From 8eaaa01186ba478d160f18ae5fa34d70705872cf Mon Sep 17 00:00:00 2001 From: Beda Schmid Date: Sun, 1 Mar 2026 21:25:37 -0300 Subject: [PATCH] update report --- REPORT.md | 167 ------------------------------------------------------ 1 file changed, 167 deletions(-) diff --git a/REPORT.md b/REPORT.md index 5debf01..6f06e7b 100644 --- a/REPORT.md +++ b/REPORT.md @@ -22,49 +22,6 @@ Performed a read-only static review of: ## Blocking Security Issues (Code-Level) -### 1) High - No abuse controls on expensive authenticated endpoints - -Impact: - -- Any authenticated user can repeatedly trigger high-cost operations (upload processing, OCR, summarization, routing, indexing), causing queue saturation, infrastructure exhaustion, and external provider cost abuse. - -Exploit path: - -- Repeated `POST /api/v1/documents/upload` -- Repeated `POST /api/v1/documents/{document_id}/reprocess` - -Evidence: - -- Upload endpoint has size limits but no rate/volume quota checks: `backend/app/api/routes_documents.py:665-985`. -- Reprocess endpoint has no rate/cooldown checks: `backend/app/api/routes_documents.py:958-985`. -- A Redis rate limiter exists but is currently used for markdown export only: `backend/app/services/rate_limiter.py:16-42` and `backend/app/api/routes_documents.py:100-120`. - -Remediation: - -- Add per-user and per-IP rate limiting to upload and reprocess endpoints. -- Add per-user daily/rolling quotas (documents, bytes, reprocess calls). -- Add queue depth backpressure and reject or defer requests when saturated. -- Add alerting for anomalous request and job-enqueue rates. - -### 2) Medium - API docs and schema are exposed by default - -Impact: - -- Unauthenticated endpoint discovery and contract reconnaissance are easier (`/docs`, `/redoc`, `/openapi.json`). - -Exploit path: - -- Remote probing of public API metadata when service is internet-reachable. - -Evidence: - -- FastAPI app is created with default docs behavior and no production gating: `backend/app/main.py:37`. - -Remediation: - -- Disable docs in production (`docs_url=None`, `redoc_url=None`, `openapi_url=None`) or -- Restrict these routes at reverse proxy / edge to trusted admin networks. - ### 3) Medium - Bearer token stored in browser sessionStorage Impact: @@ -83,127 +40,3 @@ Remediation: - Prefer HttpOnly Secure SameSite cookies for session auth, plus CSRF protection. - If bearer-in-JS remains, enforce strict CSP, remove inline script execution, and add strong dependency hygiene. - -## MUST KNOW (User-Dependent, Non-Blocking Per Request) - -### A) Current `.env` is development-oriented and exposed beyond localhost - -Why this matters: - -- Service currently binds to all interfaces and uses development settings, which is unsafe for direct internet exposure. - -Evidence: - -- `APP_ENV=development`: `.env:1` -- `HOST_BIND_IP=0.0.0.0`: `.env:2` -- `PUBLIC_BASE_URL=http://...`: `.env:33` -- Broad CORS for localhost and LAN host: `.env:34` - -Action: - -- Set production values (`APP_ENV=production`, HTTPS base URL, strict CORS, host binding behind hardened reverse proxy). - -### B) Provider SSRF protections are disabled in active env - -Why this matters: - -- Allowing HTTP, private network targets, and empty allowlist can permit unsafe outbound model-provider endpoints. - -Evidence: - -- Active `.env`: `PROVIDER_BASE_URL_ALLOW_HTTP=true`, `PROVIDER_BASE_URL_ALLOW_PRIVATE_NETWORK=true`, `PROVIDER_BASE_URL_ALLOWLIST=[]` at `.env:29-31`. -- Compose defaults are permissive if env is not hardened: `docker-compose.yml:55-57`, `docker-compose.yml:100-102`. - -Action: - -- For production set: `ALLOW_HTTP=false`, `ALLOW_PRIVATE_NETWORK=false`, explicit host allowlist. - -### C) Sensitive model and payload text logging is enabled in active env - -Why this matters: - -- Prompt/response and payload text may include confidential document content and credentials, increasing breach impact. - -Evidence: - -- Active `.env` enables both flags: `.env:22-23`. -- Code defaults are safer (`false`): `backend/app/core/config.py:60-61`. - -Action: - -- Disable in production unless explicitly required for short-term diagnostics. -- Apply strict retention and access controls if temporarily enabled. - -### D) Redis transport/auth hardening depends on env mode and URL - -Why this matters: - -- Current env uses `redis://` with auto security/tls mode; this is not suitable for untrusted network paths. - -Evidence: - -- Active `.env`: `REDIS_URL=redis://...`, `REDIS_SECURITY_MODE=auto`, `REDIS_TLS_MODE=auto` at `.env:10-12`. -- Strict checks are triggered by production mode when auto is used: `backend/app/core/config.py:157-171`. - -Action: - -- In production use `rediss://`, `REDIS_SECURITY_MODE=strict`, `REDIS_TLS_MODE=required`. - -### E) Frontend container runs a development server - -Why this matters: - -- Vite dev server is not intended as a hardened production serving layer. - -Evidence: - -- Frontend container command runs `npm run dev`: `frontend/Dockerfile:20`. -- `dev` script maps to Vite dev mode: `frontend/package.json:7`. - -Action: - -- Build static assets and serve behind a production-grade web server/reverse proxy. - -### F) Login throttle IP identity depends on proxy topology - -Why this matters: - -- Throttle identity uses `request.client.host`; if a proxy masks client IPs, lockout behavior may be inaccurate. - -Evidence: - -- IP extraction uses transport client host directly: `backend/app/api/routes_auth.py:32-35`. - -Action: - -- Ensure trusted proxy configuration preserves real client IP semantics before internet deployment. - -## Validation Commands and Outcomes - -Preflight: - -- `command -v git` -> passed (`/usr/bin/git`) -- `git rev-parse --is-inside-work-tree` -> passed (`true`) -- `git status --short` -> clean before work - -Security-related backend tests: - -- `PYTHONDONTWRITEBYTECODE=1 /Users/bedas/Developer/Python/global_venv/bin/python backend/tests/test_security_controls.py` -> passed (34 tests) -- `PYTHONDONTWRITEBYTECODE=1 /Users/bedas/Developer/Python/global_venv/bin/python backend/tests/test_upload_request_size_middleware.py` -> passed (3 tests) -- `PYTHONDONTWRITEBYTECODE=1 /Users/bedas/Developer/Python/global_venv/bin/python backend/tests/test_app_settings_provider_resilience.py` -> passed (6 tests) -- `PYTHONDONTWRITEBYTECODE=1 /Users/bedas/Developer/Python/global_venv/bin/python backend/tests/test_processing_log_retention_settings.py` -> passed (5 tests) - -Frontend auth-client test: - -- `npm run test` (in `frontend/`) -> passed - -Note: - -- `pytest` is not installed in `/Users/bedas/Developer/Python/global_venv/bin/python`, so direct module execution via `unittest` entrypoints was used for backend test files. - -## Residual Risk and Coverage Limits - -- No dynamic penetration test was run against a live deployed stack. -- No dependency CVE audit (`pip-audit`, `npm audit`) was run in this turn. -- Reverse proxy, firewall, TLS termination, and cloud/network policy were not reviewed. -