fix: platform-wide glob role assignments crash get_orgs_for_user/has_org_for_user - #38980
fix: platform-wide glob role assignments crash get_orgs_for_user/has_org_for_user#38980efortish wants to merge 7 commits into
Conversation
…org_for_user RoleBase._authz_get_orgs_for_user() read assignment.scope.org for every AuthZ assignment, but PlatformGlobData (course-v1:*, lib:*) has no .org attribute at all, unlike CourseOverviewData/OrgGlobData where it's a real (possibly None) field. Any user with a platform-wide role crashed get_orgs_for_user()/has_org_for_user() with an AttributeError. Special-case platform-wide assignments: since they cover every org, return all registered org short names instead of deriving them from individual assignments. Fixes openedx/openedx-authz#380
|
Thanks for the pull request, @efortish! This repository is currently maintained by Once you've gone through the following steps feel free to tag them in a comment and let them know that your changes are ready for engineering review. 🔘 Get product approvalIf you haven't already, check this list to see if your contribution needs to go through the product review process.
🔘 Provide contextTo help your reviewers and other members of the community understand the purpose and larger context of your changes, feel free to add as much of the following information to the PR description as you can:
🔘 Get a green buildIf one or more checks are failing, continue working on your changes until this is no longer the case and your build turns green. DetailsWhere can I find more information?If you'd like to get more details on all aspects of the review process for open source pull requests (OSPRs), check out the following resources: When can I expect my changes to be merged?Our goal is to get community contributions seen and reviewed as efficiently as possible. However, the amount of time that it takes to review and merge a PR can vary significantly based on factors such as:
💡 As a result it may take up to several weeks or months to complete a review and merge your PR. |
| RoleAssignmentData( | ||
| subject=UserData(external_key=self.student.username), | ||
| roles=[staff_authz_role], | ||
| scope=PlatformCourseOverviewGlobData(external_key="course-v1:*"), |
There was a problem hiding this comment.
| scope=PlatformCourseOverviewGlobData(external_key="course-v1:*"), | |
| scope=PlatformCourseOverviewGlobData.build_external_key(), |
And could we please update test_course_listing.py module to use the build_external_key method? I forgot to do that in some of the tests.
There was a problem hiding this comment.
Applied — wrapped it in PlatformCourseOverviewGlobData(...) since the literal suggestion would assign a bare string to scope (RoleAssignmentData.scope expects a ScopeData instance). Also updated test_course_listing.py's platform/org-glob tests to use build_external_key() as you asked. Fixed in d32aa52.
There was a problem hiding this comment.
Yes, you're right. Now that I'm taking a closer look at the test, instead of mocking the assignments, could we use the assign_role_to_user_in_scope function? Something like this:
assign_role_to_user_in_scope(
self.student.username,
COURSE_STAFF.external_key,
PlatformCourseOverviewGlobData.build_external_key(),
)There was a problem hiding this comment.
Good idea — this is actually stronger, since it exercises the real Casbin assignment + policy load instead of mocking get_user_role_assignments_filtered. It also caught a latent bug in my mock: I had RoleData(external_key=COURSE_STAFF) (the RoleData object itself) instead of COURSE_STAFF.external_key, which the mock silently tolerated since it bypassed the real filtering. Switched to assign_role_to_user_in_scope + load_policy() in 795de56.
Per review feedback from @BryanttV on openedx#38980, also applied to test_course_listing.py which had the same pattern in unrelated pre-existing tests.
…to ks/issue-380-platform-glob-org
…ents Per review feedback from @BryanttV on openedx#38980: exercising the real Casbin policy assignment + get_orgs_for_user path is more faithful than mocking get_user_role_assignments_filtered directly, and it catches issues the mock papered over (the mocked RoleData had external_key=COURSE_STAFF, the RoleData object itself, instead of COURSE_STAFF.external_key).
| # with a concrete assignment. Platform-glob scopes have no .org attribute at all | ||
| # (unlike org-glob/course/library scopes, where it's a real field that can be None). | ||
| if any(assignment.scope.IS_PLATFORM_GLOB for assignment in assignments): | ||
| return [org["short_name"] for org in get_organizations()] |
There was a problem hiding this comment.
Sorry for being late to this review. Does this return the same structure as in L641? Can we make sure of this with a test - like a test case that for a user returns a subset and for another all orgs? Not sure if we're doing that already. Thanks
There was a problem hiding this comment.
No worries about the timing! Yes — both branches return the same shape, a plain list[str] of org short names (L640 is a list comprehension over get_organizations(), L641 is list(a set) of assignment.scope.org values). Added test_get_orgs_for_user_authz_platform_glob_vs_org_scoped: it registers a third org that's never assigned to anyone, then asserts an org-scoped grant returns only its own org while a platform-wide grant returns all three, against that same shared pool of orgs — so the two branches are actually distinguished instead of just coincidentally returning the same numbers (which is what my original platform_glob test alone couldn't rule out, since it only ever registered exactly the orgs it expected back). Verified locally against a real devstack. Fixed in fbb2726.
Per review feedback from @mariajgrimaldi on openedx#38980: registers a third org that's never assigned, then asserts an org-scoped grant returns only its own org while a platform-wide grant returns all three, against the same pool of registered orgs — so the two branches (roles.py L640 vs L641, both list[str]) are actually distinguished by the test instead of coincidentally matching.
Per review feedback from @BryanttV on openedx#38986 (same feedback given earlier on openedx#38984/openedx#38980) — new test code in this repo should use plain assert, not unittest-style assertions with # noqa: PT009.
* fix: Course Auditor gets 403 navigating to a course unit xblock_outline_handler (the course outline tree) was already migrated to the AuthZ-aware user_has_course_permission(..., COURSES_VIEW_COURSE, ..., LegacyAuthoringPermission.READ) check, so it correctly recognizes AuthZ- native roles that have no legacy equivalent, like course_auditor and course_editor. xblock_container_handler (the unit/container page — what's hit when navigating to a unit) and xblock_view_handler (renders each child block's preview fragment on that page), plus xblock_edit_view, were never migrated the same way: they still called the legacy-only has_studio_read_access directly, which only recognizes roles with a legacy equivalent (staff/instructor/limited_staff). A Course Auditor has none, so get_user_permissions() returned no permissions and these handlers raised PermissionDenied, even though the outline (using the correct pattern) let the same user in. Migrate the three remaining read checks in block.py to the same user_has_course_permission pattern xblock_outline_handler already uses. The xblock_handler/handle_xblock CRUD endpoint was already correctly AuthZ-aware (via _check_xblock_permission) and needed no change. Fixes openedx/openedx-authz#384 * fix: also fix the REST API v1 container view that the Authoring MFE actually uses Manual testing against a real devstack found that the block.py fix alone wasn't enough: the modern Authoring MFE calls the REST API v1 ContainerHandlerView (/api/contentstore/v1/container_handler/...), which still 403'd for course_auditor. That view (and container_handler/container_embed_handler/xblock_edit_view in the legacy views) all route through the shared _get_item_in_course() helper in component.py, which gated on has_course_author_access — a legacy-only *write* check — even though all four callers only need read access to render a view. course_auditor has no legacy role equivalent, so it never had write access and always got PermissionDenied here, regardless of the block.py fix. Migrate _get_item_in_course() to the same user_has_course_permission read check, fixing all four callers (including the REST API v1 view) at their single shared choke point instead of patching each call site. Verified locally end-to-end: mounted this branch into a real devstack, assigned course_auditor to a test user, confirmed the unit page 403'd before this commit and loads correctly after it. Also ran the full test_block.py + test_vertical_block.py suites against the same devstack (186 passed). * fix: sort imports (ruff I001) * refactor: use plain assert instead of self.assertEqual in new tests Per review feedback from @BryanttV on #38986 (same feedback given earlier on #38984/#38980) — new test code in this repo should use plain assert, not unittest-style assertions with # noqa: PT009.
Description
Fixes openedx-authz#380
RoleBase._authz_get_orgs_for_user()(common/djangoapps/student/roles.py) readsassignment.scope.orgfor every AuthZ assignment. However,PlatformGlobData(thecourse-v1:*/lib:*platform-wide scopes) does not have an.orgattribute, unlikeCourseOverviewData/OrgGlobData, where.orgis a valid (possiblyNone) field.As a result, any user with a role assigned at a platform-wide scope would cause
get_orgs_for_user()/has_org_for_user()to fail with anAttributeError.The fix handles platform-wide assignments before accessing
.org. Since a platform-wide grant applies to every organization, the compatibility layer returns all registered organization short names instead of trying to derive them from individual assignments.Testing Instructions
authz.enable_course_authoringflag globally.course-v1:*).role.get_orgs_for_user(user)/role.has_org_for_user(user), or trigger any code path that uses them (e.g., the Studio course listing).AttributeErroris raised.Added
test_get_orgs_for_user_authz_platform_globincommon/djangoapps/student/tests/test_roles.pyto cover this case.