Update to the Extensible Authentication Protocol (EAP) Applicability Statement for Application Bridging for Federated Access Beyond Web (ABFAB)

Ongoing work in the IETF specifies the use of EAP over Generic Security Service Application Program Interface (GSS-API) for generic application layer access . In the past, using EAP in this context has met resistance due to the lack of channel bindings . Without channel bindings, a peer cannot verify if an authenticator is authorized to provide an advertised service. However, as additional services use EAP for authentication, the distinction of which service is being contacted becomes more important. Application services might have different properties. Consider an environment with multiple printers, some of which provide a confidential service to output documents to a controlled location. If a peer sent a document to the wrong service, then potentially sensitive information might be printed in an uncontrolled location and be disclosed. In addition, it might be more likely that a low-value service is compromised than some high-value service. If the high-value service could be impersonated by a low-value service then the security of the overall system would be limited by the security of the lower-value service. This distinction is present in any environment where peers' security depends on which service they reach. However, it is particularly acute in a federated environment where multiple organizations are involved. It is very likely that these organizations will have different security policies and practices. It is very likely that the goals of these organizations will not entirely be aligned. In many situations, one organization could gain value by being able to impersonate another. In this environment, authenticating the EAP server is insufficient: the peer must also validate that the contacted host is authorized to provide the requested service. In environments where EAP is used for purposes other than network access authentication: All EAP servers and all application access EAP peers MUST support channel bindings. All network access EAP peers SHOULD support channel bindings. Channel binding MUST be used for all application authentication. The EAP server MUST require that either the correct EAP lower-layer attribute or another attribute indicating the purpose of the authentication be present in the channel binding data for application authentication. Channel binding SHOULD be used for all network access authentication, and when channel binding data is present, the EAP server MUST require that it contain the correct EAP lower-layer attribute to explicitly identify the reason for authentication. Any new usage of EAP MUST use channel bindings including the EAP lower-layer attribute to prevent confusion with network access usage. Operators need to carefully consider the security implications before relaxing these requirements. One potentially serious attack exists when channel binding is not required and EAP authentication is introduced into an existing service other than network access. A device can be created that impersonates a Network Access Service (NAS) to peers, but actually proxies the authentication to the new application service that accepts EAP authentications. This may decrease the security of this service even for users who previously used non-EAP means of authentication to the service. It is REQUIRED for the application layer to prove that both the EAP peer and EAP authenticator possess the EAP Master Session Key (MSK). Failing to validate the possession of the EAP MSK can allow an attacker to insert himself into the conversation and impersonate the peer or authenticator. In addition, the application should define channel binding attributes that are sufficient to validate that the application service is being correctly represented to the peer.

In EAP, the authenticator is responsible for retransmission. By default, EAP assumes that the lower layer (the application in this context) is unreliable. The authenticator can send a packet whenever its retransmission timer triggers. In this mode, applications need to be able to receive and process EAP messages at any time during the authentication conversation. Alternatively, EAP permits a lower layer to set the retransmission timer to infinite. When this happens, the lower layer becomes responsible for reliable delivery of EAP messages. Applications that use a lock-step or client-driven authentication protocol might benefit from this approach. In addition to retransmission behavior, applications need to deal with discarded EAP messages. For example, whenever some EAP methods receive erroneous input, these methods discard the input rather than generating an error response. If the erroneous input was generated by an attacker, legitimate input can sometimes be received after the erroneous input. Applications MUST handle discarded EAP messages, although the specific way in which discarded messages will be handled depends on the characteristics of the application. Options include failing the authentication at the application level, requesting an EAP retransmit and waiting for additional EAP input. Applications designers that incorporate EAP into their application need to determine how retransmission and message discards are handled.

EAP lower layers MAY provide a mechanism for re-authentication to happen within an existing session . Re-authentication permits security associations to be updated without establishing a new session. For network access, this can be important because interrupting network access can disrupt connections and media. Some applications might not need re-authentication support. For example, if sessions are relatively short-lived or if sessions can be replaced without significant disruption, re-authentication might not provide value. Protocols like HyperText Transfer Protocol (HTTP) and Simple Mail Transport Protocol (SMTP) are examples of protocols where establishing a new connection to update security associations is likely to be sufficient. Re-authentication is likely to be valuable if sessions or connections are long-lived or if there is a significant cost to disrupting them. Another factor may make re-authentication important. Some protocols only permit one party in a protocol (for example, the client) to establish a new connection. If another party in the protocol needs the security association refreshed, then re-authentication can provide a mechanism to do so. Application designers need to determine whether re-authentication support is needed and which parties can initiate it.