]> ELVIS-PLUS

PO Box 81 Moscow (Zelenograd) 124460 Russian Federation +7 495 276 0211 svan@elvis.ru

Red Hat

pwouters@redhat.com

unauthenticated opportunistic security pervasive monitoring Peer Authorization Database PAD This document specifies the NULL Authentication method and the ID_NULL Identification Payload ID Type for Internet Key Exchange Protocol version 2 (IKEv2). This allows two IKE peers to establish single-side authenticated or mutual unauthenticated IKE sessions for those use cases where a peer is unwilling or unable to authenticate or identify itself. This ensures IKEv2 can be used for Opportunistic Security (also known as Opportunistic Encryption) to defend against Pervasive Monitoring attacks without the need to sacrifice anonymity.

Internet Key Exchange Protocol version 2 (IKEv2), specified in , provides a way for two parties to perform an authenticated key exchange. While the authentication methods used by the peers can be different, there is no method for one or both parties to remain unauthenticated and anonymous. This document extends the authentication methods to support unauthenticated and anonymous IKE sessions. In some situations, mutual authentication is undesirable, superfluous, or impossible. The following three examples illustrate these unauthenticated use cases: A user wants to establish an anonymous secure connection to a server. In this situation, the user should be able to authenticate the server without presenting or authenticating to the server with their own identity. This case uses a single-sided authentication of the responder. A sensor that periodically wakes up from a suspended state wants to send a measurement (e.g., temperature) to a collecting server. The sensor must be authenticated by the server to ensure authenticity of the measurement, but the sensor does not need to authenticate the server. This case uses a single-sided authentication of the initiator. Two peers without any trust relationship wish to defend against widespread pervasive monitoring attacks as described in . Without a trust relationship, the peers cannot authenticate each other. Opportunistic Security states that unauthenticated encrypted communication is preferred over cleartext communication. The peers want to use IKE to setup an unauthenticated encrypted connection that gives them protection against pervasive monitoring attacks. An attacker that is able and willing to send packets can still launch a man-in-the-middle (MITM) attack to obtain a copy of the unencrypted communication. This case uses a fully unauthenticated key exchange. To meet these needs, this document introduces the NULL Authentication method and the ID_NULL ID type. This allows an IKE peer to explicitly indicate that it is unwilling or unable to certify its identity.

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 .

In IKEv2, each peer independently selects the method to authenticate itself to the other side. A peer may choose to refrain from authentication by using the NULL Authentication method. If a host's local policy requires that the identity of its peer be (non-null) authenticated, and if that host receives an AUTH payload containing the NULL Authentication method type, it MUST return an AUTHENTICATION_FAILED notification. If an initiator uses the Extensible Authentication Protocol (EAP), the responder MUST NOT use the NULL Authentication method (in conformance with Section 2.16 of ). NULL authentication affects how the Authentication and the Identification payloads are formed in the IKE_AUTH exchange.

NULL authentication still requires a properly formed AUTH payload to be present in the IKE_AUTH exchange messages, as the AUTH payload cryptographically links the IKE_SA_INIT exchange messages with the other messages sent over this IKE Security Association (SA). When using NULL authentication, the content of the AUTH payload is computed using the syntax of pre-shared secret authentication, described in Section 2.15 of . The value of SK_pi for the initiator and SK_pr for the responder is used as the shared secret for the content of the AUTH payload. Implementers should note this means that authentication keys used by the two peers are different in each direction. This is identical to how the contents of the two last AUTH payloads are generated for the non-key-generating EAP methods (see Section 2.16 of for details). The IKEv2 Authentication Method value for NULL Authentication is 13.

When a remote peer is not authenticated, any ID presented in the Identification Data field of the ID payload cannot be validated. To avoid the need of sending a bogus ID Type with placeholder data, this specification defines a new ID Type, ID_NULL. The Identification Data field of the ID payload for this ID Type MUST be empty. If NULL authentication is in use and anonymity is a concern, then ID_NULL SHOULD be used in the Identification payload. Some examples of cases where a non-null identity type and value with NULL authentication can be used are logging, troubleshooting, and in scenarios where authentication takes place out of band after the IKE SA is created (like in ). The content of the Identification payload MUST NOT be used for any trust and policy checking in IKE_AUTH exchange when NULL authentication is employed (see Section 2.4 for details). ID_NULL is primarily intended to be used with NULL authentication but could be used in other situations where the content of the Identification payload is not used. For example, ID_NULL could be used when authentication is performed via raw public keys and the identities are the keys themselves. These alternative uses of ID_NULL should be described in their own respective documents. The IKEv2 Identification Payload ID Type for ID_NULL is 13.

The identity of a peer using NULL authentication cannot be used to find existing IKE SAs created by the same peer, as the peer identity is not authenticated. For that reason, the INITIAL_CONTACT notifications MUST NOT be used to delete any other IKE SAs based on the same peer identity without additional verification that the existing IKE SAs with matching identity are actually stale. The standard IKE Liveness Check procedure, described in Section 2.4 of , can be used to detect stale IKE SAs created by peers using NULL authentication. Inactive, unauthenticated IKE SAs should be checked periodically. Additionally, the event of creating a new unauthenticated IKE SA can be used to trigger an out-of-order check on existing unauthenticated IKE SAs possibly limited to identical or close-by IP addresses or to identical identities of the just created IKE SA. Implementations should weigh the resource consumption of sending Liveness Checks against the memory usage of possible orphaned IKE SAs. Implementations may choose to handle situations with thousands of unauthenticated IKE SAs differently from situations with very few such SAs.

Section 4.4.3 of defines the Peer Authorization Database (PAD), which provides the link between the Security Policy Database (SPD) and IKEv2. The PAD contains an ordered list of records with peers' identities along with corresponding authentication data and Child SA authorization data. When the IKE SA is being established, the PAD is consulted to determine how the peer should be authenticated and what Child SAs it is authorized to create. When using NULL authentication, the peer identity is not authenticated and cannot be trusted. If ID_NULL is used with NULL authentication, there is no ID at all. The processing of the PAD described in Section 4.4.3 of is updated for NULL authentication as follows. NULL authentication is added as one of the supported authentication methods. This method does not have any authentication data. ID_NULL is included into the list of allowed ID types. The matching rule for ID_NULL consists only of whether this type is used, i.e., no actual ID matching is done as ID_NULL contains no identity data. When using the NULL Authentication method, those matching rules MUST include matching of a new flag in the SPD entry specifying whether unauthenticated users are allowed to use that entry. That is, each SPD entry needs to be augmented to have a flag specifying whether it can be used with NULL authentication or not, and only those rules that explicitly have that flag turned on can be used with unauthenticated connections. The specific updates of text in Section 4.4.3 of are listed in .

Traffic Selectors and narrowing allow two IKE peers to mutually agree on a traffic range for an IPsec SA. An unauthenticated peer must not be allowed to use this mechanism to steal traffic that an IKE peer intended to be for another host. This is especially problematic when supporting anonymous IKE peers behind NAT, as such IKE peers build an IPsec SA using their pre-NAT IP address that is different from the source IP of their IKE packets. A rogue IKE peer could use malicious Traffic Selectors to trick a remote host into giving it IP traffic that the remote host never intended to be sent to remote IKE peers. For example, if the remote host uses 192.0.2.1 as the DNS server, a rogue IKE peer could set its Traffic Selector to 192.0.2.1 in an attempt to receive the remote peer's DNS traffic. Implementations SHOULD restrict and isolate all anonymous IKE peers from each other and itself and only allow it access to itself and possibly its intended network ranges. One method to achieve this is to always assign internal IP addresses to unauthenticated IKE clients, as described in Section 2.19 of . Implementations may also use other techniques such as internal NAT and connection tracking. Implementations MAY force unauthenticated IKE peers to single host-to-host IPsec SAs. When using IPv6, this is not always possible, so implementations MUST be able to assign a full /64 address block to the peer as described in , even if it is not authenticated.

If authenticated IKE sessions are possible for a certain Traffic Selector range between the peers, then unauthenticated IKE SHOULD NOT be allowed for that Traffic Selector range. When mixing authenticated and unauthenticated IKE with the same peer, policy rules should ensure the highest level of security will be used to protect the communication between the two peers. See for details. If both peers use NULL authentication, the entire key exchange becomes unauthenticated. This makes the IKE session vulnerable to active MITM attacks. Using an ID Type other than ID_NULL with the NULL Authentication method may compromise the client's anonymity in case of an active MITM attack. IKE implementations without NULL authentication have always performed mutual authentication and were not designed for use with unauthenticated IKE peers. Implementations might have made assumptions that remote peers are identified. With NULL authentication, these assumptions are no longer valid. Furthermore, the host itself might have made trust assumptions or may not be aware of the network topology changes that resulted from IPsec SAs from unauthenticated IKE peers.

With NULL authentication, an established IKE session is no longer guaranteed to provide a verifiable (authenticated) entity known to the system or network. Any logging of unproven ID payloads that were not authenticated should be clearly marked and treated as "untrusted" and possibly accompanied by logging the remote IP address of the IKE session. Rate limiting of logging might be required to prevent excessive resource consumption causing system damage.

Section 2.6 of provides guidance for mitigation of denial-of-service (DoS) attacks by issuing COOKIES in response to resource consumption of half-open IKE SAs. Furthermore, offers additional countermeasures in an attempt to distinguish attacking IKE packets from legitimate IKE peers. These defense mechanisms do not take into account IKE systems that allow unauthenticated IKE peers. An attacker using NULL authentication is a fully legitimate IKE peer that is only distinguished from authenticated IKE peers by having used NULL authentication. Implementers that implement NULL authentication should ensure their implementation does not make any assumptions that depend on IKE peers being "friendly", "trusted", or "identifiable". While implementations should have been written to account for abusive authenticated clients, any omission or error in handling abusive clients may have gone unnoticed because abusive clients have been a rare or nonexistent problem. When adding support for unauthenticated IKE peers, these implementation omissions and errors will be found and abused by attackers. For example, an unauthenticated IKE peer could send an abusive amount of Liveness probes or Delete requests.

Combining authenticated and unauthenticated IKE peers on a single host can be dangerous, assuming the authenticated IKE peer gains more or different access from unauthenticated peers (otherwise, why not only allow unauthenticated peers). An unauthenticated IKE peer MUST NOT be able to reach resources only meant for authenticated IKE peers and MUST NOT be able to replace the Child SAs of an authenticated IKE peer.

When a host relies on packet filters or firewall software to protect itself, establishing an IKE SA and installing an IPsec SA might accidentally circumvent these packet filters and firewall restrictions, as the Encapsulating Security Payload (ESP, protocol 50) or ESPinUDP (UDP port 4500) packets of the encrypted traffic do not match the packet filters defined for unencrypted traffic. IKE peers supporting unauthenticated IKE MUST pass all decrypted traffic through the same packet filters and security mechanisms as incoming plaintext traffic.

Per this document, IANA has added a new entry in the "IKEv2 Authentication Method" registry:

Per this document, IANA has added a new entry in the "IKEv2 Identification Payload ID Types" registry:

&rfc2119; &rfc4301; &rfc5739; &rfc7296; &rfc7258; &rfc7435;

This appendix lists the specific updates of the text in Section 4.4.3 of that should be followed when implementing NULL authentication. A new item is added to the list of supported ID types in Section 4.4.3.1 of NULL ID (matches ID type only) and the following text is added at the end of the section: The NULL ID type is defined as having no data. For this name type, the matching function is defined as comparing the ID type only. A new item is added to the list of authentication data types in Section 4.4.3.2 of : - NULL authentication and the next paragraph is updated as follows: For authentication based on an X.509 certificate [...] For authentication based on a pre-shared secret, the PAD contains the pre-shared secret to be used by IKE. For authentication based on an X.509 certificate [...] For authentication based on a pre-shared secret, the PAD contains the pre-shared secret to be used by IKE. For NULL authentication the PAD contains no data. In addition, the following text is added at the end of Section 4.4.3.4 of : When using the NULL Authentication method, implementations MUST make sure that they do not mix authenticated and unauthenticated SPD rules, i.e., implementations need to keep them separately; for example, by adding a flag in the SPD to tell whether NULL authentication can be used or not for the entry. That is, each SPD entry needs to be augmented to have a flag specifying whether it can be used with NULL authentication or not, and only those rules that explicitly have that flag set can be used with unauthenticated connections.

The authors would like to thank Yaron Sheffer and Tero Kivinen for their reviews, valuable comments, and contributed text.