Handover Key Management and Re-Authentication Problem Statement

The Extensible Authentication Protocol (EAP), specified in RFC 3748 is a generic framework supporting multiple authentication methods. The primary purpose of EAP is network access control. It also supports exporting session keys derived during the authentication. The EAP keying hierarchy defines two keys that are derived at the top level, the Master Session Key (MSK) and the Extended Master Session Key (EMSK). In many common deployment scenarios, an EAP peer and EAP server authenticate each other through a third party known as the pass-through authenticator (hereafter referred to as simply "authenticator"). The authenticator is responsible for encapsulating EAP packets from a network-access technology lower layer within the Authentication, Authorization, and Accounting (AAA) protocol. The authenticator does not directly participate in the EAP exchange, and simply acts as a gateway during the EAP method execution. After successful authentication, the EAP server transports the MSK to the authenticator. Note that this is performed using AAA protocols, not EAP itself. The underlying L2 or L3 protocol uses the MSK to derive additional keys, including the transient session keys (TSKs) used for per-packet encryption and authentication. Note that while the authenticator is one logical device, there can be multiple physical devices involved. For example, the CAPWAP model splits authenticators into two logical devices: Wireless Termination Points (WTPs) and Access Controllers (ACs). Depending on the configuration, authenticator features can be split in a variety of ways between physical devices; however, from the EAP perspective, there is only one logical authenticator. Wireless handover between access points or base stations is typically a complex process that involves several layers of protocol execution. Often times executing these protocols results in unacceptable delays for many real-time applications such as voice . One part of the handover process is EAP re-authentication, which can contribute significantly to the overall handover time . Thus, in many environments we can lower overall handover time by lowering EAP re-authentication time. In EAP existing implementations, when a peer arrives at the new authenticator, it runs an EAP method irrespective of whether it has been authenticated to the network recently and has unexpired keying material. This typically involves an EAP-Response/Identity message from the peer to the server, followed by one or more round trips between the EAP server and peer to perform the authentication, followed by the EAP-Success or EAP-Failure message from the EAP server to the peer. At a minimum, the EAP exchange consists of 1.5 round trips. However, given the way EAP interacts with AAA, and given that an EAP identity exchange is typically employed, at least 2 round trips are required to the EAP server. An even higher number of round trips is required by the most commonly used EAP methods. For instance, EAP-TLS (Extensible Authentication Protocol - Transport Layer Security) requires at least 3, but typically 4 or more, round trips. There have been attempts to solve the problem of efficient re-authentication in various ways. However, those solutions are either EAP-method specific or EAP lower-layer specific. Furthermore, these solutions do not deal with scenarios involving handovers to new authenticators, or they do not conform to the AAA keying requirements specified in . This document provides a detailed description of efficient EAP-based re-authentication protocol design goals. The scope of this protocol is specifically re-authentication and handover between authenticators within a single administrative domain. While the design goals presented in this document may facilitate inter-technology handover and inter-administrative-domain handover, they are outside the scope of this protocol.

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 , with the qualification that, unless otherwise stated, they apply to the design of the re-authentication protocol, not its implementation or application. With respect to EAP, this document follows the terminology that has been defined in and .

Under the existing model, any re-authentication requires a full EAP exchange with the EAP server to which the peer initially authenticated . This introduces handover latency from both network transit time and processing delay. In service provider networks, the home EAP server for a peer could be on the other side of the world, and typical intercontinental latencies across the Internet are 100 to 300 milliseconds per round trip . Processing delays average a couple of milliseconds for symmetric-key operations and hundreds of milliseconds for public-key operations. An EAP conversation with a full EAP method run can take two or more round trips to complete, causing delays in re-authentication and handover times. Some methods specify the use of keys and state from the initial authentication to finish subsequent authentications in fewer round trips and without using public-key operations (detailed in Section 6.1). However, even in those cases, multiple round trips to the EAP server are required, resulting in unacceptable handover times. In summary, it is undesirable to run an EAP Identity and complete EAP method exchange each time a peer authenticates to a new authenticator or needs to extend its current authentication with the same authenticator. Furthermore, it is desirable to specify a method-independent, efficient, re-authentication protocol. Keying material from the initial authentication can be used to enable efficient re-authentication. It is also desirable to have a local server with low-latency connectivity to the peer that can facilitate re-authentication. Lastly, a re-authentication protocol should also be capable of supporting scenarios where an EAP server passes authentication information to a remote re-authentication server, allowing a peer to re-authenticate locally, without having to communicate with its home re-authentication server. These problems are the primary issues to be resolved. In solving them, there are a number of constraints to conform to, and those result in some additional work to be done in the area of EAP keying.

The following are the goals and constraints in designing the EAP re-authentication and key management protocol: The protocol MUST be responsive to handover and re-authentication latency performance objectives within a mobile access network. A solution that reduces latency as compared to a full EAP authentication will be most favorable, since in networks that rely on reactive re-authentication this will directly impact handover times. Any keying hierarchy and protocol defined MUST be lower-layer independent in order to provide capabilities over heterogeneous technologies. The defined protocols MAY require some additional support from the lower layers that use it, but should not require any particular lower layer. Changes to existing EAP methods MUST NOT be required as a result of the re-authentication protocol. There MUST be no requirements imposed on future EAP methods, provided they satisfy and . Note that the only EAP methods for which independence is required are those that currently conform to the specifications of and . In particular, methods that do not generate the keys required by need not be supported by the re-authentication protocol. Any modifications to EAP and EAP keying MUST be compatible with RADIUS and Diameter . Extensions to both RADIUS and Diameter to support these EAP modifications are acceptable. The designs and protocols must be configurable to satisfy the AAA key management requirements specified in RFC 4962 . Compatibility and coexistence with compliant ( ) EAP deployments MUST be provided. Specifically, the protocol should be designed such that a peer not supporting fast re-reauthentication should still function in a network supporting fast re-authentication, and also a peer supporting fast re-authentication should still function in a network not supporting fast re-authentication. Any re-authentication protocol MUST support cryptographic algorithm agility, to avoid hard-coded primitives whose security may eventually prove to be compromised. The protocol MAY support cryptographic algorithm negotiation, provided it does not adversely affect overall performance (i.e., by requiring additional round trips). Any re-authentication protocol MAY make changes to the peer, authenticator, and EAP server, as necessary to meet the aforementioned design goals. In order to facilitate protocol deployment, protocols should seek to minimize the necessary changes, without sacrificing performance.

This section draws from the guidance provided in to further define the security goals to be achieved by a complete re-authentication keying solution.

Any key must have a well-defined scope and must be used in a specific context and for the intended use. This specifically means the lifetime and scope of each key must be defined clearly so that all entities that are authorized to have access to the key have the same context during the validity period. In a hierarchical key structure, the lifetime of lower-level keys must not exceed the lifetime of higher-level keys. This requirement may imply that the context and the scope parameters have to be exchanged. Furthermore, the semantics of these parameters must be defined to provide proper channel binding specifications. The definition of exact parameter syntax definition is part of the design of the transport protocol used for the parameter exchange, and that may be outside scope of this protocol. If a key hierarchy is deployed, compromising lower-level keys must not result in a compromise of higher-level keys that were used to derive the lower-level keys. The compromise of keys at each level must not result in compromise of other keys at the same level. The same principle applies to entities that hold and manage a particular key defined in the key hierarchy. Compromising keys on one authenticator must not reveal the keys of another authenticator. Note that the compromise of higher-level keys has security implications on lower levels. Guidance on parameters required, caching, and storage and deletion procedures to ensure adequate security and authorization provisioning for keying procedures must be defined in a solution document. All the keying material must be uniquely named so that it can be managed effectively.

As defines, a fresh key is one that is generated for the intended use. This would mean the key hierarchy must provide for creation of multiple cryptographically separate child keys from a root key at higher level. Furthermore, the keying solution needs to provide mechanisms for refreshing each of the keys within the key hierarchy.

Each handover keying participant must be authenticated to any other party with whom it communicates to the extent it is necessary to ensure proper key scoping, and securely provide its identity to any other entity that may require the identity for defining the key scope.

The EAP Key management document discusses several vulnerabilities that are common to handover mechanisms. One important issue arises from the way the authorization decisions might be handled at the AAA server during network access authentication. Furthermore, the reasons for making a particular authorization decision are not communicated to the authenticator. In fact, the authenticator only knows the final authorization result. The proposed solution must make efforts to document and mitigate authorization attacks.

Channel Binding procedures are needed to avoid a compromised intermediate authenticator providing unverified and conflicting service information to each of the peer and the EAP server. To support fast re-authentication, there will be intermediate entities between the peer and the back-end EAP server. Various keys need to be established and scoped between these parties and some of these keys may be parents to other keys. Hence, the channel binding for this architecture will need to consider layering intermediate entities at each level to make sure that an entity with a higher level of trust can examine the truthfulness of the claims made by intermediate parties.

Depending on the physical architecture and the functionality of the elements involved, there may be a need for multiple protocols to perform the key transport between entities involved in the handover keying architecture. Thus, a set of requirements for each of these protocols, and the parameters they will carry, must be developed. The use of existing AAA protocols for carrying EAP messages and keying material between the AAA server and AAA clients that have a role within the architecture considered for the keying problem will be carefully examined. Definition of specific parameters, required for keying procedures and for being transferred over any of the links in the architecture, are part of the scope. The relation between the identities used by the transport protocol and the identities used for keying also needs to be explored.

In order to further clarify the items listed in scope of the proposed work, this section provides some background on related work and the use cases envisioned for the proposed work.

A number of EAP methods support fast re-authentication. In this section, we examine their properties in further detail. EAP-SIM and EAP-AKA support fast re-authentication, bootstrapped by the keys generated during an initial full authentication. In response to the typical EAP-Request/Identity, the peer sends a specially formatted identity indicating a desire to perform a fast re-authentication. A single round-trip occurs to verify knowledge of the existing keys and provide fresh nonces for generating new keys. This is followed by an EAP success. In the end, it requires a single local round trip between the peer and authenticator, followed by another round trip between the peer and EAP server. AKA is based on symmetric-key cryptography, so processing latency is minimal. EAP-TTLS and PEAP (Protected EAP Protocol) support using TLS session resumption for fast re-authentication. During the TLS handshake, the client includes the message ID of the previous session he wishes to resume, and the server can echo that ID back if it agrees to resume the session. EAP-FAST also supports TLS session resumption, but additionally allows stateless session resumption as defined in . Overall, for all three protocols, there are still two round trips between the peer and EAP server, in addition to the local round trip for the Identity request and response. To improve performance, fast re-authentication needs to reduce the number of overall round trips. Optimal performance could result from eliminating the EAP-Request/Identity and EAP-Response/Identity messages observed in typical EAP method execution, and allowing a single round trip between the peer and a local re-authentication server.

One of the EAP lower layers, IEEE 802.11 , is in the process of specifying a fast handover mechanism. Access Points (APs) are grouped into mobility domains. Initial authentication to any AP in a mobility domain requires execution of EAP, but handover between APs within the mobility domain does not require the use of EAP. Internal to the mobility domain are sets of security associations to support key transfers between APs. In one model, relatively few devices, called R0-KHs, act as authenticators. All EAP traffic traverses an R0-KH, and it derives the initial IEEE 802.11 keys. It then distributes cryptographically separate keys to APs in the mobility domain, as necessary, to support the client mobility. For a deployment with M designated R0-KHs and N APs, this requires M*N security associations. For small M, this approach scales reasonably. Another approach allows any AP to act as an R0-KH, necessitating a full mesh of N2 security associations, which scales poorly. The model that utilizes designated R0-KHs is architecturally similar to the fast re-authentication model proposed by HOKEY. HOKEY, however, allows for handover between authenticators. This would allow an IEEE 802.11r-enabled peer to handover from one mobility domain to another without performing an EAP authentication.

The CAPWAP (Control and Provisioning of Wireless Access Points) protocol allows the functionality of an IEEE 802.11 access point to be split into two physical devices in enterprise deployments. Wireless Termination Points (WTPs) implement the physical and low-level Media Access Control (MAC) layers, while a centralized Access Controller (AC) provides higher-level management and protocol execution. Client authentication is handled by the AC, which acts as the AAA authenticator. One of the many features provided by CAPWAP is the ability to roam between WTPs without executing an EAP authentication. To accomplish this, the AC caches the MSK from an initial EAP authentication, and uses it to execute a separate four-way handshake with the station as it moves between WTPs. The keys resulting from the four-way handshake are then distributed to the WTP to which the station is associated. CAPWAP is transparent to the station. CAPWAP currently has no means to support roaming between ACs in an enterprise network. The proposed work on EAP efficient re-authentication addresses is an inter-authenticator handover problem from an EAP perspective, which applies during handover between ACs. Inter-AC handover is a topic yet to be addressed in great detail and the re-authentication work can potentially address it in an effective manner.

This document details the HOKEY problem statement. Since HOKEY is an authentication protocol, there is a myriad of security-related issues surrounding its development and deployment. In this document, we have detailed a variety of security properties inferred from to which HOKEY must conform, including the management of key context, scope, freshness, and transport; resistance to attacks based on the domino effect; and authentication and authorization. See for further details.

This document represents the synthesis of two problem statement documents. In this section, we acknowledge their contributions, and involvement in the early documents. Mohan Parthasarathy Nokia EMail: mohan.parthasarathy@nokia.com Julien Bournelle France Telecom R&D EMail: julien.bournelle@orange-ftgroup.com Hannes Tschofenig Siemens EMail: Hannes.Tschofenig@siemens.com Rafael Marin Lopez Universidad de Murcia EMail: rafa@dif.um.es

The authors would like to thank the participants of the HOKEY working group for their review and comments including: Glen Zorn, Dan Harkins, Joe Salowey, and Yoshi Ohba. The authors would also like to thank those that provided last-call reviews including: Bernard Aboba, Alan DeKok, Jari Arkko, and Hannes Tschofenig.