mbedtls
mbedtls copied to clipboard
Mbed-TLS
Implement mbedtls_mpi_core_fill_random
Addresses https://github.com/Mbed-TLS/mbedtls/issues/6239 (only fill_random in core, not yet random or mod).
Also in this PR: introduce a function mbedtls_test_read_mpi_core() to conveniently read a core MPI in unit tests.
Changelog: no (only core bignum stuff).
Backport: TODO (not the new bignum stuff, only the minor changes to existing test framework code and test data).
Rebased to resolve conflicts with Montgomery multiplication (only additions in the same place and changes on consecutive lines, no actual interference).
@gilles-peskine-arm this has conflicts and needs-work - do you know when you'll update it?
I'll update once https://github.com/Mbed-TLS/mbedtls/pull/6450 is merged.
Looks good to me with one minor question.
I am not quite certain of the interger range the generation is bound to. The assumption is that rng is returning an int, which can be 32 or 64 bits.
int (*f_rng)(void *, unsigned char *, size_t)
I can see that the upper bound is set by and N and used in calculating the bitsize:
size_t n_bits = mbedtls_mpi_core_bitlen( N, limbs );
Could we have situations where the rng is generating 32 bit numbers and the platform is 64 bit? Do we need to have test cases specific to different architectures?
I am not quite certain of the interger range the generation is bound to. The assumption is that rng is returning an int, which can be 32 or 64 bits.
f_rng returns a standard Mbed TLS return code, i.e. 0 for success else an error number. It just puts random bytes into the buffer it is given. Look in bignum_core.c:mbedtls_mpi_core_fill_random() to see where it is used.
There's a tech debt item to document f_rng better: https://github.com/Mbed-TLS/mbedtls/issues/5868