[dns-operations] DNS deluge for x.p.ctrc.cc
ge at linuxbox.org
Mon Feb 27 22:45:10 UTC 2006
Joe Greco wrote:
>>On Feb 27, 2006, at 2:15 PM, Joe Greco wrote:
>>>If shunning would be effective, wouldn't it make more sense to shun
>>>networks that don't implement BCP38? We could fix a wide *range* of
>>>future attack vectors, rather than just this relatively small single
>>>vector that doesn't even address all of the ways to abuse DNS for this
>>>sort of thing.
>>i agree that "fixing" via filtering would solve many problems at
>>once, and would fix this particular issue with amplification but as
>>paul noted (and this has been my and many others' experience as well)
>>getting providers (or enterprise networks, in my own experience) to
>>*do* it is very, very hard. they don't have financial incentive to
>>do so and sometimes negative financial incentive (no staff or
>>expertise to deal with it).
>>as rodney/rob pointed out, working from the other end with the
>>providers that have open/recursive servers that are used in
>>amplification attacks (and therefore impacting them financially)
>>yields fairly good results.
>>i don't disagree with much of what you've said, but aside from the
>>more difficult problem of getting bcp38 implemented, you're not
>>proposing a workable solution either.
> How do we know that one problem is "more difficult" than another?
> Why is shunning going to work to implement one solution but not another?
> What happens if you start shunning networks who haven't implemented BCP38?
> After all, if you're willing to take the collateral damage associated with
> shunning open recursers, why not just go right for the jugular and just
> start shunning anyone on a network without BCP38?
> Why bother trying to piddle around with one little problem that'll be
> replaced with another technique in *MINUTES*? I came up with another
> one that was much more exciting in less than 30 seconds of thought, some
> folks here will have seen the e-mail.
> Open recursers are a problem in an environment lacking BCP38.
> But many things are a problem in an environment lacking BCP38.
> Removing open recursers from a non-BCP38 environment fixes one attack
> vector, but also breaks useful things.
> Having a BCP38 environment fixes many attack vectors, and breaks nothing.
> So. Does it make more sense to fix the open recursers or fix the BCP38
> I am just amazed that any significant number of clueful people would
> think that it is easier to fix a ton of open recursers... these things
> are hiding all around the globe, unknown, unloved, in closets and attics,
> big and little, Linux boxes deployed by that geek college student we had
> one summer back in 2000.
> Let me make a prediction. It's one that I hate to make.
> UltraDNS starts shunning open recursers. Instantly, 100,000 recursers
> around the globe stop resolving 20% of the Internet.
> What happens next?
> It's not what you think.
> You have to remember that there's a ton of Internet stuff out there that
> was set up and has been left to run. The people who set it up have long
> ago moved on, but stuff kept churning, and of course momentum being what
> it is, it kept getting used.
> So one day, ABC Co. finds that they can't resolve 20% of the Internet.
> So they call up XYZ Internet, and say "what the futz." XYZ does the
> normal and has them examine settings, and they see that the nameserver
> isn't the XYZ blessed nameserver. So XYZ tech support has them change
> it to the XYZ blessed nameservers. ABC Co. can now resolve the Internet.
> But that old open recurser is still there. Still recursing.
> Repeat 49,999 times.
> Don't bother trying to tell me it's unlikely. I've seen too many such
> So, let me be very clear: this is a losing battle, unless you're willing
> to get the cooperation of ISP's, probably do active scanning, or do other
> really intrusive things, it's not very realistic to think that you can
> eliminate this problem. If you can really get the cooperation of ISP's,
> etc., then you're better off doing BCP38...
> ... JG
This reminds me of deliberations of pretty much everything else related
to abuse of power. Just because something can potentially be abused,
doesn't mean it doesn't need to get done. How else would we ever have
Okay, enough about our ideas.
What would YOU suggest?
More information about the dns-operations