[PATCH v2 09/28] mbedtls: add digest shim layer for MbedTLS

[Ilias Apalodimas]

Hi Raymond,


On Tue, 7 May 2024 at 20:55, Raymond Mao <raymond.mao at linaro.org> wrote:
>
> Implement digest shim layer on top of MbedTLS crypto library.
>
> Signed-off-by: Raymond Mao <raymond.mao at linaro.org>
> ---
> Changes in v2
> - Split the shim layer into separated files and use the original head
>   files instead of creating new ones.
>
>  lib/mbedtls/Makefile |   7 +++
>  lib/mbedtls/md5.c    |  68 ++++++++++++++++++++++++++
>  lib/mbedtls/sha1.c   | 111 +++++++++++++++++++++++++++++++++++++++++++
>  lib/mbedtls/sha256.c |  65 +++++++++++++++++++++++++
>  lib/mbedtls/sha512.c |  96 +++++++++++++++++++++++++++++++++++++
>  5 files changed, 347 insertions(+)
>  create mode 100644 lib/mbedtls/md5.c
>  create mode 100644 lib/mbedtls/sha1.c
>  create mode 100644 lib/mbedtls/sha256.c
>  create mode 100644 lib/mbedtls/sha512.c
>
> diff --git a/lib/mbedtls/Makefile b/lib/mbedtls/Makefile
> index 85f0a3cfd07..b8eda9638f4 100644
> --- a/lib/mbedtls/Makefile
> +++ b/lib/mbedtls/Makefile
> @@ -14,6 +14,13 @@ ccflags-y += \
>         -I$(src)/external/mbedtls/library \
>         # This line is intentionally left blank
>
> +# shim layer for hash
> +obj-$(CONFIG_MBEDTLS_LIB_CRYPTO) += hash_mbedtls.o
> +hash_mbedtls-$(CONFIG_$(SPL_)MD5) += md5.o
> +hash_mbedtls-$(CONFIG_$(SPL_)SHA1) += sha1.o
> +hash_mbedtls-$(CONFIG_$(SPL_)SHA256) += sha256.o
> +hash_mbedtls-$(CONFIG_$(SPL_)SHA512) += sha512.o
> +
>  obj-$(CONFIG_MBEDTLS_LIB_CRYPTO) += mbedtls_lib_crypto.o
>  mbedtls_lib_crypto-y := \
>         $(MBEDTLS_LIB_DIR)/aes.o \
> diff --git a/lib/mbedtls/md5.c b/lib/mbedtls/md5.c
> new file mode 100644
> index 00000000000..2488d4f5603
> --- /dev/null
> +++ b/lib/mbedtls/md5.c
> @@ -0,0 +1,68 @@
> +// SPDX-License-Identifier: GPL-2.0+
> +/*
> + * Hash shim layer on MbedTLS Crypto library
> + *
> + * Copyright (c) 2023 Linaro Limited
> + * Author: Raymond Mao <raymond.mao at linaro.org>
> + */
> +#include "compiler.h"
> +
> +#ifndef USE_HOSTCC
> +#include <watchdog.h>
> +#endif /* USE_HOSTCC */
> +#include <u-boot/md5.h>
> +
> +void MD5Init(MD5Context *ctx)
> +{
> +       mbedtls_md5_init(ctx);
> +       mbedtls_md5_starts(ctx);
> +}
> +
> +void MD5Update(MD5Context *ctx, unsigned char const *buf, unsigned int len)
> +{
> +       mbedtls_md5_update(ctx, buf, len);
> +}
> +
> +void MD5Final(unsigned char digest[16], MD5Context *ctx)
> +{
> +       mbedtls_md5_finish(ctx, digest);
> +       mbedtls_md5_free(ctx);
> +}
> +
> +void md5(unsigned char *input, int len, unsigned char output[16])
> +{
> +       MD5Context context;
> +
> +       MD5Init(&context);
> +       MD5Update(&context, input, len);
> +       MD5Final(output, &context);
> +}
> +
> +void md5_wd(const unsigned char *input, unsigned int len,
> +           unsigned char output[16], unsigned int chunk_sz)
> +{
> +       MD5Context context;
> +#if defined(CONFIG_HW_WATCHDOG) || defined(CONFIG_WATCHDOG)
> +       const unsigned char *end, *curr;
> +       int chunk;
> +#endif
> +
> +       MD5Init(&context);
> +
> +#if defined(CONFIG_HW_WATCHDOG) || defined(CONFIG_WATCHDOG)
> +       curr = input;
> +       end = input + len;
> +       while (curr < end) {
> +               chunk = end - curr;
> +               if (chunk > chunk_sz)
> +                       chunk = chunk_sz;
> +               MD5Update(&context, curr, chunk);
> +               curr += chunk;
> +               schedule();
> +       }
> +#else
> +       MD5Update(&context, input, len);
> +#endif
> +
> +       MD5Final(output, &context);
> +}

For all having functions you have a watchdog and non-watchdog variant.
Doing a quick grep, only board/gdsys/a38x/hre.c uses the non-watchdog
variant directly which seems wrong.
Instead of defining 2 functions why don't we define the watchdog one
only? We could then save enough space and make the gap smaller.



> +
> +void sha1_hmac(const unsigned char *key, int keylen,
> +              const unsigned char *input, unsigned int ilen,
> +              unsigned char *output)
> +{
> +       int i;
> +       sha1_context ctx;
> +       unsigned char k_ipad[64];
> +       unsigned char k_opad[64];
> +       unsigned char tmpbuf[20];
> +
> +       memset(k_ipad, 0x36, 64);
> +       memset(k_opad, 0x5C, 64);

0x36 and 0x5c need a define that explains the magic values.
Also, don't use lengths directly, memset(k_ipad, 0x36, sizeof(k_ipad))
is more readable.

> +
> +       for (i = 0; i < keylen; i++) {
> +               if (i >= 64)
> +                       break;

This check needs to be outside the loop and report an error in the
keylen is bigger than expected

> +
> +               k_ipad[i] ^= key[i];
> +               k_opad[i] ^= key[i];
> +       }
> +
> +       sha1_starts(&ctx);
> +       sha1_update(&ctx, k_ipad, 64);

Same here don't use lengths directly

> +       sha1_update(&ctx, input, ilen);
> +       sha1_finish(&ctx, tmpbuf);
> +
> +       sha1_starts(&ctx);
> +       sha1_update(&ctx, k_opad, 64);
> +       sha1_update(&ctx, tmpbuf, 20);
> +       sha1_finish(&ctx, output);
> +
> +       memset(k_ipad, 0, 64);
> +       memset(k_opad, 0, 64);
> +       memset(tmpbuf, 0, 20);

and here etc...

[...]

Thanks
/Ilias