Trusted Domain SRv6

SRv6 as designed has evoked interest from various parties, though its deployment is being limited by known security problems in its architecture. This document specifies a standard way to create a solution that closes some of the major security concerns, while retaining the basis of the SRv6 protocol.

Fail-Closed Domain:: synonymous with a Trusted Domain.
Trusted Domain (TD):: A domain that prevents processing of a protocol without explicit configuration, defined in detail in . This document is limited to treatment of deployment of SRv6 in the context of a trusted domain only.
Fail-Closed Protocol (FCP):: A protocol that can be deployed by establishing a fail-closed domain.
TD-SRv6:: SRv6 modified to become a FCP and with that allowing for easy deployment in a trusted domain.

SRv6 relies on the concept of limited domain. The application of this concept in the context of the draft however, suffers from a lack of security that is easily deployable in an economi and scalable fashion. Limited domains without very careful deployment will invariably leak beyond the domain and allow untrusted traffic to enter the domain and terminate on any arbitrary node. As per RFC 8402RFC8402 Section 8, SRv6 that leaks beyond the border of a trusted domain creates a security violation. An established and proven solution is to create a trusted domain that has a default fail-closed approach and a well-defined trusted/untrusted boundary. Examples of fail-closed protocols include:

mpls
clns
bier

A fail-closed domain is determined by following properties: Processing of the protocol packet on an interface requires explicit configuration. Otherwise, due to lack of packet classification, further processing and forwarding cannot be achieved. In practical terms the behavior used most often is a drop of the offending packet. In a fail-closed protocol, leaking beyond the boundary of the domain requires explicit config. Fail-closed protocols are easily identifiable by their top level (e.g. link layer) encoding early in the packet formats and often by fields at a fixed offset. In another words either their encoding or encapsulation allows such packets to be easily distinguished from other traffic. Classification of the protocol packets is completely deterministic. Confining the protocol to the trusted domaim does not require complex processing in either hardware or software to allow for scalable and economical deployment. The boundary of a trusted domain consists of a set of interfaces that exhibit default behavior.

It is impossible to differentiate SRv6 and IPv6 at the link-layer or easily at network layer by e.g. a reserved protocol number the way IPSec does since SRv6 and IPv6 share the same ethernet types and IP protocol numbers. Hence, in the event of a packet being sent into a trusted domain, either accidentally or by a malicious actor, it is possible to send the frame to a node binding the specific SID, and have the packet processed, irrespective of the content of the underlying (encapsulated) packet. The security proposals in RFC8402 section 8.2 is based on the application of filters preventing ingress traffic at the boundary routers destined towards a SID within the domain. Such filtering is prone to configuration errors and in addition, has significant impact on fast matching hardware utilization on devices that have large numbers of ingress points into the domain. The matching itself, due to the complexity and numerous possibilities of expressing a set of SIDs will likely necessitate a complete semantic parsing of such list to guarantee fully precise matching including wildcarding in different forms. In the context of a trusted domain, anything outside of the operators control should not be considered trusted. This means applying filters to prevent leakage into the domain at every customer port, every server, and every cloud stack. The scale and complexity of maintaining such a "shorewall" is daunting and at large scale will not be likely to keep up with the timing necessary in case of attacks mounted and metamorphosing in short time intervals. An attack avoiding the filter wall may evade discovery for a long time in the absence of sophisticated traffic analyis and analytics tools.

To implement SRv6 in the context of a trusted domain, it is necessary to modify it to allow deployment in a fail-closed boundary efficiently. This requires changes to the protocol encapsulation at both the boundary routers and the transit nodes. This document introduces a distinct ethertype to be used for TD-SRv6 packets.

Trusted Domain boundary routers form the point at which the new ethertype is imposed. Imposition of the ethertype happens on packet ingress, at the same point as SRv6 header imposition is performed. Boundary interfaces will, by default behavior, drop packets already containing the TD-SRv6 ethertype.

In the case of a transit or egress router, should a frame not be marked with the TD-SRv6 ethertype, the frame will be treated as a standard IPv6 packet for the purposes of handling and forwarding. Even if an SRv6 packet is introduced into such domain with an ethertype different from TD-SRv6, the according SRv6 packet handling will not occur. Hence the resulting handling of the packet is indistinguishable from standard IPv6 processing. Only frames marked with the TD-SRv6 ethertype will be processed as SRv6 packets. A router configured to process TD-SRv6 MUST drop packets containing an SRH if received on any ethertype except TD-SRv6.

It cannot be excluded that deployment of TD-SRv6 are using TD-SRv6 on external interfaces but choose to revert to standard IPv6 ether type for SRv6 packets within the domain. The mechanisms required to realize such a deployment and risks incurred are outside the scope of this document.

No IANA Considerations

TD-SRv6 Ethertype: TBD0

This draft enhances the security mechanisms required by section 8 of RFC8402, and does not impose any further security considerations of its own.

Weiqiang Cheng chengweiqiang@chinamobile.com Anthony Somerset anthony.somerset@liquid.tech