An example file, conf/examples/oidc-attribute-resolver.xml, is included that contains some examples of custom encoders, definitions that match the optional "default" claim rules, and suggested starting points for producing the "sub" claim (see also OPSubClaim).
Timing of Resolution
The Authorize, Token and Userinfo profile endpoints all trigger attribute resolution and filtering phases. The motivation for this is to maintain an implementation that is as stateless as possible to avoid complex clustering technologies. Each endpoint is able to resolve attributes independently and therefore attribute information need not be stored to server side. This also ensures up to date values for claims returned when an access token is refreshed for extended period of time.
(For long time Shibboleth deployers, this is somewhat analagous to the old days when SAML attribute queries were common.)
There are a few obvious downsides to this stateless approach. One is that, depending on the OIDC response type, the same attribute may be resolved more than once. This is particularly an issue if the resolution process is costly and the intention is only to authenticate the user. The ResultCache feature is helpful for this case.
A harder problem is when attribute resolution on the back-channel in the Token and Userinfo endpoints are unable to resolve attributes based on session/authentication/client information, as it is unavailable. This places additional requirements on the resolver configuration to fail gracefully in such cases. As an alternative, there is an encodedAttributes profile configuration property that instructs the Authorize endpoint to encode the attributes needed in later phases into the Authorization Code and/or Access Token directly, making them available to the Token and Userinfo endpoints. This naturally increases the size of the code/token but for specific cases may be tolerable.
Because OIDC is string-based, rather than reliant on XML and URIs for uniqueness, there are a number of arbitrary claim names that need to be avoided and will not be produced if used:
The "sub" claim is also semi-reserved but does come from your configuration. However it has to meet certain requirements and so cannot just contain arbitrary data without risking severe consequences to RPs. It is analagous to violating the expectations of a SAML SP regarding the content of an Attribute, but with more consistently severe problems. Please refer to the dedicated page at OPSubClaim for specifics.
The mappings from traditional LDAP (and SAML) attributes and OIDC claims is much less self-evident, but we have produced a set of plausible default rules that map from typical LDAP attributes to appropriate claims. These rules are not enabled by default, but can be found in conf/attributes/oidc-claim-rules.xml and imported into conf/attributes/default-rules.xml (with or without changes). That file also provides numerous examples to follow in building your own rules if desired.
Note that it's optional and not required to "combine" SAML and OIDC rules together. The system is intelligent enough to allow multiple "outbound" rules associated with the same IdPAttribute id with different transcoders involved.
The additional supported properties and transcoder types are described below. As with SAML, OIDC includes mechanisms to request claims, and so the transcoders support bidirectional mappings to allow decoding of requested JSON claims as well as encoding of IdPAttributes into JSON.
The claim name to map to and from (if absent, the IdPAttribute's id is used)
Encodes and decodes multiple values as a JSON array
Encodes and decodes individual values as a JSON integer
Encodes and decodes individual values as a Boolean
Encodes and decodes multiple values as a string with a specifie delimiter
There are 3 built-in types of OIDC transcoders, as follows. Each one is predefined as a Spring bean for use in rules using the "short" name of the transcoder as shown.
The simplest and most commonly used transcoder, it supports encoding and decoding internal values from and to the StringAttributeValue class. It supports the following additional optional property:
Encodes and decodes a string value as a JSON object (meaning parsing to and from JSON)
Supports encoding and decoding internal values from and to the ScopedStringAttributeValue class. It supports the following additional properties (all optional):
The character(s) to use to separate the value and scope
Supports encoding and decoding internal values from and to the ByteAttributeValue class, with a base64 transform applied. It supports no additional properties.
The alternative to the generality of the transcoding approach is the older style of embedding <AttributeEncoder> elements within the AttributeResolverConfiguration to specify individual encoding of AttributeDefinitions to claims. This is supported for OIDC via a set of extensions that add new encoder plugin types.
Note for upgraders: the older "token placement" settings that were supported here for splitting claims have been moved to profile configuration settings and are no longer valid here.
Because these are extensions, their xsi:type definitions live in a separate XML namespace and schema (unfortunately an unavoidable callback to the days when multiple namespaces were required in configurations).
In order to use any of these encoders, the root element of the resolver configuration file needs to be adjusted to declare a namespace prefix for the extension namespace and add in the schema location to address some fundamental XML bugs in Spring: