Files
reverse-proxy/docs/architecture/overview.md
glm-5.1 80d1fd0fb3 Update architecture docs to address security review #003 findings
Add three ADRs (025-027) and update five spec documents to close gaps
identified in the security and bug review:

- ADR-025: Rate limiter IP source must be ConnectInfo only (C1 fix)
- ADR-026: Connector timeout ceiling of 30s for per-site timeouts (C3 fix)
- ADR-027: Admin socket resource limits — 5s timeout, 4096 byte line limit (W4 fix)

Spec changes:
- proxy.md: add rate limiter IP source section, URI error handling
  constraint, connector ceiling description, renumber sections
- operations.md: add ConnectInfo-only IP source, in-flight counter
  architectural requirement (C2), JSON format guarantee (C4), admin
  socket resource limits, 100ms drain polling interval
- config.md: fix http_port type u32→u16 (W12), tighten upstream host
  validation (W1), tighten ACME contact validation (W2), add
  X-Forwarded-Proto cross-reference, clarify alknet ADR-030 reference
- overview.md: fix ambiguous C1 reference, add ADR/OQ cross-references
- open-questions.md: update OQ-09 resolution, add OQ-13 (acme_contact
  Vec) and OQ-14 (eviction configurability)
- README.md: add ADR-025/026/027 and OQ-13/14, update doc statuses to draft

Also fix reviewer findings: alknet ADR-030 scope clarification, RFC 2616
reference updated to RFC 7230.
2026-06-12 13:17:39 +00:00

15 KiB

status, last_updated
status last_updated
draft 2026-06-12

Overview

Vision

A memory-safe, minimal reverse proxy that replaces our vulnerable nginx instance for forwarding requests to backend services. The proxy terminates TLS, injects standard proxy headers, enforces rate limits, and forwards requests to upstream services — supporting multiple domains from initial release.

This project is open source under dual licensing: MIT OR Apache-2.0, consistent with standard Rust project licensing.

Why This Exists

Our nginx 1.24.0 installation is vulnerable to multiple actively-exploited CVEs, including CVE-2026-42945 (unauthenticated RCE via rewrite/set directives). The broader threat landscape is worsening: LLM-assisted fuzzing is accelerating bug discovery in nginx's C codebase, and security researchers report additional undisclosed vulnerabilities. Upgrading nginx patches known CVEs but does not address the structural problem — memory corruption bugs are endemic to C, and the discovery rate is accelerating.

Rust's memory safety eliminates the entire class of buffer overflow, use-after-free, and double-free bugs that constitute 6 of 7 recent nginx CVEs. Combined with rustls (pure Rust TLS, no OpenSSL dependency), this provides a fundamentally safer baseline.

See threat-landscape.md for full vulnerability details.

Scope

In Scope

  • Phase 1: Multi-site reverse proxy with TLS termination

    • Multiple independent TLS listeners via [[listeners]] configuration
    • Each listener has its own bind address, TLS config, and site routing
    • Supports both dedicated-IP (1 IP = 1 cert = 1 domain) and shared-IP (SAN certificate) deployment models (ADR-019)
    • TLS termination with ACME (Let's Encrypt) and manual certificate management
    • Cipher suite restriction matching nginx scope (ECDHE-AES-GCM + TLS 1.3)
    • HTTP/2 support on the client-facing side (between client and proxy), with ALPN-based protocol detection (ADR-023). Upstream connections remain HTTP/1.1.
    • HTTP → HTTPS redirect
    • Host-based routing to multiple upstream services
    • Reverse proxy to Gitea at 127.0.0.1:3000 (git.alk.dev)
    • Reverse proxy to Deno/Fresh container for alk.dev (simple pass-through)
    • Proxy header injection (Host, X-Real-IP, X-Forwarded-For, X-Forwarded-Proto)
    • Per-site upstream timeouts with sensible defaults (5s connect, 60s request)
    • Request rate limiting with fail2ban-compatible logging (global per-IP)
    • 100 MB body size limit (global)
    • Configurable bind addresses (must be explicit, no 0.0.0.0)
    • Local health check endpoint on separate port (default: 9900, localhost only)
    • Unix domain socket admin API for config reload with feedback
    • Graceful shutdown (SIGTERM handling with in-flight request drain)
    • Systemd unit file
    • Dual licensing: MIT OR Apache-2.0
  • Phase 2: Operational hardening

    • Per-site rate limits and body limits
    • Metrics endpoint (Prometheus-compatible)
    • Connection limits and timeouts
    • Log rotation
  • Phase 3: Future enhancements

    • Wildcard subdomain support

Out of Scope

  • HTTP/2 or HTTP/3 proxying to upstreams — the proxy communicates with upstreams over HTTP/1.1 (or HTTPS/1.1). HTTP/2 from clients is supported (see ADR-023). Services that need HTTP/2+ to their backends can handle termination themselves.
  • Load balancing or round-robin upstream selection
  • WebSocket proxying (can be added later if needed)
  • Static file serving
  • Access control beyond rate limiting (no auth, no IP allowlists in Phase 1)
  • CGI, SCGI, uWSGI, FastCGI

Architecture

                      ┌────────────────────────────────────┐
                      │  reverse-proxy container (Rust/axum)│
 config.toml ───────► │  StaticConfig + DynamicConfig       │
 (volume mount)       │  (ArcSwap for hot-reload)            │
                      │                                      │
                        │  ┌─ Listener 1 ─────────────────┐   │
   bind_addr:80  ────►  │  │  HTTP → 301 redirect           │   │
   (published)          │  └────────────────────────────────┘   │
                        │                                      │
   bind_addr:443 ────►  │  │  TLS listener (tokio-rustls)    │   │
   (published)          │  │  ├─ ACME or Manual TLS config    │   │
                        │  │  └─ axum router (per-listener)   │   │
                        │  │     ├─ Host → global site lookup  │   │
                        │  │     ├─ git.alk.dev → gitea:3000  │   │
                        │  │     └─ Rate limiting, headers     │   │
                        │  └────────────────────────────────┘   │
                        │                                      │
                        │  ┌─ Listener N ─────────────────┐   │
   bind_addr_N:80  ───► │  │  HTTP → 301 redirect           │   │
                        │  └────────────────────────────────┘   │
                        │                                      │
   bind_addr_N:443 ───► │  │  TLS listener (tokio-rustls)    │   │
                        │  │  ├─ Manual TLS cert             │   │
                        │  │  └─ axum router (per-listener)   │   │
                        │  │     ├─ Host → global site lookup  │   │
                        │  │     ├─ alk.dev → app:8080       │   │
                        │  │     └─ Rate limiting, headers     │   │
                        │  └────────────────────────────────┘   │
                        │                                      │
                        │  /health → 200 OK (port 9900)        │
                        │  Admin socket (Unix domain)           │
                       └────────────────────────────────────┘
                            │              │
                     ┌──────┘              └──────┐
                     │                             │
              Docker network              Volume mounts:
              (upstream DNS)            ├─ config (ro)
              ├─ gitea:3000              ├─ ACME cache (rw)
              ├─ app:8080                ├─ log dir (rw, fail2ban)
                                         └─ admin socket (rw)

Each listener has its own axum::Router instance with its own middleware stack, but all routers share Arc<ArcSwap<DynamicConfig>> and Arc<Mutex<HashMap<IpAddr, TokenBucket>>> via axum State. Site routing is global: the Host header is matched against a single routing table collected from all listeners' site definitions. Hostnames must be unique across all listeners. Hostnames must be unique across all listeners — see Security & Bug Review #003, finding C1, resolved by ADR-025.

In container deployments (ADR-020), the proxy runs in a minimal container with 0.0.0.0 bind address and Docker port publishing. Upstream addresses use Docker DNS names for same-host containers (e.g., gitea:3000) but also support loopback, LAN, and tunnel endpoints for multi-host deployments.

Crate Dependencies

Core

Crate Version Purpose Notes
axum 0.8 HTTP framework Routing, middleware, extractors
tokio 1 (full) Async runtime Multi-threaded runtime
hyper 1 HTTP protocol Used via axum, and directly for HTTP/2 server builder
hyper-util 0.1 Hyper utilities Client builder, TokioExecutor, auto::Builder
tower 0.5 Middleware ecosystem Service trait, layers
rustls 0.23 TLS implementation aws_lc_rs crypto provider
tokio-rustls 0.26 Async TLS I/O Wraps TCP with TLS
rustls-acme 0.12 ACME client Let's Encrypt auto-provisioning and renewal
hyper-rustls 0.27 HTTPS client Upstream HTTPS connections with rustls TLS
rustls-native-certs 0.8 Native cert loading System root certificates for upstream HTTPS validation

Supporting

Crate Version Purpose Notes
serde 1 Serialization TOML config deserialization
toml 0.8 Config format Declarative site definitions
arc-swap 1 Atomic config swap Lock-free DynamicConfig reload
tracing 0.1 Structured logging fail2ban-compatible output
tracing-subscriber 0.3 Log output File + journald support
rustls-pemfile 2 PEM parsing Manual cert loading
rustls-pki-types 1 TLS types CertificateDer, PrivateKeyDer
clap 4 CLI arguments Server startup options
signal-hook 0.3 Signal handling SIGTERM/SIGINT for shutdown, SIGHUP for config reload

Versions listed are minimum major versions. Implementation should pin exact versions in Cargo.toml per standard Rust practice.

Exports

This is a single-binary deployment. There are no library exports. The product is the reverse-proxy binary plus a systemd unit file and a config file.

Dependencies on Other Projects

  • alknet: The ArcSwap<DynamicConfig> pattern, tokio-rustls TLS acceptor construction, rustls-acme integration, and ServerConfig builder patterns are adapted from alknet's transport and config layers. These patterns are referenced as validation that the approaches work in production; all code in this project is written from scratch.

Design Decisions

All design decisions are documented as ADRs in decisions/.

ADR Decision Summary
001 Rust with axum Memory safety eliminates the bug class causing nginx CVEs; axum provides ergonomic tower integration
002 Custom proxy handler Single upstream per domain — simpler than a general proxy library
003 TOML configuration format Rust-native, unambiguous, excellent serde support
004 ACME-primary certificate management Eliminates certbot dependency; automatic provisioning and renewal
005 tokio-rustls directly, not axum-server Full control over TLS config, ACME resolver integration, cipher suite configuration
006 Token bucket rate limiting In-memory per-IP token bucket matching nginx burst semantics
007 Custom structured log format key=value pairs with RATE_LIMIT prefix for fail2ban
008 Static/dynamic config with ArcSwap Immutable StaticConfig, hot-reloadable DynamicConfig via ArcSwap
009 Signal handling strategy signal-hook for SIGTERM/SIGINT/SIGHUP
010 Multi-site in Phase 1 Multiple domains from initial release; avoids config migration later
011 Multi-domain TLS config Single SAN certificate covering all domains via rustls-acme
012 Restrict cipher suites Match nginx scope: ECDHE-AES-GCM for TLS 1.2, all TLS 1.3
013 Health check on separate local port Localhost-only HTTP health check, configurable port
014 Unix domain socket config reload API Programmatic reload with success/failure feedback
015 Per-site upstream timeouts with defaults 5s connect / 60s request defaults, per-site overrides
016 Explicit bind address required Rejects 0.0.0.0 to prevent accidental exposure
017 Upstream connection defaults HTTP/1.1, no redirects, connection pooling
018 Request body size limit 100 MB default matching nginx, Gitea push compatibility
019 Multi-config listeners [[listeners]] supporting both dedicated-IP and shared-IP deployment models
020 Container deployment model Defense-in-depth via container isolation; file-primary logging; flexible upstream addressing
021 X-Forwarded-For edge proxy model Replace, don't append — proxy is the edge, no trusted upstream proxies
022 Health check scope — local port and admin socket only No /health route on main listener; health check is port 9900/admin socket only
023 HTTP/2 client-facing support ALPN-based protocol detection; HTTP/2 to clients, HTTP/1.1 to upstreams
024 ANSI-disabled logging All log output uses with_ansi(false) for fail2ban and Docker compatibility
025 Rate limiter IP source ConnectInfo only, never client-supplied X-Forwarded-For
026 Connector timeout ceiling 30s ceiling on connector, per-site timeout via tokio::time::timeout
027 Admin socket resource limits 5s read timeout, 4096 byte line length limit

Open Questions

Open questions are tracked in open-questions.md. All questions affecting this document have been resolved:

  • OQ-01: Should cipher suites be restricted beyond rustls defaults? (resolved — ADR-012)
  • OQ-03: Should the health check endpoint be on a separate port? (resolved — ADR-013)
  • OQ-05: Should the proxy bind to multiple addresses? (resolved — single bind_addr per listener)
  • OQ-07: Should per-site TLS overrides be supported for mixed ACME/manual domains? (resolved — ADR-019: [[listeners]] with per-listener TLS config)
  • OQ-08: Should /health use a less common path? (resolved — ADR-022: no /health route on main listener; health check is port 9900/admin socket only)
  • OQ-13: Should acme_contact support multiple email addresses? (see open-questions.md)
  • OQ-14: Should rate limiter eviction interval and max age be configurable? (see open-questions.md)