?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
btrfs-progs: fix wrong data ratio for raid56 in btrfs-file-usage
[btrfs-progs-unstable/devel.git] / cmds-check.c
blob9fc141014926595e19ffc22ce019dcf91ba5fa37
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
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.
7 *
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.
12 *
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.
17 */
18
19 #define _XOPEN_SOURCE 500
20 #define _GNU_SOURCE 1
21 #include <stdio.h>
22 #include <stdlib.h>
23 #include <unistd.h>
24 #include <fcntl.h>
25 #include <sys/types.h>
26 #include <sys/stat.h>
27 #include <unistd.h>
28 #include <getopt.h>
29 #include <uuid/uuid.h>
30 #include "ctree.h"
31 #include "volumes.h"
32 #include "repair.h"
33 #include "disk-io.h"
34 #include "print-tree.h"
35 #include "transaction.h"
36 #include "version.h"
37 #include "utils.h"
38 #include "commands.h"
39 #include "free-space-cache.h"
40 #include "btrfsck.h"
41 #include "qgroup-verify.h"
42 #include "rbtree-utils.h"
43 #include "backref.h"
44 #include "ulist.h"
45
46 static u64 bytes_used = 0;
47 static u64 total_csum_bytes = 0;
48 static u64 total_btree_bytes = 0;
49 static u64 total_fs_tree_bytes = 0;
50 static u64 total_extent_tree_bytes = 0;
51 static u64 btree_space_waste = 0;
52 static u64 data_bytes_allocated = 0;
53 static u64 data_bytes_referenced = 0;
54 static int found_old_backref = 0;
55 static LIST_HEAD(duplicate_extents);
56 static LIST_HEAD(delete_items);
57 static int repair = 0;
58 static int no_holes = 0;
59 static int init_extent_tree = 0;
60 static int check_data_csum = 0;
61
62 struct extent_backref {
63 struct list_head list;
64 unsigned int is_data:1;
65 unsigned int found_extent_tree:1;
66 unsigned int full_backref:1;
67 unsigned int found_ref:1;
68 unsigned int broken:1;
69 };
70
71 struct data_backref {
72 struct extent_backref node;
73 union {
74 u64 parent;
75 u64 root;
76 };
77 u64 owner;
78 u64 offset;
79 u64 disk_bytenr;
80 u64 bytes;
81 u64 ram_bytes;
82 u32 num_refs;
83 u32 found_ref;
84 };
85
86 struct tree_backref {
87 struct extent_backref node;
88 union {
89 u64 parent;
90 u64 root;
91 };
92 };
93
94 struct extent_record {
95 struct list_head backrefs;
96 struct list_head dups;
97 struct list_head list;
98 struct cache_extent cache;
99 struct btrfs_disk_key parent_key;
100 u64 start;
101 u64 max_size;
102 u64 nr;
103 u64 refs;
104 u64 extent_item_refs;
105 u64 generation;
106 u64 parent_generation;
107 u64 info_objectid;
108 u32 num_duplicates;
109 u8 info_level;
110 unsigned int found_rec:1;
111 unsigned int content_checked:1;
112 unsigned int owner_ref_checked:1;
113 unsigned int is_root:1;
114 unsigned int metadata:1;
115 };
116
117 struct inode_backref {
118 struct list_head list;
119 unsigned int found_dir_item:1;
120 unsigned int found_dir_index:1;
121 unsigned int found_inode_ref:1;
122 unsigned int filetype:8;
123 int errors;
124 unsigned int ref_type;
125 u64 dir;
126 u64 index;
127 u16 namelen;
128 char name[0];
129 };
130
131 struct dropping_root_item_record {
132 struct list_head list;
133 struct btrfs_root_item ri;
134 struct btrfs_key found_key;
135 };
136
137 #define REF_ERR_NO_DIR_ITEM (1 << 0)
138 #define REF_ERR_NO_DIR_INDEX (1 << 1)
139 #define REF_ERR_NO_INODE_REF (1 << 2)
140 #define REF_ERR_DUP_DIR_ITEM (1 << 3)
141 #define REF_ERR_DUP_DIR_INDEX (1 << 4)
142 #define REF_ERR_DUP_INODE_REF (1 << 5)
143 #define REF_ERR_INDEX_UNMATCH (1 << 6)
144 #define REF_ERR_FILETYPE_UNMATCH (1 << 7)
145 #define REF_ERR_NAME_TOO_LONG (1 << 8) // 100
146 #define REF_ERR_NO_ROOT_REF (1 << 9)
147 #define REF_ERR_NO_ROOT_BACKREF (1 << 10)
148 #define REF_ERR_DUP_ROOT_REF (1 << 11)
149 #define REF_ERR_DUP_ROOT_BACKREF (1 << 12)
150
151 struct inode_record {
152 struct list_head backrefs;
153 unsigned int checked:1;
154 unsigned int merging:1;
155 unsigned int found_inode_item:1;
156 unsigned int found_dir_item:1;
157 unsigned int found_file_extent:1;
158 unsigned int found_csum_item:1;
159 unsigned int some_csum_missing:1;
160 unsigned int nodatasum:1;
161 int errors;
162
163 u64 ino;
164 u32 nlink;
165 u32 imode;
166 u64 isize;
167 u64 nbytes;
168
169 u32 found_link;
170 u64 found_size;
171 u64 extent_start;
172 u64 extent_end;
173 u64 first_extent_gap;
174
175 u32 refs;
176 };
177
178 #define I_ERR_NO_INODE_ITEM (1 << 0)
179 #define I_ERR_NO_ORPHAN_ITEM (1 << 1)
180 #define I_ERR_DUP_INODE_ITEM (1 << 2)
181 #define I_ERR_DUP_DIR_INDEX (1 << 3)
182 #define I_ERR_ODD_DIR_ITEM (1 << 4)
183 #define I_ERR_ODD_FILE_EXTENT (1 << 5)
184 #define I_ERR_BAD_FILE_EXTENT (1 << 6)
185 #define I_ERR_FILE_EXTENT_OVERLAP (1 << 7)
186 #define I_ERR_FILE_EXTENT_DISCOUNT (1 << 8) // 100
187 #define I_ERR_DIR_ISIZE_WRONG (1 << 9)
188 #define I_ERR_FILE_NBYTES_WRONG (1 << 10) // 400
189 #define I_ERR_ODD_CSUM_ITEM (1 << 11)
190 #define I_ERR_SOME_CSUM_MISSING (1 << 12)
191 #define I_ERR_LINK_COUNT_WRONG (1 << 13)
192
193 struct root_backref {
194 struct list_head list;
195 unsigned int found_dir_item:1;
196 unsigned int found_dir_index:1;
197 unsigned int found_back_ref:1;
198 unsigned int found_forward_ref:1;
199 unsigned int reachable:1;
200 int errors;
201 u64 ref_root;
202 u64 dir;
203 u64 index;
204 u16 namelen;
205 char name[0];
206 };
207
208 struct root_record {
209 struct list_head backrefs;
210 struct cache_extent cache;
211 unsigned int found_root_item:1;
212 u64 objectid;
213 u32 found_ref;
214 };
215
216 struct ptr_node {
217 struct cache_extent cache;
218 void *data;
219 };
220
221 struct shared_node {
222 struct cache_extent cache;
223 struct cache_tree root_cache;
224 struct cache_tree inode_cache;
225 struct inode_record *current;
226 u32 refs;
227 };
228
229 struct block_info {
230 u64 start;
231 u32 size;
232 };
233
234 struct walk_control {
235 struct cache_tree shared;
236 struct shared_node *nodes[BTRFS_MAX_LEVEL];
237 int active_node;
238 int root_level;
239 };
240
241 struct bad_item {
242 struct btrfs_key key;
243 u64 root_id;
244 struct list_head list;
245 };
246
247 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info);
248
249 static void record_root_in_trans(struct btrfs_trans_handle *trans,
250 struct btrfs_root *root)
251 {
252 if (root->last_trans != trans->transid) {
253 root->track_dirty = 1;
254 root->last_trans = trans->transid;
255 root->commit_root = root->node;
256 extent_buffer_get(root->node);
257 }
258 }
259
260 static u8 imode_to_type(u32 imode)
261 {
262 #define S_SHIFT 12
263 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
264 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
265 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
266 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
267 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
268 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
269 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
270 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
271 };
272
273 return btrfs_type_by_mode[(imode & S_IFMT) >> S_SHIFT];
274 #undef S_SHIFT
275 }
276
277 static int device_record_compare(struct rb_node *node1, struct rb_node *node2)
278 {
279 struct device_record *rec1;
280 struct device_record *rec2;
281
282 rec1 = rb_entry(node1, struct device_record, node);
283 rec2 = rb_entry(node2, struct device_record, node);
284 if (rec1->devid > rec2->devid)
285 return -1;
286 else if (rec1->devid < rec2->devid)
287 return 1;
288 else
289 return 0;
290 }
291
292 static struct inode_record *clone_inode_rec(struct inode_record *orig_rec)
293 {
294 struct inode_record *rec;
295 struct inode_backref *backref;
296 struct inode_backref *orig;
297 size_t size;
298
299 rec = malloc(sizeof(*rec));
300 memcpy(rec, orig_rec, sizeof(*rec));
301 rec->refs = 1;
302 INIT_LIST_HEAD(&rec->backrefs);
303
304 list_for_each_entry(orig, &orig_rec->backrefs, list) {
305 size = sizeof(*orig) + orig->namelen + 1;
306 backref = malloc(size);
307 memcpy(backref, orig, size);
308 list_add_tail(&backref->list, &rec->backrefs);
309 }
310 return rec;
311 }
312
313 static void print_inode_error(struct btrfs_root *root, struct inode_record *rec)
314 {
315 u64 root_objectid = root->root_key.objectid;
316 int errors = rec->errors;
317
318 if (!errors)
319 return;
320 /* reloc root errors, we print its corresponding fs root objectid*/
321 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
322 root_objectid = root->root_key.offset;
323 fprintf(stderr, "reloc");
324 }
325 fprintf(stderr, "root %llu inode %llu errors %x",
326 (unsigned long long) root_objectid,
327 (unsigned long long) rec->ino, rec->errors);
328
329 if (errors & I_ERR_NO_INODE_ITEM)
330 fprintf(stderr, ", no inode item");
331 if (errors & I_ERR_NO_ORPHAN_ITEM)
332 fprintf(stderr, ", no orphan item");
333 if (errors & I_ERR_DUP_INODE_ITEM)
334 fprintf(stderr, ", dup inode item");
335 if (errors & I_ERR_DUP_DIR_INDEX)
336 fprintf(stderr, ", dup dir index");
337 if (errors & I_ERR_ODD_DIR_ITEM)
338 fprintf(stderr, ", odd dir item");
339 if (errors & I_ERR_ODD_FILE_EXTENT)
340 fprintf(stderr, ", odd file extent");
341 if (errors & I_ERR_BAD_FILE_EXTENT)
342 fprintf(stderr, ", bad file extent");
343 if (errors & I_ERR_FILE_EXTENT_OVERLAP)
344 fprintf(stderr, ", file extent overlap");
345 if (errors & I_ERR_FILE_EXTENT_DISCOUNT)
346 fprintf(stderr, ", file extent discount");
347 if (errors & I_ERR_DIR_ISIZE_WRONG)
348 fprintf(stderr, ", dir isize wrong");
349 if (errors & I_ERR_FILE_NBYTES_WRONG)
350 fprintf(stderr, ", nbytes wrong");
351 if (errors & I_ERR_ODD_CSUM_ITEM)
352 fprintf(stderr, ", odd csum item");
353 if (errors & I_ERR_SOME_CSUM_MISSING)
354 fprintf(stderr, ", some csum missing");
355 if (errors & I_ERR_LINK_COUNT_WRONG)
356 fprintf(stderr, ", link count wrong");
357 fprintf(stderr, "\n");
358 }
359
360 static void print_ref_error(int errors)
361 {
362 if (errors & REF_ERR_NO_DIR_ITEM)
363 fprintf(stderr, ", no dir item");
364 if (errors & REF_ERR_NO_DIR_INDEX)
365 fprintf(stderr, ", no dir index");
366 if (errors & REF_ERR_NO_INODE_REF)
367 fprintf(stderr, ", no inode ref");
368 if (errors & REF_ERR_DUP_DIR_ITEM)
369 fprintf(stderr, ", dup dir item");
370 if (errors & REF_ERR_DUP_DIR_INDEX)
371 fprintf(stderr, ", dup dir index");
372 if (errors & REF_ERR_DUP_INODE_REF)
373 fprintf(stderr, ", dup inode ref");
374 if (errors & REF_ERR_INDEX_UNMATCH)
375 fprintf(stderr, ", index unmatch");
376 if (errors & REF_ERR_FILETYPE_UNMATCH)
377 fprintf(stderr, ", filetype unmatch");
378 if (errors & REF_ERR_NAME_TOO_LONG)
379 fprintf(stderr, ", name too long");
380 if (errors & REF_ERR_NO_ROOT_REF)
381 fprintf(stderr, ", no root ref");
382 if (errors & REF_ERR_NO_ROOT_BACKREF)
383 fprintf(stderr, ", no root backref");
384 if (errors & REF_ERR_DUP_ROOT_REF)
385 fprintf(stderr, ", dup root ref");
386 if (errors & REF_ERR_DUP_ROOT_BACKREF)
387 fprintf(stderr, ", dup root backref");
388 fprintf(stderr, "\n");
389 }
390
391 static struct inode_record *get_inode_rec(struct cache_tree *inode_cache,
392 u64 ino, int mod)
393 {
394 struct ptr_node *node;
395 struct cache_extent *cache;
396 struct inode_record *rec = NULL;
397 int ret;
398
399 cache = lookup_cache_extent(inode_cache, ino, 1);
400 if (cache) {
401 node = container_of(cache, struct ptr_node, cache);
402 rec = node->data;
403 if (mod && rec->refs > 1) {
404 node->data = clone_inode_rec(rec);
405 rec->refs--;
406 rec = node->data;
407 }
408 } else if (mod) {
409 rec = calloc(1, sizeof(*rec));
410 rec->ino = ino;
411 rec->extent_start = (u64)-1;
412 rec->first_extent_gap = (u64)-1;
413 rec->refs = 1;
414 INIT_LIST_HEAD(&rec->backrefs);
415
416 node = malloc(sizeof(*node));
417 node->cache.start = ino;
418 node->cache.size = 1;
419 node->data = rec;
420
421 if (ino == BTRFS_FREE_INO_OBJECTID)
422 rec->found_link = 1;
423
424 ret = insert_cache_extent(inode_cache, &node->cache);
425 BUG_ON(ret);
426 }
427 return rec;
428 }
429
430 static void free_inode_rec(struct inode_record *rec)
431 {
432 struct inode_backref *backref;
433
434 if (--rec->refs > 0)
435 return;
436
437 while (!list_empty(&rec->backrefs)) {
438 backref = list_entry(rec->backrefs.next,
439 struct inode_backref, list);
440 list_del(&backref->list);
441 free(backref);
442 }
443 free(rec);
444 }
445
446 static int can_free_inode_rec(struct inode_record *rec)
447 {
448 if (!rec->errors && rec->checked && rec->found_inode_item &&
449 rec->nlink == rec->found_link && list_empty(&rec->backrefs))
450 return 1;
451 return 0;
452 }
453
454 static void maybe_free_inode_rec(struct cache_tree *inode_cache,
455 struct inode_record *rec)
456 {
457 struct cache_extent *cache;
458 struct inode_backref *tmp, *backref;
459 struct ptr_node *node;
460 unsigned char filetype;
461
462 if (!rec->found_inode_item)
463 return;
464
465 filetype = imode_to_type(rec->imode);
466 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
467 if (backref->found_dir_item && backref->found_dir_index) {
468 if (backref->filetype != filetype)
469 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
470 if (!backref->errors && backref->found_inode_ref) {
471 list_del(&backref->list);
472 free(backref);
473 }
474 }
475 }
476
477 if (!rec->checked || rec->merging)
478 return;
479
480 if (S_ISDIR(rec->imode)) {
481 if (rec->found_size != rec->isize)
482 rec->errors |= I_ERR_DIR_ISIZE_WRONG;
483 if (rec->found_file_extent)
484 rec->errors |= I_ERR_ODD_FILE_EXTENT;
485 } else if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
486 if (rec->found_dir_item)
487 rec->errors |= I_ERR_ODD_DIR_ITEM;
488 if (rec->found_size != rec->nbytes)
489 rec->errors |= I_ERR_FILE_NBYTES_WRONG;
490 if (rec->extent_start == (u64)-1 || rec->extent_start > 0)
491 rec->first_extent_gap = 0;
492 if (rec->nlink > 0 && !no_holes &&
493 (rec->extent_end < rec->isize ||
494 rec->first_extent_gap < rec->isize))
495 rec->errors |= I_ERR_FILE_EXTENT_DISCOUNT;
496 }
497
498 if (S_ISREG(rec->imode) || S_ISLNK(rec->imode)) {
499 if (rec->found_csum_item && rec->nodatasum)
500 rec->errors |= I_ERR_ODD_CSUM_ITEM;
501 if (rec->some_csum_missing && !rec->nodatasum)
502 rec->errors |= I_ERR_SOME_CSUM_MISSING;
503 }
504
505 BUG_ON(rec->refs != 1);
506 if (can_free_inode_rec(rec)) {
507 cache = lookup_cache_extent(inode_cache, rec->ino, 1);
508 node = container_of(cache, struct ptr_node, cache);
509 BUG_ON(node->data != rec);
510 remove_cache_extent(inode_cache, &node->cache);
511 free(node);
512 free_inode_rec(rec);
513 }
514 }
515
516 static int check_orphan_item(struct btrfs_root *root, u64 ino)
517 {
518 struct btrfs_path path;
519 struct btrfs_key key;
520 int ret;
521
522 key.objectid = BTRFS_ORPHAN_OBJECTID;
523 key.type = BTRFS_ORPHAN_ITEM_KEY;
524 key.offset = ino;
525
526 btrfs_init_path(&path);
527 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
528 btrfs_release_path(&path);
529 if (ret > 0)
530 ret = -ENOENT;
531 return ret;
532 }
533
534 static int process_inode_item(struct extent_buffer *eb,
535 int slot, struct btrfs_key *key,
536 struct shared_node *active_node)
537 {
538 struct inode_record *rec;
539 struct btrfs_inode_item *item;
540
541 rec = active_node->current;
542 BUG_ON(rec->ino != key->objectid || rec->refs > 1);
543 if (rec->found_inode_item) {
544 rec->errors |= I_ERR_DUP_INODE_ITEM;
545 return 1;
546 }
547 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
548 rec->nlink = btrfs_inode_nlink(eb, item);
549 rec->isize = btrfs_inode_size(eb, item);
550 rec->nbytes = btrfs_inode_nbytes(eb, item);
551 rec->imode = btrfs_inode_mode(eb, item);
552 if (btrfs_inode_flags(eb, item) & BTRFS_INODE_NODATASUM)
553 rec->nodatasum = 1;
554 rec->found_inode_item = 1;
555 if (rec->nlink == 0)
556 rec->errors |= I_ERR_NO_ORPHAN_ITEM;
557 maybe_free_inode_rec(&active_node->inode_cache, rec);
558 return 0;
559 }
560
561 static struct inode_backref *get_inode_backref(struct inode_record *rec,
562 const char *name,
563 int namelen, u64 dir)
564 {
565 struct inode_backref *backref;
566
567 list_for_each_entry(backref, &rec->backrefs, list) {
568 if (rec->ino == BTRFS_MULTIPLE_OBJECTIDS)
569 break;
570 if (backref->dir != dir || backref->namelen != namelen)
571 continue;
572 if (memcmp(name, backref->name, namelen))
573 continue;
574 return backref;
575 }
576
577 backref = malloc(sizeof(*backref) + namelen + 1);
578 memset(backref, 0, sizeof(*backref));
579 backref->dir = dir;
580 backref->namelen = namelen;
581 memcpy(backref->name, name, namelen);
582 backref->name[namelen] = '\0';
583 list_add_tail(&backref->list, &rec->backrefs);
584 return backref;
585 }
586
587 static int add_inode_backref(struct cache_tree *inode_cache,
588 u64 ino, u64 dir, u64 index,
589 const char *name, int namelen,
590 int filetype, int itemtype, int errors)
591 {
592 struct inode_record *rec;
593 struct inode_backref *backref;
594
595 rec = get_inode_rec(inode_cache, ino, 1);
596 backref = get_inode_backref(rec, name, namelen, dir);
597 if (errors)
598 backref->errors |= errors;
599 if (itemtype == BTRFS_DIR_INDEX_KEY) {
600 if (backref->found_dir_index)
601 backref->errors |= REF_ERR_DUP_DIR_INDEX;
602 if (backref->found_inode_ref && backref->index != index)
603 backref->errors |= REF_ERR_INDEX_UNMATCH;
604 if (backref->found_dir_item && backref->filetype != filetype)
605 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
606
607 backref->index = index;
608 backref->filetype = filetype;
609 backref->found_dir_index = 1;
610 } else if (itemtype == BTRFS_DIR_ITEM_KEY) {
611 rec->found_link++;
612 if (backref->found_dir_item)
613 backref->errors |= REF_ERR_DUP_DIR_ITEM;
614 if (backref->found_dir_index && backref->filetype != filetype)
615 backref->errors |= REF_ERR_FILETYPE_UNMATCH;
616
617 backref->filetype = filetype;
618 backref->found_dir_item = 1;
619 } else if ((itemtype == BTRFS_INODE_REF_KEY) ||
620 (itemtype == BTRFS_INODE_EXTREF_KEY)) {
621 if (backref->found_inode_ref)
622 backref->errors |= REF_ERR_DUP_INODE_REF;
623 if (backref->found_dir_index && backref->index != index)
624 backref->errors |= REF_ERR_INDEX_UNMATCH;
625 else
626 backref->index = index;
627
628 backref->ref_type = itemtype;
629 backref->found_inode_ref = 1;
630 } else {
631 BUG_ON(1);
632 }
633
634 maybe_free_inode_rec(inode_cache, rec);
635 return 0;
636 }
637
638 static int merge_inode_recs(struct inode_record *src, struct inode_record *dst,
639 struct cache_tree *dst_cache)
640 {
641 struct inode_backref *backref;
642 u32 dir_count = 0;
643
644 dst->merging = 1;
645 list_for_each_entry(backref, &src->backrefs, list) {
646 if (backref->found_dir_index) {
647 add_inode_backref(dst_cache, dst->ino, backref->dir,
648 backref->index, backref->name,
649 backref->namelen, backref->filetype,
650 BTRFS_DIR_INDEX_KEY, backref->errors);
651 }
652 if (backref->found_dir_item) {
653 dir_count++;
654 add_inode_backref(dst_cache, dst->ino,
655 backref->dir, 0, backref->name,
656 backref->namelen, backref->filetype,
657 BTRFS_DIR_ITEM_KEY, backref->errors);
658 }
659 if (backref->found_inode_ref) {
660 add_inode_backref(dst_cache, dst->ino,
661 backref->dir, backref->index,
662 backref->name, backref->namelen, 0,
663 backref->ref_type, backref->errors);
664 }
665 }
666
667 if (src->found_dir_item)
668 dst->found_dir_item = 1;
669 if (src->found_file_extent)
670 dst->found_file_extent = 1;
671 if (src->found_csum_item)
672 dst->found_csum_item = 1;
673 if (src->some_csum_missing)
674 dst->some_csum_missing = 1;
675 if (dst->first_extent_gap > src->first_extent_gap)
676 dst->first_extent_gap = src->first_extent_gap;
677
678 BUG_ON(src->found_link < dir_count);
679 dst->found_link += src->found_link - dir_count;
680 dst->found_size += src->found_size;
681 if (src->extent_start != (u64)-1) {
682 if (dst->extent_start == (u64)-1) {
683 dst->extent_start = src->extent_start;
684 dst->extent_end = src->extent_end;
685 } else {
686 if (dst->extent_end > src->extent_start)
687 dst->errors |= I_ERR_FILE_EXTENT_OVERLAP;
688 else if (dst->extent_end < src->extent_start &&
689 dst->extent_end < dst->first_extent_gap)
690 dst->first_extent_gap = dst->extent_end;
691 if (dst->extent_end < src->extent_end)
692 dst->extent_end = src->extent_end;
693 }
694 }
695
696 dst->errors |= src->errors;
697 if (src->found_inode_item) {
698 if (!dst->found_inode_item) {
699 dst->nlink = src->nlink;
700 dst->isize = src->isize;
701 dst->nbytes = src->nbytes;
702 dst->imode = src->imode;
703 dst->nodatasum = src->nodatasum;
704 dst->found_inode_item = 1;
705 } else {
706 dst->errors |= I_ERR_DUP_INODE_ITEM;
707 }
708 }
709 dst->merging = 0;
710
711 return 0;
712 }
713
714 static int splice_shared_node(struct shared_node *src_node,
715 struct shared_node *dst_node)
716 {
717 struct cache_extent *cache;
718 struct ptr_node *node, *ins;
719 struct cache_tree *src, *dst;
720 struct inode_record *rec, *conflict;
721 u64 current_ino = 0;
722 int splice = 0;
723 int ret;
724
725 if (--src_node->refs == 0)
726 splice = 1;
727 if (src_node->current)
728 current_ino = src_node->current->ino;
729
730 src = &src_node->root_cache;
731 dst = &dst_node->root_cache;
732 again:
733 cache = search_cache_extent(src, 0);
734 while (cache) {
735 node = container_of(cache, struct ptr_node, cache);
736 rec = node->data;
737 cache = next_cache_extent(cache);
738
739 if (splice) {
740 remove_cache_extent(src, &node->cache);
741 ins = node;
742 } else {
743 ins = malloc(sizeof(*ins));
744 ins->cache.start = node->cache.start;
745 ins->cache.size = node->cache.size;
746 ins->data = rec;
747 rec->refs++;
748 }
749 ret = insert_cache_extent(dst, &ins->cache);
750 if (ret == -EEXIST) {
751 conflict = get_inode_rec(dst, rec->ino, 1);
752 merge_inode_recs(rec, conflict, dst);
753 if (rec->checked) {
754 conflict->checked = 1;
755 if (dst_node->current == conflict)
756 dst_node->current = NULL;
757 }
758 maybe_free_inode_rec(dst, conflict);
759 free_inode_rec(rec);
760 free(ins);
761 } else {
762 BUG_ON(ret);
763 }
764 }
765
766 if (src == &src_node->root_cache) {
767 src = &src_node->inode_cache;
768 dst = &dst_node->inode_cache;
769 goto again;
770 }
771
772 if (current_ino > 0 && (!dst_node->current ||
773 current_ino > dst_node->current->ino)) {
774 if (dst_node->current) {
775 dst_node->current->checked = 1;
776 maybe_free_inode_rec(dst, dst_node->current);
777 }
778 dst_node->current = get_inode_rec(dst, current_ino, 1);
779 }
780 return 0;
781 }
782
783 static void free_inode_ptr(struct cache_extent *cache)
784 {
785 struct ptr_node *node;
786 struct inode_record *rec;
787
788 node = container_of(cache, struct ptr_node, cache);
789 rec = node->data;
790 free_inode_rec(rec);
791 free(node);
792 }
793
794 FREE_EXTENT_CACHE_BASED_TREE(inode_recs, free_inode_ptr);
795
796 static struct shared_node *find_shared_node(struct cache_tree *shared,
797 u64 bytenr)
798 {
799 struct cache_extent *cache;
800 struct shared_node *node;
801
802 cache = lookup_cache_extent(shared, bytenr, 1);
803 if (cache) {
804 node = container_of(cache, struct shared_node, cache);
805 return node;
806 }
807 return NULL;
808 }
809
810 static int add_shared_node(struct cache_tree *shared, u64 bytenr, u32 refs)
811 {
812 int ret;
813 struct shared_node *node;
814
815 node = calloc(1, sizeof(*node));
816 node->cache.start = bytenr;
817 node->cache.size = 1;
818 cache_tree_init(&node->root_cache);
819 cache_tree_init(&node->inode_cache);
820 node->refs = refs;
821
822 ret = insert_cache_extent(shared, &node->cache);
823 BUG_ON(ret);
824 return 0;
825 }
826
827 static int enter_shared_node(struct btrfs_root *root, u64 bytenr, u32 refs,
828 struct walk_control *wc, int level)
829 {
830 struct shared_node *node;
831 struct shared_node *dest;
832
833 if (level == wc->active_node)
834 return 0;
835
836 BUG_ON(wc->active_node <= level);
837 node = find_shared_node(&wc->shared, bytenr);
838 if (!node) {
839 add_shared_node(&wc->shared, bytenr, refs);
840 node = find_shared_node(&wc->shared, bytenr);
841 wc->nodes[level] = node;
842 wc->active_node = level;
843 return 0;
844 }
845
846 if (wc->root_level == wc->active_node &&
847 btrfs_root_refs(&root->root_item) == 0) {
848 if (--node->refs == 0) {
849 free_inode_recs_tree(&node->root_cache);
850 free_inode_recs_tree(&node->inode_cache);
851 remove_cache_extent(&wc->shared, &node->cache);
852 free(node);
853 }
854 return 1;
855 }
856
857 dest = wc->nodes[wc->active_node];
858 splice_shared_node(node, dest);
859 if (node->refs == 0) {
860 remove_cache_extent(&wc->shared, &node->cache);
861 free(node);
862 }
863 return 1;
864 }
865
866 static int leave_shared_node(struct btrfs_root *root,
867 struct walk_control *wc, int level)
868 {
869 struct shared_node *node;
870 struct shared_node *dest;
871 int i;
872
873 if (level == wc->root_level)
874 return 0;
875
876 for (i = level + 1; i < BTRFS_MAX_LEVEL; i++) {
877 if (wc->nodes[i])
878 break;
879 }
880 BUG_ON(i >= BTRFS_MAX_LEVEL);
881
882 node = wc->nodes[wc->active_node];
883 wc->nodes[wc->active_node] = NULL;
884 wc->active_node = i;
885
886 dest = wc->nodes[wc->active_node];
887 if (wc->active_node < wc->root_level ||
888 btrfs_root_refs(&root->root_item) > 0) {
889 BUG_ON(node->refs <= 1);
890 splice_shared_node(node, dest);
891 } else {
892 BUG_ON(node->refs < 2);
893 node->refs--;
894 }
895 return 0;
896 }
897
898 /*
899 * Returns:
900 * < 0 - on error
901 * 1 - if the root with id child_root_id is a child of root parent_root_id
902 * 0 - if the root child_root_id isn't a child of the root parent_root_id but
903 * has other root(s) as parent(s)
904 * 2 - if the root child_root_id doesn't have any parent roots
905 */
906 static int is_child_root(struct btrfs_root *root, u64 parent_root_id,
907 u64 child_root_id)
908 {
909 struct btrfs_path path;
910 struct btrfs_key key;
911 struct extent_buffer *leaf;
912 int has_parent = 0;
913 int ret;
914
915 btrfs_init_path(&path);
916
917 key.objectid = parent_root_id;
918 key.type = BTRFS_ROOT_REF_KEY;
919 key.offset = child_root_id;
920 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
921 0, 0);
922 if (ret < 0)
923 return ret;
924 btrfs_release_path(&path);
925 if (!ret)
926 return 1;
927
928 key.objectid = child_root_id;
929 key.type = BTRFS_ROOT_BACKREF_KEY;
930 key.offset = 0;
931 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path,
932 0, 0);
933 if (ret < 0)
934 goto out;
935
936 while (1) {
937 leaf = path.nodes[0];
938 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
939 ret = btrfs_next_leaf(root->fs_info->tree_root, &path);
940 if (ret)
941 break;
942 leaf = path.nodes[0];
943 }
944
945 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
946 if (key.objectid != child_root_id ||
947 key.type != BTRFS_ROOT_BACKREF_KEY)
948 break;
949
950 has_parent = 1;
951
952 if (key.offset == parent_root_id) {
953 btrfs_release_path(&path);
954 return 1;
955 }
956
957 path.slots[0]++;
958 }
959 out:
960 btrfs_release_path(&path);
961 if (ret < 0)
962 return ret;
963 return has_parent ? 0 : 2;
964 }
965
966 static int process_dir_item(struct btrfs_root *root,
967 struct extent_buffer *eb,
968 int slot, struct btrfs_key *key,
969 struct shared_node *active_node)
970 {
971 u32 total;
972 u32 cur = 0;
973 u32 len;
974 u32 name_len;
975 u32 data_len;
976 int error;
977 int nritems = 0;
978 int filetype;
979 struct btrfs_dir_item *di;
980 struct inode_record *rec;
981 struct cache_tree *root_cache;
982 struct cache_tree *inode_cache;
983 struct btrfs_key location;
984 char namebuf[BTRFS_NAME_LEN];
985
986 root_cache = &active_node->root_cache;
987 inode_cache = &active_node->inode_cache;
988 rec = active_node->current;
989 rec->found_dir_item = 1;
990
991 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
992 total = btrfs_item_size_nr(eb, slot);
993 while (cur < total) {
994 nritems++;
995 btrfs_dir_item_key_to_cpu(eb, di, &location);
996 name_len = btrfs_dir_name_len(eb, di);
997 data_len = btrfs_dir_data_len(eb, di);
998 filetype = btrfs_dir_type(eb, di);
999
1000 rec->found_size += name_len;
1001 if (name_len <= BTRFS_NAME_LEN) {
1002 len = name_len;
1003 error = 0;
1004 } else {
1005 len = BTRFS_NAME_LEN;
1006 error = REF_ERR_NAME_TOO_LONG;
1007 }
1008 read_extent_buffer(eb, namebuf, (unsigned long)(di + 1), len);
1009
1010 if (location.type == BTRFS_INODE_ITEM_KEY) {
1011 add_inode_backref(inode_cache, location.objectid,
1012 key->objectid, key->offset, namebuf,
1013 len, filetype, key->type, error);
1014 } else if (location.type == BTRFS_ROOT_ITEM_KEY) {
1015 add_inode_backref(root_cache, location.objectid,
1016 key->objectid, key->offset,
1017 namebuf, len, filetype,
1018 key->type, error);
1019 } else {
1020 fprintf(stderr, "invalid location in dir item %u\n",
1021 location.type);
1022 add_inode_backref(inode_cache, BTRFS_MULTIPLE_OBJECTIDS,
1023 key->objectid, key->offset, namebuf,
1024 len, filetype, key->type, error);
1025 }
1026
1027 len = sizeof(*di) + name_len + data_len;
1028 di = (struct btrfs_dir_item *)((char *)di + len);
1029 cur += len;
1030 }
1031 if (key->type == BTRFS_DIR_INDEX_KEY && nritems > 1)
1032 rec->errors |= I_ERR_DUP_DIR_INDEX;
1033
1034 return 0;
1035 }
1036
1037 static int process_inode_ref(struct extent_buffer *eb,
1038 int slot, struct btrfs_key *key,
1039 struct shared_node *active_node)
1040 {
1041 u32 total;
1042 u32 cur = 0;
1043 u32 len;
1044 u32 name_len;
1045 u64 index;
1046 int error;
1047 struct cache_tree *inode_cache;
1048 struct btrfs_inode_ref *ref;
1049 char namebuf[BTRFS_NAME_LEN];
1050
1051 inode_cache = &active_node->inode_cache;
1052
1053 ref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1054 total = btrfs_item_size_nr(eb, slot);
1055 while (cur < total) {
1056 name_len = btrfs_inode_ref_name_len(eb, ref);
1057 index = btrfs_inode_ref_index(eb, ref);
1058 if (name_len <= BTRFS_NAME_LEN) {
1059 len = name_len;
1060 error = 0;
1061 } else {
1062 len = BTRFS_NAME_LEN;
1063 error = REF_ERR_NAME_TOO_LONG;
1064 }
1065 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
1066 add_inode_backref(inode_cache, key->objectid, key->offset,
1067 index, namebuf, len, 0, key->type, error);
1068
1069 len = sizeof(*ref) + name_len;
1070 ref = (struct btrfs_inode_ref *)((char *)ref + len);
1071 cur += len;
1072 }
1073 return 0;
1074 }
1075
1076 static int process_inode_extref(struct extent_buffer *eb,
1077 int slot, struct btrfs_key *key,
1078 struct shared_node *active_node)
1079 {
1080 u32 total;
1081 u32 cur = 0;
1082 u32 len;
1083 u32 name_len;
1084 u64 index;
1085 u64 parent;
1086 int error;
1087 struct cache_tree *inode_cache;
1088 struct btrfs_inode_extref *extref;
1089 char namebuf[BTRFS_NAME_LEN];
1090
1091 inode_cache = &active_node->inode_cache;
1092
1093 extref = btrfs_item_ptr(eb, slot, struct btrfs_inode_extref);
1094 total = btrfs_item_size_nr(eb, slot);
1095 while (cur < total) {
1096 name_len = btrfs_inode_extref_name_len(eb, extref);
1097 index = btrfs_inode_extref_index(eb, extref);
1098 parent = btrfs_inode_extref_parent(eb, extref);
1099 if (name_len <= BTRFS_NAME_LEN) {
1100 len = name_len;
1101 error = 0;
1102 } else {
1103 len = BTRFS_NAME_LEN;
1104 error = REF_ERR_NAME_TOO_LONG;
1105 }
1106 read_extent_buffer(eb, namebuf,
1107 (unsigned long)(extref + 1), len);
1108 add_inode_backref(inode_cache, key->objectid, parent,
1109 index, namebuf, len, 0, key->type, error);
1110
1111 len = sizeof(*extref) + name_len;
1112 extref = (struct btrfs_inode_extref *)((char *)extref + len);
1113 cur += len;
1114 }
1115 return 0;
1116
1117 }
1118
1119 static int count_csum_range(struct btrfs_root *root, u64 start,
1120 u64 len, u64 *found)
1121 {
1122 struct btrfs_key key;
1123 struct btrfs_path path;
1124 struct extent_buffer *leaf;
1125 int ret;
1126 size_t size;
1127 *found = 0;
1128 u64 csum_end;
1129 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
1130
1131 btrfs_init_path(&path);
1132
1133 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
1134 key.offset = start;
1135 key.type = BTRFS_EXTENT_CSUM_KEY;
1136
1137 ret = btrfs_search_slot(NULL, root->fs_info->csum_root,
1138 &key, &path, 0, 0);
1139 if (ret < 0)
1140 goto out;
1141 if (ret > 0 && path.slots[0] > 0) {
1142 leaf = path.nodes[0];
1143 btrfs_item_key_to_cpu(leaf, &key, path.slots[0] - 1);
1144 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
1145 key.type == BTRFS_EXTENT_CSUM_KEY)
1146 path.slots[0]--;
1147 }
1148
1149 while (len > 0) {
1150 leaf = path.nodes[0];
1151 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
1152 ret = btrfs_next_leaf(root->fs_info->csum_root, &path);
1153 if (ret > 0)
1154 break;
1155 else if (ret < 0)
1156 goto out;
1157 leaf = path.nodes[0];
1158 }
1159
1160 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1161 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
1162 key.type != BTRFS_EXTENT_CSUM_KEY)
1163 break;
1164
1165 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
1166 if (key.offset >= start + len)
1167 break;
1168
1169 if (key.offset > start)
1170 start = key.offset;
1171
1172 size = btrfs_item_size_nr(leaf, path.slots[0]);
1173 csum_end = key.offset + (size / csum_size) * root->sectorsize;
1174 if (csum_end > start) {
1175 size = min(csum_end - start, len);
1176 len -= size;
1177 start += size;
1178 *found += size;
1179 }
1180
1181 path.slots[0]++;
1182 }
1183 out:
1184 if (ret < 0)
1185 return ret;
1186 btrfs_release_path(&path);
1187 return 0;
1188 }
1189
1190 static int process_file_extent(struct btrfs_root *root,
1191 struct extent_buffer *eb,
1192 int slot, struct btrfs_key *key,
1193 struct shared_node *active_node)
1194 {
1195 struct inode_record *rec;
1196 struct btrfs_file_extent_item *fi;
1197 u64 num_bytes = 0;
1198 u64 disk_bytenr = 0;
1199 u64 extent_offset = 0;
1200 u64 mask = root->sectorsize - 1;
1201 int extent_type;
1202 int ret;
1203
1204 rec = active_node->current;
1205 BUG_ON(rec->ino != key->objectid || rec->refs > 1);
1206 rec->found_file_extent = 1;
1207
1208 if (rec->extent_start == (u64)-1) {
1209 rec->extent_start = key->offset;
1210 rec->extent_end = key->offset;
1211 }
1212
1213 if (rec->extent_end > key->offset)
1214 rec->errors |= I_ERR_FILE_EXTENT_OVERLAP;
1215 else if (rec->extent_end < key->offset &&
1216 rec->extent_end < rec->first_extent_gap)
1217 rec->first_extent_gap = rec->extent_end;
1218
1219 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
1220 extent_type = btrfs_file_extent_type(eb, fi);
1221
1222 if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1223 num_bytes = btrfs_file_extent_inline_len(eb, slot, fi);
1224 if (num_bytes == 0)
1225 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1226 rec->found_size += num_bytes;
1227 num_bytes = (num_bytes + mask) & ~mask;
1228 } else if (extent_type == BTRFS_FILE_EXTENT_REG ||
1229 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1230 num_bytes = btrfs_file_extent_num_bytes(eb, fi);
1231 disk_bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1232 extent_offset = btrfs_file_extent_offset(eb, fi);
1233 if (num_bytes == 0 || (num_bytes & mask))
1234 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1235 if (num_bytes + extent_offset >
1236 btrfs_file_extent_ram_bytes(eb, fi))
1237 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1238 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC &&
1239 (btrfs_file_extent_compression(eb, fi) ||
1240 btrfs_file_extent_encryption(eb, fi) ||
1241 btrfs_file_extent_other_encoding(eb, fi)))
1242 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1243 if (disk_bytenr > 0)
1244 rec->found_size += num_bytes;
1245 } else {
1246 rec->errors |= I_ERR_BAD_FILE_EXTENT;
1247 }
1248 rec->extent_end = key->offset + num_bytes;
1249
1250 if (disk_bytenr > 0) {
1251 u64 found;
1252 if (btrfs_file_extent_compression(eb, fi))
1253 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1254 else
1255 disk_bytenr += extent_offset;
1256
1257 ret = count_csum_range(root, disk_bytenr, num_bytes, &found);
1258 if (ret < 0)
1259 return ret;
1260 if (extent_type == BTRFS_FILE_EXTENT_REG) {
1261 if (found > 0)
1262 rec->found_csum_item = 1;
1263 if (found < num_bytes)
1264 rec->some_csum_missing = 1;
1265 } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1266 if (found > 0)
1267 rec->errors |= I_ERR_ODD_CSUM_ITEM;
1268 }
1269 }
1270 return 0;
1271 }
1272
1273 static int process_one_leaf(struct btrfs_root *root, struct extent_buffer *eb,
1274 struct walk_control *wc)
1275 {
1276 struct btrfs_key key;
1277 u32 nritems;
1278 int i;
1279 int ret = 0;
1280 struct cache_tree *inode_cache;
1281 struct shared_node *active_node;
1282
1283 if (wc->root_level == wc->active_node &&
1284 btrfs_root_refs(&root->root_item) == 0)
1285 return 0;
1286
1287 active_node = wc->nodes[wc->active_node];
1288 inode_cache = &active_node->inode_cache;
1289 nritems = btrfs_header_nritems(eb);
1290 for (i = 0; i < nritems; i++) {
1291 btrfs_item_key_to_cpu(eb, &key, i);
1292
1293 if (key.objectid == BTRFS_FREE_SPACE_OBJECTID)
1294 continue;
1295 if (key.type == BTRFS_ORPHAN_ITEM_KEY)
1296 continue;
1297
1298 if (active_node->current == NULL ||
1299 active_node->current->ino < key.objectid) {
1300 if (active_node->current) {
1301 active_node->current->checked = 1;
1302 maybe_free_inode_rec(inode_cache,
1303 active_node->current);
1304 }
1305 active_node->current = get_inode_rec(inode_cache,
1306 key.objectid, 1);
1307 }
1308 switch (key.type) {
1309 case BTRFS_DIR_ITEM_KEY:
1310 case BTRFS_DIR_INDEX_KEY:
1311 ret = process_dir_item(root, eb, i, &key, active_node);
1312 break;
1313 case BTRFS_INODE_REF_KEY:
1314 ret = process_inode_ref(eb, i, &key, active_node);
1315 break;
1316 case BTRFS_INODE_EXTREF_KEY:
1317 ret = process_inode_extref(eb, i, &key, active_node);
1318 break;
1319 case BTRFS_INODE_ITEM_KEY:
1320 ret = process_inode_item(eb, i, &key, active_node);
1321 break;
1322 case BTRFS_EXTENT_DATA_KEY:
1323 ret = process_file_extent(root, eb, i, &key,
1324 active_node);
1325 break;
1326 default:
1327 break;
1328 };
1329 }
1330 return ret;
1331 }
1332
1333 static void reada_walk_down(struct btrfs_root *root,
1334 struct extent_buffer *node, int slot)
1335 {
1336 u64 bytenr;
1337 u64 ptr_gen;
1338 u32 nritems;
1339 u32 blocksize;
1340 int i;
1341 int level;
1342
1343 level = btrfs_header_level(node);
1344 if (level != 1)
1345 return;
1346
1347 nritems = btrfs_header_nritems(node);
1348 blocksize = btrfs_level_size(root, level - 1);
1349 for (i = slot; i < nritems; i++) {
1350 bytenr = btrfs_node_blockptr(node, i);
1351 ptr_gen = btrfs_node_ptr_generation(node, i);
1352 readahead_tree_block(root, bytenr, blocksize, ptr_gen);
1353 }
1354 }
1355
1356 /*
1357 * Check the child node/leaf by the following condition:
1358 * 1. the first item key of the node/leaf should be the same with the one
1359 * in parent.
1360 * 2. block in parent node should match the child node/leaf.
1361 * 3. generation of parent node and child's header should be consistent.
1362 *
1363 * Or the child node/leaf pointed by the key in parent is not valid.
1364 *
1365 * We hope to check leaf owner too, but since subvol may share leaves,
1366 * which makes leaf owner check not so strong, key check should be
1367 * sufficient enough for that case.
1368 */
1369 static int check_child_node(struct btrfs_root *root,
1370 struct extent_buffer *parent, int slot,
1371 struct extent_buffer *child)
1372 {
1373 struct btrfs_key parent_key;
1374 struct btrfs_key child_key;
1375 int ret = 0;
1376
1377 btrfs_node_key_to_cpu(parent, &parent_key, slot);
1378 if (btrfs_header_level(child) == 0)
1379 btrfs_item_key_to_cpu(child, &child_key, 0);
1380 else
1381 btrfs_node_key_to_cpu(child, &child_key, 0);
1382
1383 if (memcmp(&parent_key, &child_key, sizeof(parent_key))) {
1384 ret = -EINVAL;
1385 fprintf(stderr,
1386 "Wrong key of child node/leaf, wanted: (%llu, %u, %llu), have: (%llu, %u, %llu)\n",
1387 parent_key.objectid, parent_key.type, parent_key.offset,
1388 child_key.objectid, child_key.type, child_key.offset);
1389 }
1390 if (btrfs_header_bytenr(child) != btrfs_node_blockptr(parent, slot)) {
1391 ret = -EINVAL;
1392 fprintf(stderr, "Wrong block of child node/leaf, wanted: %llu, have: %llu\n",
1393 btrfs_node_blockptr(parent, slot),
1394 btrfs_header_bytenr(child));
1395 }
1396 if (btrfs_node_ptr_generation(parent, slot) !=
1397 btrfs_header_generation(child)) {
1398 ret = -EINVAL;
1399 fprintf(stderr, "Wrong generation of child node/leaf, wanted: %llu, have: %llu\n",
1400 btrfs_header_generation(child),
1401 btrfs_node_ptr_generation(parent, slot));
1402 }
1403 return ret;
1404 }
1405
1406 static int walk_down_tree(struct btrfs_root *root, struct btrfs_path *path,
1407 struct walk_control *wc, int *level)
1408 {
1409 enum btrfs_tree_block_status status;
1410 u64 bytenr;
1411 u64 ptr_gen;
1412 struct extent_buffer *next;
1413 struct extent_buffer *cur;
1414 u32 blocksize;
1415 int ret, err = 0;
1416 u64 refs;
1417
1418 WARN_ON(*level < 0);
1419 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1420 ret = btrfs_lookup_extent_info(NULL, root,
1421 path->nodes[*level]->start,
1422 *level, 1, &refs, NULL);
1423 if (ret < 0) {
1424 err = ret;
1425 goto out;
1426 }
1427
1428 if (refs > 1) {
1429 ret = enter_shared_node(root, path->nodes[*level]->start,
1430 refs, wc, *level);
1431 if (ret > 0) {
1432 err = ret;
1433 goto out;
1434 }
1435 }
1436
1437 while (*level >= 0) {
1438 WARN_ON(*level < 0);
1439 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1440 cur = path->nodes[*level];
1441
1442 if (btrfs_header_level(cur) != *level)
1443 WARN_ON(1);
1444
1445 if (path->slots[*level] >= btrfs_header_nritems(cur))
1446 break;
1447 if (*level == 0) {
1448 ret = process_one_leaf(root, cur, wc);
1449 if (ret < 0)
1450 err = ret;
1451 break;
1452 }
1453 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1454 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1455 blocksize = btrfs_level_size(root, *level - 1);
1456 ret = btrfs_lookup_extent_info(NULL, root, bytenr, *level - 1,
1457 1, &refs, NULL);
1458 if (ret < 0)
1459 refs = 0;
1460
1461 if (refs > 1) {
1462 ret = enter_shared_node(root, bytenr, refs,
1463 wc, *level - 1);
1464 if (ret > 0) {
1465 path->slots[*level]++;
1466 continue;
1467 }
1468 }
1469
1470 next = btrfs_find_tree_block(root, bytenr, blocksize);
1471 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
1472 free_extent_buffer(next);
1473 reada_walk_down(root, cur, path->slots[*level]);
1474 next = read_tree_block(root, bytenr, blocksize,
1475 ptr_gen);
1476 if (!next) {
1477 err = -EIO;
1478 goto out;
1479 }
1480 }
1481
1482 ret = check_child_node(root, cur, path->slots[*level], next);
1483 if (ret) {
1484 err = ret;
1485 goto out;
1486 }
1487
1488 if (btrfs_is_leaf(next))
1489 status = btrfs_check_leaf(root, NULL, next);
1490 else
1491 status = btrfs_check_node(root, NULL, next);
1492 if (status != BTRFS_TREE_BLOCK_CLEAN) {
1493 free_extent_buffer(next);
1494 err = -EIO;
1495 goto out;
1496 }
1497
1498 *level = *level - 1;
1499 free_extent_buffer(path->nodes[*level]);
1500 path->nodes[*level] = next;
1501 path->slots[*level] = 0;
1502 }
1503 out:
1504 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
1505 return err;
1506 }
1507
1508 static int walk_up_tree(struct btrfs_root *root, struct btrfs_path *path,
1509 struct walk_control *wc, int *level)
1510 {
1511 int i;
1512 struct extent_buffer *leaf;
1513
1514 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1515 leaf = path->nodes[i];
1516 if (path->slots[i] + 1 < btrfs_header_nritems(leaf)) {
1517 path->slots[i]++;
1518 *level = i;
1519 return 0;
1520 } else {
1521 free_extent_buffer(path->nodes[*level]);
1522 path->nodes[*level] = NULL;
1523 BUG_ON(*level > wc->active_node);
1524 if (*level == wc->active_node)
1525 leave_shared_node(root, wc, *level);
1526 *level = i + 1;
1527 }
1528 }
1529 return 1;
1530 }
1531
1532 static int check_root_dir(struct inode_record *rec)
1533 {
1534 struct inode_backref *backref;
1535 int ret = -1;
1536
1537 if (!rec->found_inode_item || rec->errors)
1538 goto out;
1539 if (rec->nlink != 1 || rec->found_link != 0)
1540 goto out;
1541 if (list_empty(&rec->backrefs))
1542 goto out;
1543 backref = list_entry(rec->backrefs.next, struct inode_backref, list);
1544 if (!backref->found_inode_ref)
1545 goto out;
1546 if (backref->index != 0 || backref->namelen != 2 ||
1547 memcmp(backref->name, "..", 2))
1548 goto out;
1549 if (backref->found_dir_index || backref->found_dir_item)
1550 goto out;
1551 ret = 0;
1552 out:
1553 return ret;
1554 }
1555
1556 static int repair_inode_isize(struct btrfs_trans_handle *trans,
1557 struct btrfs_root *root, struct btrfs_path *path,
1558 struct inode_record *rec)
1559 {
1560 struct btrfs_inode_item *ei;
1561 struct btrfs_key key;
1562 int ret;
1563
1564 key.objectid = rec->ino;
1565 key.type = BTRFS_INODE_ITEM_KEY;
1566 key.offset = (u64)-1;
1567
1568 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1569 if (ret < 0)
1570 goto out;
1571 if (ret) {
1572 if (!path->slots[0]) {
1573 ret = -ENOENT;
1574 goto out;
1575 }
1576 path->slots[0]--;
1577 ret = 0;
1578 }
1579 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1580 if (key.objectid != rec->ino) {
1581 ret = -ENOENT;
1582 goto out;
1583 }
1584
1585 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
1586 struct btrfs_inode_item);
1587 btrfs_set_inode_size(path->nodes[0], ei, rec->found_size);
1588 btrfs_mark_buffer_dirty(path->nodes[0]);
1589 rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
1590 printf("reset isize for dir %Lu root %Lu\n", rec->ino,
1591 root->root_key.objectid);
1592 out:
1593 btrfs_release_path(path);
1594 return ret;
1595 }
1596
1597 static int repair_inode_orphan_item(struct btrfs_trans_handle *trans,
1598 struct btrfs_root *root,
1599 struct btrfs_path *path,
1600 struct inode_record *rec)
1601 {
1602 struct btrfs_key key;
1603 int ret;
1604
1605 key.objectid = BTRFS_ORPHAN_OBJECTID;
1606 key.type = BTRFS_ORPHAN_ITEM_KEY;
1607 key.offset = rec->ino;
1608
1609 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1610 btrfs_release_path(path);
1611 if (!ret)
1612 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
1613 return ret;
1614 }
1615
1616 static int add_missing_dir_index(struct btrfs_root *root,
1617 struct cache_tree *inode_cache,
1618 struct inode_record *rec,
1619 struct inode_backref *backref)
1620 {
1621 struct btrfs_path *path;
1622 struct btrfs_trans_handle *trans;
1623 struct btrfs_dir_item *dir_item;
1624 struct extent_buffer *leaf;
1625 struct btrfs_key key;
1626 struct btrfs_disk_key disk_key;
1627 struct inode_record *dir_rec;
1628 unsigned long name_ptr;
1629 u32 data_size = sizeof(*dir_item) + backref->namelen;
1630 int ret;
1631
1632 path = btrfs_alloc_path();
1633 if (!path)
1634 return -ENOMEM;
1635
1636 trans = btrfs_start_transaction(root, 1);
1637 if (IS_ERR(trans)) {
1638 btrfs_free_path(path);
1639 return PTR_ERR(trans);
1640 }
1641
1642 fprintf(stderr, "repairing missing dir index item for inode %llu\n",
1643 (unsigned long long)rec->ino);
1644 key.objectid = backref->dir;
1645 key.type = BTRFS_DIR_INDEX_KEY;
1646 key.offset = backref->index;
1647
1648 ret = btrfs_insert_empty_item(trans, root, path, &key, data_size);
1649 BUG_ON(ret);
1650
1651 leaf = path->nodes[0];
1652 dir_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
1653
1654 disk_key.objectid = cpu_to_le64(rec->ino);
1655 disk_key.type = BTRFS_INODE_ITEM_KEY;
1656 disk_key.offset = 0;
1657
1658 btrfs_set_dir_item_key(leaf, dir_item, &disk_key);
1659 btrfs_set_dir_type(leaf, dir_item, imode_to_type(rec->imode));
1660 btrfs_set_dir_data_len(leaf, dir_item, 0);
1661 btrfs_set_dir_name_len(leaf, dir_item, backref->namelen);
1662 name_ptr = (unsigned long)(dir_item + 1);
1663 write_extent_buffer(leaf, backref->name, name_ptr, backref->namelen);
1664 btrfs_mark_buffer_dirty(leaf);
1665 btrfs_free_path(path);
1666 btrfs_commit_transaction(trans, root);
1667
1668 backref->found_dir_index = 1;
1669 dir_rec = get_inode_rec(inode_cache, backref->dir, 0);
1670 if (!dir_rec)
1671 return 0;
1672 dir_rec->found_size += backref->namelen;
1673 if (dir_rec->found_size == dir_rec->isize &&
1674 (dir_rec->errors & I_ERR_DIR_ISIZE_WRONG))
1675 dir_rec->errors &= ~I_ERR_DIR_ISIZE_WRONG;
1676 if (dir_rec->found_size != dir_rec->isize)
1677 dir_rec->errors |= I_ERR_DIR_ISIZE_WRONG;
1678
1679 return 0;
1680 }
1681
1682 static int delete_dir_index(struct btrfs_root *root,
1683 struct cache_tree *inode_cache,
1684 struct inode_record *rec,
1685 struct inode_backref *backref)
1686 {
1687 struct btrfs_trans_handle *trans;
1688 struct btrfs_dir_item *di;
1689 struct btrfs_path *path;
1690 int ret = 0;
1691
1692 path = btrfs_alloc_path();
1693 if (!path)
1694 return -ENOMEM;
1695
1696 trans = btrfs_start_transaction(root, 1);
1697 if (IS_ERR(trans)) {
1698 btrfs_free_path(path);
1699 return PTR_ERR(trans);
1700 }
1701
1702
1703 fprintf(stderr, "Deleting bad dir index [%llu,%u,%llu] root %llu\n",
1704 (unsigned long long)backref->dir,
1705 BTRFS_DIR_INDEX_KEY, (unsigned long long)backref->index,
1706 (unsigned long long)root->objectid);
1707
1708 di = btrfs_lookup_dir_index(trans, root, path, backref->dir,
1709 backref->name, backref->namelen,
1710 backref->index, -1);
1711 if (IS_ERR(di)) {
1712 ret = PTR_ERR(di);
1713 btrfs_free_path(path);
1714 btrfs_commit_transaction(trans, root);
1715 if (ret == -ENOENT)
1716 return 0;
1717 return ret;
1718 }
1719
1720 if (!di)
1721 ret = btrfs_del_item(trans, root, path);
1722 else
1723 ret = btrfs_delete_one_dir_name(trans, root, path, di);
1724 BUG_ON(ret);
1725 btrfs_free_path(path);
1726 btrfs_commit_transaction(trans, root);
1727 return ret;
1728 }
1729
1730 static int create_inode_item(struct btrfs_root *root,
1731 struct inode_record *rec,
1732 struct inode_backref *backref, int root_dir)
1733 {
1734 struct btrfs_trans_handle *trans;
1735 struct btrfs_inode_item inode_item;
1736 time_t now = time(NULL);
1737 int ret;
1738
1739 trans = btrfs_start_transaction(root, 1);
1740 if (IS_ERR(trans)) {
1741 ret = PTR_ERR(trans);
1742 return ret;
1743 }
1744
1745 fprintf(stderr, "root %llu inode %llu recreating inode item, this may "
1746 "be incomplete, please check permissions and content after "
1747 "the fsck completes.\n", (unsigned long long)root->objectid,
1748 (unsigned long long)rec->ino);
1749
1750 memset(&inode_item, 0, sizeof(inode_item));
1751 btrfs_set_stack_inode_generation(&inode_item, trans->transid);
1752 if (root_dir)
1753 btrfs_set_stack_inode_nlink(&inode_item, 1);
1754 else
1755 btrfs_set_stack_inode_nlink(&inode_item, rec->found_link);
1756 btrfs_set_stack_inode_nbytes(&inode_item, rec->found_size);
1757 if (rec->found_dir_item) {
1758 if (rec->found_file_extent)
1759 fprintf(stderr, "root %llu inode %llu has both a dir "
1760 "item and extents, unsure if it is a dir or a "
1761 "regular file so setting it as a directory\n",
1762 (unsigned long long)root->objectid,
1763 (unsigned long long)rec->ino);
1764 btrfs_set_stack_inode_mode(&inode_item, S_IFDIR | 0755);
1765 btrfs_set_stack_inode_size(&inode_item, rec->found_size);
1766 } else if (!rec->found_dir_item) {
1767 btrfs_set_stack_inode_size(&inode_item, rec->extent_end);
1768 btrfs_set_stack_inode_mode(&inode_item, S_IFREG | 0755);
1769 }
1770 btrfs_set_stack_timespec_sec(&inode_item.atime, now);
1771 btrfs_set_stack_timespec_nsec(&inode_item.atime, 0);
1772 btrfs_set_stack_timespec_sec(&inode_item.ctime, now);
1773 btrfs_set_stack_timespec_nsec(&inode_item.ctime, 0);
1774 btrfs_set_stack_timespec_sec(&inode_item.mtime, now);
1775 btrfs_set_stack_timespec_nsec(&inode_item.mtime, 0);
1776 btrfs_set_stack_timespec_sec(&inode_item.otime, 0);
1777 btrfs_set_stack_timespec_nsec(&inode_item.otime, 0);
1778
1779 ret = btrfs_insert_inode(trans, root, rec->ino, &inode_item);
1780 BUG_ON(ret);
1781 btrfs_commit_transaction(trans, root);
1782 return 0;
1783 }
1784
1785 static int repair_inode_backrefs(struct btrfs_root *root,
1786 struct inode_record *rec,
1787 struct cache_tree *inode_cache,
1788 int delete)
1789 {
1790 struct inode_backref *tmp, *backref;
1791 u64 root_dirid = btrfs_root_dirid(&root->root_item);
1792 int ret = 0;
1793 int repaired = 0;
1794
1795 list_for_each_entry_safe(backref, tmp, &rec->backrefs, list) {
1796 if (!delete && rec->ino == root_dirid) {
1797 if (!rec->found_inode_item) {
1798 ret = create_inode_item(root, rec, backref, 1);
1799 if (ret)
1800 break;
1801 repaired++;
1802 }
1803 }
1804
1805 /* Index 0 for root dir's are special, don't mess with it */
1806 if (rec->ino == root_dirid && backref->index == 0)
1807 continue;
1808
1809 if (delete &&
1810 ((backref->found_dir_index && !backref->found_inode_ref) ||
1811 (backref->found_dir_index && backref->found_inode_ref &&
1812 (backref->errors & REF_ERR_INDEX_UNMATCH)))) {
1813 ret = delete_dir_index(root, inode_cache, rec, backref);
1814 if (ret)
1815 break;
1816 repaired++;
1817 list_del(&backref->list);
1818 free(backref);
1819 }
1820
1821 if (!delete && !backref->found_dir_index &&
1822 backref->found_dir_item && backref->found_inode_ref) {
1823 ret = add_missing_dir_index(root, inode_cache, rec,
1824 backref);
1825 if (ret)
1826 break;
1827 repaired++;
1828 if (backref->found_dir_item &&
1829 backref->found_dir_index &&
1830 backref->found_dir_index) {
1831 if (!backref->errors &&
1832 backref->found_inode_ref) {
1833 list_del(&backref->list);
1834 free(backref);
1835 }
1836 }
1837 }
1838
1839 if (!delete && (!backref->found_dir_index &&
1840 !backref->found_dir_item &&
1841 backref->found_inode_ref)) {
1842 struct btrfs_trans_handle *trans;
1843 struct btrfs_key location;
1844
1845 location.objectid = rec->ino;
1846 location.type = BTRFS_INODE_ITEM_KEY;
1847 location.offset = 0;
1848
1849 trans = btrfs_start_transaction(root, 1);
1850 if (IS_ERR(trans)) {
1851 ret = PTR_ERR(trans);
1852 break;
1853 }
1854 fprintf(stderr, "adding missing dir index/item pair "
1855 "for inode %llu\n",
1856 (unsigned long long)rec->ino);
1857 ret = btrfs_insert_dir_item(trans, root, backref->name,
1858 backref->namelen,
1859 backref->dir, &location,
1860 imode_to_type(rec->imode),
1861 backref->index);
1862 BUG_ON(ret);
1863 btrfs_commit_transaction(trans, root);
1864 repaired++;
1865 }
1866
1867 if (!delete && (backref->found_inode_ref &&
1868 backref->found_dir_index &&
1869 backref->found_dir_item &&
1870 !(backref->errors & REF_ERR_INDEX_UNMATCH) &&
1871 !rec->found_inode_item)) {
1872 ret = create_inode_item(root, rec, backref, 0);
1873 if (ret)
1874 break;
1875 repaired++;
1876 }
1877
1878 }
1879 return ret ? ret : repaired;
1880 }
1881
1882 static int try_repair_inode(struct btrfs_root *root, struct inode_record *rec)
1883 {
1884 struct btrfs_trans_handle *trans;
1885 struct btrfs_path *path;
1886 int ret = 0;
1887
1888 if (!(rec->errors & (I_ERR_DIR_ISIZE_WRONG | I_ERR_NO_ORPHAN_ITEM)))
1889 return rec->errors;
1890
1891 path = btrfs_alloc_path();
1892 if (!path)
1893 return -ENOMEM;
1894
1895 trans = btrfs_start_transaction(root, 1);
1896 if (IS_ERR(trans)) {
1897 btrfs_free_path(path);
1898 return PTR_ERR(trans);
1899 }
1900
1901 if (rec->errors & I_ERR_DIR_ISIZE_WRONG)
1902 ret = repair_inode_isize(trans, root, path, rec);
1903 if (!ret && rec->errors & I_ERR_NO_ORPHAN_ITEM)
1904 ret = repair_inode_orphan_item(trans, root, path, rec);
1905 btrfs_commit_transaction(trans, root);
1906 btrfs_free_path(path);
1907 return ret;
1908 }
1909
1910 static int check_inode_recs(struct btrfs_root *root,
1911 struct cache_tree *inode_cache)
1912 {
1913 struct cache_extent *cache;
1914 struct ptr_node *node;
1915 struct inode_record *rec;
1916 struct inode_backref *backref;
1917 int stage = 0;
1918 int ret;
1919 int err = 0;
1920 u64 error = 0;
1921 u64 root_dirid = btrfs_root_dirid(&root->root_item);
1922
1923 if (btrfs_root_refs(&root->root_item) == 0) {
1924 if (!cache_tree_empty(inode_cache))
1925 fprintf(stderr, "warning line %d\n", __LINE__);
1926 return 0;
1927 }
1928
1929 /*
1930 * We need to repair backrefs first because we could change some of the
1931 * errors in the inode recs.
1932 *
1933 * We also need to go through and delete invalid backrefs first and then
1934 * add the correct ones second. We do this because we may get EEXIST
1935 * when adding back the correct index because we hadn't yet deleted the
1936 * invalid index.
1937 *
1938 * For example, if we were missing a dir index then the directories
1939 * isize would be wrong, so if we fixed the isize to what we thought it
1940 * would be and then fixed the backref we'd still have a invalid fs, so
1941 * we need to add back the dir index and then check to see if the isize
1942 * is still wrong.
1943 */
1944 while (stage < 3) {
1945 stage++;
1946 if (stage == 3 && !err)
1947 break;
1948
1949 cache = search_cache_extent(inode_cache, 0);
1950 while (repair && cache) {
1951 node = container_of(cache, struct ptr_node, cache);
1952 rec = node->data;
1953 cache = next_cache_extent(cache);
1954
1955 /* Need to free everything up and rescan */
1956 if (stage == 3) {
1957 remove_cache_extent(inode_cache, &node->cache);
1958 free(node);
1959 free_inode_rec(rec);
1960 continue;
1961 }
1962
1963 if (list_empty(&rec->backrefs))
1964 continue;
1965
1966 ret = repair_inode_backrefs(root, rec, inode_cache,
1967 stage == 1);
1968 if (ret < 0) {
1969 err = ret;
1970 stage = 2;
1971 break;
1972 } if (ret > 0) {
1973 err = -EAGAIN;
1974 }
1975 }
1976 }
1977 if (err)
1978 return err;
1979
1980 rec = get_inode_rec(inode_cache, root_dirid, 0);
1981 if (rec) {
1982 ret = check_root_dir(rec);
1983 if (ret) {
1984 fprintf(stderr, "root %llu root dir %llu error\n",
1985 (unsigned long long)root->root_key.objectid,
1986 (unsigned long long)root_dirid);
1987 print_inode_error(root, rec);
1988 error++;
1989 }
1990 } else {
1991 if (repair) {
1992 struct btrfs_trans_handle *trans;
1993
1994 trans = btrfs_start_transaction(root, 1);
1995 if (IS_ERR(trans)) {
1996 err = PTR_ERR(trans);
1997 return err;
1998 }
1999
2000 fprintf(stderr,
2001 "root %llu missing its root dir, recreating\n",
2002 (unsigned long long)root->objectid);
2003
2004 ret = btrfs_make_root_dir(trans, root, root_dirid);
2005 BUG_ON(ret);
2006
2007 btrfs_commit_transaction(trans, root);
2008 return -EAGAIN;
2009 }
2010
2011 fprintf(stderr, "root %llu root dir %llu not found\n",
2012 (unsigned long long)root->root_key.objectid,
2013 (unsigned long long)root_dirid);
2014 }
2015
2016 while (1) {
2017 cache = search_cache_extent(inode_cache, 0);
2018 if (!cache)
2019 break;
2020 node = container_of(cache, struct ptr_node, cache);
2021 rec = node->data;
2022 remove_cache_extent(inode_cache, &node->cache);
2023 free(node);
2024 if (rec->ino == root_dirid ||
2025 rec->ino == BTRFS_ORPHAN_OBJECTID) {
2026 free_inode_rec(rec);
2027 continue;
2028 }
2029
2030 if (rec->errors & I_ERR_NO_ORPHAN_ITEM) {
2031 ret = check_orphan_item(root, rec->ino);
2032 if (ret == 0)
2033 rec->errors &= ~I_ERR_NO_ORPHAN_ITEM;
2034 if (can_free_inode_rec(rec)) {
2035 free_inode_rec(rec);
2036 continue;
2037 }
2038 }
2039
2040 if (repair) {
2041 ret = try_repair_inode(root, rec);
2042 if (ret == 0 && can_free_inode_rec(rec)) {
2043 free_inode_rec(rec);
2044 continue;
2045 }
2046 ret = 0;
2047 }
2048
2049 error++;
2050 if (!rec->found_inode_item)
2051 rec->errors |= I_ERR_NO_INODE_ITEM;
2052 if (rec->found_link != rec->nlink)
2053 rec->errors |= I_ERR_LINK_COUNT_WRONG;
2054 print_inode_error(root, rec);
2055 list_for_each_entry(backref, &rec->backrefs, list) {
2056 if (!backref->found_dir_item)
2057 backref->errors |= REF_ERR_NO_DIR_ITEM;
2058 if (!backref->found_dir_index)
2059 backref->errors |= REF_ERR_NO_DIR_INDEX;
2060 if (!backref->found_inode_ref)
2061 backref->errors |= REF_ERR_NO_INODE_REF;
2062 fprintf(stderr, "\tunresolved ref dir %llu index %llu"
2063 " namelen %u name %s filetype %d errors %x",
2064 (unsigned long long)backref->dir,
2065 (unsigned long long)backref->index,
2066 backref->namelen, backref->name,
2067 backref->filetype, backref->errors);
2068 print_ref_error(backref->errors);
2069 }
2070 free_inode_rec(rec);
2071 }
2072 return (error > 0) ? -1 : 0;
2073 }
2074
2075 static struct root_record *get_root_rec(struct cache_tree *root_cache,
2076 u64 objectid)
2077 {
2078 struct cache_extent *cache;
2079 struct root_record *rec = NULL;
2080 int ret;
2081
2082 cache = lookup_cache_extent(root_cache, objectid, 1);
2083 if (cache) {
2084 rec = container_of(cache, struct root_record, cache);
2085 } else {
2086 rec = calloc(1, sizeof(*rec));
2087 rec->objectid = objectid;
2088 INIT_LIST_HEAD(&rec->backrefs);
2089 rec->cache.start = objectid;
2090 rec->cache.size = 1;
2091
2092 ret = insert_cache_extent(root_cache, &rec->cache);
2093 BUG_ON(ret);
2094 }
2095 return rec;
2096 }
2097
2098 static struct root_backref *get_root_backref(struct root_record *rec,
2099 u64 ref_root, u64 dir, u64 index,
2100 const char *name, int namelen)
2101 {
2102 struct root_backref *backref;
2103
2104 list_for_each_entry(backref, &rec->backrefs, list) {
2105 if (backref->ref_root != ref_root || backref->dir != dir ||
2106 backref->namelen != namelen)
2107 continue;
2108 if (memcmp(name, backref->name, namelen))
2109 continue;
2110 return backref;
2111 }
2112
2113 backref = malloc(sizeof(*backref) + namelen + 1);
2114 memset(backref, 0, sizeof(*backref));
2115 backref->ref_root = ref_root;
2116 backref->dir = dir;
2117 backref->index = index;
2118 backref->namelen = namelen;
2119 memcpy(backref->name, name, namelen);
2120 backref->name[namelen] = '\0';
2121 list_add_tail(&backref->list, &rec->backrefs);
2122 return backref;
2123 }
2124
2125 static void free_root_record(struct cache_extent *cache)
2126 {
2127 struct root_record *rec;
2128 struct root_backref *backref;
2129
2130 rec = container_of(cache, struct root_record, cache);
2131 while (!list_empty(&rec->backrefs)) {
2132 backref = list_entry(rec->backrefs.next,
2133 struct root_backref, list);
2134 list_del(&backref->list);
2135 free(backref);
2136 }
2137
2138 kfree(rec);
2139 }
2140
2141 FREE_EXTENT_CACHE_BASED_TREE(root_recs, free_root_record);
2142
2143 static int add_root_backref(struct cache_tree *root_cache,
2144 u64 root_id, u64 ref_root, u64 dir, u64 index,
2145 const char *name, int namelen,
2146 int item_type, int errors)
2147 {
2148 struct root_record *rec;
2149 struct root_backref *backref;
2150
2151 rec = get_root_rec(root_cache, root_id);
2152 backref = get_root_backref(rec, ref_root, dir, index, name, namelen);
2153
2154 backref->errors |= errors;
2155
2156 if (item_type != BTRFS_DIR_ITEM_KEY) {
2157 if (backref->found_dir_index || backref->found_back_ref ||
2158 backref->found_forward_ref) {
2159 if (backref->index != index)
2160 backref->errors |= REF_ERR_INDEX_UNMATCH;
2161 } else {
2162 backref->index = index;
2163 }
2164 }
2165
2166 if (item_type == BTRFS_DIR_ITEM_KEY) {
2167 if (backref->found_forward_ref)
2168 rec->found_ref++;
2169 backref->found_dir_item = 1;
2170 } else if (item_type == BTRFS_DIR_INDEX_KEY) {
2171 backref->found_dir_index = 1;
2172 } else if (item_type == BTRFS_ROOT_REF_KEY) {
2173 if (backref->found_forward_ref)
2174 backref->errors |= REF_ERR_DUP_ROOT_REF;
2175 else if (backref->found_dir_item)
2176 rec->found_ref++;
2177 backref->found_forward_ref = 1;
2178 } else if (item_type == BTRFS_ROOT_BACKREF_KEY) {
2179 if (backref->found_back_ref)
2180 backref->errors |= REF_ERR_DUP_ROOT_BACKREF;
2181 backref->found_back_ref = 1;
2182 } else {
2183 BUG_ON(1);
2184 }
2185
2186 if (backref->found_forward_ref && backref->found_dir_item)
2187 backref->reachable = 1;
2188 return 0;
2189 }
2190
2191 static int merge_root_recs(struct btrfs_root *root,
2192 struct cache_tree *src_cache,
2193 struct cache_tree *dst_cache)
2194 {
2195 struct cache_extent *cache;
2196 struct ptr_node *node;
2197 struct inode_record *rec;
2198 struct inode_backref *backref;
2199 int ret = 0;
2200
2201 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2202 free_inode_recs_tree(src_cache);
2203 return 0;
2204 }
2205
2206 while (1) {
2207 cache = search_cache_extent(src_cache, 0);
2208 if (!cache)
2209 break;
2210 node = container_of(cache, struct ptr_node, cache);
2211 rec = node->data;
2212 remove_cache_extent(src_cache, &node->cache);
2213 free(node);
2214
2215 ret = is_child_root(root, root->objectid, rec->ino);
2216 if (ret < 0)
2217 break;
2218 else if (ret == 0)
2219 goto skip;
2220
2221 list_for_each_entry(backref, &rec->backrefs, list) {
2222 BUG_ON(backref->found_inode_ref);
2223 if (backref->found_dir_item)
2224 add_root_backref(dst_cache, rec->ino,
2225 root->root_key.objectid, backref->dir,
2226 backref->index, backref->name,
2227 backref->namelen, BTRFS_DIR_ITEM_KEY,
2228 backref->errors);
2229 if (backref->found_dir_index)
2230 add_root_backref(dst_cache, rec->ino,
2231 root->root_key.objectid, backref->dir,
2232 backref->index, backref->name,
2233 backref->namelen, BTRFS_DIR_INDEX_KEY,
2234 backref->errors);
2235 }
2236 skip:
2237 free_inode_rec(rec);
2238 }
2239 if (ret < 0)
2240 return ret;
2241 return 0;
2242 }
2243
2244 static int check_root_refs(struct btrfs_root *root,
2245 struct cache_tree *root_cache)
2246 {
2247 struct root_record *rec;
2248 struct root_record *ref_root;
2249 struct root_backref *backref;
2250 struct cache_extent *cache;
2251 int loop = 1;
2252 int ret;
2253 int error;
2254 int errors = 0;
2255
2256 rec = get_root_rec(root_cache, BTRFS_FS_TREE_OBJECTID);
2257 rec->found_ref = 1;
2258
2259 /* fixme: this can not detect circular references */
2260 while (loop) {
2261 loop = 0;
2262 cache = search_cache_extent(root_cache, 0);
2263 while (1) {
2264 if (!cache)
2265 break;
2266 rec = container_of(cache, struct root_record, cache);
2267 cache = next_cache_extent(cache);
2268
2269 if (rec->found_ref == 0)
2270 continue;
2271
2272 list_for_each_entry(backref, &rec->backrefs, list) {
2273 if (!backref->reachable)
2274 continue;
2275
2276 ref_root = get_root_rec(root_cache,
2277 backref->ref_root);
2278 if (ref_root->found_ref > 0)
2279 continue;
2280
2281 backref->reachable = 0;
2282 rec->found_ref--;
2283 if (rec->found_ref == 0)
2284 loop = 1;
2285 }
2286 }
2287 }
2288
2289 cache = search_cache_extent(root_cache, 0);
2290 while (1) {
2291 if (!cache)
2292 break;
2293 rec = container_of(cache, struct root_record, cache);
2294 cache = next_cache_extent(cache);
2295
2296 if (rec->found_ref == 0 &&
2297 rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
2298 rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
2299 ret = check_orphan_item(root->fs_info->tree_root,
2300 rec->objectid);
2301 if (ret == 0)
2302 continue;
2303
2304 /*
2305 * If we don't have a root item then we likely just have
2306 * a dir item in a snapshot for this root but no actual
2307 * ref key or anything so it's meaningless.
2308 */
2309 if (!rec->found_root_item)
2310 continue;
2311 errors++;
2312 fprintf(stderr, "fs tree %llu not referenced\n",
2313 (unsigned long long)rec->objectid);
2314 }
2315
2316 error = 0;
2317 if (rec->found_ref > 0 && !rec->found_root_item)
2318 error = 1;
2319 list_for_each_entry(backref, &rec->backrefs, list) {
2320 if (!backref->found_dir_item)
2321 backref->errors |= REF_ERR_NO_DIR_ITEM;
2322 if (!backref->found_dir_index)
2323 backref->errors |= REF_ERR_NO_DIR_INDEX;
2324 if (!backref->found_back_ref)
2325 backref->errors |= REF_ERR_NO_ROOT_BACKREF;
2326 if (!backref->found_forward_ref)
2327 backref->errors |= REF_ERR_NO_ROOT_REF;
2328 if (backref->reachable && backref->errors)
2329 error = 1;
2330 }
2331 if (!error)
2332 continue;
2333
2334 errors++;
2335 fprintf(stderr, "fs tree %llu refs %u %s\n",
2336 (unsigned long long)rec->objectid, rec->found_ref,
2337 rec->found_root_item ? "" : "not found");
2338
2339 list_for_each_entry(backref, &rec->backrefs, list) {
2340 if (!backref->reachable)
2341 continue;
2342 if (!backref->errors && rec->found_root_item)
2343 continue;
2344 fprintf(stderr, "\tunresolved ref root %llu dir %llu"
2345 " index %llu namelen %u name %s errors %x\n",
2346 (unsigned long long)backref->ref_root,
2347 (unsigned long long)backref->dir,
2348 (unsigned long long)backref->index,
2349 backref->namelen, backref->name,
2350 backref->errors);
2351 print_ref_error(backref->errors);
2352 }
2353 }
2354 return errors > 0 ? 1 : 0;
2355 }
2356
2357 static int process_root_ref(struct extent_buffer *eb, int slot,
2358 struct btrfs_key *key,
2359 struct cache_tree *root_cache)
2360 {
2361 u64 dirid;
2362 u64 index;
2363 u32 len;
2364 u32 name_len;
2365 struct btrfs_root_ref *ref;
2366 char namebuf[BTRFS_NAME_LEN];
2367 int error;
2368
2369 ref = btrfs_item_ptr(eb, slot, struct btrfs_root_ref);
2370
2371 dirid = btrfs_root_ref_dirid(eb, ref);
2372 index = btrfs_root_ref_sequence(eb, ref);
2373 name_len = btrfs_root_ref_name_len(eb, ref);
2374
2375 if (name_len <= BTRFS_NAME_LEN) {
2376 len = name_len;
2377 error = 0;
2378 } else {
2379 len = BTRFS_NAME_LEN;
2380 error = REF_ERR_NAME_TOO_LONG;
2381 }
2382 read_extent_buffer(eb, namebuf, (unsigned long)(ref + 1), len);
2383
2384 if (key->type == BTRFS_ROOT_REF_KEY) {
2385 add_root_backref(root_cache, key->offset, key->objectid, dirid,
2386 index, namebuf, len, key->type, error);
2387 } else {
2388 add_root_backref(root_cache, key->objectid, key->offset, dirid,
2389 index, namebuf, len, key->type, error);
2390 }
2391 return 0;
2392 }
2393
2394 static int check_fs_root(struct btrfs_root *root,
2395 struct cache_tree *root_cache,
2396 struct walk_control *wc)
2397 {
2398 int ret = 0;
2399 int err = 0;
2400 int wret;
2401 int level;
2402 struct btrfs_path path;
2403 struct shared_node root_node;
2404 struct root_record *rec;
2405 struct btrfs_root_item *root_item = &root->root_item;
2406 enum btrfs_tree_block_status status;
2407
2408 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2409 rec = get_root_rec(root_cache, root->root_key.objectid);
2410 if (btrfs_root_refs(root_item) > 0)
2411 rec->found_root_item = 1;
2412 }
2413
2414 btrfs_init_path(&path);
2415 memset(&root_node, 0, sizeof(root_node));
2416 cache_tree_init(&root_node.root_cache);
2417 cache_tree_init(&root_node.inode_cache);
2418
2419 level = btrfs_header_level(root->node);
2420 memset(wc->nodes, 0, sizeof(wc->nodes));
2421 wc->nodes[level] = &root_node;
2422 wc->active_node = level;
2423 wc->root_level = level;
2424
2425 /* We may not have checked the root block, lets do that now */
2426 if (btrfs_is_leaf(root->node))
2427 status = btrfs_check_leaf(root, NULL, root->node);
2428 else
2429 status = btrfs_check_node(root, NULL, root->node);
2430 if (status != BTRFS_TREE_BLOCK_CLEAN)
2431 return -EIO;
2432
2433 if (btrfs_root_refs(root_item) > 0 ||
2434 btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2435 path.nodes[level] = root->node;
2436 extent_buffer_get(root->node);
2437 path.slots[level] = 0;
2438 } else {
2439 struct btrfs_key key;
2440 struct btrfs_disk_key found_key;
2441
2442 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2443 level = root_item->drop_level;
2444 path.lowest_level = level;
2445 wret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
2446 if (wret < 0)
2447 goto skip_walking;
2448 btrfs_node_key(path.nodes[level], &found_key,
2449 path.slots[level]);
2450 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2451 sizeof(found_key)));
2452 }
2453
2454 while (1) {
2455 wret = walk_down_tree(root, &path, wc, &level);
2456 if (wret < 0)
2457 ret = wret;
2458 if (wret != 0)
2459 break;
2460
2461 wret = walk_up_tree(root, &path, wc, &level);
2462 if (wret < 0)
2463 ret = wret;
2464 if (wret != 0)
2465 break;
2466 }
2467 skip_walking:
2468 btrfs_release_path(&path);
2469
2470 err = merge_root_recs(root, &root_node.root_cache, root_cache);
2471 if (err < 0)
2472 ret = err;
2473
2474 if (root_node.current) {
2475 root_node.current->checked = 1;
2476 maybe_free_inode_rec(&root_node.inode_cache,
2477 root_node.current);
2478 }
2479
2480 err = check_inode_recs(root, &root_node.inode_cache);
2481 if (!ret)
2482 ret = err;
2483 return ret;
2484 }
2485
2486 static int fs_root_objectid(u64 objectid)
2487 {
2488 if (objectid == BTRFS_TREE_RELOC_OBJECTID ||
2489 objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2490 return 1;
2491 return is_fstree(objectid);
2492 }
2493
2494 static int check_fs_roots(struct btrfs_root *root,
2495 struct cache_tree *root_cache)
2496 {
2497 struct btrfs_path path;
2498 struct btrfs_key key;
2499 struct walk_control wc;
2500 struct extent_buffer *leaf, *tree_node;
2501 struct btrfs_root *tmp_root;
2502 struct btrfs_root *tree_root = root->fs_info->tree_root;
2503 int ret;
2504 int err = 0;
2505
2506 /*
2507 * Just in case we made any changes to the extent tree that weren't
2508 * reflected into the free space cache yet.
2509 */
2510 if (repair)
2511 reset_cached_block_groups(root->fs_info);
2512 memset(&wc, 0, sizeof(wc));
2513 cache_tree_init(&wc.shared);
2514 btrfs_init_path(&path);
2515
2516 again:
2517 key.offset = 0;
2518 key.objectid = 0;
2519 key.type = BTRFS_ROOT_ITEM_KEY;
2520 ret = btrfs_search_slot(NULL, tree_root, &key, &path, 0, 0);
2521 if (ret < 0) {
2522 err = 1;
2523 goto out;
2524 }
2525 tree_node = tree_root->node;
2526 while (1) {
2527 if (tree_node != tree_root->node) {
2528 free_root_recs_tree(root_cache);
2529 btrfs_release_path(&path);
2530 goto again;
2531 }
2532 leaf = path.nodes[0];
2533 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
2534 ret = btrfs_next_leaf(tree_root, &path);
2535 if (ret) {
2536 if (ret < 0)
2537 err = 1;
2538 break;
2539 }
2540 leaf = path.nodes[0];
2541 }
2542 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
2543 if (key.type == BTRFS_ROOT_ITEM_KEY &&
2544 fs_root_objectid(key.objectid)) {
2545 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2546 tmp_root = btrfs_read_fs_root_no_cache(
2547 root->fs_info, &key);
2548 } else {
2549 key.offset = (u64)-1;
2550 tmp_root = btrfs_read_fs_root(
2551 root->fs_info, &key);
2552 }
2553 if (IS_ERR(tmp_root)) {
2554 err = 1;
2555 goto next;
2556 }
2557 ret = check_fs_root(tmp_root, root_cache, &wc);
2558 if (ret == -EAGAIN) {
2559 free_root_recs_tree(root_cache);
2560 btrfs_release_path(&path);
2561 goto again;
2562 }
2563 if (ret)
2564 err = 1;
2565 if (key.objectid == BTRFS_TREE_RELOC_OBJECTID)
2566 btrfs_free_fs_root(tmp_root);
2567 } else if (key.type == BTRFS_ROOT_REF_KEY ||
2568 key.type == BTRFS_ROOT_BACKREF_KEY) {
2569 process_root_ref(leaf, path.slots[0], &key,
2570 root_cache);
2571 }
2572 next:
2573 path.slots[0]++;
2574 }
2575 out:
2576 btrfs_release_path(&path);
2577 if (err)
2578 free_extent_cache_tree(&wc.shared);
2579 if (!cache_tree_empty(&wc.shared))
2580 fprintf(stderr, "warning line %d\n", __LINE__);
2581
2582 return err;
2583 }
2584
2585 static int all_backpointers_checked(struct extent_record *rec, int print_errs)
2586 {
2587 struct list_head *cur = rec->backrefs.next;
2588 struct extent_backref *back;
2589 struct tree_backref *tback;
2590 struct data_backref *dback;
2591 u64 found = 0;
2592 int err = 0;
2593
2594 while(cur != &rec->backrefs) {
2595 back = list_entry(cur, struct extent_backref, list);
2596 cur = cur->next;
2597 if (!back->found_extent_tree) {
2598 err = 1;
2599 if (!print_errs)
2600 goto out;
2601 if (back->is_data) {
2602 dback = (struct data_backref *)back;
2603 fprintf(stderr, "Backref %llu %s %llu"
2604 " owner %llu offset %llu num_refs %lu"
2605 " not found in extent tree\n",
2606 (unsigned long long)rec->start,
2607 back->full_backref ?
2608 "parent" : "root",
2609 back->full_backref ?
2610 (unsigned long long)dback->parent:
2611 (unsigned long long)dback->root,
2612 (unsigned long long)dback->owner,
2613 (unsigned long long)dback->offset,
2614 (unsigned long)dback->num_refs);
2615 } else {
2616 tback = (struct tree_backref *)back;
2617 fprintf(stderr, "Backref %llu parent %llu"
2618 " root %llu not found in extent tree\n",
2619 (unsigned long long)rec->start,
2620 (unsigned long long)tback->parent,
2621 (unsigned long long)tback->root);
2622 }
2623 }
2624 if (!back->is_data && !back->found_ref) {
2625 err = 1;
2626 if (!print_errs)
2627 goto out;
2628 tback = (struct tree_backref *)back;
2629 fprintf(stderr, "Backref %llu %s %llu not referenced back %p\n",
2630 (unsigned long long)rec->start,
2631 back->full_backref ? "parent" : "root",
2632 back->full_backref ?
2633 (unsigned long long)tback->parent :
2634 (unsigned long long)tback->root, back);
2635 }
2636 if (back->is_data) {
2637 dback = (struct data_backref *)back;
2638 if (dback->found_ref != dback->num_refs) {
2639 err = 1;
2640 if (!print_errs)
2641 goto out;
2642 fprintf(stderr, "Incorrect local backref count"
2643 " on %llu %s %llu owner %llu"
2644 " offset %llu found %u wanted %u back %p\n",
2645 (unsigned long long)rec->start,
2646 back->full_backref ?
2647 "parent" : "root",
2648 back->full_backref ?
2649 (unsigned long long)dback->parent:
2650 (unsigned long long)dback->root,
2651 (unsigned long long)dback->owner,
2652 (unsigned long long)dback->offset,
2653 dback->found_ref, dback->num_refs, back);
2654 }
2655 if (dback->disk_bytenr != rec->start) {
2656 err = 1;
2657 if (!print_errs)
2658 goto out;
2659 fprintf(stderr, "Backref disk bytenr does not"
2660 " match extent record, bytenr=%llu, "
2661 "ref bytenr=%llu\n",
2662 (unsigned long long)rec->start,
2663 (unsigned long long)dback->disk_bytenr);
2664 }
2665
2666 if (dback->bytes != rec->nr) {
2667 err = 1;
2668 if (!print_errs)
2669 goto out;
2670 fprintf(stderr, "Backref bytes do not match "
2671 "extent backref, bytenr=%llu, ref "
2672 "bytes=%llu, backref bytes=%llu\n",
2673 (unsigned long long)rec->start,
2674 (unsigned long long)rec->nr,
2675 (unsigned long long)dback->bytes);
2676 }
2677 }
2678 if (!back->is_data) {
2679 found += 1;
2680 } else {
2681 dback = (struct data_backref *)back;
2682 found += dback->found_ref;
2683 }
2684 }
2685 if (found != rec->refs) {
2686 err = 1;
2687 if (!print_errs)
2688 goto out;
2689 fprintf(stderr, "Incorrect global backref count "
2690 "on %llu found %llu wanted %llu\n",
2691 (unsigned long long)rec->start,
2692 (unsigned long long)found,
2693 (unsigned long long)rec->refs);
2694 }
2695 out:
2696 return err;
2697 }
2698
2699 static int free_all_extent_backrefs(struct extent_record *rec)
2700 {
2701 struct extent_backref *back;
2702 struct list_head *cur;
2703 while (!list_empty(&rec->backrefs)) {
2704 cur = rec->backrefs.next;
2705 back = list_entry(cur, struct extent_backref, list);
2706 list_del(cur);
2707 free(back);
2708 }
2709 return 0;
2710 }
2711
2712 static void free_extent_record_cache(struct btrfs_fs_info *fs_info,
2713 struct cache_tree *extent_cache)
2714 {
2715 struct cache_extent *cache;
2716 struct extent_record *rec;
2717
2718 while (1) {
2719 cache = first_cache_extent(extent_cache);
2720 if (!cache)
2721 break;
2722 rec = container_of(cache, struct extent_record, cache);
2723 btrfs_unpin_extent(fs_info, rec->start, rec->max_size);
2724 remove_cache_extent(extent_cache, cache);
2725 free_all_extent_backrefs(rec);
2726 free(rec);
2727 }
2728 }
2729
2730 static int maybe_free_extent_rec(struct cache_tree *extent_cache,
2731 struct extent_record *rec)
2732 {
2733 if (rec->content_checked && rec->owner_ref_checked &&
2734 rec->extent_item_refs == rec->refs && rec->refs > 0 &&
2735 rec->num_duplicates == 0 && !all_backpointers_checked(rec, 0)) {
2736 remove_cache_extent(extent_cache, &rec->cache);
2737 free_all_extent_backrefs(rec);
2738 list_del_init(&rec->list);
2739 free(rec);
2740 }
2741 return 0;
2742 }
2743
2744 static int check_owner_ref(struct btrfs_root *root,
2745 struct extent_record *rec,
2746 struct extent_buffer *buf)
2747 {
2748 struct extent_backref *node;
2749 struct tree_backref *back;
2750 struct btrfs_root *ref_root;
2751 struct btrfs_key key;
2752 struct btrfs_path path;
2753 struct extent_buffer *parent;
2754 int level;
2755 int found = 0;
2756 int ret;
2757
2758 list_for_each_entry(node, &rec->backrefs, list) {
2759 if (node->is_data)
2760 continue;
2761 if (!node->found_ref)
2762 continue;
2763 if (node->full_backref)
2764 continue;
2765 back = (struct tree_backref *)node;
2766 if (btrfs_header_owner(buf) == back->root)
2767 return 0;
2768 }
2769 BUG_ON(rec->is_root);
2770
2771 /* try to find the block by search corresponding fs tree */
2772 key.objectid = btrfs_header_owner(buf);
2773 key.type = BTRFS_ROOT_ITEM_KEY;
2774 key.offset = (u64)-1;
2775
2776 ref_root = btrfs_read_fs_root(root->fs_info, &key);
2777 if (IS_ERR(ref_root))
2778 return 1;
2779
2780 level = btrfs_header_level(buf);
2781 if (level == 0)
2782 btrfs_item_key_to_cpu(buf, &key, 0);
2783 else
2784 btrfs_node_key_to_cpu(buf, &key, 0);
2785
2786 btrfs_init_path(&path);
2787 path.lowest_level = level + 1;
2788 ret = btrfs_search_slot(NULL, ref_root, &key, &path, 0, 0);
2789 if (ret < 0)
2790 return 0;
2791
2792 parent = path.nodes[level + 1];
2793 if (parent && buf->start == btrfs_node_blockptr(parent,
2794 path.slots[level + 1]))
2795 found = 1;
2796
2797 btrfs_release_path(&path);
2798 return found ? 0 : 1;
2799 }
2800
2801 static int is_extent_tree_record(struct extent_record *rec)
2802 {
2803 struct list_head *cur = rec->backrefs.next;
2804 struct extent_backref *node;
2805 struct tree_backref *back;
2806 int is_extent = 0;
2807
2808 while(cur != &rec->backrefs) {
2809 node = list_entry(cur, struct extent_backref, list);
2810 cur = cur->next;
2811 if (node->is_data)
2812 return 0;
2813 back = (struct tree_backref *)node;
2814 if (node->full_backref)
2815 return 0;
2816 if (back->root == BTRFS_EXTENT_TREE_OBJECTID)
2817 is_extent = 1;
2818 }
2819 return is_extent;
2820 }
2821
2822
2823 static int record_bad_block_io(struct btrfs_fs_info *info,
2824 struct cache_tree *extent_cache,
2825 u64 start, u64 len)
2826 {
2827 struct extent_record *rec;
2828 struct cache_extent *cache;
2829 struct btrfs_key key;
2830
2831 cache = lookup_cache_extent(extent_cache, start, len);
2832 if (!cache)
2833 return 0;
2834
2835 rec = container_of(cache, struct extent_record, cache);
2836 if (!is_extent_tree_record(rec))
2837 return 0;
2838
2839 btrfs_disk_key_to_cpu(&key, &rec->parent_key);
2840 return btrfs_add_corrupt_extent_record(info, &key, start, len, 0);
2841 }
2842
2843 static int swap_values(struct btrfs_root *root, struct btrfs_path *path,
2844 struct extent_buffer *buf, int slot)
2845 {
2846 if (btrfs_header_level(buf)) {
2847 struct btrfs_key_ptr ptr1, ptr2;
2848
2849 read_extent_buffer(buf, &ptr1, btrfs_node_key_ptr_offset(slot),
2850 sizeof(struct btrfs_key_ptr));
2851 read_extent_buffer(buf, &ptr2,
2852 btrfs_node_key_ptr_offset(slot + 1),
2853 sizeof(struct btrfs_key_ptr));
2854 write_extent_buffer(buf, &ptr1,
2855 btrfs_node_key_ptr_offset(slot + 1),
2856 sizeof(struct btrfs_key_ptr));
2857 write_extent_buffer(buf, &ptr2,
2858 btrfs_node_key_ptr_offset(slot),
2859 sizeof(struct btrfs_key_ptr));
2860 if (slot == 0) {
2861 struct btrfs_disk_key key;
2862 btrfs_node_key(buf, &key, 0);
2863 btrfs_fixup_low_keys(root, path, &key,
2864 btrfs_header_level(buf) + 1);
2865 }
2866 } else {
2867 struct btrfs_item *item1, *item2;
2868 struct btrfs_key k1, k2;
2869 char *item1_data, *item2_data;
2870 u32 item1_offset, item2_offset, item1_size, item2_size;
2871
2872 item1 = btrfs_item_nr(slot);
2873 item2 = btrfs_item_nr(slot + 1);
2874 btrfs_item_key_to_cpu(buf, &k1, slot);
2875 btrfs_item_key_to_cpu(buf, &k2, slot + 1);
2876 item1_offset = btrfs_item_offset(buf, item1);
2877 item2_offset = btrfs_item_offset(buf, item2);
2878 item1_size = btrfs_item_size(buf, item1);
2879 item2_size = btrfs_item_size(buf, item2);
2880
2881 item1_data = malloc(item1_size);
2882 if (!item1_data)
2883 return -ENOMEM;
2884 item2_data = malloc(item2_size);
2885 if (!item2_data) {
2886 free(item1_data);
2887 return -ENOMEM;
2888 }
2889
2890 read_extent_buffer(buf, item1_data, item1_offset, item1_size);
2891 read_extent_buffer(buf, item2_data, item2_offset, item2_size);
2892
2893 write_extent_buffer(buf, item1_data, item2_offset, item2_size);
2894 write_extent_buffer(buf, item2_data, item1_offset, item1_size);
2895 free(item1_data);
2896 free(item2_data);
2897
2898 btrfs_set_item_offset(buf, item1, item2_offset);
2899 btrfs_set_item_offset(buf, item2, item1_offset);
2900 btrfs_set_item_size(buf, item1, item2_size);
2901 btrfs_set_item_size(buf, item2, item1_size);
2902
2903 path->slots[0] = slot;
2904 btrfs_set_item_key_unsafe(root, path, &k2);
2905 path->slots[0] = slot + 1;
2906 btrfs_set_item_key_unsafe(root, path, &k1);
2907 }
2908 return 0;
2909 }
2910
2911 static int fix_key_order(struct btrfs_trans_handle *trans,
2912 struct btrfs_root *root,
2913 struct btrfs_path *path)
2914 {
2915 struct extent_buffer *buf;
2916 struct btrfs_key k1, k2;
2917 int i;
2918 int level = path->lowest_level;
2919 int ret;
2920
2921 buf = path->nodes[level];
2922 for (i = 0; i < btrfs_header_nritems(buf) - 1; i++) {
2923 if (level) {
2924 btrfs_node_key_to_cpu(buf, &k1, i);
2925 btrfs_node_key_to_cpu(buf, &k2, i + 1);
2926 } else {
2927 btrfs_item_key_to_cpu(buf, &k1, i);
2928 btrfs_item_key_to_cpu(buf, &k2, i + 1);
2929 }
2930 if (btrfs_comp_cpu_keys(&k1, &k2) < 0)
2931 continue;
2932 ret = swap_values(root, path, buf, i);
2933 if (ret)
2934 break;
2935 btrfs_mark_buffer_dirty(buf);
2936 i = 0;
2937 }
2938 return ret;
2939 }
2940
2941 static int delete_bogus_item(struct btrfs_trans_handle *trans,
2942 struct btrfs_root *root,
2943 struct btrfs_path *path,
2944 struct extent_buffer *buf, int slot)
2945 {
2946 struct btrfs_key key;
2947 int nritems = btrfs_header_nritems(buf);
2948
2949 btrfs_item_key_to_cpu(buf, &key, slot);
2950
2951 /* These are all the keys we can deal with missing. */
2952 if (key.type != BTRFS_DIR_INDEX_KEY &&
2953 key.type != BTRFS_EXTENT_ITEM_KEY &&
2954 key.type != BTRFS_METADATA_ITEM_KEY &&
2955 key.type != BTRFS_TREE_BLOCK_REF_KEY &&
2956 key.type != BTRFS_EXTENT_DATA_REF_KEY)
2957 return -1;
2958
2959 printf("Deleting bogus item [%llu,%u,%llu] at slot %d on block %llu\n",
2960 (unsigned long long)key.objectid, key.type,
2961 (unsigned long long)key.offset, slot, buf->start);
2962 memmove_extent_buffer(buf, btrfs_item_nr_offset(slot),
2963 btrfs_item_nr_offset(slot + 1),
2964 sizeof(struct btrfs_item) *
2965 (nritems - slot - 1));
2966 btrfs_set_header_nritems(buf, nritems - 1);
2967 if (slot == 0) {
2968 struct btrfs_disk_key disk_key;
2969
2970 btrfs_item_key(buf, &disk_key, 0);
2971 btrfs_fixup_low_keys(root, path, &disk_key, 1);
2972 }
2973 btrfs_mark_buffer_dirty(buf);
2974 return 0;
2975 }
2976
2977 static int fix_item_offset(struct btrfs_trans_handle *trans,
2978 struct btrfs_root *root,
2979 struct btrfs_path *path)
2980 {
2981 struct extent_buffer *buf;
2982 int i;
2983 int ret = 0;
2984
2985 /* We should only get this for leaves */
2986 BUG_ON(path->lowest_level);
2987 buf = path->nodes[0];
2988 again:
2989 for (i = 0; i < btrfs_header_nritems(buf); i++) {
2990 unsigned int shift = 0, offset;
2991
2992 if (i == 0 && btrfs_item_end_nr(buf, i) !=
2993 BTRFS_LEAF_DATA_SIZE(root)) {
2994 if (btrfs_item_end_nr(buf, i) >
2995 BTRFS_LEAF_DATA_SIZE(root)) {
2996 ret = delete_bogus_item(trans, root, path,
2997 buf, i);
2998 if (!ret)
2999 goto again;
3000 fprintf(stderr, "item is off the end of the "
3001 "leaf, can't fix\n");
3002 ret = -EIO;
3003 break;
3004 }
3005 shift = BTRFS_LEAF_DATA_SIZE(root) -
3006 btrfs_item_end_nr(buf, i);
3007 } else if (i > 0 && btrfs_item_end_nr(buf, i) !=
3008 btrfs_item_offset_nr(buf, i - 1)) {
3009 if (btrfs_item_end_nr(buf, i) >
3010 btrfs_item_offset_nr(buf, i - 1)) {
3011 ret = delete_bogus_item(trans, root, path,
3012 buf, i);
3013 if (!ret)
3014 goto again;
3015 fprintf(stderr, "items overlap, can't fix\n");
3016 ret = -EIO;
3017 break;
3018 }
3019 shift = btrfs_item_offset_nr(buf, i - 1) -
3020 btrfs_item_end_nr(buf, i);
3021 }
3022 if (!shift)
3023 continue;
3024
3025 printf("Shifting item nr %d by %u bytes in block %llu\n",
3026 i, shift, (unsigned long long)buf->start);
3027 offset = btrfs_item_offset_nr(buf, i);
3028 memmove_extent_buffer(buf,
3029 btrfs_leaf_data(buf) + offset + shift,
3030 btrfs_leaf_data(buf) + offset,
3031 btrfs_item_size_nr(buf, i));
3032 btrfs_set_item_offset(buf, btrfs_item_nr(i),
3033 offset + shift);
3034 btrfs_mark_buffer_dirty(buf);
3035 }
3036
3037 /*
3038 * We may have moved things, in which case we want to exit so we don't
3039 * write those changes out. Once we have proper abort functionality in
3040 * progs this can be changed to something nicer.
3041 */
3042 BUG_ON(ret);
3043 return ret;
3044 }
3045
3046 /*
3047 * Attempt to fix basic block failures. If we can't fix it for whatever reason
3048 * then just return -EIO.
3049 */
3050 static int try_to_fix_bad_block(struct btrfs_trans_handle *trans,
3051 struct btrfs_root *root,
3052 struct extent_buffer *buf,
3053 enum btrfs_tree_block_status status)
3054 {
3055 struct ulist *roots;
3056 struct ulist_node *node;
3057 struct btrfs_root *search_root;
3058 struct btrfs_path *path;
3059 struct ulist_iterator iter;
3060 struct btrfs_key root_key, key;
3061 int ret;
3062
3063 if (status != BTRFS_TREE_BLOCK_BAD_KEY_ORDER &&
3064 status != BTRFS_TREE_BLOCK_INVALID_OFFSETS)
3065 return -EIO;
3066
3067 path = btrfs_alloc_path();
3068 if (!path)
3069 return -EIO;
3070
3071 ret = btrfs_find_all_roots(trans, root->fs_info, buf->start,
3072 0, &roots);
3073 if (ret) {
3074 btrfs_free_path(path);
3075 return -EIO;
3076 }
3077
3078 ULIST_ITER_INIT(&iter);
3079 while ((node = ulist_next(roots, &iter))) {
3080 root_key.objectid = node->val;
3081 root_key.type = BTRFS_ROOT_ITEM_KEY;
3082 root_key.offset = (u64)-1;
3083
3084 search_root = btrfs_read_fs_root(root->fs_info, &root_key);
3085 if (IS_ERR(root)) {
3086 ret = -EIO;
3087 break;
3088 }
3089
3090 record_root_in_trans(trans, search_root);
3091
3092 path->lowest_level = btrfs_header_level(buf);
3093 path->skip_check_block = 1;
3094 if (path->lowest_level)
3095 btrfs_node_key_to_cpu(buf, &key, 0);
3096 else
3097 btrfs_item_key_to_cpu(buf, &key, 0);
3098 ret = btrfs_search_slot(trans, search_root, &key, path, 0, 1);
3099 if (ret) {
3100 ret = -EIO;
3101 break;
3102 }
3103 if (status == BTRFS_TREE_BLOCK_BAD_KEY_ORDER)
3104 ret = fix_key_order(trans, search_root, path);
3105 else if (status == BTRFS_TREE_BLOCK_INVALID_OFFSETS)
3106 ret = fix_item_offset(trans, search_root, path);
3107 if (ret)
3108 break;
3109 btrfs_release_path(path);
3110 }
3111 ulist_free(roots);
3112 btrfs_free_path(path);
3113 return ret;
3114 }
3115
3116 static int check_block(struct btrfs_trans_handle *trans,
3117 struct btrfs_root *root,
3118 struct cache_tree *extent_cache,
3119 struct extent_buffer *buf, u64 flags)
3120 {
3121 struct extent_record *rec;
3122 struct cache_extent *cache;
3123 struct btrfs_key key;
3124 enum btrfs_tree_block_status status;
3125 int ret = 0;
3126 int level;
3127
3128 cache = lookup_cache_extent(extent_cache, buf->start, buf->len);
3129 if (!cache)
3130 return 1;
3131 rec = container_of(cache, struct extent_record, cache);
3132 rec->generation = btrfs_header_generation(buf);
3133
3134 level = btrfs_header_level(buf);
3135 if (btrfs_header_nritems(buf) > 0) {
3136
3137 if (level == 0)
3138 btrfs_item_key_to_cpu(buf, &key, 0);
3139 else
3140 btrfs_node_key_to_cpu(buf, &key, 0);
3141
3142 rec->info_objectid = key.objectid;
3143 }
3144 rec->info_level = level;
3145
3146 if (btrfs_is_leaf(buf))
3147 status = btrfs_check_leaf(root, &rec->parent_key, buf);
3148 else
3149 status = btrfs_check_node(root, &rec->parent_key, buf);
3150
3151 if (status != BTRFS_TREE_BLOCK_CLEAN) {
3152 if (repair)
3153 status = try_to_fix_bad_block(trans, root, buf,
3154 status);
3155 if (status != BTRFS_TREE_BLOCK_CLEAN) {
3156 ret = -EIO;
3157 fprintf(stderr, "bad block %llu\n",
3158 (unsigned long long)buf->start);
3159 } else {
3160 /*
3161 * Signal to callers we need to start the scan over
3162 * again since we'll have cow'ed blocks.
3163 */
3164 ret = -EAGAIN;
3165 }
3166 } else {
3167 rec->content_checked = 1;
3168 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3169 rec->owner_ref_checked = 1;
3170 else {
3171 ret = check_owner_ref(root, rec, buf);
3172 if (!ret)
3173 rec->owner_ref_checked = 1;
3174 }
3175 }
3176 if (!ret)
3177 maybe_free_extent_rec(extent_cache, rec);
3178 return ret;
3179 }
3180
3181 static struct tree_backref *find_tree_backref(struct extent_record *rec,
3182 u64 parent, u64 root)
3183 {
3184 struct list_head *cur = rec->backrefs.next;
3185 struct extent_backref *node;
3186 struct tree_backref *back;
3187
3188 while(cur != &rec->backrefs) {
3189 node = list_entry(cur, struct extent_backref, list);
3190 cur = cur->next;
3191 if (node->is_data)
3192 continue;
3193 back = (struct tree_backref *)node;
3194 if (parent > 0) {
3195 if (!node->full_backref)
3196 continue;
3197 if (parent == back->parent)
3198 return back;
3199 } else {
3200 if (node->full_backref)
3201 continue;
3202 if (back->root == root)
3203 return back;
3204 }
3205 }
3206 return NULL;
3207 }
3208
3209 static struct tree_backref *alloc_tree_backref(struct extent_record *rec,
3210 u64 parent, u64 root)
3211 {
3212 struct tree_backref *ref = malloc(sizeof(*ref));
3213 memset(&ref->node, 0, sizeof(ref->node));
3214 if (parent > 0) {
3215 ref->parent = parent;
3216 ref->node.full_backref = 1;
3217 } else {
3218 ref->root = root;
3219 ref->node.full_backref = 0;
3220 }
3221 list_add_tail(&ref->node.list, &rec->backrefs);
3222
3223 return ref;
3224 }
3225
3226 static struct data_backref *find_data_backref(struct extent_record *rec,
3227 u64 parent, u64 root,
3228 u64 owner, u64 offset,
3229 int found_ref,
3230 u64 disk_bytenr, u64 bytes)
3231 {
3232 struct list_head *cur = rec->backrefs.next;
3233 struct extent_backref *node;
3234 struct data_backref *back;
3235
3236 while(cur != &rec->backrefs) {
3237 node = list_entry(cur, struct extent_backref, list);
3238 cur = cur->next;
3239 if (!node->is_data)
3240 continue;
3241 back = (struct data_backref *)node;
3242 if (parent > 0) {
3243 if (!node->full_backref)
3244 continue;
3245 if (parent == back->parent)
3246 return back;
3247 } else {
3248 if (node->full_backref)
3249 continue;
3250 if (back->root == root && back->owner == owner &&
3251 back->offset == offset) {
3252 if (found_ref && node->found_ref &&
3253 (back->bytes != bytes ||
3254 back->disk_bytenr != disk_bytenr))
3255 continue;
3256 return back;
3257 }
3258 }
3259 }
3260 return NULL;
3261 }
3262
3263 static struct data_backref *alloc_data_backref(struct extent_record *rec,
3264 u64 parent, u64 root,
3265 u64 owner, u64 offset,
3266 u64 max_size)
3267 {
3268 struct data_backref *ref = malloc(sizeof(*ref));
3269 memset(&ref->node, 0, sizeof(ref->node));
3270 ref->node.is_data = 1;
3271
3272 if (parent > 0) {
3273 ref->parent = parent;
3274 ref->owner = 0;
3275 ref->offset = 0;
3276 ref->node.full_backref = 1;
3277 } else {
3278 ref->root = root;
3279 ref->owner = owner;
3280 ref->offset = offset;
3281 ref->node.full_backref = 0;
3282 }
3283 ref->bytes = max_size;
3284 ref->found_ref = 0;
3285 ref->num_refs = 0;
3286 list_add_tail(&ref->node.list, &rec->backrefs);
3287 if (max_size > rec->max_size)
3288 rec->max_size = max_size;
3289 return ref;
3290 }
3291
3292 static int add_extent_rec(struct cache_tree *extent_cache,
3293 struct btrfs_key *parent_key, u64 parent_gen,
3294 u64 start, u64 nr, u64 extent_item_refs,
3295 int is_root, int inc_ref, int set_checked,
3296 int metadata, int extent_rec, u64 max_size)
3297 {
3298 struct extent_record *rec;
3299 struct cache_extent *cache;
3300 int ret = 0;
3301 int dup = 0;
3302
3303 cache = lookup_cache_extent(extent_cache, start, nr);
3304 if (cache) {
3305 rec = container_of(cache, struct extent_record, cache);
3306 if (inc_ref)
3307 rec->refs++;
3308 if (rec->nr == 1)
3309 rec->nr = max(nr, max_size);
3310
3311 /*
3312 * We need to make sure to reset nr to whatever the extent
3313 * record says was the real size, this way we can compare it to
3314 * the backrefs.
3315 */
3316 if (extent_rec) {
3317 if (start != rec->start || rec->found_rec) {
3318 struct extent_record *tmp;
3319
3320 dup = 1;
3321 if (list_empty(&rec->list))
3322 list_add_tail(&rec->list,
3323 &duplicate_extents);
3324
3325 /*
3326 * We have to do this song and dance in case we
3327 * find an extent record that falls inside of
3328 * our current extent record but does not have
3329 * the same objectid.
3330 */
3331 tmp = malloc(sizeof(*tmp));
3332 if (!tmp)
3333 return -ENOMEM;
3334 tmp->start = start;
3335 tmp->max_size = max_size;
3336 tmp->nr = nr;
3337 tmp->found_rec = 1;
3338 tmp->metadata = metadata;
3339 tmp->extent_item_refs = extent_item_refs;
3340 INIT_LIST_HEAD(&tmp->list);
3341 list_add_tail(&tmp->list, &rec->dups);
3342 rec->num_duplicates++;
3343 } else {
3344 rec->nr = nr;
3345 rec->found_rec = 1;
3346 }
3347 }
3348
3349 if (extent_item_refs && !dup) {
3350 if (rec->extent_item_refs) {
3351 fprintf(stderr, "block %llu rec "
3352 "extent_item_refs %llu, passed %llu\n",
3353 (unsigned long long)start,
3354 (unsigned long long)
3355 rec->extent_item_refs,
3356 (unsigned long long)extent_item_refs);
3357 }
3358 rec->extent_item_refs = extent_item_refs;
3359 }
3360 if (is_root)
3361 rec->is_root = 1;
3362 if (set_checked) {
3363 rec->content_checked = 1;
3364 rec->owner_ref_checked = 1;
3365 }
3366
3367 if (parent_key)
3368 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
3369 if (parent_gen)
3370 rec->parent_generation = parent_gen;
3371
3372 if (rec->max_size < max_size)
3373 rec->max_size = max_size;
3374
3375 maybe_free_extent_rec(extent_cache, rec);
3376 return ret;
3377 }
3378 rec = malloc(sizeof(*rec));
3379 rec->start = start;
3380 rec->max_size = max_size;
3381 rec->nr = max(nr, max_size);
3382 rec->found_rec = !!extent_rec;
3383 rec->content_checked = 0;
3384 rec->owner_ref_checked = 0;
3385 rec->num_duplicates = 0;
3386 rec->metadata = metadata;
3387 INIT_LIST_HEAD(&rec->backrefs);
3388 INIT_LIST_HEAD(&rec->dups);
3389 INIT_LIST_HEAD(&rec->list);
3390
3391 if (is_root)
3392 rec->is_root = 1;
3393 else
3394 rec->is_root = 0;
3395
3396 if (inc_ref)
3397 rec->refs = 1;
3398 else
3399 rec->refs = 0;
3400
3401 if (extent_item_refs)
3402 rec->extent_item_refs = extent_item_refs;
3403 else
3404 rec->extent_item_refs = 0;
3405
3406 if (parent_key)
3407 btrfs_cpu_key_to_disk(&rec->parent_key, parent_key);
3408 else
3409 memset(&rec->parent_key, 0, sizeof(*parent_key));
3410
3411 if (parent_gen)
3412 rec->parent_generation = parent_gen;
3413 else
3414 rec->parent_generation = 0;
3415
3416 rec->cache.start = start;
3417 rec->cache.size = nr;
3418 ret = insert_cache_extent(extent_cache, &rec->cache);
3419 BUG_ON(ret);
3420 bytes_used += nr;
3421 if (set_checked) {
3422 rec->content_checked = 1;
3423 rec->owner_ref_checked = 1;
3424 }
3425 return ret;
3426 }
3427
3428 static int add_tree_backref(struct cache_tree *extent_cache, u64 bytenr,
3429 u64 parent, u64 root, int found_ref)
3430 {
3431 struct extent_record *rec;
3432 struct tree_backref *back;
3433 struct cache_extent *cache;
3434
3435 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3436 if (!cache) {
3437 add_extent_rec(extent_cache, NULL, 0, bytenr,
3438 1, 0, 0, 0, 0, 1, 0, 0);
3439 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3440 if (!cache)
3441 abort();
3442 }
3443
3444 rec = container_of(cache, struct extent_record, cache);
3445 if (rec->start != bytenr) {
3446 abort();
3447 }
3448
3449 back = find_tree_backref(rec, parent, root);
3450 if (!back)
3451 back = alloc_tree_backref(rec, parent, root);
3452
3453 if (found_ref) {
3454 if (back->node.found_ref) {
3455 fprintf(stderr, "Extent back ref already exists "
3456 "for %llu parent %llu root %llu \n",
3457 (unsigned long long)bytenr,
3458 (unsigned long long)parent,
3459 (unsigned long long)root);
3460 }
3461 back->node.found_ref = 1;
3462 } else {
3463 if (back->node.found_extent_tree) {
3464 fprintf(stderr, "Extent back ref already exists "
3465 "for %llu parent %llu root %llu \n",
3466 (unsigned long long)bytenr,
3467 (unsigned long long)parent,
3468 (unsigned long long)root);
3469 }
3470 back->node.found_extent_tree = 1;
3471 }
3472 maybe_free_extent_rec(extent_cache, rec);
3473 return 0;
3474 }
3475
3476 static int add_data_backref(struct cache_tree *extent_cache, u64 bytenr,
3477 u64 parent, u64 root, u64 owner, u64 offset,
3478 u32 num_refs, int found_ref, u64 max_size)
3479 {
3480 struct extent_record *rec;
3481 struct data_backref *back;
3482 struct cache_extent *cache;
3483
3484 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3485 if (!cache) {
3486 add_extent_rec(extent_cache, NULL, 0, bytenr, 1, 0, 0, 0, 0,
3487 0, 0, max_size);
3488 cache = lookup_cache_extent(extent_cache, bytenr, 1);
3489 if (!cache)
3490 abort();
3491 }
3492
3493 rec = container_of(cache, struct extent_record, cache);
3494 if (rec->max_size < max_size)
3495 rec->max_size = max_size;
3496
3497 /*
3498 * If found_ref is set then max_size is the real size and must match the
3499 * existing refs. So if we have already found a ref then we need to
3500 * make sure that this ref matches the existing one, otherwise we need
3501 * to add a new backref so we can notice that the backrefs don't match
3502 * and we need to figure out who is telling the truth. This is to
3503 * account for that awful fsync bug I introduced where we'd end up with
3504 * a btrfs_file_extent_item that would have its length include multiple
3505 * prealloc extents or point inside of a prealloc extent.
3506 */
3507 back = find_data_backref(rec, parent, root, owner, offset, found_ref,
3508 bytenr, max_size);
3509 if (!back)
3510 back = alloc_data_backref(rec, parent, root, owner, offset,
3511 max_size);
3512
3513 if (found_ref) {
3514 BUG_ON(num_refs != 1);
3515 if (back->node.found_ref)
3516 BUG_ON(back->bytes != max_size);
3517 back->node.found_ref = 1;
3518 back->found_ref += 1;
3519 back->bytes = max_size;
3520 back->disk_bytenr = bytenr;
3521 rec->refs += 1;
3522 rec->content_checked = 1;
3523 rec->owner_ref_checked = 1;
3524 } else {
3525 if (back->node.found_extent_tree) {
3526 fprintf(stderr, "Extent back ref already exists "
3527 "for %llu parent %llu root %llu "
3528 "owner %llu offset %llu num_refs %lu\n",
3529 (unsigned long long)bytenr,
3530 (unsigned long long)parent,
3531 (unsigned long long)root,
3532 (unsigned long long)owner,
3533 (unsigned long long)offset,
3534 (unsigned long)num_refs);
3535 }
3536 back->num_refs = num_refs;
3537 back->node.found_extent_tree = 1;
3538 }
3539 maybe_free_extent_rec(extent_cache, rec);
3540 return 0;
3541 }
3542
3543 static int add_pending(struct cache_tree *pending,
3544 struct cache_tree *seen, u64 bytenr, u32 size)
3545 {
3546 int ret;
3547 ret = add_cache_extent(seen, bytenr, size);
3548 if (ret)
3549 return ret;
3550 add_cache_extent(pending, bytenr, size);
3551 return 0;
3552 }
3553
3554 static int pick_next_pending(struct cache_tree *pending,
3555 struct cache_tree *reada,
3556 struct cache_tree *nodes,
3557 u64 last, struct block_info *bits, int bits_nr,
3558 int *reada_bits)
3559 {
3560 unsigned long node_start = last;
3561 struct cache_extent *cache;
3562 int ret;
3563
3564 cache = search_cache_extent(reada, 0);
3565 if (cache) {
3566 bits[0].start = cache->start;
3567 bits[0].size = cache->size;
3568 *reada_bits = 1;
3569 return 1;
3570 }
3571 *reada_bits = 0;
3572 if (node_start > 32768)
3573 node_start -= 32768;
3574
3575 cache = search_cache_extent(nodes, node_start);
3576 if (!cache)
3577 cache = search_cache_extent(nodes, 0);
3578
3579 if (!cache) {
3580 cache = search_cache_extent(pending, 0);
3581 if (!cache)
3582 return 0;
3583 ret = 0;
3584 do {
3585 bits[ret].start = cache->start;
3586 bits[ret].size = cache->size;
3587 cache = next_cache_extent(cache);
3588 ret++;
3589 } while (cache && ret < bits_nr);
3590 return ret;
3591 }
3592
3593 ret = 0;
3594 do {
3595 bits[ret].start = cache->start;
3596 bits[ret].size = cache->size;
3597 cache = next_cache_extent(cache);
3598 ret++;
3599 } while (cache && ret < bits_nr);
3600
3601 if (bits_nr - ret > 8) {
3602 u64 lookup = bits[0].start + bits[0].size;
3603 struct cache_extent *next;
3604 next = search_cache_extent(pending, lookup);
3605 while(next) {
3606 if (next->start - lookup > 32768)
3607 break;
3608 bits[ret].start = next->start;
3609 bits[ret].size = next->size;
3610 lookup = next->start + next->size;
3611 ret++;
3612 if (ret == bits_nr)
3613 break;
3614 next = next_cache_extent(next);
3615 if (!next)
3616 break;
3617 }
3618 }
3619 return ret;
3620 }
3621
3622 static void free_chunk_record(struct cache_extent *cache)
3623 {
3624 struct chunk_record *rec;
3625
3626 rec = container_of(cache, struct chunk_record, cache);
3627 list_del_init(&rec->list);
3628 list_del_init(&rec->dextents);
3629 free(rec);
3630 }
3631
3632 void free_chunk_cache_tree(struct cache_tree *chunk_cache)
3633 {
3634 cache_tree_free_extents(chunk_cache, free_chunk_record);
3635 }
3636
3637 static void free_device_record(struct rb_node *node)
3638 {
3639 struct device_record *rec;
3640
3641 rec = container_of(node, struct device_record, node);
3642 free(rec);
3643 }
3644
3645 FREE_RB_BASED_TREE(device_cache, free_device_record);
3646
3647 int insert_block_group_record(struct block_group_tree *tree,
3648 struct block_group_record *bg_rec)
3649 {
3650 int ret;
3651
3652 ret = insert_cache_extent(&tree->tree, &bg_rec->cache);
3653 if (ret)
3654 return ret;
3655
3656 list_add_tail(&bg_rec->list, &tree->block_groups);
3657 return 0;
3658 }
3659
3660 static void free_block_group_record(struct cache_extent *cache)
3661 {
3662 struct block_group_record *rec;
3663
3664 rec = container_of(cache, struct block_group_record, cache);
3665 list_del_init(&rec->list);
3666 free(rec);
3667 }
3668
3669 void free_block_group_tree(struct block_group_tree *tree)
3670 {
3671 cache_tree_free_extents(&tree->tree, free_block_group_record);
3672 }
3673
3674 int insert_device_extent_record(struct device_extent_tree *tree,
3675 struct device_extent_record *de_rec)
3676 {
3677 int ret;
3678
3679 /*
3680 * Device extent is a bit different from the other extents, because
3681 * the extents which belong to the different devices may have the
3682 * same start and size, so we need use the special extent cache
3683 * search/insert functions.
3684 */
3685 ret = insert_cache_extent2(&tree->tree, &de_rec->cache);
3686 if (ret)
3687 return ret;
3688
3689 list_add_tail(&de_rec->chunk_list, &tree->no_chunk_orphans);
3690 list_add_tail(&de_rec->device_list, &tree->no_device_orphans);
3691 return 0;
3692 }
3693
3694 static void free_device_extent_record(struct cache_extent *cache)
3695 {
3696 struct device_extent_record *rec;
3697
3698 rec = container_of(cache, struct device_extent_record, cache);
3699 if (!list_empty(&rec->chunk_list))
3700 list_del_init(&rec->chunk_list);
3701 if (!list_empty(&rec->device_list))
3702 list_del_init(&rec->device_list);
3703 free(rec);
3704 }
3705
3706 void free_device_extent_tree(struct device_extent_tree *tree)
3707 {
3708 cache_tree_free_extents(&tree->tree, free_device_extent_record);
3709 }
3710
3711 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3712 static int process_extent_ref_v0(struct cache_tree *extent_cache,
3713 struct extent_buffer *leaf, int slot)
3714 {
3715 struct btrfs_extent_ref_v0 *ref0;
3716 struct btrfs_key key;
3717
3718 btrfs_item_key_to_cpu(leaf, &key, slot);
3719 ref0 = btrfs_item_ptr(leaf, slot, struct btrfs_extent_ref_v0);
3720 if (btrfs_ref_objectid_v0(leaf, ref0) < BTRFS_FIRST_FREE_OBJECTID) {
3721 add_tree_backref(extent_cache, key.objectid, key.offset, 0, 0);
3722 } else {
3723 add_data_backref(extent_cache, key.objectid, key.offset, 0,
3724 0, 0, btrfs_ref_count_v0(leaf, ref0), 0, 0);
3725 }
3726 return 0;
3727 }
3728 #endif
3729
3730 struct chunk_record *btrfs_new_chunk_record(struct extent_buffer *leaf,
3731 struct btrfs_key *key,
3732 int slot)
3733 {
3734 struct btrfs_chunk *ptr;
3735 struct chunk_record *rec;
3736 int num_stripes, i;
3737
3738 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3739 num_stripes = btrfs_chunk_num_stripes(leaf, ptr);
3740
3741 rec = malloc(btrfs_chunk_record_size(num_stripes));
3742 if (!rec) {
3743 fprintf(stderr, "memory allocation failed\n");
3744 exit(-1);
3745 }
3746
3747 memset(rec, 0, btrfs_chunk_record_size(num_stripes));
3748
3749 INIT_LIST_HEAD(&rec->list);
3750 INIT_LIST_HEAD(&rec->dextents);
3751 rec->bg_rec = NULL;
3752
3753 rec->cache.start = key->offset;
3754 rec->cache.size = btrfs_chunk_length(leaf, ptr);
3755
3756 rec->generation = btrfs_header_generation(leaf);
3757
3758 rec->objectid = key->objectid;
3759 rec->type = key->type;
3760 rec->offset = key->offset;
3761
3762 rec->length = rec->cache.size;
3763 rec->owner = btrfs_chunk_owner(leaf, ptr);
3764 rec->stripe_len = btrfs_chunk_stripe_len(leaf, ptr);
3765 rec->type_flags = btrfs_chunk_type(leaf, ptr);
3766 rec->io_width = btrfs_chunk_io_width(leaf, ptr);
3767 rec->io_align = btrfs_chunk_io_align(leaf, ptr);
3768 rec->sector_size = btrfs_chunk_sector_size(leaf, ptr);
3769 rec->num_stripes = num_stripes;
3770 rec->sub_stripes = btrfs_chunk_sub_stripes(leaf, ptr);
3771
3772 for (i = 0; i < rec->num_stripes; ++i) {
3773 rec->stripes[i].devid =
3774 btrfs_stripe_devid_nr(leaf, ptr, i);
3775 rec->stripes[i].offset =
3776 btrfs_stripe_offset_nr(leaf, ptr, i);
3777 read_extent_buffer(leaf, rec->stripes[i].dev_uuid,
3778 (unsigned long)btrfs_stripe_dev_uuid_nr(ptr, i),
3779 BTRFS_UUID_SIZE);
3780 }
3781
3782 return rec;
3783 }
3784
3785 static int process_chunk_item(struct cache_tree *chunk_cache,
3786 struct btrfs_key *key, struct extent_buffer *eb,
3787 int slot)
3788 {
3789 struct chunk_record *rec;
3790 int ret = 0;
3791
3792 rec = btrfs_new_chunk_record(eb, key, slot);
3793 ret = insert_cache_extent(chunk_cache, &rec->cache);
3794 if (ret) {
3795 fprintf(stderr, "Chunk[%llu, %llu] existed.\n",
3796 rec->offset, rec->length);
3797 free(rec);
3798 }
3799
3800 return ret;
3801 }
3802
3803 static int process_device_item(struct rb_root *dev_cache,
3804 struct btrfs_key *key, struct extent_buffer *eb, int slot)
3805 {
3806 struct btrfs_dev_item *ptr;
3807 struct device_record *rec;
3808 int ret = 0;
3809
3810 ptr = btrfs_item_ptr(eb,
3811 slot, struct btrfs_dev_item);
3812
3813 rec = malloc(sizeof(*rec));
3814 if (!rec) {
3815 fprintf(stderr, "memory allocation failed\n");
3816 return -ENOMEM;
3817 }
3818
3819 rec->devid = key->offset;
3820 rec->generation = btrfs_header_generation(eb);
3821
3822 rec->objectid = key->objectid;
3823 rec->type = key->type;
3824 rec->offset = key->offset;
3825
3826 rec->devid = btrfs_device_id(eb, ptr);
3827 rec->total_byte = btrfs_device_total_bytes(eb, ptr);
3828 rec->byte_used = btrfs_device_bytes_used(eb, ptr);
3829
3830 ret = rb_insert(dev_cache, &rec->node, device_record_compare);
3831 if (ret) {
3832 fprintf(stderr, "Device[%llu] existed.\n", rec->devid);
3833 free(rec);
3834 }
3835
3836 return ret;
3837 }
3838
3839 struct block_group_record *
3840 btrfs_new_block_group_record(struct extent_buffer *leaf, struct btrfs_key *key,
3841 int slot)
3842 {
3843 struct btrfs_block_group_item *ptr;
3844 struct block_group_record *rec;
3845
3846 rec = malloc(sizeof(*rec));
3847 if (!rec) {
3848 fprintf(stderr, "memory allocation failed\n");
3849 exit(-1);
3850 }
3851 memset(rec, 0, sizeof(*rec));
3852
3853 rec->cache.start = key->objectid;
3854 rec->cache.size = key->offset;
3855
3856 rec->generation = btrfs_header_generation(leaf);
3857
3858 rec->objectid = key->objectid;
3859 rec->type = key->type;
3860 rec->offset = key->offset;
3861
3862 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item);
3863 rec->flags = btrfs_disk_block_group_flags(leaf, ptr);
3864
3865 INIT_LIST_HEAD(&rec->list);
3866
3867 return rec;
3868 }
3869
3870 static int process_block_group_item(struct block_group_tree *block_group_cache,
3871 struct btrfs_key *key,
3872 struct extent_buffer *eb, int slot)
3873 {
3874 struct block_group_record *rec;
3875 int ret = 0;
3876
3877 rec = btrfs_new_block_group_record(eb, key, slot);
3878 ret = insert_block_group_record(block_group_cache, rec);
3879 if (ret) {
3880 fprintf(stderr, "Block Group[%llu, %llu] existed.\n",
3881 rec->objectid, rec->offset);
3882 free(rec);
3883 }
3884
3885 return ret;
3886 }
3887
3888 struct device_extent_record *
3889 btrfs_new_device_extent_record(struct extent_buffer *leaf,
3890 struct btrfs_key *key, int slot)
3891 {
3892 struct device_extent_record *rec;
3893 struct btrfs_dev_extent *ptr;
3894
3895 rec = malloc(sizeof(*rec));
3896 if (!rec) {
3897 fprintf(stderr, "memory allocation failed\n");
3898 exit(-1);
3899 }
3900 memset(rec, 0, sizeof(*rec));
3901
3902 rec->cache.objectid = key->objectid;
3903 rec->cache.start = key->offset;
3904
3905 rec->generation = btrfs_header_generation(leaf);
3906
3907 rec->objectid = key->objectid;
3908 rec->type = key->type;
3909 rec->offset = key->offset;
3910
3911 ptr = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
3912 rec->chunk_objecteid =
3913 btrfs_dev_extent_chunk_objectid(leaf, ptr);
3914 rec->chunk_offset =
3915 btrfs_dev_extent_chunk_offset(leaf, ptr);
3916 rec->length = btrfs_dev_extent_length(leaf, ptr);
3917 rec->cache.size = rec->length;
3918
3919 INIT_LIST_HEAD(&rec->chunk_list);
3920 INIT_LIST_HEAD(&rec->device_list);
3921
3922 return rec;
3923 }
3924
3925 static int
3926 process_device_extent_item(struct device_extent_tree *dev_extent_cache,
3927 struct btrfs_key *key, struct extent_buffer *eb,
3928 int slot)
3929 {
3930 struct device_extent_record *rec;
3931 int ret;
3932
3933 rec = btrfs_new_device_extent_record(eb, key, slot);
3934 ret = insert_device_extent_record(dev_extent_cache, rec);
3935 if (ret) {
3936 fprintf(stderr,
3937 "Device extent[%llu, %llu, %llu] existed.\n",
3938 rec->objectid, rec->offset, rec->length);
3939 free(rec);
3940 }
3941
3942 return ret;
3943 }
3944
3945 static int process_extent_item(struct btrfs_root *root,
3946 struct cache_tree *extent_cache,
3947 struct extent_buffer *eb, int slot)
3948 {
3949 struct btrfs_extent_item *ei;
3950 struct btrfs_extent_inline_ref *iref;
3951 struct btrfs_extent_data_ref *dref;
3952 struct btrfs_shared_data_ref *sref;
3953 struct btrfs_key key;
3954 unsigned long end;
3955 unsigned long ptr;
3956 int type;
3957 u32 item_size = btrfs_item_size_nr(eb, slot);
3958 u64 refs = 0;
3959 u64 offset;
3960 u64 num_bytes;
3961 int metadata = 0;
3962
3963 btrfs_item_key_to_cpu(eb, &key, slot);
3964
3965 if (key.type == BTRFS_METADATA_ITEM_KEY) {
3966 metadata = 1;
3967 num_bytes = root->leafsize;
3968 } else {
3969 num_bytes = key.offset;
3970 }
3971
3972 if (item_size < sizeof(*ei)) {
3973 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3974 struct btrfs_extent_item_v0 *ei0;
3975 BUG_ON(item_size != sizeof(*ei0));
3976 ei0 = btrfs_item_ptr(eb, slot, struct btrfs_extent_item_v0);
3977 refs = btrfs_extent_refs_v0(eb, ei0);
3978 #else
3979 BUG();
3980 #endif
3981 return add_extent_rec(extent_cache, NULL, 0, key.objectid,
3982 num_bytes, refs, 0, 0, 0, metadata, 1,
3983 num_bytes);
3984 }
3985
3986 ei = btrfs_item_ptr(eb, slot, struct btrfs_extent_item);
3987 refs = btrfs_extent_refs(eb, ei);
3988
3989 add_extent_rec(extent_cache, NULL, 0, key.objectid, num_bytes,
3990 refs, 0, 0, 0, metadata, 1, num_bytes);
3991
3992 ptr = (unsigned long)(ei + 1);
3993 if (btrfs_extent_flags(eb, ei) & BTRFS_EXTENT_FLAG_TREE_BLOCK &&
3994 key.type == BTRFS_EXTENT_ITEM_KEY)
3995 ptr += sizeof(struct btrfs_tree_block_info);
3996
3997 end = (unsigned long)ei + item_size;
3998 while (ptr < end) {
3999 iref = (struct btrfs_extent_inline_ref *)ptr;
4000 type = btrfs_extent_inline_ref_type(eb, iref);
4001 offset = btrfs_extent_inline_ref_offset(eb, iref);
4002 switch (type) {
4003 case BTRFS_TREE_BLOCK_REF_KEY:
4004 add_tree_backref(extent_cache, key.objectid,
4005 0, offset, 0);
4006 break;
4007 case BTRFS_SHARED_BLOCK_REF_KEY:
4008 add_tree_backref(extent_cache, key.objectid,
4009 offset, 0, 0);
4010 break;
4011 case BTRFS_EXTENT_DATA_REF_KEY:
4012 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
4013 add_data_backref(extent_cache, key.objectid, 0,
4014 btrfs_extent_data_ref_root(eb, dref),
4015 btrfs_extent_data_ref_objectid(eb,
4016 dref),
4017 btrfs_extent_data_ref_offset(eb, dref),
4018 btrfs_extent_data_ref_count(eb, dref),
4019 0, num_bytes);
4020 break;
4021 case BTRFS_SHARED_DATA_REF_KEY:
4022 sref = (struct btrfs_shared_data_ref *)(iref + 1);
4023 add_data_backref(extent_cache, key.objectid, offset,
4024 0, 0, 0,
4025 btrfs_shared_data_ref_count(eb, sref),
4026 0, num_bytes);
4027 break;
4028 default:
4029 fprintf(stderr, "corrupt extent record: key %Lu %u %Lu\n",
4030 key.objectid, key.type, num_bytes);
4031 goto out;
4032 }
4033 ptr += btrfs_extent_inline_ref_size(type);
4034 }
4035 WARN_ON(ptr > end);
4036 out:
4037 return 0;
4038 }
4039
4040 static int check_cache_range(struct btrfs_root *root,
4041 struct btrfs_block_group_cache *cache,
4042 u64 offset, u64 bytes)
4043 {
4044 struct btrfs_free_space *entry;
4045 u64 *logical;
4046 u64 bytenr;
4047 int stripe_len;
4048 int i, nr, ret;
4049
4050 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
4051 bytenr = btrfs_sb_offset(i);
4052 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
4053 cache->key.objectid, bytenr, 0,
4054 &logical, &nr, &stripe_len);
4055 if (ret)
4056 return ret;
4057
4058 while (nr--) {
4059 if (logical[nr] + stripe_len <= offset)
4060 continue;
4061 if (offset + bytes <= logical[nr])
4062 continue;
4063 if (logical[nr] == offset) {
4064 if (stripe_len >= bytes) {
4065 kfree(logical);
4066 return 0;
4067 }
4068 bytes -= stripe_len;
4069 offset += stripe_len;
4070 } else if (logical[nr] < offset) {
4071 if (logical[nr] + stripe_len >=
4072 offset + bytes) {
4073 kfree(logical);
4074 return 0;
4075 }
4076 bytes = (offset + bytes) -
4077 (logical[nr] + stripe_len);
4078 offset = logical[nr] + stripe_len;
4079 } else {
4080 /*
4081 * Could be tricky, the super may land in the
4082 * middle of the area we're checking. First
4083 * check the easiest case, it's at the end.
4084 */
4085 if (logical[nr] + stripe_len >=
4086 bytes + offset) {
4087 bytes = logical[nr] - offset;
4088 continue;
4089 }
4090
4091 /* Check the left side */
4092 ret = check_cache_range(root, cache,
4093 offset,
4094 logical[nr] - offset);
4095 if (ret) {
4096 kfree(logical);
4097 return ret;
4098 }
4099
4100 /* Now we continue with the right side */
4101 bytes = (offset + bytes) -
4102 (logical[nr] + stripe_len);
4103 offset = logical[nr] + stripe_len;
4104 }
4105 }
4106
4107 kfree(logical);
4108 }
4109
4110 entry = btrfs_find_free_space(cache->free_space_ctl, offset, bytes);
4111 if (!entry) {
4112 fprintf(stderr, "There is no free space entry for %Lu-%Lu\n",
4113 offset, offset+bytes);
4114 return -EINVAL;
4115 }
4116
4117 if (entry->offset != offset) {
4118 fprintf(stderr, "Wanted offset %Lu, found %Lu\n", offset,
4119 entry->offset);
4120 return -EINVAL;
4121 }
4122
4123 if (entry->bytes != bytes) {
4124 fprintf(stderr, "Wanted bytes %Lu, found %Lu for off %Lu\n",
4125 bytes, entry->bytes, offset);
4126 return -EINVAL;
4127 }
4128
4129 unlink_free_space(cache->free_space_ctl, entry);
4130 free(entry);
4131 return 0;
4132 }
4133
4134 static int verify_space_cache(struct btrfs_root *root,
4135 struct btrfs_block_group_cache *cache)
4136 {
4137 struct btrfs_path *path;
4138 struct extent_buffer *leaf;
4139 struct btrfs_key key;
4140 u64 last;
4141 int ret = 0;
4142
4143 path = btrfs_alloc_path();
4144 if (!path)
4145 return -ENOMEM;
4146
4147 root = root->fs_info->extent_root;
4148
4149 last = max_t(u64, cache->key.objectid, BTRFS_SUPER_INFO_OFFSET);
4150
4151 key.objectid = last;
4152 key.offset = 0;
4153 key.type = BTRFS_EXTENT_ITEM_KEY;
4154
4155 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4156 if (ret < 0)
4157 goto out;
4158 ret = 0;
4159 while (1) {
4160 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4161 ret = btrfs_next_leaf(root, path);
4162 if (ret < 0)
4163 goto out;
4164 if (ret > 0) {
4165 ret = 0;
4166 break;
4167 }
4168 }
4169 leaf = path->nodes[0];
4170 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4171 if (key.objectid >= cache->key.offset + cache->key.objectid)
4172 break;
4173 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
4174 key.type != BTRFS_METADATA_ITEM_KEY) {
4175 path->slots[0]++;
4176 continue;
4177 }
4178
4179 if (last == key.objectid) {
4180 if (key.type == BTRFS_EXTENT_ITEM_KEY)
4181 last = key.objectid + key.offset;
4182 else
4183 last = key.objectid + root->leafsize;
4184 path->slots[0]++;
4185 continue;
4186 }
4187
4188 ret = check_cache_range(root, cache, last,
4189 key.objectid - last);
4190 if (ret)
4191 break;
4192 if (key.type == BTRFS_EXTENT_ITEM_KEY)
4193 last = key.objectid + key.offset;
4194 else
4195 last = key.objectid + root->leafsize;
4196 path->slots[0]++;
4197 }
4198
4199 if (last < cache->key.objectid + cache->key.offset)
4200 ret = check_cache_range(root, cache, last,
4201 cache->key.objectid +
4202 cache->key.offset - last);
4203
4204 out:
4205 btrfs_free_path(path);
4206
4207 if (!ret &&
4208 !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
4209 fprintf(stderr, "There are still entries left in the space "
4210 "cache\n");
4211 ret = -EINVAL;
4212 }
4213
4214 return ret;
4215 }
4216
4217 static int check_space_cache(struct btrfs_root *root)
4218 {
4219 struct btrfs_block_group_cache *cache;
4220 u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
4221 int ret;
4222 int error = 0;
4223
4224 if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
4225 btrfs_super_generation(root->fs_info->super_copy) !=
4226 btrfs_super_cache_generation(root->fs_info->super_copy)) {
4227 printf("cache and super generation don't match, space cache "
4228 "will be invalidated\n");
4229 return 0;
4230 }
4231
4232 while (1) {
4233 cache = btrfs_lookup_first_block_group(root->fs_info, start);
4234 if (!cache)
4235 break;
4236
4237 start = cache->key.objectid + cache->key.offset;
4238 if (!cache->free_space_ctl) {
4239 if (btrfs_init_free_space_ctl(cache,
4240 root->sectorsize)) {
4241 ret = -ENOMEM;
4242 break;
4243 }
4244 } else {
4245 btrfs_remove_free_space_cache(cache);
4246 }
4247
4248 ret = load_free_space_cache(root->fs_info, cache);
4249 if (!ret)
4250 continue;
4251
4252 ret = verify_space_cache(root, cache);
4253 if (ret) {
4254 fprintf(stderr, "cache appears valid but isnt %Lu\n",
4255 cache->key.objectid);
4256 error++;
4257 }
4258 }
4259
4260 return error ? -EINVAL : 0;
4261 }
4262
4263 static int read_extent_data(struct btrfs_root *root, char *data,
4264 u64 logical, u64 *len, int mirror)
4265 {
4266 u64 offset = 0;
4267 struct btrfs_multi_bio *multi = NULL;
4268 struct btrfs_fs_info *info = root->fs_info;
4269 struct btrfs_device *device;
4270 int ret = 0;
4271 u64 max_len = *len;
4272
4273 ret = btrfs_map_block(&info->mapping_tree, READ, logical, len,
4274 &multi, mirror, NULL);
4275 if (ret) {
4276 fprintf(stderr, "Couldn't map the block %llu\n",
4277 logical + offset);
4278 goto err;
4279 }
4280 device = multi->stripes[0].dev;
4281
4282 if (device->fd == 0)
4283 goto err;
4284 if (*len > max_len)
4285 *len = max_len;
4286
4287 ret = pread64(device->fd, data, *len, multi->stripes[0].physical);
4288 if (ret != *len)
4289 ret = -EIO;
4290 else
4291 ret = 0;
4292 err:
4293 kfree(multi);
4294 return ret;
4295 }
4296
4297 static int check_extent_csums(struct btrfs_root *root, u64 bytenr,
4298 u64 num_bytes, unsigned long leaf_offset,
4299 struct extent_buffer *eb) {
4300
4301 u64 offset = 0;
4302 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
4303 char *data;
4304 unsigned long csum_offset;
4305 u32 csum;
4306 u32 csum_expected;
4307 u64 read_len;
4308 u64 data_checked = 0;
4309 u64 tmp;
4310 int ret = 0;
4311 int mirror;
4312 int num_copies;
4313
4314 if (num_bytes % root->sectorsize)
4315 return -EINVAL;
4316
4317 data = malloc(num_bytes);
4318 if (!data)
4319 return -ENOMEM;
4320
4321 while (offset < num_bytes) {
4322 mirror = 0;
4323 again:
4324 read_len = num_bytes - offset;
4325 /* read as much space once a time */
4326 ret = read_extent_data(root, data + offset,
4327 bytenr + offset, &read_len, mirror);
4328 if (ret)
4329 goto out;
4330 data_checked = 0;
4331 /* verify every 4k data's checksum */
4332 while (data_checked < read_len) {
4333 csum = ~(u32)0;
4334 tmp = offset + data_checked;
4335
4336 csum = btrfs_csum_data(NULL, (char *)data + tmp,
4337 csum, root->sectorsize);
4338 btrfs_csum_final(csum, (char *)&csum);
4339
4340 csum_offset = leaf_offset +
4341 tmp / root->sectorsize * csum_size;
4342 read_extent_buffer(eb, (char *)&csum_expected,
4343 csum_offset, csum_size);
4344 /* try another mirror */
4345 if (csum != csum_expected) {
4346 fprintf(stderr, "mirror %d bytenr %llu csum %u expected csum %u\n",
4347 mirror, bytenr + tmp,
4348 csum, csum_expected);
4349 num_copies = btrfs_num_copies(
4350 &root->fs_info->mapping_tree,
4351 bytenr, num_bytes);
4352 if (mirror < num_copies - 1) {
4353 mirror += 1;
4354 goto again;
4355 }
4356 }
4357 data_checked += root->sectorsize;
4358 }
4359 offset += read_len;
4360 }
4361 out:
4362 free(data);
4363 return ret;
4364 }
4365
4366 static int check_extent_exists(struct btrfs_root *root, u64 bytenr,
4367 u64 num_bytes)
4368 {
4369 struct btrfs_path *path;
4370 struct extent_buffer *leaf;
4371 struct btrfs_key key;
4372 int ret;
4373
4374 path = btrfs_alloc_path();
4375 if (!path) {
4376 fprintf(stderr, "Error allocing path\n");
4377 return -ENOMEM;
4378 }
4379
4380 key.objectid = bytenr;
4381 key.type = BTRFS_EXTENT_ITEM_KEY;
4382 key.offset = (u64)-1;
4383
4384 again:
4385 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
4386 0, 0);
4387 if (ret < 0) {
4388 fprintf(stderr, "Error looking up extent record %d\n", ret);
4389 btrfs_free_path(path);
4390 return ret;
4391 } else if (ret) {
4392 if (path->slots[0] > 0) {
4393 path->slots[0]--;
4394 } else {
4395 ret = btrfs_prev_leaf(root, path);
4396 if (ret < 0) {
4397 goto out;
4398 } else if (ret > 0) {
4399 ret = 0;
4400 goto out;
4401 }
4402 }
4403 }
4404
4405 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4406
4407 /*
4408 * Block group items come before extent items if they have the same
4409 * bytenr, so walk back one more just in case. Dear future traveler,
4410 * first congrats on mastering time travel. Now if it's not too much
4411 * trouble could you go back to 2006 and tell Chris to make the
4412 * BLOCK_GROUP_ITEM_KEY (and BTRFS_*_REF_KEY) lower than the
4413 * EXTENT_ITEM_KEY please?
4414 */
4415 while (key.type > BTRFS_EXTENT_ITEM_KEY) {
4416 if (path->slots[0] > 0) {
4417 path->slots[0]--;
4418 } else {
4419 ret = btrfs_prev_leaf(root, path);
4420 if (ret < 0) {
4421 goto out;
4422 } else if (ret > 0) {
4423 ret = 0;
4424 goto out;
4425 }
4426 }
4427 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4428 }
4429
4430 while (num_bytes) {
4431 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4432 ret = btrfs_next_leaf(root, path);
4433 if (ret < 0) {
4434 fprintf(stderr, "Error going to next leaf "
4435 "%d\n", ret);
4436 btrfs_free_path(path);
4437 return ret;
4438 } else if (ret) {
4439 break;
4440 }
4441 }
4442 leaf = path->nodes[0];
4443 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4444 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
4445 path->slots[0]++;
4446 continue;
4447 }
4448 if (key.objectid + key.offset < bytenr) {
4449 path->slots[0]++;
4450 continue;
4451 }
4452 if (key.objectid > bytenr + num_bytes)
4453 break;
4454
4455 if (key.objectid == bytenr) {
4456 if (key.offset >= num_bytes) {
4457 num_bytes = 0;
4458 break;
4459 }
4460 num_bytes -= key.offset;
4461 bytenr += key.offset;
4462 } else if (key.objectid < bytenr) {
4463 if (key.objectid + key.offset >= bytenr + num_bytes) {
4464 num_bytes = 0;
4465 break;
4466 }
4467 num_bytes = (bytenr + num_bytes) -
4468 (key.objectid + key.offset);
4469 bytenr = key.objectid + key.offset;
4470 } else {
4471 if (key.objectid + key.offset < bytenr + num_bytes) {
4472 u64 new_start = key.objectid + key.offset;
4473 u64 new_bytes = bytenr + num_bytes - new_start;
4474
4475 /*
4476 * Weird case, the extent is in the middle of
4477 * our range, we'll have to search one side
4478 * and then the other. Not sure if this happens
4479 * in real life, but no harm in coding it up
4480 * anyway just in case.
4481 */
4482 btrfs_release_path(path);
4483 ret = check_extent_exists(root, new_start,
4484 new_bytes);
4485 if (ret) {
4486 fprintf(stderr, "Right section didn't "
4487 "have a record\n");
4488 break;
4489 }
4490 num_bytes = key.objectid - bytenr;
4491 goto again;
4492 }
4493 num_bytes = key.objectid - bytenr;
4494 }
4495 path->slots[0]++;
4496 }
4497 ret = 0;
4498
4499 out:
4500 if (num_bytes && !ret) {
4501 fprintf(stderr, "There are no extents for csum range "
4502 "%Lu-%Lu\n", bytenr, bytenr+num_bytes);
4503 ret = 1;
4504 }
4505
4506 btrfs_free_path(path);
4507 return ret;
4508 }
4509
4510 static int check_csums(struct btrfs_root *root)
4511 {
4512 struct btrfs_path *path;
4513 struct extent_buffer *leaf;
4514 struct btrfs_key key;
4515 u64 offset = 0, num_bytes = 0;
4516 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
4517 int errors = 0;
4518 int ret;
4519 u64 data_len;
4520 unsigned long leaf_offset;
4521
4522 root = root->fs_info->csum_root;
4523 if (!extent_buffer_uptodate(root->node)) {
4524 fprintf(stderr, "No valid csum tree found\n");
4525 return -ENOENT;
4526 }
4527
4528 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
4529 key.type = BTRFS_EXTENT_CSUM_KEY;
4530 key.offset = 0;
4531
4532 path = btrfs_alloc_path();
4533 if (!path)
4534 return -ENOMEM;
4535
4536 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4537 if (ret < 0) {
4538 fprintf(stderr, "Error searching csum tree %d\n", ret);
4539 btrfs_free_path(path);
4540 return ret;
4541 }
4542
4543 if (ret > 0 && path->slots[0])
4544 path->slots[0]--;
4545 ret = 0;
4546
4547 while (1) {
4548 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4549 ret = btrfs_next_leaf(root, path);
4550 if (ret < 0) {
4551 fprintf(stderr, "Error going to next leaf "
4552 "%d\n", ret);
4553 break;
4554 }
4555 if (ret)
4556 break;
4557 }
4558 leaf = path->nodes[0];
4559
4560 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4561 if (key.type != BTRFS_EXTENT_CSUM_KEY) {
4562 path->slots[0]++;
4563 continue;
4564 }
4565
4566 data_len = (btrfs_item_size_nr(leaf, path->slots[0]) /
4567 csum_size) * root->sectorsize;
4568 if (!check_data_csum)
4569 goto skip_csum_check;
4570 leaf_offset = btrfs_item_ptr_offset(leaf, path->slots[0]);
4571 ret = check_extent_csums(root, key.offset, data_len,
4572 leaf_offset, leaf);
4573 if (ret)
4574 break;
4575 skip_csum_check:
4576 if (!num_bytes) {
4577 offset = key.offset;
4578 } else if (key.offset != offset + num_bytes) {
4579 ret = check_extent_exists(root, offset, num_bytes);
4580 if (ret) {
4581 fprintf(stderr, "Csum exists for %Lu-%Lu but "
4582 "there is no extent record\n",
4583 offset, offset+num_bytes);
4584 errors++;
4585 }
4586 offset = key.offset;
4587 num_bytes = 0;
4588 }
4589 num_bytes += data_len;
4590 path->slots[0]++;
4591 }
4592
4593 btrfs_free_path(path);
4594 return errors;
4595 }
4596
4597 static int is_dropped_key(struct btrfs_key *key,
4598 struct btrfs_key *drop_key) {
4599 if (key->objectid < drop_key->objectid)
4600 return 1;
4601 else if (key->objectid == drop_key->objectid) {
4602 if (key->type < drop_key->type)
4603 return 1;
4604 else if (key->type == drop_key->type) {
4605 if (key->offset < drop_key->offset)
4606 return 1;
4607 }
4608 }
4609 return 0;
4610 }
4611
4612 static int run_next_block(struct btrfs_trans_handle *trans,
4613 struct btrfs_root *root,
4614 struct block_info *bits,
4615 int bits_nr,
4616 u64 *last,
4617 struct cache_tree *pending,
4618 struct cache_tree *seen,
4619 struct cache_tree *reada,
4620 struct cache_tree *nodes,
4621 struct cache_tree *extent_cache,
4622 struct cache_tree *chunk_cache,
4623 struct rb_root *dev_cache,
4624 struct block_group_tree *block_group_cache,
4625 struct device_extent_tree *dev_extent_cache,
4626 struct btrfs_root_item *ri)
4627 {
4628 struct extent_buffer *buf;
4629 u64 bytenr;
4630 u32 size;
4631 u64 parent;
4632 u64 owner;
4633 u64 flags;
4634 u64 ptr;
4635 u64 gen = 0;
4636 int ret = 0;
4637 int i;
4638 int nritems;
4639 struct btrfs_key key;
4640 struct cache_extent *cache;
4641 int reada_bits;
4642
4643 nritems = pick_next_pending(pending, reada, nodes, *last, bits,
4644 bits_nr, &reada_bits);
4645 if (nritems == 0)
4646 return 1;
4647
4648 if (!reada_bits) {
4649 for(i = 0; i < nritems; i++) {
4650 ret = add_cache_extent(reada, bits[i].start,
4651 bits[i].size);
4652 if (ret == -EEXIST)
4653 continue;
4654
4655 /* fixme, get the parent transid */
4656 readahead_tree_block(root, bits[i].start,
4657 bits[i].size, 0);
4658 }
4659 }
4660 *last = bits[0].start;
4661 bytenr = bits[0].start;
4662 size = bits[0].size;
4663
4664 cache = lookup_cache_extent(pending, bytenr, size);
4665 if (cache) {
4666 remove_cache_extent(pending, cache);
4667 free(cache);
4668 }
4669 cache = lookup_cache_extent(reada, bytenr, size);
4670 if (cache) {
4671 remove_cache_extent(reada, cache);
4672 free(cache);
4673 }
4674 cache = lookup_cache_extent(nodes, bytenr, size);
4675 if (cache) {
4676 remove_cache_extent(nodes, cache);
4677 free(cache);
4678 }
4679 cache = lookup_cache_extent(extent_cache, bytenr, size);
4680 if (cache) {
4681 struct extent_record *rec;
4682
4683 rec = container_of(cache, struct extent_record, cache);
4684 gen = rec->parent_generation;
4685 }
4686
4687 /* fixme, get the real parent transid */
4688 buf = read_tree_block(root, bytenr, size, gen);
4689 if (!extent_buffer_uptodate(buf)) {
4690 record_bad_block_io(root->fs_info,
4691 extent_cache, bytenr, size);
4692 goto out;
4693 }
4694
4695 nritems = btrfs_header_nritems(buf);
4696
4697 /*
4698 * FIXME, this only works only if we don't have any full
4699 * backref mode.
4700 */
4701 if (!init_extent_tree) {
4702 ret = btrfs_lookup_extent_info(NULL, root, bytenr,
4703 btrfs_header_level(buf), 1, NULL,
4704 &flags);
4705 if (ret < 0)
4706 flags = 0;
4707 } else {
4708 flags = 0;
4709 }
4710
4711 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
4712 parent = bytenr;
4713 owner = 0;
4714 } else {
4715 parent = 0;
4716 owner = btrfs_header_owner(buf);
4717 }
4718
4719 ret = check_block(trans, root, extent_cache, buf, flags);
4720 if (ret)
4721 goto out;
4722
4723 if (btrfs_is_leaf(buf)) {
4724 btree_space_waste += btrfs_leaf_free_space(root, buf);
4725 for (i = 0; i < nritems; i++) {
4726 struct btrfs_file_extent_item *fi;
4727 btrfs_item_key_to_cpu(buf, &key, i);
4728 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
4729 process_extent_item(root, extent_cache, buf,
4730 i);
4731 continue;
4732 }
4733 if (key.type == BTRFS_METADATA_ITEM_KEY) {
4734 process_extent_item(root, extent_cache, buf,
4735 i);
4736 continue;
4737 }
4738 if (key.type == BTRFS_EXTENT_CSUM_KEY) {
4739 total_csum_bytes +=
4740 btrfs_item_size_nr(buf, i);
4741 continue;
4742 }
4743 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4744 process_chunk_item(chunk_cache, &key, buf, i);
4745 continue;
4746 }
4747 if (key.type == BTRFS_DEV_ITEM_KEY) {
4748 process_device_item(dev_cache, &key, buf, i);
4749 continue;
4750 }
4751 if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
4752 process_block_group_item(block_group_cache,
4753 &key, buf, i);
4754 continue;
4755 }
4756 if (key.type == BTRFS_DEV_EXTENT_KEY) {
4757 process_device_extent_item(dev_extent_cache,
4758 &key, buf, i);
4759 continue;
4760
4761 }
4762 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
4763 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4764 process_extent_ref_v0(extent_cache, buf, i);
4765 #else
4766 BUG();
4767 #endif
4768 continue;
4769 }
4770
4771 if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
4772 add_tree_backref(extent_cache, key.objectid, 0,
4773 key.offset, 0);
4774 continue;
4775 }
4776 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
4777 add_tree_backref(extent_cache, key.objectid,
4778 key.offset, 0, 0);
4779 continue;
4780 }
4781 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
4782 struct btrfs_extent_data_ref *ref;
4783 ref = btrfs_item_ptr(buf, i,
4784 struct btrfs_extent_data_ref);
4785 add_data_backref(extent_cache,
4786 key.objectid, 0,
4787 btrfs_extent_data_ref_root(buf, ref),
4788 btrfs_extent_data_ref_objectid(buf,
4789 ref),
4790 btrfs_extent_data_ref_offset(buf, ref),
4791 btrfs_extent_data_ref_count(buf, ref),
4792 0, root->sectorsize);
4793 continue;
4794 }
4795 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
4796 struct btrfs_shared_data_ref *ref;
4797 ref = btrfs_item_ptr(buf, i,
4798 struct btrfs_shared_data_ref);
4799 add_data_backref(extent_cache,
4800 key.objectid, key.offset, 0, 0, 0,
4801 btrfs_shared_data_ref_count(buf, ref),
4802 0, root->sectorsize);
4803 continue;
4804 }
4805 if (key.type == BTRFS_ORPHAN_ITEM_KEY) {
4806 struct bad_item *bad;
4807
4808 if (key.objectid == BTRFS_ORPHAN_OBJECTID)
4809 continue;
4810 if (!owner)
4811 continue;
4812 bad = malloc(sizeof(struct bad_item));
4813 if (!bad)
4814 continue;
4815 INIT_LIST_HEAD(&bad->list);
4816 memcpy(&bad->key, &key,
4817 sizeof(struct btrfs_key));
4818 bad->root_id = owner;
4819 list_add_tail(&bad->list, &delete_items);
4820 continue;
4821 }
4822 if (key.type != BTRFS_EXTENT_DATA_KEY)
4823 continue;
4824 fi = btrfs_item_ptr(buf, i,
4825 struct btrfs_file_extent_item);
4826 if (btrfs_file_extent_type(buf, fi) ==
4827 BTRFS_FILE_EXTENT_INLINE)
4828 continue;
4829 if (btrfs_file_extent_disk_bytenr(buf, fi) == 0)
4830 continue;
4831
4832 data_bytes_allocated +=
4833 btrfs_file_extent_disk_num_bytes(buf, fi);
4834 if (data_bytes_allocated < root->sectorsize) {
4835 abort();
4836 }
4837 data_bytes_referenced +=
4838 btrfs_file_extent_num_bytes(buf, fi);
4839 add_data_backref(extent_cache,
4840 btrfs_file_extent_disk_bytenr(buf, fi),
4841 parent, owner, key.objectid, key.offset -
4842 btrfs_file_extent_offset(buf, fi), 1, 1,
4843 btrfs_file_extent_disk_num_bytes(buf, fi));
4844 }
4845 } else {
4846 int level;
4847 struct btrfs_key first_key;
4848
4849 first_key.objectid = 0;
4850
4851 if (nritems > 0)
4852 btrfs_item_key_to_cpu(buf, &first_key, 0);
4853 level = btrfs_header_level(buf);
4854 for (i = 0; i < nritems; i++) {
4855 ptr = btrfs_node_blockptr(buf, i);
4856 size = btrfs_level_size(root, level - 1);
4857 btrfs_node_key_to_cpu(buf, &key, i);
4858 if (ri != NULL) {
4859 struct btrfs_key drop_key;
4860 btrfs_disk_key_to_cpu(&drop_key,
4861 &ri->drop_progress);
4862 if ((level == ri->drop_level)
4863 && is_dropped_key(&key, &drop_key)) {
4864 continue;
4865 }
4866 }
4867 ret = add_extent_rec(extent_cache, &key,
4868 btrfs_node_ptr_generation(buf, i),
4869 ptr, size, 0, 0, 1, 0, 1, 0,
4870 size);
4871 BUG_ON(ret);
4872
4873 add_tree_backref(extent_cache, ptr, parent, owner, 1);
4874
4875 if (level > 1) {
4876 add_pending(nodes, seen, ptr, size);
4877 } else {
4878 add_pending(pending, seen, ptr, size);
4879 }
4880 }
4881 btree_space_waste += (BTRFS_NODEPTRS_PER_BLOCK(root) -
4882 nritems) * sizeof(struct btrfs_key_ptr);
4883 }
4884 total_btree_bytes += buf->len;
4885 if (fs_root_objectid(btrfs_header_owner(buf)))
4886 total_fs_tree_bytes += buf->len;
4887 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID)
4888 total_extent_tree_bytes += buf->len;
4889 if (!found_old_backref &&
4890 btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID &&
4891 btrfs_header_backref_rev(buf) == BTRFS_MIXED_BACKREF_REV &&
4892 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))
4893 found_old_backref = 1;
4894 out:
4895 free_extent_buffer(buf);
4896 return ret;
4897 }
4898
4899 static int add_root_to_pending(struct extent_buffer *buf,
4900 struct cache_tree *extent_cache,
4901 struct cache_tree *pending,
4902 struct cache_tree *seen,
4903 struct cache_tree *nodes,
4904 struct btrfs_key *root_key)
4905 {
4906 if (btrfs_header_level(buf) > 0)
4907 add_pending(nodes, seen, buf->start, buf->len);
4908 else
4909 add_pending(pending, seen, buf->start, buf->len);
4910 add_extent_rec(extent_cache, NULL, 0, buf->start, buf->len,
4911 0, 1, 1, 0, 1, 0, buf->len);
4912
4913 if (root_key->objectid == BTRFS_TREE_RELOC_OBJECTID ||
4914 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
4915 add_tree_backref(extent_cache, buf->start, buf->start,
4916 0, 1);
4917 else
4918 add_tree_backref(extent_cache, buf->start, 0,
4919 root_key->objectid, 1);
4920 return 0;
4921 }
4922
4923 /* as we fix the tree, we might be deleting blocks that
4924 * we're tracking for repair. This hook makes sure we
4925 * remove any backrefs for blocks as we are fixing them.
4926 */
4927 static int free_extent_hook(struct btrfs_trans_handle *trans,
4928 struct btrfs_root *root,
4929 u64 bytenr, u64 num_bytes, u64 parent,
4930 u64 root_objectid, u64 owner, u64 offset,
4931 int refs_to_drop)
4932 {
4933 struct extent_record *rec;
4934 struct cache_extent *cache;
4935 int is_data;
4936 struct cache_tree *extent_cache = root->fs_info->fsck_extent_cache;
4937
4938 is_data = owner >= BTRFS_FIRST_FREE_OBJECTID;
4939 cache = lookup_cache_extent(extent_cache, bytenr, num_bytes);
4940 if (!cache)
4941 return 0;
4942
4943 rec = container_of(cache, struct extent_record, cache);
4944 if (is_data) {
4945 struct data_backref *back;
4946 back = find_data_backref(rec, parent, root_objectid, owner,
4947 offset, 1, bytenr, num_bytes);
4948 if (!back)
4949 goto out;
4950 if (back->node.found_ref) {
4951 back->found_ref -= refs_to_drop;
4952 if (rec->refs)
4953 rec->refs -= refs_to_drop;
4954 }
4955 if (back->node.found_extent_tree) {
4956 back->num_refs -= refs_to_drop;
4957 if (rec->extent_item_refs)
4958 rec->extent_item_refs -= refs_to_drop;
4959 }
4960 if (back->found_ref == 0)
4961 back->node.found_ref = 0;
4962 if (back->num_refs == 0)
4963 back->node.found_extent_tree = 0;
4964
4965 if (!back->node.found_extent_tree && back->node.found_ref) {
4966 list_del(&back->node.list);
4967 free(back);
4968 }
4969 } else {
4970 struct tree_backref *back;
4971 back = find_tree_backref(rec, parent, root_objectid);
4972 if (!back)
4973 goto out;
4974 if (back->node.found_ref) {
4975 if (rec->refs)
4976 rec->refs--;
4977 back->node.found_ref = 0;
4978 }
4979 if (back->node.found_extent_tree) {
4980 if (rec->extent_item_refs)
4981 rec->extent_item_refs--;
4982 back->node.found_extent_tree = 0;
4983 }
4984 if (!back->node.found_extent_tree && back->node.found_ref) {
4985 list_del(&back->node.list);
4986 free(back);
4987 }
4988 }
4989 maybe_free_extent_rec(extent_cache, rec);
4990 out:
4991 return 0;
4992 }
4993
4994 static int delete_extent_records(struct btrfs_trans_handle *trans,
4995 struct btrfs_root *root,
4996 struct btrfs_path *path,
4997 u64 bytenr, u64 new_len)
4998 {
4999 struct btrfs_key key;
5000 struct btrfs_key found_key;
5001 struct extent_buffer *leaf;
5002 int ret;
5003 int slot;
5004
5005
5006 key.objectid = bytenr;
5007 key.type = (u8)-1;
5008 key.offset = (u64)-1;
5009
5010 while(1) {
5011 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
5012 &key, path, 0, 1);
5013 if (ret < 0)
5014 break;
5015
5016 if (ret > 0) {
5017 ret = 0;
5018 if (path->slots[0] == 0)
5019 break;
5020 path->slots[0]--;
5021 }
5022 ret = 0;
5023
5024 leaf = path->nodes[0];
5025 slot = path->slots[0];
5026
5027 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5028 if (found_key.objectid != bytenr)
5029 break;
5030
5031 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
5032 found_key.type != BTRFS_METADATA_ITEM_KEY &&
5033 found_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
5034 found_key.type != BTRFS_EXTENT_DATA_REF_KEY &&
5035 found_key.type != BTRFS_EXTENT_REF_V0_KEY &&
5036 found_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
5037 found_key.type != BTRFS_SHARED_DATA_REF_KEY) {
5038 btrfs_release_path(path);
5039 if (found_key.type == 0) {
5040 if (found_key.offset == 0)
5041 break;
5042 key.offset = found_key.offset - 1;
5043 key.type = found_key.type;
5044 }
5045 key.type = found_key.type - 1;
5046 key.offset = (u64)-1;
5047 continue;
5048 }
5049
5050 fprintf(stderr, "repair deleting extent record: key %Lu %u %Lu\n",
5051 found_key.objectid, found_key.type, found_key.offset);
5052
5053 ret = btrfs_del_item(trans, root->fs_info->extent_root, path);
5054 if (ret)
5055 break;
5056 btrfs_release_path(path);
5057
5058 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
5059 found_key.type == BTRFS_METADATA_ITEM_KEY) {
5060 u64 bytes = (found_key.type == BTRFS_EXTENT_ITEM_KEY) ?
5061 found_key.offset : root->leafsize;
5062
5063 ret = btrfs_update_block_group(trans, root, bytenr,
5064 bytes, 0, 0);
5065 if (ret)
5066 break;
5067 }
5068 }
5069
5070 btrfs_release_path(path);
5071 return ret;
5072 }
5073
5074 /*
5075 * for a single backref, this will allocate a new extent
5076 * and add the backref to it.
5077 */
5078 static int record_extent(struct btrfs_trans_handle *trans,
5079 struct btrfs_fs_info *info,
5080 struct btrfs_path *path,
5081 struct extent_record *rec,
5082 struct extent_backref *back,
5083 int allocated, u64 flags)
5084 {
5085 int ret;
5086 struct btrfs_root *extent_root = info->extent_root;
5087 struct extent_buffer *leaf;
5088 struct btrfs_key ins_key;
5089 struct btrfs_extent_item *ei;
5090 struct tree_backref *tback;
5091 struct data_backref *dback;
5092 struct btrfs_tree_block_info *bi;
5093
5094 if (!back->is_data)
5095 rec->max_size = max_t(u64, rec->max_size,
5096 info->extent_root->leafsize);
5097
5098 if (!allocated) {
5099 u32 item_size = sizeof(*ei);
5100
5101 if (!back->is_data)
5102 item_size += sizeof(*bi);
5103
5104 ins_key.objectid = rec->start;
5105 ins_key.offset = rec->max_size;
5106 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
5107
5108 ret = btrfs_insert_empty_item(trans, extent_root, path,
5109 &ins_key, item_size);
5110 if (ret)
5111 goto fail;
5112
5113 leaf = path->nodes[0];
5114 ei = btrfs_item_ptr(leaf, path->slots[0],
5115 struct btrfs_extent_item);
5116
5117 btrfs_set_extent_refs(leaf, ei, 0);
5118 btrfs_set_extent_generation(leaf, ei, rec->generation);
5119
5120 if (back->is_data) {
5121 btrfs_set_extent_flags(leaf, ei,
5122 BTRFS_EXTENT_FLAG_DATA);
5123 } else {
5124 struct btrfs_disk_key copy_key;;
5125
5126 tback = (struct tree_backref *)back;
5127 bi = (struct btrfs_tree_block_info *)(ei + 1);
5128 memset_extent_buffer(leaf, 0, (unsigned long)bi,
5129 sizeof(*bi));
5130
5131 btrfs_set_disk_key_objectid(&copy_key,
5132 rec->info_objectid);
5133 btrfs_set_disk_key_type(&copy_key, 0);
5134 btrfs_set_disk_key_offset(&copy_key, 0);
5135
5136 btrfs_set_tree_block_level(leaf, bi, rec->info_level);
5137 btrfs_set_tree_block_key(leaf, bi, &copy_key);
5138
5139 btrfs_set_extent_flags(leaf, ei,
5140 BTRFS_EXTENT_FLAG_TREE_BLOCK | flags);
5141 }
5142
5143 btrfs_mark_buffer_dirty(leaf);
5144 ret = btrfs_update_block_group(trans, extent_root, rec->start,
5145 rec->max_size, 1, 0);
5146 if (ret)
5147 goto fail;
5148 btrfs_release_path(path);
5149 }
5150
5151 if (back->is_data) {
5152 u64 parent;
5153 int i;
5154
5155 dback = (struct data_backref *)back;
5156 if (back->full_backref)
5157 parent = dback->parent;
5158 else
5159 parent = 0;
5160
5161 for (i = 0; i < dback->found_ref; i++) {
5162 /* if parent != 0, we're doing a full backref
5163 * passing BTRFS_FIRST_FREE_OBJECTID as the owner
5164 * just makes the backref allocator create a data
5165 * backref
5166 */
5167 ret = btrfs_inc_extent_ref(trans, info->extent_root,
5168 rec->start, rec->max_size,
5169 parent,
5170 dback->root,
5171 parent ?
5172 BTRFS_FIRST_FREE_OBJECTID :
5173 dback->owner,
5174 dback->offset);
5175 if (ret)
5176 break;
5177 }
5178 fprintf(stderr, "adding new data backref"
5179 " on %llu %s %llu owner %llu"
5180 " offset %llu found %d\n",
5181 (unsigned long long)rec->start,
5182 back->full_backref ?
5183 "parent" : "root",
5184 back->full_backref ?
5185 (unsigned long long)parent :
5186 (unsigned long long)dback->root,
5187 (unsigned long long)dback->owner,
5188 (unsigned long long)dback->offset,
5189 dback->found_ref);
5190 } else {
5191 u64 parent;
5192
5193 tback = (struct tree_backref *)back;
5194 if (back->full_backref)
5195 parent = tback->parent;
5196 else
5197 parent = 0;
5198
5199 ret = btrfs_inc_extent_ref(trans, info->extent_root,
5200 rec->start, rec->max_size,
5201 parent, tback->root, 0, 0);
5202 fprintf(stderr, "adding new tree backref on "
5203 "start %llu len %llu parent %llu root %llu\n",
5204 rec->start, rec->max_size, tback->parent, tback->root);
5205 }
5206 if (ret)
5207 goto fail;
5208 fail:
5209 btrfs_release_path(path);
5210 return ret;
5211 }
5212
5213 struct extent_entry {
5214 u64 bytenr;
5215 u64 bytes;
5216 int count;
5217 int broken;
5218 struct list_head list;
5219 };
5220
5221 static struct extent_entry *find_entry(struct list_head *entries,
5222 u64 bytenr, u64 bytes)
5223 {
5224 struct extent_entry *entry = NULL;
5225
5226 list_for_each_entry(entry, entries, list) {
5227 if (entry->bytenr == bytenr && entry->bytes == bytes)
5228 return entry;
5229 }
5230
5231 return NULL;
5232 }
5233
5234 static struct extent_entry *find_most_right_entry(struct list_head *entries)
5235 {
5236 struct extent_entry *entry, *best = NULL, *prev = NULL;
5237
5238 list_for_each_entry(entry, entries, list) {
5239 if (!prev) {
5240 prev = entry;
5241 continue;
5242 }
5243
5244 /*
5245 * If there are as many broken entries as entries then we know
5246 * not to trust this particular entry.
5247 */
5248 if (entry->broken == entry->count)
5249 continue;
5250
5251 /*
5252 * If our current entry == best then we can't be sure our best
5253 * is really the best, so we need to keep searching.
5254 */
5255 if (best && best->count == entry->count) {
5256 prev = entry;
5257 best = NULL;
5258 continue;
5259 }
5260
5261 /* Prev == entry, not good enough, have to keep searching */
5262 if (!prev->broken && prev->count == entry->count)
5263 continue;
5264
5265 if (!best)
5266 best = (prev->count > entry->count) ? prev : entry;
5267 else if (best->count < entry->count)
5268 best = entry;
5269 prev = entry;
5270 }
5271
5272 return best;
5273 }
5274
5275 static int repair_ref(struct btrfs_trans_handle *trans,
5276 struct btrfs_fs_info *info, struct btrfs_path *path,
5277 struct data_backref *dback, struct extent_entry *entry)
5278 {
5279 struct btrfs_root *root;
5280 struct btrfs_file_extent_item *fi;
5281 struct extent_buffer *leaf;
5282 struct btrfs_key key;
5283 u64 bytenr, bytes;
5284 int ret;
5285
5286 key.objectid = dback->root;
5287 key.type = BTRFS_ROOT_ITEM_KEY;
5288 key.offset = (u64)-1;
5289 root = btrfs_read_fs_root(info, &key);
5290 if (IS_ERR(root)) {
5291 fprintf(stderr, "Couldn't find root for our ref\n");
5292 return -EINVAL;
5293 }
5294
5295 /*
5296 * The backref points to the original offset of the extent if it was
5297 * split, so we need to search down to the offset we have and then walk
5298 * forward until we find the backref we're looking for.
5299 */
5300 key.objectid = dback->owner;
5301 key.type = BTRFS_EXTENT_DATA_KEY;
5302 key.offset = dback->offset;
5303 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5304 if (ret < 0) {
5305 fprintf(stderr, "Error looking up ref %d\n", ret);
5306 return ret;
5307 }
5308
5309 while (1) {
5310 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5311 ret = btrfs_next_leaf(root, path);
5312 if (ret) {
5313 fprintf(stderr, "Couldn't find our ref, next\n");
5314 return -EINVAL;
5315 }
5316 }
5317 leaf = path->nodes[0];
5318 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5319 if (key.objectid != dback->owner ||
5320 key.type != BTRFS_EXTENT_DATA_KEY) {
5321 fprintf(stderr, "Couldn't find our ref, search\n");
5322 return -EINVAL;
5323 }
5324 fi = btrfs_item_ptr(leaf, path->slots[0],
5325 struct btrfs_file_extent_item);
5326 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5327 bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
5328
5329 if (bytenr == dback->disk_bytenr && bytes == dback->bytes)
5330 break;
5331 path->slots[0]++;
5332 }
5333
5334 btrfs_release_path(path);
5335
5336 /*
5337 * Have to make sure that this root gets updated when we commit the
5338 * transaction
5339 */
5340 record_root_in_trans(trans, root);
5341
5342 /*
5343 * Ok we have the key of the file extent we want to fix, now we can cow
5344 * down to the thing and fix it.
5345 */
5346 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
5347 if (ret < 0) {
5348 fprintf(stderr, "Error cowing down to ref [%Lu, %u, %Lu]: %d\n",
5349 key.objectid, key.type, key.offset, ret);
5350 return ret;
5351 }
5352 if (ret > 0) {
5353 fprintf(stderr, "Well that's odd, we just found this key "
5354 "[%Lu, %u, %Lu]\n", key.objectid, key.type,
5355 key.offset);
5356 return -EINVAL;
5357 }
5358 leaf = path->nodes[0];
5359 fi = btrfs_item_ptr(leaf, path->slots[0],
5360 struct btrfs_file_extent_item);
5361
5362 if (btrfs_file_extent_compression(leaf, fi) &&
5363 dback->disk_bytenr != entry->bytenr) {
5364 fprintf(stderr, "Ref doesn't match the record start and is "
5365 "compressed, please take a btrfs-image of this file "
5366 "system and send it to a btrfs developer so they can "
5367 "complete this functionality for bytenr %Lu\n",
5368 dback->disk_bytenr);
5369 return -EINVAL;
5370 }
5371
5372 if (dback->node.broken && dback->disk_bytenr != entry->bytenr) {
5373 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
5374 } else if (dback->disk_bytenr > entry->bytenr) {
5375 u64 off_diff, offset;
5376
5377 off_diff = dback->disk_bytenr - entry->bytenr;
5378 offset = btrfs_file_extent_offset(leaf, fi);
5379 if (dback->disk_bytenr + offset +
5380 btrfs_file_extent_num_bytes(leaf, fi) >
5381 entry->bytenr + entry->bytes) {
5382 fprintf(stderr, "Ref is past the entry end, please "
5383 "take a btrfs-image of this file system and "
5384 "send it to a btrfs developer, ref %Lu\n",
5385 dback->disk_bytenr);
5386 return -EINVAL;
5387 }
5388 offset += off_diff;
5389 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
5390 btrfs_set_file_extent_offset(leaf, fi, offset);
5391 } else if (dback->disk_bytenr < entry->bytenr) {
5392 u64 offset;
5393
5394 offset = btrfs_file_extent_offset(leaf, fi);
5395 if (dback->disk_bytenr + offset < entry->bytenr) {
5396 fprintf(stderr, "Ref is before the entry start, please"
5397 " take a btrfs-image of this file system and "
5398 "send it to a btrfs developer, ref %Lu\n",
5399 dback->disk_bytenr);
5400 return -EINVAL;
5401 }
5402
5403 offset += dback->disk_bytenr;
5404 offset -= entry->bytenr;
5405 btrfs_set_file_extent_disk_bytenr(leaf, fi, entry->bytenr);
5406 btrfs_set_file_extent_offset(leaf, fi, offset);
5407 }
5408
5409 btrfs_set_file_extent_disk_num_bytes(leaf, fi, entry->bytes);
5410
5411 /*
5412 * Chances are if disk_num_bytes were wrong then so is ram_bytes, but
5413 * only do this if we aren't using compression, otherwise it's a
5414 * trickier case.
5415 */
5416 if (!btrfs_file_extent_compression(leaf, fi))
5417 btrfs_set_file_extent_ram_bytes(leaf, fi, entry->bytes);
5418 else
5419 printf("ram bytes may be wrong?\n");
5420 btrfs_mark_buffer_dirty(leaf);
5421 btrfs_release_path(path);
5422 return 0;
5423 }
5424
5425 static int verify_backrefs(struct btrfs_trans_handle *trans,
5426 struct btrfs_fs_info *info, struct btrfs_path *path,
5427 struct extent_record *rec)
5428 {
5429 struct extent_backref *back;
5430 struct data_backref *dback;
5431 struct extent_entry *entry, *best = NULL;
5432 LIST_HEAD(entries);
5433 int nr_entries = 0;
5434 int broken_entries = 0;
5435 int ret = 0;
5436 short mismatch = 0;
5437
5438 /*
5439 * Metadata is easy and the backrefs should always agree on bytenr and
5440 * size, if not we've got bigger issues.
5441 */
5442 if (rec->metadata)
5443 return 0;
5444
5445 list_for_each_entry(back, &rec->backrefs, list) {
5446 if (back->full_backref || !back->is_data)
5447 continue;
5448
5449 dback = (struct data_backref *)back;
5450
5451 /*
5452 * We only pay attention to backrefs that we found a real
5453 * backref for.
5454 */
5455 if (dback->found_ref == 0)
5456 continue;
5457
5458 /*
5459 * For now we only catch when the bytes don't match, not the
5460 * bytenr. We can easily do this at the same time, but I want
5461 * to have a fs image to test on before we just add repair
5462 * functionality willy-nilly so we know we won't screw up the
5463 * repair.
5464 */
5465
5466 entry = find_entry(&entries, dback->disk_bytenr,
5467 dback->bytes);
5468 if (!entry) {
5469 entry = malloc(sizeof(struct extent_entry));
5470 if (!entry) {
5471 ret = -ENOMEM;
5472 goto out;
5473 }
5474 memset(entry, 0, sizeof(*entry));
5475 entry->bytenr = dback->disk_bytenr;
5476 entry->bytes = dback->bytes;
5477 list_add_tail(&entry->list, &entries);
5478 nr_entries++;
5479 }
5480
5481 /*
5482 * If we only have on entry we may think the entries agree when
5483 * in reality they don't so we have to do some extra checking.
5484 */
5485 if (dback->disk_bytenr != rec->start ||
5486 dback->bytes != rec->nr || back->broken)
5487 mismatch = 1;
5488
5489 if (back->broken) {
5490 entry->broken++;
5491 broken_entries++;
5492 }
5493
5494 entry->count++;
5495 }
5496
5497 /* Yay all the backrefs agree, carry on good sir */
5498 if (nr_entries <= 1 && !mismatch)
5499 goto out;
5500
5501 fprintf(stderr, "attempting to repair backref discrepency for bytenr "
5502 "%Lu\n", rec->start);
5503
5504 /*
5505 * First we want to see if the backrefs can agree amongst themselves who
5506 * is right, so figure out which one of the entries has the highest
5507 * count.
5508 */
5509 best = find_most_right_entry(&entries);
5510
5511 /*
5512 * Ok so we may have an even split between what the backrefs think, so
5513 * this is where we use the extent ref to see what it thinks.
5514 */
5515 if (!best) {
5516 entry = find_entry(&entries, rec->start, rec->nr);
5517 if (!entry && (!broken_entries || !rec->found_rec)) {
5518 fprintf(stderr, "Backrefs don't agree with each other "
5519 "and extent record doesn't agree with anybody,"
5520 " so we can't fix bytenr %Lu bytes %Lu\n",
5521 rec->start, rec->nr);
5522 ret = -EINVAL;
5523 goto out;
5524 } else if (!entry) {
5525 /*
5526 * Ok our backrefs were broken, we'll assume this is the
5527 * correct value and add an entry for this range.
5528 */
5529 entry = malloc(sizeof(struct extent_entry));
5530 if (!entry) {
5531 ret = -ENOMEM;
5532 goto out;
5533 }
5534 memset(entry, 0, sizeof(*entry));
5535 entry->bytenr = rec->start;
5536 entry->bytes = rec->nr;
5537 list_add_tail(&entry->list, &entries);
5538 nr_entries++;
5539 }
5540 entry->count++;
5541 best = find_most_right_entry(&entries);
5542 if (!best) {
5543 fprintf(stderr, "Backrefs and extent record evenly "
5544 "split on who is right, this is going to "
5545 "require user input to fix bytenr %Lu bytes "
5546 "%Lu\n", rec->start, rec->nr);
5547 ret = -EINVAL;
5548 goto out;
5549 }
5550 }
5551
5552 /*
5553 * I don't think this can happen currently as we'll abort() if we catch
5554 * this case higher up, but in case somebody removes that we still can't
5555 * deal with it properly here yet, so just bail out of that's the case.
5556 */
5557 if (best->bytenr != rec->start) {
5558 fprintf(stderr, "Extent start and backref starts don't match, "
5559 "please use btrfs-image on this file system and send "
5560 "it to a btrfs developer so they can make fsck fix "
5561 "this particular case. bytenr is %Lu, bytes is %Lu\n",
5562 rec->start, rec->nr);
5563 ret = -EINVAL;
5564 goto out;
5565 }
5566
5567 /*
5568 * Ok great we all agreed on an extent record, let's go find the real
5569 * references and fix up the ones that don't match.
5570 */
5571 list_for_each_entry(back, &rec->backrefs, list) {
5572 if (back->full_backref || !back->is_data)
5573 continue;
5574
5575 dback = (struct data_backref *)back;
5576
5577 /*
5578 * Still ignoring backrefs that don't have a real ref attached
5579 * to them.
5580 */
5581 if (dback->found_ref == 0)
5582 continue;
5583
5584 if (dback->bytes == best->bytes &&
5585 dback->disk_bytenr == best->bytenr)
5586 continue;
5587
5588 ret = repair_ref(trans, info, path, dback, best);
5589 if (ret)
5590 goto out;
5591 }
5592
5593 /*
5594 * Ok we messed with the actual refs, which means we need to drop our
5595 * entire cache and go back and rescan. I know this is a huge pain and
5596 * adds a lot of extra work, but it's the only way to be safe. Once all
5597 * the backrefs agree we may not need to do anything to the extent
5598 * record itself.
5599 */
5600 ret = -EAGAIN;
5601 out:
5602 while (!list_empty(&entries)) {
5603 entry = list_entry(entries.next, struct extent_entry, list);
5604 list_del_init(&entry->list);
5605 free(entry);
5606 }
5607 return ret;
5608 }
5609
5610 static int process_duplicates(struct btrfs_root *root,
5611 struct cache_tree *extent_cache,
5612 struct extent_record *rec)
5613 {
5614 struct extent_record *good, *tmp;
5615 struct cache_extent *cache;
5616 int ret;
5617
5618 /*
5619 * If we found a extent record for this extent then return, or if we
5620 * have more than one duplicate we are likely going to need to delete
5621 * something.
5622 */
5623 if (rec->found_rec || rec->num_duplicates > 1)
5624 return 0;
5625
5626 /* Shouldn't happen but just in case */
5627 BUG_ON(!rec->num_duplicates);
5628
5629 /*
5630 * So this happens if we end up with a backref that doesn't match the
5631 * actual extent entry. So either the backref is bad or the extent
5632 * entry is bad. Either way we want to have the extent_record actually
5633 * reflect what we found in the extent_tree, so we need to take the
5634 * duplicate out and use that as the extent_record since the only way we
5635 * get a duplicate is if we find a real life BTRFS_EXTENT_ITEM_KEY.
5636 */
5637 remove_cache_extent(extent_cache, &rec->cache);
5638
5639 good = list_entry(rec->dups.next, struct extent_record, list);
5640 list_del_init(&good->list);
5641 INIT_LIST_HEAD(&good->backrefs);
5642 INIT_LIST_HEAD(&good->dups);
5643 good->cache.start = good->start;
5644 good->cache.size = good->nr;
5645 good->content_checked = 0;
5646 good->owner_ref_checked = 0;
5647 good->num_duplicates = 0;
5648 good->refs = rec->refs;
5649 list_splice_init(&rec->backrefs, &good->backrefs);
5650 while (1) {
5651 cache = lookup_cache_extent(extent_cache, good->start,
5652 good->nr);
5653 if (!cache)
5654 break;
5655 tmp = container_of(cache, struct extent_record, cache);
5656
5657 /*
5658 * If we find another overlapping extent and it's found_rec is
5659 * set then it's a duplicate and we need to try and delete
5660 * something.
5661 */
5662 if (tmp->found_rec || tmp->num_duplicates > 0) {
5663 if (list_empty(&good->list))
5664 list_add_tail(&good->list,
5665 &duplicate_extents);
5666 good->num_duplicates += tmp->num_duplicates + 1;
5667 list_splice_init(&tmp->dups, &good->dups);
5668 list_del_init(&tmp->list);
5669 list_add_tail(&tmp->list, &good->dups);
5670 remove_cache_extent(extent_cache, &tmp->cache);
5671 continue;
5672 }
5673
5674 /*
5675 * Ok we have another non extent item backed extent rec, so lets
5676 * just add it to this extent and carry on like we did above.
5677 */
5678 good->refs += tmp->refs;
5679 list_splice_init(&tmp->backrefs, &good->backrefs);
5680 remove_cache_extent(extent_cache, &tmp->cache);
5681 free(tmp);
5682 }
5683 ret = insert_cache_extent(extent_cache, &good->cache);
5684 BUG_ON(ret);
5685 free(rec);
5686 return good->num_duplicates ? 0 : 1;
5687 }
5688
5689 static int delete_duplicate_records(struct btrfs_trans_handle *trans,
5690 struct btrfs_root *root,
5691 struct extent_record *rec)
5692 {
5693 LIST_HEAD(delete_list);
5694 struct btrfs_path *path;
5695 struct extent_record *tmp, *good, *n;
5696 int nr_del = 0;
5697 int ret = 0;
5698 struct btrfs_key key;
5699
5700 path = btrfs_alloc_path();
5701 if (!path) {
5702 ret = -ENOMEM;
5703 goto out;
5704 }
5705
5706 good = rec;
5707 /* Find the record that covers all of the duplicates. */
5708 list_for_each_entry(tmp, &rec->dups, list) {
5709 if (good->start < tmp->start)
5710 continue;
5711 if (good->nr > tmp->nr)
5712 continue;
5713
5714 if (tmp->start + tmp->nr < good->start + good->nr) {
5715 fprintf(stderr, "Ok we have overlapping extents that "
5716 "aren't completely covered by eachother, this "
5717 "is going to require more careful thought. "
5718 "The extents are [%Lu-%Lu] and [%Lu-%Lu]\n",
5719 tmp->start, tmp->nr, good->start, good->nr);
5720 abort();
5721 }
5722 good = tmp;
5723 }
5724
5725 if (good != rec)
5726 list_add_tail(&rec->list, &delete_list);
5727
5728 list_for_each_entry_safe(tmp, n, &rec->dups, list) {
5729 if (tmp == good)
5730 continue;
5731 list_move_tail(&tmp->list, &delete_list);
5732 }
5733
5734 root = root->fs_info->extent_root;
5735 list_for_each_entry(tmp, &delete_list, list) {
5736 if (tmp->found_rec == 0)
5737 continue;
5738 key.objectid = tmp->start;
5739 key.type = BTRFS_EXTENT_ITEM_KEY;
5740 key.offset = tmp->nr;
5741
5742 /* Shouldn't happen but just in case */
5743 if (tmp->metadata) {
5744 fprintf(stderr, "Well this shouldn't happen, extent "
5745 "record overlaps but is metadata? "
5746 "[%Lu, %Lu]\n", tmp->start, tmp->nr);
5747 abort();
5748 }
5749
5750 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
5751 if (ret) {
5752 if (ret > 0)
5753 ret = -EINVAL;
5754 goto out;
5755 }
5756 ret = btrfs_del_item(trans, root, path);
5757 if (ret)
5758 goto out;
5759 btrfs_release_path(path);
5760 nr_del++;
5761 }
5762
5763 out:
5764 while (!list_empty(&delete_list)) {
5765 tmp = list_entry(delete_list.next, struct extent_record, list);
5766 list_del_init(&tmp->list);
5767 if (tmp == rec)
5768 continue;
5769 free(tmp);
5770 }
5771
5772 while (!list_empty(&rec->dups)) {
5773 tmp = list_entry(rec->dups.next, struct extent_record, list);
5774 list_del_init(&tmp->list);
5775 free(tmp);
5776 }
5777
5778 btrfs_free_path(path);
5779
5780 if (!ret && !nr_del)
5781 rec->num_duplicates = 0;
5782
5783 return ret ? ret : nr_del;
5784 }
5785
5786 static int find_possible_backrefs(struct btrfs_trans_handle *trans,
5787 struct btrfs_fs_info *info,
5788 struct btrfs_path *path,
5789 struct cache_tree *extent_cache,
5790 struct extent_record *rec)
5791 {
5792 struct btrfs_root *root;
5793 struct extent_backref *back;
5794 struct data_backref *dback;
5795 struct cache_extent *cache;
5796 struct btrfs_file_extent_item *fi;
5797 struct btrfs_key key;
5798 u64 bytenr, bytes;
5799 int ret;
5800
5801 list_for_each_entry(back, &rec->backrefs, list) {
5802 /* Don't care about full backrefs (poor unloved backrefs) */
5803 if (back->full_backref || !back->is_data)
5804 continue;
5805
5806 dback = (struct data_backref *)back;
5807
5808 /* We found this one, we don't need to do a lookup */
5809 if (dback->found_ref)
5810 continue;
5811
5812 key.objectid = dback->root;
5813 key.type = BTRFS_ROOT_ITEM_KEY;
5814 key.offset = (u64)-1;
5815
5816 root = btrfs_read_fs_root(info, &key);
5817
5818 /* No root, definitely a bad ref, skip */
5819 if (IS_ERR(root) && PTR_ERR(root) == -ENOENT)
5820 continue;
5821 /* Other err, exit */
5822 if (IS_ERR(root))
5823 return PTR_ERR(root);
5824
5825 key.objectid = dback->owner;
5826 key.type = BTRFS_EXTENT_DATA_KEY;
5827 key.offset = dback->offset;
5828 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5829 if (ret) {
5830 btrfs_release_path(path);
5831 if (ret < 0)
5832 return ret;
5833 /* Didn't find it, we can carry on */
5834 ret = 0;
5835 continue;
5836 }
5837
5838 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
5839 struct btrfs_file_extent_item);
5840 bytenr = btrfs_file_extent_disk_bytenr(path->nodes[0], fi);
5841 bytes = btrfs_file_extent_disk_num_bytes(path->nodes[0], fi);
5842 btrfs_release_path(path);
5843 cache = lookup_cache_extent(extent_cache, bytenr, 1);
5844 if (cache) {
5845 struct extent_record *tmp;
5846 tmp = container_of(cache, struct extent_record, cache);
5847
5848 /*
5849 * If we found an extent record for the bytenr for this
5850 * particular backref then we can't add it to our
5851 * current extent record. We only want to add backrefs
5852 * that don't have a corresponding extent item in the
5853 * extent tree since they likely belong to this record
5854 * and we need to fix it if it doesn't match bytenrs.
5855 */
5856 if (tmp->found_rec)
5857 continue;
5858 }
5859
5860 dback->found_ref += 1;
5861 dback->disk_bytenr = bytenr;
5862 dback->bytes = bytes;
5863
5864 /*
5865 * Set this so the verify backref code knows not to trust the
5866 * values in this backref.
5867 */
5868 back->broken = 1;
5869 }
5870
5871 return 0;
5872 }
5873
5874 /*
5875 * when an incorrect extent item is found, this will delete
5876 * all of the existing entries for it and recreate them
5877 * based on what the tree scan found.
5878 */
5879 static int fixup_extent_refs(struct btrfs_trans_handle *trans,
5880 struct btrfs_fs_info *info,
5881 struct cache_tree *extent_cache,
5882 struct extent_record *rec)
5883 {
5884 int ret;
5885 struct btrfs_path *path;
5886 struct list_head *cur = rec->backrefs.next;
5887 struct cache_extent *cache;
5888 struct extent_backref *back;
5889 int allocated = 0;
5890 u64 flags = 0;
5891
5892 /*
5893 * remember our flags for recreating the extent.
5894 * FIXME, if we have cleared extent tree, we can not
5895 * lookup extent info in extent tree.
5896 */
5897 if (!init_extent_tree) {
5898 ret = btrfs_lookup_extent_info(NULL, info->extent_root,
5899 rec->start, rec->max_size,
5900 rec->metadata, NULL, &flags);
5901 if (ret < 0)
5902 flags = 0;
5903 } else {
5904 flags = 0;
5905 }
5906
5907 path = btrfs_alloc_path();
5908 if (!path)
5909 return -ENOMEM;
5910
5911 if (rec->refs != rec->extent_item_refs && !rec->metadata) {
5912 /*
5913 * Sometimes the backrefs themselves are so broken they don't
5914 * get attached to any meaningful rec, so first go back and
5915 * check any of our backrefs that we couldn't find and throw
5916 * them into the list if we find the backref so that
5917 * verify_backrefs can figure out what to do.
5918 */
5919 ret = find_possible_backrefs(trans, info, path, extent_cache,
5920 rec);
5921 if (ret < 0)
5922 goto out;
5923 }
5924
5925 /* step one, make sure all of the backrefs agree */
5926 ret = verify_backrefs(trans, info, path, rec);
5927 if (ret < 0)
5928 goto out;
5929
5930 /* step two, delete all the existing records */
5931 ret = delete_extent_records(trans, info->extent_root, path,
5932 rec->start, rec->max_size);
5933
5934 if (ret < 0)
5935 goto out;
5936
5937 /* was this block corrupt? If so, don't add references to it */
5938 cache = lookup_cache_extent(info->corrupt_blocks,
5939 rec->start, rec->max_size);
5940 if (cache) {
5941 ret = 0;
5942 goto out;
5943 }
5944
5945 /* step three, recreate all the refs we did find */
5946 while(cur != &rec->backrefs) {
5947 back = list_entry(cur, struct extent_backref, list);
5948 cur = cur->next;
5949
5950 /*
5951 * if we didn't find any references, don't create a
5952 * new extent record
5953 */
5954 if (!back->found_ref)
5955 continue;
5956
5957 ret = record_extent(trans, info, path, rec, back, allocated, flags);
5958 allocated = 1;
5959
5960 if (ret)
5961 goto out;
5962 }
5963 out:
5964 btrfs_free_path(path);
5965 return ret;
5966 }
5967
5968 /* right now we only prune from the extent allocation tree */
5969 static int prune_one_block(struct btrfs_trans_handle *trans,
5970 struct btrfs_fs_info *info,
5971 struct btrfs_corrupt_block *corrupt)
5972 {
5973 int ret;
5974 struct btrfs_path path;
5975 struct extent_buffer *eb;
5976 u64 found;
5977 int slot;
5978 int nritems;
5979 int level = corrupt->level + 1;
5980
5981 btrfs_init_path(&path);
5982 again:
5983 /* we want to stop at the parent to our busted block */
5984 path.lowest_level = level;
5985
5986 ret = btrfs_search_slot(trans, info->extent_root,
5987 &corrupt->key, &path, -1, 1);
5988
5989 if (ret < 0)
5990 goto out;
5991
5992 eb = path.nodes[level];
5993 if (!eb) {
5994 ret = -ENOENT;
5995 goto out;
5996 }
5997
5998 /*
5999 * hopefully the search gave us the block we want to prune,
6000 * lets try that first
6001 */
6002 slot = path.slots[level];
6003 found = btrfs_node_blockptr(eb, slot);
6004 if (found == corrupt->cache.start)
6005 goto del_ptr;
6006
6007 nritems = btrfs_header_nritems(eb);
6008
6009 /* the search failed, lets scan this node and hope we find it */
6010 for (slot = 0; slot < nritems; slot++) {
6011 found = btrfs_node_blockptr(eb, slot);
6012 if (found == corrupt->cache.start)
6013 goto del_ptr;
6014 }
6015 /*
6016 * we couldn't find the bad block. TODO, search all the nodes for pointers
6017 * to this block
6018 */
6019 if (eb == info->extent_root->node) {
6020 ret = -ENOENT;
6021 goto out;
6022 } else {
6023 level++;
6024 btrfs_release_path(&path);
6025 goto again;
6026 }
6027
6028 del_ptr:
6029 printk("deleting pointer to block %Lu\n", corrupt->cache.start);
6030 ret = btrfs_del_ptr(trans, info->extent_root, &path, level, slot);
6031
6032 out:
6033 btrfs_release_path(&path);
6034 return ret;
6035 }
6036
6037 static int prune_corrupt_blocks(struct btrfs_trans_handle *trans,
6038 struct btrfs_fs_info *info)
6039 {
6040 struct cache_extent *cache;
6041 struct btrfs_corrupt_block *corrupt;
6042
6043 cache = search_cache_extent(info->corrupt_blocks, 0);
6044 while (1) {
6045 if (!cache)
6046 break;
6047 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
6048 prune_one_block(trans, info, corrupt);
6049 cache = next_cache_extent(cache);
6050 }
6051 return 0;
6052 }
6053
6054 static void free_corrupt_block(struct cache_extent *cache)
6055 {
6056 struct btrfs_corrupt_block *corrupt;
6057
6058 corrupt = container_of(cache, struct btrfs_corrupt_block, cache);
6059 free(corrupt);
6060 }
6061
6062 FREE_EXTENT_CACHE_BASED_TREE(corrupt_blocks, free_corrupt_block);
6063
6064 static void reset_cached_block_groups(struct btrfs_fs_info *fs_info)
6065 {
6066 struct btrfs_block_group_cache *cache;
6067 u64 start, end;
6068 int ret;
6069
6070 while (1) {
6071 ret = find_first_extent_bit(&fs_info->free_space_cache, 0,
6072 &start, &end, EXTENT_DIRTY);
6073 if (ret)
6074 break;
6075 clear_extent_dirty(&fs_info->free_space_cache, start, end,
6076 GFP_NOFS);
6077 }
6078
6079 start = 0;
6080 while (1) {
6081 cache = btrfs_lookup_first_block_group(fs_info, start);
6082 if (!cache)
6083 break;
6084 if (cache->cached)
6085 cache->cached = 0;
6086 start = cache->key.objectid + cache->key.offset;
6087 }
6088 }
6089
6090 static int check_extent_refs(struct btrfs_trans_handle *trans,
6091 struct btrfs_root *root,
6092 struct cache_tree *extent_cache)
6093 {
6094 struct extent_record *rec;
6095 struct cache_extent *cache;
6096 int err = 0;
6097 int ret = 0;
6098 int fixed = 0;
6099 int had_dups = 0;
6100
6101 if (repair) {
6102 /*
6103 * if we're doing a repair, we have to make sure
6104 * we don't allocate from the problem extents.
6105 * In the worst case, this will be all the
6106 * extents in the FS
6107 */
6108 cache = search_cache_extent(extent_cache, 0);
6109 while(cache) {
6110 rec = container_of(cache, struct extent_record, cache);
6111 btrfs_pin_extent(root->fs_info,
6112 rec->start, rec->max_size);
6113 cache = next_cache_extent(cache);
6114 }
6115
6116 /* pin down all the corrupted blocks too */
6117 cache = search_cache_extent(root->fs_info->corrupt_blocks, 0);
6118 while(cache) {
6119 btrfs_pin_extent(root->fs_info,
6120 cache->start, cache->size);
6121 cache = next_cache_extent(cache);
6122 }
6123 prune_corrupt_blocks(trans, root->fs_info);
6124 reset_cached_block_groups(root->fs_info);
6125 }
6126
6127 /*
6128 * We need to delete any duplicate entries we find first otherwise we
6129 * could mess up the extent tree when we have backrefs that actually
6130 * belong to a different extent item and not the weird duplicate one.
6131 */
6132 while (repair && !list_empty(&duplicate_extents)) {
6133 rec = list_entry(duplicate_extents.next, struct extent_record,
6134 list);
6135 list_del_init(&rec->list);
6136
6137 /* Sometimes we can find a backref before we find an actual
6138 * extent, so we need to process it a little bit to see if there
6139 * truly are multiple EXTENT_ITEM_KEY's for the same range, or
6140 * if this is a backref screwup. If we need to delete stuff
6141 * process_duplicates() will return 0, otherwise it will return
6142 * 1 and we
6143 */
6144 if (process_duplicates(root, extent_cache, rec))
6145 continue;
6146 ret = delete_duplicate_records(trans, root, rec);
6147 if (ret < 0)
6148 return ret;
6149 /*
6150 * delete_duplicate_records will return the number of entries
6151 * deleted, so if it's greater than 0 then we know we actually
6152 * did something and we need to remove.
6153 */
6154 if (ret)
6155 had_dups = 1;
6156 }
6157
6158 if (had_dups)
6159 return -EAGAIN;
6160
6161 while(1) {
6162 fixed = 0;
6163 cache = search_cache_extent(extent_cache, 0);
6164 if (!cache)
6165 break;
6166 rec = container_of(cache, struct extent_record, cache);
6167 if (rec->num_duplicates) {
6168 fprintf(stderr, "extent item %llu has multiple extent "
6169 "items\n", (unsigned long long)rec->start);
6170 err = 1;
6171 }
6172
6173 if (rec->refs != rec->extent_item_refs) {
6174 fprintf(stderr, "ref mismatch on [%llu %llu] ",
6175 (unsigned long long)rec->start,
6176 (unsigned long long)rec->nr);
6177 fprintf(stderr, "extent item %llu, found %llu\n",
6178 (unsigned long long)rec->extent_item_refs,
6179 (unsigned long long)rec->refs);
6180 if (!fixed && repair) {
6181 ret = fixup_extent_refs(trans, root->fs_info,
6182 extent_cache, rec);
6183 if (ret)
6184 goto repair_abort;
6185 fixed = 1;
6186 }
6187 err = 1;
6188
6189 }
6190 if (all_backpointers_checked(rec, 1)) {
6191 fprintf(stderr, "backpointer mismatch on [%llu %llu]\n",
6192 (unsigned long long)rec->start,
6193 (unsigned long long)rec->nr);
6194
6195 if (!fixed && repair) {
6196 ret = fixup_extent_refs(trans, root->fs_info,
6197 extent_cache, rec);
6198 if (ret)
6199 goto repair_abort;
6200 fixed = 1;
6201 }
6202
6203 err = 1;
6204 }
6205 if (!rec->owner_ref_checked) {
6206 fprintf(stderr, "owner ref check failed [%llu %llu]\n",
6207 (unsigned long long)rec->start,
6208 (unsigned long long)rec->nr);
6209 if (!fixed && repair) {
6210 ret = fixup_extent_refs(trans, root->fs_info,
6211 extent_cache, rec);
6212 if (ret)
6213 goto repair_abort;
6214 fixed = 1;
6215 }
6216 err = 1;
6217 }
6218
6219 remove_cache_extent(extent_cache, cache);
6220 free_all_extent_backrefs(rec);
6221 free(rec);
6222 }
6223 repair_abort:
6224 if (repair) {
6225 if (ret && ret != -EAGAIN) {
6226 fprintf(stderr, "failed to repair damaged filesystem, aborting\n");
6227 exit(1);
6228 } else if (!ret) {
6229 btrfs_fix_block_accounting(trans, root);
6230 }
6231 if (err)
6232 fprintf(stderr, "repaired damaged extent references\n");
6233 return ret;
6234 }
6235 return err;
6236 }
6237
6238 u64 calc_stripe_length(u64 type, u64 length, int num_stripes)
6239 {
6240 u64 stripe_size;
6241
6242 if (type & BTRFS_BLOCK_GROUP_RAID0) {
6243 stripe_size = length;
6244 stripe_size /= num_stripes;
6245 } else if (type & BTRFS_BLOCK_GROUP_RAID10) {
6246 stripe_size = length * 2;
6247 stripe_size /= num_stripes;
6248 } else if (type & BTRFS_BLOCK_GROUP_RAID5) {
6249 stripe_size = length;
6250 stripe_size /= (num_stripes - 1);
6251 } else if (type & BTRFS_BLOCK_GROUP_RAID6) {
6252 stripe_size = length;
6253 stripe_size /= (num_stripes - 2);
6254 } else {
6255 stripe_size = length;
6256 }
6257 return stripe_size;
6258 }
6259
6260 static int check_chunk_refs(struct chunk_record *chunk_rec,
6261 struct block_group_tree *block_group_cache,
6262 struct device_extent_tree *dev_extent_cache,
6263 int silent)
6264 {
6265 struct cache_extent *block_group_item;
6266 struct block_group_record *block_group_rec;
6267 struct cache_extent *dev_extent_item;
6268 struct device_extent_record *dev_extent_rec;
6269 u64 devid;
6270 u64 offset;
6271 u64 length;
6272 int i;
6273 int ret = 0;
6274
6275 block_group_item = lookup_cache_extent(&block_group_cache->tree,
6276 chunk_rec->offset,
6277 chunk_rec->length);
6278 if (block_group_item) {
6279 block_group_rec = container_of(block_group_item,
6280 struct block_group_record,
6281 cache);
6282 if (chunk_rec->length != block_group_rec->offset ||
6283 chunk_rec->offset != block_group_rec->objectid ||
6284 chunk_rec->type_flags != block_group_rec->flags) {
6285 if (!silent)
6286 fprintf(stderr,
6287 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) mismatch with block group[%llu, %u, %llu]: offset(%llu), objectid(%llu), flags(%llu)\n",
6288 chunk_rec->objectid,
6289 chunk_rec->type,
6290 chunk_rec->offset,
6291 chunk_rec->length,
6292 chunk_rec->offset,
6293 chunk_rec->type_flags,
6294 block_group_rec->objectid,
6295 block_group_rec->type,
6296 block_group_rec->offset,
6297 block_group_rec->offset,
6298 block_group_rec->objectid,
6299 block_group_rec->flags);
6300 ret = -1;
6301 } else {
6302 list_del_init(&block_group_rec->list);
6303 chunk_rec->bg_rec = block_group_rec;
6304 }
6305 } else {
6306 if (!silent)
6307 fprintf(stderr,
6308 "Chunk[%llu, %u, %llu]: length(%llu), offset(%llu), type(%llu) is not found in block group\n",
6309 chunk_rec->objectid,
6310 chunk_rec->type,
6311 chunk_rec->offset,
6312 chunk_rec->length,
6313 chunk_rec->offset,
6314 chunk_rec->type_flags);
6315 ret = -1;
6316 }
6317
6318 length = calc_stripe_length(chunk_rec->type_flags, chunk_rec->length,
6319 chunk_rec->num_stripes);
6320 for (i = 0; i < chunk_rec->num_stripes; ++i) {
6321 devid = chunk_rec->stripes[i].devid;
6322 offset = chunk_rec->stripes[i].offset;
6323 dev_extent_item = lookup_cache_extent2(&dev_extent_cache->tree,
6324 devid, offset, length);
6325 if (dev_extent_item) {
6326 dev_extent_rec = container_of(dev_extent_item,
6327 struct device_extent_record,
6328 cache);
6329 if (dev_extent_rec->objectid != devid ||
6330 dev_extent_rec->offset != offset ||
6331 dev_extent_rec->chunk_offset != chunk_rec->offset ||
6332 dev_extent_rec->length != length) {
6333 if (!silent)
6334 fprintf(stderr,
6335 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] dismatch dev extent[%llu, %llu, %llu]\n",
6336 chunk_rec->objectid,
6337 chunk_rec->type,
6338 chunk_rec->offset,
6339 chunk_rec->stripes[i].devid,
6340 chunk_rec->stripes[i].offset,
6341 dev_extent_rec->objectid,
6342 dev_extent_rec->offset,
6343 dev_extent_rec->length);
6344 ret = -1;
6345 } else {
6346 list_move(&dev_extent_rec->chunk_list,
6347 &chunk_rec->dextents);
6348 }
6349 } else {
6350 if (!silent)
6351 fprintf(stderr,
6352 "Chunk[%llu, %u, %llu] stripe[%llu, %llu] is not found in dev extent\n",
6353 chunk_rec->objectid,
6354 chunk_rec->type,
6355 chunk_rec->offset,
6356 chunk_rec->stripes[i].devid,
6357 chunk_rec->stripes[i].offset);
6358 ret = -1;
6359 }
6360 }
6361 return ret;
6362 }
6363
6364 /* check btrfs_chunk -> btrfs_dev_extent / btrfs_block_group_item */
6365 int check_chunks(struct cache_tree *chunk_cache,
6366 struct block_group_tree *block_group_cache,
6367 struct device_extent_tree *dev_extent_cache,
6368 struct list_head *good, struct list_head *bad, int silent)
6369 {
6370 struct cache_extent *chunk_item;
6371 struct chunk_record *chunk_rec;
6372 struct block_group_record *bg_rec;
6373 struct device_extent_record *dext_rec;
6374 int err;
6375 int ret = 0;
6376
6377 chunk_item = first_cache_extent(chunk_cache);
6378 while (chunk_item) {
6379 chunk_rec = container_of(chunk_item, struct chunk_record,
6380 cache);
6381 err = check_chunk_refs(chunk_rec, block_group_cache,
6382 dev_extent_cache, silent);
6383 if (err) {
6384 ret = err;
6385 if (bad)
6386 list_add_tail(&chunk_rec->list, bad);
6387 } else {
6388 if (good)
6389 list_add_tail(&chunk_rec->list, good);
6390 }
6391
6392 chunk_item = next_cache_extent(chunk_item);
6393 }
6394
6395 list_for_each_entry(bg_rec, &block_group_cache->block_groups, list) {
6396 if (!silent)
6397 fprintf(stderr,
6398 "Block group[%llu, %llu] (flags = %llu) didn't find the relative chunk.\n",
6399 bg_rec->objectid,
6400 bg_rec->offset,
6401 bg_rec->flags);
6402 if (!ret)
6403 ret = 1;
6404 }
6405
6406 list_for_each_entry(dext_rec, &dev_extent_cache->no_chunk_orphans,
6407 chunk_list) {
6408 if (!silent)
6409 fprintf(stderr,
6410 "Device extent[%llu, %llu, %llu] didn't find the relative chunk.\n",
6411 dext_rec->objectid,
6412 dext_rec->offset,
6413 dext_rec->length);
6414 if (!ret)
6415 ret = 1;
6416 }
6417 return ret;
6418 }
6419
6420
6421 static int check_device_used(struct device_record *dev_rec,
6422 struct device_extent_tree *dext_cache)
6423 {
6424 struct cache_extent *cache;
6425 struct device_extent_record *dev_extent_rec;
6426 u64 total_byte = 0;
6427
6428 cache = search_cache_extent2(&dext_cache->tree, dev_rec->devid, 0);
6429 while (cache) {
6430 dev_extent_rec = container_of(cache,
6431 struct device_extent_record,
6432 cache);
6433 if (dev_extent_rec->objectid != dev_rec->devid)
6434 break;
6435
6436 list_del_init(&dev_extent_rec->device_list);
6437 total_byte += dev_extent_rec->length;
6438 cache = next_cache_extent(cache);
6439 }
6440
6441 if (total_byte != dev_rec->byte_used) {
6442 fprintf(stderr,
6443 "Dev extent's total-byte(%llu) is not equal to byte-used(%llu) in dev[%llu, %u, %llu]\n",
6444 total_byte, dev_rec->byte_used, dev_rec->objectid,
6445 dev_rec->type, dev_rec->offset);
6446 return -1;
6447 } else {
6448 return 0;
6449 }
6450 }
6451
6452 /* check btrfs_dev_item -> btrfs_dev_extent */
6453 static int check_devices(struct rb_root *dev_cache,
6454 struct device_extent_tree *dev_extent_cache)
6455 {
6456 struct rb_node *dev_node;
6457 struct device_record *dev_rec;
6458 struct device_extent_record *dext_rec;
6459 int err;
6460 int ret = 0;
6461
6462 dev_node = rb_first(dev_cache);
6463 while (dev_node) {
6464 dev_rec = container_of(dev_node, struct device_record, node);
6465 err = check_device_used(dev_rec, dev_extent_cache);
6466 if (err)
6467 ret = err;
6468
6469 dev_node = rb_next(dev_node);
6470 }
6471 list_for_each_entry(dext_rec, &dev_extent_cache->no_device_orphans,
6472 device_list) {
6473 fprintf(stderr,
6474 "Device extent[%llu, %llu, %llu] didn't find its device.\n",
6475 dext_rec->objectid, dext_rec->offset, dext_rec->length);
6476 if (!ret)
6477 ret = 1;
6478 }
6479 return ret;
6480 }
6481
6482 static int check_chunks_and_extents(struct btrfs_root *root)
6483 {
6484 struct rb_root dev_cache;
6485 struct cache_tree chunk_cache;
6486 struct block_group_tree block_group_cache;
6487 struct device_extent_tree dev_extent_cache;
6488 struct cache_tree extent_cache;
6489 struct cache_tree seen;
6490 struct cache_tree pending;
6491 struct cache_tree reada;
6492 struct cache_tree nodes;
6493 struct cache_tree corrupt_blocks;
6494 struct btrfs_path path;
6495 struct btrfs_key key;
6496 struct btrfs_key found_key;
6497 int ret, err = 0;
6498 u64 last = 0;
6499 struct block_info *bits;
6500 int bits_nr;
6501 struct extent_buffer *leaf;
6502 struct btrfs_trans_handle *trans = NULL;
6503 int slot;
6504 struct btrfs_root_item ri;
6505 struct list_head dropping_trees;
6506
6507 dev_cache = RB_ROOT;
6508 cache_tree_init(&chunk_cache);
6509 block_group_tree_init(&block_group_cache);
6510 device_extent_tree_init(&dev_extent_cache);
6511
6512 cache_tree_init(&extent_cache);
6513 cache_tree_init(&seen);
6514 cache_tree_init(&pending);
6515 cache_tree_init(&nodes);
6516 cache_tree_init(&reada);
6517 cache_tree_init(&corrupt_blocks);
6518 INIT_LIST_HEAD(&dropping_trees);
6519
6520 if (repair) {
6521 trans = btrfs_start_transaction(root, 1);
6522 if (IS_ERR(trans)) {
6523 fprintf(stderr, "Error starting transaction\n");
6524 return PTR_ERR(trans);
6525 }
6526 root->fs_info->fsck_extent_cache = &extent_cache;
6527 root->fs_info->free_extent_hook = free_extent_hook;
6528 root->fs_info->corrupt_blocks = &corrupt_blocks;
6529 }
6530
6531 bits_nr = 1024;
6532 bits = malloc(bits_nr * sizeof(struct block_info));
6533 if (!bits) {
6534 perror("malloc");
6535 exit(1);
6536 }
6537
6538 again:
6539 add_root_to_pending(root->fs_info->tree_root->node,
6540 &extent_cache, &pending, &seen, &nodes,
6541 &root->fs_info->tree_root->root_key);
6542
6543 add_root_to_pending(root->fs_info->chunk_root->node,
6544 &extent_cache, &pending, &seen, &nodes,
6545 &root->fs_info->chunk_root->root_key);
6546
6547 btrfs_init_path(&path);
6548 key.offset = 0;
6549 key.objectid = 0;
6550 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
6551 ret = btrfs_search_slot(NULL, root->fs_info->tree_root,
6552 &key, &path, 0, 0);
6553 if (ret < 0)
6554 goto out;
6555 while(1) {
6556 leaf = path.nodes[0];
6557 slot = path.slots[0];
6558 if (slot >= btrfs_header_nritems(path.nodes[0])) {
6559 ret = btrfs_next_leaf(root, &path);
6560 if (ret != 0)
6561 break;
6562 leaf = path.nodes[0];
6563 slot = path.slots[0];
6564 }
6565 btrfs_item_key_to_cpu(leaf, &found_key, path.slots[0]);
6566 if (btrfs_key_type(&found_key) == BTRFS_ROOT_ITEM_KEY) {
6567 unsigned long offset;
6568 struct extent_buffer *buf;
6569
6570 offset = btrfs_item_ptr_offset(leaf, path.slots[0]);
6571 read_extent_buffer(leaf, &ri, offset, sizeof(ri));
6572 if (btrfs_disk_key_objectid(&ri.drop_progress) == 0) {
6573 buf = read_tree_block(root->fs_info->tree_root,
6574 btrfs_root_bytenr(&ri),
6575 btrfs_level_size(root,
6576 btrfs_root_level(&ri)),
6577 0);
6578 if (!buf) {
6579 ret = -EIO;
6580 goto out;
6581 }
6582 add_root_to_pending(buf, &extent_cache,
6583 &pending, &seen, &nodes,
6584 &found_key);
6585 free_extent_buffer(buf);
6586 } else {
6587 struct dropping_root_item_record *dri_rec;
6588 dri_rec = malloc(sizeof(*dri_rec));
6589 if (!dri_rec) {
6590 perror("malloc");
6591 exit(1);
6592 }
6593 memcpy(&dri_rec->ri, &ri, sizeof(ri));
6594 memcpy(&dri_rec->found_key, &found_key,
6595 sizeof(found_key));
6596 list_add_tail(&dri_rec->list, &dropping_trees);
6597 }
6598 }
6599 path.slots[0]++;
6600 }
6601 btrfs_release_path(&path);
6602 while (1) {
6603 ret = run_next_block(trans, root, bits, bits_nr, &last,
6604 &pending, &seen, &reada, &nodes,
6605 &extent_cache, &chunk_cache, &dev_cache,
6606 &block_group_cache, &dev_extent_cache,
6607 NULL);
6608 if (ret != 0)
6609 break;
6610 }
6611
6612 while (!list_empty(&dropping_trees)) {
6613 struct dropping_root_item_record *rec;
6614 struct extent_buffer *buf;
6615 rec = list_entry(dropping_trees.next,
6616 struct dropping_root_item_record, list);
6617 last = 0;
6618 if (!bits) {
6619 perror("realloc");
6620 exit(1);
6621 }
6622 buf = read_tree_block(root->fs_info->tree_root,
6623 btrfs_root_bytenr(&rec->ri),
6624 btrfs_level_size(root,
6625 btrfs_root_level(&rec->ri)), 0);
6626 if (!buf) {
6627 ret = -EIO;
6628 goto out;
6629 }
6630 add_root_to_pending(buf, &extent_cache, &pending,
6631 &seen, &nodes, &rec->found_key);
6632 while (1) {
6633 ret = run_next_block(trans, root, bits, bits_nr, &last,
6634 &pending, &seen, &reada,
6635 &nodes, &extent_cache,
6636 &chunk_cache, &dev_cache,
6637 &block_group_cache,
6638 &dev_extent_cache,
6639 &rec->ri);
6640 if (ret != 0)
6641 break;
6642 }
6643 free_extent_buffer(buf);
6644 list_del(&rec->list);
6645 free(rec);
6646 }
6647
6648 if (ret >= 0)
6649 ret = check_extent_refs(trans, root, &extent_cache);
6650 if (ret == -EAGAIN) {
6651 ret = btrfs_commit_transaction(trans, root);
6652 if (ret)
6653 goto out;
6654
6655 trans = btrfs_start_transaction(root, 1);
6656 if (IS_ERR(trans)) {
6657 ret = PTR_ERR(trans);
6658 goto out;
6659 }
6660
6661 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
6662 free_extent_cache_tree(&seen);
6663 free_extent_cache_tree(&pending);
6664 free_extent_cache_tree(&reada);
6665 free_extent_cache_tree(&nodes);
6666 free_chunk_cache_tree(&chunk_cache);
6667 free_block_group_tree(&block_group_cache);
6668 free_device_cache_tree(&dev_cache);
6669 free_device_extent_tree(&dev_extent_cache);
6670 free_extent_record_cache(root->fs_info, &extent_cache);
6671 goto again;
6672 }
6673
6674 err = check_chunks(&chunk_cache, &block_group_cache,
6675 &dev_extent_cache, NULL, NULL, 0);
6676 if (err && !ret)
6677 ret = err;
6678
6679 err = check_devices(&dev_cache, &dev_extent_cache);
6680 if (err && !ret)
6681 ret = err;
6682
6683 out:
6684 if (trans) {
6685 err = btrfs_commit_transaction(trans, root);
6686 if (!ret)
6687 ret = err;
6688 }
6689 if (repair) {
6690 free_corrupt_blocks_tree(root->fs_info->corrupt_blocks);
6691 root->fs_info->fsck_extent_cache = NULL;
6692 root->fs_info->free_extent_hook = NULL;
6693 root->fs_info->corrupt_blocks = NULL;
6694 }
6695 free(bits);
6696 free_chunk_cache_tree(&chunk_cache);
6697 free_device_cache_tree(&dev_cache);
6698 free_block_group_tree(&block_group_cache);
6699 free_device_extent_tree(&dev_extent_cache);
6700 free_extent_cache_tree(&seen);
6701 free_extent_cache_tree(&pending);
6702 free_extent_cache_tree(&reada);
6703 free_extent_cache_tree(&nodes);
6704 return ret;
6705 }
6706
6707 static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
6708 struct btrfs_root *root, int overwrite)
6709 {
6710 struct extent_buffer *c;
6711 struct extent_buffer *old = root->node;
6712 int level;
6713 int ret;
6714 struct btrfs_disk_key disk_key = {0,0,0};
6715
6716 level = 0;
6717
6718 if (overwrite) {
6719 c = old;
6720 extent_buffer_get(c);
6721 goto init;
6722 }
6723 c = btrfs_alloc_free_block(trans, root,
6724 btrfs_level_size(root, 0),
6725 root->root_key.objectid,
6726 &disk_key, level, 0, 0);
6727 if (IS_ERR(c)) {
6728 c = old;
6729 extent_buffer_get(c);
6730 overwrite = 1;
6731 }
6732 init:
6733 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
6734 btrfs_set_header_level(c, level);
6735 btrfs_set_header_bytenr(c, c->start);
6736 btrfs_set_header_generation(c, trans->transid);
6737 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
6738 btrfs_set_header_owner(c, root->root_key.objectid);
6739
6740 write_extent_buffer(c, root->fs_info->fsid,
6741 btrfs_header_fsid(), BTRFS_FSID_SIZE);
6742
6743 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
6744 btrfs_header_chunk_tree_uuid(c),
6745 BTRFS_UUID_SIZE);
6746
6747 btrfs_mark_buffer_dirty(c);
6748 /*
6749 * this case can happen in the following case:
6750 *
6751 * 1.overwrite previous root.
6752 *
6753 * 2.reinit reloc data root, this is because we skip pin
6754 * down reloc data tree before which means we can allocate
6755 * same block bytenr here.
6756 */
6757 if (old->start == c->start) {
6758 btrfs_set_root_generation(&root->root_item,
6759 trans->transid);
6760 root->root_item.level = btrfs_header_level(root->node);
6761 ret = btrfs_update_root(trans, root->fs_info->tree_root,
6762 &root->root_key, &root->root_item);
6763 if (ret) {
6764 free_extent_buffer(c);
6765 return ret;
6766 }
6767 }
6768 free_extent_buffer(old);
6769 root->node = c;
6770 add_root_to_dirty_list(root);
6771 return 0;
6772 }
6773
6774 static int pin_down_tree_blocks(struct btrfs_fs_info *fs_info,
6775 struct extent_buffer *eb, int tree_root)
6776 {
6777 struct extent_buffer *tmp;
6778 struct btrfs_root_item *ri;
6779 struct btrfs_key key;
6780 u64 bytenr;
6781 u32 leafsize;
6782 int level = btrfs_header_level(eb);
6783 int nritems;
6784 int ret;
6785 int i;
6786
6787 /*
6788 * If we have pinned this block before, don't pin it again.
6789 * This can not only avoid forever loop with broken filesystem
6790 * but also give us some speedups.
6791 */
6792 if (test_range_bit(&fs_info->pinned_extents, eb->start,
6793 eb->start + eb->len - 1, EXTENT_DIRTY, 0))
6794 return 0;
6795
6796 btrfs_pin_extent(fs_info, eb->start, eb->len);
6797
6798 leafsize = btrfs_super_leafsize(fs_info->super_copy);
6799 nritems = btrfs_header_nritems(eb);
6800 for (i = 0; i < nritems; i++) {
6801 if (level == 0) {
6802 btrfs_item_key_to_cpu(eb, &key, i);
6803 if (key.type != BTRFS_ROOT_ITEM_KEY)
6804 continue;
6805 /* Skip the extent root and reloc roots */
6806 if (key.objectid == BTRFS_EXTENT_TREE_OBJECTID ||
6807 key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
6808 key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
6809 continue;
6810 ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
6811 bytenr = btrfs_disk_root_bytenr(eb, ri);
6812
6813 /*
6814 * If at any point we start needing the real root we
6815 * will have to build a stump root for the root we are
6816 * in, but for now this doesn't actually use the root so
6817 * just pass in extent_root.
6818 */
6819 tmp = read_tree_block(fs_info->extent_root, bytenr,
6820 leafsize, 0);
6821 if (!tmp) {
6822 fprintf(stderr, "Error reading root block\n");
6823 return -EIO;
6824 }
6825 ret = pin_down_tree_blocks(fs_info, tmp, 0);
6826 free_extent_buffer(tmp);
6827 if (ret)
6828 return ret;
6829 } else {
6830 bytenr = btrfs_node_blockptr(eb, i);
6831
6832 /* If we aren't the tree root don't read the block */
6833 if (level == 1 && !tree_root) {
6834 btrfs_pin_extent(fs_info, bytenr, leafsize);
6835 continue;
6836 }
6837
6838 tmp = read_tree_block(fs_info->extent_root, bytenr,
6839 leafsize, 0);
6840 if (!tmp) {
6841 fprintf(stderr, "Error reading tree block\n");
6842 return -EIO;
6843 }
6844 ret = pin_down_tree_blocks(fs_info, tmp, tree_root);
6845 free_extent_buffer(tmp);
6846 if (ret)
6847 return ret;
6848 }
6849 }
6850
6851 return 0;
6852 }
6853
6854 static int pin_metadata_blocks(struct btrfs_fs_info *fs_info)
6855 {
6856 int ret;
6857
6858 ret = pin_down_tree_blocks(fs_info, fs_info->chunk_root->node, 0);
6859 if (ret)
6860 return ret;
6861
6862 return pin_down_tree_blocks(fs_info, fs_info->tree_root->node, 1);
6863 }
6864
6865 static int reset_block_groups(struct btrfs_fs_info *fs_info)
6866 {
6867 struct btrfs_block_group_cache *cache;
6868 struct btrfs_path *path;
6869 struct extent_buffer *leaf;
6870 struct btrfs_chunk *chunk;
6871 struct btrfs_key key;
6872 int ret;
6873 u64 start;
6874
6875 path = btrfs_alloc_path();
6876 if (!path)
6877 return -ENOMEM;
6878
6879 key.objectid = 0;
6880 key.type = BTRFS_CHUNK_ITEM_KEY;
6881 key.offset = 0;
6882
6883 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
6884 if (ret < 0) {
6885 btrfs_free_path(path);
6886 return ret;
6887 }
6888
6889 /*
6890 * We do this in case the block groups were screwed up and had alloc
6891 * bits that aren't actually set on the chunks. This happens with
6892 * restored images every time and could happen in real life I guess.
6893 */
6894 fs_info->avail_data_alloc_bits = 0;
6895 fs_info->avail_metadata_alloc_bits = 0;
6896 fs_info->avail_system_alloc_bits = 0;
6897
6898 /* First we need to create the in-memory block groups */
6899 while (1) {
6900 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
6901 ret = btrfs_next_leaf(fs_info->chunk_root, path);
6902 if (ret < 0) {
6903 btrfs_free_path(path);
6904 return ret;
6905 }
6906 if (ret) {
6907 ret = 0;
6908 break;
6909 }
6910 }
6911 leaf = path->nodes[0];
6912 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
6913 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
6914 path->slots[0]++;
6915 continue;
6916 }
6917
6918 chunk = btrfs_item_ptr(leaf, path->slots[0],
6919 struct btrfs_chunk);
6920 btrfs_add_block_group(fs_info, 0,
6921 btrfs_chunk_type(leaf, chunk),
6922 key.objectid, key.offset,
6923 btrfs_chunk_length(leaf, chunk));
6924 set_extent_dirty(&fs_info->free_space_cache, key.offset,
6925 key.offset + btrfs_chunk_length(leaf, chunk),
6926 GFP_NOFS);
6927 path->slots[0]++;
6928 }
6929 start = 0;
6930 while (1) {
6931 cache = btrfs_lookup_first_block_group(fs_info, start);
6932 if (!cache)
6933 break;
6934 cache->cached = 1;
6935 start = cache->key.objectid + cache->key.offset;
6936 }
6937
6938 btrfs_free_path(path);
6939 return 0;
6940 }
6941
6942 static int reset_balance(struct btrfs_trans_handle *trans,
6943 struct btrfs_fs_info *fs_info)
6944 {
6945 struct btrfs_root *root = fs_info->tree_root;
6946 struct btrfs_path *path;
6947 struct extent_buffer *leaf;
6948 struct btrfs_key key;
6949 int del_slot, del_nr = 0;
6950 int ret;
6951 int found = 0;
6952
6953 path = btrfs_alloc_path();
6954 if (!path)
6955 return -ENOMEM;
6956
6957 key.objectid = BTRFS_BALANCE_OBJECTID;
6958 key.type = BTRFS_BALANCE_ITEM_KEY;
6959 key.offset = 0;
6960
6961 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6962 if (ret) {
6963 if (ret > 0)
6964 ret = 0;
6965 if (!ret)
6966 goto reinit_data_reloc;
6967 else
6968 goto out;
6969 }
6970
6971 ret = btrfs_del_item(trans, root, path);
6972 if (ret)
6973 goto out;
6974 btrfs_release_path(path);
6975
6976 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
6977 key.type = BTRFS_ROOT_ITEM_KEY;
6978 key.offset = 0;
6979
6980 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
6981 if (ret < 0)
6982 goto out;
6983 while (1) {
6984 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
6985 if (!found)
6986 break;
6987
6988 if (del_nr) {
6989 ret = btrfs_del_items(trans, root, path,
6990 del_slot, del_nr);
6991 del_nr = 0;
6992 if (ret)
6993 goto out;
6994 }
6995 key.offset++;
6996 btrfs_release_path(path);
6997
6998 found = 0;
6999 ret = btrfs_search_slot(trans, root, &key, path,
7000 -1, 1);
7001 if (ret < 0)
7002 goto out;
7003 continue;
7004 }
7005 found = 1;
7006 leaf = path->nodes[0];
7007 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7008 if (key.objectid > BTRFS_TREE_RELOC_OBJECTID)
7009 break;
7010 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
7011 path->slots[0]++;
7012 continue;
7013 }
7014 if (!del_nr) {
7015 del_slot = path->slots[0];
7016 del_nr = 1;
7017 } else {
7018 del_nr++;
7019 }
7020 path->slots[0]++;
7021 }
7022
7023 if (del_nr) {
7024 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
7025 if (ret)
7026 goto out;
7027 }
7028 btrfs_release_path(path);
7029
7030 reinit_data_reloc:
7031 key.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID;
7032 key.type = BTRFS_ROOT_ITEM_KEY;
7033 key.offset = (u64)-1;
7034 root = btrfs_read_fs_root(fs_info, &key);
7035 if (IS_ERR(root)) {
7036 fprintf(stderr, "Error reading data reloc tree\n");
7037 return PTR_ERR(root);
7038 }
7039 record_root_in_trans(trans, root);
7040 ret = btrfs_fsck_reinit_root(trans, root, 0);
7041 if (ret)
7042 goto out;
7043 ret = btrfs_make_root_dir(trans, root, BTRFS_FIRST_FREE_OBJECTID);
7044 out:
7045 btrfs_free_path(path);
7046 return ret;
7047 }
7048
7049 static int reinit_extent_tree(struct btrfs_trans_handle *trans,
7050 struct btrfs_fs_info *fs_info)
7051 {
7052 u64 start = 0;
7053 int ret;
7054
7055 /*
7056 * The only reason we don't do this is because right now we're just
7057 * walking the trees we find and pinning down their bytes, we don't look
7058 * at any of the leaves. In order to do mixed groups we'd have to check
7059 * the leaves of any fs roots and pin down the bytes for any file
7060 * extents we find. Not hard but why do it if we don't have to?
7061 */
7062 if (btrfs_fs_incompat(fs_info, BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)) {
7063 fprintf(stderr, "We don't support re-initing the extent tree "
7064 "for mixed block groups yet, please notify a btrfs "
7065 "developer you want to do this so they can add this "
7066 "functionality.\n");
7067 return -EINVAL;
7068 }
7069
7070 /*
7071 * first we need to walk all of the trees except the extent tree and pin
7072 * down the bytes that are in use so we don't overwrite any existing
7073 * metadata.
7074 */
7075 ret = pin_metadata_blocks(fs_info);
7076 if (ret) {
7077 fprintf(stderr, "error pinning down used bytes\n");
7078 return ret;
7079 }
7080
7081 /*
7082 * Need to drop all the block groups since we're going to recreate all
7083 * of them again.
7084 */
7085 btrfs_free_block_groups(fs_info);
7086 ret = reset_block_groups(fs_info);
7087 if (ret) {
7088 fprintf(stderr, "error resetting the block groups\n");
7089 return ret;
7090 }
7091
7092 /* Ok we can allocate now, reinit the extent root */
7093 ret = btrfs_fsck_reinit_root(trans, fs_info->extent_root, 0);
7094 if (ret) {
7095 fprintf(stderr, "extent root initialization failed\n");
7096 /*
7097 * When the transaction code is updated we should end the
7098 * transaction, but for now progs only knows about commit so
7099 * just return an error.
7100 */
7101 return ret;
7102 }
7103
7104 /*
7105 * Now we have all the in-memory block groups setup so we can make
7106 * allocations properly, and the metadata we care about is safe since we
7107 * pinned all of it above.
7108 */
7109 while (1) {
7110 struct btrfs_block_group_cache *cache;
7111
7112 cache = btrfs_lookup_first_block_group(fs_info, start);
7113 if (!cache)
7114 break;
7115 start = cache->key.objectid + cache->key.offset;
7116 ret = btrfs_insert_item(trans, fs_info->extent_root,
7117 &cache->key, &cache->item,
7118 sizeof(cache->item));
7119 if (ret) {
7120 fprintf(stderr, "Error adding block group\n");
7121 return ret;
7122 }
7123 btrfs_extent_post_op(trans, fs_info->extent_root);
7124 }
7125
7126 ret = reset_balance(trans, fs_info);
7127 if (ret)
7128 fprintf(stderr, "error reseting the pending balance\n");
7129
7130 return ret;
7131 }
7132
7133 static int recow_extent_buffer(struct btrfs_root *root, struct extent_buffer *eb)
7134 {
7135 struct btrfs_path *path;
7136 struct btrfs_trans_handle *trans;
7137 struct btrfs_key key;
7138 int ret;
7139
7140 printf("Recowing metadata block %llu\n", eb->start);
7141 key.objectid = btrfs_header_owner(eb);
7142 key.type = BTRFS_ROOT_ITEM_KEY;
7143 key.offset = (u64)-1;
7144
7145 root = btrfs_read_fs_root(root->fs_info, &key);
7146 if (IS_ERR(root)) {
7147 fprintf(stderr, "Couldn't find owner root %llu\n",
7148 key.objectid);
7149 return PTR_ERR(root);
7150 }
7151
7152 path = btrfs_alloc_path();
7153 if (!path)
7154 return -ENOMEM;
7155
7156 trans = btrfs_start_transaction(root, 1);
7157 if (IS_ERR(trans)) {
7158 btrfs_free_path(path);
7159 return PTR_ERR(trans);
7160 }
7161
7162 path->lowest_level = btrfs_header_level(eb);
7163 if (path->lowest_level)
7164 btrfs_node_key_to_cpu(eb, &key, 0);
7165 else
7166 btrfs_item_key_to_cpu(eb, &key, 0);
7167
7168 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
7169 btrfs_commit_transaction(trans, root);
7170 btrfs_free_path(path);
7171 return ret;
7172 }
7173
7174 static int delete_bad_item(struct btrfs_root *root, struct bad_item *bad)
7175 {
7176 struct btrfs_path *path;
7177 struct btrfs_trans_handle *trans;
7178 struct btrfs_key key;
7179 int ret;
7180
7181 printf("Deleting bad item [%llu,%u,%llu]\n", bad->key.objectid,
7182 bad->key.type, bad->key.offset);
7183 key.objectid = bad->root_id;
7184 key.type = BTRFS_ROOT_ITEM_KEY;
7185 key.offset = (u64)-1;
7186
7187 root = btrfs_read_fs_root(root->fs_info, &key);
7188 if (IS_ERR(root)) {
7189 fprintf(stderr, "Couldn't find owner root %llu\n",
7190 key.objectid);
7191 return PTR_ERR(root);
7192 }
7193
7194 path = btrfs_alloc_path();
7195 if (!path)
7196 return -ENOMEM;
7197
7198 trans = btrfs_start_transaction(root, 1);
7199 if (IS_ERR(trans)) {
7200 btrfs_free_path(path);
7201 return PTR_ERR(trans);
7202 }
7203
7204 ret = btrfs_search_slot(trans, root, &bad->key, path, -1, 1);
7205 if (ret) {
7206 if (ret > 0)
7207 ret = 0;
7208 goto out;
7209 }
7210 ret = btrfs_del_item(trans, root, path);
7211 out:
7212 btrfs_commit_transaction(trans, root);
7213 btrfs_free_path(path);
7214 return ret;
7215 }
7216
7217 static int zero_log_tree(struct btrfs_root *root)
7218 {
7219 struct btrfs_trans_handle *trans;
7220 int ret;
7221
7222 trans = btrfs_start_transaction(root, 1);
7223 if (IS_ERR(trans)) {
7224 ret = PTR_ERR(trans);
7225 return ret;
7226 }
7227 btrfs_set_super_log_root(root->fs_info->super_copy, 0);
7228 btrfs_set_super_log_root_level(root->fs_info->super_copy, 0);
7229 ret = btrfs_commit_transaction(trans, root);
7230 return ret;
7231 }
7232
7233 static int populate_csum(struct btrfs_trans_handle *trans,
7234 struct btrfs_root *csum_root, char *buf, u64 start,
7235 u64 len)
7236 {
7237 u64 offset = 0;
7238 u64 sectorsize;
7239 int ret = 0;
7240
7241 while (offset < len) {
7242 sectorsize = csum_root->sectorsize;
7243 ret = read_extent_data(csum_root, buf, start + offset,
7244 &sectorsize, 0);
7245 if (ret)
7246 break;
7247 ret = btrfs_csum_file_block(trans, csum_root, start + len,
7248 start + offset, buf, sectorsize);
7249 if (ret)
7250 break;
7251 offset += sectorsize;
7252 }
7253 return ret;
7254 }
7255
7256 static int fill_csum_tree(struct btrfs_trans_handle *trans,
7257 struct btrfs_root *csum_root)
7258 {
7259 struct btrfs_root *extent_root = csum_root->fs_info->extent_root;
7260 struct btrfs_path *path;
7261 struct btrfs_extent_item *ei;
7262 struct extent_buffer *leaf;
7263 char *buf;
7264 struct btrfs_key key;
7265 int ret;
7266
7267 path = btrfs_alloc_path();
7268 if (!path)
7269 return -ENOMEM;
7270
7271 key.objectid = 0;
7272 key.type = BTRFS_EXTENT_ITEM_KEY;
7273 key.offset = 0;
7274
7275 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
7276 if (ret < 0) {
7277 btrfs_free_path(path);
7278 return ret;
7279 }
7280
7281 buf = malloc(csum_root->sectorsize);
7282 if (!buf) {
7283 btrfs_free_path(path);
7284 return -ENOMEM;
7285 }
7286
7287 while (1) {
7288 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7289 ret = btrfs_next_leaf(extent_root, path);
7290 if (ret < 0)
7291 break;
7292 if (ret) {
7293 ret = 0;
7294 break;
7295 }
7296 }
7297 leaf = path->nodes[0];
7298
7299 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7300 if (key.type != BTRFS_EXTENT_ITEM_KEY) {
7301 path->slots[0]++;
7302 continue;
7303 }
7304
7305 ei = btrfs_item_ptr(leaf, path->slots[0],
7306 struct btrfs_extent_item);
7307 if (!(btrfs_extent_flags(leaf, ei) &
7308 BTRFS_EXTENT_FLAG_DATA)) {
7309 path->slots[0]++;
7310 continue;
7311 }
7312
7313 ret = populate_csum(trans, csum_root, buf, key.objectid,
7314 key.offset);
7315 if (ret)
7316 break;
7317 path->slots[0]++;
7318 }
7319
7320 btrfs_free_path(path);
7321 free(buf);
7322 return ret;
7323 }
7324
7325 struct root_item_info {
7326 /* level of the root */
7327 u8 level;
7328 /* number of nodes at this level, must be 1 for a root */
7329 int node_count;
7330 u64 bytenr;
7331 u64 gen;
7332 struct cache_extent cache_extent;
7333 };
7334
7335 static struct cache_tree *roots_info_cache = NULL;
7336
7337 static void free_roots_info_cache(void)
7338 {
7339 if (!roots_info_cache)
7340 return;
7341
7342 while (!cache_tree_empty(roots_info_cache)) {
7343 struct cache_extent *entry;
7344 struct root_item_info *rii;
7345
7346 entry = first_cache_extent(roots_info_cache);
7347 remove_cache_extent(roots_info_cache, entry);
7348 rii = container_of(entry, struct root_item_info, cache_extent);
7349 free(rii);
7350 }
7351
7352 free(roots_info_cache);
7353 roots_info_cache = NULL;
7354 }
7355
7356 static int build_roots_info_cache(struct btrfs_fs_info *info)
7357 {
7358 int ret = 0;
7359 struct btrfs_key key;
7360 struct extent_buffer *leaf;
7361 struct btrfs_path *path;
7362
7363 if (!roots_info_cache) {
7364 roots_info_cache = malloc(sizeof(*roots_info_cache));
7365 if (!roots_info_cache)
7366 return -ENOMEM;
7367 cache_tree_init(roots_info_cache);
7368 }
7369
7370 path = btrfs_alloc_path();
7371 if (!path)
7372 return -ENOMEM;
7373
7374 key.objectid = 0;
7375 key.type = BTRFS_EXTENT_ITEM_KEY;
7376 key.offset = 0;
7377
7378 ret = btrfs_search_slot(NULL, info->extent_root, &key, path, 0, 0);
7379 if (ret < 0)
7380 goto out;
7381 leaf = path->nodes[0];
7382
7383 while (1) {
7384 struct btrfs_key found_key;
7385 struct btrfs_extent_item *ei;
7386 struct btrfs_extent_inline_ref *iref;
7387 int slot = path->slots[0];
7388 int type;
7389 u64 flags;
7390 u64 root_id;
7391 u8 level;
7392 struct cache_extent *entry;
7393 struct root_item_info *rii;
7394
7395 if (slot >= btrfs_header_nritems(leaf)) {
7396 ret = btrfs_next_leaf(info->extent_root, path);
7397 if (ret < 0) {
7398 break;
7399 } else if (ret) {
7400 ret = 0;
7401 break;
7402 }
7403 leaf = path->nodes[0];
7404 slot = path->slots[0];
7405 }
7406
7407 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7408
7409 if (found_key.type != BTRFS_EXTENT_ITEM_KEY &&
7410 found_key.type != BTRFS_METADATA_ITEM_KEY)
7411 goto next;
7412
7413 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
7414 flags = btrfs_extent_flags(leaf, ei);
7415
7416 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
7417 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
7418 goto next;
7419
7420 if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
7421 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
7422 level = found_key.offset;
7423 } else {
7424 struct btrfs_tree_block_info *info;
7425
7426 info = (struct btrfs_tree_block_info *)(ei + 1);
7427 iref = (struct btrfs_extent_inline_ref *)(info + 1);
7428 level = btrfs_tree_block_level(leaf, info);
7429 }
7430
7431 /*
7432 * For a root extent, it must be of the following type and the
7433 * first (and only one) iref in the item.
7434 */
7435 type = btrfs_extent_inline_ref_type(leaf, iref);
7436 if (type != BTRFS_TREE_BLOCK_REF_KEY)
7437 goto next;
7438
7439 root_id = btrfs_extent_inline_ref_offset(leaf, iref);
7440 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
7441 if (!entry) {
7442 rii = malloc(sizeof(struct root_item_info));
7443 if (!rii) {
7444 ret = -ENOMEM;
7445 goto out;
7446 }
7447 rii->cache_extent.start = root_id;
7448 rii->cache_extent.size = 1;
7449 rii->level = (u8)-1;
7450 entry = &rii->cache_extent;
7451 ret = insert_cache_extent(roots_info_cache, entry);
7452 ASSERT(ret == 0);
7453 } else {
7454 rii = container_of(entry, struct root_item_info,
7455 cache_extent);
7456 }
7457
7458 ASSERT(rii->cache_extent.start == root_id);
7459 ASSERT(rii->cache_extent.size == 1);
7460
7461 if (level > rii->level || rii->level == (u8)-1) {
7462 rii->level = level;
7463 rii->bytenr = found_key.objectid;
7464 rii->gen = btrfs_extent_generation(leaf, ei);
7465 rii->node_count = 1;
7466 } else if (level == rii->level) {
7467 rii->node_count++;
7468 }
7469 next:
7470 path->slots[0]++;
7471 }
7472
7473 out:
7474 btrfs_free_path(path);
7475
7476 return ret;
7477 }
7478
7479 static int maybe_repair_root_item(struct btrfs_fs_info *info,
7480 struct btrfs_path *path,
7481 const struct btrfs_key *root_key,
7482 const int read_only_mode)
7483 {
7484 const u64 root_id = root_key->objectid;
7485 struct cache_extent *entry;
7486 struct root_item_info *rii;
7487 struct btrfs_root_item ri;
7488 unsigned long offset;
7489
7490 entry = lookup_cache_extent(roots_info_cache, root_id, 1);
7491 if (!entry) {
7492 fprintf(stderr,
7493 "Error: could not find extent items for root %llu\n",
7494 root_key->objectid);
7495 return -ENOENT;
7496 }
7497
7498 rii = container_of(entry, struct root_item_info, cache_extent);
7499 ASSERT(rii->cache_extent.start == root_id);
7500 ASSERT(rii->cache_extent.size == 1);
7501
7502 if (rii->node_count != 1) {
7503 fprintf(stderr,
7504 "Error: could not find btree root extent for root %llu\n",
7505 root_id);
7506 return -ENOENT;
7507 }
7508
7509 offset = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
7510 read_extent_buffer(path->nodes[0], &ri, offset, sizeof(ri));
7511
7512 if (btrfs_root_bytenr(&ri) != rii->bytenr ||
7513 btrfs_root_level(&ri) != rii->level ||
7514 btrfs_root_generation(&ri) != rii->gen) {
7515
7516 /*
7517 * If we're in repair mode but our caller told us to not update
7518 * the root item, i.e. just check if it needs to be updated, don't
7519 * print this message, since the caller will call us again shortly
7520 * for the same root item without read only mode (the caller will
7521 * open a transaction first).
7522 */
7523 if (!(read_only_mode && repair))
7524 fprintf(stderr,
7525 "%sroot item for root %llu,"
7526 " current bytenr %llu, current gen %llu, current level %u,"
7527 " new bytenr %llu, new gen %llu, new level %u\n",
7528 (read_only_mode ? "" : "fixing "),
7529 root_id,
7530 btrfs_root_bytenr(&ri), btrfs_root_generation(&ri),
7531 btrfs_root_level(&ri),
7532 rii->bytenr, rii->gen, rii->level);
7533
7534 if (btrfs_root_generation(&ri) > rii->gen) {
7535 fprintf(stderr,
7536 "root %llu has a root item with a more recent gen (%llu) compared to the found root node (%llu)\n",
7537 root_id, btrfs_root_generation(&ri), rii->gen);
7538 return -EINVAL;
7539 }
7540
7541 if (!read_only_mode) {
7542 btrfs_set_root_bytenr(&ri, rii->bytenr);
7543 btrfs_set_root_level(&ri, rii->level);
7544 btrfs_set_root_generation(&ri, rii->gen);
7545 write_extent_buffer(path->nodes[0], &ri,
7546 offset, sizeof(ri));
7547 }
7548
7549 return 1;
7550 }
7551
7552 return 0;
7553 }
7554
7555 /*
7556 * A regression introduced in the 3.17 kernel (more specifically in 3.17-rc2),
7557 * caused read-only snapshots to be corrupted if they were created at a moment
7558 * when the source subvolume/snapshot had orphan items. The issue was that the
7559 * on-disk root items became incorrect, referring to the pre orphan cleanup root
7560 * node instead of the post orphan cleanup root node.
7561 * So this function, and its callees, just detects and fixes those cases. Even
7562 * though the regression was for read-only snapshots, this function applies to
7563 * any snapshot/subvolume root.
7564 * This must be run before any other repair code - not doing it so, makes other
7565 * repair code delete or modify backrefs in the extent tree for example, which
7566 * will result in an inconsistent fs after repairing the root items.
7567 */
7568 static int repair_root_items(struct btrfs_fs_info *info)
7569 {
7570 struct btrfs_path *path = NULL;
7571 struct btrfs_key key;
7572 struct extent_buffer *leaf;
7573 struct btrfs_trans_handle *trans = NULL;
7574 int ret = 0;
7575 int bad_roots = 0;
7576 int need_trans = 0;
7577
7578 ret = build_roots_info_cache(info);
7579 if (ret)
7580 goto out;
7581
7582 path = btrfs_alloc_path();
7583 if (!path) {
7584 ret = -ENOMEM;
7585 goto out;
7586 }
7587
7588 key.objectid = BTRFS_FIRST_FREE_OBJECTID;
7589 key.type = BTRFS_ROOT_ITEM_KEY;
7590 key.offset = 0;
7591
7592 again:
7593 /*
7594 * Avoid opening and committing transactions if a leaf doesn't have
7595 * any root items that need to be fixed, so that we avoid rotating
7596 * backup roots unnecessarily.
7597 */
7598 if (need_trans) {
7599 trans = btrfs_start_transaction(info->tree_root, 1);
7600 if (IS_ERR(trans)) {
7601 ret = PTR_ERR(trans);
7602 goto out;
7603 }
7604 }
7605
7606 ret = btrfs_search_slot(trans, info->tree_root, &key, path,
7607 0, trans ? 1 : 0);
7608 if (ret < 0)
7609 goto out;
7610 leaf = path->nodes[0];
7611
7612 while (1) {
7613 struct btrfs_key found_key;
7614
7615 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7616 int no_more_keys = find_next_key(path, &key);
7617
7618 btrfs_release_path(path);
7619 if (trans) {
7620 ret = btrfs_commit_transaction(trans,
7621 info->tree_root);
7622 trans = NULL;
7623 if (ret < 0)
7624 goto out;
7625 }
7626 need_trans = 0;
7627 if (no_more_keys)
7628 break;
7629 goto again;
7630 }
7631
7632 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7633
7634 if (found_key.type != BTRFS_ROOT_ITEM_KEY)
7635 goto next;
7636
7637 ret = maybe_repair_root_item(info, path, &found_key,
7638 trans ? 0 : 1);
7639 if (ret < 0)
7640 goto out;
7641 if (ret) {
7642 if (!trans && repair) {
7643 need_trans = 1;
7644 key = found_key;
7645 btrfs_release_path(path);
7646 goto again;
7647 }
7648 bad_roots++;
7649 }
7650 next:
7651 path->slots[0]++;
7652 }
7653 ret = 0;
7654 out:
7655 free_roots_info_cache();
7656 if (path)
7657 btrfs_free_path(path);
7658 if (ret < 0)
7659 return ret;
7660
7661 return bad_roots;
7662 }
7663
7664 static struct option long_options[] = {
7665 { "super", 1, NULL, 's' },
7666 { "repair", 0, NULL, 0 },
7667 { "init-csum-tree", 0, NULL, 0 },
7668 { "init-extent-tree", 0, NULL, 0 },
7669 { "check-data-csum", 0, NULL, 0 },
7670 { "backup", 0, NULL, 0 },
7671 { "subvol-extents", 1, NULL, 'E' },
7672 { "qgroup-report", 0, NULL, 'Q' },
7673 { NULL, 0, NULL, 0}
7674 };
7675
7676 const char * const cmd_check_usage[] = {
7677 "btrfs check [options] <device>",
7678 "Check an unmounted btrfs filesystem.",
7679 "",
7680 "-s|--super <superblock> use this superblock copy",
7681 "-b|--backup use the backup root copy",
7682 "--repair try to repair the filesystem",
7683 "--init-csum-tree create a new CRC tree",
7684 "--init-extent-tree create a new extent tree",
7685 "--check-data-csum verify checkums of data blocks",
7686 "--qgroup-report print a report on qgroup consistency",
7687 "--subvol-extents <subvolid> print subvolume extents and sharing state",
7688 NULL
7689 };
7690
7691 int cmd_check(int argc, char **argv)
7692 {
7693 struct cache_tree root_cache;
7694 struct btrfs_root *root;
7695 struct btrfs_fs_info *info;
7696 u64 bytenr = 0;
7697 u64 subvolid = 0;
7698 char uuidbuf[BTRFS_UUID_UNPARSED_SIZE];
7699 int ret;
7700 u64 num;
7701 int option_index = 0;
7702 int init_csum_tree = 0;
7703 int qgroup_report = 0;
7704 enum btrfs_open_ctree_flags ctree_flags = OPEN_CTREE_EXCLUSIVE;
7705
7706 while(1) {
7707 int c;
7708 c = getopt_long(argc, argv, "as:b", long_options,
7709 &option_index);
7710 if (c < 0)
7711 break;
7712 switch(c) {
7713 case 'a': /* ignored */ break;
7714 case 'b':
7715 ctree_flags |= OPEN_CTREE_BACKUP_ROOT;
7716 break;
7717 case 's':
7718 num = arg_strtou64(optarg);
7719 if (num >= BTRFS_SUPER_MIRROR_MAX) {
7720 fprintf(stderr,
7721 "ERROR: super mirror should be less than: %d\n",
7722 BTRFS_SUPER_MIRROR_MAX);
7723 exit(1);
7724 }
7725 bytenr = btrfs_sb_offset(((int)num));
7726 printf("using SB copy %llu, bytenr %llu\n", num,
7727 (unsigned long long)bytenr);
7728 break;
7729 case 'Q':
7730 qgroup_report = 1;
7731 break;
7732 case 'E':
7733 subvolid = arg_strtou64(optarg);
7734 break;
7735 case '?':
7736 case 'h':
7737 usage(cmd_check_usage);
7738 }
7739 if (option_index == 1) {
7740 printf("enabling repair mode\n");
7741 repair = 1;
7742 ctree_flags |= OPEN_CTREE_WRITES;
7743 } else if (option_index == 2) {
7744 printf("Creating a new CRC tree\n");
7745 init_csum_tree = 1;
7746 repair = 1;
7747 ctree_flags |= OPEN_CTREE_WRITES;
7748 } else if (option_index == 3) {
7749 init_extent_tree = 1;
7750 ctree_flags |= (OPEN_CTREE_WRITES |
7751 OPEN_CTREE_NO_BLOCK_GROUPS);
7752 repair = 1;
7753 } else if (option_index == 4) {
7754 check_data_csum = 1;
7755 }
7756 }
7757 argc = argc - optind;
7758
7759 if (check_argc_exact(argc, 1))
7760 usage(cmd_check_usage);
7761
7762 radix_tree_init();
7763 cache_tree_init(&root_cache);
7764
7765 if((ret = check_mounted(argv[optind])) < 0) {
7766 fprintf(stderr, "Could not check mount status: %s\n", strerror(-ret));
7767 goto err_out;
7768 } else if(ret) {
7769 fprintf(stderr, "%s is currently mounted. Aborting.\n", argv[optind]);
7770 ret = -EBUSY;
7771 goto err_out;
7772 }
7773
7774 /* only allow partial opening under repair mode */
7775 if (repair)
7776 ctree_flags |= OPEN_CTREE_PARTIAL;
7777
7778 info = open_ctree_fs_info(argv[optind], bytenr, 0, ctree_flags);
7779 if (!info) {
7780 fprintf(stderr, "Couldn't open file system\n");
7781 ret = -EIO;
7782 goto err_out;
7783 }
7784
7785 root = info->fs_root;
7786
7787 ret = repair_root_items(info);
7788 if (ret < 0)
7789 goto close_out;
7790 if (repair) {
7791 fprintf(stderr, "Fixed %d roots.\n", ret);
7792 ret = 0;
7793 } else if (ret > 0) {
7794 fprintf(stderr,
7795 "Found %d roots with an outdated root item.\n",
7796 ret);
7797 fprintf(stderr,
7798 "Please run a filesystem check with the option --repair to fix them.\n");
7799 ret = 1;
7800 goto close_out;
7801 }
7802
7803 /*
7804 * repair mode will force us to commit transaction which
7805 * will make us fail to load log tree when mounting.
7806 */
7807 if (repair && btrfs_super_log_root(info->super_copy)) {
7808 ret = ask_user("repair mode will force to clear out log tree, Are you sure?");
7809 if (!ret) {
7810 ret = 1;
7811 goto close_out;
7812 }
7813 ret = zero_log_tree(root);
7814 if (ret) {
7815 fprintf(stderr, "fail to zero log tree\n");
7816 goto close_out;
7817 }
7818 }
7819
7820 uuid_unparse(info->super_copy->fsid, uuidbuf);
7821 if (qgroup_report) {
7822 printf("Print quota groups for %s\nUUID: %s\n", argv[optind],
7823 uuidbuf);
7824 ret = qgroup_verify_all(info);
7825 if (ret == 0)
7826 print_qgroup_report(1);
7827 goto close_out;
7828 }
7829 if (subvolid) {
7830 printf("Print extent state for subvolume %llu on %s\nUUID: %s\n",
7831 subvolid, argv[optind], uuidbuf);
7832 ret = print_extent_state(info, subvolid);
7833 goto close_out;
7834 }
7835 printf("Checking filesystem on %s\nUUID: %s\n", argv[optind], uuidbuf);
7836
7837 if (!extent_buffer_uptodate(info->tree_root->node) ||
7838 !extent_buffer_uptodate(info->dev_root->node) ||
7839 !extent_buffer_uptodate(info->chunk_root->node)) {
7840 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
7841 ret = -EIO;
7842 goto close_out;
7843 }
7844
7845 if (init_extent_tree || init_csum_tree) {
7846 struct btrfs_trans_handle *trans;
7847
7848 trans = btrfs_start_transaction(info->extent_root, 0);
7849 if (IS_ERR(trans)) {
7850 fprintf(stderr, "Error starting transaction\n");
7851 ret = PTR_ERR(trans);
7852 goto close_out;
7853 }
7854
7855 if (init_extent_tree) {
7856 printf("Creating a new extent tree\n");
7857 ret = reinit_extent_tree(trans, info);
7858 if (ret)
7859 goto close_out;
7860 }
7861
7862 if (init_csum_tree) {
7863 fprintf(stderr, "Reinit crc root\n");
7864 ret = btrfs_fsck_reinit_root(trans, info->csum_root, 0);
7865 if (ret) {
7866 fprintf(stderr, "crc root initialization failed\n");
7867 ret = -EIO;
7868 goto close_out;
7869 }
7870
7871 ret = fill_csum_tree(trans, info->csum_root);
7872 if (ret) {
7873 fprintf(stderr, "crc refilling failed\n");
7874 return -EIO;
7875 }
7876 }
7877 /*
7878 * Ok now we commit and run the normal fsck, which will add
7879 * extent entries for all of the items it finds.
7880 */
7881 ret = btrfs_commit_transaction(trans, info->extent_root);
7882 if (ret)
7883 goto close_out;
7884 }
7885 if (!extent_buffer_uptodate(info->extent_root->node)) {
7886 fprintf(stderr, "Critical roots corrupted, unable to fsck the FS\n");
7887 ret = -EIO;
7888 goto close_out;
7889 }
7890 if (!extent_buffer_uptodate(info->csum_root->node)) {
7891 fprintf(stderr, "Checksum root corrupted, rerun with --init-csum-tree option\n");
7892 ret = -EIO;
7893 goto close_out;
7894 }
7895
7896 fprintf(stderr, "checking extents\n");
7897 ret = check_chunks_and_extents(root);
7898 if (ret)
7899 fprintf(stderr, "Errors found in extent allocation tree or chunk allocation\n");
7900
7901 fprintf(stderr, "checking free space cache\n");
7902 ret = check_space_cache(root);
7903 if (ret)
7904 goto out;
7905
7906 /*
7907 * We used to have to have these hole extents in between our real
7908 * extents so if we don't have this flag set we need to make sure there
7909 * are no gaps in the file extents for inodes, otherwise we can just
7910 * ignore it when this happens.
7911 */
7912 no_holes = btrfs_fs_incompat(root->fs_info,
7913 BTRFS_FEATURE_INCOMPAT_NO_HOLES);
7914 fprintf(stderr, "checking fs roots\n");
7915 ret = check_fs_roots(root, &root_cache);
7916 if (ret)
7917 goto out;
7918
7919 fprintf(stderr, "checking csums\n");
7920 ret = check_csums(root);
7921 if (ret)
7922 goto out;
7923
7924 fprintf(stderr, "checking root refs\n");
7925 ret = check_root_refs(root, &root_cache);
7926 if (ret)
7927 goto out;
7928
7929 while (repair && !list_empty(&root->fs_info->recow_ebs)) {
7930 struct extent_buffer *eb;
7931
7932 eb = list_first_entry(&root->fs_info->recow_ebs,
7933 struct extent_buffer, recow);
7934 list_del_init(&eb->recow);
7935 ret = recow_extent_buffer(root, eb);
7936 if (ret)
7937 break;
7938 }
7939
7940 while (!list_empty(&delete_items)) {
7941 struct bad_item *bad;
7942
7943 bad = list_first_entry(&delete_items, struct bad_item, list);
7944 list_del_init(&bad->list);
7945 if (repair)
7946 ret = delete_bad_item(root, bad);
7947 free(bad);
7948 }
7949
7950 if (info->quota_enabled) {
7951 int err;
7952 fprintf(stderr, "checking quota groups\n");
7953 err = qgroup_verify_all(info);
7954 if (err)
7955 goto out;
7956 }
7957
7958 if (!list_empty(&root->fs_info->recow_ebs)) {
7959 fprintf(stderr, "Transid errors in file system\n");
7960 ret = 1;
7961 }
7962 out:
7963 print_qgroup_report(0);
7964 if (found_old_backref) { /*
7965 * there was a disk format change when mixed
7966 * backref was in testing tree. The old format
7967 * existed about one week.
7968 */
7969 printf("\n * Found old mixed backref format. "
7970 "The old format is not supported! *"
7971 "\n * Please mount the FS in readonly mode, "
7972 "backup data and re-format the FS. *\n\n");
7973 ret = 1;
7974 }
7975 printf("found %llu bytes used err is %d\n",
7976 (unsigned long long)bytes_used, ret);
7977 printf("total csum bytes: %llu\n",(unsigned long long)total_csum_bytes);
7978 printf("total tree bytes: %llu\n",
7979 (unsigned long long)total_btree_bytes);
7980 printf("total fs tree bytes: %llu\n",
7981 (unsigned long long)total_fs_tree_bytes);
7982 printf("total extent tree bytes: %llu\n",
7983 (unsigned long long)total_extent_tree_bytes);
7984 printf("btree space waste bytes: %llu\n",
7985 (unsigned long long)btree_space_waste);
7986 printf("file data blocks allocated: %llu\n referenced %llu\n",
7987 (unsigned long long)data_bytes_allocated,
7988 (unsigned long long)data_bytes_referenced);
7989 printf("%s\n", BTRFS_BUILD_VERSION);
7990
7991 free_root_recs_tree(&root_cache);
7992 close_out:
7993 close_ctree(root);
7994 err_out:
7995 return ret;
7996 }
(deprecated) Btrfs progs developments and patch integration