| Internet-Draft | WIMSE Architecture | October 2024 |
| Salowey, et al. | Expires 24 April 2025 | [Page] |
The increasing prevalence of cloud computing and micro service architectures has led to the rise of complex software functions being built and deployed as workloads, where a workload is defined as a running instance of software executing for a specific purpose. This document discusses an architecture for designing and standardizing protocols and payloads for conveying workload identity and security context information.¶
This note is to be removed before publishing as an RFC.¶
Discussion of this document takes place on the Workload Identity in Multi System Environments Working Group mailing list (wimse@ietf.org), which is archived at https://mailarchive.ietf.org/arch/browse/wimse/.¶
Source for this draft and an issue tracker can be found at https://github.com/jsalowey/wimse-arch.¶
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 24 April 2025.¶
Copyright (c) 2024 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
The increasing prevalence of cloud computing and micro service architectures has led to the rise of complex software functions being built and deployed as systems composed of workloads, where a workload is defined as a running instance of software executing for a specific purpose.¶
Workloads need to be provisioned with an identity when they are started. Often, additional information needs to be provided, such as trust anchors and security context details. Workload identity credential is used to authenticate communications between workloads. Workloads make use of identity information and additional context information to perform authentication and authorization.¶
This architecture considers two ways to express identity information: X.509 certificates often used in the TLS layer and JSON Web Tokens (JWTs) used at the application layer. Collectively, these are referred to as WIMSE tokens. The applicability of given token format depends on application and security context and will be explored in later sections.¶
Once the workload is started and has obtained identity information, it can start performing its functions. Once the workload is invoked it may require interaction with other workloads. An example of such interaction is shown in [I-D.ietf-oauth-transaction-tokens] where an externally-facing endpoint is invoked using conventional authorization mechanism, such as an OAuth 2.0 access token. The interaction with other workload may require the security context associated with the authorization to be passed along the call chain.¶
In the rest of the document we describe terminology and use cases, discuss details of the architecture, and discuss threats.¶
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 uses the following terms:¶
Workload¶
A workload is a running instance of software executing for a specific purpose. Workload typically interacts with other parts of a larger system. A workload may exist for a very short durations of time (fraction of a second) and run for a specific purpose such as to provide a response to an API request. Other kinds of workloads may execute for a very long duration, such as months or years. Examples include database services and machine learning training jobs.¶
Security Context¶
A security context provides information needed for a workload to perform its function. This information is often used for authorization, accounting and auditing purposes and often contains information about the request being made. Some examples include user information, software and hardware information or information about what processing has already happened for the request. Different pieces of context information may originate from different sources.¶
Identity Proxy¶
Identity proxy is an intermediary that can inspect, replace or augment workload identity and security context information. Identity proxy can be a capability of a transparent network service, such as a security gateway, or it can be implemented in a service performing explicit connection processing, such as an ingress gateway or a Content Delivery Network (CDN) service.¶
Attestation¶
Attestation is the function through which a task verifies the identity of a separate Workload. (TBD: sync definition with reference RaTS architecture)¶
Workload Identity Token¶
A token that contains a workload identifier used for service to service authentication. The token is bound to a cryptographic key and requires that the presenter provide proof of possession of the secret key material. This token is further defined in [I-D.ietf-wimse-s2s-protocol]¶
Workload identity construct consists of three basic building blocks: trust domain, workload identifier and workflow identity credentials. These components are sufficient for establishing authentication, authorization and accounting processes. More complex identity constructs can be created from these basic building blocks.¶
A trust domain is a logical grouping of systems that share a common set of security controls and policies. Workload certificates and tokens are issued under the authority of a trust domain. Trust domains SHOULD be identified by a fully qualified domain name associated with the organization defining the trust domain. THe FQDN format of trust domain helps to ensure uniqueness of the trust domain identifier. A trust domain maps to one or more trust anchors for validating X.509 certificates and a mechanism to securely obtain a JWK Set [RFC7517] for validating WIMSE WIT tokens. This mapping MUST be obtained through a secure mechanism that ensures the authenticity and integrity of the mapping is fresh and not compromised. This secure mechanism is out of scope for this document.¶
A single organization may define multiple trust domains for different purpose such as different departments or environments. Each trust domain must have a unique identifier. Workload identifiers are scoped within a trust domain. If two identifiers differ only by trust domain they still refer to two different entities.¶
The WIMSE architecture defines a workload identifier as a URI [RFC3986]. This URI is used in the subject fields in the certificates and tokens defined later in this document. The URI MUST meet the criteria for the URI type of Subject Alternative Name defined in Section 4.2.1.6 of [RFC5280].¶
The name MUST NOT be a relative URI, and it MUST follow the URI syntax and encoding rules specified in [RFC3986]. The name MUST include both a scheme and a scheme-specific-part.¶
In addition the URI MUST include an authority that identifies the trust domain within which the identifier is scoped. The trust domain SHOULD be a fully qualified domain name belonging to the organization defining the trust domain to help provide uniqueness for the trust domain identifier. The scheme and scheme specific part are not defined by this specification. An example of an identifier format that conforms to this definition is SPIFFE ID.¶
While the URI encoding rules allow host names to be specified as IP addresses, IP addresses MUT NOT be used to represent trust domains except in the case where they are needed for compatibility with existing naming schemes.¶
A workload identity only has meaning within the scope of a specific issuer. Two identities of the same value issued by different issuers may or may not refer to the same workload. In order to avoid collisions identity URIs SHOULD specify, in the URI's "authority" field, the trust domain associated with an issuer that is selected from a global name space such as host domains. However, the validator of an identity credential MUST make sure that they are using the correct issuer credential to verify the identity credential and that the issuer is trusted to issue tokens for the defined trust domain.¶
The Agent provisions the identity credentials to the workload. These credentials are represented in form of JWT tokens and/or X.509 certificates.¶
JWT bearer tokens are presented to another party as a proof of identity. They are signed to prevent forgery, however since these credentials are often not bound to other information it is possible that they could be stolen and reused elsewhere. To mitigate these risks and make the token more generally useful the WIMSE architecture defines a workload identity token that binds a JWT to a cryptographic key.¶
Both X.509 certificate and workload identity token credentials consist of two parts:¶
a certificate or WIT is a signed data structure that contains a public key and identity information¶
a corresponding private key¶
The certificate or WIT is presented during authentication, however the private key is kept secret and only used in cryptographic computation to prove that the presenter has access to the private key corresponding to the public key.¶
[TODO: this section will need to be updated to discuss workload identifier as a concept as well]¶
A workload needs to obtain its identity early in its lifecycle. This identity is the base identity upon which further identity and security context are built.¶
Identity bootstrapping often utilizes identity information provisioned through mechanisms specific to hosting platforms and orchestration services. This initial bootstrapping information is used to obtain specific identity credentials for a workload. This process may use attestation to ensure the workload receives the correct identity credentials. An example of a bootstrapping process follows.¶
Figure 1 provides an example of software layering at a host running workloads. During startup, workloads bootstrap their identity with the help of an agent. The agent may be associated with one or more workloads to help ensure that workloads are provisioned with the correct identity. The agent provides attestation evidence and other relevant information to a server. The server validates this information and provides the agent with identity credentials for the workloads it is associated with. The server can use a variety of internal and external means to validate the request against policy. After obtaining identity credentials from the Server, the agent passes them to the workload.¶
How the workload obtains its identity credentials and interacts with the agent is subject to different implementations. Some common mechanisms for obtaining this initial identity include:¶
File System - in this mechanism the identity credential is provisioned to the workload via the filesystem.¶
Local API - the identity credential is provided through an API, such as a local domain socket (for example SPIFFE or QEMU guest agent) or network API (for example Cloud Provider Metadata Server).¶
Environment Variables - identity credential may also be injected into workloads using operating system environment variables.¶
One of the most basic use cases for workload identity is authentication of one workload to another, such as in the case where one service is making a request to another service as part of a larger, more complex application. Following authentication, the request to the service offered by the workload needs to be authorized. Even in this simple case the identity of a workload is often composed of many attributes such as:¶
Trigger Information¶
Service Name¶
Instance Name¶
Role¶
Environment¶
Trust Domain¶
Software Attestation¶
Hardware Attestation¶
These attributes are used for various purposes such as:¶
ensuring the request is made to the correct service or service instance¶
authorizing access to APIs and resources¶
providing an audit trail of requests within a system¶
providing context for other decisions made within a service¶
There are several methods defined to perform this authentication. Some of the most common include:¶
TLS authentication of the server using X.509 certificates and client bearer token, encoded as JWTs.¶
Mutual TLS authentication using X.509 certificate for both client and server.¶
TLS authentication of the server and HTTP request signing using a secret key.¶
Figure 2 illustrates the communication between different workloads. Two aspects are important to highlight: First, there is a need to consider the interaction with workloads that are external to the trust domain (sometimes called cross-domain). Second, the interaction does not only occur between workloads that directly interact with each other but instead may also take place across intermediate workloads (in an end-to-end style).¶
In a typical system of workloads additional information is needed in order for the workload to perform its function. For example, it is common for a workload to require information about a user or other entity that originated the request. Other types of information may include information about the hardware or software that the workload is running or information about what processing and validation has already been done to the request. This type of information is part of the security context that the workload uses during authorization, accounting and auditing. This context is propagated and possibly augmented from workload to workload using tokens. Workload identity comes into play to ensure that the information in the context can only be used by an authorized workload and that the context information originated from an authorized workload.¶
When source workloads send authenticated requests to destination workloads, those destination workloads may rely on upstream dependencies to fulfill such requests. Such access patterns are increasingly common in a microservices architecture. While X.509 certificates can be used for point-to-point authentication, such services relying on upstream microservices for answers, may use delegation and/or impersonation semantics as described in RFC 8693 OAuth 2.0 Access Token Exchange.¶
WIMSE credentials constrain the subjects and actors identified in delegation and impersonation tokens to be bound by TrustDomains, and to follow their issuing authorities' trust configurations. Upstream workloads should consider the security context of delegation and/or impersonation tokens within and across Trust Domains, when arriving at authorization decisions.¶
Source workloads may send authenticated asynchronous and batch requests to destination workloads. A destination workload may need to fulfill such requests with requests to authorized upstream protected resources and workloads, after the source workload credentials have expired. Credentials identifying the original source workload as subject may need to be obtained from the credential issuing authority with appropriately-downscoped context needed access to upstream workloads. These credentials should identify the workload as the actor in the actor chain, but may also identify other principals that the action is taken on behalf. To mitigate risks associated with long-duration credentials, these credentials should be bound to the workload identity credential such as an X.509 certificate or Workload Identity Token (WIT) of the acting service performing asynchronous computation on the source workload's behalf.¶
As workloads often need to communicate across trust boundaries, extra care needs to be taken when it comes to identity communication to ensure scalability and privacy. (TODO: align with OAuth cross domain identity and authorization)¶
A workload communicating with a service, or another workload located outside the trust boundary may need to provide modified identity information. Detailed identity of internal workload originating the communication is relevant inside the trust boundary but could be excessive for the outside world and expose potentially sensitive internal topology information.¶
A security gateway at the edge of a trust boundary can be used to validate identity information of the workload, perform context specific authorization of the transaction and replace workload specific identity with a generalized one for a given trust domain.¶
Inbound security gateway is a common design pattern for service protection. This functionality is often found in CDN services, API gateways, load balancers, Web Application Firewalls (WAFs) and other security solutions. Workload identity verification of inbound requests should be performed as a part of these security services. After validation of workload identity, the gateway may either leave it unmodified or replace it with its own identity to be validated by the destination.¶
Workloads communicating with applications may face different threats to traffic interception in different deployments. In many deployments security controls are deployed for internal communications at lower layers to reduce the risk of traffic observation and modification for network communications. When a security layer, such as TLS, is deployed in these environments. TLS may be terminated in various places, including the workload itself, and in various middleware devices, such as load balancers, gateways, proxies, and firewalls. Therefore, protection is provided only between each adjacent pair of TLS endpoints. There are no guarantees of confidentiality, integrity and correct identity passthrough in those middleware devices and services.¶
Observation and interception of network traffic is not the only means of disclosure in these systems. Other vectors of information leakage is through disclosure in log files and other observability and troubleshooting mechanisms. For example, an application may log the contents of HTTP headers containing JWT bearer tokens. The information in these logs may be made available to other systems with less stringent access controls, which may result in this token falling into an attackers hands who then uses it to compromise a system. This creates privacy risks and potential surface for reconnaissance attacks. If observed tokens can be reused, this also may allow attackers to impersonate workloads.¶
Even the most well-designed and implemented workloads may contain security flaws that allow an attacker to gain limited or full compromise. For example, a server side request forgery may result in the ability for an attacker to force the workload to make requests of other parts of a system even though the rest of the workload functionality may be unaffected. An attacker with this advantage may be able to utilize privileges of the compromised workload to attack other parts of the system. Therefore it is important that communicating workloads apply the principle of least privilege through security controls such as authorization.¶
This document has no IANA actions.¶
Todo: Add your name here.¶