xref: /openbmc/linux/drivers/md/dm-table.c (revision 82003e04)
1 /*
2  * Copyright (C) 2001 Sistina Software (UK) Limited.
3  * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4  *
5  * This file is released under the GPL.
6  */
7 
8 #include "dm-core.h"
9 
10 #include <linux/module.h>
11 #include <linux/vmalloc.h>
12 #include <linux/blkdev.h>
13 #include <linux/namei.h>
14 #include <linux/ctype.h>
15 #include <linux/string.h>
16 #include <linux/slab.h>
17 #include <linux/interrupt.h>
18 #include <linux/mutex.h>
19 #include <linux/delay.h>
20 #include <linux/atomic.h>
21 #include <linux/blk-mq.h>
22 #include <linux/mount.h>
23 
24 #define DM_MSG_PREFIX "table"
25 
26 #define MAX_DEPTH 16
27 #define NODE_SIZE L1_CACHE_BYTES
28 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
29 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
30 
31 struct dm_table {
32 	struct mapped_device *md;
33 	unsigned type;
34 
35 	/* btree table */
36 	unsigned int depth;
37 	unsigned int counts[MAX_DEPTH];	/* in nodes */
38 	sector_t *index[MAX_DEPTH];
39 
40 	unsigned int num_targets;
41 	unsigned int num_allocated;
42 	sector_t *highs;
43 	struct dm_target *targets;
44 
45 	struct target_type *immutable_target_type;
46 
47 	bool integrity_supported:1;
48 	bool singleton:1;
49 	bool all_blk_mq:1;
50 
51 	/*
52 	 * Indicates the rw permissions for the new logical
53 	 * device.  This should be a combination of FMODE_READ
54 	 * and FMODE_WRITE.
55 	 */
56 	fmode_t mode;
57 
58 	/* a list of devices used by this table */
59 	struct list_head devices;
60 
61 	/* events get handed up using this callback */
62 	void (*event_fn)(void *);
63 	void *event_context;
64 
65 	struct dm_md_mempools *mempools;
66 
67 	struct list_head target_callbacks;
68 };
69 
70 /*
71  * Similar to ceiling(log_size(n))
72  */
73 static unsigned int int_log(unsigned int n, unsigned int base)
74 {
75 	int result = 0;
76 
77 	while (n > 1) {
78 		n = dm_div_up(n, base);
79 		result++;
80 	}
81 
82 	return result;
83 }
84 
85 /*
86  * Calculate the index of the child node of the n'th node k'th key.
87  */
88 static inline unsigned int get_child(unsigned int n, unsigned int k)
89 {
90 	return (n * CHILDREN_PER_NODE) + k;
91 }
92 
93 /*
94  * Return the n'th node of level l from table t.
95  */
96 static inline sector_t *get_node(struct dm_table *t,
97 				 unsigned int l, unsigned int n)
98 {
99 	return t->index[l] + (n * KEYS_PER_NODE);
100 }
101 
102 /*
103  * Return the highest key that you could lookup from the n'th
104  * node on level l of the btree.
105  */
106 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
107 {
108 	for (; l < t->depth - 1; l++)
109 		n = get_child(n, CHILDREN_PER_NODE - 1);
110 
111 	if (n >= t->counts[l])
112 		return (sector_t) - 1;
113 
114 	return get_node(t, l, n)[KEYS_PER_NODE - 1];
115 }
116 
117 /*
118  * Fills in a level of the btree based on the highs of the level
119  * below it.
120  */
121 static int setup_btree_index(unsigned int l, struct dm_table *t)
122 {
123 	unsigned int n, k;
124 	sector_t *node;
125 
126 	for (n = 0U; n < t->counts[l]; n++) {
127 		node = get_node(t, l, n);
128 
129 		for (k = 0U; k < KEYS_PER_NODE; k++)
130 			node[k] = high(t, l + 1, get_child(n, k));
131 	}
132 
133 	return 0;
134 }
135 
136 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
137 {
138 	unsigned long size;
139 	void *addr;
140 
141 	/*
142 	 * Check that we're not going to overflow.
143 	 */
144 	if (nmemb > (ULONG_MAX / elem_size))
145 		return NULL;
146 
147 	size = nmemb * elem_size;
148 	addr = vzalloc(size);
149 
150 	return addr;
151 }
152 EXPORT_SYMBOL(dm_vcalloc);
153 
154 /*
155  * highs, and targets are managed as dynamic arrays during a
156  * table load.
157  */
158 static int alloc_targets(struct dm_table *t, unsigned int num)
159 {
160 	sector_t *n_highs;
161 	struct dm_target *n_targets;
162 
163 	/*
164 	 * Allocate both the target array and offset array at once.
165 	 * Append an empty entry to catch sectors beyond the end of
166 	 * the device.
167 	 */
168 	n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
169 					  sizeof(sector_t));
170 	if (!n_highs)
171 		return -ENOMEM;
172 
173 	n_targets = (struct dm_target *) (n_highs + num);
174 
175 	memset(n_highs, -1, sizeof(*n_highs) * num);
176 	vfree(t->highs);
177 
178 	t->num_allocated = num;
179 	t->highs = n_highs;
180 	t->targets = n_targets;
181 
182 	return 0;
183 }
184 
185 int dm_table_create(struct dm_table **result, fmode_t mode,
186 		    unsigned num_targets, struct mapped_device *md)
187 {
188 	struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
189 
190 	if (!t)
191 		return -ENOMEM;
192 
193 	INIT_LIST_HEAD(&t->devices);
194 	INIT_LIST_HEAD(&t->target_callbacks);
195 
196 	if (!num_targets)
197 		num_targets = KEYS_PER_NODE;
198 
199 	num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
200 
201 	if (!num_targets) {
202 		kfree(t);
203 		return -ENOMEM;
204 	}
205 
206 	if (alloc_targets(t, num_targets)) {
207 		kfree(t);
208 		return -ENOMEM;
209 	}
210 
211 	t->type = DM_TYPE_NONE;
212 	t->mode = mode;
213 	t->md = md;
214 	*result = t;
215 	return 0;
216 }
217 
218 static void free_devices(struct list_head *devices, struct mapped_device *md)
219 {
220 	struct list_head *tmp, *next;
221 
222 	list_for_each_safe(tmp, next, devices) {
223 		struct dm_dev_internal *dd =
224 		    list_entry(tmp, struct dm_dev_internal, list);
225 		DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
226 		       dm_device_name(md), dd->dm_dev->name);
227 		dm_put_table_device(md, dd->dm_dev);
228 		kfree(dd);
229 	}
230 }
231 
232 void dm_table_destroy(struct dm_table *t)
233 {
234 	unsigned int i;
235 
236 	if (!t)
237 		return;
238 
239 	/* free the indexes */
240 	if (t->depth >= 2)
241 		vfree(t->index[t->depth - 2]);
242 
243 	/* free the targets */
244 	for (i = 0; i < t->num_targets; i++) {
245 		struct dm_target *tgt = t->targets + i;
246 
247 		if (tgt->type->dtr)
248 			tgt->type->dtr(tgt);
249 
250 		dm_put_target_type(tgt->type);
251 	}
252 
253 	vfree(t->highs);
254 
255 	/* free the device list */
256 	free_devices(&t->devices, t->md);
257 
258 	dm_free_md_mempools(t->mempools);
259 
260 	kfree(t);
261 }
262 
263 /*
264  * See if we've already got a device in the list.
265  */
266 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
267 {
268 	struct dm_dev_internal *dd;
269 
270 	list_for_each_entry (dd, l, list)
271 		if (dd->dm_dev->bdev->bd_dev == dev)
272 			return dd;
273 
274 	return NULL;
275 }
276 
277 /*
278  * If possible, this checks an area of a destination device is invalid.
279  */
280 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
281 				  sector_t start, sector_t len, void *data)
282 {
283 	struct request_queue *q;
284 	struct queue_limits *limits = data;
285 	struct block_device *bdev = dev->bdev;
286 	sector_t dev_size =
287 		i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
288 	unsigned short logical_block_size_sectors =
289 		limits->logical_block_size >> SECTOR_SHIFT;
290 	char b[BDEVNAME_SIZE];
291 
292 	/*
293 	 * Some devices exist without request functions,
294 	 * such as loop devices not yet bound to backing files.
295 	 * Forbid the use of such devices.
296 	 */
297 	q = bdev_get_queue(bdev);
298 	if (!q || !q->make_request_fn) {
299 		DMWARN("%s: %s is not yet initialised: "
300 		       "start=%llu, len=%llu, dev_size=%llu",
301 		       dm_device_name(ti->table->md), bdevname(bdev, b),
302 		       (unsigned long long)start,
303 		       (unsigned long long)len,
304 		       (unsigned long long)dev_size);
305 		return 1;
306 	}
307 
308 	if (!dev_size)
309 		return 0;
310 
311 	if ((start >= dev_size) || (start + len > dev_size)) {
312 		DMWARN("%s: %s too small for target: "
313 		       "start=%llu, len=%llu, dev_size=%llu",
314 		       dm_device_name(ti->table->md), bdevname(bdev, b),
315 		       (unsigned long long)start,
316 		       (unsigned long long)len,
317 		       (unsigned long long)dev_size);
318 		return 1;
319 	}
320 
321 	if (logical_block_size_sectors <= 1)
322 		return 0;
323 
324 	if (start & (logical_block_size_sectors - 1)) {
325 		DMWARN("%s: start=%llu not aligned to h/w "
326 		       "logical block size %u of %s",
327 		       dm_device_name(ti->table->md),
328 		       (unsigned long long)start,
329 		       limits->logical_block_size, bdevname(bdev, b));
330 		return 1;
331 	}
332 
333 	if (len & (logical_block_size_sectors - 1)) {
334 		DMWARN("%s: len=%llu not aligned to h/w "
335 		       "logical block size %u of %s",
336 		       dm_device_name(ti->table->md),
337 		       (unsigned long long)len,
338 		       limits->logical_block_size, bdevname(bdev, b));
339 		return 1;
340 	}
341 
342 	return 0;
343 }
344 
345 /*
346  * This upgrades the mode on an already open dm_dev, being
347  * careful to leave things as they were if we fail to reopen the
348  * device and not to touch the existing bdev field in case
349  * it is accessed concurrently inside dm_table_any_congested().
350  */
351 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
352 			struct mapped_device *md)
353 {
354 	int r;
355 	struct dm_dev *old_dev, *new_dev;
356 
357 	old_dev = dd->dm_dev;
358 
359 	r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
360 				dd->dm_dev->mode | new_mode, &new_dev);
361 	if (r)
362 		return r;
363 
364 	dd->dm_dev = new_dev;
365 	dm_put_table_device(md, old_dev);
366 
367 	return 0;
368 }
369 
370 /*
371  * Convert the path to a device
372  */
373 dev_t dm_get_dev_t(const char *path)
374 {
375 	dev_t uninitialized_var(dev);
376 	struct block_device *bdev;
377 
378 	bdev = lookup_bdev(path);
379 	if (IS_ERR(bdev))
380 		dev = name_to_dev_t(path);
381 	else {
382 		dev = bdev->bd_dev;
383 		bdput(bdev);
384 	}
385 
386 	return dev;
387 }
388 EXPORT_SYMBOL_GPL(dm_get_dev_t);
389 
390 /*
391  * Add a device to the list, or just increment the usage count if
392  * it's already present.
393  */
394 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
395 		  struct dm_dev **result)
396 {
397 	int r;
398 	dev_t dev;
399 	struct dm_dev_internal *dd;
400 	struct dm_table *t = ti->table;
401 
402 	BUG_ON(!t);
403 
404 	dev = dm_get_dev_t(path);
405 	if (!dev)
406 		return -ENODEV;
407 
408 	dd = find_device(&t->devices, dev);
409 	if (!dd) {
410 		dd = kmalloc(sizeof(*dd), GFP_KERNEL);
411 		if (!dd)
412 			return -ENOMEM;
413 
414 		if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
415 			kfree(dd);
416 			return r;
417 		}
418 
419 		atomic_set(&dd->count, 0);
420 		list_add(&dd->list, &t->devices);
421 
422 	} else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
423 		r = upgrade_mode(dd, mode, t->md);
424 		if (r)
425 			return r;
426 	}
427 	atomic_inc(&dd->count);
428 
429 	*result = dd->dm_dev;
430 	return 0;
431 }
432 EXPORT_SYMBOL(dm_get_device);
433 
434 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
435 				sector_t start, sector_t len, void *data)
436 {
437 	struct queue_limits *limits = data;
438 	struct block_device *bdev = dev->bdev;
439 	struct request_queue *q = bdev_get_queue(bdev);
440 	char b[BDEVNAME_SIZE];
441 
442 	if (unlikely(!q)) {
443 		DMWARN("%s: Cannot set limits for nonexistent device %s",
444 		       dm_device_name(ti->table->md), bdevname(bdev, b));
445 		return 0;
446 	}
447 
448 	if (bdev_stack_limits(limits, bdev, start) < 0)
449 		DMWARN("%s: adding target device %s caused an alignment inconsistency: "
450 		       "physical_block_size=%u, logical_block_size=%u, "
451 		       "alignment_offset=%u, start=%llu",
452 		       dm_device_name(ti->table->md), bdevname(bdev, b),
453 		       q->limits.physical_block_size,
454 		       q->limits.logical_block_size,
455 		       q->limits.alignment_offset,
456 		       (unsigned long long) start << SECTOR_SHIFT);
457 
458 	return 0;
459 }
460 
461 /*
462  * Decrement a device's use count and remove it if necessary.
463  */
464 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
465 {
466 	int found = 0;
467 	struct list_head *devices = &ti->table->devices;
468 	struct dm_dev_internal *dd;
469 
470 	list_for_each_entry(dd, devices, list) {
471 		if (dd->dm_dev == d) {
472 			found = 1;
473 			break;
474 		}
475 	}
476 	if (!found) {
477 		DMWARN("%s: device %s not in table devices list",
478 		       dm_device_name(ti->table->md), d->name);
479 		return;
480 	}
481 	if (atomic_dec_and_test(&dd->count)) {
482 		dm_put_table_device(ti->table->md, d);
483 		list_del(&dd->list);
484 		kfree(dd);
485 	}
486 }
487 EXPORT_SYMBOL(dm_put_device);
488 
489 /*
490  * Checks to see if the target joins onto the end of the table.
491  */
492 static int adjoin(struct dm_table *table, struct dm_target *ti)
493 {
494 	struct dm_target *prev;
495 
496 	if (!table->num_targets)
497 		return !ti->begin;
498 
499 	prev = &table->targets[table->num_targets - 1];
500 	return (ti->begin == (prev->begin + prev->len));
501 }
502 
503 /*
504  * Used to dynamically allocate the arg array.
505  *
506  * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
507  * process messages even if some device is suspended. These messages have a
508  * small fixed number of arguments.
509  *
510  * On the other hand, dm-switch needs to process bulk data using messages and
511  * excessive use of GFP_NOIO could cause trouble.
512  */
513 static char **realloc_argv(unsigned *array_size, char **old_argv)
514 {
515 	char **argv;
516 	unsigned new_size;
517 	gfp_t gfp;
518 
519 	if (*array_size) {
520 		new_size = *array_size * 2;
521 		gfp = GFP_KERNEL;
522 	} else {
523 		new_size = 8;
524 		gfp = GFP_NOIO;
525 	}
526 	argv = kmalloc(new_size * sizeof(*argv), gfp);
527 	if (argv) {
528 		memcpy(argv, old_argv, *array_size * sizeof(*argv));
529 		*array_size = new_size;
530 	}
531 
532 	kfree(old_argv);
533 	return argv;
534 }
535 
536 /*
537  * Destructively splits up the argument list to pass to ctr.
538  */
539 int dm_split_args(int *argc, char ***argvp, char *input)
540 {
541 	char *start, *end = input, *out, **argv = NULL;
542 	unsigned array_size = 0;
543 
544 	*argc = 0;
545 
546 	if (!input) {
547 		*argvp = NULL;
548 		return 0;
549 	}
550 
551 	argv = realloc_argv(&array_size, argv);
552 	if (!argv)
553 		return -ENOMEM;
554 
555 	while (1) {
556 		/* Skip whitespace */
557 		start = skip_spaces(end);
558 
559 		if (!*start)
560 			break;	/* success, we hit the end */
561 
562 		/* 'out' is used to remove any back-quotes */
563 		end = out = start;
564 		while (*end) {
565 			/* Everything apart from '\0' can be quoted */
566 			if (*end == '\\' && *(end + 1)) {
567 				*out++ = *(end + 1);
568 				end += 2;
569 				continue;
570 			}
571 
572 			if (isspace(*end))
573 				break;	/* end of token */
574 
575 			*out++ = *end++;
576 		}
577 
578 		/* have we already filled the array ? */
579 		if ((*argc + 1) > array_size) {
580 			argv = realloc_argv(&array_size, argv);
581 			if (!argv)
582 				return -ENOMEM;
583 		}
584 
585 		/* we know this is whitespace */
586 		if (*end)
587 			end++;
588 
589 		/* terminate the string and put it in the array */
590 		*out = '\0';
591 		argv[*argc] = start;
592 		(*argc)++;
593 	}
594 
595 	*argvp = argv;
596 	return 0;
597 }
598 
599 /*
600  * Impose necessary and sufficient conditions on a devices's table such
601  * that any incoming bio which respects its logical_block_size can be
602  * processed successfully.  If it falls across the boundary between
603  * two or more targets, the size of each piece it gets split into must
604  * be compatible with the logical_block_size of the target processing it.
605  */
606 static int validate_hardware_logical_block_alignment(struct dm_table *table,
607 						 struct queue_limits *limits)
608 {
609 	/*
610 	 * This function uses arithmetic modulo the logical_block_size
611 	 * (in units of 512-byte sectors).
612 	 */
613 	unsigned short device_logical_block_size_sects =
614 		limits->logical_block_size >> SECTOR_SHIFT;
615 
616 	/*
617 	 * Offset of the start of the next table entry, mod logical_block_size.
618 	 */
619 	unsigned short next_target_start = 0;
620 
621 	/*
622 	 * Given an aligned bio that extends beyond the end of a
623 	 * target, how many sectors must the next target handle?
624 	 */
625 	unsigned short remaining = 0;
626 
627 	struct dm_target *uninitialized_var(ti);
628 	struct queue_limits ti_limits;
629 	unsigned i = 0;
630 
631 	/*
632 	 * Check each entry in the table in turn.
633 	 */
634 	while (i < dm_table_get_num_targets(table)) {
635 		ti = dm_table_get_target(table, i++);
636 
637 		blk_set_stacking_limits(&ti_limits);
638 
639 		/* combine all target devices' limits */
640 		if (ti->type->iterate_devices)
641 			ti->type->iterate_devices(ti, dm_set_device_limits,
642 						  &ti_limits);
643 
644 		/*
645 		 * If the remaining sectors fall entirely within this
646 		 * table entry are they compatible with its logical_block_size?
647 		 */
648 		if (remaining < ti->len &&
649 		    remaining & ((ti_limits.logical_block_size >>
650 				  SECTOR_SHIFT) - 1))
651 			break;	/* Error */
652 
653 		next_target_start =
654 		    (unsigned short) ((next_target_start + ti->len) &
655 				      (device_logical_block_size_sects - 1));
656 		remaining = next_target_start ?
657 		    device_logical_block_size_sects - next_target_start : 0;
658 	}
659 
660 	if (remaining) {
661 		DMWARN("%s: table line %u (start sect %llu len %llu) "
662 		       "not aligned to h/w logical block size %u",
663 		       dm_device_name(table->md), i,
664 		       (unsigned long long) ti->begin,
665 		       (unsigned long long) ti->len,
666 		       limits->logical_block_size);
667 		return -EINVAL;
668 	}
669 
670 	return 0;
671 }
672 
673 int dm_table_add_target(struct dm_table *t, const char *type,
674 			sector_t start, sector_t len, char *params)
675 {
676 	int r = -EINVAL, argc;
677 	char **argv;
678 	struct dm_target *tgt;
679 
680 	if (t->singleton) {
681 		DMERR("%s: target type %s must appear alone in table",
682 		      dm_device_name(t->md), t->targets->type->name);
683 		return -EINVAL;
684 	}
685 
686 	BUG_ON(t->num_targets >= t->num_allocated);
687 
688 	tgt = t->targets + t->num_targets;
689 	memset(tgt, 0, sizeof(*tgt));
690 
691 	if (!len) {
692 		DMERR("%s: zero-length target", dm_device_name(t->md));
693 		return -EINVAL;
694 	}
695 
696 	tgt->type = dm_get_target_type(type);
697 	if (!tgt->type) {
698 		DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
699 		return -EINVAL;
700 	}
701 
702 	if (dm_target_needs_singleton(tgt->type)) {
703 		if (t->num_targets) {
704 			tgt->error = "singleton target type must appear alone in table";
705 			goto bad;
706 		}
707 		t->singleton = true;
708 	}
709 
710 	if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
711 		tgt->error = "target type may not be included in a read-only table";
712 		goto bad;
713 	}
714 
715 	if (t->immutable_target_type) {
716 		if (t->immutable_target_type != tgt->type) {
717 			tgt->error = "immutable target type cannot be mixed with other target types";
718 			goto bad;
719 		}
720 	} else if (dm_target_is_immutable(tgt->type)) {
721 		if (t->num_targets) {
722 			tgt->error = "immutable target type cannot be mixed with other target types";
723 			goto bad;
724 		}
725 		t->immutable_target_type = tgt->type;
726 	}
727 
728 	tgt->table = t;
729 	tgt->begin = start;
730 	tgt->len = len;
731 	tgt->error = "Unknown error";
732 
733 	/*
734 	 * Does this target adjoin the previous one ?
735 	 */
736 	if (!adjoin(t, tgt)) {
737 		tgt->error = "Gap in table";
738 		goto bad;
739 	}
740 
741 	r = dm_split_args(&argc, &argv, params);
742 	if (r) {
743 		tgt->error = "couldn't split parameters (insufficient memory)";
744 		goto bad;
745 	}
746 
747 	r = tgt->type->ctr(tgt, argc, argv);
748 	kfree(argv);
749 	if (r)
750 		goto bad;
751 
752 	t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
753 
754 	if (!tgt->num_discard_bios && tgt->discards_supported)
755 		DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
756 		       dm_device_name(t->md), type);
757 
758 	return 0;
759 
760  bad:
761 	DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
762 	dm_put_target_type(tgt->type);
763 	return r;
764 }
765 
766 /*
767  * Target argument parsing helpers.
768  */
769 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
770 			     unsigned *value, char **error, unsigned grouped)
771 {
772 	const char *arg_str = dm_shift_arg(arg_set);
773 	char dummy;
774 
775 	if (!arg_str ||
776 	    (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
777 	    (*value < arg->min) ||
778 	    (*value > arg->max) ||
779 	    (grouped && arg_set->argc < *value)) {
780 		*error = arg->error;
781 		return -EINVAL;
782 	}
783 
784 	return 0;
785 }
786 
787 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
788 		unsigned *value, char **error)
789 {
790 	return validate_next_arg(arg, arg_set, value, error, 0);
791 }
792 EXPORT_SYMBOL(dm_read_arg);
793 
794 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
795 		      unsigned *value, char **error)
796 {
797 	return validate_next_arg(arg, arg_set, value, error, 1);
798 }
799 EXPORT_SYMBOL(dm_read_arg_group);
800 
801 const char *dm_shift_arg(struct dm_arg_set *as)
802 {
803 	char *r;
804 
805 	if (as->argc) {
806 		as->argc--;
807 		r = *as->argv;
808 		as->argv++;
809 		return r;
810 	}
811 
812 	return NULL;
813 }
814 EXPORT_SYMBOL(dm_shift_arg);
815 
816 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
817 {
818 	BUG_ON(as->argc < num_args);
819 	as->argc -= num_args;
820 	as->argv += num_args;
821 }
822 EXPORT_SYMBOL(dm_consume_args);
823 
824 static bool __table_type_bio_based(unsigned table_type)
825 {
826 	return (table_type == DM_TYPE_BIO_BASED ||
827 		table_type == DM_TYPE_DAX_BIO_BASED);
828 }
829 
830 static bool __table_type_request_based(unsigned table_type)
831 {
832 	return (table_type == DM_TYPE_REQUEST_BASED ||
833 		table_type == DM_TYPE_MQ_REQUEST_BASED);
834 }
835 
836 void dm_table_set_type(struct dm_table *t, unsigned type)
837 {
838 	t->type = type;
839 }
840 EXPORT_SYMBOL_GPL(dm_table_set_type);
841 
842 static int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
843 			       sector_t start, sector_t len, void *data)
844 {
845 	struct request_queue *q = bdev_get_queue(dev->bdev);
846 
847 	return q && blk_queue_dax(q);
848 }
849 
850 static bool dm_table_supports_dax(struct dm_table *t)
851 {
852 	struct dm_target *ti;
853 	unsigned i = 0;
854 
855 	/* Ensure that all targets support DAX. */
856 	while (i < dm_table_get_num_targets(t)) {
857 		ti = dm_table_get_target(t, i++);
858 
859 		if (!ti->type->direct_access)
860 			return false;
861 
862 		if (!ti->type->iterate_devices ||
863 		    !ti->type->iterate_devices(ti, device_supports_dax, NULL))
864 			return false;
865 	}
866 
867 	return true;
868 }
869 
870 static int dm_table_determine_type(struct dm_table *t)
871 {
872 	unsigned i;
873 	unsigned bio_based = 0, request_based = 0, hybrid = 0;
874 	bool verify_blk_mq = false;
875 	struct dm_target *tgt;
876 	struct dm_dev_internal *dd;
877 	struct list_head *devices = dm_table_get_devices(t);
878 	unsigned live_md_type = dm_get_md_type(t->md);
879 
880 	if (t->type != DM_TYPE_NONE) {
881 		/* target already set the table's type */
882 		if (t->type == DM_TYPE_BIO_BASED)
883 			return 0;
884 		BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
885 		goto verify_rq_based;
886 	}
887 
888 	for (i = 0; i < t->num_targets; i++) {
889 		tgt = t->targets + i;
890 		if (dm_target_hybrid(tgt))
891 			hybrid = 1;
892 		else if (dm_target_request_based(tgt))
893 			request_based = 1;
894 		else
895 			bio_based = 1;
896 
897 		if (bio_based && request_based) {
898 			DMWARN("Inconsistent table: different target types"
899 			       " can't be mixed up");
900 			return -EINVAL;
901 		}
902 	}
903 
904 	if (hybrid && !bio_based && !request_based) {
905 		/*
906 		 * The targets can work either way.
907 		 * Determine the type from the live device.
908 		 * Default to bio-based if device is new.
909 		 */
910 		if (__table_type_request_based(live_md_type))
911 			request_based = 1;
912 		else
913 			bio_based = 1;
914 	}
915 
916 	if (bio_based) {
917 		/* We must use this table as bio-based */
918 		t->type = DM_TYPE_BIO_BASED;
919 		if (dm_table_supports_dax(t) ||
920 		    (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED))
921 			t->type = DM_TYPE_DAX_BIO_BASED;
922 		return 0;
923 	}
924 
925 	BUG_ON(!request_based); /* No targets in this table */
926 
927 	if (list_empty(devices) && __table_type_request_based(live_md_type)) {
928 		/* inherit live MD type */
929 		t->type = live_md_type;
930 		return 0;
931 	}
932 
933 	/*
934 	 * The only way to establish DM_TYPE_MQ_REQUEST_BASED is by
935 	 * having a compatible target use dm_table_set_type.
936 	 */
937 	t->type = DM_TYPE_REQUEST_BASED;
938 
939 verify_rq_based:
940 	/*
941 	 * Request-based dm supports only tables that have a single target now.
942 	 * To support multiple targets, request splitting support is needed,
943 	 * and that needs lots of changes in the block-layer.
944 	 * (e.g. request completion process for partial completion.)
945 	 */
946 	if (t->num_targets > 1) {
947 		DMWARN("Request-based dm doesn't support multiple targets yet");
948 		return -EINVAL;
949 	}
950 
951 	/* Non-request-stackable devices can't be used for request-based dm */
952 	list_for_each_entry(dd, devices, list) {
953 		struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
954 
955 		if (!blk_queue_stackable(q)) {
956 			DMERR("table load rejected: including"
957 			      " non-request-stackable devices");
958 			return -EINVAL;
959 		}
960 
961 		if (q->mq_ops)
962 			verify_blk_mq = true;
963 	}
964 
965 	if (verify_blk_mq) {
966 		/* verify _all_ devices in the table are blk-mq devices */
967 		list_for_each_entry(dd, devices, list)
968 			if (!bdev_get_queue(dd->dm_dev->bdev)->mq_ops) {
969 				DMERR("table load rejected: not all devices"
970 				      " are blk-mq request-stackable");
971 				return -EINVAL;
972 			}
973 
974 		t->all_blk_mq = true;
975 	}
976 
977 	return 0;
978 }
979 
980 unsigned dm_table_get_type(struct dm_table *t)
981 {
982 	return t->type;
983 }
984 
985 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
986 {
987 	return t->immutable_target_type;
988 }
989 
990 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
991 {
992 	/* Immutable target is implicitly a singleton */
993 	if (t->num_targets > 1 ||
994 	    !dm_target_is_immutable(t->targets[0].type))
995 		return NULL;
996 
997 	return t->targets;
998 }
999 
1000 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1001 {
1002 	struct dm_target *uninitialized_var(ti);
1003 	unsigned i = 0;
1004 
1005 	while (i < dm_table_get_num_targets(t)) {
1006 		ti = dm_table_get_target(t, i++);
1007 		if (dm_target_is_wildcard(ti->type))
1008 			return ti;
1009 	}
1010 
1011 	return NULL;
1012 }
1013 
1014 bool dm_table_bio_based(struct dm_table *t)
1015 {
1016 	return __table_type_bio_based(dm_table_get_type(t));
1017 }
1018 
1019 bool dm_table_request_based(struct dm_table *t)
1020 {
1021 	return __table_type_request_based(dm_table_get_type(t));
1022 }
1023 
1024 bool dm_table_all_blk_mq_devices(struct dm_table *t)
1025 {
1026 	return t->all_blk_mq;
1027 }
1028 
1029 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1030 {
1031 	unsigned type = dm_table_get_type(t);
1032 	unsigned per_io_data_size = 0;
1033 	struct dm_target *tgt;
1034 	unsigned i;
1035 
1036 	if (unlikely(type == DM_TYPE_NONE)) {
1037 		DMWARN("no table type is set, can't allocate mempools");
1038 		return -EINVAL;
1039 	}
1040 
1041 	if (__table_type_bio_based(type))
1042 		for (i = 0; i < t->num_targets; i++) {
1043 			tgt = t->targets + i;
1044 			per_io_data_size = max(per_io_data_size, tgt->per_io_data_size);
1045 		}
1046 
1047 	t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported, per_io_data_size);
1048 	if (!t->mempools)
1049 		return -ENOMEM;
1050 
1051 	return 0;
1052 }
1053 
1054 void dm_table_free_md_mempools(struct dm_table *t)
1055 {
1056 	dm_free_md_mempools(t->mempools);
1057 	t->mempools = NULL;
1058 }
1059 
1060 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1061 {
1062 	return t->mempools;
1063 }
1064 
1065 static int setup_indexes(struct dm_table *t)
1066 {
1067 	int i;
1068 	unsigned int total = 0;
1069 	sector_t *indexes;
1070 
1071 	/* allocate the space for *all* the indexes */
1072 	for (i = t->depth - 2; i >= 0; i--) {
1073 		t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1074 		total += t->counts[i];
1075 	}
1076 
1077 	indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1078 	if (!indexes)
1079 		return -ENOMEM;
1080 
1081 	/* set up internal nodes, bottom-up */
1082 	for (i = t->depth - 2; i >= 0; i--) {
1083 		t->index[i] = indexes;
1084 		indexes += (KEYS_PER_NODE * t->counts[i]);
1085 		setup_btree_index(i, t);
1086 	}
1087 
1088 	return 0;
1089 }
1090 
1091 /*
1092  * Builds the btree to index the map.
1093  */
1094 static int dm_table_build_index(struct dm_table *t)
1095 {
1096 	int r = 0;
1097 	unsigned int leaf_nodes;
1098 
1099 	/* how many indexes will the btree have ? */
1100 	leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1101 	t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1102 
1103 	/* leaf layer has already been set up */
1104 	t->counts[t->depth - 1] = leaf_nodes;
1105 	t->index[t->depth - 1] = t->highs;
1106 
1107 	if (t->depth >= 2)
1108 		r = setup_indexes(t);
1109 
1110 	return r;
1111 }
1112 
1113 static bool integrity_profile_exists(struct gendisk *disk)
1114 {
1115 	return !!blk_get_integrity(disk);
1116 }
1117 
1118 /*
1119  * Get a disk whose integrity profile reflects the table's profile.
1120  * Returns NULL if integrity support was inconsistent or unavailable.
1121  */
1122 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1123 {
1124 	struct list_head *devices = dm_table_get_devices(t);
1125 	struct dm_dev_internal *dd = NULL;
1126 	struct gendisk *prev_disk = NULL, *template_disk = NULL;
1127 
1128 	list_for_each_entry(dd, devices, list) {
1129 		template_disk = dd->dm_dev->bdev->bd_disk;
1130 		if (!integrity_profile_exists(template_disk))
1131 			goto no_integrity;
1132 		else if (prev_disk &&
1133 			 blk_integrity_compare(prev_disk, template_disk) < 0)
1134 			goto no_integrity;
1135 		prev_disk = template_disk;
1136 	}
1137 
1138 	return template_disk;
1139 
1140 no_integrity:
1141 	if (prev_disk)
1142 		DMWARN("%s: integrity not set: %s and %s profile mismatch",
1143 		       dm_device_name(t->md),
1144 		       prev_disk->disk_name,
1145 		       template_disk->disk_name);
1146 	return NULL;
1147 }
1148 
1149 /*
1150  * Register the mapped device for blk_integrity support if the
1151  * underlying devices have an integrity profile.  But all devices may
1152  * not have matching profiles (checking all devices isn't reliable
1153  * during table load because this table may use other DM device(s) which
1154  * must be resumed before they will have an initialized integity
1155  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1156  * profile validation: First pass during table load, final pass during
1157  * resume.
1158  */
1159 static int dm_table_register_integrity(struct dm_table *t)
1160 {
1161 	struct mapped_device *md = t->md;
1162 	struct gendisk *template_disk = NULL;
1163 
1164 	template_disk = dm_table_get_integrity_disk(t);
1165 	if (!template_disk)
1166 		return 0;
1167 
1168 	if (!integrity_profile_exists(dm_disk(md))) {
1169 		t->integrity_supported = true;
1170 		/*
1171 		 * Register integrity profile during table load; we can do
1172 		 * this because the final profile must match during resume.
1173 		 */
1174 		blk_integrity_register(dm_disk(md),
1175 				       blk_get_integrity(template_disk));
1176 		return 0;
1177 	}
1178 
1179 	/*
1180 	 * If DM device already has an initialized integrity
1181 	 * profile the new profile should not conflict.
1182 	 */
1183 	if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1184 		DMWARN("%s: conflict with existing integrity profile: "
1185 		       "%s profile mismatch",
1186 		       dm_device_name(t->md),
1187 		       template_disk->disk_name);
1188 		return 1;
1189 	}
1190 
1191 	/* Preserve existing integrity profile */
1192 	t->integrity_supported = true;
1193 	return 0;
1194 }
1195 
1196 /*
1197  * Prepares the table for use by building the indices,
1198  * setting the type, and allocating mempools.
1199  */
1200 int dm_table_complete(struct dm_table *t)
1201 {
1202 	int r;
1203 
1204 	r = dm_table_determine_type(t);
1205 	if (r) {
1206 		DMERR("unable to determine table type");
1207 		return r;
1208 	}
1209 
1210 	r = dm_table_build_index(t);
1211 	if (r) {
1212 		DMERR("unable to build btrees");
1213 		return r;
1214 	}
1215 
1216 	r = dm_table_register_integrity(t);
1217 	if (r) {
1218 		DMERR("could not register integrity profile.");
1219 		return r;
1220 	}
1221 
1222 	r = dm_table_alloc_md_mempools(t, t->md);
1223 	if (r)
1224 		DMERR("unable to allocate mempools");
1225 
1226 	return r;
1227 }
1228 
1229 static DEFINE_MUTEX(_event_lock);
1230 void dm_table_event_callback(struct dm_table *t,
1231 			     void (*fn)(void *), void *context)
1232 {
1233 	mutex_lock(&_event_lock);
1234 	t->event_fn = fn;
1235 	t->event_context = context;
1236 	mutex_unlock(&_event_lock);
1237 }
1238 
1239 void dm_table_event(struct dm_table *t)
1240 {
1241 	/*
1242 	 * You can no longer call dm_table_event() from interrupt
1243 	 * context, use a bottom half instead.
1244 	 */
1245 	BUG_ON(in_interrupt());
1246 
1247 	mutex_lock(&_event_lock);
1248 	if (t->event_fn)
1249 		t->event_fn(t->event_context);
1250 	mutex_unlock(&_event_lock);
1251 }
1252 EXPORT_SYMBOL(dm_table_event);
1253 
1254 sector_t dm_table_get_size(struct dm_table *t)
1255 {
1256 	return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1257 }
1258 EXPORT_SYMBOL(dm_table_get_size);
1259 
1260 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1261 {
1262 	if (index >= t->num_targets)
1263 		return NULL;
1264 
1265 	return t->targets + index;
1266 }
1267 
1268 /*
1269  * Search the btree for the correct target.
1270  *
1271  * Caller should check returned pointer with dm_target_is_valid()
1272  * to trap I/O beyond end of device.
1273  */
1274 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1275 {
1276 	unsigned int l, n = 0, k = 0;
1277 	sector_t *node;
1278 
1279 	for (l = 0; l < t->depth; l++) {
1280 		n = get_child(n, k);
1281 		node = get_node(t, l, n);
1282 
1283 		for (k = 0; k < KEYS_PER_NODE; k++)
1284 			if (node[k] >= sector)
1285 				break;
1286 	}
1287 
1288 	return &t->targets[(KEYS_PER_NODE * n) + k];
1289 }
1290 
1291 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1292 			sector_t start, sector_t len, void *data)
1293 {
1294 	unsigned *num_devices = data;
1295 
1296 	(*num_devices)++;
1297 
1298 	return 0;
1299 }
1300 
1301 /*
1302  * Check whether a table has no data devices attached using each
1303  * target's iterate_devices method.
1304  * Returns false if the result is unknown because a target doesn't
1305  * support iterate_devices.
1306  */
1307 bool dm_table_has_no_data_devices(struct dm_table *table)
1308 {
1309 	struct dm_target *uninitialized_var(ti);
1310 	unsigned i = 0, num_devices = 0;
1311 
1312 	while (i < dm_table_get_num_targets(table)) {
1313 		ti = dm_table_get_target(table, i++);
1314 
1315 		if (!ti->type->iterate_devices)
1316 			return false;
1317 
1318 		ti->type->iterate_devices(ti, count_device, &num_devices);
1319 		if (num_devices)
1320 			return false;
1321 	}
1322 
1323 	return true;
1324 }
1325 
1326 /*
1327  * Establish the new table's queue_limits and validate them.
1328  */
1329 int dm_calculate_queue_limits(struct dm_table *table,
1330 			      struct queue_limits *limits)
1331 {
1332 	struct dm_target *uninitialized_var(ti);
1333 	struct queue_limits ti_limits;
1334 	unsigned i = 0;
1335 
1336 	blk_set_stacking_limits(limits);
1337 
1338 	while (i < dm_table_get_num_targets(table)) {
1339 		blk_set_stacking_limits(&ti_limits);
1340 
1341 		ti = dm_table_get_target(table, i++);
1342 
1343 		if (!ti->type->iterate_devices)
1344 			goto combine_limits;
1345 
1346 		/*
1347 		 * Combine queue limits of all the devices this target uses.
1348 		 */
1349 		ti->type->iterate_devices(ti, dm_set_device_limits,
1350 					  &ti_limits);
1351 
1352 		/* Set I/O hints portion of queue limits */
1353 		if (ti->type->io_hints)
1354 			ti->type->io_hints(ti, &ti_limits);
1355 
1356 		/*
1357 		 * Check each device area is consistent with the target's
1358 		 * overall queue limits.
1359 		 */
1360 		if (ti->type->iterate_devices(ti, device_area_is_invalid,
1361 					      &ti_limits))
1362 			return -EINVAL;
1363 
1364 combine_limits:
1365 		/*
1366 		 * Merge this target's queue limits into the overall limits
1367 		 * for the table.
1368 		 */
1369 		if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1370 			DMWARN("%s: adding target device "
1371 			       "(start sect %llu len %llu) "
1372 			       "caused an alignment inconsistency",
1373 			       dm_device_name(table->md),
1374 			       (unsigned long long) ti->begin,
1375 			       (unsigned long long) ti->len);
1376 	}
1377 
1378 	return validate_hardware_logical_block_alignment(table, limits);
1379 }
1380 
1381 /*
1382  * Verify that all devices have an integrity profile that matches the
1383  * DM device's registered integrity profile.  If the profiles don't
1384  * match then unregister the DM device's integrity profile.
1385  */
1386 static void dm_table_verify_integrity(struct dm_table *t)
1387 {
1388 	struct gendisk *template_disk = NULL;
1389 
1390 	if (t->integrity_supported) {
1391 		/*
1392 		 * Verify that the original integrity profile
1393 		 * matches all the devices in this table.
1394 		 */
1395 		template_disk = dm_table_get_integrity_disk(t);
1396 		if (template_disk &&
1397 		    blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1398 			return;
1399 	}
1400 
1401 	if (integrity_profile_exists(dm_disk(t->md))) {
1402 		DMWARN("%s: unable to establish an integrity profile",
1403 		       dm_device_name(t->md));
1404 		blk_integrity_unregister(dm_disk(t->md));
1405 	}
1406 }
1407 
1408 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1409 				sector_t start, sector_t len, void *data)
1410 {
1411 	unsigned long flush = (unsigned long) data;
1412 	struct request_queue *q = bdev_get_queue(dev->bdev);
1413 
1414 	return q && (q->queue_flags & flush);
1415 }
1416 
1417 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1418 {
1419 	struct dm_target *ti;
1420 	unsigned i = 0;
1421 
1422 	/*
1423 	 * Require at least one underlying device to support flushes.
1424 	 * t->devices includes internal dm devices such as mirror logs
1425 	 * so we need to use iterate_devices here, which targets
1426 	 * supporting flushes must provide.
1427 	 */
1428 	while (i < dm_table_get_num_targets(t)) {
1429 		ti = dm_table_get_target(t, i++);
1430 
1431 		if (!ti->num_flush_bios)
1432 			continue;
1433 
1434 		if (ti->flush_supported)
1435 			return true;
1436 
1437 		if (ti->type->iterate_devices &&
1438 		    ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1439 			return true;
1440 	}
1441 
1442 	return false;
1443 }
1444 
1445 static bool dm_table_discard_zeroes_data(struct dm_table *t)
1446 {
1447 	struct dm_target *ti;
1448 	unsigned i = 0;
1449 
1450 	/* Ensure that all targets supports discard_zeroes_data. */
1451 	while (i < dm_table_get_num_targets(t)) {
1452 		ti = dm_table_get_target(t, i++);
1453 
1454 		if (ti->discard_zeroes_data_unsupported)
1455 			return false;
1456 	}
1457 
1458 	return true;
1459 }
1460 
1461 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1462 			    sector_t start, sector_t len, void *data)
1463 {
1464 	struct request_queue *q = bdev_get_queue(dev->bdev);
1465 
1466 	return q && blk_queue_nonrot(q);
1467 }
1468 
1469 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1470 			     sector_t start, sector_t len, void *data)
1471 {
1472 	struct request_queue *q = bdev_get_queue(dev->bdev);
1473 
1474 	return q && !blk_queue_add_random(q);
1475 }
1476 
1477 static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1478 				   sector_t start, sector_t len, void *data)
1479 {
1480 	struct request_queue *q = bdev_get_queue(dev->bdev);
1481 
1482 	return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1483 }
1484 
1485 static bool dm_table_all_devices_attribute(struct dm_table *t,
1486 					   iterate_devices_callout_fn func)
1487 {
1488 	struct dm_target *ti;
1489 	unsigned i = 0;
1490 
1491 	while (i < dm_table_get_num_targets(t)) {
1492 		ti = dm_table_get_target(t, i++);
1493 
1494 		if (!ti->type->iterate_devices ||
1495 		    !ti->type->iterate_devices(ti, func, NULL))
1496 			return false;
1497 	}
1498 
1499 	return true;
1500 }
1501 
1502 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1503 					 sector_t start, sector_t len, void *data)
1504 {
1505 	struct request_queue *q = bdev_get_queue(dev->bdev);
1506 
1507 	return q && !q->limits.max_write_same_sectors;
1508 }
1509 
1510 static bool dm_table_supports_write_same(struct dm_table *t)
1511 {
1512 	struct dm_target *ti;
1513 	unsigned i = 0;
1514 
1515 	while (i < dm_table_get_num_targets(t)) {
1516 		ti = dm_table_get_target(t, i++);
1517 
1518 		if (!ti->num_write_same_bios)
1519 			return false;
1520 
1521 		if (!ti->type->iterate_devices ||
1522 		    ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1523 			return false;
1524 	}
1525 
1526 	return true;
1527 }
1528 
1529 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1530 				  sector_t start, sector_t len, void *data)
1531 {
1532 	struct request_queue *q = bdev_get_queue(dev->bdev);
1533 
1534 	return q && blk_queue_discard(q);
1535 }
1536 
1537 static bool dm_table_supports_discards(struct dm_table *t)
1538 {
1539 	struct dm_target *ti;
1540 	unsigned i = 0;
1541 
1542 	/*
1543 	 * Unless any target used by the table set discards_supported,
1544 	 * require at least one underlying device to support discards.
1545 	 * t->devices includes internal dm devices such as mirror logs
1546 	 * so we need to use iterate_devices here, which targets
1547 	 * supporting discard selectively must provide.
1548 	 */
1549 	while (i < dm_table_get_num_targets(t)) {
1550 		ti = dm_table_get_target(t, i++);
1551 
1552 		if (!ti->num_discard_bios)
1553 			continue;
1554 
1555 		if (ti->discards_supported)
1556 			return true;
1557 
1558 		if (ti->type->iterate_devices &&
1559 		    ti->type->iterate_devices(ti, device_discard_capable, NULL))
1560 			return true;
1561 	}
1562 
1563 	return false;
1564 }
1565 
1566 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1567 			       struct queue_limits *limits)
1568 {
1569 	bool wc = false, fua = false;
1570 
1571 	/*
1572 	 * Copy table's limits to the DM device's request_queue
1573 	 */
1574 	q->limits = *limits;
1575 
1576 	if (!dm_table_supports_discards(t))
1577 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1578 	else
1579 		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1580 
1581 	if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1582 		wc = true;
1583 		if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1584 			fua = true;
1585 	}
1586 	blk_queue_write_cache(q, wc, fua);
1587 
1588 	if (!dm_table_discard_zeroes_data(t))
1589 		q->limits.discard_zeroes_data = 0;
1590 
1591 	/* Ensure that all underlying devices are non-rotational. */
1592 	if (dm_table_all_devices_attribute(t, device_is_nonrot))
1593 		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1594 	else
1595 		queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1596 
1597 	if (!dm_table_supports_write_same(t))
1598 		q->limits.max_write_same_sectors = 0;
1599 
1600 	if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1601 		queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1602 	else
1603 		queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1604 
1605 	dm_table_verify_integrity(t);
1606 
1607 	/*
1608 	 * Determine whether or not this queue's I/O timings contribute
1609 	 * to the entropy pool, Only request-based targets use this.
1610 	 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1611 	 * have it set.
1612 	 */
1613 	if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1614 		queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1615 
1616 	/*
1617 	 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1618 	 * visible to other CPUs because, once the flag is set, incoming bios
1619 	 * are processed by request-based dm, which refers to the queue
1620 	 * settings.
1621 	 * Until the flag set, bios are passed to bio-based dm and queued to
1622 	 * md->deferred where queue settings are not needed yet.
1623 	 * Those bios are passed to request-based dm at the resume time.
1624 	 */
1625 	smp_mb();
1626 	if (dm_table_request_based(t))
1627 		queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1628 }
1629 
1630 unsigned int dm_table_get_num_targets(struct dm_table *t)
1631 {
1632 	return t->num_targets;
1633 }
1634 
1635 struct list_head *dm_table_get_devices(struct dm_table *t)
1636 {
1637 	return &t->devices;
1638 }
1639 
1640 fmode_t dm_table_get_mode(struct dm_table *t)
1641 {
1642 	return t->mode;
1643 }
1644 EXPORT_SYMBOL(dm_table_get_mode);
1645 
1646 enum suspend_mode {
1647 	PRESUSPEND,
1648 	PRESUSPEND_UNDO,
1649 	POSTSUSPEND,
1650 };
1651 
1652 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1653 {
1654 	int i = t->num_targets;
1655 	struct dm_target *ti = t->targets;
1656 
1657 	while (i--) {
1658 		switch (mode) {
1659 		case PRESUSPEND:
1660 			if (ti->type->presuspend)
1661 				ti->type->presuspend(ti);
1662 			break;
1663 		case PRESUSPEND_UNDO:
1664 			if (ti->type->presuspend_undo)
1665 				ti->type->presuspend_undo(ti);
1666 			break;
1667 		case POSTSUSPEND:
1668 			if (ti->type->postsuspend)
1669 				ti->type->postsuspend(ti);
1670 			break;
1671 		}
1672 		ti++;
1673 	}
1674 }
1675 
1676 void dm_table_presuspend_targets(struct dm_table *t)
1677 {
1678 	if (!t)
1679 		return;
1680 
1681 	suspend_targets(t, PRESUSPEND);
1682 }
1683 
1684 void dm_table_presuspend_undo_targets(struct dm_table *t)
1685 {
1686 	if (!t)
1687 		return;
1688 
1689 	suspend_targets(t, PRESUSPEND_UNDO);
1690 }
1691 
1692 void dm_table_postsuspend_targets(struct dm_table *t)
1693 {
1694 	if (!t)
1695 		return;
1696 
1697 	suspend_targets(t, POSTSUSPEND);
1698 }
1699 
1700 int dm_table_resume_targets(struct dm_table *t)
1701 {
1702 	int i, r = 0;
1703 
1704 	for (i = 0; i < t->num_targets; i++) {
1705 		struct dm_target *ti = t->targets + i;
1706 
1707 		if (!ti->type->preresume)
1708 			continue;
1709 
1710 		r = ti->type->preresume(ti);
1711 		if (r) {
1712 			DMERR("%s: %s: preresume failed, error = %d",
1713 			      dm_device_name(t->md), ti->type->name, r);
1714 			return r;
1715 		}
1716 	}
1717 
1718 	for (i = 0; i < t->num_targets; i++) {
1719 		struct dm_target *ti = t->targets + i;
1720 
1721 		if (ti->type->resume)
1722 			ti->type->resume(ti);
1723 	}
1724 
1725 	return 0;
1726 }
1727 
1728 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1729 {
1730 	list_add(&cb->list, &t->target_callbacks);
1731 }
1732 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1733 
1734 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1735 {
1736 	struct dm_dev_internal *dd;
1737 	struct list_head *devices = dm_table_get_devices(t);
1738 	struct dm_target_callbacks *cb;
1739 	int r = 0;
1740 
1741 	list_for_each_entry(dd, devices, list) {
1742 		struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1743 		char b[BDEVNAME_SIZE];
1744 
1745 		if (likely(q))
1746 			r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1747 		else
1748 			DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1749 				     dm_device_name(t->md),
1750 				     bdevname(dd->dm_dev->bdev, b));
1751 	}
1752 
1753 	list_for_each_entry(cb, &t->target_callbacks, list)
1754 		if (cb->congested_fn)
1755 			r |= cb->congested_fn(cb, bdi_bits);
1756 
1757 	return r;
1758 }
1759 
1760 struct mapped_device *dm_table_get_md(struct dm_table *t)
1761 {
1762 	return t->md;
1763 }
1764 EXPORT_SYMBOL(dm_table_get_md);
1765 
1766 void dm_table_run_md_queue_async(struct dm_table *t)
1767 {
1768 	struct mapped_device *md;
1769 	struct request_queue *queue;
1770 	unsigned long flags;
1771 
1772 	if (!dm_table_request_based(t))
1773 		return;
1774 
1775 	md = dm_table_get_md(t);
1776 	queue = dm_get_md_queue(md);
1777 	if (queue) {
1778 		if (queue->mq_ops)
1779 			blk_mq_run_hw_queues(queue, true);
1780 		else {
1781 			spin_lock_irqsave(queue->queue_lock, flags);
1782 			blk_run_queue_async(queue);
1783 			spin_unlock_irqrestore(queue->queue_lock, flags);
1784 		}
1785 	}
1786 }
1787 EXPORT_SYMBOL(dm_table_run_md_queue_async);
1788 
1789