1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) Qu Wenruo 2017. All rights reserved.
4 */
5
6 /*
7 * The module is used to catch unexpected/corrupted tree block data.
8 * Such behavior can be caused either by a fuzzed image or bugs.
9 *
10 * The objective is to do leaf/node validation checks when tree block is read
11 * from disk, and check *every* possible member, so other code won't
12 * need to checking them again.
13 *
14 * Due to the potential and unwanted damage, every checker needs to be
15 * carefully reviewed otherwise so it does not prevent mount of valid images.
16 */
17
18 #include <linux/types.h>
19 #include <linux/stddef.h>
20 #include <linux/error-injection.h>
21 #include "messages.h"
22 #include "ctree.h"
23 #include "tree-checker.h"
24 #include "disk-io.h"
25 #include "compression.h"
26 #include "volumes.h"
27 #include "misc.h"
28 #include "fs.h"
29 #include "accessors.h"
30 #include "file-item.h"
31 #include "inode-item.h"
32 #include "extent-tree.h"
33
34 /*
35 * Error message should follow the following format:
36 * corrupt <type>: <identifier>, <reason>[, <bad_value>]
37 *
38 * @type: leaf or node
39 * @identifier: the necessary info to locate the leaf/node.
40 * It's recommended to decode key.objecitd/offset if it's
41 * meaningful.
42 * @reason: describe the error
43 * @bad_value: optional, it's recommended to output bad value and its
44 * expected value (range).
45 *
46 * Since comma is used to separate the components, only space is allowed
47 * inside each component.
48 */
49
50 /*
51 * Append generic "corrupt leaf/node root=%llu block=%llu slot=%d: " to @fmt.
52 * Allows callers to customize the output.
53 */
54 __printf(3, 4)
55 __cold
generic_err(const struct extent_buffer * eb,int slot,const char * fmt,...)56 static void generic_err(const struct extent_buffer *eb, int slot,
57 const char *fmt, ...)
58 {
59 const struct btrfs_fs_info *fs_info = eb->fs_info;
60 struct va_format vaf;
61 va_list args;
62
63 va_start(args, fmt);
64
65 vaf.fmt = fmt;
66 vaf.va = &args;
67
68 btrfs_crit(fs_info,
69 "corrupt %s: root=%llu block=%llu slot=%d, %pV",
70 btrfs_header_level(eb) == 0 ? "leaf" : "node",
71 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot, &vaf);
72 va_end(args);
73 }
74
75 /*
76 * Customized reporter for extent data item, since its key objectid and
77 * offset has its own meaning.
78 */
79 __printf(3, 4)
80 __cold
file_extent_err(const struct extent_buffer * eb,int slot,const char * fmt,...)81 static void file_extent_err(const struct extent_buffer *eb, int slot,
82 const char *fmt, ...)
83 {
84 const struct btrfs_fs_info *fs_info = eb->fs_info;
85 struct btrfs_key key;
86 struct va_format vaf;
87 va_list args;
88
89 btrfs_item_key_to_cpu(eb, &key, slot);
90 va_start(args, fmt);
91
92 vaf.fmt = fmt;
93 vaf.va = &args;
94
95 btrfs_crit(fs_info,
96 "corrupt %s: root=%llu block=%llu slot=%d ino=%llu file_offset=%llu, %pV",
97 btrfs_header_level(eb) == 0 ? "leaf" : "node",
98 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
99 key.objectid, key.offset, &vaf);
100 va_end(args);
101 }
102
103 /*
104 * Return 0 if the btrfs_file_extent_##name is aligned to @alignment
105 * Else return 1
106 */
107 #define CHECK_FE_ALIGNED(leaf, slot, fi, name, alignment) \
108 ({ \
109 if (unlikely(!IS_ALIGNED(btrfs_file_extent_##name((leaf), (fi)), \
110 (alignment)))) \
111 file_extent_err((leaf), (slot), \
112 "invalid %s for file extent, have %llu, should be aligned to %u", \
113 (#name), btrfs_file_extent_##name((leaf), (fi)), \
114 (alignment)); \
115 (!IS_ALIGNED(btrfs_file_extent_##name((leaf), (fi)), (alignment))); \
116 })
117
file_extent_end(struct extent_buffer * leaf,struct btrfs_key * key,struct btrfs_file_extent_item * extent)118 static u64 file_extent_end(struct extent_buffer *leaf,
119 struct btrfs_key *key,
120 struct btrfs_file_extent_item *extent)
121 {
122 u64 end;
123 u64 len;
124
125 if (btrfs_file_extent_type(leaf, extent) == BTRFS_FILE_EXTENT_INLINE) {
126 len = btrfs_file_extent_ram_bytes(leaf, extent);
127 end = ALIGN(key->offset + len, leaf->fs_info->sectorsize);
128 } else {
129 len = btrfs_file_extent_num_bytes(leaf, extent);
130 end = key->offset + len;
131 }
132 return end;
133 }
134
135 /*
136 * Customized report for dir_item, the only new important information is
137 * key->objectid, which represents inode number
138 */
139 __printf(3, 4)
140 __cold
dir_item_err(const struct extent_buffer * eb,int slot,const char * fmt,...)141 static void dir_item_err(const struct extent_buffer *eb, int slot,
142 const char *fmt, ...)
143 {
144 const struct btrfs_fs_info *fs_info = eb->fs_info;
145 struct btrfs_key key;
146 struct va_format vaf;
147 va_list args;
148
149 btrfs_item_key_to_cpu(eb, &key, slot);
150 va_start(args, fmt);
151
152 vaf.fmt = fmt;
153 vaf.va = &args;
154
155 btrfs_crit(fs_info,
156 "corrupt %s: root=%llu block=%llu slot=%d ino=%llu, %pV",
157 btrfs_header_level(eb) == 0 ? "leaf" : "node",
158 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
159 key.objectid, &vaf);
160 va_end(args);
161 }
162
163 /*
164 * This functions checks prev_key->objectid, to ensure current key and prev_key
165 * share the same objectid as inode number.
166 *
167 * This is to detect missing INODE_ITEM in subvolume trees.
168 *
169 * Return true if everything is OK or we don't need to check.
170 * Return false if anything is wrong.
171 */
check_prev_ino(struct extent_buffer * leaf,struct btrfs_key * key,int slot,struct btrfs_key * prev_key)172 static bool check_prev_ino(struct extent_buffer *leaf,
173 struct btrfs_key *key, int slot,
174 struct btrfs_key *prev_key)
175 {
176 /* No prev key, skip check */
177 if (slot == 0)
178 return true;
179
180 /* Only these key->types needs to be checked */
181 ASSERT(key->type == BTRFS_XATTR_ITEM_KEY ||
182 key->type == BTRFS_INODE_REF_KEY ||
183 key->type == BTRFS_DIR_INDEX_KEY ||
184 key->type == BTRFS_DIR_ITEM_KEY ||
185 key->type == BTRFS_EXTENT_DATA_KEY);
186
187 /*
188 * Only subvolume trees along with their reloc trees need this check.
189 * Things like log tree doesn't follow this ino requirement.
190 */
191 if (!is_fstree(btrfs_header_owner(leaf)))
192 return true;
193
194 if (key->objectid == prev_key->objectid)
195 return true;
196
197 /* Error found */
198 dir_item_err(leaf, slot,
199 "invalid previous key objectid, have %llu expect %llu",
200 prev_key->objectid, key->objectid);
201 return false;
202 }
check_extent_data_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot,struct btrfs_key * prev_key)203 static int check_extent_data_item(struct extent_buffer *leaf,
204 struct btrfs_key *key, int slot,
205 struct btrfs_key *prev_key)
206 {
207 struct btrfs_fs_info *fs_info = leaf->fs_info;
208 struct btrfs_file_extent_item *fi;
209 u32 sectorsize = fs_info->sectorsize;
210 u32 item_size = btrfs_item_size(leaf, slot);
211 u64 extent_end;
212
213 if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) {
214 file_extent_err(leaf, slot,
215 "unaligned file_offset for file extent, have %llu should be aligned to %u",
216 key->offset, sectorsize);
217 return -EUCLEAN;
218 }
219
220 /*
221 * Previous key must have the same key->objectid (ino).
222 * It can be XATTR_ITEM, INODE_ITEM or just another EXTENT_DATA.
223 * But if objectids mismatch, it means we have a missing
224 * INODE_ITEM.
225 */
226 if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
227 return -EUCLEAN;
228
229 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
230
231 /*
232 * Make sure the item contains at least inline header, so the file
233 * extent type is not some garbage.
234 */
235 if (unlikely(item_size < BTRFS_FILE_EXTENT_INLINE_DATA_START)) {
236 file_extent_err(leaf, slot,
237 "invalid item size, have %u expect [%zu, %u)",
238 item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START,
239 SZ_4K);
240 return -EUCLEAN;
241 }
242 if (unlikely(btrfs_file_extent_type(leaf, fi) >=
243 BTRFS_NR_FILE_EXTENT_TYPES)) {
244 file_extent_err(leaf, slot,
245 "invalid type for file extent, have %u expect range [0, %u]",
246 btrfs_file_extent_type(leaf, fi),
247 BTRFS_NR_FILE_EXTENT_TYPES - 1);
248 return -EUCLEAN;
249 }
250
251 /*
252 * Support for new compression/encryption must introduce incompat flag,
253 * and must be caught in open_ctree().
254 */
255 if (unlikely(btrfs_file_extent_compression(leaf, fi) >=
256 BTRFS_NR_COMPRESS_TYPES)) {
257 file_extent_err(leaf, slot,
258 "invalid compression for file extent, have %u expect range [0, %u]",
259 btrfs_file_extent_compression(leaf, fi),
260 BTRFS_NR_COMPRESS_TYPES - 1);
261 return -EUCLEAN;
262 }
263 if (unlikely(btrfs_file_extent_encryption(leaf, fi))) {
264 file_extent_err(leaf, slot,
265 "invalid encryption for file extent, have %u expect 0",
266 btrfs_file_extent_encryption(leaf, fi));
267 return -EUCLEAN;
268 }
269 if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
270 /* Inline extent must have 0 as key offset */
271 if (unlikely(key->offset)) {
272 file_extent_err(leaf, slot,
273 "invalid file_offset for inline file extent, have %llu expect 0",
274 key->offset);
275 return -EUCLEAN;
276 }
277
278 /* Compressed inline extent has no on-disk size, skip it */
279 if (btrfs_file_extent_compression(leaf, fi) !=
280 BTRFS_COMPRESS_NONE)
281 return 0;
282
283 /* Uncompressed inline extent size must match item size */
284 if (unlikely(item_size != BTRFS_FILE_EXTENT_INLINE_DATA_START +
285 btrfs_file_extent_ram_bytes(leaf, fi))) {
286 file_extent_err(leaf, slot,
287 "invalid ram_bytes for uncompressed inline extent, have %u expect %llu",
288 item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START +
289 btrfs_file_extent_ram_bytes(leaf, fi));
290 return -EUCLEAN;
291 }
292 return 0;
293 }
294
295 /* Regular or preallocated extent has fixed item size */
296 if (unlikely(item_size != sizeof(*fi))) {
297 file_extent_err(leaf, slot,
298 "invalid item size for reg/prealloc file extent, have %u expect %zu",
299 item_size, sizeof(*fi));
300 return -EUCLEAN;
301 }
302 if (unlikely(CHECK_FE_ALIGNED(leaf, slot, fi, ram_bytes, sectorsize) ||
303 CHECK_FE_ALIGNED(leaf, slot, fi, disk_bytenr, sectorsize) ||
304 CHECK_FE_ALIGNED(leaf, slot, fi, disk_num_bytes, sectorsize) ||
305 CHECK_FE_ALIGNED(leaf, slot, fi, offset, sectorsize) ||
306 CHECK_FE_ALIGNED(leaf, slot, fi, num_bytes, sectorsize)))
307 return -EUCLEAN;
308
309 /* Catch extent end overflow */
310 if (unlikely(check_add_overflow(btrfs_file_extent_num_bytes(leaf, fi),
311 key->offset, &extent_end))) {
312 file_extent_err(leaf, slot,
313 "extent end overflow, have file offset %llu extent num bytes %llu",
314 key->offset,
315 btrfs_file_extent_num_bytes(leaf, fi));
316 return -EUCLEAN;
317 }
318
319 /*
320 * Check that no two consecutive file extent items, in the same leaf,
321 * present ranges that overlap each other.
322 */
323 if (slot > 0 &&
324 prev_key->objectid == key->objectid &&
325 prev_key->type == BTRFS_EXTENT_DATA_KEY) {
326 struct btrfs_file_extent_item *prev_fi;
327 u64 prev_end;
328
329 prev_fi = btrfs_item_ptr(leaf, slot - 1,
330 struct btrfs_file_extent_item);
331 prev_end = file_extent_end(leaf, prev_key, prev_fi);
332 if (unlikely(prev_end > key->offset)) {
333 file_extent_err(leaf, slot - 1,
334 "file extent end range (%llu) goes beyond start offset (%llu) of the next file extent",
335 prev_end, key->offset);
336 return -EUCLEAN;
337 }
338 }
339
340 return 0;
341 }
342
check_csum_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot,struct btrfs_key * prev_key)343 static int check_csum_item(struct extent_buffer *leaf, struct btrfs_key *key,
344 int slot, struct btrfs_key *prev_key)
345 {
346 struct btrfs_fs_info *fs_info = leaf->fs_info;
347 u32 sectorsize = fs_info->sectorsize;
348 const u32 csumsize = fs_info->csum_size;
349
350 if (unlikely(key->objectid != BTRFS_EXTENT_CSUM_OBJECTID)) {
351 generic_err(leaf, slot,
352 "invalid key objectid for csum item, have %llu expect %llu",
353 key->objectid, BTRFS_EXTENT_CSUM_OBJECTID);
354 return -EUCLEAN;
355 }
356 if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) {
357 generic_err(leaf, slot,
358 "unaligned key offset for csum item, have %llu should be aligned to %u",
359 key->offset, sectorsize);
360 return -EUCLEAN;
361 }
362 if (unlikely(!IS_ALIGNED(btrfs_item_size(leaf, slot), csumsize))) {
363 generic_err(leaf, slot,
364 "unaligned item size for csum item, have %u should be aligned to %u",
365 btrfs_item_size(leaf, slot), csumsize);
366 return -EUCLEAN;
367 }
368 if (slot > 0 && prev_key->type == BTRFS_EXTENT_CSUM_KEY) {
369 u64 prev_csum_end;
370 u32 prev_item_size;
371
372 prev_item_size = btrfs_item_size(leaf, slot - 1);
373 prev_csum_end = (prev_item_size / csumsize) * sectorsize;
374 prev_csum_end += prev_key->offset;
375 if (unlikely(prev_csum_end > key->offset)) {
376 generic_err(leaf, slot - 1,
377 "csum end range (%llu) goes beyond the start range (%llu) of the next csum item",
378 prev_csum_end, key->offset);
379 return -EUCLEAN;
380 }
381 }
382 return 0;
383 }
384
385 /* Inode item error output has the same format as dir_item_err() */
386 #define inode_item_err(eb, slot, fmt, ...) \
387 dir_item_err(eb, slot, fmt, __VA_ARGS__)
388
check_inode_key(struct extent_buffer * leaf,struct btrfs_key * key,int slot)389 static int check_inode_key(struct extent_buffer *leaf, struct btrfs_key *key,
390 int slot)
391 {
392 struct btrfs_key item_key;
393 bool is_inode_item;
394
395 btrfs_item_key_to_cpu(leaf, &item_key, slot);
396 is_inode_item = (item_key.type == BTRFS_INODE_ITEM_KEY);
397
398 /* For XATTR_ITEM, location key should be all 0 */
399 if (item_key.type == BTRFS_XATTR_ITEM_KEY) {
400 if (unlikely(key->objectid != 0 || key->type != 0 ||
401 key->offset != 0))
402 return -EUCLEAN;
403 return 0;
404 }
405
406 if (unlikely((key->objectid < BTRFS_FIRST_FREE_OBJECTID ||
407 key->objectid > BTRFS_LAST_FREE_OBJECTID) &&
408 key->objectid != BTRFS_ROOT_TREE_DIR_OBJECTID &&
409 key->objectid != BTRFS_FREE_INO_OBJECTID)) {
410 if (is_inode_item) {
411 generic_err(leaf, slot,
412 "invalid key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
413 key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
414 BTRFS_FIRST_FREE_OBJECTID,
415 BTRFS_LAST_FREE_OBJECTID,
416 BTRFS_FREE_INO_OBJECTID);
417 } else {
418 dir_item_err(leaf, slot,
419 "invalid location key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
420 key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
421 BTRFS_FIRST_FREE_OBJECTID,
422 BTRFS_LAST_FREE_OBJECTID,
423 BTRFS_FREE_INO_OBJECTID);
424 }
425 return -EUCLEAN;
426 }
427 if (unlikely(key->offset != 0)) {
428 if (is_inode_item)
429 inode_item_err(leaf, slot,
430 "invalid key offset: has %llu expect 0",
431 key->offset);
432 else
433 dir_item_err(leaf, slot,
434 "invalid location key offset:has %llu expect 0",
435 key->offset);
436 return -EUCLEAN;
437 }
438 return 0;
439 }
440
check_root_key(struct extent_buffer * leaf,struct btrfs_key * key,int slot)441 static int check_root_key(struct extent_buffer *leaf, struct btrfs_key *key,
442 int slot)
443 {
444 struct btrfs_key item_key;
445 bool is_root_item;
446
447 btrfs_item_key_to_cpu(leaf, &item_key, slot);
448 is_root_item = (item_key.type == BTRFS_ROOT_ITEM_KEY);
449
450 /*
451 * Bad rootid for reloc trees.
452 *
453 * Reloc trees are only for subvolume trees, other trees only need
454 * to be COWed to be relocated.
455 */
456 if (unlikely(is_root_item && key->objectid == BTRFS_TREE_RELOC_OBJECTID &&
457 !is_fstree(key->offset))) {
458 generic_err(leaf, slot,
459 "invalid reloc tree for root %lld, root id is not a subvolume tree",
460 key->offset);
461 return -EUCLEAN;
462 }
463
464 /* No such tree id */
465 if (unlikely(key->objectid == 0)) {
466 if (is_root_item)
467 generic_err(leaf, slot, "invalid root id 0");
468 else
469 dir_item_err(leaf, slot,
470 "invalid location key root id 0");
471 return -EUCLEAN;
472 }
473
474 /* DIR_ITEM/INDEX/INODE_REF is not allowed to point to non-fs trees */
475 if (unlikely(!is_fstree(key->objectid) && !is_root_item)) {
476 dir_item_err(leaf, slot,
477 "invalid location key objectid, have %llu expect [%llu, %llu]",
478 key->objectid, BTRFS_FIRST_FREE_OBJECTID,
479 BTRFS_LAST_FREE_OBJECTID);
480 return -EUCLEAN;
481 }
482
483 /*
484 * ROOT_ITEM with non-zero offset means this is a snapshot, created at
485 * @offset transid.
486 * Furthermore, for location key in DIR_ITEM, its offset is always -1.
487 *
488 * So here we only check offset for reloc tree whose key->offset must
489 * be a valid tree.
490 */
491 if (unlikely(key->objectid == BTRFS_TREE_RELOC_OBJECTID &&
492 key->offset == 0)) {
493 generic_err(leaf, slot, "invalid root id 0 for reloc tree");
494 return -EUCLEAN;
495 }
496 return 0;
497 }
498
check_dir_item(struct extent_buffer * leaf,struct btrfs_key * key,struct btrfs_key * prev_key,int slot)499 static int check_dir_item(struct extent_buffer *leaf,
500 struct btrfs_key *key, struct btrfs_key *prev_key,
501 int slot)
502 {
503 struct btrfs_fs_info *fs_info = leaf->fs_info;
504 struct btrfs_dir_item *di;
505 u32 item_size = btrfs_item_size(leaf, slot);
506 u32 cur = 0;
507
508 if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
509 return -EUCLEAN;
510
511 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
512 while (cur < item_size) {
513 struct btrfs_key location_key;
514 u32 name_len;
515 u32 data_len;
516 u32 max_name_len;
517 u32 total_size;
518 u32 name_hash;
519 u8 dir_type;
520 int ret;
521
522 /* header itself should not cross item boundary */
523 if (unlikely(cur + sizeof(*di) > item_size)) {
524 dir_item_err(leaf, slot,
525 "dir item header crosses item boundary, have %zu boundary %u",
526 cur + sizeof(*di), item_size);
527 return -EUCLEAN;
528 }
529
530 /* Location key check */
531 btrfs_dir_item_key_to_cpu(leaf, di, &location_key);
532 if (location_key.type == BTRFS_ROOT_ITEM_KEY) {
533 ret = check_root_key(leaf, &location_key, slot);
534 if (unlikely(ret < 0))
535 return ret;
536 } else if (location_key.type == BTRFS_INODE_ITEM_KEY ||
537 location_key.type == 0) {
538 ret = check_inode_key(leaf, &location_key, slot);
539 if (unlikely(ret < 0))
540 return ret;
541 } else {
542 dir_item_err(leaf, slot,
543 "invalid location key type, have %u, expect %u or %u",
544 location_key.type, BTRFS_ROOT_ITEM_KEY,
545 BTRFS_INODE_ITEM_KEY);
546 return -EUCLEAN;
547 }
548
549 /* dir type check */
550 dir_type = btrfs_dir_ftype(leaf, di);
551 if (unlikely(dir_type <= BTRFS_FT_UNKNOWN ||
552 dir_type >= BTRFS_FT_MAX)) {
553 dir_item_err(leaf, slot,
554 "invalid dir item type, have %u expect (0, %u)",
555 dir_type, BTRFS_FT_MAX);
556 return -EUCLEAN;
557 }
558
559 if (unlikely(key->type == BTRFS_XATTR_ITEM_KEY &&
560 dir_type != BTRFS_FT_XATTR)) {
561 dir_item_err(leaf, slot,
562 "invalid dir item type for XATTR key, have %u expect %u",
563 dir_type, BTRFS_FT_XATTR);
564 return -EUCLEAN;
565 }
566 if (unlikely(dir_type == BTRFS_FT_XATTR &&
567 key->type != BTRFS_XATTR_ITEM_KEY)) {
568 dir_item_err(leaf, slot,
569 "xattr dir type found for non-XATTR key");
570 return -EUCLEAN;
571 }
572 if (dir_type == BTRFS_FT_XATTR)
573 max_name_len = XATTR_NAME_MAX;
574 else
575 max_name_len = BTRFS_NAME_LEN;
576
577 /* Name/data length check */
578 name_len = btrfs_dir_name_len(leaf, di);
579 data_len = btrfs_dir_data_len(leaf, di);
580 if (unlikely(name_len > max_name_len)) {
581 dir_item_err(leaf, slot,
582 "dir item name len too long, have %u max %u",
583 name_len, max_name_len);
584 return -EUCLEAN;
585 }
586 if (unlikely(name_len + data_len > BTRFS_MAX_XATTR_SIZE(fs_info))) {
587 dir_item_err(leaf, slot,
588 "dir item name and data len too long, have %u max %u",
589 name_len + data_len,
590 BTRFS_MAX_XATTR_SIZE(fs_info));
591 return -EUCLEAN;
592 }
593
594 if (unlikely(data_len && dir_type != BTRFS_FT_XATTR)) {
595 dir_item_err(leaf, slot,
596 "dir item with invalid data len, have %u expect 0",
597 data_len);
598 return -EUCLEAN;
599 }
600
601 total_size = sizeof(*di) + name_len + data_len;
602
603 /* header and name/data should not cross item boundary */
604 if (unlikely(cur + total_size > item_size)) {
605 dir_item_err(leaf, slot,
606 "dir item data crosses item boundary, have %u boundary %u",
607 cur + total_size, item_size);
608 return -EUCLEAN;
609 }
610
611 /*
612 * Special check for XATTR/DIR_ITEM, as key->offset is name
613 * hash, should match its name
614 */
615 if (key->type == BTRFS_DIR_ITEM_KEY ||
616 key->type == BTRFS_XATTR_ITEM_KEY) {
617 char namebuf[max(BTRFS_NAME_LEN, XATTR_NAME_MAX)];
618
619 read_extent_buffer(leaf, namebuf,
620 (unsigned long)(di + 1), name_len);
621 name_hash = btrfs_name_hash(namebuf, name_len);
622 if (unlikely(key->offset != name_hash)) {
623 dir_item_err(leaf, slot,
624 "name hash mismatch with key, have 0x%016x expect 0x%016llx",
625 name_hash, key->offset);
626 return -EUCLEAN;
627 }
628 }
629 cur += total_size;
630 di = (struct btrfs_dir_item *)((void *)di + total_size);
631 }
632 return 0;
633 }
634
635 __printf(3, 4)
636 __cold
block_group_err(const struct extent_buffer * eb,int slot,const char * fmt,...)637 static void block_group_err(const struct extent_buffer *eb, int slot,
638 const char *fmt, ...)
639 {
640 const struct btrfs_fs_info *fs_info = eb->fs_info;
641 struct btrfs_key key;
642 struct va_format vaf;
643 va_list args;
644
645 btrfs_item_key_to_cpu(eb, &key, slot);
646 va_start(args, fmt);
647
648 vaf.fmt = fmt;
649 vaf.va = &args;
650
651 btrfs_crit(fs_info,
652 "corrupt %s: root=%llu block=%llu slot=%d bg_start=%llu bg_len=%llu, %pV",
653 btrfs_header_level(eb) == 0 ? "leaf" : "node",
654 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
655 key.objectid, key.offset, &vaf);
656 va_end(args);
657 }
658
check_block_group_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot)659 static int check_block_group_item(struct extent_buffer *leaf,
660 struct btrfs_key *key, int slot)
661 {
662 struct btrfs_fs_info *fs_info = leaf->fs_info;
663 struct btrfs_block_group_item bgi;
664 u32 item_size = btrfs_item_size(leaf, slot);
665 u64 chunk_objectid;
666 u64 flags;
667 u64 type;
668
669 /*
670 * Here we don't really care about alignment since extent allocator can
671 * handle it. We care more about the size.
672 */
673 if (unlikely(key->offset == 0)) {
674 block_group_err(leaf, slot,
675 "invalid block group size 0");
676 return -EUCLEAN;
677 }
678
679 if (unlikely(item_size != sizeof(bgi))) {
680 block_group_err(leaf, slot,
681 "invalid item size, have %u expect %zu",
682 item_size, sizeof(bgi));
683 return -EUCLEAN;
684 }
685
686 read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot),
687 sizeof(bgi));
688 chunk_objectid = btrfs_stack_block_group_chunk_objectid(&bgi);
689 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
690 /*
691 * We don't init the nr_global_roots until we load the global
692 * roots, so this could be 0 at mount time. If it's 0 we'll
693 * just assume we're fine, and later we'll check against our
694 * actual value.
695 */
696 if (unlikely(fs_info->nr_global_roots &&
697 chunk_objectid >= fs_info->nr_global_roots)) {
698 block_group_err(leaf, slot,
699 "invalid block group global root id, have %llu, needs to be <= %llu",
700 chunk_objectid,
701 fs_info->nr_global_roots);
702 return -EUCLEAN;
703 }
704 } else if (unlikely(chunk_objectid != BTRFS_FIRST_CHUNK_TREE_OBJECTID)) {
705 block_group_err(leaf, slot,
706 "invalid block group chunk objectid, have %llu expect %llu",
707 btrfs_stack_block_group_chunk_objectid(&bgi),
708 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
709 return -EUCLEAN;
710 }
711
712 if (unlikely(btrfs_stack_block_group_used(&bgi) > key->offset)) {
713 block_group_err(leaf, slot,
714 "invalid block group used, have %llu expect [0, %llu)",
715 btrfs_stack_block_group_used(&bgi), key->offset);
716 return -EUCLEAN;
717 }
718
719 flags = btrfs_stack_block_group_flags(&bgi);
720 if (unlikely(hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) > 1)) {
721 block_group_err(leaf, slot,
722 "invalid profile flags, have 0x%llx (%lu bits set) expect no more than 1 bit set",
723 flags & BTRFS_BLOCK_GROUP_PROFILE_MASK,
724 hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK));
725 return -EUCLEAN;
726 }
727
728 type = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
729 if (unlikely(type != BTRFS_BLOCK_GROUP_DATA &&
730 type != BTRFS_BLOCK_GROUP_METADATA &&
731 type != BTRFS_BLOCK_GROUP_SYSTEM &&
732 type != (BTRFS_BLOCK_GROUP_METADATA |
733 BTRFS_BLOCK_GROUP_DATA))) {
734 block_group_err(leaf, slot,
735 "invalid type, have 0x%llx (%lu bits set) expect either 0x%llx, 0x%llx, 0x%llx or 0x%llx",
736 type, hweight64(type),
737 BTRFS_BLOCK_GROUP_DATA, BTRFS_BLOCK_GROUP_METADATA,
738 BTRFS_BLOCK_GROUP_SYSTEM,
739 BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA);
740 return -EUCLEAN;
741 }
742 return 0;
743 }
744
745 __printf(4, 5)
746 __cold
chunk_err(const struct extent_buffer * leaf,const struct btrfs_chunk * chunk,u64 logical,const char * fmt,...)747 static void chunk_err(const struct extent_buffer *leaf,
748 const struct btrfs_chunk *chunk, u64 logical,
749 const char *fmt, ...)
750 {
751 const struct btrfs_fs_info *fs_info = leaf->fs_info;
752 bool is_sb;
753 struct va_format vaf;
754 va_list args;
755 int i;
756 int slot = -1;
757
758 /* Only superblock eb is able to have such small offset */
759 is_sb = (leaf->start == BTRFS_SUPER_INFO_OFFSET);
760
761 if (!is_sb) {
762 /*
763 * Get the slot number by iterating through all slots, this
764 * would provide better readability.
765 */
766 for (i = 0; i < btrfs_header_nritems(leaf); i++) {
767 if (btrfs_item_ptr_offset(leaf, i) ==
768 (unsigned long)chunk) {
769 slot = i;
770 break;
771 }
772 }
773 }
774 va_start(args, fmt);
775 vaf.fmt = fmt;
776 vaf.va = &args;
777
778 if (is_sb)
779 btrfs_crit(fs_info,
780 "corrupt superblock syschunk array: chunk_start=%llu, %pV",
781 logical, &vaf);
782 else
783 btrfs_crit(fs_info,
784 "corrupt leaf: root=%llu block=%llu slot=%d chunk_start=%llu, %pV",
785 BTRFS_CHUNK_TREE_OBJECTID, leaf->start, slot,
786 logical, &vaf);
787 va_end(args);
788 }
789
790 /*
791 * The common chunk check which could also work on super block sys chunk array.
792 *
793 * Return -EUCLEAN if anything is corrupted.
794 * Return 0 if everything is OK.
795 */
btrfs_check_chunk_valid(struct extent_buffer * leaf,struct btrfs_chunk * chunk,u64 logical)796 int btrfs_check_chunk_valid(struct extent_buffer *leaf,
797 struct btrfs_chunk *chunk, u64 logical)
798 {
799 struct btrfs_fs_info *fs_info = leaf->fs_info;
800 u64 length;
801 u64 chunk_end;
802 u64 stripe_len;
803 u16 num_stripes;
804 u16 sub_stripes;
805 u64 type;
806 u64 features;
807 bool mixed = false;
808 int raid_index;
809 int nparity;
810 int ncopies;
811
812 length = btrfs_chunk_length(leaf, chunk);
813 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
814 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
815 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
816 type = btrfs_chunk_type(leaf, chunk);
817 raid_index = btrfs_bg_flags_to_raid_index(type);
818 ncopies = btrfs_raid_array[raid_index].ncopies;
819 nparity = btrfs_raid_array[raid_index].nparity;
820
821 if (unlikely(!num_stripes)) {
822 chunk_err(leaf, chunk, logical,
823 "invalid chunk num_stripes, have %u", num_stripes);
824 return -EUCLEAN;
825 }
826 if (unlikely(num_stripes < ncopies)) {
827 chunk_err(leaf, chunk, logical,
828 "invalid chunk num_stripes < ncopies, have %u < %d",
829 num_stripes, ncopies);
830 return -EUCLEAN;
831 }
832 if (unlikely(nparity && num_stripes == nparity)) {
833 chunk_err(leaf, chunk, logical,
834 "invalid chunk num_stripes == nparity, have %u == %d",
835 num_stripes, nparity);
836 return -EUCLEAN;
837 }
838 if (unlikely(!IS_ALIGNED(logical, fs_info->sectorsize))) {
839 chunk_err(leaf, chunk, logical,
840 "invalid chunk logical, have %llu should aligned to %u",
841 logical, fs_info->sectorsize);
842 return -EUCLEAN;
843 }
844 if (unlikely(btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize)) {
845 chunk_err(leaf, chunk, logical,
846 "invalid chunk sectorsize, have %u expect %u",
847 btrfs_chunk_sector_size(leaf, chunk),
848 fs_info->sectorsize);
849 return -EUCLEAN;
850 }
851 if (unlikely(!length || !IS_ALIGNED(length, fs_info->sectorsize))) {
852 chunk_err(leaf, chunk, logical,
853 "invalid chunk length, have %llu", length);
854 return -EUCLEAN;
855 }
856 if (unlikely(check_add_overflow(logical, length, &chunk_end))) {
857 chunk_err(leaf, chunk, logical,
858 "invalid chunk logical start and length, have logical start %llu length %llu",
859 logical, length);
860 return -EUCLEAN;
861 }
862 if (unlikely(!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN)) {
863 chunk_err(leaf, chunk, logical,
864 "invalid chunk stripe length: %llu",
865 stripe_len);
866 return -EUCLEAN;
867 }
868 /*
869 * We artificially limit the chunk size, so that the number of stripes
870 * inside a chunk can be fit into a U32. The current limit (256G) is
871 * way too large for real world usage anyway, and it's also much larger
872 * than our existing limit (10G).
873 *
874 * Thus it should be a good way to catch obvious bitflips.
875 */
876 if (unlikely(length >= btrfs_stripe_nr_to_offset(U32_MAX))) {
877 chunk_err(leaf, chunk, logical,
878 "chunk length too large: have %llu limit %llu",
879 length, btrfs_stripe_nr_to_offset(U32_MAX));
880 return -EUCLEAN;
881 }
882 if (unlikely(type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
883 BTRFS_BLOCK_GROUP_PROFILE_MASK))) {
884 chunk_err(leaf, chunk, logical,
885 "unrecognized chunk type: 0x%llx",
886 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
887 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
888 btrfs_chunk_type(leaf, chunk));
889 return -EUCLEAN;
890 }
891
892 if (unlikely(!has_single_bit_set(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
893 (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) != 0)) {
894 chunk_err(leaf, chunk, logical,
895 "invalid chunk profile flag: 0x%llx, expect 0 or 1 bit set",
896 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
897 return -EUCLEAN;
898 }
899 if (unlikely((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0)) {
900 chunk_err(leaf, chunk, logical,
901 "missing chunk type flag, have 0x%llx one bit must be set in 0x%llx",
902 type, BTRFS_BLOCK_GROUP_TYPE_MASK);
903 return -EUCLEAN;
904 }
905
906 if (unlikely((type & BTRFS_BLOCK_GROUP_SYSTEM) &&
907 (type & (BTRFS_BLOCK_GROUP_METADATA |
908 BTRFS_BLOCK_GROUP_DATA)))) {
909 chunk_err(leaf, chunk, logical,
910 "system chunk with data or metadata type: 0x%llx",
911 type);
912 return -EUCLEAN;
913 }
914
915 features = btrfs_super_incompat_flags(fs_info->super_copy);
916 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
917 mixed = true;
918
919 if (!mixed) {
920 if (unlikely((type & BTRFS_BLOCK_GROUP_METADATA) &&
921 (type & BTRFS_BLOCK_GROUP_DATA))) {
922 chunk_err(leaf, chunk, logical,
923 "mixed chunk type in non-mixed mode: 0x%llx", type);
924 return -EUCLEAN;
925 }
926 }
927
928 if (unlikely((type & BTRFS_BLOCK_GROUP_RAID10 &&
929 sub_stripes != btrfs_raid_array[BTRFS_RAID_RAID10].sub_stripes) ||
930 (type & BTRFS_BLOCK_GROUP_RAID1 &&
931 num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1].devs_min) ||
932 (type & BTRFS_BLOCK_GROUP_RAID1C3 &&
933 num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1C3].devs_min) ||
934 (type & BTRFS_BLOCK_GROUP_RAID1C4 &&
935 num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1C4].devs_min) ||
936 (type & BTRFS_BLOCK_GROUP_RAID5 &&
937 num_stripes < btrfs_raid_array[BTRFS_RAID_RAID5].devs_min) ||
938 (type & BTRFS_BLOCK_GROUP_RAID6 &&
939 num_stripes < btrfs_raid_array[BTRFS_RAID_RAID6].devs_min) ||
940 (type & BTRFS_BLOCK_GROUP_DUP &&
941 num_stripes != btrfs_raid_array[BTRFS_RAID_DUP].dev_stripes) ||
942 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
943 num_stripes != btrfs_raid_array[BTRFS_RAID_SINGLE].dev_stripes))) {
944 chunk_err(leaf, chunk, logical,
945 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
946 num_stripes, sub_stripes,
947 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
948 return -EUCLEAN;
949 }
950
951 return 0;
952 }
953
954 /*
955 * Enhanced version of chunk item checker.
956 *
957 * The common btrfs_check_chunk_valid() doesn't check item size since it needs
958 * to work on super block sys_chunk_array which doesn't have full item ptr.
959 */
check_leaf_chunk_item(struct extent_buffer * leaf,struct btrfs_chunk * chunk,struct btrfs_key * key,int slot)960 static int check_leaf_chunk_item(struct extent_buffer *leaf,
961 struct btrfs_chunk *chunk,
962 struct btrfs_key *key, int slot)
963 {
964 int num_stripes;
965
966 if (unlikely(btrfs_item_size(leaf, slot) < sizeof(struct btrfs_chunk))) {
967 chunk_err(leaf, chunk, key->offset,
968 "invalid chunk item size: have %u expect [%zu, %u)",
969 btrfs_item_size(leaf, slot),
970 sizeof(struct btrfs_chunk),
971 BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
972 return -EUCLEAN;
973 }
974
975 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
976 /* Let btrfs_check_chunk_valid() handle this error type */
977 if (num_stripes == 0)
978 goto out;
979
980 if (unlikely(btrfs_chunk_item_size(num_stripes) !=
981 btrfs_item_size(leaf, slot))) {
982 chunk_err(leaf, chunk, key->offset,
983 "invalid chunk item size: have %u expect %lu",
984 btrfs_item_size(leaf, slot),
985 btrfs_chunk_item_size(num_stripes));
986 return -EUCLEAN;
987 }
988 out:
989 return btrfs_check_chunk_valid(leaf, chunk, key->offset);
990 }
991
992 __printf(3, 4)
993 __cold
dev_item_err(const struct extent_buffer * eb,int slot,const char * fmt,...)994 static void dev_item_err(const struct extent_buffer *eb, int slot,
995 const char *fmt, ...)
996 {
997 struct btrfs_key key;
998 struct va_format vaf;
999 va_list args;
1000
1001 btrfs_item_key_to_cpu(eb, &key, slot);
1002 va_start(args, fmt);
1003
1004 vaf.fmt = fmt;
1005 vaf.va = &args;
1006
1007 btrfs_crit(eb->fs_info,
1008 "corrupt %s: root=%llu block=%llu slot=%d devid=%llu %pV",
1009 btrfs_header_level(eb) == 0 ? "leaf" : "node",
1010 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
1011 key.objectid, &vaf);
1012 va_end(args);
1013 }
1014
check_dev_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot)1015 static int check_dev_item(struct extent_buffer *leaf,
1016 struct btrfs_key *key, int slot)
1017 {
1018 struct btrfs_dev_item *ditem;
1019 const u32 item_size = btrfs_item_size(leaf, slot);
1020
1021 if (unlikely(key->objectid != BTRFS_DEV_ITEMS_OBJECTID)) {
1022 dev_item_err(leaf, slot,
1023 "invalid objectid: has=%llu expect=%llu",
1024 key->objectid, BTRFS_DEV_ITEMS_OBJECTID);
1025 return -EUCLEAN;
1026 }
1027
1028 if (unlikely(item_size != sizeof(*ditem))) {
1029 dev_item_err(leaf, slot, "invalid item size: has %u expect %zu",
1030 item_size, sizeof(*ditem));
1031 return -EUCLEAN;
1032 }
1033
1034 ditem = btrfs_item_ptr(leaf, slot, struct btrfs_dev_item);
1035 if (unlikely(btrfs_device_id(leaf, ditem) != key->offset)) {
1036 dev_item_err(leaf, slot,
1037 "devid mismatch: key has=%llu item has=%llu",
1038 key->offset, btrfs_device_id(leaf, ditem));
1039 return -EUCLEAN;
1040 }
1041
1042 /*
1043 * For device total_bytes, we don't have reliable way to check it, as
1044 * it can be 0 for device removal. Device size check can only be done
1045 * by dev extents check.
1046 */
1047 if (unlikely(btrfs_device_bytes_used(leaf, ditem) >
1048 btrfs_device_total_bytes(leaf, ditem))) {
1049 dev_item_err(leaf, slot,
1050 "invalid bytes used: have %llu expect [0, %llu]",
1051 btrfs_device_bytes_used(leaf, ditem),
1052 btrfs_device_total_bytes(leaf, ditem));
1053 return -EUCLEAN;
1054 }
1055 /*
1056 * Remaining members like io_align/type/gen/dev_group aren't really
1057 * utilized. Skip them to make later usage of them easier.
1058 */
1059 return 0;
1060 }
1061
check_inode_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot)1062 static int check_inode_item(struct extent_buffer *leaf,
1063 struct btrfs_key *key, int slot)
1064 {
1065 struct btrfs_fs_info *fs_info = leaf->fs_info;
1066 struct btrfs_inode_item *iitem;
1067 u64 super_gen = btrfs_super_generation(fs_info->super_copy);
1068 u32 valid_mask = (S_IFMT | S_ISUID | S_ISGID | S_ISVTX | 0777);
1069 const u32 item_size = btrfs_item_size(leaf, slot);
1070 u32 mode;
1071 int ret;
1072 u32 flags;
1073 u32 ro_flags;
1074
1075 ret = check_inode_key(leaf, key, slot);
1076 if (unlikely(ret < 0))
1077 return ret;
1078
1079 if (unlikely(item_size != sizeof(*iitem))) {
1080 generic_err(leaf, slot, "invalid item size: has %u expect %zu",
1081 item_size, sizeof(*iitem));
1082 return -EUCLEAN;
1083 }
1084
1085 iitem = btrfs_item_ptr(leaf, slot, struct btrfs_inode_item);
1086
1087 /* Here we use super block generation + 1 to handle log tree */
1088 if (unlikely(btrfs_inode_generation(leaf, iitem) > super_gen + 1)) {
1089 inode_item_err(leaf, slot,
1090 "invalid inode generation: has %llu expect (0, %llu]",
1091 btrfs_inode_generation(leaf, iitem),
1092 super_gen + 1);
1093 return -EUCLEAN;
1094 }
1095 /* Note for ROOT_TREE_DIR_ITEM, mkfs could set its transid 0 */
1096 if (unlikely(btrfs_inode_transid(leaf, iitem) > super_gen + 1)) {
1097 inode_item_err(leaf, slot,
1098 "invalid inode transid: has %llu expect [0, %llu]",
1099 btrfs_inode_transid(leaf, iitem), super_gen + 1);
1100 return -EUCLEAN;
1101 }
1102
1103 /*
1104 * For size and nbytes it's better not to be too strict, as for dir
1105 * item its size/nbytes can easily get wrong, but doesn't affect
1106 * anything in the fs. So here we skip the check.
1107 */
1108 mode = btrfs_inode_mode(leaf, iitem);
1109 if (unlikely(mode & ~valid_mask)) {
1110 inode_item_err(leaf, slot,
1111 "unknown mode bit detected: 0x%x",
1112 mode & ~valid_mask);
1113 return -EUCLEAN;
1114 }
1115
1116 /*
1117 * S_IFMT is not bit mapped so we can't completely rely on
1118 * is_power_of_2/has_single_bit_set, but it can save us from checking
1119 * FIFO/CHR/DIR/REG. Only needs to check BLK, LNK and SOCKS
1120 */
1121 if (!has_single_bit_set(mode & S_IFMT)) {
1122 if (unlikely(!S_ISLNK(mode) && !S_ISBLK(mode) && !S_ISSOCK(mode))) {
1123 inode_item_err(leaf, slot,
1124 "invalid mode: has 0%o expect valid S_IF* bit(s)",
1125 mode & S_IFMT);
1126 return -EUCLEAN;
1127 }
1128 }
1129 if (unlikely(S_ISDIR(mode) && btrfs_inode_nlink(leaf, iitem) > 1)) {
1130 inode_item_err(leaf, slot,
1131 "invalid nlink: has %u expect no more than 1 for dir",
1132 btrfs_inode_nlink(leaf, iitem));
1133 return -EUCLEAN;
1134 }
1135 btrfs_inode_split_flags(btrfs_inode_flags(leaf, iitem), &flags, &ro_flags);
1136 if (unlikely(flags & ~BTRFS_INODE_FLAG_MASK)) {
1137 inode_item_err(leaf, slot,
1138 "unknown incompat flags detected: 0x%x", flags);
1139 return -EUCLEAN;
1140 }
1141 if (unlikely(!sb_rdonly(fs_info->sb) &&
1142 (ro_flags & ~BTRFS_INODE_RO_FLAG_MASK))) {
1143 inode_item_err(leaf, slot,
1144 "unknown ro-compat flags detected on writeable mount: 0x%x",
1145 ro_flags);
1146 return -EUCLEAN;
1147 }
1148 return 0;
1149 }
1150
check_root_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot)1151 static int check_root_item(struct extent_buffer *leaf, struct btrfs_key *key,
1152 int slot)
1153 {
1154 struct btrfs_fs_info *fs_info = leaf->fs_info;
1155 struct btrfs_root_item ri = { 0 };
1156 const u64 valid_root_flags = BTRFS_ROOT_SUBVOL_RDONLY |
1157 BTRFS_ROOT_SUBVOL_DEAD;
1158 int ret;
1159
1160 ret = check_root_key(leaf, key, slot);
1161 if (unlikely(ret < 0))
1162 return ret;
1163
1164 if (unlikely(btrfs_item_size(leaf, slot) != sizeof(ri) &&
1165 btrfs_item_size(leaf, slot) !=
1166 btrfs_legacy_root_item_size())) {
1167 generic_err(leaf, slot,
1168 "invalid root item size, have %u expect %zu or %u",
1169 btrfs_item_size(leaf, slot), sizeof(ri),
1170 btrfs_legacy_root_item_size());
1171 return -EUCLEAN;
1172 }
1173
1174 /*
1175 * For legacy root item, the members starting at generation_v2 will be
1176 * all filled with 0.
1177 * And since we allow geneartion_v2 as 0, it will still pass the check.
1178 */
1179 read_extent_buffer(leaf, &ri, btrfs_item_ptr_offset(leaf, slot),
1180 btrfs_item_size(leaf, slot));
1181
1182 /* Generation related */
1183 if (unlikely(btrfs_root_generation(&ri) >
1184 btrfs_super_generation(fs_info->super_copy) + 1)) {
1185 generic_err(leaf, slot,
1186 "invalid root generation, have %llu expect (0, %llu]",
1187 btrfs_root_generation(&ri),
1188 btrfs_super_generation(fs_info->super_copy) + 1);
1189 return -EUCLEAN;
1190 }
1191 if (unlikely(btrfs_root_generation_v2(&ri) >
1192 btrfs_super_generation(fs_info->super_copy) + 1)) {
1193 generic_err(leaf, slot,
1194 "invalid root v2 generation, have %llu expect (0, %llu]",
1195 btrfs_root_generation_v2(&ri),
1196 btrfs_super_generation(fs_info->super_copy) + 1);
1197 return -EUCLEAN;
1198 }
1199 if (unlikely(btrfs_root_last_snapshot(&ri) >
1200 btrfs_super_generation(fs_info->super_copy) + 1)) {
1201 generic_err(leaf, slot,
1202 "invalid root last_snapshot, have %llu expect (0, %llu]",
1203 btrfs_root_last_snapshot(&ri),
1204 btrfs_super_generation(fs_info->super_copy) + 1);
1205 return -EUCLEAN;
1206 }
1207
1208 /* Alignment and level check */
1209 if (unlikely(!IS_ALIGNED(btrfs_root_bytenr(&ri), fs_info->sectorsize))) {
1210 generic_err(leaf, slot,
1211 "invalid root bytenr, have %llu expect to be aligned to %u",
1212 btrfs_root_bytenr(&ri), fs_info->sectorsize);
1213 return -EUCLEAN;
1214 }
1215 if (unlikely(btrfs_root_level(&ri) >= BTRFS_MAX_LEVEL)) {
1216 generic_err(leaf, slot,
1217 "invalid root level, have %u expect [0, %u]",
1218 btrfs_root_level(&ri), BTRFS_MAX_LEVEL - 1);
1219 return -EUCLEAN;
1220 }
1221 if (unlikely(btrfs_root_drop_level(&ri) >= BTRFS_MAX_LEVEL)) {
1222 generic_err(leaf, slot,
1223 "invalid root level, have %u expect [0, %u]",
1224 btrfs_root_drop_level(&ri), BTRFS_MAX_LEVEL - 1);
1225 return -EUCLEAN;
1226 }
1227
1228 /* Flags check */
1229 if (unlikely(btrfs_root_flags(&ri) & ~valid_root_flags)) {
1230 generic_err(leaf, slot,
1231 "invalid root flags, have 0x%llx expect mask 0x%llx",
1232 btrfs_root_flags(&ri), valid_root_flags);
1233 return -EUCLEAN;
1234 }
1235 return 0;
1236 }
1237
1238 __printf(3,4)
1239 __cold
extent_err(const struct extent_buffer * eb,int slot,const char * fmt,...)1240 static void extent_err(const struct extent_buffer *eb, int slot,
1241 const char *fmt, ...)
1242 {
1243 struct btrfs_key key;
1244 struct va_format vaf;
1245 va_list args;
1246 u64 bytenr;
1247 u64 len;
1248
1249 btrfs_item_key_to_cpu(eb, &key, slot);
1250 bytenr = key.objectid;
1251 if (key.type == BTRFS_METADATA_ITEM_KEY ||
1252 key.type == BTRFS_TREE_BLOCK_REF_KEY ||
1253 key.type == BTRFS_SHARED_BLOCK_REF_KEY)
1254 len = eb->fs_info->nodesize;
1255 else
1256 len = key.offset;
1257 va_start(args, fmt);
1258
1259 vaf.fmt = fmt;
1260 vaf.va = &args;
1261
1262 btrfs_crit(eb->fs_info,
1263 "corrupt %s: block=%llu slot=%d extent bytenr=%llu len=%llu %pV",
1264 btrfs_header_level(eb) == 0 ? "leaf" : "node",
1265 eb->start, slot, bytenr, len, &vaf);
1266 va_end(args);
1267 }
1268
check_extent_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot,struct btrfs_key * prev_key)1269 static int check_extent_item(struct extent_buffer *leaf,
1270 struct btrfs_key *key, int slot,
1271 struct btrfs_key *prev_key)
1272 {
1273 struct btrfs_fs_info *fs_info = leaf->fs_info;
1274 struct btrfs_extent_item *ei;
1275 bool is_tree_block = false;
1276 unsigned long ptr; /* Current pointer inside inline refs */
1277 unsigned long end; /* Extent item end */
1278 const u32 item_size = btrfs_item_size(leaf, slot);
1279 u8 last_type = 0;
1280 u64 last_seq = U64_MAX;
1281 u64 flags;
1282 u64 generation;
1283 u64 total_refs; /* Total refs in btrfs_extent_item */
1284 u64 inline_refs = 0; /* found total inline refs */
1285
1286 if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1287 !btrfs_fs_incompat(fs_info, SKINNY_METADATA))) {
1288 generic_err(leaf, slot,
1289 "invalid key type, METADATA_ITEM type invalid when SKINNY_METADATA feature disabled");
1290 return -EUCLEAN;
1291 }
1292 /* key->objectid is the bytenr for both key types */
1293 if (unlikely(!IS_ALIGNED(key->objectid, fs_info->sectorsize))) {
1294 generic_err(leaf, slot,
1295 "invalid key objectid, have %llu expect to be aligned to %u",
1296 key->objectid, fs_info->sectorsize);
1297 return -EUCLEAN;
1298 }
1299
1300 /* key->offset is tree level for METADATA_ITEM_KEY */
1301 if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1302 key->offset >= BTRFS_MAX_LEVEL)) {
1303 extent_err(leaf, slot,
1304 "invalid tree level, have %llu expect [0, %u]",
1305 key->offset, BTRFS_MAX_LEVEL - 1);
1306 return -EUCLEAN;
1307 }
1308
1309 /*
1310 * EXTENT/METADATA_ITEM consists of:
1311 * 1) One btrfs_extent_item
1312 * Records the total refs, type and generation of the extent.
1313 *
1314 * 2) One btrfs_tree_block_info (for EXTENT_ITEM and tree backref only)
1315 * Records the first key and level of the tree block.
1316 *
1317 * 2) Zero or more btrfs_extent_inline_ref(s)
1318 * Each inline ref has one btrfs_extent_inline_ref shows:
1319 * 2.1) The ref type, one of the 4
1320 * TREE_BLOCK_REF Tree block only
1321 * SHARED_BLOCK_REF Tree block only
1322 * EXTENT_DATA_REF Data only
1323 * SHARED_DATA_REF Data only
1324 * 2.2) Ref type specific data
1325 * Either using btrfs_extent_inline_ref::offset, or specific
1326 * data structure.
1327 *
1328 * All above inline items should follow the order:
1329 *
1330 * - All btrfs_extent_inline_ref::type should be in an ascending
1331 * order
1332 *
1333 * - Within the same type, the items should follow a descending
1334 * order by their sequence number. The sequence number is
1335 * determined by:
1336 * * btrfs_extent_inline_ref::offset for all types other than
1337 * EXTENT_DATA_REF
1338 * * hash_extent_data_ref() for EXTENT_DATA_REF
1339 */
1340 if (unlikely(item_size < sizeof(*ei))) {
1341 extent_err(leaf, slot,
1342 "invalid item size, have %u expect [%zu, %u)",
1343 item_size, sizeof(*ei),
1344 BTRFS_LEAF_DATA_SIZE(fs_info));
1345 return -EUCLEAN;
1346 }
1347 end = item_size + btrfs_item_ptr_offset(leaf, slot);
1348
1349 /* Checks against extent_item */
1350 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
1351 flags = btrfs_extent_flags(leaf, ei);
1352 total_refs = btrfs_extent_refs(leaf, ei);
1353 generation = btrfs_extent_generation(leaf, ei);
1354 if (unlikely(generation >
1355 btrfs_super_generation(fs_info->super_copy) + 1)) {
1356 extent_err(leaf, slot,
1357 "invalid generation, have %llu expect (0, %llu]",
1358 generation,
1359 btrfs_super_generation(fs_info->super_copy) + 1);
1360 return -EUCLEAN;
1361 }
1362 if (unlikely(!has_single_bit_set(flags & (BTRFS_EXTENT_FLAG_DATA |
1363 BTRFS_EXTENT_FLAG_TREE_BLOCK)))) {
1364 extent_err(leaf, slot,
1365 "invalid extent flag, have 0x%llx expect 1 bit set in 0x%llx",
1366 flags, BTRFS_EXTENT_FLAG_DATA |
1367 BTRFS_EXTENT_FLAG_TREE_BLOCK);
1368 return -EUCLEAN;
1369 }
1370 is_tree_block = !!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK);
1371 if (is_tree_block) {
1372 if (unlikely(key->type == BTRFS_EXTENT_ITEM_KEY &&
1373 key->offset != fs_info->nodesize)) {
1374 extent_err(leaf, slot,
1375 "invalid extent length, have %llu expect %u",
1376 key->offset, fs_info->nodesize);
1377 return -EUCLEAN;
1378 }
1379 } else {
1380 if (unlikely(key->type != BTRFS_EXTENT_ITEM_KEY)) {
1381 extent_err(leaf, slot,
1382 "invalid key type, have %u expect %u for data backref",
1383 key->type, BTRFS_EXTENT_ITEM_KEY);
1384 return -EUCLEAN;
1385 }
1386 if (unlikely(!IS_ALIGNED(key->offset, fs_info->sectorsize))) {
1387 extent_err(leaf, slot,
1388 "invalid extent length, have %llu expect aligned to %u",
1389 key->offset, fs_info->sectorsize);
1390 return -EUCLEAN;
1391 }
1392 if (unlikely(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
1393 extent_err(leaf, slot,
1394 "invalid extent flag, data has full backref set");
1395 return -EUCLEAN;
1396 }
1397 }
1398 ptr = (unsigned long)(struct btrfs_extent_item *)(ei + 1);
1399
1400 /* Check the special case of btrfs_tree_block_info */
1401 if (is_tree_block && key->type != BTRFS_METADATA_ITEM_KEY) {
1402 struct btrfs_tree_block_info *info;
1403
1404 info = (struct btrfs_tree_block_info *)ptr;
1405 if (unlikely(btrfs_tree_block_level(leaf, info) >= BTRFS_MAX_LEVEL)) {
1406 extent_err(leaf, slot,
1407 "invalid tree block info level, have %u expect [0, %u]",
1408 btrfs_tree_block_level(leaf, info),
1409 BTRFS_MAX_LEVEL - 1);
1410 return -EUCLEAN;
1411 }
1412 ptr = (unsigned long)(struct btrfs_tree_block_info *)(info + 1);
1413 }
1414
1415 /* Check inline refs */
1416 while (ptr < end) {
1417 struct btrfs_extent_inline_ref *iref;
1418 struct btrfs_extent_data_ref *dref;
1419 struct btrfs_shared_data_ref *sref;
1420 u64 seq;
1421 u64 dref_offset;
1422 u64 inline_offset;
1423 u8 inline_type;
1424
1425 if (unlikely(ptr + sizeof(*iref) > end)) {
1426 extent_err(leaf, slot,
1427 "inline ref item overflows extent item, ptr %lu iref size %zu end %lu",
1428 ptr, sizeof(*iref), end);
1429 return -EUCLEAN;
1430 }
1431 iref = (struct btrfs_extent_inline_ref *)ptr;
1432 inline_type = btrfs_extent_inline_ref_type(leaf, iref);
1433 inline_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1434 seq = inline_offset;
1435 if (unlikely(ptr + btrfs_extent_inline_ref_size(inline_type) > end)) {
1436 extent_err(leaf, slot,
1437 "inline ref item overflows extent item, ptr %lu iref size %u end %lu",
1438 ptr, btrfs_extent_inline_ref_size(inline_type), end);
1439 return -EUCLEAN;
1440 }
1441
1442 switch (inline_type) {
1443 /* inline_offset is subvolid of the owner, no need to check */
1444 case BTRFS_TREE_BLOCK_REF_KEY:
1445 inline_refs++;
1446 break;
1447 /* Contains parent bytenr */
1448 case BTRFS_SHARED_BLOCK_REF_KEY:
1449 if (unlikely(!IS_ALIGNED(inline_offset,
1450 fs_info->sectorsize))) {
1451 extent_err(leaf, slot,
1452 "invalid tree parent bytenr, have %llu expect aligned to %u",
1453 inline_offset, fs_info->sectorsize);
1454 return -EUCLEAN;
1455 }
1456 inline_refs++;
1457 break;
1458 /*
1459 * Contains owner subvolid, owner key objectid, adjusted offset.
1460 * The only obvious corruption can happen in that offset.
1461 */
1462 case BTRFS_EXTENT_DATA_REF_KEY:
1463 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1464 dref_offset = btrfs_extent_data_ref_offset(leaf, dref);
1465 seq = hash_extent_data_ref(
1466 btrfs_extent_data_ref_root(leaf, dref),
1467 btrfs_extent_data_ref_objectid(leaf, dref),
1468 btrfs_extent_data_ref_offset(leaf, dref));
1469 if (unlikely(!IS_ALIGNED(dref_offset,
1470 fs_info->sectorsize))) {
1471 extent_err(leaf, slot,
1472 "invalid data ref offset, have %llu expect aligned to %u",
1473 dref_offset, fs_info->sectorsize);
1474 return -EUCLEAN;
1475 }
1476 inline_refs += btrfs_extent_data_ref_count(leaf, dref);
1477 break;
1478 /* Contains parent bytenr and ref count */
1479 case BTRFS_SHARED_DATA_REF_KEY:
1480 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1481 if (unlikely(!IS_ALIGNED(inline_offset,
1482 fs_info->sectorsize))) {
1483 extent_err(leaf, slot,
1484 "invalid data parent bytenr, have %llu expect aligned to %u",
1485 inline_offset, fs_info->sectorsize);
1486 return -EUCLEAN;
1487 }
1488 inline_refs += btrfs_shared_data_ref_count(leaf, sref);
1489 break;
1490 default:
1491 extent_err(leaf, slot, "unknown inline ref type: %u",
1492 inline_type);
1493 return -EUCLEAN;
1494 }
1495 if (inline_type < last_type) {
1496 extent_err(leaf, slot,
1497 "inline ref out-of-order: has type %u, prev type %u",
1498 inline_type, last_type);
1499 return -EUCLEAN;
1500 }
1501 /* Type changed, allow the sequence starts from U64_MAX again. */
1502 if (inline_type > last_type)
1503 last_seq = U64_MAX;
1504 if (seq > last_seq) {
1505 extent_err(leaf, slot,
1506 "inline ref out-of-order: has type %u offset %llu seq 0x%llx, prev type %u seq 0x%llx",
1507 inline_type, inline_offset, seq,
1508 last_type, last_seq);
1509 return -EUCLEAN;
1510 }
1511 last_type = inline_type;
1512 last_seq = seq;
1513 ptr += btrfs_extent_inline_ref_size(inline_type);
1514 }
1515 /* No padding is allowed */
1516 if (unlikely(ptr != end)) {
1517 extent_err(leaf, slot,
1518 "invalid extent item size, padding bytes found");
1519 return -EUCLEAN;
1520 }
1521
1522 /* Finally, check the inline refs against total refs */
1523 if (unlikely(inline_refs > total_refs)) {
1524 extent_err(leaf, slot,
1525 "invalid extent refs, have %llu expect >= inline %llu",
1526 total_refs, inline_refs);
1527 return -EUCLEAN;
1528 }
1529
1530 if ((prev_key->type == BTRFS_EXTENT_ITEM_KEY) ||
1531 (prev_key->type == BTRFS_METADATA_ITEM_KEY)) {
1532 u64 prev_end = prev_key->objectid;
1533
1534 if (prev_key->type == BTRFS_METADATA_ITEM_KEY)
1535 prev_end += fs_info->nodesize;
1536 else
1537 prev_end += prev_key->offset;
1538
1539 if (unlikely(prev_end > key->objectid)) {
1540 extent_err(leaf, slot,
1541 "previous extent [%llu %u %llu] overlaps current extent [%llu %u %llu]",
1542 prev_key->objectid, prev_key->type,
1543 prev_key->offset, key->objectid, key->type,
1544 key->offset);
1545 return -EUCLEAN;
1546 }
1547 }
1548
1549 return 0;
1550 }
1551
check_simple_keyed_refs(struct extent_buffer * leaf,struct btrfs_key * key,int slot)1552 static int check_simple_keyed_refs(struct extent_buffer *leaf,
1553 struct btrfs_key *key, int slot)
1554 {
1555 u32 expect_item_size = 0;
1556
1557 if (key->type == BTRFS_SHARED_DATA_REF_KEY)
1558 expect_item_size = sizeof(struct btrfs_shared_data_ref);
1559
1560 if (unlikely(btrfs_item_size(leaf, slot) != expect_item_size)) {
1561 generic_err(leaf, slot,
1562 "invalid item size, have %u expect %u for key type %u",
1563 btrfs_item_size(leaf, slot),
1564 expect_item_size, key->type);
1565 return -EUCLEAN;
1566 }
1567 if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1568 generic_err(leaf, slot,
1569 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1570 key->objectid, leaf->fs_info->sectorsize);
1571 return -EUCLEAN;
1572 }
1573 if (unlikely(key->type != BTRFS_TREE_BLOCK_REF_KEY &&
1574 !IS_ALIGNED(key->offset, leaf->fs_info->sectorsize))) {
1575 extent_err(leaf, slot,
1576 "invalid tree parent bytenr, have %llu expect aligned to %u",
1577 key->offset, leaf->fs_info->sectorsize);
1578 return -EUCLEAN;
1579 }
1580 return 0;
1581 }
1582
check_extent_data_ref(struct extent_buffer * leaf,struct btrfs_key * key,int slot)1583 static int check_extent_data_ref(struct extent_buffer *leaf,
1584 struct btrfs_key *key, int slot)
1585 {
1586 struct btrfs_extent_data_ref *dref;
1587 unsigned long ptr = btrfs_item_ptr_offset(leaf, slot);
1588 const unsigned long end = ptr + btrfs_item_size(leaf, slot);
1589
1590 if (unlikely(btrfs_item_size(leaf, slot) % sizeof(*dref) != 0)) {
1591 generic_err(leaf, slot,
1592 "invalid item size, have %u expect aligned to %zu for key type %u",
1593 btrfs_item_size(leaf, slot),
1594 sizeof(*dref), key->type);
1595 return -EUCLEAN;
1596 }
1597 if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1598 generic_err(leaf, slot,
1599 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1600 key->objectid, leaf->fs_info->sectorsize);
1601 return -EUCLEAN;
1602 }
1603 for (; ptr < end; ptr += sizeof(*dref)) {
1604 u64 offset;
1605
1606 /*
1607 * We cannot check the extent_data_ref hash due to possible
1608 * overflow from the leaf due to hash collisions.
1609 */
1610 dref = (struct btrfs_extent_data_ref *)ptr;
1611 offset = btrfs_extent_data_ref_offset(leaf, dref);
1612 if (unlikely(!IS_ALIGNED(offset, leaf->fs_info->sectorsize))) {
1613 extent_err(leaf, slot,
1614 "invalid extent data backref offset, have %llu expect aligned to %u",
1615 offset, leaf->fs_info->sectorsize);
1616 return -EUCLEAN;
1617 }
1618 }
1619 return 0;
1620 }
1621
1622 #define inode_ref_err(eb, slot, fmt, args...) \
1623 inode_item_err(eb, slot, fmt, ##args)
check_inode_ref(struct extent_buffer * leaf,struct btrfs_key * key,struct btrfs_key * prev_key,int slot)1624 static int check_inode_ref(struct extent_buffer *leaf,
1625 struct btrfs_key *key, struct btrfs_key *prev_key,
1626 int slot)
1627 {
1628 struct btrfs_inode_ref *iref;
1629 unsigned long ptr;
1630 unsigned long end;
1631
1632 if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
1633 return -EUCLEAN;
1634 /* namelen can't be 0, so item_size == sizeof() is also invalid */
1635 if (unlikely(btrfs_item_size(leaf, slot) <= sizeof(*iref))) {
1636 inode_ref_err(leaf, slot,
1637 "invalid item size, have %u expect (%zu, %u)",
1638 btrfs_item_size(leaf, slot),
1639 sizeof(*iref), BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
1640 return -EUCLEAN;
1641 }
1642
1643 ptr = btrfs_item_ptr_offset(leaf, slot);
1644 end = ptr + btrfs_item_size(leaf, slot);
1645 while (ptr < end) {
1646 u16 namelen;
1647
1648 if (unlikely(ptr + sizeof(iref) > end)) {
1649 inode_ref_err(leaf, slot,
1650 "inode ref overflow, ptr %lu end %lu inode_ref_size %zu",
1651 ptr, end, sizeof(iref));
1652 return -EUCLEAN;
1653 }
1654
1655 iref = (struct btrfs_inode_ref *)ptr;
1656 namelen = btrfs_inode_ref_name_len(leaf, iref);
1657 if (unlikely(ptr + sizeof(*iref) + namelen > end)) {
1658 inode_ref_err(leaf, slot,
1659 "inode ref overflow, ptr %lu end %lu namelen %u",
1660 ptr, end, namelen);
1661 return -EUCLEAN;
1662 }
1663
1664 /*
1665 * NOTE: In theory we should record all found index numbers
1666 * to find any duplicated indexes, but that will be too time
1667 * consuming for inodes with too many hard links.
1668 */
1669 ptr += sizeof(*iref) + namelen;
1670 }
1671 return 0;
1672 }
1673
check_dev_extent_item(const struct extent_buffer * leaf,const struct btrfs_key * key,int slot,struct btrfs_key * prev_key)1674 static int check_dev_extent_item(const struct extent_buffer *leaf,
1675 const struct btrfs_key *key,
1676 int slot,
1677 struct btrfs_key *prev_key)
1678 {
1679 struct btrfs_dev_extent *de;
1680 const u32 sectorsize = leaf->fs_info->sectorsize;
1681
1682 de = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
1683 /* Basic fixed member checks. */
1684 if (unlikely(btrfs_dev_extent_chunk_tree(leaf, de) !=
1685 BTRFS_CHUNK_TREE_OBJECTID)) {
1686 generic_err(leaf, slot,
1687 "invalid dev extent chunk tree id, has %llu expect %llu",
1688 btrfs_dev_extent_chunk_tree(leaf, de),
1689 BTRFS_CHUNK_TREE_OBJECTID);
1690 return -EUCLEAN;
1691 }
1692 if (unlikely(btrfs_dev_extent_chunk_objectid(leaf, de) !=
1693 BTRFS_FIRST_CHUNK_TREE_OBJECTID)) {
1694 generic_err(leaf, slot,
1695 "invalid dev extent chunk objectid, has %llu expect %llu",
1696 btrfs_dev_extent_chunk_objectid(leaf, de),
1697 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1698 return -EUCLEAN;
1699 }
1700 /* Alignment check. */
1701 if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) {
1702 generic_err(leaf, slot,
1703 "invalid dev extent key.offset, has %llu not aligned to %u",
1704 key->offset, sectorsize);
1705 return -EUCLEAN;
1706 }
1707 if (unlikely(!IS_ALIGNED(btrfs_dev_extent_chunk_offset(leaf, de),
1708 sectorsize))) {
1709 generic_err(leaf, slot,
1710 "invalid dev extent chunk offset, has %llu not aligned to %u",
1711 btrfs_dev_extent_chunk_objectid(leaf, de),
1712 sectorsize);
1713 return -EUCLEAN;
1714 }
1715 if (unlikely(!IS_ALIGNED(btrfs_dev_extent_length(leaf, de),
1716 sectorsize))) {
1717 generic_err(leaf, slot,
1718 "invalid dev extent length, has %llu not aligned to %u",
1719 btrfs_dev_extent_length(leaf, de), sectorsize);
1720 return -EUCLEAN;
1721 }
1722 /* Overlap check with previous dev extent. */
1723 if (slot && prev_key->objectid == key->objectid &&
1724 prev_key->type == key->type) {
1725 struct btrfs_dev_extent *prev_de;
1726 u64 prev_len;
1727
1728 prev_de = btrfs_item_ptr(leaf, slot - 1, struct btrfs_dev_extent);
1729 prev_len = btrfs_dev_extent_length(leaf, prev_de);
1730 if (unlikely(prev_key->offset + prev_len > key->offset)) {
1731 generic_err(leaf, slot,
1732 "dev extent overlap, prev offset %llu len %llu current offset %llu",
1733 prev_key->objectid, prev_len, key->offset);
1734 return -EUCLEAN;
1735 }
1736 }
1737 return 0;
1738 }
1739
1740 /*
1741 * Common point to switch the item-specific validation.
1742 */
check_leaf_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot,struct btrfs_key * prev_key)1743 static enum btrfs_tree_block_status check_leaf_item(struct extent_buffer *leaf,
1744 struct btrfs_key *key,
1745 int slot,
1746 struct btrfs_key *prev_key)
1747 {
1748 int ret = 0;
1749 struct btrfs_chunk *chunk;
1750
1751 switch (key->type) {
1752 case BTRFS_EXTENT_DATA_KEY:
1753 ret = check_extent_data_item(leaf, key, slot, prev_key);
1754 break;
1755 case BTRFS_EXTENT_CSUM_KEY:
1756 ret = check_csum_item(leaf, key, slot, prev_key);
1757 break;
1758 case BTRFS_DIR_ITEM_KEY:
1759 case BTRFS_DIR_INDEX_KEY:
1760 case BTRFS_XATTR_ITEM_KEY:
1761 ret = check_dir_item(leaf, key, prev_key, slot);
1762 break;
1763 case BTRFS_INODE_REF_KEY:
1764 ret = check_inode_ref(leaf, key, prev_key, slot);
1765 break;
1766 case BTRFS_BLOCK_GROUP_ITEM_KEY:
1767 ret = check_block_group_item(leaf, key, slot);
1768 break;
1769 case BTRFS_CHUNK_ITEM_KEY:
1770 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1771 ret = check_leaf_chunk_item(leaf, chunk, key, slot);
1772 break;
1773 case BTRFS_DEV_ITEM_KEY:
1774 ret = check_dev_item(leaf, key, slot);
1775 break;
1776 case BTRFS_DEV_EXTENT_KEY:
1777 ret = check_dev_extent_item(leaf, key, slot, prev_key);
1778 break;
1779 case BTRFS_INODE_ITEM_KEY:
1780 ret = check_inode_item(leaf, key, slot);
1781 break;
1782 case BTRFS_ROOT_ITEM_KEY:
1783 ret = check_root_item(leaf, key, slot);
1784 break;
1785 case BTRFS_EXTENT_ITEM_KEY:
1786 case BTRFS_METADATA_ITEM_KEY:
1787 ret = check_extent_item(leaf, key, slot, prev_key);
1788 break;
1789 case BTRFS_TREE_BLOCK_REF_KEY:
1790 case BTRFS_SHARED_DATA_REF_KEY:
1791 case BTRFS_SHARED_BLOCK_REF_KEY:
1792 ret = check_simple_keyed_refs(leaf, key, slot);
1793 break;
1794 case BTRFS_EXTENT_DATA_REF_KEY:
1795 ret = check_extent_data_ref(leaf, key, slot);
1796 break;
1797 }
1798
1799 if (ret)
1800 return BTRFS_TREE_BLOCK_INVALID_ITEM;
1801 return BTRFS_TREE_BLOCK_CLEAN;
1802 }
1803
__btrfs_check_leaf(struct extent_buffer * leaf)1804 enum btrfs_tree_block_status __btrfs_check_leaf(struct extent_buffer *leaf)
1805 {
1806 struct btrfs_fs_info *fs_info = leaf->fs_info;
1807 /* No valid key type is 0, so all key should be larger than this key */
1808 struct btrfs_key prev_key = {0, 0, 0};
1809 struct btrfs_key key;
1810 u32 nritems = btrfs_header_nritems(leaf);
1811 int slot;
1812
1813 if (unlikely(btrfs_header_level(leaf) != 0)) {
1814 generic_err(leaf, 0,
1815 "invalid level for leaf, have %d expect 0",
1816 btrfs_header_level(leaf));
1817 return BTRFS_TREE_BLOCK_INVALID_LEVEL;
1818 }
1819
1820 /*
1821 * Extent buffers from a relocation tree have a owner field that
1822 * corresponds to the subvolume tree they are based on. So just from an
1823 * extent buffer alone we can not find out what is the id of the
1824 * corresponding subvolume tree, so we can not figure out if the extent
1825 * buffer corresponds to the root of the relocation tree or not. So
1826 * skip this check for relocation trees.
1827 */
1828 if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
1829 u64 owner = btrfs_header_owner(leaf);
1830
1831 /* These trees must never be empty */
1832 if (unlikely(owner == BTRFS_ROOT_TREE_OBJECTID ||
1833 owner == BTRFS_CHUNK_TREE_OBJECTID ||
1834 owner == BTRFS_DEV_TREE_OBJECTID ||
1835 owner == BTRFS_FS_TREE_OBJECTID ||
1836 owner == BTRFS_DATA_RELOC_TREE_OBJECTID)) {
1837 generic_err(leaf, 0,
1838 "invalid root, root %llu must never be empty",
1839 owner);
1840 return BTRFS_TREE_BLOCK_INVALID_NRITEMS;
1841 }
1842
1843 /* Unknown tree */
1844 if (unlikely(owner == 0)) {
1845 generic_err(leaf, 0,
1846 "invalid owner, root 0 is not defined");
1847 return BTRFS_TREE_BLOCK_INVALID_OWNER;
1848 }
1849
1850 /* EXTENT_TREE_V2 can have empty extent trees. */
1851 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1852 return BTRFS_TREE_BLOCK_CLEAN;
1853
1854 if (unlikely(owner == BTRFS_EXTENT_TREE_OBJECTID)) {
1855 generic_err(leaf, 0,
1856 "invalid root, root %llu must never be empty",
1857 owner);
1858 return BTRFS_TREE_BLOCK_INVALID_NRITEMS;
1859 }
1860
1861 return BTRFS_TREE_BLOCK_CLEAN;
1862 }
1863
1864 if (unlikely(nritems == 0))
1865 return BTRFS_TREE_BLOCK_CLEAN;
1866
1867 /*
1868 * Check the following things to make sure this is a good leaf, and
1869 * leaf users won't need to bother with similar sanity checks:
1870 *
1871 * 1) key ordering
1872 * 2) item offset and size
1873 * No overlap, no hole, all inside the leaf.
1874 * 3) item content
1875 * If possible, do comprehensive sanity check.
1876 * NOTE: All checks must only rely on the item data itself.
1877 */
1878 for (slot = 0; slot < nritems; slot++) {
1879 u32 item_end_expected;
1880 u64 item_data_end;
1881
1882 btrfs_item_key_to_cpu(leaf, &key, slot);
1883
1884 /* Make sure the keys are in the right order */
1885 if (unlikely(btrfs_comp_cpu_keys(&prev_key, &key) >= 0)) {
1886 generic_err(leaf, slot,
1887 "bad key order, prev (%llu %u %llu) current (%llu %u %llu)",
1888 prev_key.objectid, prev_key.type,
1889 prev_key.offset, key.objectid, key.type,
1890 key.offset);
1891 return BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
1892 }
1893
1894 item_data_end = (u64)btrfs_item_offset(leaf, slot) +
1895 btrfs_item_size(leaf, slot);
1896 /*
1897 * Make sure the offset and ends are right, remember that the
1898 * item data starts at the end of the leaf and grows towards the
1899 * front.
1900 */
1901 if (slot == 0)
1902 item_end_expected = BTRFS_LEAF_DATA_SIZE(fs_info);
1903 else
1904 item_end_expected = btrfs_item_offset(leaf,
1905 slot - 1);
1906 if (unlikely(item_data_end != item_end_expected)) {
1907 generic_err(leaf, slot,
1908 "unexpected item end, have %llu expect %u",
1909 item_data_end, item_end_expected);
1910 return BTRFS_TREE_BLOCK_INVALID_OFFSETS;
1911 }
1912
1913 /*
1914 * Check to make sure that we don't point outside of the leaf,
1915 * just in case all the items are consistent to each other, but
1916 * all point outside of the leaf.
1917 */
1918 if (unlikely(item_data_end > BTRFS_LEAF_DATA_SIZE(fs_info))) {
1919 generic_err(leaf, slot,
1920 "slot end outside of leaf, have %llu expect range [0, %u]",
1921 item_data_end, BTRFS_LEAF_DATA_SIZE(fs_info));
1922 return BTRFS_TREE_BLOCK_INVALID_OFFSETS;
1923 }
1924
1925 /* Also check if the item pointer overlaps with btrfs item. */
1926 if (unlikely(btrfs_item_ptr_offset(leaf, slot) <
1927 btrfs_item_nr_offset(leaf, slot) + sizeof(struct btrfs_item))) {
1928 generic_err(leaf, slot,
1929 "slot overlaps with its data, item end %lu data start %lu",
1930 btrfs_item_nr_offset(leaf, slot) +
1931 sizeof(struct btrfs_item),
1932 btrfs_item_ptr_offset(leaf, slot));
1933 return BTRFS_TREE_BLOCK_INVALID_OFFSETS;
1934 }
1935
1936 /*
1937 * We only want to do this if WRITTEN is set, otherwise the leaf
1938 * may be in some intermediate state and won't appear valid.
1939 */
1940 if (btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_WRITTEN)) {
1941 enum btrfs_tree_block_status ret;
1942
1943 /*
1944 * Check if the item size and content meet other
1945 * criteria
1946 */
1947 ret = check_leaf_item(leaf, &key, slot, &prev_key);
1948 if (unlikely(ret != BTRFS_TREE_BLOCK_CLEAN))
1949 return ret;
1950 }
1951
1952 prev_key.objectid = key.objectid;
1953 prev_key.type = key.type;
1954 prev_key.offset = key.offset;
1955 }
1956
1957 return BTRFS_TREE_BLOCK_CLEAN;
1958 }
1959
btrfs_check_leaf(struct extent_buffer * leaf)1960 int btrfs_check_leaf(struct extent_buffer *leaf)
1961 {
1962 enum btrfs_tree_block_status ret;
1963
1964 ret = __btrfs_check_leaf(leaf);
1965 if (unlikely(ret != BTRFS_TREE_BLOCK_CLEAN))
1966 return -EUCLEAN;
1967 return 0;
1968 }
1969 ALLOW_ERROR_INJECTION(btrfs_check_leaf, ERRNO);
1970
__btrfs_check_node(struct extent_buffer * node)1971 enum btrfs_tree_block_status __btrfs_check_node(struct extent_buffer *node)
1972 {
1973 struct btrfs_fs_info *fs_info = node->fs_info;
1974 unsigned long nr = btrfs_header_nritems(node);
1975 struct btrfs_key key, next_key;
1976 int slot;
1977 int level = btrfs_header_level(node);
1978 u64 bytenr;
1979
1980 if (unlikely(level <= 0 || level >= BTRFS_MAX_LEVEL)) {
1981 generic_err(node, 0,
1982 "invalid level for node, have %d expect [1, %d]",
1983 level, BTRFS_MAX_LEVEL - 1);
1984 return BTRFS_TREE_BLOCK_INVALID_LEVEL;
1985 }
1986 if (unlikely(nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(fs_info))) {
1987 btrfs_crit(fs_info,
1988 "corrupt node: root=%llu block=%llu, nritems too %s, have %lu expect range [1,%u]",
1989 btrfs_header_owner(node), node->start,
1990 nr == 0 ? "small" : "large", nr,
1991 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1992 return BTRFS_TREE_BLOCK_INVALID_NRITEMS;
1993 }
1994
1995 for (slot = 0; slot < nr - 1; slot++) {
1996 bytenr = btrfs_node_blockptr(node, slot);
1997 btrfs_node_key_to_cpu(node, &key, slot);
1998 btrfs_node_key_to_cpu(node, &next_key, slot + 1);
1999
2000 if (unlikely(!bytenr)) {
2001 generic_err(node, slot,
2002 "invalid NULL node pointer");
2003 return BTRFS_TREE_BLOCK_INVALID_BLOCKPTR;
2004 }
2005 if (unlikely(!IS_ALIGNED(bytenr, fs_info->sectorsize))) {
2006 generic_err(node, slot,
2007 "unaligned pointer, have %llu should be aligned to %u",
2008 bytenr, fs_info->sectorsize);
2009 return BTRFS_TREE_BLOCK_INVALID_BLOCKPTR;
2010 }
2011
2012 if (unlikely(btrfs_comp_cpu_keys(&key, &next_key) >= 0)) {
2013 generic_err(node, slot,
2014 "bad key order, current (%llu %u %llu) next (%llu %u %llu)",
2015 key.objectid, key.type, key.offset,
2016 next_key.objectid, next_key.type,
2017 next_key.offset);
2018 return BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
2019 }
2020 }
2021 return BTRFS_TREE_BLOCK_CLEAN;
2022 }
2023
btrfs_check_node(struct extent_buffer * node)2024 int btrfs_check_node(struct extent_buffer *node)
2025 {
2026 enum btrfs_tree_block_status ret;
2027
2028 ret = __btrfs_check_node(node);
2029 if (unlikely(ret != BTRFS_TREE_BLOCK_CLEAN))
2030 return -EUCLEAN;
2031 return 0;
2032 }
2033 ALLOW_ERROR_INJECTION(btrfs_check_node, ERRNO);
2034
btrfs_check_eb_owner(const struct extent_buffer * eb,u64 root_owner)2035 int btrfs_check_eb_owner(const struct extent_buffer *eb, u64 root_owner)
2036 {
2037 const bool is_subvol = is_fstree(root_owner);
2038 const u64 eb_owner = btrfs_header_owner(eb);
2039
2040 /*
2041 * Skip dummy fs, as selftests don't create unique ebs for each dummy
2042 * root.
2043 */
2044 if (test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &eb->fs_info->fs_state))
2045 return 0;
2046 /*
2047 * There are several call sites (backref walking, qgroup, and data
2048 * reloc) passing 0 as @root_owner, as they are not holding the
2049 * tree root. In that case, we can not do a reliable ownership check,
2050 * so just exit.
2051 */
2052 if (root_owner == 0)
2053 return 0;
2054 /*
2055 * These trees use key.offset as their owner, our callers don't have
2056 * the extra capacity to pass key.offset here. So we just skip them.
2057 */
2058 if (root_owner == BTRFS_TREE_LOG_OBJECTID ||
2059 root_owner == BTRFS_TREE_RELOC_OBJECTID)
2060 return 0;
2061
2062 if (!is_subvol) {
2063 /* For non-subvolume trees, the eb owner should match root owner */
2064 if (unlikely(root_owner != eb_owner)) {
2065 btrfs_crit(eb->fs_info,
2066 "corrupted %s, root=%llu block=%llu owner mismatch, have %llu expect %llu",
2067 btrfs_header_level(eb) == 0 ? "leaf" : "node",
2068 root_owner, btrfs_header_bytenr(eb), eb_owner,
2069 root_owner);
2070 return -EUCLEAN;
2071 }
2072 return 0;
2073 }
2074
2075 /*
2076 * For subvolume trees, owners can mismatch, but they should all belong
2077 * to subvolume trees.
2078 */
2079 if (unlikely(is_subvol != is_fstree(eb_owner))) {
2080 btrfs_crit(eb->fs_info,
2081 "corrupted %s, root=%llu block=%llu owner mismatch, have %llu expect [%llu, %llu]",
2082 btrfs_header_level(eb) == 0 ? "leaf" : "node",
2083 root_owner, btrfs_header_bytenr(eb), eb_owner,
2084 BTRFS_FIRST_FREE_OBJECTID, BTRFS_LAST_FREE_OBJECTID);
2085 return -EUCLEAN;
2086 }
2087 return 0;
2088 }
2089
btrfs_verify_level_key(struct extent_buffer * eb,int level,struct btrfs_key * first_key,u64 parent_transid)2090 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
2091 struct btrfs_key *first_key, u64 parent_transid)
2092 {
2093 struct btrfs_fs_info *fs_info = eb->fs_info;
2094 int found_level;
2095 struct btrfs_key found_key;
2096 int ret;
2097
2098 found_level = btrfs_header_level(eb);
2099 if (found_level != level) {
2100 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
2101 KERN_ERR "BTRFS: tree level check failed\n");
2102 btrfs_err(fs_info,
2103 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
2104 eb->start, level, found_level);
2105 return -EIO;
2106 }
2107
2108 if (!first_key)
2109 return 0;
2110
2111 /*
2112 * For live tree block (new tree blocks in current transaction),
2113 * we need proper lock context to avoid race, which is impossible here.
2114 * So we only checks tree blocks which is read from disk, whose
2115 * generation <= fs_info->last_trans_committed.
2116 */
2117 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
2118 return 0;
2119
2120 /* We have @first_key, so this @eb must have at least one item */
2121 if (btrfs_header_nritems(eb) == 0) {
2122 btrfs_err(fs_info,
2123 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
2124 eb->start);
2125 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
2126 return -EUCLEAN;
2127 }
2128
2129 if (found_level)
2130 btrfs_node_key_to_cpu(eb, &found_key, 0);
2131 else
2132 btrfs_item_key_to_cpu(eb, &found_key, 0);
2133 ret = btrfs_comp_cpu_keys(first_key, &found_key);
2134
2135 if (ret) {
2136 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
2137 KERN_ERR "BTRFS: tree first key check failed\n");
2138 btrfs_err(fs_info,
2139 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
2140 eb->start, parent_transid, first_key->objectid,
2141 first_key->type, first_key->offset,
2142 found_key.objectid, found_key.type,
2143 found_key.offset);
2144 }
2145 return ret;
2146 }
2147