xref: /openbmc/linux/drivers/md/dm-table.c (revision 568b9de4)
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 #include <linux/dax.h>
24 
25 #define DM_MSG_PREFIX "table"
26 
27 #define MAX_DEPTH 16
28 #define NODE_SIZE L1_CACHE_BYTES
29 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
30 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
31 
32 struct dm_table {
33 	struct mapped_device *md;
34 	enum dm_queue_mode type;
35 
36 	/* btree table */
37 	unsigned int depth;
38 	unsigned int counts[MAX_DEPTH];	/* in nodes */
39 	sector_t *index[MAX_DEPTH];
40 
41 	unsigned int num_targets;
42 	unsigned int num_allocated;
43 	sector_t *highs;
44 	struct dm_target *targets;
45 
46 	struct target_type *immutable_target_type;
47 
48 	bool integrity_supported:1;
49 	bool singleton:1;
50 	unsigned integrity_added:1;
51 
52 	/*
53 	 * Indicates the rw permissions for the new logical
54 	 * device.  This should be a combination of FMODE_READ
55 	 * and FMODE_WRITE.
56 	 */
57 	fmode_t mode;
58 
59 	/* a list of devices used by this table */
60 	struct list_head devices;
61 
62 	/* events get handed up using this callback */
63 	void (*event_fn)(void *);
64 	void *event_context;
65 
66 	struct dm_md_mempools *mempools;
67 
68 	struct list_head target_callbacks;
69 };
70 
71 /*
72  * Similar to ceiling(log_size(n))
73  */
74 static unsigned int int_log(unsigned int n, unsigned int base)
75 {
76 	int result = 0;
77 
78 	while (n > 1) {
79 		n = dm_div_up(n, base);
80 		result++;
81 	}
82 
83 	return result;
84 }
85 
86 /*
87  * Calculate the index of the child node of the n'th node k'th key.
88  */
89 static inline unsigned int get_child(unsigned int n, unsigned int k)
90 {
91 	return (n * CHILDREN_PER_NODE) + k;
92 }
93 
94 /*
95  * Return the n'th node of level l from table t.
96  */
97 static inline sector_t *get_node(struct dm_table *t,
98 				 unsigned int l, unsigned int n)
99 {
100 	return t->index[l] + (n * KEYS_PER_NODE);
101 }
102 
103 /*
104  * Return the highest key that you could lookup from the n'th
105  * node on level l of the btree.
106  */
107 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
108 {
109 	for (; l < t->depth - 1; l++)
110 		n = get_child(n, CHILDREN_PER_NODE - 1);
111 
112 	if (n >= t->counts[l])
113 		return (sector_t) - 1;
114 
115 	return get_node(t, l, n)[KEYS_PER_NODE - 1];
116 }
117 
118 /*
119  * Fills in a level of the btree based on the highs of the level
120  * below it.
121  */
122 static int setup_btree_index(unsigned int l, struct dm_table *t)
123 {
124 	unsigned int n, k;
125 	sector_t *node;
126 
127 	for (n = 0U; n < t->counts[l]; n++) {
128 		node = get_node(t, l, n);
129 
130 		for (k = 0U; k < KEYS_PER_NODE; k++)
131 			node[k] = high(t, l + 1, get_child(n, k));
132 	}
133 
134 	return 0;
135 }
136 
137 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
138 {
139 	unsigned long size;
140 	void *addr;
141 
142 	/*
143 	 * Check that we're not going to overflow.
144 	 */
145 	if (nmemb > (ULONG_MAX / elem_size))
146 		return NULL;
147 
148 	size = nmemb * elem_size;
149 	addr = vzalloc(size);
150 
151 	return addr;
152 }
153 EXPORT_SYMBOL(dm_vcalloc);
154 
155 /*
156  * highs, and targets are managed as dynamic arrays during a
157  * table load.
158  */
159 static int alloc_targets(struct dm_table *t, unsigned int num)
160 {
161 	sector_t *n_highs;
162 	struct dm_target *n_targets;
163 
164 	/*
165 	 * Allocate both the target array and offset array at once.
166 	 * Append an empty entry to catch sectors beyond the end of
167 	 * the device.
168 	 */
169 	n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
170 					  sizeof(sector_t));
171 	if (!n_highs)
172 		return -ENOMEM;
173 
174 	n_targets = (struct dm_target *) (n_highs + num);
175 
176 	memset(n_highs, -1, sizeof(*n_highs) * num);
177 	vfree(t->highs);
178 
179 	t->num_allocated = num;
180 	t->highs = n_highs;
181 	t->targets = n_targets;
182 
183 	return 0;
184 }
185 
186 int dm_table_create(struct dm_table **result, fmode_t mode,
187 		    unsigned num_targets, struct mapped_device *md)
188 {
189 	struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
190 
191 	if (!t)
192 		return -ENOMEM;
193 
194 	INIT_LIST_HEAD(&t->devices);
195 	INIT_LIST_HEAD(&t->target_callbacks);
196 
197 	if (!num_targets)
198 		num_targets = KEYS_PER_NODE;
199 
200 	num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
201 
202 	if (!num_targets) {
203 		kfree(t);
204 		return -ENOMEM;
205 	}
206 
207 	if (alloc_targets(t, num_targets)) {
208 		kfree(t);
209 		return -ENOMEM;
210 	}
211 
212 	t->type = DM_TYPE_NONE;
213 	t->mode = mode;
214 	t->md = md;
215 	*result = t;
216 	return 0;
217 }
218 
219 static void free_devices(struct list_head *devices, struct mapped_device *md)
220 {
221 	struct list_head *tmp, *next;
222 
223 	list_for_each_safe(tmp, next, devices) {
224 		struct dm_dev_internal *dd =
225 		    list_entry(tmp, struct dm_dev_internal, list);
226 		DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
227 		       dm_device_name(md), dd->dm_dev->name);
228 		dm_put_table_device(md, dd->dm_dev);
229 		kfree(dd);
230 	}
231 }
232 
233 void dm_table_destroy(struct dm_table *t)
234 {
235 	unsigned int i;
236 
237 	if (!t)
238 		return;
239 
240 	/* free the indexes */
241 	if (t->depth >= 2)
242 		vfree(t->index[t->depth - 2]);
243 
244 	/* free the targets */
245 	for (i = 0; i < t->num_targets; i++) {
246 		struct dm_target *tgt = t->targets + i;
247 
248 		if (tgt->type->dtr)
249 			tgt->type->dtr(tgt);
250 
251 		dm_put_target_type(tgt->type);
252 	}
253 
254 	vfree(t->highs);
255 
256 	/* free the device list */
257 	free_devices(&t->devices, t->md);
258 
259 	dm_free_md_mempools(t->mempools);
260 
261 	kfree(t);
262 }
263 
264 /*
265  * See if we've already got a device in the list.
266  */
267 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
268 {
269 	struct dm_dev_internal *dd;
270 
271 	list_for_each_entry (dd, l, list)
272 		if (dd->dm_dev->bdev->bd_dev == dev)
273 			return dd;
274 
275 	return NULL;
276 }
277 
278 /*
279  * If possible, this checks an area of a destination device is invalid.
280  */
281 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
282 				  sector_t start, sector_t len, void *data)
283 {
284 	struct request_queue *q;
285 	struct queue_limits *limits = data;
286 	struct block_device *bdev = dev->bdev;
287 	sector_t dev_size =
288 		i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
289 	unsigned short logical_block_size_sectors =
290 		limits->logical_block_size >> SECTOR_SHIFT;
291 	char b[BDEVNAME_SIZE];
292 
293 	/*
294 	 * Some devices exist without request functions,
295 	 * such as loop devices not yet bound to backing files.
296 	 * Forbid the use of such devices.
297 	 */
298 	q = bdev_get_queue(bdev);
299 	if (!q || !q->make_request_fn) {
300 		DMWARN("%s: %s is not yet initialised: "
301 		       "start=%llu, len=%llu, dev_size=%llu",
302 		       dm_device_name(ti->table->md), bdevname(bdev, b),
303 		       (unsigned long long)start,
304 		       (unsigned long long)len,
305 		       (unsigned long long)dev_size);
306 		return 1;
307 	}
308 
309 	if (!dev_size)
310 		return 0;
311 
312 	if ((start >= dev_size) || (start + len > dev_size)) {
313 		DMWARN("%s: %s too small for target: "
314 		       "start=%llu, len=%llu, dev_size=%llu",
315 		       dm_device_name(ti->table->md), bdevname(bdev, b),
316 		       (unsigned long long)start,
317 		       (unsigned long long)len,
318 		       (unsigned long long)dev_size);
319 		return 1;
320 	}
321 
322 	/*
323 	 * If the target is mapped to zoned block device(s), check
324 	 * that the zones are not partially mapped.
325 	 */
326 	if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) {
327 		unsigned int zone_sectors = bdev_zone_sectors(bdev);
328 
329 		if (start & (zone_sectors - 1)) {
330 			DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
331 			       dm_device_name(ti->table->md),
332 			       (unsigned long long)start,
333 			       zone_sectors, bdevname(bdev, b));
334 			return 1;
335 		}
336 
337 		/*
338 		 * Note: The last zone of a zoned block device may be smaller
339 		 * than other zones. So for a target mapping the end of a
340 		 * zoned block device with such a zone, len would not be zone
341 		 * aligned. We do not allow such last smaller zone to be part
342 		 * of the mapping here to ensure that mappings with multiple
343 		 * devices do not end up with a smaller zone in the middle of
344 		 * the sector range.
345 		 */
346 		if (len & (zone_sectors - 1)) {
347 			DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
348 			       dm_device_name(ti->table->md),
349 			       (unsigned long long)len,
350 			       zone_sectors, bdevname(bdev, b));
351 			return 1;
352 		}
353 	}
354 
355 	if (logical_block_size_sectors <= 1)
356 		return 0;
357 
358 	if (start & (logical_block_size_sectors - 1)) {
359 		DMWARN("%s: start=%llu not aligned to h/w "
360 		       "logical block size %u of %s",
361 		       dm_device_name(ti->table->md),
362 		       (unsigned long long)start,
363 		       limits->logical_block_size, bdevname(bdev, b));
364 		return 1;
365 	}
366 
367 	if (len & (logical_block_size_sectors - 1)) {
368 		DMWARN("%s: len=%llu not aligned to h/w "
369 		       "logical block size %u of %s",
370 		       dm_device_name(ti->table->md),
371 		       (unsigned long long)len,
372 		       limits->logical_block_size, bdevname(bdev, b));
373 		return 1;
374 	}
375 
376 	return 0;
377 }
378 
379 /*
380  * This upgrades the mode on an already open dm_dev, being
381  * careful to leave things as they were if we fail to reopen the
382  * device and not to touch the existing bdev field in case
383  * it is accessed concurrently inside dm_table_any_congested().
384  */
385 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
386 			struct mapped_device *md)
387 {
388 	int r;
389 	struct dm_dev *old_dev, *new_dev;
390 
391 	old_dev = dd->dm_dev;
392 
393 	r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
394 				dd->dm_dev->mode | new_mode, &new_dev);
395 	if (r)
396 		return r;
397 
398 	dd->dm_dev = new_dev;
399 	dm_put_table_device(md, old_dev);
400 
401 	return 0;
402 }
403 
404 /*
405  * Convert the path to a device
406  */
407 dev_t dm_get_dev_t(const char *path)
408 {
409 	dev_t dev;
410 	struct block_device *bdev;
411 
412 	bdev = lookup_bdev(path);
413 	if (IS_ERR(bdev))
414 		dev = name_to_dev_t(path);
415 	else {
416 		dev = bdev->bd_dev;
417 		bdput(bdev);
418 	}
419 
420 	return dev;
421 }
422 EXPORT_SYMBOL_GPL(dm_get_dev_t);
423 
424 /*
425  * Add a device to the list, or just increment the usage count if
426  * it's already present.
427  */
428 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
429 		  struct dm_dev **result)
430 {
431 	int r;
432 	dev_t dev;
433 	struct dm_dev_internal *dd;
434 	struct dm_table *t = ti->table;
435 
436 	BUG_ON(!t);
437 
438 	dev = dm_get_dev_t(path);
439 	if (!dev)
440 		return -ENODEV;
441 
442 	dd = find_device(&t->devices, dev);
443 	if (!dd) {
444 		dd = kmalloc(sizeof(*dd), GFP_KERNEL);
445 		if (!dd)
446 			return -ENOMEM;
447 
448 		if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
449 			kfree(dd);
450 			return r;
451 		}
452 
453 		refcount_set(&dd->count, 1);
454 		list_add(&dd->list, &t->devices);
455 		goto out;
456 
457 	} else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
458 		r = upgrade_mode(dd, mode, t->md);
459 		if (r)
460 			return r;
461 	}
462 	refcount_inc(&dd->count);
463 out:
464 	*result = dd->dm_dev;
465 	return 0;
466 }
467 EXPORT_SYMBOL(dm_get_device);
468 
469 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
470 				sector_t start, sector_t len, void *data)
471 {
472 	struct queue_limits *limits = data;
473 	struct block_device *bdev = dev->bdev;
474 	struct request_queue *q = bdev_get_queue(bdev);
475 	char b[BDEVNAME_SIZE];
476 
477 	if (unlikely(!q)) {
478 		DMWARN("%s: Cannot set limits for nonexistent device %s",
479 		       dm_device_name(ti->table->md), bdevname(bdev, b));
480 		return 0;
481 	}
482 
483 	if (bdev_stack_limits(limits, bdev, start) < 0)
484 		DMWARN("%s: adding target device %s caused an alignment inconsistency: "
485 		       "physical_block_size=%u, logical_block_size=%u, "
486 		       "alignment_offset=%u, start=%llu",
487 		       dm_device_name(ti->table->md), bdevname(bdev, b),
488 		       q->limits.physical_block_size,
489 		       q->limits.logical_block_size,
490 		       q->limits.alignment_offset,
491 		       (unsigned long long) start << SECTOR_SHIFT);
492 
493 	limits->zoned = blk_queue_zoned_model(q);
494 
495 	return 0;
496 }
497 
498 /*
499  * Decrement a device's use count and remove it if necessary.
500  */
501 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
502 {
503 	int found = 0;
504 	struct list_head *devices = &ti->table->devices;
505 	struct dm_dev_internal *dd;
506 
507 	list_for_each_entry(dd, devices, list) {
508 		if (dd->dm_dev == d) {
509 			found = 1;
510 			break;
511 		}
512 	}
513 	if (!found) {
514 		DMWARN("%s: device %s not in table devices list",
515 		       dm_device_name(ti->table->md), d->name);
516 		return;
517 	}
518 	if (refcount_dec_and_test(&dd->count)) {
519 		dm_put_table_device(ti->table->md, d);
520 		list_del(&dd->list);
521 		kfree(dd);
522 	}
523 }
524 EXPORT_SYMBOL(dm_put_device);
525 
526 /*
527  * Checks to see if the target joins onto the end of the table.
528  */
529 static int adjoin(struct dm_table *table, struct dm_target *ti)
530 {
531 	struct dm_target *prev;
532 
533 	if (!table->num_targets)
534 		return !ti->begin;
535 
536 	prev = &table->targets[table->num_targets - 1];
537 	return (ti->begin == (prev->begin + prev->len));
538 }
539 
540 /*
541  * Used to dynamically allocate the arg array.
542  *
543  * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
544  * process messages even if some device is suspended. These messages have a
545  * small fixed number of arguments.
546  *
547  * On the other hand, dm-switch needs to process bulk data using messages and
548  * excessive use of GFP_NOIO could cause trouble.
549  */
550 static char **realloc_argv(unsigned *size, char **old_argv)
551 {
552 	char **argv;
553 	unsigned new_size;
554 	gfp_t gfp;
555 
556 	if (*size) {
557 		new_size = *size * 2;
558 		gfp = GFP_KERNEL;
559 	} else {
560 		new_size = 8;
561 		gfp = GFP_NOIO;
562 	}
563 	argv = kmalloc_array(new_size, sizeof(*argv), gfp);
564 	if (argv && old_argv) {
565 		memcpy(argv, old_argv, *size * sizeof(*argv));
566 		*size = new_size;
567 	}
568 
569 	kfree(old_argv);
570 	return argv;
571 }
572 
573 /*
574  * Destructively splits up the argument list to pass to ctr.
575  */
576 int dm_split_args(int *argc, char ***argvp, char *input)
577 {
578 	char *start, *end = input, *out, **argv = NULL;
579 	unsigned array_size = 0;
580 
581 	*argc = 0;
582 
583 	if (!input) {
584 		*argvp = NULL;
585 		return 0;
586 	}
587 
588 	argv = realloc_argv(&array_size, argv);
589 	if (!argv)
590 		return -ENOMEM;
591 
592 	while (1) {
593 		/* Skip whitespace */
594 		start = skip_spaces(end);
595 
596 		if (!*start)
597 			break;	/* success, we hit the end */
598 
599 		/* 'out' is used to remove any back-quotes */
600 		end = out = start;
601 		while (*end) {
602 			/* Everything apart from '\0' can be quoted */
603 			if (*end == '\\' && *(end + 1)) {
604 				*out++ = *(end + 1);
605 				end += 2;
606 				continue;
607 			}
608 
609 			if (isspace(*end))
610 				break;	/* end of token */
611 
612 			*out++ = *end++;
613 		}
614 
615 		/* have we already filled the array ? */
616 		if ((*argc + 1) > array_size) {
617 			argv = realloc_argv(&array_size, argv);
618 			if (!argv)
619 				return -ENOMEM;
620 		}
621 
622 		/* we know this is whitespace */
623 		if (*end)
624 			end++;
625 
626 		/* terminate the string and put it in the array */
627 		*out = '\0';
628 		argv[*argc] = start;
629 		(*argc)++;
630 	}
631 
632 	*argvp = argv;
633 	return 0;
634 }
635 
636 /*
637  * Impose necessary and sufficient conditions on a devices's table such
638  * that any incoming bio which respects its logical_block_size can be
639  * processed successfully.  If it falls across the boundary between
640  * two or more targets, the size of each piece it gets split into must
641  * be compatible with the logical_block_size of the target processing it.
642  */
643 static int validate_hardware_logical_block_alignment(struct dm_table *table,
644 						 struct queue_limits *limits)
645 {
646 	/*
647 	 * This function uses arithmetic modulo the logical_block_size
648 	 * (in units of 512-byte sectors).
649 	 */
650 	unsigned short device_logical_block_size_sects =
651 		limits->logical_block_size >> SECTOR_SHIFT;
652 
653 	/*
654 	 * Offset of the start of the next table entry, mod logical_block_size.
655 	 */
656 	unsigned short next_target_start = 0;
657 
658 	/*
659 	 * Given an aligned bio that extends beyond the end of a
660 	 * target, how many sectors must the next target handle?
661 	 */
662 	unsigned short remaining = 0;
663 
664 	struct dm_target *uninitialized_var(ti);
665 	struct queue_limits ti_limits;
666 	unsigned i;
667 
668 	/*
669 	 * Check each entry in the table in turn.
670 	 */
671 	for (i = 0; i < dm_table_get_num_targets(table); i++) {
672 		ti = dm_table_get_target(table, i);
673 
674 		blk_set_stacking_limits(&ti_limits);
675 
676 		/* combine all target devices' limits */
677 		if (ti->type->iterate_devices)
678 			ti->type->iterate_devices(ti, dm_set_device_limits,
679 						  &ti_limits);
680 
681 		/*
682 		 * If the remaining sectors fall entirely within this
683 		 * table entry are they compatible with its logical_block_size?
684 		 */
685 		if (remaining < ti->len &&
686 		    remaining & ((ti_limits.logical_block_size >>
687 				  SECTOR_SHIFT) - 1))
688 			break;	/* Error */
689 
690 		next_target_start =
691 		    (unsigned short) ((next_target_start + ti->len) &
692 				      (device_logical_block_size_sects - 1));
693 		remaining = next_target_start ?
694 		    device_logical_block_size_sects - next_target_start : 0;
695 	}
696 
697 	if (remaining) {
698 		DMWARN("%s: table line %u (start sect %llu len %llu) "
699 		       "not aligned to h/w logical block size %u",
700 		       dm_device_name(table->md), i,
701 		       (unsigned long long) ti->begin,
702 		       (unsigned long long) ti->len,
703 		       limits->logical_block_size);
704 		return -EINVAL;
705 	}
706 
707 	return 0;
708 }
709 
710 int dm_table_add_target(struct dm_table *t, const char *type,
711 			sector_t start, sector_t len, char *params)
712 {
713 	int r = -EINVAL, argc;
714 	char **argv;
715 	struct dm_target *tgt;
716 
717 	if (t->singleton) {
718 		DMERR("%s: target type %s must appear alone in table",
719 		      dm_device_name(t->md), t->targets->type->name);
720 		return -EINVAL;
721 	}
722 
723 	BUG_ON(t->num_targets >= t->num_allocated);
724 
725 	tgt = t->targets + t->num_targets;
726 	memset(tgt, 0, sizeof(*tgt));
727 
728 	if (!len) {
729 		DMERR("%s: zero-length target", dm_device_name(t->md));
730 		return -EINVAL;
731 	}
732 
733 	tgt->type = dm_get_target_type(type);
734 	if (!tgt->type) {
735 		DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
736 		return -EINVAL;
737 	}
738 
739 	if (dm_target_needs_singleton(tgt->type)) {
740 		if (t->num_targets) {
741 			tgt->error = "singleton target type must appear alone in table";
742 			goto bad;
743 		}
744 		t->singleton = true;
745 	}
746 
747 	if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
748 		tgt->error = "target type may not be included in a read-only table";
749 		goto bad;
750 	}
751 
752 	if (t->immutable_target_type) {
753 		if (t->immutable_target_type != tgt->type) {
754 			tgt->error = "immutable target type cannot be mixed with other target types";
755 			goto bad;
756 		}
757 	} else if (dm_target_is_immutable(tgt->type)) {
758 		if (t->num_targets) {
759 			tgt->error = "immutable target type cannot be mixed with other target types";
760 			goto bad;
761 		}
762 		t->immutable_target_type = tgt->type;
763 	}
764 
765 	if (dm_target_has_integrity(tgt->type))
766 		t->integrity_added = 1;
767 
768 	tgt->table = t;
769 	tgt->begin = start;
770 	tgt->len = len;
771 	tgt->error = "Unknown error";
772 
773 	/*
774 	 * Does this target adjoin the previous one ?
775 	 */
776 	if (!adjoin(t, tgt)) {
777 		tgt->error = "Gap in table";
778 		goto bad;
779 	}
780 
781 	r = dm_split_args(&argc, &argv, params);
782 	if (r) {
783 		tgt->error = "couldn't split parameters (insufficient memory)";
784 		goto bad;
785 	}
786 
787 	r = tgt->type->ctr(tgt, argc, argv);
788 	kfree(argv);
789 	if (r)
790 		goto bad;
791 
792 	t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
793 
794 	if (!tgt->num_discard_bios && tgt->discards_supported)
795 		DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
796 		       dm_device_name(t->md), type);
797 
798 	return 0;
799 
800  bad:
801 	DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
802 	dm_put_target_type(tgt->type);
803 	return r;
804 }
805 
806 /*
807  * Target argument parsing helpers.
808  */
809 static int validate_next_arg(const struct dm_arg *arg,
810 			     struct dm_arg_set *arg_set,
811 			     unsigned *value, char **error, unsigned grouped)
812 {
813 	const char *arg_str = dm_shift_arg(arg_set);
814 	char dummy;
815 
816 	if (!arg_str ||
817 	    (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
818 	    (*value < arg->min) ||
819 	    (*value > arg->max) ||
820 	    (grouped && arg_set->argc < *value)) {
821 		*error = arg->error;
822 		return -EINVAL;
823 	}
824 
825 	return 0;
826 }
827 
828 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
829 		unsigned *value, char **error)
830 {
831 	return validate_next_arg(arg, arg_set, value, error, 0);
832 }
833 EXPORT_SYMBOL(dm_read_arg);
834 
835 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
836 		      unsigned *value, char **error)
837 {
838 	return validate_next_arg(arg, arg_set, value, error, 1);
839 }
840 EXPORT_SYMBOL(dm_read_arg_group);
841 
842 const char *dm_shift_arg(struct dm_arg_set *as)
843 {
844 	char *r;
845 
846 	if (as->argc) {
847 		as->argc--;
848 		r = *as->argv;
849 		as->argv++;
850 		return r;
851 	}
852 
853 	return NULL;
854 }
855 EXPORT_SYMBOL(dm_shift_arg);
856 
857 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
858 {
859 	BUG_ON(as->argc < num_args);
860 	as->argc -= num_args;
861 	as->argv += num_args;
862 }
863 EXPORT_SYMBOL(dm_consume_args);
864 
865 static bool __table_type_bio_based(enum dm_queue_mode table_type)
866 {
867 	return (table_type == DM_TYPE_BIO_BASED ||
868 		table_type == DM_TYPE_DAX_BIO_BASED ||
869 		table_type == DM_TYPE_NVME_BIO_BASED);
870 }
871 
872 static bool __table_type_request_based(enum dm_queue_mode table_type)
873 {
874 	return table_type == DM_TYPE_REQUEST_BASED;
875 }
876 
877 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
878 {
879 	t->type = type;
880 }
881 EXPORT_SYMBOL_GPL(dm_table_set_type);
882 
883 /* validate the dax capability of the target device span */
884 int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
885 				       sector_t start, sector_t len, void *data)
886 {
887 	int blocksize = *(int *) data;
888 
889 	return generic_fsdax_supported(dev->dax_dev, dev->bdev, blocksize,
890 			start, len);
891 }
892 
893 /* Check devices support synchronous DAX */
894 static int device_synchronous(struct dm_target *ti, struct dm_dev *dev,
895 				       sector_t start, sector_t len, void *data)
896 {
897 	return dax_synchronous(dev->dax_dev);
898 }
899 
900 bool dm_table_supports_dax(struct dm_table *t,
901 			  iterate_devices_callout_fn iterate_fn, int *blocksize)
902 {
903 	struct dm_target *ti;
904 	unsigned i;
905 
906 	/* Ensure that all targets support DAX. */
907 	for (i = 0; i < dm_table_get_num_targets(t); i++) {
908 		ti = dm_table_get_target(t, i);
909 
910 		if (!ti->type->direct_access)
911 			return false;
912 
913 		if (!ti->type->iterate_devices ||
914 			!ti->type->iterate_devices(ti, iterate_fn, blocksize))
915 			return false;
916 	}
917 
918 	return true;
919 }
920 
921 static bool dm_table_does_not_support_partial_completion(struct dm_table *t);
922 
923 struct verify_rq_based_data {
924 	unsigned sq_count;
925 	unsigned mq_count;
926 };
927 
928 static int device_is_rq_based(struct dm_target *ti, struct dm_dev *dev,
929 			      sector_t start, sector_t len, void *data)
930 {
931 	struct request_queue *q = bdev_get_queue(dev->bdev);
932 	struct verify_rq_based_data *v = data;
933 
934 	if (queue_is_mq(q))
935 		v->mq_count++;
936 	else
937 		v->sq_count++;
938 
939 	return queue_is_mq(q);
940 }
941 
942 static int dm_table_determine_type(struct dm_table *t)
943 {
944 	unsigned i;
945 	unsigned bio_based = 0, request_based = 0, hybrid = 0;
946 	struct verify_rq_based_data v = {.sq_count = 0, .mq_count = 0};
947 	struct dm_target *tgt;
948 	struct list_head *devices = dm_table_get_devices(t);
949 	enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
950 	int page_size = PAGE_SIZE;
951 
952 	if (t->type != DM_TYPE_NONE) {
953 		/* target already set the table's type */
954 		if (t->type == DM_TYPE_BIO_BASED) {
955 			/* possibly upgrade to a variant of bio-based */
956 			goto verify_bio_based;
957 		}
958 		BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
959 		BUG_ON(t->type == DM_TYPE_NVME_BIO_BASED);
960 		goto verify_rq_based;
961 	}
962 
963 	for (i = 0; i < t->num_targets; i++) {
964 		tgt = t->targets + i;
965 		if (dm_target_hybrid(tgt))
966 			hybrid = 1;
967 		else if (dm_target_request_based(tgt))
968 			request_based = 1;
969 		else
970 			bio_based = 1;
971 
972 		if (bio_based && request_based) {
973 			DMERR("Inconsistent table: different target types"
974 			      " can't be mixed up");
975 			return -EINVAL;
976 		}
977 	}
978 
979 	if (hybrid && !bio_based && !request_based) {
980 		/*
981 		 * The targets can work either way.
982 		 * Determine the type from the live device.
983 		 * Default to bio-based if device is new.
984 		 */
985 		if (__table_type_request_based(live_md_type))
986 			request_based = 1;
987 		else
988 			bio_based = 1;
989 	}
990 
991 	if (bio_based) {
992 verify_bio_based:
993 		/* We must use this table as bio-based */
994 		t->type = DM_TYPE_BIO_BASED;
995 		if (dm_table_supports_dax(t, device_supports_dax, &page_size) ||
996 		    (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
997 			t->type = DM_TYPE_DAX_BIO_BASED;
998 		} else {
999 			/* Check if upgrading to NVMe bio-based is valid or required */
1000 			tgt = dm_table_get_immutable_target(t);
1001 			if (tgt && !tgt->max_io_len && dm_table_does_not_support_partial_completion(t)) {
1002 				t->type = DM_TYPE_NVME_BIO_BASED;
1003 				goto verify_rq_based; /* must be stacked directly on NVMe (blk-mq) */
1004 			} else if (list_empty(devices) && live_md_type == DM_TYPE_NVME_BIO_BASED) {
1005 				t->type = DM_TYPE_NVME_BIO_BASED;
1006 			}
1007 		}
1008 		return 0;
1009 	}
1010 
1011 	BUG_ON(!request_based); /* No targets in this table */
1012 
1013 	t->type = DM_TYPE_REQUEST_BASED;
1014 
1015 verify_rq_based:
1016 	/*
1017 	 * Request-based dm supports only tables that have a single target now.
1018 	 * To support multiple targets, request splitting support is needed,
1019 	 * and that needs lots of changes in the block-layer.
1020 	 * (e.g. request completion process for partial completion.)
1021 	 */
1022 	if (t->num_targets > 1) {
1023 		DMERR("%s DM doesn't support multiple targets",
1024 		      t->type == DM_TYPE_NVME_BIO_BASED ? "nvme bio-based" : "request-based");
1025 		return -EINVAL;
1026 	}
1027 
1028 	if (list_empty(devices)) {
1029 		int srcu_idx;
1030 		struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
1031 
1032 		/* inherit live table's type */
1033 		if (live_table)
1034 			t->type = live_table->type;
1035 		dm_put_live_table(t->md, srcu_idx);
1036 		return 0;
1037 	}
1038 
1039 	tgt = dm_table_get_immutable_target(t);
1040 	if (!tgt) {
1041 		DMERR("table load rejected: immutable target is required");
1042 		return -EINVAL;
1043 	} else if (tgt->max_io_len) {
1044 		DMERR("table load rejected: immutable target that splits IO is not supported");
1045 		return -EINVAL;
1046 	}
1047 
1048 	/* Non-request-stackable devices can't be used for request-based dm */
1049 	if (!tgt->type->iterate_devices ||
1050 	    !tgt->type->iterate_devices(tgt, device_is_rq_based, &v)) {
1051 		DMERR("table load rejected: including non-request-stackable devices");
1052 		return -EINVAL;
1053 	}
1054 	if (v.sq_count > 0) {
1055 		DMERR("table load rejected: not all devices are blk-mq request-stackable");
1056 		return -EINVAL;
1057 	}
1058 
1059 	return 0;
1060 }
1061 
1062 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
1063 {
1064 	return t->type;
1065 }
1066 
1067 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1068 {
1069 	return t->immutable_target_type;
1070 }
1071 
1072 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1073 {
1074 	/* Immutable target is implicitly a singleton */
1075 	if (t->num_targets > 1 ||
1076 	    !dm_target_is_immutable(t->targets[0].type))
1077 		return NULL;
1078 
1079 	return t->targets;
1080 }
1081 
1082 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1083 {
1084 	struct dm_target *ti;
1085 	unsigned i;
1086 
1087 	for (i = 0; i < dm_table_get_num_targets(t); i++) {
1088 		ti = dm_table_get_target(t, i);
1089 		if (dm_target_is_wildcard(ti->type))
1090 			return ti;
1091 	}
1092 
1093 	return NULL;
1094 }
1095 
1096 bool dm_table_bio_based(struct dm_table *t)
1097 {
1098 	return __table_type_bio_based(dm_table_get_type(t));
1099 }
1100 
1101 bool dm_table_request_based(struct dm_table *t)
1102 {
1103 	return __table_type_request_based(dm_table_get_type(t));
1104 }
1105 
1106 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1107 {
1108 	enum dm_queue_mode type = dm_table_get_type(t);
1109 	unsigned per_io_data_size = 0;
1110 	unsigned min_pool_size = 0;
1111 	struct dm_target *ti;
1112 	unsigned i;
1113 
1114 	if (unlikely(type == DM_TYPE_NONE)) {
1115 		DMWARN("no table type is set, can't allocate mempools");
1116 		return -EINVAL;
1117 	}
1118 
1119 	if (__table_type_bio_based(type))
1120 		for (i = 0; i < t->num_targets; i++) {
1121 			ti = t->targets + i;
1122 			per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1123 			min_pool_size = max(min_pool_size, ti->num_flush_bios);
1124 		}
1125 
1126 	t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1127 					   per_io_data_size, min_pool_size);
1128 	if (!t->mempools)
1129 		return -ENOMEM;
1130 
1131 	return 0;
1132 }
1133 
1134 void dm_table_free_md_mempools(struct dm_table *t)
1135 {
1136 	dm_free_md_mempools(t->mempools);
1137 	t->mempools = NULL;
1138 }
1139 
1140 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1141 {
1142 	return t->mempools;
1143 }
1144 
1145 static int setup_indexes(struct dm_table *t)
1146 {
1147 	int i;
1148 	unsigned int total = 0;
1149 	sector_t *indexes;
1150 
1151 	/* allocate the space for *all* the indexes */
1152 	for (i = t->depth - 2; i >= 0; i--) {
1153 		t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1154 		total += t->counts[i];
1155 	}
1156 
1157 	indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1158 	if (!indexes)
1159 		return -ENOMEM;
1160 
1161 	/* set up internal nodes, bottom-up */
1162 	for (i = t->depth - 2; i >= 0; i--) {
1163 		t->index[i] = indexes;
1164 		indexes += (KEYS_PER_NODE * t->counts[i]);
1165 		setup_btree_index(i, t);
1166 	}
1167 
1168 	return 0;
1169 }
1170 
1171 /*
1172  * Builds the btree to index the map.
1173  */
1174 static int dm_table_build_index(struct dm_table *t)
1175 {
1176 	int r = 0;
1177 	unsigned int leaf_nodes;
1178 
1179 	/* how many indexes will the btree have ? */
1180 	leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1181 	t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1182 
1183 	/* leaf layer has already been set up */
1184 	t->counts[t->depth - 1] = leaf_nodes;
1185 	t->index[t->depth - 1] = t->highs;
1186 
1187 	if (t->depth >= 2)
1188 		r = setup_indexes(t);
1189 
1190 	return r;
1191 }
1192 
1193 static bool integrity_profile_exists(struct gendisk *disk)
1194 {
1195 	return !!blk_get_integrity(disk);
1196 }
1197 
1198 /*
1199  * Get a disk whose integrity profile reflects the table's profile.
1200  * Returns NULL if integrity support was inconsistent or unavailable.
1201  */
1202 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1203 {
1204 	struct list_head *devices = dm_table_get_devices(t);
1205 	struct dm_dev_internal *dd = NULL;
1206 	struct gendisk *prev_disk = NULL, *template_disk = NULL;
1207 	unsigned i;
1208 
1209 	for (i = 0; i < dm_table_get_num_targets(t); i++) {
1210 		struct dm_target *ti = dm_table_get_target(t, i);
1211 		if (!dm_target_passes_integrity(ti->type))
1212 			goto no_integrity;
1213 	}
1214 
1215 	list_for_each_entry(dd, devices, list) {
1216 		template_disk = dd->dm_dev->bdev->bd_disk;
1217 		if (!integrity_profile_exists(template_disk))
1218 			goto no_integrity;
1219 		else if (prev_disk &&
1220 			 blk_integrity_compare(prev_disk, template_disk) < 0)
1221 			goto no_integrity;
1222 		prev_disk = template_disk;
1223 	}
1224 
1225 	return template_disk;
1226 
1227 no_integrity:
1228 	if (prev_disk)
1229 		DMWARN("%s: integrity not set: %s and %s profile mismatch",
1230 		       dm_device_name(t->md),
1231 		       prev_disk->disk_name,
1232 		       template_disk->disk_name);
1233 	return NULL;
1234 }
1235 
1236 /*
1237  * Register the mapped device for blk_integrity support if the
1238  * underlying devices have an integrity profile.  But all devices may
1239  * not have matching profiles (checking all devices isn't reliable
1240  * during table load because this table may use other DM device(s) which
1241  * must be resumed before they will have an initialized integity
1242  * profile).  Consequently, stacked DM devices force a 2 stage integrity
1243  * profile validation: First pass during table load, final pass during
1244  * resume.
1245  */
1246 static int dm_table_register_integrity(struct dm_table *t)
1247 {
1248 	struct mapped_device *md = t->md;
1249 	struct gendisk *template_disk = NULL;
1250 
1251 	/* If target handles integrity itself do not register it here. */
1252 	if (t->integrity_added)
1253 		return 0;
1254 
1255 	template_disk = dm_table_get_integrity_disk(t);
1256 	if (!template_disk)
1257 		return 0;
1258 
1259 	if (!integrity_profile_exists(dm_disk(md))) {
1260 		t->integrity_supported = true;
1261 		/*
1262 		 * Register integrity profile during table load; we can do
1263 		 * this because the final profile must match during resume.
1264 		 */
1265 		blk_integrity_register(dm_disk(md),
1266 				       blk_get_integrity(template_disk));
1267 		return 0;
1268 	}
1269 
1270 	/*
1271 	 * If DM device already has an initialized integrity
1272 	 * profile the new profile should not conflict.
1273 	 */
1274 	if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1275 		DMWARN("%s: conflict with existing integrity profile: "
1276 		       "%s profile mismatch",
1277 		       dm_device_name(t->md),
1278 		       template_disk->disk_name);
1279 		return 1;
1280 	}
1281 
1282 	/* Preserve existing integrity profile */
1283 	t->integrity_supported = true;
1284 	return 0;
1285 }
1286 
1287 /*
1288  * Prepares the table for use by building the indices,
1289  * setting the type, and allocating mempools.
1290  */
1291 int dm_table_complete(struct dm_table *t)
1292 {
1293 	int r;
1294 
1295 	r = dm_table_determine_type(t);
1296 	if (r) {
1297 		DMERR("unable to determine table type");
1298 		return r;
1299 	}
1300 
1301 	r = dm_table_build_index(t);
1302 	if (r) {
1303 		DMERR("unable to build btrees");
1304 		return r;
1305 	}
1306 
1307 	r = dm_table_register_integrity(t);
1308 	if (r) {
1309 		DMERR("could not register integrity profile.");
1310 		return r;
1311 	}
1312 
1313 	r = dm_table_alloc_md_mempools(t, t->md);
1314 	if (r)
1315 		DMERR("unable to allocate mempools");
1316 
1317 	return r;
1318 }
1319 
1320 static DEFINE_MUTEX(_event_lock);
1321 void dm_table_event_callback(struct dm_table *t,
1322 			     void (*fn)(void *), void *context)
1323 {
1324 	mutex_lock(&_event_lock);
1325 	t->event_fn = fn;
1326 	t->event_context = context;
1327 	mutex_unlock(&_event_lock);
1328 }
1329 
1330 void dm_table_event(struct dm_table *t)
1331 {
1332 	/*
1333 	 * You can no longer call dm_table_event() from interrupt
1334 	 * context, use a bottom half instead.
1335 	 */
1336 	BUG_ON(in_interrupt());
1337 
1338 	mutex_lock(&_event_lock);
1339 	if (t->event_fn)
1340 		t->event_fn(t->event_context);
1341 	mutex_unlock(&_event_lock);
1342 }
1343 EXPORT_SYMBOL(dm_table_event);
1344 
1345 sector_t dm_table_get_size(struct dm_table *t)
1346 {
1347 	return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1348 }
1349 EXPORT_SYMBOL(dm_table_get_size);
1350 
1351 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1352 {
1353 	if (index >= t->num_targets)
1354 		return NULL;
1355 
1356 	return t->targets + index;
1357 }
1358 
1359 /*
1360  * Search the btree for the correct target.
1361  *
1362  * Caller should check returned pointer with dm_target_is_valid()
1363  * to trap I/O beyond end of device.
1364  */
1365 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1366 {
1367 	unsigned int l, n = 0, k = 0;
1368 	sector_t *node;
1369 
1370 	for (l = 0; l < t->depth; l++) {
1371 		n = get_child(n, k);
1372 		node = get_node(t, l, n);
1373 
1374 		for (k = 0; k < KEYS_PER_NODE; k++)
1375 			if (node[k] >= sector)
1376 				break;
1377 	}
1378 
1379 	return &t->targets[(KEYS_PER_NODE * n) + k];
1380 }
1381 
1382 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1383 			sector_t start, sector_t len, void *data)
1384 {
1385 	unsigned *num_devices = data;
1386 
1387 	(*num_devices)++;
1388 
1389 	return 0;
1390 }
1391 
1392 /*
1393  * Check whether a table has no data devices attached using each
1394  * target's iterate_devices method.
1395  * Returns false if the result is unknown because a target doesn't
1396  * support iterate_devices.
1397  */
1398 bool dm_table_has_no_data_devices(struct dm_table *table)
1399 {
1400 	struct dm_target *ti;
1401 	unsigned i, num_devices;
1402 
1403 	for (i = 0; i < dm_table_get_num_targets(table); i++) {
1404 		ti = dm_table_get_target(table, i);
1405 
1406 		if (!ti->type->iterate_devices)
1407 			return false;
1408 
1409 		num_devices = 0;
1410 		ti->type->iterate_devices(ti, count_device, &num_devices);
1411 		if (num_devices)
1412 			return false;
1413 	}
1414 
1415 	return true;
1416 }
1417 
1418 static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1419 				 sector_t start, sector_t len, void *data)
1420 {
1421 	struct request_queue *q = bdev_get_queue(dev->bdev);
1422 	enum blk_zoned_model *zoned_model = data;
1423 
1424 	return q && blk_queue_zoned_model(q) == *zoned_model;
1425 }
1426 
1427 static bool dm_table_supports_zoned_model(struct dm_table *t,
1428 					  enum blk_zoned_model zoned_model)
1429 {
1430 	struct dm_target *ti;
1431 	unsigned i;
1432 
1433 	for (i = 0; i < dm_table_get_num_targets(t); i++) {
1434 		ti = dm_table_get_target(t, i);
1435 
1436 		if (zoned_model == BLK_ZONED_HM &&
1437 		    !dm_target_supports_zoned_hm(ti->type))
1438 			return false;
1439 
1440 		if (!ti->type->iterate_devices ||
1441 		    !ti->type->iterate_devices(ti, device_is_zoned_model, &zoned_model))
1442 			return false;
1443 	}
1444 
1445 	return true;
1446 }
1447 
1448 static int device_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1449 				       sector_t start, sector_t len, void *data)
1450 {
1451 	struct request_queue *q = bdev_get_queue(dev->bdev);
1452 	unsigned int *zone_sectors = data;
1453 
1454 	return q && blk_queue_zone_sectors(q) == *zone_sectors;
1455 }
1456 
1457 static bool dm_table_matches_zone_sectors(struct dm_table *t,
1458 					  unsigned int zone_sectors)
1459 {
1460 	struct dm_target *ti;
1461 	unsigned i;
1462 
1463 	for (i = 0; i < dm_table_get_num_targets(t); i++) {
1464 		ti = dm_table_get_target(t, i);
1465 
1466 		if (!ti->type->iterate_devices ||
1467 		    !ti->type->iterate_devices(ti, device_matches_zone_sectors, &zone_sectors))
1468 			return false;
1469 	}
1470 
1471 	return true;
1472 }
1473 
1474 static int validate_hardware_zoned_model(struct dm_table *table,
1475 					 enum blk_zoned_model zoned_model,
1476 					 unsigned int zone_sectors)
1477 {
1478 	if (zoned_model == BLK_ZONED_NONE)
1479 		return 0;
1480 
1481 	if (!dm_table_supports_zoned_model(table, zoned_model)) {
1482 		DMERR("%s: zoned model is not consistent across all devices",
1483 		      dm_device_name(table->md));
1484 		return -EINVAL;
1485 	}
1486 
1487 	/* Check zone size validity and compatibility */
1488 	if (!zone_sectors || !is_power_of_2(zone_sectors))
1489 		return -EINVAL;
1490 
1491 	if (!dm_table_matches_zone_sectors(table, zone_sectors)) {
1492 		DMERR("%s: zone sectors is not consistent across all devices",
1493 		      dm_device_name(table->md));
1494 		return -EINVAL;
1495 	}
1496 
1497 	return 0;
1498 }
1499 
1500 /*
1501  * Establish the new table's queue_limits and validate them.
1502  */
1503 int dm_calculate_queue_limits(struct dm_table *table,
1504 			      struct queue_limits *limits)
1505 {
1506 	struct dm_target *ti;
1507 	struct queue_limits ti_limits;
1508 	unsigned i;
1509 	enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1510 	unsigned int zone_sectors = 0;
1511 
1512 	blk_set_stacking_limits(limits);
1513 
1514 	for (i = 0; i < dm_table_get_num_targets(table); i++) {
1515 		blk_set_stacking_limits(&ti_limits);
1516 
1517 		ti = dm_table_get_target(table, i);
1518 
1519 		if (!ti->type->iterate_devices)
1520 			goto combine_limits;
1521 
1522 		/*
1523 		 * Combine queue limits of all the devices this target uses.
1524 		 */
1525 		ti->type->iterate_devices(ti, dm_set_device_limits,
1526 					  &ti_limits);
1527 
1528 		if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1529 			/*
1530 			 * After stacking all limits, validate all devices
1531 			 * in table support this zoned model and zone sectors.
1532 			 */
1533 			zoned_model = ti_limits.zoned;
1534 			zone_sectors = ti_limits.chunk_sectors;
1535 		}
1536 
1537 		/* Set I/O hints portion of queue limits */
1538 		if (ti->type->io_hints)
1539 			ti->type->io_hints(ti, &ti_limits);
1540 
1541 		/*
1542 		 * Check each device area is consistent with the target's
1543 		 * overall queue limits.
1544 		 */
1545 		if (ti->type->iterate_devices(ti, device_area_is_invalid,
1546 					      &ti_limits))
1547 			return -EINVAL;
1548 
1549 combine_limits:
1550 		/*
1551 		 * Merge this target's queue limits into the overall limits
1552 		 * for the table.
1553 		 */
1554 		if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1555 			DMWARN("%s: adding target device "
1556 			       "(start sect %llu len %llu) "
1557 			       "caused an alignment inconsistency",
1558 			       dm_device_name(table->md),
1559 			       (unsigned long long) ti->begin,
1560 			       (unsigned long long) ti->len);
1561 
1562 		/*
1563 		 * FIXME: this should likely be moved to blk_stack_limits(), would
1564 		 * also eliminate limits->zoned stacking hack in dm_set_device_limits()
1565 		 */
1566 		if (limits->zoned == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1567 			/*
1568 			 * By default, the stacked limits zoned model is set to
1569 			 * BLK_ZONED_NONE in blk_set_stacking_limits(). Update
1570 			 * this model using the first target model reported
1571 			 * that is not BLK_ZONED_NONE. This will be either the
1572 			 * first target device zoned model or the model reported
1573 			 * by the target .io_hints.
1574 			 */
1575 			limits->zoned = ti_limits.zoned;
1576 		}
1577 	}
1578 
1579 	/*
1580 	 * Verify that the zoned model and zone sectors, as determined before
1581 	 * any .io_hints override, are the same across all devices in the table.
1582 	 * - this is especially relevant if .io_hints is emulating a disk-managed
1583 	 *   zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1584 	 * BUT...
1585 	 */
1586 	if (limits->zoned != BLK_ZONED_NONE) {
1587 		/*
1588 		 * ...IF the above limits stacking determined a zoned model
1589 		 * validate that all of the table's devices conform to it.
1590 		 */
1591 		zoned_model = limits->zoned;
1592 		zone_sectors = limits->chunk_sectors;
1593 	}
1594 	if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1595 		return -EINVAL;
1596 
1597 	return validate_hardware_logical_block_alignment(table, limits);
1598 }
1599 
1600 /*
1601  * Verify that all devices have an integrity profile that matches the
1602  * DM device's registered integrity profile.  If the profiles don't
1603  * match then unregister the DM device's integrity profile.
1604  */
1605 static void dm_table_verify_integrity(struct dm_table *t)
1606 {
1607 	struct gendisk *template_disk = NULL;
1608 
1609 	if (t->integrity_added)
1610 		return;
1611 
1612 	if (t->integrity_supported) {
1613 		/*
1614 		 * Verify that the original integrity profile
1615 		 * matches all the devices in this table.
1616 		 */
1617 		template_disk = dm_table_get_integrity_disk(t);
1618 		if (template_disk &&
1619 		    blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1620 			return;
1621 	}
1622 
1623 	if (integrity_profile_exists(dm_disk(t->md))) {
1624 		DMWARN("%s: unable to establish an integrity profile",
1625 		       dm_device_name(t->md));
1626 		blk_integrity_unregister(dm_disk(t->md));
1627 	}
1628 }
1629 
1630 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1631 				sector_t start, sector_t len, void *data)
1632 {
1633 	unsigned long flush = (unsigned long) data;
1634 	struct request_queue *q = bdev_get_queue(dev->bdev);
1635 
1636 	return q && (q->queue_flags & flush);
1637 }
1638 
1639 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1640 {
1641 	struct dm_target *ti;
1642 	unsigned i;
1643 
1644 	/*
1645 	 * Require at least one underlying device to support flushes.
1646 	 * t->devices includes internal dm devices such as mirror logs
1647 	 * so we need to use iterate_devices here, which targets
1648 	 * supporting flushes must provide.
1649 	 */
1650 	for (i = 0; i < dm_table_get_num_targets(t); i++) {
1651 		ti = dm_table_get_target(t, i);
1652 
1653 		if (!ti->num_flush_bios)
1654 			continue;
1655 
1656 		if (ti->flush_supported)
1657 			return true;
1658 
1659 		if (ti->type->iterate_devices &&
1660 		    ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1661 			return true;
1662 	}
1663 
1664 	return false;
1665 }
1666 
1667 static int device_dax_write_cache_enabled(struct dm_target *ti,
1668 					  struct dm_dev *dev, sector_t start,
1669 					  sector_t len, void *data)
1670 {
1671 	struct dax_device *dax_dev = dev->dax_dev;
1672 
1673 	if (!dax_dev)
1674 		return false;
1675 
1676 	if (dax_write_cache_enabled(dax_dev))
1677 		return true;
1678 	return false;
1679 }
1680 
1681 static int dm_table_supports_dax_write_cache(struct dm_table *t)
1682 {
1683 	struct dm_target *ti;
1684 	unsigned i;
1685 
1686 	for (i = 0; i < dm_table_get_num_targets(t); i++) {
1687 		ti = dm_table_get_target(t, i);
1688 
1689 		if (ti->type->iterate_devices &&
1690 		    ti->type->iterate_devices(ti,
1691 				device_dax_write_cache_enabled, NULL))
1692 			return true;
1693 	}
1694 
1695 	return false;
1696 }
1697 
1698 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1699 			    sector_t start, sector_t len, void *data)
1700 {
1701 	struct request_queue *q = bdev_get_queue(dev->bdev);
1702 
1703 	return q && blk_queue_nonrot(q);
1704 }
1705 
1706 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1707 			     sector_t start, sector_t len, void *data)
1708 {
1709 	struct request_queue *q = bdev_get_queue(dev->bdev);
1710 
1711 	return q && !blk_queue_add_random(q);
1712 }
1713 
1714 static bool dm_table_all_devices_attribute(struct dm_table *t,
1715 					   iterate_devices_callout_fn func)
1716 {
1717 	struct dm_target *ti;
1718 	unsigned i;
1719 
1720 	for (i = 0; i < dm_table_get_num_targets(t); i++) {
1721 		ti = dm_table_get_target(t, i);
1722 
1723 		if (!ti->type->iterate_devices ||
1724 		    !ti->type->iterate_devices(ti, func, NULL))
1725 			return false;
1726 	}
1727 
1728 	return true;
1729 }
1730 
1731 static int device_no_partial_completion(struct dm_target *ti, struct dm_dev *dev,
1732 					sector_t start, sector_t len, void *data)
1733 {
1734 	char b[BDEVNAME_SIZE];
1735 
1736 	/* For now, NVMe devices are the only devices of this class */
1737 	return (strncmp(bdevname(dev->bdev, b), "nvme", 4) == 0);
1738 }
1739 
1740 static bool dm_table_does_not_support_partial_completion(struct dm_table *t)
1741 {
1742 	return dm_table_all_devices_attribute(t, device_no_partial_completion);
1743 }
1744 
1745 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1746 					 sector_t start, sector_t len, void *data)
1747 {
1748 	struct request_queue *q = bdev_get_queue(dev->bdev);
1749 
1750 	return q && !q->limits.max_write_same_sectors;
1751 }
1752 
1753 static bool dm_table_supports_write_same(struct dm_table *t)
1754 {
1755 	struct dm_target *ti;
1756 	unsigned i;
1757 
1758 	for (i = 0; i < dm_table_get_num_targets(t); i++) {
1759 		ti = dm_table_get_target(t, i);
1760 
1761 		if (!ti->num_write_same_bios)
1762 			return false;
1763 
1764 		if (!ti->type->iterate_devices ||
1765 		    ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1766 			return false;
1767 	}
1768 
1769 	return true;
1770 }
1771 
1772 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1773 					   sector_t start, sector_t len, void *data)
1774 {
1775 	struct request_queue *q = bdev_get_queue(dev->bdev);
1776 
1777 	return q && !q->limits.max_write_zeroes_sectors;
1778 }
1779 
1780 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1781 {
1782 	struct dm_target *ti;
1783 	unsigned i = 0;
1784 
1785 	while (i < dm_table_get_num_targets(t)) {
1786 		ti = dm_table_get_target(t, i++);
1787 
1788 		if (!ti->num_write_zeroes_bios)
1789 			return false;
1790 
1791 		if (!ti->type->iterate_devices ||
1792 		    ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1793 			return false;
1794 	}
1795 
1796 	return true;
1797 }
1798 
1799 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1800 				      sector_t start, sector_t len, void *data)
1801 {
1802 	struct request_queue *q = bdev_get_queue(dev->bdev);
1803 
1804 	return q && !blk_queue_discard(q);
1805 }
1806 
1807 static bool dm_table_supports_discards(struct dm_table *t)
1808 {
1809 	struct dm_target *ti;
1810 	unsigned i;
1811 
1812 	for (i = 0; i < dm_table_get_num_targets(t); i++) {
1813 		ti = dm_table_get_target(t, i);
1814 
1815 		if (!ti->num_discard_bios)
1816 			return false;
1817 
1818 		/*
1819 		 * Either the target provides discard support (as implied by setting
1820 		 * 'discards_supported') or it relies on _all_ data devices having
1821 		 * discard support.
1822 		 */
1823 		if (!ti->discards_supported &&
1824 		    (!ti->type->iterate_devices ||
1825 		     ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1826 			return false;
1827 	}
1828 
1829 	return true;
1830 }
1831 
1832 static int device_not_secure_erase_capable(struct dm_target *ti,
1833 					   struct dm_dev *dev, sector_t start,
1834 					   sector_t len, void *data)
1835 {
1836 	struct request_queue *q = bdev_get_queue(dev->bdev);
1837 
1838 	return q && !blk_queue_secure_erase(q);
1839 }
1840 
1841 static bool dm_table_supports_secure_erase(struct dm_table *t)
1842 {
1843 	struct dm_target *ti;
1844 	unsigned int i;
1845 
1846 	for (i = 0; i < dm_table_get_num_targets(t); i++) {
1847 		ti = dm_table_get_target(t, i);
1848 
1849 		if (!ti->num_secure_erase_bios)
1850 			return false;
1851 
1852 		if (!ti->type->iterate_devices ||
1853 		    ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1854 			return false;
1855 	}
1856 
1857 	return true;
1858 }
1859 
1860 static int device_requires_stable_pages(struct dm_target *ti,
1861 					struct dm_dev *dev, sector_t start,
1862 					sector_t len, void *data)
1863 {
1864 	struct request_queue *q = bdev_get_queue(dev->bdev);
1865 
1866 	return q && bdi_cap_stable_pages_required(q->backing_dev_info);
1867 }
1868 
1869 /*
1870  * If any underlying device requires stable pages, a table must require
1871  * them as well.  Only targets that support iterate_devices are considered:
1872  * don't want error, zero, etc to require stable pages.
1873  */
1874 static bool dm_table_requires_stable_pages(struct dm_table *t)
1875 {
1876 	struct dm_target *ti;
1877 	unsigned i;
1878 
1879 	for (i = 0; i < dm_table_get_num_targets(t); i++) {
1880 		ti = dm_table_get_target(t, i);
1881 
1882 		if (ti->type->iterate_devices &&
1883 		    ti->type->iterate_devices(ti, device_requires_stable_pages, NULL))
1884 			return true;
1885 	}
1886 
1887 	return false;
1888 }
1889 
1890 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1891 			       struct queue_limits *limits)
1892 {
1893 	bool wc = false, fua = false;
1894 	int page_size = PAGE_SIZE;
1895 
1896 	/*
1897 	 * Copy table's limits to the DM device's request_queue
1898 	 */
1899 	q->limits = *limits;
1900 
1901 	if (!dm_table_supports_discards(t)) {
1902 		blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
1903 		/* Must also clear discard limits... */
1904 		q->limits.max_discard_sectors = 0;
1905 		q->limits.max_hw_discard_sectors = 0;
1906 		q->limits.discard_granularity = 0;
1907 		q->limits.discard_alignment = 0;
1908 		q->limits.discard_misaligned = 0;
1909 	} else
1910 		blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
1911 
1912 	if (dm_table_supports_secure_erase(t))
1913 		blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
1914 
1915 	if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1916 		wc = true;
1917 		if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1918 			fua = true;
1919 	}
1920 	blk_queue_write_cache(q, wc, fua);
1921 
1922 	if (dm_table_supports_dax(t, device_supports_dax, &page_size)) {
1923 		blk_queue_flag_set(QUEUE_FLAG_DAX, q);
1924 		if (dm_table_supports_dax(t, device_synchronous, NULL))
1925 			set_dax_synchronous(t->md->dax_dev);
1926 	}
1927 	else
1928 		blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
1929 
1930 	if (dm_table_supports_dax_write_cache(t))
1931 		dax_write_cache(t->md->dax_dev, true);
1932 
1933 	/* Ensure that all underlying devices are non-rotational. */
1934 	if (dm_table_all_devices_attribute(t, device_is_nonrot))
1935 		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
1936 	else
1937 		blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
1938 
1939 	if (!dm_table_supports_write_same(t))
1940 		q->limits.max_write_same_sectors = 0;
1941 	if (!dm_table_supports_write_zeroes(t))
1942 		q->limits.max_write_zeroes_sectors = 0;
1943 
1944 	dm_table_verify_integrity(t);
1945 
1946 	/*
1947 	 * Some devices don't use blk_integrity but still want stable pages
1948 	 * because they do their own checksumming.
1949 	 */
1950 	if (dm_table_requires_stable_pages(t))
1951 		q->backing_dev_info->capabilities |= BDI_CAP_STABLE_WRITES;
1952 	else
1953 		q->backing_dev_info->capabilities &= ~BDI_CAP_STABLE_WRITES;
1954 
1955 	/*
1956 	 * Determine whether or not this queue's I/O timings contribute
1957 	 * to the entropy pool, Only request-based targets use this.
1958 	 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1959 	 * have it set.
1960 	 */
1961 	if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1962 		blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
1963 
1964 	/*
1965 	 * For a zoned target, the number of zones should be updated for the
1966 	 * correct value to be exposed in sysfs queue/nr_zones. For a BIO based
1967 	 * target, this is all that is needed. For a request based target, the
1968 	 * queue zone bitmaps must also be updated.
1969 	 * Use blk_revalidate_disk_zones() to handle this.
1970 	 */
1971 	if (blk_queue_is_zoned(q))
1972 		blk_revalidate_disk_zones(t->md->disk);
1973 
1974 	/* Allow reads to exceed readahead limits */
1975 	q->backing_dev_info->io_pages = limits->max_sectors >> (PAGE_SHIFT - 9);
1976 }
1977 
1978 unsigned int dm_table_get_num_targets(struct dm_table *t)
1979 {
1980 	return t->num_targets;
1981 }
1982 
1983 struct list_head *dm_table_get_devices(struct dm_table *t)
1984 {
1985 	return &t->devices;
1986 }
1987 
1988 fmode_t dm_table_get_mode(struct dm_table *t)
1989 {
1990 	return t->mode;
1991 }
1992 EXPORT_SYMBOL(dm_table_get_mode);
1993 
1994 enum suspend_mode {
1995 	PRESUSPEND,
1996 	PRESUSPEND_UNDO,
1997 	POSTSUSPEND,
1998 };
1999 
2000 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
2001 {
2002 	int i = t->num_targets;
2003 	struct dm_target *ti = t->targets;
2004 
2005 	lockdep_assert_held(&t->md->suspend_lock);
2006 
2007 	while (i--) {
2008 		switch (mode) {
2009 		case PRESUSPEND:
2010 			if (ti->type->presuspend)
2011 				ti->type->presuspend(ti);
2012 			break;
2013 		case PRESUSPEND_UNDO:
2014 			if (ti->type->presuspend_undo)
2015 				ti->type->presuspend_undo(ti);
2016 			break;
2017 		case POSTSUSPEND:
2018 			if (ti->type->postsuspend)
2019 				ti->type->postsuspend(ti);
2020 			break;
2021 		}
2022 		ti++;
2023 	}
2024 }
2025 
2026 void dm_table_presuspend_targets(struct dm_table *t)
2027 {
2028 	if (!t)
2029 		return;
2030 
2031 	suspend_targets(t, PRESUSPEND);
2032 }
2033 
2034 void dm_table_presuspend_undo_targets(struct dm_table *t)
2035 {
2036 	if (!t)
2037 		return;
2038 
2039 	suspend_targets(t, PRESUSPEND_UNDO);
2040 }
2041 
2042 void dm_table_postsuspend_targets(struct dm_table *t)
2043 {
2044 	if (!t)
2045 		return;
2046 
2047 	suspend_targets(t, POSTSUSPEND);
2048 }
2049 
2050 int dm_table_resume_targets(struct dm_table *t)
2051 {
2052 	int i, r = 0;
2053 
2054 	lockdep_assert_held(&t->md->suspend_lock);
2055 
2056 	for (i = 0; i < t->num_targets; i++) {
2057 		struct dm_target *ti = t->targets + i;
2058 
2059 		if (!ti->type->preresume)
2060 			continue;
2061 
2062 		r = ti->type->preresume(ti);
2063 		if (r) {
2064 			DMERR("%s: %s: preresume failed, error = %d",
2065 			      dm_device_name(t->md), ti->type->name, r);
2066 			return r;
2067 		}
2068 	}
2069 
2070 	for (i = 0; i < t->num_targets; i++) {
2071 		struct dm_target *ti = t->targets + i;
2072 
2073 		if (ti->type->resume)
2074 			ti->type->resume(ti);
2075 	}
2076 
2077 	return 0;
2078 }
2079 
2080 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
2081 {
2082 	list_add(&cb->list, &t->target_callbacks);
2083 }
2084 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
2085 
2086 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
2087 {
2088 	struct dm_dev_internal *dd;
2089 	struct list_head *devices = dm_table_get_devices(t);
2090 	struct dm_target_callbacks *cb;
2091 	int r = 0;
2092 
2093 	list_for_each_entry(dd, devices, list) {
2094 		struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
2095 		char b[BDEVNAME_SIZE];
2096 
2097 		if (likely(q))
2098 			r |= bdi_congested(q->backing_dev_info, bdi_bits);
2099 		else
2100 			DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
2101 				     dm_device_name(t->md),
2102 				     bdevname(dd->dm_dev->bdev, b));
2103 	}
2104 
2105 	list_for_each_entry(cb, &t->target_callbacks, list)
2106 		if (cb->congested_fn)
2107 			r |= cb->congested_fn(cb, bdi_bits);
2108 
2109 	return r;
2110 }
2111 
2112 struct mapped_device *dm_table_get_md(struct dm_table *t)
2113 {
2114 	return t->md;
2115 }
2116 EXPORT_SYMBOL(dm_table_get_md);
2117 
2118 const char *dm_table_device_name(struct dm_table *t)
2119 {
2120 	return dm_device_name(t->md);
2121 }
2122 EXPORT_SYMBOL_GPL(dm_table_device_name);
2123 
2124 void dm_table_run_md_queue_async(struct dm_table *t)
2125 {
2126 	struct mapped_device *md;
2127 	struct request_queue *queue;
2128 
2129 	if (!dm_table_request_based(t))
2130 		return;
2131 
2132 	md = dm_table_get_md(t);
2133 	queue = dm_get_md_queue(md);
2134 	if (queue)
2135 		blk_mq_run_hw_queues(queue, true);
2136 }
2137 EXPORT_SYMBOL(dm_table_run_md_queue_async);
2138 
2139