1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) Qu Wenruo 2017.  All rights reserved.
4  */
5 
6 /*
7  * The module is used to catch unexpected/corrupted tree block data.
8  * Such behavior can be caused either by a fuzzed image or bugs.
9  *
10  * The objective is to do leaf/node validation checks when tree block is read
11  * from disk, and check *every* possible member, so other code won't
12  * need to checking them again.
13  *
14  * Due to the potential and unwanted damage, every checker needs to be
15  * carefully reviewed otherwise so it does not prevent mount of valid images.
16  */
17 
18 #include "ctree.h"
19 #include "tree-checker.h"
20 #include "disk-io.h"
21 #include "compression.h"
22 #include "volumes.h"
23 
24 /*
25  * Error message should follow the following format:
26  * corrupt <type>: <identifier>, <reason>[, <bad_value>]
27  *
28  * @type:	leaf or node
29  * @identifier:	the necessary info to locate the leaf/node.
30  * 		It's recommened to decode key.objecitd/offset if it's
31  * 		meaningful.
32  * @reason:	describe the error
33  * @bad_value:	optional, it's recommened to output bad value and its
34  *		expected value (range).
35  *
36  * Since comma is used to separate the components, only space is allowed
37  * inside each component.
38  */
39 
40 /*
41  * Append generic "corrupt leaf/node root=%llu block=%llu slot=%d: " to @fmt.
42  * Allows callers to customize the output.
43  */
44 __printf(4, 5)
45 __cold
generic_err(const struct btrfs_fs_info * fs_info,const struct extent_buffer * eb,int slot,const char * fmt,...)46 static void generic_err(const struct btrfs_fs_info *fs_info,
47 			const struct extent_buffer *eb, int slot,
48 			const char *fmt, ...)
49 {
50 	struct va_format vaf;
51 	va_list args;
52 
53 	va_start(args, fmt);
54 
55 	vaf.fmt = fmt;
56 	vaf.va = &args;
57 
58 	btrfs_crit(fs_info,
59 		"corrupt %s: root=%llu block=%llu slot=%d, %pV",
60 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
61 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot, &vaf);
62 	va_end(args);
63 }
64 
65 /*
66  * Customized reporter for extent data item, since its key objectid and
67  * offset has its own meaning.
68  */
69 __printf(4, 5)
70 __cold
file_extent_err(const struct btrfs_fs_info * fs_info,const struct extent_buffer * eb,int slot,const char * fmt,...)71 static void file_extent_err(const struct btrfs_fs_info *fs_info,
72 			    const struct extent_buffer *eb, int slot,
73 			    const char *fmt, ...)
74 {
75 	struct btrfs_key key;
76 	struct va_format vaf;
77 	va_list args;
78 
79 	btrfs_item_key_to_cpu(eb, &key, slot);
80 	va_start(args, fmt);
81 
82 	vaf.fmt = fmt;
83 	vaf.va = &args;
84 
85 	btrfs_crit(fs_info,
86 	"corrupt %s: root=%llu block=%llu slot=%d ino=%llu file_offset=%llu, %pV",
87 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
88 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
89 		key.objectid, key.offset, &vaf);
90 	va_end(args);
91 }
92 
93 /*
94  * Return 0 if the btrfs_file_extent_##name is aligned to @alignment
95  * Else return 1
96  */
97 #define CHECK_FE_ALIGNED(fs_info, leaf, slot, fi, name, alignment)	      \
98 ({									      \
99 	if (!IS_ALIGNED(btrfs_file_extent_##name((leaf), (fi)), (alignment))) \
100 		file_extent_err((fs_info), (leaf), (slot),		      \
101 	"invalid %s for file extent, have %llu, should be aligned to %u",     \
102 			(#name), btrfs_file_extent_##name((leaf), (fi)),      \
103 			(alignment));					      \
104 	(!IS_ALIGNED(btrfs_file_extent_##name((leaf), (fi)), (alignment)));   \
105 })
106 
check_extent_data_item(struct btrfs_fs_info * fs_info,struct extent_buffer * leaf,struct btrfs_key * key,int slot)107 static int check_extent_data_item(struct btrfs_fs_info *fs_info,
108 				  struct extent_buffer *leaf,
109 				  struct btrfs_key *key, int slot)
110 {
111 	struct btrfs_file_extent_item *fi;
112 	u32 sectorsize = fs_info->sectorsize;
113 	u32 item_size = btrfs_item_size_nr(leaf, slot);
114 
115 	if (!IS_ALIGNED(key->offset, sectorsize)) {
116 		file_extent_err(fs_info, leaf, slot,
117 "unaligned file_offset for file extent, have %llu should be aligned to %u",
118 			key->offset, sectorsize);
119 		return -EUCLEAN;
120 	}
121 
122 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
123 
124 	if (btrfs_file_extent_type(leaf, fi) > BTRFS_FILE_EXTENT_TYPES) {
125 		file_extent_err(fs_info, leaf, slot,
126 		"invalid type for file extent, have %u expect range [0, %u]",
127 			btrfs_file_extent_type(leaf, fi),
128 			BTRFS_FILE_EXTENT_TYPES);
129 		return -EUCLEAN;
130 	}
131 
132 	/*
133 	 * Support for new compression/encrption must introduce incompat flag,
134 	 * and must be caught in open_ctree().
135 	 */
136 	if (btrfs_file_extent_compression(leaf, fi) > BTRFS_COMPRESS_TYPES) {
137 		file_extent_err(fs_info, leaf, slot,
138 	"invalid compression for file extent, have %u expect range [0, %u]",
139 			btrfs_file_extent_compression(leaf, fi),
140 			BTRFS_COMPRESS_TYPES);
141 		return -EUCLEAN;
142 	}
143 	if (btrfs_file_extent_encryption(leaf, fi)) {
144 		file_extent_err(fs_info, leaf, slot,
145 			"invalid encryption for file extent, have %u expect 0",
146 			btrfs_file_extent_encryption(leaf, fi));
147 		return -EUCLEAN;
148 	}
149 	if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
150 		/* Inline extent must have 0 as key offset */
151 		if (key->offset) {
152 			file_extent_err(fs_info, leaf, slot,
153 		"invalid file_offset for inline file extent, have %llu expect 0",
154 				key->offset);
155 			return -EUCLEAN;
156 		}
157 
158 		/* Compressed inline extent has no on-disk size, skip it */
159 		if (btrfs_file_extent_compression(leaf, fi) !=
160 		    BTRFS_COMPRESS_NONE)
161 			return 0;
162 
163 		/* Uncompressed inline extent size must match item size */
164 		if (item_size != BTRFS_FILE_EXTENT_INLINE_DATA_START +
165 		    btrfs_file_extent_ram_bytes(leaf, fi)) {
166 			file_extent_err(fs_info, leaf, slot,
167 	"invalid ram_bytes for uncompressed inline extent, have %u expect %llu",
168 				item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START +
169 				btrfs_file_extent_ram_bytes(leaf, fi));
170 			return -EUCLEAN;
171 		}
172 		return 0;
173 	}
174 
175 	/* Regular or preallocated extent has fixed item size */
176 	if (item_size != sizeof(*fi)) {
177 		file_extent_err(fs_info, leaf, slot,
178 	"invalid item size for reg/prealloc file extent, have %u expect %zu",
179 			item_size, sizeof(*fi));
180 		return -EUCLEAN;
181 	}
182 	if (CHECK_FE_ALIGNED(fs_info, leaf, slot, fi, ram_bytes, sectorsize) ||
183 	    CHECK_FE_ALIGNED(fs_info, leaf, slot, fi, disk_bytenr, sectorsize) ||
184 	    CHECK_FE_ALIGNED(fs_info, leaf, slot, fi, disk_num_bytes, sectorsize) ||
185 	    CHECK_FE_ALIGNED(fs_info, leaf, slot, fi, offset, sectorsize) ||
186 	    CHECK_FE_ALIGNED(fs_info, leaf, slot, fi, num_bytes, sectorsize))
187 		return -EUCLEAN;
188 	return 0;
189 }
190 
check_csum_item(struct btrfs_fs_info * fs_info,struct extent_buffer * leaf,struct btrfs_key * key,int slot)191 static int check_csum_item(struct btrfs_fs_info *fs_info,
192 			   struct extent_buffer *leaf, struct btrfs_key *key,
193 			   int slot)
194 {
195 	u32 sectorsize = fs_info->sectorsize;
196 	u32 csumsize = btrfs_super_csum_size(fs_info->super_copy);
197 
198 	if (key->objectid != BTRFS_EXTENT_CSUM_OBJECTID) {
199 		generic_err(fs_info, leaf, slot,
200 		"invalid key objectid for csum item, have %llu expect %llu",
201 			key->objectid, BTRFS_EXTENT_CSUM_OBJECTID);
202 		return -EUCLEAN;
203 	}
204 	if (!IS_ALIGNED(key->offset, sectorsize)) {
205 		generic_err(fs_info, leaf, slot,
206 	"unaligned key offset for csum item, have %llu should be aligned to %u",
207 			key->offset, sectorsize);
208 		return -EUCLEAN;
209 	}
210 	if (!IS_ALIGNED(btrfs_item_size_nr(leaf, slot), csumsize)) {
211 		generic_err(fs_info, leaf, slot,
212 	"unaligned item size for csum item, have %u should be aligned to %u",
213 			btrfs_item_size_nr(leaf, slot), csumsize);
214 		return -EUCLEAN;
215 	}
216 	return 0;
217 }
218 
219 /*
220  * Customized reported for dir_item, only important new info is key->objectid,
221  * which represents inode number
222  */
223 __printf(4, 5)
224 __cold
dir_item_err(const struct btrfs_fs_info * fs_info,const struct extent_buffer * eb,int slot,const char * fmt,...)225 static void dir_item_err(const struct btrfs_fs_info *fs_info,
226 			 const struct extent_buffer *eb, int slot,
227 			 const char *fmt, ...)
228 {
229 	struct btrfs_key key;
230 	struct va_format vaf;
231 	va_list args;
232 
233 	btrfs_item_key_to_cpu(eb, &key, slot);
234 	va_start(args, fmt);
235 
236 	vaf.fmt = fmt;
237 	vaf.va = &args;
238 
239 	btrfs_crit(fs_info,
240 	"corrupt %s: root=%llu block=%llu slot=%d ino=%llu, %pV",
241 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
242 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
243 		key.objectid, &vaf);
244 	va_end(args);
245 }
246 
check_dir_item(struct btrfs_fs_info * fs_info,struct extent_buffer * leaf,struct btrfs_key * key,int slot)247 static int check_dir_item(struct btrfs_fs_info *fs_info,
248 			  struct extent_buffer *leaf,
249 			  struct btrfs_key *key, int slot)
250 {
251 	struct btrfs_dir_item *di;
252 	u32 item_size = btrfs_item_size_nr(leaf, slot);
253 	u32 cur = 0;
254 
255 	di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
256 	while (cur < item_size) {
257 		u32 name_len;
258 		u32 data_len;
259 		u32 max_name_len;
260 		u32 total_size;
261 		u32 name_hash;
262 		u8 dir_type;
263 
264 		/* header itself should not cross item boundary */
265 		if (cur + sizeof(*di) > item_size) {
266 			dir_item_err(fs_info, leaf, slot,
267 		"dir item header crosses item boundary, have %zu boundary %u",
268 				cur + sizeof(*di), item_size);
269 			return -EUCLEAN;
270 		}
271 
272 		/* dir type check */
273 		dir_type = btrfs_dir_type(leaf, di);
274 		if (dir_type >= BTRFS_FT_MAX) {
275 			dir_item_err(fs_info, leaf, slot,
276 			"invalid dir item type, have %u expect [0, %u)",
277 				dir_type, BTRFS_FT_MAX);
278 			return -EUCLEAN;
279 		}
280 
281 		if (key->type == BTRFS_XATTR_ITEM_KEY &&
282 		    dir_type != BTRFS_FT_XATTR) {
283 			dir_item_err(fs_info, leaf, slot,
284 		"invalid dir item type for XATTR key, have %u expect %u",
285 				dir_type, BTRFS_FT_XATTR);
286 			return -EUCLEAN;
287 		}
288 		if (dir_type == BTRFS_FT_XATTR &&
289 		    key->type != BTRFS_XATTR_ITEM_KEY) {
290 			dir_item_err(fs_info, leaf, slot,
291 			"xattr dir type found for non-XATTR key");
292 			return -EUCLEAN;
293 		}
294 		if (dir_type == BTRFS_FT_XATTR)
295 			max_name_len = XATTR_NAME_MAX;
296 		else
297 			max_name_len = BTRFS_NAME_LEN;
298 
299 		/* Name/data length check */
300 		name_len = btrfs_dir_name_len(leaf, di);
301 		data_len = btrfs_dir_data_len(leaf, di);
302 		if (name_len > max_name_len) {
303 			dir_item_err(fs_info, leaf, slot,
304 			"dir item name len too long, have %u max %u",
305 				name_len, max_name_len);
306 			return -EUCLEAN;
307 		}
308 		if (name_len + data_len > BTRFS_MAX_XATTR_SIZE(fs_info)) {
309 			dir_item_err(fs_info, leaf, slot,
310 			"dir item name and data len too long, have %u max %u",
311 				name_len + data_len,
312 				BTRFS_MAX_XATTR_SIZE(fs_info));
313 			return -EUCLEAN;
314 		}
315 
316 		if (data_len && dir_type != BTRFS_FT_XATTR) {
317 			dir_item_err(fs_info, leaf, slot,
318 			"dir item with invalid data len, have %u expect 0",
319 				data_len);
320 			return -EUCLEAN;
321 		}
322 
323 		total_size = sizeof(*di) + name_len + data_len;
324 
325 		/* header and name/data should not cross item boundary */
326 		if (cur + total_size > item_size) {
327 			dir_item_err(fs_info, leaf, slot,
328 		"dir item data crosses item boundary, have %u boundary %u",
329 				cur + total_size, item_size);
330 			return -EUCLEAN;
331 		}
332 
333 		/*
334 		 * Special check for XATTR/DIR_ITEM, as key->offset is name
335 		 * hash, should match its name
336 		 */
337 		if (key->type == BTRFS_DIR_ITEM_KEY ||
338 		    key->type == BTRFS_XATTR_ITEM_KEY) {
339 			char namebuf[max(BTRFS_NAME_LEN, XATTR_NAME_MAX)];
340 
341 			read_extent_buffer(leaf, namebuf,
342 					(unsigned long)(di + 1), name_len);
343 			name_hash = btrfs_name_hash(namebuf, name_len);
344 			if (key->offset != name_hash) {
345 				dir_item_err(fs_info, leaf, slot,
346 		"name hash mismatch with key, have 0x%016x expect 0x%016llx",
347 					name_hash, key->offset);
348 				return -EUCLEAN;
349 			}
350 		}
351 		cur += total_size;
352 		di = (struct btrfs_dir_item *)((void *)di + total_size);
353 	}
354 	return 0;
355 }
356 
357 __printf(4, 5)
358 __cold
block_group_err(const struct btrfs_fs_info * fs_info,const struct extent_buffer * eb,int slot,const char * fmt,...)359 static void block_group_err(const struct btrfs_fs_info *fs_info,
360 			    const struct extent_buffer *eb, int slot,
361 			    const char *fmt, ...)
362 {
363 	struct btrfs_key key;
364 	struct va_format vaf;
365 	va_list args;
366 
367 	btrfs_item_key_to_cpu(eb, &key, slot);
368 	va_start(args, fmt);
369 
370 	vaf.fmt = fmt;
371 	vaf.va = &args;
372 
373 	btrfs_crit(fs_info,
374 	"corrupt %s: root=%llu block=%llu slot=%d bg_start=%llu bg_len=%llu, %pV",
375 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
376 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
377 		key.objectid, key.offset, &vaf);
378 	va_end(args);
379 }
380 
check_block_group_item(struct btrfs_fs_info * fs_info,struct extent_buffer * leaf,struct btrfs_key * key,int slot)381 static int check_block_group_item(struct btrfs_fs_info *fs_info,
382 				  struct extent_buffer *leaf,
383 				  struct btrfs_key *key, int slot)
384 {
385 	struct btrfs_block_group_item bgi;
386 	u32 item_size = btrfs_item_size_nr(leaf, slot);
387 	u64 flags;
388 	u64 type;
389 
390 	/*
391 	 * Here we don't really care about alignment since extent allocator can
392 	 * handle it.  We care more about the size.
393 	 */
394 	if (key->offset == 0) {
395 		block_group_err(fs_info, leaf, slot,
396 				"invalid block group size 0");
397 		return -EUCLEAN;
398 	}
399 
400 	if (item_size != sizeof(bgi)) {
401 		block_group_err(fs_info, leaf, slot,
402 			"invalid item size, have %u expect %zu",
403 				item_size, sizeof(bgi));
404 		return -EUCLEAN;
405 	}
406 
407 	read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot),
408 			   sizeof(bgi));
409 	if (btrfs_block_group_chunk_objectid(&bgi) !=
410 	    BTRFS_FIRST_CHUNK_TREE_OBJECTID) {
411 		block_group_err(fs_info, leaf, slot,
412 		"invalid block group chunk objectid, have %llu expect %llu",
413 				btrfs_block_group_chunk_objectid(&bgi),
414 				BTRFS_FIRST_CHUNK_TREE_OBJECTID);
415 		return -EUCLEAN;
416 	}
417 
418 	if (btrfs_block_group_used(&bgi) > key->offset) {
419 		block_group_err(fs_info, leaf, slot,
420 			"invalid block group used, have %llu expect [0, %llu)",
421 				btrfs_block_group_used(&bgi), key->offset);
422 		return -EUCLEAN;
423 	}
424 
425 	flags = btrfs_block_group_flags(&bgi);
426 	if (hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) > 1) {
427 		block_group_err(fs_info, leaf, slot,
428 "invalid profile flags, have 0x%llx (%lu bits set) expect no more than 1 bit set",
429 			flags & BTRFS_BLOCK_GROUP_PROFILE_MASK,
430 			hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK));
431 		return -EUCLEAN;
432 	}
433 
434 	type = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
435 	if (type != BTRFS_BLOCK_GROUP_DATA &&
436 	    type != BTRFS_BLOCK_GROUP_METADATA &&
437 	    type != BTRFS_BLOCK_GROUP_SYSTEM &&
438 	    type != (BTRFS_BLOCK_GROUP_METADATA |
439 			   BTRFS_BLOCK_GROUP_DATA)) {
440 		block_group_err(fs_info, leaf, slot,
441 "invalid type, have 0x%llx (%lu bits set) expect either 0x%llx, 0x%llx, 0x%llx or 0x%llx",
442 			type, hweight64(type),
443 			BTRFS_BLOCK_GROUP_DATA, BTRFS_BLOCK_GROUP_METADATA,
444 			BTRFS_BLOCK_GROUP_SYSTEM,
445 			BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA);
446 		return -EUCLEAN;
447 	}
448 	return 0;
449 }
450 
451 __printf(5, 6)
452 __cold
chunk_err(const struct btrfs_fs_info * fs_info,const struct extent_buffer * leaf,const struct btrfs_chunk * chunk,u64 logical,const char * fmt,...)453 static void chunk_err(const struct btrfs_fs_info *fs_info,
454 		      const struct extent_buffer *leaf,
455 		      const struct btrfs_chunk *chunk, u64 logical,
456 		      const char *fmt, ...)
457 {
458 	bool is_sb;
459 	struct va_format vaf;
460 	va_list args;
461 	int i;
462 	int slot = -1;
463 
464 	/* Only superblock eb is able to have such small offset */
465 	is_sb = (leaf->start == BTRFS_SUPER_INFO_OFFSET);
466 
467 	if (!is_sb) {
468 		/*
469 		 * Get the slot number by iterating through all slots, this
470 		 * would provide better readability.
471 		 */
472 		for (i = 0; i < btrfs_header_nritems(leaf); i++) {
473 			if (btrfs_item_ptr_offset(leaf, i) ==
474 					(unsigned long)chunk) {
475 				slot = i;
476 				break;
477 			}
478 		}
479 	}
480 	va_start(args, fmt);
481 	vaf.fmt = fmt;
482 	vaf.va = &args;
483 
484 	if (is_sb)
485 		btrfs_crit(fs_info,
486 		"corrupt superblock syschunk array: chunk_start=%llu, %pV",
487 			   logical, &vaf);
488 	else
489 		btrfs_crit(fs_info,
490 	"corrupt leaf: root=%llu block=%llu slot=%d chunk_start=%llu, %pV",
491 			   BTRFS_CHUNK_TREE_OBJECTID, leaf->start, slot,
492 			   logical, &vaf);
493 	va_end(args);
494 }
495 
496 /*
497  * The common chunk check which could also work on super block sys chunk array.
498  *
499  * Return -EUCLEAN if anything is corrupted.
500  * Return 0 if everything is OK.
501  */
btrfs_check_chunk_valid(struct btrfs_fs_info * fs_info,struct extent_buffer * leaf,struct btrfs_chunk * chunk,u64 logical)502 int btrfs_check_chunk_valid(struct btrfs_fs_info *fs_info,
503 			    struct extent_buffer *leaf,
504 			    struct btrfs_chunk *chunk, u64 logical)
505 {
506 	u64 length;
507 	u64 stripe_len;
508 	u16 num_stripes;
509 	u16 sub_stripes;
510 	u64 type;
511 	u64 features;
512 	bool mixed = false;
513 
514 	length = btrfs_chunk_length(leaf, chunk);
515 	stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
516 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
517 	sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
518 	type = btrfs_chunk_type(leaf, chunk);
519 
520 	if (!num_stripes) {
521 		chunk_err(fs_info, leaf, chunk, logical,
522 			  "invalid chunk num_stripes, have %u", num_stripes);
523 		return -EUCLEAN;
524 	}
525 	if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
526 		chunk_err(fs_info, leaf, chunk, logical,
527 		"invalid chunk logical, have %llu should aligned to %u",
528 			  logical, fs_info->sectorsize);
529 		return -EUCLEAN;
530 	}
531 	if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
532 		chunk_err(fs_info, leaf, chunk, logical,
533 			  "invalid chunk sectorsize, have %u expect %u",
534 			  btrfs_chunk_sector_size(leaf, chunk),
535 			  fs_info->sectorsize);
536 		return -EUCLEAN;
537 	}
538 	if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
539 		chunk_err(fs_info, leaf, chunk, logical,
540 			  "invalid chunk length, have %llu", length);
541 		return -EUCLEAN;
542 	}
543 	if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
544 		chunk_err(fs_info, leaf, chunk, logical,
545 			  "invalid chunk stripe length: %llu",
546 			  stripe_len);
547 		return -EUCLEAN;
548 	}
549 	if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
550 	    type) {
551 		chunk_err(fs_info, leaf, chunk, logical,
552 			  "unrecognized chunk type: 0x%llx",
553 			  ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
554 			    BTRFS_BLOCK_GROUP_PROFILE_MASK) &
555 			  btrfs_chunk_type(leaf, chunk));
556 		return -EUCLEAN;
557 	}
558 
559 	if (!is_power_of_2(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
560 	    (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) != 0) {
561 		chunk_err(fs_info, leaf, chunk, logical,
562 		"invalid chunk profile flag: 0x%llx, expect 0 or 1 bit set",
563 			  type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
564 		return -EUCLEAN;
565 	}
566 	if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0) {
567 		chunk_err(fs_info, leaf, chunk, logical,
568 	"missing chunk type flag, have 0x%llx one bit must be set in 0x%llx",
569 			  type, BTRFS_BLOCK_GROUP_TYPE_MASK);
570 		return -EUCLEAN;
571 	}
572 
573 	if ((type & BTRFS_BLOCK_GROUP_SYSTEM) &&
574 	    (type & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA))) {
575 		chunk_err(fs_info, leaf, chunk, logical,
576 			  "system chunk with data or metadata type: 0x%llx",
577 			  type);
578 		return -EUCLEAN;
579 	}
580 
581 	features = btrfs_super_incompat_flags(fs_info->super_copy);
582 	if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
583 		mixed = true;
584 
585 	if (!mixed) {
586 		if ((type & BTRFS_BLOCK_GROUP_METADATA) &&
587 		    (type & BTRFS_BLOCK_GROUP_DATA)) {
588 			chunk_err(fs_info, leaf, chunk, logical,
589 			"mixed chunk type in non-mixed mode: 0x%llx", type);
590 			return -EUCLEAN;
591 		}
592 	}
593 
594 	if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
595 	    (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes != 2) ||
596 	    (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
597 	    (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
598 	    (type & BTRFS_BLOCK_GROUP_DUP && num_stripes != 2) ||
599 	    ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 && num_stripes != 1)) {
600 		chunk_err(fs_info, leaf, chunk, logical,
601 			"invalid num_stripes:sub_stripes %u:%u for profile %llu",
602 			num_stripes, sub_stripes,
603 			type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
604 		return -EUCLEAN;
605 	}
606 
607 	return 0;
608 }
609 
610 __printf(4, 5)
611 __cold
dev_item_err(const struct btrfs_fs_info * fs_info,const struct extent_buffer * eb,int slot,const char * fmt,...)612 static void dev_item_err(const struct btrfs_fs_info *fs_info,
613 			 const struct extent_buffer *eb, int slot,
614 			 const char *fmt, ...)
615 {
616 	struct btrfs_key key;
617 	struct va_format vaf;
618 	va_list args;
619 
620 	btrfs_item_key_to_cpu(eb, &key, slot);
621 	va_start(args, fmt);
622 
623 	vaf.fmt = fmt;
624 	vaf.va = &args;
625 
626 	btrfs_crit(fs_info,
627 	"corrupt %s: root=%llu block=%llu slot=%d devid=%llu %pV",
628 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
629 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
630 		key.objectid, &vaf);
631 	va_end(args);
632 }
633 
check_dev_item(struct btrfs_fs_info * fs_info,struct extent_buffer * leaf,struct btrfs_key * key,int slot)634 static int check_dev_item(struct btrfs_fs_info *fs_info,
635 			  struct extent_buffer *leaf,
636 			  struct btrfs_key *key, int slot)
637 {
638 	struct btrfs_dev_item *ditem;
639 
640 	if (key->objectid != BTRFS_DEV_ITEMS_OBJECTID) {
641 		dev_item_err(fs_info, leaf, slot,
642 			     "invalid objectid: has=%llu expect=%llu",
643 			     key->objectid, BTRFS_DEV_ITEMS_OBJECTID);
644 		return -EUCLEAN;
645 	}
646 	ditem = btrfs_item_ptr(leaf, slot, struct btrfs_dev_item);
647 	if (btrfs_device_id(leaf, ditem) != key->offset) {
648 		dev_item_err(fs_info, leaf, slot,
649 			     "devid mismatch: key has=%llu item has=%llu",
650 			     key->offset, btrfs_device_id(leaf, ditem));
651 		return -EUCLEAN;
652 	}
653 
654 	/*
655 	 * For device total_bytes, we don't have reliable way to check it, as
656 	 * it can be 0 for device removal. Device size check can only be done
657 	 * by dev extents check.
658 	 */
659 	if (btrfs_device_bytes_used(leaf, ditem) >
660 	    btrfs_device_total_bytes(leaf, ditem)) {
661 		dev_item_err(fs_info, leaf, slot,
662 			     "invalid bytes used: have %llu expect [0, %llu]",
663 			     btrfs_device_bytes_used(leaf, ditem),
664 			     btrfs_device_total_bytes(leaf, ditem));
665 		return -EUCLEAN;
666 	}
667 	/*
668 	 * Remaining members like io_align/type/gen/dev_group aren't really
669 	 * utilized.  Skip them to make later usage of them easier.
670 	 */
671 	return 0;
672 }
673 
674 /* Inode item error output has the same format as dir_item_err() */
675 #define inode_item_err(fs_info, eb, slot, fmt, ...)			\
676 	dir_item_err(fs_info, eb, slot, fmt, __VA_ARGS__)
677 
check_inode_item(struct btrfs_fs_info * fs_info,struct extent_buffer * leaf,struct btrfs_key * key,int slot)678 static int check_inode_item(struct btrfs_fs_info *fs_info,
679 			    struct extent_buffer *leaf,
680 			    struct btrfs_key *key, int slot)
681 {
682 	struct btrfs_inode_item *iitem;
683 	u64 super_gen = btrfs_super_generation(fs_info->super_copy);
684 	u32 valid_mask = (S_IFMT | S_ISUID | S_ISGID | S_ISVTX | 0777);
685 	u32 mode;
686 
687 	if ((key->objectid < BTRFS_FIRST_FREE_OBJECTID ||
688 	     key->objectid > BTRFS_LAST_FREE_OBJECTID) &&
689 	    key->objectid != BTRFS_ROOT_TREE_DIR_OBJECTID &&
690 	    key->objectid != BTRFS_FREE_INO_OBJECTID) {
691 		generic_err(fs_info, leaf, slot,
692 	"invalid key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
693 			    key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
694 			    BTRFS_FIRST_FREE_OBJECTID,
695 			    BTRFS_LAST_FREE_OBJECTID,
696 			    BTRFS_FREE_INO_OBJECTID);
697 		return -EUCLEAN;
698 	}
699 	if (key->offset != 0) {
700 		inode_item_err(fs_info, leaf, slot,
701 			"invalid key offset: has %llu expect 0",
702 			key->offset);
703 		return -EUCLEAN;
704 	}
705 	iitem = btrfs_item_ptr(leaf, slot, struct btrfs_inode_item);
706 
707 	/* Here we use super block generation + 1 to handle log tree */
708 	if (btrfs_inode_generation(leaf, iitem) > super_gen + 1) {
709 		inode_item_err(fs_info, leaf, slot,
710 			"invalid inode generation: has %llu expect (0, %llu]",
711 			       btrfs_inode_generation(leaf, iitem),
712 			       super_gen + 1);
713 		return -EUCLEAN;
714 	}
715 	/* Note for ROOT_TREE_DIR_ITEM, mkfs could set its transid 0 */
716 	if (btrfs_inode_transid(leaf, iitem) > super_gen + 1) {
717 		inode_item_err(fs_info, leaf, slot,
718 			"invalid inode transid: has %llu expect [0, %llu]",
719 			       btrfs_inode_transid(leaf, iitem), super_gen + 1);
720 		return -EUCLEAN;
721 	}
722 
723 	/*
724 	 * For size and nbytes it's better not to be too strict, as for dir
725 	 * item its size/nbytes can easily get wrong, but doesn't affect
726 	 * anything in the fs. So here we skip the check.
727 	 */
728 	mode = btrfs_inode_mode(leaf, iitem);
729 	if (mode & ~valid_mask) {
730 		inode_item_err(fs_info, leaf, slot,
731 			       "unknown mode bit detected: 0x%x",
732 			       mode & ~valid_mask);
733 		return -EUCLEAN;
734 	}
735 
736 	/*
737 	 * S_IFMT is not bit mapped so we can't completely rely on is_power_of_2,
738 	 * but is_power_of_2() can save us from checking FIFO/CHR/DIR/REG.
739 	 * Only needs to check BLK, LNK and SOCKS
740 	 */
741 	if (!is_power_of_2(mode & S_IFMT)) {
742 		if (!S_ISLNK(mode) && !S_ISBLK(mode) && !S_ISSOCK(mode)) {
743 			inode_item_err(fs_info, leaf, slot,
744 			"invalid mode: has 0%o expect valid S_IF* bit(s)",
745 				       mode & S_IFMT);
746 			return -EUCLEAN;
747 		}
748 	}
749 	if (S_ISDIR(mode) && btrfs_inode_nlink(leaf, iitem) > 1) {
750 		inode_item_err(fs_info, leaf, slot,
751 		       "invalid nlink: has %u expect no more than 1 for dir",
752 			btrfs_inode_nlink(leaf, iitem));
753 		return -EUCLEAN;
754 	}
755 	if (btrfs_inode_flags(leaf, iitem) & ~BTRFS_INODE_FLAG_MASK) {
756 		inode_item_err(fs_info, leaf, slot,
757 			       "unknown flags detected: 0x%llx",
758 			       btrfs_inode_flags(leaf, iitem) &
759 			       ~BTRFS_INODE_FLAG_MASK);
760 		return -EUCLEAN;
761 	}
762 	return 0;
763 }
764 
765 /*
766  * Common point to switch the item-specific validation.
767  */
check_leaf_item(struct btrfs_fs_info * fs_info,struct extent_buffer * leaf,struct btrfs_key * key,int slot)768 static int check_leaf_item(struct btrfs_fs_info *fs_info,
769 			   struct extent_buffer *leaf,
770 			   struct btrfs_key *key, int slot)
771 {
772 	int ret = 0;
773 	struct btrfs_chunk *chunk;
774 
775 	switch (key->type) {
776 	case BTRFS_EXTENT_DATA_KEY:
777 		ret = check_extent_data_item(fs_info, leaf, key, slot);
778 		break;
779 	case BTRFS_EXTENT_CSUM_KEY:
780 		ret = check_csum_item(fs_info, leaf, key, slot);
781 		break;
782 	case BTRFS_DIR_ITEM_KEY:
783 	case BTRFS_DIR_INDEX_KEY:
784 	case BTRFS_XATTR_ITEM_KEY:
785 		ret = check_dir_item(fs_info, leaf, key, slot);
786 		break;
787 	case BTRFS_BLOCK_GROUP_ITEM_KEY:
788 		ret = check_block_group_item(fs_info, leaf, key, slot);
789 		break;
790 	case BTRFS_CHUNK_ITEM_KEY:
791 		chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
792 		ret = btrfs_check_chunk_valid(fs_info, leaf, chunk,
793 					      key->offset);
794 		break;
795 	case BTRFS_DEV_ITEM_KEY:
796 		ret = check_dev_item(fs_info, leaf, key, slot);
797 		break;
798 	case BTRFS_INODE_ITEM_KEY:
799 		ret = check_inode_item(fs_info, leaf, key, slot);
800 		break;
801 	}
802 	return ret;
803 }
804 
check_leaf(struct btrfs_fs_info * fs_info,struct extent_buffer * leaf,bool check_item_data)805 static int check_leaf(struct btrfs_fs_info *fs_info, struct extent_buffer *leaf,
806 		      bool check_item_data)
807 {
808 	/* No valid key type is 0, so all key should be larger than this key */
809 	struct btrfs_key prev_key = {0, 0, 0};
810 	struct btrfs_key key;
811 	u32 nritems = btrfs_header_nritems(leaf);
812 	int slot;
813 
814 	if (btrfs_header_level(leaf) != 0) {
815 		generic_err(fs_info, leaf, 0,
816 			"invalid level for leaf, have %d expect 0",
817 			btrfs_header_level(leaf));
818 		return -EUCLEAN;
819 	}
820 
821 	/*
822 	 * Extent buffers from a relocation tree have a owner field that
823 	 * corresponds to the subvolume tree they are based on. So just from an
824 	 * extent buffer alone we can not find out what is the id of the
825 	 * corresponding subvolume tree, so we can not figure out if the extent
826 	 * buffer corresponds to the root of the relocation tree or not. So
827 	 * skip this check for relocation trees.
828 	 */
829 	if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
830 		u64 owner = btrfs_header_owner(leaf);
831 		struct btrfs_root *check_root;
832 
833 		/* These trees must never be empty */
834 		if (owner == BTRFS_ROOT_TREE_OBJECTID ||
835 		    owner == BTRFS_CHUNK_TREE_OBJECTID ||
836 		    owner == BTRFS_EXTENT_TREE_OBJECTID ||
837 		    owner == BTRFS_DEV_TREE_OBJECTID ||
838 		    owner == BTRFS_FS_TREE_OBJECTID ||
839 		    owner == BTRFS_DATA_RELOC_TREE_OBJECTID) {
840 			generic_err(fs_info, leaf, 0,
841 			"invalid root, root %llu must never be empty",
842 				    owner);
843 			return -EUCLEAN;
844 		}
845 		/* Unknown tree */
846 		if (owner == 0) {
847 			generic_err(fs_info, leaf, 0,
848 				"invalid owner, root 0 is not defined");
849 			return -EUCLEAN;
850 		}
851 		key.objectid = owner;
852 		key.type = BTRFS_ROOT_ITEM_KEY;
853 		key.offset = (u64)-1;
854 
855 		check_root = btrfs_get_fs_root(fs_info, &key, false);
856 		/*
857 		 * The only reason we also check NULL here is that during
858 		 * open_ctree() some roots has not yet been set up.
859 		 */
860 		if (!IS_ERR_OR_NULL(check_root)) {
861 			struct extent_buffer *eb;
862 
863 			eb = btrfs_root_node(check_root);
864 			/* if leaf is the root, then it's fine */
865 			if (leaf != eb) {
866 				generic_err(fs_info, leaf, 0,
867 		"invalid nritems, have %u should not be 0 for non-root leaf",
868 					nritems);
869 				free_extent_buffer(eb);
870 				return -EUCLEAN;
871 			}
872 			free_extent_buffer(eb);
873 		}
874 		return 0;
875 	}
876 
877 	if (nritems == 0)
878 		return 0;
879 
880 	/*
881 	 * Check the following things to make sure this is a good leaf, and
882 	 * leaf users won't need to bother with similar sanity checks:
883 	 *
884 	 * 1) key ordering
885 	 * 2) item offset and size
886 	 *    No overlap, no hole, all inside the leaf.
887 	 * 3) item content
888 	 *    If possible, do comprehensive sanity check.
889 	 *    NOTE: All checks must only rely on the item data itself.
890 	 */
891 	for (slot = 0; slot < nritems; slot++) {
892 		u32 item_end_expected;
893 		int ret;
894 
895 		btrfs_item_key_to_cpu(leaf, &key, slot);
896 
897 		/* Make sure the keys are in the right order */
898 		if (btrfs_comp_cpu_keys(&prev_key, &key) >= 0) {
899 			generic_err(fs_info, leaf, slot,
900 	"bad key order, prev (%llu %u %llu) current (%llu %u %llu)",
901 				prev_key.objectid, prev_key.type,
902 				prev_key.offset, key.objectid, key.type,
903 				key.offset);
904 			return -EUCLEAN;
905 		}
906 
907 		/*
908 		 * Make sure the offset and ends are right, remember that the
909 		 * item data starts at the end of the leaf and grows towards the
910 		 * front.
911 		 */
912 		if (slot == 0)
913 			item_end_expected = BTRFS_LEAF_DATA_SIZE(fs_info);
914 		else
915 			item_end_expected = btrfs_item_offset_nr(leaf,
916 								 slot - 1);
917 		if (btrfs_item_end_nr(leaf, slot) != item_end_expected) {
918 			generic_err(fs_info, leaf, slot,
919 				"unexpected item end, have %u expect %u",
920 				btrfs_item_end_nr(leaf, slot),
921 				item_end_expected);
922 			return -EUCLEAN;
923 		}
924 
925 		/*
926 		 * Check to make sure that we don't point outside of the leaf,
927 		 * just in case all the items are consistent to each other, but
928 		 * all point outside of the leaf.
929 		 */
930 		if (btrfs_item_end_nr(leaf, slot) >
931 		    BTRFS_LEAF_DATA_SIZE(fs_info)) {
932 			generic_err(fs_info, leaf, slot,
933 			"slot end outside of leaf, have %u expect range [0, %u]",
934 				btrfs_item_end_nr(leaf, slot),
935 				BTRFS_LEAF_DATA_SIZE(fs_info));
936 			return -EUCLEAN;
937 		}
938 
939 		/* Also check if the item pointer overlaps with btrfs item. */
940 		if (btrfs_item_nr_offset(slot) + sizeof(struct btrfs_item) >
941 		    btrfs_item_ptr_offset(leaf, slot)) {
942 			generic_err(fs_info, leaf, slot,
943 		"slot overlaps with its data, item end %lu data start %lu",
944 				btrfs_item_nr_offset(slot) +
945 				sizeof(struct btrfs_item),
946 				btrfs_item_ptr_offset(leaf, slot));
947 			return -EUCLEAN;
948 		}
949 
950 		if (check_item_data) {
951 			/*
952 			 * Check if the item size and content meet other
953 			 * criteria
954 			 */
955 			ret = check_leaf_item(fs_info, leaf, &key, slot);
956 			if (ret < 0)
957 				return ret;
958 		}
959 
960 		prev_key.objectid = key.objectid;
961 		prev_key.type = key.type;
962 		prev_key.offset = key.offset;
963 	}
964 
965 	return 0;
966 }
967 
btrfs_check_leaf_full(struct btrfs_fs_info * fs_info,struct extent_buffer * leaf)968 int btrfs_check_leaf_full(struct btrfs_fs_info *fs_info,
969 			  struct extent_buffer *leaf)
970 {
971 	return check_leaf(fs_info, leaf, true);
972 }
973 
btrfs_check_leaf_relaxed(struct btrfs_fs_info * fs_info,struct extent_buffer * leaf)974 int btrfs_check_leaf_relaxed(struct btrfs_fs_info *fs_info,
975 			     struct extent_buffer *leaf)
976 {
977 	return check_leaf(fs_info, leaf, false);
978 }
979 
btrfs_check_node(struct btrfs_fs_info * fs_info,struct extent_buffer * node)980 int btrfs_check_node(struct btrfs_fs_info *fs_info, struct extent_buffer *node)
981 {
982 	unsigned long nr = btrfs_header_nritems(node);
983 	struct btrfs_key key, next_key;
984 	int slot;
985 	int level = btrfs_header_level(node);
986 	u64 bytenr;
987 	int ret = 0;
988 
989 	if (level <= 0 || level >= BTRFS_MAX_LEVEL) {
990 		generic_err(fs_info, node, 0,
991 			"invalid level for node, have %d expect [1, %d]",
992 			level, BTRFS_MAX_LEVEL - 1);
993 		return -EUCLEAN;
994 	}
995 	if (nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(fs_info)) {
996 		btrfs_crit(fs_info,
997 "corrupt node: root=%llu block=%llu, nritems too %s, have %lu expect range [1,%u]",
998 			   btrfs_header_owner(node), node->start,
999 			   nr == 0 ? "small" : "large", nr,
1000 			   BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1001 		return -EUCLEAN;
1002 	}
1003 
1004 	for (slot = 0; slot < nr - 1; slot++) {
1005 		bytenr = btrfs_node_blockptr(node, slot);
1006 		btrfs_node_key_to_cpu(node, &key, slot);
1007 		btrfs_node_key_to_cpu(node, &next_key, slot + 1);
1008 
1009 		if (!bytenr) {
1010 			generic_err(fs_info, node, slot,
1011 				"invalid NULL node pointer");
1012 			ret = -EUCLEAN;
1013 			goto out;
1014 		}
1015 		if (!IS_ALIGNED(bytenr, fs_info->sectorsize)) {
1016 			generic_err(fs_info, node, slot,
1017 			"unaligned pointer, have %llu should be aligned to %u",
1018 				bytenr, fs_info->sectorsize);
1019 			ret = -EUCLEAN;
1020 			goto out;
1021 		}
1022 
1023 		if (btrfs_comp_cpu_keys(&key, &next_key) >= 0) {
1024 			generic_err(fs_info, node, slot,
1025 	"bad key order, current (%llu %u %llu) next (%llu %u %llu)",
1026 				key.objectid, key.type, key.offset,
1027 				next_key.objectid, next_key.type,
1028 				next_key.offset);
1029 			ret = -EUCLEAN;
1030 			goto out;
1031 		}
1032 	}
1033 out:
1034 	return ret;
1035 }
1036