DNS Amplification attack?

Mailman List Mirror (Read Only)
1

>> > On Tue, Jan 20, 2009 at 9:16 PM, Kameron Gasso <kgasso-lists@visp.net>=
wro=3D
>> te:
>>=20
>> > We're also seeing a great number of these, but the idiots spoofing =
the
>> > queries are hitting several non-recursive nameservers we host - and =
only
>> > generating 59-byte "REFUSED" replies.
>> >
>> > Looks like they probably just grabbed a bunch of DNS hosts out of =
WHOIS
>> > and hoped that they were recursive resolvers.
>>=20
>> First post to this list, play nice :slight_smile:
>>=20
>> Are you sure about this? I'm seeing these requests on /every/ =3D20
>> (unrelated) NS I have access to, which numbers several dozen, in =3D20
>> various countries across the world, and from various registries (.net, =
=3D20
>> .org, .com.au). The spread of servers I've checked is so random that =
=3D20
>> I'm wondering just how many NS records they've laid their hands on.
>>=20
>> I've also noticed that on a server running BIND 9.3.4-P1 with =3D20
>> recursion disabled, they're still appear to be getting the list of =
=3D20
>> root NS's from cache, which is a 272-byte response to a 61-byte =3D20
>> request, which by my definition is an amplification.
>=20
> BIND 9.3.4-P1 is past end-of-life.
>=20
> You need to properly set allow-query at both the option/view
> level and at the zone level to prevent retrieving answers
> from the cache in 9.3.x.
>=20
> option/view level "allow-query { trusted; };"
> zone level "allow-query { any; };"
>=20
> BIND 9.4.x and later have allow-query-cache make the
> configuration job easier. It also defaults to directly
> connected networks.

Another BIND-specific question since we're on the topic. I see
some of our authorative servers being hit with these spoofs, and
yes, the 9.3.5-P1 (that's what Sun supports in Solaris these
days) were sending back answers from the cache... but wait...
what cache?

  Authoritative servers need a cache. Authoritative servers
  need to ask queries. The DNS protocol has evolved since
  RFC 1034 and RFC 1035 and authoritative servers need to
  translate named to addresses for their own use.

  See RFC 1996, A Mechanism for Prompt Notification of Zone
  Changes (DNS NOTIFY).

The view the Internet gets only has our authorative zones. There
is no declaration for the root zone, master, slave, or hints.
How does BIND have the root cached in that view? Where did it
get it from? I guess it's hard coded somewhere?

Blocking this in the firewall. 1:0 amplification better than the
BIND fix, 1:1. But I'll get to the BIND fix anyway.

  The real fix is to get BCP 38 deployed. Reflection
  amplification attacks can be effective if BCP 38 measures
  have not been deployed. Go chase down the offending
  sources. BCP 38 is nearly 10 years old.

  We all should be taking this as a opportunity to find where
  the leaks are in the BCP 38 deployment and correct them.

  Mark

2

Mark Andrews <Mark_Andrews@isc.org> writes:

  Authoritative servers need a cache. Authoritative servers
  need to ask queries. The DNS protocol has evolved since
  RFC 1034 and RFC 1035 and authoritative servers need to
  translate named to addresses for their own use.

  See RFC 1996, A Mechanism for Prompt Notification of Zone
  Changes (DNS NOTIFY).

if i had RFC 1996 to do over again i would either limit outbound notifies
to in-zone servernames, or recommend that primary server operators
configure stealth slaves for servername-containing zones, or (most likely)
i would point out that the need to look up secondary servernames requires
that an authority-only nameserver be able to act as a stub resolver and
that such a server much have access to an independent recursive nameserver.

it's not too late to implement it that way. no authority-only server
should need a cache of any kind. the above text from marka represents
a BIND implementatin detail, not a protocol requirement, evolved or not.

  The real fix is to get BCP 38 deployed. Reflection
  amplification attacks can be effective if BCP 38 measures
  have not been deployed. Go chase down the offending
  sources. BCP 38 is nearly 10 years old.

my agreement with this statement is tempered by the fact that BCP38
deployment cannot be continuously assured, nor tested. therefore we will
need protocols, implementations, and operational practices that take
account of packet source address spoofing as an unduring property of the
internet.

  We all should be taking this as a opportunity to find where
  the leaks are in the BCP 38 deployment and correct them.

  Mark

yea, verily. and maybe track down rfc1918-sourced spew while you're at it.

3

* Mark Andrews:

  Authoritative servers need a cache. Authoritative servers
  need to ask queries. The DNS protocol has evolved since
  RFC 1034 and RFC 1035 and authoritative servers need to
  translate named to addresses for their own use.

  See RFC 1996, A Mechanism for Prompt Notification of Zone
  Changes (DNS NOTIFY).

Authoritative servers in typical configurations need a resolver (and
with views, you might even need a very specific resolver). This does
not mean that authoritative servers must be caches. It also does not
mean that a resolver operated from the view which contains a
particular authoritatively served zone picks up the correct data (in
other words, there are configurations where the current BIND magic
does not work).