PANA Framework

PANA design provides support for various types of deployments. Access networks can differ based on the availability of lower-layer security, placement of PANA entities, choice of client IP configuration and authentication methods, etc. PANA can be used in any access network regardless of the underlying security. For example, the network might be physically secured, or secured by means of cryptographic mechanisms after the successful client-network authentication. The PANA client, PANA authentication agent, authentication server, and enforcement point are the relevant functional entities in this design. PANA authentication agent and enforcement point(s) can be placed on various elements in the access network (e.g., access point, access router, dedicated host). IP address configuration mechanisms vary as well. Static configuration, DHCP, stateless address autoconfiguration are possible mechanisms to choose from. If the client configures an IPsec tunnel for enabling per-packet security, configuring IP addresses inside the tunnel becomes relevant, for which there are additional choices such as IKE. This I-D defines a general framework for describing how these various deployment choices are handled by PANA and the access network architectures. Additionally, two possible deployments are described in detail: using PANA over DSL networks and WLAN networks.

In this document, several words are used to signify the requirements of the specification. These words are often capitalized. The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in .

This is the address configured before starting the PANA protocol exchange. This is the address (optionally) configured after a successful authentication. This is the address used as the inner address of an IPsec tunnel mode SA. When IPsec protection is used, depending on the deployment scenario used either a PRPA or POPA becomes the IPsec-TIA. This is the address used as the outer address of an IPsec tunnel mode SA. When IPsec protection is used, a PRPA becomes the IPsec-TOA.

PANA protocol is designed to facilitate authentication and authorization of client hosts in access networks. PANA is an EAP lower-layer that carries EAP authentication methods encapsulated inside EAP between a client host and an agent in the access network. While PANA enables the authentication process between the two entities, it is only a part of an overall AAA and access control framework. A AAA and access control framework using PANA is comprised of four functional entities. The client implementation of the PANA protocol. This entity resides on the client host that is requesting access to a local area network. Example client hosts can be laptops, PDAs, cell phones, desktops that are connected to a network via wired or wireless networks. A PaC is responsible for requesting network access and engaging in authentication process using the PANA protocol. The server implementation of the PANA protocol. A PAA is in charge of interfacing with the PaCs for authenticating and authorizing them for the network access service. The PAA consults an authentication server in order to verify the credentials and rights of a PaC. If the authentication server resides on the same host as the PAA, an API is sufficient for this interaction. When they are separated (a much more common case in public access networks), a protocol needs to run between the two. LDAP and AAA protocols like RADIUS and Diameter are commonly used for this purpose. The PAA is also responsible for updating the access control state (i.e., filters) depending on the creation and deletion of the authorization state. The PAA communicates the updated state to the enforcement points in the network. If the PAA and EP are residing on the same host, an API is sufficient for this communication. Otherwise, a protocol is required to carry the authorized client attributes from the PAA to the EP. While not prohibiting other protocols, currently SNMP is suggested for this task. The PAA resides on a node that is typically called a NAS (network access server) in the local area network. PAA can be hosted on any IP-enabled node on the same IP subnet as the PaC. For example on a BAS (broadband access server) in DSL networks, or PDSN in 3GPP2 networks. The server implementation that is in charge of verifying the credentials of a PaC that is requesting the network access service. The AS receives requests from the PAA on behalf of the PaCs, and responds with the result of verification together with the authorization parameters (e.g., allowed bandwidth, IP configuration, etc). The AS might be hosted on the same host as the PAA, on a dedicated host on the access network, or on a central server somewhere on the Internet. The access control implementation that is in charge of allowing access to authorized clients while preventing access by others. An EP learns the attributes of the authorized clients from the PAA. The EP uses simple or cryptographic filters to selectively allow and discard data packets. These filters may be applied at the link-layer or the IP-layer. When cryptographic access control is used, a secure association protocol needs to run between the PaC and EP. This protocol provides the cryptographic binding between the communication channel and the authorization state. Examples of secure association protocols include the 4-way handshake in IEEE 802.11i , and IKE in IP-based access control. Link-layer ciphering or IPsec is used following the secure association to enable per-packet integrity, authentication and replay protection (and optionally confidentiality). An EP must be located strategically in a local area network to minimize the access of unauthorized clients to the network. For example, the EP can be hosted on the switch that is directly connected to the clients in a wired network. That way the EP can drop unauthorized packets before they reach any other client host or beyond the local area network. illustrates these functional entities and the interfaces (protocols, APIs) among them.

RADIUS/ Diameter/ +-----+ PANA +-----+ LDAP/ API +-----+ | PaC |<----------------->| PAA |<---------------->| AS | +-----+ +-----+ +-----+ ^ ^ | | | +-----+ | IKE/ +-------->| EP |<--------+ SNMP/ API 4-way handshake +-----+ Some of the entities may be co-located depending on the deployment scenario. For example, the PAA and EP would be on the same node (BAS) in DSL networks. In that case a simple API is sufficient between the PAA and EP. In small enterprise deployments the PAA and AS may be hosted on the same node (access router) that eliminates the need for a protocol run between the two. The decision to co-locate these entities or otherwise, and their precise location in the network topology are deployment decisions. Use of IKE or 4-way handshake protocols for secure association is only required in the absence of any lower-layer security prior to running PANA. Physically secured networks (e.g., DSL) or the networks that are already cryptographically secured on the link-layer prior to PANA run (e.g., cdma2000) do not require additional secure association and per-packet ciphering. These networks can bind the PANA authentication and authorization to the lower-layer secure channel that is already available. illustrates the signaling flow for authorizing a client for network access.

PaC EP PAA AS | | | | IP address o | | | config. | PANA | | AAA | |<------------------------------>|<-------------->| | | SNMP | | (Optional) | |<-------------->| | IP address o | | | config. | Sec.Assoc. | | | |<------------->| | | | | | | | Data traffic | | | |<-----------------> | | | | | | The EP on the access network allows general data traffic from any authorized PaC, whereas it allows only limited type of traffic (e.g., PANA, DHCP, router discovery) for the unauthorized PaCs. This ensures that the newly attached clients have the minimum access service to engage in PANA and get authorized for the unlimited service. The PaC MUST configure an IP address prior to running PANA. After the successful PANA authentication, depending on the deployment scenario PaC MAY need to re-configure its IP address or configure additional IP address(es). The additional address configuration MAY be executed as part of the secure association protocol run. An initially unauthorized PaC starts the PANA authentication by discovering the PAA on the access network, followed by the EAP exchange over PANA. The PAA interfaces with the AS during this process. Upon receiving the authentication and authorization result from the AS, the PAA informs the PaC about the result of its network access request. If the PaC is authorized to gain the access to the network, the PAA also sends the PaC-specific attributes (e.g., IP address, cryptographic keys, etc.) to the EP by using SNMP. The EP uses this information to alter its filters for allowing data traffic from and to the PaC to pass through. In case a cryptographic access control needs to be enabled after the PANA authentication, a secure association protocol runs between the PaC and the EP. The PaC should already have the input parameters to this process as a result of the successful PANA exchange. Similarly, the EP should have obtained them from the PAA via SNMP. Secure association exchange produces the required security associations between the PaC and the EP to enable per-packet protection. Finally data traffic can start flowing from and to the newly authorized PaC.

The PANA protocol can be used on any network environment whether there is a lower-layer secure channel prior to PANA, or one has to be enabled upon successful PANA authentication. The secure channel enabled by PANA may rely on link-layer ciphering or IP-layer ciphering. Two types of networks are: These are the networks where the lower-layer is already providing protection against spoofing and eavesdropping (nevertheless the client is still required to get authenticated and authorized for the network access service). One example is the DSL networks where the lower-layer security is provided by physical means (a.1). Physical protection of the network wiring ensures that practically there is only one client that can send and receive IP packets on the link. Another example is the cdma2000 networks where the lower-layer security is provided by means of cryptographic protection (a.2). By the time the client requests access to the network-layer services, it is already authenticated and authorized for accessing the radio channel, and link-layer ciphering is enabled. The presence of a secure channel before PANA exchange eliminates the need for executing a secure association protocol after PANA. The PANA session can be bound to the communication channel it was carried over. Also, the choice of EAP authentication method depends on the presence of this security during PANA run. Use of some authentication methods outside a secure channel is not recommended (e.g., EAP-MD5). These are the networks where there is no lower-layer protection prior to running PANA. A successful PANA authentication enables generation of cryptographic keys that are used with a secure association protocol to enable per-packet protection. PANA authentication is run on an insecure channel that is vulnerable to eavesdropping and spoofing. The choice of EAP method must be resilient to the possible attacks associated with such an environment. Furthermore, the EAP method must be able to create cryptographic keys that will later be used by the secure association protocol. Whether to use a link-layer per-packet security (b.1) or an IP-layer one (b.2) is a deployment decision. This decision also dictates the choice of the secure association protocol. If link-layer protection is used, the protocol would be link technology specific. If IP-layer protection is used, the secure association protocol would be IKE and the per-packet security would be provided by IPsec regardless of the link type.

PaC configures an IP address before the PANA protocol exchange begins. This address is called a pre-PANA address (PRPA). After a successful authentication, the client may have to configure a post-PANA address (POPA) for communication with other nodes, if PRPA is a local-use (e.g., link-local or private address) or a temporarily allocated IP address. An operator might choose allocating a POPA only after successful PANA authorization either to prevent waste of premium IP resources until the client is authorized, or to enable client identity based address assignment. In case PaC is a IPv4-IPv6 dual-stack host, it may configure more than one PRPA, at least one for each protocol. Nevertheless, the PaC needs only one to run PANA which is executed once regardless of the type and number of IP stacks. After successful PANA authentication, similarly PaC may configure multiple POPAs. There are different methods by which a PRPA can be configured. In some deployments (e.g., DSL networks) the PaC may be statically configured with an IP address. This address SHOULD be used as PRPA. In IPv4, most clients attempt to configure an address dynamically using DHCP . If they are unable to configure an address using DHCP, they can configure a link-local address using . When the network access provider is able to run a DHCP server on the access link, the client would configure the PRPA using DHCP. This address may be from a private address pool . Also, the lease time on the address may vary. For example, a PRPA configured solely for running PANA can have a short lease time. PRPA may be used for local-use only (i.e., only for on-link communication, such as for PANA and IPsec tunneling with EP), or also for ultimate data communication. In case there is no running DHCP server on the link, the client would fall back to configuring a PRPA via zeroconfiguration technique . This yields a long-term address that can only be used for on-link communication. In IPv6, clients configure a link-local address when they initialize an interface. This address SHOULD be used as PRPA. When a PRPA is configured, the client starts the PANA protocol exchange. By that time, a dual-stack client might have configured both an IPv4 address, and one or more IPv6 addresses as PRPAs. When the client successfully authenticates to the network, it may be required to configure POPAs for its subsequent data communication with the other nodes. If the client is already configured with a non-temporary address that can be used with data communication, it is not required to configure a POPA. Otherwise, PAA would indicate PaC the available POPA configuration methods through the PANA-Bind-Request message. PaC can choose one of the methods and execute accordingly. If the network relies on physical or link-layer security, PaC can configure a POPA using DHCP or using IPv6 stateless auto-configuration . If IPv4 is being used, PRPA is likely to be a link-local or private address, or an address with a short DHCP lease. An IPv4 PRPA SHOULD be unconfigured when the POPA is configured to prevent IPv4 address selection problem . If IPv6 is used, the link-local PRPA address SHOULD NOT be unconfigured . If the PaC is a dual-stack host, it can configure both IPv4 and IPv6 type POPAs. PRPA and the IP address of PAA are assumed to be on the same IP subnet, hence the PaC and PAA can perform on-link communication as required by PANA. When the PaC replaces its IPv4 PRPA with POPA, this assumption may not hold true. In order to maintain the same on-link communication, the PaC and PAA SHOULD create host routes to each other upon PaC's IP address change. The PaC SHOULD create a host route to the PAA, and the PAA SHOULD create a host route to the PaC (POPA). If the network uses IPsec for protecting the traffic on the link subsequent to PANA authentication , PaC would use the PRPA as the outer address of IPsec tunnel mode SA (IPsec-TOA). PaC also needs to configure an inner address (IPsec-TIA). There are different ways to configure IPsec-TIA which are indicated in Pana-Bind-Request message. When an IPv4 PRPA is configured, the same address may be used as both IPsec-TOA and IPsec-TIA. In this case, a POPA is not configured. Alternatively, an IPsec-TIA can be obtained via the configuration method available within for IPv4, and for both IPv4 and IPv6. This newly configured address constitutes a POPA. Please refer to for more details. IKEv2 can enable configuration of both IPv4 and IPv6 TIAs for the same IPsec tunnel mode SA. Therefore, IKEv2 is recommended for handling dual-stack PaCs where single execution of PANA and IKE is desired. Although there are potentially a number of different ways to configure a PRPA, and POPA when necessary, it should be noted that the ultimate decision to use one or more of these in a deployment depends on the operator. The decision is dictated by the operator's choice of per-packet protection capability (physical and link-layer vs network-layer), PRPA type (local and temporary vs global and long-term), and POPA configuration mechanisms available in the network.

Protecting data traffic of authenticated and authorized client from others is another component of providing a complete secure network access solution. Authentication, integrity and replay protection of data packets are needed to prevent spoofing when the underlying network is not physically secured. Encryption is needed when eavesdropping is a concern in the network. When the network is physically secured, or the link-layer ciphering is already enabled prior to PANA, data traffic protection is already in place. In other cases, enabling link-layer ciphering or network-layer ciphering might rely on PANA authentication. The user and network have to make sure an appropriate EAP method that can generate required keying materials is used. Once the keying material is available, it needs to be provided to the EP(s) for use with ciphering. Network-layer ciphering, i.e., IPsec, can be used when data traffic protection is required but link-layer ciphering capability is not available. Note that a simple shared secret generated by an EAP method is not readily usable by IPsec for authentication and encryption of IP packets. Fresh and unique session key derived from the EAP method is still insufficient to produce an IPsec SA since both traffic selectors and other IPsec SA parameters are missing. The shared secret can be used in conjunction with a key management protocol like IKE to turn a simple shared secret into the required IPsec SA. The details of such a mechanism is outside the scope of PANA protocol and is specified in . PANA provides bootstrapping functionality for such a mechanism by carrying EAP methods that can generate initial keying material. Using network-layer ciphers should be regarded as a substitute for link-layer ciphers when the latter is not available. Network-layer ciphering can also be used in addition to link-layer ciphering if the added benefits outweigh its cost to the user and the network. In this case, PANA bootstraps only the network-layer ciphering and link-layer is protected using any of the existing link-layer specific methods.

The PANA protocol provides client authentication and authorization functionality for securing network access. The other component of a complete solution is the access control which ensures that only authenticated and authorized clients can gain access to the network. PANA enables access control by identifying legitimate clients and generating filtering information for access control mechanisms. Access control can be achieved by placing EPs (Enforcement Points) in the network for policing the traffic flow. EPs should prevent data traffic from and to any unauthorized client unless it's either PANA or one of the other allowed traffic types (e.g., ARP, IPv6 neighbor discovery, DHCP, etc.). When a client is authenticated and authorized, PAA should notify EP(s) and ask for changing filtering rules to allow traffic for a recently authorized client. There needs to be a protocol between PAA and EP(s) when these entities are not co-located. PANA Working Group will not be defining a new protocol for this interaction. Instead, it identifies one of the existing protocols that can fit the requirements. An assessment was made in the PANA Working Group and SNMPv3 has been chosen as the mandatory PAA-EP protocol. This section describes the possible models on the location of PAA and EP, as well as the basic authorization information that needs to be exchanged between PAA and EP. When PAA and EP are not co-located in a single device, there are other issues such as dead or rebooted peer detection and consideration for specific authorization and accounting models. However, these issues are closely related to the PAA-EP protocol solution and thus not discussed in this document. See for further discussion.

EPs' location in the network topology should be appropriate for performing access control functionality. The closest IP-capable access device to the client devices is the logical choice. PAA and EPs on an access network should be aware of each other as this is necessary for access control. Generally this can be achieved by manual configuration. Dynamic discovery is another possibility, but this is left to implementations and outside the scope of this document. Since PANA allows the separation of EP and PAA, there are several models depending on the number of EPs and PAAs and their locations. This section describes all possible models on the placement of PAA(s) and EP(s).

This model corresponds to the legacy NAS model. Since the PAA and the EP are co-located, the PAA-EP communication can be implemented locally by using, e.g., IPC (Inter-Process Communication). The only difference from the legacy NAS model is the case where there are multiple co-located PAA/EP devices on the same IP link and the PAA/EP devices are L2 switches or access points. In this case, for a PaC that attaches to a given PAA/EP device, other PAA/EP devices should not be visible to the PaC even if those devices are on the same IP link. Otherwise, the PaC may result in finding a PAA that is not the closest one to it during the PANA discovery and initial handshake phase and performing PANA with the wrong PAA. To prevent this, each PAA/EP device on an L2 switch or access point should not forward multicast PANA discovery message sent by PaCs attached to it to other devices.

When PAA is separated from EP, two cases are possible with regard to whether PAA and EP are located in parallel or serial when viewed from PaC, for each of models described in this section. In the first case, PAA is located behind EP. The EP should be configured to always pass through PANA messages and address configuration protocol messages used for configuring an IP address used for initial PANA messaging. This case can typically happen when the EP is an L2 switch or an access point (the EP also has an IP stack to communicate with PAA via a PAA-EP protocol).

+---+ +---+ +---+ |PaC|--------|EP |--------|PAA| +---+ +---+\ +---+ \ +---- Internet In the other case, PAA is located in parallel to EP. Since the EP is not on the communication path between PaC and PAA, the EP does not have to configure to pass through PANA messages or address configuration protocol messages in this case. This case can typically happen when the EP is a router and the PAA is an authentication gateway without IP routing functionality.

+---+ +-----|PAA| / +---+ +---+/ |PaC| +---+\ \ +---+ +-----|EP |---- Internet +---+ In the remaining of this section, PaC is not shown in figures and the figures cover both the serial and parallel models.

The model benefits from separation of data traffic handling and AAA. The EP does not need to have a AAA protocol implementation which might be updated relatively more frequently than the per-packet access enforcement implementation.

In this model, a single PAA controls multiple EPs. The PAA may be separated from any EP or co-located with a particular EP. This model might be useful where it is preferable to run a AAA protocol at a single, manageable point. This model is particularly useful in an access network that consists of a large number of access points on which per-packet access enforcement is made. When a PaC is authenticated to the PAA, the PAA should install access control filters to each of the EPs under control of the PAA if the PAA cannot tell which EP the PaC is attached to. Even if the PAA can tell which EP the PaC is attached to, the PAA may install access control filters to those EPs if the PaC is a mobile device that can roam among the EPs. Such pre-installation of filters can reduce handoff latency. If different access authorization policies are applied to different EPs, different filter rules for a PaC may be installed on different EPs.

+---+ |EP |--+ +---+ \ \ +---+ +---+ |EP |----|PAA| +---+ +---+ / +---+ / |EP |--+ +---+

The PANA protocol allows multiple PAAs to exist on the same IP link and to be visible to PaCs on the link. PAAs may or may not be co-located with EPs as long as authorization results do not depend on whichever PAAs are chosen by a PaC .

When PAA and EP are separated and PAA is configured to be the initiator of the discovery and initial handshake phase of PANA, EP has the responsibility to detect presence of a new PaC and notifies the PAA(s) of the presence . Such a presence notification is carried in a PAA-EP protocol message [I-D.ietf-pana-snmp].

Filtering rules to be installed on EP generally include a device identifier of PaC, and also cryptographic keying material (such as keys, key pairs, and initialization values) when needed. The keying material is needed when attackers can eavesdrop and spoof on the device identifiers easily. Each keying material is uniquely identified with a keying material name and has a lifetime for key management purpose. The keying material is used with link-layer or network-layer ciphering to provide additional protection. For issues regarding data-origin authentication see . In addition to the device identifier and keying material, other filter rules, such as the IP filter rules specified in NAS-Filter-Rule AVPs carried in Diameter EAP application may be installed on EP.

The network selection problem statement is made in . The PANA protocol provides a way for networks to advertise which ISPs are available and for PaC to choose one ISP from the advertised information. When a PaC chooses an ISP in the PANA protocol exchange, the ultimate destination of the AAA exchange is determined based on the identity of the chosen service provider. It is also possible that the PaC does not choose a specific ISP in the PANA protocol exchange. In this case, both the ISP choice and the AAA destination are determined based on the PaC's identity, where the identity may be an NAI or the physical port number or L2 address of the subscriber. As described in , network selection is not only related to AAA routing but also related to payload routing. Once an ISP is chosen and the PaC is successfully authenticated and authorized, PaC is assigned an address by the ISP whose IP prefix may be different from that of the AR. This affects the routing of the subsequent data traffic between AR and PaC. A suggested solution is to add host route from AR to PaC's POPA address and host route from PaC to AR. Consider a typical DSL network where the AR, EP, and PAA are co-located on a BAS (Broadband Access Server) in the access network operated by a NAP (Network Access Provider). shows a typical model for ISP selection.

<---- NAP ----><--------- ISP ---------> +---ISP1 / +---+ +---------+/ |PaC|----|AR/EP/PAA| +---+ +---------+\ BAS \ +---ISP2 When network selection is made at L3 with the use of the PANA protocol instead of L2-specific authentication mechanisms, the IP link between PaC and PAA needs to exist prior to doing PANA (and prior to network selection). In this model, the PRPA is either given by NAP or a link-local address is auto-configured. After the successful authentication with the ISP, PaC may acquire an address (POPA) from the ISP. It also learns the address of the AR, e.g., through DHCP, to be used as its default router. The address of the AR may or may not be in the same IP subnet as that of the PaC's POPA. If they don't share the same prefix, they SHOULD use host routes to reach each other. Note that the physically secured DSL networks do not require IPsec-based access control. Therefore the PaCs use one IP address at a time where POPA replaces PRPA upon configuration.

Authentication methods' capabilities and therefore applicability to various environments differ among them. Not all methods provide support for mutual authentication, key derivation or distribution, and DoS attack resiliency that are necessary for operating in insecure networks. Such networks might be susceptible to eavesdropping and spoofing, therefore a stronger authentication method needs to be used to prevent attacks on the client and the network. The authentication method choice is a function of the underlying security of the network (e.g., physically secured, shared link, etc.). It is the responsibility of the user and the network operator to pick the right method for authentication. PANA carries EAP regardless of the EAP method used. It is outside the scope of PANA to mandate, recommend, or limit use of any authentication methods. PANA cannot increase the strength of a weak authentication method to make it suitable for an insecure environment. There are some EAP-based approaches to achieve this goal (see ,,). PANA can carry these EAP encapsulating methods but it does not concern itself with how they achieve protection for the weak methods (i.e., their EAP method payloads).

In a DSL access network, PANA is seen applicable in the following scenarios. A typical DSL access consists of a NAS device at the DSL-access provider and a DSL-modem (CPE) at the customer premises. The CPE devices support multiple modes of operation and PANA is applicable in each of these modes.

Host--+ +-------- ISP1 | DSL link | +----- CPE ---------------- NAS ----+-------- ISP2 | (Bridge/NAPT/Router) | Host--+ +-------- ISP3 <------- customer --> <------- NAP -----> <---- ISP ---> premise The devices at the customer premises have been shown as "hosts" in the above network. DSL networks are protected by physical means. Eavesdropping and spoofing attacks are prevented by keeping the unintended users physically away from the network media. Therefore, generally cryptographic protection of data traffic is not necessary. Nevertheless, if enhanced security is deemed necessary for any reason, IPsec-based access control can be enabled on DSL networks as well by using the method described in .

In the bridging mode, the CPE acts as a simple layer-2 bridge. The hosts at the customer premises will function as clients to obtain addresses from the NAS device by using DHCP or PPPoE. If PPPoE is used, authentication is typically performed using CHAP or MS-CHAP. PANA will be applicable when the hosts use DHCP to obtain IP address. DHCP does not support authentication of the devices on either side of the DSL access line. In the simplest method of address assignment, the NAS will allocate the IP address to a host with a lease time reasonably sufficient to complete a full PANA based authentication which will be triggered immediately after the address assignment. The hosts will perform the PaC function and the NAS will perform the PAA, EP and AR functions.

Host--+ (PaC) | +----- CPE ---------------- NAS ------------- ISP | (Bridge) (PAA,EP,AR) Host--+ (PaC) The DSL service provider's trunk network should not be accessible to any host that has not successfully completed the PANA authentication phase.

In the NAPT mode, the CPE is configured with the authentication parameters (i.e. username, password). In this case, the CPE acts as a "client" of the NAS. If the CPE uses DHCP to obtain the IP address, PANA will be necessary to authenticate the CPE and the NAS to each other.

Host--+ | +----- CPE ---------------- NAS ------------- ISP | (NAPT, PaC) (PAA,EP,AR) Host--+ The CPE in this case typically acts as a DHCP server for the devices at the customer premises network. It applies NAPT mechanisms to forward traffic between the customer premises network and the NAS. If the CPE is configured with a static IP address and does not perform DHCP, it will need to initiate a PANA authentication phase before gaining access to the service provider's network.

In this case, the CPE acts as a full-fledged router for the customer premises network. The CPE itself may obtain the IP address using DHCP or be configured with static IP address. Once the CPE is authenticated using PANA and is provided access to the service provider's network, the NAS should begin exchanging routing updates with the CPE. All devices at the customer premises will then have access to the service provider's network.

Host--+ | +----- CPE ---------------- NAS ------------- ISP | (Router, PaC) (PAA,EP,AR) Host--+ As in the NAPT mode, it is possible that both ends of the DSL link are configured with static IP addresses. PANA-based mutual authentication of CPE and NAS is desirable before data-traffic is exchanged between the customer premises network and the service provider network.

In some installations, a NAS device is shared by multiple service providers. Each service provider configures the NAS with a certain IP address space. The devices at the customer premises network indicate their choice of service provider and the NAS chooses the IP address from the appropriate service provider's pool. In many cases, the address is assigned not by the NAS but by the AAA server that is managed fully by the service provider. This simplifies the management of the DSL access network as it is not always necessary to configure each DSL access line with the service provider's identity. The service provider is chosen dynamically by the CPE device. This is typically known as "dynamic Internet Service Provider selection". The AAA function is usually overloaded to perform dynamic ISP selection. If the CPE device uses a PPP based protocol (PPP or PPPoE), the ISP is chosen by mapping the username field of a CHAP response to a provider. If the CPE uses DHCP, the 'client-id' field of the DHCP-discover or DHCP-request packet is mapped to the provider.

The ISP selection, therefore the IP address pool, can be conveyed based on the DHCP protocol exchange (as explained earlier), or by associating the DSL access line to the service provider before the PANA authentication begins. When any of these schemes is used, the IP address used during PANA authentication (PRPA) is the ultimate IP address and it does not have to be changed upon successful authorization.

The ISP selection of the client can be explicitly conveyed during the PANA authentication. In that case, the client can be assigned a temporary IP address (PRPA) prior to PANA authentication. This IP address might be obtained via DHCP with a lease reasonably long to complete PANA authentication, or via the zeroconf technique . In either case, successful PANA authentication signalling prompts the client to obtain a new (long term) IP address via DHCP. This new IP address (POPA) replaces the previously allocated temporary IP address.

This section describes how PANA can be used on WLAN networks. In most common WLAN deployments the IP addresses are dynamically configured. Therefore this section does not cover the scenarios where the IP address is statically configured. There are two models depending on which layer security is bootstrapped from PANA authentication, L2 or L3. When PANA authentication is used for bootstrapping L3 security, L2 security is not necessarily to exist even after PANA authentication. Instead, IPsec-based data traffic protection is bootstrapped from PANA. The PAA can indicate the PaC as to whether L2 or L3 protection is needed, by including a Protection-Capability AVP in PANA-Bind-Request message. In both cases, the most common deployment would be illustrated in , where EP is typically co-located with AP (access point) when PANA is used for bootstrapping L2 security or with AR when PANA is used for bootstrapping L3 security.

+-----+ |AP/EP|----+ +-----+ | | +---+ +-----+ | +---------+ |PaC| |AP/EP|----+----|AR/PAA/EP|----- Internet +---+ +-----+ | +---------+ | +-----+ | |AP/EP|----+ +-----+

In this model, data traffic is protected by using IPsec tunnel mode SA and an IP address is used as the device identifier of PaC (see for details). Some or all of AP, DHCPv4 Server (including PRPA DHCPv4 Server and IPsec-TIA DHCPv4 Server), DHCPv6 Server, PAA and EP may be co-located in a single device. EP is always co-located with AR and may be co-located with PAA. When EP and PAA are not co-located, PAA-EP protocol is used for communication between PAA and EP. Note that for all of the cases described in this section, PBR (PANA-Bind-Request) and PBA (PANA-Bind-Answer) exchange in PANA should occur after installing the authorization parameter to AR, so that IKE can be performed immediately after PANA is successfully completed.

In this case, the IPsec-TIA and IPsec-TOA are the same as the PRPA, and all configuration information including the IP address is obtained by using DHCPv4 . No POPA is configured. Case A is the simplest compared to other ones and might be used in a network where IP address depletion attack on DHCP is not a significant concern. The PRPA needs to be a routable address unless NAT is performed on AR.

PaC AP DHCPv4 Server PAA EP(AR) | Link-layer | | | | | association| | | | |<---------->| | | | | | | | | | DHCPv4 | | | |<-----------+------------>| | | | | | | | |PANA(Discovery and Initial Handshake phase | | & PAR-PAN exchange in Authentication phase) | |<-----------+-------------------------->| | | | | | | | | | |Authorization| | | | |[IKE-PSK, | | | | | PaC-DI, | | | | | Session-Id] | | | | |------------>| | | | | | |PANA(PBR-PBA exchange in Authentication phase) | |<-----------+-------------------------->| | | | | | | | | IKE | | |<-----------+---------------------------------------->| | | | | | | | | | | In this case, the PRPA is obtained by using DHCPv4 and used as IPsec-TOA. The POPA is obtained by using IKE (via a Configuration Payload exchange or equivalent) and used as IPsec-TIA. Like Case B, the PRPA is obtained by using DHCPv4. The difference is that the POPA (eventually used as IPsec-TIA) and other configuration parameter are configured by running DHCPv4 over a special IPsec tunnel mode SA . Note that the PRPA DHCPv4 Server and IPsec-TIA DHCPv4 Server may be co-located on the same node. Note: this case may be used only when IKEv1 is used as the IPsec key management protocol (IKEv2 does not seem to support RFC3456 equivalent case).

PRPA PaC AP DHCPv4 Server PAA | Link-layer | | | | association| | | |<---------->| | | | | | | | DHCPv4 | | |<-----------+-------->| | | | | |PANA(Discovery and initial handshake phase | & PAR-PAN exchange in authentication phase) |<-----------+-------------------------->| | | | | | EP(AR) | | | |Authorization| | | |[IKE-PSK, | | | | PaC-DI, | | | | Session-Id] | | | |<------------| | | | | |PANA(PBR-PBA exchange in authentication phase) |<-----------+-------------------------->| | | | | IKEv1 phase I & II | | (to create DHCP SA) | |<-----------+------------>| | | | | DHCP over DHCP SA | |<-----------+------------>| | | | | IKEv1 phase II | | (to create IPsec SA for data traffic) |<-----------+------------>|

In the case of IPv6, the IPsec-TOA (PRPA) is the IPv6 link-local address. IPsec-TIA (POPA) is obtained by using Configuration Payload exchange of IKE version 2 (Note that there is no standard method for configuring IPsec-TIA in IKEv1). Other configuration information may be obtained in the same Configuration Payload exchange or may be obtained by running an additional DHCPv6.

PaC AP EP(AR) PAA | Link-layer | | | | association| | | |<---------->| | | | | | | | | | | |PANA(Discovery and Initial Handshake phase | & PAR-PAN exchange in Authentication phase) |<-----------+-------------------------->| | | | | | | | | | | |Authorization| | | |[IKE-PSK, | | | | PaC-DI, | | | | Session-Id] | | | |<------------| | | | | |PANA(PBR-PBA exchange in authentication phase) |<-----------+-------------------------->| | | | | | IKEv2 | | |(w/Configuration Payload | | | exchange to obtain IPsec-TIA) | |<-----------+------------>| | | | | |

In this model, PANA is used for authentication and authorization, and L2 ciphering is used for access control, the latter is enabled by the former. The L2 ciphering is based on using PSK (Pre-Shared Key) mode of WPA (Wi-Fi Protected Access) or IEEE 802.11i , which is derived from the EAP MSK as a result of successful PANA authentication. In this document, the pre-shared key shared between station and AP is referred to as PMK (Pair-wise Master Key). In this model, MAC address is used as the device identifier in PANA. A typical purpose of using this model is to reduce AP management cost by allowing physical separation of RADIUS/Diameter client from access points, where AP management can be a significant issue when deploying a large number of access points. By bootstrapping PSK mode of WPA and IEEE 802.11i from PANA it is also possible to improve wireless LAN security by providing protected disconnection procedure at L3. This model does not require any change in the current WPA and IEEE 802.11i specifications. This also means that PANA doesn't provide any L2 security features beyond those already provided for in WPA and IEEE 802.11i.

When PAA and AP are separated, two (virtual) APs are needed, one is connected to an unauthenticated VLAN and the other to an authenticated VLAN. The former and latter APs are referred to as unauthenticated VLAN AP and authenticated VLAN AP, respectively. If a PaC does not have a PMK with the authenticated VLAN AP, it runs PANA on the unauthenticated VLAN to obtain the MAC address of the authenticated VLAN AP and derive the PMK valid for the AP. Once the PMK is obtained, the PaC associates with the authenticated VLAN AP with performing 4-way handshake. It configures an (potentially new) IP address and starts sending and receiving data traffic. Future PANA exchanges are carried on the IEEE 802.1X Controlled Port of the authenticated VLAN AP just like any other data traffic. Separating PAA and AP may be used in a network where there is a large number of APs in a single IP subnet and the network administrator want to avoid setting up RADIUS client on each AP. Note that the network administrator would still need to set up SNMPv3 security association between each AP and PAA for secure PAA-EP communication, however, in the environments where operator can rely on physical layer security between PAA and EP, this might not be the case. In a typical usage scenario, a single PAA co-located with DHCP server is connected to both unauthenticated VLAN and authenticated VLAN and that the unauthenticated VLAN AP and the authenticated VLAN AP are co-located in a single physical AP, as shown in . Note that in an environment where virtual AP are not supported, physical APs can be used instead of virtual APs.

+------------------+ | Physical AP | | +--------------+ | | |Virtual AP1 | | Unauthenticated | |(open-access) |---- VLAN \ | | | | \+-------+ +---+ | +--------------+ | |PAA/AR/| |PaC| | | |DHCP |--- Internet +---+ | +--------------+ | |Server | | |Virtual AP2 | | /+-------+ | |(WPA PSK mode)|---- Authenticated / | | | | VLAN | +--------------+ | | | +------------------+ When a PaC does not have a PMK with an authenticated VLAN AP, the following procedure is taken: The PaC associates with the unauthenticated VLAN AP. The PaC configures a PRPA by using DHCP (in the case of IPv4) or configures a link-local address (in the case of IPv6), and then runs PANA by using the address. Upon successful authentication, the PaC obtains a distinct PMK for each authenticated VLAN AP controlled by the PAA. The PaC associated with an authenticated VLAN AP, with performing 4-way handshake to establish a PTK (Pair-wise Transient Key) between the PaC and AP, by using the PMK. If the PaC is required to configure a new IP address (POPA) as signaled by the PAA at the end of successful authentication, it should configure POPA by using any method that the client normally uses. If PRPA is an IPv4 address, it should be unconfigured before POPA configuration. Note that switching from one VLAN to another is a commonly used technique in WPA (even without PANA) where the client that does not have any valid credentials first accesses the certificate server via https through the unauthenticated VLAN to perform a sign-in procedure to obtain a certificate, and then switches to the authenticated VLAN with the obtained certificate. When the MSK is updated as a result of EAP re-authentication, a new PMK is derived for and transfered to each authenticated VLAN AP. WPA/IEEE 802.11i 4-way handshake is performed between the PaC and its currently associating authenticated VLAN AP to derive a new PTK.

The IEEE 802.11 specification allows Class 1 data frames to be received in any state. Also, the latest version of IEEE 802.11i optionally allows higher-layer data traffic that is terminated between stations (including AP) are received and processed on their IEEE 802.1X Uncontrolled Ports. This feature allows processing IP-based traffic (such as ARP, IPv6 neighbor discovery, DHCP, and PANA) on IEEE 802.1X Uncontrolled Port prior to client authentication. (Note: WPA and its corresponding version of IEEE 802.11i draft do not explicitly define this operation, so it may be safer not to use this co-located PAA and AP case in WPA). Until the PaC is successfully authenticated, only a selected type of IP traffic is allowed over the IEEE 802.1X Uncontrolled Port. Any other IP traffic is dropped on the AP without forwarding to the DS (Distribution System). Upon successful PANA authentication, the traffic switches to the controlled port. Host configuration, including obtaining an (potentially new) IP address, takes place on this port. Usual DHCP-based, and also in the case of IPv6 stateless autoconfiguration, mechanism is available to the PaC. After this point, the rest of the IP traffic, including PANA exchanges, are processed on the controlled port. Since this case does not require virtual AP switching, the procedure to bootstrap PSK mode becomes simpler compared to the separate PAA and EP case, however, with a cost of configuring RADIUS/Diameter client on each AP. When a PaC does not have a PMK with an authenticated VLAN AP, the following procedure is taken: The PaC associates with the AP. The PaC configures a PRPA by using DHCP (in the case of IPv4) or configures a link-local address (in the case of IPv6), and then runs PANA by using the address. Upon successful authentication, the PaC obtains a PMK for each AP controlled by the PAA. The PaC performs 4-way handshake to establish a PTK (Pair-wise Transient Key) between the PaC and AP, by using the PMK. The PaC obtains a POPA by using any method that the client normally uses.

When a PaC is a mobile, there may be multiple APs available in its vicinity. Each AP are connected to one of the following types of access networks. There is no IEEE 802.1X or PANA authentication in this access network. There is PANA authentication in this access network. There is IEEE 802.1X authentication in this access network. Type (c) is distinguished from others by checking the capability information advertised in IEEE 802.11 Beacon frames (IEEE 802.11i defines RSN Information Element for this purpose). Types (a) and (b) are not distinguishable until the PaC associates with the AP, get an IP address, and engage in PANA discovery. The default PaC behavior would be to act as if this is a free network and attempt DHCP. This would be detected by the access network and trigger unsolicited PANA discovery. A type (b) network would send a PANA-Start-Request to the client and block general purpose data traffic. This helps the client discover whether the network is type (a) or type (b). Or if the PaC is pre-provisioned with the information that this is a PANA enabled network, it can attempt PAA discovery immediately. The PaC behavior after connecting to an AP of type (b) network is described in , and . Note that in case of using PANA for bootstrapping WPA/IEEE 802.11i PSK mode (see ), PaC needs to know whether PAA and AP are separated (see ) or co-located (see ) due to PaC having to act differently in each mode. This is achieved by checking the existence of an EP-Device-Id AVP in PANA-Bind-Request message (if an EP-Device-Id AVP is included then PAA is not co-located with AP).

Certain combination of network environments and timing cases may require further considerations. If PaC changes access point right after a successful PANA authorization but before POPA configuration, it might be confused whether the new IP address configured via the new access point is the POPA of the ongoing session or a PRPA of a new session. When the PRPA and POPA types or configuration mechanisms are different this is not a problem. One possible combination where such clues are not available is when PaC moves from a Case A of (IPsec-TIA obtained by using DHCPv4) network to a Case D of (IPsec-TIA obtained by using RFC3118) network. In another example, when PaC switches from one wired Ethernet hub to another (without the use of bootstrapping IPsec) before configuring the POPA. In this case, PaC may not able to know whether an obtained address is the POPA in the same subnet or a PRPA in a new subnet. For these cases, a mechanism to remove the ambiguity (PRPA vs. POPA) may need to be defined.

We would like to thank Bernard Aboba and Yacine El Mghazli for their valuable comments.