| RATS Working Group | G. Mandyam |
| Internet-Draft | Qualcomm Technologies Inc. |
| Intended status: Standards Track | L. Lundblade |
| Expires: August 23, 2020 | Security Theory LLC |
| M. Ballesteros | |
| J. O'Donoghue | |
| Qualcomm Technologies Inc. | |
| February 20, 2020 |
The Entity Attestation Token (EAT)
draft-ietf-rats-eat-03
An Entity Attestation Token (EAT) provides a signed (attested) set of claims that describe state and characteristics of an entity, typically a device like a phone or an IoT device. These claims are used by a relying party to determine how much it wishes to trust the entity.
An EAT is either a CWT or JWT with some attestation-oriented claims. To a large degree, all this document does is extend CWT and JWT.
TBD
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on August 23, 2020.
Copyright (c) 2020 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
Remote device attestation is a fundamental service that allows a remote device such as a mobile phone, an Internet-of-Things (IoT) device, or other endpoint to prove itself to a relying party, a server or a service. This allows the relying party to know some characteristics about the device and decide whether it trusts the device.
Remote attestation is a fundamental service that can underlie other protocols and services that need to know about the trustworthiness of the device before proceeding. One good example is biometric authentication where the biometric matching is done on the device. The relying party needs to know that the device is one that is known to do biometric matching correctly. Another example is content protection where the relying party wants to know the device will protect the data. This generalizes on to corporate enterprises that might want to know that a device is trustworthy before allowing corporate data to be accessed by it.
The notion of attestation here is large and may include, but is not limited to the following:
An EAT token is either a CWT as defined in [RFC8392] or a JWT as defined in [RFC7519]. This specification defines additional claims for entity attestation.
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 4.3.2 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).
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.
At least the following three participants exist in all EAT operating models. Some operating models have additional participants.
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.
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.
The following is not standardized for EAT, just the same they are not standardized for CWT or JWT.
EATs may be transmitted by any protocol the same as CWTs and JWTs. For example, they might be added in extension fields of other protocols, bundled into an HTTP header, or just transmitted as files. This flexibility is intentional to allow broader adoption. This flexibility is possible because EAT’s are self-secured with signing (and possibly additionally with encryption and anti-replay). The transmission protocol is not required to fulfill any additional security requirements.
For certain devices, a direct connection may not exist between the EAT-producing device and the Relying Party. In such cases, the EAT should be protected against malicious access. The use of COSE and JOSE allows for signing and encryption of the EAT. Therefore, even if the EAT is conveyed through intermediaries between the device and Relying Party, such intermediaries cannot easily modify the EAT payload or alter the signature.
The term “signing scheme” is used to refer to the system that includes end-end process of establishing signing attestation key material in the entity, signing the EAT, and verifying it. This might involve key IDs and X.509 certificate chains or something similar but different. The term “signing algorithm” refers just to the algorithm ID in the COSE signing structure. No particular signing algorithm or signing scheme is required by this standard.
There are three main implementation issues driving this. First, secure non-volatile storage space in the entity for the attestation key material may be highly limited, perhaps to only a few hundred bits, on some small IoT chips. Second, the factory cost of provisioning key material in each chip or device may be high, with even millisecond delays adding to the cost of a chip. Third, privacy-preserving signing schemes like ECDAA (Elliptic Curve Direct Anonymous Attestation) are complex and not suitable for all use cases.
Over time to faciliate interoperability, some signing schemes may be defined in EAT profiles or other documents either in the IETF or outside.
The key words “MUST”, “MUST NOT”, “REQUIRED”, “SHALL”, “SHALL NOT”, “SHOULD”, “SHOULD NOT”, “RECOMMENDED”, “NOT RECOMMENDED”, “MAY”, and “OPTIONAL” in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.
This document reuses terminology from JWT [RFC7519], COSE [RFC8152], and CWT [RFC8392].
This section describes new claims defined for attestation. It also mentions several claims defined by CWT and JWT that are particularly important for EAT.
Note also: * Any claim defined for CWT or JWT may be used in an EAT including those in the CWT [IANA.CWT.Claims] and JWT IANA [IANA.JWT.Claims] claims registries.
CDDL along with text descriptions is used to define each claim indepdent of encoding. Each claim is defined as a CDDL group (the group is a general aggregation and type definition feature of CDDL). In the encoding section Section 4, the CDDL groups turn into CBOR map entries and JSON name/value pairs.
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.
The “iat” claim defined in CWT and JWT is used to indicate the date-of-creation of the token.
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.
nonce-type = [ + bstr .size (8..64) ]
nonce-claim = (
nonce => nonce-type
)
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 6.1.
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].
| Type Byte | Type Name | Specification |
|---|---|---|
| 0x01 | RAND | This is a 128, 192 or 256 bit random number generated once and stored in the device. 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 device and stored. It may not be smaller than 128 bits. |
| 0x02 | IEEE EUI | This makes use of the IEEE company identification registry. An EUI is either an EUI-48, EUI-60 or EUI-64 and made up of an OUI, OUI-36 or a CID, different registered company identifiers, and some unique per-device identifier. 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 devices with multiple network interfaces may have multiple MAC addresses, there is only one UEID for a device) [IEEE.802-2001], [OUI.Guide] |
| 0x03 | IMEI | This is a 14-digit identifier consisting of an 8-digit Type Allocation Code and a 6-digit serial number allocated by the manufacturer, which SHALL be encoded as a binary integer over 48 bits. The IMEI value encoded SHALL NOT include Luhn checksum or SVN information. [ThreeGPP.IMEI] |
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:
ueid-claim = (
ueid => bstr .size (7..33)
)
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:
| Name | Description |
|---|---|
| Acme-TEE | The EATs are generated in the TEE authored and configured by “Acme” |
| Acme-TPM | The EATs are generated in a TPM manufactured by “Acme” |
| Acme-Linux-Kernel | The EATs are generated in a Linux kernel configured and shipped by “Acme” |
| Acme-TA | The EATs are generated in a Trusted Application (TA) authored by “Acme” |
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.
origination-claim = (
origination => string-or-uri
)
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.
oemid-claim = (
oemid => bstr
)
EATs have a claim that roughly characterizes the device / entities ability to defend against attacks aimed at capturing the signing key, forging claims and at forging EATs. This is done by roughly defining four security levels as described below. This is similar to the security levels defined in the Metadata Service defined by the Fast Identity Online (FIDO) Alliance (TODO: reference).
These claims describe security environment and countermeasures available on the end-entity / client device where the attestation key reside and the claims originate.
This claim is not intended as a replacement for a proper end-device security certification schemes such as those based on FIPS (TODO: reference) or those based on Common Criteria (TODO: reference). The claim made here is solely a self-claim made by the Entity Originator.
security-level-type = &(
unrestricted: 1,
restricted: 2,
secure-restricted: 3,
hardware: 4
)
security-level-claim = (
security-level => security-level-type
)
This claim is an array of five Boolean values indicating the boot and debug state of the entity.
This indicates whether secure boot is enabled either for an entire device or an individual submodule. If it appears at the device level, then this means that secure boot is enabled for all submodules. Secure boot enablement allows a secure boot loader to authenticate software running either in a device or a submodule prior allowing execution.
This indicates whether debug capabilities are disabled for an entity (i.e. value of ‘true’). Debug disablement is considered a prerequisite before an entity is considered operational.
This claim indicates whether debug capabilities for the entity were not disabled in any way since boot (i.e. value of ‘true’).
This claim indicates whether debug capabilities for the entity are permanently disabled (i.e. value of ‘true’). This value can be set to ‘true’ also if only the manufacturer is allowed to enabled debug, but the end user is not.
This claim indicates whether debug capabilities for the entity are permanently disabled (i.e. value of ‘true’). This value can only be set to ‘true’ if no party can enable debug capabilities for the entity. Often this is implemented by blowing a fuse on a chip as fuses cannot be restored once blown.
boot-state-type = [
secure-boot-enabled => bool,
debug-disabled => bool,
debug-disabled-since-boot => bool,
debug-permanent-disable => bool,
debug-full-permanent-disable => bool
]
boot-state-claim = (
boot-state => boot-state-type
)
The location claim is a CBOR-formatted object that describes the location of the device entity from which the attestation originates. It is comprised of a map of additional sub claims that represent the actual location coordinates (latitude, longitude and altitude). The location coordinate claims are consistent with the WGS84 coordinate system [WGS84]. In addition, a sub claim providing the estimated accuracy of the location measurement is defined.
location-type = {
latitude => number,
longitude => number,
? altitude => number,
? accuracy => number,
? altitude-accuracy => number,
? heading => number,
? speed => number
}
location-claim = (
location => location-type
)
The “age” claim contains a value that represents the number of seconds that have elapsed since the token was created, measurement was made, or location was obtained. Typical attestable values are sent as soon as they are obtained. However, in the case that such a value is buffered and sent at a later time and a sufficiently accurate time reference is unavailable for creation of a timestamp, then the age claim is provided.
age-claim = (
age => uint
)
The “uptime” claim contains a value that represents the number of seconds that have elapsed since the entity or submod was last booted.
uptime-claim = (
uptime => uint
)
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 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…
Each entry in the submod map one of two types:
Essentially this type of submodule, is just a sub-map or sub-object containing claims. It is recognized from the other type by being a data item of type map in CBOR or by being an object in JSON.
The contents are claims about the submodule of types defined in this document or anywhere else claims types are defined.
This type of submodule is a fully formed EAT as described here. In this case the submodule has key material distinct from the containing EAT token that allows it to sign on its own.
When an EAT is nested in another EAT as a submodule the nested EAT MUST use the CBOR CWT tag. This clearly distinguishes it from the non-token submodules.
The subordinate modules do not inherit anything from the containing token. The subordinate modules must explicitly include all of their claims. This is the case even for claims like the nonce and age.
This rule is in place for simplicity. It avoids complex inheritance rules that might vary from one type of claim to another. (TODO: fix the boot claim which does have inheritance as currently described).
The security level of the non-token subordinate modules should always be less than or equal to that of the containing modules in the case of non-token submodules. It makes no sense for a module of lesser security to be signing claims of a module of higher security. An example of this is a TEE signing claims made by the non-TEE parts (e.g. the high-level OS) of the device.
The opposite may be true for the nested tokens. They usually have their own more secure key material. An example of this is an embedded secure element.
The label or name for each submodule in the submods map is a text string naming the submodule. No submodules may have the same name.
submods-type = { + submodule }
submodule = (
submod_name => eat-claims / eat-token
)
submod_name = tstr / int
submods-part = (
submods => submod-type
)
This makes use of the types defined in CDDL Appendix D, Standard Prelude.
string-or-uri = uri / tstr; See JSON section below for JSON encoding of string-or-uri
This section provides CDDL for the claims defined in CWT. It is non-normative as [RFC8392] is the authoritative definition of these claims.
rfc8392-claim //= ( issuer => text ) rfc8392-claim //= ( subject => text ) rfc8392-claim //= ( audience => text ) rfc8392-claim //= ( expiration => time ) rfc8392-claim //= ( not-before => time ) rfc8392-claim //= ( issued-at => time ) rfc8392-claim //= ( cwt-id => bytes ) issuer = 1 subject = 2 audience = 3 expiration = 4 not-before = 5 issued-at = 6 cwt-id = 7 cwt-claim = rfc8392-claim
ueid = "ueid" origination = "origination" oemid = "oemid" security-level = "security-level" boot-state = "boot-state" location = "location" age = "age" uptime = "uptime" nested-eat = "nested-eat" submods = "submods" latitude = "lat" longitude = "long"" altitude = "alt" accuracy = "accry" altitude-accuracy = "alt-accry" heading = "heading" speed = "speed"
JSON should be encoded per RFC 8610 Appendix E. In addition, the following CDDL types are encoded in JSON as follows:
ueid = To_be_assigned origination = To_be_assigned oemid = To_be_assigned security-level = To_be_assigned boot-state = To_be_assigned location = To_be_assigned age = To_be_assigned uptime = To_be_assigned submods = To_be_assigned nonce = To_be_assigned latitude = 1 longitude = 2 altitude = 3 accuracy = 4 altitude-accuracy = 5 heading = 6 speed = 7
Variations in the CBOR serializations supported in CBOR encoding and decoding are allowed and suggests that CBOR-based protocols specify how this variation is handled. This section specifies what formats MUST be supported in order to achieve interoperability.
The assumption is that the entity is likely to be a constrained device and relying party is likely to be a very capable server. The approach taken is that the entity generating the token can use whatever encoding it wants, specifically encodings that are easier to implement such as indefinite lengths. The relying party receiving the token must support decoding all encodings.
These rules cover all types used in the claims in this document. They also are recommendations for additional claims.
Canonical CBOR encoding, Preferred Serialization and Deterministically Encoded CBOR are explicitly NOT required as they would place an unnecessary burden on the entity implementation, particularly if the entity implementation is implemented in hardware.
A generic-claim is any CBOR map entry or JSON name/value pair.
eat-claims = { ; the top-level payload that is signed using COSE or JOSE
* claim
}
claim = (
ueid-claim //
origination-claim //
oemid-claim //
security-level-claim //
boot-state-claim //
location-claim //
age-claim //
uptime-claim //
submods-part //
cwt-claim //
generic-claim-type //
)
eat-token ; This is a set of eat-claims signed using COSE
TODO: copy the rest of the CDDL here (wait until the CDDL is more settled so as to avoid copying multiple times)
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.
TODO: add the rest of the claims in here
Certain EAT claims can be used to track the owner of an entity and therefore, implementations should consider providing privacy-preserving options dependent on the intended usage of the EAT. Examples would include suppression of location claims for EAT’s provided to unauthenticated consumers.
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:
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.
The security considerations provided in Section 8 of [RFC8392] and Section 11 of [RFC7519] apply to EAT in its CWT and JWT form, respectively. In addition, implementors should consider the following.
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.
Regarding transmission of key material from the enclave to the entity, the key material may pass through one or more intermediaries. Therefore some form of protection (“key wrapping”) may be necessary. The transmission itself may be performed electronically, but 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.
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.
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.
This is shown in CBOR diagnostic form. Only the payload signed by COSE is shown.
{
/ nonce / 9:h'948f8860d13a463e8e',
/ UEID / 10:h'0198f50a4ff6c05861c8860d13a638ea4fe2f',
/ boot-state / 12:{true, true, true, true, false}
/ time stamp (iat) / 6:1526542894,
}
{
/ nonce / 9:h'948f8860d13a463e8e',
/ UEID / 10:h'0198f50a4ff6c05861c8860d13a638ea4fe2f',
/ boot-state / 12:{true, true, true, true, false}
/ time stamp (iat) / 6:1526542894,
/ seclevel / 11:3, / secure restricted OS /
/ submods / 17:
{
/ first submod, an Android Application / "Android App Foo" : {
/ seclevel / 11:1, / unrestricted /
/ app data / -70000:'text string'
},
/ 2nd submod, A nested EAT from a secure element / "Secure Element Eat" :
/ eat / 61( 18(
/ an embedded EAT, bytes of which are not shown /
))
/ 3rd submod, information about Linux Android / "Linux Android": {
/ seclevel / 11:1, / unrestricted /
/ custom - release / -80000:'8.0.0',
/ custom - version / -80001:'4.9.51+'
}
}
}
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.
| People | Devices / Person | Subsystems / Device | Database Portion | Database Size |
|---|---|---|---|---|
| 10 billion | 100 | 10 | 10% | trillion (10^12) |
| 10 billion | 100,000 | 10 | 10% | quadrillion (10^15) |
| 100 billion | 1,000,000 | 10 | 10% | 100 quadrillion (10^17) |
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
| 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 (10^17) | 2 * 10^-05 | 8 * 10^-25 | 5 * 10^-45 |
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
| 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 (10^17) | 8 microseconds | 10^11 years | 10^31 years |
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.
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.
The following is a list of known changes from the previous drafts. This list is non-authoritative. It is meant to help reviewers see the significant differences.
This is a fairly large change in the orientation of the document, but not new claims have been added.