RFC 9528 | EDHOC | March 2024 |
Selander, et al. | Standards Track | [Page] |
This document specifies Ephemeral Diffie-Hellman Over COSE (EDHOC), a very compact and lightweight authenticated Diffie-Hellman key exchange with ephemeral keys. EDHOC provides mutual authentication, forward secrecy, and identity protection. EDHOC is intended for usage in constrained scenarios, and a main use case is to establish an Object Security for Constrained RESTful Environments (OSCORE) security context. By reusing CBOR Object Signing and Encryption (COSE) for cryptography, Concise Binary Object Representation (CBOR) for encoding, and Constrained Application Protocol (CoAP) for transport, the additional code size can be kept very low.¶
This is an Internet Standards Track document.¶
This document is a product of the Internet Engineering Task Force (IETF). It represents the consensus of the IETF community. It has received public review and has been approved for publication by the Internet Engineering Steering Group (IESG). Further information on Internet Standards is available in Section 2 of RFC 7841.¶
Information about the current status of this document, any errata, and how to provide feedback on it may be obtained at https://www.rfc-editor.org/info/rfc9528.¶
Copyright (c) 2024 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
Many Internet of Things (IoT) deployments require technologies that 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 CoAP [RFC7252], CBOR [RFC8949], and Static Context Header Compression (SCHC) [RFC8724].¶
The need for special protocols targeting constrained IoT deployments extends also to the security domain [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, 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 updates. 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. COSE [RFC9052] specifies basic application-layer security services efficiently encoded in CBOR. Another example is 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 (RPKs) 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 (DH) 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 SIGn-and-MAc [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, IPv6 over the TSCH mode of IEEE 802.15.4e (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, and 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 that are not revealed by a later compromise of the long-term keys.¶
Examples of EDHOC message sizes are shown in Table 1, which use different kinds of authentication keys and COSE header parameters for identification, including static Diffie-Hellman keys or signature keys, either in CWT/CCS [RFC8392] identified by a key identifier using 'kid' [RFC9052] or in X.509 certificates identified by a hash value using 'x5t' [RFC9360]. EDHOC always uses ephemeral-ephemeral key exchange. As a comparison, in the case of RPK authentication and when transferred in CoAP, the EDHOC message size can be less than 1/7 of the DTLS 1.3 handshake [RFC9147] with Ephemeral Elliptic Curve Diffie-Hellman (ECDHE) and connection ID; see [CoAP-SEC-PROT].¶
Static DH Keys | Signature Keys | |||
---|---|---|---|---|
kid | x5t | kid | x5t | |
message_1 | 37 | 37 | 37 | 37 |
message_2 | 45 | 58 | 102 | 115 |
message_3 | 19 | 33 | 77 | 90 |
Total | 101 | 128 | 216 | 242 |
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; Section 7 describes EDHOC support for transport that does not handle message duplication; and Section 8 lists compliance requirements. Note that normative text is also used in appendices, in particular Appendix A.¶
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 CCS [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].¶
EDHOC supports different authentication methods of the ephemeral-ephemeral Diffie-Hellman key exchange. This document specifies authentication methods based on signature keys and static Diffie-Hellman keys. This section outlines the signature-key-based method. Further details of protocol elements and other authentication methods are provided in the remainder of this document.¶
SIGn-and-MAc (SIGMA) is a family of theoretical protocols with a number of variants [SIGMA]. Like in Internet Key Exchange Protocol Version 2 (IKEv2) [RFC7296] and (D)TLS 1.3 [RFC8446] [RFC9147], EDHOC authenticated with signature keys is built on a variant of the SIGMA protocol, SIGMA-I, which provides identity protection against active attacks on the party initiating the protocol. Also like IKEv2, EDHOC implements the MAC-then-Sign variant of the SIGMA-I protocol. The message flow (excluding an optional fourth message) is shown in Figure 1.¶
The parties exchanging messages in an EDHOC session are called the Initiator (I) and the Responder (R), where the Initiator sends message_1 (see Section 3). They exchange ephemeral public keys, compute a shared secret session key PRK_out, and derive symmetric application keys used to protect application data.¶
In order to create a "full-fledged" protocol, some additional protocol elements are needed. This specification adds:¶
EDHOC is designed to encrypt and integrity protect as much information as possible. Symmetric keys and random material used in EDHOC are derived using EDHOC_KDF with as much previous information as possible; see Figure 6. EDHOC is furthermore designed to be as compact and lightweight as possible, in terms of message sizes, processing, and the ability to reuse already existing CBOR, COSE, and CoAP libraries. Like in (D)TLS, authentication is the responsibility of the application. EDHOC identifies (and optionally transports) authentication credentials and provides proof-of-possession of the private authentication key.¶
To simplify for implementors, the use of CBOR, CDDL, and COSE in EDHOC is summarized in Appendix C. Test vectors, including CBOR diagnostic notation, are provided in [RFC9529].¶
The EDHOC protocol consists of three mandatory messages (message_1, message_2, and 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 that should be considered when the roles are assigned; see Section 9.1. All EDHOC messages are CBOR Sequences [RFC8742] and are defined to be deterministically encoded CBOR as specified in Section 4.2.1 of [RFC8949]. Figure 2 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 Sections 3 and 5. Message formatting and processing are specified in Sections 5 and 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.¶
The data item METHOD in message_1 (see Section 5.2.1) is an integer specifying the authentication method. EDHOC currently supports authentication with signature or static Diffie-Hellman keys, as defined in the four authentication methods: 0, 1, 2, and 3; see Table 2. 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 that 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 Responder need to have agreed on a single method to be used for EDHOC; see Section 3.9.¶
Method Type Value | Initiator Authentication Key | Responder Authentication Key |
---|---|---|
0 | Signature Key | Signature Key |
1 | Signature Key | Static DH Key |
2 | Static DH Key | Signature Key |
3 | Static DH Key | Static DH Key |
23 | Reserved | Reserved |
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.¶
EDHOC includes the selection of connection identifiers (C_I and 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.¶
C_I and C_R are chosen by I and R, respectively. The Initiator selects C_I and sends it in message_1 for the Responder to use as a reference to the connection in communications with the Initiator. The Responder selects C_R and sends it in message_2 for the Initiator to use as a reference to the connection in communications with the Responder.¶
If connection identifiers are used by an application protocol for which EDHOC establishes keys, then the selected connection identifiers SHALL adhere to the requirements for that protocol; see Section 3.3.3 for an example.¶
To allow identifiers with minimal overhead on the wire, certain byte strings used in connection identifiers and credential identifiers (see Section 3.5.3) are defined to have integer representations.¶
The integers with one-byte CBOR encoding are -24, ..., 23; see Figure 3.¶
The byte strings that coincide with a one-byte CBOR encoding of an integer MUST be represented by the CBOR encoding of that integer. Other byte strings are simply encoded as CBOR byte strings.¶
For example:¶
One may view this representation of byte strings as a transport encoding, i.e., a byte string that parses as the one-byte CBOR encoding of an integer (i.e., integer in the interval -24, ..., 23) is just copied directly into the message, and a byte string that does not is encoded as a CBOR byte string during transport.¶
For OSCORE, the choice of connection identifier results in the endpoint selecting its Recipient ID (see Section 3.1 of [RFC8613]) for which certain uniqueness requirements apply (see Section 3.3 of [RFC8613]). Therefore, the Initiator and Responder MUST NOT select connection identifiers such that it results in the same OSCORE Recipient ID. Since the connection identifier is a byte string, it is converted to an OSCORE Recipient ID equal to the byte string.¶
Examples:¶
Cryptographically, EDHOC does not put requirements on the underlying layers. Received messages are processed as the expected next message according to the protocol state; see Section 5. If processing fails for any reason, then typically an error message is attempted to be sent and the EDHOC session is aborted.¶
EDHOC is not bound to a particular transport layer and can even be used in environments without IP. Ultimately, the application is free to choose how to transport EDHOC messages including errors. In order to avoid unnecessary message processing or protocol termination, it is RECOMMENDED to use reliable transport, such as CoAP in reliable mode, which is the default transport; see Appendix A.2. In general, the transport SHOULD handle:¶
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. The transport does not require additional means to handle message reordering because of the lockstep processing of EDHOC.¶
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 Table 1. There is no maximum message size specified by the protocol; for example, this is dependent on the size of the authentication credentials (if they are transported, see Section 3.5). The encryption of very large content in message_2 when using certain hash algorithms is described in Appendix G.¶
The use of transport is specified in the application profile, which in particular, may specify limitations in message sizes; see Section 3.9.¶
Correlation between EDHOC messages is needed to facilitate the retrieval of the protocol state and security context during an EDHOC session. It is also helpful for the Responder to get an indication that a received EDHOC message is the beginning of a new EDHOC session, such that no existing protocol state or security context needs to be retrieved.¶
Correlation may be based on existing mechanisms in the transport protocol; for example, the CoAP Token may be used to correlate EDHOC messages in a CoAP response and in an associated CoAP request. The connection identifiers may also be used to correlate EDHOC messages.¶
If correlation between consecutive messages is not provided by other means, then the transport binding SHOULD mandate prepending of an appropriate connection identifier (when available from the EDHOC protocol) to the EDHOC message. If message_1 indication is not provided by other means, then the transport binding SHOULD mandate prepending of message_1 with the CBOR simple value true
(0xf5).¶
Transport of EDHOC in CoAP payloads is described in Appendix A.2, including how to use connection identifiers and message_1 indication with CoAP. A similar construction is possible for other client-server protocols. Protocols that do not provide any correlation at all can prescribe prepending of the peer's connection identifier to all messages.¶
Note that correlation between EDHOC messages may be obtained without transport support or connection identifiers, for example, if the endpoints only accept a single instance of the protocol at a time and execute conditionally on a correct sequence of messages.¶
EDHOC supports various settings for how the other endpoint's public key for authentication may be transported, identified, and trusted. We shall use the term "authentication key" to mean key used for authentication in general, or specifically, the public key, when there is no risk for confusion.¶
EDHOC performs the following authentication-related operations:¶
EDHOC uses the authentication credentials in two ways (see Sections 5.3.2 and 5.4.2):¶
Other authentication-related verifications are out of scope for EDHOC and are 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 credentials in ID_CRED_R and ID_CRED_I. EDHOC MUST have access to the authentication key and the authentication credential.¶
Note that the type of authentication key, the type of 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.¶
The authentication key MUST be a signature key or a static Diffie-Hellman key. The Initiator and Responder MAY use different types of authentication keys, e.g., one uses a signature key and the other uses a static Diffie-Hellman key.¶
The authentication key algorithm needs to be compatible with the method and the selected cipher suite (see Section 3.6). The authentication key algorithm needs to be compatible with the EDHOC key exchange algorithm when static Diffie-Hellman authentication is used and compatible with the EDHOC signature algorithm when signature authentication is used.¶
Note that for most signature algorithms, the signature is determined jointly by the signature algorithm and the authentication key algorithm. When using static Diffie-Hellman keys, the Initiator's and the Responder's private authentication keys are denoted as I and R, respectively, and the public authentication keys are denoted G_I and G_R, respectively.¶
For X.509 certificates, the authentication key is represented by a SubjectPublicKeyInfo field, which also contains information about authentication key algorithm. For CWT and CCS (see Section 3.5.2), the authentication key is represented by a 'cnf' claim [RFC8747] containing a COSE_Key [RFC9052], which contains information about authentication key algorithm. In EDHOC, a raw public key (RPK) is an authentication key encoded as a COSE_Key wrapped in a CCS, an example is given in Figure 4.¶
The authentication credentials, CRED_I and CRED_R, contain the public authentication key of the Initiator and Responder, respectively. We use the notation CRED_x to refer to CRED_I or CRED_R. Requirements on CRED_x applies both to CRED_I and to CRED_R. The authentication credential typically also contains other parameters that needs to be verified by the application (see Appendix D) and in particular information about the identity ("subject") of the endpoint to prevent misbinding attacks (see Appendix D.2).¶
EDHOC relies on COSE for identification of credentials (see Section 3.5.3), for example, X.509 certificates [RFC9360], C509 certificates [C509-CERTS], CWTs [RFC8392], and CCSs [RFC8392]. When the identified credential is a chain or a bag, the authentication credential CRED_x is just the end entity X.509 or C509 certificate / CWT. In the choice between a chain or a bag, it is RECOMMENDED to use a chain, since the certificates in a bag are unordered and may contain self-signed and extraneous certificates, which can add complexity to the process of extracting the end entity certificate. The Initiator and Responder MAY use different types of authentication credentials, e.g., one uses an RPK and the other uses a public key certificate.¶
Since CRED_R is used in the integrity verification (see Section 5.3.2), it needs to be specified such that it is identical when used by the Initiator or Responder. Similarly for CRED_I, see Section 5.4.2. The Initiator and Responder are expected to agree on the specific encoding of the authentication credentials; see Section 3.9. It is RECOMMENDED that the COSE 'kid' parameter, when used to identify the authentication credential, refers to such a specific encoding of the authentication credential. The Initiator and Responder SHOULD use an available authentication credential without re-encoding, i.e. an authentication credential transported in EDHOC by value, or otherwise provisioned, SHOULD be used as is. If for some reason re-encoding of an authentication credential passed by reference may occur, then a potential common encoding for CBOR-based credentials is deterministically encoded CBOR, as specified in Sections 4.2.1 and 4.2.2 of [RFC8949].¶
When the authentication credential includes a COSE_Key but is not in a CWT, CRED_x SHALL be an untagged CCS. This is how RPKs are encoded, see Figure 4 for an example.¶
An example of CRED_x is shown below:¶
The ID_CRED fields, ID_CRED_R and ID_CRED_I, are transported in message_2 and message_3, respectively; see Sections 5.3.2 and 5.4.2. We use the notation ID_CRED_x to refer to ID_CRED_I or ID_CRED_R. Requirements on ID_CRED_x applies both to ID_CRED_I and to ID_CRED_R. The ID_CRED fields are used to identify and optionally transport credentials:¶
ID_CRED_x may contain the authentication credential CRED_x, for x = I or R, but for many settings, it is not necessary to transport the authentication credential within EDHOC. For example, it may be pre-provisioned or acquired out-of-band over less constrained links. ID_CRED_I and ID_CRED_R do not have any cryptographic purpose in EDHOC since the authentication credentials are integrity protected by the Signature_or_MAC field.¶
EDHOC relies on COSE for identification of credentials and supports all credential types for which COSE header parameters are defined, including X.509 certificates [RFC9360], C509 certificates [C509-CERTS], CWTs (Section 3.5.3.1) and CCSs (Section 3.5.3.1).¶
ID_CRED_I and ID_CRED_R are of type COSE header_map, as defined in Section 3 of [RFC9052], and contain one or more COSE header parameters. If a map contains several header parameters, the labels do not need to be sorted in bytewise lexicographic order. ID_CRED_I and ID_CRED_R MAY contain different header parameters. The header parameters typically provide some information about the format of the credential.¶
Example: X.509 certificates can be identified by a hash value using the 'x5t' parameter; see Section 2 of [RFC9360]:¶
Example: CWT or CCS can be identified by a key identifier using the 'kid' parameter; see Section 3.1 of [RFC9052]:¶
Note that COSE header parameters in ID_CRED_x are used to identify the message sender's credential. Therefore, there is no reason to use the "-sender" header parameters, such as x5t-sender, defined in Section 3 of [RFC9360]. Instead, the corresponding parameter without "-sender", such as x5t, SHOULD be used.¶
As stated in Section 3.1 of [RFC9052], applications MUST NOT assume that 'kid' values are unique and several keys associated with a 'kid' may need to be checked before the correct one is found. Applications might use additional information such as 'kid context' or lower layers to determine which key to try first. Applications should strive to make ID_CRED_x as unique as possible, since the recipient may otherwise have to try several keys.¶
See Appendix C.3 for more examples.¶
This document registers two new COSE header parameters, 'kcwt' and 'kccs', for use with CBOR Web Token (CWT) [RFC8392] and CWT Claims Set (CCS) [RFC8392], respectively. The CWT/CCS MUST contain a COSE_Key in a 'cnf' claim [RFC8747]. There may be any number of additional claims present in the CWT/CCS.¶
CWTs sent in 'kcwt' are protected using a MAC or a signature and are similar to a certificate (when used with public key cryptography) or a Kerberos ticket (when used with symmetric key cryptography). CCSs sent in 'kccs' are not protected and are therefore similar to raw public keys or self-signed certificates.¶
Security considerations for 'kcwt' and 'kccs' are made in Section 9.8.¶
To comply with the Lightweight Authenticated Key Exchange (LAKE) message size requirements (see [LAKE-REQS]), two optimizations are made for the case when ID_CRED_x, for x = I or R, contains a single 'kid' parameter.¶
These optimizations MUST be applied if and only if ID_CRED_x = { 4 : kid_x } and ID_CRED_x in PLAINTEXT_y of message_y, y = 2 or 3; see Sections 5.3.2 and 5.4.2. Note that these optimizations are not applied to instances of ID_CRED_x that have no impact on message size, e.g., context_y, or the COSE protected header. For example:¶
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 COSE Algorithms [RFC9053]. Note that COSE sometimes uses peculiar names such as ES256 for Elliptic Curve Digital Signature Algorithm (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 Mechanism (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 (methods 1, 2, and 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 jointly by the signature algorithm and the authentication key algorithm; 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:¶
Each cipher suite is identified with a predefined 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 predefined cipher suites. Private cipher suites can be identified with any of the four values: -24, -23, -22, and -21. The predefined cipher suites are listed in the IANA registry (Section 10.2) with the 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 AES-CCM* mode of operation defined in Annex B of [IEEE.802.15.4-2015].¶
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 and 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.¶
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 Octet Key Pair (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.¶
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, and 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. EAD item ead is a CBOR sequence of an ead_label and an optional ead_value; see Figure 5 and Appendix C.2 for the CDDL definitions.¶
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 the sign of the 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 nonnegative 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 a 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.¶
EDHOC message_1 and the plaintext of message_2, message_3, and message_4 can be padded with the use of the corresponding EAD_x field, for x = 1, 2, 3, or 4. Padding in EAD_1 mitigates amplification attacks (see Section 9.7), and padding in EAD_2, EAD_3, and EAD_4 hides the true length of the plaintext (see Section 9.6). Padding MUST be ignored and discarded by the receiving application.¶
Padding is obtained by using an EAD item with ead_label = 0 and a (pseudo)randomly generated byte string of appropriate length as ead_value, noting that the ead_label and the CBOR encoding of ead_value also add bytes. For example:¶
One-byte padding (optional ead_value omitted):¶
EAD_x = 0x00¶
Two-byte padding, using the empty byte string (0x40) as ead_value:¶
EAD_x = 0x0040¶
Three-byte padding, constructed from the pseudorandomly generated ead_value 0xe9 encoded as byte string:¶
EAD_x = 0x0041e9¶
Multiple occurrences of EAD items with ead_label = 0 are allowed. Certain padding lengths require the use of at least two such EAD items.¶
Note that padding is non-critical because the intended behavior when receiving is to ignore it.¶
EDHOC requires certain parameters to be agreed upon between the 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 the Initiator and which is the 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 the following:¶
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.¶
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 Sections 5.2.1 and 6.3, but it may become simplified by this knowledge. EDHOC messages can be processed without the application profile, i.e., the EDHOC messages include 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, the type of the 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 the 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 the lack of compliance with the application profile was the reason for failure. For example, integrity verification in message_2 or message_3 may fail not only because of a 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 the URI to which the EDHOC message is sent, or external authorization data type.¶
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 (PRKs) 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 PRK, but how the key is derived depends on the authentication method (Section 3.2), as detailed in Section 5.¶
The pseudorandom keys (PRKs) used for EDHOC message processing are derived using EDHOC_Extract:¶
PRK = EDHOC_Extract( salt, IKM )¶
where the input keying material (IKM) and salt are defined for each PRK below.¶
The definition of EDHOC_Extract depends on the EDHOC hash algorithm of the selected cipher suite:¶
where the Keccak Message Authentication Code (KMAC) is specified in [SP800-185].¶
The rest of the section defines the pseudorandom keys PRK_2e, PRK_3e2m, and PRK_4e3m; their use is shown in Figure 6. The index of a PRK indicates its use or in what message protection operation it is involved. For example, PRK_3e2m is involved in the encryption of message 3 and in calculating the MAC of message 2.¶
The pseudorandom key PRK_2e is derived with the following input:¶
Example: Assuming the use of curve25519, the ECDH shared secret G_XY is the output of the X25519 function [RFC7748]:¶
G_XY = X25519( Y, G_X ) = X25519( X, G_Y )¶
Example: Assuming the use of SHA-256, the extract phase of the key derivation function is HKDF-Extract, which produces PRK_2e as follows:¶
PRK_2e = HMAC-SHA-256( TH_2, G_XY )¶
The pseudorandom key PRK_3e2m is derived as follows:¶
If the Responder authenticates with a static Diffie-Hellman key, then PRK_3e2m = EDHOC_Extract( SALT_3e2m, G_RX ), where¶
else PRK_3e2m = PRK_2e.¶
The pseudorandom key PRK_4e3m is derived as follows:¶
If the Initiator authenticates with a static Diffie-Hellman key, then PRK_4e3m = EDHOC_Extract( SALT_4e3m, G_IY ), where¶
else PRK_4e3m = PRK_3e2m.¶
The output keying material (OKM) -- including keys, initialization vectors (IVs), and salts -- are derived from the PRKs using the EDHOC_KDF, which is defined through EDHOC_Expand:¶
OKM = EDHOC_KDF( PRK, info_label, context, length ) = EDHOC_Expand( PRK, info, length )¶
where info is encoded as the CBOR sequence:¶
info = ( info_label : int, context : bstr, length : uint, )¶
where:¶
When EDHOC_KDF is used to derive OKM for EDHOC message processing, then the context includes one of the transcript hashes, TH_2, TH_3, or TH_4, defined in Sections 5.3.2 and 5.4.2.¶
The definition of EDHOC_Expand depends on the EDHOC hash algorithm of the selected cipher suite:¶
where L = 8 ⋅ length, the output length in bits.¶
Figure 6 lists derivations made with EDHOC_KDF, where:¶
Further details of the key derivation and how the output keying material is used are specified in Section 5.¶
h'' is CBOR diagnostic notation for the empty byte string, 0x40.¶
The pseudorandom key PRK_out, derived as shown in Figure 6, is the output session key of a completed EDHOC session.¶
Keys for applications are derived using EDHOC_Exporter (see Section 4.2.1) from PRK_exporter, which in turn is derived from PRK_out as shown in Figure 6. For the purpose of generating application keys, it is sufficient to store PRK_out or PRK_exporter. (Note that the word "store" used here does not imply that the application has access to the plaintext PRK_out since that may be reserved for code within a Trusted Execution Environment (TEE); see Section 9.8.)¶
This section defines EDHOC_Exporter in terms of EDHOC_KDF and PRK_exporter. A key update function is defined in Appendix H.¶
Keying material for the application can be derived using the EDHOC_Exporter interface defined as:¶
EDHOC_Exporter(exporter_label, context, length) = EDHOC_KDF(PRK_exporter, exporter_label, context, length)¶
where:¶
The (exporter_label, context) pair used in EDHOC_Exporter must be unique, i.e., an (exporter_label, context) MUST NOT be used for two different purposes. However, an application can re-derive the same key several times as long as it is done securely. For example, in most encryption algorithms, the same key can be reused with different nonces. The context can, for example, be the empty CBOR byte string.¶
Examples of use of the EDHOC_Exporter are given in Appendix A.¶
This section specifies formatting of the messages and processing steps. Error messages are specified in Section 6. Annotated traces of EDHOC sessions are provided in [RFC9529].¶
An EDHOC message is encoded as a sequence of CBOR data items (CBOR Sequence [RFC8742]). Additional optimizations are made to reduce message overhead.¶
While EDHOC uses the COSE_Key, COSE_Sign1, and COSE_Encrypt0 structures, only a subset of the parameters is included in the EDHOC messages; see Appendix C.3. In order to recreate the COSE object, the recipient endpoint may need to add parameters to the COSE headers not included in the EDHOC message, for example, the parameter 'alg' to COSE_Sign1 or COSE_Encrypt0.¶
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) that 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:¶
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 is 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.¶
message_1 SHALL be a CBOR Sequence (see Appendix C.1), as defined below.¶
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¶
where:¶
The processing steps are detailed below and in Section 6.3.¶
The Initiator SHALL compose message_1 as follows:¶
Construct SUITES_I as an array of cipher suites supported by I in order of preference by I with the first cipher suite in the array being the most preferred by I and the last being the one selected by I for this EDHOC session. If the cipher suite most preferred by I is selected, then SUITES_I contains only that cipher suite and is encoded as an int. All cipher suites, if any, preferred by I over the selected one MUST be included. (See also Section 6.3.)¶
The Responder SHALL process message_1 in the following order:¶
If any processing step fails, then the Responder MUST send an EDHOC error message back as defined in Section 6, and the EDHOC session MUST be aborted.¶
message_2 SHALL be a CBOR Sequence (see Appendix C.1), as defined below.¶
message_2 = ( G_Y_CIPHERTEXT_2 : bstr, )¶
where:¶
The Responder SHALL compose message_2 as follows:¶
Compute MAC_2 as in Section 4.1.2 with context_2 = << C_R, ID_CRED_R, TH_2, CRED_R, ? EAD_2 >> (see Appendix C.1 for notation).¶
If the Responder authenticates with a static Diffie-Hellman key (method equals 1 or 3), then Signature_or_MAC_2 is MAC_2. If the Responder authenticates with a signature key (method equals 0 or 2), then Signature_or_MAC_2 is the 'signature' field of a COSE_Sign1 object, computed as specified in Section 4.4 of [RFC9052] using the signature algorithm of the selected cipher suite, the private authentication key of the Responder, and the following parameters as input (see Appendix C.3 for an overview of COSE and Appendix C.1 for notation):¶
CIPHERTEXT_2 is calculated with a binary additive stream cipher, using a keystream generated with EDHOC_Expand and the following plaintext:¶
PLAINTEXT_2 = ( C_R, ID_CRED_R / bstr / -24..23, Signature_or_MAC_2, ? EAD_2 )¶
The Initiator SHALL process message_2 in the following order:¶
If any processing step fails, then the Initiator MUST send an EDHOC error message back as defined in Section 6, and the EDHOC session MUST be aborted.¶
message_3 SHALL be a CBOR Sequence (see Appendix C.1), as defined below.¶
message_3 = ( CIPHERTEXT_3 : bstr, )¶
The Initiator SHALL compose message_3 as follows:¶
Compute MAC_3 as in Section 4.1.2, with context_3 = << ID_CRED_I, TH_3, CRED_I, ? EAD_3 >>¶
If the Initiator authenticates with a static Diffie-Hellman key (method equals 2 or 3), then Signature_or_MAC_3 is MAC_3. If the Initiator authenticates with a signature key (method equals 0 or 1), then Signature_or_MAC_3 is the 'signature' field of a COSE_Sign1 object, computed as specified in Section 4.4 of [RFC9052] using the signature algorithm of the selected cipher suite, the private authentication key of the Initiator, and the following parameters as input (see Appendix C.3):¶
Compute a COSE_Encrypt0 object as defined in Sections 5.2 and 5.3 of [RFC9052], with the EDHOC AEAD algorithm of the selected cipher suite, using the encryption key K_3, the initialization vector IV_3 (if used by the AEAD algorithm), the plaintext PLAINTEXT_3, and the following parameters as input (see Appendix C.3):¶
PLAINTEXT_3 = ( ID_CRED_I / bstr / -24..23, Signature_or_MAC_3, ? EAD_3 )¶
CIPHERTEXT_3 is the 'ciphertext' of COSE_Encrypt0.¶
After creating message_3, the Initiator can compute PRK_out (see Section 4.1.3) and derive application keys using the EDHOC_Exporter interface (see Section 4.2.1). The Initiator SHOULD NOT persistently store PRK_out or application keys until the Initiator has verified message_4 or a message protected with a derived application key, such as an OSCORE message, from the Responder and the application has authenticated the Responder. This is similar to waiting for an acknowledgment (ACK) in a transport protocol. The Initiator SHOULD NOT send protected application data until the application has authenticated the Responder.¶
The Responder SHALL process message_3 in the following order:¶
After processing message_3, the Responder can compute PRK_out (see Section 4.1.3) and derive application keys using the EDHOC_Exporter interface (see Section 4.2.1). The Responder SHOULD NOT persistently store PRK_out or application keys until the application has authenticated the Initiator. The Responder SHOULD NOT send protected application data until the application has authenticated the Initiator.¶
If any processing step fails, then the Responder MUST send an EDHOC error message back as defined in Section 6, and the EDHOC session MUST be aborted.¶
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:¶
Further considerations about when to use message_4 are provided in Sections 3.9 and 9.1.¶
message_4 SHALL be a CBOR Sequence (see Appendix C.1), as defined below.¶
message_4 = ( CIPHERTEXT_4 : bstr, )¶
The Responder SHALL compose message_4 as follows:¶
Compute a COSE_Encrypt0 as defined in Sections 5.2 and 5.3 of [RFC9052], with the EDHOC AEAD algorithm of the selected cipher suite, using the encryption key K_4, the initialization vector IV_4 (if used by the AEAD algorithm), the plaintext PLAINTEXT_4, and the following parameters as input (see Appendix C.3):¶
PLAINTEXT_4 = ( ? EAD_4 )¶
CIPHERTEXT_4 is the 'ciphertext' of COSE_Encrypt0.¶
The Initiator SHALL process message_4 as follows:¶
If any processing step fails, then the Initiator MUST send an EDHOC error message back as defined in Section 6, and the EDHOC session MUST be aborted.¶
After verifying message_4, the Initiator is assured that the Responder has calculated the key PRK_out (key confirmation) and that no other party can derive the key.¶
This section defines the format for error messages and the processing associated with the currently defined error codes. Additional error codes may be registered; see Section 10.4.¶
Many kinds of errors can occur during EDHOC processing. As in CoAP, an error can be triggered by errors in the received message or internal errors in the receiving endpoint. Except for processing and formatting errors, it is up to the application when to send an error message. Sending error messages is essential for debugging but MAY be skipped if, for example, an EDHOC session cannot be found or due to denial-of-service reasons; see Section 9.7. Error messages in EDHOC are always fatal. After sending an error message, the sender MUST abort the EDHOC session. The receiver SHOULD treat an error message as an indication that the other party likely has aborted the EDHOC session. But since error messages might be forged, the receiver MAY try to continue the EDHOC session.¶
An EDHOC error message can be sent by either endpoint as a reply to any non-error EDHOC message. How errors at the EDHOC layer are transported depends on lower layers, which need to enable error messages to be sent and processed as intended.¶
error SHALL be a CBOR Sequence (see Appendix C.1), as defined below.¶
where:¶
The remainder of this section specifies the currently defined error codes; see Table 3. Additional error codes and corresponding error information may be specified.¶
ERR_CODE | ERR_INFO Type | Description |
---|---|---|
0 | Reserved for success | |
1 | tstr | Unspecified error |
2 | suites | Wrong selected cipher suite |
3 | true | Unknown credential referenced |
23 | Reserved |
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.¶
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 that SHOULD be written in English, for example, "Method not supported". The diagnostic text message is mainly intended for software engineers who 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.¶
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.¶
After receiving SUITES_R, the Initiator can determine which cipher suite to select (if any) for the next EDHOC run with the Responder. The Initiator SHOULD remember which selected cipher suite to use until the next message_1 has been sent; otherwise, the Initiator and Responder will run into an infinite loop where the Initiator selects its most preferred cipher suite and the Responder sends an error with supported cipher suites.¶
After a completed EDHOC session, the Initiator MAY remember the selected cipher suite to use in future EDHOC sessions with this Responder. Note that if the Initiator or Responder is updated with new cipher suite policies, any cached information may be outdated.¶
Note that the Initiator's list of supported cipher suites and order of preference is fixed (see Sections 5.2.1 and 5.2.2). Furthermore, the Responder SHALL only accept message_1 if the selected cipher suite is the first cipher suite in SUITES_I that the Responder also supports (see Section 5.2.3). Following this procedure ensures that the selected cipher suite is the most preferred (by the Initiator) cipher suite supported by both parties. For examples, see Section 6.3.2.¶
If the selected cipher suite is not the first cipher suite that the Responder supports in SUITES_I received in message_1, then the Responder MUST abort the EDHOC session; see Section 5.2.3. If SUITES_I in message_1 is manipulated, then the integrity verification of message_2 containing the transcript hash TH_2 will fail and the Initiator will abort the EDHOC session.¶
Assume that the Initiator supports the five cipher suites, 5, 6, 7, 8, and 9, in decreasing order of preference. Figures 8 and 9 show two examples of how the Initiator can format SUITES_I and how SUITES_R is used by Responders to give the Initiator information about the cipher suites that the Responder supports.¶
In Example 1 (Figure 8), the Responder supports cipher suite 6 but not the initially selected cipher suite 5. The Responder rejects the first message_1 with an error indicating support for suite 6 in SUITES_R. The Initiator also supports suite 6 and therefore selects suite 6 in the second message_1. The Initiator prepends in SUITES_I the selected suite 6 with the more preferred suites, in this case suite 5, to mitigate a potential attack on the cipher suite negotiation.¶
In Example 2 (Figure 9), the Responder supports cipher suites 8 and 9 but not the more preferred (by the Initiator) cipher suites 5, 6 or 7. To illustrate the negotiation mechanics, we let the Initiator first make a guess that the Responder supports suite 6 but not suite 5. Since the Responder supports neither 5 nor 6, it rejects the first message_1 with an error indicating support for suites 8 and 9 in SUITES_R (in any order). The Initiator also supports suites 8 and 9, and prefers suite 8, so it selects suite 8 in the second message_1. The Initiator prepends in SUITES_I the selected suite 8 with the more preferred suites in order of preference, in this case, suites 5, 6 and 7, to mitigate a potential attack on the cipher suite negotiation.¶
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 that 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.¶
By default, EDHOC assumes that message duplication is handled by the transport (which is exemplified by CoAP in this section); see Appendix A.2.¶
Deduplication of CoAP messages is described in Section 4.5 of [RFC7252]. This handles the case when the same Confirmable (CON) message is received multiple times due to missing acknowledgment on the CoAP messaging layer. The recommended processing in [RFC7252] is that the duplicate message is acknowledged, but the received message is only processed once by the CoAP stack.¶
Message deduplication is resource demanding and therefore not supported in all CoAP implementations. Since EDHOC is targeting constrained environments, it is desirable that EDHOC can optionally support transport layers that do not handle message duplication. Special care is needed to avoid issues with duplicate messages; see Section 5.1.¶
The guiding principle here is similar to the deduplication processing on the CoAP messaging layer, i.e., a received duplicate EDHOC message SHALL NOT result in another instance of the next EDHOC message. The result MAY be that a duplicate next EDHOC message is sent, provided it is still relevant with respect to the current protocol state. In any case, the received message MUST NOT be processed more than once in the same EDHOC session. This is called "EDHOC message deduplication".¶
An EDHOC implementation MAY store the previously sent EDHOC message to be able to resend it.¶
In principle, if the EDHOC implementation would deterministically regenerate the identical EDHOC message previously sent, it would be possible to instead store the protocol state to be able to recreate and resend the previously sent EDHOC message. However, even if the protocol state is fixed, the message generation may introduce differences that compromise security. For example, in the generation of message_3, if I is performing a (non-deterministic) ECDSA signature (say, method 0 or 1 and cipher suite 2 or 3), then PLAINTEXT_3 is randomized, but K_3 and IV_3 are the same, leading to a key and nonce reuse.¶
The EDHOC implementation MUST NOT store the previous protocol state and regenerate an EDHOC message if there is a risk that the same key and IV are used for two (or more) distinct messages.¶
The previous message or protocol state MUST NOT be kept longer than what is required for retransmission, for example, in the case of CoAP transport, no longer than the EXCHANGE_LIFETIME (see Section 4.8.2 of [RFC7252]).¶
In the absence of an application profile specifying otherwise:¶
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 an identity misbinding attack like the Duplicate Signature Key Selection (DSKS) 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 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 the 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 that is supported by both the Initiator and Responder and to abort the EDHOC session if not.¶
As required by [RFC7258], IETF protocols need to mitigate pervasive monitoring when possible. Therefore, EDHOC 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. Therefore, EDHOC 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 HKDF 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 algorithm 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 re-run 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 [EDHOC-CoAP-OSCORE], 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 the 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.¶
Earlier versions of EDHOC have been formally analyzed [Bruni18] [Norrman20] [CottierPointcheval22] [Jacomme23] [GuentherIlunga22], and the specification has been updated based on the analysis.¶
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, i.e., 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 only needs to protect against a passive attacker since 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 that is proven secure as long as the expand function is a Pseudorandom Function (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 adaptive chosen ciphertext attack (IND-CCA2) as well as integrity protection. For the encryption of message_2, any speed difference is negligible, IND-CCA2 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 public keys depend on the elliptic curve. For X25519 and X448, the requirements are defined in [RFC7748]. For X25519 and X448, the check for all-zero output as specified in Section 6 of [RFC7748] MUST be done. 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 a secure shared signature and ECDH key usage in [Degabriele11] and [Thormarker21]. Note that Section 5.6.3.2 of [SP-800-56A] allows a key agreement key pair to be used with a signature algorithm in certificate requests.¶
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., PSKs 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 an identity in each initial TOFU session is RECOMMENDED.¶
NIST SP 800-56A [SP-800-56A] forbids deriving secret and non-secret randomness from the same Key Derivation Function (KDF) instance, but this decision has been criticized by Krawczyk in [HKDFpaper] and doing so is common practice. In addition to IVs, other examples are the challenge in Extensible Authentication Protocol Tunneled Transport Layer Security (EAP-TTLS), the RAND in 3GPP Authentication and Key Agreement (AKA), 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. The more recent NIST SP 800-108 [SP-800-108] aligns with [HKDFpaper] and states that, for a secure KDF, the revelation of one portion of the derived keying material must not degrade the security of any other portion of that keying material.¶
When using a private cipher suite or registering new cipher suites, the choice of the 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 Responder should enforce a minimum security level.¶
The output size of the EDHOC hash algorithm MUST be at least 256 bits. In particular, 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 all the 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.¶
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. Two of NIST's security levels for quantum-resistant public key cryptography are based on AES-128 and SHA-256. A quantum computer will likely be expensive and slow due to heavy error correction. Grover's algorithm, which is proven to be optimal, cannot effectively be parallelized. It will provide little or no advantage in attacking AES, and AES-128 will remain secure for decades to come [NISTPQC].¶
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 the 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.¶
The Initiator and 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 are revealed to passive attackers and the ead_labels in EAD_2 are 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 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.¶
Since the publication of [RFC3552], there has been an increased awareness of the need to protect against endpoints that are compromised or malicious or whose interests simply do not align with the interests of users [THREAT-MODEL-GUIDANCE]. [RFC7624] describes an updated threat model for Internet confidentiality; see Section 9.1. [THREAT-MODEL-GUIDANCE] further expands the threat model. Implementations and users should take these threat models into account and consider actions to reduce the risk of tracking by other endpoints. 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 [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 a 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'.¶
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 the state, perform cryptographic operations, and amplify messages. To mitigate such attacks, an implementation SHOULD make use of available 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 round trip, the Initiator can reduce the amplification factor by padding message_1, i.e., using EAD_1; see Section 3.8.1. Note that while the Echo option mitigates some resource exhaustion aspects of spoofing, it does not protect against a distributed denial-of-service attack made by real, potentially compromised, clients. Similarly, limiting amplification only reduces the impact, which still may be significant because of a large number of clients engaged in the attack.¶
An attacker can also send a 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.¶
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 of [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.¶
For many constrained IoT devices, it is problematic to support several crypto primitives. Existing devices can be expected to support either ECDSA or Edwards-curve Digital Signature Algorithm (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 does 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. For example, see Section 1 of [HEDGED-ECC-SIGS] 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 [CURVE-REPR] describes how Montgomery curves, such as X25519 and X448, and (twisted) Edwards curves, such as Ed25519 and Ed448, can be mapped to and from short-Weierstrass form for implementations on platforms that accelerate elliptic curve group operations in short-Weierstrass form.¶
All private keys, symmetric keys, and IVs MUST be 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. 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 Responder are responsible for verifying the integrity and validity of certificates. Verification of validity may require the use of a Real-Time Clock (RTC). The selection of trusted certification authorities (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 the Online Certificate Status Protocol (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 inputs. 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.¶
The Initiator and Responder are allowed to select connection identifiers 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 the 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) (for example, see [RFC9397]) 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 a 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, and 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. Implementations are not required to support non-deterministic encodings and MAY abort the EDHOC session if the received EDHOC message is not encoded using deterministic CBOR. Implementations MUST abort the EDHOC session if validation of a received public key fails or if any cryptographic field has the wrong length. It is RECOMMENDED to abort the EDHOC session if the received EDHOC message is not encoded using deterministic CBOR.¶
This section gives IANA considerations and, unless otherwise noted, conforms with [RFC8126].¶
IANA has created a new registry under the new registry group "Ephemeral Diffie-Hellman Over COSE (EDHOC)" as follows:¶
Label | Description | Reference |
---|---|---|
0 | Derived OSCORE Master Secret | RFC 9528 |
1 | Derived OSCORE Master Salt | RFC 9528 |
2-22 | Unassigned | |
23 | Reserved | RFC 9528 |
24-32767 | Unassigned | |
32768-65535 | Reserved for Private Use |
This registry also has a "Change Controller" field. For registrations made by IETF documents, the IETF is listed.¶
Range | Registration Procedures |
---|---|
0-23 | Standards Action |
24-32767 | Expert Review |
32768-65535 | Private Use |
IANA has created a new registry under the new registry group "Ephemeral Diffie-Hellman Over COSE (EDHOC)" as follows:¶
The columns of the registry are Value, Array, Description, and Reference, where Value is an integer and the other columns are text strings. The initial contents of the registry are:¶
Value | Array | Description | Reference |
---|---|---|---|
-24 | N/A | Private Use | RFC 9528 |
-23 | N/A | Private Use | RFC 9528 |
-22 | N/A | Private Use | RFC 9528 |
-21 | N/A | Private Use | RFC 9528 |
0 | 10, -16, 8, 4, -8, 10, -16 | AES-CCM-16-64-128, SHA-256, 8, X25519, EdDSA, AES‑CCM‑16‑64‑128, SHA-256 | RFC 9528 |
1 | 30, -16, 16, 4, -8, 10, -16 | AES-CCM-16-128-128, SHA‑256, 16, X25519, EdDSA, AES‑CCM‑16‑64‑128, SHA-256 | RFC 9528 |
2 | 10, -16, 8, 1, -7, 10, -16 | AES-CCM-16-64-128, SHA-256, 8, P-256, ES256, AES‑CCM‑16‑64‑128, SHA-256 | RFC 9528 |
3 | 30, -16, 16, 1, -7, 10, -16 | AES-CCM-16-128-128, SHA‑256, 16, P-256, ES256, AES‑CCM‑16‑64‑128, SHA-256 | RFC 9528 |
4 | 24, -16, 16, 4, -8, 24, -16 | ChaCha20/Poly1305, SHA-256, 16, X25519, EdDSA, ChaCha20/Poly1305, SHA-256 | RFC 9528 |
5 | 24, -16, 16, 1, -7, 24, -16 | ChaCha20/Poly1305, SHA-256, 16, P-256, ES256, ChaCha20/Poly1305, SHA-256 | RFC 9528 |
6 | 1, -16, 16, 4, -7, 1, -16 | A128GCM, SHA-256, 16, X25519, ES256, A128GCM, SHA-256 | RFC 9528 |
23 | Reserved | RFC 9528 | |
24 | 3, -43, 16, 2, -35, 3, -43 | A256GCM, SHA-384, 16, P-384, ES384, A256GCM, SHA-384 | RFC 9528 |
25 | 24, -45, 16, 5, -8, 24, -45 | ChaCha20/Poly1305, SHAKE256, 16, X448, EdDSA, ChaCha20/Poly1305, SHAKE256 | RFC 9528 |
Range | Registration Procedures |
---|---|
-65536 to -25 | Specification Required |
-24 to -21 | Private Use |
-20 to 23 | Standards Action with Expert Review |
24 to 65535 | Specification Required |
IANA has created a new registry under the new registry group "Ephemeral Diffie-Hellman Over COSE (EDHOC)" as follows:¶
The columns of the registry are Value, Initiator Authentication Key, 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 Table 2. Method 23 is Reserved.¶
Range | Registration Procedures |
---|---|
-65536 to -25 | Specification Required |
-24 to 23 | Standards Action with Expert Review |
24 to 65535 | Specification Required |
IANA has created a new registry under the new registry group "Ephemeral Diffie-Hellman Over COSE (EDHOC)" as follows:¶
The columns of the registry are ERR_CODE, ERR_INFO Type, Description, Change Controller, 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 Table 3. Error code 23 is Reserved. This registry also has a "Change Controller" field. For registrations made by IETF documents, the IETF is listed.¶
Range | Registration Procedures |
---|---|
-65536 to -25 | Expert Review |
-24 to 23 | Standards Action |
24 to 65535 | Expert Review |
IANA has created a new registry under the new registry group "Ephemeral Diffie-Hellman Over COSE (EDHOC)" as follows:¶
The columns of the registry are Name, Label, Description, and Reference, where Label is a nonnegative integer and the other columns are text strings. The initial contents of the registry are shown in Table 10. EAD label 23 is Reserved.¶
Name | Label | Description | Reference |
---|---|---|---|
Padding | 0 | Randomly generated CBOR byte string | RFC 9528, Section 3.8.1 |
23 | Reserved | RFC 9528 |
Range | Registration Procedures |
---|---|
0 to 23 | Standards Action with Expert Review |
24 to 65535 | Specification Required |
IANA has registered the following entries in the "COSE Header Parameters" registry under the registry group "CBOR Object Signing and Encryption (COSE)" (see Table 12). 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 column for this item is empty and omitted from the table below.¶
Name | Label | Value Type | Description |
---|---|---|---|
kcwt | 13 | COSE_Messages | A CBOR Web Token (CWT) containing a COSE_Key in a 'cnf' claim and possibly other claims. CWT is defined in RFC 8392. COSE_Messages is defined in RFC 9052. |
kccs | 14 | map | A CWT Claims Set (CCS) containing a COSE_Key in a 'cnf' claim and possibly other claims. CCS is defined in RFC 8392. |
IANA has added the well-known URI "edhoc" to the "Well-Known URIs" registry.¶
IANA has added the media types "application/edhoc+cbor-seq" and "application/cid-edhoc+cbor-seq" to the "Media Types" registry.¶
IANA has added 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".¶
Content Type | Content Coding | ID | Reference |
---|---|---|---|
application/edhoc+cbor-seq | - | 64 | RFC 9528 |
application/cid-edhoc+cbor-seq | - | 65 | RFC 9528 |
IANA has added the resource type "core.edhoc" to the "Resource Type (rt=) Link Target Attribute Values" registry under the registry group "Constrained RESTful Environments (CoRE) Parameters".¶
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:¶
This appendix describes how to derive an OSCORE security context when EDHOC is used to key OSCORE and how to transfer EDHOC messages over CoAP. The use of CoAP or OSCORE with EDHOC is optional, but if you are using CoAP or OSCORE, then certain normative requirements apply as detailed in the subsections.¶
This section specifies how to use EDHOC output to derive the OSCORE security context.¶
After successful processing of EDHOC message_3, the 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):¶
Master Secret = EDHOC_Exporter( 0, h'', oscore_key_length ) Master Salt = EDHOC_Exporter( 1, h'', oscore_salt_length )¶
OSCORE Sender ID | OSCORE Recipient ID | |
---|---|---|
EDHOC Initiator | C_R | C_I |
EDHOC Responder | C_I | C_R |
The 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, the Client and Server retrieve the OSCORE protocol state using the Recipient ID and optionally other transport information such as the 5-tuple.¶
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, provides flow and congestion control, and handles message duplication. CoAP can also perform fragmentation and mitigate certain 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 a 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 [EDHOC-CoAP-OSCORE].¶
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 that 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). 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 mitigate certain 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:¶
true
) to any request message it sends.¶
In the forward message flow, the CoAP client is the Initiator and the CoAP server is the Responder. This flow protects the client identity against active attackers and the server identity against passive attackers.¶
In the forward message flow, the CoAP Token enables correlation on the Initiator (client) side, and the prepended C_R enables correlation on the Responder (server) side.¶
true
(0xf5), indicating a new EDHOC session.¶
An example of a completed EDHOC session over CoAP in the forward message flow is shown in Figure 10.¶
The forward message flow of EDHOC can be combined with an OSCORE exchange in a total of two round trips; see [EDHOC-CoAP-OSCORE].¶
In the reverse message flow, the CoAP client is the Responder and the CoAP server is the Initiator. This flow protects the server identity against active attackers and the client identity against passive attackers.¶
In the reverse message flow, the CoAP Token enables correlation on the Responder (client) side, and the prepended C_I enables correlation on the Initiator (server) side.¶
An example of a completed EDHOC session over CoAP in the reverse message flow is shown in Figure 11.¶
When using EDHOC over CoAP, EDHOC error messages sent as CoAP responses MUST be sent in the payload of error responses, i.e., they MUST specify a CoAP error response code. In particular, it is RECOMMENDED that such error responses have response code either 4.00 (Bad Request) in case of client error (e.g., due to a malformed EDHOC message) or 5.00 (Internal Server Error) in case of server error (e.g., due to failure in deriving EDHOC keying material). The Content-Format of the error response MUST be set to "application/edhoc+cbor-seq"; see Section 10.9.¶
This section defines a format for compact representation based on the Elliptic-Curve-Point-to-Octet-String Conversion defined in Section 2.3.3 of [SECG].¶
As described in Section 4.2 of [RFC6090], the x-coordinate of an elliptic curve public key is a suitable representative for the entire point whenever scalar multiplication is used as a one-way function. One example is ECDH with compact output, where only the x-coordinate of the computed value is used as the shared secret.¶
In EDHOC, compact representation is used for the ephemeral public keys (G_X and G_Y); see Section 3.7. Using the notation from [SECG], the output is an octet string of length ceil( (log2 q) / 8 ), where ceil(x) is the smallest integer not less than x. See [SECG] for a definition of q, M, X, xp, and ~yp. The steps in Section 2.3.3 of [SECG] are replaced with the following steps:¶
The encoding of the point at infinity is not supported.¶
Compact representation does not change any requirements on validation; see Section 9.2. Using compact representation has some security benefits. An implementation does not need to check that the point is not the point at infinity (the identity element). Similarly, as not even the sign of the y-coordinate is encoded, compact representation trivially avoids so-called "benign malleability" attacks where an attacker changes the sign; see [SECG].¶
The following may be needed for validation or compatibility with APIs that do not support compact representation or do not support the full [SECG] format:¶
For example: The curve P-256 has the parameters (using the notation in [RFC6090]):¶
Given an example x:¶
We can calculate y as the square root w = (x3 + a ⋅ x + b)((p + 1)/4) (mod p).¶
Note that this does not guarantee that (x, y) is on the correct elliptic curve. A full validation according to Section 5.6.2.3.3 of [SP-800-56A] is done by also checking that 0 ≤ x < p and that y2 ≡ x3 + a ⋅ x + b (mod p).¶
This appendix is intended to help implementors not familiar with CBOR [RFC8949], CDDL [RFC8610], COSE [RFC9052], and HKDF [RFC5869].¶
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 types of data items, 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. That is, 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].¶
Diagnostic | Encoded | Type |
---|---|---|
1 | 0x01 | unsigned integer |
24 | 0x1818 | unsigned integer |
-24 | 0x37 | negative integer |
-25 | 0x3818 | negative integer |
true | 0xf5 | simple value |
h'' | 0x40 | byte string |
h'12cd' | 0x4212cd | byte string |
'12cd' | 0x4431326364 | byte string |
"12cd" | 0x6431326364 | text string |
{ 4 : h'cd' } | 0xa10441cd | map |
<< 1, 2, true >> | 0x430102f5 | byte string |
[ 1, 2, true ] | 0x830102f5 | array |
( 1, 2, true ) | 0x0102f5 | sequence |
1, 2, true | 0x0102f5 | sequence |
This section 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, ) PLAINTEXT_2 = ( C_R, ID_CRED_R : map / bstr / -24..23, Signature_or_MAC_2 : bstr, ? EAD_2, ) message_3 = ( CIPHERTEXT_3 : bstr, ) PLAINTEXT_3 = ( ID_CRED_I : map / bstr / -24..23, Signature_or_MAC_3 : bstr, ? EAD_3, ) message_4 = ( CIPHERTEXT_4 : bstr, ) PLAINTEXT_4 = ( ? EAD_4, ) error = ( ERR_CODE : int, ERR_INFO : any, ) info = ( info_label : int, context : bstr, length : uint, )¶
CBOR Object Signing and Encryption (COSE) [RFC9052] describes how to create and process signatures, MACs, and encryptions using CBOR. COSE builds on JSON Object Signing and Encryption (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:¶
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 and 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 Sections 5.4 and 5.5, respectively) as follows:¶
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 and 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 Sections 5.3 and 5.4.¶
Different header parameters to identify X.509 or C509 certificates by reference are defined in [RFC9360] and [C509-CERTS]:¶
by a hash value with the 'x5t' or 'c5t' parameters, respectively:¶
or by a URI with the 'x5u' or 'c5u' parameters, respectively:¶
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':¶
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, as defined in [C509-CERTS] and [RFC9360]. Credentials of type CWT and CCS can be transported with the COSE header parameters registered in Section 10.6.¶
EDHOC performs certain authentication-related operations (see Section 3.5), but in general, it is necessary to make additional verifications beyond EDHOC message processing. Which verifications that are needed depend on the deployment, in particular, the trust model and the security policies, but most commonly, it can be expressed in terms of verifications of credential content.¶
EDHOC assumes the existence of mechanisms (certification authority or other trusted third party, pre-provisioning, etc.) for generating and distributing authentication credentials and other credentials, as well as the existence of trust anchors (CA certificates, trusted public keys, etc.). For example, a public key certificate or CWT may rely on a trusted third party whose public key is pre-provisioned, whereas a CCS or a self-signed certificate / CWT may be used when trust in the public key can be achieved by other means, or in the case of trust on first use, see Appendix D.5.¶
In this section, we provide some examples of such verifications. These verifications are the responsibility of the application but may be implemented as part of an EDHOC library.¶
In addition to the authentication key, the authentication credential may contain other parameters that need to be verified. For example:¶
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 that 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 that authenticates with a particular public key.¶
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.)¶
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 a 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 the trust anchor to the 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 and self-signed certificate / CWT), the trust anchor is the authentication key of the other party; in which case, there is no certification path.¶
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 Authorization Data (EAD); see Appendix E.¶
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 active on-path 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 are listed below:¶
In order to reduce the number of messages and round trips, or to simplify processing, external security applications may be integrated into EDHOC by transporting related external authorization data (EAD) in the messages.¶
The EAD format is specified in Section 3.8. This section contains examples and further details of how EAD may be used with an appropriate accompanying specification.¶
External authorization data should be considered unprotected by EDHOC, and the protection of EAD is the responsibility of the security application (third-party authorization, remote attestation, certificate enrollment, etc.). The security properties of the EAD fields (after EDHOC processing) are discussed in Section 9.1.¶
The content of the EAD field may be used in the EDHOC processing of the message in which they are contained. For example, authentication-related information, like assertions and revocation information, transported in EAD fields may provide input about trust anchors or validity of credentials relevant to the authentication processing. The EAD fields (like ID_CRED fields) are therefore made available to the application before the message is verified; see details of message processing in Section 5. In the first example above, a voucher in EAD_2 made available to the application can enable the Initiator to verify the identity or the public key of the Responder before verifying the signature. An application allowing EAD fields containing authentication information thus may need to handle authentication-related verifications associated with EAD processing.¶
Conversely, the security application may need to wait for EDHOC message verification to complete. In the third example above, the validation of a CSR carried in EAD_3 is not started by the Responder before EDHOC has successfully verified message_3 and proven the possession of the private key of the Initiator.¶
The security application may reuse EDHOC protocol fields that therefore need to be available to the application. For example, the security application may use the same crypto algorithms as in the EDHOC session and therefore needs access to the selected cipher suite (or the whole SUITES_I). The application may use the ephemeral public keys G_X and G_Y as ephemeral keys or as nonces; see [LAKE-AUTHZ].¶
The processing of the EAD item (ead_label, ? ead_value) by the security application needs to be described in the specification where the ead_label is registered (see Section 10.5), including the optional ead_value for each message and actions in case of errors. An application may support multiple security applications that make use of EAD, which may result in multiple EAD items in one EAD field; see Section 3.8. Any dependencies on security applications with previously registered EAD items need to be documented, and the processing needs to consider their simultaneous use.¶
Since data carried in EAD may not be protected, or processed by the application before the EDHOC message is verified, special considerations need to be made such that it does not violate security and privacy requirements of the service that uses this data; see Section 9.5. The content in an EAD item may impact the security properties provided by EDHOC. Security applications making use of the EAD items must perform the necessary security analysis.¶
This appendix contains a rudimentary example of an application profile; see Section 3.9.¶
For use of EDHOC with application X, the following assumptions are made:¶
CRED_I is an IEEE 802.1AR Initial Device Identifier (IDevID) encoded as a C509 certificate of type 0 [C509-CERTS].¶
CRED_R is a CCS of type OKP as specified in Section 3.5.2.¶
By the definition of encryption of PLAINTEXT_2 with KEYSTREAM_2, it is limited to lengths of PLAINTEXT_2 not exceeding the output of EDHOC_KDF; see Section 4.1.2. If the EDHOC hash algorithm is SHA-2, then HKDF-Expand is used, which limits the length of the EDHOC_KDF output to 255 ⋅ hash_length, where hash_length is the length of the output of the EDHOC hash algorithm given by the cipher suite. For example, with SHA-256 as the EDHOC hash algorithm, the length of the hash output is 32 bytes and the maximum length of PLAINTEXT_2 is 255 ⋅ 32 = 8160 bytes.¶
While PLAINTEXT_2 is expected to be much shorter than 8 kB for the intended use cases, it seems nevertheless prudent to specify a solution for the event that this should turn out to be a limitation.¶
A potential work-around is to use a cipher suite with a different hash function. In particular, the use of KMAC removes all practical limitations in this respect.¶
This section specifies a solution that works with any hash function by making use of multiple invocations of HKDF-Expand and negative values of info_label.¶
Consider the PLAINTEXT_2 partitioned in parts P(i) of length equal to M = 255 ⋅ hash_length, except possibly the last part P(last), which has 0 < length ≤ M.¶
PLAINTEXT_2 = P(0) | P(1) | ... | P(last)¶
where "|" indicates concatenation.¶
The object is to define a matching KEYSTREAM_2 of the same length and perform the encryption in the same way as defined in Section 5.3.2:¶
CIPHERTEXT_2 = PLAINTEXT_2 XOR KEYSTREAM_2¶
Define the keystream as:¶
KEYSTREAM_2 = OKM(0) | OKM(1) | ... | OKM(last)¶
where:¶
OKM(i) = EDHOC_KDF( PRK_2e, -i, TH_2, length(P(i)) )¶
Note that if length(PLAINTEXT_2) ≤ M, then P(0) = PLAINTEXT_2 and the definition of KEYSTREAM_2 = OKM(0) coincides with Figure 6.¶
This describes the processing of the Responder when sending message_2. The Initiator makes the same calculations when receiving message_2 but interchanging PLAINTEXT_2 and CIPHERTEXT_2.¶
An application profile may specify if it supports or does not support the method described in this appendix.¶
To provide forward secrecy in an even more efficient way than re-running EDHOC, this section specifies the optional function EDHOC_KeyUpdate in terms of EDHOC_KDF and PRK_out.¶
When EDHOC_KeyUpdate is called, a new PRK_out is calculated as the output of the EDHOC_Expand function with the old PRK_out as input. The change of PRK_out causes a change to PRK_exporter, which enables the derivation of new application keys superseding the old ones, using EDHOC_Exporter; see Section 4.2.1. The process is illustrated by the following pseudocode.¶
EDHOC_KeyUpdate( context ): new PRK_out = EDHOC_KDF( old PRK_out, 11, context, hash_length ) new PRK_exporter = EDHOC_KDF( new PRK_out, 10, h'', hash_length )¶
where hash_length denotes the output size in bytes of the EDHOC hash algorithm of the selected cipher suite.¶
The EDHOC_KeyUpdate takes a context as input to enable binding of the updated PRK_out to some event that triggered the key update. The Initiator and Responder need to agree on the context, which can, e.g., be a counter, a pseudorandom number, or a hash. To provide forward secrecy, the old PRK_out and keys derived from it (old PRK_exporter and old application keys) must be deleted as soon as they are not needed. When to delete the old keys and how to verify that they are not needed is up to the application. Note that the security properties depend on the type of context and the number of KeyUpdate iterations.¶
An application using EDHOC_KeyUpdate needs to store PRK_out. Compromise of PRK_out leads to compromise of all keying material derived with the EDHOC_Exporter since the last invocation of the EDHOC_KeyUpdate function.¶
While this key update method provides forward secrecy, it does not give as strong security properties as re-running EDHOC. EDHOC_KeyUpdate can be used to meet cryptographic limits and provide partial protection against key leakage, but it provides significantly weaker security properties than re-running EDHOC with ephemeral Diffie-Hellman. Even with frequent use of EDHOC_KeyUpdate, compromise of one session key compromises all future session keys, and an attacker therefore only needs to perform static key exfiltration [RFC7624], which is less complicated and has a lower risk profile than the dynamic case; see Section 9.1.¶
A similar method to do a key update for OSCORE is KUDOS; see [KUDOS].¶
This appendix describes an example protocol state machine for the Initiator and Responder. States are denoted in all capitals, and parentheses denote actions taken only in some circumstances.¶
Note that this state machine is just an example, and that details of processing are omitted. For example:¶
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 Initiator transitions from COMPLETED to PERSISTED.¶
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 an 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.¶
The authors want to thank Christian Amsüss, Karthikeyan Bhargavan, Carsten Bormann, Alessandro Bruni, Timothy Claeys, Baptiste Cottier, Roman Danyliw, Martin Disch, Martin Duke, Donald Eastlake 3rd, Lars Eggert, Stephen Farrell, Loïc Ferreira, Theis Grønbech Petersen, Felix Günther, Dan Harkins, Klaus Hartke, Russ Housley, Stefan Hristozov, Marc Ilunga, Charlie Jacomme, Elise Klein, Erik Kline, Steve Kremer, Alexandros Krontiris, Ilari Liusvaara, Rafa Marín-López, Kathleen Moriarty, David Navarro, Karl Norrman, Salvador Pérez, Radia Perlman, David Pointcheval, Maïwenn Racouchot, Eric Rescorla, Michael Richardson, Thorvald Sahl Jørgensen, Zaheduzzaman Sarker, Jim Schaad, Michael Scharf, Carsten Schürmann, John Scudder, Ludwig Seitz, Brian Sipos, Stanislav Smyshlyaev, Valery Smyslov, Peter van der Stok, Rene Struik, Vaishnavi Sundararajan, Erik Thormarker, Marco Tiloca, Sean Turner, Michel Veillette, Mališa Vučinić, Paul Wouters, and Lei Yan for reviewing and commenting on intermediate draft versions of this document.¶
We are especially indebted to the late Jim Schaad for his continuous review and implementation of draft versions of this document, as well as his work on other technologies such as COSE and OSCORE without which EDHOC would not have been.¶
Work on this document has in part been supported by the H2020 project SIFIS-Home (grant agreement 952652).¶