@webdecoy/fcaptcha 1.15.0 → 1.16.0

package/clientip.js

@@ -0,0 +1,229 @@
+/**
+ * Client IP resolution.
+ *
+ * Every IP-derived signal in FCaptcha — datacenter ranges, Tor/VPN, rate
+ * limiting, token IP binding, PoW difficulty — is only as trustworthy as the
+ * address it is handed. X-Forwarded-For and X-Real-IP are set by whoever is on
+ * the other end of the socket, so honouring them unconditionally lets any
+ * caller claim a clean residential IP and walk past all of it. They are
+ * credible only when the peer that sent them is a proxy we put there.
+ *
+ * TRUSTED_PROXIES holds the peers allowed to speak for someone else, as a
+ * comma-separated list of CIDRs and bare IPs. Unset means the private and
+ * loopback defaults below. '*' trusts every peer (only correct when an edge you
+ * control always overwrites the headers). 'none' trusts nobody, so the socket
+ * address is always used.
+ */
+
+const ipaddr = require('ipaddr.js');
+
+/**
+ * Fallback trust set: loopback plus the private and link-local ranges. A server
+ * exposed directly to the internet sees public peer addresses, which are not in
+ * this set, so forged headers are ignored. A server behind a sidecar or
+ * in-cluster reverse proxy (nginx, Railway, Fly, Kubernetes ingress) sees a
+ * loopback or private peer and behaves as before.
+ */
+const DEFAULT_TRUSTED_PROXY_CIDRS = [
+  '127.0.0.0/8',
+  '::1/128',
+  '10.0.0.0/8',
+  '172.16.0.0/12',
+  '192.168.0.0/16',
+  '169.254.0.0/16',
+  'fe80::/10',
+  'fc00::/7'
+];
+
+// hostOnly strips a trailing :port or wrapping brackets. Node's
+// socket.remoteAddress carries no port, but X-Forwarded-For hops sometimes do.
+function hostOnly(value) {
+  const s = String(value).trim();
+  if (s.startsWith('[')) {
+    const end = s.indexOf(']');
+    return end === -1 ? s : s.slice(1, end);
+  }
+  // A bare IPv6 address has several colons; only strip a port from host:port.
+  const first = s.indexOf(':');
+  if (first !== -1 && s.indexOf(':', first + 1) === -1) return s.slice(0, first);
+  return s;
+}
+
+// normalizeIP validates an address and renders IPv4-mapped IPv6
+// (::ffff:1.2.3.4) in dotted form, so the CIDR checks in detection.js match
+// either spelling. Returns null when the value is not an IP.
+function normalizeIP(value) {
+  if (!value) return null;
+  try {
+    return ipaddr.process(hostOnly(value)).toString();
+  } catch {
+    return null;
+  }
+}
+
+// Node merges repeated headers into one comma-joined string for everything
+// except set-cookie, but accept an array defensively.
+function headerValue(req, name) {
+  const raw = req.headers?.[name];
+  if (Array.isArray(raw)) return raw.join(',');
+  return raw || '';
+}
+
+// Bounds the misconfiguration hint below. A directly exposed server sees spoof
+// attempts continuously, and that is working as intended — the operator needs
+// the hint once, not a running commentary.
+const MAX_PROXY_MISCONFIG_WARNINGS = 10;
+
+/** Decides which peers may set client-identifying headers. */
+class ProxyTrust {
+  /**
+   * @param {string} spec comma-separated CIDRs and bare IPs, or '*' / 'none'.
+   * Invalid entries are warned about and skipped rather than thrown: a typo in
+   * one entry should narrow trust, not take the server down.
+   */
+  constructor(spec) {
+    this.trustAll = false;
+    this.ranges = [];
+    this.warnings = 0;
+
+    for (const raw of String(spec ?? '').split(',')) {
+      const entry = raw.trim();
+      if (!entry || entry.toLowerCase() === 'none') continue;
+      if (entry === '*') {
+        this.trustAll = true;
+        continue;
+      }
+      try {
+        if (entry.includes('/')) {
+          this.ranges.push(ipaddr.parseCIDR(entry));
+        } else {
+          // A bare IP is a single-host range.
+          const addr = ipaddr.process(entry);
+          this.ranges.push([addr, addr.kind() === 'ipv4' ? 32 : 128]);
+        }
+      } catch {
+        console.warn(`warning: ignoring invalid TRUSTED_PROXIES entry "${raw}"`);
+      }
+    }
+  }
+
+  /** Builds the trust set from TRUSTED_PROXIES, defaulting when unset. An
+   * explicitly empty value means "trust nothing", so unset and empty differ. */
+  static fromEnv(env = process.env) {
+    const spec = env.TRUSTED_PROXIES === undefined
+      ? DEFAULT_TRUSTED_PROXY_CIDRS.join(',')
+      : env.TRUSTED_PROXIES;
+    return new ProxyTrust(spec);
+  }
+
+  /** Renders the trust set for the startup log. */
+  describe() {
+    if (this.trustAll) return '* (all peers)';
+    if (this.ranges.length === 0) return 'none (forwarding headers ignored)';
+    return this.ranges.map(([addr, bits]) => `${addr.toString()}/${bits}`).join(', ');
+  }
+
+  /** Reports whether an address belongs to a configured proxy. */
+  trusts(ip) {
+    if (!ip) return false;
+    if (this.trustAll) return true;
+    let addr;
+    try {
+      addr = ipaddr.process(ip);
+    } catch {
+      return false;
+    }
+    for (const range of this.ranges) {
+      // match() throws when the address families differ, so skip those.
+      if (range[0].kind() !== addr.kind()) continue;
+      if (addr.match(range)) return true;
+    }
+    return false;
+  }
+
+  /** Reports whether the immediate peer may speak for another client. */
+  peerTrusted(req) {
+    return this.trusts(normalizeIP(req.socket?.remoteAddress));
+  }
+
+  /**
+   * Returns a header value only when the peer is a trusted proxy, and ''
+   * otherwise. Used for X-JA3-Hash: a client free to state its own TLS
+   * fingerprint would just present a stock Chrome one and erase the signal.
+   */
+  trustedHeader(req, name) {
+    if (!this.peerTrusted(req)) return '';
+    return headerValue(req, name);
+  }
+
+  /**
+   * Flags the one misconfiguration this design can cause. A peer that sends
+   * forwarding headers but is not in the trust set is either a real reverse
+   * proxy the operator forgot to list — in which case every visitor is now
+   * collapsed onto that proxy's address, silently wrecking rate limiting and IP
+   * reputation — or a client spoofing them, which is exactly what we want
+   * ignored. The two are indistinguishable here, so say both and name the peer
+   * so it can be pasted into TRUSTED_PROXIES.
+   */
+  _warnUntrustedForwarding(peer, req) {
+    if (!headerValue(req, 'x-forwarded-for') && !headerValue(req, 'x-real-ip')) return;
+    if (this.warnings >= MAX_PROXY_MISCONFIG_WARNINGS) return;
+    this.warnings++;
+    const suffix = this.warnings === MAX_PROXY_MISCONFIG_WARNINGS
+      ? ' (further warnings suppressed)'
+      : '';
+    console.warn(
+      `warning: ignoring forwarding headers from untrusted peer ${peer}. ` +
+      `If that is your reverse proxy, add it to TRUSTED_PROXIES — until then every ` +
+      `visitor is attributed to ${peer}. If not, a client is spoofing them and they ` +
+      `are correctly ignored.${suffix}`
+    );
+  }
+
+  /**
+   * Resolves the address to attribute the request to. Forwarding headers are
+   * read only when the peer is trusted; otherwise the socket address wins.
+   */
+  clientIP(req) {
+    const peer = normalizeIP(req.socket?.remoteAddress);
+    // No usable socket address means nothing to trust and nothing to forward.
+    if (!peer) return '127.0.0.1';
+    if (!this.trusts(peer)) {
+      this._warnUntrustedForwarding(peer, req);
+      return peer;
+    }
+
+    // A trusted proxy sets X-Real-IP to the address it accepted the connection
+    // from, so it needs no chain walking.
+    const realIP = normalizeIP(headerValue(req, 'x-real-ip'));
+    if (realIP) return realIP;
+
+    // Otherwise walk X-Forwarded-For right to left. Each proxy appends the peer
+    // it saw, so the rightmost entries are the ones our own infrastructure
+    // wrote. The first entry that is not a proxy we trust is the closest thing
+    // to a real client; everything to its left was supplied by that client and
+    // cannot be believed.
+    const forwarded = headerValue(req, 'x-forwarded-for');
+    if (forwarded) {
+      const parts = forwarded.split(',');
+      for (let i = parts.length - 1; i >= 0; i--) {
+        const entry = parts[i].trim();
+        if (!entry) continue;
+        const ip = normalizeIP(entry);
+        // A malformed hop makes everything to its left unreadable, so stop and
+        // fall back to the peer rather than guess.
+        if (!ip) break;
+        if (this.trusts(ip)) continue;
+        return ip;
+      }
+    }
+
+    return peer;
+  }
+}
+
+module.exports = {
+  ProxyTrust,
+  DEFAULT_TRUSTED_PROXY_CIDRS,
+  normalizeIP
+};

package/index.js

@@ -11,6 +11,7 @@
 
 const crypto = require('crypto');
 const detection = require('./detection');
+const { ProxyTrust } = require('./clientip');
 
 // =============================================================================
 // PoW Challenge Store (can be extended with Redis)
@@ -719,28 +720,25 @@ class ScoringEngine {
 
 function createMiddleware(options = {}) {
   const engine = new ScoringEngine(options);
+  const proxyTrust = options.trustedProxies !== undefined
+    ? new ProxyTrust(options.trustedProxies)
+    : ProxyTrust.fromEnv();
 
   return {
     engine,
+    proxyTrust,
 
-    // Middleware to extract IP from request
-    getIP: (req) => {
-      let ip = req.headers['x-real-ip'] || '';
-      if (!ip) {
-        const forwarded = req.headers['x-forwarded-for'];
-        if (forwarded) {
-          ip = forwarded.split(',')[0].trim();
-        } else {
-          ip = req.socket?.remoteAddress || '127.0.0.1';
-        }
-      }
-      return ip;
-    },
+    // Middleware to extract IP from request. Forwarding headers are honoured
+    // only from a peer in the trusted-proxy set (TRUSTED_PROXIES, or the
+    // private/loopback defaults) — otherwise any caller could claim a clean
+    // residential IP and skip the datacenter, Tor/VPN and rate-limit checks.
+    // Pass options.trustedProxies to override the env for this instance.
+    getIP: (req) => proxyTrust.clientIP(req),
 
     // Challenge route handler
     challengeHandler: (req, res) => {
       const siteKey = req.query.siteKey || 'default';
-      const ip = options.getIP ? options.getIP(req) : module.exports.createMiddleware({}).getIP(req);
+      const ip = options.getIP ? options.getIP(req) : proxyTrust.clientIP(req);
       const challenge = engine.generateChallenge(siteKey, ip);
 
       res.json({
@@ -755,7 +753,7 @@ function createMiddleware(options = {}) {
     // Verify route handler
     verifyHandler: (req, res) => {
       const { siteKey, signals, powSolution } = req.body;
-      const ip = options.getIP ? options.getIP(req) : module.exports.createMiddleware({}).getIP(req);
+      const ip = options.getIP ? options.getIP(req) : proxyTrust.clientIP(req);
       const userAgent = req.headers['user-agent'] || '';
 
       const headers = {};

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@webdecoy/fcaptcha",
-  "version": "1.15.0",
+  "version": "1.16.0",
   "description": "Open source CAPTCHA with PoW, bot detection, and Vision AI protection",
   "main": "index.js",
   "exports": {
@@ -11,7 +11,8 @@
   "scripts": {
     "start": "node server.js",
     "dev": "node --watch server.js",
-    "test:webbotauth": "node webbotauth.test.js"
+    "test:webbotauth": "node webbotauth.test.js",
+    "test:clientip": "node clientip.test.js"
   },
   "dependencies": {
     "cors": "^2.8.5",

package/server.js

@@ -10,6 +10,7 @@ const crypto = require('crypto');
 const path = require('path');
 const detection = require('./detection');
 const webbotauth = require('./webbotauth');
+const { ProxyTrust } = require('./clientip');
 
 const app = express();
 app.use(cors());
@@ -19,6 +20,14 @@ const SECRET_KEY = process.env.FCAPTCHA_SECRET || 'dev-secret-change-in-producti
 const PORT = process.env.PORT || 3000;
 const TRUSTED_JA4_HEADERS = detection.getTrustedJA4HeaderNames();
 
+// Which peers may speak for another client via X-Forwarded-For / X-Real-IP and
+// the TLS-fingerprint headers. See clientip.js.
+const PROXY_TRUST = ProxyTrust.fromEnv();
+
+// Express's own `trust proxy` would re-derive req.ip from the same headers on
+// its own terms; IP resolution goes through PROXY_TRUST.clientIP exclusively.
+app.set('trust proxy', false);
+
 // Verdict logging is off by default: a self-hosted FCaptcha emits no per-request
 // logs unless the operator opts in via FCAPTCHA_LOG_VERDICTS (1/true/yes/on).
 // When on, each /api/verify and /api/score logs one privacy-safe JSON line
@@ -1235,25 +1244,31 @@ app.get('/health', (req, res) => {
   res.json({ status: 'ok' });
 });
 
+// collectHeaders lowercases the request headers for the detectors, dropping the
+// TLS-fingerprint headers when the peer is not a proxy we trust. TRUSTED_JA4_HEADERS
+// is an allowlist of header *names*, which says nothing about who set them — without
+// this gate a client could send cf-ja4 itself and present a clean fingerprint.
+function collectHeaders(req) {
+  const peerTrusted = PROXY_TRUST.peerTrusted(req);
+  const headers = {};
+  for (const [key, value] of Object.entries(req.headers)) {
+    const name = key.toLowerCase();
+    if (!peerTrusted && TRUSTED_JA4_HEADERS.includes(name)) continue;
+    headers[name] = Array.isArray(value) ? value[0] : value;
+  }
+  return headers;
+}
+
 app.post('/api/verify', async (req, res) => {
   const { siteKey, signals, powSolution, signalsJson, powTiming } = req.body;
-  let ip = req.headers['x-real-ip'] || '';
-  if (!ip) {
-    const forwarded = req.headers['x-forwarded-for'];
-    if (forwarded) {
-      ip = forwarded.split(',')[0].trim();
-    } else {
-      ip = req.socket.remoteAddress;
-    }
-  }
+  const ip = PROXY_TRUST.clientIP(req);
   const userAgent = req.headers['user-agent'] || '';
-  const ja3Hash = req.headers['x-ja3-hash'] || null;
+  // Only honoured from a trusted proxy: a client that can state its own TLS
+  // fingerprint would just claim a stock Chrome one.
+  const ja3Hash = PROXY_TRUST.trustedHeader(req, 'x-ja3-hash') || null;
 
   // Collect headers for analysis
-  const headers = {};
-  for (const [key, value] of Object.entries(req.headers)) {
-    headers[key.toLowerCase()] = Array.isArray(value) ? value[0] : value;
-  }
+  const headers = collectHeaders(req);
 
   // Web Bot Auth verification needs the raw request; its verdict is scored.
   const webBotAuth = await verifyWebBotAuth(req);
@@ -1265,17 +1280,9 @@ app.post('/api/verify', async (req, res) => {
 
 app.post('/api/score', async (req, res) => {
   const { siteKey, signals, action, powSolution, signalsJson, powTiming } = req.body;
-  let ip = req.headers['x-real-ip'] || '';
-  if (!ip) {
-    const forwarded = req.headers['x-forwarded-for'];
-    if (forwarded) {
-      ip = forwarded.split(',')[0].trim();
-    } else {
-      ip = req.socket.remoteAddress;
-    }
-  }
+  const ip = PROXY_TRUST.clientIP(req);
   const userAgent = req.headers['user-agent'] || '';
-  const ja3Hash = req.headers['x-ja3-hash'] || null;
+  const ja3Hash = PROXY_TRUST.trustedHeader(req, 'x-ja3-hash') || null;
 
   const headers = {};
   for (const [key, value] of Object.entries(req.headers)) {
@@ -1300,15 +1307,7 @@ app.post('/api/token/verify', (req, res) => {
   const { token } = req.body;
 
   // Extract client IP for verification
-  let ip = req.headers['x-real-ip'] || '';
-  if (!ip) {
-    const forwarded = req.headers['x-forwarded-for'];
-    if (forwarded) {
-      ip = forwarded.split(',')[0].trim();
-    } else {
-      ip = req.socket.remoteAddress;
-    }
-  }
+  const ip = PROXY_TRUST.clientIP(req);
 
   res.json(verifyToken(token, ip));
 });
@@ -1317,15 +1316,7 @@ app.post('/api/token/verify', (req, res) => {
 app.get('/api/pow/challenge', (req, res) => {
   const siteKey = req.query.siteKey || 'default';
 
-  let ip = req.headers['x-real-ip'] || '';
-  if (!ip) {
-    const forwarded = req.headers['x-forwarded-for'];
-    if (forwarded) {
-      ip = forwarded.split(',')[0].trim();
-    } else {
-      ip = req.socket.remoteAddress;
-    }
-  }
+  const ip = PROXY_TRUST.clientIP(req);
 
   // Difficulty scaling based on IP reputation
   let difficulty = 4; // Default: ~100-500ms on average hardware
@@ -1373,4 +1364,5 @@ app.get('/api/challenge', (req, res) => {
 
 app.listen(PORT, () => {
   console.log(`FCaptcha server running on port ${PORT}`);
+  console.log(`Trusted proxies: ${PROXY_TRUST.describe()}`);
 });