A client
asked me a question last week about UDP ports and compliance with CIP-007 R1.1.
Since I didn’t know the answer, I posted the question on the Western
Interconnect Compliance Forum board, where there are often very good
discussions of CIP issues. This time didn’t disappoint; there were some very
good responses (including several I haven’t included here). Below is that
discussion, followed by my short discussion of the lesson I’ve learned from the
discussion.
My original post on WICF:
A client is
asking the following question. Does anyone have ideas on this?
“We have
numerous hardened devices such as HP iLOs (Integrated Lights Out) which don't
have any sort of NetStat command to show ports. These devices get monthly scans
from our Nessus scanning tool. But each month, the monthly scans produce a
different list of UDP ports for the devices. Since UDP ports are, by nature,
stateless, they cannot be opened or closed. What is the best way to document
these ever-changing UDP ports in our CIP-010 R1 baseline?”
Response from WECC entity 1:
Ports on any
hardware device are tough. A NETSTAT or scan at any moment doesn't tell the
whole story, especially as many listening ports may actually be ephemeral
responses to outgoing traffic.
Usually, a
good bet when you can't get clean port information from the device itself is to
go to manufacturer's documentation. But there's nothing easy or direct to basic
web searches. I get this:
But it just
gives TCP ports. What's going on with UDP?
It's possible
that the UDP ports could all be DNS responses. The iLO wants to resolve a
domain name like HPE.com, so it shoots out a request to port 53 and listens on
a dynamic port for the response. So that port will always change, and you won't
know what it's for unless you catch the port 53 UDP packet when it goes out.
The tricky
thing with UDP is that it could be "fire-and-forget." It doesn't form
a connection. There could be tons of UDP traffic for which you never see an
open port. There could be leftover open ports that don't actually do anything.
Usually,
NETSTAT and NMAP will tell you open TCP ports, but they won't tell you all the
ports being used for UDP. Instead of capturing ports, the only reliable way is
to capture traffic in transit.
You can wrap
it by putting it behind a firewall and review the firewall logs to see what
went through. You can restrict rules until you see what breaks. Or you can span
the port on the switch and pass the span port into Wireshark. Maybe you can
even find a useful command in the SSH shell. Or maybe you come from a company
with enough clout that you can get HP themselves to tell you what the ports are
for.
There's
almost always a way to get ports with some degree of justification.
Response from Kevin Perry, retired former
Chief CIP Auditor of SPP Regional Entity (note: I contacted Kevin by email, not through WICF):
The above
person's response is spot on. The difference between TCP and UDP is that UDP is
stateless. There is no negotiated, bi-directional communication session like there
is with TCP. The sender transmits the UDP packet with an IP address and port,
and hopes it is picked up by the destination. There is no acknowledgment from
the destination. On the destination device, the UDP port has to be raised to
catch the packet. Because there is no "state", there is no distinct
declaration that the port is "listening" as opposed to another state
such as "established", like there is with TCP ports. Essentially, all
raised UDP ports not tied to the local loopback address are network accessible
and are considered to be listening. And, yes, there are exploits that make use
of UDP ports. Network Time Protocol (ntp - UDP port 123) had an active exploit
a couple of years ago.
In the case
of ntp, both the client and the time server have to raise UDP port 123. The
client sends a time update request out over UDP port 123 to the time server.
The time server has UDP port 123 raised in order to receive the request. It
then sends the reply time packet back to the requester (the client) using UDP port
123. That means the client must also have UDP port 123 raised in order to get
the time hack. In some environments, the time server automatically broadcasts
the time on a defined interval and again, the devices needing the time hack
have UDP port 123 raised.
Nmap
struggles with UDP ports. It may report falsely that a UDP port is open (or
not). And, as Joseph pointed out, it may simply be an instant communication
like dns that tries to mimic TCP bi-directional communication without creating
and then tearing down a session. So, like Joseph said, vendor documentation is
the best source if you cannot interrogate the device using something like
netstat.
As far as
documentation goes, you can document a port range and associate it with the
service or application, whether TCP or UDP. In the case of the ILO card, you
cannot turn the logical ports off, so they are deemed necessary. And, the ports
need to be identified, documented, and restricted as appropriate at the
perimeter firewall, not just because CIP-005 says so, but because you need to
minimize your attack surface as a good utility and good cyber security
practice.
(Kevin’s response to follow-up email from
me)
A number of
entities have been disabused of the idea that UDP ports were excluded because
they were not in a “listening” state.
Entity
approaches I have seen vary, but they are usually some combination of vendor
documentation, internal native commands like netstat, host-run programs that do
the equivalent of internal native commands, and external scanners like nmap.
I have not seen anyone do network packet captures and analysis, but it
would certainly show what was passing back and forth.
Response from Brent Sessions, VP Risk and
Reliability Compliance, Western Area Power Administration (WAPA):
In the case
of UDP, tools that can help you demonstrate what UDP traffic is going to and
from a device could be your network IDS, a host-based IDS, or a passive
vulnerability scanner. I've used these
in the past to demonstrate the random port issue - you can set rules around
those for monitoring. Cumbersome, maybe,
but it can capture the UDP traffic if the sensors are in the right place.
Also, even
with UDP, there is a service that has to have a handle to bring up a port and
listen. Knowing what service that is and
how it behaves is a good way to demonstrate you understand your network port
baseline. Like others have said, vendor
docs are invaluable here. But if the
vendor is not responsive, looking at traffic patterns over time using the
methods described here, you might be able to build a profile. Better than nothing.
And now here’s Tom’s response to this
discussion:
It seems
clear to me that there are lots of things a NERC entity can do to identify UDP
ports, although no technique will ever provide the definitive list. So the
question really becomes “What is enough for compliance?”
If CIP-007
R1 were a risk-based requirement, this would be a very easy question to answer.
The requirement would say something like “Take appropriate steps to mitigate
the risk posed by open logical ports on BES Cyber Systems.” When deciding how
it will respond for UDP ports vs. TCP ports, the entity will need to decide the
relative risks posed by the two types of ports. Since there have been very few
successful UDP attacks and since they don’t seem to have caused much damage
when they happened (vs. – as we all know – huge amounts of damage caused by
successful TCP attacks), it would be very easy to document that the relative
levels of risk posed by UDP vs. TCP ports justified focusing the entity’s
resources (which aren’t unlimited, of course) on the latter.
But as we
all know, CIP-007 R1 isn’t a risk-based requirement, which is why the only
correct answer to the question what is enough for compliance is “Whatever your
auditor will accept.” Of course, this is the case with other CIP requirements
as well, but it seems it’s especially appropriate here, since there simply is
no such animal as a definitive list of open UDP ports. Of course, this isn't a very satisfying answer, since it shows that here - as with many other questions about the prescriptive CIP requirements - you have to rely on the "Don't ask, don't tell" NERC CIP compliance structure, in which entities are forced either to ask an auditor from their Region the answer to a compliance question (and then only receive a verbal response, which won't at all be binding on any future auditor) or take their best guess and hope future auditors will consider it close enough to what they want to see that they will give them a pass (although perhaps with an Area of Concern to address before the next audit).
So what’s
the root of the problem? Is it that the CIP v5 drafting team (who drafted this
requirement) got lazy and didn’t bother to define clear criteria for
identifying open UDP ports? No. As with a lot of questions having to do with
cybersecurity, there simply is no correct way to do this, and there doesn’t even seem to be any real meaning to the phrase “the correct way to identify open UDP
ports”.
The problem
is that NERC standards are prescriptive. If you think of the Operations and
Planning standards (most of which long predate CIP), they are prescriptive
because they have to be. They’re ultimately based on the laws of physics, which
– the last time I checked – don’t change from time to time or situation to
situation. If you do something once and the rest of the environment hasn’t
changed, you’ll get the same results if you do it again.
However, since
there are no immutable “laws” of cybersecurity, prescriptive requirements
simply don’t make sense there. If you’ve been dealing with CIP for a long time,
you know this is hardly the only instance – or even the twentieth or thirtieth
- where there simply is no good answer to a question of how to comply with a prescriptive requirement. The only approach that works for cybersecurity regulation is a
risk-based (also called “plan-based” or “objectives-based”. They amount to the
same thing) one. Currently, CIP-003 R2, CIP-007 R3, CIP-010 R4, CIP-011 R1,
CIP-013 and CIP-014 are risk based (and the drafts of CIP-012 indicate that
will be risk-based as well). Ultimately, I think all of the CIP standards will
be risk-based (and it’s interesting that literally all of the new standards and
requirements drafted since CIP v5 have been risk-based. I don’t think anyone
would even propose that a new NERC CIP requirement be anything but risk-based).
There’s simply no other way to write successful security standards.
Any opinions expressed in this blog post are strictly mine
and are not necessarily shared by any of the clients of Tom Alrich LLC.
If you would like to comment on what you have read here, I
would love to hear from you. Please email me at tom@tomalrich.com. Please keep
in mind that if you’re a NERC entity, Tom Alrich LLC can help you with NERC CIP
issues or challenges like what is discussed in this post – especially on
compliance with CIP-013. My offer of a free webinar on CIP-013, specifically for
your organization, remains open to NERC entities and vendors of hardware or
software components for BES Cyber Systems. To discuss this, you can email me at
the same address.
No comments:
Post a Comment