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,
1477 struct btrfs_key node_key;
1479 btrfs_node_key_to_cpu(path->nodes[*level],
1481 path->slots[*level]);
1482 btrfs_add_corrupt_extent_record(root->fs_info,
1484 path->nodes[*level]->start,
1485 root->leafsize, *level);
1491 ret = check_child_node(root, cur, path->slots[*level], next);
1497 if (btrfs_is_leaf(next))
1498 status = btrfs_check_leaf(root, NULL, next);
1500 status = btrfs_check_node(root, NULL, next);
1501 if (status != BTRFS_TREE_BLOCK_CLEAN) {
1502 free_extent_buffer(next);
1507 *level = *level - 1;
1508 free_extent_buffer(path->nodes[*level]);
1509 path->nodes[*level] = next;
1510 path->slots[*level] = 0;
1513 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
1517 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
1518 struct walk_control *wc, int *level)
1521 struct extent_buffer *leaf;
1523 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1524 leaf = path->nodes[i];
1525 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
1530 free_extent_buffer(path->nodes[*level]);
1531 path->nodes[*level] = NULL;
1532 BUG_ON(*level > wc->active_node);
1533 if (*level == wc->active_node)
1534 leave_shared_node(root, wc, *level);
1541 static int check_root_dir(struct inode_record *rec)
1543 struct inode_backref *backref;
1546 if (!rec->found_inode_item || rec->errors)
1548 if (rec->nlink != 1 || rec->found_link != 0)
1550 if (list_empty(&rec->backrefs))
1552 backref = list_entry(rec->backrefs.next, struct inode_backref, list);
1553 if (!backref->found_inode_ref)
1555 if (backref->index != 0 || backref->namelen != 2 ||
1556 memcmp(backref->name, "..", 2))
1558 if (backref->found_dir_index || backref->found_dir_item)
1565 static int repair_inode_isize(struct btrfs_trans_handle *trans,
1566 struct btrfs_root *root, struct btrfs_path *path,
1567 struct inode_record *rec)
1569 struct btrfs_inode_item *ei;
1570 struct btrfs_key key;
1573 key.objectid = rec->ino;
1574 key.type = BTRFS_INODE_ITEM_KEY;
1575 key.offset = (u64)-1;
1577 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1581 if (!path->slots[0]) {
1588 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1589 if (key.objectid != rec->ino) {
1594 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1595 struct btrfs_inode_item);
1596 btrfs_set_inode_size(path->nodes[0], ei, rec->found_size);
1597 btrfs_mark_buffer_dirty(path->nodes[0]);
1598 rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
1599 printf("reset isize for dir %Lu root %Lu\n", rec->ino,
1600 root->root_key.objectid);
1602 btrfs_release_path(path);
1606 static int repair_inode_orphan_item(struct btrfs_trans_handle *trans,
1607 struct btrfs_root *root,
1608 struct btrfs_path *path,
1609 struct inode_record *rec)
1613 ret = btrfs_add_orphan_item(trans, root, path, rec->ino);
1614 btrfs_release_path(path);
1616 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
1620 static int add_missing_dir_index(struct btrfs_root *root,
1621 struct cache_tree *inode_cache,
1622 struct inode_record *rec,
1623 struct inode_backref *backref)
1625 struct btrfs_path *path;
1626 struct btrfs_trans_handle *trans;
1627 struct btrfs_dir_item *dir_item;
1628 struct extent_buffer *leaf;
1629 struct btrfs_key key;
1630 struct btrfs_disk_key disk_key;
1631 struct inode_record *dir_rec;
1632 unsigned long name_ptr;
1633 u32 data_size = sizeof(*dir_item) + backref->namelen;
1636 path = btrfs_alloc_path();
1640 trans = btrfs_start_transaction(root, 1);
1641 if (IS_ERR(trans)) {
1642 btrfs_free_path(path);
1643 return PTR_ERR(trans);
1646 fprintf(stderr, "repairing missing dir index item for inode %llu\n",
1647 (unsigned long long)rec->ino);
1648 key.objectid = backref->dir;
1649 key.type = BTRFS_DIR_INDEX_KEY;
1650 key.offset = backref->index;
1652 ret = btrfs_insert_empty_item(trans, root, path, &key, data_size);
1655 leaf = path->nodes[0];
1656 dir_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
1658 disk_key.objectid = cpu_to_le64(rec->ino);
1659 disk_key.type = BTRFS_INODE_ITEM_KEY;
1660 disk_key.offset = 0;
1662 btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
1663 btrfs_set_dir_type(leaf, dir_item, imode_to_type(rec->imode));
1664 btrfs_set_dir_data_len(leaf, dir_item, 0);
1665 btrfs_set_dir_name_len(leaf, dir_item, backref->namelen);
1666 name_ptr = (unsigned long)(dir_item + 1);
1667 write_extent_buffer(leaf, backref->name, name_ptr, backref->namelen);
1668 btrfs_mark_buffer_dirty(leaf);
1669 btrfs_free_path(path);
1670 btrfs_commit_transaction(trans, root);
1672 backref->found_dir_index = 1;
1673 dir_rec = get_inode_rec(inode_cache, backref->dir, 0);
1676 dir_rec->found_size += backref->namelen;
1677 if (dir_rec->found_size == dir_rec->isize &&
1678 (dir_rec->errors & I_ERR_DIR_ISIZE_WRONG))
1679 dir_rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
1680 if (dir_rec->found_size != dir_rec->isize)
1681 dir_rec->errors |= I_ERR_DIR_ISIZE_WRONG;
1686 static int delete_dir_index(struct btrfs_root *root,
1687 struct cache_tree *inode_cache,
1688 struct inode_record *rec,
1689 struct inode_backref *backref)
1691 struct btrfs_trans_handle *trans;
1692 struct btrfs_dir_item *di;
1693 struct btrfs_path *path;
1696 path = btrfs_alloc_path();
1700 trans = btrfs_start_transaction(root, 1);
1701 if (IS_ERR(trans)) {
1702 btrfs_free_path(path);
1703 return PTR_ERR(trans);
1707 fprintf(stderr, "Deleting bad dir index [%llu,%u,%llu] root %llu\n",
1708 (unsigned long long)backref->dir,
1709 BTRFS_DIR_INDEX_KEY, (unsigned long long)backref->index,
1710 (unsigned long long)root->objectid);
1712 di = btrfs_lookup_dir_index(trans, root, path, backref->dir,
1713 backref->name, backref->namelen,
1714 backref->index, -1);
1717 btrfs_free_path(path);
1718 btrfs_commit_transaction(trans, root);
1725 ret = btrfs_del_item(trans, root, path);
1727 ret = btrfs_delete_one_dir_name(trans, root, path, di);
1729 btrfs_free_path(path);
1730 btrfs_commit_transaction(trans, root);
1734 static int create_inode_item(struct btrfs_root *root,
1735 struct inode_record *rec,
1736 struct inode_backref *backref, int root_dir)
1738 struct btrfs_trans_handle *trans;
1739 struct btrfs_inode_item inode_item;
1740 time_t now = time(NULL);
1743 trans = btrfs_start_transaction(root, 1);
1744 if (IS_ERR(trans)) {
1745 ret = PTR_ERR(trans);
1749 fprintf(stderr, "root %llu inode %llu recreating inode item, this may "
1750 "be incomplete, please check permissions and content after "
1751 "the fsck completes.\n", (unsigned long long)root->objectid,
1752 (unsigned long long)rec->ino);
1754 memset(&inode_item, 0, sizeof(inode_item));
1755 btrfs_set_stack_inode_generation(&inode_item, trans->transid);
1757 btrfs_set_stack_inode_nlink(&inode_item, 1);
1759 btrfs_set_stack_inode_nlink(&inode_item, rec->found_link);
1760 btrfs_set_stack_inode_nbytes(&inode_item, rec->found_size);
1761 if (rec->found_dir_item) {
1762 if (rec->found_file_extent)
1763 fprintf(stderr, "root %llu inode %llu has both a dir "
1764 "item and extents, unsure if it is a dir or a "
1765 "regular file so setting it as a directory\n",
1766 (unsigned long long)root->objectid,
1767 (unsigned long long)rec->ino);
1768 btrfs_set_stack_inode_mode(&inode_item, S_IFDIR | 0755);
1769 btrfs_set_stack_inode_size(&inode_item, rec->found_size);
1770 } else if (!rec->found_dir_item) {
1771 btrfs_set_stack_inode_size(&inode_item, rec->extent_end);
1772 btrfs_set_stack_inode_mode(&inode_item, S_IFREG | 0755);
1774 btrfs_set_stack_timespec_sec(&inode_item.atime, now);
1775 btrfs_set_stack_timespec_nsec(&inode_item.atime, 0);
1776 btrfs_set_stack_timespec_sec(&inode_item.ctime, now);
1777 btrfs_set_stack_timespec_nsec(&inode_item.ctime, 0);
1778 btrfs_set_stack_timespec_sec(&inode_item.mtime, now);
1779 btrfs_set_stack_timespec_nsec(&inode_item.mtime, 0);
1780 btrfs_set_stack_timespec_sec(&inode_item.otime, 0);
1781 btrfs_set_stack_timespec_nsec(&inode_item.otime, 0);
1783 ret = btrfs_insert_inode(trans, root, rec->ino, &inode_item);
1785 btrfs_commit_transaction(trans, root);
1789 static int repair_inode_backrefs(struct btrfs_root *root,
1790 struct inode_record *rec,
1791 struct cache_tree *inode_cache,
1794 struct inode_backref *tmp, *backref;
1795 u64 root_dirid = btrfs_root_dirid(&root->root_item);
1799 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
1800 if (!delete && rec->ino == root_dirid) {
1801 if (!rec->found_inode_item) {
1802 ret = create_inode_item(root, rec, backref, 1);
1809 /* Index 0 for root dir's are special, don't mess with it */
1810 if (rec->ino == root_dirid && backref->index == 0)
1814 ((backref->found_dir_index && !backref->found_inode_ref) ||
1815 (backref->found_dir_index && backref->found_inode_ref &&
1816 (backref->errors & REF_ERR_INDEX_UNMATCH)))) {
1817 ret = delete_dir_index(root, inode_cache, rec, backref);
1821 list_del(&backref->list);
1825 if (!delete && !backref->found_dir_index &&
1826 backref->found_dir_item && backref->found_inode_ref) {
1827 ret = add_missing_dir_index(root, inode_cache, rec,
1832 if (backref->found_dir_item &&
1833 backref->found_dir_index &&
1834 backref->found_dir_index) {
1835 if (!backref->errors &&
1836 backref->found_inode_ref) {
1837 list_del(&backref->list);
1843 if (!delete && (!backref->found_dir_index &&
1844 !backref->found_dir_item &&
1845 backref->found_inode_ref)) {
1846 struct btrfs_trans_handle *trans;
1847 struct btrfs_key location;
1849 ret = check_dir_conflict(root, backref->name,
1855 * let nlink fixing routine to handle it,
1856 * which can do it better.
1861 location.objectid = rec->ino;
1862 location.type = BTRFS_INODE_ITEM_KEY;
1863 location.offset = 0;
1865 trans = btrfs_start_transaction(root, 1);
1866 if (IS_ERR(trans)) {
1867 ret = PTR_ERR(trans);
1870 fprintf(stderr, "adding missing dir index/item pair "
1872 (unsigned long long)rec->ino);
1873 ret = btrfs_insert_dir_item(trans, root, backref->name,
1875 backref->dir, &location,
1876 imode_to_type(rec->imode),
1879 btrfs_commit_transaction(trans, root);
1883 if (!delete && (backref->found_inode_ref &&
1884 backref->found_dir_index &&
1885 backref->found_dir_item &&
1886 !(backref->errors & REF_ERR_INDEX_UNMATCH) &&
1887 !rec->found_inode_item)) {
1888 ret = create_inode_item(root, rec, backref, 0);
1895 return ret ? ret : repaired;
1899 * To determine the file type for nlink/inode_item repair
1901 * Return 0 if file type is found and BTRFS_FT_* is stored into type.
1902 * Return -ENOENT if file type is not found.
1904 static int find_file_type(struct inode_record *rec, u8 *type)
1906 struct inode_backref *backref;
1908 list_for_each_entry(backref, &rec->backrefs, list) {
1909 if (backref->found_dir_index || backref->found_dir_item) {
1910 *type = backref->filetype;
1918 * To determine the file name for nlink repair
1920 * Return 0 if file name is found, set name and namelen.
1921 * Return -ENOENT if file name is not found.
1923 static int find_file_name(struct inode_record *rec,
1924 char *name, int *namelen)
1926 struct inode_backref *backref;
1928 list_for_each_entry(backref, &rec->backrefs, list) {
1929 if (backref->found_dir_index || backref->found_dir_item ||
1930 backref->found_inode_ref) {
1931 memcpy(name, backref->name, backref->namelen);
1932 *namelen = backref->namelen;
1939 /* Reset the nlink of the inode to the correct one */
1940 static int reset_nlink(struct btrfs_trans_handle *trans,
1941 struct btrfs_root *root,
1942 struct btrfs_path *path,
1943 struct inode_record *rec)
1945 struct inode_backref *backref;
1946 struct inode_backref *tmp;
1947 struct btrfs_key key;
1948 struct btrfs_inode_item *inode_item;
1951 /* Remove all backref including the valid ones */
1952 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
1953 ret = btrfs_unlink(trans, root, rec->ino, backref->dir,
1954 backref->index, backref->name,
1955 backref->namelen, 0);
1959 /* remove invalid backref, so it won't be added back */
1960 if (!(backref->found_dir_index &&
1961 backref->found_dir_item &&
1962 backref->found_inode_ref)) {
1963 list_del(&backref->list);
1968 /* Set nlink to 0 */
1969 key.objectid = rec->ino;
1970 key.type = BTRFS_INODE_ITEM_KEY;
1972 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1979 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1980 struct btrfs_inode_item);
1981 btrfs_set_inode_nlink(path->nodes[0], inode_item, 0);
1982 btrfs_mark_buffer_dirty(path->nodes[0]);
1983 btrfs_release_path(path);
1986 * Add back valid inode_ref/dir_item/dir_index,
1987 * add_link() will handle the nlink inc, so new nlink must be correct
1989 list_for_each_entry(backref, &rec->backrefs, list) {
1990 ret = btrfs_add_link(trans, root, rec->ino, backref->dir,
1991 backref->name, backref->namelen,
1992 backref->ref_type, &backref->index, 1);
1997 btrfs_release_path(path);
2001 static int repair_inode_nlinks(struct btrfs_trans_handle *trans,
2002 struct btrfs_root *root,
2003 struct btrfs_path *path,
2004 struct inode_record *rec)
2006 char *dir_name = "lost+found";
2007 char namebuf[BTRFS_NAME_LEN] = {0};
2012 int name_recovered = 0;
2013 int type_recovered = 0;
2017 * Get file name and type first before these invalid inode ref
2018 * are deleted by remove_all_invalid_backref()
2020 name_recovered = !find_file_name(rec, namebuf, &namelen);
2021 type_recovered = !find_file_type(rec, &type);
2023 if (!name_recovered) {
2024 printf("Can't get file name for inode %llu, using '%llu' as fallback\n",
2025 rec->ino, rec->ino);
2026 namelen = count_digits(rec->ino);
2027 sprintf(namebuf, "%llu", rec->ino);
2030 if (!type_recovered) {
2031 printf("Can't get file type for inode %llu, using FILE as fallback\n",
2033 type = BTRFS_FT_REG_FILE;
2037 ret = reset_nlink(trans, root, path, rec);
2040 "Failed to reset nlink for inode %llu: %s\n",
2041 rec->ino, strerror(-ret));
2045 if (rec->found_link == 0) {
2046 lost_found_ino = root->highest_inode;
2047 if (lost_found_ino >= BTRFS_LAST_FREE_OBJECTID) {
2052 ret = btrfs_mkdir(trans, root, dir_name, strlen(dir_name),
2053 BTRFS_FIRST_FREE_OBJECTID, &lost_found_ino,
2056 fprintf(stderr, "Failed to create '%s' dir: %s",
2057 dir_name, strerror(-ret));
2060 ret = btrfs_add_link(trans, root, rec->ino, lost_found_ino,
2061 namebuf, namelen, type, NULL, 1);
2062 if (ret == -EEXIST) {
2064 * Conflicting file name, add ".INO" as suffix * +1 for '.'
2066 if (namelen + count_digits(rec->ino) + 1 >
2071 snprintf(namebuf + namelen, BTRFS_NAME_LEN - namelen,
2073 namelen += count_digits(rec->ino) + 1;
2074 ret = btrfs_add_link(trans, root, rec->ino,
2075 lost_found_ino, namebuf,
2076 namelen, type, NULL, 1);
2080 "Failed to link the inode %llu to %s dir: %s",
2081 rec->ino, dir_name, strerror(-ret));
2085 * Just increase the found_link, don't actually add the
2086 * backref. This will make things easier and this inode
2087 * record will be freed after the repair is done.
2088 * So fsck will not report problem about this inode.
2091 printf("Moving file '%.*s' to '%s' dir since it has no valid backref\n",
2092 namelen, namebuf, dir_name);
2094 rec->errors &= ~I_ERR_LINK_COUNT_WRONG;
2095 printf("Fixed the nlink of inode %llu\n", rec->ino);
2097 btrfs_release_path(path);
2101 static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
2103 struct btrfs_trans_handle *trans;
2104 struct btrfs_path *path;
2107 if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG |
2108 I_ERR_NO_ORPHAN_ITEM |
2109 I_ERR_LINK_COUNT_WRONG)))
2112 path = btrfs_alloc_path();
2117 * For nlink repair, it may create a dir and add link, so
2118 * 2 for parent(256)'s dir_index and dir_item
2119 * 2 for lost+found dir's inode_item and inode_ref
2120 * 1 for the new inode_ref of the file
2121 * 2 for lost+found dir's dir_index and dir_item for the file
2123 trans = btrfs_start_transaction(root, 7);
2124 if (IS_ERR(trans)) {
2125 btrfs_free_path(path);
2126 return PTR_ERR(trans);
2129 if (rec->errors & I_ERR_DIR_ISIZE_WRONG)
2130 ret = repair_inode_isize(trans, root, path, rec);
2131 if (!ret && rec->errors & I_ERR_NO_ORPHAN_ITEM)
2132 ret = repair_inode_orphan_item(trans, root, path, rec);
2133 if (!ret && rec->errors & I_ERR_LINK_COUNT_WRONG)
2134 ret = repair_inode_nlinks(trans, root, path, rec);
2135 btrfs_commit_transaction(trans, root);
2136 btrfs_free_path(path);
2140 static int check_inode_recs(struct btrfs_root *root,
2141 struct cache_tree *inode_cache)
2143 struct cache_extent *cache;
2144 struct ptr_node *node;
2145 struct inode_record *rec;
2146 struct inode_backref *backref;
2151 u64 root_dirid = btrfs_root_dirid(&root->root_item);
2153 if (btrfs_root_refs(&root->root_item) == 0) {
2154 if (!cache_tree_empty(inode_cache))
2155 fprintf(stderr, "warning line %d\n", __LINE__);
2160 * We need to record the highest inode number for later 'lost+found'
2162 * We must select a ino not used/refered by any existing inode, or
2163 * 'lost+found' ino may be a missing ino in a corrupted leaf,
2164 * this may cause 'lost+found' dir has wrong nlinks.
2166 cache = last_cache_extent(inode_cache);
2168 node = container_of(cache, struct ptr_node, cache);
2170 if (rec->ino > root->highest_inode)
2171 root->highest_inode = rec->ino;
2175 * We need to repair backrefs first because we could change some of the
2176 * errors in the inode recs.
2178 * We also need to go through and delete invalid backrefs first and then
2179 * add the correct ones second. We do this because we may get EEXIST
2180 * when adding back the correct index because we hadn't yet deleted the
2183 * For example, if we were missing a dir index then the directories
2184 * isize would be wrong, so if we fixed the isize to what we thought it
2185 * would be and then fixed the backref we'd still have a invalid fs, so
2186 * we need to add back the dir index and then check to see if the isize
2191 if (stage == 3 && !err)
2194 cache = search_cache_extent(inode_cache, 0);
2195 while (repair && cache) {
2196 node = container_of(cache, struct ptr_node, cache);
2198 cache = next_cache_extent(cache);
2200 /* Need to free everything up and rescan */
2202 remove_cache_extent(inode_cache, &node->cache);
2204 free_inode_rec(rec);
2208 if (list_empty(&rec->backrefs))
2211 ret = repair_inode_backrefs(root, rec, inode_cache,
2225 rec = get_inode_rec(inode_cache, root_dirid, 0);
2227 ret = check_root_dir(rec);
2229 fprintf(stderr, "root %llu root dir %llu error\n",
2230 (unsigned long long)root->root_key.objectid,
2231 (unsigned long long)root_dirid);
2232 print_inode_error(root, rec);
2237 struct btrfs_trans_handle *trans;
2239 trans = btrfs_start_transaction(root, 1);
2240 if (IS_ERR(trans)) {
2241 err = PTR_ERR(trans);
2246 "root %llu missing its root dir, recreating\n",
2247 (unsigned long long)root->objectid);
2249 ret = btrfs_make_root_dir(trans, root, root_dirid);
2252 btrfs_commit_transaction(trans, root);
2256 fprintf(stderr, "root %llu root dir %llu not found\n",
2257 (unsigned long long)root->root_key.objectid,
2258 (unsigned long long)root_dirid);
2262 cache = search_cache_extent(inode_cache, 0);
2265 node = container_of(cache, struct ptr_node, cache);
2267 remove_cache_extent(inode_cache, &node->cache);
2269 if (rec->ino == root_dirid ||
2270 rec->ino == BTRFS_ORPHAN_OBJECTID) {
2271 free_inode_rec(rec);
2275 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
2276 ret = check_orphan_item(root, rec->ino);
2278 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
2279 if (can_free_inode_rec(rec)) {
2280 free_inode_rec(rec);
2285 if (rec->found_link != rec->nlink)
2286 rec->errors |= I_ERR_LINK_COUNT_WRONG;
2288 ret = try_repair_inode(root, rec);
2289 if (ret == 0 && can_free_inode_rec(rec)) {
2290 free_inode_rec(rec);
2297 if (!rec->found_inode_item)
2298 rec->errors |= I_ERR_NO_INODE_ITEM;
2299 print_inode_error(root, rec);
2300 list_for_each_entry(backref, &rec->backrefs, list) {
2301 if (!backref->found_dir_item)
2302 backref->errors |= REF_ERR_NO_DIR_ITEM;
2303 if (!backref->found_dir_index)
2304 backref->errors |= REF_ERR_NO_DIR_INDEX;
2305 if (!backref->found_inode_ref)
2306 backref->errors |= REF_ERR_NO_INODE_REF;
2307 fprintf(stderr, "\tunresolved ref dir %llu index %llu"
2308 " namelen %u name %s filetype %d errors %x",
2309 (unsigned long long)backref->dir,
2310 (unsigned long long)backref->index,
2311 backref->namelen, backref->name,
2312 backref->filetype, backref->errors);
2313 print_ref_error(backref->errors);
2315 free_inode_rec(rec);
2317 return (error > 0) ? -1 : 0;
2320 static struct root_record *get_root_rec(struct cache_tree *root_cache,
2323 struct cache_extent *cache;
2324 struct root_record *rec = NULL;
2327 cache = lookup_cache_extent(root_cache, objectid, 1);
2329 rec = container_of(cache, struct root_record, cache);
2331 rec = calloc(1, sizeof(*rec));
2332 rec->objectid = objectid;
2333 INIT_LIST_HEAD(&rec->backrefs);
2334 rec->cache.start = objectid;
2335 rec->cache.size = 1;
2337 ret = insert_cache_extent(root_cache, &rec->cache);
2343 static struct root_backref *get_root_backref(struct root_record *rec,
2344 u64 ref_root, u64 dir, u64 index,
2345 const char *name, int namelen)
2347 struct root_backref *backref;
2349 list_for_each_entry(backref, &rec->backrefs, list) {
2350 if (backref->ref_root != ref_root || backref->dir != dir ||
2351 backref->namelen != namelen)
2353 if (memcmp(name, backref->name, namelen))
2358 backref = malloc(sizeof(*backref) + namelen + 1);
2359 memset(backref, 0, sizeof(*backref));
2360 backref->ref_root = ref_root;
2362 backref->index = index;
2363 backref->namelen = namelen;
2364 memcpy(backref->name, name, namelen);
2365 backref->name[namelen] = '\0';
2366 list_add_tail(&backref->list, &rec->backrefs);
2370 static void free_root_record(struct cache_extent *cache)
2372 struct root_record *rec;
2373 struct root_backref *backref;
2375 rec = container_of(cache, struct root_record, cache);
2376 while (!list_empty(&rec->backrefs)) {
2377 backref = list_entry(rec->backrefs.next,
2378 struct root_backref, list);
2379 list_del(&backref->list);
2386 FREE_EXTENT_CACHE_BASED_TREE(root_recs, free_root_record);
2388 static int add_root_backref(struct cache_tree *root_cache,
2389 u64 root_id, u64 ref_root, u64 dir, u64 index,
2390 const char *name, int namelen,
2391 int item_type, int errors)
2393 struct root_record *rec;
2394 struct root_backref *backref;
2396 rec = get_root_rec(root_cache, root_id);
2397 backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
2399 backref->errors |= errors;
2401 if (item_type != BTRFS_DIR_ITEM_KEY) {
2402 if (backref->found_dir_index || backref->found_back_ref ||
2403 backref->found_forward_ref) {
2404 if (backref->index != index)
2405 backref->errors |= REF_ERR_INDEX_UNMATCH;
2407 backref->index = index;
2411 if (item_type == BTRFS_DIR_ITEM_KEY) {
2412 if (backref->found_forward_ref)
2414 backref->found_dir_item = 1;
2415 } else if (item_type == BTRFS_DIR_INDEX_KEY) {
2416 backref->found_dir_index = 1;
2417 } else if (item_type == BTRFS_ROOT_REF_KEY) {
2418 if (backref->found_forward_ref)
2419 backref->errors |= REF_ERR_DUP_ROOT_REF;
2420 else if (backref->found_dir_item)
2422 backref->found_forward_ref = 1;
2423 } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
2424 if (backref->found_back_ref)
2425 backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
2426 backref->found_back_ref = 1;
2431 if (backref->found_forward_ref && backref->found_dir_item)
2432 backref->reachable = 1;
2436 static int merge_root_recs(struct btrfs_root *root,
2437 struct cache_tree *src_cache,
2438 struct cache_tree *dst_cache)
2440 struct cache_extent *cache;
2441 struct ptr_node *node;
2442 struct inode_record *rec;
2443 struct inode_backref *backref;
2446 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2447 free_inode_recs_tree(src_cache);
2452 cache = search_cache_extent(src_cache, 0);
2455 node = container_of(cache, struct ptr_node, cache);
2457 remove_cache_extent(src_cache, &node->cache);
2460 ret = is_child_root(root, root->objectid, rec->ino);
2466 list_for_each_entry(backref, &rec->backrefs, list) {
2467 BUG_ON(backref->found_inode_ref);
2468 if (backref->found_dir_item)
2469 add_root_backref(dst_cache, rec->ino,
2470 root->root_key.objectid, backref->dir,
2471 backref->index, backref->name,
2472 backref->namelen, BTRFS_DIR_ITEM_KEY,
2474 if (backref->found_dir_index)
2475 add_root_backref(dst_cache, rec->ino,
2476 root->root_key.objectid, backref->dir,
2477 backref->index, backref->name,
2478 backref->namelen, BTRFS_DIR_INDEX_KEY,
2482 free_inode_rec(rec);
2489 static int check_root_refs(struct btrfs_root *root,
2490 struct cache_tree *root_cache)
2492 struct root_record *rec;
2493 struct root_record *ref_root;
2494 struct root_backref *backref;
2495 struct cache_extent *cache;
2501 rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
2504 /* fixme: this can not detect circular references */
2507 cache = search_cache_extent(root_cache, 0);
2511 rec = container_of(cache, struct root_record, cache);
2512 cache = next_cache_extent(cache);
2514 if (rec->found_ref == 0)
2517 list_for_each_entry(backref, &rec->backrefs, list) {
2518 if (!backref->reachable)
2521 ref_root = get_root_rec(root_cache,
2523 if (ref_root->found_ref > 0)
2526 backref->reachable = 0;
2528 if (rec->found_ref == 0)
2534 cache = search_cache_extent(root_cache, 0);
2538 rec = container_of(cache, struct root_record, cache);
2539 cache = next_cache_extent(cache);
2541 if (rec->found_ref == 0 &&
2542 rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
2543 rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
2544 ret = check_orphan_item(root->fs_info->tree_root,
2550 * If we don't have a root item then we likely just have
2551 * a dir item in a snapshot for this root but no actual
2552 * ref key or anything so it's meaningless.
2554 if (!rec->found_root_item)
2557 fprintf(stderr, "fs tree %llu not referenced\n",
2558 (unsigned long long)rec->objectid);
2562 if (rec->found_ref > 0 && !rec->found_root_item)
2564 list_for_each_entry(backref, &rec->backrefs, list) {
2565 if (!backref->found_dir_item)
2566 backref->errors |= REF_ERR_NO_DIR_ITEM;
2567 if (!backref->found_dir_index)
2568 backref->errors |= REF_ERR_NO_DIR_INDEX;
2569 if (!backref->found_back_ref)
2570 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
2571 if (!backref->found_forward_ref)
2572 backref->errors |= REF_ERR_NO_ROOT_REF;
2573 if (backref->reachable && backref->errors)
2580 fprintf(stderr, "fs tree %llu refs %u %s\n",
2581 (unsigned long long)rec->objectid, rec->found_ref,
2582 rec->found_root_item ? "" : "not found");
2584 list_for_each_entry(backref, &rec->backrefs, list) {
2585 if (!backref->reachable)
2587 if (!backref->errors && rec->found_root_item)
2589 fprintf(stderr, "\tunresolved ref root %llu dir %llu"
2590 " index %llu namelen %u name %s errors %x\n",
2591 (unsigned long long)backref->ref_root,
2592 (unsigned long long)backref->dir,
2593 (unsigned long long)backref->index,
2594 backref->namelen, backref->name,
2596 print_ref_error(backref->errors);
2599 return errors > 0 ? 1 : 0;
2602 static int process_root_ref(struct extent_buffer *eb, int slot,
2603 struct btrfs_key *key,
2604 struct cache_tree *root_cache)
2610 struct btrfs_root_ref *ref;
2611 char namebuf[BTRFS_NAME_LEN];
2614 ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
2616 dirid = btrfs_root_ref_dirid(eb, ref);
2617 index = btrfs_root_ref_sequence(eb, ref);
2618 name_len = btrfs_root_ref_name_len(eb, ref);
2620 if (name_len <= BTRFS_NAME_LEN) {
2624 len = BTRFS_NAME_LEN;
2625 error = REF_ERR_NAME_TOO_LONG;
2627 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
2629 if (key->type == BTRFS_ROOT_REF_KEY) {
2630 add_root_backref(root_cache, key->offset, key->objectid, dirid,
2631 index, namebuf, len, key->type, error);
2633 add_root_backref(root_cache, key->objectid, key->offset, dirid,
2634 index, namebuf, len, key->type, error);
2639 static void free_corrupt_block(struct cache_extent *cache)
2641 struct btrfs_corrupt_block *corrupt;
2643 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
2647 FREE_EXTENT_CACHE_BASED_TREE(corrupt_blocks, free_corrupt_block);
2650 * Repair the btree of the given root.
2652 * The fix is to remove the node key in corrupt_blocks cache_tree.
2653 * and rebalance the tree.
2654 * After the fix, the btree should be writeable.
2656 static int repair_btree(struct btrfs_root *root,
2657 struct cache_tree *corrupt_blocks)
2659 struct btrfs_trans_handle *trans;
2660 struct btrfs_path *path;
2661 struct btrfs_corrupt_block *corrupt;
2662 struct cache_extent *cache;
2663 struct btrfs_key key;
2668 if (cache_tree_empty(corrupt_blocks))
2671 path = btrfs_alloc_path();
2675 trans = btrfs_start_transaction(root, 1);
2676 if (IS_ERR(trans)) {
2677 ret = PTR_ERR(trans);
2678 fprintf(stderr, "Error starting transaction: %s\n",
2682 cache = first_cache_extent(corrupt_blocks);
2684 corrupt = container_of(cache, struct btrfs_corrupt_block,
2686 level = corrupt->level;
2687 path->lowest_level = level;
2688 key.objectid = corrupt->key.objectid;
2689 key.type = corrupt->key.type;
2690 key.offset = corrupt->key.offset;
2693 * Here we don't want to do any tree balance, since it may
2694 * cause a balance with corrupted brother leaf/node,
2695 * so ins_len set to 0 here.
2696 * Balance will be done after all corrupt node/leaf is deleted.
2698 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2701 offset = btrfs_node_blockptr(path->nodes[level],
2702 path->slots[level]);
2704 /* Remove the ptr */
2705 ret = btrfs_del_ptr(trans, root, path, level,
2706 path->slots[level]);
2710 * Remove the corresponding extent
2711 * return value is not concerned.
2713 btrfs_release_path(path);
2714 ret = btrfs_free_extent(trans, root, offset, root->nodesize,
2715 0, root->root_key.objectid,
2717 cache = next_cache_extent(cache);
2720 /* Balance the btree using btrfs_search_slot() */
2721 cache = first_cache_extent(corrupt_blocks);
2723 corrupt = container_of(cache, struct btrfs_corrupt_block,
2725 memcpy(&key, &corrupt->key, sizeof(key));
2726 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2729 /* return will always >0 since it won't find the item */
2731 btrfs_release_path(path);
2732 cache = next_cache_extent(cache);
2735 btrfs_free_path(path);
2736 btrfs_commit_transaction(trans, root);
2740 static int check_fs_root(struct btrfs_root *root,
2741 struct cache_tree *root_cache,
2742 struct walk_control *wc)
2748 struct btrfs_path path;
2749 struct shared_node root_node;
2750 struct root_record *rec;
2751 struct btrfs_root_item *root_item = &root->root_item;
2752 struct cache_tree corrupt_blocks;
2753 enum btrfs_tree_block_status status;
2756 * Reuse the corrupt_block cache tree to record corrupted tree block
2758 * Unlike the usage in extent tree check, here we do it in a per
2759 * fs/subvol tree base.
2761 cache_tree_init(&corrupt_blocks);
2762 root->fs_info->corrupt_blocks = &corrupt_blocks;
2763 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2764 rec = get_root_rec(root_cache, root->root_key.objectid);
2765 if (btrfs_root_refs(root_item) > 0)
2766 rec->found_root_item = 1;
2769 btrfs_init_path(&path);
2770 memset(&root_node, 0, sizeof(root_node));
2771 cache_tree_init(&root_node.root_cache);
2772 cache_tree_init(&root_node.inode_cache);
2774 level = btrfs_header_level(root->node);
2775 memset(wc->nodes, 0, sizeof(wc->nodes));
2776 wc->nodes[level] = &root_node;
2777 wc->active_node = level;
2778 wc->root_level = level;
2780 /* We may not have checked the root block, lets do that now */
2781 if (btrfs_is_leaf(root->node))
2782 status = btrfs_check_leaf(root, NULL, root->node);
2784 status = btrfs_check_node(root, NULL, root->node);
2785 if (status != BTRFS_TREE_BLOCK_CLEAN)
2788 if (btrfs_root_refs(root_item) > 0 ||
2789 btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2790 path.nodes[level] = root->node;
2791 extent_buffer_get(root->node);
2792 path.slots[level] = 0;
2794 struct btrfs_key key;
2795 struct btrfs_disk_key found_key;
2797 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2798 level = root_item->drop_level;
2799 path.lowest_level = level;
2800 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
2803 btrfs_node_key(path.nodes[level], &found_key,
2805 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2806 sizeof(found_key)));
2810 wret = walk_down_tree(root, &path, wc, &level);
2816 wret = walk_up_tree(root, &path, wc, &level);
2823 btrfs_release_path(&path);
2825 if (!cache_tree_empty(&corrupt_blocks)) {
2826 struct cache_extent *cache;
2827 struct btrfs_corrupt_block *corrupt;
2829 printf("The following tree block(s) is corrupted in tree %llu:\n",
2830 root->root_key.objectid);
2831 cache = first_cache_extent(&corrupt_blocks);
2833 corrupt = container_of(cache,
2834 struct btrfs_corrupt_block,
2836 printf("\ttree block bytenr: %llu, level: %d, node key: (%llu, %u, %llu)\n",
2837 cache->start, corrupt->level,
2838 corrupt->key.objectid, corrupt->key.type,
2839 corrupt->key.offset);
2840 cache = next_cache_extent(cache);
2843 printf("Try to repair the btree for root %llu\n",
2844 root->root_key.objectid);
2845 ret = repair_btree(root, &corrupt_blocks);
2847 fprintf(stderr, "Failed to repair btree: %s\n",
2850 printf("Btree for root %llu is fixed\n",
2851 root->root_key.objectid);
2855 err = merge_root_recs(root, &root_node.root_cache, root_cache);
2859 if (root_node.current) {
2860 root_node.current->checked = 1;
2861 maybe_free_inode_rec(&root_node.inode_cache,
2865 err = check_inode_recs(root, &root_node.inode_cache);
2869 free_corrupt_blocks_tree(&corrupt_blocks);
2870 root->fs_info->corrupt_blocks = NULL;
2874 static int fs_root_objectid(u64 objectid)
2876 if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
2877 objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2879 return is_fstree(objectid);
2882 static int check_fs_roots(struct btrfs_root *root,
2883 struct cache_tree *root_cache)
2885 struct btrfs_path path;
2886 struct btrfs_key key;
2887 struct walk_control wc;
2888 struct extent_buffer *leaf, *tree_node;
2889 struct btrfs_root *tmp_root;
2890 struct btrfs_root *tree_root = root->fs_info->tree_root;
2895 * Just in case we made any changes to the extent tree that weren't
2896 * reflected into the free space cache yet.
2899 reset_cached_block_groups(root->fs_info);
2900 memset(&wc, 0, sizeof(wc));
2901 cache_tree_init(&wc.shared);
2902 btrfs_init_path(&path);
2907 key.type = BTRFS_ROOT_ITEM_KEY;
2908 ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
2913 tree_node = tree_root->node;
2915 if (tree_node != tree_root->node) {
2916 free_root_recs_tree(root_cache);
2917 btrfs_release_path(&path);
2920 leaf = path.nodes[0];
2921 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
2922 ret = btrfs_next_leaf(tree_root, &path);
2928 leaf = path.nodes[0];
2930 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
2931 if (key.type == BTRFS_ROOT_ITEM_KEY &&
2932 fs_root_objectid(key.objectid)) {
2933 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2934 tmp_root = btrfs_read_fs_root_no_cache(
2935 root->fs_info, &key);
2937 key.offset = (u64)-1;
2938 tmp_root = btrfs_read_fs_root(
2939 root->fs_info, &key);
2941 if (IS_ERR(tmp_root)) {
2945 ret = check_fs_root(tmp_root, root_cache, &wc);
2946 if (ret == -EAGAIN) {
2947 free_root_recs_tree(root_cache);
2948 btrfs_release_path(&path);
2953 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
2954 btrfs_free_fs_root(tmp_root);
2955 } else if (key.type == BTRFS_ROOT_REF_KEY ||
2956 key.type == BTRFS_ROOT_BACKREF_KEY) {
2957 process_root_ref(leaf, path.slots[0], &key,
2964 btrfs_release_path(&path);
2966 free_extent_cache_tree(&wc.shared);
2967 if (!cache_tree_empty(&wc.shared))
2968 fprintf(stderr, "warning line %d\n", __LINE__);
2973 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
2975 struct list_head *cur = rec->backrefs.next;
2976 struct extent_backref *back;
2977 struct tree_backref *tback;
2978 struct data_backref *dback;
2982 while(cur != &rec->backrefs) {
2983 back = list_entry(cur, struct extent_backref, list);
2985 if (!back->found_extent_tree) {
2989 if (back->is_data) {
2990 dback = (struct data_backref *)back;
2991 fprintf(stderr, "Backref %llu %s %llu"
2992 " owner %llu offset %llu num_refs %lu"
2993 " not found in extent tree\n",
2994 (unsigned long long)rec->start,
2995 back->full_backref ?
2997 back->full_backref ?
2998 (unsigned long long)dback->parent:
2999 (unsigned long long)dback->root,
3000 (unsigned long long)dback->owner,
3001 (unsigned long long)dback->offset,
3002 (unsigned long)dback->num_refs);
3004 tback = (struct tree_backref *)back;
3005 fprintf(stderr, "Backref %llu parent %llu"
3006 " root %llu not found in extent tree\n",
3007 (unsigned long long)rec->start,
3008 (unsigned long long)tback->parent,
3009 (unsigned long long)tback->root);
3012 if (!back->is_data && !back->found_ref) {
3016 tback = (struct tree_backref *)back;
3017 fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
3018 (unsigned long long)rec->start,
3019 back->full_backref ? "parent" : "root",
3020 back->full_backref ?
3021 (unsigned long long)tback->parent :
3022 (unsigned long long)tback->root, back);
3024 if (back->is_data) {
3025 dback = (struct data_backref *)back;
3026 if (dback->found_ref != dback->num_refs) {
3030 fprintf(stderr, "Incorrect local backref count"
3031 " on %llu %s %llu owner %llu"
3032 " offset %llu found %u wanted %u back %p\n",
3033 (unsigned long long)rec->start,
3034 back->full_backref ?
3036 back->full_backref ?
3037 (unsigned long long)dback->parent:
3038 (unsigned long long)dback->root,
3039 (unsigned long long)dback->owner,
3040 (unsigned long long)dback->offset,
3041 dback->found_ref, dback->num_refs, back);
3043 if (dback->disk_bytenr != rec->start) {
3047 fprintf(stderr, "Backref disk bytenr does not"
3048 " match extent record, bytenr=%llu, "
3049 "ref bytenr=%llu\n",
3050 (unsigned long long)rec->start,
3051 (unsigned long long)dback->disk_bytenr);
3054 if (dback->bytes != rec->nr) {
3058 fprintf(stderr, "Backref bytes do not match "
3059 "extent backref, bytenr=%llu, ref "
3060 "bytes=%llu, backref bytes=%llu\n",
3061 (unsigned long long)rec->start,
3062 (unsigned long long)rec->nr,
3063 (unsigned long long)dback->bytes);
3066 if (!back->is_data) {
3069 dback = (struct data_backref *)back;
3070 found += dback->found_ref;
3073 if (found != rec->refs) {
3077 fprintf(stderr, "Incorrect global backref count "
3078 "on %llu found %llu wanted %llu\n",
3079 (unsigned long long)rec->start,
3080 (unsigned long long)found,
3081 (unsigned long long)rec->refs);
3087 static int free_all_extent_backrefs(struct extent_record *rec)
3089 struct extent_backref *back;
3090 struct list_head *cur;
3091 while (!list_empty(&rec->backrefs)) {
3092 cur = rec->backrefs.next;
3093 back = list_entry(cur, struct extent_backref, list);
3100 static void free_extent_record_cache(struct btrfs_fs_info *fs_info,
3101 struct cache_tree *extent_cache)
3103 struct cache_extent *cache;
3104 struct extent_record *rec;
3107 cache = first_cache_extent(extent_cache);
3110 rec = container_of(cache, struct extent_record, cache);
3111 btrfs_unpin_extent(fs_info, rec->start, rec->max_size);
3112 remove_cache_extent(extent_cache, cache);
3113 free_all_extent_backrefs(rec);
3118 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
3119 struct extent_record *rec)
3121 if (rec->content_checked && rec->owner_ref_checked &&
3122 rec->extent_item_refs == rec->refs && rec->refs > 0 &&
3123 rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0)) {
3124 remove_cache_extent(extent_cache, &rec->cache);
3125 free_all_extent_backrefs(rec);
3126 list_del_init(&rec->list);
3132 static int check_owner_ref(struct btrfs_root *root,
3133 struct extent_record *rec,
3134 struct extent_buffer *buf)
3136 struct extent_backref *node;
3137 struct tree_backref *back;
3138 struct btrfs_root *ref_root;
3139 struct btrfs_key key;
3140 struct btrfs_path path;
3141 struct extent_buffer *parent;
3146 list_for_each_entry(node, &rec->backrefs, list) {
3149 if (!node->found_ref)
3151 if (node->full_backref)
3153 back = (struct tree_backref *)node;
3154 if (btrfs_header_owner(buf) == back->root)
3157 BUG_ON(rec->is_root);
3159 /* try to find the block by search corresponding fs tree */
3160 key.objectid = btrfs_header_owner(buf);
3161 key.type = BTRFS_ROOT_ITEM_KEY;
3162 key.offset = (u64)-1;
3164 ref_root = btrfs_read_fs_root(root->fs_info, &key);
3165 if (IS_ERR(ref_root))
3168 level = btrfs_header_level(buf);
3170 btrfs_item_key_to_cpu(buf, &key, 0);
3172 btrfs_node_key_to_cpu(buf, &key, 0);
3174 btrfs_init_path(&path);
3175 path.lowest_level = level + 1;
3176 ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
3180 parent = path.nodes[level + 1];
3181 if (parent && buf->start == btrfs_node_blockptr(parent,
3182 path.slots[level + 1]))
3185 btrfs_release_path(&path);
3186 return found ? 0 : 1;
3189 static int is_extent_tree_record(struct extent_record *rec)
3191 struct list_head *cur = rec->backrefs.next;
3192 struct extent_backref *node;
3193 struct tree_backref *back;
3196 while(cur != &rec->backrefs) {
3197 node = list_entry(cur, struct extent_backref, list);
3201 back = (struct tree_backref *)node;
3202 if (node->full_backref)
3204 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
3211 static int record_bad_block_io(struct btrfs_fs_info *info,
3212 struct cache_tree *extent_cache,
3215 struct extent_record *rec;
3216 struct cache_extent *cache;
3217 struct btrfs_key key;
3219 cache = lookup_cache_extent(extent_cache, start, len);
3223 rec = container_of(cache, struct extent_record, cache);
3224 if (!is_extent_tree_record(rec))
3227 btrfs_disk_key_to_cpu(&key, &rec->parent_key);
3228 return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
3231 static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
3232 struct extent_buffer *buf, int slot)
3234 if (btrfs_header_level(buf)) {
3235 struct btrfs_key_ptr ptr1, ptr2;
3237 read_extent_buffer(buf, &ptr1, btrfs_node_key_ptr_offset(slot),
3238 sizeof(struct btrfs_key_ptr));
3239 read_extent_buffer(buf, &ptr2,
3240 btrfs_node_key_ptr_offset(slot + 1),
3241 sizeof(struct btrfs_key_ptr));
3242 write_extent_buffer(buf, &ptr1,
3243 btrfs_node_key_ptr_offset(slot + 1),
3244 sizeof(struct btrfs_key_ptr));
3245 write_extent_buffer(buf, &ptr2,
3246 btrfs_node_key_ptr_offset(slot),
3247 sizeof(struct btrfs_key_ptr));
3249 struct btrfs_disk_key key;
3250 btrfs_node_key(buf, &key, 0);
3251 btrfs_fixup_low_keys(root, path, &key,
3252 btrfs_header_level(buf) + 1);
3255 struct btrfs_item *item1, *item2;
3256 struct btrfs_key k1, k2;
3257 char *item1_data, *item2_data;
3258 u32 item1_offset, item2_offset, item1_size, item2_size;
3260 item1 = btrfs_item_nr(slot);
3261 item2 = btrfs_item_nr(slot + 1);
3262 btrfs_item_key_to_cpu(buf, &k1, slot);
3263 btrfs_item_key_to_cpu(buf, &k2, slot + 1);
3264 item1_offset = btrfs_item_offset(buf, item1);
3265 item2_offset = btrfs_item_offset(buf, item2);
3266 item1_size = btrfs_item_size(buf, item1);
3267 item2_size = btrfs_item_size(buf, item2);
3269 item1_data = malloc(item1_size);
3272 item2_data = malloc(item2_size);
3278 read_extent_buffer(buf, item1_data, item1_offset, item1_size);
3279 read_extent_buffer(buf, item2_data, item2_offset, item2_size);
3281 write_extent_buffer(buf, item1_data, item2_offset, item2_size);
3282 write_extent_buffer(buf, item2_data, item1_offset, item1_size);
3286 btrfs_set_item_offset(buf, item1, item2_offset);
3287 btrfs_set_item_offset(buf, item2, item1_offset);
3288 btrfs_set_item_size(buf, item1, item2_size);
3289 btrfs_set_item_size(buf, item2, item1_size);
3291 path->slots[0] = slot;
3292 btrfs_set_item_key_unsafe(root, path, &k2);
3293 path->slots[0] = slot + 1;
3294 btrfs_set_item_key_unsafe(root, path, &k1);
3299 static int fix_key_order(struct btrfs_trans_handle *trans,
3300 struct btrfs_root *root,
3301 struct btrfs_path *path)
3303 struct extent_buffer *buf;
3304 struct btrfs_key k1, k2;
3306 int level = path->lowest_level;
3309 buf = path->nodes[level];
3310 for (i = 0; i < btrfs_header_nritems(buf) - 1; i++) {
3312 btrfs_node_key_to_cpu(buf, &k1, i);
3313 btrfs_node_key_to_cpu(buf, &k2, i + 1);
3315 btrfs_item_key_to_cpu(buf, &k1, i);
3316 btrfs_item_key_to_cpu(buf, &k2, i + 1);
3318 if (btrfs_comp_cpu_keys(&k1, &k2) < 0)
3320 ret = swap_values(root, path, buf, i);
3323 btrfs_mark_buffer_dirty(buf);
3329 static int delete_bogus_item(struct btrfs_trans_handle *trans,
3330 struct btrfs_root *root,
3331 struct btrfs_path *path,
3332 struct extent_buffer *buf, int slot)
3334 struct btrfs_key key;
3335 int nritems = btrfs_header_nritems(buf);
3337 btrfs_item_key_to_cpu(buf, &key, slot);
3339 /* These are all the keys we can deal with missing. */
3340 if (key.type != BTRFS_DIR_INDEX_KEY &&
3341 key.type != BTRFS_EXTENT_ITEM_KEY &&
3342 key.type != BTRFS_METADATA_ITEM_KEY &&
3343 key.type != BTRFS_TREE_BLOCK_REF_KEY &&
3344 key.type != BTRFS_EXTENT_DATA_REF_KEY)
3347 printf("Deleting bogus item [%llu,%u,%llu] at slot %d on block %llu\n",
3348 (unsigned long long)key.objectid, key.type,
3349 (unsigned long long)key.offset, slot, buf->start);
3350 memmove_extent_buffer(buf, btrfs_item_nr_offset(slot),
3351 btrfs_item_nr_offset(slot + 1),
3352 sizeof(struct btrfs_item) *
3353 (nritems - slot - 1));
3354 btrfs_set_header_nritems(buf, nritems - 1);
3356 struct btrfs_disk_key disk_key;
3358 btrfs_item_key(buf, &disk_key, 0);
3359 btrfs_fixup_low_keys(root, path, &disk_key, 1);
3361 btrfs_mark_buffer_dirty(buf);
3365 static int fix_item_offset(struct btrfs_trans_handle *trans,
3366 struct btrfs_root *root,
3367 struct btrfs_path *path)
3369 struct extent_buffer *buf;
3373 /* We should only get this for leaves */
3374 BUG_ON(path->lowest_level);
3375 buf = path->nodes[0];
3377 for (i = 0; i < btrfs_header_nritems(buf); i++) {
3378 unsigned int shift = 0, offset;
3380 if (i == 0 && btrfs_item_end_nr(buf, i) !=
3381 BTRFS_LEAF_DATA_SIZE(root)) {
3382 if (btrfs_item_end_nr(buf, i) >
3383 BTRFS_LEAF_DATA_SIZE(root)) {
3384 ret = delete_bogus_item(trans, root, path,
3388 fprintf(stderr, "item is off the end of the "
3389 "leaf, can't fix\n");
3393 shift = BTRFS_LEAF_DATA_SIZE(root) -
3394 btrfs_item_end_nr(buf, i);
3395 } else if (i > 0 && btrfs_item_end_nr(buf, i) !=
3396 btrfs_item_offset_nr(buf, i - 1)) {
3397 if (btrfs_item_end_nr(buf, i) >
3398 btrfs_item_offset_nr(buf, i - 1)) {
3399 ret = delete_bogus_item(trans, root, path,
3403 fprintf(stderr, "items overlap, can't fix\n");
3407 shift = btrfs_item_offset_nr(buf, i - 1) -
3408 btrfs_item_end_nr(buf, i);
3413 printf("Shifting item nr %d by %u bytes in block %llu\n",
3414 i, shift, (unsigned long long)buf->start);
3415 offset = btrfs_item_offset_nr(buf, i);
3416 memmove_extent_buffer(buf,
3417 btrfs_leaf_data(buf) + offset + shift,
3418 btrfs_leaf_data(buf) + offset,
3419 btrfs_item_size_nr(buf, i));
3420 btrfs_set_item_offset(buf, btrfs_item_nr(i),
3422 btrfs_mark_buffer_dirty(buf);
3426 * We may have moved things, in which case we want to exit so we don't
3427 * write those changes out. Once we have proper abort functionality in
3428 * progs this can be changed to something nicer.
3435 * Attempt to fix basic block failures. If we can't fix it for whatever reason
3436 * then just return -EIO.
3438 static int try_to_fix_bad_block(struct btrfs_trans_handle *trans,
3439 struct btrfs_root *root,
3440 struct extent_buffer *buf,
3441 enum btrfs_tree_block_status status)
3443 struct ulist *roots;
3444 struct ulist_node *node;
3445 struct btrfs_root *search_root;
3446 struct btrfs_path *path;
3447 struct ulist_iterator iter;
3448 struct btrfs_key root_key, key;
3451 if (status != BTRFS_TREE_BLOCK_BAD_KEY_ORDER &&
3452 status != BTRFS_TREE_BLOCK_INVALID_OFFSETS)
3455 path = btrfs_alloc_path();
3459 ret = btrfs_find_all_roots(trans, root->fs_info, buf->start,
3462 btrfs_free_path(path);
3466 ULIST_ITER_INIT(&iter);
3467 while ((node = ulist_next(roots, &iter))) {
3468 root_key.objectid = node->val;
3469 root_key.type = BTRFS_ROOT_ITEM_KEY;
3470 root_key.offset = (u64)-1;
3472 search_root = btrfs_read_fs_root(root->fs_info, &root_key);
3478 record_root_in_trans(trans, search_root);
3480 path->lowest_level = btrfs_header_level(buf);
3481 path->skip_check_block = 1;
3482 if (path->lowest_level)
3483 btrfs_node_key_to_cpu(buf, &key, 0);
3485 btrfs_item_key_to_cpu(buf, &key, 0);
3486 ret = btrfs_search_slot(trans, search_root, &key, path, 0, 1);
3491 if (status == BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
3492 ret = fix_key_order(trans, search_root, path);
3493 else if (status == BTRFS_TREE_BLOCK_INVALID_OFFSETS)
3494 ret = fix_item_offset(trans, search_root, path);
3497 btrfs_release_path(path);
3500 btrfs_free_path(path);
3504 static int check_block(struct btrfs_trans_handle *trans,
3505 struct btrfs_root *root,
3506 struct cache_tree *extent_cache,
3507 struct extent_buffer *buf, u64 flags)
3509 struct extent_record *rec;
3510 struct cache_extent *cache;
3511 struct btrfs_key key;
3512 enum btrfs_tree_block_status status;
3516 cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
3519 rec = container_of(cache, struct extent_record, cache);
3520 rec->generation = btrfs_header_generation(buf);
3522 level = btrfs_header_level(buf);
3523 if (btrfs_header_nritems(buf) > 0) {
3526 btrfs_item_key_to_cpu(buf, &key, 0);
3528 btrfs_node_key_to_cpu(buf, &key, 0);
3530 rec->info_objectid = key.objectid;
3532 rec->info_level = level;
3534 if (btrfs_is_leaf(buf))
3535 status = btrfs_check_leaf(root, &rec->parent_key, buf);
3537 status = btrfs_check_node(root, &rec->parent_key, buf);
3539 if (status != BTRFS_TREE_BLOCK_CLEAN) {
3541 status = try_to_fix_bad_block(trans, root, buf,
3543 if (status != BTRFS_TREE_BLOCK_CLEAN) {
3545 fprintf(stderr, "bad block %llu\n",
3546 (unsigned long long)buf->start);
3549 * Signal to callers we need to start the scan over
3550 * again since we'll have cow'ed blocks.
3555 rec->content_checked = 1;
3556 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3557 rec->owner_ref_checked = 1;
3559 ret = check_owner_ref(root, rec, buf);
3561 rec->owner_ref_checked = 1;
3565 maybe_free_extent_rec(extent_cache, rec);
3569 static struct tree_backref *find_tree_backref(struct extent_record *rec,
3570 u64 parent, u64 root)
3572 struct list_head *cur = rec->backrefs.next;
3573 struct extent_backref *node;
3574 struct tree_backref *back;
3576 while(cur != &rec->backrefs) {
3577 node = list_entry(cur, struct extent_backref, list);
3581 back = (struct tree_backref *)node;
3583 if (!node->full_backref)
3585 if (parent == back->parent)
3588 if (node->full_backref)
3590 if (back->root == root)
3597 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
3598 u64 parent, u64 root)
3600 struct tree_backref *ref = malloc(sizeof(*ref));
3601 memset(&ref->node, 0, sizeof(ref->node));
3603 ref->parent = parent;
3604 ref->node.full_backref = 1;
3607 ref->node.full_backref = 0;
3609 list_add_tail(&ref->node.list, &rec->backrefs);
3614 static struct data_backref *find_data_backref(struct extent_record *rec,
3615 u64 parent, u64 root,
3616 u64 owner, u64 offset,
3618 u64 disk_bytenr, u64 bytes)
3620 struct list_head *cur = rec->backrefs.next;
3621 struct extent_backref *node;
3622 struct data_backref *back;
3624 while(cur != &rec->backrefs) {
3625 node = list_entry(cur, struct extent_backref, list);
3629 back = (struct data_backref *)node;
3631 if (!node->full_backref)
3633 if (parent == back->parent)
3636 if (node->full_backref)
3638 if (back->root == root && back->owner == owner &&
3639 back->offset == offset) {
3640 if (found_ref && node->found_ref &&
3641 (back->bytes != bytes ||
3642 back->disk_bytenr != disk_bytenr))
3651 static struct data_backref *alloc_data_backref(struct extent_record *rec,
3652 u64 parent, u64 root,
3653 u64 owner, u64 offset,
3656 struct data_backref *ref = malloc(sizeof(*ref));
3657 memset(&ref->node, 0, sizeof(ref->node));
3658 ref->node.is_data = 1;
3661 ref->parent = parent;
3664 ref->node.full_backref = 1;
3668 ref->offset = offset;
3669 ref->node.full_backref = 0;
3671 ref->bytes = max_size;
3674 list_add_tail(&ref->node.list, &rec->backrefs);
3675 if (max_size > rec->max_size)
3676 rec->max_size = max_size;
3680 static int add_extent_rec(struct cache_tree *extent_cache,
3681 struct btrfs_key *parent_key, u64 parent_gen,
3682 u64 start, u64 nr, u64 extent_item_refs,
3683 int is_root, int inc_ref, int set_checked,
3684 int metadata, int extent_rec, u64 max_size)
3686 struct extent_record *rec;
3687 struct cache_extent *cache;
3691 cache = lookup_cache_extent(extent_cache, start, nr);
3693 rec = container_of(cache, struct extent_record, cache);
3697 rec->nr = max(nr, max_size);
3700 * We need to make sure to reset nr to whatever the extent
3701 * record says was the real size, this way we can compare it to
3705 if (start != rec->start || rec->found_rec) {
3706 struct extent_record *tmp;
3709 if (list_empty(&rec->list))
3710 list_add_tail(&rec->list,
3711 &duplicate_extents);
3714 * We have to do this song and dance in case we
3715 * find an extent record that falls inside of
3716 * our current extent record but does not have
3717 * the same objectid.
3719 tmp = malloc(sizeof(*tmp));
3723 tmp->max_size = max_size;
3726 tmp->metadata = metadata;
3727 tmp->extent_item_refs = extent_item_refs;
3728 INIT_LIST_HEAD(&tmp->list);
3729 list_add_tail(&tmp->list, &rec->dups);
3730 rec->num_duplicates++;
3737 if (extent_item_refs && !dup) {
3738 if (rec->extent_item_refs) {
3739 fprintf(stderr, "block %llu rec "
3740 "extent_item_refs %llu, passed %llu\n",
3741 (unsigned long long)start,
3742 (unsigned long long)
3743 rec->extent_item_refs,
3744 (unsigned long long)extent_item_refs);
3746 rec->extent_item_refs = extent_item_refs;
3751 rec->content_checked = 1;
3752 rec->owner_ref_checked = 1;
3756 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
3758 rec->parent_generation = parent_gen;
3760 if (rec->max_size < max_size)
3761 rec->max_size = max_size;
3763 maybe_free_extent_rec(extent_cache, rec);
3766 rec = malloc(sizeof(*rec));
3768 rec->max_size = max_size;
3769 rec->nr = max(nr, max_size);
3770 rec->found_rec = !!extent_rec;
3771 rec->content_checked = 0;
3772 rec->owner_ref_checked = 0;
3773 rec->num_duplicates = 0;
3774 rec->metadata = metadata;
3775 INIT_LIST_HEAD(&rec->backrefs);
3776 INIT_LIST_HEAD(&rec->dups);
3777 INIT_LIST_HEAD(&rec->list);
3789 if (extent_item_refs)
3790 rec->extent_item_refs = extent_item_refs;
3792 rec->extent_item_refs = 0;
3795 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
3797 memset(&rec->parent_key, 0, sizeof(*parent_key));
3800 rec->parent_generation = parent_gen;
3802 rec->parent_generation = 0;
3804 rec->cache.start = start;
3805 rec->cache.size = nr;
3806 ret = insert_cache_extent(extent_cache, &rec->cache);
3810 rec->content_checked = 1;
3811 rec->owner_ref_checked = 1;
3816 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
3817 u64 parent, u64 root, int found_ref)
3819 struct extent_record *rec;
3820 struct tree_backref *back;
3821 struct cache_extent *cache;
3823 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3825 add_extent_rec(extent_cache, NULL, 0, bytenr,
3826 1, 0, 0, 0, 0, 1, 0, 0);
3827 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3832 rec = container_of(cache, struct extent_record, cache);
3833 if (rec->start != bytenr) {
3837 back = find_tree_backref(rec, parent, root);
3839 back = alloc_tree_backref(rec, parent, root);
3842 if (back->node.found_ref) {
3843 fprintf(stderr, "Extent back ref already exists "
3844 "for %llu parent %llu root %llu \n",
3845 (unsigned long long)bytenr,
3846 (unsigned long long)parent,
3847 (unsigned long long)root);
3849 back->node.found_ref = 1;
3851 if (back->node.found_extent_tree) {
3852 fprintf(stderr, "Extent back ref already exists "
3853 "for %llu parent %llu root %llu \n",
3854 (unsigned long long)bytenr,
3855 (unsigned long long)parent,
3856 (unsigned long long)root);
3858 back->node.found_extent_tree = 1;
3860 maybe_free_extent_rec(extent_cache, rec);
3864 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
3865 u64 parent, u64 root, u64 owner, u64 offset,
3866 u32 num_refs, int found_ref, u64 max_size)
3868 struct extent_record *rec;
3869 struct data_backref *back;
3870 struct cache_extent *cache;
3872 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3874 add_extent_rec(extent_cache, NULL, 0, bytenr, 1, 0, 0, 0, 0,
3876 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3881 rec = container_of(cache, struct extent_record, cache);
3882 if (rec->max_size < max_size)
3883 rec->max_size = max_size;
3886 * If found_ref is set then max_size is the real size and must match the
3887 * existing refs. So if we have already found a ref then we need to
3888 * make sure that this ref matches the existing one, otherwise we need
3889 * to add a new backref so we can notice that the backrefs don't match
3890 * and we need to figure out who is telling the truth. This is to
3891 * account for that awful fsync bug I introduced where we'd end up with
3892 * a btrfs_file_extent_item that would have its length include multiple
3893 * prealloc extents or point inside of a prealloc extent.
3895 back = find_data_backref(rec, parent, root, owner, offset, found_ref,
3898 back = alloc_data_backref(rec, parent, root, owner, offset,
3902 BUG_ON(num_refs != 1);
3903 if (back->node.found_ref)
3904 BUG_ON(back->bytes != max_size);
3905 back->node.found_ref = 1;
3906 back->found_ref += 1;
3907 back->bytes = max_size;
3908 back->disk_bytenr = bytenr;
3910 rec->content_checked = 1;
3911 rec->owner_ref_checked = 1;
3913 if (back->node.found_extent_tree) {
3914 fprintf(stderr, "Extent back ref already exists "
3915 "for %llu parent %llu root %llu "
3916 "owner %llu offset %llu num_refs %lu\n",
3917 (unsigned long long)bytenr,
3918 (unsigned long long)parent,
3919 (unsigned long long)root,
3920 (unsigned long long)owner,
3921 (unsigned long long)offset,
3922 (unsigned long)num_refs);
3924 back->num_refs = num_refs;
3925 back->node.found_extent_tree = 1;
3927 maybe_free_extent_rec(extent_cache, rec);
3931 static int add_pending(struct cache_tree *pending,
3932 struct cache_tree *seen, u64 bytenr, u32 size)
3935 ret = add_cache_extent(seen, bytenr, size);
3938 add_cache_extent(pending, bytenr, size);
3942 static int pick_next_pending(struct cache_tree *pending,
3943 struct cache_tree *reada,
3944 struct cache_tree *nodes,
3945 u64 last, struct block_info *bits, int bits_nr,
3948 unsigned long node_start = last;
3949 struct cache_extent *cache;
3952 cache = search_cache_extent(reada, 0);
3954 bits[0].start = cache->start;
3955 bits[0].size = cache->size;
3960 if (node_start > 32768)
3961 node_start -= 32768;
3963 cache = search_cache_extent(nodes, node_start);
3965 cache = search_cache_extent(nodes, 0);
3968 cache = search_cache_extent(pending, 0);
3973 bits[ret].start = cache->start;
3974 bits[ret].size = cache->size;
3975 cache = next_cache_extent(cache);
3977 } while (cache && ret < bits_nr);
3983 bits[ret].start = cache->start;
3984 bits[ret].size = cache->size;
3985 cache = next_cache_extent(cache);
3987 } while (cache && ret < bits_nr);
3989 if (bits_nr - ret > 8) {
3990 u64 lookup = bits[0].start + bits[0].size;
3991 struct cache_extent *next;
3992 next = search_cache_extent(pending, lookup);
3994 if (next->start - lookup > 32768)
3996 bits[ret].start = next->start;
3997 bits[ret].size = next->size;
3998 lookup = next->start + next->size;
4002 next = next_cache_extent(next);
4010 static void free_chunk_record(struct cache_extent *cache)
4012 struct chunk_record *rec;
4014 rec = container_of(cache, struct chunk_record, cache);
4015 list_del_init(&rec->list);
4016 list_del_init(&rec->dextents);
4020 void free_chunk_cache_tree(struct cache_tree *chunk_cache)
4022 cache_tree_free_extents(chunk_cache, free_chunk_record);
4025 static void free_device_record(struct rb_node *node)
4027 struct device_record *rec;
4029 rec = container_of(node, struct device_record, node);
4033 FREE_RB_BASED_TREE(device_cache, free_device_record);
4035 int insert_block_group_record(struct block_group_tree *tree,
4036 struct block_group_record *bg_rec)
4040 ret = insert_cache_extent(&tree->tree, &bg_rec->cache);
4044 list_add_tail(&bg_rec->list, &tree->block_groups);
4048 static void free_block_group_record(struct cache_extent *cache)
4050 struct block_group_record *rec;
4052 rec = container_of(cache, struct block_group_record, cache);
4053 list_del_init(&rec->list);
4057 void free_block_group_tree(struct block_group_tree *tree)
4059 cache_tree_free_extents(&tree->tree, free_block_group_record);
4062 int insert_device_extent_record(struct device_extent_tree *tree,
4063 struct device_extent_record *de_rec)
4068 * Device extent is a bit different from the other extents, because
4069 * the extents which belong to the different devices may have the
4070 * same start and size, so we need use the special extent cache
4071 * search/insert functions.
4073 ret = insert_cache_extent2(&tree->tree, &de_rec->cache);
4077 list_add_tail(&de_rec->chunk_list, &tree->no_chunk_orphans);
4078 list_add_tail(&de_rec->device_list, &tree->no_device_orphans);
4082 static void free_device_extent_record(struct cache_extent *cache)
4084 struct device_extent_record *rec;
4086 rec = container_of(cache, struct device_extent_record, cache);
4087 if (!list_empty(&rec->chunk_list))
4088 list_del_init(&rec->chunk_list);
4089 if (!list_empty(&rec->device_list))
4090 list_del_init(&rec->device_list);
4094 void free_device_extent_tree(struct device_extent_tree *tree)
4096 cache_tree_free_extents(&tree->tree, free_device_extent_record);
4099 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4100 static int process_extent_ref_v0(struct cache_tree *extent_cache,
4101 struct extent_buffer *leaf, int slot)
4103 struct btrfs_extent_ref_v0 *ref0;
4104 struct btrfs_key key;
4106 btrfs_item_key_to_cpu(leaf, &key, slot);
4107 ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
4108 if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
4109 add_tree_backref(extent_cache, key.objectid, key.offset, 0, 0);
4111 add_data_backref(extent_cache, key.objectid, key.offset, 0,
4112 0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
4118 struct chunk_record *btrfs_new_chunk_record(struct extent_buffer *leaf,
4119 struct btrfs_key *key,
4122 struct btrfs_chunk *ptr;
4123 struct chunk_record *rec;
4126 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4127 num_stripes = btrfs_chunk_num_stripes(leaf, ptr);
4129 rec = malloc(btrfs_chunk_record_size(num_stripes));
4131 fprintf(stderr, "memory allocation failed\n");
4135 memset(rec, 0, btrfs_chunk_record_size(num_stripes));
4137 INIT_LIST_HEAD(&rec->list);
4138 INIT_LIST_HEAD(&rec->dextents);
4141 rec->cache.start = key->offset;
4142 rec->cache.size = btrfs_chunk_length(leaf, ptr);
4144 rec->generation = btrfs_header_generation(leaf);
4146 rec->objectid = key->objectid;
4147 rec->type = key->type;
4148 rec->offset = key->offset;
4150 rec->length = rec->cache.size;
4151 rec->owner = btrfs_chunk_owner(leaf, ptr);
4152 rec->stripe_len = btrfs_chunk_stripe_len(leaf, ptr);
4153 rec->type_flags = btrfs_chunk_type(leaf, ptr);
4154 rec->io_width = btrfs_chunk_io_width(leaf, ptr);
4155 rec->io_align = btrfs_chunk_io_align(leaf, ptr);
4156 rec->sector_size = btrfs_chunk_sector_size(leaf, ptr);
4157 rec->num_stripes = num_stripes;
4158 rec->sub_stripes = btrfs_chunk_sub_stripes(leaf, ptr);
4160 for (i = 0; i < rec->num_stripes; ++i) {
4161 rec->stripes[i].devid =
4162 btrfs_stripe_devid_nr(leaf, ptr, i);
4163 rec->stripes[i].offset =
4164 btrfs_stripe_offset_nr(leaf, ptr, i);
4165 read_extent_buffer(leaf, rec->stripes[i].dev_uuid,
4166 (unsigned long)btrfs_stripe_dev_uuid_nr(ptr, i),
4173 static int process_chunk_item(struct cache_tree *chunk_cache,
4174 struct btrfs_key *key, struct extent_buffer *eb,
4177 struct chunk_record *rec;
4180 rec = btrfs_new_chunk_record(eb, key, slot);
4181 ret = insert_cache_extent(chunk_cache, &rec->cache);
4183 fprintf(stderr, "Chunk[%llu, %llu] existed.\n",
4184 rec->offset, rec->length);
4191 static int process_device_item(struct rb_root *dev_cache,
4192 struct btrfs_key *key, struct extent_buffer *eb, int slot)
4194 struct btrfs_dev_item *ptr;
4195 struct device_record *rec;
4198 ptr = btrfs_item_ptr(eb,
4199 slot, struct btrfs_dev_item);
4201 rec = malloc(sizeof(*rec));
4203 fprintf(stderr, "memory allocation failed\n");
4207 rec->devid = key->offset;
4208 rec->generation = btrfs_header_generation(eb);
4210 rec->objectid = key->objectid;
4211 rec->type = key->type;
4212 rec->offset = key->offset;
4214 rec->devid = btrfs_device_id(eb, ptr);
4215 rec->total_byte = btrfs_device_total_bytes(eb, ptr);
4216 rec->byte_used = btrfs_device_bytes_used(eb, ptr);
4218 ret = rb_insert(dev_cache, &rec->node, device_record_compare);
4220 fprintf(stderr, "Device[%llu] existed.\n", rec->devid);
4227 struct block_group_record *
4228 btrfs_new_block_group_record(struct extent_buffer *leaf, struct btrfs_key *key,
4231 struct btrfs_block_group_item *ptr;
4232 struct block_group_record *rec;
4234 rec = malloc(sizeof(*rec));
4236 fprintf(stderr, "memory allocation failed\n");
4239 memset(rec, 0, sizeof(*rec));
4241 rec->cache.start = key->objectid;
4242 rec->cache.size = key->offset;
4244 rec->generation = btrfs_header_generation(leaf);
4246 rec->objectid = key->objectid;
4247 rec->type = key->type;
4248 rec->offset = key->offset;
4250 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item);
4251 rec->flags = btrfs_disk_block_group_flags(leaf, ptr);
4253 INIT_LIST_HEAD(&rec->list);
4258 static int process_block_group_item(struct block_group_tree *block_group_cache,
4259 struct btrfs_key *key,
4260 struct extent_buffer *eb, int slot)
4262 struct block_group_record *rec;
4265 rec = btrfs_new_block_group_record(eb, key, slot);
4266 ret = insert_block_group_record(block_group_cache, rec);
4268 fprintf(stderr, "Block Group[%llu, %llu] existed.\n",
4269 rec->objectid, rec->offset);
4276 struct device_extent_record *
4277 btrfs_new_device_extent_record(struct extent_buffer *leaf,
4278 struct btrfs_key *key, int slot)
4280 struct device_extent_record *rec;
4281 struct btrfs_dev_extent *ptr;
4283 rec = malloc(sizeof(*rec));
4285 fprintf(stderr, "memory allocation failed\n");
4288 memset(rec, 0, sizeof(*rec));
4290 rec->cache.objectid = key->objectid;
4291 rec->cache.start = key->offset;
4293 rec->generation = btrfs_header_generation(leaf);
4295 rec->objectid = key->objectid;
4296 rec->type = key->type;
4297 rec->offset = key->offset;
4299 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
4300 rec->chunk_objecteid =
4301 btrfs_dev_extent_chunk_objectid(leaf, ptr);
4303 btrfs_dev_extent_chunk_offset(leaf, ptr);
4304 rec->length = btrfs_dev_extent_length(leaf, ptr);
4305 rec->cache.size = rec->length;
4307 INIT_LIST_HEAD(&rec->chunk_list);
4308 INIT_LIST_HEAD(&rec->device_list);
4314 process_device_extent_item(struct device_extent_tree *dev_extent_cache,
4315 struct btrfs_key *key, struct extent_buffer *eb,
4318 struct device_extent_record *rec;
4321 rec = btrfs_new_device_extent_record(eb, key, slot);
4322 ret = insert_device_extent_record(dev_extent_cache, rec);
4325 "Device extent[%llu, %llu, %llu] existed.\n",
4326 rec->objectid, rec->offset, rec->length);
4333 static int process_extent_item(struct btrfs_root *root,
4334 struct cache_tree *extent_cache,
4335 struct extent_buffer *eb, int slot)
4337 struct btrfs_extent_item *ei;
4338 struct btrfs_extent_inline_ref *iref;
4339 struct btrfs_extent_data_ref *dref;
4340 struct btrfs_shared_data_ref *sref;
4341 struct btrfs_key key;
4345 u32 item_size = btrfs_item_size_nr(eb, slot);
4351 btrfs_item_key_to_cpu(eb, &key, slot);
4353 if (key.type == BTRFS_METADATA_ITEM_KEY) {
4355 num_bytes = root->leafsize;
4357 num_bytes = key.offset;
4360 if (item_size < sizeof(*ei)) {
4361 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4362 struct btrfs_extent_item_v0 *ei0;
4363 BUG_ON(item_size != sizeof(*ei0));
4364 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
4365 refs = btrfs_extent_refs_v0(eb, ei0);
4369 return add_extent_rec(extent_cache, NULL, 0, key.objectid,
4370 num_bytes, refs, 0, 0, 0, metadata, 1,
4374 ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
4375 refs = btrfs_extent_refs(eb, ei);
4377 add_extent_rec(extent_cache, NULL, 0, key.objectid, num_bytes,
4378 refs, 0, 0, 0, metadata, 1, num_bytes);
4380 ptr = (unsigned long)(ei + 1);
4381 if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
4382 key.type == BTRFS_EXTENT_ITEM_KEY)
4383 ptr += sizeof(struct btrfs_tree_block_info);
4385 end = (unsigned long)ei + item_size;
4387 iref = (struct btrfs_extent_inline_ref *)ptr;
4388 type = btrfs_extent_inline_ref_type(eb, iref);
4389 offset = btrfs_extent_inline_ref_offset(eb, iref);
4391 case BTRFS_TREE_BLOCK_REF_KEY:
4392 add_tree_backref(extent_cache, key.objectid,
4395 case BTRFS_SHARED_BLOCK_REF_KEY:
4396 add_tree_backref(extent_cache, key.objectid,
4399 case BTRFS_EXTENT_DATA_REF_KEY:
4400 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
4401 add_data_backref(extent_cache, key.objectid, 0,
4402 btrfs_extent_data_ref_root(eb, dref),
4403 btrfs_extent_data_ref_objectid(eb,
4405 btrfs_extent_data_ref_offset(eb, dref),
4406 btrfs_extent_data_ref_count(eb, dref),
4409 case BTRFS_SHARED_DATA_REF_KEY:
4410 sref = (struct btrfs_shared_data_ref *)(iref + 1);
4411 add_data_backref(extent_cache, key.objectid, offset,
4413 btrfs_shared_data_ref_count(eb, sref),
4417 fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
4418 key.objectid, key.type, num_bytes);
4421 ptr += btrfs_extent_inline_ref_size(type);
4428 static int check_cache_range(struct btrfs_root *root,
4429 struct btrfs_block_group_cache *cache,
4430 u64 offset, u64 bytes)
4432 struct btrfs_free_space *entry;
4438 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
4439 bytenr = btrfs_sb_offset(i);
4440 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
4441 cache->key.objectid, bytenr, 0,
4442 &logical, &nr, &stripe_len);
4447 if (logical[nr] + stripe_len <= offset)
4449 if (offset + bytes <= logical[nr])
4451 if (logical[nr] == offset) {
4452 if (stripe_len >= bytes) {
4456 bytes -= stripe_len;
4457 offset += stripe_len;
4458 } else if (logical[nr] < offset) {
4459 if (logical[nr] + stripe_len >=
4464 bytes = (offset + bytes) -
4465 (logical[nr] + stripe_len);
4466 offset = logical[nr] + stripe_len;
4469 * Could be tricky, the super may land in the
4470 * middle of the area we're checking. First
4471 * check the easiest case, it's at the end.
4473 if (logical[nr] + stripe_len >=
4475 bytes = logical[nr] - offset;
4479 /* Check the left side */
4480 ret = check_cache_range(root, cache,
4482 logical[nr] - offset);
4488 /* Now we continue with the right side */
4489 bytes = (offset + bytes) -
4490 (logical[nr] + stripe_len);
4491 offset = logical[nr] + stripe_len;
4498 entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
4500 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
4501 offset, offset+bytes);
4505 if (entry->offset != offset) {
4506 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
4511 if (entry->bytes != bytes) {
4512 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
4513 bytes, entry->bytes, offset);
4517 unlink_free_space(cache->free_space_ctl, entry);
4522 static int verify_space_cache(struct btrfs_root *root,
4523 struct btrfs_block_group_cache *cache)
4525 struct btrfs_path *path;
4526 struct extent_buffer *leaf;
4527 struct btrfs_key key;
4531 path = btrfs_alloc_path();
4535 root = root->fs_info->extent_root;
4537 last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
4539 key.objectid = last;
4541 key.type = BTRFS_EXTENT_ITEM_KEY;
4543 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4548 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4549 ret = btrfs_next_leaf(root, path);
4557 leaf = path->nodes[0];
4558 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4559 if (key.objectid >= cache->key.offset + cache->key.objectid)
4561 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
4562 key.type != BTRFS_METADATA_ITEM_KEY) {
4567 if (last == key.objectid) {
4568 if (key.type == BTRFS_EXTENT_ITEM_KEY)
4569 last = key.objectid + key.offset;
4571 last = key.objectid + root->leafsize;
4576 ret = check_cache_range(root, cache, last,
4577 key.objectid - last);
4580 if (key.type == BTRFS_EXTENT_ITEM_KEY)
4581 last = key.objectid + key.offset;
4583 last = key.objectid + root->leafsize;
4587 if (last < cache->key.objectid + cache->key.offset)
4588 ret = check_cache_range(root, cache, last,
4589 cache->key.objectid +
4590 cache->key.offset - last);
4593 btrfs_free_path(path);
4596 !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
4597 fprintf(stderr, "There are still entries left in the space "
4605 static int check_space_cache(struct btrfs_root *root)
4607 struct btrfs_block_group_cache *cache;
4608 u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
4612 if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
4613 btrfs_super_generation(root->fs_info->super_copy) !=
4614 btrfs_super_cache_generation(root->fs_info->super_copy)) {
4615 printf("cache and super generation don't match, space cache "
4616 "will be invalidated\n");
4621 cache = btrfs_lookup_first_block_group(root->fs_info, start);
4625 start = cache->key.objectid + cache->key.offset;
4626 if (!cache->free_space_ctl) {
4627 if (btrfs_init_free_space_ctl(cache,
4628 root->sectorsize)) {
4633 btrfs_remove_free_space_cache(cache);
4636 ret = load_free_space_cache(root->fs_info, cache);
4640 ret = verify_space_cache(root, cache);
4642 fprintf(stderr, "cache appears valid but isnt %Lu\n",
4643 cache->key.objectid);
4648 return error ? -EINVAL : 0;
4651 static int read_extent_data(struct btrfs_root *root, char *data,
4652 u64 logical, u64 *len, int mirror)
4655 struct btrfs_multi_bio *multi = NULL;
4656 struct btrfs_fs_info *info = root->fs_info;
4657 struct btrfs_device *device;
4661 ret = btrfs_map_block(&info->mapping_tree, READ, logical, len,
4662 &multi, mirror, NULL);
4664 fprintf(stderr, "Couldn't map the block %llu\n",
4668 device = multi->stripes[0].dev;
4670 if (device->fd == 0)
4675 ret = pread64(device->fd, data, *len, multi->stripes[0].physical);
4685 static int check_extent_csums(struct btrfs_root *root, u64 bytenr,
4686 u64 num_bytes, unsigned long leaf_offset,
4687 struct extent_buffer *eb) {
4690 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
4692 unsigned long csum_offset;
4696 u64 data_checked = 0;
4702 if (num_bytes % root->sectorsize)
4705 data = malloc(num_bytes);
4709 while (offset < num_bytes) {
4712 read_len = num_bytes - offset;
4713 /* read as much space once a time */
4714 ret = read_extent_data(root, data + offset,
4715 bytenr + offset, &read_len, mirror);
4719 /* verify every 4k data's checksum */
4720 while (data_checked < read_len) {
4722 tmp = offset + data_checked;
4724 csum = btrfs_csum_data(NULL, (char *)data + tmp,
4725 csum, root->sectorsize);
4726 btrfs_csum_final(csum, (char *)&csum);
4728 csum_offset = leaf_offset +
4729 tmp / root->sectorsize * csum_size;
4730 read_extent_buffer(eb, (char *)&csum_expected,
4731 csum_offset, csum_size);
4732 /* try another mirror */
4733 if (csum != csum_expected) {
4734 fprintf(stderr, "mirror %d bytenr %llu csum %u expected csum %u\n",
4735 mirror, bytenr + tmp,
4736 csum, csum_expected);
4737 num_copies = btrfs_num_copies(
4738 &root->fs_info->mapping_tree,
4740 if (mirror < num_copies - 1) {
4745 data_checked += root->sectorsize;
4754 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
4757 struct btrfs_path *path;
4758 struct extent_buffer *leaf;
4759 struct btrfs_key key;
4762 path = btrfs_alloc_path();
4764 fprintf(stderr, "Error allocing path\n");
4768 key.objectid = bytenr;
4769 key.type = BTRFS_EXTENT_ITEM_KEY;
4770 key.offset = (u64)-1;
4773 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
4776 fprintf(stderr, "Error looking up extent record %d\n", ret);
4777 btrfs_free_path(path);
4780 if (path->slots[0] > 0) {
4783 ret = btrfs_prev_leaf(root, path);
4786 } else if (ret > 0) {
4793 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4796 * Block group items come before extent items if they have the same
4797 * bytenr, so walk back one more just in case. Dear future traveler,
4798 * first congrats on mastering time travel. Now if it's not too much
4799 * trouble could you go back to 2006 and tell Chris to make the
4800 * BLOCK_GROUP_ITEM_KEY (and BTRFS_*_REF_KEY) lower than the
4801 * EXTENT_ITEM_KEY please?
4803 while (key.type > BTRFS_EXTENT_ITEM_KEY) {
4804 if (path->slots[0] > 0) {
4807 ret = btrfs_prev_leaf(root, path);
4810 } else if (ret > 0) {
4815 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4819 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4820 ret = btrfs_next_leaf(root, path);
4822 fprintf(stderr, "Error going to next leaf "
4824 btrfs_free_path(path);
4830 leaf = path->nodes[0];
4831 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4832 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
4836 if (key.objectid + key.offset < bytenr) {
4840 if (key.objectid > bytenr + num_bytes)
4843 if (key.objectid == bytenr) {
4844 if (key.offset >= num_bytes) {
4848 num_bytes -= key.offset;
4849 bytenr += key.offset;
4850 } else if (key.objectid < bytenr) {
4851 if (key.objectid + key.offset >= bytenr + num_bytes) {
4855 num_bytes = (bytenr + num_bytes) -
4856 (key.objectid + key.offset);
4857 bytenr = key.objectid + key.offset;
4859 if (key.objectid + key.offset < bytenr + num_bytes) {
4860 u64 new_start = key.objectid + key.offset;
4861 u64 new_bytes = bytenr + num_bytes - new_start;
4864 * Weird case, the extent is in the middle of
4865 * our range, we'll have to search one side
4866 * and then the other. Not sure if this happens
4867 * in real life, but no harm in coding it up
4868 * anyway just in case.
4870 btrfs_release_path(path);
4871 ret = check_extent_exists(root, new_start,
4874 fprintf(stderr, "Right section didn't "
4878 num_bytes = key.objectid - bytenr;
4881 num_bytes = key.objectid - bytenr;
4888 if (num_bytes && !ret) {
4889 fprintf(stderr, "There are no extents for csum range "
4890 "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
4894 btrfs_free_path(path);
4898 static int check_csums(struct btrfs_root *root)
4900 struct btrfs_path *path;
4901 struct extent_buffer *leaf;
4902 struct btrfs_key key;
4903 u64 offset = 0, num_bytes = 0;
4904 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
4908 unsigned long leaf_offset;
4910 root = root->fs_info->csum_root;
4911 if (!extent_buffer_uptodate(root->node)) {
4912 fprintf(stderr, "No valid csum tree found\n");
4916 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
4917 key.type = BTRFS_EXTENT_CSUM_KEY;
4920 path = btrfs_alloc_path();
4924 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4926 fprintf(stderr, "Error searching csum tree %d\n", ret);
4927 btrfs_free_path(path);
4931 if (ret > 0 && path->slots[0])
4936 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4937 ret = btrfs_next_leaf(root, path);
4939 fprintf(stderr, "Error going to next leaf "
4946 leaf = path->nodes[0];
4948 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4949 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
4954 data_len = (btrfs_item_size_nr(leaf, path->slots[0]) /
4955 csum_size) * root->sectorsize;
4956 if (!check_data_csum)
4957 goto skip_csum_check;
4958 leaf_offset = btrfs_item_ptr_offset(leaf, path->slots[0]);
4959 ret = check_extent_csums(root, key.offset, data_len,
4965 offset = key.offset;
4966 } else if (key.offset != offset + num_bytes) {
4967 ret = check_extent_exists(root, offset, num_bytes);
4969 fprintf(stderr, "Csum exists for %Lu-%Lu but "
4970 "there is no extent record\n",
4971 offset, offset+num_bytes);
4974 offset = key.offset;
4977 num_bytes += data_len;
4981 btrfs_free_path(path);
4985 static int is_dropped_key(struct btrfs_key *key,
4986 struct btrfs_key *drop_key) {
4987 if (key->objectid < drop_key->objectid)
4989 else if (key->objectid == drop_key->objectid) {
4990 if (key->type < drop_key->type)
4992 else if (key->type == drop_key->type) {
4993 if (key->offset < drop_key->offset)
5000 static int run_next_block(struct btrfs_trans_handle *trans,
5001 struct btrfs_root *root,
5002 struct block_info *bits,
5005 struct cache_tree *pending,
5006 struct cache_tree *seen,
5007 struct cache_tree *reada,
5008 struct cache_tree *nodes,
5009 struct cache_tree *extent_cache,
5010 struct cache_tree *chunk_cache,
5011 struct rb_root *dev_cache,
5012 struct block_group_tree *block_group_cache,
5013 struct device_extent_tree *dev_extent_cache,
5014 struct btrfs_root_item *ri)
5016 struct extent_buffer *buf;
5027 struct btrfs_key key;
5028 struct cache_extent *cache;
5031 nritems = pick_next_pending(pending, reada, nodes, *last, bits,
5032 bits_nr, &reada_bits);
5037 for(i = 0; i < nritems; i++) {
5038 ret = add_cache_extent(reada, bits[i].start,
5043 /* fixme, get the parent transid */
5044 readahead_tree_block(root, bits[i].start,
5048 *last = bits[0].start;
5049 bytenr = bits[0].start;
5050 size = bits[0].size;
5052 cache = lookup_cache_extent(pending, bytenr, size);
5054 remove_cache_extent(pending, cache);
5057 cache = lookup_cache_extent(reada, bytenr, size);
5059 remove_cache_extent(reada, cache);
5062 cache = lookup_cache_extent(nodes, bytenr, size);
5064 remove_cache_extent(nodes, cache);
5067 cache = lookup_cache_extent(extent_cache, bytenr, size);
5069 struct extent_record *rec;
5071 rec = container_of(cache, struct extent_record, cache);
5072 gen = rec->parent_generation;
5075 /* fixme, get the real parent transid */
5076 buf = read_tree_block(root, bytenr, size, gen);
5077 if (!extent_buffer_uptodate(buf)) {
5078 record_bad_block_io(root->fs_info,
5079 extent_cache, bytenr, size);
5083 nritems = btrfs_header_nritems(buf);
5086 * FIXME, this only works only if we don't have any full
5089 if (!init_extent_tree) {
5090 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
5091 btrfs_header_level(buf), 1, NULL,
5099 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5104 owner = btrfs_header_owner(buf);
5107 ret = check_block(trans, root, extent_cache, buf, flags);
5111 if (btrfs_is_leaf(buf)) {
5112 btree_space_waste += btrfs_leaf_free_space(root, buf);
5113 for (i = 0; i < nritems; i++) {
5114 struct btrfs_file_extent_item *fi;
5115 btrfs_item_key_to_cpu(buf, &key, i);
5116 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
5117 process_extent_item(root, extent_cache, buf,
5121 if (key.type == BTRFS_METADATA_ITEM_KEY) {
5122 process_extent_item(root, extent_cache, buf,
5126 if (key.type == BTRFS_EXTENT_CSUM_KEY) {
5128 btrfs_item_size_nr(buf, i);
5131 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5132 process_chunk_item(chunk_cache, &key, buf, i);
5135 if (key.type == BTRFS_DEV_ITEM_KEY) {
5136 process_device_item(dev_cache, &key, buf, i);
5139 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5140 process_block_group_item(block_group_cache,
5144 if (key.type == BTRFS_DEV_EXTENT_KEY) {
5145 process_device_extent_item(dev_extent_cache,
5150 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
5151 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5152 process_extent_ref_v0(extent_cache, buf, i);
5159 if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
5160 add_tree_backref(extent_cache, key.objectid, 0,
5164 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
5165 add_tree_backref(extent_cache, key.objectid,
5169 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
5170 struct btrfs_extent_data_ref *ref;
5171 ref = btrfs_item_ptr(buf, i,
5172 struct btrfs_extent_data_ref);
5173 add_data_backref(extent_cache,
5175 btrfs_extent_data_ref_root(buf, ref),
5176 btrfs_extent_data_ref_objectid(buf,
5178 btrfs_extent_data_ref_offset(buf, ref),
5179 btrfs_extent_data_ref_count(buf, ref),
5180 0, root->sectorsize);
5183 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
5184 struct btrfs_shared_data_ref *ref;
5185 ref = btrfs_item_ptr(buf, i,
5186 struct btrfs_shared_data_ref);
5187 add_data_backref(extent_cache,
5188 key.objectid, key.offset, 0, 0, 0,
5189 btrfs_shared_data_ref_count(buf, ref),
5190 0, root->sectorsize);
5193 if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
5194 struct bad_item *bad;
5196 if (key.objectid == BTRFS_ORPHAN_OBJECTID)
5200 bad = malloc(sizeof(struct bad_item));
5203 INIT_LIST_HEAD(&bad->list);
5204 memcpy(&bad->key, &key,
5205 sizeof(struct btrfs_key));
5206 bad->root_id = owner;
5207 list_add_tail(&bad->list, &delete_items);
5210 if (key.type != BTRFS_EXTENT_DATA_KEY)
5212 fi = btrfs_item_ptr(buf, i,
5213 struct btrfs_file_extent_item);
5214 if (btrfs_file_extent_type(buf, fi) ==
5215 BTRFS_FILE_EXTENT_INLINE)
5217 if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
5220 data_bytes_allocated +=
5221 btrfs_file_extent_disk_num_bytes(buf, fi);
5222 if (data_bytes_allocated < root->sectorsize) {
5225 data_bytes_referenced +=
5226 btrfs_file_extent_num_bytes(buf, fi);
5227 add_data_backref(extent_cache,
5228 btrfs_file_extent_disk_bytenr(buf, fi),
5229 parent, owner, key.objectid, key.offset -
5230 btrfs_file_extent_offset(buf, fi), 1, 1,
5231 btrfs_file_extent_disk_num_bytes(buf, fi));
5235 struct btrfs_key first_key;
5237 first_key.objectid = 0;
5240 btrfs_item_key_to_cpu(buf, &first_key, 0);
5241 level = btrfs_header_level(buf);
5242 for (i = 0; i < nritems; i++) {
5243 ptr = btrfs_node_blockptr(buf, i);
5244 size = btrfs_level_size(root, level - 1);
5245 btrfs_node_key_to_cpu(buf, &key, i);
5247 struct btrfs_key drop_key;
5248 btrfs_disk_key_to_cpu(&drop_key,
5249 &ri->drop_progress);
5250 if ((level == ri->drop_level)
5251 && is_dropped_key(&key, &drop_key)) {
5255 ret = add_extent_rec(extent_cache, &key,
5256 btrfs_node_ptr_generation(buf, i),
5257 ptr, size, 0, 0, 1, 0, 1, 0,
5261 add_tree_backref(extent_cache, ptr, parent, owner, 1);
5264 add_pending(nodes, seen, ptr, size);
5266 add_pending(pending, seen, ptr, size);
5269 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
5270 nritems) * sizeof(struct btrfs_key_ptr);
5272 total_btree_bytes += buf->len;
5273 if (fs_root_objectid(btrfs_header_owner(buf)))
5274 total_fs_tree_bytes += buf->len;
5275 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
5276 total_extent_tree_bytes += buf->len;
5277 if (!found_old_backref &&
5278 btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
5279 btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
5280 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
5281 found_old_backref = 1;
5283 free_extent_buffer(buf);
5287 static int add_root_to_pending(struct extent_buffer *buf,
5288 struct cache_tree *extent_cache,
5289 struct cache_tree *pending,
5290 struct cache_tree *seen,
5291 struct cache_tree *nodes,
5292 struct btrfs_key *root_key)
5294 if (btrfs_header_level(buf) > 0)
5295 add_pending(nodes, seen, buf->start, buf->len);
5297 add_pending(pending, seen, buf->start, buf->len);
5298 add_extent_rec(extent_cache, NULL, 0, buf->start, buf->len,
5299 0, 1, 1, 0, 1, 0, buf->len);
5301 if (root_key->objectid == BTRFS_TREE_RELOC_OBJECTID ||
5302 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
5303 add_tree_backref(extent_cache, buf->start, buf->start,
5306 add_tree_backref(extent_cache, buf->start, 0,
5307 root_key->objectid, 1);
5311 /* as we fix the tree, we might be deleting blocks that
5312 * we're tracking for repair. This hook makes sure we
5313 * remove any backrefs for blocks as we are fixing them.
5315 static int free_extent_hook(struct btrfs_trans_handle *trans,
5316 struct btrfs_root *root,
5317 u64 bytenr, u64 num_bytes, u64 parent,
5318 u64 root_objectid, u64 owner, u64 offset,
5321 struct extent_record *rec;
5322 struct cache_extent *cache;
5324 struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
5326 is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
5327 cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
5331 rec = container_of(cache, struct extent_record, cache);
5333 struct data_backref *back;
5334 back = find_data_backref(rec, parent, root_objectid, owner,
5335 offset, 1, bytenr, num_bytes);
5338 if (back->node.found_ref) {
5339 back->found_ref -= refs_to_drop;
5341 rec->refs -= refs_to_drop;
5343 if (back->node.found_extent_tree) {
5344 back->num_refs -= refs_to_drop;
5345 if (rec->extent_item_refs)
5346 rec->extent_item_refs -= refs_to_drop;
5348 if (back->found_ref == 0)
5349 back->node.found_ref = 0;
5350 if (back->num_refs == 0)
5351 back->node.found_extent_tree = 0;
5353 if (!back->node.found_extent_tree && back->node.found_ref) {
5354 list_del(&back->node.list);
5358 struct tree_backref *back;
5359 back = find_tree_backref(rec, parent, root_objectid);
5362 if (back->node.found_ref) {
5365 back->node.found_ref = 0;
5367 if (back->node.found_extent_tree) {
5368 if (rec->extent_item_refs)
5369 rec->extent_item_refs--;
5370 back->node.found_extent_tree = 0;
5372 if (!back->node.found_extent_tree && back->node.found_ref) {
5373 list_del(&back->node.list);
5377 maybe_free_extent_rec(extent_cache, rec);
5382 static int delete_extent_records(struct btrfs_trans_handle *trans,
5383 struct btrfs_root *root,
5384 struct btrfs_path *path,
5385 u64 bytenr, u64 new_len)
5387 struct btrfs_key key;
5388 struct btrfs_key found_key;
5389 struct extent_buffer *leaf;
5394 key.objectid = bytenr;
5396 key.offset = (u64)-1;
5399 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
5406 if (path->slots[0] == 0)
5412 leaf = path->nodes[0];
5413 slot = path->slots[0];
5415 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5416 if (found_key.objectid != bytenr)
5419 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
5420 found_key.type != BTRFS_METADATA_ITEM_KEY &&
5421 found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
5422 found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
5423 found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
5424 found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
5425 found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
5426 btrfs_release_path(path);
5427 if (found_key.type == 0) {
5428 if (found_key.offset == 0)
5430 key.offset = found_key.offset - 1;
5431 key.type = found_key.type;
5433 key.type = found_key.type - 1;
5434 key.offset = (u64)-1;
5438 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
5439 found_key.objectid, found_key.type, found_key.offset);
5441 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
5444 btrfs_release_path(path);
5446 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5447 found_key.type == BTRFS_METADATA_ITEM_KEY) {
5448 u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
5449 found_key.offset : root->leafsize;
5451 ret = btrfs_update_block_group(trans, root, bytenr,
5458 btrfs_release_path(path);
5463 * for a single backref, this will allocate a new extent
5464 * and add the backref to it.
5466 static int record_extent(struct btrfs_trans_handle *trans,
5467 struct btrfs_fs_info *info,
5468 struct btrfs_path *path,
5469 struct extent_record *rec,
5470 struct extent_backref *back,
5471 int allocated, u64 flags)
5474 struct btrfs_root *extent_root = info->extent_root;
5475 struct extent_buffer *leaf;
5476 struct btrfs_key ins_key;
5477 struct btrfs_extent_item *ei;
5478 struct tree_backref *tback;
5479 struct data_backref *dback;
5480 struct btrfs_tree_block_info *bi;
5483 rec->max_size = max_t(u64, rec->max_size,
5484 info->extent_root->leafsize);
5487 u32 item_size = sizeof(*ei);
5490 item_size += sizeof(*bi);
5492 ins_key.objectid = rec->start;
5493 ins_key.offset = rec->max_size;
5494 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
5496 ret = btrfs_insert_empty_item(trans, extent_root, path,
5497 &ins_key, item_size);
5501 leaf = path->nodes[0];
5502 ei = btrfs_item_ptr(leaf, path->slots[0],
5503 struct btrfs_extent_item);
5505 btrfs_set_extent_refs(leaf, ei, 0);
5506 btrfs_set_extent_generation(leaf, ei, rec->generation);
5508 if (back->is_data) {
5509 btrfs_set_extent_flags(leaf, ei,
5510 BTRFS_EXTENT_FLAG_DATA);
5512 struct btrfs_disk_key copy_key;;
5514 tback = (struct tree_backref *)back;
5515 bi = (struct btrfs_tree_block_info *)(ei + 1);
5516 memset_extent_buffer(leaf, 0, (unsigned long)bi,
5519 btrfs_set_disk_key_objectid(©_key,
5520 rec->info_objectid);
5521 btrfs_set_disk_key_type(©_key, 0);
5522 btrfs_set_disk_key_offset(©_key, 0);
5524 btrfs_set_tree_block_level(leaf, bi, rec->info_level);
5525 btrfs_set_tree_block_key(leaf, bi, ©_key);
5527 btrfs_set_extent_flags(leaf, ei,
5528 BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
5531 btrfs_mark_buffer_dirty(leaf);
5532 ret = btrfs_update_block_group(trans, extent_root, rec->start,
5533 rec->max_size, 1, 0);
5536 btrfs_release_path(path);
5539 if (back->is_data) {
5543 dback = (struct data_backref *)back;
5544 if (back->full_backref)
5545 parent = dback->parent;
5549 for (i = 0; i < dback->found_ref; i++) {
5550 /* if parent != 0, we're doing a full backref
5551 * passing BTRFS_FIRST_FREE_OBJECTID as the owner
5552 * just makes the backref allocator create a data
5555 ret = btrfs_inc_extent_ref(trans, info->extent_root,
5556 rec->start, rec->max_size,
5560 BTRFS_FIRST_FREE_OBJECTID :
5566 fprintf(stderr, "adding new data backref"
5567 " on %llu %s %llu owner %llu"
5568 " offset %llu found %d\n",
5569 (unsigned long long)rec->start,
5570 back->full_backref ?
5572 back->full_backref ?
5573 (unsigned long long)parent :
5574 (unsigned long long)dback->root,
5575 (unsigned long long)dback->owner,
5576 (unsigned long long)dback->offset,
5581 tback = (struct tree_backref *)back;
5582 if (back->full_backref)
5583 parent = tback->parent;
5587 ret = btrfs_inc_extent_ref(trans, info->extent_root,
5588 rec->start, rec->max_size,
5589 parent, tback->root, 0, 0);
5590 fprintf(stderr, "adding new tree backref on "
5591 "start %llu len %llu parent %llu root %llu\n",
5592 rec->start, rec->max_size, tback->parent, tback->root);
5597 btrfs_release_path(path);
5601 struct extent_entry {
5606 struct list_head list;
5609 static struct extent_entry *find_entry(struct list_head *entries,
5610 u64 bytenr, u64 bytes)
5612 struct extent_entry *entry = NULL;
5614 list_for_each_entry(entry, entries, list) {
5615 if (entry->bytenr == bytenr && entry->bytes == bytes)
5622 static struct extent_entry *find_most_right_entry(struct list_head *entries)
5624 struct extent_entry *entry, *best = NULL, *prev = NULL;
5626 list_for_each_entry(entry, entries, list) {
5633 * If there are as many broken entries as entries then we know
5634 * not to trust this particular entry.
5636 if (entry->broken == entry->count)
5640 * If our current entry == best then we can't be sure our best
5641 * is really the best, so we need to keep searching.
5643 if (best && best->count == entry->count) {
5649 /* Prev == entry, not good enough, have to keep searching */
5650 if (!prev->broken && prev->count == entry->count)
5654 best = (prev->count > entry->count) ? prev : entry;
5655 else if (best->count < entry->count)
5663 static int repair_ref(struct btrfs_trans_handle *trans,
5664 struct btrfs_fs_info *info, struct btrfs_path *path,
5665 struct data_backref *dback, struct extent_entry *entry)
5667 struct btrfs_root *root;
5668 struct btrfs_file_extent_item *fi;
5669 struct extent_buffer *leaf;
5670 struct btrfs_key key;
5674 key.objectid = dback->root;
5675 key.type = BTRFS_ROOT_ITEM_KEY;
5676 key.offset = (u64)-1;
5677 root = btrfs_read_fs_root(info, &key);
5679 fprintf(stderr, "Couldn't find root for our ref\n");
5684 * The backref points to the original offset of the extent if it was
5685 * split, so we need to search down to the offset we have and then walk
5686 * forward until we find the backref we're looking for.
5688 key.objectid = dback->owner;
5689 key.type = BTRFS_EXTENT_DATA_KEY;
5690 key.offset = dback->offset;
5691 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5693 fprintf(stderr, "Error looking up ref %d\n", ret);
5698 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5699 ret = btrfs_next_leaf(root, path);
5701 fprintf(stderr, "Couldn't find our ref, next\n");
5705 leaf = path->nodes[0];
5706 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5707 if (key.objectid != dback->owner ||
5708 key.type != BTRFS_EXTENT_DATA_KEY) {
5709 fprintf(stderr, "Couldn't find our ref, search\n");
5712 fi = btrfs_item_ptr(leaf, path->slots[0],
5713 struct btrfs_file_extent_item);
5714 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5715 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
5717 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
5722 btrfs_release_path(path);
5725 * Have to make sure that this root gets updated when we commit the
5728 record_root_in_trans(trans, root);
5731 * Ok we have the key of the file extent we want to fix, now we can cow
5732 * down to the thing and fix it.
5734 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
5736 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
5737 key.objectid, key.type, key.offset, ret);
5741 fprintf(stderr, "Well that's odd, we just found this key "
5742 "[%Lu, %u, %Lu]\n", key.objectid, key.type,
5746 leaf = path->nodes[0];
5747 fi = btrfs_item_ptr(leaf, path->slots[0],
5748 struct btrfs_file_extent_item);
5750 if (btrfs_file_extent_compression(leaf, fi) &&
5751 dback->disk_bytenr != entry->bytenr) {
5752 fprintf(stderr, "Ref doesn't match the record start and is "
5753 "compressed, please take a btrfs-image of this file "
5754 "system and send it to a btrfs developer so they can "
5755 "complete this functionality for bytenr %Lu\n",
5756 dback->disk_bytenr);
5760 if (dback->node.broken && dback->disk_bytenr != entry->bytenr) {
5761 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
5762 } else if (dback->disk_bytenr > entry->bytenr) {
5763 u64 off_diff, offset;
5765 off_diff = dback->disk_bytenr - entry->bytenr;
5766 offset = btrfs_file_extent_offset(leaf, fi);
5767 if (dback->disk_bytenr + offset +
5768 btrfs_file_extent_num_bytes(leaf, fi) >
5769 entry->bytenr + entry->bytes) {
5770 fprintf(stderr, "Ref is past the entry end, please "
5771 "take a btrfs-image of this file system and "
5772 "send it to a btrfs developer, ref %Lu\n",
5773 dback->disk_bytenr);
5777 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
5778 btrfs_set_file_extent_offset(leaf, fi, offset);
5779 } else if (dback->disk_bytenr < entry->bytenr) {
5782 offset = btrfs_file_extent_offset(leaf, fi);
5783 if (dback->disk_bytenr + offset < entry->bytenr) {
5784 fprintf(stderr, "Ref is before the entry start, please"
5785 " take a btrfs-image of this file system and "
5786 "send it to a btrfs developer, ref %Lu\n",
5787 dback->disk_bytenr);
5791 offset += dback->disk_bytenr;
5792 offset -= entry->bytenr;
5793 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
5794 btrfs_set_file_extent_offset(leaf, fi, offset);
5797 btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
5800 * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
5801 * only do this if we aren't using compression, otherwise it's a
5804 if (!btrfs_file_extent_compression(leaf, fi))
5805 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
5807 printf("ram bytes may be wrong?\n");
5808 btrfs_mark_buffer_dirty(leaf);
5809 btrfs_release_path(path);
5813 static int verify_backrefs(struct btrfs_trans_handle *trans,
5814 struct btrfs_fs_info *info, struct btrfs_path *path,
5815 struct extent_record *rec)
5817 struct extent_backref *back;
5818 struct data_backref *dback;
5819 struct extent_entry *entry, *best = NULL;
5822 int broken_entries = 0;
5827 * Metadata is easy and the backrefs should always agree on bytenr and
5828 * size, if not we've got bigger issues.
5833 list_for_each_entry(back, &rec->backrefs, list) {
5834 if (back->full_backref || !back->is_data)
5837 dback = (struct data_backref *)back;
5840 * We only pay attention to backrefs that we found a real
5843 if (dback->found_ref == 0)
5847 * For now we only catch when the bytes don't match, not the
5848 * bytenr. We can easily do this at the same time, but I want
5849 * to have a fs image to test on before we just add repair
5850 * functionality willy-nilly so we know we won't screw up the
5854 entry = find_entry(&entries, dback->disk_bytenr,
5857 entry = malloc(sizeof(struct extent_entry));
5862 memset(entry, 0, sizeof(*entry));
5863 entry->bytenr = dback->disk_bytenr;
5864 entry->bytes = dback->bytes;
5865 list_add_tail(&entry->list, &entries);
5870 * If we only have on entry we may think the entries agree when
5871 * in reality they don't so we have to do some extra checking.
5873 if (dback->disk_bytenr != rec->start ||
5874 dback->bytes != rec->nr || back->broken)
5885 /* Yay all the backrefs agree, carry on good sir */
5886 if (nr_entries <= 1 && !mismatch)
5889 fprintf(stderr, "attempting to repair backref discrepency for bytenr "
5890 "%Lu\n", rec->start);
5893 * First we want to see if the backrefs can agree amongst themselves who
5894 * is right, so figure out which one of the entries has the highest
5897 best = find_most_right_entry(&entries);
5900 * Ok so we may have an even split between what the backrefs think, so
5901 * this is where we use the extent ref to see what it thinks.
5904 entry = find_entry(&entries, rec->start, rec->nr);
5905 if (!entry && (!broken_entries || !rec->found_rec)) {
5906 fprintf(stderr, "Backrefs don't agree with each other "
5907 "and extent record doesn't agree with anybody,"
5908 " so we can't fix bytenr %Lu bytes %Lu\n",
5909 rec->start, rec->nr);
5912 } else if (!entry) {
5914 * Ok our backrefs were broken, we'll assume this is the
5915 * correct value and add an entry for this range.
5917 entry = malloc(sizeof(struct extent_entry));
5922 memset(entry, 0, sizeof(*entry));
5923 entry->bytenr = rec->start;
5924 entry->bytes = rec->nr;
5925 list_add_tail(&entry->list, &entries);
5929 best = find_most_right_entry(&entries);
5931 fprintf(stderr, "Backrefs and extent record evenly "
5932 "split on who is right, this is going to "
5933 "require user input to fix bytenr %Lu bytes "
5934 "%Lu\n", rec->start, rec->nr);
5941 * I don't think this can happen currently as we'll abort() if we catch
5942 * this case higher up, but in case somebody removes that we still can't
5943 * deal with it properly here yet, so just bail out of that's the case.
5945 if (best->bytenr != rec->start) {
5946 fprintf(stderr, "Extent start and backref starts don't match, "
5947 "please use btrfs-image on this file system and send "
5948 "it to a btrfs developer so they can make fsck fix "
5949 "this particular case. bytenr is %Lu, bytes is %Lu\n",
5950 rec->start, rec->nr);
5956 * Ok great we all agreed on an extent record, let's go find the real
5957 * references and fix up the ones that don't match.
5959 list_for_each_entry(back, &rec->backrefs, list) {
5960 if (back->full_backref || !back->is_data)
5963 dback = (struct data_backref *)back;
5966 * Still ignoring backrefs that don't have a real ref attached
5969 if (dback->found_ref == 0)
5972 if (dback->bytes == best->bytes &&
5973 dback->disk_bytenr == best->bytenr)
5976 ret = repair_ref(trans, info, path, dback, best);
5982 * Ok we messed with the actual refs, which means we need to drop our
5983 * entire cache and go back and rescan. I know this is a huge pain and
5984 * adds a lot of extra work, but it's the only way to be safe. Once all
5985 * the backrefs agree we may not need to do anything to the extent
5990 while (!list_empty(&entries)) {
5991 entry = list_entry(entries.next, struct extent_entry, list);
5992 list_del_init(&entry->list);
5998 static int process_duplicates(struct btrfs_root *root,
5999 struct cache_tree *extent_cache,
6000 struct extent_record *rec)
6002 struct extent_record *good, *tmp;
6003 struct cache_extent *cache;
6007 * If we found a extent record for this extent then return, or if we
6008 * have more than one duplicate we are likely going to need to delete
6011 if (rec->found_rec || rec->num_duplicates > 1)
6014 /* Shouldn't happen but just in case */
6015 BUG_ON(!rec->num_duplicates);
6018 * So this happens if we end up with a backref that doesn't match the
6019 * actual extent entry. So either the backref is bad or the extent
6020 * entry is bad. Either way we want to have the extent_record actually
6021 * reflect what we found in the extent_tree, so we need to take the
6022 * duplicate out and use that as the extent_record since the only way we
6023 * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
6025 remove_cache_extent(extent_cache, &rec->cache);
6027 good = list_entry(rec->dups.next, struct extent_record, list);
6028 list_del_init(&good->list);
6029 INIT_LIST_HEAD(&good->backrefs);
6030 INIT_LIST_HEAD(&good->dups);
6031 good->cache.start = good->start;
6032 good->cache.size = good->nr;
6033 good->content_checked = 0;
6034 good->owner_ref_checked = 0;
6035 good->num_duplicates = 0;
6036 good->refs = rec->refs;
6037 list_splice_init(&rec->backrefs, &good->backrefs);
6039 cache = lookup_cache_extent(extent_cache, good->start,
6043 tmp = container_of(cache, struct extent_record, cache);
6046 * If we find another overlapping extent and it's found_rec is
6047 * set then it's a duplicate and we need to try and delete
6050 if (tmp->found_rec || tmp->num_duplicates > 0) {
6051 if (list_empty(&good->list))
6052 list_add_tail(&good->list,
6053 &duplicate_extents);
6054 good->num_duplicates += tmp->num_duplicates + 1;
6055 list_splice_init(&tmp->dups, &good->dups);
6056 list_del_init(&tmp->list);
6057 list_add_tail(&tmp->list, &good->dups);
6058 remove_cache_extent(extent_cache, &tmp->cache);
6063 * Ok we have another non extent item backed extent rec, so lets
6064 * just add it to this extent and carry on like we did above.
6066 good->refs += tmp->refs;
6067 list_splice_init(&tmp->backrefs, &good->backrefs);
6068 remove_cache_extent(extent_cache, &tmp->cache);
6071 ret = insert_cache_extent(extent_cache, &good->cache);
6074 return good->num_duplicates ? 0 : 1;
6077 static int delete_duplicate_records(struct btrfs_trans_handle *trans,
6078 struct btrfs_root *root,
6079 struct extent_record *rec)
6081 LIST_HEAD(delete_list);
6082 struct btrfs_path *path;
6083 struct extent_record *tmp, *good, *n;
6086 struct btrfs_key key;
6088 path = btrfs_alloc_path();
6095 /* Find the record that covers all of the duplicates. */
6096 list_for_each_entry(tmp, &rec->dups, list) {
6097 if (good->start < tmp->start)
6099 if (good->nr > tmp->nr)
6102 if (tmp->start + tmp->nr < good->start + good->nr) {
6103 fprintf(stderr, "Ok we have overlapping extents that "
6104 "aren't completely covered by eachother, this "
6105 "is going to require more careful thought. "
6106 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
6107 tmp->start, tmp->nr, good->start, good->nr);
6114 list_add_tail(&rec->list, &delete_list);
6116 list_for_each_entry_safe(tmp, n, &rec->dups, list) {
6119 list_move_tail(&tmp->list, &delete_list);
6122 root = root->fs_info->extent_root;
6123 list_for_each_entry(tmp, &delete_list, list) {
6124 if (tmp->found_rec == 0)
6126 key.objectid = tmp->start;
6127 key.type = BTRFS_EXTENT_ITEM_KEY;
6128 key.offset = tmp->nr;
6130 /* Shouldn't happen but just in case */
6131 if (tmp->metadata) {
6132 fprintf(stderr, "Well this shouldn't happen, extent "
6133 "record overlaps but is metadata? "
6134 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
6138 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6144 ret = btrfs_del_item(trans, root, path);
6147 btrfs_release_path(path);
6152 while (!list_empty(&delete_list)) {
6153 tmp = list_entry(delete_list.next, struct extent_record, list);
6154 list_del_init(&tmp->list);
6160 while (!list_empty(&rec->dups)) {
6161 tmp = list_entry(rec->dups.next, struct extent_record, list);
6162 list_del_init(&tmp->list);
6166 btrfs_free_path(path);
6168 if (!ret && !nr_del)
6169 rec->num_duplicates = 0;
6171 return ret ? ret : nr_del;
6174 static int find_possible_backrefs(struct btrfs_trans_handle *trans,
6175 struct btrfs_fs_info *info,
6176 struct btrfs_path *path,
6177 struct cache_tree *extent_cache,
6178 struct extent_record *rec)
6180 struct btrfs_root *root;
6181 struct extent_backref *back;
6182 struct data_backref *dback;
6183 struct cache_extent *cache;
6184 struct btrfs_file_extent_item *fi;
6185 struct btrfs_key key;
6189 list_for_each_entry(back, &rec->backrefs, list) {
6190 /* Don't care about full backrefs (poor unloved backrefs) */
6191 if (back->full_backref || !back->is_data)
6194 dback = (struct data_backref *)back;
6196 /* We found this one, we don't need to do a lookup */
6197 if (dback->found_ref)
6200 key.objectid = dback->root;
6201 key.type = BTRFS_ROOT_ITEM_KEY;
6202 key.offset = (u64)-1;
6204 root = btrfs_read_fs_root(info, &key);
6206 /* No root, definitely a bad ref, skip */
6207 if (IS_ERR(root) && PTR_ERR(root) == -ENOENT)
6209 /* Other err, exit */
6211 return PTR_ERR(root);
6213 key.objectid = dback->owner;
6214 key.type = BTRFS_EXTENT_DATA_KEY;
6215 key.offset = dback->offset;
6216 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6218 btrfs_release_path(path);
6221 /* Didn't find it, we can carry on */
6226 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6227 struct btrfs_file_extent_item);
6228 bytenr = btrfs_file_extent_disk_bytenr(path->nodes[0], fi);
6229 bytes = btrfs_file_extent_disk_num_bytes(path->nodes[0], fi);
6230 btrfs_release_path(path);
6231 cache = lookup_cache_extent(extent_cache, bytenr, 1);
6233 struct extent_record *tmp;
6234 tmp = container_of(cache, struct extent_record, cache);
6237 * If we found an extent record for the bytenr for this
6238 * particular backref then we can't add it to our
6239 * current extent record. We only want to add backrefs
6240 * that don't have a corresponding extent item in the
6241 * extent tree since they likely belong to this record
6242 * and we need to fix it if it doesn't match bytenrs.
6248 dback->found_ref += 1;
6249 dback->disk_bytenr = bytenr;
6250 dback->bytes = bytes;
6253 * Set this so the verify backref code knows not to trust the
6254 * values in this backref.
6263 * when an incorrect extent item is found, this will delete
6264 * all of the existing entries for it and recreate them
6265 * based on what the tree scan found.
6267 static int fixup_extent_refs(struct btrfs_trans_handle *trans,
6268 struct btrfs_fs_info *info,
6269 struct cache_tree *extent_cache,
6270 struct extent_record *rec)
6273 struct btrfs_path *path;
6274 struct list_head *cur = rec->backrefs.next;
6275 struct cache_extent *cache;
6276 struct extent_backref *back;
6281 * remember our flags for recreating the extent.
6282 * FIXME, if we have cleared extent tree, we can not
6283 * lookup extent info in extent tree.
6285 if (!init_extent_tree) {
6286 ret = btrfs_lookup_extent_info(NULL, info->extent_root,
6287 rec->start, rec->max_size,
6288 rec->metadata, NULL, &flags);
6295 path = btrfs_alloc_path();
6299 if (rec->refs != rec->extent_item_refs && !rec->metadata) {
6301 * Sometimes the backrefs themselves are so broken they don't
6302 * get attached to any meaningful rec, so first go back and
6303 * check any of our backrefs that we couldn't find and throw
6304 * them into the list if we find the backref so that
6305 * verify_backrefs can figure out what to do.
6307 ret = find_possible_backrefs(trans, info, path, extent_cache,
6313 /* step one, make sure all of the backrefs agree */
6314 ret = verify_backrefs(trans, info, path, rec);
6318 /* step two, delete all the existing records */
6319 ret = delete_extent_records(trans, info->extent_root, path,
6320 rec->start, rec->max_size);
6325 /* was this block corrupt? If so, don't add references to it */
6326 cache = lookup_cache_extent(info->corrupt_blocks,
6327 rec->start, rec->max_size);
6333 /* step three, recreate all the refs we did find */
6334 while(cur != &rec->backrefs) {
6335 back = list_entry(cur, struct extent_backref, list);
6339 * if we didn't find any references, don't create a
6342 if (!back->found_ref)
6345 ret = record_extent(trans, info, path, rec, back, allocated, flags);
6352 btrfs_free_path(path);
6356 /* right now we only prune from the extent allocation tree */
6357 static int prune_one_block(struct btrfs_trans_handle *trans,
6358 struct btrfs_fs_info *info,
6359 struct btrfs_corrupt_block *corrupt)
6362 struct btrfs_path path;
6363 struct extent_buffer *eb;
6367 int level = corrupt->level + 1;
6369 btrfs_init_path(&path);
6371 /* we want to stop at the parent to our busted block */
6372 path.lowest_level = level;
6374 ret = btrfs_search_slot(trans, info->extent_root,
6375 &corrupt->key, &path, -1, 1);
6380 eb = path.nodes[level];
6387 * hopefully the search gave us the block we want to prune,
6388 * lets try that first
6390 slot = path.slots[level];
6391 found = btrfs_node_blockptr(eb, slot);
6392 if (found == corrupt->cache.start)
6395 nritems = btrfs_header_nritems(eb);
6397 /* the search failed, lets scan this node and hope we find it */
6398 for (slot = 0; slot < nritems; slot++) {
6399 found = btrfs_node_blockptr(eb, slot);
6400 if (found == corrupt->cache.start)
6404 * we couldn't find the bad block. TODO, search all the nodes for pointers
6407 if (eb == info->extent_root->node) {
6412 btrfs_release_path(&path);
6417 printk("deleting pointer to block %Lu\n", corrupt->cache.start);
6418 ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
6421 btrfs_release_path(&path);
6425 static int prune_corrupt_blocks(struct btrfs_trans_handle *trans,
6426 struct btrfs_fs_info *info)
6428 struct cache_extent *cache;
6429 struct btrfs_corrupt_block *corrupt;
6431 cache = search_cache_extent(info->corrupt_blocks, 0);
6435 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
6436 prune_one_block(trans, info, corrupt);
6437 cache = next_cache_extent(cache);
6442 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
6444 struct btrfs_block_group_cache *cache;
6449 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
6450 &start, &end, EXTENT_DIRTY);
6453 clear_extent_dirty(&fs_info->free_space_cache, start, end,
6459 cache = btrfs_lookup_first_block_group(fs_info, start);
6464 start = cache->key.objectid + cache->key.offset;
6468 static int check_extent_refs(struct btrfs_trans_handle *trans,
6469 struct btrfs_root *root,
6470 struct cache_tree *extent_cache)
6472 struct extent_record *rec;
6473 struct cache_extent *cache;
6481 * if we're doing a repair, we have to make sure
6482 * we don't allocate from the problem extents.
6483 * In the worst case, this will be all the
6486 cache = search_cache_extent(extent_cache, 0);
6488 rec = container_of(cache, struct extent_record, cache);
6489 btrfs_pin_extent(root->fs_info,
6490 rec->start, rec->max_size);
6491 cache = next_cache_extent(cache);
6494 /* pin down all the corrupted blocks too */
6495 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
6497 btrfs_pin_extent(root->fs_info,
6498 cache->start, cache->size);
6499 cache = next_cache_extent(cache);
6501 prune_corrupt_blocks(trans, root->fs_info);
6502 reset_cached_block_groups(root->fs_info);
6506 * We need to delete any duplicate entries we find first otherwise we
6507 * could mess up the extent tree when we have backrefs that actually
6508 * belong to a different extent item and not the weird duplicate one.
6510 while (repair && !list_empty(&duplicate_extents)) {
6511 rec = list_entry(duplicate_extents.next, struct extent_record,
6513 list_del_init(&rec->list);
6515 /* Sometimes we can find a backref before we find an actual
6516 * extent, so we need to process it a little bit to see if there
6517 * truly are multiple EXTENT_ITEM_KEY's for the same range, or
6518 * if this is a backref screwup. If we need to delete stuff
6519 * process_duplicates() will return 0, otherwise it will return
6522 if (process_duplicates(root, extent_cache, rec))
6524 ret = delete_duplicate_records(trans, root, rec);
6528 * delete_duplicate_records will return the number of entries
6529 * deleted, so if it's greater than 0 then we know we actually
6530 * did something and we need to remove.
6541 cache = search_cache_extent(extent_cache, 0);
6544 rec = container_of(cache, struct extent_record, cache);
6545 if (rec->num_duplicates) {
6546 fprintf(stderr, "extent item %llu has multiple extent "
6547 "items\n", (unsigned long long)rec->start);
6551 if (rec->refs != rec->extent_item_refs) {
6552 fprintf(stderr, "ref mismatch on [%llu %llu] ",
6553 (unsigned long long)rec->start,
6554 (unsigned long long)rec->nr);
6555 fprintf(stderr, "extent item %llu, found %llu\n",
6556 (unsigned long long)rec->extent_item_refs,
6557 (unsigned long long)rec->refs);
6558 if (!fixed && repair) {
6559 ret = fixup_extent_refs(trans, root->fs_info,
6568 if (all_backpointers_checked(rec, 1)) {
6569 fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
6570 (unsigned long long)rec->start,
6571 (unsigned long long)rec->nr);
6573 if (!fixed && repair) {
6574 ret = fixup_extent_refs(trans, root->fs_info,
6583 if (!rec->owner_ref_checked) {
6584 fprintf(stderr, "owner ref check failed [%llu %llu]\n",
6585 (unsigned long long)rec->start,
6586 (unsigned long long)rec->nr);
6587 if (!fixed && repair) {
6588 ret = fixup_extent_refs(trans, root->fs_info,
6597 remove_cache_extent(extent_cache, cache);
6598 free_all_extent_backrefs(rec);
6603 if (ret && ret != -EAGAIN) {
6604 fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
6607 btrfs_fix_block_accounting(trans, root);
6610 fprintf(stderr, "repaired damaged extent references\n");
6616 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
6620 if (type & BTRFS_BLOCK_GROUP_RAID0) {
6621 stripe_size = length;
6622 stripe_size /= num_stripes;
6623 } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
6624 stripe_size = length * 2;
6625 stripe_size /= num_stripes;
6626 } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
6627 stripe_size = length;
6628 stripe_size /= (num_stripes - 1);
6629 } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
6630 stripe_size = length;
6631 stripe_size /= (num_stripes - 2);
6633 stripe_size = length;
6639 * Check the chunk with its block group/dev list ref:
6640 * Return 0 if all refs seems valid.
6641 * Return 1 if part of refs seems valid, need later check for rebuild ref
6642 * like missing block group and needs to search extent tree to rebuild them.
6643 * Return -1 if essential refs are missing and unable to rebuild.
6645 static int check_chunk_refs(struct chunk_record *chunk_rec,
6646 struct block_group_tree *block_group_cache,
6647 struct device_extent_tree *dev_extent_cache,
6650 struct cache_extent *block_group_item;
6651 struct block_group_record *block_group_rec;
6652 struct cache_extent *dev_extent_item;
6653 struct device_extent_record *dev_extent_rec;
6660 block_group_item = lookup_cache_extent(&block_group_cache->tree,
6663 if (block_group_item) {
6664 block_group_rec = container_of(block_group_item,
6665 struct block_group_record,
6667 if (chunk_rec->length != block_group_rec->offset ||
6668 chunk_rec->offset != block_group_rec->objectid ||
6669 chunk_rec->type_flags != block_group_rec->flags) {
6672 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
6673 chunk_rec->objectid,
6678 chunk_rec->type_flags,
6679 block_group_rec->objectid,
6680 block_group_rec->type,
6681 block_group_rec->offset,
6682 block_group_rec->offset,
6683 block_group_rec->objectid,
6684 block_group_rec->flags);
6687 list_del_init(&block_group_rec->list);
6688 chunk_rec->bg_rec = block_group_rec;
6693 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
6694 chunk_rec->objectid,
6699 chunk_rec->type_flags);
6703 length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
6704 chunk_rec->num_stripes);
6705 for (i = 0; i < chunk_rec->num_stripes; ++i) {
6706 devid = chunk_rec->stripes[i].devid;
6707 offset = chunk_rec->stripes[i].offset;
6708 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
6709 devid, offset, length);
6710 if (dev_extent_item) {
6711 dev_extent_rec = container_of(dev_extent_item,
6712 struct device_extent_record,
6714 if (dev_extent_rec->objectid != devid ||
6715 dev_extent_rec->offset != offset ||
6716 dev_extent_rec->chunk_offset != chunk_rec->offset ||
6717 dev_extent_rec->length != length) {
6720 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
6721 chunk_rec->objectid,
6724 chunk_rec->stripes[i].devid,
6725 chunk_rec->stripes[i].offset,
6726 dev_extent_rec->objectid,
6727 dev_extent_rec->offset,
6728 dev_extent_rec->length);
6731 list_move(&dev_extent_rec->chunk_list,
6732 &chunk_rec->dextents);
6737 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
6738 chunk_rec->objectid,
6741 chunk_rec->stripes[i].devid,
6742 chunk_rec->stripes[i].offset);
6749 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
6750 int check_chunks(struct cache_tree *chunk_cache,
6751 struct block_group_tree *block_group_cache,
6752 struct device_extent_tree *dev_extent_cache,
6753 struct list_head *good, struct list_head *bad,
6754 struct list_head *rebuild, int silent)
6756 struct cache_extent *chunk_item;
6757 struct chunk_record *chunk_rec;
6758 struct block_group_record *bg_rec;
6759 struct device_extent_record *dext_rec;
6763 chunk_item = first_cache_extent(chunk_cache);
6764 while (chunk_item) {
6765 chunk_rec = container_of(chunk_item, struct chunk_record,
6767 err = check_chunk_refs(chunk_rec, block_group_cache,
6768 dev_extent_cache, silent);
6771 if (err == 0 && good)
6772 list_add_tail(&chunk_rec->list, good);
6773 if (err > 0 && rebuild)
6774 list_add_tail(&chunk_rec->list, rebuild);
6776 list_add_tail(&chunk_rec->list, bad);
6777 chunk_item = next_cache_extent(chunk_item);
6780 list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
6783 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
6791 list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
6795 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
6806 static int check_device_used(struct device_record *dev_rec,
6807 struct device_extent_tree *dext_cache)
6809 struct cache_extent *cache;
6810 struct device_extent_record *dev_extent_rec;
6813 cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
6815 dev_extent_rec = container_of(cache,
6816 struct device_extent_record,
6818 if (dev_extent_rec->objectid != dev_rec->devid)
6821 list_del_init(&dev_extent_rec->device_list);
6822 total_byte += dev_extent_rec->length;
6823 cache = next_cache_extent(cache);
6826 if (total_byte != dev_rec->byte_used) {
6828 "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
6829 total_byte, dev_rec->byte_used, dev_rec->objectid,
6830 dev_rec->type, dev_rec->offset);
6837 /* check btrfs_dev_item -> btrfs_dev_extent */
6838 static int check_devices(struct rb_root *dev_cache,
6839 struct device_extent_tree *dev_extent_cache)
6841 struct rb_node *dev_node;
6842 struct device_record *dev_rec;
6843 struct device_extent_record *dext_rec;
6847 dev_node = rb_first(dev_cache);
6849 dev_rec = container_of(dev_node, struct device_record, node);
6850 err = check_device_used(dev_rec, dev_extent_cache);
6854 dev_node = rb_next(dev_node);
6856 list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
6859 "Device extent[%llu, %llu, %llu] didn't find its device.\n",
6860 dext_rec->objectid, dext_rec->offset, dext_rec->length);
6867 static int check_chunks_and_extents(struct btrfs_root *root)
6869 struct rb_root dev_cache;
6870 struct cache_tree chunk_cache;
6871 struct block_group_tree block_group_cache;
6872 struct device_extent_tree dev_extent_cache;
6873 struct cache_tree extent_cache;
6874 struct cache_tree seen;
6875 struct cache_tree pending;
6876 struct cache_tree reada;
6877 struct cache_tree nodes;
6878 struct cache_tree corrupt_blocks;
6879 struct btrfs_path path;
6880 struct btrfs_key key;
6881 struct btrfs_key found_key;
6884 struct block_info *bits;
6886 struct extent_buffer *leaf;
6887 struct btrfs_trans_handle *trans = NULL;
6889 struct btrfs_root_item ri;
6890 struct list_head dropping_trees;
6892 dev_cache = RB_ROOT;
6893 cache_tree_init(&chunk_cache);
6894 block_group_tree_init(&block_group_cache);
6895 device_extent_tree_init(&dev_extent_cache);
6897 cache_tree_init(&extent_cache);
6898 cache_tree_init(&seen);
6899 cache_tree_init(&pending);
6900 cache_tree_init(&nodes);
6901 cache_tree_init(&reada);
6902 cache_tree_init(&corrupt_blocks);
6903 INIT_LIST_HEAD(&dropping_trees);
6906 trans = btrfs_start_transaction(root, 1);
6907 if (IS_ERR(trans)) {
6908 fprintf(stderr, "Error starting transaction\n");
6909 return PTR_ERR(trans);
6911 root->fs_info->fsck_extent_cache = &extent_cache;
6912 root->fs_info->free_extent_hook = free_extent_hook;
6913 root->fs_info->corrupt_blocks = &corrupt_blocks;
6917 bits = malloc(bits_nr * sizeof(struct block_info));
6924 add_root_to_pending(root->fs_info->tree_root->node,
6925 &extent_cache, &pending, &seen, &nodes,
6926 &root->fs_info->tree_root->root_key);
6928 add_root_to_pending(root->fs_info->chunk_root->node,
6929 &extent_cache, &pending, &seen, &nodes,
6930 &root->fs_info->chunk_root->root_key);
6932 btrfs_init_path(&path);
6935 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
6936 ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
6941 leaf = path.nodes[0];
6942 slot = path.slots[0];
6943 if (slot >= btrfs_header_nritems(path.nodes[0])) {
6944 ret = btrfs_next_leaf(root, &path);
6947 leaf = path.nodes[0];
6948 slot = path.slots[0];
6950 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
6951 if (btrfs_key_type(&found_key) == BTRFS_ROOT_ITEM_KEY) {
6952 unsigned long offset;
6953 struct extent_buffer *buf;
6955 offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
6956 read_extent_buffer(leaf, &ri, offset, sizeof(ri));
6957 if (btrfs_disk_key_objectid(&ri.drop_progress) == 0) {
6958 buf = read_tree_block(root->fs_info->tree_root,
6959 btrfs_root_bytenr(&ri),
6960 btrfs_level_size(root,
6961 btrfs_root_level(&ri)),
6967 add_root_to_pending(buf, &extent_cache,
6968 &pending, &seen, &nodes,
6970 free_extent_buffer(buf);
6972 struct dropping_root_item_record *dri_rec;
6973 dri_rec = malloc(sizeof(*dri_rec));
6978 memcpy(&dri_rec->ri, &ri, sizeof(ri));
6979 memcpy(&dri_rec->found_key, &found_key,
6981 list_add_tail(&dri_rec->list, &dropping_trees);
6986 btrfs_release_path(&path);
6988 ret = run_next_block(trans, root, bits, bits_nr, &last,
6989 &pending, &seen, &reada, &nodes,
6990 &extent_cache, &chunk_cache, &dev_cache,
6991 &block_group_cache, &dev_extent_cache,
6997 while (!list_empty(&dropping_trees)) {
6998 struct dropping_root_item_record *rec;
6999 struct extent_buffer *buf;
7000 rec = list_entry(dropping_trees.next,
7001 struct dropping_root_item_record, list);
7007 buf = read_tree_block(root->fs_info->tree_root,
7008 btrfs_root_bytenr(&rec->ri),
7009 btrfs_level_size(root,
7010 btrfs_root_level(&rec->ri)), 0);
7015 add_root_to_pending(buf, &extent_cache, &pending,
7016 &seen, &nodes, &rec->found_key);
7018 ret = run_next_block(trans, root, bits, bits_nr, &last,
7019 &pending, &seen, &reada,
7020 &nodes, &extent_cache,
7021 &chunk_cache, &dev_cache,
7028 free_extent_buffer(buf);
7029 list_del(&rec->list);
7034 ret = check_extent_refs(trans, root, &extent_cache);
7035 if (ret == -EAGAIN) {
7036 ret = btrfs_commit_transaction(trans, root);
7040 trans = btrfs_start_transaction(root, 1);
7041 if (IS_ERR(trans)) {
7042 ret = PTR_ERR(trans);
7046 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
7047 free_extent_cache_tree(&seen);
7048 free_extent_cache_tree(&pending);
7049 free_extent_cache_tree(&reada);
7050 free_extent_cache_tree(&nodes);
7051 free_chunk_cache_tree(&chunk_cache);
7052 free_block_group_tree(&block_group_cache);
7053 free_device_cache_tree(&dev_cache);
7054 free_device_extent_tree(&dev_extent_cache);
7055 free_extent_record_cache(root->fs_info, &extent_cache);
7059 err = check_chunks(&chunk_cache, &block_group_cache,
7060 &dev_extent_cache, NULL, NULL, NULL, 0);
7064 err = check_devices(&dev_cache, &dev_extent_cache);
7070 err = btrfs_commit_transaction(trans, root);
7075 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
7076 root->fs_info->fsck_extent_cache = NULL;
7077 root->fs_info->free_extent_hook = NULL;
7078 root->fs_info->corrupt_blocks = NULL;
7081 free_chunk_cache_tree(&chunk_cache);
7082 free_device_cache_tree(&dev_cache);
7083 free_block_group_tree(&block_group_cache);
7084 free_device_extent_tree(&dev_extent_cache);
7085 free_extent_cache_tree(&seen);
7086 free_extent_cache_tree(&pending);
7087 free_extent_cache_tree(&reada);
7088 free_extent_cache_tree(&nodes);
7092 static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
7093 struct btrfs_root *root, int overwrite)
7095 struct extent_buffer *c;
7096 struct extent_buffer *old = root->node;
7099 struct btrfs_disk_key disk_key = {0,0,0};
7105 extent_buffer_get(c);
7108 c = btrfs_alloc_free_block(trans, root,
7109 btrfs_level_size(root, 0),
7110 root->root_key.objectid,
7111 &disk_key, level, 0, 0);
7114 extent_buffer_get(c);
7118 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
7119 btrfs_set_header_level(c, level);
7120 btrfs_set_header_bytenr(c, c->start);
7121 btrfs_set_header_generation(c, trans->transid);
7122 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
7123 btrfs_set_header_owner(c, root->root_key.objectid);
7125 write_extent_buffer(c, root->fs_info->fsid,
7126 btrfs_header_fsid(), BTRFS_FSID_SIZE);
7128 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
7129 btrfs_header_chunk_tree_uuid(c),
7132 btrfs_mark_buffer_dirty(c);
7134 * this case can happen in the following case:
7136 * 1.overwrite previous root.
7138 * 2.reinit reloc data root, this is because we skip pin
7139 * down reloc data tree before which means we can allocate
7140 * same block bytenr here.
7142 if (old->start == c->start) {
7143 btrfs_set_root_generation(&root->root_item,
7145 root->root_item.level = btrfs_header_level(root->node);
7146 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7147 &root->root_key, &root->root_item);
7149 free_extent_buffer(c);
7153 free_extent_buffer(old);
7155 add_root_to_dirty_list(root);
7159 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
7160 struct extent_buffer *eb, int tree_root)
7162 struct extent_buffer *tmp;
7163 struct btrfs_root_item *ri;
7164 struct btrfs_key key;
7167 int level = btrfs_header_level(eb);
7173 * If we have pinned this block before, don't pin it again.
7174 * This can not only avoid forever loop with broken filesystem
7175 * but also give us some speedups.
7177 if (test_range_bit(&fs_info->pinned_extents, eb->start,
7178 eb->start + eb->len - 1, EXTENT_DIRTY, 0))
7181 btrfs_pin_extent(fs_info, eb->start, eb->len);
7183 leafsize = btrfs_super_leafsize(fs_info->super_copy);
7184 nritems = btrfs_header_nritems(eb);
7185 for (i = 0; i < nritems; i++) {
7187 btrfs_item_key_to_cpu(eb, &key, i);
7188 if (key.type != BTRFS_ROOT_ITEM_KEY)
7190 /* Skip the extent root and reloc roots */
7191 if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
7192 key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
7193 key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
7195 ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
7196 bytenr = btrfs_disk_root_bytenr(eb, ri);
7199 * If at any point we start needing the real root we
7200 * will have to build a stump root for the root we are
7201 * in, but for now this doesn't actually use the root so
7202 * just pass in extent_root.
7204 tmp = read_tree_block(fs_info->extent_root, bytenr,
7207 fprintf(stderr, "Error reading root block\n");
7210 ret = pin_down_tree_blocks(fs_info, tmp, 0);
7211 free_extent_buffer(tmp);
7215 bytenr = btrfs_node_blockptr(eb, i);
7217 /* If we aren't the tree root don't read the block */
7218 if (level == 1 && !tree_root) {
7219 btrfs_pin_extent(fs_info, bytenr, leafsize);
7223 tmp = read_tree_block(fs_info->extent_root, bytenr,
7226 fprintf(stderr, "Error reading tree block\n");
7229 ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
7230 free_extent_buffer(tmp);
7239 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
7243 ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
7247 return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
7250 static int reset_block_groups(struct btrfs_fs_info *fs_info)
7252 struct btrfs_block_group_cache *cache;
7253 struct btrfs_path *path;
7254 struct extent_buffer *leaf;
7255 struct btrfs_chunk *chunk;
7256 struct btrfs_key key;
7260 path = btrfs_alloc_path();
7265 key.type = BTRFS_CHUNK_ITEM_KEY;
7268 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
7270 btrfs_free_path(path);
7275 * We do this in case the block groups were screwed up and had alloc
7276 * bits that aren't actually set on the chunks. This happens with
7277 * restored images every time and could happen in real life I guess.
7279 fs_info->avail_data_alloc_bits = 0;
7280 fs_info->avail_metadata_alloc_bits = 0;
7281 fs_info->avail_system_alloc_bits = 0;
7283 /* First we need to create the in-memory block groups */
7285 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7286 ret = btrfs_next_leaf(fs_info->chunk_root, path);
7288 btrfs_free_path(path);
7296 leaf = path->nodes[0];
7297 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7298 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7303 chunk = btrfs_item_ptr(leaf, path->slots[0],
7304 struct btrfs_chunk);
7305 btrfs_add_block_group(fs_info, 0,
7306 btrfs_chunk_type(leaf, chunk),
7307 key.objectid, key.offset,
7308 btrfs_chunk_length(leaf, chunk));
7309 set_extent_dirty(&fs_info->free_space_cache, key.offset,
7310 key.offset + btrfs_chunk_length(leaf, chunk),
7316 cache = btrfs_lookup_first_block_group(fs_info, start);
7320 start = cache->key.objectid + cache->key.offset;
7323 btrfs_free_path(path);
7327 static int reset_balance(struct btrfs_trans_handle *trans,
7328 struct btrfs_fs_info *fs_info)
7330 struct btrfs_root *root = fs_info->tree_root;
7331 struct btrfs_path *path;
7332 struct extent_buffer *leaf;
7333 struct btrfs_key key;
7334 int del_slot, del_nr = 0;
7338 path = btrfs_alloc_path();
7342 key.objectid = BTRFS_BALANCE_OBJECTID;
7343 key.type = BTRFS_BALANCE_ITEM_KEY;
7346 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7351 goto reinit_data_reloc;
7356 ret = btrfs_del_item(trans, root, path);
7359 btrfs_release_path(path);
7361 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7362 key.type = BTRFS_ROOT_ITEM_KEY;
7365 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7369 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7374 ret = btrfs_del_items(trans, root, path,
7381 btrfs_release_path(path);
7384 ret = btrfs_search_slot(trans, root, &key, path,
7391 leaf = path->nodes[0];
7392 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7393 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
7395 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7400 del_slot = path->slots[0];
7409 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
7413 btrfs_release_path(path);
7416 key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7417 key.type = BTRFS_ROOT_ITEM_KEY;
7418 key.offset = (u64)-1;
7419 root = btrfs_read_fs_root(fs_info, &key);
7421 fprintf(stderr, "Error reading data reloc tree\n");
7422 return PTR_ERR(root);
7424 record_root_in_trans(trans, root);
7425 ret = btrfs_fsck_reinit_root(trans, root, 0);
7428 ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
7430 btrfs_free_path(path);
7434 static int reinit_extent_tree(struct btrfs_trans_handle *trans,
7435 struct btrfs_fs_info *fs_info)
7441 * The only reason we don't do this is because right now we're just
7442 * walking the trees we find and pinning down their bytes, we don't look
7443 * at any of the leaves. In order to do mixed groups we'd have to check
7444 * the leaves of any fs roots and pin down the bytes for any file
7445 * extents we find. Not hard but why do it if we don't have to?
7447 if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
7448 fprintf(stderr, "We don't support re-initing the extent tree "
7449 "for mixed block groups yet, please notify a btrfs "
7450 "developer you want to do this so they can add this "
7451 "functionality.\n");
7456 * first we need to walk all of the trees except the extent tree and pin
7457 * down the bytes that are in use so we don't overwrite any existing
7460 ret = pin_metadata_blocks(fs_info);
7462 fprintf(stderr, "error pinning down used bytes\n");
7467 * Need to drop all the block groups since we're going to recreate all
7470 btrfs_free_block_groups(fs_info);
7471 ret = reset_block_groups(fs_info);
7473 fprintf(stderr, "error resetting the block groups\n");
7477 /* Ok we can allocate now, reinit the extent root */
7478 ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 0);
7480 fprintf(stderr, "extent root initialization failed\n");
7482 * When the transaction code is updated we should end the
7483 * transaction, but for now progs only knows about commit so
7484 * just return an error.
7490 * Now we have all the in-memory block groups setup so we can make
7491 * allocations properly, and the metadata we care about is safe since we
7492 * pinned all of it above.
7495 struct btrfs_block_group_cache *cache;
7497 cache = btrfs_lookup_first_block_group(fs_info, start);
7500 start = cache->key.objectid + cache->key.offset;
7501 ret = btrfs_insert_item(trans, fs_info->extent_root,
7502 &cache->key, &cache->item,
7503 sizeof(cache->item));
7505 fprintf(stderr, "Error adding block group\n");
7508 btrfs_extent_post_op(trans, fs_info->extent_root);
7511 ret = reset_balance(trans, fs_info);
7513 fprintf(stderr, "error reseting the pending balance\n");
7518 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
7520 struct btrfs_path *path;
7521 struct btrfs_trans_handle *trans;
7522 struct btrfs_key key;
7525 printf("Recowing metadata block %llu\n", eb->start);
7526 key.objectid = btrfs_header_owner(eb);
7527 key.type = BTRFS_ROOT_ITEM_KEY;
7528 key.offset = (u64)-1;
7530 root = btrfs_read_fs_root(root->fs_info, &key);
7532 fprintf(stderr, "Couldn't find owner root %llu\n",
7534 return PTR_ERR(root);
7537 path = btrfs_alloc_path();
7541 trans = btrfs_start_transaction(root, 1);
7542 if (IS_ERR(trans)) {
7543 btrfs_free_path(path);
7544 return PTR_ERR(trans);
7547 path->lowest_level = btrfs_header_level(eb);
7548 if (path->lowest_level)
7549 btrfs_node_key_to_cpu(eb, &key, 0);
7551 btrfs_item_key_to_cpu(eb, &key, 0);
7553 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7554 btrfs_commit_transaction(trans, root);
7555 btrfs_free_path(path);
7559 static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
7561 struct btrfs_path *path;
7562 struct btrfs_trans_handle *trans;
7563 struct btrfs_key key;
7566 printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
7567 bad->key.type, bad->key.offset);
7568 key.objectid = bad->root_id;
7569 key.type = BTRFS_ROOT_ITEM_KEY;
7570 key.offset = (u64)-1;
7572 root = btrfs_read_fs_root(root->fs_info, &key);
7574 fprintf(stderr, "Couldn't find owner root %llu\n",
7576 return PTR_ERR(root);
7579 path = btrfs_alloc_path();
7583 trans = btrfs_start_transaction(root, 1);
7584 if (IS_ERR(trans)) {
7585 btrfs_free_path(path);
7586 return PTR_ERR(trans);
7589 ret = btrfs_search_slot(trans, root, &bad->key, path, -1, 1);
7595 ret = btrfs_del_item(trans, root, path);
7597 btrfs_commit_transaction(trans, root);
7598 btrfs_free_path(path);
7602 static int zero_log_tree(struct btrfs_root *root)
7604 struct btrfs_trans_handle *trans;
7607 trans = btrfs_start_transaction(root, 1);
7608 if (IS_ERR(trans)) {
7609 ret = PTR_ERR(trans);
7612 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
7613 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
7614 ret = btrfs_commit_transaction(trans, root);
7618 static int populate_csum(struct btrfs_trans_handle *trans,
7619 struct btrfs_root *csum_root, char *buf, u64 start,
7626 while (offset < len) {
7627 sectorsize = csum_root->sectorsize;
7628 ret = read_extent_data(csum_root, buf, start + offset,
7632 ret = btrfs_csum_file_block(trans, csum_root, start + len,
7633 start + offset, buf, sectorsize);
7636 offset += sectorsize;
7641 static int fill_csum_tree(struct btrfs_trans_handle *trans,
7642 struct btrfs_root *csum_root)
7644 struct btrfs_root *extent_root = csum_root->fs_info->extent_root;
7645 struct btrfs_path *path;
7646 struct btrfs_extent_item *ei;
7647 struct extent_buffer *leaf;
7649 struct btrfs_key key;
7652 path = btrfs_alloc_path();
7657 key.type = BTRFS_EXTENT_ITEM_KEY;
7660 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
7662 btrfs_free_path(path);
7666 buf = malloc(csum_root->sectorsize);
7668 btrfs_free_path(path);
7673 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7674 ret = btrfs_next_leaf(extent_root, path);
7682 leaf = path->nodes[0];
7684 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7685 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
7690 ei = btrfs_item_ptr(leaf, path->slots[0],
7691 struct btrfs_extent_item);
7692 if (!(btrfs_extent_flags(leaf, ei) &
7693 BTRFS_EXTENT_FLAG_DATA)) {
7698 ret = populate_csum(trans, csum_root, buf, key.objectid,
7705 btrfs_free_path(path);
7710 struct root_item_info {
7711 /* level of the root */
7713 /* number of nodes at this level, must be 1 for a root */
7717 struct cache_extent cache_extent;
7720 static struct cache_tree *roots_info_cache = NULL;
7722 static void free_roots_info_cache(void)
7724 if (!roots_info_cache)
7727 while (!cache_tree_empty(roots_info_cache)) {
7728 struct cache_extent *entry;
7729 struct root_item_info *rii;
7731 entry = first_cache_extent(roots_info_cache);
7732 remove_cache_extent(roots_info_cache, entry);
7733 rii = container_of(entry, struct root_item_info, cache_extent);
7737 free(roots_info_cache);
7738 roots_info_cache = NULL;
7741 static int build_roots_info_cache(struct btrfs_fs_info *info)
7744 struct btrfs_key key;
7745 struct extent_buffer *leaf;
7746 struct btrfs_path *path;
7748 if (!roots_info_cache) {
7749 roots_info_cache = malloc(sizeof(*roots_info_cache));
7750 if (!roots_info_cache)
7752 cache_tree_init(roots_info_cache);
7755 path = btrfs_alloc_path();
7760 key.type = BTRFS_EXTENT_ITEM_KEY;
7763 ret = btrfs_search_slot(NULL, info->extent_root, &key, path, 0, 0);
7766 leaf = path->nodes[0];
7769 struct btrfs_key found_key;
7770 struct btrfs_extent_item *ei;
7771 struct btrfs_extent_inline_ref *iref;
7772 int slot = path->slots[0];
7777 struct cache_extent *entry;
7778 struct root_item_info *rii;
7780 if (slot >= btrfs_header_nritems(leaf)) {
7781 ret = btrfs_next_leaf(info->extent_root, path);
7788 leaf = path->nodes[0];
7789 slot = path->slots[0];
7792 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7794 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
7795 found_key.type != BTRFS_METADATA_ITEM_KEY)
7798 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
7799 flags = btrfs_extent_flags(leaf, ei);
7801 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
7802 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
7805 if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
7806 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
7807 level = found_key.offset;
7809 struct btrfs_tree_block_info *info;
7811 info = (struct btrfs_tree_block_info *)(ei + 1);
7812 iref = (struct btrfs_extent_inline_ref *)(info + 1);
7813 level = btrfs_tree_block_level(leaf, info);
7817 * For a root extent, it must be of the following type and the
7818 * first (and only one) iref in the item.
7820 type = btrfs_extent_inline_ref_type(leaf, iref);
7821 if (type != BTRFS_TREE_BLOCK_REF_KEY)
7824 root_id = btrfs_extent_inline_ref_offset(leaf, iref);
7825 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
7827 rii = malloc(sizeof(struct root_item_info));
7832 rii->cache_extent.start = root_id;
7833 rii->cache_extent.size = 1;
7834 rii->level = (u8)-1;
7835 entry = &rii->cache_extent;
7836 ret = insert_cache_extent(roots_info_cache, entry);
7839 rii = container_of(entry, struct root_item_info,
7843 ASSERT(rii->cache_extent.start == root_id);
7844 ASSERT(rii->cache_extent.size == 1);
7846 if (level > rii->level || rii->level == (u8)-1) {
7848 rii->bytenr = found_key.objectid;
7849 rii->gen = btrfs_extent_generation(leaf, ei);
7850 rii->node_count = 1;
7851 } else if (level == rii->level) {
7859 btrfs_free_path(path);
7864 static int maybe_repair_root_item(struct btrfs_fs_info *info,
7865 struct btrfs_path *path,
7866 const struct btrfs_key *root_key,
7867 const int read_only_mode)
7869 const u64 root_id = root_key->objectid;
7870 struct cache_extent *entry;
7871 struct root_item_info *rii;
7872 struct btrfs_root_item ri;
7873 unsigned long offset;
7875 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
7878 "Error: could not find extent items for root %llu\n",
7879 root_key->objectid);
7883 rii = container_of(entry, struct root_item_info, cache_extent);
7884 ASSERT(rii->cache_extent.start == root_id);
7885 ASSERT(rii->cache_extent.size == 1);
7887 if (rii->node_count != 1) {
7889 "Error: could not find btree root extent for root %llu\n",
7894 offset = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
7895 read_extent_buffer(path->nodes[0], &ri, offset, sizeof(ri));
7897 if (btrfs_root_bytenr(&ri) != rii->bytenr ||
7898 btrfs_root_level(&ri) != rii->level ||
7899 btrfs_root_generation(&ri) != rii->gen) {
7902 * If we're in repair mode but our caller told us to not update
7903 * the root item, i.e. just check if it needs to be updated, don't
7904 * print this message, since the caller will call us again shortly
7905 * for the same root item without read only mode (the caller will
7906 * open a transaction first).
7908 if (!(read_only_mode && repair))
7910 "%sroot item for root %llu,"
7911 " current bytenr %llu, current gen %llu, current level %u,"
7912 " new bytenr %llu, new gen %llu, new level %u\n",
7913 (read_only_mode ? "" : "fixing "),
7915 btrfs_root_bytenr(&ri), btrfs_root_generation(&ri),
7916 btrfs_root_level(&ri),
7917 rii->bytenr, rii->gen, rii->level);
7919 if (btrfs_root_generation(&ri) > rii->gen) {
7921 "root %llu has a root item with a more recent gen (%llu) compared to the found root node (%llu)\n",
7922 root_id, btrfs_root_generation(&ri), rii->gen);
7926 if (!read_only_mode) {
7927 btrfs_set_root_bytenr(&ri, rii->bytenr);
7928 btrfs_set_root_level(&ri, rii->level);
7929 btrfs_set_root_generation(&ri, rii->gen);
7930 write_extent_buffer(path->nodes[0], &ri,
7931 offset, sizeof(ri));
7941 * A regression introduced in the 3.17 kernel (more specifically in 3.17-rc2),
7942 * caused read-only snapshots to be corrupted if they were created at a moment
7943 * when the source subvolume/snapshot had orphan items. The issue was that the
7944 * on-disk root items became incorrect, referring to the pre orphan cleanup root
7945 * node instead of the post orphan cleanup root node.
7946 * So this function, and its callees, just detects and fixes those cases. Even
7947 * though the regression was for read-only snapshots, this function applies to
7948 * any snapshot/subvolume root.
7949 * This must be run before any other repair code - not doing it so, makes other
7950 * repair code delete or modify backrefs in the extent tree for example, which
7951 * will result in an inconsistent fs after repairing the root items.
7953 static int repair_root_items(struct btrfs_fs_info *info)
7955 struct btrfs_path *path = NULL;
7956 struct btrfs_key key;
7957 struct extent_buffer *leaf;
7958 struct btrfs_trans_handle *trans = NULL;
7963 ret = build_roots_info_cache(info);
7967 path = btrfs_alloc_path();
7973 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
7974 key.type = BTRFS_ROOT_ITEM_KEY;
7979 * Avoid opening and committing transactions if a leaf doesn't have
7980 * any root items that need to be fixed, so that we avoid rotating
7981 * backup roots unnecessarily.
7984 trans = btrfs_start_transaction(info->tree_root, 1);
7985 if (IS_ERR(trans)) {
7986 ret = PTR_ERR(trans);
7991 ret = btrfs_search_slot(trans, info->tree_root, &key, path,
7995 leaf = path->nodes[0];
7998 struct btrfs_key found_key;
8000 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
8001 int no_more_keys = find_next_key(path, &key);
8003 btrfs_release_path(path);
8005 ret = btrfs_commit_transaction(trans,
8017 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8019 if (found_key.type != BTRFS_ROOT_ITEM_KEY)
8022 ret = maybe_repair_root_item(info, path, &found_key,
8027 if (!trans && repair) {
8030 btrfs_release_path(path);
8040 free_roots_info_cache();
8042 btrfs_free_path(path);
8049 static struct option long_options[] = {
8050 { "super", 1, NULL, 's' },
8051 { "repair", 0, NULL, 0 },
8052 { "init-csum-tree", 0, NULL, 0 },
8053 { "init-extent-tree", 0, NULL, 0 },
8054 { "check-data-csum", 0, NULL, 0 },
8055 { "backup", 0, NULL, 0 },
8056 { "subvol-extents", 1, NULL, 'E' },
8057 { "qgroup-report", 0, NULL, 'Q' },
8058 { "tree-root", 1, NULL, 'r' },
8062 const char * const cmd_check_usage[] = {
8063 "btrfs check [options] <device>",
8064 "Check an unmounted btrfs filesystem.",
8066 "-s|--super <superblock> use this superblock copy",
8067 "-b|--backup use the backup root copy",
8068 "--repair try to repair the filesystem",
8069 "--init-csum-tree create a new CRC tree",
8070 "--init-extent-tree create a new extent tree",
8071 "--check-data-csum verify checkums of data blocks",
8072 "--qgroup-report print a report on qgroup consistency",
8073 "--subvol-extents <subvolid> print subvolume extents and sharing state",
8074 "--tree-root <bytenr> use the given bytenr for the tree root",
8078 int cmd_check(int argc, char **argv)
8080 struct cache_tree root_cache;
8081 struct btrfs_root *root;
8082 struct btrfs_fs_info *info;
8085 u64 tree_root_bytenr = 0;
8086 char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
8089 int option_index = 0;
8090 int init_csum_tree = 0;
8091 int qgroup_report = 0;
8092 enum btrfs_open_ctree_flags ctree_flags = OPEN_CTREE_EXCLUSIVE;
8096 c = getopt_long(argc, argv, "as:br:", long_options,
8101 case 'a': /* ignored */ break;
8103 ctree_flags |= OPEN_CTREE_BACKUP_ROOT;
8106 num = arg_strtou64(optarg);
8107 if (num >= BTRFS_SUPER_MIRROR_MAX) {
8109 "ERROR: super mirror should be less than: %d\n",
8110 BTRFS_SUPER_MIRROR_MAX);
8113 bytenr = btrfs_sb_offset(((int)num));
8114 printf("using SB copy %llu, bytenr %llu\n", num,
8115 (unsigned long long)bytenr);
8121 subvolid = arg_strtou64(optarg);
8124 tree_root_bytenr = arg_strtou64(optarg);
8128 usage(cmd_check_usage);
8130 if (option_index == 1) {
8131 printf("enabling repair mode\n");
8133 ctree_flags |= OPEN_CTREE_WRITES;
8134 } else if (option_index == 2) {
8135 printf("Creating a new CRC tree\n");
8138 ctree_flags |= OPEN_CTREE_WRITES;
8139 } else if (option_index == 3) {
8140 init_extent_tree = 1;
8141 ctree_flags |= (OPEN_CTREE_WRITES |
8142 OPEN_CTREE_NO_BLOCK_GROUPS);
8144 } else if (option_index == 4) {
8145 check_data_csum = 1;
8148 argc = argc - optind;
8150 if (check_argc_exact(argc, 1))
8151 usage(cmd_check_usage);
8154 cache_tree_init(&root_cache);
8156 if((ret = check_mounted(argv[optind])) < 0) {
8157 fprintf(stderr, "Could not check mount status: %s\n", strerror(-ret));
8160 fprintf(stderr, "%s is currently mounted. Aborting.\n", argv[optind]);
8165 /* only allow partial opening under repair mode */
8167 ctree_flags |= OPEN_CTREE_PARTIAL;
8169 info = open_ctree_fs_info(argv[optind], bytenr, tree_root_bytenr,
8172 fprintf(stderr, "Couldn't open file system\n");
8177 root = info->fs_root;
8179 ret = repair_root_items(info);
8183 fprintf(stderr, "Fixed %d roots.\n", ret);
8185 } else if (ret > 0) {
8187 "Found %d roots with an outdated root item.\n",
8190 "Please run a filesystem check with the option --repair to fix them.\n");
8196 * repair mode will force us to commit transaction which
8197 * will make us fail to load log tree when mounting.
8199 if (repair && btrfs_super_log_root(info->super_copy)) {
8200 ret = ask_user("repair mode will force to clear out log tree, Are you sure?");
8205 ret = zero_log_tree(root);
8207 fprintf(stderr, "fail to zero log tree\n");
8212 uuid_unparse(info->super_copy->fsid, uuidbuf);
8213 if (qgroup_report) {
8214 printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
8216 ret = qgroup_verify_all(info);
8218 print_qgroup_report(1);
8222 printf("Print extent state for subvolume %llu on %s\nUUID: %s\n",
8223 subvolid, argv[optind], uuidbuf);
8224 ret = print_extent_state(info, subvolid);
8227 printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
8229 if (!extent_buffer_uptodate(info->tree_root->node) ||
8230 !extent_buffer_uptodate(info->dev_root->node) ||
8231 !extent_buffer_uptodate(info->chunk_root->node)) {
8232 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
8237 if (init_extent_tree || init_csum_tree) {
8238 struct btrfs_trans_handle *trans;
8240 trans = btrfs_start_transaction(info->extent_root, 0);
8241 if (IS_ERR(trans)) {
8242 fprintf(stderr, "Error starting transaction\n");
8243 ret = PTR_ERR(trans);
8247 if (init_extent_tree) {
8248 printf("Creating a new extent tree\n");
8249 ret = reinit_extent_tree(trans, info);
8254 if (init_csum_tree) {
8255 fprintf(stderr, "Reinit crc root\n");
8256 ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
8258 fprintf(stderr, "crc root initialization failed\n");
8263 ret = fill_csum_tree(trans, info->csum_root);
8265 fprintf(stderr, "crc refilling failed\n");
8270 * Ok now we commit and run the normal fsck, which will add
8271 * extent entries for all of the items it finds.
8273 ret = btrfs_commit_transaction(trans, info->extent_root);
8277 if (!extent_buffer_uptodate(info->extent_root->node)) {
8278 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
8282 if (!extent_buffer_uptodate(info->csum_root->node)) {
8283 fprintf(stderr, "Checksum root corrupted, rerun with --init-csum-tree option\n");
8288 fprintf(stderr, "checking extents\n");
8289 ret = check_chunks_and_extents(root);
8291 fprintf(stderr, "Errors found in extent allocation tree or chunk allocation\n");
8293 fprintf(stderr, "checking free space cache\n");
8294 ret = check_space_cache(root);
8299 * We used to have to have these hole extents in between our real
8300 * extents so if we don't have this flag set we need to make sure there
8301 * are no gaps in the file extents for inodes, otherwise we can just
8302 * ignore it when this happens.
8304 no_holes = btrfs_fs_incompat(root->fs_info,
8305 BTRFS_FEATURE_INCOMPAT_NO_HOLES);
8306 fprintf(stderr, "checking fs roots\n");
8307 ret = check_fs_roots(root, &root_cache);
8311 fprintf(stderr, "checking csums\n");
8312 ret = check_csums(root);
8316 fprintf(stderr, "checking root refs\n");
8317 ret = check_root_refs(root, &root_cache);
8321 while (repair && !list_empty(&root->fs_info->recow_ebs)) {
8322 struct extent_buffer *eb;
8324 eb = list_first_entry(&root->fs_info->recow_ebs,
8325 struct extent_buffer, recow);
8326 list_del_init(&eb->recow);
8327 ret = recow_extent_buffer(root, eb);
8332 while (!list_empty(&delete_items)) {
8333 struct bad_item *bad;
8335 bad = list_first_entry(&delete_items, struct bad_item, list);
8336 list_del_init(&bad->list);
8338 ret = delete_bad_item(root, bad);
8342 if (info->quota_enabled) {
8344 fprintf(stderr, "checking quota groups\n");
8345 err = qgroup_verify_all(info);
8350 if (!list_empty(&root->fs_info->recow_ebs)) {
8351 fprintf(stderr, "Transid errors in file system\n");
8355 print_qgroup_report(0);
8356 if (found_old_backref) { /*
8357 * there was a disk format change when mixed
8358 * backref was in testing tree. The old format
8359 * existed about one week.
8361 printf("\n * Found old mixed backref format. "
8362 "The old format is not supported! *"
8363 "\n * Please mount the FS in readonly mode, "
8364 "backup data and re-format the FS. *\n\n");
8367 printf("found %llu bytes used err is %d\n",
8368 (unsigned long long)bytes_used, ret);
8369 printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
8370 printf("total tree bytes: %llu\n",
8371 (unsigned long long)total_btree_bytes);
8372 printf("total fs tree bytes: %llu\n",
8373 (unsigned long long)total_fs_tree_bytes);
8374 printf("total extent tree bytes: %llu\n",
8375 (unsigned long long)total_extent_tree_bytes);
8376 printf("btree space waste bytes: %llu\n",
8377 (unsigned long long)btree_space_waste);
8378 printf("file data blocks allocated: %llu\n referenced %llu\n",
8379 (unsigned long long)data_bytes_allocated,
8380 (unsigned long long)data_bytes_referenced);
8381 printf("%s\n", BTRFS_BUILD_VERSION);
8383 free_root_recs_tree(&root_cache);