Merge tag 'regmap-fix-v5.11-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux/fpc-iii.git] / fs / btrfs / tree-checker.c
blob028e733e42f3b5fe01668b69e9c2a552722c231c
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 stripe_len;
764 u16 num_stripes;
765 u16 sub_stripes;
766 u64 type;
767 u64 features;
768 bool mixed = false;
769 int raid_index;
770 int nparity;
771 int ncopies;
773 length = btrfs_chunk_length(leaf, chunk);
774 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
775 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
776 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
777 type = btrfs_chunk_type(leaf, chunk);
778 raid_index = btrfs_bg_flags_to_raid_index(type);
779 ncopies = btrfs_raid_array[raid_index].ncopies;
780 nparity = btrfs_raid_array[raid_index].nparity;
782 if (unlikely(!num_stripes)) {
783 chunk_err(leaf, chunk, logical,
784 "invalid chunk num_stripes, have %u", num_stripes);
785 return -EUCLEAN;
787 if (unlikely(num_stripes < ncopies)) {
788 chunk_err(leaf, chunk, logical,
789 "invalid chunk num_stripes < ncopies, have %u < %d",
790 num_stripes, ncopies);
791 return -EUCLEAN;
793 if (unlikely(nparity && num_stripes == nparity)) {
794 chunk_err(leaf, chunk, logical,
795 "invalid chunk num_stripes == nparity, have %u == %d",
796 num_stripes, nparity);
797 return -EUCLEAN;
799 if (unlikely(!IS_ALIGNED(logical, fs_info->sectorsize))) {
800 chunk_err(leaf, chunk, logical,
801 "invalid chunk logical, have %llu should aligned to %u",
802 logical, fs_info->sectorsize);
803 return -EUCLEAN;
805 if (unlikely(btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize)) {
806 chunk_err(leaf, chunk, logical,
807 "invalid chunk sectorsize, have %u expect %u",
808 btrfs_chunk_sector_size(leaf, chunk),
809 fs_info->sectorsize);
810 return -EUCLEAN;
812 if (unlikely(!length || !IS_ALIGNED(length, fs_info->sectorsize))) {
813 chunk_err(leaf, chunk, logical,
814 "invalid chunk length, have %llu", length);
815 return -EUCLEAN;
817 if (unlikely(!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN)) {
818 chunk_err(leaf, chunk, logical,
819 "invalid chunk stripe length: %llu",
820 stripe_len);
821 return -EUCLEAN;
823 if (unlikely(type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
824 BTRFS_BLOCK_GROUP_PROFILE_MASK))) {
825 chunk_err(leaf, chunk, logical,
826 "unrecognized chunk type: 0x%llx",
827 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
828 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
829 btrfs_chunk_type(leaf, chunk));
830 return -EUCLEAN;
833 if (unlikely(!has_single_bit_set(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
834 (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) != 0)) {
835 chunk_err(leaf, chunk, logical,
836 "invalid chunk profile flag: 0x%llx, expect 0 or 1 bit set",
837 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
838 return -EUCLEAN;
840 if (unlikely((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0)) {
841 chunk_err(leaf, chunk, logical,
842 "missing chunk type flag, have 0x%llx one bit must be set in 0x%llx",
843 type, BTRFS_BLOCK_GROUP_TYPE_MASK);
844 return -EUCLEAN;
847 if (unlikely((type & BTRFS_BLOCK_GROUP_SYSTEM) &&
848 (type & (BTRFS_BLOCK_GROUP_METADATA |
849 BTRFS_BLOCK_GROUP_DATA)))) {
850 chunk_err(leaf, chunk, logical,
851 "system chunk with data or metadata type: 0x%llx",
852 type);
853 return -EUCLEAN;
856 features = btrfs_super_incompat_flags(fs_info->super_copy);
857 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
858 mixed = true;
860 if (!mixed) {
861 if (unlikely((type & BTRFS_BLOCK_GROUP_METADATA) &&
862 (type & BTRFS_BLOCK_GROUP_DATA))) {
863 chunk_err(leaf, chunk, logical,
864 "mixed chunk type in non-mixed mode: 0x%llx", type);
865 return -EUCLEAN;
869 if (unlikely((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
870 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes != 2) ||
871 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
872 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
873 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes != 2) ||
874 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
875 num_stripes != 1))) {
876 chunk_err(leaf, chunk, logical,
877 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
878 num_stripes, sub_stripes,
879 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
880 return -EUCLEAN;
883 return 0;
887 * Enhanced version of chunk item checker.
889 * The common btrfs_check_chunk_valid() doesn't check item size since it needs
890 * to work on super block sys_chunk_array which doesn't have full item ptr.
892 static int check_leaf_chunk_item(struct extent_buffer *leaf,
893 struct btrfs_chunk *chunk,
894 struct btrfs_key *key, int slot)
896 int num_stripes;
898 if (unlikely(btrfs_item_size_nr(leaf, slot) < sizeof(struct btrfs_chunk))) {
899 chunk_err(leaf, chunk, key->offset,
900 "invalid chunk item size: have %u expect [%zu, %u)",
901 btrfs_item_size_nr(leaf, slot),
902 sizeof(struct btrfs_chunk),
903 BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
904 return -EUCLEAN;
907 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
908 /* Let btrfs_check_chunk_valid() handle this error type */
909 if (num_stripes == 0)
910 goto out;
912 if (unlikely(btrfs_chunk_item_size(num_stripes) !=
913 btrfs_item_size_nr(leaf, slot))) {
914 chunk_err(leaf, chunk, key->offset,
915 "invalid chunk item size: have %u expect %lu",
916 btrfs_item_size_nr(leaf, slot),
917 btrfs_chunk_item_size(num_stripes));
918 return -EUCLEAN;
920 out:
921 return btrfs_check_chunk_valid(leaf, chunk, key->offset);
924 __printf(3, 4)
925 __cold
926 static void dev_item_err(const struct extent_buffer *eb, int slot,
927 const char *fmt, ...)
929 struct btrfs_key key;
930 struct va_format vaf;
931 va_list args;
933 btrfs_item_key_to_cpu(eb, &key, slot);
934 va_start(args, fmt);
936 vaf.fmt = fmt;
937 vaf.va = &args;
939 btrfs_crit(eb->fs_info,
940 "corrupt %s: root=%llu block=%llu slot=%d devid=%llu %pV",
941 btrfs_header_level(eb) == 0 ? "leaf" : "node",
942 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
943 key.objectid, &vaf);
944 va_end(args);
947 static int check_dev_item(struct extent_buffer *leaf,
948 struct btrfs_key *key, int slot)
950 struct btrfs_dev_item *ditem;
952 if (unlikely(key->objectid != BTRFS_DEV_ITEMS_OBJECTID)) {
953 dev_item_err(leaf, slot,
954 "invalid objectid: has=%llu expect=%llu",
955 key->objectid, BTRFS_DEV_ITEMS_OBJECTID);
956 return -EUCLEAN;
958 ditem = btrfs_item_ptr(leaf, slot, struct btrfs_dev_item);
959 if (unlikely(btrfs_device_id(leaf, ditem) != key->offset)) {
960 dev_item_err(leaf, slot,
961 "devid mismatch: key has=%llu item has=%llu",
962 key->offset, btrfs_device_id(leaf, ditem));
963 return -EUCLEAN;
967 * For device total_bytes, we don't have reliable way to check it, as
968 * it can be 0 for device removal. Device size check can only be done
969 * by dev extents check.
971 if (unlikely(btrfs_device_bytes_used(leaf, ditem) >
972 btrfs_device_total_bytes(leaf, ditem))) {
973 dev_item_err(leaf, slot,
974 "invalid bytes used: have %llu expect [0, %llu]",
975 btrfs_device_bytes_used(leaf, ditem),
976 btrfs_device_total_bytes(leaf, ditem));
977 return -EUCLEAN;
980 * Remaining members like io_align/type/gen/dev_group aren't really
981 * utilized. Skip them to make later usage of them easier.
983 return 0;
986 static int check_inode_item(struct extent_buffer *leaf,
987 struct btrfs_key *key, int slot)
989 struct btrfs_fs_info *fs_info = leaf->fs_info;
990 struct btrfs_inode_item *iitem;
991 u64 super_gen = btrfs_super_generation(fs_info->super_copy);
992 u32 valid_mask = (S_IFMT | S_ISUID | S_ISGID | S_ISVTX | 0777);
993 u32 mode;
994 int ret;
996 ret = check_inode_key(leaf, key, slot);
997 if (unlikely(ret < 0))
998 return ret;
1000 iitem = btrfs_item_ptr(leaf, slot, struct btrfs_inode_item);
1002 /* Here we use super block generation + 1 to handle log tree */
1003 if (unlikely(btrfs_inode_generation(leaf, iitem) > super_gen + 1)) {
1004 inode_item_err(leaf, slot,
1005 "invalid inode generation: has %llu expect (0, %llu]",
1006 btrfs_inode_generation(leaf, iitem),
1007 super_gen + 1);
1008 return -EUCLEAN;
1010 /* Note for ROOT_TREE_DIR_ITEM, mkfs could set its transid 0 */
1011 if (unlikely(btrfs_inode_transid(leaf, iitem) > super_gen + 1)) {
1012 inode_item_err(leaf, slot,
1013 "invalid inode transid: has %llu expect [0, %llu]",
1014 btrfs_inode_transid(leaf, iitem), super_gen + 1);
1015 return -EUCLEAN;
1019 * For size and nbytes it's better not to be too strict, as for dir
1020 * item its size/nbytes can easily get wrong, but doesn't affect
1021 * anything in the fs. So here we skip the check.
1023 mode = btrfs_inode_mode(leaf, iitem);
1024 if (unlikely(mode & ~valid_mask)) {
1025 inode_item_err(leaf, slot,
1026 "unknown mode bit detected: 0x%x",
1027 mode & ~valid_mask);
1028 return -EUCLEAN;
1032 * S_IFMT is not bit mapped so we can't completely rely on
1033 * is_power_of_2/has_single_bit_set, but it can save us from checking
1034 * FIFO/CHR/DIR/REG. Only needs to check BLK, LNK and SOCKS
1036 if (!has_single_bit_set(mode & S_IFMT)) {
1037 if (unlikely(!S_ISLNK(mode) && !S_ISBLK(mode) && !S_ISSOCK(mode))) {
1038 inode_item_err(leaf, slot,
1039 "invalid mode: has 0%o expect valid S_IF* bit(s)",
1040 mode & S_IFMT);
1041 return -EUCLEAN;
1044 if (unlikely(S_ISDIR(mode) && btrfs_inode_nlink(leaf, iitem) > 1)) {
1045 inode_item_err(leaf, slot,
1046 "invalid nlink: has %u expect no more than 1 for dir",
1047 btrfs_inode_nlink(leaf, iitem));
1048 return -EUCLEAN;
1050 if (unlikely(btrfs_inode_flags(leaf, iitem) & ~BTRFS_INODE_FLAG_MASK)) {
1051 inode_item_err(leaf, slot,
1052 "unknown flags detected: 0x%llx",
1053 btrfs_inode_flags(leaf, iitem) &
1054 ~BTRFS_INODE_FLAG_MASK);
1055 return -EUCLEAN;
1057 return 0;
1060 static int check_root_item(struct extent_buffer *leaf, struct btrfs_key *key,
1061 int slot)
1063 struct btrfs_fs_info *fs_info = leaf->fs_info;
1064 struct btrfs_root_item ri = { 0 };
1065 const u64 valid_root_flags = BTRFS_ROOT_SUBVOL_RDONLY |
1066 BTRFS_ROOT_SUBVOL_DEAD;
1067 int ret;
1069 ret = check_root_key(leaf, key, slot);
1070 if (unlikely(ret < 0))
1071 return ret;
1073 if (unlikely(btrfs_item_size_nr(leaf, slot) != sizeof(ri) &&
1074 btrfs_item_size_nr(leaf, slot) !=
1075 btrfs_legacy_root_item_size())) {
1076 generic_err(leaf, slot,
1077 "invalid root item size, have %u expect %zu or %u",
1078 btrfs_item_size_nr(leaf, slot), sizeof(ri),
1079 btrfs_legacy_root_item_size());
1080 return -EUCLEAN;
1084 * For legacy root item, the members starting at generation_v2 will be
1085 * all filled with 0.
1086 * And since we allow geneartion_v2 as 0, it will still pass the check.
1088 read_extent_buffer(leaf, &ri, btrfs_item_ptr_offset(leaf, slot),
1089 btrfs_item_size_nr(leaf, slot));
1091 /* Generation related */
1092 if (unlikely(btrfs_root_generation(&ri) >
1093 btrfs_super_generation(fs_info->super_copy) + 1)) {
1094 generic_err(leaf, slot,
1095 "invalid root generation, have %llu expect (0, %llu]",
1096 btrfs_root_generation(&ri),
1097 btrfs_super_generation(fs_info->super_copy) + 1);
1098 return -EUCLEAN;
1100 if (unlikely(btrfs_root_generation_v2(&ri) >
1101 btrfs_super_generation(fs_info->super_copy) + 1)) {
1102 generic_err(leaf, slot,
1103 "invalid root v2 generation, have %llu expect (0, %llu]",
1104 btrfs_root_generation_v2(&ri),
1105 btrfs_super_generation(fs_info->super_copy) + 1);
1106 return -EUCLEAN;
1108 if (unlikely(btrfs_root_last_snapshot(&ri) >
1109 btrfs_super_generation(fs_info->super_copy) + 1)) {
1110 generic_err(leaf, slot,
1111 "invalid root last_snapshot, have %llu expect (0, %llu]",
1112 btrfs_root_last_snapshot(&ri),
1113 btrfs_super_generation(fs_info->super_copy) + 1);
1114 return -EUCLEAN;
1117 /* Alignment and level check */
1118 if (unlikely(!IS_ALIGNED(btrfs_root_bytenr(&ri), fs_info->sectorsize))) {
1119 generic_err(leaf, slot,
1120 "invalid root bytenr, have %llu expect to be aligned to %u",
1121 btrfs_root_bytenr(&ri), fs_info->sectorsize);
1122 return -EUCLEAN;
1124 if (unlikely(btrfs_root_level(&ri) >= BTRFS_MAX_LEVEL)) {
1125 generic_err(leaf, slot,
1126 "invalid root level, have %u expect [0, %u]",
1127 btrfs_root_level(&ri), BTRFS_MAX_LEVEL - 1);
1128 return -EUCLEAN;
1130 if (unlikely(btrfs_root_drop_level(&ri) >= BTRFS_MAX_LEVEL)) {
1131 generic_err(leaf, slot,
1132 "invalid root level, have %u expect [0, %u]",
1133 btrfs_root_drop_level(&ri), BTRFS_MAX_LEVEL - 1);
1134 return -EUCLEAN;
1137 /* Flags check */
1138 if (unlikely(btrfs_root_flags(&ri) & ~valid_root_flags)) {
1139 generic_err(leaf, slot,
1140 "invalid root flags, have 0x%llx expect mask 0x%llx",
1141 btrfs_root_flags(&ri), valid_root_flags);
1142 return -EUCLEAN;
1144 return 0;
1147 __printf(3,4)
1148 __cold
1149 static void extent_err(const struct extent_buffer *eb, int slot,
1150 const char *fmt, ...)
1152 struct btrfs_key key;
1153 struct va_format vaf;
1154 va_list args;
1155 u64 bytenr;
1156 u64 len;
1158 btrfs_item_key_to_cpu(eb, &key, slot);
1159 bytenr = key.objectid;
1160 if (key.type == BTRFS_METADATA_ITEM_KEY ||
1161 key.type == BTRFS_TREE_BLOCK_REF_KEY ||
1162 key.type == BTRFS_SHARED_BLOCK_REF_KEY)
1163 len = eb->fs_info->nodesize;
1164 else
1165 len = key.offset;
1166 va_start(args, fmt);
1168 vaf.fmt = fmt;
1169 vaf.va = &args;
1171 btrfs_crit(eb->fs_info,
1172 "corrupt %s: block=%llu slot=%d extent bytenr=%llu len=%llu %pV",
1173 btrfs_header_level(eb) == 0 ? "leaf" : "node",
1174 eb->start, slot, bytenr, len, &vaf);
1175 va_end(args);
1178 static int check_extent_item(struct extent_buffer *leaf,
1179 struct btrfs_key *key, int slot)
1181 struct btrfs_fs_info *fs_info = leaf->fs_info;
1182 struct btrfs_extent_item *ei;
1183 bool is_tree_block = false;
1184 unsigned long ptr; /* Current pointer inside inline refs */
1185 unsigned long end; /* Extent item end */
1186 const u32 item_size = btrfs_item_size_nr(leaf, slot);
1187 u64 flags;
1188 u64 generation;
1189 u64 total_refs; /* Total refs in btrfs_extent_item */
1190 u64 inline_refs = 0; /* found total inline refs */
1192 if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1193 !btrfs_fs_incompat(fs_info, SKINNY_METADATA))) {
1194 generic_err(leaf, slot,
1195 "invalid key type, METADATA_ITEM type invalid when SKINNY_METADATA feature disabled");
1196 return -EUCLEAN;
1198 /* key->objectid is the bytenr for both key types */
1199 if (unlikely(!IS_ALIGNED(key->objectid, fs_info->sectorsize))) {
1200 generic_err(leaf, slot,
1201 "invalid key objectid, have %llu expect to be aligned to %u",
1202 key->objectid, fs_info->sectorsize);
1203 return -EUCLEAN;
1206 /* key->offset is tree level for METADATA_ITEM_KEY */
1207 if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1208 key->offset >= BTRFS_MAX_LEVEL)) {
1209 extent_err(leaf, slot,
1210 "invalid tree level, have %llu expect [0, %u]",
1211 key->offset, BTRFS_MAX_LEVEL - 1);
1212 return -EUCLEAN;
1216 * EXTENT/METADATA_ITEM consists of:
1217 * 1) One btrfs_extent_item
1218 * Records the total refs, type and generation of the extent.
1220 * 2) One btrfs_tree_block_info (for EXTENT_ITEM and tree backref only)
1221 * Records the first key and level of the tree block.
1223 * 2) Zero or more btrfs_extent_inline_ref(s)
1224 * Each inline ref has one btrfs_extent_inline_ref shows:
1225 * 2.1) The ref type, one of the 4
1226 * TREE_BLOCK_REF Tree block only
1227 * SHARED_BLOCK_REF Tree block only
1228 * EXTENT_DATA_REF Data only
1229 * SHARED_DATA_REF Data only
1230 * 2.2) Ref type specific data
1231 * Either using btrfs_extent_inline_ref::offset, or specific
1232 * data structure.
1234 if (unlikely(item_size < sizeof(*ei))) {
1235 extent_err(leaf, slot,
1236 "invalid item size, have %u expect [%zu, %u)",
1237 item_size, sizeof(*ei),
1238 BTRFS_LEAF_DATA_SIZE(fs_info));
1239 return -EUCLEAN;
1241 end = item_size + btrfs_item_ptr_offset(leaf, slot);
1243 /* Checks against extent_item */
1244 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
1245 flags = btrfs_extent_flags(leaf, ei);
1246 total_refs = btrfs_extent_refs(leaf, ei);
1247 generation = btrfs_extent_generation(leaf, ei);
1248 if (unlikely(generation >
1249 btrfs_super_generation(fs_info->super_copy) + 1)) {
1250 extent_err(leaf, slot,
1251 "invalid generation, have %llu expect (0, %llu]",
1252 generation,
1253 btrfs_super_generation(fs_info->super_copy) + 1);
1254 return -EUCLEAN;
1256 if (unlikely(!has_single_bit_set(flags & (BTRFS_EXTENT_FLAG_DATA |
1257 BTRFS_EXTENT_FLAG_TREE_BLOCK)))) {
1258 extent_err(leaf, slot,
1259 "invalid extent flag, have 0x%llx expect 1 bit set in 0x%llx",
1260 flags, BTRFS_EXTENT_FLAG_DATA |
1261 BTRFS_EXTENT_FLAG_TREE_BLOCK);
1262 return -EUCLEAN;
1264 is_tree_block = !!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK);
1265 if (is_tree_block) {
1266 if (unlikely(key->type == BTRFS_EXTENT_ITEM_KEY &&
1267 key->offset != fs_info->nodesize)) {
1268 extent_err(leaf, slot,
1269 "invalid extent length, have %llu expect %u",
1270 key->offset, fs_info->nodesize);
1271 return -EUCLEAN;
1273 } else {
1274 if (unlikely(key->type != BTRFS_EXTENT_ITEM_KEY)) {
1275 extent_err(leaf, slot,
1276 "invalid key type, have %u expect %u for data backref",
1277 key->type, BTRFS_EXTENT_ITEM_KEY);
1278 return -EUCLEAN;
1280 if (unlikely(!IS_ALIGNED(key->offset, fs_info->sectorsize))) {
1281 extent_err(leaf, slot,
1282 "invalid extent length, have %llu expect aligned to %u",
1283 key->offset, fs_info->sectorsize);
1284 return -EUCLEAN;
1287 ptr = (unsigned long)(struct btrfs_extent_item *)(ei + 1);
1289 /* Check the special case of btrfs_tree_block_info */
1290 if (is_tree_block && key->type != BTRFS_METADATA_ITEM_KEY) {
1291 struct btrfs_tree_block_info *info;
1293 info = (struct btrfs_tree_block_info *)ptr;
1294 if (unlikely(btrfs_tree_block_level(leaf, info) >= BTRFS_MAX_LEVEL)) {
1295 extent_err(leaf, slot,
1296 "invalid tree block info level, have %u expect [0, %u]",
1297 btrfs_tree_block_level(leaf, info),
1298 BTRFS_MAX_LEVEL - 1);
1299 return -EUCLEAN;
1301 ptr = (unsigned long)(struct btrfs_tree_block_info *)(info + 1);
1304 /* Check inline refs */
1305 while (ptr < end) {
1306 struct btrfs_extent_inline_ref *iref;
1307 struct btrfs_extent_data_ref *dref;
1308 struct btrfs_shared_data_ref *sref;
1309 u64 dref_offset;
1310 u64 inline_offset;
1311 u8 inline_type;
1313 if (unlikely(ptr + sizeof(*iref) > end)) {
1314 extent_err(leaf, slot,
1315 "inline ref item overflows extent item, ptr %lu iref size %zu end %lu",
1316 ptr, sizeof(*iref), end);
1317 return -EUCLEAN;
1319 iref = (struct btrfs_extent_inline_ref *)ptr;
1320 inline_type = btrfs_extent_inline_ref_type(leaf, iref);
1321 inline_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1322 if (unlikely(ptr + btrfs_extent_inline_ref_size(inline_type) > end)) {
1323 extent_err(leaf, slot,
1324 "inline ref item overflows extent item, ptr %lu iref size %u end %lu",
1325 ptr, inline_type, end);
1326 return -EUCLEAN;
1329 switch (inline_type) {
1330 /* inline_offset is subvolid of the owner, no need to check */
1331 case BTRFS_TREE_BLOCK_REF_KEY:
1332 inline_refs++;
1333 break;
1334 /* Contains parent bytenr */
1335 case BTRFS_SHARED_BLOCK_REF_KEY:
1336 if (unlikely(!IS_ALIGNED(inline_offset,
1337 fs_info->sectorsize))) {
1338 extent_err(leaf, slot,
1339 "invalid tree parent bytenr, have %llu expect aligned to %u",
1340 inline_offset, fs_info->sectorsize);
1341 return -EUCLEAN;
1343 inline_refs++;
1344 break;
1346 * Contains owner subvolid, owner key objectid, adjusted offset.
1347 * The only obvious corruption can happen in that offset.
1349 case BTRFS_EXTENT_DATA_REF_KEY:
1350 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1351 dref_offset = btrfs_extent_data_ref_offset(leaf, dref);
1352 if (unlikely(!IS_ALIGNED(dref_offset,
1353 fs_info->sectorsize))) {
1354 extent_err(leaf, slot,
1355 "invalid data ref offset, have %llu expect aligned to %u",
1356 dref_offset, fs_info->sectorsize);
1357 return -EUCLEAN;
1359 inline_refs += btrfs_extent_data_ref_count(leaf, dref);
1360 break;
1361 /* Contains parent bytenr and ref count */
1362 case BTRFS_SHARED_DATA_REF_KEY:
1363 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1364 if (unlikely(!IS_ALIGNED(inline_offset,
1365 fs_info->sectorsize))) {
1366 extent_err(leaf, slot,
1367 "invalid data parent bytenr, have %llu expect aligned to %u",
1368 inline_offset, fs_info->sectorsize);
1369 return -EUCLEAN;
1371 inline_refs += btrfs_shared_data_ref_count(leaf, sref);
1372 break;
1373 default:
1374 extent_err(leaf, slot, "unknown inline ref type: %u",
1375 inline_type);
1376 return -EUCLEAN;
1378 ptr += btrfs_extent_inline_ref_size(inline_type);
1380 /* No padding is allowed */
1381 if (unlikely(ptr != end)) {
1382 extent_err(leaf, slot,
1383 "invalid extent item size, padding bytes found");
1384 return -EUCLEAN;
1387 /* Finally, check the inline refs against total refs */
1388 if (unlikely(inline_refs > total_refs)) {
1389 extent_err(leaf, slot,
1390 "invalid extent refs, have %llu expect >= inline %llu",
1391 total_refs, inline_refs);
1392 return -EUCLEAN;
1394 return 0;
1397 static int check_simple_keyed_refs(struct extent_buffer *leaf,
1398 struct btrfs_key *key, int slot)
1400 u32 expect_item_size = 0;
1402 if (key->type == BTRFS_SHARED_DATA_REF_KEY)
1403 expect_item_size = sizeof(struct btrfs_shared_data_ref);
1405 if (unlikely(btrfs_item_size_nr(leaf, slot) != expect_item_size)) {
1406 generic_err(leaf, slot,
1407 "invalid item size, have %u expect %u for key type %u",
1408 btrfs_item_size_nr(leaf, slot),
1409 expect_item_size, key->type);
1410 return -EUCLEAN;
1412 if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1413 generic_err(leaf, slot,
1414 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1415 key->objectid, leaf->fs_info->sectorsize);
1416 return -EUCLEAN;
1418 if (unlikely(key->type != BTRFS_TREE_BLOCK_REF_KEY &&
1419 !IS_ALIGNED(key->offset, leaf->fs_info->sectorsize))) {
1420 extent_err(leaf, slot,
1421 "invalid tree parent bytenr, have %llu expect aligned to %u",
1422 key->offset, leaf->fs_info->sectorsize);
1423 return -EUCLEAN;
1425 return 0;
1428 static int check_extent_data_ref(struct extent_buffer *leaf,
1429 struct btrfs_key *key, int slot)
1431 struct btrfs_extent_data_ref *dref;
1432 unsigned long ptr = btrfs_item_ptr_offset(leaf, slot);
1433 const unsigned long end = ptr + btrfs_item_size_nr(leaf, slot);
1435 if (unlikely(btrfs_item_size_nr(leaf, slot) % sizeof(*dref) != 0)) {
1436 generic_err(leaf, slot,
1437 "invalid item size, have %u expect aligned to %zu for key type %u",
1438 btrfs_item_size_nr(leaf, slot),
1439 sizeof(*dref), key->type);
1440 return -EUCLEAN;
1442 if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1443 generic_err(leaf, slot,
1444 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1445 key->objectid, leaf->fs_info->sectorsize);
1446 return -EUCLEAN;
1448 for (; ptr < end; ptr += sizeof(*dref)) {
1449 u64 root_objectid;
1450 u64 owner;
1451 u64 offset;
1452 u64 hash;
1454 dref = (struct btrfs_extent_data_ref *)ptr;
1455 root_objectid = btrfs_extent_data_ref_root(leaf, dref);
1456 owner = btrfs_extent_data_ref_objectid(leaf, dref);
1457 offset = btrfs_extent_data_ref_offset(leaf, dref);
1458 hash = hash_extent_data_ref(root_objectid, owner, offset);
1459 if (unlikely(hash != key->offset)) {
1460 extent_err(leaf, slot,
1461 "invalid extent data ref hash, item has 0x%016llx key has 0x%016llx",
1462 hash, key->offset);
1463 return -EUCLEAN;
1465 if (unlikely(!IS_ALIGNED(offset, leaf->fs_info->sectorsize))) {
1466 extent_err(leaf, slot,
1467 "invalid extent data backref offset, have %llu expect aligned to %u",
1468 offset, leaf->fs_info->sectorsize);
1469 return -EUCLEAN;
1472 return 0;
1475 #define inode_ref_err(eb, slot, fmt, args...) \
1476 inode_item_err(eb, slot, fmt, ##args)
1477 static int check_inode_ref(struct extent_buffer *leaf,
1478 struct btrfs_key *key, struct btrfs_key *prev_key,
1479 int slot)
1481 struct btrfs_inode_ref *iref;
1482 unsigned long ptr;
1483 unsigned long end;
1485 if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
1486 return -EUCLEAN;
1487 /* namelen can't be 0, so item_size == sizeof() is also invalid */
1488 if (unlikely(btrfs_item_size_nr(leaf, slot) <= sizeof(*iref))) {
1489 inode_ref_err(leaf, slot,
1490 "invalid item size, have %u expect (%zu, %u)",
1491 btrfs_item_size_nr(leaf, slot),
1492 sizeof(*iref), BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
1493 return -EUCLEAN;
1496 ptr = btrfs_item_ptr_offset(leaf, slot);
1497 end = ptr + btrfs_item_size_nr(leaf, slot);
1498 while (ptr < end) {
1499 u16 namelen;
1501 if (unlikely(ptr + sizeof(iref) > end)) {
1502 inode_ref_err(leaf, slot,
1503 "inode ref overflow, ptr %lu end %lu inode_ref_size %zu",
1504 ptr, end, sizeof(iref));
1505 return -EUCLEAN;
1508 iref = (struct btrfs_inode_ref *)ptr;
1509 namelen = btrfs_inode_ref_name_len(leaf, iref);
1510 if (unlikely(ptr + sizeof(*iref) + namelen > end)) {
1511 inode_ref_err(leaf, slot,
1512 "inode ref overflow, ptr %lu end %lu namelen %u",
1513 ptr, end, namelen);
1514 return -EUCLEAN;
1518 * NOTE: In theory we should record all found index numbers
1519 * to find any duplicated indexes, but that will be too time
1520 * consuming for inodes with too many hard links.
1522 ptr += sizeof(*iref) + namelen;
1524 return 0;
1528 * Common point to switch the item-specific validation.
1530 static int check_leaf_item(struct extent_buffer *leaf,
1531 struct btrfs_key *key, int slot,
1532 struct btrfs_key *prev_key)
1534 int ret = 0;
1535 struct btrfs_chunk *chunk;
1537 switch (key->type) {
1538 case BTRFS_EXTENT_DATA_KEY:
1539 ret = check_extent_data_item(leaf, key, slot, prev_key);
1540 break;
1541 case BTRFS_EXTENT_CSUM_KEY:
1542 ret = check_csum_item(leaf, key, slot, prev_key);
1543 break;
1544 case BTRFS_DIR_ITEM_KEY:
1545 case BTRFS_DIR_INDEX_KEY:
1546 case BTRFS_XATTR_ITEM_KEY:
1547 ret = check_dir_item(leaf, key, prev_key, slot);
1548 break;
1549 case BTRFS_INODE_REF_KEY:
1550 ret = check_inode_ref(leaf, key, prev_key, slot);
1551 break;
1552 case BTRFS_BLOCK_GROUP_ITEM_KEY:
1553 ret = check_block_group_item(leaf, key, slot);
1554 break;
1555 case BTRFS_CHUNK_ITEM_KEY:
1556 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1557 ret = check_leaf_chunk_item(leaf, chunk, key, slot);
1558 break;
1559 case BTRFS_DEV_ITEM_KEY:
1560 ret = check_dev_item(leaf, key, slot);
1561 break;
1562 case BTRFS_INODE_ITEM_KEY:
1563 ret = check_inode_item(leaf, key, slot);
1564 break;
1565 case BTRFS_ROOT_ITEM_KEY:
1566 ret = check_root_item(leaf, key, slot);
1567 break;
1568 case BTRFS_EXTENT_ITEM_KEY:
1569 case BTRFS_METADATA_ITEM_KEY:
1570 ret = check_extent_item(leaf, key, slot);
1571 break;
1572 case BTRFS_TREE_BLOCK_REF_KEY:
1573 case BTRFS_SHARED_DATA_REF_KEY:
1574 case BTRFS_SHARED_BLOCK_REF_KEY:
1575 ret = check_simple_keyed_refs(leaf, key, slot);
1576 break;
1577 case BTRFS_EXTENT_DATA_REF_KEY:
1578 ret = check_extent_data_ref(leaf, key, slot);
1579 break;
1581 return ret;
1584 static int check_leaf(struct extent_buffer *leaf, bool check_item_data)
1586 struct btrfs_fs_info *fs_info = leaf->fs_info;
1587 /* No valid key type is 0, so all key should be larger than this key */
1588 struct btrfs_key prev_key = {0, 0, 0};
1589 struct btrfs_key key;
1590 u32 nritems = btrfs_header_nritems(leaf);
1591 int slot;
1593 if (unlikely(btrfs_header_level(leaf) != 0)) {
1594 generic_err(leaf, 0,
1595 "invalid level for leaf, have %d expect 0",
1596 btrfs_header_level(leaf));
1597 return -EUCLEAN;
1601 * Extent buffers from a relocation tree have a owner field that
1602 * corresponds to the subvolume tree they are based on. So just from an
1603 * extent buffer alone we can not find out what is the id of the
1604 * corresponding subvolume tree, so we can not figure out if the extent
1605 * buffer corresponds to the root of the relocation tree or not. So
1606 * skip this check for relocation trees.
1608 if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
1609 u64 owner = btrfs_header_owner(leaf);
1611 /* These trees must never be empty */
1612 if (unlikely(owner == BTRFS_ROOT_TREE_OBJECTID ||
1613 owner == BTRFS_CHUNK_TREE_OBJECTID ||
1614 owner == BTRFS_EXTENT_TREE_OBJECTID ||
1615 owner == BTRFS_DEV_TREE_OBJECTID ||
1616 owner == BTRFS_FS_TREE_OBJECTID ||
1617 owner == BTRFS_DATA_RELOC_TREE_OBJECTID)) {
1618 generic_err(leaf, 0,
1619 "invalid root, root %llu must never be empty",
1620 owner);
1621 return -EUCLEAN;
1623 /* Unknown tree */
1624 if (unlikely(owner == 0)) {
1625 generic_err(leaf, 0,
1626 "invalid owner, root 0 is not defined");
1627 return -EUCLEAN;
1629 return 0;
1632 if (unlikely(nritems == 0))
1633 return 0;
1636 * Check the following things to make sure this is a good leaf, and
1637 * leaf users won't need to bother with similar sanity checks:
1639 * 1) key ordering
1640 * 2) item offset and size
1641 * No overlap, no hole, all inside the leaf.
1642 * 3) item content
1643 * If possible, do comprehensive sanity check.
1644 * NOTE: All checks must only rely on the item data itself.
1646 for (slot = 0; slot < nritems; slot++) {
1647 u32 item_end_expected;
1648 int ret;
1650 btrfs_item_key_to_cpu(leaf, &key, slot);
1652 /* Make sure the keys are in the right order */
1653 if (unlikely(btrfs_comp_cpu_keys(&prev_key, &key) >= 0)) {
1654 generic_err(leaf, slot,
1655 "bad key order, prev (%llu %u %llu) current (%llu %u %llu)",
1656 prev_key.objectid, prev_key.type,
1657 prev_key.offset, key.objectid, key.type,
1658 key.offset);
1659 return -EUCLEAN;
1663 * Make sure the offset and ends are right, remember that the
1664 * item data starts at the end of the leaf and grows towards the
1665 * front.
1667 if (slot == 0)
1668 item_end_expected = BTRFS_LEAF_DATA_SIZE(fs_info);
1669 else
1670 item_end_expected = btrfs_item_offset_nr(leaf,
1671 slot - 1);
1672 if (unlikely(btrfs_item_end_nr(leaf, slot) != item_end_expected)) {
1673 generic_err(leaf, slot,
1674 "unexpected item end, have %u expect %u",
1675 btrfs_item_end_nr(leaf, slot),
1676 item_end_expected);
1677 return -EUCLEAN;
1681 * Check to make sure that we don't point outside of the leaf,
1682 * just in case all the items are consistent to each other, but
1683 * all point outside of the leaf.
1685 if (unlikely(btrfs_item_end_nr(leaf, slot) >
1686 BTRFS_LEAF_DATA_SIZE(fs_info))) {
1687 generic_err(leaf, slot,
1688 "slot end outside of leaf, have %u expect range [0, %u]",
1689 btrfs_item_end_nr(leaf, slot),
1690 BTRFS_LEAF_DATA_SIZE(fs_info));
1691 return -EUCLEAN;
1694 /* Also check if the item pointer overlaps with btrfs item. */
1695 if (unlikely(btrfs_item_ptr_offset(leaf, slot) <
1696 btrfs_item_nr_offset(slot) + sizeof(struct btrfs_item))) {
1697 generic_err(leaf, slot,
1698 "slot overlaps with its data, item end %lu data start %lu",
1699 btrfs_item_nr_offset(slot) +
1700 sizeof(struct btrfs_item),
1701 btrfs_item_ptr_offset(leaf, slot));
1702 return -EUCLEAN;
1705 if (check_item_data) {
1707 * Check if the item size and content meet other
1708 * criteria
1710 ret = check_leaf_item(leaf, &key, slot, &prev_key);
1711 if (unlikely(ret < 0))
1712 return ret;
1715 prev_key.objectid = key.objectid;
1716 prev_key.type = key.type;
1717 prev_key.offset = key.offset;
1720 return 0;
1723 int btrfs_check_leaf_full(struct extent_buffer *leaf)
1725 return check_leaf(leaf, true);
1727 ALLOW_ERROR_INJECTION(btrfs_check_leaf_full, ERRNO);
1729 int btrfs_check_leaf_relaxed(struct extent_buffer *leaf)
1731 return check_leaf(leaf, false);
1734 int btrfs_check_node(struct extent_buffer *node)
1736 struct btrfs_fs_info *fs_info = node->fs_info;
1737 unsigned long nr = btrfs_header_nritems(node);
1738 struct btrfs_key key, next_key;
1739 int slot;
1740 int level = btrfs_header_level(node);
1741 u64 bytenr;
1742 int ret = 0;
1744 if (unlikely(level <= 0 || level >= BTRFS_MAX_LEVEL)) {
1745 generic_err(node, 0,
1746 "invalid level for node, have %d expect [1, %d]",
1747 level, BTRFS_MAX_LEVEL - 1);
1748 return -EUCLEAN;
1750 if (unlikely(nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(fs_info))) {
1751 btrfs_crit(fs_info,
1752 "corrupt node: root=%llu block=%llu, nritems too %s, have %lu expect range [1,%u]",
1753 btrfs_header_owner(node), node->start,
1754 nr == 0 ? "small" : "large", nr,
1755 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1756 return -EUCLEAN;
1759 for (slot = 0; slot < nr - 1; slot++) {
1760 bytenr = btrfs_node_blockptr(node, slot);
1761 btrfs_node_key_to_cpu(node, &key, slot);
1762 btrfs_node_key_to_cpu(node, &next_key, slot + 1);
1764 if (unlikely(!bytenr)) {
1765 generic_err(node, slot,
1766 "invalid NULL node pointer");
1767 ret = -EUCLEAN;
1768 goto out;
1770 if (unlikely(!IS_ALIGNED(bytenr, fs_info->sectorsize))) {
1771 generic_err(node, slot,
1772 "unaligned pointer, have %llu should be aligned to %u",
1773 bytenr, fs_info->sectorsize);
1774 ret = -EUCLEAN;
1775 goto out;
1778 if (unlikely(btrfs_comp_cpu_keys(&key, &next_key) >= 0)) {
1779 generic_err(node, slot,
1780 "bad key order, current (%llu %u %llu) next (%llu %u %llu)",
1781 key.objectid, key.type, key.offset,
1782 next_key.objectid, next_key.type,
1783 next_key.offset);
1784 ret = -EUCLEAN;
1785 goto out;
1788 out:
1789 return ret;
1791 ALLOW_ERROR_INJECTION(btrfs_check_node, ERRNO);