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