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