WIP FPC-III support
[linux/fpc-iii.git] / fs / btrfs / tree-checker.c
blob582061c7b54716e71563790beda87baa7fb26460
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) Qu Wenruo 2017. All rights reserved.
4 */
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.
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.
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.
18 #include <linux/types.h>
19 #include <linux/stddef.h>
20 #include <linux/error-injection.h>
21 #include "ctree.h"
22 #include "tree-checker.h"
23 #include "disk-io.h"
24 #include "compression.h"
25 #include "volumes.h"
26 #include "misc.h"
29 * Error message should follow the following format:
30 * corrupt <type>: <identifier>, <reason>[, <bad_value>]
32 * @type: leaf or node
33 * @identifier: the necessary info to locate the leaf/node.
34 * It's recommended to decode key.objecitd/offset if it's
35 * meaningful.
36 * @reason: describe the error
37 * @bad_value: optional, it's recommended to output bad value and its
38 * expected value (range).
40 * Since comma is used to separate the components, only space is allowed
41 * inside each component.
45 * Append generic "corrupt leaf/node root=%llu block=%llu slot=%d: " to @fmt.
46 * Allows callers to customize the output.
48 __printf(3, 4)
49 __cold
50 static void generic_err(const struct extent_buffer *eb, int slot,
51 const char *fmt, ...)
53 const struct btrfs_fs_info *fs_info = eb->fs_info;
54 struct va_format vaf;
55 va_list args;
57 va_start(args, fmt);
59 vaf.fmt = fmt;
60 vaf.va = &args;
62 btrfs_crit(fs_info,
63 "corrupt %s: root=%llu block=%llu slot=%d, %pV",
64 btrfs_header_level(eb) == 0 ? "leaf" : "node",
65 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot, &vaf);
66 va_end(args);
70 * Customized reporter for extent data item, since its key objectid and
71 * offset has its own meaning.
73 __printf(3, 4)
74 __cold
75 static void file_extent_err(const struct extent_buffer *eb, int slot,
76 const char *fmt, ...)
78 const struct btrfs_fs_info *fs_info = eb->fs_info;
79 struct btrfs_key key;
80 struct va_format vaf;
81 va_list args;
83 btrfs_item_key_to_cpu(eb, &key, slot);
84 va_start(args, fmt);
86 vaf.fmt = fmt;
87 vaf.va = &args;
89 btrfs_crit(fs_info,
90 "corrupt %s: root=%llu block=%llu slot=%d ino=%llu file_offset=%llu, %pV",
91 btrfs_header_level(eb) == 0 ? "leaf" : "node",
92 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
93 key.objectid, key.offset, &vaf);
94 va_end(args);
98 * Return 0 if the btrfs_file_extent_##name is aligned to @alignment
99 * Else return 1
101 #define CHECK_FE_ALIGNED(leaf, slot, fi, name, alignment) \
102 ({ \
103 if (unlikely(!IS_ALIGNED(btrfs_file_extent_##name((leaf), (fi)), \
104 (alignment)))) \
105 file_extent_err((leaf), (slot), \
106 "invalid %s for file extent, have %llu, should be aligned to %u", \
107 (#name), btrfs_file_extent_##name((leaf), (fi)), \
108 (alignment)); \
109 (!IS_ALIGNED(btrfs_file_extent_##name((leaf), (fi)), (alignment))); \
112 static u64 file_extent_end(struct extent_buffer *leaf,
113 struct btrfs_key *key,
114 struct btrfs_file_extent_item *extent)
116 u64 end;
117 u64 len;
119 if (btrfs_file_extent_type(leaf, extent) == BTRFS_FILE_EXTENT_INLINE) {
120 len = btrfs_file_extent_ram_bytes(leaf, extent);
121 end = ALIGN(key->offset + len, leaf->fs_info->sectorsize);
122 } else {
123 len = btrfs_file_extent_num_bytes(leaf, extent);
124 end = key->offset + len;
126 return end;
130 * Customized report for dir_item, the only new important information is
131 * key->objectid, which represents inode number
133 __printf(3, 4)
134 __cold
135 static void dir_item_err(const struct extent_buffer *eb, int slot,
136 const char *fmt, ...)
138 const struct btrfs_fs_info *fs_info = eb->fs_info;
139 struct btrfs_key key;
140 struct va_format vaf;
141 va_list args;
143 btrfs_item_key_to_cpu(eb, &key, slot);
144 va_start(args, fmt);
146 vaf.fmt = fmt;
147 vaf.va = &args;
149 btrfs_crit(fs_info,
150 "corrupt %s: root=%llu block=%llu slot=%d ino=%llu, %pV",
151 btrfs_header_level(eb) == 0 ? "leaf" : "node",
152 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
153 key.objectid, &vaf);
154 va_end(args);
158 * This functions checks prev_key->objectid, to ensure current key and prev_key
159 * share the same objectid as inode number.
161 * This is to detect missing INODE_ITEM in subvolume trees.
163 * Return true if everything is OK or we don't need to check.
164 * Return false if anything is wrong.
166 static bool check_prev_ino(struct extent_buffer *leaf,
167 struct btrfs_key *key, int slot,
168 struct btrfs_key *prev_key)
170 /* No prev key, skip check */
171 if (slot == 0)
172 return true;
174 /* Only these key->types needs to be checked */
175 ASSERT(key->type == BTRFS_XATTR_ITEM_KEY ||
176 key->type == BTRFS_INODE_REF_KEY ||
177 key->type == BTRFS_DIR_INDEX_KEY ||
178 key->type == BTRFS_DIR_ITEM_KEY ||
179 key->type == BTRFS_EXTENT_DATA_KEY);
182 * Only subvolume trees along with their reloc trees need this check.
183 * Things like log tree doesn't follow this ino requirement.
185 if (!is_fstree(btrfs_header_owner(leaf)))
186 return true;
188 if (key->objectid == prev_key->objectid)
189 return true;
191 /* Error found */
192 dir_item_err(leaf, slot,
193 "invalid previous key objectid, have %llu expect %llu",
194 prev_key->objectid, key->objectid);
195 return false;
197 static int check_extent_data_item(struct extent_buffer *leaf,
198 struct btrfs_key *key, int slot,
199 struct btrfs_key *prev_key)
201 struct btrfs_fs_info *fs_info = leaf->fs_info;
202 struct btrfs_file_extent_item *fi;
203 u32 sectorsize = fs_info->sectorsize;
204 u32 item_size = btrfs_item_size_nr(leaf, slot);
205 u64 extent_end;
207 if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) {
208 file_extent_err(leaf, slot,
209 "unaligned file_offset for file extent, have %llu should be aligned to %u",
210 key->offset, sectorsize);
211 return -EUCLEAN;
215 * Previous key must have the same key->objectid (ino).
216 * It can be XATTR_ITEM, INODE_ITEM or just another EXTENT_DATA.
217 * But if objectids mismatch, it means we have a missing
218 * INODE_ITEM.
220 if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
221 return -EUCLEAN;
223 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
226 * Make sure the item contains at least inline header, so the file
227 * extent type is not some garbage.
229 if (unlikely(item_size < BTRFS_FILE_EXTENT_INLINE_DATA_START)) {
230 file_extent_err(leaf, slot,
231 "invalid item size, have %u expect [%zu, %u)",
232 item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START,
233 SZ_4K);
234 return -EUCLEAN;
236 if (unlikely(btrfs_file_extent_type(leaf, fi) >=
237 BTRFS_NR_FILE_EXTENT_TYPES)) {
238 file_extent_err(leaf, slot,
239 "invalid type for file extent, have %u expect range [0, %u]",
240 btrfs_file_extent_type(leaf, fi),
241 BTRFS_NR_FILE_EXTENT_TYPES - 1);
242 return -EUCLEAN;
246 * Support for new compression/encryption must introduce incompat flag,
247 * and must be caught in open_ctree().
249 if (unlikely(btrfs_file_extent_compression(leaf, fi) >=
250 BTRFS_NR_COMPRESS_TYPES)) {
251 file_extent_err(leaf, slot,
252 "invalid compression for file extent, have %u expect range [0, %u]",
253 btrfs_file_extent_compression(leaf, fi),
254 BTRFS_NR_COMPRESS_TYPES - 1);
255 return -EUCLEAN;
257 if (unlikely(btrfs_file_extent_encryption(leaf, fi))) {
258 file_extent_err(leaf, slot,
259 "invalid encryption for file extent, have %u expect 0",
260 btrfs_file_extent_encryption(leaf, fi));
261 return -EUCLEAN;
263 if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
264 /* Inline extent must have 0 as key offset */
265 if (unlikely(key->offset)) {
266 file_extent_err(leaf, slot,
267 "invalid file_offset for inline file extent, have %llu expect 0",
268 key->offset);
269 return -EUCLEAN;
272 /* Compressed inline extent has no on-disk size, skip it */
273 if (btrfs_file_extent_compression(leaf, fi) !=
274 BTRFS_COMPRESS_NONE)
275 return 0;
277 /* Uncompressed inline extent size must match item size */
278 if (unlikely(item_size != BTRFS_FILE_EXTENT_INLINE_DATA_START +
279 btrfs_file_extent_ram_bytes(leaf, fi))) {
280 file_extent_err(leaf, slot,
281 "invalid ram_bytes for uncompressed inline extent, have %u expect %llu",
282 item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START +
283 btrfs_file_extent_ram_bytes(leaf, fi));
284 return -EUCLEAN;
286 return 0;
289 /* Regular or preallocated extent has fixed item size */
290 if (unlikely(item_size != sizeof(*fi))) {
291 file_extent_err(leaf, slot,
292 "invalid item size for reg/prealloc file extent, have %u expect %zu",
293 item_size, sizeof(*fi));
294 return -EUCLEAN;
296 if (unlikely(CHECK_FE_ALIGNED(leaf, slot, fi, ram_bytes, sectorsize) ||
297 CHECK_FE_ALIGNED(leaf, slot, fi, disk_bytenr, sectorsize) ||
298 CHECK_FE_ALIGNED(leaf, slot, fi, disk_num_bytes, sectorsize) ||
299 CHECK_FE_ALIGNED(leaf, slot, fi, offset, sectorsize) ||
300 CHECK_FE_ALIGNED(leaf, slot, fi, num_bytes, sectorsize)))
301 return -EUCLEAN;
303 /* Catch extent end overflow */
304 if (unlikely(check_add_overflow(btrfs_file_extent_num_bytes(leaf, fi),
305 key->offset, &extent_end))) {
306 file_extent_err(leaf, slot,
307 "extent end overflow, have file offset %llu extent num bytes %llu",
308 key->offset,
309 btrfs_file_extent_num_bytes(leaf, fi));
310 return -EUCLEAN;
314 * Check that no two consecutive file extent items, in the same leaf,
315 * present ranges that overlap each other.
317 if (slot > 0 &&
318 prev_key->objectid == key->objectid &&
319 prev_key->type == BTRFS_EXTENT_DATA_KEY) {
320 struct btrfs_file_extent_item *prev_fi;
321 u64 prev_end;
323 prev_fi = btrfs_item_ptr(leaf, slot - 1,
324 struct btrfs_file_extent_item);
325 prev_end = file_extent_end(leaf, prev_key, prev_fi);
326 if (unlikely(prev_end > key->offset)) {
327 file_extent_err(leaf, slot - 1,
328 "file extent end range (%llu) goes beyond start offset (%llu) of the next file extent",
329 prev_end, key->offset);
330 return -EUCLEAN;
334 return 0;
337 static int check_csum_item(struct extent_buffer *leaf, struct btrfs_key *key,
338 int slot, struct btrfs_key *prev_key)
340 struct btrfs_fs_info *fs_info = leaf->fs_info;
341 u32 sectorsize = fs_info->sectorsize;
342 const u32 csumsize = fs_info->csum_size;
344 if (unlikely(key->objectid != BTRFS_EXTENT_CSUM_OBJECTID)) {
345 generic_err(leaf, slot,
346 "invalid key objectid for csum item, have %llu expect %llu",
347 key->objectid, BTRFS_EXTENT_CSUM_OBJECTID);
348 return -EUCLEAN;
350 if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) {
351 generic_err(leaf, slot,
352 "unaligned key offset for csum item, have %llu should be aligned to %u",
353 key->offset, sectorsize);
354 return -EUCLEAN;
356 if (unlikely(!IS_ALIGNED(btrfs_item_size_nr(leaf, slot), csumsize))) {
357 generic_err(leaf, slot,
358 "unaligned item size for csum item, have %u should be aligned to %u",
359 btrfs_item_size_nr(leaf, slot), csumsize);
360 return -EUCLEAN;
362 if (slot > 0 && prev_key->type == BTRFS_EXTENT_CSUM_KEY) {
363 u64 prev_csum_end;
364 u32 prev_item_size;
366 prev_item_size = btrfs_item_size_nr(leaf, slot - 1);
367 prev_csum_end = (prev_item_size / csumsize) * sectorsize;
368 prev_csum_end += prev_key->offset;
369 if (unlikely(prev_csum_end > key->offset)) {
370 generic_err(leaf, slot - 1,
371 "csum end range (%llu) goes beyond the start range (%llu) of the next csum item",
372 prev_csum_end, key->offset);
373 return -EUCLEAN;
376 return 0;
379 /* Inode item error output has the same format as dir_item_err() */
380 #define inode_item_err(eb, slot, fmt, ...) \
381 dir_item_err(eb, slot, fmt, __VA_ARGS__)
383 static int check_inode_key(struct extent_buffer *leaf, struct btrfs_key *key,
384 int slot)
386 struct btrfs_key item_key;
387 bool is_inode_item;
389 btrfs_item_key_to_cpu(leaf, &item_key, slot);
390 is_inode_item = (item_key.type == BTRFS_INODE_ITEM_KEY);
392 /* For XATTR_ITEM, location key should be all 0 */
393 if (item_key.type == BTRFS_XATTR_ITEM_KEY) {
394 if (unlikely(key->objectid != 0 || key->type != 0 ||
395 key->offset != 0))
396 return -EUCLEAN;
397 return 0;
400 if (unlikely((key->objectid < BTRFS_FIRST_FREE_OBJECTID ||
401 key->objectid > BTRFS_LAST_FREE_OBJECTID) &&
402 key->objectid != BTRFS_ROOT_TREE_DIR_OBJECTID &&
403 key->objectid != BTRFS_FREE_INO_OBJECTID)) {
404 if (is_inode_item) {
405 generic_err(leaf, slot,
406 "invalid key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
407 key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
408 BTRFS_FIRST_FREE_OBJECTID,
409 BTRFS_LAST_FREE_OBJECTID,
410 BTRFS_FREE_INO_OBJECTID);
411 } else {
412 dir_item_err(leaf, slot,
413 "invalid location key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
414 key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
415 BTRFS_FIRST_FREE_OBJECTID,
416 BTRFS_LAST_FREE_OBJECTID,
417 BTRFS_FREE_INO_OBJECTID);
419 return -EUCLEAN;
421 if (unlikely(key->offset != 0)) {
422 if (is_inode_item)
423 inode_item_err(leaf, slot,
424 "invalid key offset: has %llu expect 0",
425 key->offset);
426 else
427 dir_item_err(leaf, slot,
428 "invalid location key offset:has %llu expect 0",
429 key->offset);
430 return -EUCLEAN;
432 return 0;
435 static int check_root_key(struct extent_buffer *leaf, struct btrfs_key *key,
436 int slot)
438 struct btrfs_key item_key;
439 bool is_root_item;
441 btrfs_item_key_to_cpu(leaf, &item_key, slot);
442 is_root_item = (item_key.type == BTRFS_ROOT_ITEM_KEY);
444 /* No such tree id */
445 if (unlikely(key->objectid == 0)) {
446 if (is_root_item)
447 generic_err(leaf, slot, "invalid root id 0");
448 else
449 dir_item_err(leaf, slot,
450 "invalid location key root id 0");
451 return -EUCLEAN;
454 /* DIR_ITEM/INDEX/INODE_REF is not allowed to point to non-fs trees */
455 if (unlikely(!is_fstree(key->objectid) && !is_root_item)) {
456 dir_item_err(leaf, slot,
457 "invalid location key objectid, have %llu expect [%llu, %llu]",
458 key->objectid, BTRFS_FIRST_FREE_OBJECTID,
459 BTRFS_LAST_FREE_OBJECTID);
460 return -EUCLEAN;
464 * ROOT_ITEM with non-zero offset means this is a snapshot, created at
465 * @offset transid.
466 * Furthermore, for location key in DIR_ITEM, its offset is always -1.
468 * So here we only check offset for reloc tree whose key->offset must
469 * be a valid tree.
471 if (unlikely(key->objectid == BTRFS_TREE_RELOC_OBJECTID &&
472 key->offset == 0)) {
473 generic_err(leaf, slot, "invalid root id 0 for reloc tree");
474 return -EUCLEAN;
476 return 0;
479 static int check_dir_item(struct extent_buffer *leaf,
480 struct btrfs_key *key, struct btrfs_key *prev_key,
481 int slot)
483 struct btrfs_fs_info *fs_info = leaf->fs_info;
484 struct btrfs_dir_item *di;
485 u32 item_size = btrfs_item_size_nr(leaf, slot);
486 u32 cur = 0;
488 if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
489 return -EUCLEAN;
491 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
492 while (cur < item_size) {
493 struct btrfs_key location_key;
494 u32 name_len;
495 u32 data_len;
496 u32 max_name_len;
497 u32 total_size;
498 u32 name_hash;
499 u8 dir_type;
500 int ret;
502 /* header itself should not cross item boundary */
503 if (unlikely(cur + sizeof(*di) > item_size)) {
504 dir_item_err(leaf, slot,
505 "dir item header crosses item boundary, have %zu boundary %u",
506 cur + sizeof(*di), item_size);
507 return -EUCLEAN;
510 /* Location key check */
511 btrfs_dir_item_key_to_cpu(leaf, di, &location_key);
512 if (location_key.type == BTRFS_ROOT_ITEM_KEY) {
513 ret = check_root_key(leaf, &location_key, slot);
514 if (unlikely(ret < 0))
515 return ret;
516 } else if (location_key.type == BTRFS_INODE_ITEM_KEY ||
517 location_key.type == 0) {
518 ret = check_inode_key(leaf, &location_key, slot);
519 if (unlikely(ret < 0))
520 return ret;
521 } else {
522 dir_item_err(leaf, slot,
523 "invalid location key type, have %u, expect %u or %u",
524 location_key.type, BTRFS_ROOT_ITEM_KEY,
525 BTRFS_INODE_ITEM_KEY);
526 return -EUCLEAN;
529 /* dir type check */
530 dir_type = btrfs_dir_type(leaf, di);
531 if (unlikely(dir_type >= BTRFS_FT_MAX)) {
532 dir_item_err(leaf, slot,
533 "invalid dir item type, have %u expect [0, %u)",
534 dir_type, BTRFS_FT_MAX);
535 return -EUCLEAN;
538 if (unlikely(key->type == BTRFS_XATTR_ITEM_KEY &&
539 dir_type != BTRFS_FT_XATTR)) {
540 dir_item_err(leaf, slot,
541 "invalid dir item type for XATTR key, have %u expect %u",
542 dir_type, BTRFS_FT_XATTR);
543 return -EUCLEAN;
545 if (unlikely(dir_type == BTRFS_FT_XATTR &&
546 key->type != BTRFS_XATTR_ITEM_KEY)) {
547 dir_item_err(leaf, slot,
548 "xattr dir type found for non-XATTR key");
549 return -EUCLEAN;
551 if (dir_type == BTRFS_FT_XATTR)
552 max_name_len = XATTR_NAME_MAX;
553 else
554 max_name_len = BTRFS_NAME_LEN;
556 /* Name/data length check */
557 name_len = btrfs_dir_name_len(leaf, di);
558 data_len = btrfs_dir_data_len(leaf, di);
559 if (unlikely(name_len > max_name_len)) {
560 dir_item_err(leaf, slot,
561 "dir item name len too long, have %u max %u",
562 name_len, max_name_len);
563 return -EUCLEAN;
565 if (unlikely(name_len + data_len > BTRFS_MAX_XATTR_SIZE(fs_info))) {
566 dir_item_err(leaf, slot,
567 "dir item name and data len too long, have %u max %u",
568 name_len + data_len,
569 BTRFS_MAX_XATTR_SIZE(fs_info));
570 return -EUCLEAN;
573 if (unlikely(data_len && dir_type != BTRFS_FT_XATTR)) {
574 dir_item_err(leaf, slot,
575 "dir item with invalid data len, have %u expect 0",
576 data_len);
577 return -EUCLEAN;
580 total_size = sizeof(*di) + name_len + data_len;
582 /* header and name/data should not cross item boundary */
583 if (unlikely(cur + total_size > item_size)) {
584 dir_item_err(leaf, slot,
585 "dir item data crosses item boundary, have %u boundary %u",
586 cur + total_size, item_size);
587 return -EUCLEAN;
591 * Special check for XATTR/DIR_ITEM, as key->offset is name
592 * hash, should match its name
594 if (key->type == BTRFS_DIR_ITEM_KEY ||
595 key->type == BTRFS_XATTR_ITEM_KEY) {
596 char namebuf[max(BTRFS_NAME_LEN, XATTR_NAME_MAX)];
598 read_extent_buffer(leaf, namebuf,
599 (unsigned long)(di + 1), name_len);
600 name_hash = btrfs_name_hash(namebuf, name_len);
601 if (unlikely(key->offset != name_hash)) {
602 dir_item_err(leaf, slot,
603 "name hash mismatch with key, have 0x%016x expect 0x%016llx",
604 name_hash, key->offset);
605 return -EUCLEAN;
608 cur += total_size;
609 di = (struct btrfs_dir_item *)((void *)di + total_size);
611 return 0;
614 __printf(3, 4)
615 __cold
616 static void block_group_err(const struct extent_buffer *eb, int slot,
617 const char *fmt, ...)
619 const struct btrfs_fs_info *fs_info = eb->fs_info;
620 struct btrfs_key key;
621 struct va_format vaf;
622 va_list args;
624 btrfs_item_key_to_cpu(eb, &key, slot);
625 va_start(args, fmt);
627 vaf.fmt = fmt;
628 vaf.va = &args;
630 btrfs_crit(fs_info,
631 "corrupt %s: root=%llu block=%llu slot=%d bg_start=%llu bg_len=%llu, %pV",
632 btrfs_header_level(eb) == 0 ? "leaf" : "node",
633 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
634 key.objectid, key.offset, &vaf);
635 va_end(args);
638 static int check_block_group_item(struct extent_buffer *leaf,
639 struct btrfs_key *key, int slot)
641 struct btrfs_block_group_item bgi;
642 u32 item_size = btrfs_item_size_nr(leaf, slot);
643 u64 flags;
644 u64 type;
647 * Here we don't really care about alignment since extent allocator can
648 * handle it. We care more about the size.
650 if (unlikely(key->offset == 0)) {
651 block_group_err(leaf, slot,
652 "invalid block group size 0");
653 return -EUCLEAN;
656 if (unlikely(item_size != sizeof(bgi))) {
657 block_group_err(leaf, slot,
658 "invalid item size, have %u expect %zu",
659 item_size, sizeof(bgi));
660 return -EUCLEAN;
663 read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot),
664 sizeof(bgi));
665 if (unlikely(btrfs_stack_block_group_chunk_objectid(&bgi) !=
666 BTRFS_FIRST_CHUNK_TREE_OBJECTID)) {
667 block_group_err(leaf, slot,
668 "invalid block group chunk objectid, have %llu expect %llu",
669 btrfs_stack_block_group_chunk_objectid(&bgi),
670 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
671 return -EUCLEAN;
674 if (unlikely(btrfs_stack_block_group_used(&bgi) > key->offset)) {
675 block_group_err(leaf, slot,
676 "invalid block group used, have %llu expect [0, %llu)",
677 btrfs_stack_block_group_used(&bgi), key->offset);
678 return -EUCLEAN;
681 flags = btrfs_stack_block_group_flags(&bgi);
682 if (unlikely(hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) > 1)) {
683 block_group_err(leaf, slot,
684 "invalid profile flags, have 0x%llx (%lu bits set) expect no more than 1 bit set",
685 flags & BTRFS_BLOCK_GROUP_PROFILE_MASK,
686 hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK));
687 return -EUCLEAN;
690 type = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
691 if (unlikely(type != BTRFS_BLOCK_GROUP_DATA &&
692 type != BTRFS_BLOCK_GROUP_METADATA &&
693 type != BTRFS_BLOCK_GROUP_SYSTEM &&
694 type != (BTRFS_BLOCK_GROUP_METADATA |
695 BTRFS_BLOCK_GROUP_DATA))) {
696 block_group_err(leaf, slot,
697 "invalid type, have 0x%llx (%lu bits set) expect either 0x%llx, 0x%llx, 0x%llx or 0x%llx",
698 type, hweight64(type),
699 BTRFS_BLOCK_GROUP_DATA, BTRFS_BLOCK_GROUP_METADATA,
700 BTRFS_BLOCK_GROUP_SYSTEM,
701 BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA);
702 return -EUCLEAN;
704 return 0;
707 __printf(4, 5)
708 __cold
709 static void chunk_err(const struct extent_buffer *leaf,
710 const struct btrfs_chunk *chunk, u64 logical,
711 const char *fmt, ...)
713 const struct btrfs_fs_info *fs_info = leaf->fs_info;
714 bool is_sb;
715 struct va_format vaf;
716 va_list args;
717 int i;
718 int slot = -1;
720 /* Only superblock eb is able to have such small offset */
721 is_sb = (leaf->start == BTRFS_SUPER_INFO_OFFSET);
723 if (!is_sb) {
725 * Get the slot number by iterating through all slots, this
726 * would provide better readability.
728 for (i = 0; i < btrfs_header_nritems(leaf); i++) {
729 if (btrfs_item_ptr_offset(leaf, i) ==
730 (unsigned long)chunk) {
731 slot = i;
732 break;
736 va_start(args, fmt);
737 vaf.fmt = fmt;
738 vaf.va = &args;
740 if (is_sb)
741 btrfs_crit(fs_info,
742 "corrupt superblock syschunk array: chunk_start=%llu, %pV",
743 logical, &vaf);
744 else
745 btrfs_crit(fs_info,
746 "corrupt leaf: root=%llu block=%llu slot=%d chunk_start=%llu, %pV",
747 BTRFS_CHUNK_TREE_OBJECTID, leaf->start, slot,
748 logical, &vaf);
749 va_end(args);
753 * The common chunk check which could also work on super block sys chunk array.
755 * Return -EUCLEAN if anything is corrupted.
756 * Return 0 if everything is OK.
758 int btrfs_check_chunk_valid(struct extent_buffer *leaf,
759 struct btrfs_chunk *chunk, u64 logical)
761 struct btrfs_fs_info *fs_info = leaf->fs_info;
762 u64 length;
763 u64 chunk_end;
764 u64 stripe_len;
765 u16 num_stripes;
766 u16 sub_stripes;
767 u64 type;
768 u64 features;
769 bool mixed = false;
770 int raid_index;
771 int nparity;
772 int ncopies;
774 length = btrfs_chunk_length(leaf, chunk);
775 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
776 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
777 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
778 type = btrfs_chunk_type(leaf, chunk);
779 raid_index = btrfs_bg_flags_to_raid_index(type);
780 ncopies = btrfs_raid_array[raid_index].ncopies;
781 nparity = btrfs_raid_array[raid_index].nparity;
783 if (unlikely(!num_stripes)) {
784 chunk_err(leaf, chunk, logical,
785 "invalid chunk num_stripes, have %u", num_stripes);
786 return -EUCLEAN;
788 if (unlikely(num_stripes < ncopies)) {
789 chunk_err(leaf, chunk, logical,
790 "invalid chunk num_stripes < ncopies, have %u < %d",
791 num_stripes, ncopies);
792 return -EUCLEAN;
794 if (unlikely(nparity && num_stripes == nparity)) {
795 chunk_err(leaf, chunk, logical,
796 "invalid chunk num_stripes == nparity, have %u == %d",
797 num_stripes, nparity);
798 return -EUCLEAN;
800 if (unlikely(!IS_ALIGNED(logical, fs_info->sectorsize))) {
801 chunk_err(leaf, chunk, logical,
802 "invalid chunk logical, have %llu should aligned to %u",
803 logical, fs_info->sectorsize);
804 return -EUCLEAN;
806 if (unlikely(btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize)) {
807 chunk_err(leaf, chunk, logical,
808 "invalid chunk sectorsize, have %u expect %u",
809 btrfs_chunk_sector_size(leaf, chunk),
810 fs_info->sectorsize);
811 return -EUCLEAN;
813 if (unlikely(!length || !IS_ALIGNED(length, fs_info->sectorsize))) {
814 chunk_err(leaf, chunk, logical,
815 "invalid chunk length, have %llu", length);
816 return -EUCLEAN;
818 if (unlikely(check_add_overflow(logical, length, &chunk_end))) {
819 chunk_err(leaf, chunk, logical,
820 "invalid chunk logical start and length, have logical start %llu length %llu",
821 logical, length);
822 return -EUCLEAN;
824 if (unlikely(!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN)) {
825 chunk_err(leaf, chunk, logical,
826 "invalid chunk stripe length: %llu",
827 stripe_len);
828 return -EUCLEAN;
830 if (unlikely(type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
831 BTRFS_BLOCK_GROUP_PROFILE_MASK))) {
832 chunk_err(leaf, chunk, logical,
833 "unrecognized chunk type: 0x%llx",
834 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
835 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
836 btrfs_chunk_type(leaf, chunk));
837 return -EUCLEAN;
840 if (unlikely(!has_single_bit_set(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
841 (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) != 0)) {
842 chunk_err(leaf, chunk, logical,
843 "invalid chunk profile flag: 0x%llx, expect 0 or 1 bit set",
844 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
845 return -EUCLEAN;
847 if (unlikely((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0)) {
848 chunk_err(leaf, chunk, logical,
849 "missing chunk type flag, have 0x%llx one bit must be set in 0x%llx",
850 type, BTRFS_BLOCK_GROUP_TYPE_MASK);
851 return -EUCLEAN;
854 if (unlikely((type & BTRFS_BLOCK_GROUP_SYSTEM) &&
855 (type & (BTRFS_BLOCK_GROUP_METADATA |
856 BTRFS_BLOCK_GROUP_DATA)))) {
857 chunk_err(leaf, chunk, logical,
858 "system chunk with data or metadata type: 0x%llx",
859 type);
860 return -EUCLEAN;
863 features = btrfs_super_incompat_flags(fs_info->super_copy);
864 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
865 mixed = true;
867 if (!mixed) {
868 if (unlikely((type & BTRFS_BLOCK_GROUP_METADATA) &&
869 (type & BTRFS_BLOCK_GROUP_DATA))) {
870 chunk_err(leaf, chunk, logical,
871 "mixed chunk type in non-mixed mode: 0x%llx", type);
872 return -EUCLEAN;
876 if (unlikely((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
877 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes != 2) ||
878 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
879 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
880 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes != 2) ||
881 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
882 num_stripes != 1))) {
883 chunk_err(leaf, chunk, logical,
884 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
885 num_stripes, sub_stripes,
886 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
887 return -EUCLEAN;
890 return 0;
894 * Enhanced version of chunk item checker.
896 * The common btrfs_check_chunk_valid() doesn't check item size since it needs
897 * to work on super block sys_chunk_array which doesn't have full item ptr.
899 static int check_leaf_chunk_item(struct extent_buffer *leaf,
900 struct btrfs_chunk *chunk,
901 struct btrfs_key *key, int slot)
903 int num_stripes;
905 if (unlikely(btrfs_item_size_nr(leaf, slot) < sizeof(struct btrfs_chunk))) {
906 chunk_err(leaf, chunk, key->offset,
907 "invalid chunk item size: have %u expect [%zu, %u)",
908 btrfs_item_size_nr(leaf, slot),
909 sizeof(struct btrfs_chunk),
910 BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
911 return -EUCLEAN;
914 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
915 /* Let btrfs_check_chunk_valid() handle this error type */
916 if (num_stripes == 0)
917 goto out;
919 if (unlikely(btrfs_chunk_item_size(num_stripes) !=
920 btrfs_item_size_nr(leaf, slot))) {
921 chunk_err(leaf, chunk, key->offset,
922 "invalid chunk item size: have %u expect %lu",
923 btrfs_item_size_nr(leaf, slot),
924 btrfs_chunk_item_size(num_stripes));
925 return -EUCLEAN;
927 out:
928 return btrfs_check_chunk_valid(leaf, chunk, key->offset);
931 __printf(3, 4)
932 __cold
933 static void dev_item_err(const struct extent_buffer *eb, int slot,
934 const char *fmt, ...)
936 struct btrfs_key key;
937 struct va_format vaf;
938 va_list args;
940 btrfs_item_key_to_cpu(eb, &key, slot);
941 va_start(args, fmt);
943 vaf.fmt = fmt;
944 vaf.va = &args;
946 btrfs_crit(eb->fs_info,
947 "corrupt %s: root=%llu block=%llu slot=%d devid=%llu %pV",
948 btrfs_header_level(eb) == 0 ? "leaf" : "node",
949 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
950 key.objectid, &vaf);
951 va_end(args);
954 static int check_dev_item(struct extent_buffer *leaf,
955 struct btrfs_key *key, int slot)
957 struct btrfs_dev_item *ditem;
959 if (unlikely(key->objectid != BTRFS_DEV_ITEMS_OBJECTID)) {
960 dev_item_err(leaf, slot,
961 "invalid objectid: has=%llu expect=%llu",
962 key->objectid, BTRFS_DEV_ITEMS_OBJECTID);
963 return -EUCLEAN;
965 ditem = btrfs_item_ptr(leaf, slot, struct btrfs_dev_item);
966 if (unlikely(btrfs_device_id(leaf, ditem) != key->offset)) {
967 dev_item_err(leaf, slot,
968 "devid mismatch: key has=%llu item has=%llu",
969 key->offset, btrfs_device_id(leaf, ditem));
970 return -EUCLEAN;
974 * For device total_bytes, we don't have reliable way to check it, as
975 * it can be 0 for device removal. Device size check can only be done
976 * by dev extents check.
978 if (unlikely(btrfs_device_bytes_used(leaf, ditem) >
979 btrfs_device_total_bytes(leaf, ditem))) {
980 dev_item_err(leaf, slot,
981 "invalid bytes used: have %llu expect [0, %llu]",
982 btrfs_device_bytes_used(leaf, ditem),
983 btrfs_device_total_bytes(leaf, ditem));
984 return -EUCLEAN;
987 * Remaining members like io_align/type/gen/dev_group aren't really
988 * utilized. Skip them to make later usage of them easier.
990 return 0;
993 static int check_inode_item(struct extent_buffer *leaf,
994 struct btrfs_key *key, int slot)
996 struct btrfs_fs_info *fs_info = leaf->fs_info;
997 struct btrfs_inode_item *iitem;
998 u64 super_gen = btrfs_super_generation(fs_info->super_copy);
999 u32 valid_mask = (S_IFMT | S_ISUID | S_ISGID | S_ISVTX | 0777);
1000 u32 mode;
1001 int ret;
1003 ret = check_inode_key(leaf, key, slot);
1004 if (unlikely(ret < 0))
1005 return ret;
1007 iitem = btrfs_item_ptr(leaf, slot, struct btrfs_inode_item);
1009 /* Here we use super block generation + 1 to handle log tree */
1010 if (unlikely(btrfs_inode_generation(leaf, iitem) > super_gen + 1)) {
1011 inode_item_err(leaf, slot,
1012 "invalid inode generation: has %llu expect (0, %llu]",
1013 btrfs_inode_generation(leaf, iitem),
1014 super_gen + 1);
1015 return -EUCLEAN;
1017 /* Note for ROOT_TREE_DIR_ITEM, mkfs could set its transid 0 */
1018 if (unlikely(btrfs_inode_transid(leaf, iitem) > super_gen + 1)) {
1019 inode_item_err(leaf, slot,
1020 "invalid inode transid: has %llu expect [0, %llu]",
1021 btrfs_inode_transid(leaf, iitem), super_gen + 1);
1022 return -EUCLEAN;
1026 * For size and nbytes it's better not to be too strict, as for dir
1027 * item its size/nbytes can easily get wrong, but doesn't affect
1028 * anything in the fs. So here we skip the check.
1030 mode = btrfs_inode_mode(leaf, iitem);
1031 if (unlikely(mode & ~valid_mask)) {
1032 inode_item_err(leaf, slot,
1033 "unknown mode bit detected: 0x%x",
1034 mode & ~valid_mask);
1035 return -EUCLEAN;
1039 * S_IFMT is not bit mapped so we can't completely rely on
1040 * is_power_of_2/has_single_bit_set, but it can save us from checking
1041 * FIFO/CHR/DIR/REG. Only needs to check BLK, LNK and SOCKS
1043 if (!has_single_bit_set(mode & S_IFMT)) {
1044 if (unlikely(!S_ISLNK(mode) && !S_ISBLK(mode) && !S_ISSOCK(mode))) {
1045 inode_item_err(leaf, slot,
1046 "invalid mode: has 0%o expect valid S_IF* bit(s)",
1047 mode & S_IFMT);
1048 return -EUCLEAN;
1051 if (unlikely(S_ISDIR(mode) && btrfs_inode_nlink(leaf, iitem) > 1)) {
1052 inode_item_err(leaf, slot,
1053 "invalid nlink: has %u expect no more than 1 for dir",
1054 btrfs_inode_nlink(leaf, iitem));
1055 return -EUCLEAN;
1057 if (unlikely(btrfs_inode_flags(leaf, iitem) & ~BTRFS_INODE_FLAG_MASK)) {
1058 inode_item_err(leaf, slot,
1059 "unknown flags detected: 0x%llx",
1060 btrfs_inode_flags(leaf, iitem) &
1061 ~BTRFS_INODE_FLAG_MASK);
1062 return -EUCLEAN;
1064 return 0;
1067 static int check_root_item(struct extent_buffer *leaf, struct btrfs_key *key,
1068 int slot)
1070 struct btrfs_fs_info *fs_info = leaf->fs_info;
1071 struct btrfs_root_item ri = { 0 };
1072 const u64 valid_root_flags = BTRFS_ROOT_SUBVOL_RDONLY |
1073 BTRFS_ROOT_SUBVOL_DEAD;
1074 int ret;
1076 ret = check_root_key(leaf, key, slot);
1077 if (unlikely(ret < 0))
1078 return ret;
1080 if (unlikely(btrfs_item_size_nr(leaf, slot) != sizeof(ri) &&
1081 btrfs_item_size_nr(leaf, slot) !=
1082 btrfs_legacy_root_item_size())) {
1083 generic_err(leaf, slot,
1084 "invalid root item size, have %u expect %zu or %u",
1085 btrfs_item_size_nr(leaf, slot), sizeof(ri),
1086 btrfs_legacy_root_item_size());
1087 return -EUCLEAN;
1091 * For legacy root item, the members starting at generation_v2 will be
1092 * all filled with 0.
1093 * And since we allow geneartion_v2 as 0, it will still pass the check.
1095 read_extent_buffer(leaf, &ri, btrfs_item_ptr_offset(leaf, slot),
1096 btrfs_item_size_nr(leaf, slot));
1098 /* Generation related */
1099 if (unlikely(btrfs_root_generation(&ri) >
1100 btrfs_super_generation(fs_info->super_copy) + 1)) {
1101 generic_err(leaf, slot,
1102 "invalid root generation, have %llu expect (0, %llu]",
1103 btrfs_root_generation(&ri),
1104 btrfs_super_generation(fs_info->super_copy) + 1);
1105 return -EUCLEAN;
1107 if (unlikely(btrfs_root_generation_v2(&ri) >
1108 btrfs_super_generation(fs_info->super_copy) + 1)) {
1109 generic_err(leaf, slot,
1110 "invalid root v2 generation, have %llu expect (0, %llu]",
1111 btrfs_root_generation_v2(&ri),
1112 btrfs_super_generation(fs_info->super_copy) + 1);
1113 return -EUCLEAN;
1115 if (unlikely(btrfs_root_last_snapshot(&ri) >
1116 btrfs_super_generation(fs_info->super_copy) + 1)) {
1117 generic_err(leaf, slot,
1118 "invalid root last_snapshot, have %llu expect (0, %llu]",
1119 btrfs_root_last_snapshot(&ri),
1120 btrfs_super_generation(fs_info->super_copy) + 1);
1121 return -EUCLEAN;
1124 /* Alignment and level check */
1125 if (unlikely(!IS_ALIGNED(btrfs_root_bytenr(&ri), fs_info->sectorsize))) {
1126 generic_err(leaf, slot,
1127 "invalid root bytenr, have %llu expect to be aligned to %u",
1128 btrfs_root_bytenr(&ri), fs_info->sectorsize);
1129 return -EUCLEAN;
1131 if (unlikely(btrfs_root_level(&ri) >= BTRFS_MAX_LEVEL)) {
1132 generic_err(leaf, slot,
1133 "invalid root level, have %u expect [0, %u]",
1134 btrfs_root_level(&ri), BTRFS_MAX_LEVEL - 1);
1135 return -EUCLEAN;
1137 if (unlikely(btrfs_root_drop_level(&ri) >= BTRFS_MAX_LEVEL)) {
1138 generic_err(leaf, slot,
1139 "invalid root level, have %u expect [0, %u]",
1140 btrfs_root_drop_level(&ri), BTRFS_MAX_LEVEL - 1);
1141 return -EUCLEAN;
1144 /* Flags check */
1145 if (unlikely(btrfs_root_flags(&ri) & ~valid_root_flags)) {
1146 generic_err(leaf, slot,
1147 "invalid root flags, have 0x%llx expect mask 0x%llx",
1148 btrfs_root_flags(&ri), valid_root_flags);
1149 return -EUCLEAN;
1151 return 0;
1154 __printf(3,4)
1155 __cold
1156 static void extent_err(const struct extent_buffer *eb, int slot,
1157 const char *fmt, ...)
1159 struct btrfs_key key;
1160 struct va_format vaf;
1161 va_list args;
1162 u64 bytenr;
1163 u64 len;
1165 btrfs_item_key_to_cpu(eb, &key, slot);
1166 bytenr = key.objectid;
1167 if (key.type == BTRFS_METADATA_ITEM_KEY ||
1168 key.type == BTRFS_TREE_BLOCK_REF_KEY ||
1169 key.type == BTRFS_SHARED_BLOCK_REF_KEY)
1170 len = eb->fs_info->nodesize;
1171 else
1172 len = key.offset;
1173 va_start(args, fmt);
1175 vaf.fmt = fmt;
1176 vaf.va = &args;
1178 btrfs_crit(eb->fs_info,
1179 "corrupt %s: block=%llu slot=%d extent bytenr=%llu len=%llu %pV",
1180 btrfs_header_level(eb) == 0 ? "leaf" : "node",
1181 eb->start, slot, bytenr, len, &vaf);
1182 va_end(args);
1185 static int check_extent_item(struct extent_buffer *leaf,
1186 struct btrfs_key *key, int slot)
1188 struct btrfs_fs_info *fs_info = leaf->fs_info;
1189 struct btrfs_extent_item *ei;
1190 bool is_tree_block = false;
1191 unsigned long ptr; /* Current pointer inside inline refs */
1192 unsigned long end; /* Extent item end */
1193 const u32 item_size = btrfs_item_size_nr(leaf, slot);
1194 u64 flags;
1195 u64 generation;
1196 u64 total_refs; /* Total refs in btrfs_extent_item */
1197 u64 inline_refs = 0; /* found total inline refs */
1199 if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1200 !btrfs_fs_incompat(fs_info, SKINNY_METADATA))) {
1201 generic_err(leaf, slot,
1202 "invalid key type, METADATA_ITEM type invalid when SKINNY_METADATA feature disabled");
1203 return -EUCLEAN;
1205 /* key->objectid is the bytenr for both key types */
1206 if (unlikely(!IS_ALIGNED(key->objectid, fs_info->sectorsize))) {
1207 generic_err(leaf, slot,
1208 "invalid key objectid, have %llu expect to be aligned to %u",
1209 key->objectid, fs_info->sectorsize);
1210 return -EUCLEAN;
1213 /* key->offset is tree level for METADATA_ITEM_KEY */
1214 if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1215 key->offset >= BTRFS_MAX_LEVEL)) {
1216 extent_err(leaf, slot,
1217 "invalid tree level, have %llu expect [0, %u]",
1218 key->offset, BTRFS_MAX_LEVEL - 1);
1219 return -EUCLEAN;
1223 * EXTENT/METADATA_ITEM consists of:
1224 * 1) One btrfs_extent_item
1225 * Records the total refs, type and generation of the extent.
1227 * 2) One btrfs_tree_block_info (for EXTENT_ITEM and tree backref only)
1228 * Records the first key and level of the tree block.
1230 * 2) Zero or more btrfs_extent_inline_ref(s)
1231 * Each inline ref has one btrfs_extent_inline_ref shows:
1232 * 2.1) The ref type, one of the 4
1233 * TREE_BLOCK_REF Tree block only
1234 * SHARED_BLOCK_REF Tree block only
1235 * EXTENT_DATA_REF Data only
1236 * SHARED_DATA_REF Data only
1237 * 2.2) Ref type specific data
1238 * Either using btrfs_extent_inline_ref::offset, or specific
1239 * data structure.
1241 if (unlikely(item_size < sizeof(*ei))) {
1242 extent_err(leaf, slot,
1243 "invalid item size, have %u expect [%zu, %u)",
1244 item_size, sizeof(*ei),
1245 BTRFS_LEAF_DATA_SIZE(fs_info));
1246 return -EUCLEAN;
1248 end = item_size + btrfs_item_ptr_offset(leaf, slot);
1250 /* Checks against extent_item */
1251 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
1252 flags = btrfs_extent_flags(leaf, ei);
1253 total_refs = btrfs_extent_refs(leaf, ei);
1254 generation = btrfs_extent_generation(leaf, ei);
1255 if (unlikely(generation >
1256 btrfs_super_generation(fs_info->super_copy) + 1)) {
1257 extent_err(leaf, slot,
1258 "invalid generation, have %llu expect (0, %llu]",
1259 generation,
1260 btrfs_super_generation(fs_info->super_copy) + 1);
1261 return -EUCLEAN;
1263 if (unlikely(!has_single_bit_set(flags & (BTRFS_EXTENT_FLAG_DATA |
1264 BTRFS_EXTENT_FLAG_TREE_BLOCK)))) {
1265 extent_err(leaf, slot,
1266 "invalid extent flag, have 0x%llx expect 1 bit set in 0x%llx",
1267 flags, BTRFS_EXTENT_FLAG_DATA |
1268 BTRFS_EXTENT_FLAG_TREE_BLOCK);
1269 return -EUCLEAN;
1271 is_tree_block = !!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK);
1272 if (is_tree_block) {
1273 if (unlikely(key->type == BTRFS_EXTENT_ITEM_KEY &&
1274 key->offset != fs_info->nodesize)) {
1275 extent_err(leaf, slot,
1276 "invalid extent length, have %llu expect %u",
1277 key->offset, fs_info->nodesize);
1278 return -EUCLEAN;
1280 } else {
1281 if (unlikely(key->type != BTRFS_EXTENT_ITEM_KEY)) {
1282 extent_err(leaf, slot,
1283 "invalid key type, have %u expect %u for data backref",
1284 key->type, BTRFS_EXTENT_ITEM_KEY);
1285 return -EUCLEAN;
1287 if (unlikely(!IS_ALIGNED(key->offset, fs_info->sectorsize))) {
1288 extent_err(leaf, slot,
1289 "invalid extent length, have %llu expect aligned to %u",
1290 key->offset, fs_info->sectorsize);
1291 return -EUCLEAN;
1294 ptr = (unsigned long)(struct btrfs_extent_item *)(ei + 1);
1296 /* Check the special case of btrfs_tree_block_info */
1297 if (is_tree_block && key->type != BTRFS_METADATA_ITEM_KEY) {
1298 struct btrfs_tree_block_info *info;
1300 info = (struct btrfs_tree_block_info *)ptr;
1301 if (unlikely(btrfs_tree_block_level(leaf, info) >= BTRFS_MAX_LEVEL)) {
1302 extent_err(leaf, slot,
1303 "invalid tree block info level, have %u expect [0, %u]",
1304 btrfs_tree_block_level(leaf, info),
1305 BTRFS_MAX_LEVEL - 1);
1306 return -EUCLEAN;
1308 ptr = (unsigned long)(struct btrfs_tree_block_info *)(info + 1);
1311 /* Check inline refs */
1312 while (ptr < end) {
1313 struct btrfs_extent_inline_ref *iref;
1314 struct btrfs_extent_data_ref *dref;
1315 struct btrfs_shared_data_ref *sref;
1316 u64 dref_offset;
1317 u64 inline_offset;
1318 u8 inline_type;
1320 if (unlikely(ptr + sizeof(*iref) > end)) {
1321 extent_err(leaf, slot,
1322 "inline ref item overflows extent item, ptr %lu iref size %zu end %lu",
1323 ptr, sizeof(*iref), end);
1324 return -EUCLEAN;
1326 iref = (struct btrfs_extent_inline_ref *)ptr;
1327 inline_type = btrfs_extent_inline_ref_type(leaf, iref);
1328 inline_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1329 if (unlikely(ptr + btrfs_extent_inline_ref_size(inline_type) > end)) {
1330 extent_err(leaf, slot,
1331 "inline ref item overflows extent item, ptr %lu iref size %u end %lu",
1332 ptr, inline_type, end);
1333 return -EUCLEAN;
1336 switch (inline_type) {
1337 /* inline_offset is subvolid of the owner, no need to check */
1338 case BTRFS_TREE_BLOCK_REF_KEY:
1339 inline_refs++;
1340 break;
1341 /* Contains parent bytenr */
1342 case BTRFS_SHARED_BLOCK_REF_KEY:
1343 if (unlikely(!IS_ALIGNED(inline_offset,
1344 fs_info->sectorsize))) {
1345 extent_err(leaf, slot,
1346 "invalid tree parent bytenr, have %llu expect aligned to %u",
1347 inline_offset, fs_info->sectorsize);
1348 return -EUCLEAN;
1350 inline_refs++;
1351 break;
1353 * Contains owner subvolid, owner key objectid, adjusted offset.
1354 * The only obvious corruption can happen in that offset.
1356 case BTRFS_EXTENT_DATA_REF_KEY:
1357 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1358 dref_offset = btrfs_extent_data_ref_offset(leaf, dref);
1359 if (unlikely(!IS_ALIGNED(dref_offset,
1360 fs_info->sectorsize))) {
1361 extent_err(leaf, slot,
1362 "invalid data ref offset, have %llu expect aligned to %u",
1363 dref_offset, fs_info->sectorsize);
1364 return -EUCLEAN;
1366 inline_refs += btrfs_extent_data_ref_count(leaf, dref);
1367 break;
1368 /* Contains parent bytenr and ref count */
1369 case BTRFS_SHARED_DATA_REF_KEY:
1370 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1371 if (unlikely(!IS_ALIGNED(inline_offset,
1372 fs_info->sectorsize))) {
1373 extent_err(leaf, slot,
1374 "invalid data parent bytenr, have %llu expect aligned to %u",
1375 inline_offset, fs_info->sectorsize);
1376 return -EUCLEAN;
1378 inline_refs += btrfs_shared_data_ref_count(leaf, sref);
1379 break;
1380 default:
1381 extent_err(leaf, slot, "unknown inline ref type: %u",
1382 inline_type);
1383 return -EUCLEAN;
1385 ptr += btrfs_extent_inline_ref_size(inline_type);
1387 /* No padding is allowed */
1388 if (unlikely(ptr != end)) {
1389 extent_err(leaf, slot,
1390 "invalid extent item size, padding bytes found");
1391 return -EUCLEAN;
1394 /* Finally, check the inline refs against total refs */
1395 if (unlikely(inline_refs > total_refs)) {
1396 extent_err(leaf, slot,
1397 "invalid extent refs, have %llu expect >= inline %llu",
1398 total_refs, inline_refs);
1399 return -EUCLEAN;
1401 return 0;
1404 static int check_simple_keyed_refs(struct extent_buffer *leaf,
1405 struct btrfs_key *key, int slot)
1407 u32 expect_item_size = 0;
1409 if (key->type == BTRFS_SHARED_DATA_REF_KEY)
1410 expect_item_size = sizeof(struct btrfs_shared_data_ref);
1412 if (unlikely(btrfs_item_size_nr(leaf, slot) != expect_item_size)) {
1413 generic_err(leaf, slot,
1414 "invalid item size, have %u expect %u for key type %u",
1415 btrfs_item_size_nr(leaf, slot),
1416 expect_item_size, key->type);
1417 return -EUCLEAN;
1419 if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1420 generic_err(leaf, slot,
1421 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1422 key->objectid, leaf->fs_info->sectorsize);
1423 return -EUCLEAN;
1425 if (unlikely(key->type != BTRFS_TREE_BLOCK_REF_KEY &&
1426 !IS_ALIGNED(key->offset, leaf->fs_info->sectorsize))) {
1427 extent_err(leaf, slot,
1428 "invalid tree parent bytenr, have %llu expect aligned to %u",
1429 key->offset, leaf->fs_info->sectorsize);
1430 return -EUCLEAN;
1432 return 0;
1435 static int check_extent_data_ref(struct extent_buffer *leaf,
1436 struct btrfs_key *key, int slot)
1438 struct btrfs_extent_data_ref *dref;
1439 unsigned long ptr = btrfs_item_ptr_offset(leaf, slot);
1440 const unsigned long end = ptr + btrfs_item_size_nr(leaf, slot);
1442 if (unlikely(btrfs_item_size_nr(leaf, slot) % sizeof(*dref) != 0)) {
1443 generic_err(leaf, slot,
1444 "invalid item size, have %u expect aligned to %zu for key type %u",
1445 btrfs_item_size_nr(leaf, slot),
1446 sizeof(*dref), key->type);
1447 return -EUCLEAN;
1449 if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1450 generic_err(leaf, slot,
1451 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1452 key->objectid, leaf->fs_info->sectorsize);
1453 return -EUCLEAN;
1455 for (; ptr < end; ptr += sizeof(*dref)) {
1456 u64 root_objectid;
1457 u64 owner;
1458 u64 offset;
1459 u64 hash;
1461 dref = (struct btrfs_extent_data_ref *)ptr;
1462 root_objectid = btrfs_extent_data_ref_root(leaf, dref);
1463 owner = btrfs_extent_data_ref_objectid(leaf, dref);
1464 offset = btrfs_extent_data_ref_offset(leaf, dref);
1465 hash = hash_extent_data_ref(root_objectid, owner, offset);
1466 if (unlikely(hash != key->offset)) {
1467 extent_err(leaf, slot,
1468 "invalid extent data ref hash, item has 0x%016llx key has 0x%016llx",
1469 hash, key->offset);
1470 return -EUCLEAN;
1472 if (unlikely(!IS_ALIGNED(offset, leaf->fs_info->sectorsize))) {
1473 extent_err(leaf, slot,
1474 "invalid extent data backref offset, have %llu expect aligned to %u",
1475 offset, leaf->fs_info->sectorsize);
1476 return -EUCLEAN;
1479 return 0;
1482 #define inode_ref_err(eb, slot, fmt, args...) \
1483 inode_item_err(eb, slot, fmt, ##args)
1484 static int check_inode_ref(struct extent_buffer *leaf,
1485 struct btrfs_key *key, struct btrfs_key *prev_key,
1486 int slot)
1488 struct btrfs_inode_ref *iref;
1489 unsigned long ptr;
1490 unsigned long end;
1492 if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
1493 return -EUCLEAN;
1494 /* namelen can't be 0, so item_size == sizeof() is also invalid */
1495 if (unlikely(btrfs_item_size_nr(leaf, slot) <= sizeof(*iref))) {
1496 inode_ref_err(leaf, slot,
1497 "invalid item size, have %u expect (%zu, %u)",
1498 btrfs_item_size_nr(leaf, slot),
1499 sizeof(*iref), BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
1500 return -EUCLEAN;
1503 ptr = btrfs_item_ptr_offset(leaf, slot);
1504 end = ptr + btrfs_item_size_nr(leaf, slot);
1505 while (ptr < end) {
1506 u16 namelen;
1508 if (unlikely(ptr + sizeof(iref) > end)) {
1509 inode_ref_err(leaf, slot,
1510 "inode ref overflow, ptr %lu end %lu inode_ref_size %zu",
1511 ptr, end, sizeof(iref));
1512 return -EUCLEAN;
1515 iref = (struct btrfs_inode_ref *)ptr;
1516 namelen = btrfs_inode_ref_name_len(leaf, iref);
1517 if (unlikely(ptr + sizeof(*iref) + namelen > end)) {
1518 inode_ref_err(leaf, slot,
1519 "inode ref overflow, ptr %lu end %lu namelen %u",
1520 ptr, end, namelen);
1521 return -EUCLEAN;
1525 * NOTE: In theory we should record all found index numbers
1526 * to find any duplicated indexes, but that will be too time
1527 * consuming for inodes with too many hard links.
1529 ptr += sizeof(*iref) + namelen;
1531 return 0;
1535 * Common point to switch the item-specific validation.
1537 static int check_leaf_item(struct extent_buffer *leaf,
1538 struct btrfs_key *key, int slot,
1539 struct btrfs_key *prev_key)
1541 int ret = 0;
1542 struct btrfs_chunk *chunk;
1544 switch (key->type) {
1545 case BTRFS_EXTENT_DATA_KEY:
1546 ret = check_extent_data_item(leaf, key, slot, prev_key);
1547 break;
1548 case BTRFS_EXTENT_CSUM_KEY:
1549 ret = check_csum_item(leaf, key, slot, prev_key);
1550 break;
1551 case BTRFS_DIR_ITEM_KEY:
1552 case BTRFS_DIR_INDEX_KEY:
1553 case BTRFS_XATTR_ITEM_KEY:
1554 ret = check_dir_item(leaf, key, prev_key, slot);
1555 break;
1556 case BTRFS_INODE_REF_KEY:
1557 ret = check_inode_ref(leaf, key, prev_key, slot);
1558 break;
1559 case BTRFS_BLOCK_GROUP_ITEM_KEY:
1560 ret = check_block_group_item(leaf, key, slot);
1561 break;
1562 case BTRFS_CHUNK_ITEM_KEY:
1563 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1564 ret = check_leaf_chunk_item(leaf, chunk, key, slot);
1565 break;
1566 case BTRFS_DEV_ITEM_KEY:
1567 ret = check_dev_item(leaf, key, slot);
1568 break;
1569 case BTRFS_INODE_ITEM_KEY:
1570 ret = check_inode_item(leaf, key, slot);
1571 break;
1572 case BTRFS_ROOT_ITEM_KEY:
1573 ret = check_root_item(leaf, key, slot);
1574 break;
1575 case BTRFS_EXTENT_ITEM_KEY:
1576 case BTRFS_METADATA_ITEM_KEY:
1577 ret = check_extent_item(leaf, key, slot);
1578 break;
1579 case BTRFS_TREE_BLOCK_REF_KEY:
1580 case BTRFS_SHARED_DATA_REF_KEY:
1581 case BTRFS_SHARED_BLOCK_REF_KEY:
1582 ret = check_simple_keyed_refs(leaf, key, slot);
1583 break;
1584 case BTRFS_EXTENT_DATA_REF_KEY:
1585 ret = check_extent_data_ref(leaf, key, slot);
1586 break;
1588 return ret;
1591 static int check_leaf(struct extent_buffer *leaf, bool check_item_data)
1593 struct btrfs_fs_info *fs_info = leaf->fs_info;
1594 /* No valid key type is 0, so all key should be larger than this key */
1595 struct btrfs_key prev_key = {0, 0, 0};
1596 struct btrfs_key key;
1597 u32 nritems = btrfs_header_nritems(leaf);
1598 int slot;
1600 if (unlikely(btrfs_header_level(leaf) != 0)) {
1601 generic_err(leaf, 0,
1602 "invalid level for leaf, have %d expect 0",
1603 btrfs_header_level(leaf));
1604 return -EUCLEAN;
1608 * Extent buffers from a relocation tree have a owner field that
1609 * corresponds to the subvolume tree they are based on. So just from an
1610 * extent buffer alone we can not find out what is the id of the
1611 * corresponding subvolume tree, so we can not figure out if the extent
1612 * buffer corresponds to the root of the relocation tree or not. So
1613 * skip this check for relocation trees.
1615 if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
1616 u64 owner = btrfs_header_owner(leaf);
1618 /* These trees must never be empty */
1619 if (unlikely(owner == BTRFS_ROOT_TREE_OBJECTID ||
1620 owner == BTRFS_CHUNK_TREE_OBJECTID ||
1621 owner == BTRFS_EXTENT_TREE_OBJECTID ||
1622 owner == BTRFS_DEV_TREE_OBJECTID ||
1623 owner == BTRFS_FS_TREE_OBJECTID ||
1624 owner == BTRFS_DATA_RELOC_TREE_OBJECTID)) {
1625 generic_err(leaf, 0,
1626 "invalid root, root %llu must never be empty",
1627 owner);
1628 return -EUCLEAN;
1630 /* Unknown tree */
1631 if (unlikely(owner == 0)) {
1632 generic_err(leaf, 0,
1633 "invalid owner, root 0 is not defined");
1634 return -EUCLEAN;
1636 return 0;
1639 if (unlikely(nritems == 0))
1640 return 0;
1643 * Check the following things to make sure this is a good leaf, and
1644 * leaf users won't need to bother with similar sanity checks:
1646 * 1) key ordering
1647 * 2) item offset and size
1648 * No overlap, no hole, all inside the leaf.
1649 * 3) item content
1650 * If possible, do comprehensive sanity check.
1651 * NOTE: All checks must only rely on the item data itself.
1653 for (slot = 0; slot < nritems; slot++) {
1654 u32 item_end_expected;
1655 int ret;
1657 btrfs_item_key_to_cpu(leaf, &key, slot);
1659 /* Make sure the keys are in the right order */
1660 if (unlikely(btrfs_comp_cpu_keys(&prev_key, &key) >= 0)) {
1661 generic_err(leaf, slot,
1662 "bad key order, prev (%llu %u %llu) current (%llu %u %llu)",
1663 prev_key.objectid, prev_key.type,
1664 prev_key.offset, key.objectid, key.type,
1665 key.offset);
1666 return -EUCLEAN;
1670 * Make sure the offset and ends are right, remember that the
1671 * item data starts at the end of the leaf and grows towards the
1672 * front.
1674 if (slot == 0)
1675 item_end_expected = BTRFS_LEAF_DATA_SIZE(fs_info);
1676 else
1677 item_end_expected = btrfs_item_offset_nr(leaf,
1678 slot - 1);
1679 if (unlikely(btrfs_item_end_nr(leaf, slot) != item_end_expected)) {
1680 generic_err(leaf, slot,
1681 "unexpected item end, have %u expect %u",
1682 btrfs_item_end_nr(leaf, slot),
1683 item_end_expected);
1684 return -EUCLEAN;
1688 * Check to make sure that we don't point outside of the leaf,
1689 * just in case all the items are consistent to each other, but
1690 * all point outside of the leaf.
1692 if (unlikely(btrfs_item_end_nr(leaf, slot) >
1693 BTRFS_LEAF_DATA_SIZE(fs_info))) {
1694 generic_err(leaf, slot,
1695 "slot end outside of leaf, have %u expect range [0, %u]",
1696 btrfs_item_end_nr(leaf, slot),
1697 BTRFS_LEAF_DATA_SIZE(fs_info));
1698 return -EUCLEAN;
1701 /* Also check if the item pointer overlaps with btrfs item. */
1702 if (unlikely(btrfs_item_ptr_offset(leaf, slot) <
1703 btrfs_item_nr_offset(slot) + sizeof(struct btrfs_item))) {
1704 generic_err(leaf, slot,
1705 "slot overlaps with its data, item end %lu data start %lu",
1706 btrfs_item_nr_offset(slot) +
1707 sizeof(struct btrfs_item),
1708 btrfs_item_ptr_offset(leaf, slot));
1709 return -EUCLEAN;
1712 if (check_item_data) {
1714 * Check if the item size and content meet other
1715 * criteria
1717 ret = check_leaf_item(leaf, &key, slot, &prev_key);
1718 if (unlikely(ret < 0))
1719 return ret;
1722 prev_key.objectid = key.objectid;
1723 prev_key.type = key.type;
1724 prev_key.offset = key.offset;
1727 return 0;
1730 int btrfs_check_leaf_full(struct extent_buffer *leaf)
1732 return check_leaf(leaf, true);
1734 ALLOW_ERROR_INJECTION(btrfs_check_leaf_full, ERRNO);
1736 int btrfs_check_leaf_relaxed(struct extent_buffer *leaf)
1738 return check_leaf(leaf, false);
1741 int btrfs_check_node(struct extent_buffer *node)
1743 struct btrfs_fs_info *fs_info = node->fs_info;
1744 unsigned long nr = btrfs_header_nritems(node);
1745 struct btrfs_key key, next_key;
1746 int slot;
1747 int level = btrfs_header_level(node);
1748 u64 bytenr;
1749 int ret = 0;
1751 if (unlikely(level <= 0 || level >= BTRFS_MAX_LEVEL)) {
1752 generic_err(node, 0,
1753 "invalid level for node, have %d expect [1, %d]",
1754 level, BTRFS_MAX_LEVEL - 1);
1755 return -EUCLEAN;
1757 if (unlikely(nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(fs_info))) {
1758 btrfs_crit(fs_info,
1759 "corrupt node: root=%llu block=%llu, nritems too %s, have %lu expect range [1,%u]",
1760 btrfs_header_owner(node), node->start,
1761 nr == 0 ? "small" : "large", nr,
1762 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1763 return -EUCLEAN;
1766 for (slot = 0; slot < nr - 1; slot++) {
1767 bytenr = btrfs_node_blockptr(node, slot);
1768 btrfs_node_key_to_cpu(node, &key, slot);
1769 btrfs_node_key_to_cpu(node, &next_key, slot + 1);
1771 if (unlikely(!bytenr)) {
1772 generic_err(node, slot,
1773 "invalid NULL node pointer");
1774 ret = -EUCLEAN;
1775 goto out;
1777 if (unlikely(!IS_ALIGNED(bytenr, fs_info->sectorsize))) {
1778 generic_err(node, slot,
1779 "unaligned pointer, have %llu should be aligned to %u",
1780 bytenr, fs_info->sectorsize);
1781 ret = -EUCLEAN;
1782 goto out;
1785 if (unlikely(btrfs_comp_cpu_keys(&key, &next_key) >= 0)) {
1786 generic_err(node, slot,
1787 "bad key order, current (%llu %u %llu) next (%llu %u %llu)",
1788 key.objectid, key.type, key.offset,
1789 next_key.objectid, next_key.type,
1790 next_key.offset);
1791 ret = -EUCLEAN;
1792 goto out;
1795 out:
1796 return ret;
1798 ALLOW_ERROR_INJECTION(btrfs_check_node, ERRNO);