DHCPv6 Prefix Delegating Relay

For Internet service providers that offer native IPv6 access with prefix delegation to their customers, a common deployment architecture is to have a DHCPv6 relay agent function located in the ISP's Layer-3 customer edge device and separate, centralized DHCPv6 server infrastructure. describes the functionality of a DHCPv6 relay and Section 19.1.3 mentions this deployment scenario, but does not provide detail for all of the functional requirements that the relay needs to fulfill to operate deterministically in this deployment scenario. A DHCPv6 relay agent for prefix delegation is a function commonly implemented in routing devices, but implementations vary in their functionality and client/server inter-working. This can result in operational problems such as messages not being forwarded by the relay or unreachability of the delegated prefixes. This document provides a set of requirements for devices implementing a relay function for use with prefix delegation. The mechanisms for a relay to inject routes (including aggregated ones), on its network-facing interface based on prefixes learnt from a server via DHCP-PD are out of scope of the document. Multi-hop DHCPv6 relaying is not affected, as the requirements in this document are solely applicable to the DHCP relay agent co-located with the first-hop router that the DHCPv6 client requesting the prefix is connected to, no changes to any subsequent relays in the path are needed.

This document uses the terminology defined in , however when defining the functional elements for prefix delegation , Section 4.2 defines the term 'delegating router' as: "The router that acts as a DHCP server and responds to requests for delegated prefixes." This document is concerned with deployment scenarios in which the DHCPv6 relay and DHCPv6 server functions are separated, so the term 'delegating router' is not used. Instead, a new term is introduced to describe the relaying function: A delegating relay acts as an intermediate device, forwarding DHCPv6 messages containing IA_PD/IAPREFIX options between the client and server. The delegating relay does not implement a DHCPv6 server function. The delegating relay is also responsible for routing traffic for the delegated prefixes. Where the term 'relay' is used on its own within this document, it should be understood to be a delegating relay, unless specifically stated otherwise. In CableLabs DOCSIS environments, the Cable Modem Termination System (CMTS) would be considered a delegating relay with respect to Customer Premises Devices (CPEs) , Section 5.2.7.2. A Broadband Network Gateway (BNG) in a DSL based access network may be a delegating relay if it does not implement a local DHCPv6 server function , Section 4.10. defines the 'DHCP server', (or 'server') as: "A node that responds to requests from clients. It may or may not be on the same link as the client(s). Depending on its capabilities, if it supports prefix delegation it may also feature the functionality of a delegating router." This document serves the deployment cases where a DHCPv6 server is not located on the same link as the client (necessitating the delegating relay). The server supports prefix delegation and is capable of leasing prefixes to clients, but is not responsible for other functions required of a delegating router, such as managing routes for the delegated prefixes. The term 'requesting router' has previously been used to describe the DHCP client requesting prefixes for use. This document adopts the terminology and uses 'DHCP client' or 'client' interchangeably for this element.

The following diagram shows the deployment topology relevant to this document.

| _ ,--,_ | +--------+ +------------+ _( `' )_ +--------+ +---+ PD |-------| Delegating |--( Operator )---| DHCPv6 | | | Client | | relay | `(_ Network_)' | server | | +--------+ +----------- + `--'`---' +--------+ | | <----- downlink ------ + (client facing) Client Network ]]> The client requests prefixes via the downlink interface of the delegating relay. The resulting prefixes will be used for addressing the client network. The delegating relay is responsible for forwarding DHCP messages, including prefix delegation requests and responses between the client and server. Messages are forwarded from the delegating relay to the server using multicast or unicast via the operator uplink interface. The delegating relay provides the operator's Layer-3 edge towards the client and is responsible for routing traffic to and from clients connected to the client network using addresses from the delegated prefixes.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here. This document uses these keywords not strictly for the purpose of interoperability, but rather for the purpose of establishing industry-common baseline functionality. As such, the document points to several other specifications (preferably in RFC or stable form) to provide additional guidance to implementers regarding any protocol implementation required to produce a DHCP relaying router that functions successfully with prefix delegation.

The following sections of the document describe problems that have been observed with delegating relay implementations in commercially available devices.

Delegating relay implementations have been observed not to forward messages between the client and server. This generally occurs if a client sends a message which is unexpected by the delegating relay. For example, the delegating router already has an active PD lease entry for an existing client on a port. A new client is connected to this port and sends a Solicit message. The delegating relay then drops the Solicit messages until it receives either a DHCP Release message from the original client, or the existing lease times out. This causes a particular problem when a client device needs to be replaced due to a failure. In addition to dropping messages, in some cases the delegating relay will generate error messages and send them to the client, e.g. 'NoBinding' messages being sent in the event that the delegating relay does not have an active delegated prefix lease.

For proper routing of client traffic, the delegating relay requires a corresponding routing table entry for each active prefix delegated to a connected client. A delegating relay which does not store this state persistently across reboots will not be able to forward traffic to client's delegated leases until the state is re-established through new DHCP messages.

allows for a client to include more than one instance of OPTION_IA_PD in messages in order to request multiple prefix delegations by the server. If configured for this, the server supplies one (or more) instance of OPTION_IAPREFIX for each received instance of OPTION_IA_PD, each containing information for a different delegated prefix. In some delegating relay implementations, only a single delegated prefix per-DUID is supported. In those cases only one IPv6 route for one of the delegated prefixes is installed; meaning that other prefixes delegated to a client are unreachable.

It is an unfortunate operational reality that client devices with duplicate MAC addresses and/or DUIDs exist and have been deployed. In this situation, the operational costs of locating and swapping out such devices are prohibitive. Delegating relays have been observed to restrict forwarding client messages originating from one client DUID to a single interface. In this case if the same client DUID appears from a second client on another interface while there is already an active lease, messages originating from the second client are dropped causing the second client to be unable to obtain a prefix delegation. It should be noted that in some access networks, the MAC address and/or DUID are used as part of device identification and authentication. In such networks, enforcing MAC address/DUID uniqueness is a necessary function and not considered a problem.

If the client loses information about a prefix that it is delegated while the lease entry and associated route is still active in the delegating relay, then the relay will forward traffic to the client which the client will return to the relay (which is the client's default gateway (learnt via an RA). The loop will continue until either the client is successfully reprovisioned via DHCP, or the lease ages out in the relay.

To resolve the problems described in and pre-empt other undesirable behavior, the following section of the document describes a set of functional requirements for the delegating relay. In addition, relay implementers are reminded that makes it clear that relays MUST NOT drop (and hence not relay) packets that either contain message codes (Section 19 of ) it may not understand, or contain options that it does not understand (Section 19 of ).

The delegating relay MUST forward messages bidirectionally between the client and server without changing the contents of the message. The relay MUST allow for multiple prefixes to be delegated for the same client IA_PD. These delegations may have different lifetimes. The relay MUST allow for multiple prefixes (with or without separate IA_PDs) to be delegated to a single client connected to a single interface, identified by its DHCPv6 Client Identifier (DUID). A delegating relay may have one or more interfaces on which it acts as a relay, as well as one or more interfaces on which it does not (for example, in an ISP, it might act as a relay on all southbound interfaces, but not on the northbound interfaces). The relay SHOULD allow the same client identifier (DUID) to have active delegated prefix leases on more than one interface simultaneously, unless client DUID uniqueness is necessary for the functioning or security of the network. This is to allow client devices with duplicate DUIDs to function on separate broadcast domains. The maximum number of simultaneous prefixes delegated to a single client MUST be configurable. The relay MUST implement a mechanism to limit the maximum number of active prefix delegations on a single port for all client identifiers and IA_PDs. This value MUST be configurable. It is RECOMMENDED that delegating relays support at least 8 active delegated leases per client device and use this as the default limit. The delegating relay MUST update the lease lifetimes based on the Client Reply messages it forwards to the client and only expire the delegated prefixes when the valid lifetime has elapsed. On receipt of a Release message from the client, the delegating relay MUST expire the active leases for each of the IA_PDs in the message.

The relay MUST maintain a local routing table that is dynamically updated with leases and the associated next-hops as they are delegated to clients. When a delegated prefix is Released or expires, the associated route MUST be removed from the relay's routing table. The relay MUST provide a mechanism to dynamically update ingress filters permitting ingress traffic sourced from client delegated leases and blocking packets from invalid source prefixes. This is to implement anti-spoofing as described in . The relay MAY provide a mechanism to dynamically advertise delegated leases into a routing protocol as they are learnt. When a delegated lease is released or expires, the delegated route MUST be withdrawn from the routing protocol. The mechanism by which the routes are inserted and deleted is out of the scope of this document. If the relay has learned a route for a delegated prefix via a given interface, and receives traffic on this interface with a destination address within the delegated prefix (that is not an on-link prefix for the relay), then it MUST be dropped. This is to prevent routing loops. An ICMPv6 Type 1, Code 6 (Destination Unreachable, reject route to destination) error message MAY be sent back to the client. The ICMP policy SHOULD be configurable. The delegating relay's routing entry MUST use the same prefix length for the delegated prefix as given in the IA_PD.

In the event that the relay is restarted, active client prefix delegations will be lost. This may result in clients becoming unreachable. In order to mitigate this problem, the relay SHOULD implement at least one following: * Implement DHCPv6 bulk lease query as defined in . * Store active prefix delegations in persistent storage so they can be re-read after the reboot. If a client's next-hop link-local address becomes unreachable (e.g., due to a link-down event on the relevant physical interface), routes for the client's delegated prefixes MUST be retained by the delegating relay unless they are released or removed due to expiring DHCP timers. This is to re-establish routing for the delegated prefix if the client next-hop becomes reachable without the delegated prefixes needing to be re-learnt. The relay SHOULD implement DHCPv6 active lease query as defined in to keep the local lease database in sync with the DHCPv6 server.

The relay SHOULD implement an interface allowing the operator to view the active delegated prefixes. This SHOULD provide information about the delegated lease and client details such as client identifier, next-hop address, connected interface, and remaining lifetimes. The relay SHOULD provide a method for the operator to clear active bindings for an individual lease, client or all bindings on a port. To facilitate troubleshooting of operational problems between the delegating relay and other elements, it is RECOMMENDED that a time synchronization protocol is used by the delegating relays and DHCP servers.

The authors of this document would like to thank Bernie Volz and Ted Lemon for their valuable comments.

This memo includes no request to IANA.

This document does not add any new security considerations beyond those mentioned in Section 22 of . If the delegating relay implements filtering, then the filtering rules will need to be dynamically updated as delegated prefixes are leased. describes a method for securing traffic between the relay agent and server by sending DHCP messages over an IPSec tunnel. In this case the IPSec tunnel is functionally the server-facing interface and DHCPv6 message snooping can be carried out as described. It is RECOMMENDED that this is implemented by the delegating relay.