Keep track of the J-PAKE internal state in a more intuitive way.
Specifically, replace the current state with a struct of 5 fields:
* The round of J-PAKE we are currently in, FIRST or SECOND
* The 'mode' we are currently working in, INPUT or OUTPUT
* The number of inputs so far this round
* The number of outputs so far this round
* The PAKE step we are expecting, KEY_SHARE, ZK_PUBLIC or ZK_PROOF
This should improve the readability of the state-transformation code.
Signed-off-by: David Horstmann <david.horstmann@arm.com>
Given the size of ciL is set dependant on MBEDTLS_HAVE_INT32 /
MBEDTLS_HAVE_INT64, clang rightfully reports this as unreachable code in
32 bit builds. Fix this by using #define guards instead.
Signed-off-by: Paul Elliott <paul.elliott@arm.com>
This allows also to:
- removing the dependency on ECP_C for these functions and only rely
on PSA symbols
- removing extra header inclusing from crypto_extra.h
- return MBEDTLS_PK_USE_PSA_EC_DATA and MBEDTLS_PK_HAVE_ECC_KEYS to
their original position in pk.h
Signed-off-by: Valerio Setti <valerio.setti@nordicsemi.no>
This includes also:
- auto enabling ECP_LIGHT when MBEDTLS_PK_PARSE_EC_COMPRESSED is
defined
- replacing ECP_LIGHT guards with PK_PARSE_EC_COMPRESSED in pkparse
- disabling PK_PARSE_EC_COMPRESSED in tests with accelarated EC curves
(it get disabled also in the reference components because we want
to achieve test parity)
- remove skipped checks in analyze_outcomes.py
Signed-off-by: Valerio Setti <valerio.setti@nordicsemi.no>
When MBEDTLS_PK_USE_PSA_EC_DATA is defined, opaque and non-opaque keys
are basically stored in the same way (only a diffferent ownership for
the key itself), so they should be treated similarly in the code.
Signed-off-by: Valerio Setti <valerio.setti@nordicsemi.no>
This is an internal detail of the ECC arithmetic implementation, only
exposed for the sake of the unit tests
Mbed TLS 3.4.0 was released with the type mbedtls_ecp_modulus_type defined
in a public header, but without Doxygen documentation, and without any
public function or data structure using it. So removing it is not an API
break.
Signed-off-by: Gilles Peskine <Gilles.Peskine@arm.com>
This reverts commit f1e396c42724896b9d31ac727043da45a35d5e26.
Performance is slightly better with this reverted, especially
for AES-CBC 192.
Signed-off-by: Dave Rodgman <dave.rodgman@arm.com>
Attempt to fix failing test by dealing with overflow with three rounds,
instead of previous subtract modulus solution. Also optimise out shifts
by using memcpy / memmove instead. Remove final sub to return canonical
result, as this is not required here.
Signed-off-by: Paul Elliott <paul.elliott@arm.com>
