Files
nick-doc/01 - Architecture/Security Architecture.md
Siavash Sameni 4cf5c49274 docs(audit): align documentation with post-remediation backend reality
- Update data model enums to match backend models
- Update API reference auth requirements
- Add dispute module references and warning blocks
- Add 2026-05-24 audit remediation callout to Overview
- Generate task breakdowns and audit artifacts
- Add doc alignment report (.taskmaster/reports/)
2026-05-24 11:16:29 +04:00

232 lines
9.9 KiB
Markdown

---
title: Security Architecture
tags: [architecture, security, authentication, rbac]
created: 2026-05-23
---
# Security Architecture
How identity, authorization, transport, and integrity are handled across the platform.
> [!important]
> Read alongside [[Authentication Flow]] (user-facing), [[Passkey (WebAuthn) Flow]], and [[Payment Flow - SHKeeper]] (webhook HMAC).
---
## 1. Threat model — at a glance
| Threat | Mitigation |
|---|---|
| Credential stuffing | bcrypt 12-round hashing + account lockout + rate-limit (when enabled) |
| Session hijacking | Short-lived JWTs (7d), opaque refresh tokens (30d), token rotation |
| CSRF | JWT in `Authorization` header (not cookie), CORS allow-list |
| XSS | Helmet CSP, React auto-escaping, sanitize HTML before storage |
| SQL/NoSQL injection | Mongoose parameterized queries, no `$where` strings, schema validation |
| Webhook spoofing | HMAC SHA-256 over body + secret, constant-time compare |
| File upload abuse | Multer MIME validation, 5 MB cap, non-executable storage, served by Nginx not Node |
| Replay attacks | Per-payment idempotency on `providerPaymentId`, per-request `X-Request-Id` |
| Account takeover | Email verification required, password reset code expiry (1h), passkey support |
| Phishing | Passkey origin binding (`NEXT_PUBLIC_PASSKEY_ORIGIN`), email domain pinning |
| Data leakage | Role-gated endpoints, field-level projection (`select: false` on password), redacted logs |
---
## 2. Authentication layers
### 2.1 Email + password (primary)
```mermaid
sequenceDiagram
actor U as User
participant FE as Frontend
participant BE as Backend
participant DB as MongoDB
U->>FE: enters credentials
FE->>BE: POST /api/auth/login { email, password }
BE->>DB: User.findOne({ email })
DB-->>BE: user doc (incl. hashed password)
BE->>BE: bcrypt.compare(password, hash)
alt invalid
BE->>DB: increment loginAttempts
BE->>DB: lock account at N
BE-->>FE: 401 / 423 locked
else valid
BE->>BE: sign JWT (7d), refresh (30d)
BE->>DB: store refresh-token id
BE->>DB: clear attempts
BE-->>FE: 200 { user, token, refreshToken }
end
```
- Password rules enforced by `authValidation.ts`: ≥8 chars, mixed case + digit recommended (cite the validator for exact rules).
- bcrypt rounds = 12 (`authService.ts`).
- Lockout: after N failed attempts within window, account locked for cooldown — see `authService.ts:113-145`.
- Reset code emailed on `passwordResetCode` request; valid 1h.
### 2.2 Google OAuth 2.0
- Frontend uses `NEXT_PUBLIC_GOOGLE_CLIENT_ID` for the Sign-In with Google button.
- ID token sent to backend → `googleOAuthService.ts` verifies via Google's public keys → either links to existing User by email or creates a new one.
- See [[Google OAuth Flow]].
### 2.3 WebAuthn / Passkey
- Standards-based passwordless.
- Backend: `passkeyService.ts` orchestrates registration and assertion challenges.
- Frontend env: `NEXT_PUBLIC_PASSKEY_RP_NAME=Amn`, `NEXT_PUBLIC_PASSKEY_RP_ID=<domain>`, `NEXT_PUBLIC_PASSKEY_ORIGIN=<origin>`.
- See [[Passkey (WebAuthn) Flow]].
> [!warning]
> Dev env files ship `NEXT_PUBLIC_PASSKEY_RP_ID=localhost`. In production this MUST be the actual eTLD+1 domain (e.g., `amn.gg`) — passkeys are scoped to the RP ID and can't be transferred.
### 2.4 Refresh-token rotation
- On `POST /api/auth/refresh`, the backend:
- Verifies the supplied refresh token.
- Issues a NEW access token + a NEW refresh token.
- Invalidates the old refresh token id in MongoDB.
- If the same refresh token is presented twice → all sessions for that user are invalidated (token reuse detection).
---
## 3. Authorization (RBAC)
### 3.1 Roles
| Role | Source | Capabilities |
|---|---|---|
| `buyer` (user) | default on signup | Create requests, pay, chat, dispute, rate |
| `seller` (owner) | chosen at signup OR upgraded | Make offers, build templates, run a shop, withdraw |
| `admin` | seed / manual | Moderate, mediate disputes, manage users/blogs/levels |
| `support` | seed / manual | Read-only on most data, can reset passwords, escalate |
A single User may be `buyer` and `seller` simultaneously (combined role).
### 3.2 Enforcement points
- **Middleware** — `authMiddleware` (verifies JWT) followed by `roleGuard(role)` on every route that requires elevation.
- **Service layer** — defensive `assertRole(ctx, 'admin')` calls inside critical service methods so even mis-mounted routes can't bypass.
- **UI** — `AuthGuard` + `EmailVerificationGuard` + role-aware nav (`components/nav-section`) hide admin/seller menus for users without permission. This is convenience only — never the security boundary.
---
## 4. Transport security
- **HTTPS** terminated upstream (CloudFlare / external Nginx). Internal cluster is HTTP.
- **HSTS** header set by upstream proxy (recommended `max-age=31536000; includeSubDomains; preload`).
- **CORS** — exactly one origin allowed: `config.frontendUrl`. `credentials: true`.
- **CSP** — Helmet default, currently permissive for Web3 popup compatibility (see `frontend/next.config.ts` setting COOP=`same-origin-allow-popups`, COEP=`unsafe-none`).
---
## 5. Webhook integrity (SHKeeper)
```mermaid
sequenceDiagram
participant SHK
participant BE
SHK->>BE: POST /api/payment/shkeeper/webhook<br/>X-Signature: sha256=<hmac>
BE->>BE: hmac = HMAC_SHA256(SHKEEPER_WEBHOOK_SECRET, body)
BE->>BE: crypto.timingSafeEqual(hmac, providedSig)
alt mismatch
BE-->>SHK: 401 Unauthorized
else match
BE->>BE: process payment update
BE-->>SHK: 200 OK
end
```
- Body must be the raw bytes used for HMAC — apply `express.raw({ type: 'application/json' })` on the webhook route ONLY (the rest of the app uses parsed JSON).
- In dev (`NODE_ENV === 'development'`) signature verification can be bypassed for local testing — confirm this is gated and never reachable in prod.
- Idempotency: identical webhook delivered twice should be no-op. Check by `(providerPaymentId, status)` tuple before mutating.
See [[Payment Flow - SHKeeper]] for the full flow.
---
## 6. Input validation
- **Backend** — `express-validator` per route (e.g., `authValidation.ts`), centralised `validate` middleware that 422s on failure with `{ details: [...] }`.
- **Frontend** — `zod` schemas via `@hookform/resolvers/zod`. Same schema can be re-exported to a `shared/` package for true single-source-of-truth (not yet wired).
- **Mongoose** — schema-level `type`, `required`, `enum`, `min`/`max`, custom `validate` functions as a last line of defence.
---
## 7. File upload safety
- Stored under `uploads/{avatars|documents|products|temp}/` — non-executable, served by Nginx (no Node interpretation).
- MIME allow-list in `fileService.ts`: images for avatars/products, PDFs/docs for evidence.
- 5 MB hard cap (`MAX_FILE_SIZE=5242880`).
- Original filenames hashed → no path traversal, no clobber.
- Recommended: virus scan via ClamAV before exposing to other users (dispute evidence, chat attachments).
---
## 8. Secrets management
- Production secrets injected via host `.env`, mounted into compose `env_file`.
- Never log secrets — logger redaction recommended (winston/pino formatter).
- `.env*` files in `.gitignore`. Repo includes only `.env.development` / `.env.production` templates with **public** values (NEXT_PUBLIC_*).
- Rotate `JWT_SECRET` invalidates all existing JWTs — schedule a maintenance window.
- Rotate `SHKEEPER_WEBHOOK_SECRET` coordinated with SHKeeper dashboard (set new → verify → remove old).
See [[Environment Variables]] for the catalog.
---
## 9. Rate limiting & abuse
- Rate limiting is **enabled** as of 2026-05-24 (`app.ts`).
- Active tiers:
- `/api/auth/*` — 10 req / 15 min / IP
- `/api/payment/*` — 30 req / 15 min / IP
- `/api/ai/*` — 20 req / 15 min / IP
- global API — 100 req / 15 min / IP (skips `/health` and Request-Network webhooks)
- Counters are in-memory (Redis adapter planned for distributed deploys).
- For chat and notifications, debounce at the client to avoid spamming legitimate emits.
---
## 10. Audit logging
The codebase currently uses `morgan` (HTTP access logs) and ad-hoc `logger.info/warn/error`. For PCI-adjacent operations (payments) consider:
- Append-only audit log of every payment / payout / refund / role change.
- Include actor (userId), target, action, before/after diff, request id.
- Persist in a separate Mongo collection or external log sink with retention ≥1y.
---
## 11. Frontend session storage
- JWT and refresh token stored in `localStorage` (per current implementation — cite to verify in `frontend/src/lib/`).
- Risk: XSS = total takeover. Mitigations: strict CSP, no `dangerouslySetInnerHTML` on untrusted content, audit dependencies (`yarn audit`).
- Alternative: store refresh token in `httpOnly` cookie and keep only short-lived access token in memory — recommended for production hardening.
---
## 12. Hardening checklist (pre-launch)
- [x] Enable rate-limit middleware (done 2026-05-24)
- [x] Enforce Socket.IO JWT authentication (done 2026-05-24)
- [ ] Promote refresh tokens to `httpOnly` cookies
- [ ] Replace `localhost` passkey RP ID with production domain
- [ ] Disable `NEXT_PUBLIC_IS_DEVELOPMENT=true` and `ENABLE_DEBUG=true` in prod build
- [ ] Verify `NODE_ENV=production` in backend prod env
- [ ] Pin production Watchtower to versioned tag (not `latest`)
- [ ] Add backend Sentry SDK + source maps
- [ ] Rotate all dev-seeded credentials before public launch
- [ ] Run `yarn audit` / `npm audit` and triage CVEs
- [ ] Pentest the payment + dispute flows specifically
- [ ] Review every `> [!warning]` callout in this vault
---
## Related
- [[Authentication Flow]] · [[Google OAuth Flow]] · [[Passkey (WebAuthn) Flow]] · [[Password Reset Flow]]
- [[Backend Architecture]] · [[Frontend Architecture]] · [[Real-time Layer]]
- [[Payment Flow - SHKeeper]] — webhook HMAC details
- [[Environment Variables]] — secret catalog
- [[Incident Response]] — what to do when something goes wrong