[lwip-devel] altcp_tls_mbedtls: MBEDTLS_ECP_RESTARTABLE

[Giuseppe Modugno]

I tried to enable MBEDTLS_ECP_RESTARTABLE option in config.h of mbedTLS. 
This option (and the use of mbedtls_ecp_set_max_ops()) limits the number 
of operations made during expensive ECC crypto functions. It's important 
in my NO_SYS application, because I need to serve other "tasks" in a 
resonable time and TLS handshake could take even 5 seconds!

However I think the layer altcp_tls_mbedtls isn't compatible with 
MBEDTLS_ECP_RESTARTABLE. Indeed mbedtls_ssl_handshake() could return 
MBEDTLS_ERR_SSL_CRYPTO_IN_PROGRESS, but this return value isn't managed 
in altcp_mbedtls_lower_recv_process().
Also mbedtls_ssl_read() and mbedtls_ssl_write() could return 
MBEDTLS_ECP_RESTARTABLE.

When one of the above functions return 
MBEDTLS_ERR_SSL_CRYPTO_IN_PROGRESS, the application should call the 
function again with the same input parameters.

I'd like to invest some time to add MBEDTLS_ECP_RESTARTABLE support in 
altcp_tls_mbedtls, but I need some help how to do. I think we need a 
crypto_in_progress flag in altcp_mbedtls_state_t that is set when 
MBEDTLS_ERR_SSL_CRYPTO_IN_PROGRESS is returned.
In poll callback we call again the function when the flag is set.

However we don't have a poll callbak during handshake: the final poll 
callback will be set after successfull handshake. Do I create and use a 
temporary poll callback to set in altcp_mbedtls_setup_callbacks()?

After handshake, mbedtls_ssl_read() and mbedtls_ssl_write() could return 
MBEDTLS_ERR_SSL_CRYPTO_IN_PROGRESS. So this situation must be faced too. 
Any help?