Files
parking_solution/wiki/entities/local-jwt-auth.md
T
julian 66c1291578 docs(deploy): COOKIE_SECURE=0 runbook for the plain-HTTP appliance
Documents the deploy-time requirement that the cookie fail-safe fix (7629d5d)
introduced: the LAN appliance serves the SPA same-origin over plain http, where a
Secure cookie is never sent — so it MUST set COOKIE_SECURE=0 or operators can't log
in. A TLS deploy leaves it unset.

- wiki/concepts/disk-os-hardening.md: new "Deploy-time server configuration (runbook)"
  section listing the security-load-bearing env (JWT_SECRET, EVENT_SIGNING_KEY,
  COOKIE_SECURE=0) with the why + the network-scoped justification.
- wiki/entities/local-jwt-auth.md: corrected the stale "Secure when NODE_ENV=production"
  cookie line to the Secure-by-default / opt-out model.
- wiki/log.md: entry.

Claude-Session: https://claude.ai/code/session_01Xcm6ikLgGoCxxHrxtjkk5V
2026-06-21 23:52:21 +02:00

5.1 KiB

type, tags, sources, updated
type tags sources updated
entity
parking
stack
auth
offline-first
parking-system-architecture
2026-06-15

Local JWT Auth

Authentication and authorization, kept fully local — a direct consequence of offline-first (an air-gapped park cannot reach an external identity provider; see logto-zitadel-oidc for the rejected alternative). (See parking-system-architecture §2.)

  • @fastify/jwt signs tokens with a local secret (symmetric HMAC). The server refuses to start without a strong JWT_SECRET (≥32 chars, no placeholder) — there is deliberately no insecure default.
  • Session lifetime: valid until explicit logout — no time expiry (decision 2026-06-15, built). Booth reality breaks any fixed clock: relief arrives late, fails to show, or one operator is forced to work two shifts in a row — a token that expired mid-duty would strand an active operator. So the login persists until logout; a shift is a separate, explicit boundary, not tied to token lifetime. (Superseded the earlier "8h expiry, bound to a shift" assumption.) The JWT carries no exp; the cookie has a long fixed maxAge (30 days) so a browser restart doesn't log out an active operator, and logout clears it.
  • A users table in sqlite holds bcrypt password hashes plus a role_id FK. The first admin is seeded via pnpm --filter @parking/server seed-admin (no bootstrap endpoint); every other user is created in-app (admin → Users screen).
  • Authorization = dynamic RBAC (built 2026-06-18, replacing the old hardcoded admin/operator/cashier/readonly enum — those are now ordinary seed roles). Roles are data: roles + role_permissions tables, composed by an admin from a code-defined permission grid (@parking/shared PERMISSIONS = resource:action, e.g. tariff:update, payment:create, event:void). A preHandler requirePermission(...) per route checks a PERMISSION, not a role name. The JWT carries roleId (not the permission list); the guard resolves the role's permission set per-request from an in-memory cache (bumpPermsCache() on any role write), so editing a role applies immediately — no re-login, no token bloat. No Casbin/engine needed at this scale.
  • Protected built-in admin role (id='admin', builtin=1): non-editable, non-deletable, and always resolves to the FULL permission set in code. The app refuses to delete or downgrade the last user holding admin — administration can never be locked out of the appliance.
  • event:void is a permission, NOT a ledger delete: the append-only signed chain is untouched; the permission only gates who may APPEND a void event (there is no void API route yet — forward seam).
  • The grid is extensible — adding a feature adds its resource:action rows. Latest: log:read (a new log resource) gates the diagnostic-log viewer (GET /api/logs); admin holds it, and it's grantable to a diagnostic role. See app-logs.
  • No privilege escalation through the RBAC system itself. role:create/role:update and user:create/user:update are themselves grantable, so a non-admin could otherwise self-escalate. Guards (routes/roles.ts, routes/users.ts): a caller may only put permissions on a role that they already hold, and may only assign/modify users whose role is a SUBSET of the caller's own (so no minting a privileged role, handing out the admin role, or resetting/deleting a more- privileged account). An admin holds the full set, so it is unrestricted — the intended behaviour.

The SPA never sees the JWT. Login (POST /api/auth/login) verifies bcrypt and sets two cookies:

  • parking_token — the JWT, HttpOnly + SameSite=Strict, and Secure by default (fail-safe — a forgotten env can only make cookies more restrictive, never drop the flag). Secure is dropped ONLY for a deliberate opt-out: COOKIE_SECURE=0 (the plain-HTTP LAN appliance — see disk-os-hardening deploy checklist) or NODE_ENV=development. (Was keyed off NODE_ENV=production, which silently leaked cookies on an appliance that forgot to set it — corrected 2026-06-21.) JS can't read it; @fastify/jwt reads it from the cookie, not the Authorization header.
  • parking_csrf — a random token, readable by JS. The JWT also carries a matching csrf claim. On every mutation the SPA echoes the cookie in the X-CSRF-Token header; the guard requires header == cookie == the signed claim (double-submit CSRF). Safe reads are exempt.

Routes: login, logout (clears cookies), me (bootstraps SPA session on load). The dev react-vite-spa proxy and the prod nginx reverse proxy keep the SPA and API same-origin, so the cookies work without CORS. (This replaced an earlier dev-only SETUP_AUTH_BYPASS shim, now removed.)

Open decision: moving from the symmetric secret to an asymmetric key (RS256/EdDSA) so verifying hosts hold only a public key — open-questions #7. Relevant before any multi-host/multi-lane deployment.

Part of the technology-stack. License: MIT.