fix(auth): per-param allow must continue, not return early

A pre-existing logic bug in ``_check_banned_params``: when the
deployment-level ``configurable_clientside_auth_params`` permitted one
banned field, the loop ``return``-ed on the first match instead of
``continue``-ing, so any other banned param later in the same body or
metadata dict was never checked. This PR's metadata walk multiplies the
surface where that bypass matters — a body pairing an allowed
``api_base`` with an observability credential like ``langfuse_host``
would silently pass.

Proxy-wide ``allow_client_side_credentials`` keeps ``return`` (it's a
global opt-in for every banned param). The per-param branch becomes
``continue`` so only the one explicitly-permitted field is skipped.

Adds a regression test that exercises the api_base + langfuse_host pair.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

litellm/proxy/auth/auth_utils.py

@@ -279,6 +279,8 @@ def _check_banned_params(
         if param not in body:
             continue
         if general_settings.get("allow_client_side_credentials") is True:
+            # Proxy-wide opt-in: every banned param is permitted, exit
+            # entirely so the rest of the loop doesn't waste work.
             return
         if (
             _allow_model_level_clientside_configurable_parameters(
@@ -289,7 +291,12 @@ def _check_banned_params(
             )
             is True
         ):
-            return
+            # Per-param opt-in: only THIS param is permitted by the
+            # deployment's ``configurable_clientside_auth_params``. Skip
+            # to the next banned param so a body that pairs an allowed
+            # ``api_base`` with an unallowed ``langfuse_host`` is still
+            # rejected for the second field.
+            continue
         raise ValueError(
             f"Rejected Request: {param} is not allowed in request body. "
             "Clientside passthrough requires explicit admin opt-in via "

tests/test_litellm/proxy/auth/test_auth_utils.py

@@ -1455,6 +1455,36 @@ class TestObservabilityCallbackBans:
         )
 
 
+def test_model_level_allow_does_not_skip_subsequent_banned_params(monkeypatch):
+    """Greptile P1: ``_check_banned_params`` previously ``return``-ed when a
+    deployment's ``configurable_clientside_auth_params`` permitted one
+    banned field, exiting before any later banned field in the same body
+    was checked. The metadata walk this PR adds multiplies the surface
+    where that bypass matters: a body pairing a model-level-allowed
+    ``api_base`` with an observability credential like ``langfuse_host``
+    must still reject on the second field, not silently pass."""
+    from litellm.proxy.auth import auth_utils
+
+    monkeypatch.setattr(
+        auth_utils,
+        "_allow_model_level_clientside_configurable_parameters",
+        lambda model, param, request_body_value, llm_router: param == "api_base",
+    )
+
+    with pytest.raises(ValueError) as exc:
+        is_request_body_safe(
+            request_body={
+                "model": "gpt-4",
+                "api_base": "https://allowed-by-deployment.example",
+                "langfuse_host": "https://attacker.example",
+            },
+            general_settings={},
+            llm_router=None,
+            model="gpt-4",
+        )
+    assert "langfuse_host" in str(exc.value)
+
+
 def test_observability_ban_covers_canonical_supported_callback_params():
     """Guard test: every entry in the canonical
     ``_supported_callback_params`` allow-list must end up either banned by