Harden validate defaults for remote $ref resolution

Author: p1c2u

README.rst

@@ -54,14 +54,18 @@ Usage
 
 .. code-block:: python
 
-   validate(instance, schema, cls=OAS32Validator, **kwargs)
+   validate(instance, schema, cls=OAS32Validator, allow_remote_references=False, **kwargs)
 
 The first argument is always the value you want to validate.
 The second argument is always the OpenAPI schema object.
 The ``cls`` keyword argument is optional and defaults to ``OAS32Validator``.
 Use ``cls`` when you need a specific validator version/behavior.
 Common forwarded keyword arguments include ``registry`` (reference context)
 and ``format_checker`` (format validation behavior).
+By default, ``validate`` uses a local-only empty registry to avoid implicit
+remote ``$ref`` retrieval. To resolve external references, pass an explicit
+``registry``. Set ``allow_remote_references=True`` only if you explicitly
+accept jsonschema's default remote retrieval behavior.
 
 To validate an OpenAPI schema:
 

SECURITY.md

@@ -6,7 +6,9 @@ If you believe you have found a security vulnerability in the repository, please
 
 **Please do not report security vulnerabilities through public GitHub issues.**
 
-Instead, please report them directly to the repository maintainer.
+Instead, please report them directly to the repository maintainers or use GitHub's private vulnerability reporting flow
+(``Security Advisories``) to report security issues privately:
+https://docs.github.com/en/code-security/how-tos/report-and-fix-vulnerabilities/report-a-vulnerability/privately-reporting-a-security-vulnerability
 
 Please include the requested information listed below (as much as you can provide) to help us better understand the nature and scope of the possible issue:
 

docs/references.rst

@@ -4,6 +4,10 @@ References
 You can resolve JSON Schema references by passing registry.
 The ``validate(instance, schema, ...)`` shortcut resolves local references
 (``#/...``) against the provided ``schema`` mapping.
+By default, the shortcut uses a local-only empty registry and does not
+implicitly retrieve remote references.
+If needed, ``allow_remote_references=True`` enables jsonschema's default
+remote retrieval behavior.
 
 .. code-block:: python
 

docs/validation.rst

@@ -10,17 +10,24 @@ Validate
 
 .. code-block:: python
 
-   validate(instance, schema, cls=OAS32Validator, **kwargs)
+   validate(instance, schema, cls=OAS32Validator, allow_remote_references=False, **kwargs)
 
 The first argument is always the value you want to validate.
 The second argument is always the OpenAPI schema object.
 The ``cls`` keyword argument is optional and defaults to ``OAS32Validator``.
 Use ``cls`` when you need a specific validator version/behavior.
+The ``allow_remote_references`` keyword argument is optional and defaults to
+``False``.
 Common forwarded keyword arguments include:
 
-- ``registry`` for reference resolution context
+- ``registry`` for explicit external reference resolution context
 - ``format_checker`` to control format validation behavior
 
+By default, ``validate`` uses a local-only empty registry to avoid implicit
+remote ``$ref`` retrieval.
+Set ``allow_remote_references=True`` only if you explicitly accept
+jsonschema's default remote retrieval behavior.
+
 To validate an OpenAPI schema:
 
 .. code-block:: python

openapi_schema_validator/shortcuts.py

@@ -4,6 +4,7 @@
 
 from jsonschema.exceptions import best_match
 from jsonschema.protocols import Validator
+from referencing import Registry
 
 from openapi_schema_validator._dialects import OAS31_BASE_DIALECT_ID
 from openapi_schema_validator._dialects import OAS32_BASE_DIALECT_ID
@@ -16,6 +17,7 @@ def validate(
     schema: Mapping[str, Any],
     cls: type[Validator] = OAS32Validator,
     *args: Any,
+    allow_remote_references: bool = False,
     **kwargs: Any
 ) -> None:
     """
@@ -33,8 +35,13 @@ def validate(
             (``#/...``) are resolved against this mapping.
         cls: Validator class to use. Defaults to ``OAS32Validator``.
         *args: Positional arguments forwarded to ``cls`` constructor.
+        allow_remote_references: If ``True`` and no explicit ``registry`` is
+            provided, allow jsonschema's default remote reference retrieval
+            behavior.
         **kwargs: Keyword arguments forwarded to ``cls`` constructor
-            (for example ``registry`` and ``format_checker``).
+            (for example ``registry`` and ``format_checker``). If omitted,
+            a local-only empty ``Registry`` is used to avoid implicit remote
+            reference retrieval.
 
     Raises:
         jsonschema.exceptions.SchemaError: If ``schema`` is invalid.
@@ -53,8 +60,12 @@ def validate(
         check_openapi_schema(cls, schema_dict)
     else:
         cls.check_schema(schema_dict)
+    
+    validator_kwargs = kwargs.copy()
+    if not allow_remote_references:
+        validator_kwargs.setdefault("registry", Registry())
 
-    validator = cls(schema_dict, *args, **kwargs)
+    validator = cls(schema_dict, *args, **validator_kwargs)
     error = best_match(
         validator.evolve(schema=schema_dict).iter_errors(instance)
     )

tests/unit/test_shortcut.py

@@ -2,6 +2,8 @@
 from unittest.mock import patch
 
 import pytest
+from referencing import Registry
+from referencing import Resource
 
 from openapi_schema_validator import OAS32Validator
 from openapi_schema_validator import validate
@@ -56,3 +58,63 @@ def test_oas32_validate_does_not_fetch_remote_metaschemas(schema):
         validate({"email": "foo@bar.com"}, schema, cls=OAS32Validator)
 
     urlopen.assert_not_called()
+
+
+def test_validate_blocks_implicit_remote_http_references_by_default():
+    schema = {"$ref": "http://example.com/remote-schema.json"}
+
+    with patch("urllib.request.urlopen") as urlopen:
+        with pytest.raises(Exception, match="Unresolvable"):
+            validate({}, schema)
+
+    urlopen.assert_not_called()
+
+
+def test_validate_blocks_implicit_file_references_by_default():
+    schema = {"$ref": "file:///etc/hosts"}
+
+    with patch("urllib.request.urlopen") as urlopen:
+        with pytest.raises(Exception, match="Unresolvable"):
+            validate({}, schema)
+
+    urlopen.assert_not_called()
+
+
+def test_validate_local_references_still_work_by_default():
+    schema = {"$defs": {"Value": {"type": "integer"}}, "$ref": "#/$defs/Value"}
+
+    with patch("urllib.request.urlopen") as urlopen:
+        result = validate(1, schema)
+
+    assert result is None
+    urlopen.assert_not_called()
+
+
+def test_validate_honors_explicit_registry():
+    schema = {
+        "type": "object",
+        "properties": {"name": {"$ref": "urn:name-schema"}},
+    }
+    name_schema = Resource.from_contents(
+        {
+            "$schema": "https://json-schema.org/draft/2020-12/schema",
+            "type": "string",
+        }
+    )
+    registry = Registry().with_resources(
+        [("urn:name-schema", name_schema)],
+    )
+
+    result = validate({"name": "John"}, schema, registry=registry)
+
+    assert result is None
+
+
+def test_validate_can_allow_implicit_remote_references():
+    schema = {"$ref": "http://example.com/remote-schema.json"}
+
+    with patch("urllib.request.urlopen") as urlopen:
+        with pytest.raises(Exception):
+            validate({}, schema, allow_remote_references=True)
+
+    assert urlopen.called