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 "compression.h" 25 #include "volumes.h" 26 #include "misc.h" 27 #include "fs.h" 28 #include "accessors.h" 29 #include "file-item.h" 30 #include "inode-item.h" 31 #include "dir-item.h" 32 #include "extent-tree.h" 33 34 /* 35 * Error message should follow the following format: 36 * corrupt <type>: <identifier>, <reason>[, <bad_value>] 37 * 38 * @type: leaf or node 39 * @identifier: the necessary info to locate the leaf/node. 40 * It's recommended to decode key.objecitd/offset if it's 41 * meaningful. 42 * @reason: describe the error 43 * @bad_value: optional, it's recommended to output bad value and its 44 * expected value (range). 45 * 46 * Since comma is used to separate the components, only space is allowed 47 * inside each component. 48 */ 49 50 /* 51 * Append generic "corrupt leaf/node root=%llu block=%llu slot=%d: " to @fmt. 52 * Allows callers to customize the output. 53 */ 54 __printf(3, 4) 55 __cold 56 static void generic_err(const struct extent_buffer *eb, int slot, 57 const char *fmt, ...) 58 { 59 const struct btrfs_fs_info *fs_info = eb->fs_info; 60 struct va_format vaf; 61 va_list args; 62 63 va_start(args, fmt); 64 65 vaf.fmt = fmt; 66 vaf.va = &args; 67 68 dump_page(folio_page(eb->folios[0], 0), "eb page dump"); 69 btrfs_crit(fs_info, 70 "corrupt %s: root=%llu block=%llu slot=%d, %pV", 71 btrfs_header_level(eb) == 0 ? "leaf" : "node", 72 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot, &vaf); 73 va_end(args); 74 } 75 76 /* 77 * Customized reporter for extent data item, since its key objectid and 78 * offset has its own meaning. 79 */ 80 __printf(3, 4) 81 __cold 82 static void file_extent_err(const struct extent_buffer *eb, int slot, 83 const char *fmt, ...) 84 { 85 const struct btrfs_fs_info *fs_info = eb->fs_info; 86 struct btrfs_key key; 87 struct va_format vaf; 88 va_list args; 89 90 btrfs_item_key_to_cpu(eb, &key, slot); 91 va_start(args, fmt); 92 93 vaf.fmt = fmt; 94 vaf.va = &args; 95 96 dump_page(folio_page(eb->folios[0], 0), "eb page dump"); 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 const u64 val = btrfs_file_extent_##name((leaf), (fi)); \ 112 const bool not_aligned = !IS_ALIGNED(val, (alignment)); \ 113 \ 114 if (unlikely(not_aligned)) \ 115 file_extent_err((leaf), (slot), \ 116 "invalid %s for file extent, have %llu, should be aligned to %u", \ 117 (#name), val, (alignment)); \ 118 not_aligned; \ 119 }) 120 121 static u64 file_extent_end(struct extent_buffer *leaf, 122 struct btrfs_key *key, 123 struct btrfs_file_extent_item *extent) 124 { 125 u64 end; 126 u64 len; 127 128 if (btrfs_file_extent_type(leaf, extent) == BTRFS_FILE_EXTENT_INLINE) { 129 len = btrfs_file_extent_ram_bytes(leaf, extent); 130 end = ALIGN(key->offset + len, leaf->fs_info->sectorsize); 131 } else { 132 len = btrfs_file_extent_num_bytes(leaf, extent); 133 end = key->offset + len; 134 } 135 return end; 136 } 137 138 /* 139 * Customized report for dir_item, the only new important information is 140 * key->objectid, which represents inode number 141 */ 142 __printf(3, 4) 143 __cold 144 static void dir_item_err(const struct extent_buffer *eb, int slot, 145 const char *fmt, ...) 146 { 147 const struct btrfs_fs_info *fs_info = eb->fs_info; 148 struct btrfs_key key; 149 struct va_format vaf; 150 va_list args; 151 152 btrfs_item_key_to_cpu(eb, &key, slot); 153 va_start(args, fmt); 154 155 vaf.fmt = fmt; 156 vaf.va = &args; 157 158 dump_page(folio_page(eb->folios[0], 0), "eb page dump"); 159 btrfs_crit(fs_info, 160 "corrupt %s: root=%llu block=%llu slot=%d ino=%llu, %pV", 161 btrfs_header_level(eb) == 0 ? "leaf" : "node", 162 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot, 163 key.objectid, &vaf); 164 va_end(args); 165 } 166 167 /* Record info for the last hit inode. */ 168 struct saved_inode_info { 169 u64 ino; 170 u32 mode; 171 }; 172 173 /* 174 * This functions checks prev_key->objectid, to ensure current key and prev_key 175 * share the same objectid as inode number. 176 * 177 * This is to detect missing INODE_ITEM in subvolume trees. 178 * 179 * Return true if everything is OK or we don't need to check. 180 * Return false if anything is wrong. 181 */ 182 static bool check_prev_ino(struct extent_buffer *leaf, 183 struct btrfs_key *key, int slot, 184 struct btrfs_key *prev_key) 185 { 186 /* No prev key, skip check */ 187 if (slot == 0) 188 return true; 189 190 /* Only these key->types needs to be checked */ 191 ASSERT(key->type == BTRFS_XATTR_ITEM_KEY || 192 key->type == BTRFS_INODE_REF_KEY || 193 key->type == BTRFS_INODE_EXTREF_KEY || 194 key->type == BTRFS_DIR_INDEX_KEY || 195 key->type == BTRFS_DIR_ITEM_KEY || 196 key->type == BTRFS_EXTENT_DATA_KEY, "key->type=%u", key->type); 197 198 /* 199 * Only subvolume trees along with their reloc trees need this check. 200 * Things like log tree doesn't follow this ino requirement. 201 */ 202 if (!btrfs_is_fstree(btrfs_header_owner(leaf))) 203 return true; 204 205 if (key->objectid == prev_key->objectid) 206 return true; 207 208 /* Error found */ 209 dir_item_err(leaf, slot, 210 "invalid previous key objectid, have %llu expect %llu", 211 prev_key->objectid, key->objectid); 212 return false; 213 } 214 215 static bool can_have_extent_data(struct extent_buffer *leaf, 216 struct btrfs_key *key, int slot, u8 fi_type, 217 const struct saved_inode_info *inode_info) 218 { 219 /* No inode item in this leaf. */ 220 if (inode_info->ino != key->objectid) 221 return true; 222 if (S_ISREG(inode_info->mode)) 223 return true; 224 if (S_ISLNK(inode_info->mode)) { 225 /* For a symlink, the file extent item should always be inlined. */ 226 if (unlikely(fi_type != BTRFS_FILE_EXTENT_INLINE)) 227 return false; 228 return true; 229 } 230 231 /* 232 * The rest are special files, e.g. block/FIFO files, which cannot 233 * have any file extent. 234 */ 235 return false; 236 } 237 238 static int check_extent_data_item(struct extent_buffer *leaf, 239 struct btrfs_key *key, int slot, 240 struct btrfs_key *prev_key, 241 const struct saved_inode_info *inode_info) 242 { 243 struct btrfs_fs_info *fs_info = leaf->fs_info; 244 struct btrfs_file_extent_item *fi; 245 u32 sectorsize = fs_info->sectorsize; 246 u32 item_size = btrfs_item_size(leaf, slot); 247 u64 extent_end; 248 u8 fi_type; 249 250 if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) { 251 file_extent_err(leaf, slot, 252 "unaligned file_offset for file extent, have %llu should be aligned to %u", 253 key->offset, sectorsize); 254 return -EUCLEAN; 255 } 256 257 /* 258 * Previous key must have the same key->objectid (ino). 259 * It can be XATTR_ITEM, INODE_ITEM or just another EXTENT_DATA. 260 * But if objectids mismatch, it means we have a missing 261 * INODE_ITEM. 262 */ 263 if (unlikely(!check_prev_ino(leaf, key, slot, prev_key))) 264 return -EUCLEAN; 265 266 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); 267 268 /* 269 * Make sure the item contains at least inline header, so the file 270 * extent type is not some garbage. 271 */ 272 if (unlikely(item_size < BTRFS_FILE_EXTENT_INLINE_DATA_START)) { 273 file_extent_err(leaf, slot, 274 "invalid item size, have %u expect [%zu, %u)", 275 item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START, 276 SZ_4K); 277 return -EUCLEAN; 278 } 279 fi_type = btrfs_file_extent_type(leaf, fi); 280 if (unlikely(fi_type >= BTRFS_NR_FILE_EXTENT_TYPES)) { 281 file_extent_err(leaf, slot, 282 "invalid type for file extent, have %u expect range [0, %u]", 283 fi_type, BTRFS_NR_FILE_EXTENT_TYPES - 1); 284 return -EUCLEAN; 285 } 286 287 if (unlikely(!can_have_extent_data(leaf, key, slot, fi_type, inode_info))) { 288 file_extent_err(leaf, slot, 289 "unexpected file extent item type %u for inode mode 0%o", 290 fi_type, inode_info->mode); 291 return -EUCLEAN; 292 } 293 294 /* 295 * Support for new compression/encryption must introduce incompat flag, 296 * and must be caught in open_ctree(). 297 */ 298 if (unlikely(btrfs_file_extent_compression(leaf, fi) >= 299 BTRFS_NR_COMPRESS_TYPES)) { 300 file_extent_err(leaf, slot, 301 "invalid compression for file extent, have %u expect range [0, %u]", 302 btrfs_file_extent_compression(leaf, fi), 303 BTRFS_NR_COMPRESS_TYPES - 1); 304 return -EUCLEAN; 305 } 306 if (unlikely(btrfs_file_extent_encryption(leaf, fi))) { 307 file_extent_err(leaf, slot, 308 "invalid encryption for file extent, have %u expect 0", 309 btrfs_file_extent_encryption(leaf, fi)); 310 return -EUCLEAN; 311 } 312 313 if (fi_type == BTRFS_FILE_EXTENT_INLINE) { 314 /* Inline extent must have 0 as key offset */ 315 if (unlikely(key->offset)) { 316 file_extent_err(leaf, slot, 317 "invalid file_offset for inline file extent, have %llu expect 0", 318 key->offset); 319 return -EUCLEAN; 320 } 321 322 /* Compressed inline extent has no on-disk size, skip it */ 323 if (btrfs_file_extent_compression(leaf, fi) != 324 BTRFS_COMPRESS_NONE) 325 return 0; 326 327 /* Uncompressed inline extent size must match item size */ 328 if (unlikely(item_size != BTRFS_FILE_EXTENT_INLINE_DATA_START + 329 btrfs_file_extent_ram_bytes(leaf, fi))) { 330 file_extent_err(leaf, slot, 331 "invalid ram_bytes for uncompressed inline extent, have %u expect %llu", 332 item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START + 333 btrfs_file_extent_ram_bytes(leaf, fi)); 334 return -EUCLEAN; 335 } 336 return 0; 337 } 338 339 /* 340 * For the data reloc tree, file extent items are written by 341 * relocation's own paths. The data reloc inode is created with 342 * BTRFS_INODE_NOCOMPRESS, so insert_ordered_extent_file_extent() 343 * always leaves the compression field at 0. Encryption and 344 * other_encoding are reserved-and-zero in btrfs. A non-zero value 345 * for any of these means the leaf decoded from disk does not match 346 * what the kernel wrote, i.e. on-disk corruption. 347 * 348 * The file_extent_item's offset field is NOT a universal invariant 349 * here: partial-PREALLOC writebacks legitimately produce REG items 350 * with non-zero offset at non-boundary keys. The offset check is 351 * performed at the call site in get_new_location(), which only 352 * inspects cluster-boundary keys where offset is always 0. 353 */ 354 if (unlikely(btrfs_header_owner(leaf) == BTRFS_DATA_RELOC_TREE_OBJECTID && 355 (btrfs_file_extent_compression(leaf, fi) || 356 btrfs_file_extent_encryption(leaf, fi) || 357 btrfs_file_extent_other_encoding(leaf, fi)))) { 358 file_extent_err(leaf, slot, 359 "invalid encoding fields for data reloc tree, compression=%u encryption=%u other_encoding=%u", 360 btrfs_file_extent_compression(leaf, fi), 361 btrfs_file_extent_encryption(leaf, fi), 362 btrfs_file_extent_other_encoding(leaf, fi)); 363 return -EUCLEAN; 364 } 365 366 /* Regular or preallocated extent has fixed item size */ 367 if (unlikely(item_size != sizeof(*fi))) { 368 file_extent_err(leaf, slot, 369 "invalid item size for reg/prealloc file extent, have %u expect %zu", 370 item_size, sizeof(*fi)); 371 return -EUCLEAN; 372 } 373 if (unlikely(CHECK_FE_ALIGNED(leaf, slot, fi, ram_bytes, sectorsize) || 374 CHECK_FE_ALIGNED(leaf, slot, fi, disk_bytenr, sectorsize) || 375 CHECK_FE_ALIGNED(leaf, slot, fi, disk_num_bytes, sectorsize) || 376 CHECK_FE_ALIGNED(leaf, slot, fi, offset, sectorsize) || 377 CHECK_FE_ALIGNED(leaf, slot, fi, num_bytes, sectorsize))) 378 return -EUCLEAN; 379 380 /* Catch extent end overflow */ 381 if (unlikely(check_add_overflow(btrfs_file_extent_num_bytes(leaf, fi), 382 key->offset, &extent_end))) { 383 file_extent_err(leaf, slot, 384 "extent end overflow, have file offset %llu extent num bytes %llu", 385 key->offset, 386 btrfs_file_extent_num_bytes(leaf, fi)); 387 return -EUCLEAN; 388 } 389 390 /* 391 * Check that no two consecutive file extent items, in the same leaf, 392 * present ranges that overlap each other. 393 */ 394 if (slot > 0 && 395 prev_key->objectid == key->objectid && 396 prev_key->type == BTRFS_EXTENT_DATA_KEY) { 397 struct btrfs_file_extent_item *prev_fi; 398 u64 prev_end; 399 400 prev_fi = btrfs_item_ptr(leaf, slot - 1, 401 struct btrfs_file_extent_item); 402 prev_end = file_extent_end(leaf, prev_key, prev_fi); 403 if (unlikely(prev_end > key->offset)) { 404 file_extent_err(leaf, slot - 1, 405 "file extent end range (%llu) goes beyond start offset (%llu) of the next file extent", 406 prev_end, key->offset); 407 return -EUCLEAN; 408 } 409 } 410 411 /* 412 * For non-compressed data extents, ram_bytes should match its 413 * disk_num_bytes. 414 * However we do not really utilize ram_bytes in this case, so this check 415 * is only optional for DEBUG builds for developers to catch the 416 * unexpected behaviors. 417 */ 418 if (IS_ENABLED(CONFIG_BTRFS_DEBUG) && 419 btrfs_file_extent_compression(leaf, fi) == BTRFS_COMPRESS_NONE && 420 btrfs_file_extent_disk_bytenr(leaf, fi)) { 421 if (WARN_ON(btrfs_file_extent_ram_bytes(leaf, fi) != 422 btrfs_file_extent_disk_num_bytes(leaf, fi))) 423 file_extent_err(leaf, slot, 424 "mismatch ram_bytes (%llu) and disk_num_bytes (%llu) for non-compressed extent", 425 btrfs_file_extent_ram_bytes(leaf, fi), 426 btrfs_file_extent_disk_num_bytes(leaf, fi)); 427 } 428 429 return 0; 430 } 431 432 static int check_csum_item(struct extent_buffer *leaf, struct btrfs_key *key, 433 int slot, struct btrfs_key *prev_key) 434 { 435 struct btrfs_fs_info *fs_info = leaf->fs_info; 436 u32 sectorsize = fs_info->sectorsize; 437 const u32 csumsize = fs_info->csum_size; 438 439 if (unlikely(key->objectid != BTRFS_EXTENT_CSUM_OBJECTID)) { 440 generic_err(leaf, slot, 441 "invalid key objectid for csum item, have %llu expect %llu", 442 key->objectid, BTRFS_EXTENT_CSUM_OBJECTID); 443 return -EUCLEAN; 444 } 445 if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) { 446 generic_err(leaf, slot, 447 "unaligned key offset for csum item, have %llu should be aligned to %u", 448 key->offset, sectorsize); 449 return -EUCLEAN; 450 } 451 if (unlikely(!IS_ALIGNED(btrfs_item_size(leaf, slot), csumsize))) { 452 generic_err(leaf, slot, 453 "unaligned item size for csum item, have %u should be aligned to %u", 454 btrfs_item_size(leaf, slot), csumsize); 455 return -EUCLEAN; 456 } 457 if (slot > 0 && prev_key->type == BTRFS_EXTENT_CSUM_KEY) { 458 u64 prev_csum_end; 459 u32 prev_item_size; 460 461 prev_item_size = btrfs_item_size(leaf, slot - 1); 462 prev_csum_end = (prev_item_size / csumsize) * sectorsize; 463 prev_csum_end += prev_key->offset; 464 if (unlikely(prev_csum_end > key->offset)) { 465 generic_err(leaf, slot - 1, 466 "csum end range (%llu) goes beyond the start range (%llu) of the next csum item", 467 prev_csum_end, key->offset); 468 return -EUCLEAN; 469 } 470 } 471 return 0; 472 } 473 474 /* Inode item error output has the same format as dir_item_err() */ 475 #define inode_item_err(eb, slot, fmt, ...) \ 476 dir_item_err(eb, slot, fmt, __VA_ARGS__) 477 478 static int check_inode_key(struct extent_buffer *leaf, struct btrfs_key *key, 479 int slot) 480 { 481 struct btrfs_key item_key; 482 bool is_inode_item; 483 484 btrfs_item_key_to_cpu(leaf, &item_key, slot); 485 is_inode_item = (item_key.type == BTRFS_INODE_ITEM_KEY); 486 487 /* For XATTR_ITEM, location key should be all 0 */ 488 if (item_key.type == BTRFS_XATTR_ITEM_KEY) { 489 if (unlikely(key->objectid != 0 || key->type != 0 || 490 key->offset != 0)) 491 return -EUCLEAN; 492 return 0; 493 } 494 495 if (unlikely((key->objectid < BTRFS_FIRST_FREE_OBJECTID || 496 key->objectid > BTRFS_LAST_FREE_OBJECTID) && 497 key->objectid != BTRFS_ROOT_TREE_DIR_OBJECTID && 498 key->objectid != BTRFS_FREE_INO_OBJECTID)) { 499 if (is_inode_item) { 500 generic_err(leaf, slot, 501 "invalid key objectid: has %llu expect %llu or [%llu, %llu] or %llu", 502 key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID, 503 BTRFS_FIRST_FREE_OBJECTID, 504 BTRFS_LAST_FREE_OBJECTID, 505 BTRFS_FREE_INO_OBJECTID); 506 } else { 507 dir_item_err(leaf, slot, 508 "invalid location key objectid: has %llu expect %llu or [%llu, %llu] or %llu", 509 key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID, 510 BTRFS_FIRST_FREE_OBJECTID, 511 BTRFS_LAST_FREE_OBJECTID, 512 BTRFS_FREE_INO_OBJECTID); 513 } 514 return -EUCLEAN; 515 } 516 if (unlikely(key->offset != 0)) { 517 if (is_inode_item) 518 inode_item_err(leaf, slot, 519 "invalid key offset: has %llu expect 0", 520 key->offset); 521 else 522 dir_item_err(leaf, slot, 523 "invalid location key offset:has %llu expect 0", 524 key->offset); 525 return -EUCLEAN; 526 } 527 return 0; 528 } 529 530 static int check_root_key(struct extent_buffer *leaf, struct btrfs_key *key, 531 int slot) 532 { 533 struct btrfs_key item_key; 534 bool is_root_item; 535 536 btrfs_item_key_to_cpu(leaf, &item_key, slot); 537 is_root_item = (item_key.type == BTRFS_ROOT_ITEM_KEY); 538 539 /* 540 * Bad rootid for reloc trees. 541 * 542 * Reloc trees are only for subvolume trees, other trees only need 543 * to be COWed to be relocated. 544 */ 545 if (unlikely(is_root_item && key->objectid == BTRFS_TREE_RELOC_OBJECTID && 546 !btrfs_is_fstree(key->offset))) { 547 generic_err(leaf, slot, 548 "invalid reloc tree for root %lld, root id is not a subvolume tree", 549 key->offset); 550 return -EUCLEAN; 551 } 552 553 /* No such tree id */ 554 if (unlikely(key->objectid == 0)) { 555 if (is_root_item) 556 generic_err(leaf, slot, "invalid root id 0"); 557 else 558 dir_item_err(leaf, slot, 559 "invalid location key root id 0"); 560 return -EUCLEAN; 561 } 562 563 /* DIR_ITEM/INDEX/INODE_REF is not allowed to point to non-fs trees */ 564 if (unlikely(!btrfs_is_fstree(key->objectid) && !is_root_item)) { 565 dir_item_err(leaf, slot, 566 "invalid location key objectid, have %llu expect [%llu, %llu]", 567 key->objectid, BTRFS_FIRST_FREE_OBJECTID, 568 BTRFS_LAST_FREE_OBJECTID); 569 return -EUCLEAN; 570 } 571 572 /* 573 * ROOT_ITEM with non-zero offset means this is a snapshot, created at 574 * @offset transid. 575 * Furthermore, for location key in DIR_ITEM, its offset is always -1. 576 * 577 * So here we only check offset for reloc tree whose key->offset must 578 * be a valid tree. 579 */ 580 if (unlikely(key->objectid == BTRFS_TREE_RELOC_OBJECTID && 581 key->offset == 0)) { 582 generic_err(leaf, slot, "invalid root id 0 for reloc tree"); 583 return -EUCLEAN; 584 } 585 return 0; 586 } 587 588 static int check_dir_item(struct extent_buffer *leaf, 589 struct btrfs_key *key, struct btrfs_key *prev_key, 590 int slot) 591 { 592 struct btrfs_fs_info *fs_info = leaf->fs_info; 593 struct btrfs_dir_item *di; 594 u32 item_size = btrfs_item_size(leaf, slot); 595 u32 cur = 0; 596 597 if (unlikely(!check_prev_ino(leaf, key, slot, prev_key))) 598 return -EUCLEAN; 599 600 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item); 601 while (cur < item_size) { 602 struct btrfs_key location_key; 603 u32 name_len; 604 u32 data_len; 605 u32 max_name_len; 606 u32 total_size; 607 u32 name_hash; 608 u8 dir_type; 609 int ret; 610 611 /* header itself should not cross item boundary */ 612 if (unlikely(cur + sizeof(*di) > item_size)) { 613 dir_item_err(leaf, slot, 614 "dir item header crosses item boundary, have %zu boundary %u", 615 cur + sizeof(*di), item_size); 616 return -EUCLEAN; 617 } 618 619 /* Location key check */ 620 btrfs_dir_item_key_to_cpu(leaf, di, &location_key); 621 if (location_key.type == BTRFS_ROOT_ITEM_KEY) { 622 ret = check_root_key(leaf, &location_key, slot); 623 if (unlikely(ret < 0)) 624 return ret; 625 } else if (location_key.type == BTRFS_INODE_ITEM_KEY || 626 location_key.type == 0) { 627 ret = check_inode_key(leaf, &location_key, slot); 628 if (unlikely(ret < 0)) 629 return ret; 630 } else { 631 dir_item_err(leaf, slot, 632 "invalid location key type, have %u, expect %u or %u", 633 location_key.type, BTRFS_ROOT_ITEM_KEY, 634 BTRFS_INODE_ITEM_KEY); 635 return -EUCLEAN; 636 } 637 638 /* dir type check */ 639 dir_type = btrfs_dir_ftype(leaf, di); 640 if (unlikely(dir_type <= BTRFS_FT_UNKNOWN || 641 dir_type >= BTRFS_FT_MAX)) { 642 dir_item_err(leaf, slot, 643 "invalid dir item type, have %u expect (0, %u)", 644 dir_type, BTRFS_FT_MAX); 645 return -EUCLEAN; 646 } 647 648 if (unlikely(key->type == BTRFS_XATTR_ITEM_KEY && 649 dir_type != BTRFS_FT_XATTR)) { 650 dir_item_err(leaf, slot, 651 "invalid dir item type for XATTR key, have %u expect %u", 652 dir_type, BTRFS_FT_XATTR); 653 return -EUCLEAN; 654 } 655 if (unlikely(dir_type == BTRFS_FT_XATTR && 656 key->type != BTRFS_XATTR_ITEM_KEY)) { 657 dir_item_err(leaf, slot, 658 "xattr dir type found for non-XATTR key"); 659 return -EUCLEAN; 660 } 661 if (dir_type == BTRFS_FT_XATTR) 662 max_name_len = XATTR_NAME_MAX; 663 else 664 max_name_len = BTRFS_NAME_LEN; 665 666 /* Name/data length check */ 667 name_len = btrfs_dir_name_len(leaf, di); 668 data_len = btrfs_dir_data_len(leaf, di); 669 if (unlikely(name_len > max_name_len)) { 670 dir_item_err(leaf, slot, 671 "dir item name len too long, have %u max %u", 672 name_len, max_name_len); 673 return -EUCLEAN; 674 } 675 if (unlikely(name_len + data_len > BTRFS_MAX_XATTR_SIZE(fs_info))) { 676 dir_item_err(leaf, slot, 677 "dir item name and data len too long, have %u max %u", 678 name_len + data_len, 679 BTRFS_MAX_XATTR_SIZE(fs_info)); 680 return -EUCLEAN; 681 } 682 683 if (unlikely(data_len && dir_type != BTRFS_FT_XATTR)) { 684 dir_item_err(leaf, slot, 685 "dir item with invalid data len, have %u expect 0", 686 data_len); 687 return -EUCLEAN; 688 } 689 690 total_size = sizeof(*di) + name_len + data_len; 691 692 /* header and name/data should not cross item boundary */ 693 if (unlikely(cur + total_size > item_size)) { 694 dir_item_err(leaf, slot, 695 "dir item data crosses item boundary, have %u boundary %u", 696 cur + total_size, item_size); 697 return -EUCLEAN; 698 } 699 700 /* 701 * Special check for XATTR/DIR_ITEM, as key->offset is name 702 * hash, should match its name 703 */ 704 if (key->type == BTRFS_DIR_ITEM_KEY || 705 key->type == BTRFS_XATTR_ITEM_KEY) { 706 char namebuf[MAX(BTRFS_NAME_LEN, XATTR_NAME_MAX)]; 707 708 read_extent_buffer(leaf, namebuf, 709 (unsigned long)(di + 1), name_len); 710 name_hash = btrfs_name_hash(namebuf, name_len); 711 if (unlikely(key->offset != name_hash)) { 712 dir_item_err(leaf, slot, 713 "name hash mismatch with key, have 0x%016x expect 0x%016llx", 714 name_hash, key->offset); 715 return -EUCLEAN; 716 } 717 } 718 cur += total_size; 719 di = (struct btrfs_dir_item *)((void *)di + total_size); 720 } 721 return 0; 722 } 723 724 __printf(3, 4) 725 __cold 726 static void block_group_err(const struct extent_buffer *eb, int slot, 727 const char *fmt, ...) 728 { 729 const struct btrfs_fs_info *fs_info = eb->fs_info; 730 struct btrfs_key key; 731 struct va_format vaf; 732 va_list args; 733 734 btrfs_item_key_to_cpu(eb, &key, slot); 735 va_start(args, fmt); 736 737 vaf.fmt = fmt; 738 vaf.va = &args; 739 740 dump_page(folio_page(eb->folios[0], 0), "eb page dump"); 741 btrfs_crit(fs_info, 742 "corrupt %s: root=%llu block=%llu slot=%d bg_start=%llu bg_len=%llu, %pV", 743 btrfs_header_level(eb) == 0 ? "leaf" : "node", 744 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot, 745 key.objectid, key.offset, &vaf); 746 va_end(args); 747 } 748 749 static int check_block_group_item(struct extent_buffer *leaf, 750 struct btrfs_key *key, int slot) 751 { 752 struct btrfs_fs_info *fs_info = leaf->fs_info; 753 struct btrfs_block_group_item bgi; 754 u32 item_size = btrfs_item_size(leaf, slot); 755 u64 chunk_objectid; 756 u64 flags; 757 u64 type; 758 size_t exp_size; 759 760 /* 761 * Here we don't really care about alignment since extent allocator can 762 * handle it. We care more about the size. 763 */ 764 if (unlikely(key->offset == 0)) { 765 block_group_err(leaf, slot, 766 "invalid block group size 0"); 767 return -EUCLEAN; 768 } 769 770 if (btrfs_fs_incompat(fs_info, REMAP_TREE)) 771 exp_size = sizeof(struct btrfs_block_group_item_v2); 772 else 773 exp_size = sizeof(struct btrfs_block_group_item); 774 775 if (unlikely(item_size != exp_size)) { 776 block_group_err(leaf, slot, 777 "invalid item size, have %u expect %zu", 778 item_size, exp_size); 779 return -EUCLEAN; 780 } 781 782 read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot), 783 sizeof(bgi)); 784 chunk_objectid = btrfs_stack_block_group_chunk_objectid(&bgi); 785 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 786 /* 787 * We don't init the nr_global_roots until we load the global 788 * roots, so this could be 0 at mount time. If it's 0 we'll 789 * just assume we're fine, and later we'll check against our 790 * actual value. 791 */ 792 if (unlikely(fs_info->nr_global_roots && 793 chunk_objectid >= fs_info->nr_global_roots)) { 794 block_group_err(leaf, slot, 795 "invalid block group global root id, have %llu, needs to be <= %llu", 796 chunk_objectid, 797 fs_info->nr_global_roots); 798 return -EUCLEAN; 799 } 800 } else if (unlikely(chunk_objectid != BTRFS_FIRST_CHUNK_TREE_OBJECTID)) { 801 block_group_err(leaf, slot, 802 "invalid block group chunk objectid, have %llu expect %llu", 803 btrfs_stack_block_group_chunk_objectid(&bgi), 804 BTRFS_FIRST_CHUNK_TREE_OBJECTID); 805 return -EUCLEAN; 806 } 807 808 if (unlikely(btrfs_stack_block_group_used(&bgi) > key->offset)) { 809 block_group_err(leaf, slot, 810 "invalid block group used, have %llu expect [0, %llu)", 811 btrfs_stack_block_group_used(&bgi), key->offset); 812 return -EUCLEAN; 813 } 814 815 flags = btrfs_stack_block_group_flags(&bgi); 816 if (unlikely(hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) > 1)) { 817 block_group_err(leaf, slot, 818 "invalid profile flags, have 0x%llx (%lu bits set) expect no more than 1 bit set", 819 flags & BTRFS_BLOCK_GROUP_PROFILE_MASK, 820 hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)); 821 return -EUCLEAN; 822 } 823 824 if (unlikely(flags & BTRFS_BLOCK_GROUP_METADATA_REMAP && 825 !btrfs_fs_incompat(fs_info, REMAP_TREE))) { 826 block_group_err(leaf, slot, 827 "invalid flags, have 0x%llx (METADATA_REMAP flag set) but no remap-tree incompat flag", 828 flags); 829 return -EUCLEAN; 830 } 831 832 type = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; 833 if (unlikely(type != BTRFS_BLOCK_GROUP_DATA && 834 type != BTRFS_BLOCK_GROUP_METADATA && 835 type != BTRFS_BLOCK_GROUP_SYSTEM && 836 type != BTRFS_BLOCK_GROUP_METADATA_REMAP && 837 type != (BTRFS_BLOCK_GROUP_METADATA | 838 BTRFS_BLOCK_GROUP_DATA))) { 839 block_group_err(leaf, slot, 840 "invalid type, have 0x%llx (%lu bits set) expect either 0x%llx, 0x%llx, 0x%llx, 0x%llx or 0x%llx", 841 type, hweight64(type), 842 BTRFS_BLOCK_GROUP_DATA, BTRFS_BLOCK_GROUP_METADATA, 843 BTRFS_BLOCK_GROUP_SYSTEM, BTRFS_BLOCK_GROUP_METADATA_REMAP, 844 BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA); 845 return -EUCLEAN; 846 } 847 848 if (unlikely(!btrfs_fs_incompat(fs_info, REMAP_TREE) && 849 type == BTRFS_BLOCK_GROUP_METADATA_REMAP)) { 850 block_group_err(leaf, slot, 851 "invalid type, METADATA_REMAP set but REMAP_TREE incompat flag not set"); 852 return -EUCLEAN; 853 } 854 855 if (unlikely(!btrfs_fs_incompat(fs_info, REMAP_TREE) && 856 flags & BTRFS_BLOCK_GROUP_REMAPPED)) { 857 block_group_err(leaf, slot, 858 "invalid flags, REMAPPED set but REMAP_TREE incompat flag not set"); 859 return -EUCLEAN; 860 } 861 862 if (item_size == sizeof(struct btrfs_block_group_item_v2)) { 863 struct btrfs_block_group_item_v2 *bgi2; 864 u64 remap_bytes; 865 u32 identity_remap_count; 866 867 bgi2 = btrfs_item_ptr(leaf, slot, struct btrfs_block_group_item_v2); 868 remap_bytes = btrfs_block_group_v2_remap_bytes(leaf, bgi2); 869 870 if (unlikely(remap_bytes > key->offset)) { 871 block_group_err(leaf, slot, 872 "invalid remap_bytes, have %llu expect [0, %llu]", 873 remap_bytes, key->offset); 874 return -EUCLEAN; 875 } 876 877 identity_remap_count = btrfs_block_group_v2_identity_remap_count(leaf, bgi2); 878 if (unlikely((u64)identity_remap_count > 879 key->offset >> fs_info->sectorsize_bits)) { 880 block_group_err(leaf, slot, 881 "invalid identity_remap_count, have %u expect [0, %llu]", 882 identity_remap_count, 883 key->offset >> fs_info->sectorsize_bits); 884 return -EUCLEAN; 885 } 886 } 887 888 return 0; 889 } 890 891 __printf(5, 6) 892 __cold 893 static void chunk_err(const struct btrfs_fs_info *fs_info, 894 const struct extent_buffer *leaf, 895 const struct btrfs_chunk *chunk, u64 logical, 896 const char *fmt, ...) 897 { 898 bool is_sb = !leaf; 899 struct va_format vaf; 900 va_list args; 901 int i; 902 int slot = -1; 903 904 if (!is_sb) { 905 /* 906 * Get the slot number by iterating through all slots, this 907 * would provide better readability. 908 */ 909 for (i = 0; i < btrfs_header_nritems(leaf); i++) { 910 if (btrfs_item_ptr_offset(leaf, i) == 911 (unsigned long)chunk) { 912 slot = i; 913 break; 914 } 915 } 916 } 917 va_start(args, fmt); 918 vaf.fmt = fmt; 919 vaf.va = &args; 920 921 if (is_sb) 922 btrfs_crit(fs_info, 923 "corrupt superblock syschunk array: chunk_start=%llu, %pV", 924 logical, &vaf); 925 else 926 btrfs_crit(fs_info, 927 "corrupt leaf: root=%llu block=%llu slot=%d chunk_start=%llu, %pV", 928 BTRFS_CHUNK_TREE_OBJECTID, leaf->start, slot, 929 logical, &vaf); 930 va_end(args); 931 } 932 933 static bool valid_stripe_count(u64 profile, u16 num_stripes, u16 sub_stripes) 934 { 935 switch (profile) { 936 case BTRFS_BLOCK_GROUP_RAID0: 937 return true; 938 case BTRFS_BLOCK_GROUP_RAID10: 939 return sub_stripes == btrfs_raid_array[BTRFS_RAID_RAID10].sub_stripes; 940 case BTRFS_BLOCK_GROUP_RAID1: 941 return num_stripes == btrfs_raid_array[BTRFS_RAID_RAID1].devs_min; 942 case BTRFS_BLOCK_GROUP_RAID1C3: 943 return num_stripes == btrfs_raid_array[BTRFS_RAID_RAID1C3].devs_min; 944 case BTRFS_BLOCK_GROUP_RAID1C4: 945 return num_stripes == btrfs_raid_array[BTRFS_RAID_RAID1C4].devs_min; 946 case BTRFS_BLOCK_GROUP_RAID5: 947 return num_stripes >= btrfs_raid_array[BTRFS_RAID_RAID5].devs_min; 948 case BTRFS_BLOCK_GROUP_RAID6: 949 return num_stripes >= btrfs_raid_array[BTRFS_RAID_RAID6].devs_min; 950 case BTRFS_BLOCK_GROUP_DUP: 951 return num_stripes == btrfs_raid_array[BTRFS_RAID_DUP].dev_stripes; 952 case 0: /* SINGLE */ 953 return num_stripes == btrfs_raid_array[BTRFS_RAID_SINGLE].dev_stripes; 954 default: 955 BUG(); 956 } 957 } 958 959 /* 960 * The common chunk check which could also work on super block sys chunk array. 961 * 962 * If @leaf is NULL, then @chunk must be an on-stack chunk item. 963 * (For superblock sys_chunk array, and fs_info->sectorsize is unreliable) 964 * 965 * Return -EUCLEAN if anything is corrupted. 966 * Return 0 if everything is OK. 967 */ 968 int btrfs_check_chunk_valid(const struct btrfs_fs_info *fs_info, 969 const struct extent_buffer *leaf, 970 const struct btrfs_chunk *chunk, u64 logical, 971 u32 sectorsize) 972 { 973 u64 length; 974 u64 chunk_end; 975 u64 stripe_len; 976 u16 num_stripes; 977 u16 sub_stripes; 978 u64 type; 979 u64 features; 980 u32 chunk_sector_size; 981 bool mixed = false; 982 bool remapped; 983 int raid_index; 984 int nparity; 985 int ncopies; 986 987 if (leaf) { 988 length = btrfs_chunk_length(leaf, chunk); 989 stripe_len = btrfs_chunk_stripe_len(leaf, chunk); 990 num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 991 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); 992 type = btrfs_chunk_type(leaf, chunk); 993 chunk_sector_size = btrfs_chunk_sector_size(leaf, chunk); 994 } else { 995 length = btrfs_stack_chunk_length(chunk); 996 stripe_len = btrfs_stack_chunk_stripe_len(chunk); 997 num_stripes = btrfs_stack_chunk_num_stripes(chunk); 998 sub_stripes = btrfs_stack_chunk_sub_stripes(chunk); 999 type = btrfs_stack_chunk_type(chunk); 1000 chunk_sector_size = btrfs_stack_chunk_sector_size(chunk); 1001 } 1002 raid_index = btrfs_bg_flags_to_raid_index(type); 1003 ncopies = btrfs_raid_array[raid_index].ncopies; 1004 nparity = btrfs_raid_array[raid_index].nparity; 1005 remapped = (type & BTRFS_BLOCK_GROUP_REMAPPED); 1006 1007 if (unlikely(!remapped && !num_stripes)) { 1008 chunk_err(fs_info, leaf, chunk, logical, 1009 "invalid chunk num_stripes, have %u", num_stripes); 1010 return -EUCLEAN; 1011 } 1012 if (unlikely(num_stripes != 0 && num_stripes < ncopies)) { 1013 chunk_err(fs_info, leaf, chunk, logical, 1014 "invalid chunk num_stripes < ncopies, have %u < %d", 1015 num_stripes, ncopies); 1016 return -EUCLEAN; 1017 } 1018 if (unlikely(nparity && num_stripes == nparity)) { 1019 chunk_err(fs_info, leaf, chunk, logical, 1020 "invalid chunk num_stripes == nparity, have %u == %d", 1021 num_stripes, nparity); 1022 return -EUCLEAN; 1023 } 1024 if (unlikely(!IS_ALIGNED(logical, sectorsize))) { 1025 chunk_err(fs_info, leaf, chunk, logical, 1026 "invalid chunk logical, have %llu should aligned to %u", 1027 logical, sectorsize); 1028 return -EUCLEAN; 1029 } 1030 if (unlikely(chunk_sector_size != sectorsize)) { 1031 chunk_err(fs_info, leaf, chunk, logical, 1032 "invalid chunk sectorsize, have %u expect %u", 1033 chunk_sector_size, sectorsize); 1034 return -EUCLEAN; 1035 } 1036 if (unlikely(!length || !IS_ALIGNED(length, sectorsize))) { 1037 chunk_err(fs_info, leaf, chunk, logical, 1038 "invalid chunk length, have %llu", length); 1039 return -EUCLEAN; 1040 } 1041 if (unlikely(check_add_overflow(logical, length, &chunk_end))) { 1042 chunk_err(fs_info, leaf, chunk, logical, 1043 "invalid chunk logical start and length, have logical start %llu length %llu", 1044 logical, length); 1045 return -EUCLEAN; 1046 } 1047 if (unlikely(!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN)) { 1048 chunk_err(fs_info, leaf, chunk, logical, 1049 "invalid chunk stripe length: %llu", 1050 stripe_len); 1051 return -EUCLEAN; 1052 } 1053 /* 1054 * We artificially limit the chunk size, so that the number of stripes 1055 * inside a chunk can be fit into a U32. The current limit (256G) is 1056 * way too large for real world usage anyway, and it's also much larger 1057 * than our existing limit (10G). 1058 * 1059 * Thus it should be a good way to catch obvious bitflips. 1060 */ 1061 if (unlikely(length >= btrfs_stripe_nr_to_offset(U32_MAX))) { 1062 chunk_err(fs_info, leaf, chunk, logical, 1063 "chunk length too large: have %llu limit %llu", 1064 length, btrfs_stripe_nr_to_offset(U32_MAX)); 1065 return -EUCLEAN; 1066 } 1067 if (unlikely(type & ~BTRFS_BLOCK_GROUP_VALID)) { 1068 chunk_err(fs_info, leaf, chunk, logical, 1069 "unrecognized chunk type: 0x%llx", 1070 type & ~BTRFS_BLOCK_GROUP_VALID); 1071 return -EUCLEAN; 1072 } 1073 1074 if (unlikely(!has_single_bit_set(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) && 1075 (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) != 0)) { 1076 chunk_err(fs_info, leaf, chunk, logical, 1077 "invalid chunk profile flag: 0x%llx, expect 0 or 1 bit set", 1078 type & BTRFS_BLOCK_GROUP_PROFILE_MASK); 1079 return -EUCLEAN; 1080 } 1081 if (unlikely((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0)) { 1082 chunk_err(fs_info, leaf, chunk, logical, 1083 "missing chunk type flag, have 0x%llx one bit must be set in 0x%llx", 1084 type, BTRFS_BLOCK_GROUP_TYPE_MASK); 1085 return -EUCLEAN; 1086 } 1087 1088 if (unlikely((type & BTRFS_BLOCK_GROUP_SYSTEM) && 1089 (type & (BTRFS_BLOCK_GROUP_METADATA | 1090 BTRFS_BLOCK_GROUP_DATA)))) { 1091 chunk_err(fs_info, leaf, chunk, logical, 1092 "system chunk with data or metadata type: 0x%llx", 1093 type); 1094 return -EUCLEAN; 1095 } 1096 1097 features = btrfs_super_incompat_flags(fs_info->super_copy); 1098 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) 1099 mixed = true; 1100 1101 if (!mixed) { 1102 if (unlikely((type & BTRFS_BLOCK_GROUP_METADATA) && 1103 (type & BTRFS_BLOCK_GROUP_DATA))) { 1104 chunk_err(fs_info, leaf, chunk, logical, 1105 "mixed chunk type in non-mixed mode: 0x%llx", type); 1106 return -EUCLEAN; 1107 } 1108 } 1109 1110 if (unlikely((type & BTRFS_BLOCK_GROUP_METADATA_REMAP) && 1111 !(features & BTRFS_FEATURE_INCOMPAT_REMAP_TREE))) { 1112 chunk_err(fs_info, leaf, chunk, logical, 1113 "METADATA_REMAP chunk type without REMAP_TREE incompat bit"); 1114 return -EUCLEAN; 1115 } 1116 1117 if (unlikely(remapped && 1118 !(features & BTRFS_FEATURE_INCOMPAT_REMAP_TREE))) { 1119 chunk_err(fs_info, leaf, chunk, logical, 1120 "REMAPPED chunk flag without REMAP_TREE incompat bit"); 1121 return -EUCLEAN; 1122 } 1123 1124 if (unlikely(!remapped && 1125 !valid_stripe_count(type & BTRFS_BLOCK_GROUP_PROFILE_MASK, 1126 num_stripes, sub_stripes))) { 1127 chunk_err(fs_info, leaf, chunk, logical, 1128 "invalid num_stripes:sub_stripes %u:%u for profile %llu", 1129 num_stripes, sub_stripes, 1130 type & BTRFS_BLOCK_GROUP_PROFILE_MASK); 1131 return -EUCLEAN; 1132 } 1133 1134 return 0; 1135 } 1136 1137 /* 1138 * Enhanced version of chunk item checker. 1139 * 1140 * The common btrfs_check_chunk_valid() doesn't check item size since it needs 1141 * to work on super block sys_chunk_array which doesn't have full item ptr. 1142 */ 1143 static int check_leaf_chunk_item(struct extent_buffer *leaf, 1144 struct btrfs_chunk *chunk, 1145 struct btrfs_key *key, int slot) 1146 { 1147 struct btrfs_fs_info *fs_info = leaf->fs_info; 1148 int num_stripes; 1149 1150 if (unlikely(btrfs_item_size(leaf, slot) < offsetof(struct btrfs_chunk, stripe))) { 1151 chunk_err(fs_info, leaf, chunk, key->offset, 1152 "invalid chunk item size: have %u expect [%zu, %u)", 1153 btrfs_item_size(leaf, slot), 1154 offsetof(struct btrfs_chunk, stripe), 1155 BTRFS_LEAF_DATA_SIZE(fs_info)); 1156 return -EUCLEAN; 1157 } 1158 1159 num_stripes = btrfs_chunk_num_stripes(leaf, chunk); 1160 /* Let btrfs_check_chunk_valid() handle this error type */ 1161 if (num_stripes == 0) 1162 goto out; 1163 1164 if (unlikely(btrfs_chunk_item_size(num_stripes) != 1165 btrfs_item_size(leaf, slot))) { 1166 chunk_err(fs_info, leaf, chunk, key->offset, 1167 "invalid chunk item size: have %u expect %lu", 1168 btrfs_item_size(leaf, slot), 1169 btrfs_chunk_item_size(num_stripes)); 1170 return -EUCLEAN; 1171 } 1172 out: 1173 return btrfs_check_chunk_valid(fs_info, leaf, chunk, key->offset, 1174 fs_info->sectorsize); 1175 } 1176 1177 __printf(3, 4) 1178 __cold 1179 static void dev_item_err(const struct extent_buffer *eb, int slot, 1180 const char *fmt, ...) 1181 { 1182 struct btrfs_key key; 1183 struct va_format vaf; 1184 va_list args; 1185 1186 btrfs_item_key_to_cpu(eb, &key, slot); 1187 va_start(args, fmt); 1188 1189 vaf.fmt = fmt; 1190 vaf.va = &args; 1191 1192 dump_page(folio_page(eb->folios[0], 0), "eb page dump"); 1193 btrfs_crit(eb->fs_info, 1194 "corrupt %s: root=%llu block=%llu slot=%d devid=%llu %pV", 1195 btrfs_header_level(eb) == 0 ? "leaf" : "node", 1196 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot, 1197 key.objectid, &vaf); 1198 va_end(args); 1199 } 1200 1201 static int check_dev_item(struct extent_buffer *leaf, 1202 struct btrfs_key *key, int slot) 1203 { 1204 struct btrfs_dev_item *ditem; 1205 const u32 item_size = btrfs_item_size(leaf, slot); 1206 1207 if (unlikely(key->objectid != BTRFS_DEV_ITEMS_OBJECTID)) { 1208 dev_item_err(leaf, slot, 1209 "invalid objectid: has=%llu expect=%llu", 1210 key->objectid, BTRFS_DEV_ITEMS_OBJECTID); 1211 return -EUCLEAN; 1212 } 1213 1214 if (unlikely(item_size != sizeof(*ditem))) { 1215 dev_item_err(leaf, slot, "invalid item size: has %u expect %zu", 1216 item_size, sizeof(*ditem)); 1217 return -EUCLEAN; 1218 } 1219 1220 ditem = btrfs_item_ptr(leaf, slot, struct btrfs_dev_item); 1221 if (unlikely(btrfs_device_id(leaf, ditem) != key->offset)) { 1222 dev_item_err(leaf, slot, 1223 "devid mismatch: key has=%llu item has=%llu", 1224 key->offset, btrfs_device_id(leaf, ditem)); 1225 return -EUCLEAN; 1226 } 1227 1228 /* 1229 * For device total_bytes, we don't have reliable way to check it, as 1230 * it can be 0 for device removal. Device size check can only be done 1231 * by dev extents check. 1232 */ 1233 if (unlikely(btrfs_device_bytes_used(leaf, ditem) > 1234 btrfs_device_total_bytes(leaf, ditem))) { 1235 dev_item_err(leaf, slot, 1236 "invalid bytes used: have %llu expect [0, %llu]", 1237 btrfs_device_bytes_used(leaf, ditem), 1238 btrfs_device_total_bytes(leaf, ditem)); 1239 return -EUCLEAN; 1240 } 1241 /* 1242 * Remaining members like io_align/type/gen/dev_group aren't really 1243 * utilized. Skip them to make later usage of them easier. 1244 */ 1245 return 0; 1246 } 1247 1248 static int check_inode_item(struct extent_buffer *leaf, 1249 struct btrfs_key *key, int slot, 1250 struct saved_inode_info *inode_info) 1251 { 1252 struct btrfs_fs_info *fs_info = leaf->fs_info; 1253 struct btrfs_inode_item *iitem; 1254 u64 super_gen = btrfs_super_generation(fs_info->super_copy); 1255 u32 valid_mask = (S_IFMT | S_ISUID | S_ISGID | S_ISVTX | 0777); 1256 const u32 item_size = btrfs_item_size(leaf, slot); 1257 u32 mode; 1258 int ret; 1259 u32 flags; 1260 u32 ro_flags; 1261 1262 ret = check_inode_key(leaf, key, slot); 1263 if (unlikely(ret < 0)) 1264 return ret; 1265 1266 if (unlikely(item_size != sizeof(*iitem))) { 1267 generic_err(leaf, slot, "invalid item size: has %u expect %zu", 1268 item_size, sizeof(*iitem)); 1269 return -EUCLEAN; 1270 } 1271 1272 iitem = btrfs_item_ptr(leaf, slot, struct btrfs_inode_item); 1273 1274 /* Here we use super block generation + 1 to handle log tree */ 1275 if (unlikely(btrfs_inode_generation(leaf, iitem) > super_gen + 1)) { 1276 inode_item_err(leaf, slot, 1277 "invalid inode generation: has %llu expect (0, %llu]", 1278 btrfs_inode_generation(leaf, iitem), 1279 super_gen + 1); 1280 return -EUCLEAN; 1281 } 1282 /* Note for ROOT_TREE_DIR_ITEM, mkfs could set its transid 0 */ 1283 if (unlikely(btrfs_inode_transid(leaf, iitem) > super_gen + 1)) { 1284 inode_item_err(leaf, slot, 1285 "invalid inode transid: has %llu expect [0, %llu]", 1286 btrfs_inode_transid(leaf, iitem), super_gen + 1); 1287 return -EUCLEAN; 1288 } 1289 1290 /* 1291 * For size and nbytes it's better not to be too strict, as for dir 1292 * item its size/nbytes can easily get wrong, but doesn't affect 1293 * anything in the fs. So here we skip the check. 1294 */ 1295 mode = btrfs_inode_mode(leaf, iitem); 1296 if (unlikely(mode & ~valid_mask)) { 1297 inode_item_err(leaf, slot, 1298 "unknown mode bit detected: 0x%x", 1299 mode & ~valid_mask); 1300 return -EUCLEAN; 1301 } 1302 1303 /* 1304 * S_IFMT is not bit mapped so we can't completely rely on 1305 * is_power_of_2/has_single_bit_set, but it can save us from checking 1306 * FIFO/CHR/DIR/REG. Only needs to check BLK, LNK and SOCKS 1307 */ 1308 if (!has_single_bit_set(mode & S_IFMT)) { 1309 if (unlikely(!S_ISLNK(mode) && !S_ISBLK(mode) && !S_ISSOCK(mode))) { 1310 inode_item_err(leaf, slot, 1311 "invalid mode: has 0%o expect valid S_IF* bit(s)", 1312 mode & S_IFMT); 1313 return -EUCLEAN; 1314 } 1315 } 1316 if (unlikely(S_ISDIR(mode) && btrfs_inode_nlink(leaf, iitem) > 1)) { 1317 inode_item_err(leaf, slot, 1318 "invalid nlink: has %u expect no more than 1 for dir", 1319 btrfs_inode_nlink(leaf, iitem)); 1320 return -EUCLEAN; 1321 } 1322 btrfs_inode_split_flags(btrfs_inode_flags(leaf, iitem), &flags, &ro_flags); 1323 if (unlikely(flags & ~BTRFS_INODE_FLAG_MASK)) { 1324 inode_item_err(leaf, slot, 1325 "unknown incompat flags detected: 0x%x", flags); 1326 return -EUCLEAN; 1327 } 1328 if (unlikely(!sb_rdonly(fs_info->sb) && 1329 (ro_flags & ~BTRFS_INODE_RO_FLAG_MASK))) { 1330 inode_item_err(leaf, slot, 1331 "unknown ro-compat flags detected on writeable mount: 0x%x", 1332 ro_flags); 1333 return -EUCLEAN; 1334 } 1335 inode_info->ino = key->objectid; 1336 inode_info->mode = mode; 1337 return 0; 1338 } 1339 1340 static int check_root_item(struct extent_buffer *leaf, struct btrfs_key *key, 1341 int slot) 1342 { 1343 struct btrfs_fs_info *fs_info = leaf->fs_info; 1344 struct btrfs_root_item ri = { 0 }; 1345 const u64 valid_root_flags = BTRFS_ROOT_SUBVOL_RDONLY | 1346 BTRFS_ROOT_SUBVOL_DEAD; 1347 int ret; 1348 1349 ret = check_root_key(leaf, key, slot); 1350 if (unlikely(ret < 0)) 1351 return ret; 1352 1353 if (unlikely(btrfs_item_size(leaf, slot) != sizeof(ri) && 1354 btrfs_item_size(leaf, slot) != 1355 btrfs_legacy_root_item_size())) { 1356 generic_err(leaf, slot, 1357 "invalid root item size, have %u expect %zu or %u", 1358 btrfs_item_size(leaf, slot), sizeof(ri), 1359 btrfs_legacy_root_item_size()); 1360 return -EUCLEAN; 1361 } 1362 1363 /* 1364 * For legacy root item, the members starting at generation_v2 will be 1365 * all filled with 0. 1366 * And since we allow generation_v2 as 0, it will still pass the check. 1367 */ 1368 read_extent_buffer(leaf, &ri, btrfs_item_ptr_offset(leaf, slot), 1369 btrfs_item_size(leaf, slot)); 1370 1371 /* Generation related */ 1372 if (unlikely(btrfs_root_generation(&ri) > 1373 btrfs_super_generation(fs_info->super_copy) + 1)) { 1374 generic_err(leaf, slot, 1375 "invalid root generation, have %llu expect (0, %llu]", 1376 btrfs_root_generation(&ri), 1377 btrfs_super_generation(fs_info->super_copy) + 1); 1378 return -EUCLEAN; 1379 } 1380 if (unlikely(btrfs_root_generation_v2(&ri) > 1381 btrfs_super_generation(fs_info->super_copy) + 1)) { 1382 generic_err(leaf, slot, 1383 "invalid root v2 generation, have %llu expect (0, %llu]", 1384 btrfs_root_generation_v2(&ri), 1385 btrfs_super_generation(fs_info->super_copy) + 1); 1386 return -EUCLEAN; 1387 } 1388 if (unlikely(btrfs_root_last_snapshot(&ri) > 1389 btrfs_super_generation(fs_info->super_copy) + 1)) { 1390 generic_err(leaf, slot, 1391 "invalid root last_snapshot, have %llu expect (0, %llu]", 1392 btrfs_root_last_snapshot(&ri), 1393 btrfs_super_generation(fs_info->super_copy) + 1); 1394 return -EUCLEAN; 1395 } 1396 1397 /* Alignment and level check */ 1398 if (unlikely(!IS_ALIGNED(btrfs_root_bytenr(&ri), fs_info->sectorsize))) { 1399 generic_err(leaf, slot, 1400 "invalid root bytenr, have %llu expect to be aligned to %u", 1401 btrfs_root_bytenr(&ri), fs_info->sectorsize); 1402 return -EUCLEAN; 1403 } 1404 if (unlikely(btrfs_root_level(&ri) >= BTRFS_MAX_LEVEL)) { 1405 generic_err(leaf, slot, 1406 "invalid root level, have %u expect [0, %u]", 1407 btrfs_root_level(&ri), BTRFS_MAX_LEVEL - 1); 1408 return -EUCLEAN; 1409 } 1410 if (unlikely(btrfs_root_drop_level(&ri) >= BTRFS_MAX_LEVEL)) { 1411 generic_err(leaf, slot, 1412 "invalid root drop_level, have %u expect [0, %u]", 1413 btrfs_root_drop_level(&ri), BTRFS_MAX_LEVEL - 1); 1414 return -EUCLEAN; 1415 } 1416 /* 1417 * If drop_progress.objectid is non-zero, a btrfs_drop_snapshot() was 1418 * interrupted and the resume point was recorded in drop_progress and 1419 * drop_level. In that case drop_level must be >= 1: level 0 is the 1420 * leaf level and drop_snapshot never saves a checkpoint there (it 1421 * only records checkpoints at internal node levels in DROP_REFERENCE 1422 * stage). A zero drop_level combined with a non-zero drop_progress 1423 * objectid indicates on-disk corruption and would cause a BUG_ON in 1424 * merge_reloc_root() and btrfs_drop_snapshot() at mount time. 1425 */ 1426 if (unlikely(btrfs_disk_key_objectid(&ri.drop_progress) != 0 && 1427 btrfs_root_drop_level(&ri) == 0)) { 1428 generic_err(leaf, slot, 1429 "invalid root drop_level 0 with non-zero drop_progress objectid %llu", 1430 btrfs_disk_key_objectid(&ri.drop_progress)); 1431 return -EUCLEAN; 1432 } 1433 1434 /* Flags check */ 1435 if (unlikely(btrfs_root_flags(&ri) & ~valid_root_flags)) { 1436 generic_err(leaf, slot, 1437 "invalid root flags, have 0x%llx expect mask 0x%llx", 1438 btrfs_root_flags(&ri), valid_root_flags); 1439 return -EUCLEAN; 1440 } 1441 return 0; 1442 } 1443 1444 static int check_root_ref(struct extent_buffer *leaf, struct btrfs_key *key, int slot) 1445 { 1446 struct btrfs_root_ref *rref; 1447 u32 item_size = btrfs_item_size(leaf, slot); 1448 u32 name_len; 1449 1450 if (unlikely(item_size <= sizeof(*rref))) { 1451 generic_err(leaf, slot, 1452 "invalid root ref item size for key type %u, have %u expect > %zu", 1453 key->type, item_size, sizeof(*rref)); 1454 return -EUCLEAN; 1455 } 1456 1457 rref = btrfs_item_ptr(leaf, slot, struct btrfs_root_ref); 1458 name_len = btrfs_root_ref_name_len(leaf, rref); 1459 if (unlikely(name_len > BTRFS_NAME_LEN)) { 1460 generic_err(leaf, slot, 1461 "root ref name too long for key type %u, have %u max %u", 1462 key->type, name_len, BTRFS_NAME_LEN); 1463 return -EUCLEAN; 1464 } 1465 if (unlikely(item_size != sizeof(*rref) + name_len)) { 1466 generic_err(leaf, slot, 1467 "invalid root ref item size for key type %u, have %u expect %zu", 1468 key->type, item_size, sizeof(*rref) + name_len); 1469 return -EUCLEAN; 1470 } 1471 1472 return 0; 1473 } 1474 1475 __printf(3,4) 1476 __cold 1477 static void extent_err(const struct extent_buffer *eb, int slot, 1478 const char *fmt, ...) 1479 { 1480 struct btrfs_key key; 1481 struct va_format vaf; 1482 va_list args; 1483 u64 bytenr; 1484 u64 len; 1485 1486 btrfs_item_key_to_cpu(eb, &key, slot); 1487 bytenr = key.objectid; 1488 if (key.type == BTRFS_METADATA_ITEM_KEY || 1489 key.type == BTRFS_TREE_BLOCK_REF_KEY || 1490 key.type == BTRFS_SHARED_BLOCK_REF_KEY) 1491 len = eb->fs_info->nodesize; 1492 else 1493 len = key.offset; 1494 va_start(args, fmt); 1495 1496 vaf.fmt = fmt; 1497 vaf.va = &args; 1498 1499 dump_page(folio_page(eb->folios[0], 0), "eb page dump"); 1500 btrfs_crit(eb->fs_info, 1501 "corrupt %s: block=%llu slot=%d extent bytenr=%llu len=%llu %pV", 1502 btrfs_header_level(eb) == 0 ? "leaf" : "node", 1503 eb->start, slot, bytenr, len, &vaf); 1504 va_end(args); 1505 } 1506 1507 static bool is_valid_dref_root(u64 rootid) 1508 { 1509 /* 1510 * The following tree root objectids are allowed to have a data backref: 1511 * - subvolume trees 1512 * - data reloc tree 1513 * - tree root 1514 * For v1 space cache 1515 */ 1516 return btrfs_is_fstree(rootid) || rootid == BTRFS_DATA_RELOC_TREE_OBJECTID || 1517 rootid == BTRFS_ROOT_TREE_OBJECTID; 1518 } 1519 1520 static int check_extent_item(struct extent_buffer *leaf, 1521 struct btrfs_key *key, int slot, 1522 struct btrfs_key *prev_key) 1523 { 1524 struct btrfs_fs_info *fs_info = leaf->fs_info; 1525 struct btrfs_extent_item *ei; 1526 bool is_tree_block = false; 1527 unsigned long ptr; /* Current pointer inside inline refs */ 1528 unsigned long end; /* Extent item end */ 1529 const u32 item_size = btrfs_item_size(leaf, slot); 1530 u8 last_type = 0; 1531 u64 last_seq = U64_MAX; 1532 u64 flags; 1533 u64 generation; 1534 u64 total_refs; /* Total refs in btrfs_extent_item */ 1535 u64 inline_refs = 0; /* found total inline refs */ 1536 1537 if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY && 1538 !btrfs_fs_incompat(fs_info, SKINNY_METADATA))) { 1539 generic_err(leaf, slot, 1540 "invalid key type, METADATA_ITEM type invalid when SKINNY_METADATA feature disabled"); 1541 return -EUCLEAN; 1542 } 1543 /* key->objectid is the bytenr for both key types */ 1544 if (unlikely(!IS_ALIGNED(key->objectid, fs_info->sectorsize))) { 1545 generic_err(leaf, slot, 1546 "invalid key objectid, have %llu expect to be aligned to %u", 1547 key->objectid, fs_info->sectorsize); 1548 return -EUCLEAN; 1549 } 1550 1551 /* key->offset is tree level for METADATA_ITEM_KEY */ 1552 if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY && 1553 key->offset >= BTRFS_MAX_LEVEL)) { 1554 extent_err(leaf, slot, 1555 "invalid tree level, have %llu expect [0, %u]", 1556 key->offset, BTRFS_MAX_LEVEL - 1); 1557 return -EUCLEAN; 1558 } 1559 1560 /* 1561 * EXTENT/METADATA_ITEM consists of: 1562 * 1) One btrfs_extent_item 1563 * Records the total refs, type and generation of the extent. 1564 * 1565 * 2) One btrfs_tree_block_info (for EXTENT_ITEM and tree backref only) 1566 * Records the first key and level of the tree block. 1567 * 1568 * 2) Zero or more btrfs_extent_inline_ref(s) 1569 * Each inline ref has one btrfs_extent_inline_ref shows: 1570 * 2.1) The ref type, one of the 4 1571 * TREE_BLOCK_REF Tree block only 1572 * SHARED_BLOCK_REF Tree block only 1573 * EXTENT_DATA_REF Data only 1574 * SHARED_DATA_REF Data only 1575 * 2.2) Ref type specific data 1576 * Either using btrfs_extent_inline_ref::offset, or specific 1577 * data structure. 1578 * 1579 * All above inline items should follow the order: 1580 * 1581 * - All btrfs_extent_inline_ref::type should be in an ascending 1582 * order 1583 * 1584 * - Within the same type, the items should follow a descending 1585 * order by their sequence number. The sequence number is 1586 * determined by: 1587 * * btrfs_extent_inline_ref::offset for all types other than 1588 * EXTENT_DATA_REF 1589 * * hash_extent_data_ref() for EXTENT_DATA_REF 1590 */ 1591 if (unlikely(item_size < sizeof(*ei))) { 1592 extent_err(leaf, slot, 1593 "invalid item size, have %u expect [%zu, %u)", 1594 item_size, sizeof(*ei), 1595 BTRFS_LEAF_DATA_SIZE(fs_info)); 1596 return -EUCLEAN; 1597 } 1598 end = item_size + btrfs_item_ptr_offset(leaf, slot); 1599 1600 /* Checks against extent_item */ 1601 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); 1602 flags = btrfs_extent_flags(leaf, ei); 1603 total_refs = btrfs_extent_refs(leaf, ei); 1604 generation = btrfs_extent_generation(leaf, ei); 1605 if (unlikely(generation > 1606 btrfs_super_generation(fs_info->super_copy) + 1)) { 1607 extent_err(leaf, slot, 1608 "invalid generation, have %llu expect (0, %llu]", 1609 generation, 1610 btrfs_super_generation(fs_info->super_copy) + 1); 1611 return -EUCLEAN; 1612 } 1613 if (unlikely(!has_single_bit_set(flags & (BTRFS_EXTENT_FLAG_DATA | 1614 BTRFS_EXTENT_FLAG_TREE_BLOCK)))) { 1615 extent_err(leaf, slot, 1616 "invalid extent flag, have 0x%llx expect 1 bit set in 0x%llx", 1617 flags, BTRFS_EXTENT_FLAG_DATA | 1618 BTRFS_EXTENT_FLAG_TREE_BLOCK); 1619 return -EUCLEAN; 1620 } 1621 is_tree_block = !!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK); 1622 if (is_tree_block) { 1623 if (unlikely(key->type == BTRFS_EXTENT_ITEM_KEY && 1624 key->offset != fs_info->nodesize)) { 1625 extent_err(leaf, slot, 1626 "invalid extent length, have %llu expect %u", 1627 key->offset, fs_info->nodesize); 1628 return -EUCLEAN; 1629 } 1630 } else { 1631 if (unlikely(key->type != BTRFS_EXTENT_ITEM_KEY)) { 1632 extent_err(leaf, slot, 1633 "invalid key type, have %u expect %u for data backref", 1634 key->type, BTRFS_EXTENT_ITEM_KEY); 1635 return -EUCLEAN; 1636 } 1637 if (unlikely(!IS_ALIGNED(key->offset, fs_info->sectorsize))) { 1638 extent_err(leaf, slot, 1639 "invalid extent length, have %llu expect aligned to %u", 1640 key->offset, fs_info->sectorsize); 1641 return -EUCLEAN; 1642 } 1643 if (unlikely(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { 1644 extent_err(leaf, slot, 1645 "invalid extent flag, data has full backref set"); 1646 return -EUCLEAN; 1647 } 1648 } 1649 ptr = (unsigned long)(struct btrfs_extent_item *)(ei + 1); 1650 1651 /* Check the special case of btrfs_tree_block_info */ 1652 if (is_tree_block && key->type != BTRFS_METADATA_ITEM_KEY) { 1653 struct btrfs_tree_block_info *info; 1654 1655 info = (struct btrfs_tree_block_info *)ptr; 1656 if (unlikely(btrfs_tree_block_level(leaf, info) >= BTRFS_MAX_LEVEL)) { 1657 extent_err(leaf, slot, 1658 "invalid tree block info level, have %u expect [0, %u]", 1659 btrfs_tree_block_level(leaf, info), 1660 BTRFS_MAX_LEVEL - 1); 1661 return -EUCLEAN; 1662 } 1663 ptr = (unsigned long)(struct btrfs_tree_block_info *)(info + 1); 1664 } 1665 1666 /* Check inline refs */ 1667 while (ptr < end) { 1668 struct btrfs_extent_inline_ref *iref; 1669 struct btrfs_extent_data_ref *dref; 1670 struct btrfs_shared_data_ref *sref; 1671 u64 seq; 1672 u64 dref_root; 1673 u64 dref_objectid; 1674 u64 dref_offset; 1675 u64 inline_offset; 1676 u8 inline_type; 1677 1678 if (unlikely(ptr + sizeof(*iref) > end)) { 1679 extent_err(leaf, slot, 1680 "inline ref item overflows extent item, ptr %lu iref size %zu end %lu", 1681 ptr, sizeof(*iref), end); 1682 return -EUCLEAN; 1683 } 1684 iref = (struct btrfs_extent_inline_ref *)ptr; 1685 inline_type = btrfs_extent_inline_ref_type(leaf, iref); 1686 inline_offset = btrfs_extent_inline_ref_offset(leaf, iref); 1687 seq = inline_offset; 1688 if (unlikely(ptr + btrfs_extent_inline_ref_size(inline_type) > end)) { 1689 extent_err(leaf, slot, 1690 "inline ref item overflows extent item, ptr %lu iref size %u end %lu", 1691 ptr, btrfs_extent_inline_ref_size(inline_type), end); 1692 return -EUCLEAN; 1693 } 1694 1695 switch (inline_type) { 1696 /* inline_offset is subvolid of the owner, no need to check */ 1697 case BTRFS_TREE_BLOCK_REF_KEY: 1698 inline_refs++; 1699 break; 1700 /* Contains parent bytenr */ 1701 case BTRFS_SHARED_BLOCK_REF_KEY: 1702 if (unlikely(!IS_ALIGNED(inline_offset, 1703 fs_info->sectorsize))) { 1704 extent_err(leaf, slot, 1705 "invalid tree parent bytenr, have %llu expect aligned to %u", 1706 inline_offset, fs_info->sectorsize); 1707 return -EUCLEAN; 1708 } 1709 inline_refs++; 1710 break; 1711 /* 1712 * Contains owner subvolid, owner key objectid, adjusted offset. 1713 * The only obvious corruption can happen in that offset. 1714 */ 1715 case BTRFS_EXTENT_DATA_REF_KEY: 1716 dref = (struct btrfs_extent_data_ref *)(&iref->offset); 1717 dref_root = btrfs_extent_data_ref_root(leaf, dref); 1718 dref_objectid = btrfs_extent_data_ref_objectid(leaf, dref); 1719 dref_offset = btrfs_extent_data_ref_offset(leaf, dref); 1720 seq = hash_extent_data_ref( 1721 btrfs_extent_data_ref_root(leaf, dref), 1722 btrfs_extent_data_ref_objectid(leaf, dref), 1723 btrfs_extent_data_ref_offset(leaf, dref)); 1724 if (unlikely(!is_valid_dref_root(dref_root))) { 1725 extent_err(leaf, slot, 1726 "invalid data ref root value %llu", 1727 dref_root); 1728 return -EUCLEAN; 1729 } 1730 if (unlikely(dref_objectid < BTRFS_FIRST_FREE_OBJECTID || 1731 dref_objectid > BTRFS_LAST_FREE_OBJECTID)) { 1732 extent_err(leaf, slot, 1733 "invalid data ref objectid value %llu", 1734 dref_objectid); 1735 return -EUCLEAN; 1736 } 1737 if (unlikely(!IS_ALIGNED(dref_offset, 1738 fs_info->sectorsize))) { 1739 extent_err(leaf, slot, 1740 "invalid data ref offset, have %llu expect aligned to %u", 1741 dref_offset, fs_info->sectorsize); 1742 return -EUCLEAN; 1743 } 1744 if (unlikely(btrfs_extent_data_ref_count(leaf, dref) == 0)) { 1745 extent_err(leaf, slot, 1746 "invalid data ref count, should have non-zero value"); 1747 return -EUCLEAN; 1748 } 1749 inline_refs += btrfs_extent_data_ref_count(leaf, dref); 1750 break; 1751 /* Contains parent bytenr and ref count */ 1752 case BTRFS_SHARED_DATA_REF_KEY: 1753 sref = (struct btrfs_shared_data_ref *)(iref + 1); 1754 if (unlikely(!IS_ALIGNED(inline_offset, 1755 fs_info->sectorsize))) { 1756 extent_err(leaf, slot, 1757 "invalid data parent bytenr, have %llu expect aligned to %u", 1758 inline_offset, fs_info->sectorsize); 1759 return -EUCLEAN; 1760 } 1761 if (unlikely(btrfs_shared_data_ref_count(leaf, sref) == 0)) { 1762 extent_err(leaf, slot, 1763 "invalid shared data ref count, should have non-zero value"); 1764 return -EUCLEAN; 1765 } 1766 inline_refs += btrfs_shared_data_ref_count(leaf, sref); 1767 break; 1768 case BTRFS_EXTENT_OWNER_REF_KEY: 1769 WARN_ON(!btrfs_fs_incompat(fs_info, SIMPLE_QUOTA)); 1770 break; 1771 default: 1772 extent_err(leaf, slot, "unknown inline ref type: %u", 1773 inline_type); 1774 return -EUCLEAN; 1775 } 1776 if (unlikely(inline_type < last_type)) { 1777 extent_err(leaf, slot, 1778 "inline ref out-of-order: has type %u, prev type %u", 1779 inline_type, last_type); 1780 return -EUCLEAN; 1781 } 1782 /* Type changed, allow the sequence starts from U64_MAX again. */ 1783 if (inline_type > last_type) 1784 last_seq = U64_MAX; 1785 if (unlikely(seq > last_seq)) { 1786 extent_err(leaf, slot, 1787 "inline ref out-of-order: has type %u offset %llu seq 0x%llx, prev type %u seq 0x%llx", 1788 inline_type, inline_offset, seq, 1789 last_type, last_seq); 1790 return -EUCLEAN; 1791 } 1792 last_type = inline_type; 1793 last_seq = seq; 1794 ptr += btrfs_extent_inline_ref_size(inline_type); 1795 } 1796 /* No padding is allowed */ 1797 if (unlikely(ptr != end)) { 1798 extent_err(leaf, slot, 1799 "invalid extent item size, padding bytes found"); 1800 return -EUCLEAN; 1801 } 1802 1803 /* Finally, check the inline refs against total refs */ 1804 if (unlikely(inline_refs > total_refs)) { 1805 extent_err(leaf, slot, 1806 "invalid extent refs, have %llu expect >= inline %llu", 1807 total_refs, inline_refs); 1808 return -EUCLEAN; 1809 } 1810 1811 if ((prev_key->type == BTRFS_EXTENT_ITEM_KEY) || 1812 (prev_key->type == BTRFS_METADATA_ITEM_KEY)) { 1813 u64 prev_end = prev_key->objectid; 1814 1815 if (prev_key->type == BTRFS_METADATA_ITEM_KEY) 1816 prev_end += fs_info->nodesize; 1817 else 1818 prev_end += prev_key->offset; 1819 1820 if (unlikely(prev_end > key->objectid)) { 1821 extent_err(leaf, slot, 1822 "previous extent " BTRFS_KEY_FMT " overlaps current extent " BTRFS_KEY_FMT, 1823 BTRFS_KEY_FMT_VALUE(prev_key), 1824 BTRFS_KEY_FMT_VALUE(key)); 1825 return -EUCLEAN; 1826 } 1827 } 1828 1829 return 0; 1830 } 1831 1832 static int check_simple_keyed_refs(struct extent_buffer *leaf, 1833 struct btrfs_key *key, int slot) 1834 { 1835 u32 expect_item_size = 0; 1836 1837 if (key->type == BTRFS_SHARED_DATA_REF_KEY) { 1838 struct btrfs_shared_data_ref *sref; 1839 1840 sref = btrfs_item_ptr(leaf, slot, struct btrfs_shared_data_ref); 1841 if (unlikely(btrfs_shared_data_ref_count(leaf, sref) == 0)) { 1842 extent_err(leaf, slot, 1843 "invalid shared data backref count, should have non-zero value"); 1844 return -EUCLEAN; 1845 } 1846 1847 expect_item_size = sizeof(struct btrfs_shared_data_ref); 1848 } 1849 1850 if (unlikely(btrfs_item_size(leaf, slot) != expect_item_size)) { 1851 generic_err(leaf, slot, 1852 "invalid item size, have %u expect %u for key type %u", 1853 btrfs_item_size(leaf, slot), 1854 expect_item_size, key->type); 1855 return -EUCLEAN; 1856 } 1857 if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) { 1858 generic_err(leaf, slot, 1859 "invalid key objectid for shared block ref, have %llu expect aligned to %u", 1860 key->objectid, leaf->fs_info->sectorsize); 1861 return -EUCLEAN; 1862 } 1863 if (unlikely(key->type != BTRFS_TREE_BLOCK_REF_KEY && 1864 !IS_ALIGNED(key->offset, leaf->fs_info->sectorsize))) { 1865 extent_err(leaf, slot, 1866 "invalid tree parent bytenr, have %llu expect aligned to %u", 1867 key->offset, leaf->fs_info->sectorsize); 1868 return -EUCLEAN; 1869 } 1870 return 0; 1871 } 1872 1873 static int check_extent_data_ref(struct extent_buffer *leaf, 1874 struct btrfs_key *key, int slot) 1875 { 1876 struct btrfs_extent_data_ref *dref; 1877 unsigned long ptr = btrfs_item_ptr_offset(leaf, slot); 1878 const unsigned long end = ptr + btrfs_item_size(leaf, slot); 1879 1880 if (unlikely(btrfs_item_size(leaf, slot) % sizeof(*dref) != 0)) { 1881 generic_err(leaf, slot, 1882 "invalid item size, have %u expect aligned to %zu for key type %u", 1883 btrfs_item_size(leaf, slot), 1884 sizeof(*dref), key->type); 1885 return -EUCLEAN; 1886 } 1887 if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) { 1888 generic_err(leaf, slot, 1889 "invalid key objectid for shared block ref, have %llu expect aligned to %u", 1890 key->objectid, leaf->fs_info->sectorsize); 1891 return -EUCLEAN; 1892 } 1893 for (; ptr < end; ptr += sizeof(*dref)) { 1894 u64 root; 1895 u64 objectid; 1896 u64 offset; 1897 1898 /* 1899 * We cannot check the extent_data_ref hash due to possible 1900 * overflow from the leaf due to hash collisions. 1901 */ 1902 dref = (struct btrfs_extent_data_ref *)ptr; 1903 root = btrfs_extent_data_ref_root(leaf, dref); 1904 objectid = btrfs_extent_data_ref_objectid(leaf, dref); 1905 offset = btrfs_extent_data_ref_offset(leaf, dref); 1906 if (unlikely(!is_valid_dref_root(root))) { 1907 extent_err(leaf, slot, 1908 "invalid extent data backref root value %llu", 1909 root); 1910 return -EUCLEAN; 1911 } 1912 if (unlikely(objectid < BTRFS_FIRST_FREE_OBJECTID || 1913 objectid > BTRFS_LAST_FREE_OBJECTID)) { 1914 extent_err(leaf, slot, 1915 "invalid extent data backref objectid value %llu", 1916 objectid); 1917 return -EUCLEAN; 1918 } 1919 if (unlikely(!IS_ALIGNED(offset, leaf->fs_info->sectorsize))) { 1920 extent_err(leaf, slot, 1921 "invalid extent data backref offset, have %llu expect aligned to %u", 1922 offset, leaf->fs_info->sectorsize); 1923 return -EUCLEAN; 1924 } 1925 if (unlikely(btrfs_extent_data_ref_count(leaf, dref) == 0)) { 1926 extent_err(leaf, slot, 1927 "invalid extent data backref count, should have non-zero value"); 1928 return -EUCLEAN; 1929 } 1930 } 1931 return 0; 1932 } 1933 1934 #define inode_ref_err(eb, slot, fmt, args...) \ 1935 inode_item_err(eb, slot, fmt, ##args) 1936 static int check_inode_ref(struct extent_buffer *leaf, 1937 struct btrfs_key *key, struct btrfs_key *prev_key, 1938 int slot) 1939 { 1940 struct btrfs_inode_ref *iref; 1941 unsigned long ptr; 1942 unsigned long end; 1943 1944 if (unlikely(!check_prev_ino(leaf, key, slot, prev_key))) 1945 return -EUCLEAN; 1946 /* namelen can't be 0, so item_size == sizeof() is also invalid */ 1947 if (unlikely(btrfs_item_size(leaf, slot) <= sizeof(*iref))) { 1948 inode_ref_err(leaf, slot, 1949 "invalid item size, have %u expect (%zu, %u)", 1950 btrfs_item_size(leaf, slot), 1951 sizeof(*iref), BTRFS_LEAF_DATA_SIZE(leaf->fs_info)); 1952 return -EUCLEAN; 1953 } 1954 1955 if (unlikely(btrfs_is_fstree(btrfs_header_owner(leaf)) && 1956 (key->offset < BTRFS_FIRST_FREE_OBJECTID || 1957 key->offset > BTRFS_LAST_FREE_OBJECTID))) { 1958 inode_ref_err(leaf, slot, 1959 "invalid offset for ref key, have %llu expect [%llu, %lld]", 1960 key->offset, BTRFS_FIRST_FREE_OBJECTID, 1961 BTRFS_LAST_FREE_OBJECTID); 1962 return -EUCLEAN; 1963 } 1964 1965 ptr = btrfs_item_ptr_offset(leaf, slot); 1966 end = ptr + btrfs_item_size(leaf, slot); 1967 while (ptr < end) { 1968 u16 namelen; 1969 1970 if (unlikely(ptr + sizeof(*iref) > end)) { 1971 inode_ref_err(leaf, slot, 1972 "inode ref overflow, ptr %lu end %lu inode_ref_size %zu", 1973 ptr, end, sizeof(*iref)); 1974 return -EUCLEAN; 1975 } 1976 1977 iref = (struct btrfs_inode_ref *)ptr; 1978 namelen = btrfs_inode_ref_name_len(leaf, iref); 1979 if (unlikely(namelen == 0 || namelen > BTRFS_NAME_LEN)) { 1980 inode_ref_err(leaf, slot, 1981 "invalid inode ref name length, has %u expect [1, %u]", 1982 namelen, BTRFS_NAME_LEN); 1983 return -EUCLEAN; 1984 } 1985 if (unlikely(ptr + sizeof(*iref) + namelen > end)) { 1986 inode_ref_err(leaf, slot, 1987 "inode ref overflow, ptr %lu end %lu namelen %u", 1988 ptr, end, namelen); 1989 return -EUCLEAN; 1990 } 1991 1992 /* 1993 * NOTE: In theory we should record all found index numbers 1994 * to find any duplicated indexes, but that will be too time 1995 * consuming for inodes with too many hard links. 1996 */ 1997 ptr += sizeof(*iref) + namelen; 1998 } 1999 return 0; 2000 } 2001 2002 static int check_inode_extref(struct extent_buffer *leaf, 2003 struct btrfs_key *key, struct btrfs_key *prev_key, 2004 int slot) 2005 { 2006 unsigned long ptr = btrfs_item_ptr_offset(leaf, slot); 2007 unsigned long end = ptr + btrfs_item_size(leaf, slot); 2008 const bool is_fstree = btrfs_is_fstree(btrfs_header_owner(leaf)); 2009 2010 if (unlikely(!check_prev_ino(leaf, key, slot, prev_key))) 2011 return -EUCLEAN; 2012 2013 while (ptr < end) { 2014 struct btrfs_inode_extref *extref = (struct btrfs_inode_extref *)ptr; 2015 u64 parent; 2016 u16 namelen; 2017 2018 if (unlikely(ptr + sizeof(*extref) > end)) { 2019 inode_ref_err(leaf, slot, 2020 "inode extref overflow, ptr %lu end %lu inode_extref size %zu", 2021 ptr, end, sizeof(*extref)); 2022 return -EUCLEAN; 2023 } 2024 2025 parent = btrfs_inode_extref_parent(leaf, extref); 2026 if (unlikely(is_fstree && (parent < BTRFS_FIRST_FREE_OBJECTID || 2027 parent > BTRFS_LAST_FREE_OBJECTID))) { 2028 inode_ref_err(leaf, slot, 2029 "invalid parent for extref key, have %llu expect [%llu, %lld]", 2030 parent, BTRFS_FIRST_FREE_OBJECTID, 2031 BTRFS_LAST_FREE_OBJECTID); 2032 return -EUCLEAN; 2033 } 2034 2035 namelen = btrfs_inode_extref_name_len(leaf, extref); 2036 if (unlikely(namelen == 0 || namelen > BTRFS_NAME_LEN)) { 2037 inode_ref_err(leaf, slot, 2038 "invalid inode extref name length, has %u expect [1, %u]", 2039 namelen, BTRFS_NAME_LEN); 2040 return -EUCLEAN; 2041 } 2042 2043 if (unlikely(ptr + sizeof(*extref) + namelen > end)) { 2044 inode_ref_err(leaf, slot, 2045 "inode extref overflow, ptr %lu end %lu namelen %u", 2046 ptr, end, namelen); 2047 return -EUCLEAN; 2048 } 2049 ptr += sizeof(*extref) + namelen; 2050 } 2051 return 0; 2052 } 2053 2054 static int check_raid_stripe_extent(const struct extent_buffer *leaf, 2055 const struct btrfs_key *key, int slot) 2056 { 2057 if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) { 2058 generic_err(leaf, slot, 2059 "invalid key objectid for raid stripe extent, have %llu expect aligned to %u", 2060 key->objectid, leaf->fs_info->sectorsize); 2061 return -EUCLEAN; 2062 } 2063 2064 if (unlikely(!btrfs_fs_incompat(leaf->fs_info, RAID_STRIPE_TREE))) { 2065 generic_err(leaf, slot, 2066 "RAID_STRIPE_EXTENT present but RAID_STRIPE_TREE incompat bit unset"); 2067 return -EUCLEAN; 2068 } 2069 2070 return 0; 2071 } 2072 2073 static int check_remap_key(const struct extent_buffer *leaf, 2074 const struct btrfs_key *key, int slot) 2075 { 2076 const u32 item_size = btrfs_item_size(leaf, slot); 2077 const u32 sectorsize = leaf->fs_info->sectorsize; 2078 u64 end; 2079 2080 if (unlikely(!btrfs_fs_incompat(leaf->fs_info, REMAP_TREE))) { 2081 generic_err(leaf, slot, 2082 "remap key type %u present but REMAP_TREE incompat bit unset", 2083 key->type); 2084 return -EUCLEAN; 2085 } 2086 2087 switch (key->type) { 2088 case BTRFS_IDENTITY_REMAP_KEY: 2089 if (unlikely(item_size != 0)) { 2090 generic_err(leaf, slot, 2091 "invalid item size for IDENTITY_REMAP, have %u expect 0", 2092 item_size); 2093 return -EUCLEAN; 2094 } 2095 break; 2096 case BTRFS_REMAP_KEY: 2097 case BTRFS_REMAP_BACKREF_KEY: 2098 if (unlikely(item_size != sizeof(struct btrfs_remap_item))) { 2099 generic_err(leaf, slot, 2100 "invalid item size for remap key type %u, have %u expect %zu", 2101 key->type, item_size, 2102 sizeof(struct btrfs_remap_item)); 2103 return -EUCLEAN; 2104 } 2105 break; 2106 } 2107 2108 if (unlikely(key->offset == 0)) { 2109 generic_err(leaf, slot, 2110 "invalid remap key length, have 0 expect nonzero"); 2111 return -EUCLEAN; 2112 } 2113 2114 if (unlikely(!IS_ALIGNED(key->objectid, sectorsize))) { 2115 generic_err(leaf, slot, 2116 "invalid remap key objectid, have %llu expect aligned to %u", 2117 key->objectid, sectorsize); 2118 return -EUCLEAN; 2119 } 2120 2121 if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) { 2122 generic_err(leaf, slot, 2123 "invalid remap key offset (length), have %llu expect aligned to %u", 2124 key->offset, sectorsize); 2125 return -EUCLEAN; 2126 } 2127 2128 if (unlikely(check_add_overflow(key->objectid, key->offset, &end))) { 2129 generic_err(leaf, slot, 2130 "remap key overflow, objectid %llu + offset %llu wraps", 2131 key->objectid, key->offset); 2132 return -EUCLEAN; 2133 } 2134 2135 return 0; 2136 } 2137 2138 static int check_dev_extent_item(const struct extent_buffer *leaf, 2139 const struct btrfs_key *key, 2140 int slot, 2141 struct btrfs_key *prev_key) 2142 { 2143 struct btrfs_dev_extent *de; 2144 const u32 sectorsize = leaf->fs_info->sectorsize; 2145 2146 de = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); 2147 /* Basic fixed member checks. */ 2148 if (unlikely(btrfs_dev_extent_chunk_tree(leaf, de) != 2149 BTRFS_CHUNK_TREE_OBJECTID)) { 2150 generic_err(leaf, slot, 2151 "invalid dev extent chunk tree id, has %llu expect %llu", 2152 btrfs_dev_extent_chunk_tree(leaf, de), 2153 BTRFS_CHUNK_TREE_OBJECTID); 2154 return -EUCLEAN; 2155 } 2156 if (unlikely(btrfs_dev_extent_chunk_objectid(leaf, de) != 2157 BTRFS_FIRST_CHUNK_TREE_OBJECTID)) { 2158 generic_err(leaf, slot, 2159 "invalid dev extent chunk objectid, has %llu expect %llu", 2160 btrfs_dev_extent_chunk_objectid(leaf, de), 2161 BTRFS_FIRST_CHUNK_TREE_OBJECTID); 2162 return -EUCLEAN; 2163 } 2164 /* Alignment check. */ 2165 if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) { 2166 generic_err(leaf, slot, 2167 "invalid dev extent key.offset, has %llu not aligned to %u", 2168 key->offset, sectorsize); 2169 return -EUCLEAN; 2170 } 2171 if (unlikely(!IS_ALIGNED(btrfs_dev_extent_chunk_offset(leaf, de), 2172 sectorsize))) { 2173 generic_err(leaf, slot, 2174 "invalid dev extent chunk offset, has %llu not aligned to %u", 2175 btrfs_dev_extent_chunk_offset(leaf, de), 2176 sectorsize); 2177 return -EUCLEAN; 2178 } 2179 if (unlikely(!IS_ALIGNED(btrfs_dev_extent_length(leaf, de), 2180 sectorsize))) { 2181 generic_err(leaf, slot, 2182 "invalid dev extent length, has %llu not aligned to %u", 2183 btrfs_dev_extent_length(leaf, de), sectorsize); 2184 return -EUCLEAN; 2185 } 2186 /* Overlap check with previous dev extent. */ 2187 if (slot && prev_key->objectid == key->objectid && 2188 prev_key->type == key->type) { 2189 struct btrfs_dev_extent *prev_de; 2190 u64 prev_len; 2191 2192 prev_de = btrfs_item_ptr(leaf, slot - 1, struct btrfs_dev_extent); 2193 prev_len = btrfs_dev_extent_length(leaf, prev_de); 2194 if (unlikely(prev_key->offset + prev_len > key->offset)) { 2195 generic_err(leaf, slot, 2196 "dev extent overlap, prev offset %llu len %llu current offset %llu", 2197 prev_key->offset, prev_len, key->offset); 2198 return -EUCLEAN; 2199 } 2200 } 2201 return 0; 2202 } 2203 2204 static int check_free_space_info(struct extent_buffer *leaf, struct btrfs_key *key, 2205 int slot) 2206 { 2207 struct btrfs_fs_info *fs_info = leaf->fs_info; 2208 struct btrfs_free_space_info *fsi; 2209 const u32 blocksize = fs_info->sectorsize; 2210 u64 end; 2211 u32 flags; 2212 2213 if (unlikely(!IS_ALIGNED(key->objectid, blocksize))) { 2214 generic_err(leaf, slot, 2215 "free space info key objectid is not aligned to %u, has " BTRFS_KEY_FMT, 2216 blocksize, BTRFS_KEY_FMT_VALUE(key)); 2217 return -EUCLEAN; 2218 } 2219 if (unlikely(!IS_ALIGNED(key->offset, blocksize))) { 2220 generic_err(leaf, slot, 2221 "free space info key offset is not aligned to %u, has " BTRFS_KEY_FMT, 2222 blocksize, BTRFS_KEY_FMT_VALUE(key)); 2223 return -EUCLEAN; 2224 } 2225 if (unlikely(check_add_overflow(key->objectid, key->offset, &end))) { 2226 generic_err(leaf, slot, 2227 "free space info key overflows, has " BTRFS_KEY_FMT, 2228 BTRFS_KEY_FMT_VALUE(key)); 2229 return -EUCLEAN; 2230 } 2231 if (unlikely(btrfs_item_size(leaf, slot) != 2232 sizeof(struct btrfs_free_space_info))) { 2233 generic_err(leaf, slot, 2234 "invalid item size for free space info, has %u expect %zu", 2235 btrfs_item_size(leaf, slot), 2236 sizeof(struct btrfs_free_space_info)); 2237 return -EUCLEAN; 2238 } 2239 fsi = btrfs_item_ptr(leaf, slot, struct btrfs_free_space_info); 2240 flags = btrfs_free_space_flags(leaf, fsi); 2241 if (unlikely(flags & ~BTRFS_FREE_SPACE_FLAGS_MASK)) { 2242 generic_err(leaf, slot, 2243 "unknown flags for free space info, has 0x%x valid mask 0x%lx", 2244 flags, BTRFS_FREE_SPACE_FLAGS_MASK); 2245 return -EUCLEAN; 2246 } 2247 if (unlikely(btrfs_free_space_extent_count(leaf, fsi) > 2248 key->offset >> fs_info->sectorsize_bits)) { 2249 generic_err(leaf, slot, 2250 "suspicious extent count, has %u max valid %llu", 2251 btrfs_free_space_extent_count(leaf, fsi), 2252 key->offset >> fs_info->sectorsize_bits); 2253 return -EUCLEAN; 2254 } 2255 return 0; 2256 } 2257 2258 static int check_free_space_common_key(struct extent_buffer *leaf, struct btrfs_key *key, int slot, 2259 struct btrfs_key *prev_key) 2260 { 2261 struct btrfs_fs_info *fs_info = leaf->fs_info; 2262 const u32 blocksize = fs_info->sectorsize; 2263 const char *type_str = (key->type == BTRFS_FREE_SPACE_EXTENT_KEY) ? "extent" : "bitmap"; 2264 u64 end; 2265 2266 if (unlikely(!IS_ALIGNED(key->objectid, blocksize))) { 2267 generic_err(leaf, slot, 2268 "free space %s key objectid is not aligned to %u, has " BTRFS_KEY_FMT, 2269 type_str, blocksize, BTRFS_KEY_FMT_VALUE(key)); 2270 return -EUCLEAN; 2271 } 2272 if (unlikely(!IS_ALIGNED(key->offset, blocksize))) { 2273 generic_err(leaf, slot, 2274 "free space %s key offset is not aligned to %u, has " BTRFS_KEY_FMT, 2275 type_str, blocksize, BTRFS_KEY_FMT_VALUE(key)); 2276 return -EUCLEAN; 2277 } 2278 if (unlikely(key->offset == 0)) { 2279 generic_err(leaf, slot, "free space %s length is 0", type_str); 2280 return -EUCLEAN; 2281 } 2282 if (unlikely(check_add_overflow(key->objectid, key->offset, &end))) { 2283 generic_err(leaf, slot, 2284 "free space %s end overflow, have objectid %llu offset %llu", 2285 type_str, key->objectid, key->offset); 2286 return -EUCLEAN; 2287 } 2288 if (slot == 0) 2289 return 0; 2290 2291 /* 2292 * Make sure the current key is inside the block group, and matching 2293 * the expected info type. 2294 */ 2295 if (prev_key->type == BTRFS_FREE_SPACE_INFO_KEY) { 2296 struct btrfs_free_space_info *fsi; 2297 u32 info_flags; 2298 2299 if (unlikely(key->objectid < prev_key->objectid || 2300 key->objectid + key->offset > prev_key->objectid + prev_key->offset)) { 2301 generic_err(leaf, slot, 2302 "free space %s is not inside the space info, prev key " BTRFS_KEY_FMT " current key " BTRFS_KEY_FMT, 2303 type_str, BTRFS_KEY_FMT_VALUE(prev_key), 2304 BTRFS_KEY_FMT_VALUE(key)); 2305 return -EUCLEAN; 2306 } 2307 fsi = btrfs_item_ptr(leaf, slot - 1, struct btrfs_free_space_info); 2308 info_flags = btrfs_free_space_flags(leaf, fsi); 2309 if (unlikely((info_flags == BTRFS_FREE_SPACE_USING_BITMAPS && 2310 key->type == BTRFS_FREE_SPACE_EXTENT_KEY) || 2311 (info_flags != BTRFS_FREE_SPACE_USING_BITMAPS && 2312 key->type == BTRFS_FREE_SPACE_BITMAP_KEY))) { 2313 generic_err(leaf, slot, 2314 "free space %s key type is not matching the type of space info, key type %u space info flags %u", 2315 type_str, key->type, info_flags); 2316 return -EUCLEAN; 2317 } 2318 return 0; 2319 } 2320 /* 2321 * Previous key should be either FREE_SPACE_EXTENT or FREE_SPACE_BITMAP. 2322 * Inside the same block group the key type should match each other, and 2323 * no overlaps. 2324 */ 2325 if (unlikely(key->type != prev_key->type)) { 2326 generic_err(leaf, slot, 2327 "free space %s key type is not matching the type of previous key, key type %u prev key type %u", 2328 type_str, key->type, prev_key->type); 2329 return -EUCLEAN; 2330 } 2331 if (unlikely(prev_key->objectid + prev_key->offset > key->objectid)) { 2332 generic_err(leaf, slot, 2333 "free space %s key overlaps previous key, prev key " BTRFS_KEY_FMT " current key " BTRFS_KEY_FMT, 2334 type_str, BTRFS_KEY_FMT_VALUE(prev_key), 2335 BTRFS_KEY_FMT_VALUE(key)); 2336 return -EUCLEAN; 2337 } 2338 return 0; 2339 } 2340 2341 static int check_free_space_extent(struct extent_buffer *leaf, struct btrfs_key *key, int slot, 2342 struct btrfs_key *prev_key) 2343 { 2344 int ret; 2345 2346 ret = check_free_space_common_key(leaf, key, slot, prev_key); 2347 if (unlikely(ret < 0)) 2348 return ret; 2349 2350 if (unlikely(btrfs_item_size(leaf, slot) != 0)) { 2351 generic_err(leaf, slot, 2352 "invalid item size for free space extent, has %u expect 0", 2353 btrfs_item_size(leaf, slot)); 2354 return -EUCLEAN; 2355 } 2356 return 0; 2357 } 2358 2359 static int check_free_space_bitmap(struct extent_buffer *leaf, 2360 struct btrfs_key *key, int slot, 2361 struct btrfs_key *prev_key) 2362 { 2363 struct btrfs_fs_info *fs_info = leaf->fs_info; 2364 u32 expected_item_size; 2365 int ret; 2366 2367 ret = check_free_space_common_key(leaf, key, slot, prev_key); 2368 if (unlikely(ret < 0)) 2369 return ret; 2370 2371 /* 2372 * The item must hold exactly the right number of bitmap bytes for the 2373 * range described by key->offset. A mismatch means the item was 2374 * truncated or the key is corrupt; either way the bitmap data is not 2375 * safe to access. 2376 */ 2377 expected_item_size = DIV_ROUND_UP(key->offset >> fs_info->sectorsize_bits, 2378 BITS_PER_BYTE); 2379 if (unlikely(btrfs_item_size(leaf, slot) != expected_item_size)) { 2380 generic_err(leaf, slot, 2381 "invalid item size for free space bitmap, has %u expect %u", 2382 btrfs_item_size(leaf, slot), expected_item_size); 2383 return -EUCLEAN; 2384 } 2385 return 0; 2386 } 2387 2388 /* 2389 * Common point to switch the item-specific validation. 2390 */ 2391 static enum btrfs_tree_block_status check_leaf_item(struct extent_buffer *leaf, 2392 struct btrfs_key *key, 2393 int slot, 2394 struct btrfs_key *prev_key, 2395 struct saved_inode_info *inode_info) 2396 { 2397 int ret = 0; 2398 struct btrfs_chunk *chunk; 2399 2400 switch (key->type) { 2401 case BTRFS_EXTENT_DATA_KEY: 2402 ret = check_extent_data_item(leaf, key, slot, prev_key, inode_info); 2403 break; 2404 case BTRFS_EXTENT_CSUM_KEY: 2405 ret = check_csum_item(leaf, key, slot, prev_key); 2406 break; 2407 case BTRFS_DIR_ITEM_KEY: 2408 case BTRFS_DIR_INDEX_KEY: 2409 case BTRFS_XATTR_ITEM_KEY: 2410 ret = check_dir_item(leaf, key, prev_key, slot); 2411 break; 2412 case BTRFS_INODE_REF_KEY: 2413 ret = check_inode_ref(leaf, key, prev_key, slot); 2414 break; 2415 case BTRFS_INODE_EXTREF_KEY: 2416 ret = check_inode_extref(leaf, key, prev_key, slot); 2417 break; 2418 case BTRFS_BLOCK_GROUP_ITEM_KEY: 2419 ret = check_block_group_item(leaf, key, slot); 2420 break; 2421 case BTRFS_CHUNK_ITEM_KEY: 2422 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); 2423 ret = check_leaf_chunk_item(leaf, chunk, key, slot); 2424 break; 2425 case BTRFS_DEV_ITEM_KEY: 2426 ret = check_dev_item(leaf, key, slot); 2427 break; 2428 case BTRFS_DEV_EXTENT_KEY: 2429 ret = check_dev_extent_item(leaf, key, slot, prev_key); 2430 break; 2431 case BTRFS_INODE_ITEM_KEY: 2432 ret = check_inode_item(leaf, key, slot, inode_info); 2433 break; 2434 case BTRFS_ROOT_ITEM_KEY: 2435 ret = check_root_item(leaf, key, slot); 2436 break; 2437 case BTRFS_ROOT_REF_KEY: 2438 case BTRFS_ROOT_BACKREF_KEY: 2439 ret = check_root_ref(leaf, key, slot); 2440 break; 2441 case BTRFS_EXTENT_ITEM_KEY: 2442 case BTRFS_METADATA_ITEM_KEY: 2443 ret = check_extent_item(leaf, key, slot, prev_key); 2444 break; 2445 case BTRFS_TREE_BLOCK_REF_KEY: 2446 case BTRFS_SHARED_DATA_REF_KEY: 2447 case BTRFS_SHARED_BLOCK_REF_KEY: 2448 ret = check_simple_keyed_refs(leaf, key, slot); 2449 break; 2450 case BTRFS_EXTENT_DATA_REF_KEY: 2451 ret = check_extent_data_ref(leaf, key, slot); 2452 break; 2453 case BTRFS_RAID_STRIPE_KEY: 2454 ret = check_raid_stripe_extent(leaf, key, slot); 2455 break; 2456 case BTRFS_FREE_SPACE_INFO_KEY: 2457 ret = check_free_space_info(leaf, key, slot); 2458 break; 2459 case BTRFS_FREE_SPACE_EXTENT_KEY: 2460 ret = check_free_space_extent(leaf, key, slot, prev_key); 2461 break; 2462 case BTRFS_FREE_SPACE_BITMAP_KEY: 2463 ret = check_free_space_bitmap(leaf, key, slot, prev_key); 2464 break; 2465 case BTRFS_IDENTITY_REMAP_KEY: 2466 case BTRFS_REMAP_KEY: 2467 case BTRFS_REMAP_BACKREF_KEY: 2468 ret = check_remap_key(leaf, key, slot); 2469 break; 2470 } 2471 2472 if (unlikely(ret)) 2473 return BTRFS_TREE_BLOCK_INVALID_ITEM; 2474 return BTRFS_TREE_BLOCK_CLEAN; 2475 } 2476 2477 enum btrfs_tree_block_status __btrfs_check_leaf(struct extent_buffer *leaf) 2478 { 2479 struct btrfs_fs_info *fs_info = leaf->fs_info; 2480 struct saved_inode_info inode_info = { 0 }; 2481 /* No valid key type is 0, so all key should be larger than this key */ 2482 struct btrfs_key prev_key = {0, 0, 0}; 2483 struct btrfs_key key; 2484 u32 nritems = btrfs_header_nritems(leaf); 2485 int slot; 2486 2487 if (unlikely(btrfs_header_level(leaf) != 0)) { 2488 generic_err(leaf, 0, 2489 "invalid level for leaf, have %d expect 0", 2490 btrfs_header_level(leaf)); 2491 return BTRFS_TREE_BLOCK_INVALID_LEVEL; 2492 } 2493 2494 if (unlikely(!btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_WRITTEN))) { 2495 generic_err(leaf, 0, "invalid flag for leaf, WRITTEN not set"); 2496 return BTRFS_TREE_BLOCK_WRITTEN_NOT_SET; 2497 } 2498 2499 /* 2500 * Extent buffers from a relocation tree have a owner field that 2501 * corresponds to the subvolume tree they are based on. So just from an 2502 * extent buffer alone we can not find out what is the id of the 2503 * corresponding subvolume tree, so we can not figure out if the extent 2504 * buffer corresponds to the root of the relocation tree or not. So 2505 * skip this check for relocation trees. 2506 */ 2507 if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) { 2508 u64 owner = btrfs_header_owner(leaf); 2509 2510 /* These trees must never be empty */ 2511 if (unlikely(owner == BTRFS_ROOT_TREE_OBJECTID || 2512 owner == BTRFS_CHUNK_TREE_OBJECTID || 2513 owner == BTRFS_DEV_TREE_OBJECTID || 2514 owner == BTRFS_FS_TREE_OBJECTID || 2515 owner == BTRFS_DATA_RELOC_TREE_OBJECTID)) { 2516 generic_err(leaf, 0, 2517 "invalid root, root %llu must never be empty", 2518 owner); 2519 return BTRFS_TREE_BLOCK_INVALID_NRITEMS; 2520 } 2521 2522 /* Unknown tree */ 2523 if (unlikely(owner == 0)) { 2524 generic_err(leaf, 0, 2525 "invalid owner, root 0 is not defined"); 2526 return BTRFS_TREE_BLOCK_INVALID_OWNER; 2527 } 2528 2529 /* EXTENT_TREE_V2 can have empty extent trees. */ 2530 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) 2531 return BTRFS_TREE_BLOCK_CLEAN; 2532 2533 if (unlikely(owner == BTRFS_EXTENT_TREE_OBJECTID)) { 2534 generic_err(leaf, 0, 2535 "invalid root, root %llu must never be empty", 2536 owner); 2537 return BTRFS_TREE_BLOCK_INVALID_NRITEMS; 2538 } 2539 2540 return BTRFS_TREE_BLOCK_CLEAN; 2541 } 2542 2543 if (unlikely(nritems == 0)) 2544 return BTRFS_TREE_BLOCK_CLEAN; 2545 2546 /* 2547 * Check the following things to make sure this is a good leaf, and 2548 * leaf users won't need to bother with similar sanity checks: 2549 * 2550 * 1) key ordering 2551 * 2) item offset and size 2552 * No overlap, no hole, all inside the leaf. 2553 * 3) item content 2554 * If possible, do comprehensive sanity check. 2555 * NOTE: All checks must only rely on the item data itself. 2556 */ 2557 for (slot = 0; slot < nritems; slot++) { 2558 u32 item_end_expected; 2559 u64 item_data_end; 2560 enum btrfs_tree_block_status ret; 2561 2562 btrfs_item_key_to_cpu(leaf, &key, slot); 2563 2564 /* Make sure the keys are in the right order */ 2565 if (unlikely(btrfs_comp_cpu_keys(&prev_key, &key) >= 0)) { 2566 generic_err(leaf, slot, 2567 "bad key order, prev " BTRFS_KEY_FMT " current " BTRFS_KEY_FMT, 2568 BTRFS_KEY_FMT_VALUE(&prev_key), 2569 BTRFS_KEY_FMT_VALUE(&key)); 2570 return BTRFS_TREE_BLOCK_BAD_KEY_ORDER; 2571 } 2572 2573 item_data_end = (u64)btrfs_item_offset(leaf, slot) + 2574 btrfs_item_size(leaf, slot); 2575 /* 2576 * Make sure the offset and ends are right, remember that the 2577 * item data starts at the end of the leaf and grows towards the 2578 * front. 2579 */ 2580 if (slot == 0) 2581 item_end_expected = BTRFS_LEAF_DATA_SIZE(fs_info); 2582 else 2583 item_end_expected = btrfs_item_offset(leaf, 2584 slot - 1); 2585 if (unlikely(item_data_end != item_end_expected)) { 2586 generic_err(leaf, slot, 2587 "unexpected item end, have %llu expect %u", 2588 item_data_end, item_end_expected); 2589 return BTRFS_TREE_BLOCK_INVALID_OFFSETS; 2590 } 2591 2592 /* 2593 * Check to make sure that we don't point outside of the leaf, 2594 * just in case all the items are consistent to each other, but 2595 * all point outside of the leaf. 2596 */ 2597 if (unlikely(item_data_end > BTRFS_LEAF_DATA_SIZE(fs_info))) { 2598 generic_err(leaf, slot, 2599 "slot end outside of leaf, have %llu expect range [0, %u]", 2600 item_data_end, BTRFS_LEAF_DATA_SIZE(fs_info)); 2601 return BTRFS_TREE_BLOCK_INVALID_OFFSETS; 2602 } 2603 2604 /* Also check if the item pointer overlaps with btrfs item. */ 2605 if (unlikely(btrfs_item_ptr_offset(leaf, slot) < 2606 btrfs_item_nr_offset(leaf, slot) + sizeof(struct btrfs_item))) { 2607 generic_err(leaf, slot, 2608 "slot overlaps with its data, item end %lu data start %lu", 2609 btrfs_item_nr_offset(leaf, slot) + 2610 sizeof(struct btrfs_item), 2611 btrfs_item_ptr_offset(leaf, slot)); 2612 return BTRFS_TREE_BLOCK_INVALID_OFFSETS; 2613 } 2614 2615 /* Check if the item size and content meet other criteria. */ 2616 ret = check_leaf_item(leaf, &key, slot, &prev_key, &inode_info); 2617 if (unlikely(ret != BTRFS_TREE_BLOCK_CLEAN)) 2618 return ret; 2619 2620 prev_key.objectid = key.objectid; 2621 prev_key.type = key.type; 2622 prev_key.offset = key.offset; 2623 } 2624 2625 return BTRFS_TREE_BLOCK_CLEAN; 2626 } 2627 2628 int btrfs_check_leaf(struct extent_buffer *leaf) 2629 { 2630 enum btrfs_tree_block_status ret; 2631 2632 ret = __btrfs_check_leaf(leaf); 2633 if (unlikely(ret != BTRFS_TREE_BLOCK_CLEAN)) 2634 return -EUCLEAN; 2635 return 0; 2636 } 2637 ALLOW_ERROR_INJECTION(btrfs_check_leaf, ERRNO); 2638 2639 enum btrfs_tree_block_status __btrfs_check_node(struct extent_buffer *node) 2640 { 2641 struct btrfs_fs_info *fs_info = node->fs_info; 2642 unsigned long nr = btrfs_header_nritems(node); 2643 struct btrfs_key key, next_key; 2644 int slot; 2645 int level = btrfs_header_level(node); 2646 u64 bytenr; 2647 2648 if (unlikely(!btrfs_header_flag(node, BTRFS_HEADER_FLAG_WRITTEN))) { 2649 generic_err(node, 0, "invalid flag for node, WRITTEN not set"); 2650 return BTRFS_TREE_BLOCK_WRITTEN_NOT_SET; 2651 } 2652 2653 if (unlikely(level <= 0 || level >= BTRFS_MAX_LEVEL)) { 2654 generic_err(node, 0, 2655 "invalid level for node, have %d expect [1, %d]", 2656 level, BTRFS_MAX_LEVEL - 1); 2657 return BTRFS_TREE_BLOCK_INVALID_LEVEL; 2658 } 2659 if (unlikely(nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(fs_info))) { 2660 btrfs_crit(fs_info, 2661 "corrupt node: root=%llu block=%llu, nritems too %s, have %lu expect range [1,%u]", 2662 btrfs_header_owner(node), node->start, 2663 nr == 0 ? "small" : "large", nr, 2664 BTRFS_NODEPTRS_PER_BLOCK(fs_info)); 2665 return BTRFS_TREE_BLOCK_INVALID_NRITEMS; 2666 } 2667 2668 for (slot = 0; slot < nr - 1; slot++) { 2669 bytenr = btrfs_node_blockptr(node, slot); 2670 btrfs_node_key_to_cpu(node, &key, slot); 2671 btrfs_node_key_to_cpu(node, &next_key, slot + 1); 2672 2673 if (unlikely(!bytenr)) { 2674 generic_err(node, slot, 2675 "invalid NULL node pointer"); 2676 return BTRFS_TREE_BLOCK_INVALID_BLOCKPTR; 2677 } 2678 if (unlikely(!IS_ALIGNED(bytenr, fs_info->sectorsize))) { 2679 generic_err(node, slot, 2680 "unaligned pointer, have %llu should be aligned to %u", 2681 bytenr, fs_info->sectorsize); 2682 return BTRFS_TREE_BLOCK_INVALID_BLOCKPTR; 2683 } 2684 2685 if (unlikely(btrfs_comp_cpu_keys(&key, &next_key) >= 0)) { 2686 generic_err(node, slot, 2687 "bad key order, current " BTRFS_KEY_FMT " next " BTRFS_KEY_FMT, 2688 BTRFS_KEY_FMT_VALUE(&key), 2689 BTRFS_KEY_FMT_VALUE(&next_key)); 2690 return BTRFS_TREE_BLOCK_BAD_KEY_ORDER; 2691 } 2692 } 2693 return BTRFS_TREE_BLOCK_CLEAN; 2694 } 2695 2696 int btrfs_check_node(struct extent_buffer *node) 2697 { 2698 enum btrfs_tree_block_status ret; 2699 2700 ret = __btrfs_check_node(node); 2701 if (unlikely(ret != BTRFS_TREE_BLOCK_CLEAN)) 2702 return -EUCLEAN; 2703 return 0; 2704 } 2705 ALLOW_ERROR_INJECTION(btrfs_check_node, ERRNO); 2706 2707 int btrfs_check_eb_owner(const struct extent_buffer *eb, u64 root_owner) 2708 { 2709 const bool is_subvol = btrfs_is_fstree(root_owner); 2710 const u64 eb_owner = btrfs_header_owner(eb); 2711 2712 /* 2713 * Skip dummy fs, as selftests don't create unique ebs for each dummy 2714 * root. 2715 */ 2716 if (btrfs_is_testing(eb->fs_info)) 2717 return 0; 2718 /* 2719 * There are several call sites (backref walking, qgroup, and data 2720 * reloc) passing 0 as @root_owner, as they are not holding the 2721 * tree root. In that case, we can not do a reliable ownership check, 2722 * so just exit. 2723 */ 2724 if (root_owner == 0) 2725 return 0; 2726 /* 2727 * These trees use key.offset as their owner, our callers don't have 2728 * the extra capacity to pass key.offset here. So we just skip them. 2729 */ 2730 if (root_owner == BTRFS_TREE_LOG_OBJECTID || 2731 root_owner == BTRFS_TREE_RELOC_OBJECTID) 2732 return 0; 2733 2734 if (!is_subvol) { 2735 /* For non-subvolume trees, the eb owner should match root owner */ 2736 if (unlikely(root_owner != eb_owner)) { 2737 btrfs_crit(eb->fs_info, 2738 "corrupted %s, root=%llu block=%llu owner mismatch, have %llu expect %llu", 2739 btrfs_header_level(eb) == 0 ? "leaf" : "node", 2740 root_owner, btrfs_header_bytenr(eb), eb_owner, 2741 root_owner); 2742 return -EUCLEAN; 2743 } 2744 return 0; 2745 } 2746 2747 /* 2748 * For subvolume trees, owners can mismatch, but they should all belong 2749 * to subvolume trees. 2750 */ 2751 if (unlikely(is_subvol != btrfs_is_fstree(eb_owner))) { 2752 btrfs_crit(eb->fs_info, 2753 "corrupted %s, root=%llu block=%llu owner mismatch, have %llu expect [%llu, %llu]", 2754 btrfs_header_level(eb) == 0 ? "leaf" : "node", 2755 root_owner, btrfs_header_bytenr(eb), eb_owner, 2756 BTRFS_FIRST_FREE_OBJECTID, BTRFS_LAST_FREE_OBJECTID); 2757 return -EUCLEAN; 2758 } 2759 return 0; 2760 } 2761 2762 int btrfs_verify_level_key(struct extent_buffer *eb, 2763 const struct btrfs_tree_parent_check *check) 2764 { 2765 struct btrfs_fs_info *fs_info = eb->fs_info; 2766 int found_level; 2767 struct btrfs_key found_key; 2768 int ret; 2769 2770 found_level = btrfs_header_level(eb); 2771 if (unlikely(found_level != check->level)) { 2772 DEBUG_WARN(); 2773 btrfs_err(fs_info, 2774 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u", 2775 eb->start, check->level, found_level); 2776 return -EUCLEAN; 2777 } 2778 2779 if (!check->has_first_key) 2780 return 0; 2781 2782 /* 2783 * For live tree block (new tree blocks in current transaction), 2784 * we need proper lock context to avoid race, which is impossible here. 2785 * So we only checks tree blocks which is read from disk, whose 2786 * generation <= fs_info->last_trans_committed. 2787 */ 2788 if (btrfs_header_generation(eb) > btrfs_get_last_trans_committed(fs_info)) 2789 return 0; 2790 2791 /* We have @first_key, so this @eb must have at least one item */ 2792 if (unlikely(btrfs_header_nritems(eb) == 0)) { 2793 btrfs_err(fs_info, 2794 "invalid tree nritems, bytenr=%llu nritems=0 expect >0", 2795 eb->start); 2796 DEBUG_WARN(); 2797 return -EUCLEAN; 2798 } 2799 2800 if (found_level) 2801 btrfs_node_key_to_cpu(eb, &found_key, 0); 2802 else 2803 btrfs_item_key_to_cpu(eb, &found_key, 0); 2804 2805 ret = btrfs_comp_cpu_keys(&check->first_key, &found_key); 2806 if (unlikely(ret)) { 2807 DEBUG_WARN(); 2808 btrfs_err(fs_info, 2809 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)", 2810 eb->start, check->transid, check->first_key.objectid, 2811 check->first_key.type, check->first_key.offset, 2812 found_key.objectid, found_key.type, 2813 found_key.offset); 2814 } 2815 return ret; 2816 } 2817