Re: [rtcweb] state of libsrtp maintenance? (Re: SRTP not mandatory-to-use)

[Eric Rescorla <[e-mail]>]

On Thu, Jan 12, 2012 at 3:56 AM, Roman Shpount <[e-mail]> wrote:

> libsrtp comes with its own crypto library (which is considerably worse
> maintenance state wise and performance wise then OpenSSH or GnuTLS). It also
> implements extensive self-checks as a part of library initialization, which
> are strictly speaking not necessary. For some reason libsrtp implements two
> replay libraries (one for RTP and another for RTCP) even  though they
> suppose to do the same thing. It also implements a number of crypto routines
> that are not normally used for anything when implementing SRTP.
>
> My point is, if all is needed is an implementation of SRTP-SDES, the state
> of libsrtp is irrelevant since it is easy enough to code around. If
> SRTP-DTLS is needed, then couple of order of magnitude more code is needed
> to implement TLS handshake, which will mean using an existing mature DTLS
> implementation and adding SRTP encoding if necessary.

DTLS-SRTP was specifically designed so that one could put together a DTLS
stack and an SRTP stack with minimal modifications to both (and no necessary
modifications to the SRTP stack). In the case of OpenSSL and libsrtp, you
do the OpenSSL handshake, then use a new interface to export the keys
which you then push onto libsrtp using existing interfaces.

-Ekr