LISP Transport for Policy Distribution

When LISP ITRs are deployed with enough configuration to build a LISP overlay, they may require additional configurations such as security, QoS, and/or traffic forwarding policies. As networks continue to grow, it can be challenging to ensure these configurations are distributed to many ITRs and kept in sync. LISP network operators may wish to re-use their existing LISP architecture to distribute these configurations as opposed to configuring them by hand, using a script, or investing in a configuration management system. The configurations can be distributed via a mapping system that the network operator manages or is managed by a third-party as part of a managed service offering.

LISP related terms are defined as part of the LISP specification , notably EID, RLOC, Map-Request, Map- Reply, Map-Notify, Ingress Tunnel Router (ITR), Egress Tunnel Router (ETR), Map- Server (MS) and Map-Resolver (MR).

The ITR could use the mapping system to receive configuration policies for use cases such as: The RLOC interfaces of an ITR may be connected to WAN links that are policed at sub-line rate by its upstream provider. Using the mapping system, the ITR could receive and apply the QoS policies that would shape traffic to the correct rate on each ITR RLOC interface. ITRs use the mapping system to receive access-list (ACL) configuration(s) that would allow them to restrict traffic from authorized sources to authorized services. ITRs receive configurations that determine local forwarding policies, such as specifying ITR RLOCs to be used for egress forwarding on a per-application basis or RLOCs on different ITRs within the same LISP site to maintain application symmetry. Baseline configurations for common services (e.g., DNS, SSH, Syslog) can be maintained in a mapping system and distributed across multiple ITRs. Policy distribution is not meant to provide zero-touch provisioning for ITRs within a LISP network. At a minimum, the ITR must have a map resolver defined, IP connectivity to the map resolver, and one or more distinguished names defined for receiving specific policies from the mapping system.

The following figure illustrates a reference system used to support packet flow descriptions in this section.

The reference system contains two sites, site A and site B, with corresponding xTR-A and xTR-B providing encapsulation and decapsulation services for the overlay traffic. xTR-A uses interface-A to forward and receive encapsulated traffic through the RLOC space; and xTR-B uses interface-B for it. For packet flow purposes the reference system assumes that a network controller provides the policies to a map-server. When an ITR comes up, it requests it's designated policies with it's map-server. The MS may have this policy configured by the administrator via a network controller.

The following is an illustration of the sequence to distribute a policy registered by the controller with the mapping system, down to an ITR that requests its designated policies. In the example <ITR-A> represents the hostname of the ITR that learns a policy using this mechanism. The Mapping-System is either configured by an operator or learns a mapping sent by a controller though a Map-Register. The Mapping System learns the mapping: EID=”policy-<ITR-A>” --> RLOC= “{ "shape":{ "interface":"ifaceA", "direction":"outbound", "value":100Mbps }}”. The EID is encoded as a Distinguished Name and the RLOC as a JSON string. ITR-A is configured to dynamically learn policies from the Mapping System with the name “policy-ITR-A” (policy followed by its hostname). ITR-A sends a Map-Request to the Mapping System with EID=”policy-<ITR-A>” encoded as a Distinguished Name. The Map-Request is sent with the N-bit set. The Mapping System forwards the request to the appropriate Map-Server. The Map-Server adds ITR-A to the subscription list of EID=”policy-<ITR-A>” and sends back a Map-Notify with the mapping that the controller has registered. When ITR-A receives the Map-Notify installs the received policy locally, to shape traffic sent over the RLOC facing interface. Note that when the map-server has multiple policies associated with this ITR, it can send each one of the policies as an additional locator record (following the same JSON format) in the mapping. The locator count in the Map-Notify reflects the number of policies distributed with the mapping.

Policy distribution takes advantage of the LISP pubsub model to ensure that router updates are properly distributed when policies change. In such a case, and using the same reference sytem as above, the information exchange is as follows: The controller sends a Map-Register to the Mapping System, updating the policy mapping with: EID=”policy-<ITR-A>” --> RLOC= “{ "shape":{ "interface":"ifaceA", "direction":"outbound", "value":200Mbps }}”. When the corresponding Map-Server receives this update it checks the list of ITRs subscribed for updates of EID=”policy-<ITR-A>” and finds out that ITR-A is subscribed. The Map-Server sends a Map-Notify to ITR-A with the updated mapping information that has been registered. When ITR-A receives and validates the Map-Notify, it updates the local policy, changing the shaping rate as specified in the new JSON description. Note that if the JSON specifies the same policy that is currently applied the notification is ignored.

The mapping system that is used for distributing policy configurations can be managed by either the administrator who owns and operates their own LISP sites or a third-party administrator who offers LISP mapping system functionality as a managed service. A controller or orchestrator could be used to update and optimize policies within the mapping system based on network or ITR telemetry. Within the mapping system, the administrator must define a distinguished name that is specific to an ITR. The distinguished name is associated with the specific policy configurations that the ITR is to receive. Each ITR is configured with the minimal requirements to perform a mapping request procedure as well as a distinguished name that can be matched upon in the mapping system. Map-Servers should be able to receive policy registrations through the Map-Registration process. The Map-Registration must encode the policy following the specification in the policy distribution encoding section.

The ITR subscribes to its policy via the Map-Request procedure defined in section 5 of . The PubSub procedure is used to ensure that policies can be updated or audited after an ITR has received them. Policies are published to the ITR from the mapping system using the mapping notification procedure defined in section 6 of . EID-to-RLOC mappings used for policy distribution are of the type EID <Distinguished Name> to RLOC <JSON policy specification>. The EID is a distinguished name uniquely identifying a router in the system, while each RLOC record uses JSON encoding to specify the particular policy (or policies) that this router needs to implement.

When the ITR is configured to receive a policy using a distinguished name, the ITR sends a subscription for the EID record encoded as this Distinguished Name. When a policy has been registered with the Mapping System for this Distinguished Name, the ITR receives a publication with a list of policies as RLOC records and encoded as JSON strings (as defined in section 5.4 of . Example encoding for QoS policy that shapes traffic to 50 percent of the line-rate: EID-Record encoded as distinguished name “policy-ce-router1” RLOC-Record record encoded as JSON string “{"shape":{"interface":"ethernet1","direction":"outbound", "unit":"percent","value":50}}” Example encoding for setting the ITR’s NTP server to 10.10.10.10: EID-Record encoded as distinguished name “policy-ce-router” RLOC-Record record encoded as JSON string "{"NTP-address": "10.10.10.10"}" Multiple ITRs can be configured to use multiple distinguished names for receiving multiple sets policies. This allows for an ITR to receive specific policies and many ITRs to receive policies that can be broadly applied. Referring to the two examples above, an ITR can be configured to use a distinguished name of “policy-ce-router1” to receive a QoS configuration that is specific to that node while also using a distinguished name of “policy-ce-router” to receive configurations that are common to each ITR in the LISP network (e.g., NTP configuration). The use of multiple distinguished names per ITR reduces the amount of configuration within the mapping system.

This memo includes no request to IANA.

Thanks to James Stankiewicz for his thorough comments and suggestions.