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 "task-utils.h"
34 #include "transaction.h"
37 #include "free-space-cache.h"
38 #include "free-space-tree.h"
40 #include "qgroup-verify.h"
41 #include "rbtree-utils.h"
49 TASK_NOTHING, /* have to be the last element */
54 enum task_position tp;
56 struct task_info *info;
59 static u64 bytes_used = 0;
60 static u64 total_csum_bytes = 0;
61 static u64 total_btree_bytes = 0;
62 static u64 total_fs_tree_bytes = 0;
63 static u64 total_extent_tree_bytes = 0;
64 static u64 btree_space_waste = 0;
65 static u64 data_bytes_allocated = 0;
66 static u64 data_bytes_referenced = 0;
67 static int found_old_backref = 0;
68 static LIST_HEAD(duplicate_extents);
69 static LIST_HEAD(delete_items);
70 static int no_holes = 0;
71 static int init_extent_tree = 0;
72 static int check_data_csum = 0;
73 static struct btrfs_fs_info *global_info;
74 static struct task_ctx ctx = { 0 };
75 static struct cache_tree *roots_info_cache = NULL;
77 enum btrfs_check_mode {
81 CHECK_MODE_DEFAULT = CHECK_MODE_ORIGINAL
84 static enum btrfs_check_mode check_mode = CHECK_MODE_DEFAULT;
86 struct extent_backref {
87 struct list_head list;
88 unsigned int is_data:1;
89 unsigned int found_extent_tree:1;
90 unsigned int full_backref:1;
91 unsigned int found_ref:1;
92 unsigned int broken:1;
95 static inline struct extent_backref* to_extent_backref(struct list_head *entry)
97 return list_entry(entry, struct extent_backref, list);
100 struct data_backref {
101 struct extent_backref node;
115 static inline struct data_backref* to_data_backref(struct extent_backref *back)
117 return container_of(back, struct data_backref, node);
121 * Much like data_backref, just removed the undetermined members
122 * and change it to use list_head.
123 * During extent scan, it is stored in root->orphan_data_extent.
124 * During fs tree scan, it is then moved to inode_rec->orphan_data_extents.
126 struct orphan_data_extent {
127 struct list_head list;
135 struct tree_backref {
136 struct extent_backref node;
143 static inline struct tree_backref* to_tree_backref(struct extent_backref *back)
145 return container_of(back, struct tree_backref, node);
148 /* Explicit initialization for extent_record::flag_block_full_backref */
149 enum { FLAG_UNSET = 2 };
151 struct extent_record {
152 struct list_head backrefs;
153 struct list_head dups;
154 struct list_head list;
155 struct cache_extent cache;
156 struct btrfs_disk_key parent_key;
161 u64 extent_item_refs;
163 u64 parent_generation;
167 unsigned int flag_block_full_backref:2;
168 unsigned int found_rec:1;
169 unsigned int content_checked:1;
170 unsigned int owner_ref_checked:1;
171 unsigned int is_root:1;
172 unsigned int metadata:1;
173 unsigned int bad_full_backref:1;
174 unsigned int crossing_stripes:1;
175 unsigned int wrong_chunk_type:1;
178 static inline struct extent_record* to_extent_record(struct list_head *entry)
180 return container_of(entry, struct extent_record, list);
183 struct inode_backref {
184 struct list_head list;
185 unsigned int found_dir_item:1;
186 unsigned int found_dir_index:1;
187 unsigned int found_inode_ref:1;
197 static inline struct inode_backref* to_inode_backref(struct list_head *entry)
199 return list_entry(entry, struct inode_backref, list);
202 struct root_item_record {
203 struct list_head list;
210 struct btrfs_key drop_key;
213 #define REF_ERR_NO_DIR_ITEM (1 << 0)
214 #define REF_ERR_NO_DIR_INDEX (1 << 1)
215 #define REF_ERR_NO_INODE_REF (1 << 2)
216 #define REF_ERR_DUP_DIR_ITEM (1 << 3)
217 #define REF_ERR_DUP_DIR_INDEX (1 << 4)
218 #define REF_ERR_DUP_INODE_REF (1 << 5)
219 #define REF_ERR_INDEX_UNMATCH (1 << 6)
220 #define REF_ERR_FILETYPE_UNMATCH (1 << 7)
221 #define REF_ERR_NAME_TOO_LONG (1 << 8) // 100
222 #define REF_ERR_NO_ROOT_REF (1 << 9)
223 #define REF_ERR_NO_ROOT_BACKREF (1 << 10)
224 #define REF_ERR_DUP_ROOT_REF (1 << 11)
225 #define REF_ERR_DUP_ROOT_BACKREF (1 << 12)
227 struct file_extent_hole {
233 struct inode_record {
234 struct list_head backrefs;
235 unsigned int checked:1;
236 unsigned int merging:1;
237 unsigned int found_inode_item:1;
238 unsigned int found_dir_item:1;
239 unsigned int found_file_extent:1;
240 unsigned int found_csum_item:1;
241 unsigned int some_csum_missing:1;
242 unsigned int nodatasum:1;
255 struct rb_root holes;
256 struct list_head orphan_extents;
261 #define I_ERR_NO_INODE_ITEM (1 << 0)
262 #define I_ERR_NO_ORPHAN_ITEM (1 << 1)
263 #define I_ERR_DUP_INODE_ITEM (1 << 2)
264 #define I_ERR_DUP_DIR_INDEX (1 << 3)
265 #define I_ERR_ODD_DIR_ITEM (1 << 4)
266 #define I_ERR_ODD_FILE_EXTENT (1 << 5)
267 #define I_ERR_BAD_FILE_EXTENT (1 << 6)
268 #define I_ERR_FILE_EXTENT_OVERLAP (1 << 7)
269 #define I_ERR_FILE_EXTENT_DISCOUNT (1 << 8) // 100
270 #define I_ERR_DIR_ISIZE_WRONG (1 << 9)
271 #define I_ERR_FILE_NBYTES_WRONG (1 << 10) // 400
272 #define I_ERR_ODD_CSUM_ITEM (1 << 11)
273 #define I_ERR_SOME_CSUM_MISSING (1 << 12)
274 #define I_ERR_LINK_COUNT_WRONG (1 << 13)
275 #define I_ERR_FILE_EXTENT_ORPHAN (1 << 14)
277 struct root_backref {
278 struct list_head list;
279 unsigned int found_dir_item:1;
280 unsigned int found_dir_index:1;
281 unsigned int found_back_ref:1;
282 unsigned int found_forward_ref:1;
283 unsigned int reachable:1;
292 static inline struct root_backref* to_root_backref(struct list_head *entry)
294 return list_entry(entry, struct root_backref, list);
298 struct list_head backrefs;
299 struct cache_extent cache;
300 unsigned int found_root_item:1;
306 struct cache_extent cache;
311 struct cache_extent cache;
312 struct cache_tree root_cache;
313 struct cache_tree inode_cache;
314 struct inode_record *current;
323 struct walk_control {
324 struct cache_tree shared;
325 struct shared_node *nodes[BTRFS_MAX_LEVEL];
331 struct btrfs_key key;
333 struct list_head list;
336 struct extent_entry {
341 struct list_head list;
344 struct root_item_info {
345 /* level of the root */
347 /* number of nodes at this level, must be 1 for a root */
351 struct cache_extent cache_extent;
355 * Error bit for low memory mode check.
357 * Currently no caller cares about it yet. Just internal use for error
360 #define BACKREF_MISSING (1 << 0) /* Backref missing in extent tree */
361 #define BACKREF_MISMATCH (1 << 1) /* Backref exists but does not match */
362 #define BYTES_UNALIGNED (1 << 2) /* Some bytes are not aligned */
363 #define REFERENCER_MISSING (1 << 3) /* Referencer not found */
364 #define REFERENCER_MISMATCH (1 << 4) /* Referenceer found but does not match */
365 #define CROSSING_STRIPE_BOUNDARY (1 << 4) /* For kernel scrub workaround */
366 #define ITEM_SIZE_MISMATCH (1 << 5) /* Bad item size */
367 #define UNKNOWN_TYPE (1 << 6) /* Unknown type */
368 #define ACCOUNTING_MISMATCH (1 << 7) /* Used space accounting error */
369 #define CHUNK_TYPE_MISMATCH (1 << 8)
371 static void *print_status_check(void *p)
373 struct task_ctx *priv = p;
374 const char work_indicator[] = { '.', 'o', 'O', 'o' };
376 static char *task_position_string[] = {
378 "checking free space cache",
382 task_period_start(priv->info, 1000 /* 1s */);
384 if (priv->tp == TASK_NOTHING)
388 printf("%s [%c]\r", task_position_string[priv->tp],
389 work_indicator[count % 4]);
392 task_period_wait(priv->info);
397 static int print_status_return(void *p)
405 static enum btrfs_check_mode parse_check_mode(const char *str)
407 if (strcmp(str, "lowmem") == 0)
408 return CHECK_MODE_LOWMEM;
409 if (strcmp(str, "orig") == 0)
410 return CHECK_MODE_ORIGINAL;
411 if (strcmp(str, "original") == 0)
412 return CHECK_MODE_ORIGINAL;
414 return CHECK_MODE_UNKNOWN;
417 /* Compatible function to allow reuse of old codes */
418 static u64 first_extent_gap(struct rb_root *holes)
420 struct file_extent_hole *hole;
422 if (RB_EMPTY_ROOT(holes))
425 hole = rb_entry(rb_first(holes), struct file_extent_hole, node);
429 static int compare_hole(struct rb_node *node1, struct rb_node *node2)
431 struct file_extent_hole *hole1;
432 struct file_extent_hole *hole2;
434 hole1 = rb_entry(node1, struct file_extent_hole, node);
435 hole2 = rb_entry(node2, struct file_extent_hole, node);
437 if (hole1->start > hole2->start)
439 if (hole1->start < hole2->start)
441 /* Now hole1->start == hole2->start */
442 if (hole1->len >= hole2->len)
444 * Hole 1 will be merge center
445 * Same hole will be merged later
448 /* Hole 2 will be merge center */
453 * Add a hole to the record
455 * This will do hole merge for copy_file_extent_holes(),
456 * which will ensure there won't be continuous holes.
458 static int add_file_extent_hole(struct rb_root *holes,
461 struct file_extent_hole *hole;
462 struct file_extent_hole *prev = NULL;
463 struct file_extent_hole *next = NULL;
465 hole = malloc(sizeof(*hole));
470 /* Since compare will not return 0, no -EEXIST will happen */
471 rb_insert(holes, &hole->node, compare_hole);
473 /* simple merge with previous hole */
474 if (rb_prev(&hole->node))
475 prev = rb_entry(rb_prev(&hole->node), struct file_extent_hole,
477 if (prev && prev->start + prev->len >= hole->start) {
478 hole->len = hole->start + hole->len - prev->start;
479 hole->start = prev->start;
480 rb_erase(&prev->node, holes);
485 /* iterate merge with next holes */
487 if (!rb_next(&hole->node))
489 next = rb_entry(rb_next(&hole->node), struct file_extent_hole,
491 if (hole->start + hole->len >= next->start) {
492 if (hole->start + hole->len <= next->start + next->len)
493 hole->len = next->start + next->len -
495 rb_erase(&next->node, holes);
504 static int compare_hole_range(struct rb_node *node, void *data)
506 struct file_extent_hole *hole;
509 hole = (struct file_extent_hole *)data;
512 hole = rb_entry(node, struct file_extent_hole, node);
513 if (start < hole->start)
515 if (start >= hole->start && start < hole->start + hole->len)
521 * Delete a hole in the record
523 * This will do the hole split and is much restrict than add.
525 static int del_file_extent_hole(struct rb_root *holes,
528 struct file_extent_hole *hole;
529 struct file_extent_hole tmp;
534 struct rb_node *node;
541 node = rb_search(holes, &tmp, compare_hole_range, NULL);
544 hole = rb_entry(node, struct file_extent_hole, node);
545 if (start + len > hole->start + hole->len)
549 * Now there will be no overlap, delete the hole and re-add the
550 * split(s) if they exists.
552 if (start > hole->start) {
553 prev_start = hole->start;
554 prev_len = start - hole->start;
557 if (hole->start + hole->len > start + len) {
558 next_start = start + len;
559 next_len = hole->start + hole->len - start - len;
562 rb_erase(node, holes);
565 ret = add_file_extent_hole(holes, prev_start, prev_len);
570 ret = add_file_extent_hole(holes, next_start, next_len);
577 static int copy_file_extent_holes(struct rb_root *dst,
580 struct file_extent_hole *hole;
581 struct rb_node *node;
584 node = rb_first(src);
586 hole = rb_entry(node, struct file_extent_hole, node);
587 ret = add_file_extent_hole(dst, hole->start, hole->len);
590 node = rb_next(node);
595 static void free_file_extent_holes(struct rb_root *holes)
597 struct rb_node *node;
598 struct file_extent_hole *hole;
600 node = rb_first(holes);
602 hole = rb_entry(node, struct file_extent_hole, node);
603 rb_erase(node, holes);
605 node = rb_first(holes);
609 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info);
611 static void record_root_in_trans(struct btrfs_trans_handle *trans,
612 struct btrfs_root *root)
614 if (root->last_trans != trans->transid) {
615 root->track_dirty = 1;
616 root->last_trans = trans->transid;
617 root->commit_root = root->node;
618 extent_buffer_get(root->node);
622 static u8 imode_to_type(u32 imode)
625 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
626 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
627 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
628 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
629 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
630 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
631 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
632 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
635 return btrfs_type_by_mode[(imode & S_IFMT) >> S_SHIFT];
639 static int device_record_compare(struct rb_node *node1, struct rb_node *node2)
641 struct device_record *rec1;
642 struct device_record *rec2;
644 rec1 = rb_entry(node1, struct device_record, node);
645 rec2 = rb_entry(node2, struct device_record, node);
646 if (rec1->devid > rec2->devid)
648 else if (rec1->devid < rec2->devid)
654 static struct inode_record *clone_inode_rec(struct inode_record *orig_rec)
656 struct inode_record *rec;
657 struct inode_backref *backref;
658 struct inode_backref *orig;
659 struct inode_backref *tmp;
660 struct orphan_data_extent *src_orphan;
661 struct orphan_data_extent *dst_orphan;
666 rec = malloc(sizeof(*rec));
668 return ERR_PTR(-ENOMEM);
669 memcpy(rec, orig_rec, sizeof(*rec));
671 INIT_LIST_HEAD(&rec->backrefs);
672 INIT_LIST_HEAD(&rec->orphan_extents);
673 rec->holes = RB_ROOT;
675 list_for_each_entry(orig, &orig_rec->backrefs, list) {
676 size = sizeof(*orig) + orig->namelen + 1;
677 backref = malloc(size);
682 memcpy(backref, orig, size);
683 list_add_tail(&backref->list, &rec->backrefs);
685 list_for_each_entry(src_orphan, &orig_rec->orphan_extents, list) {
686 dst_orphan = malloc(sizeof(*dst_orphan));
691 memcpy(dst_orphan, src_orphan, sizeof(*src_orphan));
692 list_add_tail(&dst_orphan->list, &rec->orphan_extents);
694 ret = copy_file_extent_holes(&rec->holes, &orig_rec->holes);
701 rb = rb_first(&rec->holes);
703 struct file_extent_hole *hole;
705 hole = rb_entry(rb, struct file_extent_hole, node);
711 if (!list_empty(&rec->backrefs))
712 list_for_each_entry_safe(orig, tmp, &rec->backrefs, list) {
713 list_del(&orig->list);
717 if (!list_empty(&rec->orphan_extents))
718 list_for_each_entry_safe(orig, tmp, &rec->orphan_extents, list) {
719 list_del(&orig->list);
728 static void print_orphan_data_extents(struct list_head *orphan_extents,
731 struct orphan_data_extent *orphan;
733 if (list_empty(orphan_extents))
735 printf("The following data extent is lost in tree %llu:\n",
737 list_for_each_entry(orphan, orphan_extents, list) {
738 printf("\tinode: %llu, offset:%llu, disk_bytenr: %llu, disk_len: %llu\n",
739 orphan->objectid, orphan->offset, orphan->disk_bytenr,
744 static void print_inode_error(struct btrfs_root *root, struct inode_record *rec)
746 u64 root_objectid = root->root_key.objectid;
747 int errors = rec->errors;
751 /* reloc root errors, we print its corresponding fs root objectid*/
752 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
753 root_objectid = root->root_key.offset;
754 fprintf(stderr, "reloc");
756 fprintf(stderr, "root %llu inode %llu errors %x",
757 (unsigned long long) root_objectid,
758 (unsigned long long) rec->ino, rec->errors);
760 if (errors & I_ERR_NO_INODE_ITEM)
761 fprintf(stderr, ", no inode item");
762 if (errors & I_ERR_NO_ORPHAN_ITEM)
763 fprintf(stderr, ", no orphan item");
764 if (errors & I_ERR_DUP_INODE_ITEM)
765 fprintf(stderr, ", dup inode item");
766 if (errors & I_ERR_DUP_DIR_INDEX)
767 fprintf(stderr, ", dup dir index");
768 if (errors & I_ERR_ODD_DIR_ITEM)
769 fprintf(stderr, ", odd dir item");
770 if (errors & I_ERR_ODD_FILE_EXTENT)
771 fprintf(stderr, ", odd file extent");
772 if (errors & I_ERR_BAD_FILE_EXTENT)
773 fprintf(stderr, ", bad file extent");
774 if (errors & I_ERR_FILE_EXTENT_OVERLAP)
775 fprintf(stderr, ", file extent overlap");
776 if (errors & I_ERR_FILE_EXTENT_DISCOUNT)
777 fprintf(stderr, ", file extent discount");
778 if (errors & I_ERR_DIR_ISIZE_WRONG)
779 fprintf(stderr, ", dir isize wrong");
780 if (errors & I_ERR_FILE_NBYTES_WRONG)
781 fprintf(stderr, ", nbytes wrong");
782 if (errors & I_ERR_ODD_CSUM_ITEM)
783 fprintf(stderr, ", odd csum item");
784 if (errors & I_ERR_SOME_CSUM_MISSING)
785 fprintf(stderr, ", some csum missing");
786 if (errors & I_ERR_LINK_COUNT_WRONG)
787 fprintf(stderr, ", link count wrong");
788 if (errors & I_ERR_FILE_EXTENT_ORPHAN)
789 fprintf(stderr, ", orphan file extent");
790 fprintf(stderr, "\n");
791 /* Print the orphan extents if needed */
792 if (errors & I_ERR_FILE_EXTENT_ORPHAN)
793 print_orphan_data_extents(&rec->orphan_extents, root->objectid);
795 /* Print the holes if needed */
796 if (errors & I_ERR_FILE_EXTENT_DISCOUNT) {
797 struct file_extent_hole *hole;
798 struct rb_node *node;
801 node = rb_first(&rec->holes);
802 fprintf(stderr, "Found file extent holes:\n");
805 hole = rb_entry(node, struct file_extent_hole, node);
806 fprintf(stderr, "\tstart: %llu, len: %llu\n",
807 hole->start, hole->len);
808 node = rb_next(node);
811 fprintf(stderr, "\tstart: 0, len: %llu\n",
812 round_up(rec->isize, root->sectorsize));
816 static void print_ref_error(int errors)
818 if (errors & REF_ERR_NO_DIR_ITEM)
819 fprintf(stderr, ", no dir item");
820 if (errors & REF_ERR_NO_DIR_INDEX)
821 fprintf(stderr, ", no dir index");
822 if (errors & REF_ERR_NO_INODE_REF)
823 fprintf(stderr, ", no inode ref");
824 if (errors & REF_ERR_DUP_DIR_ITEM)
825 fprintf(stderr, ", dup dir item");
826 if (errors & REF_ERR_DUP_DIR_INDEX)
827 fprintf(stderr, ", dup dir index");
828 if (errors & REF_ERR_DUP_INODE_REF)
829 fprintf(stderr, ", dup inode ref");
830 if (errors & REF_ERR_INDEX_UNMATCH)
831 fprintf(stderr, ", index mismatch");
832 if (errors & REF_ERR_FILETYPE_UNMATCH)
833 fprintf(stderr, ", filetype mismatch");
834 if (errors & REF_ERR_NAME_TOO_LONG)
835 fprintf(stderr, ", name too long");
836 if (errors & REF_ERR_NO_ROOT_REF)
837 fprintf(stderr, ", no root ref");
838 if (errors & REF_ERR_NO_ROOT_BACKREF)
839 fprintf(stderr, ", no root backref");
840 if (errors & REF_ERR_DUP_ROOT_REF)
841 fprintf(stderr, ", dup root ref");
842 if (errors & REF_ERR_DUP_ROOT_BACKREF)
843 fprintf(stderr, ", dup root backref");
844 fprintf(stderr, "\n");
847 static struct inode_record *get_inode_rec(struct cache_tree *inode_cache,
850 struct ptr_node *node;
851 struct cache_extent *cache;
852 struct inode_record *rec = NULL;
855 cache = lookup_cache_extent(inode_cache, ino, 1);
857 node = container_of(cache, struct ptr_node, cache);
859 if (mod && rec->refs > 1) {
860 node->data = clone_inode_rec(rec);
861 if (IS_ERR(node->data))
867 rec = calloc(1, sizeof(*rec));
869 return ERR_PTR(-ENOMEM);
871 rec->extent_start = (u64)-1;
873 INIT_LIST_HEAD(&rec->backrefs);
874 INIT_LIST_HEAD(&rec->orphan_extents);
875 rec->holes = RB_ROOT;
877 node = malloc(sizeof(*node));
880 return ERR_PTR(-ENOMEM);
882 node->cache.start = ino;
883 node->cache.size = 1;
886 if (ino == BTRFS_FREE_INO_OBJECTID)
889 ret = insert_cache_extent(inode_cache, &node->cache);
891 return ERR_PTR(-EEXIST);
896 static void free_orphan_data_extents(struct list_head *orphan_extents)
898 struct orphan_data_extent *orphan;
900 while (!list_empty(orphan_extents)) {
901 orphan = list_entry(orphan_extents->next,
902 struct orphan_data_extent, list);
903 list_del(&orphan->list);
908 static void free_inode_rec(struct inode_record *rec)
910 struct inode_backref *backref;
915 while (!list_empty(&rec->backrefs)) {
916 backref = to_inode_backref(rec->backrefs.next);
917 list_del(&backref->list);
920 free_orphan_data_extents(&rec->orphan_extents);
921 free_file_extent_holes(&rec->holes);
925 static int can_free_inode_rec(struct inode_record *rec)
927 if (!rec->errors && rec->checked && rec->found_inode_item &&
928 rec->nlink == rec->found_link && list_empty(&rec->backrefs))
933 static void maybe_free_inode_rec(struct cache_tree *inode_cache,
934 struct inode_record *rec)
936 struct cache_extent *cache;
937 struct inode_backref *tmp, *backref;
938 struct ptr_node *node;
941 if (!rec->found_inode_item)
944 filetype = imode_to_type(rec->imode);
945 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
946 if (backref->found_dir_item && backref->found_dir_index) {
947 if (backref->filetype != filetype)
948 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
949 if (!backref->errors && backref->found_inode_ref &&
950 rec->nlink == rec->found_link) {
951 list_del(&backref->list);
957 if (!rec->checked || rec->merging)
960 if (S_ISDIR(rec->imode)) {
961 if (rec->found_size != rec->isize)
962 rec->errors |= I_ERR_DIR_ISIZE_WRONG;
963 if (rec->found_file_extent)
964 rec->errors |= I_ERR_ODD_FILE_EXTENT;
965 } else if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
966 if (rec->found_dir_item)
967 rec->errors |= I_ERR_ODD_DIR_ITEM;
968 if (rec->found_size != rec->nbytes)
969 rec->errors |= I_ERR_FILE_NBYTES_WRONG;
970 if (rec->nlink > 0 && !no_holes &&
971 (rec->extent_end < rec->isize ||
972 first_extent_gap(&rec->holes) < rec->isize))
973 rec->errors |= I_ERR_FILE_EXTENT_DISCOUNT;
976 if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
977 if (rec->found_csum_item && rec->nodatasum)
978 rec->errors |= I_ERR_ODD_CSUM_ITEM;
979 if (rec->some_csum_missing && !rec->nodatasum)
980 rec->errors |= I_ERR_SOME_CSUM_MISSING;
983 BUG_ON(rec->refs != 1);
984 if (can_free_inode_rec(rec)) {
985 cache = lookup_cache_extent(inode_cache, rec->ino, 1);
986 node = container_of(cache, struct ptr_node, cache);
987 BUG_ON(node->data != rec);
988 remove_cache_extent(inode_cache, &node->cache);
994 static int check_orphan_item(struct btrfs_root *root, u64 ino)
996 struct btrfs_path path;
997 struct btrfs_key key;
1000 key.objectid = BTRFS_ORPHAN_OBJECTID;
1001 key.type = BTRFS_ORPHAN_ITEM_KEY;
1004 btrfs_init_path(&path);
1005 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1006 btrfs_release_path(&path);
1012 static int process_inode_item(struct extent_buffer *eb,
1013 int slot, struct btrfs_key *key,
1014 struct shared_node *active_node)
1016 struct inode_record *rec;
1017 struct btrfs_inode_item *item;
1019 rec = active_node->current;
1020 BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1021 if (rec->found_inode_item) {
1022 rec->errors |= I_ERR_DUP_INODE_ITEM;
1025 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
1026 rec->nlink = btrfs_inode_nlink(eb, item);
1027 rec->isize = btrfs_inode_size(eb, item);
1028 rec->nbytes = btrfs_inode_nbytes(eb, item);
1029 rec->imode = btrfs_inode_mode(eb, item);
1030 if (btrfs_inode_flags(eb, item) & BTRFS_INODE_NODATASUM)
1032 rec->found_inode_item = 1;
1033 if (rec->nlink == 0)
1034 rec->errors |= I_ERR_NO_ORPHAN_ITEM;
1035 maybe_free_inode_rec(&active_node->inode_cache, rec);
1039 static struct inode_backref *get_inode_backref(struct inode_record *rec,
1041 int namelen, u64 dir)
1043 struct inode_backref *backref;
1045 list_for_each_entry(backref, &rec->backrefs, list) {
1046 if (rec->ino == BTRFS_MULTIPLE_OBJECTIDS)
1048 if (backref->dir != dir || backref->namelen != namelen)
1050 if (memcmp(name, backref->name, namelen))
1055 backref = malloc(sizeof(*backref) + namelen + 1);
1058 memset(backref, 0, sizeof(*backref));
1060 backref->namelen = namelen;
1061 memcpy(backref->name, name, namelen);
1062 backref->name[namelen] = '\0';
1063 list_add_tail(&backref->list, &rec->backrefs);
1067 static int add_inode_backref(struct cache_tree *inode_cache,
1068 u64 ino, u64 dir, u64 index,
1069 const char *name, int namelen,
1070 u8 filetype, u8 itemtype, int errors)
1072 struct inode_record *rec;
1073 struct inode_backref *backref;
1075 rec = get_inode_rec(inode_cache, ino, 1);
1076 BUG_ON(IS_ERR(rec));
1077 backref = get_inode_backref(rec, name, namelen, dir);
1080 backref->errors |= errors;
1081 if (itemtype == BTRFS_DIR_INDEX_KEY) {
1082 if (backref->found_dir_index)
1083 backref->errors |= REF_ERR_DUP_DIR_INDEX;
1084 if (backref->found_inode_ref && backref->index != index)
1085 backref->errors |= REF_ERR_INDEX_UNMATCH;
1086 if (backref->found_dir_item && backref->filetype != filetype)
1087 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
1089 backref->index = index;
1090 backref->filetype = filetype;
1091 backref->found_dir_index = 1;
1092 } else if (itemtype == BTRFS_DIR_ITEM_KEY) {
1094 if (backref->found_dir_item)
1095 backref->errors |= REF_ERR_DUP_DIR_ITEM;
1096 if (backref->found_dir_index && backref->filetype != filetype)
1097 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
1099 backref->filetype = filetype;
1100 backref->found_dir_item = 1;
1101 } else if ((itemtype == BTRFS_INODE_REF_KEY) ||
1102 (itemtype == BTRFS_INODE_EXTREF_KEY)) {
1103 if (backref->found_inode_ref)
1104 backref->errors |= REF_ERR_DUP_INODE_REF;
1105 if (backref->found_dir_index && backref->index != index)
1106 backref->errors |= REF_ERR_INDEX_UNMATCH;
1108 backref->index = index;
1110 backref->ref_type = itemtype;
1111 backref->found_inode_ref = 1;
1116 maybe_free_inode_rec(inode_cache, rec);
1120 static int merge_inode_recs(struct inode_record *src, struct inode_record *dst,
1121 struct cache_tree *dst_cache)
1123 struct inode_backref *backref;
1128 list_for_each_entry(backref, &src->backrefs, list) {
1129 if (backref->found_dir_index) {
1130 add_inode_backref(dst_cache, dst->ino, backref->dir,
1131 backref->index, backref->name,
1132 backref->namelen, backref->filetype,
1133 BTRFS_DIR_INDEX_KEY, backref->errors);
1135 if (backref->found_dir_item) {
1137 add_inode_backref(dst_cache, dst->ino,
1138 backref->dir, 0, backref->name,
1139 backref->namelen, backref->filetype,
1140 BTRFS_DIR_ITEM_KEY, backref->errors);
1142 if (backref->found_inode_ref) {
1143 add_inode_backref(dst_cache, dst->ino,
1144 backref->dir, backref->index,
1145 backref->name, backref->namelen, 0,
1146 backref->ref_type, backref->errors);
1150 if (src->found_dir_item)
1151 dst->found_dir_item = 1;
1152 if (src->found_file_extent)
1153 dst->found_file_extent = 1;
1154 if (src->found_csum_item)
1155 dst->found_csum_item = 1;
1156 if (src->some_csum_missing)
1157 dst->some_csum_missing = 1;
1158 if (first_extent_gap(&dst->holes) > first_extent_gap(&src->holes)) {
1159 ret = copy_file_extent_holes(&dst->holes, &src->holes);
1164 BUG_ON(src->found_link < dir_count);
1165 dst->found_link += src->found_link - dir_count;
1166 dst->found_size += src->found_size;
1167 if (src->extent_start != (u64)-1) {
1168 if (dst->extent_start == (u64)-1) {
1169 dst->extent_start = src->extent_start;
1170 dst->extent_end = src->extent_end;
1172 if (dst->extent_end > src->extent_start)
1173 dst->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1174 else if (dst->extent_end < src->extent_start) {
1175 ret = add_file_extent_hole(&dst->holes,
1177 src->extent_start - dst->extent_end);
1179 if (dst->extent_end < src->extent_end)
1180 dst->extent_end = src->extent_end;
1184 dst->errors |= src->errors;
1185 if (src->found_inode_item) {
1186 if (!dst->found_inode_item) {
1187 dst->nlink = src->nlink;
1188 dst->isize = src->isize;
1189 dst->nbytes = src->nbytes;
1190 dst->imode = src->imode;
1191 dst->nodatasum = src->nodatasum;
1192 dst->found_inode_item = 1;
1194 dst->errors |= I_ERR_DUP_INODE_ITEM;
1202 static int splice_shared_node(struct shared_node *src_node,
1203 struct shared_node *dst_node)
1205 struct cache_extent *cache;
1206 struct ptr_node *node, *ins;
1207 struct cache_tree *src, *dst;
1208 struct inode_record *rec, *conflict;
1209 u64 current_ino = 0;
1213 if (--src_node->refs == 0)
1215 if (src_node->current)
1216 current_ino = src_node->current->ino;
1218 src = &src_node->root_cache;
1219 dst = &dst_node->root_cache;
1221 cache = search_cache_extent(src, 0);
1223 node = container_of(cache, struct ptr_node, cache);
1225 cache = next_cache_extent(cache);
1228 remove_cache_extent(src, &node->cache);
1231 ins = malloc(sizeof(*ins));
1233 ins->cache.start = node->cache.start;
1234 ins->cache.size = node->cache.size;
1238 ret = insert_cache_extent(dst, &ins->cache);
1239 if (ret == -EEXIST) {
1240 conflict = get_inode_rec(dst, rec->ino, 1);
1241 BUG_ON(IS_ERR(conflict));
1242 merge_inode_recs(rec, conflict, dst);
1244 conflict->checked = 1;
1245 if (dst_node->current == conflict)
1246 dst_node->current = NULL;
1248 maybe_free_inode_rec(dst, conflict);
1249 free_inode_rec(rec);
1256 if (src == &src_node->root_cache) {
1257 src = &src_node->inode_cache;
1258 dst = &dst_node->inode_cache;
1262 if (current_ino > 0 && (!dst_node->current ||
1263 current_ino > dst_node->current->ino)) {
1264 if (dst_node->current) {
1265 dst_node->current->checked = 1;
1266 maybe_free_inode_rec(dst, dst_node->current);
1268 dst_node->current = get_inode_rec(dst, current_ino, 1);
1269 BUG_ON(IS_ERR(dst_node->current));
1274 static void free_inode_ptr(struct cache_extent *cache)
1276 struct ptr_node *node;
1277 struct inode_record *rec;
1279 node = container_of(cache, struct ptr_node, cache);
1281 free_inode_rec(rec);
1285 FREE_EXTENT_CACHE_BASED_TREE(inode_recs, free_inode_ptr);
1287 static struct shared_node *find_shared_node(struct cache_tree *shared,
1290 struct cache_extent *cache;
1291 struct shared_node *node;
1293 cache = lookup_cache_extent(shared, bytenr, 1);
1295 node = container_of(cache, struct shared_node, cache);
1301 static int add_shared_node(struct cache_tree *shared, u64 bytenr, u32 refs)
1304 struct shared_node *node;
1306 node = calloc(1, sizeof(*node));
1309 node->cache.start = bytenr;
1310 node->cache.size = 1;
1311 cache_tree_init(&node->root_cache);
1312 cache_tree_init(&node->inode_cache);
1315 ret = insert_cache_extent(shared, &node->cache);
1320 static int enter_shared_node(struct btrfs_root *root, u64 bytenr, u32 refs,
1321 struct walk_control *wc, int level)
1323 struct shared_node *node;
1324 struct shared_node *dest;
1327 if (level == wc->active_node)
1330 BUG_ON(wc->active_node <= level);
1331 node = find_shared_node(&wc->shared, bytenr);
1333 ret = add_shared_node(&wc->shared, bytenr, refs);
1335 node = find_shared_node(&wc->shared, bytenr);
1336 wc->nodes[level] = node;
1337 wc->active_node = level;
1341 if (wc->root_level == wc->active_node &&
1342 btrfs_root_refs(&root->root_item) == 0) {
1343 if (--node->refs == 0) {
1344 free_inode_recs_tree(&node->root_cache);
1345 free_inode_recs_tree(&node->inode_cache);
1346 remove_cache_extent(&wc->shared, &node->cache);
1352 dest = wc->nodes[wc->active_node];
1353 splice_shared_node(node, dest);
1354 if (node->refs == 0) {
1355 remove_cache_extent(&wc->shared, &node->cache);
1361 static int leave_shared_node(struct btrfs_root *root,
1362 struct walk_control *wc, int level)
1364 struct shared_node *node;
1365 struct shared_node *dest;
1368 if (level == wc->root_level)
1371 for (i = level + 1; i < BTRFS_MAX_LEVEL; i++) {
1375 BUG_ON(i >= BTRFS_MAX_LEVEL);
1377 node = wc->nodes[wc->active_node];
1378 wc->nodes[wc->active_node] = NULL;
1379 wc->active_node = i;
1381 dest = wc->nodes[wc->active_node];
1382 if (wc->active_node < wc->root_level ||
1383 btrfs_root_refs(&root->root_item) > 0) {
1384 BUG_ON(node->refs <= 1);
1385 splice_shared_node(node, dest);
1387 BUG_ON(node->refs < 2);
1396 * 1 - if the root with id child_root_id is a child of root parent_root_id
1397 * 0 - if the root child_root_id isn't a child of the root parent_root_id but
1398 * has other root(s) as parent(s)
1399 * 2 - if the root child_root_id doesn't have any parent roots
1401 static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
1404 struct btrfs_path path;
1405 struct btrfs_key key;
1406 struct extent_buffer *leaf;
1410 btrfs_init_path(&path);
1412 key.objectid = parent_root_id;
1413 key.type = BTRFS_ROOT_REF_KEY;
1414 key.offset = child_root_id;
1415 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
1419 btrfs_release_path(&path);
1423 key.objectid = child_root_id;
1424 key.type = BTRFS_ROOT_BACKREF_KEY;
1426 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
1432 leaf = path.nodes[0];
1433 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1434 ret = btrfs_next_leaf(root->fs_info->tree_root, &path);
1437 leaf = path.nodes[0];
1440 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1441 if (key.objectid != child_root_id ||
1442 key.type != BTRFS_ROOT_BACKREF_KEY)
1447 if (key.offset == parent_root_id) {
1448 btrfs_release_path(&path);
1455 btrfs_release_path(&path);
1458 return has_parent ? 0 : 2;
1461 static int process_dir_item(struct btrfs_root *root,
1462 struct extent_buffer *eb,
1463 int slot, struct btrfs_key *key,
1464 struct shared_node *active_node)
1474 struct btrfs_dir_item *di;
1475 struct inode_record *rec;
1476 struct cache_tree *root_cache;
1477 struct cache_tree *inode_cache;
1478 struct btrfs_key location;
1479 char namebuf[BTRFS_NAME_LEN];
1481 root_cache = &active_node->root_cache;
1482 inode_cache = &active_node->inode_cache;
1483 rec = active_node->current;
1484 rec->found_dir_item = 1;
1486 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1487 total = btrfs_item_size_nr(eb, slot);
1488 while (cur < total) {
1490 btrfs_dir_item_key_to_cpu(eb, di, &location);
1491 name_len = btrfs_dir_name_len(eb, di);
1492 data_len = btrfs_dir_data_len(eb, di);
1493 filetype = btrfs_dir_type(eb, di);
1495 rec->found_size += name_len;
1496 if (name_len <= BTRFS_NAME_LEN) {
1500 len = BTRFS_NAME_LEN;
1501 error = REF_ERR_NAME_TOO_LONG;
1503 read_extent_buffer(eb, namebuf, (unsigned long)(di + 1), len);
1505 if (location.type == BTRFS_INODE_ITEM_KEY) {
1506 add_inode_backref(inode_cache, location.objectid,
1507 key->objectid, key->offset, namebuf,
1508 len, filetype, key->type, error);
1509 } else if (location.type == BTRFS_ROOT_ITEM_KEY) {
1510 add_inode_backref(root_cache, location.objectid,
1511 key->objectid, key->offset,
1512 namebuf, len, filetype,
1515 fprintf(stderr, "invalid location in dir item %u\n",
1517 add_inode_backref(inode_cache, BTRFS_MULTIPLE_OBJECTIDS,
1518 key->objectid, key->offset, namebuf,
1519 len, filetype, key->type, error);
1522 len = sizeof(*di) + name_len + data_len;
1523 di = (struct btrfs_dir_item *)((char *)di + len);
1526 if (key->type == BTRFS_DIR_INDEX_KEY && nritems > 1)
1527 rec->errors |= I_ERR_DUP_DIR_INDEX;
1532 static int process_inode_ref(struct extent_buffer *eb,
1533 int slot, struct btrfs_key *key,
1534 struct shared_node *active_node)
1542 struct cache_tree *inode_cache;
1543 struct btrfs_inode_ref *ref;
1544 char namebuf[BTRFS_NAME_LEN];
1546 inode_cache = &active_node->inode_cache;
1548 ref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1549 total = btrfs_item_size_nr(eb, slot);
1550 while (cur < total) {
1551 name_len = btrfs_inode_ref_name_len(eb, ref);
1552 index = btrfs_inode_ref_index(eb, ref);
1553 if (name_len <= BTRFS_NAME_LEN) {
1557 len = BTRFS_NAME_LEN;
1558 error = REF_ERR_NAME_TOO_LONG;
1560 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
1561 add_inode_backref(inode_cache, key->objectid, key->offset,
1562 index, namebuf, len, 0, key->type, error);
1564 len = sizeof(*ref) + name_len;
1565 ref = (struct btrfs_inode_ref *)((char *)ref + len);
1571 static int process_inode_extref(struct extent_buffer *eb,
1572 int slot, struct btrfs_key *key,
1573 struct shared_node *active_node)
1582 struct cache_tree *inode_cache;
1583 struct btrfs_inode_extref *extref;
1584 char namebuf[BTRFS_NAME_LEN];
1586 inode_cache = &active_node->inode_cache;
1588 extref = btrfs_item_ptr(eb, slot, struct btrfs_inode_extref);
1589 total = btrfs_item_size_nr(eb, slot);
1590 while (cur < total) {
1591 name_len = btrfs_inode_extref_name_len(eb, extref);
1592 index = btrfs_inode_extref_index(eb, extref);
1593 parent = btrfs_inode_extref_parent(eb, extref);
1594 if (name_len <= BTRFS_NAME_LEN) {
1598 len = BTRFS_NAME_LEN;
1599 error = REF_ERR_NAME_TOO_LONG;
1601 read_extent_buffer(eb, namebuf,
1602 (unsigned long)(extref + 1), len);
1603 add_inode_backref(inode_cache, key->objectid, parent,
1604 index, namebuf, len, 0, key->type, error);
1606 len = sizeof(*extref) + name_len;
1607 extref = (struct btrfs_inode_extref *)((char *)extref + len);
1614 static int count_csum_range(struct btrfs_root *root, u64 start,
1615 u64 len, u64 *found)
1617 struct btrfs_key key;
1618 struct btrfs_path path;
1619 struct extent_buffer *leaf;
1624 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
1626 btrfs_init_path(&path);
1628 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1630 key.type = BTRFS_EXTENT_CSUM_KEY;
1632 ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
1636 if (ret > 0 && path.slots[0] > 0) {
1637 leaf = path.nodes[0];
1638 btrfs_item_key_to_cpu(leaf, &key, path.slots[0] - 1);
1639 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
1640 key.type == BTRFS_EXTENT_CSUM_KEY)
1645 leaf = path.nodes[0];
1646 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1647 ret = btrfs_next_leaf(root->fs_info->csum_root, &path);
1652 leaf = path.nodes[0];
1655 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1656 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1657 key.type != BTRFS_EXTENT_CSUM_KEY)
1660 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1661 if (key.offset >= start + len)
1664 if (key.offset > start)
1667 size = btrfs_item_size_nr(leaf, path.slots[0]);
1668 csum_end = key.offset + (size / csum_size) * root->sectorsize;
1669 if (csum_end > start) {
1670 size = min(csum_end - start, len);
1679 btrfs_release_path(&path);
1685 static int process_file_extent(struct btrfs_root *root,
1686 struct extent_buffer *eb,
1687 int slot, struct btrfs_key *key,
1688 struct shared_node *active_node)
1690 struct inode_record *rec;
1691 struct btrfs_file_extent_item *fi;
1693 u64 disk_bytenr = 0;
1694 u64 extent_offset = 0;
1695 u64 mask = root->sectorsize - 1;
1699 rec = active_node->current;
1700 BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1701 rec->found_file_extent = 1;
1703 if (rec->extent_start == (u64)-1) {
1704 rec->extent_start = key->offset;
1705 rec->extent_end = key->offset;
1708 if (rec->extent_end > key->offset)
1709 rec->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1710 else if (rec->extent_end < key->offset) {
1711 ret = add_file_extent_hole(&rec->holes, rec->extent_end,
1712 key->offset - rec->extent_end);
1717 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
1718 extent_type = btrfs_file_extent_type(eb, fi);
1720 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1721 num_bytes = btrfs_file_extent_inline_len(eb, slot, fi);
1723 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1724 rec->found_size += num_bytes;
1725 num_bytes = (num_bytes + mask) & ~mask;
1726 } else if (extent_type == BTRFS_FILE_EXTENT_REG ||
1727 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1728 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1729 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1730 extent_offset = btrfs_file_extent_offset(eb, fi);
1731 if (num_bytes == 0 || (num_bytes & mask))
1732 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1733 if (num_bytes + extent_offset >
1734 btrfs_file_extent_ram_bytes(eb, fi))
1735 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1736 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC &&
1737 (btrfs_file_extent_compression(eb, fi) ||
1738 btrfs_file_extent_encryption(eb, fi) ||
1739 btrfs_file_extent_other_encoding(eb, fi)))
1740 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1741 if (disk_bytenr > 0)
1742 rec->found_size += num_bytes;
1744 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1746 rec->extent_end = key->offset + num_bytes;
1749 * The data reloc tree will copy full extents into its inode and then
1750 * copy the corresponding csums. Because the extent it copied could be
1751 * a preallocated extent that hasn't been written to yet there may be no
1752 * csums to copy, ergo we won't have csums for our file extent. This is
1753 * ok so just don't bother checking csums if the inode belongs to the
1756 if (disk_bytenr > 0 &&
1757 btrfs_header_owner(eb) != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1759 if (btrfs_file_extent_compression(eb, fi))
1760 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1762 disk_bytenr += extent_offset;
1764 ret = count_csum_range(root, disk_bytenr, num_bytes, &found);
1767 if (extent_type == BTRFS_FILE_EXTENT_REG) {
1769 rec->found_csum_item = 1;
1770 if (found < num_bytes)
1771 rec->some_csum_missing = 1;
1772 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1774 rec->errors |= I_ERR_ODD_CSUM_ITEM;
1780 static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
1781 struct walk_control *wc)
1783 struct btrfs_key key;
1787 struct cache_tree *inode_cache;
1788 struct shared_node *active_node;
1790 if (wc->root_level == wc->active_node &&
1791 btrfs_root_refs(&root->root_item) == 0)
1794 active_node = wc->nodes[wc->active_node];
1795 inode_cache = &active_node->inode_cache;
1796 nritems = btrfs_header_nritems(eb);
1797 for (i = 0; i < nritems; i++) {
1798 btrfs_item_key_to_cpu(eb, &key, i);
1800 if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
1802 if (key.type == BTRFS_ORPHAN_ITEM_KEY)
1805 if (active_node->current == NULL ||
1806 active_node->current->ino < key.objectid) {
1807 if (active_node->current) {
1808 active_node->current->checked = 1;
1809 maybe_free_inode_rec(inode_cache,
1810 active_node->current);
1812 active_node->current = get_inode_rec(inode_cache,
1814 BUG_ON(IS_ERR(active_node->current));
1817 case BTRFS_DIR_ITEM_KEY:
1818 case BTRFS_DIR_INDEX_KEY:
1819 ret = process_dir_item(root, eb, i, &key, active_node);
1821 case BTRFS_INODE_REF_KEY:
1822 ret = process_inode_ref(eb, i, &key, active_node);
1824 case BTRFS_INODE_EXTREF_KEY:
1825 ret = process_inode_extref(eb, i, &key, active_node);
1827 case BTRFS_INODE_ITEM_KEY:
1828 ret = process_inode_item(eb, i, &key, active_node);
1830 case BTRFS_EXTENT_DATA_KEY:
1831 ret = process_file_extent(root, eb, i, &key,
1841 static void reada_walk_down(struct btrfs_root *root,
1842 struct extent_buffer *node, int slot)
1851 level = btrfs_header_level(node);
1855 nritems = btrfs_header_nritems(node);
1856 blocksize = root->nodesize;
1857 for (i = slot; i < nritems; i++) {
1858 bytenr = btrfs_node_blockptr(node, i);
1859 ptr_gen = btrfs_node_ptr_generation(node, i);
1860 readahead_tree_block(root, bytenr, blocksize, ptr_gen);
1865 * Check the child node/leaf by the following condition:
1866 * 1. the first item key of the node/leaf should be the same with the one
1868 * 2. block in parent node should match the child node/leaf.
1869 * 3. generation of parent node and child's header should be consistent.
1871 * Or the child node/leaf pointed by the key in parent is not valid.
1873 * We hope to check leaf owner too, but since subvol may share leaves,
1874 * which makes leaf owner check not so strong, key check should be
1875 * sufficient enough for that case.
1877 static int check_child_node(struct btrfs_root *root,
1878 struct extent_buffer *parent, int slot,
1879 struct extent_buffer *child)
1881 struct btrfs_key parent_key;
1882 struct btrfs_key child_key;
1885 btrfs_node_key_to_cpu(parent, &parent_key, slot);
1886 if (btrfs_header_level(child) == 0)
1887 btrfs_item_key_to_cpu(child, &child_key, 0);
1889 btrfs_node_key_to_cpu(child, &child_key, 0);
1891 if (memcmp(&parent_key, &child_key, sizeof(parent_key))) {
1894 "Wrong key of child node/leaf, wanted: (%llu, %u, %llu), have: (%llu, %u, %llu)\n",
1895 parent_key.objectid, parent_key.type, parent_key.offset,
1896 child_key.objectid, child_key.type, child_key.offset);
1898 if (btrfs_header_bytenr(child) != btrfs_node_blockptr(parent, slot)) {
1900 fprintf(stderr, "Wrong block of child node/leaf, wanted: %llu, have: %llu\n",
1901 btrfs_node_blockptr(parent, slot),
1902 btrfs_header_bytenr(child));
1904 if (btrfs_node_ptr_generation(parent, slot) !=
1905 btrfs_header_generation(child)) {
1907 fprintf(stderr, "Wrong generation of child node/leaf, wanted: %llu, have: %llu\n",
1908 btrfs_header_generation(child),
1909 btrfs_node_ptr_generation(parent, slot));
1915 u64 bytenr[BTRFS_MAX_LEVEL];
1916 u64 refs[BTRFS_MAX_LEVEL];
1919 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
1920 struct walk_control *wc, int *level,
1921 struct node_refs *nrefs)
1923 enum btrfs_tree_block_status status;
1926 struct extent_buffer *next;
1927 struct extent_buffer *cur;
1932 WARN_ON(*level < 0);
1933 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1935 if (path->nodes[*level]->start == nrefs->bytenr[*level]) {
1936 refs = nrefs->refs[*level];
1939 ret = btrfs_lookup_extent_info(NULL, root,
1940 path->nodes[*level]->start,
1941 *level, 1, &refs, NULL);
1946 nrefs->bytenr[*level] = path->nodes[*level]->start;
1947 nrefs->refs[*level] = refs;
1951 ret = enter_shared_node(root, path->nodes[*level]->start,
1959 while (*level >= 0) {
1960 WARN_ON(*level < 0);
1961 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1962 cur = path->nodes[*level];
1964 if (btrfs_header_level(cur) != *level)
1967 if (path->slots[*level] >= btrfs_header_nritems(cur))
1970 ret = process_one_leaf(root, cur, wc);
1975 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1976 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1977 blocksize = root->nodesize;
1979 if (bytenr == nrefs->bytenr[*level - 1]) {
1980 refs = nrefs->refs[*level - 1];
1982 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
1983 *level - 1, 1, &refs, NULL);
1987 nrefs->bytenr[*level - 1] = bytenr;
1988 nrefs->refs[*level - 1] = refs;
1993 ret = enter_shared_node(root, bytenr, refs,
1996 path->slots[*level]++;
2001 next = btrfs_find_tree_block(root, bytenr, blocksize);
2002 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2003 free_extent_buffer(next);
2004 reada_walk_down(root, cur, path->slots[*level]);
2005 next = read_tree_block(root, bytenr, blocksize,
2007 if (!extent_buffer_uptodate(next)) {
2008 struct btrfs_key node_key;
2010 btrfs_node_key_to_cpu(path->nodes[*level],
2012 path->slots[*level]);
2013 btrfs_add_corrupt_extent_record(root->fs_info,
2015 path->nodes[*level]->start,
2016 root->nodesize, *level);
2022 ret = check_child_node(root, cur, path->slots[*level], next);
2028 if (btrfs_is_leaf(next))
2029 status = btrfs_check_leaf(root, NULL, next);
2031 status = btrfs_check_node(root, NULL, next);
2032 if (status != BTRFS_TREE_BLOCK_CLEAN) {
2033 free_extent_buffer(next);
2038 *level = *level - 1;
2039 free_extent_buffer(path->nodes[*level]);
2040 path->nodes[*level] = next;
2041 path->slots[*level] = 0;
2044 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2048 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
2049 struct walk_control *wc, int *level)
2052 struct extent_buffer *leaf;
2054 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2055 leaf = path->nodes[i];
2056 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
2061 free_extent_buffer(path->nodes[*level]);
2062 path->nodes[*level] = NULL;
2063 BUG_ON(*level > wc->active_node);
2064 if (*level == wc->active_node)
2065 leave_shared_node(root, wc, *level);
2072 static int check_root_dir(struct inode_record *rec)
2074 struct inode_backref *backref;
2077 if (!rec->found_inode_item || rec->errors)
2079 if (rec->nlink != 1 || rec->found_link != 0)
2081 if (list_empty(&rec->backrefs))
2083 backref = to_inode_backref(rec->backrefs.next);
2084 if (!backref->found_inode_ref)
2086 if (backref->index != 0 || backref->namelen != 2 ||
2087 memcmp(backref->name, "..", 2))
2089 if (backref->found_dir_index || backref->found_dir_item)
2096 static int repair_inode_isize(struct btrfs_trans_handle *trans,
2097 struct btrfs_root *root, struct btrfs_path *path,
2098 struct inode_record *rec)
2100 struct btrfs_inode_item *ei;
2101 struct btrfs_key key;
2104 key.objectid = rec->ino;
2105 key.type = BTRFS_INODE_ITEM_KEY;
2106 key.offset = (u64)-1;
2108 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2112 if (!path->slots[0]) {
2119 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2120 if (key.objectid != rec->ino) {
2125 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2126 struct btrfs_inode_item);
2127 btrfs_set_inode_size(path->nodes[0], ei, rec->found_size);
2128 btrfs_mark_buffer_dirty(path->nodes[0]);
2129 rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
2130 printf("reset isize for dir %Lu root %Lu\n", rec->ino,
2131 root->root_key.objectid);
2133 btrfs_release_path(path);
2137 static int repair_inode_orphan_item(struct btrfs_trans_handle *trans,
2138 struct btrfs_root *root,
2139 struct btrfs_path *path,
2140 struct inode_record *rec)
2144 ret = btrfs_add_orphan_item(trans, root, path, rec->ino);
2145 btrfs_release_path(path);
2147 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
2151 static int repair_inode_nbytes(struct btrfs_trans_handle *trans,
2152 struct btrfs_root *root,
2153 struct btrfs_path *path,
2154 struct inode_record *rec)
2156 struct btrfs_inode_item *ei;
2157 struct btrfs_key key;
2160 key.objectid = rec->ino;
2161 key.type = BTRFS_INODE_ITEM_KEY;
2164 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2171 /* Since ret == 0, no need to check anything */
2172 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2173 struct btrfs_inode_item);
2174 btrfs_set_inode_nbytes(path->nodes[0], ei, rec->found_size);
2175 btrfs_mark_buffer_dirty(path->nodes[0]);
2176 rec->errors &= ~I_ERR_FILE_NBYTES_WRONG;
2177 printf("reset nbytes for ino %llu root %llu\n",
2178 rec->ino, root->root_key.objectid);
2180 btrfs_release_path(path);
2184 static int add_missing_dir_index(struct btrfs_root *root,
2185 struct cache_tree *inode_cache,
2186 struct inode_record *rec,
2187 struct inode_backref *backref)
2189 struct btrfs_path path;
2190 struct btrfs_trans_handle *trans;
2191 struct btrfs_dir_item *dir_item;
2192 struct extent_buffer *leaf;
2193 struct btrfs_key key;
2194 struct btrfs_disk_key disk_key;
2195 struct inode_record *dir_rec;
2196 unsigned long name_ptr;
2197 u32 data_size = sizeof(*dir_item) + backref->namelen;
2200 trans = btrfs_start_transaction(root, 1);
2202 return PTR_ERR(trans);
2204 fprintf(stderr, "repairing missing dir index item for inode %llu\n",
2205 (unsigned long long)rec->ino);
2207 btrfs_init_path(&path);
2208 key.objectid = backref->dir;
2209 key.type = BTRFS_DIR_INDEX_KEY;
2210 key.offset = backref->index;
2211 ret = btrfs_insert_empty_item(trans, root, &path, &key, data_size);
2214 leaf = path.nodes[0];
2215 dir_item = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_dir_item);
2217 disk_key.objectid = cpu_to_le64(rec->ino);
2218 disk_key.type = BTRFS_INODE_ITEM_KEY;
2219 disk_key.offset = 0;
2221 btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
2222 btrfs_set_dir_type(leaf, dir_item, imode_to_type(rec->imode));
2223 btrfs_set_dir_data_len(leaf, dir_item, 0);
2224 btrfs_set_dir_name_len(leaf, dir_item, backref->namelen);
2225 name_ptr = (unsigned long)(dir_item + 1);
2226 write_extent_buffer(leaf, backref->name, name_ptr, backref->namelen);
2227 btrfs_mark_buffer_dirty(leaf);
2228 btrfs_release_path(&path);
2229 btrfs_commit_transaction(trans, root);
2231 backref->found_dir_index = 1;
2232 dir_rec = get_inode_rec(inode_cache, backref->dir, 0);
2233 BUG_ON(IS_ERR(dir_rec));
2236 dir_rec->found_size += backref->namelen;
2237 if (dir_rec->found_size == dir_rec->isize &&
2238 (dir_rec->errors & I_ERR_DIR_ISIZE_WRONG))
2239 dir_rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
2240 if (dir_rec->found_size != dir_rec->isize)
2241 dir_rec->errors |= I_ERR_DIR_ISIZE_WRONG;
2246 static int delete_dir_index(struct btrfs_root *root,
2247 struct cache_tree *inode_cache,
2248 struct inode_record *rec,
2249 struct inode_backref *backref)
2251 struct btrfs_trans_handle *trans;
2252 struct btrfs_dir_item *di;
2253 struct btrfs_path path;
2256 trans = btrfs_start_transaction(root, 1);
2258 return PTR_ERR(trans);
2260 fprintf(stderr, "Deleting bad dir index [%llu,%u,%llu] root %llu\n",
2261 (unsigned long long)backref->dir,
2262 BTRFS_DIR_INDEX_KEY, (unsigned long long)backref->index,
2263 (unsigned long long)root->objectid);
2265 btrfs_init_path(&path);
2266 di = btrfs_lookup_dir_index(trans, root, &path, backref->dir,
2267 backref->name, backref->namelen,
2268 backref->index, -1);
2271 btrfs_release_path(&path);
2272 btrfs_commit_transaction(trans, root);
2279 ret = btrfs_del_item(trans, root, &path);
2281 ret = btrfs_delete_one_dir_name(trans, root, &path, di);
2283 btrfs_release_path(&path);
2284 btrfs_commit_transaction(trans, root);
2288 static int create_inode_item(struct btrfs_root *root,
2289 struct inode_record *rec,
2290 struct inode_backref *backref, int root_dir)
2292 struct btrfs_trans_handle *trans;
2293 struct btrfs_inode_item inode_item;
2294 time_t now = time(NULL);
2297 trans = btrfs_start_transaction(root, 1);
2298 if (IS_ERR(trans)) {
2299 ret = PTR_ERR(trans);
2303 fprintf(stderr, "root %llu inode %llu recreating inode item, this may "
2304 "be incomplete, please check permissions and content after "
2305 "the fsck completes.\n", (unsigned long long)root->objectid,
2306 (unsigned long long)rec->ino);
2308 memset(&inode_item, 0, sizeof(inode_item));
2309 btrfs_set_stack_inode_generation(&inode_item, trans->transid);
2311 btrfs_set_stack_inode_nlink(&inode_item, 1);
2313 btrfs_set_stack_inode_nlink(&inode_item, rec->found_link);
2314 btrfs_set_stack_inode_nbytes(&inode_item, rec->found_size);
2315 if (rec->found_dir_item) {
2316 if (rec->found_file_extent)
2317 fprintf(stderr, "root %llu inode %llu has both a dir "
2318 "item and extents, unsure if it is a dir or a "
2319 "regular file so setting it as a directory\n",
2320 (unsigned long long)root->objectid,
2321 (unsigned long long)rec->ino);
2322 btrfs_set_stack_inode_mode(&inode_item, S_IFDIR | 0755);
2323 btrfs_set_stack_inode_size(&inode_item, rec->found_size);
2324 } else if (!rec->found_dir_item) {
2325 btrfs_set_stack_inode_size(&inode_item, rec->extent_end);
2326 btrfs_set_stack_inode_mode(&inode_item, S_IFREG | 0755);
2328 btrfs_set_stack_timespec_sec(&inode_item.atime, now);
2329 btrfs_set_stack_timespec_nsec(&inode_item.atime, 0);
2330 btrfs_set_stack_timespec_sec(&inode_item.ctime, now);
2331 btrfs_set_stack_timespec_nsec(&inode_item.ctime, 0);
2332 btrfs_set_stack_timespec_sec(&inode_item.mtime, now);
2333 btrfs_set_stack_timespec_nsec(&inode_item.mtime, 0);
2334 btrfs_set_stack_timespec_sec(&inode_item.otime, 0);
2335 btrfs_set_stack_timespec_nsec(&inode_item.otime, 0);
2337 ret = btrfs_insert_inode(trans, root, rec->ino, &inode_item);
2339 btrfs_commit_transaction(trans, root);
2343 static int repair_inode_backrefs(struct btrfs_root *root,
2344 struct inode_record *rec,
2345 struct cache_tree *inode_cache,
2348 struct inode_backref *tmp, *backref;
2349 u64 root_dirid = btrfs_root_dirid(&root->root_item);
2353 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
2354 if (!delete && rec->ino == root_dirid) {
2355 if (!rec->found_inode_item) {
2356 ret = create_inode_item(root, rec, backref, 1);
2363 /* Index 0 for root dir's are special, don't mess with it */
2364 if (rec->ino == root_dirid && backref->index == 0)
2368 ((backref->found_dir_index && !backref->found_inode_ref) ||
2369 (backref->found_dir_index && backref->found_inode_ref &&
2370 (backref->errors & REF_ERR_INDEX_UNMATCH)))) {
2371 ret = delete_dir_index(root, inode_cache, rec, backref);
2375 list_del(&backref->list);
2379 if (!delete && !backref->found_dir_index &&
2380 backref->found_dir_item && backref->found_inode_ref) {
2381 ret = add_missing_dir_index(root, inode_cache, rec,
2386 if (backref->found_dir_item &&
2387 backref->found_dir_index &&
2388 backref->found_dir_index) {
2389 if (!backref->errors &&
2390 backref->found_inode_ref) {
2391 list_del(&backref->list);
2397 if (!delete && (!backref->found_dir_index &&
2398 !backref->found_dir_item &&
2399 backref->found_inode_ref)) {
2400 struct btrfs_trans_handle *trans;
2401 struct btrfs_key location;
2403 ret = check_dir_conflict(root, backref->name,
2409 * let nlink fixing routine to handle it,
2410 * which can do it better.
2415 location.objectid = rec->ino;
2416 location.type = BTRFS_INODE_ITEM_KEY;
2417 location.offset = 0;
2419 trans = btrfs_start_transaction(root, 1);
2420 if (IS_ERR(trans)) {
2421 ret = PTR_ERR(trans);
2424 fprintf(stderr, "adding missing dir index/item pair "
2426 (unsigned long long)rec->ino);
2427 ret = btrfs_insert_dir_item(trans, root, backref->name,
2429 backref->dir, &location,
2430 imode_to_type(rec->imode),
2433 btrfs_commit_transaction(trans, root);
2437 if (!delete && (backref->found_inode_ref &&
2438 backref->found_dir_index &&
2439 backref->found_dir_item &&
2440 !(backref->errors & REF_ERR_INDEX_UNMATCH) &&
2441 !rec->found_inode_item)) {
2442 ret = create_inode_item(root, rec, backref, 0);
2449 return ret ? ret : repaired;
2453 * To determine the file type for nlink/inode_item repair
2455 * Return 0 if file type is found and BTRFS_FT_* is stored into type.
2456 * Return -ENOENT if file type is not found.
2458 static int find_file_type(struct inode_record *rec, u8 *type)
2460 struct inode_backref *backref;
2462 /* For inode item recovered case */
2463 if (rec->found_inode_item) {
2464 *type = imode_to_type(rec->imode);
2468 list_for_each_entry(backref, &rec->backrefs, list) {
2469 if (backref->found_dir_index || backref->found_dir_item) {
2470 *type = backref->filetype;
2478 * To determine the file name for nlink repair
2480 * Return 0 if file name is found, set name and namelen.
2481 * Return -ENOENT if file name is not found.
2483 static int find_file_name(struct inode_record *rec,
2484 char *name, int *namelen)
2486 struct inode_backref *backref;
2488 list_for_each_entry(backref, &rec->backrefs, list) {
2489 if (backref->found_dir_index || backref->found_dir_item ||
2490 backref->found_inode_ref) {
2491 memcpy(name, backref->name, backref->namelen);
2492 *namelen = backref->namelen;
2499 /* Reset the nlink of the inode to the correct one */
2500 static int reset_nlink(struct btrfs_trans_handle *trans,
2501 struct btrfs_root *root,
2502 struct btrfs_path *path,
2503 struct inode_record *rec)
2505 struct inode_backref *backref;
2506 struct inode_backref *tmp;
2507 struct btrfs_key key;
2508 struct btrfs_inode_item *inode_item;
2511 /* We don't believe this either, reset it and iterate backref */
2512 rec->found_link = 0;
2514 /* Remove all backref including the valid ones */
2515 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
2516 ret = btrfs_unlink(trans, root, rec->ino, backref->dir,
2517 backref->index, backref->name,
2518 backref->namelen, 0);
2522 /* remove invalid backref, so it won't be added back */
2523 if (!(backref->found_dir_index &&
2524 backref->found_dir_item &&
2525 backref->found_inode_ref)) {
2526 list_del(&backref->list);
2533 /* Set nlink to 0 */
2534 key.objectid = rec->ino;
2535 key.type = BTRFS_INODE_ITEM_KEY;
2537 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2544 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2545 struct btrfs_inode_item);
2546 btrfs_set_inode_nlink(path->nodes[0], inode_item, 0);
2547 btrfs_mark_buffer_dirty(path->nodes[0]);
2548 btrfs_release_path(path);
2551 * Add back valid inode_ref/dir_item/dir_index,
2552 * add_link() will handle the nlink inc, so new nlink must be correct
2554 list_for_each_entry(backref, &rec->backrefs, list) {
2555 ret = btrfs_add_link(trans, root, rec->ino, backref->dir,
2556 backref->name, backref->namelen,
2557 backref->filetype, &backref->index, 1);
2562 btrfs_release_path(path);
2566 static int repair_inode_nlinks(struct btrfs_trans_handle *trans,
2567 struct btrfs_root *root,
2568 struct btrfs_path *path,
2569 struct inode_record *rec)
2571 char *dir_name = "lost+found";
2572 char namebuf[BTRFS_NAME_LEN] = {0};
2577 int name_recovered = 0;
2578 int type_recovered = 0;
2582 * Get file name and type first before these invalid inode ref
2583 * are deleted by remove_all_invalid_backref()
2585 name_recovered = !find_file_name(rec, namebuf, &namelen);
2586 type_recovered = !find_file_type(rec, &type);
2588 if (!name_recovered) {
2589 printf("Can't get file name for inode %llu, using '%llu' as fallback\n",
2590 rec->ino, rec->ino);
2591 namelen = count_digits(rec->ino);
2592 sprintf(namebuf, "%llu", rec->ino);
2595 if (!type_recovered) {
2596 printf("Can't get file type for inode %llu, using FILE as fallback\n",
2598 type = BTRFS_FT_REG_FILE;
2602 ret = reset_nlink(trans, root, path, rec);
2605 "Failed to reset nlink for inode %llu: %s\n",
2606 rec->ino, strerror(-ret));
2610 if (rec->found_link == 0) {
2611 lost_found_ino = root->highest_inode;
2612 if (lost_found_ino >= BTRFS_LAST_FREE_OBJECTID) {
2617 ret = btrfs_mkdir(trans, root, dir_name, strlen(dir_name),
2618 BTRFS_FIRST_FREE_OBJECTID, &lost_found_ino,
2621 fprintf(stderr, "Failed to create '%s' dir: %s\n",
2622 dir_name, strerror(-ret));
2625 ret = btrfs_add_link(trans, root, rec->ino, lost_found_ino,
2626 namebuf, namelen, type, NULL, 1);
2628 * Add ".INO" suffix several times to handle case where
2629 * "FILENAME.INO" is already taken by another file.
2631 while (ret == -EEXIST) {
2633 * Conflicting file name, add ".INO" as suffix * +1 for '.'
2635 if (namelen + count_digits(rec->ino) + 1 >
2640 snprintf(namebuf + namelen, BTRFS_NAME_LEN - namelen,
2642 namelen += count_digits(rec->ino) + 1;
2643 ret = btrfs_add_link(trans, root, rec->ino,
2644 lost_found_ino, namebuf,
2645 namelen, type, NULL, 1);
2649 "Failed to link the inode %llu to %s dir: %s\n",
2650 rec->ino, dir_name, strerror(-ret));
2654 * Just increase the found_link, don't actually add the
2655 * backref. This will make things easier and this inode
2656 * record will be freed after the repair is done.
2657 * So fsck will not report problem about this inode.
2660 printf("Moving file '%.*s' to '%s' dir since it has no valid backref\n",
2661 namelen, namebuf, dir_name);
2663 printf("Fixed the nlink of inode %llu\n", rec->ino);
2666 * Clear the flag anyway, or we will loop forever for the same inode
2667 * as it will not be removed from the bad inode list and the dead loop
2670 rec->errors &= ~I_ERR_LINK_COUNT_WRONG;
2671 btrfs_release_path(path);
2676 * Check if there is any normal(reg or prealloc) file extent for given
2678 * This is used to determine the file type when neither its dir_index/item or
2679 * inode_item exists.
2681 * This will *NOT* report error, if any error happens, just consider it does
2682 * not have any normal file extent.
2684 static int find_normal_file_extent(struct btrfs_root *root, u64 ino)
2686 struct btrfs_path path;
2687 struct btrfs_key key;
2688 struct btrfs_key found_key;
2689 struct btrfs_file_extent_item *fi;
2693 btrfs_init_path(&path);
2695 key.type = BTRFS_EXTENT_DATA_KEY;
2698 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
2703 if (ret && path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
2704 ret = btrfs_next_leaf(root, &path);
2711 btrfs_item_key_to_cpu(path.nodes[0], &found_key,
2713 if (found_key.objectid != ino ||
2714 found_key.type != BTRFS_EXTENT_DATA_KEY)
2716 fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
2717 struct btrfs_file_extent_item);
2718 type = btrfs_file_extent_type(path.nodes[0], fi);
2719 if (type != BTRFS_FILE_EXTENT_INLINE) {
2725 btrfs_release_path(&path);
2729 static u32 btrfs_type_to_imode(u8 type)
2731 static u32 imode_by_btrfs_type[] = {
2732 [BTRFS_FT_REG_FILE] = S_IFREG,
2733 [BTRFS_FT_DIR] = S_IFDIR,
2734 [BTRFS_FT_CHRDEV] = S_IFCHR,
2735 [BTRFS_FT_BLKDEV] = S_IFBLK,
2736 [BTRFS_FT_FIFO] = S_IFIFO,
2737 [BTRFS_FT_SOCK] = S_IFSOCK,
2738 [BTRFS_FT_SYMLINK] = S_IFLNK,
2741 return imode_by_btrfs_type[(type)];
2744 static int repair_inode_no_item(struct btrfs_trans_handle *trans,
2745 struct btrfs_root *root,
2746 struct btrfs_path *path,
2747 struct inode_record *rec)
2751 int type_recovered = 0;
2754 printf("Trying to rebuild inode:%llu\n", rec->ino);
2756 type_recovered = !find_file_type(rec, &filetype);
2759 * Try to determine inode type if type not found.
2761 * For found regular file extent, it must be FILE.
2762 * For found dir_item/index, it must be DIR.
2764 * For undetermined one, use FILE as fallback.
2767 * 1. If found backref(inode_index/item is already handled) to it,
2769 * Need new inode-inode ref structure to allow search for that.
2771 if (!type_recovered) {
2772 if (rec->found_file_extent &&
2773 find_normal_file_extent(root, rec->ino)) {
2775 filetype = BTRFS_FT_REG_FILE;
2776 } else if (rec->found_dir_item) {
2778 filetype = BTRFS_FT_DIR;
2779 } else if (!list_empty(&rec->orphan_extents)) {
2781 filetype = BTRFS_FT_REG_FILE;
2783 printf("Can't determine the filetype for inode %llu, assume it is a normal file\n",
2786 filetype = BTRFS_FT_REG_FILE;
2790 ret = btrfs_new_inode(trans, root, rec->ino,
2791 mode | btrfs_type_to_imode(filetype));
2796 * Here inode rebuild is done, we only rebuild the inode item,
2797 * don't repair the nlink(like move to lost+found).
2798 * That is the job of nlink repair.
2800 * We just fill the record and return
2802 rec->found_dir_item = 1;
2803 rec->imode = mode | btrfs_type_to_imode(filetype);
2805 rec->errors &= ~I_ERR_NO_INODE_ITEM;
2806 /* Ensure the inode_nlinks repair function will be called */
2807 rec->errors |= I_ERR_LINK_COUNT_WRONG;
2812 static int repair_inode_orphan_extent(struct btrfs_trans_handle *trans,
2813 struct btrfs_root *root,
2814 struct btrfs_path *path,
2815 struct inode_record *rec)
2817 struct orphan_data_extent *orphan;
2818 struct orphan_data_extent *tmp;
2821 list_for_each_entry_safe(orphan, tmp, &rec->orphan_extents, list) {
2823 * Check for conflicting file extents
2825 * Here we don't know whether the extents is compressed or not,
2826 * so we can only assume it not compressed nor data offset,
2827 * and use its disk_len as extent length.
2829 ret = btrfs_get_extent(NULL, root, path, orphan->objectid,
2830 orphan->offset, orphan->disk_len, 0);
2831 btrfs_release_path(path);
2836 "orphan extent (%llu, %llu) conflicts, delete the orphan\n",
2837 orphan->disk_bytenr, orphan->disk_len);
2838 ret = btrfs_free_extent(trans,
2839 root->fs_info->extent_root,
2840 orphan->disk_bytenr, orphan->disk_len,
2841 0, root->objectid, orphan->objectid,
2846 ret = btrfs_insert_file_extent(trans, root, orphan->objectid,
2847 orphan->offset, orphan->disk_bytenr,
2848 orphan->disk_len, orphan->disk_len);
2852 /* Update file size info */
2853 rec->found_size += orphan->disk_len;
2854 if (rec->found_size == rec->nbytes)
2855 rec->errors &= ~I_ERR_FILE_NBYTES_WRONG;
2857 /* Update the file extent hole info too */
2858 ret = del_file_extent_hole(&rec->holes, orphan->offset,
2862 if (RB_EMPTY_ROOT(&rec->holes))
2863 rec->errors &= ~I_ERR_FILE_EXTENT_DISCOUNT;
2865 list_del(&orphan->list);
2868 rec->errors &= ~I_ERR_FILE_EXTENT_ORPHAN;
2873 static int repair_inode_discount_extent(struct btrfs_trans_handle *trans,
2874 struct btrfs_root *root,
2875 struct btrfs_path *path,
2876 struct inode_record *rec)
2878 struct rb_node *node;
2879 struct file_extent_hole *hole;
2883 node = rb_first(&rec->holes);
2887 hole = rb_entry(node, struct file_extent_hole, node);
2888 ret = btrfs_punch_hole(trans, root, rec->ino,
2889 hole->start, hole->len);
2892 ret = del_file_extent_hole(&rec->holes, hole->start,
2896 if (RB_EMPTY_ROOT(&rec->holes))
2897 rec->errors &= ~I_ERR_FILE_EXTENT_DISCOUNT;
2898 node = rb_first(&rec->holes);
2900 /* special case for a file losing all its file extent */
2902 ret = btrfs_punch_hole(trans, root, rec->ino, 0,
2903 round_up(rec->isize, root->sectorsize));
2907 printf("Fixed discount file extents for inode: %llu in root: %llu\n",
2908 rec->ino, root->objectid);
2913 static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
2915 struct btrfs_trans_handle *trans;
2916 struct btrfs_path path;
2919 if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG |
2920 I_ERR_NO_ORPHAN_ITEM |
2921 I_ERR_LINK_COUNT_WRONG |
2922 I_ERR_NO_INODE_ITEM |
2923 I_ERR_FILE_EXTENT_ORPHAN |
2924 I_ERR_FILE_EXTENT_DISCOUNT|
2925 I_ERR_FILE_NBYTES_WRONG)))
2929 * For nlink repair, it may create a dir and add link, so
2930 * 2 for parent(256)'s dir_index and dir_item
2931 * 2 for lost+found dir's inode_item and inode_ref
2932 * 1 for the new inode_ref of the file
2933 * 2 for lost+found dir's dir_index and dir_item for the file
2935 trans = btrfs_start_transaction(root, 7);
2937 return PTR_ERR(trans);
2939 btrfs_init_path(&path);
2940 if (rec->errors & I_ERR_NO_INODE_ITEM)
2941 ret = repair_inode_no_item(trans, root, &path, rec);
2942 if (!ret && rec->errors & I_ERR_FILE_EXTENT_ORPHAN)
2943 ret = repair_inode_orphan_extent(trans, root, &path, rec);
2944 if (!ret && rec->errors & I_ERR_FILE_EXTENT_DISCOUNT)
2945 ret = repair_inode_discount_extent(trans, root, &path, rec);
2946 if (!ret && rec->errors & I_ERR_DIR_ISIZE_WRONG)
2947 ret = repair_inode_isize(trans, root, &path, rec);
2948 if (!ret && rec->errors & I_ERR_NO_ORPHAN_ITEM)
2949 ret = repair_inode_orphan_item(trans, root, &path, rec);
2950 if (!ret && rec->errors & I_ERR_LINK_COUNT_WRONG)
2951 ret = repair_inode_nlinks(trans, root, &path, rec);
2952 if (!ret && rec->errors & I_ERR_FILE_NBYTES_WRONG)
2953 ret = repair_inode_nbytes(trans, root, &path, rec);
2954 btrfs_commit_transaction(trans, root);
2955 btrfs_release_path(&path);
2959 static int check_inode_recs(struct btrfs_root *root,
2960 struct cache_tree *inode_cache)
2962 struct cache_extent *cache;
2963 struct ptr_node *node;
2964 struct inode_record *rec;
2965 struct inode_backref *backref;
2970 u64 root_dirid = btrfs_root_dirid(&root->root_item);
2972 if (btrfs_root_refs(&root->root_item) == 0) {
2973 if (!cache_tree_empty(inode_cache))
2974 fprintf(stderr, "warning line %d\n", __LINE__);
2979 * We need to record the highest inode number for later 'lost+found'
2981 * We must select an ino not used/referred by any existing inode, or
2982 * 'lost+found' ino may be a missing ino in a corrupted leaf,
2983 * this may cause 'lost+found' dir has wrong nlinks.
2985 cache = last_cache_extent(inode_cache);
2987 node = container_of(cache, struct ptr_node, cache);
2989 if (rec->ino > root->highest_inode)
2990 root->highest_inode = rec->ino;
2994 * We need to repair backrefs first because we could change some of the
2995 * errors in the inode recs.
2997 * We also need to go through and delete invalid backrefs first and then
2998 * add the correct ones second. We do this because we may get EEXIST
2999 * when adding back the correct index because we hadn't yet deleted the
3002 * For example, if we were missing a dir index then the directories
3003 * isize would be wrong, so if we fixed the isize to what we thought it
3004 * would be and then fixed the backref we'd still have a invalid fs, so
3005 * we need to add back the dir index and then check to see if the isize
3010 if (stage == 3 && !err)
3013 cache = search_cache_extent(inode_cache, 0);
3014 while (repair && cache) {
3015 node = container_of(cache, struct ptr_node, cache);
3017 cache = next_cache_extent(cache);
3019 /* Need to free everything up and rescan */
3021 remove_cache_extent(inode_cache, &node->cache);
3023 free_inode_rec(rec);
3027 if (list_empty(&rec->backrefs))
3030 ret = repair_inode_backrefs(root, rec, inode_cache,
3044 rec = get_inode_rec(inode_cache, root_dirid, 0);
3045 BUG_ON(IS_ERR(rec));
3047 ret = check_root_dir(rec);
3049 fprintf(stderr, "root %llu root dir %llu error\n",
3050 (unsigned long long)root->root_key.objectid,
3051 (unsigned long long)root_dirid);
3052 print_inode_error(root, rec);
3057 struct btrfs_trans_handle *trans;
3059 trans = btrfs_start_transaction(root, 1);
3060 if (IS_ERR(trans)) {
3061 err = PTR_ERR(trans);
3066 "root %llu missing its root dir, recreating\n",
3067 (unsigned long long)root->objectid);
3069 ret = btrfs_make_root_dir(trans, root, root_dirid);
3072 btrfs_commit_transaction(trans, root);
3076 fprintf(stderr, "root %llu root dir %llu not found\n",
3077 (unsigned long long)root->root_key.objectid,
3078 (unsigned long long)root_dirid);
3082 cache = search_cache_extent(inode_cache, 0);
3085 node = container_of(cache, struct ptr_node, cache);
3087 remove_cache_extent(inode_cache, &node->cache);
3089 if (rec->ino == root_dirid ||
3090 rec->ino == BTRFS_ORPHAN_OBJECTID) {
3091 free_inode_rec(rec);
3095 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
3096 ret = check_orphan_item(root, rec->ino);
3098 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
3099 if (can_free_inode_rec(rec)) {
3100 free_inode_rec(rec);
3105 if (!rec->found_inode_item)
3106 rec->errors |= I_ERR_NO_INODE_ITEM;
3107 if (rec->found_link != rec->nlink)
3108 rec->errors |= I_ERR_LINK_COUNT_WRONG;
3110 ret = try_repair_inode(root, rec);
3111 if (ret == 0 && can_free_inode_rec(rec)) {
3112 free_inode_rec(rec);
3118 if (!(repair && ret == 0))
3120 print_inode_error(root, rec);
3121 list_for_each_entry(backref, &rec->backrefs, list) {
3122 if (!backref->found_dir_item)
3123 backref->errors |= REF_ERR_NO_DIR_ITEM;
3124 if (!backref->found_dir_index)
3125 backref->errors |= REF_ERR_NO_DIR_INDEX;
3126 if (!backref->found_inode_ref)
3127 backref->errors |= REF_ERR_NO_INODE_REF;
3128 fprintf(stderr, "\tunresolved ref dir %llu index %llu"
3129 " namelen %u name %s filetype %d errors %x",
3130 (unsigned long long)backref->dir,
3131 (unsigned long long)backref->index,
3132 backref->namelen, backref->name,
3133 backref->filetype, backref->errors);
3134 print_ref_error(backref->errors);
3136 free_inode_rec(rec);
3138 return (error > 0) ? -1 : 0;
3141 static struct root_record *get_root_rec(struct cache_tree *root_cache,
3144 struct cache_extent *cache;
3145 struct root_record *rec = NULL;
3148 cache = lookup_cache_extent(root_cache, objectid, 1);
3150 rec = container_of(cache, struct root_record, cache);
3152 rec = calloc(1, sizeof(*rec));
3154 return ERR_PTR(-ENOMEM);
3155 rec->objectid = objectid;
3156 INIT_LIST_HEAD(&rec->backrefs);
3157 rec->cache.start = objectid;
3158 rec->cache.size = 1;
3160 ret = insert_cache_extent(root_cache, &rec->cache);
3162 return ERR_PTR(-EEXIST);
3167 static struct root_backref *get_root_backref(struct root_record *rec,
3168 u64 ref_root, u64 dir, u64 index,
3169 const char *name, int namelen)
3171 struct root_backref *backref;
3173 list_for_each_entry(backref, &rec->backrefs, list) {
3174 if (backref->ref_root != ref_root || backref->dir != dir ||
3175 backref->namelen != namelen)
3177 if (memcmp(name, backref->name, namelen))
3182 backref = calloc(1, sizeof(*backref) + namelen + 1);
3185 backref->ref_root = ref_root;
3187 backref->index = index;
3188 backref->namelen = namelen;
3189 memcpy(backref->name, name, namelen);
3190 backref->name[namelen] = '\0';
3191 list_add_tail(&backref->list, &rec->backrefs);
3195 static void free_root_record(struct cache_extent *cache)
3197 struct root_record *rec;
3198 struct root_backref *backref;
3200 rec = container_of(cache, struct root_record, cache);
3201 while (!list_empty(&rec->backrefs)) {
3202 backref = to_root_backref(rec->backrefs.next);
3203 list_del(&backref->list);
3210 FREE_EXTENT_CACHE_BASED_TREE(root_recs, free_root_record);
3212 static int add_root_backref(struct cache_tree *root_cache,
3213 u64 root_id, u64 ref_root, u64 dir, u64 index,
3214 const char *name, int namelen,
3215 int item_type, int errors)
3217 struct root_record *rec;
3218 struct root_backref *backref;
3220 rec = get_root_rec(root_cache, root_id);
3221 BUG_ON(IS_ERR(rec));
3222 backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
3225 backref->errors |= errors;
3227 if (item_type != BTRFS_DIR_ITEM_KEY) {
3228 if (backref->found_dir_index || backref->found_back_ref ||
3229 backref->found_forward_ref) {
3230 if (backref->index != index)
3231 backref->errors |= REF_ERR_INDEX_UNMATCH;
3233 backref->index = index;
3237 if (item_type == BTRFS_DIR_ITEM_KEY) {
3238 if (backref->found_forward_ref)
3240 backref->found_dir_item = 1;
3241 } else if (item_type == BTRFS_DIR_INDEX_KEY) {
3242 backref->found_dir_index = 1;
3243 } else if (item_type == BTRFS_ROOT_REF_KEY) {
3244 if (backref->found_forward_ref)
3245 backref->errors |= REF_ERR_DUP_ROOT_REF;
3246 else if (backref->found_dir_item)
3248 backref->found_forward_ref = 1;
3249 } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
3250 if (backref->found_back_ref)
3251 backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
3252 backref->found_back_ref = 1;
3257 if (backref->found_forward_ref && backref->found_dir_item)
3258 backref->reachable = 1;
3262 static int merge_root_recs(struct btrfs_root *root,
3263 struct cache_tree *src_cache,
3264 struct cache_tree *dst_cache)
3266 struct cache_extent *cache;
3267 struct ptr_node *node;
3268 struct inode_record *rec;
3269 struct inode_backref *backref;
3272 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
3273 free_inode_recs_tree(src_cache);
3278 cache = search_cache_extent(src_cache, 0);
3281 node = container_of(cache, struct ptr_node, cache);
3283 remove_cache_extent(src_cache, &node->cache);
3286 ret = is_child_root(root, root->objectid, rec->ino);
3292 list_for_each_entry(backref, &rec->backrefs, list) {
3293 BUG_ON(backref->found_inode_ref);
3294 if (backref->found_dir_item)
3295 add_root_backref(dst_cache, rec->ino,
3296 root->root_key.objectid, backref->dir,
3297 backref->index, backref->name,
3298 backref->namelen, BTRFS_DIR_ITEM_KEY,
3300 if (backref->found_dir_index)
3301 add_root_backref(dst_cache, rec->ino,
3302 root->root_key.objectid, backref->dir,
3303 backref->index, backref->name,
3304 backref->namelen, BTRFS_DIR_INDEX_KEY,
3308 free_inode_rec(rec);
3315 static int check_root_refs(struct btrfs_root *root,
3316 struct cache_tree *root_cache)
3318 struct root_record *rec;
3319 struct root_record *ref_root;
3320 struct root_backref *backref;
3321 struct cache_extent *cache;
3327 rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
3328 BUG_ON(IS_ERR(rec));
3331 /* fixme: this can not detect circular references */
3334 cache = search_cache_extent(root_cache, 0);
3338 rec = container_of(cache, struct root_record, cache);
3339 cache = next_cache_extent(cache);
3341 if (rec->found_ref == 0)
3344 list_for_each_entry(backref, &rec->backrefs, list) {
3345 if (!backref->reachable)
3348 ref_root = get_root_rec(root_cache,
3350 BUG_ON(IS_ERR(ref_root));
3351 if (ref_root->found_ref > 0)
3354 backref->reachable = 0;
3356 if (rec->found_ref == 0)
3362 cache = search_cache_extent(root_cache, 0);
3366 rec = container_of(cache, struct root_record, cache);
3367 cache = next_cache_extent(cache);
3369 if (rec->found_ref == 0 &&
3370 rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
3371 rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
3372 ret = check_orphan_item(root->fs_info->tree_root,
3378 * If we don't have a root item then we likely just have
3379 * a dir item in a snapshot for this root but no actual
3380 * ref key or anything so it's meaningless.
3382 if (!rec->found_root_item)
3385 fprintf(stderr, "fs tree %llu not referenced\n",
3386 (unsigned long long)rec->objectid);
3390 if (rec->found_ref > 0 && !rec->found_root_item)
3392 list_for_each_entry(backref, &rec->backrefs, list) {
3393 if (!backref->found_dir_item)
3394 backref->errors |= REF_ERR_NO_DIR_ITEM;
3395 if (!backref->found_dir_index)
3396 backref->errors |= REF_ERR_NO_DIR_INDEX;
3397 if (!backref->found_back_ref)
3398 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
3399 if (!backref->found_forward_ref)
3400 backref->errors |= REF_ERR_NO_ROOT_REF;
3401 if (backref->reachable && backref->errors)
3408 fprintf(stderr, "fs tree %llu refs %u %s\n",
3409 (unsigned long long)rec->objectid, rec->found_ref,
3410 rec->found_root_item ? "" : "not found");
3412 list_for_each_entry(backref, &rec->backrefs, list) {
3413 if (!backref->reachable)
3415 if (!backref->errors && rec->found_root_item)
3417 fprintf(stderr, "\tunresolved ref root %llu dir %llu"
3418 " index %llu namelen %u name %s errors %x\n",
3419 (unsigned long long)backref->ref_root,
3420 (unsigned long long)backref->dir,
3421 (unsigned long long)backref->index,
3422 backref->namelen, backref->name,
3424 print_ref_error(backref->errors);
3427 return errors > 0 ? 1 : 0;
3430 static int process_root_ref(struct extent_buffer *eb, int slot,
3431 struct btrfs_key *key,
3432 struct cache_tree *root_cache)
3438 struct btrfs_root_ref *ref;
3439 char namebuf[BTRFS_NAME_LEN];
3442 ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
3444 dirid = btrfs_root_ref_dirid(eb, ref);
3445 index = btrfs_root_ref_sequence(eb, ref);
3446 name_len = btrfs_root_ref_name_len(eb, ref);
3448 if (name_len <= BTRFS_NAME_LEN) {
3452 len = BTRFS_NAME_LEN;
3453 error = REF_ERR_NAME_TOO_LONG;
3455 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
3457 if (key->type == BTRFS_ROOT_REF_KEY) {
3458 add_root_backref(root_cache, key->offset, key->objectid, dirid,
3459 index, namebuf, len, key->type, error);
3461 add_root_backref(root_cache, key->objectid, key->offset, dirid,
3462 index, namebuf, len, key->type, error);
3467 static void free_corrupt_block(struct cache_extent *cache)
3469 struct btrfs_corrupt_block *corrupt;
3471 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
3475 FREE_EXTENT_CACHE_BASED_TREE(corrupt_blocks, free_corrupt_block);
3478 * Repair the btree of the given root.
3480 * The fix is to remove the node key in corrupt_blocks cache_tree.
3481 * and rebalance the tree.
3482 * After the fix, the btree should be writeable.
3484 static int repair_btree(struct btrfs_root *root,
3485 struct cache_tree *corrupt_blocks)
3487 struct btrfs_trans_handle *trans;
3488 struct btrfs_path path;
3489 struct btrfs_corrupt_block *corrupt;
3490 struct cache_extent *cache;
3491 struct btrfs_key key;
3496 if (cache_tree_empty(corrupt_blocks))
3499 trans = btrfs_start_transaction(root, 1);
3500 if (IS_ERR(trans)) {
3501 ret = PTR_ERR(trans);
3502 fprintf(stderr, "Error starting transaction: %s\n",
3506 btrfs_init_path(&path);
3507 cache = first_cache_extent(corrupt_blocks);
3509 corrupt = container_of(cache, struct btrfs_corrupt_block,
3511 level = corrupt->level;
3512 path.lowest_level = level;
3513 key.objectid = corrupt->key.objectid;
3514 key.type = corrupt->key.type;
3515 key.offset = corrupt->key.offset;
3518 * Here we don't want to do any tree balance, since it may
3519 * cause a balance with corrupted brother leaf/node,
3520 * so ins_len set to 0 here.
3521 * Balance will be done after all corrupt node/leaf is deleted.
3523 ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
3526 offset = btrfs_node_blockptr(path.nodes[level],
3529 /* Remove the ptr */
3530 ret = btrfs_del_ptr(trans, root, &path, level,
3535 * Remove the corresponding extent
3536 * return value is not concerned.
3538 btrfs_release_path(&path);
3539 ret = btrfs_free_extent(trans, root, offset, root->nodesize,
3540 0, root->root_key.objectid,
3542 cache = next_cache_extent(cache);
3545 /* Balance the btree using btrfs_search_slot() */
3546 cache = first_cache_extent(corrupt_blocks);
3548 corrupt = container_of(cache, struct btrfs_corrupt_block,
3550 memcpy(&key, &corrupt->key, sizeof(key));
3551 ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
3554 /* return will always >0 since it won't find the item */
3556 btrfs_release_path(&path);
3557 cache = next_cache_extent(cache);
3560 btrfs_commit_transaction(trans, root);
3561 btrfs_release_path(&path);
3565 static int check_fs_root(struct btrfs_root *root,
3566 struct cache_tree *root_cache,
3567 struct walk_control *wc)
3573 struct btrfs_path path;
3574 struct shared_node root_node;
3575 struct root_record *rec;
3576 struct btrfs_root_item *root_item = &root->root_item;
3577 struct cache_tree corrupt_blocks;
3578 struct orphan_data_extent *orphan;
3579 struct orphan_data_extent *tmp;
3580 enum btrfs_tree_block_status status;
3581 struct node_refs nrefs;
3584 * Reuse the corrupt_block cache tree to record corrupted tree block
3586 * Unlike the usage in extent tree check, here we do it in a per
3587 * fs/subvol tree base.
3589 cache_tree_init(&corrupt_blocks);
3590 root->fs_info->corrupt_blocks = &corrupt_blocks;
3592 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
3593 rec = get_root_rec(root_cache, root->root_key.objectid);
3594 BUG_ON(IS_ERR(rec));
3595 if (btrfs_root_refs(root_item) > 0)
3596 rec->found_root_item = 1;
3599 btrfs_init_path(&path);
3600 memset(&root_node, 0, sizeof(root_node));
3601 cache_tree_init(&root_node.root_cache);
3602 cache_tree_init(&root_node.inode_cache);
3603 memset(&nrefs, 0, sizeof(nrefs));
3605 /* Move the orphan extent record to corresponding inode_record */
3606 list_for_each_entry_safe(orphan, tmp,
3607 &root->orphan_data_extents, list) {
3608 struct inode_record *inode;
3610 inode = get_inode_rec(&root_node.inode_cache, orphan->objectid,
3612 BUG_ON(IS_ERR(inode));
3613 inode->errors |= I_ERR_FILE_EXTENT_ORPHAN;
3614 list_move(&orphan->list, &inode->orphan_extents);
3617 level = btrfs_header_level(root->node);
3618 memset(wc->nodes, 0, sizeof(wc->nodes));
3619 wc->nodes[level] = &root_node;
3620 wc->active_node = level;
3621 wc->root_level = level;
3623 /* We may not have checked the root block, lets do that now */
3624 if (btrfs_is_leaf(root->node))
3625 status = btrfs_check_leaf(root, NULL, root->node);
3627 status = btrfs_check_node(root, NULL, root->node);
3628 if (status != BTRFS_TREE_BLOCK_CLEAN)
3631 if (btrfs_root_refs(root_item) > 0 ||
3632 btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3633 path.nodes[level] = root->node;
3634 extent_buffer_get(root->node);
3635 path.slots[level] = 0;
3637 struct btrfs_key key;
3638 struct btrfs_disk_key found_key;
3640 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3641 level = root_item->drop_level;
3642 path.lowest_level = level;
3643 if (level > btrfs_header_level(root->node) ||
3644 level >= BTRFS_MAX_LEVEL) {
3645 error("ignoring invalid drop level: %u", level);
3648 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
3651 btrfs_node_key(path.nodes[level], &found_key,
3653 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3654 sizeof(found_key)));
3658 wret = walk_down_tree(root, &path, wc, &level, &nrefs);
3664 wret = walk_up_tree(root, &path, wc, &level);
3671 btrfs_release_path(&path);
3673 if (!cache_tree_empty(&corrupt_blocks)) {
3674 struct cache_extent *cache;
3675 struct btrfs_corrupt_block *corrupt;
3677 printf("The following tree block(s) is corrupted in tree %llu:\n",
3678 root->root_key.objectid);
3679 cache = first_cache_extent(&corrupt_blocks);
3681 corrupt = container_of(cache,
3682 struct btrfs_corrupt_block,
3684 printf("\ttree block bytenr: %llu, level: %d, node key: (%llu, %u, %llu)\n",
3685 cache->start, corrupt->level,
3686 corrupt->key.objectid, corrupt->key.type,
3687 corrupt->key.offset);
3688 cache = next_cache_extent(cache);
3691 printf("Try to repair the btree for root %llu\n",
3692 root->root_key.objectid);
3693 ret = repair_btree(root, &corrupt_blocks);
3695 fprintf(stderr, "Failed to repair btree: %s\n",
3698 printf("Btree for root %llu is fixed\n",
3699 root->root_key.objectid);
3703 err = merge_root_recs(root, &root_node.root_cache, root_cache);
3707 if (root_node.current) {
3708 root_node.current->checked = 1;
3709 maybe_free_inode_rec(&root_node.inode_cache,
3713 err = check_inode_recs(root, &root_node.inode_cache);
3717 free_corrupt_blocks_tree(&corrupt_blocks);
3718 root->fs_info->corrupt_blocks = NULL;
3719 free_orphan_data_extents(&root->orphan_data_extents);
3723 static int fs_root_objectid(u64 objectid)
3725 if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
3726 objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3728 return is_fstree(objectid);
3731 static int check_fs_roots(struct btrfs_root *root,
3732 struct cache_tree *root_cache)
3734 struct btrfs_path path;
3735 struct btrfs_key key;
3736 struct walk_control wc;
3737 struct extent_buffer *leaf, *tree_node;
3738 struct btrfs_root *tmp_root;
3739 struct btrfs_root *tree_root = root->fs_info->tree_root;
3743 if (ctx.progress_enabled) {
3744 ctx.tp = TASK_FS_ROOTS;
3745 task_start(ctx.info);
3749 * Just in case we made any changes to the extent tree that weren't
3750 * reflected into the free space cache yet.
3753 reset_cached_block_groups(root->fs_info);
3754 memset(&wc, 0, sizeof(wc));
3755 cache_tree_init(&wc.shared);
3756 btrfs_init_path(&path);
3761 key.type = BTRFS_ROOT_ITEM_KEY;
3762 ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
3767 tree_node = tree_root->node;
3769 if (tree_node != tree_root->node) {
3770 free_root_recs_tree(root_cache);
3771 btrfs_release_path(&path);
3774 leaf = path.nodes[0];
3775 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
3776 ret = btrfs_next_leaf(tree_root, &path);
3782 leaf = path.nodes[0];
3784 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
3785 if (key.type == BTRFS_ROOT_ITEM_KEY &&
3786 fs_root_objectid(key.objectid)) {
3787 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
3788 tmp_root = btrfs_read_fs_root_no_cache(
3789 root->fs_info, &key);
3791 key.offset = (u64)-1;
3792 tmp_root = btrfs_read_fs_root(
3793 root->fs_info, &key);
3795 if (IS_ERR(tmp_root)) {
3799 ret = check_fs_root(tmp_root, root_cache, &wc);
3800 if (ret == -EAGAIN) {
3801 free_root_recs_tree(root_cache);
3802 btrfs_release_path(&path);
3807 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
3808 btrfs_free_fs_root(tmp_root);
3809 } else if (key.type == BTRFS_ROOT_REF_KEY ||
3810 key.type == BTRFS_ROOT_BACKREF_KEY) {
3811 process_root_ref(leaf, path.slots[0], &key,
3818 btrfs_release_path(&path);
3820 free_extent_cache_tree(&wc.shared);
3821 if (!cache_tree_empty(&wc.shared))
3822 fprintf(stderr, "warning line %d\n", __LINE__);
3824 task_stop(ctx.info);
3829 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
3831 struct list_head *cur = rec->backrefs.next;
3832 struct extent_backref *back;
3833 struct tree_backref *tback;
3834 struct data_backref *dback;
3838 while(cur != &rec->backrefs) {
3839 back = to_extent_backref(cur);
3841 if (!back->found_extent_tree) {
3845 if (back->is_data) {
3846 dback = to_data_backref(back);
3847 fprintf(stderr, "Backref %llu %s %llu"
3848 " owner %llu offset %llu num_refs %lu"
3849 " not found in extent tree\n",
3850 (unsigned long long)rec->start,
3851 back->full_backref ?
3853 back->full_backref ?
3854 (unsigned long long)dback->parent:
3855 (unsigned long long)dback->root,
3856 (unsigned long long)dback->owner,
3857 (unsigned long long)dback->offset,
3858 (unsigned long)dback->num_refs);
3860 tback = to_tree_backref(back);
3861 fprintf(stderr, "Backref %llu parent %llu"
3862 " root %llu not found in extent tree\n",
3863 (unsigned long long)rec->start,
3864 (unsigned long long)tback->parent,
3865 (unsigned long long)tback->root);
3868 if (!back->is_data && !back->found_ref) {
3872 tback = to_tree_backref(back);
3873 fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
3874 (unsigned long long)rec->start,
3875 back->full_backref ? "parent" : "root",
3876 back->full_backref ?
3877 (unsigned long long)tback->parent :
3878 (unsigned long long)tback->root, back);
3880 if (back->is_data) {
3881 dback = to_data_backref(back);
3882 if (dback->found_ref != dback->num_refs) {
3886 fprintf(stderr, "Incorrect local backref count"
3887 " on %llu %s %llu owner %llu"
3888 " offset %llu found %u wanted %u back %p\n",
3889 (unsigned long long)rec->start,
3890 back->full_backref ?
3892 back->full_backref ?
3893 (unsigned long long)dback->parent:
3894 (unsigned long long)dback->root,
3895 (unsigned long long)dback->owner,
3896 (unsigned long long)dback->offset,
3897 dback->found_ref, dback->num_refs, back);
3899 if (dback->disk_bytenr != rec->start) {
3903 fprintf(stderr, "Backref disk bytenr does not"
3904 " match extent record, bytenr=%llu, "
3905 "ref bytenr=%llu\n",
3906 (unsigned long long)rec->start,
3907 (unsigned long long)dback->disk_bytenr);
3910 if (dback->bytes != rec->nr) {
3914 fprintf(stderr, "Backref bytes do not match "
3915 "extent backref, bytenr=%llu, ref "
3916 "bytes=%llu, backref bytes=%llu\n",
3917 (unsigned long long)rec->start,
3918 (unsigned long long)rec->nr,
3919 (unsigned long long)dback->bytes);
3922 if (!back->is_data) {
3925 dback = to_data_backref(back);
3926 found += dback->found_ref;
3929 if (found != rec->refs) {
3933 fprintf(stderr, "Incorrect global backref count "
3934 "on %llu found %llu wanted %llu\n",
3935 (unsigned long long)rec->start,
3936 (unsigned long long)found,
3937 (unsigned long long)rec->refs);
3943 static int free_all_extent_backrefs(struct extent_record *rec)
3945 struct extent_backref *back;
3946 struct list_head *cur;
3947 while (!list_empty(&rec->backrefs)) {
3948 cur = rec->backrefs.next;
3949 back = to_extent_backref(cur);
3956 static void free_extent_record_cache(struct btrfs_fs_info *fs_info,
3957 struct cache_tree *extent_cache)
3959 struct cache_extent *cache;
3960 struct extent_record *rec;
3963 cache = first_cache_extent(extent_cache);
3966 rec = container_of(cache, struct extent_record, cache);
3967 remove_cache_extent(extent_cache, cache);
3968 free_all_extent_backrefs(rec);
3973 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
3974 struct extent_record *rec)
3976 if (rec->content_checked && rec->owner_ref_checked &&
3977 rec->extent_item_refs == rec->refs && rec->refs > 0 &&
3978 rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0) &&
3979 !rec->bad_full_backref && !rec->crossing_stripes &&
3980 !rec->wrong_chunk_type) {
3981 remove_cache_extent(extent_cache, &rec->cache);
3982 free_all_extent_backrefs(rec);
3983 list_del_init(&rec->list);
3989 static int check_owner_ref(struct btrfs_root *root,
3990 struct extent_record *rec,
3991 struct extent_buffer *buf)
3993 struct extent_backref *node;
3994 struct tree_backref *back;
3995 struct btrfs_root *ref_root;
3996 struct btrfs_key key;
3997 struct btrfs_path path;
3998 struct extent_buffer *parent;
4003 list_for_each_entry(node, &rec->backrefs, list) {
4006 if (!node->found_ref)
4008 if (node->full_backref)
4010 back = to_tree_backref(node);
4011 if (btrfs_header_owner(buf) == back->root)
4014 BUG_ON(rec->is_root);
4016 /* try to find the block by search corresponding fs tree */
4017 key.objectid = btrfs_header_owner(buf);
4018 key.type = BTRFS_ROOT_ITEM_KEY;
4019 key.offset = (u64)-1;
4021 ref_root = btrfs_read_fs_root(root->fs_info, &key);
4022 if (IS_ERR(ref_root))
4025 level = btrfs_header_level(buf);
4027 btrfs_item_key_to_cpu(buf, &key, 0);
4029 btrfs_node_key_to_cpu(buf, &key, 0);
4031 btrfs_init_path(&path);
4032 path.lowest_level = level + 1;
4033 ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
4037 parent = path.nodes[level + 1];
4038 if (parent && buf->start == btrfs_node_blockptr(parent,
4039 path.slots[level + 1]))
4042 btrfs_release_path(&path);
4043 return found ? 0 : 1;
4046 static int is_extent_tree_record(struct extent_record *rec)
4048 struct list_head *cur = rec->backrefs.next;
4049 struct extent_backref *node;
4050 struct tree_backref *back;
4053 while(cur != &rec->backrefs) {
4054 node = to_extent_backref(cur);
4058 back = to_tree_backref(node);
4059 if (node->full_backref)
4061 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
4068 static int record_bad_block_io(struct btrfs_fs_info *info,
4069 struct cache_tree *extent_cache,
4072 struct extent_record *rec;
4073 struct cache_extent *cache;
4074 struct btrfs_key key;
4076 cache = lookup_cache_extent(extent_cache, start, len);
4080 rec = container_of(cache, struct extent_record, cache);
4081 if (!is_extent_tree_record(rec))
4084 btrfs_disk_key_to_cpu(&key, &rec->parent_key);
4085 return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
4088 static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
4089 struct extent_buffer *buf, int slot)
4091 if (btrfs_header_level(buf)) {
4092 struct btrfs_key_ptr ptr1, ptr2;
4094 read_extent_buffer(buf, &ptr1, btrfs_node_key_ptr_offset(slot),
4095 sizeof(struct btrfs_key_ptr));
4096 read_extent_buffer(buf, &ptr2,
4097 btrfs_node_key_ptr_offset(slot + 1),
4098 sizeof(struct btrfs_key_ptr));
4099 write_extent_buffer(buf, &ptr1,
4100 btrfs_node_key_ptr_offset(slot + 1),
4101 sizeof(struct btrfs_key_ptr));
4102 write_extent_buffer(buf, &ptr2,
4103 btrfs_node_key_ptr_offset(slot),
4104 sizeof(struct btrfs_key_ptr));
4106 struct btrfs_disk_key key;
4107 btrfs_node_key(buf, &key, 0);
4108 btrfs_fixup_low_keys(root, path, &key,
4109 btrfs_header_level(buf) + 1);
4112 struct btrfs_item *item1, *item2;
4113 struct btrfs_key k1, k2;
4114 char *item1_data, *item2_data;
4115 u32 item1_offset, item2_offset, item1_size, item2_size;
4117 item1 = btrfs_item_nr(slot);
4118 item2 = btrfs_item_nr(slot + 1);
4119 btrfs_item_key_to_cpu(buf, &k1, slot);
4120 btrfs_item_key_to_cpu(buf, &k2, slot + 1);
4121 item1_offset = btrfs_item_offset(buf, item1);
4122 item2_offset = btrfs_item_offset(buf, item2);
4123 item1_size = btrfs_item_size(buf, item1);
4124 item2_size = btrfs_item_size(buf, item2);
4126 item1_data = malloc(item1_size);
4129 item2_data = malloc(item2_size);
4135 read_extent_buffer(buf, item1_data, item1_offset, item1_size);
4136 read_extent_buffer(buf, item2_data, item2_offset, item2_size);
4138 write_extent_buffer(buf, item1_data, item2_offset, item2_size);
4139 write_extent_buffer(buf, item2_data, item1_offset, item1_size);
4143 btrfs_set_item_offset(buf, item1, item2_offset);
4144 btrfs_set_item_offset(buf, item2, item1_offset);
4145 btrfs_set_item_size(buf, item1, item2_size);
4146 btrfs_set_item_size(buf, item2, item1_size);
4148 path->slots[0] = slot;
4149 btrfs_set_item_key_unsafe(root, path, &k2);
4150 path->slots[0] = slot + 1;
4151 btrfs_set_item_key_unsafe(root, path, &k1);
4156 static int fix_key_order(struct btrfs_trans_handle *trans,
4157 struct btrfs_root *root,
4158 struct btrfs_path *path)
4160 struct extent_buffer *buf;
4161 struct btrfs_key k1, k2;
4163 int level = path->lowest_level;
4166 buf = path->nodes[level];
4167 for (i = 0; i < btrfs_header_nritems(buf) - 1; i++) {
4169 btrfs_node_key_to_cpu(buf, &k1, i);
4170 btrfs_node_key_to_cpu(buf, &k2, i + 1);
4172 btrfs_item_key_to_cpu(buf, &k1, i);
4173 btrfs_item_key_to_cpu(buf, &k2, i + 1);
4175 if (btrfs_comp_cpu_keys(&k1, &k2) < 0)
4177 ret = swap_values(root, path, buf, i);
4180 btrfs_mark_buffer_dirty(buf);
4186 static int delete_bogus_item(struct btrfs_trans_handle *trans,
4187 struct btrfs_root *root,
4188 struct btrfs_path *path,
4189 struct extent_buffer *buf, int slot)
4191 struct btrfs_key key;
4192 int nritems = btrfs_header_nritems(buf);
4194 btrfs_item_key_to_cpu(buf, &key, slot);
4196 /* These are all the keys we can deal with missing. */
4197 if (key.type != BTRFS_DIR_INDEX_KEY &&
4198 key.type != BTRFS_EXTENT_ITEM_KEY &&
4199 key.type != BTRFS_METADATA_ITEM_KEY &&
4200 key.type != BTRFS_TREE_BLOCK_REF_KEY &&
4201 key.type != BTRFS_EXTENT_DATA_REF_KEY)
4204 printf("Deleting bogus item [%llu,%u,%llu] at slot %d on block %llu\n",
4205 (unsigned long long)key.objectid, key.type,
4206 (unsigned long long)key.offset, slot, buf->start);
4207 memmove_extent_buffer(buf, btrfs_item_nr_offset(slot),
4208 btrfs_item_nr_offset(slot + 1),
4209 sizeof(struct btrfs_item) *
4210 (nritems - slot - 1));
4211 btrfs_set_header_nritems(buf, nritems - 1);
4213 struct btrfs_disk_key disk_key;
4215 btrfs_item_key(buf, &disk_key, 0);
4216 btrfs_fixup_low_keys(root, path, &disk_key, 1);
4218 btrfs_mark_buffer_dirty(buf);
4222 static int fix_item_offset(struct btrfs_trans_handle *trans,
4223 struct btrfs_root *root,
4224 struct btrfs_path *path)
4226 struct extent_buffer *buf;
4230 /* We should only get this for leaves */
4231 BUG_ON(path->lowest_level);
4232 buf = path->nodes[0];
4234 for (i = 0; i < btrfs_header_nritems(buf); i++) {
4235 unsigned int shift = 0, offset;
4237 if (i == 0 && btrfs_item_end_nr(buf, i) !=
4238 BTRFS_LEAF_DATA_SIZE(root)) {
4239 if (btrfs_item_end_nr(buf, i) >
4240 BTRFS_LEAF_DATA_SIZE(root)) {
4241 ret = delete_bogus_item(trans, root, path,
4245 fprintf(stderr, "item is off the end of the "
4246 "leaf, can't fix\n");
4250 shift = BTRFS_LEAF_DATA_SIZE(root) -
4251 btrfs_item_end_nr(buf, i);
4252 } else if (i > 0 && btrfs_item_end_nr(buf, i) !=
4253 btrfs_item_offset_nr(buf, i - 1)) {
4254 if (btrfs_item_end_nr(buf, i) >
4255 btrfs_item_offset_nr(buf, i - 1)) {
4256 ret = delete_bogus_item(trans, root, path,
4260 fprintf(stderr, "items overlap, can't fix\n");
4264 shift = btrfs_item_offset_nr(buf, i - 1) -
4265 btrfs_item_end_nr(buf, i);
4270 printf("Shifting item nr %d by %u bytes in block %llu\n",
4271 i, shift, (unsigned long long)buf->start);
4272 offset = btrfs_item_offset_nr(buf, i);
4273 memmove_extent_buffer(buf,
4274 btrfs_leaf_data(buf) + offset + shift,
4275 btrfs_leaf_data(buf) + offset,
4276 btrfs_item_size_nr(buf, i));
4277 btrfs_set_item_offset(buf, btrfs_item_nr(i),
4279 btrfs_mark_buffer_dirty(buf);
4283 * We may have moved things, in which case we want to exit so we don't
4284 * write those changes out. Once we have proper abort functionality in
4285 * progs this can be changed to something nicer.
4292 * Attempt to fix basic block failures. If we can't fix it for whatever reason
4293 * then just return -EIO.
4295 static int try_to_fix_bad_block(struct btrfs_root *root,
4296 struct extent_buffer *buf,
4297 enum btrfs_tree_block_status status)
4299 struct btrfs_trans_handle *trans;
4300 struct ulist *roots;
4301 struct ulist_node *node;
4302 struct btrfs_root *search_root;
4303 struct btrfs_path path;
4304 struct ulist_iterator iter;
4305 struct btrfs_key root_key, key;
4308 if (status != BTRFS_TREE_BLOCK_BAD_KEY_ORDER &&
4309 status != BTRFS_TREE_BLOCK_INVALID_OFFSETS)
4312 ret = btrfs_find_all_roots(NULL, root->fs_info, buf->start, 0, &roots);
4316 btrfs_init_path(&path);
4317 ULIST_ITER_INIT(&iter);
4318 while ((node = ulist_next(roots, &iter))) {
4319 root_key.objectid = node->val;
4320 root_key.type = BTRFS_ROOT_ITEM_KEY;
4321 root_key.offset = (u64)-1;
4323 search_root = btrfs_read_fs_root(root->fs_info, &root_key);
4330 trans = btrfs_start_transaction(search_root, 0);
4331 if (IS_ERR(trans)) {
4332 ret = PTR_ERR(trans);
4336 path.lowest_level = btrfs_header_level(buf);
4337 path.skip_check_block = 1;
4338 if (path.lowest_level)
4339 btrfs_node_key_to_cpu(buf, &key, 0);
4341 btrfs_item_key_to_cpu(buf, &key, 0);
4342 ret = btrfs_search_slot(trans, search_root, &key, &path, 0, 1);
4345 btrfs_commit_transaction(trans, search_root);
4348 if (status == BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
4349 ret = fix_key_order(trans, search_root, &path);
4350 else if (status == BTRFS_TREE_BLOCK_INVALID_OFFSETS)
4351 ret = fix_item_offset(trans, search_root, &path);
4353 btrfs_commit_transaction(trans, search_root);
4356 btrfs_release_path(&path);
4357 btrfs_commit_transaction(trans, search_root);
4360 btrfs_release_path(&path);
4364 static int check_block(struct btrfs_root *root,
4365 struct cache_tree *extent_cache,
4366 struct extent_buffer *buf, u64 flags)
4368 struct extent_record *rec;
4369 struct cache_extent *cache;
4370 struct btrfs_key key;
4371 enum btrfs_tree_block_status status;
4375 cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
4378 rec = container_of(cache, struct extent_record, cache);
4379 rec->generation = btrfs_header_generation(buf);
4381 level = btrfs_header_level(buf);
4382 if (btrfs_header_nritems(buf) > 0) {
4385 btrfs_item_key_to_cpu(buf, &key, 0);
4387 btrfs_node_key_to_cpu(buf, &key, 0);
4389 rec->info_objectid = key.objectid;
4391 rec->info_level = level;
4393 if (btrfs_is_leaf(buf))
4394 status = btrfs_check_leaf(root, &rec->parent_key, buf);
4396 status = btrfs_check_node(root, &rec->parent_key, buf);
4398 if (status != BTRFS_TREE_BLOCK_CLEAN) {
4400 status = try_to_fix_bad_block(root, buf, status);
4401 if (status != BTRFS_TREE_BLOCK_CLEAN) {
4403 fprintf(stderr, "bad block %llu\n",
4404 (unsigned long long)buf->start);
4407 * Signal to callers we need to start the scan over
4408 * again since we'll have cowed blocks.
4413 rec->content_checked = 1;
4414 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
4415 rec->owner_ref_checked = 1;
4417 ret = check_owner_ref(root, rec, buf);
4419 rec->owner_ref_checked = 1;
4423 maybe_free_extent_rec(extent_cache, rec);
4427 static struct tree_backref *find_tree_backref(struct extent_record *rec,
4428 u64 parent, u64 root)
4430 struct list_head *cur = rec->backrefs.next;
4431 struct extent_backref *node;
4432 struct tree_backref *back;
4434 while(cur != &rec->backrefs) {
4435 node = to_extent_backref(cur);
4439 back = to_tree_backref(node);
4441 if (!node->full_backref)
4443 if (parent == back->parent)
4446 if (node->full_backref)
4448 if (back->root == root)
4455 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
4456 u64 parent, u64 root)
4458 struct tree_backref *ref = malloc(sizeof(*ref));
4462 memset(&ref->node, 0, sizeof(ref->node));
4464 ref->parent = parent;
4465 ref->node.full_backref = 1;
4468 ref->node.full_backref = 0;
4470 list_add_tail(&ref->node.list, &rec->backrefs);
4475 static struct data_backref *find_data_backref(struct extent_record *rec,
4476 u64 parent, u64 root,
4477 u64 owner, u64 offset,
4479 u64 disk_bytenr, u64 bytes)
4481 struct list_head *cur = rec->backrefs.next;
4482 struct extent_backref *node;
4483 struct data_backref *back;
4485 while(cur != &rec->backrefs) {
4486 node = to_extent_backref(cur);
4490 back = to_data_backref(node);
4492 if (!node->full_backref)
4494 if (parent == back->parent)
4497 if (node->full_backref)
4499 if (back->root == root && back->owner == owner &&
4500 back->offset == offset) {
4501 if (found_ref && node->found_ref &&
4502 (back->bytes != bytes ||
4503 back->disk_bytenr != disk_bytenr))
4512 static struct data_backref *alloc_data_backref(struct extent_record *rec,
4513 u64 parent, u64 root,
4514 u64 owner, u64 offset,
4517 struct data_backref *ref = malloc(sizeof(*ref));
4521 memset(&ref->node, 0, sizeof(ref->node));
4522 ref->node.is_data = 1;
4525 ref->parent = parent;
4528 ref->node.full_backref = 1;
4532 ref->offset = offset;
4533 ref->node.full_backref = 0;
4535 ref->bytes = max_size;
4538 list_add_tail(&ref->node.list, &rec->backrefs);
4539 if (max_size > rec->max_size)
4540 rec->max_size = max_size;
4544 /* Check if the type of extent matches with its chunk */
4545 static void check_extent_type(struct extent_record *rec)
4547 struct btrfs_block_group_cache *bg_cache;
4549 bg_cache = btrfs_lookup_first_block_group(global_info, rec->start);
4553 /* data extent, check chunk directly*/
4554 if (!rec->metadata) {
4555 if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_DATA))
4556 rec->wrong_chunk_type = 1;
4560 /* metadata extent, check the obvious case first */
4561 if (!(bg_cache->flags & (BTRFS_BLOCK_GROUP_SYSTEM |
4562 BTRFS_BLOCK_GROUP_METADATA))) {
4563 rec->wrong_chunk_type = 1;
4568 * Check SYSTEM extent, as it's also marked as metadata, we can only
4569 * make sure it's a SYSTEM extent by its backref
4571 if (!list_empty(&rec->backrefs)) {
4572 struct extent_backref *node;
4573 struct tree_backref *tback;
4576 node = to_extent_backref(rec->backrefs.next);
4577 if (node->is_data) {
4578 /* tree block shouldn't have data backref */
4579 rec->wrong_chunk_type = 1;
4582 tback = container_of(node, struct tree_backref, node);
4584 if (tback->root == BTRFS_CHUNK_TREE_OBJECTID)
4585 bg_type = BTRFS_BLOCK_GROUP_SYSTEM;
4587 bg_type = BTRFS_BLOCK_GROUP_METADATA;
4588 if (!(bg_cache->flags & bg_type))
4589 rec->wrong_chunk_type = 1;
4594 * Allocate a new extent record, fill default values from @tmpl and insert int
4595 * @extent_cache. Caller is supposed to make sure the [start,nr) is not in
4596 * the cache, otherwise it fails.
4598 static int add_extent_rec_nolookup(struct cache_tree *extent_cache,
4599 struct extent_record *tmpl)
4601 struct extent_record *rec;
4604 rec = malloc(sizeof(*rec));
4607 rec->start = tmpl->start;
4608 rec->max_size = tmpl->max_size;
4609 rec->nr = max(tmpl->nr, tmpl->max_size);
4610 rec->found_rec = tmpl->found_rec;
4611 rec->content_checked = tmpl->content_checked;
4612 rec->owner_ref_checked = tmpl->owner_ref_checked;
4613 rec->num_duplicates = 0;
4614 rec->metadata = tmpl->metadata;
4615 rec->flag_block_full_backref = FLAG_UNSET;
4616 rec->bad_full_backref = 0;
4617 rec->crossing_stripes = 0;
4618 rec->wrong_chunk_type = 0;
4619 rec->is_root = tmpl->is_root;
4620 rec->refs = tmpl->refs;
4621 rec->extent_item_refs = tmpl->extent_item_refs;
4622 rec->parent_generation = tmpl->parent_generation;
4623 INIT_LIST_HEAD(&rec->backrefs);
4624 INIT_LIST_HEAD(&rec->dups);
4625 INIT_LIST_HEAD(&rec->list);
4626 memcpy(&rec->parent_key, &tmpl->parent_key, sizeof(tmpl->parent_key));
4627 rec->cache.start = tmpl->start;
4628 rec->cache.size = tmpl->nr;
4629 ret = insert_cache_extent(extent_cache, &rec->cache);
4634 bytes_used += rec->nr;
4637 rec->crossing_stripes = check_crossing_stripes(global_info,
4638 rec->start, global_info->tree_root->nodesize);
4639 check_extent_type(rec);
4644 * Lookup and modify an extent, some values of @tmpl are interpreted verbatim,
4646 * - refs - if found, increase refs
4647 * - is_root - if found, set
4648 * - content_checked - if found, set
4649 * - owner_ref_checked - if found, set
4651 * If not found, create a new one, initialize and insert.
4653 static int add_extent_rec(struct cache_tree *extent_cache,
4654 struct extent_record *tmpl)
4656 struct extent_record *rec;
4657 struct cache_extent *cache;
4661 cache = lookup_cache_extent(extent_cache, tmpl->start, tmpl->nr);
4663 rec = container_of(cache, struct extent_record, cache);
4667 rec->nr = max(tmpl->nr, tmpl->max_size);
4670 * We need to make sure to reset nr to whatever the extent
4671 * record says was the real size, this way we can compare it to
4674 if (tmpl->found_rec) {
4675 if (tmpl->start != rec->start || rec->found_rec) {
4676 struct extent_record *tmp;
4679 if (list_empty(&rec->list))
4680 list_add_tail(&rec->list,
4681 &duplicate_extents);
4684 * We have to do this song and dance in case we
4685 * find an extent record that falls inside of
4686 * our current extent record but does not have
4687 * the same objectid.
4689 tmp = malloc(sizeof(*tmp));
4692 tmp->start = tmpl->start;
4693 tmp->max_size = tmpl->max_size;
4696 tmp->metadata = tmpl->metadata;
4697 tmp->extent_item_refs = tmpl->extent_item_refs;
4698 INIT_LIST_HEAD(&tmp->list);
4699 list_add_tail(&tmp->list, &rec->dups);
4700 rec->num_duplicates++;
4707 if (tmpl->extent_item_refs && !dup) {
4708 if (rec->extent_item_refs) {
4709 fprintf(stderr, "block %llu rec "
4710 "extent_item_refs %llu, passed %llu\n",
4711 (unsigned long long)tmpl->start,
4712 (unsigned long long)
4713 rec->extent_item_refs,
4714 (unsigned long long)tmpl->extent_item_refs);
4716 rec->extent_item_refs = tmpl->extent_item_refs;
4720 if (tmpl->content_checked)
4721 rec->content_checked = 1;
4722 if (tmpl->owner_ref_checked)
4723 rec->owner_ref_checked = 1;
4724 memcpy(&rec->parent_key, &tmpl->parent_key,
4725 sizeof(tmpl->parent_key));
4726 if (tmpl->parent_generation)
4727 rec->parent_generation = tmpl->parent_generation;
4728 if (rec->max_size < tmpl->max_size)
4729 rec->max_size = tmpl->max_size;
4732 * A metadata extent can't cross stripe_len boundary, otherwise
4733 * kernel scrub won't be able to handle it.
4734 * As now stripe_len is fixed to BTRFS_STRIPE_LEN, just check
4738 rec->crossing_stripes = check_crossing_stripes(
4739 global_info, rec->start,
4740 global_info->tree_root->nodesize);
4741 check_extent_type(rec);
4742 maybe_free_extent_rec(extent_cache, rec);
4746 ret = add_extent_rec_nolookup(extent_cache, tmpl);
4751 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
4752 u64 parent, u64 root, int found_ref)
4754 struct extent_record *rec;
4755 struct tree_backref *back;
4756 struct cache_extent *cache;
4759 cache = lookup_cache_extent(extent_cache, bytenr, 1);
4761 struct extent_record tmpl;
4763 memset(&tmpl, 0, sizeof(tmpl));
4764 tmpl.start = bytenr;
4768 ret = add_extent_rec_nolookup(extent_cache, &tmpl);
4772 /* really a bug in cache_extent implement now */
4773 cache = lookup_cache_extent(extent_cache, bytenr, 1);
4778 rec = container_of(cache, struct extent_record, cache);
4779 if (rec->start != bytenr) {
4781 * Several cause, from unaligned bytenr to over lapping extents
4786 back = find_tree_backref(rec, parent, root);
4788 back = alloc_tree_backref(rec, parent, root);
4794 if (back->node.found_ref) {
4795 fprintf(stderr, "Extent back ref already exists "
4796 "for %llu parent %llu root %llu \n",
4797 (unsigned long long)bytenr,
4798 (unsigned long long)parent,
4799 (unsigned long long)root);
4801 back->node.found_ref = 1;
4803 if (back->node.found_extent_tree) {
4804 fprintf(stderr, "Extent back ref already exists "
4805 "for %llu parent %llu root %llu \n",
4806 (unsigned long long)bytenr,
4807 (unsigned long long)parent,
4808 (unsigned long long)root);
4810 back->node.found_extent_tree = 1;
4812 check_extent_type(rec);
4813 maybe_free_extent_rec(extent_cache, rec);
4817 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
4818 u64 parent, u64 root, u64 owner, u64 offset,
4819 u32 num_refs, int found_ref, u64 max_size)
4821 struct extent_record *rec;
4822 struct data_backref *back;
4823 struct cache_extent *cache;
4826 cache = lookup_cache_extent(extent_cache, bytenr, 1);
4828 struct extent_record tmpl;
4830 memset(&tmpl, 0, sizeof(tmpl));
4831 tmpl.start = bytenr;
4833 tmpl.max_size = max_size;
4835 ret = add_extent_rec_nolookup(extent_cache, &tmpl);
4839 cache = lookup_cache_extent(extent_cache, bytenr, 1);
4844 rec = container_of(cache, struct extent_record, cache);
4845 if (rec->max_size < max_size)
4846 rec->max_size = max_size;
4849 * If found_ref is set then max_size is the real size and must match the
4850 * existing refs. So if we have already found a ref then we need to
4851 * make sure that this ref matches the existing one, otherwise we need
4852 * to add a new backref so we can notice that the backrefs don't match
4853 * and we need to figure out who is telling the truth. This is to
4854 * account for that awful fsync bug I introduced where we'd end up with
4855 * a btrfs_file_extent_item that would have its length include multiple
4856 * prealloc extents or point inside of a prealloc extent.
4858 back = find_data_backref(rec, parent, root, owner, offset, found_ref,
4861 back = alloc_data_backref(rec, parent, root, owner, offset,
4867 BUG_ON(num_refs != 1);
4868 if (back->node.found_ref)
4869 BUG_ON(back->bytes != max_size);
4870 back->node.found_ref = 1;
4871 back->found_ref += 1;
4872 back->bytes = max_size;
4873 back->disk_bytenr = bytenr;
4875 rec->content_checked = 1;
4876 rec->owner_ref_checked = 1;
4878 if (back->node.found_extent_tree) {
4879 fprintf(stderr, "Extent back ref already exists "
4880 "for %llu parent %llu root %llu "
4881 "owner %llu offset %llu num_refs %lu\n",
4882 (unsigned long long)bytenr,
4883 (unsigned long long)parent,
4884 (unsigned long long)root,
4885 (unsigned long long)owner,
4886 (unsigned long long)offset,
4887 (unsigned long)num_refs);
4889 back->num_refs = num_refs;
4890 back->node.found_extent_tree = 1;
4892 maybe_free_extent_rec(extent_cache, rec);
4896 static int add_pending(struct cache_tree *pending,
4897 struct cache_tree *seen, u64 bytenr, u32 size)
4900 ret = add_cache_extent(seen, bytenr, size);
4903 add_cache_extent(pending, bytenr, size);
4907 static int pick_next_pending(struct cache_tree *pending,
4908 struct cache_tree *reada,
4909 struct cache_tree *nodes,
4910 u64 last, struct block_info *bits, int bits_nr,
4913 unsigned long node_start = last;
4914 struct cache_extent *cache;
4917 cache = search_cache_extent(reada, 0);
4919 bits[0].start = cache->start;
4920 bits[0].size = cache->size;
4925 if (node_start > 32768)
4926 node_start -= 32768;
4928 cache = search_cache_extent(nodes, node_start);
4930 cache = search_cache_extent(nodes, 0);
4933 cache = search_cache_extent(pending, 0);
4938 bits[ret].start = cache->start;
4939 bits[ret].size = cache->size;
4940 cache = next_cache_extent(cache);
4942 } while (cache && ret < bits_nr);
4948 bits[ret].start = cache->start;
4949 bits[ret].size = cache->size;
4950 cache = next_cache_extent(cache);
4952 } while (cache && ret < bits_nr);
4954 if (bits_nr - ret > 8) {
4955 u64 lookup = bits[0].start + bits[0].size;
4956 struct cache_extent *next;
4957 next = search_cache_extent(pending, lookup);
4959 if (next->start - lookup > 32768)
4961 bits[ret].start = next->start;
4962 bits[ret].size = next->size;
4963 lookup = next->start + next->size;
4967 next = next_cache_extent(next);
4975 static void free_chunk_record(struct cache_extent *cache)
4977 struct chunk_record *rec;
4979 rec = container_of(cache, struct chunk_record, cache);
4980 list_del_init(&rec->list);
4981 list_del_init(&rec->dextents);
4985 void free_chunk_cache_tree(struct cache_tree *chunk_cache)
4987 cache_tree_free_extents(chunk_cache, free_chunk_record);
4990 static void free_device_record(struct rb_node *node)
4992 struct device_record *rec;
4994 rec = container_of(node, struct device_record, node);
4998 FREE_RB_BASED_TREE(device_cache, free_device_record);
5000 int insert_block_group_record(struct block_group_tree *tree,
5001 struct block_group_record *bg_rec)
5005 ret = insert_cache_extent(&tree->tree, &bg_rec->cache);
5009 list_add_tail(&bg_rec->list, &tree->block_groups);
5013 static void free_block_group_record(struct cache_extent *cache)
5015 struct block_group_record *rec;
5017 rec = container_of(cache, struct block_group_record, cache);
5018 list_del_init(&rec->list);
5022 void free_block_group_tree(struct block_group_tree *tree)
5024 cache_tree_free_extents(&tree->tree, free_block_group_record);
5027 int insert_device_extent_record(struct device_extent_tree *tree,
5028 struct device_extent_record *de_rec)
5033 * Device extent is a bit different from the other extents, because
5034 * the extents which belong to the different devices may have the
5035 * same start and size, so we need use the special extent cache
5036 * search/insert functions.
5038 ret = insert_cache_extent2(&tree->tree, &de_rec->cache);
5042 list_add_tail(&de_rec->chunk_list, &tree->no_chunk_orphans);
5043 list_add_tail(&de_rec->device_list, &tree->no_device_orphans);
5047 static void free_device_extent_record(struct cache_extent *cache)
5049 struct device_extent_record *rec;
5051 rec = container_of(cache, struct device_extent_record, cache);
5052 if (!list_empty(&rec->chunk_list))
5053 list_del_init(&rec->chunk_list);
5054 if (!list_empty(&rec->device_list))
5055 list_del_init(&rec->device_list);
5059 void free_device_extent_tree(struct device_extent_tree *tree)
5061 cache_tree_free_extents(&tree->tree, free_device_extent_record);
5064 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5065 static int process_extent_ref_v0(struct cache_tree *extent_cache,
5066 struct extent_buffer *leaf, int slot)
5068 struct btrfs_extent_ref_v0 *ref0;
5069 struct btrfs_key key;
5072 btrfs_item_key_to_cpu(leaf, &key, slot);
5073 ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
5074 if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
5075 ret = add_tree_backref(extent_cache, key.objectid, key.offset,
5078 ret = add_data_backref(extent_cache, key.objectid, key.offset,
5079 0, 0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
5085 struct chunk_record *btrfs_new_chunk_record(struct extent_buffer *leaf,
5086 struct btrfs_key *key,
5089 struct btrfs_chunk *ptr;
5090 struct chunk_record *rec;
5093 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5094 num_stripes = btrfs_chunk_num_stripes(leaf, ptr);
5096 rec = calloc(1, btrfs_chunk_record_size(num_stripes));
5098 fprintf(stderr, "memory allocation failed\n");
5102 INIT_LIST_HEAD(&rec->list);
5103 INIT_LIST_HEAD(&rec->dextents);
5106 rec->cache.start = key->offset;
5107 rec->cache.size = btrfs_chunk_length(leaf, ptr);
5109 rec->generation = btrfs_header_generation(leaf);
5111 rec->objectid = key->objectid;
5112 rec->type = key->type;
5113 rec->offset = key->offset;
5115 rec->length = rec->cache.size;
5116 rec->owner = btrfs_chunk_owner(leaf, ptr);
5117 rec->stripe_len = btrfs_chunk_stripe_len(leaf, ptr);
5118 rec->type_flags = btrfs_chunk_type(leaf, ptr);
5119 rec->io_width = btrfs_chunk_io_width(leaf, ptr);
5120 rec->io_align = btrfs_chunk_io_align(leaf, ptr);
5121 rec->sector_size = btrfs_chunk_sector_size(leaf, ptr);
5122 rec->num_stripes = num_stripes;
5123 rec->sub_stripes = btrfs_chunk_sub_stripes(leaf, ptr);
5125 for (i = 0; i < rec->num_stripes; ++i) {
5126 rec->stripes[i].devid =
5127 btrfs_stripe_devid_nr(leaf, ptr, i);
5128 rec->stripes[i].offset =
5129 btrfs_stripe_offset_nr(leaf, ptr, i);
5130 read_extent_buffer(leaf, rec->stripes[i].dev_uuid,
5131 (unsigned long)btrfs_stripe_dev_uuid_nr(ptr, i),
5138 static int process_chunk_item(struct cache_tree *chunk_cache,
5139 struct btrfs_key *key, struct extent_buffer *eb,
5142 struct chunk_record *rec;
5143 struct btrfs_chunk *chunk;
5146 chunk = btrfs_item_ptr(eb, slot, struct btrfs_chunk);
5148 * Do extra check for this chunk item,
5150 * It's still possible one can craft a leaf with CHUNK_ITEM, with
5151 * wrong onwer(3) out of chunk tree, to pass both chunk tree check
5152 * and owner<->key_type check.
5154 ret = btrfs_check_chunk_valid(global_info->tree_root, eb, chunk, slot,
5157 error("chunk(%llu, %llu) is not valid, ignore it",
5158 key->offset, btrfs_chunk_length(eb, chunk));
5161 rec = btrfs_new_chunk_record(eb, key, slot);
5162 ret = insert_cache_extent(chunk_cache, &rec->cache);
5164 fprintf(stderr, "Chunk[%llu, %llu] existed.\n",
5165 rec->offset, rec->length);
5172 static int process_device_item(struct rb_root *dev_cache,
5173 struct btrfs_key *key, struct extent_buffer *eb, int slot)
5175 struct btrfs_dev_item *ptr;
5176 struct device_record *rec;
5179 ptr = btrfs_item_ptr(eb,
5180 slot, struct btrfs_dev_item);
5182 rec = malloc(sizeof(*rec));
5184 fprintf(stderr, "memory allocation failed\n");
5188 rec->devid = key->offset;
5189 rec->generation = btrfs_header_generation(eb);
5191 rec->objectid = key->objectid;
5192 rec->type = key->type;
5193 rec->offset = key->offset;
5195 rec->devid = btrfs_device_id(eb, ptr);
5196 rec->total_byte = btrfs_device_total_bytes(eb, ptr);
5197 rec->byte_used = btrfs_device_bytes_used(eb, ptr);
5199 ret = rb_insert(dev_cache, &rec->node, device_record_compare);
5201 fprintf(stderr, "Device[%llu] existed.\n", rec->devid);
5208 struct block_group_record *
5209 btrfs_new_block_group_record(struct extent_buffer *leaf, struct btrfs_key *key,
5212 struct btrfs_block_group_item *ptr;
5213 struct block_group_record *rec;
5215 rec = calloc(1, sizeof(*rec));
5217 fprintf(stderr, "memory allocation failed\n");
5221 rec->cache.start = key->objectid;
5222 rec->cache.size = key->offset;
5224 rec->generation = btrfs_header_generation(leaf);
5226 rec->objectid = key->objectid;
5227 rec->type = key->type;
5228 rec->offset = key->offset;
5230 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item);
5231 rec->flags = btrfs_disk_block_group_flags(leaf, ptr);
5233 INIT_LIST_HEAD(&rec->list);
5238 static int process_block_group_item(struct block_group_tree *block_group_cache,
5239 struct btrfs_key *key,
5240 struct extent_buffer *eb, int slot)
5242 struct block_group_record *rec;
5245 rec = btrfs_new_block_group_record(eb, key, slot);
5246 ret = insert_block_group_record(block_group_cache, rec);
5248 fprintf(stderr, "Block Group[%llu, %llu] existed.\n",
5249 rec->objectid, rec->offset);
5256 struct device_extent_record *
5257 btrfs_new_device_extent_record(struct extent_buffer *leaf,
5258 struct btrfs_key *key, int slot)
5260 struct device_extent_record *rec;
5261 struct btrfs_dev_extent *ptr;
5263 rec = calloc(1, sizeof(*rec));
5265 fprintf(stderr, "memory allocation failed\n");
5269 rec->cache.objectid = key->objectid;
5270 rec->cache.start = key->offset;
5272 rec->generation = btrfs_header_generation(leaf);
5274 rec->objectid = key->objectid;
5275 rec->type = key->type;
5276 rec->offset = key->offset;
5278 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
5279 rec->chunk_objecteid =
5280 btrfs_dev_extent_chunk_objectid(leaf, ptr);
5282 btrfs_dev_extent_chunk_offset(leaf, ptr);
5283 rec->length = btrfs_dev_extent_length(leaf, ptr);
5284 rec->cache.size = rec->length;
5286 INIT_LIST_HEAD(&rec->chunk_list);
5287 INIT_LIST_HEAD(&rec->device_list);
5293 process_device_extent_item(struct device_extent_tree *dev_extent_cache,
5294 struct btrfs_key *key, struct extent_buffer *eb,
5297 struct device_extent_record *rec;
5300 rec = btrfs_new_device_extent_record(eb, key, slot);
5301 ret = insert_device_extent_record(dev_extent_cache, rec);
5304 "Device extent[%llu, %llu, %llu] existed.\n",
5305 rec->objectid, rec->offset, rec->length);
5312 static int process_extent_item(struct btrfs_root *root,
5313 struct cache_tree *extent_cache,
5314 struct extent_buffer *eb, int slot)
5316 struct btrfs_extent_item *ei;
5317 struct btrfs_extent_inline_ref *iref;
5318 struct btrfs_extent_data_ref *dref;
5319 struct btrfs_shared_data_ref *sref;
5320 struct btrfs_key key;
5321 struct extent_record tmpl;
5326 u32 item_size = btrfs_item_size_nr(eb, slot);
5332 btrfs_item_key_to_cpu(eb, &key, slot);
5334 if (key.type == BTRFS_METADATA_ITEM_KEY) {
5336 num_bytes = root->nodesize;
5338 num_bytes = key.offset;
5341 if (!IS_ALIGNED(key.objectid, root->sectorsize)) {
5342 error("ignoring invalid extent, bytenr %llu is not aligned to %u",
5343 key.objectid, root->sectorsize);
5346 if (item_size < sizeof(*ei)) {
5347 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5348 struct btrfs_extent_item_v0 *ei0;
5349 BUG_ON(item_size != sizeof(*ei0));
5350 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
5351 refs = btrfs_extent_refs_v0(eb, ei0);
5355 memset(&tmpl, 0, sizeof(tmpl));
5356 tmpl.start = key.objectid;
5357 tmpl.nr = num_bytes;
5358 tmpl.extent_item_refs = refs;
5359 tmpl.metadata = metadata;
5361 tmpl.max_size = num_bytes;
5363 return add_extent_rec(extent_cache, &tmpl);
5366 ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
5367 refs = btrfs_extent_refs(eb, ei);
5368 if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK)
5372 if (metadata && num_bytes != root->nodesize) {
5373 error("ignore invalid metadata extent, length %llu does not equal to %u",
5374 num_bytes, root->nodesize);
5377 if (!metadata && !IS_ALIGNED(num_bytes, root->sectorsize)) {
5378 error("ignore invalid data extent, length %llu is not aligned to %u",
5379 num_bytes, root->sectorsize);
5383 memset(&tmpl, 0, sizeof(tmpl));
5384 tmpl.start = key.objectid;
5385 tmpl.nr = num_bytes;
5386 tmpl.extent_item_refs = refs;
5387 tmpl.metadata = metadata;
5389 tmpl.max_size = num_bytes;
5390 add_extent_rec(extent_cache, &tmpl);
5392 ptr = (unsigned long)(ei + 1);
5393 if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
5394 key.type == BTRFS_EXTENT_ITEM_KEY)
5395 ptr += sizeof(struct btrfs_tree_block_info);
5397 end = (unsigned long)ei + item_size;
5399 iref = (struct btrfs_extent_inline_ref *)ptr;
5400 type = btrfs_extent_inline_ref_type(eb, iref);
5401 offset = btrfs_extent_inline_ref_offset(eb, iref);
5403 case BTRFS_TREE_BLOCK_REF_KEY:
5404 ret = add_tree_backref(extent_cache, key.objectid,
5407 error("add_tree_backref failed: %s",
5410 case BTRFS_SHARED_BLOCK_REF_KEY:
5411 ret = add_tree_backref(extent_cache, key.objectid,
5414 error("add_tree_backref failed: %s",
5417 case BTRFS_EXTENT_DATA_REF_KEY:
5418 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
5419 add_data_backref(extent_cache, key.objectid, 0,
5420 btrfs_extent_data_ref_root(eb, dref),
5421 btrfs_extent_data_ref_objectid(eb,
5423 btrfs_extent_data_ref_offset(eb, dref),
5424 btrfs_extent_data_ref_count(eb, dref),
5427 case BTRFS_SHARED_DATA_REF_KEY:
5428 sref = (struct btrfs_shared_data_ref *)(iref + 1);
5429 add_data_backref(extent_cache, key.objectid, offset,
5431 btrfs_shared_data_ref_count(eb, sref),
5435 fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
5436 key.objectid, key.type, num_bytes);
5439 ptr += btrfs_extent_inline_ref_size(type);
5446 static int check_cache_range(struct btrfs_root *root,
5447 struct btrfs_block_group_cache *cache,
5448 u64 offset, u64 bytes)
5450 struct btrfs_free_space *entry;
5456 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
5457 bytenr = btrfs_sb_offset(i);
5458 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
5459 cache->key.objectid, bytenr, 0,
5460 &logical, &nr, &stripe_len);
5465 if (logical[nr] + stripe_len <= offset)
5467 if (offset + bytes <= logical[nr])
5469 if (logical[nr] == offset) {
5470 if (stripe_len >= bytes) {
5474 bytes -= stripe_len;
5475 offset += stripe_len;
5476 } else if (logical[nr] < offset) {
5477 if (logical[nr] + stripe_len >=
5482 bytes = (offset + bytes) -
5483 (logical[nr] + stripe_len);
5484 offset = logical[nr] + stripe_len;
5487 * Could be tricky, the super may land in the
5488 * middle of the area we're checking. First
5489 * check the easiest case, it's at the end.
5491 if (logical[nr] + stripe_len >=
5493 bytes = logical[nr] - offset;
5497 /* Check the left side */
5498 ret = check_cache_range(root, cache,
5500 logical[nr] - offset);
5506 /* Now we continue with the right side */
5507 bytes = (offset + bytes) -
5508 (logical[nr] + stripe_len);
5509 offset = logical[nr] + stripe_len;
5516 entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
5518 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
5519 offset, offset+bytes);
5523 if (entry->offset != offset) {
5524 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
5529 if (entry->bytes != bytes) {
5530 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
5531 bytes, entry->bytes, offset);
5535 unlink_free_space(cache->free_space_ctl, entry);
5540 static int verify_space_cache(struct btrfs_root *root,
5541 struct btrfs_block_group_cache *cache)
5543 struct btrfs_path path;
5544 struct extent_buffer *leaf;
5545 struct btrfs_key key;
5549 root = root->fs_info->extent_root;
5551 last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
5553 btrfs_init_path(&path);
5554 key.objectid = last;
5556 key.type = BTRFS_EXTENT_ITEM_KEY;
5557 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
5562 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
5563 ret = btrfs_next_leaf(root, &path);
5571 leaf = path.nodes[0];
5572 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
5573 if (key.objectid >= cache->key.offset + cache->key.objectid)
5575 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
5576 key.type != BTRFS_METADATA_ITEM_KEY) {
5581 if (last == key.objectid) {
5582 if (key.type == BTRFS_EXTENT_ITEM_KEY)
5583 last = key.objectid + key.offset;
5585 last = key.objectid + root->nodesize;
5590 ret = check_cache_range(root, cache, last,
5591 key.objectid - last);
5594 if (key.type == BTRFS_EXTENT_ITEM_KEY)
5595 last = key.objectid + key.offset;
5597 last = key.objectid + root->nodesize;
5601 if (last < cache->key.objectid + cache->key.offset)
5602 ret = check_cache_range(root, cache, last,
5603 cache->key.objectid +
5604 cache->key.offset - last);
5607 btrfs_release_path(&path);
5610 !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
5611 fprintf(stderr, "There are still entries left in the space "
5619 static int check_space_cache(struct btrfs_root *root)
5621 struct btrfs_block_group_cache *cache;
5622 u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
5626 if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
5627 btrfs_super_generation(root->fs_info->super_copy) !=
5628 btrfs_super_cache_generation(root->fs_info->super_copy)) {
5629 printf("cache and super generation don't match, space cache "
5630 "will be invalidated\n");
5634 if (ctx.progress_enabled) {
5635 ctx.tp = TASK_FREE_SPACE;
5636 task_start(ctx.info);
5640 cache = btrfs_lookup_first_block_group(root->fs_info, start);
5644 start = cache->key.objectid + cache->key.offset;
5645 if (!cache->free_space_ctl) {
5646 if (btrfs_init_free_space_ctl(cache,
5647 root->sectorsize)) {
5652 btrfs_remove_free_space_cache(cache);
5655 if (btrfs_fs_compat_ro(root->fs_info,
5656 BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE)) {
5657 ret = exclude_super_stripes(root, cache);
5659 fprintf(stderr, "could not exclude super stripes: %s\n",
5664 ret = load_free_space_tree(root->fs_info, cache);
5665 free_excluded_extents(root, cache);
5667 fprintf(stderr, "could not load free space tree: %s\n",
5674 ret = load_free_space_cache(root->fs_info, cache);
5679 ret = verify_space_cache(root, cache);
5681 fprintf(stderr, "cache appears valid but isn't %Lu\n",
5682 cache->key.objectid);
5687 task_stop(ctx.info);
5689 return error ? -EINVAL : 0;
5692 static int check_extent_csums(struct btrfs_root *root, u64 bytenr,
5693 u64 num_bytes, unsigned long leaf_offset,
5694 struct extent_buffer *eb) {
5697 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
5699 unsigned long csum_offset;
5703 u64 data_checked = 0;
5709 if (num_bytes % root->sectorsize)
5712 data = malloc(num_bytes);
5716 while (offset < num_bytes) {
5719 read_len = num_bytes - offset;
5720 /* read as much space once a time */
5721 ret = read_extent_data(root, data + offset,
5722 bytenr + offset, &read_len, mirror);
5726 /* verify every 4k data's checksum */
5727 while (data_checked < read_len) {
5729 tmp = offset + data_checked;
5731 csum = btrfs_csum_data(NULL, (char *)data + tmp,
5732 csum, root->sectorsize);
5733 btrfs_csum_final(csum, (u8 *)&csum);
5735 csum_offset = leaf_offset +
5736 tmp / root->sectorsize * csum_size;
5737 read_extent_buffer(eb, (char *)&csum_expected,
5738 csum_offset, csum_size);
5739 /* try another mirror */
5740 if (csum != csum_expected) {
5741 fprintf(stderr, "mirror %d bytenr %llu csum %u expected csum %u\n",
5742 mirror, bytenr + tmp,
5743 csum, csum_expected);
5744 num_copies = btrfs_num_copies(
5745 &root->fs_info->mapping_tree,
5747 if (mirror < num_copies - 1) {
5752 data_checked += root->sectorsize;
5761 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
5764 struct btrfs_path *path;
5765 struct extent_buffer *leaf;
5766 struct btrfs_key key;
5769 path = btrfs_alloc_path();
5771 fprintf(stderr, "Error allocating path\n");
5775 key.objectid = bytenr;
5776 key.type = BTRFS_EXTENT_ITEM_KEY;
5777 key.offset = (u64)-1;
5780 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
5783 fprintf(stderr, "Error looking up extent record %d\n", ret);
5784 btrfs_free_path(path);
5787 if (path->slots[0] > 0) {
5790 ret = btrfs_prev_leaf(root, path);
5793 } else if (ret > 0) {
5800 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5803 * Block group items come before extent items if they have the same
5804 * bytenr, so walk back one more just in case. Dear future traveller,
5805 * first congrats on mastering time travel. Now if it's not too much
5806 * trouble could you go back to 2006 and tell Chris to make the
5807 * BLOCK_GROUP_ITEM_KEY (and BTRFS_*_REF_KEY) lower than the
5808 * EXTENT_ITEM_KEY please?
5810 while (key.type > BTRFS_EXTENT_ITEM_KEY) {
5811 if (path->slots[0] > 0) {
5814 ret = btrfs_prev_leaf(root, path);
5817 } else if (ret > 0) {
5822 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
5826 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5827 ret = btrfs_next_leaf(root, path);
5829 fprintf(stderr, "Error going to next leaf "
5831 btrfs_free_path(path);
5837 leaf = path->nodes[0];
5838 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5839 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
5843 if (key.objectid + key.offset < bytenr) {
5847 if (key.objectid > bytenr + num_bytes)
5850 if (key.objectid == bytenr) {
5851 if (key.offset >= num_bytes) {
5855 num_bytes -= key.offset;
5856 bytenr += key.offset;
5857 } else if (key.objectid < bytenr) {
5858 if (key.objectid + key.offset >= bytenr + num_bytes) {
5862 num_bytes = (bytenr + num_bytes) -
5863 (key.objectid + key.offset);
5864 bytenr = key.objectid + key.offset;
5866 if (key.objectid + key.offset < bytenr + num_bytes) {
5867 u64 new_start = key.objectid + key.offset;
5868 u64 new_bytes = bytenr + num_bytes - new_start;
5871 * Weird case, the extent is in the middle of
5872 * our range, we'll have to search one side
5873 * and then the other. Not sure if this happens
5874 * in real life, but no harm in coding it up
5875 * anyway just in case.
5877 btrfs_release_path(path);
5878 ret = check_extent_exists(root, new_start,
5881 fprintf(stderr, "Right section didn't "
5885 num_bytes = key.objectid - bytenr;
5888 num_bytes = key.objectid - bytenr;
5895 if (num_bytes && !ret) {
5896 fprintf(stderr, "There are no extents for csum range "
5897 "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
5901 btrfs_free_path(path);
5905 static int check_csums(struct btrfs_root *root)
5907 struct btrfs_path *path;
5908 struct extent_buffer *leaf;
5909 struct btrfs_key key;
5910 u64 offset = 0, num_bytes = 0;
5911 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
5915 unsigned long leaf_offset;
5917 root = root->fs_info->csum_root;
5918 if (!extent_buffer_uptodate(root->node)) {
5919 fprintf(stderr, "No valid csum tree found\n");
5923 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
5924 key.type = BTRFS_EXTENT_CSUM_KEY;
5927 path = btrfs_alloc_path();
5931 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5933 fprintf(stderr, "Error searching csum tree %d\n", ret);
5934 btrfs_free_path(path);
5938 if (ret > 0 && path->slots[0])
5943 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5944 ret = btrfs_next_leaf(root, path);
5946 fprintf(stderr, "Error going to next leaf "
5953 leaf = path->nodes[0];
5955 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5956 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
5961 data_len = (btrfs_item_size_nr(leaf, path->slots[0]) /
5962 csum_size) * root->sectorsize;
5963 if (!check_data_csum)
5964 goto skip_csum_check;
5965 leaf_offset = btrfs_item_ptr_offset(leaf, path->slots[0]);
5966 ret = check_extent_csums(root, key.offset, data_len,
5972 offset = key.offset;
5973 } else if (key.offset != offset + num_bytes) {
5974 ret = check_extent_exists(root, offset, num_bytes);
5976 fprintf(stderr, "Csum exists for %Lu-%Lu but "
5977 "there is no extent record\n",
5978 offset, offset+num_bytes);
5981 offset = key.offset;
5984 num_bytes += data_len;
5988 btrfs_free_path(path);
5992 static int is_dropped_key(struct btrfs_key *key,
5993 struct btrfs_key *drop_key) {
5994 if (key->objectid < drop_key->objectid)
5996 else if (key->objectid == drop_key->objectid) {
5997 if (key->type < drop_key->type)
5999 else if (key->type == drop_key->type) {
6000 if (key->offset < drop_key->offset)
6008 * Here are the rules for FULL_BACKREF.
6010 * 1) If BTRFS_HEADER_FLAG_RELOC is set then we have FULL_BACKREF set.
6011 * 2) If btrfs_header_owner(buf) no longer points to buf then we have
6013 * 3) We cowed the block walking down a reloc tree. This is impossible to tell
6014 * if it happened after the relocation occurred since we'll have dropped the
6015 * reloc root, so it's entirely possible to have FULL_BACKREF set on buf and
6016 * have no real way to know for sure.
6018 * We process the blocks one root at a time, and we start from the lowest root
6019 * objectid and go to the highest. So we can just lookup the owner backref for
6020 * the record and if we don't find it then we know it doesn't exist and we have
6023 * FIXME: if we ever start reclaiming root objectid's then we need to fix this
6024 * assumption and simply indicate that we _think_ that the FULL BACKREF needs to
6025 * be set or not and then we can check later once we've gathered all the refs.
6027 static int calc_extent_flag(struct btrfs_root *root,
6028 struct cache_tree *extent_cache,
6029 struct extent_buffer *buf,
6030 struct root_item_record *ri,
6033 struct extent_record *rec;
6034 struct cache_extent *cache;
6035 struct tree_backref *tback;
6038 cache = lookup_cache_extent(extent_cache, buf->start, 1);
6039 /* we have added this extent before */
6043 rec = container_of(cache, struct extent_record, cache);
6046 * Except file/reloc tree, we can not have
6049 if (ri->objectid < BTRFS_FIRST_FREE_OBJECTID)
6054 if (buf->start == ri->bytenr)
6057 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
6060 owner = btrfs_header_owner(buf);
6061 if (owner == ri->objectid)
6064 tback = find_tree_backref(rec, 0, owner);
6069 if (rec->flag_block_full_backref != FLAG_UNSET &&
6070 rec->flag_block_full_backref != 0)
6071 rec->bad_full_backref = 1;
6074 *flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6075 if (rec->flag_block_full_backref != FLAG_UNSET &&
6076 rec->flag_block_full_backref != 1)
6077 rec->bad_full_backref = 1;
6081 static void report_mismatch_key_root(u8 key_type, u64 rootid)
6083 fprintf(stderr, "Invalid key type(");
6084 print_key_type(stderr, 0, key_type);
6085 fprintf(stderr, ") found in root(");
6086 print_objectid(stderr, rootid, 0);
6087 fprintf(stderr, ")\n");
6091 * Check if the key is valid with its extent buffer.
6093 * This is a early check in case invalid key exists in a extent buffer
6094 * This is not comprehensive yet, but should prevent wrong key/item passed
6097 static int check_type_with_root(u64 rootid, u8 key_type)
6100 /* Only valid in chunk tree */
6101 case BTRFS_DEV_ITEM_KEY:
6102 case BTRFS_CHUNK_ITEM_KEY:
6103 if (rootid != BTRFS_CHUNK_TREE_OBJECTID)
6106 /* valid in csum and log tree */
6107 case BTRFS_CSUM_TREE_OBJECTID:
6108 if (!(rootid == BTRFS_TREE_LOG_OBJECTID ||
6112 case BTRFS_EXTENT_ITEM_KEY:
6113 case BTRFS_METADATA_ITEM_KEY:
6114 case BTRFS_BLOCK_GROUP_ITEM_KEY:
6115 if (rootid != BTRFS_EXTENT_TREE_OBJECTID)
6118 case BTRFS_ROOT_ITEM_KEY:
6119 if (rootid != BTRFS_ROOT_TREE_OBJECTID)
6122 case BTRFS_DEV_EXTENT_KEY:
6123 if (rootid != BTRFS_DEV_TREE_OBJECTID)
6129 report_mismatch_key_root(key_type, rootid);
6133 static int run_next_block(struct btrfs_root *root,
6134 struct block_info *bits,
6137 struct cache_tree *pending,
6138 struct cache_tree *seen,
6139 struct cache_tree *reada,
6140 struct cache_tree *nodes,
6141 struct cache_tree *extent_cache,
6142 struct cache_tree *chunk_cache,
6143 struct rb_root *dev_cache,
6144 struct block_group_tree *block_group_cache,
6145 struct device_extent_tree *dev_extent_cache,
6146 struct root_item_record *ri)
6148 struct extent_buffer *buf;
6149 struct extent_record *rec = NULL;
6160 struct btrfs_key key;
6161 struct cache_extent *cache;
6164 nritems = pick_next_pending(pending, reada, nodes, *last, bits,
6165 bits_nr, &reada_bits);
6170 for(i = 0; i < nritems; i++) {
6171 ret = add_cache_extent(reada, bits[i].start,
6176 /* fixme, get the parent transid */
6177 readahead_tree_block(root, bits[i].start,
6181 *last = bits[0].start;
6182 bytenr = bits[0].start;
6183 size = bits[0].size;
6185 cache = lookup_cache_extent(pending, bytenr, size);
6187 remove_cache_extent(pending, cache);
6190 cache = lookup_cache_extent(reada, bytenr, size);
6192 remove_cache_extent(reada, cache);
6195 cache = lookup_cache_extent(nodes, bytenr, size);
6197 remove_cache_extent(nodes, cache);
6200 cache = lookup_cache_extent(extent_cache, bytenr, size);
6202 rec = container_of(cache, struct extent_record, cache);
6203 gen = rec->parent_generation;
6206 /* fixme, get the real parent transid */
6207 buf = read_tree_block(root, bytenr, size, gen);
6208 if (!extent_buffer_uptodate(buf)) {
6209 record_bad_block_io(root->fs_info,
6210 extent_cache, bytenr, size);
6214 nritems = btrfs_header_nritems(buf);
6217 if (!init_extent_tree) {
6218 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
6219 btrfs_header_level(buf), 1, NULL,
6222 ret = calc_extent_flag(root, extent_cache, buf, ri, &flags);
6224 fprintf(stderr, "Couldn't calc extent flags\n");
6225 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6230 ret = calc_extent_flag(root, extent_cache, buf, ri, &flags);
6232 fprintf(stderr, "Couldn't calc extent flags\n");
6233 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6237 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6239 ri->objectid != BTRFS_TREE_RELOC_OBJECTID &&
6240 ri->objectid == btrfs_header_owner(buf)) {
6242 * Ok we got to this block from it's original owner and
6243 * we have FULL_BACKREF set. Relocation can leave
6244 * converted blocks over so this is altogether possible,
6245 * however it's not possible if the generation > the
6246 * last snapshot, so check for this case.
6248 if (!btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC) &&
6249 btrfs_header_generation(buf) > ri->last_snapshot) {
6250 flags &= ~BTRFS_BLOCK_FLAG_FULL_BACKREF;
6251 rec->bad_full_backref = 1;
6256 (ri->objectid == BTRFS_TREE_RELOC_OBJECTID ||
6257 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) {
6258 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
6259 rec->bad_full_backref = 1;
6263 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
6264 rec->flag_block_full_backref = 1;
6268 rec->flag_block_full_backref = 0;
6270 owner = btrfs_header_owner(buf);
6273 ret = check_block(root, extent_cache, buf, flags);
6277 if (btrfs_is_leaf(buf)) {
6278 btree_space_waste += btrfs_leaf_free_space(root, buf);
6279 for (i = 0; i < nritems; i++) {
6280 struct btrfs_file_extent_item *fi;
6281 btrfs_item_key_to_cpu(buf, &key, i);
6283 * Check key type against the leaf owner.
6284 * Could filter quite a lot of early error if
6287 if (check_type_with_root(btrfs_header_owner(buf),
6289 fprintf(stderr, "ignoring invalid key\n");
6292 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
6293 process_extent_item(root, extent_cache, buf,
6297 if (key.type == BTRFS_METADATA_ITEM_KEY) {
6298 process_extent_item(root, extent_cache, buf,
6302 if (key.type == BTRFS_EXTENT_CSUM_KEY) {
6304 btrfs_item_size_nr(buf, i);
6307 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6308 process_chunk_item(chunk_cache, &key, buf, i);
6311 if (key.type == BTRFS_DEV_ITEM_KEY) {
6312 process_device_item(dev_cache, &key, buf, i);
6315 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
6316 process_block_group_item(block_group_cache,
6320 if (key.type == BTRFS_DEV_EXTENT_KEY) {
6321 process_device_extent_item(dev_extent_cache,
6326 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
6327 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6328 process_extent_ref_v0(extent_cache, buf, i);
6335 if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
6336 ret = add_tree_backref(extent_cache,
6337 key.objectid, 0, key.offset, 0);
6339 error("add_tree_backref failed: %s",
6343 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
6344 ret = add_tree_backref(extent_cache,
6345 key.objectid, key.offset, 0, 0);
6347 error("add_tree_backref failed: %s",
6351 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
6352 struct btrfs_extent_data_ref *ref;
6353 ref = btrfs_item_ptr(buf, i,
6354 struct btrfs_extent_data_ref);
6355 add_data_backref(extent_cache,
6357 btrfs_extent_data_ref_root(buf, ref),
6358 btrfs_extent_data_ref_objectid(buf,
6360 btrfs_extent_data_ref_offset(buf, ref),
6361 btrfs_extent_data_ref_count(buf, ref),
6362 0, root->sectorsize);
6365 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
6366 struct btrfs_shared_data_ref *ref;
6367 ref = btrfs_item_ptr(buf, i,
6368 struct btrfs_shared_data_ref);
6369 add_data_backref(extent_cache,
6370 key.objectid, key.offset, 0, 0, 0,
6371 btrfs_shared_data_ref_count(buf, ref),
6372 0, root->sectorsize);
6375 if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
6376 struct bad_item *bad;
6378 if (key.objectid == BTRFS_ORPHAN_OBJECTID)
6382 bad = malloc(sizeof(struct bad_item));
6385 INIT_LIST_HEAD(&bad->list);
6386 memcpy(&bad->key, &key,
6387 sizeof(struct btrfs_key));
6388 bad->root_id = owner;
6389 list_add_tail(&bad->list, &delete_items);
6392 if (key.type != BTRFS_EXTENT_DATA_KEY)
6394 fi = btrfs_item_ptr(buf, i,
6395 struct btrfs_file_extent_item);
6396 if (btrfs_file_extent_type(buf, fi) ==
6397 BTRFS_FILE_EXTENT_INLINE)
6399 if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
6402 data_bytes_allocated +=
6403 btrfs_file_extent_disk_num_bytes(buf, fi);
6404 if (data_bytes_allocated < root->sectorsize) {
6407 data_bytes_referenced +=
6408 btrfs_file_extent_num_bytes(buf, fi);
6409 add_data_backref(extent_cache,
6410 btrfs_file_extent_disk_bytenr(buf, fi),
6411 parent, owner, key.objectid, key.offset -
6412 btrfs_file_extent_offset(buf, fi), 1, 1,
6413 btrfs_file_extent_disk_num_bytes(buf, fi));
6417 struct btrfs_key first_key;
6419 first_key.objectid = 0;
6422 btrfs_item_key_to_cpu(buf, &first_key, 0);
6423 level = btrfs_header_level(buf);
6424 for (i = 0; i < nritems; i++) {
6425 struct extent_record tmpl;
6427 ptr = btrfs_node_blockptr(buf, i);
6428 size = root->nodesize;
6429 btrfs_node_key_to_cpu(buf, &key, i);
6431 if ((level == ri->drop_level)
6432 && is_dropped_key(&key, &ri->drop_key)) {
6437 memset(&tmpl, 0, sizeof(tmpl));
6438 btrfs_cpu_key_to_disk(&tmpl.parent_key, &key);
6439 tmpl.parent_generation = btrfs_node_ptr_generation(buf, i);
6444 tmpl.max_size = size;
6445 ret = add_extent_rec(extent_cache, &tmpl);
6449 ret = add_tree_backref(extent_cache, ptr, parent,
6452 error("add_tree_backref failed: %s",
6458 add_pending(nodes, seen, ptr, size);
6460 add_pending(pending, seen, ptr, size);
6463 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
6464 nritems) * sizeof(struct btrfs_key_ptr);
6466 total_btree_bytes += buf->len;
6467 if (fs_root_objectid(btrfs_header_owner(buf)))
6468 total_fs_tree_bytes += buf->len;
6469 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
6470 total_extent_tree_bytes += buf->len;
6471 if (!found_old_backref &&
6472 btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
6473 btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
6474 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
6475 found_old_backref = 1;
6477 free_extent_buffer(buf);
6481 static int add_root_to_pending(struct extent_buffer *buf,
6482 struct cache_tree *extent_cache,
6483 struct cache_tree *pending,
6484 struct cache_tree *seen,
6485 struct cache_tree *nodes,
6488 struct extent_record tmpl;
6491 if (btrfs_header_level(buf) > 0)
6492 add_pending(nodes, seen, buf->start, buf->len);
6494 add_pending(pending, seen, buf->start, buf->len);
6496 memset(&tmpl, 0, sizeof(tmpl));
6497 tmpl.start = buf->start;
6502 tmpl.max_size = buf->len;
6503 add_extent_rec(extent_cache, &tmpl);
6505 if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
6506 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
6507 ret = add_tree_backref(extent_cache, buf->start, buf->start,
6510 ret = add_tree_backref(extent_cache, buf->start, 0, objectid,
6515 /* as we fix the tree, we might be deleting blocks that
6516 * we're tracking for repair. This hook makes sure we
6517 * remove any backrefs for blocks as we are fixing them.
6519 static int free_extent_hook(struct btrfs_trans_handle *trans,
6520 struct btrfs_root *root,
6521 u64 bytenr, u64 num_bytes, u64 parent,
6522 u64 root_objectid, u64 owner, u64 offset,
6525 struct extent_record *rec;
6526 struct cache_extent *cache;
6528 struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
6530 is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
6531 cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
6535 rec = container_of(cache, struct extent_record, cache);
6537 struct data_backref *back;
6538 back = find_data_backref(rec, parent, root_objectid, owner,
6539 offset, 1, bytenr, num_bytes);
6542 if (back->node.found_ref) {
6543 back->found_ref -= refs_to_drop;
6545 rec->refs -= refs_to_drop;
6547 if (back->node.found_extent_tree) {
6548 back->num_refs -= refs_to_drop;
6549 if (rec->extent_item_refs)
6550 rec->extent_item_refs -= refs_to_drop;
6552 if (back->found_ref == 0)
6553 back->node.found_ref = 0;
6554 if (back->num_refs == 0)
6555 back->node.found_extent_tree = 0;
6557 if (!back->node.found_extent_tree && back->node.found_ref) {
6558 list_del(&back->node.list);
6562 struct tree_backref *back;
6563 back = find_tree_backref(rec, parent, root_objectid);
6566 if (back->node.found_ref) {
6569 back->node.found_ref = 0;
6571 if (back->node.found_extent_tree) {
6572 if (rec->extent_item_refs)
6573 rec->extent_item_refs--;
6574 back->node.found_extent_tree = 0;
6576 if (!back->node.found_extent_tree && back->node.found_ref) {
6577 list_del(&back->node.list);
6581 maybe_free_extent_rec(extent_cache, rec);
6586 static int delete_extent_records(struct btrfs_trans_handle *trans,
6587 struct btrfs_root *root,
6588 struct btrfs_path *path,
6589 u64 bytenr, u64 new_len)
6591 struct btrfs_key key;
6592 struct btrfs_key found_key;
6593 struct extent_buffer *leaf;
6598 key.objectid = bytenr;
6600 key.offset = (u64)-1;
6603 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
6610 if (path->slots[0] == 0)
6616 leaf = path->nodes[0];
6617 slot = path->slots[0];
6619 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6620 if (found_key.objectid != bytenr)
6623 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
6624 found_key.type != BTRFS_METADATA_ITEM_KEY &&
6625 found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
6626 found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
6627 found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
6628 found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
6629 found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
6630 btrfs_release_path(path);
6631 if (found_key.type == 0) {
6632 if (found_key.offset == 0)
6634 key.offset = found_key.offset - 1;
6635 key.type = found_key.type;
6637 key.type = found_key.type - 1;
6638 key.offset = (u64)-1;
6642 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
6643 found_key.objectid, found_key.type, found_key.offset);
6645 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
6648 btrfs_release_path(path);
6650 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
6651 found_key.type == BTRFS_METADATA_ITEM_KEY) {
6652 u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
6653 found_key.offset : root->nodesize;
6655 ret = btrfs_update_block_group(trans, root, bytenr,
6662 btrfs_release_path(path);
6667 * for a single backref, this will allocate a new extent
6668 * and add the backref to it.
6670 static int record_extent(struct btrfs_trans_handle *trans,
6671 struct btrfs_fs_info *info,
6672 struct btrfs_path *path,
6673 struct extent_record *rec,
6674 struct extent_backref *back,
6675 int allocated, u64 flags)
6678 struct btrfs_root *extent_root = info->extent_root;
6679 struct extent_buffer *leaf;
6680 struct btrfs_key ins_key;
6681 struct btrfs_extent_item *ei;
6682 struct tree_backref *tback;
6683 struct data_backref *dback;
6684 struct btrfs_tree_block_info *bi;
6687 rec->max_size = max_t(u64, rec->max_size,
6688 info->extent_root->nodesize);
6691 u32 item_size = sizeof(*ei);
6694 item_size += sizeof(*bi);
6696 ins_key.objectid = rec->start;
6697 ins_key.offset = rec->max_size;
6698 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
6700 ret = btrfs_insert_empty_item(trans, extent_root, path,
6701 &ins_key, item_size);
6705 leaf = path->nodes[0];
6706 ei = btrfs_item_ptr(leaf, path->slots[0],
6707 struct btrfs_extent_item);
6709 btrfs_set_extent_refs(leaf, ei, 0);
6710 btrfs_set_extent_generation(leaf, ei, rec->generation);
6712 if (back->is_data) {
6713 btrfs_set_extent_flags(leaf, ei,
6714 BTRFS_EXTENT_FLAG_DATA);
6716 struct btrfs_disk_key copy_key;;
6718 tback = to_tree_backref(back);
6719 bi = (struct btrfs_tree_block_info *)(ei + 1);
6720 memset_extent_buffer(leaf, 0, (unsigned long)bi,
6723 btrfs_set_disk_key_objectid(©_key,
6724 rec->info_objectid);
6725 btrfs_set_disk_key_type(©_key, 0);
6726 btrfs_set_disk_key_offset(©_key, 0);
6728 btrfs_set_tree_block_level(leaf, bi, rec->info_level);
6729 btrfs_set_tree_block_key(leaf, bi, ©_key);
6731 btrfs_set_extent_flags(leaf, ei,
6732 BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
6735 btrfs_mark_buffer_dirty(leaf);
6736 ret = btrfs_update_block_group(trans, extent_root, rec->start,
6737 rec->max_size, 1, 0);
6740 btrfs_release_path(path);
6743 if (back->is_data) {
6747 dback = to_data_backref(back);
6748 if (back->full_backref)
6749 parent = dback->parent;
6753 for (i = 0; i < dback->found_ref; i++) {
6754 /* if parent != 0, we're doing a full backref
6755 * passing BTRFS_FIRST_FREE_OBJECTID as the owner
6756 * just makes the backref allocator create a data
6759 ret = btrfs_inc_extent_ref(trans, info->extent_root,
6760 rec->start, rec->max_size,
6764 BTRFS_FIRST_FREE_OBJECTID :
6770 fprintf(stderr, "adding new data backref"
6771 " on %llu %s %llu owner %llu"
6772 " offset %llu found %d\n",
6773 (unsigned long long)rec->start,
6774 back->full_backref ?
6776 back->full_backref ?
6777 (unsigned long long)parent :
6778 (unsigned long long)dback->root,
6779 (unsigned long long)dback->owner,
6780 (unsigned long long)dback->offset,
6785 tback = to_tree_backref(back);
6786 if (back->full_backref)
6787 parent = tback->parent;
6791 ret = btrfs_inc_extent_ref(trans, info->extent_root,
6792 rec->start, rec->max_size,
6793 parent, tback->root, 0, 0);
6794 fprintf(stderr, "adding new tree backref on "
6795 "start %llu len %llu parent %llu root %llu\n",
6796 rec->start, rec->max_size, parent, tback->root);
6799 btrfs_release_path(path);
6803 static struct extent_entry *find_entry(struct list_head *entries,
6804 u64 bytenr, u64 bytes)
6806 struct extent_entry *entry = NULL;
6808 list_for_each_entry(entry, entries, list) {
6809 if (entry->bytenr == bytenr && entry->bytes == bytes)
6816 static struct extent_entry *find_most_right_entry(struct list_head *entries)
6818 struct extent_entry *entry, *best = NULL, *prev = NULL;
6820 list_for_each_entry(entry, entries, list) {
6827 * If there are as many broken entries as entries then we know
6828 * not to trust this particular entry.
6830 if (entry->broken == entry->count)
6834 * If our current entry == best then we can't be sure our best
6835 * is really the best, so we need to keep searching.
6837 if (best && best->count == entry->count) {
6843 /* Prev == entry, not good enough, have to keep searching */
6844 if (!prev->broken && prev->count == entry->count)
6848 best = (prev->count > entry->count) ? prev : entry;
6849 else if (best->count < entry->count)
6857 static int repair_ref(struct btrfs_fs_info *info, struct btrfs_path *path,
6858 struct data_backref *dback, struct extent_entry *entry)
6860 struct btrfs_trans_handle *trans;
6861 struct btrfs_root *root;
6862 struct btrfs_file_extent_item *fi;
6863 struct extent_buffer *leaf;
6864 struct btrfs_key key;
6868 key.objectid = dback->root;
6869 key.type = BTRFS_ROOT_ITEM_KEY;
6870 key.offset = (u64)-1;
6871 root = btrfs_read_fs_root(info, &key);
6873 fprintf(stderr, "Couldn't find root for our ref\n");
6878 * The backref points to the original offset of the extent if it was
6879 * split, so we need to search down to the offset we have and then walk
6880 * forward until we find the backref we're looking for.
6882 key.objectid = dback->owner;
6883 key.type = BTRFS_EXTENT_DATA_KEY;
6884 key.offset = dback->offset;
6885 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6887 fprintf(stderr, "Error looking up ref %d\n", ret);
6892 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
6893 ret = btrfs_next_leaf(root, path);
6895 fprintf(stderr, "Couldn't find our ref, next\n");
6899 leaf = path->nodes[0];
6900 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6901 if (key.objectid != dback->owner ||
6902 key.type != BTRFS_EXTENT_DATA_KEY) {
6903 fprintf(stderr, "Couldn't find our ref, search\n");
6906 fi = btrfs_item_ptr(leaf, path->slots[0],
6907 struct btrfs_file_extent_item);
6908 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6909 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
6911 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
6916 btrfs_release_path(path);
6918 trans = btrfs_start_transaction(root, 1);
6920 return PTR_ERR(trans);
6923 * Ok we have the key of the file extent we want to fix, now we can cow
6924 * down to the thing and fix it.
6926 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
6928 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
6929 key.objectid, key.type, key.offset, ret);
6933 fprintf(stderr, "Well that's odd, we just found this key "
6934 "[%Lu, %u, %Lu]\n", key.objectid, key.type,
6939 leaf = path->nodes[0];
6940 fi = btrfs_item_ptr(leaf, path->slots[0],
6941 struct btrfs_file_extent_item);
6943 if (btrfs_file_extent_compression(leaf, fi) &&
6944 dback->disk_bytenr != entry->bytenr) {
6945 fprintf(stderr, "Ref doesn't match the record start and is "
6946 "compressed, please take a btrfs-image of this file "
6947 "system and send it to a btrfs developer so they can "
6948 "complete this functionality for bytenr %Lu\n",
6949 dback->disk_bytenr);
6954 if (dback->node.broken && dback->disk_bytenr != entry->bytenr) {
6955 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6956 } else if (dback->disk_bytenr > entry->bytenr) {
6957 u64 off_diff, offset;
6959 off_diff = dback->disk_bytenr - entry->bytenr;
6960 offset = btrfs_file_extent_offset(leaf, fi);
6961 if (dback->disk_bytenr + offset +
6962 btrfs_file_extent_num_bytes(leaf, fi) >
6963 entry->bytenr + entry->bytes) {
6964 fprintf(stderr, "Ref is past the entry end, please "
6965 "take a btrfs-image of this file system and "
6966 "send it to a btrfs developer, ref %Lu\n",
6967 dback->disk_bytenr);
6972 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6973 btrfs_set_file_extent_offset(leaf, fi, offset);
6974 } else if (dback->disk_bytenr < entry->bytenr) {
6977 offset = btrfs_file_extent_offset(leaf, fi);
6978 if (dback->disk_bytenr + offset < entry->bytenr) {
6979 fprintf(stderr, "Ref is before the entry start, please"
6980 " take a btrfs-image of this file system and "
6981 "send it to a btrfs developer, ref %Lu\n",
6982 dback->disk_bytenr);
6987 offset += dback->disk_bytenr;
6988 offset -= entry->bytenr;
6989 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
6990 btrfs_set_file_extent_offset(leaf, fi, offset);
6993 btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
6996 * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
6997 * only do this if we aren't using compression, otherwise it's a
7000 if (!btrfs_file_extent_compression(leaf, fi))
7001 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
7003 printf("ram bytes may be wrong?\n");
7004 btrfs_mark_buffer_dirty(leaf);
7006 err = btrfs_commit_transaction(trans, root);
7007 btrfs_release_path(path);
7008 return ret ? ret : err;
7011 static int verify_backrefs(struct btrfs_fs_info *info, struct btrfs_path *path,
7012 struct extent_record *rec)
7014 struct extent_backref *back;
7015 struct data_backref *dback;
7016 struct extent_entry *entry, *best = NULL;
7019 int broken_entries = 0;
7024 * Metadata is easy and the backrefs should always agree on bytenr and
7025 * size, if not we've got bigger issues.
7030 list_for_each_entry(back, &rec->backrefs, list) {
7031 if (back->full_backref || !back->is_data)
7034 dback = to_data_backref(back);
7037 * We only pay attention to backrefs that we found a real
7040 if (dback->found_ref == 0)
7044 * For now we only catch when the bytes don't match, not the
7045 * bytenr. We can easily do this at the same time, but I want
7046 * to have a fs image to test on before we just add repair
7047 * functionality willy-nilly so we know we won't screw up the
7051 entry = find_entry(&entries, dback->disk_bytenr,
7054 entry = malloc(sizeof(struct extent_entry));
7059 memset(entry, 0, sizeof(*entry));
7060 entry->bytenr = dback->disk_bytenr;
7061 entry->bytes = dback->bytes;
7062 list_add_tail(&entry->list, &entries);
7067 * If we only have on entry we may think the entries agree when
7068 * in reality they don't so we have to do some extra checking.
7070 if (dback->disk_bytenr != rec->start ||
7071 dback->bytes != rec->nr || back->broken)
7082 /* Yay all the backrefs agree, carry on good sir */
7083 if (nr_entries <= 1 && !mismatch)
7086 fprintf(stderr, "attempting to repair backref discrepency for bytenr "
7087 "%Lu\n", rec->start);
7090 * First we want to see if the backrefs can agree amongst themselves who
7091 * is right, so figure out which one of the entries has the highest
7094 best = find_most_right_entry(&entries);
7097 * Ok so we may have an even split between what the backrefs think, so
7098 * this is where we use the extent ref to see what it thinks.
7101 entry = find_entry(&entries, rec->start, rec->nr);
7102 if (!entry && (!broken_entries || !rec->found_rec)) {
7103 fprintf(stderr, "Backrefs don't agree with each other "
7104 "and extent record doesn't agree with anybody,"
7105 " so we can't fix bytenr %Lu bytes %Lu\n",
7106 rec->start, rec->nr);
7109 } else if (!entry) {
7111 * Ok our backrefs were broken, we'll assume this is the
7112 * correct value and add an entry for this range.
7114 entry = malloc(sizeof(struct extent_entry));
7119 memset(entry, 0, sizeof(*entry));
7120 entry->bytenr = rec->start;
7121 entry->bytes = rec->nr;
7122 list_add_tail(&entry->list, &entries);
7126 best = find_most_right_entry(&entries);
7128 fprintf(stderr, "Backrefs and extent record evenly "
7129 "split on who is right, this is going to "
7130 "require user input to fix bytenr %Lu bytes "
7131 "%Lu\n", rec->start, rec->nr);
7138 * I don't think this can happen currently as we'll abort() if we catch
7139 * this case higher up, but in case somebody removes that we still can't
7140 * deal with it properly here yet, so just bail out of that's the case.
7142 if (best->bytenr != rec->start) {
7143 fprintf(stderr, "Extent start and backref starts don't match, "
7144 "please use btrfs-image on this file system and send "
7145 "it to a btrfs developer so they can make fsck fix "
7146 "this particular case. bytenr is %Lu, bytes is %Lu\n",
7147 rec->start, rec->nr);
7153 * Ok great we all agreed on an extent record, let's go find the real
7154 * references and fix up the ones that don't match.
7156 list_for_each_entry(back, &rec->backrefs, list) {
7157 if (back->full_backref || !back->is_data)
7160 dback = to_data_backref(back);
7163 * Still ignoring backrefs that don't have a real ref attached
7166 if (dback->found_ref == 0)
7169 if (dback->bytes == best->bytes &&
7170 dback->disk_bytenr == best->bytenr)
7173 ret = repair_ref(info, path, dback, best);
7179 * Ok we messed with the actual refs, which means we need to drop our
7180 * entire cache and go back and rescan. I know this is a huge pain and
7181 * adds a lot of extra work, but it's the only way to be safe. Once all
7182 * the backrefs agree we may not need to do anything to the extent
7187 while (!list_empty(&entries)) {
7188 entry = list_entry(entries.next, struct extent_entry, list);
7189 list_del_init(&entry->list);
7195 static int process_duplicates(struct btrfs_root *root,
7196 struct cache_tree *extent_cache,
7197 struct extent_record *rec)
7199 struct extent_record *good, *tmp;
7200 struct cache_extent *cache;
7204 * If we found a extent record for this extent then return, or if we
7205 * have more than one duplicate we are likely going to need to delete
7208 if (rec->found_rec || rec->num_duplicates > 1)
7211 /* Shouldn't happen but just in case */
7212 BUG_ON(!rec->num_duplicates);
7215 * So this happens if we end up with a backref that doesn't match the
7216 * actual extent entry. So either the backref is bad or the extent
7217 * entry is bad. Either way we want to have the extent_record actually
7218 * reflect what we found in the extent_tree, so we need to take the
7219 * duplicate out and use that as the extent_record since the only way we
7220 * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
7222 remove_cache_extent(extent_cache, &rec->cache);
7224 good = to_extent_record(rec->dups.next);
7225 list_del_init(&good->list);
7226 INIT_LIST_HEAD(&good->backrefs);
7227 INIT_LIST_HEAD(&good->dups);
7228 good->cache.start = good->start;
7229 good->cache.size = good->nr;
7230 good->content_checked = 0;
7231 good->owner_ref_checked = 0;
7232 good->num_duplicates = 0;
7233 good->refs = rec->refs;
7234 list_splice_init(&rec->backrefs, &good->backrefs);
7236 cache = lookup_cache_extent(extent_cache, good->start,
7240 tmp = container_of(cache, struct extent_record, cache);
7243 * If we find another overlapping extent and it's found_rec is
7244 * set then it's a duplicate and we need to try and delete
7247 if (tmp->found_rec || tmp->num_duplicates > 0) {
7248 if (list_empty(&good->list))
7249 list_add_tail(&good->list,
7250 &duplicate_extents);
7251 good->num_duplicates += tmp->num_duplicates + 1;
7252 list_splice_init(&tmp->dups, &good->dups);
7253 list_del_init(&tmp->list);
7254 list_add_tail(&tmp->list, &good->dups);
7255 remove_cache_extent(extent_cache, &tmp->cache);
7260 * Ok we have another non extent item backed extent rec, so lets
7261 * just add it to this extent and carry on like we did above.
7263 good->refs += tmp->refs;
7264 list_splice_init(&tmp->backrefs, &good->backrefs);
7265 remove_cache_extent(extent_cache, &tmp->cache);
7268 ret = insert_cache_extent(extent_cache, &good->cache);
7271 return good->num_duplicates ? 0 : 1;
7274 static int delete_duplicate_records(struct btrfs_root *root,
7275 struct extent_record *rec)
7277 struct btrfs_trans_handle *trans;
7278 LIST_HEAD(delete_list);
7279 struct btrfs_path *path;
7280 struct extent_record *tmp, *good, *n;
7283 struct btrfs_key key;
7285 path = btrfs_alloc_path();
7292 /* Find the record that covers all of the duplicates. */
7293 list_for_each_entry(tmp, &rec->dups, list) {
7294 if (good->start < tmp->start)
7296 if (good->nr > tmp->nr)
7299 if (tmp->start + tmp->nr < good->start + good->nr) {
7300 fprintf(stderr, "Ok we have overlapping extents that "
7301 "aren't completely covered by each other, this "
7302 "is going to require more careful thought. "
7303 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
7304 tmp->start, tmp->nr, good->start, good->nr);
7311 list_add_tail(&rec->list, &delete_list);
7313 list_for_each_entry_safe(tmp, n, &rec->dups, list) {
7316 list_move_tail(&tmp->list, &delete_list);
7319 root = root->fs_info->extent_root;
7320 trans = btrfs_start_transaction(root, 1);
7321 if (IS_ERR(trans)) {
7322 ret = PTR_ERR(trans);
7326 list_for_each_entry(tmp, &delete_list, list) {
7327 if (tmp->found_rec == 0)
7329 key.objectid = tmp->start;
7330 key.type = BTRFS_EXTENT_ITEM_KEY;
7331 key.offset = tmp->nr;
7333 /* Shouldn't happen but just in case */
7334 if (tmp->metadata) {
7335 fprintf(stderr, "Well this shouldn't happen, extent "
7336 "record overlaps but is metadata? "
7337 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
7341 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
7347 ret = btrfs_del_item(trans, root, path);
7350 btrfs_release_path(path);
7353 err = btrfs_commit_transaction(trans, root);
7357 while (!list_empty(&delete_list)) {
7358 tmp = to_extent_record(delete_list.next);
7359 list_del_init(&tmp->list);
7365 while (!list_empty(&rec->dups)) {
7366 tmp = to_extent_record(rec->dups.next);
7367 list_del_init(&tmp->list);
7371 btrfs_free_path(path);
7373 if (!ret && !nr_del)
7374 rec->num_duplicates = 0;
7376 return ret ? ret : nr_del;
7379 static int find_possible_backrefs(struct btrfs_fs_info *info,
7380 struct btrfs_path *path,
7381 struct cache_tree *extent_cache,
7382 struct extent_record *rec)
7384 struct btrfs_root *root;
7385 struct extent_backref *back;
7386 struct data_backref *dback;
7387 struct cache_extent *cache;
7388 struct btrfs_file_extent_item *fi;
7389 struct btrfs_key key;
7393 list_for_each_entry(back, &rec->backrefs, list) {
7394 /* Don't care about full backrefs (poor unloved backrefs) */
7395 if (back->full_backref || !back->is_data)
7398 dback = to_data_backref(back);
7400 /* We found this one, we don't need to do a lookup */
7401 if (dback->found_ref)
7404 key.objectid = dback->root;
7405 key.type = BTRFS_ROOT_ITEM_KEY;
7406 key.offset = (u64)-1;
7408 root = btrfs_read_fs_root(info, &key);
7410 /* No root, definitely a bad ref, skip */
7411 if (IS_ERR(root) && PTR_ERR(root) == -ENOENT)
7413 /* Other err, exit */
7415 return PTR_ERR(root);
7417 key.objectid = dback->owner;
7418 key.type = BTRFS_EXTENT_DATA_KEY;
7419 key.offset = dback->offset;
7420 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7422 btrfs_release_path(path);
7425 /* Didn't find it, we can carry on */
7430 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
7431 struct btrfs_file_extent_item);
7432 bytenr = btrfs_file_extent_disk_bytenr(path->nodes[0], fi);
7433 bytes = btrfs_file_extent_disk_num_bytes(path->nodes[0], fi);
7434 btrfs_release_path(path);
7435 cache = lookup_cache_extent(extent_cache, bytenr, 1);
7437 struct extent_record *tmp;
7438 tmp = container_of(cache, struct extent_record, cache);
7441 * If we found an extent record for the bytenr for this
7442 * particular backref then we can't add it to our
7443 * current extent record. We only want to add backrefs
7444 * that don't have a corresponding extent item in the
7445 * extent tree since they likely belong to this record
7446 * and we need to fix it if it doesn't match bytenrs.
7452 dback->found_ref += 1;
7453 dback->disk_bytenr = bytenr;
7454 dback->bytes = bytes;
7457 * Set this so the verify backref code knows not to trust the
7458 * values in this backref.
7467 * Record orphan data ref into corresponding root.
7469 * Return 0 if the extent item contains data ref and recorded.
7470 * Return 1 if the extent item contains no useful data ref
7471 * On that case, it may contains only shared_dataref or metadata backref
7472 * or the file extent exists(this should be handled by the extent bytenr
7474 * Return <0 if something goes wrong.
7476 static int record_orphan_data_extents(struct btrfs_fs_info *fs_info,
7477 struct extent_record *rec)
7479 struct btrfs_key key;
7480 struct btrfs_root *dest_root;
7481 struct extent_backref *back;
7482 struct data_backref *dback;
7483 struct orphan_data_extent *orphan;
7484 struct btrfs_path *path;
7485 int recorded_data_ref = 0;
7490 path = btrfs_alloc_path();
7493 list_for_each_entry(back, &rec->backrefs, list) {
7494 if (back->full_backref || !back->is_data ||
7495 !back->found_extent_tree)
7497 dback = to_data_backref(back);
7498 if (dback->found_ref)
7500 key.objectid = dback->root;
7501 key.type = BTRFS_ROOT_ITEM_KEY;
7502 key.offset = (u64)-1;
7504 dest_root = btrfs_read_fs_root(fs_info, &key);
7506 /* For non-exist root we just skip it */
7507 if (IS_ERR(dest_root) || !dest_root)
7510 key.objectid = dback->owner;
7511 key.type = BTRFS_EXTENT_DATA_KEY;
7512 key.offset = dback->offset;
7514 ret = btrfs_search_slot(NULL, dest_root, &key, path, 0, 0);
7515 btrfs_release_path(path);
7517 * For ret < 0, it's OK since the fs-tree may be corrupted,
7518 * we need to record it for inode/file extent rebuild.
7519 * For ret > 0, we record it only for file extent rebuild.
7520 * For ret == 0, the file extent exists but only bytenr
7521 * mismatch, let the original bytenr fix routine to handle,
7527 orphan = malloc(sizeof(*orphan));
7532 INIT_LIST_HEAD(&orphan->list);
7533 orphan->root = dback->root;
7534 orphan->objectid = dback->owner;
7535 orphan->offset = dback->offset;
7536 orphan->disk_bytenr = rec->cache.start;
7537 orphan->disk_len = rec->cache.size;
7538 list_add(&dest_root->orphan_data_extents, &orphan->list);
7539 recorded_data_ref = 1;
7542 btrfs_free_path(path);
7544 return !recorded_data_ref;
7550 * when an incorrect extent item is found, this will delete
7551 * all of the existing entries for it and recreate them
7552 * based on what the tree scan found.
7554 static int fixup_extent_refs(struct btrfs_fs_info *info,
7555 struct cache_tree *extent_cache,
7556 struct extent_record *rec)
7558 struct btrfs_trans_handle *trans = NULL;
7560 struct btrfs_path *path;
7561 struct list_head *cur = rec->backrefs.next;
7562 struct cache_extent *cache;
7563 struct extent_backref *back;
7567 if (rec->flag_block_full_backref)
7568 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7570 path = btrfs_alloc_path();
7574 if (rec->refs != rec->extent_item_refs && !rec->metadata) {
7576 * Sometimes the backrefs themselves are so broken they don't
7577 * get attached to any meaningful rec, so first go back and
7578 * check any of our backrefs that we couldn't find and throw
7579 * them into the list if we find the backref so that
7580 * verify_backrefs can figure out what to do.
7582 ret = find_possible_backrefs(info, path, extent_cache, rec);
7587 /* step one, make sure all of the backrefs agree */
7588 ret = verify_backrefs(info, path, rec);
7592 trans = btrfs_start_transaction(info->extent_root, 1);
7593 if (IS_ERR(trans)) {
7594 ret = PTR_ERR(trans);
7598 /* step two, delete all the existing records */
7599 ret = delete_extent_records(trans, info->extent_root, path,
7600 rec->start, rec->max_size);
7605 /* was this block corrupt? If so, don't add references to it */
7606 cache = lookup_cache_extent(info->corrupt_blocks,
7607 rec->start, rec->max_size);
7613 /* step three, recreate all the refs we did find */
7614 while(cur != &rec->backrefs) {
7615 back = to_extent_backref(cur);
7619 * if we didn't find any references, don't create a
7622 if (!back->found_ref)
7625 rec->bad_full_backref = 0;
7626 ret = record_extent(trans, info, path, rec, back, allocated, flags);
7634 int err = btrfs_commit_transaction(trans, info->extent_root);
7639 btrfs_free_path(path);
7643 static int fixup_extent_flags(struct btrfs_fs_info *fs_info,
7644 struct extent_record *rec)
7646 struct btrfs_trans_handle *trans;
7647 struct btrfs_root *root = fs_info->extent_root;
7648 struct btrfs_path *path;
7649 struct btrfs_extent_item *ei;
7650 struct btrfs_key key;
7654 key.objectid = rec->start;
7655 if (rec->metadata) {
7656 key.type = BTRFS_METADATA_ITEM_KEY;
7657 key.offset = rec->info_level;
7659 key.type = BTRFS_EXTENT_ITEM_KEY;
7660 key.offset = rec->max_size;
7663 path = btrfs_alloc_path();
7667 trans = btrfs_start_transaction(root, 0);
7668 if (IS_ERR(trans)) {
7669 btrfs_free_path(path);
7670 return PTR_ERR(trans);
7673 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7675 btrfs_free_path(path);
7676 btrfs_commit_transaction(trans, root);
7679 fprintf(stderr, "Didn't find extent for %llu\n",
7680 (unsigned long long)rec->start);
7681 btrfs_free_path(path);
7682 btrfs_commit_transaction(trans, root);
7686 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
7687 struct btrfs_extent_item);
7688 flags = btrfs_extent_flags(path->nodes[0], ei);
7689 if (rec->flag_block_full_backref) {
7690 fprintf(stderr, "setting full backref on %llu\n",
7691 (unsigned long long)key.objectid);
7692 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7694 fprintf(stderr, "clearing full backref on %llu\n",
7695 (unsigned long long)key.objectid);
7696 flags &= ~BTRFS_BLOCK_FLAG_FULL_BACKREF;
7698 btrfs_set_extent_flags(path->nodes[0], ei, flags);
7699 btrfs_mark_buffer_dirty(path->nodes[0]);
7700 btrfs_free_path(path);
7701 return btrfs_commit_transaction(trans, root);
7704 /* right now we only prune from the extent allocation tree */
7705 static int prune_one_block(struct btrfs_trans_handle *trans,
7706 struct btrfs_fs_info *info,
7707 struct btrfs_corrupt_block *corrupt)
7710 struct btrfs_path path;
7711 struct extent_buffer *eb;
7715 int level = corrupt->level + 1;
7717 btrfs_init_path(&path);
7719 /* we want to stop at the parent to our busted block */
7720 path.lowest_level = level;
7722 ret = btrfs_search_slot(trans, info->extent_root,
7723 &corrupt->key, &path, -1, 1);
7728 eb = path.nodes[level];
7735 * hopefully the search gave us the block we want to prune,
7736 * lets try that first
7738 slot = path.slots[level];
7739 found = btrfs_node_blockptr(eb, slot);
7740 if (found == corrupt->cache.start)
7743 nritems = btrfs_header_nritems(eb);
7745 /* the search failed, lets scan this node and hope we find it */
7746 for (slot = 0; slot < nritems; slot++) {
7747 found = btrfs_node_blockptr(eb, slot);
7748 if (found == corrupt->cache.start)
7752 * we couldn't find the bad block. TODO, search all the nodes for pointers
7755 if (eb == info->extent_root->node) {
7760 btrfs_release_path(&path);
7765 printk("deleting pointer to block %Lu\n", corrupt->cache.start);
7766 ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
7769 btrfs_release_path(&path);
7773 static int prune_corrupt_blocks(struct btrfs_fs_info *info)
7775 struct btrfs_trans_handle *trans = NULL;
7776 struct cache_extent *cache;
7777 struct btrfs_corrupt_block *corrupt;
7780 cache = search_cache_extent(info->corrupt_blocks, 0);
7784 trans = btrfs_start_transaction(info->extent_root, 1);
7786 return PTR_ERR(trans);
7788 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
7789 prune_one_block(trans, info, corrupt);
7790 remove_cache_extent(info->corrupt_blocks, cache);
7793 return btrfs_commit_transaction(trans, info->extent_root);
7797 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
7799 struct btrfs_block_group_cache *cache;
7804 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
7805 &start, &end, EXTENT_DIRTY);
7808 clear_extent_dirty(&fs_info->free_space_cache, start, end,
7814 cache = btrfs_lookup_first_block_group(fs_info, start);
7819 start = cache->key.objectid + cache->key.offset;
7823 static int check_extent_refs(struct btrfs_root *root,
7824 struct cache_tree *extent_cache)
7826 struct extent_record *rec;
7827 struct cache_extent *cache;
7836 * if we're doing a repair, we have to make sure
7837 * we don't allocate from the problem extents.
7838 * In the worst case, this will be all the
7841 cache = search_cache_extent(extent_cache, 0);
7843 rec = container_of(cache, struct extent_record, cache);
7844 set_extent_dirty(root->fs_info->excluded_extents,
7846 rec->start + rec->max_size - 1,
7848 cache = next_cache_extent(cache);
7851 /* pin down all the corrupted blocks too */
7852 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
7854 set_extent_dirty(root->fs_info->excluded_extents,
7856 cache->start + cache->size - 1,
7858 cache = next_cache_extent(cache);
7860 prune_corrupt_blocks(root->fs_info);
7861 reset_cached_block_groups(root->fs_info);
7864 reset_cached_block_groups(root->fs_info);
7867 * We need to delete any duplicate entries we find first otherwise we
7868 * could mess up the extent tree when we have backrefs that actually
7869 * belong to a different extent item and not the weird duplicate one.
7871 while (repair && !list_empty(&duplicate_extents)) {
7872 rec = to_extent_record(duplicate_extents.next);
7873 list_del_init(&rec->list);
7875 /* Sometimes we can find a backref before we find an actual
7876 * extent, so we need to process it a little bit to see if there
7877 * truly are multiple EXTENT_ITEM_KEY's for the same range, or
7878 * if this is a backref screwup. If we need to delete stuff
7879 * process_duplicates() will return 0, otherwise it will return
7882 if (process_duplicates(root, extent_cache, rec))
7884 ret = delete_duplicate_records(root, rec);
7888 * delete_duplicate_records will return the number of entries
7889 * deleted, so if it's greater than 0 then we know we actually
7890 * did something and we need to remove.
7904 cache = search_cache_extent(extent_cache, 0);
7907 rec = container_of(cache, struct extent_record, cache);
7908 if (rec->num_duplicates) {
7909 fprintf(stderr, "extent item %llu has multiple extent "
7910 "items\n", (unsigned long long)rec->start);
7915 if (rec->refs != rec->extent_item_refs) {
7916 fprintf(stderr, "ref mismatch on [%llu %llu] ",
7917 (unsigned long long)rec->start,
7918 (unsigned long long)rec->nr);
7919 fprintf(stderr, "extent item %llu, found %llu\n",
7920 (unsigned long long)rec->extent_item_refs,
7921 (unsigned long long)rec->refs);
7922 ret = record_orphan_data_extents(root->fs_info, rec);
7929 * we can't use the extent to repair file
7930 * extent, let the fallback method handle it.
7932 if (!fixed && repair) {
7933 ret = fixup_extent_refs(
7944 if (all_backpointers_checked(rec, 1)) {
7945 fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
7946 (unsigned long long)rec->start,
7947 (unsigned long long)rec->nr);
7949 if (!fixed && !recorded && repair) {
7950 ret = fixup_extent_refs(root->fs_info,
7959 if (!rec->owner_ref_checked) {
7960 fprintf(stderr, "owner ref check failed [%llu %llu]\n",
7961 (unsigned long long)rec->start,
7962 (unsigned long long)rec->nr);
7963 if (!fixed && !recorded && repair) {
7964 ret = fixup_extent_refs(root->fs_info,
7973 if (rec->bad_full_backref) {
7974 fprintf(stderr, "bad full backref, on [%llu]\n",
7975 (unsigned long long)rec->start);
7977 ret = fixup_extent_flags(root->fs_info, rec);
7986 * Although it's not a extent ref's problem, we reuse this
7987 * routine for error reporting.
7988 * No repair function yet.
7990 if (rec->crossing_stripes) {
7992 "bad metadata [%llu, %llu) crossing stripe boundary\n",
7993 rec->start, rec->start + rec->max_size);
7998 if (rec->wrong_chunk_type) {
8000 "bad extent [%llu, %llu), type mismatch with chunk\n",
8001 rec->start, rec->start + rec->max_size);
8006 remove_cache_extent(extent_cache, cache);
8007 free_all_extent_backrefs(rec);
8008 if (!init_extent_tree && repair && (!cur_err || fixed))
8009 clear_extent_dirty(root->fs_info->excluded_extents,
8011 rec->start + rec->max_size - 1,
8017 if (ret && ret != -EAGAIN) {
8018 fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
8021 struct btrfs_trans_handle *trans;
8023 root = root->fs_info->extent_root;
8024 trans = btrfs_start_transaction(root, 1);
8025 if (IS_ERR(trans)) {
8026 ret = PTR_ERR(trans);
8030 btrfs_fix_block_accounting(trans, root);
8031 ret = btrfs_commit_transaction(trans, root);
8036 fprintf(stderr, "repaired damaged extent references\n");
8042 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
8046 if (type & BTRFS_BLOCK_GROUP_RAID0) {
8047 stripe_size = length;
8048 stripe_size /= num_stripes;
8049 } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
8050 stripe_size = length * 2;
8051 stripe_size /= num_stripes;
8052 } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
8053 stripe_size = length;
8054 stripe_size /= (num_stripes - 1);
8055 } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
8056 stripe_size = length;
8057 stripe_size /= (num_stripes - 2);
8059 stripe_size = length;
8065 * Check the chunk with its block group/dev list ref:
8066 * Return 0 if all refs seems valid.
8067 * Return 1 if part of refs seems valid, need later check for rebuild ref
8068 * like missing block group and needs to search extent tree to rebuild them.
8069 * Return -1 if essential refs are missing and unable to rebuild.
8071 static int check_chunk_refs(struct chunk_record *chunk_rec,
8072 struct block_group_tree *block_group_cache,
8073 struct device_extent_tree *dev_extent_cache,
8076 struct cache_extent *block_group_item;
8077 struct block_group_record *block_group_rec;
8078 struct cache_extent *dev_extent_item;
8079 struct device_extent_record *dev_extent_rec;
8083 int metadump_v2 = 0;
8087 block_group_item = lookup_cache_extent(&block_group_cache->tree,
8090 if (block_group_item) {
8091 block_group_rec = container_of(block_group_item,
8092 struct block_group_record,
8094 if (chunk_rec->length != block_group_rec->offset ||
8095 chunk_rec->offset != block_group_rec->objectid ||
8097 chunk_rec->type_flags != block_group_rec->flags)) {
8100 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
8101 chunk_rec->objectid,
8106 chunk_rec->type_flags,
8107 block_group_rec->objectid,
8108 block_group_rec->type,
8109 block_group_rec->offset,
8110 block_group_rec->offset,
8111 block_group_rec->objectid,
8112 block_group_rec->flags);
8115 list_del_init(&block_group_rec->list);
8116 chunk_rec->bg_rec = block_group_rec;
8121 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
8122 chunk_rec->objectid,
8127 chunk_rec->type_flags);
8134 length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
8135 chunk_rec->num_stripes);
8136 for (i = 0; i < chunk_rec->num_stripes; ++i) {
8137 devid = chunk_rec->stripes[i].devid;
8138 offset = chunk_rec->stripes[i].offset;
8139 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
8140 devid, offset, length);
8141 if (dev_extent_item) {
8142 dev_extent_rec = container_of(dev_extent_item,
8143 struct device_extent_record,
8145 if (dev_extent_rec->objectid != devid ||
8146 dev_extent_rec->offset != offset ||
8147 dev_extent_rec->chunk_offset != chunk_rec->offset ||
8148 dev_extent_rec->length != length) {
8151 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
8152 chunk_rec->objectid,
8155 chunk_rec->stripes[i].devid,
8156 chunk_rec->stripes[i].offset,
8157 dev_extent_rec->objectid,
8158 dev_extent_rec->offset,
8159 dev_extent_rec->length);
8162 list_move(&dev_extent_rec->chunk_list,
8163 &chunk_rec->dextents);
8168 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
8169 chunk_rec->objectid,
8172 chunk_rec->stripes[i].devid,
8173 chunk_rec->stripes[i].offset);
8180 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
8181 int check_chunks(struct cache_tree *chunk_cache,
8182 struct block_group_tree *block_group_cache,
8183 struct device_extent_tree *dev_extent_cache,
8184 struct list_head *good, struct list_head *bad,
8185 struct list_head *rebuild, int silent)
8187 struct cache_extent *chunk_item;
8188 struct chunk_record *chunk_rec;
8189 struct block_group_record *bg_rec;
8190 struct device_extent_record *dext_rec;
8194 chunk_item = first_cache_extent(chunk_cache);
8195 while (chunk_item) {
8196 chunk_rec = container_of(chunk_item, struct chunk_record,
8198 err = check_chunk_refs(chunk_rec, block_group_cache,
8199 dev_extent_cache, silent);
8202 if (err == 0 && good)
8203 list_add_tail(&chunk_rec->list, good);
8204 if (err > 0 && rebuild)
8205 list_add_tail(&chunk_rec->list, rebuild);
8207 list_add_tail(&chunk_rec->list, bad);
8208 chunk_item = next_cache_extent(chunk_item);
8211 list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
8214 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
8222 list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
8226 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
8237 static int check_device_used(struct device_record *dev_rec,
8238 struct device_extent_tree *dext_cache)
8240 struct cache_extent *cache;
8241 struct device_extent_record *dev_extent_rec;
8244 cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
8246 dev_extent_rec = container_of(cache,
8247 struct device_extent_record,
8249 if (dev_extent_rec->objectid != dev_rec->devid)
8252 list_del_init(&dev_extent_rec->device_list);
8253 total_byte += dev_extent_rec->length;
8254 cache = next_cache_extent(cache);
8257 if (total_byte != dev_rec->byte_used) {
8259 "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
8260 total_byte, dev_rec->byte_used, dev_rec->objectid,
8261 dev_rec->type, dev_rec->offset);
8268 /* check btrfs_dev_item -> btrfs_dev_extent */
8269 static int check_devices(struct rb_root *dev_cache,
8270 struct device_extent_tree *dev_extent_cache)
8272 struct rb_node *dev_node;
8273 struct device_record *dev_rec;
8274 struct device_extent_record *dext_rec;
8278 dev_node = rb_first(dev_cache);
8280 dev_rec = container_of(dev_node, struct device_record, node);
8281 err = check_device_used(dev_rec, dev_extent_cache);
8285 dev_node = rb_next(dev_node);
8287 list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
8290 "Device extent[%llu, %llu, %llu] didn't find its device.\n",
8291 dext_rec->objectid, dext_rec->offset, dext_rec->length);
8298 static int add_root_item_to_list(struct list_head *head,
8299 u64 objectid, u64 bytenr, u64 last_snapshot,
8300 u8 level, u8 drop_level,
8301 int level_size, struct btrfs_key *drop_key)
8304 struct root_item_record *ri_rec;
8305 ri_rec = malloc(sizeof(*ri_rec));
8308 ri_rec->bytenr = bytenr;
8309 ri_rec->objectid = objectid;
8310 ri_rec->level = level;
8311 ri_rec->level_size = level_size;
8312 ri_rec->drop_level = drop_level;
8313 ri_rec->last_snapshot = last_snapshot;
8315 memcpy(&ri_rec->drop_key, drop_key, sizeof(*drop_key));
8316 list_add_tail(&ri_rec->list, head);
8321 static void free_root_item_list(struct list_head *list)
8323 struct root_item_record *ri_rec;
8325 while (!list_empty(list)) {
8326 ri_rec = list_first_entry(list, struct root_item_record,
8328 list_del_init(&ri_rec->list);
8333 static int deal_root_from_list(struct list_head *list,
8334 struct btrfs_root *root,
8335 struct block_info *bits,
8337 struct cache_tree *pending,
8338 struct cache_tree *seen,
8339 struct cache_tree *reada,
8340 struct cache_tree *nodes,
8341 struct cache_tree *extent_cache,
8342 struct cache_tree *chunk_cache,
8343 struct rb_root *dev_cache,
8344 struct block_group_tree *block_group_cache,
8345 struct device_extent_tree *dev_extent_cache)
8350 while (!list_empty(list)) {
8351 struct root_item_record *rec;
8352 struct extent_buffer *buf;
8353 rec = list_entry(list->next,
8354 struct root_item_record, list);
8356 buf = read_tree_block(root->fs_info->tree_root,
8357 rec->bytenr, rec->level_size, 0);
8358 if (!extent_buffer_uptodate(buf)) {
8359 free_extent_buffer(buf);
8363 ret = add_root_to_pending(buf, extent_cache, pending,
8364 seen, nodes, rec->objectid);
8368 * To rebuild extent tree, we need deal with snapshot
8369 * one by one, otherwise we deal with node firstly which
8370 * can maximize readahead.
8373 ret = run_next_block(root, bits, bits_nr, &last,
8374 pending, seen, reada, nodes,
8375 extent_cache, chunk_cache,
8376 dev_cache, block_group_cache,
8377 dev_extent_cache, rec);
8381 free_extent_buffer(buf);
8382 list_del(&rec->list);
8388 ret = run_next_block(root, bits, bits_nr, &last, pending, seen,
8389 reada, nodes, extent_cache, chunk_cache,
8390 dev_cache, block_group_cache,
8391 dev_extent_cache, NULL);
8401 static int check_chunks_and_extents(struct btrfs_root *root)
8403 struct rb_root dev_cache;
8404 struct cache_tree chunk_cache;
8405 struct block_group_tree block_group_cache;
8406 struct device_extent_tree dev_extent_cache;
8407 struct cache_tree extent_cache;
8408 struct cache_tree seen;
8409 struct cache_tree pending;
8410 struct cache_tree reada;
8411 struct cache_tree nodes;
8412 struct extent_io_tree excluded_extents;
8413 struct cache_tree corrupt_blocks;
8414 struct btrfs_path path;
8415 struct btrfs_key key;
8416 struct btrfs_key found_key;
8418 struct block_info *bits;
8420 struct extent_buffer *leaf;
8422 struct btrfs_root_item ri;
8423 struct list_head dropping_trees;
8424 struct list_head normal_trees;
8425 struct btrfs_root *root1;
8430 dev_cache = RB_ROOT;
8431 cache_tree_init(&chunk_cache);
8432 block_group_tree_init(&block_group_cache);
8433 device_extent_tree_init(&dev_extent_cache);
8435 cache_tree_init(&extent_cache);
8436 cache_tree_init(&seen);
8437 cache_tree_init(&pending);
8438 cache_tree_init(&nodes);
8439 cache_tree_init(&reada);
8440 cache_tree_init(&corrupt_blocks);
8441 extent_io_tree_init(&excluded_extents);
8442 INIT_LIST_HEAD(&dropping_trees);
8443 INIT_LIST_HEAD(&normal_trees);
8446 root->fs_info->excluded_extents = &excluded_extents;
8447 root->fs_info->fsck_extent_cache = &extent_cache;
8448 root->fs_info->free_extent_hook = free_extent_hook;
8449 root->fs_info->corrupt_blocks = &corrupt_blocks;
8453 bits = malloc(bits_nr * sizeof(struct block_info));
8459 if (ctx.progress_enabled) {
8460 ctx.tp = TASK_EXTENTS;
8461 task_start(ctx.info);
8465 root1 = root->fs_info->tree_root;
8466 level = btrfs_header_level(root1->node);
8467 ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
8468 root1->node->start, 0, level, 0,
8469 root1->nodesize, NULL);
8472 root1 = root->fs_info->chunk_root;
8473 level = btrfs_header_level(root1->node);
8474 ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
8475 root1->node->start, 0, level, 0,
8476 root1->nodesize, NULL);
8479 btrfs_init_path(&path);
8482 key.type = BTRFS_ROOT_ITEM_KEY;
8483 ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
8488 leaf = path.nodes[0];
8489 slot = path.slots[0];
8490 if (slot >= btrfs_header_nritems(path.nodes[0])) {
8491 ret = btrfs_next_leaf(root, &path);
8494 leaf = path.nodes[0];
8495 slot = path.slots[0];
8497 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
8498 if (found_key.type == BTRFS_ROOT_ITEM_KEY) {
8499 unsigned long offset;
8502 offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
8503 read_extent_buffer(leaf, &ri, offset, sizeof(ri));
8504 last_snapshot = btrfs_root_last_snapshot(&ri);
8505 if (btrfs_disk_key_objectid(&ri.drop_progress) == 0) {
8506 level = btrfs_root_level(&ri);
8507 level_size = root->nodesize;
8508 ret = add_root_item_to_list(&normal_trees,
8510 btrfs_root_bytenr(&ri),
8511 last_snapshot, level,
8512 0, level_size, NULL);
8516 level = btrfs_root_level(&ri);
8517 level_size = root->nodesize;
8518 objectid = found_key.objectid;
8519 btrfs_disk_key_to_cpu(&found_key,
8521 ret = add_root_item_to_list(&dropping_trees,
8523 btrfs_root_bytenr(&ri),
8524 last_snapshot, level,
8526 level_size, &found_key);
8533 btrfs_release_path(&path);
8536 * check_block can return -EAGAIN if it fixes something, please keep
8537 * this in mind when dealing with return values from these functions, if
8538 * we get -EAGAIN we want to fall through and restart the loop.
8540 ret = deal_root_from_list(&normal_trees, root, bits, bits_nr, &pending,
8541 &seen, &reada, &nodes, &extent_cache,
8542 &chunk_cache, &dev_cache, &block_group_cache,
8549 ret = deal_root_from_list(&dropping_trees, root, bits, bits_nr,
8550 &pending, &seen, &reada, &nodes,
8551 &extent_cache, &chunk_cache, &dev_cache,
8552 &block_group_cache, &dev_extent_cache);
8559 ret = check_chunks(&chunk_cache, &block_group_cache,
8560 &dev_extent_cache, NULL, NULL, NULL, 0);
8567 ret = check_extent_refs(root, &extent_cache);
8574 ret = check_devices(&dev_cache, &dev_extent_cache);
8579 task_stop(ctx.info);
8581 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
8582 extent_io_tree_cleanup(&excluded_extents);
8583 root->fs_info->fsck_extent_cache = NULL;
8584 root->fs_info->free_extent_hook = NULL;
8585 root->fs_info->corrupt_blocks = NULL;
8586 root->fs_info->excluded_extents = NULL;
8589 free_chunk_cache_tree(&chunk_cache);
8590 free_device_cache_tree(&dev_cache);
8591 free_block_group_tree(&block_group_cache);
8592 free_device_extent_tree(&dev_extent_cache);
8593 free_extent_cache_tree(&seen);
8594 free_extent_cache_tree(&pending);
8595 free_extent_cache_tree(&reada);
8596 free_extent_cache_tree(&nodes);
8599 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
8600 free_extent_cache_tree(&seen);
8601 free_extent_cache_tree(&pending);
8602 free_extent_cache_tree(&reada);
8603 free_extent_cache_tree(&nodes);
8604 free_chunk_cache_tree(&chunk_cache);
8605 free_block_group_tree(&block_group_cache);
8606 free_device_cache_tree(&dev_cache);
8607 free_device_extent_tree(&dev_extent_cache);
8608 free_extent_record_cache(root->fs_info, &extent_cache);
8609 free_root_item_list(&normal_trees);
8610 free_root_item_list(&dropping_trees);
8611 extent_io_tree_cleanup(&excluded_extents);
8616 * Check backrefs of a tree block given by @bytenr or @eb.
8618 * @root: the root containing the @bytenr or @eb
8619 * @eb: tree block extent buffer, can be NULL
8620 * @bytenr: bytenr of the tree block to search
8621 * @level: tree level of the tree block
8622 * @owner: owner of the tree block
8624 * Return >0 for any error found and output error message
8625 * Return 0 for no error found
8627 static int check_tree_block_ref(struct btrfs_root *root,
8628 struct extent_buffer *eb, u64 bytenr,
8629 int level, u64 owner)
8631 struct btrfs_key key;
8632 struct btrfs_root *extent_root = root->fs_info->extent_root;
8633 struct btrfs_path path;
8634 struct btrfs_extent_item *ei;
8635 struct btrfs_extent_inline_ref *iref;
8636 struct extent_buffer *leaf;
8642 u32 nodesize = root->nodesize;
8649 btrfs_init_path(&path);
8650 key.objectid = bytenr;
8651 if (btrfs_fs_incompat(root->fs_info,
8652 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA))
8653 key.type = BTRFS_METADATA_ITEM_KEY;
8655 key.type = BTRFS_EXTENT_ITEM_KEY;
8656 key.offset = (u64)-1;
8658 /* Search for the backref in extent tree */
8659 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
8661 err |= BACKREF_MISSING;
8664 ret = btrfs_previous_extent_item(extent_root, &path, bytenr);
8666 err |= BACKREF_MISSING;
8670 leaf = path.nodes[0];
8671 slot = path.slots[0];
8672 btrfs_item_key_to_cpu(leaf, &key, slot);
8674 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
8676 if (key.type == BTRFS_METADATA_ITEM_KEY) {
8677 skinny_level = (int)key.offset;
8678 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
8680 struct btrfs_tree_block_info *info;
8682 info = (struct btrfs_tree_block_info *)(ei + 1);
8683 skinny_level = btrfs_tree_block_level(leaf, info);
8684 iref = (struct btrfs_extent_inline_ref *)(info + 1);
8691 if (!(btrfs_extent_flags(leaf, ei) &
8692 BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
8694 "extent[%llu %u] backref type mismatch, missing bit: %llx",
8695 key.objectid, nodesize,
8696 BTRFS_EXTENT_FLAG_TREE_BLOCK);
8697 err = BACKREF_MISMATCH;
8699 header_gen = btrfs_header_generation(eb);
8700 extent_gen = btrfs_extent_generation(leaf, ei);
8701 if (header_gen != extent_gen) {
8703 "extent[%llu %u] backref generation mismatch, wanted: %llu, have: %llu",
8704 key.objectid, nodesize, header_gen,
8706 err = BACKREF_MISMATCH;
8708 if (level != skinny_level) {
8710 "extent[%llu %u] level mismatch, wanted: %u, have: %u",
8711 key.objectid, nodesize, level, skinny_level);
8712 err = BACKREF_MISMATCH;
8714 if (!is_fstree(owner) && btrfs_extent_refs(leaf, ei) != 1) {
8716 "extent[%llu %u] is referred by other roots than %llu",
8717 key.objectid, nodesize, root->objectid);
8718 err = BACKREF_MISMATCH;
8723 * Iterate the extent/metadata item to find the exact backref
8725 item_size = btrfs_item_size_nr(leaf, slot);
8726 ptr = (unsigned long)iref;
8727 end = (unsigned long)ei + item_size;
8729 iref = (struct btrfs_extent_inline_ref *)ptr;
8730 type = btrfs_extent_inline_ref_type(leaf, iref);
8731 offset = btrfs_extent_inline_ref_offset(leaf, iref);
8733 if (type == BTRFS_TREE_BLOCK_REF_KEY &&
8734 (offset == root->objectid || offset == owner)) {
8736 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
8737 /* Check if the backref points to valid referencer */
8738 found_ref = !check_tree_block_ref(root, NULL, offset,
8744 ptr += btrfs_extent_inline_ref_size(type);
8748 * Inlined extent item doesn't have what we need, check
8749 * TREE_BLOCK_REF_KEY
8752 btrfs_release_path(&path);
8753 key.objectid = bytenr;
8754 key.type = BTRFS_TREE_BLOCK_REF_KEY;
8755 key.offset = root->objectid;
8757 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
8762 err |= BACKREF_MISSING;
8764 btrfs_release_path(&path);
8765 if (eb && (err & BACKREF_MISSING))
8766 error("extent[%llu %u] backref lost (owner: %llu, level: %u)",
8767 bytenr, nodesize, owner, level);
8772 * Check EXTENT_DATA item, mainly for its dbackref in extent tree
8774 * Return >0 any error found and output error message
8775 * Return 0 for no error found
8777 static int check_extent_data_item(struct btrfs_root *root,
8778 struct extent_buffer *eb, int slot)
8780 struct btrfs_file_extent_item *fi;
8781 struct btrfs_path path;
8782 struct btrfs_root *extent_root = root->fs_info->extent_root;
8783 struct btrfs_key fi_key;
8784 struct btrfs_key dbref_key;
8785 struct extent_buffer *leaf;
8786 struct btrfs_extent_item *ei;
8787 struct btrfs_extent_inline_ref *iref;
8788 struct btrfs_extent_data_ref *dref;
8790 u64 file_extent_gen;
8793 u64 extent_num_bytes;
8801 int found_dbackref = 0;
8805 btrfs_item_key_to_cpu(eb, &fi_key, slot);
8806 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
8807 file_extent_gen = btrfs_file_extent_generation(eb, fi);
8809 /* Nothing to check for hole and inline data extents */
8810 if (btrfs_file_extent_type(eb, fi) == BTRFS_FILE_EXTENT_INLINE ||
8811 btrfs_file_extent_disk_bytenr(eb, fi) == 0)
8814 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8815 disk_num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8816 extent_num_bytes = btrfs_file_extent_num_bytes(eb, fi);
8818 /* Check unaligned disk_num_bytes and num_bytes */
8819 if (!IS_ALIGNED(disk_num_bytes, root->sectorsize)) {
8821 "file extent [%llu, %llu] has unaligned disk num bytes: %llu, should be aligned to %u",
8822 fi_key.objectid, fi_key.offset, disk_num_bytes,
8824 err |= BYTES_UNALIGNED;
8826 data_bytes_allocated += disk_num_bytes;
8828 if (!IS_ALIGNED(extent_num_bytes, root->sectorsize)) {
8830 "file extent [%llu, %llu] has unaligned num bytes: %llu, should be aligned to %u",
8831 fi_key.objectid, fi_key.offset, extent_num_bytes,
8833 err |= BYTES_UNALIGNED;
8835 data_bytes_referenced += extent_num_bytes;
8837 owner = btrfs_header_owner(eb);
8839 /* Check the extent item of the file extent in extent tree */
8840 btrfs_init_path(&path);
8841 dbref_key.objectid = btrfs_file_extent_disk_bytenr(eb, fi);
8842 dbref_key.type = BTRFS_EXTENT_ITEM_KEY;
8843 dbref_key.offset = btrfs_file_extent_disk_num_bytes(eb, fi);
8845 ret = btrfs_search_slot(NULL, extent_root, &dbref_key, &path, 0, 0);
8847 err |= BACKREF_MISSING;
8851 leaf = path.nodes[0];
8852 slot = path.slots[0];
8853 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
8855 extent_flags = btrfs_extent_flags(leaf, ei);
8856 extent_gen = btrfs_extent_generation(leaf, ei);
8858 if (!(extent_flags & BTRFS_EXTENT_FLAG_DATA)) {
8860 "extent[%llu %llu] backref type mismatch, wanted bit: %llx",
8861 disk_bytenr, disk_num_bytes,
8862 BTRFS_EXTENT_FLAG_DATA);
8863 err |= BACKREF_MISMATCH;
8866 if (file_extent_gen < extent_gen) {
8868 "extent[%llu %llu] backref generation mismatch, wanted: <=%llu, have: %llu",
8869 disk_bytenr, disk_num_bytes, file_extent_gen,
8871 err |= BACKREF_MISMATCH;
8874 /* Check data backref inside that extent item */
8875 item_size = btrfs_item_size_nr(leaf, path.slots[0]);
8876 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
8877 ptr = (unsigned long)iref;
8878 end = (unsigned long)ei + item_size;
8880 iref = (struct btrfs_extent_inline_ref *)ptr;
8881 type = btrfs_extent_inline_ref_type(leaf, iref);
8882 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
8884 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
8885 ref_root = btrfs_extent_data_ref_root(leaf, dref);
8886 if (ref_root == owner || ref_root == root->objectid)
8888 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
8889 found_dbackref = !check_tree_block_ref(root, NULL,
8890 btrfs_extent_inline_ref_offset(leaf, iref),
8896 ptr += btrfs_extent_inline_ref_size(type);
8899 /* Didn't found inlined data backref, try EXTENT_DATA_REF_KEY */
8900 if (!found_dbackref) {
8901 btrfs_release_path(&path);
8903 btrfs_init_path(&path);
8904 dbref_key.objectid = btrfs_file_extent_disk_bytenr(eb, fi);
8905 dbref_key.type = BTRFS_EXTENT_DATA_REF_KEY;
8906 dbref_key.offset = hash_extent_data_ref(root->objectid,
8907 fi_key.objectid, fi_key.offset);
8909 ret = btrfs_search_slot(NULL, root->fs_info->extent_root,
8910 &dbref_key, &path, 0, 0);
8915 if (!found_dbackref)
8916 err |= BACKREF_MISSING;
8918 btrfs_release_path(&path);
8919 if (err & BACKREF_MISSING) {
8920 error("data extent[%llu %llu] backref lost",
8921 disk_bytenr, disk_num_bytes);
8927 * Get real tree block level for the case like shared block
8928 * Return >= 0 as tree level
8929 * Return <0 for error
8931 static int query_tree_block_level(struct btrfs_fs_info *fs_info, u64 bytenr)
8933 struct extent_buffer *eb;
8934 struct btrfs_path path;
8935 struct btrfs_key key;
8936 struct btrfs_extent_item *ei;
8939 u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
8944 /* Search extent tree for extent generation and level */
8945 key.objectid = bytenr;
8946 key.type = BTRFS_METADATA_ITEM_KEY;
8947 key.offset = (u64)-1;
8949 btrfs_init_path(&path);
8950 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, &path, 0, 0);
8953 ret = btrfs_previous_extent_item(fs_info->extent_root, &path, bytenr);
8961 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
8962 ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
8963 struct btrfs_extent_item);
8964 flags = btrfs_extent_flags(path.nodes[0], ei);
8965 if (!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
8970 /* Get transid for later read_tree_block() check */
8971 transid = btrfs_extent_generation(path.nodes[0], ei);
8973 /* Get backref level as one source */
8974 if (key.type == BTRFS_METADATA_ITEM_KEY) {
8975 backref_level = key.offset;
8977 struct btrfs_tree_block_info *info;
8979 info = (struct btrfs_tree_block_info *)(ei + 1);
8980 backref_level = btrfs_tree_block_level(path.nodes[0], info);
8982 btrfs_release_path(&path);
8984 /* Get level from tree block as an alternative source */
8985 eb = read_tree_block_fs_info(fs_info, bytenr, nodesize, transid);
8986 if (!extent_buffer_uptodate(eb)) {
8987 free_extent_buffer(eb);
8990 header_level = btrfs_header_level(eb);
8991 free_extent_buffer(eb);
8993 if (header_level != backref_level)
8995 return header_level;
8998 btrfs_release_path(&path);
9003 * Check if a tree block backref is valid (points to a valid tree block)
9004 * if level == -1, level will be resolved
9005 * Return >0 for any error found and print error message
9007 static int check_tree_block_backref(struct btrfs_fs_info *fs_info, u64 root_id,
9008 u64 bytenr, int level)
9010 struct btrfs_root *root;
9011 struct btrfs_key key;
9012 struct btrfs_path path;
9013 struct extent_buffer *eb;
9014 struct extent_buffer *node;
9015 u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9019 /* Query level for level == -1 special case */
9021 level = query_tree_block_level(fs_info, bytenr);
9023 err |= REFERENCER_MISSING;
9027 key.objectid = root_id;
9028 key.type = BTRFS_ROOT_ITEM_KEY;
9029 key.offset = (u64)-1;
9031 root = btrfs_read_fs_root(fs_info, &key);
9033 err |= REFERENCER_MISSING;
9037 /* Read out the tree block to get item/node key */
9038 eb = read_tree_block(root, bytenr, root->nodesize, 0);
9039 if (!extent_buffer_uptodate(eb)) {
9040 err |= REFERENCER_MISSING;
9041 free_extent_buffer(eb);
9045 /* Empty tree, no need to check key */
9046 if (!btrfs_header_nritems(eb) && !level) {
9047 free_extent_buffer(eb);
9052 btrfs_node_key_to_cpu(eb, &key, 0);
9054 btrfs_item_key_to_cpu(eb, &key, 0);
9056 free_extent_buffer(eb);
9058 btrfs_init_path(&path);
9059 path.lowest_level = level;
9060 /* Search with the first key, to ensure we can reach it */
9061 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
9063 err |= REFERENCER_MISSING;
9067 node = path.nodes[level];
9068 if (btrfs_header_bytenr(node) != bytenr) {
9070 "extent [%llu %d] referencer bytenr mismatch, wanted: %llu, have: %llu",
9071 bytenr, nodesize, bytenr,
9072 btrfs_header_bytenr(node));
9073 err |= REFERENCER_MISMATCH;
9075 if (btrfs_header_level(node) != level) {
9077 "extent [%llu %d] referencer level mismatch, wanted: %d, have: %d",
9078 bytenr, nodesize, level,
9079 btrfs_header_level(node));
9080 err |= REFERENCER_MISMATCH;
9084 btrfs_release_path(&path);
9086 if (err & REFERENCER_MISSING) {
9088 error("extent [%llu %d] lost referencer (owner: %llu)",
9089 bytenr, nodesize, root_id);
9092 "extent [%llu %d] lost referencer (owner: %llu, level: %u)",
9093 bytenr, nodesize, root_id, level);
9100 * Check referencer for shared block backref
9101 * If level == -1, this function will resolve the level.
9103 static int check_shared_block_backref(struct btrfs_fs_info *fs_info,
9104 u64 parent, u64 bytenr, int level)
9106 struct extent_buffer *eb;
9107 u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9109 int found_parent = 0;
9112 eb = read_tree_block_fs_info(fs_info, parent, nodesize, 0);
9113 if (!extent_buffer_uptodate(eb))
9117 level = query_tree_block_level(fs_info, bytenr);
9121 if (level + 1 != btrfs_header_level(eb))
9124 nr = btrfs_header_nritems(eb);
9125 for (i = 0; i < nr; i++) {
9126 if (bytenr == btrfs_node_blockptr(eb, i)) {
9132 free_extent_buffer(eb);
9133 if (!found_parent) {
9135 "shared extent[%llu %u] lost its parent (parent: %llu, level: %u)",
9136 bytenr, nodesize, parent, level);
9137 return REFERENCER_MISSING;
9143 * Check referencer for normal (inlined) data ref
9144 * If len == 0, it will be resolved by searching in extent tree
9146 static int check_extent_data_backref(struct btrfs_fs_info *fs_info,
9147 u64 root_id, u64 objectid, u64 offset,
9148 u64 bytenr, u64 len, u32 count)
9150 struct btrfs_root *root;
9151 struct btrfs_root *extent_root = fs_info->extent_root;
9152 struct btrfs_key key;
9153 struct btrfs_path path;
9154 struct extent_buffer *leaf;
9155 struct btrfs_file_extent_item *fi;
9156 u32 found_count = 0;
9161 key.objectid = bytenr;
9162 key.type = BTRFS_EXTENT_ITEM_KEY;
9163 key.offset = (u64)-1;
9165 btrfs_init_path(&path);
9166 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
9169 ret = btrfs_previous_extent_item(extent_root, &path, bytenr);
9172 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
9173 if (key.objectid != bytenr ||
9174 key.type != BTRFS_EXTENT_ITEM_KEY)
9177 btrfs_release_path(&path);
9179 key.objectid = root_id;
9180 key.type = BTRFS_ROOT_ITEM_KEY;
9181 key.offset = (u64)-1;
9182 btrfs_init_path(&path);
9184 root = btrfs_read_fs_root(fs_info, &key);
9188 key.objectid = objectid;
9189 key.type = BTRFS_EXTENT_DATA_KEY;
9191 * It can be nasty as data backref offset is
9192 * file offset - file extent offset, which is smaller or
9193 * equal to original backref offset. The only special case is
9194 * overflow. So we need to special check and do further search.
9196 key.offset = offset & (1ULL << 63) ? 0 : offset;
9198 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
9203 * Search afterwards to get correct one
9204 * NOTE: As we must do a comprehensive check on the data backref to
9205 * make sure the dref count also matches, we must iterate all file
9206 * extents for that inode.
9209 leaf = path.nodes[0];
9210 slot = path.slots[0];
9212 btrfs_item_key_to_cpu(leaf, &key, slot);
9213 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
9215 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
9217 * Except normal disk bytenr and disk num bytes, we still
9218 * need to do extra check on dbackref offset as
9219 * dbackref offset = file_offset - file_extent_offset
9221 if (btrfs_file_extent_disk_bytenr(leaf, fi) == bytenr &&
9222 btrfs_file_extent_disk_num_bytes(leaf, fi) == len &&
9223 (u64)(key.offset - btrfs_file_extent_offset(leaf, fi)) ==
9227 ret = btrfs_next_item(root, &path);
9232 btrfs_release_path(&path);
9233 if (found_count != count) {
9235 "extent[%llu, %llu] referencer count mismatch (root: %llu, owner: %llu, offset: %llu) wanted: %u, have: %u",
9236 bytenr, len, root_id, objectid, offset, count, found_count);
9237 return REFERENCER_MISSING;
9243 * Check if the referencer of a shared data backref exists
9245 static int check_shared_data_backref(struct btrfs_fs_info *fs_info,
9246 u64 parent, u64 bytenr)
9248 struct extent_buffer *eb;
9249 struct btrfs_key key;
9250 struct btrfs_file_extent_item *fi;
9251 u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9253 int found_parent = 0;
9256 eb = read_tree_block_fs_info(fs_info, parent, nodesize, 0);
9257 if (!extent_buffer_uptodate(eb))
9260 nr = btrfs_header_nritems(eb);
9261 for (i = 0; i < nr; i++) {
9262 btrfs_item_key_to_cpu(eb, &key, i);
9263 if (key.type != BTRFS_EXTENT_DATA_KEY)
9266 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
9267 if (btrfs_file_extent_type(eb, fi) == BTRFS_FILE_EXTENT_INLINE)
9270 if (btrfs_file_extent_disk_bytenr(eb, fi) == bytenr) {
9277 free_extent_buffer(eb);
9278 if (!found_parent) {
9279 error("shared extent %llu referencer lost (parent: %llu)",
9281 return REFERENCER_MISSING;
9287 * This function will check a given extent item, including its backref and
9288 * itself (like crossing stripe boundary and type)
9290 * Since we don't use extent_record anymore, introduce new error bit
9292 static int check_extent_item(struct btrfs_fs_info *fs_info,
9293 struct extent_buffer *eb, int slot)
9295 struct btrfs_extent_item *ei;
9296 struct btrfs_extent_inline_ref *iref;
9297 struct btrfs_extent_data_ref *dref;
9301 u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9302 u32 item_size = btrfs_item_size_nr(eb, slot);
9307 struct btrfs_key key;
9311 btrfs_item_key_to_cpu(eb, &key, slot);
9312 if (key.type == BTRFS_EXTENT_ITEM_KEY)
9313 bytes_used += key.offset;
9315 bytes_used += nodesize;
9317 if (item_size < sizeof(*ei)) {
9319 * COMPAT_EXTENT_TREE_V0 case, but it's already a super
9320 * old thing when on disk format is still un-determined.
9321 * No need to care about it anymore
9323 error("unsupported COMPAT_EXTENT_TREE_V0 detected");
9327 ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
9328 flags = btrfs_extent_flags(eb, ei);
9330 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
9332 if (metadata && check_crossing_stripes(global_info, key.objectid,
9334 error("bad metadata [%llu, %llu) crossing stripe boundary",
9335 key.objectid, key.objectid + nodesize);
9336 err |= CROSSING_STRIPE_BOUNDARY;
9339 ptr = (unsigned long)(ei + 1);
9341 if (metadata && key.type == BTRFS_EXTENT_ITEM_KEY) {
9342 /* Old EXTENT_ITEM metadata */
9343 struct btrfs_tree_block_info *info;
9345 info = (struct btrfs_tree_block_info *)ptr;
9346 level = btrfs_tree_block_level(eb, info);
9347 ptr += sizeof(struct btrfs_tree_block_info);
9349 /* New METADATA_ITEM */
9352 end = (unsigned long)ei + item_size;
9355 err |= ITEM_SIZE_MISMATCH;
9359 /* Now check every backref in this extent item */
9361 iref = (struct btrfs_extent_inline_ref *)ptr;
9362 type = btrfs_extent_inline_ref_type(eb, iref);
9363 offset = btrfs_extent_inline_ref_offset(eb, iref);
9365 case BTRFS_TREE_BLOCK_REF_KEY:
9366 ret = check_tree_block_backref(fs_info, offset, key.objectid,
9370 case BTRFS_SHARED_BLOCK_REF_KEY:
9371 ret = check_shared_block_backref(fs_info, offset, key.objectid,
9375 case BTRFS_EXTENT_DATA_REF_KEY:
9376 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
9377 ret = check_extent_data_backref(fs_info,
9378 btrfs_extent_data_ref_root(eb, dref),
9379 btrfs_extent_data_ref_objectid(eb, dref),
9380 btrfs_extent_data_ref_offset(eb, dref),
9381 key.objectid, key.offset,
9382 btrfs_extent_data_ref_count(eb, dref));
9385 case BTRFS_SHARED_DATA_REF_KEY:
9386 ret = check_shared_data_backref(fs_info, offset, key.objectid);
9390 error("extent[%llu %d %llu] has unknown ref type: %d",
9391 key.objectid, key.type, key.offset, type);
9392 err |= UNKNOWN_TYPE;
9396 ptr += btrfs_extent_inline_ref_size(type);
9405 * Check if a dev extent item is referred correctly by its chunk
9407 static int check_dev_extent_item(struct btrfs_fs_info *fs_info,
9408 struct extent_buffer *eb, int slot)
9410 struct btrfs_root *chunk_root = fs_info->chunk_root;
9411 struct btrfs_dev_extent *ptr;
9412 struct btrfs_path path;
9413 struct btrfs_key chunk_key;
9414 struct btrfs_key devext_key;
9415 struct btrfs_chunk *chunk;
9416 struct extent_buffer *l;
9420 int found_chunk = 0;
9423 btrfs_item_key_to_cpu(eb, &devext_key, slot);
9424 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_extent);
9425 length = btrfs_dev_extent_length(eb, ptr);
9427 chunk_key.objectid = btrfs_dev_extent_chunk_objectid(eb, ptr);
9428 chunk_key.type = BTRFS_CHUNK_ITEM_KEY;
9429 chunk_key.offset = btrfs_dev_extent_chunk_offset(eb, ptr);
9431 btrfs_init_path(&path);
9432 ret = btrfs_search_slot(NULL, chunk_root, &chunk_key, &path, 0, 0);
9437 chunk = btrfs_item_ptr(l, path.slots[0], struct btrfs_chunk);
9438 if (btrfs_chunk_length(l, chunk) != length)
9441 num_stripes = btrfs_chunk_num_stripes(l, chunk);
9442 for (i = 0; i < num_stripes; i++) {
9443 u64 devid = btrfs_stripe_devid_nr(l, chunk, i);
9444 u64 offset = btrfs_stripe_offset_nr(l, chunk, i);
9446 if (devid == devext_key.objectid &&
9447 offset == devext_key.offset) {
9453 btrfs_release_path(&path);
9456 "device extent[%llu, %llu, %llu] did not find the related chunk",
9457 devext_key.objectid, devext_key.offset, length);
9458 return REFERENCER_MISSING;
9464 * Check if the used space is correct with the dev item
9466 static int check_dev_item(struct btrfs_fs_info *fs_info,
9467 struct extent_buffer *eb, int slot)
9469 struct btrfs_root *dev_root = fs_info->dev_root;
9470 struct btrfs_dev_item *dev_item;
9471 struct btrfs_path path;
9472 struct btrfs_key key;
9473 struct btrfs_dev_extent *ptr;
9479 dev_item = btrfs_item_ptr(eb, slot, struct btrfs_dev_item);
9480 dev_id = btrfs_device_id(eb, dev_item);
9481 used = btrfs_device_bytes_used(eb, dev_item);
9483 key.objectid = dev_id;
9484 key.type = BTRFS_DEV_EXTENT_KEY;
9487 btrfs_init_path(&path);
9488 ret = btrfs_search_slot(NULL, dev_root, &key, &path, 0, 0);
9490 btrfs_item_key_to_cpu(eb, &key, slot);
9491 error("cannot find any related dev extent for dev[%llu, %u, %llu]",
9492 key.objectid, key.type, key.offset);
9493 btrfs_release_path(&path);
9494 return REFERENCER_MISSING;
9497 /* Iterate dev_extents to calculate the used space of a device */
9499 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
9501 if (key.objectid > dev_id)
9503 if (key.type != BTRFS_DEV_EXTENT_KEY || key.objectid != dev_id)
9506 ptr = btrfs_item_ptr(path.nodes[0], path.slots[0],
9507 struct btrfs_dev_extent);
9508 total += btrfs_dev_extent_length(path.nodes[0], ptr);
9510 ret = btrfs_next_item(dev_root, &path);
9514 btrfs_release_path(&path);
9516 if (used != total) {
9517 btrfs_item_key_to_cpu(eb, &key, slot);
9519 "Dev extent's total-byte %llu is not equal to bytes-used %llu in dev[%llu, %u, %llu]",
9520 total, used, BTRFS_ROOT_TREE_OBJECTID,
9521 BTRFS_DEV_EXTENT_KEY, dev_id);
9522 return ACCOUNTING_MISMATCH;
9528 * Check a block group item with its referener (chunk) and its used space
9529 * with extent/metadata item
9531 static int check_block_group_item(struct btrfs_fs_info *fs_info,
9532 struct extent_buffer *eb, int slot)
9534 struct btrfs_root *extent_root = fs_info->extent_root;
9535 struct btrfs_root *chunk_root = fs_info->chunk_root;
9536 struct btrfs_block_group_item *bi;
9537 struct btrfs_block_group_item bg_item;
9538 struct btrfs_path path;
9539 struct btrfs_key bg_key;
9540 struct btrfs_key chunk_key;
9541 struct btrfs_key extent_key;
9542 struct btrfs_chunk *chunk;
9543 struct extent_buffer *leaf;
9544 struct btrfs_extent_item *ei;
9545 u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
9553 btrfs_item_key_to_cpu(eb, &bg_key, slot);
9554 bi = btrfs_item_ptr(eb, slot, struct btrfs_block_group_item);
9555 read_extent_buffer(eb, &bg_item, (unsigned long)bi, sizeof(bg_item));
9556 used = btrfs_block_group_used(&bg_item);
9557 bg_flags = btrfs_block_group_flags(&bg_item);
9559 chunk_key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
9560 chunk_key.type = BTRFS_CHUNK_ITEM_KEY;
9561 chunk_key.offset = bg_key.objectid;
9563 btrfs_init_path(&path);
9564 /* Search for the referencer chunk */
9565 ret = btrfs_search_slot(NULL, chunk_root, &chunk_key, &path, 0, 0);
9568 "block group[%llu %llu] did not find the related chunk item",
9569 bg_key.objectid, bg_key.offset);
9570 err |= REFERENCER_MISSING;
9572 chunk = btrfs_item_ptr(path.nodes[0], path.slots[0],
9573 struct btrfs_chunk);
9574 if (btrfs_chunk_length(path.nodes[0], chunk) !=
9577 "block group[%llu %llu] related chunk item length does not match",
9578 bg_key.objectid, bg_key.offset);
9579 err |= REFERENCER_MISMATCH;
9582 btrfs_release_path(&path);
9584 /* Search from the block group bytenr */
9585 extent_key.objectid = bg_key.objectid;
9586 extent_key.type = 0;
9587 extent_key.offset = 0;
9589 btrfs_init_path(&path);
9590 ret = btrfs_search_slot(NULL, extent_root, &extent_key, &path, 0, 0);
9594 /* Iterate extent tree to account used space */
9596 leaf = path.nodes[0];
9597 btrfs_item_key_to_cpu(leaf, &extent_key, path.slots[0]);
9598 if (extent_key.objectid >= bg_key.objectid + bg_key.offset)
9601 if (extent_key.type != BTRFS_METADATA_ITEM_KEY &&
9602 extent_key.type != BTRFS_EXTENT_ITEM_KEY)
9604 if (extent_key.objectid < bg_key.objectid)
9607 if (extent_key.type == BTRFS_METADATA_ITEM_KEY)
9610 total += extent_key.offset;
9612 ei = btrfs_item_ptr(leaf, path.slots[0],
9613 struct btrfs_extent_item);
9614 flags = btrfs_extent_flags(leaf, ei);
9615 if (flags & BTRFS_EXTENT_FLAG_DATA) {
9616 if (!(bg_flags & BTRFS_BLOCK_GROUP_DATA)) {
9618 "bad extent[%llu, %llu) type mismatch with chunk",
9619 extent_key.objectid,
9620 extent_key.objectid + extent_key.offset);
9621 err |= CHUNK_TYPE_MISMATCH;
9623 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
9624 if (!(bg_flags & (BTRFS_BLOCK_GROUP_SYSTEM |
9625 BTRFS_BLOCK_GROUP_METADATA))) {
9627 "bad extent[%llu, %llu) type mismatch with chunk",
9628 extent_key.objectid,
9629 extent_key.objectid + nodesize);
9630 err |= CHUNK_TYPE_MISMATCH;
9634 ret = btrfs_next_item(extent_root, &path);
9640 btrfs_release_path(&path);
9642 if (total != used) {
9644 "block group[%llu %llu] used %llu but extent items used %llu",
9645 bg_key.objectid, bg_key.offset, used, total);
9646 err |= ACCOUNTING_MISMATCH;
9652 * Check a chunk item.
9653 * Including checking all referred dev_extents and block group
9655 static int check_chunk_item(struct btrfs_fs_info *fs_info,
9656 struct extent_buffer *eb, int slot)
9658 struct btrfs_root *extent_root = fs_info->extent_root;
9659 struct btrfs_root *dev_root = fs_info->dev_root;
9660 struct btrfs_path path;
9661 struct btrfs_key chunk_key;
9662 struct btrfs_key bg_key;
9663 struct btrfs_key devext_key;
9664 struct btrfs_chunk *chunk;
9665 struct extent_buffer *leaf;
9666 struct btrfs_block_group_item *bi;
9667 struct btrfs_block_group_item bg_item;
9668 struct btrfs_dev_extent *ptr;
9669 u32 sectorsize = btrfs_super_sectorsize(fs_info->super_copy);
9681 btrfs_item_key_to_cpu(eb, &chunk_key, slot);
9682 chunk = btrfs_item_ptr(eb, slot, struct btrfs_chunk);
9683 length = btrfs_chunk_length(eb, chunk);
9684 chunk_end = chunk_key.offset + length;
9685 if (!IS_ALIGNED(length, sectorsize)) {
9686 error("chunk[%llu %llu) not aligned to %u",
9687 chunk_key.offset, chunk_end, sectorsize);
9688 err |= BYTES_UNALIGNED;
9692 type = btrfs_chunk_type(eb, chunk);
9693 profile = type & BTRFS_BLOCK_GROUP_PROFILE_MASK;
9694 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
9695 error("chunk[%llu %llu) has no chunk type",
9696 chunk_key.offset, chunk_end);
9697 err |= UNKNOWN_TYPE;
9699 if (profile && (profile & (profile - 1))) {
9700 error("chunk[%llu %llu) multiple profiles detected: %llx",
9701 chunk_key.offset, chunk_end, profile);
9702 err |= UNKNOWN_TYPE;
9705 bg_key.objectid = chunk_key.offset;
9706 bg_key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9707 bg_key.offset = length;
9709 btrfs_init_path(&path);
9710 ret = btrfs_search_slot(NULL, extent_root, &bg_key, &path, 0, 0);
9713 "chunk[%llu %llu) did not find the related block group item",
9714 chunk_key.offset, chunk_end);
9715 err |= REFERENCER_MISSING;
9717 leaf = path.nodes[0];
9718 bi = btrfs_item_ptr(leaf, path.slots[0],
9719 struct btrfs_block_group_item);
9720 read_extent_buffer(leaf, &bg_item, (unsigned long)bi,
9722 if (btrfs_block_group_flags(&bg_item) != type) {
9724 "chunk[%llu %llu) related block group item flags mismatch, wanted: %llu, have: %llu",
9725 chunk_key.offset, chunk_end, type,
9726 btrfs_block_group_flags(&bg_item));
9727 err |= REFERENCER_MISSING;
9731 num_stripes = btrfs_chunk_num_stripes(eb, chunk);
9732 for (i = 0; i < num_stripes; i++) {
9733 btrfs_release_path(&path);
9734 btrfs_init_path(&path);
9735 devext_key.objectid = btrfs_stripe_devid_nr(eb, chunk, i);
9736 devext_key.type = BTRFS_DEV_EXTENT_KEY;
9737 devext_key.offset = btrfs_stripe_offset_nr(eb, chunk, i);
9739 ret = btrfs_search_slot(NULL, dev_root, &devext_key, &path,
9744 leaf = path.nodes[0];
9745 ptr = btrfs_item_ptr(leaf, path.slots[0],
9746 struct btrfs_dev_extent);
9747 objectid = btrfs_dev_extent_chunk_objectid(leaf, ptr);
9748 offset = btrfs_dev_extent_chunk_offset(leaf, ptr);
9749 if (objectid != chunk_key.objectid ||
9750 offset != chunk_key.offset ||
9751 btrfs_dev_extent_length(leaf, ptr) != length)
9755 err |= BACKREF_MISSING;
9757 "chunk[%llu %llu) stripe %d did not find the related dev extent",
9758 chunk_key.objectid, chunk_end, i);
9761 btrfs_release_path(&path);
9767 * Main entry function to check known items and update related accounting info
9769 static int check_leaf_items(struct btrfs_root *root, struct extent_buffer *eb)
9771 struct btrfs_fs_info *fs_info = root->fs_info;
9772 struct btrfs_key key;
9775 struct btrfs_extent_data_ref *dref;
9780 btrfs_item_key_to_cpu(eb, &key, slot);
9784 case BTRFS_EXTENT_DATA_KEY:
9785 ret = check_extent_data_item(root, eb, slot);
9788 case BTRFS_BLOCK_GROUP_ITEM_KEY:
9789 ret = check_block_group_item(fs_info, eb, slot);
9792 case BTRFS_DEV_ITEM_KEY:
9793 ret = check_dev_item(fs_info, eb, slot);
9796 case BTRFS_CHUNK_ITEM_KEY:
9797 ret = check_chunk_item(fs_info, eb, slot);
9800 case BTRFS_DEV_EXTENT_KEY:
9801 ret = check_dev_extent_item(fs_info, eb, slot);
9804 case BTRFS_EXTENT_ITEM_KEY:
9805 case BTRFS_METADATA_ITEM_KEY:
9806 ret = check_extent_item(fs_info, eb, slot);
9809 case BTRFS_EXTENT_CSUM_KEY:
9810 total_csum_bytes += btrfs_item_size_nr(eb, slot);
9812 case BTRFS_TREE_BLOCK_REF_KEY:
9813 ret = check_tree_block_backref(fs_info, key.offset,
9817 case BTRFS_EXTENT_DATA_REF_KEY:
9818 dref = btrfs_item_ptr(eb, slot, struct btrfs_extent_data_ref);
9819 ret = check_extent_data_backref(fs_info,
9820 btrfs_extent_data_ref_root(eb, dref),
9821 btrfs_extent_data_ref_objectid(eb, dref),
9822 btrfs_extent_data_ref_offset(eb, dref),
9824 btrfs_extent_data_ref_count(eb, dref));
9827 case BTRFS_SHARED_BLOCK_REF_KEY:
9828 ret = check_shared_block_backref(fs_info, key.offset,
9832 case BTRFS_SHARED_DATA_REF_KEY:
9833 ret = check_shared_data_backref(fs_info, key.offset,
9841 if (++slot < btrfs_header_nritems(eb))
9848 * Helper function for later fs/subvol tree check. To determine if a tree
9849 * block should be checked.
9850 * This function will ensure only the direct referencer with lowest rootid to
9851 * check a fs/subvolume tree block.
9853 * Backref check at extent tree would detect errors like missing subvolume
9854 * tree, so we can do aggressive check to reduce duplicated checks.
9856 static int should_check(struct btrfs_root *root, struct extent_buffer *eb)
9858 struct btrfs_root *extent_root = root->fs_info->extent_root;
9859 struct btrfs_key key;
9860 struct btrfs_path path;
9861 struct extent_buffer *leaf;
9863 struct btrfs_extent_item *ei;
9869 struct btrfs_extent_inline_ref *iref;
9872 btrfs_init_path(&path);
9873 key.objectid = btrfs_header_bytenr(eb);
9874 key.type = BTRFS_METADATA_ITEM_KEY;
9875 key.offset = (u64)-1;
9878 * Any failure in backref resolving means we can't determine
9879 * whom the tree block belongs to.
9880 * So in that case, we need to check that tree block
9882 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
9886 ret = btrfs_previous_extent_item(extent_root, &path,
9887 btrfs_header_bytenr(eb));
9891 leaf = path.nodes[0];
9892 slot = path.slots[0];
9893 btrfs_item_key_to_cpu(leaf, &key, slot);
9894 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
9896 if (key.type == BTRFS_METADATA_ITEM_KEY) {
9897 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
9899 struct btrfs_tree_block_info *info;
9901 info = (struct btrfs_tree_block_info *)(ei + 1);
9902 iref = (struct btrfs_extent_inline_ref *)(info + 1);
9905 item_size = btrfs_item_size_nr(leaf, slot);
9906 ptr = (unsigned long)iref;
9907 end = (unsigned long)ei + item_size;
9909 iref = (struct btrfs_extent_inline_ref *)ptr;
9910 type = btrfs_extent_inline_ref_type(leaf, iref);
9911 offset = btrfs_extent_inline_ref_offset(leaf, iref);
9914 * We only check the tree block if current root is
9915 * the lowest referencer of it.
9917 if (type == BTRFS_TREE_BLOCK_REF_KEY &&
9918 offset < root->objectid) {
9919 btrfs_release_path(&path);
9923 ptr += btrfs_extent_inline_ref_size(type);
9926 * Normally we should also check keyed tree block ref, but that may be
9927 * very time consuming. Inlined ref should already make us skip a lot
9928 * of refs now. So skip search keyed tree block ref.
9932 btrfs_release_path(&path);
9937 * Traversal function for tree block. We will do:
9938 * 1) Skip shared fs/subvolume tree blocks
9939 * 2) Update related bytes accounting
9940 * 3) Pre-order traversal
9942 static int traverse_tree_block(struct btrfs_root *root,
9943 struct extent_buffer *node)
9945 struct extent_buffer *eb;
9946 struct btrfs_key key;
9947 struct btrfs_key drop_key;
9955 * Skip shared fs/subvolume tree block, in that case they will
9956 * be checked by referencer with lowest rootid
9958 if (is_fstree(root->objectid) && !should_check(root, node))
9961 /* Update bytes accounting */
9962 total_btree_bytes += node->len;
9963 if (fs_root_objectid(btrfs_header_owner(node)))
9964 total_fs_tree_bytes += node->len;
9965 if (btrfs_header_owner(node) == BTRFS_EXTENT_TREE_OBJECTID)
9966 total_extent_tree_bytes += node->len;
9967 if (!found_old_backref &&
9968 btrfs_header_owner(node) == BTRFS_TREE_RELOC_OBJECTID &&
9969 btrfs_header_backref_rev(node) == BTRFS_MIXED_BACKREF_REV &&
9970 !btrfs_header_flag(node, BTRFS_HEADER_FLAG_RELOC))
9971 found_old_backref = 1;
9973 /* pre-order tranversal, check itself first */
9974 level = btrfs_header_level(node);
9975 ret = check_tree_block_ref(root, node, btrfs_header_bytenr(node),
9976 btrfs_header_level(node),
9977 btrfs_header_owner(node));
9981 "check %s failed root %llu bytenr %llu level %d, force continue check",
9982 level ? "node":"leaf", root->objectid,
9983 btrfs_header_bytenr(node), btrfs_header_level(node));
9986 btree_space_waste += btrfs_leaf_free_space(root, node);
9987 ret = check_leaf_items(root, node);
9992 nr = btrfs_header_nritems(node);
9993 btrfs_disk_key_to_cpu(&drop_key, &root->root_item.drop_progress);
9994 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) - nr) *
9995 sizeof(struct btrfs_key_ptr);
9997 /* Then check all its children */
9998 for (i = 0; i < nr; i++) {
9999 u64 blocknr = btrfs_node_blockptr(node, i);
10001 btrfs_node_key_to_cpu(node, &key, i);
10002 if (level == root->root_item.drop_level &&
10003 is_dropped_key(&key, &drop_key))
10007 * As a btrfs tree has most 8 levels (0..7), so it's quite safe
10008 * to call the function itself.
10010 eb = read_tree_block(root, blocknr, root->nodesize, 0);
10011 if (extent_buffer_uptodate(eb)) {
10012 ret = traverse_tree_block(root, eb);
10015 free_extent_buffer(eb);
10022 * Low memory usage version check_chunks_and_extents.
10024 static int check_chunks_and_extents_v2(struct btrfs_root *root)
10026 struct btrfs_path path;
10027 struct btrfs_key key;
10028 struct btrfs_root *root1;
10029 struct btrfs_root *cur_root;
10033 root1 = root->fs_info->chunk_root;
10034 ret = traverse_tree_block(root1, root1->node);
10037 root1 = root->fs_info->tree_root;
10038 ret = traverse_tree_block(root1, root1->node);
10041 btrfs_init_path(&path);
10042 key.objectid = BTRFS_EXTENT_TREE_OBJECTID;
10044 key.type = BTRFS_ROOT_ITEM_KEY;
10046 ret = btrfs_search_slot(NULL, root1, &key, &path, 0, 0);
10048 error("cannot find extent treet in tree_root");
10053 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
10054 if (key.type != BTRFS_ROOT_ITEM_KEY)
10056 key.offset = (u64)-1;
10058 cur_root = btrfs_read_fs_root(root->fs_info, &key);
10059 if (IS_ERR(cur_root) || !cur_root) {
10060 error("failed to read tree: %lld", key.objectid);
10064 ret = traverse_tree_block(cur_root, cur_root->node);
10068 ret = btrfs_next_item(root1, &path);
10074 btrfs_release_path(&path);
10078 static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
10079 struct btrfs_root *root, int overwrite)
10081 struct extent_buffer *c;
10082 struct extent_buffer *old = root->node;
10085 struct btrfs_disk_key disk_key = {0,0,0};
10091 extent_buffer_get(c);
10094 c = btrfs_alloc_free_block(trans, root,
10096 root->root_key.objectid,
10097 &disk_key, level, 0, 0);
10100 extent_buffer_get(c);
10104 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
10105 btrfs_set_header_level(c, level);
10106 btrfs_set_header_bytenr(c, c->start);
10107 btrfs_set_header_generation(c, trans->transid);
10108 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
10109 btrfs_set_header_owner(c, root->root_key.objectid);
10111 write_extent_buffer(c, root->fs_info->fsid,
10112 btrfs_header_fsid(), BTRFS_FSID_SIZE);
10114 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
10115 btrfs_header_chunk_tree_uuid(c),
10118 btrfs_mark_buffer_dirty(c);
10120 * this case can happen in the following case:
10122 * 1.overwrite previous root.
10124 * 2.reinit reloc data root, this is because we skip pin
10125 * down reloc data tree before which means we can allocate
10126 * same block bytenr here.
10128 if (old->start == c->start) {
10129 btrfs_set_root_generation(&root->root_item,
10131 root->root_item.level = btrfs_header_level(root->node);
10132 ret = btrfs_update_root(trans, root->fs_info->tree_root,
10133 &root->root_key, &root->root_item);
10135 free_extent_buffer(c);
10139 free_extent_buffer(old);
10141 add_root_to_dirty_list(root);
10145 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
10146 struct extent_buffer *eb, int tree_root)
10148 struct extent_buffer *tmp;
10149 struct btrfs_root_item *ri;
10150 struct btrfs_key key;
10153 int level = btrfs_header_level(eb);
10159 * If we have pinned this block before, don't pin it again.
10160 * This can not only avoid forever loop with broken filesystem
10161 * but also give us some speedups.
10163 if (test_range_bit(&fs_info->pinned_extents, eb->start,
10164 eb->start + eb->len - 1, EXTENT_DIRTY, 0))
10167 btrfs_pin_extent(fs_info, eb->start, eb->len);
10169 nodesize = btrfs_super_nodesize(fs_info->super_copy);
10170 nritems = btrfs_header_nritems(eb);
10171 for (i = 0; i < nritems; i++) {
10173 btrfs_item_key_to_cpu(eb, &key, i);
10174 if (key.type != BTRFS_ROOT_ITEM_KEY)
10176 /* Skip the extent root and reloc roots */
10177 if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
10178 key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
10179 key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
10181 ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
10182 bytenr = btrfs_disk_root_bytenr(eb, ri);
10185 * If at any point we start needing the real root we
10186 * will have to build a stump root for the root we are
10187 * in, but for now this doesn't actually use the root so
10188 * just pass in extent_root.
10190 tmp = read_tree_block(fs_info->extent_root, bytenr,
10192 if (!extent_buffer_uptodate(tmp)) {
10193 fprintf(stderr, "Error reading root block\n");
10196 ret = pin_down_tree_blocks(fs_info, tmp, 0);
10197 free_extent_buffer(tmp);
10201 bytenr = btrfs_node_blockptr(eb, i);
10203 /* If we aren't the tree root don't read the block */
10204 if (level == 1 && !tree_root) {
10205 btrfs_pin_extent(fs_info, bytenr, nodesize);
10209 tmp = read_tree_block(fs_info->extent_root, bytenr,
10211 if (!extent_buffer_uptodate(tmp)) {
10212 fprintf(stderr, "Error reading tree block\n");
10215 ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
10216 free_extent_buffer(tmp);
10225 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
10229 ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
10233 return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
10236 static int reset_block_groups(struct btrfs_fs_info *fs_info)
10238 struct btrfs_block_group_cache *cache;
10239 struct btrfs_path *path;
10240 struct extent_buffer *leaf;
10241 struct btrfs_chunk *chunk;
10242 struct btrfs_key key;
10246 path = btrfs_alloc_path();
10251 key.type = BTRFS_CHUNK_ITEM_KEY;
10254 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
10256 btrfs_free_path(path);
10261 * We do this in case the block groups were screwed up and had alloc
10262 * bits that aren't actually set on the chunks. This happens with
10263 * restored images every time and could happen in real life I guess.
10265 fs_info->avail_data_alloc_bits = 0;
10266 fs_info->avail_metadata_alloc_bits = 0;
10267 fs_info->avail_system_alloc_bits = 0;
10269 /* First we need to create the in-memory block groups */
10271 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
10272 ret = btrfs_next_leaf(fs_info->chunk_root, path);
10274 btrfs_free_path(path);
10282 leaf = path->nodes[0];
10283 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
10284 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
10289 chunk = btrfs_item_ptr(leaf, path->slots[0],
10290 struct btrfs_chunk);
10291 btrfs_add_block_group(fs_info, 0,
10292 btrfs_chunk_type(leaf, chunk),
10293 key.objectid, key.offset,
10294 btrfs_chunk_length(leaf, chunk));
10295 set_extent_dirty(&fs_info->free_space_cache, key.offset,
10296 key.offset + btrfs_chunk_length(leaf, chunk),
10302 cache = btrfs_lookup_first_block_group(fs_info, start);
10306 start = cache->key.objectid + cache->key.offset;
10309 btrfs_free_path(path);
10313 static int reset_balance(struct btrfs_trans_handle *trans,
10314 struct btrfs_fs_info *fs_info)
10316 struct btrfs_root *root = fs_info->tree_root;
10317 struct btrfs_path *path;
10318 struct extent_buffer *leaf;
10319 struct btrfs_key key;
10320 int del_slot, del_nr = 0;
10324 path = btrfs_alloc_path();
10328 key.objectid = BTRFS_BALANCE_OBJECTID;
10329 key.type = BTRFS_BALANCE_ITEM_KEY;
10332 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10337 goto reinit_data_reloc;
10342 ret = btrfs_del_item(trans, root, path);
10345 btrfs_release_path(path);
10347 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
10348 key.type = BTRFS_ROOT_ITEM_KEY;
10351 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10355 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
10360 ret = btrfs_del_items(trans, root, path,
10367 btrfs_release_path(path);
10370 ret = btrfs_search_slot(trans, root, &key, path,
10377 leaf = path->nodes[0];
10378 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
10379 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
10381 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
10386 del_slot = path->slots[0];
10395 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
10399 btrfs_release_path(path);
10402 key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
10403 key.type = BTRFS_ROOT_ITEM_KEY;
10404 key.offset = (u64)-1;
10405 root = btrfs_read_fs_root(fs_info, &key);
10406 if (IS_ERR(root)) {
10407 fprintf(stderr, "Error reading data reloc tree\n");
10408 ret = PTR_ERR(root);
10411 record_root_in_trans(trans, root);
10412 ret = btrfs_fsck_reinit_root(trans, root, 0);
10415 ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
10417 btrfs_free_path(path);
10421 static int reinit_extent_tree(struct btrfs_trans_handle *trans,
10422 struct btrfs_fs_info *fs_info)
10428 * The only reason we don't do this is because right now we're just
10429 * walking the trees we find and pinning down their bytes, we don't look
10430 * at any of the leaves. In order to do mixed groups we'd have to check
10431 * the leaves of any fs roots and pin down the bytes for any file
10432 * extents we find. Not hard but why do it if we don't have to?
10434 if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
10435 fprintf(stderr, "We don't support re-initing the extent tree "
10436 "for mixed block groups yet, please notify a btrfs "
10437 "developer you want to do this so they can add this "
10438 "functionality.\n");
10443 * first we need to walk all of the trees except the extent tree and pin
10444 * down the bytes that are in use so we don't overwrite any existing
10447 ret = pin_metadata_blocks(fs_info);
10449 fprintf(stderr, "error pinning down used bytes\n");
10454 * Need to drop all the block groups since we're going to recreate all
10457 btrfs_free_block_groups(fs_info);
10458 ret = reset_block_groups(fs_info);
10460 fprintf(stderr, "error resetting the block groups\n");
10464 /* Ok we can allocate now, reinit the extent root */
10465 ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 0);
10467 fprintf(stderr, "extent root initialization failed\n");
10469 * When the transaction code is updated we should end the
10470 * transaction, but for now progs only knows about commit so
10471 * just return an error.
10477 * Now we have all the in-memory block groups setup so we can make
10478 * allocations properly, and the metadata we care about is safe since we
10479 * pinned all of it above.
10482 struct btrfs_block_group_cache *cache;
10484 cache = btrfs_lookup_first_block_group(fs_info, start);
10487 start = cache->key.objectid + cache->key.offset;
10488 ret = btrfs_insert_item(trans, fs_info->extent_root,
10489 &cache->key, &cache->item,
10490 sizeof(cache->item));
10492 fprintf(stderr, "Error adding block group\n");
10495 btrfs_extent_post_op(trans, fs_info->extent_root);
10498 ret = reset_balance(trans, fs_info);
10500 fprintf(stderr, "error resetting the pending balance\n");
10505 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
10507 struct btrfs_path *path;
10508 struct btrfs_trans_handle *trans;
10509 struct btrfs_key key;
10512 printf("Recowing metadata block %llu\n", eb->start);
10513 key.objectid = btrfs_header_owner(eb);
10514 key.type = BTRFS_ROOT_ITEM_KEY;
10515 key.offset = (u64)-1;
10517 root = btrfs_read_fs_root(root->fs_info, &key);
10518 if (IS_ERR(root)) {
10519 fprintf(stderr, "Couldn't find owner root %llu\n",
10521 return PTR_ERR(root);
10524 path = btrfs_alloc_path();
10528 trans = btrfs_start_transaction(root, 1);
10529 if (IS_ERR(trans)) {
10530 btrfs_free_path(path);
10531 return PTR_ERR(trans);
10534 path->lowest_level = btrfs_header_level(eb);
10535 if (path->lowest_level)
10536 btrfs_node_key_to_cpu(eb, &key, 0);
10538 btrfs_item_key_to_cpu(eb, &key, 0);
10540 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
10541 btrfs_commit_transaction(trans, root);
10542 btrfs_free_path(path);
10546 static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
10548 struct btrfs_path *path;
10549 struct btrfs_trans_handle *trans;
10550 struct btrfs_key key;
10553 printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
10554 bad->key.type, bad->key.offset);
10555 key.objectid = bad->root_id;
10556 key.type = BTRFS_ROOT_ITEM_KEY;
10557 key.offset = (u64)-1;
10559 root = btrfs_read_fs_root(root->fs_info, &key);
10560 if (IS_ERR(root)) {
10561 fprintf(stderr, "Couldn't find owner root %llu\n",
10563 return PTR_ERR(root);
10566 path = btrfs_alloc_path();
10570 trans = btrfs_start_transaction(root, 1);
10571 if (IS_ERR(trans)) {
10572 btrfs_free_path(path);
10573 return PTR_ERR(trans);
10576 ret = btrfs_search_slot(trans, root, &bad->key, path, -1, 1);
10582 ret = btrfs_del_item(trans, root, path);
10584 btrfs_commit_transaction(trans, root);
10585 btrfs_free_path(path);
10589 static int zero_log_tree(struct btrfs_root *root)
10591 struct btrfs_trans_handle *trans;
10594 trans = btrfs_start_transaction(root, 1);
10595 if (IS_ERR(trans)) {
10596 ret = PTR_ERR(trans);
10599 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
10600 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
10601 ret = btrfs_commit_transaction(trans, root);
10605 static int populate_csum(struct btrfs_trans_handle *trans,
10606 struct btrfs_root *csum_root, char *buf, u64 start,
10613 while (offset < len) {
10614 sectorsize = csum_root->sectorsize;
10615 ret = read_extent_data(csum_root, buf, start + offset,
10619 ret = btrfs_csum_file_block(trans, csum_root, start + len,
10620 start + offset, buf, sectorsize);
10623 offset += sectorsize;
10628 static int fill_csum_tree_from_one_fs_root(struct btrfs_trans_handle *trans,
10629 struct btrfs_root *csum_root,
10630 struct btrfs_root *cur_root)
10632 struct btrfs_path *path;
10633 struct btrfs_key key;
10634 struct extent_buffer *node;
10635 struct btrfs_file_extent_item *fi;
10642 path = btrfs_alloc_path();
10645 buf = malloc(cur_root->fs_info->csum_root->sectorsize);
10655 ret = btrfs_search_slot(NULL, cur_root, &key, path, 0, 0);
10658 /* Iterate all regular file extents and fill its csum */
10660 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
10662 if (key.type != BTRFS_EXTENT_DATA_KEY)
10664 node = path->nodes[0];
10665 slot = path->slots[0];
10666 fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
10667 if (btrfs_file_extent_type(node, fi) != BTRFS_FILE_EXTENT_REG)
10669 start = btrfs_file_extent_disk_bytenr(node, fi);
10670 len = btrfs_file_extent_disk_num_bytes(node, fi);
10672 ret = populate_csum(trans, csum_root, buf, start, len);
10673 if (ret == -EEXIST)
10679 * TODO: if next leaf is corrupted, jump to nearest next valid
10682 ret = btrfs_next_item(cur_root, path);
10692 btrfs_free_path(path);
10697 static int fill_csum_tree_from_fs(struct btrfs_trans_handle *trans,
10698 struct btrfs_root *csum_root)
10700 struct btrfs_fs_info *fs_info = csum_root->fs_info;
10701 struct btrfs_path *path;
10702 struct btrfs_root *tree_root = fs_info->tree_root;
10703 struct btrfs_root *cur_root;
10704 struct extent_buffer *node;
10705 struct btrfs_key key;
10709 path = btrfs_alloc_path();
10713 key.objectid = BTRFS_FS_TREE_OBJECTID;
10715 key.type = BTRFS_ROOT_ITEM_KEY;
10717 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
10726 node = path->nodes[0];
10727 slot = path->slots[0];
10728 btrfs_item_key_to_cpu(node, &key, slot);
10729 if (key.objectid > BTRFS_LAST_FREE_OBJECTID)
10731 if (key.type != BTRFS_ROOT_ITEM_KEY)
10733 if (!is_fstree(key.objectid))
10735 key.offset = (u64)-1;
10737 cur_root = btrfs_read_fs_root(fs_info, &key);
10738 if (IS_ERR(cur_root) || !cur_root) {
10739 fprintf(stderr, "Fail to read fs/subvol tree: %lld\n",
10743 ret = fill_csum_tree_from_one_fs_root(trans, csum_root,
10748 ret = btrfs_next_item(tree_root, path);
10758 btrfs_free_path(path);
10762 static int fill_csum_tree_from_extent(struct btrfs_trans_handle *trans,
10763 struct btrfs_root *csum_root)
10765 struct btrfs_root *extent_root = csum_root->fs_info->extent_root;
10766 struct btrfs_path *path;
10767 struct btrfs_extent_item *ei;
10768 struct extent_buffer *leaf;
10770 struct btrfs_key key;
10773 path = btrfs_alloc_path();
10778 key.type = BTRFS_EXTENT_ITEM_KEY;
10781 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
10783 btrfs_free_path(path);
10787 buf = malloc(csum_root->sectorsize);
10789 btrfs_free_path(path);
10794 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
10795 ret = btrfs_next_leaf(extent_root, path);
10803 leaf = path->nodes[0];
10805 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
10806 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
10811 ei = btrfs_item_ptr(leaf, path->slots[0],
10812 struct btrfs_extent_item);
10813 if (!(btrfs_extent_flags(leaf, ei) &
10814 BTRFS_EXTENT_FLAG_DATA)) {
10819 ret = populate_csum(trans, csum_root, buf, key.objectid,
10826 btrfs_free_path(path);
10832 * Recalculate the csum and put it into the csum tree.
10834 * Extent tree init will wipe out all the extent info, so in that case, we
10835 * can't depend on extent tree, but use fs tree. If search_fs_tree is set, we
10836 * will use fs/subvol trees to init the csum tree.
10838 static int fill_csum_tree(struct btrfs_trans_handle *trans,
10839 struct btrfs_root *csum_root,
10840 int search_fs_tree)
10842 if (search_fs_tree)
10843 return fill_csum_tree_from_fs(trans, csum_root);
10845 return fill_csum_tree_from_extent(trans, csum_root);
10848 static void free_roots_info_cache(void)
10850 if (!roots_info_cache)
10853 while (!cache_tree_empty(roots_info_cache)) {
10854 struct cache_extent *entry;
10855 struct root_item_info *rii;
10857 entry = first_cache_extent(roots_info_cache);
10860 remove_cache_extent(roots_info_cache, entry);
10861 rii = container_of(entry, struct root_item_info, cache_extent);
10865 free(roots_info_cache);
10866 roots_info_cache = NULL;
10869 static int build_roots_info_cache(struct btrfs_fs_info *info)
10872 struct btrfs_key key;
10873 struct extent_buffer *leaf;
10874 struct btrfs_path *path;
10876 if (!roots_info_cache) {
10877 roots_info_cache = malloc(sizeof(*roots_info_cache));
10878 if (!roots_info_cache)
10880 cache_tree_init(roots_info_cache);
10883 path = btrfs_alloc_path();
10888 key.type = BTRFS_EXTENT_ITEM_KEY;
10891 ret = btrfs_search_slot(NULL, info->extent_root, &key, path, 0, 0);
10894 leaf = path->nodes[0];
10897 struct btrfs_key found_key;
10898 struct btrfs_extent_item *ei;
10899 struct btrfs_extent_inline_ref *iref;
10900 int slot = path->slots[0];
10905 struct cache_extent *entry;
10906 struct root_item_info *rii;
10908 if (slot >= btrfs_header_nritems(leaf)) {
10909 ret = btrfs_next_leaf(info->extent_root, path);
10916 leaf = path->nodes[0];
10917 slot = path->slots[0];
10920 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10922 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
10923 found_key.type != BTRFS_METADATA_ITEM_KEY)
10926 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
10927 flags = btrfs_extent_flags(leaf, ei);
10929 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
10930 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
10933 if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
10934 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
10935 level = found_key.offset;
10937 struct btrfs_tree_block_info *binfo;
10939 binfo = (struct btrfs_tree_block_info *)(ei + 1);
10940 iref = (struct btrfs_extent_inline_ref *)(binfo + 1);
10941 level = btrfs_tree_block_level(leaf, binfo);
10945 * For a root extent, it must be of the following type and the
10946 * first (and only one) iref in the item.
10948 type = btrfs_extent_inline_ref_type(leaf, iref);
10949 if (type != BTRFS_TREE_BLOCK_REF_KEY)
10952 root_id = btrfs_extent_inline_ref_offset(leaf, iref);
10953 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
10955 rii = malloc(sizeof(struct root_item_info));
10960 rii->cache_extent.start = root_id;
10961 rii->cache_extent.size = 1;
10962 rii->level = (u8)-1;
10963 entry = &rii->cache_extent;
10964 ret = insert_cache_extent(roots_info_cache, entry);
10967 rii = container_of(entry, struct root_item_info,
10971 ASSERT(rii->cache_extent.start == root_id);
10972 ASSERT(rii->cache_extent.size == 1);
10974 if (level > rii->level || rii->level == (u8)-1) {
10975 rii->level = level;
10976 rii->bytenr = found_key.objectid;
10977 rii->gen = btrfs_extent_generation(leaf, ei);
10978 rii->node_count = 1;
10979 } else if (level == rii->level) {
10987 btrfs_free_path(path);
10992 static int maybe_repair_root_item(struct btrfs_fs_info *info,
10993 struct btrfs_path *path,
10994 const struct btrfs_key *root_key,
10995 const int read_only_mode)
10997 const u64 root_id = root_key->objectid;
10998 struct cache_extent *entry;
10999 struct root_item_info *rii;
11000 struct btrfs_root_item ri;
11001 unsigned long offset;
11003 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
11006 "Error: could not find extent items for root %llu\n",
11007 root_key->objectid);
11011 rii = container_of(entry, struct root_item_info, cache_extent);
11012 ASSERT(rii->cache_extent.start == root_id);
11013 ASSERT(rii->cache_extent.size == 1);
11015 if (rii->node_count != 1) {
11017 "Error: could not find btree root extent for root %llu\n",
11022 offset = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
11023 read_extent_buffer(path->nodes[0], &ri, offset, sizeof(ri));
11025 if (btrfs_root_bytenr(&ri) != rii->bytenr ||
11026 btrfs_root_level(&ri) != rii->level ||
11027 btrfs_root_generation(&ri) != rii->gen) {
11030 * If we're in repair mode but our caller told us to not update
11031 * the root item, i.e. just check if it needs to be updated, don't
11032 * print this message, since the caller will call us again shortly
11033 * for the same root item without read only mode (the caller will
11034 * open a transaction first).
11036 if (!(read_only_mode && repair))
11038 "%sroot item for root %llu,"
11039 " current bytenr %llu, current gen %llu, current level %u,"
11040 " new bytenr %llu, new gen %llu, new level %u\n",
11041 (read_only_mode ? "" : "fixing "),
11043 btrfs_root_bytenr(&ri), btrfs_root_generation(&ri),
11044 btrfs_root_level(&ri),
11045 rii->bytenr, rii->gen, rii->level);
11047 if (btrfs_root_generation(&ri) > rii->gen) {
11049 "root %llu has a root item with a more recent gen (%llu) compared to the found root node (%llu)\n",
11050 root_id, btrfs_root_generation(&ri), rii->gen);
11054 if (!read_only_mode) {
11055 btrfs_set_root_bytenr(&ri, rii->bytenr);
11056 btrfs_set_root_level(&ri, rii->level);
11057 btrfs_set_root_generation(&ri, rii->gen);
11058 write_extent_buffer(path->nodes[0], &ri,
11059 offset, sizeof(ri));
11069 * A regression introduced in the 3.17 kernel (more specifically in 3.17-rc2),
11070 * caused read-only snapshots to be corrupted if they were created at a moment
11071 * when the source subvolume/snapshot had orphan items. The issue was that the
11072 * on-disk root items became incorrect, referring to the pre orphan cleanup root
11073 * node instead of the post orphan cleanup root node.
11074 * So this function, and its callees, just detects and fixes those cases. Even
11075 * though the regression was for read-only snapshots, this function applies to
11076 * any snapshot/subvolume root.
11077 * This must be run before any other repair code - not doing it so, makes other
11078 * repair code delete or modify backrefs in the extent tree for example, which
11079 * will result in an inconsistent fs after repairing the root items.
11081 static int repair_root_items(struct btrfs_fs_info *info)
11083 struct btrfs_path *path = NULL;
11084 struct btrfs_key key;
11085 struct extent_buffer *leaf;
11086 struct btrfs_trans_handle *trans = NULL;
11089 int need_trans = 0;
11091 ret = build_roots_info_cache(info);
11095 path = btrfs_alloc_path();
11101 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
11102 key.type = BTRFS_ROOT_ITEM_KEY;
11107 * Avoid opening and committing transactions if a leaf doesn't have
11108 * any root items that need to be fixed, so that we avoid rotating
11109 * backup roots unnecessarily.
11112 trans = btrfs_start_transaction(info->tree_root, 1);
11113 if (IS_ERR(trans)) {
11114 ret = PTR_ERR(trans);
11119 ret = btrfs_search_slot(trans, info->tree_root, &key, path,
11123 leaf = path->nodes[0];
11126 struct btrfs_key found_key;
11128 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
11129 int no_more_keys = find_next_key(path, &key);
11131 btrfs_release_path(path);
11133 ret = btrfs_commit_transaction(trans,
11145 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
11147 if (found_key.type != BTRFS_ROOT_ITEM_KEY)
11149 if (found_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
11152 ret = maybe_repair_root_item(info, path, &found_key,
11157 if (!trans && repair) {
11160 btrfs_release_path(path);
11170 free_roots_info_cache();
11171 btrfs_free_path(path);
11173 btrfs_commit_transaction(trans, info->tree_root);
11180 static int clear_free_space_cache(struct btrfs_fs_info *fs_info)
11182 struct btrfs_trans_handle *trans;
11183 struct btrfs_block_group_cache *bg_cache;
11187 /* Clear all free space cache inodes and its extent data */
11189 bg_cache = btrfs_lookup_first_block_group(fs_info, current);
11192 ret = btrfs_clear_free_space_cache(fs_info, bg_cache);
11195 current = bg_cache->key.objectid + bg_cache->key.offset;
11198 /* Don't forget to set cache_generation to -1 */
11199 trans = btrfs_start_transaction(fs_info->tree_root, 0);
11200 if (IS_ERR(trans)) {
11201 error("failed to update super block cache generation");
11202 return PTR_ERR(trans);
11204 btrfs_set_super_cache_generation(fs_info->super_copy, (u64)-1);
11205 btrfs_commit_transaction(trans, fs_info->tree_root);
11210 const char * const cmd_check_usage[] = {
11211 "btrfs check [options] <device>",
11212 "Check structural integrity of a filesystem (unmounted).",
11213 "Check structural integrity of an unmounted filesystem. Verify internal",
11214 "trees' consistency and item connectivity. In the repair mode try to",
11215 "fix the problems found. ",
11216 "WARNING: the repair mode is considered dangerous",
11218 "-s|--super <superblock> use this superblock copy",
11219 "-b|--backup use the first valid backup root copy",
11220 "--repair try to repair the filesystem",
11221 "--readonly run in read-only mode (default)",
11222 "--init-csum-tree create a new CRC tree",
11223 "--init-extent-tree create a new extent tree",
11224 "--mode <MODE> allows choice of memory/IO trade-offs",
11225 " where MODE is one of:",
11226 " original - read inodes and extents to memory (requires",
11227 " more memory, does less IO)",
11228 " lowmem - try to use less memory but read blocks again",
11230 "--check-data-csum verify checksums of data blocks",
11231 "-Q|--qgroup-report print a report on qgroup consistency",
11232 "-E|--subvol-extents <subvolid>",
11233 " print subvolume extents and sharing state",
11234 "-r|--tree-root <bytenr> use the given bytenr for the tree root",
11235 "--chunk-root <bytenr> use the given bytenr for the chunk tree root",
11236 "-p|--progress indicate progress",
11237 "--clear-space-cache v1|v2 clear space cache for v1 or v2",
11238 " NOTE: v1 support implemented",
11242 int cmd_check(int argc, char **argv)
11244 struct cache_tree root_cache;
11245 struct btrfs_root *root;
11246 struct btrfs_fs_info *info;
11249 u64 tree_root_bytenr = 0;
11250 u64 chunk_root_bytenr = 0;
11251 char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
11254 int init_csum_tree = 0;
11256 int clear_space_cache = 0;
11257 int qgroup_report = 0;
11258 int qgroups_repaired = 0;
11259 unsigned ctree_flags = OPEN_CTREE_EXCLUSIVE;
11263 enum { GETOPT_VAL_REPAIR = 257, GETOPT_VAL_INIT_CSUM,
11264 GETOPT_VAL_INIT_EXTENT, GETOPT_VAL_CHECK_CSUM,
11265 GETOPT_VAL_READONLY, GETOPT_VAL_CHUNK_TREE,
11266 GETOPT_VAL_MODE, GETOPT_VAL_CLEAR_SPACE_CACHE };
11267 static const struct option long_options[] = {
11268 { "super", required_argument, NULL, 's' },
11269 { "repair", no_argument, NULL, GETOPT_VAL_REPAIR },
11270 { "readonly", no_argument, NULL, GETOPT_VAL_READONLY },
11271 { "init-csum-tree", no_argument, NULL,
11272 GETOPT_VAL_INIT_CSUM },
11273 { "init-extent-tree", no_argument, NULL,
11274 GETOPT_VAL_INIT_EXTENT },
11275 { "check-data-csum", no_argument, NULL,
11276 GETOPT_VAL_CHECK_CSUM },
11277 { "backup", no_argument, NULL, 'b' },
11278 { "subvol-extents", required_argument, NULL, 'E' },
11279 { "qgroup-report", no_argument, NULL, 'Q' },
11280 { "tree-root", required_argument, NULL, 'r' },
11281 { "chunk-root", required_argument, NULL,
11282 GETOPT_VAL_CHUNK_TREE },
11283 { "progress", no_argument, NULL, 'p' },
11284 { "mode", required_argument, NULL,
11286 { "clear-space-cache", required_argument, NULL,
11287 GETOPT_VAL_CLEAR_SPACE_CACHE},
11288 { NULL, 0, NULL, 0}
11291 c = getopt_long(argc, argv, "as:br:p", long_options, NULL);
11295 case 'a': /* ignored */ break;
11297 ctree_flags |= OPEN_CTREE_BACKUP_ROOT;
11300 num = arg_strtou64(optarg);
11301 if (num >= BTRFS_SUPER_MIRROR_MAX) {
11303 "super mirror should be less than %d",
11304 BTRFS_SUPER_MIRROR_MAX);
11307 bytenr = btrfs_sb_offset(((int)num));
11308 printf("using SB copy %llu, bytenr %llu\n", num,
11309 (unsigned long long)bytenr);
11315 subvolid = arg_strtou64(optarg);
11318 tree_root_bytenr = arg_strtou64(optarg);
11320 case GETOPT_VAL_CHUNK_TREE:
11321 chunk_root_bytenr = arg_strtou64(optarg);
11324 ctx.progress_enabled = true;
11328 usage(cmd_check_usage);
11329 case GETOPT_VAL_REPAIR:
11330 printf("enabling repair mode\n");
11332 ctree_flags |= OPEN_CTREE_WRITES;
11334 case GETOPT_VAL_READONLY:
11337 case GETOPT_VAL_INIT_CSUM:
11338 printf("Creating a new CRC tree\n");
11339 init_csum_tree = 1;
11341 ctree_flags |= OPEN_CTREE_WRITES;
11343 case GETOPT_VAL_INIT_EXTENT:
11344 init_extent_tree = 1;
11345 ctree_flags |= (OPEN_CTREE_WRITES |
11346 OPEN_CTREE_NO_BLOCK_GROUPS);
11349 case GETOPT_VAL_CHECK_CSUM:
11350 check_data_csum = 1;
11352 case GETOPT_VAL_MODE:
11353 check_mode = parse_check_mode(optarg);
11354 if (check_mode == CHECK_MODE_UNKNOWN) {
11355 error("unknown mode: %s", optarg);
11359 case GETOPT_VAL_CLEAR_SPACE_CACHE:
11360 if (strcmp(optarg, "v1") != 0) {
11362 "only v1 support implmented, unrecognized value %s",
11366 clear_space_cache = 1;
11367 ctree_flags |= OPEN_CTREE_WRITES;
11372 if (check_argc_exact(argc - optind, 1))
11373 usage(cmd_check_usage);
11375 if (ctx.progress_enabled) {
11376 ctx.tp = TASK_NOTHING;
11377 ctx.info = task_init(print_status_check, print_status_return, &ctx);
11380 /* This check is the only reason for --readonly to exist */
11381 if (readonly && repair) {
11382 error("repair options are not compatible with --readonly");
11387 * Not supported yet
11389 if (repair && check_mode == CHECK_MODE_LOWMEM) {
11390 error("low memory mode doesn't support repair yet");
11395 cache_tree_init(&root_cache);
11397 if((ret = check_mounted(argv[optind])) < 0) {
11398 error("could not check mount status: %s", strerror(-ret));
11401 error("%s is currently mounted, aborting", argv[optind]);
11406 /* only allow partial opening under repair mode */
11408 ctree_flags |= OPEN_CTREE_PARTIAL;
11410 info = open_ctree_fs_info(argv[optind], bytenr, tree_root_bytenr,
11411 chunk_root_bytenr, ctree_flags);
11413 error("cannot open file system");
11418 global_info = info;
11419 root = info->fs_root;
11420 if (clear_space_cache) {
11421 if (btrfs_fs_compat_ro(info,
11422 BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE)) {
11424 "free space cache v2 detected, clearing not implemented");
11428 printf("Clearing free space cache\n");
11429 ret = clear_free_space_cache(info);
11431 error("failed to clear free space cache");
11434 printf("Free space cache cleared\n");
11440 * repair mode will force us to commit transaction which
11441 * will make us fail to load log tree when mounting.
11443 if (repair && btrfs_super_log_root(info->super_copy)) {
11444 ret = ask_user("repair mode will force to clear out log tree, are you sure?");
11449 ret = zero_log_tree(root);
11451 error("failed to zero log tree: %d", ret);
11456 uuid_unparse(info->super_copy->fsid, uuidbuf);
11457 if (qgroup_report) {
11458 printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
11460 ret = qgroup_verify_all(info);
11466 printf("Print extent state for subvolume %llu on %s\nUUID: %s\n",
11467 subvolid, argv[optind], uuidbuf);
11468 ret = print_extent_state(info, subvolid);
11471 printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
11473 if (!extent_buffer_uptodate(info->tree_root->node) ||
11474 !extent_buffer_uptodate(info->dev_root->node) ||
11475 !extent_buffer_uptodate(info->chunk_root->node)) {
11476 error("critical roots corrupted, unable to check the filesystem");
11481 if (init_extent_tree || init_csum_tree) {
11482 struct btrfs_trans_handle *trans;
11484 trans = btrfs_start_transaction(info->extent_root, 0);
11485 if (IS_ERR(trans)) {
11486 error("error starting transaction");
11487 ret = PTR_ERR(trans);
11491 if (init_extent_tree) {
11492 printf("Creating a new extent tree\n");
11493 ret = reinit_extent_tree(trans, info);
11498 if (init_csum_tree) {
11499 printf("Reinitialize checksum tree\n");
11500 ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
11502 error("checksum tree initialization failed: %d",
11508 ret = fill_csum_tree(trans, info->csum_root,
11511 error("checksum tree refilling failed: %d", ret);
11516 * Ok now we commit and run the normal fsck, which will add
11517 * extent entries for all of the items it finds.
11519 ret = btrfs_commit_transaction(trans, info->extent_root);
11523 if (!extent_buffer_uptodate(info->extent_root->node)) {
11524 error("critical: extent_root, unable to check the filesystem");
11528 if (!extent_buffer_uptodate(info->csum_root->node)) {
11529 error("critical: csum_root, unable to check the filesystem");
11534 if (!ctx.progress_enabled)
11535 printf("checking extents");
11536 if (check_mode == CHECK_MODE_LOWMEM)
11537 ret = check_chunks_and_extents_v2(root);
11539 ret = check_chunks_and_extents(root);
11541 printf("Errors found in extent allocation tree or chunk allocation");
11543 ret = repair_root_items(info);
11547 fprintf(stderr, "Fixed %d roots.\n", ret);
11549 } else if (ret > 0) {
11551 "Found %d roots with an outdated root item.\n",
11554 "Please run a filesystem check with the option --repair to fix them.\n");
11559 if (!ctx.progress_enabled) {
11560 if (btrfs_fs_compat_ro(info, BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE))
11561 fprintf(stderr, "checking free space tree\n");
11563 fprintf(stderr, "checking free space cache\n");
11565 ret = check_space_cache(root);
11570 * We used to have to have these hole extents in between our real
11571 * extents so if we don't have this flag set we need to make sure there
11572 * are no gaps in the file extents for inodes, otherwise we can just
11573 * ignore it when this happens.
11575 no_holes = btrfs_fs_incompat(root->fs_info,
11576 BTRFS_FEATURE_INCOMPAT_NO_HOLES);
11577 if (!ctx.progress_enabled)
11578 fprintf(stderr, "checking fs roots\n");
11579 ret = check_fs_roots(root, &root_cache);
11583 fprintf(stderr, "checking csums\n");
11584 ret = check_csums(root);
11588 fprintf(stderr, "checking root refs\n");
11589 ret = check_root_refs(root, &root_cache);
11593 while (repair && !list_empty(&root->fs_info->recow_ebs)) {
11594 struct extent_buffer *eb;
11596 eb = list_first_entry(&root->fs_info->recow_ebs,
11597 struct extent_buffer, recow);
11598 list_del_init(&eb->recow);
11599 ret = recow_extent_buffer(root, eb);
11604 while (!list_empty(&delete_items)) {
11605 struct bad_item *bad;
11607 bad = list_first_entry(&delete_items, struct bad_item, list);
11608 list_del_init(&bad->list);
11610 ret = delete_bad_item(root, bad);
11614 if (info->quota_enabled) {
11616 fprintf(stderr, "checking quota groups\n");
11617 err = qgroup_verify_all(info);
11621 err = repair_qgroups(info, &qgroups_repaired);
11626 if (!list_empty(&root->fs_info->recow_ebs)) {
11627 error("transid errors in file system");
11631 /* Don't override original ret */
11632 if (!ret && qgroups_repaired)
11633 ret = qgroups_repaired;
11635 if (found_old_backref) { /*
11636 * there was a disk format change when mixed
11637 * backref was in testing tree. The old format
11638 * existed about one week.
11640 printf("\n * Found old mixed backref format. "
11641 "The old format is not supported! *"
11642 "\n * Please mount the FS in readonly mode, "
11643 "backup data and re-format the FS. *\n\n");
11646 printf("found %llu bytes used err is %d\n",
11647 (unsigned long long)bytes_used, ret);
11648 printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
11649 printf("total tree bytes: %llu\n",
11650 (unsigned long long)total_btree_bytes);
11651 printf("total fs tree bytes: %llu\n",
11652 (unsigned long long)total_fs_tree_bytes);
11653 printf("total extent tree bytes: %llu\n",
11654 (unsigned long long)total_extent_tree_bytes);
11655 printf("btree space waste bytes: %llu\n",
11656 (unsigned long long)btree_space_waste);
11657 printf("file data blocks allocated: %llu\n referenced %llu\n",
11658 (unsigned long long)data_bytes_allocated,
11659 (unsigned long long)data_bytes_referenced);
11661 free_qgroup_counts();
11662 free_root_recs_tree(&root_cache);
11666 if (ctx.progress_enabled)
11667 task_deinit(ctx.info);