xref: /openbmc/linux/drivers/md/dm-table.c (revision 93dc544c)
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 	int r;
320 	struct nameidata nd;
321 	struct inode *inode;
322 
323 	if ((r = path_lookup(path, LOOKUP_FOLLOW, &nd)))
324 		return r;
325 
326 	inode = nd.path.dentry->d_inode;
327 	if (!inode) {
328 		r = -ENOENT;
329 		goto out;
330 	}
331 
332 	if (!S_ISBLK(inode->i_mode)) {
333 		r = -ENOTBLK;
334 		goto out;
335 	}
336 
337 	*dev = inode->i_rdev;
338 
339  out:
340 	path_put(&nd.path);
341 	return r;
342 }
343 
344 /*
345  * See if we've already got a device in the list.
346  */
347 static struct dm_dev *find_device(struct list_head *l, dev_t dev)
348 {
349 	struct dm_dev *dd;
350 
351 	list_for_each_entry (dd, l, list)
352 		if (dd->bdev->bd_dev == dev)
353 			return dd;
354 
355 	return NULL;
356 }
357 
358 /*
359  * Open a device so we can use it as a map destination.
360  */
361 static int open_dev(struct dm_dev *d, dev_t dev, struct mapped_device *md)
362 {
363 	static char *_claim_ptr = "I belong to device-mapper";
364 	struct block_device *bdev;
365 
366 	int r;
367 
368 	BUG_ON(d->bdev);
369 
370 	bdev = open_by_devnum(dev, d->mode);
371 	if (IS_ERR(bdev))
372 		return PTR_ERR(bdev);
373 	r = bd_claim_by_disk(bdev, _claim_ptr, dm_disk(md));
374 	if (r)
375 		blkdev_put(bdev);
376 	else
377 		d->bdev = bdev;
378 	return r;
379 }
380 
381 /*
382  * Close a device that we've been using.
383  */
384 static void close_dev(struct dm_dev *d, struct mapped_device *md)
385 {
386 	if (!d->bdev)
387 		return;
388 
389 	bd_release_from_disk(d->bdev, dm_disk(md));
390 	blkdev_put(d->bdev);
391 	d->bdev = NULL;
392 }
393 
394 /*
395  * If possible, this checks an area of a destination device is valid.
396  */
397 static int check_device_area(struct dm_dev *dd, sector_t start, sector_t len)
398 {
399 	sector_t dev_size = dd->bdev->bd_inode->i_size >> SECTOR_SHIFT;
400 
401 	if (!dev_size)
402 		return 1;
403 
404 	return ((start < dev_size) && (len <= (dev_size - start)));
405 }
406 
407 /*
408  * This upgrades the mode on an already open dm_dev.  Being
409  * careful to leave things as they were if we fail to reopen the
410  * device.
411  */
412 static int upgrade_mode(struct dm_dev *dd, int new_mode, struct mapped_device *md)
413 {
414 	int r;
415 	struct dm_dev dd_copy;
416 	dev_t dev = dd->bdev->bd_dev;
417 
418 	dd_copy = *dd;
419 
420 	dd->mode |= new_mode;
421 	dd->bdev = NULL;
422 	r = open_dev(dd, dev, md);
423 	if (!r)
424 		close_dev(&dd_copy, md);
425 	else
426 		*dd = dd_copy;
427 
428 	return r;
429 }
430 
431 /*
432  * Add a device to the list, or just increment the usage count if
433  * it's already present.
434  */
435 static int __table_get_device(struct dm_table *t, struct dm_target *ti,
436 			      const char *path, sector_t start, sector_t len,
437 			      int mode, struct dm_dev **result)
438 {
439 	int r;
440 	dev_t uninitialized_var(dev);
441 	struct dm_dev *dd;
442 	unsigned int major, minor;
443 
444 	BUG_ON(!t);
445 
446 	if (sscanf(path, "%u:%u", &major, &minor) == 2) {
447 		/* Extract the major/minor numbers */
448 		dev = MKDEV(major, minor);
449 		if (MAJOR(dev) != major || MINOR(dev) != minor)
450 			return -EOVERFLOW;
451 	} else {
452 		/* convert the path to a device */
453 		if ((r = lookup_device(path, &dev)))
454 			return r;
455 	}
456 
457 	dd = find_device(&t->devices, dev);
458 	if (!dd) {
459 		dd = kmalloc(sizeof(*dd), GFP_KERNEL);
460 		if (!dd)
461 			return -ENOMEM;
462 
463 		dd->mode = mode;
464 		dd->bdev = NULL;
465 
466 		if ((r = open_dev(dd, dev, t->md))) {
467 			kfree(dd);
468 			return r;
469 		}
470 
471 		format_dev_t(dd->name, dev);
472 
473 		atomic_set(&dd->count, 0);
474 		list_add(&dd->list, &t->devices);
475 
476 	} else if (dd->mode != (mode | dd->mode)) {
477 		r = upgrade_mode(dd, mode, t->md);
478 		if (r)
479 			return r;
480 	}
481 	atomic_inc(&dd->count);
482 
483 	if (!check_device_area(dd, start, len)) {
484 		DMWARN("device %s too small for target", path);
485 		dm_put_device(ti, dd);
486 		return -EINVAL;
487 	}
488 
489 	*result = dd;
490 
491 	return 0;
492 }
493 
494 void dm_set_device_limits(struct dm_target *ti, struct block_device *bdev)
495 {
496 	struct request_queue *q = bdev_get_queue(bdev);
497 	struct io_restrictions *rs = &ti->limits;
498 
499 	/*
500 	 * Combine the device limits low.
501 	 *
502 	 * FIXME: if we move an io_restriction struct
503 	 *        into q this would just be a call to
504 	 *        combine_restrictions_low()
505 	 */
506 	rs->max_sectors =
507 		min_not_zero(rs->max_sectors, q->max_sectors);
508 
509 	/*
510 	 * Check if merge fn is supported.
511 	 * If not we'll force DM to use PAGE_SIZE or
512 	 * smaller I/O, just to be safe.
513 	 */
514 
515 	if (q->merge_bvec_fn && !ti->type->merge)
516 		rs->max_sectors =
517 			min_not_zero(rs->max_sectors,
518 				     (unsigned int) (PAGE_SIZE >> 9));
519 
520 	rs->max_phys_segments =
521 		min_not_zero(rs->max_phys_segments,
522 			     q->max_phys_segments);
523 
524 	rs->max_hw_segments =
525 		min_not_zero(rs->max_hw_segments, q->max_hw_segments);
526 
527 	rs->hardsect_size = max(rs->hardsect_size, q->hardsect_size);
528 
529 	rs->max_segment_size =
530 		min_not_zero(rs->max_segment_size, q->max_segment_size);
531 
532 	rs->max_hw_sectors =
533 		min_not_zero(rs->max_hw_sectors, q->max_hw_sectors);
534 
535 	rs->seg_boundary_mask =
536 		min_not_zero(rs->seg_boundary_mask,
537 			     q->seg_boundary_mask);
538 
539 	rs->bounce_pfn = min_not_zero(rs->bounce_pfn, q->bounce_pfn);
540 
541 	rs->no_cluster |= !test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
542 }
543 EXPORT_SYMBOL_GPL(dm_set_device_limits);
544 
545 int dm_get_device(struct dm_target *ti, const char *path, sector_t start,
546 		  sector_t len, int mode, struct dm_dev **result)
547 {
548 	int r = __table_get_device(ti->table, ti, path,
549 				   start, len, mode, result);
550 
551 	if (!r)
552 		dm_set_device_limits(ti, (*result)->bdev);
553 
554 	return r;
555 }
556 
557 /*
558  * Decrement a devices use count and remove it if necessary.
559  */
560 void dm_put_device(struct dm_target *ti, struct dm_dev *dd)
561 {
562 	if (atomic_dec_and_test(&dd->count)) {
563 		close_dev(dd, ti->table->md);
564 		list_del(&dd->list);
565 		kfree(dd);
566 	}
567 }
568 
569 /*
570  * Checks to see if the target joins onto the end of the table.
571  */
572 static int adjoin(struct dm_table *table, struct dm_target *ti)
573 {
574 	struct dm_target *prev;
575 
576 	if (!table->num_targets)
577 		return !ti->begin;
578 
579 	prev = &table->targets[table->num_targets - 1];
580 	return (ti->begin == (prev->begin + prev->len));
581 }
582 
583 /*
584  * Used to dynamically allocate the arg array.
585  */
586 static char **realloc_argv(unsigned *array_size, char **old_argv)
587 {
588 	char **argv;
589 	unsigned new_size;
590 
591 	new_size = *array_size ? *array_size * 2 : 64;
592 	argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
593 	if (argv) {
594 		memcpy(argv, old_argv, *array_size * sizeof(*argv));
595 		*array_size = new_size;
596 	}
597 
598 	kfree(old_argv);
599 	return argv;
600 }
601 
602 /*
603  * Destructively splits up the argument list to pass to ctr.
604  */
605 int dm_split_args(int *argc, char ***argvp, char *input)
606 {
607 	char *start, *end = input, *out, **argv = NULL;
608 	unsigned array_size = 0;
609 
610 	*argc = 0;
611 
612 	if (!input) {
613 		*argvp = NULL;
614 		return 0;
615 	}
616 
617 	argv = realloc_argv(&array_size, argv);
618 	if (!argv)
619 		return -ENOMEM;
620 
621 	while (1) {
622 		start = end;
623 
624 		/* Skip whitespace */
625 		while (*start && isspace(*start))
626 			start++;
627 
628 		if (!*start)
629 			break;	/* success, we hit the end */
630 
631 		/* 'out' is used to remove any back-quotes */
632 		end = out = start;
633 		while (*end) {
634 			/* Everything apart from '\0' can be quoted */
635 			if (*end == '\\' && *(end + 1)) {
636 				*out++ = *(end + 1);
637 				end += 2;
638 				continue;
639 			}
640 
641 			if (isspace(*end))
642 				break;	/* end of token */
643 
644 			*out++ = *end++;
645 		}
646 
647 		/* have we already filled the array ? */
648 		if ((*argc + 1) > array_size) {
649 			argv = realloc_argv(&array_size, argv);
650 			if (!argv)
651 				return -ENOMEM;
652 		}
653 
654 		/* we know this is whitespace */
655 		if (*end)
656 			end++;
657 
658 		/* terminate the string and put it in the array */
659 		*out = '\0';
660 		argv[*argc] = start;
661 		(*argc)++;
662 	}
663 
664 	*argvp = argv;
665 	return 0;
666 }
667 
668 static void check_for_valid_limits(struct io_restrictions *rs)
669 {
670 	if (!rs->max_sectors)
671 		rs->max_sectors = SAFE_MAX_SECTORS;
672 	if (!rs->max_hw_sectors)
673 		rs->max_hw_sectors = SAFE_MAX_SECTORS;
674 	if (!rs->max_phys_segments)
675 		rs->max_phys_segments = MAX_PHYS_SEGMENTS;
676 	if (!rs->max_hw_segments)
677 		rs->max_hw_segments = MAX_HW_SEGMENTS;
678 	if (!rs->hardsect_size)
679 		rs->hardsect_size = 1 << SECTOR_SHIFT;
680 	if (!rs->max_segment_size)
681 		rs->max_segment_size = MAX_SEGMENT_SIZE;
682 	if (!rs->seg_boundary_mask)
683 		rs->seg_boundary_mask = -1;
684 	if (!rs->bounce_pfn)
685 		rs->bounce_pfn = -1;
686 }
687 
688 int dm_table_add_target(struct dm_table *t, const char *type,
689 			sector_t start, sector_t len, char *params)
690 {
691 	int r = -EINVAL, argc;
692 	char **argv;
693 	struct dm_target *tgt;
694 
695 	if ((r = check_space(t)))
696 		return r;
697 
698 	tgt = t->targets + t->num_targets;
699 	memset(tgt, 0, sizeof(*tgt));
700 
701 	if (!len) {
702 		DMERR("%s: zero-length target", dm_device_name(t->md));
703 		return -EINVAL;
704 	}
705 
706 	tgt->type = dm_get_target_type(type);
707 	if (!tgt->type) {
708 		DMERR("%s: %s: unknown target type", dm_device_name(t->md),
709 		      type);
710 		return -EINVAL;
711 	}
712 
713 	tgt->table = t;
714 	tgt->begin = start;
715 	tgt->len = len;
716 	tgt->error = "Unknown error";
717 
718 	/*
719 	 * Does this target adjoin the previous one ?
720 	 */
721 	if (!adjoin(t, tgt)) {
722 		tgt->error = "Gap in table";
723 		r = -EINVAL;
724 		goto bad;
725 	}
726 
727 	r = dm_split_args(&argc, &argv, params);
728 	if (r) {
729 		tgt->error = "couldn't split parameters (insufficient memory)";
730 		goto bad;
731 	}
732 
733 	r = tgt->type->ctr(tgt, argc, argv);
734 	kfree(argv);
735 	if (r)
736 		goto bad;
737 
738 	t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
739 
740 	/* FIXME: the plan is to combine high here and then have
741 	 * the merge fn apply the target level restrictions. */
742 	combine_restrictions_low(&t->limits, &tgt->limits);
743 	return 0;
744 
745  bad:
746 	DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
747 	dm_put_target_type(tgt->type);
748 	return r;
749 }
750 
751 static int setup_indexes(struct dm_table *t)
752 {
753 	int i;
754 	unsigned int total = 0;
755 	sector_t *indexes;
756 
757 	/* allocate the space for *all* the indexes */
758 	for (i = t->depth - 2; i >= 0; i--) {
759 		t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
760 		total += t->counts[i];
761 	}
762 
763 	indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
764 	if (!indexes)
765 		return -ENOMEM;
766 
767 	/* set up internal nodes, bottom-up */
768 	for (i = t->depth - 2; i >= 0; i--) {
769 		t->index[i] = indexes;
770 		indexes += (KEYS_PER_NODE * t->counts[i]);
771 		setup_btree_index(i, t);
772 	}
773 
774 	return 0;
775 }
776 
777 /*
778  * Builds the btree to index the map.
779  */
780 int dm_table_complete(struct dm_table *t)
781 {
782 	int r = 0;
783 	unsigned int leaf_nodes;
784 
785 	check_for_valid_limits(&t->limits);
786 
787 	/* how many indexes will the btree have ? */
788 	leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
789 	t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
790 
791 	/* leaf layer has already been set up */
792 	t->counts[t->depth - 1] = leaf_nodes;
793 	t->index[t->depth - 1] = t->highs;
794 
795 	if (t->depth >= 2)
796 		r = setup_indexes(t);
797 
798 	return r;
799 }
800 
801 static DEFINE_MUTEX(_event_lock);
802 void dm_table_event_callback(struct dm_table *t,
803 			     void (*fn)(void *), void *context)
804 {
805 	mutex_lock(&_event_lock);
806 	t->event_fn = fn;
807 	t->event_context = context;
808 	mutex_unlock(&_event_lock);
809 }
810 
811 void dm_table_event(struct dm_table *t)
812 {
813 	/*
814 	 * You can no longer call dm_table_event() from interrupt
815 	 * context, use a bottom half instead.
816 	 */
817 	BUG_ON(in_interrupt());
818 
819 	mutex_lock(&_event_lock);
820 	if (t->event_fn)
821 		t->event_fn(t->event_context);
822 	mutex_unlock(&_event_lock);
823 }
824 
825 sector_t dm_table_get_size(struct dm_table *t)
826 {
827 	return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
828 }
829 
830 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
831 {
832 	if (index >= t->num_targets)
833 		return NULL;
834 
835 	return t->targets + index;
836 }
837 
838 /*
839  * Search the btree for the correct target.
840  *
841  * Caller should check returned pointer with dm_target_is_valid()
842  * to trap I/O beyond end of device.
843  */
844 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
845 {
846 	unsigned int l, n = 0, k = 0;
847 	sector_t *node;
848 
849 	for (l = 0; l < t->depth; l++) {
850 		n = get_child(n, k);
851 		node = get_node(t, l, n);
852 
853 		for (k = 0; k < KEYS_PER_NODE; k++)
854 			if (node[k] >= sector)
855 				break;
856 	}
857 
858 	return &t->targets[(KEYS_PER_NODE * n) + k];
859 }
860 
861 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q)
862 {
863 	/*
864 	 * Make sure we obey the optimistic sub devices
865 	 * restrictions.
866 	 */
867 	blk_queue_max_sectors(q, t->limits.max_sectors);
868 	q->max_phys_segments = t->limits.max_phys_segments;
869 	q->max_hw_segments = t->limits.max_hw_segments;
870 	q->hardsect_size = t->limits.hardsect_size;
871 	q->max_segment_size = t->limits.max_segment_size;
872 	q->max_hw_sectors = t->limits.max_hw_sectors;
873 	q->seg_boundary_mask = t->limits.seg_boundary_mask;
874 	q->bounce_pfn = t->limits.bounce_pfn;
875 
876 	if (t->limits.no_cluster)
877 		queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q);
878 	else
879 		queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, q);
880 
881 }
882 
883 unsigned int dm_table_get_num_targets(struct dm_table *t)
884 {
885 	return t->num_targets;
886 }
887 
888 struct list_head *dm_table_get_devices(struct dm_table *t)
889 {
890 	return &t->devices;
891 }
892 
893 int dm_table_get_mode(struct dm_table *t)
894 {
895 	return t->mode;
896 }
897 
898 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
899 {
900 	int i = t->num_targets;
901 	struct dm_target *ti = t->targets;
902 
903 	while (i--) {
904 		if (postsuspend) {
905 			if (ti->type->postsuspend)
906 				ti->type->postsuspend(ti);
907 		} else if (ti->type->presuspend)
908 			ti->type->presuspend(ti);
909 
910 		ti++;
911 	}
912 }
913 
914 void dm_table_presuspend_targets(struct dm_table *t)
915 {
916 	if (!t)
917 		return;
918 
919 	suspend_targets(t, 0);
920 }
921 
922 void dm_table_postsuspend_targets(struct dm_table *t)
923 {
924 	if (!t)
925 		return;
926 
927 	suspend_targets(t, 1);
928 }
929 
930 int dm_table_resume_targets(struct dm_table *t)
931 {
932 	int i, r = 0;
933 
934 	for (i = 0; i < t->num_targets; i++) {
935 		struct dm_target *ti = t->targets + i;
936 
937 		if (!ti->type->preresume)
938 			continue;
939 
940 		r = ti->type->preresume(ti);
941 		if (r)
942 			return r;
943 	}
944 
945 	for (i = 0; i < t->num_targets; i++) {
946 		struct dm_target *ti = t->targets + i;
947 
948 		if (ti->type->resume)
949 			ti->type->resume(ti);
950 	}
951 
952 	return 0;
953 }
954 
955 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
956 {
957 	struct dm_dev *dd;
958 	struct list_head *devices = dm_table_get_devices(t);
959 	int r = 0;
960 
961 	list_for_each_entry(dd, devices, list) {
962 		struct request_queue *q = bdev_get_queue(dd->bdev);
963 		r |= bdi_congested(&q->backing_dev_info, bdi_bits);
964 	}
965 
966 	return r;
967 }
968 
969 void dm_table_unplug_all(struct dm_table *t)
970 {
971 	struct dm_dev *dd;
972 	struct list_head *devices = dm_table_get_devices(t);
973 
974 	list_for_each_entry(dd, devices, list) {
975 		struct request_queue *q = bdev_get_queue(dd->bdev);
976 
977 		blk_unplug(q);
978 	}
979 }
980 
981 struct mapped_device *dm_table_get_md(struct dm_table *t)
982 {
983 	dm_get(t->md);
984 
985 	return t->md;
986 }
987 
988 EXPORT_SYMBOL(dm_vcalloc);
989 EXPORT_SYMBOL(dm_get_device);
990 EXPORT_SYMBOL(dm_put_device);
991 EXPORT_SYMBOL(dm_table_event);
992 EXPORT_SYMBOL(dm_table_get_size);
993 EXPORT_SYMBOL(dm_table_get_mode);
994 EXPORT_SYMBOL(dm_table_get_md);
995 EXPORT_SYMBOL(dm_table_put);
996 EXPORT_SYMBOL(dm_table_get);
997 EXPORT_SYMBOL(dm_table_unplug_all);
998