xref: /linux/drivers/md/dm-table.c (revision 98366c20a275e957416e9516db5dcb7195b4e101)
1 /*
2  * Copyright (C) 2001 Sistina Software (UK) Limited.
3  * Copyright (C) 2004 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/slab.h>
16 #include <linux/interrupt.h>
17 #include <linux/mutex.h>
18 #include <asm/atomic.h>
19 
20 #define DM_MSG_PREFIX "table"
21 
22 #define MAX_DEPTH 16
23 #define NODE_SIZE L1_CACHE_BYTES
24 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
25 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
26 
27 struct dm_table {
28 	struct mapped_device *md;
29 	atomic_t holders;
30 
31 	/* btree table */
32 	unsigned int depth;
33 	unsigned int counts[MAX_DEPTH];	/* in nodes */
34 	sector_t *index[MAX_DEPTH];
35 
36 	unsigned int num_targets;
37 	unsigned int num_allocated;
38 	sector_t *highs;
39 	struct dm_target *targets;
40 
41 	/*
42 	 * Indicates the rw permissions for the new logical
43 	 * device.  This should be a combination of FMODE_READ
44 	 * and FMODE_WRITE.
45 	 */
46 	int mode;
47 
48 	/* a list of devices used by this table */
49 	struct list_head devices;
50 
51 	/*
52 	 * These are optimistic limits taken from all the
53 	 * targets, some targets will need smaller limits.
54 	 */
55 	struct io_restrictions limits;
56 
57 	/* events get handed up using this callback */
58 	void (*event_fn)(void *);
59 	void *event_context;
60 };
61 
62 /*
63  * Similar to ceiling(log_size(n))
64  */
65 static unsigned int int_log(unsigned int n, unsigned int base)
66 {
67 	int result = 0;
68 
69 	while (n > 1) {
70 		n = dm_div_up(n, base);
71 		result++;
72 	}
73 
74 	return result;
75 }
76 
77 /*
78  * Returns the minimum that is _not_ zero, unless both are zero.
79  */
80 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
81 
82 /*
83  * Combine two io_restrictions, always taking the lower value.
84  */
85 static void combine_restrictions_low(struct io_restrictions *lhs,
86 				     struct io_restrictions *rhs)
87 {
88 	lhs->max_sectors =
89 		min_not_zero(lhs->max_sectors, rhs->max_sectors);
90 
91 	lhs->max_phys_segments =
92 		min_not_zero(lhs->max_phys_segments, rhs->max_phys_segments);
93 
94 	lhs->max_hw_segments =
95 		min_not_zero(lhs->max_hw_segments, rhs->max_hw_segments);
96 
97 	lhs->hardsect_size = max(lhs->hardsect_size, rhs->hardsect_size);
98 
99 	lhs->max_segment_size =
100 		min_not_zero(lhs->max_segment_size, rhs->max_segment_size);
101 
102 	lhs->seg_boundary_mask =
103 		min_not_zero(lhs->seg_boundary_mask, rhs->seg_boundary_mask);
104 
105 	lhs->bounce_pfn = min_not_zero(lhs->bounce_pfn, rhs->bounce_pfn);
106 
107 	lhs->no_cluster |= rhs->no_cluster;
108 }
109 
110 /*
111  * Calculate the index of the child node of the n'th node k'th key.
112  */
113 static inline unsigned int get_child(unsigned int n, unsigned int k)
114 {
115 	return (n * CHILDREN_PER_NODE) + k;
116 }
117 
118 /*
119  * Return the n'th node of level l from table t.
120  */
121 static inline sector_t *get_node(struct dm_table *t,
122 				 unsigned int l, unsigned int n)
123 {
124 	return t->index[l] + (n * KEYS_PER_NODE);
125 }
126 
127 /*
128  * Return the highest key that you could lookup from the n'th
129  * node on level l of the btree.
130  */
131 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
132 {
133 	for (; l < t->depth - 1; l++)
134 		n = get_child(n, CHILDREN_PER_NODE - 1);
135 
136 	if (n >= t->counts[l])
137 		return (sector_t) - 1;
138 
139 	return get_node(t, l, n)[KEYS_PER_NODE - 1];
140 }
141 
142 /*
143  * Fills in a level of the btree based on the highs of the level
144  * below it.
145  */
146 static int setup_btree_index(unsigned int l, struct dm_table *t)
147 {
148 	unsigned int n, k;
149 	sector_t *node;
150 
151 	for (n = 0U; n < t->counts[l]; n++) {
152 		node = get_node(t, l, n);
153 
154 		for (k = 0U; k < KEYS_PER_NODE; k++)
155 			node[k] = high(t, l + 1, get_child(n, k));
156 	}
157 
158 	return 0;
159 }
160 
161 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
162 {
163 	unsigned long size;
164 	void *addr;
165 
166 	/*
167 	 * Check that we're not going to overflow.
168 	 */
169 	if (nmemb > (ULONG_MAX / elem_size))
170 		return NULL;
171 
172 	size = nmemb * elem_size;
173 	addr = vmalloc(size);
174 	if (addr)
175 		memset(addr, 0, size);
176 
177 	return addr;
178 }
179 
180 /*
181  * highs, and targets are managed as dynamic arrays during a
182  * table load.
183  */
184 static int alloc_targets(struct dm_table *t, unsigned int num)
185 {
186 	sector_t *n_highs;
187 	struct dm_target *n_targets;
188 	int n = t->num_targets;
189 
190 	/*
191 	 * Allocate both the target array and offset array at once.
192 	 */
193 	n_highs = (sector_t *) dm_vcalloc(num, sizeof(struct dm_target) +
194 					  sizeof(sector_t));
195 	if (!n_highs)
196 		return -ENOMEM;
197 
198 	n_targets = (struct dm_target *) (n_highs + num);
199 
200 	if (n) {
201 		memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
202 		memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
203 	}
204 
205 	memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
206 	vfree(t->highs);
207 
208 	t->num_allocated = num;
209 	t->highs = n_highs;
210 	t->targets = n_targets;
211 
212 	return 0;
213 }
214 
215 int dm_table_create(struct dm_table **result, int mode,
216 		    unsigned num_targets, struct mapped_device *md)
217 {
218 	struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
219 
220 	if (!t)
221 		return -ENOMEM;
222 
223 	INIT_LIST_HEAD(&t->devices);
224 	atomic_set(&t->holders, 1);
225 
226 	if (!num_targets)
227 		num_targets = KEYS_PER_NODE;
228 
229 	num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
230 
231 	if (alloc_targets(t, num_targets)) {
232 		kfree(t);
233 		t = NULL;
234 		return -ENOMEM;
235 	}
236 
237 	t->mode = mode;
238 	t->md = md;
239 	*result = t;
240 	return 0;
241 }
242 
243 int dm_create_error_table(struct dm_table **result, struct mapped_device *md)
244 {
245 	struct dm_table *t;
246 	sector_t dev_size = 1;
247 	int r;
248 
249 	/*
250 	 * Find current size of device.
251 	 * Default to 1 sector if inactive.
252 	 */
253 	t = dm_get_table(md);
254 	if (t) {
255 		dev_size = dm_table_get_size(t);
256 		dm_table_put(t);
257 	}
258 
259 	r = dm_table_create(&t, FMODE_READ, 1, md);
260 	if (r)
261 		return r;
262 
263 	r = dm_table_add_target(t, "error", 0, dev_size, NULL);
264 	if (r)
265 		goto out;
266 
267 	r = dm_table_complete(t);
268 	if (r)
269 		goto out;
270 
271 	*result = t;
272 
273 out:
274 	if (r)
275 		dm_table_put(t);
276 
277 	return r;
278 }
279 EXPORT_SYMBOL_GPL(dm_create_error_table);
280 
281 static void free_devices(struct list_head *devices)
282 {
283 	struct list_head *tmp, *next;
284 
285 	for (tmp = devices->next; tmp != devices; tmp = next) {
286 		struct dm_dev *dd = list_entry(tmp, struct dm_dev, list);
287 		next = tmp->next;
288 		kfree(dd);
289 	}
290 }
291 
292 static void table_destroy(struct dm_table *t)
293 {
294 	unsigned int i;
295 
296 	/* free the indexes (see dm_table_complete) */
297 	if (t->depth >= 2)
298 		vfree(t->index[t->depth - 2]);
299 
300 	/* free the targets */
301 	for (i = 0; i < t->num_targets; i++) {
302 		struct dm_target *tgt = t->targets + i;
303 
304 		if (tgt->type->dtr)
305 			tgt->type->dtr(tgt);
306 
307 		dm_put_target_type(tgt->type);
308 	}
309 
310 	vfree(t->highs);
311 
312 	/* free the device list */
313 	if (t->devices.next != &t->devices) {
314 		DMWARN("devices still present during destroy: "
315 		       "dm_table_remove_device calls missing");
316 
317 		free_devices(&t->devices);
318 	}
319 
320 	kfree(t);
321 }
322 
323 void dm_table_get(struct dm_table *t)
324 {
325 	atomic_inc(&t->holders);
326 }
327 
328 void dm_table_put(struct dm_table *t)
329 {
330 	if (!t)
331 		return;
332 
333 	if (atomic_dec_and_test(&t->holders))
334 		table_destroy(t);
335 }
336 
337 /*
338  * Checks to see if we need to extend highs or targets.
339  */
340 static inline int check_space(struct dm_table *t)
341 {
342 	if (t->num_targets >= t->num_allocated)
343 		return alloc_targets(t, t->num_allocated * 2);
344 
345 	return 0;
346 }
347 
348 /*
349  * Convert a device path to a dev_t.
350  */
351 static int lookup_device(const char *path, dev_t *dev)
352 {
353 	int r;
354 	struct nameidata nd;
355 	struct inode *inode;
356 
357 	if ((r = path_lookup(path, LOOKUP_FOLLOW, &nd)))
358 		return r;
359 
360 	inode = nd.dentry->d_inode;
361 	if (!inode) {
362 		r = -ENOENT;
363 		goto out;
364 	}
365 
366 	if (!S_ISBLK(inode->i_mode)) {
367 		r = -ENOTBLK;
368 		goto out;
369 	}
370 
371 	*dev = inode->i_rdev;
372 
373  out:
374 	path_release(&nd);
375 	return r;
376 }
377 
378 /*
379  * See if we've already got a device in the list.
380  */
381 static struct dm_dev *find_device(struct list_head *l, dev_t dev)
382 {
383 	struct dm_dev *dd;
384 
385 	list_for_each_entry (dd, l, list)
386 		if (dd->bdev->bd_dev == dev)
387 			return dd;
388 
389 	return NULL;
390 }
391 
392 /*
393  * Open a device so we can use it as a map destination.
394  */
395 static int open_dev(struct dm_dev *d, dev_t dev, struct mapped_device *md)
396 {
397 	static char *_claim_ptr = "I belong to device-mapper";
398 	struct block_device *bdev;
399 
400 	int r;
401 
402 	BUG_ON(d->bdev);
403 
404 	bdev = open_by_devnum(dev, d->mode);
405 	if (IS_ERR(bdev))
406 		return PTR_ERR(bdev);
407 	r = bd_claim_by_disk(bdev, _claim_ptr, dm_disk(md));
408 	if (r)
409 		blkdev_put(bdev);
410 	else
411 		d->bdev = bdev;
412 	return r;
413 }
414 
415 /*
416  * Close a device that we've been using.
417  */
418 static void close_dev(struct dm_dev *d, struct mapped_device *md)
419 {
420 	if (!d->bdev)
421 		return;
422 
423 	bd_release_from_disk(d->bdev, dm_disk(md));
424 	blkdev_put(d->bdev);
425 	d->bdev = NULL;
426 }
427 
428 /*
429  * If possible, this checks an area of a destination device is valid.
430  */
431 static int check_device_area(struct dm_dev *dd, sector_t start, sector_t len)
432 {
433 	sector_t dev_size = dd->bdev->bd_inode->i_size >> SECTOR_SHIFT;
434 
435 	if (!dev_size)
436 		return 1;
437 
438 	return ((start < dev_size) && (len <= (dev_size - start)));
439 }
440 
441 /*
442  * This upgrades the mode on an already open dm_dev.  Being
443  * careful to leave things as they were if we fail to reopen the
444  * device.
445  */
446 static int upgrade_mode(struct dm_dev *dd, int new_mode, struct mapped_device *md)
447 {
448 	int r;
449 	struct dm_dev dd_copy;
450 	dev_t dev = dd->bdev->bd_dev;
451 
452 	dd_copy = *dd;
453 
454 	dd->mode |= new_mode;
455 	dd->bdev = NULL;
456 	r = open_dev(dd, dev, md);
457 	if (!r)
458 		close_dev(&dd_copy, md);
459 	else
460 		*dd = dd_copy;
461 
462 	return r;
463 }
464 
465 /*
466  * Add a device to the list, or just increment the usage count if
467  * it's already present.
468  */
469 static int __table_get_device(struct dm_table *t, struct dm_target *ti,
470 			      const char *path, sector_t start, sector_t len,
471 			      int mode, struct dm_dev **result)
472 {
473 	int r;
474 	dev_t dev;
475 	struct dm_dev *dd;
476 	unsigned int major, minor;
477 
478 	BUG_ON(!t);
479 
480 	if (sscanf(path, "%u:%u", &major, &minor) == 2) {
481 		/* Extract the major/minor numbers */
482 		dev = MKDEV(major, minor);
483 		if (MAJOR(dev) != major || MINOR(dev) != minor)
484 			return -EOVERFLOW;
485 	} else {
486 		/* convert the path to a device */
487 		if ((r = lookup_device(path, &dev)))
488 			return r;
489 	}
490 
491 	dd = find_device(&t->devices, dev);
492 	if (!dd) {
493 		dd = kmalloc(sizeof(*dd), GFP_KERNEL);
494 		if (!dd)
495 			return -ENOMEM;
496 
497 		dd->mode = mode;
498 		dd->bdev = NULL;
499 
500 		if ((r = open_dev(dd, dev, t->md))) {
501 			kfree(dd);
502 			return r;
503 		}
504 
505 		format_dev_t(dd->name, dev);
506 
507 		atomic_set(&dd->count, 0);
508 		list_add(&dd->list, &t->devices);
509 
510 	} else if (dd->mode != (mode | dd->mode)) {
511 		r = upgrade_mode(dd, mode, t->md);
512 		if (r)
513 			return r;
514 	}
515 	atomic_inc(&dd->count);
516 
517 	if (!check_device_area(dd, start, len)) {
518 		DMWARN("device %s too small for target", path);
519 		dm_put_device(ti, dd);
520 		return -EINVAL;
521 	}
522 
523 	*result = dd;
524 
525 	return 0;
526 }
527 
528 void dm_set_device_limits(struct dm_target *ti, struct block_device *bdev)
529 {
530 	struct request_queue *q = bdev_get_queue(bdev);
531 	struct io_restrictions *rs = &ti->limits;
532 
533 	/*
534 	 * Combine the device limits low.
535 	 *
536 	 * FIXME: if we move an io_restriction struct
537 	 *        into q this would just be a call to
538 	 *        combine_restrictions_low()
539 	 */
540 	rs->max_sectors =
541 		min_not_zero(rs->max_sectors, q->max_sectors);
542 
543 	/* FIXME: Device-Mapper on top of RAID-0 breaks because DM
544 	 *        currently doesn't honor MD's merge_bvec_fn routine.
545 	 *        In this case, we'll force DM to use PAGE_SIZE or
546 	 *        smaller I/O, just to be safe. A better fix is in the
547 	 *        works, but add this for the time being so it will at
548 	 *        least operate correctly.
549 	 */
550 	if (q->merge_bvec_fn)
551 		rs->max_sectors =
552 			min_not_zero(rs->max_sectors,
553 				     (unsigned int) (PAGE_SIZE >> 9));
554 
555 	rs->max_phys_segments =
556 		min_not_zero(rs->max_phys_segments,
557 			     q->max_phys_segments);
558 
559 	rs->max_hw_segments =
560 		min_not_zero(rs->max_hw_segments, q->max_hw_segments);
561 
562 	rs->hardsect_size = max(rs->hardsect_size, q->hardsect_size);
563 
564 	rs->max_segment_size =
565 		min_not_zero(rs->max_segment_size, q->max_segment_size);
566 
567 	rs->seg_boundary_mask =
568 		min_not_zero(rs->seg_boundary_mask,
569 			     q->seg_boundary_mask);
570 
571 	rs->bounce_pfn = min_not_zero(rs->bounce_pfn, q->bounce_pfn);
572 
573 	rs->no_cluster |= !test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
574 }
575 EXPORT_SYMBOL_GPL(dm_set_device_limits);
576 
577 int dm_get_device(struct dm_target *ti, const char *path, sector_t start,
578 		  sector_t len, int mode, struct dm_dev **result)
579 {
580 	int r = __table_get_device(ti->table, ti, path,
581 				   start, len, mode, result);
582 
583 	if (!r)
584 		dm_set_device_limits(ti, (*result)->bdev);
585 
586 	return r;
587 }
588 
589 /*
590  * Decrement a devices use count and remove it if necessary.
591  */
592 void dm_put_device(struct dm_target *ti, struct dm_dev *dd)
593 {
594 	if (atomic_dec_and_test(&dd->count)) {
595 		close_dev(dd, ti->table->md);
596 		list_del(&dd->list);
597 		kfree(dd);
598 	}
599 }
600 
601 /*
602  * Checks to see if the target joins onto the end of the table.
603  */
604 static int adjoin(struct dm_table *table, struct dm_target *ti)
605 {
606 	struct dm_target *prev;
607 
608 	if (!table->num_targets)
609 		return !ti->begin;
610 
611 	prev = &table->targets[table->num_targets - 1];
612 	return (ti->begin == (prev->begin + prev->len));
613 }
614 
615 /*
616  * Used to dynamically allocate the arg array.
617  */
618 static char **realloc_argv(unsigned *array_size, char **old_argv)
619 {
620 	char **argv;
621 	unsigned new_size;
622 
623 	new_size = *array_size ? *array_size * 2 : 64;
624 	argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
625 	if (argv) {
626 		memcpy(argv, old_argv, *array_size * sizeof(*argv));
627 		*array_size = new_size;
628 	}
629 
630 	kfree(old_argv);
631 	return argv;
632 }
633 
634 /*
635  * Destructively splits up the argument list to pass to ctr.
636  */
637 int dm_split_args(int *argc, char ***argvp, char *input)
638 {
639 	char *start, *end = input, *out, **argv = NULL;
640 	unsigned array_size = 0;
641 
642 	*argc = 0;
643 
644 	if (!input) {
645 		*argvp = NULL;
646 		return 0;
647 	}
648 
649 	argv = realloc_argv(&array_size, argv);
650 	if (!argv)
651 		return -ENOMEM;
652 
653 	while (1) {
654 		start = end;
655 
656 		/* Skip whitespace */
657 		while (*start && isspace(*start))
658 			start++;
659 
660 		if (!*start)
661 			break;	/* success, we hit the end */
662 
663 		/* 'out' is used to remove any back-quotes */
664 		end = out = start;
665 		while (*end) {
666 			/* Everything apart from '\0' can be quoted */
667 			if (*end == '\\' && *(end + 1)) {
668 				*out++ = *(end + 1);
669 				end += 2;
670 				continue;
671 			}
672 
673 			if (isspace(*end))
674 				break;	/* end of token */
675 
676 			*out++ = *end++;
677 		}
678 
679 		/* have we already filled the array ? */
680 		if ((*argc + 1) > array_size) {
681 			argv = realloc_argv(&array_size, argv);
682 			if (!argv)
683 				return -ENOMEM;
684 		}
685 
686 		/* we know this is whitespace */
687 		if (*end)
688 			end++;
689 
690 		/* terminate the string and put it in the array */
691 		*out = '\0';
692 		argv[*argc] = start;
693 		(*argc)++;
694 	}
695 
696 	*argvp = argv;
697 	return 0;
698 }
699 
700 static void check_for_valid_limits(struct io_restrictions *rs)
701 {
702 	if (!rs->max_sectors)
703 		rs->max_sectors = SAFE_MAX_SECTORS;
704 	if (!rs->max_phys_segments)
705 		rs->max_phys_segments = MAX_PHYS_SEGMENTS;
706 	if (!rs->max_hw_segments)
707 		rs->max_hw_segments = MAX_HW_SEGMENTS;
708 	if (!rs->hardsect_size)
709 		rs->hardsect_size = 1 << SECTOR_SHIFT;
710 	if (!rs->max_segment_size)
711 		rs->max_segment_size = MAX_SEGMENT_SIZE;
712 	if (!rs->seg_boundary_mask)
713 		rs->seg_boundary_mask = -1;
714 	if (!rs->bounce_pfn)
715 		rs->bounce_pfn = -1;
716 }
717 
718 int dm_table_add_target(struct dm_table *t, const char *type,
719 			sector_t start, sector_t len, char *params)
720 {
721 	int r = -EINVAL, argc;
722 	char **argv;
723 	struct dm_target *tgt;
724 
725 	if ((r = check_space(t)))
726 		return r;
727 
728 	tgt = t->targets + t->num_targets;
729 	memset(tgt, 0, sizeof(*tgt));
730 
731 	if (!len) {
732 		DMERR("%s: zero-length target", dm_device_name(t->md));
733 		return -EINVAL;
734 	}
735 
736 	tgt->type = dm_get_target_type(type);
737 	if (!tgt->type) {
738 		DMERR("%s: %s: unknown target type", dm_device_name(t->md),
739 		      type);
740 		return -EINVAL;
741 	}
742 
743 	tgt->table = t;
744 	tgt->begin = start;
745 	tgt->len = len;
746 	tgt->error = "Unknown error";
747 
748 	/*
749 	 * Does this target adjoin the previous one ?
750 	 */
751 	if (!adjoin(t, tgt)) {
752 		tgt->error = "Gap in table";
753 		r = -EINVAL;
754 		goto bad;
755 	}
756 
757 	r = dm_split_args(&argc, &argv, params);
758 	if (r) {
759 		tgt->error = "couldn't split parameters (insufficient memory)";
760 		goto bad;
761 	}
762 
763 	r = tgt->type->ctr(tgt, argc, argv);
764 	kfree(argv);
765 	if (r)
766 		goto bad;
767 
768 	t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
769 
770 	/* FIXME: the plan is to combine high here and then have
771 	 * the merge fn apply the target level restrictions. */
772 	combine_restrictions_low(&t->limits, &tgt->limits);
773 	return 0;
774 
775  bad:
776 	DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
777 	dm_put_target_type(tgt->type);
778 	return r;
779 }
780 
781 static int setup_indexes(struct dm_table *t)
782 {
783 	int i;
784 	unsigned int total = 0;
785 	sector_t *indexes;
786 
787 	/* allocate the space for *all* the indexes */
788 	for (i = t->depth - 2; i >= 0; i--) {
789 		t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
790 		total += t->counts[i];
791 	}
792 
793 	indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
794 	if (!indexes)
795 		return -ENOMEM;
796 
797 	/* set up internal nodes, bottom-up */
798 	for (i = t->depth - 2, total = 0; i >= 0; i--) {
799 		t->index[i] = indexes;
800 		indexes += (KEYS_PER_NODE * t->counts[i]);
801 		setup_btree_index(i, t);
802 	}
803 
804 	return 0;
805 }
806 
807 /*
808  * Builds the btree to index the map.
809  */
810 int dm_table_complete(struct dm_table *t)
811 {
812 	int r = 0;
813 	unsigned int leaf_nodes;
814 
815 	check_for_valid_limits(&t->limits);
816 
817 	/* how many indexes will the btree have ? */
818 	leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
819 	t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
820 
821 	/* leaf layer has already been set up */
822 	t->counts[t->depth - 1] = leaf_nodes;
823 	t->index[t->depth - 1] = t->highs;
824 
825 	if (t->depth >= 2)
826 		r = setup_indexes(t);
827 
828 	return r;
829 }
830 
831 static DEFINE_MUTEX(_event_lock);
832 void dm_table_event_callback(struct dm_table *t,
833 			     void (*fn)(void *), void *context)
834 {
835 	mutex_lock(&_event_lock);
836 	t->event_fn = fn;
837 	t->event_context = context;
838 	mutex_unlock(&_event_lock);
839 }
840 
841 void dm_table_event(struct dm_table *t)
842 {
843 	/*
844 	 * You can no longer call dm_table_event() from interrupt
845 	 * context, use a bottom half instead.
846 	 */
847 	BUG_ON(in_interrupt());
848 
849 	mutex_lock(&_event_lock);
850 	if (t->event_fn)
851 		t->event_fn(t->event_context);
852 	mutex_unlock(&_event_lock);
853 }
854 
855 sector_t dm_table_get_size(struct dm_table *t)
856 {
857 	return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
858 }
859 
860 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
861 {
862 	if (index >= t->num_targets)
863 		return NULL;
864 
865 	return t->targets + index;
866 }
867 
868 /*
869  * Search the btree for the correct target.
870  */
871 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
872 {
873 	unsigned int l, n = 0, k = 0;
874 	sector_t *node;
875 
876 	for (l = 0; l < t->depth; l++) {
877 		n = get_child(n, k);
878 		node = get_node(t, l, n);
879 
880 		for (k = 0; k < KEYS_PER_NODE; k++)
881 			if (node[k] >= sector)
882 				break;
883 	}
884 
885 	return &t->targets[(KEYS_PER_NODE * n) + k];
886 }
887 
888 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q)
889 {
890 	/*
891 	 * Make sure we obey the optimistic sub devices
892 	 * restrictions.
893 	 */
894 	blk_queue_max_sectors(q, t->limits.max_sectors);
895 	q->max_phys_segments = t->limits.max_phys_segments;
896 	q->max_hw_segments = t->limits.max_hw_segments;
897 	q->hardsect_size = t->limits.hardsect_size;
898 	q->max_segment_size = t->limits.max_segment_size;
899 	q->seg_boundary_mask = t->limits.seg_boundary_mask;
900 	q->bounce_pfn = t->limits.bounce_pfn;
901 	if (t->limits.no_cluster)
902 		q->queue_flags &= ~(1 << QUEUE_FLAG_CLUSTER);
903 	else
904 		q->queue_flags |= (1 << QUEUE_FLAG_CLUSTER);
905 
906 }
907 
908 unsigned int dm_table_get_num_targets(struct dm_table *t)
909 {
910 	return t->num_targets;
911 }
912 
913 struct list_head *dm_table_get_devices(struct dm_table *t)
914 {
915 	return &t->devices;
916 }
917 
918 int dm_table_get_mode(struct dm_table *t)
919 {
920 	return t->mode;
921 }
922 
923 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
924 {
925 	int i = t->num_targets;
926 	struct dm_target *ti = t->targets;
927 
928 	while (i--) {
929 		if (postsuspend) {
930 			if (ti->type->postsuspend)
931 				ti->type->postsuspend(ti);
932 		} else if (ti->type->presuspend)
933 			ti->type->presuspend(ti);
934 
935 		ti++;
936 	}
937 }
938 
939 void dm_table_presuspend_targets(struct dm_table *t)
940 {
941 	if (!t)
942 		return;
943 
944 	return suspend_targets(t, 0);
945 }
946 
947 void dm_table_postsuspend_targets(struct dm_table *t)
948 {
949 	if (!t)
950 		return;
951 
952 	return suspend_targets(t, 1);
953 }
954 
955 int dm_table_resume_targets(struct dm_table *t)
956 {
957 	int i, r = 0;
958 
959 	for (i = 0; i < t->num_targets; i++) {
960 		struct dm_target *ti = t->targets + i;
961 
962 		if (!ti->type->preresume)
963 			continue;
964 
965 		r = ti->type->preresume(ti);
966 		if (r)
967 			return r;
968 	}
969 
970 	for (i = 0; i < t->num_targets; i++) {
971 		struct dm_target *ti = t->targets + i;
972 
973 		if (ti->type->resume)
974 			ti->type->resume(ti);
975 	}
976 
977 	return 0;
978 }
979 
980 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
981 {
982 	struct list_head *d, *devices;
983 	int r = 0;
984 
985 	devices = dm_table_get_devices(t);
986 	for (d = devices->next; d != devices; d = d->next) {
987 		struct dm_dev *dd = list_entry(d, struct dm_dev, list);
988 		struct request_queue *q = bdev_get_queue(dd->bdev);
989 		r |= bdi_congested(&q->backing_dev_info, bdi_bits);
990 	}
991 
992 	return r;
993 }
994 
995 void dm_table_unplug_all(struct dm_table *t)
996 {
997 	struct list_head *d, *devices = dm_table_get_devices(t);
998 
999 	for (d = devices->next; d != devices; d = d->next) {
1000 		struct dm_dev *dd = list_entry(d, struct dm_dev, list);
1001 		struct request_queue *q = bdev_get_queue(dd->bdev);
1002 
1003 		if (q->unplug_fn)
1004 			q->unplug_fn(q);
1005 	}
1006 }
1007 
1008 struct mapped_device *dm_table_get_md(struct dm_table *t)
1009 {
1010 	dm_get(t->md);
1011 
1012 	return t->md;
1013 }
1014 
1015 EXPORT_SYMBOL(dm_vcalloc);
1016 EXPORT_SYMBOL(dm_get_device);
1017 EXPORT_SYMBOL(dm_put_device);
1018 EXPORT_SYMBOL(dm_table_event);
1019 EXPORT_SYMBOL(dm_table_get_size);
1020 EXPORT_SYMBOL(dm_table_get_mode);
1021 EXPORT_SYMBOL(dm_table_get_md);
1022 EXPORT_SYMBOL(dm_table_put);
1023 EXPORT_SYMBOL(dm_table_get);
1024 EXPORT_SYMBOL(dm_table_unplug_all);
1025