Merge branch 'main' into feature/demo-default-events

server/src/app.js

@@ -100,6 +100,7 @@ if (isDemo()) {
     demoLock.robots(req, res)
   })
   app.use(demoLock.refuseMultipart)
+  app.use(demoLock.refuseDataUri)
   app.use('/api/v1', demoLock.createLock(() => modules.demoLocks()))
 }
 

server/src/demo/locks.js

@@ -1,6 +1,6 @@
 // ── The demo lock (docs/website/DEMO_MODE.md §3) ───────────────────────────
 //
-// With DEMO_MODE on, app.js mounts three middlewares from this file, and with it
+// With DEMO_MODE on, app.js mounts four middlewares from this file, and with it
 // off none of them is mounted at all, so a real shard runs none of this code.
 //
 //   • `refuseMultipart` — any request with a multipart body, anywhere, before a
@@ -8,6 +8,8 @@
 //     is locked without anyone remembering to list it, and a module cannot open one.
 //   • `lock` — each write request against core's list below plus every module's
 //     declared `demoLocks`, at `/api/v1`, ahead of every router.
+//   • `refuseDataUri` — any write whose JSON body carries a data: URI in any
+//     string (DM25): a pasted image is an upload that did not use multipart.
 //   • `noindex` + `robots` — nothing a visitor writes reaches a search engine.
 //
 // **The server is the lock (DM10).** Everything here is decided from the request
@@ -174,6 +176,50 @@ function refuseMultipart(req, res, next) {
   return refuse(res)
 }
 
+// DM25: an image pasted into rich text travels inside the JSON body as a data:
+// URI, which is an upload by another name. Any string anywhere in a write's
+// body is checked, so a field added later, a setting or a module's route is
+// covered without being listed, as with multipart.
+//
+// A data: URI is `data:` (not the tail of a word like "metadata:") followed by
+// a comma, a `;`, or a type/subtype and then a comma. Prose such as
+// "Data: none, yet" does not match. Before testing, HTML character references
+// are decoded and tabs and newlines removed, because a browser does both when
+// it reads an attribute: `&#100;ata:` and `da<newline>ta:` are still data: URIs.
+const DATA_URI = /(?<![a-z0-9+.-])data:\s*(?:[,;]|[a-z]+\/[a-z0-9.+-]+[^,<>"']{0,256},)/i
+
+function browserReading(s) {
+  return s
+    .replace(/&#x([0-9a-f]+);?/gi, (m, hex) => safeChar(parseInt(hex, 16), m))
+    .replace(/&#(\d+);?/g, (m, dec) => safeChar(parseInt(dec, 10), m))
+    .replace(/&colon;/gi, ':')
+    .replace(/&(?:tab|newline);/gi, '')
+    .replace(/[\t\n\r]/g, '')
+}
+
+function safeChar(code, original) {
+  return code > 0 && code <= 0x10ffff ? String.fromCodePoint(code) : original
+}
+
+function hasDataUri(body) {
+  const stack = [body]
+  while (stack.length) {
+    const value = stack.pop()
+    if (typeof value === 'string') {
+      if (value.length > 5 && DATA_URI.test(browserReading(value))) return true
+    } else if (value && typeof value === 'object') {
+      for (const v of Object.values(value)) stack.push(v)
+    }
+  }
+  return false
+}
+
+function refuseDataUri(req, res, next) {
+  if (!WRITES.includes(req.method) || !hasDataUri(req.body)) return next()
+  log.info('refused data: URI', { method: req.method, path: req.originalUrl })
+  return refuse(res)
+}
+
 function noindex(req, res, next) {
   res.set('X-Robots-Tag', 'noindex, nofollow')
   next()
@@ -191,6 +237,8 @@ module.exports = {
   PROTECTED_SETTINGS,
   createLock,
   refuseMultipart,
+  refuseDataUri,
+  hasDataUri,
   noindex,
   robots,
   refuse,

server/test/demoApp.test.js

@@ -64,6 +64,18 @@ test('an upload is refused before multer or auth sees it', async () => {
   })
 })
 
+test('a pasted data: image is refused before auth sees it, on an open route', async () => {
+  await withServer(async (base) => {
+    const res = await fetch(`${base}/api/v1/admin/posts`, {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({ title: 'x', body: '<p><img src="data:image/png;base64,iVBORw0KGgo="></p>' }),
+    })
+    assert.equal(res.status, 403)
+    assert.equal((await res.json()).code, 'DEMO_LOCKED')
+  })
+})
+
 test('every response carries noindex, and robots.txt disallows everything', async () => {
   await withServer(async (base) => {
     const health = await fetch(`${base}/api/health`)

server/test/demoLocks.test.js

@@ -279,6 +279,74 @@ test('any multipart request is refused before a router sees it, wherever it is s
   })
 })
 
+// DM25: a data: URI in any string of a write's body is an upload too.
+async function withDataUriLock(fn) {
+  const app = await startApp((a) => {
+    a.use(locks.refuseDataUri)
+    a.all('*', (req, res) => res.json({ reached: true }))
+  })
+  try {
+    return await fn((method, path, body) =>
+      fetch(`${app.url}/api/v1${path}`, {
+        method,
+        headers: { 'Content-Type': 'application/json' },
+        body: JSON.stringify(body),
+      }))
+  } finally {
+    await app.close()
+  }
+}
+
+test('a data: URI anywhere in a write body is refused, on any route and any write method', async () => {
+  await withDataUriLock(async (req) => {
+    const img = '<p>hi</p><img src="data:image/png;base64,iVBORw0KGgo=">'
+    await expectLocked(await req('POST', '/admin/posts', { title: 't', body: img }))
+    await expectLocked(await req('PUT', '/admin/wiki/1', { content: img }))
+    await expectLocked(await req('PATCH', '/rust/anything', { deep: [{ x: { y: [img] } }] }))
+    await expectLocked(await req('PUT', '/admin/settings', { hero_layout_draft: JSON.stringify([{ src: 'data:image/webp;base64,UklG' }]) }))
+    await expectLocked(await req('DELETE', '/player/teams/x/forum/1', { reason: img }))
+  })
+})
+
+test('every data: URI shape a browser would load is refused', async () => {
+  const shapes = [
+    'DATA:image/png;base64,AAAA',
+    'data:image/svg+xml,<svg xmlns="http://www.w3.org/2000/svg"/>',
+    'data:text/html;base64,PHNjcmlwdD4=',
+    'data:,hello',
+    'data:;base64,AAAA',
+    'data: image/png;base64,AAAA',
+    '<div style="background:url(data:image/gif;base64,R0lG)">',
+    '<img src="&#100;ata:image/png;base64,AAAA">',
+    '<img src="&#x64;&#x61;&#x74;&#x61;&#x3a;image/png;base64,AAAA">',
+    '<img src="data&colon;image/png;base64,AAAA">',
+    '<img src="da\nta:image/png;base64,AAAA">',
+    '<img src="data:im\tage/png;base64,AAAA">',
+  ]
+  for (const s of shapes) assert.equal(locks.hasDataUri({ body: s }), true, s)
+})
+
+test('prose that mentions data, reads, and bodies without strings pass', async () => {
+  for (const s of ['metadata: none', 'Data: none, yet', 'the data: it is fine', 'see https://example.com/data:x', 'mydata:image/png,x']) {
+    assert.equal(locks.hasDataUri({ body: s }), false, s)
+  }
+  assert.equal(locks.hasDataUri(undefined), false)
+  assert.equal(locks.hasDataUri({ n: 1, ok: true, list: [null, 2] }), false)
+  await withDataUriLock(async (req) => {
+    await expectReached(await req('POST', '/admin/posts', { title: 'json is fine', body: '<p>metadata: none</p>' }))
+  })
+  // A read is never refused, whatever it carries.
+  const app = await startApp((a) => {
+    a.use(locks.refuseDataUri)
+    a.all('*', (req, res) => res.json({ reached: true }))
+  })
+  try {
+    await expectReached(await fetch(`${app.url}/api/v1/public/settings?x=data:image/png;base64,AAAA`))
+  } finally {
+    await app.close()
+  }
+})
+
 // ── Module locks ───────────────────────────────────────────────────────────
 
 test("a module's declared locks are refused alongside core's", async () => {