[Security] updated Jingle-XTLS proposal

Florian Zeitz florian.zeitz at gmx.de
Fri Feb 20 21:38:28 CST 2009


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Peter Saint-Andre schrieb:
> Peter Saint-Andre wrote:
>> Peter Saint-Andre wrote:
>>> Dirk Meyer has updated the Jingle-XTLS proposal:
>>>
>>> http://xmpp.org/extensions/inbox/jingle-xtls.html
>> Dirk and I have been communicating off-list about this and we will
>> provide an updated version before long -- mostly to simplify the
>> protocol, clarify the error cases, and generalize the spec so that it
>> can be used with both streaming transports (TLS) and datagram transports
>> (DTLS).
> 
> We've updated the spec to version 0.0.3. Dirk and I will probably work
> on it some more over the weekend. :)
> 
> Peter
> 
Some comments you two might take into account:

Up first the encouraging comment:
This is a really well written XEP IMHO. The reasoning for the design
seems quite clear and it is not to lengthy, but still seems to explain
everything necessary (obviously ignoring the parts that are said to be
missing). Keep it that way ;)

Criticism and nitpicks:
* Example 3 should probably have action='session-accept'
* After "The following rules apply to the initiator's handling of the
session-accept message:" only the 2. case where the certificate could
not be verified is said to require user interaction. I'd personally also
want to be asked what to do if encryption wasn't possible (1. case).
* As Non-Human Parties may also be (web-)services. Maybe add encrypted
E-mail to section 5.1. E.g. Launchpad knows your GPG-key, so they could
in theory send you encrypted mail with a PIN.
Or/and possibly something more general along the lines of: "If possible
any out-of-band method a human could use to convey the PIN is
practicable too" E.g. a Asterisk PBX may call you and 'read' a PIN to
you (whether sth. like this would be secure depends on the type of
telephony and suspected MITM attack of course, but that's a different topic)
* Example 10 might need some ellipsis. XTLS being the only feature seems
unlikely.
* Possibly add some notes about bot2bot verification of certificates
(using a CA I'd suspect)
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)

iEYEARECAAYFAkmfdy4ACgkQ0JXcdjR+9YQ9uACffp7aWcK6rSz9s2CiCvKW7PYX
1c4An1IKXjIv/pzJr7Rnxb+8kYXv+qH8
=1QhP
-----END PGP SIGNATURE-----


More information about the Security mailing list