Refactors authorization to use b.access.Check() with verb-based checks:
Repository subresources:
- CRUD: uses actual verb (get/create/update/delete)
- test: uses 'update' (write permission)
- files/refs/resources/history/status: uses 'get' (read permission)
- jobs: uses actual verb for jobs resource
Connection subresources:
- CRUD: uses actual verb
- status: uses 'get' (read permission)
The access checker maps verbs to actions defined in accesscontrol.go.
Falls back to admin role for backwards compatibility.
Also removes redundant admin check from listFolderFiles since
authorization is now properly handled at route level.
Adds connection permissions following the same pattern as repositories:
- provisioning.connections:create
- provisioning.connections:read
- provisioning.connections:write
- provisioning.connections:delete
Roles:
- fixed:provisioning.connections:reader (granted to Admin)
- fixed:provisioning.connections:writer (granted to Admin)
Simplifies authorization by:
- Adding isAccessPolicy() helper for AccessPolicy identity check
- Adding allowForAdminsOrAccessPolicy() that returns Decision directly
- Consolidating stats/settings/jobs into single switch case
- Using consistent pattern in files.go
The admin role check in listFolderFiles was redundant (route-level auth already
handles access) and broken for AccessPolicy identities which don't have org roles.
File access is controlled by the AccessClient as documented in the route-level
authorization comment.
These are read-only endpoints needed by the frontend:
- settings: returns available repository types and configuration for the wizard
- stats: returns resource counts
Authentication is verified before reaching authorization, so any user who
reaches these endpoints is already authenticated. Requiring specific org
roles failed for AccessPolicy tokens which don't carry traditional roles.
* Deprecate Legacy Storage Migration in Backend
* Change the messaging around legacy storage
* Disable cards to connect
* Commit import changes
* Block repository creation if resources are in legacy storage
* Update error message
* Prettify
* chore: uncomment unified migration
* chore: adapt and fix tests
* Remove legacy storage migration from frontend
* Refactor provisioning job options by removing legacy storage and history fields
- Removed the `History` field from `MigrateJobOptions` and related references in the codebase.
- Eliminated the `LegacyStorage` field from `RepositoryViewList` and its associated comments.
- Updated tests and generated OpenAPI schema to reflect these changes.
- Simplified the `MigrationWorker` by removing dependencies on legacy storage checks.
* Refactor OpenAPI schema and tests to remove deprecated fields
- Removed the `history` field from `MigrateJobOptions` and updated the OpenAPI schema accordingly.
- Eliminated the `legacyStorage` field from `RepositoryViewList` and its associated comments in the schema.
- Updated integration tests to reflect the removal of these fields.
* Fix typescript errors
* Refactor provisioning code to remove legacy storage dependencies
- Eliminated references to `dualwrite.Service` and related legacy storage checks across multiple files.
- Updated `APIBuilder`, `RepositoryController`, and `SyncWorker` to streamline resource handling without legacy storage considerations.
- Adjusted tests to reflect the removal of legacy storage mocks and dependencies, ensuring cleaner and more maintainable code.
* Fix unit tests
* Remove more references to legacy
* Enhance provisioning wizard with migration options
- Added a checkbox for migrating existing resources in the BootstrapStep component.
- Updated the form context to track the new migration option.
- Adjusted the SynchronizeStep and useCreateSyncJob hook to incorporate the migration logic.
- Enhanced localization with new descriptions and labels for migration features.
* Remove unused variable and dualwrite reference in provisioning code
- Eliminated an unused variable declaration in `provisioning_manifest.go`.
- Removed the `nil` reference for dualwrite in `repo_operator.go`, aligning with the standalone operator's assumption of unified storage.
* Update go.mod and go.sum to include new dependencies
- Added `github.com/grafana/grafana-app-sdk` version `0.48.5` and several indirect dependencies including `github.com/getkin/kin-openapi`, `github.com/hashicorp/errwrap`, and others.
- Updated `go.sum` to reflect the new dependencies and their respective versions.
* Refactor provisioning components for improved readability
- Simplified the import statement in HomePage.tsx by removing unnecessary line breaks.
- Consolidated props in the SynchronizeStep component for cleaner code.
- Enhanced the layout of the ProvisioningWizard component by streamlining the rendering of the SynchronizeStep.
* Deprecate MigrationWorker and clean up related comments
- Removed the deprecated MigrationWorker implementation and its associated comments from the provisioning code.
- This change reflects the ongoing effort to eliminate legacy components and improve code maintainability.
* Fix linting issues
* Add explicit comment
* Update useResourceStats hook in BootstrapStep component to accept selected target
- Modified the BootstrapStep component to pass the selected target to the useResourceStats hook.
- Updated related tests to reflect the change in expected arguments for the useResourceStats hook.
* fix(provisioning): Update migrate tests to match export-then-sync behavior for all repository types
Updates test expectations for folder-type repositories to match the
implementation changes where both folder and instance repository types
now run export followed by sync. Only the namespace cleaner is skipped
for folder-type repositories.
Changes:
- Update "should run export and sync for folder-type repositories" test to include export mocks
- Update "should fail when sync job fails for folder-type repositories" test to include export mocks
- Rename test to clarify that both export and sync run for folder types
- Add proper mock expectations for SetMessage, StrictMaxErrors, Process, and ResetResults
All migrate package tests now pass.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* Update provisioning wizard text and improve resource counting display
- Enhanced descriptions for migrating existing resources to clarify that unmanaged resources will also be included.
- Refactored BootstrapStepResourceCounting component to simplify the rendering logic and ensure both external storage and unmanaged resources are displayed correctly.
- Updated alert messages in SynchronizeStep to reflect accurate information regarding resource management during migration.
- Adjusted localization strings for consistency with the new descriptions.
* Update provisioning wizard alert messages for clarity and accuracy
- Revised alert points to indicate that resources can still be modified during migration, with a note on potential export issues.
- Clarified that resources will be marked as managed post-provisioning and that dashboards remain accessible throughout the process.
* Fix issue with trigger wrong type of job
* Fix export failure when folder already exists in repository
When exporting resources to a repository, if a folder already exists,
the Read() method would fail with "path component is empty" error.
This occurred because:
1. Folders are identified by trailing slash (e.g., "Legacy Folder/")
2. The Read() method passes this path directly to GetTreeByPath()
3. GetTreeByPath() splits the path by "/" creating empty components
4. This causes the "path component is empty" error
The fix strips the trailing slash before calling GetTreeByPath() to
avoid empty path components, while still using the trailing slash
convention to identify directories.
The Create() method already handles this correctly by appending
".keep" to directory paths, which is why the first export succeeded
but subsequent exports failed.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* Fix folder tree not updated when folder already exists in repository
When exporting resources and a folder already exists in the repository,
the folder was not being added to the FolderManager's tree. This caused
subsequent dashboard exports to fail with "folder NOT found in tree".
The fix adds the folder to fm.tree even when it already exists in the
repository, ensuring all folders are available for resource lookups.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* Revert "Merge remote-tracking branch 'origin/uncomment-unified-migration-code' into cleanup/deprecate-legacy-storage-migration-in-provisioning"
This reverts commit 6440fae342, reversing
changes made to ec39fb04f2.
* fix: handle empty folder titles in path construction
- Skip folders with empty titles in dirPath to avoid empty path components
- Skip folders with empty paths before checking if they exist in repository
- Fix unit tests to properly check useResourceStats hook calls with type annotations
* Update workspace
* Fix BootstrapStep tests after reverting unified migration merge
Updated test expectations to match the current component behavior where
resource counts are displayed for both instance and folder sync options.
- Changed 'Empty' count expectation from 3 to 4 (2 cards × 2 counts each)
- Changed '7 resources' test to use findAllByText instead of findByText
since the count appears in multiple cards
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* Remove bubbletee deps
* Fix workspace
* provisioning: update error message to reference enableMigration config
Update the error message when provisioning cannot be used due to
incompatible data format to instruct users to enable data migration
for folders and dashboards using the enableMigration configuration
introduced in PR #114857.
Also update the test helper to include EnableMigration: true for both
dashboards and folders to match the new configuration pattern.
* provisioning: add comment explaining Mode5 and EnableMigration requirement
Add a comment in the integration test helper explaining that Provisioning
requires Mode5 (unified storage) and EnableMigration (data migration) as
it expects resources to be fully migrated to unified storage.
* Remove migrate resources checkbox from folder type provisioning wizard
- Remove checkbox UI for migrating existing resources in folder type
- Remove migrateExistingResources from migration logic
- Simplify migration to only use requiresMigration flag
- Remove unused translation keys
- Update i18n strings
* Fix linting
* Remove unnecessary React Fragment wrapper in BootstrapStep
* Address comments
---------
Co-authored-by: Rafael Paulovic <rafael.paulovic@grafana.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
The settings and stats endpoints were returning 403 for users accessing via
ST->MT because the AccessPolicy identity was routed to the access checker,
which doesn't know about these resources.
This fix handles 'settings' and 'stats' resources before the access checker
path, routing them to the role-based authorization that allows:
- settings: Viewer role (read-only, needed by frontend)
- stats: Admin role (can leak information)
* Alerting: Protect sensitive fields of contact points from
unauthorized modification
- Introduce a new permission alert.notifications.receivers.protected:write. The permission is granted to contact point administrators.
- Introduce field Protected to NotifierOption
- Introduce DiffReport for models.Integrations with focus on Settings. The diff report is extended with methods that return all keys that are different between two settings.
- Add new annotation 'grafana.com/access/CanModifyProtected' to Receiver model
- Update receiver service to enforce the permission and return status 403 if unauthorized user modifies protected field
- Update receiver testing API to enforce permission and return status 403 if unauthorized user modifies protected field.
- Update UI to disable protected fields if user cannot modify them
* refactor: delegate authorization to access checker in dualwriter
- Remove role-based authorization checks (editor/admin role checks)
- Delegate all authorization to access checker which checks resource-level permissions
- Update authorizeCreateFolder to use access checker instead of role-based checks
- Add comprehensive authorization tests for viewer, editor, and admin roles
- Tests cover GET, POST, PUT, DELETE operations and folder creation
This change ensures that authorization is consistently handled through
the access checker, which checks resource-level permissions rather than
just organization roles.
* fix: format files_test.go
* fix: check error return value of resp.Body.Close()
* fix: grant permissions to all dashboards for editor role in authorization test
Use SetPermissions with wildcard to grant permissions to Editor user
for all dashboards, not just the initial one. This ensures that dashboards
created during tests (like in DELETE operations) have the necessary
permissions for the editor role.
fix: allow editors to POST jobs in provisioning API
Editors should be able to post jobs in the 'jobs' endpoint for syncing
repositories. This aligns with the requirement that syncing a repository
requires editor privileges.
- Separated 'jobs' subresource authorization from repository/test
- Allow both admins and editors to POST jobs
- Added integration tests to verify permissions
Fixes authorization bug where editors were incorrectly denied access.
* Provisioning: Deprecate single file/folder move and delete on configured branch
Reject individual file and folder move/delete operations on the configured
branch via the single files endpoints (HTTP 405 MethodNotAllowed). Users
must use the bulk operations API (jobs API) instead.
Motivation:
- Reconciliation for these operations is not reliable as it must be
recursive and cannot run synchronously since it could take a long time
- Simplifies authorization logic - fewer operations to secure and validate
- Reduces complexity and surface area for potential bugs
- Bulk operations via jobs API provide better control and observability
Operations on non-configured branches (e.g., creating PRs) continue to work
as before since they don't update the Grafana database.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* fix: remove trailing whitespace in test file
* Fix behaviour to match current behavior
* Revert changes for individual files
---------
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
Previous attempt to land this required this PR and a grafana-enterprise
PR to land at the ~same time.
This PR guards the use of `dsConfigHandlerRequestsDuration` with a nil
check, and doesn't change any existing APIs, so we can land it without
any timing issues with grafana-enterprise.
Once this has landed, we'll make a follow-up PR for grafana-enterprise.
* Fix race condition causing unhealthy repository message to be lost
This commit fixes a race condition in the provisioning repository controller
where the "Repository is unhealthy" message in the sync status could be lost
due to status updates being based on stale repository objects.
## Problem
The issue occurred in the `process` function when:
1. Repository object was fetched from cache with old status
2. `RefreshHealth` immediately patched the health status to "unhealthy"
3. `determineSyncStatusOps` used the stale object to check if unhealthy
message was already set
4. A second patch operation based on stale data would overwrite the
health status update
## Solution
Introduced `RefreshHealthWithPatchOps` method that returns patch operations
instead of immediately applying them. This allows batching all status updates
(health + sync) into a single atomic patch operation, eliminating the race
condition.
## Changes
- Added `HealthCheckerInterface` for better testability
- Added `RefreshHealthWithPatchOps` method to return patch ops without applying
- Updated `process` function to batch health and sync status updates
- Added comprehensive unit tests for the fix
Fixes the issue where unhealthy repositories don't show the "Repository is
unhealthy" message in their sync status.
* Fix staticcheck lint error: remove unnecessary nil check for slice
* `grafana-iam`: Fetch target parent folder
* WIP add different ParentProviders
* Add version
* Move code to a different file
* Instantiate resourceParentProvider
* same import name
* imports
* Add tests
* Remove unecessary test
* forgot wire
* WIP integration tests
* Add test to cover list
* Fix caching problem in integration tests
* comments
* Logger and comments
* Add lazy creation and caching
* Instantiate clients only once
* Rerun wire gen
* feat: legacy ListIterator with batches
* chore: address code review
* chore: remove nil check in nextBatch
* chore: move close before count check
* chore: add err field to batchingIterator for its own errors
* chore: remove unused import
* fix: use dsIndexProvider cache on migrations
* chore: use same comment as before
* feat: org-aware TTL cache for schemaversion migration and warmup for single tenant
* chore: use LRU cache
* chore: change DefaultCacheTTL to 1 minute
* chore: address copilot reviews
* chore: use expirable cache
* chore: remove unused import
Roberto Jimenez Sanchez
Will Browne
Mustafa Sencer Özcan
Ryan McKinley
Yuri Tseretyan
Daniele Stefano Ferru
Andres Martinez GotorandTodd Treece
Matheus Macabu
Costa Alexoglou
Gabriel MABILLE
Gonzalo Trigueros Manzanas
Charandas
beejeebus
Steve SimpsonandCharandas Batra
Misi
Gábor Farkas