Ephemeral Diffie-Hellman Over COSE (EDHOC)

1.1. Motivation

Many Internet of Things (IoT) deployments require technologies which are highly performant in constrained environments [RFC7228]. IoT devices may be constrained in various ways, including memory, storage, processing capacity, and power. The connectivity for these settings may also exhibit constraints such as unreliable and lossy channels, highly restricted bandwidth, and dynamic topology. The IETF has acknowledged this problem by standardizing a range of lightweight protocols and enablers designed for the IoT, including the Constrained Application Protocol (CoAP, [RFC7252]), Concise Binary Object Representation (CBOR, [RFC8949]), and Static Context Header Compression (SCHC, [RFC8724]).¶

The need for special protocols targeting constrained IoT deployments extends also to the security domain [I-D.ietf-lake-reqs]. Important characteristics in constrained environments are the number of round trips and protocol message sizes, which if kept low can contribute to good performance by enabling transport over a small number of radio frames, reducing latency due to fragmentation or duty cycles, etc. Another important criterion is code size, which may be prohibitively large for certain deployments due to device capabilities or network load during firmware update. Some IoT deployments also need to support a variety of underlying transport technologies, potentially even with a single connection.¶

Some security solutions for such settings exist already. CBOR Object Signing and Encryption (COSE, [RFC9052]) specifies basic application-layer security services efficiently encoded in CBOR. Another example is Object Security for Constrained RESTful Environments (OSCORE, [RFC8613]) which is a lightweight communication security extension to CoAP using CBOR and COSE. In order to establish good quality cryptographic keys for security protocols such as COSE and OSCORE, the two endpoints may run an authenticated Diffie-Hellman key exchange protocol, from which shared secret keying material can be derived. Such a key exchange protocol should also be lightweight; to prevent bad performance in case of repeated use, e.g., due to device rebooting or frequent rekeying for security reasons; or to avoid latencies in a network formation setting with many devices authenticating at the same time.¶

This document specifies Ephemeral Diffie-Hellman Over COSE (EDHOC), a lightweight authenticated key exchange protocol providing good security properties including forward secrecy, identity protection, and cipher suite negotiation. Authentication can be based on raw public keys (RPK) or public key certificates and requires the application to provide input on how to verify that endpoints are trusted. This specification supports the referencing of credentials in order to reduce message overhead, but credentials may alternatively be embedded in the messages. EDHOC does not currently support pre-shared key (PSK) authentication as authentication with static Diffie-Hellman public keys by reference produces equally small message sizes but with much simpler key distribution and identity protection.¶

EDHOC makes use of known protocol constructions, such as SIGMA [SIGMA], the Noise XX pattern [Noise], and Extract-and-Expand [RFC5869]. EDHOC uses COSE for cryptography and identification of credentials (including COSE_Key, CBOR Web Token (CWT), CWT Claims Set (CCS), X.509, and CBOR encoded X.509 (C509) certificates, see Section 3.5.2). COSE provides crypto agility and enables the use of future algorithms and credential types targeting IoT.¶

EDHOC is designed for highly constrained settings making it especially suitable for low-power networks [RFC8376] such as Cellular IoT, 6TiSCH, and LoRaWAN. A main objective for EDHOC is to be a lightweight authenticated key exchange for OSCORE, i.e., to provide authentication and session key establishment for IoT use cases such as those built on CoAP [RFC7252] involving 'things' with embedded microcontrollers, sensors, and actuators. By reusing the same lightweight primitives as OSCORE (CBOR, COSE, CoAP) the additional code size can be kept very low. Note that while CBOR and COSE primitives are built into the protocol messages, EDHOC is not bound to a particular transport.¶

A typical setting is when one of the endpoints is constrained or in a constrained network, and the other endpoint is a node on the Internet (such as a mobile phone). Thing-to-thing interactions over constrained networks are also relevant since both endpoints would then benefit from the lightweight properties of the protocol. EDHOC could, e.g., be run when a device connects for the first time, or to establish fresh keys which are not revealed by a later compromise of the long-term keys.¶

1.2. Message Size Examples

Examples of EDHOC message sizes are shown in Figure 1, using different kinds of authentication keys and COSE header parameters for identification: static Diffie-Hellman keys or signature keys, either in CBOR Web Token (CWT) / CWT Claims Set (CCS) [RFC8392] identified by a key identifier using 'kid' [RFC9052], or in X.509 certificates identified by a hash value using 'x5t' [RFC9360]. As a comparison, in the case of RPK authentication, the EDHOC message size when transferred in CoAP can be less than 1/7 of the DTLS 1.3 handshake [RFC9147] with ECDHE and connection ID, see Section 2 of [I-D.ietf-iotops-security-protocol-comparison].¶

1.3. Document Structure

The remainder of the document is organized as follows: Section 2 outlines EDHOC authenticated with signature keys, Section 3 describes the protocol elements of EDHOC, including formatting of the ephemeral public keys, Section 4 specifies the key derivation, Section 5 specifies message processing for EDHOC authenticated with signature keys or static Diffie-Hellman keys, Section 6 describes the error messages, and Section 8 lists compliance requirements. Note that normative text is also used in appendices, in particular Appendix A.¶

1.4. Terminology and 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.¶

Readers are expected to be familiar with the terms and concepts described in CBOR [RFC8949], CBOR Sequences [RFC8742], COSE structures and processing [RFC9052], COSE algorithms [RFC9053], CWT and CWT Claims Set [RFC8392], and the Concise Data Definition Language (CDDL, [RFC8610]), which is used to express CBOR data structures. Examples of CBOR and CDDL are provided in Appendix C.1. When referring to CBOR, this specification always refers to Deterministically Encoded CBOR as specified in Sections 4.2.1 and 4.2.2 of [RFC8949]. The single output from authenticated encryption (including the authentication tag) is called "ciphertext", following [RFC5116].¶

3.1. General

The EDHOC protocol consists of three mandatory messages (message_1, message_2, message_3), an optional fourth message (message_4), and an error message, between an Initiator (I) and a Responder (R). The odd messages are sent by I, the even by R. Both I and R can send error messages. The roles have slightly different security properties which should be considered when the roles are assigned, see Section 9.1. All EDHOC messages are CBOR Sequences [RFC8742], and are deterministically encoded. Figure 3 illustrates an EDHOC message flow with the optional fourth message as well as the content of each message. The protocol elements in the figure are introduced in Section 3 and Section 5. Message formatting and processing are specified in Section 5 and Section 6.¶

Application data may be protected using the agreed application algorithms (AEAD, hash) in the selected cipher suite (see Section 3.6) and the application can make use of the established connection identifiers C_I and C_R (see Section 3.3). Media types that may be used for EDHOC are defined in Section 10.8.¶

The Initiator can derive symmetric application keys after creating EDHOC message_3, see Section 4.2.1. Protected application data can therefore be sent in parallel or together with EDHOC message_3. EDHOC message_4 is typically not sent.¶

3.2. Method

The data item METHOD in message_1 (see Section 5.2.1), is an integer specifying the authentication method. EDHOC supports authentication with signature or static Diffie-Hellman keys, as defined in the four authentication methods: 0, 1, 2, and 3, see Figure 4. When using a static Diffie-Hellman key the authentication is provided by a Message Authentication Code (MAC) computed from an ephemeral-static ECDH shared secret which enables significant reductions in message sizes. Note that also in the static Diffie-Hellman based authentication methods there is an ephemeral-ephemeral Diffie-Hellman key exchange.¶

The Initiator and the Responder need to have agreed on a single method to be used for EDHOC, see Section 3.9.¶

EDHOC does not have a dedicated message field to indicate the protocol version. Breaking changes to EDHOC can be introduced by specifying and registering new methods.¶

3.3. Connection Identifiers

EDHOC includes the selection of connection identifiers (C_I, C_R) identifying a connection for which keys are agreed.¶

Connection identifiers may be used to correlate EDHOC messages and facilitate the retrieval of protocol state during an EDHOC session (see Section 3.4), or may be used in applications of EDHOC, e.g., in OSCORE (see Section 3.3.3). The connection identifiers do not have any cryptographic purpose in EDHOC and only facilitate the retrieval of security data associated with the protocol state.¶

Connection identifiers in EDHOC are intrinsically byte strings. Most constrained devices only have a few connections for which short identifiers may be sufficient. In some cases minimum length identifiers are necessary to comply with overhead requirements. However, CBOR byte strings - with the exception of the empty byte string h’’ which encodes as one byte (0x40) - are encoded as two or more bytes. To enable one-byte encoding of certain byte strings while maintaining CBOR encoding, EDHOC represents certain identifiers as CBOR integers on the wire, see Section 3.3.2.¶

Integer:                -24  -23   ...   -2   -1    0    1   ...   23
CBOR encoding (1 byte):  37   36   ...   21   20   00   01   ...   17

3.4. Transport

Cryptographically, EDHOC does not put requirements on the lower layers. EDHOC is not bound to a particular transport layer and can even be used in environments without IP. In addition to the transport of messages including errors, the transport is responsible, where necessary, to handle:¶

• message loss,¶
• message reordering,¶
• message duplication,¶
• flow control,¶
• fragmentation and reassembly,¶
• demultiplexing EDHOC messages from other types of messages,¶
• denial-of-service protection,¶
• message correlation, see Section 3.4.1.¶

EDHOC does not require error free transport since a change in message content is detected through the transcript hashes in a subsequent integrity verification, see Section 5.¶

EDHOC is designed to enable an authenticated key exchange with small messages, where the minimum message sizes are of the order illustrated in the first column of Figure 1. There is no maximum message size specified by the protocol; this is for example dependent on the size of authentication credentials (if they are transported, see Section 3.5).¶

The use of transport is specified in the application profile, which in particular may specify limitations in message sizes, see Section 3.9.¶

3.5. Authentication Parameters

EDHOC supports various settings for how the other endpoint's authentication (public) key may be transported, identified, and trusted.¶

EDHOC performs the following authentication related operations:¶

• EDHOC transports information about credentials in ID_CRED_I and ID_CRED_R (described in Section 3.5.3). Based on this information, the authentication credentials CRED_I and CRED_R (described in Section 3.5.2) can be obtained. EDHOC may also transport certain authentication related information as External Authorization Data (see Section 3.8).¶

• EDHOC uses the authentication credentials in two ways (see Section 5.3.2 and Section 5.4.2):¶

  • The authentication credential is input to the integrity verification using the MAC fields.¶
  • The authentication key of the authentication credential is used with the Signature_or_MAC field to verify proof-of-possession of the private key.¶

Other authentication related verifications are out of scope for EDHOC, and is the responsibility of the application. In particular, the authentication credential needs to be validated in the context of the connection for which EDHOC is used, see Appendix D. EDHOC MUST allow the application to read received information about credential (ID_CRED_R, ID_CRED_I). EDHOC MUST have access to the authentication key and the authentication credential.¶

Note that the type of authentication key, authentication credential, and the identification of the credential have a large impact on the message size. For example, the Signature_or_MAC field is much smaller with a static DH key than with a signature key. A CWT Claims Set (CCS) is much smaller than a self-signed certificate/CWT, but if it is possible to reference the credential with a COSE header like 'kid', then that is in turn much smaller than a CCS.¶

{                                              /CCS/
  2 : "42-50-31-FF-EF-37-32-39",               /sub/
  8 : {                                        /cnf/
    1 : {                                      /COSE_Key/
      1 : 1,                                   /kty/
      2 : h'00',                               /kid/
     -1 : 4,                                   /crv/
     -2 : h'b1a3e89460e88d3a8d54211dc95f0b90   /x/
            3ff205eb71912d6db8f4af980d2db83a'
    }
  }
}

3.6. Cipher Suites

An EDHOC cipher suite consists of an ordered set of algorithms from the "COSE Algorithms" and "COSE Elliptic Curves" registries as well as the EDHOC MAC length. All algorithm names and definitions follow from COSE algorithms [RFC9053]. Note that COSE sometimes uses peculiar names such as ES256 for ECDSA with SHA-256, A128 for AES-128, and Ed25519 for the curve edwards25519. Algorithms need to be specified with enough parameters to make them completely determined. The EDHOC MAC length MUST be at least 8 bytes. Any cryptographic algorithm used in the COSE header parameters in ID_CRED fields is selected independently of the selected cipher suite. EDHOC is currently only specified for use with key exchange algorithms of type ECDH curves, but any Key Encapsulation Method (KEM), including Post-Quantum Cryptography (PQC) KEMs, can be used in method 0, see Section 9.4. Use of other types of key exchange algorithms to replace static DH authentication (method 1,2,3) would likely require a specification updating EDHOC with new methods.¶

EDHOC supports all signature algorithms defined by COSE. Just like in (D)TLS 1.3 [RFC8446][RFC9147] and IKEv2 [RFC7296], a signature in COSE is determined by the signature algorithm and the authentication key algorithm together, see Section 3.5.1. The exact details of the authentication key algorithm depend on the type of authentication credential. COSE supports different formats for storing the public authentication keys including COSE_Key and X.509, which use different names and ways to represent the authentication key and the authentication key algorithm.¶

An EDHOC cipher suite consists of the following parameters:¶

• EDHOC AEAD algorithm¶
• EDHOC hash algorithm¶
• EDHOC MAC length in bytes (Static DH)¶
• EDHOC key exchange algorithm (ECDH curve)¶
• EDHOC signature algorithm¶
• Application AEAD algorithm¶
• Application hash algorithm¶

Each cipher suite is identified with a pre-defined integer label.¶

EDHOC can be used with all algorithms and curves defined for COSE. Implementations can either use any combination of COSE algorithms and parameters to define their own private cipher suite, or use one of the pre-defined cipher suites. Private cipher suites can be identified with any of the four values -24, -23, -22, -21. The pre-defined cipher suites are listed in the IANA registry (Section 10.2) with initial content outlined here:¶

• Cipher suites 0-3, based on AES-CCM, are intended for constrained IoT where message overhead is a very important factor. Note that AES-CCM-16-64-128 and AES-CCM-16-128-128 are compatible with the IEEE CCM* mode.¶

  • Cipher suites 1 and 3 use a larger tag length (128-bit) in EDHOC than in the Application AEAD algorithm (64-bit).¶
• Cipher suites 4 and 5, based on ChaCha20, are intended for less constrained applications and only use 128-bit tag lengths.¶
• Cipher suite 6, based on AES-GCM, is for general non-constrained applications. It consists of high performance algorithms that are widely used in non-constrained applications.¶
• Cipher suites 24 and 25 are intended for high security applications such as government use and financial applications. These cipher suites do not share any algorithms. Cipher suite 24 consists of algorithms from the CNSA suite [CNSA].¶

The different methods (Section 3.2) use the same cipher suites, but some algorithms are not used in some methods. The EDHOC signature algorithm is not used in methods without signature authentication.¶

The Initiator needs to have a list of cipher suites it supports in order of preference. The Responder needs to have a list of cipher suites it supports. SUITES_I contains cipher suites supported by the Initiator, formatted and processed as detailed in Section 5.2.1 to secure the cipher suite negotiation. Examples of cipher suite negotiation are given in Section 6.3.2.¶

3.7. Ephemeral Public Keys

The ephemeral public keys in EDHOC (G_X and G_Y) use compact representation of elliptic curve points, see Appendix B. In COSE, compact representation is achieved by formatting the ECDH ephemeral public keys as COSE_Keys of type EC2 or OKP according to Sections 7.1 and 7.2 of [RFC9053], but only including the 'x' parameter in G_X and G_Y. For Elliptic Curve Keys of type EC2, compact representation MAY be used also in the COSE_Key. COSE always uses compact output for Elliptic Curve Keys of type EC2. If the COSE implementation requires a 'y' parameter, the value y = false or a calculated y-coordinate can be used, see Appendix B.¶

3.8. External Authorization Data (EAD)

In order to reduce round trips and the number of messages, or to simplify processing, external security applications may be integrated into EDHOC by transporting authorization related data in the messages.¶

EDHOC allows processing of external authorization data (EAD) to be defined in a separate specification, and sent in dedicated fields of the four EDHOC messages (EAD_1, EAD_2, EAD_3, EAD_4). EAD is opaque data to EDHOC.¶

Each EAD field, EAD_x for x = 1, 2, 3 or 4, is a CBOR sequence (see Appendix C.1) consisting of one or more EAD items. An EAD item ead is a CBOR sequence of an ead_label and an optional ead_value, see Figure 7 and Appendix C.2 for the CDDL definitions.¶

ead = (
  ead_label : int,
  ? ead_value : bstr,
)

A security application may register one or more EAD labels, see Section 10.5, and specify the associated processing and security considerations. The IANA registry contains the absolute value of the ead_label, |ead_label|; the same ead_value applies independently of sign of ead_label.¶

An EAD item can be either critical or non-critical, determined by the sign of the ead_label in the EAD item transported in the EAD field. A negative value indicates that the EAD item is critical and a non-negative value indicates that the EAD item is non-critical.¶

If an endpoint receives a critical EAD item it does not recognize, or a critical EAD item that contains information that it cannot process, then the endpoint MUST send an EDHOC error message back as defined in Section 6, and the EDHOC session MUST be aborted. The EAD item specification defines the error processing. A non-critical EAD item can be ignored.¶

The security application registering a new EAD item needs to describe under what conditions the EAD item is critical or non-critical, and thus whether the ead_label is used with negative or positive sign. ead_label = 0 is used for padding, see Section 3.8.1.¶

The security application may define multiple uses of certain EAD items, e.g., the same EAD item may be used in different EDHOC messages. Multiple occurrences of an EAD item in one EAD field may also be specified, but the criticality of the repeated EAD item is expected to be the same.¶

The EAD fields of EDHOC MUST only be used with registered EAD items, see Section 10.5. Examples of the use of EAD are provided in Appendix E.¶

3.9. Application Profile

EDHOC requires certain parameters to be agreed upon between Initiator and Responder. Some parameters can be negotiated through the protocol execution (specifically, cipher suite, see Section 3.6) but other parameters are only communicated and may not be negotiated (e.g., which authentication method is used, see Section 3.2). Yet other parameters need to be known out-of-band to ensure successful completion, e.g., whether message_4 is used or not. The application decides which endpoint is Initiator and which is Responder.¶

The purpose of an application profile is to describe the intended use of EDHOC to allow for the relevant processing and verifications to be made, including things like:¶

1. How the endpoint detects that an EDHOC message is received. This includes how EDHOC messages are transported, for example in the payload of a CoAP message with a certain Uri-Path or Content-Format; see Appendix A.2.¶

   • The method of transporting EDHOC messages may also describe data carried along with the messages that are needed for the transport to satisfy the requirements of Section 3.4, e.g., connection identifiers used with certain messages, see Appendix A.2.¶
2. Authentication method (METHOD; see Section 3.2).¶
3. Profile for authentication credentials (CRED_I, CRED_R; see Section 3.5.2), e.g., profile for certificate or CCS, including supported authentication key algorithms (subject public key algorithm in X.509 or C509 certificate).¶
4. Type used to identify credentials (ID_CRED_I, ID_CRED_R; see Section 3.5.3).¶
5. Use and type of external authorization data (EAD_1, EAD_2, EAD_3, EAD_4; see Section 3.8).¶
6. Identifier used as the identity of the endpoint; see Appendix D.2.¶
7. If message_4 shall be sent/expected, and if not, how to ensure a protected application message is sent from the Responder to the Initiator; see Section 5.5.¶

The application profile may also contain information about supported cipher suites. The procedure for selecting and verifying a cipher suite is still performed as described in Section 5.2.1 and Section 6.3, but it may become simplified by this knowledge. EDHOC messages can be processed without the application profile, i.e., the EDHOC messages includes information about the type and length of all fields.¶

An example of an application profile is shown in Appendix F.¶

For some parameters, like METHOD, type of ID_CRED field or EAD, the receiver of an EDHOC message is able to verify compliance with the application profile, and if it needs to fail because of lack of compliance, to infer the reason why the EDHOC session failed.¶

For other encodings, like the profiling of CRED_x in the case that it is not transported, it may not be possible to verify that lack of compliance with the application profile was the reason for failure: Integrity verification in message_2 or message_3 may fail not only because of wrong credential. For example, in case the Initiator uses a public key certificate by reference (i.e., not transported within the protocol) then both endpoints need to use an identical data structure as CRED_I or else the integrity verification will fail.¶

Note that it is not necessary for the endpoints to specify a single transport for the EDHOC messages. For example, a mix of CoAP and HTTP may be used along the path, and this may still allow correlation between messages.¶

The application profile may be dependent on the identity of the other endpoint, or other information carried in an EDHOC message, but it then applies only to the later phases of the protocol when such information is known. (The Initiator does not know the identity of the Responder before having verified message_2, and the Responder does not know the identity of the Initiator before having verified message_3.)¶

Other conditions may be part of the application profile, such as what is the target application or use (if there is more than one application/use) to the extent that EDHOC can distinguish between them. In case multiple application profiles are used, the receiver needs to be able to determine which is applicable for a given EDHOC session, for example based on URI to which the EDHOC message is sent, or external authorization data type.¶

4.1. Keys for EDHOC Message Processing

EDHOC uses Extract-and-Expand [RFC5869] with the EDHOC hash algorithm in the selected cipher suite to derive keys used in message processing. This section defines EDHOC_Extract (Section 4.1.1) and EDHOC_Expand (Section 4.1.2), and how to use them to derive PRK_out (Section 4.1.3) which is the shared secret session key resulting from a completed EDHOC session.¶

EDHOC_Extract is used to derive fixed-length uniformly pseudorandom keys (PRK) from ECDH shared secrets. EDHOC_Expand is used to define EDHOC_KDF for generating MACs and for deriving output keying material (OKM) from PRKs.¶

In EDHOC a specific message is protected with a certain pseudorandom key, but how the key is derived depends on the authentication method (Section 3.2) as detailed in Section 5.¶

   PRK = EDHOC_Extract( salt, IKM )

   G_XY = X25519( Y, G_X ) = X25519( X, G_Y )

   PRK_2e = HMAC-SHA-256( TH_2, G_XY )

   OKM = EDHOC_KDF( PRK, info_label, context, length )
       = EDHOC_Expand( PRK, info, length )

info = (
  info_label : int,
  context : bstr,
  length : uint,
)

KEYSTREAM_2   = EDHOC_KDF( PRK_2e,   0, TH_2,      plaintext_length )
SALT_3e2m     = EDHOC_KDF( PRK_2e,   1, TH_2,      hash_length )
MAC_2         = EDHOC_KDF( PRK_3e2m, 2, context_2, mac_length_2 )
K_3           = EDHOC_KDF( PRK_3e2m, 3, TH_3,      key_length )
IV_3          = EDHOC_KDF( PRK_3e2m, 4, TH_3,      iv_length )
SALT_4e3m     = EDHOC_KDF( PRK_3e2m, 5, TH_3,      hash_length )
MAC_3         = EDHOC_KDF( PRK_4e3m, 6, context_3, mac_length_3 )
PRK_out       = EDHOC_KDF( PRK_4e3m, 7, TH_4,      hash_length )
K_4           = EDHOC_KDF( PRK_4e3m, 8, TH_4,      key_length )
IV_4          = EDHOC_KDF( PRK_4e3m, 9, TH_4,      iv_length )
PRK_exporter  = EDHOC_KDF( PRK_out, 10, h'',       hash_length )

4.2. Keys for EDHOC Applications

This section defines EDHOC_Exporter in terms of EDHOC_KDF and PRK_exporter. A key update function is defined in Appendix H.¶

   EDHOC_Exporter(exporter_label, context, length)
     = EDHOC_KDF(PRK_exporter, exporter_label, context, length)

5.1. EDHOC Message Processing Outline

For each new/ongoing EDHOC session, the endpoints are assumed to keep an associated protocol state containing identifiers, keying material, etc. used for subsequent processing of protocol related data. The protocol state is assumed to be associated with an application profile (Section 3.9) which provides the context for how messages are transported, identified, and processed.¶

EDHOC messages SHALL be processed according to the current protocol state. The following steps are expected to be performed at reception of an EDHOC message:¶

1. Detect that an EDHOC message has been received, for example by means of port number, URI, or media type (Section 3.9).¶
2. Retrieve the protocol state according to the message correlation, see Section 3.4.1. If there is no protocol state, in the case of message_1, a new protocol state is created. The Responder endpoint needs to make use of available Denial-of-Service mitigation (Section 9.7).¶
3. If the message received is an error message, then process it according to Section 6, else process it as the expected next message according to the protocol state.¶

The message processing steps SHALL be processed in order, unless otherwise stated. If the processing fails for some reason then, typically, an error message is sent, the EDHOC session is aborted, and the protocol state erased. When the composition and sending of one message is completed and before the next message is received, error messages SHALL NOT be sent.¶

After having successfully processed the last message (message_3 or message_4 depending on application profile) the EDHOC session is completed, after which no error messages are sent and EDHOC session output MAY be maintained even if error messages are received. Further details are provided in the following subsections and in Section 6.¶

Different instances of the same message MUST NOT be processed in one EDHOC session. Note that processing will fail if the same message appears a second time for EDHOC processing in the same EDHOC session because the state of the protocol has moved on and now expects something else. Message deduplication MUST be done by the transport protocol (see Section 3.4) or, if not supported by the transport, as described in Section 7.¶

5.2. EDHOC Message 1

message_1 = (
  METHOD : int,
  SUITES_I : suites,
  G_X : bstr,
  C_I : bstr / -24..23,
  ? EAD_1,
)

suites = [ 2* int ] / int
EAD_1 = 1* ead

5.3. EDHOC Message 2

message_2 = (
  G_Y_CIPHERTEXT_2 : bstr,
)

5.4. EDHOC Message 3

message_3 = (
  CIPHERTEXT_3 : bstr,
)

5.5. EDHOC Message 4

This section specifies message_4 which is OPTIONAL to support. Key confirmation is normally provided by sending an application message from the Responder to the Initiator protected with a key derived with the EDHOC_Exporter, e.g., using OSCORE (see Appendix A). In deployments where no protected application message is sent from the Responder to the Initiator, message_4 MUST be supported and MUST be used. Two examples of such deployments are:¶

1. When EDHOC is only used for authentication and no application data is sent.¶
2. When application data is only sent from the Initiator to the Responder.¶

Further considerations about when to use message_4 are provided in Section 3.9 and Section 9.1.¶

message_4 = (
  CIPHERTEXT_4 : bstr,
)

6.1. Success

Error code 0 MAY be used internally in an application to indicate success, i.e., as a standard value in case of no error, e.g., in status reporting or log files. Error code 0 MUST NOT be used as part of the EDHOC message exchange. If an endpoint receives an error message with error code 0, then it MUST abort the EDHOC session and MUST NOT send an error message.¶

6.2. Unspecified Error

Error code 1 is used for errors that do not have a specific error code defined. ERR_INFO MUST be a text string containing a human-readable diagnostic message written in English, for example "Method not supported". The diagnostic text message is mainly intended for software engineers that during debugging need to interpret it in the context of the EDHOC specification. The diagnostic message SHOULD be provided to the calling application where it SHOULD be logged.¶

6.3. Wrong Selected Cipher Suite

Error code 2 MUST only be used when replying to message_1 in case the cipher suite selected by the Initiator is not supported by the Responder, or if the Responder supports a cipher suite more preferred by the Initiator than the selected cipher suite, see Section 5.2.3. In this case, ERR_INFO = SUITES_R and is of type suites, see Section 5.2.1. If the Responder does not support the selected cipher suite, then SUITES_R MUST include one or more supported cipher suites. If the Responder supports a cipher suite in SUITES_I other than the selected cipher suite (independently of if the selected cipher suite is supported or not) then SUITES_R MUST include the supported cipher suite in SUITES_I which is most preferred by the Initiator. SUITES_R MAY include a single cipher suite, in which case it is encoded as an int. If the Responder does not support any cipher suite in SUITES_I, then it SHOULD include all its supported cipher suites in SUITES_R.¶

In contrast to SUITES_I, the order of the cipher suites in SUITES_R has no significance.¶

6.4. Unknown Credential Referenced

Error code 3 is used for errors due to a received credential identifier (ID_CRED_R in message_2 or ID_CRED_I message_3) containing a reference to a credential which the receiving endpoint does not have access to. The intent with this error code is that the endpoint who sent the credential identifier should for the next EDHOC session try another credential identifier supported according to the application profile.¶

For example, an application profile could list x5t and x5chain as supported credential identifiers, and state that x5t should be used if it can be assumed that the X.509 certificate is available at the receiving side. This error code thus enables the certificate chain to be sent only when needed, bearing in mind that error messages are not protected so an adversary can try to cause unnecessary large credential identifiers.¶

For the error code 3, the error information SHALL be the CBOR simple value true (0xf5). Error code 3 MUST NOT be used when the received credential identifier type is not supported.¶

9.1. Security Properties

EDHOC has similar security properties as can be expected from the theoretical SIGMA-I protocol [SIGMA] and the Noise XX pattern [Noise], which are similar to methods 0 and 3, respectively. Proven security properties are detailed in the security analysis publications referenced at the end of this section.¶

Using the terminology from [SIGMA], EDHOC provides forward secrecy, mutual authentication with aliveness, consistency, and peer awareness. As described in [SIGMA], message_3 provides peer awareness to the Responder while message_4 provides peer awareness to the Initiator. By including the authentication credentials in the transcript hash, EDHOC protects against Duplicate Signature Key Selection (DSKS)-like identity mis-binding attack that the MAC-then-Sign variant of SIGMA-I is otherwise vulnerable to.¶

As described in [SIGMA], different levels of identity protection are provided to the Initiator and the Responder. EDHOC provides identity protection of the Initiator against active attacks and identity protection of the Responder against passive attacks. An active attacker can get the credential identifier of the Responder by eavesdropping on the destination address used for transporting message_1 and then sending its own message_1 to the same address. The roles should be assigned to protect the most sensitive identity/identifier, typically that which is not possible to infer from routing information in the lower layers.¶

EDHOC messages might change in transit due to a noisy channel or through modification by an attacker. Changes in message_1 and message_2 (except Signature_or_MAC_2 when the signature scheme is not strongly unforgeable) are detected when verifying Signature_or_MAC_2. Changes to not strongly unforgeable Signature_or_MAC_2, and message_3 are detected when verifying CIPHERTEXT_3. Changes to message_4 are detected when verifying CIPHERTEXT_4.¶

Compared to [SIGMA], EDHOC adds an explicit method type and expands the message authentication coverage to additional elements such as algorithms, external authorization data, and previous plaintext messages. This protects against an attacker replaying messages or injecting messages from another EDHOC session.¶

EDHOC also adds selection of connection identifiers and downgrade protected negotiation of cryptographic parameters, i.e., an attacker cannot affect the negotiated parameters. A single session of EDHOC does not include negotiation of cipher suites, but it enables the Responder to verify that the selected cipher suite is the most preferred cipher suite by the Initiator which is supported by both the Initiator and the Responder, and to abort the EDHOC session if not.¶

As required by [RFC7258], IETF protocols need to mitigate pervasive monitoring when possible. EDHOC therefore only supports methods with ephemeral Diffie-Hellman and provides a key update function (see Appendix H) for lightweight application protocol rekeying. Either of these provides forward secrecy, in the sense that compromise of the private authentication keys does not compromise past session keys (PRK_out), and compromise of a session key does not compromise past session keys. Frequently re-running EDHOC with ephemeral Diffie-Hellman forces attackers to perform dynamic key exfiltration where the attacker must have continuous interactions with the collaborator, which is a significant sustained attack.¶

To limit the effect of breaches, it is important to limit the use of symmetric group keys for bootstrapping. EDHOC therefore strives to make the additional cost of using raw public keys and self-signed certificates as small as possible. Raw public keys and self-signed certificates are not a replacement for a public key infrastructure but SHOULD be used instead of symmetric group keys for bootstrapping.¶

Compromise of the long-term keys (private signature or static DH keys) does not compromise the security of completed EDHOC sessions. Compromising the private authentication keys of one party lets an active attacker impersonate that compromised party in EDHOC sessions with other parties but does not let the attacker impersonate other parties in EDHOC sessions with the compromised party. Compromise of the long-term keys does not enable a passive attacker to compromise future session keys (PRK_out). Compromise of the HDKF input parameters (ECDH shared secret) leads to compromise of all session keys derived from that compromised shared secret. Compromise of one session key does not compromise other session keys. Compromise of PRK_out leads to compromise of all keying material derived with the EDHOC_Exporter.¶

Based on the cryptographic algorithms requirements Section 9.3, EDHOC provides a minimum of 64-bit security against online brute force attacks and a minimum of 128-bit security against offline brute force attacks. To break 64-bit security against online brute force an attacker would on average have to send 4.3 billion messages per second for 68 years, which is infeasible in constrained IoT radio technologies. A forgery against a 64-bit MAC in EDHOC breaks the security of all future application data, while a forgery against a 64-bit MAC in the subsequent application protocol (e.g., OSCORE [RFC8613]) typically only breaks the security of the data in the forged packet.¶

As the EDHOC session is aborted when verification fails, the security against online attacks is given by the sum of the strength of the verified signatures and MACs (including MAC in AEAD). As an example, if EDHOC is used with method 3, cipher suite 2, and message_4, the Responder is authenticated with 128-bit security against online attacks (the sum of the 64-bit MACs in message_2 and message_4). The same principle applies for MACs in an application protocol keyed by EDHOC as long as EDHOC is rerun when verification of the first MACs in the application protocol fails. As an example, if EDHOC with method 3 and cipher suite 2 is used as in Figure 2 of [I-D.ietf-core-oscore-edhoc], 128-bit mutual authentication against online attacks can be achieved after completion of the first OSCORE request and response.¶

After sending message_3, the Initiator is assured that no other party than the Responder can compute the key PRK_out. While the Initiator can securely send protected application data, the Initiator SHOULD NOT persistently store the keying material PRK_out until the Initiator has verified message_4 or a message protected with a derived application key, such as an OSCORE message, from the Responder. After verifying message_3, the Responder is assured that an honest Initiator has computed the key PRK_out. The Responder can securely derive and store the keying material PRK_out, and send protected application data.¶

External authorization data sent in message_1 (EAD_1) or message_2 (EAD_2) should be considered unprotected by EDHOC, see Section 9.5. EAD_2 is encrypted but the Responder has not yet authenticated the Initiator and the encryption does not provide confidentiality against active attacks.¶

External authorization data sent in message_3 (EAD_3) or message_4 (EAD_4) is protected between Initiator and Responder by the protocol, but note that EAD fields may be used by the application before the message verification is completed, see Section 3.8. Designing a secure mechanism that uses EAD is not necessarily straightforward. This document only provides the EAD transport mechanism, but the problem of agreeing on the surrounding context and the meaning of the information passed to and from the application remains. Any new uses of EAD should be subject to careful review.¶

Key compromise impersonation (KCI): In EDHOC authenticated with signature keys, EDHOC provides KCI protection against an attacker having access to the long-term key or the ephemeral secret key. With static Diffie-Hellman key authentication, KCI protection would be provided against an attacker having access to the long-term Diffie-Hellman key, but not to an attacker having access to the ephemeral secret key. Note that the term KCI has typically been used for compromise of long-term keys, and that an attacker with access to the ephemeral secret key can only attack that specific EDHOC session.¶

Repudiation: If an endpoint authenticates with a signature, the other endpoint can prove that the endpoint performed a run of the protocol by presenting the data being signed as well as the signature itself. With static Diffie-Hellman key authentication, the authenticating endpoint can deny having participated in the protocol.¶

Earlier versions of EDHOC have been formally analyzed [Bruni18] [Norrman20] [CottierPointcheval22] [Jacomme23] [GuentherIlunga22] and the specification has been updated based on the analysis.¶

9.2. Cryptographic Considerations

The SIGMA protocol requires that the encryption of message_3 provides confidentiality against active attackers and EDHOC message_4 relies on the use of authenticated encryption. Hence the message authenticating functionality of the authenticated encryption in EDHOC is critical: authenticated encryption MUST NOT be replaced by plain encryption only, even if authentication is provided at another level or through a different mechanism.¶

To reduce message overhead EDHOC does not use explicit nonces and instead relies on the ephemeral public keys to provide randomness to each EDHOC session. A good amount of randomness is important for the key generation, to provide liveness, and to protect against interleaving attacks. For this reason, the ephemeral keys MUST NOT be used in more than one EDHOC message, and both parties SHALL generate fresh random ephemeral key pairs. Note that an ephemeral key may be used to calculate several ECDH shared secrets. When static Diffie-Hellman authentication is used the same ephemeral key is used in both ephemeral-ephemeral and ephemeral-static ECDH.¶

As discussed in [SIGMA], the encryption of message_2 does only need to protect against passive attacker as active attackers can always get the Responder's identity by sending their own message_1. EDHOC uses the EDHOC_Expand function (typically HKDF-Expand) as a binary additive stream cipher which is proven secure as long as the expand function is a PRF. HKDF-Expand is not often used as a stream cipher as it is slow on long messages, and most applications require both confidentiality with indistinguishability under chosen ciphertext (IND-CCA) as well as integrity protection. For the encryption of message_2, any speed difference is negligible, IND-CCA does not increase security, and integrity is provided by the inner MAC (and signature depending on method).¶

Requirements for how to securely generate, validate, and process the ephemeral public keys depend on the elliptic curve. For X25519 and X448, the requirements are defined in [RFC7748]. For secp256r1, secp384r1, and secp521r1, the requirements are defined in Section 5 of [SP-800-56A]. For secp256r1, secp384r1, and secp521r1, at least partial public-key validation MUST be done.¶

The same authentication credential MAY be used for both the Initiator and Responder roles. As noted in Section 12 of [RFC9052] the use of a single key for multiple algorithms is strongly discouraged unless proven secure by a dedicated cryptographic analysis. In particular this recommendation applies to using the same private key for static Diffie-Hellman authentication and digital signature authentication. A preliminary conjecture is that a minor change to EDHOC may be sufficient to fit the analysis of secure shared signature and ECDH key usage in [Degabriele11] and [Thormarker21].¶

The property that a completed EDHOC session implies that another identity has been active is upheld as long as the Initiator does not have its own identity in the set of Responder identities it is allowed to communicate with. In Trust on first use (TOFU) use cases, see Appendix D.5, the Initiator should verify that the Responder's identity is not equal to its own. Any future EDHOC methods using e.g., pre-shared keys might need to mitigate this in other ways. However, an active attacker can gain information about the set of identities an Initiator is willing to communicate with. If the Initiator is willing to communicate with all identities except its own an attacker can determine that a guessed Initiator identity is correct. To not leak any long-term identifiers, using a freshly generated authentication key as identity in each initial TOFU session is RECOMMENDED.¶

9.3. Cipher Suites and Cryptographic Algorithms

When using private cipher suite or registering new cipher suites, the choice of key length used in the different algorithms needs to be harmonized, so that a sufficient security level is maintained for authentication credentials, the EDHOC session, and the protection of application data. The Initiator and the Responder should enforce a minimum security level.¶

The output size of the EDHOC hash algorithm MUST be at least 256-bits, i.e., the hash algorithms SHA-1 and SHA-256/64 (SHA-256 truncated to 64-bits) SHALL NOT be supported for use in EDHOC except for certificate identification with x5t and c5t. For security considerations of SHA-1, see [RFC6194]. As EDHOC integrity protects the whole authentication credentials, the choice of hash algorithm in x5t and c5t does not affect security, and using the same hash algorithm as in the cipher suite, but with as much truncation as possible, is RECOMMENDED. That is, when the EDHOC hash algorithm is SHA-256, using SHA-256/64 in x5t and c5t is RECOMMENDED. The EDHOC MAC length MUST be at least 8 bytes and the tag length of the EDHOC AEAD algorithm MUST be at least 64-bits. Note that secp256k1 is only defined for use with ECDSA and not for ECDH. Note that some COSE algorithms are marked as not recommended in the COSE IANA registry.¶

9.4. Post-Quantum Considerations

As of the publication of this specification, it is unclear when or even if a quantum computer of sufficient size and power to exploit public key cryptography will exist. Deployments that need to consider risks decades into the future should transition to Post-Quantum Cryptography (PQC) in the not-too-distant future. Many other systems should take a slower wait-and-see approach where PQC is phased in when the quantum threat is more imminent. Current PQC algorithms have limitations compared to Elliptic Curve Cryptography (ECC) and the data sizes would be problematic in many constrained IoT systems.¶

Symmetric algorithms used in EDHOC such as SHA-256 and AES-CCM-16-64-128 are practically secure against even large quantum computers. EDHOC supports all signature algorithms defined by COSE, including PQC signature algorithms such as HSS-LMS. EDHOC is currently only specified for use with key exchange algorithms of type ECDH curves, but any Key Encapsulation Method (KEM), including PQC KEMs, can be used in method 0. While the key exchange in method 0 is specified with terms of the Diffie-Hellman protocol, the key exchange adheres to a KEM interface: G_X is then the public key of the Initiator, G_Y is the encapsulation, and G_XY is the shared secret. Use of PQC KEMs to replace static DH authentication would likely require a specification updating EDHOC with new methods.¶

9.5. Unprotected Data and Privacy

The Initiator and the Responder must make sure that unprotected data and metadata do not reveal any sensitive information. This also applies for encrypted data sent to an unauthenticated party. In particular, it applies to EAD_1, ID_CRED_R, EAD_2, and error messages. Using the same EAD_1 in several EDHOC sessions allows passive eavesdroppers to correlate the different sessions. Note that even if ead_value is encrypted outside of EDHOC, the ead_labels in EAD_1 is revealed to passive attackers and the ead_labels in EAD_2 is revealed to active attackers. Another consideration is that the list of supported cipher suites may potentially be used to identify the application. The Initiator and the Responder must also make sure that unauthenticated data does not trigger any harmful actions. In particular, this applies to EAD_1 and error messages.¶

An attacker observing network traffic may use connection identifiers sent in clear in EDHOC or the subsequent application protocol to correlate packets sent on different paths or at different times. The attacker may use this information for traffic flow analysis or to track an endpoint. Application protocols using connection identifiers from EDHOC SHOULD provide mechanisms to update the connection identifiers and MAY provide mechanisms to issue several simultaneously active connection identifiers. See [RFC9000] for a non-constrained example of such mechanisms. Connection identifiers can e.g., be chosen randomly among the set of unused 1-byte connection identifiers. Connection identity privacy mechanisms are only useful when there are not fixed identifiers such as IP address or MAC address in the lower layers.¶

9.6. Updated Internet Threat Model Considerations

Since the publication of [RFC3552] there has been an increased awareness of the need to protect against endpoints that are compromised, malicious, or whose interests simply do not align with the interests of users [I-D.arkko-arch-internet-threat-model-guidance]. [RFC7624] describes an updated threat model for Internet confidentiality, see Section 9.1. [I-D.arkko-arch-internet-threat-model-guidance] further expands the threat model. Implementations and users SHOULD consider these threat models. In particular, even data sent protected to the other endpoint such as ID_CRED fields and EAD fields can be used for tracking, see Section 2.7 of [I-D.arkko-arch-internet-threat-model-guidance].¶

The fields ID_CRED_I, ID_CRED_R, EAD_2, EAD_3, and EAD_4 have variable length, and information regarding the length may leak to an attacker. A passive attacker may, e.g., be able to differentiate endpoints using identifiers of different length. To mitigate this information leakage an implementation may ensure that the fields have fixed length or use padding. An implementation may, e.g., only use fixed length identifiers like 'kid' of length 1. Alternatively, padding may be used (see Section 3.8.1) to hide the true length of, e.g., certificates by value in 'x5chain' or 'c5c'.¶

9.7. Denial-of-Service

EDHOC itself does not provide countermeasures against Denial-of-Service attacks. In particular, by sending a number of new or replayed message_1 an attacker may cause the Responder to allocate state, perform cryptographic operations, and amplify messages. To mitigate such attacks, an implementation SHOULD rely on lower layer mechanisms. For instance, when EDHOC is transferred as an exchange of CoAP messages, the CoAP server can use the Echo option defined in [RFC9175] which forces the CoAP client to demonstrate reachability at its apparent network address. To avoid an additional roundtrip the Initiator can reduce the amplification factor by padding message_1, i.e., using EAD_1, see Section 3.8.1.¶

An attacker can also send faked message_2, message_3, message_4, or error in an attempt to trick the receiving party to send an error message and abort the EDHOC session. EDHOC implementations MAY evaluate if a received message is likely to have been forged by an attacker and ignore it without sending an error message or aborting the EDHOC session.¶

9.8. Implementation Considerations

The availability of a secure random number generator is essential for the security of EDHOC. If no true random number generator is available, a random seed MUST be provided from an external source and used with a cryptographically secure pseudorandom number generator. As each pseudorandom number must only be used once, an implementation needs to get a unique input to the pseudorandom number generator after reboot, or continuously store state in nonvolatile memory. Appendix B.1.1 in [RFC8613] describes issues and solution approaches for writing to nonvolatile memory. Intentionally or unintentionally weak or predictable pseudorandom number generators can be abused or exploited for malicious purposes. [RFC8937] describes a way for security protocol implementations to augment their (pseudo)random number generators using a long-term private key and a deterministic signature function. This improves randomness from broken or otherwise subverted random number generators. The same idea can be used with other secrets and functions such as a Diffie-Hellman function or a symmetric secret and a PRF like HMAC or KMAC. It is RECOMMENDED to not trust a single source of randomness and to not put unaugmented random numbers on the wire.¶

Implementations might consider deriving secret and non-secret randomness from different PRNG/PRF/KDF instances to limit the damage if the PRNG/PRF/KDF turns out to be fundamentally broken. NIST generally forbids deriving secret and non-secret randomness from the same KDF instance, but this decision has been criticized by Krawczyk [HKDFpaper] and doing so is common practice. In addition to IVs, other examples are the challenge in EAP-TTLS, the RAND in 3GPP AKAs, and the Session-Id in EAP-TLS 1.3. Note that part of KEYSTREAM_2 is also non-secret randomness as it is known or predictable to an attacker. As explained by Krawczyk, if any attack is mitigated by the NIST requirement it would mean that the KDF is fully broken and would have to be replaced anyway.¶

For many constrained IoT devices it is problematic to support several crypto primitives. Existing devices can be expected to support either ECDSA or EdDSA. If ECDSA is supported, "deterministic ECDSA" as specified in [RFC6979] MAY be used. Pure deterministic elliptic-curve signatures such as deterministic ECDSA and EdDSA have gained popularity over randomized ECDSA as their security do not depend on a source of high-quality randomness. Recent research has however found that implementations of these signature algorithms may be vulnerable to certain side-channel and fault injection attacks due to their determinism. See e.g., Section 1 of [I-D.irtf-cfrg-det-sigs-with-noise] for a list of attack papers. As suggested in Section 2.1.1 of [RFC9053] this can be addressed by combining randomness and determinism.¶

Appendix D of [I-D.ietf-lwig-curve-representations] describes how Montgomery curves such as X25519 and X448 and (twisted) Edwards curves as curves such as Ed25519 and Ed448 can mapped to and from short-Weierstrass form for implementation on platforms that accelerate elliptic curve group operations in short-Weierstrass form.¶

All private keys, symmetric keys, and IVs MUST be secret. Implementations should provide countermeasures to side-channel attacks such as timing attacks. Intermediate computed values such as ephemeral ECDH keys and ECDH shared secrets MUST be deleted after key derivation is completed.¶

The Initiator and the Responder are responsible for verifying the integrity and validity of certificates. The selection of trusted CAs should be done very carefully and certificate revocation should be supported. The choice of revocation mechanism is left to the application. For example, in case of X.509 certificates, Certificate Revocation Lists [RFC5280] or OCSP [RFC6960] may be used.¶

Similar considerations as for certificates are needed for CWT/CCS. The endpoints are responsible for verifying the integrity and validity of CWT/CCS, and to handle revocation. The application needs to determine what trust anchors are relevant, and have a well-defined trust-establishment process. A self-signed certificate/CWT or CCS appearing in the protocol cannot be a trigger to modify the set of trust anchors. One common way for a new trust anchor to be added to (or removed from) a device is by means firmware upgrade. See [RFC9360] for a longer discussion on trust and validation in constrained devices.¶

Just like for certificates, the contents of the COSE header parameters 'kcwt' and 'kccs' defined in Section 10.6 must be processed as untrusted input. Endpoints that intend to rely on the assertions made by a CWT/CCS obtained from any of these methods need to validate the contents. For 'kccs', which enables transport of raw public keys, the data structure used does not include any protection or verification data. 'kccs' may be used for unauthenticated operations, e.g. trust on first use, with the limitations and caveats entailed, see Appendix D.5.¶

Verification of validity may require the use of a Real-Time Clock (RTC). The private authentication keys MUST be kept secret, only the Responder SHALL have access to the Responder's private authentication key and only the Initiator SHALL have access to the Initiator's private authentication key.¶

The Initiator and the Responder are allowed to select the connection identifier C_I and C_R, respectively, for the other party to use in the ongoing EDHOC session as well as in a subsequent application protocol (e.g., OSCORE [RFC8613]). The choice of connection identifier is not security critical in EDHOC but intended to simplify the retrieval of the right security context in combination with using short identifiers. If the wrong connection identifier of the other party is used in a protocol message it will result in the receiving party not being able to retrieve a security context (which will abort the EDHOC session) or retrieve the wrong security context (which also aborts the EDHOC session as the message cannot be verified).¶

If two nodes unintentionally initiate two simultaneous EDHOC sessions with each other even if they only want to complete a single EDHOC session, they MAY abort the EDHOC session with the lexicographically smallest G_X. Note that in cases where several EDHOC sessions with different parameter sets (method, COSE headers, etc.) are used, an attacker can affect which parameter set will be used by blocking some of the parameter sets.¶

If supported by the device, it is RECOMMENDED that at least the long-term private keys are stored in a Trusted Execution Environment (TEE, see for example [I-D.ietf-teep-architecture]) and that sensitive operations using these keys are performed inside the TEE. To achieve even higher security it is RECOMMENDED that additional operations such as ephemeral key generation, all computations of shared secrets, and storage of the PRK keys can be done inside the TEE. The use of a TEE aims at preventing code within that environment to be tampered with, and preventing data used by such code to be read or tampered with by code outside that environment.¶

Note that HKDF-Expand has a relatively small maximum output length of 255 ⋅ hash_length, where hash_length is the output size in bytes of the EDHOC hash algorithm of the selected cipher suite. This means that when SHA-256 is used as hash algorithm, PLAINTEXT_2 cannot be longer than 8160 bytes. This is probably not a limitation for most intended applications, but to be able to support for example long certificate chains or large external authorization data, there is a backwards compatible method specified in Appendix G.¶

The sequence of transcript hashes in EDHOC (TH_2, TH_3, TH_4) does not make use of a so called running hash. This is a design choice as running hashes are often not supported on constrained platforms.¶

When parsing a received EDHOC message, implementations MUST abort the EDHOC session if the message does not comply with the CDDL for that message. It is RECOMMENDED to abort the EDHOC session if the received EDHOC message is not encoded using deterministic CBOR.¶

10.1. EDHOC Exporter Label Registry

IANA is requested to create a new registry under the new registry group "Ephemeral Diffie-Hellman Over COSE (EDHOC)" as follows:¶

Registry Name: EDHOC Exporter Label¶

Reference: [[this document]]¶

10.2. EDHOC Cipher Suites Registry

IANA is requested to create a new registry under the new registry group "Ephemeral Diffie-Hellman Over COSE (EDHOC)" as follows:¶

Registry Name: EDHOC Cipher Suites¶

Reference: [[this document]]¶

The columns of the registry are Value, Array and Description, where Value is an integer and the other columns are text strings. The initial contents of the registry are:¶

Value: -24
Array: N/A
Description: Private Use
Reference: [[this document]]

Value: -23
Array: N/A
Description: Private Use
Reference: [[this document]]

Value: -22
Array: N/A
Description: Private Use
Reference: [[this document]]

Value: -21
Array: N/A
Description: Private Use
Reference: [[this document]]

Value: 0
Array: 10, -16, 8, 4, -8, 10, -16
Description: AES-CCM-16-64-128, SHA-256, 8, X25519, EdDSA,
      AES-CCM-16-64-128, SHA-256
Reference: [[this document]]

Value: 1
Array: 30, -16, 16, 4, -8, 10, -16
Description: AES-CCM-16-128-128, SHA-256, 16, X25519, EdDSA,
      AES-CCM-16-64-128, SHA-256
Reference: [[this document]]

Value: 2
Array: 10, -16, 8, 1, -7, 10, -16
Description: AES-CCM-16-64-128, SHA-256, 8, P-256, ES256,
      AES-CCM-16-64-128, SHA-256
Reference: [[this document]]

Value: 3
Array: 30, -16, 16, 1, -7, 10, -16
Description: AES-CCM-16-128-128, SHA-256, 16, P-256, ES256,
      AES-CCM-16-64-128, SHA-256
Reference: [[this document]]

Value: 4
Array: 24, -16, 16, 4, -8, 24, -16
Description: ChaCha20/Poly1305, SHA-256, 16, X25519, EdDSA,
      ChaCha20/Poly1305, SHA-256
Reference: [[this document]]

Value: 5
Array: 24, -16, 16, 1, -7, 24, -16
Description: ChaCha20/Poly1305, SHA-256, 16, P-256, ES256,
      ChaCha20/Poly1305, SHA-256
Reference: [[this document]]

Value: 6
Array: 1, -16, 16, 4, -7, 1, -16
Description: A128GCM, SHA-256, 16, X25519, ES256,
      A128GCM, SHA-256
Reference: [[this document]]

Value: 23
Reserved
Reference: [[this document]]

Value: 24
Array: 3, -43, 16, 2, -35, 3, -43
Description: A256GCM, SHA-384, 16, P-384, ES384,
      A256GCM, SHA-384
Reference: [[this document]]

Value: 25
Array: 24, -45, 16, 5, -8, 24, -45
Description: ChaCha20/Poly1305, SHAKE256, 16, X448, EdDSA,
      ChaCha20/Poly1305, SHAKE256
Reference: [[this document]]

10.3. EDHOC Method Type Registry

IANA is requested to create a new registry under the new registry group "Ephemeral Diffie-Hellman Over COSE (EDHOC)" as follows:¶

Registry Name: EDHOC Method Type¶

Reference: [[this document]]¶

The columns of the registry are Value, Initiator Authentication Key, and Responder Authentication Key, and Reference, where Value is an integer and the key columns are text strings describing the authentication keys.¶

The initial contents of the registry are shown in Figure 4. Method 23 is Reserved.¶

10.4. EDHOC Error Codes Registry

IANA is requested to create a new registry under the new registry group "Ephemeral Diffie-Hellman Over COSE (EDHOC)" as follows:¶

Registry Name: EDHOC Error Codes¶

Reference: [[this document]]¶

The columns of the registry are ERR_CODE, ERR_INFO Type, Description, and Reference, where ERR_CODE is an integer, ERR_INFO is a CDDL defined type, and Description is a text string. The initial contents of the registry are shown in Figure 10. Error code 23 is Reserved.¶

10.5. EDHOC External Authorization Data Registry

IANA is requested to create a new registry under the new registry group "Ephemeral Diffie-Hellman Over COSE (EDHOC)" as follows:¶

Registry Name: EDHOC External Authorization Data¶

Reference: [[this document]]¶

The columns of the registry are Name, Label, Description, and Reference, where Label is a non-negative integer and the other columns are text strings. The initial contents of the registry is shown in Figure 14. EAD label 23 is Reserved.¶

10.6. COSE Header Parameters Registry

IANA is requested to register the following entries in the "COSE Header Parameters" registry under the registry group "CBOR Object Signing and Encryption (COSE)" (see Figure 15): The value of the 'kcwt' header parameter is a COSE Web Token (CWT) [RFC8392], and the value of the 'kccs' header parameter is a CWT Claims Set (CCS), see Section 1.4. The CWT/CCS must contain a COSE_Key in a 'cnf' claim [RFC8747]. The Value Registry for this item is empty and omitted from the table below.¶

10.7. The Well-Known URI Registry

IANA is requested to add the well-known URI "edhoc" to the "Well-Known URIs" registry.¶

• URI suffix: edhoc¶
• Change controller: IETF¶
• Specification document(s): [[this document]]¶
• Related information: None¶

10.8. Media Types Registry

IANA is requested to add the media types "application/edhoc+cbor-seq" and "application/cid-edhoc+cbor-seq" to the "Media Types" registry.¶

10.9. CoAP Content-Formats Registry

IANA is requested to add the media types "application/edhoc+cbor-seq" and "application/cid-edhoc+cbor-seq" to the "CoAP Content-Formats" registry under the registry group "Constrained RESTful Environments (CoRE) Parameters".¶

10.10. Resource Type (rt=) Link Target Attribute Values Registry

IANA is requested to add the resource type "core.edhoc" to the "Resource Type (rt=) Link Target Attribute Values" registry under the registry group "Constrained RESTful Environments (CoRE) Parameters".¶

• Value: "core.edhoc"¶
• Description: EDHOC resource.¶
• Reference: [[this document]]¶

10.11. Expert Review Instructions

The IANA Registries established in this document are defined as "Expert Review", "Specification Required" or "Standards Action with Expert Review". This section gives some general guidelines for what the experts should be looking for, but they are being designated as experts for a reason so they should be given substantial latitude.¶

Expert reviewers should take into consideration the following points:¶

• Clarity and correctness of registrations. Experts are expected to check the clarity of purpose and use of the requested entries. Expert needs to make sure the values of algorithms are taken from the right registry, when that is required. Experts should consider requesting an opinion on the correctness of registered parameters from relevant IETF working groups. Encodings that do not meet these objective of clarity and completeness should not be registered.¶
• Experts should take into account the expected usage of fields when approving code point assignment. The length of the encoded value should be weighed against how many code points of that length are left, the size of device it will be used on, and the number of code points left that encode to that size.¶
• Even for "Expert Review" specifications are recommended. When specifications are not provided for a request where Expert Review is the assignment policy, the description provided needs to have sufficient information to verify the code points above.¶

A.1. Deriving the OSCORE Security Context

This section specifies how to use EDHOC output to derive the OSCORE security context.¶

After successful processing of EDHOC message_3, Client and Server derive Security Context parameters for OSCORE as follows (see Section 3.2 of [RFC8613]):¶

• The Master Secret and Master Salt SHALL be derived by using the EDHOC_Exporter interface, see Section 4.2.1:¶

  • The EDHOC Exporter Labels for deriving the OSCORE Master Secret and the OSCORE Master Salt, are the uints 0 and 1, respectively.¶
  • The context parameter is h'' (0x40), the empty CBOR byte string.¶
  • By default, oscore_key_length is the key length (in bytes) of the application AEAD Algorithm of the selected cipher suite for the EDHOC session. Also by default, oscore_salt_length has value 8. The Initiator and Responder MAY agree out-of-band on a longer oscore_key_length than the default, and on shorter or longer than the default oscore_salt_length.¶

   Master Secret = EDHOC_Exporter( 0, h'', oscore_key_length )
   Master Salt   = EDHOC_Exporter( 1, h'', oscore_salt_length )

• The AEAD Algorithm SHALL be the application AEAD algorithm of the selected cipher suite for the EDHOC session.¶
• The HKDF Algorithm SHALL be the one based on the application hash algorithm of the selected cipher suite for the EDHOC session. For example, if SHA-256 is the application hash algorithm of the selected cipher suite, HKDF SHA-256 is used as HKDF Algorithm in the OSCORE Security Context.¶
• The relationship between identifiers in OSCORE and EDHOC is specified in Section 3.3.3. The OSCORE Sender ID and Recipient ID SHALL be determined by the EDHOC connection identifiers C_R and C_I for the EDHOC session as shown in Figure 17.¶

Client and Server SHALL use the parameters above to establish an OSCORE Security Context, as per Section 3.2.1 of [RFC8613].¶

From then on, Client and Server retrieve the OSCORE protocol state using the Recipient ID, and optionally other transport information such as the 5-tuple.¶

A.2. Transferring EDHOC over CoAP

This section specifies how EDHOC can be transferred as an exchange of CoAP [RFC7252] messages. CoAP provides a reliable transport that can preserve packet ordering, flow control, and handle message duplication. CoAP can also perform fragmentation and protect against denial-of-service attacks. The underlying CoAP transport should be used in reliable mode, in particular when fragmentation is used, to avoid, e.g., situations with hanging endpoints waiting for each other.¶

EDHOC may run with the Initiator either being CoAP client or CoAP server. We denote the former by the "forward message flow" (see Appendix A.2.1) and the latter by the "reverse message flow" (see Appendix A.2.2). By default we assume the forward message flow, but the roles SHOULD be chosen to protect the most sensitive identity, see Section 9.¶

According to this specification, EDHOC is transferred in POST requests to the Uri-Path: "/.well-known/edhoc" (see Section 10.7), and 2.04 (Changed) responses. An application may define its own path that can be discovered, e.g., using a resource directory [RFC9176]. Client applications can use the resource type "core.edhoc" to discover a server's EDHOC resource, i.e., where to send a request for executing the EDHOC protocol, see Section 10.10. An alternative transfer of the forward message flow is specified in [I-D.ietf-core-oscore-edhoc].¶

In order for the server to correlate a message received from a client to a message previously sent in the same EDHOC session over CoAP, messages sent by the client SHALL be prepended with the CBOR serialization of the connection identifier which the server has selected, see Section 3.4.1. This applies both to the forward and the reverse message flows. To indicate a new EDHOC session in the forward message flow, message_1 SHALL be prepended with the CBOR simple value true (0xf5), as a dummy identifier. Even if CoAP is carried over a reliable transport protocol such as TCP, the prepending of identifiers specified here SHALL be practiced to enable interoperability independent of how CoAP is transported.¶

The prepended identifiers are encoded in CBOR and thus self-delimiting. The representation of identifiers described in Section 3.3.2 SHALL be used. They are sent in front of the actual EDHOC message to keep track of messages in an EDHOC session, and only the part of the body following the identifier is used for EDHOC processing. In particular, the connection identifiers within the EDHOC messages are not impacted by the prepended identifiers.¶

An EDHOC message has media type application/edhoc+cbor-seq, whereas an EDHOC message prepended by a connection identifier has media type application/cid-edhoc+cbor-seq, see Section 10.9.¶

To protect against denial-of-service attacks, the CoAP server MAY respond to the first POST request with a 4.01 (Unauthorized) containing an Echo option [RFC9175]. This forces the Initiator to demonstrate reachability at its apparent network address. If message fragmentation is needed, the EDHOC messages may be fragmented using the CoAP Block-Wise Transfer mechanism [RFC7959].¶

EDHOC error messages need to be transported in response to a message that failed (see Section 6). EDHOC error messages transported with CoAP are carried in the payload.¶

Note that the transport over CoAP can serve as a blueprint for other client-server protocols:¶

• The client prepends the connection identifier selected by the server (or, for message_1, the CBOR simple value true) to any request message it sends.¶
• The server does not send any such indicator, as responses are matched to request by the client-server protocol design.¶

C.1. CBOR and CDDL

The Concise Binary Object Representation (CBOR) [RFC8949] is a data format designed for small code size and small message size. CBOR builds on the JSON data model but extends it by e.g., encoding binary data directly without base64 conversion. In addition to the binary CBOR encoding, CBOR also has a diagnostic notation that is readable and editable by humans. The Concise Data Definition Language (CDDL) [RFC8610] provides a way to express structures for protocol messages and APIs that use CBOR. [RFC8610] also extends the diagnostic notation.¶

CBOR data items are encoded to or decoded from byte strings using a type-length-value encoding scheme, where the three highest order bits of the initial byte contain information about the major type. CBOR supports several different types of data items, in addition to integers (int, uint), simple values, byte strings (bstr), and text strings (tstr), CBOR also supports arrays [] of data items, maps {} of pairs of data items, and sequences [RFC8742] of data items. Some examples are given below.¶

The EDHOC specification sometimes use CDDL names in CBOR diagnostic notation as in e.g., << ID_CRED_R, ? EAD_2 >>. This means that EAD_2 is optional and that ID_CRED_R and EAD_2 should be substituted with their values before evaluation. I.e., if ID_CRED_R = { 4 : h'' } and EAD_2 is omitted then << ID_CRED_R, ? EAD_2 >> = << { 4 : h'' } >>, which encodes to 0x43a10440. We also make use of the occurrence symbol "*", like in e.g., 2* int, meaning two or more CBOR integers.¶

For a complete specification and more examples, see [RFC8949] and [RFC8610]. We recommend implementors get used to CBOR by using the CBOR playground [CborMe].¶

C.2. CDDL Definitions

This sections compiles the CDDL definitions for ease of reference.¶

suites = [ 2* int ] / int

ead = (
  ead_label : int,
  ? ead_value : bstr,
)

EAD_1 = 1* ead
EAD_2 = 1* ead
EAD_3 = 1* ead
EAD_4 = 1* ead

message_1 = (
  METHOD : int,
  SUITES_I : suites,
  G_X : bstr,
  C_I : bstr / -24..23,
  ? EAD_1,
)

message_2 = (
  G_Y_CIPHERTEXT_2 : bstr,
)

message_3 = (
  CIPHERTEXT_3 : bstr,
)

message_4 = (
  CIPHERTEXT_4 : bstr,
)

error = (
  ERR_CODE : int,
  ERR_INFO : any,
)

info = (
  info_label : int,
  context : bstr,
  length : uint,
)

C.3. COSE

CBOR Object Signing and Encryption (COSE) [RFC9052] describes how to create and process signatures, message authentication codes, and encryption using CBOR. COSE builds on JOSE, but is adapted to allow more efficient processing in constrained devices. EDHOC makes use of COSE_Key, COSE_Encrypt0, and COSE_Sign1 objects in the message processing:¶

• ECDH ephemeral public keys of type EC2 or OKP in message_1 and message_2 consist of the COSE_Key parameter named 'x', see Section 7.1 and 7.2 of [RFC9053]¶

• The ciphertexts in message_3 and message_4 consist of a subset of the single recipient encrypted data object COSE_Encrypt0, which is described in Sections 5.2-5.3 of [RFC9052]. The ciphertext is computed over the plaintext and associated data, using an encryption key and an initialization vector. The associated data is an Enc_structure consisting of protected headers and externally supplied data (external_aad). COSE constructs the input to the AEAD [RFC5116] for message_i (i = 3 or 4, see Section 5.4 and Section 5.5, respectively) as follows:¶

  • Secret key K = K_i¶
  • Nonce N = IV_i¶
  • Plaintext P for message_i¶
  • Associated Data A = [ "Encrypt0", h'', TH_i ]¶

• Signatures in message_2 of method 0 and 2, and in message_3 of method 0 and 1, consist of a subset of the single signer data object COSE_Sign1, which is described in Sections 4.2-4.4 of [RFC9052]. The signature is computed over a Sig_structure containing payload, protected headers and externally supplied data (external_aad) using a private signature key and verified using the corresponding public signature key. For COSE_Sign1, the message to be signed is:¶

   [ "Signature1", protected, external_aad, payload ]
  

  ¶

  where protected, external_aad and payload are specified in Section 5.3 and Section 5.4.¶

Different header parameters to identify X.509 or C509 certificates by reference are defined in [RFC9360] and [I-D.ietf-cose-cbor-encoded-cert]:¶

• by a hash value with the 'x5t' or 'c5t' parameters, respectively:¶

  • ID_CRED_x = { 34 : COSE_CertHash }, for x = I or R,¶
  • ID_CRED_x = { TBD3 : COSE_CertHash }, for x = I or R;¶

• or by a URI with the 'x5u' or 'c5u' parameters, respectively:¶

  • ID_CRED_x = { 35 : uri }, for x = I or R,¶
  • ID_CRED_x = { TBD4 : uri }, for x = I or R.¶

When ID_CRED_x does not contain the actual credential, it may be very short, e.g., if the endpoints have agreed to use a key identifier parameter 'kid':¶

• ID_CRED_x = { 4 : kid_x }, where kid_x : kid, for x = I or R. For further optimization, see Section 3.5.3.¶

Note that ID_CRED_x can contain several header parameters, for example { x5u, x5t } or { kid, kid_context }.¶

ID_CRED_x MAY also identify the credential by value. For example, a certificate chain can be transported in an ID_CRED field with COSE header parameter c5c or x5chain, defined in [I-D.ietf-cose-cbor-encoded-cert] and [RFC9360] and credentials of type CWT and CCS can be transported with the COSE header parameters registered in Section 10.6.¶

D.1. Validating the Authentication Credential

The authentication credential may contain, in addition to the authentication key, other parameters that needs to be verified. For example:¶

• In X.509 and C509 certificates, signature keys typically have key usage "digitalSignature" and Diffie-Hellman public keys typically have key usage "keyAgreement" [RFC3279][RFC8410].¶
• In X.509 and C509 certificates validity is expressed using Not After and Not Before. In CWT and CCS, the "exp" and "nbf" claims have similar meanings.¶

D.2. Identities

The application must decide on allowing a connection or not depending on the intended endpoint, and in particular whether it is a specific identity or in a set of identities. To prevent misbinding attacks, the identity of the endpoint is included in a MAC verified through the protocol. More details and examples are provided in this section.¶

Policies for what connections to allow are typically set based on the identity of the other endpoint, and endpoints typically only allow connections from a specific identity or a small restricted set of identities. For example, in the case of a device connecting to a network, the network may only allow connections from devices which authenticate with certificates having a particular range of serial numbers and signed by a particular CA. Conversely, a device may only be allowed to connect to a network which authenticates with a particular public key.¶

• When a Public Key Infrastructure (PKI) is used with certificates, the identity is the subject whose unique name, e.g., a domain name, a Network Access Identifier (NAI), or an Extended Unique Identifier (EUI), is included in the endpoint's certificate.¶
• Similarly, when a PKI is used with CWTs, the identity is the subject identified by the relevant claim(s), such as 'sub' (subject).¶
• When PKI is not used (e.g., CCS, self-signed certificate/CWT) the identity is typically directly associated with the authentication key of the other party. For example, if identities can be expressed in the form of unique subject names assigned to public keys, then a binding to identity is achieved by including both public key and associated subject name in the authentication credential: CRED_I or CRED_R may be a self-signed certificate/CWT or CCS containing the authentication key and the subject name, see Section 3.5.2. Each endpoint thus needs to know the specific authentication key/unique associated subject name, or set of public authentication keys/unique associated subject names, which it is allowed to communicate with.¶

To prevent misbinding attacks in systems where an attacker can register public keys without proving knowledge of the private key, SIGMA [SIGMA] enforces a MAC to be calculated over the "identity". EDHOC follows SIGMA by calculating a MAC over the whole authentication credential, which in case of an X.509 or C509 certificate includes the "subject" and "subjectAltName" fields, and in the case of CWT or CCS includes the "sub" claim.¶

(While the SIGMA paper only focuses on the identity, the same principle is true for other information such as policies associated with the public key.)¶

D.3. Certification Path and Trust Anchors

When a Public Key Infrastructure (PKI) is used with certificates, the trust anchor is a Certification Authority (CA) certificate. Each party needs at least one CA public key certificate, or just the CA public key. The certification path contains proof that the subject of the certificate owns the public key in the certificate. Only validated public-key certificates are to be accepted.¶

Similarly, when a PKI is used with CWTs, each party needs to have at least one trusted third party public key as trust anchor to verify the end entity CWTs. The trusted third party public key can, e.g., be stored in a self-signed CWT or in a CCS.¶

The signature of the authentication credential needs to be verified with the public key of the issuer. X.509 and C509 certificates includes the “Issuer” field. In CWT and CCS, the “iss” claim has a similar meaning. The public key is either a trust anchor or the public key in another valid and trusted credential in a certification path from trust anchor to authentication credential.¶

Similar verifications as made with the authentication credential (see Appendix D.1) are also needed for the other credentials in the certification path.¶

When PKI is not used (CCS, self-signed certificate/CWT), the trust anchor is the authentication key of the other party, in which case there is no certification path.¶

D.4. Revocation Status

The application may need to verify that the credentials are not revoked, see Section 9.8. Some use cases may be served by short-lived credentials, for example, where the validity of the credential is on par with the interval between revocation checks. But, in general, credential lifetime and revocation checking are complementary measures to control credential status. Revocation information may be transported as External Authentication Data (EAD), see Appendix E.¶

D.5. Unauthenticated Operation

EDHOC might be used without authentication by allowing the Initiator or Responder to communicate with any identity except its own. Note that EDHOC without mutual authentication is vulnerable to man-in-the-middle attacks and therefore unsafe for general use. However, it is possible to later establish a trust relationship with an unknown or not-yet-trusted endpoint. Some examples:¶

• The EDHOC authentication credential can be verified out-of-band at a later stage.¶
• The EDHOC session key can be bound to an identity out-of-band at a later stage.¶
• Trust on first use (TOFU) can be used to verify that several EDHOC connections are made to the same identity. TOFU combined with proximity is a common IoT deployment model which provides good security if done correctly. Note that secure proximity based on short range wireless technology requires very low signal strength or very low latency.¶

I.1. Initiator State Machine

The Initiator sends message_1, triggering the state machine to transition from START to WAIT_M2, and waits for message_2.¶

If the incoming message is an error message then the Initiator transitions from WAIT_M2 to ABORTED. In case of error code 2 (Wrong Selected Cipher Suite), the Initiator remembers the supported cipher suites for this particular Responder and transitions from ABORTED to START. The message_1 that the Initiator subsequently sends takes into account the cipher suites supported by the Responder.¶

Upon receiving a non-error message, the Initiator transitions from WAIT_M2 to RCVD_M2 and processes the message. If a processing error occurs on message_2, then the Initiator transitions from RCVD_M2 to ABORTED. In case of successful processing of message_2, the Initiator transitions from RCVD_M2 to VRFD_M2.¶

The Initiator prepares and processes message_3 for sending. If any processing error is encountered, the Initiator transitions from VRFD_M2 to ABORTED. If message_3 is successfully sent, the Initiator transitions from VRFD_M2 to COMPLETED.¶

If the application profile includes message_4, then the Initiator waits for message_4. If the incoming message is an error message then the Initiator transitions from COMPLETED to ABORTED. Upon receiving a non-error message, the Initiator transitions from COMPLETED (="WAIT_M4") to RCVD_M4 and processes the message. If a processing error occurs on message_4, then the Initiator transitions from RCVD_M4 to ABORTED. In case of successful processing of message_4, the Initiator transitions from RCVD_M4 to PERSISTED (="VRFD_M4").¶

If the application profile does not include message_4, then the Initiator waits for an incoming application message. If the decryption and verification of the application message is successful, then the the Initiator transitions from COMPLETED to PERSISTED.¶

I.2. Responder State Machine

Upon receiving message_1, the Responder transitions from START to RCVD_M1.¶

If a processing error occurs on message_1, the Responder transitions from RCVD_M1 to ABORTED. This includes sending error message with error code 2 (Wrong Selected Cipher Suite) if the selected cipher suite in message_1 is not supported. In case of successful processing of message_1, the Responder transitions from RCVD_M1 to VRFD_M1.¶

The Responder prepares and processes message_2 for sending. If any processing error is encountered, the Responder transitions from VRFD_M1 to ABORTED. If message_2 is successfully sent, the Initiator transitions from VRFD_M2 to WAIT_M3, and waits for message_3.¶

If the incoming message is an error message then the Responder transitions from WAIT_M3 to ABORTED.¶

Upon receiving message_3, the Responder transitions from WAIT_M3 to RCVD_M3. If a processing error occurs on message_3, the Responder transitions from RCVD_M3 to ABORTED. In case of successful processing of message_3, the Responder transitions from RCVD_M3 to COMPLETED (="VRFD_M3").¶

If the application profile includes message_4, the Responder prepares and processes message_4 for sending. If any processing error is encountered, the Responder transitions from COMPLETED to ABORTED.¶

If message_4 is successfully sent, or if the application profile does not include message_4, the Responder transitions from COMPLETED to PERSISTED.¶