2 * Copyright (C) 2007 Oracle. All rights reserved.
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.
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.
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.
19 #define _XOPEN_SOURCE 500
25 #include <sys/types.h>
29 #include <uuid/uuid.h>
34 #include "print-tree.h"
35 #include "transaction.h"
39 #include "free-space-cache.h"
41 #include "qgroup-verify.h"
42 #include "rbtree-utils.h"
46 static u64 bytes_used = 0;
47 static u64 total_csum_bytes = 0;
48 static u64 total_btree_bytes = 0;
49 static u64 total_fs_tree_bytes = 0;
50 static u64 total_extent_tree_bytes = 0;
51 static u64 btree_space_waste = 0;
52 static u64 data_bytes_allocated = 0;
53 static u64 data_bytes_referenced = 0;
54 static int found_old_backref = 0;
55 static LIST_HEAD(duplicate_extents);
56 static LIST_HEAD(delete_items);
57 static int repair = 0;
58 static int no_holes = 0;
59 static int init_extent_tree = 0;
60 static int check_data_csum = 0;
62 struct extent_backref {
63 struct list_head list;
64 unsigned int is_data:1;
65 unsigned int found_extent_tree:1;
66 unsigned int full_backref:1;
67 unsigned int found_ref:1;
68 unsigned int broken:1;
72 struct extent_backref node;
87 struct extent_backref node;
94 struct extent_record {
95 struct list_head backrefs;
96 struct list_head dups;
97 struct list_head list;
98 struct cache_extent cache;
99 struct btrfs_disk_key parent_key;
104 u64 extent_item_refs;
106 u64 parent_generation;
110 unsigned int found_rec:1;
111 unsigned int content_checked:1;
112 unsigned int owner_ref_checked:1;
113 unsigned int is_root:1;
114 unsigned int metadata:1;
117 struct inode_backref {
118 struct list_head list;
119 unsigned int found_dir_item:1;
120 unsigned int found_dir_index:1;
121 unsigned int found_inode_ref:1;
122 unsigned int filetype:8;
124 unsigned int ref_type;
131 struct dropping_root_item_record {
132 struct list_head list;
133 struct btrfs_root_item ri;
134 struct btrfs_key found_key;
137 #define REF_ERR_NO_DIR_ITEM (1 << 0)
138 #define REF_ERR_NO_DIR_INDEX (1 << 1)
139 #define REF_ERR_NO_INODE_REF (1 << 2)
140 #define REF_ERR_DUP_DIR_ITEM (1 << 3)
141 #define REF_ERR_DUP_DIR_INDEX (1 << 4)
142 #define REF_ERR_DUP_INODE_REF (1 << 5)
143 #define REF_ERR_INDEX_UNMATCH (1 << 6)
144 #define REF_ERR_FILETYPE_UNMATCH (1 << 7)
145 #define REF_ERR_NAME_TOO_LONG (1 << 8) // 100
146 #define REF_ERR_NO_ROOT_REF (1 << 9)
147 #define REF_ERR_NO_ROOT_BACKREF (1 << 10)
148 #define REF_ERR_DUP_ROOT_REF (1 << 11)
149 #define REF_ERR_DUP_ROOT_BACKREF (1 << 12)
151 struct inode_record {
152 struct list_head backrefs;
153 unsigned int checked:1;
154 unsigned int merging:1;
155 unsigned int found_inode_item:1;
156 unsigned int found_dir_item:1;
157 unsigned int found_file_extent:1;
158 unsigned int found_csum_item:1;
159 unsigned int some_csum_missing:1;
160 unsigned int nodatasum:1;
173 u64 first_extent_gap;
178 #define I_ERR_NO_INODE_ITEM (1 << 0)
179 #define I_ERR_NO_ORPHAN_ITEM (1 << 1)
180 #define I_ERR_DUP_INODE_ITEM (1 << 2)
181 #define I_ERR_DUP_DIR_INDEX (1 << 3)
182 #define I_ERR_ODD_DIR_ITEM (1 << 4)
183 #define I_ERR_ODD_FILE_EXTENT (1 << 5)
184 #define I_ERR_BAD_FILE_EXTENT (1 << 6)
185 #define I_ERR_FILE_EXTENT_OVERLAP (1 << 7)
186 #define I_ERR_FILE_EXTENT_DISCOUNT (1 << 8) // 100
187 #define I_ERR_DIR_ISIZE_WRONG (1 << 9)
188 #define I_ERR_FILE_NBYTES_WRONG (1 << 10) // 400
189 #define I_ERR_ODD_CSUM_ITEM (1 << 11)
190 #define I_ERR_SOME_CSUM_MISSING (1 << 12)
191 #define I_ERR_LINK_COUNT_WRONG (1 << 13)
193 struct root_backref {
194 struct list_head list;
195 unsigned int found_dir_item:1;
196 unsigned int found_dir_index:1;
197 unsigned int found_back_ref:1;
198 unsigned int found_forward_ref:1;
199 unsigned int reachable:1;
209 struct list_head backrefs;
210 struct cache_extent cache;
211 unsigned int found_root_item:1;
217 struct cache_extent cache;
222 struct cache_extent cache;
223 struct cache_tree root_cache;
224 struct cache_tree inode_cache;
225 struct inode_record *current;
234 struct walk_control {
235 struct cache_tree shared;
236 struct shared_node *nodes[BTRFS_MAX_LEVEL];
242 struct btrfs_key key;
244 struct list_head list;
247 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info);
249 static void record_root_in_trans(struct btrfs_trans_handle *trans,
250 struct btrfs_root *root)
252 if (root->last_trans != trans->transid) {
253 root->track_dirty = 1;
254 root->last_trans = trans->transid;
255 root->commit_root = root->node;
256 extent_buffer_get(root->node);
260 static u8 imode_to_type(u32 imode)
263 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
264 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
265 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
266 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
267 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
268 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
269 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
270 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
273 return btrfs_type_by_mode[(imode & S_IFMT) >> S_SHIFT];
277 static int device_record_compare(struct rb_node *node1, struct rb_node *node2)
279 struct device_record *rec1;
280 struct device_record *rec2;
282 rec1 = rb_entry(node1, struct device_record, node);
283 rec2 = rb_entry(node2, struct device_record, node);
284 if (rec1->devid > rec2->devid)
286 else if (rec1->devid < rec2->devid)
292 static struct inode_record *clone_inode_rec(struct inode_record *orig_rec)
294 struct inode_record *rec;
295 struct inode_backref *backref;
296 struct inode_backref *orig;
299 rec = malloc(sizeof(*rec));
300 memcpy(rec, orig_rec, sizeof(*rec));
302 INIT_LIST_HEAD(&rec->backrefs);
304 list_for_each_entry(orig, &orig_rec->backrefs, list) {
305 size = sizeof(*orig) + orig->namelen + 1;
306 backref = malloc(size);
307 memcpy(backref, orig, size);
308 list_add_tail(&backref->list, &rec->backrefs);
313 static void print_inode_error(struct btrfs_root *root, struct inode_record *rec)
315 u64 root_objectid = root->root_key.objectid;
316 int errors = rec->errors;
320 /* reloc root errors, we print its corresponding fs root objectid*/
321 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
322 root_objectid = root->root_key.offset;
323 fprintf(stderr, "reloc");
325 fprintf(stderr, "root %llu inode %llu errors %x",
326 (unsigned long long) root_objectid,
327 (unsigned long long) rec->ino, rec->errors);
329 if (errors & I_ERR_NO_INODE_ITEM)
330 fprintf(stderr, ", no inode item");
331 if (errors & I_ERR_NO_ORPHAN_ITEM)
332 fprintf(stderr, ", no orphan item");
333 if (errors & I_ERR_DUP_INODE_ITEM)
334 fprintf(stderr, ", dup inode item");
335 if (errors & I_ERR_DUP_DIR_INDEX)
336 fprintf(stderr, ", dup dir index");
337 if (errors & I_ERR_ODD_DIR_ITEM)
338 fprintf(stderr, ", odd dir item");
339 if (errors & I_ERR_ODD_FILE_EXTENT)
340 fprintf(stderr, ", odd file extent");
341 if (errors & I_ERR_BAD_FILE_EXTENT)
342 fprintf(stderr, ", bad file extent");
343 if (errors & I_ERR_FILE_EXTENT_OVERLAP)
344 fprintf(stderr, ", file extent overlap");
345 if (errors & I_ERR_FILE_EXTENT_DISCOUNT)
346 fprintf(stderr, ", file extent discount");
347 if (errors & I_ERR_DIR_ISIZE_WRONG)
348 fprintf(stderr, ", dir isize wrong");
349 if (errors & I_ERR_FILE_NBYTES_WRONG)
350 fprintf(stderr, ", nbytes wrong");
351 if (errors & I_ERR_ODD_CSUM_ITEM)
352 fprintf(stderr, ", odd csum item");
353 if (errors & I_ERR_SOME_CSUM_MISSING)
354 fprintf(stderr, ", some csum missing");
355 if (errors & I_ERR_LINK_COUNT_WRONG)
356 fprintf(stderr, ", link count wrong");
357 fprintf(stderr, "\n");
360 static void print_ref_error(int errors)
362 if (errors & REF_ERR_NO_DIR_ITEM)
363 fprintf(stderr, ", no dir item");
364 if (errors & REF_ERR_NO_DIR_INDEX)
365 fprintf(stderr, ", no dir index");
366 if (errors & REF_ERR_NO_INODE_REF)
367 fprintf(stderr, ", no inode ref");
368 if (errors & REF_ERR_DUP_DIR_ITEM)
369 fprintf(stderr, ", dup dir item");
370 if (errors & REF_ERR_DUP_DIR_INDEX)
371 fprintf(stderr, ", dup dir index");
372 if (errors & REF_ERR_DUP_INODE_REF)
373 fprintf(stderr, ", dup inode ref");
374 if (errors & REF_ERR_INDEX_UNMATCH)
375 fprintf(stderr, ", index unmatch");
376 if (errors & REF_ERR_FILETYPE_UNMATCH)
377 fprintf(stderr, ", filetype unmatch");
378 if (errors & REF_ERR_NAME_TOO_LONG)
379 fprintf(stderr, ", name too long");
380 if (errors & REF_ERR_NO_ROOT_REF)
381 fprintf(stderr, ", no root ref");
382 if (errors & REF_ERR_NO_ROOT_BACKREF)
383 fprintf(stderr, ", no root backref");
384 if (errors & REF_ERR_DUP_ROOT_REF)
385 fprintf(stderr, ", dup root ref");
386 if (errors & REF_ERR_DUP_ROOT_BACKREF)
387 fprintf(stderr, ", dup root backref");
388 fprintf(stderr, "\n");
391 static struct inode_record *get_inode_rec(struct cache_tree *inode_cache,
394 struct ptr_node *node;
395 struct cache_extent *cache;
396 struct inode_record *rec = NULL;
399 cache = lookup_cache_extent(inode_cache, ino, 1);
401 node = container_of(cache, struct ptr_node, cache);
403 if (mod && rec->refs > 1) {
404 node->data = clone_inode_rec(rec);
409 rec = calloc(1, sizeof(*rec));
411 rec->extent_start = (u64)-1;
412 rec->first_extent_gap = (u64)-1;
414 INIT_LIST_HEAD(&rec->backrefs);
416 node = malloc(sizeof(*node));
417 node->cache.start = ino;
418 node->cache.size = 1;
421 if (ino == BTRFS_FREE_INO_OBJECTID)
424 ret = insert_cache_extent(inode_cache, &node->cache);
430 static void free_inode_rec(struct inode_record *rec)
432 struct inode_backref *backref;
437 while (!list_empty(&rec->backrefs)) {
438 backref = list_entry(rec->backrefs.next,
439 struct inode_backref, list);
440 list_del(&backref->list);
446 static int can_free_inode_rec(struct inode_record *rec)
448 if (!rec->errors && rec->checked && rec->found_inode_item &&
449 rec->nlink == rec->found_link && list_empty(&rec->backrefs))
454 static void maybe_free_inode_rec(struct cache_tree *inode_cache,
455 struct inode_record *rec)
457 struct cache_extent *cache;
458 struct inode_backref *tmp, *backref;
459 struct ptr_node *node;
460 unsigned char filetype;
462 if (!rec->found_inode_item)
465 filetype = imode_to_type(rec->imode);
466 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
467 if (backref->found_dir_item && backref->found_dir_index) {
468 if (backref->filetype != filetype)
469 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
470 if (!backref->errors && backref->found_inode_ref) {
471 list_del(&backref->list);
477 if (!rec->checked || rec->merging)
480 if (S_ISDIR(rec->imode)) {
481 if (rec->found_size != rec->isize)
482 rec->errors |= I_ERR_DIR_ISIZE_WRONG;
483 if (rec->found_file_extent)
484 rec->errors |= I_ERR_ODD_FILE_EXTENT;
485 } else if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
486 if (rec->found_dir_item)
487 rec->errors |= I_ERR_ODD_DIR_ITEM;
488 if (rec->found_size != rec->nbytes)
489 rec->errors |= I_ERR_FILE_NBYTES_WRONG;
490 if (rec->extent_start == (u64)-1 || rec->extent_start > 0)
491 rec->first_extent_gap = 0;
492 if (rec->nlink > 0 && !no_holes &&
493 (rec->extent_end < rec->isize ||
494 rec->first_extent_gap < rec->isize))
495 rec->errors |= I_ERR_FILE_EXTENT_DISCOUNT;
498 if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
499 if (rec->found_csum_item && rec->nodatasum)
500 rec->errors |= I_ERR_ODD_CSUM_ITEM;
501 if (rec->some_csum_missing && !rec->nodatasum)
502 rec->errors |= I_ERR_SOME_CSUM_MISSING;
505 BUG_ON(rec->refs != 1);
506 if (can_free_inode_rec(rec)) {
507 cache = lookup_cache_extent(inode_cache, rec->ino, 1);
508 node = container_of(cache, struct ptr_node, cache);
509 BUG_ON(node->data != rec);
510 remove_cache_extent(inode_cache, &node->cache);
516 static int check_orphan_item(struct btrfs_root *root, u64 ino)
518 struct btrfs_path path;
519 struct btrfs_key key;
522 key.objectid = BTRFS_ORPHAN_OBJECTID;
523 key.type = BTRFS_ORPHAN_ITEM_KEY;
526 btrfs_init_path(&path);
527 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
528 btrfs_release_path(&path);
534 static int process_inode_item(struct extent_buffer *eb,
535 int slot, struct btrfs_key *key,
536 struct shared_node *active_node)
538 struct inode_record *rec;
539 struct btrfs_inode_item *item;
541 rec = active_node->current;
542 BUG_ON(rec->ino != key->objectid || rec->refs > 1);
543 if (rec->found_inode_item) {
544 rec->errors |= I_ERR_DUP_INODE_ITEM;
547 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
548 rec->nlink = btrfs_inode_nlink(eb, item);
549 rec->isize = btrfs_inode_size(eb, item);
550 rec->nbytes = btrfs_inode_nbytes(eb, item);
551 rec->imode = btrfs_inode_mode(eb, item);
552 if (btrfs_inode_flags(eb, item) & BTRFS_INODE_NODATASUM)
554 rec->found_inode_item = 1;
556 rec->errors |= I_ERR_NO_ORPHAN_ITEM;
557 maybe_free_inode_rec(&active_node->inode_cache, rec);
561 static struct inode_backref *get_inode_backref(struct inode_record *rec,
563 int namelen, u64 dir)
565 struct inode_backref *backref;
567 list_for_each_entry(backref, &rec->backrefs, list) {
568 if (rec->ino == BTRFS_MULTIPLE_OBJECTIDS)
570 if (backref->dir != dir || backref->namelen != namelen)
572 if (memcmp(name, backref->name, namelen))
577 backref = malloc(sizeof(*backref) + namelen + 1);
578 memset(backref, 0, sizeof(*backref));
580 backref->namelen = namelen;
581 memcpy(backref->name, name, namelen);
582 backref->name[namelen] = '\0';
583 list_add_tail(&backref->list, &rec->backrefs);
587 static int add_inode_backref(struct cache_tree *inode_cache,
588 u64 ino, u64 dir, u64 index,
589 const char *name, int namelen,
590 int filetype, int itemtype, int errors)
592 struct inode_record *rec;
593 struct inode_backref *backref;
595 rec = get_inode_rec(inode_cache, ino, 1);
596 backref = get_inode_backref(rec, name, namelen, dir);
598 backref->errors |= errors;
599 if (itemtype == BTRFS_DIR_INDEX_KEY) {
600 if (backref->found_dir_index)
601 backref->errors |= REF_ERR_DUP_DIR_INDEX;
602 if (backref->found_inode_ref && backref->index != index)
603 backref->errors |= REF_ERR_INDEX_UNMATCH;
604 if (backref->found_dir_item && backref->filetype != filetype)
605 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
607 backref->index = index;
608 backref->filetype = filetype;
609 backref->found_dir_index = 1;
610 } else if (itemtype == BTRFS_DIR_ITEM_KEY) {
612 if (backref->found_dir_item)
613 backref->errors |= REF_ERR_DUP_DIR_ITEM;
614 if (backref->found_dir_index && backref->filetype != filetype)
615 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
617 backref->filetype = filetype;
618 backref->found_dir_item = 1;
619 } else if ((itemtype == BTRFS_INODE_REF_KEY) ||
620 (itemtype == BTRFS_INODE_EXTREF_KEY)) {
621 if (backref->found_inode_ref)
622 backref->errors |= REF_ERR_DUP_INODE_REF;
623 if (backref->found_dir_index && backref->index != index)
624 backref->errors |= REF_ERR_INDEX_UNMATCH;
626 backref->index = index;
628 backref->ref_type = itemtype;
629 backref->found_inode_ref = 1;
634 maybe_free_inode_rec(inode_cache, rec);
638 static int merge_inode_recs(struct inode_record *src, struct inode_record *dst,
639 struct cache_tree *dst_cache)
641 struct inode_backref *backref;
645 list_for_each_entry(backref, &src->backrefs, list) {
646 if (backref->found_dir_index) {
647 add_inode_backref(dst_cache, dst->ino, backref->dir,
648 backref->index, backref->name,
649 backref->namelen, backref->filetype,
650 BTRFS_DIR_INDEX_KEY, backref->errors);
652 if (backref->found_dir_item) {
654 add_inode_backref(dst_cache, dst->ino,
655 backref->dir, 0, backref->name,
656 backref->namelen, backref->filetype,
657 BTRFS_DIR_ITEM_KEY, backref->errors);
659 if (backref->found_inode_ref) {
660 add_inode_backref(dst_cache, dst->ino,
661 backref->dir, backref->index,
662 backref->name, backref->namelen, 0,
663 backref->ref_type, backref->errors);
667 if (src->found_dir_item)
668 dst->found_dir_item = 1;
669 if (src->found_file_extent)
670 dst->found_file_extent = 1;
671 if (src->found_csum_item)
672 dst->found_csum_item = 1;
673 if (src->some_csum_missing)
674 dst->some_csum_missing = 1;
675 if (dst->first_extent_gap > src->first_extent_gap)
676 dst->first_extent_gap = src->first_extent_gap;
678 BUG_ON(src->found_link < dir_count);
679 dst->found_link += src->found_link - dir_count;
680 dst->found_size += src->found_size;
681 if (src->extent_start != (u64)-1) {
682 if (dst->extent_start == (u64)-1) {
683 dst->extent_start = src->extent_start;
684 dst->extent_end = src->extent_end;
686 if (dst->extent_end > src->extent_start)
687 dst->errors |= I_ERR_FILE_EXTENT_OVERLAP;
688 else if (dst->extent_end < src->extent_start &&
689 dst->extent_end < dst->first_extent_gap)
690 dst->first_extent_gap = dst->extent_end;
691 if (dst->extent_end < src->extent_end)
692 dst->extent_end = src->extent_end;
696 dst->errors |= src->errors;
697 if (src->found_inode_item) {
698 if (!dst->found_inode_item) {
699 dst->nlink = src->nlink;
700 dst->isize = src->isize;
701 dst->nbytes = src->nbytes;
702 dst->imode = src->imode;
703 dst->nodatasum = src->nodatasum;
704 dst->found_inode_item = 1;
706 dst->errors |= I_ERR_DUP_INODE_ITEM;
714 static int splice_shared_node(struct shared_node *src_node,
715 struct shared_node *dst_node)
717 struct cache_extent *cache;
718 struct ptr_node *node, *ins;
719 struct cache_tree *src, *dst;
720 struct inode_record *rec, *conflict;
725 if (--src_node->refs == 0)
727 if (src_node->current)
728 current_ino = src_node->current->ino;
730 src = &src_node->root_cache;
731 dst = &dst_node->root_cache;
733 cache = search_cache_extent(src, 0);
735 node = container_of(cache, struct ptr_node, cache);
737 cache = next_cache_extent(cache);
740 remove_cache_extent(src, &node->cache);
743 ins = malloc(sizeof(*ins));
744 ins->cache.start = node->cache.start;
745 ins->cache.size = node->cache.size;
749 ret = insert_cache_extent(dst, &ins->cache);
750 if (ret == -EEXIST) {
751 conflict = get_inode_rec(dst, rec->ino, 1);
752 merge_inode_recs(rec, conflict, dst);
754 conflict->checked = 1;
755 if (dst_node->current == conflict)
756 dst_node->current = NULL;
758 maybe_free_inode_rec(dst, conflict);
766 if (src == &src_node->root_cache) {
767 src = &src_node->inode_cache;
768 dst = &dst_node->inode_cache;
772 if (current_ino > 0 && (!dst_node->current ||
773 current_ino > dst_node->current->ino)) {
774 if (dst_node->current) {
775 dst_node->current->checked = 1;
776 maybe_free_inode_rec(dst, dst_node->current);
778 dst_node->current = get_inode_rec(dst, current_ino, 1);
783 static void free_inode_ptr(struct cache_extent *cache)
785 struct ptr_node *node;
786 struct inode_record *rec;
788 node = container_of(cache, struct ptr_node, cache);
794 FREE_EXTENT_CACHE_BASED_TREE(inode_recs, free_inode_ptr);
796 static struct shared_node *find_shared_node(struct cache_tree *shared,
799 struct cache_extent *cache;
800 struct shared_node *node;
802 cache = lookup_cache_extent(shared, bytenr, 1);
804 node = container_of(cache, struct shared_node, cache);
810 static int add_shared_node(struct cache_tree *shared, u64 bytenr, u32 refs)
813 struct shared_node *node;
815 node = calloc(1, sizeof(*node));
816 node->cache.start = bytenr;
817 node->cache.size = 1;
818 cache_tree_init(&node->root_cache);
819 cache_tree_init(&node->inode_cache);
822 ret = insert_cache_extent(shared, &node->cache);
827 static int enter_shared_node(struct btrfs_root *root, u64 bytenr, u32 refs,
828 struct walk_control *wc, int level)
830 struct shared_node *node;
831 struct shared_node *dest;
833 if (level == wc->active_node)
836 BUG_ON(wc->active_node <= level);
837 node = find_shared_node(&wc->shared, bytenr);
839 add_shared_node(&wc->shared, bytenr, refs);
840 node = find_shared_node(&wc->shared, bytenr);
841 wc->nodes[level] = node;
842 wc->active_node = level;
846 if (wc->root_level == wc->active_node &&
847 btrfs_root_refs(&root->root_item) == 0) {
848 if (--node->refs == 0) {
849 free_inode_recs_tree(&node->root_cache);
850 free_inode_recs_tree(&node->inode_cache);
851 remove_cache_extent(&wc->shared, &node->cache);
857 dest = wc->nodes[wc->active_node];
858 splice_shared_node(node, dest);
859 if (node->refs == 0) {
860 remove_cache_extent(&wc->shared, &node->cache);
866 static int leave_shared_node(struct btrfs_root *root,
867 struct walk_control *wc, int level)
869 struct shared_node *node;
870 struct shared_node *dest;
873 if (level == wc->root_level)
876 for (i = level + 1; i < BTRFS_MAX_LEVEL; i++) {
880 BUG_ON(i >= BTRFS_MAX_LEVEL);
882 node = wc->nodes[wc->active_node];
883 wc->nodes[wc->active_node] = NULL;
886 dest = wc->nodes[wc->active_node];
887 if (wc->active_node < wc->root_level ||
888 btrfs_root_refs(&root->root_item) > 0) {
889 BUG_ON(node->refs <= 1);
890 splice_shared_node(node, dest);
892 BUG_ON(node->refs < 2);
901 * 1 - if the root with id child_root_id is a child of root parent_root_id
902 * 0 - if the root child_root_id isn't a child of the root parent_root_id but
903 * has other root(s) as parent(s)
904 * 2 - if the root child_root_id doesn't have any parent roots
906 static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
909 struct btrfs_path path;
910 struct btrfs_key key;
911 struct extent_buffer *leaf;
915 btrfs_init_path(&path);
917 key.objectid = parent_root_id;
918 key.type = BTRFS_ROOT_REF_KEY;
919 key.offset = child_root_id;
920 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
924 btrfs_release_path(&path);
928 key.objectid = child_root_id;
929 key.type = BTRFS_ROOT_BACKREF_KEY;
931 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
937 leaf = path.nodes[0];
938 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
939 ret = btrfs_next_leaf(root->fs_info->tree_root, &path);
942 leaf = path.nodes[0];
945 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
946 if (key.objectid != child_root_id ||
947 key.type != BTRFS_ROOT_BACKREF_KEY)
952 if (key.offset == parent_root_id) {
953 btrfs_release_path(&path);
960 btrfs_release_path(&path);
963 return has_parent ? 0 : 2;
966 static int process_dir_item(struct btrfs_root *root,
967 struct extent_buffer *eb,
968 int slot, struct btrfs_key *key,
969 struct shared_node *active_node)
979 struct btrfs_dir_item *di;
980 struct inode_record *rec;
981 struct cache_tree *root_cache;
982 struct cache_tree *inode_cache;
983 struct btrfs_key location;
984 char namebuf[BTRFS_NAME_LEN];
986 root_cache = &active_node->root_cache;
987 inode_cache = &active_node->inode_cache;
988 rec = active_node->current;
989 rec->found_dir_item = 1;
991 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
992 total = btrfs_item_size_nr(eb, slot);
993 while (cur < total) {
995 btrfs_dir_item_key_to_cpu(eb, di, &location);
996 name_len = btrfs_dir_name_len(eb, di);
997 data_len = btrfs_dir_data_len(eb, di);
998 filetype = btrfs_dir_type(eb, di);
1000 rec->found_size += name_len;
1001 if (name_len <= BTRFS_NAME_LEN) {
1005 len = BTRFS_NAME_LEN;
1006 error = REF_ERR_NAME_TOO_LONG;
1008 read_extent_buffer(eb, namebuf, (unsigned long)(di + 1), len);
1010 if (location.type == BTRFS_INODE_ITEM_KEY) {
1011 add_inode_backref(inode_cache, location.objectid,
1012 key->objectid, key->offset, namebuf,
1013 len, filetype, key->type, error);
1014 } else if (location.type == BTRFS_ROOT_ITEM_KEY) {
1015 add_inode_backref(root_cache, location.objectid,
1016 key->objectid, key->offset,
1017 namebuf, len, filetype,
1020 fprintf(stderr, "invalid location in dir item %u\n",
1022 add_inode_backref(inode_cache, BTRFS_MULTIPLE_OBJECTIDS,
1023 key->objectid, key->offset, namebuf,
1024 len, filetype, key->type, error);
1027 len = sizeof(*di) + name_len + data_len;
1028 di = (struct btrfs_dir_item *)((char *)di + len);
1031 if (key->type == BTRFS_DIR_INDEX_KEY && nritems > 1)
1032 rec->errors |= I_ERR_DUP_DIR_INDEX;
1037 static int process_inode_ref(struct extent_buffer *eb,
1038 int slot, struct btrfs_key *key,
1039 struct shared_node *active_node)
1047 struct cache_tree *inode_cache;
1048 struct btrfs_inode_ref *ref;
1049 char namebuf[BTRFS_NAME_LEN];
1051 inode_cache = &active_node->inode_cache;
1053 ref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1054 total = btrfs_item_size_nr(eb, slot);
1055 while (cur < total) {
1056 name_len = btrfs_inode_ref_name_len(eb, ref);
1057 index = btrfs_inode_ref_index(eb, ref);
1058 if (name_len <= BTRFS_NAME_LEN) {
1062 len = BTRFS_NAME_LEN;
1063 error = REF_ERR_NAME_TOO_LONG;
1065 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
1066 add_inode_backref(inode_cache, key->objectid, key->offset,
1067 index, namebuf, len, 0, key->type, error);
1069 len = sizeof(*ref) + name_len;
1070 ref = (struct btrfs_inode_ref *)((char *)ref + len);
1076 static int process_inode_extref(struct extent_buffer *eb,
1077 int slot, struct btrfs_key *key,
1078 struct shared_node *active_node)
1087 struct cache_tree *inode_cache;
1088 struct btrfs_inode_extref *extref;
1089 char namebuf[BTRFS_NAME_LEN];
1091 inode_cache = &active_node->inode_cache;
1093 extref = btrfs_item_ptr(eb, slot, struct btrfs_inode_extref);
1094 total = btrfs_item_size_nr(eb, slot);
1095 while (cur < total) {
1096 name_len = btrfs_inode_extref_name_len(eb, extref);
1097 index = btrfs_inode_extref_index(eb, extref);
1098 parent = btrfs_inode_extref_parent(eb, extref);
1099 if (name_len <= BTRFS_NAME_LEN) {
1103 len = BTRFS_NAME_LEN;
1104 error = REF_ERR_NAME_TOO_LONG;
1106 read_extent_buffer(eb, namebuf,
1107 (unsigned long)(extref + 1), len);
1108 add_inode_backref(inode_cache, key->objectid, parent,
1109 index, namebuf, len, 0, key->type, error);
1111 len = sizeof(*extref) + name_len;
1112 extref = (struct btrfs_inode_extref *)((char *)extref + len);
1119 static int count_csum_range(struct btrfs_root *root, u64 start,
1120 u64 len, u64 *found)
1122 struct btrfs_key key;
1123 struct btrfs_path path;
1124 struct extent_buffer *leaf;
1129 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
1131 btrfs_init_path(&path);
1133 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1135 key.type = BTRFS_EXTENT_CSUM_KEY;
1137 ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
1141 if (ret > 0 && path.slots[0] > 0) {
1142 leaf = path.nodes[0];
1143 btrfs_item_key_to_cpu(leaf, &key, path.slots[0] - 1);
1144 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
1145 key.type == BTRFS_EXTENT_CSUM_KEY)
1150 leaf = path.nodes[0];
1151 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1152 ret = btrfs_next_leaf(root->fs_info->csum_root, &path);
1157 leaf = path.nodes[0];
1160 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1161 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1162 key.type != BTRFS_EXTENT_CSUM_KEY)
1165 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1166 if (key.offset >= start + len)
1169 if (key.offset > start)
1172 size = btrfs_item_size_nr(leaf, path.slots[0]);
1173 csum_end = key.offset + (size / csum_size) * root->sectorsize;
1174 if (csum_end > start) {
1175 size = min(csum_end - start, len);
1186 btrfs_release_path(&path);
1190 static int process_file_extent(struct btrfs_root *root,
1191 struct extent_buffer *eb,
1192 int slot, struct btrfs_key *key,
1193 struct shared_node *active_node)
1195 struct inode_record *rec;
1196 struct btrfs_file_extent_item *fi;
1198 u64 disk_bytenr = 0;
1199 u64 extent_offset = 0;
1200 u64 mask = root->sectorsize - 1;
1204 rec = active_node->current;
1205 BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1206 rec->found_file_extent = 1;
1208 if (rec->extent_start == (u64)-1) {
1209 rec->extent_start = key->offset;
1210 rec->extent_end = key->offset;
1213 if (rec->extent_end > key->offset)
1214 rec->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1215 else if (rec->extent_end < key->offset &&
1216 rec->extent_end < rec->first_extent_gap)
1217 rec->first_extent_gap = rec->extent_end;
1219 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
1220 extent_type = btrfs_file_extent_type(eb, fi);
1222 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1223 num_bytes = btrfs_file_extent_inline_len(eb, slot, fi);
1225 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1226 rec->found_size += num_bytes;
1227 num_bytes = (num_bytes + mask) & ~mask;
1228 } else if (extent_type == BTRFS_FILE_EXTENT_REG ||
1229 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1230 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1231 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1232 extent_offset = btrfs_file_extent_offset(eb, fi);
1233 if (num_bytes == 0 || (num_bytes & mask))
1234 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1235 if (num_bytes + extent_offset >
1236 btrfs_file_extent_ram_bytes(eb, fi))
1237 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1238 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC &&
1239 (btrfs_file_extent_compression(eb, fi) ||
1240 btrfs_file_extent_encryption(eb, fi) ||
1241 btrfs_file_extent_other_encoding(eb, fi)))
1242 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1243 if (disk_bytenr > 0)
1244 rec->found_size += num_bytes;
1246 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1248 rec->extent_end = key->offset + num_bytes;
1250 if (disk_bytenr > 0) {
1252 if (btrfs_file_extent_compression(eb, fi))
1253 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1255 disk_bytenr += extent_offset;
1257 ret = count_csum_range(root, disk_bytenr, num_bytes, &found);
1260 if (extent_type == BTRFS_FILE_EXTENT_REG) {
1262 rec->found_csum_item = 1;
1263 if (found < num_bytes)
1264 rec->some_csum_missing = 1;
1265 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1267 rec->errors |= I_ERR_ODD_CSUM_ITEM;
1273 static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
1274 struct walk_control *wc)
1276 struct btrfs_key key;
1280 struct cache_tree *inode_cache;
1281 struct shared_node *active_node;
1283 if (wc->root_level == wc->active_node &&
1284 btrfs_root_refs(&root->root_item) == 0)
1287 active_node = wc->nodes[wc->active_node];
1288 inode_cache = &active_node->inode_cache;
1289 nritems = btrfs_header_nritems(eb);
1290 for (i = 0; i < nritems; i++) {
1291 btrfs_item_key_to_cpu(eb, &key, i);
1293 if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
1295 if (key.type == BTRFS_ORPHAN_ITEM_KEY)
1298 if (active_node->current == NULL ||
1299 active_node->current->ino < key.objectid) {
1300 if (active_node->current) {
1301 active_node->current->checked = 1;
1302 maybe_free_inode_rec(inode_cache,
1303 active_node->current);
1305 active_node->current = get_inode_rec(inode_cache,
1309 case BTRFS_DIR_ITEM_KEY:
1310 case BTRFS_DIR_INDEX_KEY:
1311 ret = process_dir_item(root, eb, i, &key, active_node);
1313 case BTRFS_INODE_REF_KEY:
1314 ret = process_inode_ref(eb, i, &key, active_node);
1316 case BTRFS_INODE_EXTREF_KEY:
1317 ret = process_inode_extref(eb, i, &key, active_node);
1319 case BTRFS_INODE_ITEM_KEY:
1320 ret = process_inode_item(eb, i, &key, active_node);
1322 case BTRFS_EXTENT_DATA_KEY:
1323 ret = process_file_extent(root, eb, i, &key,
1333 static void reada_walk_down(struct btrfs_root *root,
1334 struct extent_buffer *node, int slot)
1343 level = btrfs_header_level(node);
1347 nritems = btrfs_header_nritems(node);
1348 blocksize = btrfs_level_size(root, level - 1);
1349 for (i = slot; i < nritems; i++) {
1350 bytenr = btrfs_node_blockptr(node, i);
1351 ptr_gen = btrfs_node_ptr_generation(node, i);
1352 readahead_tree_block(root, bytenr, blocksize, ptr_gen);
1357 * Check the child node/leaf by the following condition:
1358 * 1. the first item key of the node/leaf should be the same with the one
1360 * 2. block in parent node should match the child node/leaf.
1361 * 3. generation of parent node and child's header should be consistent.
1363 * Or the child node/leaf pointed by the key in parent is not valid.
1365 * We hope to check leaf owner too, but since subvol may share leaves,
1366 * which makes leaf owner check not so strong, key check should be
1367 * sufficient enough for that case.
1369 static int check_child_node(struct btrfs_root *root,
1370 struct extent_buffer *parent, int slot,
1371 struct extent_buffer *child)
1373 struct btrfs_key parent_key;
1374 struct btrfs_key child_key;
1377 btrfs_node_key_to_cpu(parent, &parent_key, slot);
1378 if (btrfs_header_level(child) == 0)
1379 btrfs_item_key_to_cpu(child, &child_key, 0);
1381 btrfs_node_key_to_cpu(child, &child_key, 0);
1383 if (memcmp(&parent_key, &child_key, sizeof(parent_key))) {
1386 "Wrong key of child node/leaf, wanted: (%llu, %u, %llu), have: (%llu, %u, %llu)\n",
1387 parent_key.objectid, parent_key.type, parent_key.offset,
1388 child_key.objectid, child_key.type, child_key.offset);
1390 if (btrfs_header_bytenr(child) != btrfs_node_blockptr(parent, slot)) {
1392 fprintf(stderr, "Wrong block of child node/leaf, wanted: %llu, have: %llu\n",
1393 btrfs_node_blockptr(parent, slot),
1394 btrfs_header_bytenr(child));
1396 if (btrfs_node_ptr_generation(parent, slot) !=
1397 btrfs_header_generation(child)) {
1399 fprintf(stderr, "Wrong generation of child node/leaf, wanted: %llu, have: %llu\n",
1400 btrfs_header_generation(child),
1401 btrfs_node_ptr_generation(parent, slot));
1406 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
1407 struct walk_control *wc, int *level)
1409 enum btrfs_tree_block_status status;
1412 struct extent_buffer *next;
1413 struct extent_buffer *cur;
1418 WARN_ON(*level < 0);
1419 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1420 ret = btrfs_lookup_extent_info(NULL, root,
1421 path->nodes[*level]->start,
1422 *level, 1, &refs, NULL);
1429 ret = enter_shared_node(root, path->nodes[*level]->start,
1437 while (*level >= 0) {
1438 WARN_ON(*level < 0);
1439 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1440 cur = path->nodes[*level];
1442 if (btrfs_header_level(cur) != *level)
1445 if (path->slots[*level] >= btrfs_header_nritems(cur))
1448 ret = process_one_leaf(root, cur, wc);
1453 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1454 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1455 blocksize = btrfs_level_size(root, *level - 1);
1456 ret = btrfs_lookup_extent_info(NULL, root, bytenr, *level - 1,
1462 ret = enter_shared_node(root, bytenr, refs,
1465 path->slots[*level]++;
1470 next = btrfs_find_tree_block(root, bytenr, blocksize);
1471 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
1472 free_extent_buffer(next);
1473 reada_walk_down(root, cur, path->slots[*level]);
1474 next = read_tree_block(root, bytenr, blocksize,
1482 ret = check_child_node(root, cur, path->slots[*level], next);
1488 if (btrfs_is_leaf(next))
1489 status = btrfs_check_leaf(root, NULL, next);
1491 status = btrfs_check_node(root, NULL, next);
1492 if (status != BTRFS_TREE_BLOCK_CLEAN) {
1493 free_extent_buffer(next);
1498 *level = *level - 1;
1499 free_extent_buffer(path->nodes[*level]);
1500 path->nodes[*level] = next;
1501 path->slots[*level] = 0;
1504 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
1508 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
1509 struct walk_control *wc, int *level)
1512 struct extent_buffer *leaf;
1514 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1515 leaf = path->nodes[i];
1516 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
1521 free_extent_buffer(path->nodes[*level]);
1522 path->nodes[*level] = NULL;
1523 BUG_ON(*level > wc->active_node);
1524 if (*level == wc->active_node)
1525 leave_shared_node(root, wc, *level);
1532 static int check_root_dir(struct inode_record *rec)
1534 struct inode_backref *backref;
1537 if (!rec->found_inode_item || rec->errors)
1539 if (rec->nlink != 1 || rec->found_link != 0)
1541 if (list_empty(&rec->backrefs))
1543 backref = list_entry(rec->backrefs.next, struct inode_backref, list);
1544 if (!backref->found_inode_ref)
1546 if (backref->index != 0 || backref->namelen != 2 ||
1547 memcmp(backref->name, "..", 2))
1549 if (backref->found_dir_index || backref->found_dir_item)
1556 static int repair_inode_isize(struct btrfs_trans_handle *trans,
1557 struct btrfs_root *root, struct btrfs_path *path,
1558 struct inode_record *rec)
1560 struct btrfs_inode_item *ei;
1561 struct btrfs_key key;
1564 key.objectid = rec->ino;
1565 key.type = BTRFS_INODE_ITEM_KEY;
1566 key.offset = (u64)-1;
1568 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1572 if (!path->slots[0]) {
1579 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1580 if (key.objectid != rec->ino) {
1585 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1586 struct btrfs_inode_item);
1587 btrfs_set_inode_size(path->nodes[0], ei, rec->found_size);
1588 btrfs_mark_buffer_dirty(path->nodes[0]);
1589 rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
1590 printf("reset isize for dir %Lu root %Lu\n", rec->ino,
1591 root->root_key.objectid);
1593 btrfs_release_path(path);
1597 static int repair_inode_orphan_item(struct btrfs_trans_handle *trans,
1598 struct btrfs_root *root,
1599 struct btrfs_path *path,
1600 struct inode_record *rec)
1602 struct btrfs_key key;
1605 key.objectid = BTRFS_ORPHAN_OBJECTID;
1606 key.type = BTRFS_ORPHAN_ITEM_KEY;
1607 key.offset = rec->ino;
1609 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1610 btrfs_release_path(path);
1612 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
1616 static int add_missing_dir_index(struct btrfs_root *root,
1617 struct cache_tree *inode_cache,
1618 struct inode_record *rec,
1619 struct inode_backref *backref)
1621 struct btrfs_path *path;
1622 struct btrfs_trans_handle *trans;
1623 struct btrfs_dir_item *dir_item;
1624 struct extent_buffer *leaf;
1625 struct btrfs_key key;
1626 struct btrfs_disk_key disk_key;
1627 struct inode_record *dir_rec;
1628 unsigned long name_ptr;
1629 u32 data_size = sizeof(*dir_item) + backref->namelen;
1632 path = btrfs_alloc_path();
1636 trans = btrfs_start_transaction(root, 1);
1637 if (IS_ERR(trans)) {
1638 btrfs_free_path(path);
1639 return PTR_ERR(trans);
1642 fprintf(stderr, "repairing missing dir index item for inode %llu\n",
1643 (unsigned long long)rec->ino);
1644 key.objectid = backref->dir;
1645 key.type = BTRFS_DIR_INDEX_KEY;
1646 key.offset = backref->index;
1648 ret = btrfs_insert_empty_item(trans, root, path, &key, data_size);
1651 leaf = path->nodes[0];
1652 dir_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
1654 disk_key.objectid = cpu_to_le64(rec->ino);
1655 disk_key.type = BTRFS_INODE_ITEM_KEY;
1656 disk_key.offset = 0;
1658 btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
1659 btrfs_set_dir_type(leaf, dir_item, imode_to_type(rec->imode));
1660 btrfs_set_dir_data_len(leaf, dir_item, 0);
1661 btrfs_set_dir_name_len(leaf, dir_item, backref->namelen);
1662 name_ptr = (unsigned long)(dir_item + 1);
1663 write_extent_buffer(leaf, backref->name, name_ptr, backref->namelen);
1664 btrfs_mark_buffer_dirty(leaf);
1665 btrfs_free_path(path);
1666 btrfs_commit_transaction(trans, root);
1668 backref->found_dir_index = 1;
1669 dir_rec = get_inode_rec(inode_cache, backref->dir, 0);
1672 dir_rec->found_size += backref->namelen;
1673 if (dir_rec->found_size == dir_rec->isize &&
1674 (dir_rec->errors & I_ERR_DIR_ISIZE_WRONG))
1675 dir_rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
1676 if (dir_rec->found_size != dir_rec->isize)
1677 dir_rec->errors |= I_ERR_DIR_ISIZE_WRONG;
1682 static int delete_dir_index(struct btrfs_root *root,
1683 struct cache_tree *inode_cache,
1684 struct inode_record *rec,
1685 struct inode_backref *backref)
1687 struct btrfs_trans_handle *trans;
1688 struct btrfs_dir_item *di;
1689 struct btrfs_path *path;
1692 path = btrfs_alloc_path();
1696 trans = btrfs_start_transaction(root, 1);
1697 if (IS_ERR(trans)) {
1698 btrfs_free_path(path);
1699 return PTR_ERR(trans);
1703 fprintf(stderr, "Deleting bad dir index [%llu,%u,%llu] root %llu\n",
1704 (unsigned long long)backref->dir,
1705 BTRFS_DIR_INDEX_KEY, (unsigned long long)backref->index,
1706 (unsigned long long)root->objectid);
1708 di = btrfs_lookup_dir_index(trans, root, path, backref->dir,
1709 backref->name, backref->namelen,
1710 backref->index, -1);
1713 btrfs_free_path(path);
1714 btrfs_commit_transaction(trans, root);
1721 ret = btrfs_del_item(trans, root, path);
1723 ret = btrfs_delete_one_dir_name(trans, root, path, di);
1725 btrfs_free_path(path);
1726 btrfs_commit_transaction(trans, root);
1730 static int repair_inode_backrefs(struct btrfs_root *root,
1731 struct inode_record *rec,
1732 struct cache_tree *inode_cache,
1735 struct inode_backref *tmp, *backref;
1736 u64 root_dirid = btrfs_root_dirid(&root->root_item);
1740 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
1741 /* Index 0 for root dir's are special, don't mess with it */
1742 if (rec->ino == root_dirid && backref->index == 0)
1746 ((backref->found_dir_index && !backref->found_inode_ref) ||
1747 (backref->found_dir_index && backref->found_inode_ref &&
1748 (backref->errors & REF_ERR_INDEX_UNMATCH)))) {
1749 ret = delete_dir_index(root, inode_cache, rec, backref);
1753 list_del(&backref->list);
1757 if (!delete && !backref->found_dir_index &&
1758 backref->found_dir_item && backref->found_inode_ref) {
1759 ret = add_missing_dir_index(root, inode_cache, rec,
1764 if (backref->found_dir_item &&
1765 backref->found_dir_index &&
1766 backref->found_dir_index) {
1767 if (!backref->errors &&
1768 backref->found_inode_ref) {
1769 list_del(&backref->list);
1775 if (!delete && (!backref->found_dir_index &&
1776 !backref->found_dir_item &&
1777 backref->found_inode_ref)) {
1778 struct btrfs_trans_handle *trans;
1779 struct btrfs_key location;
1781 location.objectid = rec->ino;
1782 location.type = BTRFS_INODE_ITEM_KEY;
1783 location.offset = 0;
1785 trans = btrfs_start_transaction(root, 1);
1786 if (IS_ERR(trans)) {
1787 ret = PTR_ERR(trans);
1790 fprintf(stderr, "adding missing dir index/item pair "
1792 (unsigned long long)rec->ino);
1793 ret = btrfs_insert_dir_item(trans, root, backref->name,
1795 backref->dir, &location,
1796 imode_to_type(rec->imode),
1799 btrfs_commit_transaction(trans, root);
1803 return ret ? ret : repaired;
1806 static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
1808 struct btrfs_trans_handle *trans;
1809 struct btrfs_path *path;
1812 if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG | I_ERR_NO_ORPHAN_ITEM)))
1815 path = btrfs_alloc_path();
1819 trans = btrfs_start_transaction(root, 1);
1820 if (IS_ERR(trans)) {
1821 btrfs_free_path(path);
1822 return PTR_ERR(trans);
1825 if (rec->errors & I_ERR_DIR_ISIZE_WRONG)
1826 ret = repair_inode_isize(trans, root, path, rec);
1827 if (!ret && rec->errors & I_ERR_NO_ORPHAN_ITEM)
1828 ret = repair_inode_orphan_item(trans, root, path, rec);
1829 btrfs_commit_transaction(trans, root);
1830 btrfs_free_path(path);
1834 static int check_inode_recs(struct btrfs_root *root,
1835 struct cache_tree *inode_cache)
1837 struct cache_extent *cache;
1838 struct ptr_node *node;
1839 struct inode_record *rec;
1840 struct inode_backref *backref;
1845 u64 root_dirid = btrfs_root_dirid(&root->root_item);
1847 if (btrfs_root_refs(&root->root_item) == 0) {
1848 if (!cache_tree_empty(inode_cache))
1849 fprintf(stderr, "warning line %d\n", __LINE__);
1854 * We need to repair backrefs first because we could change some of the
1855 * errors in the inode recs.
1857 * We also need to go through and delete invalid backrefs first and then
1858 * add the correct ones second. We do this because we may get EEXIST
1859 * when adding back the correct index because we hadn't yet deleted the
1862 * For example, if we were missing a dir index then the directories
1863 * isize would be wrong, so if we fixed the isize to what we thought it
1864 * would be and then fixed the backref we'd still have a invalid fs, so
1865 * we need to add back the dir index and then check to see if the isize
1870 if (stage == 3 && !err)
1873 cache = search_cache_extent(inode_cache, 0);
1874 while (repair && cache) {
1875 node = container_of(cache, struct ptr_node, cache);
1877 cache = next_cache_extent(cache);
1879 /* Need to free everything up and rescan */
1881 remove_cache_extent(inode_cache, &node->cache);
1883 free_inode_rec(rec);
1887 if (list_empty(&rec->backrefs))
1890 ret = repair_inode_backrefs(root, rec, inode_cache,
1904 rec = get_inode_rec(inode_cache, root_dirid, 0);
1906 ret = check_root_dir(rec);
1908 fprintf(stderr, "root %llu root dir %llu error\n",
1909 (unsigned long long)root->root_key.objectid,
1910 (unsigned long long)root_dirid);
1914 fprintf(stderr, "root %llu root dir %llu not found\n",
1915 (unsigned long long)root->root_key.objectid,
1916 (unsigned long long)root_dirid);
1920 cache = search_cache_extent(inode_cache, 0);
1923 node = container_of(cache, struct ptr_node, cache);
1925 remove_cache_extent(inode_cache, &node->cache);
1927 if (rec->ino == root_dirid ||
1928 rec->ino == BTRFS_ORPHAN_OBJECTID) {
1929 free_inode_rec(rec);
1933 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
1934 ret = check_orphan_item(root, rec->ino);
1936 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
1937 if (can_free_inode_rec(rec)) {
1938 free_inode_rec(rec);
1944 ret = try_repair_inode(root, rec);
1945 if (ret == 0 && can_free_inode_rec(rec)) {
1946 free_inode_rec(rec);
1953 if (!rec->found_inode_item)
1954 rec->errors |= I_ERR_NO_INODE_ITEM;
1955 if (rec->found_link != rec->nlink)
1956 rec->errors |= I_ERR_LINK_COUNT_WRONG;
1957 print_inode_error(root, rec);
1958 list_for_each_entry(backref, &rec->backrefs, list) {
1959 if (!backref->found_dir_item)
1960 backref->errors |= REF_ERR_NO_DIR_ITEM;
1961 if (!backref->found_dir_index)
1962 backref->errors |= REF_ERR_NO_DIR_INDEX;
1963 if (!backref->found_inode_ref)
1964 backref->errors |= REF_ERR_NO_INODE_REF;
1965 fprintf(stderr, "\tunresolved ref dir %llu index %llu"
1966 " namelen %u name %s filetype %d errors %x",
1967 (unsigned long long)backref->dir,
1968 (unsigned long long)backref->index,
1969 backref->namelen, backref->name,
1970 backref->filetype, backref->errors);
1971 print_ref_error(backref->errors);
1973 free_inode_rec(rec);
1975 return (error > 0) ? -1 : 0;
1978 static struct root_record *get_root_rec(struct cache_tree *root_cache,
1981 struct cache_extent *cache;
1982 struct root_record *rec = NULL;
1985 cache = lookup_cache_extent(root_cache, objectid, 1);
1987 rec = container_of(cache, struct root_record, cache);
1989 rec = calloc(1, sizeof(*rec));
1990 rec->objectid = objectid;
1991 INIT_LIST_HEAD(&rec->backrefs);
1992 rec->cache.start = objectid;
1993 rec->cache.size = 1;
1995 ret = insert_cache_extent(root_cache, &rec->cache);
2001 static struct root_backref *get_root_backref(struct root_record *rec,
2002 u64 ref_root, u64 dir, u64 index,
2003 const char *name, int namelen)
2005 struct root_backref *backref;
2007 list_for_each_entry(backref, &rec->backrefs, list) {
2008 if (backref->ref_root != ref_root || backref->dir != dir ||
2009 backref->namelen != namelen)
2011 if (memcmp(name, backref->name, namelen))
2016 backref = malloc(sizeof(*backref) + namelen + 1);
2017 memset(backref, 0, sizeof(*backref));
2018 backref->ref_root = ref_root;
2020 backref->index = index;
2021 backref->namelen = namelen;
2022 memcpy(backref->name, name, namelen);
2023 backref->name[namelen] = '\0';
2024 list_add_tail(&backref->list, &rec->backrefs);
2028 static void free_root_record(struct cache_extent *cache)
2030 struct root_record *rec;
2031 struct root_backref *backref;
2033 rec = container_of(cache, struct root_record, cache);
2034 while (!list_empty(&rec->backrefs)) {
2035 backref = list_entry(rec->backrefs.next,
2036 struct root_backref, list);
2037 list_del(&backref->list);
2044 FREE_EXTENT_CACHE_BASED_TREE(root_recs, free_root_record);
2046 static int add_root_backref(struct cache_tree *root_cache,
2047 u64 root_id, u64 ref_root, u64 dir, u64 index,
2048 const char *name, int namelen,
2049 int item_type, int errors)
2051 struct root_record *rec;
2052 struct root_backref *backref;
2054 rec = get_root_rec(root_cache, root_id);
2055 backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
2057 backref->errors |= errors;
2059 if (item_type != BTRFS_DIR_ITEM_KEY) {
2060 if (backref->found_dir_index || backref->found_back_ref ||
2061 backref->found_forward_ref) {
2062 if (backref->index != index)
2063 backref->errors |= REF_ERR_INDEX_UNMATCH;
2065 backref->index = index;
2069 if (item_type == BTRFS_DIR_ITEM_KEY) {
2070 if (backref->found_forward_ref)
2072 backref->found_dir_item = 1;
2073 } else if (item_type == BTRFS_DIR_INDEX_KEY) {
2074 backref->found_dir_index = 1;
2075 } else if (item_type == BTRFS_ROOT_REF_KEY) {
2076 if (backref->found_forward_ref)
2077 backref->errors |= REF_ERR_DUP_ROOT_REF;
2078 else if (backref->found_dir_item)
2080 backref->found_forward_ref = 1;
2081 } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
2082 if (backref->found_back_ref)
2083 backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
2084 backref->found_back_ref = 1;
2089 if (backref->found_forward_ref && backref->found_dir_item)
2090 backref->reachable = 1;
2094 static int merge_root_recs(struct btrfs_root *root,
2095 struct cache_tree *src_cache,
2096 struct cache_tree *dst_cache)
2098 struct cache_extent *cache;
2099 struct ptr_node *node;
2100 struct inode_record *rec;
2101 struct inode_backref *backref;
2104 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2105 free_inode_recs_tree(src_cache);
2110 cache = search_cache_extent(src_cache, 0);
2113 node = container_of(cache, struct ptr_node, cache);
2115 remove_cache_extent(src_cache, &node->cache);
2118 ret = is_child_root(root, root->objectid, rec->ino);
2124 list_for_each_entry(backref, &rec->backrefs, list) {
2125 BUG_ON(backref->found_inode_ref);
2126 if (backref->found_dir_item)
2127 add_root_backref(dst_cache, rec->ino,
2128 root->root_key.objectid, backref->dir,
2129 backref->index, backref->name,
2130 backref->namelen, BTRFS_DIR_ITEM_KEY,
2132 if (backref->found_dir_index)
2133 add_root_backref(dst_cache, rec->ino,
2134 root->root_key.objectid, backref->dir,
2135 backref->index, backref->name,
2136 backref->namelen, BTRFS_DIR_INDEX_KEY,
2140 free_inode_rec(rec);
2147 static int check_root_refs(struct btrfs_root *root,
2148 struct cache_tree *root_cache)
2150 struct root_record *rec;
2151 struct root_record *ref_root;
2152 struct root_backref *backref;
2153 struct cache_extent *cache;
2159 rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
2162 /* fixme: this can not detect circular references */
2165 cache = search_cache_extent(root_cache, 0);
2169 rec = container_of(cache, struct root_record, cache);
2170 cache = next_cache_extent(cache);
2172 if (rec->found_ref == 0)
2175 list_for_each_entry(backref, &rec->backrefs, list) {
2176 if (!backref->reachable)
2179 ref_root = get_root_rec(root_cache,
2181 if (ref_root->found_ref > 0)
2184 backref->reachable = 0;
2186 if (rec->found_ref == 0)
2192 cache = search_cache_extent(root_cache, 0);
2196 rec = container_of(cache, struct root_record, cache);
2197 cache = next_cache_extent(cache);
2199 if (rec->found_ref == 0 &&
2200 rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
2201 rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
2202 ret = check_orphan_item(root->fs_info->tree_root,
2208 * If we don't have a root item then we likely just have
2209 * a dir item in a snapshot for this root but no actual
2210 * ref key or anything so it's meaningless.
2212 if (!rec->found_root_item)
2215 fprintf(stderr, "fs tree %llu not referenced\n",
2216 (unsigned long long)rec->objectid);
2220 if (rec->found_ref > 0 && !rec->found_root_item)
2222 list_for_each_entry(backref, &rec->backrefs, list) {
2223 if (!backref->found_dir_item)
2224 backref->errors |= REF_ERR_NO_DIR_ITEM;
2225 if (!backref->found_dir_index)
2226 backref->errors |= REF_ERR_NO_DIR_INDEX;
2227 if (!backref->found_back_ref)
2228 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
2229 if (!backref->found_forward_ref)
2230 backref->errors |= REF_ERR_NO_ROOT_REF;
2231 if (backref->reachable && backref->errors)
2238 fprintf(stderr, "fs tree %llu refs %u %s\n",
2239 (unsigned long long)rec->objectid, rec->found_ref,
2240 rec->found_root_item ? "" : "not found");
2242 list_for_each_entry(backref, &rec->backrefs, list) {
2243 if (!backref->reachable)
2245 if (!backref->errors && rec->found_root_item)
2247 fprintf(stderr, "\tunresolved ref root %llu dir %llu"
2248 " index %llu namelen %u name %s errors %x\n",
2249 (unsigned long long)backref->ref_root,
2250 (unsigned long long)backref->dir,
2251 (unsigned long long)backref->index,
2252 backref->namelen, backref->name,
2254 print_ref_error(backref->errors);
2257 return errors > 0 ? 1 : 0;
2260 static int process_root_ref(struct extent_buffer *eb, int slot,
2261 struct btrfs_key *key,
2262 struct cache_tree *root_cache)
2268 struct btrfs_root_ref *ref;
2269 char namebuf[BTRFS_NAME_LEN];
2272 ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
2274 dirid = btrfs_root_ref_dirid(eb, ref);
2275 index = btrfs_root_ref_sequence(eb, ref);
2276 name_len = btrfs_root_ref_name_len(eb, ref);
2278 if (name_len <= BTRFS_NAME_LEN) {
2282 len = BTRFS_NAME_LEN;
2283 error = REF_ERR_NAME_TOO_LONG;
2285 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
2287 if (key->type == BTRFS_ROOT_REF_KEY) {
2288 add_root_backref(root_cache, key->offset, key->objectid, dirid,
2289 index, namebuf, len, key->type, error);
2291 add_root_backref(root_cache, key->objectid, key->offset, dirid,
2292 index, namebuf, len, key->type, error);
2297 static int check_fs_root(struct btrfs_root *root,
2298 struct cache_tree *root_cache,
2299 struct walk_control *wc)
2305 struct btrfs_path path;
2306 struct shared_node root_node;
2307 struct root_record *rec;
2308 struct btrfs_root_item *root_item = &root->root_item;
2309 enum btrfs_tree_block_status status;
2311 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2312 rec = get_root_rec(root_cache, root->root_key.objectid);
2313 if (btrfs_root_refs(root_item) > 0)
2314 rec->found_root_item = 1;
2317 btrfs_init_path(&path);
2318 memset(&root_node, 0, sizeof(root_node));
2319 cache_tree_init(&root_node.root_cache);
2320 cache_tree_init(&root_node.inode_cache);
2322 level = btrfs_header_level(root->node);
2323 memset(wc->nodes, 0, sizeof(wc->nodes));
2324 wc->nodes[level] = &root_node;
2325 wc->active_node = level;
2326 wc->root_level = level;
2328 /* We may not have checked the root block, lets do that now */
2329 if (btrfs_is_leaf(root->node))
2330 status = btrfs_check_leaf(root, NULL, root->node);
2332 status = btrfs_check_node(root, NULL, root->node);
2333 if (status != BTRFS_TREE_BLOCK_CLEAN)
2336 if (btrfs_root_refs(root_item) > 0 ||
2337 btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2338 path.nodes[level] = root->node;
2339 extent_buffer_get(root->node);
2340 path.slots[level] = 0;
2342 struct btrfs_key key;
2343 struct btrfs_disk_key found_key;
2345 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2346 level = root_item->drop_level;
2347 path.lowest_level = level;
2348 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
2351 btrfs_node_key(path.nodes[level], &found_key,
2353 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2354 sizeof(found_key)));
2358 wret = walk_down_tree(root, &path, wc, &level);
2364 wret = walk_up_tree(root, &path, wc, &level);
2371 btrfs_release_path(&path);
2373 err = merge_root_recs(root, &root_node.root_cache, root_cache);
2377 if (root_node.current) {
2378 root_node.current->checked = 1;
2379 maybe_free_inode_rec(&root_node.inode_cache,
2383 err = check_inode_recs(root, &root_node.inode_cache);
2389 static int fs_root_objectid(u64 objectid)
2391 if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
2392 objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2394 return is_fstree(objectid);
2397 static int check_fs_roots(struct btrfs_root *root,
2398 struct cache_tree *root_cache)
2400 struct btrfs_path path;
2401 struct btrfs_key key;
2402 struct walk_control wc;
2403 struct extent_buffer *leaf, *tree_node;
2404 struct btrfs_root *tmp_root;
2405 struct btrfs_root *tree_root = root->fs_info->tree_root;
2410 * Just in case we made any changes to the extent tree that weren't
2411 * reflected into the free space cache yet.
2414 reset_cached_block_groups(root->fs_info);
2415 memset(&wc, 0, sizeof(wc));
2416 cache_tree_init(&wc.shared);
2417 btrfs_init_path(&path);
2422 key.type = BTRFS_ROOT_ITEM_KEY;
2423 ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
2428 tree_node = tree_root->node;
2430 if (tree_node != tree_root->node) {
2431 free_root_recs_tree(root_cache);
2432 btrfs_release_path(&path);
2435 leaf = path.nodes[0];
2436 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
2437 ret = btrfs_next_leaf(tree_root, &path);
2443 leaf = path.nodes[0];
2445 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
2446 if (key.type == BTRFS_ROOT_ITEM_KEY &&
2447 fs_root_objectid(key.objectid)) {
2448 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2449 tmp_root = btrfs_read_fs_root_no_cache(
2450 root->fs_info, &key);
2452 key.offset = (u64)-1;
2453 tmp_root = btrfs_read_fs_root(
2454 root->fs_info, &key);
2456 if (IS_ERR(tmp_root)) {
2460 ret = check_fs_root(tmp_root, root_cache, &wc);
2461 if (ret == -EAGAIN) {
2462 free_root_recs_tree(root_cache);
2463 btrfs_release_path(&path);
2468 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
2469 btrfs_free_fs_root(tmp_root);
2470 } else if (key.type == BTRFS_ROOT_REF_KEY ||
2471 key.type == BTRFS_ROOT_BACKREF_KEY) {
2472 process_root_ref(leaf, path.slots[0], &key,
2479 btrfs_release_path(&path);
2481 free_extent_cache_tree(&wc.shared);
2482 if (!cache_tree_empty(&wc.shared))
2483 fprintf(stderr, "warning line %d\n", __LINE__);
2488 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
2490 struct list_head *cur = rec->backrefs.next;
2491 struct extent_backref *back;
2492 struct tree_backref *tback;
2493 struct data_backref *dback;
2497 while(cur != &rec->backrefs) {
2498 back = list_entry(cur, struct extent_backref, list);
2500 if (!back->found_extent_tree) {
2504 if (back->is_data) {
2505 dback = (struct data_backref *)back;
2506 fprintf(stderr, "Backref %llu %s %llu"
2507 " owner %llu offset %llu num_refs %lu"
2508 " not found in extent tree\n",
2509 (unsigned long long)rec->start,
2510 back->full_backref ?
2512 back->full_backref ?
2513 (unsigned long long)dback->parent:
2514 (unsigned long long)dback->root,
2515 (unsigned long long)dback->owner,
2516 (unsigned long long)dback->offset,
2517 (unsigned long)dback->num_refs);
2519 tback = (struct tree_backref *)back;
2520 fprintf(stderr, "Backref %llu parent %llu"
2521 " root %llu not found in extent tree\n",
2522 (unsigned long long)rec->start,
2523 (unsigned long long)tback->parent,
2524 (unsigned long long)tback->root);
2527 if (!back->is_data && !back->found_ref) {
2531 tback = (struct tree_backref *)back;
2532 fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
2533 (unsigned long long)rec->start,
2534 back->full_backref ? "parent" : "root",
2535 back->full_backref ?
2536 (unsigned long long)tback->parent :
2537 (unsigned long long)tback->root, back);
2539 if (back->is_data) {
2540 dback = (struct data_backref *)back;
2541 if (dback->found_ref != dback->num_refs) {
2545 fprintf(stderr, "Incorrect local backref count"
2546 " on %llu %s %llu owner %llu"
2547 " offset %llu found %u wanted %u back %p\n",
2548 (unsigned long long)rec->start,
2549 back->full_backref ?
2551 back->full_backref ?
2552 (unsigned long long)dback->parent:
2553 (unsigned long long)dback->root,
2554 (unsigned long long)dback->owner,
2555 (unsigned long long)dback->offset,
2556 dback->found_ref, dback->num_refs, back);
2558 if (dback->disk_bytenr != rec->start) {
2562 fprintf(stderr, "Backref disk bytenr does not"
2563 " match extent record, bytenr=%llu, "
2564 "ref bytenr=%llu\n",
2565 (unsigned long long)rec->start,
2566 (unsigned long long)dback->disk_bytenr);
2569 if (dback->bytes != rec->nr) {
2573 fprintf(stderr, "Backref bytes do not match "
2574 "extent backref, bytenr=%llu, ref "
2575 "bytes=%llu, backref bytes=%llu\n",
2576 (unsigned long long)rec->start,
2577 (unsigned long long)rec->nr,
2578 (unsigned long long)dback->bytes);
2581 if (!back->is_data) {
2584 dback = (struct data_backref *)back;
2585 found += dback->found_ref;
2588 if (found != rec->refs) {
2592 fprintf(stderr, "Incorrect global backref count "
2593 "on %llu found %llu wanted %llu\n",
2594 (unsigned long long)rec->start,
2595 (unsigned long long)found,
2596 (unsigned long long)rec->refs);
2602 static int free_all_extent_backrefs(struct extent_record *rec)
2604 struct extent_backref *back;
2605 struct list_head *cur;
2606 while (!list_empty(&rec->backrefs)) {
2607 cur = rec->backrefs.next;
2608 back = list_entry(cur, struct extent_backref, list);
2615 static void free_extent_record_cache(struct btrfs_fs_info *fs_info,
2616 struct cache_tree *extent_cache)
2618 struct cache_extent *cache;
2619 struct extent_record *rec;
2622 cache = first_cache_extent(extent_cache);
2625 rec = container_of(cache, struct extent_record, cache);
2626 btrfs_unpin_extent(fs_info, rec->start, rec->max_size);
2627 remove_cache_extent(extent_cache, cache);
2628 free_all_extent_backrefs(rec);
2633 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
2634 struct extent_record *rec)
2636 if (rec->content_checked && rec->owner_ref_checked &&
2637 rec->extent_item_refs == rec->refs && rec->refs > 0 &&
2638 rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0)) {
2639 remove_cache_extent(extent_cache, &rec->cache);
2640 free_all_extent_backrefs(rec);
2641 list_del_init(&rec->list);
2647 static int check_owner_ref(struct btrfs_root *root,
2648 struct extent_record *rec,
2649 struct extent_buffer *buf)
2651 struct extent_backref *node;
2652 struct tree_backref *back;
2653 struct btrfs_root *ref_root;
2654 struct btrfs_key key;
2655 struct btrfs_path path;
2656 struct extent_buffer *parent;
2661 list_for_each_entry(node, &rec->backrefs, list) {
2664 if (!node->found_ref)
2666 if (node->full_backref)
2668 back = (struct tree_backref *)node;
2669 if (btrfs_header_owner(buf) == back->root)
2672 BUG_ON(rec->is_root);
2674 /* try to find the block by search corresponding fs tree */
2675 key.objectid = btrfs_header_owner(buf);
2676 key.type = BTRFS_ROOT_ITEM_KEY;
2677 key.offset = (u64)-1;
2679 ref_root = btrfs_read_fs_root(root->fs_info, &key);
2680 if (IS_ERR(ref_root))
2683 level = btrfs_header_level(buf);
2685 btrfs_item_key_to_cpu(buf, &key, 0);
2687 btrfs_node_key_to_cpu(buf, &key, 0);
2689 btrfs_init_path(&path);
2690 path.lowest_level = level + 1;
2691 ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
2695 parent = path.nodes[level + 1];
2696 if (parent && buf->start == btrfs_node_blockptr(parent,
2697 path.slots[level + 1]))
2700 btrfs_release_path(&path);
2701 return found ? 0 : 1;
2704 static int is_extent_tree_record(struct extent_record *rec)
2706 struct list_head *cur = rec->backrefs.next;
2707 struct extent_backref *node;
2708 struct tree_backref *back;
2711 while(cur != &rec->backrefs) {
2712 node = list_entry(cur, struct extent_backref, list);
2716 back = (struct tree_backref *)node;
2717 if (node->full_backref)
2719 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
2726 static int record_bad_block_io(struct btrfs_fs_info *info,
2727 struct cache_tree *extent_cache,
2730 struct extent_record *rec;
2731 struct cache_extent *cache;
2732 struct btrfs_key key;
2734 cache = lookup_cache_extent(extent_cache, start, len);
2738 rec = container_of(cache, struct extent_record, cache);
2739 if (!is_extent_tree_record(rec))
2742 btrfs_disk_key_to_cpu(&key, &rec->parent_key);
2743 return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
2746 static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
2747 struct extent_buffer *buf, int slot)
2749 if (btrfs_header_level(buf)) {
2750 struct btrfs_key_ptr ptr1, ptr2;
2752 read_extent_buffer(buf, &ptr1, btrfs_node_key_ptr_offset(slot),
2753 sizeof(struct btrfs_key_ptr));
2754 read_extent_buffer(buf, &ptr2,
2755 btrfs_node_key_ptr_offset(slot + 1),
2756 sizeof(struct btrfs_key_ptr));
2757 write_extent_buffer(buf, &ptr1,
2758 btrfs_node_key_ptr_offset(slot + 1),
2759 sizeof(struct btrfs_key_ptr));
2760 write_extent_buffer(buf, &ptr2,
2761 btrfs_node_key_ptr_offset(slot),
2762 sizeof(struct btrfs_key_ptr));
2764 struct btrfs_disk_key key;
2765 btrfs_node_key(buf, &key, 0);
2766 btrfs_fixup_low_keys(root, path, &key,
2767 btrfs_header_level(buf) + 1);
2770 struct btrfs_item *item1, *item2;
2771 struct btrfs_key k1, k2;
2772 char *item1_data, *item2_data;
2773 u32 item1_offset, item2_offset, item1_size, item2_size;
2775 item1 = btrfs_item_nr(slot);
2776 item2 = btrfs_item_nr(slot + 1);
2777 btrfs_item_key_to_cpu(buf, &k1, slot);
2778 btrfs_item_key_to_cpu(buf, &k2, slot + 1);
2779 item1_offset = btrfs_item_offset(buf, item1);
2780 item2_offset = btrfs_item_offset(buf, item2);
2781 item1_size = btrfs_item_size(buf, item1);
2782 item2_size = btrfs_item_size(buf, item2);
2784 item1_data = malloc(item1_size);
2787 item2_data = malloc(item2_size);
2793 read_extent_buffer(buf, item1_data, item1_offset, item1_size);
2794 read_extent_buffer(buf, item2_data, item2_offset, item2_size);
2796 write_extent_buffer(buf, item1_data, item2_offset, item2_size);
2797 write_extent_buffer(buf, item2_data, item1_offset, item1_size);
2801 btrfs_set_item_offset(buf, item1, item2_offset);
2802 btrfs_set_item_offset(buf, item2, item1_offset);
2803 btrfs_set_item_size(buf, item1, item2_size);
2804 btrfs_set_item_size(buf, item2, item1_size);
2806 path->slots[0] = slot;
2807 btrfs_set_item_key_unsafe(root, path, &k2);
2808 path->slots[0] = slot + 1;
2809 btrfs_set_item_key_unsafe(root, path, &k1);
2814 static int fix_key_order(struct btrfs_trans_handle *trans,
2815 struct btrfs_root *root,
2816 struct btrfs_path *path)
2818 struct extent_buffer *buf;
2819 struct btrfs_key k1, k2;
2821 int level = path->lowest_level;
2824 buf = path->nodes[level];
2825 for (i = 0; i < btrfs_header_nritems(buf) - 1; i++) {
2827 btrfs_node_key_to_cpu(buf, &k1, i);
2828 btrfs_node_key_to_cpu(buf, &k2, i + 1);
2830 btrfs_item_key_to_cpu(buf, &k1, i);
2831 btrfs_item_key_to_cpu(buf, &k2, i + 1);
2833 if (btrfs_comp_cpu_keys(&k1, &k2) < 0)
2835 ret = swap_values(root, path, buf, i);
2838 btrfs_mark_buffer_dirty(buf);
2844 static int delete_bogus_item(struct btrfs_trans_handle *trans,
2845 struct btrfs_root *root,
2846 struct btrfs_path *path,
2847 struct extent_buffer *buf, int slot)
2849 struct btrfs_key key;
2850 int nritems = btrfs_header_nritems(buf);
2852 btrfs_item_key_to_cpu(buf, &key, slot);
2854 /* These are all the keys we can deal with missing. */
2855 if (key.type != BTRFS_DIR_INDEX_KEY &&
2856 key.type != BTRFS_EXTENT_ITEM_KEY &&
2857 key.type != BTRFS_METADATA_ITEM_KEY &&
2858 key.type != BTRFS_TREE_BLOCK_REF_KEY &&
2859 key.type != BTRFS_EXTENT_DATA_REF_KEY)
2862 printf("Deleting bogus item [%llu,%u,%llu] at slot %d on block %llu\n",
2863 (unsigned long long)key.objectid, key.type,
2864 (unsigned long long)key.offset, slot, buf->start);
2865 memmove_extent_buffer(buf, btrfs_item_nr_offset(slot),
2866 btrfs_item_nr_offset(slot + 1),
2867 sizeof(struct btrfs_item) *
2868 (nritems - slot - 1));
2869 btrfs_set_header_nritems(buf, nritems - 1);
2871 struct btrfs_disk_key disk_key;
2873 btrfs_item_key(buf, &disk_key, 0);
2874 btrfs_fixup_low_keys(root, path, &disk_key, 1);
2876 btrfs_mark_buffer_dirty(buf);
2880 static int fix_item_offset(struct btrfs_trans_handle *trans,
2881 struct btrfs_root *root,
2882 struct btrfs_path *path)
2884 struct extent_buffer *buf;
2888 /* We should only get this for leaves */
2889 BUG_ON(path->lowest_level);
2890 buf = path->nodes[0];
2892 for (i = 0; i < btrfs_header_nritems(buf); i++) {
2893 unsigned int shift = 0, offset;
2895 if (i == 0 && btrfs_item_end_nr(buf, i) !=
2896 BTRFS_LEAF_DATA_SIZE(root)) {
2897 if (btrfs_item_end_nr(buf, i) >
2898 BTRFS_LEAF_DATA_SIZE(root)) {
2899 ret = delete_bogus_item(trans, root, path,
2903 fprintf(stderr, "item is off the end of the "
2904 "leaf, can't fix\n");
2908 shift = BTRFS_LEAF_DATA_SIZE(root) -
2909 btrfs_item_end_nr(buf, i);
2910 } else if (i > 0 && btrfs_item_end_nr(buf, i) !=
2911 btrfs_item_offset_nr(buf, i - 1)) {
2912 if (btrfs_item_end_nr(buf, i) >
2913 btrfs_item_offset_nr(buf, i - 1)) {
2914 ret = delete_bogus_item(trans, root, path,
2918 fprintf(stderr, "items overlap, can't fix\n");
2922 shift = btrfs_item_offset_nr(buf, i - 1) -
2923 btrfs_item_end_nr(buf, i);
2928 printf("Shifting item nr %d by %u bytes in block %llu\n",
2929 i, shift, (unsigned long long)buf->start);
2930 offset = btrfs_item_offset_nr(buf, i);
2931 memmove_extent_buffer(buf,
2932 btrfs_leaf_data(buf) + offset + shift,
2933 btrfs_leaf_data(buf) + offset,
2934 btrfs_item_size_nr(buf, i));
2935 btrfs_set_item_offset(buf, btrfs_item_nr(i),
2937 btrfs_mark_buffer_dirty(buf);
2941 * We may have moved things, in which case we want to exit so we don't
2942 * write those changes out. Once we have proper abort functionality in
2943 * progs this can be changed to something nicer.
2950 * Attempt to fix basic block failures. If we can't fix it for whatever reason
2951 * then just return -EIO.
2953 static int try_to_fix_bad_block(struct btrfs_trans_handle *trans,
2954 struct btrfs_root *root,
2955 struct extent_buffer *buf,
2956 enum btrfs_tree_block_status status)
2958 struct ulist *roots;
2959 struct ulist_node *node;
2960 struct btrfs_root *search_root;
2961 struct btrfs_path *path;
2962 struct ulist_iterator iter;
2963 struct btrfs_key root_key, key;
2966 if (status != BTRFS_TREE_BLOCK_BAD_KEY_ORDER &&
2967 status != BTRFS_TREE_BLOCK_INVALID_OFFSETS)
2970 path = btrfs_alloc_path();
2974 ret = btrfs_find_all_roots(trans, root->fs_info, buf->start,
2977 btrfs_free_path(path);
2981 ULIST_ITER_INIT(&iter);
2982 while ((node = ulist_next(roots, &iter))) {
2983 root_key.objectid = node->val;
2984 root_key.type = BTRFS_ROOT_ITEM_KEY;
2985 root_key.offset = (u64)-1;
2987 search_root = btrfs_read_fs_root(root->fs_info, &root_key);
2993 record_root_in_trans(trans, search_root);
2995 path->lowest_level = btrfs_header_level(buf);
2996 path->skip_check_block = 1;
2997 if (path->lowest_level)
2998 btrfs_node_key_to_cpu(buf, &key, 0);
3000 btrfs_item_key_to_cpu(buf, &key, 0);
3001 ret = btrfs_search_slot(trans, search_root, &key, path, 0, 1);
3006 if (status == BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
3007 ret = fix_key_order(trans, search_root, path);
3008 else if (status == BTRFS_TREE_BLOCK_INVALID_OFFSETS)
3009 ret = fix_item_offset(trans, search_root, path);
3012 btrfs_release_path(path);
3015 btrfs_free_path(path);
3019 static int check_block(struct btrfs_trans_handle *trans,
3020 struct btrfs_root *root,
3021 struct cache_tree *extent_cache,
3022 struct extent_buffer *buf, u64 flags)
3024 struct extent_record *rec;
3025 struct cache_extent *cache;
3026 struct btrfs_key key;
3027 enum btrfs_tree_block_status status;
3031 cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
3034 rec = container_of(cache, struct extent_record, cache);
3035 rec->generation = btrfs_header_generation(buf);
3037 level = btrfs_header_level(buf);
3038 if (btrfs_header_nritems(buf) > 0) {
3041 btrfs_item_key_to_cpu(buf, &key, 0);
3043 btrfs_node_key_to_cpu(buf, &key, 0);
3045 rec->info_objectid = key.objectid;
3047 rec->info_level = level;
3049 if (btrfs_is_leaf(buf))
3050 status = btrfs_check_leaf(root, &rec->parent_key, buf);
3052 status = btrfs_check_node(root, &rec->parent_key, buf);
3054 if (status != BTRFS_TREE_BLOCK_CLEAN) {
3056 status = try_to_fix_bad_block(trans, root, buf,
3058 if (status != BTRFS_TREE_BLOCK_CLEAN) {
3060 fprintf(stderr, "bad block %llu\n",
3061 (unsigned long long)buf->start);
3064 * Signal to callers we need to start the scan over
3065 * again since we'll have cow'ed blocks.
3070 rec->content_checked = 1;
3071 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3072 rec->owner_ref_checked = 1;
3074 ret = check_owner_ref(root, rec, buf);
3076 rec->owner_ref_checked = 1;
3080 maybe_free_extent_rec(extent_cache, rec);
3084 static struct tree_backref *find_tree_backref(struct extent_record *rec,
3085 u64 parent, u64 root)
3087 struct list_head *cur = rec->backrefs.next;
3088 struct extent_backref *node;
3089 struct tree_backref *back;
3091 while(cur != &rec->backrefs) {
3092 node = list_entry(cur, struct extent_backref, list);
3096 back = (struct tree_backref *)node;
3098 if (!node->full_backref)
3100 if (parent == back->parent)
3103 if (node->full_backref)
3105 if (back->root == root)
3112 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
3113 u64 parent, u64 root)
3115 struct tree_backref *ref = malloc(sizeof(*ref));
3116 memset(&ref->node, 0, sizeof(ref->node));
3118 ref->parent = parent;
3119 ref->node.full_backref = 1;
3122 ref->node.full_backref = 0;
3124 list_add_tail(&ref->node.list, &rec->backrefs);
3129 static struct data_backref *find_data_backref(struct extent_record *rec,
3130 u64 parent, u64 root,
3131 u64 owner, u64 offset,
3133 u64 disk_bytenr, u64 bytes)
3135 struct list_head *cur = rec->backrefs.next;
3136 struct extent_backref *node;
3137 struct data_backref *back;
3139 while(cur != &rec->backrefs) {
3140 node = list_entry(cur, struct extent_backref, list);
3144 back = (struct data_backref *)node;
3146 if (!node->full_backref)
3148 if (parent == back->parent)
3151 if (node->full_backref)
3153 if (back->root == root && back->owner == owner &&
3154 back->offset == offset) {
3155 if (found_ref && node->found_ref &&
3156 (back->bytes != bytes ||
3157 back->disk_bytenr != disk_bytenr))
3166 static struct data_backref *alloc_data_backref(struct extent_record *rec,
3167 u64 parent, u64 root,
3168 u64 owner, u64 offset,
3171 struct data_backref *ref = malloc(sizeof(*ref));
3172 memset(&ref->node, 0, sizeof(ref->node));
3173 ref->node.is_data = 1;
3176 ref->parent = parent;
3179 ref->node.full_backref = 1;
3183 ref->offset = offset;
3184 ref->node.full_backref = 0;
3186 ref->bytes = max_size;
3189 list_add_tail(&ref->node.list, &rec->backrefs);
3190 if (max_size > rec->max_size)
3191 rec->max_size = max_size;
3195 static int add_extent_rec(struct cache_tree *extent_cache,
3196 struct btrfs_key *parent_key, u64 parent_gen,
3197 u64 start, u64 nr, u64 extent_item_refs,
3198 int is_root, int inc_ref, int set_checked,
3199 int metadata, int extent_rec, u64 max_size)
3201 struct extent_record *rec;
3202 struct cache_extent *cache;
3206 cache = lookup_cache_extent(extent_cache, start, nr);
3208 rec = container_of(cache, struct extent_record, cache);
3212 rec->nr = max(nr, max_size);
3215 * We need to make sure to reset nr to whatever the extent
3216 * record says was the real size, this way we can compare it to
3220 if (start != rec->start || rec->found_rec) {
3221 struct extent_record *tmp;
3224 if (list_empty(&rec->list))
3225 list_add_tail(&rec->list,
3226 &duplicate_extents);
3229 * We have to do this song and dance in case we
3230 * find an extent record that falls inside of
3231 * our current extent record but does not have
3232 * the same objectid.
3234 tmp = malloc(sizeof(*tmp));
3238 tmp->max_size = max_size;
3241 tmp->metadata = metadata;
3242 tmp->extent_item_refs = extent_item_refs;
3243 INIT_LIST_HEAD(&tmp->list);
3244 list_add_tail(&tmp->list, &rec->dups);
3245 rec->num_duplicates++;
3252 if (extent_item_refs && !dup) {
3253 if (rec->extent_item_refs) {
3254 fprintf(stderr, "block %llu rec "
3255 "extent_item_refs %llu, passed %llu\n",
3256 (unsigned long long)start,
3257 (unsigned long long)
3258 rec->extent_item_refs,
3259 (unsigned long long)extent_item_refs);
3261 rec->extent_item_refs = extent_item_refs;
3266 rec->content_checked = 1;
3267 rec->owner_ref_checked = 1;
3271 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
3273 rec->parent_generation = parent_gen;
3275 if (rec->max_size < max_size)
3276 rec->max_size = max_size;
3278 maybe_free_extent_rec(extent_cache, rec);
3281 rec = malloc(sizeof(*rec));
3283 rec->max_size = max_size;
3284 rec->nr = max(nr, max_size);
3285 rec->found_rec = !!extent_rec;
3286 rec->content_checked = 0;
3287 rec->owner_ref_checked = 0;
3288 rec->num_duplicates = 0;
3289 rec->metadata = metadata;
3290 INIT_LIST_HEAD(&rec->backrefs);
3291 INIT_LIST_HEAD(&rec->dups);
3292 INIT_LIST_HEAD(&rec->list);
3304 if (extent_item_refs)
3305 rec->extent_item_refs = extent_item_refs;
3307 rec->extent_item_refs = 0;
3310 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
3312 memset(&rec->parent_key, 0, sizeof(*parent_key));
3315 rec->parent_generation = parent_gen;
3317 rec->parent_generation = 0;
3319 rec->cache.start = start;
3320 rec->cache.size = nr;
3321 ret = insert_cache_extent(extent_cache, &rec->cache);
3325 rec->content_checked = 1;
3326 rec->owner_ref_checked = 1;
3331 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
3332 u64 parent, u64 root, int found_ref)
3334 struct extent_record *rec;
3335 struct tree_backref *back;
3336 struct cache_extent *cache;
3338 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3340 add_extent_rec(extent_cache, NULL, 0, bytenr,
3341 1, 0, 0, 0, 0, 1, 0, 0);
3342 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3347 rec = container_of(cache, struct extent_record, cache);
3348 if (rec->start != bytenr) {
3352 back = find_tree_backref(rec, parent, root);
3354 back = alloc_tree_backref(rec, parent, root);
3357 if (back->node.found_ref) {
3358 fprintf(stderr, "Extent back ref already exists "
3359 "for %llu parent %llu root %llu \n",
3360 (unsigned long long)bytenr,
3361 (unsigned long long)parent,
3362 (unsigned long long)root);
3364 back->node.found_ref = 1;
3366 if (back->node.found_extent_tree) {
3367 fprintf(stderr, "Extent back ref already exists "
3368 "for %llu parent %llu root %llu \n",
3369 (unsigned long long)bytenr,
3370 (unsigned long long)parent,
3371 (unsigned long long)root);
3373 back->node.found_extent_tree = 1;
3375 maybe_free_extent_rec(extent_cache, rec);
3379 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
3380 u64 parent, u64 root, u64 owner, u64 offset,
3381 u32 num_refs, int found_ref, u64 max_size)
3383 struct extent_record *rec;
3384 struct data_backref *back;
3385 struct cache_extent *cache;
3387 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3389 add_extent_rec(extent_cache, NULL, 0, bytenr, 1, 0, 0, 0, 0,
3391 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3396 rec = container_of(cache, struct extent_record, cache);
3397 if (rec->max_size < max_size)
3398 rec->max_size = max_size;
3401 * If found_ref is set then max_size is the real size and must match the
3402 * existing refs. So if we have already found a ref then we need to
3403 * make sure that this ref matches the existing one, otherwise we need
3404 * to add a new backref so we can notice that the backrefs don't match
3405 * and we need to figure out who is telling the truth. This is to
3406 * account for that awful fsync bug I introduced where we'd end up with
3407 * a btrfs_file_extent_item that would have its length include multiple
3408 * prealloc extents or point inside of a prealloc extent.
3410 back = find_data_backref(rec, parent, root, owner, offset, found_ref,
3413 back = alloc_data_backref(rec, parent, root, owner, offset,
3417 BUG_ON(num_refs != 1);
3418 if (back->node.found_ref)
3419 BUG_ON(back->bytes != max_size);
3420 back->node.found_ref = 1;
3421 back->found_ref += 1;
3422 back->bytes = max_size;
3423 back->disk_bytenr = bytenr;
3425 rec->content_checked = 1;
3426 rec->owner_ref_checked = 1;
3428 if (back->node.found_extent_tree) {
3429 fprintf(stderr, "Extent back ref already exists "
3430 "for %llu parent %llu root %llu "
3431 "owner %llu offset %llu num_refs %lu\n",
3432 (unsigned long long)bytenr,
3433 (unsigned long long)parent,
3434 (unsigned long long)root,
3435 (unsigned long long)owner,
3436 (unsigned long long)offset,
3437 (unsigned long)num_refs);
3439 back->num_refs = num_refs;
3440 back->node.found_extent_tree = 1;
3442 maybe_free_extent_rec(extent_cache, rec);
3446 static int add_pending(struct cache_tree *pending,
3447 struct cache_tree *seen, u64 bytenr, u32 size)
3450 ret = add_cache_extent(seen, bytenr, size);
3453 add_cache_extent(pending, bytenr, size);
3457 static int pick_next_pending(struct cache_tree *pending,
3458 struct cache_tree *reada,
3459 struct cache_tree *nodes,
3460 u64 last, struct block_info *bits, int bits_nr,
3463 unsigned long node_start = last;
3464 struct cache_extent *cache;
3467 cache = search_cache_extent(reada, 0);
3469 bits[0].start = cache->start;
3470 bits[0].size = cache->size;
3475 if (node_start > 32768)
3476 node_start -= 32768;
3478 cache = search_cache_extent(nodes, node_start);
3480 cache = search_cache_extent(nodes, 0);
3483 cache = search_cache_extent(pending, 0);
3488 bits[ret].start = cache->start;
3489 bits[ret].size = cache->size;
3490 cache = next_cache_extent(cache);
3492 } while (cache && ret < bits_nr);
3498 bits[ret].start = cache->start;
3499 bits[ret].size = cache->size;
3500 cache = next_cache_extent(cache);
3502 } while (cache && ret < bits_nr);
3504 if (bits_nr - ret > 8) {
3505 u64 lookup = bits[0].start + bits[0].size;
3506 struct cache_extent *next;
3507 next = search_cache_extent(pending, lookup);
3509 if (next->start - lookup > 32768)
3511 bits[ret].start = next->start;
3512 bits[ret].size = next->size;
3513 lookup = next->start + next->size;
3517 next = next_cache_extent(next);
3525 static void free_chunk_record(struct cache_extent *cache)
3527 struct chunk_record *rec;
3529 rec = container_of(cache, struct chunk_record, cache);
3530 list_del_init(&rec->list);
3531 list_del_init(&rec->dextents);
3535 void free_chunk_cache_tree(struct cache_tree *chunk_cache)
3537 cache_tree_free_extents(chunk_cache, free_chunk_record);
3540 static void free_device_record(struct rb_node *node)
3542 struct device_record *rec;
3544 rec = container_of(node, struct device_record, node);
3548 FREE_RB_BASED_TREE(device_cache, free_device_record);
3550 int insert_block_group_record(struct block_group_tree *tree,
3551 struct block_group_record *bg_rec)
3555 ret = insert_cache_extent(&tree->tree, &bg_rec->cache);
3559 list_add_tail(&bg_rec->list, &tree->block_groups);
3563 static void free_block_group_record(struct cache_extent *cache)
3565 struct block_group_record *rec;
3567 rec = container_of(cache, struct block_group_record, cache);
3568 list_del_init(&rec->list);
3572 void free_block_group_tree(struct block_group_tree *tree)
3574 cache_tree_free_extents(&tree->tree, free_block_group_record);
3577 int insert_device_extent_record(struct device_extent_tree *tree,
3578 struct device_extent_record *de_rec)
3583 * Device extent is a bit different from the other extents, because
3584 * the extents which belong to the different devices may have the
3585 * same start and size, so we need use the special extent cache
3586 * search/insert functions.
3588 ret = insert_cache_extent2(&tree->tree, &de_rec->cache);
3592 list_add_tail(&de_rec->chunk_list, &tree->no_chunk_orphans);
3593 list_add_tail(&de_rec->device_list, &tree->no_device_orphans);
3597 static void free_device_extent_record(struct cache_extent *cache)
3599 struct device_extent_record *rec;
3601 rec = container_of(cache, struct device_extent_record, cache);
3602 if (!list_empty(&rec->chunk_list))
3603 list_del_init(&rec->chunk_list);
3604 if (!list_empty(&rec->device_list))
3605 list_del_init(&rec->device_list);
3609 void free_device_extent_tree(struct device_extent_tree *tree)
3611 cache_tree_free_extents(&tree->tree, free_device_extent_record);
3614 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3615 static int process_extent_ref_v0(struct cache_tree *extent_cache,
3616 struct extent_buffer *leaf, int slot)
3618 struct btrfs_extent_ref_v0 *ref0;
3619 struct btrfs_key key;
3621 btrfs_item_key_to_cpu(leaf, &key, slot);
3622 ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
3623 if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
3624 add_tree_backref(extent_cache, key.objectid, key.offset, 0, 0);
3626 add_data_backref(extent_cache, key.objectid, key.offset, 0,
3627 0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
3633 struct chunk_record *btrfs_new_chunk_record(struct extent_buffer *leaf,
3634 struct btrfs_key *key,
3637 struct btrfs_chunk *ptr;
3638 struct chunk_record *rec;
3641 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3642 num_stripes = btrfs_chunk_num_stripes(leaf, ptr);
3644 rec = malloc(btrfs_chunk_record_size(num_stripes));
3646 fprintf(stderr, "memory allocation failed\n");
3650 memset(rec, 0, btrfs_chunk_record_size(num_stripes));
3652 INIT_LIST_HEAD(&rec->list);
3653 INIT_LIST_HEAD(&rec->dextents);
3656 rec->cache.start = key->offset;
3657 rec->cache.size = btrfs_chunk_length(leaf, ptr);
3659 rec->generation = btrfs_header_generation(leaf);
3661 rec->objectid = key->objectid;
3662 rec->type = key->type;
3663 rec->offset = key->offset;
3665 rec->length = rec->cache.size;
3666 rec->owner = btrfs_chunk_owner(leaf, ptr);
3667 rec->stripe_len = btrfs_chunk_stripe_len(leaf, ptr);
3668 rec->type_flags = btrfs_chunk_type(leaf, ptr);
3669 rec->io_width = btrfs_chunk_io_width(leaf, ptr);
3670 rec->io_align = btrfs_chunk_io_align(leaf, ptr);
3671 rec->sector_size = btrfs_chunk_sector_size(leaf, ptr);
3672 rec->num_stripes = num_stripes;
3673 rec->sub_stripes = btrfs_chunk_sub_stripes(leaf, ptr);
3675 for (i = 0; i < rec->num_stripes; ++i) {
3676 rec->stripes[i].devid =
3677 btrfs_stripe_devid_nr(leaf, ptr, i);
3678 rec->stripes[i].offset =
3679 btrfs_stripe_offset_nr(leaf, ptr, i);
3680 read_extent_buffer(leaf, rec->stripes[i].dev_uuid,
3681 (unsigned long)btrfs_stripe_dev_uuid_nr(ptr, i),
3688 static int process_chunk_item(struct cache_tree *chunk_cache,
3689 struct btrfs_key *key, struct extent_buffer *eb,
3692 struct chunk_record *rec;
3695 rec = btrfs_new_chunk_record(eb, key, slot);
3696 ret = insert_cache_extent(chunk_cache, &rec->cache);
3698 fprintf(stderr, "Chunk[%llu, %llu] existed.\n",
3699 rec->offset, rec->length);
3706 static int process_device_item(struct rb_root *dev_cache,
3707 struct btrfs_key *key, struct extent_buffer *eb, int slot)
3709 struct btrfs_dev_item *ptr;
3710 struct device_record *rec;
3713 ptr = btrfs_item_ptr(eb,
3714 slot, struct btrfs_dev_item);
3716 rec = malloc(sizeof(*rec));
3718 fprintf(stderr, "memory allocation failed\n");
3722 rec->devid = key->offset;
3723 rec->generation = btrfs_header_generation(eb);
3725 rec->objectid = key->objectid;
3726 rec->type = key->type;
3727 rec->offset = key->offset;
3729 rec->devid = btrfs_device_id(eb, ptr);
3730 rec->total_byte = btrfs_device_total_bytes(eb, ptr);
3731 rec->byte_used = btrfs_device_bytes_used(eb, ptr);
3733 ret = rb_insert(dev_cache, &rec->node, device_record_compare);
3735 fprintf(stderr, "Device[%llu] existed.\n", rec->devid);
3742 struct block_group_record *
3743 btrfs_new_block_group_record(struct extent_buffer *leaf, struct btrfs_key *key,
3746 struct btrfs_block_group_item *ptr;
3747 struct block_group_record *rec;
3749 rec = malloc(sizeof(*rec));
3751 fprintf(stderr, "memory allocation failed\n");
3754 memset(rec, 0, sizeof(*rec));
3756 rec->cache.start = key->objectid;
3757 rec->cache.size = key->offset;
3759 rec->generation = btrfs_header_generation(leaf);
3761 rec->objectid = key->objectid;
3762 rec->type = key->type;
3763 rec->offset = key->offset;
3765 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item);
3766 rec->flags = btrfs_disk_block_group_flags(leaf, ptr);
3768 INIT_LIST_HEAD(&rec->list);
3773 static int process_block_group_item(struct block_group_tree *block_group_cache,
3774 struct btrfs_key *key,
3775 struct extent_buffer *eb, int slot)
3777 struct block_group_record *rec;
3780 rec = btrfs_new_block_group_record(eb, key, slot);
3781 ret = insert_block_group_record(block_group_cache, rec);
3783 fprintf(stderr, "Block Group[%llu, %llu] existed.\n",
3784 rec->objectid, rec->offset);
3791 struct device_extent_record *
3792 btrfs_new_device_extent_record(struct extent_buffer *leaf,
3793 struct btrfs_key *key, int slot)
3795 struct device_extent_record *rec;
3796 struct btrfs_dev_extent *ptr;
3798 rec = malloc(sizeof(*rec));
3800 fprintf(stderr, "memory allocation failed\n");
3803 memset(rec, 0, sizeof(*rec));
3805 rec->cache.objectid = key->objectid;
3806 rec->cache.start = key->offset;
3808 rec->generation = btrfs_header_generation(leaf);
3810 rec->objectid = key->objectid;
3811 rec->type = key->type;
3812 rec->offset = key->offset;
3814 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
3815 rec->chunk_objecteid =
3816 btrfs_dev_extent_chunk_objectid(leaf, ptr);
3818 btrfs_dev_extent_chunk_offset(leaf, ptr);
3819 rec->length = btrfs_dev_extent_length(leaf, ptr);
3820 rec->cache.size = rec->length;
3822 INIT_LIST_HEAD(&rec->chunk_list);
3823 INIT_LIST_HEAD(&rec->device_list);
3829 process_device_extent_item(struct device_extent_tree *dev_extent_cache,
3830 struct btrfs_key *key, struct extent_buffer *eb,
3833 struct device_extent_record *rec;
3836 rec = btrfs_new_device_extent_record(eb, key, slot);
3837 ret = insert_device_extent_record(dev_extent_cache, rec);
3840 "Device extent[%llu, %llu, %llu] existed.\n",
3841 rec->objectid, rec->offset, rec->length);
3848 static int process_extent_item(struct btrfs_root *root,
3849 struct cache_tree *extent_cache,
3850 struct extent_buffer *eb, int slot)
3852 struct btrfs_extent_item *ei;
3853 struct btrfs_extent_inline_ref *iref;
3854 struct btrfs_extent_data_ref *dref;
3855 struct btrfs_shared_data_ref *sref;
3856 struct btrfs_key key;
3860 u32 item_size = btrfs_item_size_nr(eb, slot);
3866 btrfs_item_key_to_cpu(eb, &key, slot);
3868 if (key.type == BTRFS_METADATA_ITEM_KEY) {
3870 num_bytes = root->leafsize;
3872 num_bytes = key.offset;
3875 if (item_size < sizeof(*ei)) {
3876 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3877 struct btrfs_extent_item_v0 *ei0;
3878 BUG_ON(item_size != sizeof(*ei0));
3879 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
3880 refs = btrfs_extent_refs_v0(eb, ei0);
3884 return add_extent_rec(extent_cache, NULL, 0, key.objectid,
3885 num_bytes, refs, 0, 0, 0, metadata, 1,
3889 ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
3890 refs = btrfs_extent_refs(eb, ei);
3892 add_extent_rec(extent_cache, NULL, 0, key.objectid, num_bytes,
3893 refs, 0, 0, 0, metadata, 1, num_bytes);
3895 ptr = (unsigned long)(ei + 1);
3896 if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
3897 key.type == BTRFS_EXTENT_ITEM_KEY)
3898 ptr += sizeof(struct btrfs_tree_block_info);
3900 end = (unsigned long)ei + item_size;
3902 iref = (struct btrfs_extent_inline_ref *)ptr;
3903 type = btrfs_extent_inline_ref_type(eb, iref);
3904 offset = btrfs_extent_inline_ref_offset(eb, iref);
3906 case BTRFS_TREE_BLOCK_REF_KEY:
3907 add_tree_backref(extent_cache, key.objectid,
3910 case BTRFS_SHARED_BLOCK_REF_KEY:
3911 add_tree_backref(extent_cache, key.objectid,
3914 case BTRFS_EXTENT_DATA_REF_KEY:
3915 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3916 add_data_backref(extent_cache, key.objectid, 0,
3917 btrfs_extent_data_ref_root(eb, dref),
3918 btrfs_extent_data_ref_objectid(eb,
3920 btrfs_extent_data_ref_offset(eb, dref),
3921 btrfs_extent_data_ref_count(eb, dref),
3924 case BTRFS_SHARED_DATA_REF_KEY:
3925 sref = (struct btrfs_shared_data_ref *)(iref + 1);
3926 add_data_backref(extent_cache, key.objectid, offset,
3928 btrfs_shared_data_ref_count(eb, sref),
3932 fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
3933 key.objectid, key.type, num_bytes);
3936 ptr += btrfs_extent_inline_ref_size(type);
3943 static int check_cache_range(struct btrfs_root *root,
3944 struct btrfs_block_group_cache *cache,
3945 u64 offset, u64 bytes)
3947 struct btrfs_free_space *entry;
3953 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
3954 bytenr = btrfs_sb_offset(i);
3955 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
3956 cache->key.objectid, bytenr, 0,
3957 &logical, &nr, &stripe_len);
3962 if (logical[nr] + stripe_len <= offset)
3964 if (offset + bytes <= logical[nr])
3966 if (logical[nr] == offset) {
3967 if (stripe_len >= bytes) {
3971 bytes -= stripe_len;
3972 offset += stripe_len;
3973 } else if (logical[nr] < offset) {
3974 if (logical[nr] + stripe_len >=
3979 bytes = (offset + bytes) -
3980 (logical[nr] + stripe_len);
3981 offset = logical[nr] + stripe_len;
3984 * Could be tricky, the super may land in the
3985 * middle of the area we're checking. First
3986 * check the easiest case, it's at the end.
3988 if (logical[nr] + stripe_len >=
3990 bytes = logical[nr] - offset;
3994 /* Check the left side */
3995 ret = check_cache_range(root, cache,
3997 logical[nr] - offset);
4003 /* Now we continue with the right side */
4004 bytes = (offset + bytes) -
4005 (logical[nr] + stripe_len);
4006 offset = logical[nr] + stripe_len;
4013 entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
4015 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
4016 offset, offset+bytes);
4020 if (entry->offset != offset) {
4021 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
4026 if (entry->bytes != bytes) {
4027 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
4028 bytes, entry->bytes, offset);
4032 unlink_free_space(cache->free_space_ctl, entry);
4037 static int verify_space_cache(struct btrfs_root *root,
4038 struct btrfs_block_group_cache *cache)
4040 struct btrfs_path *path;
4041 struct extent_buffer *leaf;
4042 struct btrfs_key key;
4046 path = btrfs_alloc_path();
4050 root = root->fs_info->extent_root;
4052 last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
4054 key.objectid = last;
4056 key.type = BTRFS_EXTENT_ITEM_KEY;
4058 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4063 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4064 ret = btrfs_next_leaf(root, path);
4072 leaf = path->nodes[0];
4073 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4074 if (key.objectid >= cache->key.offset + cache->key.objectid)
4076 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
4077 key.type != BTRFS_METADATA_ITEM_KEY) {
4082 if (last == key.objectid) {
4083 if (key.type == BTRFS_EXTENT_ITEM_KEY)
4084 last = key.objectid + key.offset;
4086 last = key.objectid + root->leafsize;
4091 ret = check_cache_range(root, cache, last,
4092 key.objectid - last);
4095 if (key.type == BTRFS_EXTENT_ITEM_KEY)
4096 last = key.objectid + key.offset;
4098 last = key.objectid + root->leafsize;
4102 if (last < cache->key.objectid + cache->key.offset)
4103 ret = check_cache_range(root, cache, last,
4104 cache->key.objectid +
4105 cache->key.offset - last);
4108 btrfs_free_path(path);
4111 !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
4112 fprintf(stderr, "There are still entries left in the space "
4120 static int check_space_cache(struct btrfs_root *root)
4122 struct btrfs_block_group_cache *cache;
4123 u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
4127 if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
4128 btrfs_super_generation(root->fs_info->super_copy) !=
4129 btrfs_super_cache_generation(root->fs_info->super_copy)) {
4130 printf("cache and super generation don't match, space cache "
4131 "will be invalidated\n");
4136 cache = btrfs_lookup_first_block_group(root->fs_info, start);
4140 start = cache->key.objectid + cache->key.offset;
4141 if (!cache->free_space_ctl) {
4142 if (btrfs_init_free_space_ctl(cache,
4143 root->sectorsize)) {
4148 btrfs_remove_free_space_cache(cache);
4151 ret = load_free_space_cache(root->fs_info, cache);
4155 ret = verify_space_cache(root, cache);
4157 fprintf(stderr, "cache appears valid but isnt %Lu\n",
4158 cache->key.objectid);
4163 return error ? -EINVAL : 0;
4166 static int read_extent_data(struct btrfs_root *root, char *data,
4167 u64 logical, u64 *len, int mirror)
4170 struct btrfs_multi_bio *multi = NULL;
4171 struct btrfs_fs_info *info = root->fs_info;
4172 struct btrfs_device *device;
4176 ret = btrfs_map_block(&info->mapping_tree, READ, logical, len,
4177 &multi, mirror, NULL);
4179 fprintf(stderr, "Couldn't map the block %llu\n",
4183 device = multi->stripes[0].dev;
4185 if (device->fd == 0)
4190 ret = pread64(device->fd, data, *len, multi->stripes[0].physical);
4200 static int check_extent_csums(struct btrfs_root *root, u64 bytenr,
4201 u64 num_bytes, unsigned long leaf_offset,
4202 struct extent_buffer *eb) {
4205 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
4207 unsigned long csum_offset;
4211 u64 data_checked = 0;
4217 if (num_bytes % root->sectorsize)
4220 data = malloc(num_bytes);
4224 while (offset < num_bytes) {
4227 read_len = num_bytes - offset;
4228 /* read as much space once a time */
4229 ret = read_extent_data(root, data + offset,
4230 bytenr + offset, &read_len, mirror);
4234 /* verify every 4k data's checksum */
4235 while (data_checked < read_len) {
4237 tmp = offset + data_checked;
4239 csum = btrfs_csum_data(NULL, (char *)data + tmp,
4240 csum, root->sectorsize);
4241 btrfs_csum_final(csum, (char *)&csum);
4243 csum_offset = leaf_offset +
4244 tmp / root->sectorsize * csum_size;
4245 read_extent_buffer(eb, (char *)&csum_expected,
4246 csum_offset, csum_size);
4247 /* try another mirror */
4248 if (csum != csum_expected) {
4249 fprintf(stderr, "mirror %d bytenr %llu csum %u expected csum %u\n",
4250 mirror, bytenr + tmp,
4251 csum, csum_expected);
4252 num_copies = btrfs_num_copies(
4253 &root->fs_info->mapping_tree,
4255 if (mirror < num_copies - 1) {
4260 data_checked += root->sectorsize;
4269 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
4272 struct btrfs_path *path;
4273 struct extent_buffer *leaf;
4274 struct btrfs_key key;
4277 path = btrfs_alloc_path();
4279 fprintf(stderr, "Error allocing path\n");
4283 key.objectid = bytenr;
4284 key.type = BTRFS_EXTENT_ITEM_KEY;
4285 key.offset = (u64)-1;
4288 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
4291 fprintf(stderr, "Error looking up extent record %d\n", ret);
4292 btrfs_free_path(path);
4295 if (path->slots[0] > 0) {
4298 ret = btrfs_prev_leaf(root, path);
4301 } else if (ret > 0) {
4308 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4311 * Block group items come before extent items if they have the same
4312 * bytenr, so walk back one more just in case. Dear future traveler,
4313 * first congrats on mastering time travel. Now if it's not too much
4314 * trouble could you go back to 2006 and tell Chris to make the
4315 * BLOCK_GROUP_ITEM_KEY (and BTRFS_*_REF_KEY) lower than the
4316 * EXTENT_ITEM_KEY please?
4318 while (key.type > BTRFS_EXTENT_ITEM_KEY) {
4319 if (path->slots[0] > 0) {
4322 ret = btrfs_prev_leaf(root, path);
4325 } else if (ret > 0) {
4330 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4334 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4335 ret = btrfs_next_leaf(root, path);
4337 fprintf(stderr, "Error going to next leaf "
4339 btrfs_free_path(path);
4345 leaf = path->nodes[0];
4346 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4347 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
4351 if (key.objectid + key.offset < bytenr) {
4355 if (key.objectid > bytenr + num_bytes)
4358 if (key.objectid == bytenr) {
4359 if (key.offset >= num_bytes) {
4363 num_bytes -= key.offset;
4364 bytenr += key.offset;
4365 } else if (key.objectid < bytenr) {
4366 if (key.objectid + key.offset >= bytenr + num_bytes) {
4370 num_bytes = (bytenr + num_bytes) -
4371 (key.objectid + key.offset);
4372 bytenr = key.objectid + key.offset;
4374 if (key.objectid + key.offset < bytenr + num_bytes) {
4375 u64 new_start = key.objectid + key.offset;
4376 u64 new_bytes = bytenr + num_bytes - new_start;
4379 * Weird case, the extent is in the middle of
4380 * our range, we'll have to search one side
4381 * and then the other. Not sure if this happens
4382 * in real life, but no harm in coding it up
4383 * anyway just in case.
4385 btrfs_release_path(path);
4386 ret = check_extent_exists(root, new_start,
4389 fprintf(stderr, "Right section didn't "
4393 num_bytes = key.objectid - bytenr;
4396 num_bytes = key.objectid - bytenr;
4403 if (num_bytes && !ret) {
4404 fprintf(stderr, "There are no extents for csum range "
4405 "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
4409 btrfs_free_path(path);
4413 static int check_csums(struct btrfs_root *root)
4415 struct btrfs_path *path;
4416 struct extent_buffer *leaf;
4417 struct btrfs_key key;
4418 u64 offset = 0, num_bytes = 0;
4419 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
4423 unsigned long leaf_offset;
4425 root = root->fs_info->csum_root;
4426 if (!extent_buffer_uptodate(root->node)) {
4427 fprintf(stderr, "No valid csum tree found\n");
4431 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
4432 key.type = BTRFS_EXTENT_CSUM_KEY;
4435 path = btrfs_alloc_path();
4439 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4441 fprintf(stderr, "Error searching csum tree %d\n", ret);
4442 btrfs_free_path(path);
4446 if (ret > 0 && path->slots[0])
4451 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4452 ret = btrfs_next_leaf(root, path);
4454 fprintf(stderr, "Error going to next leaf "
4461 leaf = path->nodes[0];
4463 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4464 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
4469 data_len = (btrfs_item_size_nr(leaf, path->slots[0]) /
4470 csum_size) * root->sectorsize;
4471 if (!check_data_csum)
4472 goto skip_csum_check;
4473 leaf_offset = btrfs_item_ptr_offset(leaf, path->slots[0]);
4474 ret = check_extent_csums(root, key.offset, data_len,
4480 offset = key.offset;
4481 } else if (key.offset != offset + num_bytes) {
4482 ret = check_extent_exists(root, offset, num_bytes);
4484 fprintf(stderr, "Csum exists for %Lu-%Lu but "
4485 "there is no extent record\n",
4486 offset, offset+num_bytes);
4489 offset = key.offset;
4492 num_bytes += data_len;
4496 btrfs_free_path(path);
4500 static int is_dropped_key(struct btrfs_key *key,
4501 struct btrfs_key *drop_key) {
4502 if (key->objectid < drop_key->objectid)
4504 else if (key->objectid == drop_key->objectid) {
4505 if (key->type < drop_key->type)
4507 else if (key->type == drop_key->type) {
4508 if (key->offset < drop_key->offset)
4515 static int run_next_block(struct btrfs_trans_handle *trans,
4516 struct btrfs_root *root,
4517 struct block_info *bits,
4520 struct cache_tree *pending,
4521 struct cache_tree *seen,
4522 struct cache_tree *reada,
4523 struct cache_tree *nodes,
4524 struct cache_tree *extent_cache,
4525 struct cache_tree *chunk_cache,
4526 struct rb_root *dev_cache,
4527 struct block_group_tree *block_group_cache,
4528 struct device_extent_tree *dev_extent_cache,
4529 struct btrfs_root_item *ri)
4531 struct extent_buffer *buf;
4542 struct btrfs_key key;
4543 struct cache_extent *cache;
4546 nritems = pick_next_pending(pending, reada, nodes, *last, bits,
4547 bits_nr, &reada_bits);
4552 for(i = 0; i < nritems; i++) {
4553 ret = add_cache_extent(reada, bits[i].start,
4558 /* fixme, get the parent transid */
4559 readahead_tree_block(root, bits[i].start,
4563 *last = bits[0].start;
4564 bytenr = bits[0].start;
4565 size = bits[0].size;
4567 cache = lookup_cache_extent(pending, bytenr, size);
4569 remove_cache_extent(pending, cache);
4572 cache = lookup_cache_extent(reada, bytenr, size);
4574 remove_cache_extent(reada, cache);
4577 cache = lookup_cache_extent(nodes, bytenr, size);
4579 remove_cache_extent(nodes, cache);
4582 cache = lookup_cache_extent(extent_cache, bytenr, size);
4584 struct extent_record *rec;
4586 rec = container_of(cache, struct extent_record, cache);
4587 gen = rec->parent_generation;
4590 /* fixme, get the real parent transid */
4591 buf = read_tree_block(root, bytenr, size, gen);
4592 if (!extent_buffer_uptodate(buf)) {
4593 record_bad_block_io(root->fs_info,
4594 extent_cache, bytenr, size);
4598 nritems = btrfs_header_nritems(buf);
4601 * FIXME, this only works only if we don't have any full
4604 if (!init_extent_tree) {
4605 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
4606 btrfs_header_level(buf), 1, NULL,
4614 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
4619 owner = btrfs_header_owner(buf);
4622 ret = check_block(trans, root, extent_cache, buf, flags);
4626 if (btrfs_is_leaf(buf)) {
4627 btree_space_waste += btrfs_leaf_free_space(root, buf);
4628 for (i = 0; i < nritems; i++) {
4629 struct btrfs_file_extent_item *fi;
4630 btrfs_item_key_to_cpu(buf, &key, i);
4631 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
4632 process_extent_item(root, extent_cache, buf,
4636 if (key.type == BTRFS_METADATA_ITEM_KEY) {
4637 process_extent_item(root, extent_cache, buf,
4641 if (key.type == BTRFS_EXTENT_CSUM_KEY) {
4643 btrfs_item_size_nr(buf, i);
4646 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4647 process_chunk_item(chunk_cache, &key, buf, i);
4650 if (key.type == BTRFS_DEV_ITEM_KEY) {
4651 process_device_item(dev_cache, &key, buf, i);
4654 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
4655 process_block_group_item(block_group_cache,
4659 if (key.type == BTRFS_DEV_EXTENT_KEY) {
4660 process_device_extent_item(dev_extent_cache,
4665 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
4666 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4667 process_extent_ref_v0(extent_cache, buf, i);
4674 if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
4675 add_tree_backref(extent_cache, key.objectid, 0,
4679 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
4680 add_tree_backref(extent_cache, key.objectid,
4684 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
4685 struct btrfs_extent_data_ref *ref;
4686 ref = btrfs_item_ptr(buf, i,
4687 struct btrfs_extent_data_ref);
4688 add_data_backref(extent_cache,
4690 btrfs_extent_data_ref_root(buf, ref),
4691 btrfs_extent_data_ref_objectid(buf,
4693 btrfs_extent_data_ref_offset(buf, ref),
4694 btrfs_extent_data_ref_count(buf, ref),
4695 0, root->sectorsize);
4698 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
4699 struct btrfs_shared_data_ref *ref;
4700 ref = btrfs_item_ptr(buf, i,
4701 struct btrfs_shared_data_ref);
4702 add_data_backref(extent_cache,
4703 key.objectid, key.offset, 0, 0, 0,
4704 btrfs_shared_data_ref_count(buf, ref),
4705 0, root->sectorsize);
4708 if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
4709 struct bad_item *bad;
4711 if (key.objectid == BTRFS_ORPHAN_OBJECTID)
4715 bad = malloc(sizeof(struct bad_item));
4718 INIT_LIST_HEAD(&bad->list);
4719 memcpy(&bad->key, &key,
4720 sizeof(struct btrfs_key));
4721 bad->root_id = owner;
4722 list_add_tail(&bad->list, &delete_items);
4725 if (key.type != BTRFS_EXTENT_DATA_KEY)
4727 fi = btrfs_item_ptr(buf, i,
4728 struct btrfs_file_extent_item);
4729 if (btrfs_file_extent_type(buf, fi) ==
4730 BTRFS_FILE_EXTENT_INLINE)
4732 if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
4735 data_bytes_allocated +=
4736 btrfs_file_extent_disk_num_bytes(buf, fi);
4737 if (data_bytes_allocated < root->sectorsize) {
4740 data_bytes_referenced +=
4741 btrfs_file_extent_num_bytes(buf, fi);
4742 add_data_backref(extent_cache,
4743 btrfs_file_extent_disk_bytenr(buf, fi),
4744 parent, owner, key.objectid, key.offset -
4745 btrfs_file_extent_offset(buf, fi), 1, 1,
4746 btrfs_file_extent_disk_num_bytes(buf, fi));
4750 struct btrfs_key first_key;
4752 first_key.objectid = 0;
4755 btrfs_item_key_to_cpu(buf, &first_key, 0);
4756 level = btrfs_header_level(buf);
4757 for (i = 0; i < nritems; i++) {
4758 ptr = btrfs_node_blockptr(buf, i);
4759 size = btrfs_level_size(root, level - 1);
4760 btrfs_node_key_to_cpu(buf, &key, i);
4762 struct btrfs_key drop_key;
4763 btrfs_disk_key_to_cpu(&drop_key,
4764 &ri->drop_progress);
4765 if ((level == ri->drop_level)
4766 && is_dropped_key(&key, &drop_key)) {
4770 ret = add_extent_rec(extent_cache, &key,
4771 btrfs_node_ptr_generation(buf, i),
4772 ptr, size, 0, 0, 1, 0, 1, 0,
4776 add_tree_backref(extent_cache, ptr, parent, owner, 1);
4779 add_pending(nodes, seen, ptr, size);
4781 add_pending(pending, seen, ptr, size);
4784 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
4785 nritems) * sizeof(struct btrfs_key_ptr);
4787 total_btree_bytes += buf->len;
4788 if (fs_root_objectid(btrfs_header_owner(buf)))
4789 total_fs_tree_bytes += buf->len;
4790 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
4791 total_extent_tree_bytes += buf->len;
4792 if (!found_old_backref &&
4793 btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
4794 btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
4795 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
4796 found_old_backref = 1;
4798 free_extent_buffer(buf);
4802 static int add_root_to_pending(struct extent_buffer *buf,
4803 struct cache_tree *extent_cache,
4804 struct cache_tree *pending,
4805 struct cache_tree *seen,
4806 struct cache_tree *nodes,
4807 struct btrfs_key *root_key)
4809 if (btrfs_header_level(buf) > 0)
4810 add_pending(nodes, seen, buf->start, buf->len);
4812 add_pending(pending, seen, buf->start, buf->len);
4813 add_extent_rec(extent_cache, NULL, 0, buf->start, buf->len,
4814 0, 1, 1, 0, 1, 0, buf->len);
4816 if (root_key->objectid == BTRFS_TREE_RELOC_OBJECTID ||
4817 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
4818 add_tree_backref(extent_cache, buf->start, buf->start,
4821 add_tree_backref(extent_cache, buf->start, 0,
4822 root_key->objectid, 1);
4826 /* as we fix the tree, we might be deleting blocks that
4827 * we're tracking for repair. This hook makes sure we
4828 * remove any backrefs for blocks as we are fixing them.
4830 static int free_extent_hook(struct btrfs_trans_handle *trans,
4831 struct btrfs_root *root,
4832 u64 bytenr, u64 num_bytes, u64 parent,
4833 u64 root_objectid, u64 owner, u64 offset,
4836 struct extent_record *rec;
4837 struct cache_extent *cache;
4839 struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
4841 is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
4842 cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
4846 rec = container_of(cache, struct extent_record, cache);
4848 struct data_backref *back;
4849 back = find_data_backref(rec, parent, root_objectid, owner,
4850 offset, 1, bytenr, num_bytes);
4853 if (back->node.found_ref) {
4854 back->found_ref -= refs_to_drop;
4856 rec->refs -= refs_to_drop;
4858 if (back->node.found_extent_tree) {
4859 back->num_refs -= refs_to_drop;
4860 if (rec->extent_item_refs)
4861 rec->extent_item_refs -= refs_to_drop;
4863 if (back->found_ref == 0)
4864 back->node.found_ref = 0;
4865 if (back->num_refs == 0)
4866 back->node.found_extent_tree = 0;
4868 if (!back->node.found_extent_tree && back->node.found_ref) {
4869 list_del(&back->node.list);
4873 struct tree_backref *back;
4874 back = find_tree_backref(rec, parent, root_objectid);
4877 if (back->node.found_ref) {
4880 back->node.found_ref = 0;
4882 if (back->node.found_extent_tree) {
4883 if (rec->extent_item_refs)
4884 rec->extent_item_refs--;
4885 back->node.found_extent_tree = 0;
4887 if (!back->node.found_extent_tree && back->node.found_ref) {
4888 list_del(&back->node.list);
4892 maybe_free_extent_rec(extent_cache, rec);
4897 static int delete_extent_records(struct btrfs_trans_handle *trans,
4898 struct btrfs_root *root,
4899 struct btrfs_path *path,
4900 u64 bytenr, u64 new_len)
4902 struct btrfs_key key;
4903 struct btrfs_key found_key;
4904 struct extent_buffer *leaf;
4909 key.objectid = bytenr;
4911 key.offset = (u64)-1;
4914 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
4921 if (path->slots[0] == 0)
4927 leaf = path->nodes[0];
4928 slot = path->slots[0];
4930 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4931 if (found_key.objectid != bytenr)
4934 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
4935 found_key.type != BTRFS_METADATA_ITEM_KEY &&
4936 found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
4937 found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
4938 found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
4939 found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
4940 found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
4941 btrfs_release_path(path);
4942 if (found_key.type == 0) {
4943 if (found_key.offset == 0)
4945 key.offset = found_key.offset - 1;
4946 key.type = found_key.type;
4948 key.type = found_key.type - 1;
4949 key.offset = (u64)-1;
4953 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
4954 found_key.objectid, found_key.type, found_key.offset);
4956 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
4959 btrfs_release_path(path);
4961 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
4962 found_key.type == BTRFS_METADATA_ITEM_KEY) {
4963 u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
4964 found_key.offset : root->leafsize;
4966 ret = btrfs_update_block_group(trans, root, bytenr,
4973 btrfs_release_path(path);
4978 * for a single backref, this will allocate a new extent
4979 * and add the backref to it.
4981 static int record_extent(struct btrfs_trans_handle *trans,
4982 struct btrfs_fs_info *info,
4983 struct btrfs_path *path,
4984 struct extent_record *rec,
4985 struct extent_backref *back,
4986 int allocated, u64 flags)
4989 struct btrfs_root *extent_root = info->extent_root;
4990 struct extent_buffer *leaf;
4991 struct btrfs_key ins_key;
4992 struct btrfs_extent_item *ei;
4993 struct tree_backref *tback;
4994 struct data_backref *dback;
4995 struct btrfs_tree_block_info *bi;
4998 rec->max_size = max_t(u64, rec->max_size,
4999 info->extent_root->leafsize);
5002 u32 item_size = sizeof(*ei);
5005 item_size += sizeof(*bi);
5007 ins_key.objectid = rec->start;
5008 ins_key.offset = rec->max_size;
5009 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
5011 ret = btrfs_insert_empty_item(trans, extent_root, path,
5012 &ins_key, item_size);
5016 leaf = path->nodes[0];
5017 ei = btrfs_item_ptr(leaf, path->slots[0],
5018 struct btrfs_extent_item);
5020 btrfs_set_extent_refs(leaf, ei, 0);
5021 btrfs_set_extent_generation(leaf, ei, rec->generation);
5023 if (back->is_data) {
5024 btrfs_set_extent_flags(leaf, ei,
5025 BTRFS_EXTENT_FLAG_DATA);
5027 struct btrfs_disk_key copy_key;;
5029 tback = (struct tree_backref *)back;
5030 bi = (struct btrfs_tree_block_info *)(ei + 1);
5031 memset_extent_buffer(leaf, 0, (unsigned long)bi,
5034 btrfs_set_disk_key_objectid(©_key,
5035 rec->info_objectid);
5036 btrfs_set_disk_key_type(©_key, 0);
5037 btrfs_set_disk_key_offset(©_key, 0);
5039 btrfs_set_tree_block_level(leaf, bi, rec->info_level);
5040 btrfs_set_tree_block_key(leaf, bi, ©_key);
5042 btrfs_set_extent_flags(leaf, ei,
5043 BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
5046 btrfs_mark_buffer_dirty(leaf);
5047 ret = btrfs_update_block_group(trans, extent_root, rec->start,
5048 rec->max_size, 1, 0);
5051 btrfs_release_path(path);
5054 if (back->is_data) {
5058 dback = (struct data_backref *)back;
5059 if (back->full_backref)
5060 parent = dback->parent;
5064 for (i = 0; i < dback->found_ref; i++) {
5065 /* if parent != 0, we're doing a full backref
5066 * passing BTRFS_FIRST_FREE_OBJECTID as the owner
5067 * just makes the backref allocator create a data
5070 ret = btrfs_inc_extent_ref(trans, info->extent_root,
5071 rec->start, rec->max_size,
5075 BTRFS_FIRST_FREE_OBJECTID :
5081 fprintf(stderr, "adding new data backref"
5082 " on %llu %s %llu owner %llu"
5083 " offset %llu found %d\n",
5084 (unsigned long long)rec->start,
5085 back->full_backref ?
5087 back->full_backref ?
5088 (unsigned long long)parent :
5089 (unsigned long long)dback->root,
5090 (unsigned long long)dback->owner,
5091 (unsigned long long)dback->offset,
5096 tback = (struct tree_backref *)back;
5097 if (back->full_backref)
5098 parent = tback->parent;
5102 ret = btrfs_inc_extent_ref(trans, info->extent_root,
5103 rec->start, rec->max_size,
5104 parent, tback->root, 0, 0);
5105 fprintf(stderr, "adding new tree backref on "
5106 "start %llu len %llu parent %llu root %llu\n",
5107 rec->start, rec->max_size, tback->parent, tback->root);
5112 btrfs_release_path(path);
5116 struct extent_entry {
5121 struct list_head list;
5124 static struct extent_entry *find_entry(struct list_head *entries,
5125 u64 bytenr, u64 bytes)
5127 struct extent_entry *entry = NULL;
5129 list_for_each_entry(entry, entries, list) {
5130 if (entry->bytenr == bytenr && entry->bytes == bytes)
5137 static struct extent_entry *find_most_right_entry(struct list_head *entries)
5139 struct extent_entry *entry, *best = NULL, *prev = NULL;
5141 list_for_each_entry(entry, entries, list) {
5148 * If there are as many broken entries as entries then we know
5149 * not to trust this particular entry.
5151 if (entry->broken == entry->count)
5155 * If our current entry == best then we can't be sure our best
5156 * is really the best, so we need to keep searching.
5158 if (best && best->count == entry->count) {
5164 /* Prev == entry, not good enough, have to keep searching */
5165 if (!prev->broken && prev->count == entry->count)
5169 best = (prev->count > entry->count) ? prev : entry;
5170 else if (best->count < entry->count)
5178 static int repair_ref(struct btrfs_trans_handle *trans,
5179 struct btrfs_fs_info *info, struct btrfs_path *path,
5180 struct data_backref *dback, struct extent_entry *entry)
5182 struct btrfs_root *root;
5183 struct btrfs_file_extent_item *fi;
5184 struct extent_buffer *leaf;
5185 struct btrfs_key key;
5189 key.objectid = dback->root;
5190 key.type = BTRFS_ROOT_ITEM_KEY;
5191 key.offset = (u64)-1;
5192 root = btrfs_read_fs_root(info, &key);
5194 fprintf(stderr, "Couldn't find root for our ref\n");
5199 * The backref points to the original offset of the extent if it was
5200 * split, so we need to search down to the offset we have and then walk
5201 * forward until we find the backref we're looking for.
5203 key.objectid = dback->owner;
5204 key.type = BTRFS_EXTENT_DATA_KEY;
5205 key.offset = dback->offset;
5206 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5208 fprintf(stderr, "Error looking up ref %d\n", ret);
5213 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5214 ret = btrfs_next_leaf(root, path);
5216 fprintf(stderr, "Couldn't find our ref, next\n");
5220 leaf = path->nodes[0];
5221 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5222 if (key.objectid != dback->owner ||
5223 key.type != BTRFS_EXTENT_DATA_KEY) {
5224 fprintf(stderr, "Couldn't find our ref, search\n");
5227 fi = btrfs_item_ptr(leaf, path->slots[0],
5228 struct btrfs_file_extent_item);
5229 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5230 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
5232 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
5237 btrfs_release_path(path);
5240 * Have to make sure that this root gets updated when we commit the
5243 record_root_in_trans(trans, root);
5246 * Ok we have the key of the file extent we want to fix, now we can cow
5247 * down to the thing and fix it.
5249 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
5251 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
5252 key.objectid, key.type, key.offset, ret);
5256 fprintf(stderr, "Well that's odd, we just found this key "
5257 "[%Lu, %u, %Lu]\n", key.objectid, key.type,
5261 leaf = path->nodes[0];
5262 fi = btrfs_item_ptr(leaf, path->slots[0],
5263 struct btrfs_file_extent_item);
5265 if (btrfs_file_extent_compression(leaf, fi) &&
5266 dback->disk_bytenr != entry->bytenr) {
5267 fprintf(stderr, "Ref doesn't match the record start and is "
5268 "compressed, please take a btrfs-image of this file "
5269 "system and send it to a btrfs developer so they can "
5270 "complete this functionality for bytenr %Lu\n",
5271 dback->disk_bytenr);
5275 if (dback->node.broken && dback->disk_bytenr != entry->bytenr) {
5276 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
5277 } else if (dback->disk_bytenr > entry->bytenr) {
5278 u64 off_diff, offset;
5280 off_diff = dback->disk_bytenr - entry->bytenr;
5281 offset = btrfs_file_extent_offset(leaf, fi);
5282 if (dback->disk_bytenr + offset +
5283 btrfs_file_extent_num_bytes(leaf, fi) >
5284 entry->bytenr + entry->bytes) {
5285 fprintf(stderr, "Ref is past the entry end, please "
5286 "take a btrfs-image of this file system and "
5287 "send it to a btrfs developer, ref %Lu\n",
5288 dback->disk_bytenr);
5292 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
5293 btrfs_set_file_extent_offset(leaf, fi, offset);
5294 } else if (dback->disk_bytenr < entry->bytenr) {
5297 offset = btrfs_file_extent_offset(leaf, fi);
5298 if (dback->disk_bytenr + offset < entry->bytenr) {
5299 fprintf(stderr, "Ref is before the entry start, please"
5300 " take a btrfs-image of this file system and "
5301 "send it to a btrfs developer, ref %Lu\n",
5302 dback->disk_bytenr);
5306 offset += dback->disk_bytenr;
5307 offset -= entry->bytenr;
5308 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
5309 btrfs_set_file_extent_offset(leaf, fi, offset);
5312 btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
5315 * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
5316 * only do this if we aren't using compression, otherwise it's a
5319 if (!btrfs_file_extent_compression(leaf, fi))
5320 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
5322 printf("ram bytes may be wrong?\n");
5323 btrfs_mark_buffer_dirty(leaf);
5324 btrfs_release_path(path);
5328 static int verify_backrefs(struct btrfs_trans_handle *trans,
5329 struct btrfs_fs_info *info, struct btrfs_path *path,
5330 struct extent_record *rec)
5332 struct extent_backref *back;
5333 struct data_backref *dback;
5334 struct extent_entry *entry, *best = NULL;
5337 int broken_entries = 0;
5342 * Metadata is easy and the backrefs should always agree on bytenr and
5343 * size, if not we've got bigger issues.
5348 list_for_each_entry(back, &rec->backrefs, list) {
5349 if (back->full_backref || !back->is_data)
5352 dback = (struct data_backref *)back;
5355 * We only pay attention to backrefs that we found a real
5358 if (dback->found_ref == 0)
5362 * For now we only catch when the bytes don't match, not the
5363 * bytenr. We can easily do this at the same time, but I want
5364 * to have a fs image to test on before we just add repair
5365 * functionality willy-nilly so we know we won't screw up the
5369 entry = find_entry(&entries, dback->disk_bytenr,
5372 entry = malloc(sizeof(struct extent_entry));
5377 memset(entry, 0, sizeof(*entry));
5378 entry->bytenr = dback->disk_bytenr;
5379 entry->bytes = dback->bytes;
5380 list_add_tail(&entry->list, &entries);
5385 * If we only have on entry we may think the entries agree when
5386 * in reality they don't so we have to do some extra checking.
5388 if (dback->disk_bytenr != rec->start ||
5389 dback->bytes != rec->nr || back->broken)
5400 /* Yay all the backrefs agree, carry on good sir */
5401 if (nr_entries <= 1 && !mismatch)
5404 fprintf(stderr, "attempting to repair backref discrepency for bytenr "
5405 "%Lu\n", rec->start);
5408 * First we want to see if the backrefs can agree amongst themselves who
5409 * is right, so figure out which one of the entries has the highest
5412 best = find_most_right_entry(&entries);
5415 * Ok so we may have an even split between what the backrefs think, so
5416 * this is where we use the extent ref to see what it thinks.
5419 entry = find_entry(&entries, rec->start, rec->nr);
5420 if (!entry && (!broken_entries || !rec->found_rec)) {
5421 fprintf(stderr, "Backrefs don't agree with each other "
5422 "and extent record doesn't agree with anybody,"
5423 " so we can't fix bytenr %Lu bytes %Lu\n",
5424 rec->start, rec->nr);
5427 } else if (!entry) {
5429 * Ok our backrefs were broken, we'll assume this is the
5430 * correct value and add an entry for this range.
5432 entry = malloc(sizeof(struct extent_entry));
5437 memset(entry, 0, sizeof(*entry));
5438 entry->bytenr = rec->start;
5439 entry->bytes = rec->nr;
5440 list_add_tail(&entry->list, &entries);
5444 best = find_most_right_entry(&entries);
5446 fprintf(stderr, "Backrefs and extent record evenly "
5447 "split on who is right, this is going to "
5448 "require user input to fix bytenr %Lu bytes "
5449 "%Lu\n", rec->start, rec->nr);
5456 * I don't think this can happen currently as we'll abort() if we catch
5457 * this case higher up, but in case somebody removes that we still can't
5458 * deal with it properly here yet, so just bail out of that's the case.
5460 if (best->bytenr != rec->start) {
5461 fprintf(stderr, "Extent start and backref starts don't match, "
5462 "please use btrfs-image on this file system and send "
5463 "it to a btrfs developer so they can make fsck fix "
5464 "this particular case. bytenr is %Lu, bytes is %Lu\n",
5465 rec->start, rec->nr);
5471 * Ok great we all agreed on an extent record, let's go find the real
5472 * references and fix up the ones that don't match.
5474 list_for_each_entry(back, &rec->backrefs, list) {
5475 if (back->full_backref || !back->is_data)
5478 dback = (struct data_backref *)back;
5481 * Still ignoring backrefs that don't have a real ref attached
5484 if (dback->found_ref == 0)
5487 if (dback->bytes == best->bytes &&
5488 dback->disk_bytenr == best->bytenr)
5491 ret = repair_ref(trans, info, path, dback, best);
5497 * Ok we messed with the actual refs, which means we need to drop our
5498 * entire cache and go back and rescan. I know this is a huge pain and
5499 * adds a lot of extra work, but it's the only way to be safe. Once all
5500 * the backrefs agree we may not need to do anything to the extent
5505 while (!list_empty(&entries)) {
5506 entry = list_entry(entries.next, struct extent_entry, list);
5507 list_del_init(&entry->list);
5513 static int process_duplicates(struct btrfs_root *root,
5514 struct cache_tree *extent_cache,
5515 struct extent_record *rec)
5517 struct extent_record *good, *tmp;
5518 struct cache_extent *cache;
5522 * If we found a extent record for this extent then return, or if we
5523 * have more than one duplicate we are likely going to need to delete
5526 if (rec->found_rec || rec->num_duplicates > 1)
5529 /* Shouldn't happen but just in case */
5530 BUG_ON(!rec->num_duplicates);
5533 * So this happens if we end up with a backref that doesn't match the
5534 * actual extent entry. So either the backref is bad or the extent
5535 * entry is bad. Either way we want to have the extent_record actually
5536 * reflect what we found in the extent_tree, so we need to take the
5537 * duplicate out and use that as the extent_record since the only way we
5538 * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
5540 remove_cache_extent(extent_cache, &rec->cache);
5542 good = list_entry(rec->dups.next, struct extent_record, list);
5543 list_del_init(&good->list);
5544 INIT_LIST_HEAD(&good->backrefs);
5545 INIT_LIST_HEAD(&good->dups);
5546 good->cache.start = good->start;
5547 good->cache.size = good->nr;
5548 good->content_checked = 0;
5549 good->owner_ref_checked = 0;
5550 good->num_duplicates = 0;
5551 good->refs = rec->refs;
5552 list_splice_init(&rec->backrefs, &good->backrefs);
5554 cache = lookup_cache_extent(extent_cache, good->start,
5558 tmp = container_of(cache, struct extent_record, cache);
5561 * If we find another overlapping extent and it's found_rec is
5562 * set then it's a duplicate and we need to try and delete
5565 if (tmp->found_rec || tmp->num_duplicates > 0) {
5566 if (list_empty(&good->list))
5567 list_add_tail(&good->list,
5568 &duplicate_extents);
5569 good->num_duplicates += tmp->num_duplicates + 1;
5570 list_splice_init(&tmp->dups, &good->dups);
5571 list_del_init(&tmp->list);
5572 list_add_tail(&tmp->list, &good->dups);
5573 remove_cache_extent(extent_cache, &tmp->cache);
5578 * Ok we have another non extent item backed extent rec, so lets
5579 * just add it to this extent and carry on like we did above.
5581 good->refs += tmp->refs;
5582 list_splice_init(&tmp->backrefs, &good->backrefs);
5583 remove_cache_extent(extent_cache, &tmp->cache);
5586 ret = insert_cache_extent(extent_cache, &good->cache);
5589 return good->num_duplicates ? 0 : 1;
5592 static int delete_duplicate_records(struct btrfs_trans_handle *trans,
5593 struct btrfs_root *root,
5594 struct extent_record *rec)
5596 LIST_HEAD(delete_list);
5597 struct btrfs_path *path;
5598 struct extent_record *tmp, *good, *n;
5601 struct btrfs_key key;
5603 path = btrfs_alloc_path();
5610 /* Find the record that covers all of the duplicates. */
5611 list_for_each_entry(tmp, &rec->dups, list) {
5612 if (good->start < tmp->start)
5614 if (good->nr > tmp->nr)
5617 if (tmp->start + tmp->nr < good->start + good->nr) {
5618 fprintf(stderr, "Ok we have overlapping extents that "
5619 "aren't completely covered by eachother, this "
5620 "is going to require more careful thought. "
5621 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
5622 tmp->start, tmp->nr, good->start, good->nr);
5629 list_add_tail(&rec->list, &delete_list);
5631 list_for_each_entry_safe(tmp, n, &rec->dups, list) {
5634 list_move_tail(&tmp->list, &delete_list);
5637 root = root->fs_info->extent_root;
5638 list_for_each_entry(tmp, &delete_list, list) {
5639 if (tmp->found_rec == 0)
5641 key.objectid = tmp->start;
5642 key.type = BTRFS_EXTENT_ITEM_KEY;
5643 key.offset = tmp->nr;
5645 /* Shouldn't happen but just in case */
5646 if (tmp->metadata) {
5647 fprintf(stderr, "Well this shouldn't happen, extent "
5648 "record overlaps but is metadata? "
5649 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
5653 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5659 ret = btrfs_del_item(trans, root, path);
5662 btrfs_release_path(path);
5667 while (!list_empty(&delete_list)) {
5668 tmp = list_entry(delete_list.next, struct extent_record, list);
5669 list_del_init(&tmp->list);
5675 while (!list_empty(&rec->dups)) {
5676 tmp = list_entry(rec->dups.next, struct extent_record, list);
5677 list_del_init(&tmp->list);
5681 btrfs_free_path(path);
5683 if (!ret && !nr_del)
5684 rec->num_duplicates = 0;
5686 return ret ? ret : nr_del;
5689 static int find_possible_backrefs(struct btrfs_trans_handle *trans,
5690 struct btrfs_fs_info *info,
5691 struct btrfs_path *path,
5692 struct cache_tree *extent_cache,
5693 struct extent_record *rec)
5695 struct btrfs_root *root;
5696 struct extent_backref *back;
5697 struct data_backref *dback;
5698 struct cache_extent *cache;
5699 struct btrfs_file_extent_item *fi;
5700 struct btrfs_key key;
5704 list_for_each_entry(back, &rec->backrefs, list) {
5705 /* Don't care about full backrefs (poor unloved backrefs) */
5706 if (back->full_backref || !back->is_data)
5709 dback = (struct data_backref *)back;
5711 /* We found this one, we don't need to do a lookup */
5712 if (dback->found_ref)
5715 key.objectid = dback->root;
5716 key.type = BTRFS_ROOT_ITEM_KEY;
5717 key.offset = (u64)-1;
5719 root = btrfs_read_fs_root(info, &key);
5721 /* No root, definitely a bad ref, skip */
5722 if (IS_ERR(root) && PTR_ERR(root) == -ENOENT)
5724 /* Other err, exit */
5726 return PTR_ERR(root);
5728 key.objectid = dback->owner;
5729 key.type = BTRFS_EXTENT_DATA_KEY;
5730 key.offset = dback->offset;
5731 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5733 btrfs_release_path(path);
5736 /* Didn't find it, we can carry on */
5741 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5742 struct btrfs_file_extent_item);
5743 bytenr = btrfs_file_extent_disk_bytenr(path->nodes[0], fi);
5744 bytes = btrfs_file_extent_disk_num_bytes(path->nodes[0], fi);
5745 btrfs_release_path(path);
5746 cache = lookup_cache_extent(extent_cache, bytenr, 1);
5748 struct extent_record *tmp;
5749 tmp = container_of(cache, struct extent_record, cache);
5752 * If we found an extent record for the bytenr for this
5753 * particular backref then we can't add it to our
5754 * current extent record. We only want to add backrefs
5755 * that don't have a corresponding extent item in the
5756 * extent tree since they likely belong to this record
5757 * and we need to fix it if it doesn't match bytenrs.
5763 dback->found_ref += 1;
5764 dback->disk_bytenr = bytenr;
5765 dback->bytes = bytes;
5768 * Set this so the verify backref code knows not to trust the
5769 * values in this backref.
5778 * when an incorrect extent item is found, this will delete
5779 * all of the existing entries for it and recreate them
5780 * based on what the tree scan found.
5782 static int fixup_extent_refs(struct btrfs_trans_handle *trans,
5783 struct btrfs_fs_info *info,
5784 struct cache_tree *extent_cache,
5785 struct extent_record *rec)
5788 struct btrfs_path *path;
5789 struct list_head *cur = rec->backrefs.next;
5790 struct cache_extent *cache;
5791 struct extent_backref *back;
5796 * remember our flags for recreating the extent.
5797 * FIXME, if we have cleared extent tree, we can not
5798 * lookup extent info in extent tree.
5800 if (!init_extent_tree) {
5801 ret = btrfs_lookup_extent_info(NULL, info->extent_root,
5802 rec->start, rec->max_size,
5803 rec->metadata, NULL, &flags);
5810 path = btrfs_alloc_path();
5814 if (rec->refs != rec->extent_item_refs && !rec->metadata) {
5816 * Sometimes the backrefs themselves are so broken they don't
5817 * get attached to any meaningful rec, so first go back and
5818 * check any of our backrefs that we couldn't find and throw
5819 * them into the list if we find the backref so that
5820 * verify_backrefs can figure out what to do.
5822 ret = find_possible_backrefs(trans, info, path, extent_cache,
5828 /* step one, make sure all of the backrefs agree */
5829 ret = verify_backrefs(trans, info, path, rec);
5833 /* step two, delete all the existing records */
5834 ret = delete_extent_records(trans, info->extent_root, path,
5835 rec->start, rec->max_size);
5840 /* was this block corrupt? If so, don't add references to it */
5841 cache = lookup_cache_extent(info->corrupt_blocks,
5842 rec->start, rec->max_size);
5848 /* step three, recreate all the refs we did find */
5849 while(cur != &rec->backrefs) {
5850 back = list_entry(cur, struct extent_backref, list);
5854 * if we didn't find any references, don't create a
5857 if (!back->found_ref)
5860 ret = record_extent(trans, info, path, rec, back, allocated, flags);
5867 btrfs_free_path(path);
5871 /* right now we only prune from the extent allocation tree */
5872 static int prune_one_block(struct btrfs_trans_handle *trans,
5873 struct btrfs_fs_info *info,
5874 struct btrfs_corrupt_block *corrupt)
5877 struct btrfs_path path;
5878 struct extent_buffer *eb;
5882 int level = corrupt->level + 1;
5884 btrfs_init_path(&path);
5886 /* we want to stop at the parent to our busted block */
5887 path.lowest_level = level;
5889 ret = btrfs_search_slot(trans, info->extent_root,
5890 &corrupt->key, &path, -1, 1);
5895 eb = path.nodes[level];
5902 * hopefully the search gave us the block we want to prune,
5903 * lets try that first
5905 slot = path.slots[level];
5906 found = btrfs_node_blockptr(eb, slot);
5907 if (found == corrupt->cache.start)
5910 nritems = btrfs_header_nritems(eb);
5912 /* the search failed, lets scan this node and hope we find it */
5913 for (slot = 0; slot < nritems; slot++) {
5914 found = btrfs_node_blockptr(eb, slot);
5915 if (found == corrupt->cache.start)
5919 * we couldn't find the bad block. TODO, search all the nodes for pointers
5922 if (eb == info->extent_root->node) {
5927 btrfs_release_path(&path);
5932 printk("deleting pointer to block %Lu\n", corrupt->cache.start);
5933 ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
5936 btrfs_release_path(&path);
5940 static int prune_corrupt_blocks(struct btrfs_trans_handle *trans,
5941 struct btrfs_fs_info *info)
5943 struct cache_extent *cache;
5944 struct btrfs_corrupt_block *corrupt;
5946 cache = search_cache_extent(info->corrupt_blocks, 0);
5950 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
5951 prune_one_block(trans, info, corrupt);
5952 cache = next_cache_extent(cache);
5957 static void free_corrupt_block(struct cache_extent *cache)
5959 struct btrfs_corrupt_block *corrupt;
5961 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
5965 FREE_EXTENT_CACHE_BASED_TREE(corrupt_blocks, free_corrupt_block);
5967 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
5969 struct btrfs_block_group_cache *cache;
5974 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
5975 &start, &end, EXTENT_DIRTY);
5978 clear_extent_dirty(&fs_info->free_space_cache, start, end,
5984 cache = btrfs_lookup_first_block_group(fs_info, start);
5989 start = cache->key.objectid + cache->key.offset;
5993 static int check_extent_refs(struct btrfs_trans_handle *trans,
5994 struct btrfs_root *root,
5995 struct cache_tree *extent_cache)
5997 struct extent_record *rec;
5998 struct cache_extent *cache;
6006 * if we're doing a repair, we have to make sure
6007 * we don't allocate from the problem extents.
6008 * In the worst case, this will be all the
6011 cache = search_cache_extent(extent_cache, 0);
6013 rec = container_of(cache, struct extent_record, cache);
6014 btrfs_pin_extent(root->fs_info,
6015 rec->start, rec->max_size);
6016 cache = next_cache_extent(cache);
6019 /* pin down all the corrupted blocks too */
6020 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
6022 btrfs_pin_extent(root->fs_info,
6023 cache->start, cache->size);
6024 cache = next_cache_extent(cache);
6026 prune_corrupt_blocks(trans, root->fs_info);
6027 reset_cached_block_groups(root->fs_info);
6031 * We need to delete any duplicate entries we find first otherwise we
6032 * could mess up the extent tree when we have backrefs that actually
6033 * belong to a different extent item and not the weird duplicate one.
6035 while (repair && !list_empty(&duplicate_extents)) {
6036 rec = list_entry(duplicate_extents.next, struct extent_record,
6038 list_del_init(&rec->list);
6040 /* Sometimes we can find a backref before we find an actual
6041 * extent, so we need to process it a little bit to see if there
6042 * truly are multiple EXTENT_ITEM_KEY's for the same range, or
6043 * if this is a backref screwup. If we need to delete stuff
6044 * process_duplicates() will return 0, otherwise it will return
6047 if (process_duplicates(root, extent_cache, rec))
6049 ret = delete_duplicate_records(trans, root, rec);
6053 * delete_duplicate_records will return the number of entries
6054 * deleted, so if it's greater than 0 then we know we actually
6055 * did something and we need to remove.
6066 cache = search_cache_extent(extent_cache, 0);
6069 rec = container_of(cache, struct extent_record, cache);
6070 if (rec->num_duplicates) {
6071 fprintf(stderr, "extent item %llu has multiple extent "
6072 "items\n", (unsigned long long)rec->start);
6076 if (rec->refs != rec->extent_item_refs) {
6077 fprintf(stderr, "ref mismatch on [%llu %llu] ",
6078 (unsigned long long)rec->start,
6079 (unsigned long long)rec->nr);
6080 fprintf(stderr, "extent item %llu, found %llu\n",
6081 (unsigned long long)rec->extent_item_refs,
6082 (unsigned long long)rec->refs);
6083 if (!fixed && repair) {
6084 ret = fixup_extent_refs(trans, root->fs_info,
6093 if (all_backpointers_checked(rec, 1)) {
6094 fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
6095 (unsigned long long)rec->start,
6096 (unsigned long long)rec->nr);
6098 if (!fixed && repair) {
6099 ret = fixup_extent_refs(trans, root->fs_info,
6108 if (!rec->owner_ref_checked) {
6109 fprintf(stderr, "owner ref check failed [%llu %llu]\n",
6110 (unsigned long long)rec->start,
6111 (unsigned long long)rec->nr);
6112 if (!fixed && repair) {
6113 ret = fixup_extent_refs(trans, root->fs_info,
6122 remove_cache_extent(extent_cache, cache);
6123 free_all_extent_backrefs(rec);
6128 if (ret && ret != -EAGAIN) {
6129 fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
6132 btrfs_fix_block_accounting(trans, root);
6135 fprintf(stderr, "repaired damaged extent references\n");
6141 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
6145 if (type & BTRFS_BLOCK_GROUP_RAID0) {
6146 stripe_size = length;
6147 stripe_size /= num_stripes;
6148 } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
6149 stripe_size = length * 2;
6150 stripe_size /= num_stripes;
6151 } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
6152 stripe_size = length;
6153 stripe_size /= (num_stripes - 1);
6154 } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
6155 stripe_size = length;
6156 stripe_size /= (num_stripes - 2);
6158 stripe_size = length;
6163 static int check_chunk_refs(struct chunk_record *chunk_rec,
6164 struct block_group_tree *block_group_cache,
6165 struct device_extent_tree *dev_extent_cache,
6168 struct cache_extent *block_group_item;
6169 struct block_group_record *block_group_rec;
6170 struct cache_extent *dev_extent_item;
6171 struct device_extent_record *dev_extent_rec;
6178 block_group_item = lookup_cache_extent(&block_group_cache->tree,
6181 if (block_group_item) {
6182 block_group_rec = container_of(block_group_item,
6183 struct block_group_record,
6185 if (chunk_rec->length != block_group_rec->offset ||
6186 chunk_rec->offset != block_group_rec->objectid ||
6187 chunk_rec->type_flags != block_group_rec->flags) {
6190 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
6191 chunk_rec->objectid,
6196 chunk_rec->type_flags,
6197 block_group_rec->objectid,
6198 block_group_rec->type,
6199 block_group_rec->offset,
6200 block_group_rec->offset,
6201 block_group_rec->objectid,
6202 block_group_rec->flags);
6205 list_del_init(&block_group_rec->list);
6206 chunk_rec->bg_rec = block_group_rec;
6211 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
6212 chunk_rec->objectid,
6217 chunk_rec->type_flags);
6221 length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
6222 chunk_rec->num_stripes);
6223 for (i = 0; i < chunk_rec->num_stripes; ++i) {
6224 devid = chunk_rec->stripes[i].devid;
6225 offset = chunk_rec->stripes[i].offset;
6226 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
6227 devid, offset, length);
6228 if (dev_extent_item) {
6229 dev_extent_rec = container_of(dev_extent_item,
6230 struct device_extent_record,
6232 if (dev_extent_rec->objectid != devid ||
6233 dev_extent_rec->offset != offset ||
6234 dev_extent_rec->chunk_offset != chunk_rec->offset ||
6235 dev_extent_rec->length != length) {
6238 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
6239 chunk_rec->objectid,
6242 chunk_rec->stripes[i].devid,
6243 chunk_rec->stripes[i].offset,
6244 dev_extent_rec->objectid,
6245 dev_extent_rec->offset,
6246 dev_extent_rec->length);
6249 list_move(&dev_extent_rec->chunk_list,
6250 &chunk_rec->dextents);
6255 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
6256 chunk_rec->objectid,
6259 chunk_rec->stripes[i].devid,
6260 chunk_rec->stripes[i].offset);
6267 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
6268 int check_chunks(struct cache_tree *chunk_cache,
6269 struct block_group_tree *block_group_cache,
6270 struct device_extent_tree *dev_extent_cache,
6271 struct list_head *good, struct list_head *bad, int silent)
6273 struct cache_extent *chunk_item;
6274 struct chunk_record *chunk_rec;
6275 struct block_group_record *bg_rec;
6276 struct device_extent_record *dext_rec;
6280 chunk_item = first_cache_extent(chunk_cache);
6281 while (chunk_item) {
6282 chunk_rec = container_of(chunk_item, struct chunk_record,
6284 err = check_chunk_refs(chunk_rec, block_group_cache,
6285 dev_extent_cache, silent);
6289 list_add_tail(&chunk_rec->list, bad);
6292 list_add_tail(&chunk_rec->list, good);
6295 chunk_item = next_cache_extent(chunk_item);
6298 list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
6301 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
6309 list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
6313 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
6324 static int check_device_used(struct device_record *dev_rec,
6325 struct device_extent_tree *dext_cache)
6327 struct cache_extent *cache;
6328 struct device_extent_record *dev_extent_rec;
6331 cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
6333 dev_extent_rec = container_of(cache,
6334 struct device_extent_record,
6336 if (dev_extent_rec->objectid != dev_rec->devid)
6339 list_del_init(&dev_extent_rec->device_list);
6340 total_byte += dev_extent_rec->length;
6341 cache = next_cache_extent(cache);
6344 if (total_byte != dev_rec->byte_used) {
6346 "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
6347 total_byte, dev_rec->byte_used, dev_rec->objectid,
6348 dev_rec->type, dev_rec->offset);
6355 /* check btrfs_dev_item -> btrfs_dev_extent */
6356 static int check_devices(struct rb_root *dev_cache,
6357 struct device_extent_tree *dev_extent_cache)
6359 struct rb_node *dev_node;
6360 struct device_record *dev_rec;
6361 struct device_extent_record *dext_rec;
6365 dev_node = rb_first(dev_cache);
6367 dev_rec = container_of(dev_node, struct device_record, node);
6368 err = check_device_used(dev_rec, dev_extent_cache);
6372 dev_node = rb_next(dev_node);
6374 list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
6377 "Device extent[%llu, %llu, %llu] didn't find its device.\n",
6378 dext_rec->objectid, dext_rec->offset, dext_rec->length);
6385 static int check_chunks_and_extents(struct btrfs_root *root)
6387 struct rb_root dev_cache;
6388 struct cache_tree chunk_cache;
6389 struct block_group_tree block_group_cache;
6390 struct device_extent_tree dev_extent_cache;
6391 struct cache_tree extent_cache;
6392 struct cache_tree seen;
6393 struct cache_tree pending;
6394 struct cache_tree reada;
6395 struct cache_tree nodes;
6396 struct cache_tree corrupt_blocks;
6397 struct btrfs_path path;
6398 struct btrfs_key key;
6399 struct btrfs_key found_key;
6402 struct block_info *bits;
6404 struct extent_buffer *leaf;
6405 struct btrfs_trans_handle *trans = NULL;
6407 struct btrfs_root_item ri;
6408 struct list_head dropping_trees;
6410 dev_cache = RB_ROOT;
6411 cache_tree_init(&chunk_cache);
6412 block_group_tree_init(&block_group_cache);
6413 device_extent_tree_init(&dev_extent_cache);
6415 cache_tree_init(&extent_cache);
6416 cache_tree_init(&seen);
6417 cache_tree_init(&pending);
6418 cache_tree_init(&nodes);
6419 cache_tree_init(&reada);
6420 cache_tree_init(&corrupt_blocks);
6421 INIT_LIST_HEAD(&dropping_trees);
6424 trans = btrfs_start_transaction(root, 1);
6425 if (IS_ERR(trans)) {
6426 fprintf(stderr, "Error starting transaction\n");
6427 return PTR_ERR(trans);
6429 root->fs_info->fsck_extent_cache = &extent_cache;
6430 root->fs_info->free_extent_hook = free_extent_hook;
6431 root->fs_info->corrupt_blocks = &corrupt_blocks;
6435 bits = malloc(bits_nr * sizeof(struct block_info));
6442 add_root_to_pending(root->fs_info->tree_root->node,
6443 &extent_cache, &pending, &seen, &nodes,
6444 &root->fs_info->tree_root->root_key);
6446 add_root_to_pending(root->fs_info->chunk_root->node,
6447 &extent_cache, &pending, &seen, &nodes,
6448 &root->fs_info->chunk_root->root_key);
6450 btrfs_init_path(&path);
6453 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
6454 ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
6459 leaf = path.nodes[0];
6460 slot = path.slots[0];
6461 if (slot >= btrfs_header_nritems(path.nodes[0])) {
6462 ret = btrfs_next_leaf(root, &path);
6465 leaf = path.nodes[0];
6466 slot = path.slots[0];
6468 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
6469 if (btrfs_key_type(&found_key) == BTRFS_ROOT_ITEM_KEY) {
6470 unsigned long offset;
6471 struct extent_buffer *buf;
6473 offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
6474 read_extent_buffer(leaf, &ri, offset, sizeof(ri));
6475 if (btrfs_disk_key_objectid(&ri.drop_progress) == 0) {
6476 buf = read_tree_block(root->fs_info->tree_root,
6477 btrfs_root_bytenr(&ri),
6478 btrfs_level_size(root,
6479 btrfs_root_level(&ri)),
6485 add_root_to_pending(buf, &extent_cache,
6486 &pending, &seen, &nodes,
6488 free_extent_buffer(buf);
6490 struct dropping_root_item_record *dri_rec;
6491 dri_rec = malloc(sizeof(*dri_rec));
6496 memcpy(&dri_rec->ri, &ri, sizeof(ri));
6497 memcpy(&dri_rec->found_key, &found_key,
6499 list_add_tail(&dri_rec->list, &dropping_trees);
6504 btrfs_release_path(&path);
6506 ret = run_next_block(trans, root, bits, bits_nr, &last,
6507 &pending, &seen, &reada, &nodes,
6508 &extent_cache, &chunk_cache, &dev_cache,
6509 &block_group_cache, &dev_extent_cache,
6515 while (!list_empty(&dropping_trees)) {
6516 struct dropping_root_item_record *rec;
6517 struct extent_buffer *buf;
6518 rec = list_entry(dropping_trees.next,
6519 struct dropping_root_item_record, list);
6525 buf = read_tree_block(root->fs_info->tree_root,
6526 btrfs_root_bytenr(&rec->ri),
6527 btrfs_level_size(root,
6528 btrfs_root_level(&rec->ri)), 0);
6533 add_root_to_pending(buf, &extent_cache, &pending,
6534 &seen, &nodes, &rec->found_key);
6536 ret = run_next_block(trans, root, bits, bits_nr, &last,
6537 &pending, &seen, &reada,
6538 &nodes, &extent_cache,
6539 &chunk_cache, &dev_cache,
6546 free_extent_buffer(buf);
6547 list_del(&rec->list);
6552 ret = check_extent_refs(trans, root, &extent_cache);
6553 if (ret == -EAGAIN) {
6554 ret = btrfs_commit_transaction(trans, root);
6558 trans = btrfs_start_transaction(root, 1);
6559 if (IS_ERR(trans)) {
6560 ret = PTR_ERR(trans);
6564 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
6565 free_extent_cache_tree(&seen);
6566 free_extent_cache_tree(&pending);
6567 free_extent_cache_tree(&reada);
6568 free_extent_cache_tree(&nodes);
6569 free_chunk_cache_tree(&chunk_cache);
6570 free_block_group_tree(&block_group_cache);
6571 free_device_cache_tree(&dev_cache);
6572 free_device_extent_tree(&dev_extent_cache);
6573 free_extent_record_cache(root->fs_info, &extent_cache);
6577 err = check_chunks(&chunk_cache, &block_group_cache,
6578 &dev_extent_cache, NULL, NULL, 0);
6582 err = check_devices(&dev_cache, &dev_extent_cache);
6588 err = btrfs_commit_transaction(trans, root);
6593 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
6594 root->fs_info->fsck_extent_cache = NULL;
6595 root->fs_info->free_extent_hook = NULL;
6596 root->fs_info->corrupt_blocks = NULL;
6599 free_chunk_cache_tree(&chunk_cache);
6600 free_device_cache_tree(&dev_cache);
6601 free_block_group_tree(&block_group_cache);
6602 free_device_extent_tree(&dev_extent_cache);
6603 free_extent_cache_tree(&seen);
6604 free_extent_cache_tree(&pending);
6605 free_extent_cache_tree(&reada);
6606 free_extent_cache_tree(&nodes);
6610 static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
6611 struct btrfs_root *root, int overwrite)
6613 struct extent_buffer *c;
6614 struct extent_buffer *old = root->node;
6617 struct btrfs_disk_key disk_key = {0,0,0};
6623 extent_buffer_get(c);
6626 c = btrfs_alloc_free_block(trans, root,
6627 btrfs_level_size(root, 0),
6628 root->root_key.objectid,
6629 &disk_key, level, 0, 0);
6632 extent_buffer_get(c);
6636 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
6637 btrfs_set_header_level(c, level);
6638 btrfs_set_header_bytenr(c, c->start);
6639 btrfs_set_header_generation(c, trans->transid);
6640 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
6641 btrfs_set_header_owner(c, root->root_key.objectid);
6643 write_extent_buffer(c, root->fs_info->fsid,
6644 btrfs_header_fsid(), BTRFS_FSID_SIZE);
6646 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
6647 btrfs_header_chunk_tree_uuid(c),
6650 btrfs_mark_buffer_dirty(c);
6652 * this case can happen in the following case:
6654 * 1.overwrite previous root.
6656 * 2.reinit reloc data root, this is because we skip pin
6657 * down reloc data tree before which means we can allocate
6658 * same block bytenr here.
6660 if (old->start == c->start) {
6661 btrfs_set_root_generation(&root->root_item,
6663 root->root_item.level = btrfs_header_level(root->node);
6664 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6665 &root->root_key, &root->root_item);
6667 free_extent_buffer(c);
6671 free_extent_buffer(old);
6673 add_root_to_dirty_list(root);
6677 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
6678 struct extent_buffer *eb, int tree_root)
6680 struct extent_buffer *tmp;
6681 struct btrfs_root_item *ri;
6682 struct btrfs_key key;
6685 int level = btrfs_header_level(eb);
6691 * If we have pinned this block before, don't pin it again.
6692 * This can not only avoid forever loop with broken filesystem
6693 * but also give us some speedups.
6695 if (test_range_bit(&fs_info->pinned_extents, eb->start,
6696 eb->start + eb->len - 1, EXTENT_DIRTY, 0))
6699 btrfs_pin_extent(fs_info, eb->start, eb->len);
6701 leafsize = btrfs_super_leafsize(fs_info->super_copy);
6702 nritems = btrfs_header_nritems(eb);
6703 for (i = 0; i < nritems; i++) {
6705 btrfs_item_key_to_cpu(eb, &key, i);
6706 if (key.type != BTRFS_ROOT_ITEM_KEY)
6708 /* Skip the extent root and reloc roots */
6709 if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6710 key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
6711 key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
6713 ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
6714 bytenr = btrfs_disk_root_bytenr(eb, ri);
6717 * If at any point we start needing the real root we
6718 * will have to build a stump root for the root we are
6719 * in, but for now this doesn't actually use the root so
6720 * just pass in extent_root.
6722 tmp = read_tree_block(fs_info->extent_root, bytenr,
6725 fprintf(stderr, "Error reading root block\n");
6728 ret = pin_down_tree_blocks(fs_info, tmp, 0);
6729 free_extent_buffer(tmp);
6733 bytenr = btrfs_node_blockptr(eb, i);
6735 /* If we aren't the tree root don't read the block */
6736 if (level == 1 && !tree_root) {
6737 btrfs_pin_extent(fs_info, bytenr, leafsize);
6741 tmp = read_tree_block(fs_info->extent_root, bytenr,
6744 fprintf(stderr, "Error reading tree block\n");
6747 ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
6748 free_extent_buffer(tmp);
6757 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
6761 ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
6765 return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
6768 static int reset_block_groups(struct btrfs_fs_info *fs_info)
6770 struct btrfs_block_group_cache *cache;
6771 struct btrfs_path *path;
6772 struct extent_buffer *leaf;
6773 struct btrfs_chunk *chunk;
6774 struct btrfs_key key;
6778 path = btrfs_alloc_path();
6783 key.type = BTRFS_CHUNK_ITEM_KEY;
6786 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
6788 btrfs_free_path(path);
6793 * We do this in case the block groups were screwed up and had alloc
6794 * bits that aren't actually set on the chunks. This happens with
6795 * restored images every time and could happen in real life I guess.
6797 fs_info->avail_data_alloc_bits = 0;
6798 fs_info->avail_metadata_alloc_bits = 0;
6799 fs_info->avail_system_alloc_bits = 0;
6801 /* First we need to create the in-memory block groups */
6803 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
6804 ret = btrfs_next_leaf(fs_info->chunk_root, path);
6806 btrfs_free_path(path);
6814 leaf = path->nodes[0];
6815 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6816 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
6821 chunk = btrfs_item_ptr(leaf, path->slots[0],
6822 struct btrfs_chunk);
6823 btrfs_add_block_group(fs_info, 0,
6824 btrfs_chunk_type(leaf, chunk),
6825 key.objectid, key.offset,
6826 btrfs_chunk_length(leaf, chunk));
6827 set_extent_dirty(&fs_info->free_space_cache, key.offset,
6828 key.offset + btrfs_chunk_length(leaf, chunk),
6834 cache = btrfs_lookup_first_block_group(fs_info, start);
6838 start = cache->key.objectid + cache->key.offset;
6841 btrfs_free_path(path);
6845 static int reset_balance(struct btrfs_trans_handle *trans,
6846 struct btrfs_fs_info *fs_info)
6848 struct btrfs_root *root = fs_info->tree_root;
6849 struct btrfs_path *path;
6850 struct extent_buffer *leaf;
6851 struct btrfs_key key;
6852 int del_slot, del_nr = 0;
6856 path = btrfs_alloc_path();
6860 key.objectid = BTRFS_BALANCE_OBJECTID;
6861 key.type = BTRFS_BALANCE_ITEM_KEY;
6864 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6869 goto reinit_data_reloc;
6874 ret = btrfs_del_item(trans, root, path);
6877 btrfs_release_path(path);
6879 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6880 key.type = BTRFS_ROOT_ITEM_KEY;
6883 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6887 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
6892 ret = btrfs_del_items(trans, root, path,
6899 btrfs_release_path(path);
6902 ret = btrfs_search_slot(trans, root, &key, path,
6909 leaf = path->nodes[0];
6910 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6911 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
6913 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6918 del_slot = path->slots[0];
6927 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
6931 btrfs_release_path(path);
6934 key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
6935 key.type = BTRFS_ROOT_ITEM_KEY;
6936 key.offset = (u64)-1;
6937 root = btrfs_read_fs_root(fs_info, &key);
6939 fprintf(stderr, "Error reading data reloc tree\n");
6940 return PTR_ERR(root);
6942 record_root_in_trans(trans, root);
6943 ret = btrfs_fsck_reinit_root(trans, root, 0);
6946 ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
6948 btrfs_free_path(path);
6952 static int reinit_extent_tree(struct btrfs_trans_handle *trans,
6953 struct btrfs_fs_info *fs_info)
6959 * The only reason we don't do this is because right now we're just
6960 * walking the trees we find and pinning down their bytes, we don't look
6961 * at any of the leaves. In order to do mixed groups we'd have to check
6962 * the leaves of any fs roots and pin down the bytes for any file
6963 * extents we find. Not hard but why do it if we don't have to?
6965 if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
6966 fprintf(stderr, "We don't support re-initing the extent tree "
6967 "for mixed block groups yet, please notify a btrfs "
6968 "developer you want to do this so they can add this "
6969 "functionality.\n");
6974 * first we need to walk all of the trees except the extent tree and pin
6975 * down the bytes that are in use so we don't overwrite any existing
6978 ret = pin_metadata_blocks(fs_info);
6980 fprintf(stderr, "error pinning down used bytes\n");
6985 * Need to drop all the block groups since we're going to recreate all
6988 btrfs_free_block_groups(fs_info);
6989 ret = reset_block_groups(fs_info);
6991 fprintf(stderr, "error resetting the block groups\n");
6995 /* Ok we can allocate now, reinit the extent root */
6996 ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 0);
6998 fprintf(stderr, "extent root initialization failed\n");
7000 * When the transaction code is updated we should end the
7001 * transaction, but for now progs only knows about commit so
7002 * just return an error.
7008 * Now we have all the in-memory block groups setup so we can make
7009 * allocations properly, and the metadata we care about is safe since we
7010 * pinned all of it above.
7013 struct btrfs_block_group_cache *cache;
7015 cache = btrfs_lookup_first_block_group(fs_info, start);
7018 start = cache->key.objectid + cache->key.offset;
7019 ret = btrfs_insert_item(trans, fs_info->extent_root,
7020 &cache->key, &cache->item,
7021 sizeof(cache->item));
7023 fprintf(stderr, "Error adding block group\n");
7026 btrfs_extent_post_op(trans, fs_info->extent_root);
7029 ret = reset_balance(trans, fs_info);
7031 fprintf(stderr, "error reseting the pending balance\n");
7036 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
7038 struct btrfs_path *path;
7039 struct btrfs_trans_handle *trans;
7040 struct btrfs_key key;
7043 printf("Recowing metadata block %llu\n", eb->start);
7044 key.objectid = btrfs_header_owner(eb);
7045 key.type = BTRFS_ROOT_ITEM_KEY;
7046 key.offset = (u64)-1;
7048 root = btrfs_read_fs_root(root->fs_info, &key);
7050 fprintf(stderr, "Couldn't find owner root %llu\n",
7052 return PTR_ERR(root);
7055 path = btrfs_alloc_path();
7059 trans = btrfs_start_transaction(root, 1);
7060 if (IS_ERR(trans)) {
7061 btrfs_free_path(path);
7062 return PTR_ERR(trans);
7065 path->lowest_level = btrfs_header_level(eb);
7066 if (path->lowest_level)
7067 btrfs_node_key_to_cpu(eb, &key, 0);
7069 btrfs_item_key_to_cpu(eb, &key, 0);
7071 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7072 btrfs_commit_transaction(trans, root);
7073 btrfs_free_path(path);
7077 static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
7079 struct btrfs_path *path;
7080 struct btrfs_trans_handle *trans;
7081 struct btrfs_key key;
7084 printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
7085 bad->key.type, bad->key.offset);
7086 key.objectid = bad->root_id;
7087 key.type = BTRFS_ROOT_ITEM_KEY;
7088 key.offset = (u64)-1;
7090 root = btrfs_read_fs_root(root->fs_info, &key);
7092 fprintf(stderr, "Couldn't find owner root %llu\n",
7094 return PTR_ERR(root);
7097 path = btrfs_alloc_path();
7101 trans = btrfs_start_transaction(root, 1);
7102 if (IS_ERR(trans)) {
7103 btrfs_free_path(path);
7104 return PTR_ERR(trans);
7107 ret = btrfs_search_slot(trans, root, &bad->key, path, -1, 1);
7113 ret = btrfs_del_item(trans, root, path);
7115 btrfs_commit_transaction(trans, root);
7116 btrfs_free_path(path);
7120 static int zero_log_tree(struct btrfs_root *root)
7122 struct btrfs_trans_handle *trans;
7125 trans = btrfs_start_transaction(root, 1);
7126 if (IS_ERR(trans)) {
7127 ret = PTR_ERR(trans);
7130 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
7131 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
7132 ret = btrfs_commit_transaction(trans, root);
7136 static int populate_csum(struct btrfs_trans_handle *trans,
7137 struct btrfs_root *csum_root, char *buf, u64 start,
7144 while (offset < len) {
7145 sectorsize = csum_root->sectorsize;
7146 ret = read_extent_data(csum_root, buf, start + offset,
7150 ret = btrfs_csum_file_block(trans, csum_root, start + len,
7151 start + offset, buf, sectorsize);
7154 offset += sectorsize;
7159 static int fill_csum_tree(struct btrfs_trans_handle *trans,
7160 struct btrfs_root *csum_root)
7162 struct btrfs_root *extent_root = csum_root->fs_info->extent_root;
7163 struct btrfs_path *path;
7164 struct btrfs_extent_item *ei;
7165 struct extent_buffer *leaf;
7167 struct btrfs_key key;
7170 path = btrfs_alloc_path();
7175 key.type = BTRFS_EXTENT_ITEM_KEY;
7178 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
7180 btrfs_free_path(path);
7184 buf = malloc(csum_root->sectorsize);
7186 btrfs_free_path(path);
7191 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7192 ret = btrfs_next_leaf(extent_root, path);
7200 leaf = path->nodes[0];
7202 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7203 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
7208 ei = btrfs_item_ptr(leaf, path->slots[0],
7209 struct btrfs_extent_item);
7210 if (!(btrfs_extent_flags(leaf, ei) &
7211 BTRFS_EXTENT_FLAG_DATA)) {
7216 ret = populate_csum(trans, csum_root, buf, key.objectid,
7223 btrfs_free_path(path);
7228 struct root_item_info {
7229 /* level of the root */
7231 /* number of nodes at this level, must be 1 for a root */
7235 struct cache_extent cache_extent;
7238 static struct cache_tree *roots_info_cache = NULL;
7240 static void free_roots_info_cache(void)
7242 if (!roots_info_cache)
7245 while (!cache_tree_empty(roots_info_cache)) {
7246 struct cache_extent *entry;
7247 struct root_item_info *rii;
7249 entry = first_cache_extent(roots_info_cache);
7250 remove_cache_extent(roots_info_cache, entry);
7251 rii = container_of(entry, struct root_item_info, cache_extent);
7255 free(roots_info_cache);
7256 roots_info_cache = NULL;
7259 static int build_roots_info_cache(struct btrfs_fs_info *info)
7262 struct btrfs_key key;
7263 struct extent_buffer *leaf;
7264 struct btrfs_path *path;
7266 if (!roots_info_cache) {
7267 roots_info_cache = malloc(sizeof(*roots_info_cache));
7268 if (!roots_info_cache)
7270 cache_tree_init(roots_info_cache);
7273 path = btrfs_alloc_path();
7278 key.type = BTRFS_EXTENT_ITEM_KEY;
7281 ret = btrfs_search_slot(NULL, info->extent_root, &key, path, 0, 0);
7284 leaf = path->nodes[0];
7287 struct btrfs_key found_key;
7288 struct btrfs_extent_item *ei;
7289 struct btrfs_extent_inline_ref *iref;
7290 int slot = path->slots[0];
7295 struct cache_extent *entry;
7296 struct root_item_info *rii;
7298 if (slot >= btrfs_header_nritems(leaf)) {
7299 ret = btrfs_next_leaf(info->extent_root, path);
7306 leaf = path->nodes[0];
7307 slot = path->slots[0];
7310 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7312 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
7313 found_key.type != BTRFS_METADATA_ITEM_KEY)
7316 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
7317 flags = btrfs_extent_flags(leaf, ei);
7319 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
7320 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
7323 if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
7324 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
7325 level = found_key.offset;
7327 struct btrfs_tree_block_info *info;
7329 info = (struct btrfs_tree_block_info *)(ei + 1);
7330 iref = (struct btrfs_extent_inline_ref *)(info + 1);
7331 level = btrfs_tree_block_level(leaf, info);
7335 * For a root extent, it must be of the following type and the
7336 * first (and only one) iref in the item.
7338 type = btrfs_extent_inline_ref_type(leaf, iref);
7339 if (type != BTRFS_TREE_BLOCK_REF_KEY)
7342 root_id = btrfs_extent_inline_ref_offset(leaf, iref);
7343 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
7345 rii = malloc(sizeof(struct root_item_info));
7350 rii->cache_extent.start = root_id;
7351 rii->cache_extent.size = 1;
7352 rii->level = (u8)-1;
7353 entry = &rii->cache_extent;
7354 ret = insert_cache_extent(roots_info_cache, entry);
7357 rii = container_of(entry, struct root_item_info,
7361 ASSERT(rii->cache_extent.start == root_id);
7362 ASSERT(rii->cache_extent.size == 1);
7364 if (level > rii->level || rii->level == (u8)-1) {
7366 rii->bytenr = found_key.objectid;
7367 rii->gen = btrfs_extent_generation(leaf, ei);
7368 rii->node_count = 1;
7369 } else if (level == rii->level) {
7377 btrfs_free_path(path);
7382 static int maybe_repair_root_item(struct btrfs_fs_info *info,
7383 struct btrfs_path *path,
7384 const struct btrfs_key *root_key,
7385 const int read_only_mode)
7387 const u64 root_id = root_key->objectid;
7388 struct cache_extent *entry;
7389 struct root_item_info *rii;
7390 struct btrfs_root_item ri;
7391 unsigned long offset;
7393 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
7396 "Error: could not find extent items for root %llu\n",
7397 root_key->objectid);
7401 rii = container_of(entry, struct root_item_info, cache_extent);
7402 ASSERT(rii->cache_extent.start == root_id);
7403 ASSERT(rii->cache_extent.size == 1);
7405 if (rii->node_count != 1) {
7407 "Error: could not find btree root extent for root %llu\n",
7412 offset = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
7413 read_extent_buffer(path->nodes[0], &ri, offset, sizeof(ri));
7415 if (btrfs_root_bytenr(&ri) != rii->bytenr ||
7416 btrfs_root_level(&ri) != rii->level ||
7417 btrfs_root_generation(&ri) != rii->gen) {
7420 * If we're in repair mode but our caller told us to not update
7421 * the root item, i.e. just check if it needs to be updated, don't
7422 * print this message, since the caller will call us again shortly
7423 * for the same root item without read only mode (the caller will
7424 * open a transaction first).
7426 if (!(read_only_mode && repair))
7428 "%sroot item for root %llu,"
7429 " current bytenr %llu, current gen %llu, current level %u,"
7430 " new bytenr %llu, new gen %llu, new level %u\n",
7431 (read_only_mode ? "" : "fixing "),
7433 btrfs_root_bytenr(&ri), btrfs_root_generation(&ri),
7434 btrfs_root_level(&ri),
7435 rii->bytenr, rii->gen, rii->level);
7437 if (btrfs_root_generation(&ri) > rii->gen) {
7439 "root %llu has a root item with a more recent gen (%llu) compared to the found root node (%llu)\n",
7440 root_id, btrfs_root_generation(&ri), rii->gen);
7444 if (!read_only_mode) {
7445 btrfs_set_root_bytenr(&ri, rii->bytenr);
7446 btrfs_set_root_level(&ri, rii->level);
7447 btrfs_set_root_generation(&ri, rii->gen);
7448 write_extent_buffer(path->nodes[0], &ri,
7449 offset, sizeof(ri));
7459 * A regression introduced in the 3.17 kernel (more specifically in 3.17-rc2),
7460 * caused read-only snapshots to be corrupted if they were created at a moment
7461 * when the source subvolume/snapshot had orphan items. The issue was that the
7462 * on-disk root items became incorrect, referring to the pre orphan cleanup root
7463 * node instead of the post orphan cleanup root node.
7464 * So this function, and its callees, just detects and fixes those cases. Even
7465 * though the regression was for read-only snapshots, this function applies to
7466 * any snapshot/subvolume root.
7467 * This must be run before any other repair code - not doing it so, makes other
7468 * repair code delete or modify backrefs in the extent tree for example, which
7469 * will result in an inconsistent fs after repairing the root items.
7471 static int repair_root_items(struct btrfs_fs_info *info)
7473 struct btrfs_path *path = NULL;
7474 struct btrfs_key key;
7475 struct extent_buffer *leaf;
7476 struct btrfs_trans_handle *trans = NULL;
7481 ret = build_roots_info_cache(info);
7485 path = btrfs_alloc_path();
7491 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
7492 key.type = BTRFS_ROOT_ITEM_KEY;
7497 * Avoid opening and committing transactions if a leaf doesn't have
7498 * any root items that need to be fixed, so that we avoid rotating
7499 * backup roots unnecessarily.
7502 trans = btrfs_start_transaction(info->tree_root, 1);
7503 if (IS_ERR(trans)) {
7504 ret = PTR_ERR(trans);
7509 ret = btrfs_search_slot(trans, info->tree_root, &key, path,
7513 leaf = path->nodes[0];
7516 struct btrfs_key found_key;
7518 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7519 int no_more_keys = find_next_key(path, &key);
7521 btrfs_release_path(path);
7523 ret = btrfs_commit_transaction(trans,
7535 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7537 if (found_key.type != BTRFS_ROOT_ITEM_KEY)
7540 ret = maybe_repair_root_item(info, path, &found_key,
7545 if (!trans && repair) {
7548 btrfs_release_path(path);
7558 free_roots_info_cache();
7560 btrfs_free_path(path);
7567 static struct option long_options[] = {
7568 { "super", 1, NULL, 's' },
7569 { "repair", 0, NULL, 0 },
7570 { "init-csum-tree", 0, NULL, 0 },
7571 { "init-extent-tree", 0, NULL, 0 },
7572 { "check-data-csum", 0, NULL, 0 },
7573 { "backup", 0, NULL, 0 },
7574 { "subvol-extents", 1, NULL, 'E' },
7575 { "qgroup-report", 0, NULL, 'Q' },
7579 const char * const cmd_check_usage[] = {
7580 "btrfs check [options] <device>",
7581 "Check an unmounted btrfs filesystem.",
7583 "-s|--super <superblock> use this superblock copy",
7584 "-b|--backup use the backup root copy",
7585 "--repair try to repair the filesystem",
7586 "--init-csum-tree create a new CRC tree",
7587 "--init-extent-tree create a new extent tree",
7588 "--check-data-csum verify checkums of data blocks",
7589 "--qgroup-report print a report on qgroup consistency",
7590 "--subvol-extents <subvolid> print subvolume extents and sharing state",
7594 int cmd_check(int argc, char **argv)
7596 struct cache_tree root_cache;
7597 struct btrfs_root *root;
7598 struct btrfs_fs_info *info;
7601 char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
7604 int option_index = 0;
7605 int init_csum_tree = 0;
7606 int qgroup_report = 0;
7607 enum btrfs_open_ctree_flags ctree_flags = OPEN_CTREE_EXCLUSIVE;
7611 c = getopt_long(argc, argv, "as:b", long_options,
7616 case 'a': /* ignored */ break;
7618 ctree_flags |= OPEN_CTREE_BACKUP_ROOT;
7621 num = arg_strtou64(optarg);
7622 if (num >= BTRFS_SUPER_MIRROR_MAX) {
7624 "ERROR: super mirror should be less than: %d\n",
7625 BTRFS_SUPER_MIRROR_MAX);
7628 bytenr = btrfs_sb_offset(((int)num));
7629 printf("using SB copy %llu, bytenr %llu\n", num,
7630 (unsigned long long)bytenr);
7636 subvolid = arg_strtou64(optarg);
7640 usage(cmd_check_usage);
7642 if (option_index == 1) {
7643 printf("enabling repair mode\n");
7645 ctree_flags |= OPEN_CTREE_WRITES;
7646 } else if (option_index == 2) {
7647 printf("Creating a new CRC tree\n");
7650 ctree_flags |= OPEN_CTREE_WRITES;
7651 } else if (option_index == 3) {
7652 init_extent_tree = 1;
7653 ctree_flags |= (OPEN_CTREE_WRITES |
7654 OPEN_CTREE_NO_BLOCK_GROUPS);
7656 } else if (option_index == 4) {
7657 check_data_csum = 1;
7660 argc = argc - optind;
7662 if (check_argc_exact(argc, 1))
7663 usage(cmd_check_usage);
7666 cache_tree_init(&root_cache);
7668 if((ret = check_mounted(argv[optind])) < 0) {
7669 fprintf(stderr, "Could not check mount status: %s\n", strerror(-ret));
7672 fprintf(stderr, "%s is currently mounted. Aborting.\n", argv[optind]);
7677 /* only allow partial opening under repair mode */
7679 ctree_flags |= OPEN_CTREE_PARTIAL;
7681 info = open_ctree_fs_info(argv[optind], bytenr, 0, ctree_flags);
7683 fprintf(stderr, "Couldn't open file system\n");
7688 root = info->fs_root;
7690 ret = repair_root_items(info);
7694 fprintf(stderr, "Fixed %d roots.\n", ret);
7696 } else if (ret > 0) {
7698 "Found %d roots with an outdated root item.\n",
7701 "Please run a filesystem check with the option --repair to fix them.\n");
7707 * repair mode will force us to commit transaction which
7708 * will make us fail to load log tree when mounting.
7710 if (repair && btrfs_super_log_root(info->super_copy)) {
7711 ret = ask_user("repair mode will force to clear out log tree, Are you sure?");
7716 ret = zero_log_tree(root);
7718 fprintf(stderr, "fail to zero log tree\n");
7723 uuid_unparse(info->super_copy->fsid, uuidbuf);
7724 if (qgroup_report) {
7725 printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
7727 ret = qgroup_verify_all(info);
7729 print_qgroup_report(1);
7733 printf("Print extent state for subvolume %llu on %s\nUUID: %s\n",
7734 subvolid, argv[optind], uuidbuf);
7735 ret = print_extent_state(info, subvolid);
7738 printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
7740 if (!extent_buffer_uptodate(info->tree_root->node) ||
7741 !extent_buffer_uptodate(info->dev_root->node) ||
7742 !extent_buffer_uptodate(info->chunk_root->node)) {
7743 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
7748 if (init_extent_tree || init_csum_tree) {
7749 struct btrfs_trans_handle *trans;
7751 trans = btrfs_start_transaction(info->extent_root, 0);
7752 if (IS_ERR(trans)) {
7753 fprintf(stderr, "Error starting transaction\n");
7754 ret = PTR_ERR(trans);
7758 if (init_extent_tree) {
7759 printf("Creating a new extent tree\n");
7760 ret = reinit_extent_tree(trans, info);
7765 if (init_csum_tree) {
7766 fprintf(stderr, "Reinit crc root\n");
7767 ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
7769 fprintf(stderr, "crc root initialization failed\n");
7774 ret = fill_csum_tree(trans, info->csum_root);
7776 fprintf(stderr, "crc refilling failed\n");
7781 * Ok now we commit and run the normal fsck, which will add
7782 * extent entries for all of the items it finds.
7784 ret = btrfs_commit_transaction(trans, info->extent_root);
7788 if (!extent_buffer_uptodate(info->extent_root->node)) {
7789 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
7793 if (!extent_buffer_uptodate(info->csum_root->node)) {
7794 fprintf(stderr, "Checksum root corrupted, rerun with --init-csum-tree option\n");
7799 fprintf(stderr, "checking extents\n");
7800 ret = check_chunks_and_extents(root);
7802 fprintf(stderr, "Errors found in extent allocation tree or chunk allocation\n");
7804 fprintf(stderr, "checking free space cache\n");
7805 ret = check_space_cache(root);
7810 * We used to have to have these hole extents in between our real
7811 * extents so if we don't have this flag set we need to make sure there
7812 * are no gaps in the file extents for inodes, otherwise we can just
7813 * ignore it when this happens.
7815 no_holes = btrfs_fs_incompat(root->fs_info,
7816 BTRFS_FEATURE_INCOMPAT_NO_HOLES);
7817 fprintf(stderr, "checking fs roots\n");
7818 ret = check_fs_roots(root, &root_cache);
7822 fprintf(stderr, "checking csums\n");
7823 ret = check_csums(root);
7827 fprintf(stderr, "checking root refs\n");
7828 ret = check_root_refs(root, &root_cache);
7832 while (repair && !list_empty(&root->fs_info->recow_ebs)) {
7833 struct extent_buffer *eb;
7835 eb = list_first_entry(&root->fs_info->recow_ebs,
7836 struct extent_buffer, recow);
7837 list_del_init(&eb->recow);
7838 ret = recow_extent_buffer(root, eb);
7843 while (!list_empty(&delete_items)) {
7844 struct bad_item *bad;
7846 bad = list_first_entry(&delete_items, struct bad_item, list);
7847 list_del_init(&bad->list);
7849 ret = delete_bad_item(root, bad);
7853 if (info->quota_enabled) {
7855 fprintf(stderr, "checking quota groups\n");
7856 err = qgroup_verify_all(info);
7861 if (!list_empty(&root->fs_info->recow_ebs)) {
7862 fprintf(stderr, "Transid errors in file system\n");
7866 print_qgroup_report(0);
7867 if (found_old_backref) { /*
7868 * there was a disk format change when mixed
7869 * backref was in testing tree. The old format
7870 * existed about one week.
7872 printf("\n * Found old mixed backref format. "
7873 "The old format is not supported! *"
7874 "\n * Please mount the FS in readonly mode, "
7875 "backup data and re-format the FS. *\n\n");
7878 printf("found %llu bytes used err is %d\n",
7879 (unsigned long long)bytes_used, ret);
7880 printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
7881 printf("total tree bytes: %llu\n",
7882 (unsigned long long)total_btree_bytes);
7883 printf("total fs tree bytes: %llu\n",
7884 (unsigned long long)total_fs_tree_bytes);
7885 printf("total extent tree bytes: %llu\n",
7886 (unsigned long long)total_extent_tree_bytes);
7887 printf("btree space waste bytes: %llu\n",
7888 (unsigned long long)btree_space_waste);
7889 printf("file data blocks allocated: %llu\n referenced %llu\n",
7890 (unsigned long long)data_bytes_allocated,
7891 (unsigned long long)data_bytes_referenced);
7892 printf("%s\n", BTRFS_BUILD_VERSION);
7894 free_root_recs_tree(&root_cache);