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"
43 #include "kernel-shared/ulist.h"
51 TASK_NOTHING, /* have to be the last element */
56 enum task_position tp;
58 struct task_info *info;
61 static u64 bytes_used = 0;
62 static u64 total_csum_bytes = 0;
63 static u64 total_btree_bytes = 0;
64 static u64 total_fs_tree_bytes = 0;
65 static u64 total_extent_tree_bytes = 0;
66 static u64 btree_space_waste = 0;
67 static u64 data_bytes_allocated = 0;
68 static u64 data_bytes_referenced = 0;
69 static int found_old_backref = 0;
70 static LIST_HEAD(duplicate_extents);
71 static LIST_HEAD(delete_items);
72 static int no_holes = 0;
73 static int init_extent_tree = 0;
74 static int check_data_csum = 0;
75 static struct btrfs_fs_info *global_info;
76 static struct task_ctx ctx = { 0 };
77 static struct cache_tree *roots_info_cache = NULL;
79 enum btrfs_check_mode {
83 CHECK_MODE_DEFAULT = CHECK_MODE_ORIGINAL
86 static enum btrfs_check_mode check_mode = CHECK_MODE_DEFAULT;
88 struct extent_backref {
89 struct list_head list;
90 unsigned int is_data:1;
91 unsigned int found_extent_tree:1;
92 unsigned int full_backref:1;
93 unsigned int found_ref:1;
94 unsigned int broken:1;
97 static inline struct extent_backref* to_extent_backref(struct list_head *entry)
99 return list_entry(entry, struct extent_backref, list);
102 struct data_backref {
103 struct extent_backref node;
117 #define ROOT_DIR_ERROR (1<<1) /* bad ROOT_DIR */
118 #define DIR_ITEM_MISSING (1<<2) /* DIR_ITEM not found */
119 #define DIR_ITEM_MISMATCH (1<<3) /* DIR_ITEM found but not match */
120 #define INODE_REF_MISSING (1<<4) /* INODE_REF/INODE_EXTREF not found */
121 #define INODE_ITEM_MISSING (1<<5) /* INODE_ITEM not found */
122 #define INODE_ITEM_MISMATCH (1<<6) /* INODE_ITEM found but not match */
123 #define FILE_EXTENT_ERROR (1<<7) /* bad FILE_EXTENT */
124 #define ODD_CSUM_ITEM (1<<8) /* CSUM_ITEM error */
125 #define CSUM_ITEM_MISSING (1<<9) /* CSUM_ITEM not found */
126 #define LINK_COUNT_ERROR (1<<10) /* INODE_ITEM nlink count error */
127 #define NBYTES_ERROR (1<<11) /* INODE_ITEM nbytes count error */
128 #define ISIZE_ERROR (1<<12) /* INODE_ITEM size count error */
129 #define ORPHAN_ITEM (1<<13) /* INODE_ITEM no reference */
130 #define NO_INODE_ITEM (1<<14) /* no inode_item */
131 #define LAST_ITEM (1<<15) /* Complete this tree traversal */
132 #define ROOT_REF_MISSING (1<<16) /* ROOT_REF not found */
133 #define ROOT_REF_MISMATCH (1<<17) /* ROOT_REF found but not match */
135 static inline struct data_backref* to_data_backref(struct extent_backref *back)
137 return container_of(back, struct data_backref, node);
141 * Much like data_backref, just removed the undetermined members
142 * and change it to use list_head.
143 * During extent scan, it is stored in root->orphan_data_extent.
144 * During fs tree scan, it is then moved to inode_rec->orphan_data_extents.
146 struct orphan_data_extent {
147 struct list_head list;
155 struct tree_backref {
156 struct extent_backref node;
163 static inline struct tree_backref* to_tree_backref(struct extent_backref *back)
165 return container_of(back, struct tree_backref, node);
168 /* Explicit initialization for extent_record::flag_block_full_backref */
169 enum { FLAG_UNSET = 2 };
171 struct extent_record {
172 struct list_head backrefs;
173 struct list_head dups;
174 struct list_head list;
175 struct cache_extent cache;
176 struct btrfs_disk_key parent_key;
181 u64 extent_item_refs;
183 u64 parent_generation;
187 unsigned int flag_block_full_backref:2;
188 unsigned int found_rec:1;
189 unsigned int content_checked:1;
190 unsigned int owner_ref_checked:1;
191 unsigned int is_root:1;
192 unsigned int metadata:1;
193 unsigned int bad_full_backref:1;
194 unsigned int crossing_stripes:1;
195 unsigned int wrong_chunk_type:1;
198 static inline struct extent_record* to_extent_record(struct list_head *entry)
200 return container_of(entry, struct extent_record, list);
203 struct inode_backref {
204 struct list_head list;
205 unsigned int found_dir_item:1;
206 unsigned int found_dir_index:1;
207 unsigned int found_inode_ref:1;
217 static inline struct inode_backref* to_inode_backref(struct list_head *entry)
219 return list_entry(entry, struct inode_backref, list);
222 struct root_item_record {
223 struct list_head list;
230 struct btrfs_key drop_key;
233 #define REF_ERR_NO_DIR_ITEM (1 << 0)
234 #define REF_ERR_NO_DIR_INDEX (1 << 1)
235 #define REF_ERR_NO_INODE_REF (1 << 2)
236 #define REF_ERR_DUP_DIR_ITEM (1 << 3)
237 #define REF_ERR_DUP_DIR_INDEX (1 << 4)
238 #define REF_ERR_DUP_INODE_REF (1 << 5)
239 #define REF_ERR_INDEX_UNMATCH (1 << 6)
240 #define REF_ERR_FILETYPE_UNMATCH (1 << 7)
241 #define REF_ERR_NAME_TOO_LONG (1 << 8) // 100
242 #define REF_ERR_NO_ROOT_REF (1 << 9)
243 #define REF_ERR_NO_ROOT_BACKREF (1 << 10)
244 #define REF_ERR_DUP_ROOT_REF (1 << 11)
245 #define REF_ERR_DUP_ROOT_BACKREF (1 << 12)
247 struct file_extent_hole {
253 struct inode_record {
254 struct list_head backrefs;
255 unsigned int checked:1;
256 unsigned int merging:1;
257 unsigned int found_inode_item:1;
258 unsigned int found_dir_item:1;
259 unsigned int found_file_extent:1;
260 unsigned int found_csum_item:1;
261 unsigned int some_csum_missing:1;
262 unsigned int nodatasum:1;
275 struct rb_root holes;
276 struct list_head orphan_extents;
281 #define I_ERR_NO_INODE_ITEM (1 << 0)
282 #define I_ERR_NO_ORPHAN_ITEM (1 << 1)
283 #define I_ERR_DUP_INODE_ITEM (1 << 2)
284 #define I_ERR_DUP_DIR_INDEX (1 << 3)
285 #define I_ERR_ODD_DIR_ITEM (1 << 4)
286 #define I_ERR_ODD_FILE_EXTENT (1 << 5)
287 #define I_ERR_BAD_FILE_EXTENT (1 << 6)
288 #define I_ERR_FILE_EXTENT_OVERLAP (1 << 7)
289 #define I_ERR_FILE_EXTENT_DISCOUNT (1 << 8) // 100
290 #define I_ERR_DIR_ISIZE_WRONG (1 << 9)
291 #define I_ERR_FILE_NBYTES_WRONG (1 << 10) // 400
292 #define I_ERR_ODD_CSUM_ITEM (1 << 11)
293 #define I_ERR_SOME_CSUM_MISSING (1 << 12)
294 #define I_ERR_LINK_COUNT_WRONG (1 << 13)
295 #define I_ERR_FILE_EXTENT_ORPHAN (1 << 14)
297 struct root_backref {
298 struct list_head list;
299 unsigned int found_dir_item:1;
300 unsigned int found_dir_index:1;
301 unsigned int found_back_ref:1;
302 unsigned int found_forward_ref:1;
303 unsigned int reachable:1;
312 static inline struct root_backref* to_root_backref(struct list_head *entry)
314 return list_entry(entry, struct root_backref, list);
318 struct list_head backrefs;
319 struct cache_extent cache;
320 unsigned int found_root_item:1;
326 struct cache_extent cache;
331 struct cache_extent cache;
332 struct cache_tree root_cache;
333 struct cache_tree inode_cache;
334 struct inode_record *current;
343 struct walk_control {
344 struct cache_tree shared;
345 struct shared_node *nodes[BTRFS_MAX_LEVEL];
351 struct btrfs_key key;
353 struct list_head list;
356 struct extent_entry {
361 struct list_head list;
364 struct root_item_info {
365 /* level of the root */
367 /* number of nodes at this level, must be 1 for a root */
371 struct cache_extent cache_extent;
375 * Error bit for low memory mode check.
377 * Currently no caller cares about it yet. Just internal use for error
380 #define BACKREF_MISSING (1 << 0) /* Backref missing in extent tree */
381 #define BACKREF_MISMATCH (1 << 1) /* Backref exists but does not match */
382 #define BYTES_UNALIGNED (1 << 2) /* Some bytes are not aligned */
383 #define REFERENCER_MISSING (1 << 3) /* Referencer not found */
384 #define REFERENCER_MISMATCH (1 << 4) /* Referenceer found but does not match */
385 #define CROSSING_STRIPE_BOUNDARY (1 << 4) /* For kernel scrub workaround */
386 #define ITEM_SIZE_MISMATCH (1 << 5) /* Bad item size */
387 #define UNKNOWN_TYPE (1 << 6) /* Unknown type */
388 #define ACCOUNTING_MISMATCH (1 << 7) /* Used space accounting error */
389 #define CHUNK_TYPE_MISMATCH (1 << 8)
391 static void *print_status_check(void *p)
393 struct task_ctx *priv = p;
394 const char work_indicator[] = { '.', 'o', 'O', 'o' };
396 static char *task_position_string[] = {
398 "checking free space cache",
402 task_period_start(priv->info, 1000 /* 1s */);
404 if (priv->tp == TASK_NOTHING)
408 printf("%s [%c]\r", task_position_string[priv->tp],
409 work_indicator[count % 4]);
412 task_period_wait(priv->info);
417 static int print_status_return(void *p)
425 static enum btrfs_check_mode parse_check_mode(const char *str)
427 if (strcmp(str, "lowmem") == 0)
428 return CHECK_MODE_LOWMEM;
429 if (strcmp(str, "orig") == 0)
430 return CHECK_MODE_ORIGINAL;
431 if (strcmp(str, "original") == 0)
432 return CHECK_MODE_ORIGINAL;
434 return CHECK_MODE_UNKNOWN;
437 /* Compatible function to allow reuse of old codes */
438 static u64 first_extent_gap(struct rb_root *holes)
440 struct file_extent_hole *hole;
442 if (RB_EMPTY_ROOT(holes))
445 hole = rb_entry(rb_first(holes), struct file_extent_hole, node);
449 static int compare_hole(struct rb_node *node1, struct rb_node *node2)
451 struct file_extent_hole *hole1;
452 struct file_extent_hole *hole2;
454 hole1 = rb_entry(node1, struct file_extent_hole, node);
455 hole2 = rb_entry(node2, struct file_extent_hole, node);
457 if (hole1->start > hole2->start)
459 if (hole1->start < hole2->start)
461 /* Now hole1->start == hole2->start */
462 if (hole1->len >= hole2->len)
464 * Hole 1 will be merge center
465 * Same hole will be merged later
468 /* Hole 2 will be merge center */
473 * Add a hole to the record
475 * This will do hole merge for copy_file_extent_holes(),
476 * which will ensure there won't be continuous holes.
478 static int add_file_extent_hole(struct rb_root *holes,
481 struct file_extent_hole *hole;
482 struct file_extent_hole *prev = NULL;
483 struct file_extent_hole *next = NULL;
485 hole = malloc(sizeof(*hole));
490 /* Since compare will not return 0, no -EEXIST will happen */
491 rb_insert(holes, &hole->node, compare_hole);
493 /* simple merge with previous hole */
494 if (rb_prev(&hole->node))
495 prev = rb_entry(rb_prev(&hole->node), struct file_extent_hole,
497 if (prev && prev->start + prev->len >= hole->start) {
498 hole->len = hole->start + hole->len - prev->start;
499 hole->start = prev->start;
500 rb_erase(&prev->node, holes);
505 /* iterate merge with next holes */
507 if (!rb_next(&hole->node))
509 next = rb_entry(rb_next(&hole->node), struct file_extent_hole,
511 if (hole->start + hole->len >= next->start) {
512 if (hole->start + hole->len <= next->start + next->len)
513 hole->len = next->start + next->len -
515 rb_erase(&next->node, holes);
524 static int compare_hole_range(struct rb_node *node, void *data)
526 struct file_extent_hole *hole;
529 hole = (struct file_extent_hole *)data;
532 hole = rb_entry(node, struct file_extent_hole, node);
533 if (start < hole->start)
535 if (start >= hole->start && start < hole->start + hole->len)
541 * Delete a hole in the record
543 * This will do the hole split and is much restrict than add.
545 static int del_file_extent_hole(struct rb_root *holes,
548 struct file_extent_hole *hole;
549 struct file_extent_hole tmp;
554 struct rb_node *node;
561 node = rb_search(holes, &tmp, compare_hole_range, NULL);
564 hole = rb_entry(node, struct file_extent_hole, node);
565 if (start + len > hole->start + hole->len)
569 * Now there will be no overlap, delete the hole and re-add the
570 * split(s) if they exists.
572 if (start > hole->start) {
573 prev_start = hole->start;
574 prev_len = start - hole->start;
577 if (hole->start + hole->len > start + len) {
578 next_start = start + len;
579 next_len = hole->start + hole->len - start - len;
582 rb_erase(node, holes);
585 ret = add_file_extent_hole(holes, prev_start, prev_len);
590 ret = add_file_extent_hole(holes, next_start, next_len);
597 static int copy_file_extent_holes(struct rb_root *dst,
600 struct file_extent_hole *hole;
601 struct rb_node *node;
604 node = rb_first(src);
606 hole = rb_entry(node, struct file_extent_hole, node);
607 ret = add_file_extent_hole(dst, hole->start, hole->len);
610 node = rb_next(node);
615 static void free_file_extent_holes(struct rb_root *holes)
617 struct rb_node *node;
618 struct file_extent_hole *hole;
620 node = rb_first(holes);
622 hole = rb_entry(node, struct file_extent_hole, node);
623 rb_erase(node, holes);
625 node = rb_first(holes);
629 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info);
631 static void record_root_in_trans(struct btrfs_trans_handle *trans,
632 struct btrfs_root *root)
634 if (root->last_trans != trans->transid) {
635 root->track_dirty = 1;
636 root->last_trans = trans->transid;
637 root->commit_root = root->node;
638 extent_buffer_get(root->node);
642 static u8 imode_to_type(u32 imode)
645 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
646 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
647 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
648 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
649 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
650 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
651 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
652 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
655 return btrfs_type_by_mode[(imode & S_IFMT) >> S_SHIFT];
659 static int device_record_compare(struct rb_node *node1, struct rb_node *node2)
661 struct device_record *rec1;
662 struct device_record *rec2;
664 rec1 = rb_entry(node1, struct device_record, node);
665 rec2 = rb_entry(node2, struct device_record, node);
666 if (rec1->devid > rec2->devid)
668 else if (rec1->devid < rec2->devid)
674 static struct inode_record *clone_inode_rec(struct inode_record *orig_rec)
676 struct inode_record *rec;
677 struct inode_backref *backref;
678 struct inode_backref *orig;
679 struct inode_backref *tmp;
680 struct orphan_data_extent *src_orphan;
681 struct orphan_data_extent *dst_orphan;
686 rec = malloc(sizeof(*rec));
688 return ERR_PTR(-ENOMEM);
689 memcpy(rec, orig_rec, sizeof(*rec));
691 INIT_LIST_HEAD(&rec->backrefs);
692 INIT_LIST_HEAD(&rec->orphan_extents);
693 rec->holes = RB_ROOT;
695 list_for_each_entry(orig, &orig_rec->backrefs, list) {
696 size = sizeof(*orig) + orig->namelen + 1;
697 backref = malloc(size);
702 memcpy(backref, orig, size);
703 list_add_tail(&backref->list, &rec->backrefs);
705 list_for_each_entry(src_orphan, &orig_rec->orphan_extents, list) {
706 dst_orphan = malloc(sizeof(*dst_orphan));
711 memcpy(dst_orphan, src_orphan, sizeof(*src_orphan));
712 list_add_tail(&dst_orphan->list, &rec->orphan_extents);
714 ret = copy_file_extent_holes(&rec->holes, &orig_rec->holes);
721 rb = rb_first(&rec->holes);
723 struct file_extent_hole *hole;
725 hole = rb_entry(rb, struct file_extent_hole, node);
731 if (!list_empty(&rec->backrefs))
732 list_for_each_entry_safe(orig, tmp, &rec->backrefs, list) {
733 list_del(&orig->list);
737 if (!list_empty(&rec->orphan_extents))
738 list_for_each_entry_safe(orig, tmp, &rec->orphan_extents, list) {
739 list_del(&orig->list);
748 static void print_orphan_data_extents(struct list_head *orphan_extents,
751 struct orphan_data_extent *orphan;
753 if (list_empty(orphan_extents))
755 printf("The following data extent is lost in tree %llu:\n",
757 list_for_each_entry(orphan, orphan_extents, list) {
758 printf("\tinode: %llu, offset:%llu, disk_bytenr: %llu, disk_len: %llu\n",
759 orphan->objectid, orphan->offset, orphan->disk_bytenr,
764 static void print_inode_error(struct btrfs_root *root, struct inode_record *rec)
766 u64 root_objectid = root->root_key.objectid;
767 int errors = rec->errors;
771 /* reloc root errors, we print its corresponding fs root objectid*/
772 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
773 root_objectid = root->root_key.offset;
774 fprintf(stderr, "reloc");
776 fprintf(stderr, "root %llu inode %llu errors %x",
777 (unsigned long long) root_objectid,
778 (unsigned long long) rec->ino, rec->errors);
780 if (errors & I_ERR_NO_INODE_ITEM)
781 fprintf(stderr, ", no inode item");
782 if (errors & I_ERR_NO_ORPHAN_ITEM)
783 fprintf(stderr, ", no orphan item");
784 if (errors & I_ERR_DUP_INODE_ITEM)
785 fprintf(stderr, ", dup inode item");
786 if (errors & I_ERR_DUP_DIR_INDEX)
787 fprintf(stderr, ", dup dir index");
788 if (errors & I_ERR_ODD_DIR_ITEM)
789 fprintf(stderr, ", odd dir item");
790 if (errors & I_ERR_ODD_FILE_EXTENT)
791 fprintf(stderr, ", odd file extent");
792 if (errors & I_ERR_BAD_FILE_EXTENT)
793 fprintf(stderr, ", bad file extent");
794 if (errors & I_ERR_FILE_EXTENT_OVERLAP)
795 fprintf(stderr, ", file extent overlap");
796 if (errors & I_ERR_FILE_EXTENT_DISCOUNT)
797 fprintf(stderr, ", file extent discount");
798 if (errors & I_ERR_DIR_ISIZE_WRONG)
799 fprintf(stderr, ", dir isize wrong");
800 if (errors & I_ERR_FILE_NBYTES_WRONG)
801 fprintf(stderr, ", nbytes wrong");
802 if (errors & I_ERR_ODD_CSUM_ITEM)
803 fprintf(stderr, ", odd csum item");
804 if (errors & I_ERR_SOME_CSUM_MISSING)
805 fprintf(stderr, ", some csum missing");
806 if (errors & I_ERR_LINK_COUNT_WRONG)
807 fprintf(stderr, ", link count wrong");
808 if (errors & I_ERR_FILE_EXTENT_ORPHAN)
809 fprintf(stderr, ", orphan file extent");
810 fprintf(stderr, "\n");
811 /* Print the orphan extents if needed */
812 if (errors & I_ERR_FILE_EXTENT_ORPHAN)
813 print_orphan_data_extents(&rec->orphan_extents, root->objectid);
815 /* Print the holes if needed */
816 if (errors & I_ERR_FILE_EXTENT_DISCOUNT) {
817 struct file_extent_hole *hole;
818 struct rb_node *node;
821 node = rb_first(&rec->holes);
822 fprintf(stderr, "Found file extent holes:\n");
825 hole = rb_entry(node, struct file_extent_hole, node);
826 fprintf(stderr, "\tstart: %llu, len: %llu\n",
827 hole->start, hole->len);
828 node = rb_next(node);
831 fprintf(stderr, "\tstart: 0, len: %llu\n",
832 round_up(rec->isize, root->sectorsize));
836 static void print_ref_error(int errors)
838 if (errors & REF_ERR_NO_DIR_ITEM)
839 fprintf(stderr, ", no dir item");
840 if (errors & REF_ERR_NO_DIR_INDEX)
841 fprintf(stderr, ", no dir index");
842 if (errors & REF_ERR_NO_INODE_REF)
843 fprintf(stderr, ", no inode ref");
844 if (errors & REF_ERR_DUP_DIR_ITEM)
845 fprintf(stderr, ", dup dir item");
846 if (errors & REF_ERR_DUP_DIR_INDEX)
847 fprintf(stderr, ", dup dir index");
848 if (errors & REF_ERR_DUP_INODE_REF)
849 fprintf(stderr, ", dup inode ref");
850 if (errors & REF_ERR_INDEX_UNMATCH)
851 fprintf(stderr, ", index mismatch");
852 if (errors & REF_ERR_FILETYPE_UNMATCH)
853 fprintf(stderr, ", filetype mismatch");
854 if (errors & REF_ERR_NAME_TOO_LONG)
855 fprintf(stderr, ", name too long");
856 if (errors & REF_ERR_NO_ROOT_REF)
857 fprintf(stderr, ", no root ref");
858 if (errors & REF_ERR_NO_ROOT_BACKREF)
859 fprintf(stderr, ", no root backref");
860 if (errors & REF_ERR_DUP_ROOT_REF)
861 fprintf(stderr, ", dup root ref");
862 if (errors & REF_ERR_DUP_ROOT_BACKREF)
863 fprintf(stderr, ", dup root backref");
864 fprintf(stderr, "\n");
867 static struct inode_record *get_inode_rec(struct cache_tree *inode_cache,
870 struct ptr_node *node;
871 struct cache_extent *cache;
872 struct inode_record *rec = NULL;
875 cache = lookup_cache_extent(inode_cache, ino, 1);
877 node = container_of(cache, struct ptr_node, cache);
879 if (mod && rec->refs > 1) {
880 node->data = clone_inode_rec(rec);
881 if (IS_ERR(node->data))
887 rec = calloc(1, sizeof(*rec));
889 return ERR_PTR(-ENOMEM);
891 rec->extent_start = (u64)-1;
893 INIT_LIST_HEAD(&rec->backrefs);
894 INIT_LIST_HEAD(&rec->orphan_extents);
895 rec->holes = RB_ROOT;
897 node = malloc(sizeof(*node));
900 return ERR_PTR(-ENOMEM);
902 node->cache.start = ino;
903 node->cache.size = 1;
906 if (ino == BTRFS_FREE_INO_OBJECTID)
909 ret = insert_cache_extent(inode_cache, &node->cache);
911 return ERR_PTR(-EEXIST);
916 static void free_orphan_data_extents(struct list_head *orphan_extents)
918 struct orphan_data_extent *orphan;
920 while (!list_empty(orphan_extents)) {
921 orphan = list_entry(orphan_extents->next,
922 struct orphan_data_extent, list);
923 list_del(&orphan->list);
928 static void free_inode_rec(struct inode_record *rec)
930 struct inode_backref *backref;
935 while (!list_empty(&rec->backrefs)) {
936 backref = to_inode_backref(rec->backrefs.next);
937 list_del(&backref->list);
940 free_orphan_data_extents(&rec->orphan_extents);
941 free_file_extent_holes(&rec->holes);
945 static int can_free_inode_rec(struct inode_record *rec)
947 if (!rec->errors && rec->checked && rec->found_inode_item &&
948 rec->nlink == rec->found_link && list_empty(&rec->backrefs))
953 static void maybe_free_inode_rec(struct cache_tree *inode_cache,
954 struct inode_record *rec)
956 struct cache_extent *cache;
957 struct inode_backref *tmp, *backref;
958 struct ptr_node *node;
961 if (!rec->found_inode_item)
964 filetype = imode_to_type(rec->imode);
965 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
966 if (backref->found_dir_item && backref->found_dir_index) {
967 if (backref->filetype != filetype)
968 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
969 if (!backref->errors && backref->found_inode_ref &&
970 rec->nlink == rec->found_link) {
971 list_del(&backref->list);
977 if (!rec->checked || rec->merging)
980 if (S_ISDIR(rec->imode)) {
981 if (rec->found_size != rec->isize)
982 rec->errors |= I_ERR_DIR_ISIZE_WRONG;
983 if (rec->found_file_extent)
984 rec->errors |= I_ERR_ODD_FILE_EXTENT;
985 } else if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
986 if (rec->found_dir_item)
987 rec->errors |= I_ERR_ODD_DIR_ITEM;
988 if (rec->found_size != rec->nbytes)
989 rec->errors |= I_ERR_FILE_NBYTES_WRONG;
990 if (rec->nlink > 0 && !no_holes &&
991 (rec->extent_end < rec->isize ||
992 first_extent_gap(&rec->holes) < rec->isize))
993 rec->errors |= I_ERR_FILE_EXTENT_DISCOUNT;
996 if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
997 if (rec->found_csum_item && rec->nodatasum)
998 rec->errors |= I_ERR_ODD_CSUM_ITEM;
999 if (rec->some_csum_missing && !rec->nodatasum)
1000 rec->errors |= I_ERR_SOME_CSUM_MISSING;
1003 BUG_ON(rec->refs != 1);
1004 if (can_free_inode_rec(rec)) {
1005 cache = lookup_cache_extent(inode_cache, rec->ino, 1);
1006 node = container_of(cache, struct ptr_node, cache);
1007 BUG_ON(node->data != rec);
1008 remove_cache_extent(inode_cache, &node->cache);
1010 free_inode_rec(rec);
1014 static int check_orphan_item(struct btrfs_root *root, u64 ino)
1016 struct btrfs_path path;
1017 struct btrfs_key key;
1020 key.objectid = BTRFS_ORPHAN_OBJECTID;
1021 key.type = BTRFS_ORPHAN_ITEM_KEY;
1024 btrfs_init_path(&path);
1025 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1026 btrfs_release_path(&path);
1032 static int process_inode_item(struct extent_buffer *eb,
1033 int slot, struct btrfs_key *key,
1034 struct shared_node *active_node)
1036 struct inode_record *rec;
1037 struct btrfs_inode_item *item;
1039 rec = active_node->current;
1040 BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1041 if (rec->found_inode_item) {
1042 rec->errors |= I_ERR_DUP_INODE_ITEM;
1045 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
1046 rec->nlink = btrfs_inode_nlink(eb, item);
1047 rec->isize = btrfs_inode_size(eb, item);
1048 rec->nbytes = btrfs_inode_nbytes(eb, item);
1049 rec->imode = btrfs_inode_mode(eb, item);
1050 if (btrfs_inode_flags(eb, item) & BTRFS_INODE_NODATASUM)
1052 rec->found_inode_item = 1;
1053 if (rec->nlink == 0)
1054 rec->errors |= I_ERR_NO_ORPHAN_ITEM;
1055 maybe_free_inode_rec(&active_node->inode_cache, rec);
1059 static struct inode_backref *get_inode_backref(struct inode_record *rec,
1061 int namelen, u64 dir)
1063 struct inode_backref *backref;
1065 list_for_each_entry(backref, &rec->backrefs, list) {
1066 if (rec->ino == BTRFS_MULTIPLE_OBJECTIDS)
1068 if (backref->dir != dir || backref->namelen != namelen)
1070 if (memcmp(name, backref->name, namelen))
1075 backref = malloc(sizeof(*backref) + namelen + 1);
1078 memset(backref, 0, sizeof(*backref));
1080 backref->namelen = namelen;
1081 memcpy(backref->name, name, namelen);
1082 backref->name[namelen] = '\0';
1083 list_add_tail(&backref->list, &rec->backrefs);
1087 static int add_inode_backref(struct cache_tree *inode_cache,
1088 u64 ino, u64 dir, u64 index,
1089 const char *name, int namelen,
1090 u8 filetype, u8 itemtype, int errors)
1092 struct inode_record *rec;
1093 struct inode_backref *backref;
1095 rec = get_inode_rec(inode_cache, ino, 1);
1096 BUG_ON(IS_ERR(rec));
1097 backref = get_inode_backref(rec, name, namelen, dir);
1100 backref->errors |= errors;
1101 if (itemtype == BTRFS_DIR_INDEX_KEY) {
1102 if (backref->found_dir_index)
1103 backref->errors |= REF_ERR_DUP_DIR_INDEX;
1104 if (backref->found_inode_ref && backref->index != index)
1105 backref->errors |= REF_ERR_INDEX_UNMATCH;
1106 if (backref->found_dir_item && backref->filetype != filetype)
1107 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
1109 backref->index = index;
1110 backref->filetype = filetype;
1111 backref->found_dir_index = 1;
1112 } else if (itemtype == BTRFS_DIR_ITEM_KEY) {
1114 if (backref->found_dir_item)
1115 backref->errors |= REF_ERR_DUP_DIR_ITEM;
1116 if (backref->found_dir_index && backref->filetype != filetype)
1117 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
1119 backref->filetype = filetype;
1120 backref->found_dir_item = 1;
1121 } else if ((itemtype == BTRFS_INODE_REF_KEY) ||
1122 (itemtype == BTRFS_INODE_EXTREF_KEY)) {
1123 if (backref->found_inode_ref)
1124 backref->errors |= REF_ERR_DUP_INODE_REF;
1125 if (backref->found_dir_index && backref->index != index)
1126 backref->errors |= REF_ERR_INDEX_UNMATCH;
1128 backref->index = index;
1130 backref->ref_type = itemtype;
1131 backref->found_inode_ref = 1;
1136 maybe_free_inode_rec(inode_cache, rec);
1140 static int merge_inode_recs(struct inode_record *src, struct inode_record *dst,
1141 struct cache_tree *dst_cache)
1143 struct inode_backref *backref;
1148 list_for_each_entry(backref, &src->backrefs, list) {
1149 if (backref->found_dir_index) {
1150 add_inode_backref(dst_cache, dst->ino, backref->dir,
1151 backref->index, backref->name,
1152 backref->namelen, backref->filetype,
1153 BTRFS_DIR_INDEX_KEY, backref->errors);
1155 if (backref->found_dir_item) {
1157 add_inode_backref(dst_cache, dst->ino,
1158 backref->dir, 0, backref->name,
1159 backref->namelen, backref->filetype,
1160 BTRFS_DIR_ITEM_KEY, backref->errors);
1162 if (backref->found_inode_ref) {
1163 add_inode_backref(dst_cache, dst->ino,
1164 backref->dir, backref->index,
1165 backref->name, backref->namelen, 0,
1166 backref->ref_type, backref->errors);
1170 if (src->found_dir_item)
1171 dst->found_dir_item = 1;
1172 if (src->found_file_extent)
1173 dst->found_file_extent = 1;
1174 if (src->found_csum_item)
1175 dst->found_csum_item = 1;
1176 if (src->some_csum_missing)
1177 dst->some_csum_missing = 1;
1178 if (first_extent_gap(&dst->holes) > first_extent_gap(&src->holes)) {
1179 ret = copy_file_extent_holes(&dst->holes, &src->holes);
1184 BUG_ON(src->found_link < dir_count);
1185 dst->found_link += src->found_link - dir_count;
1186 dst->found_size += src->found_size;
1187 if (src->extent_start != (u64)-1) {
1188 if (dst->extent_start == (u64)-1) {
1189 dst->extent_start = src->extent_start;
1190 dst->extent_end = src->extent_end;
1192 if (dst->extent_end > src->extent_start)
1193 dst->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1194 else if (dst->extent_end < src->extent_start) {
1195 ret = add_file_extent_hole(&dst->holes,
1197 src->extent_start - dst->extent_end);
1199 if (dst->extent_end < src->extent_end)
1200 dst->extent_end = src->extent_end;
1204 dst->errors |= src->errors;
1205 if (src->found_inode_item) {
1206 if (!dst->found_inode_item) {
1207 dst->nlink = src->nlink;
1208 dst->isize = src->isize;
1209 dst->nbytes = src->nbytes;
1210 dst->imode = src->imode;
1211 dst->nodatasum = src->nodatasum;
1212 dst->found_inode_item = 1;
1214 dst->errors |= I_ERR_DUP_INODE_ITEM;
1222 static int splice_shared_node(struct shared_node *src_node,
1223 struct shared_node *dst_node)
1225 struct cache_extent *cache;
1226 struct ptr_node *node, *ins;
1227 struct cache_tree *src, *dst;
1228 struct inode_record *rec, *conflict;
1229 u64 current_ino = 0;
1233 if (--src_node->refs == 0)
1235 if (src_node->current)
1236 current_ino = src_node->current->ino;
1238 src = &src_node->root_cache;
1239 dst = &dst_node->root_cache;
1241 cache = search_cache_extent(src, 0);
1243 node = container_of(cache, struct ptr_node, cache);
1245 cache = next_cache_extent(cache);
1248 remove_cache_extent(src, &node->cache);
1251 ins = malloc(sizeof(*ins));
1253 ins->cache.start = node->cache.start;
1254 ins->cache.size = node->cache.size;
1258 ret = insert_cache_extent(dst, &ins->cache);
1259 if (ret == -EEXIST) {
1260 conflict = get_inode_rec(dst, rec->ino, 1);
1261 BUG_ON(IS_ERR(conflict));
1262 merge_inode_recs(rec, conflict, dst);
1264 conflict->checked = 1;
1265 if (dst_node->current == conflict)
1266 dst_node->current = NULL;
1268 maybe_free_inode_rec(dst, conflict);
1269 free_inode_rec(rec);
1276 if (src == &src_node->root_cache) {
1277 src = &src_node->inode_cache;
1278 dst = &dst_node->inode_cache;
1282 if (current_ino > 0 && (!dst_node->current ||
1283 current_ino > dst_node->current->ino)) {
1284 if (dst_node->current) {
1285 dst_node->current->checked = 1;
1286 maybe_free_inode_rec(dst, dst_node->current);
1288 dst_node->current = get_inode_rec(dst, current_ino, 1);
1289 BUG_ON(IS_ERR(dst_node->current));
1294 static void free_inode_ptr(struct cache_extent *cache)
1296 struct ptr_node *node;
1297 struct inode_record *rec;
1299 node = container_of(cache, struct ptr_node, cache);
1301 free_inode_rec(rec);
1305 FREE_EXTENT_CACHE_BASED_TREE(inode_recs, free_inode_ptr);
1307 static struct shared_node *find_shared_node(struct cache_tree *shared,
1310 struct cache_extent *cache;
1311 struct shared_node *node;
1313 cache = lookup_cache_extent(shared, bytenr, 1);
1315 node = container_of(cache, struct shared_node, cache);
1321 static int add_shared_node(struct cache_tree *shared, u64 bytenr, u32 refs)
1324 struct shared_node *node;
1326 node = calloc(1, sizeof(*node));
1329 node->cache.start = bytenr;
1330 node->cache.size = 1;
1331 cache_tree_init(&node->root_cache);
1332 cache_tree_init(&node->inode_cache);
1335 ret = insert_cache_extent(shared, &node->cache);
1340 static int enter_shared_node(struct btrfs_root *root, u64 bytenr, u32 refs,
1341 struct walk_control *wc, int level)
1343 struct shared_node *node;
1344 struct shared_node *dest;
1347 if (level == wc->active_node)
1350 BUG_ON(wc->active_node <= level);
1351 node = find_shared_node(&wc->shared, bytenr);
1353 ret = add_shared_node(&wc->shared, bytenr, refs);
1355 node = find_shared_node(&wc->shared, bytenr);
1356 wc->nodes[level] = node;
1357 wc->active_node = level;
1361 if (wc->root_level == wc->active_node &&
1362 btrfs_root_refs(&root->root_item) == 0) {
1363 if (--node->refs == 0) {
1364 free_inode_recs_tree(&node->root_cache);
1365 free_inode_recs_tree(&node->inode_cache);
1366 remove_cache_extent(&wc->shared, &node->cache);
1372 dest = wc->nodes[wc->active_node];
1373 splice_shared_node(node, dest);
1374 if (node->refs == 0) {
1375 remove_cache_extent(&wc->shared, &node->cache);
1381 static int leave_shared_node(struct btrfs_root *root,
1382 struct walk_control *wc, int level)
1384 struct shared_node *node;
1385 struct shared_node *dest;
1388 if (level == wc->root_level)
1391 for (i = level + 1; i < BTRFS_MAX_LEVEL; i++) {
1395 BUG_ON(i >= BTRFS_MAX_LEVEL);
1397 node = wc->nodes[wc->active_node];
1398 wc->nodes[wc->active_node] = NULL;
1399 wc->active_node = i;
1401 dest = wc->nodes[wc->active_node];
1402 if (wc->active_node < wc->root_level ||
1403 btrfs_root_refs(&root->root_item) > 0) {
1404 BUG_ON(node->refs <= 1);
1405 splice_shared_node(node, dest);
1407 BUG_ON(node->refs < 2);
1416 * 1 - if the root with id child_root_id is a child of root parent_root_id
1417 * 0 - if the root child_root_id isn't a child of the root parent_root_id but
1418 * has other root(s) as parent(s)
1419 * 2 - if the root child_root_id doesn't have any parent roots
1421 static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
1424 struct btrfs_path path;
1425 struct btrfs_key key;
1426 struct extent_buffer *leaf;
1430 btrfs_init_path(&path);
1432 key.objectid = parent_root_id;
1433 key.type = BTRFS_ROOT_REF_KEY;
1434 key.offset = child_root_id;
1435 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
1439 btrfs_release_path(&path);
1443 key.objectid = child_root_id;
1444 key.type = BTRFS_ROOT_BACKREF_KEY;
1446 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
1452 leaf = path.nodes[0];
1453 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1454 ret = btrfs_next_leaf(root->fs_info->tree_root, &path);
1457 leaf = path.nodes[0];
1460 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1461 if (key.objectid != child_root_id ||
1462 key.type != BTRFS_ROOT_BACKREF_KEY)
1467 if (key.offset == parent_root_id) {
1468 btrfs_release_path(&path);
1475 btrfs_release_path(&path);
1478 return has_parent ? 0 : 2;
1481 static int process_dir_item(struct extent_buffer *eb,
1482 int slot, struct btrfs_key *key,
1483 struct shared_node *active_node)
1493 struct btrfs_dir_item *di;
1494 struct inode_record *rec;
1495 struct cache_tree *root_cache;
1496 struct cache_tree *inode_cache;
1497 struct btrfs_key location;
1498 char namebuf[BTRFS_NAME_LEN];
1500 root_cache = &active_node->root_cache;
1501 inode_cache = &active_node->inode_cache;
1502 rec = active_node->current;
1503 rec->found_dir_item = 1;
1505 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
1506 total = btrfs_item_size_nr(eb, slot);
1507 while (cur < total) {
1509 btrfs_dir_item_key_to_cpu(eb, di, &location);
1510 name_len = btrfs_dir_name_len(eb, di);
1511 data_len = btrfs_dir_data_len(eb, di);
1512 filetype = btrfs_dir_type(eb, di);
1514 rec->found_size += name_len;
1515 if (name_len <= BTRFS_NAME_LEN) {
1519 len = BTRFS_NAME_LEN;
1520 error = REF_ERR_NAME_TOO_LONG;
1522 read_extent_buffer(eb, namebuf, (unsigned long)(di + 1), len);
1524 if (location.type == BTRFS_INODE_ITEM_KEY) {
1525 add_inode_backref(inode_cache, location.objectid,
1526 key->objectid, key->offset, namebuf,
1527 len, filetype, key->type, error);
1528 } else if (location.type == BTRFS_ROOT_ITEM_KEY) {
1529 add_inode_backref(root_cache, location.objectid,
1530 key->objectid, key->offset,
1531 namebuf, len, filetype,
1534 fprintf(stderr, "invalid location in dir item %u\n",
1536 add_inode_backref(inode_cache, BTRFS_MULTIPLE_OBJECTIDS,
1537 key->objectid, key->offset, namebuf,
1538 len, filetype, key->type, error);
1541 len = sizeof(*di) + name_len + data_len;
1542 di = (struct btrfs_dir_item *)((char *)di + len);
1545 if (key->type == BTRFS_DIR_INDEX_KEY && nritems > 1)
1546 rec->errors |= I_ERR_DUP_DIR_INDEX;
1551 static int process_inode_ref(struct extent_buffer *eb,
1552 int slot, struct btrfs_key *key,
1553 struct shared_node *active_node)
1561 struct cache_tree *inode_cache;
1562 struct btrfs_inode_ref *ref;
1563 char namebuf[BTRFS_NAME_LEN];
1565 inode_cache = &active_node->inode_cache;
1567 ref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1568 total = btrfs_item_size_nr(eb, slot);
1569 while (cur < total) {
1570 name_len = btrfs_inode_ref_name_len(eb, ref);
1571 index = btrfs_inode_ref_index(eb, ref);
1572 if (name_len <= BTRFS_NAME_LEN) {
1576 len = BTRFS_NAME_LEN;
1577 error = REF_ERR_NAME_TOO_LONG;
1579 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
1580 add_inode_backref(inode_cache, key->objectid, key->offset,
1581 index, namebuf, len, 0, key->type, error);
1583 len = sizeof(*ref) + name_len;
1584 ref = (struct btrfs_inode_ref *)((char *)ref + len);
1590 static int process_inode_extref(struct extent_buffer *eb,
1591 int slot, struct btrfs_key *key,
1592 struct shared_node *active_node)
1601 struct cache_tree *inode_cache;
1602 struct btrfs_inode_extref *extref;
1603 char namebuf[BTRFS_NAME_LEN];
1605 inode_cache = &active_node->inode_cache;
1607 extref = btrfs_item_ptr(eb, slot, struct btrfs_inode_extref);
1608 total = btrfs_item_size_nr(eb, slot);
1609 while (cur < total) {
1610 name_len = btrfs_inode_extref_name_len(eb, extref);
1611 index = btrfs_inode_extref_index(eb, extref);
1612 parent = btrfs_inode_extref_parent(eb, extref);
1613 if (name_len <= BTRFS_NAME_LEN) {
1617 len = BTRFS_NAME_LEN;
1618 error = REF_ERR_NAME_TOO_LONG;
1620 read_extent_buffer(eb, namebuf,
1621 (unsigned long)(extref + 1), len);
1622 add_inode_backref(inode_cache, key->objectid, parent,
1623 index, namebuf, len, 0, key->type, error);
1625 len = sizeof(*extref) + name_len;
1626 extref = (struct btrfs_inode_extref *)((char *)extref + len);
1633 static int count_csum_range(struct btrfs_root *root, u64 start,
1634 u64 len, u64 *found)
1636 struct btrfs_key key;
1637 struct btrfs_path path;
1638 struct extent_buffer *leaf;
1643 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
1645 btrfs_init_path(&path);
1647 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1649 key.type = BTRFS_EXTENT_CSUM_KEY;
1651 ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
1655 if (ret > 0 && path.slots[0] > 0) {
1656 leaf = path.nodes[0];
1657 btrfs_item_key_to_cpu(leaf, &key, path.slots[0] - 1);
1658 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
1659 key.type == BTRFS_EXTENT_CSUM_KEY)
1664 leaf = path.nodes[0];
1665 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1666 ret = btrfs_next_leaf(root->fs_info->csum_root, &path);
1671 leaf = path.nodes[0];
1674 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1675 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1676 key.type != BTRFS_EXTENT_CSUM_KEY)
1679 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1680 if (key.offset >= start + len)
1683 if (key.offset > start)
1686 size = btrfs_item_size_nr(leaf, path.slots[0]);
1687 csum_end = key.offset + (size / csum_size) * root->sectorsize;
1688 if (csum_end > start) {
1689 size = min(csum_end - start, len);
1698 btrfs_release_path(&path);
1704 static int process_file_extent(struct btrfs_root *root,
1705 struct extent_buffer *eb,
1706 int slot, struct btrfs_key *key,
1707 struct shared_node *active_node)
1709 struct inode_record *rec;
1710 struct btrfs_file_extent_item *fi;
1712 u64 disk_bytenr = 0;
1713 u64 extent_offset = 0;
1714 u64 mask = root->sectorsize - 1;
1718 rec = active_node->current;
1719 BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1720 rec->found_file_extent = 1;
1722 if (rec->extent_start == (u64)-1) {
1723 rec->extent_start = key->offset;
1724 rec->extent_end = key->offset;
1727 if (rec->extent_end > key->offset)
1728 rec->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1729 else if (rec->extent_end < key->offset) {
1730 ret = add_file_extent_hole(&rec->holes, rec->extent_end,
1731 key->offset - rec->extent_end);
1736 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
1737 extent_type = btrfs_file_extent_type(eb, fi);
1739 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1740 num_bytes = btrfs_file_extent_inline_len(eb, slot, fi);
1742 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1743 rec->found_size += num_bytes;
1744 num_bytes = (num_bytes + mask) & ~mask;
1745 } else if (extent_type == BTRFS_FILE_EXTENT_REG ||
1746 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1747 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1748 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1749 extent_offset = btrfs_file_extent_offset(eb, fi);
1750 if (num_bytes == 0 || (num_bytes & mask))
1751 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1752 if (num_bytes + extent_offset >
1753 btrfs_file_extent_ram_bytes(eb, fi))
1754 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1755 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC &&
1756 (btrfs_file_extent_compression(eb, fi) ||
1757 btrfs_file_extent_encryption(eb, fi) ||
1758 btrfs_file_extent_other_encoding(eb, fi)))
1759 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1760 if (disk_bytenr > 0)
1761 rec->found_size += num_bytes;
1763 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1765 rec->extent_end = key->offset + num_bytes;
1768 * The data reloc tree will copy full extents into its inode and then
1769 * copy the corresponding csums. Because the extent it copied could be
1770 * a preallocated extent that hasn't been written to yet there may be no
1771 * csums to copy, ergo we won't have csums for our file extent. This is
1772 * ok so just don't bother checking csums if the inode belongs to the
1775 if (disk_bytenr > 0 &&
1776 btrfs_header_owner(eb) != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1778 if (btrfs_file_extent_compression(eb, fi))
1779 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1781 disk_bytenr += extent_offset;
1783 ret = count_csum_range(root, disk_bytenr, num_bytes, &found);
1786 if (extent_type == BTRFS_FILE_EXTENT_REG) {
1788 rec->found_csum_item = 1;
1789 if (found < num_bytes)
1790 rec->some_csum_missing = 1;
1791 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1793 rec->errors |= I_ERR_ODD_CSUM_ITEM;
1799 static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
1800 struct walk_control *wc)
1802 struct btrfs_key key;
1806 struct cache_tree *inode_cache;
1807 struct shared_node *active_node;
1809 if (wc->root_level == wc->active_node &&
1810 btrfs_root_refs(&root->root_item) == 0)
1813 active_node = wc->nodes[wc->active_node];
1814 inode_cache = &active_node->inode_cache;
1815 nritems = btrfs_header_nritems(eb);
1816 for (i = 0; i < nritems; i++) {
1817 btrfs_item_key_to_cpu(eb, &key, i);
1819 if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
1821 if (key.type == BTRFS_ORPHAN_ITEM_KEY)
1824 if (active_node->current == NULL ||
1825 active_node->current->ino < key.objectid) {
1826 if (active_node->current) {
1827 active_node->current->checked = 1;
1828 maybe_free_inode_rec(inode_cache,
1829 active_node->current);
1831 active_node->current = get_inode_rec(inode_cache,
1833 BUG_ON(IS_ERR(active_node->current));
1836 case BTRFS_DIR_ITEM_KEY:
1837 case BTRFS_DIR_INDEX_KEY:
1838 ret = process_dir_item(eb, i, &key, active_node);
1840 case BTRFS_INODE_REF_KEY:
1841 ret = process_inode_ref(eb, i, &key, active_node);
1843 case BTRFS_INODE_EXTREF_KEY:
1844 ret = process_inode_extref(eb, i, &key, active_node);
1846 case BTRFS_INODE_ITEM_KEY:
1847 ret = process_inode_item(eb, i, &key, active_node);
1849 case BTRFS_EXTENT_DATA_KEY:
1850 ret = process_file_extent(root, eb, i, &key,
1861 u64 bytenr[BTRFS_MAX_LEVEL];
1862 u64 refs[BTRFS_MAX_LEVEL];
1863 int need_check[BTRFS_MAX_LEVEL];
1866 static int update_nodes_refs(struct btrfs_root *root, u64 bytenr,
1867 struct node_refs *nrefs, u64 level);
1868 static int check_inode_item(struct btrfs_root *root, struct btrfs_path *path,
1869 unsigned int ext_ref);
1871 static int process_one_leaf_v2(struct btrfs_root *root, struct btrfs_path *path,
1872 struct node_refs *nrefs, int *level, int ext_ref)
1874 struct extent_buffer *cur = path->nodes[0];
1875 struct btrfs_key key;
1879 int root_level = btrfs_header_level(root->node);
1881 int ret = 0; /* Final return value */
1882 int err = 0; /* Positive error bitmap */
1884 cur_bytenr = cur->start;
1886 /* skip to first inode item or the first inode number change */
1887 nritems = btrfs_header_nritems(cur);
1888 for (i = 0; i < nritems; i++) {
1889 btrfs_item_key_to_cpu(cur, &key, i);
1891 first_ino = key.objectid;
1892 if (key.type == BTRFS_INODE_ITEM_KEY ||
1893 (first_ino && first_ino != key.objectid))
1897 path->slots[0] = nritems;
1903 err |= check_inode_item(root, path, ext_ref);
1905 if (err & LAST_ITEM)
1908 /* still have inode items in thie leaf */
1909 if (cur->start == cur_bytenr)
1913 * we have switched to another leaf, above nodes may
1914 * have changed, here walk down the path, if a node
1915 * or leaf is shared, check whether we can skip this
1918 for (i = root_level; i >= 0; i--) {
1919 if (path->nodes[i]->start == nrefs->bytenr[i])
1922 ret = update_nodes_refs(root,
1923 path->nodes[i]->start,
1928 if (!nrefs->need_check[i]) {
1934 for (i = 0; i < *level; i++) {
1935 free_extent_buffer(path->nodes[i]);
1936 path->nodes[i] = NULL;
1941 * Convert any error bitmap to -EIO, as we should avoid
1942 * mixing positive and negative return value to represent
1950 static void reada_walk_down(struct btrfs_root *root,
1951 struct extent_buffer *node, int slot)
1960 level = btrfs_header_level(node);
1964 nritems = btrfs_header_nritems(node);
1965 blocksize = root->nodesize;
1966 for (i = slot; i < nritems; i++) {
1967 bytenr = btrfs_node_blockptr(node, i);
1968 ptr_gen = btrfs_node_ptr_generation(node, i);
1969 readahead_tree_block(root, bytenr, blocksize, ptr_gen);
1974 * Check the child node/leaf by the following condition:
1975 * 1. the first item key of the node/leaf should be the same with the one
1977 * 2. block in parent node should match the child node/leaf.
1978 * 3. generation of parent node and child's header should be consistent.
1980 * Or the child node/leaf pointed by the key in parent is not valid.
1982 * We hope to check leaf owner too, but since subvol may share leaves,
1983 * which makes leaf owner check not so strong, key check should be
1984 * sufficient enough for that case.
1986 static int check_child_node(struct extent_buffer *parent, int slot,
1987 struct extent_buffer *child)
1989 struct btrfs_key parent_key;
1990 struct btrfs_key child_key;
1993 btrfs_node_key_to_cpu(parent, &parent_key, slot);
1994 if (btrfs_header_level(child) == 0)
1995 btrfs_item_key_to_cpu(child, &child_key, 0);
1997 btrfs_node_key_to_cpu(child, &child_key, 0);
1999 if (memcmp(&parent_key, &child_key, sizeof(parent_key))) {
2002 "Wrong key of child node/leaf, wanted: (%llu, %u, %llu), have: (%llu, %u, %llu)\n",
2003 parent_key.objectid, parent_key.type, parent_key.offset,
2004 child_key.objectid, child_key.type, child_key.offset);
2006 if (btrfs_header_bytenr(child) != btrfs_node_blockptr(parent, slot)) {
2008 fprintf(stderr, "Wrong block of child node/leaf, wanted: %llu, have: %llu\n",
2009 btrfs_node_blockptr(parent, slot),
2010 btrfs_header_bytenr(child));
2012 if (btrfs_node_ptr_generation(parent, slot) !=
2013 btrfs_header_generation(child)) {
2015 fprintf(stderr, "Wrong generation of child node/leaf, wanted: %llu, have: %llu\n",
2016 btrfs_header_generation(child),
2017 btrfs_node_ptr_generation(parent, slot));
2023 * for a tree node or leaf, if it's shared, indeed we don't need to iterate it
2024 * in every fs or file tree check. Here we find its all root ids, and only check
2025 * it in the fs or file tree which has the smallest root id.
2027 static int need_check(struct btrfs_root *root, struct ulist *roots)
2029 struct rb_node *node;
2030 struct ulist_node *u;
2032 if (roots->nnodes == 1)
2035 node = rb_first(&roots->root);
2036 u = rb_entry(node, struct ulist_node, rb_node);
2038 * current root id is not smallest, we skip it and let it be checked
2039 * in the fs or file tree who hash the smallest root id.
2041 if (root->objectid != u->val)
2048 * for a tree node or leaf, we record its reference count, so later if we still
2049 * process this node or leaf, don't need to compute its reference count again.
2051 static int update_nodes_refs(struct btrfs_root *root, u64 bytenr,
2052 struct node_refs *nrefs, u64 level)
2056 struct ulist *roots;
2058 if (nrefs->bytenr[level] != bytenr) {
2059 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
2060 level, 1, &refs, NULL);
2064 nrefs->bytenr[level] = bytenr;
2065 nrefs->refs[level] = refs;
2067 ret = btrfs_find_all_roots(NULL, root->fs_info, bytenr,
2072 check = need_check(root, roots);
2074 nrefs->need_check[level] = check;
2076 nrefs->need_check[level] = 1;
2083 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
2084 struct walk_control *wc, int *level,
2085 struct node_refs *nrefs)
2087 enum btrfs_tree_block_status status;
2090 struct extent_buffer *next;
2091 struct extent_buffer *cur;
2096 WARN_ON(*level < 0);
2097 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2099 if (path->nodes[*level]->start == nrefs->bytenr[*level]) {
2100 refs = nrefs->refs[*level];
2103 ret = btrfs_lookup_extent_info(NULL, root,
2104 path->nodes[*level]->start,
2105 *level, 1, &refs, NULL);
2110 nrefs->bytenr[*level] = path->nodes[*level]->start;
2111 nrefs->refs[*level] = refs;
2115 ret = enter_shared_node(root, path->nodes[*level]->start,
2123 while (*level >= 0) {
2124 WARN_ON(*level < 0);
2125 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2126 cur = path->nodes[*level];
2128 if (btrfs_header_level(cur) != *level)
2131 if (path->slots[*level] >= btrfs_header_nritems(cur))
2134 ret = process_one_leaf(root, cur, wc);
2139 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2140 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2141 blocksize = root->nodesize;
2143 if (bytenr == nrefs->bytenr[*level - 1]) {
2144 refs = nrefs->refs[*level - 1];
2146 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
2147 *level - 1, 1, &refs, NULL);
2151 nrefs->bytenr[*level - 1] = bytenr;
2152 nrefs->refs[*level - 1] = refs;
2157 ret = enter_shared_node(root, bytenr, refs,
2160 path->slots[*level]++;
2165 next = btrfs_find_tree_block(root, bytenr, blocksize);
2166 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2167 free_extent_buffer(next);
2168 reada_walk_down(root, cur, path->slots[*level]);
2169 next = read_tree_block(root, bytenr, blocksize,
2171 if (!extent_buffer_uptodate(next)) {
2172 struct btrfs_key node_key;
2174 btrfs_node_key_to_cpu(path->nodes[*level],
2176 path->slots[*level]);
2177 btrfs_add_corrupt_extent_record(root->fs_info,
2179 path->nodes[*level]->start,
2180 root->nodesize, *level);
2186 ret = check_child_node(cur, path->slots[*level], next);
2192 if (btrfs_is_leaf(next))
2193 status = btrfs_check_leaf(root, NULL, next);
2195 status = btrfs_check_node(root, NULL, next);
2196 if (status != BTRFS_TREE_BLOCK_CLEAN) {
2197 free_extent_buffer(next);
2202 *level = *level - 1;
2203 free_extent_buffer(path->nodes[*level]);
2204 path->nodes[*level] = next;
2205 path->slots[*level] = 0;
2208 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2212 static int check_inode_item(struct btrfs_root *root, struct btrfs_path *path,
2213 unsigned int ext_ref);
2215 static int walk_down_tree_v2(struct btrfs_root *root, struct btrfs_path *path,
2216 int *level, struct node_refs *nrefs, int ext_ref)
2218 enum btrfs_tree_block_status status;
2221 struct extent_buffer *next;
2222 struct extent_buffer *cur;
2226 WARN_ON(*level < 0);
2227 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2229 ret = update_nodes_refs(root, path->nodes[*level]->start,
2234 while (*level >= 0) {
2235 WARN_ON(*level < 0);
2236 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2237 cur = path->nodes[*level];
2239 if (btrfs_header_level(cur) != *level)
2242 if (path->slots[*level] >= btrfs_header_nritems(cur))
2244 /* Don't forgot to check leaf/node validation */
2246 ret = btrfs_check_leaf(root, NULL, cur);
2247 if (ret != BTRFS_TREE_BLOCK_CLEAN) {
2251 ret = process_one_leaf_v2(root, path, nrefs,
2255 ret = btrfs_check_node(root, NULL, cur);
2256 if (ret != BTRFS_TREE_BLOCK_CLEAN) {
2261 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2262 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2263 blocksize = root->nodesize;
2265 ret = update_nodes_refs(root, bytenr, nrefs, *level - 1);
2268 if (!nrefs->need_check[*level - 1]) {
2269 path->slots[*level]++;
2273 next = btrfs_find_tree_block(root, bytenr, blocksize);
2274 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2275 free_extent_buffer(next);
2276 reada_walk_down(root, cur, path->slots[*level]);
2277 next = read_tree_block(root, bytenr, blocksize,
2279 if (!extent_buffer_uptodate(next)) {
2280 struct btrfs_key node_key;
2282 btrfs_node_key_to_cpu(path->nodes[*level],
2284 path->slots[*level]);
2285 btrfs_add_corrupt_extent_record(root->fs_info,
2287 path->nodes[*level]->start,
2288 root->nodesize, *level);
2294 ret = check_child_node(cur, path->slots[*level], next);
2298 if (btrfs_is_leaf(next))
2299 status = btrfs_check_leaf(root, NULL, next);
2301 status = btrfs_check_node(root, NULL, next);
2302 if (status != BTRFS_TREE_BLOCK_CLEAN) {
2303 free_extent_buffer(next);
2308 *level = *level - 1;
2309 free_extent_buffer(path->nodes[*level]);
2310 path->nodes[*level] = next;
2311 path->slots[*level] = 0;
2316 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
2317 struct walk_control *wc, int *level)
2320 struct extent_buffer *leaf;
2322 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2323 leaf = path->nodes[i];
2324 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
2329 free_extent_buffer(path->nodes[*level]);
2330 path->nodes[*level] = NULL;
2331 BUG_ON(*level > wc->active_node);
2332 if (*level == wc->active_node)
2333 leave_shared_node(root, wc, *level);
2340 static int walk_up_tree_v2(struct btrfs_root *root, struct btrfs_path *path,
2344 struct extent_buffer *leaf;
2346 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2347 leaf = path->nodes[i];
2348 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
2353 free_extent_buffer(path->nodes[*level]);
2354 path->nodes[*level] = NULL;
2361 static int check_root_dir(struct inode_record *rec)
2363 struct inode_backref *backref;
2366 if (!rec->found_inode_item || rec->errors)
2368 if (rec->nlink != 1 || rec->found_link != 0)
2370 if (list_empty(&rec->backrefs))
2372 backref = to_inode_backref(rec->backrefs.next);
2373 if (!backref->found_inode_ref)
2375 if (backref->index != 0 || backref->namelen != 2 ||
2376 memcmp(backref->name, "..", 2))
2378 if (backref->found_dir_index || backref->found_dir_item)
2385 static int repair_inode_isize(struct btrfs_trans_handle *trans,
2386 struct btrfs_root *root, struct btrfs_path *path,
2387 struct inode_record *rec)
2389 struct btrfs_inode_item *ei;
2390 struct btrfs_key key;
2393 key.objectid = rec->ino;
2394 key.type = BTRFS_INODE_ITEM_KEY;
2395 key.offset = (u64)-1;
2397 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2401 if (!path->slots[0]) {
2408 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2409 if (key.objectid != rec->ino) {
2414 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2415 struct btrfs_inode_item);
2416 btrfs_set_inode_size(path->nodes[0], ei, rec->found_size);
2417 btrfs_mark_buffer_dirty(path->nodes[0]);
2418 rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
2419 printf("reset isize for dir %Lu root %Lu\n", rec->ino,
2420 root->root_key.objectid);
2422 btrfs_release_path(path);
2426 static int repair_inode_orphan_item(struct btrfs_trans_handle *trans,
2427 struct btrfs_root *root,
2428 struct btrfs_path *path,
2429 struct inode_record *rec)
2433 ret = btrfs_add_orphan_item(trans, root, path, rec->ino);
2434 btrfs_release_path(path);
2436 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
2440 static int repair_inode_nbytes(struct btrfs_trans_handle *trans,
2441 struct btrfs_root *root,
2442 struct btrfs_path *path,
2443 struct inode_record *rec)
2445 struct btrfs_inode_item *ei;
2446 struct btrfs_key key;
2449 key.objectid = rec->ino;
2450 key.type = BTRFS_INODE_ITEM_KEY;
2453 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2460 /* Since ret == 0, no need to check anything */
2461 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2462 struct btrfs_inode_item);
2463 btrfs_set_inode_nbytes(path->nodes[0], ei, rec->found_size);
2464 btrfs_mark_buffer_dirty(path->nodes[0]);
2465 rec->errors &= ~I_ERR_FILE_NBYTES_WRONG;
2466 printf("reset nbytes for ino %llu root %llu\n",
2467 rec->ino, root->root_key.objectid);
2469 btrfs_release_path(path);
2473 static int add_missing_dir_index(struct btrfs_root *root,
2474 struct cache_tree *inode_cache,
2475 struct inode_record *rec,
2476 struct inode_backref *backref)
2478 struct btrfs_path path;
2479 struct btrfs_trans_handle *trans;
2480 struct btrfs_dir_item *dir_item;
2481 struct extent_buffer *leaf;
2482 struct btrfs_key key;
2483 struct btrfs_disk_key disk_key;
2484 struct inode_record *dir_rec;
2485 unsigned long name_ptr;
2486 u32 data_size = sizeof(*dir_item) + backref->namelen;
2489 trans = btrfs_start_transaction(root, 1);
2491 return PTR_ERR(trans);
2493 fprintf(stderr, "repairing missing dir index item for inode %llu\n",
2494 (unsigned long long)rec->ino);
2496 btrfs_init_path(&path);
2497 key.objectid = backref->dir;
2498 key.type = BTRFS_DIR_INDEX_KEY;
2499 key.offset = backref->index;
2500 ret = btrfs_insert_empty_item(trans, root, &path, &key, data_size);
2503 leaf = path.nodes[0];
2504 dir_item = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_dir_item);
2506 disk_key.objectid = cpu_to_le64(rec->ino);
2507 disk_key.type = BTRFS_INODE_ITEM_KEY;
2508 disk_key.offset = 0;
2510 btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
2511 btrfs_set_dir_type(leaf, dir_item, imode_to_type(rec->imode));
2512 btrfs_set_dir_data_len(leaf, dir_item, 0);
2513 btrfs_set_dir_name_len(leaf, dir_item, backref->namelen);
2514 name_ptr = (unsigned long)(dir_item + 1);
2515 write_extent_buffer(leaf, backref->name, name_ptr, backref->namelen);
2516 btrfs_mark_buffer_dirty(leaf);
2517 btrfs_release_path(&path);
2518 btrfs_commit_transaction(trans, root);
2520 backref->found_dir_index = 1;
2521 dir_rec = get_inode_rec(inode_cache, backref->dir, 0);
2522 BUG_ON(IS_ERR(dir_rec));
2525 dir_rec->found_size += backref->namelen;
2526 if (dir_rec->found_size == dir_rec->isize &&
2527 (dir_rec->errors & I_ERR_DIR_ISIZE_WRONG))
2528 dir_rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
2529 if (dir_rec->found_size != dir_rec->isize)
2530 dir_rec->errors |= I_ERR_DIR_ISIZE_WRONG;
2535 static int delete_dir_index(struct btrfs_root *root,
2536 struct inode_backref *backref)
2538 struct btrfs_trans_handle *trans;
2539 struct btrfs_dir_item *di;
2540 struct btrfs_path path;
2543 trans = btrfs_start_transaction(root, 1);
2545 return PTR_ERR(trans);
2547 fprintf(stderr, "Deleting bad dir index [%llu,%u,%llu] root %llu\n",
2548 (unsigned long long)backref->dir,
2549 BTRFS_DIR_INDEX_KEY, (unsigned long long)backref->index,
2550 (unsigned long long)root->objectid);
2552 btrfs_init_path(&path);
2553 di = btrfs_lookup_dir_index(trans, root, &path, backref->dir,
2554 backref->name, backref->namelen,
2555 backref->index, -1);
2558 btrfs_release_path(&path);
2559 btrfs_commit_transaction(trans, root);
2566 ret = btrfs_del_item(trans, root, &path);
2568 ret = btrfs_delete_one_dir_name(trans, root, &path, di);
2570 btrfs_release_path(&path);
2571 btrfs_commit_transaction(trans, root);
2575 static int create_inode_item(struct btrfs_root *root,
2576 struct inode_record *rec,
2579 struct btrfs_trans_handle *trans;
2580 struct btrfs_inode_item inode_item;
2581 time_t now = time(NULL);
2584 trans = btrfs_start_transaction(root, 1);
2585 if (IS_ERR(trans)) {
2586 ret = PTR_ERR(trans);
2590 fprintf(stderr, "root %llu inode %llu recreating inode item, this may "
2591 "be incomplete, please check permissions and content after "
2592 "the fsck completes.\n", (unsigned long long)root->objectid,
2593 (unsigned long long)rec->ino);
2595 memset(&inode_item, 0, sizeof(inode_item));
2596 btrfs_set_stack_inode_generation(&inode_item, trans->transid);
2598 btrfs_set_stack_inode_nlink(&inode_item, 1);
2600 btrfs_set_stack_inode_nlink(&inode_item, rec->found_link);
2601 btrfs_set_stack_inode_nbytes(&inode_item, rec->found_size);
2602 if (rec->found_dir_item) {
2603 if (rec->found_file_extent)
2604 fprintf(stderr, "root %llu inode %llu has both a dir "
2605 "item and extents, unsure if it is a dir or a "
2606 "regular file so setting it as a directory\n",
2607 (unsigned long long)root->objectid,
2608 (unsigned long long)rec->ino);
2609 btrfs_set_stack_inode_mode(&inode_item, S_IFDIR | 0755);
2610 btrfs_set_stack_inode_size(&inode_item, rec->found_size);
2611 } else if (!rec->found_dir_item) {
2612 btrfs_set_stack_inode_size(&inode_item, rec->extent_end);
2613 btrfs_set_stack_inode_mode(&inode_item, S_IFREG | 0755);
2615 btrfs_set_stack_timespec_sec(&inode_item.atime, now);
2616 btrfs_set_stack_timespec_nsec(&inode_item.atime, 0);
2617 btrfs_set_stack_timespec_sec(&inode_item.ctime, now);
2618 btrfs_set_stack_timespec_nsec(&inode_item.ctime, 0);
2619 btrfs_set_stack_timespec_sec(&inode_item.mtime, now);
2620 btrfs_set_stack_timespec_nsec(&inode_item.mtime, 0);
2621 btrfs_set_stack_timespec_sec(&inode_item.otime, 0);
2622 btrfs_set_stack_timespec_nsec(&inode_item.otime, 0);
2624 ret = btrfs_insert_inode(trans, root, rec->ino, &inode_item);
2626 btrfs_commit_transaction(trans, root);
2630 static int repair_inode_backrefs(struct btrfs_root *root,
2631 struct inode_record *rec,
2632 struct cache_tree *inode_cache,
2635 struct inode_backref *tmp, *backref;
2636 u64 root_dirid = btrfs_root_dirid(&root->root_item);
2640 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
2641 if (!delete && rec->ino == root_dirid) {
2642 if (!rec->found_inode_item) {
2643 ret = create_inode_item(root, rec, 1);
2650 /* Index 0 for root dir's are special, don't mess with it */
2651 if (rec->ino == root_dirid && backref->index == 0)
2655 ((backref->found_dir_index && !backref->found_inode_ref) ||
2656 (backref->found_dir_index && backref->found_inode_ref &&
2657 (backref->errors & REF_ERR_INDEX_UNMATCH)))) {
2658 ret = delete_dir_index(root, backref);
2662 list_del(&backref->list);
2666 if (!delete && !backref->found_dir_index &&
2667 backref->found_dir_item && backref->found_inode_ref) {
2668 ret = add_missing_dir_index(root, inode_cache, rec,
2673 if (backref->found_dir_item &&
2674 backref->found_dir_index &&
2675 backref->found_dir_index) {
2676 if (!backref->errors &&
2677 backref->found_inode_ref) {
2678 list_del(&backref->list);
2684 if (!delete && (!backref->found_dir_index &&
2685 !backref->found_dir_item &&
2686 backref->found_inode_ref)) {
2687 struct btrfs_trans_handle *trans;
2688 struct btrfs_key location;
2690 ret = check_dir_conflict(root, backref->name,
2696 * let nlink fixing routine to handle it,
2697 * which can do it better.
2702 location.objectid = rec->ino;
2703 location.type = BTRFS_INODE_ITEM_KEY;
2704 location.offset = 0;
2706 trans = btrfs_start_transaction(root, 1);
2707 if (IS_ERR(trans)) {
2708 ret = PTR_ERR(trans);
2711 fprintf(stderr, "adding missing dir index/item pair "
2713 (unsigned long long)rec->ino);
2714 ret = btrfs_insert_dir_item(trans, root, backref->name,
2716 backref->dir, &location,
2717 imode_to_type(rec->imode),
2720 btrfs_commit_transaction(trans, root);
2724 if (!delete && (backref->found_inode_ref &&
2725 backref->found_dir_index &&
2726 backref->found_dir_item &&
2727 !(backref->errors & REF_ERR_INDEX_UNMATCH) &&
2728 !rec->found_inode_item)) {
2729 ret = create_inode_item(root, rec, 0);
2736 return ret ? ret : repaired;
2740 * To determine the file type for nlink/inode_item repair
2742 * Return 0 if file type is found and BTRFS_FT_* is stored into type.
2743 * Return -ENOENT if file type is not found.
2745 static int find_file_type(struct inode_record *rec, u8 *type)
2747 struct inode_backref *backref;
2749 /* For inode item recovered case */
2750 if (rec->found_inode_item) {
2751 *type = imode_to_type(rec->imode);
2755 list_for_each_entry(backref, &rec->backrefs, list) {
2756 if (backref->found_dir_index || backref->found_dir_item) {
2757 *type = backref->filetype;
2765 * To determine the file name for nlink repair
2767 * Return 0 if file name is found, set name and namelen.
2768 * Return -ENOENT if file name is not found.
2770 static int find_file_name(struct inode_record *rec,
2771 char *name, int *namelen)
2773 struct inode_backref *backref;
2775 list_for_each_entry(backref, &rec->backrefs, list) {
2776 if (backref->found_dir_index || backref->found_dir_item ||
2777 backref->found_inode_ref) {
2778 memcpy(name, backref->name, backref->namelen);
2779 *namelen = backref->namelen;
2786 /* Reset the nlink of the inode to the correct one */
2787 static int reset_nlink(struct btrfs_trans_handle *trans,
2788 struct btrfs_root *root,
2789 struct btrfs_path *path,
2790 struct inode_record *rec)
2792 struct inode_backref *backref;
2793 struct inode_backref *tmp;
2794 struct btrfs_key key;
2795 struct btrfs_inode_item *inode_item;
2798 /* We don't believe this either, reset it and iterate backref */
2799 rec->found_link = 0;
2801 /* Remove all backref including the valid ones */
2802 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
2803 ret = btrfs_unlink(trans, root, rec->ino, backref->dir,
2804 backref->index, backref->name,
2805 backref->namelen, 0);
2809 /* remove invalid backref, so it won't be added back */
2810 if (!(backref->found_dir_index &&
2811 backref->found_dir_item &&
2812 backref->found_inode_ref)) {
2813 list_del(&backref->list);
2820 /* Set nlink to 0 */
2821 key.objectid = rec->ino;
2822 key.type = BTRFS_INODE_ITEM_KEY;
2824 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2831 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2832 struct btrfs_inode_item);
2833 btrfs_set_inode_nlink(path->nodes[0], inode_item, 0);
2834 btrfs_mark_buffer_dirty(path->nodes[0]);
2835 btrfs_release_path(path);
2838 * Add back valid inode_ref/dir_item/dir_index,
2839 * add_link() will handle the nlink inc, so new nlink must be correct
2841 list_for_each_entry(backref, &rec->backrefs, list) {
2842 ret = btrfs_add_link(trans, root, rec->ino, backref->dir,
2843 backref->name, backref->namelen,
2844 backref->filetype, &backref->index, 1);
2849 btrfs_release_path(path);
2853 static int get_highest_inode(struct btrfs_trans_handle *trans,
2854 struct btrfs_root *root,
2855 struct btrfs_path *path,
2858 struct btrfs_key key, found_key;
2861 btrfs_init_path(path);
2862 key.objectid = BTRFS_LAST_FREE_OBJECTID;
2864 key.type = BTRFS_INODE_ITEM_KEY;
2865 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2867 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2868 path->slots[0] - 1);
2869 *highest_ino = found_key.objectid;
2872 if (*highest_ino >= BTRFS_LAST_FREE_OBJECTID)
2874 btrfs_release_path(path);
2878 static int repair_inode_nlinks(struct btrfs_trans_handle *trans,
2879 struct btrfs_root *root,
2880 struct btrfs_path *path,
2881 struct inode_record *rec)
2883 char *dir_name = "lost+found";
2884 char namebuf[BTRFS_NAME_LEN] = {0};
2889 int name_recovered = 0;
2890 int type_recovered = 0;
2894 * Get file name and type first before these invalid inode ref
2895 * are deleted by remove_all_invalid_backref()
2897 name_recovered = !find_file_name(rec, namebuf, &namelen);
2898 type_recovered = !find_file_type(rec, &type);
2900 if (!name_recovered) {
2901 printf("Can't get file name for inode %llu, using '%llu' as fallback\n",
2902 rec->ino, rec->ino);
2903 namelen = count_digits(rec->ino);
2904 sprintf(namebuf, "%llu", rec->ino);
2907 if (!type_recovered) {
2908 printf("Can't get file type for inode %llu, using FILE as fallback\n",
2910 type = BTRFS_FT_REG_FILE;
2914 ret = reset_nlink(trans, root, path, rec);
2917 "Failed to reset nlink for inode %llu: %s\n",
2918 rec->ino, strerror(-ret));
2922 if (rec->found_link == 0) {
2923 ret = get_highest_inode(trans, root, path, &lost_found_ino);
2927 ret = btrfs_mkdir(trans, root, dir_name, strlen(dir_name),
2928 BTRFS_FIRST_FREE_OBJECTID, &lost_found_ino,
2931 fprintf(stderr, "Failed to create '%s' dir: %s\n",
2932 dir_name, strerror(-ret));
2935 ret = btrfs_add_link(trans, root, rec->ino, lost_found_ino,
2936 namebuf, namelen, type, NULL, 1);
2938 * Add ".INO" suffix several times to handle case where
2939 * "FILENAME.INO" is already taken by another file.
2941 while (ret == -EEXIST) {
2943 * Conflicting file name, add ".INO" as suffix * +1 for '.'
2945 if (namelen + count_digits(rec->ino) + 1 >
2950 snprintf(namebuf + namelen, BTRFS_NAME_LEN - namelen,
2952 namelen += count_digits(rec->ino) + 1;
2953 ret = btrfs_add_link(trans, root, rec->ino,
2954 lost_found_ino, namebuf,
2955 namelen, type, NULL, 1);
2959 "Failed to link the inode %llu to %s dir: %s\n",
2960 rec->ino, dir_name, strerror(-ret));
2964 * Just increase the found_link, don't actually add the
2965 * backref. This will make things easier and this inode
2966 * record will be freed after the repair is done.
2967 * So fsck will not report problem about this inode.
2970 printf("Moving file '%.*s' to '%s' dir since it has no valid backref\n",
2971 namelen, namebuf, dir_name);
2973 printf("Fixed the nlink of inode %llu\n", rec->ino);
2976 * Clear the flag anyway, or we will loop forever for the same inode
2977 * as it will not be removed from the bad inode list and the dead loop
2980 rec->errors &= ~I_ERR_LINK_COUNT_WRONG;
2981 btrfs_release_path(path);
2986 * Check if there is any normal(reg or prealloc) file extent for given
2988 * This is used to determine the file type when neither its dir_index/item or
2989 * inode_item exists.
2991 * This will *NOT* report error, if any error happens, just consider it does
2992 * not have any normal file extent.
2994 static int find_normal_file_extent(struct btrfs_root *root, u64 ino)
2996 struct btrfs_path path;
2997 struct btrfs_key key;
2998 struct btrfs_key found_key;
2999 struct btrfs_file_extent_item *fi;
3003 btrfs_init_path(&path);
3005 key.type = BTRFS_EXTENT_DATA_KEY;
3008 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
3013 if (ret && path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
3014 ret = btrfs_next_leaf(root, &path);
3021 btrfs_item_key_to_cpu(path.nodes[0], &found_key,
3023 if (found_key.objectid != ino ||
3024 found_key.type != BTRFS_EXTENT_DATA_KEY)
3026 fi = btrfs_item_ptr(path.nodes[0], path.slots[0],
3027 struct btrfs_file_extent_item);
3028 type = btrfs_file_extent_type(path.nodes[0], fi);
3029 if (type != BTRFS_FILE_EXTENT_INLINE) {
3035 btrfs_release_path(&path);
3039 static u32 btrfs_type_to_imode(u8 type)
3041 static u32 imode_by_btrfs_type[] = {
3042 [BTRFS_FT_REG_FILE] = S_IFREG,
3043 [BTRFS_FT_DIR] = S_IFDIR,
3044 [BTRFS_FT_CHRDEV] = S_IFCHR,
3045 [BTRFS_FT_BLKDEV] = S_IFBLK,
3046 [BTRFS_FT_FIFO] = S_IFIFO,
3047 [BTRFS_FT_SOCK] = S_IFSOCK,
3048 [BTRFS_FT_SYMLINK] = S_IFLNK,
3051 return imode_by_btrfs_type[(type)];
3054 static int repair_inode_no_item(struct btrfs_trans_handle *trans,
3055 struct btrfs_root *root,
3056 struct btrfs_path *path,
3057 struct inode_record *rec)
3061 int type_recovered = 0;
3064 printf("Trying to rebuild inode:%llu\n", rec->ino);
3066 type_recovered = !find_file_type(rec, &filetype);
3069 * Try to determine inode type if type not found.
3071 * For found regular file extent, it must be FILE.
3072 * For found dir_item/index, it must be DIR.
3074 * For undetermined one, use FILE as fallback.
3077 * 1. If found backref(inode_index/item is already handled) to it,
3079 * Need new inode-inode ref structure to allow search for that.
3081 if (!type_recovered) {
3082 if (rec->found_file_extent &&
3083 find_normal_file_extent(root, rec->ino)) {
3085 filetype = BTRFS_FT_REG_FILE;
3086 } else if (rec->found_dir_item) {
3088 filetype = BTRFS_FT_DIR;
3089 } else if (!list_empty(&rec->orphan_extents)) {
3091 filetype = BTRFS_FT_REG_FILE;
3093 printf("Can't determine the filetype for inode %llu, assume it is a normal file\n",
3096 filetype = BTRFS_FT_REG_FILE;
3100 ret = btrfs_new_inode(trans, root, rec->ino,
3101 mode | btrfs_type_to_imode(filetype));
3106 * Here inode rebuild is done, we only rebuild the inode item,
3107 * don't repair the nlink(like move to lost+found).
3108 * That is the job of nlink repair.
3110 * We just fill the record and return
3112 rec->found_dir_item = 1;
3113 rec->imode = mode | btrfs_type_to_imode(filetype);
3115 rec->errors &= ~I_ERR_NO_INODE_ITEM;
3116 /* Ensure the inode_nlinks repair function will be called */
3117 rec->errors |= I_ERR_LINK_COUNT_WRONG;
3122 static int repair_inode_orphan_extent(struct btrfs_trans_handle *trans,
3123 struct btrfs_root *root,
3124 struct btrfs_path *path,
3125 struct inode_record *rec)
3127 struct orphan_data_extent *orphan;
3128 struct orphan_data_extent *tmp;
3131 list_for_each_entry_safe(orphan, tmp, &rec->orphan_extents, list) {
3133 * Check for conflicting file extents
3135 * Here we don't know whether the extents is compressed or not,
3136 * so we can only assume it not compressed nor data offset,
3137 * and use its disk_len as extent length.
3139 ret = btrfs_get_extent(NULL, root, path, orphan->objectid,
3140 orphan->offset, orphan->disk_len, 0);
3141 btrfs_release_path(path);
3146 "orphan extent (%llu, %llu) conflicts, delete the orphan\n",
3147 orphan->disk_bytenr, orphan->disk_len);
3148 ret = btrfs_free_extent(trans,
3149 root->fs_info->extent_root,
3150 orphan->disk_bytenr, orphan->disk_len,
3151 0, root->objectid, orphan->objectid,
3156 ret = btrfs_insert_file_extent(trans, root, orphan->objectid,
3157 orphan->offset, orphan->disk_bytenr,
3158 orphan->disk_len, orphan->disk_len);
3162 /* Update file size info */
3163 rec->found_size += orphan->disk_len;
3164 if (rec->found_size == rec->nbytes)
3165 rec->errors &= ~I_ERR_FILE_NBYTES_WRONG;
3167 /* Update the file extent hole info too */
3168 ret = del_file_extent_hole(&rec->holes, orphan->offset,
3172 if (RB_EMPTY_ROOT(&rec->holes))
3173 rec->errors &= ~I_ERR_FILE_EXTENT_DISCOUNT;
3175 list_del(&orphan->list);
3178 rec->errors &= ~I_ERR_FILE_EXTENT_ORPHAN;
3183 static int repair_inode_discount_extent(struct btrfs_trans_handle *trans,
3184 struct btrfs_root *root,
3185 struct btrfs_path *path,
3186 struct inode_record *rec)
3188 struct rb_node *node;
3189 struct file_extent_hole *hole;
3193 node = rb_first(&rec->holes);
3197 hole = rb_entry(node, struct file_extent_hole, node);
3198 ret = btrfs_punch_hole(trans, root, rec->ino,
3199 hole->start, hole->len);
3202 ret = del_file_extent_hole(&rec->holes, hole->start,
3206 if (RB_EMPTY_ROOT(&rec->holes))
3207 rec->errors &= ~I_ERR_FILE_EXTENT_DISCOUNT;
3208 node = rb_first(&rec->holes);
3210 /* special case for a file losing all its file extent */
3212 ret = btrfs_punch_hole(trans, root, rec->ino, 0,
3213 round_up(rec->isize, root->sectorsize));
3217 printf("Fixed discount file extents for inode: %llu in root: %llu\n",
3218 rec->ino, root->objectid);
3223 static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
3225 struct btrfs_trans_handle *trans;
3226 struct btrfs_path path;
3229 if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG |
3230 I_ERR_NO_ORPHAN_ITEM |
3231 I_ERR_LINK_COUNT_WRONG |
3232 I_ERR_NO_INODE_ITEM |
3233 I_ERR_FILE_EXTENT_ORPHAN |
3234 I_ERR_FILE_EXTENT_DISCOUNT|
3235 I_ERR_FILE_NBYTES_WRONG)))
3239 * For nlink repair, it may create a dir and add link, so
3240 * 2 for parent(256)'s dir_index and dir_item
3241 * 2 for lost+found dir's inode_item and inode_ref
3242 * 1 for the new inode_ref of the file
3243 * 2 for lost+found dir's dir_index and dir_item for the file
3245 trans = btrfs_start_transaction(root, 7);
3247 return PTR_ERR(trans);
3249 btrfs_init_path(&path);
3250 if (rec->errors & I_ERR_NO_INODE_ITEM)
3251 ret = repair_inode_no_item(trans, root, &path, rec);
3252 if (!ret && rec->errors & I_ERR_FILE_EXTENT_ORPHAN)
3253 ret = repair_inode_orphan_extent(trans, root, &path, rec);
3254 if (!ret && rec->errors & I_ERR_FILE_EXTENT_DISCOUNT)
3255 ret = repair_inode_discount_extent(trans, root, &path, rec);
3256 if (!ret && rec->errors & I_ERR_DIR_ISIZE_WRONG)
3257 ret = repair_inode_isize(trans, root, &path, rec);
3258 if (!ret && rec->errors & I_ERR_NO_ORPHAN_ITEM)
3259 ret = repair_inode_orphan_item(trans, root, &path, rec);
3260 if (!ret && rec->errors & I_ERR_LINK_COUNT_WRONG)
3261 ret = repair_inode_nlinks(trans, root, &path, rec);
3262 if (!ret && rec->errors & I_ERR_FILE_NBYTES_WRONG)
3263 ret = repair_inode_nbytes(trans, root, &path, rec);
3264 btrfs_commit_transaction(trans, root);
3265 btrfs_release_path(&path);
3269 static int check_inode_recs(struct btrfs_root *root,
3270 struct cache_tree *inode_cache)
3272 struct cache_extent *cache;
3273 struct ptr_node *node;
3274 struct inode_record *rec;
3275 struct inode_backref *backref;
3280 u64 root_dirid = btrfs_root_dirid(&root->root_item);
3282 if (btrfs_root_refs(&root->root_item) == 0) {
3283 if (!cache_tree_empty(inode_cache))
3284 fprintf(stderr, "warning line %d\n", __LINE__);
3289 * We need to repair backrefs first because we could change some of the
3290 * errors in the inode recs.
3292 * We also need to go through and delete invalid backrefs first and then
3293 * add the correct ones second. We do this because we may get EEXIST
3294 * when adding back the correct index because we hadn't yet deleted the
3297 * For example, if we were missing a dir index then the directories
3298 * isize would be wrong, so if we fixed the isize to what we thought it
3299 * would be and then fixed the backref we'd still have a invalid fs, so
3300 * we need to add back the dir index and then check to see if the isize
3305 if (stage == 3 && !err)
3308 cache = search_cache_extent(inode_cache, 0);
3309 while (repair && cache) {
3310 node = container_of(cache, struct ptr_node, cache);
3312 cache = next_cache_extent(cache);
3314 /* Need to free everything up and rescan */
3316 remove_cache_extent(inode_cache, &node->cache);
3318 free_inode_rec(rec);
3322 if (list_empty(&rec->backrefs))
3325 ret = repair_inode_backrefs(root, rec, inode_cache,
3339 rec = get_inode_rec(inode_cache, root_dirid, 0);
3340 BUG_ON(IS_ERR(rec));
3342 ret = check_root_dir(rec);
3344 fprintf(stderr, "root %llu root dir %llu error\n",
3345 (unsigned long long)root->root_key.objectid,
3346 (unsigned long long)root_dirid);
3347 print_inode_error(root, rec);
3352 struct btrfs_trans_handle *trans;
3354 trans = btrfs_start_transaction(root, 1);
3355 if (IS_ERR(trans)) {
3356 err = PTR_ERR(trans);
3361 "root %llu missing its root dir, recreating\n",
3362 (unsigned long long)root->objectid);
3364 ret = btrfs_make_root_dir(trans, root, root_dirid);
3367 btrfs_commit_transaction(trans, root);
3371 fprintf(stderr, "root %llu root dir %llu not found\n",
3372 (unsigned long long)root->root_key.objectid,
3373 (unsigned long long)root_dirid);
3377 cache = search_cache_extent(inode_cache, 0);
3380 node = container_of(cache, struct ptr_node, cache);
3382 remove_cache_extent(inode_cache, &node->cache);
3384 if (rec->ino == root_dirid ||
3385 rec->ino == BTRFS_ORPHAN_OBJECTID) {
3386 free_inode_rec(rec);
3390 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
3391 ret = check_orphan_item(root, rec->ino);
3393 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
3394 if (can_free_inode_rec(rec)) {
3395 free_inode_rec(rec);
3400 if (!rec->found_inode_item)
3401 rec->errors |= I_ERR_NO_INODE_ITEM;
3402 if (rec->found_link != rec->nlink)
3403 rec->errors |= I_ERR_LINK_COUNT_WRONG;
3405 ret = try_repair_inode(root, rec);
3406 if (ret == 0 && can_free_inode_rec(rec)) {
3407 free_inode_rec(rec);
3413 if (!(repair && ret == 0))
3415 print_inode_error(root, rec);
3416 list_for_each_entry(backref, &rec->backrefs, list) {
3417 if (!backref->found_dir_item)
3418 backref->errors |= REF_ERR_NO_DIR_ITEM;
3419 if (!backref->found_dir_index)
3420 backref->errors |= REF_ERR_NO_DIR_INDEX;
3421 if (!backref->found_inode_ref)
3422 backref->errors |= REF_ERR_NO_INODE_REF;
3423 fprintf(stderr, "\tunresolved ref dir %llu index %llu"
3424 " namelen %u name %s filetype %d errors %x",
3425 (unsigned long long)backref->dir,
3426 (unsigned long long)backref->index,
3427 backref->namelen, backref->name,
3428 backref->filetype, backref->errors);
3429 print_ref_error(backref->errors);
3431 free_inode_rec(rec);
3433 return (error > 0) ? -1 : 0;
3436 static struct root_record *get_root_rec(struct cache_tree *root_cache,
3439 struct cache_extent *cache;
3440 struct root_record *rec = NULL;
3443 cache = lookup_cache_extent(root_cache, objectid, 1);
3445 rec = container_of(cache, struct root_record, cache);
3447 rec = calloc(1, sizeof(*rec));
3449 return ERR_PTR(-ENOMEM);
3450 rec->objectid = objectid;
3451 INIT_LIST_HEAD(&rec->backrefs);
3452 rec->cache.start = objectid;
3453 rec->cache.size = 1;
3455 ret = insert_cache_extent(root_cache, &rec->cache);
3457 return ERR_PTR(-EEXIST);
3462 static struct root_backref *get_root_backref(struct root_record *rec,
3463 u64 ref_root, u64 dir, u64 index,
3464 const char *name, int namelen)
3466 struct root_backref *backref;
3468 list_for_each_entry(backref, &rec->backrefs, list) {
3469 if (backref->ref_root != ref_root || backref->dir != dir ||
3470 backref->namelen != namelen)
3472 if (memcmp(name, backref->name, namelen))
3477 backref = calloc(1, sizeof(*backref) + namelen + 1);
3480 backref->ref_root = ref_root;
3482 backref->index = index;
3483 backref->namelen = namelen;
3484 memcpy(backref->name, name, namelen);
3485 backref->name[namelen] = '\0';
3486 list_add_tail(&backref->list, &rec->backrefs);
3490 static void free_root_record(struct cache_extent *cache)
3492 struct root_record *rec;
3493 struct root_backref *backref;
3495 rec = container_of(cache, struct root_record, cache);
3496 while (!list_empty(&rec->backrefs)) {
3497 backref = to_root_backref(rec->backrefs.next);
3498 list_del(&backref->list);
3505 FREE_EXTENT_CACHE_BASED_TREE(root_recs, free_root_record);
3507 static int add_root_backref(struct cache_tree *root_cache,
3508 u64 root_id, u64 ref_root, u64 dir, u64 index,
3509 const char *name, int namelen,
3510 int item_type, int errors)
3512 struct root_record *rec;
3513 struct root_backref *backref;
3515 rec = get_root_rec(root_cache, root_id);
3516 BUG_ON(IS_ERR(rec));
3517 backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
3520 backref->errors |= errors;
3522 if (item_type != BTRFS_DIR_ITEM_KEY) {
3523 if (backref->found_dir_index || backref->found_back_ref ||
3524 backref->found_forward_ref) {
3525 if (backref->index != index)
3526 backref->errors |= REF_ERR_INDEX_UNMATCH;
3528 backref->index = index;
3532 if (item_type == BTRFS_DIR_ITEM_KEY) {
3533 if (backref->found_forward_ref)
3535 backref->found_dir_item = 1;
3536 } else if (item_type == BTRFS_DIR_INDEX_KEY) {
3537 backref->found_dir_index = 1;
3538 } else if (item_type == BTRFS_ROOT_REF_KEY) {
3539 if (backref->found_forward_ref)
3540 backref->errors |= REF_ERR_DUP_ROOT_REF;
3541 else if (backref->found_dir_item)
3543 backref->found_forward_ref = 1;
3544 } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
3545 if (backref->found_back_ref)
3546 backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
3547 backref->found_back_ref = 1;
3552 if (backref->found_forward_ref && backref->found_dir_item)
3553 backref->reachable = 1;
3557 static int merge_root_recs(struct btrfs_root *root,
3558 struct cache_tree *src_cache,
3559 struct cache_tree *dst_cache)
3561 struct cache_extent *cache;
3562 struct ptr_node *node;
3563 struct inode_record *rec;
3564 struct inode_backref *backref;
3567 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
3568 free_inode_recs_tree(src_cache);
3573 cache = search_cache_extent(src_cache, 0);
3576 node = container_of(cache, struct ptr_node, cache);
3578 remove_cache_extent(src_cache, &node->cache);
3581 ret = is_child_root(root, root->objectid, rec->ino);
3587 list_for_each_entry(backref, &rec->backrefs, list) {
3588 BUG_ON(backref->found_inode_ref);
3589 if (backref->found_dir_item)
3590 add_root_backref(dst_cache, rec->ino,
3591 root->root_key.objectid, backref->dir,
3592 backref->index, backref->name,
3593 backref->namelen, BTRFS_DIR_ITEM_KEY,
3595 if (backref->found_dir_index)
3596 add_root_backref(dst_cache, rec->ino,
3597 root->root_key.objectid, backref->dir,
3598 backref->index, backref->name,
3599 backref->namelen, BTRFS_DIR_INDEX_KEY,
3603 free_inode_rec(rec);
3610 static int check_root_refs(struct btrfs_root *root,
3611 struct cache_tree *root_cache)
3613 struct root_record *rec;
3614 struct root_record *ref_root;
3615 struct root_backref *backref;
3616 struct cache_extent *cache;
3622 rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
3623 BUG_ON(IS_ERR(rec));
3626 /* fixme: this can not detect circular references */
3629 cache = search_cache_extent(root_cache, 0);
3633 rec = container_of(cache, struct root_record, cache);
3634 cache = next_cache_extent(cache);
3636 if (rec->found_ref == 0)
3639 list_for_each_entry(backref, &rec->backrefs, list) {
3640 if (!backref->reachable)
3643 ref_root = get_root_rec(root_cache,
3645 BUG_ON(IS_ERR(ref_root));
3646 if (ref_root->found_ref > 0)
3649 backref->reachable = 0;
3651 if (rec->found_ref == 0)
3657 cache = search_cache_extent(root_cache, 0);
3661 rec = container_of(cache, struct root_record, cache);
3662 cache = next_cache_extent(cache);
3664 if (rec->found_ref == 0 &&
3665 rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
3666 rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
3667 ret = check_orphan_item(root->fs_info->tree_root,
3673 * If we don't have a root item then we likely just have
3674 * a dir item in a snapshot for this root but no actual
3675 * ref key or anything so it's meaningless.
3677 if (!rec->found_root_item)
3680 fprintf(stderr, "fs tree %llu not referenced\n",
3681 (unsigned long long)rec->objectid);
3685 if (rec->found_ref > 0 && !rec->found_root_item)
3687 list_for_each_entry(backref, &rec->backrefs, list) {
3688 if (!backref->found_dir_item)
3689 backref->errors |= REF_ERR_NO_DIR_ITEM;
3690 if (!backref->found_dir_index)
3691 backref->errors |= REF_ERR_NO_DIR_INDEX;
3692 if (!backref->found_back_ref)
3693 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
3694 if (!backref->found_forward_ref)
3695 backref->errors |= REF_ERR_NO_ROOT_REF;
3696 if (backref->reachable && backref->errors)
3703 fprintf(stderr, "fs tree %llu refs %u %s\n",
3704 (unsigned long long)rec->objectid, rec->found_ref,
3705 rec->found_root_item ? "" : "not found");
3707 list_for_each_entry(backref, &rec->backrefs, list) {
3708 if (!backref->reachable)
3710 if (!backref->errors && rec->found_root_item)
3712 fprintf(stderr, "\tunresolved ref root %llu dir %llu"
3713 " index %llu namelen %u name %s errors %x\n",
3714 (unsigned long long)backref->ref_root,
3715 (unsigned long long)backref->dir,
3716 (unsigned long long)backref->index,
3717 backref->namelen, backref->name,
3719 print_ref_error(backref->errors);
3722 return errors > 0 ? 1 : 0;
3725 static int process_root_ref(struct extent_buffer *eb, int slot,
3726 struct btrfs_key *key,
3727 struct cache_tree *root_cache)
3733 struct btrfs_root_ref *ref;
3734 char namebuf[BTRFS_NAME_LEN];
3737 ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
3739 dirid = btrfs_root_ref_dirid(eb, ref);
3740 index = btrfs_root_ref_sequence(eb, ref);
3741 name_len = btrfs_root_ref_name_len(eb, ref);
3743 if (name_len <= BTRFS_NAME_LEN) {
3747 len = BTRFS_NAME_LEN;
3748 error = REF_ERR_NAME_TOO_LONG;
3750 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
3752 if (key->type == BTRFS_ROOT_REF_KEY) {
3753 add_root_backref(root_cache, key->offset, key->objectid, dirid,
3754 index, namebuf, len, key->type, error);
3756 add_root_backref(root_cache, key->objectid, key->offset, dirid,
3757 index, namebuf, len, key->type, error);
3762 static void free_corrupt_block(struct cache_extent *cache)
3764 struct btrfs_corrupt_block *corrupt;
3766 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
3770 FREE_EXTENT_CACHE_BASED_TREE(corrupt_blocks, free_corrupt_block);
3773 * Repair the btree of the given root.
3775 * The fix is to remove the node key in corrupt_blocks cache_tree.
3776 * and rebalance the tree.
3777 * After the fix, the btree should be writeable.
3779 static int repair_btree(struct btrfs_root *root,
3780 struct cache_tree *corrupt_blocks)
3782 struct btrfs_trans_handle *trans;
3783 struct btrfs_path path;
3784 struct btrfs_corrupt_block *corrupt;
3785 struct cache_extent *cache;
3786 struct btrfs_key key;
3791 if (cache_tree_empty(corrupt_blocks))
3794 trans = btrfs_start_transaction(root, 1);
3795 if (IS_ERR(trans)) {
3796 ret = PTR_ERR(trans);
3797 fprintf(stderr, "Error starting transaction: %s\n",
3801 btrfs_init_path(&path);
3802 cache = first_cache_extent(corrupt_blocks);
3804 corrupt = container_of(cache, struct btrfs_corrupt_block,
3806 level = corrupt->level;
3807 path.lowest_level = level;
3808 key.objectid = corrupt->key.objectid;
3809 key.type = corrupt->key.type;
3810 key.offset = corrupt->key.offset;
3813 * Here we don't want to do any tree balance, since it may
3814 * cause a balance with corrupted brother leaf/node,
3815 * so ins_len set to 0 here.
3816 * Balance will be done after all corrupt node/leaf is deleted.
3818 ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
3821 offset = btrfs_node_blockptr(path.nodes[level],
3824 /* Remove the ptr */
3825 ret = btrfs_del_ptr(root, &path, level, path.slots[level]);
3829 * Remove the corresponding extent
3830 * return value is not concerned.
3832 btrfs_release_path(&path);
3833 ret = btrfs_free_extent(trans, root, offset, root->nodesize,
3834 0, root->root_key.objectid,
3836 cache = next_cache_extent(cache);
3839 /* Balance the btree using btrfs_search_slot() */
3840 cache = first_cache_extent(corrupt_blocks);
3842 corrupt = container_of(cache, struct btrfs_corrupt_block,
3844 memcpy(&key, &corrupt->key, sizeof(key));
3845 ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
3848 /* return will always >0 since it won't find the item */
3850 btrfs_release_path(&path);
3851 cache = next_cache_extent(cache);
3854 btrfs_commit_transaction(trans, root);
3855 btrfs_release_path(&path);
3859 static int check_fs_root(struct btrfs_root *root,
3860 struct cache_tree *root_cache,
3861 struct walk_control *wc)
3867 struct btrfs_path path;
3868 struct shared_node root_node;
3869 struct root_record *rec;
3870 struct btrfs_root_item *root_item = &root->root_item;
3871 struct cache_tree corrupt_blocks;
3872 struct orphan_data_extent *orphan;
3873 struct orphan_data_extent *tmp;
3874 enum btrfs_tree_block_status status;
3875 struct node_refs nrefs;
3878 * Reuse the corrupt_block cache tree to record corrupted tree block
3880 * Unlike the usage in extent tree check, here we do it in a per
3881 * fs/subvol tree base.
3883 cache_tree_init(&corrupt_blocks);
3884 root->fs_info->corrupt_blocks = &corrupt_blocks;
3886 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
3887 rec = get_root_rec(root_cache, root->root_key.objectid);
3888 BUG_ON(IS_ERR(rec));
3889 if (btrfs_root_refs(root_item) > 0)
3890 rec->found_root_item = 1;
3893 btrfs_init_path(&path);
3894 memset(&root_node, 0, sizeof(root_node));
3895 cache_tree_init(&root_node.root_cache);
3896 cache_tree_init(&root_node.inode_cache);
3897 memset(&nrefs, 0, sizeof(nrefs));
3899 /* Move the orphan extent record to corresponding inode_record */
3900 list_for_each_entry_safe(orphan, tmp,
3901 &root->orphan_data_extents, list) {
3902 struct inode_record *inode;
3904 inode = get_inode_rec(&root_node.inode_cache, orphan->objectid,
3906 BUG_ON(IS_ERR(inode));
3907 inode->errors |= I_ERR_FILE_EXTENT_ORPHAN;
3908 list_move(&orphan->list, &inode->orphan_extents);
3911 level = btrfs_header_level(root->node);
3912 memset(wc->nodes, 0, sizeof(wc->nodes));
3913 wc->nodes[level] = &root_node;
3914 wc->active_node = level;
3915 wc->root_level = level;
3917 /* We may not have checked the root block, lets do that now */
3918 if (btrfs_is_leaf(root->node))
3919 status = btrfs_check_leaf(root, NULL, root->node);
3921 status = btrfs_check_node(root, NULL, root->node);
3922 if (status != BTRFS_TREE_BLOCK_CLEAN)
3925 if (btrfs_root_refs(root_item) > 0 ||
3926 btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
3927 path.nodes[level] = root->node;
3928 extent_buffer_get(root->node);
3929 path.slots[level] = 0;
3931 struct btrfs_key key;
3932 struct btrfs_disk_key found_key;
3934 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
3935 level = root_item->drop_level;
3936 path.lowest_level = level;
3937 if (level > btrfs_header_level(root->node) ||
3938 level >= BTRFS_MAX_LEVEL) {
3939 error("ignoring invalid drop level: %u", level);
3942 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
3945 btrfs_node_key(path.nodes[level], &found_key,
3947 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
3948 sizeof(found_key)));
3952 wret = walk_down_tree(root, &path, wc, &level, &nrefs);
3958 wret = walk_up_tree(root, &path, wc, &level);
3965 btrfs_release_path(&path);
3967 if (!cache_tree_empty(&corrupt_blocks)) {
3968 struct cache_extent *cache;
3969 struct btrfs_corrupt_block *corrupt;
3971 printf("The following tree block(s) is corrupted in tree %llu:\n",
3972 root->root_key.objectid);
3973 cache = first_cache_extent(&corrupt_blocks);
3975 corrupt = container_of(cache,
3976 struct btrfs_corrupt_block,
3978 printf("\ttree block bytenr: %llu, level: %d, node key: (%llu, %u, %llu)\n",
3979 cache->start, corrupt->level,
3980 corrupt->key.objectid, corrupt->key.type,
3981 corrupt->key.offset);
3982 cache = next_cache_extent(cache);
3985 printf("Try to repair the btree for root %llu\n",
3986 root->root_key.objectid);
3987 ret = repair_btree(root, &corrupt_blocks);
3989 fprintf(stderr, "Failed to repair btree: %s\n",
3992 printf("Btree for root %llu is fixed\n",
3993 root->root_key.objectid);
3997 err = merge_root_recs(root, &root_node.root_cache, root_cache);
4001 if (root_node.current) {
4002 root_node.current->checked = 1;
4003 maybe_free_inode_rec(&root_node.inode_cache,
4007 err = check_inode_recs(root, &root_node.inode_cache);
4011 free_corrupt_blocks_tree(&corrupt_blocks);
4012 root->fs_info->corrupt_blocks = NULL;
4013 free_orphan_data_extents(&root->orphan_data_extents);
4017 static int fs_root_objectid(u64 objectid)
4019 if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
4020 objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
4022 return is_fstree(objectid);
4025 static int check_fs_roots(struct btrfs_root *root,
4026 struct cache_tree *root_cache)
4028 struct btrfs_path path;
4029 struct btrfs_key key;
4030 struct walk_control wc;
4031 struct extent_buffer *leaf, *tree_node;
4032 struct btrfs_root *tmp_root;
4033 struct btrfs_root *tree_root = root->fs_info->tree_root;
4037 if (ctx.progress_enabled) {
4038 ctx.tp = TASK_FS_ROOTS;
4039 task_start(ctx.info);
4043 * Just in case we made any changes to the extent tree that weren't
4044 * reflected into the free space cache yet.
4047 reset_cached_block_groups(root->fs_info);
4048 memset(&wc, 0, sizeof(wc));
4049 cache_tree_init(&wc.shared);
4050 btrfs_init_path(&path);
4055 key.type = BTRFS_ROOT_ITEM_KEY;
4056 ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
4061 tree_node = tree_root->node;
4063 if (tree_node != tree_root->node) {
4064 free_root_recs_tree(root_cache);
4065 btrfs_release_path(&path);
4068 leaf = path.nodes[0];
4069 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
4070 ret = btrfs_next_leaf(tree_root, &path);
4076 leaf = path.nodes[0];
4078 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
4079 if (key.type == BTRFS_ROOT_ITEM_KEY &&
4080 fs_root_objectid(key.objectid)) {
4081 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
4082 tmp_root = btrfs_read_fs_root_no_cache(
4083 root->fs_info, &key);
4085 key.offset = (u64)-1;
4086 tmp_root = btrfs_read_fs_root(
4087 root->fs_info, &key);
4089 if (IS_ERR(tmp_root)) {
4093 ret = check_fs_root(tmp_root, root_cache, &wc);
4094 if (ret == -EAGAIN) {
4095 free_root_recs_tree(root_cache);
4096 btrfs_release_path(&path);
4101 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
4102 btrfs_free_fs_root(tmp_root);
4103 } else if (key.type == BTRFS_ROOT_REF_KEY ||
4104 key.type == BTRFS_ROOT_BACKREF_KEY) {
4105 process_root_ref(leaf, path.slots[0], &key,
4112 btrfs_release_path(&path);
4114 free_extent_cache_tree(&wc.shared);
4115 if (!cache_tree_empty(&wc.shared))
4116 fprintf(stderr, "warning line %d\n", __LINE__);
4118 task_stop(ctx.info);
4124 * Find DIR_ITEM/DIR_INDEX for the given key and check it with the specified
4125 * INODE_REF/INODE_EXTREF match.
4127 * @root: the root of the fs/file tree
4128 * @ref_key: the key of the INODE_REF/INODE_EXTREF
4129 * @key: the key of the DIR_ITEM/DIR_INDEX
4130 * @index: the index in the INODE_REF/INODE_EXTREF, be used to
4131 * distinguish root_dir between normal dir/file
4132 * @name: the name in the INODE_REF/INODE_EXTREF
4133 * @namelen: the length of name in the INODE_REF/INODE_EXTREF
4134 * @mode: the st_mode of INODE_ITEM
4136 * Return 0 if no error occurred.
4137 * Return ROOT_DIR_ERROR if found DIR_ITEM/DIR_INDEX for root_dir.
4138 * Return DIR_ITEM_MISSING if couldn't find DIR_ITEM/DIR_INDEX for normal
4140 * Return DIR_ITEM_MISMATCH if INODE_REF/INODE_EXTREF and DIR_ITEM/DIR_INDEX
4141 * not match for normal dir/file.
4143 static int find_dir_item(struct btrfs_root *root, struct btrfs_key *ref_key,
4144 struct btrfs_key *key, u64 index, char *name,
4145 u32 namelen, u32 mode)
4147 struct btrfs_path path;
4148 struct extent_buffer *node;
4149 struct btrfs_dir_item *di;
4150 struct btrfs_key location;
4151 char namebuf[BTRFS_NAME_LEN] = {0};
4161 btrfs_init_path(&path);
4162 ret = btrfs_search_slot(NULL, root, key, &path, 0, 0);
4164 ret = DIR_ITEM_MISSING;
4168 /* Process root dir and goto out*/
4171 ret = ROOT_DIR_ERROR;
4173 "root %llu INODE %s[%llu %llu] ROOT_DIR shouldn't have %s",
4175 ref_key->type == BTRFS_INODE_REF_KEY ?
4177 ref_key->objectid, ref_key->offset,
4178 key->type == BTRFS_DIR_ITEM_KEY ?
4179 "DIR_ITEM" : "DIR_INDEX");
4187 /* Process normal file/dir */
4189 ret = DIR_ITEM_MISSING;
4191 "root %llu INODE %s[%llu %llu] doesn't have related %s[%llu %llu] namelen %u filename %s filetype %d",
4193 ref_key->type == BTRFS_INODE_REF_KEY ? "REF" : "EXTREF",
4194 ref_key->objectid, ref_key->offset,
4195 key->type == BTRFS_DIR_ITEM_KEY ?
4196 "DIR_ITEM" : "DIR_INDEX",
4197 key->objectid, key->offset, namelen, name,
4198 imode_to_type(mode));
4202 /* Check whether inode_id/filetype/name match */
4203 node = path.nodes[0];
4204 slot = path.slots[0];
4205 di = btrfs_item_ptr(node, slot, struct btrfs_dir_item);
4206 total = btrfs_item_size_nr(node, slot);
4207 while (cur < total) {
4208 ret = DIR_ITEM_MISMATCH;
4209 name_len = btrfs_dir_name_len(node, di);
4210 data_len = btrfs_dir_data_len(node, di);
4212 btrfs_dir_item_key_to_cpu(node, di, &location);
4213 if (location.objectid != ref_key->objectid ||
4214 location.type != BTRFS_INODE_ITEM_KEY ||
4215 location.offset != 0)
4218 filetype = btrfs_dir_type(node, di);
4219 if (imode_to_type(mode) != filetype)
4222 if (name_len <= BTRFS_NAME_LEN) {
4225 len = BTRFS_NAME_LEN;
4226 warning("root %llu %s[%llu %llu] name too long %u, trimmed",
4228 key->type == BTRFS_DIR_ITEM_KEY ?
4229 "DIR_ITEM" : "DIR_INDEX",
4230 key->objectid, key->offset, name_len);
4232 read_extent_buffer(node, namebuf, (unsigned long)(di + 1), len);
4233 if (len != namelen || strncmp(namebuf, name, len))
4239 len = sizeof(*di) + name_len + data_len;
4240 di = (struct btrfs_dir_item *)((char *)di + len);
4243 if (ret == DIR_ITEM_MISMATCH)
4245 "root %llu INODE %s[%llu %llu] and %s[%llu %llu] mismatch namelen %u filename %s filetype %d",
4247 ref_key->type == BTRFS_INODE_REF_KEY ? "REF" : "EXTREF",
4248 ref_key->objectid, ref_key->offset,
4249 key->type == BTRFS_DIR_ITEM_KEY ?
4250 "DIR_ITEM" : "DIR_INDEX",
4251 key->objectid, key->offset, namelen, name,
4252 imode_to_type(mode));
4254 btrfs_release_path(&path);
4259 * Traverse the given INODE_REF and call find_dir_item() to find related
4260 * DIR_ITEM/DIR_INDEX.
4262 * @root: the root of the fs/file tree
4263 * @ref_key: the key of the INODE_REF
4264 * @refs: the count of INODE_REF
4265 * @mode: the st_mode of INODE_ITEM
4267 * Return 0 if no error occurred.
4269 static int check_inode_ref(struct btrfs_root *root, struct btrfs_key *ref_key,
4270 struct extent_buffer *node, int slot, u64 *refs,
4273 struct btrfs_key key;
4274 struct btrfs_inode_ref *ref;
4275 char namebuf[BTRFS_NAME_LEN] = {0};
4283 ref = btrfs_item_ptr(node, slot, struct btrfs_inode_ref);
4284 total = btrfs_item_size_nr(node, slot);
4287 /* Update inode ref count */
4290 index = btrfs_inode_ref_index(node, ref);
4291 name_len = btrfs_inode_ref_name_len(node, ref);
4292 if (name_len <= BTRFS_NAME_LEN) {
4295 len = BTRFS_NAME_LEN;
4296 warning("root %llu INODE_REF[%llu %llu] name too long",
4297 root->objectid, ref_key->objectid, ref_key->offset);
4300 read_extent_buffer(node, namebuf, (unsigned long)(ref + 1), len);
4302 /* Check root dir ref name */
4303 if (index == 0 && strncmp(namebuf, "..", name_len)) {
4304 error("root %llu INODE_REF[%llu %llu] ROOT_DIR name shouldn't be %s",
4305 root->objectid, ref_key->objectid, ref_key->offset,
4307 err |= ROOT_DIR_ERROR;
4310 /* Find related DIR_INDEX */
4311 key.objectid = ref_key->offset;
4312 key.type = BTRFS_DIR_INDEX_KEY;
4314 ret = find_dir_item(root, ref_key, &key, index, namebuf, len, mode);
4317 /* Find related dir_item */
4318 key.objectid = ref_key->offset;
4319 key.type = BTRFS_DIR_ITEM_KEY;
4320 key.offset = btrfs_name_hash(namebuf, len);
4321 ret = find_dir_item(root, ref_key, &key, index, namebuf, len, mode);
4324 len = sizeof(*ref) + name_len;
4325 ref = (struct btrfs_inode_ref *)((char *)ref + len);
4334 * Traverse the given INODE_EXTREF and call find_dir_item() to find related
4335 * DIR_ITEM/DIR_INDEX.
4337 * @root: the root of the fs/file tree
4338 * @ref_key: the key of the INODE_EXTREF
4339 * @refs: the count of INODE_EXTREF
4340 * @mode: the st_mode of INODE_ITEM
4342 * Return 0 if no error occurred.
4344 static int check_inode_extref(struct btrfs_root *root,
4345 struct btrfs_key *ref_key,
4346 struct extent_buffer *node, int slot, u64 *refs,
4349 struct btrfs_key key;
4350 struct btrfs_inode_extref *extref;
4351 char namebuf[BTRFS_NAME_LEN] = {0};
4361 extref = btrfs_item_ptr(node, slot, struct btrfs_inode_extref);
4362 total = btrfs_item_size_nr(node, slot);
4365 /* update inode ref count */
4367 name_len = btrfs_inode_extref_name_len(node, extref);
4368 index = btrfs_inode_extref_index(node, extref);
4369 parent = btrfs_inode_extref_parent(node, extref);
4370 if (name_len <= BTRFS_NAME_LEN) {
4373 len = BTRFS_NAME_LEN;
4374 warning("root %llu INODE_EXTREF[%llu %llu] name too long",
4375 root->objectid, ref_key->objectid, ref_key->offset);
4377 read_extent_buffer(node, namebuf, (unsigned long)(extref + 1), len);
4379 /* Check root dir ref name */
4380 if (index == 0 && strncmp(namebuf, "..", name_len)) {
4381 error("root %llu INODE_EXTREF[%llu %llu] ROOT_DIR name shouldn't be %s",
4382 root->objectid, ref_key->objectid, ref_key->offset,
4384 err |= ROOT_DIR_ERROR;
4387 /* find related dir_index */
4388 key.objectid = parent;
4389 key.type = BTRFS_DIR_INDEX_KEY;
4391 ret = find_dir_item(root, ref_key, &key, index, namebuf, len, mode);
4394 /* find related dir_item */
4395 key.objectid = parent;
4396 key.type = BTRFS_DIR_ITEM_KEY;
4397 key.offset = btrfs_name_hash(namebuf, len);
4398 ret = find_dir_item(root, ref_key, &key, index, namebuf, len, mode);
4401 len = sizeof(*extref) + name_len;
4402 extref = (struct btrfs_inode_extref *)((char *)extref + len);
4412 * Find INODE_REF/INODE_EXTREF for the given key and check it with the specified
4413 * DIR_ITEM/DIR_INDEX match.
4415 * @root: the root of the fs/file tree
4416 * @key: the key of the INODE_REF/INODE_EXTREF
4417 * @name: the name in the INODE_REF/INODE_EXTREF
4418 * @namelen: the length of name in the INODE_REF/INODE_EXTREF
4419 * @index: the index in the INODE_REF/INODE_EXTREF, for DIR_ITEM set index
4421 * @ext_ref: the EXTENDED_IREF feature
4423 * Return 0 if no error occurred.
4424 * Return >0 for error bitmap
4426 static int find_inode_ref(struct btrfs_root *root, struct btrfs_key *key,
4427 char *name, int namelen, u64 index,
4428 unsigned int ext_ref)
4430 struct btrfs_path path;
4431 struct btrfs_inode_ref *ref;
4432 struct btrfs_inode_extref *extref;
4433 struct extent_buffer *node;
4434 char ref_namebuf[BTRFS_NAME_LEN] = {0};
4445 btrfs_init_path(&path);
4446 ret = btrfs_search_slot(NULL, root, key, &path, 0, 0);
4448 ret = INODE_REF_MISSING;
4452 node = path.nodes[0];
4453 slot = path.slots[0];
4455 ref = btrfs_item_ptr(node, slot, struct btrfs_inode_ref);
4456 total = btrfs_item_size_nr(node, slot);
4458 /* Iterate all entry of INODE_REF */
4459 while (cur < total) {
4460 ret = INODE_REF_MISSING;
4462 ref_namelen = btrfs_inode_ref_name_len(node, ref);
4463 ref_index = btrfs_inode_ref_index(node, ref);
4464 if (index != (u64)-1 && index != ref_index)
4467 if (ref_namelen <= BTRFS_NAME_LEN) {
4470 len = BTRFS_NAME_LEN;
4471 warning("root %llu INODE %s[%llu %llu] name too long",
4473 key->type == BTRFS_INODE_REF_KEY ?
4475 key->objectid, key->offset);
4477 read_extent_buffer(node, ref_namebuf, (unsigned long)(ref + 1),
4480 if (len != namelen || strncmp(ref_namebuf, name, len))
4486 len = sizeof(*ref) + ref_namelen;
4487 ref = (struct btrfs_inode_ref *)((char *)ref + len);
4492 /* Skip if not support EXTENDED_IREF feature */
4496 btrfs_release_path(&path);
4497 btrfs_init_path(&path);
4499 dir_id = key->offset;
4500 key->type = BTRFS_INODE_EXTREF_KEY;
4501 key->offset = btrfs_extref_hash(dir_id, name, namelen);
4503 ret = btrfs_search_slot(NULL, root, key, &path, 0, 0);
4505 ret = INODE_REF_MISSING;
4509 node = path.nodes[0];
4510 slot = path.slots[0];
4512 extref = btrfs_item_ptr(node, slot, struct btrfs_inode_extref);
4514 total = btrfs_item_size_nr(node, slot);
4516 /* Iterate all entry of INODE_EXTREF */
4517 while (cur < total) {
4518 ret = INODE_REF_MISSING;
4520 ref_namelen = btrfs_inode_extref_name_len(node, extref);
4521 ref_index = btrfs_inode_extref_index(node, extref);
4522 parent = btrfs_inode_extref_parent(node, extref);
4523 if (index != (u64)-1 && index != ref_index)
4526 if (parent != dir_id)
4529 if (ref_namelen <= BTRFS_NAME_LEN) {
4532 len = BTRFS_NAME_LEN;
4533 warning("root %llu INODE %s[%llu %llu] name too long",
4535 key->type == BTRFS_INODE_REF_KEY ?
4537 key->objectid, key->offset);
4539 read_extent_buffer(node, ref_namebuf,
4540 (unsigned long)(extref + 1), len);
4542 if (len != namelen || strncmp(ref_namebuf, name, len))
4549 len = sizeof(*extref) + ref_namelen;
4550 extref = (struct btrfs_inode_extref *)((char *)extref + len);
4555 btrfs_release_path(&path);
4560 * Traverse the given DIR_ITEM/DIR_INDEX and check related INODE_ITEM and
4561 * call find_inode_ref() to check related INODE_REF/INODE_EXTREF.
4563 * @root: the root of the fs/file tree
4564 * @key: the key of the INODE_REF/INODE_EXTREF
4565 * @size: the st_size of the INODE_ITEM
4566 * @ext_ref: the EXTENDED_IREF feature
4568 * Return 0 if no error occurred.
4570 static int check_dir_item(struct btrfs_root *root, struct btrfs_key *key,
4571 struct extent_buffer *node, int slot, u64 *size,
4572 unsigned int ext_ref)
4574 struct btrfs_dir_item *di;
4575 struct btrfs_inode_item *ii;
4576 struct btrfs_path path;
4577 struct btrfs_key location;
4578 char namebuf[BTRFS_NAME_LEN] = {0};
4591 * For DIR_ITEM set index to (u64)-1, so that find_inode_ref
4592 * ignore index check.
4594 index = (key->type == BTRFS_DIR_INDEX_KEY) ? key->offset : (u64)-1;
4596 di = btrfs_item_ptr(node, slot, struct btrfs_dir_item);
4597 total = btrfs_item_size_nr(node, slot);
4599 while (cur < total) {
4600 data_len = btrfs_dir_data_len(node, di);
4602 error("root %llu %s[%llu %llu] data_len shouldn't be %u",
4603 root->objectid, key->type == BTRFS_DIR_ITEM_KEY ?
4604 "DIR_ITEM" : "DIR_INDEX",
4605 key->objectid, key->offset, data_len);
4607 name_len = btrfs_dir_name_len(node, di);
4608 if (name_len <= BTRFS_NAME_LEN) {
4611 len = BTRFS_NAME_LEN;
4612 warning("root %llu %s[%llu %llu] name too long",
4614 key->type == BTRFS_DIR_ITEM_KEY ?
4615 "DIR_ITEM" : "DIR_INDEX",
4616 key->objectid, key->offset);
4618 (*size) += name_len;
4620 read_extent_buffer(node, namebuf, (unsigned long)(di + 1), len);
4621 filetype = btrfs_dir_type(node, di);
4623 btrfs_init_path(&path);
4624 btrfs_dir_item_key_to_cpu(node, di, &location);
4626 /* Ignore related ROOT_ITEM check */
4627 if (location.type == BTRFS_ROOT_ITEM_KEY)
4630 /* Check relative INODE_ITEM(existence/filetype) */
4631 ret = btrfs_search_slot(NULL, root, &location, &path, 0, 0);
4633 err |= INODE_ITEM_MISSING;
4634 error("root %llu %s[%llu %llu] couldn't find relative INODE_ITEM[%llu] namelen %u filename %s filetype %x",
4635 root->objectid, key->type == BTRFS_DIR_ITEM_KEY ?
4636 "DIR_ITEM" : "DIR_INDEX", key->objectid,
4637 key->offset, location.objectid, name_len,
4642 ii = btrfs_item_ptr(path.nodes[0], path.slots[0],
4643 struct btrfs_inode_item);
4644 mode = btrfs_inode_mode(path.nodes[0], ii);
4646 if (imode_to_type(mode) != filetype) {
4647 err |= INODE_ITEM_MISMATCH;
4648 error("root %llu %s[%llu %llu] relative INODE_ITEM filetype mismatch namelen %u filename %s filetype %d",
4649 root->objectid, key->type == BTRFS_DIR_ITEM_KEY ?
4650 "DIR_ITEM" : "DIR_INDEX", key->objectid,
4651 key->offset, name_len, namebuf, filetype);
4654 /* Check relative INODE_REF/INODE_EXTREF */
4655 location.type = BTRFS_INODE_REF_KEY;
4656 location.offset = key->objectid;
4657 ret = find_inode_ref(root, &location, namebuf, len,
4660 if (ret & INODE_REF_MISSING)
4661 error("root %llu %s[%llu %llu] relative INODE_REF missing namelen %u filename %s filetype %d",
4662 root->objectid, key->type == BTRFS_DIR_ITEM_KEY ?
4663 "DIR_ITEM" : "DIR_INDEX", key->objectid,
4664 key->offset, name_len, namebuf, filetype);
4667 btrfs_release_path(&path);
4668 len = sizeof(*di) + name_len + data_len;
4669 di = (struct btrfs_dir_item *)((char *)di + len);
4672 if (key->type == BTRFS_DIR_INDEX_KEY && cur < total) {
4673 error("root %llu DIR_INDEX[%llu %llu] should contain only one entry",
4674 root->objectid, key->objectid, key->offset);
4683 * Check file extent datasum/hole, update the size of the file extents,
4684 * check and update the last offset of the file extent.
4686 * @root: the root of fs/file tree.
4687 * @fkey: the key of the file extent.
4688 * @nodatasum: INODE_NODATASUM feature.
4689 * @size: the sum of all EXTENT_DATA items size for this inode.
4690 * @end: the offset of the last extent.
4692 * Return 0 if no error occurred.
4694 static int check_file_extent(struct btrfs_root *root, struct btrfs_key *fkey,
4695 struct extent_buffer *node, int slot,
4696 unsigned int nodatasum, u64 *size, u64 *end)
4698 struct btrfs_file_extent_item *fi;
4701 u64 extent_num_bytes;
4703 unsigned int extent_type;
4704 unsigned int is_hole;
4708 fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
4710 extent_type = btrfs_file_extent_type(node, fi);
4711 /* Skip if file extent is inline */
4712 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4713 struct btrfs_item *e = btrfs_item_nr(slot);
4714 u32 item_inline_len;
4716 item_inline_len = btrfs_file_extent_inline_item_len(node, e);
4717 extent_num_bytes = btrfs_file_extent_inline_len(node, slot, fi);
4718 if (extent_num_bytes == 0 ||
4719 extent_num_bytes != item_inline_len)
4720 err |= FILE_EXTENT_ERROR;
4721 *size += extent_num_bytes;
4725 /* Check extent type */
4726 if (extent_type != BTRFS_FILE_EXTENT_REG &&
4727 extent_type != BTRFS_FILE_EXTENT_PREALLOC) {
4728 err |= FILE_EXTENT_ERROR;
4729 error("root %llu EXTENT_DATA[%llu %llu] type bad",
4730 root->objectid, fkey->objectid, fkey->offset);
4734 /* Check REG_EXTENT/PREALLOC_EXTENT */
4735 disk_bytenr = btrfs_file_extent_disk_bytenr(node, fi);
4736 disk_num_bytes = btrfs_file_extent_disk_num_bytes(node, fi);
4737 extent_num_bytes = btrfs_file_extent_num_bytes(node, fi);
4738 is_hole = (disk_bytenr == 0) && (disk_num_bytes == 0);
4740 /* Check EXTENT_DATA datasum */
4741 ret = count_csum_range(root, disk_bytenr, disk_num_bytes, &found);
4742 if (found > 0 && nodatasum) {
4743 err |= ODD_CSUM_ITEM;
4744 error("root %llu EXTENT_DATA[%llu %llu] nodatasum shouldn't have datasum",
4745 root->objectid, fkey->objectid, fkey->offset);
4746 } else if (extent_type == BTRFS_FILE_EXTENT_REG && !nodatasum &&
4748 (ret < 0 || found == 0 || found < disk_num_bytes)) {
4749 err |= CSUM_ITEM_MISSING;
4750 error("root %llu EXTENT_DATA[%llu %llu] datasum missing",
4751 root->objectid, fkey->objectid, fkey->offset);
4752 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC && found > 0) {
4753 err |= ODD_CSUM_ITEM;
4754 error("root %llu EXTENT_DATA[%llu %llu] prealloc shouldn't have datasum",
4755 root->objectid, fkey->objectid, fkey->offset);
4758 /* Check EXTENT_DATA hole */
4759 if (no_holes && is_hole) {
4760 err |= FILE_EXTENT_ERROR;
4761 error("root %llu EXTENT_DATA[%llu %llu] shouldn't be hole",
4762 root->objectid, fkey->objectid, fkey->offset);
4763 } else if (!no_holes && *end != fkey->offset) {
4764 err |= FILE_EXTENT_ERROR;
4765 error("root %llu EXTENT_DATA[%llu %llu] interrupt",
4766 root->objectid, fkey->objectid, fkey->offset);
4769 *end += extent_num_bytes;
4771 *size += extent_num_bytes;
4777 * Check INODE_ITEM and related ITEMs (the same inode number)
4778 * 1. check link count
4779 * 2. check inode ref/extref
4780 * 3. check dir item/index
4782 * @ext_ref: the EXTENDED_IREF feature
4784 * Return 0 if no error occurred.
4785 * Return >0 for error or hit the traversal is done(by error bitmap)
4787 static int check_inode_item(struct btrfs_root *root, struct btrfs_path *path,
4788 unsigned int ext_ref)
4790 struct extent_buffer *node;
4791 struct btrfs_inode_item *ii;
4792 struct btrfs_key key;
4801 u64 extent_size = 0;
4803 unsigned int nodatasum;
4808 node = path->nodes[0];
4809 slot = path->slots[0];
4811 btrfs_item_key_to_cpu(node, &key, slot);
4812 inode_id = key.objectid;
4814 if (inode_id == BTRFS_ORPHAN_OBJECTID) {
4815 ret = btrfs_next_item(root, path);
4821 ii = btrfs_item_ptr(node, slot, struct btrfs_inode_item);
4822 isize = btrfs_inode_size(node, ii);
4823 nbytes = btrfs_inode_nbytes(node, ii);
4824 mode = btrfs_inode_mode(node, ii);
4825 dir = imode_to_type(mode) == BTRFS_FT_DIR;
4826 nlink = btrfs_inode_nlink(node, ii);
4827 nodatasum = btrfs_inode_flags(node, ii) & BTRFS_INODE_NODATASUM;
4830 ret = btrfs_next_item(root, path);
4832 /* out will fill 'err' rusing current statistics */
4834 } else if (ret > 0) {
4839 node = path->nodes[0];
4840 slot = path->slots[0];
4841 btrfs_item_key_to_cpu(node, &key, slot);
4842 if (key.objectid != inode_id)
4846 case BTRFS_INODE_REF_KEY:
4847 ret = check_inode_ref(root, &key, node, slot, &refs,
4851 case BTRFS_INODE_EXTREF_KEY:
4852 if (key.type == BTRFS_INODE_EXTREF_KEY && !ext_ref)
4853 warning("root %llu EXTREF[%llu %llu] isn't supported",
4854 root->objectid, key.objectid,
4856 ret = check_inode_extref(root, &key, node, slot, &refs,
4860 case BTRFS_DIR_ITEM_KEY:
4861 case BTRFS_DIR_INDEX_KEY:
4863 warning("root %llu INODE[%llu] mode %u shouldn't have DIR_INDEX[%llu %llu]",
4864 root->objectid, inode_id,
4865 imode_to_type(mode), key.objectid,
4868 ret = check_dir_item(root, &key, node, slot, &size,
4872 case BTRFS_EXTENT_DATA_KEY:
4874 warning("root %llu DIR INODE[%llu] shouldn't EXTENT_DATA[%llu %llu]",
4875 root->objectid, inode_id, key.objectid,
4878 ret = check_file_extent(root, &key, node, slot,
4879 nodatasum, &extent_size,
4883 case BTRFS_XATTR_ITEM_KEY:
4886 error("ITEM[%llu %u %llu] UNKNOWN TYPE",
4887 key.objectid, key.type, key.offset);
4892 /* verify INODE_ITEM nlink/isize/nbytes */
4895 err |= LINK_COUNT_ERROR;
4896 error("root %llu DIR INODE[%llu] shouldn't have more than one link(%llu)",
4897 root->objectid, inode_id, nlink);
4901 * Just a warning, as dir inode nbytes is just an
4902 * instructive value.
4904 if (!IS_ALIGNED(nbytes, root->nodesize)) {
4905 warning("root %llu DIR INODE[%llu] nbytes should be aligned to %u",
4906 root->objectid, inode_id, root->nodesize);
4909 if (isize != size) {
4911 error("root %llu DIR INODE [%llu] size(%llu) not equal to %llu",
4912 root->objectid, inode_id, isize, size);
4915 if (nlink != refs) {
4916 err |= LINK_COUNT_ERROR;
4917 error("root %llu INODE[%llu] nlink(%llu) not equal to inode_refs(%llu)",
4918 root->objectid, inode_id, nlink, refs);
4919 } else if (!nlink) {
4923 if (!nbytes && !no_holes && extent_end < isize) {
4924 err |= NBYTES_ERROR;
4925 error("root %llu INODE[%llu] size (%llu) should have a file extent hole",
4926 root->objectid, inode_id, isize);
4929 if (nbytes != extent_size) {
4930 err |= NBYTES_ERROR;
4931 error("root %llu INODE[%llu] nbytes(%llu) not equal to extent_size(%llu)",
4932 root->objectid, inode_id, nbytes, extent_size);
4939 static int check_fs_first_inode(struct btrfs_root *root, unsigned int ext_ref)
4941 struct btrfs_path path;
4942 struct btrfs_key key;
4946 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
4947 key.type = BTRFS_INODE_ITEM_KEY;
4950 /* For root being dropped, we don't need to check first inode */
4951 if (btrfs_root_refs(&root->root_item) == 0 &&
4952 btrfs_disk_key_objectid(&root->root_item.drop_progress) >=
4956 btrfs_init_path(&path);
4958 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
4963 err |= INODE_ITEM_MISSING;
4966 err |= check_inode_item(root, &path, ext_ref);
4971 btrfs_release_path(&path);
4976 * Iterate all item on the tree and call check_inode_item() to check.
4978 * @root: the root of the tree to be checked.
4979 * @ext_ref: the EXTENDED_IREF feature
4981 * Return 0 if no error found.
4982 * Return <0 for error.
4984 static int check_fs_root_v2(struct btrfs_root *root, unsigned int ext_ref)
4986 struct btrfs_path path;
4987 struct node_refs nrefs;
4988 struct btrfs_root_item *root_item = &root->root_item;
4993 * We need to manually check the first inode item(256)
4994 * As the following traversal function will only start from
4995 * the first inode item in the leaf, if inode item(256) is missing
4996 * we will just skip it forever.
4998 ret = check_fs_first_inode(root, ext_ref);
5002 memset(&nrefs, 0, sizeof(nrefs));
5003 level = btrfs_header_level(root->node);
5004 btrfs_init_path(&path);
5006 if (btrfs_root_refs(root_item) > 0 ||
5007 btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
5008 path.nodes[level] = root->node;
5009 path.slots[level] = 0;
5010 extent_buffer_get(root->node);
5012 struct btrfs_key key;
5014 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
5015 level = root_item->drop_level;
5016 path.lowest_level = level;
5017 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
5024 wret = walk_down_tree_v2(root, &path, &level, &nrefs, ext_ref);
5030 wret = walk_up_tree_v2(root, &path, &level);
5038 btrfs_release_path(&path);
5043 * Find the relative ref for root_ref and root_backref.
5045 * @root: the root of the root tree.
5046 * @ref_key: the key of the root ref.
5048 * Return 0 if no error occurred.
5050 static int check_root_ref(struct btrfs_root *root, struct btrfs_key *ref_key,
5051 struct extent_buffer *node, int slot)
5053 struct btrfs_path path;
5054 struct btrfs_key key;
5055 struct btrfs_root_ref *ref;
5056 struct btrfs_root_ref *backref;
5057 char ref_name[BTRFS_NAME_LEN] = {0};
5058 char backref_name[BTRFS_NAME_LEN] = {0};
5064 u32 backref_namelen;
5069 ref = btrfs_item_ptr(node, slot, struct btrfs_root_ref);
5070 ref_dirid = btrfs_root_ref_dirid(node, ref);
5071 ref_seq = btrfs_root_ref_sequence(node, ref);
5072 ref_namelen = btrfs_root_ref_name_len(node, ref);
5074 if (ref_namelen <= BTRFS_NAME_LEN) {
5077 len = BTRFS_NAME_LEN;
5078 warning("%s[%llu %llu] ref_name too long",
5079 ref_key->type == BTRFS_ROOT_REF_KEY ?
5080 "ROOT_REF" : "ROOT_BACKREF", ref_key->objectid,
5083 read_extent_buffer(node, ref_name, (unsigned long)(ref + 1), len);
5085 /* Find relative root_ref */
5086 key.objectid = ref_key->offset;
5087 key.type = BTRFS_ROOT_BACKREF_KEY + BTRFS_ROOT_REF_KEY - ref_key->type;
5088 key.offset = ref_key->objectid;
5090 btrfs_init_path(&path);
5091 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
5093 err |= ROOT_REF_MISSING;
5094 error("%s[%llu %llu] couldn't find relative ref",
5095 ref_key->type == BTRFS_ROOT_REF_KEY ?
5096 "ROOT_REF" : "ROOT_BACKREF",
5097 ref_key->objectid, ref_key->offset);
5101 backref = btrfs_item_ptr(path.nodes[0], path.slots[0],
5102 struct btrfs_root_ref);
5103 backref_dirid = btrfs_root_ref_dirid(path.nodes[0], backref);
5104 backref_seq = btrfs_root_ref_sequence(path.nodes[0], backref);
5105 backref_namelen = btrfs_root_ref_name_len(path.nodes[0], backref);
5107 if (backref_namelen <= BTRFS_NAME_LEN) {
5108 len = backref_namelen;
5110 len = BTRFS_NAME_LEN;
5111 warning("%s[%llu %llu] ref_name too long",
5112 key.type == BTRFS_ROOT_REF_KEY ?
5113 "ROOT_REF" : "ROOT_BACKREF",
5114 key.objectid, key.offset);
5116 read_extent_buffer(path.nodes[0], backref_name,
5117 (unsigned long)(backref + 1), len);
5119 if (ref_dirid != backref_dirid || ref_seq != backref_seq ||
5120 ref_namelen != backref_namelen ||
5121 strncmp(ref_name, backref_name, len)) {
5122 err |= ROOT_REF_MISMATCH;
5123 error("%s[%llu %llu] mismatch relative ref",
5124 ref_key->type == BTRFS_ROOT_REF_KEY ?
5125 "ROOT_REF" : "ROOT_BACKREF",
5126 ref_key->objectid, ref_key->offset);
5129 btrfs_release_path(&path);
5134 * Check all fs/file tree in low_memory mode.
5136 * 1. for fs tree root item, call check_fs_root_v2()
5137 * 2. for fs tree root ref/backref, call check_root_ref()
5139 * Return 0 if no error occurred.
5141 static int check_fs_roots_v2(struct btrfs_fs_info *fs_info)
5143 struct btrfs_root *tree_root = fs_info->tree_root;
5144 struct btrfs_root *cur_root = NULL;
5145 struct btrfs_path path;
5146 struct btrfs_key key;
5147 struct extent_buffer *node;
5148 unsigned int ext_ref;
5153 ext_ref = btrfs_fs_incompat(fs_info, EXTENDED_IREF);
5155 btrfs_init_path(&path);
5156 key.objectid = BTRFS_FS_TREE_OBJECTID;
5158 key.type = BTRFS_ROOT_ITEM_KEY;
5160 ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
5164 } else if (ret > 0) {
5170 node = path.nodes[0];
5171 slot = path.slots[0];
5172 btrfs_item_key_to_cpu(node, &key, slot);
5173 if (key.objectid > BTRFS_LAST_FREE_OBJECTID)
5175 if (key.type == BTRFS_ROOT_ITEM_KEY &&
5176 fs_root_objectid(key.objectid)) {
5177 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
5178 cur_root = btrfs_read_fs_root_no_cache(fs_info,
5181 key.offset = (u64)-1;
5182 cur_root = btrfs_read_fs_root(fs_info, &key);
5185 if (IS_ERR(cur_root)) {
5186 error("Fail to read fs/subvol tree: %lld",
5192 ret = check_fs_root_v2(cur_root, ext_ref);
5195 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
5196 btrfs_free_fs_root(cur_root);
5197 } else if (key.type == BTRFS_ROOT_REF_KEY ||
5198 key.type == BTRFS_ROOT_BACKREF_KEY) {
5199 ret = check_root_ref(tree_root, &key, node, slot);
5203 ret = btrfs_next_item(tree_root, &path);
5213 btrfs_release_path(&path);
5217 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
5219 struct list_head *cur = rec->backrefs.next;
5220 struct extent_backref *back;
5221 struct tree_backref *tback;
5222 struct data_backref *dback;
5226 while(cur != &rec->backrefs) {
5227 back = to_extent_backref(cur);
5229 if (!back->found_extent_tree) {
5233 if (back->is_data) {
5234 dback = to_data_backref(back);
5235 fprintf(stderr, "Backref %llu %s %llu"
5236 " owner %llu offset %llu num_refs %lu"
5237 " not found in extent tree\n",
5238 (unsigned long long)rec->start,
5239 back->full_backref ?
5241 back->full_backref ?
5242 (unsigned long long)dback->parent:
5243 (unsigned long long)dback->root,
5244 (unsigned long long)dback->owner,
5245 (unsigned long long)dback->offset,
5246 (unsigned long)dback->num_refs);
5248 tback = to_tree_backref(back);
5249 fprintf(stderr, "Backref %llu parent %llu"
5250 " root %llu not found in extent tree\n",
5251 (unsigned long long)rec->start,
5252 (unsigned long long)tback->parent,
5253 (unsigned long long)tback->root);
5256 if (!back->is_data && !back->found_ref) {
5260 tback = to_tree_backref(back);
5261 fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
5262 (unsigned long long)rec->start,
5263 back->full_backref ? "parent" : "root",
5264 back->full_backref ?
5265 (unsigned long long)tback->parent :
5266 (unsigned long long)tback->root, back);
5268 if (back->is_data) {
5269 dback = to_data_backref(back);
5270 if (dback->found_ref != dback->num_refs) {
5274 fprintf(stderr, "Incorrect local backref count"
5275 " on %llu %s %llu owner %llu"
5276 " offset %llu found %u wanted %u back %p\n",
5277 (unsigned long long)rec->start,
5278 back->full_backref ?
5280 back->full_backref ?
5281 (unsigned long long)dback->parent:
5282 (unsigned long long)dback->root,
5283 (unsigned long long)dback->owner,
5284 (unsigned long long)dback->offset,
5285 dback->found_ref, dback->num_refs, back);
5287 if (dback->disk_bytenr != rec->start) {
5291 fprintf(stderr, "Backref disk bytenr does not"
5292 " match extent record, bytenr=%llu, "
5293 "ref bytenr=%llu\n",
5294 (unsigned long long)rec->start,
5295 (unsigned long long)dback->disk_bytenr);
5298 if (dback->bytes != rec->nr) {
5302 fprintf(stderr, "Backref bytes do not match "
5303 "extent backref, bytenr=%llu, ref "
5304 "bytes=%llu, backref bytes=%llu\n",
5305 (unsigned long long)rec->start,
5306 (unsigned long long)rec->nr,
5307 (unsigned long long)dback->bytes);
5310 if (!back->is_data) {
5313 dback = to_data_backref(back);
5314 found += dback->found_ref;
5317 if (found != rec->refs) {
5321 fprintf(stderr, "Incorrect global backref count "
5322 "on %llu found %llu wanted %llu\n",
5323 (unsigned long long)rec->start,
5324 (unsigned long long)found,
5325 (unsigned long long)rec->refs);
5331 static int free_all_extent_backrefs(struct extent_record *rec)
5333 struct extent_backref *back;
5334 struct list_head *cur;
5335 while (!list_empty(&rec->backrefs)) {
5336 cur = rec->backrefs.next;
5337 back = to_extent_backref(cur);
5344 static void free_extent_record_cache(struct cache_tree *extent_cache)
5346 struct cache_extent *cache;
5347 struct extent_record *rec;
5350 cache = first_cache_extent(extent_cache);
5353 rec = container_of(cache, struct extent_record, cache);
5354 remove_cache_extent(extent_cache, cache);
5355 free_all_extent_backrefs(rec);
5360 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
5361 struct extent_record *rec)
5363 if (rec->content_checked && rec->owner_ref_checked &&
5364 rec->extent_item_refs == rec->refs && rec->refs > 0 &&
5365 rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0) &&
5366 !rec->bad_full_backref && !rec->crossing_stripes &&
5367 !rec->wrong_chunk_type) {
5368 remove_cache_extent(extent_cache, &rec->cache);
5369 free_all_extent_backrefs(rec);
5370 list_del_init(&rec->list);
5376 static int check_owner_ref(struct btrfs_root *root,
5377 struct extent_record *rec,
5378 struct extent_buffer *buf)
5380 struct extent_backref *node;
5381 struct tree_backref *back;
5382 struct btrfs_root *ref_root;
5383 struct btrfs_key key;
5384 struct btrfs_path path;
5385 struct extent_buffer *parent;
5390 list_for_each_entry(node, &rec->backrefs, list) {
5393 if (!node->found_ref)
5395 if (node->full_backref)
5397 back = to_tree_backref(node);
5398 if (btrfs_header_owner(buf) == back->root)
5401 BUG_ON(rec->is_root);
5403 /* try to find the block by search corresponding fs tree */
5404 key.objectid = btrfs_header_owner(buf);
5405 key.type = BTRFS_ROOT_ITEM_KEY;
5406 key.offset = (u64)-1;
5408 ref_root = btrfs_read_fs_root(root->fs_info, &key);
5409 if (IS_ERR(ref_root))
5412 level = btrfs_header_level(buf);
5414 btrfs_item_key_to_cpu(buf, &key, 0);
5416 btrfs_node_key_to_cpu(buf, &key, 0);
5418 btrfs_init_path(&path);
5419 path.lowest_level = level + 1;
5420 ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
5424 parent = path.nodes[level + 1];
5425 if (parent && buf->start == btrfs_node_blockptr(parent,
5426 path.slots[level + 1]))
5429 btrfs_release_path(&path);
5430 return found ? 0 : 1;
5433 static int is_extent_tree_record(struct extent_record *rec)
5435 struct list_head *cur = rec->backrefs.next;
5436 struct extent_backref *node;
5437 struct tree_backref *back;
5440 while(cur != &rec->backrefs) {
5441 node = to_extent_backref(cur);
5445 back = to_tree_backref(node);
5446 if (node->full_backref)
5448 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
5455 static int record_bad_block_io(struct btrfs_fs_info *info,
5456 struct cache_tree *extent_cache,
5459 struct extent_record *rec;
5460 struct cache_extent *cache;
5461 struct btrfs_key key;
5463 cache = lookup_cache_extent(extent_cache, start, len);
5467 rec = container_of(cache, struct extent_record, cache);
5468 if (!is_extent_tree_record(rec))
5471 btrfs_disk_key_to_cpu(&key, &rec->parent_key);
5472 return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
5475 static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
5476 struct extent_buffer *buf, int slot)
5478 if (btrfs_header_level(buf)) {
5479 struct btrfs_key_ptr ptr1, ptr2;
5481 read_extent_buffer(buf, &ptr1, btrfs_node_key_ptr_offset(slot),
5482 sizeof(struct btrfs_key_ptr));
5483 read_extent_buffer(buf, &ptr2,
5484 btrfs_node_key_ptr_offset(slot + 1),
5485 sizeof(struct btrfs_key_ptr));
5486 write_extent_buffer(buf, &ptr1,
5487 btrfs_node_key_ptr_offset(slot + 1),
5488 sizeof(struct btrfs_key_ptr));
5489 write_extent_buffer(buf, &ptr2,
5490 btrfs_node_key_ptr_offset(slot),
5491 sizeof(struct btrfs_key_ptr));
5493 struct btrfs_disk_key key;
5494 btrfs_node_key(buf, &key, 0);
5495 btrfs_fixup_low_keys(root, path, &key,
5496 btrfs_header_level(buf) + 1);
5499 struct btrfs_item *item1, *item2;
5500 struct btrfs_key k1, k2;
5501 char *item1_data, *item2_data;
5502 u32 item1_offset, item2_offset, item1_size, item2_size;
5504 item1 = btrfs_item_nr(slot);
5505 item2 = btrfs_item_nr(slot + 1);
5506 btrfs_item_key_to_cpu(buf, &k1, slot);
5507 btrfs_item_key_to_cpu(buf, &k2, slot + 1);
5508 item1_offset = btrfs_item_offset(buf, item1);
5509 item2_offset = btrfs_item_offset(buf, item2);
5510 item1_size = btrfs_item_size(buf, item1);
5511 item2_size = btrfs_item_size(buf, item2);
5513 item1_data = malloc(item1_size);
5516 item2_data = malloc(item2_size);
5522 read_extent_buffer(buf, item1_data, item1_offset, item1_size);
5523 read_extent_buffer(buf, item2_data, item2_offset, item2_size);
5525 write_extent_buffer(buf, item1_data, item2_offset, item2_size);
5526 write_extent_buffer(buf, item2_data, item1_offset, item1_size);
5530 btrfs_set_item_offset(buf, item1, item2_offset);
5531 btrfs_set_item_offset(buf, item2, item1_offset);
5532 btrfs_set_item_size(buf, item1, item2_size);
5533 btrfs_set_item_size(buf, item2, item1_size);
5535 path->slots[0] = slot;
5536 btrfs_set_item_key_unsafe(root, path, &k2);
5537 path->slots[0] = slot + 1;
5538 btrfs_set_item_key_unsafe(root, path, &k1);
5543 static int fix_key_order(struct btrfs_trans_handle *trans,
5544 struct btrfs_root *root,
5545 struct btrfs_path *path)
5547 struct extent_buffer *buf;
5548 struct btrfs_key k1, k2;
5550 int level = path->lowest_level;
5553 buf = path->nodes[level];
5554 for (i = 0; i < btrfs_header_nritems(buf) - 1; i++) {
5556 btrfs_node_key_to_cpu(buf, &k1, i);
5557 btrfs_node_key_to_cpu(buf, &k2, i + 1);
5559 btrfs_item_key_to_cpu(buf, &k1, i);
5560 btrfs_item_key_to_cpu(buf, &k2, i + 1);
5562 if (btrfs_comp_cpu_keys(&k1, &k2) < 0)
5564 ret = swap_values(root, path, buf, i);
5567 btrfs_mark_buffer_dirty(buf);
5573 static int delete_bogus_item(struct btrfs_trans_handle *trans,
5574 struct btrfs_root *root,
5575 struct btrfs_path *path,
5576 struct extent_buffer *buf, int slot)
5578 struct btrfs_key key;
5579 int nritems = btrfs_header_nritems(buf);
5581 btrfs_item_key_to_cpu(buf, &key, slot);
5583 /* These are all the keys we can deal with missing. */
5584 if (key.type != BTRFS_DIR_INDEX_KEY &&
5585 key.type != BTRFS_EXTENT_ITEM_KEY &&
5586 key.type != BTRFS_METADATA_ITEM_KEY &&
5587 key.type != BTRFS_TREE_BLOCK_REF_KEY &&
5588 key.type != BTRFS_EXTENT_DATA_REF_KEY)
5591 printf("Deleting bogus item [%llu,%u,%llu] at slot %d on block %llu\n",
5592 (unsigned long long)key.objectid, key.type,
5593 (unsigned long long)key.offset, slot, buf->start);
5594 memmove_extent_buffer(buf, btrfs_item_nr_offset(slot),
5595 btrfs_item_nr_offset(slot + 1),
5596 sizeof(struct btrfs_item) *
5597 (nritems - slot - 1));
5598 btrfs_set_header_nritems(buf, nritems - 1);
5600 struct btrfs_disk_key disk_key;
5602 btrfs_item_key(buf, &disk_key, 0);
5603 btrfs_fixup_low_keys(root, path, &disk_key, 1);
5605 btrfs_mark_buffer_dirty(buf);
5609 static int fix_item_offset(struct btrfs_trans_handle *trans,
5610 struct btrfs_root *root,
5611 struct btrfs_path *path)
5613 struct extent_buffer *buf;
5617 /* We should only get this for leaves */
5618 BUG_ON(path->lowest_level);
5619 buf = path->nodes[0];
5621 for (i = 0; i < btrfs_header_nritems(buf); i++) {
5622 unsigned int shift = 0, offset;
5624 if (i == 0 && btrfs_item_end_nr(buf, i) !=
5625 BTRFS_LEAF_DATA_SIZE(root)) {
5626 if (btrfs_item_end_nr(buf, i) >
5627 BTRFS_LEAF_DATA_SIZE(root)) {
5628 ret = delete_bogus_item(trans, root, path,
5632 fprintf(stderr, "item is off the end of the "
5633 "leaf, can't fix\n");
5637 shift = BTRFS_LEAF_DATA_SIZE(root) -
5638 btrfs_item_end_nr(buf, i);
5639 } else if (i > 0 && btrfs_item_end_nr(buf, i) !=
5640 btrfs_item_offset_nr(buf, i - 1)) {
5641 if (btrfs_item_end_nr(buf, i) >
5642 btrfs_item_offset_nr(buf, i - 1)) {
5643 ret = delete_bogus_item(trans, root, path,
5647 fprintf(stderr, "items overlap, can't fix\n");
5651 shift = btrfs_item_offset_nr(buf, i - 1) -
5652 btrfs_item_end_nr(buf, i);
5657 printf("Shifting item nr %d by %u bytes in block %llu\n",
5658 i, shift, (unsigned long long)buf->start);
5659 offset = btrfs_item_offset_nr(buf, i);
5660 memmove_extent_buffer(buf,
5661 btrfs_leaf_data(buf) + offset + shift,
5662 btrfs_leaf_data(buf) + offset,
5663 btrfs_item_size_nr(buf, i));
5664 btrfs_set_item_offset(buf, btrfs_item_nr(i),
5666 btrfs_mark_buffer_dirty(buf);
5670 * We may have moved things, in which case we want to exit so we don't
5671 * write those changes out. Once we have proper abort functionality in
5672 * progs this can be changed to something nicer.
5679 * Attempt to fix basic block failures. If we can't fix it for whatever reason
5680 * then just return -EIO.
5682 static int try_to_fix_bad_block(struct btrfs_root *root,
5683 struct extent_buffer *buf,
5684 enum btrfs_tree_block_status status)
5686 struct btrfs_trans_handle *trans;
5687 struct ulist *roots;
5688 struct ulist_node *node;
5689 struct btrfs_root *search_root;
5690 struct btrfs_path path;
5691 struct ulist_iterator iter;
5692 struct btrfs_key root_key, key;
5695 if (status != BTRFS_TREE_BLOCK_BAD_KEY_ORDER &&
5696 status != BTRFS_TREE_BLOCK_INVALID_OFFSETS)
5699 ret = btrfs_find_all_roots(NULL, root->fs_info, buf->start, 0, &roots);
5703 btrfs_init_path(&path);
5704 ULIST_ITER_INIT(&iter);
5705 while ((node = ulist_next(roots, &iter))) {
5706 root_key.objectid = node->val;
5707 root_key.type = BTRFS_ROOT_ITEM_KEY;
5708 root_key.offset = (u64)-1;
5710 search_root = btrfs_read_fs_root(root->fs_info, &root_key);
5717 trans = btrfs_start_transaction(search_root, 0);
5718 if (IS_ERR(trans)) {
5719 ret = PTR_ERR(trans);
5723 path.lowest_level = btrfs_header_level(buf);
5724 path.skip_check_block = 1;
5725 if (path.lowest_level)
5726 btrfs_node_key_to_cpu(buf, &key, 0);
5728 btrfs_item_key_to_cpu(buf, &key, 0);
5729 ret = btrfs_search_slot(trans, search_root, &key, &path, 0, 1);
5732 btrfs_commit_transaction(trans, search_root);
5735 if (status == BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
5736 ret = fix_key_order(trans, search_root, &path);
5737 else if (status == BTRFS_TREE_BLOCK_INVALID_OFFSETS)
5738 ret = fix_item_offset(trans, search_root, &path);
5740 btrfs_commit_transaction(trans, search_root);
5743 btrfs_release_path(&path);
5744 btrfs_commit_transaction(trans, search_root);
5747 btrfs_release_path(&path);
5751 static int check_block(struct btrfs_root *root,
5752 struct cache_tree *extent_cache,
5753 struct extent_buffer *buf, u64 flags)
5755 struct extent_record *rec;
5756 struct cache_extent *cache;
5757 struct btrfs_key key;
5758 enum btrfs_tree_block_status status;
5762 cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
5765 rec = container_of(cache, struct extent_record, cache);
5766 rec->generation = btrfs_header_generation(buf);
5768 level = btrfs_header_level(buf);
5769 if (btrfs_header_nritems(buf) > 0) {
5772 btrfs_item_key_to_cpu(buf, &key, 0);
5774 btrfs_node_key_to_cpu(buf, &key, 0);
5776 rec->info_objectid = key.objectid;
5778 rec->info_level = level;
5780 if (btrfs_is_leaf(buf))
5781 status = btrfs_check_leaf(root, &rec->parent_key, buf);
5783 status = btrfs_check_node(root, &rec->parent_key, buf);
5785 if (status != BTRFS_TREE_BLOCK_CLEAN) {
5787 status = try_to_fix_bad_block(root, buf, status);
5788 if (status != BTRFS_TREE_BLOCK_CLEAN) {
5790 fprintf(stderr, "bad block %llu\n",
5791 (unsigned long long)buf->start);
5794 * Signal to callers we need to start the scan over
5795 * again since we'll have cowed blocks.
5800 rec->content_checked = 1;
5801 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5802 rec->owner_ref_checked = 1;
5804 ret = check_owner_ref(root, rec, buf);
5806 rec->owner_ref_checked = 1;
5810 maybe_free_extent_rec(extent_cache, rec);
5814 static struct tree_backref *find_tree_backref(struct extent_record *rec,
5815 u64 parent, u64 root)
5817 struct list_head *cur = rec->backrefs.next;
5818 struct extent_backref *node;
5819 struct tree_backref *back;
5821 while(cur != &rec->backrefs) {
5822 node = to_extent_backref(cur);
5826 back = to_tree_backref(node);
5828 if (!node->full_backref)
5830 if (parent == back->parent)
5833 if (node->full_backref)
5835 if (back->root == root)
5842 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
5843 u64 parent, u64 root)
5845 struct tree_backref *ref = malloc(sizeof(*ref));
5849 memset(&ref->node, 0, sizeof(ref->node));
5851 ref->parent = parent;
5852 ref->node.full_backref = 1;
5855 ref->node.full_backref = 0;
5857 list_add_tail(&ref->node.list, &rec->backrefs);
5862 static struct data_backref *find_data_backref(struct extent_record *rec,
5863 u64 parent, u64 root,
5864 u64 owner, u64 offset,
5866 u64 disk_bytenr, u64 bytes)
5868 struct list_head *cur = rec->backrefs.next;
5869 struct extent_backref *node;
5870 struct data_backref *back;
5872 while(cur != &rec->backrefs) {
5873 node = to_extent_backref(cur);
5877 back = to_data_backref(node);
5879 if (!node->full_backref)
5881 if (parent == back->parent)
5884 if (node->full_backref)
5886 if (back->root == root && back->owner == owner &&
5887 back->offset == offset) {
5888 if (found_ref && node->found_ref &&
5889 (back->bytes != bytes ||
5890 back->disk_bytenr != disk_bytenr))
5899 static struct data_backref *alloc_data_backref(struct extent_record *rec,
5900 u64 parent, u64 root,
5901 u64 owner, u64 offset,
5904 struct data_backref *ref = malloc(sizeof(*ref));
5908 memset(&ref->node, 0, sizeof(ref->node));
5909 ref->node.is_data = 1;
5912 ref->parent = parent;
5915 ref->node.full_backref = 1;
5919 ref->offset = offset;
5920 ref->node.full_backref = 0;
5922 ref->bytes = max_size;
5925 list_add_tail(&ref->node.list, &rec->backrefs);
5926 if (max_size > rec->max_size)
5927 rec->max_size = max_size;
5931 /* Check if the type of extent matches with its chunk */
5932 static void check_extent_type(struct extent_record *rec)
5934 struct btrfs_block_group_cache *bg_cache;
5936 bg_cache = btrfs_lookup_first_block_group(global_info, rec->start);
5940 /* data extent, check chunk directly*/
5941 if (!rec->metadata) {
5942 if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_DATA))
5943 rec->wrong_chunk_type = 1;
5947 /* metadata extent, check the obvious case first */
5948 if (!(bg_cache->flags & (BTRFS_BLOCK_GROUP_SYSTEM |
5949 BTRFS_BLOCK_GROUP_METADATA))) {
5950 rec->wrong_chunk_type = 1;
5955 * Check SYSTEM extent, as it's also marked as metadata, we can only
5956 * make sure it's a SYSTEM extent by its backref
5958 if (!list_empty(&rec->backrefs)) {
5959 struct extent_backref *node;
5960 struct tree_backref *tback;
5963 node = to_extent_backref(rec->backrefs.next);
5964 if (node->is_data) {
5965 /* tree block shouldn't have data backref */
5966 rec->wrong_chunk_type = 1;
5969 tback = container_of(node, struct tree_backref, node);
5971 if (tback->root == BTRFS_CHUNK_TREE_OBJECTID)
5972 bg_type = BTRFS_BLOCK_GROUP_SYSTEM;
5974 bg_type = BTRFS_BLOCK_GROUP_METADATA;
5975 if (!(bg_cache->flags & bg_type))
5976 rec->wrong_chunk_type = 1;
5981 * Allocate a new extent record, fill default values from @tmpl and insert int
5982 * @extent_cache. Caller is supposed to make sure the [start,nr) is not in
5983 * the cache, otherwise it fails.
5985 static int add_extent_rec_nolookup(struct cache_tree *extent_cache,
5986 struct extent_record *tmpl)
5988 struct extent_record *rec;
5991 rec = malloc(sizeof(*rec));
5994 rec->start = tmpl->start;
5995 rec->max_size = tmpl->max_size;
5996 rec->nr = max(tmpl->nr, tmpl->max_size);
5997 rec->found_rec = tmpl->found_rec;
5998 rec->content_checked = tmpl->content_checked;
5999 rec->owner_ref_checked = tmpl->owner_ref_checked;
6000 rec->num_duplicates = 0;
6001 rec->metadata = tmpl->metadata;
6002 rec->flag_block_full_backref = FLAG_UNSET;
6003 rec->bad_full_backref = 0;
6004 rec->crossing_stripes = 0;
6005 rec->wrong_chunk_type = 0;
6006 rec->is_root = tmpl->is_root;
6007 rec->refs = tmpl->refs;
6008 rec->extent_item_refs = tmpl->extent_item_refs;
6009 rec->parent_generation = tmpl->parent_generation;
6010 INIT_LIST_HEAD(&rec->backrefs);
6011 INIT_LIST_HEAD(&rec->dups);
6012 INIT_LIST_HEAD(&rec->list);
6013 memcpy(&rec->parent_key, &tmpl->parent_key, sizeof(tmpl->parent_key));
6014 rec->cache.start = tmpl->start;
6015 rec->cache.size = tmpl->nr;
6016 ret = insert_cache_extent(extent_cache, &rec->cache);
6021 bytes_used += rec->nr;
6024 rec->crossing_stripes = check_crossing_stripes(global_info,
6025 rec->start, global_info->tree_root->nodesize);
6026 check_extent_type(rec);
6031 * Lookup and modify an extent, some values of @tmpl are interpreted verbatim,
6033 * - refs - if found, increase refs
6034 * - is_root - if found, set
6035 * - content_checked - if found, set
6036 * - owner_ref_checked - if found, set
6038 * If not found, create a new one, initialize and insert.
6040 static int add_extent_rec(struct cache_tree *extent_cache,
6041 struct extent_record *tmpl)
6043 struct extent_record *rec;
6044 struct cache_extent *cache;
6048 cache = lookup_cache_extent(extent_cache, tmpl->start, tmpl->nr);
6050 rec = container_of(cache, struct extent_record, cache);
6054 rec->nr = max(tmpl->nr, tmpl->max_size);
6057 * We need to make sure to reset nr to whatever the extent
6058 * record says was the real size, this way we can compare it to
6061 if (tmpl->found_rec) {
6062 if (tmpl->start != rec->start || rec->found_rec) {
6063 struct extent_record *tmp;
6066 if (list_empty(&rec->list))
6067 list_add_tail(&rec->list,
6068 &duplicate_extents);
6071 * We have to do this song and dance in case we
6072 * find an extent record that falls inside of
6073 * our current extent record but does not have
6074 * the same objectid.
6076 tmp = malloc(sizeof(*tmp));
6079 tmp->start = tmpl->start;
6080 tmp->max_size = tmpl->max_size;
6083 tmp->metadata = tmpl->metadata;
6084 tmp->extent_item_refs = tmpl->extent_item_refs;
6085 INIT_LIST_HEAD(&tmp->list);
6086 list_add_tail(&tmp->list, &rec->dups);
6087 rec->num_duplicates++;
6094 if (tmpl->extent_item_refs && !dup) {
6095 if (rec->extent_item_refs) {
6096 fprintf(stderr, "block %llu rec "
6097 "extent_item_refs %llu, passed %llu\n",
6098 (unsigned long long)tmpl->start,
6099 (unsigned long long)
6100 rec->extent_item_refs,
6101 (unsigned long long)tmpl->extent_item_refs);
6103 rec->extent_item_refs = tmpl->extent_item_refs;
6107 if (tmpl->content_checked)
6108 rec->content_checked = 1;
6109 if (tmpl->owner_ref_checked)
6110 rec->owner_ref_checked = 1;
6111 memcpy(&rec->parent_key, &tmpl->parent_key,
6112 sizeof(tmpl->parent_key));
6113 if (tmpl->parent_generation)
6114 rec->parent_generation = tmpl->parent_generation;
6115 if (rec->max_size < tmpl->max_size)
6116 rec->max_size = tmpl->max_size;
6119 * A metadata extent can't cross stripe_len boundary, otherwise
6120 * kernel scrub won't be able to handle it.
6121 * As now stripe_len is fixed to BTRFS_STRIPE_LEN, just check
6125 rec->crossing_stripes = check_crossing_stripes(
6126 global_info, rec->start,
6127 global_info->tree_root->nodesize);
6128 check_extent_type(rec);
6129 maybe_free_extent_rec(extent_cache, rec);
6133 ret = add_extent_rec_nolookup(extent_cache, tmpl);
6138 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
6139 u64 parent, u64 root, int found_ref)
6141 struct extent_record *rec;
6142 struct tree_backref *back;
6143 struct cache_extent *cache;
6146 cache = lookup_cache_extent(extent_cache, bytenr, 1);
6148 struct extent_record tmpl;
6150 memset(&tmpl, 0, sizeof(tmpl));
6151 tmpl.start = bytenr;
6155 ret = add_extent_rec_nolookup(extent_cache, &tmpl);
6159 /* really a bug in cache_extent implement now */
6160 cache = lookup_cache_extent(extent_cache, bytenr, 1);
6165 rec = container_of(cache, struct extent_record, cache);
6166 if (rec->start != bytenr) {
6168 * Several cause, from unaligned bytenr to over lapping extents
6173 back = find_tree_backref(rec, parent, root);
6175 back = alloc_tree_backref(rec, parent, root);
6181 if (back->node.found_ref) {
6182 fprintf(stderr, "Extent back ref already exists "
6183 "for %llu parent %llu root %llu \n",
6184 (unsigned long long)bytenr,
6185 (unsigned long long)parent,
6186 (unsigned long long)root);
6188 back->node.found_ref = 1;
6190 if (back->node.found_extent_tree) {
6191 fprintf(stderr, "Extent back ref already exists "
6192 "for %llu parent %llu root %llu \n",
6193 (unsigned long long)bytenr,
6194 (unsigned long long)parent,
6195 (unsigned long long)root);
6197 back->node.found_extent_tree = 1;
6199 check_extent_type(rec);
6200 maybe_free_extent_rec(extent_cache, rec);
6204 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
6205 u64 parent, u64 root, u64 owner, u64 offset,
6206 u32 num_refs, int found_ref, u64 max_size)
6208 struct extent_record *rec;
6209 struct data_backref *back;
6210 struct cache_extent *cache;
6213 cache = lookup_cache_extent(extent_cache, bytenr, 1);
6215 struct extent_record tmpl;
6217 memset(&tmpl, 0, sizeof(tmpl));
6218 tmpl.start = bytenr;
6220 tmpl.max_size = max_size;
6222 ret = add_extent_rec_nolookup(extent_cache, &tmpl);
6226 cache = lookup_cache_extent(extent_cache, bytenr, 1);
6231 rec = container_of(cache, struct extent_record, cache);
6232 if (rec->max_size < max_size)
6233 rec->max_size = max_size;
6236 * If found_ref is set then max_size is the real size and must match the
6237 * existing refs. So if we have already found a ref then we need to
6238 * make sure that this ref matches the existing one, otherwise we need
6239 * to add a new backref so we can notice that the backrefs don't match
6240 * and we need to figure out who is telling the truth. This is to
6241 * account for that awful fsync bug I introduced where we'd end up with
6242 * a btrfs_file_extent_item that would have its length include multiple
6243 * prealloc extents or point inside of a prealloc extent.
6245 back = find_data_backref(rec, parent, root, owner, offset, found_ref,
6248 back = alloc_data_backref(rec, parent, root, owner, offset,
6254 BUG_ON(num_refs != 1);
6255 if (back->node.found_ref)
6256 BUG_ON(back->bytes != max_size);
6257 back->node.found_ref = 1;
6258 back->found_ref += 1;
6259 back->bytes = max_size;
6260 back->disk_bytenr = bytenr;
6262 rec->content_checked = 1;
6263 rec->owner_ref_checked = 1;
6265 if (back->node.found_extent_tree) {
6266 fprintf(stderr, "Extent back ref already exists "
6267 "for %llu parent %llu root %llu "
6268 "owner %llu offset %llu num_refs %lu\n",
6269 (unsigned long long)bytenr,
6270 (unsigned long long)parent,
6271 (unsigned long long)root,
6272 (unsigned long long)owner,
6273 (unsigned long long)offset,
6274 (unsigned long)num_refs);
6276 back->num_refs = num_refs;
6277 back->node.found_extent_tree = 1;
6279 maybe_free_extent_rec(extent_cache, rec);
6283 static int add_pending(struct cache_tree *pending,
6284 struct cache_tree *seen, u64 bytenr, u32 size)
6287 ret = add_cache_extent(seen, bytenr, size);
6290 add_cache_extent(pending, bytenr, size);
6294 static int pick_next_pending(struct cache_tree *pending,
6295 struct cache_tree *reada,
6296 struct cache_tree *nodes,
6297 u64 last, struct block_info *bits, int bits_nr,
6300 unsigned long node_start = last;
6301 struct cache_extent *cache;
6304 cache = search_cache_extent(reada, 0);
6306 bits[0].start = cache->start;
6307 bits[0].size = cache->size;
6312 if (node_start > 32768)
6313 node_start -= 32768;
6315 cache = search_cache_extent(nodes, node_start);
6317 cache = search_cache_extent(nodes, 0);
6320 cache = search_cache_extent(pending, 0);
6325 bits[ret].start = cache->start;
6326 bits[ret].size = cache->size;
6327 cache = next_cache_extent(cache);
6329 } while (cache && ret < bits_nr);
6335 bits[ret].start = cache->start;
6336 bits[ret].size = cache->size;
6337 cache = next_cache_extent(cache);
6339 } while (cache && ret < bits_nr);
6341 if (bits_nr - ret > 8) {
6342 u64 lookup = bits[0].start + bits[0].size;
6343 struct cache_extent *next;
6344 next = search_cache_extent(pending, lookup);
6346 if (next->start - lookup > 32768)
6348 bits[ret].start = next->start;
6349 bits[ret].size = next->size;
6350 lookup = next->start + next->size;
6354 next = next_cache_extent(next);
6362 static void free_chunk_record(struct cache_extent *cache)
6364 struct chunk_record *rec;
6366 rec = container_of(cache, struct chunk_record, cache);
6367 list_del_init(&rec->list);
6368 list_del_init(&rec->dextents);
6372 void free_chunk_cache_tree(struct cache_tree *chunk_cache)
6374 cache_tree_free_extents(chunk_cache, free_chunk_record);
6377 static void free_device_record(struct rb_node *node)
6379 struct device_record *rec;
6381 rec = container_of(node, struct device_record, node);
6385 FREE_RB_BASED_TREE(device_cache, free_device_record);
6387 int insert_block_group_record(struct block_group_tree *tree,
6388 struct block_group_record *bg_rec)
6392 ret = insert_cache_extent(&tree->tree, &bg_rec->cache);
6396 list_add_tail(&bg_rec->list, &tree->block_groups);
6400 static void free_block_group_record(struct cache_extent *cache)
6402 struct block_group_record *rec;
6404 rec = container_of(cache, struct block_group_record, cache);
6405 list_del_init(&rec->list);
6409 void free_block_group_tree(struct block_group_tree *tree)
6411 cache_tree_free_extents(&tree->tree, free_block_group_record);
6414 int insert_device_extent_record(struct device_extent_tree *tree,
6415 struct device_extent_record *de_rec)
6420 * Device extent is a bit different from the other extents, because
6421 * the extents which belong to the different devices may have the
6422 * same start and size, so we need use the special extent cache
6423 * search/insert functions.
6425 ret = insert_cache_extent2(&tree->tree, &de_rec->cache);
6429 list_add_tail(&de_rec->chunk_list, &tree->no_chunk_orphans);
6430 list_add_tail(&de_rec->device_list, &tree->no_device_orphans);
6434 static void free_device_extent_record(struct cache_extent *cache)
6436 struct device_extent_record *rec;
6438 rec = container_of(cache, struct device_extent_record, cache);
6439 if (!list_empty(&rec->chunk_list))
6440 list_del_init(&rec->chunk_list);
6441 if (!list_empty(&rec->device_list))
6442 list_del_init(&rec->device_list);
6446 void free_device_extent_tree(struct device_extent_tree *tree)
6448 cache_tree_free_extents(&tree->tree, free_device_extent_record);
6451 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6452 static int process_extent_ref_v0(struct cache_tree *extent_cache,
6453 struct extent_buffer *leaf, int slot)
6455 struct btrfs_extent_ref_v0 *ref0;
6456 struct btrfs_key key;
6459 btrfs_item_key_to_cpu(leaf, &key, slot);
6460 ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
6461 if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
6462 ret = add_tree_backref(extent_cache, key.objectid, key.offset,
6465 ret = add_data_backref(extent_cache, key.objectid, key.offset,
6466 0, 0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
6472 struct chunk_record *btrfs_new_chunk_record(struct extent_buffer *leaf,
6473 struct btrfs_key *key,
6476 struct btrfs_chunk *ptr;
6477 struct chunk_record *rec;
6480 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6481 num_stripes = btrfs_chunk_num_stripes(leaf, ptr);
6483 rec = calloc(1, btrfs_chunk_record_size(num_stripes));
6485 fprintf(stderr, "memory allocation failed\n");
6489 INIT_LIST_HEAD(&rec->list);
6490 INIT_LIST_HEAD(&rec->dextents);
6493 rec->cache.start = key->offset;
6494 rec->cache.size = btrfs_chunk_length(leaf, ptr);
6496 rec->generation = btrfs_header_generation(leaf);
6498 rec->objectid = key->objectid;
6499 rec->type = key->type;
6500 rec->offset = key->offset;
6502 rec->length = rec->cache.size;
6503 rec->owner = btrfs_chunk_owner(leaf, ptr);
6504 rec->stripe_len = btrfs_chunk_stripe_len(leaf, ptr);
6505 rec->type_flags = btrfs_chunk_type(leaf, ptr);
6506 rec->io_width = btrfs_chunk_io_width(leaf, ptr);
6507 rec->io_align = btrfs_chunk_io_align(leaf, ptr);
6508 rec->sector_size = btrfs_chunk_sector_size(leaf, ptr);
6509 rec->num_stripes = num_stripes;
6510 rec->sub_stripes = btrfs_chunk_sub_stripes(leaf, ptr);
6512 for (i = 0; i < rec->num_stripes; ++i) {
6513 rec->stripes[i].devid =
6514 btrfs_stripe_devid_nr(leaf, ptr, i);
6515 rec->stripes[i].offset =
6516 btrfs_stripe_offset_nr(leaf, ptr, i);
6517 read_extent_buffer(leaf, rec->stripes[i].dev_uuid,
6518 (unsigned long)btrfs_stripe_dev_uuid_nr(ptr, i),
6525 static int process_chunk_item(struct cache_tree *chunk_cache,
6526 struct btrfs_key *key, struct extent_buffer *eb,
6529 struct chunk_record *rec;
6530 struct btrfs_chunk *chunk;
6533 chunk = btrfs_item_ptr(eb, slot, struct btrfs_chunk);
6535 * Do extra check for this chunk item,
6537 * It's still possible one can craft a leaf with CHUNK_ITEM, with
6538 * wrong onwer(3) out of chunk tree, to pass both chunk tree check
6539 * and owner<->key_type check.
6541 ret = btrfs_check_chunk_valid(global_info->tree_root, eb, chunk, slot,
6544 error("chunk(%llu, %llu) is not valid, ignore it",
6545 key->offset, btrfs_chunk_length(eb, chunk));
6548 rec = btrfs_new_chunk_record(eb, key, slot);
6549 ret = insert_cache_extent(chunk_cache, &rec->cache);
6551 fprintf(stderr, "Chunk[%llu, %llu] existed.\n",
6552 rec->offset, rec->length);
6559 static int process_device_item(struct rb_root *dev_cache,
6560 struct btrfs_key *key, struct extent_buffer *eb, int slot)
6562 struct btrfs_dev_item *ptr;
6563 struct device_record *rec;
6566 ptr = btrfs_item_ptr(eb,
6567 slot, struct btrfs_dev_item);
6569 rec = malloc(sizeof(*rec));
6571 fprintf(stderr, "memory allocation failed\n");
6575 rec->devid = key->offset;
6576 rec->generation = btrfs_header_generation(eb);
6578 rec->objectid = key->objectid;
6579 rec->type = key->type;
6580 rec->offset = key->offset;
6582 rec->devid = btrfs_device_id(eb, ptr);
6583 rec->total_byte = btrfs_device_total_bytes(eb, ptr);
6584 rec->byte_used = btrfs_device_bytes_used(eb, ptr);
6586 ret = rb_insert(dev_cache, &rec->node, device_record_compare);
6588 fprintf(stderr, "Device[%llu] existed.\n", rec->devid);
6595 struct block_group_record *
6596 btrfs_new_block_group_record(struct extent_buffer *leaf, struct btrfs_key *key,
6599 struct btrfs_block_group_item *ptr;
6600 struct block_group_record *rec;
6602 rec = calloc(1, sizeof(*rec));
6604 fprintf(stderr, "memory allocation failed\n");
6608 rec->cache.start = key->objectid;
6609 rec->cache.size = key->offset;
6611 rec->generation = btrfs_header_generation(leaf);
6613 rec->objectid = key->objectid;
6614 rec->type = key->type;
6615 rec->offset = key->offset;
6617 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item);
6618 rec->flags = btrfs_disk_block_group_flags(leaf, ptr);
6620 INIT_LIST_HEAD(&rec->list);
6625 static int process_block_group_item(struct block_group_tree *block_group_cache,
6626 struct btrfs_key *key,
6627 struct extent_buffer *eb, int slot)
6629 struct block_group_record *rec;
6632 rec = btrfs_new_block_group_record(eb, key, slot);
6633 ret = insert_block_group_record(block_group_cache, rec);
6635 fprintf(stderr, "Block Group[%llu, %llu] existed.\n",
6636 rec->objectid, rec->offset);
6643 struct device_extent_record *
6644 btrfs_new_device_extent_record(struct extent_buffer *leaf,
6645 struct btrfs_key *key, int slot)
6647 struct device_extent_record *rec;
6648 struct btrfs_dev_extent *ptr;
6650 rec = calloc(1, sizeof(*rec));
6652 fprintf(stderr, "memory allocation failed\n");
6656 rec->cache.objectid = key->objectid;
6657 rec->cache.start = key->offset;
6659 rec->generation = btrfs_header_generation(leaf);
6661 rec->objectid = key->objectid;
6662 rec->type = key->type;
6663 rec->offset = key->offset;
6665 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
6666 rec->chunk_objecteid =
6667 btrfs_dev_extent_chunk_objectid(leaf, ptr);
6669 btrfs_dev_extent_chunk_offset(leaf, ptr);
6670 rec->length = btrfs_dev_extent_length(leaf, ptr);
6671 rec->cache.size = rec->length;
6673 INIT_LIST_HEAD(&rec->chunk_list);
6674 INIT_LIST_HEAD(&rec->device_list);
6680 process_device_extent_item(struct device_extent_tree *dev_extent_cache,
6681 struct btrfs_key *key, struct extent_buffer *eb,
6684 struct device_extent_record *rec;
6687 rec = btrfs_new_device_extent_record(eb, key, slot);
6688 ret = insert_device_extent_record(dev_extent_cache, rec);
6691 "Device extent[%llu, %llu, %llu] existed.\n",
6692 rec->objectid, rec->offset, rec->length);
6699 static int process_extent_item(struct btrfs_root *root,
6700 struct cache_tree *extent_cache,
6701 struct extent_buffer *eb, int slot)
6703 struct btrfs_extent_item *ei;
6704 struct btrfs_extent_inline_ref *iref;
6705 struct btrfs_extent_data_ref *dref;
6706 struct btrfs_shared_data_ref *sref;
6707 struct btrfs_key key;
6708 struct extent_record tmpl;
6713 u32 item_size = btrfs_item_size_nr(eb, slot);
6719 btrfs_item_key_to_cpu(eb, &key, slot);
6721 if (key.type == BTRFS_METADATA_ITEM_KEY) {
6723 num_bytes = root->nodesize;
6725 num_bytes = key.offset;
6728 if (!IS_ALIGNED(key.objectid, root->sectorsize)) {
6729 error("ignoring invalid extent, bytenr %llu is not aligned to %u",
6730 key.objectid, root->sectorsize);
6733 if (item_size < sizeof(*ei)) {
6734 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6735 struct btrfs_extent_item_v0 *ei0;
6736 BUG_ON(item_size != sizeof(*ei0));
6737 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
6738 refs = btrfs_extent_refs_v0(eb, ei0);
6742 memset(&tmpl, 0, sizeof(tmpl));
6743 tmpl.start = key.objectid;
6744 tmpl.nr = num_bytes;
6745 tmpl.extent_item_refs = refs;
6746 tmpl.metadata = metadata;
6748 tmpl.max_size = num_bytes;
6750 return add_extent_rec(extent_cache, &tmpl);
6753 ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
6754 refs = btrfs_extent_refs(eb, ei);
6755 if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK)
6759 if (metadata && num_bytes != root->nodesize) {
6760 error("ignore invalid metadata extent, length %llu does not equal to %u",
6761 num_bytes, root->nodesize);
6764 if (!metadata && !IS_ALIGNED(num_bytes, root->sectorsize)) {
6765 error("ignore invalid data extent, length %llu is not aligned to %u",
6766 num_bytes, root->sectorsize);
6770 memset(&tmpl, 0, sizeof(tmpl));
6771 tmpl.start = key.objectid;
6772 tmpl.nr = num_bytes;
6773 tmpl.extent_item_refs = refs;
6774 tmpl.metadata = metadata;
6776 tmpl.max_size = num_bytes;
6777 add_extent_rec(extent_cache, &tmpl);
6779 ptr = (unsigned long)(ei + 1);
6780 if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
6781 key.type == BTRFS_EXTENT_ITEM_KEY)
6782 ptr += sizeof(struct btrfs_tree_block_info);
6784 end = (unsigned long)ei + item_size;
6786 iref = (struct btrfs_extent_inline_ref *)ptr;
6787 type = btrfs_extent_inline_ref_type(eb, iref);
6788 offset = btrfs_extent_inline_ref_offset(eb, iref);
6790 case BTRFS_TREE_BLOCK_REF_KEY:
6791 ret = add_tree_backref(extent_cache, key.objectid,
6794 error("add_tree_backref failed: %s",
6797 case BTRFS_SHARED_BLOCK_REF_KEY:
6798 ret = add_tree_backref(extent_cache, key.objectid,
6801 error("add_tree_backref failed: %s",
6804 case BTRFS_EXTENT_DATA_REF_KEY:
6805 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
6806 add_data_backref(extent_cache, key.objectid, 0,
6807 btrfs_extent_data_ref_root(eb, dref),
6808 btrfs_extent_data_ref_objectid(eb,
6810 btrfs_extent_data_ref_offset(eb, dref),
6811 btrfs_extent_data_ref_count(eb, dref),
6814 case BTRFS_SHARED_DATA_REF_KEY:
6815 sref = (struct btrfs_shared_data_ref *)(iref + 1);
6816 add_data_backref(extent_cache, key.objectid, offset,
6818 btrfs_shared_data_ref_count(eb, sref),
6822 fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
6823 key.objectid, key.type, num_bytes);
6826 ptr += btrfs_extent_inline_ref_size(type);
6833 static int check_cache_range(struct btrfs_root *root,
6834 struct btrfs_block_group_cache *cache,
6835 u64 offset, u64 bytes)
6837 struct btrfs_free_space *entry;
6843 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
6844 bytenr = btrfs_sb_offset(i);
6845 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
6846 cache->key.objectid, bytenr, 0,
6847 &logical, &nr, &stripe_len);
6852 if (logical[nr] + stripe_len <= offset)
6854 if (offset + bytes <= logical[nr])
6856 if (logical[nr] == offset) {
6857 if (stripe_len >= bytes) {
6861 bytes -= stripe_len;
6862 offset += stripe_len;
6863 } else if (logical[nr] < offset) {
6864 if (logical[nr] + stripe_len >=
6869 bytes = (offset + bytes) -
6870 (logical[nr] + stripe_len);
6871 offset = logical[nr] + stripe_len;
6874 * Could be tricky, the super may land in the
6875 * middle of the area we're checking. First
6876 * check the easiest case, it's at the end.
6878 if (logical[nr] + stripe_len >=
6880 bytes = logical[nr] - offset;
6884 /* Check the left side */
6885 ret = check_cache_range(root, cache,
6887 logical[nr] - offset);
6893 /* Now we continue with the right side */
6894 bytes = (offset + bytes) -
6895 (logical[nr] + stripe_len);
6896 offset = logical[nr] + stripe_len;
6903 entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
6905 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
6906 offset, offset+bytes);
6910 if (entry->offset != offset) {
6911 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
6916 if (entry->bytes != bytes) {
6917 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
6918 bytes, entry->bytes, offset);
6922 unlink_free_space(cache->free_space_ctl, entry);
6927 static int verify_space_cache(struct btrfs_root *root,
6928 struct btrfs_block_group_cache *cache)
6930 struct btrfs_path path;
6931 struct extent_buffer *leaf;
6932 struct btrfs_key key;
6936 root = root->fs_info->extent_root;
6938 last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
6940 btrfs_init_path(&path);
6941 key.objectid = last;
6943 key.type = BTRFS_EXTENT_ITEM_KEY;
6944 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
6949 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
6950 ret = btrfs_next_leaf(root, &path);
6958 leaf = path.nodes[0];
6959 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
6960 if (key.objectid >= cache->key.offset + cache->key.objectid)
6962 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
6963 key.type != BTRFS_METADATA_ITEM_KEY) {
6968 if (last == key.objectid) {
6969 if (key.type == BTRFS_EXTENT_ITEM_KEY)
6970 last = key.objectid + key.offset;
6972 last = key.objectid + root->nodesize;
6977 ret = check_cache_range(root, cache, last,
6978 key.objectid - last);
6981 if (key.type == BTRFS_EXTENT_ITEM_KEY)
6982 last = key.objectid + key.offset;
6984 last = key.objectid + root->nodesize;
6988 if (last < cache->key.objectid + cache->key.offset)
6989 ret = check_cache_range(root, cache, last,
6990 cache->key.objectid +
6991 cache->key.offset - last);
6994 btrfs_release_path(&path);
6997 !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
6998 fprintf(stderr, "There are still entries left in the space "
7006 static int check_space_cache(struct btrfs_root *root)
7008 struct btrfs_block_group_cache *cache;
7009 u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
7013 if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
7014 btrfs_super_generation(root->fs_info->super_copy) !=
7015 btrfs_super_cache_generation(root->fs_info->super_copy)) {
7016 printf("cache and super generation don't match, space cache "
7017 "will be invalidated\n");
7021 if (ctx.progress_enabled) {
7022 ctx.tp = TASK_FREE_SPACE;
7023 task_start(ctx.info);
7027 cache = btrfs_lookup_first_block_group(root->fs_info, start);
7031 start = cache->key.objectid + cache->key.offset;
7032 if (!cache->free_space_ctl) {
7033 if (btrfs_init_free_space_ctl(cache,
7034 root->sectorsize)) {
7039 btrfs_remove_free_space_cache(cache);
7042 if (btrfs_fs_compat_ro(root->fs_info, FREE_SPACE_TREE)) {
7043 ret = exclude_super_stripes(root, cache);
7045 fprintf(stderr, "could not exclude super stripes: %s\n",
7050 ret = load_free_space_tree(root->fs_info, cache);
7051 free_excluded_extents(root, cache);
7053 fprintf(stderr, "could not load free space tree: %s\n",
7060 ret = load_free_space_cache(root->fs_info, cache);
7065 ret = verify_space_cache(root, cache);
7067 fprintf(stderr, "cache appears valid but isn't %Lu\n",
7068 cache->key.objectid);
7073 task_stop(ctx.info);
7075 return error ? -EINVAL : 0;
7078 static int check_extent_csums(struct btrfs_root *root, u64 bytenr,
7079 u64 num_bytes, unsigned long leaf_offset,
7080 struct extent_buffer *eb) {
7083 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7085 unsigned long csum_offset;
7089 u64 data_checked = 0;
7095 if (num_bytes % root->sectorsize)
7098 data = malloc(num_bytes);
7102 while (offset < num_bytes) {
7105 read_len = num_bytes - offset;
7106 /* read as much space once a time */
7107 ret = read_extent_data(root, data + offset,
7108 bytenr + offset, &read_len, mirror);
7112 /* verify every 4k data's checksum */
7113 while (data_checked < read_len) {
7115 tmp = offset + data_checked;
7117 csum = btrfs_csum_data((char *)data + tmp,
7118 csum, root->sectorsize);
7119 btrfs_csum_final(csum, (u8 *)&csum);
7121 csum_offset = leaf_offset +
7122 tmp / root->sectorsize * csum_size;
7123 read_extent_buffer(eb, (char *)&csum_expected,
7124 csum_offset, csum_size);
7125 /* try another mirror */
7126 if (csum != csum_expected) {
7127 fprintf(stderr, "mirror %d bytenr %llu csum %u expected csum %u\n",
7128 mirror, bytenr + tmp,
7129 csum, csum_expected);
7130 num_copies = btrfs_num_copies(
7131 &root->fs_info->mapping_tree,
7133 if (mirror < num_copies - 1) {
7138 data_checked += root->sectorsize;
7147 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
7150 struct btrfs_path path;
7151 struct extent_buffer *leaf;
7152 struct btrfs_key key;
7155 btrfs_init_path(&path);
7156 key.objectid = bytenr;
7157 key.type = BTRFS_EXTENT_ITEM_KEY;
7158 key.offset = (u64)-1;
7161 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, &path,
7164 fprintf(stderr, "Error looking up extent record %d\n", ret);
7165 btrfs_release_path(&path);
7168 if (path.slots[0] > 0) {
7171 ret = btrfs_prev_leaf(root, &path);
7174 } else if (ret > 0) {
7181 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
7184 * Block group items come before extent items if they have the same
7185 * bytenr, so walk back one more just in case. Dear future traveller,
7186 * first congrats on mastering time travel. Now if it's not too much
7187 * trouble could you go back to 2006 and tell Chris to make the
7188 * BLOCK_GROUP_ITEM_KEY (and BTRFS_*_REF_KEY) lower than the
7189 * EXTENT_ITEM_KEY please?
7191 while (key.type > BTRFS_EXTENT_ITEM_KEY) {
7192 if (path.slots[0] > 0) {
7195 ret = btrfs_prev_leaf(root, &path);
7198 } else if (ret > 0) {
7203 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
7207 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
7208 ret = btrfs_next_leaf(root, &path);
7210 fprintf(stderr, "Error going to next leaf "
7212 btrfs_release_path(&path);
7218 leaf = path.nodes[0];
7219 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
7220 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
7224 if (key.objectid + key.offset < bytenr) {
7228 if (key.objectid > bytenr + num_bytes)
7231 if (key.objectid == bytenr) {
7232 if (key.offset >= num_bytes) {
7236 num_bytes -= key.offset;
7237 bytenr += key.offset;
7238 } else if (key.objectid < bytenr) {
7239 if (key.objectid + key.offset >= bytenr + num_bytes) {
7243 num_bytes = (bytenr + num_bytes) -
7244 (key.objectid + key.offset);
7245 bytenr = key.objectid + key.offset;
7247 if (key.objectid + key.offset < bytenr + num_bytes) {
7248 u64 new_start = key.objectid + key.offset;
7249 u64 new_bytes = bytenr + num_bytes - new_start;
7252 * Weird case, the extent is in the middle of
7253 * our range, we'll have to search one side
7254 * and then the other. Not sure if this happens
7255 * in real life, but no harm in coding it up
7256 * anyway just in case.
7258 btrfs_release_path(&path);
7259 ret = check_extent_exists(root, new_start,
7262 fprintf(stderr, "Right section didn't "
7266 num_bytes = key.objectid - bytenr;
7269 num_bytes = key.objectid - bytenr;
7276 if (num_bytes && !ret) {
7277 fprintf(stderr, "There are no extents for csum range "
7278 "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
7282 btrfs_release_path(&path);
7286 static int check_csums(struct btrfs_root *root)
7288 struct btrfs_path path;
7289 struct extent_buffer *leaf;
7290 struct btrfs_key key;
7291 u64 offset = 0, num_bytes = 0;
7292 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7296 unsigned long leaf_offset;
7298 root = root->fs_info->csum_root;
7299 if (!extent_buffer_uptodate(root->node)) {
7300 fprintf(stderr, "No valid csum tree found\n");
7304 btrfs_init_path(&path);
7305 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
7306 key.type = BTRFS_EXTENT_CSUM_KEY;
7308 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
7310 fprintf(stderr, "Error searching csum tree %d\n", ret);
7311 btrfs_release_path(&path);
7315 if (ret > 0 && path.slots[0])
7320 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
7321 ret = btrfs_next_leaf(root, &path);
7323 fprintf(stderr, "Error going to next leaf "
7330 leaf = path.nodes[0];
7332 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
7333 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
7338 data_len = (btrfs_item_size_nr(leaf, path.slots[0]) /
7339 csum_size) * root->sectorsize;
7340 if (!check_data_csum)
7341 goto skip_csum_check;
7342 leaf_offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
7343 ret = check_extent_csums(root, key.offset, data_len,
7349 offset = key.offset;
7350 } else if (key.offset != offset + num_bytes) {
7351 ret = check_extent_exists(root, offset, num_bytes);
7353 fprintf(stderr, "Csum exists for %Lu-%Lu but "
7354 "there is no extent record\n",
7355 offset, offset+num_bytes);
7358 offset = key.offset;
7361 num_bytes += data_len;
7365 btrfs_release_path(&path);
7369 static int is_dropped_key(struct btrfs_key *key,
7370 struct btrfs_key *drop_key) {
7371 if (key->objectid < drop_key->objectid)
7373 else if (key->objectid == drop_key->objectid) {
7374 if (key->type < drop_key->type)
7376 else if (key->type == drop_key->type) {
7377 if (key->offset < drop_key->offset)
7385 * Here are the rules for FULL_BACKREF.
7387 * 1) If BTRFS_HEADER_FLAG_RELOC is set then we have FULL_BACKREF set.
7388 * 2) If btrfs_header_owner(buf) no longer points to buf then we have
7390 * 3) We cowed the block walking down a reloc tree. This is impossible to tell
7391 * if it happened after the relocation occurred since we'll have dropped the
7392 * reloc root, so it's entirely possible to have FULL_BACKREF set on buf and
7393 * have no real way to know for sure.
7395 * We process the blocks one root at a time, and we start from the lowest root
7396 * objectid and go to the highest. So we can just lookup the owner backref for
7397 * the record and if we don't find it then we know it doesn't exist and we have
7400 * FIXME: if we ever start reclaiming root objectid's then we need to fix this
7401 * assumption and simply indicate that we _think_ that the FULL BACKREF needs to
7402 * be set or not and then we can check later once we've gathered all the refs.
7404 static int calc_extent_flag(struct btrfs_root *root,
7405 struct cache_tree *extent_cache,
7406 struct extent_buffer *buf,
7407 struct root_item_record *ri,
7410 struct extent_record *rec;
7411 struct cache_extent *cache;
7412 struct tree_backref *tback;
7415 cache = lookup_cache_extent(extent_cache, buf->start, 1);
7416 /* we have added this extent before */
7420 rec = container_of(cache, struct extent_record, cache);
7423 * Except file/reloc tree, we can not have
7426 if (ri->objectid < BTRFS_FIRST_FREE_OBJECTID)
7431 if (buf->start == ri->bytenr)
7434 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
7437 owner = btrfs_header_owner(buf);
7438 if (owner == ri->objectid)
7441 tback = find_tree_backref(rec, 0, owner);
7446 if (rec->flag_block_full_backref != FLAG_UNSET &&
7447 rec->flag_block_full_backref != 0)
7448 rec->bad_full_backref = 1;
7451 *flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7452 if (rec->flag_block_full_backref != FLAG_UNSET &&
7453 rec->flag_block_full_backref != 1)
7454 rec->bad_full_backref = 1;
7458 static void report_mismatch_key_root(u8 key_type, u64 rootid)
7460 fprintf(stderr, "Invalid key type(");
7461 print_key_type(stderr, 0, key_type);
7462 fprintf(stderr, ") found in root(");
7463 print_objectid(stderr, rootid, 0);
7464 fprintf(stderr, ")\n");
7468 * Check if the key is valid with its extent buffer.
7470 * This is a early check in case invalid key exists in a extent buffer
7471 * This is not comprehensive yet, but should prevent wrong key/item passed
7474 static int check_type_with_root(u64 rootid, u8 key_type)
7477 /* Only valid in chunk tree */
7478 case BTRFS_DEV_ITEM_KEY:
7479 case BTRFS_CHUNK_ITEM_KEY:
7480 if (rootid != BTRFS_CHUNK_TREE_OBJECTID)
7483 /* valid in csum and log tree */
7484 case BTRFS_CSUM_TREE_OBJECTID:
7485 if (!(rootid == BTRFS_TREE_LOG_OBJECTID ||
7489 case BTRFS_EXTENT_ITEM_KEY:
7490 case BTRFS_METADATA_ITEM_KEY:
7491 case BTRFS_BLOCK_GROUP_ITEM_KEY:
7492 if (rootid != BTRFS_EXTENT_TREE_OBJECTID)
7495 case BTRFS_ROOT_ITEM_KEY:
7496 if (rootid != BTRFS_ROOT_TREE_OBJECTID)
7499 case BTRFS_DEV_EXTENT_KEY:
7500 if (rootid != BTRFS_DEV_TREE_OBJECTID)
7506 report_mismatch_key_root(key_type, rootid);
7510 static int run_next_block(struct btrfs_root *root,
7511 struct block_info *bits,
7514 struct cache_tree *pending,
7515 struct cache_tree *seen,
7516 struct cache_tree *reada,
7517 struct cache_tree *nodes,
7518 struct cache_tree *extent_cache,
7519 struct cache_tree *chunk_cache,
7520 struct rb_root *dev_cache,
7521 struct block_group_tree *block_group_cache,
7522 struct device_extent_tree *dev_extent_cache,
7523 struct root_item_record *ri)
7525 struct extent_buffer *buf;
7526 struct extent_record *rec = NULL;
7537 struct btrfs_key key;
7538 struct cache_extent *cache;
7541 nritems = pick_next_pending(pending, reada, nodes, *last, bits,
7542 bits_nr, &reada_bits);
7547 for(i = 0; i < nritems; i++) {
7548 ret = add_cache_extent(reada, bits[i].start,
7553 /* fixme, get the parent transid */
7554 readahead_tree_block(root, bits[i].start,
7558 *last = bits[0].start;
7559 bytenr = bits[0].start;
7560 size = bits[0].size;
7562 cache = lookup_cache_extent(pending, bytenr, size);
7564 remove_cache_extent(pending, cache);
7567 cache = lookup_cache_extent(reada, bytenr, size);
7569 remove_cache_extent(reada, cache);
7572 cache = lookup_cache_extent(nodes, bytenr, size);
7574 remove_cache_extent(nodes, cache);
7577 cache = lookup_cache_extent(extent_cache, bytenr, size);
7579 rec = container_of(cache, struct extent_record, cache);
7580 gen = rec->parent_generation;
7583 /* fixme, get the real parent transid */
7584 buf = read_tree_block(root, bytenr, size, gen);
7585 if (!extent_buffer_uptodate(buf)) {
7586 record_bad_block_io(root->fs_info,
7587 extent_cache, bytenr, size);
7591 nritems = btrfs_header_nritems(buf);
7594 if (!init_extent_tree) {
7595 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
7596 btrfs_header_level(buf), 1, NULL,
7599 ret = calc_extent_flag(root, extent_cache, buf, ri, &flags);
7601 fprintf(stderr, "Couldn't calc extent flags\n");
7602 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7607 ret = calc_extent_flag(root, extent_cache, buf, ri, &flags);
7609 fprintf(stderr, "Couldn't calc extent flags\n");
7610 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7614 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7616 ri->objectid != BTRFS_TREE_RELOC_OBJECTID &&
7617 ri->objectid == btrfs_header_owner(buf)) {
7619 * Ok we got to this block from it's original owner and
7620 * we have FULL_BACKREF set. Relocation can leave
7621 * converted blocks over so this is altogether possible,
7622 * however it's not possible if the generation > the
7623 * last snapshot, so check for this case.
7625 if (!btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC) &&
7626 btrfs_header_generation(buf) > ri->last_snapshot) {
7627 flags &= ~BTRFS_BLOCK_FLAG_FULL_BACKREF;
7628 rec->bad_full_backref = 1;
7633 (ri->objectid == BTRFS_TREE_RELOC_OBJECTID ||
7634 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) {
7635 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7636 rec->bad_full_backref = 1;
7640 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7641 rec->flag_block_full_backref = 1;
7645 rec->flag_block_full_backref = 0;
7647 owner = btrfs_header_owner(buf);
7650 ret = check_block(root, extent_cache, buf, flags);
7654 if (btrfs_is_leaf(buf)) {
7655 btree_space_waste += btrfs_leaf_free_space(root, buf);
7656 for (i = 0; i < nritems; i++) {
7657 struct btrfs_file_extent_item *fi;
7658 btrfs_item_key_to_cpu(buf, &key, i);
7660 * Check key type against the leaf owner.
7661 * Could filter quite a lot of early error if
7664 if (check_type_with_root(btrfs_header_owner(buf),
7666 fprintf(stderr, "ignoring invalid key\n");
7669 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
7670 process_extent_item(root, extent_cache, buf,
7674 if (key.type == BTRFS_METADATA_ITEM_KEY) {
7675 process_extent_item(root, extent_cache, buf,
7679 if (key.type == BTRFS_EXTENT_CSUM_KEY) {
7681 btrfs_item_size_nr(buf, i);
7684 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
7685 process_chunk_item(chunk_cache, &key, buf, i);
7688 if (key.type == BTRFS_DEV_ITEM_KEY) {
7689 process_device_item(dev_cache, &key, buf, i);
7692 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
7693 process_block_group_item(block_group_cache,
7697 if (key.type == BTRFS_DEV_EXTENT_KEY) {
7698 process_device_extent_item(dev_extent_cache,
7703 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
7704 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
7705 process_extent_ref_v0(extent_cache, buf, i);
7712 if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
7713 ret = add_tree_backref(extent_cache,
7714 key.objectid, 0, key.offset, 0);
7716 error("add_tree_backref failed: %s",
7720 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
7721 ret = add_tree_backref(extent_cache,
7722 key.objectid, key.offset, 0, 0);
7724 error("add_tree_backref failed: %s",
7728 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
7729 struct btrfs_extent_data_ref *ref;
7730 ref = btrfs_item_ptr(buf, i,
7731 struct btrfs_extent_data_ref);
7732 add_data_backref(extent_cache,
7734 btrfs_extent_data_ref_root(buf, ref),
7735 btrfs_extent_data_ref_objectid(buf,
7737 btrfs_extent_data_ref_offset(buf, ref),
7738 btrfs_extent_data_ref_count(buf, ref),
7739 0, root->sectorsize);
7742 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
7743 struct btrfs_shared_data_ref *ref;
7744 ref = btrfs_item_ptr(buf, i,
7745 struct btrfs_shared_data_ref);
7746 add_data_backref(extent_cache,
7747 key.objectid, key.offset, 0, 0, 0,
7748 btrfs_shared_data_ref_count(buf, ref),
7749 0, root->sectorsize);
7752 if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
7753 struct bad_item *bad;
7755 if (key.objectid == BTRFS_ORPHAN_OBJECTID)
7759 bad = malloc(sizeof(struct bad_item));
7762 INIT_LIST_HEAD(&bad->list);
7763 memcpy(&bad->key, &key,
7764 sizeof(struct btrfs_key));
7765 bad->root_id = owner;
7766 list_add_tail(&bad->list, &delete_items);
7769 if (key.type != BTRFS_EXTENT_DATA_KEY)
7771 fi = btrfs_item_ptr(buf, i,
7772 struct btrfs_file_extent_item);
7773 if (btrfs_file_extent_type(buf, fi) ==
7774 BTRFS_FILE_EXTENT_INLINE)
7776 if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
7779 data_bytes_allocated +=
7780 btrfs_file_extent_disk_num_bytes(buf, fi);
7781 if (data_bytes_allocated < root->sectorsize) {
7784 data_bytes_referenced +=
7785 btrfs_file_extent_num_bytes(buf, fi);
7786 add_data_backref(extent_cache,
7787 btrfs_file_extent_disk_bytenr(buf, fi),
7788 parent, owner, key.objectid, key.offset -
7789 btrfs_file_extent_offset(buf, fi), 1, 1,
7790 btrfs_file_extent_disk_num_bytes(buf, fi));
7794 struct btrfs_key first_key;
7796 first_key.objectid = 0;
7799 btrfs_item_key_to_cpu(buf, &first_key, 0);
7800 level = btrfs_header_level(buf);
7801 for (i = 0; i < nritems; i++) {
7802 struct extent_record tmpl;
7804 ptr = btrfs_node_blockptr(buf, i);
7805 size = root->nodesize;
7806 btrfs_node_key_to_cpu(buf, &key, i);
7808 if ((level == ri->drop_level)
7809 && is_dropped_key(&key, &ri->drop_key)) {
7814 memset(&tmpl, 0, sizeof(tmpl));
7815 btrfs_cpu_key_to_disk(&tmpl.parent_key, &key);
7816 tmpl.parent_generation = btrfs_node_ptr_generation(buf, i);
7821 tmpl.max_size = size;
7822 ret = add_extent_rec(extent_cache, &tmpl);
7826 ret = add_tree_backref(extent_cache, ptr, parent,
7829 error("add_tree_backref failed: %s",
7835 add_pending(nodes, seen, ptr, size);
7837 add_pending(pending, seen, ptr, size);
7840 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
7841 nritems) * sizeof(struct btrfs_key_ptr);
7843 total_btree_bytes += buf->len;
7844 if (fs_root_objectid(btrfs_header_owner(buf)))
7845 total_fs_tree_bytes += buf->len;
7846 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
7847 total_extent_tree_bytes += buf->len;
7848 if (!found_old_backref &&
7849 btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
7850 btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
7851 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
7852 found_old_backref = 1;
7854 free_extent_buffer(buf);
7858 static int add_root_to_pending(struct extent_buffer *buf,
7859 struct cache_tree *extent_cache,
7860 struct cache_tree *pending,
7861 struct cache_tree *seen,
7862 struct cache_tree *nodes,
7865 struct extent_record tmpl;
7868 if (btrfs_header_level(buf) > 0)
7869 add_pending(nodes, seen, buf->start, buf->len);
7871 add_pending(pending, seen, buf->start, buf->len);
7873 memset(&tmpl, 0, sizeof(tmpl));
7874 tmpl.start = buf->start;
7879 tmpl.max_size = buf->len;
7880 add_extent_rec(extent_cache, &tmpl);
7882 if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
7883 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
7884 ret = add_tree_backref(extent_cache, buf->start, buf->start,
7887 ret = add_tree_backref(extent_cache, buf->start, 0, objectid,
7892 /* as we fix the tree, we might be deleting blocks that
7893 * we're tracking for repair. This hook makes sure we
7894 * remove any backrefs for blocks as we are fixing them.
7896 static int free_extent_hook(struct btrfs_trans_handle *trans,
7897 struct btrfs_root *root,
7898 u64 bytenr, u64 num_bytes, u64 parent,
7899 u64 root_objectid, u64 owner, u64 offset,
7902 struct extent_record *rec;
7903 struct cache_extent *cache;
7905 struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
7907 is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
7908 cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
7912 rec = container_of(cache, struct extent_record, cache);
7914 struct data_backref *back;
7915 back = find_data_backref(rec, parent, root_objectid, owner,
7916 offset, 1, bytenr, num_bytes);
7919 if (back->node.found_ref) {
7920 back->found_ref -= refs_to_drop;
7922 rec->refs -= refs_to_drop;
7924 if (back->node.found_extent_tree) {
7925 back->num_refs -= refs_to_drop;
7926 if (rec->extent_item_refs)
7927 rec->extent_item_refs -= refs_to_drop;
7929 if (back->found_ref == 0)
7930 back->node.found_ref = 0;
7931 if (back->num_refs == 0)
7932 back->node.found_extent_tree = 0;
7934 if (!back->node.found_extent_tree && back->node.found_ref) {
7935 list_del(&back->node.list);
7939 struct tree_backref *back;
7940 back = find_tree_backref(rec, parent, root_objectid);
7943 if (back->node.found_ref) {
7946 back->node.found_ref = 0;
7948 if (back->node.found_extent_tree) {
7949 if (rec->extent_item_refs)
7950 rec->extent_item_refs--;
7951 back->node.found_extent_tree = 0;
7953 if (!back->node.found_extent_tree && back->node.found_ref) {
7954 list_del(&back->node.list);
7958 maybe_free_extent_rec(extent_cache, rec);
7963 static int delete_extent_records(struct btrfs_trans_handle *trans,
7964 struct btrfs_root *root,
7965 struct btrfs_path *path,
7968 struct btrfs_key key;
7969 struct btrfs_key found_key;
7970 struct extent_buffer *leaf;
7975 key.objectid = bytenr;
7977 key.offset = (u64)-1;
7980 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
7987 if (path->slots[0] == 0)
7993 leaf = path->nodes[0];
7994 slot = path->slots[0];
7996 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7997 if (found_key.objectid != bytenr)
8000 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
8001 found_key.type != BTRFS_METADATA_ITEM_KEY &&
8002 found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
8003 found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
8004 found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
8005 found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
8006 found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
8007 btrfs_release_path(path);
8008 if (found_key.type == 0) {
8009 if (found_key.offset == 0)
8011 key.offset = found_key.offset - 1;
8012 key.type = found_key.type;
8014 key.type = found_key.type - 1;
8015 key.offset = (u64)-1;
8019 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
8020 found_key.objectid, found_key.type, found_key.offset);
8022 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
8025 btrfs_release_path(path);
8027 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
8028 found_key.type == BTRFS_METADATA_ITEM_KEY) {
8029 u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
8030 found_key.offset : root->nodesize;
8032 ret = btrfs_update_block_group(trans, root, bytenr,
8039 btrfs_release_path(path);
8044 * for a single backref, this will allocate a new extent
8045 * and add the backref to it.
8047 static int record_extent(struct btrfs_trans_handle *trans,
8048 struct btrfs_fs_info *info,
8049 struct btrfs_path *path,
8050 struct extent_record *rec,
8051 struct extent_backref *back,
8052 int allocated, u64 flags)
8055 struct btrfs_root *extent_root = info->extent_root;
8056 struct extent_buffer *leaf;
8057 struct btrfs_key ins_key;
8058 struct btrfs_extent_item *ei;
8059 struct data_backref *dback;
8060 struct btrfs_tree_block_info *bi;
8063 rec->max_size = max_t(u64, rec->max_size,
8064 info->extent_root->nodesize);
8067 u32 item_size = sizeof(*ei);
8070 item_size += sizeof(*bi);
8072 ins_key.objectid = rec->start;
8073 ins_key.offset = rec->max_size;
8074 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
8076 ret = btrfs_insert_empty_item(trans, extent_root, path,
8077 &ins_key, item_size);
8081 leaf = path->nodes[0];
8082 ei = btrfs_item_ptr(leaf, path->slots[0],
8083 struct btrfs_extent_item);
8085 btrfs_set_extent_refs(leaf, ei, 0);
8086 btrfs_set_extent_generation(leaf, ei, rec->generation);
8088 if (back->is_data) {
8089 btrfs_set_extent_flags(leaf, ei,
8090 BTRFS_EXTENT_FLAG_DATA);
8092 struct btrfs_disk_key copy_key;;
8094 bi = (struct btrfs_tree_block_info *)(ei + 1);
8095 memset_extent_buffer(leaf, 0, (unsigned long)bi,
8098 btrfs_set_disk_key_objectid(©_key,
8099 rec->info_objectid);
8100 btrfs_set_disk_key_type(©_key, 0);
8101 btrfs_set_disk_key_offset(©_key, 0);
8103 btrfs_set_tree_block_level(leaf, bi, rec->info_level);
8104 btrfs_set_tree_block_key(leaf, bi, ©_key);
8106 btrfs_set_extent_flags(leaf, ei,
8107 BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
8110 btrfs_mark_buffer_dirty(leaf);
8111 ret = btrfs_update_block_group(trans, extent_root, rec->start,
8112 rec->max_size, 1, 0);
8115 btrfs_release_path(path);
8118 if (back->is_data) {
8122 dback = to_data_backref(back);
8123 if (back->full_backref)
8124 parent = dback->parent;
8128 for (i = 0; i < dback->found_ref; i++) {
8129 /* if parent != 0, we're doing a full backref
8130 * passing BTRFS_FIRST_FREE_OBJECTID as the owner
8131 * just makes the backref allocator create a data
8134 ret = btrfs_inc_extent_ref(trans, info->extent_root,
8135 rec->start, rec->max_size,
8139 BTRFS_FIRST_FREE_OBJECTID :
8145 fprintf(stderr, "adding new data backref"
8146 " on %llu %s %llu owner %llu"
8147 " offset %llu found %d\n",
8148 (unsigned long long)rec->start,
8149 back->full_backref ?
8151 back->full_backref ?
8152 (unsigned long long)parent :
8153 (unsigned long long)dback->root,
8154 (unsigned long long)dback->owner,
8155 (unsigned long long)dback->offset,
8159 struct tree_backref *tback;
8161 tback = to_tree_backref(back);
8162 if (back->full_backref)
8163 parent = tback->parent;
8167 ret = btrfs_inc_extent_ref(trans, info->extent_root,
8168 rec->start, rec->max_size,
8169 parent, tback->root, 0, 0);
8170 fprintf(stderr, "adding new tree backref on "
8171 "start %llu len %llu parent %llu root %llu\n",
8172 rec->start, rec->max_size, parent, tback->root);
8175 btrfs_release_path(path);
8179 static struct extent_entry *find_entry(struct list_head *entries,
8180 u64 bytenr, u64 bytes)
8182 struct extent_entry *entry = NULL;
8184 list_for_each_entry(entry, entries, list) {
8185 if (entry->bytenr == bytenr && entry->bytes == bytes)
8192 static struct extent_entry *find_most_right_entry(struct list_head *entries)
8194 struct extent_entry *entry, *best = NULL, *prev = NULL;
8196 list_for_each_entry(entry, entries, list) {
8198 * If there are as many broken entries as entries then we know
8199 * not to trust this particular entry.
8201 if (entry->broken == entry->count)
8205 * Special case, when there are only two entries and 'best' is
8215 * If our current entry == best then we can't be sure our best
8216 * is really the best, so we need to keep searching.
8218 if (best && best->count == entry->count) {
8224 /* Prev == entry, not good enough, have to keep searching */
8225 if (!prev->broken && prev->count == entry->count)
8229 best = (prev->count > entry->count) ? prev : entry;
8230 else if (best->count < entry->count)
8238 static int repair_ref(struct btrfs_fs_info *info, struct btrfs_path *path,
8239 struct data_backref *dback, struct extent_entry *entry)
8241 struct btrfs_trans_handle *trans;
8242 struct btrfs_root *root;
8243 struct btrfs_file_extent_item *fi;
8244 struct extent_buffer *leaf;
8245 struct btrfs_key key;
8249 key.objectid = dback->root;
8250 key.type = BTRFS_ROOT_ITEM_KEY;
8251 key.offset = (u64)-1;
8252 root = btrfs_read_fs_root(info, &key);
8254 fprintf(stderr, "Couldn't find root for our ref\n");
8259 * The backref points to the original offset of the extent if it was
8260 * split, so we need to search down to the offset we have and then walk
8261 * forward until we find the backref we're looking for.
8263 key.objectid = dback->owner;
8264 key.type = BTRFS_EXTENT_DATA_KEY;
8265 key.offset = dback->offset;
8266 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8268 fprintf(stderr, "Error looking up ref %d\n", ret);
8273 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8274 ret = btrfs_next_leaf(root, path);
8276 fprintf(stderr, "Couldn't find our ref, next\n");
8280 leaf = path->nodes[0];
8281 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
8282 if (key.objectid != dback->owner ||
8283 key.type != BTRFS_EXTENT_DATA_KEY) {
8284 fprintf(stderr, "Couldn't find our ref, search\n");
8287 fi = btrfs_item_ptr(leaf, path->slots[0],
8288 struct btrfs_file_extent_item);
8289 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
8290 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
8292 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
8297 btrfs_release_path(path);
8299 trans = btrfs_start_transaction(root, 1);
8301 return PTR_ERR(trans);
8304 * Ok we have the key of the file extent we want to fix, now we can cow
8305 * down to the thing and fix it.
8307 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
8309 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
8310 key.objectid, key.type, key.offset, ret);
8314 fprintf(stderr, "Well that's odd, we just found this key "
8315 "[%Lu, %u, %Lu]\n", key.objectid, key.type,
8320 leaf = path->nodes[0];
8321 fi = btrfs_item_ptr(leaf, path->slots[0],
8322 struct btrfs_file_extent_item);
8324 if (btrfs_file_extent_compression(leaf, fi) &&
8325 dback->disk_bytenr != entry->bytenr) {
8326 fprintf(stderr, "Ref doesn't match the record start and is "
8327 "compressed, please take a btrfs-image of this file "
8328 "system and send it to a btrfs developer so they can "
8329 "complete this functionality for bytenr %Lu\n",
8330 dback->disk_bytenr);
8335 if (dback->node.broken && dback->disk_bytenr != entry->bytenr) {
8336 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
8337 } else if (dback->disk_bytenr > entry->bytenr) {
8338 u64 off_diff, offset;
8340 off_diff = dback->disk_bytenr - entry->bytenr;
8341 offset = btrfs_file_extent_offset(leaf, fi);
8342 if (dback->disk_bytenr + offset +
8343 btrfs_file_extent_num_bytes(leaf, fi) >
8344 entry->bytenr + entry->bytes) {
8345 fprintf(stderr, "Ref is past the entry end, please "
8346 "take a btrfs-image of this file system and "
8347 "send it to a btrfs developer, ref %Lu\n",
8348 dback->disk_bytenr);
8353 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
8354 btrfs_set_file_extent_offset(leaf, fi, offset);
8355 } else if (dback->disk_bytenr < entry->bytenr) {
8358 offset = btrfs_file_extent_offset(leaf, fi);
8359 if (dback->disk_bytenr + offset < entry->bytenr) {
8360 fprintf(stderr, "Ref is before the entry start, please"
8361 " take a btrfs-image of this file system and "
8362 "send it to a btrfs developer, ref %Lu\n",
8363 dback->disk_bytenr);
8368 offset += dback->disk_bytenr;
8369 offset -= entry->bytenr;
8370 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
8371 btrfs_set_file_extent_offset(leaf, fi, offset);
8374 btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
8377 * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
8378 * only do this if we aren't using compression, otherwise it's a
8381 if (!btrfs_file_extent_compression(leaf, fi))
8382 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
8384 printf("ram bytes may be wrong?\n");
8385 btrfs_mark_buffer_dirty(leaf);
8387 err = btrfs_commit_transaction(trans, root);
8388 btrfs_release_path(path);
8389 return ret ? ret : err;
8392 static int verify_backrefs(struct btrfs_fs_info *info, struct btrfs_path *path,
8393 struct extent_record *rec)
8395 struct extent_backref *back;
8396 struct data_backref *dback;
8397 struct extent_entry *entry, *best = NULL;
8400 int broken_entries = 0;
8405 * Metadata is easy and the backrefs should always agree on bytenr and
8406 * size, if not we've got bigger issues.
8411 list_for_each_entry(back, &rec->backrefs, list) {
8412 if (back->full_backref || !back->is_data)
8415 dback = to_data_backref(back);
8418 * We only pay attention to backrefs that we found a real
8421 if (dback->found_ref == 0)
8425 * For now we only catch when the bytes don't match, not the
8426 * bytenr. We can easily do this at the same time, but I want
8427 * to have a fs image to test on before we just add repair
8428 * functionality willy-nilly so we know we won't screw up the
8432 entry = find_entry(&entries, dback->disk_bytenr,
8435 entry = malloc(sizeof(struct extent_entry));
8440 memset(entry, 0, sizeof(*entry));
8441 entry->bytenr = dback->disk_bytenr;
8442 entry->bytes = dback->bytes;
8443 list_add_tail(&entry->list, &entries);
8448 * If we only have on entry we may think the entries agree when
8449 * in reality they don't so we have to do some extra checking.
8451 if (dback->disk_bytenr != rec->start ||
8452 dback->bytes != rec->nr || back->broken)
8463 /* Yay all the backrefs agree, carry on good sir */
8464 if (nr_entries <= 1 && !mismatch)
8467 fprintf(stderr, "attempting to repair backref discrepency for bytenr "
8468 "%Lu\n", rec->start);
8471 * First we want to see if the backrefs can agree amongst themselves who
8472 * is right, so figure out which one of the entries has the highest
8475 best = find_most_right_entry(&entries);
8478 * Ok so we may have an even split between what the backrefs think, so
8479 * this is where we use the extent ref to see what it thinks.
8482 entry = find_entry(&entries, rec->start, rec->nr);
8483 if (!entry && (!broken_entries || !rec->found_rec)) {
8484 fprintf(stderr, "Backrefs don't agree with each other "
8485 "and extent record doesn't agree with anybody,"
8486 " so we can't fix bytenr %Lu bytes %Lu\n",
8487 rec->start, rec->nr);
8490 } else if (!entry) {
8492 * Ok our backrefs were broken, we'll assume this is the
8493 * correct value and add an entry for this range.
8495 entry = malloc(sizeof(struct extent_entry));
8500 memset(entry, 0, sizeof(*entry));
8501 entry->bytenr = rec->start;
8502 entry->bytes = rec->nr;
8503 list_add_tail(&entry->list, &entries);
8507 best = find_most_right_entry(&entries);
8509 fprintf(stderr, "Backrefs and extent record evenly "
8510 "split on who is right, this is going to "
8511 "require user input to fix bytenr %Lu bytes "
8512 "%Lu\n", rec->start, rec->nr);
8519 * I don't think this can happen currently as we'll abort() if we catch
8520 * this case higher up, but in case somebody removes that we still can't
8521 * deal with it properly here yet, so just bail out of that's the case.
8523 if (best->bytenr != rec->start) {
8524 fprintf(stderr, "Extent start and backref starts don't match, "
8525 "please use btrfs-image on this file system and send "
8526 "it to a btrfs developer so they can make fsck fix "
8527 "this particular case. bytenr is %Lu, bytes is %Lu\n",
8528 rec->start, rec->nr);
8534 * Ok great we all agreed on an extent record, let's go find the real
8535 * references and fix up the ones that don't match.
8537 list_for_each_entry(back, &rec->backrefs, list) {
8538 if (back->full_backref || !back->is_data)
8541 dback = to_data_backref(back);
8544 * Still ignoring backrefs that don't have a real ref attached
8547 if (dback->found_ref == 0)
8550 if (dback->bytes == best->bytes &&
8551 dback->disk_bytenr == best->bytenr)
8554 ret = repair_ref(info, path, dback, best);
8560 * Ok we messed with the actual refs, which means we need to drop our
8561 * entire cache and go back and rescan. I know this is a huge pain and
8562 * adds a lot of extra work, but it's the only way to be safe. Once all
8563 * the backrefs agree we may not need to do anything to the extent
8568 while (!list_empty(&entries)) {
8569 entry = list_entry(entries.next, struct extent_entry, list);
8570 list_del_init(&entry->list);
8576 static int process_duplicates(struct btrfs_root *root,
8577 struct cache_tree *extent_cache,
8578 struct extent_record *rec)
8580 struct extent_record *good, *tmp;
8581 struct cache_extent *cache;
8585 * If we found a extent record for this extent then return, or if we
8586 * have more than one duplicate we are likely going to need to delete
8589 if (rec->found_rec || rec->num_duplicates > 1)
8592 /* Shouldn't happen but just in case */
8593 BUG_ON(!rec->num_duplicates);
8596 * So this happens if we end up with a backref that doesn't match the
8597 * actual extent entry. So either the backref is bad or the extent
8598 * entry is bad. Either way we want to have the extent_record actually
8599 * reflect what we found in the extent_tree, so we need to take the
8600 * duplicate out and use that as the extent_record since the only way we
8601 * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
8603 remove_cache_extent(extent_cache, &rec->cache);
8605 good = to_extent_record(rec->dups.next);
8606 list_del_init(&good->list);
8607 INIT_LIST_HEAD(&good->backrefs);
8608 INIT_LIST_HEAD(&good->dups);
8609 good->cache.start = good->start;
8610 good->cache.size = good->nr;
8611 good->content_checked = 0;
8612 good->owner_ref_checked = 0;
8613 good->num_duplicates = 0;
8614 good->refs = rec->refs;
8615 list_splice_init(&rec->backrefs, &good->backrefs);
8617 cache = lookup_cache_extent(extent_cache, good->start,
8621 tmp = container_of(cache, struct extent_record, cache);
8624 * If we find another overlapping extent and it's found_rec is
8625 * set then it's a duplicate and we need to try and delete
8628 if (tmp->found_rec || tmp->num_duplicates > 0) {
8629 if (list_empty(&good->list))
8630 list_add_tail(&good->list,
8631 &duplicate_extents);
8632 good->num_duplicates += tmp->num_duplicates + 1;
8633 list_splice_init(&tmp->dups, &good->dups);
8634 list_del_init(&tmp->list);
8635 list_add_tail(&tmp->list, &good->dups);
8636 remove_cache_extent(extent_cache, &tmp->cache);
8641 * Ok we have another non extent item backed extent rec, so lets
8642 * just add it to this extent and carry on like we did above.
8644 good->refs += tmp->refs;
8645 list_splice_init(&tmp->backrefs, &good->backrefs);
8646 remove_cache_extent(extent_cache, &tmp->cache);
8649 ret = insert_cache_extent(extent_cache, &good->cache);
8652 return good->num_duplicates ? 0 : 1;
8655 static int delete_duplicate_records(struct btrfs_root *root,
8656 struct extent_record *rec)
8658 struct btrfs_trans_handle *trans;
8659 LIST_HEAD(delete_list);
8660 struct btrfs_path path;
8661 struct extent_record *tmp, *good, *n;
8664 struct btrfs_key key;
8666 btrfs_init_path(&path);
8669 /* Find the record that covers all of the duplicates. */
8670 list_for_each_entry(tmp, &rec->dups, list) {
8671 if (good->start < tmp->start)
8673 if (good->nr > tmp->nr)
8676 if (tmp->start + tmp->nr < good->start + good->nr) {
8677 fprintf(stderr, "Ok we have overlapping extents that "
8678 "aren't completely covered by each other, this "
8679 "is going to require more careful thought. "
8680 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
8681 tmp->start, tmp->nr, good->start, good->nr);
8688 list_add_tail(&rec->list, &delete_list);
8690 list_for_each_entry_safe(tmp, n, &rec->dups, list) {
8693 list_move_tail(&tmp->list, &delete_list);
8696 root = root->fs_info->extent_root;
8697 trans = btrfs_start_transaction(root, 1);
8698 if (IS_ERR(trans)) {
8699 ret = PTR_ERR(trans);
8703 list_for_each_entry(tmp, &delete_list, list) {
8704 if (tmp->found_rec == 0)
8706 key.objectid = tmp->start;
8707 key.type = BTRFS_EXTENT_ITEM_KEY;
8708 key.offset = tmp->nr;
8710 /* Shouldn't happen but just in case */
8711 if (tmp->metadata) {
8712 fprintf(stderr, "Well this shouldn't happen, extent "
8713 "record overlaps but is metadata? "
8714 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
8718 ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
8724 ret = btrfs_del_item(trans, root, &path);
8727 btrfs_release_path(&path);
8730 err = btrfs_commit_transaction(trans, root);
8734 while (!list_empty(&delete_list)) {
8735 tmp = to_extent_record(delete_list.next);
8736 list_del_init(&tmp->list);
8742 while (!list_empty(&rec->dups)) {
8743 tmp = to_extent_record(rec->dups.next);
8744 list_del_init(&tmp->list);
8748 btrfs_release_path(&path);
8750 if (!ret && !nr_del)
8751 rec->num_duplicates = 0;
8753 return ret ? ret : nr_del;
8756 static int find_possible_backrefs(struct btrfs_fs_info *info,
8757 struct btrfs_path *path,
8758 struct cache_tree *extent_cache,
8759 struct extent_record *rec)
8761 struct btrfs_root *root;
8762 struct extent_backref *back;
8763 struct data_backref *dback;
8764 struct cache_extent *cache;
8765 struct btrfs_file_extent_item *fi;
8766 struct btrfs_key key;
8770 list_for_each_entry(back, &rec->backrefs, list) {
8771 /* Don't care about full backrefs (poor unloved backrefs) */
8772 if (back->full_backref || !back->is_data)
8775 dback = to_data_backref(back);
8777 /* We found this one, we don't need to do a lookup */
8778 if (dback->found_ref)
8781 key.objectid = dback->root;
8782 key.type = BTRFS_ROOT_ITEM_KEY;
8783 key.offset = (u64)-1;
8785 root = btrfs_read_fs_root(info, &key);
8787 /* No root, definitely a bad ref, skip */
8788 if (IS_ERR(root) && PTR_ERR(root) == -ENOENT)
8790 /* Other err, exit */
8792 return PTR_ERR(root);
8794 key.objectid = dback->owner;
8795 key.type = BTRFS_EXTENT_DATA_KEY;
8796 key.offset = dback->offset;
8797 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8799 btrfs_release_path(path);
8802 /* Didn't find it, we can carry on */
8807 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
8808 struct btrfs_file_extent_item);
8809 bytenr = btrfs_file_extent_disk_bytenr(path->nodes[0], fi);
8810 bytes = btrfs_file_extent_disk_num_bytes(path->nodes[0], fi);
8811 btrfs_release_path(path);
8812 cache = lookup_cache_extent(extent_cache, bytenr, 1);
8814 struct extent_record *tmp;
8815 tmp = container_of(cache, struct extent_record, cache);
8818 * If we found an extent record for the bytenr for this
8819 * particular backref then we can't add it to our
8820 * current extent record. We only want to add backrefs
8821 * that don't have a corresponding extent item in the
8822 * extent tree since they likely belong to this record
8823 * and we need to fix it if it doesn't match bytenrs.
8829 dback->found_ref += 1;
8830 dback->disk_bytenr = bytenr;
8831 dback->bytes = bytes;
8834 * Set this so the verify backref code knows not to trust the
8835 * values in this backref.
8844 * Record orphan data ref into corresponding root.
8846 * Return 0 if the extent item contains data ref and recorded.
8847 * Return 1 if the extent item contains no useful data ref
8848 * On that case, it may contains only shared_dataref or metadata backref
8849 * or the file extent exists(this should be handled by the extent bytenr
8851 * Return <0 if something goes wrong.
8853 static int record_orphan_data_extents(struct btrfs_fs_info *fs_info,
8854 struct extent_record *rec)
8856 struct btrfs_key key;
8857 struct btrfs_root *dest_root;
8858 struct extent_backref *back;
8859 struct data_backref *dback;
8860 struct orphan_data_extent *orphan;
8861 struct btrfs_path path;
8862 int recorded_data_ref = 0;
8867 btrfs_init_path(&path);
8868 list_for_each_entry(back, &rec->backrefs, list) {
8869 if (back->full_backref || !back->is_data ||
8870 !back->found_extent_tree)
8872 dback = to_data_backref(back);
8873 if (dback->found_ref)
8875 key.objectid = dback->root;
8876 key.type = BTRFS_ROOT_ITEM_KEY;
8877 key.offset = (u64)-1;
8879 dest_root = btrfs_read_fs_root(fs_info, &key);
8881 /* For non-exist root we just skip it */
8882 if (IS_ERR(dest_root) || !dest_root)
8885 key.objectid = dback->owner;
8886 key.type = BTRFS_EXTENT_DATA_KEY;
8887 key.offset = dback->offset;
8889 ret = btrfs_search_slot(NULL, dest_root, &key, &path, 0, 0);
8890 btrfs_release_path(&path);
8892 * For ret < 0, it's OK since the fs-tree may be corrupted,
8893 * we need to record it for inode/file extent rebuild.
8894 * For ret > 0, we record it only for file extent rebuild.
8895 * For ret == 0, the file extent exists but only bytenr
8896 * mismatch, let the original bytenr fix routine to handle,
8902 orphan = malloc(sizeof(*orphan));
8907 INIT_LIST_HEAD(&orphan->list);
8908 orphan->root = dback->root;
8909 orphan->objectid = dback->owner;
8910 orphan->offset = dback->offset;
8911 orphan->disk_bytenr = rec->cache.start;
8912 orphan->disk_len = rec->cache.size;
8913 list_add(&dest_root->orphan_data_extents, &orphan->list);
8914 recorded_data_ref = 1;
8917 btrfs_release_path(&path);
8919 return !recorded_data_ref;
8925 * when an incorrect extent item is found, this will delete
8926 * all of the existing entries for it and recreate them
8927 * based on what the tree scan found.
8929 static int fixup_extent_refs(struct btrfs_fs_info *info,
8930 struct cache_tree *extent_cache,
8931 struct extent_record *rec)
8933 struct btrfs_trans_handle *trans = NULL;
8935 struct btrfs_path path;
8936 struct list_head *cur = rec->backrefs.next;
8937 struct cache_extent *cache;
8938 struct extent_backref *back;
8942 if (rec->flag_block_full_backref)
8943 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8945 btrfs_init_path(&path);
8946 if (rec->refs != rec->extent_item_refs && !rec->metadata) {
8948 * Sometimes the backrefs themselves are so broken they don't
8949 * get attached to any meaningful rec, so first go back and
8950 * check any of our backrefs that we couldn't find and throw
8951 * them into the list if we find the backref so that
8952 * verify_backrefs can figure out what to do.
8954 ret = find_possible_backrefs(info, &path, extent_cache, rec);
8959 /* step one, make sure all of the backrefs agree */
8960 ret = verify_backrefs(info, &path, rec);
8964 trans = btrfs_start_transaction(info->extent_root, 1);
8965 if (IS_ERR(trans)) {
8966 ret = PTR_ERR(trans);
8970 /* step two, delete all the existing records */
8971 ret = delete_extent_records(trans, info->extent_root, &path,
8977 /* was this block corrupt? If so, don't add references to it */
8978 cache = lookup_cache_extent(info->corrupt_blocks,
8979 rec->start, rec->max_size);
8985 /* step three, recreate all the refs we did find */
8986 while(cur != &rec->backrefs) {
8987 back = to_extent_backref(cur);
8991 * if we didn't find any references, don't create a
8994 if (!back->found_ref)
8997 rec->bad_full_backref = 0;
8998 ret = record_extent(trans, info, &path, rec, back, allocated, flags);
9006 int err = btrfs_commit_transaction(trans, info->extent_root);
9012 fprintf(stderr, "Repaired extent references for %llu\n",
9013 (unsigned long long)rec->start);
9015 btrfs_release_path(&path);
9019 static int fixup_extent_flags(struct btrfs_fs_info *fs_info,
9020 struct extent_record *rec)
9022 struct btrfs_trans_handle *trans;
9023 struct btrfs_root *root = fs_info->extent_root;
9024 struct btrfs_path path;
9025 struct btrfs_extent_item *ei;
9026 struct btrfs_key key;
9030 key.objectid = rec->start;
9031 if (rec->metadata) {
9032 key.type = BTRFS_METADATA_ITEM_KEY;
9033 key.offset = rec->info_level;
9035 key.type = BTRFS_EXTENT_ITEM_KEY;
9036 key.offset = rec->max_size;
9039 trans = btrfs_start_transaction(root, 0);
9041 return PTR_ERR(trans);
9043 btrfs_init_path(&path);
9044 ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
9046 btrfs_release_path(&path);
9047 btrfs_commit_transaction(trans, root);
9050 fprintf(stderr, "Didn't find extent for %llu\n",
9051 (unsigned long long)rec->start);
9052 btrfs_release_path(&path);
9053 btrfs_commit_transaction(trans, root);
9057 ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
9058 struct btrfs_extent_item);
9059 flags = btrfs_extent_flags(path.nodes[0], ei);
9060 if (rec->flag_block_full_backref) {
9061 fprintf(stderr, "setting full backref on %llu\n",
9062 (unsigned long long)key.objectid);
9063 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
9065 fprintf(stderr, "clearing full backref on %llu\n",
9066 (unsigned long long)key.objectid);
9067 flags &= ~BTRFS_BLOCK_FLAG_FULL_BACKREF;
9069 btrfs_set_extent_flags(path.nodes[0], ei, flags);
9070 btrfs_mark_buffer_dirty(path.nodes[0]);
9071 btrfs_release_path(&path);
9072 ret = btrfs_commit_transaction(trans, root);
9074 fprintf(stderr, "Repaired extent flags for %llu\n",
9075 (unsigned long long)rec->start);
9080 /* right now we only prune from the extent allocation tree */
9081 static int prune_one_block(struct btrfs_trans_handle *trans,
9082 struct btrfs_fs_info *info,
9083 struct btrfs_corrupt_block *corrupt)
9086 struct btrfs_path path;
9087 struct extent_buffer *eb;
9091 int level = corrupt->level + 1;
9093 btrfs_init_path(&path);
9095 /* we want to stop at the parent to our busted block */
9096 path.lowest_level = level;
9098 ret = btrfs_search_slot(trans, info->extent_root,
9099 &corrupt->key, &path, -1, 1);
9104 eb = path.nodes[level];
9111 * hopefully the search gave us the block we want to prune,
9112 * lets try that first
9114 slot = path.slots[level];
9115 found = btrfs_node_blockptr(eb, slot);
9116 if (found == corrupt->cache.start)
9119 nritems = btrfs_header_nritems(eb);
9121 /* the search failed, lets scan this node and hope we find it */
9122 for (slot = 0; slot < nritems; slot++) {
9123 found = btrfs_node_blockptr(eb, slot);
9124 if (found == corrupt->cache.start)
9128 * we couldn't find the bad block. TODO, search all the nodes for pointers
9131 if (eb == info->extent_root->node) {
9136 btrfs_release_path(&path);
9141 printk("deleting pointer to block %Lu\n", corrupt->cache.start);
9142 ret = btrfs_del_ptr(info->extent_root, &path, level, slot);
9145 btrfs_release_path(&path);
9149 static int prune_corrupt_blocks(struct btrfs_fs_info *info)
9151 struct btrfs_trans_handle *trans = NULL;
9152 struct cache_extent *cache;
9153 struct btrfs_corrupt_block *corrupt;
9156 cache = search_cache_extent(info->corrupt_blocks, 0);
9160 trans = btrfs_start_transaction(info->extent_root, 1);
9162 return PTR_ERR(trans);
9164 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
9165 prune_one_block(trans, info, corrupt);
9166 remove_cache_extent(info->corrupt_blocks, cache);
9169 return btrfs_commit_transaction(trans, info->extent_root);
9173 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
9175 struct btrfs_block_group_cache *cache;
9180 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
9181 &start, &end, EXTENT_DIRTY);
9184 clear_extent_dirty(&fs_info->free_space_cache, start, end);
9189 cache = btrfs_lookup_first_block_group(fs_info, start);
9194 start = cache->key.objectid + cache->key.offset;
9198 static int check_extent_refs(struct btrfs_root *root,
9199 struct cache_tree *extent_cache)
9201 struct extent_record *rec;
9202 struct cache_extent *cache;
9208 * if we're doing a repair, we have to make sure
9209 * we don't allocate from the problem extents.
9210 * In the worst case, this will be all the
9213 cache = search_cache_extent(extent_cache, 0);
9215 rec = container_of(cache, struct extent_record, cache);
9216 set_extent_dirty(root->fs_info->excluded_extents,
9218 rec->start + rec->max_size - 1);
9219 cache = next_cache_extent(cache);
9222 /* pin down all the corrupted blocks too */
9223 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
9225 set_extent_dirty(root->fs_info->excluded_extents,
9227 cache->start + cache->size - 1);
9228 cache = next_cache_extent(cache);
9230 prune_corrupt_blocks(root->fs_info);
9231 reset_cached_block_groups(root->fs_info);
9234 reset_cached_block_groups(root->fs_info);
9237 * We need to delete any duplicate entries we find first otherwise we
9238 * could mess up the extent tree when we have backrefs that actually
9239 * belong to a different extent item and not the weird duplicate one.
9241 while (repair && !list_empty(&duplicate_extents)) {
9242 rec = to_extent_record(duplicate_extents.next);
9243 list_del_init(&rec->list);
9245 /* Sometimes we can find a backref before we find an actual
9246 * extent, so we need to process it a little bit to see if there
9247 * truly are multiple EXTENT_ITEM_KEY's for the same range, or
9248 * if this is a backref screwup. If we need to delete stuff
9249 * process_duplicates() will return 0, otherwise it will return
9252 if (process_duplicates(root, extent_cache, rec))
9254 ret = delete_duplicate_records(root, rec);
9258 * delete_duplicate_records will return the number of entries
9259 * deleted, so if it's greater than 0 then we know we actually
9260 * did something and we need to remove.
9273 cache = search_cache_extent(extent_cache, 0);
9276 rec = container_of(cache, struct extent_record, cache);
9277 if (rec->num_duplicates) {
9278 fprintf(stderr, "extent item %llu has multiple extent "
9279 "items\n", (unsigned long long)rec->start);
9283 if (rec->refs != rec->extent_item_refs) {
9284 fprintf(stderr, "ref mismatch on [%llu %llu] ",
9285 (unsigned long long)rec->start,
9286 (unsigned long long)rec->nr);
9287 fprintf(stderr, "extent item %llu, found %llu\n",
9288 (unsigned long long)rec->extent_item_refs,
9289 (unsigned long long)rec->refs);
9290 ret = record_orphan_data_extents(root->fs_info, rec);
9296 if (all_backpointers_checked(rec, 1)) {
9297 fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
9298 (unsigned long long)rec->start,
9299 (unsigned long long)rec->nr);
9303 if (!rec->owner_ref_checked) {
9304 fprintf(stderr, "owner ref check failed [%llu %llu]\n",
9305 (unsigned long long)rec->start,
9306 (unsigned long long)rec->nr);
9311 if (repair && fix) {
9312 ret = fixup_extent_refs(root->fs_info, extent_cache, rec);
9318 if (rec->bad_full_backref) {
9319 fprintf(stderr, "bad full backref, on [%llu]\n",
9320 (unsigned long long)rec->start);
9322 ret = fixup_extent_flags(root->fs_info, rec);
9330 * Although it's not a extent ref's problem, we reuse this
9331 * routine for error reporting.
9332 * No repair function yet.
9334 if (rec->crossing_stripes) {
9336 "bad metadata [%llu, %llu) crossing stripe boundary\n",
9337 rec->start, rec->start + rec->max_size);
9341 if (rec->wrong_chunk_type) {
9343 "bad extent [%llu, %llu), type mismatch with chunk\n",
9344 rec->start, rec->start + rec->max_size);
9348 remove_cache_extent(extent_cache, cache);
9349 free_all_extent_backrefs(rec);
9350 if (!init_extent_tree && repair && (!cur_err || fix))
9351 clear_extent_dirty(root->fs_info->excluded_extents,
9353 rec->start + rec->max_size - 1);
9358 if (ret && ret != -EAGAIN) {
9359 fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
9362 struct btrfs_trans_handle *trans;
9364 root = root->fs_info->extent_root;
9365 trans = btrfs_start_transaction(root, 1);
9366 if (IS_ERR(trans)) {
9367 ret = PTR_ERR(trans);
9371 btrfs_fix_block_accounting(trans, root);
9372 ret = btrfs_commit_transaction(trans, root);
9381 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
9385 if (type & BTRFS_BLOCK_GROUP_RAID0) {
9386 stripe_size = length;
9387 stripe_size /= num_stripes;
9388 } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
9389 stripe_size = length * 2;
9390 stripe_size /= num_stripes;
9391 } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
9392 stripe_size = length;
9393 stripe_size /= (num_stripes - 1);
9394 } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
9395 stripe_size = length;
9396 stripe_size /= (num_stripes - 2);
9398 stripe_size = length;
9404 * Check the chunk with its block group/dev list ref:
9405 * Return 0 if all refs seems valid.
9406 * Return 1 if part of refs seems valid, need later check for rebuild ref
9407 * like missing block group and needs to search extent tree to rebuild them.
9408 * Return -1 if essential refs are missing and unable to rebuild.
9410 static int check_chunk_refs(struct chunk_record *chunk_rec,
9411 struct block_group_tree *block_group_cache,
9412 struct device_extent_tree *dev_extent_cache,
9415 struct cache_extent *block_group_item;
9416 struct block_group_record *block_group_rec;
9417 struct cache_extent *dev_extent_item;
9418 struct device_extent_record *dev_extent_rec;
9422 int metadump_v2 = 0;
9426 block_group_item = lookup_cache_extent(&block_group_cache->tree,
9429 if (block_group_item) {
9430 block_group_rec = container_of(block_group_item,
9431 struct block_group_record,
9433 if (chunk_rec->length != block_group_rec->offset ||
9434 chunk_rec->offset != block_group_rec->objectid ||
9436 chunk_rec->type_flags != block_group_rec->flags)) {
9439 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
9440 chunk_rec->objectid,
9445 chunk_rec->type_flags,
9446 block_group_rec->objectid,
9447 block_group_rec->type,
9448 block_group_rec->offset,
9449 block_group_rec->offset,
9450 block_group_rec->objectid,
9451 block_group_rec->flags);
9454 list_del_init(&block_group_rec->list);
9455 chunk_rec->bg_rec = block_group_rec;
9460 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
9461 chunk_rec->objectid,
9466 chunk_rec->type_flags);
9473 length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
9474 chunk_rec->num_stripes);
9475 for (i = 0; i < chunk_rec->num_stripes; ++i) {
9476 devid = chunk_rec->stripes[i].devid;
9477 offset = chunk_rec->stripes[i].offset;
9478 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
9479 devid, offset, length);
9480 if (dev_extent_item) {
9481 dev_extent_rec = container_of(dev_extent_item,
9482 struct device_extent_record,
9484 if (dev_extent_rec->objectid != devid ||
9485 dev_extent_rec->offset != offset ||
9486 dev_extent_rec->chunk_offset != chunk_rec->offset ||
9487 dev_extent_rec->length != length) {
9490 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
9491 chunk_rec->objectid,
9494 chunk_rec->stripes[i].devid,
9495 chunk_rec->stripes[i].offset,
9496 dev_extent_rec->objectid,
9497 dev_extent_rec->offset,
9498 dev_extent_rec->length);
9501 list_move(&dev_extent_rec->chunk_list,
9502 &chunk_rec->dextents);
9507 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
9508 chunk_rec->objectid,
9511 chunk_rec->stripes[i].devid,
9512 chunk_rec->stripes[i].offset);
9519 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
9520 int check_chunks(struct cache_tree *chunk_cache,
9521 struct block_group_tree *block_group_cache,
9522 struct device_extent_tree *dev_extent_cache,
9523 struct list_head *good, struct list_head *bad,
9524 struct list_head *rebuild, int silent)
9526 struct cache_extent *chunk_item;
9527 struct chunk_record *chunk_rec;
9528 struct block_group_record *bg_rec;
9529 struct device_extent_record *dext_rec;
9533 chunk_item = first_cache_extent(chunk_cache);
9534 while (chunk_item) {
9535 chunk_rec = container_of(chunk_item, struct chunk_record,
9537 err = check_chunk_refs(chunk_rec, block_group_cache,
9538 dev_extent_cache, silent);
9541 if (err == 0 && good)
9542 list_add_tail(&chunk_rec->list, good);
9543 if (err > 0 && rebuild)
9544 list_add_tail(&chunk_rec->list, rebuild);
9546 list_add_tail(&chunk_rec->list, bad);
9547 chunk_item = next_cache_extent(chunk_item);
9550 list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
9553 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
9561 list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
9565 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
9576 static int check_device_used(struct device_record *dev_rec,
9577 struct device_extent_tree *dext_cache)
9579 struct cache_extent *cache;
9580 struct device_extent_record *dev_extent_rec;
9583 cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
9585 dev_extent_rec = container_of(cache,
9586 struct device_extent_record,
9588 if (dev_extent_rec->objectid != dev_rec->devid)
9591 list_del_init(&dev_extent_rec->device_list);
9592 total_byte += dev_extent_rec->length;
9593 cache = next_cache_extent(cache);
9596 if (total_byte != dev_rec->byte_used) {
9598 "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
9599 total_byte, dev_rec->byte_used, dev_rec->objectid,
9600 dev_rec->type, dev_rec->offset);
9607 /* check btrfs_dev_item -> btrfs_dev_extent */
9608 static int check_devices(struct rb_root *dev_cache,
9609 struct device_extent_tree *dev_extent_cache)
9611 struct rb_node *dev_node;
9612 struct device_record *dev_rec;
9613 struct device_extent_record *dext_rec;
9617 dev_node = rb_first(dev_cache);
9619 dev_rec = container_of(dev_node, struct device_record, node);
9620 err = check_device_used(dev_rec, dev_extent_cache);
9624 dev_node = rb_next(dev_node);
9626 list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
9629 "Device extent[%llu, %llu, %llu] didn't find its device.\n",
9630 dext_rec->objectid, dext_rec->offset, dext_rec->length);
9637 static int add_root_item_to_list(struct list_head *head,
9638 u64 objectid, u64 bytenr, u64 last_snapshot,
9639 u8 level, u8 drop_level,
9640 int level_size, struct btrfs_key *drop_key)
9643 struct root_item_record *ri_rec;
9644 ri_rec = malloc(sizeof(*ri_rec));
9647 ri_rec->bytenr = bytenr;
9648 ri_rec->objectid = objectid;
9649 ri_rec->level = level;
9650 ri_rec->level_size = level_size;
9651 ri_rec->drop_level = drop_level;
9652 ri_rec->last_snapshot = last_snapshot;
9654 memcpy(&ri_rec->drop_key, drop_key, sizeof(*drop_key));
9655 list_add_tail(&ri_rec->list, head);
9660 static void free_root_item_list(struct list_head *list)
9662 struct root_item_record *ri_rec;
9664 while (!list_empty(list)) {
9665 ri_rec = list_first_entry(list, struct root_item_record,
9667 list_del_init(&ri_rec->list);
9672 static int deal_root_from_list(struct list_head *list,
9673 struct btrfs_root *root,
9674 struct block_info *bits,
9676 struct cache_tree *pending,
9677 struct cache_tree *seen,
9678 struct cache_tree *reada,
9679 struct cache_tree *nodes,
9680 struct cache_tree *extent_cache,
9681 struct cache_tree *chunk_cache,
9682 struct rb_root *dev_cache,
9683 struct block_group_tree *block_group_cache,
9684 struct device_extent_tree *dev_extent_cache)
9689 while (!list_empty(list)) {
9690 struct root_item_record *rec;
9691 struct extent_buffer *buf;
9692 rec = list_entry(list->next,
9693 struct root_item_record, list);
9695 buf = read_tree_block(root->fs_info->tree_root,
9696 rec->bytenr, rec->level_size, 0);
9697 if (!extent_buffer_uptodate(buf)) {
9698 free_extent_buffer(buf);
9702 ret = add_root_to_pending(buf, extent_cache, pending,
9703 seen, nodes, rec->objectid);
9707 * To rebuild extent tree, we need deal with snapshot
9708 * one by one, otherwise we deal with node firstly which
9709 * can maximize readahead.
9712 ret = run_next_block(root, bits, bits_nr, &last,
9713 pending, seen, reada, nodes,
9714 extent_cache, chunk_cache,
9715 dev_cache, block_group_cache,
9716 dev_extent_cache, rec);
9720 free_extent_buffer(buf);
9721 list_del(&rec->list);
9727 ret = run_next_block(root, bits, bits_nr, &last, pending, seen,
9728 reada, nodes, extent_cache, chunk_cache,
9729 dev_cache, block_group_cache,
9730 dev_extent_cache, NULL);
9740 static int check_chunks_and_extents(struct btrfs_root *root)
9742 struct rb_root dev_cache;
9743 struct cache_tree chunk_cache;
9744 struct block_group_tree block_group_cache;
9745 struct device_extent_tree dev_extent_cache;
9746 struct cache_tree extent_cache;
9747 struct cache_tree seen;
9748 struct cache_tree pending;
9749 struct cache_tree reada;
9750 struct cache_tree nodes;
9751 struct extent_io_tree excluded_extents;
9752 struct cache_tree corrupt_blocks;
9753 struct btrfs_path path;
9754 struct btrfs_key key;
9755 struct btrfs_key found_key;
9757 struct block_info *bits;
9759 struct extent_buffer *leaf;
9761 struct btrfs_root_item ri;
9762 struct list_head dropping_trees;
9763 struct list_head normal_trees;
9764 struct btrfs_root *root1;
9769 dev_cache = RB_ROOT;
9770 cache_tree_init(&chunk_cache);
9771 block_group_tree_init(&block_group_cache);
9772 device_extent_tree_init(&dev_extent_cache);
9774 cache_tree_init(&extent_cache);
9775 cache_tree_init(&seen);
9776 cache_tree_init(&pending);
9777 cache_tree_init(&nodes);
9778 cache_tree_init(&reada);
9779 cache_tree_init(&corrupt_blocks);
9780 extent_io_tree_init(&excluded_extents);
9781 INIT_LIST_HEAD(&dropping_trees);
9782 INIT_LIST_HEAD(&normal_trees);
9785 root->fs_info->excluded_extents = &excluded_extents;
9786 root->fs_info->fsck_extent_cache = &extent_cache;
9787 root->fs_info->free_extent_hook = free_extent_hook;
9788 root->fs_info->corrupt_blocks = &corrupt_blocks;
9792 bits = malloc(bits_nr * sizeof(struct block_info));
9798 if (ctx.progress_enabled) {
9799 ctx.tp = TASK_EXTENTS;
9800 task_start(ctx.info);
9804 root1 = root->fs_info->tree_root;
9805 level = btrfs_header_level(root1->node);
9806 ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
9807 root1->node->start, 0, level, 0,
9808 root1->nodesize, NULL);
9811 root1 = root->fs_info->chunk_root;
9812 level = btrfs_header_level(root1->node);
9813 ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
9814 root1->node->start, 0, level, 0,
9815 root1->nodesize, NULL);
9818 btrfs_init_path(&path);
9821 key.type = BTRFS_ROOT_ITEM_KEY;
9822 ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
9827 leaf = path.nodes[0];
9828 slot = path.slots[0];
9829 if (slot >= btrfs_header_nritems(path.nodes[0])) {
9830 ret = btrfs_next_leaf(root, &path);
9833 leaf = path.nodes[0];
9834 slot = path.slots[0];
9836 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
9837 if (found_key.type == BTRFS_ROOT_ITEM_KEY) {
9838 unsigned long offset;
9841 offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
9842 read_extent_buffer(leaf, &ri, offset, sizeof(ri));
9843 last_snapshot = btrfs_root_last_snapshot(&ri);
9844 if (btrfs_disk_key_objectid(&ri.drop_progress) == 0) {
9845 level = btrfs_root_level(&ri);
9846 level_size = root->nodesize;
9847 ret = add_root_item_to_list(&normal_trees,
9849 btrfs_root_bytenr(&ri),
9850 last_snapshot, level,
9851 0, level_size, NULL);
9855 level = btrfs_root_level(&ri);
9856 level_size = root->nodesize;
9857 objectid = found_key.objectid;
9858 btrfs_disk_key_to_cpu(&found_key,
9860 ret = add_root_item_to_list(&dropping_trees,
9862 btrfs_root_bytenr(&ri),
9863 last_snapshot, level,
9865 level_size, &found_key);
9872 btrfs_release_path(&path);
9875 * check_block can return -EAGAIN if it fixes something, please keep
9876 * this in mind when dealing with return values from these functions, if
9877 * we get -EAGAIN we want to fall through and restart the loop.
9879 ret = deal_root_from_list(&normal_trees, root, bits, bits_nr, &pending,
9880 &seen, &reada, &nodes, &extent_cache,
9881 &chunk_cache, &dev_cache, &block_group_cache,
9888 ret = deal_root_from_list(&dropping_trees, root, bits, bits_nr,
9889 &pending, &seen, &reada, &nodes,
9890 &extent_cache, &chunk_cache, &dev_cache,
9891 &block_group_cache, &dev_extent_cache);
9898 ret = check_chunks(&chunk_cache, &block_group_cache,
9899 &dev_extent_cache, NULL, NULL, NULL, 0);
9906 ret = check_extent_refs(root, &extent_cache);
9913 ret = check_devices(&dev_cache, &dev_extent_cache);
9918 task_stop(ctx.info);
9920 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
9921 extent_io_tree_cleanup(&excluded_extents);
9922 root->fs_info->fsck_extent_cache = NULL;
9923 root->fs_info->free_extent_hook = NULL;
9924 root->fs_info->corrupt_blocks = NULL;
9925 root->fs_info->excluded_extents = NULL;
9928 free_chunk_cache_tree(&chunk_cache);
9929 free_device_cache_tree(&dev_cache);
9930 free_block_group_tree(&block_group_cache);
9931 free_device_extent_tree(&dev_extent_cache);
9932 free_extent_cache_tree(&seen);
9933 free_extent_cache_tree(&pending);
9934 free_extent_cache_tree(&reada);
9935 free_extent_cache_tree(&nodes);
9938 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
9939 free_extent_cache_tree(&seen);
9940 free_extent_cache_tree(&pending);
9941 free_extent_cache_tree(&reada);
9942 free_extent_cache_tree(&nodes);
9943 free_chunk_cache_tree(&chunk_cache);
9944 free_block_group_tree(&block_group_cache);
9945 free_device_cache_tree(&dev_cache);
9946 free_device_extent_tree(&dev_extent_cache);
9947 free_extent_record_cache(&extent_cache);
9948 free_root_item_list(&normal_trees);
9949 free_root_item_list(&dropping_trees);
9950 extent_io_tree_cleanup(&excluded_extents);
9955 * Check backrefs of a tree block given by @bytenr or @eb.
9957 * @root: the root containing the @bytenr or @eb
9958 * @eb: tree block extent buffer, can be NULL
9959 * @bytenr: bytenr of the tree block to search
9960 * @level: tree level of the tree block
9961 * @owner: owner of the tree block
9963 * Return >0 for any error found and output error message
9964 * Return 0 for no error found
9966 static int check_tree_block_ref(struct btrfs_root *root,
9967 struct extent_buffer *eb, u64 bytenr,
9968 int level, u64 owner)
9970 struct btrfs_key key;
9971 struct btrfs_root *extent_root = root->fs_info->extent_root;
9972 struct btrfs_path path;
9973 struct btrfs_extent_item *ei;
9974 struct btrfs_extent_inline_ref *iref;
9975 struct extent_buffer *leaf;
9981 u32 nodesize = root->nodesize;
9984 int tree_reloc_root = 0;
9989 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
9990 btrfs_header_bytenr(root->node) == bytenr)
9991 tree_reloc_root = 1;
9993 btrfs_init_path(&path);
9994 key.objectid = bytenr;
9995 if (btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
9996 key.type = BTRFS_METADATA_ITEM_KEY;
9998 key.type = BTRFS_EXTENT_ITEM_KEY;
9999 key.offset = (u64)-1;
10001 /* Search for the backref in extent tree */
10002 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
10004 err |= BACKREF_MISSING;
10007 ret = btrfs_previous_extent_item(extent_root, &path, bytenr);
10009 err |= BACKREF_MISSING;
10013 leaf = path.nodes[0];
10014 slot = path.slots[0];
10015 btrfs_item_key_to_cpu(leaf, &key, slot);
10017 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
10019 if (key.type == BTRFS_METADATA_ITEM_KEY) {
10020 skinny_level = (int)key.offset;
10021 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
10023 struct btrfs_tree_block_info *info;
10025 info = (struct btrfs_tree_block_info *)(ei + 1);
10026 skinny_level = btrfs_tree_block_level(leaf, info);
10027 iref = (struct btrfs_extent_inline_ref *)(info + 1);
10034 if (!(btrfs_extent_flags(leaf, ei) &
10035 BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
10037 "extent[%llu %u] backref type mismatch, missing bit: %llx",
10038 key.objectid, nodesize,
10039 BTRFS_EXTENT_FLAG_TREE_BLOCK);
10040 err = BACKREF_MISMATCH;
10042 header_gen = btrfs_header_generation(eb);
10043 extent_gen = btrfs_extent_generation(leaf, ei);
10044 if (header_gen != extent_gen) {
10046 "extent[%llu %u] backref generation mismatch, wanted: %llu, have: %llu",
10047 key.objectid, nodesize, header_gen,
10049 err = BACKREF_MISMATCH;
10051 if (level != skinny_level) {
10053 "extent[%llu %u] level mismatch, wanted: %u, have: %u",
10054 key.objectid, nodesize, level, skinny_level);
10055 err = BACKREF_MISMATCH;
10057 if (!is_fstree(owner) && btrfs_extent_refs(leaf, ei) != 1) {
10059 "extent[%llu %u] is referred by other roots than %llu",
10060 key.objectid, nodesize, root->objectid);
10061 err = BACKREF_MISMATCH;
10066 * Iterate the extent/metadata item to find the exact backref
10068 item_size = btrfs_item_size_nr(leaf, slot);
10069 ptr = (unsigned long)iref;
10070 end = (unsigned long)ei + item_size;
10071 while (ptr < end) {
10072 iref = (struct btrfs_extent_inline_ref *)ptr;
10073 type = btrfs_extent_inline_ref_type(leaf, iref);
10074 offset = btrfs_extent_inline_ref_offset(leaf, iref);
10076 if (type == BTRFS_TREE_BLOCK_REF_KEY &&
10077 (offset == root->objectid || offset == owner)) {
10079 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
10081 * Backref of tree reloc root points to itself, no need
10082 * to check backref any more.
10084 if (tree_reloc_root)
10087 /* Check if the backref points to valid referencer */
10088 found_ref = !check_tree_block_ref(root, NULL,
10089 offset, level + 1, owner);
10094 ptr += btrfs_extent_inline_ref_size(type);
10098 * Inlined extent item doesn't have what we need, check
10099 * TREE_BLOCK_REF_KEY
10102 btrfs_release_path(&path);
10103 key.objectid = bytenr;
10104 key.type = BTRFS_TREE_BLOCK_REF_KEY;
10105 key.offset = root->objectid;
10107 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
10112 err |= BACKREF_MISSING;
10114 btrfs_release_path(&path);
10115 if (eb && (err & BACKREF_MISSING))
10116 error("extent[%llu %u] backref lost (owner: %llu, level: %u)",
10117 bytenr, nodesize, owner, level);
10122 * Check EXTENT_DATA item, mainly for its dbackref in extent tree
10124 * Return >0 any error found and output error message
10125 * Return 0 for no error found
10127 static int check_extent_data_item(struct btrfs_root *root,
10128 struct extent_buffer *eb, int slot)
10130 struct btrfs_file_extent_item *fi;
10131 struct btrfs_path path;
10132 struct btrfs_root *extent_root = root->fs_info->extent_root;
10133 struct btrfs_key fi_key;
10134 struct btrfs_key dbref_key;
10135 struct extent_buffer *leaf;
10136 struct btrfs_extent_item *ei;
10137 struct btrfs_extent_inline_ref *iref;
10138 struct btrfs_extent_data_ref *dref;
10141 u64 disk_num_bytes;
10142 u64 extent_num_bytes;
10149 int found_dbackref = 0;
10153 btrfs_item_key_to_cpu(eb, &fi_key, slot);
10154 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
10156 /* Nothing to check for hole and inline data extents */
10157 if (btrfs_file_extent_type(eb, fi) == BTRFS_FILE_EXTENT_INLINE ||
10158 btrfs_file_extent_disk_bytenr(eb, fi) == 0)
10161 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
10162 disk_num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
10163 extent_num_bytes = btrfs_file_extent_num_bytes(eb, fi);
10165 /* Check unaligned disk_num_bytes and num_bytes */
10166 if (!IS_ALIGNED(disk_num_bytes, root->sectorsize)) {
10168 "file extent [%llu, %llu] has unaligned disk num bytes: %llu, should be aligned to %u",
10169 fi_key.objectid, fi_key.offset, disk_num_bytes,
10171 err |= BYTES_UNALIGNED;
10173 data_bytes_allocated += disk_num_bytes;
10175 if (!IS_ALIGNED(extent_num_bytes, root->sectorsize)) {
10177 "file extent [%llu, %llu] has unaligned num bytes: %llu, should be aligned to %u",
10178 fi_key.objectid, fi_key.offset, extent_num_bytes,
10180 err |= BYTES_UNALIGNED;
10182 data_bytes_referenced += extent_num_bytes;
10184 owner = btrfs_header_owner(eb);
10186 /* Check the extent item of the file extent in extent tree */
10187 btrfs_init_path(&path);
10188 dbref_key.objectid = btrfs_file_extent_disk_bytenr(eb, fi);
10189 dbref_key.type = BTRFS_EXTENT_ITEM_KEY;
10190 dbref_key.offset = btrfs_file_extent_disk_num_bytes(eb, fi);
10192 ret = btrfs_search_slot(NULL, extent_root, &dbref_key, &path, 0, 0);
10194 err |= BACKREF_MISSING;
10198 leaf = path.nodes[0];
10199 slot = path.slots[0];
10200 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
10202 extent_flags = btrfs_extent_flags(leaf, ei);
10204 if (!(extent_flags & BTRFS_EXTENT_FLAG_DATA)) {
10206 "extent[%llu %llu] backref type mismatch, wanted bit: %llx",
10207 disk_bytenr, disk_num_bytes,
10208 BTRFS_EXTENT_FLAG_DATA);
10209 err |= BACKREF_MISMATCH;
10212 /* Check data backref inside that extent item */
10213 item_size = btrfs_item_size_nr(leaf, path.slots[0]);
10214 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
10215 ptr = (unsigned long)iref;
10216 end = (unsigned long)ei + item_size;
10217 while (ptr < end) {
10218 iref = (struct btrfs_extent_inline_ref *)ptr;
10219 type = btrfs_extent_inline_ref_type(leaf, iref);
10220 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
10222 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
10223 ref_root = btrfs_extent_data_ref_root(leaf, dref);
10224 if (ref_root == owner || ref_root == root->objectid)
10225 found_dbackref = 1;
10226 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
10227 found_dbackref = !check_tree_block_ref(root, NULL,
10228 btrfs_extent_inline_ref_offset(leaf, iref),
10232 if (found_dbackref)
10234 ptr += btrfs_extent_inline_ref_size(type);
10237 /* Didn't found inlined data backref, try EXTENT_DATA_REF_KEY */
10238 if (!found_dbackref) {
10239 btrfs_release_path(&path);
10241 btrfs_init_path(&path);
10242 dbref_key.objectid = btrfs_file_extent_disk_bytenr(eb, fi);
10243 dbref_key.type = BTRFS_EXTENT_DATA_REF_KEY;
10244 dbref_key.offset = hash_extent_data_ref(root->objectid,
10245 fi_key.objectid, fi_key.offset);
10247 ret = btrfs_search_slot(NULL, root->fs_info->extent_root,
10248 &dbref_key, &path, 0, 0);
10250 found_dbackref = 1;
10253 if (!found_dbackref)
10254 err |= BACKREF_MISSING;
10256 btrfs_release_path(&path);
10257 if (err & BACKREF_MISSING) {
10258 error("data extent[%llu %llu] backref lost",
10259 disk_bytenr, disk_num_bytes);
10265 * Get real tree block level for the case like shared block
10266 * Return >= 0 as tree level
10267 * Return <0 for error
10269 static int query_tree_block_level(struct btrfs_fs_info *fs_info, u64 bytenr)
10271 struct extent_buffer *eb;
10272 struct btrfs_path path;
10273 struct btrfs_key key;
10274 struct btrfs_extent_item *ei;
10277 u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
10282 /* Search extent tree for extent generation and level */
10283 key.objectid = bytenr;
10284 key.type = BTRFS_METADATA_ITEM_KEY;
10285 key.offset = (u64)-1;
10287 btrfs_init_path(&path);
10288 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, &path, 0, 0);
10291 ret = btrfs_previous_extent_item(fs_info->extent_root, &path, bytenr);
10299 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
10300 ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
10301 struct btrfs_extent_item);
10302 flags = btrfs_extent_flags(path.nodes[0], ei);
10303 if (!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
10308 /* Get transid for later read_tree_block() check */
10309 transid = btrfs_extent_generation(path.nodes[0], ei);
10311 /* Get backref level as one source */
10312 if (key.type == BTRFS_METADATA_ITEM_KEY) {
10313 backref_level = key.offset;
10315 struct btrfs_tree_block_info *info;
10317 info = (struct btrfs_tree_block_info *)(ei + 1);
10318 backref_level = btrfs_tree_block_level(path.nodes[0], info);
10320 btrfs_release_path(&path);
10322 /* Get level from tree block as an alternative source */
10323 eb = read_tree_block_fs_info(fs_info, bytenr, nodesize, transid);
10324 if (!extent_buffer_uptodate(eb)) {
10325 free_extent_buffer(eb);
10328 header_level = btrfs_header_level(eb);
10329 free_extent_buffer(eb);
10331 if (header_level != backref_level)
10333 return header_level;
10336 btrfs_release_path(&path);
10341 * Check if a tree block backref is valid (points to a valid tree block)
10342 * if level == -1, level will be resolved
10343 * Return >0 for any error found and print error message
10345 static int check_tree_block_backref(struct btrfs_fs_info *fs_info, u64 root_id,
10346 u64 bytenr, int level)
10348 struct btrfs_root *root;
10349 struct btrfs_key key;
10350 struct btrfs_path path;
10351 struct extent_buffer *eb;
10352 struct extent_buffer *node;
10353 u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
10357 /* Query level for level == -1 special case */
10359 level = query_tree_block_level(fs_info, bytenr);
10361 err |= REFERENCER_MISSING;
10365 key.objectid = root_id;
10366 key.type = BTRFS_ROOT_ITEM_KEY;
10367 key.offset = (u64)-1;
10369 root = btrfs_read_fs_root(fs_info, &key);
10370 if (IS_ERR(root)) {
10371 err |= REFERENCER_MISSING;
10375 /* Read out the tree block to get item/node key */
10376 eb = read_tree_block(root, bytenr, root->nodesize, 0);
10377 if (!extent_buffer_uptodate(eb)) {
10378 err |= REFERENCER_MISSING;
10379 free_extent_buffer(eb);
10383 /* Empty tree, no need to check key */
10384 if (!btrfs_header_nritems(eb) && !level) {
10385 free_extent_buffer(eb);
10390 btrfs_node_key_to_cpu(eb, &key, 0);
10392 btrfs_item_key_to_cpu(eb, &key, 0);
10394 free_extent_buffer(eb);
10396 btrfs_init_path(&path);
10397 path.lowest_level = level;
10398 /* Search with the first key, to ensure we can reach it */
10399 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
10401 err |= REFERENCER_MISSING;
10405 node = path.nodes[level];
10406 if (btrfs_header_bytenr(node) != bytenr) {
10408 "extent [%llu %d] referencer bytenr mismatch, wanted: %llu, have: %llu",
10409 bytenr, nodesize, bytenr,
10410 btrfs_header_bytenr(node));
10411 err |= REFERENCER_MISMATCH;
10413 if (btrfs_header_level(node) != level) {
10415 "extent [%llu %d] referencer level mismatch, wanted: %d, have: %d",
10416 bytenr, nodesize, level,
10417 btrfs_header_level(node));
10418 err |= REFERENCER_MISMATCH;
10422 btrfs_release_path(&path);
10424 if (err & REFERENCER_MISSING) {
10426 error("extent [%llu %d] lost referencer (owner: %llu)",
10427 bytenr, nodesize, root_id);
10430 "extent [%llu %d] lost referencer (owner: %llu, level: %u)",
10431 bytenr, nodesize, root_id, level);
10438 * Check if tree block @eb is tree reloc root.
10439 * Return 0 if it's not or any problem happens
10440 * Return 1 if it's a tree reloc root
10442 static int is_tree_reloc_root(struct btrfs_fs_info *fs_info,
10443 struct extent_buffer *eb)
10445 struct btrfs_root *tree_reloc_root;
10446 struct btrfs_key key;
10447 u64 bytenr = btrfs_header_bytenr(eb);
10448 u64 owner = btrfs_header_owner(eb);
10451 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
10452 key.offset = owner;
10453 key.type = BTRFS_ROOT_ITEM_KEY;
10455 tree_reloc_root = btrfs_read_fs_root_no_cache(fs_info, &key);
10456 if (IS_ERR(tree_reloc_root))
10459 if (bytenr == btrfs_header_bytenr(tree_reloc_root->node))
10461 btrfs_free_fs_root(tree_reloc_root);
10466 * Check referencer for shared block backref
10467 * If level == -1, this function will resolve the level.
10469 static int check_shared_block_backref(struct btrfs_fs_info *fs_info,
10470 u64 parent, u64 bytenr, int level)
10472 struct extent_buffer *eb;
10473 u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
10475 int found_parent = 0;
10478 eb = read_tree_block_fs_info(fs_info, parent, nodesize, 0);
10479 if (!extent_buffer_uptodate(eb))
10483 level = query_tree_block_level(fs_info, bytenr);
10487 /* It's possible it's a tree reloc root */
10488 if (parent == bytenr) {
10489 if (is_tree_reloc_root(fs_info, eb))
10494 if (level + 1 != btrfs_header_level(eb))
10497 nr = btrfs_header_nritems(eb);
10498 for (i = 0; i < nr; i++) {
10499 if (bytenr == btrfs_node_blockptr(eb, i)) {
10505 free_extent_buffer(eb);
10506 if (!found_parent) {
10508 "shared extent[%llu %u] lost its parent (parent: %llu, level: %u)",
10509 bytenr, nodesize, parent, level);
10510 return REFERENCER_MISSING;
10516 * Check referencer for normal (inlined) data ref
10517 * If len == 0, it will be resolved by searching in extent tree
10519 static int check_extent_data_backref(struct btrfs_fs_info *fs_info,
10520 u64 root_id, u64 objectid, u64 offset,
10521 u64 bytenr, u64 len, u32 count)
10523 struct btrfs_root *root;
10524 struct btrfs_root *extent_root = fs_info->extent_root;
10525 struct btrfs_key key;
10526 struct btrfs_path path;
10527 struct extent_buffer *leaf;
10528 struct btrfs_file_extent_item *fi;
10529 u32 found_count = 0;
10534 key.objectid = bytenr;
10535 key.type = BTRFS_EXTENT_ITEM_KEY;
10536 key.offset = (u64)-1;
10538 btrfs_init_path(&path);
10539 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
10542 ret = btrfs_previous_extent_item(extent_root, &path, bytenr);
10545 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
10546 if (key.objectid != bytenr ||
10547 key.type != BTRFS_EXTENT_ITEM_KEY)
10550 btrfs_release_path(&path);
10552 key.objectid = root_id;
10553 key.type = BTRFS_ROOT_ITEM_KEY;
10554 key.offset = (u64)-1;
10555 btrfs_init_path(&path);
10557 root = btrfs_read_fs_root(fs_info, &key);
10561 key.objectid = objectid;
10562 key.type = BTRFS_EXTENT_DATA_KEY;
10564 * It can be nasty as data backref offset is
10565 * file offset - file extent offset, which is smaller or
10566 * equal to original backref offset. The only special case is
10567 * overflow. So we need to special check and do further search.
10569 key.offset = offset & (1ULL << 63) ? 0 : offset;
10571 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
10576 * Search afterwards to get correct one
10577 * NOTE: As we must do a comprehensive check on the data backref to
10578 * make sure the dref count also matches, we must iterate all file
10579 * extents for that inode.
10582 leaf = path.nodes[0];
10583 slot = path.slots[0];
10585 btrfs_item_key_to_cpu(leaf, &key, slot);
10586 if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
10588 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
10590 * Except normal disk bytenr and disk num bytes, we still
10591 * need to do extra check on dbackref offset as
10592 * dbackref offset = file_offset - file_extent_offset
10594 if (btrfs_file_extent_disk_bytenr(leaf, fi) == bytenr &&
10595 btrfs_file_extent_disk_num_bytes(leaf, fi) == len &&
10596 (u64)(key.offset - btrfs_file_extent_offset(leaf, fi)) ==
10600 ret = btrfs_next_item(root, &path);
10605 btrfs_release_path(&path);
10606 if (found_count != count) {
10608 "extent[%llu, %llu] referencer count mismatch (root: %llu, owner: %llu, offset: %llu) wanted: %u, have: %u",
10609 bytenr, len, root_id, objectid, offset, count, found_count);
10610 return REFERENCER_MISSING;
10616 * Check if the referencer of a shared data backref exists
10618 static int check_shared_data_backref(struct btrfs_fs_info *fs_info,
10619 u64 parent, u64 bytenr)
10621 struct extent_buffer *eb;
10622 struct btrfs_key key;
10623 struct btrfs_file_extent_item *fi;
10624 u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
10626 int found_parent = 0;
10629 eb = read_tree_block_fs_info(fs_info, parent, nodesize, 0);
10630 if (!extent_buffer_uptodate(eb))
10633 nr = btrfs_header_nritems(eb);
10634 for (i = 0; i < nr; i++) {
10635 btrfs_item_key_to_cpu(eb, &key, i);
10636 if (key.type != BTRFS_EXTENT_DATA_KEY)
10639 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
10640 if (btrfs_file_extent_type(eb, fi) == BTRFS_FILE_EXTENT_INLINE)
10643 if (btrfs_file_extent_disk_bytenr(eb, fi) == bytenr) {
10650 free_extent_buffer(eb);
10651 if (!found_parent) {
10652 error("shared extent %llu referencer lost (parent: %llu)",
10654 return REFERENCER_MISSING;
10660 * This function will check a given extent item, including its backref and
10661 * itself (like crossing stripe boundary and type)
10663 * Since we don't use extent_record anymore, introduce new error bit
10665 static int check_extent_item(struct btrfs_fs_info *fs_info,
10666 struct extent_buffer *eb, int slot)
10668 struct btrfs_extent_item *ei;
10669 struct btrfs_extent_inline_ref *iref;
10670 struct btrfs_extent_data_ref *dref;
10674 u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
10675 u32 item_size = btrfs_item_size_nr(eb, slot);
10680 struct btrfs_key key;
10684 btrfs_item_key_to_cpu(eb, &key, slot);
10685 if (key.type == BTRFS_EXTENT_ITEM_KEY)
10686 bytes_used += key.offset;
10688 bytes_used += nodesize;
10690 if (item_size < sizeof(*ei)) {
10692 * COMPAT_EXTENT_TREE_V0 case, but it's already a super
10693 * old thing when on disk format is still un-determined.
10694 * No need to care about it anymore
10696 error("unsupported COMPAT_EXTENT_TREE_V0 detected");
10700 ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
10701 flags = btrfs_extent_flags(eb, ei);
10703 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
10705 if (metadata && check_crossing_stripes(global_info, key.objectid,
10707 error("bad metadata [%llu, %llu) crossing stripe boundary",
10708 key.objectid, key.objectid + nodesize);
10709 err |= CROSSING_STRIPE_BOUNDARY;
10712 ptr = (unsigned long)(ei + 1);
10714 if (metadata && key.type == BTRFS_EXTENT_ITEM_KEY) {
10715 /* Old EXTENT_ITEM metadata */
10716 struct btrfs_tree_block_info *info;
10718 info = (struct btrfs_tree_block_info *)ptr;
10719 level = btrfs_tree_block_level(eb, info);
10720 ptr += sizeof(struct btrfs_tree_block_info);
10722 /* New METADATA_ITEM */
10723 level = key.offset;
10725 end = (unsigned long)ei + item_size;
10728 err |= ITEM_SIZE_MISMATCH;
10732 /* Now check every backref in this extent item */
10734 iref = (struct btrfs_extent_inline_ref *)ptr;
10735 type = btrfs_extent_inline_ref_type(eb, iref);
10736 offset = btrfs_extent_inline_ref_offset(eb, iref);
10738 case BTRFS_TREE_BLOCK_REF_KEY:
10739 ret = check_tree_block_backref(fs_info, offset, key.objectid,
10743 case BTRFS_SHARED_BLOCK_REF_KEY:
10744 ret = check_shared_block_backref(fs_info, offset, key.objectid,
10748 case BTRFS_EXTENT_DATA_REF_KEY:
10749 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
10750 ret = check_extent_data_backref(fs_info,
10751 btrfs_extent_data_ref_root(eb, dref),
10752 btrfs_extent_data_ref_objectid(eb, dref),
10753 btrfs_extent_data_ref_offset(eb, dref),
10754 key.objectid, key.offset,
10755 btrfs_extent_data_ref_count(eb, dref));
10758 case BTRFS_SHARED_DATA_REF_KEY:
10759 ret = check_shared_data_backref(fs_info, offset, key.objectid);
10763 error("extent[%llu %d %llu] has unknown ref type: %d",
10764 key.objectid, key.type, key.offset, type);
10765 err |= UNKNOWN_TYPE;
10769 ptr += btrfs_extent_inline_ref_size(type);
10778 * Check if a dev extent item is referred correctly by its chunk
10780 static int check_dev_extent_item(struct btrfs_fs_info *fs_info,
10781 struct extent_buffer *eb, int slot)
10783 struct btrfs_root *chunk_root = fs_info->chunk_root;
10784 struct btrfs_dev_extent *ptr;
10785 struct btrfs_path path;
10786 struct btrfs_key chunk_key;
10787 struct btrfs_key devext_key;
10788 struct btrfs_chunk *chunk;
10789 struct extent_buffer *l;
10793 int found_chunk = 0;
10796 btrfs_item_key_to_cpu(eb, &devext_key, slot);
10797 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_extent);
10798 length = btrfs_dev_extent_length(eb, ptr);
10800 chunk_key.objectid = btrfs_dev_extent_chunk_objectid(eb, ptr);
10801 chunk_key.type = BTRFS_CHUNK_ITEM_KEY;
10802 chunk_key.offset = btrfs_dev_extent_chunk_offset(eb, ptr);
10804 btrfs_init_path(&path);
10805 ret = btrfs_search_slot(NULL, chunk_root, &chunk_key, &path, 0, 0);
10810 chunk = btrfs_item_ptr(l, path.slots[0], struct btrfs_chunk);
10811 if (btrfs_chunk_length(l, chunk) != length)
10814 num_stripes = btrfs_chunk_num_stripes(l, chunk);
10815 for (i = 0; i < num_stripes; i++) {
10816 u64 devid = btrfs_stripe_devid_nr(l, chunk, i);
10817 u64 offset = btrfs_stripe_offset_nr(l, chunk, i);
10819 if (devid == devext_key.objectid &&
10820 offset == devext_key.offset) {
10826 btrfs_release_path(&path);
10827 if (!found_chunk) {
10829 "device extent[%llu, %llu, %llu] did not find the related chunk",
10830 devext_key.objectid, devext_key.offset, length);
10831 return REFERENCER_MISSING;
10837 * Check if the used space is correct with the dev item
10839 static int check_dev_item(struct btrfs_fs_info *fs_info,
10840 struct extent_buffer *eb, int slot)
10842 struct btrfs_root *dev_root = fs_info->dev_root;
10843 struct btrfs_dev_item *dev_item;
10844 struct btrfs_path path;
10845 struct btrfs_key key;
10846 struct btrfs_dev_extent *ptr;
10852 dev_item = btrfs_item_ptr(eb, slot, struct btrfs_dev_item);
10853 dev_id = btrfs_device_id(eb, dev_item);
10854 used = btrfs_device_bytes_used(eb, dev_item);
10856 key.objectid = dev_id;
10857 key.type = BTRFS_DEV_EXTENT_KEY;
10860 btrfs_init_path(&path);
10861 ret = btrfs_search_slot(NULL, dev_root, &key, &path, 0, 0);
10863 btrfs_item_key_to_cpu(eb, &key, slot);
10864 error("cannot find any related dev extent for dev[%llu, %u, %llu]",
10865 key.objectid, key.type, key.offset);
10866 btrfs_release_path(&path);
10867 return REFERENCER_MISSING;
10870 /* Iterate dev_extents to calculate the used space of a device */
10872 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
10874 if (key.objectid > dev_id)
10876 if (key.type != BTRFS_DEV_EXTENT_KEY || key.objectid != dev_id)
10879 ptr = btrfs_item_ptr(path.nodes[0], path.slots[0],
10880 struct btrfs_dev_extent);
10881 total += btrfs_dev_extent_length(path.nodes[0], ptr);
10883 ret = btrfs_next_item(dev_root, &path);
10887 btrfs_release_path(&path);
10889 if (used != total) {
10890 btrfs_item_key_to_cpu(eb, &key, slot);
10892 "Dev extent's total-byte %llu is not equal to bytes-used %llu in dev[%llu, %u, %llu]",
10893 total, used, BTRFS_ROOT_TREE_OBJECTID,
10894 BTRFS_DEV_EXTENT_KEY, dev_id);
10895 return ACCOUNTING_MISMATCH;
10901 * Check a block group item with its referener (chunk) and its used space
10902 * with extent/metadata item
10904 static int check_block_group_item(struct btrfs_fs_info *fs_info,
10905 struct extent_buffer *eb, int slot)
10907 struct btrfs_root *extent_root = fs_info->extent_root;
10908 struct btrfs_root *chunk_root = fs_info->chunk_root;
10909 struct btrfs_block_group_item *bi;
10910 struct btrfs_block_group_item bg_item;
10911 struct btrfs_path path;
10912 struct btrfs_key bg_key;
10913 struct btrfs_key chunk_key;
10914 struct btrfs_key extent_key;
10915 struct btrfs_chunk *chunk;
10916 struct extent_buffer *leaf;
10917 struct btrfs_extent_item *ei;
10918 u32 nodesize = btrfs_super_nodesize(fs_info->super_copy);
10926 btrfs_item_key_to_cpu(eb, &bg_key, slot);
10927 bi = btrfs_item_ptr(eb, slot, struct btrfs_block_group_item);
10928 read_extent_buffer(eb, &bg_item, (unsigned long)bi, sizeof(bg_item));
10929 used = btrfs_block_group_used(&bg_item);
10930 bg_flags = btrfs_block_group_flags(&bg_item);
10932 chunk_key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
10933 chunk_key.type = BTRFS_CHUNK_ITEM_KEY;
10934 chunk_key.offset = bg_key.objectid;
10936 btrfs_init_path(&path);
10937 /* Search for the referencer chunk */
10938 ret = btrfs_search_slot(NULL, chunk_root, &chunk_key, &path, 0, 0);
10941 "block group[%llu %llu] did not find the related chunk item",
10942 bg_key.objectid, bg_key.offset);
10943 err |= REFERENCER_MISSING;
10945 chunk = btrfs_item_ptr(path.nodes[0], path.slots[0],
10946 struct btrfs_chunk);
10947 if (btrfs_chunk_length(path.nodes[0], chunk) !=
10950 "block group[%llu %llu] related chunk item length does not match",
10951 bg_key.objectid, bg_key.offset);
10952 err |= REFERENCER_MISMATCH;
10955 btrfs_release_path(&path);
10957 /* Search from the block group bytenr */
10958 extent_key.objectid = bg_key.objectid;
10959 extent_key.type = 0;
10960 extent_key.offset = 0;
10962 btrfs_init_path(&path);
10963 ret = btrfs_search_slot(NULL, extent_root, &extent_key, &path, 0, 0);
10967 /* Iterate extent tree to account used space */
10969 leaf = path.nodes[0];
10970 btrfs_item_key_to_cpu(leaf, &extent_key, path.slots[0]);
10971 if (extent_key.objectid >= bg_key.objectid + bg_key.offset)
10974 if (extent_key.type != BTRFS_METADATA_ITEM_KEY &&
10975 extent_key.type != BTRFS_EXTENT_ITEM_KEY)
10977 if (extent_key.objectid < bg_key.objectid)
10980 if (extent_key.type == BTRFS_METADATA_ITEM_KEY)
10983 total += extent_key.offset;
10985 ei = btrfs_item_ptr(leaf, path.slots[0],
10986 struct btrfs_extent_item);
10987 flags = btrfs_extent_flags(leaf, ei);
10988 if (flags & BTRFS_EXTENT_FLAG_DATA) {
10989 if (!(bg_flags & BTRFS_BLOCK_GROUP_DATA)) {
10991 "bad extent[%llu, %llu) type mismatch with chunk",
10992 extent_key.objectid,
10993 extent_key.objectid + extent_key.offset);
10994 err |= CHUNK_TYPE_MISMATCH;
10996 } else if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
10997 if (!(bg_flags & (BTRFS_BLOCK_GROUP_SYSTEM |
10998 BTRFS_BLOCK_GROUP_METADATA))) {
11000 "bad extent[%llu, %llu) type mismatch with chunk",
11001 extent_key.objectid,
11002 extent_key.objectid + nodesize);
11003 err |= CHUNK_TYPE_MISMATCH;
11007 ret = btrfs_next_item(extent_root, &path);
11013 btrfs_release_path(&path);
11015 if (total != used) {
11017 "block group[%llu %llu] used %llu but extent items used %llu",
11018 bg_key.objectid, bg_key.offset, used, total);
11019 err |= ACCOUNTING_MISMATCH;
11025 * Check a chunk item.
11026 * Including checking all referred dev_extents and block group
11028 static int check_chunk_item(struct btrfs_fs_info *fs_info,
11029 struct extent_buffer *eb, int slot)
11031 struct btrfs_root *extent_root = fs_info->extent_root;
11032 struct btrfs_root *dev_root = fs_info->dev_root;
11033 struct btrfs_path path;
11034 struct btrfs_key chunk_key;
11035 struct btrfs_key bg_key;
11036 struct btrfs_key devext_key;
11037 struct btrfs_chunk *chunk;
11038 struct extent_buffer *leaf;
11039 struct btrfs_block_group_item *bi;
11040 struct btrfs_block_group_item bg_item;
11041 struct btrfs_dev_extent *ptr;
11042 u32 sectorsize = btrfs_super_sectorsize(fs_info->super_copy);
11054 btrfs_item_key_to_cpu(eb, &chunk_key, slot);
11055 chunk = btrfs_item_ptr(eb, slot, struct btrfs_chunk);
11056 length = btrfs_chunk_length(eb, chunk);
11057 chunk_end = chunk_key.offset + length;
11058 if (!IS_ALIGNED(length, sectorsize)) {
11059 error("chunk[%llu %llu) not aligned to %u",
11060 chunk_key.offset, chunk_end, sectorsize);
11061 err |= BYTES_UNALIGNED;
11065 type = btrfs_chunk_type(eb, chunk);
11066 profile = type & BTRFS_BLOCK_GROUP_PROFILE_MASK;
11067 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
11068 error("chunk[%llu %llu) has no chunk type",
11069 chunk_key.offset, chunk_end);
11070 err |= UNKNOWN_TYPE;
11072 if (profile && (profile & (profile - 1))) {
11073 error("chunk[%llu %llu) multiple profiles detected: %llx",
11074 chunk_key.offset, chunk_end, profile);
11075 err |= UNKNOWN_TYPE;
11078 bg_key.objectid = chunk_key.offset;
11079 bg_key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
11080 bg_key.offset = length;
11082 btrfs_init_path(&path);
11083 ret = btrfs_search_slot(NULL, extent_root, &bg_key, &path, 0, 0);
11086 "chunk[%llu %llu) did not find the related block group item",
11087 chunk_key.offset, chunk_end);
11088 err |= REFERENCER_MISSING;
11090 leaf = path.nodes[0];
11091 bi = btrfs_item_ptr(leaf, path.slots[0],
11092 struct btrfs_block_group_item);
11093 read_extent_buffer(leaf, &bg_item, (unsigned long)bi,
11095 if (btrfs_block_group_flags(&bg_item) != type) {
11097 "chunk[%llu %llu) related block group item flags mismatch, wanted: %llu, have: %llu",
11098 chunk_key.offset, chunk_end, type,
11099 btrfs_block_group_flags(&bg_item));
11100 err |= REFERENCER_MISSING;
11104 num_stripes = btrfs_chunk_num_stripes(eb, chunk);
11105 for (i = 0; i < num_stripes; i++) {
11106 btrfs_release_path(&path);
11107 btrfs_init_path(&path);
11108 devext_key.objectid = btrfs_stripe_devid_nr(eb, chunk, i);
11109 devext_key.type = BTRFS_DEV_EXTENT_KEY;
11110 devext_key.offset = btrfs_stripe_offset_nr(eb, chunk, i);
11112 ret = btrfs_search_slot(NULL, dev_root, &devext_key, &path,
11115 goto not_match_dev;
11117 leaf = path.nodes[0];
11118 ptr = btrfs_item_ptr(leaf, path.slots[0],
11119 struct btrfs_dev_extent);
11120 objectid = btrfs_dev_extent_chunk_objectid(leaf, ptr);
11121 offset = btrfs_dev_extent_chunk_offset(leaf, ptr);
11122 if (objectid != chunk_key.objectid ||
11123 offset != chunk_key.offset ||
11124 btrfs_dev_extent_length(leaf, ptr) != length)
11125 goto not_match_dev;
11128 err |= BACKREF_MISSING;
11130 "chunk[%llu %llu) stripe %d did not find the related dev extent",
11131 chunk_key.objectid, chunk_end, i);
11134 btrfs_release_path(&path);
11140 * Main entry function to check known items and update related accounting info
11142 static int check_leaf_items(struct btrfs_root *root, struct extent_buffer *eb)
11144 struct btrfs_fs_info *fs_info = root->fs_info;
11145 struct btrfs_key key;
11148 struct btrfs_extent_data_ref *dref;
11153 btrfs_item_key_to_cpu(eb, &key, slot);
11157 case BTRFS_EXTENT_DATA_KEY:
11158 ret = check_extent_data_item(root, eb, slot);
11161 case BTRFS_BLOCK_GROUP_ITEM_KEY:
11162 ret = check_block_group_item(fs_info, eb, slot);
11165 case BTRFS_DEV_ITEM_KEY:
11166 ret = check_dev_item(fs_info, eb, slot);
11169 case BTRFS_CHUNK_ITEM_KEY:
11170 ret = check_chunk_item(fs_info, eb, slot);
11173 case BTRFS_DEV_EXTENT_KEY:
11174 ret = check_dev_extent_item(fs_info, eb, slot);
11177 case BTRFS_EXTENT_ITEM_KEY:
11178 case BTRFS_METADATA_ITEM_KEY:
11179 ret = check_extent_item(fs_info, eb, slot);
11182 case BTRFS_EXTENT_CSUM_KEY:
11183 total_csum_bytes += btrfs_item_size_nr(eb, slot);
11185 case BTRFS_TREE_BLOCK_REF_KEY:
11186 ret = check_tree_block_backref(fs_info, key.offset,
11190 case BTRFS_EXTENT_DATA_REF_KEY:
11191 dref = btrfs_item_ptr(eb, slot, struct btrfs_extent_data_ref);
11192 ret = check_extent_data_backref(fs_info,
11193 btrfs_extent_data_ref_root(eb, dref),
11194 btrfs_extent_data_ref_objectid(eb, dref),
11195 btrfs_extent_data_ref_offset(eb, dref),
11197 btrfs_extent_data_ref_count(eb, dref));
11200 case BTRFS_SHARED_BLOCK_REF_KEY:
11201 ret = check_shared_block_backref(fs_info, key.offset,
11205 case BTRFS_SHARED_DATA_REF_KEY:
11206 ret = check_shared_data_backref(fs_info, key.offset,
11214 if (++slot < btrfs_header_nritems(eb))
11221 * Helper function for later fs/subvol tree check. To determine if a tree
11222 * block should be checked.
11223 * This function will ensure only the direct referencer with lowest rootid to
11224 * check a fs/subvolume tree block.
11226 * Backref check at extent tree would detect errors like missing subvolume
11227 * tree, so we can do aggressive check to reduce duplicated checks.
11229 static int should_check(struct btrfs_root *root, struct extent_buffer *eb)
11231 struct btrfs_root *extent_root = root->fs_info->extent_root;
11232 struct btrfs_key key;
11233 struct btrfs_path path;
11234 struct extent_buffer *leaf;
11236 struct btrfs_extent_item *ei;
11242 struct btrfs_extent_inline_ref *iref;
11245 btrfs_init_path(&path);
11246 key.objectid = btrfs_header_bytenr(eb);
11247 key.type = BTRFS_METADATA_ITEM_KEY;
11248 key.offset = (u64)-1;
11251 * Any failure in backref resolving means we can't determine
11252 * whom the tree block belongs to.
11253 * So in that case, we need to check that tree block
11255 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
11259 ret = btrfs_previous_extent_item(extent_root, &path,
11260 btrfs_header_bytenr(eb));
11264 leaf = path.nodes[0];
11265 slot = path.slots[0];
11266 btrfs_item_key_to_cpu(leaf, &key, slot);
11267 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
11269 if (key.type == BTRFS_METADATA_ITEM_KEY) {
11270 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
11272 struct btrfs_tree_block_info *info;
11274 info = (struct btrfs_tree_block_info *)(ei + 1);
11275 iref = (struct btrfs_extent_inline_ref *)(info + 1);
11278 item_size = btrfs_item_size_nr(leaf, slot);
11279 ptr = (unsigned long)iref;
11280 end = (unsigned long)ei + item_size;
11281 while (ptr < end) {
11282 iref = (struct btrfs_extent_inline_ref *)ptr;
11283 type = btrfs_extent_inline_ref_type(leaf, iref);
11284 offset = btrfs_extent_inline_ref_offset(leaf, iref);
11287 * We only check the tree block if current root is
11288 * the lowest referencer of it.
11290 if (type == BTRFS_TREE_BLOCK_REF_KEY &&
11291 offset < root->objectid) {
11292 btrfs_release_path(&path);
11296 ptr += btrfs_extent_inline_ref_size(type);
11299 * Normally we should also check keyed tree block ref, but that may be
11300 * very time consuming. Inlined ref should already make us skip a lot
11301 * of refs now. So skip search keyed tree block ref.
11305 btrfs_release_path(&path);
11310 * Traversal function for tree block. We will do:
11311 * 1) Skip shared fs/subvolume tree blocks
11312 * 2) Update related bytes accounting
11313 * 3) Pre-order traversal
11315 static int traverse_tree_block(struct btrfs_root *root,
11316 struct extent_buffer *node)
11318 struct extent_buffer *eb;
11319 struct btrfs_key key;
11320 struct btrfs_key drop_key;
11328 * Skip shared fs/subvolume tree block, in that case they will
11329 * be checked by referencer with lowest rootid
11331 if (is_fstree(root->objectid) && !should_check(root, node))
11334 /* Update bytes accounting */
11335 total_btree_bytes += node->len;
11336 if (fs_root_objectid(btrfs_header_owner(node)))
11337 total_fs_tree_bytes += node->len;
11338 if (btrfs_header_owner(node) == BTRFS_EXTENT_TREE_OBJECTID)
11339 total_extent_tree_bytes += node->len;
11340 if (!found_old_backref &&
11341 btrfs_header_owner(node) == BTRFS_TREE_RELOC_OBJECTID &&
11342 btrfs_header_backref_rev(node) == BTRFS_MIXED_BACKREF_REV &&
11343 !btrfs_header_flag(node, BTRFS_HEADER_FLAG_RELOC))
11344 found_old_backref = 1;
11346 /* pre-order tranversal, check itself first */
11347 level = btrfs_header_level(node);
11348 ret = check_tree_block_ref(root, node, btrfs_header_bytenr(node),
11349 btrfs_header_level(node),
11350 btrfs_header_owner(node));
11354 "check %s failed root %llu bytenr %llu level %d, force continue check",
11355 level ? "node":"leaf", root->objectid,
11356 btrfs_header_bytenr(node), btrfs_header_level(node));
11359 btree_space_waste += btrfs_leaf_free_space(root, node);
11360 ret = check_leaf_items(root, node);
11365 nr = btrfs_header_nritems(node);
11366 btrfs_disk_key_to_cpu(&drop_key, &root->root_item.drop_progress);
11367 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) - nr) *
11368 sizeof(struct btrfs_key_ptr);
11370 /* Then check all its children */
11371 for (i = 0; i < nr; i++) {
11372 u64 blocknr = btrfs_node_blockptr(node, i);
11374 btrfs_node_key_to_cpu(node, &key, i);
11375 if (level == root->root_item.drop_level &&
11376 is_dropped_key(&key, &drop_key))
11380 * As a btrfs tree has most 8 levels (0..7), so it's quite safe
11381 * to call the function itself.
11383 eb = read_tree_block(root, blocknr, root->nodesize, 0);
11384 if (extent_buffer_uptodate(eb)) {
11385 ret = traverse_tree_block(root, eb);
11388 free_extent_buffer(eb);
11395 * Low memory usage version check_chunks_and_extents.
11397 static int check_chunks_and_extents_v2(struct btrfs_root *root)
11399 struct btrfs_path path;
11400 struct btrfs_key key;
11401 struct btrfs_root *root1;
11402 struct btrfs_root *cur_root;
11406 root1 = root->fs_info->chunk_root;
11407 ret = traverse_tree_block(root1, root1->node);
11410 root1 = root->fs_info->tree_root;
11411 ret = traverse_tree_block(root1, root1->node);
11414 btrfs_init_path(&path);
11415 key.objectid = BTRFS_EXTENT_TREE_OBJECTID;
11417 key.type = BTRFS_ROOT_ITEM_KEY;
11419 ret = btrfs_search_slot(NULL, root1, &key, &path, 0, 0);
11421 error("cannot find extent treet in tree_root");
11426 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
11427 if (key.type != BTRFS_ROOT_ITEM_KEY)
11429 key.offset = (u64)-1;
11431 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
11432 cur_root = btrfs_read_fs_root_no_cache(root->fs_info,
11435 cur_root = btrfs_read_fs_root(root->fs_info, &key);
11436 if (IS_ERR(cur_root) || !cur_root) {
11437 error("failed to read tree: %lld", key.objectid);
11441 ret = traverse_tree_block(cur_root, cur_root->node);
11444 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
11445 btrfs_free_fs_root(cur_root);
11447 ret = btrfs_next_item(root1, &path);
11453 btrfs_release_path(&path);
11457 static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
11458 struct btrfs_root *root, int overwrite)
11460 struct extent_buffer *c;
11461 struct extent_buffer *old = root->node;
11464 struct btrfs_disk_key disk_key = {0,0,0};
11470 extent_buffer_get(c);
11473 c = btrfs_alloc_free_block(trans, root,
11475 root->root_key.objectid,
11476 &disk_key, level, 0, 0);
11479 extent_buffer_get(c);
11483 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
11484 btrfs_set_header_level(c, level);
11485 btrfs_set_header_bytenr(c, c->start);
11486 btrfs_set_header_generation(c, trans->transid);
11487 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
11488 btrfs_set_header_owner(c, root->root_key.objectid);
11490 write_extent_buffer(c, root->fs_info->fsid,
11491 btrfs_header_fsid(), BTRFS_FSID_SIZE);
11493 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
11494 btrfs_header_chunk_tree_uuid(c),
11497 btrfs_mark_buffer_dirty(c);
11499 * this case can happen in the following case:
11501 * 1.overwrite previous root.
11503 * 2.reinit reloc data root, this is because we skip pin
11504 * down reloc data tree before which means we can allocate
11505 * same block bytenr here.
11507 if (old->start == c->start) {
11508 btrfs_set_root_generation(&root->root_item,
11510 root->root_item.level = btrfs_header_level(root->node);
11511 ret = btrfs_update_root(trans, root->fs_info->tree_root,
11512 &root->root_key, &root->root_item);
11514 free_extent_buffer(c);
11518 free_extent_buffer(old);
11520 add_root_to_dirty_list(root);
11524 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
11525 struct extent_buffer *eb, int tree_root)
11527 struct extent_buffer *tmp;
11528 struct btrfs_root_item *ri;
11529 struct btrfs_key key;
11532 int level = btrfs_header_level(eb);
11538 * If we have pinned this block before, don't pin it again.
11539 * This can not only avoid forever loop with broken filesystem
11540 * but also give us some speedups.
11542 if (test_range_bit(&fs_info->pinned_extents, eb->start,
11543 eb->start + eb->len - 1, EXTENT_DIRTY, 0))
11546 btrfs_pin_extent(fs_info, eb->start, eb->len);
11548 nodesize = btrfs_super_nodesize(fs_info->super_copy);
11549 nritems = btrfs_header_nritems(eb);
11550 for (i = 0; i < nritems; i++) {
11552 btrfs_item_key_to_cpu(eb, &key, i);
11553 if (key.type != BTRFS_ROOT_ITEM_KEY)
11555 /* Skip the extent root and reloc roots */
11556 if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
11557 key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
11558 key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
11560 ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
11561 bytenr = btrfs_disk_root_bytenr(eb, ri);
11564 * If at any point we start needing the real root we
11565 * will have to build a stump root for the root we are
11566 * in, but for now this doesn't actually use the root so
11567 * just pass in extent_root.
11569 tmp = read_tree_block(fs_info->extent_root, bytenr,
11571 if (!extent_buffer_uptodate(tmp)) {
11572 fprintf(stderr, "Error reading root block\n");
11575 ret = pin_down_tree_blocks(fs_info, tmp, 0);
11576 free_extent_buffer(tmp);
11580 bytenr = btrfs_node_blockptr(eb, i);
11582 /* If we aren't the tree root don't read the block */
11583 if (level == 1 && !tree_root) {
11584 btrfs_pin_extent(fs_info, bytenr, nodesize);
11588 tmp = read_tree_block(fs_info->extent_root, bytenr,
11590 if (!extent_buffer_uptodate(tmp)) {
11591 fprintf(stderr, "Error reading tree block\n");
11594 ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
11595 free_extent_buffer(tmp);
11604 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
11608 ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
11612 return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
11615 static int reset_block_groups(struct btrfs_fs_info *fs_info)
11617 struct btrfs_block_group_cache *cache;
11618 struct btrfs_path path;
11619 struct extent_buffer *leaf;
11620 struct btrfs_chunk *chunk;
11621 struct btrfs_key key;
11625 btrfs_init_path(&path);
11627 key.type = BTRFS_CHUNK_ITEM_KEY;
11629 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, &path, 0, 0);
11631 btrfs_release_path(&path);
11636 * We do this in case the block groups were screwed up and had alloc
11637 * bits that aren't actually set on the chunks. This happens with
11638 * restored images every time and could happen in real life I guess.
11640 fs_info->avail_data_alloc_bits = 0;
11641 fs_info->avail_metadata_alloc_bits = 0;
11642 fs_info->avail_system_alloc_bits = 0;
11644 /* First we need to create the in-memory block groups */
11646 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
11647 ret = btrfs_next_leaf(fs_info->chunk_root, &path);
11649 btrfs_release_path(&path);
11657 leaf = path.nodes[0];
11658 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
11659 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
11664 chunk = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_chunk);
11665 btrfs_add_block_group(fs_info, 0,
11666 btrfs_chunk_type(leaf, chunk),
11667 key.objectid, key.offset,
11668 btrfs_chunk_length(leaf, chunk));
11669 set_extent_dirty(&fs_info->free_space_cache, key.offset,
11670 key.offset + btrfs_chunk_length(leaf, chunk));
11675 cache = btrfs_lookup_first_block_group(fs_info, start);
11679 start = cache->key.objectid + cache->key.offset;
11682 btrfs_release_path(&path);
11686 static int reset_balance(struct btrfs_trans_handle *trans,
11687 struct btrfs_fs_info *fs_info)
11689 struct btrfs_root *root = fs_info->tree_root;
11690 struct btrfs_path path;
11691 struct extent_buffer *leaf;
11692 struct btrfs_key key;
11693 int del_slot, del_nr = 0;
11697 btrfs_init_path(&path);
11698 key.objectid = BTRFS_BALANCE_OBJECTID;
11699 key.type = BTRFS_BALANCE_ITEM_KEY;
11701 ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
11706 goto reinit_data_reloc;
11711 ret = btrfs_del_item(trans, root, &path);
11714 btrfs_release_path(&path);
11716 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
11717 key.type = BTRFS_ROOT_ITEM_KEY;
11719 ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
11723 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
11728 ret = btrfs_del_items(trans, root, &path,
11735 btrfs_release_path(&path);
11738 ret = btrfs_search_slot(trans, root, &key, &path,
11745 leaf = path.nodes[0];
11746 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
11747 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
11749 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
11754 del_slot = path.slots[0];
11763 ret = btrfs_del_items(trans, root, &path, del_slot, del_nr);
11767 btrfs_release_path(&path);
11770 key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
11771 key.type = BTRFS_ROOT_ITEM_KEY;
11772 key.offset = (u64)-1;
11773 root = btrfs_read_fs_root(fs_info, &key);
11774 if (IS_ERR(root)) {
11775 fprintf(stderr, "Error reading data reloc tree\n");
11776 ret = PTR_ERR(root);
11779 record_root_in_trans(trans, root);
11780 ret = btrfs_fsck_reinit_root(trans, root, 0);
11783 ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
11785 btrfs_release_path(&path);
11789 static int reinit_extent_tree(struct btrfs_trans_handle *trans,
11790 struct btrfs_fs_info *fs_info)
11796 * The only reason we don't do this is because right now we're just
11797 * walking the trees we find and pinning down their bytes, we don't look
11798 * at any of the leaves. In order to do mixed groups we'd have to check
11799 * the leaves of any fs roots and pin down the bytes for any file
11800 * extents we find. Not hard but why do it if we don't have to?
11802 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
11803 fprintf(stderr, "We don't support re-initing the extent tree "
11804 "for mixed block groups yet, please notify a btrfs "
11805 "developer you want to do this so they can add this "
11806 "functionality.\n");
11811 * first we need to walk all of the trees except the extent tree and pin
11812 * down the bytes that are in use so we don't overwrite any existing
11815 ret = pin_metadata_blocks(fs_info);
11817 fprintf(stderr, "error pinning down used bytes\n");
11822 * Need to drop all the block groups since we're going to recreate all
11825 btrfs_free_block_groups(fs_info);
11826 ret = reset_block_groups(fs_info);
11828 fprintf(stderr, "error resetting the block groups\n");
11832 /* Ok we can allocate now, reinit the extent root */
11833 ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 0);
11835 fprintf(stderr, "extent root initialization failed\n");
11837 * When the transaction code is updated we should end the
11838 * transaction, but for now progs only knows about commit so
11839 * just return an error.
11845 * Now we have all the in-memory block groups setup so we can make
11846 * allocations properly, and the metadata we care about is safe since we
11847 * pinned all of it above.
11850 struct btrfs_block_group_cache *cache;
11852 cache = btrfs_lookup_first_block_group(fs_info, start);
11855 start = cache->key.objectid + cache->key.offset;
11856 ret = btrfs_insert_item(trans, fs_info->extent_root,
11857 &cache->key, &cache->item,
11858 sizeof(cache->item));
11860 fprintf(stderr, "Error adding block group\n");
11863 btrfs_extent_post_op(trans, fs_info->extent_root);
11866 ret = reset_balance(trans, fs_info);
11868 fprintf(stderr, "error resetting the pending balance\n");
11873 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
11875 struct btrfs_path path;
11876 struct btrfs_trans_handle *trans;
11877 struct btrfs_key key;
11880 printf("Recowing metadata block %llu\n", eb->start);
11881 key.objectid = btrfs_header_owner(eb);
11882 key.type = BTRFS_ROOT_ITEM_KEY;
11883 key.offset = (u64)-1;
11885 root = btrfs_read_fs_root(root->fs_info, &key);
11886 if (IS_ERR(root)) {
11887 fprintf(stderr, "Couldn't find owner root %llu\n",
11889 return PTR_ERR(root);
11892 trans = btrfs_start_transaction(root, 1);
11894 return PTR_ERR(trans);
11896 btrfs_init_path(&path);
11897 path.lowest_level = btrfs_header_level(eb);
11898 if (path.lowest_level)
11899 btrfs_node_key_to_cpu(eb, &key, 0);
11901 btrfs_item_key_to_cpu(eb, &key, 0);
11903 ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
11904 btrfs_commit_transaction(trans, root);
11905 btrfs_release_path(&path);
11909 static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
11911 struct btrfs_path path;
11912 struct btrfs_trans_handle *trans;
11913 struct btrfs_key key;
11916 printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
11917 bad->key.type, bad->key.offset);
11918 key.objectid = bad->root_id;
11919 key.type = BTRFS_ROOT_ITEM_KEY;
11920 key.offset = (u64)-1;
11922 root = btrfs_read_fs_root(root->fs_info, &key);
11923 if (IS_ERR(root)) {
11924 fprintf(stderr, "Couldn't find owner root %llu\n",
11926 return PTR_ERR(root);
11929 trans = btrfs_start_transaction(root, 1);
11931 return PTR_ERR(trans);
11933 btrfs_init_path(&path);
11934 ret = btrfs_search_slot(trans, root, &bad->key, &path, -1, 1);
11940 ret = btrfs_del_item(trans, root, &path);
11942 btrfs_commit_transaction(trans, root);
11943 btrfs_release_path(&path);
11947 static int zero_log_tree(struct btrfs_root *root)
11949 struct btrfs_trans_handle *trans;
11952 trans = btrfs_start_transaction(root, 1);
11953 if (IS_ERR(trans)) {
11954 ret = PTR_ERR(trans);
11957 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
11958 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
11959 ret = btrfs_commit_transaction(trans, root);
11963 static int populate_csum(struct btrfs_trans_handle *trans,
11964 struct btrfs_root *csum_root, char *buf, u64 start,
11971 while (offset < len) {
11972 sectorsize = csum_root->sectorsize;
11973 ret = read_extent_data(csum_root, buf, start + offset,
11977 ret = btrfs_csum_file_block(trans, csum_root, start + len,
11978 start + offset, buf, sectorsize);
11981 offset += sectorsize;
11986 static int fill_csum_tree_from_one_fs_root(struct btrfs_trans_handle *trans,
11987 struct btrfs_root *csum_root,
11988 struct btrfs_root *cur_root)
11990 struct btrfs_path path;
11991 struct btrfs_key key;
11992 struct extent_buffer *node;
11993 struct btrfs_file_extent_item *fi;
12000 buf = malloc(cur_root->fs_info->csum_root->sectorsize);
12004 btrfs_init_path(&path);
12008 ret = btrfs_search_slot(NULL, cur_root, &key, &path, 0, 0);
12011 /* Iterate all regular file extents and fill its csum */
12013 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
12015 if (key.type != BTRFS_EXTENT_DATA_KEY)
12017 node = path.nodes[0];
12018 slot = path.slots[0];
12019 fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
12020 if (btrfs_file_extent_type(node, fi) != BTRFS_FILE_EXTENT_REG)
12022 start = btrfs_file_extent_disk_bytenr(node, fi);
12023 len = btrfs_file_extent_disk_num_bytes(node, fi);
12025 ret = populate_csum(trans, csum_root, buf, start, len);
12026 if (ret == -EEXIST)
12032 * TODO: if next leaf is corrupted, jump to nearest next valid
12035 ret = btrfs_next_item(cur_root, &path);
12045 btrfs_release_path(&path);
12050 static int fill_csum_tree_from_fs(struct btrfs_trans_handle *trans,
12051 struct btrfs_root *csum_root)
12053 struct btrfs_fs_info *fs_info = csum_root->fs_info;
12054 struct btrfs_path path;
12055 struct btrfs_root *tree_root = fs_info->tree_root;
12056 struct btrfs_root *cur_root;
12057 struct extent_buffer *node;
12058 struct btrfs_key key;
12062 btrfs_init_path(&path);
12063 key.objectid = BTRFS_FS_TREE_OBJECTID;
12065 key.type = BTRFS_ROOT_ITEM_KEY;
12066 ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
12075 node = path.nodes[0];
12076 slot = path.slots[0];
12077 btrfs_item_key_to_cpu(node, &key, slot);
12078 if (key.objectid > BTRFS_LAST_FREE_OBJECTID)
12080 if (key.type != BTRFS_ROOT_ITEM_KEY)
12082 if (!is_fstree(key.objectid))
12084 key.offset = (u64)-1;
12086 cur_root = btrfs_read_fs_root(fs_info, &key);
12087 if (IS_ERR(cur_root) || !cur_root) {
12088 fprintf(stderr, "Fail to read fs/subvol tree: %lld\n",
12092 ret = fill_csum_tree_from_one_fs_root(trans, csum_root,
12097 ret = btrfs_next_item(tree_root, &path);
12107 btrfs_release_path(&path);
12111 static int fill_csum_tree_from_extent(struct btrfs_trans_handle *trans,
12112 struct btrfs_root *csum_root)
12114 struct btrfs_root *extent_root = csum_root->fs_info->extent_root;
12115 struct btrfs_path path;
12116 struct btrfs_extent_item *ei;
12117 struct extent_buffer *leaf;
12119 struct btrfs_key key;
12122 btrfs_init_path(&path);
12124 key.type = BTRFS_EXTENT_ITEM_KEY;
12126 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
12128 btrfs_release_path(&path);
12132 buf = malloc(csum_root->sectorsize);
12134 btrfs_release_path(&path);
12139 if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
12140 ret = btrfs_next_leaf(extent_root, &path);
12148 leaf = path.nodes[0];
12150 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
12151 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
12156 ei = btrfs_item_ptr(leaf, path.slots[0],
12157 struct btrfs_extent_item);
12158 if (!(btrfs_extent_flags(leaf, ei) &
12159 BTRFS_EXTENT_FLAG_DATA)) {
12164 ret = populate_csum(trans, csum_root, buf, key.objectid,
12171 btrfs_release_path(&path);
12177 * Recalculate the csum and put it into the csum tree.
12179 * Extent tree init will wipe out all the extent info, so in that case, we
12180 * can't depend on extent tree, but use fs tree. If search_fs_tree is set, we
12181 * will use fs/subvol trees to init the csum tree.
12183 static int fill_csum_tree(struct btrfs_trans_handle *trans,
12184 struct btrfs_root *csum_root,
12185 int search_fs_tree)
12187 if (search_fs_tree)
12188 return fill_csum_tree_from_fs(trans, csum_root);
12190 return fill_csum_tree_from_extent(trans, csum_root);
12193 static void free_roots_info_cache(void)
12195 if (!roots_info_cache)
12198 while (!cache_tree_empty(roots_info_cache)) {
12199 struct cache_extent *entry;
12200 struct root_item_info *rii;
12202 entry = first_cache_extent(roots_info_cache);
12205 remove_cache_extent(roots_info_cache, entry);
12206 rii = container_of(entry, struct root_item_info, cache_extent);
12210 free(roots_info_cache);
12211 roots_info_cache = NULL;
12214 static int build_roots_info_cache(struct btrfs_fs_info *info)
12217 struct btrfs_key key;
12218 struct extent_buffer *leaf;
12219 struct btrfs_path path;
12221 if (!roots_info_cache) {
12222 roots_info_cache = malloc(sizeof(*roots_info_cache));
12223 if (!roots_info_cache)
12225 cache_tree_init(roots_info_cache);
12228 btrfs_init_path(&path);
12230 key.type = BTRFS_EXTENT_ITEM_KEY;
12232 ret = btrfs_search_slot(NULL, info->extent_root, &key, &path, 0, 0);
12235 leaf = path.nodes[0];
12238 struct btrfs_key found_key;
12239 struct btrfs_extent_item *ei;
12240 struct btrfs_extent_inline_ref *iref;
12241 int slot = path.slots[0];
12246 struct cache_extent *entry;
12247 struct root_item_info *rii;
12249 if (slot >= btrfs_header_nritems(leaf)) {
12250 ret = btrfs_next_leaf(info->extent_root, &path);
12257 leaf = path.nodes[0];
12258 slot = path.slots[0];
12261 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
12263 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
12264 found_key.type != BTRFS_METADATA_ITEM_KEY)
12267 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
12268 flags = btrfs_extent_flags(leaf, ei);
12270 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
12271 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
12274 if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
12275 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
12276 level = found_key.offset;
12278 struct btrfs_tree_block_info *binfo;
12280 binfo = (struct btrfs_tree_block_info *)(ei + 1);
12281 iref = (struct btrfs_extent_inline_ref *)(binfo + 1);
12282 level = btrfs_tree_block_level(leaf, binfo);
12286 * For a root extent, it must be of the following type and the
12287 * first (and only one) iref in the item.
12289 type = btrfs_extent_inline_ref_type(leaf, iref);
12290 if (type != BTRFS_TREE_BLOCK_REF_KEY)
12293 root_id = btrfs_extent_inline_ref_offset(leaf, iref);
12294 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
12296 rii = malloc(sizeof(struct root_item_info));
12301 rii->cache_extent.start = root_id;
12302 rii->cache_extent.size = 1;
12303 rii->level = (u8)-1;
12304 entry = &rii->cache_extent;
12305 ret = insert_cache_extent(roots_info_cache, entry);
12308 rii = container_of(entry, struct root_item_info,
12312 ASSERT(rii->cache_extent.start == root_id);
12313 ASSERT(rii->cache_extent.size == 1);
12315 if (level > rii->level || rii->level == (u8)-1) {
12316 rii->level = level;
12317 rii->bytenr = found_key.objectid;
12318 rii->gen = btrfs_extent_generation(leaf, ei);
12319 rii->node_count = 1;
12320 } else if (level == rii->level) {
12328 btrfs_release_path(&path);
12333 static int maybe_repair_root_item(struct btrfs_fs_info *info,
12334 struct btrfs_path *path,
12335 const struct btrfs_key *root_key,
12336 const int read_only_mode)
12338 const u64 root_id = root_key->objectid;
12339 struct cache_extent *entry;
12340 struct root_item_info *rii;
12341 struct btrfs_root_item ri;
12342 unsigned long offset;
12344 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
12347 "Error: could not find extent items for root %llu\n",
12348 root_key->objectid);
12352 rii = container_of(entry, struct root_item_info, cache_extent);
12353 ASSERT(rii->cache_extent.start == root_id);
12354 ASSERT(rii->cache_extent.size == 1);
12356 if (rii->node_count != 1) {
12358 "Error: could not find btree root extent for root %llu\n",
12363 offset = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
12364 read_extent_buffer(path->nodes[0], &ri, offset, sizeof(ri));
12366 if (btrfs_root_bytenr(&ri) != rii->bytenr ||
12367 btrfs_root_level(&ri) != rii->level ||
12368 btrfs_root_generation(&ri) != rii->gen) {
12371 * If we're in repair mode but our caller told us to not update
12372 * the root item, i.e. just check if it needs to be updated, don't
12373 * print this message, since the caller will call us again shortly
12374 * for the same root item without read only mode (the caller will
12375 * open a transaction first).
12377 if (!(read_only_mode && repair))
12379 "%sroot item for root %llu,"
12380 " current bytenr %llu, current gen %llu, current level %u,"
12381 " new bytenr %llu, new gen %llu, new level %u\n",
12382 (read_only_mode ? "" : "fixing "),
12384 btrfs_root_bytenr(&ri), btrfs_root_generation(&ri),
12385 btrfs_root_level(&ri),
12386 rii->bytenr, rii->gen, rii->level);
12388 if (btrfs_root_generation(&ri) > rii->gen) {
12390 "root %llu has a root item with a more recent gen (%llu) compared to the found root node (%llu)\n",
12391 root_id, btrfs_root_generation(&ri), rii->gen);
12395 if (!read_only_mode) {
12396 btrfs_set_root_bytenr(&ri, rii->bytenr);
12397 btrfs_set_root_level(&ri, rii->level);
12398 btrfs_set_root_generation(&ri, rii->gen);
12399 write_extent_buffer(path->nodes[0], &ri,
12400 offset, sizeof(ri));
12410 * A regression introduced in the 3.17 kernel (more specifically in 3.17-rc2),
12411 * caused read-only snapshots to be corrupted if they were created at a moment
12412 * when the source subvolume/snapshot had orphan items. The issue was that the
12413 * on-disk root items became incorrect, referring to the pre orphan cleanup root
12414 * node instead of the post orphan cleanup root node.
12415 * So this function, and its callees, just detects and fixes those cases. Even
12416 * though the regression was for read-only snapshots, this function applies to
12417 * any snapshot/subvolume root.
12418 * This must be run before any other repair code - not doing it so, makes other
12419 * repair code delete or modify backrefs in the extent tree for example, which
12420 * will result in an inconsistent fs after repairing the root items.
12422 static int repair_root_items(struct btrfs_fs_info *info)
12424 struct btrfs_path path;
12425 struct btrfs_key key;
12426 struct extent_buffer *leaf;
12427 struct btrfs_trans_handle *trans = NULL;
12430 int need_trans = 0;
12432 btrfs_init_path(&path);
12434 ret = build_roots_info_cache(info);
12438 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
12439 key.type = BTRFS_ROOT_ITEM_KEY;
12444 * Avoid opening and committing transactions if a leaf doesn't have
12445 * any root items that need to be fixed, so that we avoid rotating
12446 * backup roots unnecessarily.
12449 trans = btrfs_start_transaction(info->tree_root, 1);
12450 if (IS_ERR(trans)) {
12451 ret = PTR_ERR(trans);
12456 ret = btrfs_search_slot(trans, info->tree_root, &key, &path,
12460 leaf = path.nodes[0];
12463 struct btrfs_key found_key;
12465 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
12466 int no_more_keys = find_next_key(&path, &key);
12468 btrfs_release_path(&path);
12470 ret = btrfs_commit_transaction(trans,
12482 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
12484 if (found_key.type != BTRFS_ROOT_ITEM_KEY)
12486 if (found_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
12489 ret = maybe_repair_root_item(info, &path, &found_key,
12494 if (!trans && repair) {
12497 btrfs_release_path(&path);
12507 free_roots_info_cache();
12508 btrfs_release_path(&path);
12510 btrfs_commit_transaction(trans, info->tree_root);
12517 static int clear_free_space_cache(struct btrfs_fs_info *fs_info)
12519 struct btrfs_trans_handle *trans;
12520 struct btrfs_block_group_cache *bg_cache;
12524 /* Clear all free space cache inodes and its extent data */
12526 bg_cache = btrfs_lookup_first_block_group(fs_info, current);
12529 ret = btrfs_clear_free_space_cache(fs_info, bg_cache);
12532 current = bg_cache->key.objectid + bg_cache->key.offset;
12535 /* Don't forget to set cache_generation to -1 */
12536 trans = btrfs_start_transaction(fs_info->tree_root, 0);
12537 if (IS_ERR(trans)) {
12538 error("failed to update super block cache generation");
12539 return PTR_ERR(trans);
12541 btrfs_set_super_cache_generation(fs_info->super_copy, (u64)-1);
12542 btrfs_commit_transaction(trans, fs_info->tree_root);
12547 const char * const cmd_check_usage[] = {
12548 "btrfs check [options] <device>",
12549 "Check structural integrity of a filesystem (unmounted).",
12550 "Check structural integrity of an unmounted filesystem. Verify internal",
12551 "trees' consistency and item connectivity. In the repair mode try to",
12552 "fix the problems found. ",
12553 "WARNING: the repair mode is considered dangerous",
12555 "-s|--super <superblock> use this superblock copy",
12556 "-b|--backup use the first valid backup root copy",
12557 "--repair try to repair the filesystem",
12558 "--readonly run in read-only mode (default)",
12559 "--init-csum-tree create a new CRC tree",
12560 "--init-extent-tree create a new extent tree",
12561 "--mode <MODE> allows choice of memory/IO trade-offs",
12562 " where MODE is one of:",
12563 " original - read inodes and extents to memory (requires",
12564 " more memory, does less IO)",
12565 " lowmem - try to use less memory but read blocks again",
12567 "--check-data-csum verify checksums of data blocks",
12568 "-Q|--qgroup-report print a report on qgroup consistency",
12569 "-E|--subvol-extents <subvolid>",
12570 " print subvolume extents and sharing state",
12571 "-r|--tree-root <bytenr> use the given bytenr for the tree root",
12572 "--chunk-root <bytenr> use the given bytenr for the chunk tree root",
12573 "-p|--progress indicate progress",
12574 "--clear-space-cache v1|v2 clear space cache for v1 or v2",
12578 int cmd_check(int argc, char **argv)
12580 struct cache_tree root_cache;
12581 struct btrfs_root *root;
12582 struct btrfs_fs_info *info;
12585 u64 tree_root_bytenr = 0;
12586 u64 chunk_root_bytenr = 0;
12587 char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
12591 int init_csum_tree = 0;
12593 int clear_space_cache = 0;
12594 int qgroup_report = 0;
12595 int qgroups_repaired = 0;
12596 unsigned ctree_flags = OPEN_CTREE_EXCLUSIVE;
12600 enum { GETOPT_VAL_REPAIR = 257, GETOPT_VAL_INIT_CSUM,
12601 GETOPT_VAL_INIT_EXTENT, GETOPT_VAL_CHECK_CSUM,
12602 GETOPT_VAL_READONLY, GETOPT_VAL_CHUNK_TREE,
12603 GETOPT_VAL_MODE, GETOPT_VAL_CLEAR_SPACE_CACHE };
12604 static const struct option long_options[] = {
12605 { "super", required_argument, NULL, 's' },
12606 { "repair", no_argument, NULL, GETOPT_VAL_REPAIR },
12607 { "readonly", no_argument, NULL, GETOPT_VAL_READONLY },
12608 { "init-csum-tree", no_argument, NULL,
12609 GETOPT_VAL_INIT_CSUM },
12610 { "init-extent-tree", no_argument, NULL,
12611 GETOPT_VAL_INIT_EXTENT },
12612 { "check-data-csum", no_argument, NULL,
12613 GETOPT_VAL_CHECK_CSUM },
12614 { "backup", no_argument, NULL, 'b' },
12615 { "subvol-extents", required_argument, NULL, 'E' },
12616 { "qgroup-report", no_argument, NULL, 'Q' },
12617 { "tree-root", required_argument, NULL, 'r' },
12618 { "chunk-root", required_argument, NULL,
12619 GETOPT_VAL_CHUNK_TREE },
12620 { "progress", no_argument, NULL, 'p' },
12621 { "mode", required_argument, NULL,
12623 { "clear-space-cache", required_argument, NULL,
12624 GETOPT_VAL_CLEAR_SPACE_CACHE},
12625 { NULL, 0, NULL, 0}
12628 c = getopt_long(argc, argv, "as:br:p", long_options, NULL);
12632 case 'a': /* ignored */ break;
12634 ctree_flags |= OPEN_CTREE_BACKUP_ROOT;
12637 num = arg_strtou64(optarg);
12638 if (num >= BTRFS_SUPER_MIRROR_MAX) {
12640 "super mirror should be less than %d",
12641 BTRFS_SUPER_MIRROR_MAX);
12644 bytenr = btrfs_sb_offset(((int)num));
12645 printf("using SB copy %llu, bytenr %llu\n", num,
12646 (unsigned long long)bytenr);
12652 subvolid = arg_strtou64(optarg);
12655 tree_root_bytenr = arg_strtou64(optarg);
12657 case GETOPT_VAL_CHUNK_TREE:
12658 chunk_root_bytenr = arg_strtou64(optarg);
12661 ctx.progress_enabled = true;
12665 usage(cmd_check_usage);
12666 case GETOPT_VAL_REPAIR:
12667 printf("enabling repair mode\n");
12669 ctree_flags |= OPEN_CTREE_WRITES;
12671 case GETOPT_VAL_READONLY:
12674 case GETOPT_VAL_INIT_CSUM:
12675 printf("Creating a new CRC tree\n");
12676 init_csum_tree = 1;
12678 ctree_flags |= OPEN_CTREE_WRITES;
12680 case GETOPT_VAL_INIT_EXTENT:
12681 init_extent_tree = 1;
12682 ctree_flags |= (OPEN_CTREE_WRITES |
12683 OPEN_CTREE_NO_BLOCK_GROUPS);
12686 case GETOPT_VAL_CHECK_CSUM:
12687 check_data_csum = 1;
12689 case GETOPT_VAL_MODE:
12690 check_mode = parse_check_mode(optarg);
12691 if (check_mode == CHECK_MODE_UNKNOWN) {
12692 error("unknown mode: %s", optarg);
12696 case GETOPT_VAL_CLEAR_SPACE_CACHE:
12697 if (strcmp(optarg, "v1") == 0) {
12698 clear_space_cache = 1;
12699 } else if (strcmp(optarg, "v2") == 0) {
12700 clear_space_cache = 2;
12701 ctree_flags |= OPEN_CTREE_INVALIDATE_FST;
12704 "invalid argument to --clear-space-cache, must be v1 or v2");
12707 ctree_flags |= OPEN_CTREE_WRITES;
12712 if (check_argc_exact(argc - optind, 1))
12713 usage(cmd_check_usage);
12715 if (ctx.progress_enabled) {
12716 ctx.tp = TASK_NOTHING;
12717 ctx.info = task_init(print_status_check, print_status_return, &ctx);
12720 /* This check is the only reason for --readonly to exist */
12721 if (readonly && repair) {
12722 error("repair options are not compatible with --readonly");
12727 * Not supported yet
12729 if (repair && check_mode == CHECK_MODE_LOWMEM) {
12730 error("low memory mode doesn't support repair yet");
12735 cache_tree_init(&root_cache);
12737 if((ret = check_mounted(argv[optind])) < 0) {
12738 error("could not check mount status: %s", strerror(-ret));
12742 error("%s is currently mounted, aborting", argv[optind]);
12748 /* only allow partial opening under repair mode */
12750 ctree_flags |= OPEN_CTREE_PARTIAL;
12752 info = open_ctree_fs_info(argv[optind], bytenr, tree_root_bytenr,
12753 chunk_root_bytenr, ctree_flags);
12755 error("cannot open file system");
12761 global_info = info;
12762 root = info->fs_root;
12763 if (clear_space_cache == 1) {
12764 if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE)) {
12766 "free space cache v2 detected, use --clear-space-cache v2");
12770 printf("Clearing free space cache\n");
12771 ret = clear_free_space_cache(info);
12773 error("failed to clear free space cache");
12776 printf("Free space cache cleared\n");
12779 } else if (clear_space_cache == 2) {
12780 if (!btrfs_fs_compat_ro(info, FREE_SPACE_TREE)) {
12781 printf("no free space cache v2 to clear\n");
12785 printf("Clear free space cache v2\n");
12786 ret = btrfs_clear_free_space_tree(info);
12788 error("failed to clear free space cache v2: %d", ret);
12791 printf("free space cache v2 cleared\n");
12797 * repair mode will force us to commit transaction which
12798 * will make us fail to load log tree when mounting.
12800 if (repair && btrfs_super_log_root(info->super_copy)) {
12801 ret = ask_user("repair mode will force to clear out log tree, are you sure?");
12807 ret = zero_log_tree(root);
12810 error("failed to zero log tree: %d", ret);
12815 uuid_unparse(info->super_copy->fsid, uuidbuf);
12816 if (qgroup_report) {
12817 printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
12819 ret = qgroup_verify_all(info);
12826 printf("Print extent state for subvolume %llu on %s\nUUID: %s\n",
12827 subvolid, argv[optind], uuidbuf);
12828 ret = print_extent_state(info, subvolid);
12832 printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
12834 if (!extent_buffer_uptodate(info->tree_root->node) ||
12835 !extent_buffer_uptodate(info->dev_root->node) ||
12836 !extent_buffer_uptodate(info->chunk_root->node)) {
12837 error("critical roots corrupted, unable to check the filesystem");
12843 if (init_extent_tree || init_csum_tree) {
12844 struct btrfs_trans_handle *trans;
12846 trans = btrfs_start_transaction(info->extent_root, 0);
12847 if (IS_ERR(trans)) {
12848 error("error starting transaction");
12849 ret = PTR_ERR(trans);
12854 if (init_extent_tree) {
12855 printf("Creating a new extent tree\n");
12856 ret = reinit_extent_tree(trans, info);
12862 if (init_csum_tree) {
12863 printf("Reinitialize checksum tree\n");
12864 ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
12866 error("checksum tree initialization failed: %d",
12873 ret = fill_csum_tree(trans, info->csum_root,
12877 error("checksum tree refilling failed: %d", ret);
12882 * Ok now we commit and run the normal fsck, which will add
12883 * extent entries for all of the items it finds.
12885 ret = btrfs_commit_transaction(trans, info->extent_root);
12890 if (!extent_buffer_uptodate(info->extent_root->node)) {
12891 error("critical: extent_root, unable to check the filesystem");
12896 if (!extent_buffer_uptodate(info->csum_root->node)) {
12897 error("critical: csum_root, unable to check the filesystem");
12903 if (!ctx.progress_enabled)
12904 fprintf(stderr, "checking extents\n");
12905 if (check_mode == CHECK_MODE_LOWMEM)
12906 ret = check_chunks_and_extents_v2(root);
12908 ret = check_chunks_and_extents(root);
12912 "errors found in extent allocation tree or chunk allocation");
12914 ret = repair_root_items(info);
12919 fprintf(stderr, "Fixed %d roots.\n", ret);
12921 } else if (ret > 0) {
12923 "Found %d roots with an outdated root item.\n",
12926 "Please run a filesystem check with the option --repair to fix them.\n");
12932 if (!ctx.progress_enabled) {
12933 if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
12934 fprintf(stderr, "checking free space tree\n");
12936 fprintf(stderr, "checking free space cache\n");
12938 ret = check_space_cache(root);
12944 * We used to have to have these hole extents in between our real
12945 * extents so if we don't have this flag set we need to make sure there
12946 * are no gaps in the file extents for inodes, otherwise we can just
12947 * ignore it when this happens.
12949 no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES);
12950 if (!ctx.progress_enabled)
12951 fprintf(stderr, "checking fs roots\n");
12952 if (check_mode == CHECK_MODE_LOWMEM)
12953 ret = check_fs_roots_v2(root->fs_info);
12955 ret = check_fs_roots(root, &root_cache);
12960 fprintf(stderr, "checking csums\n");
12961 ret = check_csums(root);
12966 fprintf(stderr, "checking root refs\n");
12967 /* For low memory mode, check_fs_roots_v2 handles root refs */
12968 if (check_mode != CHECK_MODE_LOWMEM) {
12969 ret = check_root_refs(root, &root_cache);
12975 while (repair && !list_empty(&root->fs_info->recow_ebs)) {
12976 struct extent_buffer *eb;
12978 eb = list_first_entry(&root->fs_info->recow_ebs,
12979 struct extent_buffer, recow);
12980 list_del_init(&eb->recow);
12981 ret = recow_extent_buffer(root, eb);
12987 while (!list_empty(&delete_items)) {
12988 struct bad_item *bad;
12990 bad = list_first_entry(&delete_items, struct bad_item, list);
12991 list_del_init(&bad->list);
12993 ret = delete_bad_item(root, bad);
12999 if (info->quota_enabled) {
13000 fprintf(stderr, "checking quota groups\n");
13001 ret = qgroup_verify_all(info);
13006 ret = repair_qgroups(info, &qgroups_repaired);
13013 if (!list_empty(&root->fs_info->recow_ebs)) {
13014 error("transid errors in file system");
13019 if (found_old_backref) { /*
13020 * there was a disk format change when mixed
13021 * backref was in testing tree. The old format
13022 * existed about one week.
13024 printf("\n * Found old mixed backref format. "
13025 "The old format is not supported! *"
13026 "\n * Please mount the FS in readonly mode, "
13027 "backup data and re-format the FS. *\n\n");
13030 printf("found %llu bytes used err is %d\n",
13031 (unsigned long long)bytes_used, ret);
13032 printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
13033 printf("total tree bytes: %llu\n",
13034 (unsigned long long)total_btree_bytes);
13035 printf("total fs tree bytes: %llu\n",
13036 (unsigned long long)total_fs_tree_bytes);
13037 printf("total extent tree bytes: %llu\n",
13038 (unsigned long long)total_extent_tree_bytes);
13039 printf("btree space waste bytes: %llu\n",
13040 (unsigned long long)btree_space_waste);
13041 printf("file data blocks allocated: %llu\n referenced %llu\n",
13042 (unsigned long long)data_bytes_allocated,
13043 (unsigned long long)data_bytes_referenced);
13045 free_qgroup_counts();
13046 free_root_recs_tree(&root_cache);
13050 if (ctx.progress_enabled)
13051 task_deinit(ctx.info);