xref: /linux/block/blk-core.c (revision 246a1c4c6b7ffba88a2553d2b88f7b6280f253a2)
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 	if (q->poll_stat)
267 		blk_stat_remove_callback(q, q->poll_cb);
268 	blk_stat_free_callback(q->poll_cb);
269 
270 	blk_free_queue_stats(q->stats);
271 	kfree(q->poll_stat);
272 
273 	if (queue_is_mq(q))
274 		blk_mq_release(q);
275 
276 	ida_free(&blk_queue_ida, q->id);
277 	call_rcu(&q->rcu_head, blk_free_queue_rcu);
278 }
279 
280 /**
281  * blk_put_queue - decrement the request_queue refcount
282  * @q: the request_queue structure to decrement the refcount for
283  *
284  * Decrements the refcount of the request_queue and free it when the refcount
285  * reaches 0.
286  *
287  * Context: Can sleep.
288  */
289 void blk_put_queue(struct request_queue *q)
290 {
291 	might_sleep();
292 	if (refcount_dec_and_test(&q->refs))
293 		blk_free_queue(q);
294 }
295 EXPORT_SYMBOL(blk_put_queue);
296 
297 void blk_queue_start_drain(struct request_queue *q)
298 {
299 	/*
300 	 * When queue DYING flag is set, we need to block new req
301 	 * entering queue, so we call blk_freeze_queue_start() to
302 	 * prevent I/O from crossing blk_queue_enter().
303 	 */
304 	blk_freeze_queue_start(q);
305 	if (queue_is_mq(q))
306 		blk_mq_wake_waiters(q);
307 	/* Make blk_queue_enter() reexamine the DYING flag. */
308 	wake_up_all(&q->mq_freeze_wq);
309 }
310 
311 /**
312  * blk_queue_enter() - try to increase q->q_usage_counter
313  * @q: request queue pointer
314  * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
315  */
316 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
317 {
318 	const bool pm = flags & BLK_MQ_REQ_PM;
319 
320 	while (!blk_try_enter_queue(q, pm)) {
321 		if (flags & BLK_MQ_REQ_NOWAIT)
322 			return -EAGAIN;
323 
324 		/*
325 		 * read pair of barrier in blk_freeze_queue_start(), we need to
326 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
327 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
328 		 * following wait may never return if the two reads are
329 		 * reordered.
330 		 */
331 		smp_rmb();
332 		wait_event(q->mq_freeze_wq,
333 			   (!q->mq_freeze_depth &&
334 			    blk_pm_resume_queue(pm, q)) ||
335 			   blk_queue_dying(q));
336 		if (blk_queue_dying(q))
337 			return -ENODEV;
338 	}
339 
340 	return 0;
341 }
342 
343 int __bio_queue_enter(struct request_queue *q, struct bio *bio)
344 {
345 	while (!blk_try_enter_queue(q, false)) {
346 		struct gendisk *disk = bio->bi_bdev->bd_disk;
347 
348 		if (bio->bi_opf & REQ_NOWAIT) {
349 			if (test_bit(GD_DEAD, &disk->state))
350 				goto dead;
351 			bio_wouldblock_error(bio);
352 			return -EAGAIN;
353 		}
354 
355 		/*
356 		 * read pair of barrier in blk_freeze_queue_start(), we need to
357 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
358 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
359 		 * following wait may never return if the two reads are
360 		 * reordered.
361 		 */
362 		smp_rmb();
363 		wait_event(q->mq_freeze_wq,
364 			   (!q->mq_freeze_depth &&
365 			    blk_pm_resume_queue(false, q)) ||
366 			   test_bit(GD_DEAD, &disk->state));
367 		if (test_bit(GD_DEAD, &disk->state))
368 			goto dead;
369 	}
370 
371 	return 0;
372 dead:
373 	bio_io_error(bio);
374 	return -ENODEV;
375 }
376 
377 void blk_queue_exit(struct request_queue *q)
378 {
379 	percpu_ref_put(&q->q_usage_counter);
380 }
381 
382 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
383 {
384 	struct request_queue *q =
385 		container_of(ref, struct request_queue, q_usage_counter);
386 
387 	wake_up_all(&q->mq_freeze_wq);
388 }
389 
390 static void blk_rq_timed_out_timer(struct timer_list *t)
391 {
392 	struct request_queue *q = from_timer(q, t, timeout);
393 
394 	kblockd_schedule_work(&q->timeout_work);
395 }
396 
397 static void blk_timeout_work(struct work_struct *work)
398 {
399 }
400 
401 struct request_queue *blk_alloc_queue(int node_id)
402 {
403 	struct request_queue *q;
404 
405 	q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO,
406 				  node_id);
407 	if (!q)
408 		return NULL;
409 
410 	q->last_merge = NULL;
411 
412 	q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL);
413 	if (q->id < 0)
414 		goto fail_q;
415 
416 	q->stats = blk_alloc_queue_stats();
417 	if (!q->stats)
418 		goto fail_id;
419 
420 	q->node = node_id;
421 
422 	atomic_set(&q->nr_active_requests_shared_tags, 0);
423 
424 	timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
425 	INIT_WORK(&q->timeout_work, blk_timeout_work);
426 	INIT_LIST_HEAD(&q->icq_list);
427 
428 	refcount_set(&q->refs, 1);
429 	mutex_init(&q->debugfs_mutex);
430 	mutex_init(&q->sysfs_lock);
431 	mutex_init(&q->sysfs_dir_lock);
432 	spin_lock_init(&q->queue_lock);
433 
434 	init_waitqueue_head(&q->mq_freeze_wq);
435 	mutex_init(&q->mq_freeze_lock);
436 
437 	/*
438 	 * Init percpu_ref in atomic mode so that it's faster to shutdown.
439 	 * See blk_register_queue() for details.
440 	 */
441 	if (percpu_ref_init(&q->q_usage_counter,
442 				blk_queue_usage_counter_release,
443 				PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
444 		goto fail_stats;
445 
446 	blk_set_default_limits(&q->limits);
447 	q->nr_requests = BLKDEV_DEFAULT_RQ;
448 
449 	return q;
450 
451 fail_stats:
452 	blk_free_queue_stats(q->stats);
453 fail_id:
454 	ida_free(&blk_queue_ida, q->id);
455 fail_q:
456 	kmem_cache_free(blk_requestq_cachep, q);
457 	return NULL;
458 }
459 
460 /**
461  * blk_get_queue - increment the request_queue refcount
462  * @q: the request_queue structure to increment the refcount for
463  *
464  * Increment the refcount of the request_queue kobject.
465  *
466  * Context: Any context.
467  */
468 bool blk_get_queue(struct request_queue *q)
469 {
470 	if (unlikely(blk_queue_dying(q)))
471 		return false;
472 	refcount_inc(&q->refs);
473 	return true;
474 }
475 EXPORT_SYMBOL(blk_get_queue);
476 
477 #ifdef CONFIG_FAIL_MAKE_REQUEST
478 
479 static DECLARE_FAULT_ATTR(fail_make_request);
480 
481 static int __init setup_fail_make_request(char *str)
482 {
483 	return setup_fault_attr(&fail_make_request, str);
484 }
485 __setup("fail_make_request=", setup_fail_make_request);
486 
487 bool should_fail_request(struct block_device *part, unsigned int bytes)
488 {
489 	return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
490 }
491 
492 static int __init fail_make_request_debugfs(void)
493 {
494 	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
495 						NULL, &fail_make_request);
496 
497 	return PTR_ERR_OR_ZERO(dir);
498 }
499 
500 late_initcall(fail_make_request_debugfs);
501 #endif /* CONFIG_FAIL_MAKE_REQUEST */
502 
503 static inline void bio_check_ro(struct bio *bio)
504 {
505 	if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
506 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
507 			return;
508 		pr_warn("Trying to write to read-only block-device %pg\n",
509 			bio->bi_bdev);
510 		/* Older lvm-tools actually trigger this */
511 	}
512 }
513 
514 static noinline int should_fail_bio(struct bio *bio)
515 {
516 	if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
517 		return -EIO;
518 	return 0;
519 }
520 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
521 
522 /*
523  * Check whether this bio extends beyond the end of the device or partition.
524  * This may well happen - the kernel calls bread() without checking the size of
525  * the device, e.g., when mounting a file system.
526  */
527 static inline int bio_check_eod(struct bio *bio)
528 {
529 	sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
530 	unsigned int nr_sectors = bio_sectors(bio);
531 
532 	if (nr_sectors && maxsector &&
533 	    (nr_sectors > maxsector ||
534 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
535 		pr_info_ratelimited("%s: attempt to access beyond end of device\n"
536 				    "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
537 				    current->comm, bio->bi_bdev, bio->bi_opf,
538 				    bio->bi_iter.bi_sector, nr_sectors, maxsector);
539 		return -EIO;
540 	}
541 	return 0;
542 }
543 
544 /*
545  * Remap block n of partition p to block n+start(p) of the disk.
546  */
547 static int blk_partition_remap(struct bio *bio)
548 {
549 	struct block_device *p = bio->bi_bdev;
550 
551 	if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
552 		return -EIO;
553 	if (bio_sectors(bio)) {
554 		bio->bi_iter.bi_sector += p->bd_start_sect;
555 		trace_block_bio_remap(bio, p->bd_dev,
556 				      bio->bi_iter.bi_sector -
557 				      p->bd_start_sect);
558 	}
559 	bio_set_flag(bio, BIO_REMAPPED);
560 	return 0;
561 }
562 
563 /*
564  * Check write append to a zoned block device.
565  */
566 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
567 						 struct bio *bio)
568 {
569 	int nr_sectors = bio_sectors(bio);
570 
571 	/* Only applicable to zoned block devices */
572 	if (!bdev_is_zoned(bio->bi_bdev))
573 		return BLK_STS_NOTSUPP;
574 
575 	/* The bio sector must point to the start of a sequential zone */
576 	if (bio->bi_iter.bi_sector & (bdev_zone_sectors(bio->bi_bdev) - 1) ||
577 	    !bio_zone_is_seq(bio))
578 		return BLK_STS_IOERR;
579 
580 	/*
581 	 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
582 	 * split and could result in non-contiguous sectors being written in
583 	 * different zones.
584 	 */
585 	if (nr_sectors > q->limits.chunk_sectors)
586 		return BLK_STS_IOERR;
587 
588 	/* Make sure the BIO is small enough and will not get split */
589 	if (nr_sectors > q->limits.max_zone_append_sectors)
590 		return BLK_STS_IOERR;
591 
592 	bio->bi_opf |= REQ_NOMERGE;
593 
594 	return BLK_STS_OK;
595 }
596 
597 static void __submit_bio(struct bio *bio)
598 {
599 	struct gendisk *disk = bio->bi_bdev->bd_disk;
600 
601 	if (unlikely(!blk_crypto_bio_prep(&bio)))
602 		return;
603 
604 	if (!disk->fops->submit_bio) {
605 		blk_mq_submit_bio(bio);
606 	} else if (likely(bio_queue_enter(bio) == 0)) {
607 		disk->fops->submit_bio(bio);
608 		blk_queue_exit(disk->queue);
609 	}
610 }
611 
612 /*
613  * The loop in this function may be a bit non-obvious, and so deserves some
614  * explanation:
615  *
616  *  - Before entering the loop, bio->bi_next is NULL (as all callers ensure
617  *    that), so we have a list with a single bio.
618  *  - We pretend that we have just taken it off a longer list, so we assign
619  *    bio_list to a pointer to the bio_list_on_stack, thus initialising the
620  *    bio_list of new bios to be added.  ->submit_bio() may indeed add some more
621  *    bios through a recursive call to submit_bio_noacct.  If it did, we find a
622  *    non-NULL value in bio_list and re-enter the loop from the top.
623  *  - In this case we really did just take the bio of the top of the list (no
624  *    pretending) and so remove it from bio_list, and call into ->submit_bio()
625  *    again.
626  *
627  * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
628  * bio_list_on_stack[1] contains bios that were submitted before the current
629  *	->submit_bio, but that haven't been processed yet.
630  */
631 static void __submit_bio_noacct(struct bio *bio)
632 {
633 	struct bio_list bio_list_on_stack[2];
634 
635 	BUG_ON(bio->bi_next);
636 
637 	bio_list_init(&bio_list_on_stack[0]);
638 	current->bio_list = bio_list_on_stack;
639 
640 	do {
641 		struct request_queue *q = bdev_get_queue(bio->bi_bdev);
642 		struct bio_list lower, same;
643 
644 		/*
645 		 * Create a fresh bio_list for all subordinate requests.
646 		 */
647 		bio_list_on_stack[1] = bio_list_on_stack[0];
648 		bio_list_init(&bio_list_on_stack[0]);
649 
650 		__submit_bio(bio);
651 
652 		/*
653 		 * Sort new bios into those for a lower level and those for the
654 		 * same level.
655 		 */
656 		bio_list_init(&lower);
657 		bio_list_init(&same);
658 		while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
659 			if (q == bdev_get_queue(bio->bi_bdev))
660 				bio_list_add(&same, bio);
661 			else
662 				bio_list_add(&lower, bio);
663 
664 		/*
665 		 * Now assemble so we handle the lowest level first.
666 		 */
667 		bio_list_merge(&bio_list_on_stack[0], &lower);
668 		bio_list_merge(&bio_list_on_stack[0], &same);
669 		bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
670 	} while ((bio = bio_list_pop(&bio_list_on_stack[0])));
671 
672 	current->bio_list = NULL;
673 }
674 
675 static void __submit_bio_noacct_mq(struct bio *bio)
676 {
677 	struct bio_list bio_list[2] = { };
678 
679 	current->bio_list = bio_list;
680 
681 	do {
682 		__submit_bio(bio);
683 	} while ((bio = bio_list_pop(&bio_list[0])));
684 
685 	current->bio_list = NULL;
686 }
687 
688 void submit_bio_noacct_nocheck(struct bio *bio)
689 {
690 	/*
691 	 * We only want one ->submit_bio to be active at a time, else stack
692 	 * usage with stacked devices could be a problem.  Use current->bio_list
693 	 * to collect a list of requests submited by a ->submit_bio method while
694 	 * it is active, and then process them after it returned.
695 	 */
696 	if (current->bio_list)
697 		bio_list_add(&current->bio_list[0], bio);
698 	else if (!bio->bi_bdev->bd_disk->fops->submit_bio)
699 		__submit_bio_noacct_mq(bio);
700 	else
701 		__submit_bio_noacct(bio);
702 }
703 
704 /**
705  * submit_bio_noacct - re-submit a bio to the block device layer for I/O
706  * @bio:  The bio describing the location in memory and on the device.
707  *
708  * This is a version of submit_bio() that shall only be used for I/O that is
709  * resubmitted to lower level drivers by stacking block drivers.  All file
710  * systems and other upper level users of the block layer should use
711  * submit_bio() instead.
712  */
713 void submit_bio_noacct(struct bio *bio)
714 {
715 	struct block_device *bdev = bio->bi_bdev;
716 	struct request_queue *q = bdev_get_queue(bdev);
717 	blk_status_t status = BLK_STS_IOERR;
718 	struct blk_plug *plug;
719 
720 	might_sleep();
721 
722 	plug = blk_mq_plug(bio);
723 	if (plug && plug->nowait)
724 		bio->bi_opf |= REQ_NOWAIT;
725 
726 	/*
727 	 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
728 	 * if queue does not support NOWAIT.
729 	 */
730 	if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev))
731 		goto not_supported;
732 
733 	if (should_fail_bio(bio))
734 		goto end_io;
735 	bio_check_ro(bio);
736 	if (!bio_flagged(bio, BIO_REMAPPED)) {
737 		if (unlikely(bio_check_eod(bio)))
738 			goto end_io;
739 		if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
740 			goto end_io;
741 	}
742 
743 	/*
744 	 * Filter flush bio's early so that bio based drivers without flush
745 	 * support don't have to worry about them.
746 	 */
747 	if (op_is_flush(bio->bi_opf)) {
748 		if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE &&
749 				 bio_op(bio) != REQ_OP_ZONE_APPEND))
750 			goto end_io;
751 		if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
752 			bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
753 			if (!bio_sectors(bio)) {
754 				status = BLK_STS_OK;
755 				goto end_io;
756 			}
757 		}
758 	}
759 
760 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
761 		bio_clear_polled(bio);
762 
763 	switch (bio_op(bio)) {
764 	case REQ_OP_DISCARD:
765 		if (!bdev_max_discard_sectors(bdev))
766 			goto not_supported;
767 		break;
768 	case REQ_OP_SECURE_ERASE:
769 		if (!bdev_max_secure_erase_sectors(bdev))
770 			goto not_supported;
771 		break;
772 	case REQ_OP_ZONE_APPEND:
773 		status = blk_check_zone_append(q, bio);
774 		if (status != BLK_STS_OK)
775 			goto end_io;
776 		break;
777 	case REQ_OP_ZONE_RESET:
778 	case REQ_OP_ZONE_OPEN:
779 	case REQ_OP_ZONE_CLOSE:
780 	case REQ_OP_ZONE_FINISH:
781 		if (!bdev_is_zoned(bio->bi_bdev))
782 			goto not_supported;
783 		break;
784 	case REQ_OP_ZONE_RESET_ALL:
785 		if (!bdev_is_zoned(bio->bi_bdev) || !blk_queue_zone_resetall(q))
786 			goto not_supported;
787 		break;
788 	case REQ_OP_WRITE_ZEROES:
789 		if (!q->limits.max_write_zeroes_sectors)
790 			goto not_supported;
791 		break;
792 	default:
793 		break;
794 	}
795 
796 	if (blk_throtl_bio(bio))
797 		return;
798 
799 	blk_cgroup_bio_start(bio);
800 	blkcg_bio_issue_init(bio);
801 
802 	if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
803 		trace_block_bio_queue(bio);
804 		/* Now that enqueuing has been traced, we need to trace
805 		 * completion as well.
806 		 */
807 		bio_set_flag(bio, BIO_TRACE_COMPLETION);
808 	}
809 	submit_bio_noacct_nocheck(bio);
810 	return;
811 
812 not_supported:
813 	status = BLK_STS_NOTSUPP;
814 end_io:
815 	bio->bi_status = status;
816 	bio_endio(bio);
817 }
818 EXPORT_SYMBOL(submit_bio_noacct);
819 
820 /**
821  * submit_bio - submit a bio to the block device layer for I/O
822  * @bio: The &struct bio which describes the I/O
823  *
824  * submit_bio() is used to submit I/O requests to block devices.  It is passed a
825  * fully set up &struct bio that describes the I/O that needs to be done.  The
826  * bio will be send to the device described by the bi_bdev field.
827  *
828  * The success/failure status of the request, along with notification of
829  * completion, is delivered asynchronously through the ->bi_end_io() callback
830  * in @bio.  The bio must NOT be touched by the caller until ->bi_end_io() has
831  * been called.
832  */
833 void submit_bio(struct bio *bio)
834 {
835 	if (blkcg_punt_bio_submit(bio))
836 		return;
837 
838 	if (bio_op(bio) == REQ_OP_READ) {
839 		task_io_account_read(bio->bi_iter.bi_size);
840 		count_vm_events(PGPGIN, bio_sectors(bio));
841 	} else if (bio_op(bio) == REQ_OP_WRITE) {
842 		count_vm_events(PGPGOUT, bio_sectors(bio));
843 	}
844 
845 	submit_bio_noacct(bio);
846 }
847 EXPORT_SYMBOL(submit_bio);
848 
849 /**
850  * bio_poll - poll for BIO completions
851  * @bio: bio to poll for
852  * @iob: batches of IO
853  * @flags: BLK_POLL_* flags that control the behavior
854  *
855  * Poll for completions on queue associated with the bio. Returns number of
856  * completed entries found.
857  *
858  * Note: the caller must either be the context that submitted @bio, or
859  * be in a RCU critical section to prevent freeing of @bio.
860  */
861 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
862 {
863 	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
864 	blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
865 	int ret = 0;
866 
867 	if (cookie == BLK_QC_T_NONE ||
868 	    !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
869 		return 0;
870 
871 	/*
872 	 * As the requests that require a zone lock are not plugged in the
873 	 * first place, directly accessing the plug instead of using
874 	 * blk_mq_plug() should not have any consequences during flushing for
875 	 * zoned devices.
876 	 */
877 	blk_flush_plug(current->plug, false);
878 
879 	if (bio_queue_enter(bio))
880 		return 0;
881 	if (queue_is_mq(q)) {
882 		ret = blk_mq_poll(q, cookie, iob, flags);
883 	} else {
884 		struct gendisk *disk = q->disk;
885 
886 		if (disk && disk->fops->poll_bio)
887 			ret = disk->fops->poll_bio(bio, iob, flags);
888 	}
889 	blk_queue_exit(q);
890 	return ret;
891 }
892 EXPORT_SYMBOL_GPL(bio_poll);
893 
894 /*
895  * Helper to implement file_operations.iopoll.  Requires the bio to be stored
896  * in iocb->private, and cleared before freeing the bio.
897  */
898 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
899 		    unsigned int flags)
900 {
901 	struct bio *bio;
902 	int ret = 0;
903 
904 	/*
905 	 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
906 	 * point to a freshly allocated bio at this point.  If that happens
907 	 * we have a few cases to consider:
908 	 *
909 	 *  1) the bio is beeing initialized and bi_bdev is NULL.  We can just
910 	 *     simply nothing in this case
911 	 *  2) the bio points to a not poll enabled device.  bio_poll will catch
912 	 *     this and return 0
913 	 *  3) the bio points to a poll capable device, including but not
914 	 *     limited to the one that the original bio pointed to.  In this
915 	 *     case we will call into the actual poll method and poll for I/O,
916 	 *     even if we don't need to, but it won't cause harm either.
917 	 *
918 	 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
919 	 * is still allocated. Because partitions hold a reference to the whole
920 	 * device bdev and thus disk, the disk is also still valid.  Grabbing
921 	 * a reference to the queue in bio_poll() ensures the hctxs and requests
922 	 * are still valid as well.
923 	 */
924 	rcu_read_lock();
925 	bio = READ_ONCE(kiocb->private);
926 	if (bio && bio->bi_bdev)
927 		ret = bio_poll(bio, iob, flags);
928 	rcu_read_unlock();
929 
930 	return ret;
931 }
932 EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
933 
934 void update_io_ticks(struct block_device *part, unsigned long now, bool end)
935 {
936 	unsigned long stamp;
937 again:
938 	stamp = READ_ONCE(part->bd_stamp);
939 	if (unlikely(time_after(now, stamp))) {
940 		if (likely(try_cmpxchg(&part->bd_stamp, &stamp, now)))
941 			__part_stat_add(part, io_ticks, end ? now - stamp : 1);
942 	}
943 	if (part->bd_partno) {
944 		part = bdev_whole(part);
945 		goto again;
946 	}
947 }
948 
949 unsigned long bdev_start_io_acct(struct block_device *bdev,
950 				 unsigned int sectors, enum req_op op,
951 				 unsigned long start_time)
952 {
953 	const int sgrp = op_stat_group(op);
954 
955 	part_stat_lock();
956 	update_io_ticks(bdev, start_time, false);
957 	part_stat_inc(bdev, ios[sgrp]);
958 	part_stat_add(bdev, sectors[sgrp], sectors);
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_sectors(bio),
975 				  bio_op(bio), jiffies);
976 }
977 EXPORT_SYMBOL_GPL(bio_start_io_acct);
978 
979 void bdev_end_io_acct(struct block_device *bdev, enum req_op op,
980 		      unsigned long start_time)
981 {
982 	const int sgrp = op_stat_group(op);
983 	unsigned long now = READ_ONCE(jiffies);
984 	unsigned long duration = now - start_time;
985 
986 	part_stat_lock();
987 	update_io_ticks(bdev, now, true);
988 	part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration));
989 	part_stat_local_dec(bdev, in_flight[op_is_write(op)]);
990 	part_stat_unlock();
991 }
992 EXPORT_SYMBOL(bdev_end_io_acct);
993 
994 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
995 			      struct block_device *orig_bdev)
996 {
997 	bdev_end_io_acct(orig_bdev, bio_op(bio), start_time);
998 }
999 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1000 
1001 /**
1002  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1003  * @q : the queue of the device being checked
1004  *
1005  * Description:
1006  *    Check if underlying low-level drivers of a device are busy.
1007  *    If the drivers want to export their busy state, they must set own
1008  *    exporting function using blk_queue_lld_busy() first.
1009  *
1010  *    Basically, this function is used only by request stacking drivers
1011  *    to stop dispatching requests to underlying devices when underlying
1012  *    devices are busy.  This behavior helps more I/O merging on the queue
1013  *    of the request stacking driver and prevents I/O throughput regression
1014  *    on burst I/O load.
1015  *
1016  * Return:
1017  *    0 - Not busy (The request stacking driver should dispatch request)
1018  *    1 - Busy (The request stacking driver should stop dispatching request)
1019  */
1020 int blk_lld_busy(struct request_queue *q)
1021 {
1022 	if (queue_is_mq(q) && q->mq_ops->busy)
1023 		return q->mq_ops->busy(q);
1024 
1025 	return 0;
1026 }
1027 EXPORT_SYMBOL_GPL(blk_lld_busy);
1028 
1029 int kblockd_schedule_work(struct work_struct *work)
1030 {
1031 	return queue_work(kblockd_workqueue, work);
1032 }
1033 EXPORT_SYMBOL(kblockd_schedule_work);
1034 
1035 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1036 				unsigned long delay)
1037 {
1038 	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1039 }
1040 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1041 
1042 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1043 {
1044 	struct task_struct *tsk = current;
1045 
1046 	/*
1047 	 * If this is a nested plug, don't actually assign it.
1048 	 */
1049 	if (tsk->plug)
1050 		return;
1051 
1052 	plug->mq_list = NULL;
1053 	plug->cached_rq = NULL;
1054 	plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1055 	plug->rq_count = 0;
1056 	plug->multiple_queues = false;
1057 	plug->has_elevator = false;
1058 	plug->nowait = false;
1059 	INIT_LIST_HEAD(&plug->cb_list);
1060 
1061 	/*
1062 	 * Store ordering should not be needed here, since a potential
1063 	 * preempt will imply a full memory barrier
1064 	 */
1065 	tsk->plug = plug;
1066 }
1067 
1068 /**
1069  * blk_start_plug - initialize blk_plug and track it inside the task_struct
1070  * @plug:	The &struct blk_plug that needs to be initialized
1071  *
1072  * Description:
1073  *   blk_start_plug() indicates to the block layer an intent by the caller
1074  *   to submit multiple I/O requests in a batch.  The block layer may use
1075  *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1076  *   is called.  However, the block layer may choose to submit requests
1077  *   before a call to blk_finish_plug() if the number of queued I/Os
1078  *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1079  *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1080  *   the task schedules (see below).
1081  *
1082  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1083  *   pending I/O should the task end up blocking between blk_start_plug() and
1084  *   blk_finish_plug(). This is important from a performance perspective, but
1085  *   also ensures that we don't deadlock. For instance, if the task is blocking
1086  *   for a memory allocation, memory reclaim could end up wanting to free a
1087  *   page belonging to that request that is currently residing in our private
1088  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1089  *   this kind of deadlock.
1090  */
1091 void blk_start_plug(struct blk_plug *plug)
1092 {
1093 	blk_start_plug_nr_ios(plug, 1);
1094 }
1095 EXPORT_SYMBOL(blk_start_plug);
1096 
1097 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1098 {
1099 	LIST_HEAD(callbacks);
1100 
1101 	while (!list_empty(&plug->cb_list)) {
1102 		list_splice_init(&plug->cb_list, &callbacks);
1103 
1104 		while (!list_empty(&callbacks)) {
1105 			struct blk_plug_cb *cb = list_first_entry(&callbacks,
1106 							  struct blk_plug_cb,
1107 							  list);
1108 			list_del(&cb->list);
1109 			cb->callback(cb, from_schedule);
1110 		}
1111 	}
1112 }
1113 
1114 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1115 				      int size)
1116 {
1117 	struct blk_plug *plug = current->plug;
1118 	struct blk_plug_cb *cb;
1119 
1120 	if (!plug)
1121 		return NULL;
1122 
1123 	list_for_each_entry(cb, &plug->cb_list, list)
1124 		if (cb->callback == unplug && cb->data == data)
1125 			return cb;
1126 
1127 	/* Not currently on the callback list */
1128 	BUG_ON(size < sizeof(*cb));
1129 	cb = kzalloc(size, GFP_ATOMIC);
1130 	if (cb) {
1131 		cb->data = data;
1132 		cb->callback = unplug;
1133 		list_add(&cb->list, &plug->cb_list);
1134 	}
1135 	return cb;
1136 }
1137 EXPORT_SYMBOL(blk_check_plugged);
1138 
1139 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1140 {
1141 	if (!list_empty(&plug->cb_list))
1142 		flush_plug_callbacks(plug, from_schedule);
1143 	if (!rq_list_empty(plug->mq_list))
1144 		blk_mq_flush_plug_list(plug, from_schedule);
1145 	/*
1146 	 * Unconditionally flush out cached requests, even if the unplug
1147 	 * event came from schedule. Since we know hold references to the
1148 	 * queue for cached requests, we don't want a blocked task holding
1149 	 * up a queue freeze/quiesce event.
1150 	 */
1151 	if (unlikely(!rq_list_empty(plug->cached_rq)))
1152 		blk_mq_free_plug_rqs(plug);
1153 }
1154 
1155 /**
1156  * blk_finish_plug - mark the end of a batch of submitted I/O
1157  * @plug:	The &struct blk_plug passed to blk_start_plug()
1158  *
1159  * Description:
1160  * Indicate that a batch of I/O submissions is complete.  This function
1161  * must be paired with an initial call to blk_start_plug().  The intent
1162  * is to allow the block layer to optimize I/O submission.  See the
1163  * documentation for blk_start_plug() for more information.
1164  */
1165 void blk_finish_plug(struct blk_plug *plug)
1166 {
1167 	if (plug == current->plug) {
1168 		__blk_flush_plug(plug, false);
1169 		current->plug = NULL;
1170 	}
1171 }
1172 EXPORT_SYMBOL(blk_finish_plug);
1173 
1174 void blk_io_schedule(void)
1175 {
1176 	/* Prevent hang_check timer from firing at us during very long I/O */
1177 	unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1178 
1179 	if (timeout)
1180 		io_schedule_timeout(timeout);
1181 	else
1182 		io_schedule();
1183 }
1184 EXPORT_SYMBOL_GPL(blk_io_schedule);
1185 
1186 int __init blk_dev_init(void)
1187 {
1188 	BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS));
1189 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1190 			sizeof_field(struct request, cmd_flags));
1191 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1192 			sizeof_field(struct bio, bi_opf));
1193 
1194 	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
1195 	kblockd_workqueue = alloc_workqueue("kblockd",
1196 					    WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1197 	if (!kblockd_workqueue)
1198 		panic("Failed to create kblockd\n");
1199 
1200 	blk_requestq_cachep = kmem_cache_create("request_queue",
1201 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1202 
1203 	blk_debugfs_root = debugfs_create_dir("block", NULL);
1204 
1205 	return 0;
1206 }
1207