> On Mon, Jul 25, 2011 at 9:12 AM, Dino Farinacci <[email protected]> wrote:
>> For your reference to "transition state", there isn't this same issue on a 
>> PITR because it receives and forwards packets from non-LISP sites, where 
>> there is one IP header in the packet, and the source address is subject to 
>> uRPF and forwarding rules as it is today.
> 
> The reason I mention "transition state" is you seem to be gravitating
> towards ideas that require new inspection capabilities at
> substantially all "legacy Internet" access providers in order to
> protect the new LISP infrastructure.  If I can use any source IP
> address I want when sending crafted packets, including crafted LISP
> outer-header, to your LISP site, then the chances are good that I can
> disable your site as long as I can generate enough PPS and there are
> enough mappings in the system to produce cache churn in your xTR.
> 
> So relying on ITRs to prevent "bad traffic" from ever entering the
> system will not work because there will be a large, legacy Internet
> with no ability to prevent that traffic from reaching your site.

How many times do I have to say we should not solve this on the ETR?

>> Right, so you have some hard decisions to make. The ETR can do lookup in its 
>> cache to figure out if the (source-RLOC, source-EID) is in its cache and if 
>> not, drop the packet, then do a lookup so subsequent packets can have the 
>> cache populated so the check can pass.
> 
> Except the ETR is bounded by the same cache scaling limits as an ITR
> subject to a reflection attack, meaning its cache can churn, and
> substantially all traffic needing to be decapsulated by that ETR would
> be dropped.

Right, agree. Don't do this on the ETR. DO NOT SOLVE THIS ON THE ETR.

>> I call this the "needle in the haystack" problem. And we have had to deal 
>> with this with hardware packet punting for a number of decades.
> 
> Even if you imagine that the xTR control-plane is not a bottleneck,
> presuming that the data-plane can generate MS queries directly and can

Why do you want to presume that?

> update its FIB at an unlimited rate, the result of having a limited
> FIB size plus an attack designed to take advantage of that means the
> MS infrastructure will be required to have very low latency and very
> high throughput to keep up with an xTR experiencing high PPS and low
> cache hit rate because of churn.  This essentially would mean that the
> MS infrastructure is directly involved in a substantial fraction of
> packet forwarding decisions on that xTR.

That doesn't follow at all and I don't know why you want to make this 
conclusion.

Dino

> 
> --
> Jeff S Wheeler <[email protected]>
> Sr Network Operator  /  Innovative Network Concepts
> _______________________________________________
> lisp mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/lisp

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

Reply via email to