Add support for ECDSA image signing (with test)
Subsystems (2)
v4 cover Add support for ECDSA image signing (with test)
## Purpose and intent
The purpose of this series is to enable ECDSA as an alternative to RSA for FIT signing. As new chips have built-in support for ECDSA verified boot, it makes sense to stick to one signing algorithm, instead of resorting to RSA for u-boot images.
The focus of this series is signing an existing FIT image:
mkimage -F some-existing.fit --signing-key some/key.pem
Signing while assembling a FIT is not a tested use case.
# Implementation
## Code organization
Unlike the RSA path, which mixes host and firmware code in the same, source files, this series keeps a very clear distinction. ecdsa-libcrypto.c is intended to be used for host code and only for host code. There is more opportunity for code reuse this way.
## Signing
There is one major difference from the RSA path. The 'key-name-hint' property is ignored in the ECDSA path. There are two reasons:
(1) The intent of 'key-name-hint' is not clear
(2) Initial implementation is much easier to review
There is an intentional side-effect. The RSA path takes 'key-name-hint' to decide which key file to read from disk. In the context of "which fdt node describes my signing key", this makes sense. On the other hand, 'key-name-hint' is also used as the basename of where the key is on the filesystem. This leads to some funny search paths, such as
"some/dir/(null).key"
So I am using the -K option to mkimage as the _full_ path to the key file. It doesn't have to be named .key, it doesn't have to be named .crt, and it doesn't have to exist in a particular directory (as is the case for the RSA path). I understand and recognize that this discrepancy must be resolved, but resolving it right now would make the initial implementation much harder and longer.
# Testing
test/py/tests/test_fit_ecdsa.py is implemented withe the goal to check that the signing is done correctly, and that the signature is encoded in the proper raw format. Verification is done with pyCryptodomex, so this test will catch both coding errors and openssl bugs. This is the only scope of testing proposed here.
# Things not yet resolved:
- is mkimage '-k' supposed to be a directory or file path
I'm hoping I can postpone answering this question pending further discussion.
Changes since v3:
- Don't use 'log_msg_ret()', as it's not available host-side
Changes since v1 and v2:
- Added lots of function comments
- Replaced hardcoded error numbers with more meaningful errno numbers
- Changed some error paths to use 'return log_msg_ret'
Alexandru Gagniuc (6):
lib: Rename rsa-checksum.c to hash-checksum.c lib/rsa: Make fdt_add_bignum() available outside of RSA code
lib: Add support for ECDSA image signing
doc: signature.txt: Document devicetree format for ECDSA keys test/py: Add pycryptodomex to list of required pakages test/py: ecdsa: Add test for mkimage ECDSA signing
common/image-fit-sig.c | 2 +-
common/image-sig.c | 13 +-
doc/uImage.FIT/signature.txt | 7 +-
include/image.h | 5 +-
include/u-boot/ecdsa.h | 94 ++++++
include/u-boot/fdt-libcrypto.h | 27 ++
.../{rsa-checksum.h => hash-checksum.h} | 0
lib/Makefile | 1 +
lib/crypto/pkcs7_verify.c | 2 +-
lib/crypto/x509_public_key.c | 2 +-
lib/ecdsa/ecdsa-libcrypto.c | 306 ++++++++++++++++++
lib/fdt-libcrypto.c | 72 +++++
lib/{rsa/rsa-checksum.c => hash-checksum.c} | 3 +-
lib/rsa/Makefile | 2 +-
lib/rsa/rsa-sign.c | 65 +---
test/py/requirements.txt | 1 +
test/py/tests/test_fit_ecdsa.py | 111 +++++++
tools/Makefile | 7 +-
18 files changed, 645 insertions(+), 75 deletions(-)
create mode 100644 include/u-boot/ecdsa.h
create mode 100644 include/u-boot/fdt-libcrypto.h
rename include/u-boot/{rsa-checksum.h => hash-checksum.h} (100%)
create mode 100644 lib/ecdsa/ecdsa-libcrypto.c
create mode 100644 lib/fdt-libcrypto.c
rename lib/{rsa/rsa-checksum.c => hash-checksum.c} (96%)
create mode 100644 test/py/tests/test_fit_ecdsa.pySignature · 1 line
2.26.2
PATCH v4 1/6 lib: Rename rsa-checksum.c to hash-checksum.c
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/crypto, and rename it to hash-checksum.c, to give it a more "generic" feel.
---
common/image-fit-sig.c | 2 +-
common/image-sig.c | 2 +-
include/image.h | 2 +-
include/u-boot/{rsa-checksum.h => hash-checksum.h} | 0
lib/Makefile | 1 +
lib/crypto/pkcs7_verify.c | 2 +-
lib/crypto/x509_public_key.c | 2 +-
lib/{rsa/rsa-checksum.c => hash-checksum.c} | 3 ++-
lib/rsa/Makefile | 2 +-
tools/Makefile | 3 ++-
10 files changed, 11 insertions(+), 8 deletions(-)
rename include/u-boot/{rsa-checksum.h => hash-checksum.h} (100%)
rename lib/{rsa/rsa-checksum.c => hash-checksum.c} (96%)common/image-fit-sig.c +1 -1
@@ -15,7 +15,7 @@ DECLARE_GLOBAL_DATA_PTR;#include <fdt_region.h> #include <image.h> #include <u-boot/rsa.h> -#include <u-boot/rsa-checksum.h> +#include <u-boot/hash-checksum.h> #define IMAGE_MAX_HASHED_NODES 100
common/image-sig.c +1 -1
@@ -16,7 +16,7 @@ DECLARE_GLOBAL_DATA_PTR;#endif /* !USE_HOSTCC*/ #include <image.h> #include <u-boot/rsa.h> -#include <u-boot/rsa-checksum.h> +#include <u-boot/hash-checksum.h> #define IMAGE_MAX_HASHED_NODES 100
include/image.h +1 -1
@@ -1258,7 +1258,7 @@ struct image_region { };#if IMAGE_ENABLE_VERIFY -# include <u-boot/rsa-checksum.h> +# include <u-boot/hash-checksum.h> #endif struct checksum_algo {const char *name;
include/u-boot/hash-checksum.h renamed from include/u-boot/rsa-checksum.h +0 -0
lib/Makefile +1 -0
@@ -60,6 +60,7 @@ endif obj-$(CONFIG_$(SPL_)ACPIGEN) += acpi/ obj-$(CONFIG_$(SPL_)MD5) += md5.o obj-$(CONFIG_$(SPL_)RSA) += rsa/ +obj-$(CONFIG_FIT_SIGNATURE) += hash-checksum.o obj-$(CONFIG_SHA1) += sha1.o obj-$(CONFIG_SHA256) += sha256.o obj-$(CONFIG_SHA512_ALGO) += sha512.o
lib/crypto/pkcs7_verify.c +1 -1
@@ -15,7 +15,7 @@#include <linux/bitops.h> #include <linux/compat.h> #include <linux/asn1.h> -#include <u-boot/rsa-checksum.h> +#include <u-boot/hash-checksum.h> #include <crypto/public_key.h> #include <crypto/pkcs7_parser.h> #else
lib/crypto/x509_public_key.c +1 -1
@@ -19,7 +19,7 @@#include <linux/kernel.h> #ifdef __UBOOT__#include <crypto/x509_parser.h> -#include <u-boot/rsa-checksum.h> +#include <u-boot/hash-checksum.h> #else #include <linux/slab.h> #include <keys/asymmetric-subtype.h>
lib/hash-checksum.c renamed from lib/rsa/rsa-checksum.c +2 -1
@@ -13,7 +13,8 @@#else #include "fdt_host.h" #endif -#include <u-boot/rsa.h> +#include <hash.h> +#include <image.h> int hash_calculate(const char *name, const struct image_region region[],
lib/rsa/Makefile +1 -1
@@ -5,6 +5,6 @@ # (C) Copyright 2000-2007 # Wolfgang Denk, DENX Software Engineering, wd at denx.de. -obj-$(CONFIG_$(SPL_TPL_)RSA_VERIFY) += rsa-verify.o rsa-checksum.o +obj-$(CONFIG_$(SPL_TPL_)RSA_VERIFY) += rsa-verify.o obj-$(CONFIG_$(SPL_TPL_)RSA_VERIFY_WITH_PKEY) += rsa-keyprop.o obj-$(CONFIG_RSA_SOFTWARE_EXP) += rsa-mod-exp.o
tools/Makefile +2 -1
@@ -67,7 +67,7 @@ LIBFDT_OBJS := $(addprefix libfdt/, fdt.o fdt_ro.o fdt_wip.o fdt_sw.o fdt_rw.o \ fdt_strerror.o fdt_empty_tree.o fdt_addresses.o fdt_overlay.o) RSA_OBJS-$(CONFIG_FIT_SIGNATURE) := $(addprefix lib/rsa/, \ - rsa-sign.o rsa-verify.o rsa-checksum.o \ + rsa-sign.o rsa-verify.o \ rsa-mod-exp.o) AES_OBJS-$(CONFIG_FIT_CIPHER) := $(addprefix lib/aes/, \
@@ -105,6 +105,7 @@ dumpimage-mkimage-objs := aisimage.o \ $(ROCKCHIP_OBS) \ socfpgaimage.o \ lib/crc16.o \ + lib/hash-checksum.o \ lib/sha1.o \ lib/sha256.o \ lib/sha512.o \
Signature · 1 line
2.26.2
PATCH v4 1/6 lib: Rename rsa-checksum.c to hash-checksum.c
Hi,
Quoted 4 lines · Alexandru Gagniuc
rsa-checksum.csontains thehash_calculate()implementations. Despite the "rsa-" file prefix, this function is useful for other algorithms.To prevent confusion, move this file to lib/crypto, and rename it to
Today the file is moved in lib and not in lib/crypto...
change the comment or the file location.
Quoted 66 lines
hash-checksum.c, to give it a more "generic" feel.Signed-off-by Alexandru Gagniuc <[email protected]>--- common/image-fit-sig.c | 2 +- common/image-sig.c | 2 +- include/image.h | 2 +- include/u-boot/{rsa-checksum.h => hash-checksum.h} | 0 lib/Makefile | 1 + lib/crypto/pkcs7_verify.c | 2 +- lib/crypto/x509_public_key.c | 2 +- lib/{rsa/rsa-checksum.c => hash-checksum.c} | 3 ++- lib/rsa/Makefile | 2 +- tools/Makefile | 3 ++- 10 files changed, 11 insertions(+), 8 deletions(-)rename
include/u-boot/{rsa-checksum.h=>hash-checksum.h} (100%) renamelib/{rsa/rsa-checksum.c=>hash-checksum.c} (96%)common/image-fit-sig.c +1
-1@@ -15,7 +15,7 @@ DECLARE_GLOBAL_DATA_PTR;#include <fdt_region.h> #include <image.h> #include <u-boot/rsa.h> -#include <u-boot/rsa-checksum.h> +#include <u-boot/hash-checksum.h> #define IMAGE_MAX_HASHED_NODES 100common/image-sig.c +1
-1@@ -16,7 +16,7 @@ DECLARE_GLOBAL_DATA_PTR;#endif /* !USE_HOSTCC*/ #include <image.h> #include <u-boot/rsa.h> -#include <u-boot/rsa-checksum.h> +#include <u-boot/hash-checksum.h> #define IMAGE_MAX_HASHED_NODES 100include/image.h +1
-1@@ -1258,7 +1258,7 @@ struct image_region { };#if IMAGE_ENABLE_VERIFY -# include <u-boot/rsa-checksum.h> +# include <u-boot/hash-checksum.h> #endif struct checksum_algo {const char *name;include/u-boot/hash-checksum.h renamed from include/u-boot/rsa-checksum.h +0
-0lib/Makefile +0
-0
lib/Makefile endif · line 63
Quoted 4 lines
lib/Makefile +1
-0@@ -60,6 +60,7 @@ endif obj-$(CONFIG_$(SPL_)ACPIGEN) += acpi/ obj-$(CONFIG_$(SPL_)MD5) += md5.o obj-$(CONFIG_$(SPL_)RSA) += rsa/ +obj-$(CONFIG_FIT_SIGNATURE) += hash-checksum.o
compiled if CONFIG_FIT_SIGNATURE = y
but after used in mkimage even if CONFIG_FIT_SIGNATURE is not used...
(see after dumpimage-mkimage-objs)
it could be more simple to have :
obj-y += hash-checksum.o
or add a new config CONFIG_HASH (but we need to managed dependancy in Kconfig / Makefile)
(...)
Quoted 14 lines
tools/Makefile +1
-1@@ -67,7 +67,7 @@ LIBFDT_OBJS := $(addprefix libfdt/, fdt.o fdt_ro.o fdt_wip.o fdt_sw.o fdt_rw.o \ fdt_strerror.o fdt_empty_tree.o fdt_addresses.o fdt_overlay.o) RSA_OBJS-$(CONFIG_FIT_SIGNATURE) := $(addprefix lib/rsa/, \ - rsa-sign.o rsa-verify.o rsa-checksum.o \ + rsa-sign.o rsa-verify.o \ rsa-mod-exp.o) AES_OBJS-$(CONFIG_FIT_CIPHER) := $(addprefix lib/aes/, \
tools/Makefile dumpimage-mkimage-objs := aisimage.o \ · line 111
Quoted 7 lines
tools/Makefile +1
-0@@ -105,6 +105,7 @@ dumpimage-mkimage-objs := aisimage.o \ $(ROCKCHIP_OBS) \ socfpgaimage.o \ lib/crc16.o \ + lib/hash-checksum.o \ lib/sha1.o \ lib/sha256.o \ lib/sha512.o \
lib/hash-checksum.o is required here...
Regards
Patrick
PATCH v4 2/6 lib/rsa: Make fdt_add_bignum() available outside of RSA code
In reply to Alexandru Gagniucfdt_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.
--- include/u-boot/fdt-libcrypto.h | 27 +++++++++++++ lib/fdt-libcrypto.c | 72 ++++++++++++++++++++++++++++++++++ lib/rsa/rsa-sign.c | 65 +----------------------------- tools/Makefile | 1 + 4 files changed, 101 insertions(+), 64 deletions(-) create mode 100644 include/u-boot/fdt-libcrypto.h create mode 100644 lib/fdt-libcrypto.c
include/u-boot/fdt-libcrypto.h new file +27 -0
@@ -0,0 +1,27 @@ +/* SPDX-License-Identifier: GPL-2.0+ */ +/* +* Copyright (c) 2020, Alexandru Gagniuc <[email protected]> +* Copyright (c) 2013, Google Inc. +*/ + +#ifndef _FDT_LIBCRYPTO_H +#define _FDT_LIBCRYPTO_H + +#include <openssl/bn.h> + +/** +* fdt_add_bignum() - Write a libcrypto BIGNUM as an FDT property +* +* Convert a libcrypto BIGNUM * into a big endian array of integers. +* +* @blob: FDT blob to modify +* @noffset: Offset of the FDT node +* @prop_name: What to call the property in the FDT +* @num: pointer to a libcrypto big number +* @num_bits: How big is 'num' in bits? +* @return 0 if all good all working, -ve on horror +*/ +int fdt_add_bignum (void *blob,int noffset,const char *prop_name, +BIGNUM *num,int num_bits); + +#endif /* _FDT_LIBCRYPTO_H */
lib/fdt-libcrypto.c new file +72 -0
72 lines not on this page · raw message
lib/rsa/rsa-sign.c +1 -64
77 lines not on this page · raw message
tools/Makefile +1 -0
@@ -106,6 +106,7 @@ dumpimage-mkimage-objs := aisimage.o \ socfpgaimage.o \ lib/crc16.o \ lib/hash-checksum.o \ + lib/fdt-libcrypto.o \ lib/sha1.o \ lib/sha256.o \ lib/sha512.o \
Signature · 1 line
2.26.2
PATCH v4 2/6 lib/rsa: Make fdt_add_bignum() available outside of RSA code
Hi,
Quoted 15 lines · Alexandru Gagniuc
fdt_add_bignum()is useful for algorithms other than just RSA. To allow its use for ECDSA, move it to a common file underlib/.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 <[email protected]>--- include/u-boot/fdt-libcrypto.h | 27 +++++++++++++ lib/fdt-libcrypto.c | 72 ++++++++++++++++++++++++++++++++++ lib/rsa/rsa-sign.c | 65 +----------------------------- tools/Makefile | 1 + 4 files changed, 101 insertions(+), 64 deletions(-)create mode 100644
include/u-boot/fdt-libcrypto.hcreate mode 100644lib/fdt-libcrypto.c
(...)
tools/Makefile dumpimage-mkimage-objs := aisimage.o \ · line 112
Quoted 12 lines
tools/Makefile +1
-0@@ -106,6 +106,7 @@ dumpimage-mkimage-objs := aisimage.o \ socfpgaimage.o \ lib/crc16.o \ lib/hash-checksum.o \ + lib/fdt-libcrypto.o \ lib/sha1.o \ lib/sha256.o \ lib/sha512.o \
This part cause compilation issue when CONFIG_FIT_SIGNATURE is not defined / when openssl is not used
fdt-libcrypto.c:(.text+0x2f): undefined reference to BN_new' fdt-libcrypto.c:(.text+0x37): undefined reference to BN_new' fdt-libcrypto.c:(.text+0x44): undefined reference to BN_new' fdt-libcrypto.c:(.text+0x51): undefined reference to BN_new' fdt-libcrypto.c:(.text+0x7d): undefined reference to BN_CTX_new' fdt-libcrypto.c:(.text+0x9d): undefined reference to BN_set_word' fdt-libcrypto.c:(.text+0xaf): undefined reference to BN_set_word' fdt-libcrypto.c:(.text+0xc5): undefined reference to BN_exp' fdt-libcrypto.c:(.text+0x135): undefined reference to BN_div' fdt-libcrypto.c:(.text+0x13d): undefined reference to BN_get_word' fdt-libcrypto.c:(.text+0x172): undefined reference to BN_rshift' fdt-libcrypto.c:(.text+0x1a9): undefined reference to BN_free' fdt-libcrypto.c:(.text+0x1b3): undefined reference to BN_free' fdt-libcrypto.c:(.text+0x1bd): undefined reference to BN_free' fdt-libcrypto.c:(.text+0x1c5): undefined reference to `BN_free'
libssl not always include in mkimage, see the lines:
? # MXSImage needs LibSSL
? ifneq
($(CONFIG_MX23)$(CONFIG_MX28)$(CONFIG_ARMADA_38X)$(CONFIG_ARMADA_39X)$(CONFIG_FIT_SIGNATURE)$(CONFIG_FIT_CIPHER),)
? HOSTCFLAGS_kwbimage.o += \
? ? ? $(shell pkg-config --cflags libssl libcrypto 2> /dev/null || echo "")
? HOSTLDLIBS_mkimage += \
?? ?? $(shell pkg-config --libs libssl libcrypto 2> /dev/null || echo
"-lssl -lcrypto")
=> I propose to add the added lib only when CONFIG_FIT_SIGNATURE is
????? activated by removing dumpimage-mkimage-objs udpate and with:
- FIT_SIG_OBJS-$(CONFIG_FIT_SIGNATURE) := common/image-sig.o common/image-fit-sig.o
+ FIT_SIG_OBJS-$(CONFIG_FIT_SIGNATURE) := common/image-sig.o common/image-fit-sig.o lib/fdt-libcrypto.o
Regards
Patrick
PATCH v4 2/6 lib/rsa: Make fdt_add_bignum() available outside of RSA code
In reply to Alexandru GagniucQuoted 7 lines · Alexandru Gagniuc
fdt_add_bignum()is useful for algorithms other than just RSA. To allow its use for ECDSA, move it to a common file underlib/.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 <[email protected]>
This makes a large number of platform tools fail to build such as j7200_evm_a72.
Signature · 8 lines
Tom -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 659 bytes Desc: not available URL: <https://lists.denx.de/pipermail/u-boot/attachments/20210127/8ea63b1f/attachment.sig>
PATCH v4 3/6 lib: Add support for ECDSA image signing
In reply to Alexandru Gagniucmkimage supports rsa2048, and rsa4096 signatures. With newer silicon now supporting hardware-accelerated ECDSA, it makes sense to expand signing support to elliptic curves.
Implement host-side ECDSA signing and verification with libcrypto. Device-side implementation of signature verification is beyond the scope of this patch.
--- common/image-sig.c | 11 +- include/image.h | 3 + include/u-boot/ecdsa.h | 94 +++++++++++ lib/ecdsa/ecdsa-libcrypto.c | 306 ++++++++++++++++++++++++++++++++++++ tools/Makefile | 3 + 5 files changed, 415 insertions(+), 2 deletions(-) create mode 100644 include/u-boot/ecdsa.h create mode 100644 lib/ecdsa/ecdsa-libcrypto.c
common/image-sig.c +9 -2
@@ -15,6 +15,7 @@DECLARE_GLOBAL_DATA_PTR ;#endif /* !USE_HOSTCC*/ #include <image.h> +#include <u-boot/ecdsa.h> #include <u-boot/rsa.h> #include <u-boot/hash-checksum.h>
@@ -82,8 +83,14 @@ struct crypto_algo crypto_algos[] = { .sign = rsa_sign, .add_verify_data = rsa_add_verify_data, .verify = rsa_verify, - } - + }, + { + .name = "ecdsa256", + .key_len = ECDSA256_BYTES, + .sign = ecdsa_sign, + .add_verify_data = ecdsa_add_verify_data, + .verify = ecdsa_verify, + }, };struct padding_algo padding_algos[] = {
include/image.h +3 -0
@@ -1199,16 +1199,19 @@ int calculate_hash(const void *data, int data_len, const char *algo,# if defined(CONFIG_FIT_SIGNATURE )# define IMAGE_ENABLE_SIGN 1# define IMAGE_ENABLE_VERIFY 1 +# define IMAGE_ENABLE_VERIFY_ECDSA 1# define FIT_IMAGE_ENABLE_VERIFY 1# include <openssl/evp.h> # else # define IMAGE_ENABLE_SIGN 0# define IMAGE_ENABLE_VERIFY 0 +# define IMAGE_ENABLE_VERIFY_ECDSA 0# define FIT_IMAGE_ENABLE_VERIFY 0# endif #else # define IMAGE_ENABLE_SIGN 0# define IMAGE_ENABLE_VERIFY CONFIG_IS_ENABLED(RSA_VERIFY) +# define IMAGE_ENABLE_VERIFY_ECDSA 0# define FIT_IMAGE_ENABLE_VERIFY CONFIG_IS_ENABLED(FIT_SIGNATURE)#endif
include/u-boot/ecdsa.h new file +94 -0
94 lines not on this page · raw message
lib/ecdsa/ecdsa-libcrypto.c new file +306 -0
306 lines not on this page · raw message
tools/Makefile +3 -0
@@ -70,6 +70,8 @@ RSA_OBJS-$(CONFIG_FIT_SIGNATURE) := $(addprefix lib/rsa/, \ rsa-sign.o rsa-verify.o \ rsa-mod-exp.o) +ECDSA_OBJS-$(CONFIG_FIT_SIGNATURE) := $(addprefix lib/ecdsa/, ecdsa-libcrypto.o) + AES_OBJS-$(CONFIG_FIT_CIPHER) := $(addprefix lib/aes/, \ aes-encrypt.o aes-decrypt.o)
@@ -119,6 +121,7 @@ dumpimage-mkimage-objs := aisimage.o \ gpimage.o \ gpimage-common.o \ mtk_image.o \ + $(ECDSA_OBJS-y) \ $(RSA_OBJS-y) \ $(AES_OBJS-y)
Signature · 1 line
2.26.2
PATCH v4 4/6 doc: signature.txt: Document devicetree format for ECDSA keys
In reply to Alexandru Gagniuc--- doc/uImage.FIT/signature.txt | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
doc/uImage.FIT/signature.txt +6 -1
@@ -142,7 +142,7 @@ public key in U-Boot's control FDT (using CONFIG_OF_CONTROL). Public keys should be stored as sub-nodes in a /signature node. Required properties are: -- algo: Algorithm name (e.g. "sha1,rsa2048") +- algo: Algorithm name (e.g. "sha1,rsa2048" or "sha256,ecdsa256") Optional properties are:
@@ -167,6 +167,11 @@ For RSA the following are mandatory: - rsa,r-squared: (2^num-bits)^2 as a big-endian multi-word integer - rsa,n0-inverse: -1 / modulus[0] mod 2^32 +For ECDSA the following are mandatory: +- ecdsa,curve: Name of ECDSA curve (e.g. "prime256v1") +- ecdsa,x-point: Public key X coordinate as a big-endian multi-word integer +- ecdsa,y-point: Public key Y coordinate as a big-endian multi-word integer + These parameters can be added to a binary device tree using parameter -K of the mkimage command::
Signature · 1 line
2.26.2
PATCH v4 5/6 test/py: Add pycryptodomex to list of required pakages
In reply to Alexandru GagniucWe wish to use pycryptodomex to verify code paths involving ECDSA signatures. Add it to requirements.txt so that they get picked up automatically .gitlab and .azure tasks
--- test/py/requirements.txt | 1 + 1 file changed, 1 insertion(+)
test/py/requirements.txt +1 -0
@@ -10,6 +10,7 @@ packaging==19.2 pbr==5.4.3 pluggy==0.13.0 py==1.8.0 +pycryptodomex==3.9.8 pyparsing==2.4.2 pytest==5.2.1 python-mimeparse==1.6.0
Signature · 1 line
2.26.2
PATCH v4 6/6 test/py: ecdsa: Add test for mkimage ECDSA signing
In reply to Alexandru GagniucAdd a test to make sure that the ECDSA signatures generated by mkimage can be verified successfully. pyCryptodomex was chosen as the crypto library because it integrates much better with python code. Using openssl would have been unnecessarily painful.
--- test/py/tests/test_fit_ecdsa.py | 111 ++++++++++++++++++++++++++++++++ 1 file changed, 111 insertions(+) create mode 100644 test/py/tests/test_fit_ecdsa.py
test/py/tests/test_fit_ecdsa.py new file +111 -0
@@ -0,0 +1,111 @@ +# SPDX-License-Identifier: GPL-2.0+ +# +# Copyright (c) 2020,2021 Alexandru Gagniuc <[email protected]> + +""" +Test ECDSA signing of FIT images + +This test uses mkimage to sign an existing FIT image with an ECDSA key. The +signature is then extracted, and verified against pyCryptodome. +This test doesn't run the sandbox. It only checks the host tool 'mkimage' +""" + +import pytest +import u_boot_utils as util +from Cryptodome.Hash import SHA256 +from Cryptodome.PublicKey import ECC +from Cryptodome.Signature import DSS + +class SignableFitImage(object): + """ Helper to manipulate a FIT image on disk """ + def __init__(self, cons, file_name): + self.fit = file_name + self.cons = cons + self.signable_nodes = set() + + def __fdt_list(self, path): + return util.run_and_log(self.cons, f'fdtget -l {self.fit} {path}') + + def __fdt_set(self, node, **prop_value): + for prop, value in prop_value.items(): + util.run_and_log(self.cons, f'fdtput -ts {self.fit} {node} {prop} {value}') + + def __fdt_get_binary(self, node, prop): + numbers = util.run_and_log(self.cons, f'fdtget -tbi {self.fit} {node} {prop}') + + bignum = bytearray() + for little_num in numbers.split(): + bignum.append(int(little_num)) + + return bignum + + def find_signable_image_nodes(self): + for node in self.__fdt_list('/images').split(): + image = f'/images/{node}' + if 'signature' in self.__fdt_list(image): + self.signable_nodes.add(image) + + return self.signable_nodes + + def change_signature_algo_to_ecdsa(self): + for image in self.signable_nodes: + self.__fdt_set(f'{image}/signature', algo='sha256,ecdsa256') + + def sign(self, mkimage, key_file): + util.run_and_log(self.cons, [mkimage, '-F', self.fit, f'-k{key_file}']) + + def check_signatures(self, key): + for image in self.signable_nodes: + raw_sig = self.__fdt_get_binary(f'{image}/signature', 'value') + raw_bin = self.__fdt_get_binary(image, 'data') + + sha = SHA256.new(raw_bin) + verifier = DSS.new(key, 'fips-186-3') + verifier.verify(sha, bytes(raw_sig)) + + + at pytest.mark.buildconfigspec('fit_signature') + at pytest.mark.requiredtool('dtc') + at pytest.mark.requiredtool('fdtget') + at pytest.mark.requiredtool('fdtput') +def test_fit_ecdsa(u_boot_console): + """ Test that signatures generated by mkimage are legible. """ + def generate_ecdsa_key(): + return ECC.generate(curve='prime256v1') + + def assemble_fit_image(dest_fit, its, destdir): + dtc_args = f'-I dts -O dtb -i {destdir}' + util.run_and_log(cons, [mkimage, '-D', dtc_args, '-f', its, dest_fit]) + + def dtc(dts): + dtb = dts.replace('.dts', '.dtb') + util.run_and_log(cons, f'dtc {datadir}/{dts} -O dtb -o {tempdir}/{dtb}') + + cons = u_boot_console + mkimage = cons.config.build_dir + '/tools/mkimage' + datadir = cons.config.source_dir + '/test/py/tests/vboot/' + tempdir = cons.config.result_dir + key_file = f'{tempdir}/ecdsa-test-key.pem' + fit_file = f'{tempdir}/test.fit' + dtc('sandbox-kernel.dts') + + key = generate_ecdsa_key() + + # Create a fake kernel image -- zeroes will do just fine + with open(f'{tempdir}/test-kernel.bin', 'w') as fd: + fd.write(500 * chr(0)) + + # invokations of mkimage expect to read the key from disk + with open(key_file, 'w') as f: + f.write(key.export_key(format='PEM')) + + assemble_fit_image(fit_file, f'{datadir}/sign-images-sha256.its', tempdir) + + fit = SignableFitImage(cons, fit_file) + nodes = fit.find_signable_image_nodes() + if len(nodes) == 0: + raise ValueError('FIT image has no "/image" nodes with "signature"') + + fit.change_signature_algo_to_ecdsa() + fit.sign(mkimage, key_file) + fit.check_signatures(key)
Signature · 1 line
2.26.2
v4 cover Add support for ECDSA image signing (with test)
In reply to Alexandru GagniucHi Alexandru,
Quoted 2 lines · Alexandru Gagniuc
## Purpose and intent > > The purpose of this series is to enable ECDSA as an alternative to RSA for FIT signing. As new chips have built-in support for ECDSA >
verified boot, it makes sense to stick to one signing algorithm, > instead of resorting to RSA for u-boot images. > > The focus of this series is signing an existing FIT image: > > mkimage -F some-existing.fit --signing-key some/key.pem > > Signing while assembling a FIT is not a tested use case. # > Implementation > > ## Code organization > > Unlike the RSA path, which mixes host and firmware code in the same, > source files, this series keeps a very clear distinction. > ecdsa-libcrypto.c is intended to be used for host code and only for > host code. There is more opportunity for code reuse this way. > > ## Signing > > There is one major difference from the RSA path. The 'key-name-hint' > property is ignored in the ECDSA path. There are two reasons: (1) The > intent of 'key-name-hint' is not clear (2) Initial implementation is > much easier to review >
I found in doc/uImage.FIT/signature.txt the description
??? - key-name-hint: Name of key to use for signing. The keys will
normally be in
??? a single directory (parameter -k to mkimage). For a given key
<name>, its
??? private key is stored in <name>.key and the certificate is stored in ??? <name>.crt.
And i check the code and the result is a little different:
include/image.h:1000:#define FIT_KEY_HINT??????? "key-name-hint"
common/image-fit-sig.c:89:??? info->keyname = fdt_getprop(fit, noffset, FIT_KEY_HINT, NULL);
lib/rsa/rsa-sign.c:742: ret = fdt_setprop_string(keydest, node, FIT_KEY_HINT, info->keyname);
tools/image-host.c: info->keyname = fdt_getprop(fit, noffset, FIT_KEY_HINT, NULL);
=> "keyname" provided the name oft he node in device tree for verify
lib/rsa/rsa-verify.c:518:
??????? snprintf(name, sizeof(name), "key-%s", info->keyname); ??????? node = fdt_subnode_offset(blob, sig_node, name); ??????? ret = rsa_verify_with_keynode(info, hash, sig, sig_len, node);
=> "keyname" provide the name of the key provider (pem file or key prefix for engine ) in keydir directory
lib/rsa/rsa-sign.c:537:??? ret = rsa_get_priv_key(info->keydir, info->keyname, e, &rsa);
lib/rsa/rsa-sign.c:702:??? ret = rsa_get_pub_key(info->keydir, info->keyname, e, &rsa);
dir and namle are used to get crt or key file.
=> rsa_pem_get_priv_key? = "%s/%s.key", keydir, name
=> rsa_pem_get_pub_key = "%s/%s.crt", keydir, name
so today the RSA features seens more compete based on openssl (with ENGINE and pkcs11 support for exmaple)
and keydir / keyname seens clear enought...
PS: I think the engine part could be shared between RSA and EDCSA part.
Quoted 1 line
There is an intentional side-effect. The RSA path takes > 'key-name-hint' to decide which key file to read from disk. In the >
context of "which fdt node describes my signing key", this makes > sense. On the other hand, 'key-name-hint' is also used as the > basename of where the key is on the filesystem. This leads to some > funny search paths, such as > > "some/dir/(null).key" So I am using the -K option to mkimage as the > _full_ path to the key file. It doesn't have to be named .key, it > doesn't have to be named .crt, and it doesn't have to exist in a > particular directory (as is the case for the RSA path). I understand > and recognize that this discrepancy must be resolved, but resolving > it right now would make the initial implementation much harder and > longer. marex at denx.de
This new option -K with full path will be permanent added to mkimage
or it is a temporarily solution (for initial ECDSA implementation).
If it is permanent it should be also supported in RSA mode ?
??? => for example: -K allow to override the "key-name-hint" value
Why you can't easily could use the existing -k option and "key-name-hint" to keep the current RSA behavior for EDCSA (even without ENGINE support)
For me, EDCSA part can rely on "key-name-hint" to chose the expected key file to read the pem from disk
??? info->keydir / params->keydir (from option -k)
??? info->keyname
perhaps : edcsa_get_priv_key()? => "%s/%s.pem", keydir, keyname
in prepare_ctx() function
and in the test:
??? keyname = "ecdsa-test-key"
??? keydir =? "${tempdir}"
or I miss something.....
Quoted 1 line
key_file> > # Testing > >test/py/tests/test_fit_ecdsa.pyis implemented withe
the goal to > check that the signing is done correctly, and that the signature is > encoded in the proper raw format. Verification is done with > pyCryptodomex, so this test will catch both coding errors and openssl > bugs. This is the only scope of testing proposed here. > > > # Things not yet resolved: - is mkimage '-k' supposed to be a > directory or file path I'm hoping I can postpone answering this > question pending further discussion.
Sorry to open debate.
I propose to change the test with previous proposal.
mkimage -F /local/home/frq07632/views/u-boot/build-sandbox/test.fit -k /local/home/frq07632/views/u-boot/build-sandbox/ecdsa-test-key.pem
=> -k use directory as RSA
mkimage -F /local/home/frq07632/views/u-boot/build-sandbox/test.fit -k /local/home/frq07632/views/u-boot/build-sandbox/
with test/py/tests/vboot/sign-images-sha256.its
?? ?fdt at 1 {
....
?? ??? ?signature at 1 {
??????????????????????????????? algo = "sha1,rsa2048";
-??????????????????????????????? key-name-hint = "dev"; +?????????????????????????????? key-name-hint = "ecdsa-test-key.pem";
??????????????????????? };
??????????????? };
Or change? key_file = f'{tempdir}/ecdsa-test-key.pem'
to '{tempdir}/dev.pem' file (as:? key-name-hint = "dev";)
Regards.
Patrick
v4 cover Add support for ECDSA image signing (with test)
Quoted 1 line · Patrick DELAUNAY
Hi Alexandru,
Hi Patrick
Quoted 4 lines
I found in
doc/uImage.FIT/signature.txtthe description??? - key-name-hint: Name of key to use for signing. The keys will
normally be in a single directory (parameter -k to mkimage).
[snip]
You are correct that the ECDSA path does not use the "key-name-hint". And this deviates from what RSA does. This is intentional.
[snip]> so today the RSA features seens more compete based on openssl (with
Quoted 3 lines
ENGINE and pkcs11 support for exmaple)
and keydir / keyname seens clear enought...
The the common case with image signing is that one wants to sign an image with a key. keyname comes from the console, and keydir comes from the FIT image, which contradicts this simplicity.
Another issue is incorporating this into a bigger build system like yocto. Now mkimage would impose a specific directory structure for signing keys. This would not be ideal.
Quoted 1 line
PS: I think the engine part could be shared between RSA and EDCSA part.
I don't see the benefit of using the engine, and it seems highly libcrypto specific. It would be a lot more code, with no useful functionality -- we can ecdsa sign with the simpler code.
[snip]
Quoted 8 lines
This new option -K with full path will be permanent added to mkimage
or it is a temporarily solution (for initial ECDSA implementation).
If it is permanent it should be also supported in RSA mode ?
??? => for example: -K allow to override the "key-name-hint" value
Yes and no. It is temporary in that we'd like to update the RSA path to be consistent with the ECDSA path. It's permanent in that we want to have the -'k' option to specify the key filename instead of the key dir. At least that's my understanding given the previous discussion with Simon.
[snip]
Quoted 3 lines
Sorry to open debate.
I propose to change the test with previous proposal.
[snip]
Quoted 7 lines
with test/py/tests/vboot/sign-images-sha256.its
?? ?fdt at 1 {
....
?? ??? ?signature at 1 {
??????????????????????????????? algo = "sha1,rsa2048";
-??????????????????????????????? key-name-hint = "dev"; +?????????????????????????????? key-name-hint = "ecdsa-test-key.pem";
This would go against us wanting to decouple the key filename from the key name. Consider haing several keys:
dev-key-frobnozzle.pem prov-key-frobnozzle-r1.pem prov-key-frobnozzle-r2.pem prov-key-frobnozzle-r3-after-hack-mitigation.pem
One might not want to expose those key names in the .its. The issue is, when the fit-image is created, the key filename must be known. But when the signing happens on a separate machine, the filename really isn't known.
So we should really use the "key-name-hint" as a hint rather than a filename or part of a filename.
Alex
v4 cover Add support for ECDSA image signing (with test)
Hi Alex,
Quoted 5 lines · Alex G.
Quoted 1 line · Patrick DELAUNAY
Hi Alexandru,
Hi Patrick
Quoted 9 lines
Quoted 4 lines
I found in
doc/uImage.FIT/signature.txtthe description- key-name-hint: Name of key to use for signing. The keys will
normally be in a single directory (parameter -k to mkimage).[snip]
You are correct that the ECDSA path does not use the "key-name-hint". And this deviates from what RSA does. This is intentional.
We need to join these two up though. We can't add a new way of doing things whenever we add a new algorithm.
Quoted 6 lines
[snip]> so today the RSA features seens more compete based on openssl (with
Quoted 3 lines
ENGINE and pkcs11 support for exmaple)
and keydir / keyname seens clear enought...
Quoted 7 lines
The the common case with image signing is that one wants to sign an image with a key. keyname comes from the console, and keydir comes from the FIT image, which contradicts this simplicity.
Another issue is incorporating this into a bigger build system like yocto. Now mkimage would impose a specific directory structure for signing keys. This would not be ideal.
It more likely requires a standard filename for keys...I wonder how this works in yocto at present?
Quoted 6 lines
Quoted 2 lines
PS: I think the engine part could be shared between RSA and EDCSA part.
I don't see the benefit of using the engine, and it seems highly libcrypto specific. It would be a lot more code, with no useful functionality -- we can ecdsa sign with the simpler code.
The hope was that the same API could be used...is that possible?
Quoted 7 lines
[snip]
Quoted 5 lines
This new option -K with full path will be permanent added to mkimage
or it is a temporarily solution (for initial ECDSA implementation).
Quoted 8 lines
Quoted 3 lines
If it is permanent it should be also supported in RSA mode ?
=> for example: -K allow to override the "key-name-hint" value
Yes and no. It is temporary in that we'd like to update the RSA path to be consistent with the ECDSA path. It's permanent in that we want to have the -'k' option to specify the key filename instead of the key dir. At least that's my understanding given the previous discussion with Simon.
But we do need to either move RSA over (in the same patch series ) or use a different flag. I'm leaning towards a different flag, since if we are not changing RSA, it is just going to get confusing.
Quoted 23 lines
[snip]
Quoted 3 lines
Sorry to open debate.
I propose to change the test with previous proposal.[snip]
Review
Quoted 7 lines
with test/py/tests/vboot/sign-images-sha256.its
fdt at 1 { .... signature at 1 { algo = "sha1,rsa2048";(quoted diff) +1
-1- key-name-hint = "dev"; + key-name-hint = "ecdsa-test-key.pem";This would go against us wanting to decouple the key filename from the key name. Consider haing several keys:
dev-key-frobnozzle.pem prov-key-frobnozzle-r1.pem prov-key-frobnozzle-r2.pem prov-key-frobnozzle-r3-after-hack-mitigation.pem
Quoted 6 lines
One might not want to expose those key names in the .its. The issue is, when the fit-image is created, the key filename must be known. But when the signing happens on a separate machine, the filename really isn't known.
So we should really use the "key-name-hint" as a hint rather than a filename or part of a filename.
Yes it is just a hint. We must not rely on it. Its only purpose is to hopefully speed up verification since fewer signatures might need to be tried.
Regards,
Simon