The Entity Attestation Token (EAT)

1.1. CWT, JWT and UCCS

For flexibility and ease of imlpementation in a wide variety of environments, EATs can be either CBOR [RFC7049] or JSON [ECMAScript] format. This specification simultaneously describes both formats.¶

An EAT is either a CWT as defined in [RFC8392], a UCCS as defined in [UCCS.Draft], or a JWT as defined in [RFC7519]. This specification extends those specifications with additional claims for attestation.¶

The identification of a protocol element as an EAT, whether CBOR or JSON format, follows the general conventions used by CWT, JWT and UCCS. Largely this depends on the protocol carrying the EAT. In some cases it may be by content type (e.g., MIME type). In other cases it may be through use of CBOR tags. There is no fixed mechanism across all use cases.¶

1.2. CDDL

This specification uses CDDL, [RFC8610], as the primary formalism to define each claim. The implementor then interprets the CDDL to come to either the CBOR [RFC7049] or JSON [ECMAScript] representation. In the case of JSON, Appendix E of [RFC8610] is followed. Additional rules are given in Section 5.3.2 of this document where Appendix E is insufficient. (Note that this is not to define a general means to translate between CBOR and JSON, but only to define enough such that the claims defined in this document can be rendered unambiguously in JSON).¶

The CWT specification was authored before CDDL was available and did not use it. This specification includes a CDDL definition of most of what is described in [RFC8392].¶

1.3. Entity Overview

An "entity" can be any device or device subassembly ("submodule") that can generate its own attestation in the form of an EAT. The attestation should be cryptographically verifiable by the EAT consumer. An EAT at the device-level can be composed of several submodule EAT's. It is assumed that any entity that can create an EAT does so by means of a dedicated root-of-trust (RoT).¶

Modern devices such as a mobile phone have many different execution environments operating with different security levels. For example, it is common for a mobile phone to have an "apps" environment that runs an operating system (OS) that hosts a plethora of downloadable apps. It may also have a TEE (Trusted Execution Environment) that is distinct, isolated, and hosts security-oriented functionality like biometric authentication. Additionally, it may have an eSE (embedded Secure Element) - a high security chip with defenses against HW attacks that can serve as a RoT. This device attestation format allows the attested data to be tagged at a security level from which it originates. In general, any discrete execution environment that has an identifiable security level can be considered an entity.¶

1.4. EAT Operating Models

TODO: Rewrite (or eliminate) this section in light of the RATS architecture draft.¶

At least the following three participants exist in all EAT operating models. Some operating models have additional participants.¶

The Entity.

This is the phone, the IoT device, the sensor, the sub-assembly or such that the attestation provides information about.¶

The Manufacturer.

The company that made the entity. This may be a chip vendor, a circuit board module vendor or a vendor of finished consumer products.¶

The Relying Party.

The server, service or company that makes use of the information in the EAT about the entity.¶

In all operating models, the manufacturer provisions some secret attestation key material (AKM) into the entity during manufacturing. This might be during the manufacturer of a chip at a fabrication facility (fab) or during final assembly of a consumer product or any time in between. This attestation key material is used for signing EATs.¶

In all operating models, hardware and/or software on the entity create an EAT of the format described in this document. The EAT is always signed by the attestation key material provisioned by the manufacturer.¶

In all operating models, the relying party must end up knowing that the signature on the EAT is valid and consistent with data from claims in the EAT. This can happen in many different ways. Here are some examples.¶

• The EAT is transmitted to the relying party. The relying party gets corresponding key material (e.g. a root certificate) from the manufacturer. The relying party performs the verification.¶
• The EAT is transmitted to the relying party. The relying party transmits the EAT to a verification service offered by the manufacturer. The server returns the validated claims.¶
• The EAT is transmitted directly to a verification service, perhaps operated by the manufacturer or perhaps by another party. It verifies the EAT and makes the validated claims available to the relying party. It may even modify the claims in some way and re-sign the EAT (with a different signing key).¶

All these operating models are supported and there is no preference of one over the other. It is important to support this variety of operating models to generally facilitate deployment and to allow for some special scenarios. One special scenario has a validation service that is monetized, most likely by the manufacturer. In another, a privacy proxy service processes the EAT before it is transmitted to the relying party. In yet another, symmetric key material is used for signing. In this case the manufacturer should perform the verification, because any release of the key material would enable a participant other than the entity to create valid signed EATs.¶

1.5. What is Not Standardized

The following is not standardized for EAT, just the same they are not standardized for CWT or JWT.¶

3.1. Token ID Claim (cti and jti)

CWT defines the "cti" claim. JWT defines the "jti" claim. These are equivalent to each other in EAT and carry a unique token identifier as they do in JWT and CWT. They may be used to defend against re use of the token but are distinct from the nonce that is used by the relying party to guarantee freshness and defend against replay.¶

3.2. Timestamp claim (iat)

The "iat" claim defined in CWT and JWT is used to indicate the date-of-creation of the token, the time at which 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 their age is older than the "iat" timestamp.¶

CWT allows the use floating-point for this claim. EAT disallows the use of floating-point. No token may contain an iat claim in float-point format. Any recipient of a token with a floating-point format iat claim may consider it an error. A 64-bit integer representation of epoch time can represent a range of +/- 500 billion years, so the only point of a floating-point timestamp is to have precession greater than one second. This is not needed for EAT.¶

3.3. Nonce Claim (nonce)

All EATs should have a nonce to prevent replay attacks. The nonce is generated by the relying party, the end consumer of the token. It is conveyed to the entity over whatever transport is in use before the token is generated and then included in the token as the nonce claim.¶

This documents the nonce claim for registration in the IANA CWT claims registry. This is equivalent to the JWT nonce claim that is already registered.¶

The nonce must be at least 8 bytes (64 bits) as fewer are unlikely to be secure. A maximum of 64 bytes is set to limit the memory a constrained implementation uses. This size range is not set for the already-registered JWT nonce, but it should follow this size recommendation when used in an EAT.¶

Multiple nonces are allowed to accommodate multistage verification and consumption.¶

{::include cddl/nonce.cddl}

3.4. Universal Entity ID Claim (ueid)

UEID's identify individual manufactured entities / devices such as a mobile phone, a water meter, a Bluetooth speaker or a networked security camera. It may identify the entire device or a submodule or subsystem. It does not identify types, models or classes of devices. It is akin to a serial number, though it does not have to be sequential.¶

UEID's must be universally and globally unique across manufacturers and countries. UEIDs must also be unique across protocols and systems, as tokens are intended to be embedded in many different protocols and systems. No two products anywhere, even in completely different industries made by two different manufacturers in two different countries should have the same UEID (if they are not global and universal in this way, then relying parties receiving them will have to track other characteristics of the device to keep devices distinct between manufacturers).¶

There are privacy considerations for UEID's. See Section 7.1.¶

The UEID should be permanent. It should never change for a given device / entity. In addition, it should not be reprogrammable. UEID's are variable length. All implementations MUST be able to receive UEID's that are 33 bytes long (1 type byte and 256 bits). The recommended maximum sent is also 33 bytes.¶

When the entity constructs the UEID, the first byte is a type and the following bytes the ID for that type. Several types are allowed to accommodate different industries and different manufacturing processes and to give options to avoid paying fees for certain types of manufacturer registrations.¶

Creation of new types requires a Standards Action [RFC8126].¶

UEID's are not designed for direct use by humans (e.g., printing on the case of a device), so no textual representation is defined.¶

The consumer (the relying party) of a UEID MUST treat a UEID as a completely opaque string of bytes and not make any use of its internal structure. For example, they should not use the OUI part of a type 0x02 UEID to identify the manufacturer of the device. Instead they should use the oemid claim that is defined elsewhere. The reasons for this are:¶

• UEIDs types may vary freely from one manufacturer to the next.¶
• New types of UEIDs may be created. For example, a type 0x07 UEID may be created based on some other manufacturer registration scheme.¶
• Device manufacturers are allowed to change from one type of UEID to another anytime they want. For example, they may find they can optimize their manufacturing by switching from type 0x01 to type 0x02 or vice versa. The main requirement on the manufacturer is that UEIDs be universally unique.¶

{::include cddl/ueid.cddl}

3.5. Origination Claim (origination)

TODO: this claim is likely to be dropped in favor of Endorsement identifier and locators.¶

This claim describes the parts of the device or entity that are creating the EAT. Often it will be tied back to the device or chip manufacturer. The following table gives some examples:¶

TODO: consider a more structure approach where the name and the URI and other are in separate fields.¶

TODO: This needs refinement. It is somewhat parallel to issuer claim in CWT in that it describes the authority that created the token.¶

{::include cddl/origination.cddl}

3.6. OEM Identification by IEEE (oemid)

The IEEE operates a global registry for MAC addresses and company IDs. This claim uses that database to identify OEMs. The contents of the claim may be either an IEEE MA-L, MA-M, MA-S or an IEEE 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 similarly is a 28-bit value uses as the first part of a MAC address, and MA-S, formerly known as OUI-36, a 36-bit value. Many companies already have purchased one of these. A CID is also a 24-bit value from the same space as an MA-L, but not for use as a MAC address. IEEE has published Guidelines for Use of EUI, OUI, and CID [OUI.Guide] and provides a lookup services [OUI.Lookup]¶

Companies that have more than one of these IDs or MAC address blocks should pick one and prefer that for all their devices.¶

Commonly, these are expressed in Hexadecimal Representation [IEEE.802-2001] also called the Canonical format. When this claim is encoded the order of bytes in the bstr are the same as the order in the Hexadecimal Representation. For example, an MA-L like "AC-DE-48" would be encoded in 3 bytes with values 0xAC, 0xDE, 0x48. For JSON encoded tokens, this is further base64url encoded.¶

{::include cddl/oemid.cddl}

3.7. Hardware Version Claims (hardware-version-claims)

The hardware version can be claimed at three different levels, the chip, the circuit board and the final device assembly. An EAT can include any combination these claims.¶

The hardware version is a simple text string the format of which is set by each manufacturer. The structure and sorting order of this text string can be specified using the version-scheme item from CoSWID [CoSWID].¶

The hardware version can also be given by a 13-digit European Article Number [EAN-13]. An EAN-13 is also known as an International Article Number or most commonly as a bar code. This claim is the ASCII text representation of actual digits often printed with a bar code. Use of this claim must comply with the EAN allocation and assignment rules. For example, this requires the manufacturer to obtain a manufacture code from GS1.¶

Both the simple version string and EAN-13 versions may be included for the same hardware.¶

{::include cddl/hardware-version.cddl}

3.8. Software Description and Version

TODO: Add claims that reference CoSWID.¶

3.9. The Security Level Claim (security-level)

This claim characterizes the device/entity ability to defend against attacks aimed at capturing the signing key, forging claims and at forging EATs. This is done by defining four security levels as described below. This is similar to the key protection types defined by the Fast Identity Online (FIDO) Alliance [FIDO.Registry]).¶

These claims describe security environment and countermeasures available on the end-entity / client device where the attestation key reside and the claims originate.¶

1 - Unrestricted

There is some expectation that implementor will protect the attestation signing keys at this level. Otherwise the EAT provides no meaningful security assurances.¶

2- Restricted

Entities at this level should not be general-purpose operating environments that host features such as app download systems, web browsers and complex productivity applications. It is akin to the Secure Restricted level (see below) without the security orientation. Examples include a Wi-Fi subsystem, an IoT camera, or sensor device.¶

3 - Secure Restricted

Entities at this level must meet the criteria defined by FIDO Allowed Restricted Operating Environments [FIDO.AROE]. Examples include TEE's and schemes using virtualization-based security. Like the FIDO security goal, security at this level is aimed at defending well against large-scale network / remote attacks against the device.¶

4 - Hardware

Entities at this level must include substantial defense against physical or electrical attacks against the device itself. It is assumed any potential attacker has captured the device and can disassemble it. Example include TPMs and Secure Elements.¶

The entity should claim the highest security level it achieves and no higher. This set is not extensible so as to provide a common interoperable description of security level to the relying party. If a particular implementation considers this claim to be inadequate, it can define its own proprietary claim. It may consider including both this claim as a coarse indication of security and its own proprietary claim as a refined indication.¶

This claim is not intended as a replacement for a proper end-device security certification schemes such as those based on FIPS 140 [FIPS-140] or those based on Common Criteria [Common.Criteria]. The claim made here is solely a self-claim made by the Entity Originator.¶

{::include cddl/security-level.cddl}

3.10. Secure Boot Claim (secure-boot)

The value of true indicates secure boot is enabled. Secure boot is considered enabled when base software, the firmware and operating system, are under control of the entity manufacturer identified in the oemid claimd described in Section 3.6. This may because the software is in ROM or because it is cryptographically authenticated or some combination of the two or other.¶

{::include cddl/secure-boot.cddl}

3.11. Debug Status Claim (debug-status)

This applies to system-wide or submodule-wide debug facilities of the target device / submodule like JTAG and diagnostic hardware built into chips. It applies to any software debug facilities related to root, operating system or privileged software that allow system-wide memory inspection, tracing or modification of non-system software like user mode applications.¶

This characterization assumes that debug facilities can be enabled and disabled in a dynamic way or be disabled in some permanent way such that no enabling is possible. An example of dynamic enabling is one where some authentication is required to enable debugging. An example of permanent disabling is blowing a hardware fuse in a chip. The specific type of the mechanism is not taken into account. For example, it does not matter if authentication is by a global password or by per-device public keys.¶

As with all claims, the absence of the debug level claim means it is not reported. A conservative interpretation might assume the Not Disabled state. It could however be that it is reported in a proprietary claim.¶

This claim is not extensible so as to provide a common interoperable description of debug status to the relying party. If a particular implementation considers this claim to be inadequate, it can define its own proprietary claim. It may consider including both this claim as a coarse indication of debug status and its own proprietary claim as a refined indication.¶

The higher levels of debug disabling requires that all debug disabling of the levels below it be in effect. Since the lowest level requires that all of the target's debug be currently disabled, all other levels require that too.¶

There is no inheritance of claims from a submodule to a superior module or vice versa. There is no assumption, requirement or guarantee that the target of a superior module encompasses the targets of submodules. Thus, every submodule must explicitly describe its own debug state. The verifier or relying party receiving an EAT cannot assume that debug is turned off in a submodule because there is a claim indicating it is turned off in a superior module.¶

An individual target device / submodule may have multiple debug facilities. The use of plural in the description of the states refers to that, not to any aggregation or inheritance.¶

The architecture of some chips or devices may be such that a debug facility operates for the whole chip or device. If the EAT for such a chip includes submodules, then each submodule should independently report the status of the whole-chip or whole-device debug facility. This is the only way the relying party can know the debug status of the submodules since there is no inheritance.¶

{::include cddl/debug-status.cddl}

3.12. Including Keys

An EAT may include a cryptographic key such as a public key. The signing of the EAT binds the key to all the other claims in the token.¶

The purpose for inclusion of the key may vary by use case. For example, the key may be included as part of an IoT device onboarding protocol. When the FIDO protocol includes a pubic key in its attestation message, the key represents the binding of a user, device and relying party. This document describes how claims containing keys should be defined for the various use cases. It does not define specific claims for specific use cases.¶

Keys in CBOR format tokens SHOULD be the COSE_Key format [RFC8152] and keys in JSON format tokens SHOULD be the JSON Web Key format [RFC7517]. These two formats support many common key types. Their use avoids the need to decode other serialization formats. These two formats can be extended to support further key types through their IANA registries.¶

The general confirmation claim format [RFC8747], [RFC7800] may also be used. It provides key encryption. It also allows for inclusion by reference through a key ID. The confirmation claim format may employed in the definition of some new claim for a a particular use case.¶

When the actual confirmation claim is included in an EAT, this document associates no use case semantics other than proof of posession. Different EAT use cases may choose to associate further semantics. The key in the confirmation claim MUST be protected the same as the key used to sign the EAT. That is, the same, equivalent or better hardware defenses, access controls, key generation and such must be used.¶

3.13. The Location Claim (location)

The location claim gives the location of the device entity from which the attestation originates. It is derived from the W3C Geolocation API [W3C.GeoLoc]. The latitude, longitude, altitude and accuracy must conform to [WGS84]. The altitude is in meters above the [WGS84] ellipsoid. The two accuracy values are positive numbers in meters. The heading is in degrees relative to true north. If the device is stationary, the heading is NaN (floating-point not-a-number). The speed is the horizontal component of the device velocity in meters per second.¶

When encoding floating-point numbers half-precision should not be used. It usually does not provide enough precision for a geographic location. It is not a requirement that the receiver of an EAT implement half-precision, so the receiver may not be able to decode the location.¶

The location may have been cached for a period of time before token creation. For example, it might have been minutes or hours or more since the last contact with a GPS satellite. Either the timestamp or age data item can be used to quantify the cached period. The timestamp data item is preferred as it a non-relative time.¶

The age data item can be used when the entity doesn't know what time it is either because it doesn't have a clock or it isn't set. The entity must still have a "ticker" that can measure a time interval. The age is the interval between acquisition of the location data and token creation.¶

See {#locationprivacyconsiderations} below.¶

{::include cddl/location.cddl}

3.14. The Uptime Claim (uptime)

The "uptime" claim contains a value that represents the number of seconds that have elapsed since the entity or submod was last booted.¶

{::include cddl/uptime.cddl}

3.15. The Intended Use Claim (intended-use)

EAT's may be used in the context of several different applications. The intended-use claim provides an indication to an EAT consumer about the intended usage of the token. This claim can be used as a way for an application using EAT to internally distinguish between different ways it uses EAT.¶

1 - Generic

Generic attestation describes an application where the EAT consumer requres 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 intended-use 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 device security state of the entity prior to provisioning.¶

4 - Certificate Issuance (Certificate Signing Request)

Certifying authorities (CA's) may require attestations prior to the issuance of certificates related to keypairs hosted at the entity. 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) appication. More precisely, a PoP transaction is intended to provide to the recipient cryptographically-verifiable proof that the sender has posession of a key. This kind of attestation may be neceesary to verify the security state of the entity storing the private key used in a PoP application.¶

intended-use = &(
    generic: 1,
    registration: 2,
    provisioning: 3,
    csr: 4,
    pop:  5
)

3.16. The Submodules Part of a Token (submods)

Some devices are complex, having many subsystems or submodules. A mobile phone is a good example. It may have several connectivity submodules for communications (e.g., Wi-Fi and cellular). It may have subsystems for low-power audio and video playback. It may have one or more security-oriented subsystems like a TEE or a Secure Element.¶

The claims for each these can be grouped together in a submodule.¶

The submods part of a token are in a single map/object with many entries, one per submodule. There is only one submods map in a token. It is identified by its specific label. It is a peer to other claims, but it is not called a claim because it is a container for a claim set rather than an individual claim. This submods part of a token allows what might be called recursion. It allows claim sets inside of claim sets inside of claims sets...¶

{::include cddl/submods.cddl}

5.1. Common CDDL Types

time-int is identical to the epoch-based time, but disallows floating-point representation.¶

{::include cddl/common-types.cddl}

5.2. CDDL for CWT-defined Claims

This section provides CDDL for the claims defined in CWT. It is non-normative as [RFC8392] is the authoritative definition of these claims.¶

{::include cddl/cwt.cddl}

5.3. JSON

{::include cddl/json.cddl}

5.4. CBOR

5.5. Collected CDDL

{::include cddl/eat-token.cddl}

6.1. Reuse of CBOR Web Token (CWT) Claims Registry

Claims defined for EAT are compatible with those of CWT so the CWT Claims Registry is re used. No new IANA registry is created. All EAT claims should be registered in the CWT and JWT Claims Registries.¶

6.2. Claim Characteristics

The following is design guidance for creating new EAT claims, particularly those to be registered with IANA.¶

Much of this guidance is generic and could also be considered when designing new CWT or JWT claims.¶

6.3. Claims Registered by This Document

• Claim Name: UEID¶
• Claim Description: The Universal Entity ID¶
• JWT Claim Name: N/A¶
• Claim Key: 8¶
• Claim Value Type(s): byte string¶
• Change Controller: IESG¶
• Specification Document(s): this document¶

TODO: add the rest of the claims in here¶

7.1. UEID Privacy Considerations

A UEID is usually not privacy-preserving. Any set of relying parties that receives tokens that happen to be from a single device will be able to know the tokens are all from the same device and be able to track the device. Thus, in many usage situations ueid violates governmental privacy regulation. In other usage situations UEID will not be allowed for certain products like browsers that give privacy for the end user. It will often be the case that tokens will not have a UEID for these reasons.¶

There are several strategies that can be used to still be able to put UEID's in tokens:¶

• The device obtains explicit permission from the user of the device to use the UEID. This may be through a prompt. It may also be through a license agreement. For example, agreements for some online banking and brokerage services might already cover use of a UEID.¶
• The UEID is used only in a particular context or particular use case. It is used only by one relying party.¶
• The device authenticates the relying party and generates a derived UEID just for that particular relying party. For example, the relying party could prove their identity cryptographically to the device, then the device generates a UEID just for that relying party by hashing a proofed relying party ID with the main device UEID.¶

Note that some of these privacy preservation strategies result in multiple UEIDs per device. Each UEID is used in a different context, use case or system on the device. However, from the view of the relying party, there is just one UEID and it is still globally universal across manufacturers.¶

7.2. Location Privacy Considerations

Geographic location is most always considered personally identifiable information. Implementers should consider laws and regulations governing the transmission of location data from end user devices to servers and services. Implementers should consider using location management facilities offered by the operating system on the device generating the attestation. For example, many mobile phones prompt the user for permission when before sending location data.¶

8.1. Key Provisioning

Private key material can be used to sign and/or encrypt the EAT, or can be used 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 the key material in the entity itself. The manfuacturer of any entity that is capable of producing an EAT should take care to ensure that any private key material be suitably protected prior to provisioning the key material in the entity itself. This can require creation of key material in an enclave (see [RFC4949] for definition of "enclave"), secure transmission of the key material from the enclave to the entity using an appropriate protocol, and persistence of the private key material in some form of secure storage to which (preferably) only the entity has access.¶

8.2. Transport Security

As stated in Section 8 of [RFC8392], "The security of the CWT relies upon on the protections offered by COSE". Similar considerations apply to EAT when sent as a CWT. However, EAT introduces the concept of a nonce to protect against replay. Since an EAT may be created by an entity that may not support the same type of transport security as the consumer of the EAT, intermediaries may be required to bridge communications between the entity and consumer. As a result, it is RECOMMENDED that both the consumer create a nonce, and the entity leverage the nonce along with COSE mechanisms for encryption and/or signing to create the EAT.¶

Similar considerations apply to the use of EAT as a JWT. Although the security of a JWT leverages the JSON Web Encryption (JWE) and JSON Web Signature (JWS) specifications, it is still recommended to make use of the EAT nonce.¶

8.3. 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 with an EAT. When multiple consumers are required for verification of an EAT, it is 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, 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 should leverage a secure protocol (e.g.one that uses transport-layer security, i.e. 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 transport security 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.¶

A.1. Very Simple EAT

This is shown in CBOR diagnostic form. Only the payload signed by COSE is shown.¶

{::include cddl/examples/simple.diag}

A.2. Example with Submodules, Nesting and Security Levels

{::include cddl/examples/submods.diag}

B.1. Collision Probability

This calculation is to determine the probability of a collision of UEIDs given the total possible entity population and the number of entities in a particular entity management database.¶

Three different sized databases are considered. The number of devices per person roughly models non-personal devices such as traffic lights, devices in stores they shop in, facilities they work in and so on, even considering individual light bulbs. A device may have individually attested subsystems, for example parts of a car or a mobile phone. It is assumed that the largest database will have at most 10% of the world's population of devices. Note that 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 is imaginable and should probably be accommodated. The 100 quadrillion datadbase is highly speculative perhaps involving nanorobots for every person, livestock animal and domesticated bird. It is included to round out the analysis.¶

Note that the items counted here certainly do not have IP address and are not individually connected to the network. They may be connected to internal buses, via serial links, Bluetooth and so on. This is not the same problem as sizing IP addresses.¶

This is conceptually similar to the Birthday Problem where m is the number of possible birthdays, always 365, and k is the number of people. It is also conceptually similar to 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)}

   p   Collision Probability
   n   Total possible population
   k   Actual population

However, for the very large values involved here, this formula requires floating point precision higher than commonly available in calculators and SW so this simple approximation is used. See [BirthdayAttack].¶

    p = k^2 / 2n

For this calculation:¶

    p  Collision Probability
    n  Total population based on number of bits in UEID
    k  Population in a database

Next, to calculate the probability of a collision occurring in one year's operation of a database, it is assumed that 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 since it assumes that each state of the database is completely independent from the previous state. In reality this is unlikely as state changes will be the addition or deletion of a few records.¶

The following tables gives the time interval until there is a probability of a collision based on there being one tenth the number of states per year as the number of records in the database.¶

  t = 1 / ((k / 10) * p)

  t  Time until a collision
  p  Collision probability for UEID size
  k  Database size

Clearly, 128 bits is enough for the near future thus the requirement that UEIDs be a minimum of 128 bits.¶

There is no requirement for 256 bits today as quadrillion-record databases are not expected in the near future and because this time-to-collision calculation is a very worst case. A future update of the standard may increase the requirement to 256 bits, so there is a requirement that implementations be able to receive 256-bit UEIDs.¶

B.2. No Use of UUID

A UEID is not a UUID [RFC4122] by conscious choice for the following reasons.¶

UUIDs are limited to 128 bits which may not be enough for some future use cases.¶

Today, cryptographic-quality random numbers are available from common CPUs and hardware. This hardware was introduced between 2010 and 2015. Operating systems and cryptographic libraries give access to this hardware. Consequently, there is little need for implementations to construct such random values from multiple sources on their own.¶

Version 4 UUIDs do allow for use of such cryptographic-quality random numbers, but do so by mapping into the overall UUID structure of time and clock values. This structure is of no value here yet adds complexity. It also slightly reduces the number of actual bits with entropy.¶

UUIDs seem to have been designed for scenarios where the implementor does not have full control over the environment and uniqueness has to be constructed from identifiers at hand. UEID takes the view that hardware, software and/or manufacturing process directly implement UEID in a simple and direct way. It takes the view that cryptographic quality random number generators are readily available as they are implemented in commonly used CPU hardware.¶

C.1. From draft-rats-eat-01

• Added UEID design rationale appendix¶

C.2. From draft-mandyam-rats-eat-00

This is a fairly large change in the orientation of the document, but no new claims have been added.¶

• Separate information and data model using CDDL.¶
• Say an EAT is a CWT or JWT¶
• Use a map to structure the boot_state and location claims¶

C.3. From draft-ietf-rats-eat-01

• Clarifications and corrections for OEMID claim¶
• Minor spelling and other fixes¶
• Add the nonce claim, clarify jti claim¶

C.4. From draft-ietf-rats-eat-02

• Roll all EUIs back into one UEID type¶
• UEIDs can be one of three lengths, 128, 192 and 256.¶
• Added appendix justifying UEID design and size.¶
• Submods part now includes nested eat tokens so they can be named and there can be more tha one of them¶
• Lots of fixes to the CDDL¶
• Added security considerations¶

C.5. From draft-ietf-rats-eat-03

• Split boot_state into secure-boot and debug-disable claims¶
• Debug disable is an enumerated type rather than Booleans¶

C.6. From draft-ietf-rats-eat-04

• Change IMEI-based UEIDs to be encoded as a 14-byte string¶
• CDDL cleaned up some more¶
• CDDL allows for JWTs and UCCSs¶
• CWT format submodules are byte string wrapped¶
• Allows for JWT nested in CWT and vice versa¶
• Allows UCCS (unsigned CWTs) and JWT unsecured tokens¶
• Clarify tag usage when nesting tokens¶
• Add section on key inclusion¶
• Add hardware version claims¶
• Collected CDDL is now filled in. Other CDDL corrections.¶
• Rename debug-disable to debug-status; clarify that it is not extensible¶
• Security level claim is not extensible¶
• Improve specification of location claim and added a location privacy section¶
• Add intended use claim¶