xref: /openbmc/linux/block/blk-core.c (revision 2dd6532e)
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/psi.h>
41 #include <linux/part_stat.h>
42 #include <linux/sched/sysctl.h>
43 #include <linux/blk-crypto.h>
44 
45 #define CREATE_TRACE_POINTS
46 #include <trace/events/block.h>
47 
48 #include "blk.h"
49 #include "blk-mq-sched.h"
50 #include "blk-pm.h"
51 #include "blk-cgroup.h"
52 #include "blk-throttle.h"
53 
54 struct dentry *blk_debugfs_root;
55 
56 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
57 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
58 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
59 EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
60 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
61 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert);
62 
63 DEFINE_IDA(blk_queue_ida);
64 
65 /*
66  * For queue allocation
67  */
68 struct kmem_cache *blk_requestq_cachep;
69 struct kmem_cache *blk_requestq_srcu_cachep;
70 
71 /*
72  * Controlling structure to kblockd
73  */
74 static struct workqueue_struct *kblockd_workqueue;
75 
76 /**
77  * blk_queue_flag_set - atomically set a queue flag
78  * @flag: flag to be set
79  * @q: request queue
80  */
81 void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
82 {
83 	set_bit(flag, &q->queue_flags);
84 }
85 EXPORT_SYMBOL(blk_queue_flag_set);
86 
87 /**
88  * blk_queue_flag_clear - atomically clear a queue flag
89  * @flag: flag to be cleared
90  * @q: request queue
91  */
92 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
93 {
94 	clear_bit(flag, &q->queue_flags);
95 }
96 EXPORT_SYMBOL(blk_queue_flag_clear);
97 
98 /**
99  * blk_queue_flag_test_and_set - atomically test and set a queue flag
100  * @flag: flag to be set
101  * @q: request queue
102  *
103  * Returns the previous value of @flag - 0 if the flag was not set and 1 if
104  * the flag was already set.
105  */
106 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
107 {
108 	return test_and_set_bit(flag, &q->queue_flags);
109 }
110 EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
111 
112 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
113 static const char *const blk_op_name[] = {
114 	REQ_OP_NAME(READ),
115 	REQ_OP_NAME(WRITE),
116 	REQ_OP_NAME(FLUSH),
117 	REQ_OP_NAME(DISCARD),
118 	REQ_OP_NAME(SECURE_ERASE),
119 	REQ_OP_NAME(ZONE_RESET),
120 	REQ_OP_NAME(ZONE_RESET_ALL),
121 	REQ_OP_NAME(ZONE_OPEN),
122 	REQ_OP_NAME(ZONE_CLOSE),
123 	REQ_OP_NAME(ZONE_FINISH),
124 	REQ_OP_NAME(ZONE_APPEND),
125 	REQ_OP_NAME(WRITE_ZEROES),
126 	REQ_OP_NAME(DRV_IN),
127 	REQ_OP_NAME(DRV_OUT),
128 };
129 #undef REQ_OP_NAME
130 
131 /**
132  * blk_op_str - Return string XXX in the REQ_OP_XXX.
133  * @op: REQ_OP_XXX.
134  *
135  * Description: Centralize block layer function to convert REQ_OP_XXX into
136  * string format. Useful in the debugging and tracing bio or request. For
137  * invalid REQ_OP_XXX it returns string "UNKNOWN".
138  */
139 inline const char *blk_op_str(unsigned int op)
140 {
141 	const char *op_str = "UNKNOWN";
142 
143 	if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
144 		op_str = blk_op_name[op];
145 
146 	return op_str;
147 }
148 EXPORT_SYMBOL_GPL(blk_op_str);
149 
150 static const struct {
151 	int		errno;
152 	const char	*name;
153 } blk_errors[] = {
154 	[BLK_STS_OK]		= { 0,		"" },
155 	[BLK_STS_NOTSUPP]	= { -EOPNOTSUPP, "operation not supported" },
156 	[BLK_STS_TIMEOUT]	= { -ETIMEDOUT,	"timeout" },
157 	[BLK_STS_NOSPC]		= { -ENOSPC,	"critical space allocation" },
158 	[BLK_STS_TRANSPORT]	= { -ENOLINK,	"recoverable transport" },
159 	[BLK_STS_TARGET]	= { -EREMOTEIO,	"critical target" },
160 	[BLK_STS_NEXUS]		= { -EBADE,	"critical nexus" },
161 	[BLK_STS_MEDIUM]	= { -ENODATA,	"critical medium" },
162 	[BLK_STS_PROTECTION]	= { -EILSEQ,	"protection" },
163 	[BLK_STS_RESOURCE]	= { -ENOMEM,	"kernel resource" },
164 	[BLK_STS_DEV_RESOURCE]	= { -EBUSY,	"device resource" },
165 	[BLK_STS_AGAIN]		= { -EAGAIN,	"nonblocking retry" },
166 	[BLK_STS_OFFLINE]	= { -ENODEV,	"device offline" },
167 
168 	/* device mapper special case, should not leak out: */
169 	[BLK_STS_DM_REQUEUE]	= { -EREMCHG, "dm internal retry" },
170 
171 	/* zone device specific errors */
172 	[BLK_STS_ZONE_OPEN_RESOURCE]	= { -ETOOMANYREFS, "open zones exceeded" },
173 	[BLK_STS_ZONE_ACTIVE_RESOURCE]	= { -EOVERFLOW, "active zones exceeded" },
174 
175 	/* everything else not covered above: */
176 	[BLK_STS_IOERR]		= { -EIO,	"I/O" },
177 };
178 
179 blk_status_t errno_to_blk_status(int errno)
180 {
181 	int i;
182 
183 	for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
184 		if (blk_errors[i].errno == errno)
185 			return (__force blk_status_t)i;
186 	}
187 
188 	return BLK_STS_IOERR;
189 }
190 EXPORT_SYMBOL_GPL(errno_to_blk_status);
191 
192 int blk_status_to_errno(blk_status_t status)
193 {
194 	int idx = (__force int)status;
195 
196 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
197 		return -EIO;
198 	return blk_errors[idx].errno;
199 }
200 EXPORT_SYMBOL_GPL(blk_status_to_errno);
201 
202 const char *blk_status_to_str(blk_status_t status)
203 {
204 	int idx = (__force int)status;
205 
206 	if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
207 		return "<null>";
208 	return blk_errors[idx].name;
209 }
210 
211 /**
212  * blk_sync_queue - cancel any pending callbacks on a queue
213  * @q: the queue
214  *
215  * Description:
216  *     The block layer may perform asynchronous callback activity
217  *     on a queue, such as calling the unplug function after a timeout.
218  *     A block device may call blk_sync_queue to ensure that any
219  *     such activity is cancelled, thus allowing it to release resources
220  *     that the callbacks might use. The caller must already have made sure
221  *     that its ->submit_bio will not re-add plugging prior to calling
222  *     this function.
223  *
224  *     This function does not cancel any asynchronous activity arising
225  *     out of elevator or throttling code. That would require elevator_exit()
226  *     and blkcg_exit_queue() to be called with queue lock initialized.
227  *
228  */
229 void blk_sync_queue(struct request_queue *q)
230 {
231 	del_timer_sync(&q->timeout);
232 	cancel_work_sync(&q->timeout_work);
233 }
234 EXPORT_SYMBOL(blk_sync_queue);
235 
236 /**
237  * blk_set_pm_only - increment pm_only counter
238  * @q: request queue pointer
239  */
240 void blk_set_pm_only(struct request_queue *q)
241 {
242 	atomic_inc(&q->pm_only);
243 }
244 EXPORT_SYMBOL_GPL(blk_set_pm_only);
245 
246 void blk_clear_pm_only(struct request_queue *q)
247 {
248 	int pm_only;
249 
250 	pm_only = atomic_dec_return(&q->pm_only);
251 	WARN_ON_ONCE(pm_only < 0);
252 	if (pm_only == 0)
253 		wake_up_all(&q->mq_freeze_wq);
254 }
255 EXPORT_SYMBOL_GPL(blk_clear_pm_only);
256 
257 /**
258  * blk_put_queue - decrement the request_queue refcount
259  * @q: the request_queue structure to decrement the refcount for
260  *
261  * Decrements the refcount of the request_queue kobject. When this reaches 0
262  * we'll have blk_release_queue() called.
263  *
264  * Context: Any context, but the last reference must not be dropped from
265  *          atomic context.
266  */
267 void blk_put_queue(struct request_queue *q)
268 {
269 	kobject_put(&q->kobj);
270 }
271 EXPORT_SYMBOL(blk_put_queue);
272 
273 void blk_queue_start_drain(struct request_queue *q)
274 {
275 	/*
276 	 * When queue DYING flag is set, we need to block new req
277 	 * entering queue, so we call blk_freeze_queue_start() to
278 	 * prevent I/O from crossing blk_queue_enter().
279 	 */
280 	blk_freeze_queue_start(q);
281 	if (queue_is_mq(q))
282 		blk_mq_wake_waiters(q);
283 	/* Make blk_queue_enter() reexamine the DYING flag. */
284 	wake_up_all(&q->mq_freeze_wq);
285 }
286 
287 /**
288  * blk_queue_enter() - try to increase q->q_usage_counter
289  * @q: request queue pointer
290  * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
291  */
292 int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
293 {
294 	const bool pm = flags & BLK_MQ_REQ_PM;
295 
296 	while (!blk_try_enter_queue(q, pm)) {
297 		if (flags & BLK_MQ_REQ_NOWAIT)
298 			return -EBUSY;
299 
300 		/*
301 		 * read pair of barrier in blk_freeze_queue_start(), we need to
302 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
303 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
304 		 * following wait may never return if the two reads are
305 		 * reordered.
306 		 */
307 		smp_rmb();
308 		wait_event(q->mq_freeze_wq,
309 			   (!q->mq_freeze_depth &&
310 			    blk_pm_resume_queue(pm, q)) ||
311 			   blk_queue_dying(q));
312 		if (blk_queue_dying(q))
313 			return -ENODEV;
314 	}
315 
316 	return 0;
317 }
318 
319 int __bio_queue_enter(struct request_queue *q, struct bio *bio)
320 {
321 	while (!blk_try_enter_queue(q, false)) {
322 		struct gendisk *disk = bio->bi_bdev->bd_disk;
323 
324 		if (bio->bi_opf & REQ_NOWAIT) {
325 			if (test_bit(GD_DEAD, &disk->state))
326 				goto dead;
327 			bio_wouldblock_error(bio);
328 			return -EBUSY;
329 		}
330 
331 		/*
332 		 * read pair of barrier in blk_freeze_queue_start(), we need to
333 		 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
334 		 * reading .mq_freeze_depth or queue dying flag, otherwise the
335 		 * following wait may never return if the two reads are
336 		 * reordered.
337 		 */
338 		smp_rmb();
339 		wait_event(q->mq_freeze_wq,
340 			   (!q->mq_freeze_depth &&
341 			    blk_pm_resume_queue(false, q)) ||
342 			   test_bit(GD_DEAD, &disk->state));
343 		if (test_bit(GD_DEAD, &disk->state))
344 			goto dead;
345 	}
346 
347 	return 0;
348 dead:
349 	bio_io_error(bio);
350 	return -ENODEV;
351 }
352 
353 void blk_queue_exit(struct request_queue *q)
354 {
355 	percpu_ref_put(&q->q_usage_counter);
356 }
357 
358 static void blk_queue_usage_counter_release(struct percpu_ref *ref)
359 {
360 	struct request_queue *q =
361 		container_of(ref, struct request_queue, q_usage_counter);
362 
363 	wake_up_all(&q->mq_freeze_wq);
364 }
365 
366 static void blk_rq_timed_out_timer(struct timer_list *t)
367 {
368 	struct request_queue *q = from_timer(q, t, timeout);
369 
370 	kblockd_schedule_work(&q->timeout_work);
371 }
372 
373 static void blk_timeout_work(struct work_struct *work)
374 {
375 }
376 
377 struct request_queue *blk_alloc_queue(int node_id, bool alloc_srcu)
378 {
379 	struct request_queue *q;
380 	int ret;
381 
382 	q = kmem_cache_alloc_node(blk_get_queue_kmem_cache(alloc_srcu),
383 			GFP_KERNEL | __GFP_ZERO, node_id);
384 	if (!q)
385 		return NULL;
386 
387 	if (alloc_srcu) {
388 		blk_queue_flag_set(QUEUE_FLAG_HAS_SRCU, q);
389 		if (init_srcu_struct(q->srcu) != 0)
390 			goto fail_q;
391 	}
392 
393 	q->last_merge = NULL;
394 
395 	q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL);
396 	if (q->id < 0)
397 		goto fail_srcu;
398 
399 	ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, 0);
400 	if (ret)
401 		goto fail_id;
402 
403 	q->stats = blk_alloc_queue_stats();
404 	if (!q->stats)
405 		goto fail_split;
406 
407 	q->node = node_id;
408 
409 	atomic_set(&q->nr_active_requests_shared_tags, 0);
410 
411 	timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
412 	INIT_WORK(&q->timeout_work, blk_timeout_work);
413 	INIT_LIST_HEAD(&q->icq_list);
414 
415 	kobject_init(&q->kobj, &blk_queue_ktype);
416 
417 	mutex_init(&q->debugfs_mutex);
418 	mutex_init(&q->sysfs_lock);
419 	mutex_init(&q->sysfs_dir_lock);
420 	spin_lock_init(&q->queue_lock);
421 
422 	init_waitqueue_head(&q->mq_freeze_wq);
423 	mutex_init(&q->mq_freeze_lock);
424 
425 	/*
426 	 * Init percpu_ref in atomic mode so that it's faster to shutdown.
427 	 * See blk_register_queue() for details.
428 	 */
429 	if (percpu_ref_init(&q->q_usage_counter,
430 				blk_queue_usage_counter_release,
431 				PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
432 		goto fail_stats;
433 
434 	blk_queue_dma_alignment(q, 511);
435 	blk_set_default_limits(&q->limits);
436 	q->nr_requests = BLKDEV_DEFAULT_RQ;
437 
438 	return q;
439 
440 fail_stats:
441 	blk_free_queue_stats(q->stats);
442 fail_split:
443 	bioset_exit(&q->bio_split);
444 fail_id:
445 	ida_free(&blk_queue_ida, q->id);
446 fail_srcu:
447 	if (alloc_srcu)
448 		cleanup_srcu_struct(q->srcu);
449 fail_q:
450 	kmem_cache_free(blk_get_queue_kmem_cache(alloc_srcu), q);
451 	return NULL;
452 }
453 
454 /**
455  * blk_get_queue - increment the request_queue refcount
456  * @q: the request_queue structure to increment the refcount for
457  *
458  * Increment the refcount of the request_queue kobject.
459  *
460  * Context: Any context.
461  */
462 bool blk_get_queue(struct request_queue *q)
463 {
464 	if (likely(!blk_queue_dying(q))) {
465 		__blk_get_queue(q);
466 		return true;
467 	}
468 
469 	return false;
470 }
471 EXPORT_SYMBOL(blk_get_queue);
472 
473 #ifdef CONFIG_FAIL_MAKE_REQUEST
474 
475 static DECLARE_FAULT_ATTR(fail_make_request);
476 
477 static int __init setup_fail_make_request(char *str)
478 {
479 	return setup_fault_attr(&fail_make_request, str);
480 }
481 __setup("fail_make_request=", setup_fail_make_request);
482 
483 bool should_fail_request(struct block_device *part, unsigned int bytes)
484 {
485 	return part->bd_make_it_fail && should_fail(&fail_make_request, bytes);
486 }
487 
488 static int __init fail_make_request_debugfs(void)
489 {
490 	struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
491 						NULL, &fail_make_request);
492 
493 	return PTR_ERR_OR_ZERO(dir);
494 }
495 
496 late_initcall(fail_make_request_debugfs);
497 #endif /* CONFIG_FAIL_MAKE_REQUEST */
498 
499 static inline bool bio_check_ro(struct bio *bio)
500 {
501 	if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
502 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
503 			return false;
504 		pr_warn("Trying to write to read-only block-device %pg\n",
505 			bio->bi_bdev);
506 		/* Older lvm-tools actually trigger this */
507 		return false;
508 	}
509 
510 	return false;
511 }
512 
513 static noinline int should_fail_bio(struct bio *bio)
514 {
515 	if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
516 		return -EIO;
517 	return 0;
518 }
519 ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
520 
521 /*
522  * Check whether this bio extends beyond the end of the device or partition.
523  * This may well happen - the kernel calls bread() without checking the size of
524  * the device, e.g., when mounting a file system.
525  */
526 static inline int bio_check_eod(struct bio *bio)
527 {
528 	sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
529 	unsigned int nr_sectors = bio_sectors(bio);
530 
531 	if (nr_sectors && maxsector &&
532 	    (nr_sectors > maxsector ||
533 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
534 		pr_info_ratelimited("%s: attempt to access beyond end of device\n"
535 				    "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
536 				    current->comm, bio->bi_bdev, bio->bi_opf,
537 				    bio->bi_iter.bi_sector, nr_sectors, maxsector);
538 		return -EIO;
539 	}
540 	return 0;
541 }
542 
543 /*
544  * Remap block n of partition p to block n+start(p) of the disk.
545  */
546 static int blk_partition_remap(struct bio *bio)
547 {
548 	struct block_device *p = bio->bi_bdev;
549 
550 	if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
551 		return -EIO;
552 	if (bio_sectors(bio)) {
553 		bio->bi_iter.bi_sector += p->bd_start_sect;
554 		trace_block_bio_remap(bio, p->bd_dev,
555 				      bio->bi_iter.bi_sector -
556 				      p->bd_start_sect);
557 	}
558 	bio_set_flag(bio, BIO_REMAPPED);
559 	return 0;
560 }
561 
562 /*
563  * Check write append to a zoned block device.
564  */
565 static inline blk_status_t blk_check_zone_append(struct request_queue *q,
566 						 struct bio *bio)
567 {
568 	sector_t pos = bio->bi_iter.bi_sector;
569 	int nr_sectors = bio_sectors(bio);
570 
571 	/* Only applicable to zoned block devices */
572 	if (!blk_queue_is_zoned(q))
573 		return BLK_STS_NOTSUPP;
574 
575 	/* The bio sector must point to the start of a sequential zone */
576 	if (pos & (blk_queue_zone_sectors(q) - 1) ||
577 	    !blk_queue_zone_is_seq(q, pos))
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(q, 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) && !blk_queue_nowait(q))
731 		goto not_supported;
732 
733 	if (should_fail_bio(bio))
734 		goto end_io;
735 	if (unlikely(bio_check_ro(bio)))
736 		goto end_io;
737 	if (!bio_flagged(bio, BIO_REMAPPED)) {
738 		if (unlikely(bio_check_eod(bio)))
739 			goto end_io;
740 		if (bdev->bd_partno && unlikely(blk_partition_remap(bio)))
741 			goto end_io;
742 	}
743 
744 	/*
745 	 * Filter flush bio's early so that bio based drivers without flush
746 	 * support don't have to worry about them.
747 	 */
748 	if (op_is_flush(bio->bi_opf) &&
749 	    !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
750 		bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
751 		if (!bio_sectors(bio)) {
752 			status = BLK_STS_OK;
753 			goto end_io;
754 		}
755 	}
756 
757 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
758 		bio_clear_polled(bio);
759 
760 	switch (bio_op(bio)) {
761 	case REQ_OP_DISCARD:
762 		if (!bdev_max_discard_sectors(bdev))
763 			goto not_supported;
764 		break;
765 	case REQ_OP_SECURE_ERASE:
766 		if (!bdev_max_secure_erase_sectors(bdev))
767 			goto not_supported;
768 		break;
769 	case REQ_OP_ZONE_APPEND:
770 		status = blk_check_zone_append(q, bio);
771 		if (status != BLK_STS_OK)
772 			goto end_io;
773 		break;
774 	case REQ_OP_ZONE_RESET:
775 	case REQ_OP_ZONE_OPEN:
776 	case REQ_OP_ZONE_CLOSE:
777 	case REQ_OP_ZONE_FINISH:
778 		if (!blk_queue_is_zoned(q))
779 			goto not_supported;
780 		break;
781 	case REQ_OP_ZONE_RESET_ALL:
782 		if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
783 			goto not_supported;
784 		break;
785 	case REQ_OP_WRITE_ZEROES:
786 		if (!q->limits.max_write_zeroes_sectors)
787 			goto not_supported;
788 		break;
789 	default:
790 		break;
791 	}
792 
793 	if (blk_throtl_bio(bio))
794 		return;
795 
796 	blk_cgroup_bio_start(bio);
797 	blkcg_bio_issue_init(bio);
798 
799 	if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
800 		trace_block_bio_queue(bio);
801 		/* Now that enqueuing has been traced, we need to trace
802 		 * completion as well.
803 		 */
804 		bio_set_flag(bio, BIO_TRACE_COMPLETION);
805 	}
806 	submit_bio_noacct_nocheck(bio);
807 	return;
808 
809 not_supported:
810 	status = BLK_STS_NOTSUPP;
811 end_io:
812 	bio->bi_status = status;
813 	bio_endio(bio);
814 }
815 EXPORT_SYMBOL(submit_bio_noacct);
816 
817 /**
818  * submit_bio - submit a bio to the block device layer for I/O
819  * @bio: The &struct bio which describes the I/O
820  *
821  * submit_bio() is used to submit I/O requests to block devices.  It is passed a
822  * fully set up &struct bio that describes the I/O that needs to be done.  The
823  * bio will be send to the device described by the bi_bdev field.
824  *
825  * The success/failure status of the request, along with notification of
826  * completion, is delivered asynchronously through the ->bi_end_io() callback
827  * in @bio.  The bio must NOT be touched by thecaller until ->bi_end_io() has
828  * been called.
829  */
830 void submit_bio(struct bio *bio)
831 {
832 	if (blkcg_punt_bio_submit(bio))
833 		return;
834 
835 	if (bio_op(bio) == REQ_OP_READ) {
836 		task_io_account_read(bio->bi_iter.bi_size);
837 		count_vm_events(PGPGIN, bio_sectors(bio));
838 	} else if (bio_op(bio) == REQ_OP_WRITE) {
839 		count_vm_events(PGPGOUT, bio_sectors(bio));
840 	}
841 
842 	/*
843 	 * If we're reading data that is part of the userspace workingset, count
844 	 * submission time as memory stall.  When the device is congested, or
845 	 * the submitting cgroup IO-throttled, submission can be a significant
846 	 * part of overall IO time.
847 	 */
848 	if (unlikely(bio_op(bio) == REQ_OP_READ &&
849 	    bio_flagged(bio, BIO_WORKINGSET))) {
850 		unsigned long pflags;
851 
852 		psi_memstall_enter(&pflags);
853 		submit_bio_noacct(bio);
854 		psi_memstall_leave(&pflags);
855 		return;
856 	}
857 
858 	submit_bio_noacct(bio);
859 }
860 EXPORT_SYMBOL(submit_bio);
861 
862 /**
863  * bio_poll - poll for BIO completions
864  * @bio: bio to poll for
865  * @iob: batches of IO
866  * @flags: BLK_POLL_* flags that control the behavior
867  *
868  * Poll for completions on queue associated with the bio. Returns number of
869  * completed entries found.
870  *
871  * Note: the caller must either be the context that submitted @bio, or
872  * be in a RCU critical section to prevent freeing of @bio.
873  */
874 int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
875 {
876 	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
877 	blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
878 	int ret = 0;
879 
880 	if (cookie == BLK_QC_T_NONE ||
881 	    !test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
882 		return 0;
883 
884 	blk_flush_plug(current->plug, false);
885 
886 	if (bio_queue_enter(bio))
887 		return 0;
888 	if (queue_is_mq(q)) {
889 		ret = blk_mq_poll(q, cookie, iob, flags);
890 	} else {
891 		struct gendisk *disk = q->disk;
892 
893 		if (disk && disk->fops->poll_bio)
894 			ret = disk->fops->poll_bio(bio, iob, flags);
895 	}
896 	blk_queue_exit(q);
897 	return ret;
898 }
899 EXPORT_SYMBOL_GPL(bio_poll);
900 
901 /*
902  * Helper to implement file_operations.iopoll.  Requires the bio to be stored
903  * in iocb->private, and cleared before freeing the bio.
904  */
905 int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
906 		    unsigned int flags)
907 {
908 	struct bio *bio;
909 	int ret = 0;
910 
911 	/*
912 	 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
913 	 * point to a freshly allocated bio at this point.  If that happens
914 	 * we have a few cases to consider:
915 	 *
916 	 *  1) the bio is beeing initialized and bi_bdev is NULL.  We can just
917 	 *     simply nothing in this case
918 	 *  2) the bio points to a not poll enabled device.  bio_poll will catch
919 	 *     this and return 0
920 	 *  3) the bio points to a poll capable device, including but not
921 	 *     limited to the one that the original bio pointed to.  In this
922 	 *     case we will call into the actual poll method and poll for I/O,
923 	 *     even if we don't need to, but it won't cause harm either.
924 	 *
925 	 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
926 	 * is still allocated. Because partitions hold a reference to the whole
927 	 * device bdev and thus disk, the disk is also still valid.  Grabbing
928 	 * a reference to the queue in bio_poll() ensures the hctxs and requests
929 	 * are still valid as well.
930 	 */
931 	rcu_read_lock();
932 	bio = READ_ONCE(kiocb->private);
933 	if (bio && bio->bi_bdev)
934 		ret = bio_poll(bio, iob, flags);
935 	rcu_read_unlock();
936 
937 	return ret;
938 }
939 EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
940 
941 void update_io_ticks(struct block_device *part, unsigned long now, bool end)
942 {
943 	unsigned long stamp;
944 again:
945 	stamp = READ_ONCE(part->bd_stamp);
946 	if (unlikely(time_after(now, stamp))) {
947 		if (likely(cmpxchg(&part->bd_stamp, stamp, now) == stamp))
948 			__part_stat_add(part, io_ticks, end ? now - stamp : 1);
949 	}
950 	if (part->bd_partno) {
951 		part = bdev_whole(part);
952 		goto again;
953 	}
954 }
955 
956 unsigned long bdev_start_io_acct(struct block_device *bdev,
957 				 unsigned int sectors, unsigned int op,
958 				 unsigned long start_time)
959 {
960 	const int sgrp = op_stat_group(op);
961 
962 	part_stat_lock();
963 	update_io_ticks(bdev, start_time, false);
964 	part_stat_inc(bdev, ios[sgrp]);
965 	part_stat_add(bdev, sectors[sgrp], sectors);
966 	part_stat_local_inc(bdev, in_flight[op_is_write(op)]);
967 	part_stat_unlock();
968 
969 	return start_time;
970 }
971 EXPORT_SYMBOL(bdev_start_io_acct);
972 
973 /**
974  * bio_start_io_acct_time - start I/O accounting for bio based drivers
975  * @bio:	bio to start account for
976  * @start_time:	start time that should be passed back to bio_end_io_acct().
977  */
978 void bio_start_io_acct_time(struct bio *bio, unsigned long start_time)
979 {
980 	bdev_start_io_acct(bio->bi_bdev, bio_sectors(bio),
981 			   bio_op(bio), start_time);
982 }
983 EXPORT_SYMBOL_GPL(bio_start_io_acct_time);
984 
985 /**
986  * bio_start_io_acct - start I/O accounting for bio based drivers
987  * @bio:	bio to start account for
988  *
989  * Returns the start time that should be passed back to bio_end_io_acct().
990  */
991 unsigned long bio_start_io_acct(struct bio *bio)
992 {
993 	return bdev_start_io_acct(bio->bi_bdev, bio_sectors(bio),
994 				  bio_op(bio), jiffies);
995 }
996 EXPORT_SYMBOL_GPL(bio_start_io_acct);
997 
998 void bdev_end_io_acct(struct block_device *bdev, unsigned int op,
999 		      unsigned long start_time)
1000 {
1001 	const int sgrp = op_stat_group(op);
1002 	unsigned long now = READ_ONCE(jiffies);
1003 	unsigned long duration = now - start_time;
1004 
1005 	part_stat_lock();
1006 	update_io_ticks(bdev, now, true);
1007 	part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration));
1008 	part_stat_local_dec(bdev, in_flight[op_is_write(op)]);
1009 	part_stat_unlock();
1010 }
1011 EXPORT_SYMBOL(bdev_end_io_acct);
1012 
1013 void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
1014 			      struct block_device *orig_bdev)
1015 {
1016 	bdev_end_io_acct(orig_bdev, bio_op(bio), start_time);
1017 }
1018 EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
1019 
1020 /**
1021  * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1022  * @q : the queue of the device being checked
1023  *
1024  * Description:
1025  *    Check if underlying low-level drivers of a device are busy.
1026  *    If the drivers want to export their busy state, they must set own
1027  *    exporting function using blk_queue_lld_busy() first.
1028  *
1029  *    Basically, this function is used only by request stacking drivers
1030  *    to stop dispatching requests to underlying devices when underlying
1031  *    devices are busy.  This behavior helps more I/O merging on the queue
1032  *    of the request stacking driver and prevents I/O throughput regression
1033  *    on burst I/O load.
1034  *
1035  * Return:
1036  *    0 - Not busy (The request stacking driver should dispatch request)
1037  *    1 - Busy (The request stacking driver should stop dispatching request)
1038  */
1039 int blk_lld_busy(struct request_queue *q)
1040 {
1041 	if (queue_is_mq(q) && q->mq_ops->busy)
1042 		return q->mq_ops->busy(q);
1043 
1044 	return 0;
1045 }
1046 EXPORT_SYMBOL_GPL(blk_lld_busy);
1047 
1048 int kblockd_schedule_work(struct work_struct *work)
1049 {
1050 	return queue_work(kblockd_workqueue, work);
1051 }
1052 EXPORT_SYMBOL(kblockd_schedule_work);
1053 
1054 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1055 				unsigned long delay)
1056 {
1057 	return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1058 }
1059 EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1060 
1061 void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1062 {
1063 	struct task_struct *tsk = current;
1064 
1065 	/*
1066 	 * If this is a nested plug, don't actually assign it.
1067 	 */
1068 	if (tsk->plug)
1069 		return;
1070 
1071 	plug->mq_list = NULL;
1072 	plug->cached_rq = NULL;
1073 	plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1074 	plug->rq_count = 0;
1075 	plug->multiple_queues = false;
1076 	plug->has_elevator = false;
1077 	plug->nowait = false;
1078 	INIT_LIST_HEAD(&plug->cb_list);
1079 
1080 	/*
1081 	 * Store ordering should not be needed here, since a potential
1082 	 * preempt will imply a full memory barrier
1083 	 */
1084 	tsk->plug = plug;
1085 }
1086 
1087 /**
1088  * blk_start_plug - initialize blk_plug and track it inside the task_struct
1089  * @plug:	The &struct blk_plug that needs to be initialized
1090  *
1091  * Description:
1092  *   blk_start_plug() indicates to the block layer an intent by the caller
1093  *   to submit multiple I/O requests in a batch.  The block layer may use
1094  *   this hint to defer submitting I/Os from the caller until blk_finish_plug()
1095  *   is called.  However, the block layer may choose to submit requests
1096  *   before a call to blk_finish_plug() if the number of queued I/Os
1097  *   exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1098  *   %BLK_PLUG_FLUSH_SIZE.  The queued I/Os may also be submitted early if
1099  *   the task schedules (see below).
1100  *
1101  *   Tracking blk_plug inside the task_struct will help with auto-flushing the
1102  *   pending I/O should the task end up blocking between blk_start_plug() and
1103  *   blk_finish_plug(). This is important from a performance perspective, but
1104  *   also ensures that we don't deadlock. For instance, if the task is blocking
1105  *   for a memory allocation, memory reclaim could end up wanting to free a
1106  *   page belonging to that request that is currently residing in our private
1107  *   plug. By flushing the pending I/O when the process goes to sleep, we avoid
1108  *   this kind of deadlock.
1109  */
1110 void blk_start_plug(struct blk_plug *plug)
1111 {
1112 	blk_start_plug_nr_ios(plug, 1);
1113 }
1114 EXPORT_SYMBOL(blk_start_plug);
1115 
1116 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
1117 {
1118 	LIST_HEAD(callbacks);
1119 
1120 	while (!list_empty(&plug->cb_list)) {
1121 		list_splice_init(&plug->cb_list, &callbacks);
1122 
1123 		while (!list_empty(&callbacks)) {
1124 			struct blk_plug_cb *cb = list_first_entry(&callbacks,
1125 							  struct blk_plug_cb,
1126 							  list);
1127 			list_del(&cb->list);
1128 			cb->callback(cb, from_schedule);
1129 		}
1130 	}
1131 }
1132 
1133 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1134 				      int size)
1135 {
1136 	struct blk_plug *plug = current->plug;
1137 	struct blk_plug_cb *cb;
1138 
1139 	if (!plug)
1140 		return NULL;
1141 
1142 	list_for_each_entry(cb, &plug->cb_list, list)
1143 		if (cb->callback == unplug && cb->data == data)
1144 			return cb;
1145 
1146 	/* Not currently on the callback list */
1147 	BUG_ON(size < sizeof(*cb));
1148 	cb = kzalloc(size, GFP_ATOMIC);
1149 	if (cb) {
1150 		cb->data = data;
1151 		cb->callback = unplug;
1152 		list_add(&cb->list, &plug->cb_list);
1153 	}
1154 	return cb;
1155 }
1156 EXPORT_SYMBOL(blk_check_plugged);
1157 
1158 void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
1159 {
1160 	if (!list_empty(&plug->cb_list))
1161 		flush_plug_callbacks(plug, from_schedule);
1162 	if (!rq_list_empty(plug->mq_list))
1163 		blk_mq_flush_plug_list(plug, from_schedule);
1164 	/*
1165 	 * Unconditionally flush out cached requests, even if the unplug
1166 	 * event came from schedule. Since we know hold references to the
1167 	 * queue for cached requests, we don't want a blocked task holding
1168 	 * up a queue freeze/quiesce event.
1169 	 */
1170 	if (unlikely(!rq_list_empty(plug->cached_rq)))
1171 		blk_mq_free_plug_rqs(plug);
1172 }
1173 
1174 /**
1175  * blk_finish_plug - mark the end of a batch of submitted I/O
1176  * @plug:	The &struct blk_plug passed to blk_start_plug()
1177  *
1178  * Description:
1179  * Indicate that a batch of I/O submissions is complete.  This function
1180  * must be paired with an initial call to blk_start_plug().  The intent
1181  * is to allow the block layer to optimize I/O submission.  See the
1182  * documentation for blk_start_plug() for more information.
1183  */
1184 void blk_finish_plug(struct blk_plug *plug)
1185 {
1186 	if (plug == current->plug) {
1187 		__blk_flush_plug(plug, false);
1188 		current->plug = NULL;
1189 	}
1190 }
1191 EXPORT_SYMBOL(blk_finish_plug);
1192 
1193 void blk_io_schedule(void)
1194 {
1195 	/* Prevent hang_check timer from firing at us during very long I/O */
1196 	unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1197 
1198 	if (timeout)
1199 		io_schedule_timeout(timeout);
1200 	else
1201 		io_schedule();
1202 }
1203 EXPORT_SYMBOL_GPL(blk_io_schedule);
1204 
1205 int __init blk_dev_init(void)
1206 {
1207 	BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1208 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1209 			sizeof_field(struct request, cmd_flags));
1210 	BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1211 			sizeof_field(struct bio, bi_opf));
1212 	BUILD_BUG_ON(ALIGN(offsetof(struct request_queue, srcu),
1213 			   __alignof__(struct request_queue)) !=
1214 		     sizeof(struct request_queue));
1215 
1216 	/* used for unplugging and affects IO latency/throughput - HIGHPRI */
1217 	kblockd_workqueue = alloc_workqueue("kblockd",
1218 					    WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
1219 	if (!kblockd_workqueue)
1220 		panic("Failed to create kblockd\n");
1221 
1222 	blk_requestq_cachep = kmem_cache_create("request_queue",
1223 			sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1224 
1225 	blk_requestq_srcu_cachep = kmem_cache_create("request_queue_srcu",
1226 			sizeof(struct request_queue) +
1227 			sizeof(struct srcu_struct), 0, SLAB_PANIC, NULL);
1228 
1229 	blk_debugfs_root = debugfs_create_dir("block", NULL);
1230 
1231 	return 0;
1232 }
1233