• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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(3,4)
452 __cold
extent_err(const struct extent_buffer * eb,int slot,const char * fmt,...)453 static void extent_err(const struct extent_buffer *eb, int slot,
454 		       const char *fmt, ...)
455 {
456 	struct btrfs_key key;
457 	struct va_format vaf;
458 	va_list args;
459 	u64 bytenr;
460 	u64 len;
461 
462 	btrfs_item_key_to_cpu(eb, &key, slot);
463 	bytenr = key.objectid;
464 	if (key.type == BTRFS_METADATA_ITEM_KEY)
465 		len = eb->fs_info->nodesize;
466 	else
467 		len = key.offset;
468 	va_start(args, fmt);
469 
470 	vaf.fmt = fmt;
471 	vaf.va = &args;
472 
473 	btrfs_crit(eb->fs_info,
474 	"corrupt %s: block=%llu slot=%d extent bytenr=%llu len=%llu %pV",
475 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
476 		eb->start, slot, bytenr, len, &vaf);
477 	va_end(args);
478 }
479 
check_extent_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot)480 static int check_extent_item(struct extent_buffer *leaf,
481 			     struct btrfs_key *key, int slot)
482 {
483 	struct btrfs_fs_info *fs_info = leaf->fs_info;
484 	struct btrfs_extent_item *ei;
485 	bool is_tree_block = false;
486 	unsigned long ptr;	/* Current pointer inside inline refs */
487 	unsigned long end;	/* Extent item end */
488 	const u32 item_size = btrfs_item_size_nr(leaf, slot);
489 	u64 flags;
490 	u64 generation;
491 	u64 total_refs;		/* Total refs in btrfs_extent_item */
492 	u64 inline_refs = 0;	/* found total inline refs */
493 
494 	if (key->type == BTRFS_METADATA_ITEM_KEY &&
495 	    !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
496 		generic_err(fs_info, leaf, slot,
497 "invalid key type, METADATA_ITEM type invalid when SKINNY_METADATA feature disabled");
498 		return -EUCLEAN;
499 	}
500 	/* key->objectid is the bytenr for both key types */
501 	if (!IS_ALIGNED(key->objectid, fs_info->sectorsize)) {
502 		generic_err(fs_info, leaf, slot,
503 		"invalid key objectid, have %llu expect to be aligned to %u",
504 			   key->objectid, fs_info->sectorsize);
505 		return -EUCLEAN;
506 	}
507 
508 	/* key->offset is tree level for METADATA_ITEM_KEY */
509 	if (key->type == BTRFS_METADATA_ITEM_KEY &&
510 	    key->offset >= BTRFS_MAX_LEVEL) {
511 		extent_err(leaf, slot,
512 			   "invalid tree level, have %llu expect [0, %u]",
513 			   key->offset, BTRFS_MAX_LEVEL - 1);
514 		return -EUCLEAN;
515 	}
516 
517 	/*
518 	 * EXTENT/METADATA_ITEM consists of:
519 	 * 1) One btrfs_extent_item
520 	 *    Records the total refs, type and generation of the extent.
521 	 *
522 	 * 2) One btrfs_tree_block_info (for EXTENT_ITEM and tree backref only)
523 	 *    Records the first key and level of the tree block.
524 	 *
525 	 * 2) Zero or more btrfs_extent_inline_ref(s)
526 	 *    Each inline ref has one btrfs_extent_inline_ref shows:
527 	 *    2.1) The ref type, one of the 4
528 	 *         TREE_BLOCK_REF	Tree block only
529 	 *         SHARED_BLOCK_REF	Tree block only
530 	 *         EXTENT_DATA_REF	Data only
531 	 *         SHARED_DATA_REF	Data only
532 	 *    2.2) Ref type specific data
533 	 *         Either using btrfs_extent_inline_ref::offset, or specific
534 	 *         data structure.
535 	 */
536 	if (item_size < sizeof(*ei)) {
537 		extent_err(leaf, slot,
538 			   "invalid item size, have %u expect [%zu, %u)",
539 			   item_size, sizeof(*ei),
540 			   BTRFS_LEAF_DATA_SIZE(fs_info));
541 		return -EUCLEAN;
542 	}
543 	end = item_size + btrfs_item_ptr_offset(leaf, slot);
544 
545 	/* Checks against extent_item */
546 	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
547 	flags = btrfs_extent_flags(leaf, ei);
548 	total_refs = btrfs_extent_refs(leaf, ei);
549 	generation = btrfs_extent_generation(leaf, ei);
550 	if (generation > btrfs_super_generation(fs_info->super_copy) + 1) {
551 		extent_err(leaf, slot,
552 			   "invalid generation, have %llu expect (0, %llu]",
553 			   generation,
554 			   btrfs_super_generation(fs_info->super_copy) + 1);
555 		return -EUCLEAN;
556 	}
557 	if (!is_power_of_2(flags & (BTRFS_EXTENT_FLAG_DATA |
558 				    BTRFS_EXTENT_FLAG_TREE_BLOCK))) {
559 		extent_err(leaf, slot,
560 		"invalid extent flag, have 0x%llx expect 1 bit set in 0x%llx",
561 			flags, BTRFS_EXTENT_FLAG_DATA |
562 			BTRFS_EXTENT_FLAG_TREE_BLOCK);
563 		return -EUCLEAN;
564 	}
565 	is_tree_block = !!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK);
566 	if (is_tree_block) {
567 		if (key->type == BTRFS_EXTENT_ITEM_KEY &&
568 		    key->offset != fs_info->nodesize) {
569 			extent_err(leaf, slot,
570 				   "invalid extent length, have %llu expect %u",
571 				   key->offset, fs_info->nodesize);
572 			return -EUCLEAN;
573 		}
574 	} else {
575 		if (key->type != BTRFS_EXTENT_ITEM_KEY) {
576 			extent_err(leaf, slot,
577 			"invalid key type, have %u expect %u for data backref",
578 				   key->type, BTRFS_EXTENT_ITEM_KEY);
579 			return -EUCLEAN;
580 		}
581 		if (!IS_ALIGNED(key->offset, fs_info->sectorsize)) {
582 			extent_err(leaf, slot,
583 			"invalid extent length, have %llu expect aligned to %u",
584 				   key->offset, fs_info->sectorsize);
585 			return -EUCLEAN;
586 		}
587 	}
588 	ptr = (unsigned long)(struct btrfs_extent_item *)(ei + 1);
589 
590 	/* Check the special case of btrfs_tree_block_info */
591 	if (is_tree_block && key->type != BTRFS_METADATA_ITEM_KEY) {
592 		struct btrfs_tree_block_info *info;
593 
594 		info = (struct btrfs_tree_block_info *)ptr;
595 		if (btrfs_tree_block_level(leaf, info) >= BTRFS_MAX_LEVEL) {
596 			extent_err(leaf, slot,
597 			"invalid tree block info level, have %u expect [0, %u]",
598 				   btrfs_tree_block_level(leaf, info),
599 				   BTRFS_MAX_LEVEL - 1);
600 			return -EUCLEAN;
601 		}
602 		ptr = (unsigned long)(struct btrfs_tree_block_info *)(info + 1);
603 	}
604 
605 	/* Check inline refs */
606 	while (ptr < end) {
607 		struct btrfs_extent_inline_ref *iref;
608 		struct btrfs_extent_data_ref *dref;
609 		struct btrfs_shared_data_ref *sref;
610 		u64 dref_offset;
611 		u64 inline_offset;
612 		u8 inline_type;
613 
614 		if (ptr + sizeof(*iref) > end) {
615 			extent_err(leaf, slot,
616 "inline ref item overflows extent item, ptr %lu iref size %zu end %lu",
617 				   ptr, sizeof(*iref), end);
618 			return -EUCLEAN;
619 		}
620 		iref = (struct btrfs_extent_inline_ref *)ptr;
621 		inline_type = btrfs_extent_inline_ref_type(leaf, iref);
622 		inline_offset = btrfs_extent_inline_ref_offset(leaf, iref);
623 		if (ptr + btrfs_extent_inline_ref_size(inline_type) > end) {
624 			extent_err(leaf, slot,
625 "inline ref item overflows extent item, ptr %lu iref size %u end %lu",
626 				   ptr, inline_type, end);
627 			return -EUCLEAN;
628 		}
629 
630 		switch (inline_type) {
631 		/* inline_offset is subvolid of the owner, no need to check */
632 		case BTRFS_TREE_BLOCK_REF_KEY:
633 			inline_refs++;
634 			break;
635 		/* Contains parent bytenr */
636 		case BTRFS_SHARED_BLOCK_REF_KEY:
637 			if (!IS_ALIGNED(inline_offset, fs_info->sectorsize)) {
638 				extent_err(leaf, slot,
639 		"invalid tree parent bytenr, have %llu expect aligned to %u",
640 					   inline_offset, fs_info->sectorsize);
641 				return -EUCLEAN;
642 			}
643 			inline_refs++;
644 			break;
645 		/*
646 		 * Contains owner subvolid, owner key objectid, adjusted offset.
647 		 * The only obvious corruption can happen in that offset.
648 		 */
649 		case BTRFS_EXTENT_DATA_REF_KEY:
650 			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
651 			dref_offset = btrfs_extent_data_ref_offset(leaf, dref);
652 			if (!IS_ALIGNED(dref_offset, fs_info->sectorsize)) {
653 				extent_err(leaf, slot,
654 		"invalid data ref offset, have %llu expect aligned to %u",
655 					   dref_offset, fs_info->sectorsize);
656 				return -EUCLEAN;
657 			}
658 			inline_refs += btrfs_extent_data_ref_count(leaf, dref);
659 			break;
660 		/* Contains parent bytenr and ref count */
661 		case BTRFS_SHARED_DATA_REF_KEY:
662 			sref = (struct btrfs_shared_data_ref *)(iref + 1);
663 			if (!IS_ALIGNED(inline_offset, fs_info->sectorsize)) {
664 				extent_err(leaf, slot,
665 		"invalid data parent bytenr, have %llu expect aligned to %u",
666 					   inline_offset, fs_info->sectorsize);
667 				return -EUCLEAN;
668 			}
669 			inline_refs += btrfs_shared_data_ref_count(leaf, sref);
670 			break;
671 		default:
672 			extent_err(leaf, slot, "unknown inline ref type: %u",
673 				   inline_type);
674 			return -EUCLEAN;
675 		}
676 		ptr += btrfs_extent_inline_ref_size(inline_type);
677 	}
678 	/* No padding is allowed */
679 	if (ptr != end) {
680 		extent_err(leaf, slot,
681 			   "invalid extent item size, padding bytes found");
682 		return -EUCLEAN;
683 	}
684 
685 	/* Finally, check the inline refs against total refs */
686 	if (inline_refs > total_refs) {
687 		extent_err(leaf, slot,
688 			"invalid extent refs, have %llu expect >= inline %llu",
689 			   total_refs, inline_refs);
690 		return -EUCLEAN;
691 	}
692 	return 0;
693 }
694 
695 /* Inode item error output has the same format as dir_item_err() */
696 #define inode_item_err(fs_info, eb, slot, fmt, ...)			\
697 	dir_item_err(fs_info, eb, slot, fmt, __VA_ARGS__)
698 
check_inode_item(struct btrfs_fs_info * fs_info,struct extent_buffer * leaf,struct btrfs_key * key,int slot)699 static int check_inode_item(struct btrfs_fs_info *fs_info,
700 			    struct extent_buffer *leaf,
701 			    struct btrfs_key *key, int slot)
702 {
703 	struct btrfs_inode_item *iitem;
704 	u64 super_gen = btrfs_super_generation(fs_info->super_copy);
705 	u32 valid_mask = (S_IFMT | S_ISUID | S_ISGID | S_ISVTX | 0777);
706 	u32 mode;
707 
708 	if ((key->objectid < BTRFS_FIRST_FREE_OBJECTID ||
709 	     key->objectid > BTRFS_LAST_FREE_OBJECTID) &&
710 	    key->objectid != BTRFS_ROOT_TREE_DIR_OBJECTID &&
711 	    key->objectid != BTRFS_FREE_INO_OBJECTID) {
712 		generic_err(fs_info, leaf, slot,
713 	"invalid key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
714 			    key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
715 			    BTRFS_FIRST_FREE_OBJECTID,
716 			    BTRFS_LAST_FREE_OBJECTID,
717 			    BTRFS_FREE_INO_OBJECTID);
718 		return -EUCLEAN;
719 	}
720 	if (key->offset != 0) {
721 		inode_item_err(fs_info, leaf, slot,
722 			"invalid key offset: has %llu expect 0",
723 			key->offset);
724 		return -EUCLEAN;
725 	}
726 	iitem = btrfs_item_ptr(leaf, slot, struct btrfs_inode_item);
727 
728 	/* Here we use super block generation + 1 to handle log tree */
729 	if (btrfs_inode_generation(leaf, iitem) > super_gen + 1) {
730 		inode_item_err(fs_info, leaf, slot,
731 			"invalid inode generation: has %llu expect (0, %llu]",
732 			       btrfs_inode_generation(leaf, iitem),
733 			       super_gen + 1);
734 		return -EUCLEAN;
735 	}
736 	/* Note for ROOT_TREE_DIR_ITEM, mkfs could set its transid 0 */
737 	if (btrfs_inode_transid(leaf, iitem) > super_gen + 1) {
738 		inode_item_err(fs_info, leaf, slot,
739 			"invalid inode generation: has %llu expect [0, %llu]",
740 			       btrfs_inode_transid(leaf, iitem), super_gen + 1);
741 		return -EUCLEAN;
742 	}
743 
744 	/*
745 	 * For size and nbytes it's better not to be too strict, as for dir
746 	 * item its size/nbytes can easily get wrong, but doesn't affect
747 	 * anything in the fs. So here we skip the check.
748 	 */
749 	mode = btrfs_inode_mode(leaf, iitem);
750 	if (mode & ~valid_mask) {
751 		inode_item_err(fs_info, leaf, slot,
752 			       "unknown mode bit detected: 0x%x",
753 			       mode & ~valid_mask);
754 		return -EUCLEAN;
755 	}
756 
757 	/*
758 	 * S_IFMT is not bit mapped so we can't completely rely on is_power_of_2,
759 	 * but is_power_of_2() can save us from checking FIFO/CHR/DIR/REG.
760 	 * Only needs to check BLK, LNK and SOCKS
761 	 */
762 	if (!is_power_of_2(mode & S_IFMT)) {
763 		if (!S_ISLNK(mode) && !S_ISBLK(mode) && !S_ISSOCK(mode)) {
764 			inode_item_err(fs_info, leaf, slot,
765 			"invalid mode: has 0%o expect valid S_IF* bit(s)",
766 				       mode & S_IFMT);
767 			return -EUCLEAN;
768 		}
769 	}
770 	if (S_ISDIR(mode) && btrfs_inode_nlink(leaf, iitem) > 1) {
771 		inode_item_err(fs_info, leaf, slot,
772 		       "invalid nlink: has %u expect no more than 1 for dir",
773 			btrfs_inode_nlink(leaf, iitem));
774 		return -EUCLEAN;
775 	}
776 	if (btrfs_inode_flags(leaf, iitem) & ~BTRFS_INODE_FLAG_MASK) {
777 		inode_item_err(fs_info, leaf, slot,
778 			       "unknown flags detected: 0x%llx",
779 			       btrfs_inode_flags(leaf, iitem) &
780 			       ~BTRFS_INODE_FLAG_MASK);
781 		return -EUCLEAN;
782 	}
783 	return 0;
784 }
785 
786 /*
787  * Common point to switch the item-specific validation.
788  */
check_leaf_item(struct btrfs_fs_info * fs_info,struct extent_buffer * leaf,struct btrfs_key * key,int slot)789 static int check_leaf_item(struct btrfs_fs_info *fs_info,
790 			   struct extent_buffer *leaf,
791 			   struct btrfs_key *key, int slot)
792 {
793 	int ret = 0;
794 
795 	switch (key->type) {
796 	case BTRFS_EXTENT_DATA_KEY:
797 		ret = check_extent_data_item(fs_info, leaf, key, slot);
798 		break;
799 	case BTRFS_EXTENT_CSUM_KEY:
800 		ret = check_csum_item(fs_info, leaf, key, slot);
801 		break;
802 	case BTRFS_DIR_ITEM_KEY:
803 	case BTRFS_DIR_INDEX_KEY:
804 	case BTRFS_XATTR_ITEM_KEY:
805 		ret = check_dir_item(fs_info, leaf, key, slot);
806 		break;
807 	case BTRFS_BLOCK_GROUP_ITEM_KEY:
808 		ret = check_block_group_item(fs_info, leaf, key, slot);
809 		break;
810 	case BTRFS_EXTENT_ITEM_KEY:
811 	case BTRFS_METADATA_ITEM_KEY:
812 		ret = check_extent_item(leaf, key, slot);
813 		break;
814 	case BTRFS_INODE_ITEM_KEY:
815 		ret = check_inode_item(fs_info, leaf, key, slot);
816 		break;
817 	}
818 	return ret;
819 }
820 
check_leaf(struct btrfs_fs_info * fs_info,struct extent_buffer * leaf,bool check_item_data)821 static int check_leaf(struct btrfs_fs_info *fs_info, struct extent_buffer *leaf,
822 		      bool check_item_data)
823 {
824 	/* No valid key type is 0, so all key should be larger than this key */
825 	struct btrfs_key prev_key = {0, 0, 0};
826 	struct btrfs_key key;
827 	u32 nritems = btrfs_header_nritems(leaf);
828 	int slot;
829 
830 	if (btrfs_header_level(leaf) != 0) {
831 		generic_err(fs_info, leaf, 0,
832 			"invalid level for leaf, have %d expect 0",
833 			btrfs_header_level(leaf));
834 		return -EUCLEAN;
835 	}
836 
837 	/*
838 	 * Extent buffers from a relocation tree have a owner field that
839 	 * corresponds to the subvolume tree they are based on. So just from an
840 	 * extent buffer alone we can not find out what is the id of the
841 	 * corresponding subvolume tree, so we can not figure out if the extent
842 	 * buffer corresponds to the root of the relocation tree or not. So
843 	 * skip this check for relocation trees.
844 	 */
845 	if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
846 		u64 owner = btrfs_header_owner(leaf);
847 
848 		/* These trees must never be empty */
849 		if (owner == BTRFS_ROOT_TREE_OBJECTID ||
850 		    owner == BTRFS_CHUNK_TREE_OBJECTID ||
851 		    owner == BTRFS_EXTENT_TREE_OBJECTID ||
852 		    owner == BTRFS_DEV_TREE_OBJECTID ||
853 		    owner == BTRFS_FS_TREE_OBJECTID ||
854 		    owner == BTRFS_DATA_RELOC_TREE_OBJECTID) {
855 			generic_err(fs_info, leaf, 0,
856 			"invalid root, root %llu must never be empty",
857 				    owner);
858 			return -EUCLEAN;
859 		}
860 		/* Unknown tree */
861 		if (owner == 0) {
862 			generic_err(fs_info, leaf, 0,
863 				"invalid owner, root 0 is not defined");
864 			return -EUCLEAN;
865 		}
866 		return 0;
867 	}
868 
869 	if (nritems == 0)
870 		return 0;
871 
872 	/*
873 	 * Check the following things to make sure this is a good leaf, and
874 	 * leaf users won't need to bother with similar sanity checks:
875 	 *
876 	 * 1) key ordering
877 	 * 2) item offset and size
878 	 *    No overlap, no hole, all inside the leaf.
879 	 * 3) item content
880 	 *    If possible, do comprehensive sanity check.
881 	 *    NOTE: All checks must only rely on the item data itself.
882 	 */
883 	for (slot = 0; slot < nritems; slot++) {
884 		u32 item_end_expected;
885 		int ret;
886 
887 		btrfs_item_key_to_cpu(leaf, &key, slot);
888 
889 		/* Make sure the keys are in the right order */
890 		if (btrfs_comp_cpu_keys(&prev_key, &key) >= 0) {
891 			generic_err(fs_info, leaf, slot,
892 	"bad key order, prev (%llu %u %llu) current (%llu %u %llu)",
893 				prev_key.objectid, prev_key.type,
894 				prev_key.offset, key.objectid, key.type,
895 				key.offset);
896 			return -EUCLEAN;
897 		}
898 
899 		/*
900 		 * Make sure the offset and ends are right, remember that the
901 		 * item data starts at the end of the leaf and grows towards the
902 		 * front.
903 		 */
904 		if (slot == 0)
905 			item_end_expected = BTRFS_LEAF_DATA_SIZE(fs_info);
906 		else
907 			item_end_expected = btrfs_item_offset_nr(leaf,
908 								 slot - 1);
909 		if (btrfs_item_end_nr(leaf, slot) != item_end_expected) {
910 			generic_err(fs_info, leaf, slot,
911 				"unexpected item end, have %u expect %u",
912 				btrfs_item_end_nr(leaf, slot),
913 				item_end_expected);
914 			return -EUCLEAN;
915 		}
916 
917 		/*
918 		 * Check to make sure that we don't point outside of the leaf,
919 		 * just in case all the items are consistent to each other, but
920 		 * all point outside of the leaf.
921 		 */
922 		if (btrfs_item_end_nr(leaf, slot) >
923 		    BTRFS_LEAF_DATA_SIZE(fs_info)) {
924 			generic_err(fs_info, leaf, slot,
925 			"slot end outside of leaf, have %u expect range [0, %u]",
926 				btrfs_item_end_nr(leaf, slot),
927 				BTRFS_LEAF_DATA_SIZE(fs_info));
928 			return -EUCLEAN;
929 		}
930 
931 		/* Also check if the item pointer overlaps with btrfs item. */
932 		if (btrfs_item_nr_offset(slot) + sizeof(struct btrfs_item) >
933 		    btrfs_item_ptr_offset(leaf, slot)) {
934 			generic_err(fs_info, leaf, slot,
935 		"slot overlaps with its data, item end %lu data start %lu",
936 				btrfs_item_nr_offset(slot) +
937 				sizeof(struct btrfs_item),
938 				btrfs_item_ptr_offset(leaf, slot));
939 			return -EUCLEAN;
940 		}
941 
942 		if (check_item_data) {
943 			/*
944 			 * Check if the item size and content meet other
945 			 * criteria
946 			 */
947 			ret = check_leaf_item(fs_info, leaf, &key, slot);
948 			if (ret < 0)
949 				return ret;
950 		}
951 
952 		prev_key.objectid = key.objectid;
953 		prev_key.type = key.type;
954 		prev_key.offset = key.offset;
955 	}
956 
957 	return 0;
958 }
959 
btrfs_check_leaf_full(struct btrfs_fs_info * fs_info,struct extent_buffer * leaf)960 int btrfs_check_leaf_full(struct btrfs_fs_info *fs_info,
961 			  struct extent_buffer *leaf)
962 {
963 	return check_leaf(fs_info, leaf, true);
964 }
965 
btrfs_check_leaf_relaxed(struct btrfs_fs_info * fs_info,struct extent_buffer * leaf)966 int btrfs_check_leaf_relaxed(struct btrfs_fs_info *fs_info,
967 			     struct extent_buffer *leaf)
968 {
969 	return check_leaf(fs_info, leaf, false);
970 }
971 
btrfs_check_node(struct btrfs_fs_info * fs_info,struct extent_buffer * node)972 int btrfs_check_node(struct btrfs_fs_info *fs_info, struct extent_buffer *node)
973 {
974 	unsigned long nr = btrfs_header_nritems(node);
975 	struct btrfs_key key, next_key;
976 	int slot;
977 	int level = btrfs_header_level(node);
978 	u64 bytenr;
979 	int ret = 0;
980 
981 	if (level <= 0 || level >= BTRFS_MAX_LEVEL) {
982 		generic_err(fs_info, node, 0,
983 			"invalid level for node, have %d expect [1, %d]",
984 			level, BTRFS_MAX_LEVEL - 1);
985 		return -EUCLEAN;
986 	}
987 	if (nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(fs_info)) {
988 		btrfs_crit(fs_info,
989 "corrupt node: root=%llu block=%llu, nritems too %s, have %lu expect range [1,%u]",
990 			   btrfs_header_owner(node), node->start,
991 			   nr == 0 ? "small" : "large", nr,
992 			   BTRFS_NODEPTRS_PER_BLOCK(fs_info));
993 		return -EUCLEAN;
994 	}
995 
996 	for (slot = 0; slot < nr - 1; slot++) {
997 		bytenr = btrfs_node_blockptr(node, slot);
998 		btrfs_node_key_to_cpu(node, &key, slot);
999 		btrfs_node_key_to_cpu(node, &next_key, slot + 1);
1000 
1001 		if (!bytenr) {
1002 			generic_err(fs_info, node, slot,
1003 				"invalid NULL node pointer");
1004 			ret = -EUCLEAN;
1005 			goto out;
1006 		}
1007 		if (!IS_ALIGNED(bytenr, fs_info->sectorsize)) {
1008 			generic_err(fs_info, node, slot,
1009 			"unaligned pointer, have %llu should be aligned to %u",
1010 				bytenr, fs_info->sectorsize);
1011 			ret = -EUCLEAN;
1012 			goto out;
1013 		}
1014 
1015 		if (btrfs_comp_cpu_keys(&key, &next_key) >= 0) {
1016 			generic_err(fs_info, node, slot,
1017 	"bad key order, current (%llu %u %llu) next (%llu %u %llu)",
1018 				key.objectid, key.type, key.offset,
1019 				next_key.objectid, next_key.type,
1020 				next_key.offset);
1021 			ret = -EUCLEAN;
1022 			goto out;
1023 		}
1024 	}
1025 out:
1026 	return ret;
1027 }
1028