2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #define _XOPEN_SOURCE 500
25 #include <sys/types.h>
29 #include <uuid/uuid.h>
34 #include "print-tree.h"
35 #include "transaction.h"
39 #include "free-space-cache.h"
42 static u64 bytes_used = 0;
43 static u64 total_csum_bytes = 0;
44 static u64 total_btree_bytes = 0;
45 static u64 total_fs_tree_bytes = 0;
46 static u64 total_extent_tree_bytes = 0;
47 static u64 btree_space_waste = 0;
48 static u64 data_bytes_allocated = 0;
49 static u64 data_bytes_referenced = 0;
50 static int found_old_backref = 0;
51 static LIST_HEAD(duplicate_extents);
53 struct extent_backref {
54 struct list_head list;
55 unsigned int is_data:1;
56 unsigned int found_extent_tree:1;
57 unsigned int full_backref:1;
58 unsigned int found_ref:1;
62 struct extent_backref node;
77 struct extent_backref node;
84 struct extent_record {
85 struct list_head backrefs;
86 struct list_head dups;
87 struct list_head list;
88 struct cache_extent cache;
89 struct btrfs_disk_key parent_key;
90 unsigned int found_rec;
100 unsigned int content_checked:1;
101 unsigned int owner_ref_checked:1;
102 unsigned int is_root:1;
103 unsigned int metadata:1;
106 struct inode_backref {
107 struct list_head list;
108 unsigned int found_dir_item:1;
109 unsigned int found_dir_index:1;
110 unsigned int found_inode_ref:1;
111 unsigned int filetype:8;
113 unsigned int ref_type;
120 #define REF_ERR_NO_DIR_ITEM (1 << 0)
121 #define REF_ERR_NO_DIR_INDEX (1 << 1)
122 #define REF_ERR_NO_INODE_REF (1 << 2)
123 #define REF_ERR_DUP_DIR_ITEM (1 << 3)
124 #define REF_ERR_DUP_DIR_INDEX (1 << 4)
125 #define REF_ERR_DUP_INODE_REF (1 << 5)
126 #define REF_ERR_INDEX_UNMATCH (1 << 6)
127 #define REF_ERR_FILETYPE_UNMATCH (1 << 7)
128 #define REF_ERR_NAME_TOO_LONG (1 << 8) // 100
129 #define REF_ERR_NO_ROOT_REF (1 << 9)
130 #define REF_ERR_NO_ROOT_BACKREF (1 << 10)
131 #define REF_ERR_DUP_ROOT_REF (1 << 11)
132 #define REF_ERR_DUP_ROOT_BACKREF (1 << 12)
134 struct inode_record {
135 struct list_head backrefs;
136 unsigned int checked:1;
137 unsigned int merging:1;
138 unsigned int found_inode_item:1;
139 unsigned int found_dir_item:1;
140 unsigned int found_file_extent:1;
141 unsigned int found_csum_item:1;
142 unsigned int some_csum_missing:1;
143 unsigned int nodatasum:1;
156 u64 first_extent_gap;
161 #define I_ERR_NO_INODE_ITEM (1 << 0)
162 #define I_ERR_NO_ORPHAN_ITEM (1 << 1)
163 #define I_ERR_DUP_INODE_ITEM (1 << 2)
164 #define I_ERR_DUP_DIR_INDEX (1 << 3)
165 #define I_ERR_ODD_DIR_ITEM (1 << 4)
166 #define I_ERR_ODD_FILE_EXTENT (1 << 5)
167 #define I_ERR_BAD_FILE_EXTENT (1 << 6)
168 #define I_ERR_FILE_EXTENT_OVERLAP (1 << 7)
169 #define I_ERR_FILE_EXTENT_DISCOUNT (1 << 8) // 100
170 #define I_ERR_DIR_ISIZE_WRONG (1 << 9)
171 #define I_ERR_FILE_NBYTES_WRONG (1 << 10) // 400
172 #define I_ERR_ODD_CSUM_ITEM (1 << 11)
173 #define I_ERR_SOME_CSUM_MISSING (1 << 12)
174 #define I_ERR_LINK_COUNT_WRONG (1 << 13)
176 struct root_backref {
177 struct list_head list;
178 unsigned int found_dir_item:1;
179 unsigned int found_dir_index:1;
180 unsigned int found_back_ref:1;
181 unsigned int found_forward_ref:1;
182 unsigned int reachable:1;
192 struct list_head backrefs;
193 struct cache_extent cache;
194 unsigned int found_root_item:1;
200 struct cache_extent cache;
205 struct cache_extent cache;
206 struct cache_tree root_cache;
207 struct cache_tree inode_cache;
208 struct inode_record *current;
217 struct walk_control {
218 struct cache_tree shared;
219 struct shared_node *nodes[BTRFS_MAX_LEVEL];
224 static u8 imode_to_type(u32 imode)
227 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
228 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
229 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
230 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
231 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
232 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
233 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
234 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
237 return btrfs_type_by_mode[(imode & S_IFMT) >> S_SHIFT];
241 static int device_record_compare(struct rb_node *node1, struct rb_node *node2)
243 struct device_record *rec1;
244 struct device_record *rec2;
246 rec1 = rb_entry(node1, struct device_record, node);
247 rec2 = rb_entry(node2, struct device_record, node);
248 if (rec1->devid > rec2->devid)
250 else if (rec1->devid < rec2->devid)
256 static struct inode_record *clone_inode_rec(struct inode_record *orig_rec)
258 struct inode_record *rec;
259 struct inode_backref *backref;
260 struct inode_backref *orig;
263 rec = malloc(sizeof(*rec));
264 memcpy(rec, orig_rec, sizeof(*rec));
266 INIT_LIST_HEAD(&rec->backrefs);
268 list_for_each_entry(orig, &orig_rec->backrefs, list) {
269 size = sizeof(*orig) + orig->namelen + 1;
270 backref = malloc(size);
271 memcpy(backref, orig, size);
272 list_add_tail(&backref->list, &rec->backrefs);
277 static struct inode_record *get_inode_rec(struct cache_tree *inode_cache,
280 struct ptr_node *node;
281 struct cache_extent *cache;
282 struct inode_record *rec = NULL;
285 cache = lookup_cache_extent(inode_cache, ino, 1);
287 node = container_of(cache, struct ptr_node, cache);
289 if (mod && rec->refs > 1) {
290 node->data = clone_inode_rec(rec);
295 rec = calloc(1, sizeof(*rec));
297 rec->extent_start = (u64)-1;
298 rec->first_extent_gap = (u64)-1;
300 INIT_LIST_HEAD(&rec->backrefs);
302 node = malloc(sizeof(*node));
303 node->cache.start = ino;
304 node->cache.size = 1;
307 if (ino == BTRFS_FREE_INO_OBJECTID)
310 ret = insert_cache_extent(inode_cache, &node->cache);
316 static void free_inode_rec(struct inode_record *rec)
318 struct inode_backref *backref;
323 while (!list_empty(&rec->backrefs)) {
324 backref = list_entry(rec->backrefs.next,
325 struct inode_backref, list);
326 list_del(&backref->list);
332 static int can_free_inode_rec(struct inode_record *rec)
334 if (!rec->errors && rec->checked && rec->found_inode_item &&
335 rec->nlink == rec->found_link && list_empty(&rec->backrefs))
340 static void maybe_free_inode_rec(struct cache_tree *inode_cache,
341 struct inode_record *rec)
343 struct cache_extent *cache;
344 struct inode_backref *tmp, *backref;
345 struct ptr_node *node;
346 unsigned char filetype;
348 if (!rec->found_inode_item)
351 filetype = imode_to_type(rec->imode);
352 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
353 if (backref->found_dir_item && backref->found_dir_index) {
354 if (backref->filetype != filetype)
355 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
356 if (!backref->errors && backref->found_inode_ref) {
357 list_del(&backref->list);
363 if (!rec->checked || rec->merging)
366 if (S_ISDIR(rec->imode)) {
367 if (rec->found_size != rec->isize)
368 rec->errors |= I_ERR_DIR_ISIZE_WRONG;
369 if (rec->found_file_extent)
370 rec->errors |= I_ERR_ODD_FILE_EXTENT;
371 } else if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
372 if (rec->found_dir_item)
373 rec->errors |= I_ERR_ODD_DIR_ITEM;
374 if (rec->found_size != rec->nbytes)
375 rec->errors |= I_ERR_FILE_NBYTES_WRONG;
376 if (rec->extent_start == (u64)-1 || rec->extent_start > 0)
377 rec->first_extent_gap = 0;
378 if (rec->nlink > 0 && (rec->extent_end < rec->isize ||
379 rec->first_extent_gap < rec->isize))
380 rec->errors |= I_ERR_FILE_EXTENT_DISCOUNT;
383 if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
384 if (rec->found_csum_item && rec->nodatasum)
385 rec->errors |= I_ERR_ODD_CSUM_ITEM;
386 if (rec->some_csum_missing && !rec->nodatasum)
387 rec->errors |= I_ERR_SOME_CSUM_MISSING;
390 BUG_ON(rec->refs != 1);
391 if (can_free_inode_rec(rec)) {
392 cache = lookup_cache_extent(inode_cache, rec->ino, 1);
393 node = container_of(cache, struct ptr_node, cache);
394 BUG_ON(node->data != rec);
395 remove_cache_extent(inode_cache, &node->cache);
401 static int check_orphan_item(struct btrfs_root *root, u64 ino)
403 struct btrfs_path path;
404 struct btrfs_key key;
407 key.objectid = BTRFS_ORPHAN_OBJECTID;
408 key.type = BTRFS_ORPHAN_ITEM_KEY;
411 btrfs_init_path(&path);
412 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
413 btrfs_release_path(&path);
419 static int process_inode_item(struct extent_buffer *eb,
420 int slot, struct btrfs_key *key,
421 struct shared_node *active_node)
423 struct inode_record *rec;
424 struct btrfs_inode_item *item;
426 rec = active_node->current;
427 BUG_ON(rec->ino != key->objectid || rec->refs > 1);
428 if (rec->found_inode_item) {
429 rec->errors |= I_ERR_DUP_INODE_ITEM;
432 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
433 rec->nlink = btrfs_inode_nlink(eb, item);
434 rec->isize = btrfs_inode_size(eb, item);
435 rec->nbytes = btrfs_inode_nbytes(eb, item);
436 rec->imode = btrfs_inode_mode(eb, item);
437 if (btrfs_inode_flags(eb, item) & BTRFS_INODE_NODATASUM)
439 rec->found_inode_item = 1;
441 rec->errors |= I_ERR_NO_ORPHAN_ITEM;
442 maybe_free_inode_rec(&active_node->inode_cache, rec);
446 static struct inode_backref *get_inode_backref(struct inode_record *rec,
448 int namelen, u64 dir)
450 struct inode_backref *backref;
452 list_for_each_entry(backref, &rec->backrefs, list) {
453 if (backref->dir != dir || backref->namelen != namelen)
455 if (memcmp(name, backref->name, namelen))
460 backref = malloc(sizeof(*backref) + namelen + 1);
461 memset(backref, 0, sizeof(*backref));
463 backref->namelen = namelen;
464 memcpy(backref->name, name, namelen);
465 backref->name[namelen] = '\0';
466 list_add_tail(&backref->list, &rec->backrefs);
470 static int add_inode_backref(struct cache_tree *inode_cache,
471 u64 ino, u64 dir, u64 index,
472 const char *name, int namelen,
473 int filetype, int itemtype, int errors)
475 struct inode_record *rec;
476 struct inode_backref *backref;
478 rec = get_inode_rec(inode_cache, ino, 1);
479 backref = get_inode_backref(rec, name, namelen, dir);
481 backref->errors |= errors;
482 if (itemtype == BTRFS_DIR_INDEX_KEY) {
483 if (backref->found_dir_index)
484 backref->errors |= REF_ERR_DUP_DIR_INDEX;
485 if (backref->found_inode_ref && backref->index != index)
486 backref->errors |= REF_ERR_INDEX_UNMATCH;
487 if (backref->found_dir_item && backref->filetype != filetype)
488 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
490 backref->index = index;
491 backref->filetype = filetype;
492 backref->found_dir_index = 1;
493 } else if (itemtype == BTRFS_DIR_ITEM_KEY) {
495 if (backref->found_dir_item)
496 backref->errors |= REF_ERR_DUP_DIR_ITEM;
497 if (backref->found_dir_index && backref->filetype != filetype)
498 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
500 backref->filetype = filetype;
501 backref->found_dir_item = 1;
502 } else if ((itemtype == BTRFS_INODE_REF_KEY) ||
503 (itemtype == BTRFS_INODE_EXTREF_KEY)) {
504 if (backref->found_inode_ref)
505 backref->errors |= REF_ERR_DUP_INODE_REF;
506 if (backref->found_dir_index && backref->index != index)
507 backref->errors |= REF_ERR_INDEX_UNMATCH;
509 backref->ref_type = itemtype;
510 backref->index = index;
511 backref->found_inode_ref = 1;
516 maybe_free_inode_rec(inode_cache, rec);
520 static int merge_inode_recs(struct inode_record *src, struct inode_record *dst,
521 struct cache_tree *dst_cache)
523 struct inode_backref *backref;
527 list_for_each_entry(backref, &src->backrefs, list) {
528 if (backref->found_dir_index) {
529 add_inode_backref(dst_cache, dst->ino, backref->dir,
530 backref->index, backref->name,
531 backref->namelen, backref->filetype,
532 BTRFS_DIR_INDEX_KEY, backref->errors);
534 if (backref->found_dir_item) {
536 add_inode_backref(dst_cache, dst->ino,
537 backref->dir, 0, backref->name,
538 backref->namelen, backref->filetype,
539 BTRFS_DIR_ITEM_KEY, backref->errors);
541 if (backref->found_inode_ref) {
542 add_inode_backref(dst_cache, dst->ino,
543 backref->dir, backref->index,
544 backref->name, backref->namelen, 0,
545 backref->ref_type, backref->errors);
549 if (src->found_dir_item)
550 dst->found_dir_item = 1;
551 if (src->found_file_extent)
552 dst->found_file_extent = 1;
553 if (src->found_csum_item)
554 dst->found_csum_item = 1;
555 if (src->some_csum_missing)
556 dst->some_csum_missing = 1;
557 if (dst->first_extent_gap > src->first_extent_gap)
558 dst->first_extent_gap = src->first_extent_gap;
560 BUG_ON(src->found_link < dir_count);
561 dst->found_link += src->found_link - dir_count;
562 dst->found_size += src->found_size;
563 if (src->extent_start != (u64)-1) {
564 if (dst->extent_start == (u64)-1) {
565 dst->extent_start = src->extent_start;
566 dst->extent_end = src->extent_end;
568 if (dst->extent_end > src->extent_start)
569 dst->errors |= I_ERR_FILE_EXTENT_OVERLAP;
570 else if (dst->extent_end < src->extent_start &&
571 dst->extent_end < dst->first_extent_gap)
572 dst->first_extent_gap = dst->extent_end;
573 if (dst->extent_end < src->extent_end)
574 dst->extent_end = src->extent_end;
578 dst->errors |= src->errors;
579 if (src->found_inode_item) {
580 if (!dst->found_inode_item) {
581 dst->nlink = src->nlink;
582 dst->isize = src->isize;
583 dst->nbytes = src->nbytes;
584 dst->imode = src->imode;
585 dst->nodatasum = src->nodatasum;
586 dst->found_inode_item = 1;
588 dst->errors |= I_ERR_DUP_INODE_ITEM;
596 static int splice_shared_node(struct shared_node *src_node,
597 struct shared_node *dst_node)
599 struct cache_extent *cache;
600 struct ptr_node *node, *ins;
601 struct cache_tree *src, *dst;
602 struct inode_record *rec, *conflict;
607 if (--src_node->refs == 0)
609 if (src_node->current)
610 current_ino = src_node->current->ino;
612 src = &src_node->root_cache;
613 dst = &dst_node->root_cache;
615 cache = search_cache_extent(src, 0);
617 node = container_of(cache, struct ptr_node, cache);
619 cache = next_cache_extent(cache);
622 remove_cache_extent(src, &node->cache);
625 ins = malloc(sizeof(*ins));
626 ins->cache.start = node->cache.start;
627 ins->cache.size = node->cache.size;
631 ret = insert_cache_extent(dst, &ins->cache);
632 if (ret == -EEXIST) {
633 conflict = get_inode_rec(dst, rec->ino, 1);
634 merge_inode_recs(rec, conflict, dst);
636 conflict->checked = 1;
637 if (dst_node->current == conflict)
638 dst_node->current = NULL;
640 maybe_free_inode_rec(dst, conflict);
648 if (src == &src_node->root_cache) {
649 src = &src_node->inode_cache;
650 dst = &dst_node->inode_cache;
654 if (current_ino > 0 && (!dst_node->current ||
655 current_ino > dst_node->current->ino)) {
656 if (dst_node->current) {
657 dst_node->current->checked = 1;
658 maybe_free_inode_rec(dst, dst_node->current);
660 dst_node->current = get_inode_rec(dst, current_ino, 1);
665 static void free_inode_ptr(struct cache_extent *cache)
667 struct ptr_node *node;
668 struct inode_record *rec;
670 node = container_of(cache, struct ptr_node, cache);
676 FREE_EXTENT_CACHE_BASED_TREE(inode_recs, free_inode_ptr);
678 static struct shared_node *find_shared_node(struct cache_tree *shared,
681 struct cache_extent *cache;
682 struct shared_node *node;
684 cache = lookup_cache_extent(shared, bytenr, 1);
686 node = container_of(cache, struct shared_node, cache);
692 static int add_shared_node(struct cache_tree *shared, u64 bytenr, u32 refs)
695 struct shared_node *node;
697 node = calloc(1, sizeof(*node));
698 node->cache.start = bytenr;
699 node->cache.size = 1;
700 cache_tree_init(&node->root_cache);
701 cache_tree_init(&node->inode_cache);
704 ret = insert_cache_extent(shared, &node->cache);
709 static int enter_shared_node(struct btrfs_root *root, u64 bytenr, u32 refs,
710 struct walk_control *wc, int level)
712 struct shared_node *node;
713 struct shared_node *dest;
715 if (level == wc->active_node)
718 BUG_ON(wc->active_node <= level);
719 node = find_shared_node(&wc->shared, bytenr);
721 add_shared_node(&wc->shared, bytenr, refs);
722 node = find_shared_node(&wc->shared, bytenr);
723 wc->nodes[level] = node;
724 wc->active_node = level;
728 if (wc->root_level == wc->active_node &&
729 btrfs_root_refs(&root->root_item) == 0) {
730 if (--node->refs == 0) {
731 free_inode_recs_tree(&node->root_cache);
732 free_inode_recs_tree(&node->inode_cache);
733 remove_cache_extent(&wc->shared, &node->cache);
739 dest = wc->nodes[wc->active_node];
740 splice_shared_node(node, dest);
741 if (node->refs == 0) {
742 remove_cache_extent(&wc->shared, &node->cache);
748 static int leave_shared_node(struct btrfs_root *root,
749 struct walk_control *wc, int level)
751 struct shared_node *node;
752 struct shared_node *dest;
755 if (level == wc->root_level)
758 for (i = level + 1; i < BTRFS_MAX_LEVEL; i++) {
762 BUG_ON(i >= BTRFS_MAX_LEVEL);
764 node = wc->nodes[wc->active_node];
765 wc->nodes[wc->active_node] = NULL;
768 dest = wc->nodes[wc->active_node];
769 if (wc->active_node < wc->root_level ||
770 btrfs_root_refs(&root->root_item) > 0) {
771 BUG_ON(node->refs <= 1);
772 splice_shared_node(node, dest);
774 BUG_ON(node->refs < 2);
780 static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
783 struct btrfs_path path;
784 struct btrfs_key key;
785 struct extent_buffer *leaf;
789 btrfs_init_path(&path);
791 key.objectid = parent_root_id;
792 key.type = BTRFS_ROOT_REF_KEY;
793 key.offset = child_root_id;
794 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
797 btrfs_release_path(&path);
801 key.objectid = child_root_id;
802 key.type = BTRFS_ROOT_BACKREF_KEY;
804 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
809 leaf = path.nodes[0];
810 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
811 ret = btrfs_next_leaf(root->fs_info->tree_root, &path);
816 leaf = path.nodes[0];
819 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
820 if (key.objectid != child_root_id ||
821 key.type != BTRFS_ROOT_BACKREF_KEY)
826 if (key.offset == parent_root_id) {
827 btrfs_release_path(&path);
834 btrfs_release_path(&path);
835 return has_parent? 0 : -1;
838 static int process_dir_item(struct btrfs_root *root,
839 struct extent_buffer *eb,
840 int slot, struct btrfs_key *key,
841 struct shared_node *active_node)
851 struct btrfs_dir_item *di;
852 struct inode_record *rec;
853 struct cache_tree *root_cache;
854 struct cache_tree *inode_cache;
855 struct btrfs_key location;
856 char namebuf[BTRFS_NAME_LEN];
858 root_cache = &active_node->root_cache;
859 inode_cache = &active_node->inode_cache;
860 rec = active_node->current;
861 rec->found_dir_item = 1;
863 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
864 total = btrfs_item_size_nr(eb, slot);
865 while (cur < total) {
867 btrfs_dir_item_key_to_cpu(eb, di, &location);
868 name_len = btrfs_dir_name_len(eb, di);
869 data_len = btrfs_dir_data_len(eb, di);
870 filetype = btrfs_dir_type(eb, di);
872 rec->found_size += name_len;
873 if (name_len <= BTRFS_NAME_LEN) {
877 len = BTRFS_NAME_LEN;
878 error = REF_ERR_NAME_TOO_LONG;
880 read_extent_buffer(eb, namebuf, (unsigned long)(di + 1), len);
882 if (location.type == BTRFS_INODE_ITEM_KEY) {
883 add_inode_backref(inode_cache, location.objectid,
884 key->objectid, key->offset, namebuf,
885 len, filetype, key->type, error);
886 } else if (location.type == BTRFS_ROOT_ITEM_KEY) {
887 add_inode_backref(root_cache, location.objectid,
888 key->objectid, key->offset,
889 namebuf, len, filetype,
892 fprintf(stderr, "warning line %d\n", __LINE__);
895 len = sizeof(*di) + name_len + data_len;
896 di = (struct btrfs_dir_item *)((char *)di + len);
899 if (key->type == BTRFS_DIR_INDEX_KEY && nritems > 1)
900 rec->errors |= I_ERR_DUP_DIR_INDEX;
905 static int process_inode_ref(struct extent_buffer *eb,
906 int slot, struct btrfs_key *key,
907 struct shared_node *active_node)
915 struct cache_tree *inode_cache;
916 struct btrfs_inode_ref *ref;
917 char namebuf[BTRFS_NAME_LEN];
919 inode_cache = &active_node->inode_cache;
921 ref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
922 total = btrfs_item_size_nr(eb, slot);
923 while (cur < total) {
924 name_len = btrfs_inode_ref_name_len(eb, ref);
925 index = btrfs_inode_ref_index(eb, ref);
926 if (name_len <= BTRFS_NAME_LEN) {
930 len = BTRFS_NAME_LEN;
931 error = REF_ERR_NAME_TOO_LONG;
933 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
934 add_inode_backref(inode_cache, key->objectid, key->offset,
935 index, namebuf, len, 0, key->type, error);
937 len = sizeof(*ref) + name_len;
938 ref = (struct btrfs_inode_ref *)((char *)ref + len);
944 static int process_inode_extref(struct extent_buffer *eb,
945 int slot, struct btrfs_key *key,
946 struct shared_node *active_node)
955 struct cache_tree *inode_cache;
956 struct btrfs_inode_extref *extref;
957 char namebuf[BTRFS_NAME_LEN];
959 inode_cache = &active_node->inode_cache;
961 extref = btrfs_item_ptr(eb, slot, struct btrfs_inode_extref);
962 total = btrfs_item_size_nr(eb, slot);
963 while (cur < total) {
964 name_len = btrfs_inode_extref_name_len(eb, extref);
965 index = btrfs_inode_extref_index(eb, extref);
966 parent = btrfs_inode_extref_parent(eb, extref);
967 if (name_len <= BTRFS_NAME_LEN) {
971 len = BTRFS_NAME_LEN;
972 error = REF_ERR_NAME_TOO_LONG;
974 read_extent_buffer(eb, namebuf,
975 (unsigned long)(extref + 1), len);
976 add_inode_backref(inode_cache, key->objectid, parent,
977 index, namebuf, len, 0, key->type, error);
979 len = sizeof(*extref) + name_len;
980 extref = (struct btrfs_inode_extref *)((char *)extref + len);
987 static u64 count_csum_range(struct btrfs_root *root, u64 start, u64 len)
989 struct btrfs_key key;
990 struct btrfs_path path;
991 struct extent_buffer *leaf;
996 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
998 btrfs_init_path(&path);
1000 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1002 key.type = BTRFS_EXTENT_CSUM_KEY;
1004 ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
1007 if (ret > 0 && path.slots[0] > 0) {
1008 leaf = path.nodes[0];
1009 btrfs_item_key_to_cpu(leaf, &key, path.slots[0] - 1);
1010 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
1011 key.type == BTRFS_EXTENT_CSUM_KEY)
1016 leaf = path.nodes[0];
1017 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1018 ret = btrfs_next_leaf(root->fs_info->csum_root, &path);
1022 leaf = path.nodes[0];
1025 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1026 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1027 key.type != BTRFS_EXTENT_CSUM_KEY)
1030 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1031 if (key.offset >= start + len)
1034 if (key.offset > start)
1037 size = btrfs_item_size_nr(leaf, path.slots[0]);
1038 csum_end = key.offset + (size / csum_size) * root->sectorsize;
1039 if (csum_end > start) {
1040 size = min(csum_end - start, len);
1048 btrfs_release_path(&path);
1052 static int process_file_extent(struct btrfs_root *root,
1053 struct extent_buffer *eb,
1054 int slot, struct btrfs_key *key,
1055 struct shared_node *active_node)
1057 struct inode_record *rec;
1058 struct btrfs_file_extent_item *fi;
1060 u64 disk_bytenr = 0;
1061 u64 extent_offset = 0;
1062 u64 mask = root->sectorsize - 1;
1065 rec = active_node->current;
1066 BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1067 rec->found_file_extent = 1;
1069 if (rec->extent_start == (u64)-1) {
1070 rec->extent_start = key->offset;
1071 rec->extent_end = key->offset;
1074 if (rec->extent_end > key->offset)
1075 rec->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1076 else if (rec->extent_end < key->offset &&
1077 rec->extent_end < rec->first_extent_gap)
1078 rec->first_extent_gap = rec->extent_end;
1080 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
1081 extent_type = btrfs_file_extent_type(eb, fi);
1083 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1084 num_bytes = btrfs_file_extent_inline_len(eb, fi);
1086 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1087 rec->found_size += num_bytes;
1088 num_bytes = (num_bytes + mask) & ~mask;
1089 } else if (extent_type == BTRFS_FILE_EXTENT_REG ||
1090 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1091 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1092 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1093 extent_offset = btrfs_file_extent_offset(eb, fi);
1094 if (num_bytes == 0 || (num_bytes & mask))
1095 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1096 if (num_bytes + extent_offset >
1097 btrfs_file_extent_ram_bytes(eb, fi))
1098 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1099 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC &&
1100 (btrfs_file_extent_compression(eb, fi) ||
1101 btrfs_file_extent_encryption(eb, fi) ||
1102 btrfs_file_extent_other_encoding(eb, fi)))
1103 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1104 if (disk_bytenr > 0)
1105 rec->found_size += num_bytes;
1107 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1109 rec->extent_end = key->offset + num_bytes;
1111 if (disk_bytenr > 0) {
1113 if (btrfs_file_extent_compression(eb, fi))
1114 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1116 disk_bytenr += extent_offset;
1118 found = count_csum_range(root, disk_bytenr, num_bytes);
1119 if (extent_type == BTRFS_FILE_EXTENT_REG) {
1121 rec->found_csum_item = 1;
1122 if (found < num_bytes)
1123 rec->some_csum_missing = 1;
1124 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1126 rec->errors |= I_ERR_ODD_CSUM_ITEM;
1132 static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
1133 struct walk_control *wc)
1135 struct btrfs_key key;
1139 struct cache_tree *inode_cache;
1140 struct shared_node *active_node;
1142 if (wc->root_level == wc->active_node &&
1143 btrfs_root_refs(&root->root_item) == 0)
1146 active_node = wc->nodes[wc->active_node];
1147 inode_cache = &active_node->inode_cache;
1148 nritems = btrfs_header_nritems(eb);
1149 for (i = 0; i < nritems; i++) {
1150 btrfs_item_key_to_cpu(eb, &key, i);
1152 if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
1155 if (active_node->current == NULL ||
1156 active_node->current->ino < key.objectid) {
1157 if (active_node->current) {
1158 active_node->current->checked = 1;
1159 maybe_free_inode_rec(inode_cache,
1160 active_node->current);
1162 active_node->current = get_inode_rec(inode_cache,
1166 case BTRFS_DIR_ITEM_KEY:
1167 case BTRFS_DIR_INDEX_KEY:
1168 ret = process_dir_item(root, eb, i, &key, active_node);
1170 case BTRFS_INODE_REF_KEY:
1171 ret = process_inode_ref(eb, i, &key, active_node);
1173 case BTRFS_INODE_EXTREF_KEY:
1174 ret = process_inode_extref(eb, i, &key, active_node);
1176 case BTRFS_INODE_ITEM_KEY:
1177 ret = process_inode_item(eb, i, &key, active_node);
1179 case BTRFS_EXTENT_DATA_KEY:
1180 ret = process_file_extent(root, eb, i, &key,
1190 static void reada_walk_down(struct btrfs_root *root,
1191 struct extent_buffer *node, int slot)
1201 level = btrfs_header_level(node);
1205 nritems = btrfs_header_nritems(node);
1206 blocksize = btrfs_level_size(root, level - 1);
1207 for (i = slot; i < nritems; i++) {
1208 bytenr = btrfs_node_blockptr(node, i);
1209 ptr_gen = btrfs_node_ptr_generation(node, i);
1210 ret = readahead_tree_block(root, bytenr, blocksize, ptr_gen);
1216 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
1217 struct walk_control *wc, int *level)
1221 struct extent_buffer *next;
1222 struct extent_buffer *cur;
1227 WARN_ON(*level < 0);
1228 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1229 ret = btrfs_lookup_extent_info(NULL, root,
1230 path->nodes[*level]->start,
1231 *level, 1, &refs, NULL);
1238 ret = enter_shared_node(root, path->nodes[*level]->start,
1246 while (*level >= 0) {
1247 WARN_ON(*level < 0);
1248 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1249 cur = path->nodes[*level];
1251 if (btrfs_header_level(cur) != *level)
1254 if (path->slots[*level] >= btrfs_header_nritems(cur))
1257 ret = process_one_leaf(root, cur, wc);
1260 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1261 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1262 blocksize = btrfs_level_size(root, *level - 1);
1263 ret = btrfs_lookup_extent_info(NULL, root, bytenr, *level - 1,
1269 ret = enter_shared_node(root, bytenr, refs,
1272 path->slots[*level]++;
1277 next = btrfs_find_tree_block(root, bytenr, blocksize);
1278 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
1279 free_extent_buffer(next);
1280 reada_walk_down(root, cur, path->slots[*level]);
1281 next = read_tree_block(root, bytenr, blocksize,
1289 *level = *level - 1;
1290 free_extent_buffer(path->nodes[*level]);
1291 path->nodes[*level] = next;
1292 path->slots[*level] = 0;
1295 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
1299 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
1300 struct walk_control *wc, int *level)
1303 struct extent_buffer *leaf;
1305 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1306 leaf = path->nodes[i];
1307 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
1312 free_extent_buffer(path->nodes[*level]);
1313 path->nodes[*level] = NULL;
1314 BUG_ON(*level > wc->active_node);
1315 if (*level == wc->active_node)
1316 leave_shared_node(root, wc, *level);
1323 static int check_root_dir(struct inode_record *rec)
1325 struct inode_backref *backref;
1328 if (!rec->found_inode_item || rec->errors)
1330 if (rec->nlink != 1 || rec->found_link != 0)
1332 if (list_empty(&rec->backrefs))
1334 backref = list_entry(rec->backrefs.next, struct inode_backref, list);
1335 if (!backref->found_inode_ref)
1337 if (backref->index != 0 || backref->namelen != 2 ||
1338 memcmp(backref->name, "..", 2))
1340 if (backref->found_dir_index || backref->found_dir_item)
1347 static int check_inode_recs(struct btrfs_root *root,
1348 struct cache_tree *inode_cache)
1350 struct cache_extent *cache;
1351 struct ptr_node *node;
1352 struct inode_record *rec;
1353 struct inode_backref *backref;
1356 u64 root_dirid = btrfs_root_dirid(&root->root_item);
1358 if (btrfs_root_refs(&root->root_item) == 0) {
1359 if (!cache_tree_empty(inode_cache))
1360 fprintf(stderr, "warning line %d\n", __LINE__);
1364 rec = get_inode_rec(inode_cache, root_dirid, 0);
1366 ret = check_root_dir(rec);
1368 fprintf(stderr, "root %llu root dir %llu error\n",
1369 (unsigned long long)root->root_key.objectid,
1370 (unsigned long long)root_dirid);
1374 fprintf(stderr, "root %llu root dir %llu not found\n",
1375 (unsigned long long)root->root_key.objectid,
1376 (unsigned long long)root_dirid);
1380 cache = search_cache_extent(inode_cache, 0);
1383 node = container_of(cache, struct ptr_node, cache);
1385 remove_cache_extent(inode_cache, &node->cache);
1387 if (rec->ino == root_dirid ||
1388 rec->ino == BTRFS_ORPHAN_OBJECTID) {
1389 free_inode_rec(rec);
1393 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
1394 ret = check_orphan_item(root, rec->ino);
1396 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
1397 if (can_free_inode_rec(rec)) {
1398 free_inode_rec(rec);
1404 if (!rec->found_inode_item)
1405 rec->errors |= I_ERR_NO_INODE_ITEM;
1406 if (rec->found_link != rec->nlink)
1407 rec->errors |= I_ERR_LINK_COUNT_WRONG;
1408 fprintf(stderr, "root %llu inode %llu errors %x\n",
1409 (unsigned long long) root->root_key.objectid,
1410 (unsigned long long) rec->ino, rec->errors);
1411 list_for_each_entry(backref, &rec->backrefs, list) {
1412 if (!backref->found_dir_item)
1413 backref->errors |= REF_ERR_NO_DIR_ITEM;
1414 if (!backref->found_dir_index)
1415 backref->errors |= REF_ERR_NO_DIR_INDEX;
1416 if (!backref->found_inode_ref)
1417 backref->errors |= REF_ERR_NO_INODE_REF;
1418 fprintf(stderr, "\tunresolved ref dir %llu index %llu"
1419 " namelen %u name %s filetype %d error %x\n",
1420 (unsigned long long)backref->dir,
1421 (unsigned long long)backref->index,
1422 backref->namelen, backref->name,
1423 backref->filetype, backref->errors);
1425 free_inode_rec(rec);
1427 return (error > 0) ? -1 : 0;
1430 static struct root_record *get_root_rec(struct cache_tree *root_cache,
1433 struct cache_extent *cache;
1434 struct root_record *rec = NULL;
1437 cache = lookup_cache_extent(root_cache, objectid, 1);
1439 rec = container_of(cache, struct root_record, cache);
1441 rec = calloc(1, sizeof(*rec));
1442 rec->objectid = objectid;
1443 INIT_LIST_HEAD(&rec->backrefs);
1444 rec->cache.start = objectid;
1445 rec->cache.size = 1;
1447 ret = insert_cache_extent(root_cache, &rec->cache);
1453 static struct root_backref *get_root_backref(struct root_record *rec,
1454 u64 ref_root, u64 dir, u64 index,
1455 const char *name, int namelen)
1457 struct root_backref *backref;
1459 list_for_each_entry(backref, &rec->backrefs, list) {
1460 if (backref->ref_root != ref_root || backref->dir != dir ||
1461 backref->namelen != namelen)
1463 if (memcmp(name, backref->name, namelen))
1468 backref = malloc(sizeof(*backref) + namelen + 1);
1469 memset(backref, 0, sizeof(*backref));
1470 backref->ref_root = ref_root;
1472 backref->index = index;
1473 backref->namelen = namelen;
1474 memcpy(backref->name, name, namelen);
1475 backref->name[namelen] = '\0';
1476 list_add_tail(&backref->list, &rec->backrefs);
1480 static void free_root_record(struct cache_extent *cache)
1482 struct root_record *rec;
1483 struct root_backref *backref;
1485 rec = container_of(cache, struct root_record, cache);
1486 while (!list_empty(&rec->backrefs)) {
1487 backref = list_entry(rec->backrefs.next,
1488 struct root_backref, list);
1489 list_del(&backref->list);
1496 FREE_EXTENT_CACHE_BASED_TREE(root_recs, free_root_record);
1498 static int add_root_backref(struct cache_tree *root_cache,
1499 u64 root_id, u64 ref_root, u64 dir, u64 index,
1500 const char *name, int namelen,
1501 int item_type, int errors)
1503 struct root_record *rec;
1504 struct root_backref *backref;
1506 rec = get_root_rec(root_cache, root_id);
1507 backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
1509 backref->errors |= errors;
1511 if (item_type != BTRFS_DIR_ITEM_KEY) {
1512 if (backref->found_dir_index || backref->found_back_ref ||
1513 backref->found_forward_ref) {
1514 if (backref->index != index)
1515 backref->errors |= REF_ERR_INDEX_UNMATCH;
1517 backref->index = index;
1521 if (item_type == BTRFS_DIR_ITEM_KEY) {
1522 if (backref->found_forward_ref)
1524 backref->found_dir_item = 1;
1525 } else if (item_type == BTRFS_DIR_INDEX_KEY) {
1526 backref->found_dir_index = 1;
1527 } else if (item_type == BTRFS_ROOT_REF_KEY) {
1528 if (backref->found_forward_ref)
1529 backref->errors |= REF_ERR_DUP_ROOT_REF;
1530 else if (backref->found_dir_item)
1532 backref->found_forward_ref = 1;
1533 } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
1534 if (backref->found_back_ref)
1535 backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
1536 backref->found_back_ref = 1;
1541 if (backref->found_forward_ref && backref->found_dir_item)
1542 backref->reachable = 1;
1546 static int merge_root_recs(struct btrfs_root *root,
1547 struct cache_tree *src_cache,
1548 struct cache_tree *dst_cache)
1550 struct cache_extent *cache;
1551 struct ptr_node *node;
1552 struct inode_record *rec;
1553 struct inode_backref *backref;
1555 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
1556 free_inode_recs_tree(src_cache);
1561 cache = search_cache_extent(src_cache, 0);
1564 node = container_of(cache, struct ptr_node, cache);
1566 remove_cache_extent(src_cache, &node->cache);
1569 if (!is_child_root(root, root->objectid, rec->ino))
1572 list_for_each_entry(backref, &rec->backrefs, list) {
1573 BUG_ON(backref->found_inode_ref);
1574 if (backref->found_dir_item)
1575 add_root_backref(dst_cache, rec->ino,
1576 root->root_key.objectid, backref->dir,
1577 backref->index, backref->name,
1578 backref->namelen, BTRFS_DIR_ITEM_KEY,
1580 if (backref->found_dir_index)
1581 add_root_backref(dst_cache, rec->ino,
1582 root->root_key.objectid, backref->dir,
1583 backref->index, backref->name,
1584 backref->namelen, BTRFS_DIR_INDEX_KEY,
1588 free_inode_rec(rec);
1593 static int check_root_refs(struct btrfs_root *root,
1594 struct cache_tree *root_cache)
1596 struct root_record *rec;
1597 struct root_record *ref_root;
1598 struct root_backref *backref;
1599 struct cache_extent *cache;
1605 rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
1608 /* fixme: this can not detect circular references */
1611 cache = search_cache_extent(root_cache, 0);
1615 rec = container_of(cache, struct root_record, cache);
1616 cache = next_cache_extent(cache);
1618 if (rec->found_ref == 0)
1621 list_for_each_entry(backref, &rec->backrefs, list) {
1622 if (!backref->reachable)
1625 ref_root = get_root_rec(root_cache,
1627 if (ref_root->found_ref > 0)
1630 backref->reachable = 0;
1632 if (rec->found_ref == 0)
1638 cache = search_cache_extent(root_cache, 0);
1642 rec = container_of(cache, struct root_record, cache);
1643 cache = next_cache_extent(cache);
1645 if (rec->found_ref == 0 &&
1646 rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
1647 rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
1648 ret = check_orphan_item(root->fs_info->tree_root,
1654 * If we don't have a root item then we likely just have
1655 * a dir item in a snapshot for this root but no actual
1656 * ref key or anything so it's meaningless.
1658 if (!rec->found_root_item)
1661 fprintf(stderr, "fs tree %llu not referenced\n",
1662 (unsigned long long)rec->objectid);
1666 if (rec->found_ref > 0 && !rec->found_root_item)
1668 list_for_each_entry(backref, &rec->backrefs, list) {
1669 if (!backref->found_dir_item)
1670 backref->errors |= REF_ERR_NO_DIR_ITEM;
1671 if (!backref->found_dir_index)
1672 backref->errors |= REF_ERR_NO_DIR_INDEX;
1673 if (!backref->found_back_ref)
1674 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
1675 if (!backref->found_forward_ref)
1676 backref->errors |= REF_ERR_NO_ROOT_REF;
1677 if (backref->reachable && backref->errors)
1684 fprintf(stderr, "fs tree %llu refs %u %s\n",
1685 (unsigned long long)rec->objectid, rec->found_ref,
1686 rec->found_root_item ? "" : "not found");
1688 list_for_each_entry(backref, &rec->backrefs, list) {
1689 if (!backref->reachable)
1691 if (!backref->errors && rec->found_root_item)
1693 fprintf(stderr, "\tunresolved ref root %llu dir %llu"
1694 " index %llu namelen %u name %s error %x\n",
1695 (unsigned long long)backref->ref_root,
1696 (unsigned long long)backref->dir,
1697 (unsigned long long)backref->index,
1698 backref->namelen, backref->name,
1702 return errors > 0 ? 1 : 0;
1705 static int process_root_ref(struct extent_buffer *eb, int slot,
1706 struct btrfs_key *key,
1707 struct cache_tree *root_cache)
1713 struct btrfs_root_ref *ref;
1714 char namebuf[BTRFS_NAME_LEN];
1717 ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
1719 dirid = btrfs_root_ref_dirid(eb, ref);
1720 index = btrfs_root_ref_sequence(eb, ref);
1721 name_len = btrfs_root_ref_name_len(eb, ref);
1723 if (name_len <= BTRFS_NAME_LEN) {
1727 len = BTRFS_NAME_LEN;
1728 error = REF_ERR_NAME_TOO_LONG;
1730 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
1732 if (key->type == BTRFS_ROOT_REF_KEY) {
1733 add_root_backref(root_cache, key->offset, key->objectid, dirid,
1734 index, namebuf, len, key->type, error);
1736 add_root_backref(root_cache, key->objectid, key->offset, dirid,
1737 index, namebuf, len, key->type, error);
1742 static int check_fs_root(struct btrfs_root *root,
1743 struct cache_tree *root_cache,
1744 struct walk_control *wc)
1749 struct btrfs_path path;
1750 struct shared_node root_node;
1751 struct root_record *rec;
1752 struct btrfs_root_item *root_item = &root->root_item;
1754 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1755 rec = get_root_rec(root_cache, root->root_key.objectid);
1756 if (btrfs_root_refs(root_item) > 0)
1757 rec->found_root_item = 1;
1760 btrfs_init_path(&path);
1761 memset(&root_node, 0, sizeof(root_node));
1762 cache_tree_init(&root_node.root_cache);
1763 cache_tree_init(&root_node.inode_cache);
1765 level = btrfs_header_level(root->node);
1766 memset(wc->nodes, 0, sizeof(wc->nodes));
1767 wc->nodes[level] = &root_node;
1768 wc->active_node = level;
1769 wc->root_level = level;
1771 if (btrfs_root_refs(root_item) > 0 ||
1772 btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1773 path.nodes[level] = root->node;
1774 extent_buffer_get(root->node);
1775 path.slots[level] = 0;
1777 struct btrfs_key key;
1778 struct btrfs_disk_key found_key;
1780 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1781 level = root_item->drop_level;
1782 path.lowest_level = level;
1783 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1785 btrfs_node_key(path.nodes[level], &found_key,
1787 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
1788 sizeof(found_key)));
1792 wret = walk_down_tree(root, &path, wc, &level);
1798 wret = walk_up_tree(root, &path, wc, &level);
1804 btrfs_release_path(&path);
1806 merge_root_recs(root, &root_node.root_cache, root_cache);
1808 if (root_node.current) {
1809 root_node.current->checked = 1;
1810 maybe_free_inode_rec(&root_node.inode_cache,
1814 ret = check_inode_recs(root, &root_node.inode_cache);
1818 static int fs_root_objectid(u64 objectid)
1820 if (objectid == BTRFS_FS_TREE_OBJECTID ||
1821 objectid == BTRFS_TREE_RELOC_OBJECTID ||
1822 objectid == BTRFS_DATA_RELOC_TREE_OBJECTID ||
1823 (objectid >= BTRFS_FIRST_FREE_OBJECTID &&
1824 objectid <= BTRFS_LAST_FREE_OBJECTID))
1829 static int check_fs_roots(struct btrfs_root *root,
1830 struct cache_tree *root_cache)
1832 struct btrfs_path path;
1833 struct btrfs_key key;
1834 struct walk_control wc;
1835 struct extent_buffer *leaf;
1836 struct btrfs_root *tmp_root;
1837 struct btrfs_root *tree_root = root->fs_info->tree_root;
1841 memset(&wc, 0, sizeof(wc));
1842 cache_tree_init(&wc.shared);
1843 btrfs_init_path(&path);
1847 key.type = BTRFS_ROOT_ITEM_KEY;
1848 ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
1851 leaf = path.nodes[0];
1852 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1853 ret = btrfs_next_leaf(tree_root, &path);
1856 leaf = path.nodes[0];
1858 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1859 if (key.type == BTRFS_ROOT_ITEM_KEY &&
1860 fs_root_objectid(key.objectid)) {
1861 tmp_root = btrfs_read_fs_root_no_cache(root->fs_info,
1863 if (IS_ERR(tmp_root)) {
1867 ret = check_fs_root(tmp_root, root_cache, &wc);
1870 btrfs_free_fs_root(tmp_root);
1871 } else if (key.type == BTRFS_ROOT_REF_KEY ||
1872 key.type == BTRFS_ROOT_BACKREF_KEY) {
1873 process_root_ref(leaf, path.slots[0], &key,
1879 btrfs_release_path(&path);
1881 if (!cache_tree_empty(&wc.shared))
1882 fprintf(stderr, "warning line %d\n", __LINE__);
1887 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
1889 struct list_head *cur = rec->backrefs.next;
1890 struct extent_backref *back;
1891 struct tree_backref *tback;
1892 struct data_backref *dback;
1896 while(cur != &rec->backrefs) {
1897 back = list_entry(cur, struct extent_backref, list);
1899 if (!back->found_extent_tree) {
1903 if (back->is_data) {
1904 dback = (struct data_backref *)back;
1905 fprintf(stderr, "Backref %llu %s %llu"
1906 " owner %llu offset %llu num_refs %lu"
1907 " not found in extent tree\n",
1908 (unsigned long long)rec->start,
1909 back->full_backref ?
1911 back->full_backref ?
1912 (unsigned long long)dback->parent:
1913 (unsigned long long)dback->root,
1914 (unsigned long long)dback->owner,
1915 (unsigned long long)dback->offset,
1916 (unsigned long)dback->num_refs);
1918 tback = (struct tree_backref *)back;
1919 fprintf(stderr, "Backref %llu parent %llu"
1920 " root %llu not found in extent tree\n",
1921 (unsigned long long)rec->start,
1922 (unsigned long long)tback->parent,
1923 (unsigned long long)tback->root);
1926 if (!back->is_data && !back->found_ref) {
1930 tback = (struct tree_backref *)back;
1931 fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
1932 (unsigned long long)rec->start,
1933 back->full_backref ? "parent" : "root",
1934 back->full_backref ?
1935 (unsigned long long)tback->parent :
1936 (unsigned long long)tback->root, back);
1938 if (back->is_data) {
1939 dback = (struct data_backref *)back;
1940 if (dback->found_ref != dback->num_refs) {
1944 fprintf(stderr, "Incorrect local backref count"
1945 " on %llu %s %llu owner %llu"
1946 " offset %llu found %u wanted %u back %p\n",
1947 (unsigned long long)rec->start,
1948 back->full_backref ?
1950 back->full_backref ?
1951 (unsigned long long)dback->parent:
1952 (unsigned long long)dback->root,
1953 (unsigned long long)dback->owner,
1954 (unsigned long long)dback->offset,
1955 dback->found_ref, dback->num_refs, back);
1957 if (dback->disk_bytenr != rec->start) {
1961 fprintf(stderr, "Backref disk bytenr does not"
1962 " match extent record, bytenr=%llu, "
1963 "ref bytenr=%llu\n",
1964 (unsigned long long)rec->start,
1965 (unsigned long long)dback->disk_bytenr);
1968 if (dback->bytes != rec->nr) {
1972 fprintf(stderr, "Backref bytes do not match "
1973 "extent backref, bytenr=%llu, ref "
1974 "bytes=%llu, backref bytes=%llu\n",
1975 (unsigned long long)rec->start,
1976 (unsigned long long)rec->nr,
1977 (unsigned long long)dback->bytes);
1980 if (!back->is_data) {
1983 dback = (struct data_backref *)back;
1984 found += dback->found_ref;
1987 if (found != rec->refs) {
1991 fprintf(stderr, "Incorrect global backref count "
1992 "on %llu found %llu wanted %llu\n",
1993 (unsigned long long)rec->start,
1994 (unsigned long long)found,
1995 (unsigned long long)rec->refs);
2001 static int free_all_extent_backrefs(struct extent_record *rec)
2003 struct extent_backref *back;
2004 struct list_head *cur;
2005 while (!list_empty(&rec->backrefs)) {
2006 cur = rec->backrefs.next;
2007 back = list_entry(cur, struct extent_backref, list);
2014 static void free_extent_record_cache(struct btrfs_fs_info *fs_info,
2015 struct cache_tree *extent_cache)
2017 struct cache_extent *cache;
2018 struct extent_record *rec;
2021 cache = first_cache_extent(extent_cache);
2024 rec = container_of(cache, struct extent_record, cache);
2025 btrfs_unpin_extent(fs_info, rec->start, rec->max_size);
2026 remove_cache_extent(extent_cache, cache);
2027 free_all_extent_backrefs(rec);
2032 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
2033 struct extent_record *rec)
2035 if (rec->content_checked && rec->owner_ref_checked &&
2036 rec->extent_item_refs == rec->refs && rec->refs > 0 &&
2037 rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0)) {
2038 remove_cache_extent(extent_cache, &rec->cache);
2039 free_all_extent_backrefs(rec);
2040 list_del_init(&rec->list);
2046 static int check_owner_ref(struct btrfs_root *root,
2047 struct extent_record *rec,
2048 struct extent_buffer *buf)
2050 struct extent_backref *node;
2051 struct tree_backref *back;
2052 struct btrfs_root *ref_root;
2053 struct btrfs_key key;
2054 struct btrfs_path path;
2055 struct extent_buffer *parent;
2060 list_for_each_entry(node, &rec->backrefs, list) {
2063 if (!node->found_ref)
2065 if (node->full_backref)
2067 back = (struct tree_backref *)node;
2068 if (btrfs_header_owner(buf) == back->root)
2071 BUG_ON(rec->is_root);
2073 /* try to find the block by search corresponding fs tree */
2074 key.objectid = btrfs_header_owner(buf);
2075 key.type = BTRFS_ROOT_ITEM_KEY;
2076 key.offset = (u64)-1;
2078 ref_root = btrfs_read_fs_root(root->fs_info, &key);
2079 if (IS_ERR(ref_root))
2082 level = btrfs_header_level(buf);
2084 btrfs_item_key_to_cpu(buf, &key, 0);
2086 btrfs_node_key_to_cpu(buf, &key, 0);
2088 btrfs_init_path(&path);
2089 path.lowest_level = level + 1;
2090 ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
2094 parent = path.nodes[level + 1];
2095 if (parent && buf->start == btrfs_node_blockptr(parent,
2096 path.slots[level + 1]))
2099 btrfs_release_path(&path);
2100 return found ? 0 : 1;
2103 static int is_extent_tree_record(struct extent_record *rec)
2105 struct list_head *cur = rec->backrefs.next;
2106 struct extent_backref *node;
2107 struct tree_backref *back;
2110 while(cur != &rec->backrefs) {
2111 node = list_entry(cur, struct extent_backref, list);
2115 back = (struct tree_backref *)node;
2116 if (node->full_backref)
2118 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
2125 static int record_bad_block_io(struct btrfs_fs_info *info,
2126 struct cache_tree *extent_cache,
2129 struct extent_record *rec;
2130 struct cache_extent *cache;
2131 struct btrfs_key key;
2133 cache = lookup_cache_extent(extent_cache, start, len);
2137 rec = container_of(cache, struct extent_record, cache);
2138 if (!is_extent_tree_record(rec))
2141 btrfs_disk_key_to_cpu(&key, &rec->parent_key);
2142 return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
2145 static int check_block(struct btrfs_root *root,
2146 struct cache_tree *extent_cache,
2147 struct extent_buffer *buf, u64 flags)
2149 struct extent_record *rec;
2150 struct cache_extent *cache;
2151 struct btrfs_key key;
2155 cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
2158 rec = container_of(cache, struct extent_record, cache);
2159 rec->generation = btrfs_header_generation(buf);
2161 level = btrfs_header_level(buf);
2162 if (btrfs_header_nritems(buf) > 0) {
2165 btrfs_item_key_to_cpu(buf, &key, 0);
2167 btrfs_node_key_to_cpu(buf, &key, 0);
2169 rec->info_objectid = key.objectid;
2171 rec->info_level = level;
2173 if (btrfs_is_leaf(buf))
2174 ret = btrfs_check_leaf(root, &rec->parent_key, buf);
2176 ret = btrfs_check_node(root, &rec->parent_key, buf);
2179 fprintf(stderr, "bad block %llu\n",
2180 (unsigned long long)buf->start);
2182 rec->content_checked = 1;
2183 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
2184 rec->owner_ref_checked = 1;
2186 ret = check_owner_ref(root, rec, buf);
2188 rec->owner_ref_checked = 1;
2192 maybe_free_extent_rec(extent_cache, rec);
2196 static struct tree_backref *find_tree_backref(struct extent_record *rec,
2197 u64 parent, u64 root)
2199 struct list_head *cur = rec->backrefs.next;
2200 struct extent_backref *node;
2201 struct tree_backref *back;
2203 while(cur != &rec->backrefs) {
2204 node = list_entry(cur, struct extent_backref, list);
2208 back = (struct tree_backref *)node;
2210 if (!node->full_backref)
2212 if (parent == back->parent)
2215 if (node->full_backref)
2217 if (back->root == root)
2224 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
2225 u64 parent, u64 root)
2227 struct tree_backref *ref = malloc(sizeof(*ref));
2228 memset(&ref->node, 0, sizeof(ref->node));
2230 ref->parent = parent;
2231 ref->node.full_backref = 1;
2234 ref->node.full_backref = 0;
2236 list_add_tail(&ref->node.list, &rec->backrefs);
2241 static struct data_backref *find_data_backref(struct extent_record *rec,
2242 u64 parent, u64 root,
2243 u64 owner, u64 offset,
2245 u64 disk_bytenr, u64 bytes)
2247 struct list_head *cur = rec->backrefs.next;
2248 struct extent_backref *node;
2249 struct data_backref *back;
2251 while(cur != &rec->backrefs) {
2252 node = list_entry(cur, struct extent_backref, list);
2256 back = (struct data_backref *)node;
2258 if (!node->full_backref)
2260 if (parent == back->parent)
2263 if (node->full_backref)
2265 if (back->root == root && back->owner == owner &&
2266 back->offset == offset) {
2267 if (found_ref && node->found_ref &&
2268 (back->bytes != bytes ||
2269 back->disk_bytenr != disk_bytenr))
2278 static struct data_backref *alloc_data_backref(struct extent_record *rec,
2279 u64 parent, u64 root,
2280 u64 owner, u64 offset,
2283 struct data_backref *ref = malloc(sizeof(*ref));
2284 memset(&ref->node, 0, sizeof(ref->node));
2285 ref->node.is_data = 1;
2288 ref->parent = parent;
2291 ref->node.full_backref = 1;
2295 ref->offset = offset;
2296 ref->node.full_backref = 0;
2298 ref->bytes = max_size;
2301 list_add_tail(&ref->node.list, &rec->backrefs);
2302 if (max_size > rec->max_size)
2303 rec->max_size = max_size;
2307 static int add_extent_rec(struct cache_tree *extent_cache,
2308 struct btrfs_key *parent_key,
2309 u64 start, u64 nr, u64 extent_item_refs,
2310 int is_root, int inc_ref, int set_checked,
2311 int metadata, int extent_rec, u64 max_size)
2313 struct extent_record *rec;
2314 struct cache_extent *cache;
2318 cache = lookup_cache_extent(extent_cache, start, nr);
2320 rec = container_of(cache, struct extent_record, cache);
2324 rec->nr = max(nr, max_size);
2327 * We need to make sure to reset nr to whatever the extent
2328 * record says was the real size, this way we can compare it to
2332 if (start != rec->start || rec->found_rec) {
2333 struct extent_record *tmp;
2336 if (list_empty(&rec->list))
2337 list_add_tail(&rec->list,
2338 &duplicate_extents);
2341 * We have to do this song and dance in case we
2342 * find an extent record that falls inside of
2343 * our current extent record but does not have
2344 * the same objectid.
2346 tmp = malloc(sizeof(*tmp));
2350 tmp->max_size = max_size;
2353 tmp->metadata = metadata;
2354 tmp->extent_item_refs = extent_item_refs;
2355 INIT_LIST_HEAD(&tmp->list);
2356 list_add_tail(&tmp->list, &rec->dups);
2357 rec->num_duplicates++;
2364 if (extent_item_refs && !dup) {
2365 if (rec->extent_item_refs) {
2366 fprintf(stderr, "block %llu rec "
2367 "extent_item_refs %llu, passed %llu\n",
2368 (unsigned long long)start,
2369 (unsigned long long)
2370 rec->extent_item_refs,
2371 (unsigned long long)extent_item_refs);
2373 rec->extent_item_refs = extent_item_refs;
2378 rec->content_checked = 1;
2379 rec->owner_ref_checked = 1;
2383 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
2385 if (rec->max_size < max_size)
2386 rec->max_size = max_size;
2388 maybe_free_extent_rec(extent_cache, rec);
2391 rec = malloc(sizeof(*rec));
2393 rec->max_size = max_size;
2394 rec->nr = max(nr, max_size);
2395 rec->found_rec = extent_rec;
2396 rec->content_checked = 0;
2397 rec->owner_ref_checked = 0;
2398 rec->num_duplicates = 0;
2399 rec->metadata = metadata;
2400 INIT_LIST_HEAD(&rec->backrefs);
2401 INIT_LIST_HEAD(&rec->dups);
2402 INIT_LIST_HEAD(&rec->list);
2414 if (extent_item_refs)
2415 rec->extent_item_refs = extent_item_refs;
2417 rec->extent_item_refs = 0;
2420 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
2422 memset(&rec->parent_key, 0, sizeof(*parent_key));
2424 rec->cache.start = start;
2425 rec->cache.size = nr;
2426 ret = insert_cache_extent(extent_cache, &rec->cache);
2430 rec->content_checked = 1;
2431 rec->owner_ref_checked = 1;
2436 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
2437 u64 parent, u64 root, int found_ref)
2439 struct extent_record *rec;
2440 struct tree_backref *back;
2441 struct cache_extent *cache;
2443 cache = lookup_cache_extent(extent_cache, bytenr, 1);
2445 add_extent_rec(extent_cache, NULL, bytenr,
2446 1, 0, 0, 0, 0, 1, 0, 0);
2447 cache = lookup_cache_extent(extent_cache, bytenr, 1);
2452 rec = container_of(cache, struct extent_record, cache);
2453 if (rec->start != bytenr) {
2457 back = find_tree_backref(rec, parent, root);
2459 back = alloc_tree_backref(rec, parent, root);
2462 if (back->node.found_ref) {
2463 fprintf(stderr, "Extent back ref already exists "
2464 "for %llu parent %llu root %llu \n",
2465 (unsigned long long)bytenr,
2466 (unsigned long long)parent,
2467 (unsigned long long)root);
2469 back->node.found_ref = 1;
2471 if (back->node.found_extent_tree) {
2472 fprintf(stderr, "Extent back ref already exists "
2473 "for %llu parent %llu root %llu \n",
2474 (unsigned long long)bytenr,
2475 (unsigned long long)parent,
2476 (unsigned long long)root);
2478 back->node.found_extent_tree = 1;
2483 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
2484 u64 parent, u64 root, u64 owner, u64 offset,
2485 u32 num_refs, int found_ref, u64 max_size)
2487 struct extent_record *rec;
2488 struct data_backref *back;
2489 struct cache_extent *cache;
2491 cache = lookup_cache_extent(extent_cache, bytenr, 1);
2493 add_extent_rec(extent_cache, NULL, bytenr, 1, 0, 0, 0, 0,
2495 cache = lookup_cache_extent(extent_cache, bytenr, 1);
2500 rec = container_of(cache, struct extent_record, cache);
2501 if (rec->max_size < max_size)
2502 rec->max_size = max_size;
2505 * If found_ref is set then max_size is the real size and must match the
2506 * existing refs. So if we have already found a ref then we need to
2507 * make sure that this ref matches the existing one, otherwise we need
2508 * to add a new backref so we can notice that the backrefs don't match
2509 * and we need to figure out who is telling the truth. This is to
2510 * account for that awful fsync bug I introduced where we'd end up with
2511 * a btrfs_file_extent_item that would have its length include multiple
2512 * prealloc extents or point inside of a prealloc extent.
2514 back = find_data_backref(rec, parent, root, owner, offset, found_ref,
2517 back = alloc_data_backref(rec, parent, root, owner, offset,
2521 BUG_ON(num_refs != 1);
2522 if (back->node.found_ref)
2523 BUG_ON(back->bytes != max_size);
2524 back->node.found_ref = 1;
2525 back->found_ref += 1;
2526 back->bytes = max_size;
2527 back->disk_bytenr = bytenr;
2529 rec->content_checked = 1;
2530 rec->owner_ref_checked = 1;
2532 if (back->node.found_extent_tree) {
2533 fprintf(stderr, "Extent back ref already exists "
2534 "for %llu parent %llu root %llu"
2535 "owner %llu offset %llu num_refs %lu\n",
2536 (unsigned long long)bytenr,
2537 (unsigned long long)parent,
2538 (unsigned long long)root,
2539 (unsigned long long)owner,
2540 (unsigned long long)offset,
2541 (unsigned long)num_refs);
2543 back->num_refs = num_refs;
2544 back->node.found_extent_tree = 1;
2549 static int add_pending(struct cache_tree *pending,
2550 struct cache_tree *seen, u64 bytenr, u32 size)
2553 ret = add_cache_extent(seen, bytenr, size);
2556 add_cache_extent(pending, bytenr, size);
2560 static int pick_next_pending(struct cache_tree *pending,
2561 struct cache_tree *reada,
2562 struct cache_tree *nodes,
2563 u64 last, struct block_info *bits, int bits_nr,
2566 unsigned long node_start = last;
2567 struct cache_extent *cache;
2570 cache = search_cache_extent(reada, 0);
2572 bits[0].start = cache->start;
2573 bits[1].size = cache->size;
2578 if (node_start > 32768)
2579 node_start -= 32768;
2581 cache = search_cache_extent(nodes, node_start);
2583 cache = search_cache_extent(nodes, 0);
2586 cache = search_cache_extent(pending, 0);
2591 bits[ret].start = cache->start;
2592 bits[ret].size = cache->size;
2593 cache = next_cache_extent(cache);
2595 } while (cache && ret < bits_nr);
2601 bits[ret].start = cache->start;
2602 bits[ret].size = cache->size;
2603 cache = next_cache_extent(cache);
2605 } while (cache && ret < bits_nr);
2607 if (bits_nr - ret > 8) {
2608 u64 lookup = bits[0].start + bits[0].size;
2609 struct cache_extent *next;
2610 next = search_cache_extent(pending, lookup);
2612 if (next->start - lookup > 32768)
2614 bits[ret].start = next->start;
2615 bits[ret].size = next->size;
2616 lookup = next->start + next->size;
2620 next = next_cache_extent(next);
2628 static void free_chunk_record(struct cache_extent *cache)
2630 struct chunk_record *rec;
2632 rec = container_of(cache, struct chunk_record, cache);
2636 void free_chunk_cache_tree(struct cache_tree *chunk_cache)
2638 cache_tree_free_extents(chunk_cache, free_chunk_record);
2641 static void free_device_record(struct rb_node *node)
2643 struct device_record *rec;
2645 rec = container_of(node, struct device_record, node);
2649 FREE_RB_BASED_TREE(device_cache, free_device_record);
2651 int insert_block_group_record(struct block_group_tree *tree,
2652 struct block_group_record *bg_rec)
2656 ret = insert_cache_extent(&tree->tree, &bg_rec->cache);
2660 list_add_tail(&bg_rec->list, &tree->block_groups);
2664 static void free_block_group_record(struct cache_extent *cache)
2666 struct block_group_record *rec;
2668 rec = container_of(cache, struct block_group_record, cache);
2672 void free_block_group_tree(struct block_group_tree *tree)
2674 cache_tree_free_extents(&tree->tree, free_block_group_record);
2677 int insert_device_extent_record(struct device_extent_tree *tree,
2678 struct device_extent_record *de_rec)
2683 * Device extent is a bit different from the other extents, because
2684 * the extents which belong to the different devices may have the
2685 * same start and size, so we need use the special extent cache
2686 * search/insert functions.
2688 ret = insert_cache_extent2(&tree->tree, &de_rec->cache);
2692 list_add_tail(&de_rec->chunk_list, &tree->no_chunk_orphans);
2693 list_add_tail(&de_rec->device_list, &tree->no_device_orphans);
2697 static void free_device_extent_record(struct cache_extent *cache)
2699 struct device_extent_record *rec;
2701 rec = container_of(cache, struct device_extent_record, cache);
2705 void free_device_extent_tree(struct device_extent_tree *tree)
2707 cache_tree_free_extents(&tree->tree, free_device_extent_record);
2710 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2711 static int process_extent_ref_v0(struct cache_tree *extent_cache,
2712 struct extent_buffer *leaf, int slot)
2714 struct btrfs_extent_ref_v0 *ref0;
2715 struct btrfs_key key;
2717 btrfs_item_key_to_cpu(leaf, &key, slot);
2718 ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
2719 if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
2720 add_tree_backref(extent_cache, key.objectid, key.offset, 0, 0);
2722 add_data_backref(extent_cache, key.objectid, key.offset, 0,
2723 0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
2729 struct chunk_record *btrfs_new_chunk_record(struct extent_buffer *leaf,
2730 struct btrfs_key *key,
2733 struct btrfs_chunk *ptr;
2734 struct chunk_record *rec;
2737 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2738 num_stripes = btrfs_chunk_num_stripes(leaf, ptr);
2740 rec = malloc(btrfs_chunk_record_size(num_stripes));
2742 fprintf(stderr, "memory allocation failed\n");
2746 memset(rec, 0, btrfs_chunk_record_size(num_stripes));
2748 INIT_LIST_HEAD(&rec->list);
2749 INIT_LIST_HEAD(&rec->dextents);
2752 rec->cache.start = key->offset;
2753 rec->cache.size = btrfs_chunk_length(leaf, ptr);
2755 rec->generation = btrfs_header_generation(leaf);
2757 rec->objectid = key->objectid;
2758 rec->type = key->type;
2759 rec->offset = key->offset;
2761 rec->length = rec->cache.size;
2762 rec->owner = btrfs_chunk_owner(leaf, ptr);
2763 rec->stripe_len = btrfs_chunk_stripe_len(leaf, ptr);
2764 rec->type_flags = btrfs_chunk_type(leaf, ptr);
2765 rec->io_width = btrfs_chunk_io_width(leaf, ptr);
2766 rec->io_align = btrfs_chunk_io_align(leaf, ptr);
2767 rec->sector_size = btrfs_chunk_sector_size(leaf, ptr);
2768 rec->num_stripes = num_stripes;
2769 rec->sub_stripes = btrfs_chunk_sub_stripes(leaf, ptr);
2771 for (i = 0; i < rec->num_stripes; ++i) {
2772 rec->stripes[i].devid =
2773 btrfs_stripe_devid_nr(leaf, ptr, i);
2774 rec->stripes[i].offset =
2775 btrfs_stripe_offset_nr(leaf, ptr, i);
2776 read_extent_buffer(leaf, rec->stripes[i].dev_uuid,
2777 (unsigned long)btrfs_stripe_dev_uuid_nr(ptr, i),
2784 static int process_chunk_item(struct cache_tree *chunk_cache,
2785 struct btrfs_key *key, struct extent_buffer *eb,
2788 struct chunk_record *rec;
2791 rec = btrfs_new_chunk_record(eb, key, slot);
2792 ret = insert_cache_extent(chunk_cache, &rec->cache);
2794 fprintf(stderr, "Chunk[%llu, %llu] existed.\n",
2795 rec->offset, rec->length);
2802 static int process_device_item(struct rb_root *dev_cache,
2803 struct btrfs_key *key, struct extent_buffer *eb, int slot)
2805 struct btrfs_dev_item *ptr;
2806 struct device_record *rec;
2809 ptr = btrfs_item_ptr(eb,
2810 slot, struct btrfs_dev_item);
2812 rec = malloc(sizeof(*rec));
2814 fprintf(stderr, "memory allocation failed\n");
2818 rec->devid = key->offset;
2819 rec->generation = btrfs_header_generation(eb);
2821 rec->objectid = key->objectid;
2822 rec->type = key->type;
2823 rec->offset = key->offset;
2825 rec->devid = btrfs_device_id(eb, ptr);
2826 rec->total_byte = btrfs_device_total_bytes(eb, ptr);
2827 rec->byte_used = btrfs_device_bytes_used(eb, ptr);
2829 ret = rb_insert(dev_cache, &rec->node, device_record_compare);
2831 fprintf(stderr, "Device[%llu] existed.\n", rec->devid);
2838 struct block_group_record *
2839 btrfs_new_block_group_record(struct extent_buffer *leaf, struct btrfs_key *key,
2842 struct btrfs_block_group_item *ptr;
2843 struct block_group_record *rec;
2845 rec = malloc(sizeof(*rec));
2847 fprintf(stderr, "memory allocation failed\n");
2850 memset(rec, 0, sizeof(*rec));
2852 rec->cache.start = key->objectid;
2853 rec->cache.size = key->offset;
2855 rec->generation = btrfs_header_generation(leaf);
2857 rec->objectid = key->objectid;
2858 rec->type = key->type;
2859 rec->offset = key->offset;
2861 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item);
2862 rec->flags = btrfs_disk_block_group_flags(leaf, ptr);
2864 INIT_LIST_HEAD(&rec->list);
2869 static int process_block_group_item(struct block_group_tree *block_group_cache,
2870 struct btrfs_key *key,
2871 struct extent_buffer *eb, int slot)
2873 struct block_group_record *rec;
2876 rec = btrfs_new_block_group_record(eb, key, slot);
2877 ret = insert_block_group_record(block_group_cache, rec);
2879 fprintf(stderr, "Block Group[%llu, %llu] existed.\n",
2880 rec->objectid, rec->offset);
2887 struct device_extent_record *
2888 btrfs_new_device_extent_record(struct extent_buffer *leaf,
2889 struct btrfs_key *key, int slot)
2891 struct device_extent_record *rec;
2892 struct btrfs_dev_extent *ptr;
2894 rec = malloc(sizeof(*rec));
2896 fprintf(stderr, "memory allocation failed\n");
2899 memset(rec, 0, sizeof(*rec));
2901 rec->cache.objectid = key->objectid;
2902 rec->cache.start = key->offset;
2904 rec->generation = btrfs_header_generation(leaf);
2906 rec->objectid = key->objectid;
2907 rec->type = key->type;
2908 rec->offset = key->offset;
2910 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
2911 rec->chunk_objecteid =
2912 btrfs_dev_extent_chunk_objectid(leaf, ptr);
2914 btrfs_dev_extent_chunk_offset(leaf, ptr);
2915 rec->length = btrfs_dev_extent_length(leaf, ptr);
2916 rec->cache.size = rec->length;
2918 INIT_LIST_HEAD(&rec->chunk_list);
2919 INIT_LIST_HEAD(&rec->device_list);
2925 process_device_extent_item(struct device_extent_tree *dev_extent_cache,
2926 struct btrfs_key *key, struct extent_buffer *eb,
2929 struct device_extent_record *rec;
2932 rec = btrfs_new_device_extent_record(eb, key, slot);
2933 ret = insert_device_extent_record(dev_extent_cache, rec);
2936 "Device extent[%llu, %llu, %llu] existed.\n",
2937 rec->objectid, rec->offset, rec->length);
2944 static int process_extent_item(struct btrfs_root *root,
2945 struct cache_tree *extent_cache,
2946 struct extent_buffer *eb, int slot)
2948 struct btrfs_extent_item *ei;
2949 struct btrfs_extent_inline_ref *iref;
2950 struct btrfs_extent_data_ref *dref;
2951 struct btrfs_shared_data_ref *sref;
2952 struct btrfs_key key;
2956 u32 item_size = btrfs_item_size_nr(eb, slot);
2962 btrfs_item_key_to_cpu(eb, &key, slot);
2964 if (key.type == BTRFS_METADATA_ITEM_KEY) {
2966 num_bytes = root->leafsize;
2968 num_bytes = key.offset;
2971 if (item_size < sizeof(*ei)) {
2972 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2973 struct btrfs_extent_item_v0 *ei0;
2974 BUG_ON(item_size != sizeof(*ei0));
2975 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
2976 refs = btrfs_extent_refs_v0(eb, ei0);
2980 return add_extent_rec(extent_cache, NULL, key.objectid,
2981 num_bytes, refs, 0, 0, 0, metadata, 1,
2985 ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
2986 refs = btrfs_extent_refs(eb, ei);
2988 add_extent_rec(extent_cache, NULL, key.objectid, num_bytes,
2989 refs, 0, 0, 0, metadata, 1, num_bytes);
2991 ptr = (unsigned long)(ei + 1);
2992 if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
2993 key.type == BTRFS_EXTENT_ITEM_KEY)
2994 ptr += sizeof(struct btrfs_tree_block_info);
2996 end = (unsigned long)ei + item_size;
2998 iref = (struct btrfs_extent_inline_ref *)ptr;
2999 type = btrfs_extent_inline_ref_type(eb, iref);
3000 offset = btrfs_extent_inline_ref_offset(eb, iref);
3002 case BTRFS_TREE_BLOCK_REF_KEY:
3003 add_tree_backref(extent_cache, key.objectid,
3006 case BTRFS_SHARED_BLOCK_REF_KEY:
3007 add_tree_backref(extent_cache, key.objectid,
3010 case BTRFS_EXTENT_DATA_REF_KEY:
3011 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3012 add_data_backref(extent_cache, key.objectid, 0,
3013 btrfs_extent_data_ref_root(eb, dref),
3014 btrfs_extent_data_ref_objectid(eb,
3016 btrfs_extent_data_ref_offset(eb, dref),
3017 btrfs_extent_data_ref_count(eb, dref),
3020 case BTRFS_SHARED_DATA_REF_KEY:
3021 sref = (struct btrfs_shared_data_ref *)(iref + 1);
3022 add_data_backref(extent_cache, key.objectid, offset,
3024 btrfs_shared_data_ref_count(eb, sref),
3028 fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
3029 key.objectid, key.type, num_bytes);
3032 ptr += btrfs_extent_inline_ref_size(type);
3039 static int check_cache_range(struct btrfs_root *root,
3040 struct btrfs_block_group_cache *cache,
3041 u64 offset, u64 bytes)
3043 struct btrfs_free_space *entry;
3049 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
3050 bytenr = btrfs_sb_offset(i);
3051 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
3052 cache->key.objectid, bytenr, 0,
3053 &logical, &nr, &stripe_len);
3058 if (logical[nr] + stripe_len <= offset)
3060 if (offset + bytes <= logical[nr])
3062 if (logical[nr] == offset) {
3063 if (stripe_len >= bytes) {
3067 bytes -= stripe_len;
3068 offset += stripe_len;
3069 } else if (logical[nr] < offset) {
3070 if (logical[nr] + stripe_len >=
3075 bytes = (offset + bytes) -
3076 (logical[nr] + stripe_len);
3077 offset = logical[nr] + stripe_len;
3080 * Could be tricky, the super may land in the
3081 * middle of the area we're checking. First
3082 * check the easiest case, it's at the end.
3084 if (logical[nr] + stripe_len >=
3086 bytes = logical[nr] - offset;
3090 /* Check the left side */
3091 ret = check_cache_range(root, cache,
3093 logical[nr] - offset);
3099 /* Now we continue with the right side */
3100 bytes = (offset + bytes) -
3101 (logical[nr] + stripe_len);
3102 offset = logical[nr] + stripe_len;
3109 entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
3111 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
3112 offset, offset+bytes);
3116 if (entry->offset != offset) {
3117 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
3122 if (entry->bytes != bytes) {
3123 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
3124 bytes, entry->bytes, offset);
3128 unlink_free_space(cache->free_space_ctl, entry);
3133 static int verify_space_cache(struct btrfs_root *root,
3134 struct btrfs_block_group_cache *cache)
3136 struct btrfs_path *path;
3137 struct extent_buffer *leaf;
3138 struct btrfs_key key;
3142 path = btrfs_alloc_path();
3146 root = root->fs_info->extent_root;
3148 last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
3150 key.objectid = last;
3152 key.type = BTRFS_EXTENT_ITEM_KEY;
3154 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3159 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3160 ret = btrfs_next_leaf(root, path);
3168 leaf = path->nodes[0];
3169 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3170 if (key.objectid >= cache->key.offset + cache->key.objectid)
3172 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3173 key.type != BTRFS_METADATA_ITEM_KEY) {
3178 if (last == key.objectid) {
3179 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3180 last = key.objectid + key.offset;
3182 last = key.objectid + root->leafsize;
3187 ret = check_cache_range(root, cache, last,
3188 key.objectid - last);
3191 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3192 last = key.objectid + key.offset;
3194 last = key.objectid + root->leafsize;
3198 if (last < cache->key.objectid + cache->key.offset)
3199 ret = check_cache_range(root, cache, last,
3200 cache->key.objectid +
3201 cache->key.offset - last);
3202 btrfs_free_path(path);
3205 !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
3206 fprintf(stderr, "There are still entries left in the space "
3214 static int check_space_cache(struct btrfs_root *root)
3216 struct btrfs_block_group_cache *cache;
3217 u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
3221 if (btrfs_super_generation(root->fs_info->super_copy) !=
3222 btrfs_super_cache_generation(root->fs_info->super_copy)) {
3223 printf("cache and super generation don't match, space cache "
3224 "will be invalidated\n");
3229 cache = btrfs_lookup_first_block_group(root->fs_info, start);
3233 start = cache->key.objectid + cache->key.offset;
3234 if (!cache->free_space_ctl) {
3235 if (btrfs_init_free_space_ctl(cache,
3236 root->sectorsize)) {
3241 btrfs_remove_free_space_cache(cache);
3244 ret = load_free_space_cache(root->fs_info, cache);
3248 ret = verify_space_cache(root, cache);
3250 fprintf(stderr, "cache appears valid but isnt %Lu\n",
3251 cache->key.objectid);
3256 return error ? -EINVAL : 0;
3259 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
3262 struct btrfs_path *path;
3263 struct extent_buffer *leaf;
3264 struct btrfs_key key;
3267 path = btrfs_alloc_path();
3269 fprintf(stderr, "Error allocing path\n");
3273 key.objectid = bytenr;
3274 key.type = BTRFS_EXTENT_ITEM_KEY;
3279 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
3282 fprintf(stderr, "Error looking up extent record %d\n", ret);
3283 btrfs_free_path(path);
3289 btrfs_prev_leaf(root, path);
3292 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3295 * Block group items come before extent items if they have the same
3296 * bytenr, so walk back one more just in case. Dear future traveler,
3297 * first congrats on mastering time travel. Now if it's not too much
3298 * trouble could you go back to 2006 and tell Chris to make the
3299 * BLOCK_GROUP_ITEM_KEY lower than the EXTENT_ITEM_KEY please?
3301 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3305 btrfs_prev_leaf(root, path);
3309 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3310 ret = btrfs_next_leaf(root, path);
3312 fprintf(stderr, "Error going to next leaf "
3314 btrfs_free_path(path);
3320 leaf = path->nodes[0];
3321 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3322 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
3326 if (key.objectid + key.offset < bytenr) {
3330 if (key.objectid > bytenr + num_bytes)
3333 if (key.objectid == bytenr) {
3334 if (key.offset >= num_bytes) {
3338 num_bytes -= key.offset;
3339 bytenr += key.offset;
3340 } else if (key.objectid < bytenr) {
3341 if (key.objectid + key.offset >= bytenr + num_bytes) {
3345 num_bytes = (bytenr + num_bytes) -
3346 (key.objectid + key.offset);
3347 bytenr = key.objectid + key.offset;
3349 if (key.objectid + key.offset < bytenr + num_bytes) {
3350 u64 new_start = key.objectid + key.offset;
3351 u64 new_bytes = bytenr + num_bytes - new_start;
3354 * Weird case, the extent is in the middle of
3355 * our range, we'll have to search one side
3356 * and then the other. Not sure if this happens
3357 * in real life, but no harm in coding it up
3358 * anyway just in case.
3360 btrfs_release_path(path);
3361 ret = check_extent_exists(root, new_start,
3364 fprintf(stderr, "Right section didn't "
3368 num_bytes = key.objectid - bytenr;
3371 num_bytes = key.objectid - bytenr;
3378 fprintf(stderr, "There are no extents for csum range "
3379 "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
3383 btrfs_free_path(path);
3387 static int check_csums(struct btrfs_root *root)
3389 struct btrfs_path *path;
3390 struct extent_buffer *leaf;
3391 struct btrfs_key key;
3392 u64 offset = 0, num_bytes = 0;
3393 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
3397 root = root->fs_info->csum_root;
3399 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
3400 key.type = BTRFS_EXTENT_CSUM_KEY;
3403 path = btrfs_alloc_path();
3407 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3409 fprintf(stderr, "Error searching csum tree %d\n", ret);
3410 btrfs_free_path(path);
3414 if (ret > 0 && path->slots[0])
3419 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3420 ret = btrfs_next_leaf(root, path);
3422 fprintf(stderr, "Error going to next leaf "
3429 leaf = path->nodes[0];
3431 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3432 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
3438 offset = key.offset;
3439 } else if (key.offset != offset + num_bytes) {
3440 ret = check_extent_exists(root, offset, num_bytes);
3442 fprintf(stderr, "Csum exists for %Lu-%Lu but "
3443 "there is no extent record\n",
3444 offset, offset+num_bytes);
3447 offset = key.offset;
3451 num_bytes += (btrfs_item_size_nr(leaf, path->slots[0]) /
3452 csum_size) * root->sectorsize;
3456 btrfs_free_path(path);
3460 static int run_next_block(struct btrfs_root *root,
3461 struct block_info *bits,
3464 struct cache_tree *pending,
3465 struct cache_tree *seen,
3466 struct cache_tree *reada,
3467 struct cache_tree *nodes,
3468 struct cache_tree *extent_cache,
3469 struct cache_tree *chunk_cache,
3470 struct rb_root *dev_cache,
3471 struct block_group_tree *block_group_cache,
3472 struct device_extent_tree *dev_extent_cache)
3474 struct extent_buffer *buf;
3484 struct btrfs_key key;
3485 struct cache_extent *cache;
3488 nritems = pick_next_pending(pending, reada, nodes, *last, bits,
3489 bits_nr, &reada_bits);
3494 for(i = 0; i < nritems; i++) {
3495 ret = add_cache_extent(reada, bits[i].start,
3500 /* fixme, get the parent transid */
3501 readahead_tree_block(root, bits[i].start,
3505 *last = bits[0].start;
3506 bytenr = bits[0].start;
3507 size = bits[0].size;
3509 cache = lookup_cache_extent(pending, bytenr, size);
3511 remove_cache_extent(pending, cache);
3514 cache = lookup_cache_extent(reada, bytenr, size);
3516 remove_cache_extent(reada, cache);
3519 cache = lookup_cache_extent(nodes, bytenr, size);
3521 remove_cache_extent(nodes, cache);
3525 /* fixme, get the real parent transid */
3526 buf = read_tree_block(root, bytenr, size, 0);
3527 if (!extent_buffer_uptodate(buf)) {
3528 record_bad_block_io(root->fs_info,
3529 extent_cache, bytenr, size);
3533 nritems = btrfs_header_nritems(buf);
3535 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
3536 btrfs_header_level(buf), 1, NULL,
3539 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
3541 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
3546 owner = btrfs_header_owner(buf);
3549 ret = check_block(root, extent_cache, buf, flags);
3553 if (btrfs_is_leaf(buf)) {
3554 btree_space_waste += btrfs_leaf_free_space(root, buf);
3555 for (i = 0; i < nritems; i++) {
3556 struct btrfs_file_extent_item *fi;
3557 btrfs_item_key_to_cpu(buf, &key, i);
3558 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
3559 process_extent_item(root, extent_cache, buf,
3563 if (key.type == BTRFS_METADATA_ITEM_KEY) {
3564 process_extent_item(root, extent_cache, buf,
3568 if (key.type == BTRFS_EXTENT_CSUM_KEY) {
3570 btrfs_item_size_nr(buf, i);
3573 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3574 process_chunk_item(chunk_cache, &key, buf, i);
3577 if (key.type == BTRFS_DEV_ITEM_KEY) {
3578 process_device_item(dev_cache, &key, buf, i);
3581 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
3582 process_block_group_item(block_group_cache,
3586 if (key.type == BTRFS_DEV_EXTENT_KEY) {
3587 process_device_extent_item(dev_extent_cache,
3592 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
3593 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3594 process_extent_ref_v0(extent_cache, buf, i);
3601 if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
3602 add_tree_backref(extent_cache, key.objectid, 0,
3606 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
3607 add_tree_backref(extent_cache, key.objectid,
3611 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3612 struct btrfs_extent_data_ref *ref;
3613 ref = btrfs_item_ptr(buf, i,
3614 struct btrfs_extent_data_ref);
3615 add_data_backref(extent_cache,
3617 btrfs_extent_data_ref_root(buf, ref),
3618 btrfs_extent_data_ref_objectid(buf,
3620 btrfs_extent_data_ref_offset(buf, ref),
3621 btrfs_extent_data_ref_count(buf, ref),
3622 0, root->sectorsize);
3625 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3626 struct btrfs_shared_data_ref *ref;
3627 ref = btrfs_item_ptr(buf, i,
3628 struct btrfs_shared_data_ref);
3629 add_data_backref(extent_cache,
3630 key.objectid, key.offset, 0, 0, 0,
3631 btrfs_shared_data_ref_count(buf, ref),
3632 0, root->sectorsize);
3635 if (key.type != BTRFS_EXTENT_DATA_KEY)
3637 fi = btrfs_item_ptr(buf, i,
3638 struct btrfs_file_extent_item);
3639 if (btrfs_file_extent_type(buf, fi) ==
3640 BTRFS_FILE_EXTENT_INLINE)
3642 if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
3645 data_bytes_allocated +=
3646 btrfs_file_extent_disk_num_bytes(buf, fi);
3647 if (data_bytes_allocated < root->sectorsize) {
3650 data_bytes_referenced +=
3651 btrfs_file_extent_num_bytes(buf, fi);
3652 add_data_backref(extent_cache,
3653 btrfs_file_extent_disk_bytenr(buf, fi),
3654 parent, owner, key.objectid, key.offset -
3655 btrfs_file_extent_offset(buf, fi), 1, 1,
3656 btrfs_file_extent_disk_num_bytes(buf, fi));
3661 struct btrfs_key first_key;
3663 first_key.objectid = 0;
3666 btrfs_item_key_to_cpu(buf, &first_key, 0);
3667 level = btrfs_header_level(buf);
3668 for (i = 0; i < nritems; i++) {
3669 ptr = btrfs_node_blockptr(buf, i);
3670 size = btrfs_level_size(root, level - 1);
3671 btrfs_node_key_to_cpu(buf, &key, i);
3672 ret = add_extent_rec(extent_cache, &key,
3673 ptr, size, 0, 0, 1, 0, 1, 0,
3677 add_tree_backref(extent_cache, ptr, parent, owner, 1);
3680 add_pending(nodes, seen, ptr, size);
3682 add_pending(pending, seen, ptr, size);
3685 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
3686 nritems) * sizeof(struct btrfs_key_ptr);
3688 total_btree_bytes += buf->len;
3689 if (fs_root_objectid(btrfs_header_owner(buf)))
3690 total_fs_tree_bytes += buf->len;
3691 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
3692 total_extent_tree_bytes += buf->len;
3693 if (!found_old_backref &&
3694 btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
3695 btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
3696 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
3697 found_old_backref = 1;
3699 free_extent_buffer(buf);
3703 static int add_root_to_pending(struct extent_buffer *buf,
3704 struct cache_tree *extent_cache,
3705 struct cache_tree *pending,
3706 struct cache_tree *seen,
3707 struct cache_tree *nodes,
3708 struct btrfs_key *root_key)
3710 if (btrfs_header_level(buf) > 0)
3711 add_pending(nodes, seen, buf->start, buf->len);
3713 add_pending(pending, seen, buf->start, buf->len);
3714 add_extent_rec(extent_cache, NULL, buf->start, buf->len,
3715 0, 1, 1, 0, 1, 0, buf->len);
3717 if (root_key->objectid == BTRFS_TREE_RELOC_OBJECTID ||
3718 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
3719 add_tree_backref(extent_cache, buf->start, buf->start,
3722 add_tree_backref(extent_cache, buf->start, 0,
3723 root_key->objectid, 1);
3727 /* as we fix the tree, we might be deleting blocks that
3728 * we're tracking for repair. This hook makes sure we
3729 * remove any backrefs for blocks as we are fixing them.
3731 static int free_extent_hook(struct btrfs_trans_handle *trans,
3732 struct btrfs_root *root,
3733 u64 bytenr, u64 num_bytes, u64 parent,
3734 u64 root_objectid, u64 owner, u64 offset,
3737 struct extent_record *rec;
3738 struct cache_extent *cache;
3740 struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
3742 is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
3743 cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
3747 rec = container_of(cache, struct extent_record, cache);
3749 struct data_backref *back;
3750 back = find_data_backref(rec, parent, root_objectid, owner,
3751 offset, 1, bytenr, num_bytes);
3754 if (back->node.found_ref) {
3755 back->found_ref -= refs_to_drop;
3757 rec->refs -= refs_to_drop;
3759 if (back->node.found_extent_tree) {
3760 back->num_refs -= refs_to_drop;
3761 if (rec->extent_item_refs)
3762 rec->extent_item_refs -= refs_to_drop;
3764 if (back->found_ref == 0)
3765 back->node.found_ref = 0;
3766 if (back->num_refs == 0)
3767 back->node.found_extent_tree = 0;
3769 if (!back->node.found_extent_tree && back->node.found_ref) {
3770 list_del(&back->node.list);
3774 struct tree_backref *back;
3775 back = find_tree_backref(rec, parent, root_objectid);
3778 if (back->node.found_ref) {
3781 back->node.found_ref = 0;
3783 if (back->node.found_extent_tree) {
3784 if (rec->extent_item_refs)
3785 rec->extent_item_refs--;
3786 back->node.found_extent_tree = 0;
3788 if (!back->node.found_extent_tree && back->node.found_ref) {
3789 list_del(&back->node.list);
3793 maybe_free_extent_rec(extent_cache, rec);
3798 static int delete_extent_records(struct btrfs_trans_handle *trans,
3799 struct btrfs_root *root,
3800 struct btrfs_path *path,
3801 u64 bytenr, u64 new_len)
3803 struct btrfs_key key;
3804 struct btrfs_key found_key;
3805 struct extent_buffer *leaf;
3810 key.objectid = bytenr;
3812 key.offset = (u64)-1;
3815 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
3822 if (path->slots[0] == 0)
3828 leaf = path->nodes[0];
3829 slot = path->slots[0];
3831 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3832 if (found_key.objectid != bytenr)
3835 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
3836 found_key.type != BTRFS_METADATA_ITEM_KEY &&
3837 found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
3838 found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
3839 found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
3840 found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
3841 found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
3842 btrfs_release_path(path);
3843 if (found_key.type == 0) {
3844 if (found_key.offset == 0)
3846 key.offset = found_key.offset - 1;
3847 key.type = found_key.type;
3849 key.type = found_key.type - 1;
3850 key.offset = (u64)-1;
3854 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
3855 found_key.objectid, found_key.type, found_key.offset);
3857 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
3860 btrfs_release_path(path);
3862 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
3863 found_key.type == BTRFS_METADATA_ITEM_KEY) {
3864 u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
3865 found_key.offset : root->leafsize;
3867 ret = btrfs_update_block_group(trans, root, bytenr,
3874 btrfs_release_path(path);
3879 * for a single backref, this will allocate a new extent
3880 * and add the backref to it.
3882 static int record_extent(struct btrfs_trans_handle *trans,
3883 struct btrfs_fs_info *info,
3884 struct btrfs_path *path,
3885 struct extent_record *rec,
3886 struct extent_backref *back,
3887 int allocated, u64 flags)
3890 struct btrfs_root *extent_root = info->extent_root;
3891 struct extent_buffer *leaf;
3892 struct btrfs_key ins_key;
3893 struct btrfs_extent_item *ei;
3894 struct tree_backref *tback;
3895 struct data_backref *dback;
3896 struct btrfs_tree_block_info *bi;
3899 rec->max_size = max_t(u64, rec->max_size,
3900 info->extent_root->leafsize);
3903 u32 item_size = sizeof(*ei);
3906 item_size += sizeof(*bi);
3908 ins_key.objectid = rec->start;
3909 ins_key.offset = rec->max_size;
3910 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
3912 ret = btrfs_insert_empty_item(trans, extent_root, path,
3913 &ins_key, item_size);
3917 leaf = path->nodes[0];
3918 ei = btrfs_item_ptr(leaf, path->slots[0],
3919 struct btrfs_extent_item);
3921 btrfs_set_extent_refs(leaf, ei, 0);
3922 btrfs_set_extent_generation(leaf, ei, rec->generation);
3924 if (back->is_data) {
3925 btrfs_set_extent_flags(leaf, ei,
3926 BTRFS_EXTENT_FLAG_DATA);
3928 struct btrfs_disk_key copy_key;;
3930 tback = (struct tree_backref *)back;
3931 bi = (struct btrfs_tree_block_info *)(ei + 1);
3932 memset_extent_buffer(leaf, 0, (unsigned long)bi,
3934 memset(©_key, 0, sizeof(copy_key));
3936 copy_key.objectid = le64_to_cpu(rec->info_objectid);
3937 btrfs_set_tree_block_level(leaf, bi, rec->info_level);
3938 btrfs_set_tree_block_key(leaf, bi, ©_key);
3940 btrfs_set_extent_flags(leaf, ei,
3941 BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
3944 btrfs_mark_buffer_dirty(leaf);
3945 ret = btrfs_update_block_group(trans, extent_root, rec->start,
3946 rec->max_size, 1, 0);
3949 btrfs_release_path(path);
3952 if (back->is_data) {
3956 dback = (struct data_backref *)back;
3957 if (back->full_backref)
3958 parent = dback->parent;
3962 for (i = 0; i < dback->found_ref; i++) {
3963 /* if parent != 0, we're doing a full backref
3964 * passing BTRFS_FIRST_FREE_OBJECTID as the owner
3965 * just makes the backref allocator create a data
3968 ret = btrfs_inc_extent_ref(trans, info->extent_root,
3969 rec->start, rec->max_size,
3973 BTRFS_FIRST_FREE_OBJECTID :
3979 fprintf(stderr, "adding new data backref"
3980 " on %llu %s %llu owner %llu"
3981 " offset %llu found %d\n",
3982 (unsigned long long)rec->start,
3983 back->full_backref ?
3985 back->full_backref ?
3986 (unsigned long long)parent :
3987 (unsigned long long)dback->root,
3988 (unsigned long long)dback->owner,
3989 (unsigned long long)dback->offset,
3994 tback = (struct tree_backref *)back;
3995 if (back->full_backref)
3996 parent = tback->parent;
4000 ret = btrfs_inc_extent_ref(trans, info->extent_root,
4001 rec->start, rec->max_size,
4002 parent, tback->root, 0, 0);
4003 fprintf(stderr, "adding new tree backref on "
4004 "start %llu len %llu parent %llu root %llu\n",
4005 rec->start, rec->max_size, tback->parent, tback->root);
4010 btrfs_release_path(path);
4014 struct extent_entry {
4018 struct list_head list;
4021 static struct extent_entry *find_entry(struct list_head *entries,
4022 u64 bytenr, u64 bytes)
4024 struct extent_entry *entry = NULL;
4026 list_for_each_entry(entry, entries, list) {
4027 if (entry->bytenr == bytenr && entry->bytes == bytes)
4034 static struct extent_entry *find_most_right_entry(struct list_head *entries)
4036 struct extent_entry *entry, *best = NULL, *prev = NULL;
4038 list_for_each_entry(entry, entries, list) {
4045 * If our current entry == best then we can't be sure our best
4046 * is really the best, so we need to keep searching.
4048 if (best && best->count == entry->count) {
4054 /* Prev == entry, not good enough, have to keep searching */
4055 if (prev->count == entry->count)
4059 best = (prev->count > entry->count) ? prev : entry;
4060 else if (best->count < entry->count)
4068 static int repair_ref(struct btrfs_trans_handle *trans,
4069 struct btrfs_fs_info *info, struct btrfs_path *path,
4070 struct data_backref *dback, struct extent_entry *entry)
4072 struct btrfs_root *root;
4073 struct btrfs_file_extent_item *fi;
4074 struct extent_buffer *leaf;
4075 struct btrfs_key key;
4079 key.objectid = dback->root;
4080 key.type = BTRFS_ROOT_ITEM_KEY;
4081 key.offset = (u64)-1;
4082 root = btrfs_read_fs_root(info, &key);
4084 fprintf(stderr, "Couldn't find root for our ref\n");
4089 * The backref points to the original offset of the extent if it was
4090 * split, so we need to search down to the offset we have and then walk
4091 * forward until we find the backref we're looking for.
4093 key.objectid = dback->owner;
4094 key.type = BTRFS_EXTENT_DATA_KEY;
4095 key.offset = dback->offset;
4096 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4098 fprintf(stderr, "Error looking up ref %d\n", ret);
4103 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4104 ret = btrfs_next_leaf(root, path);
4106 fprintf(stderr, "Couldn't find our ref, next\n");
4110 leaf = path->nodes[0];
4111 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4112 if (key.objectid != dback->owner ||
4113 key.type != BTRFS_EXTENT_DATA_KEY) {
4114 fprintf(stderr, "Couldn't find our ref, search\n");
4117 fi = btrfs_item_ptr(leaf, path->slots[0],
4118 struct btrfs_file_extent_item);
4119 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
4120 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
4122 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
4127 btrfs_release_path(path);
4130 * Have to make sure that this root gets updated when we commit the
4133 root->track_dirty = 1;
4134 if (root->last_trans != trans->transid) {
4135 root->last_trans = trans->transid;
4136 root->commit_root = root->node;
4137 extent_buffer_get(root->node);
4141 * Ok we have the key of the file extent we want to fix, now we can cow
4142 * down to the thing and fix it.
4144 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4146 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
4147 key.objectid, key.type, key.offset, ret);
4151 fprintf(stderr, "Well that's odd, we just found this key "
4152 "[%Lu, %u, %Lu]\n", key.objectid, key.type,
4156 leaf = path->nodes[0];
4157 fi = btrfs_item_ptr(leaf, path->slots[0],
4158 struct btrfs_file_extent_item);
4160 if (btrfs_file_extent_compression(leaf, fi) &&
4161 dback->disk_bytenr != entry->bytenr) {
4162 fprintf(stderr, "Ref doesn't match the record start and is "
4163 "compressed, please take a btrfs-image of this file "
4164 "system and send it to a btrfs developer so they can "
4165 "complete this functionality for bytenr %Lu\n",
4166 dback->disk_bytenr);
4170 if (dback->disk_bytenr > entry->bytenr) {
4171 u64 off_diff, offset;
4173 off_diff = dback->disk_bytenr - entry->bytenr;
4174 offset = btrfs_file_extent_offset(leaf, fi);
4175 if (dback->disk_bytenr + offset +
4176 btrfs_file_extent_num_bytes(leaf, fi) >
4177 entry->bytenr + entry->bytes) {
4178 fprintf(stderr, "Ref is past the entry end, please "
4179 "take a btrfs-image of this file system and "
4180 "send it to a btrfs developer, ref %Lu\n",
4181 dback->disk_bytenr);
4185 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
4186 btrfs_set_file_extent_offset(leaf, fi, offset);
4187 } else if (dback->disk_bytenr < entry->bytenr) {
4190 offset = btrfs_file_extent_offset(leaf, fi);
4191 if (dback->disk_bytenr + offset < entry->bytenr) {
4192 fprintf(stderr, "Ref is before the entry start, please"
4193 " take a btrfs-image of this file system and "
4194 "send it to a btrfs developer, ref %Lu\n",
4195 dback->disk_bytenr);
4199 offset += dback->disk_bytenr;
4200 offset -= entry->bytenr;
4201 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
4202 btrfs_set_file_extent_offset(leaf, fi, offset);
4205 btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
4208 * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
4209 * only do this if we aren't using compression, otherwise it's a
4212 if (!btrfs_file_extent_compression(leaf, fi))
4213 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
4215 printf("ram bytes may be wrong?\n");
4216 btrfs_mark_buffer_dirty(leaf);
4217 btrfs_release_path(path);
4221 static int verify_backrefs(struct btrfs_trans_handle *trans,
4222 struct btrfs_fs_info *info, struct btrfs_path *path,
4223 struct extent_record *rec)
4225 struct extent_backref *back;
4226 struct data_backref *dback;
4227 struct extent_entry *entry, *best = NULL;
4233 * Metadata is easy and the backrefs should always agree on bytenr and
4234 * size, if not we've got bigger issues.
4239 list_for_each_entry(back, &rec->backrefs, list) {
4240 dback = (struct data_backref *)back;
4242 * We only pay attention to backrefs that we found a real
4245 if (dback->found_ref == 0)
4247 if (back->full_backref)
4251 * For now we only catch when the bytes don't match, not the
4252 * bytenr. We can easily do this at the same time, but I want
4253 * to have a fs image to test on before we just add repair
4254 * functionality willy-nilly so we know we won't screw up the
4258 entry = find_entry(&entries, dback->disk_bytenr,
4261 entry = malloc(sizeof(struct extent_entry));
4266 memset(entry, 0, sizeof(*entry));
4267 entry->bytenr = dback->disk_bytenr;
4268 entry->bytes = dback->bytes;
4269 list_add_tail(&entry->list, &entries);
4275 /* Yay all the backrefs agree, carry on good sir */
4276 if (nr_entries <= 1)
4279 fprintf(stderr, "attempting to repair backref discrepency for bytenr "
4280 "%Lu\n", rec->start);
4283 * First we want to see if the backrefs can agree amongst themselves who
4284 * is right, so figure out which one of the entries has the highest
4287 best = find_most_right_entry(&entries);
4290 * Ok so we may have an even split between what the backrefs think, so
4291 * this is where we use the extent ref to see what it thinks.
4294 entry = find_entry(&entries, rec->start, rec->nr);
4296 fprintf(stderr, "Backrefs don't agree with eachother "
4297 "and extent record doesn't agree with anybody,"
4298 " so we can't fix bytenr %Lu bytes %Lu\n",
4299 rec->start, rec->nr);
4304 best = find_most_right_entry(&entries);
4306 fprintf(stderr, "Backrefs and extent record evenly "
4307 "split on who is right, this is going to "
4308 "require user input to fix bytenr %Lu bytes "
4309 "%Lu\n", rec->start, rec->nr);
4316 * I don't think this can happen currently as we'll abort() if we catch
4317 * this case higher up, but in case somebody removes that we still can't
4318 * deal with it properly here yet, so just bail out of that's the case.
4320 if (best->bytenr != rec->start) {
4321 fprintf(stderr, "Extent start and backref starts don't match, "
4322 "please use btrfs-image on this file system and send "
4323 "it to a btrfs developer so they can make fsck fix "
4324 "this particular case. bytenr is %Lu, bytes is %Lu\n",
4325 rec->start, rec->nr);
4331 * Ok great we all agreed on an extent record, let's go find the real
4332 * references and fix up the ones that don't match.
4334 list_for_each_entry(back, &rec->backrefs, list) {
4335 dback = (struct data_backref *)back;
4338 * Still ignoring backrefs that don't have a real ref attached
4341 if (dback->found_ref == 0)
4343 if (back->full_backref)
4346 if (dback->bytes == best->bytes &&
4347 dback->disk_bytenr == best->bytenr)
4350 ret = repair_ref(trans, info, path, dback, best);
4356 * Ok we messed with the actual refs, which means we need to drop our
4357 * entire cache and go back and rescan. I know this is a huge pain and
4358 * adds a lot of extra work, but it's the only way to be safe. Once all
4359 * the backrefs agree we may not need to do anything to the extent
4364 while (!list_empty(&entries)) {
4365 entry = list_entry(entries.next, struct extent_entry, list);
4366 list_del_init(&entry->list);
4372 static int process_duplicates(struct btrfs_root *root,
4373 struct cache_tree *extent_cache,
4374 struct extent_record *rec)
4376 struct extent_record *good, *tmp;
4377 struct cache_extent *cache;
4381 * If we found a extent record for this extent then return, or if we
4382 * have more than one duplicate we are likely going to need to delete
4385 if (rec->found_rec || rec->num_duplicates > 1)
4388 /* Shouldn't happen but just in case */
4389 BUG_ON(!rec->num_duplicates);
4392 * So this happens if we end up with a backref that doesn't match the
4393 * actual extent entry. So either the backref is bad or the extent
4394 * entry is bad. Either way we want to have the extent_record actually
4395 * reflect what we found in the extent_tree, so we need to take the
4396 * duplicate out and use that as the extent_record since the only way we
4397 * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
4399 remove_cache_extent(extent_cache, &rec->cache);
4401 good = list_entry(rec->dups.next, struct extent_record, list);
4402 list_del_init(&good->list);
4403 INIT_LIST_HEAD(&good->backrefs);
4404 INIT_LIST_HEAD(&good->dups);
4405 good->cache.start = good->start;
4406 good->cache.size = good->nr;
4407 good->content_checked = 0;
4408 good->owner_ref_checked = 0;
4409 good->num_duplicates = 0;
4410 good->refs = rec->refs;
4411 list_splice_init(&rec->backrefs, &good->backrefs);
4413 cache = lookup_cache_extent(extent_cache, good->start,
4417 tmp = container_of(cache, struct extent_record, cache);
4420 * If we find another overlapping extent and it's found_rec is
4421 * set then it's a duplicate and we need to try and delete
4424 if (tmp->found_rec || tmp->num_duplicates > 0) {
4425 if (list_empty(&good->list))
4426 list_add_tail(&good->list,
4427 &duplicate_extents);
4428 good->num_duplicates += tmp->num_duplicates + 1;
4429 list_splice_init(&tmp->dups, &good->dups);
4430 list_del_init(&tmp->list);
4431 list_add_tail(&tmp->list, &good->dups);
4432 remove_cache_extent(extent_cache, &tmp->cache);
4437 * Ok we have another non extent item backed extent rec, so lets
4438 * just add it to this extent and carry on like we did above.
4440 good->refs += tmp->refs;
4441 list_splice_init(&tmp->backrefs, &good->backrefs);
4442 remove_cache_extent(extent_cache, &tmp->cache);
4445 ret = insert_cache_extent(extent_cache, &good->cache);
4448 return good->num_duplicates ? 0 : 1;
4451 static int delete_duplicate_records(struct btrfs_trans_handle *trans,
4452 struct btrfs_root *root,
4453 struct extent_record *rec)
4455 LIST_HEAD(delete_list);
4456 struct btrfs_path *path;
4457 struct extent_record *tmp, *good, *n;
4460 struct btrfs_key key;
4462 path = btrfs_alloc_path();
4469 /* Find the record that covers all of the duplicates. */
4470 list_for_each_entry(tmp, &rec->dups, list) {
4471 if (good->start < tmp->start)
4473 if (good->nr > tmp->nr)
4476 if (tmp->start + tmp->nr < good->start + good->nr) {
4477 fprintf(stderr, "Ok we have overlapping extents that "
4478 "aren't completely covered by eachother, this "
4479 "is going to require more careful thought. "
4480 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
4481 tmp->start, tmp->nr, good->start, good->nr);
4488 list_add_tail(&rec->list, &delete_list);
4490 list_for_each_entry_safe(tmp, n, &rec->dups, list) {
4493 list_move_tail(&tmp->list, &delete_list);
4496 root = root->fs_info->extent_root;
4497 list_for_each_entry(tmp, &delete_list, list) {
4498 if (tmp->found_rec == 0)
4500 key.objectid = tmp->start;
4501 key.type = BTRFS_EXTENT_ITEM_KEY;
4502 key.offset = tmp->nr;
4504 /* Shouldn't happen but just in case */
4505 if (tmp->metadata) {
4506 fprintf(stderr, "Well this shouldn't happen, extent "
4507 "record overlaps but is metadata? "
4508 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
4512 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4518 ret = btrfs_del_item(trans, root, path);
4521 btrfs_release_path(path);
4526 while (!list_empty(&delete_list)) {
4527 tmp = list_entry(delete_list.next, struct extent_record, list);
4528 list_del_init(&tmp->list);
4534 while (!list_empty(&rec->dups)) {
4535 tmp = list_entry(rec->dups.next, struct extent_record, list);
4536 list_del_init(&tmp->list);
4540 btrfs_free_path(path);
4542 if (!ret && !nr_del)
4543 rec->num_duplicates = 0;
4545 return ret ? ret : nr_del;
4549 * when an incorrect extent item is found, this will delete
4550 * all of the existing entries for it and recreate them
4551 * based on what the tree scan found.
4553 static int fixup_extent_refs(struct btrfs_trans_handle *trans,
4554 struct btrfs_fs_info *info,
4555 struct extent_record *rec)
4558 struct btrfs_path *path;
4559 struct list_head *cur = rec->backrefs.next;
4560 struct cache_extent *cache;
4561 struct extent_backref *back;
4565 /* remember our flags for recreating the extent */
4566 ret = btrfs_lookup_extent_info(NULL, info->extent_root, rec->start,
4567 rec->max_size, rec->metadata, NULL,
4570 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4572 path = btrfs_alloc_path();
4576 /* step one, make sure all of the backrefs agree */
4577 ret = verify_backrefs(trans, info, path, rec);
4581 /* step two, delete all the existing records */
4582 ret = delete_extent_records(trans, info->extent_root, path,
4583 rec->start, rec->max_size);
4588 /* was this block corrupt? If so, don't add references to it */
4589 cache = lookup_cache_extent(info->corrupt_blocks,
4590 rec->start, rec->max_size);
4596 /* step three, recreate all the refs we did find */
4597 while(cur != &rec->backrefs) {
4598 back = list_entry(cur, struct extent_backref, list);
4602 * if we didn't find any references, don't create a
4605 if (!back->found_ref)
4608 ret = record_extent(trans, info, path, rec, back, allocated, flags);
4615 btrfs_free_path(path);
4619 /* right now we only prune from the extent allocation tree */
4620 static int prune_one_block(struct btrfs_trans_handle *trans,
4621 struct btrfs_fs_info *info,
4622 struct btrfs_corrupt_block *corrupt)
4625 struct btrfs_path path;
4626 struct extent_buffer *eb;
4630 int level = corrupt->level + 1;
4632 btrfs_init_path(&path);
4634 /* we want to stop at the parent to our busted block */
4635 path.lowest_level = level;
4637 ret = btrfs_search_slot(trans, info->extent_root,
4638 &corrupt->key, &path, -1, 1);
4643 eb = path.nodes[level];
4650 * hopefully the search gave us the block we want to prune,
4651 * lets try that first
4653 slot = path.slots[level];
4654 found = btrfs_node_blockptr(eb, slot);
4655 if (found == corrupt->cache.start)
4658 nritems = btrfs_header_nritems(eb);
4660 /* the search failed, lets scan this node and hope we find it */
4661 for (slot = 0; slot < nritems; slot++) {
4662 found = btrfs_node_blockptr(eb, slot);
4663 if (found == corrupt->cache.start)
4667 * we couldn't find the bad block. TODO, search all the nodes for pointers
4670 if (eb == info->extent_root->node) {
4675 btrfs_release_path(&path);
4680 printk("deleting pointer to block %Lu\n", corrupt->cache.start);
4681 ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
4684 btrfs_release_path(&path);
4688 static int prune_corrupt_blocks(struct btrfs_trans_handle *trans,
4689 struct btrfs_fs_info *info)
4691 struct cache_extent *cache;
4692 struct btrfs_corrupt_block *corrupt;
4694 cache = search_cache_extent(info->corrupt_blocks, 0);
4698 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
4699 prune_one_block(trans, info, corrupt);
4700 cache = next_cache_extent(cache);
4705 static void free_corrupt_block(struct cache_extent *cache)
4707 struct btrfs_corrupt_block *corrupt;
4709 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
4713 FREE_EXTENT_CACHE_BASED_TREE(corrupt_blocks, free_corrupt_block);
4715 static int check_block_group(struct btrfs_trans_handle *trans,
4716 struct btrfs_fs_info *info,
4717 struct map_lookup *map,
4720 struct btrfs_key key;
4721 struct btrfs_path path;
4724 key.objectid = map->ce.start;
4725 key.offset = map->ce.size;
4726 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
4728 btrfs_init_path(&path);
4729 ret = btrfs_search_slot(NULL, info->extent_root,
4731 btrfs_release_path(&path);
4735 ret = btrfs_make_block_group(trans, info->extent_root, 0, map->type,
4736 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4737 key.objectid, key.offset);
4743 static int check_block_groups(struct btrfs_trans_handle *trans,
4744 struct btrfs_fs_info *info, int *reinit)
4746 struct cache_extent *ce;
4747 struct map_lookup *map;
4748 struct btrfs_mapping_tree *map_tree = &info->mapping_tree;
4750 /* this isn't quite working */
4753 ce = search_cache_extent(&map_tree->cache_tree, 0);
4757 map = container_of(ce, struct map_lookup, ce);
4758 check_block_group(trans, info, map, reinit);
4759 ce = next_cache_extent(ce);
4764 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
4766 struct btrfs_block_group_cache *cache;
4771 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
4772 &start, &end, EXTENT_DIRTY);
4775 clear_extent_dirty(&fs_info->free_space_cache, start, end,
4781 cache = btrfs_lookup_first_block_group(fs_info, start);
4786 start = cache->key.objectid + cache->key.offset;
4790 static int check_extent_refs(struct btrfs_trans_handle *trans,
4791 struct btrfs_root *root,
4792 struct cache_tree *extent_cache, int repair)
4794 struct extent_record *rec;
4795 struct cache_extent *cache;
4804 * if we're doing a repair, we have to make sure
4805 * we don't allocate from the problem extents.
4806 * In the worst case, this will be all the
4809 cache = search_cache_extent(extent_cache, 0);
4811 rec = container_of(cache, struct extent_record, cache);
4812 btrfs_pin_extent(root->fs_info,
4813 rec->start, rec->max_size);
4814 cache = next_cache_extent(cache);
4817 /* pin down all the corrupted blocks too */
4818 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
4820 btrfs_pin_extent(root->fs_info,
4821 cache->start, cache->size);
4822 cache = next_cache_extent(cache);
4824 prune_corrupt_blocks(trans, root->fs_info);
4825 check_block_groups(trans, root->fs_info, &reinit);
4827 btrfs_read_block_groups(root->fs_info->extent_root);
4828 reset_cached_block_groups(root->fs_info);
4832 * We need to delete any duplicate entries we find first otherwise we
4833 * could mess up the extent tree when we have backrefs that actually
4834 * belong to a different extent item and not the weird duplicate one.
4836 while (repair && !list_empty(&duplicate_extents)) {
4837 rec = list_entry(duplicate_extents.next, struct extent_record,
4839 list_del_init(&rec->list);
4841 /* Sometimes we can find a backref before we find an actual
4842 * extent, so we need to process it a little bit to see if there
4843 * truly are multiple EXTENT_ITEM_KEY's for the same range, or
4844 * if this is a backref screwup. If we need to delete stuff
4845 * process_duplicates() will return 0, otherwise it will return
4848 if (process_duplicates(root, extent_cache, rec))
4850 ret = delete_duplicate_records(trans, root, rec);
4854 * delete_duplicate_records will return the number of entries
4855 * deleted, so if it's greater than 0 then we know we actually
4856 * did something and we need to remove.
4867 cache = search_cache_extent(extent_cache, 0);
4870 rec = container_of(cache, struct extent_record, cache);
4871 if (rec->num_duplicates) {
4872 fprintf(stderr, "extent item %llu has multiple extent "
4873 "items\n", (unsigned long long)rec->start);
4877 if (rec->refs != rec->extent_item_refs) {
4878 fprintf(stderr, "ref mismatch on [%llu %llu] ",
4879 (unsigned long long)rec->start,
4880 (unsigned long long)rec->nr);
4881 fprintf(stderr, "extent item %llu, found %llu\n",
4882 (unsigned long long)rec->extent_item_refs,
4883 (unsigned long long)rec->refs);
4884 if (!fixed && repair) {
4885 ret = fixup_extent_refs(trans, root->fs_info, rec);
4893 if (all_backpointers_checked(rec, 1)) {
4894 fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
4895 (unsigned long long)rec->start,
4896 (unsigned long long)rec->nr);
4898 if (!fixed && repair) {
4899 ret = fixup_extent_refs(trans, root->fs_info, rec);
4907 if (!rec->owner_ref_checked) {
4908 fprintf(stderr, "owner ref check failed [%llu %llu]\n",
4909 (unsigned long long)rec->start,
4910 (unsigned long long)rec->nr);
4911 if (!fixed && repair) {
4912 ret = fixup_extent_refs(trans, root->fs_info, rec);
4920 remove_cache_extent(extent_cache, cache);
4921 free_all_extent_backrefs(rec);
4926 if (ret && ret != -EAGAIN) {
4927 fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
4930 btrfs_fix_block_accounting(trans, root);
4933 fprintf(stderr, "repaired damaged extent references\n");
4939 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
4943 if (type & BTRFS_BLOCK_GROUP_RAID0) {
4944 stripe_size = length;
4945 stripe_size /= num_stripes;
4946 } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
4947 stripe_size = length * 2;
4948 stripe_size /= num_stripes;
4949 } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
4950 stripe_size = length;
4951 stripe_size /= (num_stripes - 1);
4952 } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
4953 stripe_size = length;
4954 stripe_size /= (num_stripes - 2);
4956 stripe_size = length;
4961 static int check_chunk_refs(struct chunk_record *chunk_rec,
4962 struct block_group_tree *block_group_cache,
4963 struct device_extent_tree *dev_extent_cache,
4966 struct cache_extent *block_group_item;
4967 struct block_group_record *block_group_rec;
4968 struct cache_extent *dev_extent_item;
4969 struct device_extent_record *dev_extent_rec;
4976 block_group_item = lookup_cache_extent(&block_group_cache->tree,
4979 if (block_group_item) {
4980 block_group_rec = container_of(block_group_item,
4981 struct block_group_record,
4983 if (chunk_rec->length != block_group_rec->offset ||
4984 chunk_rec->offset != block_group_rec->objectid ||
4985 chunk_rec->type_flags != block_group_rec->flags) {
4988 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
4989 chunk_rec->objectid,
4994 chunk_rec->type_flags,
4995 block_group_rec->objectid,
4996 block_group_rec->type,
4997 block_group_rec->offset,
4998 block_group_rec->offset,
4999 block_group_rec->objectid,
5000 block_group_rec->flags);
5003 list_del_init(&block_group_rec->list);
5004 chunk_rec->bg_rec = block_group_rec;
5009 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
5010 chunk_rec->objectid,
5015 chunk_rec->type_flags);
5019 length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
5020 chunk_rec->num_stripes);
5021 for (i = 0; i < chunk_rec->num_stripes; ++i) {
5022 devid = chunk_rec->stripes[i].devid;
5023 offset = chunk_rec->stripes[i].offset;
5024 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
5025 devid, offset, length);
5026 if (dev_extent_item) {
5027 dev_extent_rec = container_of(dev_extent_item,
5028 struct device_extent_record,
5030 if (dev_extent_rec->objectid != devid ||
5031 dev_extent_rec->offset != offset ||
5032 dev_extent_rec->chunk_offset != chunk_rec->offset ||
5033 dev_extent_rec->length != length) {
5036 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
5037 chunk_rec->objectid,
5040 chunk_rec->stripes[i].devid,
5041 chunk_rec->stripes[i].offset,
5042 dev_extent_rec->objectid,
5043 dev_extent_rec->offset,
5044 dev_extent_rec->length);
5047 list_move(&dev_extent_rec->chunk_list,
5048 &chunk_rec->dextents);
5053 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
5054 chunk_rec->objectid,
5057 chunk_rec->stripes[i].devid,
5058 chunk_rec->stripes[i].offset);
5065 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
5066 int check_chunks(struct cache_tree *chunk_cache,
5067 struct block_group_tree *block_group_cache,
5068 struct device_extent_tree *dev_extent_cache,
5069 struct list_head *good, struct list_head *bad, int silent)
5071 struct cache_extent *chunk_item;
5072 struct chunk_record *chunk_rec;
5073 struct block_group_record *bg_rec;
5074 struct device_extent_record *dext_rec;
5078 chunk_item = first_cache_extent(chunk_cache);
5079 while (chunk_item) {
5080 chunk_rec = container_of(chunk_item, struct chunk_record,
5082 err = check_chunk_refs(chunk_rec, block_group_cache,
5083 dev_extent_cache, silent);
5087 list_add_tail(&chunk_rec->list, bad);
5090 list_add_tail(&chunk_rec->list, good);
5093 chunk_item = next_cache_extent(chunk_item);
5096 list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
5099 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
5107 list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
5111 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
5122 static int check_device_used(struct device_record *dev_rec,
5123 struct device_extent_tree *dext_cache)
5125 struct cache_extent *cache;
5126 struct device_extent_record *dev_extent_rec;
5129 cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
5131 dev_extent_rec = container_of(cache,
5132 struct device_extent_record,
5134 if (dev_extent_rec->objectid != dev_rec->devid)
5137 list_del(&dev_extent_rec->device_list);
5138 total_byte += dev_extent_rec->length;
5139 cache = next_cache_extent(cache);
5142 if (total_byte != dev_rec->byte_used) {
5144 "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
5145 total_byte, dev_rec->byte_used, dev_rec->objectid,
5146 dev_rec->type, dev_rec->offset);
5153 /* check btrfs_dev_item -> btrfs_dev_extent */
5154 static int check_devices(struct rb_root *dev_cache,
5155 struct device_extent_tree *dev_extent_cache)
5157 struct rb_node *dev_node;
5158 struct device_record *dev_rec;
5159 struct device_extent_record *dext_rec;
5163 dev_node = rb_first(dev_cache);
5165 dev_rec = container_of(dev_node, struct device_record, node);
5166 err = check_device_used(dev_rec, dev_extent_cache);
5170 dev_node = rb_next(dev_node);
5172 list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
5175 "Device extent[%llu, %llu, %llu] didn't find its device.\n",
5176 dext_rec->objectid, dext_rec->offset, dext_rec->length);
5183 static int check_chunks_and_extents(struct btrfs_root *root, int repair)
5185 struct rb_root dev_cache;
5186 struct cache_tree chunk_cache;
5187 struct block_group_tree block_group_cache;
5188 struct device_extent_tree dev_extent_cache;
5189 struct cache_tree extent_cache;
5190 struct cache_tree seen;
5191 struct cache_tree pending;
5192 struct cache_tree reada;
5193 struct cache_tree nodes;
5194 struct cache_tree corrupt_blocks;
5195 struct btrfs_path path;
5196 struct btrfs_key key;
5197 struct btrfs_key found_key;
5200 struct block_info *bits;
5202 struct extent_buffer *leaf;
5203 struct btrfs_trans_handle *trans = NULL;
5205 struct btrfs_root_item ri;
5207 dev_cache = RB_ROOT;
5208 cache_tree_init(&chunk_cache);
5209 block_group_tree_init(&block_group_cache);
5210 device_extent_tree_init(&dev_extent_cache);
5212 cache_tree_init(&extent_cache);
5213 cache_tree_init(&seen);
5214 cache_tree_init(&pending);
5215 cache_tree_init(&nodes);
5216 cache_tree_init(&reada);
5217 cache_tree_init(&corrupt_blocks);
5220 trans = btrfs_start_transaction(root, 1);
5221 if (IS_ERR(trans)) {
5222 fprintf(stderr, "Error starting transaction\n");
5223 return PTR_ERR(trans);
5225 root->fs_info->fsck_extent_cache = &extent_cache;
5226 root->fs_info->free_extent_hook = free_extent_hook;
5227 root->fs_info->corrupt_blocks = &corrupt_blocks;
5231 bits = malloc(bits_nr * sizeof(struct block_info));
5238 add_root_to_pending(root->fs_info->tree_root->node,
5239 &extent_cache, &pending, &seen, &nodes,
5240 &root->fs_info->tree_root->root_key);
5242 add_root_to_pending(root->fs_info->chunk_root->node,
5243 &extent_cache, &pending, &seen, &nodes,
5244 &root->fs_info->chunk_root->root_key);
5246 btrfs_init_path(&path);
5249 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
5250 ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
5254 leaf = path.nodes[0];
5255 slot = path.slots[0];
5256 if (slot >= btrfs_header_nritems(path.nodes[0])) {
5257 ret = btrfs_next_leaf(root, &path);
5260 leaf = path.nodes[0];
5261 slot = path.slots[0];
5263 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
5264 if (btrfs_key_type(&found_key) == BTRFS_ROOT_ITEM_KEY) {
5265 unsigned long offset;
5266 struct extent_buffer *buf;
5268 offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
5269 read_extent_buffer(leaf, &ri, offset, sizeof(ri));
5270 buf = read_tree_block(root->fs_info->tree_root,
5271 btrfs_root_bytenr(&ri),
5272 btrfs_level_size(root,
5273 btrfs_root_level(&ri)), 0);
5274 add_root_to_pending(buf, &extent_cache, &pending,
5275 &seen, &nodes, &found_key);
5276 free_extent_buffer(buf);
5280 btrfs_release_path(&path);
5282 ret = run_next_block(root, bits, bits_nr, &last, &pending,
5283 &seen, &reada, &nodes, &extent_cache,
5284 &chunk_cache, &dev_cache,
5285 &block_group_cache, &dev_extent_cache);
5290 ret = check_extent_refs(trans, root, &extent_cache, repair);
5291 if (ret == -EAGAIN) {
5292 ret = btrfs_commit_transaction(trans, root);
5296 trans = btrfs_start_transaction(root, 1);
5297 if (IS_ERR(trans)) {
5298 ret = PTR_ERR(trans);
5302 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
5303 free_extent_cache_tree(&seen);
5304 free_extent_cache_tree(&pending);
5305 free_extent_cache_tree(&reada);
5306 free_extent_cache_tree(&nodes);
5307 free_extent_record_cache(root->fs_info, &extent_cache);
5311 err = check_chunks(&chunk_cache, &block_group_cache,
5312 &dev_extent_cache, NULL, NULL, 0);
5316 err = check_devices(&dev_cache, &dev_extent_cache);
5321 err = btrfs_commit_transaction(trans, root);
5327 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
5328 root->fs_info->fsck_extent_cache = NULL;
5329 root->fs_info->free_extent_hook = NULL;
5330 root->fs_info->corrupt_blocks = NULL;
5333 free_chunk_cache_tree(&chunk_cache);
5334 free_device_cache_tree(&dev_cache);
5335 free_block_group_tree(&block_group_cache);
5336 free_device_extent_tree(&dev_extent_cache);
5340 static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
5341 struct btrfs_root *root, int overwrite)
5343 struct extent_buffer *c;
5344 struct extent_buffer *old = root->node;
5346 struct btrfs_disk_key disk_key = {0,0,0};
5352 extent_buffer_get(c);
5355 c = btrfs_alloc_free_block(trans, root,
5356 btrfs_level_size(root, 0),
5357 root->root_key.objectid,
5358 &disk_key, level, 0, 0);
5361 extent_buffer_get(c);
5364 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
5365 btrfs_set_header_level(c, level);
5366 btrfs_set_header_bytenr(c, c->start);
5367 btrfs_set_header_generation(c, trans->transid);
5368 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
5369 btrfs_set_header_owner(c, root->root_key.objectid);
5371 write_extent_buffer(c, root->fs_info->fsid,
5372 (unsigned long)btrfs_header_fsid(c),
5375 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
5376 (unsigned long)btrfs_header_chunk_tree_uuid(c),
5379 btrfs_mark_buffer_dirty(c);
5381 free_extent_buffer(old);
5383 add_root_to_dirty_list(root);
5387 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
5388 struct extent_buffer *eb, int tree_root)
5390 struct extent_buffer *tmp;
5391 struct btrfs_root_item *ri;
5392 struct btrfs_key key;
5395 int level = btrfs_header_level(eb);
5400 btrfs_pin_extent(fs_info, eb->start, eb->len);
5402 leafsize = btrfs_super_leafsize(fs_info->super_copy);
5403 nritems = btrfs_header_nritems(eb);
5404 for (i = 0; i < nritems; i++) {
5406 btrfs_item_key_to_cpu(eb, &key, i);
5407 if (key.type != BTRFS_ROOT_ITEM_KEY)
5409 /* Skip the extent root and reloc roots */
5410 if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
5411 key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
5412 key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
5414 ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
5415 bytenr = btrfs_disk_root_bytenr(eb, ri);
5418 * If at any point we start needing the real root we
5419 * will have to build a stump root for the root we are
5420 * in, but for now this doesn't actually use the root so
5421 * just pass in extent_root.
5423 tmp = read_tree_block(fs_info->extent_root, bytenr,
5426 fprintf(stderr, "Error reading root block\n");
5429 ret = pin_down_tree_blocks(fs_info, tmp, 0);
5430 free_extent_buffer(tmp);
5434 bytenr = btrfs_node_blockptr(eb, i);
5436 /* If we aren't the tree root don't read the block */
5437 if (level == 1 && !tree_root) {
5438 btrfs_pin_extent(fs_info, bytenr, leafsize);
5442 tmp = read_tree_block(fs_info->extent_root, bytenr,
5445 fprintf(stderr, "Error reading tree block\n");
5448 ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
5449 free_extent_buffer(tmp);
5458 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
5462 ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
5466 return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
5469 static int reset_block_groups(struct btrfs_fs_info *fs_info)
5471 struct btrfs_path *path;
5472 struct extent_buffer *leaf;
5473 struct btrfs_chunk *chunk;
5474 struct btrfs_key key;
5477 path = btrfs_alloc_path();
5482 key.type = BTRFS_CHUNK_ITEM_KEY;
5485 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
5487 btrfs_free_path(path);
5492 * We do this in case the block groups were screwed up and had alloc
5493 * bits that aren't actually set on the chunks. This happens with
5494 * restored images every time and could happen in real life I guess.
5496 fs_info->avail_data_alloc_bits = 0;
5497 fs_info->avail_metadata_alloc_bits = 0;
5498 fs_info->avail_system_alloc_bits = 0;
5500 /* First we need to create the in-memory block groups */
5502 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5503 ret = btrfs_next_leaf(fs_info->chunk_root, path);
5505 btrfs_free_path(path);
5513 leaf = path->nodes[0];
5514 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5515 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
5520 chunk = btrfs_item_ptr(leaf, path->slots[0],
5521 struct btrfs_chunk);
5522 btrfs_add_block_group(fs_info, 0,
5523 btrfs_chunk_type(leaf, chunk),
5524 key.objectid, key.offset,
5525 btrfs_chunk_length(leaf, chunk));
5529 btrfs_free_path(path);
5533 static int reset_balance(struct btrfs_trans_handle *trans,
5534 struct btrfs_fs_info *fs_info)
5536 struct btrfs_root *root = fs_info->tree_root;
5537 struct btrfs_path *path;
5538 struct extent_buffer *leaf;
5539 struct btrfs_key key;
5540 int del_slot, del_nr = 0;
5544 path = btrfs_alloc_path();
5548 key.objectid = BTRFS_BALANCE_OBJECTID;
5549 key.type = BTRFS_BALANCE_ITEM_KEY;
5552 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5559 ret = btrfs_del_item(trans, root, path);
5562 btrfs_release_path(path);
5564 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
5565 key.type = BTRFS_ROOT_ITEM_KEY;
5568 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5572 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5577 ret = btrfs_del_items(trans, root, path,
5584 btrfs_release_path(path);
5587 ret = btrfs_search_slot(trans, root, &key, path,
5594 leaf = path->nodes[0];
5595 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5596 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
5598 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
5603 del_slot = path->slots[0];
5612 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
5616 btrfs_release_path(path);
5618 key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
5619 key.type = BTRFS_ROOT_ITEM_KEY;
5620 key.offset = (u64)-1;
5621 root = btrfs_read_fs_root(fs_info, &key);
5623 fprintf(stderr, "Error reading data reloc tree\n");
5624 return PTR_ERR(root);
5626 root->track_dirty = 1;
5627 if (root->last_trans != trans->transid) {
5628 root->last_trans = trans->transid;
5629 root->commit_root = root->node;
5630 extent_buffer_get(root->node);
5632 ret = btrfs_fsck_reinit_root(trans, root, 0);
5634 btrfs_free_path(path);
5638 static int reinit_extent_tree(struct btrfs_fs_info *fs_info)
5640 struct btrfs_trans_handle *trans;
5645 * The only reason we don't do this is because right now we're just
5646 * walking the trees we find and pinning down their bytes, we don't look
5647 * at any of the leaves. In order to do mixed groups we'd have to check
5648 * the leaves of any fs roots and pin down the bytes for any file
5649 * extents we find. Not hard but why do it if we don't have to?
5651 if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
5652 fprintf(stderr, "We don't support re-initing the extent tree "
5653 "for mixed block groups yet, please notify a btrfs "
5654 "developer you want to do this so they can add this "
5655 "functionality.\n");
5659 trans = btrfs_start_transaction(fs_info->extent_root, 1);
5660 if (IS_ERR(trans)) {
5661 fprintf(stderr, "Error starting transaction\n");
5662 return PTR_ERR(trans);
5666 * first we need to walk all of the trees except the extent tree and pin
5667 * down the bytes that are in use so we don't overwrite any existing
5670 ret = pin_metadata_blocks(fs_info);
5672 fprintf(stderr, "error pinning down used bytes\n");
5677 * Need to drop all the block groups since we're going to recreate all
5680 btrfs_free_block_groups(fs_info);
5681 ret = reset_block_groups(fs_info);
5683 fprintf(stderr, "error resetting the block groups\n");
5687 /* Ok we can allocate now, reinit the extent root */
5688 ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 1);
5690 fprintf(stderr, "extent root initialization failed\n");
5692 * When the transaction code is updated we should end the
5693 * transaction, but for now progs only knows about commit so
5694 * just return an error.
5699 ret = reset_balance(trans, fs_info);
5701 fprintf(stderr, "error reseting the pending balance\n");
5706 * Now we have all the in-memory block groups setup so we can make
5707 * allocations properly, and the metadata we care about is safe since we
5708 * pinned all of it above.
5711 struct btrfs_block_group_cache *cache;
5713 cache = btrfs_lookup_first_block_group(fs_info, start);
5716 start = cache->key.objectid + cache->key.offset;
5717 ret = btrfs_insert_item(trans, fs_info->extent_root,
5718 &cache->key, &cache->item,
5719 sizeof(cache->item));
5721 fprintf(stderr, "Error adding block group\n");
5724 btrfs_extent_post_op(trans, fs_info->extent_root);
5728 * Ok now we commit and run the normal fsck, which will add extent
5729 * entries for all of the items it finds.
5731 return btrfs_commit_transaction(trans, fs_info->extent_root);
5734 static struct option long_options[] = {
5735 { "super", 1, NULL, 's' },
5736 { "repair", 0, NULL, 0 },
5737 { "init-csum-tree", 0, NULL, 0 },
5738 { "init-extent-tree", 0, NULL, 0 },
5742 const char * const cmd_check_usage[] = {
5743 "btrfs check [options] <device>",
5744 "Check an unmounted btrfs filesystem.",
5746 "-s|--super <superblock> use this superblock copy",
5747 "--repair try to repair the filesystem",
5748 "--init-csum-tree create a new CRC tree",
5749 "--init-extent-tree create a new extent tree",
5753 int cmd_check(int argc, char **argv)
5755 struct cache_tree root_cache;
5756 struct btrfs_root *root;
5757 struct btrfs_fs_info *info;
5763 int option_index = 0;
5764 int init_csum_tree = 0;
5765 int init_extent_tree = 0;
5770 c = getopt_long(argc, argv, "as:", long_options,
5775 case 'a': /* ignored */ break;
5778 bytenr = btrfs_sb_offset(num);
5779 printf("using SB copy %d, bytenr %llu\n", num,
5780 (unsigned long long)bytenr);
5784 usage(cmd_check_usage);
5786 if (option_index == 1) {
5787 printf("enabling repair mode\n");
5790 } else if (option_index == 2) {
5791 printf("Creating a new CRC tree\n");
5794 } else if (option_index == 3) {
5795 init_extent_tree = 1;
5801 argc = argc - optind;
5804 usage(cmd_check_usage);
5807 cache_tree_init(&root_cache);
5809 if((ret = check_mounted(argv[optind])) < 0) {
5810 fprintf(stderr, "Could not check mount status: %s\n", strerror(-ret));
5813 fprintf(stderr, "%s is currently mounted. Aborting.\n", argv[optind]);
5817 info = open_ctree_fs_info(argv[optind], bytenr, 0, rw, 1);
5819 fprintf(stderr, "Couldn't open file system\n");
5823 uuid_unparse(info->super_copy->fsid, uuidbuf);
5824 printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
5826 if (!extent_buffer_uptodate(info->tree_root->node) ||
5827 !extent_buffer_uptodate(info->dev_root->node) ||
5828 !extent_buffer_uptodate(info->extent_root->node) ||
5829 !extent_buffer_uptodate(info->chunk_root->node)) {
5830 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
5834 root = info->fs_root;
5836 if (init_extent_tree) {
5837 printf("Creating a new extent tree\n");
5838 ret = reinit_extent_tree(info);
5842 fprintf(stderr, "checking extents\n");
5843 if (init_csum_tree) {
5844 struct btrfs_trans_handle *trans;
5846 fprintf(stderr, "Reinit crc root\n");
5847 trans = btrfs_start_transaction(info->csum_root, 1);
5848 if (IS_ERR(trans)) {
5849 fprintf(stderr, "Error starting transaction\n");
5850 return PTR_ERR(trans);
5853 ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
5855 fprintf(stderr, "crc root initialization failed\n");
5859 ret = btrfs_commit_transaction(trans, info->csum_root);
5864 ret = check_chunks_and_extents(root, repair);
5866 fprintf(stderr, "Errors found in extent allocation tree or chunk allocation\n");
5868 fprintf(stderr, "checking free space cache\n");
5869 ret = check_space_cache(root);
5873 fprintf(stderr, "checking fs roots\n");
5874 ret = check_fs_roots(root, &root_cache);
5878 fprintf(stderr, "checking csums\n");
5879 ret = check_csums(root);
5883 fprintf(stderr, "checking root refs\n");
5884 ret = check_root_refs(root, &root_cache);
5886 free_root_recs_tree(&root_cache);
5889 if (found_old_backref) { /*
5890 * there was a disk format change when mixed
5891 * backref was in testing tree. The old format
5892 * existed about one week.
5894 printf("\n * Found old mixed backref format. "
5895 "The old format is not supported! *"
5896 "\n * Please mount the FS in readonly mode, "
5897 "backup data and re-format the FS. *\n\n");
5900 printf("found %llu bytes used err is %d\n",
5901 (unsigned long long)bytes_used, ret);
5902 printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
5903 printf("total tree bytes: %llu\n",
5904 (unsigned long long)total_btree_bytes);
5905 printf("total fs tree bytes: %llu\n",
5906 (unsigned long long)total_fs_tree_bytes);
5907 printf("total extent tree bytes: %llu\n",
5908 (unsigned long long)total_extent_tree_bytes);
5909 printf("btree space waste bytes: %llu\n",
5910 (unsigned long long)btree_space_waste);
5911 printf("file data blocks allocated: %llu\n referenced %llu\n",
5912 (unsigned long long)data_bytes_allocated,
5913 (unsigned long long)data_bytes_referenced);
5914 printf("%s\n", BTRFS_BUILD_VERSION);