xref: /openbmc/linux/drivers/md/dm-table.c (revision a1e58bbd)
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->max_hw_sectors =
103 		min_not_zero(lhs->max_hw_sectors, rhs->max_hw_sectors);
104 
105 	lhs->seg_boundary_mask =
106 		min_not_zero(lhs->seg_boundary_mask, rhs->seg_boundary_mask);
107 
108 	lhs->bounce_pfn = min_not_zero(lhs->bounce_pfn, rhs->bounce_pfn);
109 
110 	lhs->no_cluster |= rhs->no_cluster;
111 }
112 
113 /*
114  * Calculate the index of the child node of the n'th node k'th key.
115  */
116 static inline unsigned int get_child(unsigned int n, unsigned int k)
117 {
118 	return (n * CHILDREN_PER_NODE) + k;
119 }
120 
121 /*
122  * Return the n'th node of level l from table t.
123  */
124 static inline sector_t *get_node(struct dm_table *t,
125 				 unsigned int l, unsigned int n)
126 {
127 	return t->index[l] + (n * KEYS_PER_NODE);
128 }
129 
130 /*
131  * Return the highest key that you could lookup from the n'th
132  * node on level l of the btree.
133  */
134 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
135 {
136 	for (; l < t->depth - 1; l++)
137 		n = get_child(n, CHILDREN_PER_NODE - 1);
138 
139 	if (n >= t->counts[l])
140 		return (sector_t) - 1;
141 
142 	return get_node(t, l, n)[KEYS_PER_NODE - 1];
143 }
144 
145 /*
146  * Fills in a level of the btree based on the highs of the level
147  * below it.
148  */
149 static int setup_btree_index(unsigned int l, struct dm_table *t)
150 {
151 	unsigned int n, k;
152 	sector_t *node;
153 
154 	for (n = 0U; n < t->counts[l]; n++) {
155 		node = get_node(t, l, n);
156 
157 		for (k = 0U; k < KEYS_PER_NODE; k++)
158 			node[k] = high(t, l + 1, get_child(n, k));
159 	}
160 
161 	return 0;
162 }
163 
164 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
165 {
166 	unsigned long size;
167 	void *addr;
168 
169 	/*
170 	 * Check that we're not going to overflow.
171 	 */
172 	if (nmemb > (ULONG_MAX / elem_size))
173 		return NULL;
174 
175 	size = nmemb * elem_size;
176 	addr = vmalloc(size);
177 	if (addr)
178 		memset(addr, 0, size);
179 
180 	return addr;
181 }
182 
183 /*
184  * highs, and targets are managed as dynamic arrays during a
185  * table load.
186  */
187 static int alloc_targets(struct dm_table *t, unsigned int num)
188 {
189 	sector_t *n_highs;
190 	struct dm_target *n_targets;
191 	int n = t->num_targets;
192 
193 	/*
194 	 * Allocate both the target array and offset array at once.
195 	 * Append an empty entry to catch sectors beyond the end of
196 	 * the device.
197 	 */
198 	n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
199 					  sizeof(sector_t));
200 	if (!n_highs)
201 		return -ENOMEM;
202 
203 	n_targets = (struct dm_target *) (n_highs + num);
204 
205 	if (n) {
206 		memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
207 		memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
208 	}
209 
210 	memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
211 	vfree(t->highs);
212 
213 	t->num_allocated = num;
214 	t->highs = n_highs;
215 	t->targets = n_targets;
216 
217 	return 0;
218 }
219 
220 int dm_table_create(struct dm_table **result, int mode,
221 		    unsigned num_targets, struct mapped_device *md)
222 {
223 	struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
224 
225 	if (!t)
226 		return -ENOMEM;
227 
228 	INIT_LIST_HEAD(&t->devices);
229 	atomic_set(&t->holders, 1);
230 
231 	if (!num_targets)
232 		num_targets = KEYS_PER_NODE;
233 
234 	num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
235 
236 	if (alloc_targets(t, num_targets)) {
237 		kfree(t);
238 		t = NULL;
239 		return -ENOMEM;
240 	}
241 
242 	t->mode = mode;
243 	t->md = md;
244 	*result = t;
245 	return 0;
246 }
247 
248 int dm_create_error_table(struct dm_table **result, struct mapped_device *md)
249 {
250 	struct dm_table *t;
251 	sector_t dev_size = 1;
252 	int r;
253 
254 	/*
255 	 * Find current size of device.
256 	 * Default to 1 sector if inactive.
257 	 */
258 	t = dm_get_table(md);
259 	if (t) {
260 		dev_size = dm_table_get_size(t);
261 		dm_table_put(t);
262 	}
263 
264 	r = dm_table_create(&t, FMODE_READ, 1, md);
265 	if (r)
266 		return r;
267 
268 	r = dm_table_add_target(t, "error", 0, dev_size, NULL);
269 	if (r)
270 		goto out;
271 
272 	r = dm_table_complete(t);
273 	if (r)
274 		goto out;
275 
276 	*result = t;
277 
278 out:
279 	if (r)
280 		dm_table_put(t);
281 
282 	return r;
283 }
284 EXPORT_SYMBOL_GPL(dm_create_error_table);
285 
286 static void free_devices(struct list_head *devices)
287 {
288 	struct list_head *tmp, *next;
289 
290 	list_for_each_safe(tmp, next, devices) {
291 		struct dm_dev *dd = list_entry(tmp, struct dm_dev, list);
292 		kfree(dd);
293 	}
294 }
295 
296 static void table_destroy(struct dm_table *t)
297 {
298 	unsigned int i;
299 
300 	/* free the indexes (see dm_table_complete) */
301 	if (t->depth >= 2)
302 		vfree(t->index[t->depth - 2]);
303 
304 	/* free the targets */
305 	for (i = 0; i < t->num_targets; i++) {
306 		struct dm_target *tgt = t->targets + i;
307 
308 		if (tgt->type->dtr)
309 			tgt->type->dtr(tgt);
310 
311 		dm_put_target_type(tgt->type);
312 	}
313 
314 	vfree(t->highs);
315 
316 	/* free the device list */
317 	if (t->devices.next != &t->devices) {
318 		DMWARN("devices still present during destroy: "
319 		       "dm_table_remove_device calls missing");
320 
321 		free_devices(&t->devices);
322 	}
323 
324 	kfree(t);
325 }
326 
327 void dm_table_get(struct dm_table *t)
328 {
329 	atomic_inc(&t->holders);
330 }
331 
332 void dm_table_put(struct dm_table *t)
333 {
334 	if (!t)
335 		return;
336 
337 	if (atomic_dec_and_test(&t->holders))
338 		table_destroy(t);
339 }
340 
341 /*
342  * Checks to see if we need to extend highs or targets.
343  */
344 static inline int check_space(struct dm_table *t)
345 {
346 	if (t->num_targets >= t->num_allocated)
347 		return alloc_targets(t, t->num_allocated * 2);
348 
349 	return 0;
350 }
351 
352 /*
353  * Convert a device path to a dev_t.
354  */
355 static int lookup_device(const char *path, dev_t *dev)
356 {
357 	int r;
358 	struct nameidata nd;
359 	struct inode *inode;
360 
361 	if ((r = path_lookup(path, LOOKUP_FOLLOW, &nd)))
362 		return r;
363 
364 	inode = nd.path.dentry->d_inode;
365 	if (!inode) {
366 		r = -ENOENT;
367 		goto out;
368 	}
369 
370 	if (!S_ISBLK(inode->i_mode)) {
371 		r = -ENOTBLK;
372 		goto out;
373 	}
374 
375 	*dev = inode->i_rdev;
376 
377  out:
378 	path_put(&nd.path);
379 	return r;
380 }
381 
382 /*
383  * See if we've already got a device in the list.
384  */
385 static struct dm_dev *find_device(struct list_head *l, dev_t dev)
386 {
387 	struct dm_dev *dd;
388 
389 	list_for_each_entry (dd, l, list)
390 		if (dd->bdev->bd_dev == dev)
391 			return dd;
392 
393 	return NULL;
394 }
395 
396 /*
397  * Open a device so we can use it as a map destination.
398  */
399 static int open_dev(struct dm_dev *d, dev_t dev, struct mapped_device *md)
400 {
401 	static char *_claim_ptr = "I belong to device-mapper";
402 	struct block_device *bdev;
403 
404 	int r;
405 
406 	BUG_ON(d->bdev);
407 
408 	bdev = open_by_devnum(dev, d->mode);
409 	if (IS_ERR(bdev))
410 		return PTR_ERR(bdev);
411 	r = bd_claim_by_disk(bdev, _claim_ptr, dm_disk(md));
412 	if (r)
413 		blkdev_put(bdev);
414 	else
415 		d->bdev = bdev;
416 	return r;
417 }
418 
419 /*
420  * Close a device that we've been using.
421  */
422 static void close_dev(struct dm_dev *d, struct mapped_device *md)
423 {
424 	if (!d->bdev)
425 		return;
426 
427 	bd_release_from_disk(d->bdev, dm_disk(md));
428 	blkdev_put(d->bdev);
429 	d->bdev = NULL;
430 }
431 
432 /*
433  * If possible, this checks an area of a destination device is valid.
434  */
435 static int check_device_area(struct dm_dev *dd, sector_t start, sector_t len)
436 {
437 	sector_t dev_size = dd->bdev->bd_inode->i_size >> SECTOR_SHIFT;
438 
439 	if (!dev_size)
440 		return 1;
441 
442 	return ((start < dev_size) && (len <= (dev_size - start)));
443 }
444 
445 /*
446  * This upgrades the mode on an already open dm_dev.  Being
447  * careful to leave things as they were if we fail to reopen the
448  * device.
449  */
450 static int upgrade_mode(struct dm_dev *dd, int new_mode, struct mapped_device *md)
451 {
452 	int r;
453 	struct dm_dev dd_copy;
454 	dev_t dev = dd->bdev->bd_dev;
455 
456 	dd_copy = *dd;
457 
458 	dd->mode |= new_mode;
459 	dd->bdev = NULL;
460 	r = open_dev(dd, dev, md);
461 	if (!r)
462 		close_dev(&dd_copy, md);
463 	else
464 		*dd = dd_copy;
465 
466 	return r;
467 }
468 
469 /*
470  * Add a device to the list, or just increment the usage count if
471  * it's already present.
472  */
473 static int __table_get_device(struct dm_table *t, struct dm_target *ti,
474 			      const char *path, sector_t start, sector_t len,
475 			      int mode, struct dm_dev **result)
476 {
477 	int r;
478 	dev_t uninitialized_var(dev);
479 	struct dm_dev *dd;
480 	unsigned int major, minor;
481 
482 	BUG_ON(!t);
483 
484 	if (sscanf(path, "%u:%u", &major, &minor) == 2) {
485 		/* Extract the major/minor numbers */
486 		dev = MKDEV(major, minor);
487 		if (MAJOR(dev) != major || MINOR(dev) != minor)
488 			return -EOVERFLOW;
489 	} else {
490 		/* convert the path to a device */
491 		if ((r = lookup_device(path, &dev)))
492 			return r;
493 	}
494 
495 	dd = find_device(&t->devices, dev);
496 	if (!dd) {
497 		dd = kmalloc(sizeof(*dd), GFP_KERNEL);
498 		if (!dd)
499 			return -ENOMEM;
500 
501 		dd->mode = mode;
502 		dd->bdev = NULL;
503 
504 		if ((r = open_dev(dd, dev, t->md))) {
505 			kfree(dd);
506 			return r;
507 		}
508 
509 		format_dev_t(dd->name, dev);
510 
511 		atomic_set(&dd->count, 0);
512 		list_add(&dd->list, &t->devices);
513 
514 	} else if (dd->mode != (mode | dd->mode)) {
515 		r = upgrade_mode(dd, mode, t->md);
516 		if (r)
517 			return r;
518 	}
519 	atomic_inc(&dd->count);
520 
521 	if (!check_device_area(dd, start, len)) {
522 		DMWARN("device %s too small for target", path);
523 		dm_put_device(ti, dd);
524 		return -EINVAL;
525 	}
526 
527 	*result = dd;
528 
529 	return 0;
530 }
531 
532 void dm_set_device_limits(struct dm_target *ti, struct block_device *bdev)
533 {
534 	struct request_queue *q = bdev_get_queue(bdev);
535 	struct io_restrictions *rs = &ti->limits;
536 
537 	/*
538 	 * Combine the device limits low.
539 	 *
540 	 * FIXME: if we move an io_restriction struct
541 	 *        into q this would just be a call to
542 	 *        combine_restrictions_low()
543 	 */
544 	rs->max_sectors =
545 		min_not_zero(rs->max_sectors, q->max_sectors);
546 
547 	/* FIXME: Device-Mapper on top of RAID-0 breaks because DM
548 	 *        currently doesn't honor MD's merge_bvec_fn routine.
549 	 *        In this case, we'll force DM to use PAGE_SIZE or
550 	 *        smaller I/O, just to be safe. A better fix is in the
551 	 *        works, but add this for the time being so it will at
552 	 *        least operate correctly.
553 	 */
554 	if (q->merge_bvec_fn)
555 		rs->max_sectors =
556 			min_not_zero(rs->max_sectors,
557 				     (unsigned int) (PAGE_SIZE >> 9));
558 
559 	rs->max_phys_segments =
560 		min_not_zero(rs->max_phys_segments,
561 			     q->max_phys_segments);
562 
563 	rs->max_hw_segments =
564 		min_not_zero(rs->max_hw_segments, q->max_hw_segments);
565 
566 	rs->hardsect_size = max(rs->hardsect_size, q->hardsect_size);
567 
568 	rs->max_segment_size =
569 		min_not_zero(rs->max_segment_size, q->max_segment_size);
570 
571 	rs->max_hw_sectors =
572 		min_not_zero(rs->max_hw_sectors, q->max_hw_sectors);
573 
574 	rs->seg_boundary_mask =
575 		min_not_zero(rs->seg_boundary_mask,
576 			     q->seg_boundary_mask);
577 
578 	rs->bounce_pfn = min_not_zero(rs->bounce_pfn, q->bounce_pfn);
579 
580 	rs->no_cluster |= !test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
581 }
582 EXPORT_SYMBOL_GPL(dm_set_device_limits);
583 
584 int dm_get_device(struct dm_target *ti, const char *path, sector_t start,
585 		  sector_t len, int mode, struct dm_dev **result)
586 {
587 	int r = __table_get_device(ti->table, ti, path,
588 				   start, len, mode, result);
589 
590 	if (!r)
591 		dm_set_device_limits(ti, (*result)->bdev);
592 
593 	return r;
594 }
595 
596 /*
597  * Decrement a devices use count and remove it if necessary.
598  */
599 void dm_put_device(struct dm_target *ti, struct dm_dev *dd)
600 {
601 	if (atomic_dec_and_test(&dd->count)) {
602 		close_dev(dd, ti->table->md);
603 		list_del(&dd->list);
604 		kfree(dd);
605 	}
606 }
607 
608 /*
609  * Checks to see if the target joins onto the end of the table.
610  */
611 static int adjoin(struct dm_table *table, struct dm_target *ti)
612 {
613 	struct dm_target *prev;
614 
615 	if (!table->num_targets)
616 		return !ti->begin;
617 
618 	prev = &table->targets[table->num_targets - 1];
619 	return (ti->begin == (prev->begin + prev->len));
620 }
621 
622 /*
623  * Used to dynamically allocate the arg array.
624  */
625 static char **realloc_argv(unsigned *array_size, char **old_argv)
626 {
627 	char **argv;
628 	unsigned new_size;
629 
630 	new_size = *array_size ? *array_size * 2 : 64;
631 	argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
632 	if (argv) {
633 		memcpy(argv, old_argv, *array_size * sizeof(*argv));
634 		*array_size = new_size;
635 	}
636 
637 	kfree(old_argv);
638 	return argv;
639 }
640 
641 /*
642  * Destructively splits up the argument list to pass to ctr.
643  */
644 int dm_split_args(int *argc, char ***argvp, char *input)
645 {
646 	char *start, *end = input, *out, **argv = NULL;
647 	unsigned array_size = 0;
648 
649 	*argc = 0;
650 
651 	if (!input) {
652 		*argvp = NULL;
653 		return 0;
654 	}
655 
656 	argv = realloc_argv(&array_size, argv);
657 	if (!argv)
658 		return -ENOMEM;
659 
660 	while (1) {
661 		start = end;
662 
663 		/* Skip whitespace */
664 		while (*start && isspace(*start))
665 			start++;
666 
667 		if (!*start)
668 			break;	/* success, we hit the end */
669 
670 		/* 'out' is used to remove any back-quotes */
671 		end = out = start;
672 		while (*end) {
673 			/* Everything apart from '\0' can be quoted */
674 			if (*end == '\\' && *(end + 1)) {
675 				*out++ = *(end + 1);
676 				end += 2;
677 				continue;
678 			}
679 
680 			if (isspace(*end))
681 				break;	/* end of token */
682 
683 			*out++ = *end++;
684 		}
685 
686 		/* have we already filled the array ? */
687 		if ((*argc + 1) > array_size) {
688 			argv = realloc_argv(&array_size, argv);
689 			if (!argv)
690 				return -ENOMEM;
691 		}
692 
693 		/* we know this is whitespace */
694 		if (*end)
695 			end++;
696 
697 		/* terminate the string and put it in the array */
698 		*out = '\0';
699 		argv[*argc] = start;
700 		(*argc)++;
701 	}
702 
703 	*argvp = argv;
704 	return 0;
705 }
706 
707 static void check_for_valid_limits(struct io_restrictions *rs)
708 {
709 	if (!rs->max_sectors)
710 		rs->max_sectors = SAFE_MAX_SECTORS;
711 	if (!rs->max_hw_sectors)
712 		rs->max_hw_sectors = SAFE_MAX_SECTORS;
713 	if (!rs->max_phys_segments)
714 		rs->max_phys_segments = MAX_PHYS_SEGMENTS;
715 	if (!rs->max_hw_segments)
716 		rs->max_hw_segments = MAX_HW_SEGMENTS;
717 	if (!rs->hardsect_size)
718 		rs->hardsect_size = 1 << SECTOR_SHIFT;
719 	if (!rs->max_segment_size)
720 		rs->max_segment_size = MAX_SEGMENT_SIZE;
721 	if (!rs->seg_boundary_mask)
722 		rs->seg_boundary_mask = -1;
723 	if (!rs->bounce_pfn)
724 		rs->bounce_pfn = -1;
725 }
726 
727 int dm_table_add_target(struct dm_table *t, const char *type,
728 			sector_t start, sector_t len, char *params)
729 {
730 	int r = -EINVAL, argc;
731 	char **argv;
732 	struct dm_target *tgt;
733 
734 	if ((r = check_space(t)))
735 		return r;
736 
737 	tgt = t->targets + t->num_targets;
738 	memset(tgt, 0, sizeof(*tgt));
739 
740 	if (!len) {
741 		DMERR("%s: zero-length target", dm_device_name(t->md));
742 		return -EINVAL;
743 	}
744 
745 	tgt->type = dm_get_target_type(type);
746 	if (!tgt->type) {
747 		DMERR("%s: %s: unknown target type", dm_device_name(t->md),
748 		      type);
749 		return -EINVAL;
750 	}
751 
752 	tgt->table = t;
753 	tgt->begin = start;
754 	tgt->len = len;
755 	tgt->error = "Unknown error";
756 
757 	/*
758 	 * Does this target adjoin the previous one ?
759 	 */
760 	if (!adjoin(t, tgt)) {
761 		tgt->error = "Gap in table";
762 		r = -EINVAL;
763 		goto bad;
764 	}
765 
766 	r = dm_split_args(&argc, &argv, params);
767 	if (r) {
768 		tgt->error = "couldn't split parameters (insufficient memory)";
769 		goto bad;
770 	}
771 
772 	r = tgt->type->ctr(tgt, argc, argv);
773 	kfree(argv);
774 	if (r)
775 		goto bad;
776 
777 	t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
778 
779 	/* FIXME: the plan is to combine high here and then have
780 	 * the merge fn apply the target level restrictions. */
781 	combine_restrictions_low(&t->limits, &tgt->limits);
782 	return 0;
783 
784  bad:
785 	DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
786 	dm_put_target_type(tgt->type);
787 	return r;
788 }
789 
790 static int setup_indexes(struct dm_table *t)
791 {
792 	int i;
793 	unsigned int total = 0;
794 	sector_t *indexes;
795 
796 	/* allocate the space for *all* the indexes */
797 	for (i = t->depth - 2; i >= 0; i--) {
798 		t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
799 		total += t->counts[i];
800 	}
801 
802 	indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
803 	if (!indexes)
804 		return -ENOMEM;
805 
806 	/* set up internal nodes, bottom-up */
807 	for (i = t->depth - 2; i >= 0; i--) {
808 		t->index[i] = indexes;
809 		indexes += (KEYS_PER_NODE * t->counts[i]);
810 		setup_btree_index(i, t);
811 	}
812 
813 	return 0;
814 }
815 
816 /*
817  * Builds the btree to index the map.
818  */
819 int dm_table_complete(struct dm_table *t)
820 {
821 	int r = 0;
822 	unsigned int leaf_nodes;
823 
824 	check_for_valid_limits(&t->limits);
825 
826 	/* how many indexes will the btree have ? */
827 	leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
828 	t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
829 
830 	/* leaf layer has already been set up */
831 	t->counts[t->depth - 1] = leaf_nodes;
832 	t->index[t->depth - 1] = t->highs;
833 
834 	if (t->depth >= 2)
835 		r = setup_indexes(t);
836 
837 	return r;
838 }
839 
840 static DEFINE_MUTEX(_event_lock);
841 void dm_table_event_callback(struct dm_table *t,
842 			     void (*fn)(void *), void *context)
843 {
844 	mutex_lock(&_event_lock);
845 	t->event_fn = fn;
846 	t->event_context = context;
847 	mutex_unlock(&_event_lock);
848 }
849 
850 void dm_table_event(struct dm_table *t)
851 {
852 	/*
853 	 * You can no longer call dm_table_event() from interrupt
854 	 * context, use a bottom half instead.
855 	 */
856 	BUG_ON(in_interrupt());
857 
858 	mutex_lock(&_event_lock);
859 	if (t->event_fn)
860 		t->event_fn(t->event_context);
861 	mutex_unlock(&_event_lock);
862 }
863 
864 sector_t dm_table_get_size(struct dm_table *t)
865 {
866 	return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
867 }
868 
869 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
870 {
871 	if (index >= t->num_targets)
872 		return NULL;
873 
874 	return t->targets + index;
875 }
876 
877 /*
878  * Search the btree for the correct target.
879  *
880  * Caller should check returned pointer with dm_target_is_valid()
881  * to trap I/O beyond end of device.
882  */
883 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
884 {
885 	unsigned int l, n = 0, k = 0;
886 	sector_t *node;
887 
888 	for (l = 0; l < t->depth; l++) {
889 		n = get_child(n, k);
890 		node = get_node(t, l, n);
891 
892 		for (k = 0; k < KEYS_PER_NODE; k++)
893 			if (node[k] >= sector)
894 				break;
895 	}
896 
897 	return &t->targets[(KEYS_PER_NODE * n) + k];
898 }
899 
900 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q)
901 {
902 	/*
903 	 * Make sure we obey the optimistic sub devices
904 	 * restrictions.
905 	 */
906 	blk_queue_max_sectors(q, t->limits.max_sectors);
907 	q->max_phys_segments = t->limits.max_phys_segments;
908 	q->max_hw_segments = t->limits.max_hw_segments;
909 	q->hardsect_size = t->limits.hardsect_size;
910 	q->max_segment_size = t->limits.max_segment_size;
911 	q->max_hw_sectors = t->limits.max_hw_sectors;
912 	q->seg_boundary_mask = t->limits.seg_boundary_mask;
913 	q->bounce_pfn = t->limits.bounce_pfn;
914 	if (t->limits.no_cluster)
915 		q->queue_flags &= ~(1 << QUEUE_FLAG_CLUSTER);
916 	else
917 		q->queue_flags |= (1 << QUEUE_FLAG_CLUSTER);
918 
919 }
920 
921 unsigned int dm_table_get_num_targets(struct dm_table *t)
922 {
923 	return t->num_targets;
924 }
925 
926 struct list_head *dm_table_get_devices(struct dm_table *t)
927 {
928 	return &t->devices;
929 }
930 
931 int dm_table_get_mode(struct dm_table *t)
932 {
933 	return t->mode;
934 }
935 
936 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
937 {
938 	int i = t->num_targets;
939 	struct dm_target *ti = t->targets;
940 
941 	while (i--) {
942 		if (postsuspend) {
943 			if (ti->type->postsuspend)
944 				ti->type->postsuspend(ti);
945 		} else if (ti->type->presuspend)
946 			ti->type->presuspend(ti);
947 
948 		ti++;
949 	}
950 }
951 
952 void dm_table_presuspend_targets(struct dm_table *t)
953 {
954 	if (!t)
955 		return;
956 
957 	return suspend_targets(t, 0);
958 }
959 
960 void dm_table_postsuspend_targets(struct dm_table *t)
961 {
962 	if (!t)
963 		return;
964 
965 	return suspend_targets(t, 1);
966 }
967 
968 int dm_table_resume_targets(struct dm_table *t)
969 {
970 	int i, r = 0;
971 
972 	for (i = 0; i < t->num_targets; i++) {
973 		struct dm_target *ti = t->targets + i;
974 
975 		if (!ti->type->preresume)
976 			continue;
977 
978 		r = ti->type->preresume(ti);
979 		if (r)
980 			return r;
981 	}
982 
983 	for (i = 0; i < t->num_targets; i++) {
984 		struct dm_target *ti = t->targets + i;
985 
986 		if (ti->type->resume)
987 			ti->type->resume(ti);
988 	}
989 
990 	return 0;
991 }
992 
993 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
994 {
995 	struct dm_dev *dd;
996 	struct list_head *devices = dm_table_get_devices(t);
997 	int r = 0;
998 
999 	list_for_each_entry(dd, devices, list) {
1000 		struct request_queue *q = bdev_get_queue(dd->bdev);
1001 		r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1002 	}
1003 
1004 	return r;
1005 }
1006 
1007 void dm_table_unplug_all(struct dm_table *t)
1008 {
1009 	struct dm_dev *dd;
1010 	struct list_head *devices = dm_table_get_devices(t);
1011 
1012 	list_for_each_entry(dd, devices, list) {
1013 		struct request_queue *q = bdev_get_queue(dd->bdev);
1014 
1015 		blk_unplug(q);
1016 	}
1017 }
1018 
1019 struct mapped_device *dm_table_get_md(struct dm_table *t)
1020 {
1021 	dm_get(t->md);
1022 
1023 	return t->md;
1024 }
1025 
1026 EXPORT_SYMBOL(dm_vcalloc);
1027 EXPORT_SYMBOL(dm_get_device);
1028 EXPORT_SYMBOL(dm_put_device);
1029 EXPORT_SYMBOL(dm_table_event);
1030 EXPORT_SYMBOL(dm_table_get_size);
1031 EXPORT_SYMBOL(dm_table_get_mode);
1032 EXPORT_SYMBOL(dm_table_get_md);
1033 EXPORT_SYMBOL(dm_table_put);
1034 EXPORT_SYMBOL(dm_table_get);
1035 EXPORT_SYMBOL(dm_table_unplug_all);
1036