1104 
1105 	while (!list_empty(&plug->cb_list)) {
1106 		list_splice_init(&plug->cb_list, &callbacks);
1107 
1108 		while (!list_empty(&callbacks)) {
1109 			struct blk_plug_cb *cb = list_first_entry(&callbacks,
1110 							  struct blk_plug_cb,
1111 							  list);
1112 			list_del(&cb->list);
1113 			cb->callback(cb, from_schedule);
1114 		}
1115 	}
1116 }
1117 
1118 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1119 				      int size)
1120 {
1121 	struct blk_plug *plug = current->plug;
1122 	struct blk_plug_cb *cb;
1123 
1124 	if (!plug)
1125 		return NULL;
1126 
1127 	list_for_each_entry(cb, &plug->cb_list, list)
1128 		if (cb->callback == unplug && cb->data == data)
1129 			return cb;
1130 
1131 	/* Not currently on the callback list */
1132 	BUG_ON(size < sizeof(*cb));
1133 	cb = kzalloc(size, GFP_ATOMIC);
1134 	if (cb) {
1135 		cb->data = data;
1136 		cb->callback = unplug;
1137 		list_add(&cb->list, &plug->cb_list);
1138 	}
1139 	return cb;
1140 }
1141 EXPORT_SYMBOL(blk_check_plugged);
1142 
1143 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1144 {
1145 	if (!list_empty(&plug->cb_list))
1146 		flush_plug_callbacks(plug, from_schedule);
1147 	if (!rq_list_empty(plug->mq_list))
1148 		blk_mq_flush_plug_list(plug, from_schedule);
1149 	/*
1150 	 * Unconditionally flush out cached requests, even if the unplug
1151 	 * event came from schedule. Since we know hold references to the
1152 	 * queue for cached requests, we don't want a blocked task holding
1153 	 * up a queue freeze/quiesce event.
1154 	 */
1155 	if (unlikely(!rq_list_empty(plug->cached_rq)))
1156 		blk_mq_free_plug_rqs(plug);
1157 }
1158 
1159 /**
1160  * blk_finish_plug - mark the end of a batch of submitted I/O
1161  * @plug:	The &struct blk_plug passed to blk_start_plug()
1162  *
1163  * Description:
1164  * Indicate that a batch of I/O submissions is complete.  This function
1165  * must be paired with an initial call to blk_start_plug().  The intent
1166  * is to allow the block layer to optimize I/O submission.  See the
1167  * documentation for blk_start_plug() for more information.
1168  */
1169 void blk_finish_plug(struct blk_plug *plug)
1170 {
1171 	if (plug == current->plug) {
1172 		__blk_flush_plug(plug, false);
1173 		current->plug = NULL;
1174 	}
1175 }
1176 EXPORT_SYMBOL(blk_finish_plug);
1177 
1178 void blk_io_schedule(void)
1179 {
1180 	/* Prevent hang_check timer from firing at us during very long I/O */
1181 	unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1182 
1183 	if (timeout)
1184 		io_schedule_timeout(timeout);
1185 	else
1186 		io_schedule();
1187 }
1188 EXPORT_SYMBOL_GPL(blk_io_schedule);
1189 
1190 int __init blk_dev_init(void)
1191 {
1192 	BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS));
1193 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1194 			sizeof_field(struct request, cmd_flags));
1195 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1196 			sizeof_field(struct bio, bi_opf));
1197 
1198 	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
1199 	kblockd_workqueue = alloc_workqueue("kblockd",
1200 					    WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1201 	if (!kblockd_workqueue)
1202 		panic("Failed to create kblockd\n");
1203 
1204 	blk_requestq_cachep = kmem_cache_create("request_queue",
1205 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1206 
1207 	blk_debugfs_root = debugfs_create_dir("block", NULL);
1208 
1209 	return 0;
1210 }
1211