fix: let named CORS origins send credentials (closes #155)

A bookmarklet could never reach an instance behind reverse-proxy auth.
on_prepare echoed Access-Control-Allow-Origin but never sent
Access-Control-Allow-Credentials, and hardcoded Allow-Headers to
Content-Type, so both approaches the reporter suggested in 2022 still
failed in a browser today: credentials:'include' was rejected for the
missing Allow-Credentials, and an explicit Authorization header was
rejected as not allowed by the preflight.

Naming an origin in CORS_ALLOWED_ORIGINS is a deliberate trust grant, so
a named origin may now send credentials and an Authorization header. The
'*' wildcard is not such a grant: it matches origins the operator never
enumerated, and since credentials require echoing the origin back rather
than sending '*', pairing the two would let any site the user visits
drive their instance with their own session. The wildcard therefore
keeps byte-for-byte the uncredentialed behaviour it has always had, and
a '*' anywhere in the list disables credentials for every origin in it,
with a startup warning so that combination is not silently confusing.

This matches the boundary socket.io already enforces: engineio defaults
cors_credentials to True, and its wildcard test is against the string
'*' while we pass a list, so it too grants credentials only to
explicitly listed origins.

Also sets Vary: Origin, appending rather than clobbering, so a shared
cache cannot hand one origin's Allow-Origin to another.

Verified in a real browser across the matrix: with the caller's origin
named, plain/credentialed/Authorization requests all succeed; under '*'
the credentialed and Authorization requests stay blocked; from an
unlisted origin everything stays blocked.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

README.md

@@ -98,7 +98,7 @@ Enabling `writeinfojson` or `writethumbnail` in `YTDL_OPTIONS` also writes a fee
 * __HTTPS__: Use `https` instead of `http` (__CERTFILE__ and __KEYFILE__ required). Defaults to `false`.
 * __CERTFILE__: HTTPS certificate file path.
 * __KEYFILE__: HTTPS key file path.
-* __CORS_ALLOWED_ORIGINS__: Comma-separated list of origins permitted to make cross-origin requests to the MeTube API; `*` allows all. When unset or empty, all cross-origin requests are denied. Required for browser extensions and bookmarklets — see [Sending links to MeTube](#-sending-links-to-metube).
+* __CORS_ALLOWED_ORIGINS__: Comma-separated list of origins permitted to make cross-origin requests to the MeTube API; `*` allows all. When unset or empty, all cross-origin requests are denied. Required for browser extensions and bookmarklets — see [Sending links to MeTube](#-sending-links-to-metube). Naming origins explicitly also lets them send credentials (a login cookie, or the `Authorization` header a reverse proxy checks), which `*` deliberately does not: it would let any site you visit drive your instance with your own session.
 * __ROBOTS_TXT__: A path to a `robots.txt` file mounted in the container.
 
 ## 🎛️ Configuring yt-dlp options
@@ -243,7 +243,7 @@ __Browser extensions__ allow right-clicking videos and sending them directly to
 * __Chrome:__ contributed by [Rpsl](https://github.com/rpsl) — install from the [Chrome Webstore](https://chrome.google.com/webstore/detail/metube-downloader/fbmkmdnlhacefjljljlbhkodfmfkijdh) or [from sources](https://github.com/Rpsl/metube-browser-extension).
 * __Firefox:__ contributed by [nanocortex](https://github.com/nanocortex) — install from [Firefox Addons](https://addons.mozilla.org/en-US/firefox/addon/metube-downloader) or get sources [here](https://github.com/nanocortex/metube-firefox-addon).
 
-__Bookmarklets__ send the currently open page to MeTube with one click. Add the origins of the sites where you use them to `CORS_ALLOWED_ORIGINS`, e.g. `https://www.youtube.com,https://www.vimeo.com`. The code (Chrome and Firefox variants, contributed by [kushfest](https://github.com/kushfest) and [shoonya75](https://github.com/shoonya75)) is in the [Bookmarklets wiki page](https://github.com/alexta69/metube/wiki/Bookmarklets).
+__Bookmarklets__ send the currently open page to MeTube with one click. Add the origins of the sites where you use them to `CORS_ALLOWED_ORIGINS`, e.g. `https://www.youtube.com,https://www.vimeo.com`. If your instance sits behind authentication, list the origins individually rather than using `*` — only named origins are allowed to send credentials. The code (Chrome and Firefox variants, contributed by [kushfest](https://github.com/kushfest) and [shoonya75](https://github.com/shoonya75)) is in the [Bookmarklets wiki page](https://github.com/alexta69/metube/wiki/Bookmarklets).
 
 __iOS Shortcut:__ [rithask](https://github.com/rithask) created an [iOS shortcut](https://www.icloud.com/shortcuts/66627a9f334c467baabdb2769763a1a6) for sending URLs to MeTube from Safari's share menu; it prompts for your instance address on first use.
 

app/main.py

@@ -335,6 +335,12 @@ async def state_dir_guard(request, handler):
 
 app = web.Application(middlewares=[state_dir_guard])
 _cors_origins = [o.strip() for o in config.CORS_ALLOWED_ORIGINS.split(',') if o.strip()] if config.CORS_ALLOWED_ORIGINS else []
+if '*' in _cors_origins and len(_cors_origins) > 1:
+    log.warning(
+        "CORS_ALLOWED_ORIGINS mixes '*' with named origins %s. '*' wins, and credentialed "
+        "cross-origin requests stay disabled for every origin in the list. Remove '*' if you "
+        "need a bookmarklet to reach an authenticated instance.",
+        [o for o in _cors_origins if o != '*'])
 sio = socketio.AsyncServer(cors_allowed_origins=_cors_origins if _cors_origins else [])
 sio.attach(app, socketio_path=config.URL_PREFIX + 'socket.io')
 routes = web.RouteTableDef()
@@ -1286,9 +1292,44 @@ app.router.add_route('OPTIONS', config.URL_PREFIX + 'delete-cookies', add_cors)
 
 async def on_prepare(request, response):
     origin = request.headers.get('Origin')
-    if origin and _cors_origins and ('*' in _cors_origins or origin in _cors_origins):
+    if not origin or not _cors_origins:
-        response.headers['Access-Control-Allow-Origin'] = origin
+        return
-        response.headers['Access-Control-Allow-Headers'] = 'Content-Type'
+
+    # Naming an origin in CORS_ALLOWED_ORIGINS is a deliberate trust grant, so
+    # such an origin may send credentials: the cookie or Authorization header
+    # that a reverse proxy in front of MeTube checks. Without this a bookmarklet
+    # cannot reach an authenticated instance at all (issue #155).
+    #
+    # The '*' wildcard is emphatically not such a grant — it matches origins the
+    # operator never enumerated, including every site the user happens to visit.
+    # Echoing the origin back (which we must do, since '*' is illegal alongside
+    # credentials) and allowing credentials would let any page drive the user's
+    # instance with the user's own session. So the wildcard keeps exactly the
+    # uncredentialed behaviour it has always had, and a wildcard anywhere in the
+    # list disables credentials for every origin in it.
+    #
+    # Derived here rather than held in a second module global so the wildcard
+    # test and the membership test can never disagree about the same list.
+    wildcard = '*' in _cors_origins
+    trusted = not wildcard and origin in _cors_origins
+    if not (wildcard or trusted):
+        return
+
+    response.headers['Access-Control-Allow-Origin'] = origin
+    # Authorization rides on the same grant: allowing it under the wildcard
+    # would let an arbitrary page attempt credentials against an instance it
+    # can already reach, from inside the victim's network.
+    response.headers['Access-Control-Allow-Headers'] = 'Content-Type, Authorization' if trusted else 'Content-Type'
+    if trusted:
+        response.headers['Access-Control-Allow-Credentials'] = 'true'
+
+    # The response now differs per Origin, so a shared cache must not hand one
+    # origin's Allow-Origin to another.
+    vary = response.headers.get('Vary')
+    if not vary:
+        response.headers['Vary'] = 'Origin'
+    elif 'origin' not in (v.strip().lower() for v in vary.split(',')):
+        response.headers['Vary'] = f'{vary}, Origin'
 
 app.on_response_prepare.append(on_prepare)
 

app/tests/test_api.py

@@ -525,3 +525,130 @@ async def test_download_blocks_state_dir_files(monkeypatch):
         (download_dir / "video.mp4").unlink(missing_ok=True)
         (download_dir / percent_filename).unlink(missing_ok=True)
         state_dir.rmdir()
+
+# --- CORS (issue #155) -------------------------------------------------------
+#
+# The security property under test: credentials are granted only to an origin
+# the operator named explicitly, and never under the '*' wildcard. Each test
+# builds a fresh Application because main.app binds to the first event loop
+# that runs it; the logic under test lives entirely in main.on_prepare, and the
+# real main.add_cors preflight handler is mounted so the preflight path is the
+# production one.
+
+async def _cors_version(request):
+    return web.Response(text="v")
+
+
+def _cors_app():
+    app = web.Application()
+    app.router.add_route("OPTIONS", "/add", main.add_cors)
+    app.router.add_get("/version", _cors_version)
+    app.on_response_prepare.append(main.on_prepare)
+    return app
+
+
+async def _cors_headers(monkeypatch, origins, origin, path="/add", method="OPTIONS"):
+    monkeypatch.setattr(main, "_cors_origins", origins)
+    async with TestClient(TestServer(_cors_app())) as client:
+        resp = await client.request(
+            method, path,
+            headers={
+                "Origin": origin,
+                "Access-Control-Request-Method": "POST",
+                "Access-Control-Request-Headers": "content-type,authorization",
+            },
+        )
+        return resp.headers
+
+
+@pytest.mark.asyncio
+async def test_cors_listed_origin_gets_credentials(monkeypatch):
+    h = await _cors_headers(monkeypatch, ["https://www.youtube.com"], "https://www.youtube.com")
+    assert h["Access-Control-Allow-Origin"] == "https://www.youtube.com"
+    assert h["Access-Control-Allow-Credentials"] == "true"
+    assert "Authorization" in h["Access-Control-Allow-Headers"]
+    assert "Origin" in h["Vary"]
+
+
+@pytest.mark.asyncio
+async def test_cors_wildcard_never_grants_credentials(monkeypatch):
+    h = await _cors_headers(monkeypatch, ["*"], "https://evil.example")
+    # The wildcard still reflects the origin, exactly as before...
+    assert h["Access-Control-Allow-Origin"] == "https://evil.example"
+    # ...but must not hand out the user's session, nor let a page attempt
+    # credentials of its own.
+    assert "Access-Control-Allow-Credentials" not in h
+    assert h["Access-Control-Allow-Headers"] == "Content-Type"
+
+
+@pytest.mark.asyncio
+async def test_cors_wildcard_mixed_with_named_origin_still_denies_credentials(monkeypatch):
+    # '*' anywhere in the list disables credentials for everyone in it, so a
+    # stray wildcard cannot silently widen a named grant.
+    h = await _cors_headers(monkeypatch, ["*", "https://www.youtube.com"], "https://www.youtube.com")
+    assert h["Access-Control-Allow-Origin"] == "https://www.youtube.com"
+    assert "Access-Control-Allow-Credentials" not in h
+    assert h["Access-Control-Allow-Headers"] == "Content-Type"
+
+
+@pytest.mark.asyncio
+async def test_cors_unlisted_origin_gets_nothing(monkeypatch):
+    h = await _cors_headers(monkeypatch, ["https://www.youtube.com"], "https://evil.example")
+    assert "Access-Control-Allow-Origin" not in h
+    assert "Access-Control-Allow-Credentials" not in h
+
+
+@pytest.mark.asyncio
+async def test_cors_disabled_by_default(monkeypatch):
+    h = await _cors_headers(monkeypatch, [], "https://www.youtube.com")
+    assert "Access-Control-Allow-Origin" not in h
+    assert "Access-Control-Allow-Credentials" not in h
+
+
+@pytest.mark.asyncio
+async def test_cors_credentials_apply_to_actual_response_not_just_preflight(monkeypatch):
+    # The browser checks Allow-Credentials on the real response too, so a
+    # preflight-only grant would still fail.
+    h = await _cors_headers(
+        monkeypatch, ["https://www.youtube.com"], "https://www.youtube.com",
+        path="/version", method="GET")
+    assert h["Access-Control-Allow-Credentials"] == "true"
+
+
+@pytest.mark.asyncio
+async def test_cors_origin_match_is_exact(monkeypatch):
+    # Substring or suffix matching here would be a bypass.
+    for impostor in (
+        "https://www.youtube.com.evil.example",
+        "https://evilwww.youtube.com",
+        "http://www.youtube.com",
+        "https://www.youtube.com:8443",
+    ):
+        h = await _cors_headers(monkeypatch, ["https://www.youtube.com"], impostor)
+        assert "Access-Control-Allow-Origin" not in h, impostor
+        assert "Access-Control-Allow-Credentials" not in h, impostor
+
+
+@pytest.mark.asyncio
+async def test_cors_null_origin_is_not_trusted(monkeypatch):
+    # Sandboxed iframes and some file:// contexts send Origin: null.
+    h = await _cors_headers(monkeypatch, ["https://www.youtube.com"], "null")
+    assert "Access-Control-Allow-Origin" not in h
+    assert "Access-Control-Allow-Credentials" not in h
+
+
+@pytest.mark.asyncio
+async def test_cors_vary_appends_to_existing_value(monkeypatch):
+    # Static responses can already carry a Vary; clobbering it would break
+    # content negotiation.
+    monkeypatch.setattr(main, "_cors_origins", ["https://www.youtube.com"])
+
+    async def handler(request):
+        return web.Response(text="x", headers={"Vary": "Accept-Encoding"})
+
+    app = web.Application()
+    app.router.add_get("/v", handler)
+    app.on_response_prepare.append(main.on_prepare)
+    async with TestClient(TestServer(app)) as client:
+        resp = await client.get("/v", headers={"Origin": "https://www.youtube.com"})
+        assert resp.headers["Vary"] == "Accept-Encoding, Origin"