1.. SPDX-License-Identifier: GPL-2.0 2 3.. _fsverity: 4 5======================================================= 6fs-verity: read-only file-based authenticity protection 7======================================================= 8 9Introduction 10============ 11 12fs-verity (``fs/verity/``) is a support layer that filesystems can 13hook into to support transparent integrity and authenticity protection 14of read-only files. Currently, it is supported by the ext4, f2fs, and 15btrfs filesystems. Like fscrypt, not too much filesystem-specific 16code is needed to support fs-verity. 17 18fs-verity is similar to `dm-verity 19<https://www.kernel.org/doc/Documentation/admin-guide/device-mapper/verity.rst>`_ 20but works on files rather than block devices. On regular files on 21filesystems supporting fs-verity, userspace can execute an ioctl that 22causes the filesystem to build a Merkle tree for the file and persist 23it to a filesystem-specific location associated with the file. 24 25After this, the file is made readonly, and all reads from the file are 26automatically verified against the file's Merkle tree. Reads of any 27corrupted data, including mmap reads, will fail. 28 29Userspace can use another ioctl to retrieve the root hash (actually 30the "fs-verity file digest", which is a hash that includes the Merkle 31tree root hash) that fs-verity is enforcing for the file. This ioctl 32executes in constant time, regardless of the file size. 33 34fs-verity is essentially a way to hash a file in constant time, 35subject to the caveat that reads which would violate the hash will 36fail at runtime. 37 38Use cases 39========= 40 41By itself, fs-verity only provides integrity protection, i.e. 42detection of accidental (non-malicious) corruption. 43 44However, because fs-verity makes retrieving the file hash extremely 45efficient, it's primarily meant to be used as a tool to support 46authentication (detection of malicious modifications) or auditing 47(logging file hashes before use). 48 49A standard file hash could be used instead of fs-verity. However, 50this is inefficient if the file is large and only a small portion may 51be accessed. This is often the case for Android application package 52(APK) files, for example. These typically contain many translations, 53classes, and other resources that are infrequently or even never 54accessed on a particular device. It would be slow and wasteful to 55read and hash the entire file before starting the application. 56 57Unlike an ahead-of-time hash, fs-verity also re-verifies data each 58time it's paged in. This ensures that malicious disk firmware can't 59undetectably change the contents of the file at runtime. 60 61fs-verity does not replace or obsolete dm-verity. dm-verity should 62still be used on read-only filesystems. fs-verity is for files that 63must live on a read-write filesystem because they are independently 64updated and potentially user-installed, so dm-verity cannot be used. 65 66fs-verity does not mandate a particular scheme for authenticating its 67file hashes. (Similarly, dm-verity does not mandate a particular 68scheme for authenticating its block device root hashes.) Options for 69authenticating fs-verity file hashes include: 70 71- Trusted userspace code. Often, the userspace code that accesses 72 files can be trusted to authenticate them. Consider e.g. an 73 application that wants to authenticate data files before using them, 74 or an application loader that is part of the operating system (which 75 is already authenticated in a different way, such as by being loaded 76 from a read-only partition that uses dm-verity) and that wants to 77 authenticate applications before loading them. In these cases, this 78 trusted userspace code can authenticate a file's contents by 79 retrieving its fs-verity digest using `FS_IOC_MEASURE_VERITY`_, then 80 verifying a signature of it using any userspace cryptographic 81 library that supports digital signatures. 82 83- Integrity Measurement Architecture (IMA). IMA supports fs-verity 84 file digests as an alternative to its traditional full file digests. 85 "IMA appraisal" enforces that files contain a valid, matching 86 signature in their "security.ima" extended attribute, as controlled 87 by the IMA policy. For more information, see the IMA documentation. 88 89- Integrity Policy Enforcement (IPE). IPE supports enforcing access 90 control decisions based on immutable security properties of files, 91 including those protected by fs-verity's built-in signatures. 92 "IPE policy" specifically allows for the authorization of fs-verity 93 files using properties ``fsverity_digest`` for identifying 94 files by their verity digest, and ``fsverity_signature`` to authorize 95 files with a verified fs-verity's built-in signature. For 96 details on configuring IPE policies and understanding its operational 97 modes, please refer to :doc:`IPE admin guide </admin-guide/LSM/ipe>`. 98 99- Trusted userspace code in combination with `Built-in signature 100 verification`_. This approach should be used only with great care. 101 102User API 103======== 104 105FS_IOC_ENABLE_VERITY 106-------------------- 107 108The FS_IOC_ENABLE_VERITY ioctl enables fs-verity on a file. It takes 109in a pointer to a struct fsverity_enable_arg, defined as 110follows:: 111 112 struct fsverity_enable_arg { 113 __u32 version; 114 __u32 hash_algorithm; 115 __u32 block_size; 116 __u32 salt_size; 117 __u64 salt_ptr; 118 __u32 sig_size; 119 __u32 __reserved1; 120 __u64 sig_ptr; 121 __u64 __reserved2[11]; 122 }; 123 124This structure contains the parameters of the Merkle tree to build for 125the file. It must be initialized as follows: 126 127- ``version`` must be 1. 128- ``hash_algorithm`` must be the identifier for the hash algorithm to 129 use for the Merkle tree, such as FS_VERITY_HASH_ALG_SHA256. See 130 ``include/uapi/linux/fsverity.h`` for the list of possible values. 131- ``block_size`` is the Merkle tree block size, in bytes. In Linux 132 v6.3 and later, this can be any power of 2 between (inclusively) 133 1024 and the minimum of the system page size and the filesystem 134 block size. In earlier versions, the page size was the only allowed 135 value. 136- ``salt_size`` is the size of the salt in bytes, or 0 if no salt is 137 provided. The salt is a value that is prepended to every hashed 138 block; it can be used to personalize the hashing for a particular 139 file or device. Currently the maximum salt size is 32 bytes. 140- ``salt_ptr`` is the pointer to the salt, or NULL if no salt is 141 provided. 142- ``sig_size`` is the size of the builtin signature in bytes, or 0 if no 143 builtin signature is provided. Currently the builtin signature is 144 (somewhat arbitrarily) limited to 16128 bytes. 145- ``sig_ptr`` is the pointer to the builtin signature, or NULL if no 146 builtin signature is provided. A builtin signature is only needed 147 if the `Built-in signature verification`_ feature is being used. It 148 is not needed for IMA appraisal, and it is not needed if the file 149 signature is being handled entirely in userspace. 150- All reserved fields must be zeroed. 151 152FS_IOC_ENABLE_VERITY causes the filesystem to build a Merkle tree for 153the file and persist it to a filesystem-specific location associated 154with the file, then mark the file as a verity file. This ioctl may 155take a long time to execute on large files, and it is interruptible by 156fatal signals. 157 158FS_IOC_ENABLE_VERITY checks for write access to the inode. However, 159it must be executed on an O_RDONLY file descriptor and no processes 160can have the file open for writing. Attempts to open the file for 161writing while this ioctl is executing will fail with ETXTBSY. (This 162is necessary to guarantee that no writable file descriptors will exist 163after verity is enabled, and to guarantee that the file's contents are 164stable while the Merkle tree is being built over it.) 165 166On success, FS_IOC_ENABLE_VERITY returns 0, and the file becomes a 167verity file. On failure (including the case of interruption by a 168fatal signal), no changes are made to the file. 169 170FS_IOC_ENABLE_VERITY can fail with the following errors: 171 172- ``EACCES``: the process does not have write access to the file 173- ``EBADMSG``: the builtin signature is malformed 174- ``EBUSY``: this ioctl is already running on the file 175- ``EEXIST``: the file already has verity enabled 176- ``EFAULT``: the caller provided inaccessible memory 177- ``EFBIG``: the file is too large to enable verity on 178- ``EINTR``: the operation was interrupted by a fatal signal 179- ``EINVAL``: unsupported version, hash algorithm, or block size; or 180 reserved bits are set; or the file descriptor refers to neither a 181 regular file nor a directory. 182- ``EISDIR``: the file descriptor refers to a directory 183- ``EKEYREJECTED``: the builtin signature doesn't match the file 184- ``EMSGSIZE``: the salt or builtin signature is too long 185- ``ENOKEY``: the ".fs-verity" keyring doesn't contain the certificate 186 needed to verify the builtin signature 187- ``ENOPKG``: fs-verity recognizes the hash algorithm, but it's not 188 available in the kernel's crypto API as currently configured (e.g. 189 for SHA-512, missing CONFIG_CRYPTO_SHA512). 190- ``ENOTTY``: this type of filesystem does not implement fs-verity 191- ``EOPNOTSUPP``: the kernel was not configured with fs-verity 192 support; or the filesystem superblock has not had the 'verity' 193 feature enabled on it; or the filesystem does not support fs-verity 194 on this file. (See `Filesystem support`_.) 195- ``EPERM``: the file is append-only; or, a builtin signature is 196 required and one was not provided. 197- ``EROFS``: the filesystem is read-only 198- ``ETXTBSY``: someone has the file open for writing. This can be the 199 caller's file descriptor, another open file descriptor, or the file 200 reference held by a writable memory map. 201 202FS_IOC_MEASURE_VERITY 203--------------------- 204 205The FS_IOC_MEASURE_VERITY ioctl retrieves the digest of a verity file. 206The fs-verity file digest is a cryptographic digest that identifies 207the file contents that are being enforced on reads; it is computed via 208a Merkle tree and is different from a traditional full-file digest. 209 210This ioctl takes in a pointer to a variable-length structure:: 211 212 struct fsverity_digest { 213 __u16 digest_algorithm; 214 __u16 digest_size; /* input/output */ 215 __u8 digest[]; 216 }; 217 218``digest_size`` is an input/output field. On input, it must be 219initialized to the number of bytes allocated for the variable-length 220``digest`` field. 221 222On success, 0 is returned and the kernel fills in the structure as 223follows: 224 225- ``digest_algorithm`` will be the hash algorithm used for the file 226 digest. It will match ``fsverity_enable_arg::hash_algorithm``. 227- ``digest_size`` will be the size of the digest in bytes, e.g. 32 228 for SHA-256. (This can be redundant with ``digest_algorithm``.) 229- ``digest`` will be the actual bytes of the digest. 230 231FS_IOC_MEASURE_VERITY is guaranteed to execute in constant time, 232regardless of the size of the file. 233 234FS_IOC_MEASURE_VERITY can fail with the following errors: 235 236- ``EFAULT``: the caller provided inaccessible memory 237- ``ENODATA``: the file is not a verity file 238- ``ENOTTY``: this type of filesystem does not implement fs-verity 239- ``EOPNOTSUPP``: the kernel was not configured with fs-verity 240 support, or the filesystem superblock has not had the 'verity' 241 feature enabled on it. (See `Filesystem support`_.) 242- ``EOVERFLOW``: the digest is longer than the specified 243 ``digest_size`` bytes. Try providing a larger buffer. 244 245FS_IOC_READ_VERITY_METADATA 246--------------------------- 247 248The FS_IOC_READ_VERITY_METADATA ioctl reads verity metadata from a 249verity file. This ioctl is available since Linux v5.12. 250 251This ioctl is useful for cases where the verity verification should be 252performed somewhere other than the currently running kernel. 253 254One example is a server program that takes a verity file and serves it 255to a client program, such that the client can do its own fs-verity 256compatible verification of the file. This only makes sense if the 257client doesn't trust the server and if the server needs to provide the 258storage for the client. 259 260Another example is copying verity metadata when creating filesystem 261images in userspace (such as with ``mkfs.ext4 -d``). 262 263This is a fairly specialized use case, and most fs-verity users won't 264need this ioctl. 265 266This ioctl takes in a pointer to the following structure:: 267 268 #define FS_VERITY_METADATA_TYPE_MERKLE_TREE 1 269 #define FS_VERITY_METADATA_TYPE_DESCRIPTOR 2 270 #define FS_VERITY_METADATA_TYPE_SIGNATURE 3 271 272 struct fsverity_read_metadata_arg { 273 __u64 metadata_type; 274 __u64 offset; 275 __u64 length; 276 __u64 buf_ptr; 277 __u64 __reserved; 278 }; 279 280``metadata_type`` specifies the type of metadata to read: 281 282- ``FS_VERITY_METADATA_TYPE_MERKLE_TREE`` reads the blocks of the 283 Merkle tree. The blocks are returned in order from the root level 284 to the leaf level. Within each level, the blocks are returned in 285 the same order that their hashes are themselves hashed. 286 See `Merkle tree`_ for more information. 287 288- ``FS_VERITY_METADATA_TYPE_DESCRIPTOR`` reads the fs-verity 289 descriptor. See `fs-verity descriptor`_. 290 291- ``FS_VERITY_METADATA_TYPE_SIGNATURE`` reads the builtin signature 292 which was passed to FS_IOC_ENABLE_VERITY, if any. See `Built-in 293 signature verification`_. 294 295The semantics are similar to those of ``pread()``. ``offset`` 296specifies the offset in bytes into the metadata item to read from, and 297``length`` specifies the maximum number of bytes to read from the 298metadata item. ``buf_ptr`` is the pointer to the buffer to read into, 299cast to a 64-bit integer. ``__reserved`` must be 0. On success, the 300number of bytes read is returned. 0 is returned at the end of the 301metadata item. The returned length may be less than ``length``, for 302example if the ioctl is interrupted. 303 304The metadata returned by FS_IOC_READ_VERITY_METADATA isn't guaranteed 305to be authenticated against the file digest that would be returned by 306`FS_IOC_MEASURE_VERITY`_, as the metadata is expected to be used to 307implement fs-verity compatible verification anyway (though absent a 308malicious disk, the metadata will indeed match). E.g. to implement 309this ioctl, the filesystem is allowed to just read the Merkle tree 310blocks from disk without actually verifying the path to the root node. 311 312FS_IOC_READ_VERITY_METADATA can fail with the following errors: 313 314- ``EFAULT``: the caller provided inaccessible memory 315- ``EINTR``: the ioctl was interrupted before any data was read 316- ``EINVAL``: reserved fields were set, or ``offset + length`` 317 overflowed 318- ``ENODATA``: the file is not a verity file, or 319 FS_VERITY_METADATA_TYPE_SIGNATURE was requested but the file doesn't 320 have a builtin signature 321- ``ENOTTY``: this type of filesystem does not implement fs-verity, or 322 this ioctl is not yet implemented on it 323- ``EOPNOTSUPP``: the kernel was not configured with fs-verity 324 support, or the filesystem superblock has not had the 'verity' 325 feature enabled on it. (See `Filesystem support`_.) 326 327FS_IOC_GETFLAGS 328--------------- 329 330The existing ioctl FS_IOC_GETFLAGS (which isn't specific to fs-verity) 331can also be used to check whether a file has fs-verity enabled or not. 332To do so, check for FS_VERITY_FL (0x00100000) in the returned flags. 333 334The verity flag is not settable via FS_IOC_SETFLAGS. You must use 335FS_IOC_ENABLE_VERITY instead, since parameters must be provided. 336 337statx 338----- 339 340Since Linux v5.5, the statx() system call sets STATX_ATTR_VERITY if 341the file has fs-verity enabled. This can perform better than 342FS_IOC_GETFLAGS and FS_IOC_MEASURE_VERITY because it doesn't require 343opening the file, and opening verity files can be expensive. 344 345.. _accessing_verity_files: 346 347Accessing verity files 348====================== 349 350Applications can transparently access a verity file just like a 351non-verity one, with the following exceptions: 352 353- Verity files are readonly. They cannot be opened for writing or 354 truncate()d, even if the file mode bits allow it. Attempts to do 355 one of these things will fail with EPERM. However, changes to 356 metadata such as owner, mode, timestamps, and xattrs are still 357 allowed, since these are not measured by fs-verity. Verity files 358 can also still be renamed, deleted, and linked to. 359 360- Direct I/O is not supported on verity files. Attempts to use direct 361 I/O on such files will fall back to buffered I/O. 362 363- DAX (Direct Access) is not supported on verity files, because this 364 would circumvent the data verification. 365 366- Reads of data that doesn't match the verity Merkle tree will fail 367 with EIO (for read()) or SIGBUS (for mmap() reads). 368 369- If the sysctl "fs.verity.require_signatures" is set to 1 and the 370 file is not signed by a key in the ".fs-verity" keyring, then 371 opening the file will fail. See `Built-in signature verification`_. 372 373Direct access to the Merkle tree is not supported. Therefore, if a 374verity file is copied, or is backed up and restored, then it will lose 375its "verity"-ness. fs-verity is primarily meant for files like 376executables that are managed by a package manager. 377 378File digest computation 379======================= 380 381This section describes how fs-verity hashes the file contents using a 382Merkle tree to produce the digest which cryptographically identifies 383the file contents. This algorithm is the same for all filesystems 384that support fs-verity. 385 386Userspace only needs to be aware of this algorithm if it needs to 387compute fs-verity file digests itself, e.g. in order to sign files. 388 389.. _fsverity_merkle_tree: 390 391Merkle tree 392----------- 393 394The file contents is divided into blocks, where the block size is 395configurable but is usually 4096 bytes. The end of the last block is 396zero-padded if needed. Each block is then hashed, producing the first 397level of hashes. Then, the hashes in this first level are grouped 398into 'blocksize'-byte blocks (zero-padding the ends as needed) and 399these blocks are hashed, producing the second level of hashes. This 400proceeds up the tree until only a single block remains. The hash of 401this block is the "Merkle tree root hash". 402 403If the file fits in one block and is nonempty, then the "Merkle tree 404root hash" is simply the hash of the single data block. If the file 405is empty, then the "Merkle tree root hash" is all zeroes. 406 407The "blocks" here are not necessarily the same as "filesystem blocks". 408 409If a salt was specified, then it's zero-padded to the closest multiple 410of the input size of the hash algorithm's compression function, e.g. 41164 bytes for SHA-256 or 128 bytes for SHA-512. The padded salt is 412prepended to every data or Merkle tree block that is hashed. 413 414The purpose of the block padding is to cause every hash to be taken 415over the same amount of data, which simplifies the implementation and 416keeps open more possibilities for hardware acceleration. The purpose 417of the salt padding is to make the salting "free" when the salted hash 418state is precomputed, then imported for each hash. 419 420Example: in the recommended configuration of SHA-256 and 4K blocks, 421128 hash values fit in each block. Thus, each level of the Merkle 422tree is approximately 128 times smaller than the previous, and for 423large files the Merkle tree's size converges to approximately 1/127 of 424the original file size. However, for small files, the padding is 425significant, making the space overhead proportionally more. 426 427.. _fsverity_descriptor: 428 429fs-verity descriptor 430-------------------- 431 432By itself, the Merkle tree root hash is ambiguous. For example, it 433can't a distinguish a large file from a small second file whose data 434is exactly the top-level hash block of the first file. Ambiguities 435also arise from the convention of padding to the next block boundary. 436 437To solve this problem, the fs-verity file digest is actually computed 438as a hash of the following structure, which contains the Merkle tree 439root hash as well as other fields such as the file size:: 440 441 struct fsverity_descriptor { 442 __u8 version; /* must be 1 */ 443 __u8 hash_algorithm; /* Merkle tree hash algorithm */ 444 __u8 log_blocksize; /* log2 of size of data and tree blocks */ 445 __u8 salt_size; /* size of salt in bytes; 0 if none */ 446 __le32 __reserved_0x04; /* must be 0 */ 447 __le64 data_size; /* size of file the Merkle tree is built over */ 448 __u8 root_hash[64]; /* Merkle tree root hash */ 449 __u8 salt[32]; /* salt prepended to each hashed block */ 450 __u8 __reserved[144]; /* must be 0's */ 451 }; 452 453Built-in signature verification 454=============================== 455 456CONFIG_FS_VERITY_BUILTIN_SIGNATURES=y adds supports for in-kernel 457verification of fs-verity builtin signatures. 458 459**IMPORTANT**! Please take great care before using this feature. 460It is not the only way to do signatures with fs-verity, and the 461alternatives (such as userspace signature verification, and IMA 462appraisal) can be much better. It's also easy to fall into a trap 463of thinking this feature solves more problems than it actually does. 464 465Enabling this option adds the following: 466 4671. At boot time, the kernel creates a keyring named ".fs-verity". The 468 root user can add trusted X.509 certificates to this keyring using 469 the add_key() system call. 470 4712. `FS_IOC_ENABLE_VERITY`_ accepts a pointer to a PKCS#7 formatted 472 detached signature in DER format of the file's fs-verity digest. 473 On success, the ioctl persists the signature alongside the Merkle 474 tree. Then, any time the file is opened, the kernel verifies the 475 file's actual digest against this signature, using the certificates 476 in the ".fs-verity" keyring. This verification happens as long as the 477 file's signature exists, regardless of the state of the sysctl variable 478 "fs.verity.require_signatures" described in the next item. The IPE LSM 479 relies on this behavior to recognize and label fsverity files 480 that contain a verified built-in fsverity signature. 481 4823. A new sysctl "fs.verity.require_signatures" is made available. 483 When set to 1, the kernel requires that all verity files have a 484 correctly signed digest as described in (2). 485 486The data that the signature as described in (2) must be a signature of 487is the fs-verity file digest in the following format:: 488 489 struct fsverity_formatted_digest { 490 char magic[8]; /* must be "FSVerity" */ 491 __le16 digest_algorithm; 492 __le16 digest_size; 493 __u8 digest[]; 494 }; 495 496That's it. It should be emphasized again that fs-verity builtin 497signatures are not the only way to do signatures with fs-verity. See 498`Use cases`_ for an overview of ways in which fs-verity can be used. 499fs-verity builtin signatures have some major limitations that should 500be carefully considered before using them: 501 502- Builtin signature verification does *not* make the kernel enforce 503 that any files actually have fs-verity enabled. Thus, it is not a 504 complete authentication policy. Currently, if it is used, one 505 way to complete the authentication policy is for trusted userspace 506 code to explicitly check whether files have fs-verity enabled with a 507 signature before they are accessed. (With 508 fs.verity.require_signatures=1, just checking whether fs-verity is 509 enabled suffices.) But, in this case the trusted userspace code 510 could just store the signature alongside the file and verify it 511 itself using a cryptographic library, instead of using this feature. 512 513- Another approach is to utilize fs-verity builtin signature 514 verification in conjunction with the IPE LSM, which supports defining 515 a kernel-enforced, system-wide authentication policy that allows only 516 files with a verified fs-verity builtin signature to perform certain 517 operations, such as execution. Note that IPE doesn't require 518 fs.verity.require_signatures=1. 519 Please refer to :doc:`IPE admin guide </admin-guide/LSM/ipe>` for 520 more details. 521 522- A file's builtin signature can only be set at the same time that 523 fs-verity is being enabled on the file. Changing or deleting the 524 builtin signature later requires re-creating the file. 525 526- Builtin signature verification uses the same set of public keys for 527 all fs-verity enabled files on the system. Different keys cannot be 528 trusted for different files; each key is all or nothing. 529 530- The sysctl fs.verity.require_signatures applies system-wide. 531 Setting it to 1 only works when all users of fs-verity on the system 532 agree that it should be set to 1. This limitation can prevent 533 fs-verity from being used in cases where it would be helpful. 534 535- Builtin signature verification can only use signature algorithms 536 that are supported by the kernel. For example, the kernel does not 537 yet support Ed25519, even though this is often the signature 538 algorithm that is recommended for new cryptographic designs. 539 540- fs-verity builtin signatures are in PKCS#7 format, and the public 541 keys are in X.509 format. These formats are commonly used, 542 including by some other kernel features (which is why the fs-verity 543 builtin signatures use them), and are very feature rich. 544 Unfortunately, history has shown that code that parses and handles 545 these formats (which are from the 1990s and are based on ASN.1) 546 often has vulnerabilities as a result of their complexity. This 547 complexity is not inherent to the cryptography itself. 548 549 fs-verity users who do not need advanced features of X.509 and 550 PKCS#7 should strongly consider using simpler formats, such as plain 551 Ed25519 keys and signatures, and verifying signatures in userspace. 552 553 fs-verity users who choose to use X.509 and PKCS#7 anyway should 554 still consider that verifying those signatures in userspace is more 555 flexible (for other reasons mentioned earlier in this document) and 556 eliminates the need to enable CONFIG_FS_VERITY_BUILTIN_SIGNATURES 557 and its associated increase in kernel attack surface. In some cases 558 it can even be necessary, since advanced X.509 and PKCS#7 features 559 do not always work as intended with the kernel. For example, the 560 kernel does not check X.509 certificate validity times. 561 562 Note: IMA appraisal, which supports fs-verity, does not use PKCS#7 563 for its signatures, so it partially avoids the issues discussed 564 here. IMA appraisal does use X.509. 565 566Filesystem support 567================== 568 569fs-verity is supported by several filesystems, described below. The 570CONFIG_FS_VERITY kconfig option must be enabled to use fs-verity on 571any of these filesystems. 572 573``include/linux/fsverity.h`` declares the interface between the 574``fs/verity/`` support layer and filesystems. Briefly, filesystems 575must provide an ``fsverity_operations`` structure that provides 576methods to read and write the verity metadata to a filesystem-specific 577location, including the Merkle tree blocks and 578``fsverity_descriptor``. Filesystems must also call functions in 579``fs/verity/`` at certain times, such as when a file is opened or when 580pages have been read into the pagecache. (See `Verifying data`_.) 581 582ext4 583---- 584 585ext4 supports fs-verity since Linux v5.4 and e2fsprogs v1.45.2. 586 587To create verity files on an ext4 filesystem, the filesystem must have 588been formatted with ``-O verity`` or had ``tune2fs -O verity`` run on 589it. "verity" is an RO_COMPAT filesystem feature, so once set, old 590kernels will only be able to mount the filesystem readonly, and old 591versions of e2fsck will be unable to check the filesystem. 592 593Originally, an ext4 filesystem with the "verity" feature could only be 594mounted when its block size was equal to the system page size 595(typically 4096 bytes). In Linux v6.3, this limitation was removed. 596 597ext4 sets the EXT4_VERITY_FL on-disk inode flag on verity files. It 598can only be set by `FS_IOC_ENABLE_VERITY`_, and it cannot be cleared. 599 600ext4 also supports encryption, which can be used simultaneously with 601fs-verity. In this case, the plaintext data is verified rather than 602the ciphertext. This is necessary in order to make the fs-verity file 603digest meaningful, since every file is encrypted differently. 604 605ext4 stores the verity metadata (Merkle tree and fsverity_descriptor) 606past the end of the file, starting at the first 64K boundary beyond 607i_size. This approach works because (a) verity files are readonly, 608and (b) pages fully beyond i_size aren't visible to userspace but can 609be read/written internally by ext4 with only some relatively small 610changes to ext4. This approach avoids having to depend on the 611EA_INODE feature and on rearchitecturing ext4's xattr support to 612support paging multi-gigabyte xattrs into memory, and to support 613encrypting xattrs. Note that the verity metadata *must* be encrypted 614when the file is, since it contains hashes of the plaintext data. 615 616ext4 only allows verity on extent-based files. 617 618f2fs 619---- 620 621f2fs supports fs-verity since Linux v5.4 and f2fs-tools v1.11.0. 622 623To create verity files on an f2fs filesystem, the filesystem must have 624been formatted with ``-O verity``. 625 626f2fs sets the FADVISE_VERITY_BIT on-disk inode flag on verity files. 627It can only be set by `FS_IOC_ENABLE_VERITY`_, and it cannot be 628cleared. 629 630Like ext4, f2fs stores the verity metadata (Merkle tree and 631fsverity_descriptor) past the end of the file, starting at the first 63264K boundary beyond i_size. See explanation for ext4 above. 633Moreover, f2fs supports at most 4096 bytes of xattr entries per inode 634which usually wouldn't be enough for even a single Merkle tree block. 635 636f2fs doesn't support enabling verity on files that currently have 637atomic or volatile writes pending. 638 639btrfs 640----- 641 642btrfs supports fs-verity since Linux v5.15. Verity-enabled inodes are 643marked with a RO_COMPAT inode flag, and the verity metadata is stored 644in separate btree items. 645 646Implementation details 647====================== 648 649Verifying data 650-------------- 651 652fs-verity ensures that all reads of a verity file's data are verified, 653regardless of which syscall is used to do the read (e.g. mmap(), 654read(), pread()) and regardless of whether it's the first read or a 655later read (unless the later read can return cached data that was 656already verified). Below, we describe how filesystems implement this. 657 658Pagecache 659~~~~~~~~~ 660 661For filesystems using Linux's pagecache, the ``->read_folio()`` and 662``->readahead()`` methods must be modified to verify folios before 663they are marked Uptodate. Merely hooking ``->read_iter()`` would be 664insufficient, since ``->read_iter()`` is not used for memory maps. 665 666Therefore, fs/verity/ provides the function fsverity_verify_blocks() 667which verifies data that has been read into the pagecache of a verity 668inode. The containing folio must still be locked and not Uptodate, so 669it's not yet readable by userspace. As needed to do the verification, 670fsverity_verify_blocks() will call back into the filesystem to read 671hash blocks via fsverity_operations::read_merkle_tree_page(). 672 673fsverity_verify_blocks() returns false if verification failed; in this 674case, the filesystem must not set the folio Uptodate. Following this, 675as per the usual Linux pagecache behavior, attempts by userspace to 676read() from the part of the file containing the folio will fail with 677EIO, and accesses to the folio within a memory map will raise SIGBUS. 678 679In principle, verifying a data block requires verifying the entire 680path in the Merkle tree from the data block to the root hash. 681However, for efficiency the filesystem may cache the hash blocks. 682Therefore, fsverity_verify_blocks() only ascends the tree reading hash 683blocks until an already-verified hash block is seen. It then verifies 684the path to that block. 685 686This optimization, which is also used by dm-verity, results in 687excellent sequential read performance. This is because usually (e.g. 688127 in 128 times for 4K blocks and SHA-256) the hash block from the 689bottom level of the tree will already be cached and checked from 690reading a previous data block. However, random reads perform worse. 691 692Block device based filesystems 693~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 694 695Block device based filesystems (e.g. ext4 and f2fs) in Linux also use 696the pagecache, so the above subsection applies too. However, they 697also usually read many data blocks from a file at once, grouped into a 698structure called a "bio". To make it easier for these types of 699filesystems to support fs-verity, fs/verity/ also provides a function 700fsverity_verify_bio() which verifies all data blocks in a bio. 701 702ext4 and f2fs also support encryption. If a verity file is also 703encrypted, the data must be decrypted before being verified. To 704support this, these filesystems allocate a "post-read context" for 705each bio and store it in ``->bi_private``:: 706 707 struct bio_post_read_ctx { 708 struct bio *bio; 709 struct work_struct work; 710 unsigned int cur_step; 711 unsigned int enabled_steps; 712 }; 713 714``enabled_steps`` is a bitmask that specifies whether decryption, 715verity, or both is enabled. After the bio completes, for each needed 716postprocessing step the filesystem enqueues the bio_post_read_ctx on a 717workqueue, and then the workqueue work does the decryption or 718verification. Finally, folios where no decryption or verity error 719occurred are marked Uptodate, and the folios are unlocked. 720 721On many filesystems, files can contain holes. Normally, 722``->readahead()`` simply zeroes hole blocks and considers the 723corresponding data to be up-to-date; no bios are issued. To prevent 724this case from bypassing fs-verity, filesystems use 725fsverity_verify_blocks() to verify hole blocks. 726 727Filesystems also disable direct I/O on verity files, since otherwise 728direct I/O would bypass fs-verity. 729 730Userspace utility 731================= 732 733This document focuses on the kernel, but a userspace utility for 734fs-verity can be found at: 735 736 https://git.kernel.org/pub/scm/fs/fsverity/fsverity-utils.git 737 738See the README.md file in the fsverity-utils source tree for details, 739including examples of setting up fs-verity protected files. 740 741Tests 742===== 743 744To test fs-verity, use xfstests. For example, using `kvm-xfstests 745<https://github.com/tytso/xfstests-bld/blob/master/Documentation/kvm-quickstart.md>`_:: 746 747 kvm-xfstests -c ext4,f2fs,btrfs -g verity 748 749FAQ 750=== 751 752This section answers frequently asked questions about fs-verity that 753weren't already directly answered in other parts of this document. 754 755:Q: Why isn't fs-verity part of IMA? 756:A: fs-verity and IMA (Integrity Measurement Architecture) have 757 different focuses. fs-verity is a filesystem-level mechanism for 758 hashing individual files using a Merkle tree. In contrast, IMA 759 specifies a system-wide policy that specifies which files are 760 hashed and what to do with those hashes, such as log them, 761 authenticate them, or add them to a measurement list. 762 763 IMA supports the fs-verity hashing mechanism as an alternative 764 to full file hashes, for those who want the performance and 765 security benefits of the Merkle tree based hash. However, it 766 doesn't make sense to force all uses of fs-verity to be through 767 IMA. fs-verity already meets many users' needs even as a 768 standalone filesystem feature, and it's testable like other 769 filesystem features e.g. with xfstests. 770 771:Q: Isn't fs-verity useless because the attacker can just modify the 772 hashes in the Merkle tree, which is stored on-disk? 773:A: To verify the authenticity of an fs-verity file you must verify 774 the authenticity of the "fs-verity file digest", which 775 incorporates the root hash of the Merkle tree. See `Use cases`_. 776 777:Q: Isn't fs-verity useless because the attacker can just replace a 778 verity file with a non-verity one? 779:A: See `Use cases`_. In the initial use case, it's really trusted 780 userspace code that authenticates the files; fs-verity is just a 781 tool to do this job efficiently and securely. The trusted 782 userspace code will consider non-verity files to be inauthentic. 783 784:Q: Why does the Merkle tree need to be stored on-disk? Couldn't you 785 store just the root hash? 786:A: If the Merkle tree wasn't stored on-disk, then you'd have to 787 compute the entire tree when the file is first accessed, even if 788 just one byte is being read. This is a fundamental consequence of 789 how Merkle tree hashing works. To verify a leaf node, you need to 790 verify the whole path to the root hash, including the root node 791 (the thing which the root hash is a hash of). But if the root 792 node isn't stored on-disk, you have to compute it by hashing its 793 children, and so on until you've actually hashed the entire file. 794 795 That defeats most of the point of doing a Merkle tree-based hash, 796 since if you have to hash the whole file ahead of time anyway, 797 then you could simply do sha256(file) instead. That would be much 798 simpler, and a bit faster too. 799 800 It's true that an in-memory Merkle tree could still provide the 801 advantage of verification on every read rather than just on the 802 first read. However, it would be inefficient because every time a 803 hash page gets evicted (you can't pin the entire Merkle tree into 804 memory, since it may be very large), in order to restore it you 805 again need to hash everything below it in the tree. This again 806 defeats most of the point of doing a Merkle tree-based hash, since 807 a single block read could trigger re-hashing gigabytes of data. 808 809:Q: But couldn't you store just the leaf nodes and compute the rest? 810:A: See previous answer; this really just moves up one level, since 811 one could alternatively interpret the data blocks as being the 812 leaf nodes of the Merkle tree. It's true that the tree can be 813 computed much faster if the leaf level is stored rather than just 814 the data, but that's only because each level is less than 1% the 815 size of the level below (assuming the recommended settings of 816 SHA-256 and 4K blocks). For the exact same reason, by storing 817 "just the leaf nodes" you'd already be storing over 99% of the 818 tree, so you might as well simply store the whole tree. 819 820:Q: Can the Merkle tree be built ahead of time, e.g. distributed as 821 part of a package that is installed to many computers? 822:A: This isn't currently supported. It was part of the original 823 design, but was removed to simplify the kernel UAPI and because it 824 wasn't a critical use case. Files are usually installed once and 825 used many times, and cryptographic hashing is somewhat fast on 826 most modern processors. 827 828:Q: Why doesn't fs-verity support writes? 829:A: Write support would be very difficult and would require a 830 completely different design, so it's well outside the scope of 831 fs-verity. Write support would require: 832 833 - A way to maintain consistency between the data and hashes, 834 including all levels of hashes, since corruption after a crash 835 (especially of potentially the entire file!) is unacceptable. 836 The main options for solving this are data journalling, 837 copy-on-write, and log-structured volume. But it's very hard to 838 retrofit existing filesystems with new consistency mechanisms. 839 Data journalling is available on ext4, but is very slow. 840 841 - Rebuilding the Merkle tree after every write, which would be 842 extremely inefficient. Alternatively, a different authenticated 843 dictionary structure such as an "authenticated skiplist" could 844 be used. However, this would be far more complex. 845 846 Compare it to dm-verity vs. dm-integrity. dm-verity is very 847 simple: the kernel just verifies read-only data against a 848 read-only Merkle tree. In contrast, dm-integrity supports writes 849 but is slow, is much more complex, and doesn't actually support 850 full-device authentication since it authenticates each sector 851 independently, i.e. there is no "root hash". It doesn't really 852 make sense for the same device-mapper target to support these two 853 very different cases; the same applies to fs-verity. 854 855:Q: Since verity files are immutable, why isn't the immutable bit set? 856:A: The existing "immutable" bit (FS_IMMUTABLE_FL) already has a 857 specific set of semantics which not only make the file contents 858 read-only, but also prevent the file from being deleted, renamed, 859 linked to, or having its owner or mode changed. These extra 860 properties are unwanted for fs-verity, so reusing the immutable 861 bit isn't appropriate. 862 863:Q: Why does the API use ioctls instead of setxattr() and getxattr()? 864:A: Abusing the xattr interface for basically arbitrary syscalls is 865 heavily frowned upon by most of the Linux filesystem developers. 866 An xattr should really just be an xattr on-disk, not an API to 867 e.g. magically trigger construction of a Merkle tree. 868 869:Q: Does fs-verity support remote filesystems? 870:A: So far all filesystems that have implemented fs-verity support are 871 local filesystems, but in principle any filesystem that can store 872 per-file verity metadata can support fs-verity, regardless of 873 whether it's local or remote. Some filesystems may have fewer 874 options of where to store the verity metadata; one possibility is 875 to store it past the end of the file and "hide" it from userspace 876 by manipulating i_size. The data verification functions provided 877 by ``fs/verity/`` also assume that the filesystem uses the Linux 878 pagecache, but both local and remote filesystems normally do so. 879 880:Q: Why is anything filesystem-specific at all? Shouldn't fs-verity 881 be implemented entirely at the VFS level? 882:A: There are many reasons why this is not possible or would be very 883 difficult, including the following: 884 885 - To prevent bypassing verification, folios must not be marked 886 Uptodate until they've been verified. Currently, each 887 filesystem is responsible for marking folios Uptodate via 888 ``->readahead()``. Therefore, currently it's not possible for 889 the VFS to do the verification on its own. Changing this would 890 require significant changes to the VFS and all filesystems. 891 892 - It would require defining a filesystem-independent way to store 893 the verity metadata. Extended attributes don't work for this 894 because (a) the Merkle tree may be gigabytes, but many 895 filesystems assume that all xattrs fit into a single 4K 896 filesystem block, and (b) ext4 and f2fs encryption doesn't 897 encrypt xattrs, yet the Merkle tree *must* be encrypted when the 898 file contents are, because it stores hashes of the plaintext 899 file contents. 900 901 So the verity metadata would have to be stored in an actual 902 file. Using a separate file would be very ugly, since the 903 metadata is fundamentally part of the file to be protected, and 904 it could cause problems where users could delete the real file 905 but not the metadata file or vice versa. On the other hand, 906 having it be in the same file would break applications unless 907 filesystems' notion of i_size were divorced from the VFS's, 908 which would be complex and require changes to all filesystems. 909 910 - It's desirable that FS_IOC_ENABLE_VERITY uses the filesystem's 911 transaction mechanism so that either the file ends up with 912 verity enabled, or no changes were made. Allowing intermediate 913 states to occur after a crash may cause problems. 914