Files
Charon/docs/plans/current_spec.md
GitHub Actions 77a020b4db feat: registry-driven DNS provider type discovery
Phase 1 of Custom DNS Provider Plugin Support: the /api/v1/dns-providers/types
endpoint now returns types dynamically from the dnsprovider.Global() registry
instead of a hardcoded list.

Backend handler queries registry for all provider types, metadata, and fields
Response includes is_built_in flag to distinguish plugins from built-ins
Frontend types updated with DNSProviderField interface and new response shape
Fixed flaky WAF exclusion test (isolated file-based SQLite DB)
Updated operator docs for registry-driven discovery and plugin installation
Refs: #461
2026-01-14 18:05:46 +00:00

14 KiB
Raw Blame History

Custom DNS Provider Plugin Support — Remaining Work Plan

Date: 2026-01-14

This document is a phased completion plan for the remaining work on “Custom DNS Provider Plugin Support” on branch feature/beta-release (see PR #461 context in CHANGELOG.md).

Whats Already Implemented (Verified)

Whats Missing (Verified)

  • Types endpoint is not registry-driven yet: GET /api/v1/dns-providers/types is currently hardcoded in backend/internal/api/handlers/dns_provider_handler.go and will not surface:
    • the manual providers field specs
    • any externally loaded plugin types (e.g., PowerDNS)
    • any future custom providers registered in dnsprovider.Global()
  • Plugin signature allowlist is not wired: PluginLoaderService supports an optional SHA-256 allowlist map, but backend/cmd/api/main.go passes nil.
  • Sandboxing limitation is structural: Go plugins run in-process (no OS sandbox). The only practical controls are deny-by-default plugin loading + allowlisting + secure deployment guidance.
  • No first-party webhook/script/rfc2136 provider types exist as built-in dnsprovider.ProviderPlugin implementations (this is optional and should be treated as a separate feature, because external plugins already cover the extensibility goal).

Scope

  • Make DNS provider type discovery and UI configuration registry-driven so built-in + manual + externally loaded plugins show up correctly.
  • Close the key security gap for external plugins by wiring an operator-controlled allowlist for plugin SHA-256 signatures.
  • Keep the scope aligned to repo conventions: no Python, minimal new files, and follow the repository structure rules for any new docs.

Non-Goals

  • No Python scripts or example servers.
  • No unrelated refactors of existing built-in providers.
  • No “script execution provider” inside Charon (in-process shell execution is a separate high-risk feature and is explicitly out of scope here).
  • No broad redesign of certificate issuance beyond whats required for correct provider type discovery and safe plugin loading.

Dependencies

Risks

  • Type discovery mismatch: UI uses /api/v1/dns-providers/types; if backend remains hardcoded, registry/manual/external plugin types wont be configurable.
  • Supply-chain risk (plugins): .so loading is inherently sensitive; SHA-256 allowlist must be operator-controlled and deny-by-default in hardened deployments.
  • No sandbox: Go plugins execute in-process with full memory access. Treat plugins as trusted code; document this clearly and avoid implying sandboxing.
  • SSRF / outbound calls: plugins may implement TestCredentials() with outbound HTTP. Core cannot reliably enforce SSRF policy inside plugin code; mitigate via operational controls (restricted egress, allowlisted outbound via infra) and guidance for plugin authors to reuse Charon URL validators.
  • Patch coverage gate: any production changes must maintain 100% patch coverage for modified lines.

Definition of Done (DoD) Verification Gates (Per Phase)

Repository testing protocol requires Playwright E2E before unit tests.

  • E2E (first): npx playwright test --project=chromium
  • Backend tests: VS Code task shell: Test: Backend with Coverage
  • Frontend tests: VS Code task shell: Test: Frontend with Coverage
  • TypeScript: VS Code task shell: Lint: TypeScript Check
  • Pre-commit: VS Code task shell: Lint: Pre-commit (All Files)
  • Security scans:
    • VS Code tasks shell: Security: CodeQL Go Scan (CI-Aligned) [~60s] and shell: Security: CodeQL JS Scan (CI-Aligned) [~90s]
    • VS Code task shell: Security: Trivy Scan
    • VS Code task shell: Security: Go Vulnerability Check

Patch coverage requirement: 100% for modified lines.


Phase 1 — Registry-Driven Type Discovery (Unblocks UI + plugins)

Deliverables

  • Backend GET /api/v1/dns-providers/types returns registry-driven types, names, fields, and docs URLs.
  • The types list includes: built-in providers, manual, and any external plugins loaded from CHARON_PLUGINS_DIR.
  • Unit tests cover the new type discovery logic with 100% patch coverage on modified lines.

Tasks & Owners

  • Backend_Dev
    • Replace hardcoded type list behavior in backend/internal/api/handlers/dns_provider_handler.go with registry output.
    • Use the service as the abstraction boundary:
      • h.service.GetSupportedProviderTypes() for the type list
      • h.service.GetProviderCredentialFields(type) for field specs
      • dnsprovider.Global().Get(type).Metadata() for display name + docs URL
    • Ensure the handler returns a stable, sorted list for predictable UI rendering.
    • Add/adjust tests for the types endpoint.
  • Frontend_Dev
    • Confirm getDNSProviderTypes() is used as the single source of truth where appropriate.
    • Keep the fallback schemas in frontend/src/data/dnsProviderSchemas.ts as a defensive measure, but prefer server-provided fields.
  • QA_Security
    • Validate that a newly registered provider type becomes visible in the UI without a frontend deploy.
  • Docs_Writer
    • Update operator docs explaining how types are surfaced and how plugins affect the UI.

Acceptance Criteria

  • Creating a manual provider is possible end-to-end using the types endpoint output.
  • /api/v1/dns-providers/types includes manual and any externally loaded provider types (when present).
  • 100% patch coverage for modified lines.

Verification Gates

  • If UI changed: run Playwright E2E first.
  • Run backend + frontend coverage tasks, TypeScript check, pre-commit, and security scans.

Phase 2 — Provider Implementations: rfc2136, webhook, script

This phase is optional and should only proceed if we explicitly want “first-party” provider types inside Charon (instead of shipping these as external .so plugins). External plugins already satisfy the extensibility goal.

Deliverables

  • New provider plugins implemented (as dnsprovider.ProviderPlugin):
    • rfc2136
    • webhook
    • script
  • Each provider defines:
    • Metadata() (name/description/docs)
    • CredentialFields() (field definitions for UI)
    • Validation (required fields, value constraints)
    • BuildCaddyConfig() (or explicit alternate flow) with deterministic JSON output

Tasks & Owners

  • Backend_Dev
    • Add provider plugin files under backend/pkg/dnsprovider/custom (pattern matches manual_provider.go).
    • Define clear field schemas for each type (avoid guessing provider-specific parameters not supported by the underlying runtime; keep minimal + extensible).
    • Implement validation errors that are actionable (which field, whats wrong).
    • Add unit tests for each provider plugin:
      • metadata
      • fields
      • validation
      • config generation
  • Frontend_Dev
    • Ensure provider forms render correctly from server-provided field definitions.
    • Ensure any provider-specific help text uses the docs URL from the server type info.
  • Docs_Writer
    • Add/update docs pages for each provider type describing required fields and operational expectations.

Docker/Caddy Decision Checkpoint (Only if needed)

Before changing Docker/Caddy:

  • Confirm whether the running Caddy build includes the required DNS modules for the new types.
  • If a module is required and not present, update Dockerfile xcaddy build arguments to include it.

Acceptance Criteria

  • rfc2136, webhook, and script show up in /dns-providers/types with complete field definitions.
  • Creating and saving a provider of each type succeeds with validation.
  • 100% patch coverage for modified lines.

Verification Gates

  • If UI changed: run Playwright E2E first.
  • Run backend + frontend coverage tasks, TypeScript check, pre-commit, and security scans.

Phase 3 — Plugin Security Hardening & Operator Controls

Deliverables

  • Documented and configurable plugin loading policy:
    • plugin directory (CHARON_PLUGINS_DIR already used by startup in backend/cmd/api/main.go)
    • optional SHA-256 allowlist support wired end-to-end (from config/env → NewPluginLoaderService(..., allowedSignatures))
  • Minimal operator guidance for secure deployment.

Tasks & Owners

  • Backend_Dev
    • Wire a configuration source for plugin signatures into the PluginLoaderService creation path (currently passed nil in backend/cmd/api/main.go).
    • Prefer a single env var to stay minimal (example format: JSON map of pluginNamesha256:...).
    • Add tests covering:
      • allowlist reject (plugin not in allowlist)
      • signature mismatch
      • insecure directory permissions rejection
  • DevOps
    • Ensure the plugin directory is mounted read-only where feasible.
    • Validate container permissions align with verifyDirectoryPermissions() expectations.
  • QA_Security
    • Threat model review focused on .so loading risks and expected mitigations.
  • Docs_Writer
    • Update plugin operator docs to explain allowlisting, signatures, and safe deployment patterns.

Acceptance Criteria

  • Plugins can be loaded successfully when allowed, and rejected when disallowed.
  • Misconfigured (world-writable) plugin directory is detected and prevents loading.
  • 100% patch coverage for modified lines.

Verification Gates

  • Run backend + frontend coverage tasks, TypeScript check, pre-commit, and security scans.

Phase 4 — E2E Coverage + Regression Safety

Deliverables

  • Playwright coverage for:
    • DNS provider types rendering and required-field validation (including plugin types)
    • Manual DNS challenge flow regression (existing spec: tests/manual-dns-provider.spec.ts)
    • Creating a provider for at least one external plugin type (e.g., powerdns) when a plugin is present
  • Documented smoke test steps for operators.

Tasks & Owners

  • QA_Security
    • Add/extend Playwright specs under tests.
    • Validate keyboard navigation and form errors are accessible (screen reader friendly) where tests touch UI.
  • Frontend_Dev
    • Fix any UI issues uncovered by E2E (focus order, error announcements, labels).
  • Backend_Dev
    • Fix any API contract mismatches discovered by E2E.

Acceptance Criteria

  • E2E passes reliably in Chromium.
  • No regressions to manual challenge flow.

Verification Gates

  • Run Playwright E2E first.
  • Run backend + frontend coverage tasks, TypeScript check, pre-commit, and security scans.

Open Questions (Need Explicit Decisions)

  • For plugin signature allowlisting: what is the desired configuration shape?
    • Option A (minimal): env var JSON map pluginNamesha256:... parsed by backend/cmd/api/main.go
    • Option B (operator-friendly): load from a mounted file path (adds new config surface)
  • For “first-party” providers (webhook, script, rfc2136): are these still required given external plugins already exist?

Notes on Accessibility

UI work in this plan is built with accessibility in mind, but likely still requires manual review and testing (e.g., with Accessibility Insights) as changes land.