Re: Security concerns around co-locating TLS and non-secure on same port (WGLC: draft-ietf-tsvwg-iana-ports-08)

[Geoffrey Keating <[e-mail]>]

Magnus Westerlund <[e-mail]> writes:

> https://datatracker.ietf.org/doc/draft-ietf-tsvwg-iana-ports/
..
> My summary of that comment is that STARTTLS for SMTP (RFC 3207) has
> shown to have some security issues, be complexer to implement than using
> two ports and thus less popular. Thus the registration rules should be
> less restrictive in assigning an additional port for TLS version of
> services/applications/protocols.
...
> Clearly if the security issues are serious when one multiplex TLS and
> non-secured version of the protocol on the same port we must allow such
> port allocations. However if the issues are minor and the primarily
> issue is implementation complexity then saving the limited port space is
> probably more important.
> 
> Your input into these questions would be very appreciated.

There are also security issues when two ports are used: how can an
application choose which port to communicate over?  Paul mentions the
list of security issues in RFC 3207, but each of these corresponds to
a similar issue in a two-port solution.  Eventually the RFC (and
similar specifications of other protocols) comes down to

"The decision about whether acceptable authentication or privacy was
   achieved is made locally, is implementation-dependent, and is
   beyond the scope of this document."

and in practise this leads to at least one, and sometimes more, of these:

1. Implementations which do not interoperate, because one requires
   security and the other does not implement it,

2. Automatically falling back to insecure communications when secure
   communications cannot be achieved, even if this is due to an
   attacker's denial-of-service, or

3. Asking the user questions that the user does not understand or for
   which the user cannot know the answer ("Is example.com supposed to
   have secure email?").

So, I think there are very limited additional concerns for a
single-port approach over a two-port approach, but there are
significant problems with both, and if implementation complexity and
interoperability is a concern it could and probably should be
addressed by specifying that TLS is always used.