> -----Original Message-----
> From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On 
> Behalf Of Eric Rescorla
> Sent: Tuesday, April 08, 2008 10:28 PM
> To: Dan Wing
> Cc: [email protected]
> Subject: Re: [Sip] E.164 - who owns it
> 
> At Tue, 8 Apr 2008 17:01:40 -0700,
> Dan Wing wrote:
> > > If you mean "the E.164 number +123456789", say <tel:+123456789>.
> > >
> > > I don't see anything that will go wrong if people pay attention 
> > > to what URIs are defined to mean.
> > >
> > > Now there is a problem with signing tel: URIs, how do you 
> determine
> > > that the signer has the authority to sign a particular 
> tel: URI, but
> > > that's a delegation-of-authority problem that is not inherently
> > > different from other delegation-of-authority problems -- 
> you need a
> > > publicly-known root authority and a tree of delegations.
> > 
> > That is one approach to determining who owns something.
> > 
> > I have a different approach, which I derived from how we 
> determine if 
> > someone owns an IP address when we receive a TCP SYN that is 
> > proportedly from that IP address:  challange them to prove that 
> > packets route the other way.  Afterall, the Internet does not have
> > a central authority for who owns an IP address (or else we could
> > secure BGP routing better than we secure it today).
> >
> > In SIP, you accomplish something similar to the TCP SYN cookie by
> > sending a SIP request towards the E.164, asking them to sign the 
> > request and return it to you.  Details are in 
> > draft-wing-sip-e164-rrc-01.txt and comments are welcome.
> 
> While withholding judgement on draft-wing-sip-e164-rrc-01, I don't
> think this is that accurate a description of what can be inferred
> from the fact that a peer can complete a TCP connection initiated with
> a given source IP address.
> 
> As you say, in order to allow a TCP connection to be established,
> the passive endpoint requires the active endpoint to prove 
> (for some value of prove) that it received the passive endpoint's
> SYN/ACK by ACKing the passive endpoint's ISN. This arguably serves
> at least two security purposes:
> 
> - Prevention of DoS: TCP connections consume resources on the
>   client. Forcing the client to ACK the ISN prevents blind
>   resource consumption attacks on the passive endpoint.
>   (This can also be combined with a stateless cookie technique,
>   aka SYN cookies) to produce a system where no state is consumed
>   on the passive endpoint until the liveness check is passed.
> 
> - Endpoint authentication: the ability of the active endpoint
>   to ACK the ISN demonstrates that the active endpoint is on-path
>   from the passive endpoint to the claimed initiating source
>   IP address.
> 
> This second guarantee is very weak for three reasons:
> 
> (1) Many TCP stacks do not use sufficiently random PRNGs to
>     generate the ISN, in which case sequence number prediction
>     attacks are a real problem.
> (2) Even if the ISN is randomly generated, there are only
>     32 bits of ISN, so it's at least theoretically possible
>     to guess it in a blind attack (though of course very
>     expensive).

Fortunately SIP does not suffer from a 32-bit number space, so
(1) and (2) are not a limitation with draft-wing-sip-e164-rrc.

> (3) Even if we ignore the previous two considerations, 
>     all that the ability to ACK the ISN indicates is that
>     the attacker is on-path. There are many situations
>     in which an attacker is on-path (wireless being a 
>     commonly cited case).
> 
> It's precisely for these reasons that one cannot use source IP
> authentication for TCP connections on the public Internet, that
> mechanisms such as .rhosts are deprecated, and that cryptographic
> mechanisms such as IPsec were designed.
>
> Incidentally, this third deficiency seems to me to be shared
> by draft-wing-sip-e164-rrc. Any attacker who is on path 
> between the relying party and the target of the return
> routability check (either because they are on-path in
> the SIP routing fabric or because they are able to
> observe one of the links) will be able to pass the return
> routability check.

That is correct.

A signature generated by the originating domain (e.g., RFC4474)
would protect from such an on-path attacker -- unless of course 
the on-path attacker is also generated the original message 
and is also on-path for the return routability check.  

But, then, if the attacker is on-path for the return 
routability check, they already 'own' that E.164.  Ignoring
draft-wing-sip-e164-rrc for a minute, any normal INVITE 
for that E.164 will go to that very same attacker.

-d

_______________________________________________
Sip mailing list  https://www.ietf.org/mailman/listinfo/sip
This list is for NEW development of the core SIP Protocol
Use [EMAIL PROTECTED] for questions on current sip
Use [EMAIL PROTECTED] for new developments on the application of sip

Reply via email to