xref: /openbmc/linux/drivers/md/dm-table.c (revision 9c1f8594)
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.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 
22 #define DM_MSG_PREFIX "table"
23 
24 #define MAX_DEPTH 16
25 #define NODE_SIZE L1_CACHE_BYTES
26 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
27 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
28 
29 /*
30  * The table has always exactly one reference from either mapped_device->map
31  * or hash_cell->new_map. This reference is not counted in table->holders.
32  * A pair of dm_create_table/dm_destroy_table functions is used for table
33  * creation/destruction.
34  *
35  * Temporary references from the other code increase table->holders. A pair
36  * of dm_table_get/dm_table_put functions is used to manipulate it.
37  *
38  * When the table is about to be destroyed, we wait for table->holders to
39  * drop to zero.
40  */
41 
42 struct dm_table {
43 	struct mapped_device *md;
44 	atomic_t holders;
45 	unsigned type;
46 
47 	/* btree table */
48 	unsigned int depth;
49 	unsigned int counts[MAX_DEPTH];	/* in nodes */
50 	sector_t *index[MAX_DEPTH];
51 
52 	unsigned int num_targets;
53 	unsigned int num_allocated;
54 	sector_t *highs;
55 	struct dm_target *targets;
56 
57 	unsigned integrity_supported:1;
58 
59 	/*
60 	 * Indicates the rw permissions for the new logical
61 	 * device.  This should be a combination of FMODE_READ
62 	 * and FMODE_WRITE.
63 	 */
64 	fmode_t mode;
65 
66 	/* a list of devices used by this table */
67 	struct list_head devices;
68 
69 	/* events get handed up using this callback */
70 	void (*event_fn)(void *);
71 	void *event_context;
72 
73 	struct dm_md_mempools *mempools;
74 
75 	struct list_head target_callbacks;
76 };
77 
78 /*
79  * Similar to ceiling(log_size(n))
80  */
81 static unsigned int int_log(unsigned int n, unsigned int base)
82 {
83 	int result = 0;
84 
85 	while (n > 1) {
86 		n = dm_div_up(n, base);
87 		result++;
88 	}
89 
90 	return result;
91 }
92 
93 /*
94  * Calculate the index of the child node of the n'th node k'th key.
95  */
96 static inline unsigned int get_child(unsigned int n, unsigned int k)
97 {
98 	return (n * CHILDREN_PER_NODE) + k;
99 }
100 
101 /*
102  * Return the n'th node of level l from table t.
103  */
104 static inline sector_t *get_node(struct dm_table *t,
105 				 unsigned int l, unsigned int n)
106 {
107 	return t->index[l] + (n * KEYS_PER_NODE);
108 }
109 
110 /*
111  * Return the highest key that you could lookup from the n'th
112  * node on level l of the btree.
113  */
114 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
115 {
116 	for (; l < t->depth - 1; l++)
117 		n = get_child(n, CHILDREN_PER_NODE - 1);
118 
119 	if (n >= t->counts[l])
120 		return (sector_t) - 1;
121 
122 	return get_node(t, l, n)[KEYS_PER_NODE - 1];
123 }
124 
125 /*
126  * Fills in a level of the btree based on the highs of the level
127  * below it.
128  */
129 static int setup_btree_index(unsigned int l, struct dm_table *t)
130 {
131 	unsigned int n, k;
132 	sector_t *node;
133 
134 	for (n = 0U; n < t->counts[l]; n++) {
135 		node = get_node(t, l, n);
136 
137 		for (k = 0U; k < KEYS_PER_NODE; k++)
138 			node[k] = high(t, l + 1, get_child(n, k));
139 	}
140 
141 	return 0;
142 }
143 
144 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
145 {
146 	unsigned long size;
147 	void *addr;
148 
149 	/*
150 	 * Check that we're not going to overflow.
151 	 */
152 	if (nmemb > (ULONG_MAX / elem_size))
153 		return NULL;
154 
155 	size = nmemb * elem_size;
156 	addr = vzalloc(size);
157 
158 	return addr;
159 }
160 EXPORT_SYMBOL(dm_vcalloc);
161 
162 /*
163  * highs, and targets are managed as dynamic arrays during a
164  * table load.
165  */
166 static int alloc_targets(struct dm_table *t, unsigned int num)
167 {
168 	sector_t *n_highs;
169 	struct dm_target *n_targets;
170 	int n = t->num_targets;
171 
172 	/*
173 	 * Allocate both the target array and offset array at once.
174 	 * Append an empty entry to catch sectors beyond the end of
175 	 * the device.
176 	 */
177 	n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
178 					  sizeof(sector_t));
179 	if (!n_highs)
180 		return -ENOMEM;
181 
182 	n_targets = (struct dm_target *) (n_highs + num);
183 
184 	if (n) {
185 		memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
186 		memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
187 	}
188 
189 	memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
190 	vfree(t->highs);
191 
192 	t->num_allocated = num;
193 	t->highs = n_highs;
194 	t->targets = n_targets;
195 
196 	return 0;
197 }
198 
199 int dm_table_create(struct dm_table **result, fmode_t mode,
200 		    unsigned num_targets, struct mapped_device *md)
201 {
202 	struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
203 
204 	if (!t)
205 		return -ENOMEM;
206 
207 	INIT_LIST_HEAD(&t->devices);
208 	INIT_LIST_HEAD(&t->target_callbacks);
209 	atomic_set(&t->holders, 0);
210 
211 	if (!num_targets)
212 		num_targets = KEYS_PER_NODE;
213 
214 	num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
215 
216 	if (alloc_targets(t, num_targets)) {
217 		kfree(t);
218 		t = NULL;
219 		return -ENOMEM;
220 	}
221 
222 	t->mode = mode;
223 	t->md = md;
224 	*result = t;
225 	return 0;
226 }
227 
228 static void free_devices(struct list_head *devices)
229 {
230 	struct list_head *tmp, *next;
231 
232 	list_for_each_safe(tmp, next, devices) {
233 		struct dm_dev_internal *dd =
234 		    list_entry(tmp, struct dm_dev_internal, list);
235 		DMWARN("dm_table_destroy: dm_put_device call missing for %s",
236 		       dd->dm_dev.name);
237 		kfree(dd);
238 	}
239 }
240 
241 void dm_table_destroy(struct dm_table *t)
242 {
243 	unsigned int i;
244 
245 	if (!t)
246 		return;
247 
248 	while (atomic_read(&t->holders))
249 		msleep(1);
250 	smp_mb();
251 
252 	/* free the indexes */
253 	if (t->depth >= 2)
254 		vfree(t->index[t->depth - 2]);
255 
256 	/* free the targets */
257 	for (i = 0; i < t->num_targets; i++) {
258 		struct dm_target *tgt = t->targets + i;
259 
260 		if (tgt->type->dtr)
261 			tgt->type->dtr(tgt);
262 
263 		dm_put_target_type(tgt->type);
264 	}
265 
266 	vfree(t->highs);
267 
268 	/* free the device list */
269 	if (t->devices.next != &t->devices)
270 		free_devices(&t->devices);
271 
272 	dm_free_md_mempools(t->mempools);
273 
274 	kfree(t);
275 }
276 
277 void dm_table_get(struct dm_table *t)
278 {
279 	atomic_inc(&t->holders);
280 }
281 EXPORT_SYMBOL(dm_table_get);
282 
283 void dm_table_put(struct dm_table *t)
284 {
285 	if (!t)
286 		return;
287 
288 	smp_mb__before_atomic_dec();
289 	atomic_dec(&t->holders);
290 }
291 EXPORT_SYMBOL(dm_table_put);
292 
293 /*
294  * Checks to see if we need to extend highs or targets.
295  */
296 static inline int check_space(struct dm_table *t)
297 {
298 	if (t->num_targets >= t->num_allocated)
299 		return alloc_targets(t, t->num_allocated * 2);
300 
301 	return 0;
302 }
303 
304 /*
305  * See if we've already got a device in the list.
306  */
307 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
308 {
309 	struct dm_dev_internal *dd;
310 
311 	list_for_each_entry (dd, l, list)
312 		if (dd->dm_dev.bdev->bd_dev == dev)
313 			return dd;
314 
315 	return NULL;
316 }
317 
318 /*
319  * Open a device so we can use it as a map destination.
320  */
321 static int open_dev(struct dm_dev_internal *d, dev_t dev,
322 		    struct mapped_device *md)
323 {
324 	static char *_claim_ptr = "I belong to device-mapper";
325 	struct block_device *bdev;
326 
327 	int r;
328 
329 	BUG_ON(d->dm_dev.bdev);
330 
331 	bdev = blkdev_get_by_dev(dev, d->dm_dev.mode | FMODE_EXCL, _claim_ptr);
332 	if (IS_ERR(bdev))
333 		return PTR_ERR(bdev);
334 
335 	r = bd_link_disk_holder(bdev, dm_disk(md));
336 	if (r) {
337 		blkdev_put(bdev, d->dm_dev.mode | FMODE_EXCL);
338 		return r;
339 	}
340 
341 	d->dm_dev.bdev = bdev;
342 	return 0;
343 }
344 
345 /*
346  * Close a device that we've been using.
347  */
348 static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
349 {
350 	if (!d->dm_dev.bdev)
351 		return;
352 
353 	bd_unlink_disk_holder(d->dm_dev.bdev, dm_disk(md));
354 	blkdev_put(d->dm_dev.bdev, d->dm_dev.mode | FMODE_EXCL);
355 	d->dm_dev.bdev = NULL;
356 }
357 
358 /*
359  * If possible, this checks an area of a destination device is invalid.
360  */
361 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
362 				  sector_t start, sector_t len, void *data)
363 {
364 	struct request_queue *q;
365 	struct queue_limits *limits = data;
366 	struct block_device *bdev = dev->bdev;
367 	sector_t dev_size =
368 		i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
369 	unsigned short logical_block_size_sectors =
370 		limits->logical_block_size >> SECTOR_SHIFT;
371 	char b[BDEVNAME_SIZE];
372 
373 	/*
374 	 * Some devices exist without request functions,
375 	 * such as loop devices not yet bound to backing files.
376 	 * Forbid the use of such devices.
377 	 */
378 	q = bdev_get_queue(bdev);
379 	if (!q || !q->make_request_fn) {
380 		DMWARN("%s: %s is not yet initialised: "
381 		       "start=%llu, len=%llu, dev_size=%llu",
382 		       dm_device_name(ti->table->md), bdevname(bdev, b),
383 		       (unsigned long long)start,
384 		       (unsigned long long)len,
385 		       (unsigned long long)dev_size);
386 		return 1;
387 	}
388 
389 	if (!dev_size)
390 		return 0;
391 
392 	if ((start >= dev_size) || (start + len > dev_size)) {
393 		DMWARN("%s: %s too small for target: "
394 		       "start=%llu, len=%llu, dev_size=%llu",
395 		       dm_device_name(ti->table->md), bdevname(bdev, b),
396 		       (unsigned long long)start,
397 		       (unsigned long long)len,
398 		       (unsigned long long)dev_size);
399 		return 1;
400 	}
401 
402 	if (logical_block_size_sectors <= 1)
403 		return 0;
404 
405 	if (start & (logical_block_size_sectors - 1)) {
406 		DMWARN("%s: start=%llu not aligned to h/w "
407 		       "logical block size %u of %s",
408 		       dm_device_name(ti->table->md),
409 		       (unsigned long long)start,
410 		       limits->logical_block_size, bdevname(bdev, b));
411 		return 1;
412 	}
413 
414 	if (len & (logical_block_size_sectors - 1)) {
415 		DMWARN("%s: len=%llu not aligned to h/w "
416 		       "logical block size %u of %s",
417 		       dm_device_name(ti->table->md),
418 		       (unsigned long long)len,
419 		       limits->logical_block_size, bdevname(bdev, b));
420 		return 1;
421 	}
422 
423 	return 0;
424 }
425 
426 /*
427  * This upgrades the mode on an already open dm_dev, being
428  * careful to leave things as they were if we fail to reopen the
429  * device and not to touch the existing bdev field in case
430  * it is accessed concurrently inside dm_table_any_congested().
431  */
432 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
433 			struct mapped_device *md)
434 {
435 	int r;
436 	struct dm_dev_internal dd_new, dd_old;
437 
438 	dd_new = dd_old = *dd;
439 
440 	dd_new.dm_dev.mode |= new_mode;
441 	dd_new.dm_dev.bdev = NULL;
442 
443 	r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
444 	if (r)
445 		return r;
446 
447 	dd->dm_dev.mode |= new_mode;
448 	close_dev(&dd_old, md);
449 
450 	return 0;
451 }
452 
453 /*
454  * Add a device to the list, or just increment the usage count if
455  * it's already present.
456  */
457 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
458 		  struct dm_dev **result)
459 {
460 	int r;
461 	dev_t uninitialized_var(dev);
462 	struct dm_dev_internal *dd;
463 	unsigned int major, minor;
464 	struct dm_table *t = ti->table;
465 
466 	BUG_ON(!t);
467 
468 	if (sscanf(path, "%u:%u", &major, &minor) == 2) {
469 		/* Extract the major/minor numbers */
470 		dev = MKDEV(major, minor);
471 		if (MAJOR(dev) != major || MINOR(dev) != minor)
472 			return -EOVERFLOW;
473 	} else {
474 		/* convert the path to a device */
475 		struct block_device *bdev = lookup_bdev(path);
476 
477 		if (IS_ERR(bdev))
478 			return PTR_ERR(bdev);
479 		dev = bdev->bd_dev;
480 		bdput(bdev);
481 	}
482 
483 	dd = find_device(&t->devices, dev);
484 	if (!dd) {
485 		dd = kmalloc(sizeof(*dd), GFP_KERNEL);
486 		if (!dd)
487 			return -ENOMEM;
488 
489 		dd->dm_dev.mode = mode;
490 		dd->dm_dev.bdev = NULL;
491 
492 		if ((r = open_dev(dd, dev, t->md))) {
493 			kfree(dd);
494 			return r;
495 		}
496 
497 		format_dev_t(dd->dm_dev.name, dev);
498 
499 		atomic_set(&dd->count, 0);
500 		list_add(&dd->list, &t->devices);
501 
502 	} else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
503 		r = upgrade_mode(dd, mode, t->md);
504 		if (r)
505 			return r;
506 	}
507 	atomic_inc(&dd->count);
508 
509 	*result = &dd->dm_dev;
510 	return 0;
511 }
512 EXPORT_SYMBOL(dm_get_device);
513 
514 int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
515 			 sector_t start, sector_t len, void *data)
516 {
517 	struct queue_limits *limits = data;
518 	struct block_device *bdev = dev->bdev;
519 	struct request_queue *q = bdev_get_queue(bdev);
520 	char b[BDEVNAME_SIZE];
521 
522 	if (unlikely(!q)) {
523 		DMWARN("%s: Cannot set limits for nonexistent device %s",
524 		       dm_device_name(ti->table->md), bdevname(bdev, b));
525 		return 0;
526 	}
527 
528 	if (bdev_stack_limits(limits, bdev, start) < 0)
529 		DMWARN("%s: adding target device %s caused an alignment inconsistency: "
530 		       "physical_block_size=%u, logical_block_size=%u, "
531 		       "alignment_offset=%u, start=%llu",
532 		       dm_device_name(ti->table->md), bdevname(bdev, b),
533 		       q->limits.physical_block_size,
534 		       q->limits.logical_block_size,
535 		       q->limits.alignment_offset,
536 		       (unsigned long long) start << SECTOR_SHIFT);
537 
538 	/*
539 	 * Check if merge fn is supported.
540 	 * If not we'll force DM to use PAGE_SIZE or
541 	 * smaller I/O, just to be safe.
542 	 */
543 	if (dm_queue_merge_is_compulsory(q) && !ti->type->merge)
544 		blk_limits_max_hw_sectors(limits,
545 					  (unsigned int) (PAGE_SIZE >> 9));
546 	return 0;
547 }
548 EXPORT_SYMBOL_GPL(dm_set_device_limits);
549 
550 /*
551  * Decrement a device's use count and remove it if necessary.
552  */
553 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
554 {
555 	struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
556 						  dm_dev);
557 
558 	if (atomic_dec_and_test(&dd->count)) {
559 		close_dev(dd, ti->table->md);
560 		list_del(&dd->list);
561 		kfree(dd);
562 	}
563 }
564 EXPORT_SYMBOL(dm_put_device);
565 
566 /*
567  * Checks to see if the target joins onto the end of the table.
568  */
569 static int adjoin(struct dm_table *table, struct dm_target *ti)
570 {
571 	struct dm_target *prev;
572 
573 	if (!table->num_targets)
574 		return !ti->begin;
575 
576 	prev = &table->targets[table->num_targets - 1];
577 	return (ti->begin == (prev->begin + prev->len));
578 }
579 
580 /*
581  * Used to dynamically allocate the arg array.
582  */
583 static char **realloc_argv(unsigned *array_size, char **old_argv)
584 {
585 	char **argv;
586 	unsigned new_size;
587 
588 	new_size = *array_size ? *array_size * 2 : 64;
589 	argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
590 	if (argv) {
591 		memcpy(argv, old_argv, *array_size * sizeof(*argv));
592 		*array_size = new_size;
593 	}
594 
595 	kfree(old_argv);
596 	return argv;
597 }
598 
599 /*
600  * Destructively splits up the argument list to pass to ctr.
601  */
602 int dm_split_args(int *argc, char ***argvp, char *input)
603 {
604 	char *start, *end = input, *out, **argv = NULL;
605 	unsigned array_size = 0;
606 
607 	*argc = 0;
608 
609 	if (!input) {
610 		*argvp = NULL;
611 		return 0;
612 	}
613 
614 	argv = realloc_argv(&array_size, argv);
615 	if (!argv)
616 		return -ENOMEM;
617 
618 	while (1) {
619 		/* Skip whitespace */
620 		start = skip_spaces(end);
621 
622 		if (!*start)
623 			break;	/* success, we hit the end */
624 
625 		/* 'out' is used to remove any back-quotes */
626 		end = out = start;
627 		while (*end) {
628 			/* Everything apart from '\0' can be quoted */
629 			if (*end == '\\' && *(end + 1)) {
630 				*out++ = *(end + 1);
631 				end += 2;
632 				continue;
633 			}
634 
635 			if (isspace(*end))
636 				break;	/* end of token */
637 
638 			*out++ = *end++;
639 		}
640 
641 		/* have we already filled the array ? */
642 		if ((*argc + 1) > array_size) {
643 			argv = realloc_argv(&array_size, argv);
644 			if (!argv)
645 				return -ENOMEM;
646 		}
647 
648 		/* we know this is whitespace */
649 		if (*end)
650 			end++;
651 
652 		/* terminate the string and put it in the array */
653 		*out = '\0';
654 		argv[*argc] = start;
655 		(*argc)++;
656 	}
657 
658 	*argvp = argv;
659 	return 0;
660 }
661 
662 /*
663  * Impose necessary and sufficient conditions on a devices's table such
664  * that any incoming bio which respects its logical_block_size can be
665  * processed successfully.  If it falls across the boundary between
666  * two or more targets, the size of each piece it gets split into must
667  * be compatible with the logical_block_size of the target processing it.
668  */
669 static int validate_hardware_logical_block_alignment(struct dm_table *table,
670 						 struct queue_limits *limits)
671 {
672 	/*
673 	 * This function uses arithmetic modulo the logical_block_size
674 	 * (in units of 512-byte sectors).
675 	 */
676 	unsigned short device_logical_block_size_sects =
677 		limits->logical_block_size >> SECTOR_SHIFT;
678 
679 	/*
680 	 * Offset of the start of the next table entry, mod logical_block_size.
681 	 */
682 	unsigned short next_target_start = 0;
683 
684 	/*
685 	 * Given an aligned bio that extends beyond the end of a
686 	 * target, how many sectors must the next target handle?
687 	 */
688 	unsigned short remaining = 0;
689 
690 	struct dm_target *uninitialized_var(ti);
691 	struct queue_limits ti_limits;
692 	unsigned i = 0;
693 
694 	/*
695 	 * Check each entry in the table in turn.
696 	 */
697 	while (i < dm_table_get_num_targets(table)) {
698 		ti = dm_table_get_target(table, i++);
699 
700 		blk_set_default_limits(&ti_limits);
701 
702 		/* combine all target devices' limits */
703 		if (ti->type->iterate_devices)
704 			ti->type->iterate_devices(ti, dm_set_device_limits,
705 						  &ti_limits);
706 
707 		/*
708 		 * If the remaining sectors fall entirely within this
709 		 * table entry are they compatible with its logical_block_size?
710 		 */
711 		if (remaining < ti->len &&
712 		    remaining & ((ti_limits.logical_block_size >>
713 				  SECTOR_SHIFT) - 1))
714 			break;	/* Error */
715 
716 		next_target_start =
717 		    (unsigned short) ((next_target_start + ti->len) &
718 				      (device_logical_block_size_sects - 1));
719 		remaining = next_target_start ?
720 		    device_logical_block_size_sects - next_target_start : 0;
721 	}
722 
723 	if (remaining) {
724 		DMWARN("%s: table line %u (start sect %llu len %llu) "
725 		       "not aligned to h/w logical block size %u",
726 		       dm_device_name(table->md), i,
727 		       (unsigned long long) ti->begin,
728 		       (unsigned long long) ti->len,
729 		       limits->logical_block_size);
730 		return -EINVAL;
731 	}
732 
733 	return 0;
734 }
735 
736 int dm_table_add_target(struct dm_table *t, const char *type,
737 			sector_t start, sector_t len, char *params)
738 {
739 	int r = -EINVAL, argc;
740 	char **argv;
741 	struct dm_target *tgt;
742 
743 	if ((r = check_space(t)))
744 		return r;
745 
746 	tgt = t->targets + t->num_targets;
747 	memset(tgt, 0, sizeof(*tgt));
748 
749 	if (!len) {
750 		DMERR("%s: zero-length target", dm_device_name(t->md));
751 		return -EINVAL;
752 	}
753 
754 	tgt->type = dm_get_target_type(type);
755 	if (!tgt->type) {
756 		DMERR("%s: %s: unknown target type", dm_device_name(t->md),
757 		      type);
758 		return -EINVAL;
759 	}
760 
761 	tgt->table = t;
762 	tgt->begin = start;
763 	tgt->len = len;
764 	tgt->error = "Unknown error";
765 
766 	/*
767 	 * Does this target adjoin the previous one ?
768 	 */
769 	if (!adjoin(t, tgt)) {
770 		tgt->error = "Gap in table";
771 		r = -EINVAL;
772 		goto bad;
773 	}
774 
775 	r = dm_split_args(&argc, &argv, params);
776 	if (r) {
777 		tgt->error = "couldn't split parameters (insufficient memory)";
778 		goto bad;
779 	}
780 
781 	r = tgt->type->ctr(tgt, argc, argv);
782 	kfree(argv);
783 	if (r)
784 		goto bad;
785 
786 	t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
787 
788 	if (!tgt->num_discard_requests && tgt->discards_supported)
789 		DMWARN("%s: %s: ignoring discards_supported because num_discard_requests is zero.",
790 		       dm_device_name(t->md), type);
791 
792 	return 0;
793 
794  bad:
795 	DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
796 	dm_put_target_type(tgt->type);
797 	return r;
798 }
799 
800 /*
801  * Target argument parsing helpers.
802  */
803 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
804 			     unsigned *value, char **error, unsigned grouped)
805 {
806 	const char *arg_str = dm_shift_arg(arg_set);
807 
808 	if (!arg_str ||
809 	    (sscanf(arg_str, "%u", value) != 1) ||
810 	    (*value < arg->min) ||
811 	    (*value > arg->max) ||
812 	    (grouped && arg_set->argc < *value)) {
813 		*error = arg->error;
814 		return -EINVAL;
815 	}
816 
817 	return 0;
818 }
819 
820 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
821 		unsigned *value, char **error)
822 {
823 	return validate_next_arg(arg, arg_set, value, error, 0);
824 }
825 EXPORT_SYMBOL(dm_read_arg);
826 
827 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
828 		      unsigned *value, char **error)
829 {
830 	return validate_next_arg(arg, arg_set, value, error, 1);
831 }
832 EXPORT_SYMBOL(dm_read_arg_group);
833 
834 const char *dm_shift_arg(struct dm_arg_set *as)
835 {
836 	char *r;
837 
838 	if (as->argc) {
839 		as->argc--;
840 		r = *as->argv;
841 		as->argv++;
842 		return r;
843 	}
844 
845 	return NULL;
846 }
847 EXPORT_SYMBOL(dm_shift_arg);
848 
849 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
850 {
851 	BUG_ON(as->argc < num_args);
852 	as->argc -= num_args;
853 	as->argv += num_args;
854 }
855 EXPORT_SYMBOL(dm_consume_args);
856 
857 static int dm_table_set_type(struct dm_table *t)
858 {
859 	unsigned i;
860 	unsigned bio_based = 0, request_based = 0;
861 	struct dm_target *tgt;
862 	struct dm_dev_internal *dd;
863 	struct list_head *devices;
864 
865 	for (i = 0; i < t->num_targets; i++) {
866 		tgt = t->targets + i;
867 		if (dm_target_request_based(tgt))
868 			request_based = 1;
869 		else
870 			bio_based = 1;
871 
872 		if (bio_based && request_based) {
873 			DMWARN("Inconsistent table: different target types"
874 			       " can't be mixed up");
875 			return -EINVAL;
876 		}
877 	}
878 
879 	if (bio_based) {
880 		/* We must use this table as bio-based */
881 		t->type = DM_TYPE_BIO_BASED;
882 		return 0;
883 	}
884 
885 	BUG_ON(!request_based); /* No targets in this table */
886 
887 	/* Non-request-stackable devices can't be used for request-based dm */
888 	devices = dm_table_get_devices(t);
889 	list_for_each_entry(dd, devices, list) {
890 		if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) {
891 			DMWARN("table load rejected: including"
892 			       " non-request-stackable devices");
893 			return -EINVAL;
894 		}
895 	}
896 
897 	/*
898 	 * Request-based dm supports only tables that have a single target now.
899 	 * To support multiple targets, request splitting support is needed,
900 	 * and that needs lots of changes in the block-layer.
901 	 * (e.g. request completion process for partial completion.)
902 	 */
903 	if (t->num_targets > 1) {
904 		DMWARN("Request-based dm doesn't support multiple targets yet");
905 		return -EINVAL;
906 	}
907 
908 	t->type = DM_TYPE_REQUEST_BASED;
909 
910 	return 0;
911 }
912 
913 unsigned dm_table_get_type(struct dm_table *t)
914 {
915 	return t->type;
916 }
917 
918 bool dm_table_request_based(struct dm_table *t)
919 {
920 	return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
921 }
922 
923 int dm_table_alloc_md_mempools(struct dm_table *t)
924 {
925 	unsigned type = dm_table_get_type(t);
926 
927 	if (unlikely(type == DM_TYPE_NONE)) {
928 		DMWARN("no table type is set, can't allocate mempools");
929 		return -EINVAL;
930 	}
931 
932 	t->mempools = dm_alloc_md_mempools(type, t->integrity_supported);
933 	if (!t->mempools)
934 		return -ENOMEM;
935 
936 	return 0;
937 }
938 
939 void dm_table_free_md_mempools(struct dm_table *t)
940 {
941 	dm_free_md_mempools(t->mempools);
942 	t->mempools = NULL;
943 }
944 
945 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
946 {
947 	return t->mempools;
948 }
949 
950 static int setup_indexes(struct dm_table *t)
951 {
952 	int i;
953 	unsigned int total = 0;
954 	sector_t *indexes;
955 
956 	/* allocate the space for *all* the indexes */
957 	for (i = t->depth - 2; i >= 0; i--) {
958 		t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
959 		total += t->counts[i];
960 	}
961 
962 	indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
963 	if (!indexes)
964 		return -ENOMEM;
965 
966 	/* set up internal nodes, bottom-up */
967 	for (i = t->depth - 2; i >= 0; i--) {
968 		t->index[i] = indexes;
969 		indexes += (KEYS_PER_NODE * t->counts[i]);
970 		setup_btree_index(i, t);
971 	}
972 
973 	return 0;
974 }
975 
976 /*
977  * Builds the btree to index the map.
978  */
979 static int dm_table_build_index(struct dm_table *t)
980 {
981 	int r = 0;
982 	unsigned int leaf_nodes;
983 
984 	/* how many indexes will the btree have ? */
985 	leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
986 	t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
987 
988 	/* leaf layer has already been set up */
989 	t->counts[t->depth - 1] = leaf_nodes;
990 	t->index[t->depth - 1] = t->highs;
991 
992 	if (t->depth >= 2)
993 		r = setup_indexes(t);
994 
995 	return r;
996 }
997 
998 /*
999  * Get a disk whose integrity profile reflects the table's profile.
1000  * If %match_all is true, all devices' profiles must match.
1001  * If %match_all is false, all devices must at least have an
1002  * allocated integrity profile; but uninitialized is ok.
1003  * Returns NULL if integrity support was inconsistent or unavailable.
1004  */
1005 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t,
1006 						    bool match_all)
1007 {
1008 	struct list_head *devices = dm_table_get_devices(t);
1009 	struct dm_dev_internal *dd = NULL;
1010 	struct gendisk *prev_disk = NULL, *template_disk = NULL;
1011 
1012 	list_for_each_entry(dd, devices, list) {
1013 		template_disk = dd->dm_dev.bdev->bd_disk;
1014 		if (!blk_get_integrity(template_disk))
1015 			goto no_integrity;
1016 		if (!match_all && !blk_integrity_is_initialized(template_disk))
1017 			continue; /* skip uninitialized profiles */
1018 		else if (prev_disk &&
1019 			 blk_integrity_compare(prev_disk, template_disk) < 0)
1020 			goto no_integrity;
1021 		prev_disk = template_disk;
1022 	}
1023 
1024 	return template_disk;
1025 
1026 no_integrity:
1027 	if (prev_disk)
1028 		DMWARN("%s: integrity not set: %s and %s profile mismatch",
1029 		       dm_device_name(t->md),
1030 		       prev_disk->disk_name,
1031 		       template_disk->disk_name);
1032 	return NULL;
1033 }
1034 
1035 /*
1036  * Register the mapped device for blk_integrity support if
1037  * the underlying devices have an integrity profile.  But all devices
1038  * may not have matching profiles (checking all devices isn't reliable
1039  * during table load because this table may use other DM device(s) which
1040  * must be resumed before they will have an initialized integity profile).
1041  * Stacked DM devices force a 2 stage integrity profile validation:
1042  * 1 - during load, validate all initialized integrity profiles match
1043  * 2 - during resume, validate all integrity profiles match
1044  */
1045 static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
1046 {
1047 	struct gendisk *template_disk = NULL;
1048 
1049 	template_disk = dm_table_get_integrity_disk(t, false);
1050 	if (!template_disk)
1051 		return 0;
1052 
1053 	if (!blk_integrity_is_initialized(dm_disk(md))) {
1054 		t->integrity_supported = 1;
1055 		return blk_integrity_register(dm_disk(md), NULL);
1056 	}
1057 
1058 	/*
1059 	 * If DM device already has an initalized integrity
1060 	 * profile the new profile should not conflict.
1061 	 */
1062 	if (blk_integrity_is_initialized(template_disk) &&
1063 	    blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1064 		DMWARN("%s: conflict with existing integrity profile: "
1065 		       "%s profile mismatch",
1066 		       dm_device_name(t->md),
1067 		       template_disk->disk_name);
1068 		return 1;
1069 	}
1070 
1071 	/* Preserve existing initialized integrity profile */
1072 	t->integrity_supported = 1;
1073 	return 0;
1074 }
1075 
1076 /*
1077  * Prepares the table for use by building the indices,
1078  * setting the type, and allocating mempools.
1079  */
1080 int dm_table_complete(struct dm_table *t)
1081 {
1082 	int r;
1083 
1084 	r = dm_table_set_type(t);
1085 	if (r) {
1086 		DMERR("unable to set table type");
1087 		return r;
1088 	}
1089 
1090 	r = dm_table_build_index(t);
1091 	if (r) {
1092 		DMERR("unable to build btrees");
1093 		return r;
1094 	}
1095 
1096 	r = dm_table_prealloc_integrity(t, t->md);
1097 	if (r) {
1098 		DMERR("could not register integrity profile.");
1099 		return r;
1100 	}
1101 
1102 	r = dm_table_alloc_md_mempools(t);
1103 	if (r)
1104 		DMERR("unable to allocate mempools");
1105 
1106 	return r;
1107 }
1108 
1109 static DEFINE_MUTEX(_event_lock);
1110 void dm_table_event_callback(struct dm_table *t,
1111 			     void (*fn)(void *), void *context)
1112 {
1113 	mutex_lock(&_event_lock);
1114 	t->event_fn = fn;
1115 	t->event_context = context;
1116 	mutex_unlock(&_event_lock);
1117 }
1118 
1119 void dm_table_event(struct dm_table *t)
1120 {
1121 	/*
1122 	 * You can no longer call dm_table_event() from interrupt
1123 	 * context, use a bottom half instead.
1124 	 */
1125 	BUG_ON(in_interrupt());
1126 
1127 	mutex_lock(&_event_lock);
1128 	if (t->event_fn)
1129 		t->event_fn(t->event_context);
1130 	mutex_unlock(&_event_lock);
1131 }
1132 EXPORT_SYMBOL(dm_table_event);
1133 
1134 sector_t dm_table_get_size(struct dm_table *t)
1135 {
1136 	return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1137 }
1138 EXPORT_SYMBOL(dm_table_get_size);
1139 
1140 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1141 {
1142 	if (index >= t->num_targets)
1143 		return NULL;
1144 
1145 	return t->targets + index;
1146 }
1147 
1148 /*
1149  * Search the btree for the correct target.
1150  *
1151  * Caller should check returned pointer with dm_target_is_valid()
1152  * to trap I/O beyond end of device.
1153  */
1154 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1155 {
1156 	unsigned int l, n = 0, k = 0;
1157 	sector_t *node;
1158 
1159 	for (l = 0; l < t->depth; l++) {
1160 		n = get_child(n, k);
1161 		node = get_node(t, l, n);
1162 
1163 		for (k = 0; k < KEYS_PER_NODE; k++)
1164 			if (node[k] >= sector)
1165 				break;
1166 	}
1167 
1168 	return &t->targets[(KEYS_PER_NODE * n) + k];
1169 }
1170 
1171 /*
1172  * Establish the new table's queue_limits and validate them.
1173  */
1174 int dm_calculate_queue_limits(struct dm_table *table,
1175 			      struct queue_limits *limits)
1176 {
1177 	struct dm_target *uninitialized_var(ti);
1178 	struct queue_limits ti_limits;
1179 	unsigned i = 0;
1180 
1181 	blk_set_default_limits(limits);
1182 
1183 	while (i < dm_table_get_num_targets(table)) {
1184 		blk_set_default_limits(&ti_limits);
1185 
1186 		ti = dm_table_get_target(table, i++);
1187 
1188 		if (!ti->type->iterate_devices)
1189 			goto combine_limits;
1190 
1191 		/*
1192 		 * Combine queue limits of all the devices this target uses.
1193 		 */
1194 		ti->type->iterate_devices(ti, dm_set_device_limits,
1195 					  &ti_limits);
1196 
1197 		/* Set I/O hints portion of queue limits */
1198 		if (ti->type->io_hints)
1199 			ti->type->io_hints(ti, &ti_limits);
1200 
1201 		/*
1202 		 * Check each device area is consistent with the target's
1203 		 * overall queue limits.
1204 		 */
1205 		if (ti->type->iterate_devices(ti, device_area_is_invalid,
1206 					      &ti_limits))
1207 			return -EINVAL;
1208 
1209 combine_limits:
1210 		/*
1211 		 * Merge this target's queue limits into the overall limits
1212 		 * for the table.
1213 		 */
1214 		if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1215 			DMWARN("%s: adding target device "
1216 			       "(start sect %llu len %llu) "
1217 			       "caused an alignment inconsistency",
1218 			       dm_device_name(table->md),
1219 			       (unsigned long long) ti->begin,
1220 			       (unsigned long long) ti->len);
1221 	}
1222 
1223 	return validate_hardware_logical_block_alignment(table, limits);
1224 }
1225 
1226 /*
1227  * Set the integrity profile for this device if all devices used have
1228  * matching profiles.  We're quite deep in the resume path but still
1229  * don't know if all devices (particularly DM devices this device
1230  * may be stacked on) have matching profiles.  Even if the profiles
1231  * don't match we have no way to fail (to resume) at this point.
1232  */
1233 static void dm_table_set_integrity(struct dm_table *t)
1234 {
1235 	struct gendisk *template_disk = NULL;
1236 
1237 	if (!blk_get_integrity(dm_disk(t->md)))
1238 		return;
1239 
1240 	template_disk = dm_table_get_integrity_disk(t, true);
1241 	if (!template_disk &&
1242 	    blk_integrity_is_initialized(dm_disk(t->md))) {
1243 		DMWARN("%s: device no longer has a valid integrity profile",
1244 		       dm_device_name(t->md));
1245 		return;
1246 	}
1247 	blk_integrity_register(dm_disk(t->md),
1248 			       blk_get_integrity(template_disk));
1249 }
1250 
1251 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1252 				sector_t start, sector_t len, void *data)
1253 {
1254 	unsigned flush = (*(unsigned *)data);
1255 	struct request_queue *q = bdev_get_queue(dev->bdev);
1256 
1257 	return q && (q->flush_flags & flush);
1258 }
1259 
1260 static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1261 {
1262 	struct dm_target *ti;
1263 	unsigned i = 0;
1264 
1265 	/*
1266 	 * Require at least one underlying device to support flushes.
1267 	 * t->devices includes internal dm devices such as mirror logs
1268 	 * so we need to use iterate_devices here, which targets
1269 	 * supporting flushes must provide.
1270 	 */
1271 	while (i < dm_table_get_num_targets(t)) {
1272 		ti = dm_table_get_target(t, i++);
1273 
1274 		if (!ti->num_flush_requests)
1275 			continue;
1276 
1277 		if (ti->type->iterate_devices &&
1278 		    ti->type->iterate_devices(ti, device_flush_capable, &flush))
1279 			return 1;
1280 	}
1281 
1282 	return 0;
1283 }
1284 
1285 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1286 			       struct queue_limits *limits)
1287 {
1288 	unsigned flush = 0;
1289 
1290 	/*
1291 	 * Copy table's limits to the DM device's request_queue
1292 	 */
1293 	q->limits = *limits;
1294 
1295 	if (!dm_table_supports_discards(t))
1296 		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1297 	else
1298 		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1299 
1300 	if (dm_table_supports_flush(t, REQ_FLUSH)) {
1301 		flush |= REQ_FLUSH;
1302 		if (dm_table_supports_flush(t, REQ_FUA))
1303 			flush |= REQ_FUA;
1304 	}
1305 	blk_queue_flush(q, flush);
1306 
1307 	dm_table_set_integrity(t);
1308 
1309 	/*
1310 	 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1311 	 * visible to other CPUs because, once the flag is set, incoming bios
1312 	 * are processed by request-based dm, which refers to the queue
1313 	 * settings.
1314 	 * Until the flag set, bios are passed to bio-based dm and queued to
1315 	 * md->deferred where queue settings are not needed yet.
1316 	 * Those bios are passed to request-based dm at the resume time.
1317 	 */
1318 	smp_mb();
1319 	if (dm_table_request_based(t))
1320 		queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1321 }
1322 
1323 unsigned int dm_table_get_num_targets(struct dm_table *t)
1324 {
1325 	return t->num_targets;
1326 }
1327 
1328 struct list_head *dm_table_get_devices(struct dm_table *t)
1329 {
1330 	return &t->devices;
1331 }
1332 
1333 fmode_t dm_table_get_mode(struct dm_table *t)
1334 {
1335 	return t->mode;
1336 }
1337 EXPORT_SYMBOL(dm_table_get_mode);
1338 
1339 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1340 {
1341 	int i = t->num_targets;
1342 	struct dm_target *ti = t->targets;
1343 
1344 	while (i--) {
1345 		if (postsuspend) {
1346 			if (ti->type->postsuspend)
1347 				ti->type->postsuspend(ti);
1348 		} else if (ti->type->presuspend)
1349 			ti->type->presuspend(ti);
1350 
1351 		ti++;
1352 	}
1353 }
1354 
1355 void dm_table_presuspend_targets(struct dm_table *t)
1356 {
1357 	if (!t)
1358 		return;
1359 
1360 	suspend_targets(t, 0);
1361 }
1362 
1363 void dm_table_postsuspend_targets(struct dm_table *t)
1364 {
1365 	if (!t)
1366 		return;
1367 
1368 	suspend_targets(t, 1);
1369 }
1370 
1371 int dm_table_resume_targets(struct dm_table *t)
1372 {
1373 	int i, r = 0;
1374 
1375 	for (i = 0; i < t->num_targets; i++) {
1376 		struct dm_target *ti = t->targets + i;
1377 
1378 		if (!ti->type->preresume)
1379 			continue;
1380 
1381 		r = ti->type->preresume(ti);
1382 		if (r)
1383 			return r;
1384 	}
1385 
1386 	for (i = 0; i < t->num_targets; i++) {
1387 		struct dm_target *ti = t->targets + i;
1388 
1389 		if (ti->type->resume)
1390 			ti->type->resume(ti);
1391 	}
1392 
1393 	return 0;
1394 }
1395 
1396 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1397 {
1398 	list_add(&cb->list, &t->target_callbacks);
1399 }
1400 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1401 
1402 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1403 {
1404 	struct dm_dev_internal *dd;
1405 	struct list_head *devices = dm_table_get_devices(t);
1406 	struct dm_target_callbacks *cb;
1407 	int r = 0;
1408 
1409 	list_for_each_entry(dd, devices, list) {
1410 		struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1411 		char b[BDEVNAME_SIZE];
1412 
1413 		if (likely(q))
1414 			r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1415 		else
1416 			DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1417 				     dm_device_name(t->md),
1418 				     bdevname(dd->dm_dev.bdev, b));
1419 	}
1420 
1421 	list_for_each_entry(cb, &t->target_callbacks, list)
1422 		if (cb->congested_fn)
1423 			r |= cb->congested_fn(cb, bdi_bits);
1424 
1425 	return r;
1426 }
1427 
1428 int dm_table_any_busy_target(struct dm_table *t)
1429 {
1430 	unsigned i;
1431 	struct dm_target *ti;
1432 
1433 	for (i = 0; i < t->num_targets; i++) {
1434 		ti = t->targets + i;
1435 		if (ti->type->busy && ti->type->busy(ti))
1436 			return 1;
1437 	}
1438 
1439 	return 0;
1440 }
1441 
1442 struct mapped_device *dm_table_get_md(struct dm_table *t)
1443 {
1444 	return t->md;
1445 }
1446 EXPORT_SYMBOL(dm_table_get_md);
1447 
1448 static int device_discard_capable(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 
1453 	return q && blk_queue_discard(q);
1454 }
1455 
1456 bool dm_table_supports_discards(struct dm_table *t)
1457 {
1458 	struct dm_target *ti;
1459 	unsigned i = 0;
1460 
1461 	/*
1462 	 * Unless any target used by the table set discards_supported,
1463 	 * require at least one underlying device to support discards.
1464 	 * t->devices includes internal dm devices such as mirror logs
1465 	 * so we need to use iterate_devices here, which targets
1466 	 * supporting discard selectively must provide.
1467 	 */
1468 	while (i < dm_table_get_num_targets(t)) {
1469 		ti = dm_table_get_target(t, i++);
1470 
1471 		if (!ti->num_discard_requests)
1472 			continue;
1473 
1474 		if (ti->discards_supported)
1475 			return 1;
1476 
1477 		if (ti->type->iterate_devices &&
1478 		    ti->type->iterate_devices(ti, device_discard_capable, NULL))
1479 			return 1;
1480 	}
1481 
1482 	return 0;
1483 }
1484