2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
23 #include <sys/types.h>
27 #include <uuid/uuid.h>
32 #include "print-tree.h"
33 #include "transaction.h"
36 #include "free-space-cache.h"
38 #include "qgroup-verify.h"
39 #include "rbtree-utils.h"
43 static u64 bytes_used = 0;
44 static u64 total_csum_bytes = 0;
45 static u64 total_btree_bytes = 0;
46 static u64 total_fs_tree_bytes = 0;
47 static u64 total_extent_tree_bytes = 0;
48 static u64 btree_space_waste = 0;
49 static u64 data_bytes_allocated = 0;
50 static u64 data_bytes_referenced = 0;
51 static int found_old_backref = 0;
52 static LIST_HEAD(duplicate_extents);
53 static LIST_HEAD(delete_items);
54 static int repair = 0;
55 static int no_holes = 0;
56 static int init_extent_tree = 0;
57 static int check_data_csum = 0;
59 struct extent_backref {
60 struct list_head list;
61 unsigned int is_data:1;
62 unsigned int found_extent_tree:1;
63 unsigned int full_backref:1;
64 unsigned int found_ref:1;
65 unsigned int broken:1;
69 struct extent_backref node;
84 * Much like data_backref, just removed the undetermined members
85 * and change it to use list_head.
86 * During extent scan, it is stored in root->orphan_data_extent.
87 * During fs tree scan, it is then moved to inode_rec->orphan_data_extents.
89 struct orphan_data_extent {
90 struct list_head list;
99 struct extent_backref node;
106 struct extent_record {
107 struct list_head backrefs;
108 struct list_head dups;
109 struct list_head list;
110 struct cache_extent cache;
111 struct btrfs_disk_key parent_key;
116 u64 extent_item_refs;
118 u64 parent_generation;
122 unsigned int found_rec:1;
123 unsigned int content_checked:1;
124 unsigned int owner_ref_checked:1;
125 unsigned int is_root:1;
126 unsigned int metadata:1;
127 unsigned int flag_block_full_backref:1;
130 struct inode_backref {
131 struct list_head list;
132 unsigned int found_dir_item:1;
133 unsigned int found_dir_index:1;
134 unsigned int found_inode_ref:1;
135 unsigned int filetype:8;
137 unsigned int ref_type;
144 struct root_item_record {
145 struct list_head list;
151 struct btrfs_key drop_key;
154 #define REF_ERR_NO_DIR_ITEM (1 << 0)
155 #define REF_ERR_NO_DIR_INDEX (1 << 1)
156 #define REF_ERR_NO_INODE_REF (1 << 2)
157 #define REF_ERR_DUP_DIR_ITEM (1 << 3)
158 #define REF_ERR_DUP_DIR_INDEX (1 << 4)
159 #define REF_ERR_DUP_INODE_REF (1 << 5)
160 #define REF_ERR_INDEX_UNMATCH (1 << 6)
161 #define REF_ERR_FILETYPE_UNMATCH (1 << 7)
162 #define REF_ERR_NAME_TOO_LONG (1 << 8) // 100
163 #define REF_ERR_NO_ROOT_REF (1 << 9)
164 #define REF_ERR_NO_ROOT_BACKREF (1 << 10)
165 #define REF_ERR_DUP_ROOT_REF (1 << 11)
166 #define REF_ERR_DUP_ROOT_BACKREF (1 << 12)
168 struct file_extent_hole {
174 /* Compatible function to allow reuse of old codes */
175 static u64 first_extent_gap(struct rb_root *holes)
177 struct file_extent_hole *hole;
179 if (RB_EMPTY_ROOT(holes))
182 hole = rb_entry(rb_first(holes), struct file_extent_hole, node);
186 int compare_hole(struct rb_node *node1, struct rb_node *node2)
188 struct file_extent_hole *hole1;
189 struct file_extent_hole *hole2;
191 hole1 = rb_entry(node1, struct file_extent_hole, node);
192 hole2 = rb_entry(node2, struct file_extent_hole, node);
194 if (hole1->start > hole2->start)
196 if (hole1->start < hole2->start)
198 /* Now hole1->start == hole2->start */
199 if (hole1->len >= hole2->len)
201 * Hole 1 will be merge center
202 * Same hole will be merged later
205 /* Hole 2 will be merge center */
210 * Add a hole to the record
212 * This will do hole merge for copy_file_extent_holes(),
213 * which will ensure there won't be continuous holes.
215 static int add_file_extent_hole(struct rb_root *holes,
218 struct file_extent_hole *hole;
219 struct file_extent_hole *prev = NULL;
220 struct file_extent_hole *next = NULL;
222 hole = malloc(sizeof(*hole));
227 /* Since compare will not return 0, no -EEXIST will happen */
228 rb_insert(holes, &hole->node, compare_hole);
230 /* simple merge with previous hole */
231 if (rb_prev(&hole->node))
232 prev = rb_entry(rb_prev(&hole->node), struct file_extent_hole,
234 if (prev && prev->start + prev->len >= hole->start) {
235 hole->len = hole->start + hole->len - prev->start;
236 hole->start = prev->start;
237 rb_erase(&prev->node, holes);
242 /* iterate merge with next holes */
244 if (!rb_next(&hole->node))
246 next = rb_entry(rb_next(&hole->node), struct file_extent_hole,
248 if (hole->start + hole->len >= next->start) {
249 if (hole->start + hole->len <= next->start + next->len)
250 hole->len = next->start + next->len -
252 rb_erase(&next->node, holes);
261 static int compare_hole_range(struct rb_node *node, void *data)
263 struct file_extent_hole *hole;
266 hole = (struct file_extent_hole *)data;
269 hole = rb_entry(node, struct file_extent_hole, node);
270 if (start < hole->start)
272 if (start >= hole->start && start < hole->start + hole->len)
278 * Delete a hole in the record
280 * This will do the hole split and is much restrict than add.
282 static int del_file_extent_hole(struct rb_root *holes,
285 struct file_extent_hole *hole;
286 struct file_extent_hole tmp;
287 struct file_extent_hole prev;
288 struct file_extent_hole next;
289 struct rb_node *node;
296 node = rb_search(holes, &tmp, compare_hole_range, NULL);
299 hole = rb_entry(node, struct file_extent_hole, node);
300 if (start + len > hole->start + hole->len)
304 * Now there will be no overflap, delete the hole and re-add the
305 * split(s) if they exists.
307 if (start > hole->start) {
308 prev.start = hole->start;
309 prev.len = start - hole->start;
312 if (hole->start + hole->len > start + len) {
313 next.start = start + len;
314 next.len = hole->start + hole->len - start - len;
317 rb_erase(node, holes);
320 ret = add_file_extent_hole(holes, prev.start, prev.len);
325 ret = add_file_extent_hole(holes, next.start, next.len);
332 static int copy_file_extent_holes(struct rb_root *dst,
335 struct file_extent_hole *hole;
336 struct rb_node *node;
339 node = rb_first(src);
341 hole = rb_entry(node, struct file_extent_hole, node);
342 ret = add_file_extent_hole(dst, hole->start, hole->len);
345 node = rb_next(node);
350 static void free_file_extent_holes(struct rb_root *holes)
352 struct rb_node *node;
353 struct file_extent_hole *hole;
355 node = rb_first(holes);
357 hole = rb_entry(node, struct file_extent_hole, node);
358 rb_erase(node, holes);
360 node = rb_first(holes);
364 struct inode_record {
365 struct list_head backrefs;
366 unsigned int checked:1;
367 unsigned int merging:1;
368 unsigned int found_inode_item:1;
369 unsigned int found_dir_item:1;
370 unsigned int found_file_extent:1;
371 unsigned int found_csum_item:1;
372 unsigned int some_csum_missing:1;
373 unsigned int nodatasum:1;
386 struct rb_root holes;
387 struct list_head orphan_extents;
392 #define I_ERR_NO_INODE_ITEM (1 << 0)
393 #define I_ERR_NO_ORPHAN_ITEM (1 << 1)
394 #define I_ERR_DUP_INODE_ITEM (1 << 2)
395 #define I_ERR_DUP_DIR_INDEX (1 << 3)
396 #define I_ERR_ODD_DIR_ITEM (1 << 4)
397 #define I_ERR_ODD_FILE_EXTENT (1 << 5)
398 #define I_ERR_BAD_FILE_EXTENT (1 << 6)
399 #define I_ERR_FILE_EXTENT_OVERLAP (1 << 7)
400 #define I_ERR_FILE_EXTENT_DISCOUNT (1 << 8) // 100
401 #define I_ERR_DIR_ISIZE_WRONG (1 << 9)
402 #define I_ERR_FILE_NBYTES_WRONG (1 << 10) // 400
403 #define I_ERR_ODD_CSUM_ITEM (1 << 11)
404 #define I_ERR_SOME_CSUM_MISSING (1 << 12)
405 #define I_ERR_LINK_COUNT_WRONG (1 << 13)
406 #define I_ERR_FILE_EXTENT_ORPHAN (1 << 14)
408 struct root_backref {
409 struct list_head list;
410 unsigned int found_dir_item:1;
411 unsigned int found_dir_index:1;
412 unsigned int found_back_ref:1;
413 unsigned int found_forward_ref:1;
414 unsigned int reachable:1;
424 struct list_head backrefs;
425 struct cache_extent cache;
426 unsigned int found_root_item:1;
432 struct cache_extent cache;
437 struct cache_extent cache;
438 struct cache_tree root_cache;
439 struct cache_tree inode_cache;
440 struct inode_record *current;
449 struct walk_control {
450 struct cache_tree shared;
451 struct shared_node *nodes[BTRFS_MAX_LEVEL];
457 struct btrfs_key key;
459 struct list_head list;
462 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info);
464 static void record_root_in_trans(struct btrfs_trans_handle *trans,
465 struct btrfs_root *root)
467 if (root->last_trans != trans->transid) {
468 root->track_dirty = 1;
469 root->last_trans = trans->transid;
470 root->commit_root = root->node;
471 extent_buffer_get(root->node);
475 static u8 imode_to_type(u32 imode)
478 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
479 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
480 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
481 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
482 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
483 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
484 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
485 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
488 return btrfs_type_by_mode[(imode & S_IFMT) >> S_SHIFT];
492 static int device_record_compare(struct rb_node *node1, struct rb_node *node2)
494 struct device_record *rec1;
495 struct device_record *rec2;
497 rec1 = rb_entry(node1, struct device_record, node);
498 rec2 = rb_entry(node2, struct device_record, node);
499 if (rec1->devid > rec2->devid)
501 else if (rec1->devid < rec2->devid)
507 static struct inode_record *clone_inode_rec(struct inode_record *orig_rec)
509 struct inode_record *rec;
510 struct inode_backref *backref;
511 struct inode_backref *orig;
512 struct orphan_data_extent *src_orphan;
513 struct orphan_data_extent *dst_orphan;
516 rec = malloc(sizeof(*rec));
517 memcpy(rec, orig_rec, sizeof(*rec));
519 INIT_LIST_HEAD(&rec->backrefs);
520 INIT_LIST_HEAD(&rec->orphan_extents);
522 list_for_each_entry(orig, &orig_rec->backrefs, list) {
523 size = sizeof(*orig) + orig->namelen + 1;
524 backref = malloc(size);
525 memcpy(backref, orig, size);
526 list_add_tail(&backref->list, &rec->backrefs);
528 list_for_each_entry(src_orphan, &orig_rec->orphan_extents, list) {
529 dst_orphan = malloc(sizeof(*dst_orphan));
530 /* TODO: Fix all the HELL of un-catched -ENOMEM case */
532 memcpy(dst_orphan, src_orphan, sizeof(*src_orphan));
533 list_add_tail(&dst_orphan->list, &rec->orphan_extents);
538 static void print_orphan_data_extents(struct list_head *orphan_extents,
541 struct orphan_data_extent *orphan;
543 if (list_empty(orphan_extents))
545 printf("The following data extent is lost in tree %llu:\n",
547 list_for_each_entry(orphan, orphan_extents, list) {
548 printf("\tinode: %llu, offset:%llu, disk_bytenr: %llu, disk_len: %llu\n",
549 orphan->objectid, orphan->offset, orphan->disk_bytenr,
554 static void print_inode_error(struct btrfs_root *root, struct inode_record *rec)
556 u64 root_objectid = root->root_key.objectid;
557 int errors = rec->errors;
561 /* reloc root errors, we print its corresponding fs root objectid*/
562 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
563 root_objectid = root->root_key.offset;
564 fprintf(stderr, "reloc");
566 fprintf(stderr, "root %llu inode %llu errors %x",
567 (unsigned long long) root_objectid,
568 (unsigned long long) rec->ino, rec->errors);
570 if (errors & I_ERR_NO_INODE_ITEM)
571 fprintf(stderr, ", no inode item");
572 if (errors & I_ERR_NO_ORPHAN_ITEM)
573 fprintf(stderr, ", no orphan item");
574 if (errors & I_ERR_DUP_INODE_ITEM)
575 fprintf(stderr, ", dup inode item");
576 if (errors & I_ERR_DUP_DIR_INDEX)
577 fprintf(stderr, ", dup dir index");
578 if (errors & I_ERR_ODD_DIR_ITEM)
579 fprintf(stderr, ", odd dir item");
580 if (errors & I_ERR_ODD_FILE_EXTENT)
581 fprintf(stderr, ", odd file extent");
582 if (errors & I_ERR_BAD_FILE_EXTENT)
583 fprintf(stderr, ", bad file extent");
584 if (errors & I_ERR_FILE_EXTENT_OVERLAP)
585 fprintf(stderr, ", file extent overlap");
586 if (errors & I_ERR_FILE_EXTENT_DISCOUNT)
587 fprintf(stderr, ", file extent discount");
588 if (errors & I_ERR_DIR_ISIZE_WRONG)
589 fprintf(stderr, ", dir isize wrong");
590 if (errors & I_ERR_FILE_NBYTES_WRONG)
591 fprintf(stderr, ", nbytes wrong");
592 if (errors & I_ERR_ODD_CSUM_ITEM)
593 fprintf(stderr, ", odd csum item");
594 if (errors & I_ERR_SOME_CSUM_MISSING)
595 fprintf(stderr, ", some csum missing");
596 if (errors & I_ERR_LINK_COUNT_WRONG)
597 fprintf(stderr, ", link count wrong");
598 if (errors & I_ERR_FILE_EXTENT_ORPHAN)
599 fprintf(stderr, ", orphan file extent");
600 fprintf(stderr, "\n");
601 /* Print the orphan extents if needed */
602 if (errors & I_ERR_FILE_EXTENT_ORPHAN)
603 print_orphan_data_extents(&rec->orphan_extents, root->objectid);
605 /* Print the holes if needed */
606 if (errors & I_ERR_FILE_EXTENT_DISCOUNT) {
607 struct file_extent_hole *hole;
608 struct rb_node *node;
610 node = rb_first(&rec->holes);
611 fprintf(stderr, "Found file extent holes:\n");
613 hole = rb_entry(node, struct file_extent_hole, node);
614 fprintf(stderr, "\tstart: %llu, len:%llu\n",
615 hole->start, hole->len);
616 node = rb_next(node);
621 static void print_ref_error(int errors)
623 if (errors & REF_ERR_NO_DIR_ITEM)
624 fprintf(stderr, ", no dir item");
625 if (errors & REF_ERR_NO_DIR_INDEX)
626 fprintf(stderr, ", no dir index");
627 if (errors & REF_ERR_NO_INODE_REF)
628 fprintf(stderr, ", no inode ref");
629 if (errors & REF_ERR_DUP_DIR_ITEM)
630 fprintf(stderr, ", dup dir item");
631 if (errors & REF_ERR_DUP_DIR_INDEX)
632 fprintf(stderr, ", dup dir index");
633 if (errors & REF_ERR_DUP_INODE_REF)
634 fprintf(stderr, ", dup inode ref");
635 if (errors & REF_ERR_INDEX_UNMATCH)
636 fprintf(stderr, ", index unmatch");
637 if (errors & REF_ERR_FILETYPE_UNMATCH)
638 fprintf(stderr, ", filetype unmatch");
639 if (errors & REF_ERR_NAME_TOO_LONG)
640 fprintf(stderr, ", name too long");
641 if (errors & REF_ERR_NO_ROOT_REF)
642 fprintf(stderr, ", no root ref");
643 if (errors & REF_ERR_NO_ROOT_BACKREF)
644 fprintf(stderr, ", no root backref");
645 if (errors & REF_ERR_DUP_ROOT_REF)
646 fprintf(stderr, ", dup root ref");
647 if (errors & REF_ERR_DUP_ROOT_BACKREF)
648 fprintf(stderr, ", dup root backref");
649 fprintf(stderr, "\n");
652 static struct inode_record *get_inode_rec(struct cache_tree *inode_cache,
655 struct ptr_node *node;
656 struct cache_extent *cache;
657 struct inode_record *rec = NULL;
660 cache = lookup_cache_extent(inode_cache, ino, 1);
662 node = container_of(cache, struct ptr_node, cache);
664 if (mod && rec->refs > 1) {
665 node->data = clone_inode_rec(rec);
670 rec = calloc(1, sizeof(*rec));
672 rec->extent_start = (u64)-1;
674 INIT_LIST_HEAD(&rec->backrefs);
675 INIT_LIST_HEAD(&rec->orphan_extents);
676 rec->holes = RB_ROOT;
678 node = malloc(sizeof(*node));
679 node->cache.start = ino;
680 node->cache.size = 1;
683 if (ino == BTRFS_FREE_INO_OBJECTID)
686 ret = insert_cache_extent(inode_cache, &node->cache);
692 static void free_orphan_data_extents(struct list_head *orphan_extents)
694 struct orphan_data_extent *orphan;
696 while (!list_empty(orphan_extents)) {
697 orphan = list_entry(orphan_extents->next,
698 struct orphan_data_extent, list);
699 list_del(&orphan->list);
704 static void free_inode_rec(struct inode_record *rec)
706 struct inode_backref *backref;
711 while (!list_empty(&rec->backrefs)) {
712 backref = list_entry(rec->backrefs.next,
713 struct inode_backref, list);
714 list_del(&backref->list);
717 free_orphan_data_extents(&rec->orphan_extents);
718 free_file_extent_holes(&rec->holes);
722 static int can_free_inode_rec(struct inode_record *rec)
724 if (!rec->errors && rec->checked && rec->found_inode_item &&
725 rec->nlink == rec->found_link && list_empty(&rec->backrefs))
730 static void maybe_free_inode_rec(struct cache_tree *inode_cache,
731 struct inode_record *rec)
733 struct cache_extent *cache;
734 struct inode_backref *tmp, *backref;
735 struct ptr_node *node;
736 unsigned char filetype;
738 if (!rec->found_inode_item)
741 filetype = imode_to_type(rec->imode);
742 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
743 if (backref->found_dir_item && backref->found_dir_index) {
744 if (backref->filetype != filetype)
745 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
746 if (!backref->errors && backref->found_inode_ref) {
747 list_del(&backref->list);
753 if (!rec->checked || rec->merging)
756 if (S_ISDIR(rec->imode)) {
757 if (rec->found_size != rec->isize)
758 rec->errors |= I_ERR_DIR_ISIZE_WRONG;
759 if (rec->found_file_extent)
760 rec->errors |= I_ERR_ODD_FILE_EXTENT;
761 } else if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
762 if (rec->found_dir_item)
763 rec->errors |= I_ERR_ODD_DIR_ITEM;
764 if (rec->found_size != rec->nbytes)
765 rec->errors |= I_ERR_FILE_NBYTES_WRONG;
766 if (rec->nlink > 0 && !no_holes &&
767 (rec->extent_end < rec->isize ||
768 first_extent_gap(&rec->holes) < rec->isize))
769 rec->errors |= I_ERR_FILE_EXTENT_DISCOUNT;
772 if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
773 if (rec->found_csum_item && rec->nodatasum)
774 rec->errors |= I_ERR_ODD_CSUM_ITEM;
775 if (rec->some_csum_missing && !rec->nodatasum)
776 rec->errors |= I_ERR_SOME_CSUM_MISSING;
779 BUG_ON(rec->refs != 1);
780 if (can_free_inode_rec(rec)) {
781 cache = lookup_cache_extent(inode_cache, rec->ino, 1);
782 node = container_of(cache, struct ptr_node, cache);
783 BUG_ON(node->data != rec);
784 remove_cache_extent(inode_cache, &node->cache);
790 static int check_orphan_item(struct btrfs_root *root, u64 ino)
792 struct btrfs_path path;
793 struct btrfs_key key;
796 key.objectid = BTRFS_ORPHAN_OBJECTID;
797 key.type = BTRFS_ORPHAN_ITEM_KEY;
800 btrfs_init_path(&path);
801 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
802 btrfs_release_path(&path);
808 static int process_inode_item(struct extent_buffer *eb,
809 int slot, struct btrfs_key *key,
810 struct shared_node *active_node)
812 struct inode_record *rec;
813 struct btrfs_inode_item *item;
815 rec = active_node->current;
816 BUG_ON(rec->ino != key->objectid || rec->refs > 1);
817 if (rec->found_inode_item) {
818 rec->errors |= I_ERR_DUP_INODE_ITEM;
821 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
822 rec->nlink = btrfs_inode_nlink(eb, item);
823 rec->isize = btrfs_inode_size(eb, item);
824 rec->nbytes = btrfs_inode_nbytes(eb, item);
825 rec->imode = btrfs_inode_mode(eb, item);
826 if (btrfs_inode_flags(eb, item) & BTRFS_INODE_NODATASUM)
828 rec->found_inode_item = 1;
830 rec->errors |= I_ERR_NO_ORPHAN_ITEM;
831 maybe_free_inode_rec(&active_node->inode_cache, rec);
835 static struct inode_backref *get_inode_backref(struct inode_record *rec,
837 int namelen, u64 dir)
839 struct inode_backref *backref;
841 list_for_each_entry(backref, &rec->backrefs, list) {
842 if (rec->ino == BTRFS_MULTIPLE_OBJECTIDS)
844 if (backref->dir != dir || backref->namelen != namelen)
846 if (memcmp(name, backref->name, namelen))
851 backref = malloc(sizeof(*backref) + namelen + 1);
852 memset(backref, 0, sizeof(*backref));
854 backref->namelen = namelen;
855 memcpy(backref->name, name, namelen);
856 backref->name[namelen] = '\0';
857 list_add_tail(&backref->list, &rec->backrefs);
861 static int add_inode_backref(struct cache_tree *inode_cache,
862 u64 ino, u64 dir, u64 index,
863 const char *name, int namelen,
864 int filetype, int itemtype, int errors)
866 struct inode_record *rec;
867 struct inode_backref *backref;
869 rec = get_inode_rec(inode_cache, ino, 1);
870 backref = get_inode_backref(rec, name, namelen, dir);
872 backref->errors |= errors;
873 if (itemtype == BTRFS_DIR_INDEX_KEY) {
874 if (backref->found_dir_index)
875 backref->errors |= REF_ERR_DUP_DIR_INDEX;
876 if (backref->found_inode_ref && backref->index != index)
877 backref->errors |= REF_ERR_INDEX_UNMATCH;
878 if (backref->found_dir_item && backref->filetype != filetype)
879 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
881 backref->index = index;
882 backref->filetype = filetype;
883 backref->found_dir_index = 1;
884 } else if (itemtype == BTRFS_DIR_ITEM_KEY) {
886 if (backref->found_dir_item)
887 backref->errors |= REF_ERR_DUP_DIR_ITEM;
888 if (backref->found_dir_index && backref->filetype != filetype)
889 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
891 backref->filetype = filetype;
892 backref->found_dir_item = 1;
893 } else if ((itemtype == BTRFS_INODE_REF_KEY) ||
894 (itemtype == BTRFS_INODE_EXTREF_KEY)) {
895 if (backref->found_inode_ref)
896 backref->errors |= REF_ERR_DUP_INODE_REF;
897 if (backref->found_dir_index && backref->index != index)
898 backref->errors |= REF_ERR_INDEX_UNMATCH;
900 backref->index = index;
902 backref->ref_type = itemtype;
903 backref->found_inode_ref = 1;
908 maybe_free_inode_rec(inode_cache, rec);
912 static int merge_inode_recs(struct inode_record *src, struct inode_record *dst,
913 struct cache_tree *dst_cache)
915 struct inode_backref *backref;
920 list_for_each_entry(backref, &src->backrefs, list) {
921 if (backref->found_dir_index) {
922 add_inode_backref(dst_cache, dst->ino, backref->dir,
923 backref->index, backref->name,
924 backref->namelen, backref->filetype,
925 BTRFS_DIR_INDEX_KEY, backref->errors);
927 if (backref->found_dir_item) {
929 add_inode_backref(dst_cache, dst->ino,
930 backref->dir, 0, backref->name,
931 backref->namelen, backref->filetype,
932 BTRFS_DIR_ITEM_KEY, backref->errors);
934 if (backref->found_inode_ref) {
935 add_inode_backref(dst_cache, dst->ino,
936 backref->dir, backref->index,
937 backref->name, backref->namelen, 0,
938 backref->ref_type, backref->errors);
942 if (src->found_dir_item)
943 dst->found_dir_item = 1;
944 if (src->found_file_extent)
945 dst->found_file_extent = 1;
946 if (src->found_csum_item)
947 dst->found_csum_item = 1;
948 if (src->some_csum_missing)
949 dst->some_csum_missing = 1;
950 if (first_extent_gap(&dst->holes) > first_extent_gap(&src->holes)) {
951 ret = copy_file_extent_holes(&dst->holes, &src->holes);
956 BUG_ON(src->found_link < dir_count);
957 dst->found_link += src->found_link - dir_count;
958 dst->found_size += src->found_size;
959 if (src->extent_start != (u64)-1) {
960 if (dst->extent_start == (u64)-1) {
961 dst->extent_start = src->extent_start;
962 dst->extent_end = src->extent_end;
964 if (dst->extent_end > src->extent_start)
965 dst->errors |= I_ERR_FILE_EXTENT_OVERLAP;
966 else if (dst->extent_end < src->extent_start) {
967 ret = add_file_extent_hole(&dst->holes,
969 src->extent_start - dst->extent_end);
971 if (dst->extent_end < src->extent_end)
972 dst->extent_end = src->extent_end;
976 dst->errors |= src->errors;
977 if (src->found_inode_item) {
978 if (!dst->found_inode_item) {
979 dst->nlink = src->nlink;
980 dst->isize = src->isize;
981 dst->nbytes = src->nbytes;
982 dst->imode = src->imode;
983 dst->nodatasum = src->nodatasum;
984 dst->found_inode_item = 1;
986 dst->errors |= I_ERR_DUP_INODE_ITEM;
994 static int splice_shared_node(struct shared_node *src_node,
995 struct shared_node *dst_node)
997 struct cache_extent *cache;
998 struct ptr_node *node, *ins;
999 struct cache_tree *src, *dst;
1000 struct inode_record *rec, *conflict;
1001 u64 current_ino = 0;
1005 if (--src_node->refs == 0)
1007 if (src_node->current)
1008 current_ino = src_node->current->ino;
1010 src = &src_node->root_cache;
1011 dst = &dst_node->root_cache;
1013 cache = search_cache_extent(src, 0);
1015 node = container_of(cache, struct ptr_node, cache);
1017 cache = next_cache_extent(cache);
1020 remove_cache_extent(src, &node->cache);
1023 ins = malloc(sizeof(*ins));
1024 ins->cache.start = node->cache.start;
1025 ins->cache.size = node->cache.size;
1029 ret = insert_cache_extent(dst, &ins->cache);
1030 if (ret == -EEXIST) {
1031 conflict = get_inode_rec(dst, rec->ino, 1);
1032 merge_inode_recs(rec, conflict, dst);
1034 conflict->checked = 1;
1035 if (dst_node->current == conflict)
1036 dst_node->current = NULL;
1038 maybe_free_inode_rec(dst, conflict);
1039 free_inode_rec(rec);
1046 if (src == &src_node->root_cache) {
1047 src = &src_node->inode_cache;
1048 dst = &dst_node->inode_cache;
1052 if (current_ino > 0 && (!dst_node->current ||
1053 current_ino > dst_node->current->ino)) {
1054 if (dst_node->current) {
1055 dst_node->current->checked = 1;
1056 maybe_free_inode_rec(dst, dst_node->current);
1058 dst_node->current = get_inode_rec(dst, current_ino, 1);
1063 static void free_inode_ptr(struct cache_extent *cache)
1065 struct ptr_node *node;
1066 struct inode_record *rec;
1068 node = container_of(cache, struct ptr_node, cache);
1070 free_inode_rec(rec);
1074 FREE_EXTENT_CACHE_BASED_TREE(inode_recs, free_inode_ptr);
1076 static struct shared_node *find_shared_node(struct cache_tree *shared,
1079 struct cache_extent *cache;
1080 struct shared_node *node;
1082 cache = lookup_cache_extent(shared, bytenr, 1);
1084 node = container_of(cache, struct shared_node, cache);
1090 static int add_shared_node(struct cache_tree *shared, u64 bytenr, u32 refs)
1093 struct shared_node *node;
1095 node = calloc(1, sizeof(*node));
1096 node->cache.start = bytenr;
1097 node->cache.size = 1;
1098 cache_tree_init(&node->root_cache);
1099 cache_tree_init(&node->inode_cache);
1102 ret = insert_cache_extent(shared, &node->cache);
1107 static int enter_shared_node(struct btrfs_root *root, u64 bytenr, u32 refs,
1108 struct walk_control *wc, int level)
1110 struct shared_node *node;
1111 struct shared_node *dest;
1113 if (level == wc->active_node)
1116 BUG_ON(wc->active_node <= level);
1117 node = find_shared_node(&wc->shared, bytenr);
1119 add_shared_node(&wc->shared, bytenr, refs);
1120 node = find_shared_node(&wc->shared, bytenr);
1121 wc->nodes[level] = node;
1122 wc->active_node = level;
1126 if (wc->root_level == wc->active_node &&
1127 btrfs_root_refs(&root->root_item) == 0) {
1128 if (--node->refs == 0) {
1129 free_inode_recs_tree(&node->root_cache);
1130 free_inode_recs_tree(&node->inode_cache);
1131 remove_cache_extent(&wc->shared, &node->cache);
1137 dest = wc->nodes[wc->active_node];
1138 splice_shared_node(node, dest);
1139 if (node->refs == 0) {
1140 remove_cache_extent(&wc->shared, &node->cache);
1146 static int leave_shared_node(struct btrfs_root *root,
1147 struct walk_control *wc, int level)
1149 struct shared_node *node;
1150 struct shared_node *dest;
1153 if (level == wc->root_level)
1156 for (i = level + 1; i < BTRFS_MAX_LEVEL; i++) {
1160 BUG_ON(i >= BTRFS_MAX_LEVEL);
1162 node = wc->nodes[wc->active_node];
1163 wc->nodes[wc->active_node] = NULL;
1164 wc->active_node = i;
1166 dest = wc->nodes[wc->active_node];
1167 if (wc->active_node < wc->root_level ||
1168 btrfs_root_refs(&root->root_item) > 0) {
1169 BUG_ON(node->refs <= 1);
1170 splice_shared_node(node, dest);
1172 BUG_ON(node->refs < 2);
1181 * 1 - if the root with id child_root_id is a child of root parent_root_id
1182 * 0 - if the root child_root_id isn't a child of the root parent_root_id but
1183 * has other root(s) as parent(s)
1184 * 2 - if the root child_root_id doesn't have any parent roots
1186 static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
1189 struct btrfs_path path;
1190 struct btrfs_key key;
1191 struct extent_buffer *leaf;
1195 btrfs_init_path(&path);
1197 key.objectid = parent_root_id;
1198 key.type = BTRFS_ROOT_REF_KEY;
1199 key.offset = child_root_id;
1200 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
1204 btrfs_release_path(&path);
1208 key.objectid = child_root_id;
1209 key.type = BTRFS_ROOT_BACKREF_KEY;
1211 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
1217 leaf = path.nodes[0];
1218 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1219 ret = btrfs_next_leaf(root->fs_info->tree_root, &path);
1222 leaf = path.nodes[0];
1225 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1226 if (key.objectid != child_root_id ||
1227 key.type != BTRFS_ROOT_BACKREF_KEY)
1232 if (key.offset == parent_root_id) {
1233 btrfs_release_path(&path);
1240 btrfs_release_path(&path);
1243 return has_parent ? 0 : 2;
1246 static int process_dir_item(struct btrfs_root *root,
1247 struct extent_buffer *eb,
1248 int slot, struct btrfs_key *key,
1249 struct shared_node *active_node)
1259 struct btrfs_dir_item *di;
1260 struct inode_record *rec;
1261 struct cache_tree *root_cache;
1262 struct cache_tree *inode_cache;
1263 struct btrfs_key location;
1264 char namebuf[BTRFS_NAME_LEN];
1266 root_cache = &active_node->root_cache;
1267 inode_cache = &active_node->inode_cache;
1268 rec = active_node->current;
1269 rec->found_dir_item = 1;
1271 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1272 total = btrfs_item_size_nr(eb, slot);
1273 while (cur < total) {
1275 btrfs_dir_item_key_to_cpu(eb, di, &location);
1276 name_len = btrfs_dir_name_len(eb, di);
1277 data_len = btrfs_dir_data_len(eb, di);
1278 filetype = btrfs_dir_type(eb, di);
1280 rec->found_size += name_len;
1281 if (name_len <= BTRFS_NAME_LEN) {
1285 len = BTRFS_NAME_LEN;
1286 error = REF_ERR_NAME_TOO_LONG;
1288 read_extent_buffer(eb, namebuf, (unsigned long)(di + 1), len);
1290 if (location.type == BTRFS_INODE_ITEM_KEY) {
1291 add_inode_backref(inode_cache, location.objectid,
1292 key->objectid, key->offset, namebuf,
1293 len, filetype, key->type, error);
1294 } else if (location.type == BTRFS_ROOT_ITEM_KEY) {
1295 add_inode_backref(root_cache, location.objectid,
1296 key->objectid, key->offset,
1297 namebuf, len, filetype,
1300 fprintf(stderr, "invalid location in dir item %u\n",
1302 add_inode_backref(inode_cache, BTRFS_MULTIPLE_OBJECTIDS,
1303 key->objectid, key->offset, namebuf,
1304 len, filetype, key->type, error);
1307 len = sizeof(*di) + name_len + data_len;
1308 di = (struct btrfs_dir_item *)((char *)di + len);
1311 if (key->type == BTRFS_DIR_INDEX_KEY && nritems > 1)
1312 rec->errors |= I_ERR_DUP_DIR_INDEX;
1317 static int process_inode_ref(struct extent_buffer *eb,
1318 int slot, struct btrfs_key *key,
1319 struct shared_node *active_node)
1327 struct cache_tree *inode_cache;
1328 struct btrfs_inode_ref *ref;
1329 char namebuf[BTRFS_NAME_LEN];
1331 inode_cache = &active_node->inode_cache;
1333 ref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1334 total = btrfs_item_size_nr(eb, slot);
1335 while (cur < total) {
1336 name_len = btrfs_inode_ref_name_len(eb, ref);
1337 index = btrfs_inode_ref_index(eb, ref);
1338 if (name_len <= BTRFS_NAME_LEN) {
1342 len = BTRFS_NAME_LEN;
1343 error = REF_ERR_NAME_TOO_LONG;
1345 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
1346 add_inode_backref(inode_cache, key->objectid, key->offset,
1347 index, namebuf, len, 0, key->type, error);
1349 len = sizeof(*ref) + name_len;
1350 ref = (struct btrfs_inode_ref *)((char *)ref + len);
1356 static int process_inode_extref(struct extent_buffer *eb,
1357 int slot, struct btrfs_key *key,
1358 struct shared_node *active_node)
1367 struct cache_tree *inode_cache;
1368 struct btrfs_inode_extref *extref;
1369 char namebuf[BTRFS_NAME_LEN];
1371 inode_cache = &active_node->inode_cache;
1373 extref = btrfs_item_ptr(eb, slot, struct btrfs_inode_extref);
1374 total = btrfs_item_size_nr(eb, slot);
1375 while (cur < total) {
1376 name_len = btrfs_inode_extref_name_len(eb, extref);
1377 index = btrfs_inode_extref_index(eb, extref);
1378 parent = btrfs_inode_extref_parent(eb, extref);
1379 if (name_len <= BTRFS_NAME_LEN) {
1383 len = BTRFS_NAME_LEN;
1384 error = REF_ERR_NAME_TOO_LONG;
1386 read_extent_buffer(eb, namebuf,
1387 (unsigned long)(extref + 1), len);
1388 add_inode_backref(inode_cache, key->objectid, parent,
1389 index, namebuf, len, 0, key->type, error);
1391 len = sizeof(*extref) + name_len;
1392 extref = (struct btrfs_inode_extref *)((char *)extref + len);
1399 static int count_csum_range(struct btrfs_root *root, u64 start,
1400 u64 len, u64 *found)
1402 struct btrfs_key key;
1403 struct btrfs_path path;
1404 struct extent_buffer *leaf;
1409 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
1411 btrfs_init_path(&path);
1413 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1415 key.type = BTRFS_EXTENT_CSUM_KEY;
1417 ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
1421 if (ret > 0 && path.slots[0] > 0) {
1422 leaf = path.nodes[0];
1423 btrfs_item_key_to_cpu(leaf, &key, path.slots[0] - 1);
1424 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
1425 key.type == BTRFS_EXTENT_CSUM_KEY)
1430 leaf = path.nodes[0];
1431 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1432 ret = btrfs_next_leaf(root->fs_info->csum_root, &path);
1437 leaf = path.nodes[0];
1440 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1441 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1442 key.type != BTRFS_EXTENT_CSUM_KEY)
1445 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1446 if (key.offset >= start + len)
1449 if (key.offset > start)
1452 size = btrfs_item_size_nr(leaf, path.slots[0]);
1453 csum_end = key.offset + (size / csum_size) * root->sectorsize;
1454 if (csum_end > start) {
1455 size = min(csum_end - start, len);
1464 btrfs_release_path(&path);
1470 static int process_file_extent(struct btrfs_root *root,
1471 struct extent_buffer *eb,
1472 int slot, struct btrfs_key *key,
1473 struct shared_node *active_node)
1475 struct inode_record *rec;
1476 struct btrfs_file_extent_item *fi;
1478 u64 disk_bytenr = 0;
1479 u64 extent_offset = 0;
1480 u64 mask = root->sectorsize - 1;
1484 rec = active_node->current;
1485 BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1486 rec->found_file_extent = 1;
1488 if (rec->extent_start == (u64)-1) {
1489 rec->extent_start = key->offset;
1490 rec->extent_end = key->offset;
1493 if (rec->extent_end > key->offset)
1494 rec->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1495 else if (rec->extent_end < key->offset) {
1496 ret = add_file_extent_hole(&rec->holes, rec->extent_end,
1497 key->offset - rec->extent_end);
1502 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
1503 extent_type = btrfs_file_extent_type(eb, fi);
1505 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1506 num_bytes = btrfs_file_extent_inline_len(eb, slot, fi);
1508 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1509 rec->found_size += num_bytes;
1510 num_bytes = (num_bytes + mask) & ~mask;
1511 } else if (extent_type == BTRFS_FILE_EXTENT_REG ||
1512 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1513 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1514 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1515 extent_offset = btrfs_file_extent_offset(eb, fi);
1516 if (num_bytes == 0 || (num_bytes & mask))
1517 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1518 if (num_bytes + extent_offset >
1519 btrfs_file_extent_ram_bytes(eb, fi))
1520 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1521 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC &&
1522 (btrfs_file_extent_compression(eb, fi) ||
1523 btrfs_file_extent_encryption(eb, fi) ||
1524 btrfs_file_extent_other_encoding(eb, fi)))
1525 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1526 if (disk_bytenr > 0)
1527 rec->found_size += num_bytes;
1529 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1531 rec->extent_end = key->offset + num_bytes;
1533 if (disk_bytenr > 0) {
1535 if (btrfs_file_extent_compression(eb, fi))
1536 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1538 disk_bytenr += extent_offset;
1540 ret = count_csum_range(root, disk_bytenr, num_bytes, &found);
1543 if (extent_type == BTRFS_FILE_EXTENT_REG) {
1545 rec->found_csum_item = 1;
1546 if (found < num_bytes)
1547 rec->some_csum_missing = 1;
1548 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1550 rec->errors |= I_ERR_ODD_CSUM_ITEM;
1556 static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
1557 struct walk_control *wc)
1559 struct btrfs_key key;
1563 struct cache_tree *inode_cache;
1564 struct shared_node *active_node;
1566 if (wc->root_level == wc->active_node &&
1567 btrfs_root_refs(&root->root_item) == 0)
1570 active_node = wc->nodes[wc->active_node];
1571 inode_cache = &active_node->inode_cache;
1572 nritems = btrfs_header_nritems(eb);
1573 for (i = 0; i < nritems; i++) {
1574 btrfs_item_key_to_cpu(eb, &key, i);
1576 if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
1578 if (key.type == BTRFS_ORPHAN_ITEM_KEY)
1581 if (active_node->current == NULL ||
1582 active_node->current->ino < key.objectid) {
1583 if (active_node->current) {
1584 active_node->current->checked = 1;
1585 maybe_free_inode_rec(inode_cache,
1586 active_node->current);
1588 active_node->current = get_inode_rec(inode_cache,
1592 case BTRFS_DIR_ITEM_KEY:
1593 case BTRFS_DIR_INDEX_KEY:
1594 ret = process_dir_item(root, eb, i, &key, active_node);
1596 case BTRFS_INODE_REF_KEY:
1597 ret = process_inode_ref(eb, i, &key, active_node);
1599 case BTRFS_INODE_EXTREF_KEY:
1600 ret = process_inode_extref(eb, i, &key, active_node);
1602 case BTRFS_INODE_ITEM_KEY:
1603 ret = process_inode_item(eb, i, &key, active_node);
1605 case BTRFS_EXTENT_DATA_KEY:
1606 ret = process_file_extent(root, eb, i, &key,
1616 static void reada_walk_down(struct btrfs_root *root,
1617 struct extent_buffer *node, int slot)
1626 level = btrfs_header_level(node);
1630 nritems = btrfs_header_nritems(node);
1631 blocksize = btrfs_level_size(root, level - 1);
1632 for (i = slot; i < nritems; i++) {
1633 bytenr = btrfs_node_blockptr(node, i);
1634 ptr_gen = btrfs_node_ptr_generation(node, i);
1635 readahead_tree_block(root, bytenr, blocksize, ptr_gen);
1640 * Check the child node/leaf by the following condition:
1641 * 1. the first item key of the node/leaf should be the same with the one
1643 * 2. block in parent node should match the child node/leaf.
1644 * 3. generation of parent node and child's header should be consistent.
1646 * Or the child node/leaf pointed by the key in parent is not valid.
1648 * We hope to check leaf owner too, but since subvol may share leaves,
1649 * which makes leaf owner check not so strong, key check should be
1650 * sufficient enough for that case.
1652 static int check_child_node(struct btrfs_root *root,
1653 struct extent_buffer *parent, int slot,
1654 struct extent_buffer *child)
1656 struct btrfs_key parent_key;
1657 struct btrfs_key child_key;
1660 btrfs_node_key_to_cpu(parent, &parent_key, slot);
1661 if (btrfs_header_level(child) == 0)
1662 btrfs_item_key_to_cpu(child, &child_key, 0);
1664 btrfs_node_key_to_cpu(child, &child_key, 0);
1666 if (memcmp(&parent_key, &child_key, sizeof(parent_key))) {
1669 "Wrong key of child node/leaf, wanted: (%llu, %u, %llu), have: (%llu, %u, %llu)\n",
1670 parent_key.objectid, parent_key.type, parent_key.offset,
1671 child_key.objectid, child_key.type, child_key.offset);
1673 if (btrfs_header_bytenr(child) != btrfs_node_blockptr(parent, slot)) {
1675 fprintf(stderr, "Wrong block of child node/leaf, wanted: %llu, have: %llu\n",
1676 btrfs_node_blockptr(parent, slot),
1677 btrfs_header_bytenr(child));
1679 if (btrfs_node_ptr_generation(parent, slot) !=
1680 btrfs_header_generation(child)) {
1682 fprintf(stderr, "Wrong generation of child node/leaf, wanted: %llu, have: %llu\n",
1683 btrfs_header_generation(child),
1684 btrfs_node_ptr_generation(parent, slot));
1689 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
1690 struct walk_control *wc, int *level)
1692 enum btrfs_tree_block_status status;
1695 struct extent_buffer *next;
1696 struct extent_buffer *cur;
1701 WARN_ON(*level < 0);
1702 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1703 ret = btrfs_lookup_extent_info(NULL, root,
1704 path->nodes[*level]->start,
1705 *level, 1, &refs, NULL);
1712 ret = enter_shared_node(root, path->nodes[*level]->start,
1720 while (*level >= 0) {
1721 WARN_ON(*level < 0);
1722 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1723 cur = path->nodes[*level];
1725 if (btrfs_header_level(cur) != *level)
1728 if (path->slots[*level] >= btrfs_header_nritems(cur))
1731 ret = process_one_leaf(root, cur, wc);
1736 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1737 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1738 blocksize = btrfs_level_size(root, *level - 1);
1739 ret = btrfs_lookup_extent_info(NULL, root, bytenr, *level - 1,
1745 ret = enter_shared_node(root, bytenr, refs,
1748 path->slots[*level]++;
1753 next = btrfs_find_tree_block(root, bytenr, blocksize);
1754 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
1755 free_extent_buffer(next);
1756 reada_walk_down(root, cur, path->slots[*level]);
1757 next = read_tree_block(root, bytenr, blocksize,
1759 if (!extent_buffer_uptodate(next)) {
1760 struct btrfs_key node_key;
1762 btrfs_node_key_to_cpu(path->nodes[*level],
1764 path->slots[*level]);
1765 btrfs_add_corrupt_extent_record(root->fs_info,
1767 path->nodes[*level]->start,
1768 root->leafsize, *level);
1774 ret = check_child_node(root, cur, path->slots[*level], next);
1780 if (btrfs_is_leaf(next))
1781 status = btrfs_check_leaf(root, NULL, next);
1783 status = btrfs_check_node(root, NULL, next);
1784 if (status != BTRFS_TREE_BLOCK_CLEAN) {
1785 free_extent_buffer(next);
1790 *level = *level - 1;
1791 free_extent_buffer(path->nodes[*level]);
1792 path->nodes[*level] = next;
1793 path->slots[*level] = 0;
1796 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
1800 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
1801 struct walk_control *wc, int *level)
1804 struct extent_buffer *leaf;
1806 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1807 leaf = path->nodes[i];
1808 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
1813 free_extent_buffer(path->nodes[*level]);
1814 path->nodes[*level] = NULL;
1815 BUG_ON(*level > wc->active_node);
1816 if (*level == wc->active_node)
1817 leave_shared_node(root, wc, *level);
1824 static int check_root_dir(struct inode_record *rec)
1826 struct inode_backref *backref;
1829 if (!rec->found_inode_item || rec->errors)
1831 if (rec->nlink != 1 || rec->found_link != 0)
1833 if (list_empty(&rec->backrefs))
1835 backref = list_entry(rec->backrefs.next, struct inode_backref, list);
1836 if (!backref->found_inode_ref)
1838 if (backref->index != 0 || backref->namelen != 2 ||
1839 memcmp(backref->name, "..", 2))
1841 if (backref->found_dir_index || backref->found_dir_item)
1848 static int repair_inode_isize(struct btrfs_trans_handle *trans,
1849 struct btrfs_root *root, struct btrfs_path *path,
1850 struct inode_record *rec)
1852 struct btrfs_inode_item *ei;
1853 struct btrfs_key key;
1856 key.objectid = rec->ino;
1857 key.type = BTRFS_INODE_ITEM_KEY;
1858 key.offset = (u64)-1;
1860 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1864 if (!path->slots[0]) {
1871 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1872 if (key.objectid != rec->ino) {
1877 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1878 struct btrfs_inode_item);
1879 btrfs_set_inode_size(path->nodes[0], ei, rec->found_size);
1880 btrfs_mark_buffer_dirty(path->nodes[0]);
1881 rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
1882 printf("reset isize for dir %Lu root %Lu\n", rec->ino,
1883 root->root_key.objectid);
1885 btrfs_release_path(path);
1889 static int repair_inode_orphan_item(struct btrfs_trans_handle *trans,
1890 struct btrfs_root *root,
1891 struct btrfs_path *path,
1892 struct inode_record *rec)
1896 ret = btrfs_add_orphan_item(trans, root, path, rec->ino);
1897 btrfs_release_path(path);
1899 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
1903 static int add_missing_dir_index(struct btrfs_root *root,
1904 struct cache_tree *inode_cache,
1905 struct inode_record *rec,
1906 struct inode_backref *backref)
1908 struct btrfs_path *path;
1909 struct btrfs_trans_handle *trans;
1910 struct btrfs_dir_item *dir_item;
1911 struct extent_buffer *leaf;
1912 struct btrfs_key key;
1913 struct btrfs_disk_key disk_key;
1914 struct inode_record *dir_rec;
1915 unsigned long name_ptr;
1916 u32 data_size = sizeof(*dir_item) + backref->namelen;
1919 path = btrfs_alloc_path();
1923 trans = btrfs_start_transaction(root, 1);
1924 if (IS_ERR(trans)) {
1925 btrfs_free_path(path);
1926 return PTR_ERR(trans);
1929 fprintf(stderr, "repairing missing dir index item for inode %llu\n",
1930 (unsigned long long)rec->ino);
1931 key.objectid = backref->dir;
1932 key.type = BTRFS_DIR_INDEX_KEY;
1933 key.offset = backref->index;
1935 ret = btrfs_insert_empty_item(trans, root, path, &key, data_size);
1938 leaf = path->nodes[0];
1939 dir_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
1941 disk_key.objectid = cpu_to_le64(rec->ino);
1942 disk_key.type = BTRFS_INODE_ITEM_KEY;
1943 disk_key.offset = 0;
1945 btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
1946 btrfs_set_dir_type(leaf, dir_item, imode_to_type(rec->imode));
1947 btrfs_set_dir_data_len(leaf, dir_item, 0);
1948 btrfs_set_dir_name_len(leaf, dir_item, backref->namelen);
1949 name_ptr = (unsigned long)(dir_item + 1);
1950 write_extent_buffer(leaf, backref->name, name_ptr, backref->namelen);
1951 btrfs_mark_buffer_dirty(leaf);
1952 btrfs_free_path(path);
1953 btrfs_commit_transaction(trans, root);
1955 backref->found_dir_index = 1;
1956 dir_rec = get_inode_rec(inode_cache, backref->dir, 0);
1959 dir_rec->found_size += backref->namelen;
1960 if (dir_rec->found_size == dir_rec->isize &&
1961 (dir_rec->errors & I_ERR_DIR_ISIZE_WRONG))
1962 dir_rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
1963 if (dir_rec->found_size != dir_rec->isize)
1964 dir_rec->errors |= I_ERR_DIR_ISIZE_WRONG;
1969 static int delete_dir_index(struct btrfs_root *root,
1970 struct cache_tree *inode_cache,
1971 struct inode_record *rec,
1972 struct inode_backref *backref)
1974 struct btrfs_trans_handle *trans;
1975 struct btrfs_dir_item *di;
1976 struct btrfs_path *path;
1979 path = btrfs_alloc_path();
1983 trans = btrfs_start_transaction(root, 1);
1984 if (IS_ERR(trans)) {
1985 btrfs_free_path(path);
1986 return PTR_ERR(trans);
1990 fprintf(stderr, "Deleting bad dir index [%llu,%u,%llu] root %llu\n",
1991 (unsigned long long)backref->dir,
1992 BTRFS_DIR_INDEX_KEY, (unsigned long long)backref->index,
1993 (unsigned long long)root->objectid);
1995 di = btrfs_lookup_dir_index(trans, root, path, backref->dir,
1996 backref->name, backref->namelen,
1997 backref->index, -1);
2000 btrfs_free_path(path);
2001 btrfs_commit_transaction(trans, root);
2008 ret = btrfs_del_item(trans, root, path);
2010 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2012 btrfs_free_path(path);
2013 btrfs_commit_transaction(trans, root);
2017 static int create_inode_item(struct btrfs_root *root,
2018 struct inode_record *rec,
2019 struct inode_backref *backref, int root_dir)
2021 struct btrfs_trans_handle *trans;
2022 struct btrfs_inode_item inode_item;
2023 time_t now = time(NULL);
2026 trans = btrfs_start_transaction(root, 1);
2027 if (IS_ERR(trans)) {
2028 ret = PTR_ERR(trans);
2032 fprintf(stderr, "root %llu inode %llu recreating inode item, this may "
2033 "be incomplete, please check permissions and content after "
2034 "the fsck completes.\n", (unsigned long long)root->objectid,
2035 (unsigned long long)rec->ino);
2037 memset(&inode_item, 0, sizeof(inode_item));
2038 btrfs_set_stack_inode_generation(&inode_item, trans->transid);
2040 btrfs_set_stack_inode_nlink(&inode_item, 1);
2042 btrfs_set_stack_inode_nlink(&inode_item, rec->found_link);
2043 btrfs_set_stack_inode_nbytes(&inode_item, rec->found_size);
2044 if (rec->found_dir_item) {
2045 if (rec->found_file_extent)
2046 fprintf(stderr, "root %llu inode %llu has both a dir "
2047 "item and extents, unsure if it is a dir or a "
2048 "regular file so setting it as a directory\n",
2049 (unsigned long long)root->objectid,
2050 (unsigned long long)rec->ino);
2051 btrfs_set_stack_inode_mode(&inode_item, S_IFDIR | 0755);
2052 btrfs_set_stack_inode_size(&inode_item, rec->found_size);
2053 } else if (!rec->found_dir_item) {
2054 btrfs_set_stack_inode_size(&inode_item, rec->extent_end);
2055 btrfs_set_stack_inode_mode(&inode_item, S_IFREG | 0755);
2057 btrfs_set_stack_timespec_sec(&inode_item.atime, now);
2058 btrfs_set_stack_timespec_nsec(&inode_item.atime, 0);
2059 btrfs_set_stack_timespec_sec(&inode_item.ctime, now);
2060 btrfs_set_stack_timespec_nsec(&inode_item.ctime, 0);
2061 btrfs_set_stack_timespec_sec(&inode_item.mtime, now);
2062 btrfs_set_stack_timespec_nsec(&inode_item.mtime, 0);
2063 btrfs_set_stack_timespec_sec(&inode_item.otime, 0);
2064 btrfs_set_stack_timespec_nsec(&inode_item.otime, 0);
2066 ret = btrfs_insert_inode(trans, root, rec->ino, &inode_item);
2068 btrfs_commit_transaction(trans, root);
2072 static int repair_inode_backrefs(struct btrfs_root *root,
2073 struct inode_record *rec,
2074 struct cache_tree *inode_cache,
2077 struct inode_backref *tmp, *backref;
2078 u64 root_dirid = btrfs_root_dirid(&root->root_item);
2082 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
2083 if (!delete && rec->ino == root_dirid) {
2084 if (!rec->found_inode_item) {
2085 ret = create_inode_item(root, rec, backref, 1);
2092 /* Index 0 for root dir's are special, don't mess with it */
2093 if (rec->ino == root_dirid && backref->index == 0)
2097 ((backref->found_dir_index && !backref->found_inode_ref) ||
2098 (backref->found_dir_index && backref->found_inode_ref &&
2099 (backref->errors & REF_ERR_INDEX_UNMATCH)))) {
2100 ret = delete_dir_index(root, inode_cache, rec, backref);
2104 list_del(&backref->list);
2108 if (!delete && !backref->found_dir_index &&
2109 backref->found_dir_item && backref->found_inode_ref) {
2110 ret = add_missing_dir_index(root, inode_cache, rec,
2115 if (backref->found_dir_item &&
2116 backref->found_dir_index &&
2117 backref->found_dir_index) {
2118 if (!backref->errors &&
2119 backref->found_inode_ref) {
2120 list_del(&backref->list);
2126 if (!delete && (!backref->found_dir_index &&
2127 !backref->found_dir_item &&
2128 backref->found_inode_ref)) {
2129 struct btrfs_trans_handle *trans;
2130 struct btrfs_key location;
2132 ret = check_dir_conflict(root, backref->name,
2138 * let nlink fixing routine to handle it,
2139 * which can do it better.
2144 location.objectid = rec->ino;
2145 location.type = BTRFS_INODE_ITEM_KEY;
2146 location.offset = 0;
2148 trans = btrfs_start_transaction(root, 1);
2149 if (IS_ERR(trans)) {
2150 ret = PTR_ERR(trans);
2153 fprintf(stderr, "adding missing dir index/item pair "
2155 (unsigned long long)rec->ino);
2156 ret = btrfs_insert_dir_item(trans, root, backref->name,
2158 backref->dir, &location,
2159 imode_to_type(rec->imode),
2162 btrfs_commit_transaction(trans, root);
2166 if (!delete && (backref->found_inode_ref &&
2167 backref->found_dir_index &&
2168 backref->found_dir_item &&
2169 !(backref->errors & REF_ERR_INDEX_UNMATCH) &&
2170 !rec->found_inode_item)) {
2171 ret = create_inode_item(root, rec, backref, 0);
2178 return ret ? ret : repaired;
2182 * To determine the file type for nlink/inode_item repair
2184 * Return 0 if file type is found and BTRFS_FT_* is stored into type.
2185 * Return -ENOENT if file type is not found.
2187 static int find_file_type(struct inode_record *rec, u8 *type)
2189 struct inode_backref *backref;
2191 /* For inode item recovered case */
2192 if (rec->found_inode_item) {
2193 *type = imode_to_type(rec->imode);
2197 list_for_each_entry(backref, &rec->backrefs, list) {
2198 if (backref->found_dir_index || backref->found_dir_item) {
2199 *type = backref->filetype;
2207 * To determine the file name for nlink repair
2209 * Return 0 if file name is found, set name and namelen.
2210 * Return -ENOENT if file name is not found.
2212 static int find_file_name(struct inode_record *rec,
2213 char *name, int *namelen)
2215 struct inode_backref *backref;
2217 list_for_each_entry(backref, &rec->backrefs, list) {
2218 if (backref->found_dir_index || backref->found_dir_item ||
2219 backref->found_inode_ref) {
2220 memcpy(name, backref->name, backref->namelen);
2221 *namelen = backref->namelen;
2228 /* Reset the nlink of the inode to the correct one */
2229 static int reset_nlink(struct btrfs_trans_handle *trans,
2230 struct btrfs_root *root,
2231 struct btrfs_path *path,
2232 struct inode_record *rec)
2234 struct inode_backref *backref;
2235 struct inode_backref *tmp;
2236 struct btrfs_key key;
2237 struct btrfs_inode_item *inode_item;
2240 /* We don't believe this either, reset it and iterate backref */
2241 rec->found_link = 0;
2243 /* Remove all backref including the valid ones */
2244 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
2245 ret = btrfs_unlink(trans, root, rec->ino, backref->dir,
2246 backref->index, backref->name,
2247 backref->namelen, 0);
2251 /* remove invalid backref, so it won't be added back */
2252 if (!(backref->found_dir_index &&
2253 backref->found_dir_item &&
2254 backref->found_inode_ref)) {
2255 list_del(&backref->list);
2262 /* Set nlink to 0 */
2263 key.objectid = rec->ino;
2264 key.type = BTRFS_INODE_ITEM_KEY;
2266 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2273 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2274 struct btrfs_inode_item);
2275 btrfs_set_inode_nlink(path->nodes[0], inode_item, 0);
2276 btrfs_mark_buffer_dirty(path->nodes[0]);
2277 btrfs_release_path(path);
2280 * Add back valid inode_ref/dir_item/dir_index,
2281 * add_link() will handle the nlink inc, so new nlink must be correct
2283 list_for_each_entry(backref, &rec->backrefs, list) {
2284 ret = btrfs_add_link(trans, root, rec->ino, backref->dir,
2285 backref->name, backref->namelen,
2286 backref->ref_type, &backref->index, 1);
2291 btrfs_release_path(path);
2295 static int repair_inode_nlinks(struct btrfs_trans_handle *trans,
2296 struct btrfs_root *root,
2297 struct btrfs_path *path,
2298 struct inode_record *rec)
2300 char *dir_name = "lost+found";
2301 char namebuf[BTRFS_NAME_LEN] = {0};
2306 int name_recovered = 0;
2307 int type_recovered = 0;
2311 * Get file name and type first before these invalid inode ref
2312 * are deleted by remove_all_invalid_backref()
2314 name_recovered = !find_file_name(rec, namebuf, &namelen);
2315 type_recovered = !find_file_type(rec, &type);
2317 if (!name_recovered) {
2318 printf("Can't get file name for inode %llu, using '%llu' as fallback\n",
2319 rec->ino, rec->ino);
2320 namelen = count_digits(rec->ino);
2321 sprintf(namebuf, "%llu", rec->ino);
2324 if (!type_recovered) {
2325 printf("Can't get file type for inode %llu, using FILE as fallback\n",
2327 type = BTRFS_FT_REG_FILE;
2331 ret = reset_nlink(trans, root, path, rec);
2334 "Failed to reset nlink for inode %llu: %s\n",
2335 rec->ino, strerror(-ret));
2339 if (rec->found_link == 0) {
2340 lost_found_ino = root->highest_inode;
2341 if (lost_found_ino >= BTRFS_LAST_FREE_OBJECTID) {
2346 ret = btrfs_mkdir(trans, root, dir_name, strlen(dir_name),
2347 BTRFS_FIRST_FREE_OBJECTID, &lost_found_ino,
2350 fprintf(stderr, "Failed to create '%s' dir: %s",
2351 dir_name, strerror(-ret));
2354 ret = btrfs_add_link(trans, root, rec->ino, lost_found_ino,
2355 namebuf, namelen, type, NULL, 1);
2356 if (ret == -EEXIST) {
2358 * Conflicting file name, add ".INO" as suffix * +1 for '.'
2360 if (namelen + count_digits(rec->ino) + 1 >
2365 snprintf(namebuf + namelen, BTRFS_NAME_LEN - namelen,
2367 namelen += count_digits(rec->ino) + 1;
2368 ret = btrfs_add_link(trans, root, rec->ino,
2369 lost_found_ino, namebuf,
2370 namelen, type, NULL, 1);
2374 "Failed to link the inode %llu to %s dir: %s",
2375 rec->ino, dir_name, strerror(-ret));
2379 * Just increase the found_link, don't actually add the
2380 * backref. This will make things easier and this inode
2381 * record will be freed after the repair is done.
2382 * So fsck will not report problem about this inode.
2385 printf("Moving file '%.*s' to '%s' dir since it has no valid backref\n",
2386 namelen, namebuf, dir_name);
2388 rec->errors &= ~I_ERR_LINK_COUNT_WRONG;
2389 printf("Fixed the nlink of inode %llu\n", rec->ino);
2391 btrfs_release_path(path);
2396 * Check if there is any normal(reg or prealloc) file extent for given
2398 * This is used to determine the file type when neither its dir_index/item or
2399 * inode_item exists.
2401 * This will *NOT* report error, if any error happens, just consider it does
2402 * not have any normal file extent.
2404 static int find_normal_file_extent(struct btrfs_root *root, u64 ino)
2406 struct btrfs_path *path;
2407 struct btrfs_key key;
2408 struct btrfs_key found_key;
2409 struct btrfs_file_extent_item *fi;
2413 path = btrfs_alloc_path();
2417 key.type = BTRFS_EXTENT_DATA_KEY;
2420 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2425 if (ret && path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
2426 ret = btrfs_next_leaf(root, path);
2433 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2435 if (found_key.objectid != ino ||
2436 found_key.type != BTRFS_EXTENT_DATA_KEY)
2438 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
2439 struct btrfs_file_extent_item);
2440 type = btrfs_file_extent_type(path->nodes[0], fi);
2441 if (type != BTRFS_FILE_EXTENT_INLINE) {
2447 btrfs_free_path(path);
2451 static u32 btrfs_type_to_imode(u8 type)
2453 static u32 imode_by_btrfs_type[] = {
2454 [BTRFS_FT_REG_FILE] = S_IFREG,
2455 [BTRFS_FT_DIR] = S_IFDIR,
2456 [BTRFS_FT_CHRDEV] = S_IFCHR,
2457 [BTRFS_FT_BLKDEV] = S_IFBLK,
2458 [BTRFS_FT_FIFO] = S_IFIFO,
2459 [BTRFS_FT_SOCK] = S_IFSOCK,
2460 [BTRFS_FT_SYMLINK] = S_IFLNK,
2463 return imode_by_btrfs_type[(type)];
2466 static int repair_inode_no_item(struct btrfs_trans_handle *trans,
2467 struct btrfs_root *root,
2468 struct btrfs_path *path,
2469 struct inode_record *rec)
2473 int type_recovered = 0;
2476 printf("Trying to rebuild inode:%llu\n", rec->ino);
2478 type_recovered = !find_file_type(rec, &filetype);
2481 * Try to determine inode type if type not found.
2483 * For found regular file extent, it must be FILE.
2484 * For found dir_item/index, it must be DIR.
2486 * For undetermined one, use FILE as fallback.
2489 * 1. If found backref(inode_index/item is already handled) to it,
2491 * Need new inode-inode ref structure to allow search for that.
2493 if (!type_recovered) {
2494 if (rec->found_file_extent &&
2495 find_normal_file_extent(root, rec->ino)) {
2497 filetype = BTRFS_FT_REG_FILE;
2498 } else if (rec->found_dir_item) {
2500 filetype = BTRFS_FT_DIR;
2501 } else if (!list_empty(&rec->orphan_extents)) {
2503 filetype = BTRFS_FT_REG_FILE;
2505 printf("Can't determint the filetype for inode %llu, assume it is a normal file\n",
2508 filetype = BTRFS_FT_REG_FILE;
2512 ret = btrfs_new_inode(trans, root, rec->ino,
2513 mode | btrfs_type_to_imode(filetype));
2518 * Here inode rebuild is done, we only rebuild the inode item,
2519 * don't repair the nlink(like move to lost+found).
2520 * That is the job of nlink repair.
2522 * We just fill the record and return
2524 rec->found_dir_item = 1;
2525 rec->imode = mode | btrfs_type_to_imode(filetype);
2527 rec->errors &= ~I_ERR_NO_INODE_ITEM;
2528 /* Ensure the inode_nlinks repair function will be called */
2529 rec->errors |= I_ERR_LINK_COUNT_WRONG;
2534 static int repair_inode_orphan_extent(struct btrfs_trans_handle *trans,
2535 struct btrfs_root *root,
2536 struct btrfs_path *path,
2537 struct inode_record *rec)
2539 struct orphan_data_extent *orphan;
2540 struct orphan_data_extent *tmp;
2543 list_for_each_entry_safe(orphan, tmp, &rec->orphan_extents, list) {
2545 * Check for conflicting file extents
2547 * Here we don't know whether the extents is compressed or not,
2548 * so we can only assume it not compressed nor data offset,
2549 * and use its disk_len as extent length.
2551 ret = btrfs_get_extent(NULL, root, path, orphan->objectid,
2552 orphan->offset, orphan->disk_len, 0);
2553 btrfs_release_path(path);
2558 "orphan extent (%llu, %llu) conflicts, delete the orphan\n",
2559 orphan->disk_bytenr, orphan->disk_len);
2560 ret = btrfs_free_extent(trans,
2561 root->fs_info->extent_root,
2562 orphan->disk_bytenr, orphan->disk_len,
2563 0, root->objectid, orphan->objectid,
2568 ret = btrfs_insert_file_extent(trans, root, orphan->objectid,
2569 orphan->offset, orphan->disk_bytenr,
2570 orphan->disk_len, orphan->disk_len);
2574 /* Update file size info */
2575 rec->found_size += orphan->disk_len;
2576 if (rec->found_size == rec->nbytes)
2577 rec->errors &= ~I_ERR_FILE_NBYTES_WRONG;
2579 /* Update the file extent hole info too */
2580 ret = del_file_extent_hole(&rec->holes, orphan->offset,
2584 if (RB_EMPTY_ROOT(&rec->holes))
2585 rec->errors &= ~I_ERR_FILE_EXTENT_DISCOUNT;
2587 list_del(&orphan->list);
2590 rec->errors &= ~I_ERR_FILE_EXTENT_ORPHAN;
2595 static int repair_inode_discount_extent(struct btrfs_trans_handle *trans,
2596 struct btrfs_root *root,
2597 struct btrfs_path *path,
2598 struct inode_record *rec)
2600 struct rb_node *node;
2601 struct file_extent_hole *hole;
2604 node = rb_first(&rec->holes);
2607 hole = rb_entry(node, struct file_extent_hole, node);
2608 ret = btrfs_punch_hole(trans, root, rec->ino,
2609 hole->start, hole->len);
2612 ret = del_file_extent_hole(&rec->holes, hole->start,
2616 if (RB_EMPTY_ROOT(&rec->holes))
2617 rec->errors &= ~I_ERR_FILE_EXTENT_DISCOUNT;
2618 node = rb_first(&rec->holes);
2620 printf("Fixed discount file extents for inode: %llu in root: %llu\n",
2621 rec->ino, root->objectid);
2626 static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
2628 struct btrfs_trans_handle *trans;
2629 struct btrfs_path *path;
2632 if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG |
2633 I_ERR_NO_ORPHAN_ITEM |
2634 I_ERR_LINK_COUNT_WRONG |
2635 I_ERR_NO_INODE_ITEM |
2636 I_ERR_FILE_EXTENT_ORPHAN |
2637 I_ERR_FILE_EXTENT_DISCOUNT)))
2640 path = btrfs_alloc_path();
2645 * For nlink repair, it may create a dir and add link, so
2646 * 2 for parent(256)'s dir_index and dir_item
2647 * 2 for lost+found dir's inode_item and inode_ref
2648 * 1 for the new inode_ref of the file
2649 * 2 for lost+found dir's dir_index and dir_item for the file
2651 trans = btrfs_start_transaction(root, 7);
2652 if (IS_ERR(trans)) {
2653 btrfs_free_path(path);
2654 return PTR_ERR(trans);
2657 if (rec->errors & I_ERR_NO_INODE_ITEM)
2658 ret = repair_inode_no_item(trans, root, path, rec);
2659 if (!ret && rec->errors & I_ERR_FILE_EXTENT_ORPHAN)
2660 ret = repair_inode_orphan_extent(trans, root, path, rec);
2661 if (!ret && rec->errors & I_ERR_FILE_EXTENT_DISCOUNT)
2662 ret = repair_inode_discount_extent(trans, root, path, rec);
2663 if (!ret && rec->errors & I_ERR_DIR_ISIZE_WRONG)
2664 ret = repair_inode_isize(trans, root, path, rec);
2665 if (!ret && rec->errors & I_ERR_NO_ORPHAN_ITEM)
2666 ret = repair_inode_orphan_item(trans, root, path, rec);
2667 if (!ret && rec->errors & I_ERR_LINK_COUNT_WRONG)
2668 ret = repair_inode_nlinks(trans, root, path, rec);
2669 btrfs_commit_transaction(trans, root);
2670 btrfs_free_path(path);
2674 static int check_inode_recs(struct btrfs_root *root,
2675 struct cache_tree *inode_cache)
2677 struct cache_extent *cache;
2678 struct ptr_node *node;
2679 struct inode_record *rec;
2680 struct inode_backref *backref;
2685 u64 root_dirid = btrfs_root_dirid(&root->root_item);
2687 if (btrfs_root_refs(&root->root_item) == 0) {
2688 if (!cache_tree_empty(inode_cache))
2689 fprintf(stderr, "warning line %d\n", __LINE__);
2694 * We need to record the highest inode number for later 'lost+found'
2696 * We must select a ino not used/refered by any existing inode, or
2697 * 'lost+found' ino may be a missing ino in a corrupted leaf,
2698 * this may cause 'lost+found' dir has wrong nlinks.
2700 cache = last_cache_extent(inode_cache);
2702 node = container_of(cache, struct ptr_node, cache);
2704 if (rec->ino > root->highest_inode)
2705 root->highest_inode = rec->ino;
2709 * We need to repair backrefs first because we could change some of the
2710 * errors in the inode recs.
2712 * We also need to go through and delete invalid backrefs first and then
2713 * add the correct ones second. We do this because we may get EEXIST
2714 * when adding back the correct index because we hadn't yet deleted the
2717 * For example, if we were missing a dir index then the directories
2718 * isize would be wrong, so if we fixed the isize to what we thought it
2719 * would be and then fixed the backref we'd still have a invalid fs, so
2720 * we need to add back the dir index and then check to see if the isize
2725 if (stage == 3 && !err)
2728 cache = search_cache_extent(inode_cache, 0);
2729 while (repair && cache) {
2730 node = container_of(cache, struct ptr_node, cache);
2732 cache = next_cache_extent(cache);
2734 /* Need to free everything up and rescan */
2736 remove_cache_extent(inode_cache, &node->cache);
2738 free_inode_rec(rec);
2742 if (list_empty(&rec->backrefs))
2745 ret = repair_inode_backrefs(root, rec, inode_cache,
2759 rec = get_inode_rec(inode_cache, root_dirid, 0);
2761 ret = check_root_dir(rec);
2763 fprintf(stderr, "root %llu root dir %llu error\n",
2764 (unsigned long long)root->root_key.objectid,
2765 (unsigned long long)root_dirid);
2766 print_inode_error(root, rec);
2771 struct btrfs_trans_handle *trans;
2773 trans = btrfs_start_transaction(root, 1);
2774 if (IS_ERR(trans)) {
2775 err = PTR_ERR(trans);
2780 "root %llu missing its root dir, recreating\n",
2781 (unsigned long long)root->objectid);
2783 ret = btrfs_make_root_dir(trans, root, root_dirid);
2786 btrfs_commit_transaction(trans, root);
2790 fprintf(stderr, "root %llu root dir %llu not found\n",
2791 (unsigned long long)root->root_key.objectid,
2792 (unsigned long long)root_dirid);
2796 cache = search_cache_extent(inode_cache, 0);
2799 node = container_of(cache, struct ptr_node, cache);
2801 remove_cache_extent(inode_cache, &node->cache);
2803 if (rec->ino == root_dirid ||
2804 rec->ino == BTRFS_ORPHAN_OBJECTID) {
2805 free_inode_rec(rec);
2809 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
2810 ret = check_orphan_item(root, rec->ino);
2812 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
2813 if (can_free_inode_rec(rec)) {
2814 free_inode_rec(rec);
2819 if (!rec->found_inode_item)
2820 rec->errors |= I_ERR_NO_INODE_ITEM;
2821 if (rec->found_link != rec->nlink)
2822 rec->errors |= I_ERR_LINK_COUNT_WRONG;
2824 ret = try_repair_inode(root, rec);
2825 if (ret == 0 && can_free_inode_rec(rec)) {
2826 free_inode_rec(rec);
2832 if (!(repair && ret == 0))
2834 print_inode_error(root, rec);
2835 list_for_each_entry(backref, &rec->backrefs, list) {
2836 if (!backref->found_dir_item)
2837 backref->errors |= REF_ERR_NO_DIR_ITEM;
2838 if (!backref->found_dir_index)
2839 backref->errors |= REF_ERR_NO_DIR_INDEX;
2840 if (!backref->found_inode_ref)
2841 backref->errors |= REF_ERR_NO_INODE_REF;
2842 fprintf(stderr, "\tunresolved ref dir %llu index %llu"
2843 " namelen %u name %s filetype %d errors %x",
2844 (unsigned long long)backref->dir,
2845 (unsigned long long)backref->index,
2846 backref->namelen, backref->name,
2847 backref->filetype, backref->errors);
2848 print_ref_error(backref->errors);
2850 free_inode_rec(rec);
2852 return (error > 0) ? -1 : 0;
2855 static struct root_record *get_root_rec(struct cache_tree *root_cache,
2858 struct cache_extent *cache;
2859 struct root_record *rec = NULL;
2862 cache = lookup_cache_extent(root_cache, objectid, 1);
2864 rec = container_of(cache, struct root_record, cache);
2866 rec = calloc(1, sizeof(*rec));
2867 rec->objectid = objectid;
2868 INIT_LIST_HEAD(&rec->backrefs);
2869 rec->cache.start = objectid;
2870 rec->cache.size = 1;
2872 ret = insert_cache_extent(root_cache, &rec->cache);
2878 static struct root_backref *get_root_backref(struct root_record *rec,
2879 u64 ref_root, u64 dir, u64 index,
2880 const char *name, int namelen)
2882 struct root_backref *backref;
2884 list_for_each_entry(backref, &rec->backrefs, list) {
2885 if (backref->ref_root != ref_root || backref->dir != dir ||
2886 backref->namelen != namelen)
2888 if (memcmp(name, backref->name, namelen))
2893 backref = malloc(sizeof(*backref) + namelen + 1);
2894 memset(backref, 0, sizeof(*backref));
2895 backref->ref_root = ref_root;
2897 backref->index = index;
2898 backref->namelen = namelen;
2899 memcpy(backref->name, name, namelen);
2900 backref->name[namelen] = '\0';
2901 list_add_tail(&backref->list, &rec->backrefs);
2905 static void free_root_record(struct cache_extent *cache)
2907 struct root_record *rec;
2908 struct root_backref *backref;
2910 rec = container_of(cache, struct root_record, cache);
2911 while (!list_empty(&rec->backrefs)) {
2912 backref = list_entry(rec->backrefs.next,
2913 struct root_backref, list);
2914 list_del(&backref->list);
2921 FREE_EXTENT_CACHE_BASED_TREE(root_recs, free_root_record);
2923 static int add_root_backref(struct cache_tree *root_cache,
2924 u64 root_id, u64 ref_root, u64 dir, u64 index,
2925 const char *name, int namelen,
2926 int item_type, int errors)
2928 struct root_record *rec;
2929 struct root_backref *backref;
2931 rec = get_root_rec(root_cache, root_id);
2932 backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
2934 backref->errors |= errors;
2936 if (item_type != BTRFS_DIR_ITEM_KEY) {
2937 if (backref->found_dir_index || backref->found_back_ref ||
2938 backref->found_forward_ref) {
2939 if (backref->index != index)
2940 backref->errors |= REF_ERR_INDEX_UNMATCH;
2942 backref->index = index;
2946 if (item_type == BTRFS_DIR_ITEM_KEY) {
2947 if (backref->found_forward_ref)
2949 backref->found_dir_item = 1;
2950 } else if (item_type == BTRFS_DIR_INDEX_KEY) {
2951 backref->found_dir_index = 1;
2952 } else if (item_type == BTRFS_ROOT_REF_KEY) {
2953 if (backref->found_forward_ref)
2954 backref->errors |= REF_ERR_DUP_ROOT_REF;
2955 else if (backref->found_dir_item)
2957 backref->found_forward_ref = 1;
2958 } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
2959 if (backref->found_back_ref)
2960 backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
2961 backref->found_back_ref = 1;
2966 if (backref->found_forward_ref && backref->found_dir_item)
2967 backref->reachable = 1;
2971 static int merge_root_recs(struct btrfs_root *root,
2972 struct cache_tree *src_cache,
2973 struct cache_tree *dst_cache)
2975 struct cache_extent *cache;
2976 struct ptr_node *node;
2977 struct inode_record *rec;
2978 struct inode_backref *backref;
2981 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2982 free_inode_recs_tree(src_cache);
2987 cache = search_cache_extent(src_cache, 0);
2990 node = container_of(cache, struct ptr_node, cache);
2992 remove_cache_extent(src_cache, &node->cache);
2995 ret = is_child_root(root, root->objectid, rec->ino);
3001 list_for_each_entry(backref, &rec->backrefs, list) {
3002 BUG_ON(backref->found_inode_ref);
3003 if (backref->found_dir_item)
3004 add_root_backref(dst_cache, rec->ino,
3005 root->root_key.objectid, backref->dir,
3006 backref->index, backref->name,
3007 backref->namelen, BTRFS_DIR_ITEM_KEY,
3009 if (backref->found_dir_index)
3010 add_root_backref(dst_cache, rec->ino,
3011 root->root_key.objectid, backref->dir,
3012 backref->index, backref->name,
3013 backref->namelen, BTRFS_DIR_INDEX_KEY,
3017 free_inode_rec(rec);
3024 static int check_root_refs(struct btrfs_root *root,
3025 struct cache_tree *root_cache)
3027 struct root_record *rec;
3028 struct root_record *ref_root;
3029 struct root_backref *backref;
3030 struct cache_extent *cache;
3036 rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
3039 /* fixme: this can not detect circular references */
3042 cache = search_cache_extent(root_cache, 0);
3046 rec = container_of(cache, struct root_record, cache);
3047 cache = next_cache_extent(cache);
3049 if (rec->found_ref == 0)
3052 list_for_each_entry(backref, &rec->backrefs, list) {
3053 if (!backref->reachable)
3056 ref_root = get_root_rec(root_cache,
3058 if (ref_root->found_ref > 0)
3061 backref->reachable = 0;
3063 if (rec->found_ref == 0)
3069 cache = search_cache_extent(root_cache, 0);
3073 rec = container_of(cache, struct root_record, cache);
3074 cache = next_cache_extent(cache);
3076 if (rec->found_ref == 0 &&
3077 rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
3078 rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
3079 ret = check_orphan_item(root->fs_info->tree_root,
3085 * If we don't have a root item then we likely just have
3086 * a dir item in a snapshot for this root but no actual
3087 * ref key or anything so it's meaningless.
3089 if (!rec->found_root_item)
3092 fprintf(stderr, "fs tree %llu not referenced\n",
3093 (unsigned long long)rec->objectid);
3097 if (rec->found_ref > 0 && !rec->found_root_item)
3099 list_for_each_entry(backref, &rec->backrefs, list) {
3100 if (!backref->found_dir_item)
3101 backref->errors |= REF_ERR_NO_DIR_ITEM;
3102 if (!backref->found_dir_index)
3103 backref->errors |= REF_ERR_NO_DIR_INDEX;
3104 if (!backref->found_back_ref)
3105 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
3106 if (!backref->found_forward_ref)
3107 backref->errors |= REF_ERR_NO_ROOT_REF;
3108 if (backref->reachable && backref->errors)
3115 fprintf(stderr, "fs tree %llu refs %u %s\n",
3116 (unsigned long long)rec->objectid, rec->found_ref,
3117 rec->found_root_item ? "" : "not found");
3119 list_for_each_entry(backref, &rec->backrefs, list) {
3120 if (!backref->reachable)
3122 if (!backref->errors && rec->found_root_item)
3124 fprintf(stderr, "\tunresolved ref root %llu dir %llu"
3125 " index %llu namelen %u name %s errors %x\n",
3126 (unsigned long long)backref->ref_root,
3127 (unsigned long long)backref->dir,
3128 (unsigned long long)backref->index,
3129 backref->namelen, backref->name,
3131 print_ref_error(backref->errors);
3134 return errors > 0 ? 1 : 0;
3137 static int process_root_ref(struct extent_buffer *eb, int slot,
3138 struct btrfs_key *key,
3139 struct cache_tree *root_cache)
3145 struct btrfs_root_ref *ref;
3146 char namebuf[BTRFS_NAME_LEN];
3149 ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
3151 dirid = btrfs_root_ref_dirid(eb, ref);
3152 index = btrfs_root_ref_sequence(eb, ref);
3153 name_len = btrfs_root_ref_name_len(eb, ref);
3155 if (name_len <= BTRFS_NAME_LEN) {
3159 len = BTRFS_NAME_LEN;
3160 error = REF_ERR_NAME_TOO_LONG;
3162 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
3164 if (key->type == BTRFS_ROOT_REF_KEY) {
3165 add_root_backref(root_cache, key->offset, key->objectid, dirid,
3166 index, namebuf, len, key->type, error);
3168 add_root_backref(root_cache, key->objectid, key->offset, dirid,
3169 index, namebuf, len, key->type, error);
3174 static void free_corrupt_block(struct cache_extent *cache)
3176 struct btrfs_corrupt_block *corrupt;
3178 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
3182 FREE_EXTENT_CACHE_BASED_TREE(corrupt_blocks, free_corrupt_block);
3185 * Repair the btree of the given root.
3187 * The fix is to remove the node key in corrupt_blocks cache_tree.
3188 * and rebalance the tree.
3189 * After the fix, the btree should be writeable.
3191 static int repair_btree(struct btrfs_root *root,
3192 struct cache_tree *corrupt_blocks)
3194 struct btrfs_trans_handle *trans;
3195 struct btrfs_path *path;
3196 struct btrfs_corrupt_block *corrupt;
3197 struct cache_extent *cache;
3198 struct btrfs_key key;
3203 if (cache_tree_empty(corrupt_blocks))
3206 path = btrfs_alloc_path();
3210 trans = btrfs_start_transaction(root, 1);
3211 if (IS_ERR(trans)) {
3212 ret = PTR_ERR(trans);
3213 fprintf(stderr, "Error starting transaction: %s\n",
3217 cache = first_cache_extent(corrupt_blocks);
3219 corrupt = container_of(cache, struct btrfs_corrupt_block,
3221 level = corrupt->level;
3222 path->lowest_level = level;
3223 key.objectid = corrupt->key.objectid;
3224 key.type = corrupt->key.type;
3225 key.offset = corrupt->key.offset;
3228 * Here we don't want to do any tree balance, since it may
3229 * cause a balance with corrupted brother leaf/node,
3230 * so ins_len set to 0 here.
3231 * Balance will be done after all corrupt node/leaf is deleted.
3233 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
3236 offset = btrfs_node_blockptr(path->nodes[level],
3237 path->slots[level]);
3239 /* Remove the ptr */
3240 ret = btrfs_del_ptr(trans, root, path, level,
3241 path->slots[level]);
3245 * Remove the corresponding extent
3246 * return value is not concerned.
3248 btrfs_release_path(path);
3249 ret = btrfs_free_extent(trans, root, offset, root->nodesize,
3250 0, root->root_key.objectid,
3252 cache = next_cache_extent(cache);
3255 /* Balance the btree using btrfs_search_slot() */
3256 cache = first_cache_extent(corrupt_blocks);
3258 corrupt = container_of(cache, struct btrfs_corrupt_block,
3260 memcpy(&key, &corrupt->key, sizeof(key));
3261 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3264 /* return will always >0 since it won't find the item */
3266 btrfs_release_path(path);
3267 cache = next_cache_extent(cache);
3270 btrfs_commit_transaction(trans, root);
3272 btrfs_free_path(path);
3276 static int check_fs_root(struct btrfs_root *root,
3277 struct cache_tree *root_cache,
3278 struct walk_control *wc)
3284 struct btrfs_path path;
3285 struct shared_node root_node;
3286 struct root_record *rec;
3287 struct btrfs_root_item *root_item = &root->root_item;
3288 struct cache_tree corrupt_blocks;
3289 struct orphan_data_extent *orphan;
3290 struct orphan_data_extent *tmp;
3291 enum btrfs_tree_block_status status;
3294 * Reuse the corrupt_block cache tree to record corrupted tree block
3296 * Unlike the usage in extent tree check, here we do it in a per
3297 * fs/subvol tree base.
3299 cache_tree_init(&corrupt_blocks);
3300 root->fs_info->corrupt_blocks = &corrupt_blocks;
3302 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
3303 rec = get_root_rec(root_cache, root->root_key.objectid);
3304 if (btrfs_root_refs(root_item) > 0)
3305 rec->found_root_item = 1;
3308 btrfs_init_path(&path);
3309 memset(&root_node, 0, sizeof(root_node));
3310 cache_tree_init(&root_node.root_cache);
3311 cache_tree_init(&root_node.inode_cache);
3313 /* Move the orphan extent record to corresponding inode_record */
3314 list_for_each_entry_safe(orphan, tmp,
3315 &root->orphan_data_extents, list) {
3316 struct inode_record *inode;
3318 inode = get_inode_rec(&root_node.inode_cache, orphan->objectid,
3320 inode->errors |= I_ERR_FILE_EXTENT_ORPHAN;
3321 list_move(&orphan->list, &inode->orphan_extents);
3324 level = btrfs_header_level(root->node);
3325 memset(wc->nodes, 0, sizeof(wc->nodes));
3326 wc->nodes[level] = &root_node;
3327 wc->active_node = level;
3328 wc->root_level = level;
3330 /* We may not have checked the root block, lets do that now */
3331 if (btrfs_is_leaf(root->node))
3332 status = btrfs_check_leaf(root, NULL, root->node);
3334 status = btrfs_check_node(root, NULL, root->node);
3335 if (status != BTRFS_TREE_BLOCK_CLEAN)
3338 if (btrfs_root_refs(root_item) > 0 ||
3339 btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3340 path.nodes[level] = root->node;
3341 extent_buffer_get(root->node);
3342 path.slots[level] = 0;
3344 struct btrfs_key key;
3345 struct btrfs_disk_key found_key;
3347 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3348 level = root_item->drop_level;
3349 path.lowest_level = level;
3350 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
3353 btrfs_node_key(path.nodes[level], &found_key,
3355 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3356 sizeof(found_key)));
3360 wret = walk_down_tree(root, &path, wc, &level);
3366 wret = walk_up_tree(root, &path, wc, &level);
3373 btrfs_release_path(&path);
3375 if (!cache_tree_empty(&corrupt_blocks)) {
3376 struct cache_extent *cache;
3377 struct btrfs_corrupt_block *corrupt;
3379 printf("The following tree block(s) is corrupted in tree %llu:\n",
3380 root->root_key.objectid);
3381 cache = first_cache_extent(&corrupt_blocks);
3383 corrupt = container_of(cache,
3384 struct btrfs_corrupt_block,
3386 printf("\ttree block bytenr: %llu, level: %d, node key: (%llu, %u, %llu)\n",
3387 cache->start, corrupt->level,
3388 corrupt->key.objectid, corrupt->key.type,
3389 corrupt->key.offset);
3390 cache = next_cache_extent(cache);
3393 printf("Try to repair the btree for root %llu\n",
3394 root->root_key.objectid);
3395 ret = repair_btree(root, &corrupt_blocks);
3397 fprintf(stderr, "Failed to repair btree: %s\n",
3400 printf("Btree for root %llu is fixed\n",
3401 root->root_key.objectid);
3405 err = merge_root_recs(root, &root_node.root_cache, root_cache);
3409 if (root_node.current) {
3410 root_node.current->checked = 1;
3411 maybe_free_inode_rec(&root_node.inode_cache,
3415 err = check_inode_recs(root, &root_node.inode_cache);
3419 free_corrupt_blocks_tree(&corrupt_blocks);
3420 root->fs_info->corrupt_blocks = NULL;
3421 free_orphan_data_extents(&root->orphan_data_extents);
3425 static int fs_root_objectid(u64 objectid)
3427 if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
3428 objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3430 return is_fstree(objectid);
3433 static int check_fs_roots(struct btrfs_root *root,
3434 struct cache_tree *root_cache)
3436 struct btrfs_path path;
3437 struct btrfs_key key;
3438 struct walk_control wc;
3439 struct extent_buffer *leaf, *tree_node;
3440 struct btrfs_root *tmp_root;
3441 struct btrfs_root *tree_root = root->fs_info->tree_root;
3446 * Just in case we made any changes to the extent tree that weren't
3447 * reflected into the free space cache yet.
3450 reset_cached_block_groups(root->fs_info);
3451 memset(&wc, 0, sizeof(wc));
3452 cache_tree_init(&wc.shared);
3453 btrfs_init_path(&path);
3458 key.type = BTRFS_ROOT_ITEM_KEY;
3459 ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
3464 tree_node = tree_root->node;
3466 if (tree_node != tree_root->node) {
3467 free_root_recs_tree(root_cache);
3468 btrfs_release_path(&path);
3471 leaf = path.nodes[0];
3472 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
3473 ret = btrfs_next_leaf(tree_root, &path);
3479 leaf = path.nodes[0];
3481 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
3482 if (key.type == BTRFS_ROOT_ITEM_KEY &&
3483 fs_root_objectid(key.objectid)) {
3484 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
3485 tmp_root = btrfs_read_fs_root_no_cache(
3486 root->fs_info, &key);
3488 key.offset = (u64)-1;
3489 tmp_root = btrfs_read_fs_root(
3490 root->fs_info, &key);
3492 if (IS_ERR(tmp_root)) {
3496 ret = check_fs_root(tmp_root, root_cache, &wc);
3497 if (ret == -EAGAIN) {
3498 free_root_recs_tree(root_cache);
3499 btrfs_release_path(&path);
3504 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
3505 btrfs_free_fs_root(tmp_root);
3506 } else if (key.type == BTRFS_ROOT_REF_KEY ||
3507 key.type == BTRFS_ROOT_BACKREF_KEY) {
3508 process_root_ref(leaf, path.slots[0], &key,
3515 btrfs_release_path(&path);
3517 free_extent_cache_tree(&wc.shared);
3518 if (!cache_tree_empty(&wc.shared))
3519 fprintf(stderr, "warning line %d\n", __LINE__);
3524 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
3526 struct list_head *cur = rec->backrefs.next;
3527 struct extent_backref *back;
3528 struct tree_backref *tback;
3529 struct data_backref *dback;
3533 while(cur != &rec->backrefs) {
3534 back = list_entry(cur, struct extent_backref, list);
3536 if (!back->found_extent_tree) {
3540 if (back->is_data) {
3541 dback = (struct data_backref *)back;
3542 fprintf(stderr, "Backref %llu %s %llu"
3543 " owner %llu offset %llu num_refs %lu"
3544 " not found in extent tree\n",
3545 (unsigned long long)rec->start,
3546 back->full_backref ?
3548 back->full_backref ?
3549 (unsigned long long)dback->parent:
3550 (unsigned long long)dback->root,
3551 (unsigned long long)dback->owner,
3552 (unsigned long long)dback->offset,
3553 (unsigned long)dback->num_refs);
3555 tback = (struct tree_backref *)back;
3556 fprintf(stderr, "Backref %llu parent %llu"
3557 " root %llu not found in extent tree\n",
3558 (unsigned long long)rec->start,
3559 (unsigned long long)tback->parent,
3560 (unsigned long long)tback->root);
3563 if (!back->is_data && !back->found_ref) {
3567 tback = (struct tree_backref *)back;
3568 fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
3569 (unsigned long long)rec->start,
3570 back->full_backref ? "parent" : "root",
3571 back->full_backref ?
3572 (unsigned long long)tback->parent :
3573 (unsigned long long)tback->root, back);
3575 if (back->is_data) {
3576 dback = (struct data_backref *)back;
3577 if (dback->found_ref != dback->num_refs) {
3581 fprintf(stderr, "Incorrect local backref count"
3582 " on %llu %s %llu owner %llu"
3583 " offset %llu found %u wanted %u back %p\n",
3584 (unsigned long long)rec->start,
3585 back->full_backref ?
3587 back->full_backref ?
3588 (unsigned long long)dback->parent:
3589 (unsigned long long)dback->root,
3590 (unsigned long long)dback->owner,
3591 (unsigned long long)dback->offset,
3592 dback->found_ref, dback->num_refs, back);
3594 if (dback->disk_bytenr != rec->start) {
3598 fprintf(stderr, "Backref disk bytenr does not"
3599 " match extent record, bytenr=%llu, "
3600 "ref bytenr=%llu\n",
3601 (unsigned long long)rec->start,
3602 (unsigned long long)dback->disk_bytenr);
3605 if (dback->bytes != rec->nr) {
3609 fprintf(stderr, "Backref bytes do not match "
3610 "extent backref, bytenr=%llu, ref "
3611 "bytes=%llu, backref bytes=%llu\n",
3612 (unsigned long long)rec->start,
3613 (unsigned long long)rec->nr,
3614 (unsigned long long)dback->bytes);
3617 if (!back->is_data) {
3620 dback = (struct data_backref *)back;
3621 found += dback->found_ref;
3624 if (found != rec->refs) {
3628 fprintf(stderr, "Incorrect global backref count "
3629 "on %llu found %llu wanted %llu\n",
3630 (unsigned long long)rec->start,
3631 (unsigned long long)found,
3632 (unsigned long long)rec->refs);
3638 static int free_all_extent_backrefs(struct extent_record *rec)
3640 struct extent_backref *back;
3641 struct list_head *cur;
3642 while (!list_empty(&rec->backrefs)) {
3643 cur = rec->backrefs.next;
3644 back = list_entry(cur, struct extent_backref, list);
3651 static void free_extent_record_cache(struct btrfs_fs_info *fs_info,
3652 struct cache_tree *extent_cache)
3654 struct cache_extent *cache;
3655 struct extent_record *rec;
3658 cache = first_cache_extent(extent_cache);
3661 rec = container_of(cache, struct extent_record, cache);
3662 btrfs_unpin_extent(fs_info, rec->start, rec->max_size);
3663 remove_cache_extent(extent_cache, cache);
3664 free_all_extent_backrefs(rec);
3669 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
3670 struct extent_record *rec)
3672 if (rec->content_checked && rec->owner_ref_checked &&
3673 rec->extent_item_refs == rec->refs && rec->refs > 0 &&
3674 rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0)) {
3675 remove_cache_extent(extent_cache, &rec->cache);
3676 free_all_extent_backrefs(rec);
3677 list_del_init(&rec->list);
3683 static int check_owner_ref(struct btrfs_root *root,
3684 struct extent_record *rec,
3685 struct extent_buffer *buf)
3687 struct extent_backref *node;
3688 struct tree_backref *back;
3689 struct btrfs_root *ref_root;
3690 struct btrfs_key key;
3691 struct btrfs_path path;
3692 struct extent_buffer *parent;
3697 list_for_each_entry(node, &rec->backrefs, list) {
3700 if (!node->found_ref)
3702 if (node->full_backref)
3704 back = (struct tree_backref *)node;
3705 if (btrfs_header_owner(buf) == back->root)
3708 BUG_ON(rec->is_root);
3710 /* try to find the block by search corresponding fs tree */
3711 key.objectid = btrfs_header_owner(buf);
3712 key.type = BTRFS_ROOT_ITEM_KEY;
3713 key.offset = (u64)-1;
3715 ref_root = btrfs_read_fs_root(root->fs_info, &key);
3716 if (IS_ERR(ref_root))
3719 level = btrfs_header_level(buf);
3721 btrfs_item_key_to_cpu(buf, &key, 0);
3723 btrfs_node_key_to_cpu(buf, &key, 0);
3725 btrfs_init_path(&path);
3726 path.lowest_level = level + 1;
3727 ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
3731 parent = path.nodes[level + 1];
3732 if (parent && buf->start == btrfs_node_blockptr(parent,
3733 path.slots[level + 1]))
3736 btrfs_release_path(&path);
3737 return found ? 0 : 1;
3740 static int is_extent_tree_record(struct extent_record *rec)
3742 struct list_head *cur = rec->backrefs.next;
3743 struct extent_backref *node;
3744 struct tree_backref *back;
3747 while(cur != &rec->backrefs) {
3748 node = list_entry(cur, struct extent_backref, list);
3752 back = (struct tree_backref *)node;
3753 if (node->full_backref)
3755 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
3762 static int record_bad_block_io(struct btrfs_fs_info *info,
3763 struct cache_tree *extent_cache,
3766 struct extent_record *rec;
3767 struct cache_extent *cache;
3768 struct btrfs_key key;
3770 cache = lookup_cache_extent(extent_cache, start, len);
3774 rec = container_of(cache, struct extent_record, cache);
3775 if (!is_extent_tree_record(rec))
3778 btrfs_disk_key_to_cpu(&key, &rec->parent_key);
3779 return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
3782 static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
3783 struct extent_buffer *buf, int slot)
3785 if (btrfs_header_level(buf)) {
3786 struct btrfs_key_ptr ptr1, ptr2;
3788 read_extent_buffer(buf, &ptr1, btrfs_node_key_ptr_offset(slot),
3789 sizeof(struct btrfs_key_ptr));
3790 read_extent_buffer(buf, &ptr2,
3791 btrfs_node_key_ptr_offset(slot + 1),
3792 sizeof(struct btrfs_key_ptr));
3793 write_extent_buffer(buf, &ptr1,
3794 btrfs_node_key_ptr_offset(slot + 1),
3795 sizeof(struct btrfs_key_ptr));
3796 write_extent_buffer(buf, &ptr2,
3797 btrfs_node_key_ptr_offset(slot),
3798 sizeof(struct btrfs_key_ptr));
3800 struct btrfs_disk_key key;
3801 btrfs_node_key(buf, &key, 0);
3802 btrfs_fixup_low_keys(root, path, &key,
3803 btrfs_header_level(buf) + 1);
3806 struct btrfs_item *item1, *item2;
3807 struct btrfs_key k1, k2;
3808 char *item1_data, *item2_data;
3809 u32 item1_offset, item2_offset, item1_size, item2_size;
3811 item1 = btrfs_item_nr(slot);
3812 item2 = btrfs_item_nr(slot + 1);
3813 btrfs_item_key_to_cpu(buf, &k1, slot);
3814 btrfs_item_key_to_cpu(buf, &k2, slot + 1);
3815 item1_offset = btrfs_item_offset(buf, item1);
3816 item2_offset = btrfs_item_offset(buf, item2);
3817 item1_size = btrfs_item_size(buf, item1);
3818 item2_size = btrfs_item_size(buf, item2);
3820 item1_data = malloc(item1_size);
3823 item2_data = malloc(item2_size);
3829 read_extent_buffer(buf, item1_data, item1_offset, item1_size);
3830 read_extent_buffer(buf, item2_data, item2_offset, item2_size);
3832 write_extent_buffer(buf, item1_data, item2_offset, item2_size);
3833 write_extent_buffer(buf, item2_data, item1_offset, item1_size);
3837 btrfs_set_item_offset(buf, item1, item2_offset);
3838 btrfs_set_item_offset(buf, item2, item1_offset);
3839 btrfs_set_item_size(buf, item1, item2_size);
3840 btrfs_set_item_size(buf, item2, item1_size);
3842 path->slots[0] = slot;
3843 btrfs_set_item_key_unsafe(root, path, &k2);
3844 path->slots[0] = slot + 1;
3845 btrfs_set_item_key_unsafe(root, path, &k1);
3850 static int fix_key_order(struct btrfs_trans_handle *trans,
3851 struct btrfs_root *root,
3852 struct btrfs_path *path)
3854 struct extent_buffer *buf;
3855 struct btrfs_key k1, k2;
3857 int level = path->lowest_level;
3860 buf = path->nodes[level];
3861 for (i = 0; i < btrfs_header_nritems(buf) - 1; i++) {
3863 btrfs_node_key_to_cpu(buf, &k1, i);
3864 btrfs_node_key_to_cpu(buf, &k2, i + 1);
3866 btrfs_item_key_to_cpu(buf, &k1, i);
3867 btrfs_item_key_to_cpu(buf, &k2, i + 1);
3869 if (btrfs_comp_cpu_keys(&k1, &k2) < 0)
3871 ret = swap_values(root, path, buf, i);
3874 btrfs_mark_buffer_dirty(buf);
3880 static int delete_bogus_item(struct btrfs_trans_handle *trans,
3881 struct btrfs_root *root,
3882 struct btrfs_path *path,
3883 struct extent_buffer *buf, int slot)
3885 struct btrfs_key key;
3886 int nritems = btrfs_header_nritems(buf);
3888 btrfs_item_key_to_cpu(buf, &key, slot);
3890 /* These are all the keys we can deal with missing. */
3891 if (key.type != BTRFS_DIR_INDEX_KEY &&
3892 key.type != BTRFS_EXTENT_ITEM_KEY &&
3893 key.type != BTRFS_METADATA_ITEM_KEY &&
3894 key.type != BTRFS_TREE_BLOCK_REF_KEY &&
3895 key.type != BTRFS_EXTENT_DATA_REF_KEY)
3898 printf("Deleting bogus item [%llu,%u,%llu] at slot %d on block %llu\n",
3899 (unsigned long long)key.objectid, key.type,
3900 (unsigned long long)key.offset, slot, buf->start);
3901 memmove_extent_buffer(buf, btrfs_item_nr_offset(slot),
3902 btrfs_item_nr_offset(slot + 1),
3903 sizeof(struct btrfs_item) *
3904 (nritems - slot - 1));
3905 btrfs_set_header_nritems(buf, nritems - 1);
3907 struct btrfs_disk_key disk_key;
3909 btrfs_item_key(buf, &disk_key, 0);
3910 btrfs_fixup_low_keys(root, path, &disk_key, 1);
3912 btrfs_mark_buffer_dirty(buf);
3916 static int fix_item_offset(struct btrfs_trans_handle *trans,
3917 struct btrfs_root *root,
3918 struct btrfs_path *path)
3920 struct extent_buffer *buf;
3924 /* We should only get this for leaves */
3925 BUG_ON(path->lowest_level);
3926 buf = path->nodes[0];
3928 for (i = 0; i < btrfs_header_nritems(buf); i++) {
3929 unsigned int shift = 0, offset;
3931 if (i == 0 && btrfs_item_end_nr(buf, i) !=
3932 BTRFS_LEAF_DATA_SIZE(root)) {
3933 if (btrfs_item_end_nr(buf, i) >
3934 BTRFS_LEAF_DATA_SIZE(root)) {
3935 ret = delete_bogus_item(trans, root, path,
3939 fprintf(stderr, "item is off the end of the "
3940 "leaf, can't fix\n");
3944 shift = BTRFS_LEAF_DATA_SIZE(root) -
3945 btrfs_item_end_nr(buf, i);
3946 } else if (i > 0 && btrfs_item_end_nr(buf, i) !=
3947 btrfs_item_offset_nr(buf, i - 1)) {
3948 if (btrfs_item_end_nr(buf, i) >
3949 btrfs_item_offset_nr(buf, i - 1)) {
3950 ret = delete_bogus_item(trans, root, path,
3954 fprintf(stderr, "items overlap, can't fix\n");
3958 shift = btrfs_item_offset_nr(buf, i - 1) -
3959 btrfs_item_end_nr(buf, i);
3964 printf("Shifting item nr %d by %u bytes in block %llu\n",
3965 i, shift, (unsigned long long)buf->start);
3966 offset = btrfs_item_offset_nr(buf, i);
3967 memmove_extent_buffer(buf,
3968 btrfs_leaf_data(buf) + offset + shift,
3969 btrfs_leaf_data(buf) + offset,
3970 btrfs_item_size_nr(buf, i));
3971 btrfs_set_item_offset(buf, btrfs_item_nr(i),
3973 btrfs_mark_buffer_dirty(buf);
3977 * We may have moved things, in which case we want to exit so we don't
3978 * write those changes out. Once we have proper abort functionality in
3979 * progs this can be changed to something nicer.
3986 * Attempt to fix basic block failures. If we can't fix it for whatever reason
3987 * then just return -EIO.
3989 static int try_to_fix_bad_block(struct btrfs_trans_handle *trans,
3990 struct btrfs_root *root,
3991 struct extent_buffer *buf,
3992 enum btrfs_tree_block_status status)
3994 struct ulist *roots;
3995 struct ulist_node *node;
3996 struct btrfs_root *search_root;
3997 struct btrfs_path *path;
3998 struct ulist_iterator iter;
3999 struct btrfs_key root_key, key;
4002 if (status != BTRFS_TREE_BLOCK_BAD_KEY_ORDER &&
4003 status != BTRFS_TREE_BLOCK_INVALID_OFFSETS)
4006 path = btrfs_alloc_path();
4010 ret = btrfs_find_all_roots(trans, root->fs_info, buf->start,
4013 btrfs_free_path(path);
4017 ULIST_ITER_INIT(&iter);
4018 while ((node = ulist_next(roots, &iter))) {
4019 root_key.objectid = node->val;
4020 root_key.type = BTRFS_ROOT_ITEM_KEY;
4021 root_key.offset = (u64)-1;
4023 search_root = btrfs_read_fs_root(root->fs_info, &root_key);
4029 record_root_in_trans(trans, search_root);
4031 path->lowest_level = btrfs_header_level(buf);
4032 path->skip_check_block = 1;
4033 if (path->lowest_level)
4034 btrfs_node_key_to_cpu(buf, &key, 0);
4036 btrfs_item_key_to_cpu(buf, &key, 0);
4037 ret = btrfs_search_slot(trans, search_root, &key, path, 0, 1);
4042 if (status == BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
4043 ret = fix_key_order(trans, search_root, path);
4044 else if (status == BTRFS_TREE_BLOCK_INVALID_OFFSETS)
4045 ret = fix_item_offset(trans, search_root, path);
4048 btrfs_release_path(path);
4051 btrfs_free_path(path);
4055 static int check_block(struct btrfs_trans_handle *trans,
4056 struct btrfs_root *root,
4057 struct cache_tree *extent_cache,
4058 struct extent_buffer *buf, u64 flags)
4060 struct extent_record *rec;
4061 struct cache_extent *cache;
4062 struct btrfs_key key;
4063 enum btrfs_tree_block_status status;
4067 cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
4070 rec = container_of(cache, struct extent_record, cache);
4071 rec->generation = btrfs_header_generation(buf);
4073 level = btrfs_header_level(buf);
4074 if (btrfs_header_nritems(buf) > 0) {
4077 btrfs_item_key_to_cpu(buf, &key, 0);
4079 btrfs_node_key_to_cpu(buf, &key, 0);
4081 rec->info_objectid = key.objectid;
4083 rec->info_level = level;
4085 if (btrfs_is_leaf(buf))
4086 status = btrfs_check_leaf(root, &rec->parent_key, buf);
4088 status = btrfs_check_node(root, &rec->parent_key, buf);
4090 if (status != BTRFS_TREE_BLOCK_CLEAN) {
4092 status = try_to_fix_bad_block(trans, root, buf,
4094 if (status != BTRFS_TREE_BLOCK_CLEAN) {
4096 fprintf(stderr, "bad block %llu\n",
4097 (unsigned long long)buf->start);
4100 * Signal to callers we need to start the scan over
4101 * again since we'll have cow'ed blocks.
4106 rec->content_checked = 1;
4107 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4108 rec->owner_ref_checked = 1;
4110 ret = check_owner_ref(root, rec, buf);
4112 rec->owner_ref_checked = 1;
4116 maybe_free_extent_rec(extent_cache, rec);
4120 static struct tree_backref *find_tree_backref(struct extent_record *rec,
4121 u64 parent, u64 root)
4123 struct list_head *cur = rec->backrefs.next;
4124 struct extent_backref *node;
4125 struct tree_backref *back;
4127 while(cur != &rec->backrefs) {
4128 node = list_entry(cur, struct extent_backref, list);
4132 back = (struct tree_backref *)node;
4134 if (!node->full_backref)
4136 if (parent == back->parent)
4139 if (node->full_backref)
4141 if (back->root == root)
4148 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
4149 u64 parent, u64 root)
4151 struct tree_backref *ref = malloc(sizeof(*ref));
4152 memset(&ref->node, 0, sizeof(ref->node));
4154 ref->parent = parent;
4155 ref->node.full_backref = 1;
4158 ref->node.full_backref = 0;
4160 list_add_tail(&ref->node.list, &rec->backrefs);
4165 static struct data_backref *find_data_backref(struct extent_record *rec,
4166 u64 parent, u64 root,
4167 u64 owner, u64 offset,
4169 u64 disk_bytenr, u64 bytes)
4171 struct list_head *cur = rec->backrefs.next;
4172 struct extent_backref *node;
4173 struct data_backref *back;
4175 while(cur != &rec->backrefs) {
4176 node = list_entry(cur, struct extent_backref, list);
4180 back = (struct data_backref *)node;
4182 if (!node->full_backref)
4184 if (parent == back->parent)
4187 if (node->full_backref)
4189 if (back->root == root && back->owner == owner &&
4190 back->offset == offset) {
4191 if (found_ref && node->found_ref &&
4192 (back->bytes != bytes ||
4193 back->disk_bytenr != disk_bytenr))
4202 static struct data_backref *alloc_data_backref(struct extent_record *rec,
4203 u64 parent, u64 root,
4204 u64 owner, u64 offset,
4207 struct data_backref *ref = malloc(sizeof(*ref));
4208 memset(&ref->node, 0, sizeof(ref->node));
4209 ref->node.is_data = 1;
4212 ref->parent = parent;
4215 ref->node.full_backref = 1;
4219 ref->offset = offset;
4220 ref->node.full_backref = 0;
4222 ref->bytes = max_size;
4225 list_add_tail(&ref->node.list, &rec->backrefs);
4226 if (max_size > rec->max_size)
4227 rec->max_size = max_size;
4231 static int add_extent_rec(struct cache_tree *extent_cache,
4232 struct btrfs_key *parent_key, u64 parent_gen,
4233 u64 start, u64 nr, u64 extent_item_refs,
4234 int is_root, int inc_ref, int set_checked,
4235 int metadata, int extent_rec, u64 max_size)
4237 struct extent_record *rec;
4238 struct cache_extent *cache;
4242 cache = lookup_cache_extent(extent_cache, start, nr);
4244 rec = container_of(cache, struct extent_record, cache);
4248 rec->nr = max(nr, max_size);
4251 * We need to make sure to reset nr to whatever the extent
4252 * record says was the real size, this way we can compare it to
4256 if (start != rec->start || rec->found_rec) {
4257 struct extent_record *tmp;
4260 if (list_empty(&rec->list))
4261 list_add_tail(&rec->list,
4262 &duplicate_extents);
4265 * We have to do this song and dance in case we
4266 * find an extent record that falls inside of
4267 * our current extent record but does not have
4268 * the same objectid.
4270 tmp = malloc(sizeof(*tmp));
4274 tmp->max_size = max_size;
4277 tmp->metadata = metadata;
4278 tmp->extent_item_refs = extent_item_refs;
4279 INIT_LIST_HEAD(&tmp->list);
4280 list_add_tail(&tmp->list, &rec->dups);
4281 rec->num_duplicates++;
4288 if (extent_item_refs && !dup) {
4289 if (rec->extent_item_refs) {
4290 fprintf(stderr, "block %llu rec "
4291 "extent_item_refs %llu, passed %llu\n",
4292 (unsigned long long)start,
4293 (unsigned long long)
4294 rec->extent_item_refs,
4295 (unsigned long long)extent_item_refs);
4297 rec->extent_item_refs = extent_item_refs;
4302 rec->content_checked = 1;
4303 rec->owner_ref_checked = 1;
4307 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
4309 rec->parent_generation = parent_gen;
4311 if (rec->max_size < max_size)
4312 rec->max_size = max_size;
4314 maybe_free_extent_rec(extent_cache, rec);
4317 rec = malloc(sizeof(*rec));
4319 rec->max_size = max_size;
4320 rec->nr = max(nr, max_size);
4321 rec->found_rec = !!extent_rec;
4322 rec->content_checked = 0;
4323 rec->owner_ref_checked = 0;
4324 rec->num_duplicates = 0;
4325 rec->metadata = metadata;
4326 INIT_LIST_HEAD(&rec->backrefs);
4327 INIT_LIST_HEAD(&rec->dups);
4328 INIT_LIST_HEAD(&rec->list);
4340 if (extent_item_refs)
4341 rec->extent_item_refs = extent_item_refs;
4343 rec->extent_item_refs = 0;
4346 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
4348 memset(&rec->parent_key, 0, sizeof(*parent_key));
4351 rec->parent_generation = parent_gen;
4353 rec->parent_generation = 0;
4355 rec->cache.start = start;
4356 rec->cache.size = nr;
4357 ret = insert_cache_extent(extent_cache, &rec->cache);
4361 rec->content_checked = 1;
4362 rec->owner_ref_checked = 1;
4367 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
4368 u64 parent, u64 root, int found_ref)
4370 struct extent_record *rec;
4371 struct tree_backref *back;
4372 struct cache_extent *cache;
4374 cache = lookup_cache_extent(extent_cache, bytenr, 1);
4376 add_extent_rec(extent_cache, NULL, 0, bytenr,
4377 1, 0, 0, 0, 0, 1, 0, 0);
4378 cache = lookup_cache_extent(extent_cache, bytenr, 1);
4383 rec = container_of(cache, struct extent_record, cache);
4384 if (rec->start != bytenr) {
4388 back = find_tree_backref(rec, parent, root);
4390 back = alloc_tree_backref(rec, parent, root);
4393 if (back->node.found_ref) {
4394 fprintf(stderr, "Extent back ref already exists "
4395 "for %llu parent %llu root %llu \n",
4396 (unsigned long long)bytenr,
4397 (unsigned long long)parent,
4398 (unsigned long long)root);
4400 back->node.found_ref = 1;
4402 if (back->node.found_extent_tree) {
4403 fprintf(stderr, "Extent back ref already exists "
4404 "for %llu parent %llu root %llu \n",
4405 (unsigned long long)bytenr,
4406 (unsigned long long)parent,
4407 (unsigned long long)root);
4409 back->node.found_extent_tree = 1;
4411 maybe_free_extent_rec(extent_cache, rec);
4415 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
4416 u64 parent, u64 root, u64 owner, u64 offset,
4417 u32 num_refs, int found_ref, u64 max_size)
4419 struct extent_record *rec;
4420 struct data_backref *back;
4421 struct cache_extent *cache;
4423 cache = lookup_cache_extent(extent_cache, bytenr, 1);
4425 add_extent_rec(extent_cache, NULL, 0, bytenr, 1, 0, 0, 0, 0,
4427 cache = lookup_cache_extent(extent_cache, bytenr, 1);
4432 rec = container_of(cache, struct extent_record, cache);
4433 if (rec->max_size < max_size)
4434 rec->max_size = max_size;
4437 * If found_ref is set then max_size is the real size and must match the
4438 * existing refs. So if we have already found a ref then we need to
4439 * make sure that this ref matches the existing one, otherwise we need
4440 * to add a new backref so we can notice that the backrefs don't match
4441 * and we need to figure out who is telling the truth. This is to
4442 * account for that awful fsync bug I introduced where we'd end up with
4443 * a btrfs_file_extent_item that would have its length include multiple
4444 * prealloc extents or point inside of a prealloc extent.
4446 back = find_data_backref(rec, parent, root, owner, offset, found_ref,
4449 back = alloc_data_backref(rec, parent, root, owner, offset,
4453 BUG_ON(num_refs != 1);
4454 if (back->node.found_ref)
4455 BUG_ON(back->bytes != max_size);
4456 back->node.found_ref = 1;
4457 back->found_ref += 1;
4458 back->bytes = max_size;
4459 back->disk_bytenr = bytenr;
4461 rec->content_checked = 1;
4462 rec->owner_ref_checked = 1;
4464 if (back->node.found_extent_tree) {
4465 fprintf(stderr, "Extent back ref already exists "
4466 "for %llu parent %llu root %llu "
4467 "owner %llu offset %llu num_refs %lu\n",
4468 (unsigned long long)bytenr,
4469 (unsigned long long)parent,
4470 (unsigned long long)root,
4471 (unsigned long long)owner,
4472 (unsigned long long)offset,
4473 (unsigned long)num_refs);
4475 back->num_refs = num_refs;
4476 back->node.found_extent_tree = 1;
4478 maybe_free_extent_rec(extent_cache, rec);
4482 static int add_pending(struct cache_tree *pending,
4483 struct cache_tree *seen, u64 bytenr, u32 size)
4486 ret = add_cache_extent(seen, bytenr, size);
4489 add_cache_extent(pending, bytenr, size);
4493 static int pick_next_pending(struct cache_tree *pending,
4494 struct cache_tree *reada,
4495 struct cache_tree *nodes,
4496 u64 last, struct block_info *bits, int bits_nr,
4499 unsigned long node_start = last;
4500 struct cache_extent *cache;
4503 cache = search_cache_extent(reada, 0);
4505 bits[0].start = cache->start;
4506 bits[0].size = cache->size;
4511 if (node_start > 32768)
4512 node_start -= 32768;
4514 cache = search_cache_extent(nodes, node_start);
4516 cache = search_cache_extent(nodes, 0);
4519 cache = search_cache_extent(pending, 0);
4524 bits[ret].start = cache->start;
4525 bits[ret].size = cache->size;
4526 cache = next_cache_extent(cache);
4528 } while (cache && ret < bits_nr);
4534 bits[ret].start = cache->start;
4535 bits[ret].size = cache->size;
4536 cache = next_cache_extent(cache);
4538 } while (cache && ret < bits_nr);
4540 if (bits_nr - ret > 8) {
4541 u64 lookup = bits[0].start + bits[0].size;
4542 struct cache_extent *next;
4543 next = search_cache_extent(pending, lookup);
4545 if (next->start - lookup > 32768)
4547 bits[ret].start = next->start;
4548 bits[ret].size = next->size;
4549 lookup = next->start + next->size;
4553 next = next_cache_extent(next);
4561 static void free_chunk_record(struct cache_extent *cache)
4563 struct chunk_record *rec;
4565 rec = container_of(cache, struct chunk_record, cache);
4566 list_del_init(&rec->list);
4567 list_del_init(&rec->dextents);
4571 void free_chunk_cache_tree(struct cache_tree *chunk_cache)
4573 cache_tree_free_extents(chunk_cache, free_chunk_record);
4576 static void free_device_record(struct rb_node *node)
4578 struct device_record *rec;
4580 rec = container_of(node, struct device_record, node);
4584 FREE_RB_BASED_TREE(device_cache, free_device_record);
4586 int insert_block_group_record(struct block_group_tree *tree,
4587 struct block_group_record *bg_rec)
4591 ret = insert_cache_extent(&tree->tree, &bg_rec->cache);
4595 list_add_tail(&bg_rec->list, &tree->block_groups);
4599 static void free_block_group_record(struct cache_extent *cache)
4601 struct block_group_record *rec;
4603 rec = container_of(cache, struct block_group_record, cache);
4604 list_del_init(&rec->list);
4608 void free_block_group_tree(struct block_group_tree *tree)
4610 cache_tree_free_extents(&tree->tree, free_block_group_record);
4613 int insert_device_extent_record(struct device_extent_tree *tree,
4614 struct device_extent_record *de_rec)
4619 * Device extent is a bit different from the other extents, because
4620 * the extents which belong to the different devices may have the
4621 * same start and size, so we need use the special extent cache
4622 * search/insert functions.
4624 ret = insert_cache_extent2(&tree->tree, &de_rec->cache);
4628 list_add_tail(&de_rec->chunk_list, &tree->no_chunk_orphans);
4629 list_add_tail(&de_rec->device_list, &tree->no_device_orphans);
4633 static void free_device_extent_record(struct cache_extent *cache)
4635 struct device_extent_record *rec;
4637 rec = container_of(cache, struct device_extent_record, cache);
4638 if (!list_empty(&rec->chunk_list))
4639 list_del_init(&rec->chunk_list);
4640 if (!list_empty(&rec->device_list))
4641 list_del_init(&rec->device_list);
4645 void free_device_extent_tree(struct device_extent_tree *tree)
4647 cache_tree_free_extents(&tree->tree, free_device_extent_record);
4650 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4651 static int process_extent_ref_v0(struct cache_tree *extent_cache,
4652 struct extent_buffer *leaf, int slot)
4654 struct btrfs_extent_ref_v0 *ref0;
4655 struct btrfs_key key;
4657 btrfs_item_key_to_cpu(leaf, &key, slot);
4658 ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
4659 if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
4660 add_tree_backref(extent_cache, key.objectid, key.offset, 0, 0);
4662 add_data_backref(extent_cache, key.objectid, key.offset, 0,
4663 0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
4669 struct chunk_record *btrfs_new_chunk_record(struct extent_buffer *leaf,
4670 struct btrfs_key *key,
4673 struct btrfs_chunk *ptr;
4674 struct chunk_record *rec;
4677 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4678 num_stripes = btrfs_chunk_num_stripes(leaf, ptr);
4680 rec = malloc(btrfs_chunk_record_size(num_stripes));
4682 fprintf(stderr, "memory allocation failed\n");
4686 memset(rec, 0, btrfs_chunk_record_size(num_stripes));
4688 INIT_LIST_HEAD(&rec->list);
4689 INIT_LIST_HEAD(&rec->dextents);
4692 rec->cache.start = key->offset;
4693 rec->cache.size = btrfs_chunk_length(leaf, ptr);
4695 rec->generation = btrfs_header_generation(leaf);
4697 rec->objectid = key->objectid;
4698 rec->type = key->type;
4699 rec->offset = key->offset;
4701 rec->length = rec->cache.size;
4702 rec->owner = btrfs_chunk_owner(leaf, ptr);
4703 rec->stripe_len = btrfs_chunk_stripe_len(leaf, ptr);
4704 rec->type_flags = btrfs_chunk_type(leaf, ptr);
4705 rec->io_width = btrfs_chunk_io_width(leaf, ptr);
4706 rec->io_align = btrfs_chunk_io_align(leaf, ptr);
4707 rec->sector_size = btrfs_chunk_sector_size(leaf, ptr);
4708 rec->num_stripes = num_stripes;
4709 rec->sub_stripes = btrfs_chunk_sub_stripes(leaf, ptr);
4711 for (i = 0; i < rec->num_stripes; ++i) {
4712 rec->stripes[i].devid =
4713 btrfs_stripe_devid_nr(leaf, ptr, i);
4714 rec->stripes[i].offset =
4715 btrfs_stripe_offset_nr(leaf, ptr, i);
4716 read_extent_buffer(leaf, rec->stripes[i].dev_uuid,
4717 (unsigned long)btrfs_stripe_dev_uuid_nr(ptr, i),
4724 static int process_chunk_item(struct cache_tree *chunk_cache,
4725 struct btrfs_key *key, struct extent_buffer *eb,
4728 struct chunk_record *rec;
4731 rec = btrfs_new_chunk_record(eb, key, slot);
4732 ret = insert_cache_extent(chunk_cache, &rec->cache);
4734 fprintf(stderr, "Chunk[%llu, %llu] existed.\n",
4735 rec->offset, rec->length);
4742 static int process_device_item(struct rb_root *dev_cache,
4743 struct btrfs_key *key, struct extent_buffer *eb, int slot)
4745 struct btrfs_dev_item *ptr;
4746 struct device_record *rec;
4749 ptr = btrfs_item_ptr(eb,
4750 slot, struct btrfs_dev_item);
4752 rec = malloc(sizeof(*rec));
4754 fprintf(stderr, "memory allocation failed\n");
4758 rec->devid = key->offset;
4759 rec->generation = btrfs_header_generation(eb);
4761 rec->objectid = key->objectid;
4762 rec->type = key->type;
4763 rec->offset = key->offset;
4765 rec->devid = btrfs_device_id(eb, ptr);
4766 rec->total_byte = btrfs_device_total_bytes(eb, ptr);
4767 rec->byte_used = btrfs_device_bytes_used(eb, ptr);
4769 ret = rb_insert(dev_cache, &rec->node, device_record_compare);
4771 fprintf(stderr, "Device[%llu] existed.\n", rec->devid);
4778 struct block_group_record *
4779 btrfs_new_block_group_record(struct extent_buffer *leaf, struct btrfs_key *key,
4782 struct btrfs_block_group_item *ptr;
4783 struct block_group_record *rec;
4785 rec = malloc(sizeof(*rec));
4787 fprintf(stderr, "memory allocation failed\n");
4790 memset(rec, 0, sizeof(*rec));
4792 rec->cache.start = key->objectid;
4793 rec->cache.size = key->offset;
4795 rec->generation = btrfs_header_generation(leaf);
4797 rec->objectid = key->objectid;
4798 rec->type = key->type;
4799 rec->offset = key->offset;
4801 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item);
4802 rec->flags = btrfs_disk_block_group_flags(leaf, ptr);
4804 INIT_LIST_HEAD(&rec->list);
4809 static int process_block_group_item(struct block_group_tree *block_group_cache,
4810 struct btrfs_key *key,
4811 struct extent_buffer *eb, int slot)
4813 struct block_group_record *rec;
4816 rec = btrfs_new_block_group_record(eb, key, slot);
4817 ret = insert_block_group_record(block_group_cache, rec);
4819 fprintf(stderr, "Block Group[%llu, %llu] existed.\n",
4820 rec->objectid, rec->offset);
4827 struct device_extent_record *
4828 btrfs_new_device_extent_record(struct extent_buffer *leaf,
4829 struct btrfs_key *key, int slot)
4831 struct device_extent_record *rec;
4832 struct btrfs_dev_extent *ptr;
4834 rec = malloc(sizeof(*rec));
4836 fprintf(stderr, "memory allocation failed\n");
4839 memset(rec, 0, sizeof(*rec));
4841 rec->cache.objectid = key->objectid;
4842 rec->cache.start = key->offset;
4844 rec->generation = btrfs_header_generation(leaf);
4846 rec->objectid = key->objectid;
4847 rec->type = key->type;
4848 rec->offset = key->offset;
4850 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
4851 rec->chunk_objecteid =
4852 btrfs_dev_extent_chunk_objectid(leaf, ptr);
4854 btrfs_dev_extent_chunk_offset(leaf, ptr);
4855 rec->length = btrfs_dev_extent_length(leaf, ptr);
4856 rec->cache.size = rec->length;
4858 INIT_LIST_HEAD(&rec->chunk_list);
4859 INIT_LIST_HEAD(&rec->device_list);
4865 process_device_extent_item(struct device_extent_tree *dev_extent_cache,
4866 struct btrfs_key *key, struct extent_buffer *eb,
4869 struct device_extent_record *rec;
4872 rec = btrfs_new_device_extent_record(eb, key, slot);
4873 ret = insert_device_extent_record(dev_extent_cache, rec);
4876 "Device extent[%llu, %llu, %llu] existed.\n",
4877 rec->objectid, rec->offset, rec->length);
4884 static int process_extent_item(struct btrfs_root *root,
4885 struct cache_tree *extent_cache,
4886 struct extent_buffer *eb, int slot)
4888 struct btrfs_extent_item *ei;
4889 struct btrfs_extent_inline_ref *iref;
4890 struct btrfs_extent_data_ref *dref;
4891 struct btrfs_shared_data_ref *sref;
4892 struct btrfs_key key;
4896 u32 item_size = btrfs_item_size_nr(eb, slot);
4902 btrfs_item_key_to_cpu(eb, &key, slot);
4904 if (key.type == BTRFS_METADATA_ITEM_KEY) {
4906 num_bytes = root->leafsize;
4908 num_bytes = key.offset;
4911 if (item_size < sizeof(*ei)) {
4912 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4913 struct btrfs_extent_item_v0 *ei0;
4914 BUG_ON(item_size != sizeof(*ei0));
4915 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
4916 refs = btrfs_extent_refs_v0(eb, ei0);
4920 return add_extent_rec(extent_cache, NULL, 0, key.objectid,
4921 num_bytes, refs, 0, 0, 0, metadata, 1,
4925 ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
4926 refs = btrfs_extent_refs(eb, ei);
4928 add_extent_rec(extent_cache, NULL, 0, key.objectid, num_bytes,
4929 refs, 0, 0, 0, metadata, 1, num_bytes);
4931 ptr = (unsigned long)(ei + 1);
4932 if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
4933 key.type == BTRFS_EXTENT_ITEM_KEY)
4934 ptr += sizeof(struct btrfs_tree_block_info);
4936 end = (unsigned long)ei + item_size;
4938 iref = (struct btrfs_extent_inline_ref *)ptr;
4939 type = btrfs_extent_inline_ref_type(eb, iref);
4940 offset = btrfs_extent_inline_ref_offset(eb, iref);
4942 case BTRFS_TREE_BLOCK_REF_KEY:
4943 add_tree_backref(extent_cache, key.objectid,
4946 case BTRFS_SHARED_BLOCK_REF_KEY:
4947 add_tree_backref(extent_cache, key.objectid,
4950 case BTRFS_EXTENT_DATA_REF_KEY:
4951 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
4952 add_data_backref(extent_cache, key.objectid, 0,
4953 btrfs_extent_data_ref_root(eb, dref),
4954 btrfs_extent_data_ref_objectid(eb,
4956 btrfs_extent_data_ref_offset(eb, dref),
4957 btrfs_extent_data_ref_count(eb, dref),
4960 case BTRFS_SHARED_DATA_REF_KEY:
4961 sref = (struct btrfs_shared_data_ref *)(iref + 1);
4962 add_data_backref(extent_cache, key.objectid, offset,
4964 btrfs_shared_data_ref_count(eb, sref),
4968 fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
4969 key.objectid, key.type, num_bytes);
4972 ptr += btrfs_extent_inline_ref_size(type);
4979 static int check_cache_range(struct btrfs_root *root,
4980 struct btrfs_block_group_cache *cache,
4981 u64 offset, u64 bytes)
4983 struct btrfs_free_space *entry;
4989 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
4990 bytenr = btrfs_sb_offset(i);
4991 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
4992 cache->key.objectid, bytenr, 0,
4993 &logical, &nr, &stripe_len);
4998 if (logical[nr] + stripe_len <= offset)
5000 if (offset + bytes <= logical[nr])
5002 if (logical[nr] == offset) {
5003 if (stripe_len >= bytes) {
5007 bytes -= stripe_len;
5008 offset += stripe_len;
5009 } else if (logical[nr] < offset) {
5010 if (logical[nr] + stripe_len >=
5015 bytes = (offset + bytes) -
5016 (logical[nr] + stripe_len);
5017 offset = logical[nr] + stripe_len;
5020 * Could be tricky, the super may land in the
5021 * middle of the area we're checking. First
5022 * check the easiest case, it's at the end.
5024 if (logical[nr] + stripe_len >=
5026 bytes = logical[nr] - offset;
5030 /* Check the left side */
5031 ret = check_cache_range(root, cache,
5033 logical[nr] - offset);
5039 /* Now we continue with the right side */
5040 bytes = (offset + bytes) -
5041 (logical[nr] + stripe_len);
5042 offset = logical[nr] + stripe_len;
5049 entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
5051 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
5052 offset, offset+bytes);
5056 if (entry->offset != offset) {
5057 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
5062 if (entry->bytes != bytes) {
5063 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
5064 bytes, entry->bytes, offset);
5068 unlink_free_space(cache->free_space_ctl, entry);
5073 static int verify_space_cache(struct btrfs_root *root,
5074 struct btrfs_block_group_cache *cache)
5076 struct btrfs_path *path;
5077 struct extent_buffer *leaf;
5078 struct btrfs_key key;
5082 path = btrfs_alloc_path();
5086 root = root->fs_info->extent_root;
5088 last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
5090 key.objectid = last;
5092 key.type = BTRFS_EXTENT_ITEM_KEY;
5094 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5099 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5100 ret = btrfs_next_leaf(root, path);
5108 leaf = path->nodes[0];
5109 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5110 if (key.objectid >= cache->key.offset + cache->key.objectid)
5112 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
5113 key.type != BTRFS_METADATA_ITEM_KEY) {
5118 if (last == key.objectid) {
5119 if (key.type == BTRFS_EXTENT_ITEM_KEY)
5120 last = key.objectid + key.offset;
5122 last = key.objectid + root->leafsize;
5127 ret = check_cache_range(root, cache, last,
5128 key.objectid - last);
5131 if (key.type == BTRFS_EXTENT_ITEM_KEY)
5132 last = key.objectid + key.offset;
5134 last = key.objectid + root->leafsize;
5138 if (last < cache->key.objectid + cache->key.offset)
5139 ret = check_cache_range(root, cache, last,
5140 cache->key.objectid +
5141 cache->key.offset - last);
5144 btrfs_free_path(path);
5147 !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
5148 fprintf(stderr, "There are still entries left in the space "
5156 static int check_space_cache(struct btrfs_root *root)
5158 struct btrfs_block_group_cache *cache;
5159 u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
5163 if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
5164 btrfs_super_generation(root->fs_info->super_copy) !=
5165 btrfs_super_cache_generation(root->fs_info->super_copy)) {
5166 printf("cache and super generation don't match, space cache "
5167 "will be invalidated\n");
5172 cache = btrfs_lookup_first_block_group(root->fs_info, start);
5176 start = cache->key.objectid + cache->key.offset;
5177 if (!cache->free_space_ctl) {
5178 if (btrfs_init_free_space_ctl(cache,
5179 root->sectorsize)) {
5184 btrfs_remove_free_space_cache(cache);
5187 ret = load_free_space_cache(root->fs_info, cache);
5191 ret = verify_space_cache(root, cache);
5193 fprintf(stderr, "cache appears valid but isnt %Lu\n",
5194 cache->key.objectid);
5199 return error ? -EINVAL : 0;
5202 static int read_extent_data(struct btrfs_root *root, char *data,
5203 u64 logical, u64 *len, int mirror)
5206 struct btrfs_multi_bio *multi = NULL;
5207 struct btrfs_fs_info *info = root->fs_info;
5208 struct btrfs_device *device;
5212 ret = btrfs_map_block(&info->mapping_tree, READ, logical, len,
5213 &multi, mirror, NULL);
5215 fprintf(stderr, "Couldn't map the block %llu\n",
5219 device = multi->stripes[0].dev;
5221 if (device->fd == 0)
5226 ret = pread64(device->fd, data, *len, multi->stripes[0].physical);
5236 static int check_extent_csums(struct btrfs_root *root, u64 bytenr,
5237 u64 num_bytes, unsigned long leaf_offset,
5238 struct extent_buffer *eb) {
5241 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
5243 unsigned long csum_offset;
5247 u64 data_checked = 0;
5253 if (num_bytes % root->sectorsize)
5256 data = malloc(num_bytes);
5260 while (offset < num_bytes) {
5263 read_len = num_bytes - offset;
5264 /* read as much space once a time */
5265 ret = read_extent_data(root, data + offset,
5266 bytenr + offset, &read_len, mirror);
5270 /* verify every 4k data's checksum */
5271 while (data_checked < read_len) {
5273 tmp = offset + data_checked;
5275 csum = btrfs_csum_data(NULL, (char *)data + tmp,
5276 csum, root->sectorsize);
5277 btrfs_csum_final(csum, (char *)&csum);
5279 csum_offset = leaf_offset +
5280 tmp / root->sectorsize * csum_size;
5281 read_extent_buffer(eb, (char *)&csum_expected,
5282 csum_offset, csum_size);
5283 /* try another mirror */
5284 if (csum != csum_expected) {
5285 fprintf(stderr, "mirror %d bytenr %llu csum %u expected csum %u\n",
5286 mirror, bytenr + tmp,
5287 csum, csum_expected);
5288 num_copies = btrfs_num_copies(
5289 &root->fs_info->mapping_tree,
5291 if (mirror < num_copies - 1) {
5296 data_checked += root->sectorsize;
5305 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
5308 struct btrfs_path *path;
5309 struct extent_buffer *leaf;
5310 struct btrfs_key key;
5313 path = btrfs_alloc_path();
5315 fprintf(stderr, "Error allocing path\n");
5319 key.objectid = bytenr;
5320 key.type = BTRFS_EXTENT_ITEM_KEY;
5321 key.offset = (u64)-1;
5324 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
5327 fprintf(stderr, "Error looking up extent record %d\n", ret);
5328 btrfs_free_path(path);
5331 if (path->slots[0] > 0) {
5334 ret = btrfs_prev_leaf(root, path);
5337 } else if (ret > 0) {
5344 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5347 * Block group items come before extent items if they have the same
5348 * bytenr, so walk back one more just in case. Dear future traveler,
5349 * first congrats on mastering time travel. Now if it's not too much
5350 * trouble could you go back to 2006 and tell Chris to make the
5351 * BLOCK_GROUP_ITEM_KEY (and BTRFS_*_REF_KEY) lower than the
5352 * EXTENT_ITEM_KEY please?
5354 while (key.type > BTRFS_EXTENT_ITEM_KEY) {
5355 if (path->slots[0] > 0) {
5358 ret = btrfs_prev_leaf(root, path);
5361 } else if (ret > 0) {
5366 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5370 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5371 ret = btrfs_next_leaf(root, path);
5373 fprintf(stderr, "Error going to next leaf "
5375 btrfs_free_path(path);
5381 leaf = path->nodes[0];
5382 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5383 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
5387 if (key.objectid + key.offset < bytenr) {
5391 if (key.objectid > bytenr + num_bytes)
5394 if (key.objectid == bytenr) {
5395 if (key.offset >= num_bytes) {
5399 num_bytes -= key.offset;
5400 bytenr += key.offset;
5401 } else if (key.objectid < bytenr) {
5402 if (key.objectid + key.offset >= bytenr + num_bytes) {
5406 num_bytes = (bytenr + num_bytes) -
5407 (key.objectid + key.offset);
5408 bytenr = key.objectid + key.offset;
5410 if (key.objectid + key.offset < bytenr + num_bytes) {
5411 u64 new_start = key.objectid + key.offset;
5412 u64 new_bytes = bytenr + num_bytes - new_start;
5415 * Weird case, the extent is in the middle of
5416 * our range, we'll have to search one side
5417 * and then the other. Not sure if this happens
5418 * in real life, but no harm in coding it up
5419 * anyway just in case.
5421 btrfs_release_path(path);
5422 ret = check_extent_exists(root, new_start,
5425 fprintf(stderr, "Right section didn't "
5429 num_bytes = key.objectid - bytenr;
5432 num_bytes = key.objectid - bytenr;
5439 if (num_bytes && !ret) {
5440 fprintf(stderr, "There are no extents for csum range "
5441 "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
5445 btrfs_free_path(path);
5449 static int check_csums(struct btrfs_root *root)
5451 struct btrfs_path *path;
5452 struct extent_buffer *leaf;
5453 struct btrfs_key key;
5454 u64 offset = 0, num_bytes = 0;
5455 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
5459 unsigned long leaf_offset;
5461 root = root->fs_info->csum_root;
5462 if (!extent_buffer_uptodate(root->node)) {
5463 fprintf(stderr, "No valid csum tree found\n");
5467 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
5468 key.type = BTRFS_EXTENT_CSUM_KEY;
5471 path = btrfs_alloc_path();
5475 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5477 fprintf(stderr, "Error searching csum tree %d\n", ret);
5478 btrfs_free_path(path);
5482 if (ret > 0 && path->slots[0])
5487 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5488 ret = btrfs_next_leaf(root, path);
5490 fprintf(stderr, "Error going to next leaf "
5497 leaf = path->nodes[0];
5499 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5500 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
5505 data_len = (btrfs_item_size_nr(leaf, path->slots[0]) /
5506 csum_size) * root->sectorsize;
5507 if (!check_data_csum)
5508 goto skip_csum_check;
5509 leaf_offset = btrfs_item_ptr_offset(leaf, path->slots[0]);
5510 ret = check_extent_csums(root, key.offset, data_len,
5516 offset = key.offset;
5517 } else if (key.offset != offset + num_bytes) {
5518 ret = check_extent_exists(root, offset, num_bytes);
5520 fprintf(stderr, "Csum exists for %Lu-%Lu but "
5521 "there is no extent record\n",
5522 offset, offset+num_bytes);
5525 offset = key.offset;
5528 num_bytes += data_len;
5532 btrfs_free_path(path);
5536 static int is_dropped_key(struct btrfs_key *key,
5537 struct btrfs_key *drop_key) {
5538 if (key->objectid < drop_key->objectid)
5540 else if (key->objectid == drop_key->objectid) {
5541 if (key->type < drop_key->type)
5543 else if (key->type == drop_key->type) {
5544 if (key->offset < drop_key->offset)
5551 static int calc_extent_flag(struct btrfs_root *root,
5552 struct cache_tree *extent_cache,
5553 struct extent_buffer *buf,
5554 struct root_item_record *ri,
5558 int nritems = btrfs_header_nritems(buf);
5559 struct btrfs_key key;
5560 struct extent_record *rec;
5561 struct cache_extent *cache;
5562 struct data_backref *dback;
5563 struct tree_backref *tback;
5564 struct extent_buffer *new_buf;
5574 * Except file/reloc tree, we can not have
5577 if (ri->objectid < BTRFS_FIRST_FREE_OBJECTID)
5582 if (buf->start == ri->bytenr)
5584 if (btrfs_is_leaf(buf)) {
5586 * we are searching from original root, world
5587 * peace is achieved, we use normal backref.
5589 owner = btrfs_header_owner(buf);
5590 if (owner == ri->objectid)
5593 * we check every eb here, and if any of
5594 * eb dosen't have original root refers
5595 * to this eb, we set full backref flag for
5596 * this extent, otherwise normal backref.
5598 for (i = 0; i < nritems; i++) {
5599 struct btrfs_file_extent_item *fi;
5600 btrfs_item_key_to_cpu(buf, &key, i);
5602 if (key.type != BTRFS_EXTENT_DATA_KEY)
5604 fi = btrfs_item_ptr(buf, i,
5605 struct btrfs_file_extent_item);
5606 if (btrfs_file_extent_type(buf, fi) ==
5607 BTRFS_FILE_EXTENT_INLINE)
5609 if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
5611 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
5612 cache = lookup_cache_extent(extent_cache, bytenr, 1);
5615 offset = btrfs_file_extent_offset(buf, fi);
5616 rec = container_of(cache, struct extent_record, cache);
5617 dback = find_data_backref(rec, 0, ri->objectid, owner,
5618 key.offset - offset, 1, bytenr, bytenr);
5624 level = btrfs_header_level(buf);
5625 for (i = 0; i < nritems; i++) {
5626 ptr = btrfs_node_blockptr(buf, i);
5627 size = btrfs_level_size(root, level);
5629 new_buf = read_tree_block(root, ptr, size, 0);
5630 if (!extent_buffer_uptodate(new_buf)) {
5631 free_extent_buffer(new_buf);
5636 * we are searching from origin root, world
5637 * peace is achieved, we use normal backref.
5639 owner = btrfs_header_owner(new_buf);
5640 free_extent_buffer(new_buf);
5641 if (owner == ri->objectid)
5644 cache = lookup_cache_extent(extent_cache, ptr, size);
5647 rec = container_of(cache, struct extent_record, cache);
5648 tback = find_tree_backref(rec, 0, owner);
5656 cache = lookup_cache_extent(extent_cache, buf->start, 1);
5657 /* we have added this extent before */
5659 rec = container_of(cache, struct extent_record, cache);
5660 rec->flag_block_full_backref = 0;
5663 *flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
5664 cache = lookup_cache_extent(extent_cache, buf->start, 1);
5665 /* we have added this extent before */
5667 rec = container_of(cache, struct extent_record, cache);
5668 rec->flag_block_full_backref = 1;
5672 static int run_next_block(struct btrfs_trans_handle *trans,
5673 struct btrfs_root *root,
5674 struct block_info *bits,
5677 struct cache_tree *pending,
5678 struct cache_tree *seen,
5679 struct cache_tree *reada,
5680 struct cache_tree *nodes,
5681 struct cache_tree *extent_cache,
5682 struct cache_tree *chunk_cache,
5683 struct rb_root *dev_cache,
5684 struct block_group_tree *block_group_cache,
5685 struct device_extent_tree *dev_extent_cache,
5686 struct root_item_record *ri)
5688 struct extent_buffer *buf;
5699 struct btrfs_key key;
5700 struct cache_extent *cache;
5703 nritems = pick_next_pending(pending, reada, nodes, *last, bits,
5704 bits_nr, &reada_bits);
5709 for(i = 0; i < nritems; i++) {
5710 ret = add_cache_extent(reada, bits[i].start,
5715 /* fixme, get the parent transid */
5716 readahead_tree_block(root, bits[i].start,
5720 *last = bits[0].start;
5721 bytenr = bits[0].start;
5722 size = bits[0].size;
5724 cache = lookup_cache_extent(pending, bytenr, size);
5726 remove_cache_extent(pending, cache);
5729 cache = lookup_cache_extent(reada, bytenr, size);
5731 remove_cache_extent(reada, cache);
5734 cache = lookup_cache_extent(nodes, bytenr, size);
5736 remove_cache_extent(nodes, cache);
5739 cache = lookup_cache_extent(extent_cache, bytenr, size);
5741 struct extent_record *rec;
5743 rec = container_of(cache, struct extent_record, cache);
5744 gen = rec->parent_generation;
5747 /* fixme, get the real parent transid */
5748 buf = read_tree_block(root, bytenr, size, gen);
5749 if (!extent_buffer_uptodate(buf)) {
5750 record_bad_block_io(root->fs_info,
5751 extent_cache, bytenr, size);
5755 nritems = btrfs_header_nritems(buf);
5758 * FIXME, this only works only if we don't have any full
5761 if (!init_extent_tree) {
5762 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
5763 btrfs_header_level(buf), 1, NULL,
5769 ret = calc_extent_flag(root, extent_cache, buf, ri, &flags);
5774 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
5779 owner = btrfs_header_owner(buf);
5782 ret = check_block(trans, root, extent_cache, buf, flags);
5786 if (btrfs_is_leaf(buf)) {
5787 btree_space_waste += btrfs_leaf_free_space(root, buf);
5788 for (i = 0; i < nritems; i++) {
5789 struct btrfs_file_extent_item *fi;
5790 btrfs_item_key_to_cpu(buf, &key, i);
5791 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
5792 process_extent_item(root, extent_cache, buf,
5796 if (key.type == BTRFS_METADATA_ITEM_KEY) {
5797 process_extent_item(root, extent_cache, buf,
5801 if (key.type == BTRFS_EXTENT_CSUM_KEY) {
5803 btrfs_item_size_nr(buf, i);
5806 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5807 process_chunk_item(chunk_cache, &key, buf, i);
5810 if (key.type == BTRFS_DEV_ITEM_KEY) {
5811 process_device_item(dev_cache, &key, buf, i);
5814 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
5815 process_block_group_item(block_group_cache,
5819 if (key.type == BTRFS_DEV_EXTENT_KEY) {
5820 process_device_extent_item(dev_extent_cache,
5825 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
5826 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5827 process_extent_ref_v0(extent_cache, buf, i);
5834 if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
5835 add_tree_backref(extent_cache, key.objectid, 0,
5839 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
5840 add_tree_backref(extent_cache, key.objectid,
5844 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
5845 struct btrfs_extent_data_ref *ref;
5846 ref = btrfs_item_ptr(buf, i,
5847 struct btrfs_extent_data_ref);
5848 add_data_backref(extent_cache,
5850 btrfs_extent_data_ref_root(buf, ref),
5851 btrfs_extent_data_ref_objectid(buf,
5853 btrfs_extent_data_ref_offset(buf, ref),
5854 btrfs_extent_data_ref_count(buf, ref),
5855 0, root->sectorsize);
5858 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
5859 struct btrfs_shared_data_ref *ref;
5860 ref = btrfs_item_ptr(buf, i,
5861 struct btrfs_shared_data_ref);
5862 add_data_backref(extent_cache,
5863 key.objectid, key.offset, 0, 0, 0,
5864 btrfs_shared_data_ref_count(buf, ref),
5865 0, root->sectorsize);
5868 if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
5869 struct bad_item *bad;
5871 if (key.objectid == BTRFS_ORPHAN_OBJECTID)
5875 bad = malloc(sizeof(struct bad_item));
5878 INIT_LIST_HEAD(&bad->list);
5879 memcpy(&bad->key, &key,
5880 sizeof(struct btrfs_key));
5881 bad->root_id = owner;
5882 list_add_tail(&bad->list, &delete_items);
5885 if (key.type != BTRFS_EXTENT_DATA_KEY)
5887 fi = btrfs_item_ptr(buf, i,
5888 struct btrfs_file_extent_item);
5889 if (btrfs_file_extent_type(buf, fi) ==
5890 BTRFS_FILE_EXTENT_INLINE)
5892 if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
5895 data_bytes_allocated +=
5896 btrfs_file_extent_disk_num_bytes(buf, fi);
5897 if (data_bytes_allocated < root->sectorsize) {
5900 data_bytes_referenced +=
5901 btrfs_file_extent_num_bytes(buf, fi);
5902 add_data_backref(extent_cache,
5903 btrfs_file_extent_disk_bytenr(buf, fi),
5904 parent, owner, key.objectid, key.offset -
5905 btrfs_file_extent_offset(buf, fi), 1, 1,
5906 btrfs_file_extent_disk_num_bytes(buf, fi));
5910 struct btrfs_key first_key;
5912 first_key.objectid = 0;
5915 btrfs_item_key_to_cpu(buf, &first_key, 0);
5916 level = btrfs_header_level(buf);
5917 for (i = 0; i < nritems; i++) {
5918 ptr = btrfs_node_blockptr(buf, i);
5919 size = btrfs_level_size(root, level - 1);
5920 btrfs_node_key_to_cpu(buf, &key, i);
5922 if ((level == ri->drop_level)
5923 && is_dropped_key(&key, &ri->drop_key)) {
5927 ret = add_extent_rec(extent_cache, &key,
5928 btrfs_node_ptr_generation(buf, i),
5929 ptr, size, 0, 0, 1, 0, 1, 0,
5933 add_tree_backref(extent_cache, ptr, parent, owner, 1);
5936 add_pending(nodes, seen, ptr, size);
5938 add_pending(pending, seen, ptr, size);
5941 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
5942 nritems) * sizeof(struct btrfs_key_ptr);
5944 total_btree_bytes += buf->len;
5945 if (fs_root_objectid(btrfs_header_owner(buf)))
5946 total_fs_tree_bytes += buf->len;
5947 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
5948 total_extent_tree_bytes += buf->len;
5949 if (!found_old_backref &&
5950 btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
5951 btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
5952 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
5953 found_old_backref = 1;
5955 free_extent_buffer(buf);
5959 static int add_root_to_pending(struct extent_buffer *buf,
5960 struct cache_tree *extent_cache,
5961 struct cache_tree *pending,
5962 struct cache_tree *seen,
5963 struct cache_tree *nodes,
5966 if (btrfs_header_level(buf) > 0)
5967 add_pending(nodes, seen, buf->start, buf->len);
5969 add_pending(pending, seen, buf->start, buf->len);
5970 add_extent_rec(extent_cache, NULL, 0, buf->start, buf->len,
5971 0, 1, 1, 0, 1, 0, buf->len);
5973 if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
5974 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
5975 add_tree_backref(extent_cache, buf->start, buf->start,
5978 add_tree_backref(extent_cache, buf->start, 0, objectid, 1);
5982 /* as we fix the tree, we might be deleting blocks that
5983 * we're tracking for repair. This hook makes sure we
5984 * remove any backrefs for blocks as we are fixing them.
5986 static int free_extent_hook(struct btrfs_trans_handle *trans,
5987 struct btrfs_root *root,
5988 u64 bytenr, u64 num_bytes, u64 parent,
5989 u64 root_objectid, u64 owner, u64 offset,
5992 struct extent_record *rec;
5993 struct cache_extent *cache;
5995 struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
5997 is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
5998 cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
6002 rec = container_of(cache, struct extent_record, cache);
6004 struct data_backref *back;
6005 back = find_data_backref(rec, parent, root_objectid, owner,
6006 offset, 1, bytenr, num_bytes);
6009 if (back->node.found_ref) {
6010 back->found_ref -= refs_to_drop;
6012 rec->refs -= refs_to_drop;
6014 if (back->node.found_extent_tree) {
6015 back->num_refs -= refs_to_drop;
6016 if (rec->extent_item_refs)
6017 rec->extent_item_refs -= refs_to_drop;
6019 if (back->found_ref == 0)
6020 back->node.found_ref = 0;
6021 if (back->num_refs == 0)
6022 back->node.found_extent_tree = 0;
6024 if (!back->node.found_extent_tree && back->node.found_ref) {
6025 list_del(&back->node.list);
6029 struct tree_backref *back;
6030 back = find_tree_backref(rec, parent, root_objectid);
6033 if (back->node.found_ref) {
6036 back->node.found_ref = 0;
6038 if (back->node.found_extent_tree) {
6039 if (rec->extent_item_refs)
6040 rec->extent_item_refs--;
6041 back->node.found_extent_tree = 0;
6043 if (!back->node.found_extent_tree && back->node.found_ref) {
6044 list_del(&back->node.list);
6048 maybe_free_extent_rec(extent_cache, rec);
6053 static int delete_extent_records(struct btrfs_trans_handle *trans,
6054 struct btrfs_root *root,
6055 struct btrfs_path *path,
6056 u64 bytenr, u64 new_len)
6058 struct btrfs_key key;
6059 struct btrfs_key found_key;
6060 struct extent_buffer *leaf;
6065 key.objectid = bytenr;
6067 key.offset = (u64)-1;
6070 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
6077 if (path->slots[0] == 0)
6083 leaf = path->nodes[0];
6084 slot = path->slots[0];
6086 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6087 if (found_key.objectid != bytenr)
6090 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
6091 found_key.type != BTRFS_METADATA_ITEM_KEY &&
6092 found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
6093 found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
6094 found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
6095 found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
6096 found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
6097 btrfs_release_path(path);
6098 if (found_key.type == 0) {
6099 if (found_key.offset == 0)
6101 key.offset = found_key.offset - 1;
6102 key.type = found_key.type;
6104 key.type = found_key.type - 1;
6105 key.offset = (u64)-1;
6109 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
6110 found_key.objectid, found_key.type, found_key.offset);
6112 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
6115 btrfs_release_path(path);
6117 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
6118 found_key.type == BTRFS_METADATA_ITEM_KEY) {
6119 u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
6120 found_key.offset : root->leafsize;
6122 ret = btrfs_update_block_group(trans, root, bytenr,
6129 btrfs_release_path(path);
6134 * for a single backref, this will allocate a new extent
6135 * and add the backref to it.
6137 static int record_extent(struct btrfs_trans_handle *trans,
6138 struct btrfs_fs_info *info,
6139 struct btrfs_path *path,
6140 struct extent_record *rec,
6141 struct extent_backref *back,
6142 int allocated, u64 flags)
6145 struct btrfs_root *extent_root = info->extent_root;
6146 struct extent_buffer *leaf;
6147 struct btrfs_key ins_key;
6148 struct btrfs_extent_item *ei;
6149 struct tree_backref *tback;
6150 struct data_backref *dback;
6151 struct btrfs_tree_block_info *bi;
6154 rec->max_size = max_t(u64, rec->max_size,
6155 info->extent_root->leafsize);
6158 u32 item_size = sizeof(*ei);
6161 item_size += sizeof(*bi);
6163 ins_key.objectid = rec->start;
6164 ins_key.offset = rec->max_size;
6165 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
6167 ret = btrfs_insert_empty_item(trans, extent_root, path,
6168 &ins_key, item_size);
6172 leaf = path->nodes[0];
6173 ei = btrfs_item_ptr(leaf, path->slots[0],
6174 struct btrfs_extent_item);
6176 btrfs_set_extent_refs(leaf, ei, 0);
6177 btrfs_set_extent_generation(leaf, ei, rec->generation);
6179 if (back->is_data) {
6180 btrfs_set_extent_flags(leaf, ei,
6181 BTRFS_EXTENT_FLAG_DATA);
6183 struct btrfs_disk_key copy_key;;
6185 tback = (struct tree_backref *)back;
6186 bi = (struct btrfs_tree_block_info *)(ei + 1);
6187 memset_extent_buffer(leaf, 0, (unsigned long)bi,
6190 btrfs_set_disk_key_objectid(©_key,
6191 rec->info_objectid);
6192 btrfs_set_disk_key_type(©_key, 0);
6193 btrfs_set_disk_key_offset(©_key, 0);
6195 btrfs_set_tree_block_level(leaf, bi, rec->info_level);
6196 btrfs_set_tree_block_key(leaf, bi, ©_key);
6198 btrfs_set_extent_flags(leaf, ei,
6199 BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
6202 btrfs_mark_buffer_dirty(leaf);
6203 ret = btrfs_update_block_group(trans, extent_root, rec->start,
6204 rec->max_size, 1, 0);
6207 btrfs_release_path(path);
6210 if (back->is_data) {
6214 dback = (struct data_backref *)back;
6215 if (back->full_backref)
6216 parent = dback->parent;
6220 for (i = 0; i < dback->found_ref; i++) {
6221 /* if parent != 0, we're doing a full backref
6222 * passing BTRFS_FIRST_FREE_OBJECTID as the owner
6223 * just makes the backref allocator create a data
6226 ret = btrfs_inc_extent_ref(trans, info->extent_root,
6227 rec->start, rec->max_size,
6231 BTRFS_FIRST_FREE_OBJECTID :
6237 fprintf(stderr, "adding new data backref"
6238 " on %llu %s %llu owner %llu"
6239 " offset %llu found %d\n",
6240 (unsigned long long)rec->start,
6241 back->full_backref ?
6243 back->full_backref ?
6244 (unsigned long long)parent :
6245 (unsigned long long)dback->root,
6246 (unsigned long long)dback->owner,
6247 (unsigned long long)dback->offset,
6252 tback = (struct tree_backref *)back;
6253 if (back->full_backref)
6254 parent = tback->parent;
6258 ret = btrfs_inc_extent_ref(trans, info->extent_root,
6259 rec->start, rec->max_size,
6260 parent, tback->root, 0, 0);
6261 fprintf(stderr, "adding new tree backref on "
6262 "start %llu len %llu parent %llu root %llu\n",
6263 rec->start, rec->max_size, tback->parent, tback->root);
6268 btrfs_release_path(path);
6272 struct extent_entry {
6277 struct list_head list;
6280 static struct extent_entry *find_entry(struct list_head *entries,
6281 u64 bytenr, u64 bytes)
6283 struct extent_entry *entry = NULL;
6285 list_for_each_entry(entry, entries, list) {
6286 if (entry->bytenr == bytenr && entry->bytes == bytes)
6293 static struct extent_entry *find_most_right_entry(struct list_head *entries)
6295 struct extent_entry *entry, *best = NULL, *prev = NULL;
6297 list_for_each_entry(entry, entries, list) {
6304 * If there are as many broken entries as entries then we know
6305 * not to trust this particular entry.
6307 if (entry->broken == entry->count)
6311 * If our current entry == best then we can't be sure our best
6312 * is really the best, so we need to keep searching.
6314 if (best && best->count == entry->count) {
6320 /* Prev == entry, not good enough, have to keep searching */
6321 if (!prev->broken && prev->count == entry->count)
6325 best = (prev->count > entry->count) ? prev : entry;
6326 else if (best->count < entry->count)
6334 static int repair_ref(struct btrfs_trans_handle *trans,
6335 struct btrfs_fs_info *info, struct btrfs_path *path,
6336 struct data_backref *dback, struct extent_entry *entry)
6338 struct btrfs_root *root;
6339 struct btrfs_file_extent_item *fi;
6340 struct extent_buffer *leaf;
6341 struct btrfs_key key;
6345 key.objectid = dback->root;
6346 key.type = BTRFS_ROOT_ITEM_KEY;
6347 key.offset = (u64)-1;
6348 root = btrfs_read_fs_root(info, &key);
6350 fprintf(stderr, "Couldn't find root for our ref\n");
6355 * The backref points to the original offset of the extent if it was
6356 * split, so we need to search down to the offset we have and then walk
6357 * forward until we find the backref we're looking for.
6359 key.objectid = dback->owner;
6360 key.type = BTRFS_EXTENT_DATA_KEY;
6361 key.offset = dback->offset;
6362 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6364 fprintf(stderr, "Error looking up ref %d\n", ret);
6369 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
6370 ret = btrfs_next_leaf(root, path);
6372 fprintf(stderr, "Couldn't find our ref, next\n");
6376 leaf = path->nodes[0];
6377 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6378 if (key.objectid != dback->owner ||
6379 key.type != BTRFS_EXTENT_DATA_KEY) {
6380 fprintf(stderr, "Couldn't find our ref, search\n");
6383 fi = btrfs_item_ptr(leaf, path->slots[0],
6384 struct btrfs_file_extent_item);
6385 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6386 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6388 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
6393 btrfs_release_path(path);
6396 * Have to make sure that this root gets updated when we commit the
6399 record_root_in_trans(trans, root);
6402 * Ok we have the key of the file extent we want to fix, now we can cow
6403 * down to the thing and fix it.
6405 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6407 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
6408 key.objectid, key.type, key.offset, ret);
6412 fprintf(stderr, "Well that's odd, we just found this key "
6413 "[%Lu, %u, %Lu]\n", key.objectid, key.type,
6417 leaf = path->nodes[0];
6418 fi = btrfs_item_ptr(leaf, path->slots[0],
6419 struct btrfs_file_extent_item);
6421 if (btrfs_file_extent_compression(leaf, fi) &&
6422 dback->disk_bytenr != entry->bytenr) {
6423 fprintf(stderr, "Ref doesn't match the record start and is "
6424 "compressed, please take a btrfs-image of this file "
6425 "system and send it to a btrfs developer so they can "
6426 "complete this functionality for bytenr %Lu\n",
6427 dback->disk_bytenr);
6431 if (dback->node.broken && dback->disk_bytenr != entry->bytenr) {
6432 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6433 } else if (dback->disk_bytenr > entry->bytenr) {
6434 u64 off_diff, offset;
6436 off_diff = dback->disk_bytenr - entry->bytenr;
6437 offset = btrfs_file_extent_offset(leaf, fi);
6438 if (dback->disk_bytenr + offset +
6439 btrfs_file_extent_num_bytes(leaf, fi) >
6440 entry->bytenr + entry->bytes) {
6441 fprintf(stderr, "Ref is past the entry end, please "
6442 "take a btrfs-image of this file system and "
6443 "send it to a btrfs developer, ref %Lu\n",
6444 dback->disk_bytenr);
6448 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6449 btrfs_set_file_extent_offset(leaf, fi, offset);
6450 } else if (dback->disk_bytenr < entry->bytenr) {
6453 offset = btrfs_file_extent_offset(leaf, fi);
6454 if (dback->disk_bytenr + offset < entry->bytenr) {
6455 fprintf(stderr, "Ref is before the entry start, please"
6456 " take a btrfs-image of this file system and "
6457 "send it to a btrfs developer, ref %Lu\n",
6458 dback->disk_bytenr);
6462 offset += dback->disk_bytenr;
6463 offset -= entry->bytenr;
6464 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6465 btrfs_set_file_extent_offset(leaf, fi, offset);
6468 btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
6471 * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
6472 * only do this if we aren't using compression, otherwise it's a
6475 if (!btrfs_file_extent_compression(leaf, fi))
6476 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
6478 printf("ram bytes may be wrong?\n");
6479 btrfs_mark_buffer_dirty(leaf);
6480 btrfs_release_path(path);
6484 static int verify_backrefs(struct btrfs_trans_handle *trans,
6485 struct btrfs_fs_info *info, struct btrfs_path *path,
6486 struct extent_record *rec)
6488 struct extent_backref *back;
6489 struct data_backref *dback;
6490 struct extent_entry *entry, *best = NULL;
6493 int broken_entries = 0;
6498 * Metadata is easy and the backrefs should always agree on bytenr and
6499 * size, if not we've got bigger issues.
6504 list_for_each_entry(back, &rec->backrefs, list) {
6505 if (back->full_backref || !back->is_data)
6508 dback = (struct data_backref *)back;
6511 * We only pay attention to backrefs that we found a real
6514 if (dback->found_ref == 0)
6518 * For now we only catch when the bytes don't match, not the
6519 * bytenr. We can easily do this at the same time, but I want
6520 * to have a fs image to test on before we just add repair
6521 * functionality willy-nilly so we know we won't screw up the
6525 entry = find_entry(&entries, dback->disk_bytenr,
6528 entry = malloc(sizeof(struct extent_entry));
6533 memset(entry, 0, sizeof(*entry));
6534 entry->bytenr = dback->disk_bytenr;
6535 entry->bytes = dback->bytes;
6536 list_add_tail(&entry->list, &entries);
6541 * If we only have on entry we may think the entries agree when
6542 * in reality they don't so we have to do some extra checking.
6544 if (dback->disk_bytenr != rec->start ||
6545 dback->bytes != rec->nr || back->broken)
6556 /* Yay all the backrefs agree, carry on good sir */
6557 if (nr_entries <= 1 && !mismatch)
6560 fprintf(stderr, "attempting to repair backref discrepency for bytenr "
6561 "%Lu\n", rec->start);
6564 * First we want to see if the backrefs can agree amongst themselves who
6565 * is right, so figure out which one of the entries has the highest
6568 best = find_most_right_entry(&entries);
6571 * Ok so we may have an even split between what the backrefs think, so
6572 * this is where we use the extent ref to see what it thinks.
6575 entry = find_entry(&entries, rec->start, rec->nr);
6576 if (!entry && (!broken_entries || !rec->found_rec)) {
6577 fprintf(stderr, "Backrefs don't agree with each other "
6578 "and extent record doesn't agree with anybody,"
6579 " so we can't fix bytenr %Lu bytes %Lu\n",
6580 rec->start, rec->nr);
6583 } else if (!entry) {
6585 * Ok our backrefs were broken, we'll assume this is the
6586 * correct value and add an entry for this range.
6588 entry = malloc(sizeof(struct extent_entry));
6593 memset(entry, 0, sizeof(*entry));
6594 entry->bytenr = rec->start;
6595 entry->bytes = rec->nr;
6596 list_add_tail(&entry->list, &entries);
6600 best = find_most_right_entry(&entries);
6602 fprintf(stderr, "Backrefs and extent record evenly "
6603 "split on who is right, this is going to "
6604 "require user input to fix bytenr %Lu bytes "
6605 "%Lu\n", rec->start, rec->nr);
6612 * I don't think this can happen currently as we'll abort() if we catch
6613 * this case higher up, but in case somebody removes that we still can't
6614 * deal with it properly here yet, so just bail out of that's the case.
6616 if (best->bytenr != rec->start) {
6617 fprintf(stderr, "Extent start and backref starts don't match, "
6618 "please use btrfs-image on this file system and send "
6619 "it to a btrfs developer so they can make fsck fix "
6620 "this particular case. bytenr is %Lu, bytes is %Lu\n",
6621 rec->start, rec->nr);
6627 * Ok great we all agreed on an extent record, let's go find the real
6628 * references and fix up the ones that don't match.
6630 list_for_each_entry(back, &rec->backrefs, list) {
6631 if (back->full_backref || !back->is_data)
6634 dback = (struct data_backref *)back;
6637 * Still ignoring backrefs that don't have a real ref attached
6640 if (dback->found_ref == 0)
6643 if (dback->bytes == best->bytes &&
6644 dback->disk_bytenr == best->bytenr)
6647 ret = repair_ref(trans, info, path, dback, best);
6653 * Ok we messed with the actual refs, which means we need to drop our
6654 * entire cache and go back and rescan. I know this is a huge pain and
6655 * adds a lot of extra work, but it's the only way to be safe. Once all
6656 * the backrefs agree we may not need to do anything to the extent
6661 while (!list_empty(&entries)) {
6662 entry = list_entry(entries.next, struct extent_entry, list);
6663 list_del_init(&entry->list);
6669 static int process_duplicates(struct btrfs_root *root,
6670 struct cache_tree *extent_cache,
6671 struct extent_record *rec)
6673 struct extent_record *good, *tmp;
6674 struct cache_extent *cache;
6678 * If we found a extent record for this extent then return, or if we
6679 * have more than one duplicate we are likely going to need to delete
6682 if (rec->found_rec || rec->num_duplicates > 1)
6685 /* Shouldn't happen but just in case */
6686 BUG_ON(!rec->num_duplicates);
6689 * So this happens if we end up with a backref that doesn't match the
6690 * actual extent entry. So either the backref is bad or the extent
6691 * entry is bad. Either way we want to have the extent_record actually
6692 * reflect what we found in the extent_tree, so we need to take the
6693 * duplicate out and use that as the extent_record since the only way we
6694 * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
6696 remove_cache_extent(extent_cache, &rec->cache);
6698 good = list_entry(rec->dups.next, struct extent_record, list);
6699 list_del_init(&good->list);
6700 INIT_LIST_HEAD(&good->backrefs);
6701 INIT_LIST_HEAD(&good->dups);
6702 good->cache.start = good->start;
6703 good->cache.size = good->nr;
6704 good->content_checked = 0;
6705 good->owner_ref_checked = 0;
6706 good->num_duplicates = 0;
6707 good->refs = rec->refs;
6708 list_splice_init(&rec->backrefs, &good->backrefs);
6710 cache = lookup_cache_extent(extent_cache, good->start,
6714 tmp = container_of(cache, struct extent_record, cache);
6717 * If we find another overlapping extent and it's found_rec is
6718 * set then it's a duplicate and we need to try and delete
6721 if (tmp->found_rec || tmp->num_duplicates > 0) {
6722 if (list_empty(&good->list))
6723 list_add_tail(&good->list,
6724 &duplicate_extents);
6725 good->num_duplicates += tmp->num_duplicates + 1;
6726 list_splice_init(&tmp->dups, &good->dups);
6727 list_del_init(&tmp->list);
6728 list_add_tail(&tmp->list, &good->dups);
6729 remove_cache_extent(extent_cache, &tmp->cache);
6734 * Ok we have another non extent item backed extent rec, so lets
6735 * just add it to this extent and carry on like we did above.
6737 good->refs += tmp->refs;
6738 list_splice_init(&tmp->backrefs, &good->backrefs);
6739 remove_cache_extent(extent_cache, &tmp->cache);
6742 ret = insert_cache_extent(extent_cache, &good->cache);
6745 return good->num_duplicates ? 0 : 1;
6748 static int delete_duplicate_records(struct btrfs_trans_handle *trans,
6749 struct btrfs_root *root,
6750 struct extent_record *rec)
6752 LIST_HEAD(delete_list);
6753 struct btrfs_path *path;
6754 struct extent_record *tmp, *good, *n;
6757 struct btrfs_key key;
6759 path = btrfs_alloc_path();
6766 /* Find the record that covers all of the duplicates. */
6767 list_for_each_entry(tmp, &rec->dups, list) {
6768 if (good->start < tmp->start)
6770 if (good->nr > tmp->nr)
6773 if (tmp->start + tmp->nr < good->start + good->nr) {
6774 fprintf(stderr, "Ok we have overlapping extents that "
6775 "aren't completely covered by eachother, this "
6776 "is going to require more careful thought. "
6777 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
6778 tmp->start, tmp->nr, good->start, good->nr);
6785 list_add_tail(&rec->list, &delete_list);
6787 list_for_each_entry_safe(tmp, n, &rec->dups, list) {
6790 list_move_tail(&tmp->list, &delete_list);
6793 root = root->fs_info->extent_root;
6794 list_for_each_entry(tmp, &delete_list, list) {
6795 if (tmp->found_rec == 0)
6797 key.objectid = tmp->start;
6798 key.type = BTRFS_EXTENT_ITEM_KEY;
6799 key.offset = tmp->nr;
6801 /* Shouldn't happen but just in case */
6802 if (tmp->metadata) {
6803 fprintf(stderr, "Well this shouldn't happen, extent "
6804 "record overlaps but is metadata? "
6805 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
6809 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6815 ret = btrfs_del_item(trans, root, path);
6818 btrfs_release_path(path);
6823 while (!list_empty(&delete_list)) {
6824 tmp = list_entry(delete_list.next, struct extent_record, list);
6825 list_del_init(&tmp->list);
6831 while (!list_empty(&rec->dups)) {
6832 tmp = list_entry(rec->dups.next, struct extent_record, list);
6833 list_del_init(&tmp->list);
6837 btrfs_free_path(path);
6839 if (!ret && !nr_del)
6840 rec->num_duplicates = 0;
6842 return ret ? ret : nr_del;
6845 static int find_possible_backrefs(struct btrfs_trans_handle *trans,
6846 struct btrfs_fs_info *info,
6847 struct btrfs_path *path,
6848 struct cache_tree *extent_cache,
6849 struct extent_record *rec)
6851 struct btrfs_root *root;
6852 struct extent_backref *back;
6853 struct data_backref *dback;
6854 struct cache_extent *cache;
6855 struct btrfs_file_extent_item *fi;
6856 struct btrfs_key key;
6860 list_for_each_entry(back, &rec->backrefs, list) {
6861 /* Don't care about full backrefs (poor unloved backrefs) */
6862 if (back->full_backref || !back->is_data)
6865 dback = (struct data_backref *)back;
6867 /* We found this one, we don't need to do a lookup */
6868 if (dback->found_ref)
6871 key.objectid = dback->root;
6872 key.type = BTRFS_ROOT_ITEM_KEY;
6873 key.offset = (u64)-1;
6875 root = btrfs_read_fs_root(info, &key);
6877 /* No root, definitely a bad ref, skip */
6878 if (IS_ERR(root) && PTR_ERR(root) == -ENOENT)
6880 /* Other err, exit */
6882 return PTR_ERR(root);
6884 key.objectid = dback->owner;
6885 key.type = BTRFS_EXTENT_DATA_KEY;
6886 key.offset = dback->offset;
6887 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6889 btrfs_release_path(path);
6892 /* Didn't find it, we can carry on */
6897 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6898 struct btrfs_file_extent_item);
6899 bytenr = btrfs_file_extent_disk_bytenr(path->nodes[0], fi);
6900 bytes = btrfs_file_extent_disk_num_bytes(path->nodes[0], fi);
6901 btrfs_release_path(path);
6902 cache = lookup_cache_extent(extent_cache, bytenr, 1);
6904 struct extent_record *tmp;
6905 tmp = container_of(cache, struct extent_record, cache);
6908 * If we found an extent record for the bytenr for this
6909 * particular backref then we can't add it to our
6910 * current extent record. We only want to add backrefs
6911 * that don't have a corresponding extent item in the
6912 * extent tree since they likely belong to this record
6913 * and we need to fix it if it doesn't match bytenrs.
6919 dback->found_ref += 1;
6920 dback->disk_bytenr = bytenr;
6921 dback->bytes = bytes;
6924 * Set this so the verify backref code knows not to trust the
6925 * values in this backref.
6934 * Record orphan data ref into corresponding root.
6936 * Return 0 if the extent item contains data ref and recorded.
6937 * Return 1 if the extent item contains no useful data ref
6938 * On that case, it may contains only shared_dataref or metadata backref
6939 * or the file extent exists(this should be handled by the extent bytenr
6941 * Return <0 if something goes wrong.
6943 static int record_orphan_data_extents(struct btrfs_fs_info *fs_info,
6944 struct extent_record *rec)
6946 struct btrfs_key key;
6947 struct btrfs_root *dest_root;
6948 struct extent_backref *back;
6949 struct data_backref *dback;
6950 struct orphan_data_extent *orphan;
6951 struct btrfs_path *path;
6952 int recorded_data_ref = 0;
6957 path = btrfs_alloc_path();
6960 list_for_each_entry(back, &rec->backrefs, list) {
6961 if (back->full_backref || !back->is_data ||
6962 !back->found_extent_tree)
6964 dback = (struct data_backref *)back;
6965 if (dback->found_ref)
6967 key.objectid = dback->root;
6968 key.type = BTRFS_ROOT_ITEM_KEY;
6969 key.offset = (u64)-1;
6971 dest_root = btrfs_read_fs_root(fs_info, &key);
6973 /* For non-exist root we just skip it */
6974 if (IS_ERR(dest_root) || !dest_root)
6977 key.objectid = dback->owner;
6978 key.type = BTRFS_EXTENT_DATA_KEY;
6979 key.offset = dback->offset;
6981 ret = btrfs_search_slot(NULL, dest_root, &key, path, 0, 0);
6983 * For ret < 0, it's OK since the fs-tree may be corrupted,
6984 * we need to record it for inode/file extent rebuild.
6985 * For ret > 0, we record it only for file extent rebuild.
6986 * For ret == 0, the file extent exists but only bytenr
6987 * mismatch, let the original bytenr fix routine to handle,
6993 orphan = malloc(sizeof(*orphan));
6998 INIT_LIST_HEAD(&orphan->list);
6999 orphan->root = dback->root;
7000 orphan->objectid = dback->owner;
7001 orphan->offset = dback->offset;
7002 orphan->disk_bytenr = rec->cache.start;
7003 orphan->disk_len = rec->cache.size;
7004 list_add(&dest_root->orphan_data_extents, &orphan->list);
7005 recorded_data_ref = 1;
7008 btrfs_free_path(path);
7010 return !recorded_data_ref;
7016 * when an incorrect extent item is found, this will delete
7017 * all of the existing entries for it and recreate them
7018 * based on what the tree scan found.
7020 static int fixup_extent_refs(struct btrfs_trans_handle *trans,
7021 struct btrfs_fs_info *info,
7022 struct cache_tree *extent_cache,
7023 struct extent_record *rec)
7026 struct btrfs_path *path;
7027 struct list_head *cur = rec->backrefs.next;
7028 struct cache_extent *cache;
7029 struct extent_backref *back;
7034 * remember our flags for recreating the extent.
7035 * FIXME, if we have cleared extent tree, we can not
7036 * lookup extent info in extent tree.
7038 if (!init_extent_tree) {
7039 ret = btrfs_lookup_extent_info(NULL, info->extent_root,
7040 rec->start, rec->max_size,
7041 rec->metadata, NULL, &flags);
7045 if (rec->flag_block_full_backref)
7046 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7049 path = btrfs_alloc_path();
7053 if (rec->refs != rec->extent_item_refs && !rec->metadata) {
7055 * Sometimes the backrefs themselves are so broken they don't
7056 * get attached to any meaningful rec, so first go back and
7057 * check any of our backrefs that we couldn't find and throw
7058 * them into the list if we find the backref so that
7059 * verify_backrefs can figure out what to do.
7061 ret = find_possible_backrefs(trans, info, path, extent_cache,
7067 /* step one, make sure all of the backrefs agree */
7068 ret = verify_backrefs(trans, info, path, rec);
7072 /* step two, delete all the existing records */
7073 ret = delete_extent_records(trans, info->extent_root, path,
7074 rec->start, rec->max_size);
7079 /* was this block corrupt? If so, don't add references to it */
7080 cache = lookup_cache_extent(info->corrupt_blocks,
7081 rec->start, rec->max_size);
7087 /* step three, recreate all the refs we did find */
7088 while(cur != &rec->backrefs) {
7089 back = list_entry(cur, struct extent_backref, list);
7093 * if we didn't find any references, don't create a
7096 if (!back->found_ref)
7099 ret = record_extent(trans, info, path, rec, back, allocated, flags);
7106 btrfs_free_path(path);
7110 /* right now we only prune from the extent allocation tree */
7111 static int prune_one_block(struct btrfs_trans_handle *trans,
7112 struct btrfs_fs_info *info,
7113 struct btrfs_corrupt_block *corrupt)
7116 struct btrfs_path path;
7117 struct extent_buffer *eb;
7121 int level = corrupt->level + 1;
7123 btrfs_init_path(&path);
7125 /* we want to stop at the parent to our busted block */
7126 path.lowest_level = level;
7128 ret = btrfs_search_slot(trans, info->extent_root,
7129 &corrupt->key, &path, -1, 1);
7134 eb = path.nodes[level];
7141 * hopefully the search gave us the block we want to prune,
7142 * lets try that first
7144 slot = path.slots[level];
7145 found = btrfs_node_blockptr(eb, slot);
7146 if (found == corrupt->cache.start)
7149 nritems = btrfs_header_nritems(eb);
7151 /* the search failed, lets scan this node and hope we find it */
7152 for (slot = 0; slot < nritems; slot++) {
7153 found = btrfs_node_blockptr(eb, slot);
7154 if (found == corrupt->cache.start)
7158 * we couldn't find the bad block. TODO, search all the nodes for pointers
7161 if (eb == info->extent_root->node) {
7166 btrfs_release_path(&path);
7171 printk("deleting pointer to block %Lu\n", corrupt->cache.start);
7172 ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
7175 btrfs_release_path(&path);
7179 static int prune_corrupt_blocks(struct btrfs_trans_handle *trans,
7180 struct btrfs_fs_info *info)
7182 struct cache_extent *cache;
7183 struct btrfs_corrupt_block *corrupt;
7185 cache = search_cache_extent(info->corrupt_blocks, 0);
7189 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
7190 prune_one_block(trans, info, corrupt);
7191 cache = next_cache_extent(cache);
7196 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
7198 struct btrfs_block_group_cache *cache;
7203 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
7204 &start, &end, EXTENT_DIRTY);
7207 clear_extent_dirty(&fs_info->free_space_cache, start, end,
7213 cache = btrfs_lookup_first_block_group(fs_info, start);
7218 start = cache->key.objectid + cache->key.offset;
7222 static int check_extent_refs(struct btrfs_trans_handle *trans,
7223 struct btrfs_root *root,
7224 struct cache_tree *extent_cache)
7226 struct extent_record *rec;
7227 struct cache_extent *cache;
7236 * if we're doing a repair, we have to make sure
7237 * we don't allocate from the problem extents.
7238 * In the worst case, this will be all the
7241 cache = search_cache_extent(extent_cache, 0);
7243 rec = container_of(cache, struct extent_record, cache);
7244 btrfs_pin_extent(root->fs_info,
7245 rec->start, rec->max_size);
7246 cache = next_cache_extent(cache);
7249 /* pin down all the corrupted blocks too */
7250 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
7252 btrfs_pin_extent(root->fs_info,
7253 cache->start, cache->size);
7254 cache = next_cache_extent(cache);
7256 prune_corrupt_blocks(trans, root->fs_info);
7257 reset_cached_block_groups(root->fs_info);
7261 * We need to delete any duplicate entries we find first otherwise we
7262 * could mess up the extent tree when we have backrefs that actually
7263 * belong to a different extent item and not the weird duplicate one.
7265 while (repair && !list_empty(&duplicate_extents)) {
7266 rec = list_entry(duplicate_extents.next, struct extent_record,
7268 list_del_init(&rec->list);
7270 /* Sometimes we can find a backref before we find an actual
7271 * extent, so we need to process it a little bit to see if there
7272 * truly are multiple EXTENT_ITEM_KEY's for the same range, or
7273 * if this is a backref screwup. If we need to delete stuff
7274 * process_duplicates() will return 0, otherwise it will return
7277 if (process_duplicates(root, extent_cache, rec))
7279 ret = delete_duplicate_records(trans, root, rec);
7283 * delete_duplicate_records will return the number of entries
7284 * deleted, so if it's greater than 0 then we know we actually
7285 * did something and we need to remove.
7297 cache = search_cache_extent(extent_cache, 0);
7300 rec = container_of(cache, struct extent_record, cache);
7301 if (rec->num_duplicates) {
7302 fprintf(stderr, "extent item %llu has multiple extent "
7303 "items\n", (unsigned long long)rec->start);
7307 if (rec->refs != rec->extent_item_refs) {
7308 fprintf(stderr, "ref mismatch on [%llu %llu] ",
7309 (unsigned long long)rec->start,
7310 (unsigned long long)rec->nr);
7311 fprintf(stderr, "extent item %llu, found %llu\n",
7312 (unsigned long long)rec->extent_item_refs,
7313 (unsigned long long)rec->refs);
7314 ret = record_orphan_data_extents(root->fs_info, rec);
7321 * we can't use the extent to repair file
7322 * extent, let the fallback method handle it.
7324 if (!fixed && repair) {
7325 ret = fixup_extent_refs(trans,
7336 if (all_backpointers_checked(rec, 1)) {
7337 fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
7338 (unsigned long long)rec->start,
7339 (unsigned long long)rec->nr);
7341 if (!fixed && !recorded && repair) {
7342 ret = fixup_extent_refs(trans, root->fs_info,
7350 if (!rec->owner_ref_checked) {
7351 fprintf(stderr, "owner ref check failed [%llu %llu]\n",
7352 (unsigned long long)rec->start,
7353 (unsigned long long)rec->nr);
7354 if (!fixed && !recorded && repair) {
7355 ret = fixup_extent_refs(trans, root->fs_info,
7364 remove_cache_extent(extent_cache, cache);
7365 free_all_extent_backrefs(rec);
7370 if (ret && ret != -EAGAIN) {
7371 fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
7374 btrfs_fix_block_accounting(trans, root);
7377 fprintf(stderr, "repaired damaged extent references\n");
7383 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
7387 if (type & BTRFS_BLOCK_GROUP_RAID0) {
7388 stripe_size = length;
7389 stripe_size /= num_stripes;
7390 } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
7391 stripe_size = length * 2;
7392 stripe_size /= num_stripes;
7393 } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
7394 stripe_size = length;
7395 stripe_size /= (num_stripes - 1);
7396 } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
7397 stripe_size = length;
7398 stripe_size /= (num_stripes - 2);
7400 stripe_size = length;
7406 * Check the chunk with its block group/dev list ref:
7407 * Return 0 if all refs seems valid.
7408 * Return 1 if part of refs seems valid, need later check for rebuild ref
7409 * like missing block group and needs to search extent tree to rebuild them.
7410 * Return -1 if essential refs are missing and unable to rebuild.
7412 static int check_chunk_refs(struct chunk_record *chunk_rec,
7413 struct block_group_tree *block_group_cache,
7414 struct device_extent_tree *dev_extent_cache,
7417 struct cache_extent *block_group_item;
7418 struct block_group_record *block_group_rec;
7419 struct cache_extent *dev_extent_item;
7420 struct device_extent_record *dev_extent_rec;
7427 block_group_item = lookup_cache_extent(&block_group_cache->tree,
7430 if (block_group_item) {
7431 block_group_rec = container_of(block_group_item,
7432 struct block_group_record,
7434 if (chunk_rec->length != block_group_rec->offset ||
7435 chunk_rec->offset != block_group_rec->objectid ||
7436 chunk_rec->type_flags != block_group_rec->flags) {
7439 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
7440 chunk_rec->objectid,
7445 chunk_rec->type_flags,
7446 block_group_rec->objectid,
7447 block_group_rec->type,
7448 block_group_rec->offset,
7449 block_group_rec->offset,
7450 block_group_rec->objectid,
7451 block_group_rec->flags);
7454 list_del_init(&block_group_rec->list);
7455 chunk_rec->bg_rec = block_group_rec;
7460 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
7461 chunk_rec->objectid,
7466 chunk_rec->type_flags);
7470 length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
7471 chunk_rec->num_stripes);
7472 for (i = 0; i < chunk_rec->num_stripes; ++i) {
7473 devid = chunk_rec->stripes[i].devid;
7474 offset = chunk_rec->stripes[i].offset;
7475 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
7476 devid, offset, length);
7477 if (dev_extent_item) {
7478 dev_extent_rec = container_of(dev_extent_item,
7479 struct device_extent_record,
7481 if (dev_extent_rec->objectid != devid ||
7482 dev_extent_rec->offset != offset ||
7483 dev_extent_rec->chunk_offset != chunk_rec->offset ||
7484 dev_extent_rec->length != length) {
7487 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
7488 chunk_rec->objectid,
7491 chunk_rec->stripes[i].devid,
7492 chunk_rec->stripes[i].offset,
7493 dev_extent_rec->objectid,
7494 dev_extent_rec->offset,
7495 dev_extent_rec->length);
7498 list_move(&dev_extent_rec->chunk_list,
7499 &chunk_rec->dextents);
7504 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
7505 chunk_rec->objectid,
7508 chunk_rec->stripes[i].devid,
7509 chunk_rec->stripes[i].offset);
7516 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
7517 int check_chunks(struct cache_tree *chunk_cache,
7518 struct block_group_tree *block_group_cache,
7519 struct device_extent_tree *dev_extent_cache,
7520 struct list_head *good, struct list_head *bad,
7521 struct list_head *rebuild, int silent)
7523 struct cache_extent *chunk_item;
7524 struct chunk_record *chunk_rec;
7525 struct block_group_record *bg_rec;
7526 struct device_extent_record *dext_rec;
7530 chunk_item = first_cache_extent(chunk_cache);
7531 while (chunk_item) {
7532 chunk_rec = container_of(chunk_item, struct chunk_record,
7534 err = check_chunk_refs(chunk_rec, block_group_cache,
7535 dev_extent_cache, silent);
7538 if (err == 0 && good)
7539 list_add_tail(&chunk_rec->list, good);
7540 if (err > 0 && rebuild)
7541 list_add_tail(&chunk_rec->list, rebuild);
7543 list_add_tail(&chunk_rec->list, bad);
7544 chunk_item = next_cache_extent(chunk_item);
7547 list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
7550 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
7558 list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
7562 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
7573 static int check_device_used(struct device_record *dev_rec,
7574 struct device_extent_tree *dext_cache)
7576 struct cache_extent *cache;
7577 struct device_extent_record *dev_extent_rec;
7580 cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
7582 dev_extent_rec = container_of(cache,
7583 struct device_extent_record,
7585 if (dev_extent_rec->objectid != dev_rec->devid)
7588 list_del_init(&dev_extent_rec->device_list);
7589 total_byte += dev_extent_rec->length;
7590 cache = next_cache_extent(cache);
7593 if (total_byte != dev_rec->byte_used) {
7595 "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
7596 total_byte, dev_rec->byte_used, dev_rec->objectid,
7597 dev_rec->type, dev_rec->offset);
7604 /* check btrfs_dev_item -> btrfs_dev_extent */
7605 static int check_devices(struct rb_root *dev_cache,
7606 struct device_extent_tree *dev_extent_cache)
7608 struct rb_node *dev_node;
7609 struct device_record *dev_rec;
7610 struct device_extent_record *dext_rec;
7614 dev_node = rb_first(dev_cache);
7616 dev_rec = container_of(dev_node, struct device_record, node);
7617 err = check_device_used(dev_rec, dev_extent_cache);
7621 dev_node = rb_next(dev_node);
7623 list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
7626 "Device extent[%llu, %llu, %llu] didn't find its device.\n",
7627 dext_rec->objectid, dext_rec->offset, dext_rec->length);
7634 static int add_root_item_to_list(struct list_head *head,
7635 u64 objectid, u64 bytenr,
7636 u8 level, u8 drop_level,
7637 int level_size, struct btrfs_key *drop_key)
7640 struct root_item_record *ri_rec;
7641 ri_rec = malloc(sizeof(*ri_rec));
7644 ri_rec->bytenr = bytenr;
7645 ri_rec->objectid = objectid;
7646 ri_rec->level = level;
7647 ri_rec->level_size = level_size;
7648 ri_rec->drop_level = drop_level;
7650 memcpy(&ri_rec->drop_key, drop_key, sizeof(*drop_key));
7651 list_add_tail(&ri_rec->list, head);
7656 static int deal_root_from_list(struct list_head *list,
7657 struct btrfs_trans_handle *trans,
7658 struct btrfs_root *root,
7659 struct block_info *bits,
7661 struct cache_tree *pending,
7662 struct cache_tree *seen,
7663 struct cache_tree *reada,
7664 struct cache_tree *nodes,
7665 struct cache_tree *extent_cache,
7666 struct cache_tree *chunk_cache,
7667 struct rb_root *dev_cache,
7668 struct block_group_tree *block_group_cache,
7669 struct device_extent_tree *dev_extent_cache)
7674 while (!list_empty(list)) {
7675 struct root_item_record *rec;
7676 struct extent_buffer *buf;
7677 rec = list_entry(list->next,
7678 struct root_item_record, list);
7680 buf = read_tree_block(root->fs_info->tree_root,
7681 rec->bytenr, rec->level_size, 0);
7682 if (!extent_buffer_uptodate(buf)) {
7683 free_extent_buffer(buf);
7687 add_root_to_pending(buf, extent_cache, pending,
7688 seen, nodes, rec->objectid);
7690 * To rebuild extent tree, we need deal with snapshot
7691 * one by one, otherwise we deal with node firstly which
7692 * can maximize readahead.
7694 if (!init_extent_tree && !rec->drop_level)
7697 ret = run_next_block(trans, root, bits, bits_nr, &last,
7698 pending, seen, reada,
7699 nodes, extent_cache,
7700 chunk_cache, dev_cache,
7702 dev_extent_cache, rec);
7707 free_extent_buffer(buf);
7708 list_del(&rec->list);
7712 ret = run_next_block(trans, root, bits, bits_nr, &last,
7713 pending, seen, reada,
7714 nodes, extent_cache,
7715 chunk_cache, dev_cache,
7717 dev_extent_cache, NULL);
7727 static int check_chunks_and_extents(struct btrfs_root *root)
7729 struct rb_root dev_cache;
7730 struct cache_tree chunk_cache;
7731 struct block_group_tree block_group_cache;
7732 struct device_extent_tree dev_extent_cache;
7733 struct cache_tree extent_cache;
7734 struct cache_tree seen;
7735 struct cache_tree pending;
7736 struct cache_tree reada;
7737 struct cache_tree nodes;
7738 struct cache_tree corrupt_blocks;
7739 struct btrfs_path path;
7740 struct btrfs_key key;
7741 struct btrfs_key found_key;
7743 struct block_info *bits;
7745 struct extent_buffer *leaf;
7746 struct btrfs_trans_handle *trans = NULL;
7748 struct btrfs_root_item ri;
7749 struct list_head dropping_trees;
7750 struct list_head normal_trees;
7751 struct btrfs_root *root1;
7756 dev_cache = RB_ROOT;
7757 cache_tree_init(&chunk_cache);
7758 block_group_tree_init(&block_group_cache);
7759 device_extent_tree_init(&dev_extent_cache);
7761 cache_tree_init(&extent_cache);
7762 cache_tree_init(&seen);
7763 cache_tree_init(&pending);
7764 cache_tree_init(&nodes);
7765 cache_tree_init(&reada);
7766 cache_tree_init(&corrupt_blocks);
7767 INIT_LIST_HEAD(&dropping_trees);
7768 INIT_LIST_HEAD(&normal_trees);
7771 trans = btrfs_start_transaction(root, 1);
7772 if (IS_ERR(trans)) {
7773 fprintf(stderr, "Error starting transaction\n");
7774 return PTR_ERR(trans);
7776 root->fs_info->fsck_extent_cache = &extent_cache;
7777 root->fs_info->free_extent_hook = free_extent_hook;
7778 root->fs_info->corrupt_blocks = &corrupt_blocks;
7782 bits = malloc(bits_nr * sizeof(struct block_info));
7789 root1 = root->fs_info->tree_root;
7790 level = btrfs_header_level(root1->node);
7791 ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
7792 root1->node->start, level, 0,
7793 btrfs_level_size(root1, level), NULL);
7796 root1 = root->fs_info->chunk_root;
7797 level = btrfs_header_level(root1->node);
7798 ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
7799 root1->node->start, level, 0,
7800 btrfs_level_size(root1, level), NULL);
7803 btrfs_init_path(&path);
7806 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
7807 ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
7812 leaf = path.nodes[0];
7813 slot = path.slots[0];
7814 if (slot >= btrfs_header_nritems(path.nodes[0])) {
7815 ret = btrfs_next_leaf(root, &path);
7818 leaf = path.nodes[0];
7819 slot = path.slots[0];
7821 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
7822 if (btrfs_key_type(&found_key) == BTRFS_ROOT_ITEM_KEY) {
7823 unsigned long offset;
7825 offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
7826 read_extent_buffer(leaf, &ri, offset, sizeof(ri));
7827 if (btrfs_disk_key_objectid(&ri.drop_progress) == 0) {
7828 level = btrfs_root_level(&ri);
7829 level_size = btrfs_level_size(root, level);
7830 ret = add_root_item_to_list(&normal_trees,
7832 btrfs_root_bytenr(&ri), level,
7833 0, level_size, NULL);
7837 level = btrfs_root_level(&ri);
7838 level_size = btrfs_level_size(root, level);
7839 objectid = found_key.objectid;
7840 btrfs_disk_key_to_cpu(&found_key,
7842 ret = add_root_item_to_list(&dropping_trees,
7844 btrfs_root_bytenr(&ri),
7845 level, ri.drop_level,
7846 level_size, &found_key);
7853 btrfs_release_path(&path);
7854 ret = deal_root_from_list(&normal_trees, trans, root,
7855 bits, bits_nr, &pending, &seen,
7856 &reada, &nodes, &extent_cache,
7857 &chunk_cache, &dev_cache, &block_group_cache,
7861 ret = deal_root_from_list(&dropping_trees, trans, root,
7862 bits, bits_nr, &pending, &seen,
7863 &reada, &nodes, &extent_cache,
7864 &chunk_cache, &dev_cache, &block_group_cache,
7869 ret = check_extent_refs(trans, root, &extent_cache);
7870 if (ret == -EAGAIN) {
7871 ret = btrfs_commit_transaction(trans, root);
7875 trans = btrfs_start_transaction(root, 1);
7876 if (IS_ERR(trans)) {
7877 ret = PTR_ERR(trans);
7881 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
7882 free_extent_cache_tree(&seen);
7883 free_extent_cache_tree(&pending);
7884 free_extent_cache_tree(&reada);
7885 free_extent_cache_tree(&nodes);
7886 free_chunk_cache_tree(&chunk_cache);
7887 free_block_group_tree(&block_group_cache);
7888 free_device_cache_tree(&dev_cache);
7889 free_device_extent_tree(&dev_extent_cache);
7890 free_extent_record_cache(root->fs_info, &extent_cache);
7894 err = check_chunks(&chunk_cache, &block_group_cache,
7895 &dev_extent_cache, NULL, NULL, NULL, 0);
7899 err = check_devices(&dev_cache, &dev_extent_cache);
7905 err = btrfs_commit_transaction(trans, root);
7910 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
7911 root->fs_info->fsck_extent_cache = NULL;
7912 root->fs_info->free_extent_hook = NULL;
7913 root->fs_info->corrupt_blocks = NULL;
7916 free_chunk_cache_tree(&chunk_cache);
7917 free_device_cache_tree(&dev_cache);
7918 free_block_group_tree(&block_group_cache);
7919 free_device_extent_tree(&dev_extent_cache);
7920 free_extent_cache_tree(&seen);
7921 free_extent_cache_tree(&pending);
7922 free_extent_cache_tree(&reada);
7923 free_extent_cache_tree(&nodes);
7927 static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
7928 struct btrfs_root *root, int overwrite)
7930 struct extent_buffer *c;
7931 struct extent_buffer *old = root->node;
7934 struct btrfs_disk_key disk_key = {0,0,0};
7940 extent_buffer_get(c);
7943 c = btrfs_alloc_free_block(trans, root,
7944 btrfs_level_size(root, 0),
7945 root->root_key.objectid,
7946 &disk_key, level, 0, 0);
7949 extent_buffer_get(c);
7953 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
7954 btrfs_set_header_level(c, level);
7955 btrfs_set_header_bytenr(c, c->start);
7956 btrfs_set_header_generation(c, trans->transid);
7957 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
7958 btrfs_set_header_owner(c, root->root_key.objectid);
7960 write_extent_buffer(c, root->fs_info->fsid,
7961 btrfs_header_fsid(), BTRFS_FSID_SIZE);
7963 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
7964 btrfs_header_chunk_tree_uuid(c),
7967 btrfs_mark_buffer_dirty(c);
7969 * this case can happen in the following case:
7971 * 1.overwrite previous root.
7973 * 2.reinit reloc data root, this is because we skip pin
7974 * down reloc data tree before which means we can allocate
7975 * same block bytenr here.
7977 if (old->start == c->start) {
7978 btrfs_set_root_generation(&root->root_item,
7980 root->root_item.level = btrfs_header_level(root->node);
7981 ret = btrfs_update_root(trans, root->fs_info->tree_root,
7982 &root->root_key, &root->root_item);
7984 free_extent_buffer(c);
7988 free_extent_buffer(old);
7990 add_root_to_dirty_list(root);
7994 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
7995 struct extent_buffer *eb, int tree_root)
7997 struct extent_buffer *tmp;
7998 struct btrfs_root_item *ri;
7999 struct btrfs_key key;
8002 int level = btrfs_header_level(eb);
8008 * If we have pinned this block before, don't pin it again.
8009 * This can not only avoid forever loop with broken filesystem
8010 * but also give us some speedups.
8012 if (test_range_bit(&fs_info->pinned_extents, eb->start,
8013 eb->start + eb->len - 1, EXTENT_DIRTY, 0))
8016 btrfs_pin_extent(fs_info, eb->start, eb->len);
8018 leafsize = btrfs_super_leafsize(fs_info->super_copy);
8019 nritems = btrfs_header_nritems(eb);
8020 for (i = 0; i < nritems; i++) {
8022 btrfs_item_key_to_cpu(eb, &key, i);
8023 if (key.type != BTRFS_ROOT_ITEM_KEY)
8025 /* Skip the extent root and reloc roots */
8026 if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
8027 key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
8028 key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
8030 ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
8031 bytenr = btrfs_disk_root_bytenr(eb, ri);
8034 * If at any point we start needing the real root we
8035 * will have to build a stump root for the root we are
8036 * in, but for now this doesn't actually use the root so
8037 * just pass in extent_root.
8039 tmp = read_tree_block(fs_info->extent_root, bytenr,
8041 if (!extent_buffer_uptodate(tmp)) {
8042 fprintf(stderr, "Error reading root block\n");
8045 ret = pin_down_tree_blocks(fs_info, tmp, 0);
8046 free_extent_buffer(tmp);
8050 bytenr = btrfs_node_blockptr(eb, i);
8052 /* If we aren't the tree root don't read the block */
8053 if (level == 1 && !tree_root) {
8054 btrfs_pin_extent(fs_info, bytenr, leafsize);
8058 tmp = read_tree_block(fs_info->extent_root, bytenr,
8060 if (!extent_buffer_uptodate(tmp)) {
8061 fprintf(stderr, "Error reading tree block\n");
8064 ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
8065 free_extent_buffer(tmp);
8074 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
8078 ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
8082 return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
8085 static int reset_block_groups(struct btrfs_fs_info *fs_info)
8087 struct btrfs_block_group_cache *cache;
8088 struct btrfs_path *path;
8089 struct extent_buffer *leaf;
8090 struct btrfs_chunk *chunk;
8091 struct btrfs_key key;
8095 path = btrfs_alloc_path();
8100 key.type = BTRFS_CHUNK_ITEM_KEY;
8103 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
8105 btrfs_free_path(path);
8110 * We do this in case the block groups were screwed up and had alloc
8111 * bits that aren't actually set on the chunks. This happens with
8112 * restored images every time and could happen in real life I guess.
8114 fs_info->avail_data_alloc_bits = 0;
8115 fs_info->avail_metadata_alloc_bits = 0;
8116 fs_info->avail_system_alloc_bits = 0;
8118 /* First we need to create the in-memory block groups */
8120 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8121 ret = btrfs_next_leaf(fs_info->chunk_root, path);
8123 btrfs_free_path(path);
8131 leaf = path->nodes[0];
8132 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
8133 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
8138 chunk = btrfs_item_ptr(leaf, path->slots[0],
8139 struct btrfs_chunk);
8140 btrfs_add_block_group(fs_info, 0,
8141 btrfs_chunk_type(leaf, chunk),
8142 key.objectid, key.offset,
8143 btrfs_chunk_length(leaf, chunk));
8144 set_extent_dirty(&fs_info->free_space_cache, key.offset,
8145 key.offset + btrfs_chunk_length(leaf, chunk),
8151 cache = btrfs_lookup_first_block_group(fs_info, start);
8155 start = cache->key.objectid + cache->key.offset;
8158 btrfs_free_path(path);
8162 static int reset_balance(struct btrfs_trans_handle *trans,
8163 struct btrfs_fs_info *fs_info)
8165 struct btrfs_root *root = fs_info->tree_root;
8166 struct btrfs_path *path;
8167 struct extent_buffer *leaf;
8168 struct btrfs_key key;
8169 int del_slot, del_nr = 0;
8173 path = btrfs_alloc_path();
8177 key.objectid = BTRFS_BALANCE_OBJECTID;
8178 key.type = BTRFS_BALANCE_ITEM_KEY;
8181 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8186 goto reinit_data_reloc;
8191 ret = btrfs_del_item(trans, root, path);
8194 btrfs_release_path(path);
8196 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
8197 key.type = BTRFS_ROOT_ITEM_KEY;
8200 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
8204 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8209 ret = btrfs_del_items(trans, root, path,
8216 btrfs_release_path(path);
8219 ret = btrfs_search_slot(trans, root, &key, path,
8226 leaf = path->nodes[0];
8227 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
8228 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
8230 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8235 del_slot = path->slots[0];
8244 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
8248 btrfs_release_path(path);
8251 key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
8252 key.type = BTRFS_ROOT_ITEM_KEY;
8253 key.offset = (u64)-1;
8254 root = btrfs_read_fs_root(fs_info, &key);
8256 fprintf(stderr, "Error reading data reloc tree\n");
8257 ret = PTR_ERR(root);
8260 record_root_in_trans(trans, root);
8261 ret = btrfs_fsck_reinit_root(trans, root, 0);
8264 ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
8266 btrfs_free_path(path);
8270 static int reinit_extent_tree(struct btrfs_trans_handle *trans,
8271 struct btrfs_fs_info *fs_info)
8277 * The only reason we don't do this is because right now we're just
8278 * walking the trees we find and pinning down their bytes, we don't look
8279 * at any of the leaves. In order to do mixed groups we'd have to check
8280 * the leaves of any fs roots and pin down the bytes for any file
8281 * extents we find. Not hard but why do it if we don't have to?
8283 if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
8284 fprintf(stderr, "We don't support re-initing the extent tree "
8285 "for mixed block groups yet, please notify a btrfs "
8286 "developer you want to do this so they can add this "
8287 "functionality.\n");
8292 * first we need to walk all of the trees except the extent tree and pin
8293 * down the bytes that are in use so we don't overwrite any existing
8296 ret = pin_metadata_blocks(fs_info);
8298 fprintf(stderr, "error pinning down used bytes\n");
8303 * Need to drop all the block groups since we're going to recreate all
8306 btrfs_free_block_groups(fs_info);
8307 ret = reset_block_groups(fs_info);
8309 fprintf(stderr, "error resetting the block groups\n");
8313 /* Ok we can allocate now, reinit the extent root */
8314 ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 0);
8316 fprintf(stderr, "extent root initialization failed\n");
8318 * When the transaction code is updated we should end the
8319 * transaction, but for now progs only knows about commit so
8320 * just return an error.
8326 * Now we have all the in-memory block groups setup so we can make
8327 * allocations properly, and the metadata we care about is safe since we
8328 * pinned all of it above.
8331 struct btrfs_block_group_cache *cache;
8333 cache = btrfs_lookup_first_block_group(fs_info, start);
8336 start = cache->key.objectid + cache->key.offset;
8337 ret = btrfs_insert_item(trans, fs_info->extent_root,
8338 &cache->key, &cache->item,
8339 sizeof(cache->item));
8341 fprintf(stderr, "Error adding block group\n");
8344 btrfs_extent_post_op(trans, fs_info->extent_root);
8347 ret = reset_balance(trans, fs_info);
8349 fprintf(stderr, "error reseting the pending balance\n");
8354 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
8356 struct btrfs_path *path;
8357 struct btrfs_trans_handle *trans;
8358 struct btrfs_key key;
8361 printf("Recowing metadata block %llu\n", eb->start);
8362 key.objectid = btrfs_header_owner(eb);
8363 key.type = BTRFS_ROOT_ITEM_KEY;
8364 key.offset = (u64)-1;
8366 root = btrfs_read_fs_root(root->fs_info, &key);
8368 fprintf(stderr, "Couldn't find owner root %llu\n",
8370 return PTR_ERR(root);
8373 path = btrfs_alloc_path();
8377 trans = btrfs_start_transaction(root, 1);
8378 if (IS_ERR(trans)) {
8379 btrfs_free_path(path);
8380 return PTR_ERR(trans);
8383 path->lowest_level = btrfs_header_level(eb);
8384 if (path->lowest_level)
8385 btrfs_node_key_to_cpu(eb, &key, 0);
8387 btrfs_item_key_to_cpu(eb, &key, 0);
8389 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
8390 btrfs_commit_transaction(trans, root);
8391 btrfs_free_path(path);
8395 static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
8397 struct btrfs_path *path;
8398 struct btrfs_trans_handle *trans;
8399 struct btrfs_key key;
8402 printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
8403 bad->key.type, bad->key.offset);
8404 key.objectid = bad->root_id;
8405 key.type = BTRFS_ROOT_ITEM_KEY;
8406 key.offset = (u64)-1;
8408 root = btrfs_read_fs_root(root->fs_info, &key);
8410 fprintf(stderr, "Couldn't find owner root %llu\n",
8412 return PTR_ERR(root);
8415 path = btrfs_alloc_path();
8419 trans = btrfs_start_transaction(root, 1);
8420 if (IS_ERR(trans)) {
8421 btrfs_free_path(path);
8422 return PTR_ERR(trans);
8425 ret = btrfs_search_slot(trans, root, &bad->key, path, -1, 1);
8431 ret = btrfs_del_item(trans, root, path);
8433 btrfs_commit_transaction(trans, root);
8434 btrfs_free_path(path);
8438 static int zero_log_tree(struct btrfs_root *root)
8440 struct btrfs_trans_handle *trans;
8443 trans = btrfs_start_transaction(root, 1);
8444 if (IS_ERR(trans)) {
8445 ret = PTR_ERR(trans);
8448 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
8449 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
8450 ret = btrfs_commit_transaction(trans, root);
8454 static int populate_csum(struct btrfs_trans_handle *trans,
8455 struct btrfs_root *csum_root, char *buf, u64 start,
8462 while (offset < len) {
8463 sectorsize = csum_root->sectorsize;
8464 ret = read_extent_data(csum_root, buf, start + offset,
8468 ret = btrfs_csum_file_block(trans, csum_root, start + len,
8469 start + offset, buf, sectorsize);
8472 offset += sectorsize;
8477 static int fill_csum_tree(struct btrfs_trans_handle *trans,
8478 struct btrfs_root *csum_root)
8480 struct btrfs_root *extent_root = csum_root->fs_info->extent_root;
8481 struct btrfs_path *path;
8482 struct btrfs_extent_item *ei;
8483 struct extent_buffer *leaf;
8485 struct btrfs_key key;
8488 path = btrfs_alloc_path();
8493 key.type = BTRFS_EXTENT_ITEM_KEY;
8496 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
8498 btrfs_free_path(path);
8502 buf = malloc(csum_root->sectorsize);
8504 btrfs_free_path(path);
8509 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8510 ret = btrfs_next_leaf(extent_root, path);
8518 leaf = path->nodes[0];
8520 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
8521 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
8526 ei = btrfs_item_ptr(leaf, path->slots[0],
8527 struct btrfs_extent_item);
8528 if (!(btrfs_extent_flags(leaf, ei) &
8529 BTRFS_EXTENT_FLAG_DATA)) {
8534 ret = populate_csum(trans, csum_root, buf, key.objectid,
8541 btrfs_free_path(path);
8546 struct root_item_info {
8547 /* level of the root */
8549 /* number of nodes at this level, must be 1 for a root */
8553 struct cache_extent cache_extent;
8556 static struct cache_tree *roots_info_cache = NULL;
8558 static void free_roots_info_cache(void)
8560 if (!roots_info_cache)
8563 while (!cache_tree_empty(roots_info_cache)) {
8564 struct cache_extent *entry;
8565 struct root_item_info *rii;
8567 entry = first_cache_extent(roots_info_cache);
8570 remove_cache_extent(roots_info_cache, entry);
8571 rii = container_of(entry, struct root_item_info, cache_extent);
8575 free(roots_info_cache);
8576 roots_info_cache = NULL;
8579 static int build_roots_info_cache(struct btrfs_fs_info *info)
8582 struct btrfs_key key;
8583 struct extent_buffer *leaf;
8584 struct btrfs_path *path;
8586 if (!roots_info_cache) {
8587 roots_info_cache = malloc(sizeof(*roots_info_cache));
8588 if (!roots_info_cache)
8590 cache_tree_init(roots_info_cache);
8593 path = btrfs_alloc_path();
8598 key.type = BTRFS_EXTENT_ITEM_KEY;
8601 ret = btrfs_search_slot(NULL, info->extent_root, &key, path, 0, 0);
8604 leaf = path->nodes[0];
8607 struct btrfs_key found_key;
8608 struct btrfs_extent_item *ei;
8609 struct btrfs_extent_inline_ref *iref;
8610 int slot = path->slots[0];
8615 struct cache_extent *entry;
8616 struct root_item_info *rii;
8618 if (slot >= btrfs_header_nritems(leaf)) {
8619 ret = btrfs_next_leaf(info->extent_root, path);
8626 leaf = path->nodes[0];
8627 slot = path->slots[0];
8630 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8632 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
8633 found_key.type != BTRFS_METADATA_ITEM_KEY)
8636 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
8637 flags = btrfs_extent_flags(leaf, ei);
8639 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
8640 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
8643 if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
8644 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
8645 level = found_key.offset;
8647 struct btrfs_tree_block_info *info;
8649 info = (struct btrfs_tree_block_info *)(ei + 1);
8650 iref = (struct btrfs_extent_inline_ref *)(info + 1);
8651 level = btrfs_tree_block_level(leaf, info);
8655 * For a root extent, it must be of the following type and the
8656 * first (and only one) iref in the item.
8658 type = btrfs_extent_inline_ref_type(leaf, iref);
8659 if (type != BTRFS_TREE_BLOCK_REF_KEY)
8662 root_id = btrfs_extent_inline_ref_offset(leaf, iref);
8663 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
8665 rii = malloc(sizeof(struct root_item_info));
8670 rii->cache_extent.start = root_id;
8671 rii->cache_extent.size = 1;
8672 rii->level = (u8)-1;
8673 entry = &rii->cache_extent;
8674 ret = insert_cache_extent(roots_info_cache, entry);
8677 rii = container_of(entry, struct root_item_info,
8681 ASSERT(rii->cache_extent.start == root_id);
8682 ASSERT(rii->cache_extent.size == 1);
8684 if (level > rii->level || rii->level == (u8)-1) {
8686 rii->bytenr = found_key.objectid;
8687 rii->gen = btrfs_extent_generation(leaf, ei);
8688 rii->node_count = 1;
8689 } else if (level == rii->level) {
8697 btrfs_free_path(path);
8702 static int maybe_repair_root_item(struct btrfs_fs_info *info,
8703 struct btrfs_path *path,
8704 const struct btrfs_key *root_key,
8705 const int read_only_mode)
8707 const u64 root_id = root_key->objectid;
8708 struct cache_extent *entry;
8709 struct root_item_info *rii;
8710 struct btrfs_root_item ri;
8711 unsigned long offset;
8713 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
8716 "Error: could not find extent items for root %llu\n",
8717 root_key->objectid);
8721 rii = container_of(entry, struct root_item_info, cache_extent);
8722 ASSERT(rii->cache_extent.start == root_id);
8723 ASSERT(rii->cache_extent.size == 1);
8725 if (rii->node_count != 1) {
8727 "Error: could not find btree root extent for root %llu\n",
8732 offset = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
8733 read_extent_buffer(path->nodes[0], &ri, offset, sizeof(ri));
8735 if (btrfs_root_bytenr(&ri) != rii->bytenr ||
8736 btrfs_root_level(&ri) != rii->level ||
8737 btrfs_root_generation(&ri) != rii->gen) {
8740 * If we're in repair mode but our caller told us to not update
8741 * the root item, i.e. just check if it needs to be updated, don't
8742 * print this message, since the caller will call us again shortly
8743 * for the same root item without read only mode (the caller will
8744 * open a transaction first).
8746 if (!(read_only_mode && repair))
8748 "%sroot item for root %llu,"
8749 " current bytenr %llu, current gen %llu, current level %u,"
8750 " new bytenr %llu, new gen %llu, new level %u\n",
8751 (read_only_mode ? "" : "fixing "),
8753 btrfs_root_bytenr(&ri), btrfs_root_generation(&ri),
8754 btrfs_root_level(&ri),
8755 rii->bytenr, rii->gen, rii->level);
8757 if (btrfs_root_generation(&ri) > rii->gen) {
8759 "root %llu has a root item with a more recent gen (%llu) compared to the found root node (%llu)\n",
8760 root_id, btrfs_root_generation(&ri), rii->gen);
8764 if (!read_only_mode) {
8765 btrfs_set_root_bytenr(&ri, rii->bytenr);
8766 btrfs_set_root_level(&ri, rii->level);
8767 btrfs_set_root_generation(&ri, rii->gen);
8768 write_extent_buffer(path->nodes[0], &ri,
8769 offset, sizeof(ri));
8779 * A regression introduced in the 3.17 kernel (more specifically in 3.17-rc2),
8780 * caused read-only snapshots to be corrupted if they were created at a moment
8781 * when the source subvolume/snapshot had orphan items. The issue was that the
8782 * on-disk root items became incorrect, referring to the pre orphan cleanup root
8783 * node instead of the post orphan cleanup root node.
8784 * So this function, and its callees, just detects and fixes those cases. Even
8785 * though the regression was for read-only snapshots, this function applies to
8786 * any snapshot/subvolume root.
8787 * This must be run before any other repair code - not doing it so, makes other
8788 * repair code delete or modify backrefs in the extent tree for example, which
8789 * will result in an inconsistent fs after repairing the root items.
8791 static int repair_root_items(struct btrfs_fs_info *info)
8793 struct btrfs_path *path = NULL;
8794 struct btrfs_key key;
8795 struct extent_buffer *leaf;
8796 struct btrfs_trans_handle *trans = NULL;
8801 ret = build_roots_info_cache(info);
8805 path = btrfs_alloc_path();
8811 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
8812 key.type = BTRFS_ROOT_ITEM_KEY;
8817 * Avoid opening and committing transactions if a leaf doesn't have
8818 * any root items that need to be fixed, so that we avoid rotating
8819 * backup roots unnecessarily.
8822 trans = btrfs_start_transaction(info->tree_root, 1);
8823 if (IS_ERR(trans)) {
8824 ret = PTR_ERR(trans);
8829 ret = btrfs_search_slot(trans, info->tree_root, &key, path,
8833 leaf = path->nodes[0];
8836 struct btrfs_key found_key;
8838 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
8839 int no_more_keys = find_next_key(path, &key);
8841 btrfs_release_path(path);
8843 ret = btrfs_commit_transaction(trans,
8855 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
8857 if (found_key.type != BTRFS_ROOT_ITEM_KEY)
8860 ret = maybe_repair_root_item(info, path, &found_key,
8865 if (!trans && repair) {
8868 btrfs_release_path(path);
8878 free_roots_info_cache();
8880 btrfs_free_path(path);
8887 const char * const cmd_check_usage[] = {
8888 "btrfs check [options] <device>",
8889 "Check an unmounted btrfs filesystem.",
8891 "-s|--super <superblock> use this superblock copy",
8892 "-b|--backup use the backup root copy",
8893 "--repair try to repair the filesystem",
8894 "--init-csum-tree create a new CRC tree",
8895 "--init-extent-tree create a new extent tree",
8896 "--check-data-csum verify checkums of data blocks",
8897 "--qgroup-report print a report on qgroup consistency",
8898 "--subvol-extents <subvolid> print subvolume extents and sharing state",
8899 "--tree-root <bytenr> use the given bytenr for the tree root",
8903 int cmd_check(int argc, char **argv)
8905 struct cache_tree root_cache;
8906 struct btrfs_root *root;
8907 struct btrfs_fs_info *info;
8910 u64 tree_root_bytenr = 0;
8911 char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
8914 int init_csum_tree = 0;
8916 int qgroup_report = 0;
8917 enum btrfs_open_ctree_flags ctree_flags = OPEN_CTREE_EXCLUSIVE;
8921 int option_index = 0;
8922 enum { OPT_REPAIR = 257, OPT_INIT_CSUM, OPT_INIT_EXTENT,
8923 OPT_CHECK_CSUM, OPT_READONLY };
8924 static const struct option long_options[] = {
8925 { "super", 1, NULL, 's' },
8926 { "repair", 0, NULL, OPT_REPAIR },
8927 { "readonly", 0, NULL, OPT_READONLY },
8928 { "init-csum-tree", 0, NULL, OPT_INIT_CSUM },
8929 { "init-extent-tree", 0, NULL, OPT_INIT_EXTENT },
8930 { "check-data-csum", 0, NULL, OPT_CHECK_CSUM },
8931 { "backup", 0, NULL, 'b' },
8932 { "subvol-extents", 1, NULL, 'E' },
8933 { "qgroup-report", 0, NULL, 'Q' },
8934 { "tree-root", 1, NULL, 'r' },
8938 c = getopt_long(argc, argv, "as:br:", long_options,
8943 case 'a': /* ignored */ break;
8945 ctree_flags |= OPEN_CTREE_BACKUP_ROOT;
8948 num = arg_strtou64(optarg);
8949 if (num >= BTRFS_SUPER_MIRROR_MAX) {
8951 "ERROR: super mirror should be less than: %d\n",
8952 BTRFS_SUPER_MIRROR_MAX);
8955 bytenr = btrfs_sb_offset(((int)num));
8956 printf("using SB copy %llu, bytenr %llu\n", num,
8957 (unsigned long long)bytenr);
8963 subvolid = arg_strtou64(optarg);
8966 tree_root_bytenr = arg_strtou64(optarg);
8970 usage(cmd_check_usage);
8972 printf("enabling repair mode\n");
8974 ctree_flags |= OPEN_CTREE_WRITES;
8980 printf("Creating a new CRC tree\n");
8983 ctree_flags |= OPEN_CTREE_WRITES;
8985 case OPT_INIT_EXTENT:
8986 init_extent_tree = 1;
8987 ctree_flags |= (OPEN_CTREE_WRITES |
8988 OPEN_CTREE_NO_BLOCK_GROUPS);
8991 case OPT_CHECK_CSUM:
8992 check_data_csum = 1;
8996 argc = argc - optind;
8998 if (check_argc_exact(argc, 1))
8999 usage(cmd_check_usage);
9001 /* This check is the only reason for --readonly to exist */
9002 if (readonly && repair) {
9003 fprintf(stderr, "Repair options are not compatible with --readonly\n");
9008 cache_tree_init(&root_cache);
9010 if((ret = check_mounted(argv[optind])) < 0) {
9011 fprintf(stderr, "Could not check mount status: %s\n", strerror(-ret));
9014 fprintf(stderr, "%s is currently mounted. Aborting.\n", argv[optind]);
9019 /* only allow partial opening under repair mode */
9021 ctree_flags |= OPEN_CTREE_PARTIAL;
9023 info = open_ctree_fs_info(argv[optind], bytenr, tree_root_bytenr,
9026 fprintf(stderr, "Couldn't open file system\n");
9031 root = info->fs_root;
9034 * repair mode will force us to commit transaction which
9035 * will make us fail to load log tree when mounting.
9037 if (repair && btrfs_super_log_root(info->super_copy)) {
9038 ret = ask_user("repair mode will force to clear out log tree, Are you sure?");
9043 ret = zero_log_tree(root);
9045 fprintf(stderr, "fail to zero log tree\n");
9050 uuid_unparse(info->super_copy->fsid, uuidbuf);
9051 if (qgroup_report) {
9052 printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
9054 ret = qgroup_verify_all(info);
9056 print_qgroup_report(1);
9060 printf("Print extent state for subvolume %llu on %s\nUUID: %s\n",
9061 subvolid, argv[optind], uuidbuf);
9062 ret = print_extent_state(info, subvolid);
9065 printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
9067 if (!extent_buffer_uptodate(info->tree_root->node) ||
9068 !extent_buffer_uptodate(info->dev_root->node) ||
9069 !extent_buffer_uptodate(info->chunk_root->node)) {
9070 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
9075 if (init_extent_tree || init_csum_tree) {
9076 struct btrfs_trans_handle *trans;
9078 trans = btrfs_start_transaction(info->extent_root, 0);
9079 if (IS_ERR(trans)) {
9080 fprintf(stderr, "Error starting transaction\n");
9081 ret = PTR_ERR(trans);
9085 if (init_extent_tree) {
9086 printf("Creating a new extent tree\n");
9087 ret = reinit_extent_tree(trans, info);
9092 if (init_csum_tree) {
9093 fprintf(stderr, "Reinit crc root\n");
9094 ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
9096 fprintf(stderr, "crc root initialization failed\n");
9101 ret = fill_csum_tree(trans, info->csum_root);
9103 fprintf(stderr, "crc refilling failed\n");
9108 * Ok now we commit and run the normal fsck, which will add
9109 * extent entries for all of the items it finds.
9111 ret = btrfs_commit_transaction(trans, info->extent_root);
9115 if (!extent_buffer_uptodate(info->extent_root->node)) {
9116 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
9120 if (!extent_buffer_uptodate(info->csum_root->node)) {
9121 fprintf(stderr, "Checksum root corrupted, rerun with --init-csum-tree option\n");
9126 fprintf(stderr, "checking extents\n");
9127 ret = check_chunks_and_extents(root);
9129 fprintf(stderr, "Errors found in extent allocation tree or chunk allocation\n");
9131 ret = repair_root_items(info);
9135 fprintf(stderr, "Fixed %d roots.\n", ret);
9137 } else if (ret > 0) {
9139 "Found %d roots with an outdated root item.\n",
9142 "Please run a filesystem check with the option --repair to fix them.\n");
9147 fprintf(stderr, "checking free space cache\n");
9148 ret = check_space_cache(root);
9153 * We used to have to have these hole extents in between our real
9154 * extents so if we don't have this flag set we need to make sure there
9155 * are no gaps in the file extents for inodes, otherwise we can just
9156 * ignore it when this happens.
9158 no_holes = btrfs_fs_incompat(root->fs_info,
9159 BTRFS_FEATURE_INCOMPAT_NO_HOLES);
9160 fprintf(stderr, "checking fs roots\n");
9161 ret = check_fs_roots(root, &root_cache);
9165 fprintf(stderr, "checking csums\n");
9166 ret = check_csums(root);
9170 fprintf(stderr, "checking root refs\n");
9171 ret = check_root_refs(root, &root_cache);
9175 while (repair && !list_empty(&root->fs_info->recow_ebs)) {
9176 struct extent_buffer *eb;
9178 eb = list_first_entry(&root->fs_info->recow_ebs,
9179 struct extent_buffer, recow);
9180 list_del_init(&eb->recow);
9181 ret = recow_extent_buffer(root, eb);
9186 while (!list_empty(&delete_items)) {
9187 struct bad_item *bad;
9189 bad = list_first_entry(&delete_items, struct bad_item, list);
9190 list_del_init(&bad->list);
9192 ret = delete_bad_item(root, bad);
9196 if (info->quota_enabled) {
9198 fprintf(stderr, "checking quota groups\n");
9199 err = qgroup_verify_all(info);
9204 if (!list_empty(&root->fs_info->recow_ebs)) {
9205 fprintf(stderr, "Transid errors in file system\n");
9209 print_qgroup_report(0);
9210 if (found_old_backref) { /*
9211 * there was a disk format change when mixed
9212 * backref was in testing tree. The old format
9213 * existed about one week.
9215 printf("\n * Found old mixed backref format. "
9216 "The old format is not supported! *"
9217 "\n * Please mount the FS in readonly mode, "
9218 "backup data and re-format the FS. *\n\n");
9221 printf("found %llu bytes used err is %d\n",
9222 (unsigned long long)bytes_used, ret);
9223 printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
9224 printf("total tree bytes: %llu\n",
9225 (unsigned long long)total_btree_bytes);
9226 printf("total fs tree bytes: %llu\n",
9227 (unsigned long long)total_fs_tree_bytes);
9228 printf("total extent tree bytes: %llu\n",
9229 (unsigned long long)total_extent_tree_bytes);
9230 printf("btree space waste bytes: %llu\n",
9231 (unsigned long long)btree_space_waste);
9232 printf("file data blocks allocated: %llu\n referenced %llu\n",
9233 (unsigned long long)data_bytes_allocated,
9234 (unsigned long long)data_bytes_referenced);
9235 printf("%s\n", PACKAGE_STRING);
9237 free_root_recs_tree(&root_cache);