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 "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_nr(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_nr(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_nr(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_nr(leaf, slot - 1);
 368                 prev_csum_end = (prev_item_size / csumsize) * sectorsize;
 369                 prev_csum_end += prev_key->offset;
 370                 if (unlikely(prev_csum_end > key->offset)) {
 371                         generic_err(leaf, slot - 1,
 372 "csum end range (%llu) goes beyond the start range (%llu) of the next csum item",
 373                                     prev_csum_end, key->offset);
 374                         return -EUCLEAN;
 375                 }
 376         }
 377         return 0;
 378 }
 379
 380 /* Inode item error output has the same format as dir_item_err() */
 381 #define inode_item_err(eb, slot, fmt, ...)                      \
 382         dir_item_err(eb, slot, fmt, __VA_ARGS__)
 383
 384 static int check_inode_key(struct extent_buffer *leaf, struct btrfs_key *key,
 385                            int slot)
 386 {
 387         struct btrfs_key item_key;
 388         bool is_inode_item;
 389
 390         btrfs_item_key_to_cpu(leaf, &item_key, slot);
 391         is_inode_item = (item_key.type == BTRFS_INODE_ITEM_KEY);
 392
 393         /* For XATTR_ITEM, location key should be all 0 */
 394         if (item_key.type == BTRFS_XATTR_ITEM_KEY) {
 395                 if (unlikely(key->objectid != 0 || key->type != 0 ||
 396                              key->offset != 0))
 397                         return -EUCLEAN;
 398                 return 0;
 399         }
 400
 401         if (unlikely((key->objectid < BTRFS_FIRST_FREE_OBJECTID ||
 402                       key->objectid > BTRFS_LAST_FREE_OBJECTID) &&
 403                      key->objectid != BTRFS_ROOT_TREE_DIR_OBJECTID &&
 404                      key->objectid != BTRFS_FREE_INO_OBJECTID)) {
 405                 if (is_inode_item) {
 406                         generic_err(leaf, slot,
 407         "invalid key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
 408                                 key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
 409                                 BTRFS_FIRST_FREE_OBJECTID,
 410                                 BTRFS_LAST_FREE_OBJECTID,
 411                                 BTRFS_FREE_INO_OBJECTID);
 412                 } else {
 413                         dir_item_err(leaf, slot,
 414 "invalid location key objectid: has %llu expect %llu or [%llu, %llu] or %llu",
 415                                 key->objectid, BTRFS_ROOT_TREE_DIR_OBJECTID,
 416                                 BTRFS_FIRST_FREE_OBJECTID,
 417                                 BTRFS_LAST_FREE_OBJECTID,
 418                                 BTRFS_FREE_INO_OBJECTID);
 419                 }
 420                 return -EUCLEAN;
 421         }
 422         if (unlikely(key->offset != 0)) {
 423                 if (is_inode_item)
 424                         inode_item_err(leaf, slot,
 425                                        "invalid key offset: has %llu expect 0",
 426                                        key->offset);
 427                 else
 428                         dir_item_err(leaf, slot,
 429                                 "invalid location key offset:has %llu expect 0",
 430                                 key->offset);
 431                 return -EUCLEAN;
 432         }
 433         return 0;
 434 }
 435
 436 static int check_root_key(struct extent_buffer *leaf, struct btrfs_key *key,
 437                           int slot)
 438 {
 439         struct btrfs_key item_key;
 440         bool is_root_item;
 441
 442         btrfs_item_key_to_cpu(leaf, &item_key, slot);
 443         is_root_item = (item_key.type == BTRFS_ROOT_ITEM_KEY);
 444
 445         /* No such tree id */
 446         if (unlikely(key->objectid == 0)) {
 447                 if (is_root_item)
 448                         generic_err(leaf, slot, "invalid root id 0");
 449                 else
 450                         dir_item_err(leaf, slot,
 451                                      "invalid location key root id 0");
 452                 return -EUCLEAN;
 453         }
 454
 455         /* DIR_ITEM/INDEX/INODE_REF is not allowed to point to non-fs trees */
 456         if (unlikely(!is_fstree(key->objectid) && !is_root_item)) {
 457                 dir_item_err(leaf, slot,
 458                 "invalid location key objectid, have %llu expect [%llu, %llu]",
 459                                 key->objectid, BTRFS_FIRST_FREE_OBJECTID,
 460                                 BTRFS_LAST_FREE_OBJECTID);
 461                 return -EUCLEAN;
 462         }
 463
 464         /*
 465          * ROOT_ITEM with non-zero offset means this is a snapshot, created at
 466          * @offset transid.
 467          * Furthermore, for location key in DIR_ITEM, its offset is always -1.
 468          *
 469          * So here we only check offset for reloc tree whose key->offset must
 470          * be a valid tree.
 471          */
 472         if (unlikely(key->objectid == BTRFS_TREE_RELOC_OBJECTID &&
 473                      key->offset == 0)) {
 474                 generic_err(leaf, slot, "invalid root id 0 for reloc tree");
 475                 return -EUCLEAN;
 476         }
 477         return 0;
 478 }
 479
 480 static int check_dir_item(struct extent_buffer *leaf,
 481                           struct btrfs_key *key, struct btrfs_key *prev_key,
 482                           int slot)
 483 {
 484         struct btrfs_fs_info *fs_info = leaf->fs_info;
 485         struct btrfs_dir_item *di;
 486         u32 item_size = btrfs_item_size_nr(leaf, slot);
 487         u32 cur = 0;
 488
 489         if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
 490                 return -EUCLEAN;
 491
 492         di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
 493         while (cur < item_size) {
 494                 struct btrfs_key location_key;
 495                 u32 name_len;
 496                 u32 data_len;
 497                 u32 max_name_len;
 498                 u32 total_size;
 499                 u32 name_hash;
 500                 u8 dir_type;
 501                 int ret;
 502
 503                 /* header itself should not cross item boundary */
 504                 if (unlikely(cur + sizeof(*di) > item_size)) {
 505                         dir_item_err(leaf, slot,
 506                 "dir item header crosses item boundary, have %zu boundary %u",
 507                                 cur + sizeof(*di), item_size);
 508                         return -EUCLEAN;
 509                 }
 510
 511                 /* Location key check */
 512                 btrfs_dir_item_key_to_cpu(leaf, di, &location_key);
 513                 if (location_key.type == BTRFS_ROOT_ITEM_KEY) {
 514                         ret = check_root_key(leaf, &location_key, slot);
 515                         if (unlikely(ret < 0))
 516                                 return ret;
 517                 } else if (location_key.type == BTRFS_INODE_ITEM_KEY ||
 518                            location_key.type == 0) {
 519                         ret = check_inode_key(leaf, &location_key, slot);
 520                         if (unlikely(ret < 0))
 521                                 return ret;
 522                 } else {
 523                         dir_item_err(leaf, slot,
 524                         "invalid location key type, have %u, expect %u or %u",
 525                                      location_key.type, BTRFS_ROOT_ITEM_KEY,
 526                                      BTRFS_INODE_ITEM_KEY);
 527                         return -EUCLEAN;
 528                 }
 529
 530                 /* dir type check */
 531                 dir_type = btrfs_dir_type(leaf, di);
 532                 if (unlikely(dir_type >= BTRFS_FT_MAX)) {
 533                         dir_item_err(leaf, slot,
 534                         "invalid dir item type, have %u expect [0, %u)",
 535                                 dir_type, BTRFS_FT_MAX);
 536                         return -EUCLEAN;
 537                 }
 538
 539                 if (unlikely(key->type == BTRFS_XATTR_ITEM_KEY &&
 540                              dir_type != BTRFS_FT_XATTR)) {
 541                         dir_item_err(leaf, slot,
 542                 "invalid dir item type for XATTR key, have %u expect %u",
 543                                 dir_type, BTRFS_FT_XATTR);
 544                         return -EUCLEAN;
 545                 }
 546                 if (unlikely(dir_type == BTRFS_FT_XATTR &&
 547                              key->type != BTRFS_XATTR_ITEM_KEY)) {
 548                         dir_item_err(leaf, slot,
 549                         "xattr dir type found for non-XATTR key");
 550                         return -EUCLEAN;
 551                 }
 552                 if (dir_type == BTRFS_FT_XATTR)
 553                         max_name_len = XATTR_NAME_MAX;
 554                 else
 555                         max_name_len = BTRFS_NAME_LEN;
 556
 557                 /* Name/data length check */
 558                 name_len = btrfs_dir_name_len(leaf, di);
 559                 data_len = btrfs_dir_data_len(leaf, di);
 560                 if (unlikely(name_len > max_name_len)) {
 561                         dir_item_err(leaf, slot,
 562                         "dir item name len too long, have %u max %u",
 563                                 name_len, max_name_len);
 564                         return -EUCLEAN;
 565                 }
 566                 if (unlikely(name_len + data_len > BTRFS_MAX_XATTR_SIZE(fs_info))) {
 567                         dir_item_err(leaf, slot,
 568                         "dir item name and data len too long, have %u max %u",
 569                                 name_len + data_len,
 570                                 BTRFS_MAX_XATTR_SIZE(fs_info));
 571                         return -EUCLEAN;
 572                 }
 573
 574                 if (unlikely(data_len && dir_type != BTRFS_FT_XATTR)) {
 575                         dir_item_err(leaf, slot,
 576                         "dir item with invalid data len, have %u expect 0",
 577                                 data_len);
 578                         return -EUCLEAN;
 579                 }
 580
 581                 total_size = sizeof(*di) + name_len + data_len;
 582
 583                 /* header and name/data should not cross item boundary */
 584                 if (unlikely(cur + total_size > item_size)) {
 585                         dir_item_err(leaf, slot,
 586                 "dir item data crosses item boundary, have %u boundary %u",
 587                                 cur + total_size, item_size);
 588                         return -EUCLEAN;
 589                 }
 590
 591                 /*
 592                  * Special check for XATTR/DIR_ITEM, as key->offset is name
 593                  * hash, should match its name
 594                  */
 595                 if (key->type == BTRFS_DIR_ITEM_KEY ||
 596                     key->type == BTRFS_XATTR_ITEM_KEY) {
 597                         char namebuf[max(BTRFS_NAME_LEN, XATTR_NAME_MAX)];
 598
 599                         read_extent_buffer(leaf, namebuf,
 600                                         (unsigned long)(di + 1), name_len);
 601                         name_hash = btrfs_name_hash(namebuf, name_len);
 602                         if (unlikely(key->offset != name_hash)) {
 603                                 dir_item_err(leaf, slot,
 604                 "name hash mismatch with key, have 0x%016x expect 0x%016llx",
 605                                         name_hash, key->offset);
 606                                 return -EUCLEAN;
 607                         }
 608                 }
 609                 cur += total_size;
 610                 di = (struct btrfs_dir_item *)((void *)di + total_size);
 611         }
 612         return 0;
 613 }
 614
 615 __printf(3, 4)
 616 __cold
 617 static void block_group_err(const struct extent_buffer *eb, int slot,
 618                             const char *fmt, ...)
 619 {
 620         const struct btrfs_fs_info *fs_info = eb->fs_info;
 621         struct btrfs_key key;
 622         struct va_format vaf;
 623         va_list args;
 624
 625         btrfs_item_key_to_cpu(eb, &key, slot);
 626         va_start(args, fmt);
 627
 628         vaf.fmt = fmt;
 629         vaf.va = &args;
 630
 631         btrfs_crit(fs_info,
 632         "corrupt %s: root=%llu block=%llu slot=%d bg_start=%llu bg_len=%llu, %pV",
 633                 btrfs_header_level(eb) == 0 ? "leaf" : "node",
 634                 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
 635                 key.objectid, key.offset, &vaf);
 636         va_end(args);
 637 }
 638
 639 static int check_block_group_item(struct extent_buffer *leaf,
 640                                   struct btrfs_key *key, int slot)
 641 {
 642         struct btrfs_block_group_item bgi;
 643         u32 item_size = btrfs_item_size_nr(leaf, slot);
 644         u64 flags;
 645         u64 type;
 646
 647         /*
 648          * Here we don't really care about alignment since extent allocator can
 649          * handle it.  We care more about the size.
 650          */
 651         if (unlikely(key->offset == 0)) {
 652                 block_group_err(leaf, slot,
 653                                 "invalid block group size 0");
 654                 return -EUCLEAN;
 655         }
 656
 657         if (unlikely(item_size != sizeof(bgi))) {
 658                 block_group_err(leaf, slot,
 659                         "invalid item size, have %u expect %zu",
 660                                 item_size, sizeof(bgi));
 661                 return -EUCLEAN;
 662         }
 663
 664         read_extent_buffer(leaf, &bgi, btrfs_item_ptr_offset(leaf, slot),
 665                            sizeof(bgi));
 666         if (unlikely(btrfs_stack_block_group_chunk_objectid(&bgi) !=
 667                      BTRFS_FIRST_CHUNK_TREE_OBJECTID)) {
 668                 block_group_err(leaf, slot,
 669                 "invalid block group chunk objectid, have %llu expect %llu",
 670                                 btrfs_stack_block_group_chunk_objectid(&bgi),
 671                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
 672                 return -EUCLEAN;
 673         }
 674
 675         if (unlikely(btrfs_stack_block_group_used(&bgi) > key->offset)) {
 676                 block_group_err(leaf, slot,
 677                         "invalid block group used, have %llu expect [0, %llu)",
 678                                 btrfs_stack_block_group_used(&bgi), key->offset);
 679                 return -EUCLEAN;
 680         }
 681
 682         flags = btrfs_stack_block_group_flags(&bgi);
 683         if (unlikely(hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) > 1)) {
 684                 block_group_err(leaf, slot,
 685 "invalid profile flags, have 0x%llx (%lu bits set) expect no more than 1 bit set",
 686                         flags & BTRFS_BLOCK_GROUP_PROFILE_MASK,
 687                         hweight64(flags & BTRFS_BLOCK_GROUP_PROFILE_MASK));
 688                 return -EUCLEAN;
 689         }
 690
 691         type = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
 692         if (unlikely(type != BTRFS_BLOCK_GROUP_DATA &&
 693                      type != BTRFS_BLOCK_GROUP_METADATA &&
 694                      type != BTRFS_BLOCK_GROUP_SYSTEM &&
 695                      type != (BTRFS_BLOCK_GROUP_METADATA |
 696                               BTRFS_BLOCK_GROUP_DATA))) {
 697                 block_group_err(leaf, slot,
 698 "invalid type, have 0x%llx (%lu bits set) expect either 0x%llx, 0x%llx, 0x%llx or 0x%llx",
 699                         type, hweight64(type),
 700                         BTRFS_BLOCK_GROUP_DATA, BTRFS_BLOCK_GROUP_METADATA,
 701                         BTRFS_BLOCK_GROUP_SYSTEM,
 702                         BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA);
 703                 return -EUCLEAN;
 704         }
 705         return 0;
 706 }
 707
 708 __printf(4, 5)
 709 __cold
 710 static void chunk_err(const struct extent_buffer *leaf,
 711                       const struct btrfs_chunk *chunk, u64 logical,
 712                       const char *fmt, ...)
 713 {
 714         const struct btrfs_fs_info *fs_info = leaf->fs_info;
 715         bool is_sb;
 716         struct va_format vaf;
 717         va_list args;
 718         int i;
 719         int slot = -1;
 720
 721         /* Only superblock eb is able to have such small offset */
 722         is_sb = (leaf->start == BTRFS_SUPER_INFO_OFFSET);
 723
 724         if (!is_sb) {
 725                 /*
 726                  * Get the slot number by iterating through all slots, this
 727                  * would provide better readability.
 728                  */
 729                 for (i = 0; i < btrfs_header_nritems(leaf); i++) {
 730                         if (btrfs_item_ptr_offset(leaf, i) ==
 731                                         (unsigned long)chunk) {
 732                                 slot = i;
 733                                 break;
 734                         }
 735                 }
 736         }
 737         va_start(args, fmt);
 738         vaf.fmt = fmt;
 739         vaf.va = &args;
 740
 741         if (is_sb)
 742                 btrfs_crit(fs_info,
 743                 "corrupt superblock syschunk array: chunk_start=%llu, %pV",
 744                            logical, &vaf);
 745         else
 746                 btrfs_crit(fs_info,
 747         "corrupt leaf: root=%llu block=%llu slot=%d chunk_start=%llu, %pV",
 748                            BTRFS_CHUNK_TREE_OBJECTID, leaf->start, slot,
 749                            logical, &vaf);
 750         va_end(args);
 751 }
 752
 753 /*
 754  * The common chunk check which could also work on super block sys chunk array.
 755  *
 756  * Return -EUCLEAN if anything is corrupted.
 757  * Return 0 if everything is OK.
 758  */
 759 int btrfs_check_chunk_valid(struct extent_buffer *leaf,
 760                             struct btrfs_chunk *chunk, u64 logical)
 761 {
 762         struct btrfs_fs_info *fs_info = leaf->fs_info;
 763         u64 length;
 764         u64 chunk_end;
 765         u64 stripe_len;
 766         u16 num_stripes;
 767         u16 sub_stripes;
 768         u64 type;
 769         u64 features;
 770         bool mixed = false;
 771         int raid_index;
 772         int nparity;
 773         int ncopies;
 774
 775         length = btrfs_chunk_length(leaf, chunk);
 776         stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
 777         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
 778         sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
 779         type = btrfs_chunk_type(leaf, chunk);
 780         raid_index = btrfs_bg_flags_to_raid_index(type);
 781         ncopies = btrfs_raid_array[raid_index].ncopies;
 782         nparity = btrfs_raid_array[raid_index].nparity;
 783
 784         if (unlikely(!num_stripes)) {
 785                 chunk_err(leaf, chunk, logical,
 786                           "invalid chunk num_stripes, have %u", num_stripes);
 787                 return -EUCLEAN;
 788         }
 789         if (unlikely(num_stripes < ncopies)) {
 790                 chunk_err(leaf, chunk, logical,
 791                           "invalid chunk num_stripes < ncopies, have %u < %d",
 792                           num_stripes, ncopies);
 793                 return -EUCLEAN;
 794         }
 795         if (unlikely(nparity && num_stripes == nparity)) {
 796                 chunk_err(leaf, chunk, logical,
 797                           "invalid chunk num_stripes == nparity, have %u == %d",
 798                           num_stripes, nparity);
 799                 return -EUCLEAN;
 800         }
 801         if (unlikely(!IS_ALIGNED(logical, fs_info->sectorsize))) {
 802                 chunk_err(leaf, chunk, logical,
 803                 "invalid chunk logical, have %llu should aligned to %u",
 804                           logical, fs_info->sectorsize);
 805                 return -EUCLEAN;
 806         }
 807         if (unlikely(btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize)) {
 808                 chunk_err(leaf, chunk, logical,
 809                           "invalid chunk sectorsize, have %u expect %u",
 810                           btrfs_chunk_sector_size(leaf, chunk),
 811                           fs_info->sectorsize);
 812                 return -EUCLEAN;
 813         }
 814         if (unlikely(!length || !IS_ALIGNED(length, fs_info->sectorsize))) {
 815                 chunk_err(leaf, chunk, logical,
 816                           "invalid chunk length, have %llu", length);
 817                 return -EUCLEAN;
 818         }
 819         if (unlikely(check_add_overflow(logical, length, &chunk_end))) {
 820                 chunk_err(leaf, chunk, logical,
 821 "invalid chunk logical start and length, have logical start %llu length %llu",
 822                           logical, length);
 823                 return -EUCLEAN;
 824         }
 825         if (unlikely(!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN)) {
 826                 chunk_err(leaf, chunk, logical,
 827                           "invalid chunk stripe length: %llu",
 828                           stripe_len);
 829                 return -EUCLEAN;
 830         }
 831         if (unlikely(type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
 832                               BTRFS_BLOCK_GROUP_PROFILE_MASK))) {
 833                 chunk_err(leaf, chunk, logical,
 834                           "unrecognized chunk type: 0x%llx",
 835                           ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
 836                             BTRFS_BLOCK_GROUP_PROFILE_MASK) &
 837                           btrfs_chunk_type(leaf, chunk));
 838                 return -EUCLEAN;
 839         }
 840
 841         if (unlikely(!has_single_bit_set(type & BTRFS_BLOCK_GROUP_PROFILE_MASK) &&
 842                      (type & BTRFS_BLOCK_GROUP_PROFILE_MASK) != 0)) {
 843                 chunk_err(leaf, chunk, logical,
 844                 "invalid chunk profile flag: 0x%llx, expect 0 or 1 bit set",
 845                           type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
 846                 return -EUCLEAN;
 847         }
 848         if (unlikely((type & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0)) {
 849                 chunk_err(leaf, chunk, logical,
 850         "missing chunk type flag, have 0x%llx one bit must be set in 0x%llx",
 851                           type, BTRFS_BLOCK_GROUP_TYPE_MASK);
 852                 return -EUCLEAN;
 853         }
 854
 855         if (unlikely((type & BTRFS_BLOCK_GROUP_SYSTEM) &&
 856                      (type & (BTRFS_BLOCK_GROUP_METADATA |
 857                               BTRFS_BLOCK_GROUP_DATA)))) {
 858                 chunk_err(leaf, chunk, logical,
 859                           "system chunk with data or metadata type: 0x%llx",
 860                           type);
 861                 return -EUCLEAN;
 862         }
 863
 864         features = btrfs_super_incompat_flags(fs_info->super_copy);
 865         if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
 866                 mixed = true;
 867
 868         if (!mixed) {
 869                 if (unlikely((type & BTRFS_BLOCK_GROUP_METADATA) &&
 870                              (type & BTRFS_BLOCK_GROUP_DATA))) {
 871                         chunk_err(leaf, chunk, logical,
 872                         "mixed chunk type in non-mixed mode: 0x%llx", type);
 873                         return -EUCLEAN;
 874                 }
 875         }
 876
 877         if (unlikely((type & BTRFS_BLOCK_GROUP_RAID10 &&
 878                       sub_stripes != btrfs_raid_array[BTRFS_RAID_RAID10].sub_stripes) ||
 879                      (type & BTRFS_BLOCK_GROUP_RAID1 &&
 880                       num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1].devs_min) ||
 881                      (type & BTRFS_BLOCK_GROUP_RAID1C3 &&
 882                       num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1C3].devs_min) ||
 883                      (type & BTRFS_BLOCK_GROUP_RAID1C4 &&
 884                       num_stripes != btrfs_raid_array[BTRFS_RAID_RAID1C4].devs_min) ||
 885                      (type & BTRFS_BLOCK_GROUP_RAID5 &&
 886                       num_stripes < btrfs_raid_array[BTRFS_RAID_RAID5].devs_min) ||
 887                      (type & BTRFS_BLOCK_GROUP_RAID6 &&
 888                       num_stripes < btrfs_raid_array[BTRFS_RAID_RAID6].devs_min) ||
 889                      (type & BTRFS_BLOCK_GROUP_DUP &&
 890                       num_stripes != btrfs_raid_array[BTRFS_RAID_DUP].dev_stripes) ||
 891                      ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
 892                       num_stripes != btrfs_raid_array[BTRFS_RAID_SINGLE].dev_stripes))) {
 893                 chunk_err(leaf, chunk, logical,
 894                         "invalid num_stripes:sub_stripes %u:%u for profile %llu",
 895                         num_stripes, sub_stripes,
 896                         type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
 897                 return -EUCLEAN;
 898         }
 899
 900         return 0;
 901 }
 902
 903 /*
 904  * Enhanced version of chunk item checker.
 905  *
 906  * The common btrfs_check_chunk_valid() doesn't check item size since it needs
 907  * to work on super block sys_chunk_array which doesn't have full item ptr.
 908  */
 909 static int check_leaf_chunk_item(struct extent_buffer *leaf,
 910                                  struct btrfs_chunk *chunk,
 911                                  struct btrfs_key *key, int slot)
 912 {
 913         int num_stripes;
 914
 915         if (unlikely(btrfs_item_size_nr(leaf, slot) < sizeof(struct btrfs_chunk))) {
 916                 chunk_err(leaf, chunk, key->offset,
 917                         "invalid chunk item size: have %u expect [%zu, %u)",
 918                         btrfs_item_size_nr(leaf, slot),
 919                         sizeof(struct btrfs_chunk),
 920                         BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
 921                 return -EUCLEAN;
 922         }
 923
 924         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
 925         /* Let btrfs_check_chunk_valid() handle this error type */
 926         if (num_stripes == 0)
 927                 goto out;
 928
 929         if (unlikely(btrfs_chunk_item_size(num_stripes) !=
 930                      btrfs_item_size_nr(leaf, slot))) {
 931                 chunk_err(leaf, chunk, key->offset,
 932                         "invalid chunk item size: have %u expect %lu",
 933                         btrfs_item_size_nr(leaf, slot),
 934                         btrfs_chunk_item_size(num_stripes));
 935                 return -EUCLEAN;
 936         }
 937 out:
 938         return btrfs_check_chunk_valid(leaf, chunk, key->offset);
 939 }
 940
 941 __printf(3, 4)
 942 __cold
 943 static void dev_item_err(const struct extent_buffer *eb, int slot,
 944                          const char *fmt, ...)
 945 {
 946         struct btrfs_key key;
 947         struct va_format vaf;
 948         va_list args;
 949
 950         btrfs_item_key_to_cpu(eb, &key, slot);
 951         va_start(args, fmt);
 952
 953         vaf.fmt = fmt;
 954         vaf.va = &args;
 955
 956         btrfs_crit(eb->fs_info,
 957         "corrupt %s: root=%llu block=%llu slot=%d devid=%llu %pV",
 958                 btrfs_header_level(eb) == 0 ? "leaf" : "node",
 959                 btrfs_header_owner(eb), btrfs_header_bytenr(eb), slot,
 960                 key.objectid, &vaf);
 961         va_end(args);
 962 }
 963
 964 static int check_dev_item(struct extent_buffer *leaf,
 965                           struct btrfs_key *key, int slot)
 966 {
 967         struct btrfs_dev_item *ditem;
 968
 969         if (unlikely(key->objectid != BTRFS_DEV_ITEMS_OBJECTID)) {
 970                 dev_item_err(leaf, slot,
 971                              "invalid objectid: has=%llu expect=%llu",
 972                              key->objectid, BTRFS_DEV_ITEMS_OBJECTID);
 973                 return -EUCLEAN;
 974         }
 975         ditem = btrfs_item_ptr(leaf, slot, struct btrfs_dev_item);
 976         if (unlikely(btrfs_device_id(leaf, ditem) != key->offset)) {
 977                 dev_item_err(leaf, slot,
 978                              "devid mismatch: key has=%llu item has=%llu",
 979                              key->offset, btrfs_device_id(leaf, ditem));
 980                 return -EUCLEAN;
 981         }
 982
 983         /*
 984          * For device total_bytes, we don't have reliable way to check it, as
 985          * it can be 0 for device removal. Device size check can only be done
 986          * by dev extents check.
 987          */
 988         if (unlikely(btrfs_device_bytes_used(leaf, ditem) >
 989                      btrfs_device_total_bytes(leaf, ditem))) {
 990                 dev_item_err(leaf, slot,
 991                              "invalid bytes used: have %llu expect [0, %llu]",
 992                              btrfs_device_bytes_used(leaf, ditem),
 993                              btrfs_device_total_bytes(leaf, ditem));
 994                 return -EUCLEAN;
 995         }
 996         /*
 997          * Remaining members like io_align/type/gen/dev_group aren't really
 998          * utilized.  Skip them to make later usage of them easier.
 999          */
1000         return 0;
1001 }
1002
1003 static int check_inode_item(struct extent_buffer *leaf,
1004                             struct btrfs_key *key, int slot)
1005 {
1006         struct btrfs_fs_info *fs_info = leaf->fs_info;
1007         struct btrfs_inode_item *iitem;
1008         u64 super_gen = btrfs_super_generation(fs_info->super_copy);
1009         u32 valid_mask = (S_IFMT | S_ISUID | S_ISGID | S_ISVTX | 0777);
1010         u32 mode;
1011         int ret;
1012         u32 flags;
1013         u32 ro_flags;
1014
1015         ret = check_inode_key(leaf, key, slot);
1016         if (unlikely(ret < 0))
1017                 return ret;
1018
1019         iitem = btrfs_item_ptr(leaf, slot, struct btrfs_inode_item);
1020
1021         /* Here we use super block generation + 1 to handle log tree */
1022         if (unlikely(btrfs_inode_generation(leaf, iitem) > super_gen + 1)) {
1023                 inode_item_err(leaf, slot,
1024                         "invalid inode generation: has %llu expect (0, %llu]",
1025                                btrfs_inode_generation(leaf, iitem),
1026                                super_gen + 1);
1027                 return -EUCLEAN;
1028         }
1029         /* Note for ROOT_TREE_DIR_ITEM, mkfs could set its transid 0 */
1030         if (unlikely(btrfs_inode_transid(leaf, iitem) > super_gen + 1)) {
1031                 inode_item_err(leaf, slot,
1032                         "invalid inode transid: has %llu expect [0, %llu]",
1033                                btrfs_inode_transid(leaf, iitem), super_gen + 1);
1034                 return -EUCLEAN;
1035         }
1036
1037         /*
1038          * For size and nbytes it's better not to be too strict, as for dir
1039          * item its size/nbytes can easily get wrong, but doesn't affect
1040          * anything in the fs. So here we skip the check.
1041          */
1042         mode = btrfs_inode_mode(leaf, iitem);
1043         if (unlikely(mode & ~valid_mask)) {
1044                 inode_item_err(leaf, slot,
1045                                "unknown mode bit detected: 0x%x",
1046                                mode & ~valid_mask);
1047                 return -EUCLEAN;
1048         }
1049
1050         /*
1051          * S_IFMT is not bit mapped so we can't completely rely on
1052          * is_power_of_2/has_single_bit_set, but it can save us from checking
1053          * FIFO/CHR/DIR/REG.  Only needs to check BLK, LNK and SOCKS
1054          */
1055         if (!has_single_bit_set(mode & S_IFMT)) {
1056                 if (unlikely(!S_ISLNK(mode) && !S_ISBLK(mode) && !S_ISSOCK(mode))) {
1057                         inode_item_err(leaf, slot,
1058                         "invalid mode: has 0%o expect valid S_IF* bit(s)",
1059                                        mode & S_IFMT);
1060                         return -EUCLEAN;
1061                 }
1062         }
1063         if (unlikely(S_ISDIR(mode) && btrfs_inode_nlink(leaf, iitem) > 1)) {
1064                 inode_item_err(leaf, slot,
1065                        "invalid nlink: has %u expect no more than 1 for dir",
1066                         btrfs_inode_nlink(leaf, iitem));
1067                 return -EUCLEAN;
1068         }
1069         btrfs_inode_split_flags(btrfs_inode_flags(leaf, iitem), &flags, &ro_flags);
1070         if (unlikely(flags & ~BTRFS_INODE_FLAG_MASK)) {
1071                 inode_item_err(leaf, slot,
1072                                "unknown incompat flags detected: 0x%x", flags);
1073                 return -EUCLEAN;
1074         }
1075         if (unlikely(!sb_rdonly(fs_info->sb) &&
1076                      (ro_flags & ~BTRFS_INODE_RO_FLAG_MASK))) {
1077                 inode_item_err(leaf, slot,
1078                         "unknown ro-compat flags detected on writeable mount: 0x%x",
1079                         ro_flags);
1080                 return -EUCLEAN;
1081         }
1082         return 0;
1083 }
1084
1085 static int check_root_item(struct extent_buffer *leaf, struct btrfs_key *key,
1086                            int slot)
1087 {
1088         struct btrfs_fs_info *fs_info = leaf->fs_info;
1089         struct btrfs_root_item ri = { 0 };
1090         const u64 valid_root_flags = BTRFS_ROOT_SUBVOL_RDONLY |
1091                                      BTRFS_ROOT_SUBVOL_DEAD;
1092         int ret;
1093
1094         ret = check_root_key(leaf, key, slot);
1095         if (unlikely(ret < 0))
1096                 return ret;
1097
1098         if (unlikely(btrfs_item_size_nr(leaf, slot) != sizeof(ri) &&
1099                      btrfs_item_size_nr(leaf, slot) !=
1100                      btrfs_legacy_root_item_size())) {
1101                 generic_err(leaf, slot,
1102                             "invalid root item size, have %u expect %zu or %u",
1103                             btrfs_item_size_nr(leaf, slot), sizeof(ri),
1104                             btrfs_legacy_root_item_size());
1105                 return -EUCLEAN;
1106         }
1107
1108         /*
1109          * For legacy root item, the members starting at generation_v2 will be
1110          * all filled with 0.
1111          * And since we allow geneartion_v2 as 0, it will still pass the check.
1112          */
1113         read_extent_buffer(leaf, &ri, btrfs_item_ptr_offset(leaf, slot),
1114                            btrfs_item_size_nr(leaf, slot));
1115
1116         /* Generation related */
1117         if (unlikely(btrfs_root_generation(&ri) >
1118                      btrfs_super_generation(fs_info->super_copy) + 1)) {
1119                 generic_err(leaf, slot,
1120                         "invalid root generation, have %llu expect (0, %llu]",
1121                             btrfs_root_generation(&ri),
1122                             btrfs_super_generation(fs_info->super_copy) + 1);
1123                 return -EUCLEAN;
1124         }
1125         if (unlikely(btrfs_root_generation_v2(&ri) >
1126                      btrfs_super_generation(fs_info->super_copy) + 1)) {
1127                 generic_err(leaf, slot,
1128                 "invalid root v2 generation, have %llu expect (0, %llu]",
1129                             btrfs_root_generation_v2(&ri),
1130                             btrfs_super_generation(fs_info->super_copy) + 1);
1131                 return -EUCLEAN;
1132         }
1133         if (unlikely(btrfs_root_last_snapshot(&ri) >
1134                      btrfs_super_generation(fs_info->super_copy) + 1)) {
1135                 generic_err(leaf, slot,
1136                 "invalid root last_snapshot, have %llu expect (0, %llu]",
1137                             btrfs_root_last_snapshot(&ri),
1138                             btrfs_super_generation(fs_info->super_copy) + 1);
1139                 return -EUCLEAN;
1140         }
1141
1142         /* Alignment and level check */
1143         if (unlikely(!IS_ALIGNED(btrfs_root_bytenr(&ri), fs_info->sectorsize))) {
1144                 generic_err(leaf, slot,
1145                 "invalid root bytenr, have %llu expect to be aligned to %u",
1146                             btrfs_root_bytenr(&ri), fs_info->sectorsize);
1147                 return -EUCLEAN;
1148         }
1149         if (unlikely(btrfs_root_level(&ri) >= BTRFS_MAX_LEVEL)) {
1150                 generic_err(leaf, slot,
1151                             "invalid root level, have %u expect [0, %u]",
1152                             btrfs_root_level(&ri), BTRFS_MAX_LEVEL - 1);
1153                 return -EUCLEAN;
1154         }
1155         if (unlikely(btrfs_root_drop_level(&ri) >= BTRFS_MAX_LEVEL)) {
1156                 generic_err(leaf, slot,
1157                             "invalid root level, have %u expect [0, %u]",
1158                             btrfs_root_drop_level(&ri), BTRFS_MAX_LEVEL - 1);
1159                 return -EUCLEAN;
1160         }
1161
1162         /* Flags check */
1163         if (unlikely(btrfs_root_flags(&ri) & ~valid_root_flags)) {
1164                 generic_err(leaf, slot,
1165                             "invalid root flags, have 0x%llx expect mask 0x%llx",
1166                             btrfs_root_flags(&ri), valid_root_flags);
1167                 return -EUCLEAN;
1168         }
1169         return 0;
1170 }
1171
1172 __printf(3,4)
1173 __cold
1174 static void extent_err(const struct extent_buffer *eb, int slot,
1175                        const char *fmt, ...)
1176 {
1177         struct btrfs_key key;
1178         struct va_format vaf;
1179         va_list args;
1180         u64 bytenr;
1181         u64 len;
1182
1183         btrfs_item_key_to_cpu(eb, &key, slot);
1184         bytenr = key.objectid;
1185         if (key.type == BTRFS_METADATA_ITEM_KEY ||
1186             key.type == BTRFS_TREE_BLOCK_REF_KEY ||
1187             key.type == BTRFS_SHARED_BLOCK_REF_KEY)
1188                 len = eb->fs_info->nodesize;
1189         else
1190                 len = key.offset;
1191         va_start(args, fmt);
1192
1193         vaf.fmt = fmt;
1194         vaf.va = &args;
1195
1196         btrfs_crit(eb->fs_info,
1197         "corrupt %s: block=%llu slot=%d extent bytenr=%llu len=%llu %pV",
1198                 btrfs_header_level(eb) == 0 ? "leaf" : "node",
1199                 eb->start, slot, bytenr, len, &vaf);
1200         va_end(args);
1201 }
1202
1203 static int check_extent_item(struct extent_buffer *leaf,
1204                              struct btrfs_key *key, int slot)
1205 {
1206         struct btrfs_fs_info *fs_info = leaf->fs_info;
1207         struct btrfs_extent_item *ei;
1208         bool is_tree_block = false;
1209         unsigned long ptr;      /* Current pointer inside inline refs */
1210         unsigned long end;      /* Extent item end */
1211         const u32 item_size = btrfs_item_size_nr(leaf, slot);
1212         u64 flags;
1213         u64 generation;
1214         u64 total_refs;         /* Total refs in btrfs_extent_item */
1215         u64 inline_refs = 0;    /* found total inline refs */
1216
1217         if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1218                      !btrfs_fs_incompat(fs_info, SKINNY_METADATA))) {
1219                 generic_err(leaf, slot,
1220 "invalid key type, METADATA_ITEM type invalid when SKINNY_METADATA feature disabled");
1221                 return -EUCLEAN;
1222         }
1223         /* key->objectid is the bytenr for both key types */
1224         if (unlikely(!IS_ALIGNED(key->objectid, fs_info->sectorsize))) {
1225                 generic_err(leaf, slot,
1226                 "invalid key objectid, have %llu expect to be aligned to %u",
1227                            key->objectid, fs_info->sectorsize);
1228                 return -EUCLEAN;
1229         }
1230
1231         /* key->offset is tree level for METADATA_ITEM_KEY */
1232         if (unlikely(key->type == BTRFS_METADATA_ITEM_KEY &&
1233                      key->offset >= BTRFS_MAX_LEVEL)) {
1234                 extent_err(leaf, slot,
1235                            "invalid tree level, have %llu expect [0, %u]",
1236                            key->offset, BTRFS_MAX_LEVEL - 1);
1237                 return -EUCLEAN;
1238         }
1239
1240         /*
1241          * EXTENT/METADATA_ITEM consists of:
1242          * 1) One btrfs_extent_item
1243          *    Records the total refs, type and generation of the extent.
1244          *
1245          * 2) One btrfs_tree_block_info (for EXTENT_ITEM and tree backref only)
1246          *    Records the first key and level of the tree block.
1247          *
1248          * 2) Zero or more btrfs_extent_inline_ref(s)
1249          *    Each inline ref has one btrfs_extent_inline_ref shows:
1250          *    2.1) The ref type, one of the 4
1251          *         TREE_BLOCK_REF       Tree block only
1252          *         SHARED_BLOCK_REF     Tree block only
1253          *         EXTENT_DATA_REF      Data only
1254          *         SHARED_DATA_REF      Data only
1255          *    2.2) Ref type specific data
1256          *         Either using btrfs_extent_inline_ref::offset, or specific
1257          *         data structure.
1258          */
1259         if (unlikely(item_size < sizeof(*ei))) {
1260                 extent_err(leaf, slot,
1261                            "invalid item size, have %u expect [%zu, %u)",
1262                            item_size, sizeof(*ei),
1263                            BTRFS_LEAF_DATA_SIZE(fs_info));
1264                 return -EUCLEAN;
1265         }
1266         end = item_size + btrfs_item_ptr_offset(leaf, slot);
1267
1268         /* Checks against extent_item */
1269         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
1270         flags = btrfs_extent_flags(leaf, ei);
1271         total_refs = btrfs_extent_refs(leaf, ei);
1272         generation = btrfs_extent_generation(leaf, ei);
1273         if (unlikely(generation >
1274                      btrfs_super_generation(fs_info->super_copy) + 1)) {
1275                 extent_err(leaf, slot,
1276                            "invalid generation, have %llu expect (0, %llu]",
1277                            generation,
1278                            btrfs_super_generation(fs_info->super_copy) + 1);
1279                 return -EUCLEAN;
1280         }
1281         if (unlikely(!has_single_bit_set(flags & (BTRFS_EXTENT_FLAG_DATA |
1282                                                   BTRFS_EXTENT_FLAG_TREE_BLOCK)))) {
1283                 extent_err(leaf, slot,
1284                 "invalid extent flag, have 0x%llx expect 1 bit set in 0x%llx",
1285                         flags, BTRFS_EXTENT_FLAG_DATA |
1286                         BTRFS_EXTENT_FLAG_TREE_BLOCK);
1287                 return -EUCLEAN;
1288         }
1289         is_tree_block = !!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK);
1290         if (is_tree_block) {
1291                 if (unlikely(key->type == BTRFS_EXTENT_ITEM_KEY &&
1292                              key->offset != fs_info->nodesize)) {
1293                         extent_err(leaf, slot,
1294                                    "invalid extent length, have %llu expect %u",
1295                                    key->offset, fs_info->nodesize);
1296                         return -EUCLEAN;
1297                 }
1298         } else {
1299                 if (unlikely(key->type != BTRFS_EXTENT_ITEM_KEY)) {
1300                         extent_err(leaf, slot,
1301                         "invalid key type, have %u expect %u for data backref",
1302                                    key->type, BTRFS_EXTENT_ITEM_KEY);
1303                         return -EUCLEAN;
1304                 }
1305                 if (unlikely(!IS_ALIGNED(key->offset, fs_info->sectorsize))) {
1306                         extent_err(leaf, slot,
1307                         "invalid extent length, have %llu expect aligned to %u",
1308                                    key->offset, fs_info->sectorsize);
1309                         return -EUCLEAN;
1310                 }
1311                 if (unlikely(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
1312                         extent_err(leaf, slot,
1313                         "invalid extent flag, data has full backref set");
1314                         return -EUCLEAN;
1315                 }
1316         }
1317         ptr = (unsigned long)(struct btrfs_extent_item *)(ei + 1);
1318
1319         /* Check the special case of btrfs_tree_block_info */
1320         if (is_tree_block && key->type != BTRFS_METADATA_ITEM_KEY) {
1321                 struct btrfs_tree_block_info *info;
1322
1323                 info = (struct btrfs_tree_block_info *)ptr;
1324                 if (unlikely(btrfs_tree_block_level(leaf, info) >= BTRFS_MAX_LEVEL)) {
1325                         extent_err(leaf, slot,
1326                         "invalid tree block info level, have %u expect [0, %u]",
1327                                    btrfs_tree_block_level(leaf, info),
1328                                    BTRFS_MAX_LEVEL - 1);
1329                         return -EUCLEAN;
1330                 }
1331                 ptr = (unsigned long)(struct btrfs_tree_block_info *)(info + 1);
1332         }
1333
1334         /* Check inline refs */
1335         while (ptr < end) {
1336                 struct btrfs_extent_inline_ref *iref;
1337                 struct btrfs_extent_data_ref *dref;
1338                 struct btrfs_shared_data_ref *sref;
1339                 u64 dref_offset;
1340                 u64 inline_offset;
1341                 u8 inline_type;
1342
1343                 if (unlikely(ptr + sizeof(*iref) > end)) {
1344                         extent_err(leaf, slot,
1345 "inline ref item overflows extent item, ptr %lu iref size %zu end %lu",
1346                                    ptr, sizeof(*iref), end);
1347                         return -EUCLEAN;
1348                 }
1349                 iref = (struct btrfs_extent_inline_ref *)ptr;
1350                 inline_type = btrfs_extent_inline_ref_type(leaf, iref);
1351                 inline_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1352                 if (unlikely(ptr + btrfs_extent_inline_ref_size(inline_type) > end)) {
1353                         extent_err(leaf, slot,
1354 "inline ref item overflows extent item, ptr %lu iref size %u end %lu",
1355                                    ptr, inline_type, end);
1356                         return -EUCLEAN;
1357                 }
1358
1359                 switch (inline_type) {
1360                 /* inline_offset is subvolid of the owner, no need to check */
1361                 case BTRFS_TREE_BLOCK_REF_KEY:
1362                         inline_refs++;
1363                         break;
1364                 /* Contains parent bytenr */
1365                 case BTRFS_SHARED_BLOCK_REF_KEY:
1366                         if (unlikely(!IS_ALIGNED(inline_offset,
1367                                                  fs_info->sectorsize))) {
1368                                 extent_err(leaf, slot,
1369                 "invalid tree parent bytenr, have %llu expect aligned to %u",
1370                                            inline_offset, fs_info->sectorsize);
1371                                 return -EUCLEAN;
1372                         }
1373                         inline_refs++;
1374                         break;
1375                 /*
1376                  * Contains owner subvolid, owner key objectid, adjusted offset.
1377                  * The only obvious corruption can happen in that offset.
1378                  */
1379                 case BTRFS_EXTENT_DATA_REF_KEY:
1380                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1381                         dref_offset = btrfs_extent_data_ref_offset(leaf, dref);
1382                         if (unlikely(!IS_ALIGNED(dref_offset,
1383                                                  fs_info->sectorsize))) {
1384                                 extent_err(leaf, slot,
1385                 "invalid data ref offset, have %llu expect aligned to %u",
1386                                            dref_offset, fs_info->sectorsize);
1387                                 return -EUCLEAN;
1388                         }
1389                         inline_refs += btrfs_extent_data_ref_count(leaf, dref);
1390                         break;
1391                 /* Contains parent bytenr and ref count */
1392                 case BTRFS_SHARED_DATA_REF_KEY:
1393                         sref = (struct btrfs_shared_data_ref *)(iref + 1);
1394                         if (unlikely(!IS_ALIGNED(inline_offset,
1395                                                  fs_info->sectorsize))) {
1396                                 extent_err(leaf, slot,
1397                 "invalid data parent bytenr, have %llu expect aligned to %u",
1398                                            inline_offset, fs_info->sectorsize);
1399                                 return -EUCLEAN;
1400                         }
1401                         inline_refs += btrfs_shared_data_ref_count(leaf, sref);
1402                         break;
1403                 default:
1404                         extent_err(leaf, slot, "unknown inline ref type: %u",
1405                                    inline_type);
1406                         return -EUCLEAN;
1407                 }
1408                 ptr += btrfs_extent_inline_ref_size(inline_type);
1409         }
1410         /* No padding is allowed */
1411         if (unlikely(ptr != end)) {
1412                 extent_err(leaf, slot,
1413                            "invalid extent item size, padding bytes found");
1414                 return -EUCLEAN;
1415         }
1416
1417         /* Finally, check the inline refs against total refs */
1418         if (unlikely(inline_refs > total_refs)) {
1419                 extent_err(leaf, slot,
1420                         "invalid extent refs, have %llu expect >= inline %llu",
1421                            total_refs, inline_refs);
1422                 return -EUCLEAN;
1423         }
1424         return 0;
1425 }
1426
1427 static int check_simple_keyed_refs(struct extent_buffer *leaf,
1428                                    struct btrfs_key *key, int slot)
1429 {
1430         u32 expect_item_size = 0;
1431
1432         if (key->type == BTRFS_SHARED_DATA_REF_KEY)
1433                 expect_item_size = sizeof(struct btrfs_shared_data_ref);
1434
1435         if (unlikely(btrfs_item_size_nr(leaf, slot) != expect_item_size)) {
1436                 generic_err(leaf, slot,
1437                 "invalid item size, have %u expect %u for key type %u",
1438                             btrfs_item_size_nr(leaf, slot),
1439                             expect_item_size, key->type);
1440                 return -EUCLEAN;
1441         }
1442         if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1443                 generic_err(leaf, slot,
1444 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1445                             key->objectid, leaf->fs_info->sectorsize);
1446                 return -EUCLEAN;
1447         }
1448         if (unlikely(key->type != BTRFS_TREE_BLOCK_REF_KEY &&
1449                      !IS_ALIGNED(key->offset, leaf->fs_info->sectorsize))) {
1450                 extent_err(leaf, slot,
1451                 "invalid tree parent bytenr, have %llu expect aligned to %u",
1452                            key->offset, leaf->fs_info->sectorsize);
1453                 return -EUCLEAN;
1454         }
1455         return 0;
1456 }
1457
1458 static int check_extent_data_ref(struct extent_buffer *leaf,
1459                                  struct btrfs_key *key, int slot)
1460 {
1461         struct btrfs_extent_data_ref *dref;
1462         unsigned long ptr = btrfs_item_ptr_offset(leaf, slot);
1463         const unsigned long end = ptr + btrfs_item_size_nr(leaf, slot);
1464
1465         if (unlikely(btrfs_item_size_nr(leaf, slot) % sizeof(*dref) != 0)) {
1466                 generic_err(leaf, slot,
1467         "invalid item size, have %u expect aligned to %zu for key type %u",
1468                             btrfs_item_size_nr(leaf, slot),
1469                             sizeof(*dref), key->type);
1470                 return -EUCLEAN;
1471         }
1472         if (unlikely(!IS_ALIGNED(key->objectid, leaf->fs_info->sectorsize))) {
1473                 generic_err(leaf, slot,
1474 "invalid key objectid for shared block ref, have %llu expect aligned to %u",
1475                             key->objectid, leaf->fs_info->sectorsize);
1476                 return -EUCLEAN;
1477         }
1478         for (; ptr < end; ptr += sizeof(*dref)) {
1479                 u64 offset;
1480
1481                 /*
1482                  * We cannot check the extent_data_ref hash due to possible
1483                  * overflow from the leaf due to hash collisions.
1484                  */
1485                 dref = (struct btrfs_extent_data_ref *)ptr;
1486                 offset = btrfs_extent_data_ref_offset(leaf, dref);
1487                 if (unlikely(!IS_ALIGNED(offset, leaf->fs_info->sectorsize))) {
1488                         extent_err(leaf, slot,
1489         "invalid extent data backref offset, have %llu expect aligned to %u",
1490                                    offset, leaf->fs_info->sectorsize);
1491                         return -EUCLEAN;
1492                 }
1493         }
1494         return 0;
1495 }
1496
1497 #define inode_ref_err(eb, slot, fmt, args...)                   \
1498         inode_item_err(eb, slot, fmt, ##args)
1499 static int check_inode_ref(struct extent_buffer *leaf,
1500                            struct btrfs_key *key, struct btrfs_key *prev_key,
1501                            int slot)
1502 {
1503         struct btrfs_inode_ref *iref;
1504         unsigned long ptr;
1505         unsigned long end;
1506
1507         if (unlikely(!check_prev_ino(leaf, key, slot, prev_key)))
1508                 return -EUCLEAN;
1509         /* namelen can't be 0, so item_size == sizeof() is also invalid */
1510         if (unlikely(btrfs_item_size_nr(leaf, slot) <= sizeof(*iref))) {
1511                 inode_ref_err(leaf, slot,
1512                         "invalid item size, have %u expect (%zu, %u)",
1513                         btrfs_item_size_nr(leaf, slot),
1514                         sizeof(*iref), BTRFS_LEAF_DATA_SIZE(leaf->fs_info));
1515                 return -EUCLEAN;
1516         }
1517
1518         ptr = btrfs_item_ptr_offset(leaf, slot);
1519         end = ptr + btrfs_item_size_nr(leaf, slot);
1520         while (ptr < end) {
1521                 u16 namelen;
1522
1523                 if (unlikely(ptr + sizeof(iref) > end)) {
1524                         inode_ref_err(leaf, slot,
1525                         "inode ref overflow, ptr %lu end %lu inode_ref_size %zu",
1526                                 ptr, end, sizeof(iref));
1527                         return -EUCLEAN;
1528                 }
1529
1530                 iref = (struct btrfs_inode_ref *)ptr;
1531                 namelen = btrfs_inode_ref_name_len(leaf, iref);
1532                 if (unlikely(ptr + sizeof(*iref) + namelen > end)) {
1533                         inode_ref_err(leaf, slot,
1534                                 "inode ref overflow, ptr %lu end %lu namelen %u",
1535                                 ptr, end, namelen);
1536                         return -EUCLEAN;
1537                 }
1538
1539                 /*
1540                  * NOTE: In theory we should record all found index numbers
1541                  * to find any duplicated indexes, but that will be too time
1542                  * consuming for inodes with too many hard links.
1543                  */
1544                 ptr += sizeof(*iref) + namelen;
1545         }
1546         return 0;
1547 }
1548
1549 /*
1550  * Common point to switch the item-specific validation.
1551  */
1552 static int check_leaf_item(struct extent_buffer *leaf,
1553                            struct btrfs_key *key, int slot,
1554                            struct btrfs_key *prev_key)
1555 {
1556         int ret = 0;
1557         struct btrfs_chunk *chunk;
1558
1559         switch (key->type) {
1560         case BTRFS_EXTENT_DATA_KEY:
1561                 ret = check_extent_data_item(leaf, key, slot, prev_key);
1562                 break;
1563         case BTRFS_EXTENT_CSUM_KEY:
1564                 ret = check_csum_item(leaf, key, slot, prev_key);
1565                 break;
1566         case BTRFS_DIR_ITEM_KEY:
1567         case BTRFS_DIR_INDEX_KEY:
1568         case BTRFS_XATTR_ITEM_KEY:
1569                 ret = check_dir_item(leaf, key, prev_key, slot);
1570                 break;
1571         case BTRFS_INODE_REF_KEY:
1572                 ret = check_inode_ref(leaf, key, prev_key, slot);
1573                 break;
1574         case BTRFS_BLOCK_GROUP_ITEM_KEY:
1575                 ret = check_block_group_item(leaf, key, slot);
1576                 break;
1577         case BTRFS_CHUNK_ITEM_KEY:
1578                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1579                 ret = check_leaf_chunk_item(leaf, chunk, key, slot);
1580                 break;
1581         case BTRFS_DEV_ITEM_KEY:
1582                 ret = check_dev_item(leaf, key, slot);
1583                 break;
1584         case BTRFS_INODE_ITEM_KEY:
1585                 ret = check_inode_item(leaf, key, slot);
1586                 break;
1587         case BTRFS_ROOT_ITEM_KEY:
1588                 ret = check_root_item(leaf, key, slot);
1589                 break;
1590         case BTRFS_EXTENT_ITEM_KEY:
1591         case BTRFS_METADATA_ITEM_KEY:
1592                 ret = check_extent_item(leaf, key, slot);
1593                 break;
1594         case BTRFS_TREE_BLOCK_REF_KEY:
1595         case BTRFS_SHARED_DATA_REF_KEY:
1596         case BTRFS_SHARED_BLOCK_REF_KEY:
1597                 ret = check_simple_keyed_refs(leaf, key, slot);
1598                 break;
1599         case BTRFS_EXTENT_DATA_REF_KEY:
1600                 ret = check_extent_data_ref(leaf, key, slot);
1601                 break;
1602         }
1603         return ret;
1604 }
1605
1606 static int check_leaf(struct extent_buffer *leaf, bool check_item_data)
1607 {
1608         struct btrfs_fs_info *fs_info = leaf->fs_info;
1609         /* No valid key type is 0, so all key should be larger than this key */
1610         struct btrfs_key prev_key = {0, 0, 0};
1611         struct btrfs_key key;
1612         u32 nritems = btrfs_header_nritems(leaf);
1613         int slot;
1614
1615         if (unlikely(btrfs_header_level(leaf) != 0)) {
1616                 generic_err(leaf, 0,
1617                         "invalid level for leaf, have %d expect 0",
1618                         btrfs_header_level(leaf));
1619                 return -EUCLEAN;
1620         }
1621
1622         /*
1623          * Extent buffers from a relocation tree have a owner field that
1624          * corresponds to the subvolume tree they are based on. So just from an
1625          * extent buffer alone we can not find out what is the id of the
1626          * corresponding subvolume tree, so we can not figure out if the extent
1627          * buffer corresponds to the root of the relocation tree or not. So
1628          * skip this check for relocation trees.
1629          */
1630         if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
1631                 u64 owner = btrfs_header_owner(leaf);
1632
1633                 /* These trees must never be empty */
1634                 if (unlikely(owner == BTRFS_ROOT_TREE_OBJECTID ||
1635                              owner == BTRFS_CHUNK_TREE_OBJECTID ||
1636                              owner == BTRFS_EXTENT_TREE_OBJECTID ||
1637                              owner == BTRFS_DEV_TREE_OBJECTID ||
1638                              owner == BTRFS_FS_TREE_OBJECTID ||
1639                              owner == BTRFS_DATA_RELOC_TREE_OBJECTID)) {
1640                         generic_err(leaf, 0,
1641                         "invalid root, root %llu must never be empty",
1642                                     owner);
1643                         return -EUCLEAN;
1644                 }
1645                 /* Unknown tree */
1646                 if (unlikely(owner == 0)) {
1647                         generic_err(leaf, 0,
1648                                 "invalid owner, root 0 is not defined");
1649                         return -EUCLEAN;
1650                 }
1651                 return 0;
1652         }
1653
1654         if (unlikely(nritems == 0))
1655                 return 0;
1656
1657         /*
1658          * Check the following things to make sure this is a good leaf, and
1659          * leaf users won't need to bother with similar sanity checks:
1660          *
1661          * 1) key ordering
1662          * 2) item offset and size
1663          *    No overlap, no hole, all inside the leaf.
1664          * 3) item content
1665          *    If possible, do comprehensive sanity check.
1666          *    NOTE: All checks must only rely on the item data itself.
1667          */
1668         for (slot = 0; slot < nritems; slot++) {
1669                 u32 item_end_expected;
1670                 int ret;
1671
1672                 btrfs_item_key_to_cpu(leaf, &key, slot);
1673
1674                 /* Make sure the keys are in the right order */
1675                 if (unlikely(btrfs_comp_cpu_keys(&prev_key, &key) >= 0)) {
1676                         generic_err(leaf, slot,
1677         "bad key order, prev (%llu %u %llu) current (%llu %u %llu)",
1678                                 prev_key.objectid, prev_key.type,
1679                                 prev_key.offset, key.objectid, key.type,
1680                                 key.offset);
1681                         return -EUCLEAN;
1682                 }
1683
1684                 /*
1685                  * Make sure the offset and ends are right, remember that the
1686                  * item data starts at the end of the leaf and grows towards the
1687                  * front.
1688                  */
1689                 if (slot == 0)
1690                         item_end_expected = BTRFS_LEAF_DATA_SIZE(fs_info);
1691                 else
1692                         item_end_expected = btrfs_item_offset_nr(leaf,
1693                                                                  slot - 1);
1694                 if (unlikely(btrfs_item_end_nr(leaf, slot) != item_end_expected)) {
1695                         generic_err(leaf, slot,
1696                                 "unexpected item end, have %u expect %u",
1697                                 btrfs_item_end_nr(leaf, slot),
1698                                 item_end_expected);
1699                         return -EUCLEAN;
1700                 }
1701
1702                 /*
1703                  * Check to make sure that we don't point outside of the leaf,
1704                  * just in case all the items are consistent to each other, but
1705                  * all point outside of the leaf.
1706                  */
1707                 if (unlikely(btrfs_item_end_nr(leaf, slot) >
1708                              BTRFS_LEAF_DATA_SIZE(fs_info))) {
1709                         generic_err(leaf, slot,
1710                         "slot end outside of leaf, have %u expect range [0, %u]",
1711                                 btrfs_item_end_nr(leaf, slot),
1712                                 BTRFS_LEAF_DATA_SIZE(fs_info));
1713                         return -EUCLEAN;
1714                 }
1715
1716                 /* Also check if the item pointer overlaps with btrfs item. */
1717                 if (unlikely(btrfs_item_ptr_offset(leaf, slot) <
1718                              btrfs_item_nr_offset(slot) + sizeof(struct btrfs_item))) {
1719                         generic_err(leaf, slot,
1720                 "slot overlaps with its data, item end %lu data start %lu",
1721                                 btrfs_item_nr_offset(slot) +
1722                                 sizeof(struct btrfs_item),
1723                                 btrfs_item_ptr_offset(leaf, slot));
1724                         return -EUCLEAN;
1725                 }
1726
1727                 if (check_item_data) {
1728                         /*
1729                          * Check if the item size and content meet other
1730                          * criteria
1731                          */
1732                         ret = check_leaf_item(leaf, &key, slot, &prev_key);
1733                         if (unlikely(ret < 0))
1734                                 return ret;
1735                 }
1736
1737                 prev_key.objectid = key.objectid;
1738                 prev_key.type = key.type;
1739                 prev_key.offset = key.offset;
1740         }
1741
1742         return 0;
1743 }
1744
1745 int btrfs_check_leaf_full(struct extent_buffer *leaf)
1746 {
1747         return check_leaf(leaf, true);
1748 }
1749 ALLOW_ERROR_INJECTION(btrfs_check_leaf_full, ERRNO);
1750
1751 int btrfs_check_leaf_relaxed(struct extent_buffer *leaf)
1752 {
1753         return check_leaf(leaf, false);
1754 }
1755
1756 int btrfs_check_node(struct extent_buffer *node)
1757 {
1758         struct btrfs_fs_info *fs_info = node->fs_info;
1759         unsigned long nr = btrfs_header_nritems(node);
1760         struct btrfs_key key, next_key;
1761         int slot;
1762         int level = btrfs_header_level(node);
1763         u64 bytenr;
1764         int ret = 0;
1765
1766         if (unlikely(level <= 0 || level >= BTRFS_MAX_LEVEL)) {
1767                 generic_err(node, 0,
1768                         "invalid level for node, have %d expect [1, %d]",
1769                         level, BTRFS_MAX_LEVEL - 1);
1770                 return -EUCLEAN;
1771         }
1772         if (unlikely(nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(fs_info))) {
1773                 btrfs_crit(fs_info,
1774 "corrupt node: root=%llu block=%llu, nritems too %s, have %lu expect range [1,%u]",
1775                            btrfs_header_owner(node), node->start,
1776                            nr == 0 ? "small" : "large", nr,
1777                            BTRFS_NODEPTRS_PER_BLOCK(fs_info));
1778                 return -EUCLEAN;
1779         }
1780
1781         for (slot = 0; slot < nr - 1; slot++) {
1782                 bytenr = btrfs_node_blockptr(node, slot);
1783                 btrfs_node_key_to_cpu(node, &key, slot);
1784                 btrfs_node_key_to_cpu(node, &next_key, slot + 1);
1785
1786                 if (unlikely(!bytenr)) {
1787                         generic_err(node, slot,
1788                                 "invalid NULL node pointer");
1789                         ret = -EUCLEAN;
1790                         goto out;
1791                 }
1792                 if (unlikely(!IS_ALIGNED(bytenr, fs_info->sectorsize))) {
1793                         generic_err(node, slot,
1794                         "unaligned pointer, have %llu should be aligned to %u",
1795                                 bytenr, fs_info->sectorsize);
1796                         ret = -EUCLEAN;
1797                         goto out;
1798                 }
1799
1800                 if (unlikely(btrfs_comp_cpu_keys(&key, &next_key) >= 0)) {
1801                         generic_err(node, slot,
1802         "bad key order, current (%llu %u %llu) next (%llu %u %llu)",
1803                                 key.objectid, key.type, key.offset,
1804                                 next_key.objectid, next_key.type,
1805                                 next_key.offset);
1806                         ret = -EUCLEAN;
1807                         goto out;
1808                 }
1809         }
1810 out:
1811         return ret;
1812 }
1813 ALLOW_ERROR_INJECTION(btrfs_check_node, ERRNO);