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