Re: IPv6 and DHCP and ICMP

[John Ladwig <John.Ladwig () SO MNSCU EDU>]

The point of this thread is, if you "block most or all ICMP traffic in order to eliminate an attack surface," and you 
are running an IPv6 network and expect it to communicate with the rest of the IPv6 Internet, YOU WILL BE DISAPPOINTED.

The phone WILL ring, when some intermediate carrier decides that they're doing a code upgrade which happens to break 
jumbo frame support, and you have a researcher talking to a computing grid halfway across town or he world, and there's 
jumbo frame capability otherwise on the path, but now there's no possibility of IPv6 nodes discovering the path MTU.  
Then, the network guys will have the day they enjoy so much, blaming the security guys for making them adhere to a 
policy that is in violation of how the network *must* be configured in order to work.  And then you'll get all the 
invective that that carrier so richly deserved.

The message is, you have to learn more about what ICMPv6 messages and codes than you did with IPv4 before you set up a 
traffic policy.  Start with RFC4890 and have a series of unrushed discussion with your IPv6-literate network architects 
and engineers.

   -jml

From: The EDUCAUSE Security Constituent Group Listserv [mailto:SECURITY () LISTSERV EDUCAUSE EDU] On Behalf Of Everett, 
Alex D
Sent: Thursday, May 24, 2012 8:16 AM
To: SECURITY () LISTSERV EDUCAUSE EDU
Subject: Re: [SECURITY] IPv6 and DHCP and ICMP

Very good points and I agree with them.
However, for some important systems the concept of least functionality may need to be applied.
An extension of that may be to block most or all ICMP traffic in order to eliminate an attack surface.
For example, you may be concerned about exploits over IP/ICMP such as CVE-2011-1871, either before they are published 
or after.
Another example, it would be hard for an attacker to conduct a ping flood against this server if an ACL drops pings for 
that server.

Sincerely,

Alex Everett, CISSP, CCNA
University of North Carolina
Chapel Hill

On May 23, 2012, at 10:31 PM, randy marchany wrote:


One of the "prime directives" in any security  strategy is asking a) what is the purpose of a particular security 
control b) how effective is the control?

Which brings me to ICMP blocking? I believe it's totally ineffective for the reasons below.

What is the purpose of ICMP blocking? To keep someone from mapping your network? Do you think someone can't map your 
net if ICMP is blocked? Do you have wireless nets? Yes? Then your network is mapped. Do you have web servers? Yes? Then 
your net can be mapped. Do you have stateless firewalls at the border? Yes? Your net can be mapped by inverse mapping. 
Do you prevent "outbound" connections? Yes? then why not disconnect from the Internet :-)? No? your net has been mapped.
IMHO, blocking ICMP v4 or v6 accomplished nothing from a security perspective. There are far more effective strategies 
to accomplish the same goals. Our preliminary work with v6 (and by extension v4) is shows you can't hide a machine on 
the net if there's wireless connectivity. So why bother? Accept the fact that a machine can be identified on the net 
and change your focus to protecting the data on a machine rather than the machine itself.

So to answer the 2 initial questions I raised at the beginning of the post:

1. What is the purpose of ICMP blocking? To keep someone from mapping your net.
2. How effective is the control? Not effective at all because there are multiple ways to map a network and we cannot 
block all of them without interfering with the "business" purpose of your organization.

My personal opinion is that I don't care what comes into my net. I care about what "leaves" my net.  Protect the data.

-r.
On Wed, May 23, 2012 at 8:10 PM, Michael Sinatra <michael () rancid berkeley edu<mailto:michael () rancid berkeley 
edu>> wrote:
On 05/23/2012 14:22, John Ladwig wrote:
ICMPv4 should **never** have been "completely eliminated" from public

network (interacting with local network), but there's only a small set
of messages that **need** to pass an Internet/local policy boundary.

Limited, yes, but I've seen way to many blanket drop policies that I'm a
little touchy on the subject.

There's a slightly larger set of required ICMPv6 messages that must
cross an Internet/local policy boundary to enable, for example, path-MTU
discovery.

Our current proposals, LAN and WAN testbed configurations follow RFC
4890 ICMPv6 recommendations for firewall transit "must not be dropped"
and "normally should not be dropped" pretty closely, although we're not
currently testing mobile IPv6, and haven't decided whether to support it
in the near term.

+1 on RFC 4890--it's a really good resource both for firewalls and router ACLs.  Keep in mind that blocking all ICMPv6 
means blocking all IPv6.  You simply won't have connectivity if you block ND, for example.

michael


Sincerely,

Alex Everett, CISSP, CCNA
Information Security Office
University of North Carolina at Chapel Hill
919.445.9393