Internet Engineering Task Force (IETF) M. Behringer Request for Comments: 7404 E. Vyncke Category: Informational Cisco ISSN: 2070-1721 November 2014
Using Only Link-Local Addressing inside an IPv6 Network
In an IPv6 network, it is possible to use only link-local addresses on infrastructure links between routers. This document discusses the advantages and disadvantages of this approach to facilitate the decision process for a given network.
Status of This Memo
This document is not an Internet Standards Track specification; it is published for informational purposes.
This document is a product of the Internet Engineering Task Force (IETF). It represents the consensus of the IETF community. It has received public review and has been approved for publication by the Internet Engineering Steering Group (IESG). Not all documents approved by the IESG are a candidate for any level of Internet Standard; see Section 2 of RFC 5741.
Copyright (c) 2014 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
An infrastructure link between a set of routers typically does not require global or unique local addresses [RFC4193]. Using only link- local addressing on such links has a number of advantages; for example, routing tables do not need to carry link addressing and can therefore be significantly smaller. This helps to decrease failover times in certain routing convergence events. An interface of a router is also not reachable beyond the link boundaries, therefore reducing the attack surface.
This document discusses the advantages and caveats of this approach.
Note that some traditional techniques used to operate a network, such as pinging interfaces or seeing interface information in a traceroute, do not work with this approach. Details are discussed below.
During WG and IETF last call, the technical correctness of the document was reviewed; however, debate exists as to whether to recommend this technique. The deployment of this technique is appropriate where it is found to be necessary.
2. Using Link-Local Addressing on Infrastructure Links
This document discusses the approach of using only link-local addresses (LLAs) on all router interfaces on infrastructure links. Routers don't typically need to receive packets from hosts or nodes outside the network. For a network operator, there may be reasons to use addresses that are greater than link-local scope on infrastructure interfaces for certain operational tasks, such as pings to an interface or traceroutes across the network. This document discusses such cases and proposes alternative procedures.
In this approach, neither globally routed IPv6 addresses nor unique local addresses are configured on infrastructure links. In the absence of specific global or unique local address definitions, the default behavior of routers is to use link-local addresses, notably for routing protocols.
The sending of ICMPv6 [RFC4443] error messages ("packet-too-big", "time-exceeded", etc.) is required for routers. Therefore, another interface must be configured with an IPv6 address that has a greater scope than link-local. This address will usually be a loopback interface with a global scope address belonging to the operator and part of an announced prefix (with a suitable prefix length) to avoid being dropped by other routers implementing ingress filtering [RFC3704]. This is implementation dependent. For the remainder of this document, we will refer to this interface as a "loopback interface".
[RFC6724] recommends that IPv6 addresses that are greater than link- local scope be used as the source IPv6 address for all generated ICMPv6 messages sent to a non-link-local address, with the exception of ICMPv6 redirect messages (as defined in Section 4.5 of [RFC4861]).
The effect on specific traffic types is as follows:
o Most control plane protocols (such as BGP [RFC4271], IS-IS [IS-IS], OSPFv3 [RFC5340], Routing Information Protocol Next Generation (RIPng) [RFC2080], and PIM [RFC4609]) work by default or can be configured to work with link-local addresses. Exceptions are explained in the caveats section (Section 2.3).
o Management plane traffic (such as Secure SHell (SSH) Protocol [RFC4251], Telnet [RFC0495], Simple Network Management Protocol (SNMP) [RFC1157], and ICMPv6 Echo Request [RFC4443]) can use the address of the router loopback interface as the destination address. Router management can also be done over out-of-band channels.
o ICMP error messages are usually sourced from a loopback interface with a scope that is greater than link-local. Section 4.5 of [RFC4861] explains one exception: ICMP redirect messages can also be sourced from a link-local address.
o Data plane traffic is forwarded independently of the link address type.
Behringer & Vyncke Informational [Page 3]
RFC 7404 Link-Local Only November 2014
o Neighbor discovery (neighbor solicitation and neighbor advertisement) is done by using link-local unicast and multicast addresses. Therefore, neighbor discovery is not affected.
Thus, we conclude that it is possible to construct a working network in this way.
The following list of advantages is in no particular order.
Smaller routing tables: Since the routing protocol only needs to carry one global address (the loopback interface) per router, it is smaller than the traditional approach where every infrastructure link address is carried in the routing protocol. This reduces memory consumption and increases the convergence speed in some routing failover cases. Because the Forwarding Information Base to be downloaded to line cards is smaller, and there are fewer prefixes in the Routing Information Base, the routing algorithm is accelerated. Note that smaller routing tables can also be achieved by putting interfaces in passive mode for the Interior Gateway Protocol (IGP).
Simpler address management: Only loopback interface addresses need to be considered in an addressing plan. This also allows for easier renumbering.
Lower configuration complexity: Link-local addresses require no specific configuration, thereby lowering the complexity and size of router configurations. This also reduces the likelihood of configuration mistakes.
Simpler DNS: Less routable address space in use also means less reverse and forward mapping DNS resource records to maintain. Of course, if the operator selects not to enter any global interface addresses in the DNS anyway, then this is less of an advantage.
Reduced attack surface: Every routable address on a router constitutes a potential attack point; a remote attacker can send traffic to that address, for example, a TCP SYN flood (see [RFC4987]). If a network only uses the addresses of the router loopback interface(s), only those addresses need to be protected from outside the network. This may ease protection measures, such as Infrastructure Access Control Lists (iACL). Without using link-local addresses, it is still possible to achieve the simple iACL if the network addressing scheme is set up such that all link and loopback interfaces have addresses that are greater than link-local and are aggregatable, and if the infrastructure access list covers that entire aggregated space. See also [RFC6752] for further discussion
Behringer & Vyncke Informational [Page 4]
RFC 7404 Link-Local Only November 2014
on this topic. [RFC6860] describes another approach to hide addressing on infrastructure links for OSPFv2 and OSPFv3 by modifying the existing protocols. This document does not modify any protocol and applies only to IPv6.
The caveats listed in this section are in no particular order.
Interface ping: If an interface doesn't have a routable address, it can only be pinged from a node on the same link. Therefore, it is not possible to ping a specific link interface remotely. A possible workaround is to ping the loopback address of a router instead. In most cases today, it is not possible to see which link the packet was received on; however, [RFC5837] suggests including the interface identifier of the interface a packet was received on in the ICMPv6 response. It must be noted that there are few implementations of this ICMPv6 extension. With this approach, it would be possible to ping a router on the addresses of loopback interfaces, yet see which interface the packet was received on. To check liveliness of a specific interface, it may be necessary to use other methods, such as connecting to the router via SSH and checking locally or using SNMP.
Traceroute: Similar to the ping case, a reply to a traceroute packet would come from the address of a loopback interface, and current implementations do not display the specific interface the packets came in on. Again, [RFC5837] provides a solution. As in the ping case above, it is not possible to traceroute to a particular interface if it only has a link-local address. Conversely, this approach may make network topology discovery from outside the network simpler: instead of responding with multiple different interface IP addresses, which have to be correlated by the outsider, a router will always respond with the same loopback address. If reverse DNS mapping is used, the mapping is trivial in either case.
Hardware dependency: LLAs have usually been based on 64-bit Extended Unique Identifiers (EUI-64); hence, they change when the Message Authentication Code (MAC) address is changed. This could pose a problem in a case where the routing neighbor must be configured explicitly (e.g., BGP) and a line card needs to be physically replaced, hence changing the EUI-64 LLA and breaking the routing neighborship. LLAs can be statically configured, such as fe80::1 and fe80::2, which can be used to configure any required static routing neighborship. However, this static LLA configuration may be more complex to operate than statically configured addresses that are greater than link-local scope. This is because LLAs are inherently ambiguous. For a multi-link node, such as a router, to deal with the
Behringer & Vyncke Informational [Page 5]
RFC 7404 Link-Local Only November 2014
ambiguity, the link zone index must also be considered explicitly, e.g., using the extended textual notation described in [RFC4007], as in this example, 'BGP neighbor fe80::1%eth0 is down'.
Network Management System (NMS) toolkits: If there is any NMS tool that makes use of an interface IP address of a router to carry out any of its NMS functions, then it would no longer work if the interface does not have a routable address. A possible workaround for such tools is to use the routable address of the router loopback interface instead. Most vendor implementations allow the specification of loopback interface addresses for SYSLOG, IPFIX, and SNMP. The Link Layer Discovery Protocol (LLDP) (IEEE 802.1AB-2009) runs directly over Ethernet and does not require any IPv6 address, so dynamic network discovery is not hindered by using only LLA when using LLDP. But, network discovery based on Neighbor Discovery Protocol (NDP) cache content will only display the link-local addresses and not the addresses of the loopback interfaces; therefore, network discovery should rather be based on the Route Information Base to detect adjacent nodes.
MPLS and RSVP-Traffic Engineering (RSVP-TE) [RFC3209] allow the establishment of an MPLS Label Switched Path (LSP) on a path that is explicitly identified by a strict sequence of IP prefixes or addresses (each pertaining to an interface or a router on the path). This is commonly used for Fast Reroute (FRR). However, if an interface uses only a link-local address, then such LSPs cannot be established. At the time of writing this document, there is no workaround for this case; therefore, where RSVP-TE is being used, the approach described in this document does not work.
Internet Exchange Points (IXPs) have a special importance in the global Internet because they connect a high number of networks in a single location and because a significant part of Internet traffic passes through at least one IXP. An IXP requires, therefore, a very high level of security. The address space used on an IXP is generally known, as it is registered in the global Internet Route Registry, or it is easily discoverable through traceroute. The IXP prefix is especially critical because practically all addresses on this prefix are critical systems in the Internet.
Behringer & Vyncke Informational [Page 6]
RFC 7404 Link-Local Only November 2014
Apart from general device security guidelines, there are basically two additional ways to raise security (see also [BGP-OPSEC]):
2. To drop all traffic from remote locations destined to the IXP prefixes.
Not announcing the prefix of the IXP would frequently result in traceroute and similar packets (required for Path MTU Discovery (PMTUD)) being dropped due to unicast Reverse Path Forwarding (uRPF) checks. Given that PMTUD is critical, this is generally not acceptable. Dropping all external traffic to the IXP prefix is hard to implement because if only one service provider connected to an IXP does not filter correctly, then all IXP routers are reachable from at least that service provider network.
As the prefix used in the IXP is usually longer than a /48, it is frequently dropped by route filters on the Internet having the same net effect as not announcing the prefix.
Using link-local addresses on the IXP may help in this scenario. In this case, the generated ICMPv6 packets would be generated from loopback interfaces or from any other interface with a globally routable address without any configuration. However, in this case, each service provider would use their own address space, making a generic attack against all devices on the IXP harder. All of an IXP's loopback interface addresses can be discovered by a potential attacker with a simple traceroute; a generic attack is, therefore, still possible, but it would require more work.
In some cases, service providers carry the IXP addresses in their IGP for certain forms of traffic engineering across multiple exit points. Link-local addresses cannot be used for this purpose; in this case, the service provider would have to employ other methods of traffic engineering.
If an Internet Exchange Point is using a global prefix registered for this purpose, a traceroute will indicate whether the trace crosses an IXP rather than a private interconnect. If link-local addressing is used instead, a traceroute will not provide this distinction.
Exclusively using link-local addressing on infrastructure links has a number of advantages and disadvantages, both of which are described in detail in this document. A network operator can use this document to evaluate whether or not using link-local addressing on
Behringer & Vyncke Informational [Page 7]
RFC 7404 Link-Local Only November 2014
infrastructure links is a good idea in the context of his/her network. This document makes no particular recommendation either in favor or against.
Using only LLAs on infrastructure links reduces the attack surface of a router. Loopback interfaces with routed addresses are still reachable and must be secured, but infrastructure links can only be attacked from the local link. This simplifies security of control and management planes. The approach does not impact the security of the data plane. The link-local-only approach does not address control plane [RFC6192] attacks generated by data plane packets (such as hop-limit expiration or packets containing a hop-by-hop extension header).
For additional security considerations, as previously stated, see also [RFC5837] and [BGP-OPSEC].
[BGP-OPSEC] Durand, J., Pepelnjak, I., and G. Doering, "BGP operations and security", Work in Progress, draft-ietf-opsec-bgp- security-05, August 2014.
[IS-IS] International Organization for Standardization, "Intermediate System to Intermediate System intra-domain routeing information exchange protocol for use in conjunction with the protocol for providing the connectionless-mode network service (ISO 8473)", ISO Standard 10589, 2002.
The authors would like to thank Salman Asadullah, Brian Carpenter, Bill Cerveny, Benoit Claise, Rama Darbha, Simon Eng, Wes George, Fernando Gont, Jen Linkova, Harald Michl, Janos Mohacsi, Ivan Pepelnjak, Alvaro Retana, Jinmei Tatuya, and Peter Yee for their useful comments about this work.
Michael Behringer Cisco Building D, 45 Allee des Ormes Mougins 06250 France
Eric Vyncke Cisco De Kleetlaan, 6A Diegem 1831 Belgium