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/device-mapper/verity.txt>`_ 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, the base fs-verity feature only provides integrity 42protection, i.e. detection 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 49Trusted userspace code (e.g. operating system code running on a 50read-only partition that is itself authenticated by dm-verity) can 51authenticate the contents of an fs-verity file by using the 52`FS_IOC_MEASURE_VERITY`_ ioctl to retrieve its hash, then verifying a 53digital signature of it. 54 55A standard file hash could be used instead of fs-verity. However, 56this is inefficient if the file is large and only a small portion may 57be accessed. This is often the case for Android application package 58(APK) files, for example. These typically contain many translations, 59classes, and other resources that are infrequently or even never 60accessed on a particular device. It would be slow and wasteful to 61read and hash the entire file before starting the application. 62 63Unlike an ahead-of-time hash, fs-verity also re-verifies data each 64time it's paged in. This ensures that malicious disk firmware can't 65undetectably change the contents of the file at runtime. 66 67fs-verity does not replace or obsolete dm-verity. dm-verity should 68still be used on read-only filesystems. fs-verity is for files that 69must live on a read-write filesystem because they are independently 70updated and potentially user-installed, so dm-verity cannot be used. 71 72The base fs-verity feature is a hashing mechanism only; actually 73authenticating the files may be done by: 74 75* Userspace-only 76 77* Builtin signature verification + userspace policy 78 79 fs-verity optionally supports a simple signature verification 80 mechanism where users can configure the kernel to require that 81 all fs-verity files be signed by a key loaded into a keyring; 82 see `Built-in signature verification`_. 83 84* Integrity Measurement Architecture (IMA) 85 86 IMA supports including fs-verity file digests and signatures in the 87 IMA measurement list and verifying fs-verity based file signatures 88 stored as security.ima xattrs, based on policy. 89 90 91User API 92======== 93 94FS_IOC_ENABLE_VERITY 95-------------------- 96 97The FS_IOC_ENABLE_VERITY ioctl enables fs-verity on a file. It takes 98in a pointer to a struct fsverity_enable_arg, defined as 99follows:: 100 101 struct fsverity_enable_arg { 102 __u32 version; 103 __u32 hash_algorithm; 104 __u32 block_size; 105 __u32 salt_size; 106 __u64 salt_ptr; 107 __u32 sig_size; 108 __u32 __reserved1; 109 __u64 sig_ptr; 110 __u64 __reserved2[11]; 111 }; 112 113This structure contains the parameters of the Merkle tree to build for 114the file, and optionally contains a signature. It must be initialized 115as follows: 116 117- ``version`` must be 1. 118- ``hash_algorithm`` must be the identifier for the hash algorithm to 119 use for the Merkle tree, such as FS_VERITY_HASH_ALG_SHA256. See 120 ``include/uapi/linux/fsverity.h`` for the list of possible values. 121- ``block_size`` must be the Merkle tree block size. Currently, this 122 must be equal to the system page size, which is usually 4096 bytes. 123 Other sizes may be supported in the future. This value is not 124 necessarily the same as the filesystem block size. 125- ``salt_size`` is the size of the salt in bytes, or 0 if no salt is 126 provided. The salt is a value that is prepended to every hashed 127 block; it can be used to personalize the hashing for a particular 128 file or device. Currently the maximum salt size is 32 bytes. 129- ``salt_ptr`` is the pointer to the salt, or NULL if no salt is 130 provided. 131- ``sig_size`` is the size of the signature in bytes, or 0 if no 132 signature is provided. Currently the signature is (somewhat 133 arbitrarily) limited to 16128 bytes. See `Built-in signature 134 verification`_ for more information. 135- ``sig_ptr`` is the pointer to the signature, or NULL if no 136 signature is provided. 137- All reserved fields must be zeroed. 138 139FS_IOC_ENABLE_VERITY causes the filesystem to build a Merkle tree for 140the file and persist it to a filesystem-specific location associated 141with the file, then mark the file as a verity file. This ioctl may 142take a long time to execute on large files, and it is interruptible by 143fatal signals. 144 145FS_IOC_ENABLE_VERITY checks for write access to the inode. However, 146it must be executed on an O_RDONLY file descriptor and no processes 147can have the file open for writing. Attempts to open the file for 148writing while this ioctl is executing will fail with ETXTBSY. (This 149is necessary to guarantee that no writable file descriptors will exist 150after verity is enabled, and to guarantee that the file's contents are 151stable while the Merkle tree is being built over it.) 152 153On success, FS_IOC_ENABLE_VERITY returns 0, and the file becomes a 154verity file. On failure (including the case of interruption by a 155fatal signal), no changes are made to the file. 156 157FS_IOC_ENABLE_VERITY can fail with the following errors: 158 159- ``EACCES``: the process does not have write access to the file 160- ``EBADMSG``: the signature is malformed 161- ``EBUSY``: this ioctl is already running on the file 162- ``EEXIST``: the file already has verity enabled 163- ``EFAULT``: the caller provided inaccessible memory 164- ``EINTR``: the operation was interrupted by a fatal signal 165- ``EINVAL``: unsupported version, hash algorithm, or block size; or 166 reserved bits are set; or the file descriptor refers to neither a 167 regular file nor a directory. 168- ``EISDIR``: the file descriptor refers to a directory 169- ``EKEYREJECTED``: the signature doesn't match the file 170- ``EMSGSIZE``: the salt or signature is too long 171- ``ENOKEY``: the fs-verity keyring doesn't contain the certificate 172 needed to verify the signature 173- ``ENOPKG``: fs-verity recognizes the hash algorithm, but it's not 174 available in the kernel's crypto API as currently configured (e.g. 175 for SHA-512, missing CONFIG_CRYPTO_SHA512). 176- ``ENOTTY``: this type of filesystem does not implement fs-verity 177- ``EOPNOTSUPP``: the kernel was not configured with fs-verity 178 support; or the filesystem superblock has not had the 'verity' 179 feature enabled on it; or the filesystem does not support fs-verity 180 on this file. (See `Filesystem support`_.) 181- ``EPERM``: the file is append-only; or, a signature is required and 182 one was not provided. 183- ``EROFS``: the filesystem is read-only 184- ``ETXTBSY``: someone has the file open for writing. This can be the 185 caller's file descriptor, another open file descriptor, or the file 186 reference held by a writable memory map. 187 188FS_IOC_MEASURE_VERITY 189--------------------- 190 191The FS_IOC_MEASURE_VERITY ioctl retrieves the digest of a verity file. 192The fs-verity file digest is a cryptographic digest that identifies 193the file contents that are being enforced on reads; it is computed via 194a Merkle tree and is different from a traditional full-file digest. 195 196This ioctl takes in a pointer to a variable-length structure:: 197 198 struct fsverity_digest { 199 __u16 digest_algorithm; 200 __u16 digest_size; /* input/output */ 201 __u8 digest[]; 202 }; 203 204``digest_size`` is an input/output field. On input, it must be 205initialized to the number of bytes allocated for the variable-length 206``digest`` field. 207 208On success, 0 is returned and the kernel fills in the structure as 209follows: 210 211- ``digest_algorithm`` will be the hash algorithm used for the file 212 digest. It will match ``fsverity_enable_arg::hash_algorithm``. 213- ``digest_size`` will be the size of the digest in bytes, e.g. 32 214 for SHA-256. (This can be redundant with ``digest_algorithm``.) 215- ``digest`` will be the actual bytes of the digest. 216 217FS_IOC_MEASURE_VERITY is guaranteed to execute in constant time, 218regardless of the size of the file. 219 220FS_IOC_MEASURE_VERITY can fail with the following errors: 221 222- ``EFAULT``: the caller provided inaccessible memory 223- ``ENODATA``: the file is not a verity file 224- ``ENOTTY``: this type of filesystem does not implement fs-verity 225- ``EOPNOTSUPP``: the kernel was not configured with fs-verity 226 support, or the filesystem superblock has not had the 'verity' 227 feature enabled on it. (See `Filesystem support`_.) 228- ``EOVERFLOW``: the digest is longer than the specified 229 ``digest_size`` bytes. Try providing a larger buffer. 230 231FS_IOC_READ_VERITY_METADATA 232--------------------------- 233 234The FS_IOC_READ_VERITY_METADATA ioctl reads verity metadata from a 235verity file. This ioctl is available since Linux v5.12. 236 237This ioctl allows writing a server program that takes a verity file 238and serves it to a client program, such that the client can do its own 239fs-verity compatible verification of the file. This only makes sense 240if the client doesn't trust the server and if the server needs to 241provide the storage for the client. 242 243This is a fairly specialized use case, and most fs-verity users won't 244need this ioctl. 245 246This ioctl takes in a pointer to the following structure:: 247 248 #define FS_VERITY_METADATA_TYPE_MERKLE_TREE 1 249 #define FS_VERITY_METADATA_TYPE_DESCRIPTOR 2 250 #define FS_VERITY_METADATA_TYPE_SIGNATURE 3 251 252 struct fsverity_read_metadata_arg { 253 __u64 metadata_type; 254 __u64 offset; 255 __u64 length; 256 __u64 buf_ptr; 257 __u64 __reserved; 258 }; 259 260``metadata_type`` specifies the type of metadata to read: 261 262- ``FS_VERITY_METADATA_TYPE_MERKLE_TREE`` reads the blocks of the 263 Merkle tree. The blocks are returned in order from the root level 264 to the leaf level. Within each level, the blocks are returned in 265 the same order that their hashes are themselves hashed. 266 See `Merkle tree`_ for more information. 267 268- ``FS_VERITY_METADATA_TYPE_DESCRIPTOR`` reads the fs-verity 269 descriptor. See `fs-verity descriptor`_. 270 271- ``FS_VERITY_METADATA_TYPE_SIGNATURE`` reads the signature which was 272 passed to FS_IOC_ENABLE_VERITY, if any. See `Built-in signature 273 verification`_. 274 275The semantics are similar to those of ``pread()``. ``offset`` 276specifies the offset in bytes into the metadata item to read from, and 277``length`` specifies the maximum number of bytes to read from the 278metadata item. ``buf_ptr`` is the pointer to the buffer to read into, 279cast to a 64-bit integer. ``__reserved`` must be 0. On success, the 280number of bytes read is returned. 0 is returned at the end of the 281metadata item. The returned length may be less than ``length``, for 282example if the ioctl is interrupted. 283 284The metadata returned by FS_IOC_READ_VERITY_METADATA isn't guaranteed 285to be authenticated against the file digest that would be returned by 286`FS_IOC_MEASURE_VERITY`_, as the metadata is expected to be used to 287implement fs-verity compatible verification anyway (though absent a 288malicious disk, the metadata will indeed match). E.g. to implement 289this ioctl, the filesystem is allowed to just read the Merkle tree 290blocks from disk without actually verifying the path to the root node. 291 292FS_IOC_READ_VERITY_METADATA can fail with the following errors: 293 294- ``EFAULT``: the caller provided inaccessible memory 295- ``EINTR``: the ioctl was interrupted before any data was read 296- ``EINVAL``: reserved fields were set, or ``offset + length`` 297 overflowed 298- ``ENODATA``: the file is not a verity file, or 299 FS_VERITY_METADATA_TYPE_SIGNATURE was requested but the file doesn't 300 have a built-in signature 301- ``ENOTTY``: this type of filesystem does not implement fs-verity, or 302 this ioctl is not yet implemented on it 303- ``EOPNOTSUPP``: the kernel was not configured with fs-verity 304 support, or the filesystem superblock has not had the 'verity' 305 feature enabled on it. (See `Filesystem support`_.) 306 307FS_IOC_GETFLAGS 308--------------- 309 310The existing ioctl FS_IOC_GETFLAGS (which isn't specific to fs-verity) 311can also be used to check whether a file has fs-verity enabled or not. 312To do so, check for FS_VERITY_FL (0x00100000) in the returned flags. 313 314The verity flag is not settable via FS_IOC_SETFLAGS. You must use 315FS_IOC_ENABLE_VERITY instead, since parameters must be provided. 316 317statx 318----- 319 320Since Linux v5.5, the statx() system call sets STATX_ATTR_VERITY if 321the file has fs-verity enabled. This can perform better than 322FS_IOC_GETFLAGS and FS_IOC_MEASURE_VERITY because it doesn't require 323opening the file, and opening verity files can be expensive. 324 325Accessing verity files 326====================== 327 328Applications can transparently access a verity file just like a 329non-verity one, with the following exceptions: 330 331- Verity files are readonly. They cannot be opened for writing or 332 truncate()d, even if the file mode bits allow it. Attempts to do 333 one of these things will fail with EPERM. However, changes to 334 metadata such as owner, mode, timestamps, and xattrs are still 335 allowed, since these are not measured by fs-verity. Verity files 336 can also still be renamed, deleted, and linked to. 337 338- Direct I/O is not supported on verity files. Attempts to use direct 339 I/O on such files will fall back to buffered I/O. 340 341- DAX (Direct Access) is not supported on verity files, because this 342 would circumvent the data verification. 343 344- Reads of data that doesn't match the verity Merkle tree will fail 345 with EIO (for read()) or SIGBUS (for mmap() reads). 346 347- If the sysctl "fs.verity.require_signatures" is set to 1 and the 348 file is not signed by a key in the fs-verity keyring, then opening 349 the file will fail. See `Built-in signature verification`_. 350 351Direct access to the Merkle tree is not supported. Therefore, if a 352verity file is copied, or is backed up and restored, then it will lose 353its "verity"-ness. fs-verity is primarily meant for files like 354executables that are managed by a package manager. 355 356File digest computation 357======================= 358 359This section describes how fs-verity hashes the file contents using a 360Merkle tree to produce the digest which cryptographically identifies 361the file contents. This algorithm is the same for all filesystems 362that support fs-verity. 363 364Userspace only needs to be aware of this algorithm if it needs to 365compute fs-verity file digests itself, e.g. in order to sign files. 366 367.. _fsverity_merkle_tree: 368 369Merkle tree 370----------- 371 372The file contents is divided into blocks, where the block size is 373configurable but is usually 4096 bytes. The end of the last block is 374zero-padded if needed. Each block is then hashed, producing the first 375level of hashes. Then, the hashes in this first level are grouped 376into 'blocksize'-byte blocks (zero-padding the ends as needed) and 377these blocks are hashed, producing the second level of hashes. This 378proceeds up the tree until only a single block remains. The hash of 379this block is the "Merkle tree root hash". 380 381If the file fits in one block and is nonempty, then the "Merkle tree 382root hash" is simply the hash of the single data block. If the file 383is empty, then the "Merkle tree root hash" is all zeroes. 384 385The "blocks" here are not necessarily the same as "filesystem blocks". 386 387If a salt was specified, then it's zero-padded to the closest multiple 388of the input size of the hash algorithm's compression function, e.g. 38964 bytes for SHA-256 or 128 bytes for SHA-512. The padded salt is 390prepended to every data or Merkle tree block that is hashed. 391 392The purpose of the block padding is to cause every hash to be taken 393over the same amount of data, which simplifies the implementation and 394keeps open more possibilities for hardware acceleration. The purpose 395of the salt padding is to make the salting "free" when the salted hash 396state is precomputed, then imported for each hash. 397 398Example: in the recommended configuration of SHA-256 and 4K blocks, 399128 hash values fit in each block. Thus, each level of the Merkle 400tree is approximately 128 times smaller than the previous, and for 401large files the Merkle tree's size converges to approximately 1/127 of 402the original file size. However, for small files, the padding is 403significant, making the space overhead proportionally more. 404 405.. _fsverity_descriptor: 406 407fs-verity descriptor 408-------------------- 409 410By itself, the Merkle tree root hash is ambiguous. For example, it 411can't a distinguish a large file from a small second file whose data 412is exactly the top-level hash block of the first file. Ambiguities 413also arise from the convention of padding to the next block boundary. 414 415To solve this problem, the fs-verity file digest is actually computed 416as a hash of the following structure, which contains the Merkle tree 417root hash as well as other fields such as the file size:: 418 419 struct fsverity_descriptor { 420 __u8 version; /* must be 1 */ 421 __u8 hash_algorithm; /* Merkle tree hash algorithm */ 422 __u8 log_blocksize; /* log2 of size of data and tree blocks */ 423 __u8 salt_size; /* size of salt in bytes; 0 if none */ 424 __le32 __reserved_0x04; /* must be 0 */ 425 __le64 data_size; /* size of file the Merkle tree is built over */ 426 __u8 root_hash[64]; /* Merkle tree root hash */ 427 __u8 salt[32]; /* salt prepended to each hashed block */ 428 __u8 __reserved[144]; /* must be 0's */ 429 }; 430 431Built-in signature verification 432=============================== 433 434With CONFIG_FS_VERITY_BUILTIN_SIGNATURES=y, fs-verity supports putting 435a portion of an authentication policy (see `Use cases`_) in the 436kernel. Specifically, it adds support for: 437 4381. At fs-verity module initialization time, a keyring ".fs-verity" is 439 created. The root user can add trusted X.509 certificates to this 440 keyring using the add_key() system call, then (when done) 441 optionally use keyctl_restrict_keyring() to prevent additional 442 certificates from being added. 443 4442. `FS_IOC_ENABLE_VERITY`_ accepts a pointer to a PKCS#7 formatted 445 detached signature in DER format of the file's fs-verity digest. 446 On success, this signature is persisted alongside the Merkle tree. 447 Then, any time the file is opened, the kernel will verify the 448 file's actual digest against this signature, using the certificates 449 in the ".fs-verity" keyring. 450 4513. A new sysctl "fs.verity.require_signatures" is made available. 452 When set to 1, the kernel requires that all verity files have a 453 correctly signed digest as described in (2). 454 455fs-verity file digests must be signed in the following format, which 456is similar to the structure used by `FS_IOC_MEASURE_VERITY`_:: 457 458 struct fsverity_formatted_digest { 459 char magic[8]; /* must be "FSVerity" */ 460 __le16 digest_algorithm; 461 __le16 digest_size; 462 __u8 digest[]; 463 }; 464 465fs-verity's built-in signature verification support is meant as a 466relatively simple mechanism that can be used to provide some level of 467authenticity protection for verity files, as an alternative to doing 468the signature verification in userspace or using IMA-appraisal. 469However, with this mechanism, userspace programs still need to check 470that the verity bit is set, and there is no protection against verity 471files being swapped around. 472 473Filesystem support 474================== 475 476fs-verity is supported by several filesystems, described below. The 477CONFIG_FS_VERITY kconfig option must be enabled to use fs-verity on 478any of these filesystems. 479 480``include/linux/fsverity.h`` declares the interface between the 481``fs/verity/`` support layer and filesystems. Briefly, filesystems 482must provide an ``fsverity_operations`` structure that provides 483methods to read and write the verity metadata to a filesystem-specific 484location, including the Merkle tree blocks and 485``fsverity_descriptor``. Filesystems must also call functions in 486``fs/verity/`` at certain times, such as when a file is opened or when 487pages have been read into the pagecache. (See `Verifying data`_.) 488 489ext4 490---- 491 492ext4 supports fs-verity since Linux v5.4 and e2fsprogs v1.45.2. 493 494To create verity files on an ext4 filesystem, the filesystem must have 495been formatted with ``-O verity`` or had ``tune2fs -O verity`` run on 496it. "verity" is an RO_COMPAT filesystem feature, so once set, old 497kernels will only be able to mount the filesystem readonly, and old 498versions of e2fsck will be unable to check the filesystem. Moreover, 499currently ext4 only supports mounting a filesystem with the "verity" 500feature when its block size is equal to PAGE_SIZE (often 4096 bytes). 501 502ext4 sets the EXT4_VERITY_FL on-disk inode flag on verity files. It 503can only be set by `FS_IOC_ENABLE_VERITY`_, and it cannot be cleared. 504 505ext4 also supports encryption, which can be used simultaneously with 506fs-verity. In this case, the plaintext data is verified rather than 507the ciphertext. This is necessary in order to make the fs-verity file 508digest meaningful, since every file is encrypted differently. 509 510ext4 stores the verity metadata (Merkle tree and fsverity_descriptor) 511past the end of the file, starting at the first 64K boundary beyond 512i_size. This approach works because (a) verity files are readonly, 513and (b) pages fully beyond i_size aren't visible to userspace but can 514be read/written internally by ext4 with only some relatively small 515changes to ext4. This approach avoids having to depend on the 516EA_INODE feature and on rearchitecturing ext4's xattr support to 517support paging multi-gigabyte xattrs into memory, and to support 518encrypting xattrs. Note that the verity metadata *must* be encrypted 519when the file is, since it contains hashes of the plaintext data. 520 521Currently, ext4 verity only supports the case where the Merkle tree 522block size, filesystem block size, and page size are all the same. It 523also only supports extent-based files. 524 525f2fs 526---- 527 528f2fs supports fs-verity since Linux v5.4 and f2fs-tools v1.11.0. 529 530To create verity files on an f2fs filesystem, the filesystem must have 531been formatted with ``-O verity``. 532 533f2fs sets the FADVISE_VERITY_BIT on-disk inode flag on verity files. 534It can only be set by `FS_IOC_ENABLE_VERITY`_, and it cannot be 535cleared. 536 537Like ext4, f2fs stores the verity metadata (Merkle tree and 538fsverity_descriptor) past the end of the file, starting at the first 53964K boundary beyond i_size. See explanation for ext4 above. 540Moreover, f2fs supports at most 4096 bytes of xattr entries per inode 541which wouldn't be enough for even a single Merkle tree block. 542 543Currently, f2fs verity only supports a Merkle tree block size of 4096. 544Also, f2fs doesn't support enabling verity on files that currently 545have atomic or volatile writes pending. 546 547btrfs 548----- 549 550btrfs supports fs-verity since Linux v5.15. Verity-enabled inodes are 551marked with a RO_COMPAT inode flag, and the verity metadata is stored 552in separate btree items. 553 554Implementation details 555====================== 556 557Verifying data 558-------------- 559 560fs-verity ensures that all reads of a verity file's data are verified, 561regardless of which syscall is used to do the read (e.g. mmap(), 562read(), pread()) and regardless of whether it's the first read or a 563later read (unless the later read can return cached data that was 564already verified). Below, we describe how filesystems implement this. 565 566Pagecache 567~~~~~~~~~ 568 569For filesystems using Linux's pagecache, the ``->read_folio()`` and 570``->readahead()`` methods must be modified to verify pages before they 571are marked Uptodate. Merely hooking ``->read_iter()`` would be 572insufficient, since ``->read_iter()`` is not used for memory maps. 573 574Therefore, fs/verity/ provides a function fsverity_verify_page() which 575verifies a page that has been read into the pagecache of a verity 576inode, but is still locked and not Uptodate, so it's not yet readable 577by userspace. As needed to do the verification, 578fsverity_verify_page() will call back into the filesystem to read 579Merkle tree pages via fsverity_operations::read_merkle_tree_page(). 580 581fsverity_verify_page() returns false if verification failed; in this 582case, the filesystem must not set the page Uptodate. Following this, 583as per the usual Linux pagecache behavior, attempts by userspace to 584read() from the part of the file containing the page will fail with 585EIO, and accesses to the page within a memory map will raise SIGBUS. 586 587fsverity_verify_page() currently only supports the case where the 588Merkle tree block size is equal to PAGE_SIZE (often 4096 bytes). 589 590In principle, fsverity_verify_page() verifies the entire path in the 591Merkle tree from the data page to the root hash. However, for 592efficiency the filesystem may cache the hash pages. Therefore, 593fsverity_verify_page() only ascends the tree reading hash pages until 594an already-verified hash page is seen, as indicated by the PageChecked 595bit being set. It then verifies the path to that page. 596 597This optimization, which is also used by dm-verity, results in 598excellent sequential read performance. This is because usually (e.g. 599127 in 128 times for 4K blocks and SHA-256) the hash page from the 600bottom level of the tree will already be cached and checked from 601reading a previous data page. However, random reads perform worse. 602 603Block device based filesystems 604~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 605 606Block device based filesystems (e.g. ext4 and f2fs) in Linux also use 607the pagecache, so the above subsection applies too. However, they 608also usually read many pages from a file at once, grouped into a 609structure called a "bio". To make it easier for these types of 610filesystems to support fs-verity, fs/verity/ also provides a function 611fsverity_verify_bio() which verifies all pages in a bio. 612 613ext4 and f2fs also support encryption. If a verity file is also 614encrypted, the pages must be decrypted before being verified. To 615support this, these filesystems allocate a "post-read context" for 616each bio and store it in ``->bi_private``:: 617 618 struct bio_post_read_ctx { 619 struct bio *bio; 620 struct work_struct work; 621 unsigned int cur_step; 622 unsigned int enabled_steps; 623 }; 624 625``enabled_steps`` is a bitmask that specifies whether decryption, 626verity, or both is enabled. After the bio completes, for each needed 627postprocessing step the filesystem enqueues the bio_post_read_ctx on a 628workqueue, and then the workqueue work does the decryption or 629verification. Finally, pages where no decryption or verity error 630occurred are marked Uptodate, and the pages are unlocked. 631 632On many filesystems, files can contain holes. Normally, 633``->readahead()`` simply zeroes holes and sets the corresponding pages 634Uptodate; no bios are issued. To prevent this case from bypassing 635fs-verity, these filesystems use fsverity_verify_page() to verify hole 636pages. 637 638Filesystems also disable direct I/O on verity files, since otherwise 639direct I/O would bypass fs-verity. 640 641Userspace utility 642================= 643 644This document focuses on the kernel, but a userspace utility for 645fs-verity can be found at: 646 647 https://git.kernel.org/pub/scm/linux/kernel/git/ebiggers/fsverity-utils.git 648 649See the README.md file in the fsverity-utils source tree for details, 650including examples of setting up fs-verity protected files. 651 652Tests 653===== 654 655To test fs-verity, use xfstests. For example, using `kvm-xfstests 656<https://github.com/tytso/xfstests-bld/blob/master/Documentation/kvm-quickstart.md>`_:: 657 658 kvm-xfstests -c ext4,f2fs,btrfs -g verity 659 660FAQ 661=== 662 663This section answers frequently asked questions about fs-verity that 664weren't already directly answered in other parts of this document. 665 666:Q: Why isn't fs-verity part of IMA? 667:A: fs-verity and IMA (Integrity Measurement Architecture) have 668 different focuses. fs-verity is a filesystem-level mechanism for 669 hashing individual files using a Merkle tree. In contrast, IMA 670 specifies a system-wide policy that specifies which files are 671 hashed and what to do with those hashes, such as log them, 672 authenticate them, or add them to a measurement list. 673 674 IMA supports the fs-verity hashing mechanism as an alternative 675 to full file hashes, for those who want the performance and 676 security benefits of the Merkle tree based hash. However, it 677 doesn't make sense to force all uses of fs-verity to be through 678 IMA. fs-verity already meets many users' needs even as a 679 standalone filesystem feature, and it's testable like other 680 filesystem features e.g. with xfstests. 681 682:Q: Isn't fs-verity useless because the attacker can just modify the 683 hashes in the Merkle tree, which is stored on-disk? 684:A: To verify the authenticity of an fs-verity file you must verify 685 the authenticity of the "fs-verity file digest", which 686 incorporates the root hash of the Merkle tree. See `Use cases`_. 687 688:Q: Isn't fs-verity useless because the attacker can just replace a 689 verity file with a non-verity one? 690:A: See `Use cases`_. In the initial use case, it's really trusted 691 userspace code that authenticates the files; fs-verity is just a 692 tool to do this job efficiently and securely. The trusted 693 userspace code will consider non-verity files to be inauthentic. 694 695:Q: Why does the Merkle tree need to be stored on-disk? Couldn't you 696 store just the root hash? 697:A: If the Merkle tree wasn't stored on-disk, then you'd have to 698 compute the entire tree when the file is first accessed, even if 699 just one byte is being read. This is a fundamental consequence of 700 how Merkle tree hashing works. To verify a leaf node, you need to 701 verify the whole path to the root hash, including the root node 702 (the thing which the root hash is a hash of). But if the root 703 node isn't stored on-disk, you have to compute it by hashing its 704 children, and so on until you've actually hashed the entire file. 705 706 That defeats most of the point of doing a Merkle tree-based hash, 707 since if you have to hash the whole file ahead of time anyway, 708 then you could simply do sha256(file) instead. That would be much 709 simpler, and a bit faster too. 710 711 It's true that an in-memory Merkle tree could still provide the 712 advantage of verification on every read rather than just on the 713 first read. However, it would be inefficient because every time a 714 hash page gets evicted (you can't pin the entire Merkle tree into 715 memory, since it may be very large), in order to restore it you 716 again need to hash everything below it in the tree. This again 717 defeats most of the point of doing a Merkle tree-based hash, since 718 a single block read could trigger re-hashing gigabytes of data. 719 720:Q: But couldn't you store just the leaf nodes and compute the rest? 721:A: See previous answer; this really just moves up one level, since 722 one could alternatively interpret the data blocks as being the 723 leaf nodes of the Merkle tree. It's true that the tree can be 724 computed much faster if the leaf level is stored rather than just 725 the data, but that's only because each level is less than 1% the 726 size of the level below (assuming the recommended settings of 727 SHA-256 and 4K blocks). For the exact same reason, by storing 728 "just the leaf nodes" you'd already be storing over 99% of the 729 tree, so you might as well simply store the whole tree. 730 731:Q: Can the Merkle tree be built ahead of time, e.g. distributed as 732 part of a package that is installed to many computers? 733:A: This isn't currently supported. It was part of the original 734 design, but was removed to simplify the kernel UAPI and because it 735 wasn't a critical use case. Files are usually installed once and 736 used many times, and cryptographic hashing is somewhat fast on 737 most modern processors. 738 739:Q: Why doesn't fs-verity support writes? 740:A: Write support would be very difficult and would require a 741 completely different design, so it's well outside the scope of 742 fs-verity. Write support would require: 743 744 - A way to maintain consistency between the data and hashes, 745 including all levels of hashes, since corruption after a crash 746 (especially of potentially the entire file!) is unacceptable. 747 The main options for solving this are data journalling, 748 copy-on-write, and log-structured volume. But it's very hard to 749 retrofit existing filesystems with new consistency mechanisms. 750 Data journalling is available on ext4, but is very slow. 751 752 - Rebuilding the Merkle tree after every write, which would be 753 extremely inefficient. Alternatively, a different authenticated 754 dictionary structure such as an "authenticated skiplist" could 755 be used. However, this would be far more complex. 756 757 Compare it to dm-verity vs. dm-integrity. dm-verity is very 758 simple: the kernel just verifies read-only data against a 759 read-only Merkle tree. In contrast, dm-integrity supports writes 760 but is slow, is much more complex, and doesn't actually support 761 full-device authentication since it authenticates each sector 762 independently, i.e. there is no "root hash". It doesn't really 763 make sense for the same device-mapper target to support these two 764 very different cases; the same applies to fs-verity. 765 766:Q: Since verity files are immutable, why isn't the immutable bit set? 767:A: The existing "immutable" bit (FS_IMMUTABLE_FL) already has a 768 specific set of semantics which not only make the file contents 769 read-only, but also prevent the file from being deleted, renamed, 770 linked to, or having its owner or mode changed. These extra 771 properties are unwanted for fs-verity, so reusing the immutable 772 bit isn't appropriate. 773 774:Q: Why does the API use ioctls instead of setxattr() and getxattr()? 775:A: Abusing the xattr interface for basically arbitrary syscalls is 776 heavily frowned upon by most of the Linux filesystem developers. 777 An xattr should really just be an xattr on-disk, not an API to 778 e.g. magically trigger construction of a Merkle tree. 779 780:Q: Does fs-verity support remote filesystems? 781:A: So far all filesystems that have implemented fs-verity support are 782 local filesystems, but in principle any filesystem that can store 783 per-file verity metadata can support fs-verity, regardless of 784 whether it's local or remote. Some filesystems may have fewer 785 options of where to store the verity metadata; one possibility is 786 to store it past the end of the file and "hide" it from userspace 787 by manipulating i_size. The data verification functions provided 788 by ``fs/verity/`` also assume that the filesystem uses the Linux 789 pagecache, but both local and remote filesystems normally do so. 790 791:Q: Why is anything filesystem-specific at all? Shouldn't fs-verity 792 be implemented entirely at the VFS level? 793:A: There are many reasons why this is not possible or would be very 794 difficult, including the following: 795 796 - To prevent bypassing verification, pages must not be marked 797 Uptodate until they've been verified. Currently, each 798 filesystem is responsible for marking pages Uptodate via 799 ``->readahead()``. Therefore, currently it's not possible for 800 the VFS to do the verification on its own. Changing this would 801 require significant changes to the VFS and all filesystems. 802 803 - It would require defining a filesystem-independent way to store 804 the verity metadata. Extended attributes don't work for this 805 because (a) the Merkle tree may be gigabytes, but many 806 filesystems assume that all xattrs fit into a single 4K 807 filesystem block, and (b) ext4 and f2fs encryption doesn't 808 encrypt xattrs, yet the Merkle tree *must* be encrypted when the 809 file contents are, because it stores hashes of the plaintext 810 file contents. 811 812 So the verity metadata would have to be stored in an actual 813 file. Using a separate file would be very ugly, since the 814 metadata is fundamentally part of the file to be protected, and 815 it could cause problems where users could delete the real file 816 but not the metadata file or vice versa. On the other hand, 817 having it be in the same file would break applications unless 818 filesystems' notion of i_size were divorced from the VFS's, 819 which would be complex and require changes to all filesystems. 820 821 - It's desirable that FS_IOC_ENABLE_VERITY uses the filesystem's 822 transaction mechanism so that either the file ends up with 823 verity enabled, or no changes were made. Allowing intermediate 824 states to occur after a crash may cause problems. 825