Re: [pcp] Comparison of PCP authentication

["Dan Wing" <[e-mail]>]

> -----Original Message-----
> From: Sam Hartman [mailto:[e-mail]]
> Sent: Thursday, August 09, 2012 6:25 AM
> To: Dan Wing
> Cc: 'Zhangdacheng (Dacheng)'; 'Margaret Wasserman'; [e-mail]
> Subject: Re: [pcp] Comparison of PCP authentication
> 
> >>>>> "Dan" == Dan Wing <[e-mail]> writes:
> 
> 
>     Dan> Doing PANA first is a good optimization only if the PCP client
>     Dan> knows that PCP authentication will be required by the PCP
>     Dan> server on that particular network.  If PCP authentication is
>     Dan> not needed on a particular network, requesting PCP
>     Dan> authentication incurs an extra round trip.
> 
> Thinking about this as an optimization seems wrong to me.
> 
> consider the case where the client wants authentication but the server
> does not require it.  For example, the client wants integrity
> protection
> of the result of the request.
> In this case it's about security requirements *not* optimization.
> The client needs a way to force the server to require authentication.

I am struggling to imagine a PCP client that would refuse to do MAP
or PEER because its PCP server doesn't support authentication.

> Another case where this becomes important is where a server will accept
> requests from unauthenticated clients but treat them differently. For
> example an authenticated client is allowed to ask for longer lifetimes,
> etc.

That sounds reasonable.  But we could solve that differently than trying
authentication first all the time, such as for example including an 
"I WILL DO AUTHENTICATION IF I GET BETTER CHOCOLATE COOKIES" in the 
initial PCP request, which triggers a PCP authentication required error.

> Especially since interacting with firewalls is in-scope for PCP,
> supporting clients with higher security requirements seems and
> important design goal for a PCP authentication mechanism.
> 
> So, I don't think we should be thinking of this purely in terms of
> optimization.

-d