2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #define _XOPEN_SOURCE 500
25 #include <sys/types.h>
29 #include <uuid/uuid.h>
34 #include "print-tree.h"
35 #include "transaction.h"
39 #include "free-space-cache.h"
41 #include "qgroup-verify.h"
42 #include "rbtree-utils.h"
46 static u64 bytes_used = 0;
47 static u64 total_csum_bytes = 0;
48 static u64 total_btree_bytes = 0;
49 static u64 total_fs_tree_bytes = 0;
50 static u64 total_extent_tree_bytes = 0;
51 static u64 btree_space_waste = 0;
52 static u64 data_bytes_allocated = 0;
53 static u64 data_bytes_referenced = 0;
54 static int found_old_backref = 0;
55 static LIST_HEAD(duplicate_extents);
56 static LIST_HEAD(delete_items);
57 static int repair = 0;
58 static int no_holes = 0;
59 static int init_extent_tree = 0;
60 static int check_data_csum = 0;
62 struct extent_backref {
63 struct list_head list;
64 unsigned int is_data:1;
65 unsigned int found_extent_tree:1;
66 unsigned int full_backref:1;
67 unsigned int found_ref:1;
68 unsigned int broken:1;
72 struct extent_backref node;
87 struct extent_backref node;
94 struct extent_record {
95 struct list_head backrefs;
96 struct list_head dups;
97 struct list_head list;
98 struct cache_extent cache;
99 struct btrfs_disk_key parent_key;
104 u64 extent_item_refs;
106 u64 parent_generation;
110 unsigned int found_rec:1;
111 unsigned int content_checked:1;
112 unsigned int owner_ref_checked:1;
113 unsigned int is_root:1;
114 unsigned int metadata:1;
117 struct inode_backref {
118 struct list_head list;
119 unsigned int found_dir_item:1;
120 unsigned int found_dir_index:1;
121 unsigned int found_inode_ref:1;
122 unsigned int filetype:8;
124 unsigned int ref_type;
131 struct dropping_root_item_record {
132 struct list_head list;
133 struct btrfs_root_item ri;
134 struct btrfs_key found_key;
137 #define REF_ERR_NO_DIR_ITEM (1 << 0)
138 #define REF_ERR_NO_DIR_INDEX (1 << 1)
139 #define REF_ERR_NO_INODE_REF (1 << 2)
140 #define REF_ERR_DUP_DIR_ITEM (1 << 3)
141 #define REF_ERR_DUP_DIR_INDEX (1 << 4)
142 #define REF_ERR_DUP_INODE_REF (1 << 5)
143 #define REF_ERR_INDEX_UNMATCH (1 << 6)
144 #define REF_ERR_FILETYPE_UNMATCH (1 << 7)
145 #define REF_ERR_NAME_TOO_LONG (1 << 8) // 100
146 #define REF_ERR_NO_ROOT_REF (1 << 9)
147 #define REF_ERR_NO_ROOT_BACKREF (1 << 10)
148 #define REF_ERR_DUP_ROOT_REF (1 << 11)
149 #define REF_ERR_DUP_ROOT_BACKREF (1 << 12)
151 struct inode_record {
152 struct list_head backrefs;
153 unsigned int checked:1;
154 unsigned int merging:1;
155 unsigned int found_inode_item:1;
156 unsigned int found_dir_item:1;
157 unsigned int found_file_extent:1;
158 unsigned int found_csum_item:1;
159 unsigned int some_csum_missing:1;
160 unsigned int nodatasum:1;
173 u64 first_extent_gap;
178 #define I_ERR_NO_INODE_ITEM (1 << 0)
179 #define I_ERR_NO_ORPHAN_ITEM (1 << 1)
180 #define I_ERR_DUP_INODE_ITEM (1 << 2)
181 #define I_ERR_DUP_DIR_INDEX (1 << 3)
182 #define I_ERR_ODD_DIR_ITEM (1 << 4)
183 #define I_ERR_ODD_FILE_EXTENT (1 << 5)
184 #define I_ERR_BAD_FILE_EXTENT (1 << 6)
185 #define I_ERR_FILE_EXTENT_OVERLAP (1 << 7)
186 #define I_ERR_FILE_EXTENT_DISCOUNT (1 << 8) // 100
187 #define I_ERR_DIR_ISIZE_WRONG (1 << 9)
188 #define I_ERR_FILE_NBYTES_WRONG (1 << 10) // 400
189 #define I_ERR_ODD_CSUM_ITEM (1 << 11)
190 #define I_ERR_SOME_CSUM_MISSING (1 << 12)
191 #define I_ERR_LINK_COUNT_WRONG (1 << 13)
193 struct root_backref {
194 struct list_head list;
195 unsigned int found_dir_item:1;
196 unsigned int found_dir_index:1;
197 unsigned int found_back_ref:1;
198 unsigned int found_forward_ref:1;
199 unsigned int reachable:1;
209 struct list_head backrefs;
210 struct cache_extent cache;
211 unsigned int found_root_item:1;
217 struct cache_extent cache;
222 struct cache_extent cache;
223 struct cache_tree root_cache;
224 struct cache_tree inode_cache;
225 struct inode_record *current;
234 struct walk_control {
235 struct cache_tree shared;
236 struct shared_node *nodes[BTRFS_MAX_LEVEL];
242 struct btrfs_key key;
244 struct list_head list;
247 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info);
249 static void record_root_in_trans(struct btrfs_trans_handle *trans,
250 struct btrfs_root *root)
252 if (root->last_trans != trans->transid) {
253 root->track_dirty = 1;
254 root->last_trans = trans->transid;
255 root->commit_root = root->node;
256 extent_buffer_get(root->node);
260 static u8 imode_to_type(u32 imode)
263 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
264 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
265 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
266 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
267 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
268 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
269 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
270 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
273 return btrfs_type_by_mode[(imode & S_IFMT) >> S_SHIFT];
277 static int device_record_compare(struct rb_node *node1, struct rb_node *node2)
279 struct device_record *rec1;
280 struct device_record *rec2;
282 rec1 = rb_entry(node1, struct device_record, node);
283 rec2 = rb_entry(node2, struct device_record, node);
284 if (rec1->devid > rec2->devid)
286 else if (rec1->devid < rec2->devid)
292 static struct inode_record *clone_inode_rec(struct inode_record *orig_rec)
294 struct inode_record *rec;
295 struct inode_backref *backref;
296 struct inode_backref *orig;
299 rec = malloc(sizeof(*rec));
300 memcpy(rec, orig_rec, sizeof(*rec));
302 INIT_LIST_HEAD(&rec->backrefs);
304 list_for_each_entry(orig, &orig_rec->backrefs, list) {
305 size = sizeof(*orig) + orig->namelen + 1;
306 backref = malloc(size);
307 memcpy(backref, orig, size);
308 list_add_tail(&backref->list, &rec->backrefs);
313 static void print_inode_error(struct btrfs_root *root, struct inode_record *rec)
315 u64 root_objectid = root->root_key.objectid;
316 int errors = rec->errors;
320 /* reloc root errors, we print its corresponding fs root objectid*/
321 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
322 root_objectid = root->root_key.offset;
323 fprintf(stderr, "reloc");
325 fprintf(stderr, "root %llu inode %llu errors %x",
326 (unsigned long long) root_objectid,
327 (unsigned long long) rec->ino, rec->errors);
329 if (errors & I_ERR_NO_INODE_ITEM)
330 fprintf(stderr, ", no inode item");
331 if (errors & I_ERR_NO_ORPHAN_ITEM)
332 fprintf(stderr, ", no orphan item");
333 if (errors & I_ERR_DUP_INODE_ITEM)
334 fprintf(stderr, ", dup inode item");
335 if (errors & I_ERR_DUP_DIR_INDEX)
336 fprintf(stderr, ", dup dir index");
337 if (errors & I_ERR_ODD_DIR_ITEM)
338 fprintf(stderr, ", odd dir item");
339 if (errors & I_ERR_ODD_FILE_EXTENT)
340 fprintf(stderr, ", odd file extent");
341 if (errors & I_ERR_BAD_FILE_EXTENT)
342 fprintf(stderr, ", bad file extent");
343 if (errors & I_ERR_FILE_EXTENT_OVERLAP)
344 fprintf(stderr, ", file extent overlap");
345 if (errors & I_ERR_FILE_EXTENT_DISCOUNT)
346 fprintf(stderr, ", file extent discount");
347 if (errors & I_ERR_DIR_ISIZE_WRONG)
348 fprintf(stderr, ", dir isize wrong");
349 if (errors & I_ERR_FILE_NBYTES_WRONG)
350 fprintf(stderr, ", nbytes wrong");
351 if (errors & I_ERR_ODD_CSUM_ITEM)
352 fprintf(stderr, ", odd csum item");
353 if (errors & I_ERR_SOME_CSUM_MISSING)
354 fprintf(stderr, ", some csum missing");
355 if (errors & I_ERR_LINK_COUNT_WRONG)
356 fprintf(stderr, ", link count wrong");
357 fprintf(stderr, "\n");
360 static void print_ref_error(int errors)
362 if (errors & REF_ERR_NO_DIR_ITEM)
363 fprintf(stderr, ", no dir item");
364 if (errors & REF_ERR_NO_DIR_INDEX)
365 fprintf(stderr, ", no dir index");
366 if (errors & REF_ERR_NO_INODE_REF)
367 fprintf(stderr, ", no inode ref");
368 if (errors & REF_ERR_DUP_DIR_ITEM)
369 fprintf(stderr, ", dup dir item");
370 if (errors & REF_ERR_DUP_DIR_INDEX)
371 fprintf(stderr, ", dup dir index");
372 if (errors & REF_ERR_DUP_INODE_REF)
373 fprintf(stderr, ", dup inode ref");
374 if (errors & REF_ERR_INDEX_UNMATCH)
375 fprintf(stderr, ", index unmatch");
376 if (errors & REF_ERR_FILETYPE_UNMATCH)
377 fprintf(stderr, ", filetype unmatch");
378 if (errors & REF_ERR_NAME_TOO_LONG)
379 fprintf(stderr, ", name too long");
380 if (errors & REF_ERR_NO_ROOT_REF)
381 fprintf(stderr, ", no root ref");
382 if (errors & REF_ERR_NO_ROOT_BACKREF)
383 fprintf(stderr, ", no root backref");
384 if (errors & REF_ERR_DUP_ROOT_REF)
385 fprintf(stderr, ", dup root ref");
386 if (errors & REF_ERR_DUP_ROOT_BACKREF)
387 fprintf(stderr, ", dup root backref");
388 fprintf(stderr, "\n");
391 static struct inode_record *get_inode_rec(struct cache_tree *inode_cache,
394 struct ptr_node *node;
395 struct cache_extent *cache;
396 struct inode_record *rec = NULL;
399 cache = lookup_cache_extent(inode_cache, ino, 1);
401 node = container_of(cache, struct ptr_node, cache);
403 if (mod && rec->refs > 1) {
404 node->data = clone_inode_rec(rec);
409 rec = calloc(1, sizeof(*rec));
411 rec->extent_start = (u64)-1;
412 rec->first_extent_gap = (u64)-1;
414 INIT_LIST_HEAD(&rec->backrefs);
416 node = malloc(sizeof(*node));
417 node->cache.start = ino;
418 node->cache.size = 1;
421 if (ino == BTRFS_FREE_INO_OBJECTID)
424 ret = insert_cache_extent(inode_cache, &node->cache);
430 static void free_inode_rec(struct inode_record *rec)
432 struct inode_backref *backref;
437 while (!list_empty(&rec->backrefs)) {
438 backref = list_entry(rec->backrefs.next,
439 struct inode_backref, list);
440 list_del(&backref->list);
446 static int can_free_inode_rec(struct inode_record *rec)
448 if (!rec->errors && rec->checked && rec->found_inode_item &&
449 rec->nlink == rec->found_link && list_empty(&rec->backrefs))
454 static void maybe_free_inode_rec(struct cache_tree *inode_cache,
455 struct inode_record *rec)
457 struct cache_extent *cache;
458 struct inode_backref *tmp, *backref;
459 struct ptr_node *node;
460 unsigned char filetype;
462 if (!rec->found_inode_item)
465 filetype = imode_to_type(rec->imode);
466 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
467 if (backref->found_dir_item && backref->found_dir_index) {
468 if (backref->filetype != filetype)
469 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
470 if (!backref->errors && backref->found_inode_ref) {
471 list_del(&backref->list);
477 if (!rec->checked || rec->merging)
480 if (S_ISDIR(rec->imode)) {
481 if (rec->found_size != rec->isize)
482 rec->errors |= I_ERR_DIR_ISIZE_WRONG;
483 if (rec->found_file_extent)
484 rec->errors |= I_ERR_ODD_FILE_EXTENT;
485 } else if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
486 if (rec->found_dir_item)
487 rec->errors |= I_ERR_ODD_DIR_ITEM;
488 if (rec->found_size != rec->nbytes)
489 rec->errors |= I_ERR_FILE_NBYTES_WRONG;
490 if (rec->extent_start == (u64)-1 || rec->extent_start > 0)
491 rec->first_extent_gap = 0;
492 if (rec->nlink > 0 && !no_holes &&
493 (rec->extent_end < rec->isize ||
494 rec->first_extent_gap < rec->isize))
495 rec->errors |= I_ERR_FILE_EXTENT_DISCOUNT;
498 if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
499 if (rec->found_csum_item && rec->nodatasum)
500 rec->errors |= I_ERR_ODD_CSUM_ITEM;
501 if (rec->some_csum_missing && !rec->nodatasum)
502 rec->errors |= I_ERR_SOME_CSUM_MISSING;
505 BUG_ON(rec->refs != 1);
506 if (can_free_inode_rec(rec)) {
507 cache = lookup_cache_extent(inode_cache, rec->ino, 1);
508 node = container_of(cache, struct ptr_node, cache);
509 BUG_ON(node->data != rec);
510 remove_cache_extent(inode_cache, &node->cache);
516 static int check_orphan_item(struct btrfs_root *root, u64 ino)
518 struct btrfs_path path;
519 struct btrfs_key key;
522 key.objectid = BTRFS_ORPHAN_OBJECTID;
523 key.type = BTRFS_ORPHAN_ITEM_KEY;
526 btrfs_init_path(&path);
527 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
528 btrfs_release_path(&path);
534 static int process_inode_item(struct extent_buffer *eb,
535 int slot, struct btrfs_key *key,
536 struct shared_node *active_node)
538 struct inode_record *rec;
539 struct btrfs_inode_item *item;
541 rec = active_node->current;
542 BUG_ON(rec->ino != key->objectid || rec->refs > 1);
543 if (rec->found_inode_item) {
544 rec->errors |= I_ERR_DUP_INODE_ITEM;
547 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
548 rec->nlink = btrfs_inode_nlink(eb, item);
549 rec->isize = btrfs_inode_size(eb, item);
550 rec->nbytes = btrfs_inode_nbytes(eb, item);
551 rec->imode = btrfs_inode_mode(eb, item);
552 if (btrfs_inode_flags(eb, item) & BTRFS_INODE_NODATASUM)
554 rec->found_inode_item = 1;
556 rec->errors |= I_ERR_NO_ORPHAN_ITEM;
557 maybe_free_inode_rec(&active_node->inode_cache, rec);
561 static struct inode_backref *get_inode_backref(struct inode_record *rec,
563 int namelen, u64 dir)
565 struct inode_backref *backref;
567 list_for_each_entry(backref, &rec->backrefs, list) {
568 if (rec->ino == BTRFS_MULTIPLE_OBJECTIDS)
570 if (backref->dir != dir || backref->namelen != namelen)
572 if (memcmp(name, backref->name, namelen))
577 backref = malloc(sizeof(*backref) + namelen + 1);
578 memset(backref, 0, sizeof(*backref));
580 backref->namelen = namelen;
581 memcpy(backref->name, name, namelen);
582 backref->name[namelen] = '\0';
583 list_add_tail(&backref->list, &rec->backrefs);
587 static int add_inode_backref(struct cache_tree *inode_cache,
588 u64 ino, u64 dir, u64 index,
589 const char *name, int namelen,
590 int filetype, int itemtype, int errors)
592 struct inode_record *rec;
593 struct inode_backref *backref;
595 rec = get_inode_rec(inode_cache, ino, 1);
596 backref = get_inode_backref(rec, name, namelen, dir);
598 backref->errors |= errors;
599 if (itemtype == BTRFS_DIR_INDEX_KEY) {
600 if (backref->found_dir_index)
601 backref->errors |= REF_ERR_DUP_DIR_INDEX;
602 if (backref->found_inode_ref && backref->index != index)
603 backref->errors |= REF_ERR_INDEX_UNMATCH;
604 if (backref->found_dir_item && backref->filetype != filetype)
605 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
607 backref->index = index;
608 backref->filetype = filetype;
609 backref->found_dir_index = 1;
610 } else if (itemtype == BTRFS_DIR_ITEM_KEY) {
612 if (backref->found_dir_item)
613 backref->errors |= REF_ERR_DUP_DIR_ITEM;
614 if (backref->found_dir_index && backref->filetype != filetype)
615 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
617 backref->filetype = filetype;
618 backref->found_dir_item = 1;
619 } else if ((itemtype == BTRFS_INODE_REF_KEY) ||
620 (itemtype == BTRFS_INODE_EXTREF_KEY)) {
621 if (backref->found_inode_ref)
622 backref->errors |= REF_ERR_DUP_INODE_REF;
623 if (backref->found_dir_index && backref->index != index)
624 backref->errors |= REF_ERR_INDEX_UNMATCH;
626 backref->index = index;
628 backref->ref_type = itemtype;
629 backref->found_inode_ref = 1;
634 maybe_free_inode_rec(inode_cache, rec);
638 static int merge_inode_recs(struct inode_record *src, struct inode_record *dst,
639 struct cache_tree *dst_cache)
641 struct inode_backref *backref;
645 list_for_each_entry(backref, &src->backrefs, list) {
646 if (backref->found_dir_index) {
647 add_inode_backref(dst_cache, dst->ino, backref->dir,
648 backref->index, backref->name,
649 backref->namelen, backref->filetype,
650 BTRFS_DIR_INDEX_KEY, backref->errors);
652 if (backref->found_dir_item) {
654 add_inode_backref(dst_cache, dst->ino,
655 backref->dir, 0, backref->name,
656 backref->namelen, backref->filetype,
657 BTRFS_DIR_ITEM_KEY, backref->errors);
659 if (backref->found_inode_ref) {
660 add_inode_backref(dst_cache, dst->ino,
661 backref->dir, backref->index,
662 backref->name, backref->namelen, 0,
663 backref->ref_type, backref->errors);
667 if (src->found_dir_item)
668 dst->found_dir_item = 1;
669 if (src->found_file_extent)
670 dst->found_file_extent = 1;
671 if (src->found_csum_item)
672 dst->found_csum_item = 1;
673 if (src->some_csum_missing)
674 dst->some_csum_missing = 1;
675 if (dst->first_extent_gap > src->first_extent_gap)
676 dst->first_extent_gap = src->first_extent_gap;
678 BUG_ON(src->found_link < dir_count);
679 dst->found_link += src->found_link - dir_count;
680 dst->found_size += src->found_size;
681 if (src->extent_start != (u64)-1) {
682 if (dst->extent_start == (u64)-1) {
683 dst->extent_start = src->extent_start;
684 dst->extent_end = src->extent_end;
686 if (dst->extent_end > src->extent_start)
687 dst->errors |= I_ERR_FILE_EXTENT_OVERLAP;
688 else if (dst->extent_end < src->extent_start &&
689 dst->extent_end < dst->first_extent_gap)
690 dst->first_extent_gap = dst->extent_end;
691 if (dst->extent_end < src->extent_end)
692 dst->extent_end = src->extent_end;
696 dst->errors |= src->errors;
697 if (src->found_inode_item) {
698 if (!dst->found_inode_item) {
699 dst->nlink = src->nlink;
700 dst->isize = src->isize;
701 dst->nbytes = src->nbytes;
702 dst->imode = src->imode;
703 dst->nodatasum = src->nodatasum;
704 dst->found_inode_item = 1;
706 dst->errors |= I_ERR_DUP_INODE_ITEM;
714 static int splice_shared_node(struct shared_node *src_node,
715 struct shared_node *dst_node)
717 struct cache_extent *cache;
718 struct ptr_node *node, *ins;
719 struct cache_tree *src, *dst;
720 struct inode_record *rec, *conflict;
725 if (--src_node->refs == 0)
727 if (src_node->current)
728 current_ino = src_node->current->ino;
730 src = &src_node->root_cache;
731 dst = &dst_node->root_cache;
733 cache = search_cache_extent(src, 0);
735 node = container_of(cache, struct ptr_node, cache);
737 cache = next_cache_extent(cache);
740 remove_cache_extent(src, &node->cache);
743 ins = malloc(sizeof(*ins));
744 ins->cache.start = node->cache.start;
745 ins->cache.size = node->cache.size;
749 ret = insert_cache_extent(dst, &ins->cache);
750 if (ret == -EEXIST) {
751 conflict = get_inode_rec(dst, rec->ino, 1);
752 merge_inode_recs(rec, conflict, dst);
754 conflict->checked = 1;
755 if (dst_node->current == conflict)
756 dst_node->current = NULL;
758 maybe_free_inode_rec(dst, conflict);
766 if (src == &src_node->root_cache) {
767 src = &src_node->inode_cache;
768 dst = &dst_node->inode_cache;
772 if (current_ino > 0 && (!dst_node->current ||
773 current_ino > dst_node->current->ino)) {
774 if (dst_node->current) {
775 dst_node->current->checked = 1;
776 maybe_free_inode_rec(dst, dst_node->current);
778 dst_node->current = get_inode_rec(dst, current_ino, 1);
783 static void free_inode_ptr(struct cache_extent *cache)
785 struct ptr_node *node;
786 struct inode_record *rec;
788 node = container_of(cache, struct ptr_node, cache);
794 FREE_EXTENT_CACHE_BASED_TREE(inode_recs, free_inode_ptr);
796 static struct shared_node *find_shared_node(struct cache_tree *shared,
799 struct cache_extent *cache;
800 struct shared_node *node;
802 cache = lookup_cache_extent(shared, bytenr, 1);
804 node = container_of(cache, struct shared_node, cache);
810 static int add_shared_node(struct cache_tree *shared, u64 bytenr, u32 refs)
813 struct shared_node *node;
815 node = calloc(1, sizeof(*node));
816 node->cache.start = bytenr;
817 node->cache.size = 1;
818 cache_tree_init(&node->root_cache);
819 cache_tree_init(&node->inode_cache);
822 ret = insert_cache_extent(shared, &node->cache);
827 static int enter_shared_node(struct btrfs_root *root, u64 bytenr, u32 refs,
828 struct walk_control *wc, int level)
830 struct shared_node *node;
831 struct shared_node *dest;
833 if (level == wc->active_node)
836 BUG_ON(wc->active_node <= level);
837 node = find_shared_node(&wc->shared, bytenr);
839 add_shared_node(&wc->shared, bytenr, refs);
840 node = find_shared_node(&wc->shared, bytenr);
841 wc->nodes[level] = node;
842 wc->active_node = level;
846 if (wc->root_level == wc->active_node &&
847 btrfs_root_refs(&root->root_item) == 0) {
848 if (--node->refs == 0) {
849 free_inode_recs_tree(&node->root_cache);
850 free_inode_recs_tree(&node->inode_cache);
851 remove_cache_extent(&wc->shared, &node->cache);
857 dest = wc->nodes[wc->active_node];
858 splice_shared_node(node, dest);
859 if (node->refs == 0) {
860 remove_cache_extent(&wc->shared, &node->cache);
866 static int leave_shared_node(struct btrfs_root *root,
867 struct walk_control *wc, int level)
869 struct shared_node *node;
870 struct shared_node *dest;
873 if (level == wc->root_level)
876 for (i = level + 1; i < BTRFS_MAX_LEVEL; i++) {
880 BUG_ON(i >= BTRFS_MAX_LEVEL);
882 node = wc->nodes[wc->active_node];
883 wc->nodes[wc->active_node] = NULL;
886 dest = wc->nodes[wc->active_node];
887 if (wc->active_node < wc->root_level ||
888 btrfs_root_refs(&root->root_item) > 0) {
889 BUG_ON(node->refs <= 1);
890 splice_shared_node(node, dest);
892 BUG_ON(node->refs < 2);
901 * 1 - if the root with id child_root_id is a child of root parent_root_id
902 * 0 - if the root child_root_id isn't a child of the root parent_root_id but
903 * has other root(s) as parent(s)
904 * 2 - if the root child_root_id doesn't have any parent roots
906 static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
909 struct btrfs_path path;
910 struct btrfs_key key;
911 struct extent_buffer *leaf;
915 btrfs_init_path(&path);
917 key.objectid = parent_root_id;
918 key.type = BTRFS_ROOT_REF_KEY;
919 key.offset = child_root_id;
920 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
924 btrfs_release_path(&path);
928 key.objectid = child_root_id;
929 key.type = BTRFS_ROOT_BACKREF_KEY;
931 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
937 leaf = path.nodes[0];
938 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
939 ret = btrfs_next_leaf(root->fs_info->tree_root, &path);
942 leaf = path.nodes[0];
945 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
946 if (key.objectid != child_root_id ||
947 key.type != BTRFS_ROOT_BACKREF_KEY)
952 if (key.offset == parent_root_id) {
953 btrfs_release_path(&path);
960 btrfs_release_path(&path);
963 return has_parent ? 0 : 2;
966 static int process_dir_item(struct btrfs_root *root,
967 struct extent_buffer *eb,
968 int slot, struct btrfs_key *key,
969 struct shared_node *active_node)
979 struct btrfs_dir_item *di;
980 struct inode_record *rec;
981 struct cache_tree *root_cache;
982 struct cache_tree *inode_cache;
983 struct btrfs_key location;
984 char namebuf[BTRFS_NAME_LEN];
986 root_cache = &active_node->root_cache;
987 inode_cache = &active_node->inode_cache;
988 rec = active_node->current;
989 rec->found_dir_item = 1;
991 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
992 total = btrfs_item_size_nr(eb, slot);
993 while (cur < total) {
995 btrfs_dir_item_key_to_cpu(eb, di, &location);
996 name_len = btrfs_dir_name_len(eb, di);
997 data_len = btrfs_dir_data_len(eb, di);
998 filetype = btrfs_dir_type(eb, di);
1000 rec->found_size += name_len;
1001 if (name_len <= BTRFS_NAME_LEN) {
1005 len = BTRFS_NAME_LEN;
1006 error = REF_ERR_NAME_TOO_LONG;
1008 read_extent_buffer(eb, namebuf, (unsigned long)(di + 1), len);
1010 if (location.type == BTRFS_INODE_ITEM_KEY) {
1011 add_inode_backref(inode_cache, location.objectid,
1012 key->objectid, key->offset, namebuf,
1013 len, filetype, key->type, error);
1014 } else if (location.type == BTRFS_ROOT_ITEM_KEY) {
1015 add_inode_backref(root_cache, location.objectid,
1016 key->objectid, key->offset,
1017 namebuf, len, filetype,
1020 fprintf(stderr, "invalid location in dir item %u\n",
1022 add_inode_backref(inode_cache, BTRFS_MULTIPLE_OBJECTIDS,
1023 key->objectid, key->offset, namebuf,
1024 len, filetype, key->type, error);
1027 len = sizeof(*di) + name_len + data_len;
1028 di = (struct btrfs_dir_item *)((char *)di + len);
1031 if (key->type == BTRFS_DIR_INDEX_KEY && nritems > 1)
1032 rec->errors |= I_ERR_DUP_DIR_INDEX;
1037 static int process_inode_ref(struct extent_buffer *eb,
1038 int slot, struct btrfs_key *key,
1039 struct shared_node *active_node)
1047 struct cache_tree *inode_cache;
1048 struct btrfs_inode_ref *ref;
1049 char namebuf[BTRFS_NAME_LEN];
1051 inode_cache = &active_node->inode_cache;
1053 ref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1054 total = btrfs_item_size_nr(eb, slot);
1055 while (cur < total) {
1056 name_len = btrfs_inode_ref_name_len(eb, ref);
1057 index = btrfs_inode_ref_index(eb, ref);
1058 if (name_len <= BTRFS_NAME_LEN) {
1062 len = BTRFS_NAME_LEN;
1063 error = REF_ERR_NAME_TOO_LONG;
1065 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
1066 add_inode_backref(inode_cache, key->objectid, key->offset,
1067 index, namebuf, len, 0, key->type, error);
1069 len = sizeof(*ref) + name_len;
1070 ref = (struct btrfs_inode_ref *)((char *)ref + len);
1076 static int process_inode_extref(struct extent_buffer *eb,
1077 int slot, struct btrfs_key *key,
1078 struct shared_node *active_node)
1087 struct cache_tree *inode_cache;
1088 struct btrfs_inode_extref *extref;
1089 char namebuf[BTRFS_NAME_LEN];
1091 inode_cache = &active_node->inode_cache;
1093 extref = btrfs_item_ptr(eb, slot, struct btrfs_inode_extref);
1094 total = btrfs_item_size_nr(eb, slot);
1095 while (cur < total) {
1096 name_len = btrfs_inode_extref_name_len(eb, extref);
1097 index = btrfs_inode_extref_index(eb, extref);
1098 parent = btrfs_inode_extref_parent(eb, extref);
1099 if (name_len <= BTRFS_NAME_LEN) {
1103 len = BTRFS_NAME_LEN;
1104 error = REF_ERR_NAME_TOO_LONG;
1106 read_extent_buffer(eb, namebuf,
1107 (unsigned long)(extref + 1), len);
1108 add_inode_backref(inode_cache, key->objectid, parent,
1109 index, namebuf, len, 0, key->type, error);
1111 len = sizeof(*extref) + name_len;
1112 extref = (struct btrfs_inode_extref *)((char *)extref + len);
1119 static int count_csum_range(struct btrfs_root *root, u64 start,
1120 u64 len, u64 *found)
1122 struct btrfs_key key;
1123 struct btrfs_path path;
1124 struct extent_buffer *leaf;
1129 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
1131 btrfs_init_path(&path);
1133 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1135 key.type = BTRFS_EXTENT_CSUM_KEY;
1137 ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
1141 if (ret > 0 && path.slots[0] > 0) {
1142 leaf = path.nodes[0];
1143 btrfs_item_key_to_cpu(leaf, &key, path.slots[0] - 1);
1144 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
1145 key.type == BTRFS_EXTENT_CSUM_KEY)
1150 leaf = path.nodes[0];
1151 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1152 ret = btrfs_next_leaf(root->fs_info->csum_root, &path);
1157 leaf = path.nodes[0];
1160 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1161 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1162 key.type != BTRFS_EXTENT_CSUM_KEY)
1165 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1166 if (key.offset >= start + len)
1169 if (key.offset > start)
1172 size = btrfs_item_size_nr(leaf, path.slots[0]);
1173 csum_end = key.offset + (size / csum_size) * root->sectorsize;
1174 if (csum_end > start) {
1175 size = min(csum_end - start, len);
1186 btrfs_release_path(&path);
1190 static int process_file_extent(struct btrfs_root *root,
1191 struct extent_buffer *eb,
1192 int slot, struct btrfs_key *key,
1193 struct shared_node *active_node)
1195 struct inode_record *rec;
1196 struct btrfs_file_extent_item *fi;
1198 u64 disk_bytenr = 0;
1199 u64 extent_offset = 0;
1200 u64 mask = root->sectorsize - 1;
1204 rec = active_node->current;
1205 BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1206 rec->found_file_extent = 1;
1208 if (rec->extent_start == (u64)-1) {
1209 rec->extent_start = key->offset;
1210 rec->extent_end = key->offset;
1213 if (rec->extent_end > key->offset)
1214 rec->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1215 else if (rec->extent_end < key->offset &&
1216 rec->extent_end < rec->first_extent_gap)
1217 rec->first_extent_gap = rec->extent_end;
1219 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
1220 extent_type = btrfs_file_extent_type(eb, fi);
1222 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1223 num_bytes = btrfs_file_extent_inline_len(eb, slot, fi);
1225 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1226 rec->found_size += num_bytes;
1227 num_bytes = (num_bytes + mask) & ~mask;
1228 } else if (extent_type == BTRFS_FILE_EXTENT_REG ||
1229 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1230 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1231 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1232 extent_offset = btrfs_file_extent_offset(eb, fi);
1233 if (num_bytes == 0 || (num_bytes & mask))
1234 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1235 if (num_bytes + extent_offset >
1236 btrfs_file_extent_ram_bytes(eb, fi))
1237 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1238 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC &&
1239 (btrfs_file_extent_compression(eb, fi) ||
1240 btrfs_file_extent_encryption(eb, fi) ||
1241 btrfs_file_extent_other_encoding(eb, fi)))
1242 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1243 if (disk_bytenr > 0)
1244 rec->found_size += num_bytes;
1246 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1248 rec->extent_end = key->offset + num_bytes;
1250 if (disk_bytenr > 0) {
1252 if (btrfs_file_extent_compression(eb, fi))
1253 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1255 disk_bytenr += extent_offset;
1257 ret = count_csum_range(root, disk_bytenr, num_bytes, &found);
1260 if (extent_type == BTRFS_FILE_EXTENT_REG) {
1262 rec->found_csum_item = 1;
1263 if (found < num_bytes)
1264 rec->some_csum_missing = 1;
1265 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1267 rec->errors |= I_ERR_ODD_CSUM_ITEM;
1273 static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
1274 struct walk_control *wc)
1276 struct btrfs_key key;
1280 struct cache_tree *inode_cache;
1281 struct shared_node *active_node;
1283 if (wc->root_level == wc->active_node &&
1284 btrfs_root_refs(&root->root_item) == 0)
1287 active_node = wc->nodes[wc->active_node];
1288 inode_cache = &active_node->inode_cache;
1289 nritems = btrfs_header_nritems(eb);
1290 for (i = 0; i < nritems; i++) {
1291 btrfs_item_key_to_cpu(eb, &key, i);
1293 if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
1295 if (key.type == BTRFS_ORPHAN_ITEM_KEY)
1298 if (active_node->current == NULL ||
1299 active_node->current->ino < key.objectid) {
1300 if (active_node->current) {
1301 active_node->current->checked = 1;
1302 maybe_free_inode_rec(inode_cache,
1303 active_node->current);
1305 active_node->current = get_inode_rec(inode_cache,
1309 case BTRFS_DIR_ITEM_KEY:
1310 case BTRFS_DIR_INDEX_KEY:
1311 ret = process_dir_item(root, eb, i, &key, active_node);
1313 case BTRFS_INODE_REF_KEY:
1314 ret = process_inode_ref(eb, i, &key, active_node);
1316 case BTRFS_INODE_EXTREF_KEY:
1317 ret = process_inode_extref(eb, i, &key, active_node);
1319 case BTRFS_INODE_ITEM_KEY:
1320 ret = process_inode_item(eb, i, &key, active_node);
1322 case BTRFS_EXTENT_DATA_KEY:
1323 ret = process_file_extent(root, eb, i, &key,
1333 static void reada_walk_down(struct btrfs_root *root,
1334 struct extent_buffer *node, int slot)
1343 level = btrfs_header_level(node);
1347 nritems = btrfs_header_nritems(node);
1348 blocksize = btrfs_level_size(root, level - 1);
1349 for (i = slot; i < nritems; i++) {
1350 bytenr = btrfs_node_blockptr(node, i);
1351 ptr_gen = btrfs_node_ptr_generation(node, i);
1352 readahead_tree_block(root, bytenr, blocksize, ptr_gen);
1357 * Check the child node/leaf by the following condition:
1358 * 1. the first item key of the node/leaf should be the same with the one
1360 * 2. block in parent node should match the child node/leaf.
1361 * 3. generation of parent node and child's header should be consistent.
1363 * Or the child node/leaf pointed by the key in parent is not valid.
1365 * We hope to check leaf owner too, but since subvol may share leaves,
1366 * which makes leaf owner check not so strong, key check should be
1367 * sufficient enough for that case.
1369 static int check_child_node(struct btrfs_root *root,
1370 struct extent_buffer *parent, int slot,
1371 struct extent_buffer *child)
1373 struct btrfs_key parent_key;
1374 struct btrfs_key child_key;
1377 btrfs_node_key_to_cpu(parent, &parent_key, slot);
1378 if (btrfs_header_level(child) == 0)
1379 btrfs_item_key_to_cpu(child, &child_key, 0);
1381 btrfs_node_key_to_cpu(child, &child_key, 0);
1383 if (memcmp(&parent_key, &child_key, sizeof(parent_key))) {
1386 "Wrong key of child node/leaf, wanted: (%llu, %u, %llu), have: (%llu, %u, %llu)\n",
1387 parent_key.objectid, parent_key.type, parent_key.offset,
1388 child_key.objectid, child_key.type, child_key.offset);
1390 if (btrfs_header_bytenr(child) != btrfs_node_blockptr(parent, slot)) {
1392 fprintf(stderr, "Wrong block of child node/leaf, wanted: %llu, have: %llu\n",
1393 btrfs_node_blockptr(parent, slot),
1394 btrfs_header_bytenr(child));
1396 if (btrfs_node_ptr_generation(parent, slot) !=
1397 btrfs_header_generation(child)) {
1399 fprintf(stderr, "Wrong generation of child node/leaf, wanted: %llu, have: %llu\n",
1400 btrfs_header_generation(child),
1401 btrfs_node_ptr_generation(parent, slot));
1406 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
1407 struct walk_control *wc, int *level)
1409 enum btrfs_tree_block_status status;
1412 struct extent_buffer *next;
1413 struct extent_buffer *cur;
1418 WARN_ON(*level < 0);
1419 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1420 ret = btrfs_lookup_extent_info(NULL, root,
1421 path->nodes[*level]->start,
1422 *level, 1, &refs, NULL);
1429 ret = enter_shared_node(root, path->nodes[*level]->start,
1437 while (*level >= 0) {
1438 WARN_ON(*level < 0);
1439 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1440 cur = path->nodes[*level];
1442 if (btrfs_header_level(cur) != *level)
1445 if (path->slots[*level] >= btrfs_header_nritems(cur))
1448 ret = process_one_leaf(root, cur, wc);
1453 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1454 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1455 blocksize = btrfs_level_size(root, *level - 1);
1456 ret = btrfs_lookup_extent_info(NULL, root, bytenr, *level - 1,
1462 ret = enter_shared_node(root, bytenr, refs,
1465 path->slots[*level]++;
1470 next = btrfs_find_tree_block(root, bytenr, blocksize);
1471 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
1472 free_extent_buffer(next);
1473 reada_walk_down(root, cur, path->slots[*level]);
1474 next = read_tree_block(root, bytenr, blocksize,
1482 ret = check_child_node(root, cur, path->slots[*level], next);
1488 if (btrfs_is_leaf(next))
1489 status = btrfs_check_leaf(root, NULL, next);
1491 status = btrfs_check_node(root, NULL, next);
1492 if (status != BTRFS_TREE_BLOCK_CLEAN) {
1493 free_extent_buffer(next);
1498 *level = *level - 1;
1499 free_extent_buffer(path->nodes[*level]);
1500 path->nodes[*level] = next;
1501 path->slots[*level] = 0;
1504 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
1508 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
1509 struct walk_control *wc, int *level)
1512 struct extent_buffer *leaf;
1514 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1515 leaf = path->nodes[i];
1516 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
1521 free_extent_buffer(path->nodes[*level]);
1522 path->nodes[*level] = NULL;
1523 BUG_ON(*level > wc->active_node);
1524 if (*level == wc->active_node)
1525 leave_shared_node(root, wc, *level);
1532 static int check_root_dir(struct inode_record *rec)
1534 struct inode_backref *backref;
1537 if (!rec->found_inode_item || rec->errors)
1539 if (rec->nlink != 1 || rec->found_link != 0)
1541 if (list_empty(&rec->backrefs))
1543 backref = list_entry(rec->backrefs.next, struct inode_backref, list);
1544 if (!backref->found_inode_ref)
1546 if (backref->index != 0 || backref->namelen != 2 ||
1547 memcmp(backref->name, "..", 2))
1549 if (backref->found_dir_index || backref->found_dir_item)
1556 static int repair_inode_isize(struct btrfs_trans_handle *trans,
1557 struct btrfs_root *root, struct btrfs_path *path,
1558 struct inode_record *rec)
1560 struct btrfs_inode_item *ei;
1561 struct btrfs_key key;
1564 key.objectid = rec->ino;
1565 key.type = BTRFS_INODE_ITEM_KEY;
1566 key.offset = (u64)-1;
1568 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1572 if (!path->slots[0]) {
1579 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1580 if (key.objectid != rec->ino) {
1585 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1586 struct btrfs_inode_item);
1587 btrfs_set_inode_size(path->nodes[0], ei, rec->found_size);
1588 btrfs_mark_buffer_dirty(path->nodes[0]);
1589 rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
1590 printf("reset isize for dir %Lu root %Lu\n", rec->ino,
1591 root->root_key.objectid);
1593 btrfs_release_path(path);
1597 static int repair_inode_orphan_item(struct btrfs_trans_handle *trans,
1598 struct btrfs_root *root,
1599 struct btrfs_path *path,
1600 struct inode_record *rec)
1602 struct btrfs_key key;
1605 key.objectid = BTRFS_ORPHAN_OBJECTID;
1606 key.type = BTRFS_ORPHAN_ITEM_KEY;
1607 key.offset = rec->ino;
1609 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1610 btrfs_release_path(path);
1612 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
1616 static int add_missing_dir_index(struct btrfs_root *root,
1617 struct cache_tree *inode_cache,
1618 struct inode_record *rec,
1619 struct inode_backref *backref)
1621 struct btrfs_path *path;
1622 struct btrfs_trans_handle *trans;
1623 struct btrfs_dir_item *dir_item;
1624 struct extent_buffer *leaf;
1625 struct btrfs_key key;
1626 struct btrfs_disk_key disk_key;
1627 struct inode_record *dir_rec;
1628 unsigned long name_ptr;
1629 u32 data_size = sizeof(*dir_item) + backref->namelen;
1632 path = btrfs_alloc_path();
1636 trans = btrfs_start_transaction(root, 1);
1637 if (IS_ERR(trans)) {
1638 btrfs_free_path(path);
1639 return PTR_ERR(trans);
1642 fprintf(stderr, "repairing missing dir index item for inode %llu\n",
1643 (unsigned long long)rec->ino);
1644 key.objectid = backref->dir;
1645 key.type = BTRFS_DIR_INDEX_KEY;
1646 key.offset = backref->index;
1648 ret = btrfs_insert_empty_item(trans, root, path, &key, data_size);
1651 leaf = path->nodes[0];
1652 dir_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
1654 disk_key.objectid = cpu_to_le64(rec->ino);
1655 disk_key.type = BTRFS_INODE_ITEM_KEY;
1656 disk_key.offset = 0;
1658 btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
1659 btrfs_set_dir_type(leaf, dir_item, imode_to_type(rec->imode));
1660 btrfs_set_dir_data_len(leaf, dir_item, 0);
1661 btrfs_set_dir_name_len(leaf, dir_item, backref->namelen);
1662 name_ptr = (unsigned long)(dir_item + 1);
1663 write_extent_buffer(leaf, backref->name, name_ptr, backref->namelen);
1664 btrfs_mark_buffer_dirty(leaf);
1665 btrfs_free_path(path);
1666 btrfs_commit_transaction(trans, root);
1668 backref->found_dir_index = 1;
1669 dir_rec = get_inode_rec(inode_cache, backref->dir, 0);
1672 dir_rec->found_size += backref->namelen;
1673 if (dir_rec->found_size == dir_rec->isize &&
1674 (dir_rec->errors & I_ERR_DIR_ISIZE_WRONG))
1675 dir_rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
1676 if (dir_rec->found_size != dir_rec->isize)
1677 dir_rec->errors |= I_ERR_DIR_ISIZE_WRONG;
1682 static int delete_dir_index(struct btrfs_root *root,
1683 struct cache_tree *inode_cache,
1684 struct inode_record *rec,
1685 struct inode_backref *backref)
1687 struct btrfs_trans_handle *trans;
1688 struct btrfs_dir_item *di;
1689 struct btrfs_path *path;
1692 path = btrfs_alloc_path();
1696 trans = btrfs_start_transaction(root, 1);
1697 if (IS_ERR(trans)) {
1698 btrfs_free_path(path);
1699 return PTR_ERR(trans);
1703 fprintf(stderr, "Deleting bad dir index [%llu,%u,%llu] root %llu\n",
1704 (unsigned long long)backref->dir,
1705 BTRFS_DIR_INDEX_KEY, (unsigned long long)backref->index,
1706 (unsigned long long)root->objectid);
1708 di = btrfs_lookup_dir_index(trans, root, path, backref->dir,
1709 backref->name, backref->namelen,
1710 backref->index, -1);
1713 btrfs_free_path(path);
1714 btrfs_commit_transaction(trans, root);
1721 ret = btrfs_del_item(trans, root, path);
1723 ret = btrfs_delete_one_dir_name(trans, root, path, di);
1725 btrfs_free_path(path);
1726 btrfs_commit_transaction(trans, root);
1730 static int create_inode_item(struct btrfs_root *root,
1731 struct inode_record *rec,
1732 struct inode_backref *backref, int root_dir)
1734 struct btrfs_trans_handle *trans;
1735 struct btrfs_inode_item inode_item;
1736 time_t now = time(NULL);
1739 trans = btrfs_start_transaction(root, 1);
1740 if (IS_ERR(trans)) {
1741 ret = PTR_ERR(trans);
1745 fprintf(stderr, "root %llu inode %llu recreating inode item, this may "
1746 "be incomplete, please check permissions and content after "
1747 "the fsck completes.\n", (unsigned long long)root->objectid,
1748 (unsigned long long)rec->ino);
1750 memset(&inode_item, 0, sizeof(inode_item));
1751 btrfs_set_stack_inode_generation(&inode_item, trans->transid);
1753 btrfs_set_stack_inode_nlink(&inode_item, 1);
1755 btrfs_set_stack_inode_nlink(&inode_item, rec->found_link);
1756 btrfs_set_stack_inode_nbytes(&inode_item, rec->found_size);
1757 if (rec->found_dir_item) {
1758 if (rec->found_file_extent)
1759 fprintf(stderr, "root %llu inode %llu has both a dir "
1760 "item and extents, unsure if it is a dir or a "
1761 "regular file so setting it as a directory\n",
1762 (unsigned long long)root->objectid,
1763 (unsigned long long)rec->ino);
1764 btrfs_set_stack_inode_mode(&inode_item, S_IFDIR | 0755);
1765 btrfs_set_stack_inode_size(&inode_item, rec->found_size);
1766 } else if (!rec->found_dir_item) {
1767 btrfs_set_stack_inode_size(&inode_item, rec->extent_end);
1768 btrfs_set_stack_inode_mode(&inode_item, S_IFREG | 0755);
1770 btrfs_set_stack_timespec_sec(&inode_item.atime, now);
1771 btrfs_set_stack_timespec_nsec(&inode_item.atime, 0);
1772 btrfs_set_stack_timespec_sec(&inode_item.ctime, now);
1773 btrfs_set_stack_timespec_nsec(&inode_item.ctime, 0);
1774 btrfs_set_stack_timespec_sec(&inode_item.mtime, now);
1775 btrfs_set_stack_timespec_nsec(&inode_item.mtime, 0);
1776 btrfs_set_stack_timespec_sec(&inode_item.otime, 0);
1777 btrfs_set_stack_timespec_nsec(&inode_item.otime, 0);
1779 ret = btrfs_insert_inode(trans, root, rec->ino, &inode_item);
1781 btrfs_commit_transaction(trans, root);
1785 static int repair_inode_backrefs(struct btrfs_root *root,
1786 struct inode_record *rec,
1787 struct cache_tree *inode_cache,
1790 struct inode_backref *tmp, *backref;
1791 u64 root_dirid = btrfs_root_dirid(&root->root_item);
1795 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
1796 if (!delete && rec->ino == root_dirid) {
1797 if (!rec->found_inode_item) {
1798 ret = create_inode_item(root, rec, backref, 1);
1805 /* Index 0 for root dir's are special, don't mess with it */
1806 if (rec->ino == root_dirid && backref->index == 0)
1810 ((backref->found_dir_index && !backref->found_inode_ref) ||
1811 (backref->found_dir_index && backref->found_inode_ref &&
1812 (backref->errors & REF_ERR_INDEX_UNMATCH)))) {
1813 ret = delete_dir_index(root, inode_cache, rec, backref);
1817 list_del(&backref->list);
1821 if (!delete && !backref->found_dir_index &&
1822 backref->found_dir_item && backref->found_inode_ref) {
1823 ret = add_missing_dir_index(root, inode_cache, rec,
1828 if (backref->found_dir_item &&
1829 backref->found_dir_index &&
1830 backref->found_dir_index) {
1831 if (!backref->errors &&
1832 backref->found_inode_ref) {
1833 list_del(&backref->list);
1839 if (!delete && (!backref->found_dir_index &&
1840 !backref->found_dir_item &&
1841 backref->found_inode_ref)) {
1842 struct btrfs_trans_handle *trans;
1843 struct btrfs_key location;
1845 location.objectid = rec->ino;
1846 location.type = BTRFS_INODE_ITEM_KEY;
1847 location.offset = 0;
1849 trans = btrfs_start_transaction(root, 1);
1850 if (IS_ERR(trans)) {
1851 ret = PTR_ERR(trans);
1854 fprintf(stderr, "adding missing dir index/item pair "
1856 (unsigned long long)rec->ino);
1857 ret = btrfs_insert_dir_item(trans, root, backref->name,
1859 backref->dir, &location,
1860 imode_to_type(rec->imode),
1863 btrfs_commit_transaction(trans, root);
1867 if (!delete && (backref->found_inode_ref &&
1868 backref->found_dir_index &&
1869 backref->found_dir_item &&
1870 !(backref->errors & REF_ERR_INDEX_UNMATCH) &&
1871 !rec->found_inode_item)) {
1872 ret = create_inode_item(root, rec, backref, 0);
1879 return ret ? ret : repaired;
1882 static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
1884 struct btrfs_trans_handle *trans;
1885 struct btrfs_path *path;
1888 if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG | I_ERR_NO_ORPHAN_ITEM)))
1891 path = btrfs_alloc_path();
1895 trans = btrfs_start_transaction(root, 1);
1896 if (IS_ERR(trans)) {
1897 btrfs_free_path(path);
1898 return PTR_ERR(trans);
1901 if (rec->errors & I_ERR_DIR_ISIZE_WRONG)
1902 ret = repair_inode_isize(trans, root, path, rec);
1903 if (!ret && rec->errors & I_ERR_NO_ORPHAN_ITEM)
1904 ret = repair_inode_orphan_item(trans, root, path, rec);
1905 btrfs_commit_transaction(trans, root);
1906 btrfs_free_path(path);
1910 static int check_inode_recs(struct btrfs_root *root,
1911 struct cache_tree *inode_cache)
1913 struct cache_extent *cache;
1914 struct ptr_node *node;
1915 struct inode_record *rec;
1916 struct inode_backref *backref;
1921 u64 root_dirid = btrfs_root_dirid(&root->root_item);
1923 if (btrfs_root_refs(&root->root_item) == 0) {
1924 if (!cache_tree_empty(inode_cache))
1925 fprintf(stderr, "warning line %d\n", __LINE__);
1930 * We need to repair backrefs first because we could change some of the
1931 * errors in the inode recs.
1933 * We also need to go through and delete invalid backrefs first and then
1934 * add the correct ones second. We do this because we may get EEXIST
1935 * when adding back the correct index because we hadn't yet deleted the
1938 * For example, if we were missing a dir index then the directories
1939 * isize would be wrong, so if we fixed the isize to what we thought it
1940 * would be and then fixed the backref we'd still have a invalid fs, so
1941 * we need to add back the dir index and then check to see if the isize
1946 if (stage == 3 && !err)
1949 cache = search_cache_extent(inode_cache, 0);
1950 while (repair && cache) {
1951 node = container_of(cache, struct ptr_node, cache);
1953 cache = next_cache_extent(cache);
1955 /* Need to free everything up and rescan */
1957 remove_cache_extent(inode_cache, &node->cache);
1959 free_inode_rec(rec);
1963 if (list_empty(&rec->backrefs))
1966 ret = repair_inode_backrefs(root, rec, inode_cache,
1980 rec = get_inode_rec(inode_cache, root_dirid, 0);
1982 ret = check_root_dir(rec);
1984 fprintf(stderr, "root %llu root dir %llu error\n",
1985 (unsigned long long)root->root_key.objectid,
1986 (unsigned long long)root_dirid);
1990 fprintf(stderr, "root %llu root dir %llu not found\n",
1991 (unsigned long long)root->root_key.objectid,
1992 (unsigned long long)root_dirid);
1996 cache = search_cache_extent(inode_cache, 0);
1999 node = container_of(cache, struct ptr_node, cache);
2001 remove_cache_extent(inode_cache, &node->cache);
2003 if (rec->ino == root_dirid ||
2004 rec->ino == BTRFS_ORPHAN_OBJECTID) {
2005 free_inode_rec(rec);
2009 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
2010 ret = check_orphan_item(root, rec->ino);
2012 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
2013 if (can_free_inode_rec(rec)) {
2014 free_inode_rec(rec);
2020 ret = try_repair_inode(root, rec);
2021 if (ret == 0 && can_free_inode_rec(rec)) {
2022 free_inode_rec(rec);
2029 if (!rec->found_inode_item)
2030 rec->errors |= I_ERR_NO_INODE_ITEM;
2031 if (rec->found_link != rec->nlink)
2032 rec->errors |= I_ERR_LINK_COUNT_WRONG;
2033 print_inode_error(root, rec);
2034 list_for_each_entry(backref, &rec->backrefs, list) {
2035 if (!backref->found_dir_item)
2036 backref->errors |= REF_ERR_NO_DIR_ITEM;
2037 if (!backref->found_dir_index)
2038 backref->errors |= REF_ERR_NO_DIR_INDEX;
2039 if (!backref->found_inode_ref)
2040 backref->errors |= REF_ERR_NO_INODE_REF;
2041 fprintf(stderr, "\tunresolved ref dir %llu index %llu"
2042 " namelen %u name %s filetype %d errors %x",
2043 (unsigned long long)backref->dir,
2044 (unsigned long long)backref->index,
2045 backref->namelen, backref->name,
2046 backref->filetype, backref->errors);
2047 print_ref_error(backref->errors);
2049 free_inode_rec(rec);
2051 return (error > 0) ? -1 : 0;
2054 static struct root_record *get_root_rec(struct cache_tree *root_cache,
2057 struct cache_extent *cache;
2058 struct root_record *rec = NULL;
2061 cache = lookup_cache_extent(root_cache, objectid, 1);
2063 rec = container_of(cache, struct root_record, cache);
2065 rec = calloc(1, sizeof(*rec));
2066 rec->objectid = objectid;
2067 INIT_LIST_HEAD(&rec->backrefs);
2068 rec->cache.start = objectid;
2069 rec->cache.size = 1;
2071 ret = insert_cache_extent(root_cache, &rec->cache);
2077 static struct root_backref *get_root_backref(struct root_record *rec,
2078 u64 ref_root, u64 dir, u64 index,
2079 const char *name, int namelen)
2081 struct root_backref *backref;
2083 list_for_each_entry(backref, &rec->backrefs, list) {
2084 if (backref->ref_root != ref_root || backref->dir != dir ||
2085 backref->namelen != namelen)
2087 if (memcmp(name, backref->name, namelen))
2092 backref = malloc(sizeof(*backref) + namelen + 1);
2093 memset(backref, 0, sizeof(*backref));
2094 backref->ref_root = ref_root;
2096 backref->index = index;
2097 backref->namelen = namelen;
2098 memcpy(backref->name, name, namelen);
2099 backref->name[namelen] = '\0';
2100 list_add_tail(&backref->list, &rec->backrefs);
2104 static void free_root_record(struct cache_extent *cache)
2106 struct root_record *rec;
2107 struct root_backref *backref;
2109 rec = container_of(cache, struct root_record, cache);
2110 while (!list_empty(&rec->backrefs)) {
2111 backref = list_entry(rec->backrefs.next,
2112 struct root_backref, list);
2113 list_del(&backref->list);
2120 FREE_EXTENT_CACHE_BASED_TREE(root_recs, free_root_record);
2122 static int add_root_backref(struct cache_tree *root_cache,
2123 u64 root_id, u64 ref_root, u64 dir, u64 index,
2124 const char *name, int namelen,
2125 int item_type, int errors)
2127 struct root_record *rec;
2128 struct root_backref *backref;
2130 rec = get_root_rec(root_cache, root_id);
2131 backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
2133 backref->errors |= errors;
2135 if (item_type != BTRFS_DIR_ITEM_KEY) {
2136 if (backref->found_dir_index || backref->found_back_ref ||
2137 backref->found_forward_ref) {
2138 if (backref->index != index)
2139 backref->errors |= REF_ERR_INDEX_UNMATCH;
2141 backref->index = index;
2145 if (item_type == BTRFS_DIR_ITEM_KEY) {
2146 if (backref->found_forward_ref)
2148 backref->found_dir_item = 1;
2149 } else if (item_type == BTRFS_DIR_INDEX_KEY) {
2150 backref->found_dir_index = 1;
2151 } else if (item_type == BTRFS_ROOT_REF_KEY) {
2152 if (backref->found_forward_ref)
2153 backref->errors |= REF_ERR_DUP_ROOT_REF;
2154 else if (backref->found_dir_item)
2156 backref->found_forward_ref = 1;
2157 } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
2158 if (backref->found_back_ref)
2159 backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
2160 backref->found_back_ref = 1;
2165 if (backref->found_forward_ref && backref->found_dir_item)
2166 backref->reachable = 1;
2170 static int merge_root_recs(struct btrfs_root *root,
2171 struct cache_tree *src_cache,
2172 struct cache_tree *dst_cache)
2174 struct cache_extent *cache;
2175 struct ptr_node *node;
2176 struct inode_record *rec;
2177 struct inode_backref *backref;
2180 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2181 free_inode_recs_tree(src_cache);
2186 cache = search_cache_extent(src_cache, 0);
2189 node = container_of(cache, struct ptr_node, cache);
2191 remove_cache_extent(src_cache, &node->cache);
2194 ret = is_child_root(root, root->objectid, rec->ino);
2200 list_for_each_entry(backref, &rec->backrefs, list) {
2201 BUG_ON(backref->found_inode_ref);
2202 if (backref->found_dir_item)
2203 add_root_backref(dst_cache, rec->ino,
2204 root->root_key.objectid, backref->dir,
2205 backref->index, backref->name,
2206 backref->namelen, BTRFS_DIR_ITEM_KEY,
2208 if (backref->found_dir_index)
2209 add_root_backref(dst_cache, rec->ino,
2210 root->root_key.objectid, backref->dir,
2211 backref->index, backref->name,
2212 backref->namelen, BTRFS_DIR_INDEX_KEY,
2216 free_inode_rec(rec);
2223 static int check_root_refs(struct btrfs_root *root,
2224 struct cache_tree *root_cache)
2226 struct root_record *rec;
2227 struct root_record *ref_root;
2228 struct root_backref *backref;
2229 struct cache_extent *cache;
2235 rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
2238 /* fixme: this can not detect circular references */
2241 cache = search_cache_extent(root_cache, 0);
2245 rec = container_of(cache, struct root_record, cache);
2246 cache = next_cache_extent(cache);
2248 if (rec->found_ref == 0)
2251 list_for_each_entry(backref, &rec->backrefs, list) {
2252 if (!backref->reachable)
2255 ref_root = get_root_rec(root_cache,
2257 if (ref_root->found_ref > 0)
2260 backref->reachable = 0;
2262 if (rec->found_ref == 0)
2268 cache = search_cache_extent(root_cache, 0);
2272 rec = container_of(cache, struct root_record, cache);
2273 cache = next_cache_extent(cache);
2275 if (rec->found_ref == 0 &&
2276 rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
2277 rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
2278 ret = check_orphan_item(root->fs_info->tree_root,
2284 * If we don't have a root item then we likely just have
2285 * a dir item in a snapshot for this root but no actual
2286 * ref key or anything so it's meaningless.
2288 if (!rec->found_root_item)
2291 fprintf(stderr, "fs tree %llu not referenced\n",
2292 (unsigned long long)rec->objectid);
2296 if (rec->found_ref > 0 && !rec->found_root_item)
2298 list_for_each_entry(backref, &rec->backrefs, list) {
2299 if (!backref->found_dir_item)
2300 backref->errors |= REF_ERR_NO_DIR_ITEM;
2301 if (!backref->found_dir_index)
2302 backref->errors |= REF_ERR_NO_DIR_INDEX;
2303 if (!backref->found_back_ref)
2304 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
2305 if (!backref->found_forward_ref)
2306 backref->errors |= REF_ERR_NO_ROOT_REF;
2307 if (backref->reachable && backref->errors)
2314 fprintf(stderr, "fs tree %llu refs %u %s\n",
2315 (unsigned long long)rec->objectid, rec->found_ref,
2316 rec->found_root_item ? "" : "not found");
2318 list_for_each_entry(backref, &rec->backrefs, list) {
2319 if (!backref->reachable)
2321 if (!backref->errors && rec->found_root_item)
2323 fprintf(stderr, "\tunresolved ref root %llu dir %llu"
2324 " index %llu namelen %u name %s errors %x\n",
2325 (unsigned long long)backref->ref_root,
2326 (unsigned long long)backref->dir,
2327 (unsigned long long)backref->index,
2328 backref->namelen, backref->name,
2330 print_ref_error(backref->errors);
2333 return errors > 0 ? 1 : 0;
2336 static int process_root_ref(struct extent_buffer *eb, int slot,
2337 struct btrfs_key *key,
2338 struct cache_tree *root_cache)
2344 struct btrfs_root_ref *ref;
2345 char namebuf[BTRFS_NAME_LEN];
2348 ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
2350 dirid = btrfs_root_ref_dirid(eb, ref);
2351 index = btrfs_root_ref_sequence(eb, ref);
2352 name_len = btrfs_root_ref_name_len(eb, ref);
2354 if (name_len <= BTRFS_NAME_LEN) {
2358 len = BTRFS_NAME_LEN;
2359 error = REF_ERR_NAME_TOO_LONG;
2361 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
2363 if (key->type == BTRFS_ROOT_REF_KEY) {
2364 add_root_backref(root_cache, key->offset, key->objectid, dirid,
2365 index, namebuf, len, key->type, error);
2367 add_root_backref(root_cache, key->objectid, key->offset, dirid,
2368 index, namebuf, len, key->type, error);
2373 static int check_fs_root(struct btrfs_root *root,
2374 struct cache_tree *root_cache,
2375 struct walk_control *wc)
2381 struct btrfs_path path;
2382 struct shared_node root_node;
2383 struct root_record *rec;
2384 struct btrfs_root_item *root_item = &root->root_item;
2385 enum btrfs_tree_block_status status;
2387 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2388 rec = get_root_rec(root_cache, root->root_key.objectid);
2389 if (btrfs_root_refs(root_item) > 0)
2390 rec->found_root_item = 1;
2393 btrfs_init_path(&path);
2394 memset(&root_node, 0, sizeof(root_node));
2395 cache_tree_init(&root_node.root_cache);
2396 cache_tree_init(&root_node.inode_cache);
2398 level = btrfs_header_level(root->node);
2399 memset(wc->nodes, 0, sizeof(wc->nodes));
2400 wc->nodes[level] = &root_node;
2401 wc->active_node = level;
2402 wc->root_level = level;
2404 /* We may not have checked the root block, lets do that now */
2405 if (btrfs_is_leaf(root->node))
2406 status = btrfs_check_leaf(root, NULL, root->node);
2408 status = btrfs_check_node(root, NULL, root->node);
2409 if (status != BTRFS_TREE_BLOCK_CLEAN)
2412 if (btrfs_root_refs(root_item) > 0 ||
2413 btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2414 path.nodes[level] = root->node;
2415 extent_buffer_get(root->node);
2416 path.slots[level] = 0;
2418 struct btrfs_key key;
2419 struct btrfs_disk_key found_key;
2421 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2422 level = root_item->drop_level;
2423 path.lowest_level = level;
2424 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
2427 btrfs_node_key(path.nodes[level], &found_key,
2429 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2430 sizeof(found_key)));
2434 wret = walk_down_tree(root, &path, wc, &level);
2440 wret = walk_up_tree(root, &path, wc, &level);
2447 btrfs_release_path(&path);
2449 err = merge_root_recs(root, &root_node.root_cache, root_cache);
2453 if (root_node.current) {
2454 root_node.current->checked = 1;
2455 maybe_free_inode_rec(&root_node.inode_cache,
2459 err = check_inode_recs(root, &root_node.inode_cache);
2465 static int fs_root_objectid(u64 objectid)
2467 if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
2468 objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2470 return is_fstree(objectid);
2473 static int check_fs_roots(struct btrfs_root *root,
2474 struct cache_tree *root_cache)
2476 struct btrfs_path path;
2477 struct btrfs_key key;
2478 struct walk_control wc;
2479 struct extent_buffer *leaf, *tree_node;
2480 struct btrfs_root *tmp_root;
2481 struct btrfs_root *tree_root = root->fs_info->tree_root;
2486 * Just in case we made any changes to the extent tree that weren't
2487 * reflected into the free space cache yet.
2490 reset_cached_block_groups(root->fs_info);
2491 memset(&wc, 0, sizeof(wc));
2492 cache_tree_init(&wc.shared);
2493 btrfs_init_path(&path);
2498 key.type = BTRFS_ROOT_ITEM_KEY;
2499 ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
2504 tree_node = tree_root->node;
2506 if (tree_node != tree_root->node) {
2507 free_root_recs_tree(root_cache);
2508 btrfs_release_path(&path);
2511 leaf = path.nodes[0];
2512 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
2513 ret = btrfs_next_leaf(tree_root, &path);
2519 leaf = path.nodes[0];
2521 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
2522 if (key.type == BTRFS_ROOT_ITEM_KEY &&
2523 fs_root_objectid(key.objectid)) {
2524 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2525 tmp_root = btrfs_read_fs_root_no_cache(
2526 root->fs_info, &key);
2528 key.offset = (u64)-1;
2529 tmp_root = btrfs_read_fs_root(
2530 root->fs_info, &key);
2532 if (IS_ERR(tmp_root)) {
2536 ret = check_fs_root(tmp_root, root_cache, &wc);
2537 if (ret == -EAGAIN) {
2538 free_root_recs_tree(root_cache);
2539 btrfs_release_path(&path);
2544 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
2545 btrfs_free_fs_root(tmp_root);
2546 } else if (key.type == BTRFS_ROOT_REF_KEY ||
2547 key.type == BTRFS_ROOT_BACKREF_KEY) {
2548 process_root_ref(leaf, path.slots[0], &key,
2555 btrfs_release_path(&path);
2557 free_extent_cache_tree(&wc.shared);
2558 if (!cache_tree_empty(&wc.shared))
2559 fprintf(stderr, "warning line %d\n", __LINE__);
2564 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
2566 struct list_head *cur = rec->backrefs.next;
2567 struct extent_backref *back;
2568 struct tree_backref *tback;
2569 struct data_backref *dback;
2573 while(cur != &rec->backrefs) {
2574 back = list_entry(cur, struct extent_backref, list);
2576 if (!back->found_extent_tree) {
2580 if (back->is_data) {
2581 dback = (struct data_backref *)back;
2582 fprintf(stderr, "Backref %llu %s %llu"
2583 " owner %llu offset %llu num_refs %lu"
2584 " not found in extent tree\n",
2585 (unsigned long long)rec->start,
2586 back->full_backref ?
2588 back->full_backref ?
2589 (unsigned long long)dback->parent:
2590 (unsigned long long)dback->root,
2591 (unsigned long long)dback->owner,
2592 (unsigned long long)dback->offset,
2593 (unsigned long)dback->num_refs);
2595 tback = (struct tree_backref *)back;
2596 fprintf(stderr, "Backref %llu parent %llu"
2597 " root %llu not found in extent tree\n",
2598 (unsigned long long)rec->start,
2599 (unsigned long long)tback->parent,
2600 (unsigned long long)tback->root);
2603 if (!back->is_data && !back->found_ref) {
2607 tback = (struct tree_backref *)back;
2608 fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
2609 (unsigned long long)rec->start,
2610 back->full_backref ? "parent" : "root",
2611 back->full_backref ?
2612 (unsigned long long)tback->parent :
2613 (unsigned long long)tback->root, back);
2615 if (back->is_data) {
2616 dback = (struct data_backref *)back;
2617 if (dback->found_ref != dback->num_refs) {
2621 fprintf(stderr, "Incorrect local backref count"
2622 " on %llu %s %llu owner %llu"
2623 " offset %llu found %u wanted %u back %p\n",
2624 (unsigned long long)rec->start,
2625 back->full_backref ?
2627 back->full_backref ?
2628 (unsigned long long)dback->parent:
2629 (unsigned long long)dback->root,
2630 (unsigned long long)dback->owner,
2631 (unsigned long long)dback->offset,
2632 dback->found_ref, dback->num_refs, back);
2634 if (dback->disk_bytenr != rec->start) {
2638 fprintf(stderr, "Backref disk bytenr does not"
2639 " match extent record, bytenr=%llu, "
2640 "ref bytenr=%llu\n",
2641 (unsigned long long)rec->start,
2642 (unsigned long long)dback->disk_bytenr);
2645 if (dback->bytes != rec->nr) {
2649 fprintf(stderr, "Backref bytes do not match "
2650 "extent backref, bytenr=%llu, ref "
2651 "bytes=%llu, backref bytes=%llu\n",
2652 (unsigned long long)rec->start,
2653 (unsigned long long)rec->nr,
2654 (unsigned long long)dback->bytes);
2657 if (!back->is_data) {
2660 dback = (struct data_backref *)back;
2661 found += dback->found_ref;
2664 if (found != rec->refs) {
2668 fprintf(stderr, "Incorrect global backref count "
2669 "on %llu found %llu wanted %llu\n",
2670 (unsigned long long)rec->start,
2671 (unsigned long long)found,
2672 (unsigned long long)rec->refs);
2678 static int free_all_extent_backrefs(struct extent_record *rec)
2680 struct extent_backref *back;
2681 struct list_head *cur;
2682 while (!list_empty(&rec->backrefs)) {
2683 cur = rec->backrefs.next;
2684 back = list_entry(cur, struct extent_backref, list);
2691 static void free_extent_record_cache(struct btrfs_fs_info *fs_info,
2692 struct cache_tree *extent_cache)
2694 struct cache_extent *cache;
2695 struct extent_record *rec;
2698 cache = first_cache_extent(extent_cache);
2701 rec = container_of(cache, struct extent_record, cache);
2702 btrfs_unpin_extent(fs_info, rec->start, rec->max_size);
2703 remove_cache_extent(extent_cache, cache);
2704 free_all_extent_backrefs(rec);
2709 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
2710 struct extent_record *rec)
2712 if (rec->content_checked && rec->owner_ref_checked &&
2713 rec->extent_item_refs == rec->refs && rec->refs > 0 &&
2714 rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0)) {
2715 remove_cache_extent(extent_cache, &rec->cache);
2716 free_all_extent_backrefs(rec);
2717 list_del_init(&rec->list);
2723 static int check_owner_ref(struct btrfs_root *root,
2724 struct extent_record *rec,
2725 struct extent_buffer *buf)
2727 struct extent_backref *node;
2728 struct tree_backref *back;
2729 struct btrfs_root *ref_root;
2730 struct btrfs_key key;
2731 struct btrfs_path path;
2732 struct extent_buffer *parent;
2737 list_for_each_entry(node, &rec->backrefs, list) {
2740 if (!node->found_ref)
2742 if (node->full_backref)
2744 back = (struct tree_backref *)node;
2745 if (btrfs_header_owner(buf) == back->root)
2748 BUG_ON(rec->is_root);
2750 /* try to find the block by search corresponding fs tree */
2751 key.objectid = btrfs_header_owner(buf);
2752 key.type = BTRFS_ROOT_ITEM_KEY;
2753 key.offset = (u64)-1;
2755 ref_root = btrfs_read_fs_root(root->fs_info, &key);
2756 if (IS_ERR(ref_root))
2759 level = btrfs_header_level(buf);
2761 btrfs_item_key_to_cpu(buf, &key, 0);
2763 btrfs_node_key_to_cpu(buf, &key, 0);
2765 btrfs_init_path(&path);
2766 path.lowest_level = level + 1;
2767 ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
2771 parent = path.nodes[level + 1];
2772 if (parent && buf->start == btrfs_node_blockptr(parent,
2773 path.slots[level + 1]))
2776 btrfs_release_path(&path);
2777 return found ? 0 : 1;
2780 static int is_extent_tree_record(struct extent_record *rec)
2782 struct list_head *cur = rec->backrefs.next;
2783 struct extent_backref *node;
2784 struct tree_backref *back;
2787 while(cur != &rec->backrefs) {
2788 node = list_entry(cur, struct extent_backref, list);
2792 back = (struct tree_backref *)node;
2793 if (node->full_backref)
2795 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
2802 static int record_bad_block_io(struct btrfs_fs_info *info,
2803 struct cache_tree *extent_cache,
2806 struct extent_record *rec;
2807 struct cache_extent *cache;
2808 struct btrfs_key key;
2810 cache = lookup_cache_extent(extent_cache, start, len);
2814 rec = container_of(cache, struct extent_record, cache);
2815 if (!is_extent_tree_record(rec))
2818 btrfs_disk_key_to_cpu(&key, &rec->parent_key);
2819 return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
2822 static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
2823 struct extent_buffer *buf, int slot)
2825 if (btrfs_header_level(buf)) {
2826 struct btrfs_key_ptr ptr1, ptr2;
2828 read_extent_buffer(buf, &ptr1, btrfs_node_key_ptr_offset(slot),
2829 sizeof(struct btrfs_key_ptr));
2830 read_extent_buffer(buf, &ptr2,
2831 btrfs_node_key_ptr_offset(slot + 1),
2832 sizeof(struct btrfs_key_ptr));
2833 write_extent_buffer(buf, &ptr1,
2834 btrfs_node_key_ptr_offset(slot + 1),
2835 sizeof(struct btrfs_key_ptr));
2836 write_extent_buffer(buf, &ptr2,
2837 btrfs_node_key_ptr_offset(slot),
2838 sizeof(struct btrfs_key_ptr));
2840 struct btrfs_disk_key key;
2841 btrfs_node_key(buf, &key, 0);
2842 btrfs_fixup_low_keys(root, path, &key,
2843 btrfs_header_level(buf) + 1);
2846 struct btrfs_item *item1, *item2;
2847 struct btrfs_key k1, k2;
2848 char *item1_data, *item2_data;
2849 u32 item1_offset, item2_offset, item1_size, item2_size;
2851 item1 = btrfs_item_nr(slot);
2852 item2 = btrfs_item_nr(slot + 1);
2853 btrfs_item_key_to_cpu(buf, &k1, slot);
2854 btrfs_item_key_to_cpu(buf, &k2, slot + 1);
2855 item1_offset = btrfs_item_offset(buf, item1);
2856 item2_offset = btrfs_item_offset(buf, item2);
2857 item1_size = btrfs_item_size(buf, item1);
2858 item2_size = btrfs_item_size(buf, item2);
2860 item1_data = malloc(item1_size);
2863 item2_data = malloc(item2_size);
2869 read_extent_buffer(buf, item1_data, item1_offset, item1_size);
2870 read_extent_buffer(buf, item2_data, item2_offset, item2_size);
2872 write_extent_buffer(buf, item1_data, item2_offset, item2_size);
2873 write_extent_buffer(buf, item2_data, item1_offset, item1_size);
2877 btrfs_set_item_offset(buf, item1, item2_offset);
2878 btrfs_set_item_offset(buf, item2, item1_offset);
2879 btrfs_set_item_size(buf, item1, item2_size);
2880 btrfs_set_item_size(buf, item2, item1_size);
2882 path->slots[0] = slot;
2883 btrfs_set_item_key_unsafe(root, path, &k2);
2884 path->slots[0] = slot + 1;
2885 btrfs_set_item_key_unsafe(root, path, &k1);
2890 static int fix_key_order(struct btrfs_trans_handle *trans,
2891 struct btrfs_root *root,
2892 struct btrfs_path *path)
2894 struct extent_buffer *buf;
2895 struct btrfs_key k1, k2;
2897 int level = path->lowest_level;
2900 buf = path->nodes[level];
2901 for (i = 0; i < btrfs_header_nritems(buf) - 1; i++) {
2903 btrfs_node_key_to_cpu(buf, &k1, i);
2904 btrfs_node_key_to_cpu(buf, &k2, i + 1);
2906 btrfs_item_key_to_cpu(buf, &k1, i);
2907 btrfs_item_key_to_cpu(buf, &k2, i + 1);
2909 if (btrfs_comp_cpu_keys(&k1, &k2) < 0)
2911 ret = swap_values(root, path, buf, i);
2914 btrfs_mark_buffer_dirty(buf);
2920 static int delete_bogus_item(struct btrfs_trans_handle *trans,
2921 struct btrfs_root *root,
2922 struct btrfs_path *path,
2923 struct extent_buffer *buf, int slot)
2925 struct btrfs_key key;
2926 int nritems = btrfs_header_nritems(buf);
2928 btrfs_item_key_to_cpu(buf, &key, slot);
2930 /* These are all the keys we can deal with missing. */
2931 if (key.type != BTRFS_DIR_INDEX_KEY &&
2932 key.type != BTRFS_EXTENT_ITEM_KEY &&
2933 key.type != BTRFS_METADATA_ITEM_KEY &&
2934 key.type != BTRFS_TREE_BLOCK_REF_KEY &&
2935 key.type != BTRFS_EXTENT_DATA_REF_KEY)
2938 printf("Deleting bogus item [%llu,%u,%llu] at slot %d on block %llu\n",
2939 (unsigned long long)key.objectid, key.type,
2940 (unsigned long long)key.offset, slot, buf->start);
2941 memmove_extent_buffer(buf, btrfs_item_nr_offset(slot),
2942 btrfs_item_nr_offset(slot + 1),
2943 sizeof(struct btrfs_item) *
2944 (nritems - slot - 1));
2945 btrfs_set_header_nritems(buf, nritems - 1);
2947 struct btrfs_disk_key disk_key;
2949 btrfs_item_key(buf, &disk_key, 0);
2950 btrfs_fixup_low_keys(root, path, &disk_key, 1);
2952 btrfs_mark_buffer_dirty(buf);
2956 static int fix_item_offset(struct btrfs_trans_handle *trans,
2957 struct btrfs_root *root,
2958 struct btrfs_path *path)
2960 struct extent_buffer *buf;
2964 /* We should only get this for leaves */
2965 BUG_ON(path->lowest_level);
2966 buf = path->nodes[0];
2968 for (i = 0; i < btrfs_header_nritems(buf); i++) {
2969 unsigned int shift = 0, offset;
2971 if (i == 0 && btrfs_item_end_nr(buf, i) !=
2972 BTRFS_LEAF_DATA_SIZE(root)) {
2973 if (btrfs_item_end_nr(buf, i) >
2974 BTRFS_LEAF_DATA_SIZE(root)) {
2975 ret = delete_bogus_item(trans, root, path,
2979 fprintf(stderr, "item is off the end of the "
2980 "leaf, can't fix\n");
2984 shift = BTRFS_LEAF_DATA_SIZE(root) -
2985 btrfs_item_end_nr(buf, i);
2986 } else if (i > 0 && btrfs_item_end_nr(buf, i) !=
2987 btrfs_item_offset_nr(buf, i - 1)) {
2988 if (btrfs_item_end_nr(buf, i) >
2989 btrfs_item_offset_nr(buf, i - 1)) {
2990 ret = delete_bogus_item(trans, root, path,
2994 fprintf(stderr, "items overlap, can't fix\n");
2998 shift = btrfs_item_offset_nr(buf, i - 1) -
2999 btrfs_item_end_nr(buf, i);
3004 printf("Shifting item nr %d by %u bytes in block %llu\n",
3005 i, shift, (unsigned long long)buf->start);
3006 offset = btrfs_item_offset_nr(buf, i);
3007 memmove_extent_buffer(buf,
3008 btrfs_leaf_data(buf) + offset + shift,
3009 btrfs_leaf_data(buf) + offset,
3010 btrfs_item_size_nr(buf, i));
3011 btrfs_set_item_offset(buf, btrfs_item_nr(i),
3013 btrfs_mark_buffer_dirty(buf);
3017 * We may have moved things, in which case we want to exit so we don't
3018 * write those changes out. Once we have proper abort functionality in
3019 * progs this can be changed to something nicer.
3026 * Attempt to fix basic block failures. If we can't fix it for whatever reason
3027 * then just return -EIO.
3029 static int try_to_fix_bad_block(struct btrfs_trans_handle *trans,
3030 struct btrfs_root *root,
3031 struct extent_buffer *buf,
3032 enum btrfs_tree_block_status status)
3034 struct ulist *roots;
3035 struct ulist_node *node;
3036 struct btrfs_root *search_root;
3037 struct btrfs_path *path;
3038 struct ulist_iterator iter;
3039 struct btrfs_key root_key, key;
3042 if (status != BTRFS_TREE_BLOCK_BAD_KEY_ORDER &&
3043 status != BTRFS_TREE_BLOCK_INVALID_OFFSETS)
3046 path = btrfs_alloc_path();
3050 ret = btrfs_find_all_roots(trans, root->fs_info, buf->start,
3053 btrfs_free_path(path);
3057 ULIST_ITER_INIT(&iter);
3058 while ((node = ulist_next(roots, &iter))) {
3059 root_key.objectid = node->val;
3060 root_key.type = BTRFS_ROOT_ITEM_KEY;
3061 root_key.offset = (u64)-1;
3063 search_root = btrfs_read_fs_root(root->fs_info, &root_key);
3069 record_root_in_trans(trans, search_root);
3071 path->lowest_level = btrfs_header_level(buf);
3072 path->skip_check_block = 1;
3073 if (path->lowest_level)
3074 btrfs_node_key_to_cpu(buf, &key, 0);
3076 btrfs_item_key_to_cpu(buf, &key, 0);
3077 ret = btrfs_search_slot(trans, search_root, &key, path, 0, 1);
3082 if (status == BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
3083 ret = fix_key_order(trans, search_root, path);
3084 else if (status == BTRFS_TREE_BLOCK_INVALID_OFFSETS)
3085 ret = fix_item_offset(trans, search_root, path);
3088 btrfs_release_path(path);
3091 btrfs_free_path(path);
3095 static int check_block(struct btrfs_trans_handle *trans,
3096 struct btrfs_root *root,
3097 struct cache_tree *extent_cache,
3098 struct extent_buffer *buf, u64 flags)
3100 struct extent_record *rec;
3101 struct cache_extent *cache;
3102 struct btrfs_key key;
3103 enum btrfs_tree_block_status status;
3107 cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
3110 rec = container_of(cache, struct extent_record, cache);
3111 rec->generation = btrfs_header_generation(buf);
3113 level = btrfs_header_level(buf);
3114 if (btrfs_header_nritems(buf) > 0) {
3117 btrfs_item_key_to_cpu(buf, &key, 0);
3119 btrfs_node_key_to_cpu(buf, &key, 0);
3121 rec->info_objectid = key.objectid;
3123 rec->info_level = level;
3125 if (btrfs_is_leaf(buf))
3126 status = btrfs_check_leaf(root, &rec->parent_key, buf);
3128 status = btrfs_check_node(root, &rec->parent_key, buf);
3130 if (status != BTRFS_TREE_BLOCK_CLEAN) {
3132 status = try_to_fix_bad_block(trans, root, buf,
3134 if (status != BTRFS_TREE_BLOCK_CLEAN) {
3136 fprintf(stderr, "bad block %llu\n",
3137 (unsigned long long)buf->start);
3140 * Signal to callers we need to start the scan over
3141 * again since we'll have cow'ed blocks.
3146 rec->content_checked = 1;
3147 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3148 rec->owner_ref_checked = 1;
3150 ret = check_owner_ref(root, rec, buf);
3152 rec->owner_ref_checked = 1;
3156 maybe_free_extent_rec(extent_cache, rec);
3160 static struct tree_backref *find_tree_backref(struct extent_record *rec,
3161 u64 parent, u64 root)
3163 struct list_head *cur = rec->backrefs.next;
3164 struct extent_backref *node;
3165 struct tree_backref *back;
3167 while(cur != &rec->backrefs) {
3168 node = list_entry(cur, struct extent_backref, list);
3172 back = (struct tree_backref *)node;
3174 if (!node->full_backref)
3176 if (parent == back->parent)
3179 if (node->full_backref)
3181 if (back->root == root)
3188 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
3189 u64 parent, u64 root)
3191 struct tree_backref *ref = malloc(sizeof(*ref));
3192 memset(&ref->node, 0, sizeof(ref->node));
3194 ref->parent = parent;
3195 ref->node.full_backref = 1;
3198 ref->node.full_backref = 0;
3200 list_add_tail(&ref->node.list, &rec->backrefs);
3205 static struct data_backref *find_data_backref(struct extent_record *rec,
3206 u64 parent, u64 root,
3207 u64 owner, u64 offset,
3209 u64 disk_bytenr, u64 bytes)
3211 struct list_head *cur = rec->backrefs.next;
3212 struct extent_backref *node;
3213 struct data_backref *back;
3215 while(cur != &rec->backrefs) {
3216 node = list_entry(cur, struct extent_backref, list);
3220 back = (struct data_backref *)node;
3222 if (!node->full_backref)
3224 if (parent == back->parent)
3227 if (node->full_backref)
3229 if (back->root == root && back->owner == owner &&
3230 back->offset == offset) {
3231 if (found_ref && node->found_ref &&
3232 (back->bytes != bytes ||
3233 back->disk_bytenr != disk_bytenr))
3242 static struct data_backref *alloc_data_backref(struct extent_record *rec,
3243 u64 parent, u64 root,
3244 u64 owner, u64 offset,
3247 struct data_backref *ref = malloc(sizeof(*ref));
3248 memset(&ref->node, 0, sizeof(ref->node));
3249 ref->node.is_data = 1;
3252 ref->parent = parent;
3255 ref->node.full_backref = 1;
3259 ref->offset = offset;
3260 ref->node.full_backref = 0;
3262 ref->bytes = max_size;
3265 list_add_tail(&ref->node.list, &rec->backrefs);
3266 if (max_size > rec->max_size)
3267 rec->max_size = max_size;
3271 static int add_extent_rec(struct cache_tree *extent_cache,
3272 struct btrfs_key *parent_key, u64 parent_gen,
3273 u64 start, u64 nr, u64 extent_item_refs,
3274 int is_root, int inc_ref, int set_checked,
3275 int metadata, int extent_rec, u64 max_size)
3277 struct extent_record *rec;
3278 struct cache_extent *cache;
3282 cache = lookup_cache_extent(extent_cache, start, nr);
3284 rec = container_of(cache, struct extent_record, cache);
3288 rec->nr = max(nr, max_size);
3291 * We need to make sure to reset nr to whatever the extent
3292 * record says was the real size, this way we can compare it to
3296 if (start != rec->start || rec->found_rec) {
3297 struct extent_record *tmp;
3300 if (list_empty(&rec->list))
3301 list_add_tail(&rec->list,
3302 &duplicate_extents);
3305 * We have to do this song and dance in case we
3306 * find an extent record that falls inside of
3307 * our current extent record but does not have
3308 * the same objectid.
3310 tmp = malloc(sizeof(*tmp));
3314 tmp->max_size = max_size;
3317 tmp->metadata = metadata;
3318 tmp->extent_item_refs = extent_item_refs;
3319 INIT_LIST_HEAD(&tmp->list);
3320 list_add_tail(&tmp->list, &rec->dups);
3321 rec->num_duplicates++;
3328 if (extent_item_refs && !dup) {
3329 if (rec->extent_item_refs) {
3330 fprintf(stderr, "block %llu rec "
3331 "extent_item_refs %llu, passed %llu\n",
3332 (unsigned long long)start,
3333 (unsigned long long)
3334 rec->extent_item_refs,
3335 (unsigned long long)extent_item_refs);
3337 rec->extent_item_refs = extent_item_refs;
3342 rec->content_checked = 1;
3343 rec->owner_ref_checked = 1;
3347 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
3349 rec->parent_generation = parent_gen;
3351 if (rec->max_size < max_size)
3352 rec->max_size = max_size;
3354 maybe_free_extent_rec(extent_cache, rec);
3357 rec = malloc(sizeof(*rec));
3359 rec->max_size = max_size;
3360 rec->nr = max(nr, max_size);
3361 rec->found_rec = !!extent_rec;
3362 rec->content_checked = 0;
3363 rec->owner_ref_checked = 0;
3364 rec->num_duplicates = 0;
3365 rec->metadata = metadata;
3366 INIT_LIST_HEAD(&rec->backrefs);
3367 INIT_LIST_HEAD(&rec->dups);
3368 INIT_LIST_HEAD(&rec->list);
3380 if (extent_item_refs)
3381 rec->extent_item_refs = extent_item_refs;
3383 rec->extent_item_refs = 0;
3386 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
3388 memset(&rec->parent_key, 0, sizeof(*parent_key));
3391 rec->parent_generation = parent_gen;
3393 rec->parent_generation = 0;
3395 rec->cache.start = start;
3396 rec->cache.size = nr;
3397 ret = insert_cache_extent(extent_cache, &rec->cache);
3401 rec->content_checked = 1;
3402 rec->owner_ref_checked = 1;
3407 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
3408 u64 parent, u64 root, int found_ref)
3410 struct extent_record *rec;
3411 struct tree_backref *back;
3412 struct cache_extent *cache;
3414 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3416 add_extent_rec(extent_cache, NULL, 0, bytenr,
3417 1, 0, 0, 0, 0, 1, 0, 0);
3418 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3423 rec = container_of(cache, struct extent_record, cache);
3424 if (rec->start != bytenr) {
3428 back = find_tree_backref(rec, parent, root);
3430 back = alloc_tree_backref(rec, parent, root);
3433 if (back->node.found_ref) {
3434 fprintf(stderr, "Extent back ref already exists "
3435 "for %llu parent %llu root %llu \n",
3436 (unsigned long long)bytenr,
3437 (unsigned long long)parent,
3438 (unsigned long long)root);
3440 back->node.found_ref = 1;
3442 if (back->node.found_extent_tree) {
3443 fprintf(stderr, "Extent back ref already exists "
3444 "for %llu parent %llu root %llu \n",
3445 (unsigned long long)bytenr,
3446 (unsigned long long)parent,
3447 (unsigned long long)root);
3449 back->node.found_extent_tree = 1;
3451 maybe_free_extent_rec(extent_cache, rec);
3455 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
3456 u64 parent, u64 root, u64 owner, u64 offset,
3457 u32 num_refs, int found_ref, u64 max_size)
3459 struct extent_record *rec;
3460 struct data_backref *back;
3461 struct cache_extent *cache;
3463 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3465 add_extent_rec(extent_cache, NULL, 0, bytenr, 1, 0, 0, 0, 0,
3467 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3472 rec = container_of(cache, struct extent_record, cache);
3473 if (rec->max_size < max_size)
3474 rec->max_size = max_size;
3477 * If found_ref is set then max_size is the real size and must match the
3478 * existing refs. So if we have already found a ref then we need to
3479 * make sure that this ref matches the existing one, otherwise we need
3480 * to add a new backref so we can notice that the backrefs don't match
3481 * and we need to figure out who is telling the truth. This is to
3482 * account for that awful fsync bug I introduced where we'd end up with
3483 * a btrfs_file_extent_item that would have its length include multiple
3484 * prealloc extents or point inside of a prealloc extent.
3486 back = find_data_backref(rec, parent, root, owner, offset, found_ref,
3489 back = alloc_data_backref(rec, parent, root, owner, offset,
3493 BUG_ON(num_refs != 1);
3494 if (back->node.found_ref)
3495 BUG_ON(back->bytes != max_size);
3496 back->node.found_ref = 1;
3497 back->found_ref += 1;
3498 back->bytes = max_size;
3499 back->disk_bytenr = bytenr;
3501 rec->content_checked = 1;
3502 rec->owner_ref_checked = 1;
3504 if (back->node.found_extent_tree) {
3505 fprintf(stderr, "Extent back ref already exists "
3506 "for %llu parent %llu root %llu "
3507 "owner %llu offset %llu num_refs %lu\n",
3508 (unsigned long long)bytenr,
3509 (unsigned long long)parent,
3510 (unsigned long long)root,
3511 (unsigned long long)owner,
3512 (unsigned long long)offset,
3513 (unsigned long)num_refs);
3515 back->num_refs = num_refs;
3516 back->node.found_extent_tree = 1;
3518 maybe_free_extent_rec(extent_cache, rec);
3522 static int add_pending(struct cache_tree *pending,
3523 struct cache_tree *seen, u64 bytenr, u32 size)
3526 ret = add_cache_extent(seen, bytenr, size);
3529 add_cache_extent(pending, bytenr, size);
3533 static int pick_next_pending(struct cache_tree *pending,
3534 struct cache_tree *reada,
3535 struct cache_tree *nodes,
3536 u64 last, struct block_info *bits, int bits_nr,
3539 unsigned long node_start = last;
3540 struct cache_extent *cache;
3543 cache = search_cache_extent(reada, 0);
3545 bits[0].start = cache->start;
3546 bits[0].size = cache->size;
3551 if (node_start > 32768)
3552 node_start -= 32768;
3554 cache = search_cache_extent(nodes, node_start);
3556 cache = search_cache_extent(nodes, 0);
3559 cache = search_cache_extent(pending, 0);
3564 bits[ret].start = cache->start;
3565 bits[ret].size = cache->size;
3566 cache = next_cache_extent(cache);
3568 } while (cache && ret < bits_nr);
3574 bits[ret].start = cache->start;
3575 bits[ret].size = cache->size;
3576 cache = next_cache_extent(cache);
3578 } while (cache && ret < bits_nr);
3580 if (bits_nr - ret > 8) {
3581 u64 lookup = bits[0].start + bits[0].size;
3582 struct cache_extent *next;
3583 next = search_cache_extent(pending, lookup);
3585 if (next->start - lookup > 32768)
3587 bits[ret].start = next->start;
3588 bits[ret].size = next->size;
3589 lookup = next->start + next->size;
3593 next = next_cache_extent(next);
3601 static void free_chunk_record(struct cache_extent *cache)
3603 struct chunk_record *rec;
3605 rec = container_of(cache, struct chunk_record, cache);
3606 list_del_init(&rec->list);
3607 list_del_init(&rec->dextents);
3611 void free_chunk_cache_tree(struct cache_tree *chunk_cache)
3613 cache_tree_free_extents(chunk_cache, free_chunk_record);
3616 static void free_device_record(struct rb_node *node)
3618 struct device_record *rec;
3620 rec = container_of(node, struct device_record, node);
3624 FREE_RB_BASED_TREE(device_cache, free_device_record);
3626 int insert_block_group_record(struct block_group_tree *tree,
3627 struct block_group_record *bg_rec)
3631 ret = insert_cache_extent(&tree->tree, &bg_rec->cache);
3635 list_add_tail(&bg_rec->list, &tree->block_groups);
3639 static void free_block_group_record(struct cache_extent *cache)
3641 struct block_group_record *rec;
3643 rec = container_of(cache, struct block_group_record, cache);
3644 list_del_init(&rec->list);
3648 void free_block_group_tree(struct block_group_tree *tree)
3650 cache_tree_free_extents(&tree->tree, free_block_group_record);
3653 int insert_device_extent_record(struct device_extent_tree *tree,
3654 struct device_extent_record *de_rec)
3659 * Device extent is a bit different from the other extents, because
3660 * the extents which belong to the different devices may have the
3661 * same start and size, so we need use the special extent cache
3662 * search/insert functions.
3664 ret = insert_cache_extent2(&tree->tree, &de_rec->cache);
3668 list_add_tail(&de_rec->chunk_list, &tree->no_chunk_orphans);
3669 list_add_tail(&de_rec->device_list, &tree->no_device_orphans);
3673 static void free_device_extent_record(struct cache_extent *cache)
3675 struct device_extent_record *rec;
3677 rec = container_of(cache, struct device_extent_record, cache);
3678 if (!list_empty(&rec->chunk_list))
3679 list_del_init(&rec->chunk_list);
3680 if (!list_empty(&rec->device_list))
3681 list_del_init(&rec->device_list);
3685 void free_device_extent_tree(struct device_extent_tree *tree)
3687 cache_tree_free_extents(&tree->tree, free_device_extent_record);
3690 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3691 static int process_extent_ref_v0(struct cache_tree *extent_cache,
3692 struct extent_buffer *leaf, int slot)
3694 struct btrfs_extent_ref_v0 *ref0;
3695 struct btrfs_key key;
3697 btrfs_item_key_to_cpu(leaf, &key, slot);
3698 ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
3699 if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
3700 add_tree_backref(extent_cache, key.objectid, key.offset, 0, 0);
3702 add_data_backref(extent_cache, key.objectid, key.offset, 0,
3703 0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
3709 struct chunk_record *btrfs_new_chunk_record(struct extent_buffer *leaf,
3710 struct btrfs_key *key,
3713 struct btrfs_chunk *ptr;
3714 struct chunk_record *rec;
3717 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3718 num_stripes = btrfs_chunk_num_stripes(leaf, ptr);
3720 rec = malloc(btrfs_chunk_record_size(num_stripes));
3722 fprintf(stderr, "memory allocation failed\n");
3726 memset(rec, 0, btrfs_chunk_record_size(num_stripes));
3728 INIT_LIST_HEAD(&rec->list);
3729 INIT_LIST_HEAD(&rec->dextents);
3732 rec->cache.start = key->offset;
3733 rec->cache.size = btrfs_chunk_length(leaf, ptr);
3735 rec->generation = btrfs_header_generation(leaf);
3737 rec->objectid = key->objectid;
3738 rec->type = key->type;
3739 rec->offset = key->offset;
3741 rec->length = rec->cache.size;
3742 rec->owner = btrfs_chunk_owner(leaf, ptr);
3743 rec->stripe_len = btrfs_chunk_stripe_len(leaf, ptr);
3744 rec->type_flags = btrfs_chunk_type(leaf, ptr);
3745 rec->io_width = btrfs_chunk_io_width(leaf, ptr);
3746 rec->io_align = btrfs_chunk_io_align(leaf, ptr);
3747 rec->sector_size = btrfs_chunk_sector_size(leaf, ptr);
3748 rec->num_stripes = num_stripes;
3749 rec->sub_stripes = btrfs_chunk_sub_stripes(leaf, ptr);
3751 for (i = 0; i < rec->num_stripes; ++i) {
3752 rec->stripes[i].devid =
3753 btrfs_stripe_devid_nr(leaf, ptr, i);
3754 rec->stripes[i].offset =
3755 btrfs_stripe_offset_nr(leaf, ptr, i);
3756 read_extent_buffer(leaf, rec->stripes[i].dev_uuid,
3757 (unsigned long)btrfs_stripe_dev_uuid_nr(ptr, i),
3764 static int process_chunk_item(struct cache_tree *chunk_cache,
3765 struct btrfs_key *key, struct extent_buffer *eb,
3768 struct chunk_record *rec;
3771 rec = btrfs_new_chunk_record(eb, key, slot);
3772 ret = insert_cache_extent(chunk_cache, &rec->cache);
3774 fprintf(stderr, "Chunk[%llu, %llu] existed.\n",
3775 rec->offset, rec->length);
3782 static int process_device_item(struct rb_root *dev_cache,
3783 struct btrfs_key *key, struct extent_buffer *eb, int slot)
3785 struct btrfs_dev_item *ptr;
3786 struct device_record *rec;
3789 ptr = btrfs_item_ptr(eb,
3790 slot, struct btrfs_dev_item);
3792 rec = malloc(sizeof(*rec));
3794 fprintf(stderr, "memory allocation failed\n");
3798 rec->devid = key->offset;
3799 rec->generation = btrfs_header_generation(eb);
3801 rec->objectid = key->objectid;
3802 rec->type = key->type;
3803 rec->offset = key->offset;
3805 rec->devid = btrfs_device_id(eb, ptr);
3806 rec->total_byte = btrfs_device_total_bytes(eb, ptr);
3807 rec->byte_used = btrfs_device_bytes_used(eb, ptr);
3809 ret = rb_insert(dev_cache, &rec->node, device_record_compare);
3811 fprintf(stderr, "Device[%llu] existed.\n", rec->devid);
3818 struct block_group_record *
3819 btrfs_new_block_group_record(struct extent_buffer *leaf, struct btrfs_key *key,
3822 struct btrfs_block_group_item *ptr;
3823 struct block_group_record *rec;
3825 rec = malloc(sizeof(*rec));
3827 fprintf(stderr, "memory allocation failed\n");
3830 memset(rec, 0, sizeof(*rec));
3832 rec->cache.start = key->objectid;
3833 rec->cache.size = key->offset;
3835 rec->generation = btrfs_header_generation(leaf);
3837 rec->objectid = key->objectid;
3838 rec->type = key->type;
3839 rec->offset = key->offset;
3841 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item);
3842 rec->flags = btrfs_disk_block_group_flags(leaf, ptr);
3844 INIT_LIST_HEAD(&rec->list);
3849 static int process_block_group_item(struct block_group_tree *block_group_cache,
3850 struct btrfs_key *key,
3851 struct extent_buffer *eb, int slot)
3853 struct block_group_record *rec;
3856 rec = btrfs_new_block_group_record(eb, key, slot);
3857 ret = insert_block_group_record(block_group_cache, rec);
3859 fprintf(stderr, "Block Group[%llu, %llu] existed.\n",
3860 rec->objectid, rec->offset);
3867 struct device_extent_record *
3868 btrfs_new_device_extent_record(struct extent_buffer *leaf,
3869 struct btrfs_key *key, int slot)
3871 struct device_extent_record *rec;
3872 struct btrfs_dev_extent *ptr;
3874 rec = malloc(sizeof(*rec));
3876 fprintf(stderr, "memory allocation failed\n");
3879 memset(rec, 0, sizeof(*rec));
3881 rec->cache.objectid = key->objectid;
3882 rec->cache.start = key->offset;
3884 rec->generation = btrfs_header_generation(leaf);
3886 rec->objectid = key->objectid;
3887 rec->type = key->type;
3888 rec->offset = key->offset;
3890 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
3891 rec->chunk_objecteid =
3892 btrfs_dev_extent_chunk_objectid(leaf, ptr);
3894 btrfs_dev_extent_chunk_offset(leaf, ptr);
3895 rec->length = btrfs_dev_extent_length(leaf, ptr);
3896 rec->cache.size = rec->length;
3898 INIT_LIST_HEAD(&rec->chunk_list);
3899 INIT_LIST_HEAD(&rec->device_list);
3905 process_device_extent_item(struct device_extent_tree *dev_extent_cache,
3906 struct btrfs_key *key, struct extent_buffer *eb,
3909 struct device_extent_record *rec;
3912 rec = btrfs_new_device_extent_record(eb, key, slot);
3913 ret = insert_device_extent_record(dev_extent_cache, rec);
3916 "Device extent[%llu, %llu, %llu] existed.\n",
3917 rec->objectid, rec->offset, rec->length);
3924 static int process_extent_item(struct btrfs_root *root,
3925 struct cache_tree *extent_cache,
3926 struct extent_buffer *eb, int slot)
3928 struct btrfs_extent_item *ei;
3929 struct btrfs_extent_inline_ref *iref;
3930 struct btrfs_extent_data_ref *dref;
3931 struct btrfs_shared_data_ref *sref;
3932 struct btrfs_key key;
3936 u32 item_size = btrfs_item_size_nr(eb, slot);
3942 btrfs_item_key_to_cpu(eb, &key, slot);
3944 if (key.type == BTRFS_METADATA_ITEM_KEY) {
3946 num_bytes = root->leafsize;
3948 num_bytes = key.offset;
3951 if (item_size < sizeof(*ei)) {
3952 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3953 struct btrfs_extent_item_v0 *ei0;
3954 BUG_ON(item_size != sizeof(*ei0));
3955 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
3956 refs = btrfs_extent_refs_v0(eb, ei0);
3960 return add_extent_rec(extent_cache, NULL, 0, key.objectid,
3961 num_bytes, refs, 0, 0, 0, metadata, 1,
3965 ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
3966 refs = btrfs_extent_refs(eb, ei);
3968 add_extent_rec(extent_cache, NULL, 0, key.objectid, num_bytes,
3969 refs, 0, 0, 0, metadata, 1, num_bytes);
3971 ptr = (unsigned long)(ei + 1);
3972 if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
3973 key.type == BTRFS_EXTENT_ITEM_KEY)
3974 ptr += sizeof(struct btrfs_tree_block_info);
3976 end = (unsigned long)ei + item_size;
3978 iref = (struct btrfs_extent_inline_ref *)ptr;
3979 type = btrfs_extent_inline_ref_type(eb, iref);
3980 offset = btrfs_extent_inline_ref_offset(eb, iref);
3982 case BTRFS_TREE_BLOCK_REF_KEY:
3983 add_tree_backref(extent_cache, key.objectid,
3986 case BTRFS_SHARED_BLOCK_REF_KEY:
3987 add_tree_backref(extent_cache, key.objectid,
3990 case BTRFS_EXTENT_DATA_REF_KEY:
3991 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3992 add_data_backref(extent_cache, key.objectid, 0,
3993 btrfs_extent_data_ref_root(eb, dref),
3994 btrfs_extent_data_ref_objectid(eb,
3996 btrfs_extent_data_ref_offset(eb, dref),
3997 btrfs_extent_data_ref_count(eb, dref),
4000 case BTRFS_SHARED_DATA_REF_KEY:
4001 sref = (struct btrfs_shared_data_ref *)(iref + 1);
4002 add_data_backref(extent_cache, key.objectid, offset,
4004 btrfs_shared_data_ref_count(eb, sref),
4008 fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
4009 key.objectid, key.type, num_bytes);
4012 ptr += btrfs_extent_inline_ref_size(type);
4019 static int check_cache_range(struct btrfs_root *root,
4020 struct btrfs_block_group_cache *cache,
4021 u64 offset, u64 bytes)
4023 struct btrfs_free_space *entry;
4029 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
4030 bytenr = btrfs_sb_offset(i);
4031 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
4032 cache->key.objectid, bytenr, 0,
4033 &logical, &nr, &stripe_len);
4038 if (logical[nr] + stripe_len <= offset)
4040 if (offset + bytes <= logical[nr])
4042 if (logical[nr] == offset) {
4043 if (stripe_len >= bytes) {
4047 bytes -= stripe_len;
4048 offset += stripe_len;
4049 } else if (logical[nr] < offset) {
4050 if (logical[nr] + stripe_len >=
4055 bytes = (offset + bytes) -
4056 (logical[nr] + stripe_len);
4057 offset = logical[nr] + stripe_len;
4060 * Could be tricky, the super may land in the
4061 * middle of the area we're checking. First
4062 * check the easiest case, it's at the end.
4064 if (logical[nr] + stripe_len >=
4066 bytes = logical[nr] - offset;
4070 /* Check the left side */
4071 ret = check_cache_range(root, cache,
4073 logical[nr] - offset);
4079 /* Now we continue with the right side */
4080 bytes = (offset + bytes) -
4081 (logical[nr] + stripe_len);
4082 offset = logical[nr] + stripe_len;
4089 entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
4091 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
4092 offset, offset+bytes);
4096 if (entry->offset != offset) {
4097 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
4102 if (entry->bytes != bytes) {
4103 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
4104 bytes, entry->bytes, offset);
4108 unlink_free_space(cache->free_space_ctl, entry);
4113 static int verify_space_cache(struct btrfs_root *root,
4114 struct btrfs_block_group_cache *cache)
4116 struct btrfs_path *path;
4117 struct extent_buffer *leaf;
4118 struct btrfs_key key;
4122 path = btrfs_alloc_path();
4126 root = root->fs_info->extent_root;
4128 last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
4130 key.objectid = last;
4132 key.type = BTRFS_EXTENT_ITEM_KEY;
4134 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4139 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4140 ret = btrfs_next_leaf(root, path);
4148 leaf = path->nodes[0];
4149 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4150 if (key.objectid >= cache->key.offset + cache->key.objectid)
4152 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
4153 key.type != BTRFS_METADATA_ITEM_KEY) {
4158 if (last == key.objectid) {
4159 if (key.type == BTRFS_EXTENT_ITEM_KEY)
4160 last = key.objectid + key.offset;
4162 last = key.objectid + root->leafsize;
4167 ret = check_cache_range(root, cache, last,
4168 key.objectid - last);
4171 if (key.type == BTRFS_EXTENT_ITEM_KEY)
4172 last = key.objectid + key.offset;
4174 last = key.objectid + root->leafsize;
4178 if (last < cache->key.objectid + cache->key.offset)
4179 ret = check_cache_range(root, cache, last,
4180 cache->key.objectid +
4181 cache->key.offset - last);
4184 btrfs_free_path(path);
4187 !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
4188 fprintf(stderr, "There are still entries left in the space "
4196 static int check_space_cache(struct btrfs_root *root)
4198 struct btrfs_block_group_cache *cache;
4199 u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
4203 if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
4204 btrfs_super_generation(root->fs_info->super_copy) !=
4205 btrfs_super_cache_generation(root->fs_info->super_copy)) {
4206 printf("cache and super generation don't match, space cache "
4207 "will be invalidated\n");
4212 cache = btrfs_lookup_first_block_group(root->fs_info, start);
4216 start = cache->key.objectid + cache->key.offset;
4217 if (!cache->free_space_ctl) {
4218 if (btrfs_init_free_space_ctl(cache,
4219 root->sectorsize)) {
4224 btrfs_remove_free_space_cache(cache);
4227 ret = load_free_space_cache(root->fs_info, cache);
4231 ret = verify_space_cache(root, cache);
4233 fprintf(stderr, "cache appears valid but isnt %Lu\n",
4234 cache->key.objectid);
4239 return error ? -EINVAL : 0;
4242 static int read_extent_data(struct btrfs_root *root, char *data,
4243 u64 logical, u64 *len, int mirror)
4246 struct btrfs_multi_bio *multi = NULL;
4247 struct btrfs_fs_info *info = root->fs_info;
4248 struct btrfs_device *device;
4252 ret = btrfs_map_block(&info->mapping_tree, READ, logical, len,
4253 &multi, mirror, NULL);
4255 fprintf(stderr, "Couldn't map the block %llu\n",
4259 device = multi->stripes[0].dev;
4261 if (device->fd == 0)
4266 ret = pread64(device->fd, data, *len, multi->stripes[0].physical);
4276 static int check_extent_csums(struct btrfs_root *root, u64 bytenr,
4277 u64 num_bytes, unsigned long leaf_offset,
4278 struct extent_buffer *eb) {
4281 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
4283 unsigned long csum_offset;
4287 u64 data_checked = 0;
4293 if (num_bytes % root->sectorsize)
4296 data = malloc(num_bytes);
4300 while (offset < num_bytes) {
4303 read_len = num_bytes - offset;
4304 /* read as much space once a time */
4305 ret = read_extent_data(root, data + offset,
4306 bytenr + offset, &read_len, mirror);
4310 /* verify every 4k data's checksum */
4311 while (data_checked < read_len) {
4313 tmp = offset + data_checked;
4315 csum = btrfs_csum_data(NULL, (char *)data + tmp,
4316 csum, root->sectorsize);
4317 btrfs_csum_final(csum, (char *)&csum);
4319 csum_offset = leaf_offset +
4320 tmp / root->sectorsize * csum_size;
4321 read_extent_buffer(eb, (char *)&csum_expected,
4322 csum_offset, csum_size);
4323 /* try another mirror */
4324 if (csum != csum_expected) {
4325 fprintf(stderr, "mirror %d bytenr %llu csum %u expected csum %u\n",
4326 mirror, bytenr + tmp,
4327 csum, csum_expected);
4328 num_copies = btrfs_num_copies(
4329 &root->fs_info->mapping_tree,
4331 if (mirror < num_copies - 1) {
4336 data_checked += root->sectorsize;
4345 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
4348 struct btrfs_path *path;
4349 struct extent_buffer *leaf;
4350 struct btrfs_key key;
4353 path = btrfs_alloc_path();
4355 fprintf(stderr, "Error allocing path\n");
4359 key.objectid = bytenr;
4360 key.type = BTRFS_EXTENT_ITEM_KEY;
4361 key.offset = (u64)-1;
4364 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
4367 fprintf(stderr, "Error looking up extent record %d\n", ret);
4368 btrfs_free_path(path);
4371 if (path->slots[0] > 0) {
4374 ret = btrfs_prev_leaf(root, path);
4377 } else if (ret > 0) {
4384 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4387 * Block group items come before extent items if they have the same
4388 * bytenr, so walk back one more just in case. Dear future traveler,
4389 * first congrats on mastering time travel. Now if it's not too much
4390 * trouble could you go back to 2006 and tell Chris to make the
4391 * BLOCK_GROUP_ITEM_KEY (and BTRFS_*_REF_KEY) lower than the
4392 * EXTENT_ITEM_KEY please?
4394 while (key.type > BTRFS_EXTENT_ITEM_KEY) {
4395 if (path->slots[0] > 0) {
4398 ret = btrfs_prev_leaf(root, path);
4401 } else if (ret > 0) {
4406 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4410 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4411 ret = btrfs_next_leaf(root, path);
4413 fprintf(stderr, "Error going to next leaf "
4415 btrfs_free_path(path);
4421 leaf = path->nodes[0];
4422 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4423 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
4427 if (key.objectid + key.offset < bytenr) {
4431 if (key.objectid > bytenr + num_bytes)
4434 if (key.objectid == bytenr) {
4435 if (key.offset >= num_bytes) {
4439 num_bytes -= key.offset;
4440 bytenr += key.offset;
4441 } else if (key.objectid < bytenr) {
4442 if (key.objectid + key.offset >= bytenr + num_bytes) {
4446 num_bytes = (bytenr + num_bytes) -
4447 (key.objectid + key.offset);
4448 bytenr = key.objectid + key.offset;
4450 if (key.objectid + key.offset < bytenr + num_bytes) {
4451 u64 new_start = key.objectid + key.offset;
4452 u64 new_bytes = bytenr + num_bytes - new_start;
4455 * Weird case, the extent is in the middle of
4456 * our range, we'll have to search one side
4457 * and then the other. Not sure if this happens
4458 * in real life, but no harm in coding it up
4459 * anyway just in case.
4461 btrfs_release_path(path);
4462 ret = check_extent_exists(root, new_start,
4465 fprintf(stderr, "Right section didn't "
4469 num_bytes = key.objectid - bytenr;
4472 num_bytes = key.objectid - bytenr;
4479 if (num_bytes && !ret) {
4480 fprintf(stderr, "There are no extents for csum range "
4481 "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
4485 btrfs_free_path(path);
4489 static int check_csums(struct btrfs_root *root)
4491 struct btrfs_path *path;
4492 struct extent_buffer *leaf;
4493 struct btrfs_key key;
4494 u64 offset = 0, num_bytes = 0;
4495 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
4499 unsigned long leaf_offset;
4501 root = root->fs_info->csum_root;
4502 if (!extent_buffer_uptodate(root->node)) {
4503 fprintf(stderr, "No valid csum tree found\n");
4507 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
4508 key.type = BTRFS_EXTENT_CSUM_KEY;
4511 path = btrfs_alloc_path();
4515 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4517 fprintf(stderr, "Error searching csum tree %d\n", ret);
4518 btrfs_free_path(path);
4522 if (ret > 0 && path->slots[0])
4527 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4528 ret = btrfs_next_leaf(root, path);
4530 fprintf(stderr, "Error going to next leaf "
4537 leaf = path->nodes[0];
4539 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4540 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
4545 data_len = (btrfs_item_size_nr(leaf, path->slots[0]) /
4546 csum_size) * root->sectorsize;
4547 if (!check_data_csum)
4548 goto skip_csum_check;
4549 leaf_offset = btrfs_item_ptr_offset(leaf, path->slots[0]);
4550 ret = check_extent_csums(root, key.offset, data_len,
4556 offset = key.offset;
4557 } else if (key.offset != offset + num_bytes) {
4558 ret = check_extent_exists(root, offset, num_bytes);
4560 fprintf(stderr, "Csum exists for %Lu-%Lu but "
4561 "there is no extent record\n",
4562 offset, offset+num_bytes);
4565 offset = key.offset;
4568 num_bytes += data_len;
4572 btrfs_free_path(path);
4576 static int is_dropped_key(struct btrfs_key *key,
4577 struct btrfs_key *drop_key) {
4578 if (key->objectid < drop_key->objectid)
4580 else if (key->objectid == drop_key->objectid) {
4581 if (key->type < drop_key->type)
4583 else if (key->type == drop_key->type) {
4584 if (key->offset < drop_key->offset)
4591 static int run_next_block(struct btrfs_trans_handle *trans,
4592 struct btrfs_root *root,
4593 struct block_info *bits,
4596 struct cache_tree *pending,
4597 struct cache_tree *seen,
4598 struct cache_tree *reada,
4599 struct cache_tree *nodes,
4600 struct cache_tree *extent_cache,
4601 struct cache_tree *chunk_cache,
4602 struct rb_root *dev_cache,
4603 struct block_group_tree *block_group_cache,
4604 struct device_extent_tree *dev_extent_cache,
4605 struct btrfs_root_item *ri)
4607 struct extent_buffer *buf;
4618 struct btrfs_key key;
4619 struct cache_extent *cache;
4622 nritems = pick_next_pending(pending, reada, nodes, *last, bits,
4623 bits_nr, &reada_bits);
4628 for(i = 0; i < nritems; i++) {
4629 ret = add_cache_extent(reada, bits[i].start,
4634 /* fixme, get the parent transid */
4635 readahead_tree_block(root, bits[i].start,
4639 *last = bits[0].start;
4640 bytenr = bits[0].start;
4641 size = bits[0].size;
4643 cache = lookup_cache_extent(pending, bytenr, size);
4645 remove_cache_extent(pending, cache);
4648 cache = lookup_cache_extent(reada, bytenr, size);
4650 remove_cache_extent(reada, cache);
4653 cache = lookup_cache_extent(nodes, bytenr, size);
4655 remove_cache_extent(nodes, cache);
4658 cache = lookup_cache_extent(extent_cache, bytenr, size);
4660 struct extent_record *rec;
4662 rec = container_of(cache, struct extent_record, cache);
4663 gen = rec->parent_generation;
4666 /* fixme, get the real parent transid */
4667 buf = read_tree_block(root, bytenr, size, gen);
4668 if (!extent_buffer_uptodate(buf)) {
4669 record_bad_block_io(root->fs_info,
4670 extent_cache, bytenr, size);
4674 nritems = btrfs_header_nritems(buf);
4677 * FIXME, this only works only if we don't have any full
4680 if (!init_extent_tree) {
4681 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
4682 btrfs_header_level(buf), 1, NULL,
4690 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
4695 owner = btrfs_header_owner(buf);
4698 ret = check_block(trans, root, extent_cache, buf, flags);
4702 if (btrfs_is_leaf(buf)) {
4703 btree_space_waste += btrfs_leaf_free_space(root, buf);
4704 for (i = 0; i < nritems; i++) {
4705 struct btrfs_file_extent_item *fi;
4706 btrfs_item_key_to_cpu(buf, &key, i);
4707 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
4708 process_extent_item(root, extent_cache, buf,
4712 if (key.type == BTRFS_METADATA_ITEM_KEY) {
4713 process_extent_item(root, extent_cache, buf,
4717 if (key.type == BTRFS_EXTENT_CSUM_KEY) {
4719 btrfs_item_size_nr(buf, i);
4722 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4723 process_chunk_item(chunk_cache, &key, buf, i);
4726 if (key.type == BTRFS_DEV_ITEM_KEY) {
4727 process_device_item(dev_cache, &key, buf, i);
4730 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
4731 process_block_group_item(block_group_cache,
4735 if (key.type == BTRFS_DEV_EXTENT_KEY) {
4736 process_device_extent_item(dev_extent_cache,
4741 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
4742 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4743 process_extent_ref_v0(extent_cache, buf, i);
4750 if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
4751 add_tree_backref(extent_cache, key.objectid, 0,
4755 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
4756 add_tree_backref(extent_cache, key.objectid,
4760 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
4761 struct btrfs_extent_data_ref *ref;
4762 ref = btrfs_item_ptr(buf, i,
4763 struct btrfs_extent_data_ref);
4764 add_data_backref(extent_cache,
4766 btrfs_extent_data_ref_root(buf, ref),
4767 btrfs_extent_data_ref_objectid(buf,
4769 btrfs_extent_data_ref_offset(buf, ref),
4770 btrfs_extent_data_ref_count(buf, ref),
4771 0, root->sectorsize);
4774 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
4775 struct btrfs_shared_data_ref *ref;
4776 ref = btrfs_item_ptr(buf, i,
4777 struct btrfs_shared_data_ref);
4778 add_data_backref(extent_cache,
4779 key.objectid, key.offset, 0, 0, 0,
4780 btrfs_shared_data_ref_count(buf, ref),
4781 0, root->sectorsize);
4784 if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
4785 struct bad_item *bad;
4787 if (key.objectid == BTRFS_ORPHAN_OBJECTID)
4791 bad = malloc(sizeof(struct bad_item));
4794 INIT_LIST_HEAD(&bad->list);
4795 memcpy(&bad->key, &key,
4796 sizeof(struct btrfs_key));
4797 bad->root_id = owner;
4798 list_add_tail(&bad->list, &delete_items);
4801 if (key.type != BTRFS_EXTENT_DATA_KEY)
4803 fi = btrfs_item_ptr(buf, i,
4804 struct btrfs_file_extent_item);
4805 if (btrfs_file_extent_type(buf, fi) ==
4806 BTRFS_FILE_EXTENT_INLINE)
4808 if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
4811 data_bytes_allocated +=
4812 btrfs_file_extent_disk_num_bytes(buf, fi);
4813 if (data_bytes_allocated < root->sectorsize) {
4816 data_bytes_referenced +=
4817 btrfs_file_extent_num_bytes(buf, fi);
4818 add_data_backref(extent_cache,
4819 btrfs_file_extent_disk_bytenr(buf, fi),
4820 parent, owner, key.objectid, key.offset -
4821 btrfs_file_extent_offset(buf, fi), 1, 1,
4822 btrfs_file_extent_disk_num_bytes(buf, fi));
4826 struct btrfs_key first_key;
4828 first_key.objectid = 0;
4831 btrfs_item_key_to_cpu(buf, &first_key, 0);
4832 level = btrfs_header_level(buf);
4833 for (i = 0; i < nritems; i++) {
4834 ptr = btrfs_node_blockptr(buf, i);
4835 size = btrfs_level_size(root, level - 1);
4836 btrfs_node_key_to_cpu(buf, &key, i);
4838 struct btrfs_key drop_key;
4839 btrfs_disk_key_to_cpu(&drop_key,
4840 &ri->drop_progress);
4841 if ((level == ri->drop_level)
4842 && is_dropped_key(&key, &drop_key)) {
4846 ret = add_extent_rec(extent_cache, &key,
4847 btrfs_node_ptr_generation(buf, i),
4848 ptr, size, 0, 0, 1, 0, 1, 0,
4852 add_tree_backref(extent_cache, ptr, parent, owner, 1);
4855 add_pending(nodes, seen, ptr, size);
4857 add_pending(pending, seen, ptr, size);
4860 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
4861 nritems) * sizeof(struct btrfs_key_ptr);
4863 total_btree_bytes += buf->len;
4864 if (fs_root_objectid(btrfs_header_owner(buf)))
4865 total_fs_tree_bytes += buf->len;
4866 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
4867 total_extent_tree_bytes += buf->len;
4868 if (!found_old_backref &&
4869 btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
4870 btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
4871 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
4872 found_old_backref = 1;
4874 free_extent_buffer(buf);
4878 static int add_root_to_pending(struct extent_buffer *buf,
4879 struct cache_tree *extent_cache,
4880 struct cache_tree *pending,
4881 struct cache_tree *seen,
4882 struct cache_tree *nodes,
4883 struct btrfs_key *root_key)
4885 if (btrfs_header_level(buf) > 0)
4886 add_pending(nodes, seen, buf->start, buf->len);
4888 add_pending(pending, seen, buf->start, buf->len);
4889 add_extent_rec(extent_cache, NULL, 0, buf->start, buf->len,
4890 0, 1, 1, 0, 1, 0, buf->len);
4892 if (root_key->objectid == BTRFS_TREE_RELOC_OBJECTID ||
4893 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
4894 add_tree_backref(extent_cache, buf->start, buf->start,
4897 add_tree_backref(extent_cache, buf->start, 0,
4898 root_key->objectid, 1);
4902 /* as we fix the tree, we might be deleting blocks that
4903 * we're tracking for repair. This hook makes sure we
4904 * remove any backrefs for blocks as we are fixing them.
4906 static int free_extent_hook(struct btrfs_trans_handle *trans,
4907 struct btrfs_root *root,
4908 u64 bytenr, u64 num_bytes, u64 parent,
4909 u64 root_objectid, u64 owner, u64 offset,
4912 struct extent_record *rec;
4913 struct cache_extent *cache;
4915 struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
4917 is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
4918 cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
4922 rec = container_of(cache, struct extent_record, cache);
4924 struct data_backref *back;
4925 back = find_data_backref(rec, parent, root_objectid, owner,
4926 offset, 1, bytenr, num_bytes);
4929 if (back->node.found_ref) {
4930 back->found_ref -= refs_to_drop;
4932 rec->refs -= refs_to_drop;
4934 if (back->node.found_extent_tree) {
4935 back->num_refs -= refs_to_drop;
4936 if (rec->extent_item_refs)
4937 rec->extent_item_refs -= refs_to_drop;
4939 if (back->found_ref == 0)
4940 back->node.found_ref = 0;
4941 if (back->num_refs == 0)
4942 back->node.found_extent_tree = 0;
4944 if (!back->node.found_extent_tree && back->node.found_ref) {
4945 list_del(&back->node.list);
4949 struct tree_backref *back;
4950 back = find_tree_backref(rec, parent, root_objectid);
4953 if (back->node.found_ref) {
4956 back->node.found_ref = 0;
4958 if (back->node.found_extent_tree) {
4959 if (rec->extent_item_refs)
4960 rec->extent_item_refs--;
4961 back->node.found_extent_tree = 0;
4963 if (!back->node.found_extent_tree && back->node.found_ref) {
4964 list_del(&back->node.list);
4968 maybe_free_extent_rec(extent_cache, rec);
4973 static int delete_extent_records(struct btrfs_trans_handle *trans,
4974 struct btrfs_root *root,
4975 struct btrfs_path *path,
4976 u64 bytenr, u64 new_len)
4978 struct btrfs_key key;
4979 struct btrfs_key found_key;
4980 struct extent_buffer *leaf;
4985 key.objectid = bytenr;
4987 key.offset = (u64)-1;
4990 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
4997 if (path->slots[0] == 0)
5003 leaf = path->nodes[0];
5004 slot = path->slots[0];
5006 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5007 if (found_key.objectid != bytenr)
5010 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
5011 found_key.type != BTRFS_METADATA_ITEM_KEY &&
5012 found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
5013 found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
5014 found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
5015 found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
5016 found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
5017 btrfs_release_path(path);
5018 if (found_key.type == 0) {
5019 if (found_key.offset == 0)
5021 key.offset = found_key.offset - 1;
5022 key.type = found_key.type;
5024 key.type = found_key.type - 1;
5025 key.offset = (u64)-1;
5029 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
5030 found_key.objectid, found_key.type, found_key.offset);
5032 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
5035 btrfs_release_path(path);
5037 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5038 found_key.type == BTRFS_METADATA_ITEM_KEY) {
5039 u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
5040 found_key.offset : root->leafsize;
5042 ret = btrfs_update_block_group(trans, root, bytenr,
5049 btrfs_release_path(path);
5054 * for a single backref, this will allocate a new extent
5055 * and add the backref to it.
5057 static int record_extent(struct btrfs_trans_handle *trans,
5058 struct btrfs_fs_info *info,
5059 struct btrfs_path *path,
5060 struct extent_record *rec,
5061 struct extent_backref *back,
5062 int allocated, u64 flags)
5065 struct btrfs_root *extent_root = info->extent_root;
5066 struct extent_buffer *leaf;
5067 struct btrfs_key ins_key;
5068 struct btrfs_extent_item *ei;
5069 struct tree_backref *tback;
5070 struct data_backref *dback;
5071 struct btrfs_tree_block_info *bi;
5074 rec->max_size = max_t(u64, rec->max_size,
5075 info->extent_root->leafsize);
5078 u32 item_size = sizeof(*ei);
5081 item_size += sizeof(*bi);
5083 ins_key.objectid = rec->start;
5084 ins_key.offset = rec->max_size;
5085 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
5087 ret = btrfs_insert_empty_item(trans, extent_root, path,
5088 &ins_key, item_size);
5092 leaf = path->nodes[0];
5093 ei = btrfs_item_ptr(leaf, path->slots[0],
5094 struct btrfs_extent_item);
5096 btrfs_set_extent_refs(leaf, ei, 0);
5097 btrfs_set_extent_generation(leaf, ei, rec->generation);
5099 if (back->is_data) {
5100 btrfs_set_extent_flags(leaf, ei,
5101 BTRFS_EXTENT_FLAG_DATA);
5103 struct btrfs_disk_key copy_key;;
5105 tback = (struct tree_backref *)back;
5106 bi = (struct btrfs_tree_block_info *)(ei + 1);
5107 memset_extent_buffer(leaf, 0, (unsigned long)bi,
5110 btrfs_set_disk_key_objectid(©_key,
5111 rec->info_objectid);
5112 btrfs_set_disk_key_type(©_key, 0);
5113 btrfs_set_disk_key_offset(©_key, 0);
5115 btrfs_set_tree_block_level(leaf, bi, rec->info_level);
5116 btrfs_set_tree_block_key(leaf, bi, ©_key);
5118 btrfs_set_extent_flags(leaf, ei,
5119 BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
5122 btrfs_mark_buffer_dirty(leaf);
5123 ret = btrfs_update_block_group(trans, extent_root, rec->start,
5124 rec->max_size, 1, 0);
5127 btrfs_release_path(path);
5130 if (back->is_data) {
5134 dback = (struct data_backref *)back;
5135 if (back->full_backref)
5136 parent = dback->parent;
5140 for (i = 0; i < dback->found_ref; i++) {
5141 /* if parent != 0, we're doing a full backref
5142 * passing BTRFS_FIRST_FREE_OBJECTID as the owner
5143 * just makes the backref allocator create a data
5146 ret = btrfs_inc_extent_ref(trans, info->extent_root,
5147 rec->start, rec->max_size,
5151 BTRFS_FIRST_FREE_OBJECTID :
5157 fprintf(stderr, "adding new data backref"
5158 " on %llu %s %llu owner %llu"
5159 " offset %llu found %d\n",
5160 (unsigned long long)rec->start,
5161 back->full_backref ?
5163 back->full_backref ?
5164 (unsigned long long)parent :
5165 (unsigned long long)dback->root,
5166 (unsigned long long)dback->owner,
5167 (unsigned long long)dback->offset,
5172 tback = (struct tree_backref *)back;
5173 if (back->full_backref)
5174 parent = tback->parent;
5178 ret = btrfs_inc_extent_ref(trans, info->extent_root,
5179 rec->start, rec->max_size,
5180 parent, tback->root, 0, 0);
5181 fprintf(stderr, "adding new tree backref on "
5182 "start %llu len %llu parent %llu root %llu\n",
5183 rec->start, rec->max_size, tback->parent, tback->root);
5188 btrfs_release_path(path);
5192 struct extent_entry {
5197 struct list_head list;
5200 static struct extent_entry *find_entry(struct list_head *entries,
5201 u64 bytenr, u64 bytes)
5203 struct extent_entry *entry = NULL;
5205 list_for_each_entry(entry, entries, list) {
5206 if (entry->bytenr == bytenr && entry->bytes == bytes)
5213 static struct extent_entry *find_most_right_entry(struct list_head *entries)
5215 struct extent_entry *entry, *best = NULL, *prev = NULL;
5217 list_for_each_entry(entry, entries, list) {
5224 * If there are as many broken entries as entries then we know
5225 * not to trust this particular entry.
5227 if (entry->broken == entry->count)
5231 * If our current entry == best then we can't be sure our best
5232 * is really the best, so we need to keep searching.
5234 if (best && best->count == entry->count) {
5240 /* Prev == entry, not good enough, have to keep searching */
5241 if (!prev->broken && prev->count == entry->count)
5245 best = (prev->count > entry->count) ? prev : entry;
5246 else if (best->count < entry->count)
5254 static int repair_ref(struct btrfs_trans_handle *trans,
5255 struct btrfs_fs_info *info, struct btrfs_path *path,
5256 struct data_backref *dback, struct extent_entry *entry)
5258 struct btrfs_root *root;
5259 struct btrfs_file_extent_item *fi;
5260 struct extent_buffer *leaf;
5261 struct btrfs_key key;
5265 key.objectid = dback->root;
5266 key.type = BTRFS_ROOT_ITEM_KEY;
5267 key.offset = (u64)-1;
5268 root = btrfs_read_fs_root(info, &key);
5270 fprintf(stderr, "Couldn't find root for our ref\n");
5275 * The backref points to the original offset of the extent if it was
5276 * split, so we need to search down to the offset we have and then walk
5277 * forward until we find the backref we're looking for.
5279 key.objectid = dback->owner;
5280 key.type = BTRFS_EXTENT_DATA_KEY;
5281 key.offset = dback->offset;
5282 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5284 fprintf(stderr, "Error looking up ref %d\n", ret);
5289 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5290 ret = btrfs_next_leaf(root, path);
5292 fprintf(stderr, "Couldn't find our ref, next\n");
5296 leaf = path->nodes[0];
5297 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5298 if (key.objectid != dback->owner ||
5299 key.type != BTRFS_EXTENT_DATA_KEY) {
5300 fprintf(stderr, "Couldn't find our ref, search\n");
5303 fi = btrfs_item_ptr(leaf, path->slots[0],
5304 struct btrfs_file_extent_item);
5305 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5306 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
5308 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
5313 btrfs_release_path(path);
5316 * Have to make sure that this root gets updated when we commit the
5319 record_root_in_trans(trans, root);
5322 * Ok we have the key of the file extent we want to fix, now we can cow
5323 * down to the thing and fix it.
5325 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
5327 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
5328 key.objectid, key.type, key.offset, ret);
5332 fprintf(stderr, "Well that's odd, we just found this key "
5333 "[%Lu, %u, %Lu]\n", key.objectid, key.type,
5337 leaf = path->nodes[0];
5338 fi = btrfs_item_ptr(leaf, path->slots[0],
5339 struct btrfs_file_extent_item);
5341 if (btrfs_file_extent_compression(leaf, fi) &&
5342 dback->disk_bytenr != entry->bytenr) {
5343 fprintf(stderr, "Ref doesn't match the record start and is "
5344 "compressed, please take a btrfs-image of this file "
5345 "system and send it to a btrfs developer so they can "
5346 "complete this functionality for bytenr %Lu\n",
5347 dback->disk_bytenr);
5351 if (dback->node.broken && dback->disk_bytenr != entry->bytenr) {
5352 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
5353 } else if (dback->disk_bytenr > entry->bytenr) {
5354 u64 off_diff, offset;
5356 off_diff = dback->disk_bytenr - entry->bytenr;
5357 offset = btrfs_file_extent_offset(leaf, fi);
5358 if (dback->disk_bytenr + offset +
5359 btrfs_file_extent_num_bytes(leaf, fi) >
5360 entry->bytenr + entry->bytes) {
5361 fprintf(stderr, "Ref is past the entry end, please "
5362 "take a btrfs-image of this file system and "
5363 "send it to a btrfs developer, ref %Lu\n",
5364 dback->disk_bytenr);
5368 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
5369 btrfs_set_file_extent_offset(leaf, fi, offset);
5370 } else if (dback->disk_bytenr < entry->bytenr) {
5373 offset = btrfs_file_extent_offset(leaf, fi);
5374 if (dback->disk_bytenr + offset < entry->bytenr) {
5375 fprintf(stderr, "Ref is before the entry start, please"
5376 " take a btrfs-image of this file system and "
5377 "send it to a btrfs developer, ref %Lu\n",
5378 dback->disk_bytenr);
5382 offset += dback->disk_bytenr;
5383 offset -= entry->bytenr;
5384 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
5385 btrfs_set_file_extent_offset(leaf, fi, offset);
5388 btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
5391 * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
5392 * only do this if we aren't using compression, otherwise it's a
5395 if (!btrfs_file_extent_compression(leaf, fi))
5396 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
5398 printf("ram bytes may be wrong?\n");
5399 btrfs_mark_buffer_dirty(leaf);
5400 btrfs_release_path(path);
5404 static int verify_backrefs(struct btrfs_trans_handle *trans,
5405 struct btrfs_fs_info *info, struct btrfs_path *path,
5406 struct extent_record *rec)
5408 struct extent_backref *back;
5409 struct data_backref *dback;
5410 struct extent_entry *entry, *best = NULL;
5413 int broken_entries = 0;
5418 * Metadata is easy and the backrefs should always agree on bytenr and
5419 * size, if not we've got bigger issues.
5424 list_for_each_entry(back, &rec->backrefs, list) {
5425 if (back->full_backref || !back->is_data)
5428 dback = (struct data_backref *)back;
5431 * We only pay attention to backrefs that we found a real
5434 if (dback->found_ref == 0)
5438 * For now we only catch when the bytes don't match, not the
5439 * bytenr. We can easily do this at the same time, but I want
5440 * to have a fs image to test on before we just add repair
5441 * functionality willy-nilly so we know we won't screw up the
5445 entry = find_entry(&entries, dback->disk_bytenr,
5448 entry = malloc(sizeof(struct extent_entry));
5453 memset(entry, 0, sizeof(*entry));
5454 entry->bytenr = dback->disk_bytenr;
5455 entry->bytes = dback->bytes;
5456 list_add_tail(&entry->list, &entries);
5461 * If we only have on entry we may think the entries agree when
5462 * in reality they don't so we have to do some extra checking.
5464 if (dback->disk_bytenr != rec->start ||
5465 dback->bytes != rec->nr || back->broken)
5476 /* Yay all the backrefs agree, carry on good sir */
5477 if (nr_entries <= 1 && !mismatch)
5480 fprintf(stderr, "attempting to repair backref discrepency for bytenr "
5481 "%Lu\n", rec->start);
5484 * First we want to see if the backrefs can agree amongst themselves who
5485 * is right, so figure out which one of the entries has the highest
5488 best = find_most_right_entry(&entries);
5491 * Ok so we may have an even split between what the backrefs think, so
5492 * this is where we use the extent ref to see what it thinks.
5495 entry = find_entry(&entries, rec->start, rec->nr);
5496 if (!entry && (!broken_entries || !rec->found_rec)) {
5497 fprintf(stderr, "Backrefs don't agree with each other "
5498 "and extent record doesn't agree with anybody,"
5499 " so we can't fix bytenr %Lu bytes %Lu\n",
5500 rec->start, rec->nr);
5503 } else if (!entry) {
5505 * Ok our backrefs were broken, we'll assume this is the
5506 * correct value and add an entry for this range.
5508 entry = malloc(sizeof(struct extent_entry));
5513 memset(entry, 0, sizeof(*entry));
5514 entry->bytenr = rec->start;
5515 entry->bytes = rec->nr;
5516 list_add_tail(&entry->list, &entries);
5520 best = find_most_right_entry(&entries);
5522 fprintf(stderr, "Backrefs and extent record evenly "
5523 "split on who is right, this is going to "
5524 "require user input to fix bytenr %Lu bytes "
5525 "%Lu\n", rec->start, rec->nr);
5532 * I don't think this can happen currently as we'll abort() if we catch
5533 * this case higher up, but in case somebody removes that we still can't
5534 * deal with it properly here yet, so just bail out of that's the case.
5536 if (best->bytenr != rec->start) {
5537 fprintf(stderr, "Extent start and backref starts don't match, "
5538 "please use btrfs-image on this file system and send "
5539 "it to a btrfs developer so they can make fsck fix "
5540 "this particular case. bytenr is %Lu, bytes is %Lu\n",
5541 rec->start, rec->nr);
5547 * Ok great we all agreed on an extent record, let's go find the real
5548 * references and fix up the ones that don't match.
5550 list_for_each_entry(back, &rec->backrefs, list) {
5551 if (back->full_backref || !back->is_data)
5554 dback = (struct data_backref *)back;
5557 * Still ignoring backrefs that don't have a real ref attached
5560 if (dback->found_ref == 0)
5563 if (dback->bytes == best->bytes &&
5564 dback->disk_bytenr == best->bytenr)
5567 ret = repair_ref(trans, info, path, dback, best);
5573 * Ok we messed with the actual refs, which means we need to drop our
5574 * entire cache and go back and rescan. I know this is a huge pain and
5575 * adds a lot of extra work, but it's the only way to be safe. Once all
5576 * the backrefs agree we may not need to do anything to the extent
5581 while (!list_empty(&entries)) {
5582 entry = list_entry(entries.next, struct extent_entry, list);
5583 list_del_init(&entry->list);
5589 static int process_duplicates(struct btrfs_root *root,
5590 struct cache_tree *extent_cache,
5591 struct extent_record *rec)
5593 struct extent_record *good, *tmp;
5594 struct cache_extent *cache;
5598 * If we found a extent record for this extent then return, or if we
5599 * have more than one duplicate we are likely going to need to delete
5602 if (rec->found_rec || rec->num_duplicates > 1)
5605 /* Shouldn't happen but just in case */
5606 BUG_ON(!rec->num_duplicates);
5609 * So this happens if we end up with a backref that doesn't match the
5610 * actual extent entry. So either the backref is bad or the extent
5611 * entry is bad. Either way we want to have the extent_record actually
5612 * reflect what we found in the extent_tree, so we need to take the
5613 * duplicate out and use that as the extent_record since the only way we
5614 * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
5616 remove_cache_extent(extent_cache, &rec->cache);
5618 good = list_entry(rec->dups.next, struct extent_record, list);
5619 list_del_init(&good->list);
5620 INIT_LIST_HEAD(&good->backrefs);
5621 INIT_LIST_HEAD(&good->dups);
5622 good->cache.start = good->start;
5623 good->cache.size = good->nr;
5624 good->content_checked = 0;
5625 good->owner_ref_checked = 0;
5626 good->num_duplicates = 0;
5627 good->refs = rec->refs;
5628 list_splice_init(&rec->backrefs, &good->backrefs);
5630 cache = lookup_cache_extent(extent_cache, good->start,
5634 tmp = container_of(cache, struct extent_record, cache);
5637 * If we find another overlapping extent and it's found_rec is
5638 * set then it's a duplicate and we need to try and delete
5641 if (tmp->found_rec || tmp->num_duplicates > 0) {
5642 if (list_empty(&good->list))
5643 list_add_tail(&good->list,
5644 &duplicate_extents);
5645 good->num_duplicates += tmp->num_duplicates + 1;
5646 list_splice_init(&tmp->dups, &good->dups);
5647 list_del_init(&tmp->list);
5648 list_add_tail(&tmp->list, &good->dups);
5649 remove_cache_extent(extent_cache, &tmp->cache);
5654 * Ok we have another non extent item backed extent rec, so lets
5655 * just add it to this extent and carry on like we did above.
5657 good->refs += tmp->refs;
5658 list_splice_init(&tmp->backrefs, &good->backrefs);
5659 remove_cache_extent(extent_cache, &tmp->cache);
5662 ret = insert_cache_extent(extent_cache, &good->cache);
5665 return good->num_duplicates ? 0 : 1;
5668 static int delete_duplicate_records(struct btrfs_trans_handle *trans,
5669 struct btrfs_root *root,
5670 struct extent_record *rec)
5672 LIST_HEAD(delete_list);
5673 struct btrfs_path *path;
5674 struct extent_record *tmp, *good, *n;
5677 struct btrfs_key key;
5679 path = btrfs_alloc_path();
5686 /* Find the record that covers all of the duplicates. */
5687 list_for_each_entry(tmp, &rec->dups, list) {
5688 if (good->start < tmp->start)
5690 if (good->nr > tmp->nr)
5693 if (tmp->start + tmp->nr < good->start + good->nr) {
5694 fprintf(stderr, "Ok we have overlapping extents that "
5695 "aren't completely covered by eachother, this "
5696 "is going to require more careful thought. "
5697 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
5698 tmp->start, tmp->nr, good->start, good->nr);
5705 list_add_tail(&rec->list, &delete_list);
5707 list_for_each_entry_safe(tmp, n, &rec->dups, list) {
5710 list_move_tail(&tmp->list, &delete_list);
5713 root = root->fs_info->extent_root;
5714 list_for_each_entry(tmp, &delete_list, list) {
5715 if (tmp->found_rec == 0)
5717 key.objectid = tmp->start;
5718 key.type = BTRFS_EXTENT_ITEM_KEY;
5719 key.offset = tmp->nr;
5721 /* Shouldn't happen but just in case */
5722 if (tmp->metadata) {
5723 fprintf(stderr, "Well this shouldn't happen, extent "
5724 "record overlaps but is metadata? "
5725 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
5729 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5735 ret = btrfs_del_item(trans, root, path);
5738 btrfs_release_path(path);
5743 while (!list_empty(&delete_list)) {
5744 tmp = list_entry(delete_list.next, struct extent_record, list);
5745 list_del_init(&tmp->list);
5751 while (!list_empty(&rec->dups)) {
5752 tmp = list_entry(rec->dups.next, struct extent_record, list);
5753 list_del_init(&tmp->list);
5757 btrfs_free_path(path);
5759 if (!ret && !nr_del)
5760 rec->num_duplicates = 0;
5762 return ret ? ret : nr_del;
5765 static int find_possible_backrefs(struct btrfs_trans_handle *trans,
5766 struct btrfs_fs_info *info,
5767 struct btrfs_path *path,
5768 struct cache_tree *extent_cache,
5769 struct extent_record *rec)
5771 struct btrfs_root *root;
5772 struct extent_backref *back;
5773 struct data_backref *dback;
5774 struct cache_extent *cache;
5775 struct btrfs_file_extent_item *fi;
5776 struct btrfs_key key;
5780 list_for_each_entry(back, &rec->backrefs, list) {
5781 /* Don't care about full backrefs (poor unloved backrefs) */
5782 if (back->full_backref || !back->is_data)
5785 dback = (struct data_backref *)back;
5787 /* We found this one, we don't need to do a lookup */
5788 if (dback->found_ref)
5791 key.objectid = dback->root;
5792 key.type = BTRFS_ROOT_ITEM_KEY;
5793 key.offset = (u64)-1;
5795 root = btrfs_read_fs_root(info, &key);
5797 /* No root, definitely a bad ref, skip */
5798 if (IS_ERR(root) && PTR_ERR(root) == -ENOENT)
5800 /* Other err, exit */
5802 return PTR_ERR(root);
5804 key.objectid = dback->owner;
5805 key.type = BTRFS_EXTENT_DATA_KEY;
5806 key.offset = dback->offset;
5807 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5809 btrfs_release_path(path);
5812 /* Didn't find it, we can carry on */
5817 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5818 struct btrfs_file_extent_item);
5819 bytenr = btrfs_file_extent_disk_bytenr(path->nodes[0], fi);
5820 bytes = btrfs_file_extent_disk_num_bytes(path->nodes[0], fi);
5821 btrfs_release_path(path);
5822 cache = lookup_cache_extent(extent_cache, bytenr, 1);
5824 struct extent_record *tmp;
5825 tmp = container_of(cache, struct extent_record, cache);
5828 * If we found an extent record for the bytenr for this
5829 * particular backref then we can't add it to our
5830 * current extent record. We only want to add backrefs
5831 * that don't have a corresponding extent item in the
5832 * extent tree since they likely belong to this record
5833 * and we need to fix it if it doesn't match bytenrs.
5839 dback->found_ref += 1;
5840 dback->disk_bytenr = bytenr;
5841 dback->bytes = bytes;
5844 * Set this so the verify backref code knows not to trust the
5845 * values in this backref.
5854 * when an incorrect extent item is found, this will delete
5855 * all of the existing entries for it and recreate them
5856 * based on what the tree scan found.
5858 static int fixup_extent_refs(struct btrfs_trans_handle *trans,
5859 struct btrfs_fs_info *info,
5860 struct cache_tree *extent_cache,
5861 struct extent_record *rec)
5864 struct btrfs_path *path;
5865 struct list_head *cur = rec->backrefs.next;
5866 struct cache_extent *cache;
5867 struct extent_backref *back;
5872 * remember our flags for recreating the extent.
5873 * FIXME, if we have cleared extent tree, we can not
5874 * lookup extent info in extent tree.
5876 if (!init_extent_tree) {
5877 ret = btrfs_lookup_extent_info(NULL, info->extent_root,
5878 rec->start, rec->max_size,
5879 rec->metadata, NULL, &flags);
5886 path = btrfs_alloc_path();
5890 if (rec->refs != rec->extent_item_refs && !rec->metadata) {
5892 * Sometimes the backrefs themselves are so broken they don't
5893 * get attached to any meaningful rec, so first go back and
5894 * check any of our backrefs that we couldn't find and throw
5895 * them into the list if we find the backref so that
5896 * verify_backrefs can figure out what to do.
5898 ret = find_possible_backrefs(trans, info, path, extent_cache,
5904 /* step one, make sure all of the backrefs agree */
5905 ret = verify_backrefs(trans, info, path, rec);
5909 /* step two, delete all the existing records */
5910 ret = delete_extent_records(trans, info->extent_root, path,
5911 rec->start, rec->max_size);
5916 /* was this block corrupt? If so, don't add references to it */
5917 cache = lookup_cache_extent(info->corrupt_blocks,
5918 rec->start, rec->max_size);
5924 /* step three, recreate all the refs we did find */
5925 while(cur != &rec->backrefs) {
5926 back = list_entry(cur, struct extent_backref, list);
5930 * if we didn't find any references, don't create a
5933 if (!back->found_ref)
5936 ret = record_extent(trans, info, path, rec, back, allocated, flags);
5943 btrfs_free_path(path);
5947 /* right now we only prune from the extent allocation tree */
5948 static int prune_one_block(struct btrfs_trans_handle *trans,
5949 struct btrfs_fs_info *info,
5950 struct btrfs_corrupt_block *corrupt)
5953 struct btrfs_path path;
5954 struct extent_buffer *eb;
5958 int level = corrupt->level + 1;
5960 btrfs_init_path(&path);
5962 /* we want to stop at the parent to our busted block */
5963 path.lowest_level = level;
5965 ret = btrfs_search_slot(trans, info->extent_root,
5966 &corrupt->key, &path, -1, 1);
5971 eb = path.nodes[level];
5978 * hopefully the search gave us the block we want to prune,
5979 * lets try that first
5981 slot = path.slots[level];
5982 found = btrfs_node_blockptr(eb, slot);
5983 if (found == corrupt->cache.start)
5986 nritems = btrfs_header_nritems(eb);
5988 /* the search failed, lets scan this node and hope we find it */
5989 for (slot = 0; slot < nritems; slot++) {
5990 found = btrfs_node_blockptr(eb, slot);
5991 if (found == corrupt->cache.start)
5995 * we couldn't find the bad block. TODO, search all the nodes for pointers
5998 if (eb == info->extent_root->node) {
6003 btrfs_release_path(&path);
6008 printk("deleting pointer to block %Lu\n", corrupt->cache.start);
6009 ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
6012 btrfs_release_path(&path);
6016 static int prune_corrupt_blocks(struct btrfs_trans_handle *trans,
6017 struct btrfs_fs_info *info)
6019 struct cache_extent *cache;
6020 struct btrfs_corrupt_block *corrupt;
6022 cache = search_cache_extent(info->corrupt_blocks, 0);
6026 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
6027 prune_one_block(trans, info, corrupt);
6028 cache = next_cache_extent(cache);
6033 static void free_corrupt_block(struct cache_extent *cache)
6035 struct btrfs_corrupt_block *corrupt;
6037 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
6041 FREE_EXTENT_CACHE_BASED_TREE(corrupt_blocks, free_corrupt_block);
6043 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
6045 struct btrfs_block_group_cache *cache;
6050 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
6051 &start, &end, EXTENT_DIRTY);
6054 clear_extent_dirty(&fs_info->free_space_cache, start, end,
6060 cache = btrfs_lookup_first_block_group(fs_info, start);
6065 start = cache->key.objectid + cache->key.offset;
6069 static int check_extent_refs(struct btrfs_trans_handle *trans,
6070 struct btrfs_root *root,
6071 struct cache_tree *extent_cache)
6073 struct extent_record *rec;
6074 struct cache_extent *cache;
6082 * if we're doing a repair, we have to make sure
6083 * we don't allocate from the problem extents.
6084 * In the worst case, this will be all the
6087 cache = search_cache_extent(extent_cache, 0);
6089 rec = container_of(cache, struct extent_record, cache);
6090 btrfs_pin_extent(root->fs_info,
6091 rec->start, rec->max_size);
6092 cache = next_cache_extent(cache);
6095 /* pin down all the corrupted blocks too */
6096 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
6098 btrfs_pin_extent(root->fs_info,
6099 cache->start, cache->size);
6100 cache = next_cache_extent(cache);
6102 prune_corrupt_blocks(trans, root->fs_info);
6103 reset_cached_block_groups(root->fs_info);
6107 * We need to delete any duplicate entries we find first otherwise we
6108 * could mess up the extent tree when we have backrefs that actually
6109 * belong to a different extent item and not the weird duplicate one.
6111 while (repair && !list_empty(&duplicate_extents)) {
6112 rec = list_entry(duplicate_extents.next, struct extent_record,
6114 list_del_init(&rec->list);
6116 /* Sometimes we can find a backref before we find an actual
6117 * extent, so we need to process it a little bit to see if there
6118 * truly are multiple EXTENT_ITEM_KEY's for the same range, or
6119 * if this is a backref screwup. If we need to delete stuff
6120 * process_duplicates() will return 0, otherwise it will return
6123 if (process_duplicates(root, extent_cache, rec))
6125 ret = delete_duplicate_records(trans, root, rec);
6129 * delete_duplicate_records will return the number of entries
6130 * deleted, so if it's greater than 0 then we know we actually
6131 * did something and we need to remove.
6142 cache = search_cache_extent(extent_cache, 0);
6145 rec = container_of(cache, struct extent_record, cache);
6146 if (rec->num_duplicates) {
6147 fprintf(stderr, "extent item %llu has multiple extent "
6148 "items\n", (unsigned long long)rec->start);
6152 if (rec->refs != rec->extent_item_refs) {
6153 fprintf(stderr, "ref mismatch on [%llu %llu] ",
6154 (unsigned long long)rec->start,
6155 (unsigned long long)rec->nr);
6156 fprintf(stderr, "extent item %llu, found %llu\n",
6157 (unsigned long long)rec->extent_item_refs,
6158 (unsigned long long)rec->refs);
6159 if (!fixed && repair) {
6160 ret = fixup_extent_refs(trans, root->fs_info,
6169 if (all_backpointers_checked(rec, 1)) {
6170 fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
6171 (unsigned long long)rec->start,
6172 (unsigned long long)rec->nr);
6174 if (!fixed && repair) {
6175 ret = fixup_extent_refs(trans, root->fs_info,
6184 if (!rec->owner_ref_checked) {
6185 fprintf(stderr, "owner ref check failed [%llu %llu]\n",
6186 (unsigned long long)rec->start,
6187 (unsigned long long)rec->nr);
6188 if (!fixed && repair) {
6189 ret = fixup_extent_refs(trans, root->fs_info,
6198 remove_cache_extent(extent_cache, cache);
6199 free_all_extent_backrefs(rec);
6204 if (ret && ret != -EAGAIN) {
6205 fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
6208 btrfs_fix_block_accounting(trans, root);
6211 fprintf(stderr, "repaired damaged extent references\n");
6217 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
6221 if (type & BTRFS_BLOCK_GROUP_RAID0) {
6222 stripe_size = length;
6223 stripe_size /= num_stripes;
6224 } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
6225 stripe_size = length * 2;
6226 stripe_size /= num_stripes;
6227 } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
6228 stripe_size = length;
6229 stripe_size /= (num_stripes - 1);
6230 } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
6231 stripe_size = length;
6232 stripe_size /= (num_stripes - 2);
6234 stripe_size = length;
6239 static int check_chunk_refs(struct chunk_record *chunk_rec,
6240 struct block_group_tree *block_group_cache,
6241 struct device_extent_tree *dev_extent_cache,
6244 struct cache_extent *block_group_item;
6245 struct block_group_record *block_group_rec;
6246 struct cache_extent *dev_extent_item;
6247 struct device_extent_record *dev_extent_rec;
6254 block_group_item = lookup_cache_extent(&block_group_cache->tree,
6257 if (block_group_item) {
6258 block_group_rec = container_of(block_group_item,
6259 struct block_group_record,
6261 if (chunk_rec->length != block_group_rec->offset ||
6262 chunk_rec->offset != block_group_rec->objectid ||
6263 chunk_rec->type_flags != block_group_rec->flags) {
6266 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
6267 chunk_rec->objectid,
6272 chunk_rec->type_flags,
6273 block_group_rec->objectid,
6274 block_group_rec->type,
6275 block_group_rec->offset,
6276 block_group_rec->offset,
6277 block_group_rec->objectid,
6278 block_group_rec->flags);
6281 list_del_init(&block_group_rec->list);
6282 chunk_rec->bg_rec = block_group_rec;
6287 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
6288 chunk_rec->objectid,
6293 chunk_rec->type_flags);
6297 length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
6298 chunk_rec->num_stripes);
6299 for (i = 0; i < chunk_rec->num_stripes; ++i) {
6300 devid = chunk_rec->stripes[i].devid;
6301 offset = chunk_rec->stripes[i].offset;
6302 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
6303 devid, offset, length);
6304 if (dev_extent_item) {
6305 dev_extent_rec = container_of(dev_extent_item,
6306 struct device_extent_record,
6308 if (dev_extent_rec->objectid != devid ||
6309 dev_extent_rec->offset != offset ||
6310 dev_extent_rec->chunk_offset != chunk_rec->offset ||
6311 dev_extent_rec->length != length) {
6314 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
6315 chunk_rec->objectid,
6318 chunk_rec->stripes[i].devid,
6319 chunk_rec->stripes[i].offset,
6320 dev_extent_rec->objectid,
6321 dev_extent_rec->offset,
6322 dev_extent_rec->length);
6325 list_move(&dev_extent_rec->chunk_list,
6326 &chunk_rec->dextents);
6331 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
6332 chunk_rec->objectid,
6335 chunk_rec->stripes[i].devid,
6336 chunk_rec->stripes[i].offset);
6343 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
6344 int check_chunks(struct cache_tree *chunk_cache,
6345 struct block_group_tree *block_group_cache,
6346 struct device_extent_tree *dev_extent_cache,
6347 struct list_head *good, struct list_head *bad, int silent)
6349 struct cache_extent *chunk_item;
6350 struct chunk_record *chunk_rec;
6351 struct block_group_record *bg_rec;
6352 struct device_extent_record *dext_rec;
6356 chunk_item = first_cache_extent(chunk_cache);
6357 while (chunk_item) {
6358 chunk_rec = container_of(chunk_item, struct chunk_record,
6360 err = check_chunk_refs(chunk_rec, block_group_cache,
6361 dev_extent_cache, silent);
6365 list_add_tail(&chunk_rec->list, bad);
6368 list_add_tail(&chunk_rec->list, good);
6371 chunk_item = next_cache_extent(chunk_item);
6374 list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
6377 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
6385 list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
6389 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
6400 static int check_device_used(struct device_record *dev_rec,
6401 struct device_extent_tree *dext_cache)
6403 struct cache_extent *cache;
6404 struct device_extent_record *dev_extent_rec;
6407 cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
6409 dev_extent_rec = container_of(cache,
6410 struct device_extent_record,
6412 if (dev_extent_rec->objectid != dev_rec->devid)
6415 list_del_init(&dev_extent_rec->device_list);
6416 total_byte += dev_extent_rec->length;
6417 cache = next_cache_extent(cache);
6420 if (total_byte != dev_rec->byte_used) {
6422 "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
6423 total_byte, dev_rec->byte_used, dev_rec->objectid,
6424 dev_rec->type, dev_rec->offset);
6431 /* check btrfs_dev_item -> btrfs_dev_extent */
6432 static int check_devices(struct rb_root *dev_cache,
6433 struct device_extent_tree *dev_extent_cache)
6435 struct rb_node *dev_node;
6436 struct device_record *dev_rec;
6437 struct device_extent_record *dext_rec;
6441 dev_node = rb_first(dev_cache);
6443 dev_rec = container_of(dev_node, struct device_record, node);
6444 err = check_device_used(dev_rec, dev_extent_cache);
6448 dev_node = rb_next(dev_node);
6450 list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
6453 "Device extent[%llu, %llu, %llu] didn't find its device.\n",
6454 dext_rec->objectid, dext_rec->offset, dext_rec->length);
6461 static int check_chunks_and_extents(struct btrfs_root *root)
6463 struct rb_root dev_cache;
6464 struct cache_tree chunk_cache;
6465 struct block_group_tree block_group_cache;
6466 struct device_extent_tree dev_extent_cache;
6467 struct cache_tree extent_cache;
6468 struct cache_tree seen;
6469 struct cache_tree pending;
6470 struct cache_tree reada;
6471 struct cache_tree nodes;
6472 struct cache_tree corrupt_blocks;
6473 struct btrfs_path path;
6474 struct btrfs_key key;
6475 struct btrfs_key found_key;
6478 struct block_info *bits;
6480 struct extent_buffer *leaf;
6481 struct btrfs_trans_handle *trans = NULL;
6483 struct btrfs_root_item ri;
6484 struct list_head dropping_trees;
6486 dev_cache = RB_ROOT;
6487 cache_tree_init(&chunk_cache);
6488 block_group_tree_init(&block_group_cache);
6489 device_extent_tree_init(&dev_extent_cache);
6491 cache_tree_init(&extent_cache);
6492 cache_tree_init(&seen);
6493 cache_tree_init(&pending);
6494 cache_tree_init(&nodes);
6495 cache_tree_init(&reada);
6496 cache_tree_init(&corrupt_blocks);
6497 INIT_LIST_HEAD(&dropping_trees);
6500 trans = btrfs_start_transaction(root, 1);
6501 if (IS_ERR(trans)) {
6502 fprintf(stderr, "Error starting transaction\n");
6503 return PTR_ERR(trans);
6505 root->fs_info->fsck_extent_cache = &extent_cache;
6506 root->fs_info->free_extent_hook = free_extent_hook;
6507 root->fs_info->corrupt_blocks = &corrupt_blocks;
6511 bits = malloc(bits_nr * sizeof(struct block_info));
6518 add_root_to_pending(root->fs_info->tree_root->node,
6519 &extent_cache, &pending, &seen, &nodes,
6520 &root->fs_info->tree_root->root_key);
6522 add_root_to_pending(root->fs_info->chunk_root->node,
6523 &extent_cache, &pending, &seen, &nodes,
6524 &root->fs_info->chunk_root->root_key);
6526 btrfs_init_path(&path);
6529 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
6530 ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
6535 leaf = path.nodes[0];
6536 slot = path.slots[0];
6537 if (slot >= btrfs_header_nritems(path.nodes[0])) {
6538 ret = btrfs_next_leaf(root, &path);
6541 leaf = path.nodes[0];
6542 slot = path.slots[0];
6544 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
6545 if (btrfs_key_type(&found_key) == BTRFS_ROOT_ITEM_KEY) {
6546 unsigned long offset;
6547 struct extent_buffer *buf;
6549 offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
6550 read_extent_buffer(leaf, &ri, offset, sizeof(ri));
6551 if (btrfs_disk_key_objectid(&ri.drop_progress) == 0) {
6552 buf = read_tree_block(root->fs_info->tree_root,
6553 btrfs_root_bytenr(&ri),
6554 btrfs_level_size(root,
6555 btrfs_root_level(&ri)),
6561 add_root_to_pending(buf, &extent_cache,
6562 &pending, &seen, &nodes,
6564 free_extent_buffer(buf);
6566 struct dropping_root_item_record *dri_rec;
6567 dri_rec = malloc(sizeof(*dri_rec));
6572 memcpy(&dri_rec->ri, &ri, sizeof(ri));
6573 memcpy(&dri_rec->found_key, &found_key,
6575 list_add_tail(&dri_rec->list, &dropping_trees);
6580 btrfs_release_path(&path);
6582 ret = run_next_block(trans, root, bits, bits_nr, &last,
6583 &pending, &seen, &reada, &nodes,
6584 &extent_cache, &chunk_cache, &dev_cache,
6585 &block_group_cache, &dev_extent_cache,
6591 while (!list_empty(&dropping_trees)) {
6592 struct dropping_root_item_record *rec;
6593 struct extent_buffer *buf;
6594 rec = list_entry(dropping_trees.next,
6595 struct dropping_root_item_record, list);
6601 buf = read_tree_block(root->fs_info->tree_root,
6602 btrfs_root_bytenr(&rec->ri),
6603 btrfs_level_size(root,
6604 btrfs_root_level(&rec->ri)), 0);
6609 add_root_to_pending(buf, &extent_cache, &pending,
6610 &seen, &nodes, &rec->found_key);
6612 ret = run_next_block(trans, root, bits, bits_nr, &last,
6613 &pending, &seen, &reada,
6614 &nodes, &extent_cache,
6615 &chunk_cache, &dev_cache,
6622 free_extent_buffer(buf);
6623 list_del(&rec->list);
6628 ret = check_extent_refs(trans, root, &extent_cache);
6629 if (ret == -EAGAIN) {
6630 ret = btrfs_commit_transaction(trans, root);
6634 trans = btrfs_start_transaction(root, 1);
6635 if (IS_ERR(trans)) {
6636 ret = PTR_ERR(trans);
6640 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
6641 free_extent_cache_tree(&seen);
6642 free_extent_cache_tree(&pending);
6643 free_extent_cache_tree(&reada);
6644 free_extent_cache_tree(&nodes);
6645 free_chunk_cache_tree(&chunk_cache);
6646 free_block_group_tree(&block_group_cache);
6647 free_device_cache_tree(&dev_cache);
6648 free_device_extent_tree(&dev_extent_cache);
6649 free_extent_record_cache(root->fs_info, &extent_cache);
6653 err = check_chunks(&chunk_cache, &block_group_cache,
6654 &dev_extent_cache, NULL, NULL, 0);
6658 err = check_devices(&dev_cache, &dev_extent_cache);
6664 err = btrfs_commit_transaction(trans, root);
6669 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
6670 root->fs_info->fsck_extent_cache = NULL;
6671 root->fs_info->free_extent_hook = NULL;
6672 root->fs_info->corrupt_blocks = NULL;
6675 free_chunk_cache_tree(&chunk_cache);
6676 free_device_cache_tree(&dev_cache);
6677 free_block_group_tree(&block_group_cache);
6678 free_device_extent_tree(&dev_extent_cache);
6679 free_extent_cache_tree(&seen);
6680 free_extent_cache_tree(&pending);
6681 free_extent_cache_tree(&reada);
6682 free_extent_cache_tree(&nodes);
6686 static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
6687 struct btrfs_root *root, int overwrite)
6689 struct extent_buffer *c;
6690 struct extent_buffer *old = root->node;
6693 struct btrfs_disk_key disk_key = {0,0,0};
6699 extent_buffer_get(c);
6702 c = btrfs_alloc_free_block(trans, root,
6703 btrfs_level_size(root, 0),
6704 root->root_key.objectid,
6705 &disk_key, level, 0, 0);
6708 extent_buffer_get(c);
6712 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
6713 btrfs_set_header_level(c, level);
6714 btrfs_set_header_bytenr(c, c->start);
6715 btrfs_set_header_generation(c, trans->transid);
6716 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
6717 btrfs_set_header_owner(c, root->root_key.objectid);
6719 write_extent_buffer(c, root->fs_info->fsid,
6720 btrfs_header_fsid(), BTRFS_FSID_SIZE);
6722 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
6723 btrfs_header_chunk_tree_uuid(c),
6726 btrfs_mark_buffer_dirty(c);
6728 * this case can happen in the following case:
6730 * 1.overwrite previous root.
6732 * 2.reinit reloc data root, this is because we skip pin
6733 * down reloc data tree before which means we can allocate
6734 * same block bytenr here.
6736 if (old->start == c->start) {
6737 btrfs_set_root_generation(&root->root_item,
6739 root->root_item.level = btrfs_header_level(root->node);
6740 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6741 &root->root_key, &root->root_item);
6743 free_extent_buffer(c);
6747 free_extent_buffer(old);
6749 add_root_to_dirty_list(root);
6753 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
6754 struct extent_buffer *eb, int tree_root)
6756 struct extent_buffer *tmp;
6757 struct btrfs_root_item *ri;
6758 struct btrfs_key key;
6761 int level = btrfs_header_level(eb);
6767 * If we have pinned this block before, don't pin it again.
6768 * This can not only avoid forever loop with broken filesystem
6769 * but also give us some speedups.
6771 if (test_range_bit(&fs_info->pinned_extents, eb->start,
6772 eb->start + eb->len - 1, EXTENT_DIRTY, 0))
6775 btrfs_pin_extent(fs_info, eb->start, eb->len);
6777 leafsize = btrfs_super_leafsize(fs_info->super_copy);
6778 nritems = btrfs_header_nritems(eb);
6779 for (i = 0; i < nritems; i++) {
6781 btrfs_item_key_to_cpu(eb, &key, i);
6782 if (key.type != BTRFS_ROOT_ITEM_KEY)
6784 /* Skip the extent root and reloc roots */
6785 if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6786 key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
6787 key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
6789 ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
6790 bytenr = btrfs_disk_root_bytenr(eb, ri);
6793 * If at any point we start needing the real root we
6794 * will have to build a stump root for the root we are
6795 * in, but for now this doesn't actually use the root so
6796 * just pass in extent_root.
6798 tmp = read_tree_block(fs_info->extent_root, bytenr,
6801 fprintf(stderr, "Error reading root block\n");
6804 ret = pin_down_tree_blocks(fs_info, tmp, 0);
6805 free_extent_buffer(tmp);
6809 bytenr = btrfs_node_blockptr(eb, i);
6811 /* If we aren't the tree root don't read the block */
6812 if (level == 1 && !tree_root) {
6813 btrfs_pin_extent(fs_info, bytenr, leafsize);
6817 tmp = read_tree_block(fs_info->extent_root, bytenr,
6820 fprintf(stderr, "Error reading tree block\n");
6823 ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
6824 free_extent_buffer(tmp);
6833 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
6837 ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
6841 return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
6844 static int reset_block_groups(struct btrfs_fs_info *fs_info)
6846 struct btrfs_block_group_cache *cache;
6847 struct btrfs_path *path;
6848 struct extent_buffer *leaf;
6849 struct btrfs_chunk *chunk;
6850 struct btrfs_key key;
6854 path = btrfs_alloc_path();
6859 key.type = BTRFS_CHUNK_ITEM_KEY;
6862 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
6864 btrfs_free_path(path);
6869 * We do this in case the block groups were screwed up and had alloc
6870 * bits that aren't actually set on the chunks. This happens with
6871 * restored images every time and could happen in real life I guess.
6873 fs_info->avail_data_alloc_bits = 0;
6874 fs_info->avail_metadata_alloc_bits = 0;
6875 fs_info->avail_system_alloc_bits = 0;
6877 /* First we need to create the in-memory block groups */
6879 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
6880 ret = btrfs_next_leaf(fs_info->chunk_root, path);
6882 btrfs_free_path(path);
6890 leaf = path->nodes[0];
6891 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6892 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
6897 chunk = btrfs_item_ptr(leaf, path->slots[0],
6898 struct btrfs_chunk);
6899 btrfs_add_block_group(fs_info, 0,
6900 btrfs_chunk_type(leaf, chunk),
6901 key.objectid, key.offset,
6902 btrfs_chunk_length(leaf, chunk));
6903 set_extent_dirty(&fs_info->free_space_cache, key.offset,
6904 key.offset + btrfs_chunk_length(leaf, chunk),
6910 cache = btrfs_lookup_first_block_group(fs_info, start);
6914 start = cache->key.objectid + cache->key.offset;
6917 btrfs_free_path(path);
6921 static int reset_balance(struct btrfs_trans_handle *trans,
6922 struct btrfs_fs_info *fs_info)
6924 struct btrfs_root *root = fs_info->tree_root;
6925 struct btrfs_path *path;
6926 struct extent_buffer *leaf;
6927 struct btrfs_key key;
6928 int del_slot, del_nr = 0;
6932 path = btrfs_alloc_path();
6936 key.objectid = BTRFS_BALANCE_OBJECTID;
6937 key.type = BTRFS_BALANCE_ITEM_KEY;
6940 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6945 goto reinit_data_reloc;
6950 ret = btrfs_del_item(trans, root, path);
6953 btrfs_release_path(path);
6955 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6956 key.type = BTRFS_ROOT_ITEM_KEY;
6959 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6963 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
6968 ret = btrfs_del_items(trans, root, path,
6975 btrfs_release_path(path);
6978 ret = btrfs_search_slot(trans, root, &key, path,
6985 leaf = path->nodes[0];
6986 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6987 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
6989 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
6994 del_slot = path->slots[0];
7003 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
7007 btrfs_release_path(path);
7010 key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7011 key.type = BTRFS_ROOT_ITEM_KEY;
7012 key.offset = (u64)-1;
7013 root = btrfs_read_fs_root(fs_info, &key);
7015 fprintf(stderr, "Error reading data reloc tree\n");
7016 return PTR_ERR(root);
7018 record_root_in_trans(trans, root);
7019 ret = btrfs_fsck_reinit_root(trans, root, 0);
7022 ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
7024 btrfs_free_path(path);
7028 static int reinit_extent_tree(struct btrfs_trans_handle *trans,
7029 struct btrfs_fs_info *fs_info)
7035 * The only reason we don't do this is because right now we're just
7036 * walking the trees we find and pinning down their bytes, we don't look
7037 * at any of the leaves. In order to do mixed groups we'd have to check
7038 * the leaves of any fs roots and pin down the bytes for any file
7039 * extents we find. Not hard but why do it if we don't have to?
7041 if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
7042 fprintf(stderr, "We don't support re-initing the extent tree "
7043 "for mixed block groups yet, please notify a btrfs "
7044 "developer you want to do this so they can add this "
7045 "functionality.\n");
7050 * first we need to walk all of the trees except the extent tree and pin
7051 * down the bytes that are in use so we don't overwrite any existing
7054 ret = pin_metadata_blocks(fs_info);
7056 fprintf(stderr, "error pinning down used bytes\n");
7061 * Need to drop all the block groups since we're going to recreate all
7064 btrfs_free_block_groups(fs_info);
7065 ret = reset_block_groups(fs_info);
7067 fprintf(stderr, "error resetting the block groups\n");
7071 /* Ok we can allocate now, reinit the extent root */
7072 ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 0);
7074 fprintf(stderr, "extent root initialization failed\n");
7076 * When the transaction code is updated we should end the
7077 * transaction, but for now progs only knows about commit so
7078 * just return an error.
7084 * Now we have all the in-memory block groups setup so we can make
7085 * allocations properly, and the metadata we care about is safe since we
7086 * pinned all of it above.
7089 struct btrfs_block_group_cache *cache;
7091 cache = btrfs_lookup_first_block_group(fs_info, start);
7094 start = cache->key.objectid + cache->key.offset;
7095 ret = btrfs_insert_item(trans, fs_info->extent_root,
7096 &cache->key, &cache->item,
7097 sizeof(cache->item));
7099 fprintf(stderr, "Error adding block group\n");
7102 btrfs_extent_post_op(trans, fs_info->extent_root);
7105 ret = reset_balance(trans, fs_info);
7107 fprintf(stderr, "error reseting the pending balance\n");
7112 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
7114 struct btrfs_path *path;
7115 struct btrfs_trans_handle *trans;
7116 struct btrfs_key key;
7119 printf("Recowing metadata block %llu\n", eb->start);
7120 key.objectid = btrfs_header_owner(eb);
7121 key.type = BTRFS_ROOT_ITEM_KEY;
7122 key.offset = (u64)-1;
7124 root = btrfs_read_fs_root(root->fs_info, &key);
7126 fprintf(stderr, "Couldn't find owner root %llu\n",
7128 return PTR_ERR(root);
7131 path = btrfs_alloc_path();
7135 trans = btrfs_start_transaction(root, 1);
7136 if (IS_ERR(trans)) {
7137 btrfs_free_path(path);
7138 return PTR_ERR(trans);
7141 path->lowest_level = btrfs_header_level(eb);
7142 if (path->lowest_level)
7143 btrfs_node_key_to_cpu(eb, &key, 0);
7145 btrfs_item_key_to_cpu(eb, &key, 0);
7147 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7148 btrfs_commit_transaction(trans, root);
7149 btrfs_free_path(path);
7153 static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
7155 struct btrfs_path *path;
7156 struct btrfs_trans_handle *trans;
7157 struct btrfs_key key;
7160 printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
7161 bad->key.type, bad->key.offset);
7162 key.objectid = bad->root_id;
7163 key.type = BTRFS_ROOT_ITEM_KEY;
7164 key.offset = (u64)-1;
7166 root = btrfs_read_fs_root(root->fs_info, &key);
7168 fprintf(stderr, "Couldn't find owner root %llu\n",
7170 return PTR_ERR(root);
7173 path = btrfs_alloc_path();
7177 trans = btrfs_start_transaction(root, 1);
7178 if (IS_ERR(trans)) {
7179 btrfs_free_path(path);
7180 return PTR_ERR(trans);
7183 ret = btrfs_search_slot(trans, root, &bad->key, path, -1, 1);
7189 ret = btrfs_del_item(trans, root, path);
7191 btrfs_commit_transaction(trans, root);
7192 btrfs_free_path(path);
7196 static int zero_log_tree(struct btrfs_root *root)
7198 struct btrfs_trans_handle *trans;
7201 trans = btrfs_start_transaction(root, 1);
7202 if (IS_ERR(trans)) {
7203 ret = PTR_ERR(trans);
7206 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
7207 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
7208 ret = btrfs_commit_transaction(trans, root);
7212 static int populate_csum(struct btrfs_trans_handle *trans,
7213 struct btrfs_root *csum_root, char *buf, u64 start,
7220 while (offset < len) {
7221 sectorsize = csum_root->sectorsize;
7222 ret = read_extent_data(csum_root, buf, start + offset,
7226 ret = btrfs_csum_file_block(trans, csum_root, start + len,
7227 start + offset, buf, sectorsize);
7230 offset += sectorsize;
7235 static int fill_csum_tree(struct btrfs_trans_handle *trans,
7236 struct btrfs_root *csum_root)
7238 struct btrfs_root *extent_root = csum_root->fs_info->extent_root;
7239 struct btrfs_path *path;
7240 struct btrfs_extent_item *ei;
7241 struct extent_buffer *leaf;
7243 struct btrfs_key key;
7246 path = btrfs_alloc_path();
7251 key.type = BTRFS_EXTENT_ITEM_KEY;
7254 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
7256 btrfs_free_path(path);
7260 buf = malloc(csum_root->sectorsize);
7262 btrfs_free_path(path);
7267 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7268 ret = btrfs_next_leaf(extent_root, path);
7276 leaf = path->nodes[0];
7278 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7279 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
7284 ei = btrfs_item_ptr(leaf, path->slots[0],
7285 struct btrfs_extent_item);
7286 if (!(btrfs_extent_flags(leaf, ei) &
7287 BTRFS_EXTENT_FLAG_DATA)) {
7292 ret = populate_csum(trans, csum_root, buf, key.objectid,
7299 btrfs_free_path(path);
7304 struct root_item_info {
7305 /* level of the root */
7307 /* number of nodes at this level, must be 1 for a root */
7311 struct cache_extent cache_extent;
7314 static struct cache_tree *roots_info_cache = NULL;
7316 static void free_roots_info_cache(void)
7318 if (!roots_info_cache)
7321 while (!cache_tree_empty(roots_info_cache)) {
7322 struct cache_extent *entry;
7323 struct root_item_info *rii;
7325 entry = first_cache_extent(roots_info_cache);
7326 remove_cache_extent(roots_info_cache, entry);
7327 rii = container_of(entry, struct root_item_info, cache_extent);
7331 free(roots_info_cache);
7332 roots_info_cache = NULL;
7335 static int build_roots_info_cache(struct btrfs_fs_info *info)
7338 struct btrfs_key key;
7339 struct extent_buffer *leaf;
7340 struct btrfs_path *path;
7342 if (!roots_info_cache) {
7343 roots_info_cache = malloc(sizeof(*roots_info_cache));
7344 if (!roots_info_cache)
7346 cache_tree_init(roots_info_cache);
7349 path = btrfs_alloc_path();
7354 key.type = BTRFS_EXTENT_ITEM_KEY;
7357 ret = btrfs_search_slot(NULL, info->extent_root, &key, path, 0, 0);
7360 leaf = path->nodes[0];
7363 struct btrfs_key found_key;
7364 struct btrfs_extent_item *ei;
7365 struct btrfs_extent_inline_ref *iref;
7366 int slot = path->slots[0];
7371 struct cache_extent *entry;
7372 struct root_item_info *rii;
7374 if (slot >= btrfs_header_nritems(leaf)) {
7375 ret = btrfs_next_leaf(info->extent_root, path);
7382 leaf = path->nodes[0];
7383 slot = path->slots[0];
7386 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7388 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
7389 found_key.type != BTRFS_METADATA_ITEM_KEY)
7392 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
7393 flags = btrfs_extent_flags(leaf, ei);
7395 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
7396 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
7399 if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
7400 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
7401 level = found_key.offset;
7403 struct btrfs_tree_block_info *info;
7405 info = (struct btrfs_tree_block_info *)(ei + 1);
7406 iref = (struct btrfs_extent_inline_ref *)(info + 1);
7407 level = btrfs_tree_block_level(leaf, info);
7411 * For a root extent, it must be of the following type and the
7412 * first (and only one) iref in the item.
7414 type = btrfs_extent_inline_ref_type(leaf, iref);
7415 if (type != BTRFS_TREE_BLOCK_REF_KEY)
7418 root_id = btrfs_extent_inline_ref_offset(leaf, iref);
7419 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
7421 rii = malloc(sizeof(struct root_item_info));
7426 rii->cache_extent.start = root_id;
7427 rii->cache_extent.size = 1;
7428 rii->level = (u8)-1;
7429 entry = &rii->cache_extent;
7430 ret = insert_cache_extent(roots_info_cache, entry);
7433 rii = container_of(entry, struct root_item_info,
7437 ASSERT(rii->cache_extent.start == root_id);
7438 ASSERT(rii->cache_extent.size == 1);
7440 if (level > rii->level || rii->level == (u8)-1) {
7442 rii->bytenr = found_key.objectid;
7443 rii->gen = btrfs_extent_generation(leaf, ei);
7444 rii->node_count = 1;
7445 } else if (level == rii->level) {
7453 btrfs_free_path(path);
7458 static int maybe_repair_root_item(struct btrfs_fs_info *info,
7459 struct btrfs_path *path,
7460 const struct btrfs_key *root_key,
7461 const int read_only_mode)
7463 const u64 root_id = root_key->objectid;
7464 struct cache_extent *entry;
7465 struct root_item_info *rii;
7466 struct btrfs_root_item ri;
7467 unsigned long offset;
7469 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
7472 "Error: could not find extent items for root %llu\n",
7473 root_key->objectid);
7477 rii = container_of(entry, struct root_item_info, cache_extent);
7478 ASSERT(rii->cache_extent.start == root_id);
7479 ASSERT(rii->cache_extent.size == 1);
7481 if (rii->node_count != 1) {
7483 "Error: could not find btree root extent for root %llu\n",
7488 offset = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
7489 read_extent_buffer(path->nodes[0], &ri, offset, sizeof(ri));
7491 if (btrfs_root_bytenr(&ri) != rii->bytenr ||
7492 btrfs_root_level(&ri) != rii->level ||
7493 btrfs_root_generation(&ri) != rii->gen) {
7496 * If we're in repair mode but our caller told us to not update
7497 * the root item, i.e. just check if it needs to be updated, don't
7498 * print this message, since the caller will call us again shortly
7499 * for the same root item without read only mode (the caller will
7500 * open a transaction first).
7502 if (!(read_only_mode && repair))
7504 "%sroot item for root %llu,"
7505 " current bytenr %llu, current gen %llu, current level %u,"
7506 " new bytenr %llu, new gen %llu, new level %u\n",
7507 (read_only_mode ? "" : "fixing "),
7509 btrfs_root_bytenr(&ri), btrfs_root_generation(&ri),
7510 btrfs_root_level(&ri),
7511 rii->bytenr, rii->gen, rii->level);
7513 if (btrfs_root_generation(&ri) > rii->gen) {
7515 "root %llu has a root item with a more recent gen (%llu) compared to the found root node (%llu)\n",
7516 root_id, btrfs_root_generation(&ri), rii->gen);
7520 if (!read_only_mode) {
7521 btrfs_set_root_bytenr(&ri, rii->bytenr);
7522 btrfs_set_root_level(&ri, rii->level);
7523 btrfs_set_root_generation(&ri, rii->gen);
7524 write_extent_buffer(path->nodes[0], &ri,
7525 offset, sizeof(ri));
7535 * A regression introduced in the 3.17 kernel (more specifically in 3.17-rc2),
7536 * caused read-only snapshots to be corrupted if they were created at a moment
7537 * when the source subvolume/snapshot had orphan items. The issue was that the
7538 * on-disk root items became incorrect, referring to the pre orphan cleanup root
7539 * node instead of the post orphan cleanup root node.
7540 * So this function, and its callees, just detects and fixes those cases. Even
7541 * though the regression was for read-only snapshots, this function applies to
7542 * any snapshot/subvolume root.
7543 * This must be run before any other repair code - not doing it so, makes other
7544 * repair code delete or modify backrefs in the extent tree for example, which
7545 * will result in an inconsistent fs after repairing the root items.
7547 static int repair_root_items(struct btrfs_fs_info *info)
7549 struct btrfs_path *path = NULL;
7550 struct btrfs_key key;
7551 struct extent_buffer *leaf;
7552 struct btrfs_trans_handle *trans = NULL;
7557 ret = build_roots_info_cache(info);
7561 path = btrfs_alloc_path();
7567 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
7568 key.type = BTRFS_ROOT_ITEM_KEY;
7573 * Avoid opening and committing transactions if a leaf doesn't have
7574 * any root items that need to be fixed, so that we avoid rotating
7575 * backup roots unnecessarily.
7578 trans = btrfs_start_transaction(info->tree_root, 1);
7579 if (IS_ERR(trans)) {
7580 ret = PTR_ERR(trans);
7585 ret = btrfs_search_slot(trans, info->tree_root, &key, path,
7589 leaf = path->nodes[0];
7592 struct btrfs_key found_key;
7594 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7595 int no_more_keys = find_next_key(path, &key);
7597 btrfs_release_path(path);
7599 ret = btrfs_commit_transaction(trans,
7611 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7613 if (found_key.type != BTRFS_ROOT_ITEM_KEY)
7616 ret = maybe_repair_root_item(info, path, &found_key,
7621 if (!trans && repair) {
7624 btrfs_release_path(path);
7634 free_roots_info_cache();
7636 btrfs_free_path(path);
7643 static struct option long_options[] = {
7644 { "super", 1, NULL, 's' },
7645 { "repair", 0, NULL, 0 },
7646 { "init-csum-tree", 0, NULL, 0 },
7647 { "init-extent-tree", 0, NULL, 0 },
7648 { "check-data-csum", 0, NULL, 0 },
7649 { "backup", 0, NULL, 0 },
7650 { "subvol-extents", 1, NULL, 'E' },
7651 { "qgroup-report", 0, NULL, 'Q' },
7655 const char * const cmd_check_usage[] = {
7656 "btrfs check [options] <device>",
7657 "Check an unmounted btrfs filesystem.",
7659 "-s|--super <superblock> use this superblock copy",
7660 "-b|--backup use the backup root copy",
7661 "--repair try to repair the filesystem",
7662 "--init-csum-tree create a new CRC tree",
7663 "--init-extent-tree create a new extent tree",
7664 "--check-data-csum verify checkums of data blocks",
7665 "--qgroup-report print a report on qgroup consistency",
7666 "--subvol-extents <subvolid> print subvolume extents and sharing state",
7670 int cmd_check(int argc, char **argv)
7672 struct cache_tree root_cache;
7673 struct btrfs_root *root;
7674 struct btrfs_fs_info *info;
7677 char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
7680 int option_index = 0;
7681 int init_csum_tree = 0;
7682 int qgroup_report = 0;
7683 enum btrfs_open_ctree_flags ctree_flags = OPEN_CTREE_EXCLUSIVE;
7687 c = getopt_long(argc, argv, "as:b", long_options,
7692 case 'a': /* ignored */ break;
7694 ctree_flags |= OPEN_CTREE_BACKUP_ROOT;
7697 num = arg_strtou64(optarg);
7698 if (num >= BTRFS_SUPER_MIRROR_MAX) {
7700 "ERROR: super mirror should be less than: %d\n",
7701 BTRFS_SUPER_MIRROR_MAX);
7704 bytenr = btrfs_sb_offset(((int)num));
7705 printf("using SB copy %llu, bytenr %llu\n", num,
7706 (unsigned long long)bytenr);
7712 subvolid = arg_strtou64(optarg);
7716 usage(cmd_check_usage);
7718 if (option_index == 1) {
7719 printf("enabling repair mode\n");
7721 ctree_flags |= OPEN_CTREE_WRITES;
7722 } else if (option_index == 2) {
7723 printf("Creating a new CRC tree\n");
7726 ctree_flags |= OPEN_CTREE_WRITES;
7727 } else if (option_index == 3) {
7728 init_extent_tree = 1;
7729 ctree_flags |= (OPEN_CTREE_WRITES |
7730 OPEN_CTREE_NO_BLOCK_GROUPS);
7732 } else if (option_index == 4) {
7733 check_data_csum = 1;
7736 argc = argc - optind;
7738 if (check_argc_exact(argc, 1))
7739 usage(cmd_check_usage);
7742 cache_tree_init(&root_cache);
7744 if((ret = check_mounted(argv[optind])) < 0) {
7745 fprintf(stderr, "Could not check mount status: %s\n", strerror(-ret));
7748 fprintf(stderr, "%s is currently mounted. Aborting.\n", argv[optind]);
7753 /* only allow partial opening under repair mode */
7755 ctree_flags |= OPEN_CTREE_PARTIAL;
7757 info = open_ctree_fs_info(argv[optind], bytenr, 0, ctree_flags);
7759 fprintf(stderr, "Couldn't open file system\n");
7764 root = info->fs_root;
7766 ret = repair_root_items(info);
7770 fprintf(stderr, "Fixed %d roots.\n", ret);
7772 } else if (ret > 0) {
7774 "Found %d roots with an outdated root item.\n",
7777 "Please run a filesystem check with the option --repair to fix them.\n");
7783 * repair mode will force us to commit transaction which
7784 * will make us fail to load log tree when mounting.
7786 if (repair && btrfs_super_log_root(info->super_copy)) {
7787 ret = ask_user("repair mode will force to clear out log tree, Are you sure?");
7792 ret = zero_log_tree(root);
7794 fprintf(stderr, "fail to zero log tree\n");
7799 uuid_unparse(info->super_copy->fsid, uuidbuf);
7800 if (qgroup_report) {
7801 printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
7803 ret = qgroup_verify_all(info);
7805 print_qgroup_report(1);
7809 printf("Print extent state for subvolume %llu on %s\nUUID: %s\n",
7810 subvolid, argv[optind], uuidbuf);
7811 ret = print_extent_state(info, subvolid);
7814 printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
7816 if (!extent_buffer_uptodate(info->tree_root->node) ||
7817 !extent_buffer_uptodate(info->dev_root->node) ||
7818 !extent_buffer_uptodate(info->chunk_root->node)) {
7819 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
7824 if (init_extent_tree || init_csum_tree) {
7825 struct btrfs_trans_handle *trans;
7827 trans = btrfs_start_transaction(info->extent_root, 0);
7828 if (IS_ERR(trans)) {
7829 fprintf(stderr, "Error starting transaction\n");
7830 ret = PTR_ERR(trans);
7834 if (init_extent_tree) {
7835 printf("Creating a new extent tree\n");
7836 ret = reinit_extent_tree(trans, info);
7841 if (init_csum_tree) {
7842 fprintf(stderr, "Reinit crc root\n");
7843 ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
7845 fprintf(stderr, "crc root initialization failed\n");
7850 ret = fill_csum_tree(trans, info->csum_root);
7852 fprintf(stderr, "crc refilling failed\n");
7857 * Ok now we commit and run the normal fsck, which will add
7858 * extent entries for all of the items it finds.
7860 ret = btrfs_commit_transaction(trans, info->extent_root);
7864 if (!extent_buffer_uptodate(info->extent_root->node)) {
7865 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
7869 if (!extent_buffer_uptodate(info->csum_root->node)) {
7870 fprintf(stderr, "Checksum root corrupted, rerun with --init-csum-tree option\n");
7875 fprintf(stderr, "checking extents\n");
7876 ret = check_chunks_and_extents(root);
7878 fprintf(stderr, "Errors found in extent allocation tree or chunk allocation\n");
7880 fprintf(stderr, "checking free space cache\n");
7881 ret = check_space_cache(root);
7886 * We used to have to have these hole extents in between our real
7887 * extents so if we don't have this flag set we need to make sure there
7888 * are no gaps in the file extents for inodes, otherwise we can just
7889 * ignore it when this happens.
7891 no_holes = btrfs_fs_incompat(root->fs_info,
7892 BTRFS_FEATURE_INCOMPAT_NO_HOLES);
7893 fprintf(stderr, "checking fs roots\n");
7894 ret = check_fs_roots(root, &root_cache);
7898 fprintf(stderr, "checking csums\n");
7899 ret = check_csums(root);
7903 fprintf(stderr, "checking root refs\n");
7904 ret = check_root_refs(root, &root_cache);
7908 while (repair && !list_empty(&root->fs_info->recow_ebs)) {
7909 struct extent_buffer *eb;
7911 eb = list_first_entry(&root->fs_info->recow_ebs,
7912 struct extent_buffer, recow);
7913 list_del_init(&eb->recow);
7914 ret = recow_extent_buffer(root, eb);
7919 while (!list_empty(&delete_items)) {
7920 struct bad_item *bad;
7922 bad = list_first_entry(&delete_items, struct bad_item, list);
7923 list_del_init(&bad->list);
7925 ret = delete_bad_item(root, bad);
7929 if (info->quota_enabled) {
7931 fprintf(stderr, "checking quota groups\n");
7932 err = qgroup_verify_all(info);
7937 if (!list_empty(&root->fs_info->recow_ebs)) {
7938 fprintf(stderr, "Transid errors in file system\n");
7942 print_qgroup_report(0);
7943 if (found_old_backref) { /*
7944 * there was a disk format change when mixed
7945 * backref was in testing tree. The old format
7946 * existed about one week.
7948 printf("\n * Found old mixed backref format. "
7949 "The old format is not supported! *"
7950 "\n * Please mount the FS in readonly mode, "
7951 "backup data and re-format the FS. *\n\n");
7954 printf("found %llu bytes used err is %d\n",
7955 (unsigned long long)bytes_used, ret);
7956 printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
7957 printf("total tree bytes: %llu\n",
7958 (unsigned long long)total_btree_bytes);
7959 printf("total fs tree bytes: %llu\n",
7960 (unsigned long long)total_fs_tree_bytes);
7961 printf("total extent tree bytes: %llu\n",
7962 (unsigned long long)total_extent_tree_bytes);
7963 printf("btree space waste bytes: %llu\n",
7964 (unsigned long long)btree_space_waste);
7965 printf("file data blocks allocated: %llu\n referenced %llu\n",
7966 (unsigned long long)data_bytes_allocated,
7967 (unsigned long long)data_bytes_referenced);
7968 printf("%s\n", BTRFS_BUILD_VERSION);
7970 free_root_recs_tree(&root_cache);