Re: [kitten] [saag] SSH Protocol Extensions

[Nico Williams <[e-mail]>]

On Wed, Aug 12, 2015 at 07:54:37PM +0200, Simon Josefsson wrote:
> Or look into already published RFC 6616 for OpenID in SASL or RFC 6595
> on SAML in SASL.  SAML-EC avoids the web loop, and may indeed be more
> relevant depending on use-case.

If Phil needs a SAML, one-message authentiation method with no proof of
possession, then a new SSHv2 userauth method is probably best (since the
GSS userauth method uses MIC tokens for channel binding).  Would the
SAML token be possible to bind to an SSHv2 session?  I would hope so.

> I believe the problem with these mechanisms has been:
> 
> 1) They are stateless -- if you need to open many connections, or open
> them often, you need one OpenID/SAML authentication per connection which
> results in poor user experience.

Probably not an issue for SSHv2.

> 2) They are tied to a particular federation system.  Sometimes this is
> not a problem, but I imagine sometimes the client/server does not know
> what system is the most appriate to use until a web loop has been
> invoked.
> 
> The OAuth SASL mechanism (long delayed, still being worked on) may also
> be relevant to look into, but it does not support GSS-API so SSH support
> is not a given.

See above comment about MIC tokens and channel binding.  Also, without a
server-side proof of possession the channel binding can only protect
against token reuse, and the user will still depend on the use of server
host public keys for server authentication (which may not be a problem).

Nico
--