xref: /linux/drivers/md/dm-table.c (revision c0c9209ddd96bc4f1d70a8b9958710671e076080)
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 static void free_devices(struct list_head *devices)
249 {
250 	struct list_head *tmp, *next;
251 
252 	list_for_each_safe(tmp, next, devices) {
253 		struct dm_dev *dd = list_entry(tmp, struct dm_dev, list);
254 		kfree(dd);
255 	}
256 }
257 
258 static void table_destroy(struct dm_table *t)
259 {
260 	unsigned int i;
261 
262 	/* free the indexes (see dm_table_complete) */
263 	if (t->depth >= 2)
264 		vfree(t->index[t->depth - 2]);
265 
266 	/* free the targets */
267 	for (i = 0; i < t->num_targets; i++) {
268 		struct dm_target *tgt = t->targets + i;
269 
270 		if (tgt->type->dtr)
271 			tgt->type->dtr(tgt);
272 
273 		dm_put_target_type(tgt->type);
274 	}
275 
276 	vfree(t->highs);
277 
278 	/* free the device list */
279 	if (t->devices.next != &t->devices) {
280 		DMWARN("devices still present during destroy: "
281 		       "dm_table_remove_device calls missing");
282 
283 		free_devices(&t->devices);
284 	}
285 
286 	kfree(t);
287 }
288 
289 void dm_table_get(struct dm_table *t)
290 {
291 	atomic_inc(&t->holders);
292 }
293 
294 void dm_table_put(struct dm_table *t)
295 {
296 	if (!t)
297 		return;
298 
299 	if (atomic_dec_and_test(&t->holders))
300 		table_destroy(t);
301 }
302 
303 /*
304  * Checks to see if we need to extend highs or targets.
305  */
306 static inline int check_space(struct dm_table *t)
307 {
308 	if (t->num_targets >= t->num_allocated)
309 		return alloc_targets(t, t->num_allocated * 2);
310 
311 	return 0;
312 }
313 
314 /*
315  * Convert a device path to a dev_t.
316  */
317 static int lookup_device(const char *path, dev_t *dev)
318 {
319 	struct block_device *bdev = lookup_bdev(path);
320 	if (IS_ERR(bdev))
321 		return PTR_ERR(bdev);
322 	*dev = bdev->bd_dev;
323 	bdput(bdev);
324 	return 0;
325 }
326 
327 /*
328  * See if we've already got a device in the list.
329  */
330 static struct dm_dev *find_device(struct list_head *l, dev_t dev)
331 {
332 	struct dm_dev *dd;
333 
334 	list_for_each_entry (dd, l, list)
335 		if (dd->bdev->bd_dev == dev)
336 			return dd;
337 
338 	return NULL;
339 }
340 
341 /*
342  * Open a device so we can use it as a map destination.
343  */
344 static int open_dev(struct dm_dev *d, dev_t dev, struct mapped_device *md)
345 {
346 	static char *_claim_ptr = "I belong to device-mapper";
347 	struct block_device *bdev;
348 
349 	int r;
350 
351 	BUG_ON(d->bdev);
352 
353 	bdev = open_by_devnum(dev, d->mode);
354 	if (IS_ERR(bdev))
355 		return PTR_ERR(bdev);
356 	r = bd_claim_by_disk(bdev, _claim_ptr, dm_disk(md));
357 	if (r)
358 		blkdev_put(bdev);
359 	else
360 		d->bdev = bdev;
361 	return r;
362 }
363 
364 /*
365  * Close a device that we've been using.
366  */
367 static void close_dev(struct dm_dev *d, struct mapped_device *md)
368 {
369 	if (!d->bdev)
370 		return;
371 
372 	bd_release_from_disk(d->bdev, dm_disk(md));
373 	blkdev_put(d->bdev);
374 	d->bdev = NULL;
375 }
376 
377 /*
378  * If possible, this checks an area of a destination device is valid.
379  */
380 static int check_device_area(struct dm_dev *dd, sector_t start, sector_t len)
381 {
382 	sector_t dev_size = dd->bdev->bd_inode->i_size >> SECTOR_SHIFT;
383 
384 	if (!dev_size)
385 		return 1;
386 
387 	return ((start < dev_size) && (len <= (dev_size - start)));
388 }
389 
390 /*
391  * This upgrades the mode on an already open dm_dev.  Being
392  * careful to leave things as they were if we fail to reopen the
393  * device.
394  */
395 static int upgrade_mode(struct dm_dev *dd, int new_mode, struct mapped_device *md)
396 {
397 	int r;
398 	struct dm_dev dd_copy;
399 	dev_t dev = dd->bdev->bd_dev;
400 
401 	dd_copy = *dd;
402 
403 	dd->mode |= new_mode;
404 	dd->bdev = NULL;
405 	r = open_dev(dd, dev, md);
406 	if (!r)
407 		close_dev(&dd_copy, md);
408 	else
409 		*dd = dd_copy;
410 
411 	return r;
412 }
413 
414 /*
415  * Add a device to the list, or just increment the usage count if
416  * it's already present.
417  */
418 static int __table_get_device(struct dm_table *t, struct dm_target *ti,
419 			      const char *path, sector_t start, sector_t len,
420 			      int mode, struct dm_dev **result)
421 {
422 	int r;
423 	dev_t uninitialized_var(dev);
424 	struct dm_dev *dd;
425 	unsigned int major, minor;
426 
427 	BUG_ON(!t);
428 
429 	if (sscanf(path, "%u:%u", &major, &minor) == 2) {
430 		/* Extract the major/minor numbers */
431 		dev = MKDEV(major, minor);
432 		if (MAJOR(dev) != major || MINOR(dev) != minor)
433 			return -EOVERFLOW;
434 	} else {
435 		/* convert the path to a device */
436 		if ((r = lookup_device(path, &dev)))
437 			return r;
438 	}
439 
440 	dd = find_device(&t->devices, dev);
441 	if (!dd) {
442 		dd = kmalloc(sizeof(*dd), GFP_KERNEL);
443 		if (!dd)
444 			return -ENOMEM;
445 
446 		dd->mode = mode;
447 		dd->bdev = NULL;
448 
449 		if ((r = open_dev(dd, dev, t->md))) {
450 			kfree(dd);
451 			return r;
452 		}
453 
454 		format_dev_t(dd->name, dev);
455 
456 		atomic_set(&dd->count, 0);
457 		list_add(&dd->list, &t->devices);
458 
459 	} else if (dd->mode != (mode | dd->mode)) {
460 		r = upgrade_mode(dd, mode, t->md);
461 		if (r)
462 			return r;
463 	}
464 	atomic_inc(&dd->count);
465 
466 	if (!check_device_area(dd, start, len)) {
467 		DMWARN("device %s too small for target", path);
468 		dm_put_device(ti, dd);
469 		return -EINVAL;
470 	}
471 
472 	*result = dd;
473 
474 	return 0;
475 }
476 
477 void dm_set_device_limits(struct dm_target *ti, struct block_device *bdev)
478 {
479 	struct request_queue *q = bdev_get_queue(bdev);
480 	struct io_restrictions *rs = &ti->limits;
481 
482 	/*
483 	 * Combine the device limits low.
484 	 *
485 	 * FIXME: if we move an io_restriction struct
486 	 *        into q this would just be a call to
487 	 *        combine_restrictions_low()
488 	 */
489 	rs->max_sectors =
490 		min_not_zero(rs->max_sectors, q->max_sectors);
491 
492 	/*
493 	 * Check if merge fn is supported.
494 	 * If not we'll force DM to use PAGE_SIZE or
495 	 * smaller I/O, just to be safe.
496 	 */
497 
498 	if (q->merge_bvec_fn && !ti->type->merge)
499 		rs->max_sectors =
500 			min_not_zero(rs->max_sectors,
501 				     (unsigned int) (PAGE_SIZE >> 9));
502 
503 	rs->max_phys_segments =
504 		min_not_zero(rs->max_phys_segments,
505 			     q->max_phys_segments);
506 
507 	rs->max_hw_segments =
508 		min_not_zero(rs->max_hw_segments, q->max_hw_segments);
509 
510 	rs->hardsect_size = max(rs->hardsect_size, q->hardsect_size);
511 
512 	rs->max_segment_size =
513 		min_not_zero(rs->max_segment_size, q->max_segment_size);
514 
515 	rs->max_hw_sectors =
516 		min_not_zero(rs->max_hw_sectors, q->max_hw_sectors);
517 
518 	rs->seg_boundary_mask =
519 		min_not_zero(rs->seg_boundary_mask,
520 			     q->seg_boundary_mask);
521 
522 	rs->bounce_pfn = min_not_zero(rs->bounce_pfn, q->bounce_pfn);
523 
524 	rs->no_cluster |= !test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
525 }
526 EXPORT_SYMBOL_GPL(dm_set_device_limits);
527 
528 int dm_get_device(struct dm_target *ti, const char *path, sector_t start,
529 		  sector_t len, int mode, struct dm_dev **result)
530 {
531 	int r = __table_get_device(ti->table, ti, path,
532 				   start, len, mode, result);
533 
534 	if (!r)
535 		dm_set_device_limits(ti, (*result)->bdev);
536 
537 	return r;
538 }
539 
540 /*
541  * Decrement a devices use count and remove it if necessary.
542  */
543 void dm_put_device(struct dm_target *ti, struct dm_dev *dd)
544 {
545 	if (atomic_dec_and_test(&dd->count)) {
546 		close_dev(dd, ti->table->md);
547 		list_del(&dd->list);
548 		kfree(dd);
549 	}
550 }
551 
552 /*
553  * Checks to see if the target joins onto the end of the table.
554  */
555 static int adjoin(struct dm_table *table, struct dm_target *ti)
556 {
557 	struct dm_target *prev;
558 
559 	if (!table->num_targets)
560 		return !ti->begin;
561 
562 	prev = &table->targets[table->num_targets - 1];
563 	return (ti->begin == (prev->begin + prev->len));
564 }
565 
566 /*
567  * Used to dynamically allocate the arg array.
568  */
569 static char **realloc_argv(unsigned *array_size, char **old_argv)
570 {
571 	char **argv;
572 	unsigned new_size;
573 
574 	new_size = *array_size ? *array_size * 2 : 64;
575 	argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
576 	if (argv) {
577 		memcpy(argv, old_argv, *array_size * sizeof(*argv));
578 		*array_size = new_size;
579 	}
580 
581 	kfree(old_argv);
582 	return argv;
583 }
584 
585 /*
586  * Destructively splits up the argument list to pass to ctr.
587  */
588 int dm_split_args(int *argc, char ***argvp, char *input)
589 {
590 	char *start, *end = input, *out, **argv = NULL;
591 	unsigned array_size = 0;
592 
593 	*argc = 0;
594 
595 	if (!input) {
596 		*argvp = NULL;
597 		return 0;
598 	}
599 
600 	argv = realloc_argv(&array_size, argv);
601 	if (!argv)
602 		return -ENOMEM;
603 
604 	while (1) {
605 		start = end;
606 
607 		/* Skip whitespace */
608 		while (*start && isspace(*start))
609 			start++;
610 
611 		if (!*start)
612 			break;	/* success, we hit the end */
613 
614 		/* 'out' is used to remove any back-quotes */
615 		end = out = start;
616 		while (*end) {
617 			/* Everything apart from '\0' can be quoted */
618 			if (*end == '\\' && *(end + 1)) {
619 				*out++ = *(end + 1);
620 				end += 2;
621 				continue;
622 			}
623 
624 			if (isspace(*end))
625 				break;	/* end of token */
626 
627 			*out++ = *end++;
628 		}
629 
630 		/* have we already filled the array ? */
631 		if ((*argc + 1) > array_size) {
632 			argv = realloc_argv(&array_size, argv);
633 			if (!argv)
634 				return -ENOMEM;
635 		}
636 
637 		/* we know this is whitespace */
638 		if (*end)
639 			end++;
640 
641 		/* terminate the string and put it in the array */
642 		*out = '\0';
643 		argv[*argc] = start;
644 		(*argc)++;
645 	}
646 
647 	*argvp = argv;
648 	return 0;
649 }
650 
651 static void check_for_valid_limits(struct io_restrictions *rs)
652 {
653 	if (!rs->max_sectors)
654 		rs->max_sectors = SAFE_MAX_SECTORS;
655 	if (!rs->max_hw_sectors)
656 		rs->max_hw_sectors = SAFE_MAX_SECTORS;
657 	if (!rs->max_phys_segments)
658 		rs->max_phys_segments = MAX_PHYS_SEGMENTS;
659 	if (!rs->max_hw_segments)
660 		rs->max_hw_segments = MAX_HW_SEGMENTS;
661 	if (!rs->hardsect_size)
662 		rs->hardsect_size = 1 << SECTOR_SHIFT;
663 	if (!rs->max_segment_size)
664 		rs->max_segment_size = MAX_SEGMENT_SIZE;
665 	if (!rs->seg_boundary_mask)
666 		rs->seg_boundary_mask = -1;
667 	if (!rs->bounce_pfn)
668 		rs->bounce_pfn = -1;
669 }
670 
671 int dm_table_add_target(struct dm_table *t, const char *type,
672 			sector_t start, sector_t len, char *params)
673 {
674 	int r = -EINVAL, argc;
675 	char **argv;
676 	struct dm_target *tgt;
677 
678 	if ((r = check_space(t)))
679 		return r;
680 
681 	tgt = t->targets + t->num_targets;
682 	memset(tgt, 0, sizeof(*tgt));
683 
684 	if (!len) {
685 		DMERR("%s: zero-length target", dm_device_name(t->md));
686 		return -EINVAL;
687 	}
688 
689 	tgt->type = dm_get_target_type(type);
690 	if (!tgt->type) {
691 		DMERR("%s: %s: unknown target type", dm_device_name(t->md),
692 		      type);
693 		return -EINVAL;
694 	}
695 
696 	tgt->table = t;
697 	tgt->begin = start;
698 	tgt->len = len;
699 	tgt->error = "Unknown error";
700 
701 	/*
702 	 * Does this target adjoin the previous one ?
703 	 */
704 	if (!adjoin(t, tgt)) {
705 		tgt->error = "Gap in table";
706 		r = -EINVAL;
707 		goto bad;
708 	}
709 
710 	r = dm_split_args(&argc, &argv, params);
711 	if (r) {
712 		tgt->error = "couldn't split parameters (insufficient memory)";
713 		goto bad;
714 	}
715 
716 	r = tgt->type->ctr(tgt, argc, argv);
717 	kfree(argv);
718 	if (r)
719 		goto bad;
720 
721 	t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
722 
723 	/* FIXME: the plan is to combine high here and then have
724 	 * the merge fn apply the target level restrictions. */
725 	combine_restrictions_low(&t->limits, &tgt->limits);
726 	return 0;
727 
728  bad:
729 	DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
730 	dm_put_target_type(tgt->type);
731 	return r;
732 }
733 
734 static int setup_indexes(struct dm_table *t)
735 {
736 	int i;
737 	unsigned int total = 0;
738 	sector_t *indexes;
739 
740 	/* allocate the space for *all* the indexes */
741 	for (i = t->depth - 2; i >= 0; i--) {
742 		t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
743 		total += t->counts[i];
744 	}
745 
746 	indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
747 	if (!indexes)
748 		return -ENOMEM;
749 
750 	/* set up internal nodes, bottom-up */
751 	for (i = t->depth - 2; i >= 0; i--) {
752 		t->index[i] = indexes;
753 		indexes += (KEYS_PER_NODE * t->counts[i]);
754 		setup_btree_index(i, t);
755 	}
756 
757 	return 0;
758 }
759 
760 /*
761  * Builds the btree to index the map.
762  */
763 int dm_table_complete(struct dm_table *t)
764 {
765 	int r = 0;
766 	unsigned int leaf_nodes;
767 
768 	check_for_valid_limits(&t->limits);
769 
770 	/* how many indexes will the btree have ? */
771 	leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
772 	t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
773 
774 	/* leaf layer has already been set up */
775 	t->counts[t->depth - 1] = leaf_nodes;
776 	t->index[t->depth - 1] = t->highs;
777 
778 	if (t->depth >= 2)
779 		r = setup_indexes(t);
780 
781 	return r;
782 }
783 
784 static DEFINE_MUTEX(_event_lock);
785 void dm_table_event_callback(struct dm_table *t,
786 			     void (*fn)(void *), void *context)
787 {
788 	mutex_lock(&_event_lock);
789 	t->event_fn = fn;
790 	t->event_context = context;
791 	mutex_unlock(&_event_lock);
792 }
793 
794 void dm_table_event(struct dm_table *t)
795 {
796 	/*
797 	 * You can no longer call dm_table_event() from interrupt
798 	 * context, use a bottom half instead.
799 	 */
800 	BUG_ON(in_interrupt());
801 
802 	mutex_lock(&_event_lock);
803 	if (t->event_fn)
804 		t->event_fn(t->event_context);
805 	mutex_unlock(&_event_lock);
806 }
807 
808 sector_t dm_table_get_size(struct dm_table *t)
809 {
810 	return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
811 }
812 
813 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
814 {
815 	if (index >= t->num_targets)
816 		return NULL;
817 
818 	return t->targets + index;
819 }
820 
821 /*
822  * Search the btree for the correct target.
823  *
824  * Caller should check returned pointer with dm_target_is_valid()
825  * to trap I/O beyond end of device.
826  */
827 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
828 {
829 	unsigned int l, n = 0, k = 0;
830 	sector_t *node;
831 
832 	for (l = 0; l < t->depth; l++) {
833 		n = get_child(n, k);
834 		node = get_node(t, l, n);
835 
836 		for (k = 0; k < KEYS_PER_NODE; k++)
837 			if (node[k] >= sector)
838 				break;
839 	}
840 
841 	return &t->targets[(KEYS_PER_NODE * n) + k];
842 }
843 
844 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q)
845 {
846 	/*
847 	 * Make sure we obey the optimistic sub devices
848 	 * restrictions.
849 	 */
850 	blk_queue_max_sectors(q, t->limits.max_sectors);
851 	q->max_phys_segments = t->limits.max_phys_segments;
852 	q->max_hw_segments = t->limits.max_hw_segments;
853 	q->hardsect_size = t->limits.hardsect_size;
854 	q->max_segment_size = t->limits.max_segment_size;
855 	q->max_hw_sectors = t->limits.max_hw_sectors;
856 	q->seg_boundary_mask = t->limits.seg_boundary_mask;
857 	q->bounce_pfn = t->limits.bounce_pfn;
858 
859 	if (t->limits.no_cluster)
860 		queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q);
861 	else
862 		queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, q);
863 
864 }
865 
866 unsigned int dm_table_get_num_targets(struct dm_table *t)
867 {
868 	return t->num_targets;
869 }
870 
871 struct list_head *dm_table_get_devices(struct dm_table *t)
872 {
873 	return &t->devices;
874 }
875 
876 int dm_table_get_mode(struct dm_table *t)
877 {
878 	return t->mode;
879 }
880 
881 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
882 {
883 	int i = t->num_targets;
884 	struct dm_target *ti = t->targets;
885 
886 	while (i--) {
887 		if (postsuspend) {
888 			if (ti->type->postsuspend)
889 				ti->type->postsuspend(ti);
890 		} else if (ti->type->presuspend)
891 			ti->type->presuspend(ti);
892 
893 		ti++;
894 	}
895 }
896 
897 void dm_table_presuspend_targets(struct dm_table *t)
898 {
899 	if (!t)
900 		return;
901 
902 	suspend_targets(t, 0);
903 }
904 
905 void dm_table_postsuspend_targets(struct dm_table *t)
906 {
907 	if (!t)
908 		return;
909 
910 	suspend_targets(t, 1);
911 }
912 
913 int dm_table_resume_targets(struct dm_table *t)
914 {
915 	int i, r = 0;
916 
917 	for (i = 0; i < t->num_targets; i++) {
918 		struct dm_target *ti = t->targets + i;
919 
920 		if (!ti->type->preresume)
921 			continue;
922 
923 		r = ti->type->preresume(ti);
924 		if (r)
925 			return r;
926 	}
927 
928 	for (i = 0; i < t->num_targets; i++) {
929 		struct dm_target *ti = t->targets + i;
930 
931 		if (ti->type->resume)
932 			ti->type->resume(ti);
933 	}
934 
935 	return 0;
936 }
937 
938 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
939 {
940 	struct dm_dev *dd;
941 	struct list_head *devices = dm_table_get_devices(t);
942 	int r = 0;
943 
944 	list_for_each_entry(dd, devices, list) {
945 		struct request_queue *q = bdev_get_queue(dd->bdev);
946 		r |= bdi_congested(&q->backing_dev_info, bdi_bits);
947 	}
948 
949 	return r;
950 }
951 
952 void dm_table_unplug_all(struct dm_table *t)
953 {
954 	struct dm_dev *dd;
955 	struct list_head *devices = dm_table_get_devices(t);
956 
957 	list_for_each_entry(dd, devices, list) {
958 		struct request_queue *q = bdev_get_queue(dd->bdev);
959 
960 		blk_unplug(q);
961 	}
962 }
963 
964 struct mapped_device *dm_table_get_md(struct dm_table *t)
965 {
966 	dm_get(t->md);
967 
968 	return t->md;
969 }
970 
971 EXPORT_SYMBOL(dm_vcalloc);
972 EXPORT_SYMBOL(dm_get_device);
973 EXPORT_SYMBOL(dm_put_device);
974 EXPORT_SYMBOL(dm_table_event);
975 EXPORT_SYMBOL(dm_table_get_size);
976 EXPORT_SYMBOL(dm_table_get_mode);
977 EXPORT_SYMBOL(dm_table_get_md);
978 EXPORT_SYMBOL(dm_table_put);
979 EXPORT_SYMBOL(dm_table_get);
980 EXPORT_SYMBOL(dm_table_unplug_all);
981