Remote Attestation Procedures (RATS) Endorsements

Introduction provides an overview of the roles and conceptual messages in the IETF Remote ATtestation procedureS (RATS) architecture. As discussed in that document, a Verifier accepts a well-defined set of RATS conceptual messages: Evidence, Endorsements, and Reference Values, as well as Appraisal Policy for Evidence. A Verifier appraises Evidence using Appraisal Policy for Evidence, typically against a set of Reference Values. This document updates to provide further details on Reference Values and Endorsements. Specifically, this document updates as follows:

It classifies the conceptual messages defined in into actual state, reference state, and appraisal policy ().
It clarifies how Endorsements differ from Evidence and Reference Values ().
It refines the treatment of Endorsements in by adding considerations on conditionally endorsed values, the endorsement of verification keys, timeliness, and the handling of multiple Endorsements.

Readers of are expected to read it together with this document; where the two differ in their description of Endorsements and Reference Values, this document takes precedence.

Actual State vs Reference State Appraisal policies (Appraisal Policy for Evidence, and Appraisal Policy for Attestation Results) involve comparing the actual state of an Attester against desired or undesired states, to determine how trustworthy the Attester is for the Relying Party's purposes. The state of an Attester includes the structural composition of the Attester as the arrangement of its various execution environments, which are typically organized hierarchically. The state of an Attester also encompasses the combination of static and dynamic composition (e.g., provisioned and deployed software, firmware, and microcode), static and dynamic configuration, and the resulting operational state of its components at a certain point in time. Thus, a Verifier needs to receive conceptual messages with information about actual state, and information about desired/undesired states, and an appraisal policy that controls how the two are compared. Each Attester in general has at least one Attesting Environment and one Target Environment (e.g., hardware, firmware, or operating system). Typically, each Attester has multiple Target Environments, each with their own "Claims sets" representing their actual state. Additionally, there can be multiple Target Environments and Attesting Environments that are components of an Attester. "Actual state" is a group of Claims sets about the actual state of the Attester at a given point in time. Each Claims set holds Claims about a specific Target Environment that is essential to determining trustworthiness. Each Claim has a name and a singleton value, being a value collected from a Target Environment of a specific Attester at a given point in time. Some Claims may inherently have multiple values, such as a list of files in a given location on the device, but in the context of this document such a list is treated as a single unit, representing one Attester at one point in time. "Reference state" is a group of Claims sets about the desired or undesired state of an Attester. Typically, each Claim has a name and a set of potential values, being the values that are allowed/disallowed when determining the trustworthiness of the Attester. Generally, there may be varying degrees of gradation beyond just "allowed" or "disallowed." Reference state can have a set of values per claim per Target Environment. This is contrasted with actual state, which has a single value per claim per Target Environment. Actual state applies to one device at one point in time. Appraisal policy then specifies how to match the actual state values against a set of Reference Values. Some examples of such matching include:

An actual value must be in the set of allowed Reference Values.
An actual value must not be in the set of disallowed Reference Values.
An actual value must be in a range where two Reference Values are the min and max.

RATS Conceptual Messages RATS "conceptual messages" (see ) fall into the above categories as follows:

Actual state: Evidence, Endorsements, Attestation Results
Reference state: Reference Values
Appraisal policy: Appraisal Policy for Evidence, Appraisal Policy for Attestation Results

Evidence and Endorsements are both "actual state" messages, but they differ in who asserts the Claims and about which Environment. Evidence is a set of Claims about a Target Environment, generated (and typically signed) by the Attesting Environment. Endorsements, by contrast, are provided by an Endorser -- a role distinct from the Attester, such as a manufacturer or certification body -- and typically carry Claims about an Attesting Environment that has no lower-layer Attesting Environment of its own to vouch for it, or other Claims about the Attester's features that the Attester is not in a position to self-assert. See also . Hints (or suggestions) for how to perform a comparison might be supplied by a Reference Value Provider (as part of Reference Values), an Endorser (in an Endorsement), and/or an Attester (in Evidence), but the Verifier Owner is authoritative for Appraisal Policy for Evidence, and the Relying Party Owner is authoritative for Appraisal Policy for Attestation Results as depicted in . shows an example of Verifier input for a layered Attester as discussed in .

Example Verifier Input .-----------------. | e | |Actual state| Comparison | Reference state | | | | (layer 1) | Rules | (layer 1) | | V '-- '------------' '-----------------' | a | l .-- .------------. .-----------------. | u Endorsements | |Actual state| | Reference state | | e | | (layer 0) | | (layer 0) | | s '-- '------------' '-----------------' --' ]]> shows Endorsements only at layer 0 as a typical example, not as a general rule. A Trust Anchor Store is a special case of state above, where the Reference State would be the set of trust anchors accepted (or rejected) by the Verifier, and the Actual State would be a trust anchor used to appraise Evidence or Endorsements. In layered attestation using DICE , for example, the actual state of each layer is signed by a key held by the next lower layer. Thus in the example diagram above, the layer 2 actual state (e.g., OS state) is signed by a layer 1 key (e.g., a signing key used by the firmware), the layer 1 actual state (e.g., firmware state) is signed by a layer 0 key (e.g., a hardware key stored in ROM), and the layer 0 actual state (hardware specs and key ID) is signed by a layer 0 key (e.g., a vendor key) which is matched against the Verifier's trust anchor store, which is part of the layer 0 reference state depicted above.

Conditionally Endorsed Values The example depicted in shows Evidence containing actual state for layers 1 through N, and an Endorsement containing actual state for layer 0. However, some claims in Endorsements might be conditional and so are only treated as actual state if a condition is met. A claim is conditional if it only applies if other actual state matches Reference Values, according to some matching policy. For example, an Endorser for a given CPU might provide additional information about the CPU's supported features based on the current firmware configuration. Alternatively, an Endorser might provide information indicating that a specific security certification (e.g., and ) is associated with the device if the serial number falls within a given range and the firmware is configured in a specific way.

Conditional Endorsements | Condition | claims-sets | '--------------' Endorsement '-----------+---+---------' ^ Matching | | Rules [condition is met] | | | .-------------. | '----------+ claims-sets +----------' '-------------' ]]> Thus, actual state is determined by starting with a collection of unconditional Claims and adding any conditional Claims whose conditions are met based on the actual state (). This process is then repeated until no more conditional Claims are added. Verifier policies around matching actual state against reference state are normally expressed in Appraisal Policy for Evidence. Similarly, reference state is normally expressed in the Reference Values conceptual message. Such policies allow a Verifier and Relying Parties to make their decisions about the trustworthiness of an Attester. The use of conditionally endorsed values, however, is different in that a matching policy is not about trustworthiness (and hence not "appraisal" per se) but rather about whether an Endorser's claim is applicable or not, and thus usable as input to trustworthiness appraisal or not. As such the matching policy for conditionally endorsed values must be up to the Endorser not the Verifier Owner and/or Relying Party Owner. Thus, an Endorsement format that supports conditionally endorsed values (e.g., ) would probably include some minimal matching policy (e.g., exact match against a singleton reference value). This unfortunately complicates the Verifier design as it may need multiple parsers for matching policies.

Endorsing Verification Keys Attesting Environments have cryptographic keys that allow authenticating the Evidence that they produce. Typically, the bottom-most Attesting Environment in an Attester will sign Claims about one or more Target Environments (see also the DICE example at the end of ) with a private key that the Attesting Environment possesses, and the Verifier will appraise the resulting Evidence with a public key it possesses, called a verification key. While use of public key cryptography is typical for a verification key, cryptography other than public key may also be used. Endorsing the linkage between such verification keys and their associated Attesting Environments is crucial to the appraisal process. The Verifier must have access to a verification key for each Attesting Environment. Such a key could be provisioned directly in the Verifier. However, for scalability the Verifier typically is provisioned with a trusted root CA certificate (a trust anchor). This means that an Endorsement from an Endorser includes the Attesting Environment's verification key material in the form of a certificate that chains up to that trust anchor (a certification path). Such a certificate might be stored in the Verifier, or might be resolved on demand via some protocol, or might be passed to the Verifier along with the Evidence to appraise, depending on the protocol or general remote attestation procedure. Details are out of scope of this document and left to protocol or procedure specifications. A specific detail that is out of scope is whether an unreachable Endorser only postpones the appraisal of Evidence or renders it permanently impossible, which depends on whether the Endorsement remains retrievable and applicable to the previously collected Evidence. Specific protocol documents are also responsible for documenting what particular algorithm or cryptographic protocol is used for the appraisal of the Evidence. The verification key (i.e., a key with the purpose of signature checking) could be, typically, a symmetric key, a raw public key, or a certified public key. Evidence can contain an identifier for the Attester (e.g., ueid ) in a dedicated "identity claim" that can be used by the Verifier to look up its verification key for the Attester. The granularity at which such identifiers, and therefore the signature-checking keys endorsed for them, apply (e.g., per instance, class, or other Claims) is out of scope of this document. While identity Claims are another type of Claims that may be endorsed, some implementations might treat them differently. For example, a Verifier might perform a first step to cryptographically appraise that the Evidence has been generated by the Attester that has the key material associated with the identifier in the identity Claim(s) before spending effort on another step to appraise other Claims for determining trustworthiness. This document treats identity Claims the same as other Claims but allows Appraisal Policy for Evidence to have multiple phases if desired.

Timeliness discusses timeliness of Claims in Evidence. When additional "static" Claims (i.e., Claims representing invariant properties of the environment) are provided in Endorsements, no additional steps are needed for timeliness of those Claims since they are static rather than dynamically varying over time. Once timeliness of Evidence is appraised, any matching conditionally endorsed values can be applied. If Endorsements ever carry dynamic Claims in the future (e.g., whether any vulnerabilities in the version of firmware are currently known), then the same timeliness considerations as for Claims in Evidence would apply, and would be the responsibility of specific protocol documents. See and for further discussion. This distinction between static and dynamic Claims is about the invariance of properties of the environment, not about the Endorser's assessment of those properties. A condition in a Conditionally Endorsed Value () is static in the sense that, once matched against sufficiently timely Evidence, it deterministically selects a claims-set; however, the verdict an Endorser attaches to a given condition can itself change over time as the Endorser's own knowledge evolves, even though the condition's matching value does not. For example, an Endorser might issue a Conditional Endorsement stating that if a given RoT firmware measurement matches value H, the device is trusted, and later, after a vulnerability is discovered in that firmware, issue a further Conditional Endorsement stating that if the same value H is matched, the device is untrusted. Both Endorsements may be signed by an Endorser that remains in good standing, so this is a different problem than Endorser standing: it is about which of two (or more) temporally scoped, and potentially contradictory, Endorsements apply at the time of appraisal. Endorsement formats therefore need a way to bind a validity period to Endorsement content, in addition to any validity information about the Endorser's standing (discussed below), so that a Verifier can determine which Endorsement supersedes others given the same condition. For example, provides a rim-validity window in the corim-map for this purpose. Conversely, an attacker, or a party that merely relays Endorsements (e.g., a CoSERV server ), might present only an older Endorsement and withhold a newer one that supersedes or revokes it. A validity period alone does not prevent this, because the older Endorsement can still be within its validity window. A Verifier therefore needs to obtain Endorsements from a source it trusts to supply the current set (e.g., directly from the Endorser over an authenticated channel, see ), or to learn of revocations or supersession through a separate mechanism (e.g., a revocation list or a sequence number covering the Endorser's Endorsements), as determined by the specific protocol or appraisal policy. The Endorser's standing (i.e., the fact that its signing key or certificate is still valid or its trust anchor is still recognized) is a further, independent consideration, logically separate from the validity of the Endorsement content discussed above. Whether the Endorser's standing is evaluated relative to the time the Evidence was generated or the time of appraisal is a decision for the specific protocol or appraisal policy, and needs to be documented as part of the timeliness of the Endorsement itself. For example, the CoRIM data model provides a signature-validity window that bounds the validity of the Endorser's signature, distinct from the rim-validity window discussed above, and the CoRIM processor described in checks it, together with revocation and trust anchor status, relative to the time of appraisal.

Multiple Endorsements shows an example with an Endorsement at layer 0, such as a hardware manufacturer providing claims about the hardware. However, the same could be done at other layers in addition. For example, an OS vendor might provide additional static Claims about the OS software it provides, and application developers might provide additional static Claims about the applications they release. depicts an example with an Attester consisting of an application, OS, firmware, and hardware, each from a different vendor that provides an Endorsement for their own Target Environment, containing additional Claims about that Target Environment. Thus each Target Environment (application, OS, firmware, and hardware) has one set of Claims ("claims set 1") in the Evidence, and an additional set of Claims ("claims set 2") in the Endorsement from its manufacturer. A Verifier that trusts each Endorser would thus use the Claims sets from both conceptual messages (Endorsements and Evidence) when comparing against reference state for a given Target Environment.

Multiple Endorsements |Endorsement | app | | | | app | | | |claims set 2| | | |claims set 1| | | '------------' | | '------------' E| '---------------------------' | v| | i| .---------------------------. | d| OS | .------------. | | .------------. e| Endorser ----> |Endorsement | OS | | | | OS | n| | |claims set 2| | | |claims set 1| c| | '------------' | | '------------' e| '---------------------------' | | | | .---------------------------. | | Firmware | .------------. | | .------------. | Endorser ----> |Endorsement | firmware | | | | firmware | | | |claims set 2| | | |claims set 1| | | '------------' | | '------------' | '---------------------------' | | | | .---------------------------. | | Hardware | .------------. | | .------------. | Endorser ----> |Endorsement | hardware | | | | hardware | | | |claims set 2| | | |claims set 1| | | '------------' | | '------------' | '---------------------------' '-----------------' ^ | Attester -------------------------------------------' ]]> When Target Environments from different vendors each have their own Endorser, a Verifier must be able to distinguish which Endorser is allowed to provide an Endorsement about which Target Environment. For example, the OS Endorser might be trusted to provide additional claims about the OS, but not about the hardware. Thus, it is not as simple as saying that a Verifier has a trusted set of Endorsers. The authority for the binding between Target Environment and Endorser comes from the Appraisal Policy for Evidence, which is independent of the Endorser, the Target Environment, and the Evidence. Evidence cannot establish that authority on its own: a Verifier that relied solely on a Claim in Evidence naming the Endorser for a Target Environment would be trusting the Attester to select the party allowed to vouch for it. Evidence can still play a supporting role, e.g., Claims from a Target Environment might include an identifier that the Verifier uses to locate candidate Endorsements, but the Verifier accepts an Endorsement only if the Appraisal Policy for Evidence authorizes that Endorser for that Target Environment (for example, by listing the Endorser's trust anchor as acceptable for that class of Target Environment). An Endorsement format specification should explain how this concern is addressed.

Endorsement Format Considerations This section discusses considerations around formats for Endorsements.

Security Considerations for Formats In many scenarios, a Verifier can also support a variety of different formats, and while code size may not be a huge concern, simplicity and correctness of code is essential to security. "Complexity is the enemy of security" is a popular security mantra and hence to increase security, any decrease in complexity helps. As such, using the same format for both Evidence and Endorsements can reduce complexity and hence increase security.

Scalability Considerations for Formats This document assumes that Reference Value Providers typically provide the same information to a potentially large number of clients (Verifiers, or potentially to other entities for later relay to a Verifier), and are generally on devices that are not constrained nodes, and hence additional scalability, including code size, is not a significant concern. This document also assumes the same is true of Endorsers. The scenario where scalability in terms of code size is strongest, however, is when a Verifier is embedded into a constrained node. For example, when a constrained node is a Relying Party for most purposes, but still needs a way to establish trust in the Verifier it will use. In such a case, the Relying Party may have a constrained Verifier embedded in it that is only capable of appraising Evidence provided by its desired Verifier. Thus, the Relying Party uses its embedded Verifier for purposes of appraising its desired Verifier which it treats as only an Attester, and once appraised, then uses it for appraisal of all other Attesters. In this scenario, the embedded Verifier may have code and data size constraints, and a very simple (by comparison) Appraisal Policy for Evidence and desired state (e.g., a required trust anchor that Evidence must be signed with and little else). Using the same message format for Evidence, Endorsements, and (later) Attestation Results received from the later Verifier, can provide code size savings due to having only a single parser in this limited case. Similarly, an embedded constrained Verifier can choose to not support conditionally endorsed values to avoid the complexity introduced by them.

Security Considerations discusses how a Verifier stores one or more trust anchors in its trust anchor store. A Verifier expresses its trust in an Endorser by storing a trust anchor for that Endorser. The binding from an Endorsement to a given Target Environment is done as discussed in of this document. (especially Sections 3.2 and 12) also discusses security considerations around the remote attestation of layers, and sources of appraisal policies. of this document covers additional considerations in these areas, while covers additional considerations around Endorsement formats. The integrity of public and private key material and the secrecy of private key material must be ensured at all times. This includes public keys that identify trusted supply chain actors. For more detailed information on protecting trust anchors, refer to . A Verifier can use cryptographically protected, mutually authenticated secure channels to all its trusted input sources, particularly, Endorsers and Reference Value Providers. Signing the Endorsement or Reference Values themselves protects their integrity and authenticates their source, but a mutually authenticated channel additionally lets the source authenticate and authorize the requesting Verifier and protects potentially sensitive content (see ) against disclosure to unauthorized parties. These links should reach as deep as possible into the Verifier, potentially terminating within the appraisal session context, to avoid on-path attacks. Minimizing the use of intermediaries is also important, as each intermediary is another party that might need to be trusted. Refer to for information on conceptual message protection.

Privacy Considerations The privacy considerations regarding conceptual messages, as discussed in , apply. In particular, since Endorsements and Reference Values can contain privacy-sensitive information (e.g., Personally Identifiable Information (PII)) about a large number of devices, strong confidentiality protection is required at the time of conveyance. Utilizing the public part of an asymmetric key pair that is used for Evidence generation to identify an Attesting Environment raises privacy considerations that must be carefully considered.

IANA Considerations This document does not require any actions by IANA.

References Normative References Remote ATtestation procedureS (RATS) Architecture In network protocol exchanges, it is often useful for one end of a communication to know whether the other end is in an intended operating state. This document provides an architectural overview of the entities involved that make such tests possible through the process of generating, conveying, and evaluating evidentiary Claims. It provides a model that is neutral toward processor architectures, the content of Claims, and protocols. Remote ATtestation procedureS (RATS) Conceptual Message Wrapper (CMW) The conceptual messages introduced by the Remote ATtestation procedureS (RATS) architecture (RFC 9334) are protocol-agnostic data units that are conveyed between RATS roles during RATS interactions. Conceptual messages describe the meaning and function of such data units within RATS data flows without specifying a wire format, encoding, transport mechanism, or processing details. The initial set of conceptual messages is defined in Section 8 of RFC 9334 and includes Evidence, Attestation Results, Endorsements, Reference Values, and Appraisal Policies. This document introduces the Conceptual Message Wrapper (CMW) that provides a common structure to encapsulate these messages. It defines a dedicated Concise Binary Object Representation (CBOR) tag, corresponding JSON Web Token (JWT) and CBOR Web Token (CWT) claims, and an X.509 extension. Together, these mechanisms allow CMWs to be used in CBOR-based protocols, web APIs using JWTs and CWTs, and PKIX artifacts such as X.509 certificates. Additionally, this document defines media types and CoAP Content-Formats that may be used to identify CMWs when transported over protocols such as HTTP, MIME, and CoAP. The goal is to improve the interoperability and flexibility of remote attestation protocols. Introducing a shared message format such as CMW enables consistent support for different attestation message types, enables the evolution of message serialization formats without breaking compatibility, and avoids the need to redefine how messages are handled within each protocol. Informative References The Entity Attestation Token (EAT) An Entity Attestation Token (EAT) provides an attested claims set that describes the state and characteristics of an entity, a device such as a smartphone, an Internet of Things (IoT) device, network equipment, or such. This claims set is used by a relying party, server, or service to determine the type and degree of trust placed in the entity. An EAT is either a CBOR Web Token (CWT) or a JSON Web Token (JWT) with attestation-oriented claims. Concise Reference Integrity Manifest Fraunhofer SIT Linaro arm Independent Huawei Technologies Remote Attestation Procedures (RATS) enable Relying Parties to assess the trustworthiness of a remote Attester and therefore to decide whether or not to engage in secure interactions with it. Evidence about trustworthiness can be rather complex and it is deemed unrealistic that every Relying Party is capable of the appraisal of Evidence. Therefore that burden is typically offloaded to a Verifier. In order to conduct Evidence appraisal, a Verifier requires not only fresh Evidence from an Attester, but also trusted Endorsements and Reference Values from Endorsers and Reference Value Providers, such as manufacturers, distributors, or device owners. This document specifies the information elements for representing Endorsements and Reference Values in CBOR format. Concise Selector for Endorsements and Reference Values Arm Linaro Fraunhofer SIT Fujitsu AMD Alibaba Cloud In the Remote Attestation Procedures (RATS) architecture, Verifiers require Endorsements and Reference Values to assess the trustworthiness of Attesters. This document specifies the Concise Selector for Endorsements and Reference Values (CoSERV), a structured query/result format designed to facilitate the discovery and retrieval of these artifacts from various providers. CoSERV defines a query language and corresponding result structure using CDDL, which can be serialized in CBOR format, enabling efficient interoperability across diverse systems. DICE Attestation Architecture Trusted Computing Group Security Certification GlobalPlatform The OCP Security Appraisal Framework and Enablement (S.A.F.E.) Program Open Compute Project

Acknowledgements The authors wish to thank the following individuals for feedback and ideas that contributed to this document: , , , , , , , , , , , , , , , and