We are not guaranteed to have the padding_pkcs_15_verify symbol since
commit 92c960bc1d ("lib: rsa: Remove #ifdefs from rsa.h"), and
commit 61416fe9df ("Kconfig: FIT_SIGNATURE should not select RSA_VERIFY")
The padding_algos only make sense with RSA verification, which can now
be disabled in lieu of ECDSA. In fact this will lead to build failures
because of the missing symbol mentioned earlier.
To resolve this, move the padding_algos to a linker list, with
declarations moved to rsa_verify.c. This is consistent with commit
6909edb4ce ("image: rsa: Move verification algorithm to a linker list")
One could argue that the added #ifdef USE_HOSTCC is ugly, and should
be hidden within the U_BOOT_PADDING_ALGO() macro. However, this would
be inconsistent with the "cryptos" list. This logic for was not
previously explored:
Without knowledge of the U_BOOT_PADDING_ALGO() macro, its use is
similar to something being declared. However, should #ifndef
USE_HOSTCC be part of the macro, it would not be obvious that it
behaves differently on host code and target code. Having the #ifndef
outside the macro makes this obvious.
Also, the #ifdef is not always necessary. For example ecda-verify
makes use of U_BOOT_CRYPTO_ALGO() without any accompanying #ifdefs.
The fundamental issue is a lack of separation of host and target code
in rsa_verify. Therefore, the declaration of a padding algo with the
external #ifdef is more readable and consistent.
Signed-off-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>
If keydir is not provided but name is we want to use name as key_id.
But with the current coding name is only used on its own if it is NULL
and keydir is provided which never occurs.
Fixes: 824ee745fb ("lib/rsa: Use the 'keyfile' argument from mkimage")
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
With glibc 2.33 (Ubuntu package glibc6 2.33-0ubuntu9) building
sifive_unmatched_defconfig results in:
In file included from /usr/include/stdio.h:866,
from ././include/compiler.h:26,
from <command-line>:
In function ‘snprintf’,
inlined from ‘rsa_engine_get_priv_key’ at ./tools/../^:273:4:
/usr/include/riscv64-linux-gnu/bits/stdio2.h:71:10: warning:
‘%s’ directive argument is null [-Wformat-truncation=]
71 | return __builtin___snprintf_chk (__s, __n,
__USE_FORTIFY_LEVEL - 1,
|
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
72 | __glibc_objsize (__s), __fmt,
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
73 | __va_arg_pack ());
| ~~~~~~~~~~~~~~~~~
Avoid passing a NULL string.
Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Older OpenSSL and libressl versions have a slightly different API.
This require #ifdefs to support. However, we still can't support it
because the ECDSA path does not compile with these older versions.
These #ifdefs are truly a vestigial appendage.
Alternatively, the ECDSA path could be updated for older libraries,
but this requires significant extra code, and #ifdefs. Those libraries
are over three years old, and there concerns whether it makes sense to
build modern software for real world use against such old libraries.
Thusly, remove #ifdefs and code for old OpenSSL and LibreSSL support.
Signed-off-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>
This patch adds the possibility to pass the PIN the OpenSSL Engine
used during signing via the environment variable MKIMAGE_SIGN_PIN.
This follows the approach used during kernel module
signing ("KBUILD_SIGN_PIN") or UBIFS image
signing ("MKIMAGE_SIGN_PIN").
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
*sig_size isn't set until later so use the correct variables.
Signed-off-by: Donald Chan <hoiho@lab126.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Move the RSA verification crytpo_algo structure out of the
crypto_algos array, and into a linker list.
Although it appears we are adding an #ifdef to rsa-verify.c, the gains
outweigh this small inconvenience. This is because rsa_verify() is
defined differently based on #ifdefs. This change allows us to have
a single definition of rsa_verify().
Signed-off-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Drop the ENABLE and SUPPORT parts of this, which are redundant.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>
Signed-off-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>
Most modern OpenSSL engines have methods overridden at the EVP level rather
than at RSA level, to make these engines work properly with mkimage, the RSA
signing code needs to switch to using EVP_* APIs as much as possible.
Signed-off-by: Donald Chan <hoiho@lab126.com>
[trini: Rebase on top of keyfile changes]
Signed-off-by: Tom Rini <trini@konsulko.com>
Keys can be derived from keydir, and the "key-name-hint" property of
the FIT. They can also be specified ad-literam via 'keyfile'. Update
the RSA signing path to use the appropriate one.
Signed-off-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
fdt_add_bignum() is useful for algorithms other than just RSA. To
allow its use for ECDSA, move it to a common file under lib/.
The new file is suffixed with '-libcrypto' because it has a direct
dependency on openssl. This is due to the use of the "BIGNUM *" type.
Signed-off-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
rsa-checksum.c sontains the hash_calculate() implementations. Despite
the "rsa-" file prefix, this function is useful for other algorithms.
To prevent confusion, move this file to lib/, and rename it to
hash-checksum.c, to give it a more "generic" feel.
Signed-off-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Currently we fail silently if there is an algorithm mismatch. To help
distinguish this failure condition.
Signed-off-by: Sean Anderson <sean.anderson@seco.com>
In the function rsa_verify_hash, if the "main" key doesn't
work, u-boot try others keys. But it searches those keys
in the FIT image instead of the u-boot device tree.
Signed-off-by: Philippe Reynes <philippe.reynes@softathome.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Change parameter type to avoid compilation error:
In file included from ./tools/../lib/rsa/rsa-verify.c:23:0,
from tools/lib/rsa/rsa-verify.c:1:
include/u-boot/rsa-mod-exp.h:69:18: error: unknown type name ‘u32’; did you mean ‘__u32’?
int zynq_pow_mod(u32 *keyptr, u32 *inout);
^~~
__u32
include/u-boot/rsa-mod-exp.h:69:31: error: unknown type name ‘u32’; did you mean ‘__u32’?
int zynq_pow_mod(u32 *keyptr, u32 *inout);
^~~
__u32
Fixes: 37e3a36a54 ("xilinx: zynq: Add support to secure images")
Signed-off-by: Michal Simek <michal.simek@xilinx.com>
Commit fdf0819afb (rsa: fix alignment issue when getting public
exponent) changed the logic to avoid doing an 8-byte access to a
possibly-not-8-byte-aligned address.
However, using rsa_convert_big_endian is wrong: That function converts
an array of big-endian (32-bit) words with the most significant word
first (aka a BE byte array) to an array of cpu-endian words with the
least significant word first. While the exponent is indeed _stored_ as
a big-endian 64-bit word (two BE words with MSW first), we want to
extract it as a cpu-endian 64 bit word. On a little-endian host,
swapping the words and byte-swapping each 32-bit word works, because
that's the same as byte-swapping the whole 64 bit word. But on a
big-endian host, the fdt32_to_cpu are no-ops, but
rsa_convert_big_endian() still does the word-swapping, breaking
verified boot.
To fix that, while still ensuring we don't do unaligned accesses, add
a little helper that first memcpy's the bytes to a local fdt64_t, then
applies fdt64_to_cpu(). [The name is chosen based on the
[bl]eXX_to_cpup in linux/byteorder/generic.h].
Fixes: fdf0819afb ("rsa: fix alignment issue when getting public exponent")
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Reviewed-by: Simon Glass <sjg@chromium.org>
The algo name should match between the FIT's signature node and the
U-Boot's control FDT.
If we do not check it, U-Boot's control FDT can expect sha512 hash but
nothing will prevent to accept image with sha1 hash if the signature is correct.
Signed-off-by: Matthieu CASTET <castet.matthieu@free.fr>
Fixes problem for unaligned 32bit big-endian access in
lib/rsa/rsa-keyprop.c.
Exchanges br_i32_decode() with get_unaligned_be32().
This will keep the unaligned access for architectures capable and will do
some byte-shift magic for the not so capable ones.
Reported-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Signed-by: Robert Reither <robert.reither@external.thalesgroup.com>
Remove unused include.
Reviewed-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
After allocating to pointer ctx we should check that pointer and not
another pointer already checked above.
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Reviewed-by: Simon Glass <sjg@chromium.org>
This function will be used to implement public_key_verify_signature()
in a later patch. rsa_verify() is not suitable here because calculation
of message digest is not necessary.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Reviewed-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
rsa_verify() expects a memory region and wants to do the hashing itself,
but there may be cases where the hashing is done via other means,
like hashing a squashfs rootfs.
So add rsa_verify_hash() to allow verifiying a signature against
an existing hash. As this entails the same verification routines
we can just move the relevant code over from rsa_verify() and also
call rsa_verify_hash() from there.
Signed-off-by: Heiko Stuebner <heiko.stuebner@theobroma-systems.com>
padding_pss_verify only works with the default pss salt setting of -2
(length to be automatically determined based on the PSS block structure)
not -1 (salt length set to the maximum permissible value), which makes
verifications of signatures with that saltlen fail.
Until this gets implemented at least document this behaviour.
Signed-off-by: Heiko Stuebner <heiko.stuebner@theobroma-systems.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
n, rr and rrtmp are used for internal calculations, but in the end
the results are copied into separately allocated elements of the
actual key_prop, so the n, rr and rrtmp elements are not used anymore
when returning from the function and should of course be freed.
Signed-off-by: Heiko Stuebner <heiko.stuebner@theobroma-systems.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
When calculating rrtmp/rr rsa_gen_key_prop() tries to make
(((rlen + 31) >> 5) + 1) steps in the rr uint32_t array and
(((rlen + 7) >> 3) + 1) / 4 steps in uint32_t rrtmp[]
with rlen being num_bits * 2
On a 4096bit key this comes down to to 257 uint32_t elements
in rr and 256 elements in rrtmp but with the current allocation
rr and rrtmp only have 129 uint32_t elements.
On 2048bit keys this works by chance as the defined max_rsa_size=4096
allocates a suitable number of elements, but with an actual 4096bit key
this results in other memory parts getting overwritten.
So as suggested by Heinrich Schuchardt just use the actual bit-size
of the key as base for the size calculation, in turn making the code
compatible to any future keysizes.
Suggested-by: Heinrich Schuchardt <xypron.debian@gmx.de>
Signed-off-by: Heiko Stuebner <heiko.stuebner@theobroma-systems.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
rrtmp needs 2 + (((*prop)->num_bits * 2) >> 5) array elements.
Reviewed-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
The exponent field of struct key_prop gets allocated an uint64_t,
and the contents are positioned from the back, so an exponent of
"0x01 0x00 0x01" becomes 0x0 0x0 0x0 0x0 0x0 0x1 0x0 0x1"
Right now rsa_gen_key_prop() allocates a uint64_t but sets exp_len
to the size returned from the parser, while on the other hand the
when getting the key from the devicetree exp_len always gets set to
sizeof(uint64_t).
So bring that in line with the established code.
Signed-off-by: Heiko Stuebner <heiko.stuebner@theobroma-systems.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Right now in multiple places there are only checks for the full
CONFIG_RSA_VERIFY_WITH_PKEY option, not split into main,spl,tpl variants.
This breaks when the rsa functions get enabled for SPL, for example to
verify u-boot proper from spl.
So fix this by using the existing helpers to distinguis between
build-steps.
Signed-off-by: Heiko Stuebner <heiko.stuebner@theobroma-systems.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
While the SPL may want to do signature checking this won't be
the case for TPL in all cases, as TPL is mostly used when the
amount of initial memory is not enough for a full SPL.
So on a system where SPL uses DM but TPL does not we currently
end up with a TPL compile error of:
lib/rsa/rsa-verify.c:48:25: error: dereferencing pointer to incomplete type ‘struct checksum_algo’
To prevent that change the $(SPL_) to $(SPL_TPL_) to distinguish
between both. If someone really needs FIT signature checking in
TPL as well, a new TPL_RSA_VERIFY config symbol needs to be added.
Signed-off-by: Heiko Stuebner <heiko.stuebner@theobroma-systems.com>
Reviewed-by: Philipp Tomsich <philipp.tomsich@theobroma-systems.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
If "object=" is specified in "keydir" when using the pkcs11 engine do
not append another "object=<key-name-hint>". This makes it possible to
use object names other than the key name hint. These two string
identifiers are not necessarily equal.
Signed-off-by: Jan Luebbe <jlu@pengutronix.de>
Signed-off-by: Bastian Krause <bst@pengutronix.de>
Reviewed-by: George McCollister <george.mccollister@gmail.com>
stdlib.h is the header for malloc since at least c89/c90.
There is no system malloc.h on OpenBSD and trying to use malloc.h here
falls back to the U-Boot malloc.h breaking the build of qemu_arm64.
Fixes: 336d4615f8 ("dm: core: Create a new header file for 'compat' features")
Signed-off-by: Jonathan Gray <jsg@jsg.id.au>
To fill the exponent field of the rsa_public_key struct, rsa_mod_exp_sw
did a cast to uint64_t of the key_prop->public_exponent field.
But that alignment is not guaranteed in all cases.
This came to light when in my spl-fit-signature the key-name exceeded
a certain length and with it the verification then started failing.
(naming it "integrity" worked fine, "integrity-uboot" failed)
key_prop.public_exponent itself is actually a void-pointer, fdt_getprop()
also just returns such a void-pointer and inside the devicetree the 64bit
exponent is represented as 2 32bit numbers, so assuming a 64bit alignment
can lead to false reads.
So just use the already existing rsa_convert_big_endian() to do the actual
conversion from the dt's big-endian to the needed uint64 value.
Fixes: fc2f4246b4 ("rsa: Split the rsa-verify to separate the modular exponentiation")
Signed-off-by: Heiko Stuebner <heiko.stuebner@theobroma-systems.com>
Reviewed-by: Philipp Tomsich <philipp.tomsich@theobroma-systems.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
These are used in multiple places so update them to use a shared #define.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Philippe Reynes <philippe.reynes@softathome.com>
This function, and hence rsa_verify(), will perform RSA verification
with two essential parameters for a RSA public key in contract of
rsa_verify_with_keynode(), which requires additional three parameters
stored in FIT image.
It will be used in implementing UEFI secure boot, i.e. image authentication
and variable authentication.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Reviewed-by: Simon Glass <sjg@chromium.org>
In the current implementation of FIT_SIGNATURE, five parameters for
a RSA public key are required while only two of them are essential.
(See rsa-mod-exp.h and uImage.FIT/signature.txt)
This is a result of considering relatively limited computer power
and resources on embedded systems, while such a assumption may not
be quite practical for other use cases.
In this patch, added is a function, rsa_gen_key_prop(), which will
generate additional parameters for other uses, in particular
UEFI secure boot, on the fly.
Note: the current code uses some "big number" routines from BearSSL
for the calculation.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
In the next couple of commits, under new CONFIG_RSA_VERIFY_WITH_PKEY,
rsa_verify() will be extended to be able to perform RSA decryption without
additional RSA key properties from FIT image, i.e. rr and n0inv.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Reviewed-by: Simon Glass <sjg@chromium.org>
Introduce new configuration, CONFIG_RSA_VERIFY which will decouple building
RSA functions from FIT verification and allow for adding a RSA-based
signature verification for other file formats, in particular PE file
for UEFI secure boot.
Signed-off-by: AKASHI Takahiro <takahiro.akashi@linaro.org>
Reviewed-by: Simon Glass <sjg@chromium.org>
At present dm/device.h includes the linux-compatible features. This
requires including linux/compat.h which in turn includes a lot of headers.
One of these is malloc.h which we thus end up including in every file in
U-Boot. Apart from the inefficiency of this, it is problematic for sandbox
which needs to use the system malloc() in some files.
Move the compatibility features into a separate header file.
Signed-off-by: Simon Glass <sjg@chromium.org>
Currently, if image verification with a required key fails, rsa_verify()
code tries to find another key to verify the FIT image. This however, is
not the intended behavior as the documentation says that required keys
"must be verified for the image / configuration to be considered valid".
This patch fixes the issue by making rsa_verify() return immediately if
the verification of a required key fails.
Signed-off-by: Daniele Alessandrelli <daniele.alessandrelli@gmail.com>
There are multiple other openssl engines used by HSMs that can be used to
sign FIT images instead of forcing users to use pkcs11 type of service.
Relax engine selection so that other openssl engines can be specified and
use generic key id definition formula.
Signed-off-by: Vesa Jääskeläinen <vesa.jaaskelainen@vaisala.com>
Cc: Tom Rini <trini@konsulko.com>
The pointer checksum were used before checking that it
isn't NULL. We move the code that use it after the check.
Reported-by: Coverity (CID: 185835)
Signed-off-by: Philippe Reynes <philippe.reynes@softathome.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
We add the support of the padding pss for rsa signature.
This new padding is often recommended instead of pkcs-1.5.
Signed-off-by: Philippe Reynes <philippe.reynes@softathome.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
The rsa signature use a padding algorithm. By default, we use the
padding pkcs-1.5. In order to add some new padding algorithm, we
add a padding framework to manage several padding algorithm.
The choice of the padding is done in the file .its.
Signed-off-by: Philippe Reynes <philippe.reynes@softathome.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Previous implementation of the rsa signature was using
the openssl API EVP_Sign*, but the new openssl API
EVP_DigestSign* is more flexible. So we move to this
new API.
Signed-off-by: Philippe Reynes <philippe.reynes@softathome.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
