Internet-Draft | WIMSE S2S Auth | July 2024 |
Campbell, et al. | Expires 4 January 2025 | [Page] |
The WIMSE architecture defines authentication and authorization for software workloads in a variety of runtime environments, from the most basic ones up to complex multi-service, multi-cloud, multi-tenant deployments. This document defines the simplest, atomic unit of this architecture: the protocol between two workloads that need to verify each other's identity in order to communicate securely. The scope of this protocol is a single HTTP request-and-response pair. To address the needs of different setups, we propose two protocols, one at the application level and one that makes use of trusted TLS transport. These two protocols are compatible, in the sense that a single call chain can have some calls use one protocol and some use the other. Service A can call Service B with mutual TLS authentication, while the next call from Service B to Service C would be authenticated at the application level.¶
This note is to be removed before publishing as an RFC.¶
The latest revision of this draft can be found at https://yaronf.github.io/wimse-s2s/draft-sheffer-wimse-s2s-protocol.html. Status information for this document may be found at https://datatracker.ietf.org/doc/draft-sheffer-wimse-s2s-protocol/.¶
Discussion of this document takes place on the Workload Identity in Multi System Environments Working Group mailing list (mailto:wimse@ietf.org), which is archived at https://mailarchive.ietf.org/arch/browse/wimse/. Subscribe at https://www.ietf.org/mailman/listinfo/wimse/.¶
Source for this draft and an issue tracker can be found at https://github.com/yaronf/wimse-s2s.¶
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 4 January 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.¶
This document defines authentication and authorization in the context of interaction between two workloads. This is the core component of the WIMSE architecture [I-D.ietf-wimse-arch]. Assume that Service A needs to call Service B. For simplicity, this document focuses on HTTP-based services, and the service-to-service call consists of a single HTTP request and its response. We define the credentials that both services should possess and how they are used to protect the HTTP exchange.¶
There are multiple deployment styles in use today, and they result in different security properties. We propose to address them differently.¶
Many use cases have various middleboxes inserted between pairs of services, resulting in a transport layer that is not end-to-end encrypted. We propose to address these use cases by protecting the HTTP messages at the application level (Section 4).¶
The other commonly deployed architecture has a mutual-TLS connection between each pair of services. This setup can be addressed by a simpler solution (Section 5).¶
It is an explicit goal of this protocol that a service deployment can include both architectures across a multi-chain call. In other words, Service A can call Service B with mutual TLS protection, while the next call to Service C is protected at the application level.¶
For application-level protection we currently propose two alternative solutions, one inspired by DPoP [RFC9449] in Section 4.2 and one which is a profile of HTTP Message Signatures [RFC9421] in Section 4.3. The design team believes that we need to pick one of these two alternatives for standardization, once we have understood their pros and cons.¶
Regardless of the transport between the workloads, we assume the following logical architecture:¶
The Identity Server provisions credentials to each of the workloads. At least Workload A (and possibly both) must be provisioned with a credential before the call can proceed. Details of communication with the Identity Server are out of scope of this document, however we do describe the credential received by the workload.¶
PEP is a Policy Enforcement Point, the component that allows the call to go through or blocks it. PDP is an optional Policy Decision Point, which may be deployed in architectures where policy management is centralized. All details of policy management and message authorization are out of scope of this document.¶
The high-level message flow is as follows:¶
Workload A obtains a credential from the Identity Server. This happens periodically, e.g. once every 24 hours.¶
Workload A makes an HTTP call into Workload B. This is a regular HTTP request, with the additional protection mechanisms defined below.¶
Workload B now authenticates Workload A and decides whether to authorize the call. In certain architectures, Workload B may need to consult with an external server to decide whether to accept the call.¶
Workload B returns a response to Workload A, which may be an error response or a regular one.¶
This document uses "service" and "workload" interchangeably. Otherwise, all terms are as defined by [I-D.ietf-wimse-arch].¶
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 defines a workload identity as a URI [RFC3986]. This URI is used in the subject fields in the certificates and tokens defined later in this document. This specification treats the URI as opaque. The format of the URI and the namespace for the URI are at the discretion of the deployment at large. Other specifications may define specific URI structures for particular use cases. An example of a defined identity format is the SPIFFE ID.¶
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.¶
As noted in the Introduction, for many deployments communication between workloads cannot use end-to-end TLS. For these deployment styles, this document proposes application-level protections.¶
The current version of the document includes two alternatives, both using the newly introduced Workload Identity Token (Section 4.1). The first alternative (Section 4.2) is inspired by the OAuth DPoP specification. The second (Section 4.3) is based on the HTTP Message Signatures RFC. We present both alternatives and expect the working group to select one of them as this document progresses towards IETF consensus.¶
The Workload Identity Token (WIT) is a JWS [RFC7515] signed JWT [RFC7519] that represents the identity of a workload. It is issued by the Identity Server and binds a public key to the workload identity. A WIT MUST contain the following:¶
in the JOSE header:¶
alg
: An identifier for a JWS asymmetric digital signature algorithm
(registered algorithm identifiers are listed in the IANA JOSE Algorithms registry [IANA.JOSE.ALGS]). The value none
MUST NOT be used.¶
typ
: the WIT is explicitly typed, as recommended in Section 3.11 of [RFC8725], using the wimse-id+jwt
media type.¶
in the JWT claims:¶
iss
: The issuer of the token, which is the Identity Server, represented by a URI.¶
sub
: The subject of the token, which is the identity of the workload, represented by a URI.¶
exp
: The expiration time of the token (as defined in Section 4.1.4 of [RFC7519]).
WITs should be refreshed regularly, e.g. on the order of hours.¶
jti
: A unique identifier for the token.¶
cnf
: A confirmation claim containing the public key of the workload using the jwk
member as defined in Section 3.2 of [RFC7800].
The workload MUST prove possession of the corresponding private key when presenting the WIT to another party, which can be accomplished by using it in conjunction with one of the methods in Section 4.2 or Section 4.3. As such, it MUST NOT be used as a bearer token and is not intended for use in the Authorization
header.¶
An example WIT might look like this (all examples, of course, are non-normative and with line breaks and extra space for readability):¶
The decoded JOSE header of the WIT from the example above is shown here:¶
The decoded JWT claims of the WIT from the example above are shown here:¶
The claims indicate that the example WIT:¶
was issued by an Identity Server known as wimse://example.com/trusted-central-authority
.¶
is valid until May 15, 2024 3:28:45 PM GMT-06:00 (represented as NumericDate Section 2 of [RFC7519] value 1717612470
).¶
identifies the workload to which the token was issued as wimse://example.com/specific-workload
.¶
has a unique identifier of x-_1CTL2cca3CSE4cwb__
.¶
binds the public key represented by the jwk
confirmation method to the workload wimse://example.com/specific-workload
.¶
For elucidative purposes only, the workload's key, including the private part, is shown below in JWK [RFC7517] format:¶
The afore-exampled WIT is signed with the private key of the Identity Server. The public key(s) of the Identity Server need to be known to all workloads in order to verify the signature of the WIT. The Identity Server's public key from this example is shown below in JWK [RFC7517] format:¶
A WIT is conveyed in an HTTP header field named Workload-Identity-Token
.¶
For those who celebrate, ABNF [RFC5234] for the value of Workload-Identity-Token
header field is provided in Figure 6:¶
The following shows the WIT from Figure 1 in an example of a Workload-Identity-Token
header field:¶
Note that per [RFC9110], header field names are case insensitive;
thus, Workload-Identity-Token
, workload-identity-token
, WORKLOAD-IDENTITY-TOKEN
,
etc., are all valid and equivalent header field names. However, case is significant in the header field value.¶
This option, inspired by the OAuth DPoP specification [RFC9449], uses a DPoP-like mechanism to authenticate
the calling workload in the context of the request. The WIMSE Identity Token (Section 4.1) is sent in the request as
described in Section 4.1.1. An additional JWT, the Workload Proof Token (WPT), is signed by the private key
corresponding to the public key in the WIT. The WPT is sent in the Workload-Proof-Token
header field of the request.
A WPT contains the following:¶
in the JOSE header:¶
alg
: An identifier for an appropriate JWS asymmetric digital signature algorithm corresponding to
the confirmation key in the associated WIT.¶
typ
: the WPT is explicitly typed, as recommended in Section 3.11 of [RFC8725],
using the application/wimse-proof+jwt
media type.¶
in the JWT claims:¶
iss
: The issuer of the token, which is the calling workload, represented by the same value as the sub
claim
of the associated WIT.¶
aud
: The audience of the token contains the HTTP target URI (Section 7.1 of [RFC9110]) of the request
to which the WPT is attached, without query or fragment parts.¶
exp
: The expiration time of the WIT (as defined in Section 4.1.4 of [RFC7519]). WPT lifetimes MUST be short,
e.g., on the order of minutes or seconds.¶
jti
: A unique identifier for the token.¶
ath
: Hash of the OAuth access token, if present in the request, which might convey end-user identity and
authorization context of the request. The value, as per Section 4.1 of [RFC9449],
is the base64url encoding of the SHA-256 hash of the ASCII encoding of the access token's value.¶
tth
: Hash of the Txn-Token [I-D.ietf-oauth-transaction-tokens], if present in the request,
which might convey end-user identity and authorization context of the request. The value MUST be the result of
a base64url encoding (as defined in Section 2 of [RFC7515]) of the SHA-256 hash of
the ASCII encoding of the associated token's value.¶
oth
: Hash of any other token in the request that might convey end-user identity and authorization context of the
request. The value MUST be the result of a base64url encoding (as defined in Section 2 of [RFC7515]) of the
SHA-256 hash of the ASCII encoding of the associated token's value.
(note: this is less than ideal but seems we need something like this for extensibility)¶
An example WPT might look like the following:¶
The decoded JOSE header of the WPT from the example above is shown here:¶
The decoded JWT claims of the WPT from the example above are shown here:¶
An example of an HTTP request with both the WIT and WPT from prior examples is shown below:¶
To validate the WPT in the request, the recipient MUST ensure the following:¶
There is exactly one Workload-Proof-Token
header field in the request.¶
The Workload-Proof-Token
header field value is a single and well-formed JWT.¶
The WPT signature is valid using the public key from the confirmation claim of the WIT.¶
The typ
JOSE header parameter of the WPT conveys a media type of wimse-proof+jwt
.¶
The iss
claim of the WPT matches the sub
claim of the WIT. (note: not sure iss
in the WPT is useful or necessary)¶
The aud
claim of the WPT matches the target URI, or an acceptable alias or normalization thereof, of the HTTP request
in which the WPT was received, ignoring any query and fragment parts.¶
The exp
claim is present and conveys a time that has not passed. WPTs with an expiration time unreasonably
far in the future SHOULD be rejected.¶
Optionally, check that the value of the jti
claim has not been used before in the time window in which the
respective WPT would be considered valid.¶
If presented in conjunction with an OAuth access token, the value of the ath
claim matches the hash of that token's value.¶
If presented in conjunction with a Txn-Token, the value of the tth
claim matches the hash of that token's value.¶
If presented in conjunction with a token conveying end-user identity or authorization context, the value of
the oth
claim matches the hash of that token's value.¶
This option uses the WIMSE Identity Token (Section 4.1) to sign the request and optionally, the response. This specification defines a profile of the Message Signatures specification [RFC9421].¶
The request is signed as per [RFC9421]. The following derived components MUST be signed:¶
In addition, the following request headers MUST be signed when they exist:¶
Content-Type
¶
Content-Digest
¶
Authorization
¶
Txn-Token
[I-D.ietf-oauth-transaction-tokens]¶
Workload-Identity-Token
¶
If the response is signed, the following components MUST be signed:¶
@status
¶
@method;req
¶
@request-target;req
¶
Content-Type
if it exists¶
Content-Digest
if it exists¶
Workload-Identity-Token
¶
For both requests and responses, the following signature parameters MUST be included:¶
created
¶
expires
- expiration MUST be short, e.g. on the order of minutes. The WIMSE architecture will provide separate
mechanisms in support of long-lived compute processes.¶
nonce
¶
tag
- the value for implementations of this specification is wimse-service-to-service
¶
Since the signing key is sent along with the message, the keyid
parameter SHOULD NOT be used.¶
It is RECOMMENDED to include only one signature with the HTTP message.
If multiple ones are included, then the signature label included in both the Signature-Input
and Signature
headers SHOULD
be wimse
.¶
A sender MUST ensure that each nonce it generates is unique, at least among messages sent to the same recipient. To detect message replays, a recipient MAY reject a message (request or response) if a nonce is repeated.¶
To promote interoperability, the ecdsa-p256-sha256
signing algorithm MUST be implemented
by general purpose implementations of this spec.¶
OPEN ISSUE: do we use the `Accept-Signature` field to signal that the response must be signed?¶
Following is a non-normative example of a signed request and a signed response, where the caller is using the keys specified in Figure 4.¶
Assuming that the workload being called has the following keypair:¶
A signed response would be:¶
The WIMSE workload identity may be carried within an X.509 certificate. The WIMSE workload identity MUST be encoded in a SubjectAltName extension of type URI. There MUST be only one SubjectAltName extension of type URI in a WIMSE certificate. The WIMSE certificate may contain SubjectAltName extensions of other types such as DNSName.¶
WIMSE identities may be used to validate server and client connections. When validating a WIMSE identity the relying party MUST validate that the CA issuer for the WIMSE identity is authorized to issue certificates for the trust domain of the WIMSE identity in the certificate. Other PKIX path validation rules apply.¶
Servers wishing to use the WIMSE identity for authorizing the client MUST require client certificate authentication in the TLS handshake. Other methods of post handshake authentication are not specified by this document.¶
WIMSE clients and servers MUST validate that the trust domain portion of the WIMSE certificate matches the expected trust domain for the other side of the connection.¶
TODO: need to define trust root used to validate the certificate is appropriate for the trust domain.¶
It is RECOMMENDED that the server certificate contain a DNS SAN that the client can use to perform standard host name validation (Section 6.3 of [RFC9525]). The client SHOULD also extract the WIMSE identity from the certificate if it is present and validate that the WIMSE trust domain matches the intended trust domain for the server. The client MAY then further use the WIMSE identity in applying authorization policy to the server. If the client does not use the DNS SAN then the client MUST match the WIMSE identity in the certificate against the WIMSE identity of the workload of the intended server according to a locally defined policy. The host portion of the WIMSE URI is NOT treated as a host name as specified in section 6.4 of [RFC9525] but rather as a trust domain. The server identity is encoded in the path portion of the WIMSE identity in a deployment specific way.¶
The WIMSE ID is scoped within an issuer and therefore any sub-components (path portion of ID) are only unique within a trust domain defined by the issuer. Using a WIMSE ID without taking into account the trust domain could allow one domain to issue tokens to spoof identities in another domain. Additionally, the trust domain must be tied to an authorized issuer cryptographic trust root through some mechanism such as a JWKS or X.509 certificate chain. The association of an issuer, trust domain and a cryptographic trust root MUST be communicated securely out of band.¶
TODO: Should there be a DNS name to Trust domain mapping defined or some other discovery mechanism?¶
The Workload ID token (WIT) is bound to a secret cryptographic key and is always presented with a proof of possession as described in Section 4.1. The WIT is a general purpose token that can be presented in multiple contexts. The WIT and its PoP are only used in the application-level options, and both are not used in MTLS. The WIT MUST NOT be used as a bearer token. While this helps reduce the sensitivity of the token it is still possible that a token and its proof of possession may be captured and replayed within the PoP's lifetime. The following are some mitigations for the capture and reuse of the proof of possession (PoP):¶
Preventing Eavesdropping and Interception with TLS¶
An attacker observing or intercepting the communication channel can view the token and its proof of possession and attempt to replay it to gain an advantage. In order to prevent this the token and proof of possession MUST be sent over a secure, server authenticated TLS connection unless a secure channel is provided by some other mechanisms. Host name validation according to Section 5 MUST be performed. The WIT itself is not usable without a proof of possession.¶
Limiting Proof of Possession Lifespan¶
The proof of possession MUST be time limited. A PoP should only be valid over the time necessary for it to be successfully used for the purpose it is needed. This will typically be on the order of minutes. PoPs received outside their validity time MUST be rejected.¶
Limiting Proof of Possession Scope¶
In order to reduce the risk of theft and replay the PoP should have a limited scope. For example, a PoP may be targeted for use with a specific workload and even a specific transaction to reduce the impact of a stolen PoP. In some cases a workload may wish to reuse a PoP for a period of time or have it accepted by multiple target workloads. A careful analysis is warranted to understand the impacts to the system if a PoP is disclosed allowing it to be presented by an attacker along with a captured WIT.¶
Binding to a Timestamp or Nonce¶
A proof of possession should include information that can be used to uniquely identify it such as a unique timestamp or nonce. This can be used by the receiver to perform basic replay protection against tokens it has already seen. Depending upon the design of the system it may be difficult to synchronize the replay cache across all token validators. In this case, if the PoP is not sufficiently scoped it may be usable with another workload. While a fresh nonce could be included in the PoP, a mechanism for distributing a fresh challenge nonce from the validator to provide single use properties of a PoP is outside the scope of this specification.¶
Binding to TLS Endpoint¶
The POP MAY be bound to a transport layer sender such as the client identity of a TLS session or TLS channel binding parameters. The mechanisms for binding are outside the scope of this specification.¶
In some deployments the WIMSE token and proof of possession may pass through multiple systems. The communication between the systems is over TLS, but the token and PoP are available in the clear at each intermediary. While the intermediary cannot modify the token or the information within the PoP they can attempt to capture and replay the token or modify the data not protected by the PoP. Mitigations listed in the previous section can be used to provide some protection from middle boxes. Deployments should perform analysis on their situation to determine if it is appropriate to trust and allow traffic to pass through a middle box.¶
WITs and the proofs of possession may contain private information such as user names or other identities. Care should be taken to prevent the disclosure of this information. The use of TLS helps protect the privacy of WITs and proofs of possession.¶
WITs and certificates with WIMSE identifiers are typically associated with a workload and not a specific user, however in some deployments the workload may be associated directly to a user. While these are exceptional cases a deployment should evaluate if the disclosure of WITs or certificates can be used to track a user.¶
TODO: maybe a URI Scheme registration of wimse
in URI schemes per [RFC7595] but it's only being used in an example right now and might not even be appropriate. Or maybe use an ietf URI scheme a la URN Namespace for IETF Use somehow. Or maybe nothing. Or maybe something else.¶
TODO: tth
and maybe oth
claim in JSON Web Token Claims Registry¶
TODO: application/wimse-id+jwt
or appropriately bikeshedded media type name (despite my ongoing unease with using media types for typing JWTs) in Media Types.¶
TODO: application/wimse-proof+jwt
...¶
TODO: Workload-Identity-Token
from Section 4.1.1¶
TODO: Workload-Proof-Token
from Section 4.2¶
TODO acknowledge.¶