RPKI-Based Policy Without Route Refresh

Memory constraints in early BGP speakers caused classic BGP implementations to not keep a full Adj-RIB-In (Sec. 1.1). When doing RPKI-based Route Origin Validation (ROV) ( and ), and similar RPKI-based policy, if such a BGP speaker receives new RPKI data, it might not have kept paths previously marked as Invalid etc. Such an implementation must then request a Route Refresh, and , from its neighbors to recover the paths which might be covered by these new RPKI data. This will be perceived as rude by those neighbors as it passes a serious resource burden on to them. This document recommends implementations keep and mark paths affected by RPKI-based policy, so Route Refresh is no longer needed.

It is assumed that the reader understands BGP, and Route Refresh , the RPKI , Route Origin Authorizations (ROAs), , The Resource Public Key Infrastructure (RPKI) to Router Protocol , RPKI-based Prefix Validation, , and Origin Validation Clarifications, .

As Route Origin Validation dropping Invalids has deployed, some BGP speaker implementations have been found which, when receiving new RPKI data (VRPs, see ) issue a BGP Route Refresh to all sending BGP peers so that it can reevaluate the received paths against the new data. In actual deployment this has been found to be very destructive, transferring a serious resource burden to the unsuspecting peers. In reaction, RPKI based Route Origin Validation (ROV) has been turned off. There have been actual de-peerings. As RPKI registration and ROA creation have steadily increased, this problem has increased, not just proportionally, but on the order of the in-degree of ROV implementing BGP speakers. As ASPA () becomes used, the problem will increase. Other mechanisms, such as automated policy provisioning, which have flux rates similar to ROV (i.e. on the order of minutes), could very well cause similar problems. Therefore this document updates by describing how to avoid this problem.

If new RPKI data arrive which cause operator policy to invalidate the best route, and the BGP speaker did not keep the dropped routes, then it would issue a route refresh, which this feature aims to prevent. A route that is dropped by operator policy due to ROV is, by nature, considered ineligible to compete for best route, and MUST be kept in the Adj-RIB-In for potential future evaluation. Ameliorating the Route Refresh problem by keeping a full Adj-RIB-In can be a problem for resource constrained BGP speakers. In reality, only some data need be retained. If an implementation chooses not to retain the full Adj-RIB-In, it MUST retain at least routes dropped due to ROV, for potential future evaluation. As storing these routes could cause problems in resource constrained devices, there MUST be a global operation, CLI, YANG, etc. allowing the operator to enable this feature, storing the dropped routes. Such an operator control MUST NOT be per peer, as this could cause inconsistent behavior. As a side note: policy which may drop routes due to RPKI-based checks such as ROV (and ASPA, BGPsec , etc. in the future) MUST be run, and the dropped routes saved per this section, before non-RPKI policies are run, as the latter may change path attributes.

Operators deploying ROV and/or other RPKI based policies should ensure that the BGP speaker implementation is not causing Route Refresh requests to neighbors. BGP Speakers MUST either keep the full Adj-RIB-In or implement the specification in . Conformance to this behavior is a additional, mandatory capability for BGP speakers performing ROV. If the BGP speaker does not implement these recommendations, the operator should enable the vendor's control to keep the full Adj-RIB-In, sometimes referred to as "soft reconfiguration inbound". The operator should then measure to ensure that there are no unnecessary Route Refresh requests sent to neighbors. If the BGP speaker's equipment has insufficient resources to support either of the two proposed options (keeping a full AdjRibIn or at least the dropped routes), the equipment should either be replaced with capable equipment or not use ROV. I.e. the configuration setting in should only be used in very well known and controlled circumstances. Such circumstances could include operational work-arounds such as the informed consent of the affected peers. Operators using the specification in should be aware that a misconfigured neighbor might erroneously send a massive number of paths, thus consuming a lot of memory. Hence pre-policy filtering such as described in could be used to reduce this exposure. If Route Refresh has been issued toward more than one peer, the order of receipt of the refresh data can cause churn in both best route selection and in outbound signaling. Internet Exchange Points (IXPs) which provide Route Servers should be aware that some members could be causing an undue Route Refresh load on the Route Servers and take appropriate administrative and/or technical measures. IXPs using BGP speakers as route servers should ensure that they are not generating excessive route refresh requests.

This document describes a denial of service which Route Origin Validation or other RPKI policy may place on a BGP neighbor, and describes how it may be ameliorated. Otherwise, this document adds no additional security considerations to those already described by the referenced documents.

None

The authors wish to thank Alvaro Retana, Ben Maddison, Derek Yeung, John Heasley, John Scudder, Matthias Waehlisch, Nick Hilliard, Saku Ytti, and Ties de Kock.