Merge pull request #690 from daikaginza/fix/gbfs-rate-limit

fix(gbfs): rate-limit /api/gbfs

CHANGELOG.md

@@ -1,5 +1,14 @@
 # Changelog
 
+- Bound how often the GBFS proxy will relay. `/api/gbfs` already restricted
+  where it could go — https, an allowlisted host, a station feed path, no
+  redirects, a 12-second deadline and a 5 MB body cap — but nothing restricted
+  how often, so the dev server would make unlimited requests at a third-party
+  bikeshare operator on any caller's behalf. The route now admits 120
+  requests/minute per client address, refusing the rest with 429 and
+  `Retry-After` before any outbound fetch. The bikeshare layer needs about 15 a
+  minute with every catalogued system in range, so ordinary use is unchanged.
+
 - Add nine Ontario transit feeds: TTC, MiWay, Hamilton HSR, Durham Region,
   YRT/Viva, Guelph, Burlington, Barrie and Kingston, each carrying its own
   licence quote and attribution. A feed can now name its own modes, so Toronto's

docs/CURRENT-STATE.md

@@ -1262,6 +1262,15 @@ cancels the body unread. A rejected or timed-out body is an upstream miss and
 follows each route's existing error and cached-data paths. FIRMS map-key
 status, the OpenSky token exchange and CCTV media keep their current handling.
 
+The route also bounds request frequency: 120 requests/minute per client
+address, with a global backstop of three times that. A refused request answers
+429 with `Retry-After` and `Cache-Control: no-store` and performs no outbound
+fetch. The bikeshare layer's own demand is about 15 requests a minute with
+every catalogued system in range, so an ordinary session never meets the limit;
+the method, host and path guards run unchanged.
+
+Updated: September 21, 2026
+
 ## Remaining local service modules
 
 Overpass query validation, geometry simplification, disk caching and upstream

scripts/format-scope.json

@@ -1053,5 +1053,6 @@
   "scripts/generate-icons.mjs",
   "src/tooling/webAppIcons.test.mjs",
   "src/ui/flyToDataSource.js",
-  "src/ui/flyToDataSource.test.mjs"
+  "src/ui/flyToDataSource.test.mjs",
+  "src/data/gbfsProxyRateLimit.test.mjs"
 ]

scripts/package-boundaries.json

@@ -158,6 +158,8 @@
     "modules": [
       "server/providers/gbfs.js",
       "server/providers/common/http.js",
+      "server/providers/common/rate-limit.js",
+      "src/sources/rateLimit.js",
       "src/data/gbfsSource.js",
       "src/sources/httpBody.js"
     ],

server/providers/gbfs.js

@@ -1,4 +1,5 @@
 import { readResponseTextCapped } from './common/http.js';
+import { makeRateLimiter, clientKey } from './common/rate-limit.js';
 import {
   isAllowedGbfsHost,
   isAllowedGbfsPath,
@@ -17,6 +18,19 @@ export const GBFS_CLIENT_IDENTIFIER = 'gods-eye-view';
 
 export const GBFS_MAX_BODY_BYTES = 5 * 1024 * 1024; // 5 MB
 
+/**
+ * Requests/minute/IP admitted by the proxy.
+ *
+ * The bikeshare layer's own demand is small and known: `src/layers/bikeshare/
+ * registry.js` catalogues 15 systems, station information is cached per city
+ * after its first fetch, and station status refreshes once per city every
+ * `STATUS_POLL_MS` (60 s) with in-flight requests deduplicated. Even with every
+ * catalogued system in camera range at once that is about 15 requests a minute,
+ * so 120 is eight times a maximal real session while still capping a caller
+ * using this proxy as an open relay against a third-party GBFS host.
+ */
+export const GBFS_RATE_PER_MIN = 120;
+
 function gbfsRedirectHost(location, requestUrl) {
   if (!location) return '';
   try {
@@ -121,7 +135,12 @@ export async function fetchGbfsUpstream(
  *
  * @returns {import('vite').Plugin}
  */
-export function gbfsProxy() {
+export function gbfsProxy({ ratePerMinute = GBFS_RATE_PER_MIN } = {}) {
+  const allow = makeRateLimiter({
+    windowMs: 60_000,
+    max: ratePerMinute,
+    globalMax: ratePerMinute * 3,
+  });
   const installMiddleware = (server) => {
     server.middlewares.use('/api/gbfs', async (req, res) => {
       try {
@@ -134,6 +153,19 @@ export function gbfsProxy() {
           return;
         }
 
+        // Frequency, checked before the target is even parsed: the allowlist
+        // already says WHERE this proxy may go, nothing said HOW OFTEN. A
+        // refused request performs no outbound fetch and starts no timer.
+        if (!allow(clientKey(req))) {
+          res.writeHead(429, {
+            'Content-Type': 'application/json',
+            'Cache-Control': 'no-store',
+            'Retry-After': '10',
+          });
+          res.end(JSON.stringify({ error: 'Rate limit exceeded' }));
+          return;
+        }
+
         const url = new URL(req.url || '/', 'http://localhost');
         const encodedTarget = url.pathname.replace(/^\/+/, '');
         if (!encodedTarget) {

src/data/gbfsProxyRateLimit.test.mjs

@@ -0,0 +1,152 @@
+// GBFS PROXY — the frequency the allowlist never bounded.
+//
+// `/api/gbfs` already says WHERE it may go: an https allowlisted host, a
+// station_information/station_status path, a 12 s timeout, a 5 MB body cap and
+// no redirects. Nothing said HOW OFTEN, so the dev server would relay an
+// unbounded stream of requests at a third-party bikeshare operator on any
+// caller's behalf — one outbound fetch and one timeout timer each.
+//
+// These cases drive the real mounted middleware. `fetch` is stubbed per test
+// and restored afterwards, so a case that reaches the network fails loudly,
+// and the stub's call count is what proves a refused request costs nothing.
+//
+// Run with: npm test
+import { test } from 'node:test';
+import assert from 'node:assert/strict';
+import { gbfsProxy, GBFS_RATE_PER_MIN } from '../../server/providers/gbfs.js';
+
+const STATUS_URL = 'https://gbfs.lyft.com/gbfs/2.3/bay/en/station_status.json';
+const STATION_BODY = '{"data":{"stations":[]}}';
+
+/** Mount the proxy the way the dev server does; returns its route handler. */
+function mount(options) {
+  const routes = new Map();
+  gbfsProxy(options).configureServer({
+    middlewares: {
+      use(route, handler) {
+        routes.set(route, handler);
+      },
+    },
+  });
+  return routes.get('/api/gbfs');
+}
+
+function request(handler, { target = STATUS_URL, remoteAddress, method } = {}) {
+  return new Promise((resolve, reject) => {
+    const req = {
+      method: method || 'GET',
+      // connect strips the mount prefix before the handler sees the URL.
+      url: '/' + encodeURIComponent(target),
+      headers: {},
+      socket: remoteAddress ? { remoteAddress } : undefined,
+    };
+    let status = 200;
+    let headers = {};
+    const res = {
+      writeHead(code, sent) {
+        status = code;
+        headers = sent || {};
+        return res;
+      },
+      end(body = '') {
+        let parsed = null;
+        try {
+          parsed = body ? JSON.parse(String(body)) : null;
+        } catch {
+          parsed = String(body);
+        }
+        resolve({ status, headers, body: parsed });
+      },
+    };
+    Promise.resolve(handler(req, res)).catch(reject);
+  });
+}
+
+/** Answer every allowlisted upstream, counting the calls that got through. */
+function stubUpstream(t) {
+  const asked = [];
+  const original = globalThis.fetch;
+  globalThis.fetch = async (url) => {
+    asked.push(String(url));
+    return new Response(STATION_BODY, {
+      status: 200,
+      headers: { 'Content-Type': 'application/json' },
+    });
+  };
+  t.after(() => {
+    globalThis.fetch = original;
+  });
+  return asked;
+}
+
+test('a maximal real session is never throttled', async (t) => {
+  const asked = stubUpstream(t);
+  const handler = mount();
+
+  // The bikeshare layer catalogues 15 systems and refreshes each one's status
+  // once every 60 s, so a minute of the busiest possible session is ~15
+  // requests. Run eight times that and expect every one of them served.
+  let served = 0;
+  for (let i = 0; i < 120; i += 1) {
+    const res = await request(handler);
+    if (res.status === 200) served += 1;
+  }
+  assert.equal(served, 120, 'the shipped ceiling leaves a real session alone');
+  assert.equal(asked.length, 120);
+  assert.equal(GBFS_RATE_PER_MIN, 120, 'the ceiling this case relies on');
+});
+
+test('an unbounded relay is refused, and refusal costs no outbound fetch', async (t) => {
+  const asked = stubUpstream(t);
+  const handler = mount({ ratePerMinute: 5 });
+
+  const responses = [];
+  for (let i = 0; i < 12; i += 1) responses.push(await request(handler));
+
+  const served = responses.filter((r) => r.status === 200);
+  const refused = responses.filter((r) => r.status === 429);
+  assert.equal(served.length, 5, 'exactly the admitted budget is served');
+  assert.equal(refused.length, 7);
+  assert.equal(refused[0].body.error, 'Rate limit exceeded');
+  assert.equal(refused[0].headers['Retry-After'], '10');
+  assert.equal(refused[0].headers['Cache-Control'], 'no-store');
+  assert.equal(
+    asked.length,
+    5,
+    'a refused request never reaches the GBFS operator',
+  );
+});
+
+test('a different client keeps its own budget', async (t) => {
+  stubUpstream(t);
+  const handler = mount({ ratePerMinute: 3 });
+
+  for (let i = 0; i < 6; i += 1)
+    await request(handler, { remoteAddress: '10.0.0.1' });
+  const noisy = await request(handler, { remoteAddress: '10.0.0.1' });
+  assert.equal(noisy.status, 429, 'the noisy client spent its own budget');
+
+  const quiet = await request(handler, { remoteAddress: '10.0.0.2' });
+  assert.equal(quiet.status, 200, 'a quiet client is not collateral');
+});
+
+test('the limiter does not displace the existing guards', async (t) => {
+  const asked = stubUpstream(t);
+  const handler = mount({ ratePerMinute: 10 });
+
+  const wrongMethod = await request(handler, { method: 'POST' });
+  assert.equal(wrongMethod.status, 405);
+
+  const offAllowlist = await request(handler, {
+    target: 'https://example.com/gbfs/station_status.json',
+  });
+  assert.equal(offAllowlist.status, 403);
+  assert.equal(offAllowlist.body.error, 'GBFS host not allowed');
+
+  const wrongPath = await request(handler, {
+    target: 'https://gbfs.lyft.com/gbfs/2.3/bay/en/system_regions.json',
+  });
+  assert.equal(wrongPath.status, 400);
+
+  assert.equal(asked.length, 0, 'none of those reached upstream');
+});