Certificate selection for servers is missing

[TechnikEmpire]

SSL_CTX_set_tlsext_servername_callback is not implemented.

To write a modern TLS enabled server, this callback is required. Why? Because a modern server should be capable of being initialized with a default SSL_CTX that implements the SNI callback and, if executed (when SNI is present in the client hello), you should be able to fetch or generate a SSL_CTX specifically for that host (fetched with SSL_get_servername(ssl, TLSEXT_NAMETYPE_host_name)) and then also dynamically change the context in use for that connected client SSL object via SSL_set_SSL_CTX() before returning from the callback.

Without this, you can only write servers like its 2002 (SNI was introduced in 2003).

I'll see if I can do a PR but it may not come soon. It should be trivial to implement for someone who already knows their way around this library.

[Dragollla]

It is possible to use SNIHandler and insert NIOSSLServerHandler with required context (certificate)

[Lukasa]

It is indeed. However, we should have a plan to supplement this, as the SNIHandler is not capable of handling an encrypted ClientHello, and doesn’t work with QUIC.

[TechnikEmpire]

Just a quick FYI, the working group responsible has paused ECH draft after a heated argument with me where I explained to some of the members how stupid it was on openssl ticket. Whether it is moving forward will be discussed next month at their meeting.

If anyone cares, the stupidity of it is that any executable code will be able to establish fully opaque encrypted connections from start to finish with even the host device admin / ring 0 not being capable of knowing what is in it.

DoH to resolve endpoint -> opaque
ECH -> opaque
DoH to do dns loop jumping to complete ECH -> opaque

There's a billion dollar blackhat industry drooling over themselves waiting for a free license to do whatever they'd like over your home and corporate network. Anyway just fyi, it's not certain whether ECH will progress.

[TechnikEmpire]

To be clear, its not stupidity if your goal is to transfer privacy to megacorps who then will have an absolute monopoly on the cybersec market, which the rep from akamai admitted was a factor in their motivations. Comments were deleted after but I slapped them up on archive.org first. Anyway.