PKCS#11 API allows us to use opaque keys confined in hardware security
modules (HSMs) and similar hardware tokens without direct access to the
key material, providing logical separation of the keys from the
cryptographic operations performed using them.
This commit allows using the popular libp11 pkcs11 module for the
OpenSSL library with `fsverity` so that direct access to a private key
file isn't necessary to sign files.
The user needs to supply the path to the engine shared library
(typically the libp11 shared object file) and the PKCS#11 module library
(a shared object file specific to the given hardware token). The user
may also supply a token-specific key identifier.
Test evidence with a hardware PKCS#11 token:
$ echo test > dummy
$ ./fsverity sign dummy dummy.sig \
--pkcs11-engine=/usr/lib64/engines-1.1/libpkcs11.so \
--pkcs11-module=/usr/local/lib64/pkcs11_module.so \
--cert=test-pkcs11-cert.pem && echo OK;
Signed file 'dummy'
(sha256:c497326752e21b3992b57f7eff159102d474a97d972dc2c2d99d23e0f5fbdb65)
OK
Test evidence for regression check (checking that regular file-based key
signing still works):
$ ./fsverity sign dummy dummy.sig --key=key.pem --cert=cert.pem && \
echo OK;
Signed file 'dummy'
(sha256:c497326752e21b3992b57f7eff159102d474a97d972dc2c2d99d23e0f5fbdb65)
OK
Signed-off-by: Aleksander Adamowski <olo@fb.com>
[EB: Avoided overloading the --key option and keyfile field, clarified
the documentation, removed logic from cmd_sign.c that libfsverity
already handles, and many other improvements.]
Link: https://lore.kernel.org/r/20210909212731.1151190-1-olo@fb.com
Signed-off-by: Eric Biggers <ebiggers@google.com>
Make 'fsverity digest' and 'fsverity sign' support writing the Merkle
tree and fs-verity descriptor to files, using new options
'--out-merkle-tree=FILE' and '--out-descriptor=FILE'.
Normally these new options aren't useful, but they can be needed in
cases where the fs-verity metadata needs to be consumed by something
other than one of the native Linux kernel implementations of fs-verity.
This is different from 'fsverity dump_metadata' in that
'fsverity dump_metadata' only works on a file with fs-verity enabled,
whereas these new options are for the userspace file digest computation.
Link: https://lore.kernel.org/r/20210603195812.50838-5-ebiggers@kernel.org
Reviewed-by: Victor Hsieh <victorhsieh@google.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Add a 'fsverity dump_metadata' subcommand which calls
FS_IOC_READ_VERITY_METADATA on a file and prints the returned metadata
to stdout. There are three subsubcommands, one for each type of
metadata that can be read using the ioctl:
fsverity dump_metadata merkle_tree FILE
fsverity dump_metadata descriptor FILE
fsverity dump_metadata signature FILE
By default the whole metadata item is dumped. --length and --offset can
be specified to dump only a particular range of the item.
This subcommand will be used by xfstests to test the
FS_IOC_READ_VERITY_METADATA ioctl.
Link: https://lore.kernel.org/r/20210115182402.35691-3-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Add convenience functions that wrap FS_IOC_ENABLE_VERITY but take a
'struct libfsverity_merkle_tree_params' instead of
'struct fsverity_enable_arg'. This is useful because it allows
libfsverity users to deal with one common struct, and also get the
default parameter handling that libfsverity_compute_digest() does.
Acked-by: Luca Boccassi <luca.boccassi@microsoft.com>
Link: https://lore.kernel.org/r/20201116205628.262173-4-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Even though the kernel currently only supports PAGE_SIZE == Merkle tree
block size, PAGE_SIZE isn't a good default Merkle tree block size for
fsverity-utils, since it means that if someone doesn't explicitly
specify the block size, then the results of 'fsverity sign' and
'fsverity enable' will differ between different architectures.
So change the default Merkle tree block size to 4096, which is the most
common PAGE_SIZE. This will break anyone using the fsverity program
without the --block-size option on an architecture with a non-4K page
size. But I don't think anyone is actually doing that yet anyway.
Acked-by: Luca Boccassi <luca.boccassi@microsoft.com>
Link: https://lore.kernel.org/r/20201116205628.262173-2-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Add a 'fsverity digest' subcommand that prints the hex-encoded digest of
the specified file(s), ready to be signed offline. By default the
output is like 'fsverity measure', but if the --for-builtin-sig option
is given, the output is the hex-encoded "struct fsverity_signed_digest"
that the kernel expects to be signed for builtin signatures.
This subcommand is useful in case the integrated signing mechanism with
local cert/key cannot be used.
Signed-off-by: Luca Boccassi <luca.boccassi@microsoft.com>
Link: https://lore.kernel.org/r/20201026191839.3329948-1-luca.boccassi@gmail.com
[EB: commit message, comment, and whitespace tweaks]
Signed-off-by: Eric Biggers <ebiggers@google.com>
libfsverity.h is the public API, but the other headers in common/ are
private headers for fsverity-utils. Move libfsverity.h to its own
directory to make this clear. This is also needed for Android's build
system in order to restrict the exported headers to libfsverity.h.
This doesn't affect users who are using 'make install', since
'make install' still installs libfsverity.h to the same place,
and it doesn't install any private headers.
Link: https://lore.kernel.org/r/20200923202328.16310-1-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Add three test programs: 'test_hash_algs', 'test_compute_digest', and
'test_sign_digest'. Nothing fancy yet, just some basic tests to test
each library function.
With the new Makefile, these get run by 'make check'.
Reviewed-by: Jes Sorensen <jsorensen@fb.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
From the 'fsverity' program, split out a library 'libfsverity'.
Currently it supports computing file measurements ("digests"), and
signing those file measurements for use with the fs-verity builtin
signature verification feature.
Rewritten from patches by Jes Sorensen <jsorensen@fb.com>.
I made a lot of improvements, e.g.:
- Separated library and program source into different directories.
- Drastically improved the Makefile.
- Added 'make check' target and rules to build test programs.
- In the shared lib, only export the functions intended to be public.
- Prefixed global functions with "libfsverity_" so that they don't cause
conflicts when the library is built as a static library.
- Made library error messages be sent to a user-specified callback
rather than always be printed to stderr.
- Keep showing OpenSSL error messages.
- Stopped abort()ing in library code, when possible.
- Made libfsverity_digest use native endianness.
- Moved file_size into the merkle_tree_params.
- Made libfsverity_get_hash_name() just return the static strings.
- Made some variables in the API uint32_t instead of uint16_t.
- Shared parse_hash_alg_option() between cmd_enable and cmd_sign.
- Lots of other fixes.
(Folded in a couple Makefile fixes from Jes.)
Reviewed-by: Jes Sorensen <jsorensen@fb.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
fundavid