xref: /openbmc/linux/fs/btrfs/ref-verify.c (revision adb19164)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2014 Facebook.  All rights reserved.
4  */
5 
6 #include <linux/sched.h>
7 #include <linux/stacktrace.h>
8 #include "messages.h"
9 #include "ctree.h"
10 #include "disk-io.h"
11 #include "locking.h"
12 #include "delayed-ref.h"
13 #include "ref-verify.h"
14 #include "fs.h"
15 #include "accessors.h"
16 
17 /*
18  * Used to keep track the roots and number of refs each root has for a given
19  * bytenr.  This just tracks the number of direct references, no shared
20  * references.
21  */
22 struct root_entry {
23 	u64 root_objectid;
24 	u64 num_refs;
25 	struct rb_node node;
26 };
27 
28 /*
29  * These are meant to represent what should exist in the extent tree, these can
30  * be used to verify the extent tree is consistent as these should all match
31  * what the extent tree says.
32  */
33 struct ref_entry {
34 	u64 root_objectid;
35 	u64 parent;
36 	u64 owner;
37 	u64 offset;
38 	u64 num_refs;
39 	struct rb_node node;
40 };
41 
42 #define MAX_TRACE	16
43 
44 /*
45  * Whenever we add/remove a reference we record the action.  The action maps
46  * back to the delayed ref action.  We hold the ref we are changing in the
47  * action so we can account for the history properly, and we record the root we
48  * were called with since it could be different from ref_root.  We also store
49  * stack traces because that's how I roll.
50  */
51 struct ref_action {
52 	int action;
53 	u64 root;
54 	struct ref_entry ref;
55 	struct list_head list;
56 	unsigned long trace[MAX_TRACE];
57 	unsigned int trace_len;
58 };
59 
60 /*
61  * One of these for every block we reference, it holds the roots and references
62  * to it as well as all of the ref actions that have occurred to it.  We never
63  * free it until we unmount the file system in order to make sure re-allocations
64  * are happening properly.
65  */
66 struct block_entry {
67 	u64 bytenr;
68 	u64 len;
69 	u64 num_refs;
70 	int metadata;
71 	int from_disk;
72 	struct rb_root roots;
73 	struct rb_root refs;
74 	struct rb_node node;
75 	struct list_head actions;
76 };
77 
78 static struct block_entry *insert_block_entry(struct rb_root *root,
79 					      struct block_entry *be)
80 {
81 	struct rb_node **p = &root->rb_node;
82 	struct rb_node *parent_node = NULL;
83 	struct block_entry *entry;
84 
85 	while (*p) {
86 		parent_node = *p;
87 		entry = rb_entry(parent_node, struct block_entry, node);
88 		if (entry->bytenr > be->bytenr)
89 			p = &(*p)->rb_left;
90 		else if (entry->bytenr < be->bytenr)
91 			p = &(*p)->rb_right;
92 		else
93 			return entry;
94 	}
95 
96 	rb_link_node(&be->node, parent_node, p);
97 	rb_insert_color(&be->node, root);
98 	return NULL;
99 }
100 
101 static struct block_entry *lookup_block_entry(struct rb_root *root, u64 bytenr)
102 {
103 	struct rb_node *n;
104 	struct block_entry *entry = NULL;
105 
106 	n = root->rb_node;
107 	while (n) {
108 		entry = rb_entry(n, struct block_entry, node);
109 		if (entry->bytenr < bytenr)
110 			n = n->rb_right;
111 		else if (entry->bytenr > bytenr)
112 			n = n->rb_left;
113 		else
114 			return entry;
115 	}
116 	return NULL;
117 }
118 
119 static struct root_entry *insert_root_entry(struct rb_root *root,
120 					    struct root_entry *re)
121 {
122 	struct rb_node **p = &root->rb_node;
123 	struct rb_node *parent_node = NULL;
124 	struct root_entry *entry;
125 
126 	while (*p) {
127 		parent_node = *p;
128 		entry = rb_entry(parent_node, struct root_entry, node);
129 		if (entry->root_objectid > re->root_objectid)
130 			p = &(*p)->rb_left;
131 		else if (entry->root_objectid < re->root_objectid)
132 			p = &(*p)->rb_right;
133 		else
134 			return entry;
135 	}
136 
137 	rb_link_node(&re->node, parent_node, p);
138 	rb_insert_color(&re->node, root);
139 	return NULL;
140 
141 }
142 
143 static int comp_refs(struct ref_entry *ref1, struct ref_entry *ref2)
144 {
145 	if (ref1->root_objectid < ref2->root_objectid)
146 		return -1;
147 	if (ref1->root_objectid > ref2->root_objectid)
148 		return 1;
149 	if (ref1->parent < ref2->parent)
150 		return -1;
151 	if (ref1->parent > ref2->parent)
152 		return 1;
153 	if (ref1->owner < ref2->owner)
154 		return -1;
155 	if (ref1->owner > ref2->owner)
156 		return 1;
157 	if (ref1->offset < ref2->offset)
158 		return -1;
159 	if (ref1->offset > ref2->offset)
160 		return 1;
161 	return 0;
162 }
163 
164 static struct ref_entry *insert_ref_entry(struct rb_root *root,
165 					  struct ref_entry *ref)
166 {
167 	struct rb_node **p = &root->rb_node;
168 	struct rb_node *parent_node = NULL;
169 	struct ref_entry *entry;
170 	int cmp;
171 
172 	while (*p) {
173 		parent_node = *p;
174 		entry = rb_entry(parent_node, struct ref_entry, node);
175 		cmp = comp_refs(entry, ref);
176 		if (cmp > 0)
177 			p = &(*p)->rb_left;
178 		else if (cmp < 0)
179 			p = &(*p)->rb_right;
180 		else
181 			return entry;
182 	}
183 
184 	rb_link_node(&ref->node, parent_node, p);
185 	rb_insert_color(&ref->node, root);
186 	return NULL;
187 
188 }
189 
190 static struct root_entry *lookup_root_entry(struct rb_root *root, u64 objectid)
191 {
192 	struct rb_node *n;
193 	struct root_entry *entry = NULL;
194 
195 	n = root->rb_node;
196 	while (n) {
197 		entry = rb_entry(n, struct root_entry, node);
198 		if (entry->root_objectid < objectid)
199 			n = n->rb_right;
200 		else if (entry->root_objectid > objectid)
201 			n = n->rb_left;
202 		else
203 			return entry;
204 	}
205 	return NULL;
206 }
207 
208 #ifdef CONFIG_STACKTRACE
209 static void __save_stack_trace(struct ref_action *ra)
210 {
211 	ra->trace_len = stack_trace_save(ra->trace, MAX_TRACE, 2);
212 }
213 
214 static void __print_stack_trace(struct btrfs_fs_info *fs_info,
215 				struct ref_action *ra)
216 {
217 	if (ra->trace_len == 0) {
218 		btrfs_err(fs_info, "  ref-verify: no stacktrace");
219 		return;
220 	}
221 	stack_trace_print(ra->trace, ra->trace_len, 2);
222 }
223 #else
224 static inline void __save_stack_trace(struct ref_action *ra)
225 {
226 }
227 
228 static inline void __print_stack_trace(struct btrfs_fs_info *fs_info,
229 				       struct ref_action *ra)
230 {
231 	btrfs_err(fs_info, "  ref-verify: no stacktrace support");
232 }
233 #endif
234 
235 static void free_block_entry(struct block_entry *be)
236 {
237 	struct root_entry *re;
238 	struct ref_entry *ref;
239 	struct ref_action *ra;
240 	struct rb_node *n;
241 
242 	while ((n = rb_first(&be->roots))) {
243 		re = rb_entry(n, struct root_entry, node);
244 		rb_erase(&re->node, &be->roots);
245 		kfree(re);
246 	}
247 
248 	while((n = rb_first(&be->refs))) {
249 		ref = rb_entry(n, struct ref_entry, node);
250 		rb_erase(&ref->node, &be->refs);
251 		kfree(ref);
252 	}
253 
254 	while (!list_empty(&be->actions)) {
255 		ra = list_first_entry(&be->actions, struct ref_action,
256 				      list);
257 		list_del(&ra->list);
258 		kfree(ra);
259 	}
260 	kfree(be);
261 }
262 
263 static struct block_entry *add_block_entry(struct btrfs_fs_info *fs_info,
264 					   u64 bytenr, u64 len,
265 					   u64 root_objectid)
266 {
267 	struct block_entry *be = NULL, *exist;
268 	struct root_entry *re = NULL;
269 
270 	re = kzalloc(sizeof(struct root_entry), GFP_NOFS);
271 	be = kzalloc(sizeof(struct block_entry), GFP_NOFS);
272 	if (!be || !re) {
273 		kfree(re);
274 		kfree(be);
275 		return ERR_PTR(-ENOMEM);
276 	}
277 	be->bytenr = bytenr;
278 	be->len = len;
279 
280 	re->root_objectid = root_objectid;
281 	re->num_refs = 0;
282 
283 	spin_lock(&fs_info->ref_verify_lock);
284 	exist = insert_block_entry(&fs_info->block_tree, be);
285 	if (exist) {
286 		if (root_objectid) {
287 			struct root_entry *exist_re;
288 
289 			exist_re = insert_root_entry(&exist->roots, re);
290 			if (exist_re)
291 				kfree(re);
292 		} else {
293 			kfree(re);
294 		}
295 		kfree(be);
296 		return exist;
297 	}
298 
299 	be->num_refs = 0;
300 	be->metadata = 0;
301 	be->from_disk = 0;
302 	be->roots = RB_ROOT;
303 	be->refs = RB_ROOT;
304 	INIT_LIST_HEAD(&be->actions);
305 	if (root_objectid)
306 		insert_root_entry(&be->roots, re);
307 	else
308 		kfree(re);
309 	return be;
310 }
311 
312 static int add_tree_block(struct btrfs_fs_info *fs_info, u64 ref_root,
313 			  u64 parent, u64 bytenr, int level)
314 {
315 	struct block_entry *be;
316 	struct root_entry *re;
317 	struct ref_entry *ref = NULL, *exist;
318 
319 	ref = kmalloc(sizeof(struct ref_entry), GFP_NOFS);
320 	if (!ref)
321 		return -ENOMEM;
322 
323 	if (parent)
324 		ref->root_objectid = 0;
325 	else
326 		ref->root_objectid = ref_root;
327 	ref->parent = parent;
328 	ref->owner = level;
329 	ref->offset = 0;
330 	ref->num_refs = 1;
331 
332 	be = add_block_entry(fs_info, bytenr, fs_info->nodesize, ref_root);
333 	if (IS_ERR(be)) {
334 		kfree(ref);
335 		return PTR_ERR(be);
336 	}
337 	be->num_refs++;
338 	be->from_disk = 1;
339 	be->metadata = 1;
340 
341 	if (!parent) {
342 		ASSERT(ref_root);
343 		re = lookup_root_entry(&be->roots, ref_root);
344 		ASSERT(re);
345 		re->num_refs++;
346 	}
347 	exist = insert_ref_entry(&be->refs, ref);
348 	if (exist) {
349 		exist->num_refs++;
350 		kfree(ref);
351 	}
352 	spin_unlock(&fs_info->ref_verify_lock);
353 
354 	return 0;
355 }
356 
357 static int add_shared_data_ref(struct btrfs_fs_info *fs_info,
358 			       u64 parent, u32 num_refs, u64 bytenr,
359 			       u64 num_bytes)
360 {
361 	struct block_entry *be;
362 	struct ref_entry *ref;
363 
364 	ref = kzalloc(sizeof(struct ref_entry), GFP_NOFS);
365 	if (!ref)
366 		return -ENOMEM;
367 	be = add_block_entry(fs_info, bytenr, num_bytes, 0);
368 	if (IS_ERR(be)) {
369 		kfree(ref);
370 		return PTR_ERR(be);
371 	}
372 	be->num_refs += num_refs;
373 
374 	ref->parent = parent;
375 	ref->num_refs = num_refs;
376 	if (insert_ref_entry(&be->refs, ref)) {
377 		spin_unlock(&fs_info->ref_verify_lock);
378 		btrfs_err(fs_info, "existing shared ref when reading from disk?");
379 		kfree(ref);
380 		return -EINVAL;
381 	}
382 	spin_unlock(&fs_info->ref_verify_lock);
383 	return 0;
384 }
385 
386 static int add_extent_data_ref(struct btrfs_fs_info *fs_info,
387 			       struct extent_buffer *leaf,
388 			       struct btrfs_extent_data_ref *dref,
389 			       u64 bytenr, u64 num_bytes)
390 {
391 	struct block_entry *be;
392 	struct ref_entry *ref;
393 	struct root_entry *re;
394 	u64 ref_root = btrfs_extent_data_ref_root(leaf, dref);
395 	u64 owner = btrfs_extent_data_ref_objectid(leaf, dref);
396 	u64 offset = btrfs_extent_data_ref_offset(leaf, dref);
397 	u32 num_refs = btrfs_extent_data_ref_count(leaf, dref);
398 
399 	ref = kzalloc(sizeof(struct ref_entry), GFP_NOFS);
400 	if (!ref)
401 		return -ENOMEM;
402 	be = add_block_entry(fs_info, bytenr, num_bytes, ref_root);
403 	if (IS_ERR(be)) {
404 		kfree(ref);
405 		return PTR_ERR(be);
406 	}
407 	be->num_refs += num_refs;
408 
409 	ref->parent = 0;
410 	ref->owner = owner;
411 	ref->root_objectid = ref_root;
412 	ref->offset = offset;
413 	ref->num_refs = num_refs;
414 	if (insert_ref_entry(&be->refs, ref)) {
415 		spin_unlock(&fs_info->ref_verify_lock);
416 		btrfs_err(fs_info, "existing ref when reading from disk?");
417 		kfree(ref);
418 		return -EINVAL;
419 	}
420 
421 	re = lookup_root_entry(&be->roots, ref_root);
422 	if (!re) {
423 		spin_unlock(&fs_info->ref_verify_lock);
424 		btrfs_err(fs_info, "missing root in new block entry?");
425 		return -EINVAL;
426 	}
427 	re->num_refs += num_refs;
428 	spin_unlock(&fs_info->ref_verify_lock);
429 	return 0;
430 }
431 
432 static int process_extent_item(struct btrfs_fs_info *fs_info,
433 			       struct btrfs_path *path, struct btrfs_key *key,
434 			       int slot, int *tree_block_level)
435 {
436 	struct btrfs_extent_item *ei;
437 	struct btrfs_extent_inline_ref *iref;
438 	struct btrfs_extent_data_ref *dref;
439 	struct btrfs_shared_data_ref *sref;
440 	struct extent_buffer *leaf = path->nodes[0];
441 	u32 item_size = btrfs_item_size(leaf, slot);
442 	unsigned long end, ptr;
443 	u64 offset, flags, count;
444 	int type, ret;
445 
446 	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
447 	flags = btrfs_extent_flags(leaf, ei);
448 
449 	if ((key->type == BTRFS_EXTENT_ITEM_KEY) &&
450 	    flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
451 		struct btrfs_tree_block_info *info;
452 
453 		info = (struct btrfs_tree_block_info *)(ei + 1);
454 		*tree_block_level = btrfs_tree_block_level(leaf, info);
455 		iref = (struct btrfs_extent_inline_ref *)(info + 1);
456 	} else {
457 		if (key->type == BTRFS_METADATA_ITEM_KEY)
458 			*tree_block_level = key->offset;
459 		iref = (struct btrfs_extent_inline_ref *)(ei + 1);
460 	}
461 
462 	ptr = (unsigned long)iref;
463 	end = (unsigned long)ei + item_size;
464 	while (ptr < end) {
465 		iref = (struct btrfs_extent_inline_ref *)ptr;
466 		type = btrfs_extent_inline_ref_type(leaf, iref);
467 		offset = btrfs_extent_inline_ref_offset(leaf, iref);
468 		switch (type) {
469 		case BTRFS_TREE_BLOCK_REF_KEY:
470 			ret = add_tree_block(fs_info, offset, 0, key->objectid,
471 					     *tree_block_level);
472 			break;
473 		case BTRFS_SHARED_BLOCK_REF_KEY:
474 			ret = add_tree_block(fs_info, 0, offset, key->objectid,
475 					     *tree_block_level);
476 			break;
477 		case BTRFS_EXTENT_DATA_REF_KEY:
478 			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
479 			ret = add_extent_data_ref(fs_info, leaf, dref,
480 						  key->objectid, key->offset);
481 			break;
482 		case BTRFS_SHARED_DATA_REF_KEY:
483 			sref = (struct btrfs_shared_data_ref *)(iref + 1);
484 			count = btrfs_shared_data_ref_count(leaf, sref);
485 			ret = add_shared_data_ref(fs_info, offset, count,
486 						  key->objectid, key->offset);
487 			break;
488 		default:
489 			btrfs_err(fs_info, "invalid key type in iref");
490 			ret = -EINVAL;
491 			break;
492 		}
493 		if (ret)
494 			break;
495 		ptr += btrfs_extent_inline_ref_size(type);
496 	}
497 	return ret;
498 }
499 
500 static int process_leaf(struct btrfs_root *root,
501 			struct btrfs_path *path, u64 *bytenr, u64 *num_bytes,
502 			int *tree_block_level)
503 {
504 	struct btrfs_fs_info *fs_info = root->fs_info;
505 	struct extent_buffer *leaf = path->nodes[0];
506 	struct btrfs_extent_data_ref *dref;
507 	struct btrfs_shared_data_ref *sref;
508 	u32 count;
509 	int i = 0, ret = 0;
510 	struct btrfs_key key;
511 	int nritems = btrfs_header_nritems(leaf);
512 
513 	for (i = 0; i < nritems; i++) {
514 		btrfs_item_key_to_cpu(leaf, &key, i);
515 		switch (key.type) {
516 		case BTRFS_EXTENT_ITEM_KEY:
517 			*num_bytes = key.offset;
518 			fallthrough;
519 		case BTRFS_METADATA_ITEM_KEY:
520 			*bytenr = key.objectid;
521 			ret = process_extent_item(fs_info, path, &key, i,
522 						  tree_block_level);
523 			break;
524 		case BTRFS_TREE_BLOCK_REF_KEY:
525 			ret = add_tree_block(fs_info, key.offset, 0,
526 					     key.objectid, *tree_block_level);
527 			break;
528 		case BTRFS_SHARED_BLOCK_REF_KEY:
529 			ret = add_tree_block(fs_info, 0, key.offset,
530 					     key.objectid, *tree_block_level);
531 			break;
532 		case BTRFS_EXTENT_DATA_REF_KEY:
533 			dref = btrfs_item_ptr(leaf, i,
534 					      struct btrfs_extent_data_ref);
535 			ret = add_extent_data_ref(fs_info, leaf, dref, *bytenr,
536 						  *num_bytes);
537 			break;
538 		case BTRFS_SHARED_DATA_REF_KEY:
539 			sref = btrfs_item_ptr(leaf, i,
540 					      struct btrfs_shared_data_ref);
541 			count = btrfs_shared_data_ref_count(leaf, sref);
542 			ret = add_shared_data_ref(fs_info, key.offset, count,
543 						  *bytenr, *num_bytes);
544 			break;
545 		default:
546 			break;
547 		}
548 		if (ret)
549 			break;
550 	}
551 	return ret;
552 }
553 
554 /* Walk down to the leaf from the given level */
555 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
556 			  int level, u64 *bytenr, u64 *num_bytes,
557 			  int *tree_block_level)
558 {
559 	struct extent_buffer *eb;
560 	int ret = 0;
561 
562 	while (level >= 0) {
563 		if (level) {
564 			eb = btrfs_read_node_slot(path->nodes[level],
565 						  path->slots[level]);
566 			if (IS_ERR(eb))
567 				return PTR_ERR(eb);
568 			btrfs_tree_read_lock(eb);
569 			path->nodes[level-1] = eb;
570 			path->slots[level-1] = 0;
571 			path->locks[level-1] = BTRFS_READ_LOCK;
572 		} else {
573 			ret = process_leaf(root, path, bytenr, num_bytes,
574 					   tree_block_level);
575 			if (ret)
576 				break;
577 		}
578 		level--;
579 	}
580 	return ret;
581 }
582 
583 /* Walk up to the next node that needs to be processed */
584 static int walk_up_tree(struct btrfs_path *path, int *level)
585 {
586 	int l;
587 
588 	for (l = 0; l < BTRFS_MAX_LEVEL; l++) {
589 		if (!path->nodes[l])
590 			continue;
591 		if (l) {
592 			path->slots[l]++;
593 			if (path->slots[l] <
594 			    btrfs_header_nritems(path->nodes[l])) {
595 				*level = l;
596 				return 0;
597 			}
598 		}
599 		btrfs_tree_unlock_rw(path->nodes[l], path->locks[l]);
600 		free_extent_buffer(path->nodes[l]);
601 		path->nodes[l] = NULL;
602 		path->slots[l] = 0;
603 		path->locks[l] = 0;
604 	}
605 
606 	return 1;
607 }
608 
609 static void dump_ref_action(struct btrfs_fs_info *fs_info,
610 			    struct ref_action *ra)
611 {
612 	btrfs_err(fs_info,
613 "  Ref action %d, root %llu, ref_root %llu, parent %llu, owner %llu, offset %llu, num_refs %llu",
614 		  ra->action, ra->root, ra->ref.root_objectid, ra->ref.parent,
615 		  ra->ref.owner, ra->ref.offset, ra->ref.num_refs);
616 	__print_stack_trace(fs_info, ra);
617 }
618 
619 /*
620  * Dumps all the information from the block entry to printk, it's going to be
621  * awesome.
622  */
623 static void dump_block_entry(struct btrfs_fs_info *fs_info,
624 			     struct block_entry *be)
625 {
626 	struct ref_entry *ref;
627 	struct root_entry *re;
628 	struct ref_action *ra;
629 	struct rb_node *n;
630 
631 	btrfs_err(fs_info,
632 "dumping block entry [%llu %llu], num_refs %llu, metadata %d, from disk %d",
633 		  be->bytenr, be->len, be->num_refs, be->metadata,
634 		  be->from_disk);
635 
636 	for (n = rb_first(&be->refs); n; n = rb_next(n)) {
637 		ref = rb_entry(n, struct ref_entry, node);
638 		btrfs_err(fs_info,
639 "  ref root %llu, parent %llu, owner %llu, offset %llu, num_refs %llu",
640 			  ref->root_objectid, ref->parent, ref->owner,
641 			  ref->offset, ref->num_refs);
642 	}
643 
644 	for (n = rb_first(&be->roots); n; n = rb_next(n)) {
645 		re = rb_entry(n, struct root_entry, node);
646 		btrfs_err(fs_info, "  root entry %llu, num_refs %llu",
647 			  re->root_objectid, re->num_refs);
648 	}
649 
650 	list_for_each_entry(ra, &be->actions, list)
651 		dump_ref_action(fs_info, ra);
652 }
653 
654 /*
655  * btrfs_ref_tree_mod: called when we modify a ref for a bytenr
656  *
657  * This will add an action item to the given bytenr and do sanity checks to make
658  * sure we haven't messed something up.  If we are making a new allocation and
659  * this block entry has history we will delete all previous actions as long as
660  * our sanity checks pass as they are no longer needed.
661  */
662 int btrfs_ref_tree_mod(struct btrfs_fs_info *fs_info,
663 		       struct btrfs_ref *generic_ref)
664 {
665 	struct ref_entry *ref = NULL, *exist;
666 	struct ref_action *ra = NULL;
667 	struct block_entry *be = NULL;
668 	struct root_entry *re = NULL;
669 	int action = generic_ref->action;
670 	int ret = 0;
671 	bool metadata;
672 	u64 bytenr = generic_ref->bytenr;
673 	u64 num_bytes = generic_ref->len;
674 	u64 parent = generic_ref->parent;
675 	u64 ref_root = 0;
676 	u64 owner = 0;
677 	u64 offset = 0;
678 
679 	if (!btrfs_test_opt(fs_info, REF_VERIFY))
680 		return 0;
681 
682 	if (generic_ref->type == BTRFS_REF_METADATA) {
683 		if (!parent)
684 			ref_root = generic_ref->tree_ref.owning_root;
685 		owner = generic_ref->tree_ref.level;
686 	} else if (!parent) {
687 		ref_root = generic_ref->data_ref.owning_root;
688 		owner = generic_ref->data_ref.ino;
689 		offset = generic_ref->data_ref.offset;
690 	}
691 	metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
692 
693 	ref = kzalloc(sizeof(struct ref_entry), GFP_NOFS);
694 	ra = kmalloc(sizeof(struct ref_action), GFP_NOFS);
695 	if (!ra || !ref) {
696 		kfree(ref);
697 		kfree(ra);
698 		ret = -ENOMEM;
699 		goto out;
700 	}
701 
702 	ref->parent = parent;
703 	ref->owner = owner;
704 	ref->root_objectid = ref_root;
705 	ref->offset = offset;
706 	ref->num_refs = (action == BTRFS_DROP_DELAYED_REF) ? -1 : 1;
707 
708 	memcpy(&ra->ref, ref, sizeof(struct ref_entry));
709 	/*
710 	 * Save the extra info from the delayed ref in the ref action to make it
711 	 * easier to figure out what is happening.  The real ref's we add to the
712 	 * ref tree need to reflect what we save on disk so it matches any
713 	 * on-disk refs we pre-loaded.
714 	 */
715 	ra->ref.owner = owner;
716 	ra->ref.offset = offset;
717 	ra->ref.root_objectid = ref_root;
718 	__save_stack_trace(ra);
719 
720 	INIT_LIST_HEAD(&ra->list);
721 	ra->action = action;
722 	ra->root = generic_ref->real_root;
723 
724 	/*
725 	 * This is an allocation, preallocate the block_entry in case we haven't
726 	 * used it before.
727 	 */
728 	ret = -EINVAL;
729 	if (action == BTRFS_ADD_DELAYED_EXTENT) {
730 		/*
731 		 * For subvol_create we'll just pass in whatever the parent root
732 		 * is and the new root objectid, so let's not treat the passed
733 		 * in root as if it really has a ref for this bytenr.
734 		 */
735 		be = add_block_entry(fs_info, bytenr, num_bytes, ref_root);
736 		if (IS_ERR(be)) {
737 			kfree(ref);
738 			kfree(ra);
739 			ret = PTR_ERR(be);
740 			goto out;
741 		}
742 		be->num_refs++;
743 		if (metadata)
744 			be->metadata = 1;
745 
746 		if (be->num_refs != 1) {
747 			btrfs_err(fs_info,
748 			"re-allocated a block that still has references to it!");
749 			dump_block_entry(fs_info, be);
750 			dump_ref_action(fs_info, ra);
751 			kfree(ref);
752 			kfree(ra);
753 			goto out_unlock;
754 		}
755 
756 		while (!list_empty(&be->actions)) {
757 			struct ref_action *tmp;
758 
759 			tmp = list_first_entry(&be->actions, struct ref_action,
760 					       list);
761 			list_del(&tmp->list);
762 			kfree(tmp);
763 		}
764 	} else {
765 		struct root_entry *tmp;
766 
767 		if (!parent) {
768 			re = kmalloc(sizeof(struct root_entry), GFP_NOFS);
769 			if (!re) {
770 				kfree(ref);
771 				kfree(ra);
772 				ret = -ENOMEM;
773 				goto out;
774 			}
775 			/*
776 			 * This is the root that is modifying us, so it's the
777 			 * one we want to lookup below when we modify the
778 			 * re->num_refs.
779 			 */
780 			ref_root = generic_ref->real_root;
781 			re->root_objectid = generic_ref->real_root;
782 			re->num_refs = 0;
783 		}
784 
785 		spin_lock(&fs_info->ref_verify_lock);
786 		be = lookup_block_entry(&fs_info->block_tree, bytenr);
787 		if (!be) {
788 			btrfs_err(fs_info,
789 "trying to do action %d to bytenr %llu num_bytes %llu but there is no existing entry!",
790 				  action, bytenr, num_bytes);
791 			dump_ref_action(fs_info, ra);
792 			kfree(ref);
793 			kfree(ra);
794 			kfree(re);
795 			goto out_unlock;
796 		} else if (be->num_refs == 0) {
797 			btrfs_err(fs_info,
798 		"trying to do action %d for a bytenr that has 0 total references",
799 				action);
800 			dump_block_entry(fs_info, be);
801 			dump_ref_action(fs_info, ra);
802 			kfree(ref);
803 			kfree(ra);
804 			kfree(re);
805 			goto out_unlock;
806 		}
807 
808 		if (!parent) {
809 			tmp = insert_root_entry(&be->roots, re);
810 			if (tmp) {
811 				kfree(re);
812 				re = tmp;
813 			}
814 		}
815 	}
816 
817 	exist = insert_ref_entry(&be->refs, ref);
818 	if (exist) {
819 		if (action == BTRFS_DROP_DELAYED_REF) {
820 			if (exist->num_refs == 0) {
821 				btrfs_err(fs_info,
822 "dropping a ref for a existing root that doesn't have a ref on the block");
823 				dump_block_entry(fs_info, be);
824 				dump_ref_action(fs_info, ra);
825 				kfree(ref);
826 				kfree(ra);
827 				goto out_unlock;
828 			}
829 			exist->num_refs--;
830 			if (exist->num_refs == 0) {
831 				rb_erase(&exist->node, &be->refs);
832 				kfree(exist);
833 			}
834 		} else if (!be->metadata) {
835 			exist->num_refs++;
836 		} else {
837 			btrfs_err(fs_info,
838 "attempting to add another ref for an existing ref on a tree block");
839 			dump_block_entry(fs_info, be);
840 			dump_ref_action(fs_info, ra);
841 			kfree(ref);
842 			kfree(ra);
843 			goto out_unlock;
844 		}
845 		kfree(ref);
846 	} else {
847 		if (action == BTRFS_DROP_DELAYED_REF) {
848 			btrfs_err(fs_info,
849 "dropping a ref for a root that doesn't have a ref on the block");
850 			dump_block_entry(fs_info, be);
851 			dump_ref_action(fs_info, ra);
852 			kfree(ref);
853 			kfree(ra);
854 			goto out_unlock;
855 		}
856 	}
857 
858 	if (!parent && !re) {
859 		re = lookup_root_entry(&be->roots, ref_root);
860 		if (!re) {
861 			/*
862 			 * This shouldn't happen because we will add our re
863 			 * above when we lookup the be with !parent, but just in
864 			 * case catch this case so we don't panic because I
865 			 * didn't think of some other corner case.
866 			 */
867 			btrfs_err(fs_info, "failed to find root %llu for %llu",
868 				  generic_ref->real_root, be->bytenr);
869 			dump_block_entry(fs_info, be);
870 			dump_ref_action(fs_info, ra);
871 			kfree(ra);
872 			goto out_unlock;
873 		}
874 	}
875 	if (action == BTRFS_DROP_DELAYED_REF) {
876 		if (re)
877 			re->num_refs--;
878 		be->num_refs--;
879 	} else if (action == BTRFS_ADD_DELAYED_REF) {
880 		be->num_refs++;
881 		if (re)
882 			re->num_refs++;
883 	}
884 	list_add_tail(&ra->list, &be->actions);
885 	ret = 0;
886 out_unlock:
887 	spin_unlock(&fs_info->ref_verify_lock);
888 out:
889 	if (ret) {
890 		btrfs_free_ref_cache(fs_info);
891 		btrfs_clear_opt(fs_info->mount_opt, REF_VERIFY);
892 	}
893 	return ret;
894 }
895 
896 /* Free up the ref cache */
897 void btrfs_free_ref_cache(struct btrfs_fs_info *fs_info)
898 {
899 	struct block_entry *be;
900 	struct rb_node *n;
901 
902 	if (!btrfs_test_opt(fs_info, REF_VERIFY))
903 		return;
904 
905 	spin_lock(&fs_info->ref_verify_lock);
906 	while ((n = rb_first(&fs_info->block_tree))) {
907 		be = rb_entry(n, struct block_entry, node);
908 		rb_erase(&be->node, &fs_info->block_tree);
909 		free_block_entry(be);
910 		cond_resched_lock(&fs_info->ref_verify_lock);
911 	}
912 	spin_unlock(&fs_info->ref_verify_lock);
913 }
914 
915 void btrfs_free_ref_tree_range(struct btrfs_fs_info *fs_info, u64 start,
916 			       u64 len)
917 {
918 	struct block_entry *be = NULL, *entry;
919 	struct rb_node *n;
920 
921 	if (!btrfs_test_opt(fs_info, REF_VERIFY))
922 		return;
923 
924 	spin_lock(&fs_info->ref_verify_lock);
925 	n = fs_info->block_tree.rb_node;
926 	while (n) {
927 		entry = rb_entry(n, struct block_entry, node);
928 		if (entry->bytenr < start) {
929 			n = n->rb_right;
930 		} else if (entry->bytenr > start) {
931 			n = n->rb_left;
932 		} else {
933 			be = entry;
934 			break;
935 		}
936 		/* We want to get as close to start as possible */
937 		if (be == NULL ||
938 		    (entry->bytenr < start && be->bytenr > start) ||
939 		    (entry->bytenr < start && entry->bytenr > be->bytenr))
940 			be = entry;
941 	}
942 
943 	/*
944 	 * Could have an empty block group, maybe have something to check for
945 	 * this case to verify we were actually empty?
946 	 */
947 	if (!be) {
948 		spin_unlock(&fs_info->ref_verify_lock);
949 		return;
950 	}
951 
952 	n = &be->node;
953 	while (n) {
954 		be = rb_entry(n, struct block_entry, node);
955 		n = rb_next(n);
956 		if (be->bytenr < start && be->bytenr + be->len > start) {
957 			btrfs_err(fs_info,
958 				"block entry overlaps a block group [%llu,%llu]!",
959 				start, len);
960 			dump_block_entry(fs_info, be);
961 			continue;
962 		}
963 		if (be->bytenr < start)
964 			continue;
965 		if (be->bytenr >= start + len)
966 			break;
967 		if (be->bytenr + be->len > start + len) {
968 			btrfs_err(fs_info,
969 				"block entry overlaps a block group [%llu,%llu]!",
970 				start, len);
971 			dump_block_entry(fs_info, be);
972 		}
973 		rb_erase(&be->node, &fs_info->block_tree);
974 		free_block_entry(be);
975 	}
976 	spin_unlock(&fs_info->ref_verify_lock);
977 }
978 
979 /* Walk down all roots and build the ref tree, meant to be called at mount */
980 int btrfs_build_ref_tree(struct btrfs_fs_info *fs_info)
981 {
982 	struct btrfs_root *extent_root;
983 	struct btrfs_path *path;
984 	struct extent_buffer *eb;
985 	int tree_block_level = 0;
986 	u64 bytenr = 0, num_bytes = 0;
987 	int ret, level;
988 
989 	if (!btrfs_test_opt(fs_info, REF_VERIFY))
990 		return 0;
991 
992 	path = btrfs_alloc_path();
993 	if (!path)
994 		return -ENOMEM;
995 
996 	extent_root = btrfs_extent_root(fs_info, 0);
997 	eb = btrfs_read_lock_root_node(extent_root);
998 	level = btrfs_header_level(eb);
999 	path->nodes[level] = eb;
1000 	path->slots[level] = 0;
1001 	path->locks[level] = BTRFS_READ_LOCK;
1002 
1003 	while (1) {
1004 		/*
1005 		 * We have to keep track of the bytenr/num_bytes we last hit
1006 		 * because we could have run out of space for an inline ref, and
1007 		 * would have had to added a ref key item which may appear on a
1008 		 * different leaf from the original extent item.
1009 		 */
1010 		ret = walk_down_tree(extent_root, path, level,
1011 				     &bytenr, &num_bytes, &tree_block_level);
1012 		if (ret)
1013 			break;
1014 		ret = walk_up_tree(path, &level);
1015 		if (ret < 0)
1016 			break;
1017 		if (ret > 0) {
1018 			ret = 0;
1019 			break;
1020 		}
1021 	}
1022 	if (ret) {
1023 		btrfs_free_ref_cache(fs_info);
1024 		btrfs_clear_opt(fs_info->mount_opt, REF_VERIFY);
1025 	}
1026 	btrfs_free_path(path);
1027 	return ret;
1028 }
1029