1.1. Requirements Language

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶

2.1. Key Derivation

The key derivation for TLS-based EAP methods depends on the value of the EAP Type as defined by [IANA] in the "Extensible Authentication Protocol (EAP) Registry". The most important definition is of the Type field, as first defined in [RFC3748], Section 2:¶

Type = value of the EAP Method type¶

For the purposes of this specification, when we refer to logical Type, we mean that the logical Type is defined as one octet for values smaller than 254 (the value for the Expanded Type). When Expanded EAP Types are used, the logical Type is defined as the concatenation of the fields required to define the Expanded Type, including the Type with value 0xfe, Vendor-Id (in network byte order), and Vendor-Type fields (in network byte order) defined in [RFC3748], Section 5.7, as given below:¶

Type = 0xFE || Vendor-Id || Vendor-Type

This definition does not alter the meaning of Type in [RFC3748] or change the structure of EAP packets. Instead, this definition allows us to simplify references to EAP Types by using a logical "Type" instead of referring to "the Type field or the Type field with value 0xfe, plus the Vendor-ID and Vendor-Type". For example, the value of Type for PEAP is simply 0x19.¶

Note that unlike TLS 1.2 and earlier, the calculation of the TLS-Exporter function depends on the length passed to it. Therefore, implementations MUST pass the correct length instead of passing a large length and truncating the output. Any output calculated using a larger length value, which is then truncated, will be different from the output that was calculated using the correct length.¶

Unless otherwise discussed below, the key derivation functions for all TLS-based EAP Types are defined in [RFC9190], Section 2.3 and reproduced here for clarity. These definitions include ones for the Master Session Key (MSK) and the Extended Master Session Key (EMSK):¶

Key_Material = TLS-Exporter("EXPORTER_EAP_TLS_Key_Material",
                             Type, 128)
Method-Id    = TLS-Exporter("EXPORTER_EAP_TLS_Method-Id",
                             Type, 64)
Session-Id   = Type || Method-Id
MSK          = Key_Material(0, 63)
EMSK         = Key_Material(64, 127)

We note that these definitions reuse the EAP-TLS exporter labels and change the derivation only by adding a dependency on the logical Type. The reason for this change is simplicity. The inclusion of the EAP Type makes the derivation method specific. There is no need to use different labels for different EAP Types as was done earlier.¶

These definitions apply in their entirety to EAP-TTLS [RFC5281] and PEAP as defined in [PEAP] and [MSPEAP]. Some definitions apply to EAP-FAST and TEAP with exceptions as noted below.¶

It is RECOMMENDED that vendor-defined and TLS-based EAP methods use the above definitions for TLS 1.3. There is no compelling reason to use different definitions.¶

2.2. TEAP

TEAP previously used a Protected Access Credential (PAC), which is functionally equivalent to session tickets provided by TLS 1.3 that contain a pre-shared key (PSK) along with other data. As such, the use of a PAC is deprecated for TEAP in TLS 1.3. PAC provisioning, as defined in [RFC7170], Section 3.8.1, is also no longer part of TEAP when TLS 1.3 is used.¶

[RFC7170], Section 5.2 gives a definition for the Inner Method Session Key (IMSK), which depends on the TLS Pseudorandom Function (PRF) (also known as TLS-PRF). When the j'th inner method generates an EMSK, we update that definition for TLS 1.3 as:¶

IMSK[j] = TLS-Exporter("TEAPbindkey@ietf.org", secret, 32)

The secret is the EMSK or MSK from the j'th inner method. When an inner method does not provide an EMSK or MSK, IMSK[j] is 32 octets of zero.¶

The other key derivations for TEAP are given here. All derivations not given here are the same as given above in the previous section. These derivations are also used for EAP-FAST, but using the EAP-FAST Type.¶

The derivation of the IMSKs, Inner Method Compound Keys (IMCKs), and Compound Session Keys (CMKs) is given below.¶

session_key_seed = TLS-Exporter("EXPORTER: teap session key seed",
                                Type, 40)

S-IMCK[0] = session_key_seed
For j = 1 to n-1 do
  IMCK[j] = TLS-Exporter("EXPORTER: Inner Methods Compound Keys",
                         S-IMCK[j-1] || IMSK[j], 60)
  S-IMCK[j] = first 40 octets of IMCK[j]
  CMK[j] = last 20 octets of IMCK[j]

Note: In these definitions, || denotes concatenation.¶

In TLS 1.3, the derivation of IMCK[j] uses both a different label and a different order of concatenating fields than what was used by TEAP with TLS 1.2. Similarly, the session_key_seed in TLS 1.3 uses the Type as the context. In TLS 1.2, the context was a zero-length field.¶

The outer MSK and EMSK are then derived from the final ("n"th) inner method, as follows:¶

MSK  = TLS-Exporter(
     "EXPORTER: Session Key Generating Function",
     S-IMCK[n], 64)

EMSK = TLS-Exporter(
     "EXPORTER: Extended Session Key Generating Function",
     S-IMCK[n], 64)

The TEAP Compound Message Authentication Code (MAC) defined in [RFC7170], Section 5.3 remains the same, but the MAC for TLS 1.3 is computed with the Hashed Message Authentication Code (HMAC) algorithm negotiated for the HMAC-based Key Derivation Function (HKDF) in the key schedule, as per [RFC8446], Section 7.1. That is, the MAC used is the MAC derived from the TLS handshake:¶

Compound-MAC = MAC( CMK[n], BUFFER )

where we define CMK[n] as the CMK taken from the final ("n"th) inner method.¶

For TLS 1.3, the MAC is computed with the HMAC algorithm negotiated for HKDF in the key schedule, as per [RFC8446], Section 7.1. That is, the MAC used is the MAC derived from the TLS handshake.¶

The definition of BUFFER is unchanged from [RFC7170], Section 5.3.¶

2.3. EAP-FAST

For EAP-FAST, the session_key_seed is also part of the key_block as defined in [RFC4851], Section 5.1.¶

The definitions of S-IMCK[n], MSK, and EMSK are the same as given above for TEAP. We reiterate that the EAP-FAST Type must be used when deriving the session_key_seed and not the TEAP Type.¶

Unlike [RFC4851], Section 5.2, the definition of IMCK[j] places the reference to S-IMCK after the textual label and then concatenates the IMSK instead of the MSK.¶

EAP-FAST previously used a PAC that is functionally equivalent to session tickets provided by TLS 1.3, which contain a PSK along with other data. As such, the use of a PAC is deprecated for EAP-FAST in TLS 1.3. PAC provisioning [RFC5422] is also no longer part of EAP-FAST when TLS 1.3 is used.¶

The T-PRF given in [RFC4851], Section 5.5 is not used for TLS 1.3. Instead, it is replaced with the TLS 1.3 TLS-Exporter function.¶

2.4. EAP-TTLS

[RFC5281], Section 11.1 defines an implicit challenge when the inner methods of the Challenge Handshake Authentication Protocol (CHAP) [RFC1994], Microsoft CHAP (MS-CHAP) [RFC2433], or MS-CHAPv2 [RFC2759] are used. The derivation for TLS 1.3 is instead given as:¶

EAP-TTLS_challenge = TLS-Exporter("ttls challenge",, n)

There is no "context_value" ([RFC8446], Section 7.5) passed to the TLS-Exporter function. The value "n" given here is the length of the data required; [RFC5281] requires it to be 17 octets for CHAP ([RFC5281], Section 11.2.2) and MS-CHAPv2 ([RFC5281], Section 11.2.4), and 9 octets for MS-CHAP ([RFC5281], Section 11.2.3).¶

When the Password Authentication Protocol (PAP), CHAP, or MS-CHAPv1 are used as inner authentication methods, there is no opportunity for the EAP server to send a protected success indication, as is done in [RFC9190], Section 2.5. Instead, when TLS session tickets are disabled, the response from the EAP server MUST be either EAP-Success or EAP-Failure. These responses are unprotected and can be forged by a skilled attacker.¶

Where TLS session tickets are enabled, the response from the EAP server may also continue TLS negotiation with a TLS NewSessionTicket message. Since this message is protected by TLS, it can serve as the protected success indication.¶

Therefore, it is RECOMMENDED that EAP servers always send a TLS NewSessionTicket message, even if resumption is not configured. When the EAP peer attempts to use the ticket, the EAP server can instead request a full authentication. As noted earlier, implementations SHOULD NOT send TLS NewSessionTicket messages until the "inner tunnel" authentication has completed in order to take full advantage of the message as a protected success indication.¶

When resumption is not used, the TLS NewSessionTicket message is not available and some authentication methods will not have a protected success indication. While we would like to always have a protected success indication, limitations of the underlying protocols, implementations, and deployment requirements make that impossible.¶

EAP peers MUST continue running their EAP state machine until they receive either an EAP-Success or an EAP-Failure. Receiving a TLS NewSessionTicket message in response to inner method PAP, CHAP, or MS-CHAP authentication is normal and MUST NOT be treated as a failure.¶

2.5. PEAP

When PEAP uses crypto binding, it uses a different key calculation defined in [PEAP-MPPE] that consumes inner EAP method keying material. The PRF+ function used in [PEAP-MPPE] is not taken from the TLS exporter but is instead calculated via a different method that is given in [PEAP-PRF]. That derivation remains unchanged in this specification.¶

Note that the above derivation uses SHA-1, which may be formally deprecated in the near future.¶

However, the PRF+ calculation uses a PEAP Tunnel Key (TK), which is defined in [PEAP-TK] as:¶

... the TK is the first 60 octets of the Key_Material, as specified in [RFC5216]: TLS-PRF-128 (master secret, "client EAP encryption", client.random || server.random).¶

We note that the text in [PEAP-PRF] does not define Key_Material. Instead, it defines TK as the first octets of Key_Material and gives a definition of Key_Material that is appropriate for TLS versions before TLS 1.3.¶

For TLS 1.3, the TK should be derived from the Key_Material defined here in Section 2.1 instead of using the TLS-PRF-128 derivation given in [PEAP-PRF]. The method defined in [PEAP-TK] MUST NOT be used.¶

3.1. Identities

For EAP-TLS, Sections 2.1.3 and 2.1.7 of [RFC9190] recommend the use of anonymous Network Access Identifiers (NAIs) [RFC7542] in the EAP Response/Identity packet. However, as EAP-TLS does not send application data inside of the TLS tunnel, that specification does not address the subject of "inner" identities in tunneled EAP methods. However, this subject must be addressed for the tunneled methods.¶

Using an anonymous NAI for the outer identity as per [RFC7542], Section 2.4 has a few benefits. An NAI allows the EAP session to be routed in a AAA framework as described in [RFC7542], Section 3. Using an anonymous realm also ensures that user identifiers are kept private.¶

As for the inner identity, we define it generically as the identification information carried inside of the TLS tunnel. For PEAP, that identity may be an EAP Response/Identity. For EAP-TTLS, it may be the User-Name attribute. Vendor-specific EAP methods that use TLS will generally also have an inner identity. This identity is carried inside of the TLS tunnel and is therefore both routed to the correct destination by the outer identity and kept private by the use of TLS.¶

In other words, we can view the outer TLS layer of tunneled EAP methods as a secure transport layer that is responsible for getting the actual (inner) authentication credentials securely from the EAP peer to the EAP server. The EAP server then uses the inner identity and inner authentication data to identify and authenticate a particular user.¶

As the authentication data is routed to the correct destination, there is little reason for the inner identity to also contain a realm. Therefore, we have a few recommendations on the inner and outer identities, along with their relationship to each other.¶

The outer identity SHOULD use an anonymous NAI realm that allows for both user privacy and for the EAP session to be routed in a AAA framework as described in [RFC7542], Section 3. Where NAI realms are not used, packets will not be routable outside of the local organization.¶

The inner identity MUST NOT use an anonymous NAI realm. If anonymous network access is desired, EAP peers MUST use EAP-TLS without peer authentication, as per [RFC9190], Section 2.1.5. EAP servers MUST cause authentication to fail if an EAP peer uses an anonymous "inner" identity for any TLS-based EAP method.¶

Implementations SHOULD NOT use inner identities that contain an NAI realm. Many organizations typically use only one realm for all user accounts.¶

However, there are situations where it is useful for an inner identity to contain a realm. For example, an organization may have multiple independent sub-organizations, each with a different and unique realm. These realms may be independent of one another, or the realms may be a subdomain (or subdomains) of the public outer realm.¶

In that case, an organization can configure one public "routing" realm and multiple separate "inner" realms. This separation of realms also allows an organization to split users into logical groups by realm, where the "user" portion of the NAI may otherwise conflict. For example, "user@example.com" and "user@example.org" are different NAIs that can both be used as inner identities.¶

Using only one public realm both keeps internal information private and simplifies realm management for external entities by minimizing the number of realms that have to be tracked by them.¶

In most situations, routing identifiers should be associated with the authentication data that they are routing. For example, if a user has an inner identity of "user@example.com", then it generally makes little sense to have an outer identity of "@example.org". The authentication request would then be routed to the "example.org" domain, which may have no idea what to do with the credentials for "user@example.com". At best, the authentication request would be discarded. At worst, the "example.org" domain could harvest user credentials for later use in attacks on "example.com".¶

When an EAP server receives an inner identity for a realm which it is not authoritative, it MUST reject the authentication. There is no reason for one organization to authenticate users from a different (and independent) organization.¶

In addition, associating inner/outer identities from different organizations in the same EAP authentication session means that otherwise unrelated realms are tied together, which can make networks more fragile.¶

For example, an organization that uses a "hosted" AAA provider may choose to use the realm of the AAA provider as the outer identity for user authentication. The inner identity can then be fully qualified: username plus realm of the organization. This practice may result in successful authentications, but it has practical difficulties.¶

Additionally, an organization may host their own AAA servers but use a "cloud" identity provider to hold user accounts. In that situation, the organizations could try to use their own realm as the outer (routing) identity and then use an identity from the "cloud" provider as the inner identity.¶

This practice is NOT RECOMMENDED. User accounts for an organization should be qualified as belonging to that organization and not to an unrelated third party. There is no reason to tie the configuration of user systems to public realm routing; that configuration more properly belongs in the network.¶

Both of these practices mean that changing "cloud" providers is difficult. When such a change happens, each individual EAP peer must be updated with a different outer identity that points to the new "cloud" provider. This process can be expensive, and some EAP peers may not be online when this changeover happens. The result could be devices or users who are unable to obtain network access, even if all relevant network systems are online and functional.¶

Further, standards such as [RFC7585] allow for dynamic discovery of home servers for authentication. This specification has been widely deployed and means that there is minimal cost to routing authentication to a particular domain. The authentication can also be routed to a particular identity provider and changed at will with no loss of functionality. That specification is also scalable since it does not require changes to many systems when a domain updates its configuration. Instead, only one thing has to change: the configuration of that domain. Everything else is discovered dynamically.¶

That is, changing the configuration for one domain is significantly simpler and more scalable than changing the configuration for potentially millions of end-user devices.¶

We recognize that there may be existing use cases where the inner and outer identities use different realms. As such, we cannot forbid that practice. We hope that the discussion above shows not only why such practices are problematic, but how alternative methods are more flexible, more scalable, and are easier to manage.¶

5.1. Handling of TLS NewSessionTicket Messages

In some cases, client certificates are not used for TLS-based EAP methods. In those cases, the user is authenticated only after successful completion of the inner tunnel authentication. However, [RFC8446], Section 4.6.1 states that "at any time after the server has received the client Finished message, it MAY send a NewSessionTicket message." This message is sent by the server before the inner authentication method has been run and therefore before the user has been authenticated.¶

This separation of data allows for a "time of use, time of check" security issue. Malicious clients can begin a session and receive a NewSessionTicket message. The malicious client can then abort the authentication session and use the obtained NewSessionTicket to "resume" the previous session. If the server allows the session to resume without verifying that the user had first been authenticated, the malicious client can then obtain network access without ever being authenticated.¶

As a result, EAP servers MUST NOT assume that a user has been authenticated simply because a TLS session is being resumed. Even if a session is being resumed, an EAP server MAY have policies that still force the inner authentication methods to be run. For example, the user's password may have expired in the time interval between first authentication and session resumption.¶

Therefore, the guidelines given here describe situations where an EAP server is permitted to allow session resumption rather than where an EAP server is required to allow session resumption. An EAP server could simply refuse to issue session tickets or could run the full inner authentication, even if a session was resumed.¶

Where session tickets are used, the EAP server SHOULD track the successful completion of an inner authentication and associate that status with any session tickets issued for that session. This requirement can be met in a number of different ways.¶

One way is for the EAP server to simply not send any TLS NewSessionTicket messages until the inner authentication has completed successfully. The EAP server then knows that the existence of a session ticket is proof that a user was authenticated, and the session can be resumed.¶

Another way is for the EAP server to simply discard or invalidate any session tickets until after the inner authentication has completed successfully. When the user is authenticated, a new TLS NewSessionTicket message can be sent to the client, and the new ticket can be cached and/or validated.¶

Another way is for the EAP server to associate the inner authentication status with each session ticket. When a session ticket is used, the authentication status is checked. When a session ticket shows that the inner authentication did not succeed, the EAP server MUST run the inner authentication method(s) in the resumed tunnel and only grant access based on the success or failure of those inner methods.¶

However, the interaction between EAP implementations and any underlying TLS library may be complex, and the EAP server may not be able to make the above guarantees. Where the EAP server is unable to determine the user's authentication status from the session ticket, it MUST assume that inner authentication has not completed, and it MUST run the inner authentication method(s) successfully in the resumed tunnel before granting access.¶

This issue is not relevant for EAP-TLS, which only uses client certificates for authentication in the TLS handshake. It is only relevant for TLS-based EAP methods that do not use the TLS layer to authenticate.¶

5.2. Protected Success and Failure Indications

[RFC9190] provides for protected success and failure indications as discussed in [RFC4137], Section 4.1.1. These result indications are provided for both full authentication and resumption.¶

Other TLS-based EAP methods provide these result indications only for resumption.¶

For full authentication, the other TLS-based EAP methods do not provide for protected success and failure indications as part of the outer TLS exchange. That is, the protected result indication is not used, and there is no TLS-layer alert sent when the inner authentication fails. Instead, there is simply either an EAP-Success or an EAP-Failure sent. This behavior is the same as for previous TLS versions; therefore, it introduces no new security issues.¶

We note that most TLS-based EAP methods provide for success and failure indications as part of the authentication exchange performed inside of the TLS tunnel. These result indications are therefore protected, as they cannot be modified or forged.¶

However, some inner methods do not provide for success or failure indications. For example, the use of EAP-TTLS with inner PAP, CHAP, or MS-CHAP. Those methods send authentication credentials to the EAP server via the inner tunnel with no method to signal success or failure inside of the tunnel.¶

There are functionally equivalent authentication methods that can be used to provide protected result indications. PAP can often be replaced with EAP-Generic Token Card (EAP-GTC), CHAP with EAP-MD5, and MS-CHAPv1 with MS-CHAPv2 or EAP-MSCHAPv2. All of the replacement methods provide for similar functionality and have protected success and failure indication. The main cost to this change is additional round trips.¶

It is RECOMMENDED that implementations deprecate inner tunnel methods that do not provide protected success and failure indications when TLS session tickets cannot be used. Implementations SHOULD use EAP- GTC instead of PAP and EAP-MD5 instead of CHAP. Implementations SHOULD use MS-CHAPv2 or EAP-MSCHAPv2 instead of MS-CHAPv1. New TLS- based EAP methods MUST provide protected success and failure indications inside of the TLS tunnel.¶

When the inner authentication protocol indicates that authentication has failed, then implementations MUST fail authentication for the entire session. There may be additional protocol exchanges in order to exchange more detailed failure indications, but the final result MUST be a failed authentication. As noted earlier, any session tickets for this failed authentication MUST be either invalidated or discarded.¶

Similarly, when the inner authentication protocol indicates that authentication has succeeded, implementations SHOULD cause authentication to succeed for the entire session. There MAY be additional protocol exchanges that could still cause failure, so we cannot mandate sending success on successful authentication.¶

In both of these cases, the EAP server MUST send an EAP-Failure or EAP-Success message, as indicated by Step 4 in Section 2 of [RFC3748]. Even though both parties have already determined the final authentication status, the full EAP state machine must still be followed.¶