fs/btrfs/tree-checker.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) Qu Wenruo 2017.  All rights reserved.
   4  */
   5
   6 /*
   7  * The module is used to catch unexpected/corrupted tree block data.
   8  * Such behavior can be caused either by a fuzzed image or bugs.
   9  *
  10  * The objective is to do leaf/node validation checks when tree block is read
  11  * from disk, and check *every* possible member, so other code won't
  12  * need to checking them again.
  13  *
  14  * Due to the potential and unwanted damage, every checker needs to be
  15  * carefully reviewed otherwise so it does not prevent mount of valid images.
  16  */
  17
  18 #include <linux/types.h>
  19 #include <linux/stddef.h>
  20 #include <linux/error-injection.h>
  21 #include "messages.h"
  22 #include "ctree.h"
  23 #include "tree-checker.h"
  24 #include "disk-io.h"
  25 #include "compression.h"
  26 #include "volumes.h"
  27 #include "misc.h"
  28 #include "btrfs_inode.h"
  29 #include "fs.h"
  30 #include "accessors.h"
  31 #include "file-item.h"
  32
  33 /*
  34  * Error message should follow the following format:
  35  * corrupt <type>: <identifier>, <reason>[, <bad_value>]
  36  *
  37  * @type:       leaf or node
  38  * @identifier: the necessary info to locate the leaf/node.
  39  *              It's recommended to decode key.objecitd/offset if it's
  40  *              meaningful.
  41  * @reason:     describe the error
  42  * @bad_value:  optional, it's recommended to output bad value and its
  43  *              expected value (range).
  44  *
  45  * Since comma is used to separate the components, only space is allowed
  46  * inside each component.
  47  */
  48
  49 /*
  50  * Append generic "corrupt leaf/node root=%llu block=%llu slot=%d: " to @fmt.
  51  * Allows callers to customize the output.
  52  */
  53 __printf(3, 4)
  54 __cold
  55 static void generic_err(const struct extent_buffer *eb, int slot,
  56                         const char *fmt, ...)
  57 {
  58         const struct btrfs_fs_info *fs_info = eb->fs_info;
  59         struct va_format vaf;
  60         va_list args;
  61
  62         va_start(args, fmt);
  63
  64         vaf.fmt = fmt;
  65         vaf.va = &args;
  66
  67         btrfs_crit(fs_info,
  68                 "corrupt %s: root=%llu block=%llu slot=%d, %pV",
  69                 btrfs_header_level(eb) == 0 ? "leaf" : "node",
  70                 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot, &vaf);
  71         va_end(args);
  72 }
  73
  74 /*
  75  * Customized reporter for extent data item, since its key objectid and
  76  * offset has its own meaning.
  77  */
  78 __printf(3, 4)
  79 __cold
  80 static void file_extent_err(const struct extent_buffer *eb, int slot,
  81                             const char *fmt, ...)
  82 {
  83         const struct btrfs_fs_info *fs_info = eb->fs_info;
  84         struct btrfs_key key;
  85         struct va_format vaf;
  86         va_list args;
  87
  88         btrfs_item_key_to_cpu(eb, &key, slot);
  89         va_start(args, fmt);
  90
  91         vaf.fmt = fmt;
  92         vaf.va = &args;
  93
  94         btrfs_crit(fs_info,
  95         "corrupt %s: root=%llu block=%llu slot=%d ino=%llu file_offset=%llu, %pV",
  96                 btrfs_header_level(eb) == 0 ? "leaf" : "node",
  97                 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
  98                 key.objectid, key.offset, &vaf);
  99         va_end(args);
 100 }
 101
 102 /*
 103  * Return 0 if the btrfs_file_extent_##name is aligned to @alignment
 104  * Else return 1
 105  */
 106 #define CHECK_FE_ALIGNED(leaf, slot, fi, name, alignment)                     \
 107 ({                                                                            \
 108         if (unlikely(!IS_ALIGNED(btrfs_file_extent_##name((leaf), (fi)),      \
 109                                  (alignment))))                               \
 110                 file_extent_err((leaf), (slot),                               \
 111         "invalid %s for file extent, have %llu, should be aligned to %u",     \
 112                         (#name), btrfs_file_extent_##name((leaf), (fi)),      \
 113                         (alignment));                                         \
 114         (!IS_ALIGNED(btrfs_file_extent_##name((leaf), (fi)), (alignment)));   \
 115 })
 116
 117 static u64 file_extent_end(struct extent_buffer *leaf,
 118                            struct btrfs_key *key,
 119                            struct btrfs_file_extent_item *extent)
 120 {
 121         u64 end;
 122         u64 len;
 123
 124         if (btrfs_file_extent_type(leaf, extent) == BTRFS_FILE_EXTENT_INLINE) {
 125                 len = btrfs_file_extent_ram_bytes(leaf, extent);
 126                 end = ALIGN(key->offset + len, leaf->fs_info->sectorsize);
 127         } else {
 128                 len = btrfs_file_extent_num_bytes(leaf, extent);
 129                 end = key->offset + len;
 130         }
 131         return end;
 132 }
 133
 134 /*
 135  * Customized report for dir_item, the only new important information is
 136  * key->objectid, which represents inode number
 137  */
 138 __printf(3, 4)
 139 __cold
 140 static void dir_item_err(const struct extent_buffer *eb, int slot,
 141                          const char *fmt, ...)
 142 {
 143         const struct btrfs_fs_info *fs_info = eb->fs_info;
 144         struct btrfs_key key;
 145         struct va_format vaf;
 146         va_list args;
 147
 148         btrfs_item_key_to_cpu(eb, &key, slot);
 149         va_start(args, fmt);
 150
 151         vaf.fmt = fmt;
 152         vaf.va = &args;
 153
 154         btrfs_crit(fs_info,
 155                 "corrupt %s: root=%llu block=%llu slot=%d ino=%llu, %pV",
 156                 btrfs_header_level(eb) == 0 ? "leaf" : "node",
 157                 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
 158                 key.objectid, &vaf);
 159         va_end(args);
 160 }
 161
 162 /*
 163  * This functions checks prev_key->objectid, to ensure current key and prev_key
 164  * share the same objectid as inode number.
 165  *
 166  * This is to detect missing INODE_ITEM in subvolume trees.
 167  *
 168  * Return true if everything is OK or we don't need to check.
 169  * Return false if anything is wrong.
 170  */
 171 static bool check_prev_ino(struct extent_buffer *leaf,
 172                            struct btrfs_key *key, int slot,
 173                            struct btrfs_key *prev_key)
 174 {
 175         /* No prev key, skip check */
 176         if (slot == 0)
 177                 return true;
 178
 179         /* Only these key->types needs to be checked */
 180         ASSERT(key->type == BTRFS_XATTR_ITEM_KEY ||
 181                key->type == BTRFS_INODE_REF_KEY ||
 182                key->type == BTRFS_DIR_INDEX_KEY ||
 183                key->type == BTRFS_DIR_ITEM_KEY ||
 184                key->type == BTRFS_EXTENT_DATA_KEY);
 185
 186         /*
 187          * Only subvolume trees along with their reloc trees need this check.
 188          * Things like log tree doesn't follow this ino requirement.
 189          */
 190         if (!is_fstree(btrfs_header_owner(leaf)))
 191                 return true;
 192
 193         if (key->objectid == prev_key->objectid)
 194                 return true;
 195
 196         /* Error found */
 197         dir_item_err(leaf, slot,
 198                 "invalid previous key objectid, have %llu expect %llu",
 199                 prev_key->objectid, key->objectid);
 200         return false;
 201 }
 202 static int check_extent_data_item(struct extent_buffer *leaf,
 203                                   struct btrfs_key *key, int slot,
 204                                   struct btrfs_key *prev_key)
 205 {
 206         struct btrfs_fs_info *fs_info = leaf->fs_info;
 207         struct btrfs_file_extent_item *fi;
 208         u32 sectorsize = fs_info->sectorsize;
 209         u32 item_size = btrfs_item_size(leaf, slot);
 210         u64 extent_end;
 211
 212         if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) {
 213                 file_extent_err(leaf, slot,
 214 "unaligned file_offset for file extent, have %llu should be aligned to %u",
 215                         key->offset, sectorsize);
 216                 return -EUCLEAN;
 217         }
 218
 219         /*
 220          * Previous key must have the same key->objectid (ino).
 221          * It can be XATTR_ITEM, INODE_ITEM or just another EXTENT_DATA.
 222          * But if objectids mismatch, it means we have a missing
 223          * INODE_ITEM.
 224          */
 225         if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
 226                 return -EUCLEAN;
 227
 228         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
 229
 230         /*
 231          * Make sure the item contains at least inline header, so the file
 232          * extent type is not some garbage.
 233          */
 234         if (unlikely(item_size < BTRFS_FILE_EXTENT_INLINE_DATA_START)) {
 235                 file_extent_err(leaf, slot,
 236                                 "invalid item size, have %u expect [%zu, %u)",
 237                                 item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START,
 238                                 SZ_4K);
 239                 return -EUCLEAN;
 240         }
 241         if (unlikely(btrfs_file_extent_type(leaf, fi) >=
 242                      BTRFS_NR_FILE_EXTENT_TYPES)) {
 243                 file_extent_err(leaf, slot,
 244                 "invalid type for file extent, have %u expect range [0, %u]",
 245                         btrfs_file_extent_type(leaf, fi),
 246                         BTRFS_NR_FILE_EXTENT_TYPES - 1);
 247                 return -EUCLEAN;
 248         }
 249
 250         /*
 251          * Support for new compression/encryption must introduce incompat flag,
 252          * and must be caught in open_ctree().
 253          */
 254         if (unlikely(btrfs_file_extent_compression(leaf, fi) >=
 255                      BTRFS_NR_COMPRESS_TYPES)) {
 256                 file_extent_err(leaf, slot,
 257         "invalid compression for file extent, have %u expect range [0, %u]",
 258                         btrfs_file_extent_compression(leaf, fi),
 259                         BTRFS_NR_COMPRESS_TYPES - 1);
 260                 return -EUCLEAN;
 261         }
 262         if (unlikely(btrfs_file_extent_encryption(leaf, fi))) {
 263                 file_extent_err(leaf, slot,
 264                         "invalid encryption for file extent, have %u expect 0",
 265                         btrfs_file_extent_encryption(leaf, fi));
 266                 return -EUCLEAN;
 267         }
 268         if (btrfs_file_extent_type(leaf, fi) == BTRFS_FILE_EXTENT_INLINE) {
 269                 /* Inline extent must have 0 as key offset */
 270                 if (unlikely(key->offset)) {
 271                         file_extent_err(leaf, slot,
 272                 "invalid file_offset for inline file extent, have %llu expect 0",
 273                                 key->offset);
 274                         return -EUCLEAN;
 275                 }
 276
 277                 /* Compressed inline extent has no on-disk size, skip it */
 278                 if (btrfs_file_extent_compression(leaf, fi) !=
 279                     BTRFS_COMPRESS_NONE)
 280                         return 0;
 281
 282                 /* Uncompressed inline extent size must match item size */
 283                 if (unlikely(item_size != BTRFS_FILE_EXTENT_INLINE_DATA_START +
 284                                           btrfs_file_extent_ram_bytes(leaf, fi))) {
 285                         file_extent_err(leaf, slot,
 286         "invalid ram_bytes for uncompressed inline extent, have %u expect %llu",
 287                                 item_size, BTRFS_FILE_EXTENT_INLINE_DATA_START +
 288                                 btrfs_file_extent_ram_bytes(leaf, fi));
 289                         return -EUCLEAN;
 290                 }
 291                 return 0;
 292         }
 293
 294         /* Regular or preallocated extent has fixed item size */
 295         if (unlikely(item_size != sizeof(*fi))) {
 296                 file_extent_err(leaf, slot,
 297         "invalid item size for reg/prealloc file extent, have %u expect %zu",
 298                         item_size, sizeof(*fi));
 299                 return -EUCLEAN;
 300         }
 301         if (unlikely(CHECK_FE_ALIGNED(leaf, slot, fi, ram_bytes, sectorsize) ||
 302                      CHECK_FE_ALIGNED(leaf, slot, fi, disk_bytenr, sectorsize) ||
 303                      CHECK_FE_ALIGNED(leaf, slot, fi, disk_num_bytes, sectorsize) ||
 304                      CHECK_FE_ALIGNED(leaf, slot, fi, offset, sectorsize) ||
 305                      CHECK_FE_ALIGNED(leaf, slot, fi, num_bytes, sectorsize)))
 306                 return -EUCLEAN;
 307
 308         /* Catch extent end overflow */
 309         if (unlikely(check_add_overflow(btrfs_file_extent_num_bytes(leaf, fi),
 310                                         key->offset, &extent_end))) {
 311                 file_extent_err(leaf, slot,
 312         "extent end overflow, have file offset %llu extent num bytes %llu",
 313                                 key->offset,
 314                                 btrfs_file_extent_num_bytes(leaf, fi));
 315                 return -EUCLEAN;
 316         }
 317
 318         /*
 319          * Check that no two consecutive file extent items, in the same leaf,
 320          * present ranges that overlap each other.
 321          */
 322         if (slot > 0 &&
 323             prev_key->objectid == key->objectid &&
 324             prev_key->type == BTRFS_EXTENT_DATA_KEY) {
 325                 struct btrfs_file_extent_item *prev_fi;
 326                 u64 prev_end;
 327
 328                 prev_fi = btrfs_item_ptr(leaf, slot - 1,
 329                                          struct btrfs_file_extent_item);
 330                 prev_end = file_extent_end(leaf, prev_key, prev_fi);
 331                 if (unlikely(prev_end > key->offset)) {
 332                         file_extent_err(leaf, slot - 1,
 333 "file extent end range (%llu) goes beyond start offset (%llu) of the next file extent",
 334                                         prev_end, key->offset);
 335                         return -EUCLEAN;
 336                 }
 337         }
 338
 339         return 0;
 340 }
 341
 342 static int check_csum_item(struct extent_buffer *leaf, struct btrfs_key *key,
 343                            int slot, struct btrfs_key *prev_key)
 344 {
 345         struct btrfs_fs_info *fs_info = leaf->fs_info;
 346         u32 sectorsize = fs_info->sectorsize;
 347         const u32 csumsize = fs_info->csum_size;
 348
 349         if (unlikely(key->objectid != BTRFS_EXTENT_CSUM_OBJECTID)) {
 350                 generic_err(leaf, slot,
 351                 "invalid key objectid for csum item, have %llu expect %llu",
 352                         key->objectid, BTRFS_EXTENT_CSUM_OBJECTID);
 353                 return -EUCLEAN;
 354         }
 355         if (unlikely(!IS_ALIGNED(key->offset, sectorsize))) {
 356                 generic_err(leaf, slot,
 357         "unaligned key offset for csum item, have %llu should be aligned to %u",
 358                         key->offset, sectorsize);
 359                 return -EUCLEAN;
 360         }
 361         if (unlikely(!IS_ALIGNED(btrfs_item_size(leaf, slot), csumsize))) {
 362                 generic_err(leaf, slot,
 363         "unaligned item size for csum item, have %u should be aligned to %u",
 364                         btrfs_item_size(leaf, slot), csumsize);
 365                 return -EUCLEAN;
 366         }
 367         if (slot > 0 && prev_key->type == BTRFS_EXTENT_CSUM_KEY) {
 368                 u64 prev_csum_end;
 369                 u32 prev_item_size;
 370
 371                 prev_item_size = btrfs_item_size(leaf, slot - 1);
 372                 prev_csum_end = (prev_item_size / csumsize) * sectorsize;
 373                 prev_csum_end += prev_key->offset;
 374                 if (unlikely(prev_csum_end > key->offset)) {
 375                         generic_err(leaf, slot - 1,
 376 "csum end range (%llu) goes beyond the start range (%llu) of the next csum item",
 377                                     prev_csum_end, key->offset);
 378                         return -EUCLEAN;
 379                 }
 380         }
 381         return 0;
 382 }
 383
 384 /* Inode item error output has the same format as dir_item_err() */
 385 #define inode_item_err(eb, slot, fmt, ...)                      \
 386         dir_item_err(eb, slot, fmt, __VA_ARGS__)
 387
 388 static int check_inode_key(struct extent_buffer *leaf, struct btrfs_key *key,
 389                            int slot)
 390 {
 391         struct btrfs_key item_key;
 392         bool is_inode_item;
 393
 394         btrfs_item_key_to_cpu(leaf, &item_key, slot);
 395         is_inode_item = (item_key.type == BTRFS_INODE_ITEM_KEY);
 396
 397         /* For XATTR_ITEM, location key should be all 0 */
 398         if (item_key.type == BTRFS_XATTR_ITEM_KEY) {
 399                 if (unlikely(key->objectid != 0 || key->type != 0 ||
 400                              key->offset != 0))
 401                         return -EUCLEAN;
 402                 return 0;
 403         }
 404
 405         if (unlikely((key->objectid < BTRFS_FIRST_FREE_OBJECTID ||
 406                       key->objectid > BTRFS_LAST_FREE_OBJECTID) &&
 407                      key->objectid != BTRFS_ROOT_TREE_DIR_OBJECTID &&
 408                      key->objectid != BTRFS_FREE_INO_OBJECTID)) {
 409                 if (is_inode_item) {
 410                         generic_err(leaf, slot,
 411         "invalid key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
 412                                 key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
 413                                 BTRFS_FIRST_FREE_OBJECTID,
 414                                 BTRFS_LAST_FREE_OBJECTID,
 415                                 BTRFS_FREE_INO_OBJECTID);
 416                 } else {
 417                         dir_item_err(leaf, slot,
 418 "invalid location key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
 419                                 key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
 420                                 BTRFS_FIRST_FREE_OBJECTID,
 421                                 BTRFS_LAST_FREE_OBJECTID,
 422                                 BTRFS_FREE_INO_OBJECTID);
 423                 }
 424                 return -EUCLEAN;
 425         }
 426         if (unlikely(key->offset != 0)) {
 427                 if (is_inode_item)
 428                         inode_item_err(leaf, slot,
 429                                        "invalid key offset: has %llu expect 0",
 430                                        key->offset);
 431                 else
 432                         dir_item_err(leaf, slot,
 433                                 "invalid location key offset:has %llu expect 0",
 434                                 key->offset);
 435                 return -EUCLEAN;
 436         }
 437         return 0;
 438 }
 439
 440 static int check_root_key(struct extent_buffer *leaf, struct btrfs_key *key,
 441                           int slot)
 442 {
 443         struct btrfs_key item_key;
 444         bool is_root_item;
 445
 446         btrfs_item_key_to_cpu(leaf, &item_key, slot);
 447         is_root_item = (item_key.type == BTRFS_ROOT_ITEM_KEY);
 448
 449         /* No such tree id */
 450         if (unlikely(key->objectid == 0)) {
 451                 if (is_root_item)
 452                         generic_err(leaf, slot, "invalid root id 0");
 453                 else
 454                         dir_item_err(leaf, slot,
 455                                      "invalid location key root id 0");
 456                 return -EUCLEAN;
 457         }
 458
 459         /* DIR_ITEM/INDEX/INODE_REF is not allowed to point to non-fs trees */
 460         if (unlikely(!is_fstree(key->objectid) && !is_root_item)) {
 461                 dir_item_err(leaf, slot,
 462                 "invalid location key objectid, have %llu expect [%llu, %llu]",
 463                                 key->objectid, BTRFS_FIRST_FREE_OBJECTID,
 464                                 BTRFS_LAST_FREE_OBJECTID);
 465                 return -EUCLEAN;
 466         }
 467
 468         /*
 469          * ROOT_ITEM with non-zero offset means this is a snapshot, created at
 470          * @offset transid.
 471          * Furthermore, for location key in DIR_ITEM, its offset is always -1.
 472          *
 473          * So here we only check offset for reloc tree whose key->offset must
 474          * be a valid tree.
 475          */
 476         if (unlikely(key->objectid == BTRFS_TREE_RELOC_OBJECTID &&
 477                      key->offset == 0)) {
 478                 generic_err(leaf, slot, "invalid root id 0 for reloc tree");
 479                 return -EUCLEAN;
 480         }
 481         return 0;
 482 }
 483
 484 static int check_dir_item(struct extent_buffer *leaf,
 485                           struct btrfs_key *key, struct btrfs_key *prev_key,
 486                           int slot)
 487 {
 488         struct btrfs_fs_info *fs_info = leaf->fs_info;
 489         struct btrfs_dir_item *di;
 490         u32 item_size = btrfs_item_size(leaf, slot);
 491         u32 cur = 0;
 492
 493         if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
 494                 return -EUCLEAN;
 495
 496         di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
 497         while (cur < item_size) {
 498                 struct btrfs_key location_key;
 499                 u32 name_len;
 500                 u32 data_len;
 501                 u32 max_name_len;
 502                 u32 total_size;
 503                 u32 name_hash;
 504                 u8 dir_type;
 505                 int ret;
 506
 507                 /* header itself should not cross item boundary */
 508                 if (unlikely(cur + sizeof(*di) > item_size)) {
 509                         dir_item_err(leaf, slot,
 510                 "dir item header crosses item boundary, have %zu boundary %u",
 511                                 cur + sizeof(*di), item_size);
 512                         return -EUCLEAN;
 513                 }
 514
 515                 /* Location key check */
 516                 btrfs_dir_item_key_to_cpu(leaf, di, &location_key);
 517                 if (location_key.type == BTRFS_ROOT_ITEM_KEY) {
 518                         ret = check_root_key(leaf, &location_key, slot);
 519                         if (unlikely(ret < 0))
 520                                 return ret;
 521                 } else if (location_key.type == BTRFS_INODE_ITEM_KEY ||
 522                            location_key.type == 0) {
 523                         ret = check_inode_key(leaf, &location_key, slot);
 524                         if (unlikely(ret < 0))
 525                                 return ret;
 526                 } else {
 527                         dir_item_err(leaf, slot,
 528                         "invalid location key type, have %u, expect %u or %u",
 529                                      location_key.type, BTRFS_ROOT_ITEM_KEY,
 530                                      BTRFS_INODE_ITEM_KEY);
 531                         return -EUCLEAN;
 532                 }
 533
 534                 /* dir type check */
 535                 dir_type = btrfs_dir_ftype(leaf, di);
 536                 if (unlikely(dir_type >= BTRFS_FT_MAX)) {
 537                         dir_item_err(leaf, slot,
 538                         "invalid dir item type, have %u expect [0, %u)",
 539                                 dir_type, BTRFS_FT_MAX);
 540                         return -EUCLEAN;
 541                 }
 542
 543                 if (unlikely(key->type == BTRFS_XATTR_ITEM_KEY &&
 544                              dir_type != BTRFS_FT_XATTR)) {
 545                         dir_item_err(leaf, slot,
 546                 "invalid dir item type for XATTR key, have %u expect %u",
 547                                 dir_type, BTRFS_FT_XATTR);
 548                         return -EUCLEAN;
 549                 }
 550                 if (unlikely(dir_type == BTRFS_FT_XATTR &&
 551                              key->type != BTRFS_XATTR_ITEM_KEY)) {
 552                         dir_item_err(leaf, slot,
 553                         "xattr dir type found for non-XATTR key");
 554                         return -EUCLEAN;
 555                 }
 556                 if (dir_type == BTRFS_FT_XATTR)
 557                         max_name_len = XATTR_NAME_MAX;
 558                 else
 559                         max_name_len = BTRFS_NAME_LEN;
 560
 561                 /* Name/data length check */
 562                 name_len = btrfs_dir_name_len(leaf, di);
 563                 data_len = btrfs_dir_data_len(leaf, di);
 564                 if (unlikely(name_len > max_name_len)) {
 565                         dir_item_err(leaf, slot,
 566                         "dir item name len too long, have %u max %u",
 567                                 name_len, max_name_len);
 568                         return -EUCLEAN;
 569                 }
 570                 if (unlikely(name_len + data_len > BTRFS_MAX_XATTR_SIZE(fs_info))) {
 571                         dir_item_err(leaf, slot,
 572                         "dir item name and data len too long, have %u max %u",
 573                                 name_len + data_len,
 574                                 BTRFS_MAX_XATTR_SIZE(fs_info));
 575                         return -EUCLEAN;
 576                 }
 577
 578                 if (unlikely(data_len && dir_type != BTRFS_FT_XATTR)) {
 579                         dir_item_err(leaf, slot,
 580                         "dir item with invalid data len, have %u expect 0",
 581                                 data_len);
 582                         return -EUCLEAN;
 583                 }
 584
 585                 total_size = sizeof(*di) + name_len + data_len;
 586
 587                 /* header and name/data should not cross item boundary */
 588                 if (unlikely(cur + total_size > item_size)) {
 589                         dir_item_err(leaf, slot,
 590                 "dir item data crosses item boundary, have %u boundary %u",
 591                                 cur + total_size, item_size);
 592                         return -EUCLEAN;
 593                 }
 594
 595                 /*
 596                  * Special check for XATTR/DIR_ITEM, as key->offset is name
 597                  * hash, should match its name
 598                  */
 599                 if (key->type == BTRFS_DIR_ITEM_KEY ||
 600                     key->type == BTRFS_XATTR_ITEM_KEY) {
 601                         char namebuf[max(BTRFS_NAME_LEN, XATTR_NAME_MAX)];
 602
 603                         read_extent_buffer(leaf, namebuf,
 604                                         (unsigned long)(di + 1), name_len);
 605                         name_hash = btrfs_name_hash(namebuf, name_len);
 606                         if (unlikely(key->offset != name_hash)) {
 607                                 dir_item_err(leaf, slot,
 608                 "name hash mismatch with key, have 0x%016x expect 0x%016llx",
 609                                         name_hash, key->offset);
 610                                 return -EUCLEAN;
 611                         }
 612                 }
 613                 cur += total_size;
 614                 di = (struct btrfs_dir_item *)((void *)di + total_size);
 615         }
 616         return 0;
 617 }
 618
 619 __printf(3, 4)
 620 __cold
 621 static void block_group_err(const struct extent_buffer *eb, int slot,
 622                             const char *fmt, ...)
 623 {
 624         const struct btrfs_fs_info *fs_info = eb->fs_info;
 625         struct btrfs_key key;
 626         struct va_format vaf;
 627         va_list args;
 628
 629         btrfs_item_key_to_cpu(eb, &key, slot);
 630         va_start(args, fmt);
 631
 632         vaf.fmt = fmt;
 633         vaf.va = &args;
 634
 635         btrfs_crit(fs_info,
 636         "corrupt %s: root=%llu block=%llu slot=%d bg_start=%llu bg_len=%llu, %pV",
 637                 btrfs_header_level(eb) == 0 ? "leaf" : "node",
 638                 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
 639                 key.objectid, key.offset, &vaf);
 640         va_end(args);
 641 }
 642
 643 static int check_block_group_item(struct extent_buffer *leaf,
 644                                   struct btrfs_key *key, int slot)
 645 {
 646         struct btrfs_fs_info *fs_info = leaf->fs_info;
 647         struct btrfs_block_group_item bgi;
 648         u32 item_size = btrfs_item_size(leaf, slot);
 649         u64 chunk_objectid;
 650         u64 flags;
 651         u64 type;
 652
 653         /*
 654          * Here we don't really care about alignment since extent allocator can
 655          * handle it.  We care more about the size.
 656          */
 657         if (unlikely(key->offset == 0)) {
 658                 block_group_err(leaf, slot,
 659                                 "invalid block group size 0");
 660                 return -EUCLEAN;
 661         }
 662
 663         if (unlikely(item_size != sizeof(bgi))) {
 664                 block_group_err(leaf, slot,
 665                         "invalid item size, have %u expect %zu",
 666                                 item_size, sizeof(bgi));
 667                 return -EUCLEAN;
 668         }
 669
 670         read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot),
 671                            sizeof(bgi));
 672         chunk_objectid = btrfs_stack_block_group_chunk_objectid(&bgi);
 673         if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
 674                 /*
 675                  * We don't init the nr_global_roots until we load the global
 676                  * roots, so this could be 0 at mount time.  If it's 0 we'll
 677                  * just assume we're fine, and later we'll check against our
 678                  * actual value.
 679                  */
 680                 if (unlikely(fs_info->nr_global_roots &&
 681                              chunk_objectid >= fs_info->nr_global_roots)) {
 682                         block_group_err(leaf, slot,
 683         "invalid block group global root id, have %llu, needs to be <= %llu",
 684                                         chunk_objectid,
 685                                         fs_info->nr_global_roots);
 686                         return -EUCLEAN;
 687                 }
 688         } else if (unlikely(chunk_objectid != BTRFS_FIRST_CHUNK_TREE_OBJECTID)) {
 689                 block_group_err(leaf, slot,
 690                 "invalid block group chunk objectid, have %llu expect %llu",
 691                                 btrfs_stack_block_group_chunk_objectid(&bgi),
 692                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
 693                 return -EUCLEAN;
 694         }
 695
 696         if (unlikely(btrfs_stack_block_group_used(&bgi) > key->offset)) {
 697                 block_group_err(leaf, slot,
 698                         "invalid block group used, have %llu expect [0, %llu)",
 699                                 btrfs_stack_block_group_used(&bgi), key->offset);
 700                 return -EUCLEAN;
 701         }
 702
 703         flags = btrfs_stack_block_group_flags(&bgi);
 704         if (unlikely(hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) > 1)) {
 705                 block_group_err(leaf, slot,
 706 "invalid profile flags, have 0x%llx (%lu bits set) expect no more than 1 bit set",
 707                         flags & BTRFS_BLOCK_GROUP_PROFILE_MASK,
 708                         hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK));
 709                 return -EUCLEAN;
 710         }
 711
 712         type = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
 713         if (unlikely(type != BTRFS_BLOCK_GROUP_DATA &&
 714                      type != BTRFS_BLOCK_GROUP_METADATA &&
 715                      type != BTRFS_BLOCK_GROUP_SYSTEM &&
 716                      type != (BTRFS_BLOCK_GROUP_METADATA |
 717                               BTRFS_BLOCK_GROUP_DATA))) {
 718                 block_group_err(leaf, slot,
 719 "invalid type, have 0x%llx (%lu bits set) expect either 0x%llx, 0x%llx, 0x%llx or 0x%llx",
 720                         type, hweight64(type),
 721                         BTRFS_BLOCK_GROUP_DATA, BTRFS_BLOCK_GROUP_METADATA,
 722                         BTRFS_BLOCK_GROUP_SYSTEM,
 723                         BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA);
 724                 return -EUCLEAN;
 725         }
 726         return 0;
 727 }
 728
 729 __printf(4, 5)
 730 __cold
 731 static void chunk_err(const struct extent_buffer *leaf,
 732                       const struct btrfs_chunk *chunk, u64 logical,
 733                       const char *fmt, ...)
 734 {
 735         const struct btrfs_fs_info *fs_info = leaf->fs_info;
 736         bool is_sb;
 737         struct va_format vaf;
 738         va_list args;
 739         int i;
 740         int slot = -1;
 741
 742         /* Only superblock eb is able to have such small offset */
 743         is_sb = (leaf->start == BTRFS_SUPER_INFO_OFFSET);
 744
 745         if (!is_sb) {
 746                 /*
 747                  * Get the slot number by iterating through all slots, this
 748                  * would provide better readability.
 749                  */
 750                 for (i = 0; i < btrfs_header_nritems(leaf); i++) {
 751                         if (btrfs_item_ptr_offset(leaf, i) ==
 752                                         (unsigned long)chunk) {
 753                                 slot = i;
 754                                 break;
 755                         }
 756                 }
 757         }
 758         va_start(args, fmt);
 759         vaf.fmt = fmt;
 760         vaf.va = &args;
 761
 762         if (is_sb)
 763                 btrfs_crit(fs_info,
 764                 "corrupt superblock syschunk array: chunk_start=%llu, %pV",
 765                            logical, &vaf);
 766         else
 767                 btrfs_crit(fs_info,
 768         "corrupt leaf: root=%llu block=%llu slot=%d chunk_start=%llu, %pV",
 769                            BTRFS_CHUNK_TREE_OBJECTID, leaf->start, slot,
 770                            logical, &vaf);
 771         va_end(args);
 772 }
 773
 774 /*
 775  * The common chunk check which could also work on super block sys chunk array.
 776  *
 777  * Return -EUCLEAN if anything is corrupted.
 778  * Return 0 if everything is OK.
 779  */
 780 int btrfs_check_chunk_valid(struct extent_buffer *leaf,
 781                             struct btrfs_chunk *chunk, u64 logical)
 782 {
 783         struct btrfs_fs_info *fs_info = leaf->fs_info;
 784         u64 length;
 785         u64 chunk_end;
 786         u64 stripe_len;
 787         u16 num_stripes;
 788         u16 sub_stripes;
 789         u64 type;
 790         u64 features;
 791         bool mixed = false;
 792         int raid_index;
 793         int nparity;
 794         int ncopies;
 795
 796         length = btrfs_chunk_length(leaf, chunk);
 797         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
 798         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
 799         sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
 800         type = btrfs_chunk_type(leaf, chunk);
 801         raid_index = btrfs_bg_flags_to_raid_index(type);
 802         ncopies = btrfs_raid_array[raid_index].ncopies;
 803         nparity = btrfs_raid_array[raid_index].nparity;
 804
 805         if (unlikely(!num_stripes)) {
 806                 chunk_err(leaf, chunk, logical,
 807                           "invalid chunk num_stripes, have %u", num_stripes);
 808                 return -EUCLEAN;
 809         }
 810         if (unlikely(num_stripes < ncopies)) {
 811                 chunk_err(leaf, chunk, logical,
 812                           "invalid chunk num_stripes < ncopies, have %u < %d",
 813                           num_stripes, ncopies);
 814                 return -EUCLEAN;
 815         }
 816         if (unlikely(nparity && num_stripes == nparity)) {
 817                 chunk_err(leaf, chunk, logical,
 818                           "invalid chunk num_stripes == nparity, have %u == %d",
 819                           num_stripes, nparity);
 820                 return -EUCLEAN;
 821         }
 822         if (unlikely(!IS_ALIGNED(logical, fs_info->sectorsize))) {
 823                 chunk_err(leaf, chunk, logical,
 824                 "invalid chunk logical, have %llu should aligned to %u",
 825                           logical, fs_info->sectorsize);
 826                 return -EUCLEAN;
 827         }
 828         if (unlikely(btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize)) {
 829                 chunk_err(leaf, chunk, logical,
 830                           "invalid chunk sectorsize, have %u expect %u",
 831                           btrfs_chunk_sector_size(leaf, chunk),
 832                           fs_info->sectorsize);
 833                 return -EUCLEAN;
 834         }
 835         if (unlikely(!length || !IS_ALIGNED(length, fs_info->sectorsize))) {
 836                 chunk_err(leaf, chunk, logical,
 837                           "invalid chunk length, have %llu", length);
 838                 return -EUCLEAN;
 839         }
 840         if (unlikely(check_add_overflow(logical, length, &chunk_end))) {
 841                 chunk_err(leaf, chunk, logical,
 842 "invalid chunk logical start and length, have logical start %llu length %llu",
 843                           logical, length);
 844                 return -EUCLEAN;
 845         }
 846         if (unlikely(!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN)) {
 847                 chunk_err(leaf, chunk, logical,
 848                           "invalid chunk stripe length: %llu",
 849                           stripe_len);
 850                 return -EUCLEAN;
 851         }
 852         /*
 853          * We artificially limit the chunk size, so that the number of stripes
 854          * inside a chunk can be fit into a U32.  The current limit (256G) is
 855          * way too large for real world usage anyway, and it's also much larger
 856          * than our existing limit (10G).
 857          *
 858          * Thus it should be a good way to catch obvious bitflips.
 859          */
 860         if (unlikely(length >= btrfs_stripe_nr_to_offset(U32_MAX))) {
 861                 chunk_err(leaf, chunk, logical,
 862                           "chunk length too large: have %llu limit %llu",
 863                           length, btrfs_stripe_nr_to_offset(U32_MAX));
 864                 return -EUCLEAN;
 865         }
 866         if (unlikely(type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
 867                               BTRFS_BLOCK_GROUP_PROFILE_MASK))) {
 868                 chunk_err(leaf, chunk, logical,
 869                           "unrecognized chunk type: 0x%llx",
 870                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
 871                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
 872                           btrfs_chunk_type(leaf, chunk));
 873                 return -EUCLEAN;
 874         }
 875
 876         if (unlikely(!has_single_bit_set(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
 877                      (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) != 0)) {
 878                 chunk_err(leaf, chunk, logical,
 879                 "invalid chunk profile flag: 0x%llx, expect 0 or 1 bit set",
 880                           type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
 881                 return -EUCLEAN;
 882         }
 883         if (unlikely((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0)) {
 884                 chunk_err(leaf, chunk, logical,
 885         "missing chunk type flag, have 0x%llx one bit must be set in 0x%llx",
 886                           type, BTRFS_BLOCK_GROUP_TYPE_MASK);
 887                 return -EUCLEAN;
 888         }
 889
 890         if (unlikely((type & BTRFS_BLOCK_GROUP_SYSTEM) &&
 891                      (type & (BTRFS_BLOCK_GROUP_METADATA |
 892                               BTRFS_BLOCK_GROUP_DATA)))) {
 893                 chunk_err(leaf, chunk, logical,
 894                           "system chunk with data or metadata type: 0x%llx",
 895                           type);
 896                 return -EUCLEAN;
 897         }
 898
 899         features = btrfs_super_incompat_flags(fs_info->super_copy);
 900         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
 901                 mixed = true;
 902
 903         if (!mixed) {
 904                 if (unlikely((type & BTRFS_BLOCK_GROUP_METADATA) &&
 905                              (type & BTRFS_BLOCK_GROUP_DATA))) {
 906                         chunk_err(leaf, chunk, logical,
 907                         "mixed chunk type in non-mixed mode: 0x%llx", type);
 908                         return -EUCLEAN;
 909                 }
 910         }
 911
 912         if (unlikely((type & BTRFS_BLOCK_GROUP_RAID10 &&
 913                       sub_stripes != btrfs_raid_array[BTRFS_RAID_RAID10].sub_stripes) ||
 914                      (type & BTRFS_BLOCK_GROUP_RAID1 &&
 915                       num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1].devs_min) ||
 916                      (type & BTRFS_BLOCK_GROUP_RAID1C3 &&
 917                       num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1C3].devs_min) ||
 918                      (type & BTRFS_BLOCK_GROUP_RAID1C4 &&
 919                       num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1C4].devs_min) ||
 920                      (type & BTRFS_BLOCK_GROUP_RAID5 &&
 921                       num_stripes < btrfs_raid_array[BTRFS_RAID_RAID5].devs_min) ||
 922                      (type & BTRFS_BLOCK_GROUP_RAID6 &&
 923                       num_stripes < btrfs_raid_array[BTRFS_RAID_RAID6].devs_min) ||
 924                      (type & BTRFS_BLOCK_GROUP_DUP &&
 925                       num_stripes != btrfs_raid_array[BTRFS_RAID_DUP].dev_stripes) ||
 926                      ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
 927                       num_stripes != btrfs_raid_array[BTRFS_RAID_SINGLE].dev_stripes))) {
 928                 chunk_err(leaf, chunk, logical,
 929                         "invalid num_stripes:sub_stripes %u:%u for profile %llu",
 930                         num_stripes, sub_stripes,
 931                         type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
 932                 return -EUCLEAN;
 933         }
 934
 935         return 0;
 936 }
 937
 938 /*
 939  * Enhanced version of chunk item checker.
 940  *
 941  * The common btrfs_check_chunk_valid() doesn't check item size since it needs
 942  * to work on super block sys_chunk_array which doesn't have full item ptr.
 943  */
 944 static int check_leaf_chunk_item(struct extent_buffer *leaf,
 945                                  struct btrfs_chunk *chunk,
 946                                  struct btrfs_key *key, int slot)
 947 {
 948         int num_stripes;
 949
 950         if (unlikely(btrfs_item_size(leaf, slot) < sizeof(struct btrfs_chunk))) {
 951                 chunk_err(leaf, chunk, key->offset,
 952                         "invalid chunk item size: have %u expect [%zu, %u)",
 953                         btrfs_item_size(leaf, slot),
 954                         sizeof(struct btrfs_chunk),
 955                         BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
 956                 return -EUCLEAN;
 957         }
 958
 959         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
 960         /* Let btrfs_check_chunk_valid() handle this error type */
 961         if (num_stripes == 0)
 962                 goto out;
 963
 964         if (unlikely(btrfs_chunk_item_size(num_stripes) !=
 965                      btrfs_item_size(leaf, slot))) {
 966                 chunk_err(leaf, chunk, key->offset,
 967                         "invalid chunk item size: have %u expect %lu",
 968                         btrfs_item_size(leaf, slot),
 969                         btrfs_chunk_item_size(num_stripes));
 970                 return -EUCLEAN;
 971         }
 972 out:
 973         return btrfs_check_chunk_valid(leaf, chunk, key->offset);
 974 }
 975
 976 __printf(3, 4)
 977 __cold
 978 static void dev_item_err(const struct extent_buffer *eb, int slot,
 979                          const char *fmt, ...)
 980 {
 981         struct btrfs_key key;
 982         struct va_format vaf;
 983         va_list args;
 984
 985         btrfs_item_key_to_cpu(eb, &key, slot);
 986         va_start(args, fmt);
 987
 988         vaf.fmt = fmt;
 989         vaf.va = &args;
 990
 991         btrfs_crit(eb->fs_info,
 992         "corrupt %s: root=%llu block=%llu slot=%d devid=%llu %pV",
 993                 btrfs_header_level(eb) == 0 ? "leaf" : "node",
 994                 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
 995                 key.objectid, &vaf);
 996         va_end(args);
 997 }
 998
 999 static int check_dev_item(struct extent_buffer *leaf,
1000                           struct btrfs_key *key, int slot)
1001 {
1002         struct btrfs_dev_item *ditem;
1003         const u32 item_size = btrfs_item_size(leaf, slot);
1004
1005         if (unlikely(key->objectid != BTRFS_DEV_ITEMS_OBJECTID)) {
1006                 dev_item_err(leaf, slot,
1007                              "invalid objectid: has=%llu expect=%llu",
1008                              key->objectid, BTRFS_DEV_ITEMS_OBJECTID);
1009                 return -EUCLEAN;
1010         }
1011
1012         if (unlikely(item_size != sizeof(*ditem))) {
1013                 dev_item_err(leaf, slot, "invalid item size: has %u expect %zu",
1014                              item_size, sizeof(*ditem));
1015                 return -EUCLEAN;
1016         }
1017
1018         ditem = btrfs_item_ptr(leaf, slot, struct btrfs_dev_item);
1019         if (unlikely(btrfs_device_id(leaf, ditem) != key->offset)) {
1020                 dev_item_err(leaf, slot,
1021                              "devid mismatch: key has=%llu item has=%llu",
1022                              key->offset, btrfs_device_id(leaf, ditem));
1023                 return -EUCLEAN;
1024         }
1025
1026         /*
1027          * For device total_bytes, we don't have reliable way to check it, as
1028          * it can be 0 for device removal. Device size check can only be done
1029          * by dev extents check.
1030          */
1031         if (unlikely(btrfs_device_bytes_used(leaf, ditem) >
1032                      btrfs_device_total_bytes(leaf, ditem))) {
1033                 dev_item_err(leaf, slot,
1034                              "invalid bytes used: have %llu expect [0, %llu]",
1035                              btrfs_device_bytes_used(leaf, ditem),
1036                              btrfs_device_total_bytes(leaf, ditem));
1037                 return -EUCLEAN;
1038         }
1039         /*
1040          * Remaining members like io_align/type/gen/dev_group aren't really
1041          * utilized.  Skip them to make later usage of them easier.
1042          */
1043         return 0;
1044 }
1045
1046 static int check_inode_item(struct extent_buffer *leaf,
1047                             struct btrfs_key *key, int slot)
1048 {
1049         struct btrfs_fs_info *fs_info = leaf->fs_info;
1050         struct btrfs_inode_item *iitem;
1051         u64 super_gen = btrfs_super_generation(fs_info->super_copy);
1052         u32 valid_mask = (S_IFMT | S_ISUID | S_ISGID | S_ISVTX | 0777);
1053         const u32 item_size = btrfs_item_size(leaf, slot);
1054         u32 mode;
1055         int ret;
1056         u32 flags;
1057         u32 ro_flags;
1058
1059         ret = check_inode_key(leaf, key, slot);
1060         if (unlikely(ret < 0))
1061                 return ret;
1062
1063         if (unlikely(item_size != sizeof(*iitem))) {
1064                 generic_err(leaf, slot, "invalid item size: has %u expect %zu",
1065                             item_size, sizeof(*iitem));
1066                 return -EUCLEAN;
1067         }
1068
1069         iitem = btrfs_item_ptr(leaf, slot, struct btrfs_inode_item);
1070
1071         /* Here we use super block generation + 1 to handle log tree */
1072         if (unlikely(btrfs_inode_generation(leaf, iitem) > super_gen + 1)) {
1073                 inode_item_err(leaf, slot,
1074                         "invalid inode generation: has %llu expect (0, %llu]",
1075                                btrfs_inode_generation(leaf, iitem),
1076                                super_gen + 1);
1077                 return -EUCLEAN;
1078         }
1079         /* Note for ROOT_TREE_DIR_ITEM, mkfs could set its transid 0 */
1080         if (unlikely(btrfs_inode_transid(leaf, iitem) > super_gen + 1)) {
1081                 inode_item_err(leaf, slot,
1082                         "invalid inode transid: has %llu expect [0, %llu]",
1083                                btrfs_inode_transid(leaf, iitem), super_gen + 1);
1084                 return -EUCLEAN;
1085         }
1086
1087         /*
1088          * For size and nbytes it's better not to be too strict, as for dir
1089          * item its size/nbytes can easily get wrong, but doesn't affect
1090          * anything in the fs. So here we skip the check.
1091          */
1092         mode = btrfs_inode_mode(leaf, iitem);
1093         if (unlikely(mode & ~valid_mask)) {
1094                 inode_item_err(leaf, slot,
1095                                "unknown mode bit detected: 0x%x",
1096                                mode & ~valid_mask);
1097                 return -EUCLEAN;
1098         }
1099
1100         /*
1101          * S_IFMT is not bit mapped so we can't completely rely on
1102          * is_power_of_2/has_single_bit_set, but it can save us from checking
1103          * FIFO/CHR/DIR/REG.  Only needs to check BLK, LNK and SOCKS
1104          */
1105         if (!has_single_bit_set(mode & S_IFMT)) {
1106                 if (unlikely(!S_ISLNK(mode) && !S_ISBLK(mode) && !S_ISSOCK(mode))) {
1107                         inode_item_err(leaf, slot,
1108                         "invalid mode: has 0%o expect valid S_IF* bit(s)",
1109                                        mode & S_IFMT);
1110                         return -EUCLEAN;
1111                 }
1112         }
1113         if (unlikely(S_ISDIR(mode) && btrfs_inode_nlink(leaf, iitem) > 1)) {
1114                 inode_item_err(leaf, slot,
1115                        "invalid nlink: has %u expect no more than 1 for dir",
1116                         btrfs_inode_nlink(leaf, iitem));
1117                 return -EUCLEAN;
1118         }
1119         btrfs_inode_split_flags(btrfs_inode_flags(leaf, iitem), &flags, &ro_flags);
1120         if (unlikely(flags & ~BTRFS_INODE_FLAG_MASK)) {
1121                 inode_item_err(leaf, slot,
1122                                "unknown incompat flags detected: 0x%x", flags);
1123                 return -EUCLEAN;
1124         }
1125         if (unlikely(!sb_rdonly(fs_info->sb) &&
1126                      (ro_flags & ~BTRFS_INODE_RO_FLAG_MASK))) {
1127                 inode_item_err(leaf, slot,
1128                         "unknown ro-compat flags detected on writeable mount: 0x%x",
1129                         ro_flags);
1130                 return -EUCLEAN;
1131         }
1132         return 0;
1133 }
1134
1135 static int check_root_item(struct extent_buffer *leaf, struct btrfs_key *key,
1136                            int slot)
1137 {
1138         struct btrfs_fs_info *fs_info = leaf->fs_info;
1139         struct btrfs_root_item ri = { 0 };
1140         const u64 valid_root_flags = BTRFS_ROOT_SUBVOL_RDONLY |
1141                                      BTRFS_ROOT_SUBVOL_DEAD;
1142         int ret;
1143
1144         ret = check_root_key(leaf, key, slot);
1145         if (unlikely(ret < 0))
1146                 return ret;
1147
1148         if (unlikely(btrfs_item_size(leaf, slot) != sizeof(ri) &&
1149                      btrfs_item_size(leaf, slot) !=
1150                      btrfs_legacy_root_item_size())) {
1151                 generic_err(leaf, slot,
1152                             "invalid root item size, have %u expect %zu or %u",
1153                             btrfs_item_size(leaf, slot), sizeof(ri),
1154                             btrfs_legacy_root_item_size());
1155                 return -EUCLEAN;
1156         }
1157
1158         /*
1159          * For legacy root item, the members starting at generation_v2 will be
1160          * all filled with 0.
1161          * And since we allow geneartion_v2 as 0, it will still pass the check.
1162          */
1163         read_extent_buffer(leaf, &ri, btrfs_item_ptr_offset(leaf, slot),
1164                            btrfs_item_size(leaf, slot));
1165
1166         /* Generation related */
1167         if (unlikely(btrfs_root_generation(&ri) >
1168                      btrfs_super_generation(fs_info->super_copy) + 1)) {
1169                 generic_err(leaf, slot,
1170                         "invalid root generation, have %llu expect (0, %llu]",
1171                             btrfs_root_generation(&ri),
1172                             btrfs_super_generation(fs_info->super_copy) + 1);
1173                 return -EUCLEAN;
1174         }
1175         if (unlikely(btrfs_root_generation_v2(&ri) >
1176                      btrfs_super_generation(fs_info->super_copy) + 1)) {
1177                 generic_err(leaf, slot,
1178                 "invalid root v2 generation, have %llu expect (0, %llu]",
1179                             btrfs_root_generation_v2(&ri),
1180                             btrfs_super_generation(fs_info->super_copy) + 1);
1181                 return -EUCLEAN;
1182         }
1183         if (unlikely(btrfs_root_last_snapshot(&ri) >
1184                      btrfs_super_generation(fs_info->super_copy) + 1)) {
1185                 generic_err(leaf, slot,
1186                 "invalid root last_snapshot, have %llu expect (0, %llu]",
1187                             btrfs_root_last_snapshot(&ri),
1188                             btrfs_super_generation(fs_info->super_copy) + 1);
1189                 return -EUCLEAN;
1190         }
1191
1192         /* Alignment and level check */
1193         if (unlikely(!IS_ALIGNED(btrfs_root_bytenr(&ri), fs_info->sectorsize))) {
1194                 generic_err(leaf, slot,
1195                 "invalid root bytenr, have %llu expect to be aligned to %u",
1196                             btrfs_root_bytenr(&ri), fs_info->sectorsize);
1197                 return -EUCLEAN;
1198         }
1199         if (unlikely(btrfs_root_level(&ri) >= BTRFS_MAX_LEVEL)) {
1200                 generic_err(leaf, slot,
1201                             "invalid root level, have %u expect [0, %u]",
1202                             btrfs_root_level(&ri), BTRFS_MAX_LEVEL - 1);
1203                 return -EUCLEAN;
1204         }
1205         if (unlikely(btrfs_root_drop_level(&ri) >= BTRFS_MAX_LEVEL)) {
1206                 generic_err(leaf, slot,
1207                             "invalid root level, have %u expect [0, %u]",
1208                             btrfs_root_drop_level(&ri), BTRFS_MAX_LEVEL - 1);
1209                 return -EUCLEAN;
1210         }
1211
1212         /* Flags check */
1213         if (unlikely(btrfs_root_flags(&ri) & ~valid_root_flags)) {
1214                 generic_err(leaf, slot,
1215                             "invalid root flags, have 0x%llx expect mask 0x%llx",
1216                             btrfs_root_flags(&ri), valid_root_flags);
1217                 return -EUCLEAN;
1218         }
1219         return 0;
1220 }
1221
1222 __printf(3,4)
1223 __cold
1224 static void extent_err(const struct extent_buffer *eb, int slot,
1225                        const char *fmt, ...)
1226 {
1227         struct btrfs_key key;
1228         struct va_format vaf;
1229         va_list args;
1230         u64 bytenr;
1231         u64 len;
1232
1233         btrfs_item_key_to_cpu(eb, &key, slot);
1234         bytenr = key.objectid;
1235         if (key.type == BTRFS_METADATA_ITEM_KEY ||
1236             key.type == BTRFS_TREE_BLOCK_REF_KEY ||
1237             key.type == BTRFS_SHARED_BLOCK_REF_KEY)
1238                 len = eb->fs_info->nodesize;
1239         else
1240                 len = key.offset;
1241         va_start(args, fmt);
1242
1243         vaf.fmt = fmt;
1244         vaf.va = &args;
1245
1246         btrfs_crit(eb->fs_info,
1247         "corrupt %s: block=%llu slot=%d extent bytenr=%llu len=%llu %pV",
1248                 btrfs_header_level(eb) == 0 ? "leaf" : "node",
1249                 eb->start, slot, bytenr, len, &vaf);
1250         va_end(args);
1251 }
1252
1253 static int check_extent_item(struct extent_buffer *leaf,
1254                              struct btrfs_key *key, int slot,
1255                              struct btrfs_key *prev_key)
1256 {
1257         struct btrfs_fs_info *fs_info = leaf->fs_info;
1258         struct btrfs_extent_item *ei;
1259         bool is_tree_block = false;
1260         unsigned long ptr;      /* Current pointer inside inline refs */
1261         unsigned long end;      /* Extent item end */
1262         const u32 item_size = btrfs_item_size(leaf, slot);
1263         u64 flags;
1264         u64 generation;
1265         u64 total_refs;         /* Total refs in btrfs_extent_item */
1266         u64 inline_refs = 0;    /* found total inline refs */
1267
1268         if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1269                      !btrfs_fs_incompat(fs_info, SKINNY_METADATA))) {
1270                 generic_err(leaf, slot,
1271 "invalid key type, METADATA_ITEM type invalid when SKINNY_METADATA feature disabled");
1272                 return -EUCLEAN;
1273         }
1274         /* key->objectid is the bytenr for both key types */
1275         if (unlikely(!IS_ALIGNED(key->objectid, fs_info->sectorsize))) {
1276                 generic_err(leaf, slot,
1277                 "invalid key objectid, have %llu expect to be aligned to %u",
1278                            key->objectid, fs_info->sectorsize);
1279                 return -EUCLEAN;
1280         }
1281
1282         /* key->offset is tree level for METADATA_ITEM_KEY */
1283         if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1284                      key->offset >= BTRFS_MAX_LEVEL)) {
1285                 extent_err(leaf, slot,
1286                            "invalid tree level, have %llu expect [0, %u]",
1287                            key->offset, BTRFS_MAX_LEVEL - 1);
1288                 return -EUCLEAN;
1289         }
1290
1291         /*
1292          * EXTENT/METADATA_ITEM consists of:
1293          * 1) One btrfs_extent_item
1294          *    Records the total refs, type and generation of the extent.
1295          *
1296          * 2) One btrfs_tree_block_info (for EXTENT_ITEM and tree backref only)
1297          *    Records the first key and level of the tree block.
1298          *
1299          * 2) Zero or more btrfs_extent_inline_ref(s)
1300          *    Each inline ref has one btrfs_extent_inline_ref shows:
1301          *    2.1) The ref type, one of the 4
1302          *         TREE_BLOCK_REF       Tree block only
1303          *         SHARED_BLOCK_REF     Tree block only
1304          *         EXTENT_DATA_REF      Data only
1305          *         SHARED_DATA_REF      Data only
1306          *    2.2) Ref type specific data
1307          *         Either using btrfs_extent_inline_ref::offset, or specific
1308          *         data structure.
1309          */
1310         if (unlikely(item_size < sizeof(*ei))) {
1311                 extent_err(leaf, slot,
1312                            "invalid item size, have %u expect [%zu, %u)",
1313                            item_size, sizeof(*ei),
1314                            BTRFS_LEAF_DATA_SIZE(fs_info));
1315                 return -EUCLEAN;
1316         }
1317         end = item_size + btrfs_item_ptr_offset(leaf, slot);
1318
1319         /* Checks against extent_item */
1320         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
1321         flags = btrfs_extent_flags(leaf, ei);
1322         total_refs = btrfs_extent_refs(leaf, ei);
1323         generation = btrfs_extent_generation(leaf, ei);
1324         if (unlikely(generation >
1325                      btrfs_super_generation(fs_info->super_copy) + 1)) {
1326                 extent_err(leaf, slot,
1327                            "invalid generation, have %llu expect (0, %llu]",
1328                            generation,
1329                            btrfs_super_generation(fs_info->super_copy) + 1);
1330                 return -EUCLEAN;
1331         }
1332         if (unlikely(!has_single_bit_set(flags & (BTRFS_EXTENT_FLAG_DATA |
1333                                                   BTRFS_EXTENT_FLAG_TREE_BLOCK)))) {
1334                 extent_err(leaf, slot,
1335                 "invalid extent flag, have 0x%llx expect 1 bit set in 0x%llx",
1336                         flags, BTRFS_EXTENT_FLAG_DATA |
1337                         BTRFS_EXTENT_FLAG_TREE_BLOCK);
1338                 return -EUCLEAN;
1339         }
1340         is_tree_block = !!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK);
1341         if (is_tree_block) {
1342                 if (unlikely(key->type == BTRFS_EXTENT_ITEM_KEY &&
1343                              key->offset != fs_info->nodesize)) {
1344                         extent_err(leaf, slot,
1345                                    "invalid extent length, have %llu expect %u",
1346                                    key->offset, fs_info->nodesize);
1347                         return -EUCLEAN;
1348                 }
1349         } else {
1350                 if (unlikely(key->type != BTRFS_EXTENT_ITEM_KEY)) {
1351                         extent_err(leaf, slot,
1352                         "invalid key type, have %u expect %u for data backref",
1353                                    key->type, BTRFS_EXTENT_ITEM_KEY);
1354                         return -EUCLEAN;
1355                 }
1356                 if (unlikely(!IS_ALIGNED(key->offset, fs_info->sectorsize))) {
1357                         extent_err(leaf, slot,
1358                         "invalid extent length, have %llu expect aligned to %u",
1359                                    key->offset, fs_info->sectorsize);
1360                         return -EUCLEAN;
1361                 }
1362                 if (unlikely(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
1363                         extent_err(leaf, slot,
1364                         "invalid extent flag, data has full backref set");
1365                         return -EUCLEAN;
1366                 }
1367         }
1368         ptr = (unsigned long)(struct btrfs_extent_item *)(ei + 1);
1369
1370         /* Check the special case of btrfs_tree_block_info */
1371         if (is_tree_block && key->type != BTRFS_METADATA_ITEM_KEY) {
1372                 struct btrfs_tree_block_info *info;
1373
1374                 info = (struct btrfs_tree_block_info *)ptr;
1375                 if (unlikely(btrfs_tree_block_level(leaf, info) >= BTRFS_MAX_LEVEL)) {
1376                         extent_err(leaf, slot,
1377                         "invalid tree block info level, have %u expect [0, %u]",
1378                                    btrfs_tree_block_level(leaf, info),
1379                                    BTRFS_MAX_LEVEL - 1);
1380                         return -EUCLEAN;
1381                 }
1382                 ptr = (unsigned long)(struct btrfs_tree_block_info *)(info + 1);
1383         }
1384
1385         /* Check inline refs */
1386         while (ptr < end) {
1387                 struct btrfs_extent_inline_ref *iref;
1388                 struct btrfs_extent_data_ref *dref;
1389                 struct btrfs_shared_data_ref *sref;
1390                 u64 dref_offset;
1391                 u64 inline_offset;
1392                 u8 inline_type;
1393
1394                 if (unlikely(ptr + sizeof(*iref) > end)) {
1395                         extent_err(leaf, slot,
1396 "inline ref item overflows extent item, ptr %lu iref size %zu end %lu",
1397                                    ptr, sizeof(*iref), end);
1398                         return -EUCLEAN;
1399                 }
1400                 iref = (struct btrfs_extent_inline_ref *)ptr;
1401                 inline_type = btrfs_extent_inline_ref_type(leaf, iref);
1402                 inline_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1403                 if (unlikely(ptr + btrfs_extent_inline_ref_size(inline_type) > end)) {
1404                         extent_err(leaf, slot,
1405 "inline ref item overflows extent item, ptr %lu iref size %u end %lu",
1406                                    ptr, inline_type, end);
1407                         return -EUCLEAN;
1408                 }
1409
1410                 switch (inline_type) {
1411                 /* inline_offset is subvolid of the owner, no need to check */
1412                 case BTRFS_TREE_BLOCK_REF_KEY:
1413                         inline_refs++;
1414                         break;
1415                 /* Contains parent bytenr */
1416                 case BTRFS_SHARED_BLOCK_REF_KEY:
1417                         if (unlikely(!IS_ALIGNED(inline_offset,
1418                                                  fs_info->sectorsize))) {
1419                                 extent_err(leaf, slot,
1420                 "invalid tree parent bytenr, have %llu expect aligned to %u",
1421                                            inline_offset, fs_info->sectorsize);
1422                                 return -EUCLEAN;
1423                         }
1424                         inline_refs++;
1425                         break;
1426                 /*
1427                  * Contains owner subvolid, owner key objectid, adjusted offset.
1428                  * The only obvious corruption can happen in that offset.
1429                  */
1430                 case BTRFS_EXTENT_DATA_REF_KEY:
1431                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1432                         dref_offset = btrfs_extent_data_ref_offset(leaf, dref);
1433                         if (unlikely(!IS_ALIGNED(dref_offset,
1434                                                  fs_info->sectorsize))) {
1435                                 extent_err(leaf, slot,
1436                 "invalid data ref offset, have %llu expect aligned to %u",
1437                                            dref_offset, fs_info->sectorsize);
1438                                 return -EUCLEAN;
1439                         }
1440                         inline_refs += btrfs_extent_data_ref_count(leaf, dref);
1441                         break;
1442                 /* Contains parent bytenr and ref count */
1443                 case BTRFS_SHARED_DATA_REF_KEY:
1444                         sref = (struct btrfs_shared_data_ref *)(iref + 1);
1445                         if (unlikely(!IS_ALIGNED(inline_offset,
1446                                                  fs_info->sectorsize))) {
1447                                 extent_err(leaf, slot,
1448                 "invalid data parent bytenr, have %llu expect aligned to %u",
1449                                            inline_offset, fs_info->sectorsize);
1450                                 return -EUCLEAN;
1451                         }
1452                         inline_refs += btrfs_shared_data_ref_count(leaf, sref);
1453                         break;
1454                 default:
1455                         extent_err(leaf, slot, "unknown inline ref type: %u",
1456                                    inline_type);
1457                         return -EUCLEAN;
1458                 }
1459                 ptr += btrfs_extent_inline_ref_size(inline_type);
1460         }
1461         /* No padding is allowed */
1462         if (unlikely(ptr != end)) {
1463                 extent_err(leaf, slot,
1464                            "invalid extent item size, padding bytes found");
1465                 return -EUCLEAN;
1466         }
1467
1468         /* Finally, check the inline refs against total refs */
1469         if (unlikely(inline_refs > total_refs)) {
1470                 extent_err(leaf, slot,
1471                         "invalid extent refs, have %llu expect >= inline %llu",
1472                            total_refs, inline_refs);
1473                 return -EUCLEAN;
1474         }
1475
1476         if ((prev_key->type == BTRFS_EXTENT_ITEM_KEY) ||
1477             (prev_key->type == BTRFS_METADATA_ITEM_KEY)) {
1478                 u64 prev_end = prev_key->objectid;
1479
1480                 if (prev_key->type == BTRFS_METADATA_ITEM_KEY)
1481                         prev_end += fs_info->nodesize;
1482                 else
1483                         prev_end += prev_key->offset;
1484
1485                 if (unlikely(prev_end > key->objectid)) {
1486                         extent_err(leaf, slot,
1487         "previous extent [%llu %u %llu] overlaps current extent [%llu %u %llu]",
1488                                    prev_key->objectid, prev_key->type,
1489                                    prev_key->offset, key->objectid, key->type,
1490                                    key->offset);
1491                         return -EUCLEAN;
1492                 }
1493         }
1494
1495         return 0;
1496 }
1497
1498 static int check_simple_keyed_refs(struct extent_buffer *leaf,
1499                                    struct btrfs_key *key, int slot)
1500 {
1501         u32 expect_item_size = 0;
1502
1503         if (key->type == BTRFS_SHARED_DATA_REF_KEY)
1504                 expect_item_size = sizeof(struct btrfs_shared_data_ref);
1505
1506         if (unlikely(btrfs_item_size(leaf, slot) != expect_item_size)) {
1507                 generic_err(leaf, slot,
1508                 "invalid item size, have %u expect %u for key type %u",
1509                             btrfs_item_size(leaf, slot),
1510                             expect_item_size, key->type);
1511                 return -EUCLEAN;
1512         }
1513         if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1514                 generic_err(leaf, slot,
1515 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1516                             key->objectid, leaf->fs_info->sectorsize);
1517                 return -EUCLEAN;
1518         }
1519         if (unlikely(key->type != BTRFS_TREE_BLOCK_REF_KEY &&
1520                      !IS_ALIGNED(key->offset, leaf->fs_info->sectorsize))) {
1521                 extent_err(leaf, slot,
1522                 "invalid tree parent bytenr, have %llu expect aligned to %u",
1523                            key->offset, leaf->fs_info->sectorsize);
1524                 return -EUCLEAN;
1525         }
1526         return 0;
1527 }
1528
1529 static int check_extent_data_ref(struct extent_buffer *leaf,
1530                                  struct btrfs_key *key, int slot)
1531 {
1532         struct btrfs_extent_data_ref *dref;
1533         unsigned long ptr = btrfs_item_ptr_offset(leaf, slot);
1534         const unsigned long end = ptr + btrfs_item_size(leaf, slot);
1535
1536         if (unlikely(btrfs_item_size(leaf, slot) % sizeof(*dref) != 0)) {
1537                 generic_err(leaf, slot,
1538         "invalid item size, have %u expect aligned to %zu for key type %u",
1539                             btrfs_item_size(leaf, slot),
1540                             sizeof(*dref), key->type);
1541                 return -EUCLEAN;
1542         }
1543         if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1544                 generic_err(leaf, slot,
1545 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1546                             key->objectid, leaf->fs_info->sectorsize);
1547                 return -EUCLEAN;
1548         }
1549         for (; ptr < end; ptr += sizeof(*dref)) {
1550                 u64 offset;
1551
1552                 /*
1553                  * We cannot check the extent_data_ref hash due to possible
1554                  * overflow from the leaf due to hash collisions.
1555                  */
1556                 dref = (struct btrfs_extent_data_ref *)ptr;
1557                 offset = btrfs_extent_data_ref_offset(leaf, dref);
1558                 if (unlikely(!IS_ALIGNED(offset, leaf->fs_info->sectorsize))) {
1559                         extent_err(leaf, slot,
1560         "invalid extent data backref offset, have %llu expect aligned to %u",
1561                                    offset, leaf->fs_info->sectorsize);
1562                         return -EUCLEAN;
1563                 }
1564         }
1565         return 0;
1566 }
1567
1568 #define inode_ref_err(eb, slot, fmt, args...)                   \
1569         inode_item_err(eb, slot, fmt, ##args)
1570 static int check_inode_ref(struct extent_buffer *leaf,
1571                            struct btrfs_key *key, struct btrfs_key *prev_key,
1572                            int slot)
1573 {
1574         struct btrfs_inode_ref *iref;
1575         unsigned long ptr;
1576         unsigned long end;
1577
1578         if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
1579                 return -EUCLEAN;
1580         /* namelen can't be 0, so item_size == sizeof() is also invalid */
1581         if (unlikely(btrfs_item_size(leaf, slot) <= sizeof(*iref))) {
1582                 inode_ref_err(leaf, slot,
1583                         "invalid item size, have %u expect (%zu, %u)",
1584                         btrfs_item_size(leaf, slot),
1585                         sizeof(*iref), BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
1586                 return -EUCLEAN;
1587         }
1588
1589         ptr = btrfs_item_ptr_offset(leaf, slot);
1590         end = ptr + btrfs_item_size(leaf, slot);
1591         while (ptr < end) {
1592                 u16 namelen;
1593
1594                 if (unlikely(ptr + sizeof(iref) > end)) {
1595                         inode_ref_err(leaf, slot,
1596                         "inode ref overflow, ptr %lu end %lu inode_ref_size %zu",
1597                                 ptr, end, sizeof(iref));
1598                         return -EUCLEAN;
1599                 }
1600
1601                 iref = (struct btrfs_inode_ref *)ptr;
1602                 namelen = btrfs_inode_ref_name_len(leaf, iref);
1603                 if (unlikely(ptr + sizeof(*iref) + namelen > end)) {
1604                         inode_ref_err(leaf, slot,
1605                                 "inode ref overflow, ptr %lu end %lu namelen %u",
1606                                 ptr, end, namelen);
1607                         return -EUCLEAN;
1608                 }
1609
1610                 /*
1611                  * NOTE: In theory we should record all found index numbers
1612                  * to find any duplicated indexes, but that will be too time
1613                  * consuming for inodes with too many hard links.
1614                  */
1615                 ptr += sizeof(*iref) + namelen;
1616         }
1617         return 0;
1618 }
1619
1620 /*
1621  * Common point to switch the item-specific validation.
1622  */
1623 static int check_leaf_item(struct extent_buffer *leaf,
1624                            struct btrfs_key *key, int slot,
1625                            struct btrfs_key *prev_key)
1626 {
1627         int ret = 0;
1628         struct btrfs_chunk *chunk;
1629
1630         switch (key->type) {
1631         case BTRFS_EXTENT_DATA_KEY:
1632                 ret = check_extent_data_item(leaf, key, slot, prev_key);
1633                 break;
1634         case BTRFS_EXTENT_CSUM_KEY:
1635                 ret = check_csum_item(leaf, key, slot, prev_key);
1636                 break;
1637         case BTRFS_DIR_ITEM_KEY:
1638         case BTRFS_DIR_INDEX_KEY:
1639         case BTRFS_XATTR_ITEM_KEY:
1640                 ret = check_dir_item(leaf, key, prev_key, slot);
1641                 break;
1642         case BTRFS_INODE_REF_KEY:
1643                 ret = check_inode_ref(leaf, key, prev_key, slot);
1644                 break;
1645         case BTRFS_BLOCK_GROUP_ITEM_KEY:
1646                 ret = check_block_group_item(leaf, key, slot);
1647                 break;
1648         case BTRFS_CHUNK_ITEM_KEY:
1649                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1650                 ret = check_leaf_chunk_item(leaf, chunk, key, slot);
1651                 break;
1652         case BTRFS_DEV_ITEM_KEY:
1653                 ret = check_dev_item(leaf, key, slot);
1654                 break;
1655         case BTRFS_INODE_ITEM_KEY:
1656                 ret = check_inode_item(leaf, key, slot);
1657                 break;
1658         case BTRFS_ROOT_ITEM_KEY:
1659                 ret = check_root_item(leaf, key, slot);
1660                 break;
1661         case BTRFS_EXTENT_ITEM_KEY:
1662         case BTRFS_METADATA_ITEM_KEY:
1663                 ret = check_extent_item(leaf, key, slot, prev_key);
1664                 break;
1665         case BTRFS_TREE_BLOCK_REF_KEY:
1666         case BTRFS_SHARED_DATA_REF_KEY:
1667         case BTRFS_SHARED_BLOCK_REF_KEY:
1668                 ret = check_simple_keyed_refs(leaf, key, slot);
1669                 break;
1670         case BTRFS_EXTENT_DATA_REF_KEY:
1671                 ret = check_extent_data_ref(leaf, key, slot);
1672                 break;
1673         }
1674         return ret;
1675 }
1676
1677 static int check_leaf(struct extent_buffer *leaf, bool check_item_data)
1678 {
1679         struct btrfs_fs_info *fs_info = leaf->fs_info;
1680         /* No valid key type is 0, so all key should be larger than this key */
1681         struct btrfs_key prev_key = {0, 0, 0};
1682         struct btrfs_key key;
1683         u32 nritems = btrfs_header_nritems(leaf);
1684         int slot;
1685
1686         if (unlikely(btrfs_header_level(leaf) != 0)) {
1687                 generic_err(leaf, 0,
1688                         "invalid level for leaf, have %d expect 0",
1689                         btrfs_header_level(leaf));
1690                 return -EUCLEAN;
1691         }
1692
1693         /*
1694          * Extent buffers from a relocation tree have a owner field that
1695          * corresponds to the subvolume tree they are based on. So just from an
1696          * extent buffer alone we can not find out what is the id of the
1697          * corresponding subvolume tree, so we can not figure out if the extent
1698          * buffer corresponds to the root of the relocation tree or not. So
1699          * skip this check for relocation trees.
1700          */
1701         if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
1702                 u64 owner = btrfs_header_owner(leaf);
1703
1704                 /* These trees must never be empty */
1705                 if (unlikely(owner == BTRFS_ROOT_TREE_OBJECTID ||
1706                              owner == BTRFS_CHUNK_TREE_OBJECTID ||
1707                              owner == BTRFS_DEV_TREE_OBJECTID ||
1708                              owner == BTRFS_FS_TREE_OBJECTID ||
1709                              owner == BTRFS_DATA_RELOC_TREE_OBJECTID)) {
1710                         generic_err(leaf, 0,
1711                         "invalid root, root %llu must never be empty",
1712                                     owner);
1713                         return -EUCLEAN;
1714                 }
1715
1716                 /* Unknown tree */
1717                 if (unlikely(owner == 0)) {
1718                         generic_err(leaf, 0,
1719                                 "invalid owner, root 0 is not defined");
1720                         return -EUCLEAN;
1721                 }
1722
1723                 /* EXTENT_TREE_V2 can have empty extent trees. */
1724                 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
1725                         return 0;
1726
1727                 if (unlikely(owner == BTRFS_EXTENT_TREE_OBJECTID)) {
1728                         generic_err(leaf, 0,
1729                         "invalid root, root %llu must never be empty",
1730                                     owner);
1731                         return -EUCLEAN;
1732                 }
1733
1734                 return 0;
1735         }
1736
1737         if (unlikely(nritems == 0))
1738                 return 0;
1739
1740         /*
1741          * Check the following things to make sure this is a good leaf, and
1742          * leaf users won't need to bother with similar sanity checks:
1743          *
1744          * 1) key ordering
1745          * 2) item offset and size
1746          *    No overlap, no hole, all inside the leaf.
1747          * 3) item content
1748          *    If possible, do comprehensive sanity check.
1749          *    NOTE: All checks must only rely on the item data itself.
1750          */
1751         for (slot = 0; slot < nritems; slot++) {
1752                 u32 item_end_expected;
1753                 u64 item_data_end;
1754                 int ret;
1755
1756                 btrfs_item_key_to_cpu(leaf, &key, slot);
1757
1758                 /* Make sure the keys are in the right order */
1759                 if (unlikely(btrfs_comp_cpu_keys(&prev_key, &key) >= 0)) {
1760                         generic_err(leaf, slot,
1761         "bad key order, prev (%llu %u %llu) current (%llu %u %llu)",
1762                                 prev_key.objectid, prev_key.type,
1763                                 prev_key.offset, key.objectid, key.type,
1764                                 key.offset);
1765                         return -EUCLEAN;
1766                 }
1767
1768                 item_data_end = (u64)btrfs_item_offset(leaf, slot) +
1769                                 btrfs_item_size(leaf, slot);
1770                 /*
1771                  * Make sure the offset and ends are right, remember that the
1772                  * item data starts at the end of the leaf and grows towards the
1773                  * front.
1774                  */
1775                 if (slot == 0)
1776                         item_end_expected = BTRFS_LEAF_DATA_SIZE(fs_info);
1777                 else
1778                         item_end_expected = btrfs_item_offset(leaf,
1779                                                                  slot - 1);
1780                 if (unlikely(item_data_end != item_end_expected)) {
1781                         generic_err(leaf, slot,
1782                                 "unexpected item end, have %llu expect %u",
1783                                 item_data_end, item_end_expected);
1784                         return -EUCLEAN;
1785                 }
1786
1787                 /*
1788                  * Check to make sure that we don't point outside of the leaf,
1789                  * just in case all the items are consistent to each other, but
1790                  * all point outside of the leaf.
1791                  */
1792                 if (unlikely(item_data_end > BTRFS_LEAF_DATA_SIZE(fs_info))) {
1793                         generic_err(leaf, slot,
1794                         "slot end outside of leaf, have %llu expect range [0, %u]",
1795                                 item_data_end, BTRFS_LEAF_DATA_SIZE(fs_info));
1796                         return -EUCLEAN;
1797                 }
1798
1799                 /* Also check if the item pointer overlaps with btrfs item. */
1800                 if (unlikely(btrfs_item_ptr_offset(leaf, slot) <
1801                              btrfs_item_nr_offset(leaf, slot) + sizeof(struct btrfs_item))) {
1802                         generic_err(leaf, slot,
1803                 "slot overlaps with its data, item end %lu data start %lu",
1804                                 btrfs_item_nr_offset(leaf, slot) +
1805                                 sizeof(struct btrfs_item),
1806                                 btrfs_item_ptr_offset(leaf, slot));
1807                         return -EUCLEAN;
1808                 }
1809
1810                 if (check_item_data) {
1811                         /*
1812                          * Check if the item size and content meet other
1813                          * criteria
1814                          */
1815                         ret = check_leaf_item(leaf, &key, slot, &prev_key);
1816                         if (unlikely(ret < 0))
1817                                 return ret;
1818                 }
1819
1820                 prev_key.objectid = key.objectid;
1821                 prev_key.type = key.type;
1822                 prev_key.offset = key.offset;
1823         }
1824
1825         return 0;
1826 }
1827
1828 int btrfs_check_leaf_full(struct extent_buffer *leaf)
1829 {
1830         return check_leaf(leaf, true);
1831 }
1832 ALLOW_ERROR_INJECTION(btrfs_check_leaf_full, ERRNO);
1833
1834 int btrfs_check_leaf_relaxed(struct extent_buffer *leaf)
1835 {
1836         return check_leaf(leaf, false);
1837 }
1838
1839 int btrfs_check_node(struct extent_buffer *node)
1840 {
1841         struct btrfs_fs_info *fs_info = node->fs_info;
1842         unsigned long nr = btrfs_header_nritems(node);
1843         struct btrfs_key key, next_key;
1844         int slot;
1845         int level = btrfs_header_level(node);
1846         u64 bytenr;
1847         int ret = 0;
1848
1849         if (unlikely(level <= 0 || level >= BTRFS_MAX_LEVEL)) {
1850                 generic_err(node, 0,
1851                         "invalid level for node, have %d expect [1, %d]",
1852                         level, BTRFS_MAX_LEVEL - 1);
1853                 return -EUCLEAN;
1854         }
1855         if (unlikely(nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(fs_info))) {
1856                 btrfs_crit(fs_info,
1857 "corrupt node: root=%llu block=%llu, nritems too %s, have %lu expect range [1,%u]",
1858                            btrfs_header_owner(node), node->start,
1859                            nr == 0 ? "small" : "large", nr,
1860                            BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1861                 return -EUCLEAN;
1862         }
1863
1864         for (slot = 0; slot < nr - 1; slot++) {
1865                 bytenr = btrfs_node_blockptr(node, slot);
1866                 btrfs_node_key_to_cpu(node, &key, slot);
1867                 btrfs_node_key_to_cpu(node, &next_key, slot + 1);
1868
1869                 if (unlikely(!bytenr)) {
1870                         generic_err(node, slot,
1871                                 "invalid NULL node pointer");
1872                         ret = -EUCLEAN;
1873                         goto out;
1874                 }
1875                 if (unlikely(!IS_ALIGNED(bytenr, fs_info->sectorsize))) {
1876                         generic_err(node, slot,
1877                         "unaligned pointer, have %llu should be aligned to %u",
1878                                 bytenr, fs_info->sectorsize);
1879                         ret = -EUCLEAN;
1880                         goto out;
1881                 }
1882
1883                 if (unlikely(btrfs_comp_cpu_keys(&key, &next_key) >= 0)) {
1884                         generic_err(node, slot,
1885         "bad key order, current (%llu %u %llu) next (%llu %u %llu)",
1886                                 key.objectid, key.type, key.offset,
1887                                 next_key.objectid, next_key.type,
1888                                 next_key.offset);
1889                         ret = -EUCLEAN;
1890                         goto out;
1891                 }
1892         }
1893 out:
1894         return ret;
1895 }
1896 ALLOW_ERROR_INJECTION(btrfs_check_node, ERRNO);
1897
1898 int btrfs_check_eb_owner(const struct extent_buffer *eb, u64 root_owner)
1899 {
1900         const bool is_subvol = is_fstree(root_owner);
1901         const u64 eb_owner = btrfs_header_owner(eb);
1902
1903         /*
1904          * Skip dummy fs, as selftests don't create unique ebs for each dummy
1905          * root.
1906          */
1907         if (test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &eb->fs_info->fs_state))
1908                 return 0;
1909         /*
1910          * There are several call sites (backref walking, qgroup, and data
1911          * reloc) passing 0 as @root_owner, as they are not holding the
1912          * tree root.  In that case, we can not do a reliable ownership check,
1913          * so just exit.
1914          */
1915         if (root_owner == 0)
1916                 return 0;
1917         /*
1918          * These trees use key.offset as their owner, our callers don't have
1919          * the extra capacity to pass key.offset here.  So we just skip them.
1920          */
1921         if (root_owner == BTRFS_TREE_LOG_OBJECTID ||
1922             root_owner == BTRFS_TREE_RELOC_OBJECTID)
1923                 return 0;
1924
1925         if (!is_subvol) {
1926                 /* For non-subvolume trees, the eb owner should match root owner */
1927                 if (unlikely(root_owner != eb_owner)) {
1928                         btrfs_crit(eb->fs_info,
1929 "corrupted %s, root=%llu block=%llu owner mismatch, have %llu expect %llu",
1930                                 btrfs_header_level(eb) == 0 ? "leaf" : "node",
1931                                 root_owner, btrfs_header_bytenr(eb), eb_owner,
1932                                 root_owner);
1933                         return -EUCLEAN;
1934                 }
1935                 return 0;
1936         }
1937
1938         /*
1939          * For subvolume trees, owners can mismatch, but they should all belong
1940          * to subvolume trees.
1941          */
1942         if (unlikely(is_subvol != is_fstree(eb_owner))) {
1943                 btrfs_crit(eb->fs_info,
1944 "corrupted %s, root=%llu block=%llu owner mismatch, have %llu expect [%llu, %llu]",
1945                         btrfs_header_level(eb) == 0 ? "leaf" : "node",
1946                         root_owner, btrfs_header_bytenr(eb), eb_owner,
1947                         BTRFS_FIRST_FREE_OBJECTID, BTRFS_LAST_FREE_OBJECTID);
1948                 return -EUCLEAN;
1949         }
1950         return 0;
1951 }