xref: /linux/fs/btrfs/ref-verify.c (revision e60e1ee60630cafef5e430c2ae364877e061d980)
1 /*
2  * Copyright (C) 2014 Facebook.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 #include <linux/sched.h>
20 #include <linux/stacktrace.h>
21 #include "ctree.h"
22 #include "disk-io.h"
23 #include "locking.h"
24 #include "delayed-ref.h"
25 #include "ref-verify.h"
26 
27 /*
28  * Used to keep track the roots and number of refs each root has for a given
29  * bytenr.  This just tracks the number of direct references, no shared
30  * references.
31  */
32 struct root_entry {
33 	u64 root_objectid;
34 	u64 num_refs;
35 	struct rb_node node;
36 };
37 
38 /*
39  * These are meant to represent what should exist in the extent tree, these can
40  * be used to verify the extent tree is consistent as these should all match
41  * what the extent tree says.
42  */
43 struct ref_entry {
44 	u64 root_objectid;
45 	u64 parent;
46 	u64 owner;
47 	u64 offset;
48 	u64 num_refs;
49 	struct rb_node node;
50 };
51 
52 #define MAX_TRACE	16
53 
54 /*
55  * Whenever we add/remove a reference we record the action.  The action maps
56  * back to the delayed ref action.  We hold the ref we are changing in the
57  * action so we can account for the history properly, and we record the root we
58  * were called with since it could be different from ref_root.  We also store
59  * stack traces because thats how I roll.
60  */
61 struct ref_action {
62 	int action;
63 	u64 root;
64 	struct ref_entry ref;
65 	struct list_head list;
66 	unsigned long trace[MAX_TRACE];
67 	unsigned int trace_len;
68 };
69 
70 /*
71  * One of these for every block we reference, it holds the roots and references
72  * to it as well as all of the ref actions that have occured to it.  We never
73  * free it until we unmount the file system in order to make sure re-allocations
74  * are happening properly.
75  */
76 struct block_entry {
77 	u64 bytenr;
78 	u64 len;
79 	u64 num_refs;
80 	int metadata;
81 	int from_disk;
82 	struct rb_root roots;
83 	struct rb_root refs;
84 	struct rb_node node;
85 	struct list_head actions;
86 };
87 
88 static struct block_entry *insert_block_entry(struct rb_root *root,
89 					      struct block_entry *be)
90 {
91 	struct rb_node **p = &root->rb_node;
92 	struct rb_node *parent_node = NULL;
93 	struct block_entry *entry;
94 
95 	while (*p) {
96 		parent_node = *p;
97 		entry = rb_entry(parent_node, struct block_entry, node);
98 		if (entry->bytenr > be->bytenr)
99 			p = &(*p)->rb_left;
100 		else if (entry->bytenr < be->bytenr)
101 			p = &(*p)->rb_right;
102 		else
103 			return entry;
104 	}
105 
106 	rb_link_node(&be->node, parent_node, p);
107 	rb_insert_color(&be->node, root);
108 	return NULL;
109 }
110 
111 static struct block_entry *lookup_block_entry(struct rb_root *root, u64 bytenr)
112 {
113 	struct rb_node *n;
114 	struct block_entry *entry = NULL;
115 
116 	n = root->rb_node;
117 	while (n) {
118 		entry = rb_entry(n, struct block_entry, node);
119 		if (entry->bytenr < bytenr)
120 			n = n->rb_right;
121 		else if (entry->bytenr > bytenr)
122 			n = n->rb_left;
123 		else
124 			return entry;
125 	}
126 	return NULL;
127 }
128 
129 static struct root_entry *insert_root_entry(struct rb_root *root,
130 					    struct root_entry *re)
131 {
132 	struct rb_node **p = &root->rb_node;
133 	struct rb_node *parent_node = NULL;
134 	struct root_entry *entry;
135 
136 	while (*p) {
137 		parent_node = *p;
138 		entry = rb_entry(parent_node, struct root_entry, node);
139 		if (entry->root_objectid > re->root_objectid)
140 			p = &(*p)->rb_left;
141 		else if (entry->root_objectid < re->root_objectid)
142 			p = &(*p)->rb_right;
143 		else
144 			return entry;
145 	}
146 
147 	rb_link_node(&re->node, parent_node, p);
148 	rb_insert_color(&re->node, root);
149 	return NULL;
150 
151 }
152 
153 static int comp_refs(struct ref_entry *ref1, struct ref_entry *ref2)
154 {
155 	if (ref1->root_objectid < ref2->root_objectid)
156 		return -1;
157 	if (ref1->root_objectid > ref2->root_objectid)
158 		return 1;
159 	if (ref1->parent < ref2->parent)
160 		return -1;
161 	if (ref1->parent > ref2->parent)
162 		return 1;
163 	if (ref1->owner < ref2->owner)
164 		return -1;
165 	if (ref1->owner > ref2->owner)
166 		return 1;
167 	if (ref1->offset < ref2->offset)
168 		return -1;
169 	if (ref1->offset > ref2->offset)
170 		return 1;
171 	return 0;
172 }
173 
174 static struct ref_entry *insert_ref_entry(struct rb_root *root,
175 					  struct ref_entry *ref)
176 {
177 	struct rb_node **p = &root->rb_node;
178 	struct rb_node *parent_node = NULL;
179 	struct ref_entry *entry;
180 	int cmp;
181 
182 	while (*p) {
183 		parent_node = *p;
184 		entry = rb_entry(parent_node, struct ref_entry, node);
185 		cmp = comp_refs(entry, ref);
186 		if (cmp > 0)
187 			p = &(*p)->rb_left;
188 		else if (cmp < 0)
189 			p = &(*p)->rb_right;
190 		else
191 			return entry;
192 	}
193 
194 	rb_link_node(&ref->node, parent_node, p);
195 	rb_insert_color(&ref->node, root);
196 	return NULL;
197 
198 }
199 
200 static struct root_entry *lookup_root_entry(struct rb_root *root, u64 objectid)
201 {
202 	struct rb_node *n;
203 	struct root_entry *entry = NULL;
204 
205 	n = root->rb_node;
206 	while (n) {
207 		entry = rb_entry(n, struct root_entry, node);
208 		if (entry->root_objectid < objectid)
209 			n = n->rb_right;
210 		else if (entry->root_objectid > objectid)
211 			n = n->rb_left;
212 		else
213 			return entry;
214 	}
215 	return NULL;
216 }
217 
218 #ifdef CONFIG_STACKTRACE
219 static void __save_stack_trace(struct ref_action *ra)
220 {
221 	struct stack_trace stack_trace;
222 
223 	stack_trace.max_entries = MAX_TRACE;
224 	stack_trace.nr_entries = 0;
225 	stack_trace.entries = ra->trace;
226 	stack_trace.skip = 2;
227 	save_stack_trace(&stack_trace);
228 	ra->trace_len = stack_trace.nr_entries;
229 }
230 
231 static void __print_stack_trace(struct btrfs_fs_info *fs_info,
232 				struct ref_action *ra)
233 {
234 	struct stack_trace trace;
235 
236 	if (ra->trace_len == 0) {
237 		btrfs_err(fs_info, "  ref-verify: no stacktrace");
238 		return;
239 	}
240 	trace.nr_entries = ra->trace_len;
241 	trace.entries = ra->trace;
242 	print_stack_trace(&trace, 2);
243 }
244 #else
245 static void inline __save_stack_trace(struct ref_action *ra)
246 {
247 }
248 
249 static void inline __print_stack_trace(struct btrfs_fs_info *fs_info,
250 				       struct ref_action *ra)
251 {
252 	btrfs_err(fs_info, "  ref-verify: no stacktrace support");
253 }
254 #endif
255 
256 static void free_block_entry(struct block_entry *be)
257 {
258 	struct root_entry *re;
259 	struct ref_entry *ref;
260 	struct ref_action *ra;
261 	struct rb_node *n;
262 
263 	while ((n = rb_first(&be->roots))) {
264 		re = rb_entry(n, struct root_entry, node);
265 		rb_erase(&re->node, &be->roots);
266 		kfree(re);
267 	}
268 
269 	while((n = rb_first(&be->refs))) {
270 		ref = rb_entry(n, struct ref_entry, node);
271 		rb_erase(&ref->node, &be->refs);
272 		kfree(ref);
273 	}
274 
275 	while (!list_empty(&be->actions)) {
276 		ra = list_first_entry(&be->actions, struct ref_action,
277 				      list);
278 		list_del(&ra->list);
279 		kfree(ra);
280 	}
281 	kfree(be);
282 }
283 
284 static struct block_entry *add_block_entry(struct btrfs_fs_info *fs_info,
285 					   u64 bytenr, u64 len,
286 					   u64 root_objectid)
287 {
288 	struct block_entry *be = NULL, *exist;
289 	struct root_entry *re = NULL;
290 
291 	re = kzalloc(sizeof(struct root_entry), GFP_KERNEL);
292 	be = kzalloc(sizeof(struct block_entry), GFP_KERNEL);
293 	if (!be || !re) {
294 		kfree(re);
295 		kfree(be);
296 		return ERR_PTR(-ENOMEM);
297 	}
298 	be->bytenr = bytenr;
299 	be->len = len;
300 
301 	re->root_objectid = root_objectid;
302 	re->num_refs = 0;
303 
304 	spin_lock(&fs_info->ref_verify_lock);
305 	exist = insert_block_entry(&fs_info->block_tree, be);
306 	if (exist) {
307 		if (root_objectid) {
308 			struct root_entry *exist_re;
309 
310 			exist_re = insert_root_entry(&exist->roots, re);
311 			if (exist_re)
312 				kfree(re);
313 		}
314 		kfree(be);
315 		return exist;
316 	}
317 
318 	be->num_refs = 0;
319 	be->metadata = 0;
320 	be->from_disk = 0;
321 	be->roots = RB_ROOT;
322 	be->refs = RB_ROOT;
323 	INIT_LIST_HEAD(&be->actions);
324 	if (root_objectid)
325 		insert_root_entry(&be->roots, re);
326 	else
327 		kfree(re);
328 	return be;
329 }
330 
331 static int add_tree_block(struct btrfs_fs_info *fs_info, u64 ref_root,
332 			  u64 parent, u64 bytenr, int level)
333 {
334 	struct block_entry *be;
335 	struct root_entry *re;
336 	struct ref_entry *ref = NULL, *exist;
337 
338 	ref = kmalloc(sizeof(struct ref_entry), GFP_KERNEL);
339 	if (!ref)
340 		return -ENOMEM;
341 
342 	if (parent)
343 		ref->root_objectid = 0;
344 	else
345 		ref->root_objectid = ref_root;
346 	ref->parent = parent;
347 	ref->owner = level;
348 	ref->offset = 0;
349 	ref->num_refs = 1;
350 
351 	be = add_block_entry(fs_info, bytenr, fs_info->nodesize, ref_root);
352 	if (IS_ERR(be)) {
353 		kfree(ref);
354 		return PTR_ERR(be);
355 	}
356 	be->num_refs++;
357 	be->from_disk = 1;
358 	be->metadata = 1;
359 
360 	if (!parent) {
361 		ASSERT(ref_root);
362 		re = lookup_root_entry(&be->roots, ref_root);
363 		ASSERT(re);
364 		re->num_refs++;
365 	}
366 	exist = insert_ref_entry(&be->refs, ref);
367 	if (exist) {
368 		exist->num_refs++;
369 		kfree(ref);
370 	}
371 	spin_unlock(&fs_info->ref_verify_lock);
372 
373 	return 0;
374 }
375 
376 static int add_shared_data_ref(struct btrfs_fs_info *fs_info,
377 			       u64 parent, u32 num_refs, u64 bytenr,
378 			       u64 num_bytes)
379 {
380 	struct block_entry *be;
381 	struct ref_entry *ref;
382 
383 	ref = kzalloc(sizeof(struct ref_entry), GFP_KERNEL);
384 	if (!ref)
385 		return -ENOMEM;
386 	be = add_block_entry(fs_info, bytenr, num_bytes, 0);
387 	if (IS_ERR(be)) {
388 		kfree(ref);
389 		return PTR_ERR(be);
390 	}
391 	be->num_refs += num_refs;
392 
393 	ref->parent = parent;
394 	ref->num_refs = num_refs;
395 	if (insert_ref_entry(&be->refs, ref)) {
396 		spin_unlock(&fs_info->ref_verify_lock);
397 		btrfs_err(fs_info, "existing shared ref when reading from disk?");
398 		kfree(ref);
399 		return -EINVAL;
400 	}
401 	spin_unlock(&fs_info->ref_verify_lock);
402 	return 0;
403 }
404 
405 static int add_extent_data_ref(struct btrfs_fs_info *fs_info,
406 			       struct extent_buffer *leaf,
407 			       struct btrfs_extent_data_ref *dref,
408 			       u64 bytenr, u64 num_bytes)
409 {
410 	struct block_entry *be;
411 	struct ref_entry *ref;
412 	struct root_entry *re;
413 	u64 ref_root = btrfs_extent_data_ref_root(leaf, dref);
414 	u64 owner = btrfs_extent_data_ref_objectid(leaf, dref);
415 	u64 offset = btrfs_extent_data_ref_offset(leaf, dref);
416 	u32 num_refs = btrfs_extent_data_ref_count(leaf, dref);
417 
418 	ref = kzalloc(sizeof(struct ref_entry), GFP_KERNEL);
419 	if (!ref)
420 		return -ENOMEM;
421 	be = add_block_entry(fs_info, bytenr, num_bytes, ref_root);
422 	if (IS_ERR(be)) {
423 		kfree(ref);
424 		return PTR_ERR(be);
425 	}
426 	be->num_refs += num_refs;
427 
428 	ref->parent = 0;
429 	ref->owner = owner;
430 	ref->root_objectid = ref_root;
431 	ref->offset = offset;
432 	ref->num_refs = num_refs;
433 	if (insert_ref_entry(&be->refs, ref)) {
434 		spin_unlock(&fs_info->ref_verify_lock);
435 		btrfs_err(fs_info, "existing ref when reading from disk?");
436 		kfree(ref);
437 		return -EINVAL;
438 	}
439 
440 	re = lookup_root_entry(&be->roots, ref_root);
441 	if (!re) {
442 		spin_unlock(&fs_info->ref_verify_lock);
443 		btrfs_err(fs_info, "missing root in new block entry?");
444 		return -EINVAL;
445 	}
446 	re->num_refs += num_refs;
447 	spin_unlock(&fs_info->ref_verify_lock);
448 	return 0;
449 }
450 
451 static int process_extent_item(struct btrfs_fs_info *fs_info,
452 			       struct btrfs_path *path, struct btrfs_key *key,
453 			       int slot, int *tree_block_level)
454 {
455 	struct btrfs_extent_item *ei;
456 	struct btrfs_extent_inline_ref *iref;
457 	struct btrfs_extent_data_ref *dref;
458 	struct btrfs_shared_data_ref *sref;
459 	struct extent_buffer *leaf = path->nodes[0];
460 	u32 item_size = btrfs_item_size_nr(leaf, slot);
461 	unsigned long end, ptr;
462 	u64 offset, flags, count;
463 	int type, ret;
464 
465 	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
466 	flags = btrfs_extent_flags(leaf, ei);
467 
468 	if ((key->type == BTRFS_EXTENT_ITEM_KEY) &&
469 	    flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
470 		struct btrfs_tree_block_info *info;
471 
472 		info = (struct btrfs_tree_block_info *)(ei + 1);
473 		*tree_block_level = btrfs_tree_block_level(leaf, info);
474 		iref = (struct btrfs_extent_inline_ref *)(info + 1);
475 	} else {
476 		if (key->type == BTRFS_METADATA_ITEM_KEY)
477 			*tree_block_level = key->offset;
478 		iref = (struct btrfs_extent_inline_ref *)(ei + 1);
479 	}
480 
481 	ptr = (unsigned long)iref;
482 	end = (unsigned long)ei + item_size;
483 	while (ptr < end) {
484 		iref = (struct btrfs_extent_inline_ref *)ptr;
485 		type = btrfs_extent_inline_ref_type(leaf, iref);
486 		offset = btrfs_extent_inline_ref_offset(leaf, iref);
487 		switch (type) {
488 		case BTRFS_TREE_BLOCK_REF_KEY:
489 			ret = add_tree_block(fs_info, offset, 0, key->objectid,
490 					     *tree_block_level);
491 			break;
492 		case BTRFS_SHARED_BLOCK_REF_KEY:
493 			ret = add_tree_block(fs_info, 0, offset, key->objectid,
494 					     *tree_block_level);
495 			break;
496 		case BTRFS_EXTENT_DATA_REF_KEY:
497 			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
498 			ret = add_extent_data_ref(fs_info, leaf, dref,
499 						  key->objectid, key->offset);
500 			break;
501 		case BTRFS_SHARED_DATA_REF_KEY:
502 			sref = (struct btrfs_shared_data_ref *)(iref + 1);
503 			count = btrfs_shared_data_ref_count(leaf, sref);
504 			ret = add_shared_data_ref(fs_info, offset, count,
505 						  key->objectid, key->offset);
506 			break;
507 		default:
508 			btrfs_err(fs_info, "invalid key type in iref");
509 			ret = -EINVAL;
510 			break;
511 		}
512 		if (ret)
513 			break;
514 		ptr += btrfs_extent_inline_ref_size(type);
515 	}
516 	return ret;
517 }
518 
519 static int process_leaf(struct btrfs_root *root,
520 			struct btrfs_path *path, u64 *bytenr, u64 *num_bytes)
521 {
522 	struct btrfs_fs_info *fs_info = root->fs_info;
523 	struct extent_buffer *leaf = path->nodes[0];
524 	struct btrfs_extent_data_ref *dref;
525 	struct btrfs_shared_data_ref *sref;
526 	u32 count;
527 	int i = 0, tree_block_level = 0, ret;
528 	struct btrfs_key key;
529 	int nritems = btrfs_header_nritems(leaf);
530 
531 	for (i = 0; i < nritems; i++) {
532 		btrfs_item_key_to_cpu(leaf, &key, i);
533 		switch (key.type) {
534 		case BTRFS_EXTENT_ITEM_KEY:
535 			*num_bytes = key.offset;
536 		case BTRFS_METADATA_ITEM_KEY:
537 			*bytenr = key.objectid;
538 			ret = process_extent_item(fs_info, path, &key, i,
539 						  &tree_block_level);
540 			break;
541 		case BTRFS_TREE_BLOCK_REF_KEY:
542 			ret = add_tree_block(fs_info, key.offset, 0,
543 					     key.objectid, tree_block_level);
544 			break;
545 		case BTRFS_SHARED_BLOCK_REF_KEY:
546 			ret = add_tree_block(fs_info, 0, key.offset,
547 					     key.objectid, tree_block_level);
548 			break;
549 		case BTRFS_EXTENT_DATA_REF_KEY:
550 			dref = btrfs_item_ptr(leaf, i,
551 					      struct btrfs_extent_data_ref);
552 			ret = add_extent_data_ref(fs_info, leaf, dref, *bytenr,
553 						  *num_bytes);
554 			break;
555 		case BTRFS_SHARED_DATA_REF_KEY:
556 			sref = btrfs_item_ptr(leaf, i,
557 					      struct btrfs_shared_data_ref);
558 			count = btrfs_shared_data_ref_count(leaf, sref);
559 			ret = add_shared_data_ref(fs_info, key.offset, count,
560 						  *bytenr, *num_bytes);
561 			break;
562 		default:
563 			break;
564 		}
565 		if (ret)
566 			break;
567 	}
568 	return ret;
569 }
570 
571 /* Walk down to the leaf from the given level */
572 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
573 			  int level, u64 *bytenr, u64 *num_bytes)
574 {
575 	struct btrfs_fs_info *fs_info = root->fs_info;
576 	struct extent_buffer *eb;
577 	u64 block_bytenr, gen;
578 	int ret = 0;
579 
580 	while (level >= 0) {
581 		if (level) {
582 			block_bytenr = btrfs_node_blockptr(path->nodes[level],
583 							   path->slots[level]);
584 			gen = btrfs_node_ptr_generation(path->nodes[level],
585 							path->slots[level]);
586 			eb = read_tree_block(fs_info, block_bytenr, gen);
587 			if (IS_ERR(eb))
588 				return PTR_ERR(eb);
589 			if (!extent_buffer_uptodate(eb)) {
590 				free_extent_buffer(eb);
591 				return -EIO;
592 			}
593 			btrfs_tree_read_lock(eb);
594 			btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
595 			path->nodes[level-1] = eb;
596 			path->slots[level-1] = 0;
597 			path->locks[level-1] = BTRFS_READ_LOCK_BLOCKING;
598 		} else {
599 			ret = process_leaf(root, path, bytenr, num_bytes);
600 			if (ret)
601 				break;
602 		}
603 		level--;
604 	}
605 	return ret;
606 }
607 
608 /* Walk up to the next node that needs to be processed */
609 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
610 			int *level)
611 {
612 	int l;
613 
614 	for (l = 0; l < BTRFS_MAX_LEVEL; l++) {
615 		if (!path->nodes[l])
616 			continue;
617 		if (l) {
618 			path->slots[l]++;
619 			if (path->slots[l] <
620 			    btrfs_header_nritems(path->nodes[l])) {
621 				*level = l;
622 				return 0;
623 			}
624 		}
625 		btrfs_tree_unlock_rw(path->nodes[l], path->locks[l]);
626 		free_extent_buffer(path->nodes[l]);
627 		path->nodes[l] = NULL;
628 		path->slots[l] = 0;
629 		path->locks[l] = 0;
630 	}
631 
632 	return 1;
633 }
634 
635 static void dump_ref_action(struct btrfs_fs_info *fs_info,
636 			    struct ref_action *ra)
637 {
638 	btrfs_err(fs_info,
639 "  Ref action %d, root %llu, ref_root %llu, parent %llu, owner %llu, offset %llu, num_refs %llu",
640 		  ra->action, ra->root, ra->ref.root_objectid, ra->ref.parent,
641 		  ra->ref.owner, ra->ref.offset, ra->ref.num_refs);
642 	__print_stack_trace(fs_info, ra);
643 }
644 
645 /*
646  * Dumps all the information from the block entry to printk, it's going to be
647  * awesome.
648  */
649 static void dump_block_entry(struct btrfs_fs_info *fs_info,
650 			     struct block_entry *be)
651 {
652 	struct ref_entry *ref;
653 	struct root_entry *re;
654 	struct ref_action *ra;
655 	struct rb_node *n;
656 
657 	btrfs_err(fs_info,
658 "dumping block entry [%llu %llu], num_refs %llu, metadata %d, from disk %d",
659 		  be->bytenr, be->len, be->num_refs, be->metadata,
660 		  be->from_disk);
661 
662 	for (n = rb_first(&be->refs); n; n = rb_next(n)) {
663 		ref = rb_entry(n, struct ref_entry, node);
664 		btrfs_err(fs_info,
665 "  ref root %llu, parent %llu, owner %llu, offset %llu, num_refs %llu",
666 			  ref->root_objectid, ref->parent, ref->owner,
667 			  ref->offset, ref->num_refs);
668 	}
669 
670 	for (n = rb_first(&be->roots); n; n = rb_next(n)) {
671 		re = rb_entry(n, struct root_entry, node);
672 		btrfs_err(fs_info, "  root entry %llu, num_refs %llu",
673 			  re->root_objectid, re->num_refs);
674 	}
675 
676 	list_for_each_entry(ra, &be->actions, list)
677 		dump_ref_action(fs_info, ra);
678 }
679 
680 /*
681  * btrfs_ref_tree_mod: called when we modify a ref for a bytenr
682  * @root: the root we are making this modification from.
683  * @bytenr: the bytenr we are modifying.
684  * @num_bytes: number of bytes.
685  * @parent: the parent bytenr.
686  * @ref_root: the original root owner of the bytenr.
687  * @owner: level in the case of metadata, inode in the case of data.
688  * @offset: 0 for metadata, file offset for data.
689  * @action: the action that we are doing, this is the same as the delayed ref
690  *	action.
691  *
692  * This will add an action item to the given bytenr and do sanity checks to make
693  * sure we haven't messed something up.  If we are making a new allocation and
694  * this block entry has history we will delete all previous actions as long as
695  * our sanity checks pass as they are no longer needed.
696  */
697 int btrfs_ref_tree_mod(struct btrfs_root *root, u64 bytenr, u64 num_bytes,
698 		       u64 parent, u64 ref_root, u64 owner, u64 offset,
699 		       int action)
700 {
701 	struct btrfs_fs_info *fs_info = root->fs_info;
702 	struct ref_entry *ref = NULL, *exist;
703 	struct ref_action *ra = NULL;
704 	struct block_entry *be = NULL;
705 	struct root_entry *re = NULL;
706 	int ret = 0;
707 	bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
708 
709 	if (!btrfs_test_opt(root->fs_info, REF_VERIFY))
710 		return 0;
711 
712 	ref = kzalloc(sizeof(struct ref_entry), GFP_NOFS);
713 	ra = kmalloc(sizeof(struct ref_action), GFP_NOFS);
714 	if (!ra || !ref) {
715 		kfree(ref);
716 		kfree(ra);
717 		ret = -ENOMEM;
718 		goto out;
719 	}
720 
721 	if (parent) {
722 		ref->parent = parent;
723 	} else {
724 		ref->root_objectid = ref_root;
725 		ref->owner = owner;
726 		ref->offset = offset;
727 	}
728 	ref->num_refs = (action == BTRFS_DROP_DELAYED_REF) ? -1 : 1;
729 
730 	memcpy(&ra->ref, ref, sizeof(struct ref_entry));
731 	/*
732 	 * Save the extra info from the delayed ref in the ref action to make it
733 	 * easier to figure out what is happening.  The real ref's we add to the
734 	 * ref tree need to reflect what we save on disk so it matches any
735 	 * on-disk refs we pre-loaded.
736 	 */
737 	ra->ref.owner = owner;
738 	ra->ref.offset = offset;
739 	ra->ref.root_objectid = ref_root;
740 	__save_stack_trace(ra);
741 
742 	INIT_LIST_HEAD(&ra->list);
743 	ra->action = action;
744 	ra->root = root->objectid;
745 
746 	/*
747 	 * This is an allocation, preallocate the block_entry in case we haven't
748 	 * used it before.
749 	 */
750 	ret = -EINVAL;
751 	if (action == BTRFS_ADD_DELAYED_EXTENT) {
752 		/*
753 		 * For subvol_create we'll just pass in whatever the parent root
754 		 * is and the new root objectid, so let's not treat the passed
755 		 * in root as if it really has a ref for this bytenr.
756 		 */
757 		be = add_block_entry(root->fs_info, bytenr, num_bytes, ref_root);
758 		if (IS_ERR(be)) {
759 			kfree(ra);
760 			ret = PTR_ERR(be);
761 			goto out;
762 		}
763 		be->num_refs++;
764 		if (metadata)
765 			be->metadata = 1;
766 
767 		if (be->num_refs != 1) {
768 			btrfs_err(fs_info,
769 			"re-allocated a block that still has references to it!");
770 			dump_block_entry(fs_info, be);
771 			dump_ref_action(fs_info, ra);
772 			goto out_unlock;
773 		}
774 
775 		while (!list_empty(&be->actions)) {
776 			struct ref_action *tmp;
777 
778 			tmp = list_first_entry(&be->actions, struct ref_action,
779 					       list);
780 			list_del(&tmp->list);
781 			kfree(tmp);
782 		}
783 	} else {
784 		struct root_entry *tmp;
785 
786 		if (!parent) {
787 			re = kmalloc(sizeof(struct root_entry), GFP_NOFS);
788 			if (!re) {
789 				kfree(ref);
790 				kfree(ra);
791 				ret = -ENOMEM;
792 				goto out;
793 			}
794 			/*
795 			 * This is the root that is modifying us, so it's the
796 			 * one we want to lookup below when we modify the
797 			 * re->num_refs.
798 			 */
799 			ref_root = root->objectid;
800 			re->root_objectid = root->objectid;
801 			re->num_refs = 0;
802 		}
803 
804 		spin_lock(&root->fs_info->ref_verify_lock);
805 		be = lookup_block_entry(&root->fs_info->block_tree, bytenr);
806 		if (!be) {
807 			btrfs_err(fs_info,
808 "trying to do action %d to bytenr %llu num_bytes %llu but there is no existing entry!",
809 				  action, (unsigned long long)bytenr,
810 				  (unsigned long long)num_bytes);
811 			dump_ref_action(fs_info, ra);
812 			kfree(ref);
813 			kfree(ra);
814 			goto out_unlock;
815 		}
816 
817 		if (!parent) {
818 			tmp = insert_root_entry(&be->roots, re);
819 			if (tmp) {
820 				kfree(re);
821 				re = tmp;
822 			}
823 		}
824 	}
825 
826 	exist = insert_ref_entry(&be->refs, ref);
827 	if (exist) {
828 		if (action == BTRFS_DROP_DELAYED_REF) {
829 			if (exist->num_refs == 0) {
830 				btrfs_err(fs_info,
831 "dropping a ref for a existing root that doesn't have a ref on the block");
832 				dump_block_entry(fs_info, be);
833 				dump_ref_action(fs_info, ra);
834 				kfree(ra);
835 				goto out_unlock;
836 			}
837 			exist->num_refs--;
838 			if (exist->num_refs == 0) {
839 				rb_erase(&exist->node, &be->refs);
840 				kfree(exist);
841 			}
842 		} else if (!be->metadata) {
843 			exist->num_refs++;
844 		} else {
845 			btrfs_err(fs_info,
846 "attempting to add another ref for an existing ref on a tree block");
847 			dump_block_entry(fs_info, be);
848 			dump_ref_action(fs_info, ra);
849 			kfree(ra);
850 			goto out_unlock;
851 		}
852 		kfree(ref);
853 	} else {
854 		if (action == BTRFS_DROP_DELAYED_REF) {
855 			btrfs_err(fs_info,
856 "dropping a ref for a root that doesn't have a ref on the block");
857 			dump_block_entry(fs_info, be);
858 			dump_ref_action(fs_info, ra);
859 			kfree(ra);
860 			goto out_unlock;
861 		}
862 	}
863 
864 	if (!parent && !re) {
865 		re = lookup_root_entry(&be->roots, ref_root);
866 		if (!re) {
867 			/*
868 			 * This shouldn't happen because we will add our re
869 			 * above when we lookup the be with !parent, but just in
870 			 * case catch this case so we don't panic because I
871 			 * didn't thik of some other corner case.
872 			 */
873 			btrfs_err(fs_info, "failed to find root %llu for %llu",
874 				  root->objectid, be->bytenr);
875 			dump_block_entry(fs_info, be);
876 			dump_ref_action(fs_info, ra);
877 			kfree(ra);
878 			goto out_unlock;
879 		}
880 	}
881 	if (action == BTRFS_DROP_DELAYED_REF) {
882 		if (re)
883 			re->num_refs--;
884 		be->num_refs--;
885 	} else if (action == BTRFS_ADD_DELAYED_REF) {
886 		be->num_refs++;
887 		if (re)
888 			re->num_refs++;
889 	}
890 	list_add_tail(&ra->list, &be->actions);
891 	ret = 0;
892 out_unlock:
893 	spin_unlock(&root->fs_info->ref_verify_lock);
894 out:
895 	if (ret)
896 		btrfs_clear_opt(fs_info->mount_opt, REF_VERIFY);
897 	return ret;
898 }
899 
900 /* Free up the ref cache */
901 void btrfs_free_ref_cache(struct btrfs_fs_info *fs_info)
902 {
903 	struct block_entry *be;
904 	struct rb_node *n;
905 
906 	if (!btrfs_test_opt(fs_info, REF_VERIFY))
907 		return;
908 
909 	spin_lock(&fs_info->ref_verify_lock);
910 	while ((n = rb_first(&fs_info->block_tree))) {
911 		be = rb_entry(n, struct block_entry, node);
912 		rb_erase(&be->node, &fs_info->block_tree);
913 		free_block_entry(be);
914 		cond_resched_lock(&fs_info->ref_verify_lock);
915 	}
916 	spin_unlock(&fs_info->ref_verify_lock);
917 }
918 
919 void btrfs_free_ref_tree_range(struct btrfs_fs_info *fs_info, u64 start,
920 			       u64 len)
921 {
922 	struct block_entry *be = NULL, *entry;
923 	struct rb_node *n;
924 
925 	if (!btrfs_test_opt(fs_info, REF_VERIFY))
926 		return;
927 
928 	spin_lock(&fs_info->ref_verify_lock);
929 	n = fs_info->block_tree.rb_node;
930 	while (n) {
931 		entry = rb_entry(n, struct block_entry, node);
932 		if (entry->bytenr < start) {
933 			n = n->rb_right;
934 		} else if (entry->bytenr > start) {
935 			n = n->rb_left;
936 		} else {
937 			be = entry;
938 			break;
939 		}
940 		/* We want to get as close to start as possible */
941 		if (be == NULL ||
942 		    (entry->bytenr < start && be->bytenr > start) ||
943 		    (entry->bytenr < start && entry->bytenr > be->bytenr))
944 			be = entry;
945 	}
946 
947 	/*
948 	 * Could have an empty block group, maybe have something to check for
949 	 * this case to verify we were actually empty?
950 	 */
951 	if (!be) {
952 		spin_unlock(&fs_info->ref_verify_lock);
953 		return;
954 	}
955 
956 	n = &be->node;
957 	while (n) {
958 		be = rb_entry(n, struct block_entry, node);
959 		n = rb_next(n);
960 		if (be->bytenr < start && be->bytenr + be->len > start) {
961 			btrfs_err(fs_info,
962 				"block entry overlaps a block group [%llu,%llu]!",
963 				start, len);
964 			dump_block_entry(fs_info, be);
965 			continue;
966 		}
967 		if (be->bytenr < start)
968 			continue;
969 		if (be->bytenr >= start + len)
970 			break;
971 		if (be->bytenr + be->len > start + len) {
972 			btrfs_err(fs_info,
973 				"block entry overlaps a block group [%llu,%llu]!",
974 				start, len);
975 			dump_block_entry(fs_info, be);
976 		}
977 		rb_erase(&be->node, &fs_info->block_tree);
978 		free_block_entry(be);
979 	}
980 	spin_unlock(&fs_info->ref_verify_lock);
981 }
982 
983 /* Walk down all roots and build the ref tree, meant to be called at mount */
984 int btrfs_build_ref_tree(struct btrfs_fs_info *fs_info)
985 {
986 	struct btrfs_path *path;
987 	struct btrfs_root *root;
988 	struct extent_buffer *eb;
989 	u64 bytenr = 0, num_bytes = 0;
990 	int ret, level;
991 
992 	if (!btrfs_test_opt(fs_info, REF_VERIFY))
993 		return 0;
994 
995 	path = btrfs_alloc_path();
996 	if (!path)
997 		return -ENOMEM;
998 
999 	eb = btrfs_read_lock_root_node(fs_info->extent_root);
1000 	btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1001 	level = btrfs_header_level(eb);
1002 	path->nodes[level] = eb;
1003 	path->slots[level] = 0;
1004 	path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1005 
1006 	while (1) {
1007 		/*
1008 		 * We have to keep track of the bytenr/num_bytes we last hit
1009 		 * because we could have run out of space for an inline ref, and
1010 		 * would have had to added a ref key item which may appear on a
1011 		 * different leaf from the original extent item.
1012 		 */
1013 		ret = walk_down_tree(fs_info->extent_root, path, level,
1014 				     &bytenr, &num_bytes);
1015 		if (ret)
1016 			break;
1017 		ret = walk_up_tree(root, path, &level);
1018 		if (ret < 0)
1019 			break;
1020 		if (ret > 0) {
1021 			ret = 0;
1022 			break;
1023 		}
1024 	}
1025 	if (ret) {
1026 		btrfs_clear_opt(fs_info->mount_opt, REF_VERIFY);
1027 		btrfs_free_ref_cache(fs_info);
1028 	}
1029 	btrfs_free_path(path);
1030 	return ret;
1031 }
1032