xref: /openbmc/linux/drivers/md/dm-table.c (revision 96de0e252cedffad61b3cb5e05662c591898e69a)
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->no_cluster |= rhs->no_cluster;
106 }
107 
108 /*
109  * Calculate the index of the child node of the n'th node k'th key.
110  */
111 static inline unsigned int get_child(unsigned int n, unsigned int k)
112 {
113 	return (n * CHILDREN_PER_NODE) + k;
114 }
115 
116 /*
117  * Return the n'th node of level l from table t.
118  */
119 static inline sector_t *get_node(struct dm_table *t,
120 				 unsigned int l, unsigned int n)
121 {
122 	return t->index[l] + (n * KEYS_PER_NODE);
123 }
124 
125 /*
126  * Return the highest key that you could lookup from the n'th
127  * node on level l of the btree.
128  */
129 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
130 {
131 	for (; l < t->depth - 1; l++)
132 		n = get_child(n, CHILDREN_PER_NODE - 1);
133 
134 	if (n >= t->counts[l])
135 		return (sector_t) - 1;
136 
137 	return get_node(t, l, n)[KEYS_PER_NODE - 1];
138 }
139 
140 /*
141  * Fills in a level of the btree based on the highs of the level
142  * below it.
143  */
144 static int setup_btree_index(unsigned int l, struct dm_table *t)
145 {
146 	unsigned int n, k;
147 	sector_t *node;
148 
149 	for (n = 0U; n < t->counts[l]; n++) {
150 		node = get_node(t, l, n);
151 
152 		for (k = 0U; k < KEYS_PER_NODE; k++)
153 			node[k] = high(t, l + 1, get_child(n, k));
154 	}
155 
156 	return 0;
157 }
158 
159 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
160 {
161 	unsigned long size;
162 	void *addr;
163 
164 	/*
165 	 * Check that we're not going to overflow.
166 	 */
167 	if (nmemb > (ULONG_MAX / elem_size))
168 		return NULL;
169 
170 	size = nmemb * elem_size;
171 	addr = vmalloc(size);
172 	if (addr)
173 		memset(addr, 0, size);
174 
175 	return addr;
176 }
177 
178 /*
179  * highs, and targets are managed as dynamic arrays during a
180  * table load.
181  */
182 static int alloc_targets(struct dm_table *t, unsigned int num)
183 {
184 	sector_t *n_highs;
185 	struct dm_target *n_targets;
186 	int n = t->num_targets;
187 
188 	/*
189 	 * Allocate both the target array and offset array at once.
190 	 */
191 	n_highs = (sector_t *) dm_vcalloc(num, sizeof(struct dm_target) +
192 					  sizeof(sector_t));
193 	if (!n_highs)
194 		return -ENOMEM;
195 
196 	n_targets = (struct dm_target *) (n_highs + num);
197 
198 	if (n) {
199 		memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
200 		memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
201 	}
202 
203 	memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
204 	vfree(t->highs);
205 
206 	t->num_allocated = num;
207 	t->highs = n_highs;
208 	t->targets = n_targets;
209 
210 	return 0;
211 }
212 
213 int dm_table_create(struct dm_table **result, int mode,
214 		    unsigned num_targets, struct mapped_device *md)
215 {
216 	struct dm_table *t = kmalloc(sizeof(*t), GFP_KERNEL);
217 
218 	if (!t)
219 		return -ENOMEM;
220 
221 	memset(t, 0, sizeof(*t));
222 	INIT_LIST_HEAD(&t->devices);
223 	atomic_set(&t->holders, 1);
224 
225 	if (!num_targets)
226 		num_targets = KEYS_PER_NODE;
227 
228 	num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
229 
230 	if (alloc_targets(t, num_targets)) {
231 		kfree(t);
232 		t = NULL;
233 		return -ENOMEM;
234 	}
235 
236 	t->mode = mode;
237 	t->md = md;
238 	*result = t;
239 	return 0;
240 }
241 
242 int dm_create_error_table(struct dm_table **result, struct mapped_device *md)
243 {
244 	struct dm_table *t;
245 	sector_t dev_size = 1;
246 	int r;
247 
248 	/*
249 	 * Find current size of device.
250 	 * Default to 1 sector if inactive.
251 	 */
252 	t = dm_get_table(md);
253 	if (t) {
254 		dev_size = dm_table_get_size(t);
255 		dm_table_put(t);
256 	}
257 
258 	r = dm_table_create(&t, FMODE_READ, 1, md);
259 	if (r)
260 		return r;
261 
262 	r = dm_table_add_target(t, "error", 0, dev_size, NULL);
263 	if (r)
264 		goto out;
265 
266 	r = dm_table_complete(t);
267 	if (r)
268 		goto out;
269 
270 	*result = t;
271 
272 out:
273 	if (r)
274 		dm_table_put(t);
275 
276 	return r;
277 }
278 EXPORT_SYMBOL_GPL(dm_create_error_table);
279 
280 static void free_devices(struct list_head *devices)
281 {
282 	struct list_head *tmp, *next;
283 
284 	for (tmp = devices->next; tmp != devices; tmp = next) {
285 		struct dm_dev *dd = list_entry(tmp, struct dm_dev, list);
286 		next = tmp->next;
287 		kfree(dd);
288 	}
289 }
290 
291 static void table_destroy(struct dm_table *t)
292 {
293 	unsigned int i;
294 
295 	/* free the indexes (see dm_table_complete) */
296 	if (t->depth >= 2)
297 		vfree(t->index[t->depth - 2]);
298 
299 	/* free the targets */
300 	for (i = 0; i < t->num_targets; i++) {
301 		struct dm_target *tgt = t->targets + i;
302 
303 		if (tgt->type->dtr)
304 			tgt->type->dtr(tgt);
305 
306 		dm_put_target_type(tgt->type);
307 	}
308 
309 	vfree(t->highs);
310 
311 	/* free the device list */
312 	if (t->devices.next != &t->devices) {
313 		DMWARN("devices still present during destroy: "
314 		       "dm_table_remove_device calls missing");
315 
316 		free_devices(&t->devices);
317 	}
318 
319 	kfree(t);
320 }
321 
322 void dm_table_get(struct dm_table *t)
323 {
324 	atomic_inc(&t->holders);
325 }
326 
327 void dm_table_put(struct dm_table *t)
328 {
329 	if (!t)
330 		return;
331 
332 	if (atomic_dec_and_test(&t->holders))
333 		table_destroy(t);
334 }
335 
336 /*
337  * Checks to see if we need to extend highs or targets.
338  */
339 static inline int check_space(struct dm_table *t)
340 {
341 	if (t->num_targets >= t->num_allocated)
342 		return alloc_targets(t, t->num_allocated * 2);
343 
344 	return 0;
345 }
346 
347 /*
348  * Convert a device path to a dev_t.
349  */
350 static int lookup_device(const char *path, dev_t *dev)
351 {
352 	int r;
353 	struct nameidata nd;
354 	struct inode *inode;
355 
356 	if ((r = path_lookup(path, LOOKUP_FOLLOW, &nd)))
357 		return r;
358 
359 	inode = nd.dentry->d_inode;
360 	if (!inode) {
361 		r = -ENOENT;
362 		goto out;
363 	}
364 
365 	if (!S_ISBLK(inode->i_mode)) {
366 		r = -ENOTBLK;
367 		goto out;
368 	}
369 
370 	*dev = inode->i_rdev;
371 
372  out:
373 	path_release(&nd);
374 	return r;
375 }
376 
377 /*
378  * See if we've already got a device in the list.
379  */
380 static struct dm_dev *find_device(struct list_head *l, dev_t dev)
381 {
382 	struct dm_dev *dd;
383 
384 	list_for_each_entry (dd, l, list)
385 		if (dd->bdev->bd_dev == dev)
386 			return dd;
387 
388 	return NULL;
389 }
390 
391 /*
392  * Open a device so we can use it as a map destination.
393  */
394 static int open_dev(struct dm_dev *d, dev_t dev, struct mapped_device *md)
395 {
396 	static char *_claim_ptr = "I belong to device-mapper";
397 	struct block_device *bdev;
398 
399 	int r;
400 
401 	BUG_ON(d->bdev);
402 
403 	bdev = open_by_devnum(dev, d->mode);
404 	if (IS_ERR(bdev))
405 		return PTR_ERR(bdev);
406 	r = bd_claim_by_disk(bdev, _claim_ptr, dm_disk(md));
407 	if (r)
408 		blkdev_put(bdev);
409 	else
410 		d->bdev = bdev;
411 	return r;
412 }
413 
414 /*
415  * Close a device that we've been using.
416  */
417 static void close_dev(struct dm_dev *d, struct mapped_device *md)
418 {
419 	if (!d->bdev)
420 		return;
421 
422 	bd_release_from_disk(d->bdev, dm_disk(md));
423 	blkdev_put(d->bdev);
424 	d->bdev = NULL;
425 }
426 
427 /*
428  * If possible, this checks an area of a destination device is valid.
429  */
430 static int check_device_area(struct dm_dev *dd, sector_t start, sector_t len)
431 {
432 	sector_t dev_size = dd->bdev->bd_inode->i_size >> SECTOR_SHIFT;
433 
434 	if (!dev_size)
435 		return 1;
436 
437 	return ((start < dev_size) && (len <= (dev_size - start)));
438 }
439 
440 /*
441  * This upgrades the mode on an already open dm_dev.  Being
442  * careful to leave things as they were if we fail to reopen the
443  * device.
444  */
445 static int upgrade_mode(struct dm_dev *dd, int new_mode, struct mapped_device *md)
446 {
447 	int r;
448 	struct dm_dev dd_copy;
449 	dev_t dev = dd->bdev->bd_dev;
450 
451 	dd_copy = *dd;
452 
453 	dd->mode |= new_mode;
454 	dd->bdev = NULL;
455 	r = open_dev(dd, dev, md);
456 	if (!r)
457 		close_dev(&dd_copy, md);
458 	else
459 		*dd = dd_copy;
460 
461 	return r;
462 }
463 
464 /*
465  * Add a device to the list, or just increment the usage count if
466  * it's already present.
467  */
468 static int __table_get_device(struct dm_table *t, struct dm_target *ti,
469 			      const char *path, sector_t start, sector_t len,
470 			      int mode, struct dm_dev **result)
471 {
472 	int r;
473 	dev_t dev;
474 	struct dm_dev *dd;
475 	unsigned int major, minor;
476 
477 	BUG_ON(!t);
478 
479 	if (sscanf(path, "%u:%u", &major, &minor) == 2) {
480 		/* Extract the major/minor numbers */
481 		dev = MKDEV(major, minor);
482 		if (MAJOR(dev) != major || MINOR(dev) != minor)
483 			return -EOVERFLOW;
484 	} else {
485 		/* convert the path to a device */
486 		if ((r = lookup_device(path, &dev)))
487 			return r;
488 	}
489 
490 	dd = find_device(&t->devices, dev);
491 	if (!dd) {
492 		dd = kmalloc(sizeof(*dd), GFP_KERNEL);
493 		if (!dd)
494 			return -ENOMEM;
495 
496 		dd->mode = mode;
497 		dd->bdev = NULL;
498 
499 		if ((r = open_dev(dd, dev, t->md))) {
500 			kfree(dd);
501 			return r;
502 		}
503 
504 		format_dev_t(dd->name, dev);
505 
506 		atomic_set(&dd->count, 0);
507 		list_add(&dd->list, &t->devices);
508 
509 	} else if (dd->mode != (mode | dd->mode)) {
510 		r = upgrade_mode(dd, mode, t->md);
511 		if (r)
512 			return r;
513 	}
514 	atomic_inc(&dd->count);
515 
516 	if (!check_device_area(dd, start, len)) {
517 		DMWARN("device %s too small for target", path);
518 		dm_put_device(ti, dd);
519 		return -EINVAL;
520 	}
521 
522 	*result = dd;
523 
524 	return 0;
525 }
526 
527 void dm_set_device_limits(struct dm_target *ti, struct block_device *bdev)
528 {
529 	struct request_queue *q = bdev_get_queue(bdev);
530 	struct io_restrictions *rs = &ti->limits;
531 
532 	/*
533 	 * Combine the device limits low.
534 	 *
535 	 * FIXME: if we move an io_restriction struct
536 	 *        into q this would just be a call to
537 	 *        combine_restrictions_low()
538 	 */
539 	rs->max_sectors =
540 		min_not_zero(rs->max_sectors, q->max_sectors);
541 
542 	/* FIXME: Device-Mapper on top of RAID-0 breaks because DM
543 	 *        currently doesn't honor MD's merge_bvec_fn routine.
544 	 *        In this case, we'll force DM to use PAGE_SIZE or
545 	 *        smaller I/O, just to be safe. A better fix is in the
546 	 *        works, but add this for the time being so it will at
547 	 *        least operate correctly.
548 	 */
549 	if (q->merge_bvec_fn)
550 		rs->max_sectors =
551 			min_not_zero(rs->max_sectors,
552 				     (unsigned int) (PAGE_SIZE >> 9));
553 
554 	rs->max_phys_segments =
555 		min_not_zero(rs->max_phys_segments,
556 			     q->max_phys_segments);
557 
558 	rs->max_hw_segments =
559 		min_not_zero(rs->max_hw_segments, q->max_hw_segments);
560 
561 	rs->hardsect_size = max(rs->hardsect_size, q->hardsect_size);
562 
563 	rs->max_segment_size =
564 		min_not_zero(rs->max_segment_size, q->max_segment_size);
565 
566 	rs->seg_boundary_mask =
567 		min_not_zero(rs->seg_boundary_mask,
568 			     q->seg_boundary_mask);
569 
570 	rs->no_cluster |= !test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
571 }
572 EXPORT_SYMBOL_GPL(dm_set_device_limits);
573 
574 int dm_get_device(struct dm_target *ti, const char *path, sector_t start,
575 		  sector_t len, int mode, struct dm_dev **result)
576 {
577 	int r = __table_get_device(ti->table, ti, path,
578 				   start, len, mode, result);
579 
580 	if (!r)
581 		dm_set_device_limits(ti, (*result)->bdev);
582 
583 	return r;
584 }
585 
586 /*
587  * Decrement a devices use count and remove it if necessary.
588  */
589 void dm_put_device(struct dm_target *ti, struct dm_dev *dd)
590 {
591 	if (atomic_dec_and_test(&dd->count)) {
592 		close_dev(dd, ti->table->md);
593 		list_del(&dd->list);
594 		kfree(dd);
595 	}
596 }
597 
598 /*
599  * Checks to see if the target joins onto the end of the table.
600  */
601 static int adjoin(struct dm_table *table, struct dm_target *ti)
602 {
603 	struct dm_target *prev;
604 
605 	if (!table->num_targets)
606 		return !ti->begin;
607 
608 	prev = &table->targets[table->num_targets - 1];
609 	return (ti->begin == (prev->begin + prev->len));
610 }
611 
612 /*
613  * Used to dynamically allocate the arg array.
614  */
615 static char **realloc_argv(unsigned *array_size, char **old_argv)
616 {
617 	char **argv;
618 	unsigned new_size;
619 
620 	new_size = *array_size ? *array_size * 2 : 64;
621 	argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
622 	if (argv) {
623 		memcpy(argv, old_argv, *array_size * sizeof(*argv));
624 		*array_size = new_size;
625 	}
626 
627 	kfree(old_argv);
628 	return argv;
629 }
630 
631 /*
632  * Destructively splits up the argument list to pass to ctr.
633  */
634 int dm_split_args(int *argc, char ***argvp, char *input)
635 {
636 	char *start, *end = input, *out, **argv = NULL;
637 	unsigned array_size = 0;
638 
639 	*argc = 0;
640 
641 	if (!input) {
642 		*argvp = NULL;
643 		return 0;
644 	}
645 
646 	argv = realloc_argv(&array_size, argv);
647 	if (!argv)
648 		return -ENOMEM;
649 
650 	while (1) {
651 		start = end;
652 
653 		/* Skip whitespace */
654 		while (*start && isspace(*start))
655 			start++;
656 
657 		if (!*start)
658 			break;	/* success, we hit the end */
659 
660 		/* 'out' is used to remove any back-quotes */
661 		end = out = start;
662 		while (*end) {
663 			/* Everything apart from '\0' can be quoted */
664 			if (*end == '\\' && *(end + 1)) {
665 				*out++ = *(end + 1);
666 				end += 2;
667 				continue;
668 			}
669 
670 			if (isspace(*end))
671 				break;	/* end of token */
672 
673 			*out++ = *end++;
674 		}
675 
676 		/* have we already filled the array ? */
677 		if ((*argc + 1) > array_size) {
678 			argv = realloc_argv(&array_size, argv);
679 			if (!argv)
680 				return -ENOMEM;
681 		}
682 
683 		/* we know this is whitespace */
684 		if (*end)
685 			end++;
686 
687 		/* terminate the string and put it in the array */
688 		*out = '\0';
689 		argv[*argc] = start;
690 		(*argc)++;
691 	}
692 
693 	*argvp = argv;
694 	return 0;
695 }
696 
697 static void check_for_valid_limits(struct io_restrictions *rs)
698 {
699 	if (!rs->max_sectors)
700 		rs->max_sectors = SAFE_MAX_SECTORS;
701 	if (!rs->max_phys_segments)
702 		rs->max_phys_segments = MAX_PHYS_SEGMENTS;
703 	if (!rs->max_hw_segments)
704 		rs->max_hw_segments = MAX_HW_SEGMENTS;
705 	if (!rs->hardsect_size)
706 		rs->hardsect_size = 1 << SECTOR_SHIFT;
707 	if (!rs->max_segment_size)
708 		rs->max_segment_size = MAX_SEGMENT_SIZE;
709 	if (!rs->seg_boundary_mask)
710 		rs->seg_boundary_mask = -1;
711 }
712 
713 int dm_table_add_target(struct dm_table *t, const char *type,
714 			sector_t start, sector_t len, char *params)
715 {
716 	int r = -EINVAL, argc;
717 	char **argv;
718 	struct dm_target *tgt;
719 
720 	if ((r = check_space(t)))
721 		return r;
722 
723 	tgt = t->targets + t->num_targets;
724 	memset(tgt, 0, sizeof(*tgt));
725 
726 	if (!len) {
727 		DMERR("%s: zero-length target", dm_device_name(t->md));
728 		return -EINVAL;
729 	}
730 
731 	tgt->type = dm_get_target_type(type);
732 	if (!tgt->type) {
733 		DMERR("%s: %s: unknown target type", dm_device_name(t->md),
734 		      type);
735 		return -EINVAL;
736 	}
737 
738 	tgt->table = t;
739 	tgt->begin = start;
740 	tgt->len = len;
741 	tgt->error = "Unknown error";
742 
743 	/*
744 	 * Does this target adjoin the previous one ?
745 	 */
746 	if (!adjoin(t, tgt)) {
747 		tgt->error = "Gap in table";
748 		r = -EINVAL;
749 		goto bad;
750 	}
751 
752 	r = dm_split_args(&argc, &argv, params);
753 	if (r) {
754 		tgt->error = "couldn't split parameters (insufficient memory)";
755 		goto bad;
756 	}
757 
758 	r = tgt->type->ctr(tgt, argc, argv);
759 	kfree(argv);
760 	if (r)
761 		goto bad;
762 
763 	t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
764 
765 	/* FIXME: the plan is to combine high here and then have
766 	 * the merge fn apply the target level restrictions. */
767 	combine_restrictions_low(&t->limits, &tgt->limits);
768 	return 0;
769 
770  bad:
771 	DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
772 	dm_put_target_type(tgt->type);
773 	return r;
774 }
775 
776 static int setup_indexes(struct dm_table *t)
777 {
778 	int i;
779 	unsigned int total = 0;
780 	sector_t *indexes;
781 
782 	/* allocate the space for *all* the indexes */
783 	for (i = t->depth - 2; i >= 0; i--) {
784 		t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
785 		total += t->counts[i];
786 	}
787 
788 	indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
789 	if (!indexes)
790 		return -ENOMEM;
791 
792 	/* set up internal nodes, bottom-up */
793 	for (i = t->depth - 2, total = 0; i >= 0; i--) {
794 		t->index[i] = indexes;
795 		indexes += (KEYS_PER_NODE * t->counts[i]);
796 		setup_btree_index(i, t);
797 	}
798 
799 	return 0;
800 }
801 
802 /*
803  * Builds the btree to index the map.
804  */
805 int dm_table_complete(struct dm_table *t)
806 {
807 	int r = 0;
808 	unsigned int leaf_nodes;
809 
810 	check_for_valid_limits(&t->limits);
811 
812 	/* how many indexes will the btree have ? */
813 	leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
814 	t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
815 
816 	/* leaf layer has already been set up */
817 	t->counts[t->depth - 1] = leaf_nodes;
818 	t->index[t->depth - 1] = t->highs;
819 
820 	if (t->depth >= 2)
821 		r = setup_indexes(t);
822 
823 	return r;
824 }
825 
826 static DEFINE_MUTEX(_event_lock);
827 void dm_table_event_callback(struct dm_table *t,
828 			     void (*fn)(void *), void *context)
829 {
830 	mutex_lock(&_event_lock);
831 	t->event_fn = fn;
832 	t->event_context = context;
833 	mutex_unlock(&_event_lock);
834 }
835 
836 void dm_table_event(struct dm_table *t)
837 {
838 	/*
839 	 * You can no longer call dm_table_event() from interrupt
840 	 * context, use a bottom half instead.
841 	 */
842 	BUG_ON(in_interrupt());
843 
844 	mutex_lock(&_event_lock);
845 	if (t->event_fn)
846 		t->event_fn(t->event_context);
847 	mutex_unlock(&_event_lock);
848 }
849 
850 sector_t dm_table_get_size(struct dm_table *t)
851 {
852 	return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
853 }
854 
855 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
856 {
857 	if (index >= t->num_targets)
858 		return NULL;
859 
860 	return t->targets + index;
861 }
862 
863 /*
864  * Search the btree for the correct target.
865  */
866 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
867 {
868 	unsigned int l, n = 0, k = 0;
869 	sector_t *node;
870 
871 	for (l = 0; l < t->depth; l++) {
872 		n = get_child(n, k);
873 		node = get_node(t, l, n);
874 
875 		for (k = 0; k < KEYS_PER_NODE; k++)
876 			if (node[k] >= sector)
877 				break;
878 	}
879 
880 	return &t->targets[(KEYS_PER_NODE * n) + k];
881 }
882 
883 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q)
884 {
885 	/*
886 	 * Make sure we obey the optimistic sub devices
887 	 * restrictions.
888 	 */
889 	blk_queue_max_sectors(q, t->limits.max_sectors);
890 	q->max_phys_segments = t->limits.max_phys_segments;
891 	q->max_hw_segments = t->limits.max_hw_segments;
892 	q->hardsect_size = t->limits.hardsect_size;
893 	q->max_segment_size = t->limits.max_segment_size;
894 	q->seg_boundary_mask = t->limits.seg_boundary_mask;
895 	if (t->limits.no_cluster)
896 		q->queue_flags &= ~(1 << QUEUE_FLAG_CLUSTER);
897 	else
898 		q->queue_flags |= (1 << QUEUE_FLAG_CLUSTER);
899 
900 }
901 
902 unsigned int dm_table_get_num_targets(struct dm_table *t)
903 {
904 	return t->num_targets;
905 }
906 
907 struct list_head *dm_table_get_devices(struct dm_table *t)
908 {
909 	return &t->devices;
910 }
911 
912 int dm_table_get_mode(struct dm_table *t)
913 {
914 	return t->mode;
915 }
916 
917 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
918 {
919 	int i = t->num_targets;
920 	struct dm_target *ti = t->targets;
921 
922 	while (i--) {
923 		if (postsuspend) {
924 			if (ti->type->postsuspend)
925 				ti->type->postsuspend(ti);
926 		} else if (ti->type->presuspend)
927 			ti->type->presuspend(ti);
928 
929 		ti++;
930 	}
931 }
932 
933 void dm_table_presuspend_targets(struct dm_table *t)
934 {
935 	if (!t)
936 		return;
937 
938 	return suspend_targets(t, 0);
939 }
940 
941 void dm_table_postsuspend_targets(struct dm_table *t)
942 {
943 	if (!t)
944 		return;
945 
946 	return suspend_targets(t, 1);
947 }
948 
949 int dm_table_resume_targets(struct dm_table *t)
950 {
951 	int i, r = 0;
952 
953 	for (i = 0; i < t->num_targets; i++) {
954 		struct dm_target *ti = t->targets + i;
955 
956 		if (!ti->type->preresume)
957 			continue;
958 
959 		r = ti->type->preresume(ti);
960 		if (r)
961 			return r;
962 	}
963 
964 	for (i = 0; i < t->num_targets; i++) {
965 		struct dm_target *ti = t->targets + i;
966 
967 		if (ti->type->resume)
968 			ti->type->resume(ti);
969 	}
970 
971 	return 0;
972 }
973 
974 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
975 {
976 	struct list_head *d, *devices;
977 	int r = 0;
978 
979 	devices = dm_table_get_devices(t);
980 	for (d = devices->next; d != devices; d = d->next) {
981 		struct dm_dev *dd = list_entry(d, struct dm_dev, list);
982 		struct request_queue *q = bdev_get_queue(dd->bdev);
983 		r |= bdi_congested(&q->backing_dev_info, bdi_bits);
984 	}
985 
986 	return r;
987 }
988 
989 void dm_table_unplug_all(struct dm_table *t)
990 {
991 	struct list_head *d, *devices = dm_table_get_devices(t);
992 
993 	for (d = devices->next; d != devices; d = d->next) {
994 		struct dm_dev *dd = list_entry(d, struct dm_dev, list);
995 		struct request_queue *q = bdev_get_queue(dd->bdev);
996 
997 		if (q->unplug_fn)
998 			q->unplug_fn(q);
999 	}
1000 }
1001 
1002 struct mapped_device *dm_table_get_md(struct dm_table *t)
1003 {
1004 	dm_get(t->md);
1005 
1006 	return t->md;
1007 }
1008 
1009 EXPORT_SYMBOL(dm_vcalloc);
1010 EXPORT_SYMBOL(dm_get_device);
1011 EXPORT_SYMBOL(dm_put_device);
1012 EXPORT_SYMBOL(dm_table_event);
1013 EXPORT_SYMBOL(dm_table_get_size);
1014 EXPORT_SYMBOL(dm_table_get_mode);
1015 EXPORT_SYMBOL(dm_table_get_md);
1016 EXPORT_SYMBOL(dm_table_put);
1017 EXPORT_SYMBOL(dm_table_get);
1018 EXPORT_SYMBOL(dm_table_unplug_all);
1019