fix(proxy): trigger async_pre_call_hook on POST /v1/files uploads

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/proxy/openai_files_endpoints/files_endpoints.py

@@ -526,6 +526,18 @@ async def create_file(
             proxy_config=proxy_config,
         )
 
+        hook_data: Final[dict] = {
+            **data,
+            "purpose": purpose,
+            "file": {"filename": file.filename, "content_type": file.content_type, "size": file.size},
+        }
+        hooked_data: Final[dict] = await proxy_logging_obj.pre_call_hook(
+            user_api_key_dict=user_api_key_dict,
+            data=hook_data,
+            call_type="acreate_file",
+        )
+        data = {k: v for k, v in hooked_data.items() if k not in ("purpose", "file")}
+
         # /v1/files stores its proxy metadata under litellm_metadata, not metadata
         request_metadata: Final = data.get("metadata") or data.get("litellm_metadata") or EMPTY_MAPPING
         scan_result: Final = await _scan_batch_upload(

litellm/types/utils.py

@@ -543,6 +543,8 @@ CallTypesLiteral = Literal[
     "_arealtime",
     "create_batch",
     "acreate_batch",
+    "create_file",
+    "acreate_file",
     "pass_through_endpoint",
     "allm_passthrough_route",
     "anthropic_messages",

tests/test_litellm/proxy/openai_files_endpoint/test_files_endpoint.py

@@ -4476,3 +4476,82 @@ def test_scoped_list_files_still_resolves_deployment_credentials(
     provider_list.assert_awaited_once()
     assert provider_list.await_args.kwargs["custom_llm_provider"] == "openai"
     assert provider_list.await_args.kwargs["api_key"] == "openai_api_key"
+
+
+def _post_user_data_file() -> httpx.Response:
+    return client.post(
+        "/v1/files",
+        files={"file": ("labels.jsonl", b'{"label": "restricted"}', "application/json")},
+        data={"purpose": "user_data"},
+        headers={"Authorization": "Bearer test-key"},
+    )
+
+
+def _setup_create_file_over_pre_call_hook(mocker, monkeypatch, llm_router, hook):
+    setup_proxy_logging_object(monkeypatch, llm_router)
+    monkeypatch.setattr("litellm.proxy.proxy_server.llm_router", llm_router)
+    monkeypatch.setattr(litellm, "callbacks", [hook])
+    monkeypatch.setattr(
+        "litellm.proxy.openai_files_endpoints.files_endpoints.files_config",
+        [{"custom_llm_provider": "openai", "api_key": "sk-test"}],
+    )
+    return mocker.patch(
+        "litellm.proxy.openai_files_endpoints.files_endpoints.litellm.acreate_file",
+        new=mocker.AsyncMock(
+            return_value=OpenAIFileObject(
+                id="file-hooked",
+                object="file",
+                bytes=23,
+                created_at=1234567890,
+                filename="labels.jsonl",
+                purpose="user_data",
+                status="uploaded",
+            )
+        ),
+    )
+
+
+def test_create_file_triggers_async_pre_call_hook(mocker: MockerFixture, monkeypatch, llm_router: Router):
+    """`POST /v1/files` must run `async_pre_call_hook` so a hook can inspect the upload
+    before it reaches the provider (LIT-5916)."""
+    from litellm.integrations.custom_logger import CustomLogger
+
+    recorded: dict = {}
+
+    class RecordingHook(CustomLogger):
+        async def async_pre_call_hook(self, user_api_key_dict, cache, data, call_type):
+            recorded["call_type"] = call_type
+            recorded["purpose"] = data.get("purpose")
+            recorded["file"] = data.get("file")
+
+    provider_create = _setup_create_file_over_pre_call_hook(mocker, monkeypatch, llm_router, RecordingHook())
+
+    response = _post_user_data_file()
+
+    assert response.status_code == 200, response.text
+    assert recorded["call_type"] == "acreate_file"
+    assert recorded["purpose"] == "user_data"
+    assert recorded["file"]["filename"] == "labels.jsonl"
+    provider_create.assert_awaited_once()
+    forwarded = provider_create.await_args.kwargs
+    assert forwarded["purpose"] == "user_data"
+    assert forwarded["file"][0] == "labels.jsonl"
+
+
+def test_create_file_async_pre_call_hook_rejection_blocks_upload(
+    mocker: MockerFixture, monkeypatch, llm_router: Router
+):
+    """A hook rejecting the upload must 400 before the file reaches the provider."""
+    from litellm.integrations.custom_logger import CustomLogger
+
+    class RejectingHook(CustomLogger):
+        async def async_pre_call_hook(self, user_api_key_dict, cache, data, call_type):
+            return "file upload not allowed"
+
+    provider_create = _setup_create_file_over_pre_call_hook(mocker, monkeypatch, llm_router, RejectingHook())
+
+    response = _post_user_data_file()
+
+    assert response.status_code == 400, response.text
+    assert "file upload not allowed" in response.text
+    provider_create.assert_not_awaited()