about DNS RRL

Phil Mayers p.mayers at imperial.ac.uk
Wed Oct 17 08:34:26 UTC 2012


On 10/17/2012 09:17 AM, pangj wrote:
> I have read the document of redbarn RRL for BIND and this NSD RRL:
> https://www.nlnetlabs.nl/blog/2012/10/11/nsd-ratelimit/
>
> I have a question that, since the DDoS to DNS are coming from spoofed
> IPs. But RRL is working based on source IP. So how can it stop the real
> life attack?

It doesn't stop it (indeed, can't). It mitigates the impact.

The DDoS tend to come from a fixed set of spoofed source at any one 
time. RRL helps, in that it:

  1. punts early in the path, lowering resolver CPU use, and
  2. returns a minimal response, which prevents amplification.

Remember the DDoS is actually directed at the spoofed source, not the 
DNS server. The DNS server is merely an unwilling participant. RRL helps 
prevent that participation.

There is, as I understand it, some spotty evidence that the attackers 
will move to a different server if RRL seems to be in use. How this 
happens I don't know - maybe they probe with real IPs? - but I've heard 
others emphatically claim this is not the case, and attackers will 
continue to blindly flail at you until the attacking node goes down.

The only solution to these kinds of attacks is for providers to 
implement BCP 38, and for upstream providers to start de-peering 
providers who don't. I rate this about as likely as... a very unlikely 
thing.

S/RTBH can help the DNS provider, if they're being overwhelmed and their 
upstream supports it.



More information about the bind-users mailing list