Keying Material Exporters for Transport Layer Security (TLS)

The mechanism described in this document was previously known as "TLS Extractors" but was changed to avoid a name conflict with the use of the term "Extractor" in the cryptographic community. A number of protocols wish to leverage Transport Layer Security (TLS) or Datagram TLS (DTLS) to perform key establishment but then use some of the keying material for their own purposes. A typical example is DTLS-SRTP , which uses DTLS to perform a key exchange and negotiate the SRTP protection suite and then uses the DTLS master_secret to generate the SRTP keys. These applications imply a need to be able to export keying material (later called Exported Keying Material or EKM) from TLS/DTLS, and securely agree on the upper-layer context where the keying material will be used. The mechanism for exporting the keying material has the following requirements: Both client and server need to be able to export the same EKM value. EKM values should be indistinguishable from random by attackers who don't know the master_secret. It should be possible to export multiple EKM values from the same TLS/DTLS association. Knowing one EKM value should not reveal any information about the master_secret or about other EKM values. The mechanism described in this document is intended to fulfill these requirements.

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 addition to exporting keying material, an application using the keying material has to securely establish the upper-layer context where the keying material will be used. The details of this context depend on the application, but it could include things such as algorithms and parameters that will be used with the keys, identifier(s) for the endpoint(s) who will use the keys, identifier(s) for the session(s) where the keys will be used, and the lifetime(s) for the context and/or keys. At minimum, there should be some mechanism for signalling that an exporter will be used. This specification does not mandate a single mechanism for agreeing on such context; instead, there are several possibilities that can be used (and can complement each other). For example: One important part of the context -- which application will use the exported keys -- is given by the disambiguating label string (see Section 4). Information about the upper-layer context can be included in the optional data after the exporter label (see Section 4). Information about the upper-layer context can be exchanged in TLS extensions included in the ClientHello and ServerHello messages. This approach is used in [DTLS-SRTP]. The handshake messages are protected by the Finished messages, so once the handshake completes, the peers will have the same view of the information. Extensions also allow a limited form of negotiation: for example, the TLS client could propose several alternatives for some context parameters, and the TLS server could select one of them. The upper-layer protocol can include its own handshake which can be protected using the keys exported from TLS. It is important to note that just embedding TLS messages in the upper-layer protocol may not automatically secure all the important context information, since the upper-layer messages are not covered by TLS Finished messages.

The output of the exporter is intended to be used in a single scope, which is associated with the TLS session, the label, and the context value. A disambiguating label string A per-association context value provided by the application using the exporter A length value It then computes:

Where PRF is the TLS PRF in use for the session. The output is a pseudorandom bit string of length bytes generated from the master_secret. Labels here have the same definition as in TLS, i.e., an ASCII string with no terminating NULL. Label values beginning with "EXPERIMENTAL" MAY be used for private use without registration. All other label values MUST be registered via Specification Required as described by RFC 5226 . Note that exporter labels have the potential to collide with existing PRF labels. In order to prevent this, labels SHOULD begin with "EXPORTER". This is not a MUST because there are existing uses which have labels which do not begin with this prefix.

; ]]> The context value allows the application using the exporter to mix its own data with the TLS PRF for the exporter output. One example of where this might be useful is an authentication setting where the client credentials are valid for more than one identity; the context value could then be used to mix the expected identity into the keying material, thus preventing substitution attacks. The context value length is encoded as an unsigned 16-bit quantity (uint16) representing the length of the context value. The context MAY be zero length.

The prime security requirement for exporter outputs is that they be independent. More formally, after a particular TLS session, if an adversary is allowed to choose multiple (label, context value) pairs and is given the output of the PRF for those values, the attacker is still unable to distinguish between the output of the PRF for a (label, context value) pair (different from the ones that it submitted) and a random value of the same length. In particular, there may be settings, such as the one described in , where the attacker can control the context value; such an attacker MUST not be able to predict the output of the exporter. Similarly, an attacker who does not know the master secret should not be able to distinguish valid exporter outputs from random values. The current set of TLS PRFs is believed to meet this objective, provided the master secret is randomly generated. Because an exporter produces the same value if applied twice with the same label to the same master_secret, it is critical that two EKM values generated with the same label not be used for two different purposes--hence the requirement for IANA registration. However, because exporters depend on the TLS PRF, it is not a threat to the use of an EKM value generated from one label to reveal an EKM value generated from another label.

IANA is requested to create (has created) a TLS Exporter Label registry for this purpose. The initial contents of the registry are given below:

Value Reference ----- ------------ client finished [RFC5246] server finished [RFC5246] master secret [RFC5246] key expansion [RFC5246] client EAP encryption [RFC2716] ttls keying material [RFC5281] ttls challenge [RFC5281] Future values are allocated via RFC5226 Specification Required policy. The label is a string consisting of printable ASCII characters. IANA MUST also verify that one label is not a prefix of any other label. For example, labels "key" or "master secretary" are forbidden.

Thanks to Pasi Eronen for valuable comments and the contents of the IANA section and . Thanks to David McGrew for helpful discussion of the security considerations and Alfred Hoenes for editorial comments.