Andrew,
Response inline.
Sriram

> -----Original Message-----
> From: Andrew Lange [mailto:[email protected]]
> 
> Sriram,
> 
> Why would you accept a route de-aggregated by an upstream?  If signed 
> route-object says
> AS_A owns and announces only the route 1.2.3.0/20 and I'm seeing 1.2.3.0/21 
> from AS_B,
> the route filter should be configured not to accept more specifics.   If AS_A 
> wants to de-
> aggregate, it can split it's route-objects into the more specifics and 
> register them.
> 
> Andrew

I am sure you understand that ROA allows the prefix owner to 
register not only a prefix-origin pair (say, {1.2.0.0/20, AS_A})
but also a maxlength to allow any subprefixes (more specifics) 
that she may originate at times from the same AS_A.
So the registered ROA info may look like {1.2.0.0/20, AS_A, maxlength = 22}.
Then if AS_A needs to announce more specifics 1.2.0.0/21 and 1.2.8.0/21,
it can readily do so while depending on the same ROA for validation.
The reason for this may be traffic engineering, e.g., when AS_A 
is multihomed to two other upstream ASs, say, AS_X and AS_Y, then
AS_A can announce more specifics 1.2.0.0/21 and 1.2.8.0/21 to 
AS_X and AS_Y with some clever prepending so that inbound traffic 
is equally distributed over the links from AS_X and AS_Y to AS_A.
This is just one example of the usability of maxlength.

Now as one can see, there are meaningful uses of maxlength 
for the legitimate originator, but then that opens the door for
subprefix hijacks (with your proposed approach 
to securing partial path) as I explained before.
Such subprefix hijacking would be detectable in a different
approach, for example, if the whole update were signed including 
the NLRI (prefix) and AS_PATH 
(in addition to use of the normal ROA information).

Sriram     
   
> 
> On Feb 24, 2011, at 3:51 PM, Sriram, Kotikalapudi wrote:
> 
> > Andrew,
> >
> > Comment below.
> >
> > Sriram
> >
> >> -----Original Message-----
> >> From: [email protected] [mailto:[email protected]] On Behalf Of 
> >> Andrew
> Lange
> >> Sent: Wednesday, February 23, 2011 6:17 PM
> >> To: Sandra Murphy
> >> Cc: [email protected]
> >> Subject: Re: [sidr] SIDR ReCharter - to capture/cover path validation work
> >>
> > --- snip ---
> >>
> >> From a work item perspective, it would be useful to have a relationship 
> >> object signed
> that
> >> says "I'm AS_A, and I have AS_B and AS_Q as legitimate connections."  This
> information,
> >> combined with AS_A's signed route-objects means that downstream ASes would 
> >> be able
> to
> >> verify that a route 1.2.0.0/16 coming in with an AS PATH of [AS_A, AS_B] 
> >> or [AS_A,
> >> AS_Q] would be AS PATHs that could be authorized.  Effectively solving the 
> >> Origin +
> >> Upstream issue would solve most of the problems.  Yes, someone could then 
> >> spoof an
> >> announcement of origin+upstream+self, but then we're looking at a minimum 
> >> of a 3-AS
> >> path and the likelihood of that being best drops.  We can also do this 
> >> with a simple
> route-
> >> filter and does not require on-system crypto calculations.
> >>
> >> Andrew
> >>
> > Yes, the path length goes up by 2.
> > But normally an adversary would spoof an announcement with a subprefix (more
> specific)
> > of the prefix that the legitimate prefix owner has announced
> > (while still trying to keep within the maxlength as specified in the ROA).
> > I that case the increased path length would not matter.
> > The spoofed announcement would be accepted and propagated widely.
> >
> > Sriram
> >

_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr

Reply via email to