?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 6.13-rc4
[linux.git] / fs / btrfs / tree-checker.c
blobdfeee033f31fb93e2dfbe42be48926f6056465de
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_UNKNOWN ||
573 dir_type >= BTRFS_FT_MAX)) {
574 dir_item_err(leaf, slot,
575 "invalid dir item type, have %u expect (0, %u)",
576 dir_type, BTRFS_FT_MAX);
577 return -EUCLEAN;
578 }
579
580 if (unlikely(key->type == BTRFS_XATTR_ITEM_KEY &&
581 dir_type != BTRFS_FT_XATTR)) {
582 dir_item_err(leaf, slot,
583 "invalid dir item type for XATTR key, have %u expect %u",
584 dir_type, BTRFS_FT_XATTR);
585 return -EUCLEAN;
586 }
587 if (unlikely(dir_type == BTRFS_FT_XATTR &&
588 key->type != BTRFS_XATTR_ITEM_KEY)) {
589 dir_item_err(leaf, slot,
590 "xattr dir type found for non-XATTR key");
591 return -EUCLEAN;
592 }
593 if (dir_type == BTRFS_FT_XATTR)
594 max_name_len = XATTR_NAME_MAX;
595 else
596 max_name_len = BTRFS_NAME_LEN;
597
598 /* Name/data length check */
599 name_len = btrfs_dir_name_len(leaf, di);
600 data_len = btrfs_dir_data_len(leaf, di);
601 if (unlikely(name_len > max_name_len)) {
602 dir_item_err(leaf, slot,
603 "dir item name len too long, have %u max %u",
604 name_len, max_name_len);
605 return -EUCLEAN;
606 }
607 if (unlikely(name_len + data_len > BTRFS_MAX_XATTR_SIZE(fs_info))) {
608 dir_item_err(leaf, slot,
609 "dir item name and data len too long, have %u max %u",
610 name_len + data_len,
611 BTRFS_MAX_XATTR_SIZE(fs_info));
612 return -EUCLEAN;
613 }
614
615 if (unlikely(data_len && dir_type != BTRFS_FT_XATTR)) {
616 dir_item_err(leaf, slot,
617 "dir item with invalid data len, have %u expect 0",
618 data_len);
619 return -EUCLEAN;
620 }
621
622 total_size = sizeof(*di) + name_len + data_len;
623
624 /* header and name/data should not cross item boundary */
625 if (unlikely(cur + total_size > item_size)) {
626 dir_item_err(leaf, slot,
627 "dir item data crosses item boundary, have %u boundary %u",
628 cur + total_size, item_size);
629 return -EUCLEAN;
630 }
631
632 /*
633 * Special check for XATTR/DIR_ITEM, as key->offset is name
634 * hash, should match its name
635 */
636 if (key->type == BTRFS_DIR_ITEM_KEY ||
637 key->type == BTRFS_XATTR_ITEM_KEY) {
638 char namebuf[MAX(BTRFS_NAME_LEN, XATTR_NAME_MAX)];
639
640 read_extent_buffer(leaf, namebuf,
641 (unsigned long)(di + 1), name_len);
642 name_hash = btrfs_name_hash(namebuf, name_len);
643 if (unlikely(key->offset != name_hash)) {
644 dir_item_err(leaf, slot,
645 "name hash mismatch with key, have 0x%016x expect 0x%016llx",
646 name_hash, key->offset);
647 return -EUCLEAN;
648 }
649 }
650 cur += total_size;
651 di = (struct btrfs_dir_item *)((void *)di + total_size);
652 }
653 return 0;
654 }
655
656 __printf(3, 4)
657 __cold
658 static void block_group_err(const struct extent_buffer *eb, int slot,
659 const char *fmt, ...)
660 {
661 const struct btrfs_fs_info *fs_info = eb->fs_info;
662 struct btrfs_key key;
663 struct va_format vaf;
664 va_list args;
665
666 btrfs_item_key_to_cpu(eb, &key, slot);
667 va_start(args, fmt);
668
669 vaf.fmt = fmt;
670 vaf.va = &args;
671
672 dump_page(folio_page(eb->folios[0], 0), "eb page dump");
673 btrfs_crit(fs_info,
674 "corrupt %s: root=%llu block=%llu slot=%d bg_start=%llu bg_len=%llu, %pV",
675 btrfs_header_level(eb) == 0 ? "leaf" : "node",
676 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
677 key.objectid, key.offset, &vaf);
678 va_end(args);
679 }
680
681 static int check_block_group_item(struct extent_buffer *leaf,
682 struct btrfs_key *key, int slot)
683 {
684 struct btrfs_fs_info *fs_info = leaf->fs_info;
685 struct btrfs_block_group_item bgi;
686 u32 item_size = btrfs_item_size(leaf, slot);
687 u64 chunk_objectid;
688 u64 flags;
689 u64 type;
690
691 /*
692 * Here we don't really care about alignment since extent allocator can
693 * handle it. We care more about the size.
694 */
695 if (unlikely(key->offset == 0)) {
696 block_group_err(leaf, slot,
697 "invalid block group size 0");
698 return -EUCLEAN;
699 }
700
701 if (unlikely(item_size != sizeof(bgi))) {
702 block_group_err(leaf, slot,
703 "invalid item size, have %u expect %zu",
704 item_size, sizeof(bgi));
705 return -EUCLEAN;
706 }
707
708 read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot),
709 sizeof(bgi));
710 chunk_objectid = btrfs_stack_block_group_chunk_objectid(&bgi);
711 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
712 /*
713 * We don't init the nr_global_roots until we load the global
714 * roots, so this could be 0 at mount time. If it's 0 we'll
715 * just assume we're fine, and later we'll check against our
716 * actual value.
717 */
718 if (unlikely(fs_info->nr_global_roots &&
719 chunk_objectid >= fs_info->nr_global_roots)) {
720 block_group_err(leaf, slot,
721 "invalid block group global root id, have %llu, needs to be <= %llu",
722 chunk_objectid,
723 fs_info->nr_global_roots);
724 return -EUCLEAN;
725 }
726 } else if (unlikely(chunk_objectid != BTRFS_FIRST_CHUNK_TREE_OBJECTID)) {
727 block_group_err(leaf, slot,
728 "invalid block group chunk objectid, have %llu expect %llu",
729 btrfs_stack_block_group_chunk_objectid(&bgi),
730 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
731 return -EUCLEAN;
732 }
733
734 if (unlikely(btrfs_stack_block_group_used(&bgi) > key->offset)) {
735 block_group_err(leaf, slot,
736 "invalid block group used, have %llu expect [0, %llu)",
737 btrfs_stack_block_group_used(&bgi), key->offset);
738 return -EUCLEAN;
739 }
740
741 flags = btrfs_stack_block_group_flags(&bgi);
742 if (unlikely(hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) > 1)) {
743 block_group_err(leaf, slot,
744 "invalid profile flags, have 0x%llx (%lu bits set) expect no more than 1 bit set",
745 flags & BTRFS_BLOCK_GROUP_PROFILE_MASK,
746 hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK));
747 return -EUCLEAN;
748 }
749
750 type = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
751 if (unlikely(type != BTRFS_BLOCK_GROUP_DATA &&
752 type != BTRFS_BLOCK_GROUP_METADATA &&
753 type != BTRFS_BLOCK_GROUP_SYSTEM &&
754 type != (BTRFS_BLOCK_GROUP_METADATA |
755 BTRFS_BLOCK_GROUP_DATA))) {
756 block_group_err(leaf, slot,
757 "invalid type, have 0x%llx (%lu bits set) expect either 0x%llx, 0x%llx, 0x%llx or 0x%llx",
758 type, hweight64(type),
759 BTRFS_BLOCK_GROUP_DATA, BTRFS_BLOCK_GROUP_METADATA,
760 BTRFS_BLOCK_GROUP_SYSTEM,
761 BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA);
762 return -EUCLEAN;
763 }
764 return 0;
765 }
766
767 __printf(4, 5)
768 __cold
769 static void chunk_err(const struct extent_buffer *leaf,
770 const struct btrfs_chunk *chunk, u64 logical,
771 const char *fmt, ...)
772 {
773 const struct btrfs_fs_info *fs_info = leaf->fs_info;
774 bool is_sb;
775 struct va_format vaf;
776 va_list args;
777 int i;
778 int slot = -1;
779
780 /* Only superblock eb is able to have such small offset */
781 is_sb = (leaf->start == BTRFS_SUPER_INFO_OFFSET);
782
783 if (!is_sb) {
784 /*
785 * Get the slot number by iterating through all slots, this
786 * would provide better readability.
787 */
788 for (i = 0; i < btrfs_header_nritems(leaf); i++) {
789 if (btrfs_item_ptr_offset(leaf, i) ==
790 (unsigned long)chunk) {
791 slot = i;
792 break;
793 }
794 }
795 }
796 va_start(args, fmt);
797 vaf.fmt = fmt;
798 vaf.va = &args;
799
800 if (is_sb)
801 btrfs_crit(fs_info,
802 "corrupt superblock syschunk array: chunk_start=%llu, %pV",
803 logical, &vaf);
804 else
805 btrfs_crit(fs_info,
806 "corrupt leaf: root=%llu block=%llu slot=%d chunk_start=%llu, %pV",
807 BTRFS_CHUNK_TREE_OBJECTID, leaf->start, slot,
808 logical, &vaf);
809 va_end(args);
810 }
811
812 /*
813 * The common chunk check which could also work on super block sys chunk array.
814 *
815 * Return -EUCLEAN if anything is corrupted.
816 * Return 0 if everything is OK.
817 */
818 int btrfs_check_chunk_valid(struct extent_buffer *leaf,
819 struct btrfs_chunk *chunk, u64 logical)
820 {
821 struct btrfs_fs_info *fs_info = leaf->fs_info;
822 u64 length;
823 u64 chunk_end;
824 u64 stripe_len;
825 u16 num_stripes;
826 u16 sub_stripes;
827 u64 type;
828 u64 features;
829 bool mixed = false;
830 int raid_index;
831 int nparity;
832 int ncopies;
833
834 length = btrfs_chunk_length(leaf, chunk);
835 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
836 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
837 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
838 type = btrfs_chunk_type(leaf, chunk);
839 raid_index = btrfs_bg_flags_to_raid_index(type);
840 ncopies = btrfs_raid_array[raid_index].ncopies;
841 nparity = btrfs_raid_array[raid_index].nparity;
842
843 if (unlikely(!num_stripes)) {
844 chunk_err(leaf, chunk, logical,
845 "invalid chunk num_stripes, have %u", num_stripes);
846 return -EUCLEAN;
847 }
848 if (unlikely(num_stripes < ncopies)) {
849 chunk_err(leaf, chunk, logical,
850 "invalid chunk num_stripes < ncopies, have %u < %d",
851 num_stripes, ncopies);
852 return -EUCLEAN;
853 }
854 if (unlikely(nparity && num_stripes == nparity)) {
855 chunk_err(leaf, chunk, logical,
856 "invalid chunk num_stripes == nparity, have %u == %d",
857 num_stripes, nparity);
858 return -EUCLEAN;
859 }
860 if (unlikely(!IS_ALIGNED(logical, fs_info->sectorsize))) {
861 chunk_err(leaf, chunk, logical,
862 "invalid chunk logical, have %llu should aligned to %u",
863 logical, fs_info->sectorsize);
864 return -EUCLEAN;
865 }
866 if (unlikely(btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize)) {
867 chunk_err(leaf, chunk, logical,
868 "invalid chunk sectorsize, have %u expect %u",
869 btrfs_chunk_sector_size(leaf, chunk),
870 fs_info->sectorsize);
871 return -EUCLEAN;
872 }
873 if (unlikely(!length || !IS_ALIGNED(length, fs_info->sectorsize))) {
874 chunk_err(leaf, chunk, logical,
875 "invalid chunk length, have %llu", length);
876 return -EUCLEAN;
877 }
878 if (unlikely(check_add_overflow(logical, length, &chunk_end))) {
879 chunk_err(leaf, chunk, logical,
880 "invalid chunk logical start and length, have logical start %llu length %llu",
881 logical, length);
882 return -EUCLEAN;
883 }
884 if (unlikely(!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN)) {
885 chunk_err(leaf, chunk, logical,
886 "invalid chunk stripe length: %llu",
887 stripe_len);
888 return -EUCLEAN;
889 }
890 /*
891 * We artificially limit the chunk size, so that the number of stripes
892 * inside a chunk can be fit into a U32. The current limit (256G) is
893 * way too large for real world usage anyway, and it's also much larger
894 * than our existing limit (10G).
895 *
896 * Thus it should be a good way to catch obvious bitflips.
897 */
898 if (unlikely(length >= btrfs_stripe_nr_to_offset(U32_MAX))) {
899 chunk_err(leaf, chunk, logical,
900 "chunk length too large: have %llu limit %llu",
901 length, btrfs_stripe_nr_to_offset(U32_MAX));
902 return -EUCLEAN;
903 }
904 if (unlikely(type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
905 BTRFS_BLOCK_GROUP_PROFILE_MASK))) {
906 chunk_err(leaf, chunk, logical,
907 "unrecognized chunk type: 0x%llx",
908 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
909 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
910 btrfs_chunk_type(leaf, chunk));
911 return -EUCLEAN;
912 }
913
914 if (unlikely(!has_single_bit_set(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
915 (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) != 0)) {
916 chunk_err(leaf, chunk, logical,
917 "invalid chunk profile flag: 0x%llx, expect 0 or 1 bit set",
918 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
919 return -EUCLEAN;
920 }
921 if (unlikely((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0)) {
922 chunk_err(leaf, chunk, logical,
923 "missing chunk type flag, have 0x%llx one bit must be set in 0x%llx",
924 type, BTRFS_BLOCK_GROUP_TYPE_MASK);
925 return -EUCLEAN;
926 }
927
928 if (unlikely((type & BTRFS_BLOCK_GROUP_SYSTEM) &&
929 (type & (BTRFS_BLOCK_GROUP_METADATA |
930 BTRFS_BLOCK_GROUP_DATA)))) {
931 chunk_err(leaf, chunk, logical,
932 "system chunk with data or metadata type: 0x%llx",
933 type);
934 return -EUCLEAN;
935 }
936
937 features = btrfs_super_incompat_flags(fs_info->super_copy);
938 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
939 mixed = true;
940
941 if (!mixed) {
942 if (unlikely((type & BTRFS_BLOCK_GROUP_METADATA) &&
943 (type & BTRFS_BLOCK_GROUP_DATA))) {
944 chunk_err(leaf, chunk, logical,
945 "mixed chunk type in non-mixed mode: 0x%llx", type);
946 return -EUCLEAN;
947 }
948 }
949
950 if (unlikely((type & BTRFS_BLOCK_GROUP_RAID10 &&
951 sub_stripes != btrfs_raid_array[BTRFS_RAID_RAID10].sub_stripes) ||
952 (type & BTRFS_BLOCK_GROUP_RAID1 &&
953 num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1].devs_min) ||
954 (type & BTRFS_BLOCK_GROUP_RAID1C3 &&
955 num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1C3].devs_min) ||
956 (type & BTRFS_BLOCK_GROUP_RAID1C4 &&
957 num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1C4].devs_min) ||
958 (type & BTRFS_BLOCK_GROUP_RAID5 &&
959 num_stripes < btrfs_raid_array[BTRFS_RAID_RAID5].devs_min) ||
960 (type & BTRFS_BLOCK_GROUP_RAID6 &&
961 num_stripes < btrfs_raid_array[BTRFS_RAID_RAID6].devs_min) ||
962 (type & BTRFS_BLOCK_GROUP_DUP &&
963 num_stripes != btrfs_raid_array[BTRFS_RAID_DUP].dev_stripes) ||
964 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
965 num_stripes != btrfs_raid_array[BTRFS_RAID_SINGLE].dev_stripes))) {
966 chunk_err(leaf, chunk, logical,
967 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
968 num_stripes, sub_stripes,
969 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
970 return -EUCLEAN;
971 }
972
973 return 0;
974 }
975
976 /*
977 * Enhanced version of chunk item checker.
978 *
979 * The common btrfs_check_chunk_valid() doesn't check item size since it needs
980 * to work on super block sys_chunk_array which doesn't have full item ptr.
981 */
982 static int check_leaf_chunk_item(struct extent_buffer *leaf,
983 struct btrfs_chunk *chunk,
984 struct btrfs_key *key, int slot)
985 {
986 int num_stripes;
987
988 if (unlikely(btrfs_item_size(leaf, slot) < sizeof(struct btrfs_chunk))) {
989 chunk_err(leaf, chunk, key->offset,
990 "invalid chunk item size: have %u expect [%zu, %u)",
991 btrfs_item_size(leaf, slot),
992 sizeof(struct btrfs_chunk),
993 BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
994 return -EUCLEAN;
995 }
996
997 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
998 /* Let btrfs_check_chunk_valid() handle this error type */
999 if (num_stripes == 0)
1000 goto out;
1001
1002 if (unlikely(btrfs_chunk_item_size(num_stripes) !=
1003 btrfs_item_size(leaf, slot))) {
1004 chunk_err(leaf, chunk, key->offset,
1005 "invalid chunk item size: have %u expect %lu",
1006 btrfs_item_size(leaf, slot),
1007 btrfs_chunk_item_size(num_stripes));
1008 return -EUCLEAN;
1009 }
1010 out:
1011 return btrfs_check_chunk_valid(leaf, chunk, key->offset);
1012 }
1013
1014 __printf(3, 4)
1015 __cold
1016 static void dev_item_err(const struct extent_buffer *eb, int slot,
1017 const char *fmt, ...)
1018 {
1019 struct btrfs_key key;
1020 struct va_format vaf;
1021 va_list args;
1022
1023 btrfs_item_key_to_cpu(eb, &key, slot);
1024 va_start(args, fmt);
1025
1026 vaf.fmt = fmt;
1027 vaf.va = &args;
1028
1029 dump_page(folio_page(eb->folios[0], 0), "eb page dump");
1030 btrfs_crit(eb->fs_info,
1031 "corrupt %s: root=%llu block=%llu slot=%d devid=%llu %pV",
1032 btrfs_header_level(eb) == 0 ? "leaf" : "node",
1033 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
1034 key.objectid, &vaf);
1035 va_end(args);
1036 }
1037
1038 static int check_dev_item(struct extent_buffer *leaf,
1039 struct btrfs_key *key, int slot)
1040 {
1041 struct btrfs_dev_item *ditem;
1042 const u32 item_size = btrfs_item_size(leaf, slot);
1043
1044 if (unlikely(key->objectid != BTRFS_DEV_ITEMS_OBJECTID)) {
1045 dev_item_err(leaf, slot,
1046 "invalid objectid: has=%llu expect=%llu",
1047 key->objectid, BTRFS_DEV_ITEMS_OBJECTID);
1048 return -EUCLEAN;
1049 }
1050
1051 if (unlikely(item_size != sizeof(*ditem))) {
1052 dev_item_err(leaf, slot, "invalid item size: has %u expect %zu",
1053 item_size, sizeof(*ditem));
1054 return -EUCLEAN;
1055 }
1056
1057 ditem = btrfs_item_ptr(leaf, slot, struct btrfs_dev_item);
1058 if (unlikely(btrfs_device_id(leaf, ditem) != key->offset)) {
1059 dev_item_err(leaf, slot,
1060 "devid mismatch: key has=%llu item has=%llu",
1061 key->offset, btrfs_device_id(leaf, ditem));
1062 return -EUCLEAN;
1063 }
1064
1065 /*
1066 * For device total_bytes, we don't have reliable way to check it, as
1067 * it can be 0 for device removal. Device size check can only be done
1068 * by dev extents check.
1069 */
1070 if (unlikely(btrfs_device_bytes_used(leaf, ditem) >
1071 btrfs_device_total_bytes(leaf, ditem))) {
1072 dev_item_err(leaf, slot,
1073 "invalid bytes used: have %llu expect [0, %llu]",
1074 btrfs_device_bytes_used(leaf, ditem),
1075 btrfs_device_total_bytes(leaf, ditem));
1076 return -EUCLEAN;
1077 }
1078 /*
1079 * Remaining members like io_align/type/gen/dev_group aren't really
1080 * utilized. Skip them to make later usage of them easier.
1081 */
1082 return 0;
1083 }
1084
1085 static int check_inode_item(struct extent_buffer *leaf,
1086 struct btrfs_key *key, int slot)
1087 {
1088 struct btrfs_fs_info *fs_info = leaf->fs_info;
1089 struct btrfs_inode_item *iitem;
1090 u64 super_gen = btrfs_super_generation(fs_info->super_copy);
1091 u32 valid_mask = (S_IFMT | S_ISUID | S_ISGID | S_ISVTX | 0777);
1092 const u32 item_size = btrfs_item_size(leaf, slot);
1093 u32 mode;
1094 int ret;
1095 u32 flags;
1096 u32 ro_flags;
1097
1098 ret = check_inode_key(leaf, key, slot);
1099 if (unlikely(ret < 0))
1100 return ret;
1101
1102 if (unlikely(item_size != sizeof(*iitem))) {
1103 generic_err(leaf, slot, "invalid item size: has %u expect %zu",
1104 item_size, sizeof(*iitem));
1105 return -EUCLEAN;
1106 }
1107
1108 iitem = btrfs_item_ptr(leaf, slot, struct btrfs_inode_item);
1109
1110 /* Here we use super block generation + 1 to handle log tree */
1111 if (unlikely(btrfs_inode_generation(leaf, iitem) > super_gen + 1)) {
1112 inode_item_err(leaf, slot,
1113 "invalid inode generation: has %llu expect (0, %llu]",
1114 btrfs_inode_generation(leaf, iitem),
1115 super_gen + 1);
1116 return -EUCLEAN;
1117 }
1118 /* Note for ROOT_TREE_DIR_ITEM, mkfs could set its transid 0 */
1119 if (unlikely(btrfs_inode_transid(leaf, iitem) > super_gen + 1)) {
1120 inode_item_err(leaf, slot,
1121 "invalid inode transid: has %llu expect [0, %llu]",
1122 btrfs_inode_transid(leaf, iitem), super_gen + 1);
1123 return -EUCLEAN;
1124 }
1125
1126 /*
1127 * For size and nbytes it's better not to be too strict, as for dir
1128 * item its size/nbytes can easily get wrong, but doesn't affect
1129 * anything in the fs. So here we skip the check.
1130 */
1131 mode = btrfs_inode_mode(leaf, iitem);
1132 if (unlikely(mode & ~valid_mask)) {
1133 inode_item_err(leaf, slot,
1134 "unknown mode bit detected: 0x%x",
1135 mode & ~valid_mask);
1136 return -EUCLEAN;
1137 }
1138
1139 /*
1140 * S_IFMT is not bit mapped so we can't completely rely on
1141 * is_power_of_2/has_single_bit_set, but it can save us from checking
1142 * FIFO/CHR/DIR/REG. Only needs to check BLK, LNK and SOCKS
1143 */
1144 if (!has_single_bit_set(mode & S_IFMT)) {
1145 if (unlikely(!S_ISLNK(mode) && !S_ISBLK(mode) && !S_ISSOCK(mode))) {
1146 inode_item_err(leaf, slot,
1147 "invalid mode: has 0%o expect valid S_IF* bit(s)",
1148 mode & S_IFMT);
1149 return -EUCLEAN;
1150 }
1151 }
1152 if (unlikely(S_ISDIR(mode) && btrfs_inode_nlink(leaf, iitem) > 1)) {
1153 inode_item_err(leaf, slot,
1154 "invalid nlink: has %u expect no more than 1 for dir",
1155 btrfs_inode_nlink(leaf, iitem));
1156 return -EUCLEAN;
1157 }
1158 btrfs_inode_split_flags(btrfs_inode_flags(leaf, iitem), &flags, &ro_flags);
1159 if (unlikely(flags & ~BTRFS_INODE_FLAG_MASK)) {
1160 inode_item_err(leaf, slot,
1161 "unknown incompat flags detected: 0x%x", flags);
1162 return -EUCLEAN;
1163 }
1164 if (unlikely(!sb_rdonly(fs_info->sb) &&
1165 (ro_flags & ~BTRFS_INODE_RO_FLAG_MASK))) {
1166 inode_item_err(leaf, slot,
1167 "unknown ro-compat flags detected on writeable mount: 0x%x",
1168 ro_flags);
1169 return -EUCLEAN;
1170 }
1171 return 0;
1172 }
1173
1174 static int check_root_item(struct extent_buffer *leaf, struct btrfs_key *key,
1175 int slot)
1176 {
1177 struct btrfs_fs_info *fs_info = leaf->fs_info;
1178 struct btrfs_root_item ri = { 0 };
1179 const u64 valid_root_flags = BTRFS_ROOT_SUBVOL_RDONLY |
1180 BTRFS_ROOT_SUBVOL_DEAD;
1181 int ret;
1182
1183 ret = check_root_key(leaf, key, slot);
1184 if (unlikely(ret < 0))
1185 return ret;
1186
1187 if (unlikely(btrfs_item_size(leaf, slot) != sizeof(ri) &&
1188 btrfs_item_size(leaf, slot) !=
1189 btrfs_legacy_root_item_size())) {
1190 generic_err(leaf, slot,
1191 "invalid root item size, have %u expect %zu or %u",
1192 btrfs_item_size(leaf, slot), sizeof(ri),
1193 btrfs_legacy_root_item_size());
1194 return -EUCLEAN;
1195 }
1196
1197 /*
1198 * For legacy root item, the members starting at generation_v2 will be
1199 * all filled with 0.
1200 * And since we allow geneartion_v2 as 0, it will still pass the check.
1201 */
1202 read_extent_buffer(leaf, &ri, btrfs_item_ptr_offset(leaf, slot),
1203 btrfs_item_size(leaf, slot));
1204
1205 /* Generation related */
1206 if (unlikely(btrfs_root_generation(&ri) >
1207 btrfs_super_generation(fs_info->super_copy) + 1)) {
1208 generic_err(leaf, slot,
1209 "invalid root generation, have %llu expect (0, %llu]",
1210 btrfs_root_generation(&ri),
1211 btrfs_super_generation(fs_info->super_copy) + 1);
1212 return -EUCLEAN;
1213 }
1214 if (unlikely(btrfs_root_generation_v2(&ri) >
1215 btrfs_super_generation(fs_info->super_copy) + 1)) {
1216 generic_err(leaf, slot,
1217 "invalid root v2 generation, have %llu expect (0, %llu]",
1218 btrfs_root_generation_v2(&ri),
1219 btrfs_super_generation(fs_info->super_copy) + 1);
1220 return -EUCLEAN;
1221 }
1222 if (unlikely(btrfs_root_last_snapshot(&ri) >
1223 btrfs_super_generation(fs_info->super_copy) + 1)) {
1224 generic_err(leaf, slot,
1225 "invalid root last_snapshot, have %llu expect (0, %llu]",
1226 btrfs_root_last_snapshot(&ri),
1227 btrfs_super_generation(fs_info->super_copy) + 1);
1228 return -EUCLEAN;
1229 }
1230
1231 /* Alignment and level check */
1232 if (unlikely(!IS_ALIGNED(btrfs_root_bytenr(&ri), fs_info->sectorsize))) {
1233 generic_err(leaf, slot,
1234 "invalid root bytenr, have %llu expect to be aligned to %u",
1235 btrfs_root_bytenr(&ri), fs_info->sectorsize);
1236 return -EUCLEAN;
1237 }
1238 if (unlikely(btrfs_root_level(&ri) >= BTRFS_MAX_LEVEL)) {
1239 generic_err(leaf, slot,
1240 "invalid root level, have %u expect [0, %u]",
1241 btrfs_root_level(&ri), BTRFS_MAX_LEVEL - 1);
1242 return -EUCLEAN;
1243 }
1244 if (unlikely(btrfs_root_drop_level(&ri) >= BTRFS_MAX_LEVEL)) {
1245 generic_err(leaf, slot,
1246 "invalid root level, have %u expect [0, %u]",
1247 btrfs_root_drop_level(&ri), BTRFS_MAX_LEVEL - 1);
1248 return -EUCLEAN;
1249 }
1250
1251 /* Flags check */
1252 if (unlikely(btrfs_root_flags(&ri) & ~valid_root_flags)) {
1253 generic_err(leaf, slot,
1254 "invalid root flags, have 0x%llx expect mask 0x%llx",
1255 btrfs_root_flags(&ri), valid_root_flags);
1256 return -EUCLEAN;
1257 }
1258 return 0;
1259 }
1260
1261 __printf(3,4)
1262 __cold
1263 static void extent_err(const struct extent_buffer *eb, int slot,
1264 const char *fmt, ...)
1265 {
1266 struct btrfs_key key;
1267 struct va_format vaf;
1268 va_list args;
1269 u64 bytenr;
1270 u64 len;
1271
1272 btrfs_item_key_to_cpu(eb, &key, slot);
1273 bytenr = key.objectid;
1274 if (key.type == BTRFS_METADATA_ITEM_KEY ||
1275 key.type == BTRFS_TREE_BLOCK_REF_KEY ||
1276 key.type == BTRFS_SHARED_BLOCK_REF_KEY)
1277 len = eb->fs_info->nodesize;
1278 else
1279 len = key.offset;
1280 va_start(args, fmt);
1281
1282 vaf.fmt = fmt;
1283 vaf.va = &args;
1284
1285 dump_page(folio_page(eb->folios[0], 0), "eb page dump");
1286 btrfs_crit(eb->fs_info,
1287 "corrupt %s: block=%llu slot=%d extent bytenr=%llu len=%llu %pV",
1288 btrfs_header_level(eb) == 0 ? "leaf" : "node",
1289 eb->start, slot, bytenr, len, &vaf);
1290 va_end(args);
1291 }
1292
1293 static bool is_valid_dref_root(u64 rootid)
1294 {
1295 /*
1296 * The following tree root objectids are allowed to have a data backref:
1297 * - subvolume trees
1298 * - data reloc tree
1299 * - tree root
1300 * For v1 space cache
1301 */
1302 return is_fstree(rootid) || rootid == BTRFS_DATA_RELOC_TREE_OBJECTID ||
1303 rootid == BTRFS_ROOT_TREE_OBJECTID;
1304 }
1305
1306 static int check_extent_item(struct extent_buffer *leaf,
1307 struct btrfs_key *key, int slot,
1308 struct btrfs_key *prev_key)
1309 {
1310 struct btrfs_fs_info *fs_info = leaf->fs_info;
1311 struct btrfs_extent_item *ei;
1312 bool is_tree_block = false;
1313 unsigned long ptr; /* Current pointer inside inline refs */
1314 unsigned long end; /* Extent item end */
1315 const u32 item_size = btrfs_item_size(leaf, slot);
1316 u8 last_type = 0;
1317 u64 last_seq = U64_MAX;
1318 u64 flags;
1319 u64 generation;
1320 u64 total_refs; /* Total refs in btrfs_extent_item */
1321 u64 inline_refs = 0; /* found total inline refs */
1322
1323 if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1324 !btrfs_fs_incompat(fs_info, SKINNY_METADATA))) {
1325 generic_err(leaf, slot,
1326 "invalid key type, METADATA_ITEM type invalid when SKINNY_METADATA feature disabled");
1327 return -EUCLEAN;
1328 }
1329 /* key->objectid is the bytenr for both key types */
1330 if (unlikely(!IS_ALIGNED(key->objectid, fs_info->sectorsize))) {
1331 generic_err(leaf, slot,
1332 "invalid key objectid, have %llu expect to be aligned to %u",
1333 key->objectid, fs_info->sectorsize);
1334 return -EUCLEAN;
1335 }
1336
1337 /* key->offset is tree level for METADATA_ITEM_KEY */
1338 if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1339 key->offset >= BTRFS_MAX_LEVEL)) {
1340 extent_err(leaf, slot,
1341 "invalid tree level, have %llu expect [0, %u]",
1342 key->offset, BTRFS_MAX_LEVEL - 1);
1343 return -EUCLEAN;
1344 }
1345
1346 /*
1347 * EXTENT/METADATA_ITEM consists of:
1348 * 1) One btrfs_extent_item
1349 * Records the total refs, type and generation of the extent.
1350 *
1351 * 2) One btrfs_tree_block_info (for EXTENT_ITEM and tree backref only)
1352 * Records the first key and level of the tree block.
1353 *
1354 * 2) Zero or more btrfs_extent_inline_ref(s)
1355 * Each inline ref has one btrfs_extent_inline_ref shows:
1356 * 2.1) The ref type, one of the 4
1357 * TREE_BLOCK_REF Tree block only
1358 * SHARED_BLOCK_REF Tree block only
1359 * EXTENT_DATA_REF Data only
1360 * SHARED_DATA_REF Data only
1361 * 2.2) Ref type specific data
1362 * Either using btrfs_extent_inline_ref::offset, or specific
1363 * data structure.
1364 *
1365 * All above inline items should follow the order:
1366 *
1367 * - All btrfs_extent_inline_ref::type should be in an ascending
1368 * order
1369 *
1370 * - Within the same type, the items should follow a descending
1371 * order by their sequence number. The sequence number is
1372 * determined by:
1373 * * btrfs_extent_inline_ref::offset for all types other than
1374 * EXTENT_DATA_REF
1375 * * hash_extent_data_ref() for EXTENT_DATA_REF
1376 */
1377 if (unlikely(item_size < sizeof(*ei))) {
1378 extent_err(leaf, slot,
1379 "invalid item size, have %u expect [%zu, %u)",
1380 item_size, sizeof(*ei),
1381 BTRFS_LEAF_DATA_SIZE(fs_info));
1382 return -EUCLEAN;
1383 }
1384 end = item_size + btrfs_item_ptr_offset(leaf, slot);
1385
1386 /* Checks against extent_item */
1387 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
1388 flags = btrfs_extent_flags(leaf, ei);
1389 total_refs = btrfs_extent_refs(leaf, ei);
1390 generation = btrfs_extent_generation(leaf, ei);
1391 if (unlikely(generation >
1392 btrfs_super_generation(fs_info->super_copy) + 1)) {
1393 extent_err(leaf, slot,
1394 "invalid generation, have %llu expect (0, %llu]",
1395 generation,
1396 btrfs_super_generation(fs_info->super_copy) + 1);
1397 return -EUCLEAN;
1398 }
1399 if (unlikely(!has_single_bit_set(flags & (BTRFS_EXTENT_FLAG_DATA |
1400 BTRFS_EXTENT_FLAG_TREE_BLOCK)))) {
1401 extent_err(leaf, slot,
1402 "invalid extent flag, have 0x%llx expect 1 bit set in 0x%llx",
1403 flags, BTRFS_EXTENT_FLAG_DATA |
1404 BTRFS_EXTENT_FLAG_TREE_BLOCK);
1405 return -EUCLEAN;
1406 }
1407 is_tree_block = !!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK);
1408 if (is_tree_block) {
1409 if (unlikely(key->type == BTRFS_EXTENT_ITEM_KEY &&
1410 key->offset != fs_info->nodesize)) {
1411 extent_err(leaf, slot,
1412 "invalid extent length, have %llu expect %u",
1413 key->offset, fs_info->nodesize);
1414 return -EUCLEAN;
1415 }
1416 } else {
1417 if (unlikely(key->type != BTRFS_EXTENT_ITEM_KEY)) {
1418 extent_err(leaf, slot,
1419 "invalid key type, have %u expect %u for data backref",
1420 key->type, BTRFS_EXTENT_ITEM_KEY);
1421 return -EUCLEAN;
1422 }
1423 if (unlikely(!IS_ALIGNED(key->offset, fs_info->sectorsize))) {
1424 extent_err(leaf, slot,
1425 "invalid extent length, have %llu expect aligned to %u",
1426 key->offset, fs_info->sectorsize);
1427 return -EUCLEAN;
1428 }
1429 if (unlikely(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
1430 extent_err(leaf, slot,
1431 "invalid extent flag, data has full backref set");
1432 return -EUCLEAN;
1433 }
1434 }
1435 ptr = (unsigned long)(struct btrfs_extent_item *)(ei + 1);
1436
1437 /* Check the special case of btrfs_tree_block_info */
1438 if (is_tree_block && key->type != BTRFS_METADATA_ITEM_KEY) {
1439 struct btrfs_tree_block_info *info;
1440
1441 info = (struct btrfs_tree_block_info *)ptr;
1442 if (unlikely(btrfs_tree_block_level(leaf, info) >= BTRFS_MAX_LEVEL)) {
1443 extent_err(leaf, slot,
1444 "invalid tree block info level, have %u expect [0, %u]",
1445 btrfs_tree_block_level(leaf, info),
1446 BTRFS_MAX_LEVEL - 1);
1447 return -EUCLEAN;
1448 }
1449 ptr = (unsigned long)(struct btrfs_tree_block_info *)(info + 1);
1450 }
1451
1452 /* Check inline refs */
1453 while (ptr < end) {
1454 struct btrfs_extent_inline_ref *iref;
1455 struct btrfs_extent_data_ref *dref;
1456 struct btrfs_shared_data_ref *sref;
1457 u64 seq;
1458 u64 dref_root;
1459 u64 dref_objectid;
1460 u64 dref_offset;
1461 u64 inline_offset;
1462 u8 inline_type;
1463
1464 if (unlikely(ptr + sizeof(*iref) > end)) {
1465 extent_err(leaf, slot,
1466 "inline ref item overflows extent item, ptr %lu iref size %zu end %lu",
1467 ptr, sizeof(*iref), end);
1468 return -EUCLEAN;
1469 }
1470 iref = (struct btrfs_extent_inline_ref *)ptr;
1471 inline_type = btrfs_extent_inline_ref_type(leaf, iref);
1472 inline_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1473 seq = inline_offset;
1474 if (unlikely(ptr + btrfs_extent_inline_ref_size(inline_type) > end)) {
1475 extent_err(leaf, slot,
1476 "inline ref item overflows extent item, ptr %lu iref size %u end %lu",
1477 ptr, btrfs_extent_inline_ref_size(inline_type), end);
1478 return -EUCLEAN;
1479 }
1480
1481 switch (inline_type) {
1482 /* inline_offset is subvolid of the owner, no need to check */
1483 case BTRFS_TREE_BLOCK_REF_KEY:
1484 inline_refs++;
1485 break;
1486 /* Contains parent bytenr */
1487 case BTRFS_SHARED_BLOCK_REF_KEY:
1488 if (unlikely(!IS_ALIGNED(inline_offset,
1489 fs_info->sectorsize))) {
1490 extent_err(leaf, slot,
1491 "invalid tree parent bytenr, have %llu expect aligned to %u",
1492 inline_offset, fs_info->sectorsize);
1493 return -EUCLEAN;
1494 }
1495 inline_refs++;
1496 break;
1497 /*
1498 * Contains owner subvolid, owner key objectid, adjusted offset.
1499 * The only obvious corruption can happen in that offset.
1500 */
1501 case BTRFS_EXTENT_DATA_REF_KEY:
1502 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1503 dref_root = btrfs_extent_data_ref_root(leaf, dref);
1504 dref_objectid = btrfs_extent_data_ref_objectid(leaf, dref);
1505 dref_offset = btrfs_extent_data_ref_offset(leaf, dref);
1506 seq = hash_extent_data_ref(
1507 btrfs_extent_data_ref_root(leaf, dref),
1508 btrfs_extent_data_ref_objectid(leaf, dref),
1509 btrfs_extent_data_ref_offset(leaf, dref));
1510 if (unlikely(!is_valid_dref_root(dref_root))) {
1511 extent_err(leaf, slot,
1512 "invalid data ref root value %llu",
1513 dref_root);
1514 return -EUCLEAN;
1515 }
1516 if (unlikely(dref_objectid < BTRFS_FIRST_FREE_OBJECTID ||
1517 dref_objectid > BTRFS_LAST_FREE_OBJECTID)) {
1518 extent_err(leaf, slot,
1519 "invalid data ref objectid value %llu",
1520 dref_objectid);
1521 return -EUCLEAN;
1522 }
1523 if (unlikely(!IS_ALIGNED(dref_offset,
1524 fs_info->sectorsize))) {
1525 extent_err(leaf, slot,
1526 "invalid data ref offset, have %llu expect aligned to %u",
1527 dref_offset, fs_info->sectorsize);
1528 return -EUCLEAN;
1529 }
1530 if (unlikely(btrfs_extent_data_ref_count(leaf, dref) == 0)) {
1531 extent_err(leaf, slot,
1532 "invalid data ref count, should have non-zero value");
1533 return -EUCLEAN;
1534 }
1535 inline_refs += btrfs_extent_data_ref_count(leaf, dref);
1536 break;
1537 /* Contains parent bytenr and ref count */
1538 case BTRFS_SHARED_DATA_REF_KEY:
1539 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1540 if (unlikely(!IS_ALIGNED(inline_offset,
1541 fs_info->sectorsize))) {
1542 extent_err(leaf, slot,
1543 "invalid data parent bytenr, have %llu expect aligned to %u",
1544 inline_offset, fs_info->sectorsize);
1545 return -EUCLEAN;
1546 }
1547 if (unlikely(btrfs_shared_data_ref_count(leaf, sref) == 0)) {
1548 extent_err(leaf, slot,
1549 "invalid shared data ref count, should have non-zero value");
1550 return -EUCLEAN;
1551 }
1552 inline_refs += btrfs_shared_data_ref_count(leaf, sref);
1553 break;
1554 case BTRFS_EXTENT_OWNER_REF_KEY:
1555 WARN_ON(!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));
1556 break;
1557 default:
1558 extent_err(leaf, slot, "unknown inline ref type: %u",
1559 inline_type);
1560 return -EUCLEAN;
1561 }
1562 if (inline_type < last_type) {
1563 extent_err(leaf, slot,
1564 "inline ref out-of-order: has type %u, prev type %u",
1565 inline_type, last_type);
1566 return -EUCLEAN;
1567 }
1568 /* Type changed, allow the sequence starts from U64_MAX again. */
1569 if (inline_type > last_type)
1570 last_seq = U64_MAX;
1571 if (seq > last_seq) {
1572 extent_err(leaf, slot,
1573 "inline ref out-of-order: has type %u offset %llu seq 0x%llx, prev type %u seq 0x%llx",
1574 inline_type, inline_offset, seq,
1575 last_type, last_seq);
1576 return -EUCLEAN;
1577 }
1578 last_type = inline_type;
1579 last_seq = seq;
1580 ptr += btrfs_extent_inline_ref_size(inline_type);
1581 }
1582 /* No padding is allowed */
1583 if (unlikely(ptr != end)) {
1584 extent_err(leaf, slot,
1585 "invalid extent item size, padding bytes found");
1586 return -EUCLEAN;
1587 }
1588
1589 /* Finally, check the inline refs against total refs */
1590 if (unlikely(inline_refs > total_refs)) {
1591 extent_err(leaf, slot,
1592 "invalid extent refs, have %llu expect >= inline %llu",
1593 total_refs, inline_refs);
1594 return -EUCLEAN;
1595 }
1596
1597 if ((prev_key->type == BTRFS_EXTENT_ITEM_KEY) ||
1598 (prev_key->type == BTRFS_METADATA_ITEM_KEY)) {
1599 u64 prev_end = prev_key->objectid;
1600
1601 if (prev_key->type == BTRFS_METADATA_ITEM_KEY)
1602 prev_end += fs_info->nodesize;
1603 else
1604 prev_end += prev_key->offset;
1605
1606 if (unlikely(prev_end > key->objectid)) {
1607 extent_err(leaf, slot,
1608 "previous extent [%llu %u %llu] overlaps current extent [%llu %u %llu]",
1609 prev_key->objectid, prev_key->type,
1610 prev_key->offset, key->objectid, key->type,
1611 key->offset);
1612 return -EUCLEAN;
1613 }
1614 }
1615
1616 return 0;
1617 }
1618
1619 static int check_simple_keyed_refs(struct extent_buffer *leaf,
1620 struct btrfs_key *key, int slot)
1621 {
1622 u32 expect_item_size = 0;
1623
1624 if (key->type == BTRFS_SHARED_DATA_REF_KEY) {
1625 struct btrfs_shared_data_ref *sref;
1626
1627 sref = btrfs_item_ptr(leaf, slot, struct btrfs_shared_data_ref);
1628 if (unlikely(btrfs_shared_data_ref_count(leaf, sref) == 0)) {
1629 extent_err(leaf, slot,
1630 "invalid shared data backref count, should have non-zero value");
1631 return -EUCLEAN;
1632 }
1633
1634 expect_item_size = sizeof(struct btrfs_shared_data_ref);
1635 }
1636
1637 if (unlikely(btrfs_item_size(leaf, slot) != expect_item_size)) {
1638 generic_err(leaf, slot,
1639 "invalid item size, have %u expect %u for key type %u",
1640 btrfs_item_size(leaf, slot),
1641 expect_item_size, key->type);
1642 return -EUCLEAN;
1643 }
1644 if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1645 generic_err(leaf, slot,
1646 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1647 key->objectid, leaf->fs_info->sectorsize);
1648 return -EUCLEAN;
1649 }
1650 if (unlikely(key->type != BTRFS_TREE_BLOCK_REF_KEY &&
1651 !IS_ALIGNED(key->offset, leaf->fs_info->sectorsize))) {
1652 extent_err(leaf, slot,
1653 "invalid tree parent bytenr, have %llu expect aligned to %u",
1654 key->offset, leaf->fs_info->sectorsize);
1655 return -EUCLEAN;
1656 }
1657 return 0;
1658 }
1659
1660 static int check_extent_data_ref(struct extent_buffer *leaf,
1661 struct btrfs_key *key, int slot)
1662 {
1663 struct btrfs_extent_data_ref *dref;
1664 unsigned long ptr = btrfs_item_ptr_offset(leaf, slot);
1665 const unsigned long end = ptr + btrfs_item_size(leaf, slot);
1666
1667 if (unlikely(btrfs_item_size(leaf, slot) % sizeof(*dref) != 0)) {
1668 generic_err(leaf, slot,
1669 "invalid item size, have %u expect aligned to %zu for key type %u",
1670 btrfs_item_size(leaf, slot),
1671 sizeof(*dref), key->type);
1672 return -EUCLEAN;
1673 }
1674 if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1675 generic_err(leaf, slot,
1676 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1677 key->objectid, leaf->fs_info->sectorsize);
1678 return -EUCLEAN;
1679 }
1680 for (; ptr < end; ptr += sizeof(*dref)) {
1681 u64 root;
1682 u64 objectid;
1683 u64 offset;
1684
1685 /*
1686 * We cannot check the extent_data_ref hash due to possible
1687 * overflow from the leaf due to hash collisions.
1688 */
1689 dref = (struct btrfs_extent_data_ref *)ptr;
1690 root = btrfs_extent_data_ref_root(leaf, dref);
1691 objectid = btrfs_extent_data_ref_objectid(leaf, dref);
1692 offset = btrfs_extent_data_ref_offset(leaf, dref);
1693 if (unlikely(!is_valid_dref_root(root))) {
1694 extent_err(leaf, slot,
1695 "invalid extent data backref root value %llu",
1696 root);
1697 return -EUCLEAN;
1698 }
1699 if (unlikely(objectid < BTRFS_FIRST_FREE_OBJECTID ||
1700 objectid > BTRFS_LAST_FREE_OBJECTID)) {
1701 extent_err(leaf, slot,
1702 "invalid extent data backref objectid value %llu",
1703 root);
1704 return -EUCLEAN;
1705 }
1706 if (unlikely(!IS_ALIGNED(offset, leaf->fs_info->sectorsize))) {
1707 extent_err(leaf, slot,
1708 "invalid extent data backref offset, have %llu expect aligned to %u",
1709 offset, leaf->fs_info->sectorsize);
1710 return -EUCLEAN;
1711 }
1712 if (unlikely(btrfs_extent_data_ref_count(leaf, dref) == 0)) {
1713 extent_err(leaf, slot,
1714 "invalid extent data backref count, should have non-zero value");
1715 return -EUCLEAN;
1716 }
1717 }
1718 return 0;
1719 }
1720
1721 #define inode_ref_err(eb, slot, fmt, args...) \
1722 inode_item_err(eb, slot, fmt, ##args)
1723 static int check_inode_ref(struct extent_buffer *leaf,
1724 struct btrfs_key *key, struct btrfs_key *prev_key,
1725 int slot)
1726 {
1727 struct btrfs_inode_ref *iref;
1728 unsigned long ptr;
1729 unsigned long end;
1730
1731 if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
1732 return -EUCLEAN;
1733 /* namelen can't be 0, so item_size == sizeof() is also invalid */
1734 if (unlikely(btrfs_item_size(leaf, slot) <= sizeof(*iref))) {
1735 inode_ref_err(leaf, slot,
1736 "invalid item size, have %u expect (%zu, %u)",
1737 btrfs_item_size(leaf, slot),
1738 sizeof(*iref), BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
1739 return -EUCLEAN;
1740 }
1741
1742 ptr = btrfs_item_ptr_offset(leaf, slot);
1743 end = ptr + btrfs_item_size(leaf, slot);
1744 while (ptr < end) {
1745 u16 namelen;
1746
1747 if (unlikely(ptr + sizeof(iref) > end)) {
1748 inode_ref_err(leaf, slot,
1749 "inode ref overflow, ptr %lu end %lu inode_ref_size %zu",
1750 ptr, end, sizeof(iref));
1751 return -EUCLEAN;
1752 }
1753
1754 iref = (struct btrfs_inode_ref *)ptr;
1755 namelen = btrfs_inode_ref_name_len(leaf, iref);
1756 if (unlikely(ptr + sizeof(*iref) + namelen > end)) {
1757 inode_ref_err(leaf, slot,
1758 "inode ref overflow, ptr %lu end %lu namelen %u",
1759 ptr, end, namelen);
1760 return -EUCLEAN;
1761 }
1762
1763 /*
1764 * NOTE: In theory we should record all found index numbers
1765 * to find any duplicated indexes, but that will be too time
1766 * consuming for inodes with too many hard links.
1767 */
1768 ptr += sizeof(*iref) + namelen;
1769 }
1770 return 0;
1771 }
1772
1773 static int check_raid_stripe_extent(const struct extent_buffer *leaf,
1774 const struct btrfs_key *key, int slot)
1775 {
1776 if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1777 generic_err(leaf, slot,
1778 "invalid key objectid for raid stripe extent, have %llu expect aligned to %u",
1779 key->objectid, leaf->fs_info->sectorsize);
1780 return -EUCLEAN;
1781 }
1782
1783 if (unlikely(!btrfs_fs_incompat(leaf->fs_info, RAID_STRIPE_TREE))) {
1784 generic_err(leaf, slot,
1785 "RAID_STRIPE_EXTENT present but RAID_STRIPE_TREE incompat bit unset");
1786 return -EUCLEAN;
1787 }
1788
1789 return 0;
1790 }
1791
1792 static int check_dev_extent_item(const struct extent_buffer *leaf,
1793 const struct btrfs_key *key,
1794 int slot,
1795 struct btrfs_key *prev_key)
1796 {
1797 struct btrfs_dev_extent *de;
1798 const u32 sectorsize = leaf->fs_info->sectorsize;
1799
1800 de = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
1801 /* Basic fixed member checks. */
1802 if (unlikely(btrfs_dev_extent_chunk_tree(leaf, de) !=
1803 BTRFS_CHUNK_TREE_OBJECTID)) {
1804 generic_err(leaf, slot,
1805 "invalid dev extent chunk tree id, has %llu expect %llu",
1806 btrfs_dev_extent_chunk_tree(leaf, de),
1807 BTRFS_CHUNK_TREE_OBJECTID);
1808 return -EUCLEAN;
1809 }
1810 if (unlikely(btrfs_dev_extent_chunk_objectid(leaf, de) !=
1811 BTRFS_FIRST_CHUNK_TREE_OBJECTID)) {
1812 generic_err(leaf, slot,
1813 "invalid dev extent chunk objectid, has %llu expect %llu",
1814 btrfs_dev_extent_chunk_objectid(leaf, de),
1815 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1816 return -EUCLEAN;
1817 }
1818 /* Alignment check. */
1819 if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) {
1820 generic_err(leaf, slot,
1821 "invalid dev extent key.offset, has %llu not aligned to %u",
1822 key->offset, sectorsize);
1823 return -EUCLEAN;
1824 }
1825 if (unlikely(!IS_ALIGNED(btrfs_dev_extent_chunk_offset(leaf, de),
1826 sectorsize))) {
1827 generic_err(leaf, slot,
1828 "invalid dev extent chunk offset, has %llu not aligned to %u",
1829 btrfs_dev_extent_chunk_objectid(leaf, de),
1830 sectorsize);
1831 return -EUCLEAN;
1832 }
1833 if (unlikely(!IS_ALIGNED(btrfs_dev_extent_length(leaf, de),
1834 sectorsize))) {
1835 generic_err(leaf, slot,
1836 "invalid dev extent length, has %llu not aligned to %u",
1837 btrfs_dev_extent_length(leaf, de), sectorsize);
1838 return -EUCLEAN;
1839 }
1840 /* Overlap check with previous dev extent. */
1841 if (slot && prev_key->objectid == key->objectid &&
1842 prev_key->type == key->type) {
1843 struct btrfs_dev_extent *prev_de;
1844 u64 prev_len;
1845
1846 prev_de = btrfs_item_ptr(leaf, slot - 1, struct btrfs_dev_extent);
1847 prev_len = btrfs_dev_extent_length(leaf, prev_de);
1848 if (unlikely(prev_key->offset + prev_len > key->offset)) {
1849 generic_err(leaf, slot,
1850 "dev extent overlap, prev offset %llu len %llu current offset %llu",
1851 prev_key->objectid, prev_len, key->offset);
1852 return -EUCLEAN;
1853 }
1854 }
1855 return 0;
1856 }
1857
1858 /*
1859 * Common point to switch the item-specific validation.
1860 */
1861 static enum btrfs_tree_block_status check_leaf_item(struct extent_buffer *leaf,
1862 struct btrfs_key *key,
1863 int slot,
1864 struct btrfs_key *prev_key)
1865 {
1866 int ret = 0;
1867 struct btrfs_chunk *chunk;
1868
1869 switch (key->type) {
1870 case BTRFS_EXTENT_DATA_KEY:
1871 ret = check_extent_data_item(leaf, key, slot, prev_key);
1872 break;
1873 case BTRFS_EXTENT_CSUM_KEY:
1874 ret = check_csum_item(leaf, key, slot, prev_key);
1875 break;
1876 case BTRFS_DIR_ITEM_KEY:
1877 case BTRFS_DIR_INDEX_KEY:
1878 case BTRFS_XATTR_ITEM_KEY:
1879 ret = check_dir_item(leaf, key, prev_key, slot);
1880 break;
1881 case BTRFS_INODE_REF_KEY:
1882 ret = check_inode_ref(leaf, key, prev_key, slot);
1883 break;
1884 case BTRFS_BLOCK_GROUP_ITEM_KEY:
1885 ret = check_block_group_item(leaf, key, slot);
1886 break;
1887 case BTRFS_CHUNK_ITEM_KEY:
1888 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1889 ret = check_leaf_chunk_item(leaf, chunk, key, slot);
1890 break;
1891 case BTRFS_DEV_ITEM_KEY:
1892 ret = check_dev_item(leaf, key, slot);
1893 break;
1894 case BTRFS_DEV_EXTENT_KEY:
1895 ret = check_dev_extent_item(leaf, key, slot, prev_key);
1896 break;
1897 case BTRFS_INODE_ITEM_KEY:
1898 ret = check_inode_item(leaf, key, slot);
1899 break;
1900 case BTRFS_ROOT_ITEM_KEY:
1901 ret = check_root_item(leaf, key, slot);
1902 break;
1903 case BTRFS_EXTENT_ITEM_KEY:
1904 case BTRFS_METADATA_ITEM_KEY:
1905 ret = check_extent_item(leaf, key, slot, prev_key);
1906 break;
1907 case BTRFS_TREE_BLOCK_REF_KEY:
1908 case BTRFS_SHARED_DATA_REF_KEY:
1909 case BTRFS_SHARED_BLOCK_REF_KEY:
1910 ret = check_simple_keyed_refs(leaf, key, slot);
1911 break;
1912 case BTRFS_EXTENT_DATA_REF_KEY:
1913 ret = check_extent_data_ref(leaf, key, slot);
1914 break;
1915 case BTRFS_RAID_STRIPE_KEY:
1916 ret = check_raid_stripe_extent(leaf, key, slot);
1917 break;
1918 }
1919
1920 if (ret)
1921 return BTRFS_TREE_BLOCK_INVALID_ITEM;
1922 return BTRFS_TREE_BLOCK_CLEAN;
1923 }
1924
1925 enum btrfs_tree_block_status __btrfs_check_leaf(struct extent_buffer *leaf)
1926 {
1927 struct btrfs_fs_info *fs_info = leaf->fs_info;
1928 /* No valid key type is 0, so all key should be larger than this key */
1929 struct btrfs_key prev_key = {0, 0, 0};
1930 struct btrfs_key key;
1931 u32 nritems = btrfs_header_nritems(leaf);
1932 int slot;
1933
1934 if (unlikely(btrfs_header_level(leaf) != 0)) {
1935 generic_err(leaf, 0,
1936 "invalid level for leaf, have %d expect 0",
1937 btrfs_header_level(leaf));
1938 return BTRFS_TREE_BLOCK_INVALID_LEVEL;
1939 }
1940
1941 if (unlikely(!btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_WRITTEN))) {
1942 generic_err(leaf, 0, "invalid flag for leaf, WRITTEN not set");
1943 return BTRFS_TREE_BLOCK_WRITTEN_NOT_SET;
1944 }
1945
1946 /*
1947 * Extent buffers from a relocation tree have a owner field that
1948 * corresponds to the subvolume tree they are based on. So just from an
1949 * extent buffer alone we can not find out what is the id of the
1950 * corresponding subvolume tree, so we can not figure out if the extent
1951 * buffer corresponds to the root of the relocation tree or not. So
1952 * skip this check for relocation trees.
1953 */
1954 if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
1955 u64 owner = btrfs_header_owner(leaf);
1956
1957 /* These trees must never be empty */
1958 if (unlikely(owner == BTRFS_ROOT_TREE_OBJECTID ||
1959 owner == BTRFS_CHUNK_TREE_OBJECTID ||
1960 owner == BTRFS_DEV_TREE_OBJECTID ||
1961 owner == BTRFS_FS_TREE_OBJECTID ||
1962 owner == BTRFS_DATA_RELOC_TREE_OBJECTID)) {
1963 generic_err(leaf, 0,
1964 "invalid root, root %llu must never be empty",
1965 owner);
1966 return BTRFS_TREE_BLOCK_INVALID_NRITEMS;
1967 }
1968
1969 /* Unknown tree */
1970 if (unlikely(owner == 0)) {
1971 generic_err(leaf, 0,
1972 "invalid owner, root 0 is not defined");
1973 return BTRFS_TREE_BLOCK_INVALID_OWNER;
1974 }
1975
1976 /* EXTENT_TREE_V2 can have empty extent trees. */
1977 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1978 return BTRFS_TREE_BLOCK_CLEAN;
1979
1980 if (unlikely(owner == BTRFS_EXTENT_TREE_OBJECTID)) {
1981 generic_err(leaf, 0,
1982 "invalid root, root %llu must never be empty",
1983 owner);
1984 return BTRFS_TREE_BLOCK_INVALID_NRITEMS;
1985 }
1986
1987 return BTRFS_TREE_BLOCK_CLEAN;
1988 }
1989
1990 if (unlikely(nritems == 0))
1991 return BTRFS_TREE_BLOCK_CLEAN;
1992
1993 /*
1994 * Check the following things to make sure this is a good leaf, and
1995 * leaf users won't need to bother with similar sanity checks:
1996 *
1997 * 1) key ordering
1998 * 2) item offset and size
1999 * No overlap, no hole, all inside the leaf.
2000 * 3) item content
2001 * If possible, do comprehensive sanity check.
2002 * NOTE: All checks must only rely on the item data itself.
2003 */
2004 for (slot = 0; slot < nritems; slot++) {
2005 u32 item_end_expected;
2006 u64 item_data_end;
2007 enum btrfs_tree_block_status ret;
2008
2009 btrfs_item_key_to_cpu(leaf, &key, slot);
2010
2011 /* Make sure the keys are in the right order */
2012 if (unlikely(btrfs_comp_cpu_keys(&prev_key, &key) >= 0)) {
2013 generic_err(leaf, slot,
2014 "bad key order, prev (%llu %u %llu) current (%llu %u %llu)",
2015 prev_key.objectid, prev_key.type,
2016 prev_key.offset, key.objectid, key.type,
2017 key.offset);
2018 return BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
2019 }
2020
2021 item_data_end = (u64)btrfs_item_offset(leaf, slot) +
2022 btrfs_item_size(leaf, slot);
2023 /*
2024 * Make sure the offset and ends are right, remember that the
2025 * item data starts at the end of the leaf and grows towards the
2026 * front.
2027 */
2028 if (slot == 0)
2029 item_end_expected = BTRFS_LEAF_DATA_SIZE(fs_info);
2030 else
2031 item_end_expected = btrfs_item_offset(leaf,
2032 slot - 1);
2033 if (unlikely(item_data_end != item_end_expected)) {
2034 generic_err(leaf, slot,
2035 "unexpected item end, have %llu expect %u",
2036 item_data_end, item_end_expected);
2037 return BTRFS_TREE_BLOCK_INVALID_OFFSETS;
2038 }
2039
2040 /*
2041 * Check to make sure that we don't point outside of the leaf,
2042 * just in case all the items are consistent to each other, but
2043 * all point outside of the leaf.
2044 */
2045 if (unlikely(item_data_end > BTRFS_LEAF_DATA_SIZE(fs_info))) {
2046 generic_err(leaf, slot,
2047 "slot end outside of leaf, have %llu expect range [0, %u]",
2048 item_data_end, BTRFS_LEAF_DATA_SIZE(fs_info));
2049 return BTRFS_TREE_BLOCK_INVALID_OFFSETS;
2050 }
2051
2052 /* Also check if the item pointer overlaps with btrfs item. */
2053 if (unlikely(btrfs_item_ptr_offset(leaf, slot) <
2054 btrfs_item_nr_offset(leaf, slot) + sizeof(struct btrfs_item))) {
2055 generic_err(leaf, slot,
2056 "slot overlaps with its data, item end %lu data start %lu",
2057 btrfs_item_nr_offset(leaf, slot) +
2058 sizeof(struct btrfs_item),
2059 btrfs_item_ptr_offset(leaf, slot));
2060 return BTRFS_TREE_BLOCK_INVALID_OFFSETS;
2061 }
2062
2063 /* Check if the item size and content meet other criteria. */
2064 ret = check_leaf_item(leaf, &key, slot, &prev_key);
2065 if (unlikely(ret != BTRFS_TREE_BLOCK_CLEAN))
2066 return ret;
2067
2068 prev_key.objectid = key.objectid;
2069 prev_key.type = key.type;
2070 prev_key.offset = key.offset;
2071 }
2072
2073 return BTRFS_TREE_BLOCK_CLEAN;
2074 }
2075
2076 int btrfs_check_leaf(struct extent_buffer *leaf)
2077 {
2078 enum btrfs_tree_block_status ret;
2079
2080 ret = __btrfs_check_leaf(leaf);
2081 if (unlikely(ret != BTRFS_TREE_BLOCK_CLEAN))
2082 return -EUCLEAN;
2083 return 0;
2084 }
2085 ALLOW_ERROR_INJECTION(btrfs_check_leaf, ERRNO);
2086
2087 enum btrfs_tree_block_status __btrfs_check_node(struct extent_buffer *node)
2088 {
2089 struct btrfs_fs_info *fs_info = node->fs_info;
2090 unsigned long nr = btrfs_header_nritems(node);
2091 struct btrfs_key key, next_key;
2092 int slot;
2093 int level = btrfs_header_level(node);
2094 u64 bytenr;
2095
2096 if (unlikely(!btrfs_header_flag(node, BTRFS_HEADER_FLAG_WRITTEN))) {
2097 generic_err(node, 0, "invalid flag for node, WRITTEN not set");
2098 return BTRFS_TREE_BLOCK_WRITTEN_NOT_SET;
2099 }
2100
2101 if (unlikely(level <= 0 || level >= BTRFS_MAX_LEVEL)) {
2102 generic_err(node, 0,
2103 "invalid level for node, have %d expect [1, %d]",
2104 level, BTRFS_MAX_LEVEL - 1);
2105 return BTRFS_TREE_BLOCK_INVALID_LEVEL;
2106 }
2107 if (unlikely(nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(fs_info))) {
2108 btrfs_crit(fs_info,
2109 "corrupt node: root=%llu block=%llu, nritems too %s, have %lu expect range [1,%u]",
2110 btrfs_header_owner(node), node->start,
2111 nr == 0 ? "small" : "large", nr,
2112 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
2113 return BTRFS_TREE_BLOCK_INVALID_NRITEMS;
2114 }
2115
2116 for (slot = 0; slot < nr - 1; slot++) {
2117 bytenr = btrfs_node_blockptr(node, slot);
2118 btrfs_node_key_to_cpu(node, &key, slot);
2119 btrfs_node_key_to_cpu(node, &next_key, slot + 1);
2120
2121 if (unlikely(!bytenr)) {
2122 generic_err(node, slot,
2123 "invalid NULL node pointer");
2124 return BTRFS_TREE_BLOCK_INVALID_BLOCKPTR;
2125 }
2126 if (unlikely(!IS_ALIGNED(bytenr, fs_info->sectorsize))) {
2127 generic_err(node, slot,
2128 "unaligned pointer, have %llu should be aligned to %u",
2129 bytenr, fs_info->sectorsize);
2130 return BTRFS_TREE_BLOCK_INVALID_BLOCKPTR;
2131 }
2132
2133 if (unlikely(btrfs_comp_cpu_keys(&key, &next_key) >= 0)) {
2134 generic_err(node, slot,
2135 "bad key order, current (%llu %u %llu) next (%llu %u %llu)",
2136 key.objectid, key.type, key.offset,
2137 next_key.objectid, next_key.type,
2138 next_key.offset);
2139 return BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
2140 }
2141 }
2142 return BTRFS_TREE_BLOCK_CLEAN;
2143 }
2144
2145 int btrfs_check_node(struct extent_buffer *node)
2146 {
2147 enum btrfs_tree_block_status ret;
2148
2149 ret = __btrfs_check_node(node);
2150 if (unlikely(ret != BTRFS_TREE_BLOCK_CLEAN))
2151 return -EUCLEAN;
2152 return 0;
2153 }
2154 ALLOW_ERROR_INJECTION(btrfs_check_node, ERRNO);
2155
2156 int btrfs_check_eb_owner(const struct extent_buffer *eb, u64 root_owner)
2157 {
2158 const bool is_subvol = is_fstree(root_owner);
2159 const u64 eb_owner = btrfs_header_owner(eb);
2160
2161 /*
2162 * Skip dummy fs, as selftests don't create unique ebs for each dummy
2163 * root.
2164 */
2165 if (btrfs_is_testing(eb->fs_info))
2166 return 0;
2167 /*
2168 * There are several call sites (backref walking, qgroup, and data
2169 * reloc) passing 0 as @root_owner, as they are not holding the
2170 * tree root. In that case, we can not do a reliable ownership check,
2171 * so just exit.
2172 */
2173 if (root_owner == 0)
2174 return 0;
2175 /*
2176 * These trees use key.offset as their owner, our callers don't have
2177 * the extra capacity to pass key.offset here. So we just skip them.
2178 */
2179 if (root_owner == BTRFS_TREE_LOG_OBJECTID ||
2180 root_owner == BTRFS_TREE_RELOC_OBJECTID)
2181 return 0;
2182
2183 if (!is_subvol) {
2184 /* For non-subvolume trees, the eb owner should match root owner */
2185 if (unlikely(root_owner != eb_owner)) {
2186 btrfs_crit(eb->fs_info,
2187 "corrupted %s, root=%llu block=%llu owner mismatch, have %llu expect %llu",
2188 btrfs_header_level(eb) == 0 ? "leaf" : "node",
2189 root_owner, btrfs_header_bytenr(eb), eb_owner,
2190 root_owner);
2191 return -EUCLEAN;
2192 }
2193 return 0;
2194 }
2195
2196 /*
2197 * For subvolume trees, owners can mismatch, but they should all belong
2198 * to subvolume trees.
2199 */
2200 if (unlikely(is_subvol != is_fstree(eb_owner))) {
2201 btrfs_crit(eb->fs_info,
2202 "corrupted %s, root=%llu block=%llu owner mismatch, have %llu expect [%llu, %llu]",
2203 btrfs_header_level(eb) == 0 ? "leaf" : "node",
2204 root_owner, btrfs_header_bytenr(eb), eb_owner,
2205 BTRFS_FIRST_FREE_OBJECTID, BTRFS_LAST_FREE_OBJECTID);
2206 return -EUCLEAN;
2207 }
2208 return 0;
2209 }
2210
2211 int btrfs_verify_level_key(struct extent_buffer *eb,
2212 const struct btrfs_tree_parent_check *check)
2213 {
2214 struct btrfs_fs_info *fs_info = eb->fs_info;
2215 int found_level;
2216 struct btrfs_key found_key;
2217 int ret;
2218
2219 found_level = btrfs_header_level(eb);
2220 if (found_level != check->level) {
2221 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
2222 KERN_ERR "BTRFS: tree level check failed\n");
2223 btrfs_err(fs_info,
2224 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
2225 eb->start, check->level, found_level);
2226 return -EIO;
2227 }
2228
2229 if (!check->has_first_key)
2230 return 0;
2231
2232 /*
2233 * For live tree block (new tree blocks in current transaction),
2234 * we need proper lock context to avoid race, which is impossible here.
2235 * So we only checks tree blocks which is read from disk, whose
2236 * generation <= fs_info->last_trans_committed.
2237 */
2238 if (btrfs_header_generation(eb) > btrfs_get_last_trans_committed(fs_info))
2239 return 0;
2240
2241 /* We have @first_key, so this @eb must have at least one item */
2242 if (btrfs_header_nritems(eb) == 0) {
2243 btrfs_err(fs_info,
2244 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
2245 eb->start);
2246 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
2247 return -EUCLEAN;
2248 }
2249
2250 if (found_level)
2251 btrfs_node_key_to_cpu(eb, &found_key, 0);
2252 else
2253 btrfs_item_key_to_cpu(eb, &found_key, 0);
2254 ret = btrfs_comp_cpu_keys(&check->first_key, &found_key);
2255
2256 if (ret) {
2257 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
2258 KERN_ERR "BTRFS: tree first key check failed\n");
2259 btrfs_err(fs_info,
2260 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
2261 eb->start, check->transid, check->first_key.objectid,
2262 check->first_key.type, check->first_key.offset,
2263 found_key.objectid, found_key.type,
2264 found_key.offset);
2265 }
2266 return ret;
2267 }
Linux Kernel