xref: /linux/fs/btrfs/tree-checker.c (revision 2d7f3d1a5866705be2393150e1ffdf67030ab88d)
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 "messages.h"
22 #include "ctree.h"
23 #include "tree-checker.h"
24 #include "disk-io.h"
25 #include "compression.h"
26 #include "volumes.h"
27 #include "misc.h"
28 #include "fs.h"
29 #include "accessors.h"
30 #include "file-item.h"
31 #include "inode-item.h"
32 #include "dir-item.h"
33 #include "raid-stripe-tree.h"
34 #include "extent-tree.h"
35 
36 /*
37  * Error message should follow the following format:
38  * corrupt <type>: <identifier>, <reason>[, <bad_value>]
39  *
40  * @type:	leaf or node
41  * @identifier:	the necessary info to locate the leaf/node.
42  * 		It's recommended to decode key.objecitd/offset if it's
43  * 		meaningful.
44  * @reason:	describe the error
45  * @bad_value:	optional, it's recommended to output bad value and its
46  *		expected value (range).
47  *
48  * Since comma is used to separate the components, only space is allowed
49  * inside each component.
50  */
51 
52 /*
53  * Append generic "corrupt leaf/node root=%llu block=%llu slot=%d: " to @fmt.
54  * Allows callers to customize the output.
55  */
56 __printf(3, 4)
57 __cold
58 static void generic_err(const struct extent_buffer *eb, int slot,
59 			const char *fmt, ...)
60 {
61 	const struct btrfs_fs_info *fs_info = eb->fs_info;
62 	struct va_format vaf;
63 	va_list args;
64 
65 	va_start(args, fmt);
66 
67 	vaf.fmt = fmt;
68 	vaf.va = &args;
69 
70 	btrfs_crit(fs_info,
71 		"corrupt %s: root=%llu block=%llu slot=%d, %pV",
72 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
73 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot, &vaf);
74 	va_end(args);
75 }
76 
77 /*
78  * Customized reporter for extent data item, since its key objectid and
79  * offset has its own meaning.
80  */
81 __printf(3, 4)
82 __cold
83 static void file_extent_err(const struct extent_buffer *eb, int slot,
84 			    const char *fmt, ...)
85 {
86 	const struct btrfs_fs_info *fs_info = eb->fs_info;
87 	struct btrfs_key key;
88 	struct va_format vaf;
89 	va_list args;
90 
91 	btrfs_item_key_to_cpu(eb, &key, slot);
92 	va_start(args, fmt);
93 
94 	vaf.fmt = fmt;
95 	vaf.va = &args;
96 
97 	btrfs_crit(fs_info,
98 	"corrupt %s: root=%llu block=%llu slot=%d ino=%llu file_offset=%llu, %pV",
99 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
100 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
101 		key.objectid, key.offset, &vaf);
102 	va_end(args);
103 }
104 
105 /*
106  * Return 0 if the btrfs_file_extent_##name is aligned to @alignment
107  * Else return 1
108  */
109 #define CHECK_FE_ALIGNED(leaf, slot, fi, name, alignment)		      \
110 ({									      \
111 	if (unlikely(!IS_ALIGNED(btrfs_file_extent_##name((leaf), (fi)),      \
112 				 (alignment))))				      \
113 		file_extent_err((leaf), (slot),				      \
114 	"invalid %s for file extent, have %llu, should be aligned to %u",     \
115 			(#name), btrfs_file_extent_##name((leaf), (fi)),      \
116 			(alignment));					      \
117 	(!IS_ALIGNED(btrfs_file_extent_##name((leaf), (fi)), (alignment)));   \
118 })
119 
120 static u64 file_extent_end(struct extent_buffer *leaf,
121 			   struct btrfs_key *key,
122 			   struct btrfs_file_extent_item *extent)
123 {
124 	u64 end;
125 	u64 len;
126 
127 	if (btrfs_file_extent_type(leaf, extent) == BTRFS_FILE_EXTENT_INLINE) {
128 		len = btrfs_file_extent_ram_bytes(leaf, extent);
129 		end = ALIGN(key->offset + len, leaf->fs_info->sectorsize);
130 	} else {
131 		len = btrfs_file_extent_num_bytes(leaf, extent);
132 		end = key->offset + len;
133 	}
134 	return end;
135 }
136 
137 /*
138  * Customized report for dir_item, the only new important information is
139  * key->objectid, which represents inode number
140  */
141 __printf(3, 4)
142 __cold
143 static void dir_item_err(const struct extent_buffer *eb, int slot,
144 			 const char *fmt, ...)
145 {
146 	const struct btrfs_fs_info *fs_info = eb->fs_info;
147 	struct btrfs_key key;
148 	struct va_format vaf;
149 	va_list args;
150 
151 	btrfs_item_key_to_cpu(eb, &key, slot);
152 	va_start(args, fmt);
153 
154 	vaf.fmt = fmt;
155 	vaf.va = &args;
156 
157 	btrfs_crit(fs_info,
158 		"corrupt %s: root=%llu block=%llu slot=%d ino=%llu, %pV",
159 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
160 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
161 		key.objectid, &vaf);
162 	va_end(args);
163 }
164 
165 /*
166  * This functions checks prev_key->objectid, to ensure current key and prev_key
167  * share the same objectid as inode number.
168  *
169  * This is to detect missing INODE_ITEM in subvolume trees.
170  *
171  * Return true if everything is OK or we don't need to check.
172  * Return false if anything is wrong.
173  */
174 static bool check_prev_ino(struct extent_buffer *leaf,
175 			   struct btrfs_key *key, int slot,
176 			   struct btrfs_key *prev_key)
177 {
178 	/* No prev key, skip check */
179 	if (slot == 0)
180 		return true;
181 
182 	/* Only these key->types needs to be checked */
183 	ASSERT(key->type == BTRFS_XATTR_ITEM_KEY ||
184 	       key->type == BTRFS_INODE_REF_KEY ||
185 	       key->type == BTRFS_DIR_INDEX_KEY ||
186 	       key->type == BTRFS_DIR_ITEM_KEY ||
187 	       key->type == BTRFS_EXTENT_DATA_KEY);
188 
189 	/*
190 	 * Only subvolume trees along with their reloc trees need this check.
191 	 * Things like log tree doesn't follow this ino requirement.
192 	 */
193 	if (!is_fstree(btrfs_header_owner(leaf)))
194 		return true;
195 
196 	if (key->objectid == prev_key->objectid)
197 		return true;
198 
199 	/* Error found */
200 	dir_item_err(leaf, slot,
201 		"invalid previous key objectid, have %llu expect %llu",
202 		prev_key->objectid, key->objectid);
203 	return false;
204 }
205 static int check_extent_data_item(struct extent_buffer *leaf,
206 				  struct btrfs_key *key, int slot,
207 				  struct btrfs_key *prev_key)
208 {
209 	struct btrfs_fs_info *fs_info = leaf->fs_info;
210 	struct btrfs_file_extent_item *fi;
211 	u32 sectorsize = fs_info->sectorsize;
212 	u32 item_size = btrfs_item_size(leaf, slot);
213 	u64 extent_end;
214 
215 	if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) {
216 		file_extent_err(leaf, slot,
217 "unaligned file_offset for file extent, have %llu should be aligned to %u",
218 			key->offset, sectorsize);
219 		return -EUCLEAN;
220 	}
221 
222 	/*
223 	 * Previous key must have the same key->objectid (ino).
224 	 * It can be XATTR_ITEM, INODE_ITEM or just another EXTENT_DATA.
225 	 * But if objectids mismatch, it means we have a missing
226 	 * INODE_ITEM.
227 	 */
228 	if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
229 		return -EUCLEAN;
230 
231 	fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
232 
233 	/*
234 	 * Make sure the item contains at least inline header, so the file
235 	 * extent type is not some garbage.
236 	 */
237 	if (unlikely(item_size < BTRFS_FILE_EXTENT_INLINE_DATA_START)) {
238 		file_extent_err(leaf, slot,
239 				"invalid item size, have %u expect [%zu, %u)",
240 				item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START,
241 				SZ_4K);
242 		return -EUCLEAN;
243 	}
244 	if (unlikely(btrfs_file_extent_type(leaf, fi) >=
245 		     BTRFS_NR_FILE_EXTENT_TYPES)) {
246 		file_extent_err(leaf, slot,
247 		"invalid type for file extent, have %u expect range [0, %u]",
248 			btrfs_file_extent_type(leaf, fi),
249 			BTRFS_NR_FILE_EXTENT_TYPES - 1);
250 		return -EUCLEAN;
251 	}
252 
253 	/*
254 	 * Support for new compression/encryption must introduce incompat flag,
255 	 * and must be caught in open_ctree().
256 	 */
257 	if (unlikely(btrfs_file_extent_compression(leaf, fi) >=
258 		     BTRFS_NR_COMPRESS_TYPES)) {
259 		file_extent_err(leaf, slot,
260 	"invalid compression for file extent, have %u expect range [0, %u]",
261 			btrfs_file_extent_compression(leaf, fi),
262 			BTRFS_NR_COMPRESS_TYPES - 1);
263 		return -EUCLEAN;
264 	}
265 	if (unlikely(btrfs_file_extent_encryption(leaf, fi))) {
266 		file_extent_err(leaf, slot,
267 			"invalid encryption for file extent, have %u expect 0",
268 			btrfs_file_extent_encryption(leaf, fi));
269 		return -EUCLEAN;
270 	}
271 	if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
272 		/* Inline extent must have 0 as key offset */
273 		if (unlikely(key->offset)) {
274 			file_extent_err(leaf, slot,
275 		"invalid file_offset for inline file extent, have %llu expect 0",
276 				key->offset);
277 			return -EUCLEAN;
278 		}
279 
280 		/* Compressed inline extent has no on-disk size, skip it */
281 		if (btrfs_file_extent_compression(leaf, fi) !=
282 		    BTRFS_COMPRESS_NONE)
283 			return 0;
284 
285 		/* Uncompressed inline extent size must match item size */
286 		if (unlikely(item_size != BTRFS_FILE_EXTENT_INLINE_DATA_START +
287 					  btrfs_file_extent_ram_bytes(leaf, fi))) {
288 			file_extent_err(leaf, slot,
289 	"invalid ram_bytes for uncompressed inline extent, have %u expect %llu",
290 				item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START +
291 				btrfs_file_extent_ram_bytes(leaf, fi));
292 			return -EUCLEAN;
293 		}
294 		return 0;
295 	}
296 
297 	/* Regular or preallocated extent has fixed item size */
298 	if (unlikely(item_size != sizeof(*fi))) {
299 		file_extent_err(leaf, slot,
300 	"invalid item size for reg/prealloc file extent, have %u expect %zu",
301 			item_size, sizeof(*fi));
302 		return -EUCLEAN;
303 	}
304 	if (unlikely(CHECK_FE_ALIGNED(leaf, slot, fi, ram_bytes, sectorsize) ||
305 		     CHECK_FE_ALIGNED(leaf, slot, fi, disk_bytenr, sectorsize) ||
306 		     CHECK_FE_ALIGNED(leaf, slot, fi, disk_num_bytes, sectorsize) ||
307 		     CHECK_FE_ALIGNED(leaf, slot, fi, offset, sectorsize) ||
308 		     CHECK_FE_ALIGNED(leaf, slot, fi, num_bytes, sectorsize)))
309 		return -EUCLEAN;
310 
311 	/* Catch extent end overflow */
312 	if (unlikely(check_add_overflow(btrfs_file_extent_num_bytes(leaf, fi),
313 					key->offset, &extent_end))) {
314 		file_extent_err(leaf, slot,
315 	"extent end overflow, have file offset %llu extent num bytes %llu",
316 				key->offset,
317 				btrfs_file_extent_num_bytes(leaf, fi));
318 		return -EUCLEAN;
319 	}
320 
321 	/*
322 	 * Check that no two consecutive file extent items, in the same leaf,
323 	 * present ranges that overlap each other.
324 	 */
325 	if (slot > 0 &&
326 	    prev_key->objectid == key->objectid &&
327 	    prev_key->type == BTRFS_EXTENT_DATA_KEY) {
328 		struct btrfs_file_extent_item *prev_fi;
329 		u64 prev_end;
330 
331 		prev_fi = btrfs_item_ptr(leaf, slot - 1,
332 					 struct btrfs_file_extent_item);
333 		prev_end = file_extent_end(leaf, prev_key, prev_fi);
334 		if (unlikely(prev_end > key->offset)) {
335 			file_extent_err(leaf, slot - 1,
336 "file extent end range (%llu) goes beyond start offset (%llu) of the next file extent",
337 					prev_end, key->offset);
338 			return -EUCLEAN;
339 		}
340 	}
341 
342 	return 0;
343 }
344 
345 static int check_csum_item(struct extent_buffer *leaf, struct btrfs_key *key,
346 			   int slot, struct btrfs_key *prev_key)
347 {
348 	struct btrfs_fs_info *fs_info = leaf->fs_info;
349 	u32 sectorsize = fs_info->sectorsize;
350 	const u32 csumsize = fs_info->csum_size;
351 
352 	if (unlikely(key->objectid != BTRFS_EXTENT_CSUM_OBJECTID)) {
353 		generic_err(leaf, slot,
354 		"invalid key objectid for csum item, have %llu expect %llu",
355 			key->objectid, BTRFS_EXTENT_CSUM_OBJECTID);
356 		return -EUCLEAN;
357 	}
358 	if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) {
359 		generic_err(leaf, slot,
360 	"unaligned key offset for csum item, have %llu should be aligned to %u",
361 			key->offset, sectorsize);
362 		return -EUCLEAN;
363 	}
364 	if (unlikely(!IS_ALIGNED(btrfs_item_size(leaf, slot), csumsize))) {
365 		generic_err(leaf, slot,
366 	"unaligned item size for csum item, have %u should be aligned to %u",
367 			btrfs_item_size(leaf, slot), csumsize);
368 		return -EUCLEAN;
369 	}
370 	if (slot > 0 && prev_key->type == BTRFS_EXTENT_CSUM_KEY) {
371 		u64 prev_csum_end;
372 		u32 prev_item_size;
373 
374 		prev_item_size = btrfs_item_size(leaf, slot - 1);
375 		prev_csum_end = (prev_item_size / csumsize) * sectorsize;
376 		prev_csum_end += prev_key->offset;
377 		if (unlikely(prev_csum_end > key->offset)) {
378 			generic_err(leaf, slot - 1,
379 "csum end range (%llu) goes beyond the start range (%llu) of the next csum item",
380 				    prev_csum_end, key->offset);
381 			return -EUCLEAN;
382 		}
383 	}
384 	return 0;
385 }
386 
387 /* Inode item error output has the same format as dir_item_err() */
388 #define inode_item_err(eb, slot, fmt, ...)			\
389 	dir_item_err(eb, slot, fmt, __VA_ARGS__)
390 
391 static int check_inode_key(struct extent_buffer *leaf, struct btrfs_key *key,
392 			   int slot)
393 {
394 	struct btrfs_key item_key;
395 	bool is_inode_item;
396 
397 	btrfs_item_key_to_cpu(leaf, &item_key, slot);
398 	is_inode_item = (item_key.type == BTRFS_INODE_ITEM_KEY);
399 
400 	/* For XATTR_ITEM, location key should be all 0 */
401 	if (item_key.type == BTRFS_XATTR_ITEM_KEY) {
402 		if (unlikely(key->objectid != 0 || key->type != 0 ||
403 			     key->offset != 0))
404 			return -EUCLEAN;
405 		return 0;
406 	}
407 
408 	if (unlikely((key->objectid < BTRFS_FIRST_FREE_OBJECTID ||
409 		      key->objectid > BTRFS_LAST_FREE_OBJECTID) &&
410 		     key->objectid != BTRFS_ROOT_TREE_DIR_OBJECTID &&
411 		     key->objectid != BTRFS_FREE_INO_OBJECTID)) {
412 		if (is_inode_item) {
413 			generic_err(leaf, slot,
414 	"invalid 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 		} else {
420 			dir_item_err(leaf, slot,
421 "invalid location key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
422 				key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
423 				BTRFS_FIRST_FREE_OBJECTID,
424 				BTRFS_LAST_FREE_OBJECTID,
425 				BTRFS_FREE_INO_OBJECTID);
426 		}
427 		return -EUCLEAN;
428 	}
429 	if (unlikely(key->offset != 0)) {
430 		if (is_inode_item)
431 			inode_item_err(leaf, slot,
432 				       "invalid key offset: has %llu expect 0",
433 				       key->offset);
434 		else
435 			dir_item_err(leaf, slot,
436 				"invalid location key offset:has %llu expect 0",
437 				key->offset);
438 		return -EUCLEAN;
439 	}
440 	return 0;
441 }
442 
443 static int check_root_key(struct extent_buffer *leaf, struct btrfs_key *key,
444 			  int slot)
445 {
446 	struct btrfs_key item_key;
447 	bool is_root_item;
448 
449 	btrfs_item_key_to_cpu(leaf, &item_key, slot);
450 	is_root_item = (item_key.type == BTRFS_ROOT_ITEM_KEY);
451 
452 	/*
453 	 * Bad rootid for reloc trees.
454 	 *
455 	 * Reloc trees are only for subvolume trees, other trees only need
456 	 * to be COWed to be relocated.
457 	 */
458 	if (unlikely(is_root_item && key->objectid == BTRFS_TREE_RELOC_OBJECTID &&
459 		     !is_fstree(key->offset))) {
460 		generic_err(leaf, slot,
461 		"invalid reloc tree for root %lld, root id is not a subvolume tree",
462 			    key->offset);
463 		return -EUCLEAN;
464 	}
465 
466 	/* No such tree id */
467 	if (unlikely(key->objectid == 0)) {
468 		if (is_root_item)
469 			generic_err(leaf, slot, "invalid root id 0");
470 		else
471 			dir_item_err(leaf, slot,
472 				     "invalid location key root id 0");
473 		return -EUCLEAN;
474 	}
475 
476 	/* DIR_ITEM/INDEX/INODE_REF is not allowed to point to non-fs trees */
477 	if (unlikely(!is_fstree(key->objectid) && !is_root_item)) {
478 		dir_item_err(leaf, slot,
479 		"invalid location key objectid, have %llu expect [%llu, %llu]",
480 				key->objectid, BTRFS_FIRST_FREE_OBJECTID,
481 				BTRFS_LAST_FREE_OBJECTID);
482 		return -EUCLEAN;
483 	}
484 
485 	/*
486 	 * ROOT_ITEM with non-zero offset means this is a snapshot, created at
487 	 * @offset transid.
488 	 * Furthermore, for location key in DIR_ITEM, its offset is always -1.
489 	 *
490 	 * So here we only check offset for reloc tree whose key->offset must
491 	 * be a valid tree.
492 	 */
493 	if (unlikely(key->objectid == BTRFS_TREE_RELOC_OBJECTID &&
494 		     key->offset == 0)) {
495 		generic_err(leaf, slot, "invalid root id 0 for reloc tree");
496 		return -EUCLEAN;
497 	}
498 	return 0;
499 }
500 
501 static int check_dir_item(struct extent_buffer *leaf,
502 			  struct btrfs_key *key, struct btrfs_key *prev_key,
503 			  int slot)
504 {
505 	struct btrfs_fs_info *fs_info = leaf->fs_info;
506 	struct btrfs_dir_item *di;
507 	u32 item_size = btrfs_item_size(leaf, slot);
508 	u32 cur = 0;
509 
510 	if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
511 		return -EUCLEAN;
512 
513 	di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
514 	while (cur < item_size) {
515 		struct btrfs_key location_key;
516 		u32 name_len;
517 		u32 data_len;
518 		u32 max_name_len;
519 		u32 total_size;
520 		u32 name_hash;
521 		u8 dir_type;
522 		int ret;
523 
524 		/* header itself should not cross item boundary */
525 		if (unlikely(cur + sizeof(*di) > item_size)) {
526 			dir_item_err(leaf, slot,
527 		"dir item header crosses item boundary, have %zu boundary %u",
528 				cur + sizeof(*di), item_size);
529 			return -EUCLEAN;
530 		}
531 
532 		/* Location key check */
533 		btrfs_dir_item_key_to_cpu(leaf, di, &location_key);
534 		if (location_key.type == BTRFS_ROOT_ITEM_KEY) {
535 			ret = check_root_key(leaf, &location_key, slot);
536 			if (unlikely(ret < 0))
537 				return ret;
538 		} else if (location_key.type == BTRFS_INODE_ITEM_KEY ||
539 			   location_key.type == 0) {
540 			ret = check_inode_key(leaf, &location_key, slot);
541 			if (unlikely(ret < 0))
542 				return ret;
543 		} else {
544 			dir_item_err(leaf, slot,
545 			"invalid location key type, have %u, expect %u or %u",
546 				     location_key.type, BTRFS_ROOT_ITEM_KEY,
547 				     BTRFS_INODE_ITEM_KEY);
548 			return -EUCLEAN;
549 		}
550 
551 		/* dir type check */
552 		dir_type = btrfs_dir_ftype(leaf, di);
553 		if (unlikely(dir_type >= BTRFS_FT_MAX)) {
554 			dir_item_err(leaf, slot,
555 			"invalid dir item type, have %u expect [0, %u)",
556 				dir_type, BTRFS_FT_MAX);
557 			return -EUCLEAN;
558 		}
559 
560 		if (unlikely(key->type == BTRFS_XATTR_ITEM_KEY &&
561 			     dir_type != BTRFS_FT_XATTR)) {
562 			dir_item_err(leaf, slot,
563 		"invalid dir item type for XATTR key, have %u expect %u",
564 				dir_type, BTRFS_FT_XATTR);
565 			return -EUCLEAN;
566 		}
567 		if (unlikely(dir_type == BTRFS_FT_XATTR &&
568 			     key->type != BTRFS_XATTR_ITEM_KEY)) {
569 			dir_item_err(leaf, slot,
570 			"xattr dir type found for non-XATTR key");
571 			return -EUCLEAN;
572 		}
573 		if (dir_type == BTRFS_FT_XATTR)
574 			max_name_len = XATTR_NAME_MAX;
575 		else
576 			max_name_len = BTRFS_NAME_LEN;
577 
578 		/* Name/data length check */
579 		name_len = btrfs_dir_name_len(leaf, di);
580 		data_len = btrfs_dir_data_len(leaf, di);
581 		if (unlikely(name_len > max_name_len)) {
582 			dir_item_err(leaf, slot,
583 			"dir item name len too long, have %u max %u",
584 				name_len, max_name_len);
585 			return -EUCLEAN;
586 		}
587 		if (unlikely(name_len + data_len > BTRFS_MAX_XATTR_SIZE(fs_info))) {
588 			dir_item_err(leaf, slot,
589 			"dir item name and data len too long, have %u max %u",
590 				name_len + data_len,
591 				BTRFS_MAX_XATTR_SIZE(fs_info));
592 			return -EUCLEAN;
593 		}
594 
595 		if (unlikely(data_len && dir_type != BTRFS_FT_XATTR)) {
596 			dir_item_err(leaf, slot,
597 			"dir item with invalid data len, have %u expect 0",
598 				data_len);
599 			return -EUCLEAN;
600 		}
601 
602 		total_size = sizeof(*di) + name_len + data_len;
603 
604 		/* header and name/data should not cross item boundary */
605 		if (unlikely(cur + total_size > item_size)) {
606 			dir_item_err(leaf, slot,
607 		"dir item data crosses item boundary, have %u boundary %u",
608 				cur + total_size, item_size);
609 			return -EUCLEAN;
610 		}
611 
612 		/*
613 		 * Special check for XATTR/DIR_ITEM, as key->offset is name
614 		 * hash, should match its name
615 		 */
616 		if (key->type == BTRFS_DIR_ITEM_KEY ||
617 		    key->type == BTRFS_XATTR_ITEM_KEY) {
618 			char namebuf[max(BTRFS_NAME_LEN, XATTR_NAME_MAX)];
619 
620 			read_extent_buffer(leaf, namebuf,
621 					(unsigned long)(di + 1), name_len);
622 			name_hash = btrfs_name_hash(namebuf, name_len);
623 			if (unlikely(key->offset != name_hash)) {
624 				dir_item_err(leaf, slot,
625 		"name hash mismatch with key, have 0x%016x expect 0x%016llx",
626 					name_hash, key->offset);
627 				return -EUCLEAN;
628 			}
629 		}
630 		cur += total_size;
631 		di = (struct btrfs_dir_item *)((void *)di + total_size);
632 	}
633 	return 0;
634 }
635 
636 __printf(3, 4)
637 __cold
638 static void block_group_err(const struct extent_buffer *eb, int slot,
639 			    const char *fmt, ...)
640 {
641 	const struct btrfs_fs_info *fs_info = eb->fs_info;
642 	struct btrfs_key key;
643 	struct va_format vaf;
644 	va_list args;
645 
646 	btrfs_item_key_to_cpu(eb, &key, slot);
647 	va_start(args, fmt);
648 
649 	vaf.fmt = fmt;
650 	vaf.va = &args;
651 
652 	btrfs_crit(fs_info,
653 	"corrupt %s: root=%llu block=%llu slot=%d bg_start=%llu bg_len=%llu, %pV",
654 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
655 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
656 		key.objectid, key.offset, &vaf);
657 	va_end(args);
658 }
659 
660 static int check_block_group_item(struct extent_buffer *leaf,
661 				  struct btrfs_key *key, int slot)
662 {
663 	struct btrfs_fs_info *fs_info = leaf->fs_info;
664 	struct btrfs_block_group_item bgi;
665 	u32 item_size = btrfs_item_size(leaf, slot);
666 	u64 chunk_objectid;
667 	u64 flags;
668 	u64 type;
669 
670 	/*
671 	 * Here we don't really care about alignment since extent allocator can
672 	 * handle it.  We care more about the size.
673 	 */
674 	if (unlikely(key->offset == 0)) {
675 		block_group_err(leaf, slot,
676 				"invalid block group size 0");
677 		return -EUCLEAN;
678 	}
679 
680 	if (unlikely(item_size != sizeof(bgi))) {
681 		block_group_err(leaf, slot,
682 			"invalid item size, have %u expect %zu",
683 				item_size, sizeof(bgi));
684 		return -EUCLEAN;
685 	}
686 
687 	read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot),
688 			   sizeof(bgi));
689 	chunk_objectid = btrfs_stack_block_group_chunk_objectid(&bgi);
690 	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
691 		/*
692 		 * We don't init the nr_global_roots until we load the global
693 		 * roots, so this could be 0 at mount time.  If it's 0 we'll
694 		 * just assume we're fine, and later we'll check against our
695 		 * actual value.
696 		 */
697 		if (unlikely(fs_info->nr_global_roots &&
698 			     chunk_objectid >= fs_info->nr_global_roots)) {
699 			block_group_err(leaf, slot,
700 	"invalid block group global root id, have %llu, needs to be <= %llu",
701 					chunk_objectid,
702 					fs_info->nr_global_roots);
703 			return -EUCLEAN;
704 		}
705 	} else if (unlikely(chunk_objectid != BTRFS_FIRST_CHUNK_TREE_OBJECTID)) {
706 		block_group_err(leaf, slot,
707 		"invalid block group chunk objectid, have %llu expect %llu",
708 				btrfs_stack_block_group_chunk_objectid(&bgi),
709 				BTRFS_FIRST_CHUNK_TREE_OBJECTID);
710 		return -EUCLEAN;
711 	}
712 
713 	if (unlikely(btrfs_stack_block_group_used(&bgi) > key->offset)) {
714 		block_group_err(leaf, slot,
715 			"invalid block group used, have %llu expect [0, %llu)",
716 				btrfs_stack_block_group_used(&bgi), key->offset);
717 		return -EUCLEAN;
718 	}
719 
720 	flags = btrfs_stack_block_group_flags(&bgi);
721 	if (unlikely(hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) > 1)) {
722 		block_group_err(leaf, slot,
723 "invalid profile flags, have 0x%llx (%lu bits set) expect no more than 1 bit set",
724 			flags & BTRFS_BLOCK_GROUP_PROFILE_MASK,
725 			hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK));
726 		return -EUCLEAN;
727 	}
728 
729 	type = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
730 	if (unlikely(type != BTRFS_BLOCK_GROUP_DATA &&
731 		     type != BTRFS_BLOCK_GROUP_METADATA &&
732 		     type != BTRFS_BLOCK_GROUP_SYSTEM &&
733 		     type != (BTRFS_BLOCK_GROUP_METADATA |
734 			      BTRFS_BLOCK_GROUP_DATA))) {
735 		block_group_err(leaf, slot,
736 "invalid type, have 0x%llx (%lu bits set) expect either 0x%llx, 0x%llx, 0x%llx or 0x%llx",
737 			type, hweight64(type),
738 			BTRFS_BLOCK_GROUP_DATA, BTRFS_BLOCK_GROUP_METADATA,
739 			BTRFS_BLOCK_GROUP_SYSTEM,
740 			BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA);
741 		return -EUCLEAN;
742 	}
743 	return 0;
744 }
745 
746 __printf(4, 5)
747 __cold
748 static void chunk_err(const struct extent_buffer *leaf,
749 		      const struct btrfs_chunk *chunk, u64 logical,
750 		      const char *fmt, ...)
751 {
752 	const struct btrfs_fs_info *fs_info = leaf->fs_info;
753 	bool is_sb;
754 	struct va_format vaf;
755 	va_list args;
756 	int i;
757 	int slot = -1;
758 
759 	/* Only superblock eb is able to have such small offset */
760 	is_sb = (leaf->start == BTRFS_SUPER_INFO_OFFSET);
761 
762 	if (!is_sb) {
763 		/*
764 		 * Get the slot number by iterating through all slots, this
765 		 * would provide better readability.
766 		 */
767 		for (i = 0; i < btrfs_header_nritems(leaf); i++) {
768 			if (btrfs_item_ptr_offset(leaf, i) ==
769 					(unsigned long)chunk) {
770 				slot = i;
771 				break;
772 			}
773 		}
774 	}
775 	va_start(args, fmt);
776 	vaf.fmt = fmt;
777 	vaf.va = &args;
778 
779 	if (is_sb)
780 		btrfs_crit(fs_info,
781 		"corrupt superblock syschunk array: chunk_start=%llu, %pV",
782 			   logical, &vaf);
783 	else
784 		btrfs_crit(fs_info,
785 	"corrupt leaf: root=%llu block=%llu slot=%d chunk_start=%llu, %pV",
786 			   BTRFS_CHUNK_TREE_OBJECTID, leaf->start, slot,
787 			   logical, &vaf);
788 	va_end(args);
789 }
790 
791 /*
792  * The common chunk check which could also work on super block sys chunk array.
793  *
794  * Return -EUCLEAN if anything is corrupted.
795  * Return 0 if everything is OK.
796  */
797 int btrfs_check_chunk_valid(struct extent_buffer *leaf,
798 			    struct btrfs_chunk *chunk, u64 logical)
799 {
800 	struct btrfs_fs_info *fs_info = leaf->fs_info;
801 	u64 length;
802 	u64 chunk_end;
803 	u64 stripe_len;
804 	u16 num_stripes;
805 	u16 sub_stripes;
806 	u64 type;
807 	u64 features;
808 	bool mixed = false;
809 	int raid_index;
810 	int nparity;
811 	int ncopies;
812 
813 	length = btrfs_chunk_length(leaf, chunk);
814 	stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
815 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
816 	sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
817 	type = btrfs_chunk_type(leaf, chunk);
818 	raid_index = btrfs_bg_flags_to_raid_index(type);
819 	ncopies = btrfs_raid_array[raid_index].ncopies;
820 	nparity = btrfs_raid_array[raid_index].nparity;
821 
822 	if (unlikely(!num_stripes)) {
823 		chunk_err(leaf, chunk, logical,
824 			  "invalid chunk num_stripes, have %u", num_stripes);
825 		return -EUCLEAN;
826 	}
827 	if (unlikely(num_stripes < ncopies)) {
828 		chunk_err(leaf, chunk, logical,
829 			  "invalid chunk num_stripes < ncopies, have %u < %d",
830 			  num_stripes, ncopies);
831 		return -EUCLEAN;
832 	}
833 	if (unlikely(nparity && num_stripes == nparity)) {
834 		chunk_err(leaf, chunk, logical,
835 			  "invalid chunk num_stripes == nparity, have %u == %d",
836 			  num_stripes, nparity);
837 		return -EUCLEAN;
838 	}
839 	if (unlikely(!IS_ALIGNED(logical, fs_info->sectorsize))) {
840 		chunk_err(leaf, chunk, logical,
841 		"invalid chunk logical, have %llu should aligned to %u",
842 			  logical, fs_info->sectorsize);
843 		return -EUCLEAN;
844 	}
845 	if (unlikely(btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize)) {
846 		chunk_err(leaf, chunk, logical,
847 			  "invalid chunk sectorsize, have %u expect %u",
848 			  btrfs_chunk_sector_size(leaf, chunk),
849 			  fs_info->sectorsize);
850 		return -EUCLEAN;
851 	}
852 	if (unlikely(!length || !IS_ALIGNED(length, fs_info->sectorsize))) {
853 		chunk_err(leaf, chunk, logical,
854 			  "invalid chunk length, have %llu", length);
855 		return -EUCLEAN;
856 	}
857 	if (unlikely(check_add_overflow(logical, length, &chunk_end))) {
858 		chunk_err(leaf, chunk, logical,
859 "invalid chunk logical start and length, have logical start %llu length %llu",
860 			  logical, length);
861 		return -EUCLEAN;
862 	}
863 	if (unlikely(!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN)) {
864 		chunk_err(leaf, chunk, logical,
865 			  "invalid chunk stripe length: %llu",
866 			  stripe_len);
867 		return -EUCLEAN;
868 	}
869 	/*
870 	 * We artificially limit the chunk size, so that the number of stripes
871 	 * inside a chunk can be fit into a U32.  The current limit (256G) is
872 	 * way too large for real world usage anyway, and it's also much larger
873 	 * than our existing limit (10G).
874 	 *
875 	 * Thus it should be a good way to catch obvious bitflips.
876 	 */
877 	if (unlikely(length >= btrfs_stripe_nr_to_offset(U32_MAX))) {
878 		chunk_err(leaf, chunk, logical,
879 			  "chunk length too large: have %llu limit %llu",
880 			  length, btrfs_stripe_nr_to_offset(U32_MAX));
881 		return -EUCLEAN;
882 	}
883 	if (unlikely(type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
884 			      BTRFS_BLOCK_GROUP_PROFILE_MASK))) {
885 		chunk_err(leaf, chunk, logical,
886 			  "unrecognized chunk type: 0x%llx",
887 			  ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
888 			    BTRFS_BLOCK_GROUP_PROFILE_MASK) &
889 			  btrfs_chunk_type(leaf, chunk));
890 		return -EUCLEAN;
891 	}
892 
893 	if (unlikely(!has_single_bit_set(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
894 		     (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) != 0)) {
895 		chunk_err(leaf, chunk, logical,
896 		"invalid chunk profile flag: 0x%llx, expect 0 or 1 bit set",
897 			  type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
898 		return -EUCLEAN;
899 	}
900 	if (unlikely((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0)) {
901 		chunk_err(leaf, chunk, logical,
902 	"missing chunk type flag, have 0x%llx one bit must be set in 0x%llx",
903 			  type, BTRFS_BLOCK_GROUP_TYPE_MASK);
904 		return -EUCLEAN;
905 	}
906 
907 	if (unlikely((type & BTRFS_BLOCK_GROUP_SYSTEM) &&
908 		     (type & (BTRFS_BLOCK_GROUP_METADATA |
909 			      BTRFS_BLOCK_GROUP_DATA)))) {
910 		chunk_err(leaf, chunk, logical,
911 			  "system chunk with data or metadata type: 0x%llx",
912 			  type);
913 		return -EUCLEAN;
914 	}
915 
916 	features = btrfs_super_incompat_flags(fs_info->super_copy);
917 	if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
918 		mixed = true;
919 
920 	if (!mixed) {
921 		if (unlikely((type & BTRFS_BLOCK_GROUP_METADATA) &&
922 			     (type & BTRFS_BLOCK_GROUP_DATA))) {
923 			chunk_err(leaf, chunk, logical,
924 			"mixed chunk type in non-mixed mode: 0x%llx", type);
925 			return -EUCLEAN;
926 		}
927 	}
928 
929 	if (unlikely((type & BTRFS_BLOCK_GROUP_RAID10 &&
930 		      sub_stripes != btrfs_raid_array[BTRFS_RAID_RAID10].sub_stripes) ||
931 		     (type & BTRFS_BLOCK_GROUP_RAID1 &&
932 		      num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1].devs_min) ||
933 		     (type & BTRFS_BLOCK_GROUP_RAID1C3 &&
934 		      num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1C3].devs_min) ||
935 		     (type & BTRFS_BLOCK_GROUP_RAID1C4 &&
936 		      num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1C4].devs_min) ||
937 		     (type & BTRFS_BLOCK_GROUP_RAID5 &&
938 		      num_stripes < btrfs_raid_array[BTRFS_RAID_RAID5].devs_min) ||
939 		     (type & BTRFS_BLOCK_GROUP_RAID6 &&
940 		      num_stripes < btrfs_raid_array[BTRFS_RAID_RAID6].devs_min) ||
941 		     (type & BTRFS_BLOCK_GROUP_DUP &&
942 		      num_stripes != btrfs_raid_array[BTRFS_RAID_DUP].dev_stripes) ||
943 		     ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
944 		      num_stripes != btrfs_raid_array[BTRFS_RAID_SINGLE].dev_stripes))) {
945 		chunk_err(leaf, chunk, logical,
946 			"invalid num_stripes:sub_stripes %u:%u for profile %llu",
947 			num_stripes, sub_stripes,
948 			type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
949 		return -EUCLEAN;
950 	}
951 
952 	return 0;
953 }
954 
955 /*
956  * Enhanced version of chunk item checker.
957  *
958  * The common btrfs_check_chunk_valid() doesn't check item size since it needs
959  * to work on super block sys_chunk_array which doesn't have full item ptr.
960  */
961 static int check_leaf_chunk_item(struct extent_buffer *leaf,
962 				 struct btrfs_chunk *chunk,
963 				 struct btrfs_key *key, int slot)
964 {
965 	int num_stripes;
966 
967 	if (unlikely(btrfs_item_size(leaf, slot) < sizeof(struct btrfs_chunk))) {
968 		chunk_err(leaf, chunk, key->offset,
969 			"invalid chunk item size: have %u expect [%zu, %u)",
970 			btrfs_item_size(leaf, slot),
971 			sizeof(struct btrfs_chunk),
972 			BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
973 		return -EUCLEAN;
974 	}
975 
976 	num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
977 	/* Let btrfs_check_chunk_valid() handle this error type */
978 	if (num_stripes == 0)
979 		goto out;
980 
981 	if (unlikely(btrfs_chunk_item_size(num_stripes) !=
982 		     btrfs_item_size(leaf, slot))) {
983 		chunk_err(leaf, chunk, key->offset,
984 			"invalid chunk item size: have %u expect %lu",
985 			btrfs_item_size(leaf, slot),
986 			btrfs_chunk_item_size(num_stripes));
987 		return -EUCLEAN;
988 	}
989 out:
990 	return btrfs_check_chunk_valid(leaf, chunk, key->offset);
991 }
992 
993 __printf(3, 4)
994 __cold
995 static void dev_item_err(const struct extent_buffer *eb, int slot,
996 			 const char *fmt, ...)
997 {
998 	struct btrfs_key key;
999 	struct va_format vaf;
1000 	va_list args;
1001 
1002 	btrfs_item_key_to_cpu(eb, &key, slot);
1003 	va_start(args, fmt);
1004 
1005 	vaf.fmt = fmt;
1006 	vaf.va = &args;
1007 
1008 	btrfs_crit(eb->fs_info,
1009 	"corrupt %s: root=%llu block=%llu slot=%d devid=%llu %pV",
1010 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
1011 		btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
1012 		key.objectid, &vaf);
1013 	va_end(args);
1014 }
1015 
1016 static int check_dev_item(struct extent_buffer *leaf,
1017 			  struct btrfs_key *key, int slot)
1018 {
1019 	struct btrfs_dev_item *ditem;
1020 	const u32 item_size = btrfs_item_size(leaf, slot);
1021 
1022 	if (unlikely(key->objectid != BTRFS_DEV_ITEMS_OBJECTID)) {
1023 		dev_item_err(leaf, slot,
1024 			     "invalid objectid: has=%llu expect=%llu",
1025 			     key->objectid, BTRFS_DEV_ITEMS_OBJECTID);
1026 		return -EUCLEAN;
1027 	}
1028 
1029 	if (unlikely(item_size != sizeof(*ditem))) {
1030 		dev_item_err(leaf, slot, "invalid item size: has %u expect %zu",
1031 			     item_size, sizeof(*ditem));
1032 		return -EUCLEAN;
1033 	}
1034 
1035 	ditem = btrfs_item_ptr(leaf, slot, struct btrfs_dev_item);
1036 	if (unlikely(btrfs_device_id(leaf, ditem) != key->offset)) {
1037 		dev_item_err(leaf, slot,
1038 			     "devid mismatch: key has=%llu item has=%llu",
1039 			     key->offset, btrfs_device_id(leaf, ditem));
1040 		return -EUCLEAN;
1041 	}
1042 
1043 	/*
1044 	 * For device total_bytes, we don't have reliable way to check it, as
1045 	 * it can be 0 for device removal. Device size check can only be done
1046 	 * by dev extents check.
1047 	 */
1048 	if (unlikely(btrfs_device_bytes_used(leaf, ditem) >
1049 		     btrfs_device_total_bytes(leaf, ditem))) {
1050 		dev_item_err(leaf, slot,
1051 			     "invalid bytes used: have %llu expect [0, %llu]",
1052 			     btrfs_device_bytes_used(leaf, ditem),
1053 			     btrfs_device_total_bytes(leaf, ditem));
1054 		return -EUCLEAN;
1055 	}
1056 	/*
1057 	 * Remaining members like io_align/type/gen/dev_group aren't really
1058 	 * utilized.  Skip them to make later usage of them easier.
1059 	 */
1060 	return 0;
1061 }
1062 
1063 static int check_inode_item(struct extent_buffer *leaf,
1064 			    struct btrfs_key *key, int slot)
1065 {
1066 	struct btrfs_fs_info *fs_info = leaf->fs_info;
1067 	struct btrfs_inode_item *iitem;
1068 	u64 super_gen = btrfs_super_generation(fs_info->super_copy);
1069 	u32 valid_mask = (S_IFMT | S_ISUID | S_ISGID | S_ISVTX | 0777);
1070 	const u32 item_size = btrfs_item_size(leaf, slot);
1071 	u32 mode;
1072 	int ret;
1073 	u32 flags;
1074 	u32 ro_flags;
1075 
1076 	ret = check_inode_key(leaf, key, slot);
1077 	if (unlikely(ret < 0))
1078 		return ret;
1079 
1080 	if (unlikely(item_size != sizeof(*iitem))) {
1081 		generic_err(leaf, slot, "invalid item size: has %u expect %zu",
1082 			    item_size, sizeof(*iitem));
1083 		return -EUCLEAN;
1084 	}
1085 
1086 	iitem = btrfs_item_ptr(leaf, slot, struct btrfs_inode_item);
1087 
1088 	/* Here we use super block generation + 1 to handle log tree */
1089 	if (unlikely(btrfs_inode_generation(leaf, iitem) > super_gen + 1)) {
1090 		inode_item_err(leaf, slot,
1091 			"invalid inode generation: has %llu expect (0, %llu]",
1092 			       btrfs_inode_generation(leaf, iitem),
1093 			       super_gen + 1);
1094 		return -EUCLEAN;
1095 	}
1096 	/* Note for ROOT_TREE_DIR_ITEM, mkfs could set its transid 0 */
1097 	if (unlikely(btrfs_inode_transid(leaf, iitem) > super_gen + 1)) {
1098 		inode_item_err(leaf, slot,
1099 			"invalid inode transid: has %llu expect [0, %llu]",
1100 			       btrfs_inode_transid(leaf, iitem), super_gen + 1);
1101 		return -EUCLEAN;
1102 	}
1103 
1104 	/*
1105 	 * For size and nbytes it's better not to be too strict, as for dir
1106 	 * item its size/nbytes can easily get wrong, but doesn't affect
1107 	 * anything in the fs. So here we skip the check.
1108 	 */
1109 	mode = btrfs_inode_mode(leaf, iitem);
1110 	if (unlikely(mode & ~valid_mask)) {
1111 		inode_item_err(leaf, slot,
1112 			       "unknown mode bit detected: 0x%x",
1113 			       mode & ~valid_mask);
1114 		return -EUCLEAN;
1115 	}
1116 
1117 	/*
1118 	 * S_IFMT is not bit mapped so we can't completely rely on
1119 	 * is_power_of_2/has_single_bit_set, but it can save us from checking
1120 	 * FIFO/CHR/DIR/REG.  Only needs to check BLK, LNK and SOCKS
1121 	 */
1122 	if (!has_single_bit_set(mode & S_IFMT)) {
1123 		if (unlikely(!S_ISLNK(mode) && !S_ISBLK(mode) && !S_ISSOCK(mode))) {
1124 			inode_item_err(leaf, slot,
1125 			"invalid mode: has 0%o expect valid S_IF* bit(s)",
1126 				       mode & S_IFMT);
1127 			return -EUCLEAN;
1128 		}
1129 	}
1130 	if (unlikely(S_ISDIR(mode) && btrfs_inode_nlink(leaf, iitem) > 1)) {
1131 		inode_item_err(leaf, slot,
1132 		       "invalid nlink: has %u expect no more than 1 for dir",
1133 			btrfs_inode_nlink(leaf, iitem));
1134 		return -EUCLEAN;
1135 	}
1136 	btrfs_inode_split_flags(btrfs_inode_flags(leaf, iitem), &flags, &ro_flags);
1137 	if (unlikely(flags & ~BTRFS_INODE_FLAG_MASK)) {
1138 		inode_item_err(leaf, slot,
1139 			       "unknown incompat flags detected: 0x%x", flags);
1140 		return -EUCLEAN;
1141 	}
1142 	if (unlikely(!sb_rdonly(fs_info->sb) &&
1143 		     (ro_flags & ~BTRFS_INODE_RO_FLAG_MASK))) {
1144 		inode_item_err(leaf, slot,
1145 			"unknown ro-compat flags detected on writeable mount: 0x%x",
1146 			ro_flags);
1147 		return -EUCLEAN;
1148 	}
1149 	return 0;
1150 }
1151 
1152 static int check_root_item(struct extent_buffer *leaf, struct btrfs_key *key,
1153 			   int slot)
1154 {
1155 	struct btrfs_fs_info *fs_info = leaf->fs_info;
1156 	struct btrfs_root_item ri = { 0 };
1157 	const u64 valid_root_flags = BTRFS_ROOT_SUBVOL_RDONLY |
1158 				     BTRFS_ROOT_SUBVOL_DEAD;
1159 	int ret;
1160 
1161 	ret = check_root_key(leaf, key, slot);
1162 	if (unlikely(ret < 0))
1163 		return ret;
1164 
1165 	if (unlikely(btrfs_item_size(leaf, slot) != sizeof(ri) &&
1166 		     btrfs_item_size(leaf, slot) !=
1167 		     btrfs_legacy_root_item_size())) {
1168 		generic_err(leaf, slot,
1169 			    "invalid root item size, have %u expect %zu or %u",
1170 			    btrfs_item_size(leaf, slot), sizeof(ri),
1171 			    btrfs_legacy_root_item_size());
1172 		return -EUCLEAN;
1173 	}
1174 
1175 	/*
1176 	 * For legacy root item, the members starting at generation_v2 will be
1177 	 * all filled with 0.
1178 	 * And since we allow geneartion_v2 as 0, it will still pass the check.
1179 	 */
1180 	read_extent_buffer(leaf, &ri, btrfs_item_ptr_offset(leaf, slot),
1181 			   btrfs_item_size(leaf, slot));
1182 
1183 	/* Generation related */
1184 	if (unlikely(btrfs_root_generation(&ri) >
1185 		     btrfs_super_generation(fs_info->super_copy) + 1)) {
1186 		generic_err(leaf, slot,
1187 			"invalid root generation, have %llu expect (0, %llu]",
1188 			    btrfs_root_generation(&ri),
1189 			    btrfs_super_generation(fs_info->super_copy) + 1);
1190 		return -EUCLEAN;
1191 	}
1192 	if (unlikely(btrfs_root_generation_v2(&ri) >
1193 		     btrfs_super_generation(fs_info->super_copy) + 1)) {
1194 		generic_err(leaf, slot,
1195 		"invalid root v2 generation, have %llu expect (0, %llu]",
1196 			    btrfs_root_generation_v2(&ri),
1197 			    btrfs_super_generation(fs_info->super_copy) + 1);
1198 		return -EUCLEAN;
1199 	}
1200 	if (unlikely(btrfs_root_last_snapshot(&ri) >
1201 		     btrfs_super_generation(fs_info->super_copy) + 1)) {
1202 		generic_err(leaf, slot,
1203 		"invalid root last_snapshot, have %llu expect (0, %llu]",
1204 			    btrfs_root_last_snapshot(&ri),
1205 			    btrfs_super_generation(fs_info->super_copy) + 1);
1206 		return -EUCLEAN;
1207 	}
1208 
1209 	/* Alignment and level check */
1210 	if (unlikely(!IS_ALIGNED(btrfs_root_bytenr(&ri), fs_info->sectorsize))) {
1211 		generic_err(leaf, slot,
1212 		"invalid root bytenr, have %llu expect to be aligned to %u",
1213 			    btrfs_root_bytenr(&ri), fs_info->sectorsize);
1214 		return -EUCLEAN;
1215 	}
1216 	if (unlikely(btrfs_root_level(&ri) >= BTRFS_MAX_LEVEL)) {
1217 		generic_err(leaf, slot,
1218 			    "invalid root level, have %u expect [0, %u]",
1219 			    btrfs_root_level(&ri), BTRFS_MAX_LEVEL - 1);
1220 		return -EUCLEAN;
1221 	}
1222 	if (unlikely(btrfs_root_drop_level(&ri) >= BTRFS_MAX_LEVEL)) {
1223 		generic_err(leaf, slot,
1224 			    "invalid root level, have %u expect [0, %u]",
1225 			    btrfs_root_drop_level(&ri), BTRFS_MAX_LEVEL - 1);
1226 		return -EUCLEAN;
1227 	}
1228 
1229 	/* Flags check */
1230 	if (unlikely(btrfs_root_flags(&ri) & ~valid_root_flags)) {
1231 		generic_err(leaf, slot,
1232 			    "invalid root flags, have 0x%llx expect mask 0x%llx",
1233 			    btrfs_root_flags(&ri), valid_root_flags);
1234 		return -EUCLEAN;
1235 	}
1236 	return 0;
1237 }
1238 
1239 __printf(3,4)
1240 __cold
1241 static void extent_err(const struct extent_buffer *eb, int slot,
1242 		       const char *fmt, ...)
1243 {
1244 	struct btrfs_key key;
1245 	struct va_format vaf;
1246 	va_list args;
1247 	u64 bytenr;
1248 	u64 len;
1249 
1250 	btrfs_item_key_to_cpu(eb, &key, slot);
1251 	bytenr = key.objectid;
1252 	if (key.type == BTRFS_METADATA_ITEM_KEY ||
1253 	    key.type == BTRFS_TREE_BLOCK_REF_KEY ||
1254 	    key.type == BTRFS_SHARED_BLOCK_REF_KEY)
1255 		len = eb->fs_info->nodesize;
1256 	else
1257 		len = key.offset;
1258 	va_start(args, fmt);
1259 
1260 	vaf.fmt = fmt;
1261 	vaf.va = &args;
1262 
1263 	btrfs_crit(eb->fs_info,
1264 	"corrupt %s: block=%llu slot=%d extent bytenr=%llu len=%llu %pV",
1265 		btrfs_header_level(eb) == 0 ? "leaf" : "node",
1266 		eb->start, slot, bytenr, len, &vaf);
1267 	va_end(args);
1268 }
1269 
1270 static int check_extent_item(struct extent_buffer *leaf,
1271 			     struct btrfs_key *key, int slot,
1272 			     struct btrfs_key *prev_key)
1273 {
1274 	struct btrfs_fs_info *fs_info = leaf->fs_info;
1275 	struct btrfs_extent_item *ei;
1276 	bool is_tree_block = false;
1277 	unsigned long ptr;	/* Current pointer inside inline refs */
1278 	unsigned long end;	/* Extent item end */
1279 	const u32 item_size = btrfs_item_size(leaf, slot);
1280 	u8 last_type = 0;
1281 	u64 last_seq = U64_MAX;
1282 	u64 flags;
1283 	u64 generation;
1284 	u64 total_refs;		/* Total refs in btrfs_extent_item */
1285 	u64 inline_refs = 0;	/* found total inline refs */
1286 
1287 	if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1288 		     !btrfs_fs_incompat(fs_info, SKINNY_METADATA))) {
1289 		generic_err(leaf, slot,
1290 "invalid key type, METADATA_ITEM type invalid when SKINNY_METADATA feature disabled");
1291 		return -EUCLEAN;
1292 	}
1293 	/* key->objectid is the bytenr for both key types */
1294 	if (unlikely(!IS_ALIGNED(key->objectid, fs_info->sectorsize))) {
1295 		generic_err(leaf, slot,
1296 		"invalid key objectid, have %llu expect to be aligned to %u",
1297 			   key->objectid, fs_info->sectorsize);
1298 		return -EUCLEAN;
1299 	}
1300 
1301 	/* key->offset is tree level for METADATA_ITEM_KEY */
1302 	if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1303 		     key->offset >= BTRFS_MAX_LEVEL)) {
1304 		extent_err(leaf, slot,
1305 			   "invalid tree level, have %llu expect [0, %u]",
1306 			   key->offset, BTRFS_MAX_LEVEL - 1);
1307 		return -EUCLEAN;
1308 	}
1309 
1310 	/*
1311 	 * EXTENT/METADATA_ITEM consists of:
1312 	 * 1) One btrfs_extent_item
1313 	 *    Records the total refs, type and generation of the extent.
1314 	 *
1315 	 * 2) One btrfs_tree_block_info (for EXTENT_ITEM and tree backref only)
1316 	 *    Records the first key and level of the tree block.
1317 	 *
1318 	 * 2) Zero or more btrfs_extent_inline_ref(s)
1319 	 *    Each inline ref has one btrfs_extent_inline_ref shows:
1320 	 *    2.1) The ref type, one of the 4
1321 	 *         TREE_BLOCK_REF	Tree block only
1322 	 *         SHARED_BLOCK_REF	Tree block only
1323 	 *         EXTENT_DATA_REF	Data only
1324 	 *         SHARED_DATA_REF	Data only
1325 	 *    2.2) Ref type specific data
1326 	 *         Either using btrfs_extent_inline_ref::offset, or specific
1327 	 *         data structure.
1328 	 *
1329 	 *    All above inline items should follow the order:
1330 	 *
1331 	 *    - All btrfs_extent_inline_ref::type should be in an ascending
1332 	 *      order
1333 	 *
1334 	 *    - Within the same type, the items should follow a descending
1335 	 *      order by their sequence number. The sequence number is
1336 	 *      determined by:
1337 	 *      * btrfs_extent_inline_ref::offset for all types  other than
1338 	 *        EXTENT_DATA_REF
1339 	 *      * hash_extent_data_ref() for EXTENT_DATA_REF
1340 	 */
1341 	if (unlikely(item_size < sizeof(*ei))) {
1342 		extent_err(leaf, slot,
1343 			   "invalid item size, have %u expect [%zu, %u)",
1344 			   item_size, sizeof(*ei),
1345 			   BTRFS_LEAF_DATA_SIZE(fs_info));
1346 		return -EUCLEAN;
1347 	}
1348 	end = item_size + btrfs_item_ptr_offset(leaf, slot);
1349 
1350 	/* Checks against extent_item */
1351 	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
1352 	flags = btrfs_extent_flags(leaf, ei);
1353 	total_refs = btrfs_extent_refs(leaf, ei);
1354 	generation = btrfs_extent_generation(leaf, ei);
1355 	if (unlikely(generation >
1356 		     btrfs_super_generation(fs_info->super_copy) + 1)) {
1357 		extent_err(leaf, slot,
1358 			   "invalid generation, have %llu expect (0, %llu]",
1359 			   generation,
1360 			   btrfs_super_generation(fs_info->super_copy) + 1);
1361 		return -EUCLEAN;
1362 	}
1363 	if (unlikely(!has_single_bit_set(flags & (BTRFS_EXTENT_FLAG_DATA |
1364 						  BTRFS_EXTENT_FLAG_TREE_BLOCK)))) {
1365 		extent_err(leaf, slot,
1366 		"invalid extent flag, have 0x%llx expect 1 bit set in 0x%llx",
1367 			flags, BTRFS_EXTENT_FLAG_DATA |
1368 			BTRFS_EXTENT_FLAG_TREE_BLOCK);
1369 		return -EUCLEAN;
1370 	}
1371 	is_tree_block = !!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK);
1372 	if (is_tree_block) {
1373 		if (unlikely(key->type == BTRFS_EXTENT_ITEM_KEY &&
1374 			     key->offset != fs_info->nodesize)) {
1375 			extent_err(leaf, slot,
1376 				   "invalid extent length, have %llu expect %u",
1377 				   key->offset, fs_info->nodesize);
1378 			return -EUCLEAN;
1379 		}
1380 	} else {
1381 		if (unlikely(key->type != BTRFS_EXTENT_ITEM_KEY)) {
1382 			extent_err(leaf, slot,
1383 			"invalid key type, have %u expect %u for data backref",
1384 				   key->type, BTRFS_EXTENT_ITEM_KEY);
1385 			return -EUCLEAN;
1386 		}
1387 		if (unlikely(!IS_ALIGNED(key->offset, fs_info->sectorsize))) {
1388 			extent_err(leaf, slot,
1389 			"invalid extent length, have %llu expect aligned to %u",
1390 				   key->offset, fs_info->sectorsize);
1391 			return -EUCLEAN;
1392 		}
1393 		if (unlikely(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
1394 			extent_err(leaf, slot,
1395 			"invalid extent flag, data has full backref set");
1396 			return -EUCLEAN;
1397 		}
1398 	}
1399 	ptr = (unsigned long)(struct btrfs_extent_item *)(ei + 1);
1400 
1401 	/* Check the special case of btrfs_tree_block_info */
1402 	if (is_tree_block && key->type != BTRFS_METADATA_ITEM_KEY) {
1403 		struct btrfs_tree_block_info *info;
1404 
1405 		info = (struct btrfs_tree_block_info *)ptr;
1406 		if (unlikely(btrfs_tree_block_level(leaf, info) >= BTRFS_MAX_LEVEL)) {
1407 			extent_err(leaf, slot,
1408 			"invalid tree block info level, have %u expect [0, %u]",
1409 				   btrfs_tree_block_level(leaf, info),
1410 				   BTRFS_MAX_LEVEL - 1);
1411 			return -EUCLEAN;
1412 		}
1413 		ptr = (unsigned long)(struct btrfs_tree_block_info *)(info + 1);
1414 	}
1415 
1416 	/* Check inline refs */
1417 	while (ptr < end) {
1418 		struct btrfs_extent_inline_ref *iref;
1419 		struct btrfs_extent_data_ref *dref;
1420 		struct btrfs_shared_data_ref *sref;
1421 		u64 seq;
1422 		u64 dref_offset;
1423 		u64 inline_offset;
1424 		u8 inline_type;
1425 
1426 		if (unlikely(ptr + sizeof(*iref) > end)) {
1427 			extent_err(leaf, slot,
1428 "inline ref item overflows extent item, ptr %lu iref size %zu end %lu",
1429 				   ptr, sizeof(*iref), end);
1430 			return -EUCLEAN;
1431 		}
1432 		iref = (struct btrfs_extent_inline_ref *)ptr;
1433 		inline_type = btrfs_extent_inline_ref_type(leaf, iref);
1434 		inline_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1435 		seq = inline_offset;
1436 		if (unlikely(ptr + btrfs_extent_inline_ref_size(inline_type) > end)) {
1437 			extent_err(leaf, slot,
1438 "inline ref item overflows extent item, ptr %lu iref size %u end %lu",
1439 				   ptr, btrfs_extent_inline_ref_size(inline_type), end);
1440 			return -EUCLEAN;
1441 		}
1442 
1443 		switch (inline_type) {
1444 		/* inline_offset is subvolid of the owner, no need to check */
1445 		case BTRFS_TREE_BLOCK_REF_KEY:
1446 			inline_refs++;
1447 			break;
1448 		/* Contains parent bytenr */
1449 		case BTRFS_SHARED_BLOCK_REF_KEY:
1450 			if (unlikely(!IS_ALIGNED(inline_offset,
1451 						 fs_info->sectorsize))) {
1452 				extent_err(leaf, slot,
1453 		"invalid tree parent bytenr, have %llu expect aligned to %u",
1454 					   inline_offset, fs_info->sectorsize);
1455 				return -EUCLEAN;
1456 			}
1457 			inline_refs++;
1458 			break;
1459 		/*
1460 		 * Contains owner subvolid, owner key objectid, adjusted offset.
1461 		 * The only obvious corruption can happen in that offset.
1462 		 */
1463 		case BTRFS_EXTENT_DATA_REF_KEY:
1464 			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1465 			dref_offset = btrfs_extent_data_ref_offset(leaf, dref);
1466 			seq = hash_extent_data_ref(
1467 					btrfs_extent_data_ref_root(leaf, dref),
1468 					btrfs_extent_data_ref_objectid(leaf, dref),
1469 					btrfs_extent_data_ref_offset(leaf, dref));
1470 			if (unlikely(!IS_ALIGNED(dref_offset,
1471 						 fs_info->sectorsize))) {
1472 				extent_err(leaf, slot,
1473 		"invalid data ref offset, have %llu expect aligned to %u",
1474 					   dref_offset, fs_info->sectorsize);
1475 				return -EUCLEAN;
1476 			}
1477 			inline_refs += btrfs_extent_data_ref_count(leaf, dref);
1478 			break;
1479 		/* Contains parent bytenr and ref count */
1480 		case BTRFS_SHARED_DATA_REF_KEY:
1481 			sref = (struct btrfs_shared_data_ref *)(iref + 1);
1482 			if (unlikely(!IS_ALIGNED(inline_offset,
1483 						 fs_info->sectorsize))) {
1484 				extent_err(leaf, slot,
1485 		"invalid data parent bytenr, have %llu expect aligned to %u",
1486 					   inline_offset, fs_info->sectorsize);
1487 				return -EUCLEAN;
1488 			}
1489 			inline_refs += btrfs_shared_data_ref_count(leaf, sref);
1490 			break;
1491 		case BTRFS_EXTENT_OWNER_REF_KEY:
1492 			WARN_ON(!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));
1493 			break;
1494 		default:
1495 			extent_err(leaf, slot, "unknown inline ref type: %u",
1496 				   inline_type);
1497 			return -EUCLEAN;
1498 		}
1499 		if (inline_type < last_type) {
1500 			extent_err(leaf, slot,
1501 				   "inline ref out-of-order: has type %u, prev type %u",
1502 				   inline_type, last_type);
1503 			return -EUCLEAN;
1504 		}
1505 		/* Type changed, allow the sequence starts from U64_MAX again. */
1506 		if (inline_type > last_type)
1507 			last_seq = U64_MAX;
1508 		if (seq > last_seq) {
1509 			extent_err(leaf, slot,
1510 "inline ref out-of-order: has type %u offset %llu seq 0x%llx, prev type %u seq 0x%llx",
1511 				   inline_type, inline_offset, seq,
1512 				   last_type, last_seq);
1513 			return -EUCLEAN;
1514 		}
1515 		last_type = inline_type;
1516 		last_seq = seq;
1517 		ptr += btrfs_extent_inline_ref_size(inline_type);
1518 	}
1519 	/* No padding is allowed */
1520 	if (unlikely(ptr != end)) {
1521 		extent_err(leaf, slot,
1522 			   "invalid extent item size, padding bytes found");
1523 		return -EUCLEAN;
1524 	}
1525 
1526 	/* Finally, check the inline refs against total refs */
1527 	if (unlikely(inline_refs > total_refs)) {
1528 		extent_err(leaf, slot,
1529 			"invalid extent refs, have %llu expect >= inline %llu",
1530 			   total_refs, inline_refs);
1531 		return -EUCLEAN;
1532 	}
1533 
1534 	if ((prev_key->type == BTRFS_EXTENT_ITEM_KEY) ||
1535 	    (prev_key->type == BTRFS_METADATA_ITEM_KEY)) {
1536 		u64 prev_end = prev_key->objectid;
1537 
1538 		if (prev_key->type == BTRFS_METADATA_ITEM_KEY)
1539 			prev_end += fs_info->nodesize;
1540 		else
1541 			prev_end += prev_key->offset;
1542 
1543 		if (unlikely(prev_end > key->objectid)) {
1544 			extent_err(leaf, slot,
1545 	"previous extent [%llu %u %llu] overlaps current extent [%llu %u %llu]",
1546 				   prev_key->objectid, prev_key->type,
1547 				   prev_key->offset, key->objectid, key->type,
1548 				   key->offset);
1549 			return -EUCLEAN;
1550 		}
1551 	}
1552 
1553 	return 0;
1554 }
1555 
1556 static int check_simple_keyed_refs(struct extent_buffer *leaf,
1557 				   struct btrfs_key *key, int slot)
1558 {
1559 	u32 expect_item_size = 0;
1560 
1561 	if (key->type == BTRFS_SHARED_DATA_REF_KEY)
1562 		expect_item_size = sizeof(struct btrfs_shared_data_ref);
1563 
1564 	if (unlikely(btrfs_item_size(leaf, slot) != expect_item_size)) {
1565 		generic_err(leaf, slot,
1566 		"invalid item size, have %u expect %u for key type %u",
1567 			    btrfs_item_size(leaf, slot),
1568 			    expect_item_size, key->type);
1569 		return -EUCLEAN;
1570 	}
1571 	if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1572 		generic_err(leaf, slot,
1573 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1574 			    key->objectid, leaf->fs_info->sectorsize);
1575 		return -EUCLEAN;
1576 	}
1577 	if (unlikely(key->type != BTRFS_TREE_BLOCK_REF_KEY &&
1578 		     !IS_ALIGNED(key->offset, leaf->fs_info->sectorsize))) {
1579 		extent_err(leaf, slot,
1580 		"invalid tree parent bytenr, have %llu expect aligned to %u",
1581 			   key->offset, leaf->fs_info->sectorsize);
1582 		return -EUCLEAN;
1583 	}
1584 	return 0;
1585 }
1586 
1587 static int check_extent_data_ref(struct extent_buffer *leaf,
1588 				 struct btrfs_key *key, int slot)
1589 {
1590 	struct btrfs_extent_data_ref *dref;
1591 	unsigned long ptr = btrfs_item_ptr_offset(leaf, slot);
1592 	const unsigned long end = ptr + btrfs_item_size(leaf, slot);
1593 
1594 	if (unlikely(btrfs_item_size(leaf, slot) % sizeof(*dref) != 0)) {
1595 		generic_err(leaf, slot,
1596 	"invalid item size, have %u expect aligned to %zu for key type %u",
1597 			    btrfs_item_size(leaf, slot),
1598 			    sizeof(*dref), key->type);
1599 		return -EUCLEAN;
1600 	}
1601 	if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1602 		generic_err(leaf, slot,
1603 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1604 			    key->objectid, leaf->fs_info->sectorsize);
1605 		return -EUCLEAN;
1606 	}
1607 	for (; ptr < end; ptr += sizeof(*dref)) {
1608 		u64 offset;
1609 
1610 		/*
1611 		 * We cannot check the extent_data_ref hash due to possible
1612 		 * overflow from the leaf due to hash collisions.
1613 		 */
1614 		dref = (struct btrfs_extent_data_ref *)ptr;
1615 		offset = btrfs_extent_data_ref_offset(leaf, dref);
1616 		if (unlikely(!IS_ALIGNED(offset, leaf->fs_info->sectorsize))) {
1617 			extent_err(leaf, slot,
1618 	"invalid extent data backref offset, have %llu expect aligned to %u",
1619 				   offset, leaf->fs_info->sectorsize);
1620 			return -EUCLEAN;
1621 		}
1622 	}
1623 	return 0;
1624 }
1625 
1626 #define inode_ref_err(eb, slot, fmt, args...)			\
1627 	inode_item_err(eb, slot, fmt, ##args)
1628 static int check_inode_ref(struct extent_buffer *leaf,
1629 			   struct btrfs_key *key, struct btrfs_key *prev_key,
1630 			   int slot)
1631 {
1632 	struct btrfs_inode_ref *iref;
1633 	unsigned long ptr;
1634 	unsigned long end;
1635 
1636 	if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
1637 		return -EUCLEAN;
1638 	/* namelen can't be 0, so item_size == sizeof() is also invalid */
1639 	if (unlikely(btrfs_item_size(leaf, slot) <= sizeof(*iref))) {
1640 		inode_ref_err(leaf, slot,
1641 			"invalid item size, have %u expect (%zu, %u)",
1642 			btrfs_item_size(leaf, slot),
1643 			sizeof(*iref), BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
1644 		return -EUCLEAN;
1645 	}
1646 
1647 	ptr = btrfs_item_ptr_offset(leaf, slot);
1648 	end = ptr + btrfs_item_size(leaf, slot);
1649 	while (ptr < end) {
1650 		u16 namelen;
1651 
1652 		if (unlikely(ptr + sizeof(iref) > end)) {
1653 			inode_ref_err(leaf, slot,
1654 			"inode ref overflow, ptr %lu end %lu inode_ref_size %zu",
1655 				ptr, end, sizeof(iref));
1656 			return -EUCLEAN;
1657 		}
1658 
1659 		iref = (struct btrfs_inode_ref *)ptr;
1660 		namelen = btrfs_inode_ref_name_len(leaf, iref);
1661 		if (unlikely(ptr + sizeof(*iref) + namelen > end)) {
1662 			inode_ref_err(leaf, slot,
1663 				"inode ref overflow, ptr %lu end %lu namelen %u",
1664 				ptr, end, namelen);
1665 			return -EUCLEAN;
1666 		}
1667 
1668 		/*
1669 		 * NOTE: In theory we should record all found index numbers
1670 		 * to find any duplicated indexes, but that will be too time
1671 		 * consuming for inodes with too many hard links.
1672 		 */
1673 		ptr += sizeof(*iref) + namelen;
1674 	}
1675 	return 0;
1676 }
1677 
1678 static int check_raid_stripe_extent(const struct extent_buffer *leaf,
1679 				    const struct btrfs_key *key, int slot)
1680 {
1681 	struct btrfs_stripe_extent *stripe_extent =
1682 		btrfs_item_ptr(leaf, slot, struct btrfs_stripe_extent);
1683 
1684 	if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1685 		generic_err(leaf, slot,
1686 "invalid key objectid for raid stripe extent, have %llu expect aligned to %u",
1687 			    key->objectid, leaf->fs_info->sectorsize);
1688 		return -EUCLEAN;
1689 	}
1690 
1691 	if (unlikely(!btrfs_fs_incompat(leaf->fs_info, RAID_STRIPE_TREE))) {
1692 		generic_err(leaf, slot,
1693 	"RAID_STRIPE_EXTENT present but RAID_STRIPE_TREE incompat bit unset");
1694 		return -EUCLEAN;
1695 	}
1696 
1697 	switch (btrfs_stripe_extent_encoding(leaf, stripe_extent)) {
1698 	case BTRFS_STRIPE_RAID0:
1699 	case BTRFS_STRIPE_RAID1:
1700 	case BTRFS_STRIPE_DUP:
1701 	case BTRFS_STRIPE_RAID10:
1702 	case BTRFS_STRIPE_RAID5:
1703 	case BTRFS_STRIPE_RAID6:
1704 	case BTRFS_STRIPE_RAID1C3:
1705 	case BTRFS_STRIPE_RAID1C4:
1706 		break;
1707 	default:
1708 		generic_err(leaf, slot, "invalid raid stripe encoding %u",
1709 			    btrfs_stripe_extent_encoding(leaf, stripe_extent));
1710 		return -EUCLEAN;
1711 	}
1712 
1713 	return 0;
1714 }
1715 
1716 /*
1717  * Common point to switch the item-specific validation.
1718  */
1719 static enum btrfs_tree_block_status check_leaf_item(struct extent_buffer *leaf,
1720 						    struct btrfs_key *key,
1721 						    int slot,
1722 						    struct btrfs_key *prev_key)
1723 {
1724 	int ret = 0;
1725 	struct btrfs_chunk *chunk;
1726 
1727 	switch (key->type) {
1728 	case BTRFS_EXTENT_DATA_KEY:
1729 		ret = check_extent_data_item(leaf, key, slot, prev_key);
1730 		break;
1731 	case BTRFS_EXTENT_CSUM_KEY:
1732 		ret = check_csum_item(leaf, key, slot, prev_key);
1733 		break;
1734 	case BTRFS_DIR_ITEM_KEY:
1735 	case BTRFS_DIR_INDEX_KEY:
1736 	case BTRFS_XATTR_ITEM_KEY:
1737 		ret = check_dir_item(leaf, key, prev_key, slot);
1738 		break;
1739 	case BTRFS_INODE_REF_KEY:
1740 		ret = check_inode_ref(leaf, key, prev_key, slot);
1741 		break;
1742 	case BTRFS_BLOCK_GROUP_ITEM_KEY:
1743 		ret = check_block_group_item(leaf, key, slot);
1744 		break;
1745 	case BTRFS_CHUNK_ITEM_KEY:
1746 		chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1747 		ret = check_leaf_chunk_item(leaf, chunk, key, slot);
1748 		break;
1749 	case BTRFS_DEV_ITEM_KEY:
1750 		ret = check_dev_item(leaf, key, slot);
1751 		break;
1752 	case BTRFS_INODE_ITEM_KEY:
1753 		ret = check_inode_item(leaf, key, slot);
1754 		break;
1755 	case BTRFS_ROOT_ITEM_KEY:
1756 		ret = check_root_item(leaf, key, slot);
1757 		break;
1758 	case BTRFS_EXTENT_ITEM_KEY:
1759 	case BTRFS_METADATA_ITEM_KEY:
1760 		ret = check_extent_item(leaf, key, slot, prev_key);
1761 		break;
1762 	case BTRFS_TREE_BLOCK_REF_KEY:
1763 	case BTRFS_SHARED_DATA_REF_KEY:
1764 	case BTRFS_SHARED_BLOCK_REF_KEY:
1765 		ret = check_simple_keyed_refs(leaf, key, slot);
1766 		break;
1767 	case BTRFS_EXTENT_DATA_REF_KEY:
1768 		ret = check_extent_data_ref(leaf, key, slot);
1769 		break;
1770 	case BTRFS_RAID_STRIPE_KEY:
1771 		ret = check_raid_stripe_extent(leaf, key, slot);
1772 		break;
1773 	}
1774 
1775 	if (ret)
1776 		return BTRFS_TREE_BLOCK_INVALID_ITEM;
1777 	return BTRFS_TREE_BLOCK_CLEAN;
1778 }
1779 
1780 enum btrfs_tree_block_status __btrfs_check_leaf(struct extent_buffer *leaf)
1781 {
1782 	struct btrfs_fs_info *fs_info = leaf->fs_info;
1783 	/* No valid key type is 0, so all key should be larger than this key */
1784 	struct btrfs_key prev_key = {0, 0, 0};
1785 	struct btrfs_key key;
1786 	u32 nritems = btrfs_header_nritems(leaf);
1787 	int slot;
1788 
1789 	if (unlikely(btrfs_header_level(leaf) != 0)) {
1790 		generic_err(leaf, 0,
1791 			"invalid level for leaf, have %d expect 0",
1792 			btrfs_header_level(leaf));
1793 		return BTRFS_TREE_BLOCK_INVALID_LEVEL;
1794 	}
1795 
1796 	/*
1797 	 * Extent buffers from a relocation tree have a owner field that
1798 	 * corresponds to the subvolume tree they are based on. So just from an
1799 	 * extent buffer alone we can not find out what is the id of the
1800 	 * corresponding subvolume tree, so we can not figure out if the extent
1801 	 * buffer corresponds to the root of the relocation tree or not. So
1802 	 * skip this check for relocation trees.
1803 	 */
1804 	if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
1805 		u64 owner = btrfs_header_owner(leaf);
1806 
1807 		/* These trees must never be empty */
1808 		if (unlikely(owner == BTRFS_ROOT_TREE_OBJECTID ||
1809 			     owner == BTRFS_CHUNK_TREE_OBJECTID ||
1810 			     owner == BTRFS_DEV_TREE_OBJECTID ||
1811 			     owner == BTRFS_FS_TREE_OBJECTID ||
1812 			     owner == BTRFS_DATA_RELOC_TREE_OBJECTID)) {
1813 			generic_err(leaf, 0,
1814 			"invalid root, root %llu must never be empty",
1815 				    owner);
1816 			return BTRFS_TREE_BLOCK_INVALID_NRITEMS;
1817 		}
1818 
1819 		/* Unknown tree */
1820 		if (unlikely(owner == 0)) {
1821 			generic_err(leaf, 0,
1822 				"invalid owner, root 0 is not defined");
1823 			return BTRFS_TREE_BLOCK_INVALID_OWNER;
1824 		}
1825 
1826 		/* EXTENT_TREE_V2 can have empty extent trees. */
1827 		if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1828 			return BTRFS_TREE_BLOCK_CLEAN;
1829 
1830 		if (unlikely(owner == BTRFS_EXTENT_TREE_OBJECTID)) {
1831 			generic_err(leaf, 0,
1832 			"invalid root, root %llu must never be empty",
1833 				    owner);
1834 			return BTRFS_TREE_BLOCK_INVALID_NRITEMS;
1835 		}
1836 
1837 		return BTRFS_TREE_BLOCK_CLEAN;
1838 	}
1839 
1840 	if (unlikely(nritems == 0))
1841 		return BTRFS_TREE_BLOCK_CLEAN;
1842 
1843 	/*
1844 	 * Check the following things to make sure this is a good leaf, and
1845 	 * leaf users won't need to bother with similar sanity checks:
1846 	 *
1847 	 * 1) key ordering
1848 	 * 2) item offset and size
1849 	 *    No overlap, no hole, all inside the leaf.
1850 	 * 3) item content
1851 	 *    If possible, do comprehensive sanity check.
1852 	 *    NOTE: All checks must only rely on the item data itself.
1853 	 */
1854 	for (slot = 0; slot < nritems; slot++) {
1855 		u32 item_end_expected;
1856 		u64 item_data_end;
1857 
1858 		btrfs_item_key_to_cpu(leaf, &key, slot);
1859 
1860 		/* Make sure the keys are in the right order */
1861 		if (unlikely(btrfs_comp_cpu_keys(&prev_key, &key) >= 0)) {
1862 			generic_err(leaf, slot,
1863 	"bad key order, prev (%llu %u %llu) current (%llu %u %llu)",
1864 				prev_key.objectid, prev_key.type,
1865 				prev_key.offset, key.objectid, key.type,
1866 				key.offset);
1867 			return BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
1868 		}
1869 
1870 		item_data_end = (u64)btrfs_item_offset(leaf, slot) +
1871 				btrfs_item_size(leaf, slot);
1872 		/*
1873 		 * Make sure the offset and ends are right, remember that the
1874 		 * item data starts at the end of the leaf and grows towards the
1875 		 * front.
1876 		 */
1877 		if (slot == 0)
1878 			item_end_expected = BTRFS_LEAF_DATA_SIZE(fs_info);
1879 		else
1880 			item_end_expected = btrfs_item_offset(leaf,
1881 								 slot - 1);
1882 		if (unlikely(item_data_end != item_end_expected)) {
1883 			generic_err(leaf, slot,
1884 				"unexpected item end, have %llu expect %u",
1885 				item_data_end, item_end_expected);
1886 			return BTRFS_TREE_BLOCK_INVALID_OFFSETS;
1887 		}
1888 
1889 		/*
1890 		 * Check to make sure that we don't point outside of the leaf,
1891 		 * just in case all the items are consistent to each other, but
1892 		 * all point outside of the leaf.
1893 		 */
1894 		if (unlikely(item_data_end > BTRFS_LEAF_DATA_SIZE(fs_info))) {
1895 			generic_err(leaf, slot,
1896 			"slot end outside of leaf, have %llu expect range [0, %u]",
1897 				item_data_end, BTRFS_LEAF_DATA_SIZE(fs_info));
1898 			return BTRFS_TREE_BLOCK_INVALID_OFFSETS;
1899 		}
1900 
1901 		/* Also check if the item pointer overlaps with btrfs item. */
1902 		if (unlikely(btrfs_item_ptr_offset(leaf, slot) <
1903 			     btrfs_item_nr_offset(leaf, slot) + sizeof(struct btrfs_item))) {
1904 			generic_err(leaf, slot,
1905 		"slot overlaps with its data, item end %lu data start %lu",
1906 				btrfs_item_nr_offset(leaf, slot) +
1907 				sizeof(struct btrfs_item),
1908 				btrfs_item_ptr_offset(leaf, slot));
1909 			return BTRFS_TREE_BLOCK_INVALID_OFFSETS;
1910 		}
1911 
1912 		/*
1913 		 * We only want to do this if WRITTEN is set, otherwise the leaf
1914 		 * may be in some intermediate state and won't appear valid.
1915 		 */
1916 		if (btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_WRITTEN)) {
1917 			enum btrfs_tree_block_status ret;
1918 
1919 			/*
1920 			 * Check if the item size and content meet other
1921 			 * criteria
1922 			 */
1923 			ret = check_leaf_item(leaf, &key, slot, &prev_key);
1924 			if (unlikely(ret != BTRFS_TREE_BLOCK_CLEAN))
1925 				return ret;
1926 		}
1927 
1928 		prev_key.objectid = key.objectid;
1929 		prev_key.type = key.type;
1930 		prev_key.offset = key.offset;
1931 	}
1932 
1933 	return BTRFS_TREE_BLOCK_CLEAN;
1934 }
1935 
1936 int btrfs_check_leaf(struct extent_buffer *leaf)
1937 {
1938 	enum btrfs_tree_block_status ret;
1939 
1940 	ret = __btrfs_check_leaf(leaf);
1941 	if (unlikely(ret != BTRFS_TREE_BLOCK_CLEAN))
1942 		return -EUCLEAN;
1943 	return 0;
1944 }
1945 ALLOW_ERROR_INJECTION(btrfs_check_leaf, ERRNO);
1946 
1947 enum btrfs_tree_block_status __btrfs_check_node(struct extent_buffer *node)
1948 {
1949 	struct btrfs_fs_info *fs_info = node->fs_info;
1950 	unsigned long nr = btrfs_header_nritems(node);
1951 	struct btrfs_key key, next_key;
1952 	int slot;
1953 	int level = btrfs_header_level(node);
1954 	u64 bytenr;
1955 
1956 	if (unlikely(level <= 0 || level >= BTRFS_MAX_LEVEL)) {
1957 		generic_err(node, 0,
1958 			"invalid level for node, have %d expect [1, %d]",
1959 			level, BTRFS_MAX_LEVEL - 1);
1960 		return BTRFS_TREE_BLOCK_INVALID_LEVEL;
1961 	}
1962 	if (unlikely(nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(fs_info))) {
1963 		btrfs_crit(fs_info,
1964 "corrupt node: root=%llu block=%llu, nritems too %s, have %lu expect range [1,%u]",
1965 			   btrfs_header_owner(node), node->start,
1966 			   nr == 0 ? "small" : "large", nr,
1967 			   BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1968 		return BTRFS_TREE_BLOCK_INVALID_NRITEMS;
1969 	}
1970 
1971 	for (slot = 0; slot < nr - 1; slot++) {
1972 		bytenr = btrfs_node_blockptr(node, slot);
1973 		btrfs_node_key_to_cpu(node, &key, slot);
1974 		btrfs_node_key_to_cpu(node, &next_key, slot + 1);
1975 
1976 		if (unlikely(!bytenr)) {
1977 			generic_err(node, slot,
1978 				"invalid NULL node pointer");
1979 			return BTRFS_TREE_BLOCK_INVALID_BLOCKPTR;
1980 		}
1981 		if (unlikely(!IS_ALIGNED(bytenr, fs_info->sectorsize))) {
1982 			generic_err(node, slot,
1983 			"unaligned pointer, have %llu should be aligned to %u",
1984 				bytenr, fs_info->sectorsize);
1985 			return BTRFS_TREE_BLOCK_INVALID_BLOCKPTR;
1986 		}
1987 
1988 		if (unlikely(btrfs_comp_cpu_keys(&key, &next_key) >= 0)) {
1989 			generic_err(node, slot,
1990 	"bad key order, current (%llu %u %llu) next (%llu %u %llu)",
1991 				key.objectid, key.type, key.offset,
1992 				next_key.objectid, next_key.type,
1993 				next_key.offset);
1994 			return BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
1995 		}
1996 	}
1997 	return BTRFS_TREE_BLOCK_CLEAN;
1998 }
1999 
2000 int btrfs_check_node(struct extent_buffer *node)
2001 {
2002 	enum btrfs_tree_block_status ret;
2003 
2004 	ret = __btrfs_check_node(node);
2005 	if (unlikely(ret != BTRFS_TREE_BLOCK_CLEAN))
2006 		return -EUCLEAN;
2007 	return 0;
2008 }
2009 ALLOW_ERROR_INJECTION(btrfs_check_node, ERRNO);
2010 
2011 int btrfs_check_eb_owner(const struct extent_buffer *eb, u64 root_owner)
2012 {
2013 	const bool is_subvol = is_fstree(root_owner);
2014 	const u64 eb_owner = btrfs_header_owner(eb);
2015 
2016 	/*
2017 	 * Skip dummy fs, as selftests don't create unique ebs for each dummy
2018 	 * root.
2019 	 */
2020 	if (test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &eb->fs_info->fs_state))
2021 		return 0;
2022 	/*
2023 	 * There are several call sites (backref walking, qgroup, and data
2024 	 * reloc) passing 0 as @root_owner, as they are not holding the
2025 	 * tree root.  In that case, we can not do a reliable ownership check,
2026 	 * so just exit.
2027 	 */
2028 	if (root_owner == 0)
2029 		return 0;
2030 	/*
2031 	 * These trees use key.offset as their owner, our callers don't have
2032 	 * the extra capacity to pass key.offset here.  So we just skip them.
2033 	 */
2034 	if (root_owner == BTRFS_TREE_LOG_OBJECTID ||
2035 	    root_owner == BTRFS_TREE_RELOC_OBJECTID)
2036 		return 0;
2037 
2038 	if (!is_subvol) {
2039 		/* For non-subvolume trees, the eb owner should match root owner */
2040 		if (unlikely(root_owner != eb_owner)) {
2041 			btrfs_crit(eb->fs_info,
2042 "corrupted %s, root=%llu block=%llu owner mismatch, have %llu expect %llu",
2043 				btrfs_header_level(eb) == 0 ? "leaf" : "node",
2044 				root_owner, btrfs_header_bytenr(eb), eb_owner,
2045 				root_owner);
2046 			return -EUCLEAN;
2047 		}
2048 		return 0;
2049 	}
2050 
2051 	/*
2052 	 * For subvolume trees, owners can mismatch, but they should all belong
2053 	 * to subvolume trees.
2054 	 */
2055 	if (unlikely(is_subvol != is_fstree(eb_owner))) {
2056 		btrfs_crit(eb->fs_info,
2057 "corrupted %s, root=%llu block=%llu owner mismatch, have %llu expect [%llu, %llu]",
2058 			btrfs_header_level(eb) == 0 ? "leaf" : "node",
2059 			root_owner, btrfs_header_bytenr(eb), eb_owner,
2060 			BTRFS_FIRST_FREE_OBJECTID, BTRFS_LAST_FREE_OBJECTID);
2061 		return -EUCLEAN;
2062 	}
2063 	return 0;
2064 }
2065 
2066 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
2067 			   struct btrfs_key *first_key, u64 parent_transid)
2068 {
2069 	struct btrfs_fs_info *fs_info = eb->fs_info;
2070 	int found_level;
2071 	struct btrfs_key found_key;
2072 	int ret;
2073 
2074 	found_level = btrfs_header_level(eb);
2075 	if (found_level != level) {
2076 		WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
2077 		     KERN_ERR "BTRFS: tree level check failed\n");
2078 		btrfs_err(fs_info,
2079 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
2080 			  eb->start, level, found_level);
2081 		return -EIO;
2082 	}
2083 
2084 	if (!first_key)
2085 		return 0;
2086 
2087 	/*
2088 	 * For live tree block (new tree blocks in current transaction),
2089 	 * we need proper lock context to avoid race, which is impossible here.
2090 	 * So we only checks tree blocks which is read from disk, whose
2091 	 * generation <= fs_info->last_trans_committed.
2092 	 */
2093 	if (btrfs_header_generation(eb) > btrfs_get_last_trans_committed(fs_info))
2094 		return 0;
2095 
2096 	/* We have @first_key, so this @eb must have at least one item */
2097 	if (btrfs_header_nritems(eb) == 0) {
2098 		btrfs_err(fs_info,
2099 		"invalid tree nritems, bytenr=%llu nritems=0 expect >0",
2100 			  eb->start);
2101 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
2102 		return -EUCLEAN;
2103 	}
2104 
2105 	if (found_level)
2106 		btrfs_node_key_to_cpu(eb, &found_key, 0);
2107 	else
2108 		btrfs_item_key_to_cpu(eb, &found_key, 0);
2109 	ret = btrfs_comp_cpu_keys(first_key, &found_key);
2110 
2111 	if (ret) {
2112 		WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
2113 		     KERN_ERR "BTRFS: tree first key check failed\n");
2114 		btrfs_err(fs_info,
2115 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
2116 			  eb->start, parent_transid, first_key->objectid,
2117 			  first_key->type, first_key->offset,
2118 			  found_key.objectid, found_key.type,
2119 			  found_key.offset);
2120 	}
2121 	return ret;
2122 }
2123