Add a start_tls option to specify the minimum SSL version required

[fabiokung]

These patches allow EM code to disable SSLv2, or force TLSv1 only.

Usage:

module Handler
  def post_init
    start_tls(:min_version => :sslv3)
  end
end

:min_version is case insensitive and can be :sslv2 (any, default), :sslv3 (disables SSLv2), or :tlsv1 (TLSv1 only).

This is backwards compatible and the default is the same as before (:min_version => :SSLv2). @tmaher can fill us in with reasons why someone would want to completely disable SSLv2 and/or SSLv3 nowadays.

[ibc]

This is backwards compatible and the default is still SSLv23. @tmaher can fill us in with reasons why someone would want to completely disable SSLv2 nowadays.

Because not all in the world is HTTP and there are other protocols (i.e.
SIP) in which just TLS is allowed and SSL is not.

[fabiokung]

I completely reworked these patches to allow a SSL :min_version, instead of only forcing SSLv3. That way, eventmachine client and/or server code can specify what is the minimum ssl version required: :sslv2 (any), :sslv3 (no SSLv2) or :tlsv1 (TLSv1 only).

I also included some tests using a client other than EM forcing specific protocol versions.

/cc @tmaher

[elijh]

+1

This would be a very welcome addition to eventmachine. It is certainly best practice to disable SSLv2 whenever possible. I don't know if it would help this pull request get approved or not, but it would also be cool to add cipher selection. The hard coded cipher list in ssl.cpp is pretty horrible:

SSL_CTX_set_cipher_list (pCtx, "ALL:!ADH:!LOW:!EXP:!DES-CBC3-SHA:@STRENGTH");

[sodabrew]

I don't think a min version is the right way to go. It's entirely possible that a bug would be discovered in a "middle version" -- for example, TLSv1.1 could become compromised while TLSv1 and TLSv1.2 are safe. We'll need a way to handle that scenario.

I suggest a version string that allows us to whitelist / blacklist a set of protocols. See #359 (comment)

[jonbrenner]

I agree with @sodabrew. In addition to specifying specific versions due to security concerns, there are interoperability reasons for specifying specific versions. E.g., embedded systems with poor implementations of specific protocol versions.

[sodabrew]

This is cherry-picked and further developed in #654