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)
1604 ret = btrfs_add_orphan_item(trans, root, path, rec->ino);
1605 btrfs_release_path(path);
1607 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
1611 static int add_missing_dir_index(struct btrfs_root *root,
1612 struct cache_tree *inode_cache,
1613 struct inode_record *rec,
1614 struct inode_backref *backref)
1616 struct btrfs_path *path;
1617 struct btrfs_trans_handle *trans;
1618 struct btrfs_dir_item *dir_item;
1619 struct extent_buffer *leaf;
1620 struct btrfs_key key;
1621 struct btrfs_disk_key disk_key;
1622 struct inode_record *dir_rec;
1623 unsigned long name_ptr;
1624 u32 data_size = sizeof(*dir_item) + backref->namelen;
1627 path = btrfs_alloc_path();
1631 trans = btrfs_start_transaction(root, 1);
1632 if (IS_ERR(trans)) {
1633 btrfs_free_path(path);
1634 return PTR_ERR(trans);
1637 fprintf(stderr, "repairing missing dir index item for inode %llu\n",
1638 (unsigned long long)rec->ino);
1639 key.objectid = backref->dir;
1640 key.type = BTRFS_DIR_INDEX_KEY;
1641 key.offset = backref->index;
1643 ret = btrfs_insert_empty_item(trans, root, path, &key, data_size);
1646 leaf = path->nodes[0];
1647 dir_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
1649 disk_key.objectid = cpu_to_le64(rec->ino);
1650 disk_key.type = BTRFS_INODE_ITEM_KEY;
1651 disk_key.offset = 0;
1653 btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
1654 btrfs_set_dir_type(leaf, dir_item, imode_to_type(rec->imode));
1655 btrfs_set_dir_data_len(leaf, dir_item, 0);
1656 btrfs_set_dir_name_len(leaf, dir_item, backref->namelen);
1657 name_ptr = (unsigned long)(dir_item + 1);
1658 write_extent_buffer(leaf, backref->name, name_ptr, backref->namelen);
1659 btrfs_mark_buffer_dirty(leaf);
1660 btrfs_free_path(path);
1661 btrfs_commit_transaction(trans, root);
1663 backref->found_dir_index = 1;
1664 dir_rec = get_inode_rec(inode_cache, backref->dir, 0);
1667 dir_rec->found_size += backref->namelen;
1668 if (dir_rec->found_size == dir_rec->isize &&
1669 (dir_rec->errors & I_ERR_DIR_ISIZE_WRONG))
1670 dir_rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
1671 if (dir_rec->found_size != dir_rec->isize)
1672 dir_rec->errors |= I_ERR_DIR_ISIZE_WRONG;
1677 static int delete_dir_index(struct btrfs_root *root,
1678 struct cache_tree *inode_cache,
1679 struct inode_record *rec,
1680 struct inode_backref *backref)
1682 struct btrfs_trans_handle *trans;
1683 struct btrfs_dir_item *di;
1684 struct btrfs_path *path;
1687 path = btrfs_alloc_path();
1691 trans = btrfs_start_transaction(root, 1);
1692 if (IS_ERR(trans)) {
1693 btrfs_free_path(path);
1694 return PTR_ERR(trans);
1698 fprintf(stderr, "Deleting bad dir index [%llu,%u,%llu] root %llu\n",
1699 (unsigned long long)backref->dir,
1700 BTRFS_DIR_INDEX_KEY, (unsigned long long)backref->index,
1701 (unsigned long long)root->objectid);
1703 di = btrfs_lookup_dir_index(trans, root, path, backref->dir,
1704 backref->name, backref->namelen,
1705 backref->index, -1);
1708 btrfs_free_path(path);
1709 btrfs_commit_transaction(trans, root);
1716 ret = btrfs_del_item(trans, root, path);
1718 ret = btrfs_delete_one_dir_name(trans, root, path, di);
1720 btrfs_free_path(path);
1721 btrfs_commit_transaction(trans, root);
1725 static int create_inode_item(struct btrfs_root *root,
1726 struct inode_record *rec,
1727 struct inode_backref *backref, int root_dir)
1729 struct btrfs_trans_handle *trans;
1730 struct btrfs_inode_item inode_item;
1731 time_t now = time(NULL);
1734 trans = btrfs_start_transaction(root, 1);
1735 if (IS_ERR(trans)) {
1736 ret = PTR_ERR(trans);
1740 fprintf(stderr, "root %llu inode %llu recreating inode item, this may "
1741 "be incomplete, please check permissions and content after "
1742 "the fsck completes.\n", (unsigned long long)root->objectid,
1743 (unsigned long long)rec->ino);
1745 memset(&inode_item, 0, sizeof(inode_item));
1746 btrfs_set_stack_inode_generation(&inode_item, trans->transid);
1748 btrfs_set_stack_inode_nlink(&inode_item, 1);
1750 btrfs_set_stack_inode_nlink(&inode_item, rec->found_link);
1751 btrfs_set_stack_inode_nbytes(&inode_item, rec->found_size);
1752 if (rec->found_dir_item) {
1753 if (rec->found_file_extent)
1754 fprintf(stderr, "root %llu inode %llu has both a dir "
1755 "item and extents, unsure if it is a dir or a "
1756 "regular file so setting it as a directory\n",
1757 (unsigned long long)root->objectid,
1758 (unsigned long long)rec->ino);
1759 btrfs_set_stack_inode_mode(&inode_item, S_IFDIR | 0755);
1760 btrfs_set_stack_inode_size(&inode_item, rec->found_size);
1761 } else if (!rec->found_dir_item) {
1762 btrfs_set_stack_inode_size(&inode_item, rec->extent_end);
1763 btrfs_set_stack_inode_mode(&inode_item, S_IFREG | 0755);
1765 btrfs_set_stack_timespec_sec(&inode_item.atime, now);
1766 btrfs_set_stack_timespec_nsec(&inode_item.atime, 0);
1767 btrfs_set_stack_timespec_sec(&inode_item.ctime, now);
1768 btrfs_set_stack_timespec_nsec(&inode_item.ctime, 0);
1769 btrfs_set_stack_timespec_sec(&inode_item.mtime, now);
1770 btrfs_set_stack_timespec_nsec(&inode_item.mtime, 0);
1771 btrfs_set_stack_timespec_sec(&inode_item.otime, 0);
1772 btrfs_set_stack_timespec_nsec(&inode_item.otime, 0);
1774 ret = btrfs_insert_inode(trans, root, rec->ino, &inode_item);
1776 btrfs_commit_transaction(trans, root);
1780 static int repair_inode_backrefs(struct btrfs_root *root,
1781 struct inode_record *rec,
1782 struct cache_tree *inode_cache,
1785 struct inode_backref *tmp, *backref;
1786 u64 root_dirid = btrfs_root_dirid(&root->root_item);
1790 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
1791 if (!delete && rec->ino == root_dirid) {
1792 if (!rec->found_inode_item) {
1793 ret = create_inode_item(root, rec, backref, 1);
1800 /* Index 0 for root dir's are special, don't mess with it */
1801 if (rec->ino == root_dirid && backref->index == 0)
1805 ((backref->found_dir_index && !backref->found_inode_ref) ||
1806 (backref->found_dir_index && backref->found_inode_ref &&
1807 (backref->errors & REF_ERR_INDEX_UNMATCH)))) {
1808 ret = delete_dir_index(root, inode_cache, rec, backref);
1812 list_del(&backref->list);
1816 if (!delete && !backref->found_dir_index &&
1817 backref->found_dir_item && backref->found_inode_ref) {
1818 ret = add_missing_dir_index(root, inode_cache, rec,
1823 if (backref->found_dir_item &&
1824 backref->found_dir_index &&
1825 backref->found_dir_index) {
1826 if (!backref->errors &&
1827 backref->found_inode_ref) {
1828 list_del(&backref->list);
1834 if (!delete && (!backref->found_dir_index &&
1835 !backref->found_dir_item &&
1836 backref->found_inode_ref)) {
1837 struct btrfs_trans_handle *trans;
1838 struct btrfs_key location;
1840 location.objectid = rec->ino;
1841 location.type = BTRFS_INODE_ITEM_KEY;
1842 location.offset = 0;
1844 trans = btrfs_start_transaction(root, 1);
1845 if (IS_ERR(trans)) {
1846 ret = PTR_ERR(trans);
1849 fprintf(stderr, "adding missing dir index/item pair "
1851 (unsigned long long)rec->ino);
1852 ret = btrfs_insert_dir_item(trans, root, backref->name,
1854 backref->dir, &location,
1855 imode_to_type(rec->imode),
1858 btrfs_commit_transaction(trans, root);
1862 if (!delete && (backref->found_inode_ref &&
1863 backref->found_dir_index &&
1864 backref->found_dir_item &&
1865 !(backref->errors & REF_ERR_INDEX_UNMATCH) &&
1866 !rec->found_inode_item)) {
1867 ret = create_inode_item(root, rec, backref, 0);
1874 return ret ? ret : repaired;
1877 static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
1879 struct btrfs_trans_handle *trans;
1880 struct btrfs_path *path;
1883 if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG | I_ERR_NO_ORPHAN_ITEM)))
1886 path = btrfs_alloc_path();
1890 trans = btrfs_start_transaction(root, 1);
1891 if (IS_ERR(trans)) {
1892 btrfs_free_path(path);
1893 return PTR_ERR(trans);
1896 if (rec->errors & I_ERR_DIR_ISIZE_WRONG)
1897 ret = repair_inode_isize(trans, root, path, rec);
1898 if (!ret && rec->errors & I_ERR_NO_ORPHAN_ITEM)
1899 ret = repair_inode_orphan_item(trans, root, path, rec);
1900 btrfs_commit_transaction(trans, root);
1901 btrfs_free_path(path);
1905 static int check_inode_recs(struct btrfs_root *root,
1906 struct cache_tree *inode_cache)
1908 struct cache_extent *cache;
1909 struct ptr_node *node;
1910 struct inode_record *rec;
1911 struct inode_backref *backref;
1916 u64 root_dirid = btrfs_root_dirid(&root->root_item);
1918 if (btrfs_root_refs(&root->root_item) == 0) {
1919 if (!cache_tree_empty(inode_cache))
1920 fprintf(stderr, "warning line %d\n", __LINE__);
1925 * We need to record the highest inode number for later 'lost+found'
1927 * We must select a ino not used/refered by any existing inode, or
1928 * 'lost+found' ino may be a missing ino in a corrupted leaf,
1929 * this may cause 'lost+found' dir has wrong nlinks.
1931 cache = last_cache_extent(inode_cache);
1933 node = container_of(cache, struct ptr_node, cache);
1935 if (rec->ino > root->highest_inode)
1936 root->highest_inode = rec->ino;
1940 * We need to repair backrefs first because we could change some of the
1941 * errors in the inode recs.
1943 * We also need to go through and delete invalid backrefs first and then
1944 * add the correct ones second. We do this because we may get EEXIST
1945 * when adding back the correct index because we hadn't yet deleted the
1948 * For example, if we were missing a dir index then the directories
1949 * isize would be wrong, so if we fixed the isize to what we thought it
1950 * would be and then fixed the backref we'd still have a invalid fs, so
1951 * we need to add back the dir index and then check to see if the isize
1956 if (stage == 3 && !err)
1959 cache = search_cache_extent(inode_cache, 0);
1960 while (repair && cache) {
1961 node = container_of(cache, struct ptr_node, cache);
1963 cache = next_cache_extent(cache);
1965 /* Need to free everything up and rescan */
1967 remove_cache_extent(inode_cache, &node->cache);
1969 free_inode_rec(rec);
1973 if (list_empty(&rec->backrefs))
1976 ret = repair_inode_backrefs(root, rec, inode_cache,
1990 rec = get_inode_rec(inode_cache, root_dirid, 0);
1992 ret = check_root_dir(rec);
1994 fprintf(stderr, "root %llu root dir %llu error\n",
1995 (unsigned long long)root->root_key.objectid,
1996 (unsigned long long)root_dirid);
1997 print_inode_error(root, rec);
2002 struct btrfs_trans_handle *trans;
2004 trans = btrfs_start_transaction(root, 1);
2005 if (IS_ERR(trans)) {
2006 err = PTR_ERR(trans);
2011 "root %llu missing its root dir, recreating\n",
2012 (unsigned long long)root->objectid);
2014 ret = btrfs_make_root_dir(trans, root, root_dirid);
2017 btrfs_commit_transaction(trans, root);
2021 fprintf(stderr, "root %llu root dir %llu not found\n",
2022 (unsigned long long)root->root_key.objectid,
2023 (unsigned long long)root_dirid);
2027 cache = search_cache_extent(inode_cache, 0);
2030 node = container_of(cache, struct ptr_node, cache);
2032 remove_cache_extent(inode_cache, &node->cache);
2034 if (rec->ino == root_dirid ||
2035 rec->ino == BTRFS_ORPHAN_OBJECTID) {
2036 free_inode_rec(rec);
2040 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
2041 ret = check_orphan_item(root, rec->ino);
2043 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
2044 if (can_free_inode_rec(rec)) {
2045 free_inode_rec(rec);
2051 ret = try_repair_inode(root, rec);
2052 if (ret == 0 && can_free_inode_rec(rec)) {
2053 free_inode_rec(rec);
2060 if (!rec->found_inode_item)
2061 rec->errors |= I_ERR_NO_INODE_ITEM;
2062 if (rec->found_link != rec->nlink)
2063 rec->errors |= I_ERR_LINK_COUNT_WRONG;
2064 print_inode_error(root, rec);
2065 list_for_each_entry(backref, &rec->backrefs, list) {
2066 if (!backref->found_dir_item)
2067 backref->errors |= REF_ERR_NO_DIR_ITEM;
2068 if (!backref->found_dir_index)
2069 backref->errors |= REF_ERR_NO_DIR_INDEX;
2070 if (!backref->found_inode_ref)
2071 backref->errors |= REF_ERR_NO_INODE_REF;
2072 fprintf(stderr, "\tunresolved ref dir %llu index %llu"
2073 " namelen %u name %s filetype %d errors %x",
2074 (unsigned long long)backref->dir,
2075 (unsigned long long)backref->index,
2076 backref->namelen, backref->name,
2077 backref->filetype, backref->errors);
2078 print_ref_error(backref->errors);
2080 free_inode_rec(rec);
2082 return (error > 0) ? -1 : 0;
2085 static struct root_record *get_root_rec(struct cache_tree *root_cache,
2088 struct cache_extent *cache;
2089 struct root_record *rec = NULL;
2092 cache = lookup_cache_extent(root_cache, objectid, 1);
2094 rec = container_of(cache, struct root_record, cache);
2096 rec = calloc(1, sizeof(*rec));
2097 rec->objectid = objectid;
2098 INIT_LIST_HEAD(&rec->backrefs);
2099 rec->cache.start = objectid;
2100 rec->cache.size = 1;
2102 ret = insert_cache_extent(root_cache, &rec->cache);
2108 static struct root_backref *get_root_backref(struct root_record *rec,
2109 u64 ref_root, u64 dir, u64 index,
2110 const char *name, int namelen)
2112 struct root_backref *backref;
2114 list_for_each_entry(backref, &rec->backrefs, list) {
2115 if (backref->ref_root != ref_root || backref->dir != dir ||
2116 backref->namelen != namelen)
2118 if (memcmp(name, backref->name, namelen))
2123 backref = malloc(sizeof(*backref) + namelen + 1);
2124 memset(backref, 0, sizeof(*backref));
2125 backref->ref_root = ref_root;
2127 backref->index = index;
2128 backref->namelen = namelen;
2129 memcpy(backref->name, name, namelen);
2130 backref->name[namelen] = '\0';
2131 list_add_tail(&backref->list, &rec->backrefs);
2135 static void free_root_record(struct cache_extent *cache)
2137 struct root_record *rec;
2138 struct root_backref *backref;
2140 rec = container_of(cache, struct root_record, cache);
2141 while (!list_empty(&rec->backrefs)) {
2142 backref = list_entry(rec->backrefs.next,
2143 struct root_backref, list);
2144 list_del(&backref->list);
2151 FREE_EXTENT_CACHE_BASED_TREE(root_recs, free_root_record);
2153 static int add_root_backref(struct cache_tree *root_cache,
2154 u64 root_id, u64 ref_root, u64 dir, u64 index,
2155 const char *name, int namelen,
2156 int item_type, int errors)
2158 struct root_record *rec;
2159 struct root_backref *backref;
2161 rec = get_root_rec(root_cache, root_id);
2162 backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
2164 backref->errors |= errors;
2166 if (item_type != BTRFS_DIR_ITEM_KEY) {
2167 if (backref->found_dir_index || backref->found_back_ref ||
2168 backref->found_forward_ref) {
2169 if (backref->index != index)
2170 backref->errors |= REF_ERR_INDEX_UNMATCH;
2172 backref->index = index;
2176 if (item_type == BTRFS_DIR_ITEM_KEY) {
2177 if (backref->found_forward_ref)
2179 backref->found_dir_item = 1;
2180 } else if (item_type == BTRFS_DIR_INDEX_KEY) {
2181 backref->found_dir_index = 1;
2182 } else if (item_type == BTRFS_ROOT_REF_KEY) {
2183 if (backref->found_forward_ref)
2184 backref->errors |= REF_ERR_DUP_ROOT_REF;
2185 else if (backref->found_dir_item)
2187 backref->found_forward_ref = 1;
2188 } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
2189 if (backref->found_back_ref)
2190 backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
2191 backref->found_back_ref = 1;
2196 if (backref->found_forward_ref && backref->found_dir_item)
2197 backref->reachable = 1;
2201 static int merge_root_recs(struct btrfs_root *root,
2202 struct cache_tree *src_cache,
2203 struct cache_tree *dst_cache)
2205 struct cache_extent *cache;
2206 struct ptr_node *node;
2207 struct inode_record *rec;
2208 struct inode_backref *backref;
2211 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2212 free_inode_recs_tree(src_cache);
2217 cache = search_cache_extent(src_cache, 0);
2220 node = container_of(cache, struct ptr_node, cache);
2222 remove_cache_extent(src_cache, &node->cache);
2225 ret = is_child_root(root, root->objectid, rec->ino);
2231 list_for_each_entry(backref, &rec->backrefs, list) {
2232 BUG_ON(backref->found_inode_ref);
2233 if (backref->found_dir_item)
2234 add_root_backref(dst_cache, rec->ino,
2235 root->root_key.objectid, backref->dir,
2236 backref->index, backref->name,
2237 backref->namelen, BTRFS_DIR_ITEM_KEY,
2239 if (backref->found_dir_index)
2240 add_root_backref(dst_cache, rec->ino,
2241 root->root_key.objectid, backref->dir,
2242 backref->index, backref->name,
2243 backref->namelen, BTRFS_DIR_INDEX_KEY,
2247 free_inode_rec(rec);
2254 static int check_root_refs(struct btrfs_root *root,
2255 struct cache_tree *root_cache)
2257 struct root_record *rec;
2258 struct root_record *ref_root;
2259 struct root_backref *backref;
2260 struct cache_extent *cache;
2266 rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
2269 /* fixme: this can not detect circular references */
2272 cache = search_cache_extent(root_cache, 0);
2276 rec = container_of(cache, struct root_record, cache);
2277 cache = next_cache_extent(cache);
2279 if (rec->found_ref == 0)
2282 list_for_each_entry(backref, &rec->backrefs, list) {
2283 if (!backref->reachable)
2286 ref_root = get_root_rec(root_cache,
2288 if (ref_root->found_ref > 0)
2291 backref->reachable = 0;
2293 if (rec->found_ref == 0)
2299 cache = search_cache_extent(root_cache, 0);
2303 rec = container_of(cache, struct root_record, cache);
2304 cache = next_cache_extent(cache);
2306 if (rec->found_ref == 0 &&
2307 rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
2308 rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
2309 ret = check_orphan_item(root->fs_info->tree_root,
2315 * If we don't have a root item then we likely just have
2316 * a dir item in a snapshot for this root but no actual
2317 * ref key or anything so it's meaningless.
2319 if (!rec->found_root_item)
2322 fprintf(stderr, "fs tree %llu not referenced\n",
2323 (unsigned long long)rec->objectid);
2327 if (rec->found_ref > 0 && !rec->found_root_item)
2329 list_for_each_entry(backref, &rec->backrefs, list) {
2330 if (!backref->found_dir_item)
2331 backref->errors |= REF_ERR_NO_DIR_ITEM;
2332 if (!backref->found_dir_index)
2333 backref->errors |= REF_ERR_NO_DIR_INDEX;
2334 if (!backref->found_back_ref)
2335 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
2336 if (!backref->found_forward_ref)
2337 backref->errors |= REF_ERR_NO_ROOT_REF;
2338 if (backref->reachable && backref->errors)
2345 fprintf(stderr, "fs tree %llu refs %u %s\n",
2346 (unsigned long long)rec->objectid, rec->found_ref,
2347 rec->found_root_item ? "" : "not found");
2349 list_for_each_entry(backref, &rec->backrefs, list) {
2350 if (!backref->reachable)
2352 if (!backref->errors && rec->found_root_item)
2354 fprintf(stderr, "\tunresolved ref root %llu dir %llu"
2355 " index %llu namelen %u name %s errors %x\n",
2356 (unsigned long long)backref->ref_root,
2357 (unsigned long long)backref->dir,
2358 (unsigned long long)backref->index,
2359 backref->namelen, backref->name,
2361 print_ref_error(backref->errors);
2364 return errors > 0 ? 1 : 0;
2367 static int process_root_ref(struct extent_buffer *eb, int slot,
2368 struct btrfs_key *key,
2369 struct cache_tree *root_cache)
2375 struct btrfs_root_ref *ref;
2376 char namebuf[BTRFS_NAME_LEN];
2379 ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
2381 dirid = btrfs_root_ref_dirid(eb, ref);
2382 index = btrfs_root_ref_sequence(eb, ref);
2383 name_len = btrfs_root_ref_name_len(eb, ref);
2385 if (name_len <= BTRFS_NAME_LEN) {
2389 len = BTRFS_NAME_LEN;
2390 error = REF_ERR_NAME_TOO_LONG;
2392 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
2394 if (key->type == BTRFS_ROOT_REF_KEY) {
2395 add_root_backref(root_cache, key->offset, key->objectid, dirid,
2396 index, namebuf, len, key->type, error);
2398 add_root_backref(root_cache, key->objectid, key->offset, dirid,
2399 index, namebuf, len, key->type, error);
2404 static int check_fs_root(struct btrfs_root *root,
2405 struct cache_tree *root_cache,
2406 struct walk_control *wc)
2412 struct btrfs_path path;
2413 struct shared_node root_node;
2414 struct root_record *rec;
2415 struct btrfs_root_item *root_item = &root->root_item;
2416 enum btrfs_tree_block_status status;
2418 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2419 rec = get_root_rec(root_cache, root->root_key.objectid);
2420 if (btrfs_root_refs(root_item) > 0)
2421 rec->found_root_item = 1;
2424 btrfs_init_path(&path);
2425 memset(&root_node, 0, sizeof(root_node));
2426 cache_tree_init(&root_node.root_cache);
2427 cache_tree_init(&root_node.inode_cache);
2429 level = btrfs_header_level(root->node);
2430 memset(wc->nodes, 0, sizeof(wc->nodes));
2431 wc->nodes[level] = &root_node;
2432 wc->active_node = level;
2433 wc->root_level = level;
2435 /* We may not have checked the root block, lets do that now */
2436 if (btrfs_is_leaf(root->node))
2437 status = btrfs_check_leaf(root, NULL, root->node);
2439 status = btrfs_check_node(root, NULL, root->node);
2440 if (status != BTRFS_TREE_BLOCK_CLEAN)
2443 if (btrfs_root_refs(root_item) > 0 ||
2444 btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2445 path.nodes[level] = root->node;
2446 extent_buffer_get(root->node);
2447 path.slots[level] = 0;
2449 struct btrfs_key key;
2450 struct btrfs_disk_key found_key;
2452 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2453 level = root_item->drop_level;
2454 path.lowest_level = level;
2455 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
2458 btrfs_node_key(path.nodes[level], &found_key,
2460 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2461 sizeof(found_key)));
2465 wret = walk_down_tree(root, &path, wc, &level);
2471 wret = walk_up_tree(root, &path, wc, &level);
2478 btrfs_release_path(&path);
2480 err = merge_root_recs(root, &root_node.root_cache, root_cache);
2484 if (root_node.current) {
2485 root_node.current->checked = 1;
2486 maybe_free_inode_rec(&root_node.inode_cache,
2490 err = check_inode_recs(root, &root_node.inode_cache);
2496 static int fs_root_objectid(u64 objectid)
2498 if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
2499 objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2501 return is_fstree(objectid);
2504 static int check_fs_roots(struct btrfs_root *root,
2505 struct cache_tree *root_cache)
2507 struct btrfs_path path;
2508 struct btrfs_key key;
2509 struct walk_control wc;
2510 struct extent_buffer *leaf, *tree_node;
2511 struct btrfs_root *tmp_root;
2512 struct btrfs_root *tree_root = root->fs_info->tree_root;
2517 * Just in case we made any changes to the extent tree that weren't
2518 * reflected into the free space cache yet.
2521 reset_cached_block_groups(root->fs_info);
2522 memset(&wc, 0, sizeof(wc));
2523 cache_tree_init(&wc.shared);
2524 btrfs_init_path(&path);
2529 key.type = BTRFS_ROOT_ITEM_KEY;
2530 ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
2535 tree_node = tree_root->node;
2537 if (tree_node != tree_root->node) {
2538 free_root_recs_tree(root_cache);
2539 btrfs_release_path(&path);
2542 leaf = path.nodes[0];
2543 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
2544 ret = btrfs_next_leaf(tree_root, &path);
2550 leaf = path.nodes[0];
2552 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
2553 if (key.type == BTRFS_ROOT_ITEM_KEY &&
2554 fs_root_objectid(key.objectid)) {
2555 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2556 tmp_root = btrfs_read_fs_root_no_cache(
2557 root->fs_info, &key);
2559 key.offset = (u64)-1;
2560 tmp_root = btrfs_read_fs_root(
2561 root->fs_info, &key);
2563 if (IS_ERR(tmp_root)) {
2567 ret = check_fs_root(tmp_root, root_cache, &wc);
2568 if (ret == -EAGAIN) {
2569 free_root_recs_tree(root_cache);
2570 btrfs_release_path(&path);
2575 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
2576 btrfs_free_fs_root(tmp_root);
2577 } else if (key.type == BTRFS_ROOT_REF_KEY ||
2578 key.type == BTRFS_ROOT_BACKREF_KEY) {
2579 process_root_ref(leaf, path.slots[0], &key,
2586 btrfs_release_path(&path);
2588 free_extent_cache_tree(&wc.shared);
2589 if (!cache_tree_empty(&wc.shared))
2590 fprintf(stderr, "warning line %d\n", __LINE__);
2595 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
2597 struct list_head *cur = rec->backrefs.next;
2598 struct extent_backref *back;
2599 struct tree_backref *tback;
2600 struct data_backref *dback;
2604 while(cur != &rec->backrefs) {
2605 back = list_entry(cur, struct extent_backref, list);
2607 if (!back->found_extent_tree) {
2611 if (back->is_data) {
2612 dback = (struct data_backref *)back;
2613 fprintf(stderr, "Backref %llu %s %llu"
2614 " owner %llu offset %llu num_refs %lu"
2615 " not found in extent tree\n",
2616 (unsigned long long)rec->start,
2617 back->full_backref ?
2619 back->full_backref ?
2620 (unsigned long long)dback->parent:
2621 (unsigned long long)dback->root,
2622 (unsigned long long)dback->owner,
2623 (unsigned long long)dback->offset,
2624 (unsigned long)dback->num_refs);
2626 tback = (struct tree_backref *)back;
2627 fprintf(stderr, "Backref %llu parent %llu"
2628 " root %llu not found in extent tree\n",
2629 (unsigned long long)rec->start,
2630 (unsigned long long)tback->parent,
2631 (unsigned long long)tback->root);
2634 if (!back->is_data && !back->found_ref) {
2638 tback = (struct tree_backref *)back;
2639 fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
2640 (unsigned long long)rec->start,
2641 back->full_backref ? "parent" : "root",
2642 back->full_backref ?
2643 (unsigned long long)tback->parent :
2644 (unsigned long long)tback->root, back);
2646 if (back->is_data) {
2647 dback = (struct data_backref *)back;
2648 if (dback->found_ref != dback->num_refs) {
2652 fprintf(stderr, "Incorrect local backref count"
2653 " on %llu %s %llu owner %llu"
2654 " offset %llu found %u wanted %u back %p\n",
2655 (unsigned long long)rec->start,
2656 back->full_backref ?
2658 back->full_backref ?
2659 (unsigned long long)dback->parent:
2660 (unsigned long long)dback->root,
2661 (unsigned long long)dback->owner,
2662 (unsigned long long)dback->offset,
2663 dback->found_ref, dback->num_refs, back);
2665 if (dback->disk_bytenr != rec->start) {
2669 fprintf(stderr, "Backref disk bytenr does not"
2670 " match extent record, bytenr=%llu, "
2671 "ref bytenr=%llu\n",
2672 (unsigned long long)rec->start,
2673 (unsigned long long)dback->disk_bytenr);
2676 if (dback->bytes != rec->nr) {
2680 fprintf(stderr, "Backref bytes do not match "
2681 "extent backref, bytenr=%llu, ref "
2682 "bytes=%llu, backref bytes=%llu\n",
2683 (unsigned long long)rec->start,
2684 (unsigned long long)rec->nr,
2685 (unsigned long long)dback->bytes);
2688 if (!back->is_data) {
2691 dback = (struct data_backref *)back;
2692 found += dback->found_ref;
2695 if (found != rec->refs) {
2699 fprintf(stderr, "Incorrect global backref count "
2700 "on %llu found %llu wanted %llu\n",
2701 (unsigned long long)rec->start,
2702 (unsigned long long)found,
2703 (unsigned long long)rec->refs);
2709 static int free_all_extent_backrefs(struct extent_record *rec)
2711 struct extent_backref *back;
2712 struct list_head *cur;
2713 while (!list_empty(&rec->backrefs)) {
2714 cur = rec->backrefs.next;
2715 back = list_entry(cur, struct extent_backref, list);
2722 static void free_extent_record_cache(struct btrfs_fs_info *fs_info,
2723 struct cache_tree *extent_cache)
2725 struct cache_extent *cache;
2726 struct extent_record *rec;
2729 cache = first_cache_extent(extent_cache);
2732 rec = container_of(cache, struct extent_record, cache);
2733 btrfs_unpin_extent(fs_info, rec->start, rec->max_size);
2734 remove_cache_extent(extent_cache, cache);
2735 free_all_extent_backrefs(rec);
2740 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
2741 struct extent_record *rec)
2743 if (rec->content_checked && rec->owner_ref_checked &&
2744 rec->extent_item_refs == rec->refs && rec->refs > 0 &&
2745 rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0)) {
2746 remove_cache_extent(extent_cache, &rec->cache);
2747 free_all_extent_backrefs(rec);
2748 list_del_init(&rec->list);
2754 static int check_owner_ref(struct btrfs_root *root,
2755 struct extent_record *rec,
2756 struct extent_buffer *buf)
2758 struct extent_backref *node;
2759 struct tree_backref *back;
2760 struct btrfs_root *ref_root;
2761 struct btrfs_key key;
2762 struct btrfs_path path;
2763 struct extent_buffer *parent;
2768 list_for_each_entry(node, &rec->backrefs, list) {
2771 if (!node->found_ref)
2773 if (node->full_backref)
2775 back = (struct tree_backref *)node;
2776 if (btrfs_header_owner(buf) == back->root)
2779 BUG_ON(rec->is_root);
2781 /* try to find the block by search corresponding fs tree */
2782 key.objectid = btrfs_header_owner(buf);
2783 key.type = BTRFS_ROOT_ITEM_KEY;
2784 key.offset = (u64)-1;
2786 ref_root = btrfs_read_fs_root(root->fs_info, &key);
2787 if (IS_ERR(ref_root))
2790 level = btrfs_header_level(buf);
2792 btrfs_item_key_to_cpu(buf, &key, 0);
2794 btrfs_node_key_to_cpu(buf, &key, 0);
2796 btrfs_init_path(&path);
2797 path.lowest_level = level + 1;
2798 ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
2802 parent = path.nodes[level + 1];
2803 if (parent && buf->start == btrfs_node_blockptr(parent,
2804 path.slots[level + 1]))
2807 btrfs_release_path(&path);
2808 return found ? 0 : 1;
2811 static int is_extent_tree_record(struct extent_record *rec)
2813 struct list_head *cur = rec->backrefs.next;
2814 struct extent_backref *node;
2815 struct tree_backref *back;
2818 while(cur != &rec->backrefs) {
2819 node = list_entry(cur, struct extent_backref, list);
2823 back = (struct tree_backref *)node;
2824 if (node->full_backref)
2826 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
2833 static int record_bad_block_io(struct btrfs_fs_info *info,
2834 struct cache_tree *extent_cache,
2837 struct extent_record *rec;
2838 struct cache_extent *cache;
2839 struct btrfs_key key;
2841 cache = lookup_cache_extent(extent_cache, start, len);
2845 rec = container_of(cache, struct extent_record, cache);
2846 if (!is_extent_tree_record(rec))
2849 btrfs_disk_key_to_cpu(&key, &rec->parent_key);
2850 return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
2853 static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
2854 struct extent_buffer *buf, int slot)
2856 if (btrfs_header_level(buf)) {
2857 struct btrfs_key_ptr ptr1, ptr2;
2859 read_extent_buffer(buf, &ptr1, btrfs_node_key_ptr_offset(slot),
2860 sizeof(struct btrfs_key_ptr));
2861 read_extent_buffer(buf, &ptr2,
2862 btrfs_node_key_ptr_offset(slot + 1),
2863 sizeof(struct btrfs_key_ptr));
2864 write_extent_buffer(buf, &ptr1,
2865 btrfs_node_key_ptr_offset(slot + 1),
2866 sizeof(struct btrfs_key_ptr));
2867 write_extent_buffer(buf, &ptr2,
2868 btrfs_node_key_ptr_offset(slot),
2869 sizeof(struct btrfs_key_ptr));
2871 struct btrfs_disk_key key;
2872 btrfs_node_key(buf, &key, 0);
2873 btrfs_fixup_low_keys(root, path, &key,
2874 btrfs_header_level(buf) + 1);
2877 struct btrfs_item *item1, *item2;
2878 struct btrfs_key k1, k2;
2879 char *item1_data, *item2_data;
2880 u32 item1_offset, item2_offset, item1_size, item2_size;
2882 item1 = btrfs_item_nr(slot);
2883 item2 = btrfs_item_nr(slot + 1);
2884 btrfs_item_key_to_cpu(buf, &k1, slot);
2885 btrfs_item_key_to_cpu(buf, &k2, slot + 1);
2886 item1_offset = btrfs_item_offset(buf, item1);
2887 item2_offset = btrfs_item_offset(buf, item2);
2888 item1_size = btrfs_item_size(buf, item1);
2889 item2_size = btrfs_item_size(buf, item2);
2891 item1_data = malloc(item1_size);
2894 item2_data = malloc(item2_size);
2900 read_extent_buffer(buf, item1_data, item1_offset, item1_size);
2901 read_extent_buffer(buf, item2_data, item2_offset, item2_size);
2903 write_extent_buffer(buf, item1_data, item2_offset, item2_size);
2904 write_extent_buffer(buf, item2_data, item1_offset, item1_size);
2908 btrfs_set_item_offset(buf, item1, item2_offset);
2909 btrfs_set_item_offset(buf, item2, item1_offset);
2910 btrfs_set_item_size(buf, item1, item2_size);
2911 btrfs_set_item_size(buf, item2, item1_size);
2913 path->slots[0] = slot;
2914 btrfs_set_item_key_unsafe(root, path, &k2);
2915 path->slots[0] = slot + 1;
2916 btrfs_set_item_key_unsafe(root, path, &k1);
2921 static int fix_key_order(struct btrfs_trans_handle *trans,
2922 struct btrfs_root *root,
2923 struct btrfs_path *path)
2925 struct extent_buffer *buf;
2926 struct btrfs_key k1, k2;
2928 int level = path->lowest_level;
2931 buf = path->nodes[level];
2932 for (i = 0; i < btrfs_header_nritems(buf) - 1; i++) {
2934 btrfs_node_key_to_cpu(buf, &k1, i);
2935 btrfs_node_key_to_cpu(buf, &k2, i + 1);
2937 btrfs_item_key_to_cpu(buf, &k1, i);
2938 btrfs_item_key_to_cpu(buf, &k2, i + 1);
2940 if (btrfs_comp_cpu_keys(&k1, &k2) < 0)
2942 ret = swap_values(root, path, buf, i);
2945 btrfs_mark_buffer_dirty(buf);
2951 static int delete_bogus_item(struct btrfs_trans_handle *trans,
2952 struct btrfs_root *root,
2953 struct btrfs_path *path,
2954 struct extent_buffer *buf, int slot)
2956 struct btrfs_key key;
2957 int nritems = btrfs_header_nritems(buf);
2959 btrfs_item_key_to_cpu(buf, &key, slot);
2961 /* These are all the keys we can deal with missing. */
2962 if (key.type != BTRFS_DIR_INDEX_KEY &&
2963 key.type != BTRFS_EXTENT_ITEM_KEY &&
2964 key.type != BTRFS_METADATA_ITEM_KEY &&
2965 key.type != BTRFS_TREE_BLOCK_REF_KEY &&
2966 key.type != BTRFS_EXTENT_DATA_REF_KEY)
2969 printf("Deleting bogus item [%llu,%u,%llu] at slot %d on block %llu\n",
2970 (unsigned long long)key.objectid, key.type,
2971 (unsigned long long)key.offset, slot, buf->start);
2972 memmove_extent_buffer(buf, btrfs_item_nr_offset(slot),
2973 btrfs_item_nr_offset(slot + 1),
2974 sizeof(struct btrfs_item) *
2975 (nritems - slot - 1));
2976 btrfs_set_header_nritems(buf, nritems - 1);
2978 struct btrfs_disk_key disk_key;
2980 btrfs_item_key(buf, &disk_key, 0);
2981 btrfs_fixup_low_keys(root, path, &disk_key, 1);
2983 btrfs_mark_buffer_dirty(buf);
2987 static int fix_item_offset(struct btrfs_trans_handle *trans,
2988 struct btrfs_root *root,
2989 struct btrfs_path *path)
2991 struct extent_buffer *buf;
2995 /* We should only get this for leaves */
2996 BUG_ON(path->lowest_level);
2997 buf = path->nodes[0];
2999 for (i = 0; i < btrfs_header_nritems(buf); i++) {
3000 unsigned int shift = 0, offset;
3002 if (i == 0 && btrfs_item_end_nr(buf, i) !=
3003 BTRFS_LEAF_DATA_SIZE(root)) {
3004 if (btrfs_item_end_nr(buf, i) >
3005 BTRFS_LEAF_DATA_SIZE(root)) {
3006 ret = delete_bogus_item(trans, root, path,
3010 fprintf(stderr, "item is off the end of the "
3011 "leaf, can't fix\n");
3015 shift = BTRFS_LEAF_DATA_SIZE(root) -
3016 btrfs_item_end_nr(buf, i);
3017 } else if (i > 0 && btrfs_item_end_nr(buf, i) !=
3018 btrfs_item_offset_nr(buf, i - 1)) {
3019 if (btrfs_item_end_nr(buf, i) >
3020 btrfs_item_offset_nr(buf, i - 1)) {
3021 ret = delete_bogus_item(trans, root, path,
3025 fprintf(stderr, "items overlap, can't fix\n");
3029 shift = btrfs_item_offset_nr(buf, i - 1) -
3030 btrfs_item_end_nr(buf, i);
3035 printf("Shifting item nr %d by %u bytes in block %llu\n",
3036 i, shift, (unsigned long long)buf->start);
3037 offset = btrfs_item_offset_nr(buf, i);
3038 memmove_extent_buffer(buf,
3039 btrfs_leaf_data(buf) + offset + shift,
3040 btrfs_leaf_data(buf) + offset,
3041 btrfs_item_size_nr(buf, i));
3042 btrfs_set_item_offset(buf, btrfs_item_nr(i),
3044 btrfs_mark_buffer_dirty(buf);
3048 * We may have moved things, in which case we want to exit so we don't
3049 * write those changes out. Once we have proper abort functionality in
3050 * progs this can be changed to something nicer.
3057 * Attempt to fix basic block failures. If we can't fix it for whatever reason
3058 * then just return -EIO.
3060 static int try_to_fix_bad_block(struct btrfs_trans_handle *trans,
3061 struct btrfs_root *root,
3062 struct extent_buffer *buf,
3063 enum btrfs_tree_block_status status)
3065 struct ulist *roots;
3066 struct ulist_node *node;
3067 struct btrfs_root *search_root;
3068 struct btrfs_path *path;
3069 struct ulist_iterator iter;
3070 struct btrfs_key root_key, key;
3073 if (status != BTRFS_TREE_BLOCK_BAD_KEY_ORDER &&
3074 status != BTRFS_TREE_BLOCK_INVALID_OFFSETS)
3077 path = btrfs_alloc_path();
3081 ret = btrfs_find_all_roots(trans, root->fs_info, buf->start,
3084 btrfs_free_path(path);
3088 ULIST_ITER_INIT(&iter);
3089 while ((node = ulist_next(roots, &iter))) {
3090 root_key.objectid = node->val;
3091 root_key.type = BTRFS_ROOT_ITEM_KEY;
3092 root_key.offset = (u64)-1;
3094 search_root = btrfs_read_fs_root(root->fs_info, &root_key);
3100 record_root_in_trans(trans, search_root);
3102 path->lowest_level = btrfs_header_level(buf);
3103 path->skip_check_block = 1;
3104 if (path->lowest_level)
3105 btrfs_node_key_to_cpu(buf, &key, 0);
3107 btrfs_item_key_to_cpu(buf, &key, 0);
3108 ret = btrfs_search_slot(trans, search_root, &key, path, 0, 1);
3113 if (status == BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
3114 ret = fix_key_order(trans, search_root, path);
3115 else if (status == BTRFS_TREE_BLOCK_INVALID_OFFSETS)
3116 ret = fix_item_offset(trans, search_root, path);
3119 btrfs_release_path(path);
3122 btrfs_free_path(path);
3126 static int check_block(struct btrfs_trans_handle *trans,
3127 struct btrfs_root *root,
3128 struct cache_tree *extent_cache,
3129 struct extent_buffer *buf, u64 flags)
3131 struct extent_record *rec;
3132 struct cache_extent *cache;
3133 struct btrfs_key key;
3134 enum btrfs_tree_block_status status;
3138 cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
3141 rec = container_of(cache, struct extent_record, cache);
3142 rec->generation = btrfs_header_generation(buf);
3144 level = btrfs_header_level(buf);
3145 if (btrfs_header_nritems(buf) > 0) {
3148 btrfs_item_key_to_cpu(buf, &key, 0);
3150 btrfs_node_key_to_cpu(buf, &key, 0);
3152 rec->info_objectid = key.objectid;
3154 rec->info_level = level;
3156 if (btrfs_is_leaf(buf))
3157 status = btrfs_check_leaf(root, &rec->parent_key, buf);
3159 status = btrfs_check_node(root, &rec->parent_key, buf);
3161 if (status != BTRFS_TREE_BLOCK_CLEAN) {
3163 status = try_to_fix_bad_block(trans, root, buf,
3165 if (status != BTRFS_TREE_BLOCK_CLEAN) {
3167 fprintf(stderr, "bad block %llu\n",
3168 (unsigned long long)buf->start);
3171 * Signal to callers we need to start the scan over
3172 * again since we'll have cow'ed blocks.
3177 rec->content_checked = 1;
3178 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3179 rec->owner_ref_checked = 1;
3181 ret = check_owner_ref(root, rec, buf);
3183 rec->owner_ref_checked = 1;
3187 maybe_free_extent_rec(extent_cache, rec);
3191 static struct tree_backref *find_tree_backref(struct extent_record *rec,
3192 u64 parent, u64 root)
3194 struct list_head *cur = rec->backrefs.next;
3195 struct extent_backref *node;
3196 struct tree_backref *back;
3198 while(cur != &rec->backrefs) {
3199 node = list_entry(cur, struct extent_backref, list);
3203 back = (struct tree_backref *)node;
3205 if (!node->full_backref)
3207 if (parent == back->parent)
3210 if (node->full_backref)
3212 if (back->root == root)
3219 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
3220 u64 parent, u64 root)
3222 struct tree_backref *ref = malloc(sizeof(*ref));
3223 memset(&ref->node, 0, sizeof(ref->node));
3225 ref->parent = parent;
3226 ref->node.full_backref = 1;
3229 ref->node.full_backref = 0;
3231 list_add_tail(&ref->node.list, &rec->backrefs);
3236 static struct data_backref *find_data_backref(struct extent_record *rec,
3237 u64 parent, u64 root,
3238 u64 owner, u64 offset,
3240 u64 disk_bytenr, u64 bytes)
3242 struct list_head *cur = rec->backrefs.next;
3243 struct extent_backref *node;
3244 struct data_backref *back;
3246 while(cur != &rec->backrefs) {
3247 node = list_entry(cur, struct extent_backref, list);
3251 back = (struct data_backref *)node;
3253 if (!node->full_backref)
3255 if (parent == back->parent)
3258 if (node->full_backref)
3260 if (back->root == root && back->owner == owner &&
3261 back->offset == offset) {
3262 if (found_ref && node->found_ref &&
3263 (back->bytes != bytes ||
3264 back->disk_bytenr != disk_bytenr))
3273 static struct data_backref *alloc_data_backref(struct extent_record *rec,
3274 u64 parent, u64 root,
3275 u64 owner, u64 offset,
3278 struct data_backref *ref = malloc(sizeof(*ref));
3279 memset(&ref->node, 0, sizeof(ref->node));
3280 ref->node.is_data = 1;
3283 ref->parent = parent;
3286 ref->node.full_backref = 1;
3290 ref->offset = offset;
3291 ref->node.full_backref = 0;
3293 ref->bytes = max_size;
3296 list_add_tail(&ref->node.list, &rec->backrefs);
3297 if (max_size > rec->max_size)
3298 rec->max_size = max_size;
3302 static int add_extent_rec(struct cache_tree *extent_cache,
3303 struct btrfs_key *parent_key, u64 parent_gen,
3304 u64 start, u64 nr, u64 extent_item_refs,
3305 int is_root, int inc_ref, int set_checked,
3306 int metadata, int extent_rec, u64 max_size)
3308 struct extent_record *rec;
3309 struct cache_extent *cache;
3313 cache = lookup_cache_extent(extent_cache, start, nr);
3315 rec = container_of(cache, struct extent_record, cache);
3319 rec->nr = max(nr, max_size);
3322 * We need to make sure to reset nr to whatever the extent
3323 * record says was the real size, this way we can compare it to
3327 if (start != rec->start || rec->found_rec) {
3328 struct extent_record *tmp;
3331 if (list_empty(&rec->list))
3332 list_add_tail(&rec->list,
3333 &duplicate_extents);
3336 * We have to do this song and dance in case we
3337 * find an extent record that falls inside of
3338 * our current extent record but does not have
3339 * the same objectid.
3341 tmp = malloc(sizeof(*tmp));
3345 tmp->max_size = max_size;
3348 tmp->metadata = metadata;
3349 tmp->extent_item_refs = extent_item_refs;
3350 INIT_LIST_HEAD(&tmp->list);
3351 list_add_tail(&tmp->list, &rec->dups);
3352 rec->num_duplicates++;
3359 if (extent_item_refs && !dup) {
3360 if (rec->extent_item_refs) {
3361 fprintf(stderr, "block %llu rec "
3362 "extent_item_refs %llu, passed %llu\n",
3363 (unsigned long long)start,
3364 (unsigned long long)
3365 rec->extent_item_refs,
3366 (unsigned long long)extent_item_refs);
3368 rec->extent_item_refs = extent_item_refs;
3373 rec->content_checked = 1;
3374 rec->owner_ref_checked = 1;
3378 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
3380 rec->parent_generation = parent_gen;
3382 if (rec->max_size < max_size)
3383 rec->max_size = max_size;
3385 maybe_free_extent_rec(extent_cache, rec);
3388 rec = malloc(sizeof(*rec));
3390 rec->max_size = max_size;
3391 rec->nr = max(nr, max_size);
3392 rec->found_rec = !!extent_rec;
3393 rec->content_checked = 0;
3394 rec->owner_ref_checked = 0;
3395 rec->num_duplicates = 0;
3396 rec->metadata = metadata;
3397 INIT_LIST_HEAD(&rec->backrefs);
3398 INIT_LIST_HEAD(&rec->dups);
3399 INIT_LIST_HEAD(&rec->list);
3411 if (extent_item_refs)
3412 rec->extent_item_refs = extent_item_refs;
3414 rec->extent_item_refs = 0;
3417 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
3419 memset(&rec->parent_key, 0, sizeof(*parent_key));
3422 rec->parent_generation = parent_gen;
3424 rec->parent_generation = 0;
3426 rec->cache.start = start;
3427 rec->cache.size = nr;
3428 ret = insert_cache_extent(extent_cache, &rec->cache);
3432 rec->content_checked = 1;
3433 rec->owner_ref_checked = 1;
3438 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
3439 u64 parent, u64 root, int found_ref)
3441 struct extent_record *rec;
3442 struct tree_backref *back;
3443 struct cache_extent *cache;
3445 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3447 add_extent_rec(extent_cache, NULL, 0, bytenr,
3448 1, 0, 0, 0, 0, 1, 0, 0);
3449 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3454 rec = container_of(cache, struct extent_record, cache);
3455 if (rec->start != bytenr) {
3459 back = find_tree_backref(rec, parent, root);
3461 back = alloc_tree_backref(rec, parent, root);
3464 if (back->node.found_ref) {
3465 fprintf(stderr, "Extent back ref already exists "
3466 "for %llu parent %llu root %llu \n",
3467 (unsigned long long)bytenr,
3468 (unsigned long long)parent,
3469 (unsigned long long)root);
3471 back->node.found_ref = 1;
3473 if (back->node.found_extent_tree) {
3474 fprintf(stderr, "Extent back ref already exists "
3475 "for %llu parent %llu root %llu \n",
3476 (unsigned long long)bytenr,
3477 (unsigned long long)parent,
3478 (unsigned long long)root);
3480 back->node.found_extent_tree = 1;
3482 maybe_free_extent_rec(extent_cache, rec);
3486 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
3487 u64 parent, u64 root, u64 owner, u64 offset,
3488 u32 num_refs, int found_ref, u64 max_size)
3490 struct extent_record *rec;
3491 struct data_backref *back;
3492 struct cache_extent *cache;
3494 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3496 add_extent_rec(extent_cache, NULL, 0, bytenr, 1, 0, 0, 0, 0,
3498 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3503 rec = container_of(cache, struct extent_record, cache);
3504 if (rec->max_size < max_size)
3505 rec->max_size = max_size;
3508 * If found_ref is set then max_size is the real size and must match the
3509 * existing refs. So if we have already found a ref then we need to
3510 * make sure that this ref matches the existing one, otherwise we need
3511 * to add a new backref so we can notice that the backrefs don't match
3512 * and we need to figure out who is telling the truth. This is to
3513 * account for that awful fsync bug I introduced where we'd end up with
3514 * a btrfs_file_extent_item that would have its length include multiple
3515 * prealloc extents or point inside of a prealloc extent.
3517 back = find_data_backref(rec, parent, root, owner, offset, found_ref,
3520 back = alloc_data_backref(rec, parent, root, owner, offset,
3524 BUG_ON(num_refs != 1);
3525 if (back->node.found_ref)
3526 BUG_ON(back->bytes != max_size);
3527 back->node.found_ref = 1;
3528 back->found_ref += 1;
3529 back->bytes = max_size;
3530 back->disk_bytenr = bytenr;
3532 rec->content_checked = 1;
3533 rec->owner_ref_checked = 1;
3535 if (back->node.found_extent_tree) {
3536 fprintf(stderr, "Extent back ref already exists "
3537 "for %llu parent %llu root %llu "
3538 "owner %llu offset %llu num_refs %lu\n",
3539 (unsigned long long)bytenr,
3540 (unsigned long long)parent,
3541 (unsigned long long)root,
3542 (unsigned long long)owner,
3543 (unsigned long long)offset,
3544 (unsigned long)num_refs);
3546 back->num_refs = num_refs;
3547 back->node.found_extent_tree = 1;
3549 maybe_free_extent_rec(extent_cache, rec);
3553 static int add_pending(struct cache_tree *pending,
3554 struct cache_tree *seen, u64 bytenr, u32 size)
3557 ret = add_cache_extent(seen, bytenr, size);
3560 add_cache_extent(pending, bytenr, size);
3564 static int pick_next_pending(struct cache_tree *pending,
3565 struct cache_tree *reada,
3566 struct cache_tree *nodes,
3567 u64 last, struct block_info *bits, int bits_nr,
3570 unsigned long node_start = last;
3571 struct cache_extent *cache;
3574 cache = search_cache_extent(reada, 0);
3576 bits[0].start = cache->start;
3577 bits[0].size = cache->size;
3582 if (node_start > 32768)
3583 node_start -= 32768;
3585 cache = search_cache_extent(nodes, node_start);
3587 cache = search_cache_extent(nodes, 0);
3590 cache = search_cache_extent(pending, 0);
3595 bits[ret].start = cache->start;
3596 bits[ret].size = cache->size;
3597 cache = next_cache_extent(cache);
3599 } while (cache && ret < bits_nr);
3605 bits[ret].start = cache->start;
3606 bits[ret].size = cache->size;
3607 cache = next_cache_extent(cache);
3609 } while (cache && ret < bits_nr);
3611 if (bits_nr - ret > 8) {
3612 u64 lookup = bits[0].start + bits[0].size;
3613 struct cache_extent *next;
3614 next = search_cache_extent(pending, lookup);
3616 if (next->start - lookup > 32768)
3618 bits[ret].start = next->start;
3619 bits[ret].size = next->size;
3620 lookup = next->start + next->size;
3624 next = next_cache_extent(next);
3632 static void free_chunk_record(struct cache_extent *cache)
3634 struct chunk_record *rec;
3636 rec = container_of(cache, struct chunk_record, cache);
3637 list_del_init(&rec->list);
3638 list_del_init(&rec->dextents);
3642 void free_chunk_cache_tree(struct cache_tree *chunk_cache)
3644 cache_tree_free_extents(chunk_cache, free_chunk_record);
3647 static void free_device_record(struct rb_node *node)
3649 struct device_record *rec;
3651 rec = container_of(node, struct device_record, node);
3655 FREE_RB_BASED_TREE(device_cache, free_device_record);
3657 int insert_block_group_record(struct block_group_tree *tree,
3658 struct block_group_record *bg_rec)
3662 ret = insert_cache_extent(&tree->tree, &bg_rec->cache);
3666 list_add_tail(&bg_rec->list, &tree->block_groups);
3670 static void free_block_group_record(struct cache_extent *cache)
3672 struct block_group_record *rec;
3674 rec = container_of(cache, struct block_group_record, cache);
3675 list_del_init(&rec->list);
3679 void free_block_group_tree(struct block_group_tree *tree)
3681 cache_tree_free_extents(&tree->tree, free_block_group_record);
3684 int insert_device_extent_record(struct device_extent_tree *tree,
3685 struct device_extent_record *de_rec)
3690 * Device extent is a bit different from the other extents, because
3691 * the extents which belong to the different devices may have the
3692 * same start and size, so we need use the special extent cache
3693 * search/insert functions.
3695 ret = insert_cache_extent2(&tree->tree, &de_rec->cache);
3699 list_add_tail(&de_rec->chunk_list, &tree->no_chunk_orphans);
3700 list_add_tail(&de_rec->device_list, &tree->no_device_orphans);
3704 static void free_device_extent_record(struct cache_extent *cache)
3706 struct device_extent_record *rec;
3708 rec = container_of(cache, struct device_extent_record, cache);
3709 if (!list_empty(&rec->chunk_list))
3710 list_del_init(&rec->chunk_list);
3711 if (!list_empty(&rec->device_list))
3712 list_del_init(&rec->device_list);
3716 void free_device_extent_tree(struct device_extent_tree *tree)
3718 cache_tree_free_extents(&tree->tree, free_device_extent_record);
3721 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3722 static int process_extent_ref_v0(struct cache_tree *extent_cache,
3723 struct extent_buffer *leaf, int slot)
3725 struct btrfs_extent_ref_v0 *ref0;
3726 struct btrfs_key key;
3728 btrfs_item_key_to_cpu(leaf, &key, slot);
3729 ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
3730 if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
3731 add_tree_backref(extent_cache, key.objectid, key.offset, 0, 0);
3733 add_data_backref(extent_cache, key.objectid, key.offset, 0,
3734 0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
3740 struct chunk_record *btrfs_new_chunk_record(struct extent_buffer *leaf,
3741 struct btrfs_key *key,
3744 struct btrfs_chunk *ptr;
3745 struct chunk_record *rec;
3748 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3749 num_stripes = btrfs_chunk_num_stripes(leaf, ptr);
3751 rec = malloc(btrfs_chunk_record_size(num_stripes));
3753 fprintf(stderr, "memory allocation failed\n");
3757 memset(rec, 0, btrfs_chunk_record_size(num_stripes));
3759 INIT_LIST_HEAD(&rec->list);
3760 INIT_LIST_HEAD(&rec->dextents);
3763 rec->cache.start = key->offset;
3764 rec->cache.size = btrfs_chunk_length(leaf, ptr);
3766 rec->generation = btrfs_header_generation(leaf);
3768 rec->objectid = key->objectid;
3769 rec->type = key->type;
3770 rec->offset = key->offset;
3772 rec->length = rec->cache.size;
3773 rec->owner = btrfs_chunk_owner(leaf, ptr);
3774 rec->stripe_len = btrfs_chunk_stripe_len(leaf, ptr);
3775 rec->type_flags = btrfs_chunk_type(leaf, ptr);
3776 rec->io_width = btrfs_chunk_io_width(leaf, ptr);
3777 rec->io_align = btrfs_chunk_io_align(leaf, ptr);
3778 rec->sector_size = btrfs_chunk_sector_size(leaf, ptr);
3779 rec->num_stripes = num_stripes;
3780 rec->sub_stripes = btrfs_chunk_sub_stripes(leaf, ptr);
3782 for (i = 0; i < rec->num_stripes; ++i) {
3783 rec->stripes[i].devid =
3784 btrfs_stripe_devid_nr(leaf, ptr, i);
3785 rec->stripes[i].offset =
3786 btrfs_stripe_offset_nr(leaf, ptr, i);
3787 read_extent_buffer(leaf, rec->stripes[i].dev_uuid,
3788 (unsigned long)btrfs_stripe_dev_uuid_nr(ptr, i),
3795 static int process_chunk_item(struct cache_tree *chunk_cache,
3796 struct btrfs_key *key, struct extent_buffer *eb,
3799 struct chunk_record *rec;
3802 rec = btrfs_new_chunk_record(eb, key, slot);
3803 ret = insert_cache_extent(chunk_cache, &rec->cache);
3805 fprintf(stderr, "Chunk[%llu, %llu] existed.\n",
3806 rec->offset, rec->length);
3813 static int process_device_item(struct rb_root *dev_cache,
3814 struct btrfs_key *key, struct extent_buffer *eb, int slot)
3816 struct btrfs_dev_item *ptr;
3817 struct device_record *rec;
3820 ptr = btrfs_item_ptr(eb,
3821 slot, struct btrfs_dev_item);
3823 rec = malloc(sizeof(*rec));
3825 fprintf(stderr, "memory allocation failed\n");
3829 rec->devid = key->offset;
3830 rec->generation = btrfs_header_generation(eb);
3832 rec->objectid = key->objectid;
3833 rec->type = key->type;
3834 rec->offset = key->offset;
3836 rec->devid = btrfs_device_id(eb, ptr);
3837 rec->total_byte = btrfs_device_total_bytes(eb, ptr);
3838 rec->byte_used = btrfs_device_bytes_used(eb, ptr);
3840 ret = rb_insert(dev_cache, &rec->node, device_record_compare);
3842 fprintf(stderr, "Device[%llu] existed.\n", rec->devid);
3849 struct block_group_record *
3850 btrfs_new_block_group_record(struct extent_buffer *leaf, struct btrfs_key *key,
3853 struct btrfs_block_group_item *ptr;
3854 struct block_group_record *rec;
3856 rec = malloc(sizeof(*rec));
3858 fprintf(stderr, "memory allocation failed\n");
3861 memset(rec, 0, sizeof(*rec));
3863 rec->cache.start = key->objectid;
3864 rec->cache.size = key->offset;
3866 rec->generation = btrfs_header_generation(leaf);
3868 rec->objectid = key->objectid;
3869 rec->type = key->type;
3870 rec->offset = key->offset;
3872 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item);
3873 rec->flags = btrfs_disk_block_group_flags(leaf, ptr);
3875 INIT_LIST_HEAD(&rec->list);
3880 static int process_block_group_item(struct block_group_tree *block_group_cache,
3881 struct btrfs_key *key,
3882 struct extent_buffer *eb, int slot)
3884 struct block_group_record *rec;
3887 rec = btrfs_new_block_group_record(eb, key, slot);
3888 ret = insert_block_group_record(block_group_cache, rec);
3890 fprintf(stderr, "Block Group[%llu, %llu] existed.\n",
3891 rec->objectid, rec->offset);
3898 struct device_extent_record *
3899 btrfs_new_device_extent_record(struct extent_buffer *leaf,
3900 struct btrfs_key *key, int slot)
3902 struct device_extent_record *rec;
3903 struct btrfs_dev_extent *ptr;
3905 rec = malloc(sizeof(*rec));
3907 fprintf(stderr, "memory allocation failed\n");
3910 memset(rec, 0, sizeof(*rec));
3912 rec->cache.objectid = key->objectid;
3913 rec->cache.start = key->offset;
3915 rec->generation = btrfs_header_generation(leaf);
3917 rec->objectid = key->objectid;
3918 rec->type = key->type;
3919 rec->offset = key->offset;
3921 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
3922 rec->chunk_objecteid =
3923 btrfs_dev_extent_chunk_objectid(leaf, ptr);
3925 btrfs_dev_extent_chunk_offset(leaf, ptr);
3926 rec->length = btrfs_dev_extent_length(leaf, ptr);
3927 rec->cache.size = rec->length;
3929 INIT_LIST_HEAD(&rec->chunk_list);
3930 INIT_LIST_HEAD(&rec->device_list);
3936 process_device_extent_item(struct device_extent_tree *dev_extent_cache,
3937 struct btrfs_key *key, struct extent_buffer *eb,
3940 struct device_extent_record *rec;
3943 rec = btrfs_new_device_extent_record(eb, key, slot);
3944 ret = insert_device_extent_record(dev_extent_cache, rec);
3947 "Device extent[%llu, %llu, %llu] existed.\n",
3948 rec->objectid, rec->offset, rec->length);
3955 static int process_extent_item(struct btrfs_root *root,
3956 struct cache_tree *extent_cache,
3957 struct extent_buffer *eb, int slot)
3959 struct btrfs_extent_item *ei;
3960 struct btrfs_extent_inline_ref *iref;
3961 struct btrfs_extent_data_ref *dref;
3962 struct btrfs_shared_data_ref *sref;
3963 struct btrfs_key key;
3967 u32 item_size = btrfs_item_size_nr(eb, slot);
3973 btrfs_item_key_to_cpu(eb, &key, slot);
3975 if (key.type == BTRFS_METADATA_ITEM_KEY) {
3977 num_bytes = root->leafsize;
3979 num_bytes = key.offset;
3982 if (item_size < sizeof(*ei)) {
3983 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3984 struct btrfs_extent_item_v0 *ei0;
3985 BUG_ON(item_size != sizeof(*ei0));
3986 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
3987 refs = btrfs_extent_refs_v0(eb, ei0);
3991 return add_extent_rec(extent_cache, NULL, 0, key.objectid,
3992 num_bytes, refs, 0, 0, 0, metadata, 1,
3996 ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
3997 refs = btrfs_extent_refs(eb, ei);
3999 add_extent_rec(extent_cache, NULL, 0, key.objectid, num_bytes,
4000 refs, 0, 0, 0, metadata, 1, num_bytes);
4002 ptr = (unsigned long)(ei + 1);
4003 if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
4004 key.type == BTRFS_EXTENT_ITEM_KEY)
4005 ptr += sizeof(struct btrfs_tree_block_info);
4007 end = (unsigned long)ei + item_size;
4009 iref = (struct btrfs_extent_inline_ref *)ptr;
4010 type = btrfs_extent_inline_ref_type(eb, iref);
4011 offset = btrfs_extent_inline_ref_offset(eb, iref);
4013 case BTRFS_TREE_BLOCK_REF_KEY:
4014 add_tree_backref(extent_cache, key.objectid,
4017 case BTRFS_SHARED_BLOCK_REF_KEY:
4018 add_tree_backref(extent_cache, key.objectid,
4021 case BTRFS_EXTENT_DATA_REF_KEY:
4022 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
4023 add_data_backref(extent_cache, key.objectid, 0,
4024 btrfs_extent_data_ref_root(eb, dref),
4025 btrfs_extent_data_ref_objectid(eb,
4027 btrfs_extent_data_ref_offset(eb, dref),
4028 btrfs_extent_data_ref_count(eb, dref),
4031 case BTRFS_SHARED_DATA_REF_KEY:
4032 sref = (struct btrfs_shared_data_ref *)(iref + 1);
4033 add_data_backref(extent_cache, key.objectid, offset,
4035 btrfs_shared_data_ref_count(eb, sref),
4039 fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
4040 key.objectid, key.type, num_bytes);
4043 ptr += btrfs_extent_inline_ref_size(type);
4050 static int check_cache_range(struct btrfs_root *root,
4051 struct btrfs_block_group_cache *cache,
4052 u64 offset, u64 bytes)
4054 struct btrfs_free_space *entry;
4060 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
4061 bytenr = btrfs_sb_offset(i);
4062 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
4063 cache->key.objectid, bytenr, 0,
4064 &logical, &nr, &stripe_len);
4069 if (logical[nr] + stripe_len <= offset)
4071 if (offset + bytes <= logical[nr])
4073 if (logical[nr] == offset) {
4074 if (stripe_len >= bytes) {
4078 bytes -= stripe_len;
4079 offset += stripe_len;
4080 } else if (logical[nr] < offset) {
4081 if (logical[nr] + stripe_len >=
4086 bytes = (offset + bytes) -
4087 (logical[nr] + stripe_len);
4088 offset = logical[nr] + stripe_len;
4091 * Could be tricky, the super may land in the
4092 * middle of the area we're checking. First
4093 * check the easiest case, it's at the end.
4095 if (logical[nr] + stripe_len >=
4097 bytes = logical[nr] - offset;
4101 /* Check the left side */
4102 ret = check_cache_range(root, cache,
4104 logical[nr] - offset);
4110 /* Now we continue with the right side */
4111 bytes = (offset + bytes) -
4112 (logical[nr] + stripe_len);
4113 offset = logical[nr] + stripe_len;
4120 entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
4122 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
4123 offset, offset+bytes);
4127 if (entry->offset != offset) {
4128 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
4133 if (entry->bytes != bytes) {
4134 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
4135 bytes, entry->bytes, offset);
4139 unlink_free_space(cache->free_space_ctl, entry);
4144 static int verify_space_cache(struct btrfs_root *root,
4145 struct btrfs_block_group_cache *cache)
4147 struct btrfs_path *path;
4148 struct extent_buffer *leaf;
4149 struct btrfs_key key;
4153 path = btrfs_alloc_path();
4157 root = root->fs_info->extent_root;
4159 last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
4161 key.objectid = last;
4163 key.type = BTRFS_EXTENT_ITEM_KEY;
4165 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4170 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4171 ret = btrfs_next_leaf(root, path);
4179 leaf = path->nodes[0];
4180 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4181 if (key.objectid >= cache->key.offset + cache->key.objectid)
4183 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
4184 key.type != BTRFS_METADATA_ITEM_KEY) {
4189 if (last == key.objectid) {
4190 if (key.type == BTRFS_EXTENT_ITEM_KEY)
4191 last = key.objectid + key.offset;
4193 last = key.objectid + root->leafsize;
4198 ret = check_cache_range(root, cache, last,
4199 key.objectid - last);
4202 if (key.type == BTRFS_EXTENT_ITEM_KEY)
4203 last = key.objectid + key.offset;
4205 last = key.objectid + root->leafsize;
4209 if (last < cache->key.objectid + cache->key.offset)
4210 ret = check_cache_range(root, cache, last,
4211 cache->key.objectid +
4212 cache->key.offset - last);
4215 btrfs_free_path(path);
4218 !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
4219 fprintf(stderr, "There are still entries left in the space "
4227 static int check_space_cache(struct btrfs_root *root)
4229 struct btrfs_block_group_cache *cache;
4230 u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
4234 if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
4235 btrfs_super_generation(root->fs_info->super_copy) !=
4236 btrfs_super_cache_generation(root->fs_info->super_copy)) {
4237 printf("cache and super generation don't match, space cache "
4238 "will be invalidated\n");
4243 cache = btrfs_lookup_first_block_group(root->fs_info, start);
4247 start = cache->key.objectid + cache->key.offset;
4248 if (!cache->free_space_ctl) {
4249 if (btrfs_init_free_space_ctl(cache,
4250 root->sectorsize)) {
4255 btrfs_remove_free_space_cache(cache);
4258 ret = load_free_space_cache(root->fs_info, cache);
4262 ret = verify_space_cache(root, cache);
4264 fprintf(stderr, "cache appears valid but isnt %Lu\n",
4265 cache->key.objectid);
4270 return error ? -EINVAL : 0;
4273 static int read_extent_data(struct btrfs_root *root, char *data,
4274 u64 logical, u64 *len, int mirror)
4277 struct btrfs_multi_bio *multi = NULL;
4278 struct btrfs_fs_info *info = root->fs_info;
4279 struct btrfs_device *device;
4283 ret = btrfs_map_block(&info->mapping_tree, READ, logical, len,
4284 &multi, mirror, NULL);
4286 fprintf(stderr, "Couldn't map the block %llu\n",
4290 device = multi->stripes[0].dev;
4292 if (device->fd == 0)
4297 ret = pread64(device->fd, data, *len, multi->stripes[0].physical);
4307 static int check_extent_csums(struct btrfs_root *root, u64 bytenr,
4308 u64 num_bytes, unsigned long leaf_offset,
4309 struct extent_buffer *eb) {
4312 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
4314 unsigned long csum_offset;
4318 u64 data_checked = 0;
4324 if (num_bytes % root->sectorsize)
4327 data = malloc(num_bytes);
4331 while (offset < num_bytes) {
4334 read_len = num_bytes - offset;
4335 /* read as much space once a time */
4336 ret = read_extent_data(root, data + offset,
4337 bytenr + offset, &read_len, mirror);
4341 /* verify every 4k data's checksum */
4342 while (data_checked < read_len) {
4344 tmp = offset + data_checked;
4346 csum = btrfs_csum_data(NULL, (char *)data + tmp,
4347 csum, root->sectorsize);
4348 btrfs_csum_final(csum, (char *)&csum);
4350 csum_offset = leaf_offset +
4351 tmp / root->sectorsize * csum_size;
4352 read_extent_buffer(eb, (char *)&csum_expected,
4353 csum_offset, csum_size);
4354 /* try another mirror */
4355 if (csum != csum_expected) {
4356 fprintf(stderr, "mirror %d bytenr %llu csum %u expected csum %u\n",
4357 mirror, bytenr + tmp,
4358 csum, csum_expected);
4359 num_copies = btrfs_num_copies(
4360 &root->fs_info->mapping_tree,
4362 if (mirror < num_copies - 1) {
4367 data_checked += root->sectorsize;
4376 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
4379 struct btrfs_path *path;
4380 struct extent_buffer *leaf;
4381 struct btrfs_key key;
4384 path = btrfs_alloc_path();
4386 fprintf(stderr, "Error allocing path\n");
4390 key.objectid = bytenr;
4391 key.type = BTRFS_EXTENT_ITEM_KEY;
4392 key.offset = (u64)-1;
4395 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
4398 fprintf(stderr, "Error looking up extent record %d\n", ret);
4399 btrfs_free_path(path);
4402 if (path->slots[0] > 0) {
4405 ret = btrfs_prev_leaf(root, path);
4408 } else if (ret > 0) {
4415 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4418 * Block group items come before extent items if they have the same
4419 * bytenr, so walk back one more just in case. Dear future traveler,
4420 * first congrats on mastering time travel. Now if it's not too much
4421 * trouble could you go back to 2006 and tell Chris to make the
4422 * BLOCK_GROUP_ITEM_KEY (and BTRFS_*_REF_KEY) lower than the
4423 * EXTENT_ITEM_KEY please?
4425 while (key.type > BTRFS_EXTENT_ITEM_KEY) {
4426 if (path->slots[0] > 0) {
4429 ret = btrfs_prev_leaf(root, path);
4432 } else if (ret > 0) {
4437 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4441 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4442 ret = btrfs_next_leaf(root, path);
4444 fprintf(stderr, "Error going to next leaf "
4446 btrfs_free_path(path);
4452 leaf = path->nodes[0];
4453 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4454 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
4458 if (key.objectid + key.offset < bytenr) {
4462 if (key.objectid > bytenr + num_bytes)
4465 if (key.objectid == bytenr) {
4466 if (key.offset >= num_bytes) {
4470 num_bytes -= key.offset;
4471 bytenr += key.offset;
4472 } else if (key.objectid < bytenr) {
4473 if (key.objectid + key.offset >= bytenr + num_bytes) {
4477 num_bytes = (bytenr + num_bytes) -
4478 (key.objectid + key.offset);
4479 bytenr = key.objectid + key.offset;
4481 if (key.objectid + key.offset < bytenr + num_bytes) {
4482 u64 new_start = key.objectid + key.offset;
4483 u64 new_bytes = bytenr + num_bytes - new_start;
4486 * Weird case, the extent is in the middle of
4487 * our range, we'll have to search one side
4488 * and then the other. Not sure if this happens
4489 * in real life, but no harm in coding it up
4490 * anyway just in case.
4492 btrfs_release_path(path);
4493 ret = check_extent_exists(root, new_start,
4496 fprintf(stderr, "Right section didn't "
4500 num_bytes = key.objectid - bytenr;
4503 num_bytes = key.objectid - bytenr;
4510 if (num_bytes && !ret) {
4511 fprintf(stderr, "There are no extents for csum range "
4512 "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
4516 btrfs_free_path(path);
4520 static int check_csums(struct btrfs_root *root)
4522 struct btrfs_path *path;
4523 struct extent_buffer *leaf;
4524 struct btrfs_key key;
4525 u64 offset = 0, num_bytes = 0;
4526 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
4530 unsigned long leaf_offset;
4532 root = root->fs_info->csum_root;
4533 if (!extent_buffer_uptodate(root->node)) {
4534 fprintf(stderr, "No valid csum tree found\n");
4538 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
4539 key.type = BTRFS_EXTENT_CSUM_KEY;
4542 path = btrfs_alloc_path();
4546 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4548 fprintf(stderr, "Error searching csum tree %d\n", ret);
4549 btrfs_free_path(path);
4553 if (ret > 0 && path->slots[0])
4558 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4559 ret = btrfs_next_leaf(root, path);
4561 fprintf(stderr, "Error going to next leaf "
4568 leaf = path->nodes[0];
4570 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4571 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
4576 data_len = (btrfs_item_size_nr(leaf, path->slots[0]) /
4577 csum_size) * root->sectorsize;
4578 if (!check_data_csum)
4579 goto skip_csum_check;
4580 leaf_offset = btrfs_item_ptr_offset(leaf, path->slots[0]);
4581 ret = check_extent_csums(root, key.offset, data_len,
4587 offset = key.offset;
4588 } else if (key.offset != offset + num_bytes) {
4589 ret = check_extent_exists(root, offset, num_bytes);
4591 fprintf(stderr, "Csum exists for %Lu-%Lu but "
4592 "there is no extent record\n",
4593 offset, offset+num_bytes);
4596 offset = key.offset;
4599 num_bytes += data_len;
4603 btrfs_free_path(path);
4607 static int is_dropped_key(struct btrfs_key *key,
4608 struct btrfs_key *drop_key) {
4609 if (key->objectid < drop_key->objectid)
4611 else if (key->objectid == drop_key->objectid) {
4612 if (key->type < drop_key->type)
4614 else if (key->type == drop_key->type) {
4615 if (key->offset < drop_key->offset)
4622 static int run_next_block(struct btrfs_trans_handle *trans,
4623 struct btrfs_root *root,
4624 struct block_info *bits,
4627 struct cache_tree *pending,
4628 struct cache_tree *seen,
4629 struct cache_tree *reada,
4630 struct cache_tree *nodes,
4631 struct cache_tree *extent_cache,
4632 struct cache_tree *chunk_cache,
4633 struct rb_root *dev_cache,
4634 struct block_group_tree *block_group_cache,
4635 struct device_extent_tree *dev_extent_cache,
4636 struct btrfs_root_item *ri)
4638 struct extent_buffer *buf;
4649 struct btrfs_key key;
4650 struct cache_extent *cache;
4653 nritems = pick_next_pending(pending, reada, nodes, *last, bits,
4654 bits_nr, &reada_bits);
4659 for(i = 0; i < nritems; i++) {
4660 ret = add_cache_extent(reada, bits[i].start,
4665 /* fixme, get the parent transid */
4666 readahead_tree_block(root, bits[i].start,
4670 *last = bits[0].start;
4671 bytenr = bits[0].start;
4672 size = bits[0].size;
4674 cache = lookup_cache_extent(pending, bytenr, size);
4676 remove_cache_extent(pending, cache);
4679 cache = lookup_cache_extent(reada, bytenr, size);
4681 remove_cache_extent(reada, cache);
4684 cache = lookup_cache_extent(nodes, bytenr, size);
4686 remove_cache_extent(nodes, cache);
4689 cache = lookup_cache_extent(extent_cache, bytenr, size);
4691 struct extent_record *rec;
4693 rec = container_of(cache, struct extent_record, cache);
4694 gen = rec->parent_generation;
4697 /* fixme, get the real parent transid */
4698 buf = read_tree_block(root, bytenr, size, gen);
4699 if (!extent_buffer_uptodate(buf)) {
4700 record_bad_block_io(root->fs_info,
4701 extent_cache, bytenr, size);
4705 nritems = btrfs_header_nritems(buf);
4708 * FIXME, this only works only if we don't have any full
4711 if (!init_extent_tree) {
4712 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
4713 btrfs_header_level(buf), 1, NULL,
4721 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
4726 owner = btrfs_header_owner(buf);
4729 ret = check_block(trans, root, extent_cache, buf, flags);
4733 if (btrfs_is_leaf(buf)) {
4734 btree_space_waste += btrfs_leaf_free_space(root, buf);
4735 for (i = 0; i < nritems; i++) {
4736 struct btrfs_file_extent_item *fi;
4737 btrfs_item_key_to_cpu(buf, &key, i);
4738 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
4739 process_extent_item(root, extent_cache, buf,
4743 if (key.type == BTRFS_METADATA_ITEM_KEY) {
4744 process_extent_item(root, extent_cache, buf,
4748 if (key.type == BTRFS_EXTENT_CSUM_KEY) {
4750 btrfs_item_size_nr(buf, i);
4753 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4754 process_chunk_item(chunk_cache, &key, buf, i);
4757 if (key.type == BTRFS_DEV_ITEM_KEY) {
4758 process_device_item(dev_cache, &key, buf, i);
4761 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
4762 process_block_group_item(block_group_cache,
4766 if (key.type == BTRFS_DEV_EXTENT_KEY) {
4767 process_device_extent_item(dev_extent_cache,
4772 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
4773 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4774 process_extent_ref_v0(extent_cache, buf, i);
4781 if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
4782 add_tree_backref(extent_cache, key.objectid, 0,
4786 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
4787 add_tree_backref(extent_cache, key.objectid,
4791 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
4792 struct btrfs_extent_data_ref *ref;
4793 ref = btrfs_item_ptr(buf, i,
4794 struct btrfs_extent_data_ref);
4795 add_data_backref(extent_cache,
4797 btrfs_extent_data_ref_root(buf, ref),
4798 btrfs_extent_data_ref_objectid(buf,
4800 btrfs_extent_data_ref_offset(buf, ref),
4801 btrfs_extent_data_ref_count(buf, ref),
4802 0, root->sectorsize);
4805 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
4806 struct btrfs_shared_data_ref *ref;
4807 ref = btrfs_item_ptr(buf, i,
4808 struct btrfs_shared_data_ref);
4809 add_data_backref(extent_cache,
4810 key.objectid, key.offset, 0, 0, 0,
4811 btrfs_shared_data_ref_count(buf, ref),
4812 0, root->sectorsize);
4815 if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
4816 struct bad_item *bad;
4818 if (key.objectid == BTRFS_ORPHAN_OBJECTID)
4822 bad = malloc(sizeof(struct bad_item));
4825 INIT_LIST_HEAD(&bad->list);
4826 memcpy(&bad->key, &key,
4827 sizeof(struct btrfs_key));
4828 bad->root_id = owner;
4829 list_add_tail(&bad->list, &delete_items);
4832 if (key.type != BTRFS_EXTENT_DATA_KEY)
4834 fi = btrfs_item_ptr(buf, i,
4835 struct btrfs_file_extent_item);
4836 if (btrfs_file_extent_type(buf, fi) ==
4837 BTRFS_FILE_EXTENT_INLINE)
4839 if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
4842 data_bytes_allocated +=
4843 btrfs_file_extent_disk_num_bytes(buf, fi);
4844 if (data_bytes_allocated < root->sectorsize) {
4847 data_bytes_referenced +=
4848 btrfs_file_extent_num_bytes(buf, fi);
4849 add_data_backref(extent_cache,
4850 btrfs_file_extent_disk_bytenr(buf, fi),
4851 parent, owner, key.objectid, key.offset -
4852 btrfs_file_extent_offset(buf, fi), 1, 1,
4853 btrfs_file_extent_disk_num_bytes(buf, fi));
4857 struct btrfs_key first_key;
4859 first_key.objectid = 0;
4862 btrfs_item_key_to_cpu(buf, &first_key, 0);
4863 level = btrfs_header_level(buf);
4864 for (i = 0; i < nritems; i++) {
4865 ptr = btrfs_node_blockptr(buf, i);
4866 size = btrfs_level_size(root, level - 1);
4867 btrfs_node_key_to_cpu(buf, &key, i);
4869 struct btrfs_key drop_key;
4870 btrfs_disk_key_to_cpu(&drop_key,
4871 &ri->drop_progress);
4872 if ((level == ri->drop_level)
4873 && is_dropped_key(&key, &drop_key)) {
4877 ret = add_extent_rec(extent_cache, &key,
4878 btrfs_node_ptr_generation(buf, i),
4879 ptr, size, 0, 0, 1, 0, 1, 0,
4883 add_tree_backref(extent_cache, ptr, parent, owner, 1);
4886 add_pending(nodes, seen, ptr, size);
4888 add_pending(pending, seen, ptr, size);
4891 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
4892 nritems) * sizeof(struct btrfs_key_ptr);
4894 total_btree_bytes += buf->len;
4895 if (fs_root_objectid(btrfs_header_owner(buf)))
4896 total_fs_tree_bytes += buf->len;
4897 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
4898 total_extent_tree_bytes += buf->len;
4899 if (!found_old_backref &&
4900 btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
4901 btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
4902 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
4903 found_old_backref = 1;
4905 free_extent_buffer(buf);
4909 static int add_root_to_pending(struct extent_buffer *buf,
4910 struct cache_tree *extent_cache,
4911 struct cache_tree *pending,
4912 struct cache_tree *seen,
4913 struct cache_tree *nodes,
4914 struct btrfs_key *root_key)
4916 if (btrfs_header_level(buf) > 0)
4917 add_pending(nodes, seen, buf->start, buf->len);
4919 add_pending(pending, seen, buf->start, buf->len);
4920 add_extent_rec(extent_cache, NULL, 0, buf->start, buf->len,
4921 0, 1, 1, 0, 1, 0, buf->len);
4923 if (root_key->objectid == BTRFS_TREE_RELOC_OBJECTID ||
4924 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
4925 add_tree_backref(extent_cache, buf->start, buf->start,
4928 add_tree_backref(extent_cache, buf->start, 0,
4929 root_key->objectid, 1);
4933 /* as we fix the tree, we might be deleting blocks that
4934 * we're tracking for repair. This hook makes sure we
4935 * remove any backrefs for blocks as we are fixing them.
4937 static int free_extent_hook(struct btrfs_trans_handle *trans,
4938 struct btrfs_root *root,
4939 u64 bytenr, u64 num_bytes, u64 parent,
4940 u64 root_objectid, u64 owner, u64 offset,
4943 struct extent_record *rec;
4944 struct cache_extent *cache;
4946 struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
4948 is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
4949 cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
4953 rec = container_of(cache, struct extent_record, cache);
4955 struct data_backref *back;
4956 back = find_data_backref(rec, parent, root_objectid, owner,
4957 offset, 1, bytenr, num_bytes);
4960 if (back->node.found_ref) {
4961 back->found_ref -= refs_to_drop;
4963 rec->refs -= refs_to_drop;
4965 if (back->node.found_extent_tree) {
4966 back->num_refs -= refs_to_drop;
4967 if (rec->extent_item_refs)
4968 rec->extent_item_refs -= refs_to_drop;
4970 if (back->found_ref == 0)
4971 back->node.found_ref = 0;
4972 if (back->num_refs == 0)
4973 back->node.found_extent_tree = 0;
4975 if (!back->node.found_extent_tree && back->node.found_ref) {
4976 list_del(&back->node.list);
4980 struct tree_backref *back;
4981 back = find_tree_backref(rec, parent, root_objectid);
4984 if (back->node.found_ref) {
4987 back->node.found_ref = 0;
4989 if (back->node.found_extent_tree) {
4990 if (rec->extent_item_refs)
4991 rec->extent_item_refs--;
4992 back->node.found_extent_tree = 0;
4994 if (!back->node.found_extent_tree && back->node.found_ref) {
4995 list_del(&back->node.list);
4999 maybe_free_extent_rec(extent_cache, rec);
5004 static int delete_extent_records(struct btrfs_trans_handle *trans,
5005 struct btrfs_root *root,
5006 struct btrfs_path *path,
5007 u64 bytenr, u64 new_len)
5009 struct btrfs_key key;
5010 struct btrfs_key found_key;
5011 struct extent_buffer *leaf;
5016 key.objectid = bytenr;
5018 key.offset = (u64)-1;
5021 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
5028 if (path->slots[0] == 0)
5034 leaf = path->nodes[0];
5035 slot = path->slots[0];
5037 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5038 if (found_key.objectid != bytenr)
5041 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
5042 found_key.type != BTRFS_METADATA_ITEM_KEY &&
5043 found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
5044 found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
5045 found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
5046 found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
5047 found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
5048 btrfs_release_path(path);
5049 if (found_key.type == 0) {
5050 if (found_key.offset == 0)
5052 key.offset = found_key.offset - 1;
5053 key.type = found_key.type;
5055 key.type = found_key.type - 1;
5056 key.offset = (u64)-1;
5060 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
5061 found_key.objectid, found_key.type, found_key.offset);
5063 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
5066 btrfs_release_path(path);
5068 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5069 found_key.type == BTRFS_METADATA_ITEM_KEY) {
5070 u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
5071 found_key.offset : root->leafsize;
5073 ret = btrfs_update_block_group(trans, root, bytenr,
5080 btrfs_release_path(path);
5085 * for a single backref, this will allocate a new extent
5086 * and add the backref to it.
5088 static int record_extent(struct btrfs_trans_handle *trans,
5089 struct btrfs_fs_info *info,
5090 struct btrfs_path *path,
5091 struct extent_record *rec,
5092 struct extent_backref *back,
5093 int allocated, u64 flags)
5096 struct btrfs_root *extent_root = info->extent_root;
5097 struct extent_buffer *leaf;
5098 struct btrfs_key ins_key;
5099 struct btrfs_extent_item *ei;
5100 struct tree_backref *tback;
5101 struct data_backref *dback;
5102 struct btrfs_tree_block_info *bi;
5105 rec->max_size = max_t(u64, rec->max_size,
5106 info->extent_root->leafsize);
5109 u32 item_size = sizeof(*ei);
5112 item_size += sizeof(*bi);
5114 ins_key.objectid = rec->start;
5115 ins_key.offset = rec->max_size;
5116 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
5118 ret = btrfs_insert_empty_item(trans, extent_root, path,
5119 &ins_key, item_size);
5123 leaf = path->nodes[0];
5124 ei = btrfs_item_ptr(leaf, path->slots[0],
5125 struct btrfs_extent_item);
5127 btrfs_set_extent_refs(leaf, ei, 0);
5128 btrfs_set_extent_generation(leaf, ei, rec->generation);
5130 if (back->is_data) {
5131 btrfs_set_extent_flags(leaf, ei,
5132 BTRFS_EXTENT_FLAG_DATA);
5134 struct btrfs_disk_key copy_key;;
5136 tback = (struct tree_backref *)back;
5137 bi = (struct btrfs_tree_block_info *)(ei + 1);
5138 memset_extent_buffer(leaf, 0, (unsigned long)bi,
5141 btrfs_set_disk_key_objectid(©_key,
5142 rec->info_objectid);
5143 btrfs_set_disk_key_type(©_key, 0);
5144 btrfs_set_disk_key_offset(©_key, 0);
5146 btrfs_set_tree_block_level(leaf, bi, rec->info_level);
5147 btrfs_set_tree_block_key(leaf, bi, ©_key);
5149 btrfs_set_extent_flags(leaf, ei,
5150 BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
5153 btrfs_mark_buffer_dirty(leaf);
5154 ret = btrfs_update_block_group(trans, extent_root, rec->start,
5155 rec->max_size, 1, 0);
5158 btrfs_release_path(path);
5161 if (back->is_data) {
5165 dback = (struct data_backref *)back;
5166 if (back->full_backref)
5167 parent = dback->parent;
5171 for (i = 0; i < dback->found_ref; i++) {
5172 /* if parent != 0, we're doing a full backref
5173 * passing BTRFS_FIRST_FREE_OBJECTID as the owner
5174 * just makes the backref allocator create a data
5177 ret = btrfs_inc_extent_ref(trans, info->extent_root,
5178 rec->start, rec->max_size,
5182 BTRFS_FIRST_FREE_OBJECTID :
5188 fprintf(stderr, "adding new data backref"
5189 " on %llu %s %llu owner %llu"
5190 " offset %llu found %d\n",
5191 (unsigned long long)rec->start,
5192 back->full_backref ?
5194 back->full_backref ?
5195 (unsigned long long)parent :
5196 (unsigned long long)dback->root,
5197 (unsigned long long)dback->owner,
5198 (unsigned long long)dback->offset,
5203 tback = (struct tree_backref *)back;
5204 if (back->full_backref)
5205 parent = tback->parent;
5209 ret = btrfs_inc_extent_ref(trans, info->extent_root,
5210 rec->start, rec->max_size,
5211 parent, tback->root, 0, 0);
5212 fprintf(stderr, "adding new tree backref on "
5213 "start %llu len %llu parent %llu root %llu\n",
5214 rec->start, rec->max_size, tback->parent, tback->root);
5219 btrfs_release_path(path);
5223 struct extent_entry {
5228 struct list_head list;
5231 static struct extent_entry *find_entry(struct list_head *entries,
5232 u64 bytenr, u64 bytes)
5234 struct extent_entry *entry = NULL;
5236 list_for_each_entry(entry, entries, list) {
5237 if (entry->bytenr == bytenr && entry->bytes == bytes)
5244 static struct extent_entry *find_most_right_entry(struct list_head *entries)
5246 struct extent_entry *entry, *best = NULL, *prev = NULL;
5248 list_for_each_entry(entry, entries, list) {
5255 * If there are as many broken entries as entries then we know
5256 * not to trust this particular entry.
5258 if (entry->broken == entry->count)
5262 * If our current entry == best then we can't be sure our best
5263 * is really the best, so we need to keep searching.
5265 if (best && best->count == entry->count) {
5271 /* Prev == entry, not good enough, have to keep searching */
5272 if (!prev->broken && prev->count == entry->count)
5276 best = (prev->count > entry->count) ? prev : entry;
5277 else if (best->count < entry->count)
5285 static int repair_ref(struct btrfs_trans_handle *trans,
5286 struct btrfs_fs_info *info, struct btrfs_path *path,
5287 struct data_backref *dback, struct extent_entry *entry)
5289 struct btrfs_root *root;
5290 struct btrfs_file_extent_item *fi;
5291 struct extent_buffer *leaf;
5292 struct btrfs_key key;
5296 key.objectid = dback->root;
5297 key.type = BTRFS_ROOT_ITEM_KEY;
5298 key.offset = (u64)-1;
5299 root = btrfs_read_fs_root(info, &key);
5301 fprintf(stderr, "Couldn't find root for our ref\n");
5306 * The backref points to the original offset of the extent if it was
5307 * split, so we need to search down to the offset we have and then walk
5308 * forward until we find the backref we're looking for.
5310 key.objectid = dback->owner;
5311 key.type = BTRFS_EXTENT_DATA_KEY;
5312 key.offset = dback->offset;
5313 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5315 fprintf(stderr, "Error looking up ref %d\n", ret);
5320 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5321 ret = btrfs_next_leaf(root, path);
5323 fprintf(stderr, "Couldn't find our ref, next\n");
5327 leaf = path->nodes[0];
5328 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5329 if (key.objectid != dback->owner ||
5330 key.type != BTRFS_EXTENT_DATA_KEY) {
5331 fprintf(stderr, "Couldn't find our ref, search\n");
5334 fi = btrfs_item_ptr(leaf, path->slots[0],
5335 struct btrfs_file_extent_item);
5336 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5337 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
5339 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
5344 btrfs_release_path(path);
5347 * Have to make sure that this root gets updated when we commit the
5350 record_root_in_trans(trans, root);
5353 * Ok we have the key of the file extent we want to fix, now we can cow
5354 * down to the thing and fix it.
5356 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
5358 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
5359 key.objectid, key.type, key.offset, ret);
5363 fprintf(stderr, "Well that's odd, we just found this key "
5364 "[%Lu, %u, %Lu]\n", key.objectid, key.type,
5368 leaf = path->nodes[0];
5369 fi = btrfs_item_ptr(leaf, path->slots[0],
5370 struct btrfs_file_extent_item);
5372 if (btrfs_file_extent_compression(leaf, fi) &&
5373 dback->disk_bytenr != entry->bytenr) {
5374 fprintf(stderr, "Ref doesn't match the record start and is "
5375 "compressed, please take a btrfs-image of this file "
5376 "system and send it to a btrfs developer so they can "
5377 "complete this functionality for bytenr %Lu\n",
5378 dback->disk_bytenr);
5382 if (dback->node.broken && dback->disk_bytenr != entry->bytenr) {
5383 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
5384 } else if (dback->disk_bytenr > entry->bytenr) {
5385 u64 off_diff, offset;
5387 off_diff = dback->disk_bytenr - entry->bytenr;
5388 offset = btrfs_file_extent_offset(leaf, fi);
5389 if (dback->disk_bytenr + offset +
5390 btrfs_file_extent_num_bytes(leaf, fi) >
5391 entry->bytenr + entry->bytes) {
5392 fprintf(stderr, "Ref is past the entry end, please "
5393 "take a btrfs-image of this file system and "
5394 "send it to a btrfs developer, ref %Lu\n",
5395 dback->disk_bytenr);
5399 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
5400 btrfs_set_file_extent_offset(leaf, fi, offset);
5401 } else if (dback->disk_bytenr < entry->bytenr) {
5404 offset = btrfs_file_extent_offset(leaf, fi);
5405 if (dback->disk_bytenr + offset < entry->bytenr) {
5406 fprintf(stderr, "Ref is before the entry start, please"
5407 " take a btrfs-image of this file system and "
5408 "send it to a btrfs developer, ref %Lu\n",
5409 dback->disk_bytenr);
5413 offset += dback->disk_bytenr;
5414 offset -= entry->bytenr;
5415 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
5416 btrfs_set_file_extent_offset(leaf, fi, offset);
5419 btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
5422 * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
5423 * only do this if we aren't using compression, otherwise it's a
5426 if (!btrfs_file_extent_compression(leaf, fi))
5427 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
5429 printf("ram bytes may be wrong?\n");
5430 btrfs_mark_buffer_dirty(leaf);
5431 btrfs_release_path(path);
5435 static int verify_backrefs(struct btrfs_trans_handle *trans,
5436 struct btrfs_fs_info *info, struct btrfs_path *path,
5437 struct extent_record *rec)
5439 struct extent_backref *back;
5440 struct data_backref *dback;
5441 struct extent_entry *entry, *best = NULL;
5444 int broken_entries = 0;
5449 * Metadata is easy and the backrefs should always agree on bytenr and
5450 * size, if not we've got bigger issues.
5455 list_for_each_entry(back, &rec->backrefs, list) {
5456 if (back->full_backref || !back->is_data)
5459 dback = (struct data_backref *)back;
5462 * We only pay attention to backrefs that we found a real
5465 if (dback->found_ref == 0)
5469 * For now we only catch when the bytes don't match, not the
5470 * bytenr. We can easily do this at the same time, but I want
5471 * to have a fs image to test on before we just add repair
5472 * functionality willy-nilly so we know we won't screw up the
5476 entry = find_entry(&entries, dback->disk_bytenr,
5479 entry = malloc(sizeof(struct extent_entry));
5484 memset(entry, 0, sizeof(*entry));
5485 entry->bytenr = dback->disk_bytenr;
5486 entry->bytes = dback->bytes;
5487 list_add_tail(&entry->list, &entries);
5492 * If we only have on entry we may think the entries agree when
5493 * in reality they don't so we have to do some extra checking.
5495 if (dback->disk_bytenr != rec->start ||
5496 dback->bytes != rec->nr || back->broken)
5507 /* Yay all the backrefs agree, carry on good sir */
5508 if (nr_entries <= 1 && !mismatch)
5511 fprintf(stderr, "attempting to repair backref discrepency for bytenr "
5512 "%Lu\n", rec->start);
5515 * First we want to see if the backrefs can agree amongst themselves who
5516 * is right, so figure out which one of the entries has the highest
5519 best = find_most_right_entry(&entries);
5522 * Ok so we may have an even split between what the backrefs think, so
5523 * this is where we use the extent ref to see what it thinks.
5526 entry = find_entry(&entries, rec->start, rec->nr);
5527 if (!entry && (!broken_entries || !rec->found_rec)) {
5528 fprintf(stderr, "Backrefs don't agree with each other "
5529 "and extent record doesn't agree with anybody,"
5530 " so we can't fix bytenr %Lu bytes %Lu\n",
5531 rec->start, rec->nr);
5534 } else if (!entry) {
5536 * Ok our backrefs were broken, we'll assume this is the
5537 * correct value and add an entry for this range.
5539 entry = malloc(sizeof(struct extent_entry));
5544 memset(entry, 0, sizeof(*entry));
5545 entry->bytenr = rec->start;
5546 entry->bytes = rec->nr;
5547 list_add_tail(&entry->list, &entries);
5551 best = find_most_right_entry(&entries);
5553 fprintf(stderr, "Backrefs and extent record evenly "
5554 "split on who is right, this is going to "
5555 "require user input to fix bytenr %Lu bytes "
5556 "%Lu\n", rec->start, rec->nr);
5563 * I don't think this can happen currently as we'll abort() if we catch
5564 * this case higher up, but in case somebody removes that we still can't
5565 * deal with it properly here yet, so just bail out of that's the case.
5567 if (best->bytenr != rec->start) {
5568 fprintf(stderr, "Extent start and backref starts don't match, "
5569 "please use btrfs-image on this file system and send "
5570 "it to a btrfs developer so they can make fsck fix "
5571 "this particular case. bytenr is %Lu, bytes is %Lu\n",
5572 rec->start, rec->nr);
5578 * Ok great we all agreed on an extent record, let's go find the real
5579 * references and fix up the ones that don't match.
5581 list_for_each_entry(back, &rec->backrefs, list) {
5582 if (back->full_backref || !back->is_data)
5585 dback = (struct data_backref *)back;
5588 * Still ignoring backrefs that don't have a real ref attached
5591 if (dback->found_ref == 0)
5594 if (dback->bytes == best->bytes &&
5595 dback->disk_bytenr == best->bytenr)
5598 ret = repair_ref(trans, info, path, dback, best);
5604 * Ok we messed with the actual refs, which means we need to drop our
5605 * entire cache and go back and rescan. I know this is a huge pain and
5606 * adds a lot of extra work, but it's the only way to be safe. Once all
5607 * the backrefs agree we may not need to do anything to the extent
5612 while (!list_empty(&entries)) {
5613 entry = list_entry(entries.next, struct extent_entry, list);
5614 list_del_init(&entry->list);
5620 static int process_duplicates(struct btrfs_root *root,
5621 struct cache_tree *extent_cache,
5622 struct extent_record *rec)
5624 struct extent_record *good, *tmp;
5625 struct cache_extent *cache;
5629 * If we found a extent record for this extent then return, or if we
5630 * have more than one duplicate we are likely going to need to delete
5633 if (rec->found_rec || rec->num_duplicates > 1)
5636 /* Shouldn't happen but just in case */
5637 BUG_ON(!rec->num_duplicates);
5640 * So this happens if we end up with a backref that doesn't match the
5641 * actual extent entry. So either the backref is bad or the extent
5642 * entry is bad. Either way we want to have the extent_record actually
5643 * reflect what we found in the extent_tree, so we need to take the
5644 * duplicate out and use that as the extent_record since the only way we
5645 * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
5647 remove_cache_extent(extent_cache, &rec->cache);
5649 good = list_entry(rec->dups.next, struct extent_record, list);
5650 list_del_init(&good->list);
5651 INIT_LIST_HEAD(&good->backrefs);
5652 INIT_LIST_HEAD(&good->dups);
5653 good->cache.start = good->start;
5654 good->cache.size = good->nr;
5655 good->content_checked = 0;
5656 good->owner_ref_checked = 0;
5657 good->num_duplicates = 0;
5658 good->refs = rec->refs;
5659 list_splice_init(&rec->backrefs, &good->backrefs);
5661 cache = lookup_cache_extent(extent_cache, good->start,
5665 tmp = container_of(cache, struct extent_record, cache);
5668 * If we find another overlapping extent and it's found_rec is
5669 * set then it's a duplicate and we need to try and delete
5672 if (tmp->found_rec || tmp->num_duplicates > 0) {
5673 if (list_empty(&good->list))
5674 list_add_tail(&good->list,
5675 &duplicate_extents);
5676 good->num_duplicates += tmp->num_duplicates + 1;
5677 list_splice_init(&tmp->dups, &good->dups);
5678 list_del_init(&tmp->list);
5679 list_add_tail(&tmp->list, &good->dups);
5680 remove_cache_extent(extent_cache, &tmp->cache);
5685 * Ok we have another non extent item backed extent rec, so lets
5686 * just add it to this extent and carry on like we did above.
5688 good->refs += tmp->refs;
5689 list_splice_init(&tmp->backrefs, &good->backrefs);
5690 remove_cache_extent(extent_cache, &tmp->cache);
5693 ret = insert_cache_extent(extent_cache, &good->cache);
5696 return good->num_duplicates ? 0 : 1;
5699 static int delete_duplicate_records(struct btrfs_trans_handle *trans,
5700 struct btrfs_root *root,
5701 struct extent_record *rec)
5703 LIST_HEAD(delete_list);
5704 struct btrfs_path *path;
5705 struct extent_record *tmp, *good, *n;
5708 struct btrfs_key key;
5710 path = btrfs_alloc_path();
5717 /* Find the record that covers all of the duplicates. */
5718 list_for_each_entry(tmp, &rec->dups, list) {
5719 if (good->start < tmp->start)
5721 if (good->nr > tmp->nr)
5724 if (tmp->start + tmp->nr < good->start + good->nr) {
5725 fprintf(stderr, "Ok we have overlapping extents that "
5726 "aren't completely covered by eachother, this "
5727 "is going to require more careful thought. "
5728 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
5729 tmp->start, tmp->nr, good->start, good->nr);
5736 list_add_tail(&rec->list, &delete_list);
5738 list_for_each_entry_safe(tmp, n, &rec->dups, list) {
5741 list_move_tail(&tmp->list, &delete_list);
5744 root = root->fs_info->extent_root;
5745 list_for_each_entry(tmp, &delete_list, list) {
5746 if (tmp->found_rec == 0)
5748 key.objectid = tmp->start;
5749 key.type = BTRFS_EXTENT_ITEM_KEY;
5750 key.offset = tmp->nr;
5752 /* Shouldn't happen but just in case */
5753 if (tmp->metadata) {
5754 fprintf(stderr, "Well this shouldn't happen, extent "
5755 "record overlaps but is metadata? "
5756 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
5760 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5766 ret = btrfs_del_item(trans, root, path);
5769 btrfs_release_path(path);
5774 while (!list_empty(&delete_list)) {
5775 tmp = list_entry(delete_list.next, struct extent_record, list);
5776 list_del_init(&tmp->list);
5782 while (!list_empty(&rec->dups)) {
5783 tmp = list_entry(rec->dups.next, struct extent_record, list);
5784 list_del_init(&tmp->list);
5788 btrfs_free_path(path);
5790 if (!ret && !nr_del)
5791 rec->num_duplicates = 0;
5793 return ret ? ret : nr_del;
5796 static int find_possible_backrefs(struct btrfs_trans_handle *trans,
5797 struct btrfs_fs_info *info,
5798 struct btrfs_path *path,
5799 struct cache_tree *extent_cache,
5800 struct extent_record *rec)
5802 struct btrfs_root *root;
5803 struct extent_backref *back;
5804 struct data_backref *dback;
5805 struct cache_extent *cache;
5806 struct btrfs_file_extent_item *fi;
5807 struct btrfs_key key;
5811 list_for_each_entry(back, &rec->backrefs, list) {
5812 /* Don't care about full backrefs (poor unloved backrefs) */
5813 if (back->full_backref || !back->is_data)
5816 dback = (struct data_backref *)back;
5818 /* We found this one, we don't need to do a lookup */
5819 if (dback->found_ref)
5822 key.objectid = dback->root;
5823 key.type = BTRFS_ROOT_ITEM_KEY;
5824 key.offset = (u64)-1;
5826 root = btrfs_read_fs_root(info, &key);
5828 /* No root, definitely a bad ref, skip */
5829 if (IS_ERR(root) && PTR_ERR(root) == -ENOENT)
5831 /* Other err, exit */
5833 return PTR_ERR(root);
5835 key.objectid = dback->owner;
5836 key.type = BTRFS_EXTENT_DATA_KEY;
5837 key.offset = dback->offset;
5838 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5840 btrfs_release_path(path);
5843 /* Didn't find it, we can carry on */
5848 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5849 struct btrfs_file_extent_item);
5850 bytenr = btrfs_file_extent_disk_bytenr(path->nodes[0], fi);
5851 bytes = btrfs_file_extent_disk_num_bytes(path->nodes[0], fi);
5852 btrfs_release_path(path);
5853 cache = lookup_cache_extent(extent_cache, bytenr, 1);
5855 struct extent_record *tmp;
5856 tmp = container_of(cache, struct extent_record, cache);
5859 * If we found an extent record for the bytenr for this
5860 * particular backref then we can't add it to our
5861 * current extent record. We only want to add backrefs
5862 * that don't have a corresponding extent item in the
5863 * extent tree since they likely belong to this record
5864 * and we need to fix it if it doesn't match bytenrs.
5870 dback->found_ref += 1;
5871 dback->disk_bytenr = bytenr;
5872 dback->bytes = bytes;
5875 * Set this so the verify backref code knows not to trust the
5876 * values in this backref.
5885 * when an incorrect extent item is found, this will delete
5886 * all of the existing entries for it and recreate them
5887 * based on what the tree scan found.
5889 static int fixup_extent_refs(struct btrfs_trans_handle *trans,
5890 struct btrfs_fs_info *info,
5891 struct cache_tree *extent_cache,
5892 struct extent_record *rec)
5895 struct btrfs_path *path;
5896 struct list_head *cur = rec->backrefs.next;
5897 struct cache_extent *cache;
5898 struct extent_backref *back;
5903 * remember our flags for recreating the extent.
5904 * FIXME, if we have cleared extent tree, we can not
5905 * lookup extent info in extent tree.
5907 if (!init_extent_tree) {
5908 ret = btrfs_lookup_extent_info(NULL, info->extent_root,
5909 rec->start, rec->max_size,
5910 rec->metadata, NULL, &flags);
5917 path = btrfs_alloc_path();
5921 if (rec->refs != rec->extent_item_refs && !rec->metadata) {
5923 * Sometimes the backrefs themselves are so broken they don't
5924 * get attached to any meaningful rec, so first go back and
5925 * check any of our backrefs that we couldn't find and throw
5926 * them into the list if we find the backref so that
5927 * verify_backrefs can figure out what to do.
5929 ret = find_possible_backrefs(trans, info, path, extent_cache,
5935 /* step one, make sure all of the backrefs agree */
5936 ret = verify_backrefs(trans, info, path, rec);
5940 /* step two, delete all the existing records */
5941 ret = delete_extent_records(trans, info->extent_root, path,
5942 rec->start, rec->max_size);
5947 /* was this block corrupt? If so, don't add references to it */
5948 cache = lookup_cache_extent(info->corrupt_blocks,
5949 rec->start, rec->max_size);
5955 /* step three, recreate all the refs we did find */
5956 while(cur != &rec->backrefs) {
5957 back = list_entry(cur, struct extent_backref, list);
5961 * if we didn't find any references, don't create a
5964 if (!back->found_ref)
5967 ret = record_extent(trans, info, path, rec, back, allocated, flags);
5974 btrfs_free_path(path);
5978 /* right now we only prune from the extent allocation tree */
5979 static int prune_one_block(struct btrfs_trans_handle *trans,
5980 struct btrfs_fs_info *info,
5981 struct btrfs_corrupt_block *corrupt)
5984 struct btrfs_path path;
5985 struct extent_buffer *eb;
5989 int level = corrupt->level + 1;
5991 btrfs_init_path(&path);
5993 /* we want to stop at the parent to our busted block */
5994 path.lowest_level = level;
5996 ret = btrfs_search_slot(trans, info->extent_root,
5997 &corrupt->key, &path, -1, 1);
6002 eb = path.nodes[level];
6009 * hopefully the search gave us the block we want to prune,
6010 * lets try that first
6012 slot = path.slots[level];
6013 found = btrfs_node_blockptr(eb, slot);
6014 if (found == corrupt->cache.start)
6017 nritems = btrfs_header_nritems(eb);
6019 /* the search failed, lets scan this node and hope we find it */
6020 for (slot = 0; slot < nritems; slot++) {
6021 found = btrfs_node_blockptr(eb, slot);
6022 if (found == corrupt->cache.start)
6026 * we couldn't find the bad block. TODO, search all the nodes for pointers
6029 if (eb == info->extent_root->node) {
6034 btrfs_release_path(&path);
6039 printk("deleting pointer to block %Lu\n", corrupt->cache.start);
6040 ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
6043 btrfs_release_path(&path);
6047 static int prune_corrupt_blocks(struct btrfs_trans_handle *trans,
6048 struct btrfs_fs_info *info)
6050 struct cache_extent *cache;
6051 struct btrfs_corrupt_block *corrupt;
6053 cache = search_cache_extent(info->corrupt_blocks, 0);
6057 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
6058 prune_one_block(trans, info, corrupt);
6059 cache = next_cache_extent(cache);
6064 static void free_corrupt_block(struct cache_extent *cache)
6066 struct btrfs_corrupt_block *corrupt;
6068 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
6072 FREE_EXTENT_CACHE_BASED_TREE(corrupt_blocks, free_corrupt_block);
6074 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
6076 struct btrfs_block_group_cache *cache;
6081 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
6082 &start, &end, EXTENT_DIRTY);
6085 clear_extent_dirty(&fs_info->free_space_cache, start, end,
6091 cache = btrfs_lookup_first_block_group(fs_info, start);
6096 start = cache->key.objectid + cache->key.offset;
6100 static int check_extent_refs(struct btrfs_trans_handle *trans,
6101 struct btrfs_root *root,
6102 struct cache_tree *extent_cache)
6104 struct extent_record *rec;
6105 struct cache_extent *cache;
6113 * if we're doing a repair, we have to make sure
6114 * we don't allocate from the problem extents.
6115 * In the worst case, this will be all the
6118 cache = search_cache_extent(extent_cache, 0);
6120 rec = container_of(cache, struct extent_record, cache);
6121 btrfs_pin_extent(root->fs_info,
6122 rec->start, rec->max_size);
6123 cache = next_cache_extent(cache);
6126 /* pin down all the corrupted blocks too */
6127 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
6129 btrfs_pin_extent(root->fs_info,
6130 cache->start, cache->size);
6131 cache = next_cache_extent(cache);
6133 prune_corrupt_blocks(trans, root->fs_info);
6134 reset_cached_block_groups(root->fs_info);
6138 * We need to delete any duplicate entries we find first otherwise we
6139 * could mess up the extent tree when we have backrefs that actually
6140 * belong to a different extent item and not the weird duplicate one.
6142 while (repair && !list_empty(&duplicate_extents)) {
6143 rec = list_entry(duplicate_extents.next, struct extent_record,
6145 list_del_init(&rec->list);
6147 /* Sometimes we can find a backref before we find an actual
6148 * extent, so we need to process it a little bit to see if there
6149 * truly are multiple EXTENT_ITEM_KEY's for the same range, or
6150 * if this is a backref screwup. If we need to delete stuff
6151 * process_duplicates() will return 0, otherwise it will return
6154 if (process_duplicates(root, extent_cache, rec))
6156 ret = delete_duplicate_records(trans, root, rec);
6160 * delete_duplicate_records will return the number of entries
6161 * deleted, so if it's greater than 0 then we know we actually
6162 * did something and we need to remove.
6173 cache = search_cache_extent(extent_cache, 0);
6176 rec = container_of(cache, struct extent_record, cache);
6177 if (rec->num_duplicates) {
6178 fprintf(stderr, "extent item %llu has multiple extent "
6179 "items\n", (unsigned long long)rec->start);
6183 if (rec->refs != rec->extent_item_refs) {
6184 fprintf(stderr, "ref mismatch on [%llu %llu] ",
6185 (unsigned long long)rec->start,
6186 (unsigned long long)rec->nr);
6187 fprintf(stderr, "extent item %llu, found %llu\n",
6188 (unsigned long long)rec->extent_item_refs,
6189 (unsigned long long)rec->refs);
6190 if (!fixed && repair) {
6191 ret = fixup_extent_refs(trans, root->fs_info,
6200 if (all_backpointers_checked(rec, 1)) {
6201 fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
6202 (unsigned long long)rec->start,
6203 (unsigned long long)rec->nr);
6205 if (!fixed && repair) {
6206 ret = fixup_extent_refs(trans, root->fs_info,
6215 if (!rec->owner_ref_checked) {
6216 fprintf(stderr, "owner ref check failed [%llu %llu]\n",
6217 (unsigned long long)rec->start,
6218 (unsigned long long)rec->nr);
6219 if (!fixed && repair) {
6220 ret = fixup_extent_refs(trans, root->fs_info,
6229 remove_cache_extent(extent_cache, cache);
6230 free_all_extent_backrefs(rec);
6235 if (ret && ret != -EAGAIN) {
6236 fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
6239 btrfs_fix_block_accounting(trans, root);
6242 fprintf(stderr, "repaired damaged extent references\n");
6248 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
6252 if (type & BTRFS_BLOCK_GROUP_RAID0) {
6253 stripe_size = length;
6254 stripe_size /= num_stripes;
6255 } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
6256 stripe_size = length * 2;
6257 stripe_size /= num_stripes;
6258 } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
6259 stripe_size = length;
6260 stripe_size /= (num_stripes - 1);
6261 } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
6262 stripe_size = length;
6263 stripe_size /= (num_stripes - 2);
6265 stripe_size = length;
6271 * Check the chunk with its block group/dev list ref:
6272 * Return 0 if all refs seems valid.
6273 * Return 1 if part of refs seems valid, need later check for rebuild ref
6274 * like missing block group and needs to search extent tree to rebuild them.
6275 * Return -1 if essential refs are missing and unable to rebuild.
6277 static int check_chunk_refs(struct chunk_record *chunk_rec,
6278 struct block_group_tree *block_group_cache,
6279 struct device_extent_tree *dev_extent_cache,
6282 struct cache_extent *block_group_item;
6283 struct block_group_record *block_group_rec;
6284 struct cache_extent *dev_extent_item;
6285 struct device_extent_record *dev_extent_rec;
6292 block_group_item = lookup_cache_extent(&block_group_cache->tree,
6295 if (block_group_item) {
6296 block_group_rec = container_of(block_group_item,
6297 struct block_group_record,
6299 if (chunk_rec->length != block_group_rec->offset ||
6300 chunk_rec->offset != block_group_rec->objectid ||
6301 chunk_rec->type_flags != block_group_rec->flags) {
6304 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
6305 chunk_rec->objectid,
6310 chunk_rec->type_flags,
6311 block_group_rec->objectid,
6312 block_group_rec->type,
6313 block_group_rec->offset,
6314 block_group_rec->offset,
6315 block_group_rec->objectid,
6316 block_group_rec->flags);
6319 list_del_init(&block_group_rec->list);
6320 chunk_rec->bg_rec = block_group_rec;
6325 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
6326 chunk_rec->objectid,
6331 chunk_rec->type_flags);
6335 length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
6336 chunk_rec->num_stripes);
6337 for (i = 0; i < chunk_rec->num_stripes; ++i) {
6338 devid = chunk_rec->stripes[i].devid;
6339 offset = chunk_rec->stripes[i].offset;
6340 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
6341 devid, offset, length);
6342 if (dev_extent_item) {
6343 dev_extent_rec = container_of(dev_extent_item,
6344 struct device_extent_record,
6346 if (dev_extent_rec->objectid != devid ||
6347 dev_extent_rec->offset != offset ||
6348 dev_extent_rec->chunk_offset != chunk_rec->offset ||
6349 dev_extent_rec->length != length) {
6352 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
6353 chunk_rec->objectid,
6356 chunk_rec->stripes[i].devid,
6357 chunk_rec->stripes[i].offset,
6358 dev_extent_rec->objectid,
6359 dev_extent_rec->offset,
6360 dev_extent_rec->length);
6363 list_move(&dev_extent_rec->chunk_list,
6364 &chunk_rec->dextents);
6369 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
6370 chunk_rec->objectid,
6373 chunk_rec->stripes[i].devid,
6374 chunk_rec->stripes[i].offset);
6381 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
6382 int check_chunks(struct cache_tree *chunk_cache,
6383 struct block_group_tree *block_group_cache,
6384 struct device_extent_tree *dev_extent_cache,
6385 struct list_head *good, struct list_head *bad,
6386 struct list_head *rebuild, int silent)
6388 struct cache_extent *chunk_item;
6389 struct chunk_record *chunk_rec;
6390 struct block_group_record *bg_rec;
6391 struct device_extent_record *dext_rec;
6395 chunk_item = first_cache_extent(chunk_cache);
6396 while (chunk_item) {
6397 chunk_rec = container_of(chunk_item, struct chunk_record,
6399 err = check_chunk_refs(chunk_rec, block_group_cache,
6400 dev_extent_cache, silent);
6403 if (err == 0 && good)
6404 list_add_tail(&chunk_rec->list, good);
6405 if (err > 0 && rebuild)
6406 list_add_tail(&chunk_rec->list, rebuild);
6408 list_add_tail(&chunk_rec->list, bad);
6409 chunk_item = next_cache_extent(chunk_item);
6412 list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
6415 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
6423 list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
6427 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
6438 static int check_device_used(struct device_record *dev_rec,
6439 struct device_extent_tree *dext_cache)
6441 struct cache_extent *cache;
6442 struct device_extent_record *dev_extent_rec;
6445 cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
6447 dev_extent_rec = container_of(cache,
6448 struct device_extent_record,
6450 if (dev_extent_rec->objectid != dev_rec->devid)
6453 list_del_init(&dev_extent_rec->device_list);
6454 total_byte += dev_extent_rec->length;
6455 cache = next_cache_extent(cache);
6458 if (total_byte != dev_rec->byte_used) {
6460 "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
6461 total_byte, dev_rec->byte_used, dev_rec->objectid,
6462 dev_rec->type, dev_rec->offset);
6469 /* check btrfs_dev_item -> btrfs_dev_extent */
6470 static int check_devices(struct rb_root *dev_cache,
6471 struct device_extent_tree *dev_extent_cache)
6473 struct rb_node *dev_node;
6474 struct device_record *dev_rec;
6475 struct device_extent_record *dext_rec;
6479 dev_node = rb_first(dev_cache);
6481 dev_rec = container_of(dev_node, struct device_record, node);
6482 err = check_device_used(dev_rec, dev_extent_cache);
6486 dev_node = rb_next(dev_node);
6488 list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
6491 "Device extent[%llu, %llu, %llu] didn't find its device.\n",
6492 dext_rec->objectid, dext_rec->offset, dext_rec->length);
6499 static int check_chunks_and_extents(struct btrfs_root *root)
6501 struct rb_root dev_cache;
6502 struct cache_tree chunk_cache;
6503 struct block_group_tree block_group_cache;
6504 struct device_extent_tree dev_extent_cache;
6505 struct cache_tree extent_cache;
6506 struct cache_tree seen;
6507 struct cache_tree pending;
6508 struct cache_tree reada;
6509 struct cache_tree nodes;
6510 struct cache_tree corrupt_blocks;
6511 struct btrfs_path path;
6512 struct btrfs_key key;
6513 struct btrfs_key found_key;
6516 struct block_info *bits;
6518 struct extent_buffer *leaf;
6519 struct btrfs_trans_handle *trans = NULL;
6521 struct btrfs_root_item ri;
6522 struct list_head dropping_trees;
6524 dev_cache = RB_ROOT;
6525 cache_tree_init(&chunk_cache);
6526 block_group_tree_init(&block_group_cache);
6527 device_extent_tree_init(&dev_extent_cache);
6529 cache_tree_init(&extent_cache);
6530 cache_tree_init(&seen);
6531 cache_tree_init(&pending);
6532 cache_tree_init(&nodes);
6533 cache_tree_init(&reada);
6534 cache_tree_init(&corrupt_blocks);
6535 INIT_LIST_HEAD(&dropping_trees);
6538 trans = btrfs_start_transaction(root, 1);
6539 if (IS_ERR(trans)) {
6540 fprintf(stderr, "Error starting transaction\n");
6541 return PTR_ERR(trans);
6543 root->fs_info->fsck_extent_cache = &extent_cache;
6544 root->fs_info->free_extent_hook = free_extent_hook;
6545 root->fs_info->corrupt_blocks = &corrupt_blocks;
6549 bits = malloc(bits_nr * sizeof(struct block_info));
6556 add_root_to_pending(root->fs_info->tree_root->node,
6557 &extent_cache, &pending, &seen, &nodes,
6558 &root->fs_info->tree_root->root_key);
6560 add_root_to_pending(root->fs_info->chunk_root->node,
6561 &extent_cache, &pending, &seen, &nodes,
6562 &root->fs_info->chunk_root->root_key);
6564 btrfs_init_path(&path);
6567 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
6568 ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
6573 leaf = path.nodes[0];
6574 slot = path.slots[0];
6575 if (slot >= btrfs_header_nritems(path.nodes[0])) {
6576 ret = btrfs_next_leaf(root, &path);
6579 leaf = path.nodes[0];
6580 slot = path.slots[0];
6582 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
6583 if (btrfs_key_type(&found_key) == BTRFS_ROOT_ITEM_KEY) {
6584 unsigned long offset;
6585 struct extent_buffer *buf;
6587 offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
6588 read_extent_buffer(leaf, &ri, offset, sizeof(ri));
6589 if (btrfs_disk_key_objectid(&ri.drop_progress) == 0) {
6590 buf = read_tree_block(root->fs_info->tree_root,
6591 btrfs_root_bytenr(&ri),
6592 btrfs_level_size(root,
6593 btrfs_root_level(&ri)),
6599 add_root_to_pending(buf, &extent_cache,
6600 &pending, &seen, &nodes,
6602 free_extent_buffer(buf);
6604 struct dropping_root_item_record *dri_rec;
6605 dri_rec = malloc(sizeof(*dri_rec));
6610 memcpy(&dri_rec->ri, &ri, sizeof(ri));
6611 memcpy(&dri_rec->found_key, &found_key,
6613 list_add_tail(&dri_rec->list, &dropping_trees);
6618 btrfs_release_path(&path);
6620 ret = run_next_block(trans, root, bits, bits_nr, &last,
6621 &pending, &seen, &reada, &nodes,
6622 &extent_cache, &chunk_cache, &dev_cache,
6623 &block_group_cache, &dev_extent_cache,
6629 while (!list_empty(&dropping_trees)) {
6630 struct dropping_root_item_record *rec;
6631 struct extent_buffer *buf;
6632 rec = list_entry(dropping_trees.next,
6633 struct dropping_root_item_record, list);
6639 buf = read_tree_block(root->fs_info->tree_root,
6640 btrfs_root_bytenr(&rec->ri),
6641 btrfs_level_size(root,
6642 btrfs_root_level(&rec->ri)), 0);
6647 add_root_to_pending(buf, &extent_cache, &pending,
6648 &seen, &nodes, &rec->found_key);
6650 ret = run_next_block(trans, root, bits, bits_nr, &last,
6651 &pending, &seen, &reada,
6652 &nodes, &extent_cache,
6653 &chunk_cache, &dev_cache,
6660 free_extent_buffer(buf);
6661 list_del(&rec->list);
6666 ret = check_extent_refs(trans, root, &extent_cache);
6667 if (ret == -EAGAIN) {
6668 ret = btrfs_commit_transaction(trans, root);
6672 trans = btrfs_start_transaction(root, 1);
6673 if (IS_ERR(trans)) {
6674 ret = PTR_ERR(trans);
6678 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
6679 free_extent_cache_tree(&seen);
6680 free_extent_cache_tree(&pending);
6681 free_extent_cache_tree(&reada);
6682 free_extent_cache_tree(&nodes);
6683 free_chunk_cache_tree(&chunk_cache);
6684 free_block_group_tree(&block_group_cache);
6685 free_device_cache_tree(&dev_cache);
6686 free_device_extent_tree(&dev_extent_cache);
6687 free_extent_record_cache(root->fs_info, &extent_cache);
6691 err = check_chunks(&chunk_cache, &block_group_cache,
6692 &dev_extent_cache, NULL, NULL, NULL, 0);
6696 err = check_devices(&dev_cache, &dev_extent_cache);
6702 err = btrfs_commit_transaction(trans, root);
6707 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
6708 root->fs_info->fsck_extent_cache = NULL;
6709 root->fs_info->free_extent_hook = NULL;
6710 root->fs_info->corrupt_blocks = NULL;
6713 free_chunk_cache_tree(&chunk_cache);
6714 free_device_cache_tree(&dev_cache);
6715 free_block_group_tree(&block_group_cache);
6716 free_device_extent_tree(&dev_extent_cache);
6717 free_extent_cache_tree(&seen);
6718 free_extent_cache_tree(&pending);
6719 free_extent_cache_tree(&reada);
6720 free_extent_cache_tree(&nodes);
6724 static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
6725 struct btrfs_root *root, int overwrite)
6727 struct extent_buffer *c;
6728 struct extent_buffer *old = root->node;
6731 struct btrfs_disk_key disk_key = {0,0,0};
6737 extent_buffer_get(c);
6740 c = btrfs_alloc_free_block(trans, root,
6741 btrfs_level_size(root, 0),
6742 root->root_key.objectid,
6743 &disk_key, level, 0, 0);
6746 extent_buffer_get(c);
6750 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
6751 btrfs_set_header_level(c, level);
6752 btrfs_set_header_bytenr(c, c->start);
6753 btrfs_set_header_generation(c, trans->transid);
6754 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
6755 btrfs_set_header_owner(c, root->root_key.objectid);
6757 write_extent_buffer(c, root->fs_info->fsid,
6758 btrfs_header_fsid(), BTRFS_FSID_SIZE);
6760 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
6761 btrfs_header_chunk_tree_uuid(c),
6764 btrfs_mark_buffer_dirty(c);
6766 * this case can happen in the following case:
6768 * 1.overwrite previous root.
6770 * 2.reinit reloc data root, this is because we skip pin
6771 * down reloc data tree before which means we can allocate
6772 * same block bytenr here.
6774 if (old->start == c->start) {
6775 btrfs_set_root_generation(&root->root_item,
6777 root->root_item.level = btrfs_header_level(root->node);
6778 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6779 &root->root_key, &root->root_item);
6781 free_extent_buffer(c);
6785 free_extent_buffer(old);
6787 add_root_to_dirty_list(root);
6791 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
6792 struct extent_buffer *eb, int tree_root)
6794 struct extent_buffer *tmp;
6795 struct btrfs_root_item *ri;
6796 struct btrfs_key key;
6799 int level = btrfs_header_level(eb);
6805 * If we have pinned this block before, don't pin it again.
6806 * This can not only avoid forever loop with broken filesystem
6807 * but also give us some speedups.
6809 if (test_range_bit(&fs_info->pinned_extents, eb->start,
6810 eb->start + eb->len - 1, EXTENT_DIRTY, 0))
6813 btrfs_pin_extent(fs_info, eb->start, eb->len);
6815 leafsize = btrfs_super_leafsize(fs_info->super_copy);
6816 nritems = btrfs_header_nritems(eb);
6817 for (i = 0; i < nritems; i++) {
6819 btrfs_item_key_to_cpu(eb, &key, i);
6820 if (key.type != BTRFS_ROOT_ITEM_KEY)
6822 /* Skip the extent root and reloc roots */
6823 if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6824 key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
6825 key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
6827 ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
6828 bytenr = btrfs_disk_root_bytenr(eb, ri);
6831 * If at any point we start needing the real root we
6832 * will have to build a stump root for the root we are
6833 * in, but for now this doesn't actually use the root so
6834 * just pass in extent_root.
6836 tmp = read_tree_block(fs_info->extent_root, bytenr,
6839 fprintf(stderr, "Error reading root block\n");
6842 ret = pin_down_tree_blocks(fs_info, tmp, 0);
6843 free_extent_buffer(tmp);
6847 bytenr = btrfs_node_blockptr(eb, i);
6849 /* If we aren't the tree root don't read the block */
6850 if (level == 1 && !tree_root) {
6851 btrfs_pin_extent(fs_info, bytenr, leafsize);
6855 tmp = read_tree_block(fs_info->extent_root, bytenr,
6858 fprintf(stderr, "Error reading tree block\n");
6861 ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
6862 free_extent_buffer(tmp);
6871 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
6875 ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
6879 return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
6882 static int reset_block_groups(struct btrfs_fs_info *fs_info)
6884 struct btrfs_block_group_cache *cache;
6885 struct btrfs_path *path;
6886 struct extent_buffer *leaf;
6887 struct btrfs_chunk *chunk;
6888 struct btrfs_key key;
6892 path = btrfs_alloc_path();
6897 key.type = BTRFS_CHUNK_ITEM_KEY;
6900 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
6902 btrfs_free_path(path);
6907 * We do this in case the block groups were screwed up and had alloc
6908 * bits that aren't actually set on the chunks. This happens with
6909 * restored images every time and could happen in real life I guess.
6911 fs_info->avail_data_alloc_bits = 0;
6912 fs_info->avail_metadata_alloc_bits = 0;
6913 fs_info->avail_system_alloc_bits = 0;
6915 /* First we need to create the in-memory block groups */
6917 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
6918 ret = btrfs_next_leaf(fs_info->chunk_root, path);
6920 btrfs_free_path(path);
6928 leaf = path->nodes[0];
6929 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6930 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
6935 chunk = btrfs_item_ptr(leaf, path->slots[0],
6936 struct btrfs_chunk);
6937 btrfs_add_block_group(fs_info, 0,
6938 btrfs_chunk_type(leaf, chunk),
6939 key.objectid, key.offset,
6940 btrfs_chunk_length(leaf, chunk));
6941 set_extent_dirty(&fs_info->free_space_cache, key.offset,
6942 key.offset + btrfs_chunk_length(leaf, chunk),
6948 cache = btrfs_lookup_first_block_group(fs_info, start);
6952 start = cache->key.objectid + cache->key.offset;
6955 btrfs_free_path(path);
6959 static int reset_balance(struct btrfs_trans_handle *trans,
6960 struct btrfs_fs_info *fs_info)
6962 struct btrfs_root *root = fs_info->tree_root;
6963 struct btrfs_path *path;
6964 struct extent_buffer *leaf;
6965 struct btrfs_key key;
6966 int del_slot, del_nr = 0;
6970 path = btrfs_alloc_path();
6974 key.objectid = BTRFS_BALANCE_OBJECTID;
6975 key.type = BTRFS_BALANCE_ITEM_KEY;
6978 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6983 goto reinit_data_reloc;
6988 ret = btrfs_del_item(trans, root, path);
6991 btrfs_release_path(path);
6993 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6994 key.type = BTRFS_ROOT_ITEM_KEY;
6997 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7001 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7006 ret = btrfs_del_items(trans, root, path,
7013 btrfs_release_path(path);
7016 ret = btrfs_search_slot(trans, root, &key, path,
7023 leaf = path->nodes[0];
7024 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7025 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
7027 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7032 del_slot = path->slots[0];
7041 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
7045 btrfs_release_path(path);
7048 key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7049 key.type = BTRFS_ROOT_ITEM_KEY;
7050 key.offset = (u64)-1;
7051 root = btrfs_read_fs_root(fs_info, &key);
7053 fprintf(stderr, "Error reading data reloc tree\n");
7054 return PTR_ERR(root);
7056 record_root_in_trans(trans, root);
7057 ret = btrfs_fsck_reinit_root(trans, root, 0);
7060 ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
7062 btrfs_free_path(path);
7066 static int reinit_extent_tree(struct btrfs_trans_handle *trans,
7067 struct btrfs_fs_info *fs_info)
7073 * The only reason we don't do this is because right now we're just
7074 * walking the trees we find and pinning down their bytes, we don't look
7075 * at any of the leaves. In order to do mixed groups we'd have to check
7076 * the leaves of any fs roots and pin down the bytes for any file
7077 * extents we find. Not hard but why do it if we don't have to?
7079 if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
7080 fprintf(stderr, "We don't support re-initing the extent tree "
7081 "for mixed block groups yet, please notify a btrfs "
7082 "developer you want to do this so they can add this "
7083 "functionality.\n");
7088 * first we need to walk all of the trees except the extent tree and pin
7089 * down the bytes that are in use so we don't overwrite any existing
7092 ret = pin_metadata_blocks(fs_info);
7094 fprintf(stderr, "error pinning down used bytes\n");
7099 * Need to drop all the block groups since we're going to recreate all
7102 btrfs_free_block_groups(fs_info);
7103 ret = reset_block_groups(fs_info);
7105 fprintf(stderr, "error resetting the block groups\n");
7109 /* Ok we can allocate now, reinit the extent root */
7110 ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 0);
7112 fprintf(stderr, "extent root initialization failed\n");
7114 * When the transaction code is updated we should end the
7115 * transaction, but for now progs only knows about commit so
7116 * just return an error.
7122 * Now we have all the in-memory block groups setup so we can make
7123 * allocations properly, and the metadata we care about is safe since we
7124 * pinned all of it above.
7127 struct btrfs_block_group_cache *cache;
7129 cache = btrfs_lookup_first_block_group(fs_info, start);
7132 start = cache->key.objectid + cache->key.offset;
7133 ret = btrfs_insert_item(trans, fs_info->extent_root,
7134 &cache->key, &cache->item,
7135 sizeof(cache->item));
7137 fprintf(stderr, "Error adding block group\n");
7140 btrfs_extent_post_op(trans, fs_info->extent_root);
7143 ret = reset_balance(trans, fs_info);
7145 fprintf(stderr, "error reseting the pending balance\n");
7150 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
7152 struct btrfs_path *path;
7153 struct btrfs_trans_handle *trans;
7154 struct btrfs_key key;
7157 printf("Recowing metadata block %llu\n", eb->start);
7158 key.objectid = btrfs_header_owner(eb);
7159 key.type = BTRFS_ROOT_ITEM_KEY;
7160 key.offset = (u64)-1;
7162 root = btrfs_read_fs_root(root->fs_info, &key);
7164 fprintf(stderr, "Couldn't find owner root %llu\n",
7166 return PTR_ERR(root);
7169 path = btrfs_alloc_path();
7173 trans = btrfs_start_transaction(root, 1);
7174 if (IS_ERR(trans)) {
7175 btrfs_free_path(path);
7176 return PTR_ERR(trans);
7179 path->lowest_level = btrfs_header_level(eb);
7180 if (path->lowest_level)
7181 btrfs_node_key_to_cpu(eb, &key, 0);
7183 btrfs_item_key_to_cpu(eb, &key, 0);
7185 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7186 btrfs_commit_transaction(trans, root);
7187 btrfs_free_path(path);
7191 static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
7193 struct btrfs_path *path;
7194 struct btrfs_trans_handle *trans;
7195 struct btrfs_key key;
7198 printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
7199 bad->key.type, bad->key.offset);
7200 key.objectid = bad->root_id;
7201 key.type = BTRFS_ROOT_ITEM_KEY;
7202 key.offset = (u64)-1;
7204 root = btrfs_read_fs_root(root->fs_info, &key);
7206 fprintf(stderr, "Couldn't find owner root %llu\n",
7208 return PTR_ERR(root);
7211 path = btrfs_alloc_path();
7215 trans = btrfs_start_transaction(root, 1);
7216 if (IS_ERR(trans)) {
7217 btrfs_free_path(path);
7218 return PTR_ERR(trans);
7221 ret = btrfs_search_slot(trans, root, &bad->key, path, -1, 1);
7227 ret = btrfs_del_item(trans, root, path);
7229 btrfs_commit_transaction(trans, root);
7230 btrfs_free_path(path);
7234 static int zero_log_tree(struct btrfs_root *root)
7236 struct btrfs_trans_handle *trans;
7239 trans = btrfs_start_transaction(root, 1);
7240 if (IS_ERR(trans)) {
7241 ret = PTR_ERR(trans);
7244 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
7245 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
7246 ret = btrfs_commit_transaction(trans, root);
7250 static int populate_csum(struct btrfs_trans_handle *trans,
7251 struct btrfs_root *csum_root, char *buf, u64 start,
7258 while (offset < len) {
7259 sectorsize = csum_root->sectorsize;
7260 ret = read_extent_data(csum_root, buf, start + offset,
7264 ret = btrfs_csum_file_block(trans, csum_root, start + len,
7265 start + offset, buf, sectorsize);
7268 offset += sectorsize;
7273 static int fill_csum_tree(struct btrfs_trans_handle *trans,
7274 struct btrfs_root *csum_root)
7276 struct btrfs_root *extent_root = csum_root->fs_info->extent_root;
7277 struct btrfs_path *path;
7278 struct btrfs_extent_item *ei;
7279 struct extent_buffer *leaf;
7281 struct btrfs_key key;
7284 path = btrfs_alloc_path();
7289 key.type = BTRFS_EXTENT_ITEM_KEY;
7292 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
7294 btrfs_free_path(path);
7298 buf = malloc(csum_root->sectorsize);
7300 btrfs_free_path(path);
7305 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7306 ret = btrfs_next_leaf(extent_root, path);
7314 leaf = path->nodes[0];
7316 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7317 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
7322 ei = btrfs_item_ptr(leaf, path->slots[0],
7323 struct btrfs_extent_item);
7324 if (!(btrfs_extent_flags(leaf, ei) &
7325 BTRFS_EXTENT_FLAG_DATA)) {
7330 ret = populate_csum(trans, csum_root, buf, key.objectid,
7337 btrfs_free_path(path);
7342 struct root_item_info {
7343 /* level of the root */
7345 /* number of nodes at this level, must be 1 for a root */
7349 struct cache_extent cache_extent;
7352 static struct cache_tree *roots_info_cache = NULL;
7354 static void free_roots_info_cache(void)
7356 if (!roots_info_cache)
7359 while (!cache_tree_empty(roots_info_cache)) {
7360 struct cache_extent *entry;
7361 struct root_item_info *rii;
7363 entry = first_cache_extent(roots_info_cache);
7364 remove_cache_extent(roots_info_cache, entry);
7365 rii = container_of(entry, struct root_item_info, cache_extent);
7369 free(roots_info_cache);
7370 roots_info_cache = NULL;
7373 static int build_roots_info_cache(struct btrfs_fs_info *info)
7376 struct btrfs_key key;
7377 struct extent_buffer *leaf;
7378 struct btrfs_path *path;
7380 if (!roots_info_cache) {
7381 roots_info_cache = malloc(sizeof(*roots_info_cache));
7382 if (!roots_info_cache)
7384 cache_tree_init(roots_info_cache);
7387 path = btrfs_alloc_path();
7392 key.type = BTRFS_EXTENT_ITEM_KEY;
7395 ret = btrfs_search_slot(NULL, info->extent_root, &key, path, 0, 0);
7398 leaf = path->nodes[0];
7401 struct btrfs_key found_key;
7402 struct btrfs_extent_item *ei;
7403 struct btrfs_extent_inline_ref *iref;
7404 int slot = path->slots[0];
7409 struct cache_extent *entry;
7410 struct root_item_info *rii;
7412 if (slot >= btrfs_header_nritems(leaf)) {
7413 ret = btrfs_next_leaf(info->extent_root, path);
7420 leaf = path->nodes[0];
7421 slot = path->slots[0];
7424 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7426 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
7427 found_key.type != BTRFS_METADATA_ITEM_KEY)
7430 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
7431 flags = btrfs_extent_flags(leaf, ei);
7433 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
7434 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
7437 if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
7438 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
7439 level = found_key.offset;
7441 struct btrfs_tree_block_info *info;
7443 info = (struct btrfs_tree_block_info *)(ei + 1);
7444 iref = (struct btrfs_extent_inline_ref *)(info + 1);
7445 level = btrfs_tree_block_level(leaf, info);
7449 * For a root extent, it must be of the following type and the
7450 * first (and only one) iref in the item.
7452 type = btrfs_extent_inline_ref_type(leaf, iref);
7453 if (type != BTRFS_TREE_BLOCK_REF_KEY)
7456 root_id = btrfs_extent_inline_ref_offset(leaf, iref);
7457 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
7459 rii = malloc(sizeof(struct root_item_info));
7464 rii->cache_extent.start = root_id;
7465 rii->cache_extent.size = 1;
7466 rii->level = (u8)-1;
7467 entry = &rii->cache_extent;
7468 ret = insert_cache_extent(roots_info_cache, entry);
7471 rii = container_of(entry, struct root_item_info,
7475 ASSERT(rii->cache_extent.start == root_id);
7476 ASSERT(rii->cache_extent.size == 1);
7478 if (level > rii->level || rii->level == (u8)-1) {
7480 rii->bytenr = found_key.objectid;
7481 rii->gen = btrfs_extent_generation(leaf, ei);
7482 rii->node_count = 1;
7483 } else if (level == rii->level) {
7491 btrfs_free_path(path);
7496 static int maybe_repair_root_item(struct btrfs_fs_info *info,
7497 struct btrfs_path *path,
7498 const struct btrfs_key *root_key,
7499 const int read_only_mode)
7501 const u64 root_id = root_key->objectid;
7502 struct cache_extent *entry;
7503 struct root_item_info *rii;
7504 struct btrfs_root_item ri;
7505 unsigned long offset;
7507 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
7510 "Error: could not find extent items for root %llu\n",
7511 root_key->objectid);
7515 rii = container_of(entry, struct root_item_info, cache_extent);
7516 ASSERT(rii->cache_extent.start == root_id);
7517 ASSERT(rii->cache_extent.size == 1);
7519 if (rii->node_count != 1) {
7521 "Error: could not find btree root extent for root %llu\n",
7526 offset = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
7527 read_extent_buffer(path->nodes[0], &ri, offset, sizeof(ri));
7529 if (btrfs_root_bytenr(&ri) != rii->bytenr ||
7530 btrfs_root_level(&ri) != rii->level ||
7531 btrfs_root_generation(&ri) != rii->gen) {
7534 * If we're in repair mode but our caller told us to not update
7535 * the root item, i.e. just check if it needs to be updated, don't
7536 * print this message, since the caller will call us again shortly
7537 * for the same root item without read only mode (the caller will
7538 * open a transaction first).
7540 if (!(read_only_mode && repair))
7542 "%sroot item for root %llu,"
7543 " current bytenr %llu, current gen %llu, current level %u,"
7544 " new bytenr %llu, new gen %llu, new level %u\n",
7545 (read_only_mode ? "" : "fixing "),
7547 btrfs_root_bytenr(&ri), btrfs_root_generation(&ri),
7548 btrfs_root_level(&ri),
7549 rii->bytenr, rii->gen, rii->level);
7551 if (btrfs_root_generation(&ri) > rii->gen) {
7553 "root %llu has a root item with a more recent gen (%llu) compared to the found root node (%llu)\n",
7554 root_id, btrfs_root_generation(&ri), rii->gen);
7558 if (!read_only_mode) {
7559 btrfs_set_root_bytenr(&ri, rii->bytenr);
7560 btrfs_set_root_level(&ri, rii->level);
7561 btrfs_set_root_generation(&ri, rii->gen);
7562 write_extent_buffer(path->nodes[0], &ri,
7563 offset, sizeof(ri));
7573 * A regression introduced in the 3.17 kernel (more specifically in 3.17-rc2),
7574 * caused read-only snapshots to be corrupted if they were created at a moment
7575 * when the source subvolume/snapshot had orphan items. The issue was that the
7576 * on-disk root items became incorrect, referring to the pre orphan cleanup root
7577 * node instead of the post orphan cleanup root node.
7578 * So this function, and its callees, just detects and fixes those cases. Even
7579 * though the regression was for read-only snapshots, this function applies to
7580 * any snapshot/subvolume root.
7581 * This must be run before any other repair code - not doing it so, makes other
7582 * repair code delete or modify backrefs in the extent tree for example, which
7583 * will result in an inconsistent fs after repairing the root items.
7585 static int repair_root_items(struct btrfs_fs_info *info)
7587 struct btrfs_path *path = NULL;
7588 struct btrfs_key key;
7589 struct extent_buffer *leaf;
7590 struct btrfs_trans_handle *trans = NULL;
7595 ret = build_roots_info_cache(info);
7599 path = btrfs_alloc_path();
7605 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
7606 key.type = BTRFS_ROOT_ITEM_KEY;
7611 * Avoid opening and committing transactions if a leaf doesn't have
7612 * any root items that need to be fixed, so that we avoid rotating
7613 * backup roots unnecessarily.
7616 trans = btrfs_start_transaction(info->tree_root, 1);
7617 if (IS_ERR(trans)) {
7618 ret = PTR_ERR(trans);
7623 ret = btrfs_search_slot(trans, info->tree_root, &key, path,
7627 leaf = path->nodes[0];
7630 struct btrfs_key found_key;
7632 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7633 int no_more_keys = find_next_key(path, &key);
7635 btrfs_release_path(path);
7637 ret = btrfs_commit_transaction(trans,
7649 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7651 if (found_key.type != BTRFS_ROOT_ITEM_KEY)
7654 ret = maybe_repair_root_item(info, path, &found_key,
7659 if (!trans && repair) {
7662 btrfs_release_path(path);
7672 free_roots_info_cache();
7674 btrfs_free_path(path);
7681 static struct option long_options[] = {
7682 { "super", 1, NULL, 's' },
7683 { "repair", 0, NULL, 0 },
7684 { "init-csum-tree", 0, NULL, 0 },
7685 { "init-extent-tree", 0, NULL, 0 },
7686 { "check-data-csum", 0, NULL, 0 },
7687 { "backup", 0, NULL, 0 },
7688 { "subvol-extents", 1, NULL, 'E' },
7689 { "qgroup-report", 0, NULL, 'Q' },
7690 { "tree-root", 1, NULL, 'r' },
7694 const char * const cmd_check_usage[] = {
7695 "btrfs check [options] <device>",
7696 "Check an unmounted btrfs filesystem.",
7698 "-s|--super <superblock> use this superblock copy",
7699 "-b|--backup use the backup root copy",
7700 "--repair try to repair the filesystem",
7701 "--init-csum-tree create a new CRC tree",
7702 "--init-extent-tree create a new extent tree",
7703 "--check-data-csum verify checkums of data blocks",
7704 "--qgroup-report print a report on qgroup consistency",
7705 "--subvol-extents <subvolid> print subvolume extents and sharing state",
7706 "--tree-root <bytenr> use the given bytenr for the tree root",
7710 int cmd_check(int argc, char **argv)
7712 struct cache_tree root_cache;
7713 struct btrfs_root *root;
7714 struct btrfs_fs_info *info;
7717 u64 tree_root_bytenr = 0;
7718 char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
7721 int option_index = 0;
7722 int init_csum_tree = 0;
7723 int qgroup_report = 0;
7724 enum btrfs_open_ctree_flags ctree_flags = OPEN_CTREE_EXCLUSIVE;
7728 c = getopt_long(argc, argv, "as:br:", long_options,
7733 case 'a': /* ignored */ break;
7735 ctree_flags |= OPEN_CTREE_BACKUP_ROOT;
7738 num = arg_strtou64(optarg);
7739 if (num >= BTRFS_SUPER_MIRROR_MAX) {
7741 "ERROR: super mirror should be less than: %d\n",
7742 BTRFS_SUPER_MIRROR_MAX);
7745 bytenr = btrfs_sb_offset(((int)num));
7746 printf("using SB copy %llu, bytenr %llu\n", num,
7747 (unsigned long long)bytenr);
7753 subvolid = arg_strtou64(optarg);
7756 tree_root_bytenr = arg_strtou64(optarg);
7760 usage(cmd_check_usage);
7762 if (option_index == 1) {
7763 printf("enabling repair mode\n");
7765 ctree_flags |= OPEN_CTREE_WRITES;
7766 } else if (option_index == 2) {
7767 printf("Creating a new CRC tree\n");
7770 ctree_flags |= OPEN_CTREE_WRITES;
7771 } else if (option_index == 3) {
7772 init_extent_tree = 1;
7773 ctree_flags |= (OPEN_CTREE_WRITES |
7774 OPEN_CTREE_NO_BLOCK_GROUPS);
7776 } else if (option_index == 4) {
7777 check_data_csum = 1;
7780 argc = argc - optind;
7782 if (check_argc_exact(argc, 1))
7783 usage(cmd_check_usage);
7786 cache_tree_init(&root_cache);
7788 if((ret = check_mounted(argv[optind])) < 0) {
7789 fprintf(stderr, "Could not check mount status: %s\n", strerror(-ret));
7792 fprintf(stderr, "%s is currently mounted. Aborting.\n", argv[optind]);
7797 /* only allow partial opening under repair mode */
7799 ctree_flags |= OPEN_CTREE_PARTIAL;
7801 info = open_ctree_fs_info(argv[optind], bytenr, tree_root_bytenr,
7804 fprintf(stderr, "Couldn't open file system\n");
7809 root = info->fs_root;
7811 ret = repair_root_items(info);
7815 fprintf(stderr, "Fixed %d roots.\n", ret);
7817 } else if (ret > 0) {
7819 "Found %d roots with an outdated root item.\n",
7822 "Please run a filesystem check with the option --repair to fix them.\n");
7828 * repair mode will force us to commit transaction which
7829 * will make us fail to load log tree when mounting.
7831 if (repair && btrfs_super_log_root(info->super_copy)) {
7832 ret = ask_user("repair mode will force to clear out log tree, Are you sure?");
7837 ret = zero_log_tree(root);
7839 fprintf(stderr, "fail to zero log tree\n");
7844 uuid_unparse(info->super_copy->fsid, uuidbuf);
7845 if (qgroup_report) {
7846 printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
7848 ret = qgroup_verify_all(info);
7850 print_qgroup_report(1);
7854 printf("Print extent state for subvolume %llu on %s\nUUID: %s\n",
7855 subvolid, argv[optind], uuidbuf);
7856 ret = print_extent_state(info, subvolid);
7859 printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
7861 if (!extent_buffer_uptodate(info->tree_root->node) ||
7862 !extent_buffer_uptodate(info->dev_root->node) ||
7863 !extent_buffer_uptodate(info->chunk_root->node)) {
7864 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
7869 if (init_extent_tree || init_csum_tree) {
7870 struct btrfs_trans_handle *trans;
7872 trans = btrfs_start_transaction(info->extent_root, 0);
7873 if (IS_ERR(trans)) {
7874 fprintf(stderr, "Error starting transaction\n");
7875 ret = PTR_ERR(trans);
7879 if (init_extent_tree) {
7880 printf("Creating a new extent tree\n");
7881 ret = reinit_extent_tree(trans, info);
7886 if (init_csum_tree) {
7887 fprintf(stderr, "Reinit crc root\n");
7888 ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
7890 fprintf(stderr, "crc root initialization failed\n");
7895 ret = fill_csum_tree(trans, info->csum_root);
7897 fprintf(stderr, "crc refilling failed\n");
7902 * Ok now we commit and run the normal fsck, which will add
7903 * extent entries for all of the items it finds.
7905 ret = btrfs_commit_transaction(trans, info->extent_root);
7909 if (!extent_buffer_uptodate(info->extent_root->node)) {
7910 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
7914 if (!extent_buffer_uptodate(info->csum_root->node)) {
7915 fprintf(stderr, "Checksum root corrupted, rerun with --init-csum-tree option\n");
7920 fprintf(stderr, "checking extents\n");
7921 ret = check_chunks_and_extents(root);
7923 fprintf(stderr, "Errors found in extent allocation tree or chunk allocation\n");
7925 fprintf(stderr, "checking free space cache\n");
7926 ret = check_space_cache(root);
7931 * We used to have to have these hole extents in between our real
7932 * extents so if we don't have this flag set we need to make sure there
7933 * are no gaps in the file extents for inodes, otherwise we can just
7934 * ignore it when this happens.
7936 no_holes = btrfs_fs_incompat(root->fs_info,
7937 BTRFS_FEATURE_INCOMPAT_NO_HOLES);
7938 fprintf(stderr, "checking fs roots\n");
7939 ret = check_fs_roots(root, &root_cache);
7943 fprintf(stderr, "checking csums\n");
7944 ret = check_csums(root);
7948 fprintf(stderr, "checking root refs\n");
7949 ret = check_root_refs(root, &root_cache);
7953 while (repair && !list_empty(&root->fs_info->recow_ebs)) {
7954 struct extent_buffer *eb;
7956 eb = list_first_entry(&root->fs_info->recow_ebs,
7957 struct extent_buffer, recow);
7958 list_del_init(&eb->recow);
7959 ret = recow_extent_buffer(root, eb);
7964 while (!list_empty(&delete_items)) {
7965 struct bad_item *bad;
7967 bad = list_first_entry(&delete_items, struct bad_item, list);
7968 list_del_init(&bad->list);
7970 ret = delete_bad_item(root, bad);
7974 if (info->quota_enabled) {
7976 fprintf(stderr, "checking quota groups\n");
7977 err = qgroup_verify_all(info);
7982 if (!list_empty(&root->fs_info->recow_ebs)) {
7983 fprintf(stderr, "Transid errors in file system\n");
7987 print_qgroup_report(0);
7988 if (found_old_backref) { /*
7989 * there was a disk format change when mixed
7990 * backref was in testing tree. The old format
7991 * existed about one week.
7993 printf("\n * Found old mixed backref format. "
7994 "The old format is not supported! *"
7995 "\n * Please mount the FS in readonly mode, "
7996 "backup data and re-format the FS. *\n\n");
7999 printf("found %llu bytes used err is %d\n",
8000 (unsigned long long)bytes_used, ret);
8001 printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
8002 printf("total tree bytes: %llu\n",
8003 (unsigned long long)total_btree_bytes);
8004 printf("total fs tree bytes: %llu\n",
8005 (unsigned long long)total_fs_tree_bytes);
8006 printf("total extent tree bytes: %llu\n",
8007 (unsigned long long)total_extent_tree_bytes);
8008 printf("btree space waste bytes: %llu\n",
8009 (unsigned long long)btree_space_waste);
8010 printf("file data blocks allocated: %llu\n referenced %llu\n",
8011 (unsigned long long)data_bytes_allocated,
8012 (unsigned long long)data_bytes_referenced);
8013 printf("%s\n", BTRFS_BUILD_VERSION);
8015 free_root_recs_tree(&root_cache);