docs: add ADR for glob support for role assignments

CHANGELOG.rst

@@ -14,6 +14,8 @@ Change Log
 Unreleased
 **********
 
+* Add ADR for global scope support for role assignments.
+
 0.23.0 - 2026-02-18
 ********************
 

docs/decisions/0012-glob-support-for-role-assignments.rst

@@ -0,0 +1,154 @@
+0012: Glob Support For Role Assignments
+#######################################
+
+Status
+******
+
+**Draft** - *2026-03-18*
+
+Context
+*******
+
+The current authorization system is based on Casbin and models:
+
+- **Permissions per role** (``p`` policies), where the ``scope`` field may already use patterns (for example, ``lib^*``) through matcher functions in the model.
+- **Role assignments** (``g`` policies), which link a subject to a role within a scope.
+
+The current Casbin model treats the ``scope`` field in ``g`` policies as an **exact match**. This is sufficient when roles are granted for a single, concrete scope value, but it is limiting when operators need to:
+
+- Assign roles that are valid for a **set of resources** that share a common prefix (for example, all courses or all content libraries belonging to a given organization).
+- Avoid enumerating a large or evolving set of resources one by one, which increases operational overhead and risk of drift.
+
+This ADR proposes enabling glob-like matching for role assignments, so that a single ``g`` policy can represent a multi-scope assignment such as:
+
+.. code:: text
+
+   g, user^contributor, role^course_staff, course-v1^course-v1:OpenedX+*
+
+In that example, checking a permission such as:
+
+.. code:: text
+
+   authz_api.is_user_allowed("contributor", "courses.manage_advanced_settings", "course-v1:OpenedX+Some+Course")
+
+should be allowed if the user's role assignment matches the ``course-v1:OpenedX+*`` pattern.
+
+At the same time, we must preserve the guarantees of the authorization model:
+
+- **Safety**: Glob patterns must not accidentally grant permissions outside of the intended boundary.
+- **Clarity**: Patterns must be easy to understand and reason about for operators and auditors.
+- **Extensibility**: The mechanism should be general enough to support future use cases without requiring a redesign of the model.
+
+Decision
+********
+
+We will introduce support for glob-like matching on the ``scope`` field of **role assignments** (``g`` policies), combined with explicit validation in the public APIs that manage those assignments.
+
+The decision is intentionally **general**: the core change is to allow glob matching for scopes in ``g`` policies, and to guard its usage with well-defined, namespace-specific validation rules. This creates a foundation that can be extended later without changing the Casbin model again.
+
+1. Enable glob matching on ``g`` scopes in the enforcer
+=======================================================
+
+We will configure the ``AuthzEnforcer`` to use a domain/scope matching function for ``g`` policies that supports glob-like suffixes. Concretely:
+
+- The enforcer will register a domain matching function for the ``g`` (grouping) function (for example, using ``key_match_func``).
+- This matching function will treat ``*`` as a wildcard at the **end** of the string. That is, patterns such as ``course-v1:OpenedX+*`` will match ``course-v1:OpenedX+SOME+COURSE``, but the model will not rely on complex patterns or regular expressions.
+- Existing ``g`` policies that use exact scopes remain valid and continue to behave identically.
+
+This change allows the Casbin engine to evaluate role assignments that apply to a family of scopes instead of a single exact value, without modifying the underlying storage schema (``CasbinRule``) or the overall request format (``r = sub, act, scope``).
+
+2. Validate glob scopes at the API boundary
+===========================================
+
+All APIs that create, update, or delete role assignments (i.e., policies of type ``g``) must validate any scope that includes a glob pattern. The goals of validation are:
+
+- **Constrain** what forms of glob are permitted for each namespace.
+- **Reject** malformed or overly broad patterns that would be difficult to reason about or audit.
+
+The following rules apply initially:
+
+- The glob character (``*``) is only supported as a **suffix wildcard**. It cannot appear in the middle of a scope identifier.
+- A glob pattern represents a **bounded prefix match** for the external key portion of a scope within its namespace. The API validation ensures the prefix is meaningful (i.e., it corresponds to a valid identifier boundary for that namespace), so the glob cannot be used to accidentally broaden access.
+- For any glob patterns (courses, libraries, or future namespaces), malformed inputs (such as mid-string wildcards or prefixes that do not match the expected key format/boundaries) are **rejected**.
+- Additional namespaces must define their own, explicit validation rules before accepting glob scopes.
+- As needs evolve, more glob types can be added safely by introducing namespace-specific semantics and validations (for example, additional prefix boundaries such as program/tenant prefixes, or narrower matching strategies if required).
+
+These validation rules are implemented in the Open edX layer (API / data layer), not in the Casbin matcher itself. The enforcer remains general-purpose. The domain-specific semantics of what constitutes an acceptable glob pattern are enforced at the boundary where user/operator input is turned into policies.
+
+3. Keep the model general and extensible
+========================================
+
+By introducing glob support in role assignments in a constrained way, we unlock a set of future extensions without redesigning the model:
+
+- **Other scope types**
+
+  - Scope types with hierarchical or prefix-based identifiers (for example, libraries or other content groupings) can adopt glob support by:
+
+    - Defining their own namespace-specific rules for valid suffix globs.
+    - Reusing the same enforcer-level domain matching capability.
+
+- **Future matching strategies**
+
+  - If, in the future, there is a strong need for more expressive matching (for example, segment-based matching or multiple wildcards), these can be introduced as **new, explicitly-scoped features** with their own validation rules and migration story.
+  - For now, we deliberately keep glob support simple and limited (single trailing ``*``) to minimize complexity and security risk.
+
+Consequences
+************
+
+Positive consequences
+=====================
+
+- **Increased expressiveness**: Operators can express multi-scope role assignments (for example, "course staff for all courses in organization OpenedX") without enumerating each course in individual ``g`` policies.
+- **Reduced operational overhead**: New resources that fall under an existing glob pattern automatically inherit the appropriate role assignments, reducing the need for ongoing manual updates.
+- **Better alignment with real-world use cases**: Many organizational setups naturally require "all resources under this prefix" semantics. Glob support maps directly to those needs.
+- **Clear extension path**: The mechanism is generic enough to be reused for other namespaces (such as organization or library scopes), as long as each namespace defines and enforces its own validation rules.
+
+Negative consequences / risks
+=============================
+
+- **Security and safety**: If validation is misconfigured or bypassed, glob patterns could unintentionally grant access beyond the intended boundary. This risk is mitigated by:
+
+  - Enforcing validation in the Open edX API layer.
+  - Restricting globs to trailing ``*`` patterns.
+  - Defining precise, namespace-specific rules.
+
+- **Complexity in mental model**: Operators and developers must understand that some role assignments apply to families of scopes instead of a single scope. This can be addressed by:
+
+  - Providing clear documentation and examples for glob-based assignments.
+  - Exposing introspection tooling that explains which policies matched a given decision.
+
+- **Performance considerations**: Glob matching adds some overhead to Casbin evaluations. However:
+
+  - The cost of simple suffix matching is low.
+  - The policy store still uses the same schema and indexing strategy.
+  - The feature should be used primarily for coarse-grained groupings (e.g., by organization), not for highly fragmented patterns.
+
+Rejected Alternatives
+**********************
+
+- **Keep exact matching only for 'g' scopes**
+
+  - Pros:
+
+    - Simpler to reason about.
+    - No changes to matcher configuration.
+  - Cons:
+
+    - Does not scale for environments with many resources per organization.
+    - Forces operators to maintain large numbers of nearly-identical assignments.
+
+- **Introduce full regular-expression support on scopes**
+
+  - Pros:
+
+    - Maximum flexibility for expressing patterns.
+  - Cons:
+
+    - Harder to reason about and audit.
+    - Higher risk of misconfiguration and security overshoot.
+    - Potentially worse performance.
+
+References
+**********
+
+- `Casbin function documentation (matching functions) <https://casbin.org/docs/function/>`_