fix(overpass): bound query validation and reject cross-site requests

docs/CURRENT-STATE.md

@@ -2733,11 +2733,16 @@ down | auth-failed` (plus the unchanged `missing-key`/`unsupported`) with
 >   first later successful fetch. Smaller chunks reduce timeout risk; complete
 >   camera batches still wait for their sequential requests, and unresolved
 >   placement continues to use the existing prior until real heights arrive.
-> - **Overpass cache admission:** `/api/overpass` parses and sanitizes requests,
->   then checks fresh memory, identical in-flight work, and fresh disk entries
->   before invoking its local 90/min limiter. Cache and single-flight responses
->   therefore do not spend quota; upstream-bound misses retain the existing
->   limiter, configured-upstream, stale, and sanitization behavior. An unconfigured transport serves dated stale data or a non-retryable capability error.
+> - **Overpass cache admission:** `/api/overpass` refuses cross-site and forwarded
+>   POSTs through the shared local-request gate, then limits requests to 300/min
+>   per socket peer and 900/min globally before reading or validating bodies.
+>   Tag-filter brackets are scanned once; nested or unmatched brackets return 400.
+>   It then checks fresh memory, identical in-flight work, and fresh disk entries
+>   before invoking its separate local 90/min upstream limiter. Cache and
+>   single-flight responses therefore do not spend upstream quota; upstream-bound
+>   misses retain the existing limiter, configured-upstream, stale, and sanitization
+>   behavior. An unconfigured transport serves dated stale data or a non-retryable
+>   capability error.
 > - **CCTV world-click focus:** clicking an in-world CCTV icon or ambient card
 >   activates it and routes the camera flight through the panel FOCUS policy.
 >   Aircraft/satellite tracking releases outside cockpit; cockpit retains the

scripts/format-scope.json

@@ -1070,5 +1070,6 @@
   "scripts/assemble-changelog.mjs",
   "src/tooling/changelogFragments.test.mjs",
   "src/data/gbfsProxyRateLimit.test.mjs",
-  "src/terrainHeightsBounds.test.mjs"
+  "src/terrainHeightsBounds.test.mjs",
+  "server/providers/overpass/requestAdmission.test.mjs"
 ]

scripts/package-boundaries.json

@@ -253,6 +253,7 @@
       "server/providers/common/http.js",
       "server/providers/common/rate-limit.js",
       "server/providers/common/request.js",
+      "server/providers/common/same-site.js",
       "server/providers/overpass.js",
       "server/providers/overpass/cache.js",
       "server/providers/overpass/constants.js",
@@ -262,6 +263,7 @@
       "server/providers/places/routes.js",
       "src/data/placeProviderPayloads.js",
       "src/data/routeSteps.js",
+      "src/localRequestGate.mjs",
       "src/sources/httpBody.js",
       "src/i18n/marker.js",
       "src/sources/overpass.js",

server/providers/overpass.js

@@ -1,5 +1,6 @@
 import { makeRateLimiter, clientKey } from './common/rate-limit.js';
 import { readRequestBodyCapped } from './common/request.js';
+import { admitSameSite } from './common/same-site.js';
 import {
   OVERPASS_MAX_BODY_BYTES,
   OVERPASS_MAX_CONCURRENT,
@@ -27,6 +28,14 @@ const _overpassInFlight = new Map();
 
 let _overpassConcurrent = 0;
 
+// Bound body-reading and validation work, including malformed requests and
+// cache hits. Keep the tighter upstream budget below independent of this gate.
+const _overpassRequestLimiter = makeRateLimiter({
+  windowMs: 60_000,
+  max: 300,
+  globalMax: 900,
+});
+
 const _overpassRateLimiter = makeRateLimiter({
   windowMs: 60_000,
   max: 90,
@@ -94,6 +103,17 @@ function overpassProxy({ routing = {} } = {}) {
           return;
         }
 
+        if (admitSameSite(req, res)) return;
+        if (!_overpassRequestLimiter(clientKey(req))) {
+          res.writeHead(429, {
+            'Content-Type': 'application/json',
+            'Cache-Control': 'no-store',
+            'Retry-After': '60',
+          });
+          res.end(JSON.stringify({ error: 'Rate limit exceeded' }));
+          return;
+        }
+
         // Collect POST body with a hard byte cap (Overpass QL queries are small)
         let body;
         try {

server/providers/overpass/query.js

@@ -18,6 +18,29 @@ function isOverpassBoundaryQuery(cacheKey) {
   return /is_in\s*\(|\bpivot\b/i.test(String(cacheKey || ''));
 }
 
+// Input has already had quoted literals and comments removed. Consume each
+// character once: a regex that searches for a closing bracket from every `[` is
+// quadratic on an unterminated filter. Nested or unmatched brackets are invalid.
+function stripTagFilters(text) {
+  const parts = [];
+  let start = 0;
+  let inFilter = false;
+  for (let i = 0; i < text.length; i += 1) {
+    if (text[i] === '[') {
+      if (inFilter) return null;
+      parts.push(text.slice(start, i), ' ');
+      inFilter = true;
+    } else if (text[i] === ']') {
+      if (!inFilter) return null;
+      inFilter = false;
+      start = i + 1;
+    }
+  }
+  if (inFilter) return null;
+  parts.push(text.slice(start));
+  return parts.join('');
+}
+
 /**
  * Validate + clamp an Overpass form body. Defends the generic proxy against
  * planet-scale abuse: requires exactly one `data` query in which EVERY element
@@ -50,6 +73,10 @@ function sanitizeOverpassBody(rawBody) {
   // Blank quoted literals + strip comments in one lexer pass so a fake bound or a
   // `//` inside a string can't hide an unbounded selector (or satisfy a bound).
   const stripped = stripOverpassNoise(data);
+  const withoutTags = stripTagFilters(stripped);
+  if (withoutTags === null) {
+    return { ok: false, error: 'Malformed Overpass tag filter' };
+  }
 
   // Reject oversized radii in EVERY around form — point `around:r,lat,lon` AND the
   // input-set form `around.set:r` — and parse the full numeric token so scientific
@@ -93,20 +120,19 @@ function sanitizeOverpassBody(rawBody) {
   // statement. So `way[...]->.a` (global assigned to a set) is rejected, while the
   // app's `is_in(...)->.a; area.a[...]` and `area(id)->.x; rel(pivot.x)` validate.
   const boundedSets = new Set();
-  for (let stmt of stripped.split(';')) {
+  for (let stmt of withoutTags.split(';')) {
     stmt = stmt.trim();
     if (!stmt || stmt.startsWith('[') || /^out\b/.test(stmt)) continue;
 
-    // Strip output-set assignments (NOT input bounds), then strip bracket tag
-    // filters so a tag KEY/value (e.g. `way[is_in]`, `node[around]`) can never be
+    // Strip output-set assignments (NOT input bounds). Bracket tag filters were
+    // removed above so a tag KEY/value (e.g. `way[is_in]`, `node[around]`) cannot be
     // misread as a spatial bound. Bounds live in (...) / function calls / set
     // refs, never inside [...], so the probe loses nothing real.
     const outSets = [];
-    const body = stmt.replace(/->\s*\.(\w+)/g, (_, name) => {
+    const probe = stmt.replace(/->\s*\.(\w+)/g, (_, name) => {
       outSets.push(name);
       return ' ';
     });
-    const probe = body.replace(/\[[^\]]*\]/g, ' ');
 
     // Reject element-in-area scans on the TAG-STRIPPED probe, so a tag filter
     // between the selector and the area filter (way["highway"](area.a)) can't hide

server/providers/overpass/requestAdmission.test.mjs

@@ -0,0 +1,178 @@
+import test from 'node:test';
+import assert from 'node:assert/strict';
+import { Readable } from 'node:stream';
+import { spawnSync } from 'node:child_process';
+import { overpassProxy } from '../overpass.js';
+import { sanitizeOverpassBody } from './query.js';
+
+const form = (query) => `data=${encodeURIComponent(query)}`;
+
+test('bracket validation rejects malformed filters without trusting their contents as bounds', () => {
+  for (const query of [
+    'node[',
+    'node]',
+    'node[[name]](around:20,1,1);out;',
+    'node[name=around:20];out;',
+    'node[is_in];out;',
+    'node[name="(around:20,1,1)"];out;',
+    'node(around:20,1,1);way[name];out;',
+    'way[highway](area.a);out;',
+    'node[->.a](around:20,1,1);way.a;out;',
+  ]) {
+    assert.equal(sanitizeOverpassBody(form(query)).ok, false, query);
+  }
+});
+
+test('bounded queries preserve quoted brackets, comments, multiple tags and timeout clamping', () => {
+  for (const query of [
+    '[out:json][timeout:999];node[name="[Town]; ]"](around:20,1,1);out;',
+    '/* [ */way[highway][name](30.1,-97.9,30.5,-97.5);out geom;',
+    'is_in(30.2,-97.7)->.a;area.a[boundary="administrative"];out tags;',
+    'area(3600039152)->.x;rel(pivot.x);out geom;',
+  ]) {
+    const result = sanitizeOverpassBody(form(query));
+    assert.equal(result.ok, true, query);
+    assert.equal(
+      new URLSearchParams(result.body).get('data'),
+      query.replace('[timeout:999]', '[timeout:30]'),
+    );
+  }
+});
+
+test('a burst of maximum-size unterminated filters completes within a bounded process deadline', () => {
+  // A child deadline can stop synchronous regex backtracking; a test timeout in
+  // the same event loop cannot. The old parser takes several seconds for this.
+  const result = spawnSync(
+    process.execPath,
+    [
+      '--input-type=module',
+      '-e',
+      `import { sanitizeOverpassBody } from ${JSON.stringify(new URL('./query.js', import.meta.url).href)};
+       for (let i = 0; i < 20; i++) {
+         if (sanitizeOverpassBody('data=node' + '['.repeat(24000)).ok) process.exit(1);
+       }`,
+    ],
+    { timeout: 2000, encoding: 'utf8' },
+  );
+  assert.ifError(result.error);
+  assert.equal(result.status, 0, result.stderr);
+});
+
+function handler(hook = 'configureServer') {
+  let route;
+  overpassProxy()[hook]({
+    middlewares: {
+      use(path, fn) {
+        if (path === '/api/overpass') route = fn;
+      },
+    },
+  });
+  return route;
+}
+
+async function call(
+  route,
+  {
+    headers = {},
+    peer = '127.0.0.1',
+    body = 'data=node[',
+    forbidRead = false,
+    method = 'POST',
+    url = '/',
+  } = {},
+) {
+  let reads = 0;
+  const req = new Readable({
+    read() {
+      reads++;
+      assert.equal(forbidRead, false, 'refused request must not read its body');
+      this.push(Buffer.from(body));
+      this.push(null);
+    },
+  });
+  Object.assign(req, {
+    method,
+    url,
+    headers: { host: 'localhost:4173', ...headers },
+    socket: { remoteAddress: peer },
+  });
+  const res = {
+    statusCode: 200,
+    headers: {},
+    setHeader(key, value) {
+      this.headers[key.toLowerCase()] = value;
+    },
+    writeHead(status, values) {
+      this.statusCode = status;
+      for (const [key, value] of Object.entries(values))
+        this.setHeader(key, value);
+    },
+    end(value) {
+      this.body = value;
+    },
+  };
+  await route(req, res);
+  return { ...res, reads };
+}
+
+for (const hook of ['configureServer', 'configurePreviewServer']) {
+  test(`${hook}: cross-site requests are rejected before consuming a body or fetching`, async (t) => {
+    t.mock.method(globalThis, 'fetch', () =>
+      assert.fail('unexpected upstream fetch'),
+    );
+    const route = handler(hook);
+    for (const headers of [
+      { origin: 'https://foreign.example', 'content-type': 'text/plain' },
+      { 'sec-fetch-site': 'cross-site' },
+      { 'sec-fetch-site': 'same-site' },
+      { origin: 'null' },
+      { 'x-forwarded-for': '127.0.0.1' },
+    ]) {
+      const response = await call(route, { headers, forbidRead: true });
+      assert.equal(response.statusCode, 403);
+      assert.equal(response.reads, 0);
+      assert.equal(response.headers['cache-control'], 'no-store');
+    }
+  });
+}
+
+test('same-origin, LAN and non-browser requests still reach validation; GET status stays available', async () => {
+  const route = handler();
+  for (const headers of [
+    {},
+    { origin: 'http://localhost:4173', 'sec-fetch-site': 'same-origin' },
+    { host: '192.168.1.10:4173', origin: 'http://192.168.1.10:4173' },
+  ]) {
+    const response = await call(route, { headers });
+    assert.equal(response.statusCode, 400);
+    assert.equal(response.reads, 1);
+    assert.match(response.body, /Malformed Overpass tag filter/);
+  }
+  const response = await call(route, {
+    method: 'GET',
+    url: '/status',
+    forbidRead: true,
+  });
+  assert.equal(response.statusCode, 200);
+  assert.equal(typeof JSON.parse(response.body).configured, 'boolean');
+});
+
+test('invalid requests spend admission quota before parsing, including on an unconfigured server', async (t) => {
+  let now = Date.now() + 120000;
+  t.mock.method(Date, 'now', () => now);
+  t.mock.method(globalThis, 'fetch', () =>
+    assert.fail('unexpected upstream fetch'),
+  );
+  const route = handler();
+  const peer = 'admission-limit-test';
+  for (let i = 0; i < 300; i++) {
+    assert.equal((await call(route, { peer })).statusCode, 400);
+  }
+  const denied = await call(route, { peer, forbidRead: true });
+  assert.equal(denied.statusCode, 429);
+  assert.equal(denied.reads, 0);
+  assert.equal(denied.headers['retry-after'], '60');
+  assert.equal((await call(route, { peer: 'other-peer' })).statusCode, 400);
+  now += 60001;
+  assert.equal((await call(route, { peer })).statusCode, 400);
+});