• 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 <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"
27 
28 /*
29  * Error message should follow the following format:
30  * corrupt <type>: <identifier>, <reason>[, <bad_value>]
31  *
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).
39  *
40  * Since comma is used to separate the components, only space is allowed
41  * inside each component.
42  */
43 
44 /*
45  * Append generic "corrupt leaf/node root=%llu block=%llu slot=%d: " to @fmt.
46  * Allows callers to customize the output.
47  */
48 __printf(3, 4)
49 __cold
generic_err(const struct extent_buffer * eb,int slot,const char * fmt,...)50 static void generic_err(const struct extent_buffer *eb, int slot,
51 			const char *fmt, ...)
52 {
53 	const struct btrfs_fs_info *fs_info = eb->fs_info;
54 	struct va_format vaf;
55 	va_list args;
56 
57 	va_start(args, fmt);
58 
59 	vaf.fmt = fmt;
60 	vaf.va = &args;
61 
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);
67 }
68 
69 /*
70  * Customized reporter for extent data item, since its key objectid and
71  * offset has its own meaning.
72  */
73 __printf(3, 4)
74 __cold
file_extent_err(const struct extent_buffer * eb,int slot,const char * fmt,...)75 static void file_extent_err(const struct extent_buffer *eb, int slot,
76 			    const char *fmt, ...)
77 {
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;
82 
83 	btrfs_item_key_to_cpu(eb, &key, slot);
84 	va_start(args, fmt);
85 
86 	vaf.fmt = fmt;
87 	vaf.va = &args;
88 
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);
95 }
96 
97 /*
98  * Return 0 if the btrfs_file_extent_##name is aligned to @alignment
99  * Else return 1
100  */
101 #define CHECK_FE_ALIGNED(leaf, slot, fi, name, alignment)		      \
102 ({									      \
103 	if (!IS_ALIGNED(btrfs_file_extent_##name((leaf), (fi)), (alignment))) \
104 		file_extent_err((leaf), (slot),				      \
105 	"invalid %s for file extent, have %llu, should be aligned to %u",     \
106 			(#name), btrfs_file_extent_##name((leaf), (fi)),      \
107 			(alignment));					      \
108 	(!IS_ALIGNED(btrfs_file_extent_##name((leaf), (fi)), (alignment)));   \
109 })
110 
file_extent_end(struct extent_buffer * leaf,struct btrfs_key * key,struct btrfs_file_extent_item * extent)111 static u64 file_extent_end(struct extent_buffer *leaf,
112 			   struct btrfs_key *key,
113 			   struct btrfs_file_extent_item *extent)
114 {
115 	u64 end;
116 	u64 len;
117 
118 	if (btrfs_file_extent_type(leaf, extent) == BTRFS_FILE_EXTENT_INLINE) {
119 		len = btrfs_file_extent_ram_bytes(leaf, extent);
120 		end = ALIGN(key->offset + len, leaf->fs_info->sectorsize);
121 	} else {
122 		len = btrfs_file_extent_num_bytes(leaf, extent);
123 		end = key->offset + len;
124 	}
125 	return end;
126 }
127 
128 /*
129  * Customized report for dir_item, the only new important information is
130  * key->objectid, which represents inode number
131  */
132 __printf(3, 4)
133 __cold
dir_item_err(const struct extent_buffer * eb,int slot,const char * fmt,...)134 static void dir_item_err(const struct extent_buffer *eb, int slot,
135 			 const char *fmt, ...)
136 {
137 	const struct btrfs_fs_info *fs_info = eb->fs_info;
138 	struct btrfs_key key;
139 	struct va_format vaf;
140 	va_list args;
141 
142 	btrfs_item_key_to_cpu(eb, &key, slot);
143 	va_start(args, fmt);
144 
145 	vaf.fmt = fmt;
146 	vaf.va = &args;
147 
148 	btrfs_crit(fs_info,
149 		"corrupt %s: root=%llu block=%llu slot=%d ino=%llu, %pV",
150 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
151 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
152 		key.objectid, &vaf);
153 	va_end(args);
154 }
155 
156 /*
157  * This functions checks prev_key->objectid, to ensure current key and prev_key
158  * share the same objectid as inode number.
159  *
160  * This is to detect missing INODE_ITEM in subvolume trees.
161  *
162  * Return true if everything is OK or we don't need to check.
163  * Return false if anything is wrong.
164  */
check_prev_ino(struct extent_buffer * leaf,struct btrfs_key * key,int slot,struct btrfs_key * prev_key)165 static bool check_prev_ino(struct extent_buffer *leaf,
166 			   struct btrfs_key *key, int slot,
167 			   struct btrfs_key *prev_key)
168 {
169 	/* No prev key, skip check */
170 	if (slot == 0)
171 		return true;
172 
173 	/* Only these key->types needs to be checked */
174 	ASSERT(key->type == BTRFS_XATTR_ITEM_KEY ||
175 	       key->type == BTRFS_INODE_REF_KEY ||
176 	       key->type == BTRFS_DIR_INDEX_KEY ||
177 	       key->type == BTRFS_DIR_ITEM_KEY ||
178 	       key->type == BTRFS_EXTENT_DATA_KEY);
179 
180 	/*
181 	 * Only subvolume trees along with their reloc trees need this check.
182 	 * Things like log tree doesn't follow this ino requirement.
183 	 */
184 	if (!is_fstree(btrfs_header_owner(leaf)))
185 		return true;
186 
187 	if (key->objectid == prev_key->objectid)
188 		return true;
189 
190 	/* Error found */
191 	dir_item_err(leaf, slot,
192 		"invalid previous key objectid, have %llu expect %llu",
193 		prev_key->objectid, key->objectid);
194 	return false;
195 }
check_extent_data_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot,struct btrfs_key * prev_key)196 static int check_extent_data_item(struct extent_buffer *leaf,
197 				  struct btrfs_key *key, int slot,
198 				  struct btrfs_key *prev_key)
199 {
200 	struct btrfs_fs_info *fs_info = leaf->fs_info;
201 	struct btrfs_file_extent_item *fi;
202 	u32 sectorsize = fs_info->sectorsize;
203 	u32 item_size = btrfs_item_size_nr(leaf, slot);
204 	u64 extent_end;
205 
206 	if (!IS_ALIGNED(key->offset, sectorsize)) {
207 		file_extent_err(leaf, slot,
208 "unaligned file_offset for file extent, have %llu should be aligned to %u",
209 			key->offset, sectorsize);
210 		return -EUCLEAN;
211 	}
212 
213 	/*
214 	 * Previous key must have the same key->objectid (ino).
215 	 * It can be XATTR_ITEM, INODE_ITEM or just another EXTENT_DATA.
216 	 * But if objectids mismatch, it means we have a missing
217 	 * INODE_ITEM.
218 	 */
219 	if (!check_prev_ino(leaf, key, slot, prev_key))
220 		return -EUCLEAN;
221 
222 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
223 
224 	/*
225 	 * Make sure the item contains at least inline header, so the file
226 	 * extent type is not some garbage.
227 	 */
228 	if (item_size < BTRFS_FILE_EXTENT_INLINE_DATA_START) {
229 		file_extent_err(leaf, slot,
230 				"invalid item size, have %u expect [%zu, %u)",
231 				item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START,
232 				SZ_4K);
233 		return -EUCLEAN;
234 	}
235 	if (btrfs_file_extent_type(leaf, fi) >= BTRFS_NR_FILE_EXTENT_TYPES) {
236 		file_extent_err(leaf, slot,
237 		"invalid type for file extent, have %u expect range [0, %u]",
238 			btrfs_file_extent_type(leaf, fi),
239 			BTRFS_NR_FILE_EXTENT_TYPES - 1);
240 		return -EUCLEAN;
241 	}
242 
243 	/*
244 	 * Support for new compression/encryption must introduce incompat flag,
245 	 * and must be caught in open_ctree().
246 	 */
247 	if (btrfs_file_extent_compression(leaf, fi) >= BTRFS_NR_COMPRESS_TYPES) {
248 		file_extent_err(leaf, slot,
249 	"invalid compression for file extent, have %u expect range [0, %u]",
250 			btrfs_file_extent_compression(leaf, fi),
251 			BTRFS_NR_COMPRESS_TYPES - 1);
252 		return -EUCLEAN;
253 	}
254 	if (btrfs_file_extent_encryption(leaf, fi)) {
255 		file_extent_err(leaf, slot,
256 			"invalid encryption for file extent, have %u expect 0",
257 			btrfs_file_extent_encryption(leaf, fi));
258 		return -EUCLEAN;
259 	}
260 	if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
261 		/* Inline extent must have 0 as key offset */
262 		if (key->offset) {
263 			file_extent_err(leaf, slot,
264 		"invalid file_offset for inline file extent, have %llu expect 0",
265 				key->offset);
266 			return -EUCLEAN;
267 		}
268 
269 		/* Compressed inline extent has no on-disk size, skip it */
270 		if (btrfs_file_extent_compression(leaf, fi) !=
271 		    BTRFS_COMPRESS_NONE)
272 			return 0;
273 
274 		/* Uncompressed inline extent size must match item size */
275 		if (item_size != BTRFS_FILE_EXTENT_INLINE_DATA_START +
276 		    btrfs_file_extent_ram_bytes(leaf, fi)) {
277 			file_extent_err(leaf, slot,
278 	"invalid ram_bytes for uncompressed inline extent, have %u expect %llu",
279 				item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START +
280 				btrfs_file_extent_ram_bytes(leaf, fi));
281 			return -EUCLEAN;
282 		}
283 		return 0;
284 	}
285 
286 	/* Regular or preallocated extent has fixed item size */
287 	if (item_size != sizeof(*fi)) {
288 		file_extent_err(leaf, slot,
289 	"invalid item size for reg/prealloc file extent, have %u expect %zu",
290 			item_size, sizeof(*fi));
291 		return -EUCLEAN;
292 	}
293 	if (CHECK_FE_ALIGNED(leaf, slot, fi, ram_bytes, sectorsize) ||
294 	    CHECK_FE_ALIGNED(leaf, slot, fi, disk_bytenr, sectorsize) ||
295 	    CHECK_FE_ALIGNED(leaf, slot, fi, disk_num_bytes, sectorsize) ||
296 	    CHECK_FE_ALIGNED(leaf, slot, fi, offset, sectorsize) ||
297 	    CHECK_FE_ALIGNED(leaf, slot, fi, num_bytes, sectorsize))
298 		return -EUCLEAN;
299 
300 	/* Catch extent end overflow */
301 	if (check_add_overflow(btrfs_file_extent_num_bytes(leaf, fi),
302 			       key->offset, &extent_end)) {
303 		file_extent_err(leaf, slot,
304 	"extent end overflow, have file offset %llu extent num bytes %llu",
305 				key->offset,
306 				btrfs_file_extent_num_bytes(leaf, fi));
307 		return -EUCLEAN;
308 	}
309 
310 	/*
311 	 * Check that no two consecutive file extent items, in the same leaf,
312 	 * present ranges that overlap each other.
313 	 */
314 	if (slot > 0 &&
315 	    prev_key->objectid == key->objectid &&
316 	    prev_key->type == BTRFS_EXTENT_DATA_KEY) {
317 		struct btrfs_file_extent_item *prev_fi;
318 		u64 prev_end;
319 
320 		prev_fi = btrfs_item_ptr(leaf, slot - 1,
321 					 struct btrfs_file_extent_item);
322 		prev_end = file_extent_end(leaf, prev_key, prev_fi);
323 		if (prev_end > key->offset) {
324 			file_extent_err(leaf, slot - 1,
325 "file extent end range (%llu) goes beyond start offset (%llu) of the next file extent",
326 					prev_end, key->offset);
327 			return -EUCLEAN;
328 		}
329 	}
330 
331 	return 0;
332 }
333 
check_csum_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot,struct btrfs_key * prev_key)334 static int check_csum_item(struct extent_buffer *leaf, struct btrfs_key *key,
335 			   int slot, struct btrfs_key *prev_key)
336 {
337 	struct btrfs_fs_info *fs_info = leaf->fs_info;
338 	u32 sectorsize = fs_info->sectorsize;
339 	u32 csumsize = btrfs_super_csum_size(fs_info->super_copy);
340 
341 	if (key->objectid != BTRFS_EXTENT_CSUM_OBJECTID) {
342 		generic_err(leaf, slot,
343 		"invalid key objectid for csum item, have %llu expect %llu",
344 			key->objectid, BTRFS_EXTENT_CSUM_OBJECTID);
345 		return -EUCLEAN;
346 	}
347 	if (!IS_ALIGNED(key->offset, sectorsize)) {
348 		generic_err(leaf, slot,
349 	"unaligned key offset for csum item, have %llu should be aligned to %u",
350 			key->offset, sectorsize);
351 		return -EUCLEAN;
352 	}
353 	if (!IS_ALIGNED(btrfs_item_size_nr(leaf, slot), csumsize)) {
354 		generic_err(leaf, slot,
355 	"unaligned item size for csum item, have %u should be aligned to %u",
356 			btrfs_item_size_nr(leaf, slot), csumsize);
357 		return -EUCLEAN;
358 	}
359 	if (slot > 0 && prev_key->type == BTRFS_EXTENT_CSUM_KEY) {
360 		u64 prev_csum_end;
361 		u32 prev_item_size;
362 
363 		prev_item_size = btrfs_item_size_nr(leaf, slot - 1);
364 		prev_csum_end = (prev_item_size / csumsize) * sectorsize;
365 		prev_csum_end += prev_key->offset;
366 		if (prev_csum_end > key->offset) {
367 			generic_err(leaf, slot - 1,
368 "csum end range (%llu) goes beyond the start range (%llu) of the next csum item",
369 				    prev_csum_end, key->offset);
370 			return -EUCLEAN;
371 		}
372 	}
373 	return 0;
374 }
375 
376 /* Inode item error output has the same format as dir_item_err() */
377 #define inode_item_err(eb, slot, fmt, ...)			\
378 	dir_item_err(eb, slot, fmt, __VA_ARGS__)
379 
check_inode_key(struct extent_buffer * leaf,struct btrfs_key * key,int slot)380 static int check_inode_key(struct extent_buffer *leaf, struct btrfs_key *key,
381 			   int slot)
382 {
383 	struct btrfs_key item_key;
384 	bool is_inode_item;
385 
386 	btrfs_item_key_to_cpu(leaf, &item_key, slot);
387 	is_inode_item = (item_key.type == BTRFS_INODE_ITEM_KEY);
388 
389 	/* For XATTR_ITEM, location key should be all 0 */
390 	if (item_key.type == BTRFS_XATTR_ITEM_KEY) {
391 		if (key->type != 0 || key->objectid != 0 || key->offset != 0)
392 			return -EUCLEAN;
393 		return 0;
394 	}
395 
396 	if ((key->objectid < BTRFS_FIRST_FREE_OBJECTID ||
397 	     key->objectid > BTRFS_LAST_FREE_OBJECTID) &&
398 	    key->objectid != BTRFS_ROOT_TREE_DIR_OBJECTID &&
399 	    key->objectid != BTRFS_FREE_INO_OBJECTID) {
400 		if (is_inode_item) {
401 			generic_err(leaf, slot,
402 	"invalid key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
403 				key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
404 				BTRFS_FIRST_FREE_OBJECTID,
405 				BTRFS_LAST_FREE_OBJECTID,
406 				BTRFS_FREE_INO_OBJECTID);
407 		} else {
408 			dir_item_err(leaf, slot,
409 "invalid location key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
410 				key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
411 				BTRFS_FIRST_FREE_OBJECTID,
412 				BTRFS_LAST_FREE_OBJECTID,
413 				BTRFS_FREE_INO_OBJECTID);
414 		}
415 		return -EUCLEAN;
416 	}
417 	if (key->offset != 0) {
418 		if (is_inode_item)
419 			inode_item_err(leaf, slot,
420 				       "invalid key offset: has %llu expect 0",
421 				       key->offset);
422 		else
423 			dir_item_err(leaf, slot,
424 				"invalid location key offset:has %llu expect 0",
425 				key->offset);
426 		return -EUCLEAN;
427 	}
428 	return 0;
429 }
430 
check_root_key(struct extent_buffer * leaf,struct btrfs_key * key,int slot)431 static int check_root_key(struct extent_buffer *leaf, struct btrfs_key *key,
432 			  int slot)
433 {
434 	struct btrfs_key item_key;
435 	bool is_root_item;
436 
437 	btrfs_item_key_to_cpu(leaf, &item_key, slot);
438 	is_root_item = (item_key.type == BTRFS_ROOT_ITEM_KEY);
439 
440 	/* No such tree id */
441 	if (key->objectid == 0) {
442 		if (is_root_item)
443 			generic_err(leaf, slot, "invalid root id 0");
444 		else
445 			dir_item_err(leaf, slot,
446 				     "invalid location key root id 0");
447 		return -EUCLEAN;
448 	}
449 
450 	/* DIR_ITEM/INDEX/INODE_REF is not allowed to point to non-fs trees */
451 	if (!is_fstree(key->objectid) && !is_root_item) {
452 		dir_item_err(leaf, slot,
453 		"invalid location key objectid, have %llu expect [%llu, %llu]",
454 				key->objectid, BTRFS_FIRST_FREE_OBJECTID,
455 				BTRFS_LAST_FREE_OBJECTID);
456 		return -EUCLEAN;
457 	}
458 
459 	/*
460 	 * ROOT_ITEM with non-zero offset means this is a snapshot, created at
461 	 * @offset transid.
462 	 * Furthermore, for location key in DIR_ITEM, its offset is always -1.
463 	 *
464 	 * So here we only check offset for reloc tree whose key->offset must
465 	 * be a valid tree.
466 	 */
467 	if (key->objectid == BTRFS_TREE_RELOC_OBJECTID && key->offset == 0) {
468 		generic_err(leaf, slot, "invalid root id 0 for reloc tree");
469 		return -EUCLEAN;
470 	}
471 	return 0;
472 }
473 
check_dir_item(struct extent_buffer * leaf,struct btrfs_key * key,struct btrfs_key * prev_key,int slot)474 static int check_dir_item(struct extent_buffer *leaf,
475 			  struct btrfs_key *key, struct btrfs_key *prev_key,
476 			  int slot)
477 {
478 	struct btrfs_fs_info *fs_info = leaf->fs_info;
479 	struct btrfs_dir_item *di;
480 	u32 item_size = btrfs_item_size_nr(leaf, slot);
481 	u32 cur = 0;
482 
483 	if (!check_prev_ino(leaf, key, slot, prev_key))
484 		return -EUCLEAN;
485 	di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
486 	while (cur < item_size) {
487 		struct btrfs_key location_key;
488 		u32 name_len;
489 		u32 data_len;
490 		u32 max_name_len;
491 		u32 total_size;
492 		u32 name_hash;
493 		u8 dir_type;
494 		int ret;
495 
496 		/* header itself should not cross item boundary */
497 		if (cur + sizeof(*di) > item_size) {
498 			dir_item_err(leaf, slot,
499 		"dir item header crosses item boundary, have %zu boundary %u",
500 				cur + sizeof(*di), item_size);
501 			return -EUCLEAN;
502 		}
503 
504 		/* Location key check */
505 		btrfs_dir_item_key_to_cpu(leaf, di, &location_key);
506 		if (location_key.type == BTRFS_ROOT_ITEM_KEY) {
507 			ret = check_root_key(leaf, &location_key, slot);
508 			if (ret < 0)
509 				return ret;
510 		} else if (location_key.type == BTRFS_INODE_ITEM_KEY ||
511 			   location_key.type == 0) {
512 			ret = check_inode_key(leaf, &location_key, slot);
513 			if (ret < 0)
514 				return ret;
515 		} else {
516 			dir_item_err(leaf, slot,
517 			"invalid location key type, have %u, expect %u or %u",
518 				     location_key.type, BTRFS_ROOT_ITEM_KEY,
519 				     BTRFS_INODE_ITEM_KEY);
520 			return -EUCLEAN;
521 		}
522 
523 		/* dir type check */
524 		dir_type = btrfs_dir_type(leaf, di);
525 		if (dir_type >= BTRFS_FT_MAX) {
526 			dir_item_err(leaf, slot,
527 			"invalid dir item type, have %u expect [0, %u)",
528 				dir_type, BTRFS_FT_MAX);
529 			return -EUCLEAN;
530 		}
531 
532 		if (key->type == BTRFS_XATTR_ITEM_KEY &&
533 		    dir_type != BTRFS_FT_XATTR) {
534 			dir_item_err(leaf, slot,
535 		"invalid dir item type for XATTR key, have %u expect %u",
536 				dir_type, BTRFS_FT_XATTR);
537 			return -EUCLEAN;
538 		}
539 		if (dir_type == BTRFS_FT_XATTR &&
540 		    key->type != BTRFS_XATTR_ITEM_KEY) {
541 			dir_item_err(leaf, slot,
542 			"xattr dir type found for non-XATTR key");
543 			return -EUCLEAN;
544 		}
545 		if (dir_type == BTRFS_FT_XATTR)
546 			max_name_len = XATTR_NAME_MAX;
547 		else
548 			max_name_len = BTRFS_NAME_LEN;
549 
550 		/* Name/data length check */
551 		name_len = btrfs_dir_name_len(leaf, di);
552 		data_len = btrfs_dir_data_len(leaf, di);
553 		if (name_len > max_name_len) {
554 			dir_item_err(leaf, slot,
555 			"dir item name len too long, have %u max %u",
556 				name_len, max_name_len);
557 			return -EUCLEAN;
558 		}
559 		if (name_len + data_len > BTRFS_MAX_XATTR_SIZE(fs_info)) {
560 			dir_item_err(leaf, slot,
561 			"dir item name and data len too long, have %u max %u",
562 				name_len + data_len,
563 				BTRFS_MAX_XATTR_SIZE(fs_info));
564 			return -EUCLEAN;
565 		}
566 
567 		if (data_len && dir_type != BTRFS_FT_XATTR) {
568 			dir_item_err(leaf, slot,
569 			"dir item with invalid data len, have %u expect 0",
570 				data_len);
571 			return -EUCLEAN;
572 		}
573 
574 		total_size = sizeof(*di) + name_len + data_len;
575 
576 		/* header and name/data should not cross item boundary */
577 		if (cur + total_size > item_size) {
578 			dir_item_err(leaf, slot,
579 		"dir item data crosses item boundary, have %u boundary %u",
580 				cur + total_size, item_size);
581 			return -EUCLEAN;
582 		}
583 
584 		/*
585 		 * Special check for XATTR/DIR_ITEM, as key->offset is name
586 		 * hash, should match its name
587 		 */
588 		if (key->type == BTRFS_DIR_ITEM_KEY ||
589 		    key->type == BTRFS_XATTR_ITEM_KEY) {
590 			char namebuf[max(BTRFS_NAME_LEN, XATTR_NAME_MAX)];
591 
592 			read_extent_buffer(leaf, namebuf,
593 					(unsigned long)(di + 1), name_len);
594 			name_hash = btrfs_name_hash(namebuf, name_len);
595 			if (key->offset != name_hash) {
596 				dir_item_err(leaf, slot,
597 		"name hash mismatch with key, have 0x%016x expect 0x%016llx",
598 					name_hash, key->offset);
599 				return -EUCLEAN;
600 			}
601 		}
602 		cur += total_size;
603 		di = (struct btrfs_dir_item *)((void *)di + total_size);
604 	}
605 	return 0;
606 }
607 
608 __printf(3, 4)
609 __cold
block_group_err(const struct extent_buffer * eb,int slot,const char * fmt,...)610 static void block_group_err(const struct extent_buffer *eb, int slot,
611 			    const char *fmt, ...)
612 {
613 	const struct btrfs_fs_info *fs_info = eb->fs_info;
614 	struct btrfs_key key;
615 	struct va_format vaf;
616 	va_list args;
617 
618 	btrfs_item_key_to_cpu(eb, &key, slot);
619 	va_start(args, fmt);
620 
621 	vaf.fmt = fmt;
622 	vaf.va = &args;
623 
624 	btrfs_crit(fs_info,
625 	"corrupt %s: root=%llu block=%llu slot=%d bg_start=%llu bg_len=%llu, %pV",
626 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
627 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
628 		key.objectid, key.offset, &vaf);
629 	va_end(args);
630 }
631 
check_block_group_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot)632 static int check_block_group_item(struct extent_buffer *leaf,
633 				  struct btrfs_key *key, int slot)
634 {
635 	struct btrfs_block_group_item bgi;
636 	u32 item_size = btrfs_item_size_nr(leaf, slot);
637 	u64 flags;
638 	u64 type;
639 
640 	/*
641 	 * Here we don't really care about alignment since extent allocator can
642 	 * handle it.  We care more about the size.
643 	 */
644 	if (key->offset == 0) {
645 		block_group_err(leaf, slot,
646 				"invalid block group size 0");
647 		return -EUCLEAN;
648 	}
649 
650 	if (item_size != sizeof(bgi)) {
651 		block_group_err(leaf, slot,
652 			"invalid item size, have %u expect %zu",
653 				item_size, sizeof(bgi));
654 		return -EUCLEAN;
655 	}
656 
657 	read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot),
658 			   sizeof(bgi));
659 	if (btrfs_stack_block_group_chunk_objectid(&bgi) !=
660 	    BTRFS_FIRST_CHUNK_TREE_OBJECTID) {
661 		block_group_err(leaf, slot,
662 		"invalid block group chunk objectid, have %llu expect %llu",
663 				btrfs_stack_block_group_chunk_objectid(&bgi),
664 				BTRFS_FIRST_CHUNK_TREE_OBJECTID);
665 		return -EUCLEAN;
666 	}
667 
668 	if (btrfs_stack_block_group_used(&bgi) > key->offset) {
669 		block_group_err(leaf, slot,
670 			"invalid block group used, have %llu expect [0, %llu)",
671 				btrfs_stack_block_group_used(&bgi), key->offset);
672 		return -EUCLEAN;
673 	}
674 
675 	flags = btrfs_stack_block_group_flags(&bgi);
676 	if (hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) > 1) {
677 		block_group_err(leaf, slot,
678 "invalid profile flags, have 0x%llx (%lu bits set) expect no more than 1 bit set",
679 			flags & BTRFS_BLOCK_GROUP_PROFILE_MASK,
680 			hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK));
681 		return -EUCLEAN;
682 	}
683 
684 	type = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
685 	if (type != BTRFS_BLOCK_GROUP_DATA &&
686 	    type != BTRFS_BLOCK_GROUP_METADATA &&
687 	    type != BTRFS_BLOCK_GROUP_SYSTEM &&
688 	    type != (BTRFS_BLOCK_GROUP_METADATA |
689 			   BTRFS_BLOCK_GROUP_DATA)) {
690 		block_group_err(leaf, slot,
691 "invalid type, have 0x%llx (%lu bits set) expect either 0x%llx, 0x%llx, 0x%llx or 0x%llx",
692 			type, hweight64(type),
693 			BTRFS_BLOCK_GROUP_DATA, BTRFS_BLOCK_GROUP_METADATA,
694 			BTRFS_BLOCK_GROUP_SYSTEM,
695 			BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA);
696 		return -EUCLEAN;
697 	}
698 	return 0;
699 }
700 
701 __printf(4, 5)
702 __cold
chunk_err(const struct extent_buffer * leaf,const struct btrfs_chunk * chunk,u64 logical,const char * fmt,...)703 static void chunk_err(const struct extent_buffer *leaf,
704 		      const struct btrfs_chunk *chunk, u64 logical,
705 		      const char *fmt, ...)
706 {
707 	const struct btrfs_fs_info *fs_info = leaf->fs_info;
708 	bool is_sb;
709 	struct va_format vaf;
710 	va_list args;
711 	int i;
712 	int slot = -1;
713 
714 	/* Only superblock eb is able to have such small offset */
715 	is_sb = (leaf->start == BTRFS_SUPER_INFO_OFFSET);
716 
717 	if (!is_sb) {
718 		/*
719 		 * Get the slot number by iterating through all slots, this
720 		 * would provide better readability.
721 		 */
722 		for (i = 0; i < btrfs_header_nritems(leaf); i++) {
723 			if (btrfs_item_ptr_offset(leaf, i) ==
724 					(unsigned long)chunk) {
725 				slot = i;
726 				break;
727 			}
728 		}
729 	}
730 	va_start(args, fmt);
731 	vaf.fmt = fmt;
732 	vaf.va = &args;
733 
734 	if (is_sb)
735 		btrfs_crit(fs_info,
736 		"corrupt superblock syschunk array: chunk_start=%llu, %pV",
737 			   logical, &vaf);
738 	else
739 		btrfs_crit(fs_info,
740 	"corrupt leaf: root=%llu block=%llu slot=%d chunk_start=%llu, %pV",
741 			   BTRFS_CHUNK_TREE_OBJECTID, leaf->start, slot,
742 			   logical, &vaf);
743 	va_end(args);
744 }
745 
746 /*
747  * The common chunk check which could also work on super block sys chunk array.
748  *
749  * Return -EUCLEAN if anything is corrupted.
750  * Return 0 if everything is OK.
751  */
btrfs_check_chunk_valid(struct extent_buffer * leaf,struct btrfs_chunk * chunk,u64 logical)752 int btrfs_check_chunk_valid(struct extent_buffer *leaf,
753 			    struct btrfs_chunk *chunk, u64 logical)
754 {
755 	struct btrfs_fs_info *fs_info = leaf->fs_info;
756 	u64 length;
757 	u64 chunk_end;
758 	u64 stripe_len;
759 	u16 num_stripes;
760 	u16 sub_stripes;
761 	u64 type;
762 	u64 features;
763 	bool mixed = false;
764 	int raid_index;
765 	int nparity;
766 	int ncopies;
767 
768 	length = btrfs_chunk_length(leaf, chunk);
769 	stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
770 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
771 	sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
772 	type = btrfs_chunk_type(leaf, chunk);
773 	raid_index = btrfs_bg_flags_to_raid_index(type);
774 	ncopies = btrfs_raid_array[raid_index].ncopies;
775 	nparity = btrfs_raid_array[raid_index].nparity;
776 
777 	if (!num_stripes) {
778 		chunk_err(leaf, chunk, logical,
779 			  "invalid chunk num_stripes, have %u", num_stripes);
780 		return -EUCLEAN;
781 	}
782 	if (num_stripes < ncopies) {
783 		chunk_err(leaf, chunk, logical,
784 			  "invalid chunk num_stripes < ncopies, have %u < %d",
785 			  num_stripes, ncopies);
786 		return -EUCLEAN;
787 	}
788 	if (nparity && num_stripes == nparity) {
789 		chunk_err(leaf, chunk, logical,
790 			  "invalid chunk num_stripes == nparity, have %u == %d",
791 			  num_stripes, nparity);
792 		return -EUCLEAN;
793 	}
794 	if (!IS_ALIGNED(logical, fs_info->sectorsize)) {
795 		chunk_err(leaf, chunk, logical,
796 		"invalid chunk logical, have %llu should aligned to %u",
797 			  logical, fs_info->sectorsize);
798 		return -EUCLEAN;
799 	}
800 	if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) {
801 		chunk_err(leaf, chunk, logical,
802 			  "invalid chunk sectorsize, have %u expect %u",
803 			  btrfs_chunk_sector_size(leaf, chunk),
804 			  fs_info->sectorsize);
805 		return -EUCLEAN;
806 	}
807 	if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) {
808 		chunk_err(leaf, chunk, logical,
809 			  "invalid chunk length, have %llu", length);
810 		return -EUCLEAN;
811 	}
812 	if (unlikely(check_add_overflow(logical, length, &chunk_end))) {
813 		chunk_err(leaf, chunk, logical,
814 "invalid chunk logical start and length, have logical start %llu length %llu",
815 			  logical, length);
816 		return -EUCLEAN;
817 	}
818 	if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
819 		chunk_err(leaf, chunk, logical,
820 			  "invalid chunk stripe length: %llu",
821 			  stripe_len);
822 		return -EUCLEAN;
823 	}
824 	if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
825 	    type) {
826 		chunk_err(leaf, chunk, logical,
827 			  "unrecognized chunk type: 0x%llx",
828 			  ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
829 			    BTRFS_BLOCK_GROUP_PROFILE_MASK) &
830 			  btrfs_chunk_type(leaf, chunk));
831 		return -EUCLEAN;
832 	}
833 
834 	if (!has_single_bit_set(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
835 	    (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) != 0) {
836 		chunk_err(leaf, chunk, logical,
837 		"invalid chunk profile flag: 0x%llx, expect 0 or 1 bit set",
838 			  type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
839 		return -EUCLEAN;
840 	}
841 	if ((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0) {
842 		chunk_err(leaf, chunk, logical,
843 	"missing chunk type flag, have 0x%llx one bit must be set in 0x%llx",
844 			  type, BTRFS_BLOCK_GROUP_TYPE_MASK);
845 		return -EUCLEAN;
846 	}
847 
848 	if ((type & BTRFS_BLOCK_GROUP_SYSTEM) &&
849 	    (type & (BTRFS_BLOCK_GROUP_METADATA | 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;
854 	}
855 
856 	features = btrfs_super_incompat_flags(fs_info->super_copy);
857 	if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
858 		mixed = true;
859 
860 	if (!mixed) {
861 		if ((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;
866 		}
867 	}
868 
869 	if ((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 && num_stripes != 1)) {
875 		chunk_err(leaf, chunk, logical,
876 			"invalid num_stripes:sub_stripes %u:%u for profile %llu",
877 			num_stripes, sub_stripes,
878 			type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
879 		return -EUCLEAN;
880 	}
881 
882 	return 0;
883 }
884 
885 /*
886  * Enhanced version of chunk item checker.
887  *
888  * The common btrfs_check_chunk_valid() doesn't check item size since it needs
889  * to work on super block sys_chunk_array which doesn't have full item ptr.
890  */
check_leaf_chunk_item(struct extent_buffer * leaf,struct btrfs_chunk * chunk,struct btrfs_key * key,int slot)891 static int check_leaf_chunk_item(struct extent_buffer *leaf,
892 				 struct btrfs_chunk *chunk,
893 				 struct btrfs_key *key, int slot)
894 {
895 	int num_stripes;
896 
897 	if (btrfs_item_size_nr(leaf, slot) < sizeof(struct btrfs_chunk)) {
898 		chunk_err(leaf, chunk, key->offset,
899 			"invalid chunk item size: have %u expect [%zu, %u)",
900 			btrfs_item_size_nr(leaf, slot),
901 			sizeof(struct btrfs_chunk),
902 			BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
903 		return -EUCLEAN;
904 	}
905 
906 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
907 	/* Let btrfs_check_chunk_valid() handle this error type */
908 	if (num_stripes == 0)
909 		goto out;
910 
911 	if (btrfs_chunk_item_size(num_stripes) !=
912 	    btrfs_item_size_nr(leaf, slot)) {
913 		chunk_err(leaf, chunk, key->offset,
914 			"invalid chunk item size: have %u expect %lu",
915 			btrfs_item_size_nr(leaf, slot),
916 			btrfs_chunk_item_size(num_stripes));
917 		return -EUCLEAN;
918 	}
919 out:
920 	return btrfs_check_chunk_valid(leaf, chunk, key->offset);
921 }
922 
923 __printf(3, 4)
924 __cold
dev_item_err(const struct extent_buffer * eb,int slot,const char * fmt,...)925 static void dev_item_err(const struct extent_buffer *eb, int slot,
926 			 const char *fmt, ...)
927 {
928 	struct btrfs_key key;
929 	struct va_format vaf;
930 	va_list args;
931 
932 	btrfs_item_key_to_cpu(eb, &key, slot);
933 	va_start(args, fmt);
934 
935 	vaf.fmt = fmt;
936 	vaf.va = &args;
937 
938 	btrfs_crit(eb->fs_info,
939 	"corrupt %s: root=%llu block=%llu slot=%d devid=%llu %pV",
940 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
941 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
942 		key.objectid, &vaf);
943 	va_end(args);
944 }
945 
check_dev_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot)946 static int check_dev_item(struct extent_buffer *leaf,
947 			  struct btrfs_key *key, int slot)
948 {
949 	struct btrfs_dev_item *ditem;
950 	const u32 item_size = btrfs_item_size_nr(leaf, slot);
951 
952 	if (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;
957 	}
958 
959 	if (unlikely(item_size != sizeof(*ditem))) {
960 		dev_item_err(leaf, slot, "invalid item size: has %u expect %zu",
961 			     item_size, sizeof(*ditem));
962 		return -EUCLEAN;
963 	}
964 
965 	ditem = btrfs_item_ptr(leaf, slot, struct btrfs_dev_item);
966 	if (btrfs_device_id(leaf, ditem) != key->offset) {
967 		dev_item_err(leaf, slot,
968 			     "devid mismatch: key has=%llu item has=%llu",
969 			     key->offset, btrfs_device_id(leaf, ditem));
970 		return -EUCLEAN;
971 	}
972 
973 	/*
974 	 * For device total_bytes, we don't have reliable way to check it, as
975 	 * it can be 0 for device removal. Device size check can only be done
976 	 * by dev extents check.
977 	 */
978 	if (btrfs_device_bytes_used(leaf, ditem) >
979 	    btrfs_device_total_bytes(leaf, ditem)) {
980 		dev_item_err(leaf, slot,
981 			     "invalid bytes used: have %llu expect [0, %llu]",
982 			     btrfs_device_bytes_used(leaf, ditem),
983 			     btrfs_device_total_bytes(leaf, ditem));
984 		return -EUCLEAN;
985 	}
986 	/*
987 	 * Remaining members like io_align/type/gen/dev_group aren't really
988 	 * utilized.  Skip them to make later usage of them easier.
989 	 */
990 	return 0;
991 }
992 
check_inode_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot)993 static int check_inode_item(struct extent_buffer *leaf,
994 			    struct btrfs_key *key, int slot)
995 {
996 	struct btrfs_fs_info *fs_info = leaf->fs_info;
997 	struct btrfs_inode_item *iitem;
998 	u64 super_gen = btrfs_super_generation(fs_info->super_copy);
999 	u32 valid_mask = (S_IFMT | S_ISUID | S_ISGID | S_ISVTX | 0777);
1000 	const u32 item_size = btrfs_item_size_nr(leaf, slot);
1001 	u32 mode;
1002 	int ret;
1003 
1004 	ret = check_inode_key(leaf, key, slot);
1005 	if (ret < 0)
1006 		return ret;
1007 
1008 	if (unlikely(item_size != sizeof(*iitem))) {
1009 		generic_err(leaf, slot, "invalid item size: has %u expect %zu",
1010 			    item_size, sizeof(*iitem));
1011 		return -EUCLEAN;
1012 	}
1013 
1014 	iitem = btrfs_item_ptr(leaf, slot, struct btrfs_inode_item);
1015 
1016 	/* Here we use super block generation + 1 to handle log tree */
1017 	if (btrfs_inode_generation(leaf, iitem) > super_gen + 1) {
1018 		inode_item_err(leaf, slot,
1019 			"invalid inode generation: has %llu expect (0, %llu]",
1020 			       btrfs_inode_generation(leaf, iitem),
1021 			       super_gen + 1);
1022 		return -EUCLEAN;
1023 	}
1024 	/* Note for ROOT_TREE_DIR_ITEM, mkfs could set its transid 0 */
1025 	if (btrfs_inode_transid(leaf, iitem) > super_gen + 1) {
1026 		inode_item_err(leaf, slot,
1027 			"invalid inode transid: has %llu expect [0, %llu]",
1028 			       btrfs_inode_transid(leaf, iitem), super_gen + 1);
1029 		return -EUCLEAN;
1030 	}
1031 
1032 	/*
1033 	 * For size and nbytes it's better not to be too strict, as for dir
1034 	 * item its size/nbytes can easily get wrong, but doesn't affect
1035 	 * anything in the fs. So here we skip the check.
1036 	 */
1037 	mode = btrfs_inode_mode(leaf, iitem);
1038 	if (mode & ~valid_mask) {
1039 		inode_item_err(leaf, slot,
1040 			       "unknown mode bit detected: 0x%x",
1041 			       mode & ~valid_mask);
1042 		return -EUCLEAN;
1043 	}
1044 
1045 	/*
1046 	 * S_IFMT is not bit mapped so we can't completely rely on
1047 	 * is_power_of_2/has_single_bit_set, but it can save us from checking
1048 	 * FIFO/CHR/DIR/REG.  Only needs to check BLK, LNK and SOCKS
1049 	 */
1050 	if (!has_single_bit_set(mode & S_IFMT)) {
1051 		if (!S_ISLNK(mode) && !S_ISBLK(mode) && !S_ISSOCK(mode)) {
1052 			inode_item_err(leaf, slot,
1053 			"invalid mode: has 0%o expect valid S_IF* bit(s)",
1054 				       mode & S_IFMT);
1055 			return -EUCLEAN;
1056 		}
1057 	}
1058 	if (S_ISDIR(mode) && btrfs_inode_nlink(leaf, iitem) > 1) {
1059 		inode_item_err(leaf, slot,
1060 		       "invalid nlink: has %u expect no more than 1 for dir",
1061 			btrfs_inode_nlink(leaf, iitem));
1062 		return -EUCLEAN;
1063 	}
1064 	if (btrfs_inode_flags(leaf, iitem) & ~BTRFS_INODE_FLAG_MASK) {
1065 		inode_item_err(leaf, slot,
1066 			       "unknown flags detected: 0x%llx",
1067 			       btrfs_inode_flags(leaf, iitem) &
1068 			       ~BTRFS_INODE_FLAG_MASK);
1069 		return -EUCLEAN;
1070 	}
1071 	return 0;
1072 }
1073 
check_root_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot)1074 static int check_root_item(struct extent_buffer *leaf, struct btrfs_key *key,
1075 			   int slot)
1076 {
1077 	struct btrfs_fs_info *fs_info = leaf->fs_info;
1078 	struct btrfs_root_item ri = { 0 };
1079 	const u64 valid_root_flags = BTRFS_ROOT_SUBVOL_RDONLY |
1080 				     BTRFS_ROOT_SUBVOL_DEAD;
1081 	int ret;
1082 
1083 	ret = check_root_key(leaf, key, slot);
1084 	if (ret < 0)
1085 		return ret;
1086 
1087 	if (btrfs_item_size_nr(leaf, slot) != sizeof(ri) &&
1088 	    btrfs_item_size_nr(leaf, slot) != btrfs_legacy_root_item_size()) {
1089 		generic_err(leaf, slot,
1090 			    "invalid root item size, have %u expect %zu or %u",
1091 			    btrfs_item_size_nr(leaf, slot), sizeof(ri),
1092 			    btrfs_legacy_root_item_size());
1093 		return -EUCLEAN;
1094 	}
1095 
1096 	/*
1097 	 * For legacy root item, the members starting at generation_v2 will be
1098 	 * all filled with 0.
1099 	 * And since we allow geneartion_v2 as 0, it will still pass the check.
1100 	 */
1101 	read_extent_buffer(leaf, &ri, btrfs_item_ptr_offset(leaf, slot),
1102 			   btrfs_item_size_nr(leaf, slot));
1103 
1104 	/* Generation related */
1105 	if (btrfs_root_generation(&ri) >
1106 	    btrfs_super_generation(fs_info->super_copy) + 1) {
1107 		generic_err(leaf, slot,
1108 			"invalid root generation, have %llu expect (0, %llu]",
1109 			    btrfs_root_generation(&ri),
1110 			    btrfs_super_generation(fs_info->super_copy) + 1);
1111 		return -EUCLEAN;
1112 	}
1113 	if (btrfs_root_generation_v2(&ri) >
1114 	    btrfs_super_generation(fs_info->super_copy) + 1) {
1115 		generic_err(leaf, slot,
1116 		"invalid root v2 generation, have %llu expect (0, %llu]",
1117 			    btrfs_root_generation_v2(&ri),
1118 			    btrfs_super_generation(fs_info->super_copy) + 1);
1119 		return -EUCLEAN;
1120 	}
1121 	if (btrfs_root_last_snapshot(&ri) >
1122 	    btrfs_super_generation(fs_info->super_copy) + 1) {
1123 		generic_err(leaf, slot,
1124 		"invalid root last_snapshot, have %llu expect (0, %llu]",
1125 			    btrfs_root_last_snapshot(&ri),
1126 			    btrfs_super_generation(fs_info->super_copy) + 1);
1127 		return -EUCLEAN;
1128 	}
1129 
1130 	/* Alignment and level check */
1131 	if (!IS_ALIGNED(btrfs_root_bytenr(&ri), fs_info->sectorsize)) {
1132 		generic_err(leaf, slot,
1133 		"invalid root bytenr, have %llu expect to be aligned to %u",
1134 			    btrfs_root_bytenr(&ri), fs_info->sectorsize);
1135 		return -EUCLEAN;
1136 	}
1137 	if (btrfs_root_level(&ri) >= BTRFS_MAX_LEVEL) {
1138 		generic_err(leaf, slot,
1139 			    "invalid root level, have %u expect [0, %u]",
1140 			    btrfs_root_level(&ri), BTRFS_MAX_LEVEL - 1);
1141 		return -EUCLEAN;
1142 	}
1143 	if (ri.drop_level >= BTRFS_MAX_LEVEL) {
1144 		generic_err(leaf, slot,
1145 			    "invalid root level, have %u expect [0, %u]",
1146 			    ri.drop_level, BTRFS_MAX_LEVEL - 1);
1147 		return -EUCLEAN;
1148 	}
1149 
1150 	/* Flags check */
1151 	if (btrfs_root_flags(&ri) & ~valid_root_flags) {
1152 		generic_err(leaf, slot,
1153 			    "invalid root flags, have 0x%llx expect mask 0x%llx",
1154 			    btrfs_root_flags(&ri), valid_root_flags);
1155 		return -EUCLEAN;
1156 	}
1157 	return 0;
1158 }
1159 
1160 __printf(3,4)
1161 __cold
extent_err(const struct extent_buffer * eb,int slot,const char * fmt,...)1162 static void extent_err(const struct extent_buffer *eb, int slot,
1163 		       const char *fmt, ...)
1164 {
1165 	struct btrfs_key key;
1166 	struct va_format vaf;
1167 	va_list args;
1168 	u64 bytenr;
1169 	u64 len;
1170 
1171 	btrfs_item_key_to_cpu(eb, &key, slot);
1172 	bytenr = key.objectid;
1173 	if (key.type == BTRFS_METADATA_ITEM_KEY ||
1174 	    key.type == BTRFS_TREE_BLOCK_REF_KEY ||
1175 	    key.type == BTRFS_SHARED_BLOCK_REF_KEY)
1176 		len = eb->fs_info->nodesize;
1177 	else
1178 		len = key.offset;
1179 	va_start(args, fmt);
1180 
1181 	vaf.fmt = fmt;
1182 	vaf.va = &args;
1183 
1184 	btrfs_crit(eb->fs_info,
1185 	"corrupt %s: block=%llu slot=%d extent bytenr=%llu len=%llu %pV",
1186 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
1187 		eb->start, slot, bytenr, len, &vaf);
1188 	va_end(args);
1189 }
1190 
check_extent_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot,struct btrfs_key * prev_key)1191 static int check_extent_item(struct extent_buffer *leaf,
1192 			     struct btrfs_key *key, int slot,
1193 			     struct btrfs_key *prev_key)
1194 {
1195 	struct btrfs_fs_info *fs_info = leaf->fs_info;
1196 	struct btrfs_extent_item *ei;
1197 	bool is_tree_block = false;
1198 	unsigned long ptr;	/* Current pointer inside inline refs */
1199 	unsigned long end;	/* Extent item end */
1200 	const u32 item_size = btrfs_item_size_nr(leaf, slot);
1201 	u64 flags;
1202 	u64 generation;
1203 	u64 total_refs;		/* Total refs in btrfs_extent_item */
1204 	u64 inline_refs = 0;	/* found total inline refs */
1205 
1206 	if (key->type == BTRFS_METADATA_ITEM_KEY &&
1207 	    !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
1208 		generic_err(leaf, slot,
1209 "invalid key type, METADATA_ITEM type invalid when SKINNY_METADATA feature disabled");
1210 		return -EUCLEAN;
1211 	}
1212 	/* key->objectid is the bytenr for both key types */
1213 	if (!IS_ALIGNED(key->objectid, fs_info->sectorsize)) {
1214 		generic_err(leaf, slot,
1215 		"invalid key objectid, have %llu expect to be aligned to %u",
1216 			   key->objectid, fs_info->sectorsize);
1217 		return -EUCLEAN;
1218 	}
1219 
1220 	/* key->offset is tree level for METADATA_ITEM_KEY */
1221 	if (key->type == BTRFS_METADATA_ITEM_KEY &&
1222 	    key->offset >= BTRFS_MAX_LEVEL) {
1223 		extent_err(leaf, slot,
1224 			   "invalid tree level, have %llu expect [0, %u]",
1225 			   key->offset, BTRFS_MAX_LEVEL - 1);
1226 		return -EUCLEAN;
1227 	}
1228 
1229 	/*
1230 	 * EXTENT/METADATA_ITEM consists of:
1231 	 * 1) One btrfs_extent_item
1232 	 *    Records the total refs, type and generation of the extent.
1233 	 *
1234 	 * 2) One btrfs_tree_block_info (for EXTENT_ITEM and tree backref only)
1235 	 *    Records the first key and level of the tree block.
1236 	 *
1237 	 * 2) Zero or more btrfs_extent_inline_ref(s)
1238 	 *    Each inline ref has one btrfs_extent_inline_ref shows:
1239 	 *    2.1) The ref type, one of the 4
1240 	 *         TREE_BLOCK_REF	Tree block only
1241 	 *         SHARED_BLOCK_REF	Tree block only
1242 	 *         EXTENT_DATA_REF	Data only
1243 	 *         SHARED_DATA_REF	Data only
1244 	 *    2.2) Ref type specific data
1245 	 *         Either using btrfs_extent_inline_ref::offset, or specific
1246 	 *         data structure.
1247 	 */
1248 	if (item_size < sizeof(*ei)) {
1249 		extent_err(leaf, slot,
1250 			   "invalid item size, have %u expect [%zu, %u)",
1251 			   item_size, sizeof(*ei),
1252 			   BTRFS_LEAF_DATA_SIZE(fs_info));
1253 		return -EUCLEAN;
1254 	}
1255 	end = item_size + btrfs_item_ptr_offset(leaf, slot);
1256 
1257 	/* Checks against extent_item */
1258 	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
1259 	flags = btrfs_extent_flags(leaf, ei);
1260 	total_refs = btrfs_extent_refs(leaf, ei);
1261 	generation = btrfs_extent_generation(leaf, ei);
1262 	if (generation > btrfs_super_generation(fs_info->super_copy) + 1) {
1263 		extent_err(leaf, slot,
1264 			   "invalid generation, have %llu expect (0, %llu]",
1265 			   generation,
1266 			   btrfs_super_generation(fs_info->super_copy) + 1);
1267 		return -EUCLEAN;
1268 	}
1269 	if (!has_single_bit_set(flags & (BTRFS_EXTENT_FLAG_DATA |
1270 					 BTRFS_EXTENT_FLAG_TREE_BLOCK))) {
1271 		extent_err(leaf, slot,
1272 		"invalid extent flag, have 0x%llx expect 1 bit set in 0x%llx",
1273 			flags, BTRFS_EXTENT_FLAG_DATA |
1274 			BTRFS_EXTENT_FLAG_TREE_BLOCK);
1275 		return -EUCLEAN;
1276 	}
1277 	is_tree_block = !!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK);
1278 	if (is_tree_block) {
1279 		if (key->type == BTRFS_EXTENT_ITEM_KEY &&
1280 		    key->offset != fs_info->nodesize) {
1281 			extent_err(leaf, slot,
1282 				   "invalid extent length, have %llu expect %u",
1283 				   key->offset, fs_info->nodesize);
1284 			return -EUCLEAN;
1285 		}
1286 	} else {
1287 		if (key->type != BTRFS_EXTENT_ITEM_KEY) {
1288 			extent_err(leaf, slot,
1289 			"invalid key type, have %u expect %u for data backref",
1290 				   key->type, BTRFS_EXTENT_ITEM_KEY);
1291 			return -EUCLEAN;
1292 		}
1293 		if (!IS_ALIGNED(key->offset, fs_info->sectorsize)) {
1294 			extent_err(leaf, slot,
1295 			"invalid extent length, have %llu expect aligned to %u",
1296 				   key->offset, fs_info->sectorsize);
1297 			return -EUCLEAN;
1298 		}
1299 	}
1300 	ptr = (unsigned long)(struct btrfs_extent_item *)(ei + 1);
1301 
1302 	/* Check the special case of btrfs_tree_block_info */
1303 	if (is_tree_block && key->type != BTRFS_METADATA_ITEM_KEY) {
1304 		struct btrfs_tree_block_info *info;
1305 
1306 		info = (struct btrfs_tree_block_info *)ptr;
1307 		if (btrfs_tree_block_level(leaf, info) >= BTRFS_MAX_LEVEL) {
1308 			extent_err(leaf, slot,
1309 			"invalid tree block info level, have %u expect [0, %u]",
1310 				   btrfs_tree_block_level(leaf, info),
1311 				   BTRFS_MAX_LEVEL - 1);
1312 			return -EUCLEAN;
1313 		}
1314 		ptr = (unsigned long)(struct btrfs_tree_block_info *)(info + 1);
1315 	}
1316 
1317 	/* Check inline refs */
1318 	while (ptr < end) {
1319 		struct btrfs_extent_inline_ref *iref;
1320 		struct btrfs_extent_data_ref *dref;
1321 		struct btrfs_shared_data_ref *sref;
1322 		u64 dref_offset;
1323 		u64 inline_offset;
1324 		u8 inline_type;
1325 
1326 		if (ptr + sizeof(*iref) > end) {
1327 			extent_err(leaf, slot,
1328 "inline ref item overflows extent item, ptr %lu iref size %zu end %lu",
1329 				   ptr, sizeof(*iref), end);
1330 			return -EUCLEAN;
1331 		}
1332 		iref = (struct btrfs_extent_inline_ref *)ptr;
1333 		inline_type = btrfs_extent_inline_ref_type(leaf, iref);
1334 		inline_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1335 		if (ptr + btrfs_extent_inline_ref_size(inline_type) > end) {
1336 			extent_err(leaf, slot,
1337 "inline ref item overflows extent item, ptr %lu iref size %u end %lu",
1338 				   ptr, btrfs_extent_inline_ref_size(inline_type), end);
1339 			return -EUCLEAN;
1340 		}
1341 
1342 		switch (inline_type) {
1343 		/* inline_offset is subvolid of the owner, no need to check */
1344 		case BTRFS_TREE_BLOCK_REF_KEY:
1345 			inline_refs++;
1346 			break;
1347 		/* Contains parent bytenr */
1348 		case BTRFS_SHARED_BLOCK_REF_KEY:
1349 			if (!IS_ALIGNED(inline_offset, fs_info->sectorsize)) {
1350 				extent_err(leaf, slot,
1351 		"invalid tree parent bytenr, have %llu expect aligned to %u",
1352 					   inline_offset, fs_info->sectorsize);
1353 				return -EUCLEAN;
1354 			}
1355 			inline_refs++;
1356 			break;
1357 		/*
1358 		 * Contains owner subvolid, owner key objectid, adjusted offset.
1359 		 * The only obvious corruption can happen in that offset.
1360 		 */
1361 		case BTRFS_EXTENT_DATA_REF_KEY:
1362 			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1363 			dref_offset = btrfs_extent_data_ref_offset(leaf, dref);
1364 			if (!IS_ALIGNED(dref_offset, fs_info->sectorsize)) {
1365 				extent_err(leaf, slot,
1366 		"invalid data ref offset, have %llu expect aligned to %u",
1367 					   dref_offset, fs_info->sectorsize);
1368 				return -EUCLEAN;
1369 			}
1370 			inline_refs += btrfs_extent_data_ref_count(leaf, dref);
1371 			break;
1372 		/* Contains parent bytenr and ref count */
1373 		case BTRFS_SHARED_DATA_REF_KEY:
1374 			sref = (struct btrfs_shared_data_ref *)(iref + 1);
1375 			if (!IS_ALIGNED(inline_offset, fs_info->sectorsize)) {
1376 				extent_err(leaf, slot,
1377 		"invalid data parent bytenr, have %llu expect aligned to %u",
1378 					   inline_offset, fs_info->sectorsize);
1379 				return -EUCLEAN;
1380 			}
1381 			inline_refs += btrfs_shared_data_ref_count(leaf, sref);
1382 			break;
1383 		default:
1384 			extent_err(leaf, slot, "unknown inline ref type: %u",
1385 				   inline_type);
1386 			return -EUCLEAN;
1387 		}
1388 		ptr += btrfs_extent_inline_ref_size(inline_type);
1389 	}
1390 	/* No padding is allowed */
1391 	if (ptr != end) {
1392 		extent_err(leaf, slot,
1393 			   "invalid extent item size, padding bytes found");
1394 		return -EUCLEAN;
1395 	}
1396 
1397 	/* Finally, check the inline refs against total refs */
1398 	if (inline_refs > total_refs) {
1399 		extent_err(leaf, slot,
1400 			"invalid extent refs, have %llu expect >= inline %llu",
1401 			   total_refs, inline_refs);
1402 		return -EUCLEAN;
1403 	}
1404 
1405 	if ((prev_key->type == BTRFS_EXTENT_ITEM_KEY) ||
1406 	    (prev_key->type == BTRFS_METADATA_ITEM_KEY)) {
1407 		u64 prev_end = prev_key->objectid;
1408 
1409 		if (prev_key->type == BTRFS_METADATA_ITEM_KEY)
1410 			prev_end += fs_info->nodesize;
1411 		else
1412 			prev_end += prev_key->offset;
1413 
1414 		if (unlikely(prev_end > key->objectid)) {
1415 			extent_err(leaf, slot,
1416 	"previous extent [%llu %u %llu] overlaps current extent [%llu %u %llu]",
1417 				   prev_key->objectid, prev_key->type,
1418 				   prev_key->offset, key->objectid, key->type,
1419 				   key->offset);
1420 			return -EUCLEAN;
1421 		}
1422 	}
1423 
1424 	return 0;
1425 }
1426 
check_simple_keyed_refs(struct extent_buffer * leaf,struct btrfs_key * key,int slot)1427 static int check_simple_keyed_refs(struct extent_buffer *leaf,
1428 				   struct btrfs_key *key, int slot)
1429 {
1430 	u32 expect_item_size = 0;
1431 
1432 	if (key->type == BTRFS_SHARED_DATA_REF_KEY)
1433 		expect_item_size = sizeof(struct btrfs_shared_data_ref);
1434 
1435 	if (btrfs_item_size_nr(leaf, slot) != expect_item_size) {
1436 		generic_err(leaf, slot,
1437 		"invalid item size, have %u expect %u for key type %u",
1438 			    btrfs_item_size_nr(leaf, slot),
1439 			    expect_item_size, key->type);
1440 		return -EUCLEAN;
1441 	}
1442 	if (!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;
1447 	}
1448 	if (key->type != BTRFS_TREE_BLOCK_REF_KEY &&
1449 	    !IS_ALIGNED(key->offset, leaf->fs_info->sectorsize)) {
1450 		extent_err(leaf, slot,
1451 		"invalid tree parent bytenr, have %llu expect aligned to %u",
1452 			   key->offset, leaf->fs_info->sectorsize);
1453 		return -EUCLEAN;
1454 	}
1455 	return 0;
1456 }
1457 
check_extent_data_ref(struct extent_buffer * leaf,struct btrfs_key * key,int slot)1458 static int check_extent_data_ref(struct extent_buffer *leaf,
1459 				 struct btrfs_key *key, int slot)
1460 {
1461 	struct btrfs_extent_data_ref *dref;
1462 	unsigned long ptr = btrfs_item_ptr_offset(leaf, slot);
1463 	const unsigned long end = ptr + btrfs_item_size_nr(leaf, slot);
1464 
1465 	if (btrfs_item_size_nr(leaf, slot) % sizeof(*dref) != 0) {
1466 		generic_err(leaf, slot,
1467 	"invalid item size, have %u expect aligned to %zu for key type %u",
1468 			    btrfs_item_size_nr(leaf, slot),
1469 			    sizeof(*dref), key->type);
1470 		return -EUCLEAN;
1471 	}
1472 	if (!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize)) {
1473 		generic_err(leaf, slot,
1474 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1475 			    key->objectid, leaf->fs_info->sectorsize);
1476 		return -EUCLEAN;
1477 	}
1478 	for (; ptr < end; ptr += sizeof(*dref)) {
1479 		u64 offset;
1480 
1481 		/*
1482 		 * We cannot check the extent_data_ref hash due to possible
1483 		 * overflow from the leaf due to hash collisions.
1484 		 */
1485 		dref = (struct btrfs_extent_data_ref *)ptr;
1486 		offset = btrfs_extent_data_ref_offset(leaf, dref);
1487 		if (!IS_ALIGNED(offset, leaf->fs_info->sectorsize)) {
1488 			extent_err(leaf, slot,
1489 	"invalid extent data backref offset, have %llu expect aligned to %u",
1490 				   offset, leaf->fs_info->sectorsize);
1491 			return -EUCLEAN;
1492 		}
1493 	}
1494 	return 0;
1495 }
1496 
1497 #define inode_ref_err(eb, slot, fmt, args...)			\
1498 	inode_item_err(eb, slot, fmt, ##args)
check_inode_ref(struct extent_buffer * leaf,struct btrfs_key * key,struct btrfs_key * prev_key,int slot)1499 static int check_inode_ref(struct extent_buffer *leaf,
1500 			   struct btrfs_key *key, struct btrfs_key *prev_key,
1501 			   int slot)
1502 {
1503 	struct btrfs_inode_ref *iref;
1504 	unsigned long ptr;
1505 	unsigned long end;
1506 
1507 	if (!check_prev_ino(leaf, key, slot, prev_key))
1508 		return -EUCLEAN;
1509 	/* namelen can't be 0, so item_size == sizeof() is also invalid */
1510 	if (btrfs_item_size_nr(leaf, slot) <= sizeof(*iref)) {
1511 		inode_ref_err(leaf, slot,
1512 			"invalid item size, have %u expect (%zu, %u)",
1513 			btrfs_item_size_nr(leaf, slot),
1514 			sizeof(*iref), BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
1515 		return -EUCLEAN;
1516 	}
1517 
1518 	ptr = btrfs_item_ptr_offset(leaf, slot);
1519 	end = ptr + btrfs_item_size_nr(leaf, slot);
1520 	while (ptr < end) {
1521 		u16 namelen;
1522 
1523 		if (ptr + sizeof(iref) > end) {
1524 			inode_ref_err(leaf, slot,
1525 			"inode ref overflow, ptr %lu end %lu inode_ref_size %zu",
1526 				ptr, end, sizeof(iref));
1527 			return -EUCLEAN;
1528 		}
1529 
1530 		iref = (struct btrfs_inode_ref *)ptr;
1531 		namelen = btrfs_inode_ref_name_len(leaf, iref);
1532 		if (ptr + sizeof(*iref) + namelen > end) {
1533 			inode_ref_err(leaf, slot,
1534 				"inode ref overflow, ptr %lu end %lu namelen %u",
1535 				ptr, end, namelen);
1536 			return -EUCLEAN;
1537 		}
1538 
1539 		/*
1540 		 * NOTE: In theory we should record all found index numbers
1541 		 * to find any duplicated indexes, but that will be too time
1542 		 * consuming for inodes with too many hard links.
1543 		 */
1544 		ptr += sizeof(*iref) + namelen;
1545 	}
1546 	return 0;
1547 }
1548 
1549 /*
1550  * Common point to switch the item-specific validation.
1551  */
check_leaf_item(struct extent_buffer * leaf,struct btrfs_key * key,int slot,struct btrfs_key * prev_key)1552 static int check_leaf_item(struct extent_buffer *leaf,
1553 			   struct btrfs_key *key, int slot,
1554 			   struct btrfs_key *prev_key)
1555 {
1556 	int ret = 0;
1557 	struct btrfs_chunk *chunk;
1558 
1559 	switch (key->type) {
1560 	case BTRFS_EXTENT_DATA_KEY:
1561 		ret = check_extent_data_item(leaf, key, slot, prev_key);
1562 		break;
1563 	case BTRFS_EXTENT_CSUM_KEY:
1564 		ret = check_csum_item(leaf, key, slot, prev_key);
1565 		break;
1566 	case BTRFS_DIR_ITEM_KEY:
1567 	case BTRFS_DIR_INDEX_KEY:
1568 	case BTRFS_XATTR_ITEM_KEY:
1569 		ret = check_dir_item(leaf, key, prev_key, slot);
1570 		break;
1571 	case BTRFS_INODE_REF_KEY:
1572 		ret = check_inode_ref(leaf, key, prev_key, slot);
1573 		break;
1574 	case BTRFS_BLOCK_GROUP_ITEM_KEY:
1575 		ret = check_block_group_item(leaf, key, slot);
1576 		break;
1577 	case BTRFS_CHUNK_ITEM_KEY:
1578 		chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1579 		ret = check_leaf_chunk_item(leaf, chunk, key, slot);
1580 		break;
1581 	case BTRFS_DEV_ITEM_KEY:
1582 		ret = check_dev_item(leaf, key, slot);
1583 		break;
1584 	case BTRFS_INODE_ITEM_KEY:
1585 		ret = check_inode_item(leaf, key, slot);
1586 		break;
1587 	case BTRFS_ROOT_ITEM_KEY:
1588 		ret = check_root_item(leaf, key, slot);
1589 		break;
1590 	case BTRFS_EXTENT_ITEM_KEY:
1591 	case BTRFS_METADATA_ITEM_KEY:
1592 		ret = check_extent_item(leaf, key, slot, prev_key);
1593 		break;
1594 	case BTRFS_TREE_BLOCK_REF_KEY:
1595 	case BTRFS_SHARED_DATA_REF_KEY:
1596 	case BTRFS_SHARED_BLOCK_REF_KEY:
1597 		ret = check_simple_keyed_refs(leaf, key, slot);
1598 		break;
1599 	case BTRFS_EXTENT_DATA_REF_KEY:
1600 		ret = check_extent_data_ref(leaf, key, slot);
1601 		break;
1602 	}
1603 	return ret;
1604 }
1605 
check_leaf(struct extent_buffer * leaf,bool check_item_data)1606 static int check_leaf(struct extent_buffer *leaf, bool check_item_data)
1607 {
1608 	struct btrfs_fs_info *fs_info = leaf->fs_info;
1609 	/* No valid key type is 0, so all key should be larger than this key */
1610 	struct btrfs_key prev_key = {0, 0, 0};
1611 	struct btrfs_key key;
1612 	u32 nritems = btrfs_header_nritems(leaf);
1613 	int slot;
1614 
1615 	if (btrfs_header_level(leaf) != 0) {
1616 		generic_err(leaf, 0,
1617 			"invalid level for leaf, have %d expect 0",
1618 			btrfs_header_level(leaf));
1619 		return -EUCLEAN;
1620 	}
1621 
1622 	/*
1623 	 * Extent buffers from a relocation tree have a owner field that
1624 	 * corresponds to the subvolume tree they are based on. So just from an
1625 	 * extent buffer alone we can not find out what is the id of the
1626 	 * corresponding subvolume tree, so we can not figure out if the extent
1627 	 * buffer corresponds to the root of the relocation tree or not. So
1628 	 * skip this check for relocation trees.
1629 	 */
1630 	if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
1631 		u64 owner = btrfs_header_owner(leaf);
1632 
1633 		/* These trees must never be empty */
1634 		if (owner == BTRFS_ROOT_TREE_OBJECTID ||
1635 		    owner == BTRFS_CHUNK_TREE_OBJECTID ||
1636 		    owner == BTRFS_EXTENT_TREE_OBJECTID ||
1637 		    owner == BTRFS_DEV_TREE_OBJECTID ||
1638 		    owner == BTRFS_FS_TREE_OBJECTID ||
1639 		    owner == BTRFS_DATA_RELOC_TREE_OBJECTID) {
1640 			generic_err(leaf, 0,
1641 			"invalid root, root %llu must never be empty",
1642 				    owner);
1643 			return -EUCLEAN;
1644 		}
1645 		/* Unknown tree */
1646 		if (owner == 0) {
1647 			generic_err(leaf, 0,
1648 				"invalid owner, root 0 is not defined");
1649 			return -EUCLEAN;
1650 		}
1651 		return 0;
1652 	}
1653 
1654 	if (nritems == 0)
1655 		return 0;
1656 
1657 	/*
1658 	 * Check the following things to make sure this is a good leaf, and
1659 	 * leaf users won't need to bother with similar sanity checks:
1660 	 *
1661 	 * 1) key ordering
1662 	 * 2) item offset and size
1663 	 *    No overlap, no hole, all inside the leaf.
1664 	 * 3) item content
1665 	 *    If possible, do comprehensive sanity check.
1666 	 *    NOTE: All checks must only rely on the item data itself.
1667 	 */
1668 	for (slot = 0; slot < nritems; slot++) {
1669 		u32 item_end_expected;
1670 		int ret;
1671 
1672 		btrfs_item_key_to_cpu(leaf, &key, slot);
1673 
1674 		/* Make sure the keys are in the right order */
1675 		if (btrfs_comp_cpu_keys(&prev_key, &key) >= 0) {
1676 			generic_err(leaf, slot,
1677 	"bad key order, prev (%llu %u %llu) current (%llu %u %llu)",
1678 				prev_key.objectid, prev_key.type,
1679 				prev_key.offset, key.objectid, key.type,
1680 				key.offset);
1681 			return -EUCLEAN;
1682 		}
1683 
1684 		/*
1685 		 * Make sure the offset and ends are right, remember that the
1686 		 * item data starts at the end of the leaf and grows towards the
1687 		 * front.
1688 		 */
1689 		if (slot == 0)
1690 			item_end_expected = BTRFS_LEAF_DATA_SIZE(fs_info);
1691 		else
1692 			item_end_expected = btrfs_item_offset_nr(leaf,
1693 								 slot - 1);
1694 		if (btrfs_item_end_nr(leaf, slot) != item_end_expected) {
1695 			generic_err(leaf, slot,
1696 				"unexpected item end, have %u expect %u",
1697 				btrfs_item_end_nr(leaf, slot),
1698 				item_end_expected);
1699 			return -EUCLEAN;
1700 		}
1701 
1702 		/*
1703 		 * Check to make sure that we don't point outside of the leaf,
1704 		 * just in case all the items are consistent to each other, but
1705 		 * all point outside of the leaf.
1706 		 */
1707 		if (btrfs_item_end_nr(leaf, slot) >
1708 		    BTRFS_LEAF_DATA_SIZE(fs_info)) {
1709 			generic_err(leaf, slot,
1710 			"slot end outside of leaf, have %u expect range [0, %u]",
1711 				btrfs_item_end_nr(leaf, slot),
1712 				BTRFS_LEAF_DATA_SIZE(fs_info));
1713 			return -EUCLEAN;
1714 		}
1715 
1716 		/* Also check if the item pointer overlaps with btrfs item. */
1717 		if (btrfs_item_nr_offset(slot) + sizeof(struct btrfs_item) >
1718 		    btrfs_item_ptr_offset(leaf, slot)) {
1719 			generic_err(leaf, slot,
1720 		"slot overlaps with its data, item end %lu data start %lu",
1721 				btrfs_item_nr_offset(slot) +
1722 				sizeof(struct btrfs_item),
1723 				btrfs_item_ptr_offset(leaf, slot));
1724 			return -EUCLEAN;
1725 		}
1726 
1727 		if (check_item_data) {
1728 			/*
1729 			 * Check if the item size and content meet other
1730 			 * criteria
1731 			 */
1732 			ret = check_leaf_item(leaf, &key, slot, &prev_key);
1733 			if (ret < 0)
1734 				return ret;
1735 		}
1736 
1737 		prev_key.objectid = key.objectid;
1738 		prev_key.type = key.type;
1739 		prev_key.offset = key.offset;
1740 	}
1741 
1742 	return 0;
1743 }
1744 
btrfs_check_leaf_full(struct extent_buffer * leaf)1745 int btrfs_check_leaf_full(struct extent_buffer *leaf)
1746 {
1747 	return check_leaf(leaf, true);
1748 }
1749 ALLOW_ERROR_INJECTION(btrfs_check_leaf_full, ERRNO);
1750 
btrfs_check_leaf_relaxed(struct extent_buffer * leaf)1751 int btrfs_check_leaf_relaxed(struct extent_buffer *leaf)
1752 {
1753 	return check_leaf(leaf, false);
1754 }
1755 
btrfs_check_node(struct extent_buffer * node)1756 int btrfs_check_node(struct extent_buffer *node)
1757 {
1758 	struct btrfs_fs_info *fs_info = node->fs_info;
1759 	unsigned long nr = btrfs_header_nritems(node);
1760 	struct btrfs_key key, next_key;
1761 	int slot;
1762 	int level = btrfs_header_level(node);
1763 	u64 bytenr;
1764 	int ret = 0;
1765 
1766 	if (level <= 0 || level >= BTRFS_MAX_LEVEL) {
1767 		generic_err(node, 0,
1768 			"invalid level for node, have %d expect [1, %d]",
1769 			level, BTRFS_MAX_LEVEL - 1);
1770 		return -EUCLEAN;
1771 	}
1772 	if (nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(fs_info)) {
1773 		btrfs_crit(fs_info,
1774 "corrupt node: root=%llu block=%llu, nritems too %s, have %lu expect range [1,%u]",
1775 			   btrfs_header_owner(node), node->start,
1776 			   nr == 0 ? "small" : "large", nr,
1777 			   BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1778 		return -EUCLEAN;
1779 	}
1780 
1781 	for (slot = 0; slot < nr - 1; slot++) {
1782 		bytenr = btrfs_node_blockptr(node, slot);
1783 		btrfs_node_key_to_cpu(node, &key, slot);
1784 		btrfs_node_key_to_cpu(node, &next_key, slot + 1);
1785 
1786 		if (!bytenr) {
1787 			generic_err(node, slot,
1788 				"invalid NULL node pointer");
1789 			ret = -EUCLEAN;
1790 			goto out;
1791 		}
1792 		if (!IS_ALIGNED(bytenr, fs_info->sectorsize)) {
1793 			generic_err(node, slot,
1794 			"unaligned pointer, have %llu should be aligned to %u",
1795 				bytenr, fs_info->sectorsize);
1796 			ret = -EUCLEAN;
1797 			goto out;
1798 		}
1799 
1800 		if (btrfs_comp_cpu_keys(&key, &next_key) >= 0) {
1801 			generic_err(node, slot,
1802 	"bad key order, current (%llu %u %llu) next (%llu %u %llu)",
1803 				key.objectid, key.type, key.offset,
1804 				next_key.objectid, next_key.type,
1805 				next_key.offset);
1806 			ret = -EUCLEAN;
1807 			goto out;
1808 		}
1809 	}
1810 out:
1811 	return ret;
1812 }
1813 ALLOW_ERROR_INJECTION(btrfs_check_node, ERRNO);
1814