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.
23 #include <sys/types.h>
27 #include <uuid/uuid.h>
32 #include "print-tree.h"
33 #include "transaction.h"
37 #include "free-space-cache.h"
39 #include "qgroup-verify.h"
40 #include "rbtree-utils.h"
44 static u64 bytes_used = 0;
45 static u64 total_csum_bytes = 0;
46 static u64 total_btree_bytes = 0;
47 static u64 total_fs_tree_bytes = 0;
48 static u64 total_extent_tree_bytes = 0;
49 static u64 btree_space_waste = 0;
50 static u64 data_bytes_allocated = 0;
51 static u64 data_bytes_referenced = 0;
52 static int found_old_backref = 0;
53 static LIST_HEAD(duplicate_extents);
54 static LIST_HEAD(delete_items);
55 static int repair = 0;
56 static int no_holes = 0;
57 static int init_extent_tree = 0;
58 static int check_data_csum = 0;
60 struct extent_backref {
61 struct list_head list;
62 unsigned int is_data:1;
63 unsigned int found_extent_tree:1;
64 unsigned int full_backref:1;
65 unsigned int found_ref:1;
66 unsigned int broken:1;
70 struct extent_backref node;
85 struct extent_backref node;
92 struct extent_record {
93 struct list_head backrefs;
94 struct list_head dups;
95 struct list_head list;
96 struct cache_extent cache;
97 struct btrfs_disk_key parent_key;
102 u64 extent_item_refs;
104 u64 parent_generation;
108 unsigned int found_rec:1;
109 unsigned int content_checked:1;
110 unsigned int owner_ref_checked:1;
111 unsigned int is_root:1;
112 unsigned int metadata:1;
113 unsigned int flag_block_full_backref:1;
116 struct inode_backref {
117 struct list_head list;
118 unsigned int found_dir_item:1;
119 unsigned int found_dir_index:1;
120 unsigned int found_inode_ref:1;
121 unsigned int filetype:8;
123 unsigned int ref_type;
130 struct root_item_record {
131 struct list_head list;
137 struct btrfs_key drop_key;
140 #define REF_ERR_NO_DIR_ITEM (1 << 0)
141 #define REF_ERR_NO_DIR_INDEX (1 << 1)
142 #define REF_ERR_NO_INODE_REF (1 << 2)
143 #define REF_ERR_DUP_DIR_ITEM (1 << 3)
144 #define REF_ERR_DUP_DIR_INDEX (1 << 4)
145 #define REF_ERR_DUP_INODE_REF (1 << 5)
146 #define REF_ERR_INDEX_UNMATCH (1 << 6)
147 #define REF_ERR_FILETYPE_UNMATCH (1 << 7)
148 #define REF_ERR_NAME_TOO_LONG (1 << 8) // 100
149 #define REF_ERR_NO_ROOT_REF (1 << 9)
150 #define REF_ERR_NO_ROOT_BACKREF (1 << 10)
151 #define REF_ERR_DUP_ROOT_REF (1 << 11)
152 #define REF_ERR_DUP_ROOT_BACKREF (1 << 12)
154 struct inode_record {
155 struct list_head backrefs;
156 unsigned int checked:1;
157 unsigned int merging:1;
158 unsigned int found_inode_item:1;
159 unsigned int found_dir_item:1;
160 unsigned int found_file_extent:1;
161 unsigned int found_csum_item:1;
162 unsigned int some_csum_missing:1;
163 unsigned int nodatasum:1;
176 u64 first_extent_gap;
181 #define I_ERR_NO_INODE_ITEM (1 << 0)
182 #define I_ERR_NO_ORPHAN_ITEM (1 << 1)
183 #define I_ERR_DUP_INODE_ITEM (1 << 2)
184 #define I_ERR_DUP_DIR_INDEX (1 << 3)
185 #define I_ERR_ODD_DIR_ITEM (1 << 4)
186 #define I_ERR_ODD_FILE_EXTENT (1 << 5)
187 #define I_ERR_BAD_FILE_EXTENT (1 << 6)
188 #define I_ERR_FILE_EXTENT_OVERLAP (1 << 7)
189 #define I_ERR_FILE_EXTENT_DISCOUNT (1 << 8) // 100
190 #define I_ERR_DIR_ISIZE_WRONG (1 << 9)
191 #define I_ERR_FILE_NBYTES_WRONG (1 << 10) // 400
192 #define I_ERR_ODD_CSUM_ITEM (1 << 11)
193 #define I_ERR_SOME_CSUM_MISSING (1 << 12)
194 #define I_ERR_LINK_COUNT_WRONG (1 << 13)
196 struct root_backref {
197 struct list_head list;
198 unsigned int found_dir_item:1;
199 unsigned int found_dir_index:1;
200 unsigned int found_back_ref:1;
201 unsigned int found_forward_ref:1;
202 unsigned int reachable:1;
212 struct list_head backrefs;
213 struct cache_extent cache;
214 unsigned int found_root_item:1;
220 struct cache_extent cache;
225 struct cache_extent cache;
226 struct cache_tree root_cache;
227 struct cache_tree inode_cache;
228 struct inode_record *current;
237 struct walk_control {
238 struct cache_tree shared;
239 struct shared_node *nodes[BTRFS_MAX_LEVEL];
245 struct btrfs_key key;
247 struct list_head list;
250 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info);
252 static void record_root_in_trans(struct btrfs_trans_handle *trans,
253 struct btrfs_root *root)
255 if (root->last_trans != trans->transid) {
256 root->track_dirty = 1;
257 root->last_trans = trans->transid;
258 root->commit_root = root->node;
259 extent_buffer_get(root->node);
263 static u8 imode_to_type(u32 imode)
266 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
267 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
268 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
269 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
270 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
271 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
272 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
273 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
276 return btrfs_type_by_mode[(imode & S_IFMT) >> S_SHIFT];
280 static int device_record_compare(struct rb_node *node1, struct rb_node *node2)
282 struct device_record *rec1;
283 struct device_record *rec2;
285 rec1 = rb_entry(node1, struct device_record, node);
286 rec2 = rb_entry(node2, struct device_record, node);
287 if (rec1->devid > rec2->devid)
289 else if (rec1->devid < rec2->devid)
295 static struct inode_record *clone_inode_rec(struct inode_record *orig_rec)
297 struct inode_record *rec;
298 struct inode_backref *backref;
299 struct inode_backref *orig;
302 rec = malloc(sizeof(*rec));
303 memcpy(rec, orig_rec, sizeof(*rec));
305 INIT_LIST_HEAD(&rec->backrefs);
307 list_for_each_entry(orig, &orig_rec->backrefs, list) {
308 size = sizeof(*orig) + orig->namelen + 1;
309 backref = malloc(size);
310 memcpy(backref, orig, size);
311 list_add_tail(&backref->list, &rec->backrefs);
316 static void print_inode_error(struct btrfs_root *root, struct inode_record *rec)
318 u64 root_objectid = root->root_key.objectid;
319 int errors = rec->errors;
323 /* reloc root errors, we print its corresponding fs root objectid*/
324 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
325 root_objectid = root->root_key.offset;
326 fprintf(stderr, "reloc");
328 fprintf(stderr, "root %llu inode %llu errors %x",
329 (unsigned long long) root_objectid,
330 (unsigned long long) rec->ino, rec->errors);
332 if (errors & I_ERR_NO_INODE_ITEM)
333 fprintf(stderr, ", no inode item");
334 if (errors & I_ERR_NO_ORPHAN_ITEM)
335 fprintf(stderr, ", no orphan item");
336 if (errors & I_ERR_DUP_INODE_ITEM)
337 fprintf(stderr, ", dup inode item");
338 if (errors & I_ERR_DUP_DIR_INDEX)
339 fprintf(stderr, ", dup dir index");
340 if (errors & I_ERR_ODD_DIR_ITEM)
341 fprintf(stderr, ", odd dir item");
342 if (errors & I_ERR_ODD_FILE_EXTENT)
343 fprintf(stderr, ", odd file extent");
344 if (errors & I_ERR_BAD_FILE_EXTENT)
345 fprintf(stderr, ", bad file extent");
346 if (errors & I_ERR_FILE_EXTENT_OVERLAP)
347 fprintf(stderr, ", file extent overlap");
348 if (errors & I_ERR_FILE_EXTENT_DISCOUNT)
349 fprintf(stderr, ", file extent discount");
350 if (errors & I_ERR_DIR_ISIZE_WRONG)
351 fprintf(stderr, ", dir isize wrong");
352 if (errors & I_ERR_FILE_NBYTES_WRONG)
353 fprintf(stderr, ", nbytes wrong");
354 if (errors & I_ERR_ODD_CSUM_ITEM)
355 fprintf(stderr, ", odd csum item");
356 if (errors & I_ERR_SOME_CSUM_MISSING)
357 fprintf(stderr, ", some csum missing");
358 if (errors & I_ERR_LINK_COUNT_WRONG)
359 fprintf(stderr, ", link count wrong");
360 fprintf(stderr, "\n");
363 static void print_ref_error(int errors)
365 if (errors & REF_ERR_NO_DIR_ITEM)
366 fprintf(stderr, ", no dir item");
367 if (errors & REF_ERR_NO_DIR_INDEX)
368 fprintf(stderr, ", no dir index");
369 if (errors & REF_ERR_NO_INODE_REF)
370 fprintf(stderr, ", no inode ref");
371 if (errors & REF_ERR_DUP_DIR_ITEM)
372 fprintf(stderr, ", dup dir item");
373 if (errors & REF_ERR_DUP_DIR_INDEX)
374 fprintf(stderr, ", dup dir index");
375 if (errors & REF_ERR_DUP_INODE_REF)
376 fprintf(stderr, ", dup inode ref");
377 if (errors & REF_ERR_INDEX_UNMATCH)
378 fprintf(stderr, ", index unmatch");
379 if (errors & REF_ERR_FILETYPE_UNMATCH)
380 fprintf(stderr, ", filetype unmatch");
381 if (errors & REF_ERR_NAME_TOO_LONG)
382 fprintf(stderr, ", name too long");
383 if (errors & REF_ERR_NO_ROOT_REF)
384 fprintf(stderr, ", no root ref");
385 if (errors & REF_ERR_NO_ROOT_BACKREF)
386 fprintf(stderr, ", no root backref");
387 if (errors & REF_ERR_DUP_ROOT_REF)
388 fprintf(stderr, ", dup root ref");
389 if (errors & REF_ERR_DUP_ROOT_BACKREF)
390 fprintf(stderr, ", dup root backref");
391 fprintf(stderr, "\n");
394 static struct inode_record *get_inode_rec(struct cache_tree *inode_cache,
397 struct ptr_node *node;
398 struct cache_extent *cache;
399 struct inode_record *rec = NULL;
402 cache = lookup_cache_extent(inode_cache, ino, 1);
404 node = container_of(cache, struct ptr_node, cache);
406 if (mod && rec->refs > 1) {
407 node->data = clone_inode_rec(rec);
412 rec = calloc(1, sizeof(*rec));
414 rec->extent_start = (u64)-1;
415 rec->first_extent_gap = (u64)-1;
417 INIT_LIST_HEAD(&rec->backrefs);
419 node = malloc(sizeof(*node));
420 node->cache.start = ino;
421 node->cache.size = 1;
424 if (ino == BTRFS_FREE_INO_OBJECTID)
427 ret = insert_cache_extent(inode_cache, &node->cache);
433 static void free_inode_rec(struct inode_record *rec)
435 struct inode_backref *backref;
440 while (!list_empty(&rec->backrefs)) {
441 backref = list_entry(rec->backrefs.next,
442 struct inode_backref, list);
443 list_del(&backref->list);
449 static int can_free_inode_rec(struct inode_record *rec)
451 if (!rec->errors && rec->checked && rec->found_inode_item &&
452 rec->nlink == rec->found_link && list_empty(&rec->backrefs))
457 static void maybe_free_inode_rec(struct cache_tree *inode_cache,
458 struct inode_record *rec)
460 struct cache_extent *cache;
461 struct inode_backref *tmp, *backref;
462 struct ptr_node *node;
463 unsigned char filetype;
465 if (!rec->found_inode_item)
468 filetype = imode_to_type(rec->imode);
469 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
470 if (backref->found_dir_item && backref->found_dir_index) {
471 if (backref->filetype != filetype)
472 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
473 if (!backref->errors && backref->found_inode_ref) {
474 list_del(&backref->list);
480 if (!rec->checked || rec->merging)
483 if (S_ISDIR(rec->imode)) {
484 if (rec->found_size != rec->isize)
485 rec->errors |= I_ERR_DIR_ISIZE_WRONG;
486 if (rec->found_file_extent)
487 rec->errors |= I_ERR_ODD_FILE_EXTENT;
488 } else if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
489 if (rec->found_dir_item)
490 rec->errors |= I_ERR_ODD_DIR_ITEM;
491 if (rec->found_size != rec->nbytes)
492 rec->errors |= I_ERR_FILE_NBYTES_WRONG;
493 if (rec->extent_start == (u64)-1 || rec->extent_start > 0)
494 rec->first_extent_gap = 0;
495 if (rec->nlink > 0 && !no_holes &&
496 (rec->extent_end < rec->isize ||
497 rec->first_extent_gap < rec->isize))
498 rec->errors |= I_ERR_FILE_EXTENT_DISCOUNT;
501 if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
502 if (rec->found_csum_item && rec->nodatasum)
503 rec->errors |= I_ERR_ODD_CSUM_ITEM;
504 if (rec->some_csum_missing && !rec->nodatasum)
505 rec->errors |= I_ERR_SOME_CSUM_MISSING;
508 BUG_ON(rec->refs != 1);
509 if (can_free_inode_rec(rec)) {
510 cache = lookup_cache_extent(inode_cache, rec->ino, 1);
511 node = container_of(cache, struct ptr_node, cache);
512 BUG_ON(node->data != rec);
513 remove_cache_extent(inode_cache, &node->cache);
519 static int check_orphan_item(struct btrfs_root *root, u64 ino)
521 struct btrfs_path path;
522 struct btrfs_key key;
525 key.objectid = BTRFS_ORPHAN_OBJECTID;
526 key.type = BTRFS_ORPHAN_ITEM_KEY;
529 btrfs_init_path(&path);
530 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
531 btrfs_release_path(&path);
537 static int process_inode_item(struct extent_buffer *eb,
538 int slot, struct btrfs_key *key,
539 struct shared_node *active_node)
541 struct inode_record *rec;
542 struct btrfs_inode_item *item;
544 rec = active_node->current;
545 BUG_ON(rec->ino != key->objectid || rec->refs > 1);
546 if (rec->found_inode_item) {
547 rec->errors |= I_ERR_DUP_INODE_ITEM;
550 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
551 rec->nlink = btrfs_inode_nlink(eb, item);
552 rec->isize = btrfs_inode_size(eb, item);
553 rec->nbytes = btrfs_inode_nbytes(eb, item);
554 rec->imode = btrfs_inode_mode(eb, item);
555 if (btrfs_inode_flags(eb, item) & BTRFS_INODE_NODATASUM)
557 rec->found_inode_item = 1;
559 rec->errors |= I_ERR_NO_ORPHAN_ITEM;
560 maybe_free_inode_rec(&active_node->inode_cache, rec);
564 static struct inode_backref *get_inode_backref(struct inode_record *rec,
566 int namelen, u64 dir)
568 struct inode_backref *backref;
570 list_for_each_entry(backref, &rec->backrefs, list) {
571 if (rec->ino == BTRFS_MULTIPLE_OBJECTIDS)
573 if (backref->dir != dir || backref->namelen != namelen)
575 if (memcmp(name, backref->name, namelen))
580 backref = malloc(sizeof(*backref) + namelen + 1);
581 memset(backref, 0, sizeof(*backref));
583 backref->namelen = namelen;
584 memcpy(backref->name, name, namelen);
585 backref->name[namelen] = '\0';
586 list_add_tail(&backref->list, &rec->backrefs);
590 static int add_inode_backref(struct cache_tree *inode_cache,
591 u64 ino, u64 dir, u64 index,
592 const char *name, int namelen,
593 int filetype, int itemtype, int errors)
595 struct inode_record *rec;
596 struct inode_backref *backref;
598 rec = get_inode_rec(inode_cache, ino, 1);
599 backref = get_inode_backref(rec, name, namelen, dir);
601 backref->errors |= errors;
602 if (itemtype == BTRFS_DIR_INDEX_KEY) {
603 if (backref->found_dir_index)
604 backref->errors |= REF_ERR_DUP_DIR_INDEX;
605 if (backref->found_inode_ref && backref->index != index)
606 backref->errors |= REF_ERR_INDEX_UNMATCH;
607 if (backref->found_dir_item && backref->filetype != filetype)
608 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
610 backref->index = index;
611 backref->filetype = filetype;
612 backref->found_dir_index = 1;
613 } else if (itemtype == BTRFS_DIR_ITEM_KEY) {
615 if (backref->found_dir_item)
616 backref->errors |= REF_ERR_DUP_DIR_ITEM;
617 if (backref->found_dir_index && backref->filetype != filetype)
618 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
620 backref->filetype = filetype;
621 backref->found_dir_item = 1;
622 } else if ((itemtype == BTRFS_INODE_REF_KEY) ||
623 (itemtype == BTRFS_INODE_EXTREF_KEY)) {
624 if (backref->found_inode_ref)
625 backref->errors |= REF_ERR_DUP_INODE_REF;
626 if (backref->found_dir_index && backref->index != index)
627 backref->errors |= REF_ERR_INDEX_UNMATCH;
629 backref->index = index;
631 backref->ref_type = itemtype;
632 backref->found_inode_ref = 1;
637 maybe_free_inode_rec(inode_cache, rec);
641 static int merge_inode_recs(struct inode_record *src, struct inode_record *dst,
642 struct cache_tree *dst_cache)
644 struct inode_backref *backref;
648 list_for_each_entry(backref, &src->backrefs, list) {
649 if (backref->found_dir_index) {
650 add_inode_backref(dst_cache, dst->ino, backref->dir,
651 backref->index, backref->name,
652 backref->namelen, backref->filetype,
653 BTRFS_DIR_INDEX_KEY, backref->errors);
655 if (backref->found_dir_item) {
657 add_inode_backref(dst_cache, dst->ino,
658 backref->dir, 0, backref->name,
659 backref->namelen, backref->filetype,
660 BTRFS_DIR_ITEM_KEY, backref->errors);
662 if (backref->found_inode_ref) {
663 add_inode_backref(dst_cache, dst->ino,
664 backref->dir, backref->index,
665 backref->name, backref->namelen, 0,
666 backref->ref_type, backref->errors);
670 if (src->found_dir_item)
671 dst->found_dir_item = 1;
672 if (src->found_file_extent)
673 dst->found_file_extent = 1;
674 if (src->found_csum_item)
675 dst->found_csum_item = 1;
676 if (src->some_csum_missing)
677 dst->some_csum_missing = 1;
678 if (dst->first_extent_gap > src->first_extent_gap)
679 dst->first_extent_gap = src->first_extent_gap;
681 BUG_ON(src->found_link < dir_count);
682 dst->found_link += src->found_link - dir_count;
683 dst->found_size += src->found_size;
684 if (src->extent_start != (u64)-1) {
685 if (dst->extent_start == (u64)-1) {
686 dst->extent_start = src->extent_start;
687 dst->extent_end = src->extent_end;
689 if (dst->extent_end > src->extent_start)
690 dst->errors |= I_ERR_FILE_EXTENT_OVERLAP;
691 else if (dst->extent_end < src->extent_start &&
692 dst->extent_end < dst->first_extent_gap)
693 dst->first_extent_gap = dst->extent_end;
694 if (dst->extent_end < src->extent_end)
695 dst->extent_end = src->extent_end;
699 dst->errors |= src->errors;
700 if (src->found_inode_item) {
701 if (!dst->found_inode_item) {
702 dst->nlink = src->nlink;
703 dst->isize = src->isize;
704 dst->nbytes = src->nbytes;
705 dst->imode = src->imode;
706 dst->nodatasum = src->nodatasum;
707 dst->found_inode_item = 1;
709 dst->errors |= I_ERR_DUP_INODE_ITEM;
717 static int splice_shared_node(struct shared_node *src_node,
718 struct shared_node *dst_node)
720 struct cache_extent *cache;
721 struct ptr_node *node, *ins;
722 struct cache_tree *src, *dst;
723 struct inode_record *rec, *conflict;
728 if (--src_node->refs == 0)
730 if (src_node->current)
731 current_ino = src_node->current->ino;
733 src = &src_node->root_cache;
734 dst = &dst_node->root_cache;
736 cache = search_cache_extent(src, 0);
738 node = container_of(cache, struct ptr_node, cache);
740 cache = next_cache_extent(cache);
743 remove_cache_extent(src, &node->cache);
746 ins = malloc(sizeof(*ins));
747 ins->cache.start = node->cache.start;
748 ins->cache.size = node->cache.size;
752 ret = insert_cache_extent(dst, &ins->cache);
753 if (ret == -EEXIST) {
754 conflict = get_inode_rec(dst, rec->ino, 1);
755 merge_inode_recs(rec, conflict, dst);
757 conflict->checked = 1;
758 if (dst_node->current == conflict)
759 dst_node->current = NULL;
761 maybe_free_inode_rec(dst, conflict);
769 if (src == &src_node->root_cache) {
770 src = &src_node->inode_cache;
771 dst = &dst_node->inode_cache;
775 if (current_ino > 0 && (!dst_node->current ||
776 current_ino > dst_node->current->ino)) {
777 if (dst_node->current) {
778 dst_node->current->checked = 1;
779 maybe_free_inode_rec(dst, dst_node->current);
781 dst_node->current = get_inode_rec(dst, current_ino, 1);
786 static void free_inode_ptr(struct cache_extent *cache)
788 struct ptr_node *node;
789 struct inode_record *rec;
791 node = container_of(cache, struct ptr_node, cache);
797 FREE_EXTENT_CACHE_BASED_TREE(inode_recs, free_inode_ptr);
799 static struct shared_node *find_shared_node(struct cache_tree *shared,
802 struct cache_extent *cache;
803 struct shared_node *node;
805 cache = lookup_cache_extent(shared, bytenr, 1);
807 node = container_of(cache, struct shared_node, cache);
813 static int add_shared_node(struct cache_tree *shared, u64 bytenr, u32 refs)
816 struct shared_node *node;
818 node = calloc(1, sizeof(*node));
819 node->cache.start = bytenr;
820 node->cache.size = 1;
821 cache_tree_init(&node->root_cache);
822 cache_tree_init(&node->inode_cache);
825 ret = insert_cache_extent(shared, &node->cache);
830 static int enter_shared_node(struct btrfs_root *root, u64 bytenr, u32 refs,
831 struct walk_control *wc, int level)
833 struct shared_node *node;
834 struct shared_node *dest;
836 if (level == wc->active_node)
839 BUG_ON(wc->active_node <= level);
840 node = find_shared_node(&wc->shared, bytenr);
842 add_shared_node(&wc->shared, bytenr, refs);
843 node = find_shared_node(&wc->shared, bytenr);
844 wc->nodes[level] = node;
845 wc->active_node = level;
849 if (wc->root_level == wc->active_node &&
850 btrfs_root_refs(&root->root_item) == 0) {
851 if (--node->refs == 0) {
852 free_inode_recs_tree(&node->root_cache);
853 free_inode_recs_tree(&node->inode_cache);
854 remove_cache_extent(&wc->shared, &node->cache);
860 dest = wc->nodes[wc->active_node];
861 splice_shared_node(node, dest);
862 if (node->refs == 0) {
863 remove_cache_extent(&wc->shared, &node->cache);
869 static int leave_shared_node(struct btrfs_root *root,
870 struct walk_control *wc, int level)
872 struct shared_node *node;
873 struct shared_node *dest;
876 if (level == wc->root_level)
879 for (i = level + 1; i < BTRFS_MAX_LEVEL; i++) {
883 BUG_ON(i >= BTRFS_MAX_LEVEL);
885 node = wc->nodes[wc->active_node];
886 wc->nodes[wc->active_node] = NULL;
889 dest = wc->nodes[wc->active_node];
890 if (wc->active_node < wc->root_level ||
891 btrfs_root_refs(&root->root_item) > 0) {
892 BUG_ON(node->refs <= 1);
893 splice_shared_node(node, dest);
895 BUG_ON(node->refs < 2);
904 * 1 - if the root with id child_root_id is a child of root parent_root_id
905 * 0 - if the root child_root_id isn't a child of the root parent_root_id but
906 * has other root(s) as parent(s)
907 * 2 - if the root child_root_id doesn't have any parent roots
909 static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
912 struct btrfs_path path;
913 struct btrfs_key key;
914 struct extent_buffer *leaf;
918 btrfs_init_path(&path);
920 key.objectid = parent_root_id;
921 key.type = BTRFS_ROOT_REF_KEY;
922 key.offset = child_root_id;
923 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
927 btrfs_release_path(&path);
931 key.objectid = child_root_id;
932 key.type = BTRFS_ROOT_BACKREF_KEY;
934 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
940 leaf = path.nodes[0];
941 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
942 ret = btrfs_next_leaf(root->fs_info->tree_root, &path);
945 leaf = path.nodes[0];
948 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
949 if (key.objectid != child_root_id ||
950 key.type != BTRFS_ROOT_BACKREF_KEY)
955 if (key.offset == parent_root_id) {
956 btrfs_release_path(&path);
963 btrfs_release_path(&path);
966 return has_parent ? 0 : 2;
969 static int process_dir_item(struct btrfs_root *root,
970 struct extent_buffer *eb,
971 int slot, struct btrfs_key *key,
972 struct shared_node *active_node)
982 struct btrfs_dir_item *di;
983 struct inode_record *rec;
984 struct cache_tree *root_cache;
985 struct cache_tree *inode_cache;
986 struct btrfs_key location;
987 char namebuf[BTRFS_NAME_LEN];
989 root_cache = &active_node->root_cache;
990 inode_cache = &active_node->inode_cache;
991 rec = active_node->current;
992 rec->found_dir_item = 1;
994 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
995 total = btrfs_item_size_nr(eb, slot);
996 while (cur < total) {
998 btrfs_dir_item_key_to_cpu(eb, di, &location);
999 name_len = btrfs_dir_name_len(eb, di);
1000 data_len = btrfs_dir_data_len(eb, di);
1001 filetype = btrfs_dir_type(eb, di);
1003 rec->found_size += name_len;
1004 if (name_len <= BTRFS_NAME_LEN) {
1008 len = BTRFS_NAME_LEN;
1009 error = REF_ERR_NAME_TOO_LONG;
1011 read_extent_buffer(eb, namebuf, (unsigned long)(di + 1), len);
1013 if (location.type == BTRFS_INODE_ITEM_KEY) {
1014 add_inode_backref(inode_cache, location.objectid,
1015 key->objectid, key->offset, namebuf,
1016 len, filetype, key->type, error);
1017 } else if (location.type == BTRFS_ROOT_ITEM_KEY) {
1018 add_inode_backref(root_cache, location.objectid,
1019 key->objectid, key->offset,
1020 namebuf, len, filetype,
1023 fprintf(stderr, "invalid location in dir item %u\n",
1025 add_inode_backref(inode_cache, BTRFS_MULTIPLE_OBJECTIDS,
1026 key->objectid, key->offset, namebuf,
1027 len, filetype, key->type, error);
1030 len = sizeof(*di) + name_len + data_len;
1031 di = (struct btrfs_dir_item *)((char *)di + len);
1034 if (key->type == BTRFS_DIR_INDEX_KEY && nritems > 1)
1035 rec->errors |= I_ERR_DUP_DIR_INDEX;
1040 static int process_inode_ref(struct extent_buffer *eb,
1041 int slot, struct btrfs_key *key,
1042 struct shared_node *active_node)
1050 struct cache_tree *inode_cache;
1051 struct btrfs_inode_ref *ref;
1052 char namebuf[BTRFS_NAME_LEN];
1054 inode_cache = &active_node->inode_cache;
1056 ref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1057 total = btrfs_item_size_nr(eb, slot);
1058 while (cur < total) {
1059 name_len = btrfs_inode_ref_name_len(eb, ref);
1060 index = btrfs_inode_ref_index(eb, ref);
1061 if (name_len <= BTRFS_NAME_LEN) {
1065 len = BTRFS_NAME_LEN;
1066 error = REF_ERR_NAME_TOO_LONG;
1068 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
1069 add_inode_backref(inode_cache, key->objectid, key->offset,
1070 index, namebuf, len, 0, key->type, error);
1072 len = sizeof(*ref) + name_len;
1073 ref = (struct btrfs_inode_ref *)((char *)ref + len);
1079 static int process_inode_extref(struct extent_buffer *eb,
1080 int slot, struct btrfs_key *key,
1081 struct shared_node *active_node)
1090 struct cache_tree *inode_cache;
1091 struct btrfs_inode_extref *extref;
1092 char namebuf[BTRFS_NAME_LEN];
1094 inode_cache = &active_node->inode_cache;
1096 extref = btrfs_item_ptr(eb, slot, struct btrfs_inode_extref);
1097 total = btrfs_item_size_nr(eb, slot);
1098 while (cur < total) {
1099 name_len = btrfs_inode_extref_name_len(eb, extref);
1100 index = btrfs_inode_extref_index(eb, extref);
1101 parent = btrfs_inode_extref_parent(eb, extref);
1102 if (name_len <= BTRFS_NAME_LEN) {
1106 len = BTRFS_NAME_LEN;
1107 error = REF_ERR_NAME_TOO_LONG;
1109 read_extent_buffer(eb, namebuf,
1110 (unsigned long)(extref + 1), len);
1111 add_inode_backref(inode_cache, key->objectid, parent,
1112 index, namebuf, len, 0, key->type, error);
1114 len = sizeof(*extref) + name_len;
1115 extref = (struct btrfs_inode_extref *)((char *)extref + len);
1122 static int count_csum_range(struct btrfs_root *root, u64 start,
1123 u64 len, u64 *found)
1125 struct btrfs_key key;
1126 struct btrfs_path path;
1127 struct extent_buffer *leaf;
1132 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
1134 btrfs_init_path(&path);
1136 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1138 key.type = BTRFS_EXTENT_CSUM_KEY;
1140 ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
1144 if (ret > 0 && path.slots[0] > 0) {
1145 leaf = path.nodes[0];
1146 btrfs_item_key_to_cpu(leaf, &key, path.slots[0] - 1);
1147 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
1148 key.type == BTRFS_EXTENT_CSUM_KEY)
1153 leaf = path.nodes[0];
1154 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1155 ret = btrfs_next_leaf(root->fs_info->csum_root, &path);
1160 leaf = path.nodes[0];
1163 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1164 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1165 key.type != BTRFS_EXTENT_CSUM_KEY)
1168 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1169 if (key.offset >= start + len)
1172 if (key.offset > start)
1175 size = btrfs_item_size_nr(leaf, path.slots[0]);
1176 csum_end = key.offset + (size / csum_size) * root->sectorsize;
1177 if (csum_end > start) {
1178 size = min(csum_end - start, len);
1187 btrfs_release_path(&path);
1193 static int process_file_extent(struct btrfs_root *root,
1194 struct extent_buffer *eb,
1195 int slot, struct btrfs_key *key,
1196 struct shared_node *active_node)
1198 struct inode_record *rec;
1199 struct btrfs_file_extent_item *fi;
1201 u64 disk_bytenr = 0;
1202 u64 extent_offset = 0;
1203 u64 mask = root->sectorsize - 1;
1207 rec = active_node->current;
1208 BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1209 rec->found_file_extent = 1;
1211 if (rec->extent_start == (u64)-1) {
1212 rec->extent_start = key->offset;
1213 rec->extent_end = key->offset;
1216 if (rec->extent_end > key->offset)
1217 rec->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1218 else if (rec->extent_end < key->offset &&
1219 rec->extent_end < rec->first_extent_gap)
1220 rec->first_extent_gap = rec->extent_end;
1222 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
1223 extent_type = btrfs_file_extent_type(eb, fi);
1225 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1226 num_bytes = btrfs_file_extent_inline_len(eb, slot, fi);
1228 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1229 rec->found_size += num_bytes;
1230 num_bytes = (num_bytes + mask) & ~mask;
1231 } else if (extent_type == BTRFS_FILE_EXTENT_REG ||
1232 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1233 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1234 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1235 extent_offset = btrfs_file_extent_offset(eb, fi);
1236 if (num_bytes == 0 || (num_bytes & mask))
1237 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1238 if (num_bytes + extent_offset >
1239 btrfs_file_extent_ram_bytes(eb, fi))
1240 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1241 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC &&
1242 (btrfs_file_extent_compression(eb, fi) ||
1243 btrfs_file_extent_encryption(eb, fi) ||
1244 btrfs_file_extent_other_encoding(eb, fi)))
1245 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1246 if (disk_bytenr > 0)
1247 rec->found_size += num_bytes;
1249 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1251 rec->extent_end = key->offset + num_bytes;
1253 if (disk_bytenr > 0) {
1255 if (btrfs_file_extent_compression(eb, fi))
1256 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1258 disk_bytenr += extent_offset;
1260 ret = count_csum_range(root, disk_bytenr, num_bytes, &found);
1263 if (extent_type == BTRFS_FILE_EXTENT_REG) {
1265 rec->found_csum_item = 1;
1266 if (found < num_bytes)
1267 rec->some_csum_missing = 1;
1268 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1270 rec->errors |= I_ERR_ODD_CSUM_ITEM;
1276 static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
1277 struct walk_control *wc)
1279 struct btrfs_key key;
1283 struct cache_tree *inode_cache;
1284 struct shared_node *active_node;
1286 if (wc->root_level == wc->active_node &&
1287 btrfs_root_refs(&root->root_item) == 0)
1290 active_node = wc->nodes[wc->active_node];
1291 inode_cache = &active_node->inode_cache;
1292 nritems = btrfs_header_nritems(eb);
1293 for (i = 0; i < nritems; i++) {
1294 btrfs_item_key_to_cpu(eb, &key, i);
1296 if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
1298 if (key.type == BTRFS_ORPHAN_ITEM_KEY)
1301 if (active_node->current == NULL ||
1302 active_node->current->ino < key.objectid) {
1303 if (active_node->current) {
1304 active_node->current->checked = 1;
1305 maybe_free_inode_rec(inode_cache,
1306 active_node->current);
1308 active_node->current = get_inode_rec(inode_cache,
1312 case BTRFS_DIR_ITEM_KEY:
1313 case BTRFS_DIR_INDEX_KEY:
1314 ret = process_dir_item(root, eb, i, &key, active_node);
1316 case BTRFS_INODE_REF_KEY:
1317 ret = process_inode_ref(eb, i, &key, active_node);
1319 case BTRFS_INODE_EXTREF_KEY:
1320 ret = process_inode_extref(eb, i, &key, active_node);
1322 case BTRFS_INODE_ITEM_KEY:
1323 ret = process_inode_item(eb, i, &key, active_node);
1325 case BTRFS_EXTENT_DATA_KEY:
1326 ret = process_file_extent(root, eb, i, &key,
1336 static void reada_walk_down(struct btrfs_root *root,
1337 struct extent_buffer *node, int slot)
1346 level = btrfs_header_level(node);
1350 nritems = btrfs_header_nritems(node);
1351 blocksize = btrfs_level_size(root, level - 1);
1352 for (i = slot; i < nritems; i++) {
1353 bytenr = btrfs_node_blockptr(node, i);
1354 ptr_gen = btrfs_node_ptr_generation(node, i);
1355 readahead_tree_block(root, bytenr, blocksize, ptr_gen);
1360 * Check the child node/leaf by the following condition:
1361 * 1. the first item key of the node/leaf should be the same with the one
1363 * 2. block in parent node should match the child node/leaf.
1364 * 3. generation of parent node and child's header should be consistent.
1366 * Or the child node/leaf pointed by the key in parent is not valid.
1368 * We hope to check leaf owner too, but since subvol may share leaves,
1369 * which makes leaf owner check not so strong, key check should be
1370 * sufficient enough for that case.
1372 static int check_child_node(struct btrfs_root *root,
1373 struct extent_buffer *parent, int slot,
1374 struct extent_buffer *child)
1376 struct btrfs_key parent_key;
1377 struct btrfs_key child_key;
1380 btrfs_node_key_to_cpu(parent, &parent_key, slot);
1381 if (btrfs_header_level(child) == 0)
1382 btrfs_item_key_to_cpu(child, &child_key, 0);
1384 btrfs_node_key_to_cpu(child, &child_key, 0);
1386 if (memcmp(&parent_key, &child_key, sizeof(parent_key))) {
1389 "Wrong key of child node/leaf, wanted: (%llu, %u, %llu), have: (%llu, %u, %llu)\n",
1390 parent_key.objectid, parent_key.type, parent_key.offset,
1391 child_key.objectid, child_key.type, child_key.offset);
1393 if (btrfs_header_bytenr(child) != btrfs_node_blockptr(parent, slot)) {
1395 fprintf(stderr, "Wrong block of child node/leaf, wanted: %llu, have: %llu\n",
1396 btrfs_node_blockptr(parent, slot),
1397 btrfs_header_bytenr(child));
1399 if (btrfs_node_ptr_generation(parent, slot) !=
1400 btrfs_header_generation(child)) {
1402 fprintf(stderr, "Wrong generation of child node/leaf, wanted: %llu, have: %llu\n",
1403 btrfs_header_generation(child),
1404 btrfs_node_ptr_generation(parent, slot));
1409 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
1410 struct walk_control *wc, int *level)
1412 enum btrfs_tree_block_status status;
1415 struct extent_buffer *next;
1416 struct extent_buffer *cur;
1421 WARN_ON(*level < 0);
1422 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1423 ret = btrfs_lookup_extent_info(NULL, root,
1424 path->nodes[*level]->start,
1425 *level, 1, &refs, NULL);
1432 ret = enter_shared_node(root, path->nodes[*level]->start,
1440 while (*level >= 0) {
1441 WARN_ON(*level < 0);
1442 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1443 cur = path->nodes[*level];
1445 if (btrfs_header_level(cur) != *level)
1448 if (path->slots[*level] >= btrfs_header_nritems(cur))
1451 ret = process_one_leaf(root, cur, wc);
1456 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1457 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1458 blocksize = btrfs_level_size(root, *level - 1);
1459 ret = btrfs_lookup_extent_info(NULL, root, bytenr, *level - 1,
1465 ret = enter_shared_node(root, bytenr, refs,
1468 path->slots[*level]++;
1473 next = btrfs_find_tree_block(root, bytenr, blocksize);
1474 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
1475 free_extent_buffer(next);
1476 reada_walk_down(root, cur, path->slots[*level]);
1477 next = read_tree_block(root, bytenr, blocksize,
1480 struct btrfs_key node_key;
1482 btrfs_node_key_to_cpu(path->nodes[*level],
1484 path->slots[*level]);
1485 btrfs_add_corrupt_extent_record(root->fs_info,
1487 path->nodes[*level]->start,
1488 root->leafsize, *level);
1494 ret = check_child_node(root, cur, path->slots[*level], next);
1500 if (btrfs_is_leaf(next))
1501 status = btrfs_check_leaf(root, NULL, next);
1503 status = btrfs_check_node(root, NULL, next);
1504 if (status != BTRFS_TREE_BLOCK_CLEAN) {
1505 free_extent_buffer(next);
1510 *level = *level - 1;
1511 free_extent_buffer(path->nodes[*level]);
1512 path->nodes[*level] = next;
1513 path->slots[*level] = 0;
1516 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
1520 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
1521 struct walk_control *wc, int *level)
1524 struct extent_buffer *leaf;
1526 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1527 leaf = path->nodes[i];
1528 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
1533 free_extent_buffer(path->nodes[*level]);
1534 path->nodes[*level] = NULL;
1535 BUG_ON(*level > wc->active_node);
1536 if (*level == wc->active_node)
1537 leave_shared_node(root, wc, *level);
1544 static int check_root_dir(struct inode_record *rec)
1546 struct inode_backref *backref;
1549 if (!rec->found_inode_item || rec->errors)
1551 if (rec->nlink != 1 || rec->found_link != 0)
1553 if (list_empty(&rec->backrefs))
1555 backref = list_entry(rec->backrefs.next, struct inode_backref, list);
1556 if (!backref->found_inode_ref)
1558 if (backref->index != 0 || backref->namelen != 2 ||
1559 memcmp(backref->name, "..", 2))
1561 if (backref->found_dir_index || backref->found_dir_item)
1568 static int repair_inode_isize(struct btrfs_trans_handle *trans,
1569 struct btrfs_root *root, struct btrfs_path *path,
1570 struct inode_record *rec)
1572 struct btrfs_inode_item *ei;
1573 struct btrfs_key key;
1576 key.objectid = rec->ino;
1577 key.type = BTRFS_INODE_ITEM_KEY;
1578 key.offset = (u64)-1;
1580 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1584 if (!path->slots[0]) {
1591 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1592 if (key.objectid != rec->ino) {
1597 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1598 struct btrfs_inode_item);
1599 btrfs_set_inode_size(path->nodes[0], ei, rec->found_size);
1600 btrfs_mark_buffer_dirty(path->nodes[0]);
1601 rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
1602 printf("reset isize for dir %Lu root %Lu\n", rec->ino,
1603 root->root_key.objectid);
1605 btrfs_release_path(path);
1609 static int repair_inode_orphan_item(struct btrfs_trans_handle *trans,
1610 struct btrfs_root *root,
1611 struct btrfs_path *path,
1612 struct inode_record *rec)
1616 ret = btrfs_add_orphan_item(trans, root, path, rec->ino);
1617 btrfs_release_path(path);
1619 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
1623 static int add_missing_dir_index(struct btrfs_root *root,
1624 struct cache_tree *inode_cache,
1625 struct inode_record *rec,
1626 struct inode_backref *backref)
1628 struct btrfs_path *path;
1629 struct btrfs_trans_handle *trans;
1630 struct btrfs_dir_item *dir_item;
1631 struct extent_buffer *leaf;
1632 struct btrfs_key key;
1633 struct btrfs_disk_key disk_key;
1634 struct inode_record *dir_rec;
1635 unsigned long name_ptr;
1636 u32 data_size = sizeof(*dir_item) + backref->namelen;
1639 path = btrfs_alloc_path();
1643 trans = btrfs_start_transaction(root, 1);
1644 if (IS_ERR(trans)) {
1645 btrfs_free_path(path);
1646 return PTR_ERR(trans);
1649 fprintf(stderr, "repairing missing dir index item for inode %llu\n",
1650 (unsigned long long)rec->ino);
1651 key.objectid = backref->dir;
1652 key.type = BTRFS_DIR_INDEX_KEY;
1653 key.offset = backref->index;
1655 ret = btrfs_insert_empty_item(trans, root, path, &key, data_size);
1658 leaf = path->nodes[0];
1659 dir_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
1661 disk_key.objectid = cpu_to_le64(rec->ino);
1662 disk_key.type = BTRFS_INODE_ITEM_KEY;
1663 disk_key.offset = 0;
1665 btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
1666 btrfs_set_dir_type(leaf, dir_item, imode_to_type(rec->imode));
1667 btrfs_set_dir_data_len(leaf, dir_item, 0);
1668 btrfs_set_dir_name_len(leaf, dir_item, backref->namelen);
1669 name_ptr = (unsigned long)(dir_item + 1);
1670 write_extent_buffer(leaf, backref->name, name_ptr, backref->namelen);
1671 btrfs_mark_buffer_dirty(leaf);
1672 btrfs_free_path(path);
1673 btrfs_commit_transaction(trans, root);
1675 backref->found_dir_index = 1;
1676 dir_rec = get_inode_rec(inode_cache, backref->dir, 0);
1679 dir_rec->found_size += backref->namelen;
1680 if (dir_rec->found_size == dir_rec->isize &&
1681 (dir_rec->errors & I_ERR_DIR_ISIZE_WRONG))
1682 dir_rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
1683 if (dir_rec->found_size != dir_rec->isize)
1684 dir_rec->errors |= I_ERR_DIR_ISIZE_WRONG;
1689 static int delete_dir_index(struct btrfs_root *root,
1690 struct cache_tree *inode_cache,
1691 struct inode_record *rec,
1692 struct inode_backref *backref)
1694 struct btrfs_trans_handle *trans;
1695 struct btrfs_dir_item *di;
1696 struct btrfs_path *path;
1699 path = btrfs_alloc_path();
1703 trans = btrfs_start_transaction(root, 1);
1704 if (IS_ERR(trans)) {
1705 btrfs_free_path(path);
1706 return PTR_ERR(trans);
1710 fprintf(stderr, "Deleting bad dir index [%llu,%u,%llu] root %llu\n",
1711 (unsigned long long)backref->dir,
1712 BTRFS_DIR_INDEX_KEY, (unsigned long long)backref->index,
1713 (unsigned long long)root->objectid);
1715 di = btrfs_lookup_dir_index(trans, root, path, backref->dir,
1716 backref->name, backref->namelen,
1717 backref->index, -1);
1720 btrfs_free_path(path);
1721 btrfs_commit_transaction(trans, root);
1728 ret = btrfs_del_item(trans, root, path);
1730 ret = btrfs_delete_one_dir_name(trans, root, path, di);
1732 btrfs_free_path(path);
1733 btrfs_commit_transaction(trans, root);
1737 static int create_inode_item(struct btrfs_root *root,
1738 struct inode_record *rec,
1739 struct inode_backref *backref, int root_dir)
1741 struct btrfs_trans_handle *trans;
1742 struct btrfs_inode_item inode_item;
1743 time_t now = time(NULL);
1746 trans = btrfs_start_transaction(root, 1);
1747 if (IS_ERR(trans)) {
1748 ret = PTR_ERR(trans);
1752 fprintf(stderr, "root %llu inode %llu recreating inode item, this may "
1753 "be incomplete, please check permissions and content after "
1754 "the fsck completes.\n", (unsigned long long)root->objectid,
1755 (unsigned long long)rec->ino);
1757 memset(&inode_item, 0, sizeof(inode_item));
1758 btrfs_set_stack_inode_generation(&inode_item, trans->transid);
1760 btrfs_set_stack_inode_nlink(&inode_item, 1);
1762 btrfs_set_stack_inode_nlink(&inode_item, rec->found_link);
1763 btrfs_set_stack_inode_nbytes(&inode_item, rec->found_size);
1764 if (rec->found_dir_item) {
1765 if (rec->found_file_extent)
1766 fprintf(stderr, "root %llu inode %llu has both a dir "
1767 "item and extents, unsure if it is a dir or a "
1768 "regular file so setting it as a directory\n",
1769 (unsigned long long)root->objectid,
1770 (unsigned long long)rec->ino);
1771 btrfs_set_stack_inode_mode(&inode_item, S_IFDIR | 0755);
1772 btrfs_set_stack_inode_size(&inode_item, rec->found_size);
1773 } else if (!rec->found_dir_item) {
1774 btrfs_set_stack_inode_size(&inode_item, rec->extent_end);
1775 btrfs_set_stack_inode_mode(&inode_item, S_IFREG | 0755);
1777 btrfs_set_stack_timespec_sec(&inode_item.atime, now);
1778 btrfs_set_stack_timespec_nsec(&inode_item.atime, 0);
1779 btrfs_set_stack_timespec_sec(&inode_item.ctime, now);
1780 btrfs_set_stack_timespec_nsec(&inode_item.ctime, 0);
1781 btrfs_set_stack_timespec_sec(&inode_item.mtime, now);
1782 btrfs_set_stack_timespec_nsec(&inode_item.mtime, 0);
1783 btrfs_set_stack_timespec_sec(&inode_item.otime, 0);
1784 btrfs_set_stack_timespec_nsec(&inode_item.otime, 0);
1786 ret = btrfs_insert_inode(trans, root, rec->ino, &inode_item);
1788 btrfs_commit_transaction(trans, root);
1792 static int repair_inode_backrefs(struct btrfs_root *root,
1793 struct inode_record *rec,
1794 struct cache_tree *inode_cache,
1797 struct inode_backref *tmp, *backref;
1798 u64 root_dirid = btrfs_root_dirid(&root->root_item);
1802 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
1803 if (!delete && rec->ino == root_dirid) {
1804 if (!rec->found_inode_item) {
1805 ret = create_inode_item(root, rec, backref, 1);
1812 /* Index 0 for root dir's are special, don't mess with it */
1813 if (rec->ino == root_dirid && backref->index == 0)
1817 ((backref->found_dir_index && !backref->found_inode_ref) ||
1818 (backref->found_dir_index && backref->found_inode_ref &&
1819 (backref->errors & REF_ERR_INDEX_UNMATCH)))) {
1820 ret = delete_dir_index(root, inode_cache, rec, backref);
1824 list_del(&backref->list);
1828 if (!delete && !backref->found_dir_index &&
1829 backref->found_dir_item && backref->found_inode_ref) {
1830 ret = add_missing_dir_index(root, inode_cache, rec,
1835 if (backref->found_dir_item &&
1836 backref->found_dir_index &&
1837 backref->found_dir_index) {
1838 if (!backref->errors &&
1839 backref->found_inode_ref) {
1840 list_del(&backref->list);
1846 if (!delete && (!backref->found_dir_index &&
1847 !backref->found_dir_item &&
1848 backref->found_inode_ref)) {
1849 struct btrfs_trans_handle *trans;
1850 struct btrfs_key location;
1852 ret = check_dir_conflict(root, backref->name,
1858 * let nlink fixing routine to handle it,
1859 * which can do it better.
1864 location.objectid = rec->ino;
1865 location.type = BTRFS_INODE_ITEM_KEY;
1866 location.offset = 0;
1868 trans = btrfs_start_transaction(root, 1);
1869 if (IS_ERR(trans)) {
1870 ret = PTR_ERR(trans);
1873 fprintf(stderr, "adding missing dir index/item pair "
1875 (unsigned long long)rec->ino);
1876 ret = btrfs_insert_dir_item(trans, root, backref->name,
1878 backref->dir, &location,
1879 imode_to_type(rec->imode),
1882 btrfs_commit_transaction(trans, root);
1886 if (!delete && (backref->found_inode_ref &&
1887 backref->found_dir_index &&
1888 backref->found_dir_item &&
1889 !(backref->errors & REF_ERR_INDEX_UNMATCH) &&
1890 !rec->found_inode_item)) {
1891 ret = create_inode_item(root, rec, backref, 0);
1898 return ret ? ret : repaired;
1902 * To determine the file type for nlink/inode_item repair
1904 * Return 0 if file type is found and BTRFS_FT_* is stored into type.
1905 * Return -ENOENT if file type is not found.
1907 static int find_file_type(struct inode_record *rec, u8 *type)
1909 struct inode_backref *backref;
1911 /* For inode item recovered case */
1912 if (rec->found_inode_item) {
1913 *type = imode_to_type(rec->imode);
1917 list_for_each_entry(backref, &rec->backrefs, list) {
1918 if (backref->found_dir_index || backref->found_dir_item) {
1919 *type = backref->filetype;
1927 * To determine the file name for nlink repair
1929 * Return 0 if file name is found, set name and namelen.
1930 * Return -ENOENT if file name is not found.
1932 static int find_file_name(struct inode_record *rec,
1933 char *name, int *namelen)
1935 struct inode_backref *backref;
1937 list_for_each_entry(backref, &rec->backrefs, list) {
1938 if (backref->found_dir_index || backref->found_dir_item ||
1939 backref->found_inode_ref) {
1940 memcpy(name, backref->name, backref->namelen);
1941 *namelen = backref->namelen;
1948 /* Reset the nlink of the inode to the correct one */
1949 static int reset_nlink(struct btrfs_trans_handle *trans,
1950 struct btrfs_root *root,
1951 struct btrfs_path *path,
1952 struct inode_record *rec)
1954 struct inode_backref *backref;
1955 struct inode_backref *tmp;
1956 struct btrfs_key key;
1957 struct btrfs_inode_item *inode_item;
1960 /* We don't believe this either, reset it and iterate backref */
1961 rec->found_link = 0;
1963 /* Remove all backref including the valid ones */
1964 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
1965 ret = btrfs_unlink(trans, root, rec->ino, backref->dir,
1966 backref->index, backref->name,
1967 backref->namelen, 0);
1971 /* remove invalid backref, so it won't be added back */
1972 if (!(backref->found_dir_index &&
1973 backref->found_dir_item &&
1974 backref->found_inode_ref)) {
1975 list_del(&backref->list);
1982 /* Set nlink to 0 */
1983 key.objectid = rec->ino;
1984 key.type = BTRFS_INODE_ITEM_KEY;
1986 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1993 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1994 struct btrfs_inode_item);
1995 btrfs_set_inode_nlink(path->nodes[0], inode_item, 0);
1996 btrfs_mark_buffer_dirty(path->nodes[0]);
1997 btrfs_release_path(path);
2000 * Add back valid inode_ref/dir_item/dir_index,
2001 * add_link() will handle the nlink inc, so new nlink must be correct
2003 list_for_each_entry(backref, &rec->backrefs, list) {
2004 ret = btrfs_add_link(trans, root, rec->ino, backref->dir,
2005 backref->name, backref->namelen,
2006 backref->ref_type, &backref->index, 1);
2011 btrfs_release_path(path);
2015 static int repair_inode_nlinks(struct btrfs_trans_handle *trans,
2016 struct btrfs_root *root,
2017 struct btrfs_path *path,
2018 struct inode_record *rec)
2020 char *dir_name = "lost+found";
2021 char namebuf[BTRFS_NAME_LEN] = {0};
2026 int name_recovered = 0;
2027 int type_recovered = 0;
2031 * Get file name and type first before these invalid inode ref
2032 * are deleted by remove_all_invalid_backref()
2034 name_recovered = !find_file_name(rec, namebuf, &namelen);
2035 type_recovered = !find_file_type(rec, &type);
2037 if (!name_recovered) {
2038 printf("Can't get file name for inode %llu, using '%llu' as fallback\n",
2039 rec->ino, rec->ino);
2040 namelen = count_digits(rec->ino);
2041 sprintf(namebuf, "%llu", rec->ino);
2044 if (!type_recovered) {
2045 printf("Can't get file type for inode %llu, using FILE as fallback\n",
2047 type = BTRFS_FT_REG_FILE;
2051 ret = reset_nlink(trans, root, path, rec);
2054 "Failed to reset nlink for inode %llu: %s\n",
2055 rec->ino, strerror(-ret));
2059 if (rec->found_link == 0) {
2060 lost_found_ino = root->highest_inode;
2061 if (lost_found_ino >= BTRFS_LAST_FREE_OBJECTID) {
2066 ret = btrfs_mkdir(trans, root, dir_name, strlen(dir_name),
2067 BTRFS_FIRST_FREE_OBJECTID, &lost_found_ino,
2070 fprintf(stderr, "Failed to create '%s' dir: %s",
2071 dir_name, strerror(-ret));
2074 ret = btrfs_add_link(trans, root, rec->ino, lost_found_ino,
2075 namebuf, namelen, type, NULL, 1);
2076 if (ret == -EEXIST) {
2078 * Conflicting file name, add ".INO" as suffix * +1 for '.'
2080 if (namelen + count_digits(rec->ino) + 1 >
2085 snprintf(namebuf + namelen, BTRFS_NAME_LEN - namelen,
2087 namelen += count_digits(rec->ino) + 1;
2088 ret = btrfs_add_link(trans, root, rec->ino,
2089 lost_found_ino, namebuf,
2090 namelen, type, NULL, 1);
2094 "Failed to link the inode %llu to %s dir: %s",
2095 rec->ino, dir_name, strerror(-ret));
2099 * Just increase the found_link, don't actually add the
2100 * backref. This will make things easier and this inode
2101 * record will be freed after the repair is done.
2102 * So fsck will not report problem about this inode.
2105 printf("Moving file '%.*s' to '%s' dir since it has no valid backref\n",
2106 namelen, namebuf, dir_name);
2108 rec->errors &= ~I_ERR_LINK_COUNT_WRONG;
2109 printf("Fixed the nlink of inode %llu\n", rec->ino);
2111 btrfs_release_path(path);
2116 * Check if there is any normal(reg or prealloc) file extent for given
2118 * This is used to determine the file type when neither its dir_index/item or
2119 * inode_item exists.
2121 * This will *NOT* report error, if any error happens, just consider it does
2122 * not have any normal file extent.
2124 static int find_normal_file_extent(struct btrfs_root *root, u64 ino)
2126 struct btrfs_path *path;
2127 struct btrfs_key key;
2128 struct btrfs_key found_key;
2129 struct btrfs_file_extent_item *fi;
2133 path = btrfs_alloc_path();
2137 key.type = BTRFS_EXTENT_DATA_KEY;
2140 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2145 if (ret && path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2146 ret = btrfs_next_leaf(root, path);
2153 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2155 if (found_key.objectid != ino ||
2156 found_key.type != BTRFS_EXTENT_DATA_KEY)
2158 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
2159 struct btrfs_file_extent_item);
2160 type = btrfs_file_extent_type(path->nodes[0], fi);
2161 if (type != BTRFS_FILE_EXTENT_INLINE) {
2167 btrfs_free_path(path);
2171 static u32 btrfs_type_to_imode(u8 type)
2173 static u32 imode_by_btrfs_type[] = {
2174 [BTRFS_FT_REG_FILE] = S_IFREG,
2175 [BTRFS_FT_DIR] = S_IFDIR,
2176 [BTRFS_FT_CHRDEV] = S_IFCHR,
2177 [BTRFS_FT_BLKDEV] = S_IFBLK,
2178 [BTRFS_FT_FIFO] = S_IFIFO,
2179 [BTRFS_FT_SOCK] = S_IFSOCK,
2180 [BTRFS_FT_SYMLINK] = S_IFLNK,
2183 return imode_by_btrfs_type[(type)];
2186 static int repair_inode_no_item(struct btrfs_trans_handle *trans,
2187 struct btrfs_root *root,
2188 struct btrfs_path *path,
2189 struct inode_record *rec)
2193 int type_recovered = 0;
2198 * 1. salvage data from existing file extent and
2199 * punch hole to keep fi ext consistent.
2200 * 2. salvage data from extent tree
2202 printf("Trying to rebuild inode:%llu\n", rec->ino);
2204 type_recovered = !find_file_type(rec, &filetype);
2207 * Try to determine inode type if type not found.
2209 * For found regular file extent, it must be FILE.
2210 * For found dir_item/index, it must be DIR.
2212 * For undetermined one, use FILE as fallback.
2215 * 1. If found extent belong to it in extent tree, it must be FILE
2216 * Need extra hook in extent tree scan.
2217 * 2. If found backref(inode_index/item is already handled) to it,
2219 * Need new inode-inode ref structure to allow search for that.
2221 if (!type_recovered) {
2222 if (rec->found_file_extent &&
2223 find_normal_file_extent(root, rec->ino)) {
2225 filetype = BTRFS_FT_REG_FILE;
2226 } else if (rec->found_dir_item) {
2228 filetype = BTRFS_FT_DIR;
2230 printf("Can't determint the filetype for inode %llu, assume it is a normal file\n",
2233 filetype = BTRFS_FT_REG_FILE;
2237 ret = btrfs_new_inode(trans, root, rec->ino,
2238 mode | btrfs_type_to_imode(filetype));
2243 * Here inode rebuild is done, we only rebuild the inode item,
2244 * don't repair the nlink(like move to lost+found).
2245 * That is the job of nlink repair.
2247 * We just fill the record and return
2249 rec->found_dir_item = 1;
2250 rec->imode = mode | btrfs_type_to_imode(filetype);
2252 rec->errors &= ~I_ERR_NO_INODE_ITEM;
2253 /* Ensure the inode_nlinks repair function will be called */
2254 rec->errors |= I_ERR_LINK_COUNT_WRONG;
2259 static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
2261 struct btrfs_trans_handle *trans;
2262 struct btrfs_path *path;
2265 if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG |
2266 I_ERR_NO_ORPHAN_ITEM |
2267 I_ERR_LINK_COUNT_WRONG |
2268 I_ERR_NO_INODE_ITEM)))
2271 path = btrfs_alloc_path();
2276 * For nlink repair, it may create a dir and add link, so
2277 * 2 for parent(256)'s dir_index and dir_item
2278 * 2 for lost+found dir's inode_item and inode_ref
2279 * 1 for the new inode_ref of the file
2280 * 2 for lost+found dir's dir_index and dir_item for the file
2282 trans = btrfs_start_transaction(root, 7);
2283 if (IS_ERR(trans)) {
2284 btrfs_free_path(path);
2285 return PTR_ERR(trans);
2288 if (rec->errors & I_ERR_NO_INODE_ITEM)
2289 ret = repair_inode_no_item(trans, root, path, rec);
2290 if (!ret && rec->errors & I_ERR_DIR_ISIZE_WRONG)
2291 ret = repair_inode_isize(trans, root, path, rec);
2292 if (!ret && rec->errors & I_ERR_NO_ORPHAN_ITEM)
2293 ret = repair_inode_orphan_item(trans, root, path, rec);
2294 if (!ret && rec->errors & I_ERR_LINK_COUNT_WRONG)
2295 ret = repair_inode_nlinks(trans, root, path, rec);
2296 btrfs_commit_transaction(trans, root);
2297 btrfs_free_path(path);
2301 static int check_inode_recs(struct btrfs_root *root,
2302 struct cache_tree *inode_cache)
2304 struct cache_extent *cache;
2305 struct ptr_node *node;
2306 struct inode_record *rec;
2307 struct inode_backref *backref;
2312 u64 root_dirid = btrfs_root_dirid(&root->root_item);
2314 if (btrfs_root_refs(&root->root_item) == 0) {
2315 if (!cache_tree_empty(inode_cache))
2316 fprintf(stderr, "warning line %d\n", __LINE__);
2321 * We need to record the highest inode number for later 'lost+found'
2323 * We must select a ino not used/refered by any existing inode, or
2324 * 'lost+found' ino may be a missing ino in a corrupted leaf,
2325 * this may cause 'lost+found' dir has wrong nlinks.
2327 cache = last_cache_extent(inode_cache);
2329 node = container_of(cache, struct ptr_node, cache);
2331 if (rec->ino > root->highest_inode)
2332 root->highest_inode = rec->ino;
2336 * We need to repair backrefs first because we could change some of the
2337 * errors in the inode recs.
2339 * We also need to go through and delete invalid backrefs first and then
2340 * add the correct ones second. We do this because we may get EEXIST
2341 * when adding back the correct index because we hadn't yet deleted the
2344 * For example, if we were missing a dir index then the directories
2345 * isize would be wrong, so if we fixed the isize to what we thought it
2346 * would be and then fixed the backref we'd still have a invalid fs, so
2347 * we need to add back the dir index and then check to see if the isize
2352 if (stage == 3 && !err)
2355 cache = search_cache_extent(inode_cache, 0);
2356 while (repair && cache) {
2357 node = container_of(cache, struct ptr_node, cache);
2359 cache = next_cache_extent(cache);
2361 /* Need to free everything up and rescan */
2363 remove_cache_extent(inode_cache, &node->cache);
2365 free_inode_rec(rec);
2369 if (list_empty(&rec->backrefs))
2372 ret = repair_inode_backrefs(root, rec, inode_cache,
2386 rec = get_inode_rec(inode_cache, root_dirid, 0);
2388 ret = check_root_dir(rec);
2390 fprintf(stderr, "root %llu root dir %llu error\n",
2391 (unsigned long long)root->root_key.objectid,
2392 (unsigned long long)root_dirid);
2393 print_inode_error(root, rec);
2398 struct btrfs_trans_handle *trans;
2400 trans = btrfs_start_transaction(root, 1);
2401 if (IS_ERR(trans)) {
2402 err = PTR_ERR(trans);
2407 "root %llu missing its root dir, recreating\n",
2408 (unsigned long long)root->objectid);
2410 ret = btrfs_make_root_dir(trans, root, root_dirid);
2413 btrfs_commit_transaction(trans, root);
2417 fprintf(stderr, "root %llu root dir %llu not found\n",
2418 (unsigned long long)root->root_key.objectid,
2419 (unsigned long long)root_dirid);
2423 cache = search_cache_extent(inode_cache, 0);
2426 node = container_of(cache, struct ptr_node, cache);
2428 remove_cache_extent(inode_cache, &node->cache);
2430 if (rec->ino == root_dirid ||
2431 rec->ino == BTRFS_ORPHAN_OBJECTID) {
2432 free_inode_rec(rec);
2436 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
2437 ret = check_orphan_item(root, rec->ino);
2439 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
2440 if (can_free_inode_rec(rec)) {
2441 free_inode_rec(rec);
2446 if (!rec->found_inode_item)
2447 rec->errors |= I_ERR_NO_INODE_ITEM;
2448 if (rec->found_link != rec->nlink)
2449 rec->errors |= I_ERR_LINK_COUNT_WRONG;
2451 ret = try_repair_inode(root, rec);
2452 if (ret == 0 && can_free_inode_rec(rec)) {
2453 free_inode_rec(rec);
2459 if (!(repair && ret == 0))
2461 print_inode_error(root, rec);
2462 list_for_each_entry(backref, &rec->backrefs, list) {
2463 if (!backref->found_dir_item)
2464 backref->errors |= REF_ERR_NO_DIR_ITEM;
2465 if (!backref->found_dir_index)
2466 backref->errors |= REF_ERR_NO_DIR_INDEX;
2467 if (!backref->found_inode_ref)
2468 backref->errors |= REF_ERR_NO_INODE_REF;
2469 fprintf(stderr, "\tunresolved ref dir %llu index %llu"
2470 " namelen %u name %s filetype %d errors %x",
2471 (unsigned long long)backref->dir,
2472 (unsigned long long)backref->index,
2473 backref->namelen, backref->name,
2474 backref->filetype, backref->errors);
2475 print_ref_error(backref->errors);
2477 free_inode_rec(rec);
2479 return (error > 0) ? -1 : 0;
2482 static struct root_record *get_root_rec(struct cache_tree *root_cache,
2485 struct cache_extent *cache;
2486 struct root_record *rec = NULL;
2489 cache = lookup_cache_extent(root_cache, objectid, 1);
2491 rec = container_of(cache, struct root_record, cache);
2493 rec = calloc(1, sizeof(*rec));
2494 rec->objectid = objectid;
2495 INIT_LIST_HEAD(&rec->backrefs);
2496 rec->cache.start = objectid;
2497 rec->cache.size = 1;
2499 ret = insert_cache_extent(root_cache, &rec->cache);
2505 static struct root_backref *get_root_backref(struct root_record *rec,
2506 u64 ref_root, u64 dir, u64 index,
2507 const char *name, int namelen)
2509 struct root_backref *backref;
2511 list_for_each_entry(backref, &rec->backrefs, list) {
2512 if (backref->ref_root != ref_root || backref->dir != dir ||
2513 backref->namelen != namelen)
2515 if (memcmp(name, backref->name, namelen))
2520 backref = malloc(sizeof(*backref) + namelen + 1);
2521 memset(backref, 0, sizeof(*backref));
2522 backref->ref_root = ref_root;
2524 backref->index = index;
2525 backref->namelen = namelen;
2526 memcpy(backref->name, name, namelen);
2527 backref->name[namelen] = '\0';
2528 list_add_tail(&backref->list, &rec->backrefs);
2532 static void free_root_record(struct cache_extent *cache)
2534 struct root_record *rec;
2535 struct root_backref *backref;
2537 rec = container_of(cache, struct root_record, cache);
2538 while (!list_empty(&rec->backrefs)) {
2539 backref = list_entry(rec->backrefs.next,
2540 struct root_backref, list);
2541 list_del(&backref->list);
2548 FREE_EXTENT_CACHE_BASED_TREE(root_recs, free_root_record);
2550 static int add_root_backref(struct cache_tree *root_cache,
2551 u64 root_id, u64 ref_root, u64 dir, u64 index,
2552 const char *name, int namelen,
2553 int item_type, int errors)
2555 struct root_record *rec;
2556 struct root_backref *backref;
2558 rec = get_root_rec(root_cache, root_id);
2559 backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
2561 backref->errors |= errors;
2563 if (item_type != BTRFS_DIR_ITEM_KEY) {
2564 if (backref->found_dir_index || backref->found_back_ref ||
2565 backref->found_forward_ref) {
2566 if (backref->index != index)
2567 backref->errors |= REF_ERR_INDEX_UNMATCH;
2569 backref->index = index;
2573 if (item_type == BTRFS_DIR_ITEM_KEY) {
2574 if (backref->found_forward_ref)
2576 backref->found_dir_item = 1;
2577 } else if (item_type == BTRFS_DIR_INDEX_KEY) {
2578 backref->found_dir_index = 1;
2579 } else if (item_type == BTRFS_ROOT_REF_KEY) {
2580 if (backref->found_forward_ref)
2581 backref->errors |= REF_ERR_DUP_ROOT_REF;
2582 else if (backref->found_dir_item)
2584 backref->found_forward_ref = 1;
2585 } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
2586 if (backref->found_back_ref)
2587 backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
2588 backref->found_back_ref = 1;
2593 if (backref->found_forward_ref && backref->found_dir_item)
2594 backref->reachable = 1;
2598 static int merge_root_recs(struct btrfs_root *root,
2599 struct cache_tree *src_cache,
2600 struct cache_tree *dst_cache)
2602 struct cache_extent *cache;
2603 struct ptr_node *node;
2604 struct inode_record *rec;
2605 struct inode_backref *backref;
2608 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2609 free_inode_recs_tree(src_cache);
2614 cache = search_cache_extent(src_cache, 0);
2617 node = container_of(cache, struct ptr_node, cache);
2619 remove_cache_extent(src_cache, &node->cache);
2622 ret = is_child_root(root, root->objectid, rec->ino);
2628 list_for_each_entry(backref, &rec->backrefs, list) {
2629 BUG_ON(backref->found_inode_ref);
2630 if (backref->found_dir_item)
2631 add_root_backref(dst_cache, rec->ino,
2632 root->root_key.objectid, backref->dir,
2633 backref->index, backref->name,
2634 backref->namelen, BTRFS_DIR_ITEM_KEY,
2636 if (backref->found_dir_index)
2637 add_root_backref(dst_cache, rec->ino,
2638 root->root_key.objectid, backref->dir,
2639 backref->index, backref->name,
2640 backref->namelen, BTRFS_DIR_INDEX_KEY,
2644 free_inode_rec(rec);
2651 static int check_root_refs(struct btrfs_root *root,
2652 struct cache_tree *root_cache)
2654 struct root_record *rec;
2655 struct root_record *ref_root;
2656 struct root_backref *backref;
2657 struct cache_extent *cache;
2663 rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
2666 /* fixme: this can not detect circular references */
2669 cache = search_cache_extent(root_cache, 0);
2673 rec = container_of(cache, struct root_record, cache);
2674 cache = next_cache_extent(cache);
2676 if (rec->found_ref == 0)
2679 list_for_each_entry(backref, &rec->backrefs, list) {
2680 if (!backref->reachable)
2683 ref_root = get_root_rec(root_cache,
2685 if (ref_root->found_ref > 0)
2688 backref->reachable = 0;
2690 if (rec->found_ref == 0)
2696 cache = search_cache_extent(root_cache, 0);
2700 rec = container_of(cache, struct root_record, cache);
2701 cache = next_cache_extent(cache);
2703 if (rec->found_ref == 0 &&
2704 rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
2705 rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
2706 ret = check_orphan_item(root->fs_info->tree_root,
2712 * If we don't have a root item then we likely just have
2713 * a dir item in a snapshot for this root but no actual
2714 * ref key or anything so it's meaningless.
2716 if (!rec->found_root_item)
2719 fprintf(stderr, "fs tree %llu not referenced\n",
2720 (unsigned long long)rec->objectid);
2724 if (rec->found_ref > 0 && !rec->found_root_item)
2726 list_for_each_entry(backref, &rec->backrefs, list) {
2727 if (!backref->found_dir_item)
2728 backref->errors |= REF_ERR_NO_DIR_ITEM;
2729 if (!backref->found_dir_index)
2730 backref->errors |= REF_ERR_NO_DIR_INDEX;
2731 if (!backref->found_back_ref)
2732 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
2733 if (!backref->found_forward_ref)
2734 backref->errors |= REF_ERR_NO_ROOT_REF;
2735 if (backref->reachable && backref->errors)
2742 fprintf(stderr, "fs tree %llu refs %u %s\n",
2743 (unsigned long long)rec->objectid, rec->found_ref,
2744 rec->found_root_item ? "" : "not found");
2746 list_for_each_entry(backref, &rec->backrefs, list) {
2747 if (!backref->reachable)
2749 if (!backref->errors && rec->found_root_item)
2751 fprintf(stderr, "\tunresolved ref root %llu dir %llu"
2752 " index %llu namelen %u name %s errors %x\n",
2753 (unsigned long long)backref->ref_root,
2754 (unsigned long long)backref->dir,
2755 (unsigned long long)backref->index,
2756 backref->namelen, backref->name,
2758 print_ref_error(backref->errors);
2761 return errors > 0 ? 1 : 0;
2764 static int process_root_ref(struct extent_buffer *eb, int slot,
2765 struct btrfs_key *key,
2766 struct cache_tree *root_cache)
2772 struct btrfs_root_ref *ref;
2773 char namebuf[BTRFS_NAME_LEN];
2776 ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
2778 dirid = btrfs_root_ref_dirid(eb, ref);
2779 index = btrfs_root_ref_sequence(eb, ref);
2780 name_len = btrfs_root_ref_name_len(eb, ref);
2782 if (name_len <= BTRFS_NAME_LEN) {
2786 len = BTRFS_NAME_LEN;
2787 error = REF_ERR_NAME_TOO_LONG;
2789 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
2791 if (key->type == BTRFS_ROOT_REF_KEY) {
2792 add_root_backref(root_cache, key->offset, key->objectid, dirid,
2793 index, namebuf, len, key->type, error);
2795 add_root_backref(root_cache, key->objectid, key->offset, dirid,
2796 index, namebuf, len, key->type, error);
2801 static void free_corrupt_block(struct cache_extent *cache)
2803 struct btrfs_corrupt_block *corrupt;
2805 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
2809 FREE_EXTENT_CACHE_BASED_TREE(corrupt_blocks, free_corrupt_block);
2812 * Repair the btree of the given root.
2814 * The fix is to remove the node key in corrupt_blocks cache_tree.
2815 * and rebalance the tree.
2816 * After the fix, the btree should be writeable.
2818 static int repair_btree(struct btrfs_root *root,
2819 struct cache_tree *corrupt_blocks)
2821 struct btrfs_trans_handle *trans;
2822 struct btrfs_path *path;
2823 struct btrfs_corrupt_block *corrupt;
2824 struct cache_extent *cache;
2825 struct btrfs_key key;
2830 if (cache_tree_empty(corrupt_blocks))
2833 path = btrfs_alloc_path();
2837 trans = btrfs_start_transaction(root, 1);
2838 if (IS_ERR(trans)) {
2839 ret = PTR_ERR(trans);
2840 fprintf(stderr, "Error starting transaction: %s\n",
2844 cache = first_cache_extent(corrupt_blocks);
2846 corrupt = container_of(cache, struct btrfs_corrupt_block,
2848 level = corrupt->level;
2849 path->lowest_level = level;
2850 key.objectid = corrupt->key.objectid;
2851 key.type = corrupt->key.type;
2852 key.offset = corrupt->key.offset;
2855 * Here we don't want to do any tree balance, since it may
2856 * cause a balance with corrupted brother leaf/node,
2857 * so ins_len set to 0 here.
2858 * Balance will be done after all corrupt node/leaf is deleted.
2860 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2863 offset = btrfs_node_blockptr(path->nodes[level],
2864 path->slots[level]);
2866 /* Remove the ptr */
2867 ret = btrfs_del_ptr(trans, root, path, level,
2868 path->slots[level]);
2872 * Remove the corresponding extent
2873 * return value is not concerned.
2875 btrfs_release_path(path);
2876 ret = btrfs_free_extent(trans, root, offset, root->nodesize,
2877 0, root->root_key.objectid,
2879 cache = next_cache_extent(cache);
2882 /* Balance the btree using btrfs_search_slot() */
2883 cache = first_cache_extent(corrupt_blocks);
2885 corrupt = container_of(cache, struct btrfs_corrupt_block,
2887 memcpy(&key, &corrupt->key, sizeof(key));
2888 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2891 /* return will always >0 since it won't find the item */
2893 btrfs_release_path(path);
2894 cache = next_cache_extent(cache);
2897 btrfs_commit_transaction(trans, root);
2899 btrfs_free_path(path);
2903 static int check_fs_root(struct btrfs_root *root,
2904 struct cache_tree *root_cache,
2905 struct walk_control *wc)
2911 struct btrfs_path path;
2912 struct shared_node root_node;
2913 struct root_record *rec;
2914 struct btrfs_root_item *root_item = &root->root_item;
2915 struct cache_tree corrupt_blocks;
2916 enum btrfs_tree_block_status status;
2919 * Reuse the corrupt_block cache tree to record corrupted tree block
2921 * Unlike the usage in extent tree check, here we do it in a per
2922 * fs/subvol tree base.
2924 cache_tree_init(&corrupt_blocks);
2925 root->fs_info->corrupt_blocks = &corrupt_blocks;
2926 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2927 rec = get_root_rec(root_cache, root->root_key.objectid);
2928 if (btrfs_root_refs(root_item) > 0)
2929 rec->found_root_item = 1;
2932 btrfs_init_path(&path);
2933 memset(&root_node, 0, sizeof(root_node));
2934 cache_tree_init(&root_node.root_cache);
2935 cache_tree_init(&root_node.inode_cache);
2937 level = btrfs_header_level(root->node);
2938 memset(wc->nodes, 0, sizeof(wc->nodes));
2939 wc->nodes[level] = &root_node;
2940 wc->active_node = level;
2941 wc->root_level = level;
2943 /* We may not have checked the root block, lets do that now */
2944 if (btrfs_is_leaf(root->node))
2945 status = btrfs_check_leaf(root, NULL, root->node);
2947 status = btrfs_check_node(root, NULL, root->node);
2948 if (status != BTRFS_TREE_BLOCK_CLEAN)
2951 if (btrfs_root_refs(root_item) > 0 ||
2952 btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2953 path.nodes[level] = root->node;
2954 extent_buffer_get(root->node);
2955 path.slots[level] = 0;
2957 struct btrfs_key key;
2958 struct btrfs_disk_key found_key;
2960 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2961 level = root_item->drop_level;
2962 path.lowest_level = level;
2963 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
2966 btrfs_node_key(path.nodes[level], &found_key,
2968 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2969 sizeof(found_key)));
2973 wret = walk_down_tree(root, &path, wc, &level);
2979 wret = walk_up_tree(root, &path, wc, &level);
2986 btrfs_release_path(&path);
2988 if (!cache_tree_empty(&corrupt_blocks)) {
2989 struct cache_extent *cache;
2990 struct btrfs_corrupt_block *corrupt;
2992 printf("The following tree block(s) is corrupted in tree %llu:\n",
2993 root->root_key.objectid);
2994 cache = first_cache_extent(&corrupt_blocks);
2996 corrupt = container_of(cache,
2997 struct btrfs_corrupt_block,
2999 printf("\ttree block bytenr: %llu, level: %d, node key: (%llu, %u, %llu)\n",
3000 cache->start, corrupt->level,
3001 corrupt->key.objectid, corrupt->key.type,
3002 corrupt->key.offset);
3003 cache = next_cache_extent(cache);
3006 printf("Try to repair the btree for root %llu\n",
3007 root->root_key.objectid);
3008 ret = repair_btree(root, &corrupt_blocks);
3010 fprintf(stderr, "Failed to repair btree: %s\n",
3013 printf("Btree for root %llu is fixed\n",
3014 root->root_key.objectid);
3018 err = merge_root_recs(root, &root_node.root_cache, root_cache);
3022 if (root_node.current) {
3023 root_node.current->checked = 1;
3024 maybe_free_inode_rec(&root_node.inode_cache,
3028 err = check_inode_recs(root, &root_node.inode_cache);
3032 free_corrupt_blocks_tree(&corrupt_blocks);
3033 root->fs_info->corrupt_blocks = NULL;
3037 static int fs_root_objectid(u64 objectid)
3039 if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
3040 objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3042 return is_fstree(objectid);
3045 static int check_fs_roots(struct btrfs_root *root,
3046 struct cache_tree *root_cache)
3048 struct btrfs_path path;
3049 struct btrfs_key key;
3050 struct walk_control wc;
3051 struct extent_buffer *leaf, *tree_node;
3052 struct btrfs_root *tmp_root;
3053 struct btrfs_root *tree_root = root->fs_info->tree_root;
3058 * Just in case we made any changes to the extent tree that weren't
3059 * reflected into the free space cache yet.
3062 reset_cached_block_groups(root->fs_info);
3063 memset(&wc, 0, sizeof(wc));
3064 cache_tree_init(&wc.shared);
3065 btrfs_init_path(&path);
3070 key.type = BTRFS_ROOT_ITEM_KEY;
3071 ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
3076 tree_node = tree_root->node;
3078 if (tree_node != tree_root->node) {
3079 free_root_recs_tree(root_cache);
3080 btrfs_release_path(&path);
3083 leaf = path.nodes[0];
3084 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
3085 ret = btrfs_next_leaf(tree_root, &path);
3091 leaf = path.nodes[0];
3093 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
3094 if (key.type == BTRFS_ROOT_ITEM_KEY &&
3095 fs_root_objectid(key.objectid)) {
3096 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
3097 tmp_root = btrfs_read_fs_root_no_cache(
3098 root->fs_info, &key);
3100 key.offset = (u64)-1;
3101 tmp_root = btrfs_read_fs_root(
3102 root->fs_info, &key);
3104 if (IS_ERR(tmp_root)) {
3108 ret = check_fs_root(tmp_root, root_cache, &wc);
3109 if (ret == -EAGAIN) {
3110 free_root_recs_tree(root_cache);
3111 btrfs_release_path(&path);
3116 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
3117 btrfs_free_fs_root(tmp_root);
3118 } else if (key.type == BTRFS_ROOT_REF_KEY ||
3119 key.type == BTRFS_ROOT_BACKREF_KEY) {
3120 process_root_ref(leaf, path.slots[0], &key,
3127 btrfs_release_path(&path);
3129 free_extent_cache_tree(&wc.shared);
3130 if (!cache_tree_empty(&wc.shared))
3131 fprintf(stderr, "warning line %d\n", __LINE__);
3136 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
3138 struct list_head *cur = rec->backrefs.next;
3139 struct extent_backref *back;
3140 struct tree_backref *tback;
3141 struct data_backref *dback;
3145 while(cur != &rec->backrefs) {
3146 back = list_entry(cur, struct extent_backref, list);
3148 if (!back->found_extent_tree) {
3152 if (back->is_data) {
3153 dback = (struct data_backref *)back;
3154 fprintf(stderr, "Backref %llu %s %llu"
3155 " owner %llu offset %llu num_refs %lu"
3156 " not found in extent tree\n",
3157 (unsigned long long)rec->start,
3158 back->full_backref ?
3160 back->full_backref ?
3161 (unsigned long long)dback->parent:
3162 (unsigned long long)dback->root,
3163 (unsigned long long)dback->owner,
3164 (unsigned long long)dback->offset,
3165 (unsigned long)dback->num_refs);
3167 tback = (struct tree_backref *)back;
3168 fprintf(stderr, "Backref %llu parent %llu"
3169 " root %llu not found in extent tree\n",
3170 (unsigned long long)rec->start,
3171 (unsigned long long)tback->parent,
3172 (unsigned long long)tback->root);
3175 if (!back->is_data && !back->found_ref) {
3179 tback = (struct tree_backref *)back;
3180 fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
3181 (unsigned long long)rec->start,
3182 back->full_backref ? "parent" : "root",
3183 back->full_backref ?
3184 (unsigned long long)tback->parent :
3185 (unsigned long long)tback->root, back);
3187 if (back->is_data) {
3188 dback = (struct data_backref *)back;
3189 if (dback->found_ref != dback->num_refs) {
3193 fprintf(stderr, "Incorrect local backref count"
3194 " on %llu %s %llu owner %llu"
3195 " offset %llu found %u wanted %u back %p\n",
3196 (unsigned long long)rec->start,
3197 back->full_backref ?
3199 back->full_backref ?
3200 (unsigned long long)dback->parent:
3201 (unsigned long long)dback->root,
3202 (unsigned long long)dback->owner,
3203 (unsigned long long)dback->offset,
3204 dback->found_ref, dback->num_refs, back);
3206 if (dback->disk_bytenr != rec->start) {
3210 fprintf(stderr, "Backref disk bytenr does not"
3211 " match extent record, bytenr=%llu, "
3212 "ref bytenr=%llu\n",
3213 (unsigned long long)rec->start,
3214 (unsigned long long)dback->disk_bytenr);
3217 if (dback->bytes != rec->nr) {
3221 fprintf(stderr, "Backref bytes do not match "
3222 "extent backref, bytenr=%llu, ref "
3223 "bytes=%llu, backref bytes=%llu\n",
3224 (unsigned long long)rec->start,
3225 (unsigned long long)rec->nr,
3226 (unsigned long long)dback->bytes);
3229 if (!back->is_data) {
3232 dback = (struct data_backref *)back;
3233 found += dback->found_ref;
3236 if (found != rec->refs) {
3240 fprintf(stderr, "Incorrect global backref count "
3241 "on %llu found %llu wanted %llu\n",
3242 (unsigned long long)rec->start,
3243 (unsigned long long)found,
3244 (unsigned long long)rec->refs);
3250 static int free_all_extent_backrefs(struct extent_record *rec)
3252 struct extent_backref *back;
3253 struct list_head *cur;
3254 while (!list_empty(&rec->backrefs)) {
3255 cur = rec->backrefs.next;
3256 back = list_entry(cur, struct extent_backref, list);
3263 static void free_extent_record_cache(struct btrfs_fs_info *fs_info,
3264 struct cache_tree *extent_cache)
3266 struct cache_extent *cache;
3267 struct extent_record *rec;
3270 cache = first_cache_extent(extent_cache);
3273 rec = container_of(cache, struct extent_record, cache);
3274 btrfs_unpin_extent(fs_info, rec->start, rec->max_size);
3275 remove_cache_extent(extent_cache, cache);
3276 free_all_extent_backrefs(rec);
3281 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
3282 struct extent_record *rec)
3284 if (rec->content_checked && rec->owner_ref_checked &&
3285 rec->extent_item_refs == rec->refs && rec->refs > 0 &&
3286 rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0)) {
3287 remove_cache_extent(extent_cache, &rec->cache);
3288 free_all_extent_backrefs(rec);
3289 list_del_init(&rec->list);
3295 static int check_owner_ref(struct btrfs_root *root,
3296 struct extent_record *rec,
3297 struct extent_buffer *buf)
3299 struct extent_backref *node;
3300 struct tree_backref *back;
3301 struct btrfs_root *ref_root;
3302 struct btrfs_key key;
3303 struct btrfs_path path;
3304 struct extent_buffer *parent;
3309 list_for_each_entry(node, &rec->backrefs, list) {
3312 if (!node->found_ref)
3314 if (node->full_backref)
3316 back = (struct tree_backref *)node;
3317 if (btrfs_header_owner(buf) == back->root)
3320 BUG_ON(rec->is_root);
3322 /* try to find the block by search corresponding fs tree */
3323 key.objectid = btrfs_header_owner(buf);
3324 key.type = BTRFS_ROOT_ITEM_KEY;
3325 key.offset = (u64)-1;
3327 ref_root = btrfs_read_fs_root(root->fs_info, &key);
3328 if (IS_ERR(ref_root))
3331 level = btrfs_header_level(buf);
3333 btrfs_item_key_to_cpu(buf, &key, 0);
3335 btrfs_node_key_to_cpu(buf, &key, 0);
3337 btrfs_init_path(&path);
3338 path.lowest_level = level + 1;
3339 ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
3343 parent = path.nodes[level + 1];
3344 if (parent && buf->start == btrfs_node_blockptr(parent,
3345 path.slots[level + 1]))
3348 btrfs_release_path(&path);
3349 return found ? 0 : 1;
3352 static int is_extent_tree_record(struct extent_record *rec)
3354 struct list_head *cur = rec->backrefs.next;
3355 struct extent_backref *node;
3356 struct tree_backref *back;
3359 while(cur != &rec->backrefs) {
3360 node = list_entry(cur, struct extent_backref, list);
3364 back = (struct tree_backref *)node;
3365 if (node->full_backref)
3367 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
3374 static int record_bad_block_io(struct btrfs_fs_info *info,
3375 struct cache_tree *extent_cache,
3378 struct extent_record *rec;
3379 struct cache_extent *cache;
3380 struct btrfs_key key;
3382 cache = lookup_cache_extent(extent_cache, start, len);
3386 rec = container_of(cache, struct extent_record, cache);
3387 if (!is_extent_tree_record(rec))
3390 btrfs_disk_key_to_cpu(&key, &rec->parent_key);
3391 return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
3394 static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
3395 struct extent_buffer *buf, int slot)
3397 if (btrfs_header_level(buf)) {
3398 struct btrfs_key_ptr ptr1, ptr2;
3400 read_extent_buffer(buf, &ptr1, btrfs_node_key_ptr_offset(slot),
3401 sizeof(struct btrfs_key_ptr));
3402 read_extent_buffer(buf, &ptr2,
3403 btrfs_node_key_ptr_offset(slot + 1),
3404 sizeof(struct btrfs_key_ptr));
3405 write_extent_buffer(buf, &ptr1,
3406 btrfs_node_key_ptr_offset(slot + 1),
3407 sizeof(struct btrfs_key_ptr));
3408 write_extent_buffer(buf, &ptr2,
3409 btrfs_node_key_ptr_offset(slot),
3410 sizeof(struct btrfs_key_ptr));
3412 struct btrfs_disk_key key;
3413 btrfs_node_key(buf, &key, 0);
3414 btrfs_fixup_low_keys(root, path, &key,
3415 btrfs_header_level(buf) + 1);
3418 struct btrfs_item *item1, *item2;
3419 struct btrfs_key k1, k2;
3420 char *item1_data, *item2_data;
3421 u32 item1_offset, item2_offset, item1_size, item2_size;
3423 item1 = btrfs_item_nr(slot);
3424 item2 = btrfs_item_nr(slot + 1);
3425 btrfs_item_key_to_cpu(buf, &k1, slot);
3426 btrfs_item_key_to_cpu(buf, &k2, slot + 1);
3427 item1_offset = btrfs_item_offset(buf, item1);
3428 item2_offset = btrfs_item_offset(buf, item2);
3429 item1_size = btrfs_item_size(buf, item1);
3430 item2_size = btrfs_item_size(buf, item2);
3432 item1_data = malloc(item1_size);
3435 item2_data = malloc(item2_size);
3441 read_extent_buffer(buf, item1_data, item1_offset, item1_size);
3442 read_extent_buffer(buf, item2_data, item2_offset, item2_size);
3444 write_extent_buffer(buf, item1_data, item2_offset, item2_size);
3445 write_extent_buffer(buf, item2_data, item1_offset, item1_size);
3449 btrfs_set_item_offset(buf, item1, item2_offset);
3450 btrfs_set_item_offset(buf, item2, item1_offset);
3451 btrfs_set_item_size(buf, item1, item2_size);
3452 btrfs_set_item_size(buf, item2, item1_size);
3454 path->slots[0] = slot;
3455 btrfs_set_item_key_unsafe(root, path, &k2);
3456 path->slots[0] = slot + 1;
3457 btrfs_set_item_key_unsafe(root, path, &k1);
3462 static int fix_key_order(struct btrfs_trans_handle *trans,
3463 struct btrfs_root *root,
3464 struct btrfs_path *path)
3466 struct extent_buffer *buf;
3467 struct btrfs_key k1, k2;
3469 int level = path->lowest_level;
3472 buf = path->nodes[level];
3473 for (i = 0; i < btrfs_header_nritems(buf) - 1; i++) {
3475 btrfs_node_key_to_cpu(buf, &k1, i);
3476 btrfs_node_key_to_cpu(buf, &k2, i + 1);
3478 btrfs_item_key_to_cpu(buf, &k1, i);
3479 btrfs_item_key_to_cpu(buf, &k2, i + 1);
3481 if (btrfs_comp_cpu_keys(&k1, &k2) < 0)
3483 ret = swap_values(root, path, buf, i);
3486 btrfs_mark_buffer_dirty(buf);
3492 static int delete_bogus_item(struct btrfs_trans_handle *trans,
3493 struct btrfs_root *root,
3494 struct btrfs_path *path,
3495 struct extent_buffer *buf, int slot)
3497 struct btrfs_key key;
3498 int nritems = btrfs_header_nritems(buf);
3500 btrfs_item_key_to_cpu(buf, &key, slot);
3502 /* These are all the keys we can deal with missing. */
3503 if (key.type != BTRFS_DIR_INDEX_KEY &&
3504 key.type != BTRFS_EXTENT_ITEM_KEY &&
3505 key.type != BTRFS_METADATA_ITEM_KEY &&
3506 key.type != BTRFS_TREE_BLOCK_REF_KEY &&
3507 key.type != BTRFS_EXTENT_DATA_REF_KEY)
3510 printf("Deleting bogus item [%llu,%u,%llu] at slot %d on block %llu\n",
3511 (unsigned long long)key.objectid, key.type,
3512 (unsigned long long)key.offset, slot, buf->start);
3513 memmove_extent_buffer(buf, btrfs_item_nr_offset(slot),
3514 btrfs_item_nr_offset(slot + 1),
3515 sizeof(struct btrfs_item) *
3516 (nritems - slot - 1));
3517 btrfs_set_header_nritems(buf, nritems - 1);
3519 struct btrfs_disk_key disk_key;
3521 btrfs_item_key(buf, &disk_key, 0);
3522 btrfs_fixup_low_keys(root, path, &disk_key, 1);
3524 btrfs_mark_buffer_dirty(buf);
3528 static int fix_item_offset(struct btrfs_trans_handle *trans,
3529 struct btrfs_root *root,
3530 struct btrfs_path *path)
3532 struct extent_buffer *buf;
3536 /* We should only get this for leaves */
3537 BUG_ON(path->lowest_level);
3538 buf = path->nodes[0];
3540 for (i = 0; i < btrfs_header_nritems(buf); i++) {
3541 unsigned int shift = 0, offset;
3543 if (i == 0 && btrfs_item_end_nr(buf, i) !=
3544 BTRFS_LEAF_DATA_SIZE(root)) {
3545 if (btrfs_item_end_nr(buf, i) >
3546 BTRFS_LEAF_DATA_SIZE(root)) {
3547 ret = delete_bogus_item(trans, root, path,
3551 fprintf(stderr, "item is off the end of the "
3552 "leaf, can't fix\n");
3556 shift = BTRFS_LEAF_DATA_SIZE(root) -
3557 btrfs_item_end_nr(buf, i);
3558 } else if (i > 0 && btrfs_item_end_nr(buf, i) !=
3559 btrfs_item_offset_nr(buf, i - 1)) {
3560 if (btrfs_item_end_nr(buf, i) >
3561 btrfs_item_offset_nr(buf, i - 1)) {
3562 ret = delete_bogus_item(trans, root, path,
3566 fprintf(stderr, "items overlap, can't fix\n");
3570 shift = btrfs_item_offset_nr(buf, i - 1) -
3571 btrfs_item_end_nr(buf, i);
3576 printf("Shifting item nr %d by %u bytes in block %llu\n",
3577 i, shift, (unsigned long long)buf->start);
3578 offset = btrfs_item_offset_nr(buf, i);
3579 memmove_extent_buffer(buf,
3580 btrfs_leaf_data(buf) + offset + shift,
3581 btrfs_leaf_data(buf) + offset,
3582 btrfs_item_size_nr(buf, i));
3583 btrfs_set_item_offset(buf, btrfs_item_nr(i),
3585 btrfs_mark_buffer_dirty(buf);
3589 * We may have moved things, in which case we want to exit so we don't
3590 * write those changes out. Once we have proper abort functionality in
3591 * progs this can be changed to something nicer.
3598 * Attempt to fix basic block failures. If we can't fix it for whatever reason
3599 * then just return -EIO.
3601 static int try_to_fix_bad_block(struct btrfs_trans_handle *trans,
3602 struct btrfs_root *root,
3603 struct extent_buffer *buf,
3604 enum btrfs_tree_block_status status)
3606 struct ulist *roots;
3607 struct ulist_node *node;
3608 struct btrfs_root *search_root;
3609 struct btrfs_path *path;
3610 struct ulist_iterator iter;
3611 struct btrfs_key root_key, key;
3614 if (status != BTRFS_TREE_BLOCK_BAD_KEY_ORDER &&
3615 status != BTRFS_TREE_BLOCK_INVALID_OFFSETS)
3618 path = btrfs_alloc_path();
3622 ret = btrfs_find_all_roots(trans, root->fs_info, buf->start,
3625 btrfs_free_path(path);
3629 ULIST_ITER_INIT(&iter);
3630 while ((node = ulist_next(roots, &iter))) {
3631 root_key.objectid = node->val;
3632 root_key.type = BTRFS_ROOT_ITEM_KEY;
3633 root_key.offset = (u64)-1;
3635 search_root = btrfs_read_fs_root(root->fs_info, &root_key);
3641 record_root_in_trans(trans, search_root);
3643 path->lowest_level = btrfs_header_level(buf);
3644 path->skip_check_block = 1;
3645 if (path->lowest_level)
3646 btrfs_node_key_to_cpu(buf, &key, 0);
3648 btrfs_item_key_to_cpu(buf, &key, 0);
3649 ret = btrfs_search_slot(trans, search_root, &key, path, 0, 1);
3654 if (status == BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
3655 ret = fix_key_order(trans, search_root, path);
3656 else if (status == BTRFS_TREE_BLOCK_INVALID_OFFSETS)
3657 ret = fix_item_offset(trans, search_root, path);
3660 btrfs_release_path(path);
3663 btrfs_free_path(path);
3667 static int check_block(struct btrfs_trans_handle *trans,
3668 struct btrfs_root *root,
3669 struct cache_tree *extent_cache,
3670 struct extent_buffer *buf, u64 flags)
3672 struct extent_record *rec;
3673 struct cache_extent *cache;
3674 struct btrfs_key key;
3675 enum btrfs_tree_block_status status;
3679 cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
3682 rec = container_of(cache, struct extent_record, cache);
3683 rec->generation = btrfs_header_generation(buf);
3685 level = btrfs_header_level(buf);
3686 if (btrfs_header_nritems(buf) > 0) {
3689 btrfs_item_key_to_cpu(buf, &key, 0);
3691 btrfs_node_key_to_cpu(buf, &key, 0);
3693 rec->info_objectid = key.objectid;
3695 rec->info_level = level;
3697 if (btrfs_is_leaf(buf))
3698 status = btrfs_check_leaf(root, &rec->parent_key, buf);
3700 status = btrfs_check_node(root, &rec->parent_key, buf);
3702 if (status != BTRFS_TREE_BLOCK_CLEAN) {
3704 status = try_to_fix_bad_block(trans, root, buf,
3706 if (status != BTRFS_TREE_BLOCK_CLEAN) {
3708 fprintf(stderr, "bad block %llu\n",
3709 (unsigned long long)buf->start);
3712 * Signal to callers we need to start the scan over
3713 * again since we'll have cow'ed blocks.
3718 rec->content_checked = 1;
3719 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3720 rec->owner_ref_checked = 1;
3722 ret = check_owner_ref(root, rec, buf);
3724 rec->owner_ref_checked = 1;
3728 maybe_free_extent_rec(extent_cache, rec);
3732 static struct tree_backref *find_tree_backref(struct extent_record *rec,
3733 u64 parent, u64 root)
3735 struct list_head *cur = rec->backrefs.next;
3736 struct extent_backref *node;
3737 struct tree_backref *back;
3739 while(cur != &rec->backrefs) {
3740 node = list_entry(cur, struct extent_backref, list);
3744 back = (struct tree_backref *)node;
3746 if (!node->full_backref)
3748 if (parent == back->parent)
3751 if (node->full_backref)
3753 if (back->root == root)
3760 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
3761 u64 parent, u64 root)
3763 struct tree_backref *ref = malloc(sizeof(*ref));
3764 memset(&ref->node, 0, sizeof(ref->node));
3766 ref->parent = parent;
3767 ref->node.full_backref = 1;
3770 ref->node.full_backref = 0;
3772 list_add_tail(&ref->node.list, &rec->backrefs);
3777 static struct data_backref *find_data_backref(struct extent_record *rec,
3778 u64 parent, u64 root,
3779 u64 owner, u64 offset,
3781 u64 disk_bytenr, u64 bytes)
3783 struct list_head *cur = rec->backrefs.next;
3784 struct extent_backref *node;
3785 struct data_backref *back;
3787 while(cur != &rec->backrefs) {
3788 node = list_entry(cur, struct extent_backref, list);
3792 back = (struct data_backref *)node;
3794 if (!node->full_backref)
3796 if (parent == back->parent)
3799 if (node->full_backref)
3801 if (back->root == root && back->owner == owner &&
3802 back->offset == offset) {
3803 if (found_ref && node->found_ref &&
3804 (back->bytes != bytes ||
3805 back->disk_bytenr != disk_bytenr))
3814 static struct data_backref *alloc_data_backref(struct extent_record *rec,
3815 u64 parent, u64 root,
3816 u64 owner, u64 offset,
3819 struct data_backref *ref = malloc(sizeof(*ref));
3820 memset(&ref->node, 0, sizeof(ref->node));
3821 ref->node.is_data = 1;
3824 ref->parent = parent;
3827 ref->node.full_backref = 1;
3831 ref->offset = offset;
3832 ref->node.full_backref = 0;
3834 ref->bytes = max_size;
3837 list_add_tail(&ref->node.list, &rec->backrefs);
3838 if (max_size > rec->max_size)
3839 rec->max_size = max_size;
3843 static int add_extent_rec(struct cache_tree *extent_cache,
3844 struct btrfs_key *parent_key, u64 parent_gen,
3845 u64 start, u64 nr, u64 extent_item_refs,
3846 int is_root, int inc_ref, int set_checked,
3847 int metadata, int extent_rec, u64 max_size)
3849 struct extent_record *rec;
3850 struct cache_extent *cache;
3854 cache = lookup_cache_extent(extent_cache, start, nr);
3856 rec = container_of(cache, struct extent_record, cache);
3860 rec->nr = max(nr, max_size);
3863 * We need to make sure to reset nr to whatever the extent
3864 * record says was the real size, this way we can compare it to
3868 if (start != rec->start || rec->found_rec) {
3869 struct extent_record *tmp;
3872 if (list_empty(&rec->list))
3873 list_add_tail(&rec->list,
3874 &duplicate_extents);
3877 * We have to do this song and dance in case we
3878 * find an extent record that falls inside of
3879 * our current extent record but does not have
3880 * the same objectid.
3882 tmp = malloc(sizeof(*tmp));
3886 tmp->max_size = max_size;
3889 tmp->metadata = metadata;
3890 tmp->extent_item_refs = extent_item_refs;
3891 INIT_LIST_HEAD(&tmp->list);
3892 list_add_tail(&tmp->list, &rec->dups);
3893 rec->num_duplicates++;
3900 if (extent_item_refs && !dup) {
3901 if (rec->extent_item_refs) {
3902 fprintf(stderr, "block %llu rec "
3903 "extent_item_refs %llu, passed %llu\n",
3904 (unsigned long long)start,
3905 (unsigned long long)
3906 rec->extent_item_refs,
3907 (unsigned long long)extent_item_refs);
3909 rec->extent_item_refs = extent_item_refs;
3914 rec->content_checked = 1;
3915 rec->owner_ref_checked = 1;
3919 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
3921 rec->parent_generation = parent_gen;
3923 if (rec->max_size < max_size)
3924 rec->max_size = max_size;
3926 maybe_free_extent_rec(extent_cache, rec);
3929 rec = malloc(sizeof(*rec));
3931 rec->max_size = max_size;
3932 rec->nr = max(nr, max_size);
3933 rec->found_rec = !!extent_rec;
3934 rec->content_checked = 0;
3935 rec->owner_ref_checked = 0;
3936 rec->num_duplicates = 0;
3937 rec->metadata = metadata;
3938 INIT_LIST_HEAD(&rec->backrefs);
3939 INIT_LIST_HEAD(&rec->dups);
3940 INIT_LIST_HEAD(&rec->list);
3952 if (extent_item_refs)
3953 rec->extent_item_refs = extent_item_refs;
3955 rec->extent_item_refs = 0;
3958 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
3960 memset(&rec->parent_key, 0, sizeof(*parent_key));
3963 rec->parent_generation = parent_gen;
3965 rec->parent_generation = 0;
3967 rec->cache.start = start;
3968 rec->cache.size = nr;
3969 ret = insert_cache_extent(extent_cache, &rec->cache);
3973 rec->content_checked = 1;
3974 rec->owner_ref_checked = 1;
3979 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
3980 u64 parent, u64 root, int found_ref)
3982 struct extent_record *rec;
3983 struct tree_backref *back;
3984 struct cache_extent *cache;
3986 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3988 add_extent_rec(extent_cache, NULL, 0, bytenr,
3989 1, 0, 0, 0, 0, 1, 0, 0);
3990 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3995 rec = container_of(cache, struct extent_record, cache);
3996 if (rec->start != bytenr) {
4000 back = find_tree_backref(rec, parent, root);
4002 back = alloc_tree_backref(rec, parent, root);
4005 if (back->node.found_ref) {
4006 fprintf(stderr, "Extent back ref already exists "
4007 "for %llu parent %llu root %llu \n",
4008 (unsigned long long)bytenr,
4009 (unsigned long long)parent,
4010 (unsigned long long)root);
4012 back->node.found_ref = 1;
4014 if (back->node.found_extent_tree) {
4015 fprintf(stderr, "Extent back ref already exists "
4016 "for %llu parent %llu root %llu \n",
4017 (unsigned long long)bytenr,
4018 (unsigned long long)parent,
4019 (unsigned long long)root);
4021 back->node.found_extent_tree = 1;
4023 maybe_free_extent_rec(extent_cache, rec);
4027 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
4028 u64 parent, u64 root, u64 owner, u64 offset,
4029 u32 num_refs, int found_ref, u64 max_size)
4031 struct extent_record *rec;
4032 struct data_backref *back;
4033 struct cache_extent *cache;
4035 cache = lookup_cache_extent(extent_cache, bytenr, 1);
4037 add_extent_rec(extent_cache, NULL, 0, bytenr, 1, 0, 0, 0, 0,
4039 cache = lookup_cache_extent(extent_cache, bytenr, 1);
4044 rec = container_of(cache, struct extent_record, cache);
4045 if (rec->max_size < max_size)
4046 rec->max_size = max_size;
4049 * If found_ref is set then max_size is the real size and must match the
4050 * existing refs. So if we have already found a ref then we need to
4051 * make sure that this ref matches the existing one, otherwise we need
4052 * to add a new backref so we can notice that the backrefs don't match
4053 * and we need to figure out who is telling the truth. This is to
4054 * account for that awful fsync bug I introduced where we'd end up with
4055 * a btrfs_file_extent_item that would have its length include multiple
4056 * prealloc extents or point inside of a prealloc extent.
4058 back = find_data_backref(rec, parent, root, owner, offset, found_ref,
4061 back = alloc_data_backref(rec, parent, root, owner, offset,
4065 BUG_ON(num_refs != 1);
4066 if (back->node.found_ref)
4067 BUG_ON(back->bytes != max_size);
4068 back->node.found_ref = 1;
4069 back->found_ref += 1;
4070 back->bytes = max_size;
4071 back->disk_bytenr = bytenr;
4073 rec->content_checked = 1;
4074 rec->owner_ref_checked = 1;
4076 if (back->node.found_extent_tree) {
4077 fprintf(stderr, "Extent back ref already exists "
4078 "for %llu parent %llu root %llu "
4079 "owner %llu offset %llu num_refs %lu\n",
4080 (unsigned long long)bytenr,
4081 (unsigned long long)parent,
4082 (unsigned long long)root,
4083 (unsigned long long)owner,
4084 (unsigned long long)offset,
4085 (unsigned long)num_refs);
4087 back->num_refs = num_refs;
4088 back->node.found_extent_tree = 1;
4090 maybe_free_extent_rec(extent_cache, rec);
4094 static int add_pending(struct cache_tree *pending,
4095 struct cache_tree *seen, u64 bytenr, u32 size)
4098 ret = add_cache_extent(seen, bytenr, size);
4101 add_cache_extent(pending, bytenr, size);
4105 static int pick_next_pending(struct cache_tree *pending,
4106 struct cache_tree *reada,
4107 struct cache_tree *nodes,
4108 u64 last, struct block_info *bits, int bits_nr,
4111 unsigned long node_start = last;
4112 struct cache_extent *cache;
4115 cache = search_cache_extent(reada, 0);
4117 bits[0].start = cache->start;
4118 bits[0].size = cache->size;
4123 if (node_start > 32768)
4124 node_start -= 32768;
4126 cache = search_cache_extent(nodes, node_start);
4128 cache = search_cache_extent(nodes, 0);
4131 cache = search_cache_extent(pending, 0);
4136 bits[ret].start = cache->start;
4137 bits[ret].size = cache->size;
4138 cache = next_cache_extent(cache);
4140 } while (cache && ret < bits_nr);
4146 bits[ret].start = cache->start;
4147 bits[ret].size = cache->size;
4148 cache = next_cache_extent(cache);
4150 } while (cache && ret < bits_nr);
4152 if (bits_nr - ret > 8) {
4153 u64 lookup = bits[0].start + bits[0].size;
4154 struct cache_extent *next;
4155 next = search_cache_extent(pending, lookup);
4157 if (next->start - lookup > 32768)
4159 bits[ret].start = next->start;
4160 bits[ret].size = next->size;
4161 lookup = next->start + next->size;
4165 next = next_cache_extent(next);
4173 static void free_chunk_record(struct cache_extent *cache)
4175 struct chunk_record *rec;
4177 rec = container_of(cache, struct chunk_record, cache);
4178 list_del_init(&rec->list);
4179 list_del_init(&rec->dextents);
4183 void free_chunk_cache_tree(struct cache_tree *chunk_cache)
4185 cache_tree_free_extents(chunk_cache, free_chunk_record);
4188 static void free_device_record(struct rb_node *node)
4190 struct device_record *rec;
4192 rec = container_of(node, struct device_record, node);
4196 FREE_RB_BASED_TREE(device_cache, free_device_record);
4198 int insert_block_group_record(struct block_group_tree *tree,
4199 struct block_group_record *bg_rec)
4203 ret = insert_cache_extent(&tree->tree, &bg_rec->cache);
4207 list_add_tail(&bg_rec->list, &tree->block_groups);
4211 static void free_block_group_record(struct cache_extent *cache)
4213 struct block_group_record *rec;
4215 rec = container_of(cache, struct block_group_record, cache);
4216 list_del_init(&rec->list);
4220 void free_block_group_tree(struct block_group_tree *tree)
4222 cache_tree_free_extents(&tree->tree, free_block_group_record);
4225 int insert_device_extent_record(struct device_extent_tree *tree,
4226 struct device_extent_record *de_rec)
4231 * Device extent is a bit different from the other extents, because
4232 * the extents which belong to the different devices may have the
4233 * same start and size, so we need use the special extent cache
4234 * search/insert functions.
4236 ret = insert_cache_extent2(&tree->tree, &de_rec->cache);
4240 list_add_tail(&de_rec->chunk_list, &tree->no_chunk_orphans);
4241 list_add_tail(&de_rec->device_list, &tree->no_device_orphans);
4245 static void free_device_extent_record(struct cache_extent *cache)
4247 struct device_extent_record *rec;
4249 rec = container_of(cache, struct device_extent_record, cache);
4250 if (!list_empty(&rec->chunk_list))
4251 list_del_init(&rec->chunk_list);
4252 if (!list_empty(&rec->device_list))
4253 list_del_init(&rec->device_list);
4257 void free_device_extent_tree(struct device_extent_tree *tree)
4259 cache_tree_free_extents(&tree->tree, free_device_extent_record);
4262 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4263 static int process_extent_ref_v0(struct cache_tree *extent_cache,
4264 struct extent_buffer *leaf, int slot)
4266 struct btrfs_extent_ref_v0 *ref0;
4267 struct btrfs_key key;
4269 btrfs_item_key_to_cpu(leaf, &key, slot);
4270 ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
4271 if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
4272 add_tree_backref(extent_cache, key.objectid, key.offset, 0, 0);
4274 add_data_backref(extent_cache, key.objectid, key.offset, 0,
4275 0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
4281 struct chunk_record *btrfs_new_chunk_record(struct extent_buffer *leaf,
4282 struct btrfs_key *key,
4285 struct btrfs_chunk *ptr;
4286 struct chunk_record *rec;
4289 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4290 num_stripes = btrfs_chunk_num_stripes(leaf, ptr);
4292 rec = malloc(btrfs_chunk_record_size(num_stripes));
4294 fprintf(stderr, "memory allocation failed\n");
4298 memset(rec, 0, btrfs_chunk_record_size(num_stripes));
4300 INIT_LIST_HEAD(&rec->list);
4301 INIT_LIST_HEAD(&rec->dextents);
4304 rec->cache.start = key->offset;
4305 rec->cache.size = btrfs_chunk_length(leaf, ptr);
4307 rec->generation = btrfs_header_generation(leaf);
4309 rec->objectid = key->objectid;
4310 rec->type = key->type;
4311 rec->offset = key->offset;
4313 rec->length = rec->cache.size;
4314 rec->owner = btrfs_chunk_owner(leaf, ptr);
4315 rec->stripe_len = btrfs_chunk_stripe_len(leaf, ptr);
4316 rec->type_flags = btrfs_chunk_type(leaf, ptr);
4317 rec->io_width = btrfs_chunk_io_width(leaf, ptr);
4318 rec->io_align = btrfs_chunk_io_align(leaf, ptr);
4319 rec->sector_size = btrfs_chunk_sector_size(leaf, ptr);
4320 rec->num_stripes = num_stripes;
4321 rec->sub_stripes = btrfs_chunk_sub_stripes(leaf, ptr);
4323 for (i = 0; i < rec->num_stripes; ++i) {
4324 rec->stripes[i].devid =
4325 btrfs_stripe_devid_nr(leaf, ptr, i);
4326 rec->stripes[i].offset =
4327 btrfs_stripe_offset_nr(leaf, ptr, i);
4328 read_extent_buffer(leaf, rec->stripes[i].dev_uuid,
4329 (unsigned long)btrfs_stripe_dev_uuid_nr(ptr, i),
4336 static int process_chunk_item(struct cache_tree *chunk_cache,
4337 struct btrfs_key *key, struct extent_buffer *eb,
4340 struct chunk_record *rec;
4343 rec = btrfs_new_chunk_record(eb, key, slot);
4344 ret = insert_cache_extent(chunk_cache, &rec->cache);
4346 fprintf(stderr, "Chunk[%llu, %llu] existed.\n",
4347 rec->offset, rec->length);
4354 static int process_device_item(struct rb_root *dev_cache,
4355 struct btrfs_key *key, struct extent_buffer *eb, int slot)
4357 struct btrfs_dev_item *ptr;
4358 struct device_record *rec;
4361 ptr = btrfs_item_ptr(eb,
4362 slot, struct btrfs_dev_item);
4364 rec = malloc(sizeof(*rec));
4366 fprintf(stderr, "memory allocation failed\n");
4370 rec->devid = key->offset;
4371 rec->generation = btrfs_header_generation(eb);
4373 rec->objectid = key->objectid;
4374 rec->type = key->type;
4375 rec->offset = key->offset;
4377 rec->devid = btrfs_device_id(eb, ptr);
4378 rec->total_byte = btrfs_device_total_bytes(eb, ptr);
4379 rec->byte_used = btrfs_device_bytes_used(eb, ptr);
4381 ret = rb_insert(dev_cache, &rec->node, device_record_compare);
4383 fprintf(stderr, "Device[%llu] existed.\n", rec->devid);
4390 struct block_group_record *
4391 btrfs_new_block_group_record(struct extent_buffer *leaf, struct btrfs_key *key,
4394 struct btrfs_block_group_item *ptr;
4395 struct block_group_record *rec;
4397 rec = malloc(sizeof(*rec));
4399 fprintf(stderr, "memory allocation failed\n");
4402 memset(rec, 0, sizeof(*rec));
4404 rec->cache.start = key->objectid;
4405 rec->cache.size = key->offset;
4407 rec->generation = btrfs_header_generation(leaf);
4409 rec->objectid = key->objectid;
4410 rec->type = key->type;
4411 rec->offset = key->offset;
4413 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item);
4414 rec->flags = btrfs_disk_block_group_flags(leaf, ptr);
4416 INIT_LIST_HEAD(&rec->list);
4421 static int process_block_group_item(struct block_group_tree *block_group_cache,
4422 struct btrfs_key *key,
4423 struct extent_buffer *eb, int slot)
4425 struct block_group_record *rec;
4428 rec = btrfs_new_block_group_record(eb, key, slot);
4429 ret = insert_block_group_record(block_group_cache, rec);
4431 fprintf(stderr, "Block Group[%llu, %llu] existed.\n",
4432 rec->objectid, rec->offset);
4439 struct device_extent_record *
4440 btrfs_new_device_extent_record(struct extent_buffer *leaf,
4441 struct btrfs_key *key, int slot)
4443 struct device_extent_record *rec;
4444 struct btrfs_dev_extent *ptr;
4446 rec = malloc(sizeof(*rec));
4448 fprintf(stderr, "memory allocation failed\n");
4451 memset(rec, 0, sizeof(*rec));
4453 rec->cache.objectid = key->objectid;
4454 rec->cache.start = key->offset;
4456 rec->generation = btrfs_header_generation(leaf);
4458 rec->objectid = key->objectid;
4459 rec->type = key->type;
4460 rec->offset = key->offset;
4462 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
4463 rec->chunk_objecteid =
4464 btrfs_dev_extent_chunk_objectid(leaf, ptr);
4466 btrfs_dev_extent_chunk_offset(leaf, ptr);
4467 rec->length = btrfs_dev_extent_length(leaf, ptr);
4468 rec->cache.size = rec->length;
4470 INIT_LIST_HEAD(&rec->chunk_list);
4471 INIT_LIST_HEAD(&rec->device_list);
4477 process_device_extent_item(struct device_extent_tree *dev_extent_cache,
4478 struct btrfs_key *key, struct extent_buffer *eb,
4481 struct device_extent_record *rec;
4484 rec = btrfs_new_device_extent_record(eb, key, slot);
4485 ret = insert_device_extent_record(dev_extent_cache, rec);
4488 "Device extent[%llu, %llu, %llu] existed.\n",
4489 rec->objectid, rec->offset, rec->length);
4496 static int process_extent_item(struct btrfs_root *root,
4497 struct cache_tree *extent_cache,
4498 struct extent_buffer *eb, int slot)
4500 struct btrfs_extent_item *ei;
4501 struct btrfs_extent_inline_ref *iref;
4502 struct btrfs_extent_data_ref *dref;
4503 struct btrfs_shared_data_ref *sref;
4504 struct btrfs_key key;
4508 u32 item_size = btrfs_item_size_nr(eb, slot);
4514 btrfs_item_key_to_cpu(eb, &key, slot);
4516 if (key.type == BTRFS_METADATA_ITEM_KEY) {
4518 num_bytes = root->leafsize;
4520 num_bytes = key.offset;
4523 if (item_size < sizeof(*ei)) {
4524 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4525 struct btrfs_extent_item_v0 *ei0;
4526 BUG_ON(item_size != sizeof(*ei0));
4527 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
4528 refs = btrfs_extent_refs_v0(eb, ei0);
4532 return add_extent_rec(extent_cache, NULL, 0, key.objectid,
4533 num_bytes, refs, 0, 0, 0, metadata, 1,
4537 ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
4538 refs = btrfs_extent_refs(eb, ei);
4540 add_extent_rec(extent_cache, NULL, 0, key.objectid, num_bytes,
4541 refs, 0, 0, 0, metadata, 1, num_bytes);
4543 ptr = (unsigned long)(ei + 1);
4544 if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
4545 key.type == BTRFS_EXTENT_ITEM_KEY)
4546 ptr += sizeof(struct btrfs_tree_block_info);
4548 end = (unsigned long)ei + item_size;
4550 iref = (struct btrfs_extent_inline_ref *)ptr;
4551 type = btrfs_extent_inline_ref_type(eb, iref);
4552 offset = btrfs_extent_inline_ref_offset(eb, iref);
4554 case BTRFS_TREE_BLOCK_REF_KEY:
4555 add_tree_backref(extent_cache, key.objectid,
4558 case BTRFS_SHARED_BLOCK_REF_KEY:
4559 add_tree_backref(extent_cache, key.objectid,
4562 case BTRFS_EXTENT_DATA_REF_KEY:
4563 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
4564 add_data_backref(extent_cache, key.objectid, 0,
4565 btrfs_extent_data_ref_root(eb, dref),
4566 btrfs_extent_data_ref_objectid(eb,
4568 btrfs_extent_data_ref_offset(eb, dref),
4569 btrfs_extent_data_ref_count(eb, dref),
4572 case BTRFS_SHARED_DATA_REF_KEY:
4573 sref = (struct btrfs_shared_data_ref *)(iref + 1);
4574 add_data_backref(extent_cache, key.objectid, offset,
4576 btrfs_shared_data_ref_count(eb, sref),
4580 fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
4581 key.objectid, key.type, num_bytes);
4584 ptr += btrfs_extent_inline_ref_size(type);
4591 static int check_cache_range(struct btrfs_root *root,
4592 struct btrfs_block_group_cache *cache,
4593 u64 offset, u64 bytes)
4595 struct btrfs_free_space *entry;
4601 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
4602 bytenr = btrfs_sb_offset(i);
4603 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
4604 cache->key.objectid, bytenr, 0,
4605 &logical, &nr, &stripe_len);
4610 if (logical[nr] + stripe_len <= offset)
4612 if (offset + bytes <= logical[nr])
4614 if (logical[nr] == offset) {
4615 if (stripe_len >= bytes) {
4619 bytes -= stripe_len;
4620 offset += stripe_len;
4621 } else if (logical[nr] < offset) {
4622 if (logical[nr] + stripe_len >=
4627 bytes = (offset + bytes) -
4628 (logical[nr] + stripe_len);
4629 offset = logical[nr] + stripe_len;
4632 * Could be tricky, the super may land in the
4633 * middle of the area we're checking. First
4634 * check the easiest case, it's at the end.
4636 if (logical[nr] + stripe_len >=
4638 bytes = logical[nr] - offset;
4642 /* Check the left side */
4643 ret = check_cache_range(root, cache,
4645 logical[nr] - offset);
4651 /* Now we continue with the right side */
4652 bytes = (offset + bytes) -
4653 (logical[nr] + stripe_len);
4654 offset = logical[nr] + stripe_len;
4661 entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
4663 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
4664 offset, offset+bytes);
4668 if (entry->offset != offset) {
4669 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
4674 if (entry->bytes != bytes) {
4675 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
4676 bytes, entry->bytes, offset);
4680 unlink_free_space(cache->free_space_ctl, entry);
4685 static int verify_space_cache(struct btrfs_root *root,
4686 struct btrfs_block_group_cache *cache)
4688 struct btrfs_path *path;
4689 struct extent_buffer *leaf;
4690 struct btrfs_key key;
4694 path = btrfs_alloc_path();
4698 root = root->fs_info->extent_root;
4700 last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
4702 key.objectid = last;
4704 key.type = BTRFS_EXTENT_ITEM_KEY;
4706 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4711 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4712 ret = btrfs_next_leaf(root, path);
4720 leaf = path->nodes[0];
4721 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4722 if (key.objectid >= cache->key.offset + cache->key.objectid)
4724 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
4725 key.type != BTRFS_METADATA_ITEM_KEY) {
4730 if (last == key.objectid) {
4731 if (key.type == BTRFS_EXTENT_ITEM_KEY)
4732 last = key.objectid + key.offset;
4734 last = key.objectid + root->leafsize;
4739 ret = check_cache_range(root, cache, last,
4740 key.objectid - last);
4743 if (key.type == BTRFS_EXTENT_ITEM_KEY)
4744 last = key.objectid + key.offset;
4746 last = key.objectid + root->leafsize;
4750 if (last < cache->key.objectid + cache->key.offset)
4751 ret = check_cache_range(root, cache, last,
4752 cache->key.objectid +
4753 cache->key.offset - last);
4756 btrfs_free_path(path);
4759 !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
4760 fprintf(stderr, "There are still entries left in the space "
4768 static int check_space_cache(struct btrfs_root *root)
4770 struct btrfs_block_group_cache *cache;
4771 u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
4775 if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
4776 btrfs_super_generation(root->fs_info->super_copy) !=
4777 btrfs_super_cache_generation(root->fs_info->super_copy)) {
4778 printf("cache and super generation don't match, space cache "
4779 "will be invalidated\n");
4784 cache = btrfs_lookup_first_block_group(root->fs_info, start);
4788 start = cache->key.objectid + cache->key.offset;
4789 if (!cache->free_space_ctl) {
4790 if (btrfs_init_free_space_ctl(cache,
4791 root->sectorsize)) {
4796 btrfs_remove_free_space_cache(cache);
4799 ret = load_free_space_cache(root->fs_info, cache);
4803 ret = verify_space_cache(root, cache);
4805 fprintf(stderr, "cache appears valid but isnt %Lu\n",
4806 cache->key.objectid);
4811 return error ? -EINVAL : 0;
4814 static int read_extent_data(struct btrfs_root *root, char *data,
4815 u64 logical, u64 *len, int mirror)
4818 struct btrfs_multi_bio *multi = NULL;
4819 struct btrfs_fs_info *info = root->fs_info;
4820 struct btrfs_device *device;
4824 ret = btrfs_map_block(&info->mapping_tree, READ, logical, len,
4825 &multi, mirror, NULL);
4827 fprintf(stderr, "Couldn't map the block %llu\n",
4831 device = multi->stripes[0].dev;
4833 if (device->fd == 0)
4838 ret = pread64(device->fd, data, *len, multi->stripes[0].physical);
4848 static int check_extent_csums(struct btrfs_root *root, u64 bytenr,
4849 u64 num_bytes, unsigned long leaf_offset,
4850 struct extent_buffer *eb) {
4853 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
4855 unsigned long csum_offset;
4859 u64 data_checked = 0;
4865 if (num_bytes % root->sectorsize)
4868 data = malloc(num_bytes);
4872 while (offset < num_bytes) {
4875 read_len = num_bytes - offset;
4876 /* read as much space once a time */
4877 ret = read_extent_data(root, data + offset,
4878 bytenr + offset, &read_len, mirror);
4882 /* verify every 4k data's checksum */
4883 while (data_checked < read_len) {
4885 tmp = offset + data_checked;
4887 csum = btrfs_csum_data(NULL, (char *)data + tmp,
4888 csum, root->sectorsize);
4889 btrfs_csum_final(csum, (char *)&csum);
4891 csum_offset = leaf_offset +
4892 tmp / root->sectorsize * csum_size;
4893 read_extent_buffer(eb, (char *)&csum_expected,
4894 csum_offset, csum_size);
4895 /* try another mirror */
4896 if (csum != csum_expected) {
4897 fprintf(stderr, "mirror %d bytenr %llu csum %u expected csum %u\n",
4898 mirror, bytenr + tmp,
4899 csum, csum_expected);
4900 num_copies = btrfs_num_copies(
4901 &root->fs_info->mapping_tree,
4903 if (mirror < num_copies - 1) {
4908 data_checked += root->sectorsize;
4917 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
4920 struct btrfs_path *path;
4921 struct extent_buffer *leaf;
4922 struct btrfs_key key;
4925 path = btrfs_alloc_path();
4927 fprintf(stderr, "Error allocing path\n");
4931 key.objectid = bytenr;
4932 key.type = BTRFS_EXTENT_ITEM_KEY;
4933 key.offset = (u64)-1;
4936 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
4939 fprintf(stderr, "Error looking up extent record %d\n", ret);
4940 btrfs_free_path(path);
4943 if (path->slots[0] > 0) {
4946 ret = btrfs_prev_leaf(root, path);
4949 } else if (ret > 0) {
4956 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4959 * Block group items come before extent items if they have the same
4960 * bytenr, so walk back one more just in case. Dear future traveler,
4961 * first congrats on mastering time travel. Now if it's not too much
4962 * trouble could you go back to 2006 and tell Chris to make the
4963 * BLOCK_GROUP_ITEM_KEY (and BTRFS_*_REF_KEY) lower than the
4964 * EXTENT_ITEM_KEY please?
4966 while (key.type > BTRFS_EXTENT_ITEM_KEY) {
4967 if (path->slots[0] > 0) {
4970 ret = btrfs_prev_leaf(root, path);
4973 } else if (ret > 0) {
4978 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4982 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4983 ret = btrfs_next_leaf(root, path);
4985 fprintf(stderr, "Error going to next leaf "
4987 btrfs_free_path(path);
4993 leaf = path->nodes[0];
4994 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4995 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
4999 if (key.objectid + key.offset < bytenr) {
5003 if (key.objectid > bytenr + num_bytes)
5006 if (key.objectid == bytenr) {
5007 if (key.offset >= num_bytes) {
5011 num_bytes -= key.offset;
5012 bytenr += key.offset;
5013 } else if (key.objectid < bytenr) {
5014 if (key.objectid + key.offset >= bytenr + num_bytes) {
5018 num_bytes = (bytenr + num_bytes) -
5019 (key.objectid + key.offset);
5020 bytenr = key.objectid + key.offset;
5022 if (key.objectid + key.offset < bytenr + num_bytes) {
5023 u64 new_start = key.objectid + key.offset;
5024 u64 new_bytes = bytenr + num_bytes - new_start;
5027 * Weird case, the extent is in the middle of
5028 * our range, we'll have to search one side
5029 * and then the other. Not sure if this happens
5030 * in real life, but no harm in coding it up
5031 * anyway just in case.
5033 btrfs_release_path(path);
5034 ret = check_extent_exists(root, new_start,
5037 fprintf(stderr, "Right section didn't "
5041 num_bytes = key.objectid - bytenr;
5044 num_bytes = key.objectid - bytenr;
5051 if (num_bytes && !ret) {
5052 fprintf(stderr, "There are no extents for csum range "
5053 "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
5057 btrfs_free_path(path);
5061 static int check_csums(struct btrfs_root *root)
5063 struct btrfs_path *path;
5064 struct extent_buffer *leaf;
5065 struct btrfs_key key;
5066 u64 offset = 0, num_bytes = 0;
5067 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
5071 unsigned long leaf_offset;
5073 root = root->fs_info->csum_root;
5074 if (!extent_buffer_uptodate(root->node)) {
5075 fprintf(stderr, "No valid csum tree found\n");
5079 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
5080 key.type = BTRFS_EXTENT_CSUM_KEY;
5083 path = btrfs_alloc_path();
5087 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5089 fprintf(stderr, "Error searching csum tree %d\n", ret);
5090 btrfs_free_path(path);
5094 if (ret > 0 && path->slots[0])
5099 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5100 ret = btrfs_next_leaf(root, path);
5102 fprintf(stderr, "Error going to next leaf "
5109 leaf = path->nodes[0];
5111 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5112 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
5117 data_len = (btrfs_item_size_nr(leaf, path->slots[0]) /
5118 csum_size) * root->sectorsize;
5119 if (!check_data_csum)
5120 goto skip_csum_check;
5121 leaf_offset = btrfs_item_ptr_offset(leaf, path->slots[0]);
5122 ret = check_extent_csums(root, key.offset, data_len,
5128 offset = key.offset;
5129 } else if (key.offset != offset + num_bytes) {
5130 ret = check_extent_exists(root, offset, num_bytes);
5132 fprintf(stderr, "Csum exists for %Lu-%Lu but "
5133 "there is no extent record\n",
5134 offset, offset+num_bytes);
5137 offset = key.offset;
5140 num_bytes += data_len;
5144 btrfs_free_path(path);
5148 static int is_dropped_key(struct btrfs_key *key,
5149 struct btrfs_key *drop_key) {
5150 if (key->objectid < drop_key->objectid)
5152 else if (key->objectid == drop_key->objectid) {
5153 if (key->type < drop_key->type)
5155 else if (key->type == drop_key->type) {
5156 if (key->offset < drop_key->offset)
5163 static int calc_extent_flag(struct btrfs_root *root,
5164 struct cache_tree *extent_cache,
5165 struct extent_buffer *buf,
5166 struct root_item_record *ri,
5170 int nritems = btrfs_header_nritems(buf);
5171 struct btrfs_key key;
5172 struct extent_record *rec;
5173 struct cache_extent *cache;
5174 struct data_backref *dback;
5175 struct tree_backref *tback;
5176 struct extent_buffer *new_buf;
5186 * Except file/reloc tree, we can not have
5189 if (ri->objectid < BTRFS_FIRST_FREE_OBJECTID)
5194 if (buf->start == ri->bytenr)
5196 if (btrfs_is_leaf(buf)) {
5198 * we are searching from original root, world
5199 * peace is achieved, we use normal backref.
5201 owner = btrfs_header_owner(buf);
5202 if (owner == ri->objectid)
5205 * we check every eb here, and if any of
5206 * eb dosen't have original root refers
5207 * to this eb, we set full backref flag for
5208 * this extent, otherwise normal backref.
5210 for (i = 0; i < nritems; i++) {
5211 struct btrfs_file_extent_item *fi;
5212 btrfs_item_key_to_cpu(buf, &key, i);
5214 if (key.type != BTRFS_EXTENT_DATA_KEY)
5216 fi = btrfs_item_ptr(buf, i,
5217 struct btrfs_file_extent_item);
5218 if (btrfs_file_extent_type(buf, fi) ==
5219 BTRFS_FILE_EXTENT_INLINE)
5221 if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
5223 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
5224 cache = lookup_cache_extent(extent_cache, bytenr, 1);
5227 offset = btrfs_file_extent_offset(buf, fi);
5228 rec = container_of(cache, struct extent_record, cache);
5229 dback = find_data_backref(rec, 0, ri->objectid, owner,
5230 key.offset - offset, 1, bytenr, bytenr);
5236 level = btrfs_header_level(buf);
5237 for (i = 0; i < nritems; i++) {
5238 ptr = btrfs_node_blockptr(buf, i);
5239 size = btrfs_level_size(root, level);
5241 new_buf = read_tree_block(root, ptr, size, 0);
5242 if (!extent_buffer_uptodate(new_buf)) {
5243 free_extent_buffer(new_buf);
5248 * we are searching from origin root, world
5249 * peace is achieved, we use normal backref.
5251 owner = btrfs_header_owner(new_buf);
5252 free_extent_buffer(new_buf);
5253 if (owner == ri->objectid)
5256 cache = lookup_cache_extent(extent_cache, ptr, size);
5259 rec = container_of(cache, struct extent_record, cache);
5260 tback = find_tree_backref(rec, 0, owner);
5268 cache = lookup_cache_extent(extent_cache, buf->start, 1);
5269 /* we have added this extent before */
5271 rec = container_of(cache, struct extent_record, cache);
5272 rec->flag_block_full_backref = 0;
5275 *flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5276 cache = lookup_cache_extent(extent_cache, buf->start, 1);
5277 /* we have added this extent before */
5279 rec = container_of(cache, struct extent_record, cache);
5280 rec->flag_block_full_backref = 1;
5284 static int run_next_block(struct btrfs_trans_handle *trans,
5285 struct btrfs_root *root,
5286 struct block_info *bits,
5289 struct cache_tree *pending,
5290 struct cache_tree *seen,
5291 struct cache_tree *reada,
5292 struct cache_tree *nodes,
5293 struct cache_tree *extent_cache,
5294 struct cache_tree *chunk_cache,
5295 struct rb_root *dev_cache,
5296 struct block_group_tree *block_group_cache,
5297 struct device_extent_tree *dev_extent_cache,
5298 struct root_item_record *ri)
5300 struct extent_buffer *buf;
5311 struct btrfs_key key;
5312 struct cache_extent *cache;
5315 nritems = pick_next_pending(pending, reada, nodes, *last, bits,
5316 bits_nr, &reada_bits);
5321 for(i = 0; i < nritems; i++) {
5322 ret = add_cache_extent(reada, bits[i].start,
5327 /* fixme, get the parent transid */
5328 readahead_tree_block(root, bits[i].start,
5332 *last = bits[0].start;
5333 bytenr = bits[0].start;
5334 size = bits[0].size;
5336 cache = lookup_cache_extent(pending, bytenr, size);
5338 remove_cache_extent(pending, cache);
5341 cache = lookup_cache_extent(reada, bytenr, size);
5343 remove_cache_extent(reada, cache);
5346 cache = lookup_cache_extent(nodes, bytenr, size);
5348 remove_cache_extent(nodes, cache);
5351 cache = lookup_cache_extent(extent_cache, bytenr, size);
5353 struct extent_record *rec;
5355 rec = container_of(cache, struct extent_record, cache);
5356 gen = rec->parent_generation;
5359 /* fixme, get the real parent transid */
5360 buf = read_tree_block(root, bytenr, size, gen);
5361 if (!extent_buffer_uptodate(buf)) {
5362 record_bad_block_io(root->fs_info,
5363 extent_cache, bytenr, size);
5367 nritems = btrfs_header_nritems(buf);
5370 * FIXME, this only works only if we don't have any full
5373 if (!init_extent_tree) {
5374 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
5375 btrfs_header_level(buf), 1, NULL,
5381 ret = calc_extent_flag(root, extent_cache, buf, ri, &flags);
5386 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5391 owner = btrfs_header_owner(buf);
5394 ret = check_block(trans, root, extent_cache, buf, flags);
5398 if (btrfs_is_leaf(buf)) {
5399 btree_space_waste += btrfs_leaf_free_space(root, buf);
5400 for (i = 0; i < nritems; i++) {
5401 struct btrfs_file_extent_item *fi;
5402 btrfs_item_key_to_cpu(buf, &key, i);
5403 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
5404 process_extent_item(root, extent_cache, buf,
5408 if (key.type == BTRFS_METADATA_ITEM_KEY) {
5409 process_extent_item(root, extent_cache, buf,
5413 if (key.type == BTRFS_EXTENT_CSUM_KEY) {
5415 btrfs_item_size_nr(buf, i);
5418 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5419 process_chunk_item(chunk_cache, &key, buf, i);
5422 if (key.type == BTRFS_DEV_ITEM_KEY) {
5423 process_device_item(dev_cache, &key, buf, i);
5426 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5427 process_block_group_item(block_group_cache,
5431 if (key.type == BTRFS_DEV_EXTENT_KEY) {
5432 process_device_extent_item(dev_extent_cache,
5437 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
5438 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5439 process_extent_ref_v0(extent_cache, buf, i);
5446 if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
5447 add_tree_backref(extent_cache, key.objectid, 0,
5451 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
5452 add_tree_backref(extent_cache, key.objectid,
5456 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
5457 struct btrfs_extent_data_ref *ref;
5458 ref = btrfs_item_ptr(buf, i,
5459 struct btrfs_extent_data_ref);
5460 add_data_backref(extent_cache,
5462 btrfs_extent_data_ref_root(buf, ref),
5463 btrfs_extent_data_ref_objectid(buf,
5465 btrfs_extent_data_ref_offset(buf, ref),
5466 btrfs_extent_data_ref_count(buf, ref),
5467 0, root->sectorsize);
5470 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
5471 struct btrfs_shared_data_ref *ref;
5472 ref = btrfs_item_ptr(buf, i,
5473 struct btrfs_shared_data_ref);
5474 add_data_backref(extent_cache,
5475 key.objectid, key.offset, 0, 0, 0,
5476 btrfs_shared_data_ref_count(buf, ref),
5477 0, root->sectorsize);
5480 if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
5481 struct bad_item *bad;
5483 if (key.objectid == BTRFS_ORPHAN_OBJECTID)
5487 bad = malloc(sizeof(struct bad_item));
5490 INIT_LIST_HEAD(&bad->list);
5491 memcpy(&bad->key, &key,
5492 sizeof(struct btrfs_key));
5493 bad->root_id = owner;
5494 list_add_tail(&bad->list, &delete_items);
5497 if (key.type != BTRFS_EXTENT_DATA_KEY)
5499 fi = btrfs_item_ptr(buf, i,
5500 struct btrfs_file_extent_item);
5501 if (btrfs_file_extent_type(buf, fi) ==
5502 BTRFS_FILE_EXTENT_INLINE)
5504 if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
5507 data_bytes_allocated +=
5508 btrfs_file_extent_disk_num_bytes(buf, fi);
5509 if (data_bytes_allocated < root->sectorsize) {
5512 data_bytes_referenced +=
5513 btrfs_file_extent_num_bytes(buf, fi);
5514 add_data_backref(extent_cache,
5515 btrfs_file_extent_disk_bytenr(buf, fi),
5516 parent, owner, key.objectid, key.offset -
5517 btrfs_file_extent_offset(buf, fi), 1, 1,
5518 btrfs_file_extent_disk_num_bytes(buf, fi));
5522 struct btrfs_key first_key;
5524 first_key.objectid = 0;
5527 btrfs_item_key_to_cpu(buf, &first_key, 0);
5528 level = btrfs_header_level(buf);
5529 for (i = 0; i < nritems; i++) {
5530 ptr = btrfs_node_blockptr(buf, i);
5531 size = btrfs_level_size(root, level - 1);
5532 btrfs_node_key_to_cpu(buf, &key, i);
5534 if ((level == ri->drop_level)
5535 && is_dropped_key(&key, &ri->drop_key)) {
5539 ret = add_extent_rec(extent_cache, &key,
5540 btrfs_node_ptr_generation(buf, i),
5541 ptr, size, 0, 0, 1, 0, 1, 0,
5545 add_tree_backref(extent_cache, ptr, parent, owner, 1);
5548 add_pending(nodes, seen, ptr, size);
5550 add_pending(pending, seen, ptr, size);
5553 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
5554 nritems) * sizeof(struct btrfs_key_ptr);
5556 total_btree_bytes += buf->len;
5557 if (fs_root_objectid(btrfs_header_owner(buf)))
5558 total_fs_tree_bytes += buf->len;
5559 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
5560 total_extent_tree_bytes += buf->len;
5561 if (!found_old_backref &&
5562 btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
5563 btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
5564 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
5565 found_old_backref = 1;
5567 free_extent_buffer(buf);
5571 static int add_root_to_pending(struct extent_buffer *buf,
5572 struct cache_tree *extent_cache,
5573 struct cache_tree *pending,
5574 struct cache_tree *seen,
5575 struct cache_tree *nodes,
5578 if (btrfs_header_level(buf) > 0)
5579 add_pending(nodes, seen, buf->start, buf->len);
5581 add_pending(pending, seen, buf->start, buf->len);
5582 add_extent_rec(extent_cache, NULL, 0, buf->start, buf->len,
5583 0, 1, 1, 0, 1, 0, buf->len);
5585 if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
5586 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
5587 add_tree_backref(extent_cache, buf->start, buf->start,
5590 add_tree_backref(extent_cache, buf->start, 0, objectid, 1);
5594 /* as we fix the tree, we might be deleting blocks that
5595 * we're tracking for repair. This hook makes sure we
5596 * remove any backrefs for blocks as we are fixing them.
5598 static int free_extent_hook(struct btrfs_trans_handle *trans,
5599 struct btrfs_root *root,
5600 u64 bytenr, u64 num_bytes, u64 parent,
5601 u64 root_objectid, u64 owner, u64 offset,
5604 struct extent_record *rec;
5605 struct cache_extent *cache;
5607 struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
5609 is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
5610 cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
5614 rec = container_of(cache, struct extent_record, cache);
5616 struct data_backref *back;
5617 back = find_data_backref(rec, parent, root_objectid, owner,
5618 offset, 1, bytenr, num_bytes);
5621 if (back->node.found_ref) {
5622 back->found_ref -= refs_to_drop;
5624 rec->refs -= refs_to_drop;
5626 if (back->node.found_extent_tree) {
5627 back->num_refs -= refs_to_drop;
5628 if (rec->extent_item_refs)
5629 rec->extent_item_refs -= refs_to_drop;
5631 if (back->found_ref == 0)
5632 back->node.found_ref = 0;
5633 if (back->num_refs == 0)
5634 back->node.found_extent_tree = 0;
5636 if (!back->node.found_extent_tree && back->node.found_ref) {
5637 list_del(&back->node.list);
5641 struct tree_backref *back;
5642 back = find_tree_backref(rec, parent, root_objectid);
5645 if (back->node.found_ref) {
5648 back->node.found_ref = 0;
5650 if (back->node.found_extent_tree) {
5651 if (rec->extent_item_refs)
5652 rec->extent_item_refs--;
5653 back->node.found_extent_tree = 0;
5655 if (!back->node.found_extent_tree && back->node.found_ref) {
5656 list_del(&back->node.list);
5660 maybe_free_extent_rec(extent_cache, rec);
5665 static int delete_extent_records(struct btrfs_trans_handle *trans,
5666 struct btrfs_root *root,
5667 struct btrfs_path *path,
5668 u64 bytenr, u64 new_len)
5670 struct btrfs_key key;
5671 struct btrfs_key found_key;
5672 struct extent_buffer *leaf;
5677 key.objectid = bytenr;
5679 key.offset = (u64)-1;
5682 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
5689 if (path->slots[0] == 0)
5695 leaf = path->nodes[0];
5696 slot = path->slots[0];
5698 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5699 if (found_key.objectid != bytenr)
5702 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
5703 found_key.type != BTRFS_METADATA_ITEM_KEY &&
5704 found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
5705 found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
5706 found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
5707 found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
5708 found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
5709 btrfs_release_path(path);
5710 if (found_key.type == 0) {
5711 if (found_key.offset == 0)
5713 key.offset = found_key.offset - 1;
5714 key.type = found_key.type;
5716 key.type = found_key.type - 1;
5717 key.offset = (u64)-1;
5721 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
5722 found_key.objectid, found_key.type, found_key.offset);
5724 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
5727 btrfs_release_path(path);
5729 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5730 found_key.type == BTRFS_METADATA_ITEM_KEY) {
5731 u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
5732 found_key.offset : root->leafsize;
5734 ret = btrfs_update_block_group(trans, root, bytenr,
5741 btrfs_release_path(path);
5746 * for a single backref, this will allocate a new extent
5747 * and add the backref to it.
5749 static int record_extent(struct btrfs_trans_handle *trans,
5750 struct btrfs_fs_info *info,
5751 struct btrfs_path *path,
5752 struct extent_record *rec,
5753 struct extent_backref *back,
5754 int allocated, u64 flags)
5757 struct btrfs_root *extent_root = info->extent_root;
5758 struct extent_buffer *leaf;
5759 struct btrfs_key ins_key;
5760 struct btrfs_extent_item *ei;
5761 struct tree_backref *tback;
5762 struct data_backref *dback;
5763 struct btrfs_tree_block_info *bi;
5766 rec->max_size = max_t(u64, rec->max_size,
5767 info->extent_root->leafsize);
5770 u32 item_size = sizeof(*ei);
5773 item_size += sizeof(*bi);
5775 ins_key.objectid = rec->start;
5776 ins_key.offset = rec->max_size;
5777 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
5779 ret = btrfs_insert_empty_item(trans, extent_root, path,
5780 &ins_key, item_size);
5784 leaf = path->nodes[0];
5785 ei = btrfs_item_ptr(leaf, path->slots[0],
5786 struct btrfs_extent_item);
5788 btrfs_set_extent_refs(leaf, ei, 0);
5789 btrfs_set_extent_generation(leaf, ei, rec->generation);
5791 if (back->is_data) {
5792 btrfs_set_extent_flags(leaf, ei,
5793 BTRFS_EXTENT_FLAG_DATA);
5795 struct btrfs_disk_key copy_key;;
5797 tback = (struct tree_backref *)back;
5798 bi = (struct btrfs_tree_block_info *)(ei + 1);
5799 memset_extent_buffer(leaf, 0, (unsigned long)bi,
5802 btrfs_set_disk_key_objectid(©_key,
5803 rec->info_objectid);
5804 btrfs_set_disk_key_type(©_key, 0);
5805 btrfs_set_disk_key_offset(©_key, 0);
5807 btrfs_set_tree_block_level(leaf, bi, rec->info_level);
5808 btrfs_set_tree_block_key(leaf, bi, ©_key);
5810 btrfs_set_extent_flags(leaf, ei,
5811 BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
5814 btrfs_mark_buffer_dirty(leaf);
5815 ret = btrfs_update_block_group(trans, extent_root, rec->start,
5816 rec->max_size, 1, 0);
5819 btrfs_release_path(path);
5822 if (back->is_data) {
5826 dback = (struct data_backref *)back;
5827 if (back->full_backref)
5828 parent = dback->parent;
5832 for (i = 0; i < dback->found_ref; i++) {
5833 /* if parent != 0, we're doing a full backref
5834 * passing BTRFS_FIRST_FREE_OBJECTID as the owner
5835 * just makes the backref allocator create a data
5838 ret = btrfs_inc_extent_ref(trans, info->extent_root,
5839 rec->start, rec->max_size,
5843 BTRFS_FIRST_FREE_OBJECTID :
5849 fprintf(stderr, "adding new data backref"
5850 " on %llu %s %llu owner %llu"
5851 " offset %llu found %d\n",
5852 (unsigned long long)rec->start,
5853 back->full_backref ?
5855 back->full_backref ?
5856 (unsigned long long)parent :
5857 (unsigned long long)dback->root,
5858 (unsigned long long)dback->owner,
5859 (unsigned long long)dback->offset,
5864 tback = (struct tree_backref *)back;
5865 if (back->full_backref)
5866 parent = tback->parent;
5870 ret = btrfs_inc_extent_ref(trans, info->extent_root,
5871 rec->start, rec->max_size,
5872 parent, tback->root, 0, 0);
5873 fprintf(stderr, "adding new tree backref on "
5874 "start %llu len %llu parent %llu root %llu\n",
5875 rec->start, rec->max_size, tback->parent, tback->root);
5880 btrfs_release_path(path);
5884 struct extent_entry {
5889 struct list_head list;
5892 static struct extent_entry *find_entry(struct list_head *entries,
5893 u64 bytenr, u64 bytes)
5895 struct extent_entry *entry = NULL;
5897 list_for_each_entry(entry, entries, list) {
5898 if (entry->bytenr == bytenr && entry->bytes == bytes)
5905 static struct extent_entry *find_most_right_entry(struct list_head *entries)
5907 struct extent_entry *entry, *best = NULL, *prev = NULL;
5909 list_for_each_entry(entry, entries, list) {
5916 * If there are as many broken entries as entries then we know
5917 * not to trust this particular entry.
5919 if (entry->broken == entry->count)
5923 * If our current entry == best then we can't be sure our best
5924 * is really the best, so we need to keep searching.
5926 if (best && best->count == entry->count) {
5932 /* Prev == entry, not good enough, have to keep searching */
5933 if (!prev->broken && prev->count == entry->count)
5937 best = (prev->count > entry->count) ? prev : entry;
5938 else if (best->count < entry->count)
5946 static int repair_ref(struct btrfs_trans_handle *trans,
5947 struct btrfs_fs_info *info, struct btrfs_path *path,
5948 struct data_backref *dback, struct extent_entry *entry)
5950 struct btrfs_root *root;
5951 struct btrfs_file_extent_item *fi;
5952 struct extent_buffer *leaf;
5953 struct btrfs_key key;
5957 key.objectid = dback->root;
5958 key.type = BTRFS_ROOT_ITEM_KEY;
5959 key.offset = (u64)-1;
5960 root = btrfs_read_fs_root(info, &key);
5962 fprintf(stderr, "Couldn't find root for our ref\n");
5967 * The backref points to the original offset of the extent if it was
5968 * split, so we need to search down to the offset we have and then walk
5969 * forward until we find the backref we're looking for.
5971 key.objectid = dback->owner;
5972 key.type = BTRFS_EXTENT_DATA_KEY;
5973 key.offset = dback->offset;
5974 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5976 fprintf(stderr, "Error looking up ref %d\n", ret);
5981 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5982 ret = btrfs_next_leaf(root, path);
5984 fprintf(stderr, "Couldn't find our ref, next\n");
5988 leaf = path->nodes[0];
5989 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5990 if (key.objectid != dback->owner ||
5991 key.type != BTRFS_EXTENT_DATA_KEY) {
5992 fprintf(stderr, "Couldn't find our ref, search\n");
5995 fi = btrfs_item_ptr(leaf, path->slots[0],
5996 struct btrfs_file_extent_item);
5997 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5998 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6000 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
6005 btrfs_release_path(path);
6008 * Have to make sure that this root gets updated when we commit the
6011 record_root_in_trans(trans, root);
6014 * Ok we have the key of the file extent we want to fix, now we can cow
6015 * down to the thing and fix it.
6017 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6019 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
6020 key.objectid, key.type, key.offset, ret);
6024 fprintf(stderr, "Well that's odd, we just found this key "
6025 "[%Lu, %u, %Lu]\n", key.objectid, key.type,
6029 leaf = path->nodes[0];
6030 fi = btrfs_item_ptr(leaf, path->slots[0],
6031 struct btrfs_file_extent_item);
6033 if (btrfs_file_extent_compression(leaf, fi) &&
6034 dback->disk_bytenr != entry->bytenr) {
6035 fprintf(stderr, "Ref doesn't match the record start and is "
6036 "compressed, please take a btrfs-image of this file "
6037 "system and send it to a btrfs developer so they can "
6038 "complete this functionality for bytenr %Lu\n",
6039 dback->disk_bytenr);
6043 if (dback->node.broken && dback->disk_bytenr != entry->bytenr) {
6044 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6045 } else if (dback->disk_bytenr > entry->bytenr) {
6046 u64 off_diff, offset;
6048 off_diff = dback->disk_bytenr - entry->bytenr;
6049 offset = btrfs_file_extent_offset(leaf, fi);
6050 if (dback->disk_bytenr + offset +
6051 btrfs_file_extent_num_bytes(leaf, fi) >
6052 entry->bytenr + entry->bytes) {
6053 fprintf(stderr, "Ref is past the entry end, please "
6054 "take a btrfs-image of this file system and "
6055 "send it to a btrfs developer, ref %Lu\n",
6056 dback->disk_bytenr);
6060 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6061 btrfs_set_file_extent_offset(leaf, fi, offset);
6062 } else if (dback->disk_bytenr < entry->bytenr) {
6065 offset = btrfs_file_extent_offset(leaf, fi);
6066 if (dback->disk_bytenr + offset < entry->bytenr) {
6067 fprintf(stderr, "Ref is before the entry start, please"
6068 " take a btrfs-image of this file system and "
6069 "send it to a btrfs developer, ref %Lu\n",
6070 dback->disk_bytenr);
6074 offset += dback->disk_bytenr;
6075 offset -= entry->bytenr;
6076 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6077 btrfs_set_file_extent_offset(leaf, fi, offset);
6080 btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
6083 * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
6084 * only do this if we aren't using compression, otherwise it's a
6087 if (!btrfs_file_extent_compression(leaf, fi))
6088 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
6090 printf("ram bytes may be wrong?\n");
6091 btrfs_mark_buffer_dirty(leaf);
6092 btrfs_release_path(path);
6096 static int verify_backrefs(struct btrfs_trans_handle *trans,
6097 struct btrfs_fs_info *info, struct btrfs_path *path,
6098 struct extent_record *rec)
6100 struct extent_backref *back;
6101 struct data_backref *dback;
6102 struct extent_entry *entry, *best = NULL;
6105 int broken_entries = 0;
6110 * Metadata is easy and the backrefs should always agree on bytenr and
6111 * size, if not we've got bigger issues.
6116 list_for_each_entry(back, &rec->backrefs, list) {
6117 if (back->full_backref || !back->is_data)
6120 dback = (struct data_backref *)back;
6123 * We only pay attention to backrefs that we found a real
6126 if (dback->found_ref == 0)
6130 * For now we only catch when the bytes don't match, not the
6131 * bytenr. We can easily do this at the same time, but I want
6132 * to have a fs image to test on before we just add repair
6133 * functionality willy-nilly so we know we won't screw up the
6137 entry = find_entry(&entries, dback->disk_bytenr,
6140 entry = malloc(sizeof(struct extent_entry));
6145 memset(entry, 0, sizeof(*entry));
6146 entry->bytenr = dback->disk_bytenr;
6147 entry->bytes = dback->bytes;
6148 list_add_tail(&entry->list, &entries);
6153 * If we only have on entry we may think the entries agree when
6154 * in reality they don't so we have to do some extra checking.
6156 if (dback->disk_bytenr != rec->start ||
6157 dback->bytes != rec->nr || back->broken)
6168 /* Yay all the backrefs agree, carry on good sir */
6169 if (nr_entries <= 1 && !mismatch)
6172 fprintf(stderr, "attempting to repair backref discrepency for bytenr "
6173 "%Lu\n", rec->start);
6176 * First we want to see if the backrefs can agree amongst themselves who
6177 * is right, so figure out which one of the entries has the highest
6180 best = find_most_right_entry(&entries);
6183 * Ok so we may have an even split between what the backrefs think, so
6184 * this is where we use the extent ref to see what it thinks.
6187 entry = find_entry(&entries, rec->start, rec->nr);
6188 if (!entry && (!broken_entries || !rec->found_rec)) {
6189 fprintf(stderr, "Backrefs don't agree with each other "
6190 "and extent record doesn't agree with anybody,"
6191 " so we can't fix bytenr %Lu bytes %Lu\n",
6192 rec->start, rec->nr);
6195 } else if (!entry) {
6197 * Ok our backrefs were broken, we'll assume this is the
6198 * correct value and add an entry for this range.
6200 entry = malloc(sizeof(struct extent_entry));
6205 memset(entry, 0, sizeof(*entry));
6206 entry->bytenr = rec->start;
6207 entry->bytes = rec->nr;
6208 list_add_tail(&entry->list, &entries);
6212 best = find_most_right_entry(&entries);
6214 fprintf(stderr, "Backrefs and extent record evenly "
6215 "split on who is right, this is going to "
6216 "require user input to fix bytenr %Lu bytes "
6217 "%Lu\n", rec->start, rec->nr);
6224 * I don't think this can happen currently as we'll abort() if we catch
6225 * this case higher up, but in case somebody removes that we still can't
6226 * deal with it properly here yet, so just bail out of that's the case.
6228 if (best->bytenr != rec->start) {
6229 fprintf(stderr, "Extent start and backref starts don't match, "
6230 "please use btrfs-image on this file system and send "
6231 "it to a btrfs developer so they can make fsck fix "
6232 "this particular case. bytenr is %Lu, bytes is %Lu\n",
6233 rec->start, rec->nr);
6239 * Ok great we all agreed on an extent record, let's go find the real
6240 * references and fix up the ones that don't match.
6242 list_for_each_entry(back, &rec->backrefs, list) {
6243 if (back->full_backref || !back->is_data)
6246 dback = (struct data_backref *)back;
6249 * Still ignoring backrefs that don't have a real ref attached
6252 if (dback->found_ref == 0)
6255 if (dback->bytes == best->bytes &&
6256 dback->disk_bytenr == best->bytenr)
6259 ret = repair_ref(trans, info, path, dback, best);
6265 * Ok we messed with the actual refs, which means we need to drop our
6266 * entire cache and go back and rescan. I know this is a huge pain and
6267 * adds a lot of extra work, but it's the only way to be safe. Once all
6268 * the backrefs agree we may not need to do anything to the extent
6273 while (!list_empty(&entries)) {
6274 entry = list_entry(entries.next, struct extent_entry, list);
6275 list_del_init(&entry->list);
6281 static int process_duplicates(struct btrfs_root *root,
6282 struct cache_tree *extent_cache,
6283 struct extent_record *rec)
6285 struct extent_record *good, *tmp;
6286 struct cache_extent *cache;
6290 * If we found a extent record for this extent then return, or if we
6291 * have more than one duplicate we are likely going to need to delete
6294 if (rec->found_rec || rec->num_duplicates > 1)
6297 /* Shouldn't happen but just in case */
6298 BUG_ON(!rec->num_duplicates);
6301 * So this happens if we end up with a backref that doesn't match the
6302 * actual extent entry. So either the backref is bad or the extent
6303 * entry is bad. Either way we want to have the extent_record actually
6304 * reflect what we found in the extent_tree, so we need to take the
6305 * duplicate out and use that as the extent_record since the only way we
6306 * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
6308 remove_cache_extent(extent_cache, &rec->cache);
6310 good = list_entry(rec->dups.next, struct extent_record, list);
6311 list_del_init(&good->list);
6312 INIT_LIST_HEAD(&good->backrefs);
6313 INIT_LIST_HEAD(&good->dups);
6314 good->cache.start = good->start;
6315 good->cache.size = good->nr;
6316 good->content_checked = 0;
6317 good->owner_ref_checked = 0;
6318 good->num_duplicates = 0;
6319 good->refs = rec->refs;
6320 list_splice_init(&rec->backrefs, &good->backrefs);
6322 cache = lookup_cache_extent(extent_cache, good->start,
6326 tmp = container_of(cache, struct extent_record, cache);
6329 * If we find another overlapping extent and it's found_rec is
6330 * set then it's a duplicate and we need to try and delete
6333 if (tmp->found_rec || tmp->num_duplicates > 0) {
6334 if (list_empty(&good->list))
6335 list_add_tail(&good->list,
6336 &duplicate_extents);
6337 good->num_duplicates += tmp->num_duplicates + 1;
6338 list_splice_init(&tmp->dups, &good->dups);
6339 list_del_init(&tmp->list);
6340 list_add_tail(&tmp->list, &good->dups);
6341 remove_cache_extent(extent_cache, &tmp->cache);
6346 * Ok we have another non extent item backed extent rec, so lets
6347 * just add it to this extent and carry on like we did above.
6349 good->refs += tmp->refs;
6350 list_splice_init(&tmp->backrefs, &good->backrefs);
6351 remove_cache_extent(extent_cache, &tmp->cache);
6354 ret = insert_cache_extent(extent_cache, &good->cache);
6357 return good->num_duplicates ? 0 : 1;
6360 static int delete_duplicate_records(struct btrfs_trans_handle *trans,
6361 struct btrfs_root *root,
6362 struct extent_record *rec)
6364 LIST_HEAD(delete_list);
6365 struct btrfs_path *path;
6366 struct extent_record *tmp, *good, *n;
6369 struct btrfs_key key;
6371 path = btrfs_alloc_path();
6378 /* Find the record that covers all of the duplicates. */
6379 list_for_each_entry(tmp, &rec->dups, list) {
6380 if (good->start < tmp->start)
6382 if (good->nr > tmp->nr)
6385 if (tmp->start + tmp->nr < good->start + good->nr) {
6386 fprintf(stderr, "Ok we have overlapping extents that "
6387 "aren't completely covered by eachother, this "
6388 "is going to require more careful thought. "
6389 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
6390 tmp->start, tmp->nr, good->start, good->nr);
6397 list_add_tail(&rec->list, &delete_list);
6399 list_for_each_entry_safe(tmp, n, &rec->dups, list) {
6402 list_move_tail(&tmp->list, &delete_list);
6405 root = root->fs_info->extent_root;
6406 list_for_each_entry(tmp, &delete_list, list) {
6407 if (tmp->found_rec == 0)
6409 key.objectid = tmp->start;
6410 key.type = BTRFS_EXTENT_ITEM_KEY;
6411 key.offset = tmp->nr;
6413 /* Shouldn't happen but just in case */
6414 if (tmp->metadata) {
6415 fprintf(stderr, "Well this shouldn't happen, extent "
6416 "record overlaps but is metadata? "
6417 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
6421 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6427 ret = btrfs_del_item(trans, root, path);
6430 btrfs_release_path(path);
6435 while (!list_empty(&delete_list)) {
6436 tmp = list_entry(delete_list.next, struct extent_record, list);
6437 list_del_init(&tmp->list);
6443 while (!list_empty(&rec->dups)) {
6444 tmp = list_entry(rec->dups.next, struct extent_record, list);
6445 list_del_init(&tmp->list);
6449 btrfs_free_path(path);
6451 if (!ret && !nr_del)
6452 rec->num_duplicates = 0;
6454 return ret ? ret : nr_del;
6457 static int find_possible_backrefs(struct btrfs_trans_handle *trans,
6458 struct btrfs_fs_info *info,
6459 struct btrfs_path *path,
6460 struct cache_tree *extent_cache,
6461 struct extent_record *rec)
6463 struct btrfs_root *root;
6464 struct extent_backref *back;
6465 struct data_backref *dback;
6466 struct cache_extent *cache;
6467 struct btrfs_file_extent_item *fi;
6468 struct btrfs_key key;
6472 list_for_each_entry(back, &rec->backrefs, list) {
6473 /* Don't care about full backrefs (poor unloved backrefs) */
6474 if (back->full_backref || !back->is_data)
6477 dback = (struct data_backref *)back;
6479 /* We found this one, we don't need to do a lookup */
6480 if (dback->found_ref)
6483 key.objectid = dback->root;
6484 key.type = BTRFS_ROOT_ITEM_KEY;
6485 key.offset = (u64)-1;
6487 root = btrfs_read_fs_root(info, &key);
6489 /* No root, definitely a bad ref, skip */
6490 if (IS_ERR(root) && PTR_ERR(root) == -ENOENT)
6492 /* Other err, exit */
6494 return PTR_ERR(root);
6496 key.objectid = dback->owner;
6497 key.type = BTRFS_EXTENT_DATA_KEY;
6498 key.offset = dback->offset;
6499 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6501 btrfs_release_path(path);
6504 /* Didn't find it, we can carry on */
6509 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6510 struct btrfs_file_extent_item);
6511 bytenr = btrfs_file_extent_disk_bytenr(path->nodes[0], fi);
6512 bytes = btrfs_file_extent_disk_num_bytes(path->nodes[0], fi);
6513 btrfs_release_path(path);
6514 cache = lookup_cache_extent(extent_cache, bytenr, 1);
6516 struct extent_record *tmp;
6517 tmp = container_of(cache, struct extent_record, cache);
6520 * If we found an extent record for the bytenr for this
6521 * particular backref then we can't add it to our
6522 * current extent record. We only want to add backrefs
6523 * that don't have a corresponding extent item in the
6524 * extent tree since they likely belong to this record
6525 * and we need to fix it if it doesn't match bytenrs.
6531 dback->found_ref += 1;
6532 dback->disk_bytenr = bytenr;
6533 dback->bytes = bytes;
6536 * Set this so the verify backref code knows not to trust the
6537 * values in this backref.
6546 * when an incorrect extent item is found, this will delete
6547 * all of the existing entries for it and recreate them
6548 * based on what the tree scan found.
6550 static int fixup_extent_refs(struct btrfs_trans_handle *trans,
6551 struct btrfs_fs_info *info,
6552 struct cache_tree *extent_cache,
6553 struct extent_record *rec)
6556 struct btrfs_path *path;
6557 struct list_head *cur = rec->backrefs.next;
6558 struct cache_extent *cache;
6559 struct extent_backref *back;
6564 * remember our flags for recreating the extent.
6565 * FIXME, if we have cleared extent tree, we can not
6566 * lookup extent info in extent tree.
6568 if (!init_extent_tree) {
6569 ret = btrfs_lookup_extent_info(NULL, info->extent_root,
6570 rec->start, rec->max_size,
6571 rec->metadata, NULL, &flags);
6575 if (rec->flag_block_full_backref)
6576 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6579 path = btrfs_alloc_path();
6583 if (rec->refs != rec->extent_item_refs && !rec->metadata) {
6585 * Sometimes the backrefs themselves are so broken they don't
6586 * get attached to any meaningful rec, so first go back and
6587 * check any of our backrefs that we couldn't find and throw
6588 * them into the list if we find the backref so that
6589 * verify_backrefs can figure out what to do.
6591 ret = find_possible_backrefs(trans, info, path, extent_cache,
6597 /* step one, make sure all of the backrefs agree */
6598 ret = verify_backrefs(trans, info, path, rec);
6602 /* step two, delete all the existing records */
6603 ret = delete_extent_records(trans, info->extent_root, path,
6604 rec->start, rec->max_size);
6609 /* was this block corrupt? If so, don't add references to it */
6610 cache = lookup_cache_extent(info->corrupt_blocks,
6611 rec->start, rec->max_size);
6617 /* step three, recreate all the refs we did find */
6618 while(cur != &rec->backrefs) {
6619 back = list_entry(cur, struct extent_backref, list);
6623 * if we didn't find any references, don't create a
6626 if (!back->found_ref)
6629 ret = record_extent(trans, info, path, rec, back, allocated, flags);
6636 btrfs_free_path(path);
6640 /* right now we only prune from the extent allocation tree */
6641 static int prune_one_block(struct btrfs_trans_handle *trans,
6642 struct btrfs_fs_info *info,
6643 struct btrfs_corrupt_block *corrupt)
6646 struct btrfs_path path;
6647 struct extent_buffer *eb;
6651 int level = corrupt->level + 1;
6653 btrfs_init_path(&path);
6655 /* we want to stop at the parent to our busted block */
6656 path.lowest_level = level;
6658 ret = btrfs_search_slot(trans, info->extent_root,
6659 &corrupt->key, &path, -1, 1);
6664 eb = path.nodes[level];
6671 * hopefully the search gave us the block we want to prune,
6672 * lets try that first
6674 slot = path.slots[level];
6675 found = btrfs_node_blockptr(eb, slot);
6676 if (found == corrupt->cache.start)
6679 nritems = btrfs_header_nritems(eb);
6681 /* the search failed, lets scan this node and hope we find it */
6682 for (slot = 0; slot < nritems; slot++) {
6683 found = btrfs_node_blockptr(eb, slot);
6684 if (found == corrupt->cache.start)
6688 * we couldn't find the bad block. TODO, search all the nodes for pointers
6691 if (eb == info->extent_root->node) {
6696 btrfs_release_path(&path);
6701 printk("deleting pointer to block %Lu\n", corrupt->cache.start);
6702 ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
6705 btrfs_release_path(&path);
6709 static int prune_corrupt_blocks(struct btrfs_trans_handle *trans,
6710 struct btrfs_fs_info *info)
6712 struct cache_extent *cache;
6713 struct btrfs_corrupt_block *corrupt;
6715 cache = search_cache_extent(info->corrupt_blocks, 0);
6719 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
6720 prune_one_block(trans, info, corrupt);
6721 cache = next_cache_extent(cache);
6726 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
6728 struct btrfs_block_group_cache *cache;
6733 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
6734 &start, &end, EXTENT_DIRTY);
6737 clear_extent_dirty(&fs_info->free_space_cache, start, end,
6743 cache = btrfs_lookup_first_block_group(fs_info, start);
6748 start = cache->key.objectid + cache->key.offset;
6752 static int check_extent_refs(struct btrfs_trans_handle *trans,
6753 struct btrfs_root *root,
6754 struct cache_tree *extent_cache)
6756 struct extent_record *rec;
6757 struct cache_extent *cache;
6765 * if we're doing a repair, we have to make sure
6766 * we don't allocate from the problem extents.
6767 * In the worst case, this will be all the
6770 cache = search_cache_extent(extent_cache, 0);
6772 rec = container_of(cache, struct extent_record, cache);
6773 btrfs_pin_extent(root->fs_info,
6774 rec->start, rec->max_size);
6775 cache = next_cache_extent(cache);
6778 /* pin down all the corrupted blocks too */
6779 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
6781 btrfs_pin_extent(root->fs_info,
6782 cache->start, cache->size);
6783 cache = next_cache_extent(cache);
6785 prune_corrupt_blocks(trans, root->fs_info);
6786 reset_cached_block_groups(root->fs_info);
6790 * We need to delete any duplicate entries we find first otherwise we
6791 * could mess up the extent tree when we have backrefs that actually
6792 * belong to a different extent item and not the weird duplicate one.
6794 while (repair && !list_empty(&duplicate_extents)) {
6795 rec = list_entry(duplicate_extents.next, struct extent_record,
6797 list_del_init(&rec->list);
6799 /* Sometimes we can find a backref before we find an actual
6800 * extent, so we need to process it a little bit to see if there
6801 * truly are multiple EXTENT_ITEM_KEY's for the same range, or
6802 * if this is a backref screwup. If we need to delete stuff
6803 * process_duplicates() will return 0, otherwise it will return
6806 if (process_duplicates(root, extent_cache, rec))
6808 ret = delete_duplicate_records(trans, root, rec);
6812 * delete_duplicate_records will return the number of entries
6813 * deleted, so if it's greater than 0 then we know we actually
6814 * did something and we need to remove.
6825 cache = search_cache_extent(extent_cache, 0);
6828 rec = container_of(cache, struct extent_record, cache);
6829 if (rec->num_duplicates) {
6830 fprintf(stderr, "extent item %llu has multiple extent "
6831 "items\n", (unsigned long long)rec->start);
6835 if (rec->refs != rec->extent_item_refs) {
6836 fprintf(stderr, "ref mismatch on [%llu %llu] ",
6837 (unsigned long long)rec->start,
6838 (unsigned long long)rec->nr);
6839 fprintf(stderr, "extent item %llu, found %llu\n",
6840 (unsigned long long)rec->extent_item_refs,
6841 (unsigned long long)rec->refs);
6842 if (!fixed && repair) {
6843 ret = fixup_extent_refs(trans, root->fs_info,
6852 if (all_backpointers_checked(rec, 1)) {
6853 fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
6854 (unsigned long long)rec->start,
6855 (unsigned long long)rec->nr);
6857 if (!fixed && repair) {
6858 ret = fixup_extent_refs(trans, root->fs_info,
6867 if (!rec->owner_ref_checked) {
6868 fprintf(stderr, "owner ref check failed [%llu %llu]\n",
6869 (unsigned long long)rec->start,
6870 (unsigned long long)rec->nr);
6871 if (!fixed && repair) {
6872 ret = fixup_extent_refs(trans, root->fs_info,
6881 remove_cache_extent(extent_cache, cache);
6882 free_all_extent_backrefs(rec);
6887 if (ret && ret != -EAGAIN) {
6888 fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
6891 btrfs_fix_block_accounting(trans, root);
6894 fprintf(stderr, "repaired damaged extent references\n");
6900 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
6904 if (type & BTRFS_BLOCK_GROUP_RAID0) {
6905 stripe_size = length;
6906 stripe_size /= num_stripes;
6907 } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
6908 stripe_size = length * 2;
6909 stripe_size /= num_stripes;
6910 } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
6911 stripe_size = length;
6912 stripe_size /= (num_stripes - 1);
6913 } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
6914 stripe_size = length;
6915 stripe_size /= (num_stripes - 2);
6917 stripe_size = length;
6923 * Check the chunk with its block group/dev list ref:
6924 * Return 0 if all refs seems valid.
6925 * Return 1 if part of refs seems valid, need later check for rebuild ref
6926 * like missing block group and needs to search extent tree to rebuild them.
6927 * Return -1 if essential refs are missing and unable to rebuild.
6929 static int check_chunk_refs(struct chunk_record *chunk_rec,
6930 struct block_group_tree *block_group_cache,
6931 struct device_extent_tree *dev_extent_cache,
6934 struct cache_extent *block_group_item;
6935 struct block_group_record *block_group_rec;
6936 struct cache_extent *dev_extent_item;
6937 struct device_extent_record *dev_extent_rec;
6944 block_group_item = lookup_cache_extent(&block_group_cache->tree,
6947 if (block_group_item) {
6948 block_group_rec = container_of(block_group_item,
6949 struct block_group_record,
6951 if (chunk_rec->length != block_group_rec->offset ||
6952 chunk_rec->offset != block_group_rec->objectid ||
6953 chunk_rec->type_flags != block_group_rec->flags) {
6956 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
6957 chunk_rec->objectid,
6962 chunk_rec->type_flags,
6963 block_group_rec->objectid,
6964 block_group_rec->type,
6965 block_group_rec->offset,
6966 block_group_rec->offset,
6967 block_group_rec->objectid,
6968 block_group_rec->flags);
6971 list_del_init(&block_group_rec->list);
6972 chunk_rec->bg_rec = block_group_rec;
6977 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
6978 chunk_rec->objectid,
6983 chunk_rec->type_flags);
6987 length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
6988 chunk_rec->num_stripes);
6989 for (i = 0; i < chunk_rec->num_stripes; ++i) {
6990 devid = chunk_rec->stripes[i].devid;
6991 offset = chunk_rec->stripes[i].offset;
6992 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
6993 devid, offset, length);
6994 if (dev_extent_item) {
6995 dev_extent_rec = container_of(dev_extent_item,
6996 struct device_extent_record,
6998 if (dev_extent_rec->objectid != devid ||
6999 dev_extent_rec->offset != offset ||
7000 dev_extent_rec->chunk_offset != chunk_rec->offset ||
7001 dev_extent_rec->length != length) {
7004 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
7005 chunk_rec->objectid,
7008 chunk_rec->stripes[i].devid,
7009 chunk_rec->stripes[i].offset,
7010 dev_extent_rec->objectid,
7011 dev_extent_rec->offset,
7012 dev_extent_rec->length);
7015 list_move(&dev_extent_rec->chunk_list,
7016 &chunk_rec->dextents);
7021 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
7022 chunk_rec->objectid,
7025 chunk_rec->stripes[i].devid,
7026 chunk_rec->stripes[i].offset);
7033 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
7034 int check_chunks(struct cache_tree *chunk_cache,
7035 struct block_group_tree *block_group_cache,
7036 struct device_extent_tree *dev_extent_cache,
7037 struct list_head *good, struct list_head *bad,
7038 struct list_head *rebuild, int silent)
7040 struct cache_extent *chunk_item;
7041 struct chunk_record *chunk_rec;
7042 struct block_group_record *bg_rec;
7043 struct device_extent_record *dext_rec;
7047 chunk_item = first_cache_extent(chunk_cache);
7048 while (chunk_item) {
7049 chunk_rec = container_of(chunk_item, struct chunk_record,
7051 err = check_chunk_refs(chunk_rec, block_group_cache,
7052 dev_extent_cache, silent);
7055 if (err == 0 && good)
7056 list_add_tail(&chunk_rec->list, good);
7057 if (err > 0 && rebuild)
7058 list_add_tail(&chunk_rec->list, rebuild);
7060 list_add_tail(&chunk_rec->list, bad);
7061 chunk_item = next_cache_extent(chunk_item);
7064 list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
7067 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
7075 list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
7079 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
7090 static int check_device_used(struct device_record *dev_rec,
7091 struct device_extent_tree *dext_cache)
7093 struct cache_extent *cache;
7094 struct device_extent_record *dev_extent_rec;
7097 cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
7099 dev_extent_rec = container_of(cache,
7100 struct device_extent_record,
7102 if (dev_extent_rec->objectid != dev_rec->devid)
7105 list_del_init(&dev_extent_rec->device_list);
7106 total_byte += dev_extent_rec->length;
7107 cache = next_cache_extent(cache);
7110 if (total_byte != dev_rec->byte_used) {
7112 "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
7113 total_byte, dev_rec->byte_used, dev_rec->objectid,
7114 dev_rec->type, dev_rec->offset);
7121 /* check btrfs_dev_item -> btrfs_dev_extent */
7122 static int check_devices(struct rb_root *dev_cache,
7123 struct device_extent_tree *dev_extent_cache)
7125 struct rb_node *dev_node;
7126 struct device_record *dev_rec;
7127 struct device_extent_record *dext_rec;
7131 dev_node = rb_first(dev_cache);
7133 dev_rec = container_of(dev_node, struct device_record, node);
7134 err = check_device_used(dev_rec, dev_extent_cache);
7138 dev_node = rb_next(dev_node);
7140 list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
7143 "Device extent[%llu, %llu, %llu] didn't find its device.\n",
7144 dext_rec->objectid, dext_rec->offset, dext_rec->length);
7151 static int add_root_item_to_list(struct list_head *head,
7152 u64 objectid, u64 bytenr,
7153 u8 level, u8 drop_level,
7154 int level_size, struct btrfs_key *drop_key)
7157 struct root_item_record *ri_rec;
7158 ri_rec = malloc(sizeof(*ri_rec));
7161 ri_rec->bytenr = bytenr;
7162 ri_rec->objectid = objectid;
7163 ri_rec->level = level;
7164 ri_rec->level_size = level_size;
7165 ri_rec->drop_level = drop_level;
7167 memcpy(&ri_rec->drop_key, drop_key, sizeof(*drop_key));
7168 list_add_tail(&ri_rec->list, head);
7173 static int deal_root_from_list(struct list_head *list,
7174 struct btrfs_trans_handle *trans,
7175 struct btrfs_root *root,
7176 struct block_info *bits,
7178 struct cache_tree *pending,
7179 struct cache_tree *seen,
7180 struct cache_tree *reada,
7181 struct cache_tree *nodes,
7182 struct cache_tree *extent_cache,
7183 struct cache_tree *chunk_cache,
7184 struct rb_root *dev_cache,
7185 struct block_group_tree *block_group_cache,
7186 struct device_extent_tree *dev_extent_cache)
7191 while (!list_empty(list)) {
7192 struct root_item_record *rec;
7193 struct extent_buffer *buf;
7194 rec = list_entry(list->next,
7195 struct root_item_record, list);
7197 buf = read_tree_block(root->fs_info->tree_root,
7198 rec->bytenr, rec->level_size, 0);
7199 if (!extent_buffer_uptodate(buf)) {
7200 free_extent_buffer(buf);
7204 add_root_to_pending(buf, extent_cache, pending,
7205 seen, nodes, rec->objectid);
7207 * To rebuild extent tree, we need deal with snapshot
7208 * one by one, otherwise we deal with node firstly which
7209 * can maximize readahead.
7211 if (!init_extent_tree && !rec->drop_level)
7214 ret = run_next_block(trans, root, bits, bits_nr, &last,
7215 pending, seen, reada,
7216 nodes, extent_cache,
7217 chunk_cache, dev_cache,
7219 dev_extent_cache, rec);
7224 free_extent_buffer(buf);
7225 list_del(&rec->list);
7229 ret = run_next_block(trans, root, bits, bits_nr, &last,
7230 pending, seen, reada,
7231 nodes, extent_cache,
7232 chunk_cache, dev_cache,
7234 dev_extent_cache, NULL);
7244 static int check_chunks_and_extents(struct btrfs_root *root)
7246 struct rb_root dev_cache;
7247 struct cache_tree chunk_cache;
7248 struct block_group_tree block_group_cache;
7249 struct device_extent_tree dev_extent_cache;
7250 struct cache_tree extent_cache;
7251 struct cache_tree seen;
7252 struct cache_tree pending;
7253 struct cache_tree reada;
7254 struct cache_tree nodes;
7255 struct cache_tree corrupt_blocks;
7256 struct btrfs_path path;
7257 struct btrfs_key key;
7258 struct btrfs_key found_key;
7260 struct block_info *bits;
7262 struct extent_buffer *leaf;
7263 struct btrfs_trans_handle *trans = NULL;
7265 struct btrfs_root_item ri;
7266 struct list_head dropping_trees;
7267 struct list_head normal_trees;
7268 struct btrfs_root *root1;
7273 dev_cache = RB_ROOT;
7274 cache_tree_init(&chunk_cache);
7275 block_group_tree_init(&block_group_cache);
7276 device_extent_tree_init(&dev_extent_cache);
7278 cache_tree_init(&extent_cache);
7279 cache_tree_init(&seen);
7280 cache_tree_init(&pending);
7281 cache_tree_init(&nodes);
7282 cache_tree_init(&reada);
7283 cache_tree_init(&corrupt_blocks);
7284 INIT_LIST_HEAD(&dropping_trees);
7285 INIT_LIST_HEAD(&normal_trees);
7288 trans = btrfs_start_transaction(root, 1);
7289 if (IS_ERR(trans)) {
7290 fprintf(stderr, "Error starting transaction\n");
7291 return PTR_ERR(trans);
7293 root->fs_info->fsck_extent_cache = &extent_cache;
7294 root->fs_info->free_extent_hook = free_extent_hook;
7295 root->fs_info->corrupt_blocks = &corrupt_blocks;
7299 bits = malloc(bits_nr * sizeof(struct block_info));
7306 root1 = root->fs_info->tree_root;
7307 level = btrfs_header_level(root1->node);
7308 ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
7309 root1->node->start, level, 0,
7310 btrfs_level_size(root1, level), NULL);
7313 root1 = root->fs_info->chunk_root;
7314 level = btrfs_header_level(root1->node);
7315 ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
7316 root1->node->start, level, 0,
7317 btrfs_level_size(root1, level), NULL);
7320 btrfs_init_path(&path);
7323 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
7324 ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
7329 leaf = path.nodes[0];
7330 slot = path.slots[0];
7331 if (slot >= btrfs_header_nritems(path.nodes[0])) {
7332 ret = btrfs_next_leaf(root, &path);
7335 leaf = path.nodes[0];
7336 slot = path.slots[0];
7338 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
7339 if (btrfs_key_type(&found_key) == BTRFS_ROOT_ITEM_KEY) {
7340 unsigned long offset;
7342 offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
7343 read_extent_buffer(leaf, &ri, offset, sizeof(ri));
7344 if (btrfs_disk_key_objectid(&ri.drop_progress) == 0) {
7345 level = btrfs_root_level(&ri);
7346 level_size = btrfs_level_size(root, level);
7347 ret = add_root_item_to_list(&normal_trees,
7349 btrfs_root_bytenr(&ri), level,
7350 0, level_size, NULL);
7354 level = btrfs_root_level(&ri);
7355 level_size = btrfs_level_size(root, level);
7356 objectid = found_key.objectid;
7357 btrfs_disk_key_to_cpu(&found_key,
7359 ret = add_root_item_to_list(&dropping_trees,
7361 btrfs_root_bytenr(&ri),
7362 level, ri.drop_level,
7363 level_size, &found_key);
7370 btrfs_release_path(&path);
7371 ret = deal_root_from_list(&normal_trees, trans, root,
7372 bits, bits_nr, &pending, &seen,
7373 &reada, &nodes, &extent_cache,
7374 &chunk_cache, &dev_cache, &block_group_cache,
7378 ret = deal_root_from_list(&dropping_trees, trans, root,
7379 bits, bits_nr, &pending, &seen,
7380 &reada, &nodes, &extent_cache,
7381 &chunk_cache, &dev_cache, &block_group_cache,
7386 ret = check_extent_refs(trans, root, &extent_cache);
7387 if (ret == -EAGAIN) {
7388 ret = btrfs_commit_transaction(trans, root);
7392 trans = btrfs_start_transaction(root, 1);
7393 if (IS_ERR(trans)) {
7394 ret = PTR_ERR(trans);
7398 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
7399 free_extent_cache_tree(&seen);
7400 free_extent_cache_tree(&pending);
7401 free_extent_cache_tree(&reada);
7402 free_extent_cache_tree(&nodes);
7403 free_chunk_cache_tree(&chunk_cache);
7404 free_block_group_tree(&block_group_cache);
7405 free_device_cache_tree(&dev_cache);
7406 free_device_extent_tree(&dev_extent_cache);
7407 free_extent_record_cache(root->fs_info, &extent_cache);
7411 err = check_chunks(&chunk_cache, &block_group_cache,
7412 &dev_extent_cache, NULL, NULL, NULL, 0);
7416 err = check_devices(&dev_cache, &dev_extent_cache);
7422 err = btrfs_commit_transaction(trans, root);
7427 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
7428 root->fs_info->fsck_extent_cache = NULL;
7429 root->fs_info->free_extent_hook = NULL;
7430 root->fs_info->corrupt_blocks = NULL;
7433 free_chunk_cache_tree(&chunk_cache);
7434 free_device_cache_tree(&dev_cache);
7435 free_block_group_tree(&block_group_cache);
7436 free_device_extent_tree(&dev_extent_cache);
7437 free_extent_cache_tree(&seen);
7438 free_extent_cache_tree(&pending);
7439 free_extent_cache_tree(&reada);
7440 free_extent_cache_tree(&nodes);
7444 static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
7445 struct btrfs_root *root, int overwrite)
7447 struct extent_buffer *c;
7448 struct extent_buffer *old = root->node;
7451 struct btrfs_disk_key disk_key = {0,0,0};
7457 extent_buffer_get(c);
7460 c = btrfs_alloc_free_block(trans, root,
7461 btrfs_level_size(root, 0),
7462 root->root_key.objectid,
7463 &disk_key, level, 0, 0);
7466 extent_buffer_get(c);
7470 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
7471 btrfs_set_header_level(c, level);
7472 btrfs_set_header_bytenr(c, c->start);
7473 btrfs_set_header_generation(c, trans->transid);
7474 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
7475 btrfs_set_header_owner(c, root->root_key.objectid);
7477 write_extent_buffer(c, root->fs_info->fsid,
7478 btrfs_header_fsid(), BTRFS_FSID_SIZE);
7480 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
7481 btrfs_header_chunk_tree_uuid(c),
7484 btrfs_mark_buffer_dirty(c);
7486 * this case can happen in the following case:
7488 * 1.overwrite previous root.
7490 * 2.reinit reloc data root, this is because we skip pin
7491 * down reloc data tree before which means we can allocate
7492 * same block bytenr here.
7494 if (old->start == c->start) {
7495 btrfs_set_root_generation(&root->root_item,
7497 root->root_item.level = btrfs_header_level(root->node);
7498 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7499 &root->root_key, &root->root_item);
7501 free_extent_buffer(c);
7505 free_extent_buffer(old);
7507 add_root_to_dirty_list(root);
7511 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
7512 struct extent_buffer *eb, int tree_root)
7514 struct extent_buffer *tmp;
7515 struct btrfs_root_item *ri;
7516 struct btrfs_key key;
7519 int level = btrfs_header_level(eb);
7525 * If we have pinned this block before, don't pin it again.
7526 * This can not only avoid forever loop with broken filesystem
7527 * but also give us some speedups.
7529 if (test_range_bit(&fs_info->pinned_extents, eb->start,
7530 eb->start + eb->len - 1, EXTENT_DIRTY, 0))
7533 btrfs_pin_extent(fs_info, eb->start, eb->len);
7535 leafsize = btrfs_super_leafsize(fs_info->super_copy);
7536 nritems = btrfs_header_nritems(eb);
7537 for (i = 0; i < nritems; i++) {
7539 btrfs_item_key_to_cpu(eb, &key, i);
7540 if (key.type != BTRFS_ROOT_ITEM_KEY)
7542 /* Skip the extent root and reloc roots */
7543 if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
7544 key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
7545 key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
7547 ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
7548 bytenr = btrfs_disk_root_bytenr(eb, ri);
7551 * If at any point we start needing the real root we
7552 * will have to build a stump root for the root we are
7553 * in, but for now this doesn't actually use the root so
7554 * just pass in extent_root.
7556 tmp = read_tree_block(fs_info->extent_root, bytenr,
7559 fprintf(stderr, "Error reading root block\n");
7562 ret = pin_down_tree_blocks(fs_info, tmp, 0);
7563 free_extent_buffer(tmp);
7567 bytenr = btrfs_node_blockptr(eb, i);
7569 /* If we aren't the tree root don't read the block */
7570 if (level == 1 && !tree_root) {
7571 btrfs_pin_extent(fs_info, bytenr, leafsize);
7575 tmp = read_tree_block(fs_info->extent_root, bytenr,
7578 fprintf(stderr, "Error reading tree block\n");
7581 ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
7582 free_extent_buffer(tmp);
7591 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
7595 ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
7599 return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
7602 static int reset_block_groups(struct btrfs_fs_info *fs_info)
7604 struct btrfs_block_group_cache *cache;
7605 struct btrfs_path *path;
7606 struct extent_buffer *leaf;
7607 struct btrfs_chunk *chunk;
7608 struct btrfs_key key;
7612 path = btrfs_alloc_path();
7617 key.type = BTRFS_CHUNK_ITEM_KEY;
7620 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
7622 btrfs_free_path(path);
7627 * We do this in case the block groups were screwed up and had alloc
7628 * bits that aren't actually set on the chunks. This happens with
7629 * restored images every time and could happen in real life I guess.
7631 fs_info->avail_data_alloc_bits = 0;
7632 fs_info->avail_metadata_alloc_bits = 0;
7633 fs_info->avail_system_alloc_bits = 0;
7635 /* First we need to create the in-memory block groups */
7637 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7638 ret = btrfs_next_leaf(fs_info->chunk_root, path);
7640 btrfs_free_path(path);
7648 leaf = path->nodes[0];
7649 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7650 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7655 chunk = btrfs_item_ptr(leaf, path->slots[0],
7656 struct btrfs_chunk);
7657 btrfs_add_block_group(fs_info, 0,
7658 btrfs_chunk_type(leaf, chunk),
7659 key.objectid, key.offset,
7660 btrfs_chunk_length(leaf, chunk));
7661 set_extent_dirty(&fs_info->free_space_cache, key.offset,
7662 key.offset + btrfs_chunk_length(leaf, chunk),
7668 cache = btrfs_lookup_first_block_group(fs_info, start);
7672 start = cache->key.objectid + cache->key.offset;
7675 btrfs_free_path(path);
7679 static int reset_balance(struct btrfs_trans_handle *trans,
7680 struct btrfs_fs_info *fs_info)
7682 struct btrfs_root *root = fs_info->tree_root;
7683 struct btrfs_path *path;
7684 struct extent_buffer *leaf;
7685 struct btrfs_key key;
7686 int del_slot, del_nr = 0;
7690 path = btrfs_alloc_path();
7694 key.objectid = BTRFS_BALANCE_OBJECTID;
7695 key.type = BTRFS_BALANCE_ITEM_KEY;
7698 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7703 goto reinit_data_reloc;
7708 ret = btrfs_del_item(trans, root, path);
7711 btrfs_release_path(path);
7713 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
7714 key.type = BTRFS_ROOT_ITEM_KEY;
7717 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7721 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7726 ret = btrfs_del_items(trans, root, path,
7733 btrfs_release_path(path);
7736 ret = btrfs_search_slot(trans, root, &key, path,
7743 leaf = path->nodes[0];
7744 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7745 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
7747 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7752 del_slot = path->slots[0];
7761 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
7765 btrfs_release_path(path);
7768 key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7769 key.type = BTRFS_ROOT_ITEM_KEY;
7770 key.offset = (u64)-1;
7771 root = btrfs_read_fs_root(fs_info, &key);
7773 fprintf(stderr, "Error reading data reloc tree\n");
7774 ret = PTR_ERR(root);
7777 record_root_in_trans(trans, root);
7778 ret = btrfs_fsck_reinit_root(trans, root, 0);
7781 ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
7783 btrfs_free_path(path);
7787 static int reinit_extent_tree(struct btrfs_trans_handle *trans,
7788 struct btrfs_fs_info *fs_info)
7794 * The only reason we don't do this is because right now we're just
7795 * walking the trees we find and pinning down their bytes, we don't look
7796 * at any of the leaves. In order to do mixed groups we'd have to check
7797 * the leaves of any fs roots and pin down the bytes for any file
7798 * extents we find. Not hard but why do it if we don't have to?
7800 if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
7801 fprintf(stderr, "We don't support re-initing the extent tree "
7802 "for mixed block groups yet, please notify a btrfs "
7803 "developer you want to do this so they can add this "
7804 "functionality.\n");
7809 * first we need to walk all of the trees except the extent tree and pin
7810 * down the bytes that are in use so we don't overwrite any existing
7813 ret = pin_metadata_blocks(fs_info);
7815 fprintf(stderr, "error pinning down used bytes\n");
7820 * Need to drop all the block groups since we're going to recreate all
7823 btrfs_free_block_groups(fs_info);
7824 ret = reset_block_groups(fs_info);
7826 fprintf(stderr, "error resetting the block groups\n");
7830 /* Ok we can allocate now, reinit the extent root */
7831 ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 0);
7833 fprintf(stderr, "extent root initialization failed\n");
7835 * When the transaction code is updated we should end the
7836 * transaction, but for now progs only knows about commit so
7837 * just return an error.
7843 * Now we have all the in-memory block groups setup so we can make
7844 * allocations properly, and the metadata we care about is safe since we
7845 * pinned all of it above.
7848 struct btrfs_block_group_cache *cache;
7850 cache = btrfs_lookup_first_block_group(fs_info, start);
7853 start = cache->key.objectid + cache->key.offset;
7854 ret = btrfs_insert_item(trans, fs_info->extent_root,
7855 &cache->key, &cache->item,
7856 sizeof(cache->item));
7858 fprintf(stderr, "Error adding block group\n");
7861 btrfs_extent_post_op(trans, fs_info->extent_root);
7864 ret = reset_balance(trans, fs_info);
7866 fprintf(stderr, "error reseting the pending balance\n");
7871 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
7873 struct btrfs_path *path;
7874 struct btrfs_trans_handle *trans;
7875 struct btrfs_key key;
7878 printf("Recowing metadata block %llu\n", eb->start);
7879 key.objectid = btrfs_header_owner(eb);
7880 key.type = BTRFS_ROOT_ITEM_KEY;
7881 key.offset = (u64)-1;
7883 root = btrfs_read_fs_root(root->fs_info, &key);
7885 fprintf(stderr, "Couldn't find owner root %llu\n",
7887 return PTR_ERR(root);
7890 path = btrfs_alloc_path();
7894 trans = btrfs_start_transaction(root, 1);
7895 if (IS_ERR(trans)) {
7896 btrfs_free_path(path);
7897 return PTR_ERR(trans);
7900 path->lowest_level = btrfs_header_level(eb);
7901 if (path->lowest_level)
7902 btrfs_node_key_to_cpu(eb, &key, 0);
7904 btrfs_item_key_to_cpu(eb, &key, 0);
7906 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7907 btrfs_commit_transaction(trans, root);
7908 btrfs_free_path(path);
7912 static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
7914 struct btrfs_path *path;
7915 struct btrfs_trans_handle *trans;
7916 struct btrfs_key key;
7919 printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
7920 bad->key.type, bad->key.offset);
7921 key.objectid = bad->root_id;
7922 key.type = BTRFS_ROOT_ITEM_KEY;
7923 key.offset = (u64)-1;
7925 root = btrfs_read_fs_root(root->fs_info, &key);
7927 fprintf(stderr, "Couldn't find owner root %llu\n",
7929 return PTR_ERR(root);
7932 path = btrfs_alloc_path();
7936 trans = btrfs_start_transaction(root, 1);
7937 if (IS_ERR(trans)) {
7938 btrfs_free_path(path);
7939 return PTR_ERR(trans);
7942 ret = btrfs_search_slot(trans, root, &bad->key, path, -1, 1);
7948 ret = btrfs_del_item(trans, root, path);
7950 btrfs_commit_transaction(trans, root);
7951 btrfs_free_path(path);
7955 static int zero_log_tree(struct btrfs_root *root)
7957 struct btrfs_trans_handle *trans;
7960 trans = btrfs_start_transaction(root, 1);
7961 if (IS_ERR(trans)) {
7962 ret = PTR_ERR(trans);
7965 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
7966 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
7967 ret = btrfs_commit_transaction(trans, root);
7971 static int populate_csum(struct btrfs_trans_handle *trans,
7972 struct btrfs_root *csum_root, char *buf, u64 start,
7979 while (offset < len) {
7980 sectorsize = csum_root->sectorsize;
7981 ret = read_extent_data(csum_root, buf, start + offset,
7985 ret = btrfs_csum_file_block(trans, csum_root, start + len,
7986 start + offset, buf, sectorsize);
7989 offset += sectorsize;
7994 static int fill_csum_tree(struct btrfs_trans_handle *trans,
7995 struct btrfs_root *csum_root)
7997 struct btrfs_root *extent_root = csum_root->fs_info->extent_root;
7998 struct btrfs_path *path;
7999 struct btrfs_extent_item *ei;
8000 struct extent_buffer *leaf;
8002 struct btrfs_key key;
8005 path = btrfs_alloc_path();
8010 key.type = BTRFS_EXTENT_ITEM_KEY;
8013 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
8015 btrfs_free_path(path);
8019 buf = malloc(csum_root->sectorsize);
8021 btrfs_free_path(path);
8026 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8027 ret = btrfs_next_leaf(extent_root, path);
8035 leaf = path->nodes[0];
8037 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
8038 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
8043 ei = btrfs_item_ptr(leaf, path->slots[0],
8044 struct btrfs_extent_item);
8045 if (!(btrfs_extent_flags(leaf, ei) &
8046 BTRFS_EXTENT_FLAG_DATA)) {
8051 ret = populate_csum(trans, csum_root, buf, key.objectid,
8058 btrfs_free_path(path);
8063 struct root_item_info {
8064 /* level of the root */
8066 /* number of nodes at this level, must be 1 for a root */
8070 struct cache_extent cache_extent;
8073 static struct cache_tree *roots_info_cache = NULL;
8075 static void free_roots_info_cache(void)
8077 if (!roots_info_cache)
8080 while (!cache_tree_empty(roots_info_cache)) {
8081 struct cache_extent *entry;
8082 struct root_item_info *rii;
8084 entry = first_cache_extent(roots_info_cache);
8087 remove_cache_extent(roots_info_cache, entry);
8088 rii = container_of(entry, struct root_item_info, cache_extent);
8092 free(roots_info_cache);
8093 roots_info_cache = NULL;
8096 static int build_roots_info_cache(struct btrfs_fs_info *info)
8099 struct btrfs_key key;
8100 struct extent_buffer *leaf;
8101 struct btrfs_path *path;
8103 if (!roots_info_cache) {
8104 roots_info_cache = malloc(sizeof(*roots_info_cache));
8105 if (!roots_info_cache)
8107 cache_tree_init(roots_info_cache);
8110 path = btrfs_alloc_path();
8115 key.type = BTRFS_EXTENT_ITEM_KEY;
8118 ret = btrfs_search_slot(NULL, info->extent_root, &key, path, 0, 0);
8121 leaf = path->nodes[0];
8124 struct btrfs_key found_key;
8125 struct btrfs_extent_item *ei;
8126 struct btrfs_extent_inline_ref *iref;
8127 int slot = path->slots[0];
8132 struct cache_extent *entry;
8133 struct root_item_info *rii;
8135 if (slot >= btrfs_header_nritems(leaf)) {
8136 ret = btrfs_next_leaf(info->extent_root, path);
8143 leaf = path->nodes[0];
8144 slot = path->slots[0];
8147 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8149 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
8150 found_key.type != BTRFS_METADATA_ITEM_KEY)
8153 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
8154 flags = btrfs_extent_flags(leaf, ei);
8156 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
8157 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
8160 if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
8161 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
8162 level = found_key.offset;
8164 struct btrfs_tree_block_info *info;
8166 info = (struct btrfs_tree_block_info *)(ei + 1);
8167 iref = (struct btrfs_extent_inline_ref *)(info + 1);
8168 level = btrfs_tree_block_level(leaf, info);
8172 * For a root extent, it must be of the following type and the
8173 * first (and only one) iref in the item.
8175 type = btrfs_extent_inline_ref_type(leaf, iref);
8176 if (type != BTRFS_TREE_BLOCK_REF_KEY)
8179 root_id = btrfs_extent_inline_ref_offset(leaf, iref);
8180 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
8182 rii = malloc(sizeof(struct root_item_info));
8187 rii->cache_extent.start = root_id;
8188 rii->cache_extent.size = 1;
8189 rii->level = (u8)-1;
8190 entry = &rii->cache_extent;
8191 ret = insert_cache_extent(roots_info_cache, entry);
8194 rii = container_of(entry, struct root_item_info,
8198 ASSERT(rii->cache_extent.start == root_id);
8199 ASSERT(rii->cache_extent.size == 1);
8201 if (level > rii->level || rii->level == (u8)-1) {
8203 rii->bytenr = found_key.objectid;
8204 rii->gen = btrfs_extent_generation(leaf, ei);
8205 rii->node_count = 1;
8206 } else if (level == rii->level) {
8214 btrfs_free_path(path);
8219 static int maybe_repair_root_item(struct btrfs_fs_info *info,
8220 struct btrfs_path *path,
8221 const struct btrfs_key *root_key,
8222 const int read_only_mode)
8224 const u64 root_id = root_key->objectid;
8225 struct cache_extent *entry;
8226 struct root_item_info *rii;
8227 struct btrfs_root_item ri;
8228 unsigned long offset;
8230 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
8233 "Error: could not find extent items for root %llu\n",
8234 root_key->objectid);
8238 rii = container_of(entry, struct root_item_info, cache_extent);
8239 ASSERT(rii->cache_extent.start == root_id);
8240 ASSERT(rii->cache_extent.size == 1);
8242 if (rii->node_count != 1) {
8244 "Error: could not find btree root extent for root %llu\n",
8249 offset = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
8250 read_extent_buffer(path->nodes[0], &ri, offset, sizeof(ri));
8252 if (btrfs_root_bytenr(&ri) != rii->bytenr ||
8253 btrfs_root_level(&ri) != rii->level ||
8254 btrfs_root_generation(&ri) != rii->gen) {
8257 * If we're in repair mode but our caller told us to not update
8258 * the root item, i.e. just check if it needs to be updated, don't
8259 * print this message, since the caller will call us again shortly
8260 * for the same root item without read only mode (the caller will
8261 * open a transaction first).
8263 if (!(read_only_mode && repair))
8265 "%sroot item for root %llu,"
8266 " current bytenr %llu, current gen %llu, current level %u,"
8267 " new bytenr %llu, new gen %llu, new level %u\n",
8268 (read_only_mode ? "" : "fixing "),
8270 btrfs_root_bytenr(&ri), btrfs_root_generation(&ri),
8271 btrfs_root_level(&ri),
8272 rii->bytenr, rii->gen, rii->level);
8274 if (btrfs_root_generation(&ri) > rii->gen) {
8276 "root %llu has a root item with a more recent gen (%llu) compared to the found root node (%llu)\n",
8277 root_id, btrfs_root_generation(&ri), rii->gen);
8281 if (!read_only_mode) {
8282 btrfs_set_root_bytenr(&ri, rii->bytenr);
8283 btrfs_set_root_level(&ri, rii->level);
8284 btrfs_set_root_generation(&ri, rii->gen);
8285 write_extent_buffer(path->nodes[0], &ri,
8286 offset, sizeof(ri));
8296 * A regression introduced in the 3.17 kernel (more specifically in 3.17-rc2),
8297 * caused read-only snapshots to be corrupted if they were created at a moment
8298 * when the source subvolume/snapshot had orphan items. The issue was that the
8299 * on-disk root items became incorrect, referring to the pre orphan cleanup root
8300 * node instead of the post orphan cleanup root node.
8301 * So this function, and its callees, just detects and fixes those cases. Even
8302 * though the regression was for read-only snapshots, this function applies to
8303 * any snapshot/subvolume root.
8304 * This must be run before any other repair code - not doing it so, makes other
8305 * repair code delete or modify backrefs in the extent tree for example, which
8306 * will result in an inconsistent fs after repairing the root items.
8308 static int repair_root_items(struct btrfs_fs_info *info)
8310 struct btrfs_path *path = NULL;
8311 struct btrfs_key key;
8312 struct extent_buffer *leaf;
8313 struct btrfs_trans_handle *trans = NULL;
8318 ret = build_roots_info_cache(info);
8322 path = btrfs_alloc_path();
8328 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
8329 key.type = BTRFS_ROOT_ITEM_KEY;
8334 * Avoid opening and committing transactions if a leaf doesn't have
8335 * any root items that need to be fixed, so that we avoid rotating
8336 * backup roots unnecessarily.
8339 trans = btrfs_start_transaction(info->tree_root, 1);
8340 if (IS_ERR(trans)) {
8341 ret = PTR_ERR(trans);
8346 ret = btrfs_search_slot(trans, info->tree_root, &key, path,
8350 leaf = path->nodes[0];
8353 struct btrfs_key found_key;
8355 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
8356 int no_more_keys = find_next_key(path, &key);
8358 btrfs_release_path(path);
8360 ret = btrfs_commit_transaction(trans,
8372 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8374 if (found_key.type != BTRFS_ROOT_ITEM_KEY)
8377 ret = maybe_repair_root_item(info, path, &found_key,
8382 if (!trans && repair) {
8385 btrfs_release_path(path);
8395 free_roots_info_cache();
8397 btrfs_free_path(path);
8404 const char * const cmd_check_usage[] = {
8405 "btrfs check [options] <device>",
8406 "Check an unmounted btrfs filesystem.",
8408 "-s|--super <superblock> use this superblock copy",
8409 "-b|--backup use the backup root copy",
8410 "--repair try to repair the filesystem",
8411 "--init-csum-tree create a new CRC tree",
8412 "--init-extent-tree create a new extent tree",
8413 "--check-data-csum verify checkums of data blocks",
8414 "--qgroup-report print a report on qgroup consistency",
8415 "--subvol-extents <subvolid> print subvolume extents and sharing state",
8416 "--tree-root <bytenr> use the given bytenr for the tree root",
8420 int cmd_check(int argc, char **argv)
8422 struct cache_tree root_cache;
8423 struct btrfs_root *root;
8424 struct btrfs_fs_info *info;
8427 u64 tree_root_bytenr = 0;
8428 char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
8431 int init_csum_tree = 0;
8433 int qgroup_report = 0;
8434 enum btrfs_open_ctree_flags ctree_flags = OPEN_CTREE_EXCLUSIVE;
8438 int option_index = 0;
8439 enum { OPT_REPAIR = 257, OPT_INIT_CSUM, OPT_INIT_EXTENT,
8440 OPT_CHECK_CSUM, OPT_READONLY };
8441 static const struct option long_options[] = {
8442 { "super", 1, NULL, 's' },
8443 { "repair", 0, NULL, OPT_REPAIR },
8444 { "readonly", 0, NULL, OPT_READONLY },
8445 { "init-csum-tree", 0, NULL, OPT_INIT_CSUM },
8446 { "init-extent-tree", 0, NULL, OPT_INIT_EXTENT },
8447 { "check-data-csum", 0, NULL, OPT_CHECK_CSUM },
8448 { "backup", 0, NULL, 'b' },
8449 { "subvol-extents", 1, NULL, 'E' },
8450 { "qgroup-report", 0, NULL, 'Q' },
8451 { "tree-root", 1, NULL, 'r' },
8455 c = getopt_long(argc, argv, "as:br:", long_options,
8460 case 'a': /* ignored */ break;
8462 ctree_flags |= OPEN_CTREE_BACKUP_ROOT;
8465 num = arg_strtou64(optarg);
8466 if (num >= BTRFS_SUPER_MIRROR_MAX) {
8468 "ERROR: super mirror should be less than: %d\n",
8469 BTRFS_SUPER_MIRROR_MAX);
8472 bytenr = btrfs_sb_offset(((int)num));
8473 printf("using SB copy %llu, bytenr %llu\n", num,
8474 (unsigned long long)bytenr);
8480 subvolid = arg_strtou64(optarg);
8483 tree_root_bytenr = arg_strtou64(optarg);
8487 usage(cmd_check_usage);
8489 printf("enabling repair mode\n");
8491 ctree_flags |= OPEN_CTREE_WRITES;
8497 printf("Creating a new CRC tree\n");
8500 ctree_flags |= OPEN_CTREE_WRITES;
8502 case OPT_INIT_EXTENT:
8503 init_extent_tree = 1;
8504 ctree_flags |= (OPEN_CTREE_WRITES |
8505 OPEN_CTREE_NO_BLOCK_GROUPS);
8508 case OPT_CHECK_CSUM:
8509 check_data_csum = 1;
8513 argc = argc - optind;
8515 if (check_argc_exact(argc, 1))
8516 usage(cmd_check_usage);
8518 /* This check is the only reason for --readonly to exist */
8519 if (readonly && repair) {
8520 fprintf(stderr, "Repair options are not compatible with --readonly\n");
8525 cache_tree_init(&root_cache);
8527 if((ret = check_mounted(argv[optind])) < 0) {
8528 fprintf(stderr, "Could not check mount status: %s\n", strerror(-ret));
8531 fprintf(stderr, "%s is currently mounted. Aborting.\n", argv[optind]);
8536 /* only allow partial opening under repair mode */
8538 ctree_flags |= OPEN_CTREE_PARTIAL;
8540 info = open_ctree_fs_info(argv[optind], bytenr, tree_root_bytenr,
8543 fprintf(stderr, "Couldn't open file system\n");
8548 root = info->fs_root;
8551 * repair mode will force us to commit transaction which
8552 * will make us fail to load log tree when mounting.
8554 if (repair && btrfs_super_log_root(info->super_copy)) {
8555 ret = ask_user("repair mode will force to clear out log tree, Are you sure?");
8560 ret = zero_log_tree(root);
8562 fprintf(stderr, "fail to zero log tree\n");
8567 uuid_unparse(info->super_copy->fsid, uuidbuf);
8568 if (qgroup_report) {
8569 printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
8571 ret = qgroup_verify_all(info);
8573 print_qgroup_report(1);
8577 printf("Print extent state for subvolume %llu on %s\nUUID: %s\n",
8578 subvolid, argv[optind], uuidbuf);
8579 ret = print_extent_state(info, subvolid);
8582 printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
8584 if (!extent_buffer_uptodate(info->tree_root->node) ||
8585 !extent_buffer_uptodate(info->dev_root->node) ||
8586 !extent_buffer_uptodate(info->chunk_root->node)) {
8587 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
8592 if (init_extent_tree || init_csum_tree) {
8593 struct btrfs_trans_handle *trans;
8595 trans = btrfs_start_transaction(info->extent_root, 0);
8596 if (IS_ERR(trans)) {
8597 fprintf(stderr, "Error starting transaction\n");
8598 ret = PTR_ERR(trans);
8602 if (init_extent_tree) {
8603 printf("Creating a new extent tree\n");
8604 ret = reinit_extent_tree(trans, info);
8609 if (init_csum_tree) {
8610 fprintf(stderr, "Reinit crc root\n");
8611 ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
8613 fprintf(stderr, "crc root initialization failed\n");
8618 ret = fill_csum_tree(trans, info->csum_root);
8620 fprintf(stderr, "crc refilling failed\n");
8625 * Ok now we commit and run the normal fsck, which will add
8626 * extent entries for all of the items it finds.
8628 ret = btrfs_commit_transaction(trans, info->extent_root);
8632 if (!extent_buffer_uptodate(info->extent_root->node)) {
8633 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
8637 if (!extent_buffer_uptodate(info->csum_root->node)) {
8638 fprintf(stderr, "Checksum root corrupted, rerun with --init-csum-tree option\n");
8643 fprintf(stderr, "checking extents\n");
8644 ret = check_chunks_and_extents(root);
8646 fprintf(stderr, "Errors found in extent allocation tree or chunk allocation\n");
8648 ret = repair_root_items(info);
8652 fprintf(stderr, "Fixed %d roots.\n", ret);
8654 } else if (ret > 0) {
8656 "Found %d roots with an outdated root item.\n",
8659 "Please run a filesystem check with the option --repair to fix them.\n");
8664 fprintf(stderr, "checking free space cache\n");
8665 ret = check_space_cache(root);
8670 * We used to have to have these hole extents in between our real
8671 * extents so if we don't have this flag set we need to make sure there
8672 * are no gaps in the file extents for inodes, otherwise we can just
8673 * ignore it when this happens.
8675 no_holes = btrfs_fs_incompat(root->fs_info,
8676 BTRFS_FEATURE_INCOMPAT_NO_HOLES);
8677 fprintf(stderr, "checking fs roots\n");
8678 ret = check_fs_roots(root, &root_cache);
8682 fprintf(stderr, "checking csums\n");
8683 ret = check_csums(root);
8687 fprintf(stderr, "checking root refs\n");
8688 ret = check_root_refs(root, &root_cache);
8692 while (repair && !list_empty(&root->fs_info->recow_ebs)) {
8693 struct extent_buffer *eb;
8695 eb = list_first_entry(&root->fs_info->recow_ebs,
8696 struct extent_buffer, recow);
8697 list_del_init(&eb->recow);
8698 ret = recow_extent_buffer(root, eb);
8703 while (!list_empty(&delete_items)) {
8704 struct bad_item *bad;
8706 bad = list_first_entry(&delete_items, struct bad_item, list);
8707 list_del_init(&bad->list);
8709 ret = delete_bad_item(root, bad);
8713 if (info->quota_enabled) {
8715 fprintf(stderr, "checking quota groups\n");
8716 err = qgroup_verify_all(info);
8721 if (!list_empty(&root->fs_info->recow_ebs)) {
8722 fprintf(stderr, "Transid errors in file system\n");
8726 print_qgroup_report(0);
8727 if (found_old_backref) { /*
8728 * there was a disk format change when mixed
8729 * backref was in testing tree. The old format
8730 * existed about one week.
8732 printf("\n * Found old mixed backref format. "
8733 "The old format is not supported! *"
8734 "\n * Please mount the FS in readonly mode, "
8735 "backup data and re-format the FS. *\n\n");
8738 printf("found %llu bytes used err is %d\n",
8739 (unsigned long long)bytes_used, ret);
8740 printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
8741 printf("total tree bytes: %llu\n",
8742 (unsigned long long)total_btree_bytes);
8743 printf("total fs tree bytes: %llu\n",
8744 (unsigned long long)total_fs_tree_bytes);
8745 printf("total extent tree bytes: %llu\n",
8746 (unsigned long long)total_extent_tree_bytes);
8747 printf("btree space waste bytes: %llu\n",
8748 (unsigned long long)btree_space_waste);
8749 printf("file data blocks allocated: %llu\n referenced %llu\n",
8750 (unsigned long long)data_bytes_allocated,
8751 (unsigned long long)data_bytes_referenced);
8752 printf("%s\n", BTRFS_BUILD_VERSION);
8754 free_root_recs_tree(&root_cache);