fix(openai): use SSRF-safe transport for image token counting (#36819)

Author: ccurme

libs/partners/openai/langchain_openai/chat_models/base.py

@@ -141,6 +141,7 @@
 from langchain_openai.data._profiles import _PROFILES
 
 if TYPE_CHECKING:
+    import httpx
     from langchain_core.language_models import ModelProfile
     from openai.types.responses import Response
 
@@ -150,6 +151,20 @@
 # https://www.python-httpx.org/advanced/ssl/#configuring-client-instances
 global_ssl_context = ssl.create_default_context(cafile=certifi.where())
 
+_ssrf_client: httpx.Client | None = None
+
+
+def _get_ssrf_safe_client() -> httpx.Client:
+    global _ssrf_client
+    if _ssrf_client is None:
+        from langchain_core._security._transport import ssrf_safe_client
+
+        _ssrf_client = ssrf_safe_client(
+            verify=global_ssl_context, follow_redirects=False
+        )
+    return _ssrf_client
+
+
 _MODEL_PROFILES = cast(ModelProfileRegistry, _PROFILES)
 
 
@@ -3638,28 +3653,7 @@ def _url_to_size(image_source: str) -> tuple[int, int] | None:
         )
         return None
     if _is_url(image_source):
-        try:
-            import httpx
-        except ImportError:
-            logger.info(
-                "Unable to count image tokens. To count image tokens please install "
-                "`pip install -U httpx`."
-            )
-            return None
-
-        # Validate URL for SSRF protection
-        try:
-            from langchain_core._security._ssrf_protection import validate_safe_url
-
-            validate_safe_url(image_source, allow_private=False, allow_http=True)
-        except ImportError:
-            logger.warning(
-                "SSRF protection not available. "
-                "Update langchain-core to get SSRF protection."
-            )
-        except ValueError as e:
-            logger.warning("Image URL failed SSRF validation: %s", e)
-            return None
+        import httpx
 
         # Set reasonable limits to prevent resource exhaustion
         # Timeout prevents indefinite hangs on slow/malicious servers
@@ -3668,10 +3662,7 @@ def _url_to_size(image_source: str) -> tuple[int, int] | None:
         max_size = 50 * 1024 * 1024  # 50 MB
 
         try:
-            response = httpx.get(
-                image_source,
-                timeout=timeout,
-            )
+            response = _get_ssrf_safe_client().get(image_source, timeout=timeout)
             response.raise_for_status()
 
             # Check response size before loading into memory

libs/partners/openai/pyproject.toml

@@ -23,7 +23,7 @@ classifiers = [
 version = "1.1.13"
 requires-python = ">=3.10.0,<4.0.0"
 dependencies = [
-    "langchain-core>=1.2.29,<2.0.0",
+    "langchain-core>=1.2.31,<2.0.0",
     "openai>=2.26.0,<3.0.0",
     "tiktoken>=0.7.0,<1.0.0",
 ]