xref: /openbmc/linux/fs/btrfs/tree-checker.c (revision 2bc7d3e0)
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 #include "btrfs_inode.h"
28 
29 /*
30  * Error message should follow the following format:
31  * corrupt <type>: <identifier>, <reason>[, <bad_value>]
32  *
33  * @type:	leaf or node
34  * @identifier:	the necessary info to locate the leaf/node.
35  * 		It's recommended to decode key.objecitd/offset if it's
36  * 		meaningful.
37  * @reason:	describe the error
38  * @bad_value:	optional, it's recommended to output bad value and its
39  *		expected value (range).
40  *
41  * Since comma is used to separate the components, only space is allowed
42  * inside each component.
43  */
44 
45 /*
46  * Append generic "corrupt leaf/node root=%llu block=%llu slot=%d: " to @fmt.
47  * Allows callers to customize the output.
48  */
49 __printf(3, 4)
50 __cold
51 static void generic_err(const struct extent_buffer *eb, int slot,
52 			const char *fmt, ...)
53 {
54 	const struct btrfs_fs_info *fs_info = eb->fs_info;
55 	struct va_format vaf;
56 	va_list args;
57 
58 	va_start(args, fmt);
59 
60 	vaf.fmt = fmt;
61 	vaf.va = &args;
62 
63 	btrfs_crit(fs_info,
64 		"corrupt %s: root=%llu block=%llu slot=%d, %pV",
65 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
66 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot, &vaf);
67 	va_end(args);
68 }
69 
70 /*
71  * Customized reporter for extent data item, since its key objectid and
72  * offset has its own meaning.
73  */
74 __printf(3, 4)
75 __cold
76 static void file_extent_err(const struct extent_buffer *eb, int slot,
77 			    const char *fmt, ...)
78 {
79 	const struct btrfs_fs_info *fs_info = eb->fs_info;
80 	struct btrfs_key key;
81 	struct va_format vaf;
82 	va_list args;
83 
84 	btrfs_item_key_to_cpu(eb, &key, slot);
85 	va_start(args, fmt);
86 
87 	vaf.fmt = fmt;
88 	vaf.va = &args;
89 
90 	btrfs_crit(fs_info,
91 	"corrupt %s: root=%llu block=%llu slot=%d ino=%llu file_offset=%llu, %pV",
92 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
93 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
94 		key.objectid, key.offset, &vaf);
95 	va_end(args);
96 }
97 
98 /*
99  * Return 0 if the btrfs_file_extent_##name is aligned to @alignment
100  * Else return 1
101  */
102 #define CHECK_FE_ALIGNED(leaf, slot, fi, name, alignment)		      \
103 ({									      \
104 	if (unlikely(!IS_ALIGNED(btrfs_file_extent_##name((leaf), (fi)),      \
105 				 (alignment))))				      \
106 		file_extent_err((leaf), (slot),				      \
107 	"invalid %s for file extent, have %llu, should be aligned to %u",     \
108 			(#name), btrfs_file_extent_##name((leaf), (fi)),      \
109 			(alignment));					      \
110 	(!IS_ALIGNED(btrfs_file_extent_##name((leaf), (fi)), (alignment)));   \
111 })
112 
113 static u64 file_extent_end(struct extent_buffer *leaf,
114 			   struct btrfs_key *key,
115 			   struct btrfs_file_extent_item *extent)
116 {
117 	u64 end;
118 	u64 len;
119 
120 	if (btrfs_file_extent_type(leaf, extent) == BTRFS_FILE_EXTENT_INLINE) {
121 		len = btrfs_file_extent_ram_bytes(leaf, extent);
122 		end = ALIGN(key->offset + len, leaf->fs_info->sectorsize);
123 	} else {
124 		len = btrfs_file_extent_num_bytes(leaf, extent);
125 		end = key->offset + len;
126 	}
127 	return end;
128 }
129 
130 /*
131  * Customized report for dir_item, the only new important information is
132  * key->objectid, which represents inode number
133  */
134 __printf(3, 4)
135 __cold
136 static void dir_item_err(const struct extent_buffer *eb, int slot,
137 			 const char *fmt, ...)
138 {
139 	const struct btrfs_fs_info *fs_info = eb->fs_info;
140 	struct btrfs_key key;
141 	struct va_format vaf;
142 	va_list args;
143 
144 	btrfs_item_key_to_cpu(eb, &key, slot);
145 	va_start(args, fmt);
146 
147 	vaf.fmt = fmt;
148 	vaf.va = &args;
149 
150 	btrfs_crit(fs_info,
151 		"corrupt %s: root=%llu block=%llu slot=%d ino=%llu, %pV",
152 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
153 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
154 		key.objectid, &vaf);
155 	va_end(args);
156 }
157 
158 /*
159  * This functions checks prev_key->objectid, to ensure current key and prev_key
160  * share the same objectid as inode number.
161  *
162  * This is to detect missing INODE_ITEM in subvolume trees.
163  *
164  * Return true if everything is OK or we don't need to check.
165  * Return false if anything is wrong.
166  */
167 static bool check_prev_ino(struct extent_buffer *leaf,
168 			   struct btrfs_key *key, int slot,
169 			   struct btrfs_key *prev_key)
170 {
171 	/* No prev key, skip check */
172 	if (slot == 0)
173 		return true;
174 
175 	/* Only these key->types needs to be checked */
176 	ASSERT(key->type == BTRFS_XATTR_ITEM_KEY ||
177 	       key->type == BTRFS_INODE_REF_KEY ||
178 	       key->type == BTRFS_DIR_INDEX_KEY ||
179 	       key->type == BTRFS_DIR_ITEM_KEY ||
180 	       key->type == BTRFS_EXTENT_DATA_KEY);
181 
182 	/*
183 	 * Only subvolume trees along with their reloc trees need this check.
184 	 * Things like log tree doesn't follow this ino requirement.
185 	 */
186 	if (!is_fstree(btrfs_header_owner(leaf)))
187 		return true;
188 
189 	if (key->objectid == prev_key->objectid)
190 		return true;
191 
192 	/* Error found */
193 	dir_item_err(leaf, slot,
194 		"invalid previous key objectid, have %llu expect %llu",
195 		prev_key->objectid, key->objectid);
196 	return false;
197 }
198 static int check_extent_data_item(struct extent_buffer *leaf,
199 				  struct btrfs_key *key, int slot,
200 				  struct btrfs_key *prev_key)
201 {
202 	struct btrfs_fs_info *fs_info = leaf->fs_info;
203 	struct btrfs_file_extent_item *fi;
204 	u32 sectorsize = fs_info->sectorsize;
205 	u32 item_size = btrfs_item_size(leaf, slot);
206 	u64 extent_end;
207 
208 	if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) {
209 		file_extent_err(leaf, slot,
210 "unaligned file_offset for file extent, have %llu should be aligned to %u",
211 			key->offset, sectorsize);
212 		return -EUCLEAN;
213 	}
214 
215 	/*
216 	 * Previous key must have the same key->objectid (ino).
217 	 * It can be XATTR_ITEM, INODE_ITEM or just another EXTENT_DATA.
218 	 * But if objectids mismatch, it means we have a missing
219 	 * INODE_ITEM.
220 	 */
221 	if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
222 		return -EUCLEAN;
223 
224 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
225 
226 	/*
227 	 * Make sure the item contains at least inline header, so the file
228 	 * extent type is not some garbage.
229 	 */
230 	if (unlikely(item_size < BTRFS_FILE_EXTENT_INLINE_DATA_START)) {
231 		file_extent_err(leaf, slot,
232 				"invalid item size, have %u expect [%zu, %u)",
233 				item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START,
234 				SZ_4K);
235 		return -EUCLEAN;
236 	}
237 	if (unlikely(btrfs_file_extent_type(leaf, fi) >=
238 		     BTRFS_NR_FILE_EXTENT_TYPES)) {
239 		file_extent_err(leaf, slot,
240 		"invalid type for file extent, have %u expect range [0, %u]",
241 			btrfs_file_extent_type(leaf, fi),
242 			BTRFS_NR_FILE_EXTENT_TYPES - 1);
243 		return -EUCLEAN;
244 	}
245 
246 	/*
247 	 * Support for new compression/encryption must introduce incompat flag,
248 	 * and must be caught in open_ctree().
249 	 */
250 	if (unlikely(btrfs_file_extent_compression(leaf, fi) >=
251 		     BTRFS_NR_COMPRESS_TYPES)) {
252 		file_extent_err(leaf, slot,
253 	"invalid compression for file extent, have %u expect range [0, %u]",
254 			btrfs_file_extent_compression(leaf, fi),
255 			BTRFS_NR_COMPRESS_TYPES - 1);
256 		return -EUCLEAN;
257 	}
258 	if (unlikely(btrfs_file_extent_encryption(leaf, fi))) {
259 		file_extent_err(leaf, slot,
260 			"invalid encryption for file extent, have %u expect 0",
261 			btrfs_file_extent_encryption(leaf, fi));
262 		return -EUCLEAN;
263 	}
264 	if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
265 		/* Inline extent must have 0 as key offset */
266 		if (unlikely(key->offset)) {
267 			file_extent_err(leaf, slot,
268 		"invalid file_offset for inline file extent, have %llu expect 0",
269 				key->offset);
270 			return -EUCLEAN;
271 		}
272 
273 		/* Compressed inline extent has no on-disk size, skip it */
274 		if (btrfs_file_extent_compression(leaf, fi) !=
275 		    BTRFS_COMPRESS_NONE)
276 			return 0;
277 
278 		/* Uncompressed inline extent size must match item size */
279 		if (unlikely(item_size != BTRFS_FILE_EXTENT_INLINE_DATA_START +
280 					  btrfs_file_extent_ram_bytes(leaf, fi))) {
281 			file_extent_err(leaf, slot,
282 	"invalid ram_bytes for uncompressed inline extent, have %u expect %llu",
283 				item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START +
284 				btrfs_file_extent_ram_bytes(leaf, fi));
285 			return -EUCLEAN;
286 		}
287 		return 0;
288 	}
289 
290 	/* Regular or preallocated extent has fixed item size */
291 	if (unlikely(item_size != sizeof(*fi))) {
292 		file_extent_err(leaf, slot,
293 	"invalid item size for reg/prealloc file extent, have %u expect %zu",
294 			item_size, sizeof(*fi));
295 		return -EUCLEAN;
296 	}
297 	if (unlikely(CHECK_FE_ALIGNED(leaf, slot, fi, ram_bytes, sectorsize) ||
298 		     CHECK_FE_ALIGNED(leaf, slot, fi, disk_bytenr, sectorsize) ||
299 		     CHECK_FE_ALIGNED(leaf, slot, fi, disk_num_bytes, sectorsize) ||
300 		     CHECK_FE_ALIGNED(leaf, slot, fi, offset, sectorsize) ||
301 		     CHECK_FE_ALIGNED(leaf, slot, fi, num_bytes, sectorsize)))
302 		return -EUCLEAN;
303 
304 	/* Catch extent end overflow */
305 	if (unlikely(check_add_overflow(btrfs_file_extent_num_bytes(leaf, fi),
306 					key->offset, &extent_end))) {
307 		file_extent_err(leaf, slot,
308 	"extent end overflow, have file offset %llu extent num bytes %llu",
309 				key->offset,
310 				btrfs_file_extent_num_bytes(leaf, fi));
311 		return -EUCLEAN;
312 	}
313 
314 	/*
315 	 * Check that no two consecutive file extent items, in the same leaf,
316 	 * present ranges that overlap each other.
317 	 */
318 	if (slot > 0 &&
319 	    prev_key->objectid == key->objectid &&
320 	    prev_key->type == BTRFS_EXTENT_DATA_KEY) {
321 		struct btrfs_file_extent_item *prev_fi;
322 		u64 prev_end;
323 
324 		prev_fi = btrfs_item_ptr(leaf, slot - 1,
325 					 struct btrfs_file_extent_item);
326 		prev_end = file_extent_end(leaf, prev_key, prev_fi);
327 		if (unlikely(prev_end > key->offset)) {
328 			file_extent_err(leaf, slot - 1,
329 "file extent end range (%llu) goes beyond start offset (%llu) of the next file extent",
330 					prev_end, key->offset);
331 			return -EUCLEAN;
332 		}
333 	}
334 
335 	return 0;
336 }
337 
338 static int check_csum_item(struct extent_buffer *leaf, struct btrfs_key *key,
339 			   int slot, struct btrfs_key *prev_key)
340 {
341 	struct btrfs_fs_info *fs_info = leaf->fs_info;
342 	u32 sectorsize = fs_info->sectorsize;
343 	const u32 csumsize = fs_info->csum_size;
344 
345 	if (unlikely(key->objectid != BTRFS_EXTENT_CSUM_OBJECTID)) {
346 		generic_err(leaf, slot,
347 		"invalid key objectid for csum item, have %llu expect %llu",
348 			key->objectid, BTRFS_EXTENT_CSUM_OBJECTID);
349 		return -EUCLEAN;
350 	}
351 	if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) {
352 		generic_err(leaf, slot,
353 	"unaligned key offset for csum item, have %llu should be aligned to %u",
354 			key->offset, sectorsize);
355 		return -EUCLEAN;
356 	}
357 	if (unlikely(!IS_ALIGNED(btrfs_item_size(leaf, slot), csumsize))) {
358 		generic_err(leaf, slot,
359 	"unaligned item size for csum item, have %u should be aligned to %u",
360 			btrfs_item_size(leaf, slot), csumsize);
361 		return -EUCLEAN;
362 	}
363 	if (slot > 0 && prev_key->type == BTRFS_EXTENT_CSUM_KEY) {
364 		u64 prev_csum_end;
365 		u32 prev_item_size;
366 
367 		prev_item_size = btrfs_item_size(leaf, slot - 1);
368 		prev_csum_end = (prev_item_size / csumsize) * sectorsize;
369 		prev_csum_end += prev_key->offset;
370 		if (unlikely(prev_csum_end > key->offset)) {
371 			generic_err(leaf, slot - 1,
372 "csum end range (%llu) goes beyond the start range (%llu) of the next csum item",
373 				    prev_csum_end, key->offset);
374 			return -EUCLEAN;
375 		}
376 	}
377 	return 0;
378 }
379 
380 /* Inode item error output has the same format as dir_item_err() */
381 #define inode_item_err(eb, slot, fmt, ...)			\
382 	dir_item_err(eb, slot, fmt, __VA_ARGS__)
383 
384 static int check_inode_key(struct extent_buffer *leaf, struct btrfs_key *key,
385 			   int slot)
386 {
387 	struct btrfs_key item_key;
388 	bool is_inode_item;
389 
390 	btrfs_item_key_to_cpu(leaf, &item_key, slot);
391 	is_inode_item = (item_key.type == BTRFS_INODE_ITEM_KEY);
392 
393 	/* For XATTR_ITEM, location key should be all 0 */
394 	if (item_key.type == BTRFS_XATTR_ITEM_KEY) {
395 		if (unlikely(key->objectid != 0 || key->type != 0 ||
396 			     key->offset != 0))
397 			return -EUCLEAN;
398 		return 0;
399 	}
400 
401 	if (unlikely((key->objectid < BTRFS_FIRST_FREE_OBJECTID ||
402 		      key->objectid > BTRFS_LAST_FREE_OBJECTID) &&
403 		     key->objectid != BTRFS_ROOT_TREE_DIR_OBJECTID &&
404 		     key->objectid != BTRFS_FREE_INO_OBJECTID)) {
405 		if (is_inode_item) {
406 			generic_err(leaf, slot,
407 	"invalid key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
408 				key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
409 				BTRFS_FIRST_FREE_OBJECTID,
410 				BTRFS_LAST_FREE_OBJECTID,
411 				BTRFS_FREE_INO_OBJECTID);
412 		} else {
413 			dir_item_err(leaf, slot,
414 "invalid location key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
415 				key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
416 				BTRFS_FIRST_FREE_OBJECTID,
417 				BTRFS_LAST_FREE_OBJECTID,
418 				BTRFS_FREE_INO_OBJECTID);
419 		}
420 		return -EUCLEAN;
421 	}
422 	if (unlikely(key->offset != 0)) {
423 		if (is_inode_item)
424 			inode_item_err(leaf, slot,
425 				       "invalid key offset: has %llu expect 0",
426 				       key->offset);
427 		else
428 			dir_item_err(leaf, slot,
429 				"invalid location key offset:has %llu expect 0",
430 				key->offset);
431 		return -EUCLEAN;
432 	}
433 	return 0;
434 }
435 
436 static int check_root_key(struct extent_buffer *leaf, struct btrfs_key *key,
437 			  int slot)
438 {
439 	struct btrfs_key item_key;
440 	bool is_root_item;
441 
442 	btrfs_item_key_to_cpu(leaf, &item_key, slot);
443 	is_root_item = (item_key.type == BTRFS_ROOT_ITEM_KEY);
444 
445 	/* No such tree id */
446 	if (unlikely(key->objectid == 0)) {
447 		if (is_root_item)
448 			generic_err(leaf, slot, "invalid root id 0");
449 		else
450 			dir_item_err(leaf, slot,
451 				     "invalid location key root id 0");
452 		return -EUCLEAN;
453 	}
454 
455 	/* DIR_ITEM/INDEX/INODE_REF is not allowed to point to non-fs trees */
456 	if (unlikely(!is_fstree(key->objectid) && !is_root_item)) {
457 		dir_item_err(leaf, slot,
458 		"invalid location key objectid, have %llu expect [%llu, %llu]",
459 				key->objectid, BTRFS_FIRST_FREE_OBJECTID,
460 				BTRFS_LAST_FREE_OBJECTID);
461 		return -EUCLEAN;
462 	}
463 
464 	/*
465 	 * ROOT_ITEM with non-zero offset means this is a snapshot, created at
466 	 * @offset transid.
467 	 * Furthermore, for location key in DIR_ITEM, its offset is always -1.
468 	 *
469 	 * So here we only check offset for reloc tree whose key->offset must
470 	 * be a valid tree.
471 	 */
472 	if (unlikely(key->objectid == BTRFS_TREE_RELOC_OBJECTID &&
473 		     key->offset == 0)) {
474 		generic_err(leaf, slot, "invalid root id 0 for reloc tree");
475 		return -EUCLEAN;
476 	}
477 	return 0;
478 }
479 
480 static int check_dir_item(struct extent_buffer *leaf,
481 			  struct btrfs_key *key, struct btrfs_key *prev_key,
482 			  int slot)
483 {
484 	struct btrfs_fs_info *fs_info = leaf->fs_info;
485 	struct btrfs_dir_item *di;
486 	u32 item_size = btrfs_item_size(leaf, slot);
487 	u32 cur = 0;
488 
489 	if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
490 		return -EUCLEAN;
491 
492 	di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
493 	while (cur < item_size) {
494 		struct btrfs_key location_key;
495 		u32 name_len;
496 		u32 data_len;
497 		u32 max_name_len;
498 		u32 total_size;
499 		u32 name_hash;
500 		u8 dir_type;
501 		int ret;
502 
503 		/* header itself should not cross item boundary */
504 		if (unlikely(cur + sizeof(*di) > item_size)) {
505 			dir_item_err(leaf, slot,
506 		"dir item header crosses item boundary, have %zu boundary %u",
507 				cur + sizeof(*di), item_size);
508 			return -EUCLEAN;
509 		}
510 
511 		/* Location key check */
512 		btrfs_dir_item_key_to_cpu(leaf, di, &location_key);
513 		if (location_key.type == BTRFS_ROOT_ITEM_KEY) {
514 			ret = check_root_key(leaf, &location_key, slot);
515 			if (unlikely(ret < 0))
516 				return ret;
517 		} else if (location_key.type == BTRFS_INODE_ITEM_KEY ||
518 			   location_key.type == 0) {
519 			ret = check_inode_key(leaf, &location_key, slot);
520 			if (unlikely(ret < 0))
521 				return ret;
522 		} else {
523 			dir_item_err(leaf, slot,
524 			"invalid location key type, have %u, expect %u or %u",
525 				     location_key.type, BTRFS_ROOT_ITEM_KEY,
526 				     BTRFS_INODE_ITEM_KEY);
527 			return -EUCLEAN;
528 		}
529 
530 		/* dir type check */
531 		dir_type = btrfs_dir_type(leaf, di);
532 		if (unlikely(dir_type >= BTRFS_FT_MAX)) {
533 			dir_item_err(leaf, slot,
534 			"invalid dir item type, have %u expect [0, %u)",
535 				dir_type, BTRFS_FT_MAX);
536 			return -EUCLEAN;
537 		}
538 
539 		if (unlikely(key->type == BTRFS_XATTR_ITEM_KEY &&
540 			     dir_type != BTRFS_FT_XATTR)) {
541 			dir_item_err(leaf, slot,
542 		"invalid dir item type for XATTR key, have %u expect %u",
543 				dir_type, BTRFS_FT_XATTR);
544 			return -EUCLEAN;
545 		}
546 		if (unlikely(dir_type == BTRFS_FT_XATTR &&
547 			     key->type != BTRFS_XATTR_ITEM_KEY)) {
548 			dir_item_err(leaf, slot,
549 			"xattr dir type found for non-XATTR key");
550 			return -EUCLEAN;
551 		}
552 		if (dir_type == BTRFS_FT_XATTR)
553 			max_name_len = XATTR_NAME_MAX;
554 		else
555 			max_name_len = BTRFS_NAME_LEN;
556 
557 		/* Name/data length check */
558 		name_len = btrfs_dir_name_len(leaf, di);
559 		data_len = btrfs_dir_data_len(leaf, di);
560 		if (unlikely(name_len > max_name_len)) {
561 			dir_item_err(leaf, slot,
562 			"dir item name len too long, have %u max %u",
563 				name_len, max_name_len);
564 			return -EUCLEAN;
565 		}
566 		if (unlikely(name_len + data_len > BTRFS_MAX_XATTR_SIZE(fs_info))) {
567 			dir_item_err(leaf, slot,
568 			"dir item name and data len too long, have %u max %u",
569 				name_len + data_len,
570 				BTRFS_MAX_XATTR_SIZE(fs_info));
571 			return -EUCLEAN;
572 		}
573 
574 		if (unlikely(data_len && dir_type != BTRFS_FT_XATTR)) {
575 			dir_item_err(leaf, slot,
576 			"dir item with invalid data len, have %u expect 0",
577 				data_len);
578 			return -EUCLEAN;
579 		}
580 
581 		total_size = sizeof(*di) + name_len + data_len;
582 
583 		/* header and name/data should not cross item boundary */
584 		if (unlikely(cur + total_size > item_size)) {
585 			dir_item_err(leaf, slot,
586 		"dir item data crosses item boundary, have %u boundary %u",
587 				cur + total_size, item_size);
588 			return -EUCLEAN;
589 		}
590 
591 		/*
592 		 * Special check for XATTR/DIR_ITEM, as key->offset is name
593 		 * hash, should match its name
594 		 */
595 		if (key->type == BTRFS_DIR_ITEM_KEY ||
596 		    key->type == BTRFS_XATTR_ITEM_KEY) {
597 			char namebuf[max(BTRFS_NAME_LEN, XATTR_NAME_MAX)];
598 
599 			read_extent_buffer(leaf, namebuf,
600 					(unsigned long)(di + 1), name_len);
601 			name_hash = btrfs_name_hash(namebuf, name_len);
602 			if (unlikely(key->offset != name_hash)) {
603 				dir_item_err(leaf, slot,
604 		"name hash mismatch with key, have 0x%016x expect 0x%016llx",
605 					name_hash, key->offset);
606 				return -EUCLEAN;
607 			}
608 		}
609 		cur += total_size;
610 		di = (struct btrfs_dir_item *)((void *)di + total_size);
611 	}
612 	return 0;
613 }
614 
615 __printf(3, 4)
616 __cold
617 static void block_group_err(const struct extent_buffer *eb, int slot,
618 			    const char *fmt, ...)
619 {
620 	const struct btrfs_fs_info *fs_info = eb->fs_info;
621 	struct btrfs_key key;
622 	struct va_format vaf;
623 	va_list args;
624 
625 	btrfs_item_key_to_cpu(eb, &key, slot);
626 	va_start(args, fmt);
627 
628 	vaf.fmt = fmt;
629 	vaf.va = &args;
630 
631 	btrfs_crit(fs_info,
632 	"corrupt %s: root=%llu block=%llu slot=%d bg_start=%llu bg_len=%llu, %pV",
633 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
634 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
635 		key.objectid, key.offset, &vaf);
636 	va_end(args);
637 }
638 
639 static int check_block_group_item(struct extent_buffer *leaf,
640 				  struct btrfs_key *key, int slot)
641 {
642 	struct btrfs_fs_info *fs_info = leaf->fs_info;
643 	struct btrfs_block_group_item bgi;
644 	u32 item_size = btrfs_item_size(leaf, slot);
645 	u64 chunk_objectid;
646 	u64 flags;
647 	u64 type;
648 
649 	/*
650 	 * Here we don't really care about alignment since extent allocator can
651 	 * handle it.  We care more about the size.
652 	 */
653 	if (unlikely(key->offset == 0)) {
654 		block_group_err(leaf, slot,
655 				"invalid block group size 0");
656 		return -EUCLEAN;
657 	}
658 
659 	if (unlikely(item_size != sizeof(bgi))) {
660 		block_group_err(leaf, slot,
661 			"invalid item size, have %u expect %zu",
662 				item_size, sizeof(bgi));
663 		return -EUCLEAN;
664 	}
665 
666 	read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot),
667 			   sizeof(bgi));
668 	chunk_objectid = btrfs_stack_block_group_chunk_objectid(&bgi);
669 	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
670 		/*
671 		 * We don't init the nr_global_roots until we load the global
672 		 * roots, so this could be 0 at mount time.  If it's 0 we'll
673 		 * just assume we're fine, and later we'll check against our
674 		 * actual value.
675 		 */
676 		if (unlikely(fs_info->nr_global_roots &&
677 			     chunk_objectid >= fs_info->nr_global_roots)) {
678 			block_group_err(leaf, slot,
679 	"invalid block group global root id, have %llu, needs to be <= %llu",
680 					chunk_objectid,
681 					fs_info->nr_global_roots);
682 			return -EUCLEAN;
683 		}
684 	} else if (unlikely(chunk_objectid != BTRFS_FIRST_CHUNK_TREE_OBJECTID)) {
685 		block_group_err(leaf, slot,
686 		"invalid block group chunk objectid, have %llu expect %llu",
687 				btrfs_stack_block_group_chunk_objectid(&bgi),
688 				BTRFS_FIRST_CHUNK_TREE_OBJECTID);
689 		return -EUCLEAN;
690 	}
691 
692 	if (unlikely(btrfs_stack_block_group_used(&bgi) > key->offset)) {
693 		block_group_err(leaf, slot,
694 			"invalid block group used, have %llu expect [0, %llu)",
695 				btrfs_stack_block_group_used(&bgi), key->offset);
696 		return -EUCLEAN;
697 	}
698 
699 	flags = btrfs_stack_block_group_flags(&bgi);
700 	if (unlikely(hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) > 1)) {
701 		block_group_err(leaf, slot,
702 "invalid profile flags, have 0x%llx (%lu bits set) expect no more than 1 bit set",
703 			flags & BTRFS_BLOCK_GROUP_PROFILE_MASK,
704 			hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK));
705 		return -EUCLEAN;
706 	}
707 
708 	type = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
709 	if (unlikely(type != BTRFS_BLOCK_GROUP_DATA &&
710 		     type != BTRFS_BLOCK_GROUP_METADATA &&
711 		     type != BTRFS_BLOCK_GROUP_SYSTEM &&
712 		     type != (BTRFS_BLOCK_GROUP_METADATA |
713 			      BTRFS_BLOCK_GROUP_DATA))) {
714 		block_group_err(leaf, slot,
715 "invalid type, have 0x%llx (%lu bits set) expect either 0x%llx, 0x%llx, 0x%llx or 0x%llx",
716 			type, hweight64(type),
717 			BTRFS_BLOCK_GROUP_DATA, BTRFS_BLOCK_GROUP_METADATA,
718 			BTRFS_BLOCK_GROUP_SYSTEM,
719 			BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA);
720 		return -EUCLEAN;
721 	}
722 	return 0;
723 }
724 
725 __printf(4, 5)
726 __cold
727 static void chunk_err(const struct extent_buffer *leaf,
728 		      const struct btrfs_chunk *chunk, u64 logical,
729 		      const char *fmt, ...)
730 {
731 	const struct btrfs_fs_info *fs_info = leaf->fs_info;
732 	bool is_sb;
733 	struct va_format vaf;
734 	va_list args;
735 	int i;
736 	int slot = -1;
737 
738 	/* Only superblock eb is able to have such small offset */
739 	is_sb = (leaf->start == BTRFS_SUPER_INFO_OFFSET);
740 
741 	if (!is_sb) {
742 		/*
743 		 * Get the slot number by iterating through all slots, this
744 		 * would provide better readability.
745 		 */
746 		for (i = 0; i < btrfs_header_nritems(leaf); i++) {
747 			if (btrfs_item_ptr_offset(leaf, i) ==
748 					(unsigned long)chunk) {
749 				slot = i;
750 				break;
751 			}
752 		}
753 	}
754 	va_start(args, fmt);
755 	vaf.fmt = fmt;
756 	vaf.va = &args;
757 
758 	if (is_sb)
759 		btrfs_crit(fs_info,
760 		"corrupt superblock syschunk array: chunk_start=%llu, %pV",
761 			   logical, &vaf);
762 	else
763 		btrfs_crit(fs_info,
764 	"corrupt leaf: root=%llu block=%llu slot=%d chunk_start=%llu, %pV",
765 			   BTRFS_CHUNK_TREE_OBJECTID, leaf->start, slot,
766 			   logical, &vaf);
767 	va_end(args);
768 }
769 
770 /*
771  * The common chunk check which could also work on super block sys chunk array.
772  *
773  * Return -EUCLEAN if anything is corrupted.
774  * Return 0 if everything is OK.
775  */
776 int btrfs_check_chunk_valid(struct extent_buffer *leaf,
777 			    struct btrfs_chunk *chunk, u64 logical)
778 {
779 	struct btrfs_fs_info *fs_info = leaf->fs_info;
780 	u64 length;
781 	u64 chunk_end;
782 	u64 stripe_len;
783 	u16 num_stripes;
784 	u16 sub_stripes;
785 	u64 type;
786 	u64 features;
787 	bool mixed = false;
788 	int raid_index;
789 	int nparity;
790 	int ncopies;
791 
792 	length = btrfs_chunk_length(leaf, chunk);
793 	stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
794 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
795 	sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
796 	type = btrfs_chunk_type(leaf, chunk);
797 	raid_index = btrfs_bg_flags_to_raid_index(type);
798 	ncopies = btrfs_raid_array[raid_index].ncopies;
799 	nparity = btrfs_raid_array[raid_index].nparity;
800 
801 	if (unlikely(!num_stripes)) {
802 		chunk_err(leaf, chunk, logical,
803 			  "invalid chunk num_stripes, have %u", num_stripes);
804 		return -EUCLEAN;
805 	}
806 	if (unlikely(num_stripes < ncopies)) {
807 		chunk_err(leaf, chunk, logical,
808 			  "invalid chunk num_stripes < ncopies, have %u < %d",
809 			  num_stripes, ncopies);
810 		return -EUCLEAN;
811 	}
812 	if (unlikely(nparity && num_stripes == nparity)) {
813 		chunk_err(leaf, chunk, logical,
814 			  "invalid chunk num_stripes == nparity, have %u == %d",
815 			  num_stripes, nparity);
816 		return -EUCLEAN;
817 	}
818 	if (unlikely(!IS_ALIGNED(logical, fs_info->sectorsize))) {
819 		chunk_err(leaf, chunk, logical,
820 		"invalid chunk logical, have %llu should aligned to %u",
821 			  logical, fs_info->sectorsize);
822 		return -EUCLEAN;
823 	}
824 	if (unlikely(btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize)) {
825 		chunk_err(leaf, chunk, logical,
826 			  "invalid chunk sectorsize, have %u expect %u",
827 			  btrfs_chunk_sector_size(leaf, chunk),
828 			  fs_info->sectorsize);
829 		return -EUCLEAN;
830 	}
831 	if (unlikely(!length || !IS_ALIGNED(length, fs_info->sectorsize))) {
832 		chunk_err(leaf, chunk, logical,
833 			  "invalid chunk length, have %llu", length);
834 		return -EUCLEAN;
835 	}
836 	if (unlikely(check_add_overflow(logical, length, &chunk_end))) {
837 		chunk_err(leaf, chunk, logical,
838 "invalid chunk logical start and length, have logical start %llu length %llu",
839 			  logical, length);
840 		return -EUCLEAN;
841 	}
842 	if (unlikely(!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN)) {
843 		chunk_err(leaf, chunk, logical,
844 			  "invalid chunk stripe length: %llu",
845 			  stripe_len);
846 		return -EUCLEAN;
847 	}
848 	if (unlikely(type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
849 			      BTRFS_BLOCK_GROUP_PROFILE_MASK))) {
850 		chunk_err(leaf, chunk, logical,
851 			  "unrecognized chunk type: 0x%llx",
852 			  ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
853 			    BTRFS_BLOCK_GROUP_PROFILE_MASK) &
854 			  btrfs_chunk_type(leaf, chunk));
855 		return -EUCLEAN;
856 	}
857 
858 	if (unlikely(!has_single_bit_set(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
859 		     (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) != 0)) {
860 		chunk_err(leaf, chunk, logical,
861 		"invalid chunk profile flag: 0x%llx, expect 0 or 1 bit set",
862 			  type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
863 		return -EUCLEAN;
864 	}
865 	if (unlikely((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0)) {
866 		chunk_err(leaf, chunk, logical,
867 	"missing chunk type flag, have 0x%llx one bit must be set in 0x%llx",
868 			  type, BTRFS_BLOCK_GROUP_TYPE_MASK);
869 		return -EUCLEAN;
870 	}
871 
872 	if (unlikely((type & BTRFS_BLOCK_GROUP_SYSTEM) &&
873 		     (type & (BTRFS_BLOCK_GROUP_METADATA |
874 			      BTRFS_BLOCK_GROUP_DATA)))) {
875 		chunk_err(leaf, chunk, logical,
876 			  "system chunk with data or metadata type: 0x%llx",
877 			  type);
878 		return -EUCLEAN;
879 	}
880 
881 	features = btrfs_super_incompat_flags(fs_info->super_copy);
882 	if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
883 		mixed = true;
884 
885 	if (!mixed) {
886 		if (unlikely((type & BTRFS_BLOCK_GROUP_METADATA) &&
887 			     (type & BTRFS_BLOCK_GROUP_DATA))) {
888 			chunk_err(leaf, chunk, logical,
889 			"mixed chunk type in non-mixed mode: 0x%llx", type);
890 			return -EUCLEAN;
891 		}
892 	}
893 
894 	if (unlikely((type & BTRFS_BLOCK_GROUP_RAID10 &&
895 		      sub_stripes != btrfs_raid_array[BTRFS_RAID_RAID10].sub_stripes) ||
896 		     (type & BTRFS_BLOCK_GROUP_RAID1 &&
897 		      num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1].devs_min) ||
898 		     (type & BTRFS_BLOCK_GROUP_RAID1C3 &&
899 		      num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1C3].devs_min) ||
900 		     (type & BTRFS_BLOCK_GROUP_RAID1C4 &&
901 		      num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1C4].devs_min) ||
902 		     (type & BTRFS_BLOCK_GROUP_RAID5 &&
903 		      num_stripes < btrfs_raid_array[BTRFS_RAID_RAID5].devs_min) ||
904 		     (type & BTRFS_BLOCK_GROUP_RAID6 &&
905 		      num_stripes < btrfs_raid_array[BTRFS_RAID_RAID6].devs_min) ||
906 		     (type & BTRFS_BLOCK_GROUP_DUP &&
907 		      num_stripes != btrfs_raid_array[BTRFS_RAID_DUP].dev_stripes) ||
908 		     ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
909 		      num_stripes != btrfs_raid_array[BTRFS_RAID_SINGLE].dev_stripes))) {
910 		chunk_err(leaf, chunk, logical,
911 			"invalid num_stripes:sub_stripes %u:%u for profile %llu",
912 			num_stripes, sub_stripes,
913 			type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
914 		return -EUCLEAN;
915 	}
916 
917 	return 0;
918 }
919 
920 /*
921  * Enhanced version of chunk item checker.
922  *
923  * The common btrfs_check_chunk_valid() doesn't check item size since it needs
924  * to work on super block sys_chunk_array which doesn't have full item ptr.
925  */
926 static int check_leaf_chunk_item(struct extent_buffer *leaf,
927 				 struct btrfs_chunk *chunk,
928 				 struct btrfs_key *key, int slot)
929 {
930 	int num_stripes;
931 
932 	if (unlikely(btrfs_item_size(leaf, slot) < sizeof(struct btrfs_chunk))) {
933 		chunk_err(leaf, chunk, key->offset,
934 			"invalid chunk item size: have %u expect [%zu, %u)",
935 			btrfs_item_size(leaf, slot),
936 			sizeof(struct btrfs_chunk),
937 			BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
938 		return -EUCLEAN;
939 	}
940 
941 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
942 	/* Let btrfs_check_chunk_valid() handle this error type */
943 	if (num_stripes == 0)
944 		goto out;
945 
946 	if (unlikely(btrfs_chunk_item_size(num_stripes) !=
947 		     btrfs_item_size(leaf, slot))) {
948 		chunk_err(leaf, chunk, key->offset,
949 			"invalid chunk item size: have %u expect %lu",
950 			btrfs_item_size(leaf, slot),
951 			btrfs_chunk_item_size(num_stripes));
952 		return -EUCLEAN;
953 	}
954 out:
955 	return btrfs_check_chunk_valid(leaf, chunk, key->offset);
956 }
957 
958 __printf(3, 4)
959 __cold
960 static void dev_item_err(const struct extent_buffer *eb, int slot,
961 			 const char *fmt, ...)
962 {
963 	struct btrfs_key key;
964 	struct va_format vaf;
965 	va_list args;
966 
967 	btrfs_item_key_to_cpu(eb, &key, slot);
968 	va_start(args, fmt);
969 
970 	vaf.fmt = fmt;
971 	vaf.va = &args;
972 
973 	btrfs_crit(eb->fs_info,
974 	"corrupt %s: root=%llu block=%llu slot=%d devid=%llu %pV",
975 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
976 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
977 		key.objectid, &vaf);
978 	va_end(args);
979 }
980 
981 static int check_dev_item(struct extent_buffer *leaf,
982 			  struct btrfs_key *key, int slot)
983 {
984 	struct btrfs_dev_item *ditem;
985 	const u32 item_size = btrfs_item_size(leaf, slot);
986 
987 	if (unlikely(key->objectid != BTRFS_DEV_ITEMS_OBJECTID)) {
988 		dev_item_err(leaf, slot,
989 			     "invalid objectid: has=%llu expect=%llu",
990 			     key->objectid, BTRFS_DEV_ITEMS_OBJECTID);
991 		return -EUCLEAN;
992 	}
993 
994 	if (unlikely(item_size != sizeof(*ditem))) {
995 		dev_item_err(leaf, slot, "invalid item size: has %u expect %zu",
996 			     item_size, sizeof(*ditem));
997 		return -EUCLEAN;
998 	}
999 
1000 	ditem = btrfs_item_ptr(leaf, slot, struct btrfs_dev_item);
1001 	if (unlikely(btrfs_device_id(leaf, ditem) != key->offset)) {
1002 		dev_item_err(leaf, slot,
1003 			     "devid mismatch: key has=%llu item has=%llu",
1004 			     key->offset, btrfs_device_id(leaf, ditem));
1005 		return -EUCLEAN;
1006 	}
1007 
1008 	/*
1009 	 * For device total_bytes, we don't have reliable way to check it, as
1010 	 * it can be 0 for device removal. Device size check can only be done
1011 	 * by dev extents check.
1012 	 */
1013 	if (unlikely(btrfs_device_bytes_used(leaf, ditem) >
1014 		     btrfs_device_total_bytes(leaf, ditem))) {
1015 		dev_item_err(leaf, slot,
1016 			     "invalid bytes used: have %llu expect [0, %llu]",
1017 			     btrfs_device_bytes_used(leaf, ditem),
1018 			     btrfs_device_total_bytes(leaf, ditem));
1019 		return -EUCLEAN;
1020 	}
1021 	/*
1022 	 * Remaining members like io_align/type/gen/dev_group aren't really
1023 	 * utilized.  Skip them to make later usage of them easier.
1024 	 */
1025 	return 0;
1026 }
1027 
1028 static int check_inode_item(struct extent_buffer *leaf,
1029 			    struct btrfs_key *key, int slot)
1030 {
1031 	struct btrfs_fs_info *fs_info = leaf->fs_info;
1032 	struct btrfs_inode_item *iitem;
1033 	u64 super_gen = btrfs_super_generation(fs_info->super_copy);
1034 	u32 valid_mask = (S_IFMT | S_ISUID | S_ISGID | S_ISVTX | 0777);
1035 	const u32 item_size = btrfs_item_size(leaf, slot);
1036 	u32 mode;
1037 	int ret;
1038 	u32 flags;
1039 	u32 ro_flags;
1040 
1041 	ret = check_inode_key(leaf, key, slot);
1042 	if (unlikely(ret < 0))
1043 		return ret;
1044 
1045 	if (unlikely(item_size != sizeof(*iitem))) {
1046 		generic_err(leaf, slot, "invalid item size: has %u expect %zu",
1047 			    item_size, sizeof(*iitem));
1048 		return -EUCLEAN;
1049 	}
1050 
1051 	iitem = btrfs_item_ptr(leaf, slot, struct btrfs_inode_item);
1052 
1053 	/* Here we use super block generation + 1 to handle log tree */
1054 	if (unlikely(btrfs_inode_generation(leaf, iitem) > super_gen + 1)) {
1055 		inode_item_err(leaf, slot,
1056 			"invalid inode generation: has %llu expect (0, %llu]",
1057 			       btrfs_inode_generation(leaf, iitem),
1058 			       super_gen + 1);
1059 		return -EUCLEAN;
1060 	}
1061 	/* Note for ROOT_TREE_DIR_ITEM, mkfs could set its transid 0 */
1062 	if (unlikely(btrfs_inode_transid(leaf, iitem) > super_gen + 1)) {
1063 		inode_item_err(leaf, slot,
1064 			"invalid inode transid: has %llu expect [0, %llu]",
1065 			       btrfs_inode_transid(leaf, iitem), super_gen + 1);
1066 		return -EUCLEAN;
1067 	}
1068 
1069 	/*
1070 	 * For size and nbytes it's better not to be too strict, as for dir
1071 	 * item its size/nbytes can easily get wrong, but doesn't affect
1072 	 * anything in the fs. So here we skip the check.
1073 	 */
1074 	mode = btrfs_inode_mode(leaf, iitem);
1075 	if (unlikely(mode & ~valid_mask)) {
1076 		inode_item_err(leaf, slot,
1077 			       "unknown mode bit detected: 0x%x",
1078 			       mode & ~valid_mask);
1079 		return -EUCLEAN;
1080 	}
1081 
1082 	/*
1083 	 * S_IFMT is not bit mapped so we can't completely rely on
1084 	 * is_power_of_2/has_single_bit_set, but it can save us from checking
1085 	 * FIFO/CHR/DIR/REG.  Only needs to check BLK, LNK and SOCKS
1086 	 */
1087 	if (!has_single_bit_set(mode & S_IFMT)) {
1088 		if (unlikely(!S_ISLNK(mode) && !S_ISBLK(mode) && !S_ISSOCK(mode))) {
1089 			inode_item_err(leaf, slot,
1090 			"invalid mode: has 0%o expect valid S_IF* bit(s)",
1091 				       mode & S_IFMT);
1092 			return -EUCLEAN;
1093 		}
1094 	}
1095 	if (unlikely(S_ISDIR(mode) && btrfs_inode_nlink(leaf, iitem) > 1)) {
1096 		inode_item_err(leaf, slot,
1097 		       "invalid nlink: has %u expect no more than 1 for dir",
1098 			btrfs_inode_nlink(leaf, iitem));
1099 		return -EUCLEAN;
1100 	}
1101 	btrfs_inode_split_flags(btrfs_inode_flags(leaf, iitem), &flags, &ro_flags);
1102 	if (unlikely(flags & ~BTRFS_INODE_FLAG_MASK)) {
1103 		inode_item_err(leaf, slot,
1104 			       "unknown incompat flags detected: 0x%x", flags);
1105 		return -EUCLEAN;
1106 	}
1107 	if (unlikely(!sb_rdonly(fs_info->sb) &&
1108 		     (ro_flags & ~BTRFS_INODE_RO_FLAG_MASK))) {
1109 		inode_item_err(leaf, slot,
1110 			"unknown ro-compat flags detected on writeable mount: 0x%x",
1111 			ro_flags);
1112 		return -EUCLEAN;
1113 	}
1114 	return 0;
1115 }
1116 
1117 static int check_root_item(struct extent_buffer *leaf, struct btrfs_key *key,
1118 			   int slot)
1119 {
1120 	struct btrfs_fs_info *fs_info = leaf->fs_info;
1121 	struct btrfs_root_item ri = { 0 };
1122 	const u64 valid_root_flags = BTRFS_ROOT_SUBVOL_RDONLY |
1123 				     BTRFS_ROOT_SUBVOL_DEAD;
1124 	int ret;
1125 
1126 	ret = check_root_key(leaf, key, slot);
1127 	if (unlikely(ret < 0))
1128 		return ret;
1129 
1130 	if (unlikely(btrfs_item_size(leaf, slot) != sizeof(ri) &&
1131 		     btrfs_item_size(leaf, slot) !=
1132 		     btrfs_legacy_root_item_size())) {
1133 		generic_err(leaf, slot,
1134 			    "invalid root item size, have %u expect %zu or %u",
1135 			    btrfs_item_size(leaf, slot), sizeof(ri),
1136 			    btrfs_legacy_root_item_size());
1137 		return -EUCLEAN;
1138 	}
1139 
1140 	/*
1141 	 * For legacy root item, the members starting at generation_v2 will be
1142 	 * all filled with 0.
1143 	 * And since we allow geneartion_v2 as 0, it will still pass the check.
1144 	 */
1145 	read_extent_buffer(leaf, &ri, btrfs_item_ptr_offset(leaf, slot),
1146 			   btrfs_item_size(leaf, slot));
1147 
1148 	/* Generation related */
1149 	if (unlikely(btrfs_root_generation(&ri) >
1150 		     btrfs_super_generation(fs_info->super_copy) + 1)) {
1151 		generic_err(leaf, slot,
1152 			"invalid root generation, have %llu expect (0, %llu]",
1153 			    btrfs_root_generation(&ri),
1154 			    btrfs_super_generation(fs_info->super_copy) + 1);
1155 		return -EUCLEAN;
1156 	}
1157 	if (unlikely(btrfs_root_generation_v2(&ri) >
1158 		     btrfs_super_generation(fs_info->super_copy) + 1)) {
1159 		generic_err(leaf, slot,
1160 		"invalid root v2 generation, have %llu expect (0, %llu]",
1161 			    btrfs_root_generation_v2(&ri),
1162 			    btrfs_super_generation(fs_info->super_copy) + 1);
1163 		return -EUCLEAN;
1164 	}
1165 	if (unlikely(btrfs_root_last_snapshot(&ri) >
1166 		     btrfs_super_generation(fs_info->super_copy) + 1)) {
1167 		generic_err(leaf, slot,
1168 		"invalid root last_snapshot, have %llu expect (0, %llu]",
1169 			    btrfs_root_last_snapshot(&ri),
1170 			    btrfs_super_generation(fs_info->super_copy) + 1);
1171 		return -EUCLEAN;
1172 	}
1173 
1174 	/* Alignment and level check */
1175 	if (unlikely(!IS_ALIGNED(btrfs_root_bytenr(&ri), fs_info->sectorsize))) {
1176 		generic_err(leaf, slot,
1177 		"invalid root bytenr, have %llu expect to be aligned to %u",
1178 			    btrfs_root_bytenr(&ri), fs_info->sectorsize);
1179 		return -EUCLEAN;
1180 	}
1181 	if (unlikely(btrfs_root_level(&ri) >= BTRFS_MAX_LEVEL)) {
1182 		generic_err(leaf, slot,
1183 			    "invalid root level, have %u expect [0, %u]",
1184 			    btrfs_root_level(&ri), BTRFS_MAX_LEVEL - 1);
1185 		return -EUCLEAN;
1186 	}
1187 	if (unlikely(btrfs_root_drop_level(&ri) >= BTRFS_MAX_LEVEL)) {
1188 		generic_err(leaf, slot,
1189 			    "invalid root level, have %u expect [0, %u]",
1190 			    btrfs_root_drop_level(&ri), BTRFS_MAX_LEVEL - 1);
1191 		return -EUCLEAN;
1192 	}
1193 
1194 	/* Flags check */
1195 	if (unlikely(btrfs_root_flags(&ri) & ~valid_root_flags)) {
1196 		generic_err(leaf, slot,
1197 			    "invalid root flags, have 0x%llx expect mask 0x%llx",
1198 			    btrfs_root_flags(&ri), valid_root_flags);
1199 		return -EUCLEAN;
1200 	}
1201 	return 0;
1202 }
1203 
1204 __printf(3,4)
1205 __cold
1206 static void extent_err(const struct extent_buffer *eb, int slot,
1207 		       const char *fmt, ...)
1208 {
1209 	struct btrfs_key key;
1210 	struct va_format vaf;
1211 	va_list args;
1212 	u64 bytenr;
1213 	u64 len;
1214 
1215 	btrfs_item_key_to_cpu(eb, &key, slot);
1216 	bytenr = key.objectid;
1217 	if (key.type == BTRFS_METADATA_ITEM_KEY ||
1218 	    key.type == BTRFS_TREE_BLOCK_REF_KEY ||
1219 	    key.type == BTRFS_SHARED_BLOCK_REF_KEY)
1220 		len = eb->fs_info->nodesize;
1221 	else
1222 		len = key.offset;
1223 	va_start(args, fmt);
1224 
1225 	vaf.fmt = fmt;
1226 	vaf.va = &args;
1227 
1228 	btrfs_crit(eb->fs_info,
1229 	"corrupt %s: block=%llu slot=%d extent bytenr=%llu len=%llu %pV",
1230 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
1231 		eb->start, slot, bytenr, len, &vaf);
1232 	va_end(args);
1233 }
1234 
1235 static int check_extent_item(struct extent_buffer *leaf,
1236 			     struct btrfs_key *key, int slot)
1237 {
1238 	struct btrfs_fs_info *fs_info = leaf->fs_info;
1239 	struct btrfs_extent_item *ei;
1240 	bool is_tree_block = false;
1241 	unsigned long ptr;	/* Current pointer inside inline refs */
1242 	unsigned long end;	/* Extent item end */
1243 	const u32 item_size = btrfs_item_size(leaf, slot);
1244 	u64 flags;
1245 	u64 generation;
1246 	u64 total_refs;		/* Total refs in btrfs_extent_item */
1247 	u64 inline_refs = 0;	/* found total inline refs */
1248 
1249 	if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1250 		     !btrfs_fs_incompat(fs_info, SKINNY_METADATA))) {
1251 		generic_err(leaf, slot,
1252 "invalid key type, METADATA_ITEM type invalid when SKINNY_METADATA feature disabled");
1253 		return -EUCLEAN;
1254 	}
1255 	/* key->objectid is the bytenr for both key types */
1256 	if (unlikely(!IS_ALIGNED(key->objectid, fs_info->sectorsize))) {
1257 		generic_err(leaf, slot,
1258 		"invalid key objectid, have %llu expect to be aligned to %u",
1259 			   key->objectid, fs_info->sectorsize);
1260 		return -EUCLEAN;
1261 	}
1262 
1263 	/* key->offset is tree level for METADATA_ITEM_KEY */
1264 	if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1265 		     key->offset >= BTRFS_MAX_LEVEL)) {
1266 		extent_err(leaf, slot,
1267 			   "invalid tree level, have %llu expect [0, %u]",
1268 			   key->offset, BTRFS_MAX_LEVEL - 1);
1269 		return -EUCLEAN;
1270 	}
1271 
1272 	/*
1273 	 * EXTENT/METADATA_ITEM consists of:
1274 	 * 1) One btrfs_extent_item
1275 	 *    Records the total refs, type and generation of the extent.
1276 	 *
1277 	 * 2) One btrfs_tree_block_info (for EXTENT_ITEM and tree backref only)
1278 	 *    Records the first key and level of the tree block.
1279 	 *
1280 	 * 2) Zero or more btrfs_extent_inline_ref(s)
1281 	 *    Each inline ref has one btrfs_extent_inline_ref shows:
1282 	 *    2.1) The ref type, one of the 4
1283 	 *         TREE_BLOCK_REF	Tree block only
1284 	 *         SHARED_BLOCK_REF	Tree block only
1285 	 *         EXTENT_DATA_REF	Data only
1286 	 *         SHARED_DATA_REF	Data only
1287 	 *    2.2) Ref type specific data
1288 	 *         Either using btrfs_extent_inline_ref::offset, or specific
1289 	 *         data structure.
1290 	 */
1291 	if (unlikely(item_size < sizeof(*ei))) {
1292 		extent_err(leaf, slot,
1293 			   "invalid item size, have %u expect [%zu, %u)",
1294 			   item_size, sizeof(*ei),
1295 			   BTRFS_LEAF_DATA_SIZE(fs_info));
1296 		return -EUCLEAN;
1297 	}
1298 	end = item_size + btrfs_item_ptr_offset(leaf, slot);
1299 
1300 	/* Checks against extent_item */
1301 	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
1302 	flags = btrfs_extent_flags(leaf, ei);
1303 	total_refs = btrfs_extent_refs(leaf, ei);
1304 	generation = btrfs_extent_generation(leaf, ei);
1305 	if (unlikely(generation >
1306 		     btrfs_super_generation(fs_info->super_copy) + 1)) {
1307 		extent_err(leaf, slot,
1308 			   "invalid generation, have %llu expect (0, %llu]",
1309 			   generation,
1310 			   btrfs_super_generation(fs_info->super_copy) + 1);
1311 		return -EUCLEAN;
1312 	}
1313 	if (unlikely(!has_single_bit_set(flags & (BTRFS_EXTENT_FLAG_DATA |
1314 						  BTRFS_EXTENT_FLAG_TREE_BLOCK)))) {
1315 		extent_err(leaf, slot,
1316 		"invalid extent flag, have 0x%llx expect 1 bit set in 0x%llx",
1317 			flags, BTRFS_EXTENT_FLAG_DATA |
1318 			BTRFS_EXTENT_FLAG_TREE_BLOCK);
1319 		return -EUCLEAN;
1320 	}
1321 	is_tree_block = !!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK);
1322 	if (is_tree_block) {
1323 		if (unlikely(key->type == BTRFS_EXTENT_ITEM_KEY &&
1324 			     key->offset != fs_info->nodesize)) {
1325 			extent_err(leaf, slot,
1326 				   "invalid extent length, have %llu expect %u",
1327 				   key->offset, fs_info->nodesize);
1328 			return -EUCLEAN;
1329 		}
1330 	} else {
1331 		if (unlikely(key->type != BTRFS_EXTENT_ITEM_KEY)) {
1332 			extent_err(leaf, slot,
1333 			"invalid key type, have %u expect %u for data backref",
1334 				   key->type, BTRFS_EXTENT_ITEM_KEY);
1335 			return -EUCLEAN;
1336 		}
1337 		if (unlikely(!IS_ALIGNED(key->offset, fs_info->sectorsize))) {
1338 			extent_err(leaf, slot,
1339 			"invalid extent length, have %llu expect aligned to %u",
1340 				   key->offset, fs_info->sectorsize);
1341 			return -EUCLEAN;
1342 		}
1343 		if (unlikely(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
1344 			extent_err(leaf, slot,
1345 			"invalid extent flag, data has full backref set");
1346 			return -EUCLEAN;
1347 		}
1348 	}
1349 	ptr = (unsigned long)(struct btrfs_extent_item *)(ei + 1);
1350 
1351 	/* Check the special case of btrfs_tree_block_info */
1352 	if (is_tree_block && key->type != BTRFS_METADATA_ITEM_KEY) {
1353 		struct btrfs_tree_block_info *info;
1354 
1355 		info = (struct btrfs_tree_block_info *)ptr;
1356 		if (unlikely(btrfs_tree_block_level(leaf, info) >= BTRFS_MAX_LEVEL)) {
1357 			extent_err(leaf, slot,
1358 			"invalid tree block info level, have %u expect [0, %u]",
1359 				   btrfs_tree_block_level(leaf, info),
1360 				   BTRFS_MAX_LEVEL - 1);
1361 			return -EUCLEAN;
1362 		}
1363 		ptr = (unsigned long)(struct btrfs_tree_block_info *)(info + 1);
1364 	}
1365 
1366 	/* Check inline refs */
1367 	while (ptr < end) {
1368 		struct btrfs_extent_inline_ref *iref;
1369 		struct btrfs_extent_data_ref *dref;
1370 		struct btrfs_shared_data_ref *sref;
1371 		u64 dref_offset;
1372 		u64 inline_offset;
1373 		u8 inline_type;
1374 
1375 		if (unlikely(ptr + sizeof(*iref) > end)) {
1376 			extent_err(leaf, slot,
1377 "inline ref item overflows extent item, ptr %lu iref size %zu end %lu",
1378 				   ptr, sizeof(*iref), end);
1379 			return -EUCLEAN;
1380 		}
1381 		iref = (struct btrfs_extent_inline_ref *)ptr;
1382 		inline_type = btrfs_extent_inline_ref_type(leaf, iref);
1383 		inline_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1384 		if (unlikely(ptr + btrfs_extent_inline_ref_size(inline_type) > end)) {
1385 			extent_err(leaf, slot,
1386 "inline ref item overflows extent item, ptr %lu iref size %u end %lu",
1387 				   ptr, inline_type, end);
1388 			return -EUCLEAN;
1389 		}
1390 
1391 		switch (inline_type) {
1392 		/* inline_offset is subvolid of the owner, no need to check */
1393 		case BTRFS_TREE_BLOCK_REF_KEY:
1394 			inline_refs++;
1395 			break;
1396 		/* Contains parent bytenr */
1397 		case BTRFS_SHARED_BLOCK_REF_KEY:
1398 			if (unlikely(!IS_ALIGNED(inline_offset,
1399 						 fs_info->sectorsize))) {
1400 				extent_err(leaf, slot,
1401 		"invalid tree parent bytenr, have %llu expect aligned to %u",
1402 					   inline_offset, fs_info->sectorsize);
1403 				return -EUCLEAN;
1404 			}
1405 			inline_refs++;
1406 			break;
1407 		/*
1408 		 * Contains owner subvolid, owner key objectid, adjusted offset.
1409 		 * The only obvious corruption can happen in that offset.
1410 		 */
1411 		case BTRFS_EXTENT_DATA_REF_KEY:
1412 			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1413 			dref_offset = btrfs_extent_data_ref_offset(leaf, dref);
1414 			if (unlikely(!IS_ALIGNED(dref_offset,
1415 						 fs_info->sectorsize))) {
1416 				extent_err(leaf, slot,
1417 		"invalid data ref offset, have %llu expect aligned to %u",
1418 					   dref_offset, fs_info->sectorsize);
1419 				return -EUCLEAN;
1420 			}
1421 			inline_refs += btrfs_extent_data_ref_count(leaf, dref);
1422 			break;
1423 		/* Contains parent bytenr and ref count */
1424 		case BTRFS_SHARED_DATA_REF_KEY:
1425 			sref = (struct btrfs_shared_data_ref *)(iref + 1);
1426 			if (unlikely(!IS_ALIGNED(inline_offset,
1427 						 fs_info->sectorsize))) {
1428 				extent_err(leaf, slot,
1429 		"invalid data parent bytenr, have %llu expect aligned to %u",
1430 					   inline_offset, fs_info->sectorsize);
1431 				return -EUCLEAN;
1432 			}
1433 			inline_refs += btrfs_shared_data_ref_count(leaf, sref);
1434 			break;
1435 		default:
1436 			extent_err(leaf, slot, "unknown inline ref type: %u",
1437 				   inline_type);
1438 			return -EUCLEAN;
1439 		}
1440 		ptr += btrfs_extent_inline_ref_size(inline_type);
1441 	}
1442 	/* No padding is allowed */
1443 	if (unlikely(ptr != end)) {
1444 		extent_err(leaf, slot,
1445 			   "invalid extent item size, padding bytes found");
1446 		return -EUCLEAN;
1447 	}
1448 
1449 	/* Finally, check the inline refs against total refs */
1450 	if (unlikely(inline_refs > total_refs)) {
1451 		extent_err(leaf, slot,
1452 			"invalid extent refs, have %llu expect >= inline %llu",
1453 			   total_refs, inline_refs);
1454 		return -EUCLEAN;
1455 	}
1456 	return 0;
1457 }
1458 
1459 static int check_simple_keyed_refs(struct extent_buffer *leaf,
1460 				   struct btrfs_key *key, int slot)
1461 {
1462 	u32 expect_item_size = 0;
1463 
1464 	if (key->type == BTRFS_SHARED_DATA_REF_KEY)
1465 		expect_item_size = sizeof(struct btrfs_shared_data_ref);
1466 
1467 	if (unlikely(btrfs_item_size(leaf, slot) != expect_item_size)) {
1468 		generic_err(leaf, slot,
1469 		"invalid item size, have %u expect %u for key type %u",
1470 			    btrfs_item_size(leaf, slot),
1471 			    expect_item_size, key->type);
1472 		return -EUCLEAN;
1473 	}
1474 	if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1475 		generic_err(leaf, slot,
1476 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1477 			    key->objectid, leaf->fs_info->sectorsize);
1478 		return -EUCLEAN;
1479 	}
1480 	if (unlikely(key->type != BTRFS_TREE_BLOCK_REF_KEY &&
1481 		     !IS_ALIGNED(key->offset, leaf->fs_info->sectorsize))) {
1482 		extent_err(leaf, slot,
1483 		"invalid tree parent bytenr, have %llu expect aligned to %u",
1484 			   key->offset, leaf->fs_info->sectorsize);
1485 		return -EUCLEAN;
1486 	}
1487 	return 0;
1488 }
1489 
1490 static int check_extent_data_ref(struct extent_buffer *leaf,
1491 				 struct btrfs_key *key, int slot)
1492 {
1493 	struct btrfs_extent_data_ref *dref;
1494 	unsigned long ptr = btrfs_item_ptr_offset(leaf, slot);
1495 	const unsigned long end = ptr + btrfs_item_size(leaf, slot);
1496 
1497 	if (unlikely(btrfs_item_size(leaf, slot) % sizeof(*dref) != 0)) {
1498 		generic_err(leaf, slot,
1499 	"invalid item size, have %u expect aligned to %zu for key type %u",
1500 			    btrfs_item_size(leaf, slot),
1501 			    sizeof(*dref), key->type);
1502 		return -EUCLEAN;
1503 	}
1504 	if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1505 		generic_err(leaf, slot,
1506 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1507 			    key->objectid, leaf->fs_info->sectorsize);
1508 		return -EUCLEAN;
1509 	}
1510 	for (; ptr < end; ptr += sizeof(*dref)) {
1511 		u64 offset;
1512 
1513 		/*
1514 		 * We cannot check the extent_data_ref hash due to possible
1515 		 * overflow from the leaf due to hash collisions.
1516 		 */
1517 		dref = (struct btrfs_extent_data_ref *)ptr;
1518 		offset = btrfs_extent_data_ref_offset(leaf, dref);
1519 		if (unlikely(!IS_ALIGNED(offset, leaf->fs_info->sectorsize))) {
1520 			extent_err(leaf, slot,
1521 	"invalid extent data backref offset, have %llu expect aligned to %u",
1522 				   offset, leaf->fs_info->sectorsize);
1523 			return -EUCLEAN;
1524 		}
1525 	}
1526 	return 0;
1527 }
1528 
1529 #define inode_ref_err(eb, slot, fmt, args...)			\
1530 	inode_item_err(eb, slot, fmt, ##args)
1531 static int check_inode_ref(struct extent_buffer *leaf,
1532 			   struct btrfs_key *key, struct btrfs_key *prev_key,
1533 			   int slot)
1534 {
1535 	struct btrfs_inode_ref *iref;
1536 	unsigned long ptr;
1537 	unsigned long end;
1538 
1539 	if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
1540 		return -EUCLEAN;
1541 	/* namelen can't be 0, so item_size == sizeof() is also invalid */
1542 	if (unlikely(btrfs_item_size(leaf, slot) <= sizeof(*iref))) {
1543 		inode_ref_err(leaf, slot,
1544 			"invalid item size, have %u expect (%zu, %u)",
1545 			btrfs_item_size(leaf, slot),
1546 			sizeof(*iref), BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
1547 		return -EUCLEAN;
1548 	}
1549 
1550 	ptr = btrfs_item_ptr_offset(leaf, slot);
1551 	end = ptr + btrfs_item_size(leaf, slot);
1552 	while (ptr < end) {
1553 		u16 namelen;
1554 
1555 		if (unlikely(ptr + sizeof(iref) > end)) {
1556 			inode_ref_err(leaf, slot,
1557 			"inode ref overflow, ptr %lu end %lu inode_ref_size %zu",
1558 				ptr, end, sizeof(iref));
1559 			return -EUCLEAN;
1560 		}
1561 
1562 		iref = (struct btrfs_inode_ref *)ptr;
1563 		namelen = btrfs_inode_ref_name_len(leaf, iref);
1564 		if (unlikely(ptr + sizeof(*iref) + namelen > end)) {
1565 			inode_ref_err(leaf, slot,
1566 				"inode ref overflow, ptr %lu end %lu namelen %u",
1567 				ptr, end, namelen);
1568 			return -EUCLEAN;
1569 		}
1570 
1571 		/*
1572 		 * NOTE: In theory we should record all found index numbers
1573 		 * to find any duplicated indexes, but that will be too time
1574 		 * consuming for inodes with too many hard links.
1575 		 */
1576 		ptr += sizeof(*iref) + namelen;
1577 	}
1578 	return 0;
1579 }
1580 
1581 /*
1582  * Common point to switch the item-specific validation.
1583  */
1584 static int check_leaf_item(struct extent_buffer *leaf,
1585 			   struct btrfs_key *key, int slot,
1586 			   struct btrfs_key *prev_key)
1587 {
1588 	int ret = 0;
1589 	struct btrfs_chunk *chunk;
1590 
1591 	switch (key->type) {
1592 	case BTRFS_EXTENT_DATA_KEY:
1593 		ret = check_extent_data_item(leaf, key, slot, prev_key);
1594 		break;
1595 	case BTRFS_EXTENT_CSUM_KEY:
1596 		ret = check_csum_item(leaf, key, slot, prev_key);
1597 		break;
1598 	case BTRFS_DIR_ITEM_KEY:
1599 	case BTRFS_DIR_INDEX_KEY:
1600 	case BTRFS_XATTR_ITEM_KEY:
1601 		ret = check_dir_item(leaf, key, prev_key, slot);
1602 		break;
1603 	case BTRFS_INODE_REF_KEY:
1604 		ret = check_inode_ref(leaf, key, prev_key, slot);
1605 		break;
1606 	case BTRFS_BLOCK_GROUP_ITEM_KEY:
1607 		ret = check_block_group_item(leaf, key, slot);
1608 		break;
1609 	case BTRFS_CHUNK_ITEM_KEY:
1610 		chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1611 		ret = check_leaf_chunk_item(leaf, chunk, key, slot);
1612 		break;
1613 	case BTRFS_DEV_ITEM_KEY:
1614 		ret = check_dev_item(leaf, key, slot);
1615 		break;
1616 	case BTRFS_INODE_ITEM_KEY:
1617 		ret = check_inode_item(leaf, key, slot);
1618 		break;
1619 	case BTRFS_ROOT_ITEM_KEY:
1620 		ret = check_root_item(leaf, key, slot);
1621 		break;
1622 	case BTRFS_EXTENT_ITEM_KEY:
1623 	case BTRFS_METADATA_ITEM_KEY:
1624 		ret = check_extent_item(leaf, key, slot);
1625 		break;
1626 	case BTRFS_TREE_BLOCK_REF_KEY:
1627 	case BTRFS_SHARED_DATA_REF_KEY:
1628 	case BTRFS_SHARED_BLOCK_REF_KEY:
1629 		ret = check_simple_keyed_refs(leaf, key, slot);
1630 		break;
1631 	case BTRFS_EXTENT_DATA_REF_KEY:
1632 		ret = check_extent_data_ref(leaf, key, slot);
1633 		break;
1634 	}
1635 	return ret;
1636 }
1637 
1638 static int check_leaf(struct extent_buffer *leaf, bool check_item_data)
1639 {
1640 	struct btrfs_fs_info *fs_info = leaf->fs_info;
1641 	/* No valid key type is 0, so all key should be larger than this key */
1642 	struct btrfs_key prev_key = {0, 0, 0};
1643 	struct btrfs_key key;
1644 	u32 nritems = btrfs_header_nritems(leaf);
1645 	int slot;
1646 
1647 	if (unlikely(btrfs_header_level(leaf) != 0)) {
1648 		generic_err(leaf, 0,
1649 			"invalid level for leaf, have %d expect 0",
1650 			btrfs_header_level(leaf));
1651 		return -EUCLEAN;
1652 	}
1653 
1654 	/*
1655 	 * Extent buffers from a relocation tree have a owner field that
1656 	 * corresponds to the subvolume tree they are based on. So just from an
1657 	 * extent buffer alone we can not find out what is the id of the
1658 	 * corresponding subvolume tree, so we can not figure out if the extent
1659 	 * buffer corresponds to the root of the relocation tree or not. So
1660 	 * skip this check for relocation trees.
1661 	 */
1662 	if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
1663 		u64 owner = btrfs_header_owner(leaf);
1664 
1665 		/* These trees must never be empty */
1666 		if (unlikely(owner == BTRFS_ROOT_TREE_OBJECTID ||
1667 			     owner == BTRFS_CHUNK_TREE_OBJECTID ||
1668 			     owner == BTRFS_DEV_TREE_OBJECTID ||
1669 			     owner == BTRFS_FS_TREE_OBJECTID ||
1670 			     owner == BTRFS_DATA_RELOC_TREE_OBJECTID)) {
1671 			generic_err(leaf, 0,
1672 			"invalid root, root %llu must never be empty",
1673 				    owner);
1674 			return -EUCLEAN;
1675 		}
1676 
1677 		/* Unknown tree */
1678 		if (unlikely(owner == 0)) {
1679 			generic_err(leaf, 0,
1680 				"invalid owner, root 0 is not defined");
1681 			return -EUCLEAN;
1682 		}
1683 
1684 		/* EXTENT_TREE_V2 can have empty extent trees. */
1685 		if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1686 			return 0;
1687 
1688 		if (unlikely(owner == BTRFS_EXTENT_TREE_OBJECTID)) {
1689 			generic_err(leaf, 0,
1690 			"invalid root, root %llu must never be empty",
1691 				    owner);
1692 			return -EUCLEAN;
1693 		}
1694 
1695 		return 0;
1696 	}
1697 
1698 	if (unlikely(nritems == 0))
1699 		return 0;
1700 
1701 	/*
1702 	 * Check the following things to make sure this is a good leaf, and
1703 	 * leaf users won't need to bother with similar sanity checks:
1704 	 *
1705 	 * 1) key ordering
1706 	 * 2) item offset and size
1707 	 *    No overlap, no hole, all inside the leaf.
1708 	 * 3) item content
1709 	 *    If possible, do comprehensive sanity check.
1710 	 *    NOTE: All checks must only rely on the item data itself.
1711 	 */
1712 	for (slot = 0; slot < nritems; slot++) {
1713 		u32 item_end_expected;
1714 		u64 item_data_end;
1715 		int ret;
1716 
1717 		btrfs_item_key_to_cpu(leaf, &key, slot);
1718 
1719 		/* Make sure the keys are in the right order */
1720 		if (unlikely(btrfs_comp_cpu_keys(&prev_key, &key) >= 0)) {
1721 			generic_err(leaf, slot,
1722 	"bad key order, prev (%llu %u %llu) current (%llu %u %llu)",
1723 				prev_key.objectid, prev_key.type,
1724 				prev_key.offset, key.objectid, key.type,
1725 				key.offset);
1726 			return -EUCLEAN;
1727 		}
1728 
1729 		item_data_end = (u64)btrfs_item_offset(leaf, slot) +
1730 				btrfs_item_size(leaf, slot);
1731 		/*
1732 		 * Make sure the offset and ends are right, remember that the
1733 		 * item data starts at the end of the leaf and grows towards the
1734 		 * front.
1735 		 */
1736 		if (slot == 0)
1737 			item_end_expected = BTRFS_LEAF_DATA_SIZE(fs_info);
1738 		else
1739 			item_end_expected = btrfs_item_offset(leaf,
1740 								 slot - 1);
1741 		if (unlikely(item_data_end != item_end_expected)) {
1742 			generic_err(leaf, slot,
1743 				"unexpected item end, have %llu expect %u",
1744 				item_data_end, item_end_expected);
1745 			return -EUCLEAN;
1746 		}
1747 
1748 		/*
1749 		 * Check to make sure that we don't point outside of the leaf,
1750 		 * just in case all the items are consistent to each other, but
1751 		 * all point outside of the leaf.
1752 		 */
1753 		if (unlikely(item_data_end > BTRFS_LEAF_DATA_SIZE(fs_info))) {
1754 			generic_err(leaf, slot,
1755 			"slot end outside of leaf, have %llu expect range [0, %u]",
1756 				item_data_end, BTRFS_LEAF_DATA_SIZE(fs_info));
1757 			return -EUCLEAN;
1758 		}
1759 
1760 		/* Also check if the item pointer overlaps with btrfs item. */
1761 		if (unlikely(btrfs_item_ptr_offset(leaf, slot) <
1762 			     btrfs_item_nr_offset(slot) + sizeof(struct btrfs_item))) {
1763 			generic_err(leaf, slot,
1764 		"slot overlaps with its data, item end %lu data start %lu",
1765 				btrfs_item_nr_offset(slot) +
1766 				sizeof(struct btrfs_item),
1767 				btrfs_item_ptr_offset(leaf, slot));
1768 			return -EUCLEAN;
1769 		}
1770 
1771 		if (check_item_data) {
1772 			/*
1773 			 * Check if the item size and content meet other
1774 			 * criteria
1775 			 */
1776 			ret = check_leaf_item(leaf, &key, slot, &prev_key);
1777 			if (unlikely(ret < 0))
1778 				return ret;
1779 		}
1780 
1781 		prev_key.objectid = key.objectid;
1782 		prev_key.type = key.type;
1783 		prev_key.offset = key.offset;
1784 	}
1785 
1786 	return 0;
1787 }
1788 
1789 int btrfs_check_leaf_full(struct extent_buffer *leaf)
1790 {
1791 	return check_leaf(leaf, true);
1792 }
1793 ALLOW_ERROR_INJECTION(btrfs_check_leaf_full, ERRNO);
1794 
1795 int btrfs_check_leaf_relaxed(struct extent_buffer *leaf)
1796 {
1797 	return check_leaf(leaf, false);
1798 }
1799 
1800 int btrfs_check_node(struct extent_buffer *node)
1801 {
1802 	struct btrfs_fs_info *fs_info = node->fs_info;
1803 	unsigned long nr = btrfs_header_nritems(node);
1804 	struct btrfs_key key, next_key;
1805 	int slot;
1806 	int level = btrfs_header_level(node);
1807 	u64 bytenr;
1808 	int ret = 0;
1809 
1810 	if (unlikely(level <= 0 || level >= BTRFS_MAX_LEVEL)) {
1811 		generic_err(node, 0,
1812 			"invalid level for node, have %d expect [1, %d]",
1813 			level, BTRFS_MAX_LEVEL - 1);
1814 		return -EUCLEAN;
1815 	}
1816 	if (unlikely(nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(fs_info))) {
1817 		btrfs_crit(fs_info,
1818 "corrupt node: root=%llu block=%llu, nritems too %s, have %lu expect range [1,%u]",
1819 			   btrfs_header_owner(node), node->start,
1820 			   nr == 0 ? "small" : "large", nr,
1821 			   BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1822 		return -EUCLEAN;
1823 	}
1824 
1825 	for (slot = 0; slot < nr - 1; slot++) {
1826 		bytenr = btrfs_node_blockptr(node, slot);
1827 		btrfs_node_key_to_cpu(node, &key, slot);
1828 		btrfs_node_key_to_cpu(node, &next_key, slot + 1);
1829 
1830 		if (unlikely(!bytenr)) {
1831 			generic_err(node, slot,
1832 				"invalid NULL node pointer");
1833 			ret = -EUCLEAN;
1834 			goto out;
1835 		}
1836 		if (unlikely(!IS_ALIGNED(bytenr, fs_info->sectorsize))) {
1837 			generic_err(node, slot,
1838 			"unaligned pointer, have %llu should be aligned to %u",
1839 				bytenr, fs_info->sectorsize);
1840 			ret = -EUCLEAN;
1841 			goto out;
1842 		}
1843 
1844 		if (unlikely(btrfs_comp_cpu_keys(&key, &next_key) >= 0)) {
1845 			generic_err(node, slot,
1846 	"bad key order, current (%llu %u %llu) next (%llu %u %llu)",
1847 				key.objectid, key.type, key.offset,
1848 				next_key.objectid, next_key.type,
1849 				next_key.offset);
1850 			ret = -EUCLEAN;
1851 			goto out;
1852 		}
1853 	}
1854 out:
1855 	return ret;
1856 }
1857 ALLOW_ERROR_INJECTION(btrfs_check_node, ERRNO);
1858 
1859 int btrfs_check_eb_owner(const struct extent_buffer *eb, u64 root_owner)
1860 {
1861 	const bool is_subvol = is_fstree(root_owner);
1862 	const u64 eb_owner = btrfs_header_owner(eb);
1863 
1864 	/*
1865 	 * Skip dummy fs, as selftests don't create unique ebs for each dummy
1866 	 * root.
1867 	 */
1868 	if (test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &eb->fs_info->fs_state))
1869 		return 0;
1870 	/*
1871 	 * There are several call sites (backref walking, qgroup, and data
1872 	 * reloc) passing 0 as @root_owner, as they are not holding the
1873 	 * tree root.  In that case, we can not do a reliable ownership check,
1874 	 * so just exit.
1875 	 */
1876 	if (root_owner == 0)
1877 		return 0;
1878 	/*
1879 	 * These trees use key.offset as their owner, our callers don't have
1880 	 * the extra capacity to pass key.offset here.  So we just skip them.
1881 	 */
1882 	if (root_owner == BTRFS_TREE_LOG_OBJECTID ||
1883 	    root_owner == BTRFS_TREE_RELOC_OBJECTID)
1884 		return 0;
1885 
1886 	if (!is_subvol) {
1887 		/* For non-subvolume trees, the eb owner should match root owner */
1888 		if (unlikely(root_owner != eb_owner)) {
1889 			btrfs_crit(eb->fs_info,
1890 "corrupted %s, root=%llu block=%llu owner mismatch, have %llu expect %llu",
1891 				btrfs_header_level(eb) == 0 ? "leaf" : "node",
1892 				root_owner, btrfs_header_bytenr(eb), eb_owner,
1893 				root_owner);
1894 			return -EUCLEAN;
1895 		}
1896 		return 0;
1897 	}
1898 
1899 	/*
1900 	 * For subvolume trees, owners can mismatch, but they should all belong
1901 	 * to subvolume trees.
1902 	 */
1903 	if (unlikely(is_subvol != is_fstree(eb_owner))) {
1904 		btrfs_crit(eb->fs_info,
1905 "corrupted %s, root=%llu block=%llu owner mismatch, have %llu expect [%llu, %llu]",
1906 			btrfs_header_level(eb) == 0 ? "leaf" : "node",
1907 			root_owner, btrfs_header_bytenr(eb), eb_owner,
1908 			BTRFS_FIRST_FREE_OBJECTID, BTRFS_LAST_FREE_OBJECTID);
1909 		return -EUCLEAN;
1910 	}
1911 	return 0;
1912 }
1913