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 "compression.h"
25 #include "volumes.h"
26 #include "misc.h"
27 #include "fs.h"
28 #include "accessors.h"
29 #include "file-item.h"
30 #include "inode-item.h"
31 #include "dir-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
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 dump_page(folio_page(eb->folios[0], 0), "eb page dump");
69 btrfs_crit(fs_info,
70 "corrupt %s: root=%llu block=%llu slot=%d, %pV",
71 btrfs_header_level(eb) == 0 ? "leaf" : "node",
72 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot, &vaf);
73 va_end(args);
74 }
75
76 /*
77 * Customized reporter for extent data item, since its key objectid and
78 * offset has its own meaning.
79 */
80 __printf(3, 4)
81 __cold
82 static void file_extent_err(const struct extent_buffer *eb, int slot,
83 const char *fmt, ...)
84 {
85 const struct btrfs_fs_info *fs_info = eb->fs_info;
86 struct btrfs_key key;
87 struct va_format vaf;
88 va_list args;
89
90 btrfs_item_key_to_cpu(eb, &key, slot);
91 va_start(args, fmt);
92
93 vaf.fmt = fmt;
94 vaf.va = &args;
95
96 dump_page(folio_page(eb->folios[0], 0), "eb page dump");
97 btrfs_crit(fs_info,
98 "corrupt %s: root=%llu block=%llu slot=%d ino=%llu file_offset=%llu, %pV",
99 btrfs_header_level(eb) == 0 ? "leaf" : "node",
100 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
101 key.objectid, key.offset, &vaf);
102 va_end(args);
103 }
104
105 /*
106 * Return 0 if the btrfs_file_extent_##name is aligned to @alignment
107 * Else return 1
108 */
109 #define CHECK_FE_ALIGNED(leaf, slot, fi, name, alignment) \
110 ({ \
111 if (unlikely(!IS_ALIGNED(btrfs_file_extent_##name((leaf), (fi)), \
112 (alignment)))) \
113 file_extent_err((leaf), (slot), \
114 "invalid %s for file extent, have %llu, should be aligned to %u", \
115 (#name), btrfs_file_extent_##name((leaf), (fi)), \
116 (alignment)); \
117 (!IS_ALIGNED(btrfs_file_extent_##name((leaf), (fi)), (alignment))); \
118 })
119
120 static u64 file_extent_end(struct extent_buffer *leaf,
121 struct btrfs_key *key,
122 struct btrfs_file_extent_item *extent)
123 {
124 u64 end;
125 u64 len;
126
127 if (btrfs_file_extent_type(leaf, extent) == BTRFS_FILE_EXTENT_INLINE) {
128 len = btrfs_file_extent_ram_bytes(leaf, extent);
129 end = ALIGN(key->offset + len, leaf->fs_info->sectorsize);
130 } else {
131 len = btrfs_file_extent_num_bytes(leaf, extent);
132 end = key->offset + len;
133 }
134 return end;
135 }
136
137 /*
138 * Customized report for dir_item, the only new important information is
139 * key->objectid, which represents inode number
140 */
141 __printf(3, 4)
142 __cold
143 static void dir_item_err(const struct extent_buffer *eb, int slot,
144 const char *fmt, ...)
145 {
146 const struct btrfs_fs_info *fs_info = eb->fs_info;
147 struct btrfs_key key;
148 struct va_format vaf;
149 va_list args;
150
151 btrfs_item_key_to_cpu(eb, &key, slot);
152 va_start(args, fmt);
153
154 vaf.fmt = fmt;
155 vaf.va = &args;
156
157 dump_page(folio_page(eb->folios[0], 0), "eb page dump");
158 btrfs_crit(fs_info,
159 "corrupt %s: root=%llu block=%llu slot=%d ino=%llu, %pV",
160 btrfs_header_level(eb) == 0 ? "leaf" : "node",
161 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
162 key.objectid, &vaf);
163 va_end(args);
164 }
165
166 /*
167 * This functions checks prev_key->objectid, to ensure current key and prev_key
168 * share the same objectid as inode number.
169 *
170 * This is to detect missing INODE_ITEM in subvolume trees.
171 *
172 * Return true if everything is OK or we don't need to check.
173 * Return false if anything is wrong.
174 */
175 static bool check_prev_ino(struct extent_buffer *leaf,
176 struct btrfs_key *key, int slot,
177 struct btrfs_key *prev_key)
178 {
179 /* No prev key, skip check */
180 if (slot == 0)
181 return true;
182
183 /* Only these key->types needs to be checked */
184 ASSERT(key->type == BTRFS_XATTR_ITEM_KEY ||
185 key->type == BTRFS_INODE_REF_KEY ||
186 key->type == BTRFS_DIR_INDEX_KEY ||
187 key->type == BTRFS_DIR_ITEM_KEY ||
188 key->type == BTRFS_EXTENT_DATA_KEY);
189
190 /*
191 * Only subvolume trees along with their reloc trees need this check.
192 * Things like log tree doesn't follow this ino requirement.
193 */
194 if (!is_fstree(btrfs_header_owner(leaf)))
195 return true;
196
197 if (key->objectid == prev_key->objectid)
198 return true;
199
200 /* Error found */
201 dir_item_err(leaf, slot,
202 "invalid previous key objectid, have %llu expect %llu",
203 prev_key->objectid, key->objectid);
204 return false;
205 }
206 static int check_extent_data_item(struct extent_buffer *leaf,
207 struct btrfs_key *key, int slot,
208 struct btrfs_key *prev_key)
209 {
210 struct btrfs_fs_info *fs_info = leaf->fs_info;
211 struct btrfs_file_extent_item *fi;
212 u32 sectorsize = fs_info->sectorsize;
213 u32 item_size = btrfs_item_size(leaf, slot);
214 u64 extent_end;
215
216 if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) {
217 file_extent_err(leaf, slot,
218 "unaligned file_offset for file extent, have %llu should be aligned to %u",
219 key->offset, sectorsize);
220 return -EUCLEAN;
221 }
222
223 /*
224 * Previous key must have the same key->objectid (ino).
225 * It can be XATTR_ITEM, INODE_ITEM or just another EXTENT_DATA.
226 * But if objectids mismatch, it means we have a missing
227 * INODE_ITEM.
228 */
229 if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
230 return -EUCLEAN;
231
232 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
233
234 /*
235 * Make sure the item contains at least inline header, so the file
236 * extent type is not some garbage.
237 */
238 if (unlikely(item_size < BTRFS_FILE_EXTENT_INLINE_DATA_START)) {
239 file_extent_err(leaf, slot,
240 "invalid item size, have %u expect [%zu, %u)",
241 item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START,
242 SZ_4K);
243 return -EUCLEAN;
244 }
245 if (unlikely(btrfs_file_extent_type(leaf, fi) >=
246 BTRFS_NR_FILE_EXTENT_TYPES)) {
247 file_extent_err(leaf, slot,
248 "invalid type for file extent, have %u expect range [0, %u]",
249 btrfs_file_extent_type(leaf, fi),
250 BTRFS_NR_FILE_EXTENT_TYPES - 1);
251 return -EUCLEAN;
252 }
253
254 /*
255 * Support for new compression/encryption must introduce incompat flag,
256 * and must be caught in open_ctree().
257 */
258 if (unlikely(btrfs_file_extent_compression(leaf, fi) >=
259 BTRFS_NR_COMPRESS_TYPES)) {
260 file_extent_err(leaf, slot,
261 "invalid compression for file extent, have %u expect range [0, %u]",
262 btrfs_file_extent_compression(leaf, fi),
263 BTRFS_NR_COMPRESS_TYPES - 1);
264 return -EUCLEAN;
265 }
266 if (unlikely(btrfs_file_extent_encryption(leaf, fi))) {
267 file_extent_err(leaf, slot,
268 "invalid encryption for file extent, have %u expect 0",
269 btrfs_file_extent_encryption(leaf, fi));
270 return -EUCLEAN;
271 }
272 if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
273 /* Inline extent must have 0 as key offset */
274 if (unlikely(key->offset)) {
275 file_extent_err(leaf, slot,
276 "invalid file_offset for inline file extent, have %llu expect 0",
277 key->offset);
278 return -EUCLEAN;
279 }
280
281 /* Compressed inline extent has no on-disk size, skip it */
282 if (btrfs_file_extent_compression(leaf, fi) !=
283 BTRFS_COMPRESS_NONE)
284 return 0;
285
286 /* Uncompressed inline extent size must match item size */
287 if (unlikely(item_size != BTRFS_FILE_EXTENT_INLINE_DATA_START +
288 btrfs_file_extent_ram_bytes(leaf, fi))) {
289 file_extent_err(leaf, slot,
290 "invalid ram_bytes for uncompressed inline extent, have %u expect %llu",
291 item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START +
292 btrfs_file_extent_ram_bytes(leaf, fi));
293 return -EUCLEAN;
294 }
295 return 0;
296 }
297
298 /* Regular or preallocated extent has fixed item size */
299 if (unlikely(item_size != sizeof(*fi))) {
300 file_extent_err(leaf, slot,
301 "invalid item size for reg/prealloc file extent, have %u expect %zu",
302 item_size, sizeof(*fi));
303 return -EUCLEAN;
304 }
305 if (unlikely(CHECK_FE_ALIGNED(leaf, slot, fi, ram_bytes, sectorsize) ||
306 CHECK_FE_ALIGNED(leaf, slot, fi, disk_bytenr, sectorsize) ||
307 CHECK_FE_ALIGNED(leaf, slot, fi, disk_num_bytes, sectorsize) ||
308 CHECK_FE_ALIGNED(leaf, slot, fi, offset, sectorsize) ||
309 CHECK_FE_ALIGNED(leaf, slot, fi, num_bytes, sectorsize)))
310 return -EUCLEAN;
311
312 /* Catch extent end overflow */
313 if (unlikely(check_add_overflow(btrfs_file_extent_num_bytes(leaf, fi),
314 key->offset, &extent_end))) {
315 file_extent_err(leaf, slot,
316 "extent end overflow, have file offset %llu extent num bytes %llu",
317 key->offset,
318 btrfs_file_extent_num_bytes(leaf, fi));
319 return -EUCLEAN;
320 }
321
322 /*
323 * Check that no two consecutive file extent items, in the same leaf,
324 * present ranges that overlap each other.
325 */
326 if (slot > 0 &&
327 prev_key->objectid == key->objectid &&
328 prev_key->type == BTRFS_EXTENT_DATA_KEY) {
329 struct btrfs_file_extent_item *prev_fi;
330 u64 prev_end;
331
332 prev_fi = btrfs_item_ptr(leaf, slot - 1,
333 struct btrfs_file_extent_item);
334 prev_end = file_extent_end(leaf, prev_key, prev_fi);
335 if (unlikely(prev_end > key->offset)) {
336 file_extent_err(leaf, slot - 1,
337 "file extent end range (%llu) goes beyond start offset (%llu) of the next file extent",
338 prev_end, key->offset);
339 return -EUCLEAN;
340 }
341 }
342
343 /*
344 * For non-compressed data extents, ram_bytes should match its
345 * disk_num_bytes.
346 * However we do not really utilize ram_bytes in this case, so this check
347 * is only optional for DEBUG builds for developers to catch the
348 * unexpected behaviors.
349 */
350 if (IS_ENABLED(CONFIG_BTRFS_DEBUG) &&
351 btrfs_file_extent_compression(leaf, fi) == BTRFS_COMPRESS_NONE &&
352 btrfs_file_extent_disk_bytenr(leaf, fi)) {
353 if (WARN_ON(btrfs_file_extent_ram_bytes(leaf, fi) !=
354 btrfs_file_extent_disk_num_bytes(leaf, fi)))
355 file_extent_err(leaf, slot,
356 "mismatch ram_bytes (%llu) and disk_num_bytes (%llu) for non-compressed extent",
357 btrfs_file_extent_ram_bytes(leaf, fi),
358 btrfs_file_extent_disk_num_bytes(leaf, fi));
359 }
360
361 return 0;
362 }
363
364 static int check_csum_item(struct extent_buffer *leaf, struct btrfs_key *key,
365 int slot, struct btrfs_key *prev_key)
366 {
367 struct btrfs_fs_info *fs_info = leaf->fs_info;
368 u32 sectorsize = fs_info->sectorsize;
369 const u32 csumsize = fs_info->csum_size;
370
371 if (unlikely(key->objectid != BTRFS_EXTENT_CSUM_OBJECTID)) {
372 generic_err(leaf, slot,
373 "invalid key objectid for csum item, have %llu expect %llu",
374 key->objectid, BTRFS_EXTENT_CSUM_OBJECTID);
375 return -EUCLEAN;
376 }
377 if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) {
378 generic_err(leaf, slot,
379 "unaligned key offset for csum item, have %llu should be aligned to %u",
380 key->offset, sectorsize);
381 return -EUCLEAN;
382 }
383 if (unlikely(!IS_ALIGNED(btrfs_item_size(leaf, slot), csumsize))) {
384 generic_err(leaf, slot,
385 "unaligned item size for csum item, have %u should be aligned to %u",
386 btrfs_item_size(leaf, slot), csumsize);
387 return -EUCLEAN;
388 }
389 if (slot > 0 && prev_key->type == BTRFS_EXTENT_CSUM_KEY) {
390 u64 prev_csum_end;
391 u32 prev_item_size;
392
393 prev_item_size = btrfs_item_size(leaf, slot - 1);
394 prev_csum_end = (prev_item_size / csumsize) * sectorsize;
395 prev_csum_end += prev_key->offset;
396 if (unlikely(prev_csum_end > key->offset)) {
397 generic_err(leaf, slot - 1,
398 "csum end range (%llu) goes beyond the start range (%llu) of the next csum item",
399 prev_csum_end, key->offset);
400 return -EUCLEAN;
401 }
402 }
403 return 0;
404 }
405
406 /* Inode item error output has the same format as dir_item_err() */
407 #define inode_item_err(eb, slot, fmt, ...) \
408 dir_item_err(eb, slot, fmt, __VA_ARGS__)
409
410 static int check_inode_key(struct extent_buffer *leaf, struct btrfs_key *key,
411 int slot)
412 {
413 struct btrfs_key item_key;
414 bool is_inode_item;
415
416 btrfs_item_key_to_cpu(leaf, &item_key, slot);
417 is_inode_item = (item_key.type == BTRFS_INODE_ITEM_KEY);
418
419 /* For XATTR_ITEM, location key should be all 0 */
420 if (item_key.type == BTRFS_XATTR_ITEM_KEY) {
421 if (unlikely(key->objectid != 0 || key->type != 0 ||
422 key->offset != 0))
423 return -EUCLEAN;
424 return 0;
425 }
426
427 if (unlikely((key->objectid < BTRFS_FIRST_FREE_OBJECTID ||
428 key->objectid > BTRFS_LAST_FREE_OBJECTID) &&
429 key->objectid != BTRFS_ROOT_TREE_DIR_OBJECTID &&
430 key->objectid != BTRFS_FREE_INO_OBJECTID)) {
431 if (is_inode_item) {
432 generic_err(leaf, slot,
433 "invalid key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
434 key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
435 BTRFS_FIRST_FREE_OBJECTID,
436 BTRFS_LAST_FREE_OBJECTID,
437 BTRFS_FREE_INO_OBJECTID);
438 } else {
439 dir_item_err(leaf, slot,
440 "invalid location key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
441 key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
442 BTRFS_FIRST_FREE_OBJECTID,
443 BTRFS_LAST_FREE_OBJECTID,
444 BTRFS_FREE_INO_OBJECTID);
445 }
446 return -EUCLEAN;
447 }
448 if (unlikely(key->offset != 0)) {
449 if (is_inode_item)
450 inode_item_err(leaf, slot,
451 "invalid key offset: has %llu expect 0",
452 key->offset);
453 else
454 dir_item_err(leaf, slot,
455 "invalid location key offset:has %llu expect 0",
456 key->offset);
457 return -EUCLEAN;
458 }
459 return 0;
460 }
461
462 static int check_root_key(struct extent_buffer *leaf, struct btrfs_key *key,
463 int slot)
464 {
465 struct btrfs_key item_key;
466 bool is_root_item;
467
468 btrfs_item_key_to_cpu(leaf, &item_key, slot);
469 is_root_item = (item_key.type == BTRFS_ROOT_ITEM_KEY);
470
471 /*
472 * Bad rootid for reloc trees.
473 *
474 * Reloc trees are only for subvolume trees, other trees only need
475 * to be COWed to be relocated.
476 */
477 if (unlikely(is_root_item && key->objectid == BTRFS_TREE_RELOC_OBJECTID &&
478 !is_fstree(key->offset))) {
479 generic_err(leaf, slot,
480 "invalid reloc tree for root %lld, root id is not a subvolume tree",
481 key->offset);
482 return -EUCLEAN;
483 }
484
485 /* No such tree id */
486 if (unlikely(key->objectid == 0)) {
487 if (is_root_item)
488 generic_err(leaf, slot, "invalid root id 0");
489 else
490 dir_item_err(leaf, slot,
491 "invalid location key root id 0");
492 return -EUCLEAN;
493 }
494
495 /* DIR_ITEM/INDEX/INODE_REF is not allowed to point to non-fs trees */
496 if (unlikely(!is_fstree(key->objectid) && !is_root_item)) {
497 dir_item_err(leaf, slot,
498 "invalid location key objectid, have %llu expect [%llu, %llu]",
499 key->objectid, BTRFS_FIRST_FREE_OBJECTID,
500 BTRFS_LAST_FREE_OBJECTID);
501 return -EUCLEAN;
502 }
503
504 /*
505 * ROOT_ITEM with non-zero offset means this is a snapshot, created at
506 * @offset transid.
507 * Furthermore, for location key in DIR_ITEM, its offset is always -1.
508 *
509 * So here we only check offset for reloc tree whose key->offset must
510 * be a valid tree.
511 */
512 if (unlikely(key->objectid == BTRFS_TREE_RELOC_OBJECTID &&
513 key->offset == 0)) {
514 generic_err(leaf, slot, "invalid root id 0 for reloc tree");
515 return -EUCLEAN;
516 }
517 return 0;
518 }
519
520 static int check_dir_item(struct extent_buffer *leaf,
521 struct btrfs_key *key, struct btrfs_key *prev_key,
522 int slot)
523 {
524 struct btrfs_fs_info *fs_info = leaf->fs_info;
525 struct btrfs_dir_item *di;
526 u32 item_size = btrfs_item_size(leaf, slot);
527 u32 cur = 0;
528
529 if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
530 return -EUCLEAN;
531
532 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
533 while (cur < item_size) {
534 struct btrfs_key location_key;
535 u32 name_len;
536 u32 data_len;
537 u32 max_name_len;
538 u32 total_size;
539 u32 name_hash;
540 u8 dir_type;
541 int ret;
542
543 /* header itself should not cross item boundary */
544 if (unlikely(cur + sizeof(*di) > item_size)) {
545 dir_item_err(leaf, slot,
546 "dir item header crosses item boundary, have %zu boundary %u",
547 cur + sizeof(*di), item_size);
548 return -EUCLEAN;
549 }
550
551 /* Location key check */
552 btrfs_dir_item_key_to_cpu(leaf, di, &location_key);
553 if (location_key.type == BTRFS_ROOT_ITEM_KEY) {
554 ret = check_root_key(leaf, &location_key, slot);
555 if (unlikely(ret < 0))
556 return ret;
557 } else if (location_key.type == BTRFS_INODE_ITEM_KEY ||
558 location_key.type == 0) {
559 ret = check_inode_key(leaf, &location_key, slot);
560 if (unlikely(ret < 0))
561 return ret;
562 } else {
563 dir_item_err(leaf, slot,
564 "invalid location key type, have %u, expect %u or %u",
565 location_key.type, BTRFS_ROOT_ITEM_KEY,
566 BTRFS_INODE_ITEM_KEY);
567 return -EUCLEAN;
568 }
569
570 /* dir type check */
571 dir_type = btrfs_dir_ftype(leaf, di);
572 if (unlikely(dir_type >= BTRFS_FT_MAX)) {
573 dir_item_err(leaf, slot,
574 "invalid dir item type, have %u expect [0, %u)",
575 dir_type, BTRFS_FT_MAX);
576 return -EUCLEAN;
577 }
578
579 if (unlikely(key->type == BTRFS_XATTR_ITEM_KEY &&
580 dir_type != BTRFS_FT_XATTR)) {
581 dir_item_err(leaf, slot,
582 "invalid dir item type for XATTR key, have %u expect %u",
583 dir_type, BTRFS_FT_XATTR);
584 return -EUCLEAN;
585 }
586 if (unlikely(dir_type == BTRFS_FT_XATTR &&
587 key->type != BTRFS_XATTR_ITEM_KEY)) {
588 dir_item_err(leaf, slot,
589 "xattr dir type found for non-XATTR key");
590 return -EUCLEAN;
591 }
592 if (dir_type == BTRFS_FT_XATTR)
593 max_name_len = XATTR_NAME_MAX;
594 else
595 max_name_len = BTRFS_NAME_LEN;
596
597 /* Name/data length check */
598 name_len = btrfs_dir_name_len(leaf, di);
599 data_len = btrfs_dir_data_len(leaf, di);
600 if (unlikely(name_len > max_name_len)) {
601 dir_item_err(leaf, slot,
602 "dir item name len too long, have %u max %u",
603 name_len, max_name_len);
604 return -EUCLEAN;
605 }
606 if (unlikely(name_len + data_len > BTRFS_MAX_XATTR_SIZE(fs_info))) {
607 dir_item_err(leaf, slot,
608 "dir item name and data len too long, have %u max %u",
609 name_len + data_len,
610 BTRFS_MAX_XATTR_SIZE(fs_info));
611 return -EUCLEAN;
612 }
613
614 if (unlikely(data_len && dir_type != BTRFS_FT_XATTR)) {
615 dir_item_err(leaf, slot,
616 "dir item with invalid data len, have %u expect 0",
617 data_len);
618 return -EUCLEAN;
619 }
620
621 total_size = sizeof(*di) + name_len + data_len;
622
623 /* header and name/data should not cross item boundary */
624 if (unlikely(cur + total_size > item_size)) {
625 dir_item_err(leaf, slot,
626 "dir item data crosses item boundary, have %u boundary %u",
627 cur + total_size, item_size);
628 return -EUCLEAN;
629 }
630
631 /*
632 * Special check for XATTR/DIR_ITEM, as key->offset is name
633 * hash, should match its name
634 */
635 if (key->type == BTRFS_DIR_ITEM_KEY ||
636 key->type == BTRFS_XATTR_ITEM_KEY) {
637 char namebuf[MAX(BTRFS_NAME_LEN, XATTR_NAME_MAX)];
638
639 read_extent_buffer(leaf, namebuf,
640 (unsigned long)(di + 1), name_len);
641 name_hash = btrfs_name_hash(namebuf, name_len);
642 if (unlikely(key->offset != name_hash)) {
643 dir_item_err(leaf, slot,
644 "name hash mismatch with key, have 0x%016x expect 0x%016llx",
645 name_hash, key->offset);
646 return -EUCLEAN;
647 }
648 }
649 cur += total_size;
650 di = (struct btrfs_dir_item *)((void *)di + total_size);
651 }
652 return 0;
653 }
654
655 __printf(3, 4)
656 __cold
657 static void block_group_err(const struct extent_buffer *eb, int slot,
658 const char *fmt, ...)
659 {
660 const struct btrfs_fs_info *fs_info = eb->fs_info;
661 struct btrfs_key key;
662 struct va_format vaf;
663 va_list args;
664
665 btrfs_item_key_to_cpu(eb, &key, slot);
666 va_start(args, fmt);
667
668 vaf.fmt = fmt;
669 vaf.va = &args;
670
671 dump_page(folio_page(eb->folios[0], 0), "eb page dump");
672 btrfs_crit(fs_info,
673 "corrupt %s: root=%llu block=%llu slot=%d bg_start=%llu bg_len=%llu, %pV",
674 btrfs_header_level(eb) == 0 ? "leaf" : "node",
675 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
676 key.objectid, key.offset, &vaf);
677 va_end(args);
678 }
679
680 static int check_block_group_item(struct extent_buffer *leaf,
681 struct btrfs_key *key, int slot)
682 {
683 struct btrfs_fs_info *fs_info = leaf->fs_info;
684 struct btrfs_block_group_item bgi;
685 u32 item_size = btrfs_item_size(leaf, slot);
686 u64 chunk_objectid;
687 u64 flags;
688 u64 type;
689
690 /*
691 * Here we don't really care about alignment since extent allocator can
692 * handle it. We care more about the size.
693 */
694 if (unlikely(key->offset == 0)) {
695 block_group_err(leaf, slot,
696 "invalid block group size 0");
697 return -EUCLEAN;
698 }
699
700 if (unlikely(item_size != sizeof(bgi))) {
701 block_group_err(leaf, slot,
702 "invalid item size, have %u expect %zu",
703 item_size, sizeof(bgi));
704 return -EUCLEAN;
705 }
706
707 read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot),
708 sizeof(bgi));
709 chunk_objectid = btrfs_stack_block_group_chunk_objectid(&bgi);
710 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
711 /*
712 * We don't init the nr_global_roots until we load the global
713 * roots, so this could be 0 at mount time. If it's 0 we'll
714 * just assume we're fine, and later we'll check against our
715 * actual value.
716 */
717 if (unlikely(fs_info->nr_global_roots &&
718 chunk_objectid >= fs_info->nr_global_roots)) {
719 block_group_err(leaf, slot,
720 "invalid block group global root id, have %llu, needs to be <= %llu",
721 chunk_objectid,
722 fs_info->nr_global_roots);
723 return -EUCLEAN;
724 }
725 } else if (unlikely(chunk_objectid != BTRFS_FIRST_CHUNK_TREE_OBJECTID)) {
726 block_group_err(leaf, slot,
727 "invalid block group chunk objectid, have %llu expect %llu",
728 btrfs_stack_block_group_chunk_objectid(&bgi),
729 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
730 return -EUCLEAN;
731 }
732
733 if (unlikely(btrfs_stack_block_group_used(&bgi) > key->offset)) {
734 block_group_err(leaf, slot,
735 "invalid block group used, have %llu expect [0, %llu)",
736 btrfs_stack_block_group_used(&bgi), key->offset);
737 return -EUCLEAN;
738 }
739
740 flags = btrfs_stack_block_group_flags(&bgi);
741 if (unlikely(hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) > 1)) {
742 block_group_err(leaf, slot,
743 "invalid profile flags, have 0x%llx (%lu bits set) expect no more than 1 bit set",
744 flags & BTRFS_BLOCK_GROUP_PROFILE_MASK,
745 hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK));
746 return -EUCLEAN;
747 }
748
749 type = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
750 if (unlikely(type != BTRFS_BLOCK_GROUP_DATA &&
751 type != BTRFS_BLOCK_GROUP_METADATA &&
752 type != BTRFS_BLOCK_GROUP_SYSTEM &&
753 type != (BTRFS_BLOCK_GROUP_METADATA |
754 BTRFS_BLOCK_GROUP_DATA))) {
755 block_group_err(leaf, slot,
756 "invalid type, have 0x%llx (%lu bits set) expect either 0x%llx, 0x%llx, 0x%llx or 0x%llx",
757 type, hweight64(type),
758 BTRFS_BLOCK_GROUP_DATA, BTRFS_BLOCK_GROUP_METADATA,
759 BTRFS_BLOCK_GROUP_SYSTEM,
760 BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA);
761 return -EUCLEAN;
762 }
763 return 0;
764 }
765
766 __printf(4, 5)
767 __cold
768 static void chunk_err(const struct extent_buffer *leaf,
769 const struct btrfs_chunk *chunk, u64 logical,
770 const char *fmt, ...)
771 {
772 const struct btrfs_fs_info *fs_info = leaf->fs_info;
773 bool is_sb;
774 struct va_format vaf;
775 va_list args;
776 int i;
777 int slot = -1;
778
779 /* Only superblock eb is able to have such small offset */
780 is_sb = (leaf->start == BTRFS_SUPER_INFO_OFFSET);
781
782 if (!is_sb) {
783 /*
784 * Get the slot number by iterating through all slots, this
785 * would provide better readability.
786 */
787 for (i = 0; i < btrfs_header_nritems(leaf); i++) {
788 if (btrfs_item_ptr_offset(leaf, i) ==
789 (unsigned long)chunk) {
790 slot = i;
791 break;
792 }
793 }
794 }
795 va_start(args, fmt);
796 vaf.fmt = fmt;
797 vaf.va = &args;
798
799 if (is_sb)
800 btrfs_crit(fs_info,
801 "corrupt superblock syschunk array: chunk_start=%llu, %pV",
802 logical, &vaf);
803 else
804 btrfs_crit(fs_info,
805 "corrupt leaf: root=%llu block=%llu slot=%d chunk_start=%llu, %pV",
806 BTRFS_CHUNK_TREE_OBJECTID, leaf->start, slot,
807 logical, &vaf);
808 va_end(args);
809 }
810
811 /*
812 * The common chunk check which could also work on super block sys chunk array.
813 *
814 * Return -EUCLEAN if anything is corrupted.
815 * Return 0 if everything is OK.
816 */
817 int btrfs_check_chunk_valid(struct extent_buffer *leaf,
818 struct btrfs_chunk *chunk, u64 logical)
819 {
820 struct btrfs_fs_info *fs_info = leaf->fs_info;
821 u64 length;
822 u64 chunk_end;
823 u64 stripe_len;
824 u16 num_stripes;
825 u16 sub_stripes;
826 u64 type;
827 u64 features;
828 bool mixed = false;
829 int raid_index;
830 int nparity;
831 int ncopies;
832
833 length = btrfs_chunk_length(leaf, chunk);
834 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
835 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
836 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
837 type = btrfs_chunk_type(leaf, chunk);
838 raid_index = btrfs_bg_flags_to_raid_index(type);
839 ncopies = btrfs_raid_array[raid_index].ncopies;
840 nparity = btrfs_raid_array[raid_index].nparity;
841
842 if (unlikely(!num_stripes)) {
843 chunk_err(leaf, chunk, logical,
844 "invalid chunk num_stripes, have %u", num_stripes);
845 return -EUCLEAN;
846 }
847 if (unlikely(num_stripes < ncopies)) {
848 chunk_err(leaf, chunk, logical,
849 "invalid chunk num_stripes < ncopies, have %u < %d",
850 num_stripes, ncopies);
851 return -EUCLEAN;
852 }
853 if (unlikely(nparity && num_stripes == nparity)) {
854 chunk_err(leaf, chunk, logical,
855 "invalid chunk num_stripes == nparity, have %u == %d",
856 num_stripes, nparity);
857 return -EUCLEAN;
858 }
859 if (unlikely(!IS_ALIGNED(logical, fs_info->sectorsize))) {
860 chunk_err(leaf, chunk, logical,
861 "invalid chunk logical, have %llu should aligned to %u",
862 logical, fs_info->sectorsize);
863 return -EUCLEAN;
864 }
865 if (unlikely(btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize)) {
866 chunk_err(leaf, chunk, logical,
867 "invalid chunk sectorsize, have %u expect %u",
868 btrfs_chunk_sector_size(leaf, chunk),
869 fs_info->sectorsize);
870 return -EUCLEAN;
871 }
872 if (unlikely(!length || !IS_ALIGNED(length, fs_info->sectorsize))) {
873 chunk_err(leaf, chunk, logical,
874 "invalid chunk length, have %llu", length);
875 return -EUCLEAN;
876 }
877 if (unlikely(check_add_overflow(logical, length, &chunk_end))) {
878 chunk_err(leaf, chunk, logical,
879 "invalid chunk logical start and length, have logical start %llu length %llu",
880 logical, length);
881 return -EUCLEAN;
882 }
883 if (unlikely(!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN)) {
884 chunk_err(leaf, chunk, logical,
885 "invalid chunk stripe length: %llu",
886 stripe_len);
887 return -EUCLEAN;
888 }
889 /*
890 * We artificially limit the chunk size, so that the number of stripes
891 * inside a chunk can be fit into a U32. The current limit (256G) is
892 * way too large for real world usage anyway, and it's also much larger
893 * than our existing limit (10G).
894 *
895 * Thus it should be a good way to catch obvious bitflips.
896 */
897 if (unlikely(length >= btrfs_stripe_nr_to_offset(U32_MAX))) {
898 chunk_err(leaf, chunk, logical,
899 "chunk length too large: have %llu limit %llu",
900 length, btrfs_stripe_nr_to_offset(U32_MAX));
901 return -EUCLEAN;
902 }
903 if (unlikely(type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
904 BTRFS_BLOCK_GROUP_PROFILE_MASK))) {
905 chunk_err(leaf, chunk, logical,
906 "unrecognized chunk type: 0x%llx",
907 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
908 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
909 btrfs_chunk_type(leaf, chunk));
910 return -EUCLEAN;
911 }
912
913 if (unlikely(!has_single_bit_set(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
914 (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) != 0)) {
915 chunk_err(leaf, chunk, logical,
916 "invalid chunk profile flag: 0x%llx, expect 0 or 1 bit set",
917 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
918 return -EUCLEAN;
919 }
920 if (unlikely((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0)) {
921 chunk_err(leaf, chunk, logical,
922 "missing chunk type flag, have 0x%llx one bit must be set in 0x%llx",
923 type, BTRFS_BLOCK_GROUP_TYPE_MASK);
924 return -EUCLEAN;
925 }
926
927 if (unlikely((type & BTRFS_BLOCK_GROUP_SYSTEM) &&
928 (type & (BTRFS_BLOCK_GROUP_METADATA |
929 BTRFS_BLOCK_GROUP_DATA)))) {
930 chunk_err(leaf, chunk, logical,
931 "system chunk with data or metadata type: 0x%llx",
932 type);
933 return -EUCLEAN;
934 }
935
936 features = btrfs_super_incompat_flags(fs_info->super_copy);
937 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
938 mixed = true;
939
940 if (!mixed) {
941 if (unlikely((type & BTRFS_BLOCK_GROUP_METADATA) &&
942 (type & BTRFS_BLOCK_GROUP_DATA))) {
943 chunk_err(leaf, chunk, logical,
944 "mixed chunk type in non-mixed mode: 0x%llx", type);
945 return -EUCLEAN;
946 }
947 }
948
949 if (unlikely((type & BTRFS_BLOCK_GROUP_RAID10 &&
950 sub_stripes != btrfs_raid_array[BTRFS_RAID_RAID10].sub_stripes) ||
951 (type & BTRFS_BLOCK_GROUP_RAID1 &&
952 num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1].devs_min) ||
953 (type & BTRFS_BLOCK_GROUP_RAID1C3 &&
954 num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1C3].devs_min) ||
955 (type & BTRFS_BLOCK_GROUP_RAID1C4 &&
956 num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1C4].devs_min) ||
957 (type & BTRFS_BLOCK_GROUP_RAID5 &&
958 num_stripes < btrfs_raid_array[BTRFS_RAID_RAID5].devs_min) ||
959 (type & BTRFS_BLOCK_GROUP_RAID6 &&
960 num_stripes < btrfs_raid_array[BTRFS_RAID_RAID6].devs_min) ||
961 (type & BTRFS_BLOCK_GROUP_DUP &&
962 num_stripes != btrfs_raid_array[BTRFS_RAID_DUP].dev_stripes) ||
963 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
964 num_stripes != btrfs_raid_array[BTRFS_RAID_SINGLE].dev_stripes))) {
965 chunk_err(leaf, chunk, logical,
966 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
967 num_stripes, sub_stripes,
968 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
969 return -EUCLEAN;
970 }
971
972 return 0;
973 }
974
975 /*
976 * Enhanced version of chunk item checker.
977 *
978 * The common btrfs_check_chunk_valid() doesn't check item size since it needs
979 * to work on super block sys_chunk_array which doesn't have full item ptr.
980 */
981 static int check_leaf_chunk_item(struct extent_buffer *leaf,
982 struct btrfs_chunk *chunk,
983 struct btrfs_key *key, int slot)
984 {
985 int num_stripes;
986
987 if (unlikely(btrfs_item_size(leaf, slot) < sizeof(struct btrfs_chunk))) {
988 chunk_err(leaf, chunk, key->offset,
989 "invalid chunk item size: have %u expect [%zu, %u)",
990 btrfs_item_size(leaf, slot),
991 sizeof(struct btrfs_chunk),
992 BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
993 return -EUCLEAN;
994 }
995
996 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
997 /* Let btrfs_check_chunk_valid() handle this error type */
998 if (num_stripes == 0)
999 goto out;
1000
1001 if (unlikely(btrfs_chunk_item_size(num_stripes) !=
1002 btrfs_item_size(leaf, slot))) {
1003 chunk_err(leaf, chunk, key->offset,
1004 "invalid chunk item size: have %u expect %lu",
1005 btrfs_item_size(leaf, slot),
1006 btrfs_chunk_item_size(num_stripes));
1007 return -EUCLEAN;
1008 }
1009 out:
1010 return btrfs_check_chunk_valid(leaf, chunk, key->offset);
1011 }
1012
1013 __printf(3, 4)
1014 __cold
1015 static void dev_item_err(const struct extent_buffer *eb, int slot,
1016 const char *fmt, ...)
1017 {
1018 struct btrfs_key key;
1019 struct va_format vaf;
1020 va_list args;
1021
1022 btrfs_item_key_to_cpu(eb, &key, slot);
1023 va_start(args, fmt);
1024
1025 vaf.fmt = fmt;
1026 vaf.va = &args;
1027
1028 dump_page(folio_page(eb->folios[0], 0), "eb page dump");
1029 btrfs_crit(eb->fs_info,
1030 "corrupt %s: root=%llu block=%llu slot=%d devid=%llu %pV",
1031 btrfs_header_level(eb) == 0 ? "leaf" : "node",
1032 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
1033 key.objectid, &vaf);
1034 va_end(args);
1035 }
1036
1037 static int check_dev_item(struct extent_buffer *leaf,
1038 struct btrfs_key *key, int slot)
1039 {
1040 struct btrfs_dev_item *ditem;
1041 const u32 item_size = btrfs_item_size(leaf, slot);
1042
1043 if (unlikely(key->objectid != BTRFS_DEV_ITEMS_OBJECTID)) {
1044 dev_item_err(leaf, slot,
1045 "invalid objectid: has=%llu expect=%llu",
1046 key->objectid, BTRFS_DEV_ITEMS_OBJECTID);
1047 return -EUCLEAN;
1048 }
1049
1050 if (unlikely(item_size != sizeof(*ditem))) {
1051 dev_item_err(leaf, slot, "invalid item size: has %u expect %zu",
1052 item_size, sizeof(*ditem));
1053 return -EUCLEAN;
1054 }
1055
1056 ditem = btrfs_item_ptr(leaf, slot, struct btrfs_dev_item);
1057 if (unlikely(btrfs_device_id(leaf, ditem) != key->offset)) {
1058 dev_item_err(leaf, slot,
1059 "devid mismatch: key has=%llu item has=%llu",
1060 key->offset, btrfs_device_id(leaf, ditem));
1061 return -EUCLEAN;
1062 }
1063
1064 /*
1065 * For device total_bytes, we don't have reliable way to check it, as
1066 * it can be 0 for device removal. Device size check can only be done
1067 * by dev extents check.
1068 */
1069 if (unlikely(btrfs_device_bytes_used(leaf, ditem) >
1070 btrfs_device_total_bytes(leaf, ditem))) {
1071 dev_item_err(leaf, slot,
1072 "invalid bytes used: have %llu expect [0, %llu]",
1073 btrfs_device_bytes_used(leaf, ditem),
1074 btrfs_device_total_bytes(leaf, ditem));
1075 return -EUCLEAN;
1076 }
1077 /*
1078 * Remaining members like io_align/type/gen/dev_group aren't really
1079 * utilized. Skip them to make later usage of them easier.
1080 */
1081 return 0;
1082 }
1083
1084 static int check_inode_item(struct extent_buffer *leaf,
1085 struct btrfs_key *key, int slot)
1086 {
1087 struct btrfs_fs_info *fs_info = leaf->fs_info;
1088 struct btrfs_inode_item *iitem;
1089 u64 super_gen = btrfs_super_generation(fs_info->super_copy);
1090 u32 valid_mask = (S_IFMT | S_ISUID | S_ISGID | S_ISVTX | 0777);
1091 const u32 item_size = btrfs_item_size(leaf, slot);
1092 u32 mode;
1093 int ret;
1094 u32 flags;
1095 u32 ro_flags;
1096
1097 ret = check_inode_key(leaf, key, slot);
1098 if (unlikely(ret < 0))
1099 return ret;
1100
1101 if (unlikely(item_size != sizeof(*iitem))) {
1102 generic_err(leaf, slot, "invalid item size: has %u expect %zu",
1103 item_size, sizeof(*iitem));
1104 return -EUCLEAN;
1105 }
1106
1107 iitem = btrfs_item_ptr(leaf, slot, struct btrfs_inode_item);
1108
1109 /* Here we use super block generation + 1 to handle log tree */
1110 if (unlikely(btrfs_inode_generation(leaf, iitem) > super_gen + 1)) {
1111 inode_item_err(leaf, slot,
1112 "invalid inode generation: has %llu expect (0, %llu]",
1113 btrfs_inode_generation(leaf, iitem),
1114 super_gen + 1);
1115 return -EUCLEAN;
1116 }
1117 /* Note for ROOT_TREE_DIR_ITEM, mkfs could set its transid 0 */
1118 if (unlikely(btrfs_inode_transid(leaf, iitem) > super_gen + 1)) {
1119 inode_item_err(leaf, slot,
1120 "invalid inode transid: has %llu expect [0, %llu]",
1121 btrfs_inode_transid(leaf, iitem), super_gen + 1);
1122 return -EUCLEAN;
1123 }
1124
1125 /*
1126 * For size and nbytes it's better not to be too strict, as for dir
1127 * item its size/nbytes can easily get wrong, but doesn't affect
1128 * anything in the fs. So here we skip the check.
1129 */
1130 mode = btrfs_inode_mode(leaf, iitem);
1131 if (unlikely(mode & ~valid_mask)) {
1132 inode_item_err(leaf, slot,
1133 "unknown mode bit detected: 0x%x",
1134 mode & ~valid_mask);
1135 return -EUCLEAN;
1136 }
1137
1138 /*
1139 * S_IFMT is not bit mapped so we can't completely rely on
1140 * is_power_of_2/has_single_bit_set, but it can save us from checking
1141 * FIFO/CHR/DIR/REG. Only needs to check BLK, LNK and SOCKS
1142 */
1143 if (!has_single_bit_set(mode & S_IFMT)) {
1144 if (unlikely(!S_ISLNK(mode) && !S_ISBLK(mode) && !S_ISSOCK(mode))) {
1145 inode_item_err(leaf, slot,
1146 "invalid mode: has 0%o expect valid S_IF* bit(s)",
1147 mode & S_IFMT);
1148 return -EUCLEAN;
1149 }
1150 }
1151 if (unlikely(S_ISDIR(mode) && btrfs_inode_nlink(leaf, iitem) > 1)) {
1152 inode_item_err(leaf, slot,
1153 "invalid nlink: has %u expect no more than 1 for dir",
1154 btrfs_inode_nlink(leaf, iitem));
1155 return -EUCLEAN;
1156 }
1157 btrfs_inode_split_flags(btrfs_inode_flags(leaf, iitem), &flags, &ro_flags);
1158 if (unlikely(flags & ~BTRFS_INODE_FLAG_MASK)) {
1159 inode_item_err(leaf, slot,
1160 "unknown incompat flags detected: 0x%x", flags);
1161 return -EUCLEAN;
1162 }
1163 if (unlikely(!sb_rdonly(fs_info->sb) &&
1164 (ro_flags & ~BTRFS_INODE_RO_FLAG_MASK))) {
1165 inode_item_err(leaf, slot,
1166 "unknown ro-compat flags detected on writeable mount: 0x%x",
1167 ro_flags);
1168 return -EUCLEAN;
1169 }
1170 return 0;
1171 }
1172
1173 static int check_root_item(struct extent_buffer *leaf, struct btrfs_key *key,
1174 int slot)
1175 {
1176 struct btrfs_fs_info *fs_info = leaf->fs_info;
1177 struct btrfs_root_item ri = { 0 };
1178 const u64 valid_root_flags = BTRFS_ROOT_SUBVOL_RDONLY |
1179 BTRFS_ROOT_SUBVOL_DEAD;
1180 int ret;
1181
1182 ret = check_root_key(leaf, key, slot);
1183 if (unlikely(ret < 0))
1184 return ret;
1185
1186 if (unlikely(btrfs_item_size(leaf, slot) != sizeof(ri) &&
1187 btrfs_item_size(leaf, slot) !=
1188 btrfs_legacy_root_item_size())) {
1189 generic_err(leaf, slot,
1190 "invalid root item size, have %u expect %zu or %u",
1191 btrfs_item_size(leaf, slot), sizeof(ri),
1192 btrfs_legacy_root_item_size());
1193 return -EUCLEAN;
1194 }
1195
1196 /*
1197 * For legacy root item, the members starting at generation_v2 will be
1198 * all filled with 0.
1199 * And since we allow geneartion_v2 as 0, it will still pass the check.
1200 */
1201 read_extent_buffer(leaf, &ri, btrfs_item_ptr_offset(leaf, slot),
1202 btrfs_item_size(leaf, slot));
1203
1204 /* Generation related */
1205 if (unlikely(btrfs_root_generation(&ri) >
1206 btrfs_super_generation(fs_info->super_copy) + 1)) {
1207 generic_err(leaf, slot,
1208 "invalid root generation, have %llu expect (0, %llu]",
1209 btrfs_root_generation(&ri),
1210 btrfs_super_generation(fs_info->super_copy) + 1);
1211 return -EUCLEAN;
1212 }
1213 if (unlikely(btrfs_root_generation_v2(&ri) >
1214 btrfs_super_generation(fs_info->super_copy) + 1)) {
1215 generic_err(leaf, slot,
1216 "invalid root v2 generation, have %llu expect (0, %llu]",
1217 btrfs_root_generation_v2(&ri),
1218 btrfs_super_generation(fs_info->super_copy) + 1);
1219 return -EUCLEAN;
1220 }
1221 if (unlikely(btrfs_root_last_snapshot(&ri) >
1222 btrfs_super_generation(fs_info->super_copy) + 1)) {
1223 generic_err(leaf, slot,
1224 "invalid root last_snapshot, have %llu expect (0, %llu]",
1225 btrfs_root_last_snapshot(&ri),
1226 btrfs_super_generation(fs_info->super_copy) + 1);
1227 return -EUCLEAN;
1228 }
1229
1230 /* Alignment and level check */
1231 if (unlikely(!IS_ALIGNED(btrfs_root_bytenr(&ri), fs_info->sectorsize))) {
1232 generic_err(leaf, slot,
1233 "invalid root bytenr, have %llu expect to be aligned to %u",
1234 btrfs_root_bytenr(&ri), fs_info->sectorsize);
1235 return -EUCLEAN;
1236 }
1237 if (unlikely(btrfs_root_level(&ri) >= BTRFS_MAX_LEVEL)) {
1238 generic_err(leaf, slot,
1239 "invalid root level, have %u expect [0, %u]",
1240 btrfs_root_level(&ri), BTRFS_MAX_LEVEL - 1);
1241 return -EUCLEAN;
1242 }
1243 if (unlikely(btrfs_root_drop_level(&ri) >= BTRFS_MAX_LEVEL)) {
1244 generic_err(leaf, slot,
1245 "invalid root level, have %u expect [0, %u]",
1246 btrfs_root_drop_level(&ri), BTRFS_MAX_LEVEL - 1);
1247 return -EUCLEAN;
1248 }
1249
1250 /* Flags check */
1251 if (unlikely(btrfs_root_flags(&ri) & ~valid_root_flags)) {
1252 generic_err(leaf, slot,
1253 "invalid root flags, have 0x%llx expect mask 0x%llx",
1254 btrfs_root_flags(&ri), valid_root_flags);
1255 return -EUCLEAN;
1256 }
1257 return 0;
1258 }
1259
1260 __printf(3,4)
1261 __cold
1262 static void extent_err(const struct extent_buffer *eb, int slot,
1263 const char *fmt, ...)
1264 {
1265 struct btrfs_key key;
1266 struct va_format vaf;
1267 va_list args;
1268 u64 bytenr;
1269 u64 len;
1270
1271 btrfs_item_key_to_cpu(eb, &key, slot);
1272 bytenr = key.objectid;
1273 if (key.type == BTRFS_METADATA_ITEM_KEY ||
1274 key.type == BTRFS_TREE_BLOCK_REF_KEY ||
1275 key.type == BTRFS_SHARED_BLOCK_REF_KEY)
1276 len = eb->fs_info->nodesize;
1277 else
1278 len = key.offset;
1279 va_start(args, fmt);
1280
1281 vaf.fmt = fmt;
1282 vaf.va = &args;
1283
1284 dump_page(folio_page(eb->folios[0], 0), "eb page dump");
1285 btrfs_crit(eb->fs_info,
1286 "corrupt %s: block=%llu slot=%d extent bytenr=%llu len=%llu %pV",
1287 btrfs_header_level(eb) == 0 ? "leaf" : "node",
1288 eb->start, slot, bytenr, len, &vaf);
1289 va_end(args);
1290 }
1291
1292 static bool is_valid_dref_root(u64 rootid)
1293 {
1294 /*
1295 * The following tree root objectids are allowed to have a data backref:
1296 * - subvolume trees
1297 * - data reloc tree
1298 * - tree root
1299 * For v1 space cache
1300 */
1301 return is_fstree(rootid) || rootid == BTRFS_DATA_RELOC_TREE_OBJECTID ||
1302 rootid == BTRFS_ROOT_TREE_OBJECTID;
1303 }
1304
1305 static int check_extent_item(struct extent_buffer *leaf,
1306 struct btrfs_key *key, int slot,
1307 struct btrfs_key *prev_key)
1308 {
1309 struct btrfs_fs_info *fs_info = leaf->fs_info;
1310 struct btrfs_extent_item *ei;
1311 bool is_tree_block = false;
1312 unsigned long ptr; /* Current pointer inside inline refs */
1313 unsigned long end; /* Extent item end */
1314 const u32 item_size = btrfs_item_size(leaf, slot);
1315 u8 last_type = 0;
1316 u64 last_seq = U64_MAX;
1317 u64 flags;
1318 u64 generation;
1319 u64 total_refs; /* Total refs in btrfs_extent_item */
1320 u64 inline_refs = 0; /* found total inline refs */
1321
1322 if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1323 !btrfs_fs_incompat(fs_info, SKINNY_METADATA))) {
1324 generic_err(leaf, slot,
1325 "invalid key type, METADATA_ITEM type invalid when SKINNY_METADATA feature disabled");
1326 return -EUCLEAN;
1327 }
1328 /* key->objectid is the bytenr for both key types */
1329 if (unlikely(!IS_ALIGNED(key->objectid, fs_info->sectorsize))) {
1330 generic_err(leaf, slot,
1331 "invalid key objectid, have %llu expect to be aligned to %u",
1332 key->objectid, fs_info->sectorsize);
1333 return -EUCLEAN;
1334 }
1335
1336 /* key->offset is tree level for METADATA_ITEM_KEY */
1337 if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1338 key->offset >= BTRFS_MAX_LEVEL)) {
1339 extent_err(leaf, slot,
1340 "invalid tree level, have %llu expect [0, %u]",
1341 key->offset, BTRFS_MAX_LEVEL - 1);
1342 return -EUCLEAN;
1343 }
1344
1345 /*
1346 * EXTENT/METADATA_ITEM consists of:
1347 * 1) One btrfs_extent_item
1348 * Records the total refs, type and generation of the extent.
1349 *
1350 * 2) One btrfs_tree_block_info (for EXTENT_ITEM and tree backref only)
1351 * Records the first key and level of the tree block.
1352 *
1353 * 2) Zero or more btrfs_extent_inline_ref(s)
1354 * Each inline ref has one btrfs_extent_inline_ref shows:
1355 * 2.1) The ref type, one of the 4
1356 * TREE_BLOCK_REF Tree block only
1357 * SHARED_BLOCK_REF Tree block only
1358 * EXTENT_DATA_REF Data only
1359 * SHARED_DATA_REF Data only
1360 * 2.2) Ref type specific data
1361 * Either using btrfs_extent_inline_ref::offset, or specific
1362 * data structure.
1363 *
1364 * All above inline items should follow the order:
1365 *
1366 * - All btrfs_extent_inline_ref::type should be in an ascending
1367 * order
1368 *
1369 * - Within the same type, the items should follow a descending
1370 * order by their sequence number. The sequence number is
1371 * determined by:
1372 * * btrfs_extent_inline_ref::offset for all types other than
1373 * EXTENT_DATA_REF
1374 * * hash_extent_data_ref() for EXTENT_DATA_REF
1375 */
1376 if (unlikely(item_size < sizeof(*ei))) {
1377 extent_err(leaf, slot,
1378 "invalid item size, have %u expect [%zu, %u)",
1379 item_size, sizeof(*ei),
1380 BTRFS_LEAF_DATA_SIZE(fs_info));
1381 return -EUCLEAN;
1382 }
1383 end = item_size + btrfs_item_ptr_offset(leaf, slot);
1384
1385 /* Checks against extent_item */
1386 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
1387 flags = btrfs_extent_flags(leaf, ei);
1388 total_refs = btrfs_extent_refs(leaf, ei);
1389 generation = btrfs_extent_generation(leaf, ei);
1390 if (unlikely(generation >
1391 btrfs_super_generation(fs_info->super_copy) + 1)) {
1392 extent_err(leaf, slot,
1393 "invalid generation, have %llu expect (0, %llu]",
1394 generation,
1395 btrfs_super_generation(fs_info->super_copy) + 1);
1396 return -EUCLEAN;
1397 }
1398 if (unlikely(!has_single_bit_set(flags & (BTRFS_EXTENT_FLAG_DATA |
1399 BTRFS_EXTENT_FLAG_TREE_BLOCK)))) {
1400 extent_err(leaf, slot,
1401 "invalid extent flag, have 0x%llx expect 1 bit set in 0x%llx",
1402 flags, BTRFS_EXTENT_FLAG_DATA |
1403 BTRFS_EXTENT_FLAG_TREE_BLOCK);
1404 return -EUCLEAN;
1405 }
1406 is_tree_block = !!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK);
1407 if (is_tree_block) {
1408 if (unlikely(key->type == BTRFS_EXTENT_ITEM_KEY &&
1409 key->offset != fs_info->nodesize)) {
1410 extent_err(leaf, slot,
1411 "invalid extent length, have %llu expect %u",
1412 key->offset, fs_info->nodesize);
1413 return -EUCLEAN;
1414 }
1415 } else {
1416 if (unlikely(key->type != BTRFS_EXTENT_ITEM_KEY)) {
1417 extent_err(leaf, slot,
1418 "invalid key type, have %u expect %u for data backref",
1419 key->type, BTRFS_EXTENT_ITEM_KEY);
1420 return -EUCLEAN;
1421 }
1422 if (unlikely(!IS_ALIGNED(key->offset, fs_info->sectorsize))) {
1423 extent_err(leaf, slot,
1424 "invalid extent length, have %llu expect aligned to %u",
1425 key->offset, fs_info->sectorsize);
1426 return -EUCLEAN;
1427 }
1428 if (unlikely(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
1429 extent_err(leaf, slot,
1430 "invalid extent flag, data has full backref set");
1431 return -EUCLEAN;
1432 }
1433 }
1434 ptr = (unsigned long)(struct btrfs_extent_item *)(ei + 1);
1435
1436 /* Check the special case of btrfs_tree_block_info */
1437 if (is_tree_block && key->type != BTRFS_METADATA_ITEM_KEY) {
1438 struct btrfs_tree_block_info *info;
1439
1440 info = (struct btrfs_tree_block_info *)ptr;
1441 if (unlikely(btrfs_tree_block_level(leaf, info) >= BTRFS_MAX_LEVEL)) {
1442 extent_err(leaf, slot,
1443 "invalid tree block info level, have %u expect [0, %u]",
1444 btrfs_tree_block_level(leaf, info),
1445 BTRFS_MAX_LEVEL - 1);
1446 return -EUCLEAN;
1447 }
1448 ptr = (unsigned long)(struct btrfs_tree_block_info *)(info + 1);
1449 }
1450
1451 /* Check inline refs */
1452 while (ptr < end) {
1453 struct btrfs_extent_inline_ref *iref;
1454 struct btrfs_extent_data_ref *dref;
1455 struct btrfs_shared_data_ref *sref;
1456 u64 seq;
1457 u64 dref_root;
1458 u64 dref_objectid;
1459 u64 dref_offset;
1460 u64 inline_offset;
1461 u8 inline_type;
1462
1463 if (unlikely(ptr + sizeof(*iref) > end)) {
1464 extent_err(leaf, slot,
1465 "inline ref item overflows extent item, ptr %lu iref size %zu end %lu",
1466 ptr, sizeof(*iref), end);
1467 return -EUCLEAN;
1468 }
1469 iref = (struct btrfs_extent_inline_ref *)ptr;
1470 inline_type = btrfs_extent_inline_ref_type(leaf, iref);
1471 inline_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1472 seq = inline_offset;
1473 if (unlikely(ptr + btrfs_extent_inline_ref_size(inline_type) > end)) {
1474 extent_err(leaf, slot,
1475 "inline ref item overflows extent item, ptr %lu iref size %u end %lu",
1476 ptr, btrfs_extent_inline_ref_size(inline_type), end);
1477 return -EUCLEAN;
1478 }
1479
1480 switch (inline_type) {
1481 /* inline_offset is subvolid of the owner, no need to check */
1482 case BTRFS_TREE_BLOCK_REF_KEY:
1483 inline_refs++;
1484 break;
1485 /* Contains parent bytenr */
1486 case BTRFS_SHARED_BLOCK_REF_KEY:
1487 if (unlikely(!IS_ALIGNED(inline_offset,
1488 fs_info->sectorsize))) {
1489 extent_err(leaf, slot,
1490 "invalid tree parent bytenr, have %llu expect aligned to %u",
1491 inline_offset, fs_info->sectorsize);
1492 return -EUCLEAN;
1493 }
1494 inline_refs++;
1495 break;
1496 /*
1497 * Contains owner subvolid, owner key objectid, adjusted offset.
1498 * The only obvious corruption can happen in that offset.
1499 */
1500 case BTRFS_EXTENT_DATA_REF_KEY:
1501 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1502 dref_root = btrfs_extent_data_ref_root(leaf, dref);
1503 dref_objectid = btrfs_extent_data_ref_objectid(leaf, dref);
1504 dref_offset = btrfs_extent_data_ref_offset(leaf, dref);
1505 seq = hash_extent_data_ref(
1506 btrfs_extent_data_ref_root(leaf, dref),
1507 btrfs_extent_data_ref_objectid(leaf, dref),
1508 btrfs_extent_data_ref_offset(leaf, dref));
1509 if (unlikely(!is_valid_dref_root(dref_root))) {
1510 extent_err(leaf, slot,
1511 "invalid data ref root value %llu",
1512 dref_root);
1513 return -EUCLEAN;
1514 }
1515 if (unlikely(dref_objectid < BTRFS_FIRST_FREE_OBJECTID ||
1516 dref_objectid > BTRFS_LAST_FREE_OBJECTID)) {
1517 extent_err(leaf, slot,
1518 "invalid data ref objectid value %llu",
1519 dref_root);
1520 return -EUCLEAN;
1521 }
1522 if (unlikely(!IS_ALIGNED(dref_offset,
1523 fs_info->sectorsize))) {
1524 extent_err(leaf, slot,
1525 "invalid data ref offset, have %llu expect aligned to %u",
1526 dref_offset, fs_info->sectorsize);
1527 return -EUCLEAN;
1528 }
1529 inline_refs += btrfs_extent_data_ref_count(leaf, dref);
1530 break;
1531 /* Contains parent bytenr and ref count */
1532 case BTRFS_SHARED_DATA_REF_KEY:
1533 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1534 if (unlikely(!IS_ALIGNED(inline_offset,
1535 fs_info->sectorsize))) {
1536 extent_err(leaf, slot,
1537 "invalid data parent bytenr, have %llu expect aligned to %u",
1538 inline_offset, fs_info->sectorsize);
1539 return -EUCLEAN;
1540 }
1541 inline_refs += btrfs_shared_data_ref_count(leaf, sref);
1542 break;
1543 case BTRFS_EXTENT_OWNER_REF_KEY:
1544 WARN_ON(!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));
1545 break;
1546 default:
1547 extent_err(leaf, slot, "unknown inline ref type: %u",
1548 inline_type);
1549 return -EUCLEAN;
1550 }
1551 if (inline_type < last_type) {
1552 extent_err(leaf, slot,
1553 "inline ref out-of-order: has type %u, prev type %u",
1554 inline_type, last_type);
1555 return -EUCLEAN;
1556 }
1557 /* Type changed, allow the sequence starts from U64_MAX again. */
1558 if (inline_type > last_type)
1559 last_seq = U64_MAX;
1560 if (seq > last_seq) {
1561 extent_err(leaf, slot,
1562 "inline ref out-of-order: has type %u offset %llu seq 0x%llx, prev type %u seq 0x%llx",
1563 inline_type, inline_offset, seq,
1564 last_type, last_seq);
1565 return -EUCLEAN;
1566 }
1567 last_type = inline_type;
1568 last_seq = seq;
1569 ptr += btrfs_extent_inline_ref_size(inline_type);
1570 }
1571 /* No padding is allowed */
1572 if (unlikely(ptr != end)) {
1573 extent_err(leaf, slot,
1574 "invalid extent item size, padding bytes found");
1575 return -EUCLEAN;
1576 }
1577
1578 /* Finally, check the inline refs against total refs */
1579 if (unlikely(inline_refs > total_refs)) {
1580 extent_err(leaf, slot,
1581 "invalid extent refs, have %llu expect >= inline %llu",
1582 total_refs, inline_refs);
1583 return -EUCLEAN;
1584 }
1585
1586 if ((prev_key->type == BTRFS_EXTENT_ITEM_KEY) ||
1587 (prev_key->type == BTRFS_METADATA_ITEM_KEY)) {
1588 u64 prev_end = prev_key->objectid;
1589
1590 if (prev_key->type == BTRFS_METADATA_ITEM_KEY)
1591 prev_end += fs_info->nodesize;
1592 else
1593 prev_end += prev_key->offset;
1594
1595 if (unlikely(prev_end > key->objectid)) {
1596 extent_err(leaf, slot,
1597 "previous extent [%llu %u %llu] overlaps current extent [%llu %u %llu]",
1598 prev_key->objectid, prev_key->type,
1599 prev_key->offset, key->objectid, key->type,
1600 key->offset);
1601 return -EUCLEAN;
1602 }
1603 }
1604
1605 return 0;
1606 }
1607
1608 static int check_simple_keyed_refs(struct extent_buffer *leaf,
1609 struct btrfs_key *key, int slot)
1610 {
1611 u32 expect_item_size = 0;
1612
1613 if (key->type == BTRFS_SHARED_DATA_REF_KEY)
1614 expect_item_size = sizeof(struct btrfs_shared_data_ref);
1615
1616 if (unlikely(btrfs_item_size(leaf, slot) != expect_item_size)) {
1617 generic_err(leaf, slot,
1618 "invalid item size, have %u expect %u for key type %u",
1619 btrfs_item_size(leaf, slot),
1620 expect_item_size, key->type);
1621 return -EUCLEAN;
1622 }
1623 if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1624 generic_err(leaf, slot,
1625 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1626 key->objectid, leaf->fs_info->sectorsize);
1627 return -EUCLEAN;
1628 }
1629 if (unlikely(key->type != BTRFS_TREE_BLOCK_REF_KEY &&
1630 !IS_ALIGNED(key->offset, leaf->fs_info->sectorsize))) {
1631 extent_err(leaf, slot,
1632 "invalid tree parent bytenr, have %llu expect aligned to %u",
1633 key->offset, leaf->fs_info->sectorsize);
1634 return -EUCLEAN;
1635 }
1636 return 0;
1637 }
1638
1639 static int check_extent_data_ref(struct extent_buffer *leaf,
1640 struct btrfs_key *key, int slot)
1641 {
1642 struct btrfs_extent_data_ref *dref;
1643 unsigned long ptr = btrfs_item_ptr_offset(leaf, slot);
1644 const unsigned long end = ptr + btrfs_item_size(leaf, slot);
1645
1646 if (unlikely(btrfs_item_size(leaf, slot) % sizeof(*dref) != 0)) {
1647 generic_err(leaf, slot,
1648 "invalid item size, have %u expect aligned to %zu for key type %u",
1649 btrfs_item_size(leaf, slot),
1650 sizeof(*dref), key->type);
1651 return -EUCLEAN;
1652 }
1653 if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1654 generic_err(leaf, slot,
1655 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1656 key->objectid, leaf->fs_info->sectorsize);
1657 return -EUCLEAN;
1658 }
1659 for (; ptr < end; ptr += sizeof(*dref)) {
1660 u64 root;
1661 u64 objectid;
1662 u64 offset;
1663
1664 /*
1665 * We cannot check the extent_data_ref hash due to possible
1666 * overflow from the leaf due to hash collisions.
1667 */
1668 dref = (struct btrfs_extent_data_ref *)ptr;
1669 root = btrfs_extent_data_ref_root(leaf, dref);
1670 objectid = btrfs_extent_data_ref_objectid(leaf, dref);
1671 offset = btrfs_extent_data_ref_offset(leaf, dref);
1672 if (unlikely(!is_valid_dref_root(root))) {
1673 extent_err(leaf, slot,
1674 "invalid extent data backref root value %llu",
1675 root);
1676 return -EUCLEAN;
1677 }
1678 if (unlikely(objectid < BTRFS_FIRST_FREE_OBJECTID ||
1679 objectid > BTRFS_LAST_FREE_OBJECTID)) {
1680 extent_err(leaf, slot,
1681 "invalid extent data backref objectid value %llu",
1682 root);
1683 return -EUCLEAN;
1684 }
1685 if (unlikely(!IS_ALIGNED(offset, leaf->fs_info->sectorsize))) {
1686 extent_err(leaf, slot,
1687 "invalid extent data backref offset, have %llu expect aligned to %u",
1688 offset, leaf->fs_info->sectorsize);
1689 return -EUCLEAN;
1690 }
1691 }
1692 return 0;
1693 }
1694
1695 #define inode_ref_err(eb, slot, fmt, args...) \
1696 inode_item_err(eb, slot, fmt, ##args)
1697 static int check_inode_ref(struct extent_buffer *leaf,
1698 struct btrfs_key *key, struct btrfs_key *prev_key,
1699 int slot)
1700 {
1701 struct btrfs_inode_ref *iref;
1702 unsigned long ptr;
1703 unsigned long end;
1704
1705 if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
1706 return -EUCLEAN;
1707 /* namelen can't be 0, so item_size == sizeof() is also invalid */
1708 if (unlikely(btrfs_item_size(leaf, slot) <= sizeof(*iref))) {
1709 inode_ref_err(leaf, slot,
1710 "invalid item size, have %u expect (%zu, %u)",
1711 btrfs_item_size(leaf, slot),
1712 sizeof(*iref), BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
1713 return -EUCLEAN;
1714 }
1715
1716 ptr = btrfs_item_ptr_offset(leaf, slot);
1717 end = ptr + btrfs_item_size(leaf, slot);
1718 while (ptr < end) {
1719 u16 namelen;
1720
1721 if (unlikely(ptr + sizeof(iref) > end)) {
1722 inode_ref_err(leaf, slot,
1723 "inode ref overflow, ptr %lu end %lu inode_ref_size %zu",
1724 ptr, end, sizeof(iref));
1725 return -EUCLEAN;
1726 }
1727
1728 iref = (struct btrfs_inode_ref *)ptr;
1729 namelen = btrfs_inode_ref_name_len(leaf, iref);
1730 if (unlikely(ptr + sizeof(*iref) + namelen > end)) {
1731 inode_ref_err(leaf, slot,
1732 "inode ref overflow, ptr %lu end %lu namelen %u",
1733 ptr, end, namelen);
1734 return -EUCLEAN;
1735 }
1736
1737 /*
1738 * NOTE: In theory we should record all found index numbers
1739 * to find any duplicated indexes, but that will be too time
1740 * consuming for inodes with too many hard links.
1741 */
1742 ptr += sizeof(*iref) + namelen;
1743 }
1744 return 0;
1745 }
1746
1747 static int check_raid_stripe_extent(const struct extent_buffer *leaf,
1748 const struct btrfs_key *key, int slot)
1749 {
1750 if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1751 generic_err(leaf, slot,
1752 "invalid key objectid for raid stripe extent, have %llu expect aligned to %u",
1753 key->objectid, leaf->fs_info->sectorsize);
1754 return -EUCLEAN;
1755 }
1756
1757 if (unlikely(!btrfs_fs_incompat(leaf->fs_info, RAID_STRIPE_TREE))) {
1758 generic_err(leaf, slot,
1759 "RAID_STRIPE_EXTENT present but RAID_STRIPE_TREE incompat bit unset");
1760 return -EUCLEAN;
1761 }
1762
1763 return 0;
1764 }
1765
1766 /*
1767 * Common point to switch the item-specific validation.
1768 */
1769 static enum btrfs_tree_block_status check_leaf_item(struct extent_buffer *leaf,
1770 struct btrfs_key *key,
1771 int slot,
1772 struct btrfs_key *prev_key)
1773 {
1774 int ret = 0;
1775 struct btrfs_chunk *chunk;
1776
1777 switch (key->type) {
1778 case BTRFS_EXTENT_DATA_KEY:
1779 ret = check_extent_data_item(leaf, key, slot, prev_key);
1780 break;
1781 case BTRFS_EXTENT_CSUM_KEY:
1782 ret = check_csum_item(leaf, key, slot, prev_key);
1783 break;
1784 case BTRFS_DIR_ITEM_KEY:
1785 case BTRFS_DIR_INDEX_KEY:
1786 case BTRFS_XATTR_ITEM_KEY:
1787 ret = check_dir_item(leaf, key, prev_key, slot);
1788 break;
1789 case BTRFS_INODE_REF_KEY:
1790 ret = check_inode_ref(leaf, key, prev_key, slot);
1791 break;
1792 case BTRFS_BLOCK_GROUP_ITEM_KEY:
1793 ret = check_block_group_item(leaf, key, slot);
1794 break;
1795 case BTRFS_CHUNK_ITEM_KEY:
1796 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1797 ret = check_leaf_chunk_item(leaf, chunk, key, slot);
1798 break;
1799 case BTRFS_DEV_ITEM_KEY:
1800 ret = check_dev_item(leaf, key, slot);
1801 break;
1802 case BTRFS_INODE_ITEM_KEY:
1803 ret = check_inode_item(leaf, key, slot);
1804 break;
1805 case BTRFS_ROOT_ITEM_KEY:
1806 ret = check_root_item(leaf, key, slot);
1807 break;
1808 case BTRFS_EXTENT_ITEM_KEY:
1809 case BTRFS_METADATA_ITEM_KEY:
1810 ret = check_extent_item(leaf, key, slot, prev_key);
1811 break;
1812 case BTRFS_TREE_BLOCK_REF_KEY:
1813 case BTRFS_SHARED_DATA_REF_KEY:
1814 case BTRFS_SHARED_BLOCK_REF_KEY:
1815 ret = check_simple_keyed_refs(leaf, key, slot);
1816 break;
1817 case BTRFS_EXTENT_DATA_REF_KEY:
1818 ret = check_extent_data_ref(leaf, key, slot);
1819 break;
1820 case BTRFS_RAID_STRIPE_KEY:
1821 ret = check_raid_stripe_extent(leaf, key, slot);
1822 break;
1823 }
1824
1825 if (ret)
1826 return BTRFS_TREE_BLOCK_INVALID_ITEM;
1827 return BTRFS_TREE_BLOCK_CLEAN;
1828 }
1829
1830 enum btrfs_tree_block_status __btrfs_check_leaf(struct extent_buffer *leaf)
1831 {
1832 struct btrfs_fs_info *fs_info = leaf->fs_info;
1833 /* No valid key type is 0, so all key should be larger than this key */
1834 struct btrfs_key prev_key = {0, 0, 0};
1835 struct btrfs_key key;
1836 u32 nritems = btrfs_header_nritems(leaf);
1837 int slot;
1838
1839 if (unlikely(btrfs_header_level(leaf) != 0)) {
1840 generic_err(leaf, 0,
1841 "invalid level for leaf, have %d expect 0",
1842 btrfs_header_level(leaf));
1843 return BTRFS_TREE_BLOCK_INVALID_LEVEL;
1844 }
1845
1846 if (unlikely(!btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_WRITTEN))) {
1847 generic_err(leaf, 0, "invalid flag for leaf, WRITTEN not set");
1848 return BTRFS_TREE_BLOCK_WRITTEN_NOT_SET;
1849 }
1850
1851 /*
1852 * Extent buffers from a relocation tree have a owner field that
1853 * corresponds to the subvolume tree they are based on. So just from an
1854 * extent buffer alone we can not find out what is the id of the
1855 * corresponding subvolume tree, so we can not figure out if the extent
1856 * buffer corresponds to the root of the relocation tree or not. So
1857 * skip this check for relocation trees.
1858 */
1859 if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
1860 u64 owner = btrfs_header_owner(leaf);
1861
1862 /* These trees must never be empty */
1863 if (unlikely(owner == BTRFS_ROOT_TREE_OBJECTID ||
1864 owner == BTRFS_CHUNK_TREE_OBJECTID ||
1865 owner == BTRFS_DEV_TREE_OBJECTID ||
1866 owner == BTRFS_FS_TREE_OBJECTID ||
1867 owner == BTRFS_DATA_RELOC_TREE_OBJECTID)) {
1868 generic_err(leaf, 0,
1869 "invalid root, root %llu must never be empty",
1870 owner);
1871 return BTRFS_TREE_BLOCK_INVALID_NRITEMS;
1872 }
1873
1874 /* Unknown tree */
1875 if (unlikely(owner == 0)) {
1876 generic_err(leaf, 0,
1877 "invalid owner, root 0 is not defined");
1878 return BTRFS_TREE_BLOCK_INVALID_OWNER;
1879 }
1880
1881 /* EXTENT_TREE_V2 can have empty extent trees. */
1882 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1883 return BTRFS_TREE_BLOCK_CLEAN;
1884
1885 if (unlikely(owner == BTRFS_EXTENT_TREE_OBJECTID)) {
1886 generic_err(leaf, 0,
1887 "invalid root, root %llu must never be empty",
1888 owner);
1889 return BTRFS_TREE_BLOCK_INVALID_NRITEMS;
1890 }
1891
1892 return BTRFS_TREE_BLOCK_CLEAN;
1893 }
1894
1895 if (unlikely(nritems == 0))
1896 return BTRFS_TREE_BLOCK_CLEAN;
1897
1898 /*
1899 * Check the following things to make sure this is a good leaf, and
1900 * leaf users won't need to bother with similar sanity checks:
1901 *
1902 * 1) key ordering
1903 * 2) item offset and size
1904 * No overlap, no hole, all inside the leaf.
1905 * 3) item content
1906 * If possible, do comprehensive sanity check.
1907 * NOTE: All checks must only rely on the item data itself.
1908 */
1909 for (slot = 0; slot < nritems; slot++) {
1910 u32 item_end_expected;
1911 u64 item_data_end;
1912 enum btrfs_tree_block_status ret;
1913
1914 btrfs_item_key_to_cpu(leaf, &key, slot);
1915
1916 /* Make sure the keys are in the right order */
1917 if (unlikely(btrfs_comp_cpu_keys(&prev_key, &key) >= 0)) {
1918 generic_err(leaf, slot,
1919 "bad key order, prev (%llu %u %llu) current (%llu %u %llu)",
1920 prev_key.objectid, prev_key.type,
1921 prev_key.offset, key.objectid, key.type,
1922 key.offset);
1923 return BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
1924 }
1925
1926 item_data_end = (u64)btrfs_item_offset(leaf, slot) +
1927 btrfs_item_size(leaf, slot);
1928 /*
1929 * Make sure the offset and ends are right, remember that the
1930 * item data starts at the end of the leaf and grows towards the
1931 * front.
1932 */
1933 if (slot == 0)
1934 item_end_expected = BTRFS_LEAF_DATA_SIZE(fs_info);
1935 else
1936 item_end_expected = btrfs_item_offset(leaf,
1937 slot - 1);
1938 if (unlikely(item_data_end != item_end_expected)) {
1939 generic_err(leaf, slot,
1940 "unexpected item end, have %llu expect %u",
1941 item_data_end, item_end_expected);
1942 return BTRFS_TREE_BLOCK_INVALID_OFFSETS;
1943 }
1944
1945 /*
1946 * Check to make sure that we don't point outside of the leaf,
1947 * just in case all the items are consistent to each other, but
1948 * all point outside of the leaf.
1949 */
1950 if (unlikely(item_data_end > BTRFS_LEAF_DATA_SIZE(fs_info))) {
1951 generic_err(leaf, slot,
1952 "slot end outside of leaf, have %llu expect range [0, %u]",
1953 item_data_end, BTRFS_LEAF_DATA_SIZE(fs_info));
1954 return BTRFS_TREE_BLOCK_INVALID_OFFSETS;
1955 }
1956
1957 /* Also check if the item pointer overlaps with btrfs item. */
1958 if (unlikely(btrfs_item_ptr_offset(leaf, slot) <
1959 btrfs_item_nr_offset(leaf, slot) + sizeof(struct btrfs_item))) {
1960 generic_err(leaf, slot,
1961 "slot overlaps with its data, item end %lu data start %lu",
1962 btrfs_item_nr_offset(leaf, slot) +
1963 sizeof(struct btrfs_item),
1964 btrfs_item_ptr_offset(leaf, slot));
1965 return BTRFS_TREE_BLOCK_INVALID_OFFSETS;
1966 }
1967
1968 /* Check if the item size and content meet other criteria. */
1969 ret = check_leaf_item(leaf, &key, slot, &prev_key);
1970 if (unlikely(ret != BTRFS_TREE_BLOCK_CLEAN))
1971 return ret;
1972
1973 prev_key.objectid = key.objectid;
1974 prev_key.type = key.type;
1975 prev_key.offset = key.offset;
1976 }
1977
1978 return BTRFS_TREE_BLOCK_CLEAN;
1979 }
1980
1981 int btrfs_check_leaf(struct extent_buffer *leaf)
1982 {
1983 enum btrfs_tree_block_status ret;
1984
1985 ret = __btrfs_check_leaf(leaf);
1986 if (unlikely(ret != BTRFS_TREE_BLOCK_CLEAN))
1987 return -EUCLEAN;
1988 return 0;
1989 }
1990 ALLOW_ERROR_INJECTION(btrfs_check_leaf, ERRNO);
1991
1992 enum btrfs_tree_block_status __btrfs_check_node(struct extent_buffer *node)
1993 {
1994 struct btrfs_fs_info *fs_info = node->fs_info;
1995 unsigned long nr = btrfs_header_nritems(node);
1996 struct btrfs_key key, next_key;
1997 int slot;
1998 int level = btrfs_header_level(node);
1999 u64 bytenr;
2000
2001 if (unlikely(!btrfs_header_flag(node, BTRFS_HEADER_FLAG_WRITTEN))) {
2002 generic_err(node, 0, "invalid flag for node, WRITTEN not set");
2003 return BTRFS_TREE_BLOCK_WRITTEN_NOT_SET;
2004 }
2005
2006 if (unlikely(level <= 0 || level >= BTRFS_MAX_LEVEL)) {
2007 generic_err(node, 0,
2008 "invalid level for node, have %d expect [1, %d]",
2009 level, BTRFS_MAX_LEVEL - 1);
2010 return BTRFS_TREE_BLOCK_INVALID_LEVEL;
2011 }
2012 if (unlikely(nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(fs_info))) {
2013 btrfs_crit(fs_info,
2014 "corrupt node: root=%llu block=%llu, nritems too %s, have %lu expect range [1,%u]",
2015 btrfs_header_owner(node), node->start,
2016 nr == 0 ? "small" : "large", nr,
2017 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
2018 return BTRFS_TREE_BLOCK_INVALID_NRITEMS;
2019 }
2020
2021 for (slot = 0; slot < nr - 1; slot++) {
2022 bytenr = btrfs_node_blockptr(node, slot);
2023 btrfs_node_key_to_cpu(node, &key, slot);
2024 btrfs_node_key_to_cpu(node, &next_key, slot + 1);
2025
2026 if (unlikely(!bytenr)) {
2027 generic_err(node, slot,
2028 "invalid NULL node pointer");
2029 return BTRFS_TREE_BLOCK_INVALID_BLOCKPTR;
2030 }
2031 if (unlikely(!IS_ALIGNED(bytenr, fs_info->sectorsize))) {
2032 generic_err(node, slot,
2033 "unaligned pointer, have %llu should be aligned to %u",
2034 bytenr, fs_info->sectorsize);
2035 return BTRFS_TREE_BLOCK_INVALID_BLOCKPTR;
2036 }
2037
2038 if (unlikely(btrfs_comp_cpu_keys(&key, &next_key) >= 0)) {
2039 generic_err(node, slot,
2040 "bad key order, current (%llu %u %llu) next (%llu %u %llu)",
2041 key.objectid, key.type, key.offset,
2042 next_key.objectid, next_key.type,
2043 next_key.offset);
2044 return BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
2045 }
2046 }
2047 return BTRFS_TREE_BLOCK_CLEAN;
2048 }
2049
2050 int btrfs_check_node(struct extent_buffer *node)
2051 {
2052 enum btrfs_tree_block_status ret;
2053
2054 ret = __btrfs_check_node(node);
2055 if (unlikely(ret != BTRFS_TREE_BLOCK_CLEAN))
2056 return -EUCLEAN;
2057 return 0;
2058 }
2059 ALLOW_ERROR_INJECTION(btrfs_check_node, ERRNO);
2060
2061 int btrfs_check_eb_owner(const struct extent_buffer *eb, u64 root_owner)
2062 {
2063 const bool is_subvol = is_fstree(root_owner);
2064 const u64 eb_owner = btrfs_header_owner(eb);
2065
2066 /*
2067 * Skip dummy fs, as selftests don't create unique ebs for each dummy
2068 * root.
2069 */
2070 if (btrfs_is_testing(eb->fs_info))
2071 return 0;
2072 /*
2073 * There are several call sites (backref walking, qgroup, and data
2074 * reloc) passing 0 as @root_owner, as they are not holding the
2075 * tree root. In that case, we can not do a reliable ownership check,
2076 * so just exit.
2077 */
2078 if (root_owner == 0)
2079 return 0;
2080 /*
2081 * These trees use key.offset as their owner, our callers don't have
2082 * the extra capacity to pass key.offset here. So we just skip them.
2083 */
2084 if (root_owner == BTRFS_TREE_LOG_OBJECTID ||
2085 root_owner == BTRFS_TREE_RELOC_OBJECTID)
2086 return 0;
2087
2088 if (!is_subvol) {
2089 /* For non-subvolume trees, the eb owner should match root owner */
2090 if (unlikely(root_owner != eb_owner)) {
2091 btrfs_crit(eb->fs_info,
2092 "corrupted %s, root=%llu block=%llu owner mismatch, have %llu expect %llu",
2093 btrfs_header_level(eb) == 0 ? "leaf" : "node",
2094 root_owner, btrfs_header_bytenr(eb), eb_owner,
2095 root_owner);
2096 return -EUCLEAN;
2097 }
2098 return 0;
2099 }
2100
2101 /*
2102 * For subvolume trees, owners can mismatch, but they should all belong
2103 * to subvolume trees.
2104 */
2105 if (unlikely(is_subvol != is_fstree(eb_owner))) {
2106 btrfs_crit(eb->fs_info,
2107 "corrupted %s, root=%llu block=%llu owner mismatch, have %llu expect [%llu, %llu]",
2108 btrfs_header_level(eb) == 0 ? "leaf" : "node",
2109 root_owner, btrfs_header_bytenr(eb), eb_owner,
2110 BTRFS_FIRST_FREE_OBJECTID, BTRFS_LAST_FREE_OBJECTID);
2111 return -EUCLEAN;
2112 }
2113 return 0;
2114 }
2115
2116 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
2117 struct btrfs_key *first_key, u64 parent_transid)
2118 {
2119 struct btrfs_fs_info *fs_info = eb->fs_info;
2120 int found_level;
2121 struct btrfs_key found_key;
2122 int ret;
2123
2124 found_level = btrfs_header_level(eb);
2125 if (found_level != level) {
2126 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
2127 KERN_ERR "BTRFS: tree level check failed\n");
2128 btrfs_err(fs_info,
2129 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
2130 eb->start, level, found_level);
2131 return -EIO;
2132 }
2133
2134 if (!first_key)
2135 return 0;
2136
2137 /*
2138 * For live tree block (new tree blocks in current transaction),
2139 * we need proper lock context to avoid race, which is impossible here.
2140 * So we only checks tree blocks which is read from disk, whose
2141 * generation <= fs_info->last_trans_committed.
2142 */
2143 if (btrfs_header_generation(eb) > btrfs_get_last_trans_committed(fs_info))
2144 return 0;
2145
2146 /* We have @first_key, so this @eb must have at least one item */
2147 if (btrfs_header_nritems(eb) == 0) {
2148 btrfs_err(fs_info,
2149 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
2150 eb->start);
2151 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
2152 return -EUCLEAN;
2153 }
2154
2155 if (found_level)
2156 btrfs_node_key_to_cpu(eb, &found_key, 0);
2157 else
2158 btrfs_item_key_to_cpu(eb, &found_key, 0);
2159 ret = btrfs_comp_cpu_keys(first_key, &found_key);
2160
2161 if (ret) {
2162 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
2163 KERN_ERR "BTRFS: tree first key check failed\n");
2164 btrfs_err(fs_info,
2165 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
2166 eb->start, parent_transid, first_key->objectid,
2167 first_key->type, first_key->offset,
2168 found_key.objectid, found_key.type,
2169 found_key.offset);
2170 }
2171 return ret;
2172 }
2173
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.