• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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  *--------------------------------------------------------------*/
memcpy_disk(void * dest,const void * src,size_t len)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 
array_insert(void * base,size_t elt_size,unsigned nr_elts,unsigned index,void * elt)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. */
bsearch(struct btree_node * n,uint64_t key,int want_hi)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 
lower_bound(struct btree_node * n,uint64_t key)61 int lower_bound(struct btree_node *n, uint64_t key)
62 {
63 	return bsearch(n, key, 0);
64 }
65 
upper_bound(struct btree_node * n,uint64_t key)66 static int upper_bound(struct btree_node *n, uint64_t key)
67 {
68 	return bsearch(n, key, 1);
69 }
70 
inc_children(struct dm_transaction_manager * tm,struct btree_node * n,struct dm_btree_value_type * vt)71 void inc_children(struct dm_transaction_manager *tm, struct btree_node *n,
72 		  struct dm_btree_value_type *vt)
73 {
74 	unsigned i;
75 	uint32_t nr_entries = le32_to_cpu(n->header.nr_entries);
76 
77 	if (le32_to_cpu(n->header.flags) & INTERNAL_NODE)
78 		for (i = 0; i < nr_entries; i++)
79 			dm_tm_inc(tm, value64(n, i));
80 	else if (vt->inc)
81 		for (i = 0; i < nr_entries; i++)
82 			vt->inc(vt->context, value_ptr(n, i));
83 }
84 
insert_at(size_t value_size,struct btree_node * node,unsigned index,uint64_t key,void * value)85 static int insert_at(size_t value_size, struct btree_node *node, unsigned index,
86 		     uint64_t key, void *value)
87 	__dm_written_to_disk(value)
88 {
89 	uint32_t nr_entries = le32_to_cpu(node->header.nr_entries);
90 	uint32_t max_entries = le32_to_cpu(node->header.max_entries);
91 	__le64 key_le = cpu_to_le64(key);
92 
93 	if (index > nr_entries ||
94 	    index >= max_entries ||
95 	    nr_entries >= max_entries) {
96 		DMERR("too many entries in btree node for insert");
97 		__dm_unbless_for_disk(value);
98 		return -ENOMEM;
99 	}
100 
101 	__dm_bless_for_disk(&key_le);
102 
103 	array_insert(node->keys, sizeof(*node->keys), nr_entries, index, &key_le);
104 	array_insert(value_base(node), value_size, nr_entries, index, value);
105 	node->header.nr_entries = cpu_to_le32(nr_entries + 1);
106 
107 	return 0;
108 }
109 
110 /*----------------------------------------------------------------*/
111 
112 /*
113  * We want 3n entries (for some n).  This works more nicely for repeated
114  * insert remove loops than (2n + 1).
115  */
calc_max_entries(size_t value_size,size_t block_size)116 static uint32_t calc_max_entries(size_t value_size, size_t block_size)
117 {
118 	uint32_t total, n;
119 	size_t elt_size = sizeof(uint64_t) + value_size; /* key + value */
120 
121 	block_size -= sizeof(struct node_header);
122 	total = block_size / elt_size;
123 	n = total / 3;		/* rounds down */
124 
125 	return 3 * n;
126 }
127 
dm_btree_empty(struct dm_btree_info * info,dm_block_t * root)128 int dm_btree_empty(struct dm_btree_info *info, dm_block_t *root)
129 {
130 	int r;
131 	struct dm_block *b;
132 	struct btree_node *n;
133 	size_t block_size;
134 	uint32_t max_entries;
135 
136 	r = new_block(info, &b);
137 	if (r < 0)
138 		return r;
139 
140 	block_size = dm_bm_block_size(dm_tm_get_bm(info->tm));
141 	max_entries = calc_max_entries(info->value_type.size, block_size);
142 
143 	n = dm_block_data(b);
144 	memset(n, 0, block_size);
145 	n->header.flags = cpu_to_le32(LEAF_NODE);
146 	n->header.nr_entries = cpu_to_le32(0);
147 	n->header.max_entries = cpu_to_le32(max_entries);
148 	n->header.value_size = cpu_to_le32(info->value_type.size);
149 
150 	*root = dm_block_location(b);
151 	unlock_block(info, b);
152 
153 	return 0;
154 }
155 EXPORT_SYMBOL_GPL(dm_btree_empty);
156 
157 /*----------------------------------------------------------------*/
158 
159 /*
160  * Deletion uses a recursive algorithm, since we have limited stack space
161  * we explicitly manage our own stack on the heap.
162  */
163 #define MAX_SPINE_DEPTH 64
164 struct frame {
165 	struct dm_block *b;
166 	struct btree_node *n;
167 	unsigned level;
168 	unsigned nr_children;
169 	unsigned current_child;
170 };
171 
172 struct del_stack {
173 	struct dm_btree_info *info;
174 	struct dm_transaction_manager *tm;
175 	int top;
176 	struct frame spine[MAX_SPINE_DEPTH];
177 };
178 
top_frame(struct del_stack * s,struct frame ** f)179 static int top_frame(struct del_stack *s, struct frame **f)
180 {
181 	if (s->top < 0) {
182 		DMERR("btree deletion stack empty");
183 		return -EINVAL;
184 	}
185 
186 	*f = s->spine + s->top;
187 
188 	return 0;
189 }
190 
unprocessed_frames(struct del_stack * s)191 static int unprocessed_frames(struct del_stack *s)
192 {
193 	return s->top >= 0;
194 }
195 
prefetch_children(struct del_stack * s,struct frame * f)196 static void prefetch_children(struct del_stack *s, struct frame *f)
197 {
198 	unsigned i;
199 	struct dm_block_manager *bm = dm_tm_get_bm(s->tm);
200 
201 	for (i = 0; i < f->nr_children; i++)
202 		dm_bm_prefetch(bm, value64(f->n, i));
203 }
204 
is_internal_level(struct dm_btree_info * info,struct frame * f)205 static bool is_internal_level(struct dm_btree_info *info, struct frame *f)
206 {
207 	return f->level < (info->levels - 1);
208 }
209 
push_frame(struct del_stack * s,dm_block_t b,unsigned level)210 static int push_frame(struct del_stack *s, dm_block_t b, unsigned level)
211 {
212 	int r;
213 	uint32_t ref_count;
214 
215 	if (s->top >= MAX_SPINE_DEPTH - 1) {
216 		DMERR("btree deletion stack out of memory");
217 		return -ENOMEM;
218 	}
219 
220 	r = dm_tm_ref(s->tm, b, &ref_count);
221 	if (r)
222 		return r;
223 
224 	if (ref_count > 1)
225 		/*
226 		 * This is a shared node, so we can just decrement it's
227 		 * reference counter and leave the children.
228 		 */
229 		dm_tm_dec(s->tm, b);
230 
231 	else {
232 		uint32_t flags;
233 		struct frame *f = s->spine + ++s->top;
234 
235 		r = dm_tm_read_lock(s->tm, b, &btree_node_validator, &f->b);
236 		if (r) {
237 			s->top--;
238 			return r;
239 		}
240 
241 		f->n = dm_block_data(f->b);
242 		f->level = level;
243 		f->nr_children = le32_to_cpu(f->n->header.nr_entries);
244 		f->current_child = 0;
245 
246 		flags = le32_to_cpu(f->n->header.flags);
247 		if (flags & INTERNAL_NODE || is_internal_level(s->info, f))
248 			prefetch_children(s, f);
249 	}
250 
251 	return 0;
252 }
253 
pop_frame(struct del_stack * s)254 static void pop_frame(struct del_stack *s)
255 {
256 	struct frame *f = s->spine + s->top--;
257 
258 	dm_tm_dec(s->tm, dm_block_location(f->b));
259 	dm_tm_unlock(s->tm, f->b);
260 }
261 
unlock_all_frames(struct del_stack * s)262 static void unlock_all_frames(struct del_stack *s)
263 {
264 	struct frame *f;
265 
266 	while (unprocessed_frames(s)) {
267 		f = s->spine + s->top--;
268 		dm_tm_unlock(s->tm, f->b);
269 	}
270 }
271 
dm_btree_del(struct dm_btree_info * info,dm_block_t root)272 int dm_btree_del(struct dm_btree_info *info, dm_block_t root)
273 {
274 	int r;
275 	struct del_stack *s;
276 
277 	/*
278 	 * dm_btree_del() is called via an ioctl, as such should be
279 	 * considered an FS op.  We can't recurse back into the FS, so we
280 	 * allocate GFP_NOFS.
281 	 */
282 	s = kmalloc(sizeof(*s), GFP_NOFS);
283 	if (!s)
284 		return -ENOMEM;
285 	s->info = info;
286 	s->tm = info->tm;
287 	s->top = -1;
288 
289 	r = push_frame(s, root, 0);
290 	if (r)
291 		goto out;
292 
293 	while (unprocessed_frames(s)) {
294 		uint32_t flags;
295 		struct frame *f;
296 		dm_block_t b;
297 
298 		r = top_frame(s, &f);
299 		if (r)
300 			goto out;
301 
302 		if (f->current_child >= f->nr_children) {
303 			pop_frame(s);
304 			continue;
305 		}
306 
307 		flags = le32_to_cpu(f->n->header.flags);
308 		if (flags & INTERNAL_NODE) {
309 			b = value64(f->n, f->current_child);
310 			f->current_child++;
311 			r = push_frame(s, b, f->level);
312 			if (r)
313 				goto out;
314 
315 		} else if (is_internal_level(info, f)) {
316 			b = value64(f->n, f->current_child);
317 			f->current_child++;
318 			r = push_frame(s, b, f->level + 1);
319 			if (r)
320 				goto out;
321 
322 		} else {
323 			if (info->value_type.dec) {
324 				unsigned i;
325 
326 				for (i = 0; i < f->nr_children; i++)
327 					info->value_type.dec(info->value_type.context,
328 							     value_ptr(f->n, i));
329 			}
330 			pop_frame(s);
331 		}
332 	}
333 out:
334 	if (r) {
335 		/* cleanup all frames of del_stack */
336 		unlock_all_frames(s);
337 	}
338 	kfree(s);
339 
340 	return r;
341 }
342 EXPORT_SYMBOL_GPL(dm_btree_del);
343 
344 /*----------------------------------------------------------------*/
345 
btree_lookup_raw(struct ro_spine * s,dm_block_t block,uint64_t key,int (* search_fn)(struct btree_node *,uint64_t),uint64_t * result_key,void * v,size_t value_size)346 static int btree_lookup_raw(struct ro_spine *s, dm_block_t block, uint64_t key,
347 			    int (*search_fn)(struct btree_node *, uint64_t),
348 			    uint64_t *result_key, void *v, size_t value_size)
349 {
350 	int i, r;
351 	uint32_t flags, nr_entries;
352 
353 	do {
354 		r = ro_step(s, block);
355 		if (r < 0)
356 			return r;
357 
358 		i = search_fn(ro_node(s), key);
359 
360 		flags = le32_to_cpu(ro_node(s)->header.flags);
361 		nr_entries = le32_to_cpu(ro_node(s)->header.nr_entries);
362 		if (i < 0 || i >= nr_entries)
363 			return -ENODATA;
364 
365 		if (flags & INTERNAL_NODE)
366 			block = value64(ro_node(s), i);
367 
368 	} while (!(flags & LEAF_NODE));
369 
370 	*result_key = le64_to_cpu(ro_node(s)->keys[i]);
371 	if (v)
372 		memcpy(v, value_ptr(ro_node(s), i), value_size);
373 
374 	return 0;
375 }
376 
dm_btree_lookup(struct dm_btree_info * info,dm_block_t root,uint64_t * keys,void * value_le)377 int dm_btree_lookup(struct dm_btree_info *info, dm_block_t root,
378 		    uint64_t *keys, void *value_le)
379 {
380 	unsigned level, last_level = info->levels - 1;
381 	int r = -ENODATA;
382 	uint64_t rkey;
383 	__le64 internal_value_le;
384 	struct ro_spine spine;
385 
386 	init_ro_spine(&spine, info);
387 	for (level = 0; level < info->levels; level++) {
388 		size_t size;
389 		void *value_p;
390 
391 		if (level == last_level) {
392 			value_p = value_le;
393 			size = info->value_type.size;
394 
395 		} else {
396 			value_p = &internal_value_le;
397 			size = sizeof(uint64_t);
398 		}
399 
400 		r = btree_lookup_raw(&spine, root, keys[level],
401 				     lower_bound, &rkey,
402 				     value_p, size);
403 
404 		if (!r) {
405 			if (rkey != keys[level]) {
406 				exit_ro_spine(&spine);
407 				return -ENODATA;
408 			}
409 		} else {
410 			exit_ro_spine(&spine);
411 			return r;
412 		}
413 
414 		root = le64_to_cpu(internal_value_le);
415 	}
416 	exit_ro_spine(&spine);
417 
418 	return r;
419 }
420 EXPORT_SYMBOL_GPL(dm_btree_lookup);
421 
dm_btree_lookup_next_single(struct dm_btree_info * info,dm_block_t root,uint64_t key,uint64_t * rkey,void * value_le)422 static int dm_btree_lookup_next_single(struct dm_btree_info *info, dm_block_t root,
423 				       uint64_t key, uint64_t *rkey, void *value_le)
424 {
425 	int r, i;
426 	uint32_t flags, nr_entries;
427 	struct dm_block *node;
428 	struct btree_node *n;
429 
430 	r = bn_read_lock(info, root, &node);
431 	if (r)
432 		return r;
433 
434 	n = dm_block_data(node);
435 	flags = le32_to_cpu(n->header.flags);
436 	nr_entries = le32_to_cpu(n->header.nr_entries);
437 
438 	if (flags & INTERNAL_NODE) {
439 		i = lower_bound(n, key);
440 		if (i < 0) {
441 			/*
442 			 * avoid early -ENODATA return when all entries are
443 			 * higher than the search @key.
444 			 */
445 			i = 0;
446 		}
447 		if (i >= nr_entries) {
448 			r = -ENODATA;
449 			goto out;
450 		}
451 
452 		r = dm_btree_lookup_next_single(info, value64(n, i), key, rkey, value_le);
453 		if (r == -ENODATA && i < (nr_entries - 1)) {
454 			i++;
455 			r = dm_btree_lookup_next_single(info, value64(n, i), key, rkey, value_le);
456 		}
457 
458 	} else {
459 		i = upper_bound(n, key);
460 		if (i < 0 || i >= nr_entries) {
461 			r = -ENODATA;
462 			goto out;
463 		}
464 
465 		*rkey = le64_to_cpu(n->keys[i]);
466 		memcpy(value_le, value_ptr(n, i), info->value_type.size);
467 	}
468 out:
469 	dm_tm_unlock(info->tm, node);
470 	return r;
471 }
472 
dm_btree_lookup_next(struct dm_btree_info * info,dm_block_t root,uint64_t * keys,uint64_t * rkey,void * value_le)473 int dm_btree_lookup_next(struct dm_btree_info *info, dm_block_t root,
474 			 uint64_t *keys, uint64_t *rkey, void *value_le)
475 {
476 	unsigned level;
477 	int r = -ENODATA;
478 	__le64 internal_value_le;
479 	struct ro_spine spine;
480 
481 	init_ro_spine(&spine, info);
482 	for (level = 0; level < info->levels - 1u; level++) {
483 		r = btree_lookup_raw(&spine, root, keys[level],
484 				     lower_bound, rkey,
485 				     &internal_value_le, sizeof(uint64_t));
486 		if (r)
487 			goto out;
488 
489 		if (*rkey != keys[level]) {
490 			r = -ENODATA;
491 			goto out;
492 		}
493 
494 		root = le64_to_cpu(internal_value_le);
495 	}
496 
497 	r = dm_btree_lookup_next_single(info, root, keys[level], rkey, value_le);
498 out:
499 	exit_ro_spine(&spine);
500 	return r;
501 }
502 
503 EXPORT_SYMBOL_GPL(dm_btree_lookup_next);
504 
505 /*
506  * Splits a node by creating a sibling node and shifting half the nodes
507  * contents across.  Assumes there is a parent node, and it has room for
508  * another child.
509  *
510  * Before:
511  *	  +--------+
512  *	  | Parent |
513  *	  +--------+
514  *	     |
515  *	     v
516  *	+----------+
517  *	| A ++++++ |
518  *	+----------+
519  *
520  *
521  * After:
522  *		+--------+
523  *		| Parent |
524  *		+--------+
525  *		  |	|
526  *		  v	+------+
527  *	    +---------+	       |
528  *	    | A* +++  |	       v
529  *	    +---------+	  +-------+
530  *			  | B +++ |
531  *			  +-------+
532  *
533  * Where A* is a shadow of A.
534  */
btree_split_sibling(struct shadow_spine * s,unsigned parent_index,uint64_t key)535 static int btree_split_sibling(struct shadow_spine *s, unsigned parent_index,
536 			       uint64_t key)
537 {
538 	int r;
539 	size_t size;
540 	unsigned nr_left, nr_right;
541 	struct dm_block *left, *right, *parent;
542 	struct btree_node *ln, *rn, *pn;
543 	__le64 location;
544 
545 	left = shadow_current(s);
546 
547 	r = new_block(s->info, &right);
548 	if (r < 0)
549 		return r;
550 
551 	ln = dm_block_data(left);
552 	rn = dm_block_data(right);
553 
554 	nr_left = le32_to_cpu(ln->header.nr_entries) / 2;
555 	nr_right = le32_to_cpu(ln->header.nr_entries) - nr_left;
556 
557 	ln->header.nr_entries = cpu_to_le32(nr_left);
558 
559 	rn->header.flags = ln->header.flags;
560 	rn->header.nr_entries = cpu_to_le32(nr_right);
561 	rn->header.max_entries = ln->header.max_entries;
562 	rn->header.value_size = ln->header.value_size;
563 	memcpy(rn->keys, ln->keys + nr_left, nr_right * sizeof(rn->keys[0]));
564 
565 	size = le32_to_cpu(ln->header.flags) & INTERNAL_NODE ?
566 		sizeof(uint64_t) : s->info->value_type.size;
567 	memcpy(value_ptr(rn, 0), value_ptr(ln, nr_left),
568 	       size * nr_right);
569 
570 	/*
571 	 * Patch up the parent
572 	 */
573 	parent = shadow_parent(s);
574 
575 	pn = dm_block_data(parent);
576 	location = cpu_to_le64(dm_block_location(left));
577 	__dm_bless_for_disk(&location);
578 	memcpy_disk(value_ptr(pn, parent_index),
579 		    &location, sizeof(__le64));
580 
581 	location = cpu_to_le64(dm_block_location(right));
582 	__dm_bless_for_disk(&location);
583 
584 	r = insert_at(sizeof(__le64), pn, parent_index + 1,
585 		      le64_to_cpu(rn->keys[0]), &location);
586 	if (r) {
587 		unlock_block(s->info, right);
588 		return r;
589 	}
590 
591 	if (key < le64_to_cpu(rn->keys[0])) {
592 		unlock_block(s->info, right);
593 		s->nodes[1] = left;
594 	} else {
595 		unlock_block(s->info, left);
596 		s->nodes[1] = right;
597 	}
598 
599 	return 0;
600 }
601 
602 /*
603  * Splits a node by creating two new children beneath the given node.
604  *
605  * Before:
606  *	  +----------+
607  *	  | A ++++++ |
608  *	  +----------+
609  *
610  *
611  * After:
612  *	+------------+
613  *	| A (shadow) |
614  *	+------------+
615  *	    |	|
616  *   +------+	+----+
617  *   |		     |
618  *   v		     v
619  * +-------+	 +-------+
620  * | B +++ |	 | C +++ |
621  * +-------+	 +-------+
622  */
btree_split_beneath(struct shadow_spine * s,uint64_t key)623 static int btree_split_beneath(struct shadow_spine *s, uint64_t key)
624 {
625 	int r;
626 	size_t size;
627 	unsigned nr_left, nr_right;
628 	struct dm_block *left, *right, *new_parent;
629 	struct btree_node *pn, *ln, *rn;
630 	__le64 val;
631 
632 	new_parent = shadow_current(s);
633 
634 	pn = dm_block_data(new_parent);
635 	size = le32_to_cpu(pn->header.flags) & INTERNAL_NODE ?
636 		sizeof(__le64) : s->info->value_type.size;
637 
638 	/* create & init the left block */
639 	r = new_block(s->info, &left);
640 	if (r < 0)
641 		return r;
642 
643 	ln = dm_block_data(left);
644 	nr_left = le32_to_cpu(pn->header.nr_entries) / 2;
645 
646 	ln->header.flags = pn->header.flags;
647 	ln->header.nr_entries = cpu_to_le32(nr_left);
648 	ln->header.max_entries = pn->header.max_entries;
649 	ln->header.value_size = pn->header.value_size;
650 	memcpy(ln->keys, pn->keys, nr_left * sizeof(pn->keys[0]));
651 	memcpy(value_ptr(ln, 0), value_ptr(pn, 0), nr_left * size);
652 
653 	/* create & init the right block */
654 	r = new_block(s->info, &right);
655 	if (r < 0) {
656 		unlock_block(s->info, left);
657 		return r;
658 	}
659 
660 	rn = dm_block_data(right);
661 	nr_right = le32_to_cpu(pn->header.nr_entries) - nr_left;
662 
663 	rn->header.flags = pn->header.flags;
664 	rn->header.nr_entries = cpu_to_le32(nr_right);
665 	rn->header.max_entries = pn->header.max_entries;
666 	rn->header.value_size = pn->header.value_size;
667 	memcpy(rn->keys, pn->keys + nr_left, nr_right * sizeof(pn->keys[0]));
668 	memcpy(value_ptr(rn, 0), value_ptr(pn, nr_left),
669 	       nr_right * size);
670 
671 	/* new_parent should just point to l and r now */
672 	pn->header.flags = cpu_to_le32(INTERNAL_NODE);
673 	pn->header.nr_entries = cpu_to_le32(2);
674 	pn->header.max_entries = cpu_to_le32(
675 		calc_max_entries(sizeof(__le64),
676 				 dm_bm_block_size(
677 					 dm_tm_get_bm(s->info->tm))));
678 	pn->header.value_size = cpu_to_le32(sizeof(__le64));
679 
680 	val = cpu_to_le64(dm_block_location(left));
681 	__dm_bless_for_disk(&val);
682 	pn->keys[0] = ln->keys[0];
683 	memcpy_disk(value_ptr(pn, 0), &val, sizeof(__le64));
684 
685 	val = cpu_to_le64(dm_block_location(right));
686 	__dm_bless_for_disk(&val);
687 	pn->keys[1] = rn->keys[0];
688 	memcpy_disk(value_ptr(pn, 1), &val, sizeof(__le64));
689 
690 	unlock_block(s->info, left);
691 	unlock_block(s->info, right);
692 	return 0;
693 }
694 
btree_insert_raw(struct shadow_spine * s,dm_block_t root,struct dm_btree_value_type * vt,uint64_t key,unsigned * index)695 static int btree_insert_raw(struct shadow_spine *s, dm_block_t root,
696 			    struct dm_btree_value_type *vt,
697 			    uint64_t key, unsigned *index)
698 {
699 	int r, i = *index, top = 1;
700 	struct btree_node *node;
701 
702 	for (;;) {
703 		r = shadow_step(s, root, vt);
704 		if (r < 0)
705 			return r;
706 
707 		node = dm_block_data(shadow_current(s));
708 
709 		/*
710 		 * We have to patch up the parent node, ugly, but I don't
711 		 * see a way to do this automatically as part of the spine
712 		 * op.
713 		 */
714 		if (shadow_has_parent(s) && i >= 0) { /* FIXME: second clause unness. */
715 			__le64 location = cpu_to_le64(dm_block_location(shadow_current(s)));
716 
717 			__dm_bless_for_disk(&location);
718 			memcpy_disk(value_ptr(dm_block_data(shadow_parent(s)), i),
719 				    &location, sizeof(__le64));
720 		}
721 
722 		node = dm_block_data(shadow_current(s));
723 
724 		if (node->header.nr_entries == node->header.max_entries) {
725 			if (top)
726 				r = btree_split_beneath(s, key);
727 			else
728 				r = btree_split_sibling(s, i, key);
729 
730 			if (r < 0)
731 				return r;
732 		}
733 
734 		node = dm_block_data(shadow_current(s));
735 
736 		i = lower_bound(node, key);
737 
738 		if (le32_to_cpu(node->header.flags) & LEAF_NODE)
739 			break;
740 
741 		if (i < 0) {
742 			/* change the bounds on the lowest key */
743 			node->keys[0] = cpu_to_le64(key);
744 			i = 0;
745 		}
746 
747 		root = value64(node, i);
748 		top = 0;
749 	}
750 
751 	if (i < 0 || le64_to_cpu(node->keys[i]) != key)
752 		i++;
753 
754 	*index = i;
755 	return 0;
756 }
757 
need_insert(struct btree_node * node,uint64_t * keys,unsigned level,unsigned index)758 static bool need_insert(struct btree_node *node, uint64_t *keys,
759 			unsigned level, unsigned index)
760 {
761         return ((index >= le32_to_cpu(node->header.nr_entries)) ||
762 		(le64_to_cpu(node->keys[index]) != keys[level]));
763 }
764 
insert(struct dm_btree_info * info,dm_block_t root,uint64_t * keys,void * value,dm_block_t * new_root,int * inserted)765 static int insert(struct dm_btree_info *info, dm_block_t root,
766 		  uint64_t *keys, void *value, dm_block_t *new_root,
767 		  int *inserted)
768 		  __dm_written_to_disk(value)
769 {
770 	int r;
771 	unsigned level, index = -1, last_level = info->levels - 1;
772 	dm_block_t block = root;
773 	struct shadow_spine spine;
774 	struct btree_node *n;
775 	struct dm_btree_value_type le64_type;
776 
777 	init_le64_type(info->tm, &le64_type);
778 	init_shadow_spine(&spine, info);
779 
780 	for (level = 0; level < (info->levels - 1); level++) {
781 		r = btree_insert_raw(&spine, block, &le64_type, keys[level], &index);
782 		if (r < 0)
783 			goto bad;
784 
785 		n = dm_block_data(shadow_current(&spine));
786 
787 		if (need_insert(n, keys, level, index)) {
788 			dm_block_t new_tree;
789 			__le64 new_le;
790 
791 			r = dm_btree_empty(info, &new_tree);
792 			if (r < 0)
793 				goto bad;
794 
795 			new_le = cpu_to_le64(new_tree);
796 			__dm_bless_for_disk(&new_le);
797 
798 			r = insert_at(sizeof(uint64_t), n, index,
799 				      keys[level], &new_le);
800 			if (r)
801 				goto bad;
802 		}
803 
804 		if (level < last_level)
805 			block = value64(n, index);
806 	}
807 
808 	r = btree_insert_raw(&spine, block, &info->value_type,
809 			     keys[level], &index);
810 	if (r < 0)
811 		goto bad;
812 
813 	n = dm_block_data(shadow_current(&spine));
814 
815 	if (need_insert(n, keys, level, index)) {
816 		if (inserted)
817 			*inserted = 1;
818 
819 		r = insert_at(info->value_type.size, n, index,
820 			      keys[level], value);
821 		if (r)
822 			goto bad_unblessed;
823 	} else {
824 		if (inserted)
825 			*inserted = 0;
826 
827 		if (info->value_type.dec &&
828 		    (!info->value_type.equal ||
829 		     !info->value_type.equal(
830 			     info->value_type.context,
831 			     value_ptr(n, index),
832 			     value))) {
833 			info->value_type.dec(info->value_type.context,
834 					     value_ptr(n, index));
835 		}
836 		memcpy_disk(value_ptr(n, index),
837 			    value, info->value_type.size);
838 	}
839 
840 	*new_root = shadow_root(&spine);
841 	exit_shadow_spine(&spine);
842 
843 	return 0;
844 
845 bad:
846 	__dm_unbless_for_disk(value);
847 bad_unblessed:
848 	exit_shadow_spine(&spine);
849 	return r;
850 }
851 
dm_btree_insert(struct dm_btree_info * info,dm_block_t root,uint64_t * keys,void * value,dm_block_t * new_root)852 int dm_btree_insert(struct dm_btree_info *info, dm_block_t root,
853 		    uint64_t *keys, void *value, dm_block_t *new_root)
854 		    __dm_written_to_disk(value)
855 {
856 	return insert(info, root, keys, value, new_root, NULL);
857 }
858 EXPORT_SYMBOL_GPL(dm_btree_insert);
859 
dm_btree_insert_notify(struct dm_btree_info * info,dm_block_t root,uint64_t * keys,void * value,dm_block_t * new_root,int * inserted)860 int dm_btree_insert_notify(struct dm_btree_info *info, dm_block_t root,
861 			   uint64_t *keys, void *value, dm_block_t *new_root,
862 			   int *inserted)
863 			   __dm_written_to_disk(value)
864 {
865 	return insert(info, root, keys, value, new_root, inserted);
866 }
867 EXPORT_SYMBOL_GPL(dm_btree_insert_notify);
868 
869 /*----------------------------------------------------------------*/
870 
find_key(struct ro_spine * s,dm_block_t block,bool find_highest,uint64_t * result_key,dm_block_t * next_block)871 static int find_key(struct ro_spine *s, dm_block_t block, bool find_highest,
872 		    uint64_t *result_key, dm_block_t *next_block)
873 {
874 	int i, r;
875 	uint32_t flags;
876 
877 	do {
878 		r = ro_step(s, block);
879 		if (r < 0)
880 			return r;
881 
882 		flags = le32_to_cpu(ro_node(s)->header.flags);
883 		i = le32_to_cpu(ro_node(s)->header.nr_entries);
884 		if (!i)
885 			return -ENODATA;
886 		else
887 			i--;
888 
889 		if (find_highest)
890 			*result_key = le64_to_cpu(ro_node(s)->keys[i]);
891 		else
892 			*result_key = le64_to_cpu(ro_node(s)->keys[0]);
893 
894 		if (next_block || flags & INTERNAL_NODE) {
895 			if (find_highest)
896 				block = value64(ro_node(s), i);
897 			else
898 				block = value64(ro_node(s), 0);
899 		}
900 
901 	} while (flags & INTERNAL_NODE);
902 
903 	if (next_block)
904 		*next_block = block;
905 	return 0;
906 }
907 
dm_btree_find_key(struct dm_btree_info * info,dm_block_t root,bool find_highest,uint64_t * result_keys)908 static int dm_btree_find_key(struct dm_btree_info *info, dm_block_t root,
909 			     bool find_highest, uint64_t *result_keys)
910 {
911 	int r = 0, count = 0, level;
912 	struct ro_spine spine;
913 
914 	init_ro_spine(&spine, info);
915 	for (level = 0; level < info->levels; level++) {
916 		r = find_key(&spine, root, find_highest, result_keys + level,
917 			     level == info->levels - 1 ? NULL : &root);
918 		if (r == -ENODATA) {
919 			r = 0;
920 			break;
921 
922 		} else if (r)
923 			break;
924 
925 		count++;
926 	}
927 	exit_ro_spine(&spine);
928 
929 	return r ? r : count;
930 }
931 
dm_btree_find_highest_key(struct dm_btree_info * info,dm_block_t root,uint64_t * result_keys)932 int dm_btree_find_highest_key(struct dm_btree_info *info, dm_block_t root,
933 			      uint64_t *result_keys)
934 {
935 	return dm_btree_find_key(info, root, true, result_keys);
936 }
937 EXPORT_SYMBOL_GPL(dm_btree_find_highest_key);
938 
dm_btree_find_lowest_key(struct dm_btree_info * info,dm_block_t root,uint64_t * result_keys)939 int dm_btree_find_lowest_key(struct dm_btree_info *info, dm_block_t root,
940 			     uint64_t *result_keys)
941 {
942 	return dm_btree_find_key(info, root, false, result_keys);
943 }
944 EXPORT_SYMBOL_GPL(dm_btree_find_lowest_key);
945 
946 /*----------------------------------------------------------------*/
947 
948 /*
949  * FIXME: We shouldn't use a recursive algorithm when we have limited stack
950  * space.  Also this only works for single level trees.
951  */
walk_node(struct dm_btree_info * info,dm_block_t block,int (* fn)(void * context,uint64_t * keys,void * leaf),void * context)952 static int walk_node(struct dm_btree_info *info, dm_block_t block,
953 		     int (*fn)(void *context, uint64_t *keys, void *leaf),
954 		     void *context)
955 {
956 	int r;
957 	unsigned i, nr;
958 	struct dm_block *node;
959 	struct btree_node *n;
960 	uint64_t keys;
961 
962 	r = bn_read_lock(info, block, &node);
963 	if (r)
964 		return r;
965 
966 	n = dm_block_data(node);
967 
968 	nr = le32_to_cpu(n->header.nr_entries);
969 	for (i = 0; i < nr; i++) {
970 		if (le32_to_cpu(n->header.flags) & INTERNAL_NODE) {
971 			r = walk_node(info, value64(n, i), fn, context);
972 			if (r)
973 				goto out;
974 		} else {
975 			keys = le64_to_cpu(*key_ptr(n, i));
976 			r = fn(context, &keys, value_ptr(n, i));
977 			if (r)
978 				goto out;
979 		}
980 	}
981 
982 out:
983 	dm_tm_unlock(info->tm, node);
984 	return r;
985 }
986 
dm_btree_walk(struct dm_btree_info * info,dm_block_t root,int (* fn)(void * context,uint64_t * keys,void * leaf),void * context)987 int dm_btree_walk(struct dm_btree_info *info, dm_block_t root,
988 		  int (*fn)(void *context, uint64_t *keys, void *leaf),
989 		  void *context)
990 {
991 	BUG_ON(info->levels > 1);
992 	return walk_node(info, root, fn, context);
993 }
994 EXPORT_SYMBOL_GPL(dm_btree_walk);
995 
996 /*----------------------------------------------------------------*/
997 
prefetch_values(struct dm_btree_cursor * c)998 static void prefetch_values(struct dm_btree_cursor *c)
999 {
1000 	unsigned i, nr;
1001 	__le64 value_le;
1002 	struct cursor_node *n = c->nodes + c->depth - 1;
1003 	struct btree_node *bn = dm_block_data(n->b);
1004 	struct dm_block_manager *bm = dm_tm_get_bm(c->info->tm);
1005 
1006 	BUG_ON(c->info->value_type.size != sizeof(value_le));
1007 
1008 	nr = le32_to_cpu(bn->header.nr_entries);
1009 	for (i = 0; i < nr; i++) {
1010 		memcpy(&value_le, value_ptr(bn, i), sizeof(value_le));
1011 		dm_bm_prefetch(bm, le64_to_cpu(value_le));
1012 	}
1013 }
1014 
leaf_node(struct dm_btree_cursor * c)1015 static bool leaf_node(struct dm_btree_cursor *c)
1016 {
1017 	struct cursor_node *n = c->nodes + c->depth - 1;
1018 	struct btree_node *bn = dm_block_data(n->b);
1019 
1020 	return le32_to_cpu(bn->header.flags) & LEAF_NODE;
1021 }
1022 
push_node(struct dm_btree_cursor * c,dm_block_t b)1023 static int push_node(struct dm_btree_cursor *c, dm_block_t b)
1024 {
1025 	int r;
1026 	struct cursor_node *n = c->nodes + c->depth;
1027 
1028 	if (c->depth >= DM_BTREE_CURSOR_MAX_DEPTH - 1) {
1029 		DMERR("couldn't push cursor node, stack depth too high");
1030 		return -EINVAL;
1031 	}
1032 
1033 	r = bn_read_lock(c->info, b, &n->b);
1034 	if (r)
1035 		return r;
1036 
1037 	n->index = 0;
1038 	c->depth++;
1039 
1040 	if (c->prefetch_leaves || !leaf_node(c))
1041 		prefetch_values(c);
1042 
1043 	return 0;
1044 }
1045 
pop_node(struct dm_btree_cursor * c)1046 static void pop_node(struct dm_btree_cursor *c)
1047 {
1048 	c->depth--;
1049 	unlock_block(c->info, c->nodes[c->depth].b);
1050 }
1051 
inc_or_backtrack(struct dm_btree_cursor * c)1052 static int inc_or_backtrack(struct dm_btree_cursor *c)
1053 {
1054 	struct cursor_node *n;
1055 	struct btree_node *bn;
1056 
1057 	for (;;) {
1058 		if (!c->depth)
1059 			return -ENODATA;
1060 
1061 		n = c->nodes + c->depth - 1;
1062 		bn = dm_block_data(n->b);
1063 
1064 		n->index++;
1065 		if (n->index < le32_to_cpu(bn->header.nr_entries))
1066 			break;
1067 
1068 		pop_node(c);
1069 	}
1070 
1071 	return 0;
1072 }
1073 
find_leaf(struct dm_btree_cursor * c)1074 static int find_leaf(struct dm_btree_cursor *c)
1075 {
1076 	int r = 0;
1077 	struct cursor_node *n;
1078 	struct btree_node *bn;
1079 	__le64 value_le;
1080 
1081 	for (;;) {
1082 		n = c->nodes + c->depth - 1;
1083 		bn = dm_block_data(n->b);
1084 
1085 		if (le32_to_cpu(bn->header.flags) & LEAF_NODE)
1086 			break;
1087 
1088 		memcpy(&value_le, value_ptr(bn, n->index), sizeof(value_le));
1089 		r = push_node(c, le64_to_cpu(value_le));
1090 		if (r) {
1091 			DMERR("push_node failed");
1092 			break;
1093 		}
1094 	}
1095 
1096 	if (!r && (le32_to_cpu(bn->header.nr_entries) == 0))
1097 		return -ENODATA;
1098 
1099 	return r;
1100 }
1101 
dm_btree_cursor_begin(struct dm_btree_info * info,dm_block_t root,bool prefetch_leaves,struct dm_btree_cursor * c)1102 int dm_btree_cursor_begin(struct dm_btree_info *info, dm_block_t root,
1103 			  bool prefetch_leaves, struct dm_btree_cursor *c)
1104 {
1105 	int r;
1106 
1107 	c->info = info;
1108 	c->root = root;
1109 	c->depth = 0;
1110 	c->prefetch_leaves = prefetch_leaves;
1111 
1112 	r = push_node(c, root);
1113 	if (r)
1114 		return r;
1115 
1116 	return find_leaf(c);
1117 }
1118 EXPORT_SYMBOL_GPL(dm_btree_cursor_begin);
1119 
dm_btree_cursor_end(struct dm_btree_cursor * c)1120 void dm_btree_cursor_end(struct dm_btree_cursor *c)
1121 {
1122 	while (c->depth)
1123 		pop_node(c);
1124 }
1125 EXPORT_SYMBOL_GPL(dm_btree_cursor_end);
1126 
dm_btree_cursor_next(struct dm_btree_cursor * c)1127 int dm_btree_cursor_next(struct dm_btree_cursor *c)
1128 {
1129 	int r = inc_or_backtrack(c);
1130 	if (!r) {
1131 		r = find_leaf(c);
1132 		if (r)
1133 			DMERR("find_leaf failed");
1134 	}
1135 
1136 	return r;
1137 }
1138 EXPORT_SYMBOL_GPL(dm_btree_cursor_next);
1139 
dm_btree_cursor_skip(struct dm_btree_cursor * c,uint32_t count)1140 int dm_btree_cursor_skip(struct dm_btree_cursor *c, uint32_t count)
1141 {
1142 	int r = 0;
1143 
1144 	while (count-- && !r)
1145 		r = dm_btree_cursor_next(c);
1146 
1147 	return r;
1148 }
1149 EXPORT_SYMBOL_GPL(dm_btree_cursor_skip);
1150 
dm_btree_cursor_get_value(struct dm_btree_cursor * c,uint64_t * key,void * value_le)1151 int dm_btree_cursor_get_value(struct dm_btree_cursor *c, uint64_t *key, void *value_le)
1152 {
1153 	if (c->depth) {
1154 		struct cursor_node *n = c->nodes + c->depth - 1;
1155 		struct btree_node *bn = dm_block_data(n->b);
1156 
1157 		if (le32_to_cpu(bn->header.flags) & INTERNAL_NODE)
1158 			return -EINVAL;
1159 
1160 		*key = le64_to_cpu(*key_ptr(bn, n->index));
1161 		memcpy(value_le, value_ptr(bn, n->index), c->info->value_type.size);
1162 		return 0;
1163 
1164 	} else
1165 		return -ENODATA;
1166 }
1167 EXPORT_SYMBOL_GPL(dm_btree_cursor_get_value);
1168