xref: /linux/drivers/md/persistent-data/dm-btree.c (revision b85d45947951d23cb22d90caecf4c1eb81342c96)
1 /*
2  * Copyright (C) 2011 Red Hat, Inc.
3  *
4  * This file is released under the GPL.
5  */
6 
7 #include "dm-btree-internal.h"
8 #include "dm-space-map.h"
9 #include "dm-transaction-manager.h"
10 
11 #include <linux/export.h>
12 #include <linux/device-mapper.h>
13 
14 #define DM_MSG_PREFIX "btree"
15 
16 /*----------------------------------------------------------------
17  * Array manipulation
18  *--------------------------------------------------------------*/
19 static void memcpy_disk(void *dest, const void *src, size_t len)
20 	__dm_written_to_disk(src)
21 {
22 	memcpy(dest, src, len);
23 	__dm_unbless_for_disk(src);
24 }
25 
26 static void array_insert(void *base, size_t elt_size, unsigned nr_elts,
27 			 unsigned index, void *elt)
28 	__dm_written_to_disk(elt)
29 {
30 	if (index < nr_elts)
31 		memmove(base + (elt_size * (index + 1)),
32 			base + (elt_size * index),
33 			(nr_elts - index) * elt_size);
34 
35 	memcpy_disk(base + (elt_size * index), elt, elt_size);
36 }
37 
38 /*----------------------------------------------------------------*/
39 
40 /* makes the assumption that no two keys are the same. */
41 static int bsearch(struct btree_node *n, uint64_t key, int want_hi)
42 {
43 	int lo = -1, hi = le32_to_cpu(n->header.nr_entries);
44 
45 	while (hi - lo > 1) {
46 		int mid = lo + ((hi - lo) / 2);
47 		uint64_t mid_key = le64_to_cpu(n->keys[mid]);
48 
49 		if (mid_key == key)
50 			return mid;
51 
52 		if (mid_key < key)
53 			lo = mid;
54 		else
55 			hi = mid;
56 	}
57 
58 	return want_hi ? hi : lo;
59 }
60 
61 int lower_bound(struct btree_node *n, uint64_t key)
62 {
63 	return bsearch(n, key, 0);
64 }
65 
66 void inc_children(struct dm_transaction_manager *tm, struct btree_node *n,
67 		  struct dm_btree_value_type *vt)
68 {
69 	unsigned i;
70 	uint32_t nr_entries = le32_to_cpu(n->header.nr_entries);
71 
72 	if (le32_to_cpu(n->header.flags) & INTERNAL_NODE)
73 		for (i = 0; i < nr_entries; i++)
74 			dm_tm_inc(tm, value64(n, i));
75 	else if (vt->inc)
76 		for (i = 0; i < nr_entries; i++)
77 			vt->inc(vt->context, value_ptr(n, i));
78 }
79 
80 static int insert_at(size_t value_size, struct btree_node *node, unsigned index,
81 		      uint64_t key, void *value)
82 		      __dm_written_to_disk(value)
83 {
84 	uint32_t nr_entries = le32_to_cpu(node->header.nr_entries);
85 	__le64 key_le = cpu_to_le64(key);
86 
87 	if (index > nr_entries ||
88 	    index >= le32_to_cpu(node->header.max_entries)) {
89 		DMERR("too many entries in btree node for insert");
90 		__dm_unbless_for_disk(value);
91 		return -ENOMEM;
92 	}
93 
94 	__dm_bless_for_disk(&key_le);
95 
96 	array_insert(node->keys, sizeof(*node->keys), nr_entries, index, &key_le);
97 	array_insert(value_base(node), value_size, nr_entries, index, value);
98 	node->header.nr_entries = cpu_to_le32(nr_entries + 1);
99 
100 	return 0;
101 }
102 
103 /*----------------------------------------------------------------*/
104 
105 /*
106  * We want 3n entries (for some n).  This works more nicely for repeated
107  * insert remove loops than (2n + 1).
108  */
109 static uint32_t calc_max_entries(size_t value_size, size_t block_size)
110 {
111 	uint32_t total, n;
112 	size_t elt_size = sizeof(uint64_t) + value_size; /* key + value */
113 
114 	block_size -= sizeof(struct node_header);
115 	total = block_size / elt_size;
116 	n = total / 3;		/* rounds down */
117 
118 	return 3 * n;
119 }
120 
121 int dm_btree_empty(struct dm_btree_info *info, dm_block_t *root)
122 {
123 	int r;
124 	struct dm_block *b;
125 	struct btree_node *n;
126 	size_t block_size;
127 	uint32_t max_entries;
128 
129 	r = new_block(info, &b);
130 	if (r < 0)
131 		return r;
132 
133 	block_size = dm_bm_block_size(dm_tm_get_bm(info->tm));
134 	max_entries = calc_max_entries(info->value_type.size, block_size);
135 
136 	n = dm_block_data(b);
137 	memset(n, 0, block_size);
138 	n->header.flags = cpu_to_le32(LEAF_NODE);
139 	n->header.nr_entries = cpu_to_le32(0);
140 	n->header.max_entries = cpu_to_le32(max_entries);
141 	n->header.value_size = cpu_to_le32(info->value_type.size);
142 
143 	*root = dm_block_location(b);
144 	return unlock_block(info, b);
145 }
146 EXPORT_SYMBOL_GPL(dm_btree_empty);
147 
148 /*----------------------------------------------------------------*/
149 
150 /*
151  * Deletion uses a recursive algorithm, since we have limited stack space
152  * we explicitly manage our own stack on the heap.
153  */
154 #define MAX_SPINE_DEPTH 64
155 struct frame {
156 	struct dm_block *b;
157 	struct btree_node *n;
158 	unsigned level;
159 	unsigned nr_children;
160 	unsigned current_child;
161 };
162 
163 struct del_stack {
164 	struct dm_btree_info *info;
165 	struct dm_transaction_manager *tm;
166 	int top;
167 	struct frame spine[MAX_SPINE_DEPTH];
168 };
169 
170 static int top_frame(struct del_stack *s, struct frame **f)
171 {
172 	if (s->top < 0) {
173 		DMERR("btree deletion stack empty");
174 		return -EINVAL;
175 	}
176 
177 	*f = s->spine + s->top;
178 
179 	return 0;
180 }
181 
182 static int unprocessed_frames(struct del_stack *s)
183 {
184 	return s->top >= 0;
185 }
186 
187 static void prefetch_children(struct del_stack *s, struct frame *f)
188 {
189 	unsigned i;
190 	struct dm_block_manager *bm = dm_tm_get_bm(s->tm);
191 
192 	for (i = 0; i < f->nr_children; i++)
193 		dm_bm_prefetch(bm, value64(f->n, i));
194 }
195 
196 static bool is_internal_level(struct dm_btree_info *info, struct frame *f)
197 {
198 	return f->level < (info->levels - 1);
199 }
200 
201 static int push_frame(struct del_stack *s, dm_block_t b, unsigned level)
202 {
203 	int r;
204 	uint32_t ref_count;
205 
206 	if (s->top >= MAX_SPINE_DEPTH - 1) {
207 		DMERR("btree deletion stack out of memory");
208 		return -ENOMEM;
209 	}
210 
211 	r = dm_tm_ref(s->tm, b, &ref_count);
212 	if (r)
213 		return r;
214 
215 	if (ref_count > 1)
216 		/*
217 		 * This is a shared node, so we can just decrement it's
218 		 * reference counter and leave the children.
219 		 */
220 		dm_tm_dec(s->tm, b);
221 
222 	else {
223 		uint32_t flags;
224 		struct frame *f = s->spine + ++s->top;
225 
226 		r = dm_tm_read_lock(s->tm, b, &btree_node_validator, &f->b);
227 		if (r) {
228 			s->top--;
229 			return r;
230 		}
231 
232 		f->n = dm_block_data(f->b);
233 		f->level = level;
234 		f->nr_children = le32_to_cpu(f->n->header.nr_entries);
235 		f->current_child = 0;
236 
237 		flags = le32_to_cpu(f->n->header.flags);
238 		if (flags & INTERNAL_NODE || is_internal_level(s->info, f))
239 			prefetch_children(s, f);
240 	}
241 
242 	return 0;
243 }
244 
245 static void pop_frame(struct del_stack *s)
246 {
247 	struct frame *f = s->spine + s->top--;
248 
249 	dm_tm_dec(s->tm, dm_block_location(f->b));
250 	dm_tm_unlock(s->tm, f->b);
251 }
252 
253 int dm_btree_del(struct dm_btree_info *info, dm_block_t root)
254 {
255 	int r;
256 	struct del_stack *s;
257 
258 	s = kmalloc(sizeof(*s), GFP_NOIO);
259 	if (!s)
260 		return -ENOMEM;
261 	s->info = info;
262 	s->tm = info->tm;
263 	s->top = -1;
264 
265 	r = push_frame(s, root, 0);
266 	if (r)
267 		goto out;
268 
269 	while (unprocessed_frames(s)) {
270 		uint32_t flags;
271 		struct frame *f;
272 		dm_block_t b;
273 
274 		r = top_frame(s, &f);
275 		if (r)
276 			goto out;
277 
278 		if (f->current_child >= f->nr_children) {
279 			pop_frame(s);
280 			continue;
281 		}
282 
283 		flags = le32_to_cpu(f->n->header.flags);
284 		if (flags & INTERNAL_NODE) {
285 			b = value64(f->n, f->current_child);
286 			f->current_child++;
287 			r = push_frame(s, b, f->level);
288 			if (r)
289 				goto out;
290 
291 		} else if (is_internal_level(info, f)) {
292 			b = value64(f->n, f->current_child);
293 			f->current_child++;
294 			r = push_frame(s, b, f->level + 1);
295 			if (r)
296 				goto out;
297 
298 		} else {
299 			if (info->value_type.dec) {
300 				unsigned i;
301 
302 				for (i = 0; i < f->nr_children; i++)
303 					info->value_type.dec(info->value_type.context,
304 							     value_ptr(f->n, i));
305 			}
306 			pop_frame(s);
307 		}
308 	}
309 
310 out:
311 	kfree(s);
312 	return r;
313 }
314 EXPORT_SYMBOL_GPL(dm_btree_del);
315 
316 /*----------------------------------------------------------------*/
317 
318 static int btree_lookup_raw(struct ro_spine *s, dm_block_t block, uint64_t key,
319 			    int (*search_fn)(struct btree_node *, uint64_t),
320 			    uint64_t *result_key, void *v, size_t value_size)
321 {
322 	int i, r;
323 	uint32_t flags, nr_entries;
324 
325 	do {
326 		r = ro_step(s, block);
327 		if (r < 0)
328 			return r;
329 
330 		i = search_fn(ro_node(s), key);
331 
332 		flags = le32_to_cpu(ro_node(s)->header.flags);
333 		nr_entries = le32_to_cpu(ro_node(s)->header.nr_entries);
334 		if (i < 0 || i >= nr_entries)
335 			return -ENODATA;
336 
337 		if (flags & INTERNAL_NODE)
338 			block = value64(ro_node(s), i);
339 
340 	} while (!(flags & LEAF_NODE));
341 
342 	*result_key = le64_to_cpu(ro_node(s)->keys[i]);
343 	memcpy(v, value_ptr(ro_node(s), i), value_size);
344 
345 	return 0;
346 }
347 
348 int dm_btree_lookup(struct dm_btree_info *info, dm_block_t root,
349 		    uint64_t *keys, void *value_le)
350 {
351 	unsigned level, last_level = info->levels - 1;
352 	int r = -ENODATA;
353 	uint64_t rkey;
354 	__le64 internal_value_le;
355 	struct ro_spine spine;
356 
357 	init_ro_spine(&spine, info);
358 	for (level = 0; level < info->levels; level++) {
359 		size_t size;
360 		void *value_p;
361 
362 		if (level == last_level) {
363 			value_p = value_le;
364 			size = info->value_type.size;
365 
366 		} else {
367 			value_p = &internal_value_le;
368 			size = sizeof(uint64_t);
369 		}
370 
371 		r = btree_lookup_raw(&spine, root, keys[level],
372 				     lower_bound, &rkey,
373 				     value_p, size);
374 
375 		if (!r) {
376 			if (rkey != keys[level]) {
377 				exit_ro_spine(&spine);
378 				return -ENODATA;
379 			}
380 		} else {
381 			exit_ro_spine(&spine);
382 			return r;
383 		}
384 
385 		root = le64_to_cpu(internal_value_le);
386 	}
387 	exit_ro_spine(&spine);
388 
389 	return r;
390 }
391 EXPORT_SYMBOL_GPL(dm_btree_lookup);
392 
393 /*
394  * Splits a node by creating a sibling node and shifting half the nodes
395  * contents across.  Assumes there is a parent node, and it has room for
396  * another child.
397  *
398  * Before:
399  *	  +--------+
400  *	  | Parent |
401  *	  +--------+
402  *	     |
403  *	     v
404  *	+----------+
405  *	| A ++++++ |
406  *	+----------+
407  *
408  *
409  * After:
410  *		+--------+
411  *		| Parent |
412  *		+--------+
413  *		  |	|
414  *		  v	+------+
415  *	    +---------+	       |
416  *	    | A* +++  |	       v
417  *	    +---------+	  +-------+
418  *			  | B +++ |
419  *			  +-------+
420  *
421  * Where A* is a shadow of A.
422  */
423 static int btree_split_sibling(struct shadow_spine *s, unsigned parent_index,
424 			       uint64_t key)
425 {
426 	int r;
427 	size_t size;
428 	unsigned nr_left, nr_right;
429 	struct dm_block *left, *right, *parent;
430 	struct btree_node *ln, *rn, *pn;
431 	__le64 location;
432 
433 	left = shadow_current(s);
434 
435 	r = new_block(s->info, &right);
436 	if (r < 0)
437 		return r;
438 
439 	ln = dm_block_data(left);
440 	rn = dm_block_data(right);
441 
442 	nr_left = le32_to_cpu(ln->header.nr_entries) / 2;
443 	nr_right = le32_to_cpu(ln->header.nr_entries) - nr_left;
444 
445 	ln->header.nr_entries = cpu_to_le32(nr_left);
446 
447 	rn->header.flags = ln->header.flags;
448 	rn->header.nr_entries = cpu_to_le32(nr_right);
449 	rn->header.max_entries = ln->header.max_entries;
450 	rn->header.value_size = ln->header.value_size;
451 	memcpy(rn->keys, ln->keys + nr_left, nr_right * sizeof(rn->keys[0]));
452 
453 	size = le32_to_cpu(ln->header.flags) & INTERNAL_NODE ?
454 		sizeof(uint64_t) : s->info->value_type.size;
455 	memcpy(value_ptr(rn, 0), value_ptr(ln, nr_left),
456 	       size * nr_right);
457 
458 	/*
459 	 * Patch up the parent
460 	 */
461 	parent = shadow_parent(s);
462 
463 	pn = dm_block_data(parent);
464 	location = cpu_to_le64(dm_block_location(left));
465 	__dm_bless_for_disk(&location);
466 	memcpy_disk(value_ptr(pn, parent_index),
467 		    &location, sizeof(__le64));
468 
469 	location = cpu_to_le64(dm_block_location(right));
470 	__dm_bless_for_disk(&location);
471 
472 	r = insert_at(sizeof(__le64), pn, parent_index + 1,
473 		      le64_to_cpu(rn->keys[0]), &location);
474 	if (r)
475 		return r;
476 
477 	if (key < le64_to_cpu(rn->keys[0])) {
478 		unlock_block(s->info, right);
479 		s->nodes[1] = left;
480 	} else {
481 		unlock_block(s->info, left);
482 		s->nodes[1] = right;
483 	}
484 
485 	return 0;
486 }
487 
488 /*
489  * Splits a node by creating two new children beneath the given node.
490  *
491  * Before:
492  *	  +----------+
493  *	  | A ++++++ |
494  *	  +----------+
495  *
496  *
497  * After:
498  *	+------------+
499  *	| A (shadow) |
500  *	+------------+
501  *	    |	|
502  *   +------+	+----+
503  *   |		     |
504  *   v		     v
505  * +-------+	 +-------+
506  * | B +++ |	 | C +++ |
507  * +-------+	 +-------+
508  */
509 static int btree_split_beneath(struct shadow_spine *s, uint64_t key)
510 {
511 	int r;
512 	size_t size;
513 	unsigned nr_left, nr_right;
514 	struct dm_block *left, *right, *new_parent;
515 	struct btree_node *pn, *ln, *rn;
516 	__le64 val;
517 
518 	new_parent = shadow_current(s);
519 
520 	r = new_block(s->info, &left);
521 	if (r < 0)
522 		return r;
523 
524 	r = new_block(s->info, &right);
525 	if (r < 0) {
526 		unlock_block(s->info, left);
527 		return r;
528 	}
529 
530 	pn = dm_block_data(new_parent);
531 	ln = dm_block_data(left);
532 	rn = dm_block_data(right);
533 
534 	nr_left = le32_to_cpu(pn->header.nr_entries) / 2;
535 	nr_right = le32_to_cpu(pn->header.nr_entries) - nr_left;
536 
537 	ln->header.flags = pn->header.flags;
538 	ln->header.nr_entries = cpu_to_le32(nr_left);
539 	ln->header.max_entries = pn->header.max_entries;
540 	ln->header.value_size = pn->header.value_size;
541 
542 	rn->header.flags = pn->header.flags;
543 	rn->header.nr_entries = cpu_to_le32(nr_right);
544 	rn->header.max_entries = pn->header.max_entries;
545 	rn->header.value_size = pn->header.value_size;
546 
547 	memcpy(ln->keys, pn->keys, nr_left * sizeof(pn->keys[0]));
548 	memcpy(rn->keys, pn->keys + nr_left, nr_right * sizeof(pn->keys[0]));
549 
550 	size = le32_to_cpu(pn->header.flags) & INTERNAL_NODE ?
551 		sizeof(__le64) : s->info->value_type.size;
552 	memcpy(value_ptr(ln, 0), value_ptr(pn, 0), nr_left * size);
553 	memcpy(value_ptr(rn, 0), value_ptr(pn, nr_left),
554 	       nr_right * size);
555 
556 	/* new_parent should just point to l and r now */
557 	pn->header.flags = cpu_to_le32(INTERNAL_NODE);
558 	pn->header.nr_entries = cpu_to_le32(2);
559 	pn->header.max_entries = cpu_to_le32(
560 		calc_max_entries(sizeof(__le64),
561 				 dm_bm_block_size(
562 					 dm_tm_get_bm(s->info->tm))));
563 	pn->header.value_size = cpu_to_le32(sizeof(__le64));
564 
565 	val = cpu_to_le64(dm_block_location(left));
566 	__dm_bless_for_disk(&val);
567 	pn->keys[0] = ln->keys[0];
568 	memcpy_disk(value_ptr(pn, 0), &val, sizeof(__le64));
569 
570 	val = cpu_to_le64(dm_block_location(right));
571 	__dm_bless_for_disk(&val);
572 	pn->keys[1] = rn->keys[0];
573 	memcpy_disk(value_ptr(pn, 1), &val, sizeof(__le64));
574 
575 	/*
576 	 * rejig the spine.  This is ugly, since it knows too
577 	 * much about the spine
578 	 */
579 	if (s->nodes[0] != new_parent) {
580 		unlock_block(s->info, s->nodes[0]);
581 		s->nodes[0] = new_parent;
582 	}
583 	if (key < le64_to_cpu(rn->keys[0])) {
584 		unlock_block(s->info, right);
585 		s->nodes[1] = left;
586 	} else {
587 		unlock_block(s->info, left);
588 		s->nodes[1] = right;
589 	}
590 	s->count = 2;
591 
592 	return 0;
593 }
594 
595 static int btree_insert_raw(struct shadow_spine *s, dm_block_t root,
596 			    struct dm_btree_value_type *vt,
597 			    uint64_t key, unsigned *index)
598 {
599 	int r, i = *index, top = 1;
600 	struct btree_node *node;
601 
602 	for (;;) {
603 		r = shadow_step(s, root, vt);
604 		if (r < 0)
605 			return r;
606 
607 		node = dm_block_data(shadow_current(s));
608 
609 		/*
610 		 * We have to patch up the parent node, ugly, but I don't
611 		 * see a way to do this automatically as part of the spine
612 		 * op.
613 		 */
614 		if (shadow_has_parent(s) && i >= 0) { /* FIXME: second clause unness. */
615 			__le64 location = cpu_to_le64(dm_block_location(shadow_current(s)));
616 
617 			__dm_bless_for_disk(&location);
618 			memcpy_disk(value_ptr(dm_block_data(shadow_parent(s)), i),
619 				    &location, sizeof(__le64));
620 		}
621 
622 		node = dm_block_data(shadow_current(s));
623 
624 		if (node->header.nr_entries == node->header.max_entries) {
625 			if (top)
626 				r = btree_split_beneath(s, key);
627 			else
628 				r = btree_split_sibling(s, i, key);
629 
630 			if (r < 0)
631 				return r;
632 		}
633 
634 		node = dm_block_data(shadow_current(s));
635 
636 		i = lower_bound(node, key);
637 
638 		if (le32_to_cpu(node->header.flags) & LEAF_NODE)
639 			break;
640 
641 		if (i < 0) {
642 			/* change the bounds on the lowest key */
643 			node->keys[0] = cpu_to_le64(key);
644 			i = 0;
645 		}
646 
647 		root = value64(node, i);
648 		top = 0;
649 	}
650 
651 	if (i < 0 || le64_to_cpu(node->keys[i]) != key)
652 		i++;
653 
654 	*index = i;
655 	return 0;
656 }
657 
658 static int insert(struct dm_btree_info *info, dm_block_t root,
659 		  uint64_t *keys, void *value, dm_block_t *new_root,
660 		  int *inserted)
661 		  __dm_written_to_disk(value)
662 {
663 	int r, need_insert;
664 	unsigned level, index = -1, last_level = info->levels - 1;
665 	dm_block_t block = root;
666 	struct shadow_spine spine;
667 	struct btree_node *n;
668 	struct dm_btree_value_type le64_type;
669 
670 	init_le64_type(info->tm, &le64_type);
671 	init_shadow_spine(&spine, info);
672 
673 	for (level = 0; level < (info->levels - 1); level++) {
674 		r = btree_insert_raw(&spine, block, &le64_type, keys[level], &index);
675 		if (r < 0)
676 			goto bad;
677 
678 		n = dm_block_data(shadow_current(&spine));
679 		need_insert = ((index >= le32_to_cpu(n->header.nr_entries)) ||
680 			       (le64_to_cpu(n->keys[index]) != keys[level]));
681 
682 		if (need_insert) {
683 			dm_block_t new_tree;
684 			__le64 new_le;
685 
686 			r = dm_btree_empty(info, &new_tree);
687 			if (r < 0)
688 				goto bad;
689 
690 			new_le = cpu_to_le64(new_tree);
691 			__dm_bless_for_disk(&new_le);
692 
693 			r = insert_at(sizeof(uint64_t), n, index,
694 				      keys[level], &new_le);
695 			if (r)
696 				goto bad;
697 		}
698 
699 		if (level < last_level)
700 			block = value64(n, index);
701 	}
702 
703 	r = btree_insert_raw(&spine, block, &info->value_type,
704 			     keys[level], &index);
705 	if (r < 0)
706 		goto bad;
707 
708 	n = dm_block_data(shadow_current(&spine));
709 	need_insert = ((index >= le32_to_cpu(n->header.nr_entries)) ||
710 		       (le64_to_cpu(n->keys[index]) != keys[level]));
711 
712 	if (need_insert) {
713 		if (inserted)
714 			*inserted = 1;
715 
716 		r = insert_at(info->value_type.size, n, index,
717 			      keys[level], value);
718 		if (r)
719 			goto bad_unblessed;
720 	} else {
721 		if (inserted)
722 			*inserted = 0;
723 
724 		if (info->value_type.dec &&
725 		    (!info->value_type.equal ||
726 		     !info->value_type.equal(
727 			     info->value_type.context,
728 			     value_ptr(n, index),
729 			     value))) {
730 			info->value_type.dec(info->value_type.context,
731 					     value_ptr(n, index));
732 		}
733 		memcpy_disk(value_ptr(n, index),
734 			    value, info->value_type.size);
735 	}
736 
737 	*new_root = shadow_root(&spine);
738 	exit_shadow_spine(&spine);
739 
740 	return 0;
741 
742 bad:
743 	__dm_unbless_for_disk(value);
744 bad_unblessed:
745 	exit_shadow_spine(&spine);
746 	return r;
747 }
748 
749 int dm_btree_insert(struct dm_btree_info *info, dm_block_t root,
750 		    uint64_t *keys, void *value, dm_block_t *new_root)
751 		    __dm_written_to_disk(value)
752 {
753 	return insert(info, root, keys, value, new_root, NULL);
754 }
755 EXPORT_SYMBOL_GPL(dm_btree_insert);
756 
757 int dm_btree_insert_notify(struct dm_btree_info *info, dm_block_t root,
758 			   uint64_t *keys, void *value, dm_block_t *new_root,
759 			   int *inserted)
760 			   __dm_written_to_disk(value)
761 {
762 	return insert(info, root, keys, value, new_root, inserted);
763 }
764 EXPORT_SYMBOL_GPL(dm_btree_insert_notify);
765 
766 /*----------------------------------------------------------------*/
767 
768 static int find_key(struct ro_spine *s, dm_block_t block, bool find_highest,
769 		    uint64_t *result_key, dm_block_t *next_block)
770 {
771 	int i, r;
772 	uint32_t flags;
773 
774 	do {
775 		r = ro_step(s, block);
776 		if (r < 0)
777 			return r;
778 
779 		flags = le32_to_cpu(ro_node(s)->header.flags);
780 		i = le32_to_cpu(ro_node(s)->header.nr_entries);
781 		if (!i)
782 			return -ENODATA;
783 		else
784 			i--;
785 
786 		if (find_highest)
787 			*result_key = le64_to_cpu(ro_node(s)->keys[i]);
788 		else
789 			*result_key = le64_to_cpu(ro_node(s)->keys[0]);
790 
791 		if (next_block || flags & INTERNAL_NODE)
792 			block = value64(ro_node(s), i);
793 
794 	} while (flags & INTERNAL_NODE);
795 
796 	if (next_block)
797 		*next_block = block;
798 	return 0;
799 }
800 
801 static int dm_btree_find_key(struct dm_btree_info *info, dm_block_t root,
802 			     bool find_highest, uint64_t *result_keys)
803 {
804 	int r = 0, count = 0, level;
805 	struct ro_spine spine;
806 
807 	init_ro_spine(&spine, info);
808 	for (level = 0; level < info->levels; level++) {
809 		r = find_key(&spine, root, find_highest, result_keys + level,
810 			     level == info->levels - 1 ? NULL : &root);
811 		if (r == -ENODATA) {
812 			r = 0;
813 			break;
814 
815 		} else if (r)
816 			break;
817 
818 		count++;
819 	}
820 	exit_ro_spine(&spine);
821 
822 	return r ? r : count;
823 }
824 
825 int dm_btree_find_highest_key(struct dm_btree_info *info, dm_block_t root,
826 			      uint64_t *result_keys)
827 {
828 	return dm_btree_find_key(info, root, true, result_keys);
829 }
830 EXPORT_SYMBOL_GPL(dm_btree_find_highest_key);
831 
832 int dm_btree_find_lowest_key(struct dm_btree_info *info, dm_block_t root,
833 			     uint64_t *result_keys)
834 {
835 	return dm_btree_find_key(info, root, false, result_keys);
836 }
837 EXPORT_SYMBOL_GPL(dm_btree_find_lowest_key);
838 
839 /*----------------------------------------------------------------*/
840 
841 /*
842  * FIXME: We shouldn't use a recursive algorithm when we have limited stack
843  * space.  Also this only works for single level trees.
844  */
845 static int walk_node(struct dm_btree_info *info, dm_block_t block,
846 		     int (*fn)(void *context, uint64_t *keys, void *leaf),
847 		     void *context)
848 {
849 	int r;
850 	unsigned i, nr;
851 	struct dm_block *node;
852 	struct btree_node *n;
853 	uint64_t keys;
854 
855 	r = bn_read_lock(info, block, &node);
856 	if (r)
857 		return r;
858 
859 	n = dm_block_data(node);
860 
861 	nr = le32_to_cpu(n->header.nr_entries);
862 	for (i = 0; i < nr; i++) {
863 		if (le32_to_cpu(n->header.flags) & INTERNAL_NODE) {
864 			r = walk_node(info, value64(n, i), fn, context);
865 			if (r)
866 				goto out;
867 		} else {
868 			keys = le64_to_cpu(*key_ptr(n, i));
869 			r = fn(context, &keys, value_ptr(n, i));
870 			if (r)
871 				goto out;
872 		}
873 	}
874 
875 out:
876 	dm_tm_unlock(info->tm, node);
877 	return r;
878 }
879 
880 int dm_btree_walk(struct dm_btree_info *info, dm_block_t root,
881 		  int (*fn)(void *context, uint64_t *keys, void *leaf),
882 		  void *context)
883 {
884 	BUG_ON(info->levels > 1);
885 	return walk_node(info, root, fn, context);
886 }
887 EXPORT_SYMBOL_GPL(dm_btree_walk);
888