Passive IP Addresses
Cisco Systems
Santa Barbara
93117
California
USA
fred@cisco.com
Cisco Systems
De Kleetlaan 6a
Diegem
Belgium
1831
+32 2704 5473
gvandeve@cisco.com
Operations and Management
Operational Security
This note suggests an approach to minimizing the attack surface of
the network elements - routers, switches, and middleware - of a
network.
This note suggests an approach to minimizing the attack surface of
the network elements - routers, switches, and middleware - of a
network.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in .
The problem, at least in its first instance, is a side effect of
diagnostics used in the Internet. Tools such as mtr, traceroute, and
pingplotter operate by sending streams of packets to a remote address
with varying hop limit values in IPv6 or
Time to Live in IPv4, and receiving
ICMP or ICMPv6 messages that indicate which interfaces
the packet stream traversed in the forward direction. Path MTU
discovery depends on ICMP/ICMPv6 Packet Too Big. Various ICMP/ICMPv6
"unreachable" messages respond when routing fails, which are intended
to trigger applications to try other peer addresses , and so on. The IP
addresses of these responders can be looked up in Reverse DNS to
build a name that indicates the operator, POP, and equipment in
question, which is useful in identifying potential problems in the
path.
Unfortunately, those addresses can also be used in another way. A
motivated adversary can subject routers to TCP RST attacks, load-based
DDOS, and other attacks.
An alternate way to reduce this potential attack vector is to
not use addresses that are valid beyond the link it is attached towards.
A sollution describing considerations around this is given in
while passive IPv6
addresses will provide network path visibility withought increasing large extend
of vulnerability for the devices using down the traffic path.
To pick one example, attacks are being reported in which
residential broadband customer's CPE Router is targeted with large
volume SNMP GET Requests. The address of the router is not generally
known; in IPv4, that may be a result of NAPT use, with the address
being harvested from exchanges. It may be obtained from a traceroute
to a server behind the router, or it may be determined by analysis of
SMTP envelopes.
Another example is attacks on BGP peering. BGP neighbors often peer
between the loopback addresses of neighboring routers, to make the TCP
session stable in the presence of link outages, but may peer using
interface addresses. If a router is configured to use interface
addresses in ICMP/ICMPv6 messages and to peer using those same
addresses, the ICMP response exposes information that can be used in a
RST attack on routing. It also facilitates any other kind of attack on
the router, such as the previously noted SNMP attack (even if the
router knows to refuse the message, it consumes CPU). If global
addresses are not used - routers use link-local or private addresses -
that makes it harder for an attacker to attack the router, but it
means that traceroute and other uses are compromised, which is an
attack on network forensics. If link-local addresses are used on the
interfaces and ICMP is configured to use the loopback address, the
router is again exposed to RST attacks.
The simplest solution seems to be to enable the router to hide in
plain sight - to use an address as the source address in ICMP and other
messages that is identifiable using Reverse DNS (and therefore, through
the name, useful for network diagnostics and communication between
operators), but does not facilitate attacks.
The fundamental theory behind this proposal is the Principle of Least
Privilege, which in this application is that an entity in the Internet
must be able to access only the information and resources that are
necessary for its legitimate purpose. In this case, it is reasonable,
for various reasons, to enable a random user to identify the path his or
her traffic is using or to identify a system in his path when reporting
operational issues to an administration. It is not reasonable, or at
least not required, that the user be able to specifically interact with
any of those systems in the general case.
We propose that the source IP address in an ICMP/ICMPv6 message, or
indeed any message sent to a host that has no inherent need to contact
the specific system, be useful for Reverse DNS, but not for touching the
system. Ideally, it is not routable to the system in the first place;
The passive character of this type of addres address comes to play if
a packet with this address as destination address on a targetted device and
is delivered to the interface, it is summarily dropped. Such an
address is referred to as a "passive address", and if it comes from a
specific prefix, the prefix is referred to as a "passive prefix".
Addresses that are routable and not dropped on receipt will, for the
purposes of this specification, be called "active" IP addresses.
A passive address is sementically non-disguisable from any other type
of address and has no requirement for any new type of address-family. Any IPv4
or IPv6 address can become a passive address by a configuration knob when
specifying the interface IP address for the Interface or device.
Every interface in the Internet has an address, with the exception
of IPv4 unnumbered interfaces; even those have addresses that they
use, which are the actual address of some other interface on the same
system. Increasingly, this is in fact a list of addresses, some of
which are IPv4 and some of which are IPv6.
We propose that any address allocated to an interface on
infrastructure equipment be given two binary attributes:
If the address has this attribute TRUE,
the corresponding address may be used as the source address of
ICMP or ICMPv6 messages and other messages sent to hosts that have
no need to actually touch the system. It is otherwise FALSE.
If the address has this attribute TRUE, the
device will process and respond to packets it receives that have
this as a destination address; it is an active address. If the
attribute is FALSE, the address is a passive address.
If UseInICMP is set TRUE on a Global Unicast Address or Unique
Local Address, the address will be available for use in ICMP messages.
If "Respond" is set TRUE, traffic sent to the address will be served
in the usual way. This describes the present Internet usage. If
Respond is set FALSE, traffic sent to the address will be summarily
discarded, in effect presenting a "local firewall" blockage related to
the address.
An address that has UseInICMP set FALSE will not be used as the
source address of an ICMP message. That address will be indiscoverable
via ICMP messages. If Respond is TRUE and the address becomes known by
other means, such as DNS, traffic sent to the address will be served
in the usual way. If Respond is set FALSE, traffic sent to the address
will be summarily discarded, in effect presenting a "local firewall"
blockage related to the address.
The scenario in view here is that
an address that is used to access the system would have
UseInICMP FALSE (the address is not leaked in such messages) and
Respond TRUE (messages sent to the address MAY be operated on by
the system).
an address that is used in ICMP and similar messages would have
UseInICMP TRUE (the address MAY be leaked in such messages) and
Respond FALSE (messages sent to the address will be dropped on
receipt).
Per , an ICMP Response such as Time
Exceeded or Parameter Problem is sent from "the" source address of the
interface that detected the issue. This specification narrows that: it
SHOULD use one of the source addresses that have the attribute
UseInICMP set to TRUE. If no address has that attribute TRUE, it
SHOULD NOT send the message.
If the passive address is taken from any prefix that is not
advertised in routing, it will be difficult for an adversary to route
to the address, which simplifies the treatment of certain forms of
attacks. It is not impossible; a system on the same LAN could send a
crafted packet that would arrive anyway. However, especially in
inter-domain routing, it is often quite reasonable to believe that
addresses exist that need not be advertised to a neighboring
network.
One example of such an address, in IPv6, might be a Unique Local IPv6 Unicast Address, or a global
unicast address or prefix. There are obvious operational issues in the
use of a global prefix; it is easy to accidentally advertise it. In an
IPv4 network, the counterpart might be to use an address, or to use another prefix that one
chooses to not advertise.
Link Local addresses SHOULD NOT be used in this context; while they
are obviously unroutable except on the local LAN, they are not useful
in Reverse DNS.
One problem with this relates to Ingress
Filtering. If the prefix used for passive addresses is not
advertised to the neighboring network and the neighboring network is
using unicast reverse path filtering, it will filter these responses.
For this reason, a network doing this SHOULD advise neighboring
networks of passive prefixes for the purpose of inclusion in ingress
filters.
recommends that "For every IP
address, there should be a matching PTR record in the in-addr.arpa
domain." In IPv6, there is an important special case, in that
link-local addresses are not reflected there, and are used in routing
protocols for local communication among IPv6 routers. Like other
addresses, passive IP addresses SHOULD have a corresponding Reverse
DNS entry; these names help with traceroute and in fault diagnosis.
While active addresses may be expected to have A or AAAA records in
the administration's own DNS, there is little point for doing so for
passive addresses, as they are unresponsive and very likely
unreachable.
However, the names given to passive addresses SHOULD NOT be
directly similar to the names given active IP addresses. For example,
it may be useful to name the interfaces on a certain router so as to
identify the router - "ethernet7.card3.router5.lax.example.com". If
the correlation to the name of the loopback interface
("router5.lax.example.com") is obviously derivative, the security
value is largely forfeit, although it might require human interaction.
Such names should differ enough that they are not readily intuited,
such as "rack12.lax.example.com".
This memo asks the IANA for no new parameters.
Note to RFC Editor: This section will have served its purpose if it
correctly tells IANA that no new assignments or registries are required,
or if those assignments or registries are created during the RFC
publication process. From the author's perspective, it may therefore be
removed upon publication as an RFC at the RFC Editor's discretion.
This entire note could be described as addressing a set of security
considerations. It is not a complete solution to attacks on
infrastructure - if loopback addresses, which are used for network
management and other purposes are generally known, the infrastructure
can still be attacked. However, it is an important reduction of the
attack surface. It creates no attack surface that did not already
exist.
This proposal also introduces no new privacy issues.
This document grew from a conversation among the authors, John
Brzozowski, and Thienpondt Hans. Merike Keao's review was very
helpful.
1 March 2012
7 October 2012
Using Only Link-Local Addressing Inside an IPv6 Network