RFC 9711
The Entity Attestation Token (EAT)
RFC 9711: The Entity Attestation Token (EAT) 💖
- Who wrote it: L. Lundblade (Security Theory LLC), G. Mandyam (Qualcomm Technologies Inc.), J. O'Donoghue (Qualcomm Technologies Inc.), and C. Wallace (Red Hound Software, Inc.) 👑
- Category: Standards Track 🎀
- Date: April 2025 🌸
- ISSN: 2070-1721 ✨
- Obsoletes/Updates: the header doesn't list anything this one obsoletes or updates 🦋
- Published by: the Internet Engineering Task Force (IETF) 🧚
Abstract ✨
okay bestie, so an Entity Attestation Token (EAT) is an attested claims set 💅 that describes the state and characteristics of an entity. An entity can be a smartphone, an Internet of Things (IoT) device, network equipment, or something similar 🌈
The relying party, server, or service uses this claims set to decide how much trust to place in the entity. It's giving "check the receipts before you trust" energy 🔥
An EAT is either a CBOR Web Token (CWT) or a JSON Web Token (JWT) with attestation-oriented claims 🫶
Status of This Memo 📜
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/rfc9711 🔗
Copyright Notice 🎀
Copyright (c) 2025 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, because 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 they are provided without warranty as described in the Revised BSD License 🌸
Table of Contents 🧚
1. Introduction
1.1. Entity Overview
1.2. EAT as a Framework
1.3. Operating Model and RATS Architecture
1.3.1. Relationship between Evidence and Attestation Results
2. Terminology
3. Top-Level Token Definition
4. The Claims
4.1. eat_nonce (EAT Nonce) Claim
4.2. Claims Describing the Entity
4.2.1. ueid (Universal Entity ID) Claim
4.2.1.1. Rules for Creating UEIDs
4.2.1.2. Rules for Consuming UEIDs
4.2.2. sueids (Semipermanent UEIDs) Claim
4.2.3. oemid (Hardware OEM ID) Claim
4.2.3.1. Random Number-Based OEM ID
4.2.3.2. IEEE-Based OEM ID
4.2.3.3. IANA Private Enterprise Number-Based OEM ID
4.2.4. hwmodel (Hardware Model) Claim
4.2.5. hwversion (Hardware Version) Claim
4.2.6. swname (Software Name) Claim
4.2.7. swversion (Software Version) Claim
4.2.8. oemboot (OEM Authorized Boot) Claim
4.2.9. dbgstat (Debug Status) Claim
4.2.9.1. Enabled
4.2.9.2. Disabled
4.2.9.3. Disabled Since Boot
4.2.9.4. Disabled Permanently
4.2.9.5. Disabled Fully and Permanently
4.2.10. location (Location) Claim
4.2.11. uptime (Uptime) Claim
4.2.12. bootcount (Boot Count) Claim
4.2.13. bootseed (Boot Seed) Claim
4.2.14. dloas (Digital Letters of Approval) Claim
4.2.15. manifests (Software Manifests) Claim
4.2.16. measurements (Measurements) Claim
4.2.17. measres (Software Measurement Results) Claim
4.2.18. submods (Submodules) Claim
4.2.18.1. Submodule Claims-Set
4.2.18.2. Detached Submodule Digest
4.2.18.3. Nested Tokens
4.3. Claims Describing the Token
4.3.1. iat (Timestamp) Claim
4.3.2. eat_profile (EAT Profile) Claim
4.3.3. intuse (Intended Use) Claim
5. Detached EAT Bundles
6. Profiles
6.1. Format of a Profile Document
6.2. Full and Partial Profiles
6.3. List of Profile Issues
6.3.1. Use of JSON, CBOR, or Both
6.3.2. CBOR Map and Array Encoding
6.3.3. CBOR String Encoding
6.3.4. CBOR Preferred Serialization
6.3.5. CBOR Tags
6.3.6. COSE/JOSE Protection
6.3.7. COSE/JOSE Algorithms
6.3.8. Detached EAT Bundle Support
6.3.9. Key Identification
6.3.10. Endorsement Identification
6.3.11. Freshness
6.3.12. Claims Requirements
6.4. The Constrained Device Standard Profile
7. Encoding and Collected CDDL
7.1. Claims-Set and CDDL for CWT and JWT
7.2. Encoding Data Types
7.2.1. Common Data Types
7.2.2. JSON Interoperability
7.2.3. Labels
7.2.4. CBOR Interoperability
7.3. Collected CDDL
7.3.1. Payload CDDL
7.3.2. CBOR-Specific CDDL
7.3.3. JSON-Specific CDDL
8. Privacy Considerations
8.1. UEID and SUEID Privacy Considerations
8.2. Location Privacy Considerations
8.3. Boot Seed Privacy Considerations
8.4. Replay Protection and Privacy
9. Security Considerations
9.1. Claim Trustworthiness
9.2. Key Provisioning
9.2.1. Transmission of Key Material
9.3. Freshness
9.4. Multiple EAT Consumers
9.5. Detached EAT Bundle Digest Security Considerations
9.6. Verification Keys
10. IANA Considerations
10.1. Reuse of CBOR and JSON Web Token (CWT and JWT) Claims
Registries
10.2. CWT and JWT Claims Registered by This Document
10.3. UEID URNs Registered by This Document
10.4. CBOR Tag for Detached EAT Bundle Registered by This
Document
10.5. Intended Use Registry
11. References
11.1. Normative References
11.2. Informative References
Appendix A. Examples
A.1. Claims Set Examples
A.1.1. Simple TEE Attestation
A.1.2. Submodules for Board and Device
A.1.3. EAT Produced by an Attestation Hardware Block
A.1.4. Key / Key Store Attestation
A.1.5. Software Measurements of an IoT Device
A.1.6. Attestation Results in JSON
A.1.7. JSON-Encoded Token with Submodules
A.2. Signed Token Examples
A.2.1. Basic CWT Example
A.2.2. CBOR-Encoded Detached EAT Bundle
A.2.3. JSON-Encoded Detached EAT Bundle
Appendix B. UEID Design Rationale
B.1. Collision Probability
B.2. No Use of UUID
Appendix C. EAT Relation to IEEE.802.1AR Secure Device Identity
(DevID)
C.1. DevID Used with EAT
C.2. How EAT Provides an Equivalent Secure Device Identity
C.3. An X.509 Format EAT
C.4. Device Identifier Permanence
Appendix D. CDDL for CWT and JWT
Appendix E. New Claim Design Considerations
E.1. Interoperability and Relying Party Orientation
E.2. Operating System and Technology Neutral
E.3. Security Level Neutral
E.4. Reuse of Extant Data Formats
E.5. Proprietary Claims
Appendix F. Endorsements and Verification Keys
F.1. Identification Methods
F.1.1. COSE/JWS Key ID
F.1.2. JWS and COSE X.509 Header Parameters
F.1.3. CBOR Certificate COSE Header Parameters
F.1.4. Claim-Based Key Identification
Contributors
Authors' Addresses1. Introduction ✨
okay bestie, so an Entity Attestation Token (EAT) is basically a message full of claims about an entity 🎀 An entity can be a device, some hardware, or some software 🖥️ The claims eventually get used by a relying party, who decides if and how it's gonna interact with the entity. They can choose to trust it, not trust it, or partially trust it 💅 Partial trust could look like only letting a monetary transaction through up to a limit 💸
The security model and goal for attestation are giving unique 🦋 They are not the same as the ones for other security standards like server authentication, user authentication, and secured messaging. Here's one example of how they're different: think about the association and life cycle of key material 🔑 For authentication, keys belong to a user or a service and get set up by actions a user or a service operator takes. For attestation, the keys belong to specific devices and get configured by device manufacturers 🏭 The reader is assumed to already know the goals and security model for attestation, which are described in "Remote ATtestation procedureS (RATS) Architecture" [RFC9334], so we're not repeating them here 🙈
This document defines some common claims that could be useful to a lot of people 🌸 EAT also allows proprietary claims, and more claims can be standardized later. Here are some examples:
- Make and model of manufactured consumer device
- Make and model of a chip or processor, particularly for a security-oriented chip
- Identification and measurement of the software running on a device
- Configuration and state of a device
- Environmental characteristics of a device such as its Global Positioning System (GPS) location
- Formal certifications received
EAT is built to support a wide range of use cases, and it's honestly so iconic for that 👑
No single set of claims can cover every use case, so EAT is a framework for defining specific attestation tokens for specific use cases 🧚 In particular, EAT has a profile mechanism that lets you clearly specify the claims needed, the cryptographic algorithms that should be used, and other characteristics for a particular token and use case. [Section 6](#) describes profile contents and gives a profile that's suitable for constrained device use cases. Section 6 describes profile contents and provides a profile that is suitable for constrained device use cases 💖
The entity's EAT implementation generates the claims and typically signs them with an attestation key. It's responsible for protecting that attestation key 🔐 Some EAT implementations use components with very high resistance to attack, like Trusted Platform Modules or Secure Elements 🛡️ Others may rely only on simple software defenses. Not gonna lie, that's a big range 😭
Nesting of tokens and claims sets is supported for composite devices that have multiple subsystems 🧩
An EAT may be encoded in either JavaScript Object Notation (JSON) [RFC8259] or Concise Binary Object Representation (CBOR) [RFC8949], depending on what each use case needs 📦 EAT is built on the CBOR Web Token (CWT) [RFC8392] and the JSON Web Token (JWT) [RFC7519], and it inherits all of their characteristics and their security mechanisms 🫶 Just like CWT and JWT, EAT does not imply any message flow. Nothing is forced on you 🌈
Here's a very simple example. It's in JSON so it's easy to read, but it could also be CBOR. Only the Claims-Set, which is the payload for the JWT, is shown 🍓
{
"eat_nonce": "MIDBNH28iioisjPy",
"ueid": "AgAEizrK3Q",
"oemid": 76543,
"swname": "Acme IoT OS",
"swversion": "3.1.4"
}This example has a nonce for freshness ✨ The nonce is the base64url encoding of a 12-byte random binary byte string. The ueid (Universal Entity ID) is basically a serial number that uniquely identifies the device 🆔 This ueid is the base64url encoding of a 48-bit Media Access Control (MAC) address, preceded by the type byte 0x02. The oemid (Hardware OEM ID) identifies the manufacturer using a Private Enterprise Number (PEN) [PEN] 🏷️ The software is identified by a simple string name and version. It could be identified by a full manifest, but this is a minimal example, so we keep it chill 😌
1.1. Entity Overview
This document uses the term "entity" to mean the target of an EAT 🎯 Most of the claims defined in this document are claims about an entity. An entity is equivalent to a target environment in an attester, as defined in [RFC9334] 💜
Layered attestation and composite devices, as described in [RFC9334], are supported by a submodule mechanism (see Section 4.2.18). Submodules let you nest EATs and Claims-Sets so those hierarchies can be modeled 🪆
An entity is the same as a "system component," as defined in the Internet Security Glossary [RFC4949] 📖
Note that [RFC4949] defines "entity" and "system entity" as synonyms, and that they may be a person or organization in addition to being a system component. But in the EAT context, "entity" never refers to a person or organization 🙅♀️ The hardware and software that run a website server or service may be an entity in the EAT sense, but the organization that operates, maintains, or hosts the website is not an entity. Boundaries, babe 💅
Some examples of entities:
- A Secure Element
- A Trusted Execution Environment (TEE)
- A network card in a router
- A router, perhaps with each network card in the router being a submodule
- An IoT device
- An individual process
- An app on a smartphone
- A smartphone with many submodules for its many subsystems
- A subsystem in a smartphone such as the modem or the camera
An entity may have strong security defenses against hardware-invasive attacks. It may also have low security, i.e., no special security defenses at all 🥲 There is no minimum security requirement to be an entity. Literally anything can qualify 🌸
1.2. EAT as a Framework
EAT is a framework used for defining attestation tokens for specific use cases. It is not used for defining one specific token 🧠 While EAT is based on and compatible with CWT and JWT, it can also be described as:
- An identification and type system for claims in Claims-Sets
- Definitions of common attestation-oriented claims
- Claims defined in Concise Data Definition Language (CDDL) [RFC8610] and serialized using CBOR or JSON
- Security envelopes based on CBOR Object Signing and Encryption (COSE) and JSON Object Signing and Encryption (JOSE)
- The nesting of claims sets and tokens to represent complex and compound devices
- A profile mechanism for specifying and identifying specific tokens for specific use cases
EAT uses name/value pairs to identify individual claims, the same way CWT and JWT do 🏷️ Section 4 defines common attestation-oriented claims that have been added to the "CBOR Web Token (CWT) Claims" and "JSON Web Token Claims" IANA registries. As with CWT and JWT, no claims are mandatory, and claims that aren't recognized should be ignored 🤷♀️
Unlike (but still compatible with) CWT and JWT, EAT defines claims using CDDL [RFC8610]. In most cases, the same CDDL definition is used for both the CBOR/CWT serialization and the JSON/JWT serialization 🪄
Like CWT and JWT, EAT uses COSE and JOSE to provide authenticity, integrity, and, optionally, confidentiality 🔒 EAT doesn't put any new restrictions on cryptographic algorithms. It keeps all the cryptographic flexibility of CWT, COSE, JWT, and JOSE, which is very slay 💪
EAT defines a way to nest tokens and claims sets to accommodate composite devices that have multiple subsystems and multiple attesters 🪺 Tokens with security envelopes, or bare claims sets, may be embedded in an enclosing token. The nested token and the enclosing token don't have to use the same encoding (for example, a CWT may be enclosed in a JWT) 🎁
EAT also adds the ability to detach claims sets and send them separately from a security-enveloped EAT that contains a digest of the detached claims set 📬
This document registers no media or content types for identifying the EAT type, serialization encoding, or security envelope. The definition and registration of EAT media types are addressed in [EAT.media-types] 📝
Finally, this document introduces the notion of an EAT profile, which makes it easier to create narrowed definitions of EATs for specific use cases in later documents 🎯 One basic profile for constrained devices is normatively defined. Main character energy for constrained devices 👑
1.3. Operating Model and RATS Architecture
EAT follows the operational model described in Figure 1 of RATS Architecture (Section 3 of [RFC9334]) 🔄 To sum it up: an attester generates evidence, which is a claims set describing various characteristics of an entity. Evidence is usually signed by a key that proves the attester and the evidence it produces are authentic 🔏 The claims set either includes a received nonce or uses some other means to make sure it's fresh 🍃
A verifier confirms an EAT is valid by verifying the signature, and it may vet some claims using reference values 🕵️♀️ The verifier then produces attestation results, which may also be represented as an EAT. The attestation results go to the relying party, which is the ultimate consumer of the Remote Attestation Procedure 🎀 The relying party uses the attestation results however its use case needs, maybe letting an entity onto a network, approving a financial transaction, or something like that 💸 In some cases, the verifier and the relying party are not distinct entities. Two roles, one vibe 🫶
1.3.1. Relationship between Evidence and Attestation Results
Any claim defined in this document, or in the IANA "CBOR Web Token (CWT) Claims" or "JSON Web Token Claims" registries, may be used in evidence or attestation results 🗂️ The relationship between claims in attestation results and evidence is fundamentally governed by the verifier and the verifier's policy 📜
A common use case is for the verifier and its policy to run checks, calculations, and processing, with evidence as the input, to produce a summary result in the attestation results. That summary shows the overall health and status of the entity 🩺 For example, measurements in evidence may be compared to reference values, and the result of each comparison is represented as a simple pass/fail in the attestation results ✅❌
It's also possible that some claims in the evidence will be forwarded unmodified to the relying party in attestation results. This forwarding is subject to the verifier's implementation and policy. The relying party should be aware of the verifier's policy so it knows what checks have been performed on the claims it receives 👀
The verifier may modify claims it forwards, for example to implement a privacy preservation functionality 🫣 It's also possible the verifier will put claims in the attestation results that give details about the entity, which it has computed or looked up in a database. For example, the verifier may be able to put an oemid claim in the attestation results by performing a lookup based on a ueid claim (for example, a serial number) it received in evidence 🔍
This specification does not establish any normative rules for the verifier to follow, because those are a matter of local policy 🙌 It is up to each relying party to understand the processing rules of each verifier, so it knows how to interpret the claims in attestation results 🧚
2. Terminology ✨
okay bestie, let's get into the vocab, because if you mix up these words the whole spec gets messy 🫶
The BCP 14 key words here are "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL". They mean what BCP 14 [RFC2119] [RFC8174] says they mean, but ONLY when they show up in ALL CAPS like that 💅 Lowercase versions are just normal words.
The data structures are written in CDDL [RFC8610] [RFC9165] 🎀
The examples in Appendix A use CBOR diagnostic notation from Section 8 of [RFC8949] and Appendix G of [RFC8610] 🧚
This doc borrows a bunch of words from JWT [RFC7519] and CWT [RFC8392] 💜
base64url encoding: base64 encoding using the URL- and filename-safe character set from Section 5 of [RFC4648], with all trailing '=' characters dropped and no line breaks, whitespace, or other extra characters added [RFC7515] 🦋
Claim: a piece of info asserted about a subject. It's represented as a value plus either a name or a key so you can identify it 🌸
Claim Name: a unique text string that identifies the claim. It's the claim name used for JSON encoding 🍓
Claim Key: the CBOR map key used to identify a claim. (The term "Claim Key" comes from CWT. Like COSE [RFC9052], this doc uses the word "label" for CBOR map keys so nobody confuses them with cryptographic keys 🔐 no cap, that's a smart move 👑)
Claim Value: the value part of the claim. It can be any CBOR data item or JSON value 🌈
Claims Set: the CBOR map or JSON object that holds the claims carried by the CWT or JWT 💖
This doc also borrows terms from RATS Architecture [RFC9334]. Heads up: EAT doesn't capitalize RATS terms like "evidence" because it's easier to read 🫧
Attester: a role played by an entity (usually a device) whose evidence has to be appraised so we can figure out how trustworthy it is, like when deciding whether it's allowed to do some operation ⭐
Verifier: a role that appraises the validity of evidence about an attester and produces attestation results for a relying party to use 🔥
Relying Party: a role played by an entity that depends on the validity of info about an attester so it can reliably apply app-specific actions. Compare: relying party [RFC4949] 🎀
Evidence: a set of claims generated by an attester so a verifier can appraise it. Evidence may include configuration data, measurements, telemetry, or inferences 🌸
Attestation Results: the output a verifier generates, usually including info about an attester, where the verifier vouches for the validity of those results 💅
Reference Values: a set of values that claim values can be compared against as part of applying an appraisal policy for evidence. Other docs sometimes call these "known-good values", "golden measurements", or "nominal values". Those terms usually assume an equality check, but reference values can be used in any sort of comparison, so they're more general 🦋
Endorsement: a secure statement that an endorser vouches for the integrity of an attester's capabilities, like claims collection and evidence signing 🫶
This doc also reuses terms from CDDL [RFC8610]:
Group Socket: the mechanism by which a CDDL definition gets extended, as described in [RFC8610] and [RFC9165] ✨
3. Top-Level Token Definition 🎀
okay so this is where we define what an EAT actually is 💖
An "EAT" is an encoded (serialized) message whose job is to transfer a Claims-Set between two parties. An EAT MUST contain a Claims-Set. In this document, an EAT is always a CWT or JWT 🧚
An EAT MUST have authenticity and integrity protection. CWT and JWT provide that in this document 🔐 no skipping this part, bestie!
Further documents may define other encodings and security mechanisms for EAT 🌈
Identifying a protocol element as an EAT follows the general conventions used for CWTs and JWTs. How you identify it depends on the protocol carrying the EAT. Sometimes it's by media type (like in an HTTP Content-Type field). Other times it's through CBOR tags. There's no single fixed mechanism across all use cases 🦋
This doc also defines another message, the detached EAT bundle (see Section 5). It holds a collection of detached claims sets plus an EAT that gives integrity and authenticity protection for them. Detached EAT bundles can be CBOR or JSON encoded 💜
The following CDDL defines the top-level $CBOR-Tagged-Token, $EAT-CBOR-Untagged-Token, and $EAT-JSON-Token-Formats sockets (see Section 3.9 of [RFC8610]), so future token formats can be defined. Any new format that plugs into one or more of these sockets MUST be defined by an IETF Standards Action [RFC8126]. Of particular use may be a token type that provides no direct authenticity or integrity protection, for use with transport mechanisms that do provide the necessary security services [UCCS] 🌸
Nesting of EATs is allowed and defined in Section 4.2.18.3. This includes nesting an EAT in a different format than the enclosing EAT, i.e., the nested EAT may be CBOR-encoded while the enclosing EAT is JSON-encoded, or vice versa. The definition of Nested-Token references the CDDL defined in this section. When new token formats are defined, the means for identifying a nested token MUST also be defined 🪆
The top-level CDDL types are EAT-CBOR-Token for CBOR-encoded EATs and EAT-JSON-Token for JSON-encoded EATs. (CDDL and its tools support shared definitions of most items in this doc, but they don't support that sharing at the top level, which is why this is split up 🙈)
EAT-CBOR-Token = $CBOR-Tagged-Token / $EAT-CBOR-Untagged-Token
$CBOR-Tagged-Token /= CWT-Tagged-Message
$CBOR-Tagged-Token /= BUNDLE-Tagged-Message
$EAT-CBOR-Untagged-Token /= CWT-Untagged-Message
$EAT-CBOR-Untagged-Token /= BUNDLE-Untagged-Message
EAT-JSON-Token = $EAT-JSON-Token-Formats
$EAT-JSON-Token-Formats /= JWT-Message
$EAT-JSON-Token-Formats /= BUNDLE-Untagged-Message4. The Claims 👑
this section is where all the claims live, and they're giving main character energy 🌟
It describes new claims defined for attestation that have been added to the IANA "CBOR Web Token (CWT) Claims" [IANA.CWT.Claims] and "JSON Web Token Claims" [IANA.JWT.Claims] registries 📋
Everything from CWT and JWT (definitions, requirements, creation and validation procedures, security considerations, IANA registrations, and so on) carries over to EAT ✨
This section also describes how several existing CWT and JWT claims apply in EAT 💅
The set of claims an EAT must contain to be valid depends on context and is outside the scope of this spec. Specific applications of EATs will need implementations to understand and process some claims in particular ways. But if there are no such requirements, every claim that implementations don't understand MUST be ignored 🫧
CDDL plus a text description is used to define each claim, independent of encoding. Each claim is a CDDL group. In "Encoding and Collected CDDL" (Section 7), the CDDL groups turn into CBOR map entries and JSON name/value pairs 🦋
Each claim defined in this doc gets added to the $$Claims-Set-Claims group socket. Claims defined by other specs MUST also be added to that socket 🌸
All claims in an EAT MUST use the same encoding, except where it's explicitly stated otherwise (for example, in a CBOR-encoded token, all claims must be CBOR-encoded) 🔥
This spec gives a CDDL definition for most of the elements defined in [RFC7519] and [RFC8392]. Those definitions are in Appendix D and are NOT normative 💖
Each claim has a unique text string and integer that identifies it. CBOR-encoded tokens MUST only use the integer for claim keys. JSON-encoded tokens MUST only use the text string for claim names 🎀
4.1. eat_nonce (EAT Nonce) Claim 🌈
In JSON, an EAT nonce is either a text string or an array of text strings. In CBOR, an EAT nonce is either a byte string or an array of byte strings. The array option supports multistage EAT verification and consumption 🧚
A claim named "nonce" was defined for JWT and registered with IANA in the "JSON Web Token Claims" registry, but it MUST NOT be used, because it doesn't support multiple nonces 🙅♀️ No earlier "nonce" claim was defined for CWT. To tell it apart from the JWT "nonce" claim, this claim is named "eat_nonce" in JSON-encoded EATs. The CWT nonce defined here is meant for general-purpose use and keeps the "Nonce" claim name instead of an EAT-specific name 💜
An EAT nonce MUST have at least 64 bits of entropy. A maximum EAT nonce size is set to limit the memory an implementation needs. All receivers MUST be able to accommodate the maximum size 🔐
In CBOR, an EAT nonce is a byte string between 8 and 64 bytes long. In JSON, an EAT nonce is a text string between 8 and 88 bytes long 🍓
$$Claims-Set-Claims //=
(nonce-label => nonce-type / [ 2* nonce-type ])
nonce-type = JC< tstr .size (8..88), bstr .size (8..64)>4.2. Claims Describing the Entity 🦋
The claims in this section describe the entity itself. They describe the entity whether they show up in evidence or in attestation results. See Section 1.3.1 for how attestation results relate to evidence ✨
4.2.1. ueid (Universal Entity ID) Claim 💅
The "ueid" claim conveys a UEID, which identifies an individual manufactured entity, like a mobile phone, water meter, Bluetooth speaker, or networked security camera. It may identify the entire entity or a submodule. It does NOT identify types, models, or classes of entities. It's kind of like a serial number, though it doesn't have to be sequential 🌸
UEIDs MUST be universally and globally unique across manufacturers and countries, as described in Section 4.2.1.1. UEIDs MUST also be unique across protocols and systems, because tokens are meant to be embedded in lots of different protocols and systems. No two products anywhere, not even in totally different industries made by two different manufacturers in two different countries, should have the same UEID. (If they aren't global and universal like this, relying parties receiving them will have to track other characteristics of the entity to keep entities distinct between manufacturers 😭)
UEIDs aren't designed for direct human use (for example, printing on the case of a device), so no such representation is defined 🙈
There are privacy considerations for UEIDs. See Section 8.1 🫶
A Device Identifier (DevID) URN is registered for UEIDs. See Section 10.3 🌈
$$Claims-Set-Claims //= (ueid-label => ueid-type)
ueid-type = JC<base64-url-text .size (10..44) , bstr .size (7..33)>4.2.1.1. Rules for Creating UEIDs 🛠️
Okay so these rules are ONLY for whoever is making UEIDs. The EAT consumer doesn't need to know any of this, bestie 💅
A UEID is one type byte followed by the unique bytes for that type. That type byte is what keeps it globally unique, even if two different types happen to have the same unique bytes 🔐
UEIDs are variable length, so they can fit the types defined here and any future types too ✨
UEIDs SHOULD NOT be longer than 33 bytes. If they are longer, there's no guarantee a receiver can accept them. See Appendix B. 🙈
A UEID is permanent. It MUST NOT change for a given entity. Main character permanence 👑
The different types exist to accommodate different manufacturing processes, to allow small UEIDs, and to offer an option that needs no registration fees or central administration 🎀
If a brand new UEID type is ever needed (unlikely!), it MUST be defined in an update to this document on the Standards Track.
A manufacturer MAY use different types for different products. They MAY also switch from one type to another for a given product, or use one type for some items of a product and a different type for others. So flexible 🦋
+======+======+====================================================+
| Type | Type | Specification |
| Byte | Name | |
+======+======+====================================================+
| 0x01 | RAND | This is a 128-, 192-, or 256-bit random number |
| | | generated once and stored in the entity. This may |
| | | be constructed by concatenating enough identifiers |
| | | to make up an equivalent number of random bits and |
| | | then feeding the concatenation through a |
| | | cryptographic hash function. It may also be a |
| | | cryptographic quality random number generated once |
| | | at the beginning of the life of the entity and |
| | | stored. It MUST NOT be smaller than 128 bits. |
| | | See the length analysis in Appendix B. |
+------+------+----------------------------------------------------+
| 0x02 | IEEE | This makes use of the device identification scheme |
| | EUI | operated by the IEEE. An Extended Unique |
| | | Identifier (EUI) can be an EUI-48, EUI-60, or |
| | | EUI-64 and consists of two parts. The first part |
| | | is a registered company identifier, an |
| | | Organizationally Unique Identifier (OUI), an OUI- |
| | | 36, or a Company ID (CID). The second part |
| | | ensures uniqueness within the registered company. |
| | | EUIs are often the same as or similar to MAC |
| | | addresses. This type includes MAC-48, an obsolete |
| | | name for EUI-48. (Note that while entities with |
| | | multiple network interfaces may have multiple MAC |
| | | addresses, there is only one UEID for an entity; |
| | | changeable MAC addresses that do not meet the |
| | | permanence requirements in this document MUST NOT |
| | | be used for the UEID or Semipermanent UEID |
| | | (SUEID).) See [IEEE.802-2014] and [OUI.Guide]. |
+------+------+----------------------------------------------------+
| 0x03 | IMEI | This makes use of the International Mobile |
| | | Equipment Identity (IMEI) scheme operated by the |
| | | Global System for Mobile Communications |
| | | Association (GSMA). This is a 14-digit identifier |
| | | consisting of an 8-digit Type Allocation Code |
| | | (TAC) and a 6-digit serial number allocated by the |
| | | manufacturer, which SHALL be encoded as a byte |
| | | string of length 14 with each byte as the digit's |
| | | value (not the ASCII encoding of the digit; the |
| | | digit 3 encodes as 0x03, not 0x33). The IMEI |
| | | encoded value SHALL NOT include the Luhn checksum |
| | | or Software Version Number (SVN) information. See |
| | | [ThreeGPP.IMEI]. |
+------+------+----------------------------------------------------+Table 1: UEID Composition Types
4.2.1.2. Rules for Consuming UEIDs 🔍
For the consumer, a UEID is just a globally unique opaque identifier. The consumer doesn't need to know, and shouldn't need to know, the rules and structure used to make it globally unique.
All implementations MUST be able to receive UEIDs up to 33 bytes long. 33 bytes is the longest one defined in this document, and it gives the entropy needed for probabilistic uniqueness 💜
The consumer of a UEID MUST treat it as a completely opaque string of bytes and MUST NOT make any use of its internal structure. Here's why:
- UEID types vary freely from one manufacturer to the next.
- New types of UEIDs may be defined.
- The manufacturer of an entity is allowed to change from one type of UEID to another anytime they want.
For example, when the consumer gets a type 0x02 UEID, they should not use the OUI part to figure out the device's manufacturer, since there's no guarantee all UEIDs will be type 0x02. Different manufacturers may use different types. A manufacturer may make some of their product with one type and others with a different one, or even switch types for newer versions. Instead, the consumer should use the oemid claim. ✨
4.2.2. sueids (Semipermanent UEIDs) Claim 🌸
The sueids claim carries one or more semipermanent UEIDs (SUEIDs). An SUEID has the same format, characteristics, and requirements as a UEID, but it MAY change to a different value on entity life-cycle events. An entity MAY have both a UEID and SUEIDs, neither, or just one of the two.
Life-cycle events include things like a change of ownership, a factory reset, and onboarding into an IoT device management system. Specifying particular types of SUEIDs, and which life-cycle events trigger their change, is beyond this document's scope. An EAT profile MAY provide this specification.
There MAY be multiple SUEIDs. Each one has a text string label that tells it apart from the others. The label MAY name the purpose, application, or type of the SUEID. For example, the label for the SUEID used by the XYZ Onboarding Protocol could be "XYZ". Specifying any SUEID labeling schemes is also out of scope. They're use-case specific and MAY be specified in an EAT profile.
If there's only one SUEID, the claim stays a map and there still MUST be a label.
An SUEID gives you functionality similar to an IEEE Local Device Identifier (LDevID) [IEEE.802.1AR].
There are privacy considerations for SUEIDs; see Section 8.1.
A DevID URN is registered for SUEIDs; see Section 10.3.
$$Claims-Set-Claims //= (sueids-label => sueids-type)
sueids-type = {
+ tstr => ueid-type
}4.2.3. oemid (Hardware OEM ID) Claim 🏭
The oemid claim identifies the Original Equipment Manufacturer (OEM) of the hardware. Any of the three forms described below MAY be used, whichever one the claim sender likes best. The receiver MUST be able to handle all three forms 🫶
Heads up: the hwmodel claim in Section 4.2.4, the oemboot claim in Section 4.2.8, and the dbgstat claim in Section 4.2.9 all depend on this claim.
Sometimes one manufacturer acquires or merges with another. Depending on the situation and use case, newly made devices may keep the old OEM ID or switch to a new one. That's up to the manufacturers, but they should think about how it affects the claims above and the attestation ecosystem for their devices. These considerations are the same for all three forms of this claim.
4.2.3.1. Random Number-Based OEM ID 🎲
The random number-based OEM ID MUST be 16 bytes (128 bits) long.
The OEM can create its own ID using a cryptographic-quality random number generator. They only do this once in the life of the company, to generate the single ID for that company, and then use that same ID in every entity they make. This uniquely identifies the OEM on a statistical basis, and it's big enough even if there are ten billion companies 🌍
In JSON-encoded tokens, this MUST be base64url encoded.
4.2.3.2. IEEE-Based OEM ID 🧠
The IEEE runs a global registry for MAC addresses and company IDs, and this claim uses that database to identify OEMs. The contents may be an IEEE MA-L, MA-M, MA-S, or CID [IEEE-RA]. An MA-L (formerly known as an OUI) is a 24-bit value used as the first half of a MAC address. MA-M is similar: a 28-bit value used as the first part of a MAC address. MA-S (formerly known as OUI-36) is a 36-bit value. Lots of companies have already obtained an OEM ID from the IEEE registry. A CID is also a 24-bit value from the same space as an MA-L, but it is not for use as a MAC address. IEEE has published Guidelines for Use of EUI, OUI, and CID [OUI.Guide] and provides a lookup service [OUI.Lookup].
Companies that have more than one of these IDs or MAC address blocks SHOULD select one as preferred and use that for all their entities.
These are commonly written in "hexadecimal representation" as described in [IEEE.802-2014]. When this claim is encoded, the order of bytes in the bstr matches the order in the "hexadecimal representation". For example, an MA-L like "AC-DE-48" gets encoded in 3 bytes with values 0xAC, 0xDE, and 0x48.
This format is always 3 bytes in size in CBOR.
In JSON-encoded tokens, this MUST be base64url encoded and always 4 bytes.
4.2.3.3. IANA Private Enterprise Number-Based OEM ID 🎟️
IANA maintains a registry for Private Enterprise Numbers [PEN]. A PEN is an integer that identifies an enterprise, and it may be used to construct an object identifier (OID) relative to the following OID arc managed by IANA: iso(1) identified-organization(3) dod(6) internet(1) private(4) enterprise(1).
For EAT purposes, only the integer value IANA assigned as the PEN matters, not the full OID value.
In CBOR, this value MUST be encoded as a major type 0 integer and is typically 3 bytes. In JSON, this value MUST be encoded as a number.
$$Claims-Set-Claims //= (
oemid-label => oemid-type
)
oemid-type => oemid-pen / oemid-ieee / oemid-random
oemid-pen = int
oemid-ieee = JC<oemid-ieee-json, oemid-ieee-cbor>
oemid-ieee-cbor = bstr .size 3
oemid-ieee-json = base64-url-text .size 4
oemid-random = JC<oemid-random-json, oemid-random-cbor>
oemid-random-cbor = bstr .size 16
oemid-random-json = base64-url-text .size 244.2.4. hwmodel (Hardware Model) Claim
okay so the "hwmodel" claim is how you tell apart the hardware models, products, and variants that one OEM makes 🔧✨. That OEM is the one named by the OEM ID in Section 4.2.3. The hwmodel MUST be unique within a given OEM ID 💅. Stick the OEM ID and the hwmodel together and you get a global identifier for one specific product 🦋. The "hwmodel" claim MUST only be present if an "oemid" claim from Section 4.2.3 is also present 🫶.
how detailed the model ID gets is up to each OEM 🌸. It can be super granular, maybe even with version info baked in. Or it can be really general and only name the top-level products 🎀.
the "hwmodel" claim is for protocols, not for humans 🤖, so its format and encoding should not be human readable. that's how you discourage people from using it for anything else 🙈. if the claim is going to come from an identifier that already exists and is human readable, you can run it through a hash function 🔐.
there's no minimum length, so an OEM with only a few models can totally use a one-byte encoding 🍓. the maximum length is 32 bytes, and every receiver of this claim MUST be able to handle that full 32-byte size 💪.
the receiver MUST treat the hwmodel as a completely opaque string of bytes 🙅♀️, even if it looks like it has some kind of naming or structure. the OEM is free to rearrange the bytes inside however they like, as long as the claim still uniquely identifies each model ✨.
$$Claims-Set-Claims //= (
hardware-model-label => hardware-model-type
)
hardware-model-type = JC<base64-url-text .size (4..44),
bytes .size (1..32)>4.2.5. hwversion (Hardware Version) Claim
the "hwversion" claim is a text string, and each manufacturer gets to pick its format 📝. If you want to spell out the structure and the sorting order, you can use the version-scheme item from Concise Software Identification (CoSWID) [RFC9393] 🧩. Knowing how to sort versions is important, because it's how you can tell the newer ones from the older ones 👀. A "hwversion" claim MUST only be present if a "hwmodel" claim, as described in Section 4.2.4, is present 💅.
$$Claims-Set-Claims //= (
hardware-version-label => hardware-version-type
)
hardware-version-type = [
version: tstr,
? scheme: $version-scheme
]4.2.6. swname (Software Name) Claim
the "swname" claim is a super simple, free-form text value that names the software running on the entity 🎀. We intentionally didn't set any general rules or structure for it, so it's not the right tool if you need precise naming 🙈. if you do need strict, rigorous naming for the entity's software, use the "manifests" claim from Section 4.2.15 instead 🫶.
$$Claims-Set-Claims //= ( sw-name-label => tstr )4.2.7. swversion (Software Version) Claim
the "swversion" claim uses the CoSWID version-scheme defined in [RFC9393] to give a simple version for the software 🔢✨. A "swversion" claim MUST only be present if a "swname" claim, as described in Section 4.2.6, is present 💖. if a simple version is too basic for you, the "manifests" claim (Section 4.2.15) is there as an alternative 🧚.
$$Claims-Set-Claims //= (sw-version-label => sw-version-type)
sw-version-type = [
version: tstr
? scheme: $version-scheme
]4.2.8. oemboot (OEM Authorized Boot) Claim
if the "oemboot" claim is "true", it means the entity booted with software that the manufacturer authorized, as shown by the "oemid" claim in Section 4.2.3 👑. The firmware and operating system are fully under the OEM's control 🔒. Not even the end user, or the enterprise that owns the device, can swap them out 😭. The control can come from cryptographic authentication of the software, from the software living in Read-Only Memory (ROM), from a mix of both, or from something else entirely 🌈. If this claim is present, the "oemid" claim MUST be present too 💅.
$$Claims-Set-Claims //= (oem-boot-label => bool)4.2.9. dbgstat (Debug Status) Claim
the "dbgstat" claim is about the entity-wide or submodule-wide debug stuff, like [JTAG] and the diagnostic hardware built into chips 🔍. It also covers any software debug facilities tied to privileged software that can do system-wide memory inspection, tracing, or modification of non-system software, like user-mode apps 👀.
this assumes debug facilities can be turned on and off dynamically, or can be turned off permanently so that nobody can ever turn them back on 🔐. an example of dynamic enabling is when you need some authentication before debugging can be enabled 🗝️. an example of permanent disabling is blowing a hardware fuse in a chip 💥. the exact type of mechanism doesn't matter 🤷♀️. it doesn't matter if the authentication is one global password or per-entity public keys 🙌.
like with every claim, if the "dbgstat" claim is missing, it just means it isn't reported 🫠.
this claim isn't extensible, because it's not meant to be a fully interoperable way to describe debug status 🚫. if an implementation thinks this claim isn't good enough, it can define its own proprietary claim instead 🧁. it can even include both: this claim as a coarse indication, and its own proprietary claim as a refined one 💜.
the higher debug-disabling levels need all the debug disabling from the levels below them to be in effect 📶. the lowest level says all of the target's debug is currently disabled, so every other level needs that too 💯.
claims do NOT get inherited from a submodule up to a superior module, or the other way around 🙅♀️. there's also no assumption, requirement, or guarantee that a superior module's target covers its submodules' targets 🫥. so every submodule has to explicitly describe its own debug state 📝. the receiver of an EAT MUST NOT assume that debug is turned off in a submodule just because a superior module has a claim saying it's off 🚨.
an entity can have multiple debug facilities, and the plural "states" in these descriptions is talking about that 🦋. it has nothing to do with aggregation or inheritance 🙃.
some chips or devices are built so that one debug facility covers the whole chip or device 🧠. if the EAT for a chip like that includes submodules, each submodule should independently report the status of that whole-chip or whole-device debug facility 🎀. that's the only way the receiver can figure out the debug status of the submodules, since nothing gets inherited 😤.
4.2.9.1. Enabled
if any debug facility is currently enabled, even manufacturer hardware diagnostics, then this level must be indicated 🔥.
4.2.9.2. Disabled
this level means all debug facilities are currently disabled 💤. they might be turned on again in the future 🔮. it could also be that they were enabled in the past but are off right now 🌙.
4.2.9.3. Disabled Since Boot
this level means all debug facilities are currently disabled and have been the whole time since the entity booted or started 🥾✨.
4.2.9.4. Disabled Permanently
this level means all the non-manufacturer facilities are permanently disabled, so no end user or developer can turn them on 🔒. only the manufacturer named in the "oemid" claim can enable them 👑. it also means all debug facilities are currently disabled and have been since boot or start 💖. if this debug state is reported, the "oemid" claim MUST be present 🫶.
4.2.9.5. Disabled Fully and Permanently
this level means that all debug facilities for the entity are permanently disabled 💀✨.
$$Claims-Set-Claims //= ( debug-status-label => debug-status-type )
debug-status-type = ds-enabled /
disabled /
disabled-since-boot /
disabled-permanently /
disabled-fully-and-permanently
ds-enabled = JC< "enabled", 0 >
disabled = JC< "disabled", 1 >
disabled-since-boot = JC< "disabled-since-boot", 2 >
disabled-permanently = JC< "disabled-permanently", 3 >
disabled-fully-and-permanently =
JC< "disabled-fully-and-permanently", 4 >4.2.10. location (Location) Claim
the "location" claim gives the geographic position of the entity where the attestation came from 🌍🗺️. latitude, longitude, altitude, accuracy, altitude-accuracy, heading, and speed MUST be defined the way the W3C Geolocation API [W3C.GeoLoc] defines them (and that API is based on [WGS84]) 📍. if the entity is stationary, the heading is 'null' 🧭. latitude and longitude MUST be provided 💖. if any of the other values are unknown, you just leave them out 🫶.
the location might have been cached for a while before the token got created ⏳. it could have been minutes, hours, or even longer since the last contact with a satellite in the Global Navigation Satellite System (GNSS) 🛰️. you can use either the timestamp or the age data item to show how long it's been cached ⏱️. the timestamp is the one to pick, though, because it's a non-relative time ⭐. if the entity doesn't have a clock, or the clock isn't set but it can measure the time between acquiring the location and creating the token, it can report the age instead 🧚. the age is in seconds ✨.
see the location-related privacy considerations in Section 8.2 🙈🔐.
$$Claims-Set-Claims //= (location-label => location-type)
location-type = {
latitude => number,
longitude => number,
? altitude => number,
? accuracy => number,
? altitude-accuracy => number,
? heading => number / null,
? speed => number,
? timestamp => ~time-int,
? age => uint
}
latitude = JC< "latitude", 1 >
longitude = JC< "longitude", 2 >
altitude = JC< "altitude", 3 >
accuracy = JC< "accuracy", 4 >
altitude-accuracy = JC< "altitude-accuracy", 5 >
heading = JC< "heading", 6 >
speed = JC< "speed", 7 >
timestamp = JC< "timestamp", 8 >
age = JC< "age", 9 >4.2.11. uptime (Uptime) Claim
the "uptime" claim holds the number of seconds that have passed since the entity or submodule was last booted ⏰☀️.
$$Claims-Set-Claims //= (uptime-label => uint)4.2.12. bootcount (Boot Count) Claim
the "bootcount" claim is a count of how many times the entity or submodule has been booted 🔁. to support this claim, the device needs persistent storage 💾✨.
$$Claims-Set-Claims //= (boot-count-label => uint)4.2.13. bootseed (Boot Seed) Claim
the "bootseed" claim is a value created at system boot time 🌱. it lets you tell attestation reports from different boot sessions of a particular entity apart (for example, a certain UEID) 🦋.
this value is usually public 📢. it is NOT a secret, so it MUST NOT be used for anything that needs a secret seed, like seeding a random number generator 🚫🎲. there are privacy considerations for this claim too, so check Section 8.3 🫶🔐.
$$Claims-Set-Claims //= (boot-seed-label => binary-data)4.2.14. dloas (Digital Letters of Approval) Claim ✨
Okay bestie, the "dloas" claim carries one or more Digital Letters of Approval (DLOAs) 💌. A DLOA [DLOA] is a document that describes a certification an entity has received. Examples of certifications a DLOA can represent include ones issued by GlobalPlatform [GP-Example] and ones based on Common Criteria [CC-Example]. A DLOA isn't tied to any one certification type or to any particular organization 🌸.
This claim is usually issued by a verifier, not an attester 💅. Verifiers MUST NOT issue this claim unless the entity has received the certification the DLOA indicates.
This claim MAY contain more than one DLOA. If there are multiple DLOAs, verifiers MUST NOT issue this claim unless the entity has received all of the certifications 🎀.
DLOA documents are always fetched from a registrar that stores them. The claim has a few data items that get used to build a Uniform Resource Locator (URL) for grabbing the DLOA from that particular registrar 🦋.
This claim MUST be encoded as an array with either two or three elements. The first element MUST be the URL for the registrar. The second element MUST be a platform label saying which platform was certified. If the DLOA applies to an application, the third element gets added, and it MUST be an application label. How to build the registrar URL, the platform label, and maybe the application label is specified in [DLOA].
The retriever of a DLOA MUST follow the recommendation in [DLOA] and use Transport Layer Security (TLS) or some other means to be sure the DLOA registrar they are accessing is authentic 🔐. The platform and application labels in the claim indicate the correct DLOA for the entity.
$$Claims-Set-Claims //= (
dloas-label => [ + dloa-type ]
)
dloa-type = [
dloa_registrar: general-uri
dloa_platform_label: text
? dloa_application_label: text
]4.2.15. manifests (Software Manifests) Claim 📦
The "manifests" claim contains descriptions of software on the entity. These manifests are installed on the entity when the software is installed, or they're created as part of the installation process. Installation is anything that adds software to the entity: possibly factory installation, the user installing elective applications, and so on. The defining trait of a manifest is that the software manufacturer created it. The whole point of this claim is to pass unmodified manifests along to the verifier and possibly to the relying party 🧚.
Some manifests are signed by their software manufacturer independently, and some aren't, either because they don't support signing or because the manufacturer chose not to sign them. For example, a CoSWID might be signed independently before it's included in an EAT. When signed manifests go into an EAT, the manufacturer's signature SHOULD be included, even though the EAT's signature also covers the manifest. That way the receiver can directly verify the manufacturer-originated manifest 🔍.
This claim allows multiple manifest formats. For example, the manifest could be a CBOR-encoded CoSWID, an XML-encoded Software Identification (SWID) tag, or something else entirely. The type of manifest is always identified by a Constrained Application Protocol (CoAP) Content-Format identifier [RFC7252]. If there's no CoAP identifier registered for a manifest format, one MUST be registered.
This claim MUST be an array of one or more manifests. Each manifest in the claim MUST be an array of two. The first item in that array of two MUST be a CoAP Content-Format identifier. The second item is MUST be the actual manifest.
In JSON-encoded tokens, the manifest, whatever its encoding, MUST be placed in a text string. When a non-text encoded manifest, such as a CBOR-encoded CoSWID, goes into a JSON-encoded token, the manifest MUST be base64 encoded 🍓.
This claim allows multiple manifests in one token, since multiple software packages are likely to be present. The multiple manifests MAY be of different encodings. In some cases, EAT submodules may be used instead of the array structure in this claim for multiple manifests ✨.
A CoSWID manifest MUST be a payload CoSWID, not an evidence CoSWID. These are defined in [RFC9393].
This claim is extensible, so it can be used for manifest formats beyond the ones mentioned in this document. No particular manifest format is preferred. For manifest interoperability, an EAT profile, as defined in Section 6, should be used to specify which manifest format(s) is allowed 📜.
$$Claims-Set-Claims //= (
manifests-label => manifests-type
)
manifests-type = [+ manifest-format]
manifest-format = [
content-type: coap-content-format,
content-format: JC< $manifest-body-json,
$manifest-body-cbor >
]
$manifest-body-cbor /= bytes .cbor untagged-coswid
$manifest-body-json /= base64-url-text4.2.16. measurements (Measurements) Claim 📏
The "measurements" claim contains descriptions, lists, evidence, or measurements of the software that exists on the entity, or on any other measurable subsystem of the entity (for example, a hash of sections of a file system or non-volatile memory). The defining trait of this claim is that its contents are created by processes on the entity that inventory, measure, or otherwise characterize the software on the entity. The contents don't come from the manufacturer of the measurable subsystem (for example, the developer of a software library) 🦋.
This claim can be a CoSWID [RFC9393]. When the CoSWID format is used, it MUST be an evidence CoSWID, not a payload CoSWID.
Formats other than CoSWID MAY be used. The format is identified by a CoAP Content-Format identifier, which is the same one used for the "manifests" claim in Section 4.2.15 💜.
$$Claims-Set-Claims //= (
measurements-label => measurements-type
)
measurements-type = [+ measurements-format]
measurements-format = [
content-type: coap-content-format,
content-format: JC< $measurements-body-json,
$measurements-body-cbor >
]
$measurements-body-cbor /= bytes .cbor untagged-coswid
$measurements-body-json /= base64-url-text4.2.17. measres (Software Measurement Results) Claim 🔥
The "measres" claim is a general-purpose structure for reporting how measurements compare to expected reference values. It gives you a simple, standard way to report a comparison as a success, a failure, not run, or absent 🎯.
Here's the tea: measurement systems are super specific to the operating system, software, and hardware of the entity being measured. It's not possible to standardize what gets measured and how across every platform, OS, software, and hardware combo 🙈. The recipient has to get the details about what was measured, and what it means for the entity's security, from whoever provided the measurement system. What this claim does provide is a standard way to report the basic success or failure of a measurement. In some use cases, it's valuable to know in a general way whether measurements succeeded or failed, even if the details of what was measured aren't characterized.
This claim MAY be generated by the verifier and sent to the relying party. For example, it could hold the results of the verifier comparing the contents of the "measurements" claim (Section 4.2.16) to reference values.
This claim MAY also be generated on the entity, if the entity can have one subsystem measure and evaluate another subsystem. For example, a TEE might be able to measure the software of the rich OS and might have the reference values for the rich OS 👑.
Within an entity, attestation target, or submodule, multiple results can be reported. For example, it may be desirable to report results for measurements of the file system, chip configuration, installed software, running software, and so on 🌈.
Note that this claim is not for reporting the overall result of a verifier. It's solely for reporting the result of a comparison to reference values ⚠️.
An individual measurement result (individual-result) is an array of two elements: an identifier of the measurement (result-id), and an enumerated type of the result (result). Different measurement systems will measure different things, and maybe measure the same thing in different ways. It's up to each measurement system to define the identifiers (result-id) for the measurements it reports 💖.
Each individual measurement result is part of a group that may contain many individual results. Each group has a text string that names it, typically the name of the measurement scheme or system.
The claim itself consists of one or more groups.
The values for the results enumerated type are as follows:
1 -- comparison success: The comparison to reference values was
successful.
2 -- comparison failure: The comparison was completed but did not
compare correctly to the reference values.
3 -- comparison not run: The comparison was not run. This includes
error conditions such as running out of memory.
4 -- measurement absent: The particular measurement was not
available for comparison.$$Claims-Set-Claims //= (
measurement-results-label =>
[ + measurement-results-group ] )
measurement-results-group = [
measurement-system: tstr,
measurement-results: [ + individual-result ]
]
individual-result = [
result-id: tstr / binary-data,
result: result-type,
]
result-type = comparison-success /
comparison-fail /
comparison-not-run /
measurement-absent
comparison-success = JC< "success", 1 >
comparison-fail = JC< "fail", 2 >
comparison-not-run = JC< "not-run", 3 >
measurement-absent = JC< "absent", 4 >4.2.18. submods (Submodules) Claim ✨
okay bestie, some devices are literally so complicated 😭 a mobile phone is the perfect example. it's got subsystems for communications (Wi-Fi and cellular 📶), low-power audio and video playback 🎵, and multiple security-oriented subsystems like a TEE and a Secure Element 🔐. the claims for a subsystem can be grouped together in a submodule 🧩 so it's giving super organized.
submodules may be used in either evidence or attestation results 📜.
since system architecture varies wildly from use case to use case, there are no set requirements for what a submodule represents, either in evidence or in attestation results 🤷♀️. profiles (Section 6) may wish to impose requirements 💅. an attester that outputs evidence with submodules should document the semantics it associates with particular submodules for the verifier 📝. likewise, a verifier that outputs attestation results with submodules should document the semantics it associates with the submodules for the relying party 🫶.
a submodule claim is a map that holds some number of submodules 🗺️. each submodule is named by its label in the submodule claim map. the value of each entry in a submodule may be a Claims-Set, a nested token, or a Detached-Submodule-Digest 🎀. this allows the submodule to serve as its own attester or not, and allows its claims to be represented directly or indirectly (aka detached) 🦋.
a submodule may include a submodule, so you get arbitrary levels of nesting 🪆. however, submodules do not inherit anything from the containing token and must explicitly include all claims 🙈. submodules may contain claims that are present in any surrounding token or submodule. for example, the top level of the token may have a UEID, a submodule may have a different UEID, and a further subordinate submodule may also have a UEID 👑.
the following subsections define the three types for representing submodules:
- a submodule Claims-Set
- the digest of a detached Claims-Set
- a nested token, which can be any EAT
the Submodule type and Nested-Token type definitions change depending on the type of encoding. for CBOR-encoded EATs, they look like this:
Nested-Token = CBOR-Nested-Token
CBOR-Nested-Token =
JSON-Token-Inside-CBOR-Token /
CBOR-Token-Inside-CBOR-Token
CBOR-Token-Inside-CBOR-Token = bstr .cbor $CBOR-Tagged-Token
JSON-Token-Inside-CBOR-Token = tstr
$$Claims-Set-Claims //= (submods-label => { + text => Submodule })
Submodule = Claims-Set / CBOR-Nested-Token /
Detached-Submodule-Digestthe Submodule and Nested-Token definitions for JSON-encoded EATs are below. these are necessarily different from CBOR because JSON has no tag mechanism and no byte-string type to help indicate that the nested token is CBOR 🙃
Nested-Token = JSON-Selector
JSON-Selector = $JSON-Selector
$JSON-Selector /= [type: "JWT", nested-token: JWT-Message]
$JSON-Selector /= [type: "CBOR", nested-token:
CBOR-Token-Inside-JSON-Token]
$JSON-Selector /= [type: "BUNDLE", nested-token: Detached-EAT-Bundle]
$JSON-Selector /= [type: "DIGEST", nested-token:
Detached-Submodule-Digest]
CBOR-Token-Inside-JSON-Token = base64-url-text
$$Claims-Set-Claims //= (submods-label => { + text => Submodule })
Submodule = Claims-Set / JSON-Selectorthe Detached-Submodule-Digest type is defined like this:
Detached-Submodule-Digest = [
hash-algorithm : text / int,
digest : binary-data
]nested tokens can be one of three types defined in this document, or types standardized in later documents (like [UCCS]) 🌈. nested tokens are the only mechanism by which JSON can be embedded in CBOR and vice versa, so they're kind of a big deal 💜.
adding more types is done by augmenting the $CBOR-Tagged-Token socket or the $JSON-Selector socket ✨.
when decoding a JSON-encoded EAT, the type of submodule works like this. a JSON object means the submodule is a Claims-Set. in all other cases it's a JSON-Selector, which is an array of two elements that tells you whether the submodule is a nested token or a Detached-Submodule-Digest. the first element names the type present in the second element. if the value is "JWT", "CBOR", "BUNDLE", or a future-standardized token type (e.g., see [UCCS]), the submodule is a nested token of that indicated type: JWT-Message, CBOR-Token-Inside-JSON-Token, Detached-EAT-Bundle, or a future type. if the value is "DIGEST", the submodule is a Detached-Submodule-Digest. any other value indicates a standardized extension to this specification 🧠.
when decoding a CBOR-encoded EAT, the CBOR item type tells you the type of the submodule. a map means a CBOR-encoded submodule Claims-Set. an array means a CBOR-encoded Detached-Submodule-Digest. a byte string means a CBOR-encoded CBOR-Nested-Token. a text string means a JSON-encoded JSON-Selector. where JSON-Selector is used in a CBOR-encoded EAT, the "DIGEST" type and corresponding Detached-Submodule-Digest type MUST NOT be used 🚫.
the type of a CBOR-encoded nested token is always determined by the CBOR tag encountered after the byte string wrapping is removed, in a CBOR-encoded enclosing token, or after the base64 wrapping is removed, in a JSON-encoded enclosing token 🎁.
the type of a JSON-encoded nested token is always determined by the string name in JSON-Selector, and it's always "JWT", "BUNDLE", or a new name standardized outside this document for a further type (e.g., "UCCS"). this string name may also be "CBOR" to indicate the nested token is CBOR encoded 🍓.
"JWT": The second array item MUST be a JWT formatted according to [RFC7519]."CBOR": The second array item MUST be some base64url-encoded CBOR that is a tag, typically a CWT or CBOR-encoded detached EAT bundle."BUNDLE": The second array item MUST be a JSON-encoded detached EAT bundle as defined in this document."DIGEST": The second array item MUST be a JSON-encoded Detached-Submodule-Digest as defined in this document.
as noted elsewhere, additional EAT types may be defined by a Standards Action. new type specifications MUST address the integration of the new type into the submodule claim type for submodules 🧩.
4.2.18.1. Submodule Claims-Set
the Claims-Set type is a way to represent claims from a submodule that doesn't have its own attesting environment, i.e., it has no keys distinct from the attester producing the surrounding token 🔑. claims are represented as a Claims-Set. submodule claims represented this way are secured by the same mechanism as the enclosing token (e.g., they're signed by the same attestation key) ✍️.
the encoding of a submodule Claims-Set MUST be the same as the encoding of the surrounding EAT, e.g., all submodule Claims-Sets in a CBOR-encoded token must be CBOR encoded 🔁.
4.2.18.2. Detached Submodule Digest
the Detached-Submodule-Digest type is similar to a submodule Claims-Set, except a digest of the Claims-Set is included in the claim while the Claims-Set contents get conveyed separately 📦. the separately conveyed Claims-Set is called a "detached claims set". the input to the digest algorithm is the CBOR or the JSON-encoded Claims-Set for the submodule. there's no byte string wrapping or base64 encoding, so keep it clean ✨.
the data type for this kind of submodule is an array with two data items: an algorithm identifier and a byte string containing the digest. the hash algorithm identifier always comes from the "COSE Algorithms" registry [IANA.COSE.Algorithms]. either the integer or string identifier may be used. the hash algorithm identifier is never from any other algorithm registry 🚫.
a detached EAT bundle, as described in Section 5, may be used to convey detached claims sets and the EAT containing the corresponding detached digests. however, EAT does not require the use of a detached EAT bundle 🙅♀️. any other protocols may be used to convey detached claims sets and the EAT containing the corresponding detached digests. if detached Claims-Sets are modified in transit, then validation can fail 💀.
4.2.18.3. Nested Tokens
the CBOR-Nested-Token and JSON-Selector types represent claims from a submodule that has its own attesting environment, i.e., it has keys distinct from the attester producing the surrounding token 🔐. claims are represented in a signed EAT token.
including a signed EAT as a claim cryptographically binds the EAT to the surrounding token. if it were conveyed in parallel with the surrounding token, there would be no such binding, and attackers could swap in a good attestation from another device for the attestation of an errant subsystem 😱 this is the scary part, bestie.
a nested token doesn't need to use the same encoding as the enclosing token. this lets composite devices get built without worrying about which encoding each component uses. so a CBOR-encoded EAT can have a JSON-encoded EAT as a nested token, and vice versa 🧩.
4.3. Claims Describing the Token
the claims in this section give metadata about the token they occur in 📜. they don't describe the entity. they may appear in evidence or attestation results.
4.3.1. iat (Timestamp) Claim
the "iat" claim defined in CWT and JWT is used to indicate the date-of-creation of the token, which is the time the claims are collected and the token is composed and signed ⏰.
the data for some claims may be held or cached for some period of time before the token is created. this period may be long, even days 🕰️. examples are measurements taken at boot, or a geographic position fix taken the last time a satellite signal was received 🛰️. there are individual timestamps associated with these claims to indicate that their age is older than the "iat" timestamp.
CWT allows floating-point for this claim, whereas EAT disallows floating-point. an EAT token MUST NOT contain an "iat" claim in floating-point format. any recipient of a token with a floating-point format "iat" claim MUST consider it an error 🚨.
a 64-bit integer representation of the CBOR epoch-based time [RFC8949] used by this claim can represent a range of +/- 500 billion years 🌌. so the only point of a floating-point timestamp is to get precision finer than one second 🤓. this is not needed for EAT.
4.3.2. eat_profile (EAT Profile) Claim
see Section 6 for the detailed description of an EAT profile 👀.
the "eat_profile" claim identifies an EAT profile by either a Uniform Resource Identifier (URI) or an OID. typically the URI will reference a document describing the profile. an OID is just a unique identifier for the profile. it may exist anywhere in the OID tree 🌳. there is no requirement that the named document be publicly accessible. the primary purpose of the "eat_profile" claim is to uniquely identify the profile, even if it's a private profile 🤫.
the OID is always absolute and never relative 🎯.
see Section 7.2.1 for OID and URI encoding 🔤.
$$Claims-Set-Claims //= (profile-label => general-uri / general-oid)4.3.3. intuse (Intended Use) Claim 🎀
okay bestie, EATs can show up in lots of different apps 🦋 and the "intuse" claim is how a token tells its consumer what it's meant for. an app can use it to tell its own flavours of EAT apart, which is giving organised 💅 The possible values live in the "Entity Attestation Token (EAT) Intended Uses" registry defined in Section 10.5 ✨
$$Claims-Set-Claims //= ( intended-use-label => intended-use-type )
intended-use-type = JC< text, int>5. Detached EAT Bundles 📦
okay so a detached EAT bundle is a message that carries an EAT plus detached claims sets secured by that EAT 💖 It's a top-level message, just like a CWT or JWT. It can show up anywhere a CWT or JWT can, like as a submodule nested token (see Section 4.2.18.3) 🫶
A detached EAT bundle may be either CBOR or JSON encoded 🌈
It has two parts, bestie ✨
The first part is an encoded EAT that:
- MUST have at least one submodule that is a detached submodule digest as defined in Section 4.2.18.2
- MAY be either CBOR or JSON encoded and does not have to be the same as the encoding of the bundle
- MAY be a CWT, JWT, or some future-defined token type, but it MUST NOT be a detached EAT bundle
- MUST be authenticity and integrity protected
The same mechanism for telling apart the type of nested token submodules is used here too 🔥
The second part is a map/object that:
- MUST be a Claims-Set
- MUST use the same encoding as the bundle
- MUST be wrapped in a byte string when the encoding is CBOR and be base64url encoded when the encoding is JSON
For a CBOR-encoded detached EAT bundle, tag 602 can be used to identify it 🍓 The standard rules apply for whether you use a tag or not. When it's sent as a submodule, it's always sent as a tag so it doesn't get mixed up with the other kinds of nested tokens 💅
The digests of the detached claims sets are linked to the detached Claims-Sets by label/name. It's up to whoever builds the detached EAT bundle to make sure the names uniquely identify the detached claims sets. Since the names are only used inside the detached EAT bundle, they can be super short, maybe one byte 🧚
BUNDLE-Messages = BUNDLE-Tagged-Message / BUNDLE-Untagged-Message
BUNDLE-Tagged-Message = #6.602(BUNDLE-Untagged-Message)
BUNDLE-Untagged-Message = Detached-EAT-Bundle
Detached-EAT-Bundle = [
main-token : Nested-Token,
detached-claims-sets: {
+ tstr => JC<json-wrapped-claims-set,
cbor-wrapped-claims-set>
}
]
json-wrapped-claims-set = base64-url-text
cbor-wrapped-claims-set = bstr .cbor Claims-Set6. Profiles 👑
EAT makes normative use of CBOR, JSON, COSE, JOSE, CWT, and JWT 🌸 Most of these have a bunch of implementation options so they can cover a range of use cases, which is very flexible but also very much a lot 😭
For example, COSE doesn't require a particular set of cryptographic algorithms. That lets it handle different usage scenarios and algorithms changing over time 🦋 Section 10 of [RFC9052] describes the profiling considerations for COSE.
Encryption is optional for both CWT and JWT. Section 8 of [RFC7519] describes implementation requirements and recommendations for JWT 💜
CBOR also has indefinite-length encoding. It isn't commonly used, but it's valuable for very constrained devices 🧚 For EAT itself, some claims get used in one particular use case and others don't. Section 4 of [RFC8949] describes serialization considerations for CBOR ✨
For example, a mobile phone use case might need the device make and model, and might ban UEID and location for privacy reasons 🙈 The general EAT standard keeps all that flexibility, because it's also built to cover a broad range of use cases.
Here's the thing though: you have to narrow these implementation options down to guarantee interoperability 💖 EAT picks one general, explicit mechanism for this, called the profile. A profile indicates the choices made for these implementation options for all aspects of the token ⭐
Below is a list of the issues a profile should address 🫶
The "eat_profile" claim in Section 4.3.2 gives a unique identifier for the profile a particular token uses 🎀
A profile can apply to evidence, attestation results, or both 🌈
6.1. Format of a Profile Document 📝
A profile document doesn't have to be in any particular format. It can be simple text, something more formal, or a mix of both ✨
A profile may define, and possibly register, one or more new claims if needed. A profile may also reuse one or more already defined claims, either as is or with values constrained to a subset or subrange 💅
6.2. Full and Partial Profiles 🦋
For a "full" profile, the receiver will be able to decode and verify every possible EAT sent when a sender and receiver both stick to it. For a "partial" profile, some protocol options are still left undecided 🤷♀️
For example, a profile that lets the sender use signing algorithms the receiver isn't required to support is a partial profile. The sender might pick a signing algorithm that some receivers don't support 😬
Full profiles MUST be complete, so that a complying receiver can decode, verify, and check for freshness for every EAT created by a complying sender 👑 Full profiles don't need to require the receiver to fully handle every claim in an EAT from a complying sender. Profile specifications may assume the receiver has access to the necessary verification keys, or they may go into specific detail on how to access verification keys 🔑
The "eat_profile" claim MUST NOT be used to identify partial profiles 🙅♀️
While fewer profiles are preferable, sometimes a use case needs several 🌸 One way to handle variation in devices is to define several full profiles that are variants of each other. Defining profiles is relatively easy and cheap, since they don't have to be published on the Standards Track or registered anywhere 🍓 For example, flexibility for post-quantum algorithms can be handled like this: first, define a full profile for a set of non-post-quantum algorithms for current use. Then, once post-quantum algorithms are settled, define another full profile derived from the first 🔮
6.3. List of Profile Issues 📋
The following is a list of EAT, CWT, JWT, COSE, JOSE, and CBOR options that a profile should address 💜
6.3.1. Use of JSON, CBOR, or Both 🧾
A profile should specify whether CBOR, JSON, or both may be sent. A profile should specify that the receiver can accept all encodings the sender is allowed to send 🫶
This should be specified for the top level and all nested tokens. For example, a profile might require all nested tokens to use the same encoding as the top-level token 🎀
6.3.2. CBOR Map and Array Encoding 🧺
A profile should specify whether definite-length arrays/maps, indefinite-length arrays/maps, or both may be sent. A profile should specify that the receiver accepts all length encodings the sender is allowed to send 🌈
This applies to individual EAT claims, CWT, and COSE parts of the implementation.
For most use cases, specifying that only definite-length arrays/maps may be sent is suitable 💅
6.3.3. CBOR String Encoding 🧵
A profile should specify whether definite-length strings, indefinite-length strings, or both may be sent. A profile should specify that the receiver accepts all types of string encoding the sender is allowed to send 🦋
For most use cases, specifying that only definite-length strings may be sent is suitable ✨
6.3.4. CBOR Preferred Serialization 🌟
A profile should specify whether or not CBOR preferred serialization must be sent. A profile should specify that the receiver accepts preferred and/or non-preferred serialization, so it can accept anything the sender sends 🙌
6.3.5. CBOR Tags 🏷️
The profile should specify whether the token should be a CWT tag or not 🎀
When COSE protection is used, the profile should specify whether COSE tags are used or not. Note that [RFC8392] requires COSE tags be used in a CWT tag 🧠
Often a tag is unnecessary, because the surrounding or carrying protocol already identifies the object as an EAT 💖
6.3.6. COSE/JOSE Protection 🔐
COSE and JOSE have several options for signed, MACed, and encrypted messages. It may be an Unsecured JWT as described in Section 6 of [RFC7519]. It's possible to have no protection, sign only, MAC only, sign then encrypt, and so on. All combinations allowed by COSE, JOSE, JWT, and CWT are allowed by EAT 🌸
A profile should specify all signing, encryption, and MAC message formats that may be sent. For example, a profile might allow only COSE_Sign1 to be sent. As another example, a profile might allow COSE_Sign and COSE_Encrypt to be sent, to carry multiple signatures for post-quantum cryptography and to use encryption for confidentiality 🔥
A profile should specify that the receiver accepts all message formats that are allowed to be sent 🫶
When both signing and encryption are allowed, a profile should specify which one is applied first 🤔
6.3.7. COSE/JOSE Algorithms 🧮
See "Application Profiling Considerations" (Section 10 of [RFC9052]) for a discussion of selecting cryptographic algorithms and related issues 📚
The profile MAY require the protocol or system using EAT to provide an algorithm negotiation mechanism 🤝
If not, the profile document should list a set of algorithms for each COSE and JOSE message type the profile allows, per Section 6.3.6. The verifier should implement all of them. The attester may implement any of them it wishes, possibly just one for each message type 🦋
If detached submodule digests are used, the profile should address how the hash algorithm(s) for the digests get determined 🍓
6.3.8. Detached EAT Bundle Support 📦
A profile should specify whether or not a detached EAT bundle (Section 5) can be sent. A profile should specify that a receiver accepts a detached EAT bundle if the sender is allowed to send one 💜
6.3.9. Key Identification 🔑
A profile should specify what must be sent to identify the verification, decryption, or MAC key(s). If multiple methods of key identification may be sent, a profile should require the receiver to support them all 🫶
Appendix F describes a number of methods for identifying verification keys. When encryption is used, there are further considerations. In some cases key identification is very simple, and in other cases multiple components are involved. For example, it might be simple, using a COSE key ID, or it might be complex, using an X.509 certificate hierarchy 🧩
While not always possible, a profile should specify, or make reference to, a full end-to-end specification for key identification. For example, a profile should specify in full detail how COSE key IDs are created, their life cycle, and so on, rather than just saying a COSE key ID is used. Likewise, a profile should specify the full details of an X.509 hierarchy, including extension processing, algorithms allowed, and so on, rather than just saying X.509 certificates are used 🌈
6.3.10. Endorsement Identification ✨
okay so, verification key identification and endorsement identification are kinda twins, or maybe they're literally the same thing 👯♀️ the profile may wanna spell out how endorsements get identified. BUT ngl, endorsement identification is optional, whereas key identification is NOT. so keys are a must-have and endorsements are a maybe 💅
6.3.11. Freshness 🕰️
okay this one is so important bc security considerations (see Section 9.3) need some kind of freshness mechanism 😤 this could be the EAT nonce claim from Section 4.1, or some claim or mechanism that's defined outside this doc. there are a bunch of options described in "Freshness" (Section 10 of [RFC9334]), and a profile should say which freshness mechanism or mechanisms are allowed. no guessing, bestie 🫶
if the EAT nonce claim is being used, a profile should say whether multiple nonces can be sent. and if it does allow multiple nonces, the receiver should process all of them. ✨
6.3.12. Claims Requirements 🧁
a profile is allowed to define brand new claims that this doc doesn't cover. slay 💖
this doc says an EAT receiver must accept tokens with claims it doesn't understand 🥹 but a profile for a specific use case can flip that and let the receiver reject tokens with unknown claims. it can also ban specific claims outright. 🚫
a profile for a specific use case can also tweak things and say some claims are required. 📋
and a profile can constrain claims that are defined in this doc or somewhere else. like, it could require the EAT nonce to be a certain length, or make the "location" claim always include the altitude. the vibe is very "you do you, but with receipts" 💅
now the pluggable claims are the main character here 🦋 they let you use different formats for their content. the "manifests" claim (Section 4.2.15) and the "measurements" claim (Section 4.2.16) are the classic examples, since they allow CoSWID and other formats. a profile should spell out which formats can be sent, assuming the matching CoAP content types have been registered. and a profile should make the receiver accept every format that's allowed to be sent 🎀
if a format has variations allowed inside it, the profile should also say which variations can be sent. e.g. CoSWID has variations, so a profile might make the receiver accept all the variations that are allowed to be sent. it's giving organization 📚
6.4. The Constrained Device Standard Profile 🔋
it's anticipated that there will be LOTS of profiles for EAT across lots of different use cases 🌈 this section gives a normative definition of one profile that's good for a bunch of constrained device use cases. this is the one we're obsessed with rn 😭
the identifier for this profile is urn:ietf:rfc:rfc9711. bestie, write it down 📝
+================+==================================================+
| Issue | Profile Definition |
+================+==================================================+
| CBOR/JSON | CBOR MUST be used. |
+----------------+--------------------------------------------------+
| CBOR Encoding | Definite-length maps and arrays MUST be |
| | used. |
+----------------+--------------------------------------------------+
| CBOR Encoding | Definite-length strings MUST be used. |
+----------------+--------------------------------------------------+
| CBOR | Preferred serialization MUST be used. |
| Serialization | |
+----------------+--------------------------------------------------+
| COSE | COSE_Sign1 MUST be used. |
| Protection | |
+----------------+--------------------------------------------------+
| Algorithms | The receiver MUST accept ES256, ES384, |
| | and ES512; the sender MUST send one of |
| | these. |
+----------------+--------------------------------------------------+
| Detached EAT | Detached EAT bundles MUST NOT be sent |
| Bundle Usage | with this profile. |
+----------------+--------------------------------------------------+
| Verification | Either the COSE key identifier (kid) or |
| Key | the UEID MUST be used to identify the |
| Identification | verification key. If both are present, |
| | the kid takes precedence. (It is assumed |
| | the receiver has access to a database of |
| | trusted verification keys, which allows a |
| | lookup of the verification key ID; the |
| | key format and means of distribution are |
| | beyond the scope of this profile.) |
+----------------+--------------------------------------------------+
| Endorsements | This profile contains no endorsement |
| | identifier. |
+----------------+--------------------------------------------------+
| Freshness | A new single unique nonce MUST be used |
| | for every token request. |
+----------------+--------------------------------------------------+
| Claims | No requirement is made for the presence |
| | or absence of claims other than requiring |
| | an EAT nonce. As per general EAT rules, |
| | the receiver MUST NOT error out on claims |
| | it does not understand. |
+----------------+--------------------------------------------------+Table 2: Constrained Device Profile Definition 📊
any profile with different requirements than the ones above MUST have a different profile identifier. no exceptions, periodt 🛑
also, lots of claims can show up in tokens that follow this profile, even ones this doc doesn't define 🌸 and even a slight deviation from the requirements above counts as a different profile, which MUST have its own identifier. like, if a kid (key identifier) or UEID isn't used for key identification, it's NOT in conformance with this profile 💀 another example: requiring some claim to be present is also not in conformance and needs another profile.
derivations of this profile are encouraged!! for example, another profile could just be "The Constrained Device Standard Profile" plus a requirement that claim xxxx and claim yyyy are present. easy peasy 🧚
7. Encoding and Collected CDDL 🎀
an EAT is fundamentally defined using CDDL, and this doc specifies how to encode that CDDL in CBOR or JSON. since CBOR can express some stuff JSON can't (like tags) or does some stuff differently (like labels), there's some CDDL that's specific to the encoding. it's a whole thing, but iykyk 🤭
7.1. Claims-Set and CDDL for CWT and JWT 🔐
CDDL was not used to define CWT or JWT. it wasn't even available at the time 😭
this doc defines CDDL for both CWT and JWT, and it does NOT change the encoding or semantics of anything in a CWT or JWT. the vibe is respect the originals 🫶
a Claims-Set is the central data structure for EAT, CWT, and JWT. it holds all the claims and it's the thing that gets secured by signing or other means. you literally can't define EAT, CWT, or JWT in CDDL without it. the CDDL definition of Claims-Set in here applies to EAT, CWT, and JWT ✨
this doc specifies how to encode a Claims-Set in CBOR or JSON.
with the exception of nested tokens and some other externally defined structures (e.g., SWIDs), an entire Claims-Set must be encoded in either CBOR or JSON, never a mixture. no mixing it up, girlies 🙅♀️
CDDL for the seven claims defined by [RFC8392] and [RFC7519] is also specified in this doc. 🌟
7.2. Encoding Data Types 🧩
the following subsections use the types defined in "Standard Prelude" (Appendix D of [RFC8610]). 💜
7.2.1. Common Data Types 💎
time-int is identical to the epoch-based time, but it disallows floating-point representation. no decimals allowed, sorry not sorry 🙅♀️
for CBOR-encoded tokens, OIDs are specified using the CDDL type name "oid" from [RFC9090]. they're encoded without the tag number. for JSON-encoded tokens, OIDs are text strings in the common form of "nn.nn.nn...". 🍓
unless it's explicitly indicated, URIs are NOT the URI tag defined in [RFC8949]. they're just text strings that contain a URI conforming to the format defined in [RFC3986]. ✨
here's the CDDL for all that, bestie 👇
time-int = #6.1(int)
binary-data = JC< base64-url-text, bstr>
base64-url-text = tstr .regexp "[A-Za-z0-9_-]+"
general-oid = JC< json-oid, ~oid >
json-oid = tstr .regexp "([0-2])((\\.0)|(\\.[1-9][0-9]*))*"
general-uri = JC< text, ~uri >
coap-content-format = uint .le 655357.2.2. JSON Interoperability 🫧
JSON should be encoded per Appendix E of [RFC8610]. on top of that, the following CDDL types get encoded in JSON like this:
- bstr -- MUST be base64url encoded.
- time -- MUST be encoded as NumericDate as described in Section 2 of [RFC7519].
- string-or-uri -- MUST be encoded as StringOrURI as described in Section 2 of [RFC7519].
- uri -- MUST be a URI [RFC3986].
- oid -- MUST be encoded as a string using the well-established dotted-decimal notation (e.g., the text "1.2.250.1") [RFC4517].
the CDDL generic "JC<>" gets used in most places where CBOR and JSON are different. the first argument is the CDDL for JSON, and the second is the CDDL for CBOR. it's like two outfits for one look 👗👗
7.2.3. Labels 🏷️
most map labels, Claims-Keys, Claim-Names, and enumerated-type values are integers for CBOR-encoded tokens and strings for JSON-encoded tokens. when that's the case, the JC<> CDDL construct is used to give both the integer and string values. two names, one vibe 💕
7.2.4. CBOR Interoperability 🌷
CBOR lets data items be serialized in more than one form so it can cover a bunch of use cases. this is addressed in Section 6. 🙌
7.3. Collected CDDL 📚
see [EAT-GitHub] for extra info and stub files when you're using the CDDL in this section to validate EAT protocol messages. 🧁
7.3.1. Payload CDDL ✨
okay so this is the payload CDDL 💅 It defines all the EAT claims that get added onto the main Claims-Set definition in Appendix D. Claims-Set is the payload for CWT, JWT, and potentially other token types 🦋. This covers both CBOR and JSON. When CBOR and JSON differ, the JC<> generic from Appendix D does the work. ngl the JC<> generic uses the ".feature" control operator from [RFC9165] 🧚
This CDDL uses Submodule and nested tokens but doesn't define them, because those definitions are different for CBOR and JSON and JC<> can't handle them 😭 The submodule claim is the one spot where a CBOR token can be nested inside a JSON token, and the other way around too 🫶 The encoding-specific definitions come in the next sections.
time-int = #6.1(int)
binary-data = JC< base64-url-text, bstr>
base64-url-text = tstr .regexp "[A-Za-z0-9_-]+"
general-oid = JC< json-oid, ~oid >
json-oid = tstr .regexp "([0-2])((\\.0)|(\\.[1-9][0-9]*))*"
general-uri = JC< text, ~uri >
coap-content-format = uint .le 65535
$$Claims-Set-Claims //=
(nonce-label => nonce-type / [ 2* nonce-type ])
nonce-type = JC< tstr .size (8..88), bstr .size (8..64)>
$$Claims-Set-Claims //= (ueid-label => ueid-type)
ueid-type = JC<base64-url-text .size (10..44) , bstr .size (7..33)>
$$Claims-Set-Claims //= (sueids-label => sueids-type)
sueids-type = {
+ tstr => ueid-type
}
$$Claims-Set-Claims //= (
oemid-label => oemid-pen / oemid-ieee / oemid-random
)
oemid-pen = int
oemid-ieee = JC<oemid-ieee-json, oemid-ieee-cbor>
oemid-ieee-cbor = bstr .size 3
oemid-ieee-json = base64-url-text .size 4
oemid-random = JC<oemid-random-json, oemid-random-cbor>
oemid-random-cbor = bstr .size 16
oemid-random-json = base64-url-text .size 24
$$Claims-Set-Claims //= (
hardware-version-label => hardware-version-type
)
hardware-version-type = [
version: tstr,
? scheme: $version-scheme
]
$$Claims-Set-Claims //= (
hardware-model-label => hardware-model-type
)
hardware-model-type = JC<base64-url-text .size (4..44),
bytes .size (1..32)>
$$Claims-Set-Claims //= ( sw-name-label => tstr )
$$Claims-Set-Claims //= (sw-version-label => sw-version-type)
sw-version-type = [
version: tstr
? scheme: $version-scheme
]
$$Claims-Set-Claims //= (oem-boot-label => bool)
$$Claims-Set-Claims //= ( debug-status-label => debug-status-type )
debug-status-type = ds-enabled /
disabled /
disabled-since-boot /
disabled-permanently /
disabled-fully-and-permanently
ds-enabled = JC< "enabled", 0 >
disabled = JC< "disabled", 1 >
disabled-since-boot = JC< "disabled-since-boot", 2 >
disabled-permanently = JC< "disabled-permanently", 3 >
disabled-fully-and-permanently =
JC< "disabled-fully-and-permanently", 4 >
$$Claims-Set-Claims //= (location-label => location-type)
location-type = {
latitude => number,
longitude => number,
? altitude => number,
? accuracy => number,
? altitude-accuracy => number,
? heading => number,
? speed => number,
? timestamp => ~time-int,
? age => uint
}
latitude = JC< "latitude", 1 >
longitude = JC< "longitude", 2 >
altitude = JC< "altitude", 3 >
accuracy = JC< "accuracy", 4 >
altitude-accuracy = JC< "altitude-accuracy", 5 >
heading = JC< "heading", 6 >
speed = JC< "speed", 7 >
timestamp = JC< "timestamp", 8 >
age = JC< "age", 9 >
$$Claims-Set-Claims //= (uptime-label => uint)
$$Claims-Set-Claims //= (boot-seed-label => binary-data)
$$Claims-Set-Claims //= (boot-count-label => uint)
$$Claims-Set-Claims //= ( intended-use-label => intended-use-type )
intended-use-type = JC< text, int>
$$Claims-Set-Claims //= (
dloas-label => [ + dloa-type ]
)
dloa-type = [
dloa_registrar: general-uri
dloa_platform_label: text
? dloa_application_label: text
]
$$Claims-Set-Claims //= (profile-label => general-uri / general-oid)
$$Claims-Set-Claims //= (
manifests-label => manifests-type
)
manifests-type = [+ manifest-format]
manifest-format = [
content-type: coap-content-format,
content-format: JC< $manifest-body-json,
$manifest-body-cbor >
]
$manifest-body-cbor /= bytes .cbor untagged-coswid
$manifest-body-json /= base64-url-text
$$Claims-Set-Claims //= (
measurements-label => measurements-type
)
measurements-type = [+ measurements-format]
measurements-format = [
content-type: coap-content-format,
content-format: JC< $measurements-body-json,
$measurements-body-cbor >
]
$measurements-body-cbor /= bytes .cbor untagged-coswid
$measurements-body-json /= base64-url-text
$$Claims-Set-Claims //= (
measurement-results-label =>
[ + measurement-results-group ] )
measurement-results-group = [
measurement-system: tstr,
measurement-results: [ + individual-result ]
]
individual-result = [
result-id: tstr / binary-data,
result: result-type,
]
result-type = comparison-success /
comparison-fail /
comparison-not-run /
measurement-absent
comparison-success = JC< "success", 1 >
comparison-fail = JC< "fail", 2 >
comparison-not-run = JC< "not-run", 3 >
measurement-absent = JC< "absent", 4 >
Detached-Submodule-Digest = [
hash-algorithm : text / int,
digest : binary-data
]
BUNDLE-Messages = BUNDLE-Tagged-Message / BUNDLE-Untagged-Message
BUNDLE-Tagged-Message = #6.602(BUNDLE-Untagged-Message)
BUNDLE-Untagged-Message = Detached-EAT-Bundle
Detached-EAT-Bundle = [
main-token : Nested-Token,
detached-claims-sets: {
+ tstr => JC<json-wrapped-claims-set,
cbor-wrapped-claims-set>
}
]
json-wrapped-claims-set = base64-url-text
cbor-wrapped-claims-set = bstr .cbor Claims-Set
nonce-label = JC< "eat_nonce", 10 >
ueid-label = JC< "ueid", 256 >
sueids-label = JC< "sueids", 257 >
oemid-label = JC< "oemid", 258 >
hardware-model-label = JC< "hwmodel", 259 >
hardware-version-label = JC< "hwversion", 260 >
uptime-label = JC< "uptime", 261 >
oem-boot-label = JC< "oemboot", 262 >
debug-status-label = JC< "dbgstat", 263 >
location-label = JC< "location", 264 >
profile-label = JC< "eat_profile", 265 >
submods-label = JC< "submods", 266 >
boot-count-label = JC< "bootcount", 267 >
boot-seed-label = JC< "bootseed", 268 >
dloas-label = JC< "dloas", 269 >
sw-name-label = JC< "swname", 270 >
sw-version-label = JC< "swversion", 271 >
manifests-label = JC< "manifests", 272 >
measurements-label = JC< "measurements", 273 >
measurement-results-label = JC< "measres" , 274 >
intended-use-label = JC< "intuse", 275 >7.3.2. CBOR-Specific CDDL 🎀
this part is the CBOR-only stuff 🦋 It sets up which CBOR tokens are allowed and how nesting works for CBOR, plus the submodule shape.
EAT-CBOR-Token = $CBOR-Tagged-Token / $EAT-CBOR-Untagged-Token
$CBOR-Tagged-Token /= CWT-Tagged-Message
$CBOR-Tagged-Token /= BUNDLE-Tagged-Message
$EAT-CBOR-Untagged-Token /= CWT-Untagged-Message
$EAT-CBOR-Untagged-Token /= BUNDLE-Untagged-Message
Nested-Token = CBOR-Nested-Token
CBOR-Nested-Token =
JSON-Token-Inside-CBOR-Token /
CBOR-Token-Inside-CBOR-Token
CBOR-Token-Inside-CBOR-Token = bstr .cbor $CBOR-Tagged-Token
JSON-Token-Inside-CBOR-Token = tstr
$$Claims-Set-Claims //= (submods-label => { + text => Submodule })
Submodule = Claims-Set / CBOR-Nested-Token /
Detached-Submodule-Digest7.3.3. JSON-Specific CDDL 🌈
and now the JSON side, which is basically the same energy but with its own nesting selector 💅 Same submods label, different Submodule options.
EAT-JSON-Token = $EAT-JSON-Token-Formats
$EAT-JSON-Token-Formats /= JWT-Message
$EAT-JSON-Token-Formats /= BUNDLE-Untagged-Message
Nested-Token = JSON-Selector
JSON-Selector = $JSON-Selector
$JSON-Selector /= [type: "JWT", nested-token: JWT-Message]
$JSON-Selector /= [type: "CBOR", nested-token:
CBOR-Token-Inside-JSON-Token]
$JSON-Selector /= [type: "BUNDLE", nested-token: Detached-EAT-Bundle]
$JSON-Selector /= [type: "DIGEST", nested-token:
Detached-Submodule-Digest]
CBOR-Token-Inside-JSON-Token = base64-url-text
$$Claims-Set-Claims //= (submods-label => { + text => Submodule })
Submodule = Claims-Set / JSON-Selector8. Privacy Considerations 🙈
certain EAT claims can be used to track the owner of an entity, so implementations should think about privacy-preserving options depending on how the EAT is being used 🔒 For example, the location claim might be suppressed in EATs sent to unauthenticated consumers.
8.1. UEID and SUEID Privacy Considerations 🫣
okay so a UEID is usually not privacy-preserving 😬 Relying parties that get tokens from a particular entity can tell the tokens came from the same entity and figure out who issued them.
so using the claim may violate privacy policies 💀 In other situations, a UEID won't be allowed for certain products, like browsers that protect the end user's privacy. It'll often be the case that tokens don't have a UEID for these reasons.
an SUEID is also usually not privacy-preserving. Depending on when and how it's generated, it may have fewer privacy issues than a UEID 🧚
there are a few strategies you can use to still put UEIDs and SUEIDs in tokens 🌸
- The entity gets explicit permission from the user of the entity to use the UEID/SUEID. This might be through a prompt or a license agreement. For example, agreements for some online banking and brokerage services might already cover use of a UEID/SUEID.
- The UEID/SUEID is only used in a particular context or use case. It is only used by one relying party.
- The entity authenticates the relying party and generates a derived UEID/SUEID just for that particular relying party. For example, the relying party could prove their identity cryptographically to the entity, and then the entity generates a UEID just for that relying party by hashing a proofed relying party ID with the main entity UEID/SUEID.
note that some of these strategies end up with multiple UEIDs and SUEIDs per entity 🦋 Each UEID/SUEID is used in a different context, use case, or system on the entity. From the view of the relying party, though, there is still just one UEID, and it's still globally universal across manufacturers ✨
8.2. Location Privacy Considerations 📍
Okay bestie, geographic location is almost always treated as personally identifiable information 😭 It's literally where you are, so it needs extra care. Implementors should consider the laws and regulations that govern sending location data from end-user devices to servers and services. Implementors should also consider using the location management facilities that the operating system offers on the entity generating the attestation. For example, lots of mobile phones ask the user for permission before they send location data 🙈
8.3. Boot Seed Privacy Considerations 🌱
The bootseed claim is basically a stable entity identifier within a given boot epoch 👀 So it's not suitable for attestation schemes that are privacy-preserving. Not the move if privacy is the vibe 💀
8.4. Replay Protection and Privacy 🛡️
EAT defines the EAT nonce claim for replay protection and token freshness ✨ The nonce claim is based on a value that's usually derived remotely (outside of the entity). This claim might accidentally, or on purpose, carry personally identifying information along for the ride. For instance, an implementor may choose a nonce that's equivalent to a username associated with the device (e.g., account login) 😬 If the token is inspected by a third party, then that information could be used to identify the source of the token or an account associated with the token. So to avoid sneaking privacy-related information into the nonce claim, it should be derived using a salt that comes from a true and reliable random number generator, or any other source of randomness that would still meet the target system requirements for replay protection and token freshness 🎀
9. Security Considerations 🔐
The security considerations in Section 8 of [RFC8392] and Section 11 of [RFC7519] apply to EAT in its CWT and JWT form, respectively. Moreover, Section 12 of [RFC9334] also applies to implementations of EAT 🧚 In addition, implementors should consider the information in the following subsections. Take this part seriously, babes, it's the security stuff 😤
9.1. Claim Trustworthiness 🕵️
This specification defines semantics for each claim. It does not require any particular level of security in the implementation of the claims, or even for the attester itself. That kind of specification is far beyond the scope of this document, which is about a message format, not the security level of an implementation 📦
The receiver of an EAT knows the trustworthiness of the claims in it by understanding the implementation made by the attester vendor and/or understanding the checks and processing performed by the verifier 🔍
For example, this document states that a UEID is permanent and that it must not change. But it does not describe any security requirements or a level of defense that would stop an attacker from changing the UEID 😭 Not the spec's job, but still wild to think about.
The degree of security will vary from use case to use case. In some cases, the receiver may only need to know something about the implementation, like that it was implemented in a TEE. In other cases, the receiver may require the attester to be certified by a particular certification program. Or maybe the receiver is totally content with very little security 🤷♀️
9.2. Key Provisioning 🗝️
Private key material can be used to sign and/or encrypt the EAT, or to derive the keys used for signing and/or encryption. In some instances, the manufacturer of the entity may create the key material separately and provision it in the entity itself. The manufacturer of any entity that is capable of producing an EAT should take care to make sure that any private key material is suitably protected before it's provisioned in the entity itself 🔒 This can require creating the key material in an enclave (see [RFC4949] for the definition of "enclave"), securely transmitting the key material from the enclave to the entity using an appropriate protocol, and persisting the private key material in some form of secure storage to which (preferably) only the entity has access 💜
9.2.1. Transmission of Key Material 🚚
Regarding transmission of key material from the enclave to the entity, the key material may pass through one or more intermediaries. So some form of protection (e.g., key wrapping) may be necessary 🎁 The transmission itself may be done electronically, but it can also be done by human courier. In the latter case, there should be minimal to no exposure of the key material to the human (e.g., encrypted portable memory) 🙈 Moreover, the human should transport the key material directly from the secure enclave where it was created to a destination secure enclave where it can be provisioned. No detours, no sketchy stops 🚫
9.3. Freshness ⏰
All EAT use MUST provide a freshness mechanism to prevent replay and related attacks. The extensive discussions in [RFC9334] on freshness, as well as the security considerations, apply here. One option to provide freshness is the EAT nonce claim (Section 4.1) 💖
9.4. Multiple EAT Consumers 👯♀️
In many cases, more than one EAT consumer may be required to fully verify the entity attestation. Examples include individual consumers for nested EATs, or consumers for individual claims within an EAT. When multiple consumers are required to verify an EAT, it's important to minimize information exposure to each consumer. In addition, the communication between multiple consumers should be secure 🔐
For instance, consider the example of an encrypted and signed EAT with multiple claims. A consumer may receive the EAT (denoted as the "receiving consumer"), decrypt its payload, and verify its signature, but then pass specific subsets of claims to other consumers for evaluation ("downstream consumers"). Since any COSE encryption will be removed by the receiving consumer, the communication of claim subsets to any downstream consumer MUST leverage an equivalent communication security protocol (e.g., TLS) 🫶
However, assume the EAT of the previous example is hierarchical, and each claim subset for a downstream consumer is created in the form of a nested EAT. Then the nested EAT itself is encrypted and cryptographically verifiable (due to its COSE envelope) by a downstream consumer. That's unlike the previous example, where a claims set without a COSE envelope is sent to a downstream consumer 🦋 Therefore, TLS between the receiving and downstream consumers is not strictly required. Nevertheless, downstream consumers of a nested EAT should provide a nonce unique to the EAT they are consuming ✨
9.5. Detached EAT Bundle Digest Security Considerations 📎
A detached EAT bundle is composed of a nested EAT and a claims set, as per Section 5. Although the attached claims set is vulnerable to modification in transit, any modification can be detected by the receiver through the associated digest, which is a claim fully contained within an EAT 🕵️♀️ Moreover, the digest itself can only be derived using an appropriate COSE hash algorithm. That means an attacker can't induce false detection of modified detached claims, because the algorithms in the COSE registry are assumed to be of sufficient cryptographic strength 💪
9.6. Verification Keys 🔑
In all cases, there must be some way that the verification key itself is verified or determined to be trustworthy. The key identification itself is never enough 🚫 This will always be done by some out-of-band mechanism that is not described here. For example, the verifier may be configured with a root certificate or a master key by the verifier system administrator 👑
Often, an X.509 certificate or an endorsement carries more than just the verification key. For example, an X.509 certificate might have key usage constraints, and an endorsement might have reference values. When this is the case, the key identifier must be either a protected header or in the payload, so that it is cryptographically bound to the EAT. This is in line with the requirements in "Key Identification" of JSON Web Signature (Section 6 of [RFC7515]) 🎀
10. IANA Considerations 📚
10.1. Reuse of CBOR and JSON Web Token (CWT and JWT) Claims Registries 🌈
Claims defined for EAT are compatible with those of CWT and JWT, so the CWT and JWT Claims registries, [IANA.CWT.Claims] and [IANA.JWT.Claims], are reused. No new IANA registry is created. Iconic reuse energy 👏
All EAT claims defined in this document have been placed in both registries. All new EAT claims defined subsequently should be placed in both registries 🍓
Appendix E describes some considerations when defining new claims 🧚
10.2. CWT and JWT Claims Registered by This Document
Okay so this section is where the new claims get officially added to two registries at once, the JWT one from [RFC7519] and the CBOR one from [RFC8392] 💅✨. Every entry below goes into both. 🦋
The "Claim Description", "Change Controller", and "Reference" fields are the same for JWT and CWT 🎀. The "Claim Key" and "Claim Value Type" fields are only for the CWT registry 🧚. The "Claim Name" field is defined for the CWT registry, not JWT 💜. The "JWT Claim Name" field is the same thing as the "Claim Name" field in the JWT registry 🌸.
IANA has registered the following claims 🫶:
Claim Name: Nonce
Claim Description: Nonce
JWT Claim Name: eat_nonce
Claim Key: 10
Claim Value Type: bstr or array
Change Controller: IETF
Reference: RFC 9711
Claim Name: UEID
Claim Description: Universal Entity ID
JWT Claim Name: ueid
CWT Claim Key: 256
Claim Value Type: bstr
Change Controller: IETF
Reference: RFC 9711
Claim Name: SUEIDs
Claim Description: Semipermanent UEIDs
JWT Claim Name: sueids
CWT Claim Key: 257
Claim Value Type: map
Change Controller: IETF
Reference: RFC 9711
Claim Name: Hardware OEM ID
Claim Description: Hardware OEM ID
JWT Claim Name: oemid
Claim Key: 258
Claim Value Type: bstr or int
Change Controller: IETF
Reference: RFC 9711
Claim Name: Hardware Model
Claim Description: Model identifier for hardware
JWT Claim Name: hwmodel
Claim Key: 259
Claim Value Type: bstr
Change Controller: IETF
Reference: RFC 9711
Claim Name: Hardware Version
Claim Description: Hardware Version Identifier
JWT Claim Name: hwversion
Claim Key: 260
Claim Value Type: array
Change Controller: IETF
Reference: RFC 9711
Claim Name: Uptime
Claim Description: Uptime
JWT Claim Name: uptime
Claim Key: 261
Claim Value Type: uint
Change Controller: IETF
Reference: RFC 9711
Claim Name: OEM Authorized Boot
Claim Description: Indicates whether the software booted was OEM
authorized
JWT Claim Name: oemboot
Claim Key: 262
Claim Value Type: bool
Change Controller: IETF
Reference: RFC 9711
Claim Name: Debug Status
Claim Description: The status of debug facilities
JWT Claim Name: dbgstat
Claim Key: 263
Claim Value Type: uint
Change Controller: IETF
Reference: RFC 9711
Claim Name: Location
Claim Description: The geographic location
JWT Claim Name: location
Claim Key: 264
Claim Value Type: map
Change Controller: IETF
Reference: RFC 9711
Claim Name: EAT Profile
Claim Description: The EAT profile followed
JWT Claim Name: eat_profile
Claim Key: 265
Claim Value Type: uri or oid
Change Controller: IETF
Reference: RFC 9711
Claim Name: Submodules Section
Claim Description: The section containing submodules
JWT Claim Name: submods
Claim Key: 266
Claim Value Type: map
Change Controller: IETF
Reference: RFC 9711
Claim Name: Boot Count
Claim Description: The number of times the entity or submodule has
been booted
JWT Claim Name: bootcount
Claim Key: 267
Claim Value Type: uint
Change Controller: IETF
Reference: RFC 9711
Claim Name: Boot Seed
Claim Description: Identifies a boot cycle
JWT Claim Name: bootseed
Claim Key: 268
Claim Value Type: bstr
Change Controller: IETF
Reference: RFC 9711
Claim Name: DLOAs
Claim Description: Certifications received as Digital Letters of
Approval
JWT Claim Name: dloas
Claim Key: 269
Claim Value Type: array
Change Controller: IETF
Reference: RFC 9711
Claim Name: Software Name
Claim Description: The name of the software running in the entity
JWT Claim Name: swname
Claim Key: 270
Claim Value Type: tstr
Change Controller: IETF
Reference: RFC 9711
Claim Name: Software Version
Claim Description: The version of software running in the entity
JWT Claim Name: swversion
Claim Key: 271
Claim Value Type: array
Change Controller: IETF
Reference: RFC 9711
Claim Name: Software Manifests
Claim Description: Manifests describing the software installed on
the entity
JWT Claim Name: manifests
Claim Key: 272
Claim Value Type: array
Change Controller: IETF
Reference: RFC 9711
Claim Name: Measurements
Claim Description: Measurements of the software, memory
configuration, and such on the entity
JWT Claim Name: measurements
Claim Key: 273
Claim Value Type: array
Change Controller: IETF
Reference: RFC 9711
Claim Name: Software Measurement Results
Claim Description: The results of comparing software measurements to
reference values
JWT Claim Name: measres
Claim Key: 274
Claim Value Type: array
Change Controller: IETF
Reference: RFC 9711
Claim Name: Intended Use
Claim Description: The intended use of the EAT
JWT Claim Name: intuse
Claim Key: 275
Claim Value Type: uint
Change Controller: IETF
Reference: RFC 971110.3. UEID URNs Registered by This Document
IANA has registered the following new subtypes in the "DEV URN Subtypes" registry [IANA.DEV-URNs] under the "Device Identification" registry group; see [RFC9039] ✨.
+=========+===================================+===========+
| Subtype | Description | Reference |
+=========+===================================+===========+
| ueid | Universal Entity ID | RFC 9711 |
+---------+-----------------------------------+-----------+
| sueid | Semipermanent Universal Entity ID | RFC 9711 |
+---------+-----------------------------------+-----------+
Table 3: UEID URN RegistrationThe ABNF [RFC5234] [RFC7405] for these two URNs is as follows, where b64ueid is the base64url-encoded binary byte string for the UEID or SUEID 🌈:
body =/ ueidbody
ueidbody = %s"ueid:" b64ueid10.4. CBOR Tag for Detached EAT Bundle Registered by This Document
In the "CBOR Tags" registry [IANA.cbor-tags], IANA has allocated the following tag from the Specification Required range, with the present document as the reference 🍓.
+=====+===========+=====================+=====================+
| Tag | Data Item | Semantics | Reference |
+=====+===========+=====================+=====================+
| 602 | array | Detached EAT Bundle | RFC 9711, Section 5 |
+-----+-----------+---------------------+---------------------+
Table 4: Detached EAT Bundle Tag Registration10.5. Intended Use Registry
IANA has created a new registry titled "Entity Attestation Token (EAT) Intended Uses" under the new "Remote Attestation Procedures (RATS)" registry group 👑. The registry uses the Expert Review registration procedure [RFC8126].
Guidelines for designated experts:
- Each intended use should be clearly described so a user knows what it means.
- Each intended use should be distinct from others that are registered.
- Point squatting is discouraged.
The three columns for the registry are:
- Value: This is a unique integer that is used to identify the intended use in CBOR-encoded tokens.
- Description: This is one or more text paragraphs that sufficiently define what the intended use means. It may also be a reference to another document.
- Reference: This field contains a reference to the defining specification.
The following 5 values represent the initial content of the registry. Note that 0 will be marked as "reserved" for the CBOR value, and the maximum CBOR value for assignment is 255 🔥.
1 -- Generic: Generic attestation describes an application where the
EAT consumer requires the most up-to-date security assessment of
the attesting entity. It is expected that this is the most
commonly used application of EAT.
2 -- Registration: Entities that are registering for a new service
may be expected to provide an attestation as part of the
registration process. This "intuse" setting indicates that the
attestation is not intended for any use but registration.
3 -- Provisioning: Entities may be provisioned with different values
or settings by an EAT consumer. Examples include key material or
device management trees. The consumer may require an EAT to
assess entity security state of the entity prior to provisioning.
4 -- Certificate Issuance: Certification Authorities (CAs) may
require attestation results (which in a background check model
might require receiving evidence to be passed to a verifier) to
make decisions about the issuance of certificates. An EAT may be
used as part of the certificate signing request (CSR).
5 -- Proof of Possession: An EAT consumer may require an attestation
as part of an accompanying proof-of-possession (PoP) application.
More precisely, a PoP transaction is intended to provide the
recipient with cryptographically verifiable proof that the sender
has possession of a key. This kind of attestation may be
necessary to verify the security state of the entity storing the
private key used in a PoP application.11. References 🧚
11.1. Normative References 📚✨
bestie, the reference list is where all the receipts live 🧾💅 these are the docs the spec leans on, so they're mandatory reading 🫶
[DLOA] GlobalPlatform, "GlobalPlatform Card: Digital Letter of Approval", Public Release Version 1.0, GPC_SPE_095, November 2015, <https://globalplatform.org/wp-content/uploads/2015/12/GPC_DigitalLetterOfApproval_v1.0.pdf>.
[IANA.cbor-tags] IANA, "CBOR Tags", <https://www.iana.org/assignments/cbor-tags>.
[IANA.COSE.Algorithms] IANA, "COSE Algorithms", <https://www.iana.org/assignments/cose>.
[IANA.CWT.Claims] IANA, "CBOR Web Token (CWT) Claims", <https://www.iana.org/assignments/cwt>.
[IANA.DEV-URNs] IANA, "DEV URN Subtypes", <https://www.iana.org/assignments/device-identification>.
[IANA.JWT.Claims] IANA, "JSON Web Token Claims", <https://www.iana.org/assignments/jwt>.
[PEN] IANA, "Private Enterprise Numbers (PENs)", <https://www.iana.org/assignments/enterprise-numbers/>.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997, <https://www.rfc-editor.org/info/rfc2119>.
[RFC3986] Berners-Lee, T., Fielding, R., and L. Masinter, "Uniform Resource Identifier (URI): Generic Syntax", STD 66, RFC 3986, DOI 10.17487/RFC3986, January 2005, <https://www.rfc-editor.org/info/rfc3986>.
[RFC4517] Legg, S., Ed., "Lightweight Directory Access Protocol (LDAP): Syntaxes and Matching Rules", RFC 4517, DOI 10.17487/RFC4517, June 2006, <https://www.rfc-editor.org/info/rfc4517>.
[RFC4648] Josefsson, S., "The Base16, Base32, and Base64 Data Encodings", RFC 4648, DOI 10.17487/RFC4648, October 2006, <https://www.rfc-editor.org/info/rfc4648>.
[RFC5234] Crocker, D., Ed. and P. Overell, "Augmented BNF for Syntax Specifications: ABNF", STD 68, RFC 5234, DOI 10.17487/RFC5234, January 2008, <https://www.rfc-editor.org/info/rfc5234>.
[RFC7252] Shelby, Z., Hartke, K., and C. Bormann, "The Constrained Application Protocol (CoAP)", RFC 7252, DOI 10.17487/RFC7252, June 2014, <https://www.rfc-editor.org/info/rfc7252>.
[RFC7405] Kyzivat, P., "Case-Sensitive String Support in ABNF", RFC 7405, DOI 10.17487/RFC7405, December 2014, <https://www.rfc-editor.org/info/rfc7405>.
[RFC7515] Jones, M., Bradley, J., and N. Sakimura, "JSON Web Signature (JWS)", RFC 7515, DOI 10.17487/RFC7515, May 2015, <https://www.rfc-editor.org/info/rfc7515>.
[RFC7519] Jones, M., Bradley, J., and N. Sakimura, "JSON Web Token (JWT)", RFC 7519, DOI 10.17487/RFC7519, May 2015, <https://www.rfc-editor.org/info/rfc7519>.
[RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words", BCP 14, RFC 8174, DOI 10.17487/RFC8174, May 2017, <https://www.rfc-editor.org/info/rfc8174>.
[RFC8259] Bray, T., Ed., "The JavaScript Object Notation (JSON) Data Interchange Format", STD 90, RFC 8259, DOI 10.17487/RFC8259, December 2017, <https://www.rfc-editor.org/info/rfc8259>.
[RFC8392] Jones, M., Wahlstroem, E., Erdtman, S., and H. Tschofenig, "CBOR Web Token (CWT)", RFC 8392, DOI 10.17487/RFC8392, May 2018, <https://www.rfc-editor.org/info/rfc8392>.
[RFC8610] Birkholz, H., Vigano, C., and C. Bormann, "Concise Data Definition Language (CDDL): A Notational Convention to Express Concise Binary Object Representation (CBOR) and JSON Data Structures", RFC 8610, DOI 10.17487/RFC8610, June 2019, <https://www.rfc-editor.org/info/rfc8610>.
[RFC8792] Watsen, K., Auerswald, E., Farrel, A., and Q. Wu, "Handling Long Lines in Content of Internet-Drafts and RFCs", RFC 8792, DOI 10.17487/RFC8792, June 2020, <https://www.rfc-editor.org/info/rfc8792>.
[RFC8949] Bormann, C. and P. Hoffman, "Concise Binary Object Representation (CBOR)", STD 94, RFC 8949, DOI 10.17487/RFC8949, December 2020, <https://www.rfc-editor.org/info/rfc8949>.
[RFC9052] Schaad, J., "CBOR Object Signing and Encryption (COSE): Structures and Process", STD 96, RFC 9052, DOI 10.17487/RFC9052, August 2022, <https://www.rfc-editor.org/info/rfc9052>.
[RFC9090] Bormann, C., "Concise Binary Object Representation (CBOR) Tags for Object Identifiers", RFC 9090, DOI 10.17487/RFC9090, July 2021, <https://www.rfc-editor.org/info/rfc9090>.
[RFC9165] Bormann, C., "Additional Control Operators for the Concise Data Definition Language (CDDL)", RFC 9165, DOI 10.17487/RFC9165, December 2021, <https://www.rfc-editor.org/info/rfc9165>.
[RFC9334] Birkholz, H., Thaler, D., Richardson, M., Smith, N., and W. Pan, "Remote ATtestation procedureS (RATS) Architecture", RFC 9334, DOI 10.17487/RFC9334, January 2023, <https://www.rfc-editor.org/info/rfc9334>.
[RFC9393] Birkholz, H., Fitzgerald-McKay, J., Schmidt, C., and D. Waltermire, "Concise Software Identification Tags", RFC 9393, DOI 10.17487/RFC9393, June 2023, <https://www.rfc-editor.org/info/rfc9393>.
[ThreeGPP.IMEI] 3GPP, "Numbering, addressing and identification", Version 19, 3GPP TS 23.003, September 2024, <https://portal.3gpp.org/desktopmodules/Specifications/SpecificationDetails.aspx?specificationId=729>.
[W3C.GeoLoc] Cáceres, M. and R. Grant, "Geolocation", W3C Recommendation, September 2024, <https://www.w3.org/TR/geolocation/>.
[WGS84] National Geospatial-Intelligence Agency (NGA), "Department of Defense World Geodetic System 1984: Its Definition and Relationships with Local Geodetic Systems", NGA.STND.0036_1.0.0_WGS84, July 2014, <https://nsgreg.nga.mil/doc/view?i=4085>.
11.2. Informative References 🌸
these ones are the background reading 📖 handy for context, but the spec doesn't depend on them the same way 🫶
[BirthdayAttack] Wikipedia, "Birthday attack", October 2024, <https://en.wikipedia.org/w/index.php?title=Birthday_attack&oldid=1249270346>.
[CBOR.Certs] Mattsson, J. P., Selander, G., Raza, S., Höglund, J., and M. Furuhed, "CBOR Encoded X.509 Certificates (C509 Certificates)", Work in Progress, Internet-Draft, draft-ietf-cose-cbor-encoded-cert-13, 3 March 2025, <https://datatracker.ietf.org/doc/html/draft-ietf-cose-cbor-encoded-cert-13>.
[CC-Example] Eurosmart, "Secure Sub-System in System-on-Chip (3S in SoC) Protection Profile", Version 1.8, October 2023, <https://commoncriteriaportal.org/nfs/ccpfiles/files/ppfiles/pp0117V2b_pdf.pdf>.
[EAT-GitHub] "The Entity Attestation Token (EAT)", commit 62c726b, January 2024, <https://github.com/ietf-rats-wg/eat>.
[EAT.media-types] Lundblade, L., Birkholz, H., and T. Fossati, "EAT Media Types", Work in Progress, Internet-Draft, draft-ietf-rats-eat-media-type-12, 3 November 2024, <https://datatracker.ietf.org/doc/html/draft-ietf-rats-eat-media-type-12>.
[GP-Example] GlobalPlatform, "GlobalPlatform Technology: TEE Certification Process", Public Release Version 2.0, GP_PRO_023, January 2021, <https://globalplatform.org/wp-content/uploads/2021/01/GP_TEECertificationProcess_v2.0_PublicRelease.pdf>.
[IEEE-RA] IEEE, "IEEE Registration Authority", <https://standards.ieee.org/products-services/regauth/index.html>.
[IEEE.802-2014] IEEE, "IEEE Standard for Local and Metropolitan Area Networks: Overview and Architecture", IEEE Std 802-2014, DOI 10.1109/IEEESTD.2014.6847097, June 2014, <https://ieeexplore.ieee.org/document/6847097>.
[IEEE.802.1AR] IEEE, "IEEE Standard for Local and Metropolitan Area Networks - Secure Device Identity", IEEE Std 802.1AR-2018, DOI 10.1109/IEEESTD.2018.8423794, August 2018, <https://ieeexplore.ieee.org/document/8423794>.
[JTAG] IEEE, "IEEE Standard for Reduced-Pin and Enhanced-Functionality Test Access Port and Boundary-Scan Architecture", IEEE Std 1149.7-2009, DOI 10.1109/IEEESTD.2010.5412866, February 2010, <https://ieeexplore.ieee.org/document/5412866>.
[OUI.Guide] IEEE, "Guidelines for Use of Extended Unique Identifier (EUI), Organizationally Unique Identifier (OUI), and Company ID (CID)", August 2017, <https://standards.ieee.org/content/dam/ieee-standards/standards/web/documents/tutorials/eui.pdf>.
[OUI.Lookup] IEEE, "IEEE Registration Authority: Assignments", <https://regauth.standards.ieee.org/standards-ra-web/pub/view.html#registries>.
[RFC4949] Shirey, R., "Internet Security Glossary, Version 2", FYI 36, RFC 4949, DOI 10.17487/RFC4949, August 2007, <https://www.rfc-editor.org/info/rfc4949>.
[RFC8126] Cotton, M., Leiba, B., and T. Narten, "Guidelines for Writing an IANA Considerations Section in RFCs", BCP 26, RFC 8126, DOI 10.17487/RFC8126, June 2017, <https://www.rfc-editor.org/info/rfc8126>.
[RFC9039] Arkko, J., Jennings, C., and Z. Shelby, "Uniform Resource Names for Device Identifiers", RFC 9039, DOI 10.17487/RFC9039, June 2021, <https://www.rfc-editor.org/info/rfc9039>.
[RFC9360] Schaad, J., "CBOR Object Signing and Encryption (COSE): Header Parameters for Carrying and Referencing X.509 Certificates", RFC 9360, DOI 10.17487/RFC9360, February 2023, <https://www.rfc-editor.org/info/rfc9360>.
[RFC9562] Davis, K., Peabody, B., and P. Leach, "Universally Unique IDentifiers (UUIDs)", RFC 9562, DOI 10.17487/RFC9562, May 2024, <https://www.rfc-editor.org/info/rfc9562>.
[UCCS] Birkholz, H., O'Donoghue, J., Cam-Winget, N., and C. Bormann, "A CBOR Tag for Unprotected CWT Claims Sets", Work in Progress, Internet-Draft, draft-ietf-rats-uccs-12, 3 November 2024, <https://datatracker.ietf.org/doc/html/draft-ietf-rats-uccs-12>.
Appendix A. Examples 🎀
okay so the examples are the fun part 🧚 most of them are shown as a Claims-Set, which is the payload you'd put in a CWT, a JWT, a detached EAT bundle, or whatever future token types show up 🦋 the signing is left off so the Claims-Set is easier to see, but a few signed tokens get shown too 🔥
A.1. Claims Set Examples 💜
A.1.1. Simple TEE Attestation ✨
okay so this one is a simple attestation of a TEE 🛡️, and it comes with a manifest that's a payload CoSWID describing the TEE's software. no cap, it's giving clean and minimal 💅
The first chunk is the EAT payload describing this simple TEE 🌸
{
/ eat_nonce / 10: h'48df7b172d70b5a18935d0460a73dd71',
/ oemboot / 262: true,
/ dbgstat / 263: 2, / disabled-since-boot /
/ manifests / 272: [
[
258, / CoAP Content ID for CoSWID /
/ This is a byte-string-wrapped /
/ payload CoSWID. It gives the TEE /
/ software name, the version, and /
/ the name of the file it is in. /
/ {0: "3a24", /
/ 12: 1, /
/ 1: "Acme TEE OS", /
/ 13: "3.1.4", /
/ 2: [{31: "Acme TEE OS", 33: 1}, /
/ {31: "Acme TEE OS", 33: 2}], /
/ 6: { /
/ 17: { /
/ 24: "acme_tee_3.exe" /
/ } /
/ } /
/ } /
h' a60064336132340c01016b
41636d6520544545204f530d65332e31
2e340282a2181f6b41636d6520544545
204f53182101a2181f6b41636d652054
4545204f5318210206a111a118186e61
636d655f7465655f332e657865'
]
]
}And this is the payload CoSWID made by the software (SW) vendor 🎀. All it really does is name the TEE SW, name its version, and list the one file that makes up the TEE. Simple, iconic, done 🍓
1398229316({
/ Unique CoSWID ID / 0: "3a24",
/ tag-version / 12: 1,
/ software-name / 1: "Acme TEE OS",
/ software-version / 13: "3.1.4",
/ entity / 2: [
{
/ entity-name / 31: "Acme TEE OS",
/ role / 33: 1 / tag-creator /
},
{
/ entity-name / 31: "Acme TEE OS",
/ role / 33: 2 / software-creator /
}
],
/ payload / 6: {
/ ...file / 17: {
/ ...fs-name / 24: "acme_tee_3.exe"
}
}
})A.1.2. Submodules for Board and Device 🧚
this example is all about using submodules to describe the chip, the board, and the overall device 🔌. the main attestation belongs to the chip that has the CPU and runs the main OS. that chip is the one holding the keys and producing the token 🔑. the board comes from a different vendor than the chip (think generic IoT board), and the device is a specific appliance made by yet another vendor 🏠
the board and device submodules aren't the typical target environments in RATS Architecture (RFC 9334), but they're still a valid use of submodules, so we love to see it ✨
/ This example shows use of submodules to give information /
/ about the chip, board, and overall device. /
/ /
/ The main attestation is associated with the chip /
/ containing the CPU and running the main OS. It is what /
/ has the keys and produces the token. /
/ /
/ The board is made by a different vendor than the chip; /
/ perhaps it is some generic IoT board. /
/ /
/ The device is some specific appliance that is made by a /
/ different vendor than either the chip or the board. /
/ /
/ Here, the board and device submodules aren't the typical /
/ target environments as described by RATS Architecture /
/ (RFC 9334), but they are a valid use of submodules. /
{
/ eat_nonce / 10: h'e253cabedc9eec24ac4e25bcbeaf7765',
/ ueid / 256: h'0198f50a4ff6c05861c8860d13a638ea',
/ oemid / 258: h'894823', / IEEE OUI format OEM ID /
/ hwmodel / 259: h'549dcecc8b987c737b44e40f7c635ce8'
/ Hash of chip model name /,
/ hwversion / 260: ["1.3.4", 1], / Multipartnumeric /
/ swname / 270: "Acme OS",
/ swversion / 271: ["3.5.5", 1],
/ oemboot / 262: true,
/ dbgstat / 263: 3, / permanent-disable /
/ timestamp (iat) / 6: 1526542894,
/ submods / 266: {
/ A submodule to hold some claims about the circuit board /
"board" : {
/ oemid / 258: h'9bef8787eba13e2c8f6e7cb4b1f4619a',
/ hwmodel / 259: h'ee80f5a66c1fb9742999a8fdab930893'
/ Hash of board module name /,
/ hwversion / 260: ["2.0a", 2] / multipartnumeric+sfx /
},
/ A submodule to hold claims about the overall device /
"device" : {
/ oemid / 258: 61234, / PEN Format OEM ID /
/ hwversion / 260: ["4.0", 1] / Multipartnumeric /
}
}
}A.1.3. EAT Produced by an Attestation Hardware Block 💜
this one is a token made by a hardware block that was purpose-built for attestation 🤖. only the nonce changes from one attestation to the next, because everything else comes from the hardware itself or from one-time-programmable memory (like a fuse) 🔥. the whole encoded token is 47 bytes, with 8 of those bytes being the nonce and 16 being the UEID. small but mighty 🦋
/ This is an example of a token produced by a hardware block /
/ purposely built for attestation. Only the nonce claim changes /
/ from one attestation to the next as the rest come from either /
/ the hardware directly or from one-time-programmable memory /
/ (e.g., a fuse). The entire encoded token is 47 bytes, 8 of /
/ which are the nonce and 16 of which are the UEID. /
{
/ eat_nonce / 10: h'd79b964ddd5471c1393c8888',
/ ueid / 256: h'0198f50a4ff6c05861c8860d13a638ea',
/ oemid / 258: 64242, / Private Enterprise Number /
/ oemboot / 262: true,
/ dbgstat / 263: 3, / disabled-permanently /
/ hwversion / 260: [ "3.1", 1 ] / Type is multipartnumeric /
}A.1.4. Key / Key Store Attestation 🔐
this is an attestation of a public key plus the key store implementation that protects and manages it 🗝️. the key store lives in a security-oriented execution environment that's separate from the high-level OS (HLOS), like a Trusted Execution Environment (TEE), and the key store is the attester. main character energy 👑
the HLOS gets some attestation too, just its version and its boot and debug status. it's a Claims-Set submodule because it has a lower security level than the key store. since the key store's implementation can see info about the HLOS, it's able to include it 🫶
a key and an indication of the user authentication used to get access to that key are also included. the labels for these are in the private space, because this is a hypothetical example and not part of a standard protocol 🙈
/ This is an attestation of a public key and the key store /
/ implementation that protects and manages it. The key store /
/ implementation is in a security-oriented execution /
/ environment separate from the high-level OS (HLOS), for /
/ example, a Trusted Execution Environment (TEE). The key /
/ store is the attester. /
/ /
/ There is some attestation of the HLOS, just version and /
/ boot and debug status. It is a Claims-Set submodule because /
/ it has a lower security level than the key store. The key /
/ store's implementation has access to info about the HLOS, so /
/ it is able to include it. /
/ /
/ A key and an indication of the user authentication given to /
/ allow access to the key is given. The labels for these are /
/ in the private space as this is a hypothetical example, /
/ not part of a standard protocol. /
{
/ eat_nonce / 10: h'99b67438dba40743266f70bf75feb1026d5134
97a229bfe8',
/ oemboot / 262: true,
/ dbgstat / 263: 2, / disabled-since-boot /
/ manifests / 272: [
[ 258, / CoAP Content ID. /
h'a600683762623334383766
0c000169436172626f6e6974650d6331
2e320e0102a2181f75496e6475737472
69616c204175746f6d6174696f6e1821
02'
]
/ Above is an encoded CoSWID /
/ with the following data /
/ SW Name: "Carbonite" /
/ SW Vers: "1.2" /
/ SW Creator: /
/ "Industrial Automation" /
],
/ exp / 4: 1634324274, / 2021-10-15T18:57:54Z /
/ iat / 6: 1634317080, / 2021-10-15T16:58:00Z /
-80000 : "fingerprint",
-80001 : { / The key -- A COSE_Key /
/ kty / 1: 2, / EC2, elliptic curve with x & y/
/ kid / 2: h'36675c206f96236c3f51f54637b94ced',
/ curve / -1: 2, / curve is P-256 /
/ x-coord / -2: h'65eda5a12577c2bae829437fe338701a
10aaa375e1bb5b5de108de439c08551d',
/ y-coord / -3: h'1e52ed75701163f7f9e40ddf9f341b3d
c9ba860af7e0ca7ca7e9eecd0084d19c'
},
/ submods / 266 : {
"HLOS" : { / submod for high-level OS /
/ eat_nonce / 10: h'8b0b28782a23d3f6',
/ oemboot / 262: true,
/ manifests / 272: [
[ 258, / CoAP Content ID. /
h'a600687337
6537346b78380c000168
44726f6964204f530d65
52322e44320e0302a218
1F75496E647573747269
616c204175746f6d6174
696f6e182102'
]
/ Above is an encoded CoSWID /
/ with the following data: /
/ SW Name: "Droid OS" /
/ SW Vers: "R2.D2" /
/ SW Creator: /
/ "Industrial Automation"/
]
}
}
}A.1.5. Software Measurements of an IoT Device ✨
okay bestie, this one is a simple token for an IoT device 🧚 It carries CoSWID format measurements of the software (SW) 💾 The CoSWID is a byte string wrapped inside the token, and we also show it in diagnostic form so we can actually read it 👀
Here's the payload. The attestation is produced by the TEE 🔐 There's a submodule for the IoT OS, which is the device's main OS but not as locked down as the TEE, and the TEE also measures that OS and puts the measurements in the submodule 💅
/ This EAT payload is for an IoT device with a TEE. The attestation /
/ is produced by the TEE. There is a submodule for the IoT OS (the /
/ main OS of the IoT device that is not as secure as the TEE). The /
/ submodule contains claims for the IoT OS. The TEE also measures /
/ the IoT OS and puts the measurements in the submodule. /
{
/ eat_nonce / 10: h'5e19fba4483c7896',
/ oemboot / 262: true,
/ dbgstat / 263: 2, / disabled-since-boot /
/ oemid / 258: h'8945ad', / IEEE CID based /
/ ueid / 256: h'0198f50a4ff6c05861c8860d13a638ea',
/ submods / 266: {
"OS" : {
/ oemboot / 262: true,
/ dbgstat / 263: 2, / disabled-since-boot /
/ measurements / 273: [
[
258, / CoAP Content ID /
/ This is a byte-string-wrapped /
/ evidence CoSWID. It has /
/ hashes of the main files of /
/ the IoT OS. /
h'a600663463613234350c
17016d41636d6520522d496f542d4f
530d65332e312e3402a2181f724163
6d6520426173652041747465737465
7218210103a11183a318187161636d
655f725f696f745f6f732e65786514
1a0044b349078201582005f6b327c1
73b4192bd2c3ec248a292215eab456
611bf7a783e25c1782479905a31818
6d7265736f75726365732e72736314
1a000c38b10782015820c142b9aba4
280c4bb8c75f716a43c99526694caa
be529571f5569bb7dc542f98a31818
6a636f6d6d6f6e2e6c6962141a0023
3d3b0782015820a6a9dcdfb3884da5
f884e4e1e8e8629958c2dbc7027414
43a913e34de9333be6'
]
]
}
}
}And here's the evidence CoSWID itself, the bit that describes the software 🌸
/ This is an evidence CoSWID created for the "Acme R-IoT-OS" /
/ that is created by the "Acme Base Attester" (both fictitious /
/ names). It provides measurements of the SW (other than the /
/ attester SW) on the device. /
1398229316({
/ Unique CoSWID ID / 0: "4ca245",
/ tag-version / 12: 23, / Attester-maintained counter /
/ software-name / 1: "Acme R-IoT-OS",
/ software-version / 13: "3.1.4",
/ entity / 2: {
/ entity-name / 31: "Acme Base Attester",
/ role / 33: 1 / tag-creator /
},
/ evidence / 3: {
/ ...file / 17: [
{
/ ...fs-name / 24: "acme_r_iot_os.exe",
/ ...size / 20: 4502345,
/ ...hash / 7: [
1, / SHA-256 /
h'05f6b327c173b419
2bd2c3ec248a2922
15eab456611bf7a7
83e25c1782479905'
]
},
{
/ ...fs-name / 24: "resources.rsc",
/ ...size / 20: 800945,
/ ...hash / 7: [
1, / SHA-256 /
h'c142b9aba4280c4b
b8c75f716a43c995
26694caabe529571
f5569bb7dc542f98'
]
},
{
/ ...fs-name / 24: "common.lib",
/ ...size / 20: 2309435,
/ ...hash / 7: [
1, / SHA-256 /
h'a6a9dcdfb3884da5
f884e4e1e8e86299
58c2dbc702741443
a913e34de9333be6'
]
}
]
}
})A.1.6. Attestation Results in JSON 💜
this is a JSON-encoded payload that could be the output of a verifier that checked the IoT Attestation example right above 🫶
This particular verifier knows enough about the TEE attester to pass claims like debug status straight through to the relying party 🔥 It also knows the reference values for the measured software components and can check them. It tells the relying party they were correct in the "measres" claim ✨ "Trustus Verifications" is the name of the service that verifies the software component measurements 💖
{
"eat_nonce": "jkd8KL-8xQk",
"oemboot": true,
"dbgstat": "disabled-since-boot",
"oemid": "iUWt",
"ueid": "AZj1Ck_2wFhhyIYNE6Y4",
"swname": "Acme R-IoT-OS",
"swversion": [
"3.1.4"
],
"measres": [
[
"Trustus Measurements",
[
[
"all",
"success"
]
]
]
]
}A.1.7. JSON-Encoded Token with Submodules 🧚
the lines in this example are wrapped per [RFC8792] 🌈
=============== NOTE: '\' line wrapping per RFC 8792 ================
{
"eat_nonce": "lI-IYNE6Rj6O",
"ueid": "AJj1Ck_2wFhhyIYNE6Y46g==",
"secboot": true,
"dbgstat": "disabled-permanently",
"iat": 1526542894,
"submods": {
"Android App Foo": {
"swname": "Foo.app"
},
"Secure Element Eat": [
"CBOR",
"2D3ShEOhASagWGaoCkiUj4hg0TpGPhkBAFABmPUKT_bAWGHIhg0TpjjqGQ\
ECGfryGQEFBBkBBvUZAQcDGQEEgmMzLjEBGQEKoWNURUWCL1gg5c-V_ST6txRGdC3VjU\
Pa4XjlX-K5QpGpKRCC_8JjWgtYQPaQywOIZ3-mJKN3X9fLxOhAnsmBa-MvpHRzOw-Ywn\
-67bvJljuctezAPD41s6_At7NbSV3qwJlxIuqGfwe41es="
],
"Linux Android": {
"swname": "Android"
},
"Subsystem J": [
"JWT",
"eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJpc3MiOiJKLUF0dGVzd\
GVyIiwiaWF0IjoxNjUxNzc0ODY4LCJleHAiOm51bGwsImF1ZCI6IiIsInN1YiI6IiJ9.\
gjw4nFMhLpJUuPXvMPzK1GMjhyJq2vWXg1416XKszwQ"
]
}
}A.2. Signed Token Examples 🎀
A.2.1. Basic CWT Example 🦋
this is a simple CWT-format token signed with the Elliptic Curve Digital Signature Algorithm (ECDSA) 🔐 Here's the full thing. The payload is the attestation hardware block from Appendix A.1.3 of RFC 9711, and the main structure you can see is the COSE_Sign1 🍓
/ This is a full CWT-format token. The payload is the /
/ attestation hardware block in Appendix A.1.3 of /
/ RFC 9711. The main structure that is visible is /
/ that of the COSE_Sign1. /
61( 18( [
h'A10126', / protected headers /
{}, / empty unprotected headers /
h'A60A4CD79B964DDD5471C1393C88881901005001
98F50A4FF6C05861C8860D13A638EA19010219FA
F2190106F5190107031901048263332E3101', / payload /
h'9B9B2F5E470000F6A20C8A4157B5763FC45BE759
9A5334028517768C21AFFB845A56AB557E0C8973
A07417391243A79C478562D285612E292C622162
AB233787' / signature /
] ) )A.2.2. CBOR-Encoded Detached EAT Bundle ✨
okay so in this detached EAT bundle 🎀 the main token comes from a hardware (HW) attestation block 🔐. the detached Claims-Set comes from a TEE, and it's basically the same vibe as the simple TEE example in Appendix A.1.1 💅. the TEE digests its Claims-Set and hands that digest over to the HW block.
in a better example 🌟, the attestation from the HW block would be a CWT, so the HW block signs and secures it. and since that signature covers the TEE's digest, the Claims-Set is secured too 🫶
and the detached EAT bundle itself can be assembled by untrusted software 💀 (yes, really).
/ This is a detached EAT bundle tag. /
602([
/ The first part is a full EAT token with claims like nonce /
/ and UEID. Most importantly, it includes a submodule that /
/ is a detached digest, which is the hash of the "TEE" /
/ claims set in the next part of the detached EAT bundle. /
/ The COSE payload is as follows: /
/ { /
/ 10: h'948F8860D13A463E', /
/ 256: h'0198F50A4FF6C05861C8860D13A638EA', /
/ 258: 64242, /
/ 262: true, /
/ 263: 3, /
/ 260: ["3.1", 1], /
/ 266: { /
/ "TEE": [ /
/ -16, /
/ h'ab86f765643aabfd09c84eebe150b7f6 /
/ 1bc24804cee75e90c5f99cb850fe808f' /
/ ] /
/ } /
/ } /
h'D83DD28443A10126A05866A80A48948F8860D13A463E1901
00500198F50A4FF6C05861C8860D13A638EA19010219FAF2
19010504190106F5190107031901048263332E310119010A
A163544545822F58208DEF652F47000710D9F466A4C666E2
09DD74F927A1CEA352B03143E188838ABE5840F690CB0388
677FA624A3775FD7CBC4E8409EC9816BE32FA474733B0F98
C27FBAEDBBC9963B9CB5ECC03C3E35B3AFC0B7B35B495DEA
C0997122EA867F07B8D5EB',
{
/ A CBOR-encoded byte-string-wrapped EAT claims-set. /
/ It contains claims for a simple TEE attestation. /
"TEE" : h'a40a5048df7b172d70b5a18935d0460a73dd7119
0106f51901070219011081821901025858a60064
336132340c01016b41636d6520544545204f530d
65332e312e340282a2181f6b41636d6520544545
204f53182101a2181f6b41636d6520544545204f
5318210206a111a118186e61636d655f7465655f
332e657865'
}
])this example has a submodule that is a detached digest, which is the hash of a Claims-Set sent outside this token 📦. it's also an example of a token from an attestation hardware block 🔧.
/ This example contains a submodule that is a detached digest, /
/ which is the hash of a Claims-Set conveyed outside this /
/ token. It is also an example of a token from an attestation /
/ hardware block. /
{
/ eat_nonce / 10: h'3515744961254b41a6cf9c02',
/ ueid / 256: h'0198f50a4ff6c05861c8860d13a638ea',
/ oemid / 258: 64242, / Private Enterprise Number /
/ oemboot / 262: true,
/ dbgstat / 263: 3, / disabled-permanently /
/ hwversion / 260: [ "3.1", 1 ], / multipartnumeric /
/ submods/ 266: {
"TEE": [ / detached digest submod /
-16, / SHA-256 /
h'ab86f765643aabfd09c8
4eebe150b7f61bc24804
cee75e90c5f99cb850fe
808f'
]
}
}A.2.3. JSON-Encoded Detached EAT Bundle 🧃
in this bundle, there are two detached Claims-Sets: "Audio Subsystem" and "Graphics Subsystem" 🎧🖥️. the JWT at the start of the bundle has detached signature submodules with hashes that cover these two Claims-Sets. the JWT itself is protected using the Hashed Message Authentication Code (HMAC) with a key of "xxxxxx" 🔑.
the lines in this example are wrapped per [RFC8792] ✂️.
=============== NOTE: '\' line wrapping per RFC 8792 ================
[
[
"JWT",
"eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJlYXRfbm9uY2UiOiJ5dT\
c2Tk44SXVWNmUiLCJzdWJtb2RzIjp7IkF1ZGlvIFN1YnN5c3RlbSI6WyJESUdFU1QiLF\
siU0hBLTI1NiIsIkZSRW4yVlR3aTk5cWNNRVFzYmxtTVFnM2I1b2ZYUG5OM1BJYW5CME\
5RT3MiXV0sIkdyYXBoaWNzIFN1YnN5c3RlbSI6WyJESUdFU1QiLFsiU0hBLTI1NiIsIk\
52M3NqUVU3Q1Z0RFRka0RTUlhWcFZDNUNMVFBCWmVQWWhTLUhoVlZWMXMiXV19fQ.FYs\
7R-TKhgAk85NyCOPQlbtGGjFM_3chnhBEOuM6qCo"
],
{
"Audio Subsystem" : "ewogICAgImVhdF9ub25jZSI6ICJsSStJWU5FNlJ\
qNk8iLAogICAgInVlaWQiOiAiQWROSlU0b1lYdFVwQStIeDNqQTcvRFEiCiAgICAib2V\
taWQiOiAiaVVXdCIsCiAgICAib2VtYm9vdCI6IHRydWUsIAogICAgInN3bmFtZSI6ICJ\
BdWRpbyBQcm9jZXNzb3IgT1MiCn0K",
"Graphics Subsystem" : "ewogICAgImVhdF9ub25jZSI6ICJZWStJWU5F\
NlJqNk8iLAogICAgInVlaWQiOiAiQWVUTUlRQ1NVMnhWQmtVdGlndHU3bGUiCiAgICAi\
b2VtaWQiOiA3NTAwMCwKICAgICJvZW1ib290IjogdHJ1ZSwgCiAgICAic3duYW1lIjog\
IkdyYXBoaWNzIE9TIgp9Cg"
}
]Appendix B. UEID Design Rationale 🦋
B.1. Collision Probability
this one's about figuring out the chance of a collision for type 0x01 UEIDs 🎲, based on the total possible entity population and how many entities live in a particular entity management database.
we look at three different-sized databases 📊. the number of devices per person roughly models non-personal devices like traffic lights, devices in stores they shop in, facilities they work in, and so on, even counting individual light bulbs 💡. a device may have individually attested subsystems, like parts of a car or a mobile phone 🚗. it's assumed that the largest database will have at most 10% of the world's population of devices. and yes, databases that handle more than a trillion records exist today 🤯.
the trillion-record database size models an easy-to-imagine reality over the next decades 🔮. the quadrillion-record database is roughly at the limit of what's imaginable and should probably be accommodated 🌈. the 100 quadrillion database is highly speculative 👽, maybe involving nanorobots for every person, livestock animals, and domesticated birds 🐔. it's included just to round out the analysis.
note that the items counted here certainly do not have IP addresses and are not individually connected to the network 🙅♀️. they may be connected to internal buses, via serial links, via Bluetooth, and so on. this is not the same problem as sizing IP addresses.
+=========+===========+=============+==========+=================+
| People | Devices/ | Subsystems/ | Database | Database Size |
| | Person | Device | Portion | |
+=========+===========+=============+==========+=================+
| 10 | 100 | 10 | 10% | trillion |
| billion | | | | (10^12) |
+---------+-----------+-------------+----------+-----------------+
| 10 | 100,000 | 10 | 10% | quadrillion |
| billion | | | | (10^15) |
+---------+-----------+-------------+----------+-----------------+
| 100 | 1,000,000 | 10 | 10% | 100 quadrillion |
| billion | | | | (10^17) |
+---------+-----------+-------------+----------+-----------------+Table 5: Entity Database Size Examples
this is kinda like the Birthday Problem 🎂, where m is the number of possible birthdays (always 365) and k is the number of people. it's also kinda like the Birthday Attack, where collisions of the output of hash functions are considered 🔐.
the proper formula for the collision calculation is:
p = 1 - e^{-k^2/(2n)}for this calculation:
p: Collision probabilityn: Total possible populationk: Actual population
however, for the very large values involved here, this formula needs floating-point precision higher than what's commonly available in calculators and software 🧮, so this simple approximation is used. see [BirthdayAttack].
p = k^2 / 2nfor this calculation:
p: Collision probabilityn: Total population based on number of bits in UEIDk: Population in a database
+=====================+==============+==============+==============+
| Database Size | 128-bit UEID | 192-bit UEID | 256-bit UEID |
+=====================+==============+==============+==============+
| trillion (10^12) | 2 * 10^-15 | 8 * 10^-35 | 5 * 10^-55 |
+---------------------+--------------+--------------+--------------+
| quadrillion (10^15) | 2 * 10^-09 | 8 * 10^-29 | 5 * 10^-49 |
+---------------------+--------------+--------------+--------------+
| 100 quadrillion | 2 * 10^-05 | 8 * 10^-25 | 5 * 10^-45 |
| (10^17) | | | |
+---------------------+--------------+--------------+--------------+Table 6: UEID Size Options
next, to work out the probability of a collision happening in one year's operation of a database, we assume the database size is in a steady state and that 10% of the database changes per year 📅. for example, a trillion record database would have 100 billion states per year. each of those states has the above calculated probability of a collision.
this assumption is a worst-case scenario 😬, since it assumes each state of the database is completely independent from the previous state. in reality, that's unlikely, since state changes will be the addition or deletion of a few records.
the following table gives the time interval until there's a probability of a collision. it's based on there being one tenth of the number of states per year as the number of records in the database ⏳.
t = 1 / ((k / 10) * p)for this calculation:
t: Time until a collisionp: Collision probability for UEID sizek: Database size
+=====================+==============+==============+==============+
| Database Size | 128-bit UEID | 192-bit UEID | 256-bit UEID |
+=====================+==============+==============+==============+
| trillion (10^12) | 60,000 years | 10^24 years | 10^44 years |
+---------------------+--------------+--------------+--------------+
| quadrillion (10^15) | 8 seconds | 10^14 years | 10^34 years |
+---------------------+--------------+--------------+--------------+
| 100 quadrillion | 8 | 10^11 years | 10^31 years |
| (10^17) | microseconds | | |
+---------------------+--------------+--------------+--------------+Table 7: UEID Collision Probability
clearly, 128 bits is enough for the near future 💖, which is why type 0x01 UEIDs have a minimum of 128 bits. there's no need for 256 bits today, since quadrillion-record databases aren't expected in the near future, and this time-to-collision calculation is a very worst-case scenario 🙈. a future update of the standard may increase the requirement to 256 bits, so implementations need to be able to receive 256-bit UEIDs 🫶.
B.2. No Use of UUID ✨
okay bestie, a UEID is not a Universally Unique Identifier (UUID) [RFC9562] 🫶 and that's a conscious choice, not an accident. here's why 💅
UUIDs are limited to 128 bits, which might not be enough for some future use cases 😬
today, cryptographic-quality random numbers are available from common computing platforms 🎲 hardware randomness sources showed up in CPUs between 2010 and 2015, and operating systems and cryptographic libraries actually use that hardware now. so protocols don't really need to build random numbers out of multiple sources on their own. it's giving "already handled" 💖
version 4 UUIDs do allow for those cryptographic-quality random numbers, but they get shoved into the overall UUID structure of time and clock values 🕰️ and honestly that structure is of no value here, yet it still adds complexity 🙄 it also slightly reduces the number of actual bits with entropy. lowkey wasteful 💀
the UUID design lets you build a unique identifier by combining several identifiers that separately aren't unique enough 🧩 the UEID design philosophy assumes that kind of combining is no longer needed, mostly because cryptographic-quality random number generators are readily available. so hardware, software, and/or manufacturing processes can implement UEID in a simple and direct way ✨ clean and iconic 👑
also, a type 2 UEID (EUI/MAC) is only 7 bytes, whereas a UUID is 16 🌸
Appendix C. EAT Relation to IEEE.802.1AR Secure Device Identity (DevID) 🧚
this section is about the different ways an IEEE Initial Device Identifier (IDevID) [IEEE.802.1AR] connects to EAT, especially to UEID and SUEID 🦋
[IEEE.802.1AR] is all about an implementation called a "DevID Module" 🔐 it explains how IDevIDs and LDevIDs get stored, protected, and accessed using a DevID Module. it also defines a particular level of defense against attack that should be achieved before something counts as a DevID 🛡️ the goal is that IDevIDs and LDevIDs can be used with any network protocol or message format. in those protocols and message formats, the DevID secret is used to sign a nonce or something similar, to prove the DevID certificates belong to the device 💜
EAT, by contrast, standardizes a message format that gets sent to a relying party. that's the exact thing [IEEE.802.1AR] does not define 📨 EAT also doesn't give details on how keys, data, and such get stored, protected, and accessed. it's meant to work with a whole variety of on-device implementations, from minimal protection of assets all the way up to the highest levels of asset protection 🌈 it doesn't define any particular level of defense against attack. instead, it provides a set of security considerations ⚠️
EAT and DevID can be viewed as complimentary when used together, or as competing to provide a device identity service 🤼♀️ (main character energy for both of them ngl 💅)
C.1. DevID Used with EAT 🎀
like we said, EAT standardizes a message format, but [IEEE.802.1AR] doesn't. and the other way around: EAT doesn't define a device implementation, but DevID does 🧠
so EAT can be the message format that a DevID is used with 💫 the DevID secret becomes the attestation key used to sign EATs, and the DevID and its certificate chain become the endorsement sent to the verifier 📜
in this case, the EAT and the DevID are both likely to provide a device identifier (for example, a serial number) 🔢 in the EAT, that's the UEID (or SUEID). in the DevID (used as an endorsement), it's a device serial number included in the subject field of the DevID certificate. for this use, it's a good idea for the serial numbers to be the same, or for the UEID to be a hash of the DevID serial number 🫶
C.2. How EAT Provides an Equivalent Secure Device Identity 🪞
the UEID, SUEID, and other claims such as OEM ID are equivalent to the secure device identity that goes into the subject field of a DevID certificate 🍓 these EAT claims can represent all the same fields and values that can be put in a DevID certificate subject. and EAT explicitly and carefully defines a variety of useful claims 💖
EAT secures the conveyance of these claims by having them signed on the device by the attestation key when the EAT is generated 🔏 EAT also signs the nonce that gives freshness at that time. because these claims get signed for every EAT generated, they can include things that vary over time, such as GPS location 🛰️ wild and so useful 😭
DevID secures the device identity fields by embedding them in a certificate and signing it. the certificate is created once during manufacturing and stays unchanged 🏭
so in one case, the signing of the identity happens on the device, and in the other case, it happens in a manufacturing facility 🏢 however, in both cases, the signing of the nonce that proves the binding to the actual device happens on the device 🙌
while EAT doesn't specify how the signing keys, signature process, and storage of the identity values should be secured against attack, an EAT implementation may have equal defenses against attack 🔒 one reason EAT uses CBOR is because it's simple enough that a basic EAT implementation can be built entirely in hardware 🔧 this lets EAT be implemented with the strongest defenses possible 💪
C.3. An X.509 Format EAT 📃
it's possible to define a way to encode EAT claims in an X.509 certificate 🎀 for example, the EAT claims might be mapped to X.509 v3 extensions. it's even possible to stuff a whole CBOR-encoded unsigned EAT token into an X.509 certificate 😳
if that X.509 certificate is an IDevID or LDevID, it becomes another way to use EAT and DevID together 🤝
note that the DevID must still be used with an authentication protocol that has a nonce or equivalent. the EAT here is not being used as the protocol to interact with the relying party 🚫
C.4. Device Identifier Permanence 🔐
when it comes to permanence, an IDevID is similar to a UEID in that they don't change over the life of the device. they only cease to exist when the device is destroyed 💀
an SUEID is similar to an LDevID. they change on device life-cycle events 🔄
[IEEE.802.1AR] describes much of this permanence as resistant to attacks that try to change the ID 🛡️ IDevID permanence can be described this way because [IEEE.802.1AR] is oriented around the definition of an implementation with a particular level of defense against attack.
EAT isn't defined around a particular implementation, and it must work on a range of devices that have a range of defenses against attack 🌍 so for EAT, permanence isn't defined in terms of resistance to attacks. instead, it's defined in the context of operational functionality and the device life cycle ✨
Appendix D. CDDL for CWT and JWT 📚
[RFC8392] was published before CDDL was available, so it's specified in prose, not CDDL 🙈 in the following example, CDDL specifies CWT as it's needed to complete this specification. this CDDL also covers the Claims-Set for JWT 💜
note that Section 4.3.1 requires that the "iat" claim be the type ~time-int (Section 7.2.1), not the type ~time when it's used in an EAT, because floating-point values aren't allowed for the "iat" claim in EAT 🙅♀️
the COSE-related types in this CDDL are defined in [RFC9052] 🧁
this, however, is not a normative or standard definition of CWT or JWT in CDDL 🚨 the prose in CWT and JWT remains the normative definition. see also [UCCS] 📖
Claims-Set = {
* $$Claims-Set-Claims
* Claim-Label .feature "extended-claims-label" => any
}
Claim-Label = int / text
string-or-uri = text
$$Claims-Set-Claims //= ( iss-claim-label => string-or-uri )
$$Claims-Set-Claims //= ( sub-claim-label => string-or-uri )
$$Claims-Set-Claims //= ( aud-claim-label => string-or-uri )
$$Claims-Set-Claims //= ( exp-claim-label => ~time )
$$Claims-Set-Claims //= ( nbf-claim-label => ~time )
$$Claims-Set-Claims //= ( iat-claim-label => ~time )
$$Claims-Set-Claims //= ( cti-claim-label => bytes )
iss-claim-label = JC<"iss", 1>
sub-claim-label = JC<"sub", 2>
aud-claim-label = JC<"aud", 3>
exp-claim-label = JC<"exp", 4>
nbf-claim-label = JC<"nbf", 5>
iat-claim-label = JC<"iat", 6>
cti-claim-label = CBOR-ONLY<7> ; jti in JWT: different name and text
JSON-ONLY<J> = J .feature "json"
CBOR-ONLY<C> = C .feature "cbor"
JC<J,C> = JSON-ONLY<J> / CBOR-ONLY<C>
; A JWT message is either a JSON Web Signature (JWS) or a JSON Web
; Encryption (JWE) in compact serialization form with the payload
; as a Claims-Set. Compact serialization is the protected headers,
; payload, and signature that are each b64url-encoded and separated
; by a ".". This CDDL simply matches the top-level syntax of a JWS
; or JWE as it is not possible to do more in CDDL.
JWT-Message =
text .regexp "[A-Za-z0-9_-]+\\.[A-Za-z0-9_-]+\\.[A-Za-z0-9_-]+"
; Note that the payload of a JWT is defined in the CDDL description
; of claims-set. That definition is common to CBOR and JSON.
; This is some CDDL describing a CWT at the top level. This is
; not normative. RFC 8392 is the normative definition of CWT.
CWT-Messages = CWT-Tagged-Message / CWT-Untagged-Message
; The payload of the COSE_Message is always a Claims-Set.
; The contents of a CWT tag must always be a COSE tag.
CWT-Tagged-Message = #6.61(COSE_Tagged_Message)
; An untagged CWT may be a COSE tag or not.
CWT-Untagged-Message = COSE_MessagesAppendix E. New Claim Design Considerations 🧸
the following are design considerations that may be helpful to keep in mind when creating new EAT claims 💭 this is the product of discussion in the RATS Working Group 🤓
EAT reuses the CWT and JWT claims registries. there's no registry exclusively for EAT claims. this isn't an update to the expert review criteria for the JWT and CWT claims registries, because that would be an overreach for this document 🙅♀️
E.1. Interoperability and Relying Party Orientation 🌐
it's a broad goal that EATs can be processed by relying parties in a general way, no matter the type, manufacturer, or technology of the device they came from 🌸 it's also a goal that there be general-purpose verification implementations that can verify tokens for large numbers of use cases, with special cases and configurations for different device types. this is a goal of interoperability of the semantics of claims themselves, not just of the signing, encoding, and serialization formats 🧠✨
this is a lofty goal and really difficult to achieve broadly, since it requires careful definition of claims in a technology-neutral way 😮💨 sometimes it'll be hard to design a claim that can represent the semantics of data from very different device types. however, the goal remains even when it's difficult 🫡
E.2. Operating System and Technology Neutral 🖥️
claims should be defined so they're not specific to an operating system. they should be applicable to multiple large high-level operating systems from different vendors, as well as to multiple small embedded operating systems from multiple vendors, and everything in between 🌈
claims should not be defined so they're specific to a software environment or programming language 🙅♀️
claims should not be defined so they're specific to a chip or particular hardware. for example, they shouldn't just be the contents of some HW status register, since it's unlikely that the same HW status register with the same bits exists on a chip from a different manufacturer 🔧
the boot and debug state claims in this document are an example of a claim that's been defined in this neutral way 🦋
E.3. Security Level Neutral 🛡️
okay so some use cases are gonna have EATs generated by the most secure hardware and software that exists 🔐 Secure Elements and smart cards are the big examples. BUT EAT is meant for low-security use cases just as much as high-security ones 💅 like, an app on a phone can generate EATs all on its own, no cap.
Claims should be defined and registered based on whether they are useful and interoperable, not based on security level ✨ The vibe is useful + interoperable = in. So there should be no exclusion of claims just because they only show up in low-security environments 🙅♀️
E.4. Reuse of Extant Data Formats 🧩
Where possible, claims should use data items, identifiers, and formats that are already standardized 🤓 This takes advantage of all the expertise that went into making those formats, and it makes interoperability way better.
Often, extant claims won't be defined in an encoding or serialization format that EAT uses. It is preferred to define a CBOR and JSON encoding for them, so EAT implementations don't need a whole plethora of encoders and decoders for serialization formats 🎀 (a plethora is basically a giant pile, bestie 😭)
In some cases, it may be better to use the encoding and serialization as is 🌸 For example, signed X.509 certificates and Certificate Revocation Lists (CRLs) can be carried as is in a byte string. This keeps interoperability with the massive infrastructure that already exists for creating and processing X.509 certificates and CRLs. Iconic behavior 👑
E.5. Proprietary Claims 🦋
It is not always possible or convenient to hit all the goals above, so using proprietary claims is an option. It's giving flexible 🫶
For example, a device manufacturer may generate a token with proprietary claims that are only meant to be verified by a service that the same manufacturer runs. This is a supported use case, periodt 💖
In many cases proprietary claims will be the easiest and most obvious route ✨ however, for better interoperability, using general standardized claims is preferred. Choose wisely, main character 🧚
Appendix F. Endorsements and Verification Keys 🔑
The verifier must possess the correct key when it performs the cryptographic part of an EAT verification (e.g., verifying the COSE/JOSE signature). This section describes several ways to identify the verification key. There is not one standard method, so don't panic if you see different approaches 🙈
The verification key itself may be a public key, a symmetric key, or something complicated in the case of a scheme such as Direct Anonymous Attestation (DAA) 🌈
RATS Architecture [RFC9334] describes what is called an endorsement 💜 This is an input to the verifier that is usually the basis of the trust placed in an EAT and the attester that generated it. It may contain the public key for verification of the signature on the EAT, and it may contain implied claims, i.e., those that are passed on to the relying party in attestation results.
There is not yet any standard format for an endorsement. One format that may be used for an endorsement is an X.509 certificate. Endorsement data such as reference values and implied claims can be carried in X.509 v3 extensions. In this use, the public key in the X.509 certificate becomes the verification key, so identifying the endorsement also identifies the verification key. Two-for-one, bestie 🍓
The verification key identification and establishment of trust in the EAT and the attester may also be done by some other means than an endorsement 🌸
For the components (attester, verifier, relying party, etc.) of a particular end-to-end attestation system to reliably interoperate, its definition should specify how the verification key is identified. Usually, this will be in the profile document for that particular attestation system 📄
See also the security considerations in Section 9.6. Seriously, read them 😭🔥
F.1. Identification Methods 🗂️
Following is a list of possible methods of key identification. A specific attestation system may employ any one of these, or one not listed here ✨
The following assumes endorsements are X.509 certificates or equivalent, and thus does not mention or define any identifier for endorsements in other formats. If such an endorsement format is created, new identifiers for them will also need to be created 🧠
F.1.1. COSE/JWS Key ID 🪪
The COSE standard header parameter for Key ID (kid) may be used; see [RFC9052] and [RFC7515].
COSE leaves the semantics of the key ID open-ended 🌀 It could be a record locator in a database, a hash of a public key, an input to a Key Derivation Function (KDF), an Authority Key Identifier (AKI) for an X.509 certificate, or something else entirely. The profile document should specify what the key ID's semantics are 💅
F.1.2. JWS and COSE X.509 Header Parameters 📜
COSE X.509 [RFC9360] and JSON Web Signature [RFC7515] define several header parameters (x5t, x5u,...) for referencing or carrying X.509 certificates, and any of them may be used 🎀
The X.509 certificate may be an endorsement, and thus carries additional input to the verifier. It may also just be an X.509 certificate and not an endorsement. The same header parameters are used in both cases, and it is up to the attestation system design and the verifier to figure out which one it is 🕵️♀️
F.1.3. CBOR Certificate COSE Header Parameters 🧊
Compressed X.509 and CBOR Native certificates are defined by CBOR Certificates [CBOR.Certs]. These are semantically compatible with X.509, so they can be used as an equivalent to X.509 as described above ✨
These are identified by their own header parameters (c5t, c5u, etc.).
F.1.4. Claim-Based Key Identification 🆔
For some attestation systems, a claim may be reused as a key identifier. For example, the UEID uniquely identifies the entity, and therefore it can work well as a key identifier or endorsement identifier 🦄
An advantage of this is that key identification needs no additional bytes in the EAT, which makes the EAT smaller. Lean and mean 💪
A disadvantage of this is that the unverified EAT must be substantially decoded to get the identifier, since the identifier is in the COSE/JOSE payload and not in the headers 😩
Contributors 🎀
Many thanks to the following for their contributions to earlier draft versions of this document 🫶
Henk Birkholz
Fraunhofer SIT
Email: henk.birkholz@sit.fraunhofer.de
Thomas Fossati
Arm Limited
Email: thomas.fossati@arm.com
Miguel Ballesteros
Michael Richardson
Sandelman Software Works
Email: mcr+ietf@sandelman.ca
Patrick Uiterwijk
Mathias Brossard
Hannes Tschofenig
Arm Limited
Email: hannes.tschofenig@arm.com
Paul CrowleyAuthors' Addresses 💌
Laurence Lundblade
Security Theory LLC
Email: lgl@securitytheory.com
Giridhar Mandyam
Email: giridhar.mandyam@gmail.com
Jeremy O'Donoghue
Qualcomm Technologies Inc.
279 Farnborough Road
Farnborough
GU14 7LS
United Kingdom
Phone: +44 1252 363189
Email: jodonogh@qti.qualcomm.com
Carl Wallace
Red Hound Software, Inc.
Email: carl@redhoundsoftware.com📦 where this page comes from
- ietf: rfc9711 2025-04
- 💖 girlified by claude-haiku-5-5 on : a language model's rewrite of the original, under its terms