Towards deprecating TLS 1.0 & 1.1

[emillon]

Yesterday, all major browser vendors released a plan to deprecate TLS 1.0 & TLS 1.1:

• Apple
• Google
• Microsoft
• Mozilla

The key points are I believe:

• TLS 1.0 usage is about 1%;
• TLS 1.1 is virtually inexistent as most hosts that support it also support TLS 1.2;
• these vendors plan to remove support for these protocol versions in the first half of 2020.

Since ocaml-tls is not in a position where supporting legacy protocols is required, I suggest that we deprecate TLS 1.0 and 1.1 and remove them at some point in 2019.

There is some ongoing work to add TLS 1.3 support, so it will probably make sense to wait until this is added before removing the old versions in order to avoid having a single supported version.

Thoughts on this?

[hannesm]

Thanks for opening this. Tentative plan would be:

• release version X (end of 2018) with TLS 1.3 support, removing 1.0 and 1.1 from default configuration
• release version X+1 (in 2019) where TLS 1.0 and 1.1 support has been dropped entirely

[cfcs]

I semi-frequently need to talk to servers that only support legacy protocols, SSL 3.0 and SSHv1 being particularly painful, so I'm a little bit sad about the prospect of support being dropped entirely, but I completely agree with removing them from the default configuration. I guess forking to a -legacy repo would also be an option for such real-world use cases.

[hannesm]

Forgot to comment here, but 0.12.1 (released June 12th 2020) raised the lowest support TLS version in the default configuration to 1.2. Since March 2021, RFC 8996 has been published which deprecates TLS 1.0 and 1.1.

[hannesm]

I don't see any point to act on this issue - the default configuration since years does not use 1.0 and 1.1 -- but I agree with @cfcs that it is nice to be able to talk earlier protocol versions (for old hardware/software) -- the incentive to remove this code is at the moment not there. This may change in the future, but the original report does not provide any details for a "why". Feel free to reopen if you have a convincing argument why the world would be better if we drop the protocol support (e.g. compilation times / resulting binary size / performance).