fix(proxy): match a wildcard's own row pattern before asking its server

The /v1/model/info pre-filter matched a key's models against both the
deployment name and its litellm_params.model, so a key granted
hosted_vllm/* still asked a vllm-secure/* -> hosted_vllm/* server whose
rows it can never see. It now matches only the pattern expansion names
the rows with, picked by one shared helper.

litellm/proxy/auth/model_checks.py

@@ -564,17 +564,21 @@ def _names_no_deployment_serves_by_name(
     return tuple(name for name in names if not deployed_by_name(name, team_id))
 
 
-def _matches_a_pattern(name: str, patterns: Sequence[str]) -> bool:
-    return any(_wildcard_name_regex(pattern).match(name) for pattern in patterns)
+def _wildcard_pattern_to_expand(model_name: str, litellm_model: str) -> str | None:
+    if not _check_wildcard_routing(model_name):
+        return None
+    if "/" in model_name or model_name == "*":
+        return model_name
+    if _check_wildcard_routing(litellm_model):
+        return litellm_model
+    return model_name
 
 
-def _can_expand_to_an_allowed_name(
-    model_name: str, litellm_model: str, allowed_model_names: Collection[str] | None
-) -> bool:
-    if allowed_model_names is None or not _check_wildcard_routing(model_name):
+def _can_expand_to_an_allowed_name(wildcard_pattern: str, allowed_model_names: Collection[str] | None) -> bool:
+    if allowed_model_names is None:
         return True
-    patterns: Final = tuple(pattern for pattern in (model_name, litellm_model) if _check_wildcard_routing(pattern))
-    return any(_matches_a_pattern(name, patterns) for name in allowed_model_names)
+    pattern: Final = _wildcard_name_regex(wildcard_pattern)
+    return any(pattern.match(name) for name in allowed_model_names)
 
 
 def expand_wildcard_deployments_for_model_info(
@@ -597,21 +601,11 @@ def expand_wildcard_deployments_for_model_info(
         raw_params = deployment.get("litellm_params")
         litellm_params_dict: dict[str, object] = raw_params if isinstance(raw_params, dict) else {}
         litellm_model = str(litellm_params_dict.get("model") or "")
-        if not _can_expand_to_an_allowed_name(model_name, litellm_model, allowed_model_names):
-            continue
-
-        # Determine the wildcard pattern to expand.
-        # Branch order matters: only fall to litellm_model when model_name is
-        # also a wildcard, so a concrete model_name is never overwritten.
-        if _check_wildcard_routing(model_name) and ("/" in model_name or model_name == "*"):
-            wildcard_pattern = model_name
-        elif _check_wildcard_routing(model_name) and _check_wildcard_routing(litellm_model):
-            wildcard_pattern = litellm_model
-        elif _check_wildcard_routing(model_name):
-            wildcard_pattern = model_name
-        else:
+        if (wildcard_pattern := _wildcard_pattern_to_expand(model_name, litellm_model)) is None:
             expanded.append(deployment)
             continue
+        if not _can_expand_to_an_allowed_name(wildcard_pattern, allowed_model_names):
+            continue
 
         try:
             litellm_params = LiteLLM_Params.model_validate(litellm_params_dict) if litellm_params_dict else None

tests/unit/proxy/proxy_server/test_routes_model_info.py

@@ -588,6 +588,7 @@ def test_v1_model_info_asks_a_wildcard_server_only_for_keys_that_can_call_it(
     from litellm.proxy._types import LitellmUserRoles
 
     wildcard_server: Final = f"http://vllm-{request.node.callspec.id}.test:8000/v1"
+    other_alias_server: Final = f"http://vllm-secure-{request.node.callspec.id}.test:8000/v1"
     monkeypatch.setattr(litellm, "check_provider_endpoint", True)
     router: Final = litellm.Router(
         model_list=[
@@ -596,6 +597,10 @@ def test_v1_model_info_asks_a_wildcard_server_only_for_keys_that_can_call_it(
                 "litellm_params": {"model": "hosted_vllm/qwen2.5", "api_base": "http://vllm-open.test:8000/v1"},
             },
             {"model_name": "hosted_vllm/*", "litellm_params": {"model": "hosted_vllm/*", "api_base": wildcard_server}},
+            {
+                "model_name": "vllm-secure/*",
+                "litellm_params": {"model": "hosted_vllm/*", "api_base": other_alias_server},
+            },
         ]
     )
     monkeypatch.setattr(proxy_server, "llm_router", router)
@@ -607,14 +612,18 @@ def test_v1_model_info_asks_a_wildcard_server_only_for_keys_that_can_call_it(
         listing: Final = upstream.get(f"{wildcard_server}/models").mock(
             return_value=httpx.Response(200, json={"data": [{"id": "qwen2.5"}]})
         )
+        other_alias_listing: Final = upstream.get(f"{other_alias_server}/models").mock(
+            return_value=httpx.Response(200, json={"data": [{"id": "qwen2.5"}]})
+        )
         with auth_as(role=LitellmUserRoles.INTERNAL_USER, models=key_models):
             response: Final = client.get("/v1/model/info")
 
     assert response.status_code == 200, response.text
-    assert ([model["model_name"] for model in response.json()["data"]], listing.call_count) == (
-        expected_names,
-        expected_listings,
-    )
+    assert (
+        [model["model_name"] for model in response.json()["data"]],
+        listing.call_count,
+        other_alias_listing.call_count,
+    ) == (expected_names, expected_listings, 0)
 
 
 # ---------------------------------------------------------------------------