[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: POP handling commands given in wrong state
--On July 29, 2011 7:04:36 +0300 Mykyta Yevstifeyev <evnikita2@xxxxxxxxx>
I would really be happy if existing POP-over-TLS implementations adhered
usual POP behavior as defined in RFC 1939, and I would be happy to
describe it in the corresponding document. However, if we want to define
the current practices, we should document the technology as-is. If we
want to give POP-over-TLs a standard definition of operations, I don't
really think those implementation which used the discussed POP-over-TLS
algorithm will break their behavior.
We have a choice to define POPS-with-client-certs based on what has been
deployed or define it based on how we think it should work. The latter is
an architecturally cleaner choice, but is unlikely to cause the deployed
implementations with the former behavior to change.
The "discouragement" of RFC 2595 actually has no considerable reasons.
I disagree. I happen to support documenting POPS because I believe people
will continue to use it as they do today regardless of what anyone thinks
should be done, and I'd like our documents to give accurate guidance for
interoperability as long as it's possible and what's deployed is not
egregiously broken. I consider POP+STLS a superior architecture that is
cleaner are more interoperable and consider it preferable to pops in almost
all cases. I believe section 7 of RFC 2595 remains largely correct today,
but accept those arguments are not sufficiently compelling to prevent
deployment of pops, so I'd prefer to document pops for interoperability
reasons despite its flawed architecture.
Those reasons mentioned in Section 7 of RFC 2595 concern that (1) ports
are limited resource, (2) new URI scheme needed, (3) there might be
confusion of users, (4) choose "to use or not to use POP-over-TLS" is
worse than "use TLS when possible, and negotiate it with STLS".
Regarding these points, I should say: (1) isn't a problem now; moreover,
995 port number is already assigned,
IPv4 address space is not a problem right now, either. But that does not
mean it will never be a problem and it's OK to unnecessarily waste that
resource. But port 995 is registered and won't go away, so I grant that
part of your point.
(2) new URI scheme isn't a problem, too; http/https schemes illustrate
this as well,
I disagree. Are you sure in all cases when you need to use http: and when
you need to use https:? Do you ever enter sensitive data into a web page
when "http:" is in the address bar? Are you sure that's safe? If you're an
ISP would you prefer to give your users just an email address and password
(RFC 6186), an email address and pop server (POP + STLS) or an email
address and pop URL (pop, pops). I'd say the latter is the least desirable
of these choices as it will result in the most transcription errors and
(3) currently, the users
don't usually set use of POP-over-TLS due to their willing, but rather
because this is how the user agent should be configured to match the
server, as server's administration points the user to do, and
It intrudes into the UI forcing end-users to make choices related to
security they may not understand. That problem can be worked around with an
extra level of complexity such as that provided by RFC 6186, but it is
still an architectural design error.
(4) is eliminated when the server (server's administration) clearly
POP-over-TLS should or should not be used. POP-over-TLS isn't worse that
I am not convinced. The correct model is for clients to be
secure-by-default without the need for any statement from server
administration or any configuration by client users. That is, the first
time I run a mail client against a new server it will "permanently latch
on" SSL/TLS for that account if it is available (in the event SSL/TLS is
not available, the client should confirm with the user that an insecure
connection to that server is acceptable). This can be done via RFC 6186 for
servers that offer both POP and POPS and can also be done for
POP-with-STLS. It's faster and easier to code in the latter case.