diff --git a/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py b/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py index 3c2ae02dc52..3c5f7da1859 100644 --- a/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py +++ b/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py @@ -708,8 +708,7 @@ async def anthropic_proxy_route( endpoint_func: Final = create_pass_through_route( endpoint=endpoint, target=str(updated_url), - custom_headers=auth_header if auth_header is not None else {}, - _forward_headers=True, + custom_headers=_upstream_headers_for_anthropic_route(request, user_api_key_dict, auth_header), is_streaming_request=is_streaming_request, ) # dynamically construct pass-through endpoint based on incoming path received_value: Final = await endpoint_func( @@ -1909,6 +1908,19 @@ _HEADERS_NEVER_FORWARDED_TO_VERTEX: Final = frozenset({"content-length", "host"} SpecialHeaders.litellm_credential_header_names() - _VERTEX_UPSTREAM_CREDENTIAL_HEADERS ) +_CREDENTIALLESS_ANTHROPIC_MISSING_CREDENTIAL_DETAIL: Final = ( + "No Anthropic credential is configured on this proxy and the request carried no upstream " + "Anthropic credential. The LiteLLM virtual key is not forwarded to Anthropic. Configure an " + "Anthropic credential (ANTHROPIC_API_KEY or ANTHROPIC_AUTH_TOKEN, or a model with " + "use_in_pass_through: true), or send your own Anthropic API key in the x-api-key header or " + "your own Anthropic OAuth token in the Authorization header." +) + +_ANTHROPIC_UPSTREAM_CREDENTIAL_HEADERS: Final = frozenset({"authorization", "x-api-key"}) +_HEADERS_NEVER_FORWARDED_TO_ANTHROPIC: Final = frozenset({"content-length", "host", "accept-encoding"}) | ( + SpecialHeaders.litellm_credential_header_names() - _ANTHROPIC_UPSTREAM_CREDENTIAL_HEADERS +) + _MAPPED_ROUTE_CALLER_KEY_HEADER: Final = "litellm_user_api_key" @@ -1964,26 +1976,47 @@ def _is_authenticated_caller_secret(value: str, user_api_key_dict: UserAPIKeyAut return hmac.compare_digest(stored_representation.encode(), authenticated_key.encode()) +def _caller_headers_without_litellm_secrets( + request: Request, user_api_key_dict: UserAPIKeyAuth, never_forwarded: frozenset[str] +) -> Mapping[str, str]: + """Incoming headers minus the ones only LiteLLM consumes and minus whatever value authenticated the caller.""" + incoming: Final = _safe_get_request_headers(request) + dropped_by_name: Final = never_forwarded.union( + (_MAPPED_ROUTE_CALLER_KEY_HEADER, *_operator_configured_caller_key_header_names()) + ) + return MappingProxyType( + { + name: value + for name, value in incoming.items() + if name not in dropped_by_name and not _is_authenticated_caller_secret(value, user_api_key_dict) + } + ) + + def _forwarded_headers_for_credentialless_vertex_passthrough( request: Request, user_api_key_dict: UserAPIKeyAuth ) -> Mapping[str, str]: """Caller headers to forward on the bring-your-own-credentials Vertex branch, minus LiteLLM secrets.""" - incoming: Final = _safe_get_request_headers(request) - never_forwarded: Final = _HEADERS_NEVER_FORWARDED_TO_VERTEX.union( - (_MAPPED_ROUTE_CALLER_KEY_HEADER, *_operator_configured_caller_key_header_names()) + forwarded: Final = _caller_headers_without_litellm_secrets( + request, user_api_key_dict, _HEADERS_NEVER_FORWARDED_TO_VERTEX ) - forwarded: Final = MappingProxyType( - { - name: value - for name, value in incoming.items() - if name not in never_forwarded and not _is_authenticated_caller_secret(value, user_api_key_dict) - } - ) - if "authorization" not in forwarded and "x-goog-api-key" not in forwarded: + if _VERTEX_UPSTREAM_CREDENTIAL_HEADERS.isdisjoint(forwarded): raise HTTPException(status_code=401, detail=_CREDENTIALLESS_VERTEX_MISSING_CREDENTIAL_DETAIL) return forwarded +def _upstream_headers_for_anthropic_route( + request: Request, user_api_key_dict: UserAPIKeyAuth, proxy_auth_header: Mapping[str, str] | None +) -> Mapping[str, str]: + """Caller headers minus LiteLLM secrets, with the proxy's own Anthropic credential layered on top.""" + caller_headers: Final = _caller_headers_without_litellm_secrets( + request, user_api_key_dict, _HEADERS_NEVER_FORWARDED_TO_ANTHROPIC + ) + if proxy_auth_header is None and _ANTHROPIC_UPSTREAM_CREDENTIAL_HEADERS.isdisjoint(caller_headers): + raise HTTPException(status_code=401, detail=_CREDENTIALLESS_ANTHROPIC_MISSING_CREDENTIAL_DETAIL) + return MappingProxyType({**caller_headers, **(proxy_auth_header or {})}) + + async def _prepare_vertex_auth_headers( request: Request, vertex_credentials: VertexPassThroughCredentials | None, diff --git a/tests/test_litellm/proxy/pass_through_endpoints/test_llm_pass_through_endpoints.py b/tests/test_litellm/proxy/pass_through_endpoints/test_llm_pass_through_endpoints.py index 7b285674145..8cb9afc1efb 100644 --- a/tests/test_litellm/proxy/pass_through_endpoints/test_llm_pass_through_endpoints.py +++ b/tests/test_litellm/proxy/pass_through_endpoints/test_llm_pass_through_endpoints.py @@ -28,6 +28,7 @@ from litellm.proxy.pass_through_endpoints.llm_passthrough_endpoints import ( BaseOpenAIPassThroughHandler, RouteChecks, _join_url_paths, + anthropic_proxy_route, azure_proxy_route, bedrock_llm_proxy_route, bedrock_proxy_route, @@ -4285,6 +4286,298 @@ class TestVertexCredentiallessPassthroughVirtualKeyLeak: assert "sk-master-1234" not in " ".join(f"{name}:{value}" for name, value in forwarded.items()) +class TestAnthropicPassthroughVirtualKeyLeak: + """Regression coverage for LIT-3550. + + ``/anthropic/{endpoint}`` forwarded every incoming header to Anthropic, so the + header that carried the caller's LiteLLM virtual key (``Authorization``, + ``x-api-key``, ``x-litellm-api-key``, or an operator-configured name) reached + Anthropic and was rejected there as an invalid credential, with or without a + proxy-side Anthropic key layered on top. The virtual key must never leave the + proxy: it is dropped by value from the headers Anthropic reads as credentials + (``Authorization`` / ``x-api-key``), the proxy-only credential headers are + dropped by name, a caller's own Anthropic credential still passes through, and + a request with neither a proxy credential nor a caller credential fails with a + clean 401 instead of reaching ``create_pass_through_route``. + + The forwarded set is rebuilt the way ``pass_through_request`` builds it from + the captured ``create_pass_through_route`` kwargs, so a route that re-enables + ``_forward_headers`` fails these tests the same way the original bug did. + """ + + VKEY = "sk-litellm-victim-key" + PROXY_KEY = "sk-ant-api03-proxy-configured-key" + ENDPOINT = "v1/messages" + + async def _run( + self, + monkeypatch, + headers: list[tuple[bytes, bytes]], + authenticated: UserAPIKeyAuth | None = None, + master_key: str | None = "sk-master-1234", + proxy_api_key: str | None = None, + ) -> tuple[HTTPException | None, dict | None]: + from litellm.proxy.pass_through_endpoints.pass_through_endpoints import HttpPassThroughEndpointHelpers + from litellm.proxy.pass_through_endpoints.passthrough_endpoint_router import ( + PassthroughEndpointRouter, + ) + + monkeypatch.setattr("litellm.proxy.proxy_server.master_key", master_key) + monkeypatch.delenv("ANTHROPIC_AUTH_TOKEN", raising=False) + if proxy_api_key is None: + monkeypatch.delenv("ANTHROPIC_API_KEY", raising=False) + else: + monkeypatch.setenv("ANTHROPIC_API_KEY", proxy_api_key) + caller: Final = authenticated if authenticated is not None else UserAPIKeyAuth(api_key=self.VKEY) + + async def receive(): + return {"type": "http.request", "body": b"{}", "more_body": False} + + request = Request( + { + "type": "http", + "method": "POST", + "path": f"/anthropic/{self.ENDPOINT}", + "headers": headers, + "query_string": b"", + }, + receive=receive, + ) + + captured: dict = {} + + def fake_create_pass_through_route(**kwargs): + captured.update(kwargs) + return AsyncMock(return_value={"status": "success"}) + + module = "litellm.proxy.pass_through_endpoints.llm_passthrough_endpoints" + monkeypatch.setattr(f"{module}.passthrough_endpoint_router", PassthroughEndpointRouter(lambda: None)) + raised: HTTPException | None = None + with ( + mock.patch(f"{module}.create_pass_through_route", side_effect=fake_create_pass_through_route), + mock.patch(f"{module}.user_api_key_auth", new=AsyncMock(return_value=caller)), + ): + try: + await anthropic_proxy_route( + endpoint=self.ENDPOINT, + request=request, + fastapi_response=Response(), + user_api_key_dict=caller, + ) + except HTTPException as exc: + raised = exc + + if not captured: + return raised, None + upstream: Final = HttpPassThroughEndpointHelpers.forward_headers_from_request( + request_headers=dict(request.headers), + headers=dict(captured["custom_headers"] or {}), + forward_headers=captured.get("_forward_headers", False), + ) + return raised, upstream + + @staticmethod + def _blob(forwarded: dict) -> str: + return " ".join(f"{name}:{value}" for name, value in forwarded.items()) + + @pytest.mark.asyncio + async def test_authorization_bearer_virtual_key_is_rejected_not_forwarded(self, monkeypatch): + raised, forwarded = await self._run( + monkeypatch, + [(b"authorization", f"Bearer {self.VKEY}".encode()), (b"content-type", b"application/json")], + ) + assert forwarded is None, "credential-less request must never reach the upstream forwarder" + assert raised is not None and raised.status_code == 401 + assert "ANTHROPIC_API_KEY" in str(raised.detail) and "use_in_pass_through" in str(raised.detail) + + @pytest.mark.asyncio + async def test_x_api_key_virtual_key_is_rejected_not_forwarded(self, monkeypatch): + raised, forwarded = await self._run( + monkeypatch, + [(b"x-api-key", self.VKEY.encode()), (b"content-type", b"application/json")], + ) + assert forwarded is None, "a virtual key that authenticated via x-api-key must be stripped, not forwarded" + assert raised is not None and raised.status_code == 401 + + @pytest.mark.asyncio + async def test_x_litellm_api_key_virtual_key_is_rejected_not_forwarded(self, monkeypatch): + raised, forwarded = await self._run( + monkeypatch, + [(b"x-litellm-api-key", self.VKEY.encode()), (b"content-type", b"application/json")], + ) + assert forwarded is None, "credential-less request must never reach the upstream forwarder" + assert raised is not None and raised.status_code == 401 + + @pytest.mark.asyncio + async def test_master_key_in_authorization_is_rejected_not_forwarded(self, monkeypatch): + raised, forwarded = await self._run( + monkeypatch, + [(b"authorization", b"Bearer sk-master-1234"), (b"content-type", b"application/json")], + authenticated=UserAPIKeyAuth(api_key="sk-master-1234", user_role=LitellmUserRoles.PROXY_ADMIN), + ) + assert forwarded is None, "the master key must never reach Anthropic" + assert raised is not None and raised.status_code == 401 + + @pytest.mark.asyncio + async def test_byo_anthropic_oauth_token_still_forwards_without_virtual_key(self, monkeypatch): + raised, forwarded = await self._run( + monkeypatch, + [ + (b"x-litellm-api-key", self.VKEY.encode()), + (b"authorization", b"Bearer sk-ant-oat01-caller-oauth-token"), + (b"anthropic-version", b"2023-06-01"), + (b"content-type", b"application/json"), + ], + ) + assert raised is None + assert forwarded is not None + assert forwarded.get("authorization") == "Bearer sk-ant-oat01-caller-oauth-token" + assert forwarded.get("anthropic-version") == "2023-06-01" + assert "x-litellm-api-key" not in forwarded + assert self.VKEY not in self._blob(forwarded) + + @pytest.mark.asyncio + async def test_byo_x_api_key_still_forwards_without_virtual_key(self, monkeypatch): + raised, forwarded = await self._run( + monkeypatch, + [ + (b"authorization", f"Bearer {self.VKEY}".encode()), + (b"x-api-key", b"sk-ant-api03-caller-own-key"), + (b"content-type", b"application/json"), + ], + ) + assert raised is None + assert forwarded is not None + assert forwarded.get("x-api-key") == "sk-ant-api03-caller-own-key" + assert "authorization" not in forwarded + assert self.VKEY not in self._blob(forwarded) + + @pytest.mark.asyncio + async def test_custom_auth_caller_keeps_own_authorization_token(self, monkeypatch): + raised, forwarded = await self._run( + monkeypatch, + [(b"authorization", b"Bearer sk-ant-oat01-caller-oauth-token"), (b"content-type", b"application/json")], + authenticated=UserAPIKeyAuth(api_key=None), + master_key=None, + ) + assert raised is None + assert forwarded is not None + assert forwarded.get("authorization") == "Bearer sk-ant-oat01-caller-oauth-token" + + @pytest.mark.asyncio + @pytest.mark.parametrize( + "credential_header", + sorted(SpecialHeaders.litellm_credential_header_names() - {"authorization", "x-api-key", "x-litellm-api-key"}), + ) + async def test_every_non_anthropic_credential_header_is_dropped_by_name(self, monkeypatch, credential_header): + raised, forwarded = await self._run( + monkeypatch, + [ + (b"x-litellm-api-key", self.VKEY.encode()), + (b"x-api-key", b"sk-ant-api03-caller-own-key"), + (credential_header.encode(), b"some-distinct-caller-secret-value"), + (b"content-type", b"application/json"), + ], + ) + assert raised is None + assert forwarded is not None + assert forwarded.get("x-api-key") == "sk-ant-api03-caller-own-key" + assert credential_header not in forwarded + assert "x-litellm-api-key" not in forwarded + assert self.VKEY not in self._blob(forwarded) + assert "some-distinct-caller-secret-value" not in self._blob(forwarded) + + @pytest.mark.asyncio + async def test_virtual_key_in_operator_configured_header_is_stripped(self, monkeypatch): + with mock.patch.dict( # test-quality-ok: general_settings is the real proxy config surface for litellm_key_header_name; no injection seam exists on this route + "litellm.proxy.proxy_server.general_settings", + {"litellm_key_header_name": "x-company-key"}, + ): + raised, forwarded = await self._run( + monkeypatch, + [ + (b"x-company-key", f"Bearer {self.VKEY}".encode()), + (b"x-api-key", b"sk-ant-api03-caller-own-key"), + (b"content-type", b"application/json"), + ], + ) + assert raised is None + assert forwarded is not None + assert forwarded.get("x-api-key") == "sk-ant-api03-caller-own-key" + assert "x-company-key" not in forwarded + assert self.VKEY not in self._blob(forwarded) + + @pytest.mark.asyncio + async def test_proxy_credential_replaces_virtual_key_sent_as_bearer(self, monkeypatch): + raised, forwarded = await self._run( + monkeypatch, + [ + (b"authorization", f"Bearer {self.VKEY}".encode()), + (b"anthropic-version", b"2023-06-01"), + (b"content-type", b"application/json"), + ], + proxy_api_key=self.PROXY_KEY, + ) + assert raised is None + assert forwarded is not None + assert forwarded.get("x-api-key") == self.PROXY_KEY + assert "authorization" not in forwarded + assert forwarded.get("anthropic-version") == "2023-06-01" + assert self.VKEY not in self._blob(forwarded) + + @pytest.mark.asyncio + async def test_proxy_credential_replaces_virtual_key_sent_as_x_api_key(self, monkeypatch): + raised, forwarded = await self._run( + monkeypatch, + [(b"x-api-key", self.VKEY.encode()), (b"content-type", b"application/json")], + proxy_api_key=self.PROXY_KEY, + ) + assert raised is None + assert forwarded is not None + assert forwarded.get("x-api-key") == self.PROXY_KEY + assert self.VKEY not in self._blob(forwarded) + + @pytest.mark.asyncio + async def test_proxy_credential_wins_over_callers_own_x_api_key(self, monkeypatch): + raised, forwarded = await self._run( + monkeypatch, + [ + (b"x-litellm-api-key", self.VKEY.encode()), + (b"x-api-key", b"sk-ant-api03-caller-own-key"), + (b"content-type", b"application/json"), + ], + proxy_api_key=self.PROXY_KEY, + ) + assert raised is None + assert forwarded is not None + assert forwarded.get("x-api-key") == self.PROXY_KEY + assert "sk-ant-api03-caller-own-key" not in self._blob(forwarded) + + @pytest.mark.asyncio + async def test_x_pass_and_hop_by_hop_handling_is_unchanged(self, monkeypatch): + raised, forwarded = await self._run( + monkeypatch, + [ + (b"authorization", f"Bearer {self.VKEY}".encode()), + (b"x-pass-anthropic-beta", b"interleaved-thinking-2025-05-14"), + (b"x-pass-authorization", b"Bearer smuggled"), + (b"content-length", b"2"), + (b"host", b"proxy.internal"), + (b"accept-encoding", b"br"), + (b"user-agent", b"curl/8.7.1"), + ], + proxy_api_key=self.PROXY_KEY, + ) + assert raised is None + assert forwarded is not None + assert forwarded.get("anthropic-beta") == "interleaved-thinking-2025-05-14" + assert forwarded.get("user-agent") == "curl/8.7.1" + assert "authorization" not in forwarded + assert "content-length" not in forwarded + assert "host" not in forwarded + assert "accept-encoding" not in forwarded + + class TestVertexPassthroughDefaultLocationOnShortRoutes: PROJECT = "test-project" SHORT_ROUTE = "publishers/google/models/gemini-2.5-flash:generateContent"