python-kwargs-setattr-security
When reviewing or fixing Python code that uses setattr() with user-controlled kwargs to configure C++ extension objects (SessionOptions, RunOptions, etc.) in ONNX Runtime. Use this to apply the allowlist pattern that prevents arbitrary file writes and other attacks via reflected property access.
- License 许可
- MIT
- 最近更新
- 2026-11-12
- Stars
- ★ 2.2w
- Forks
- 4.4k
01 前言 Front Matter
- name
- python-kwargs-setattr-security
- description
- When reviewing or fixing Python code that uses setattr() with user-controlled kwargs to configure C++ extension objects (SessionOptions, RunOptions, etc.) in ONNX Runtime. Use this to apply the allowlist pattern that prevents arbitrary file writes and other attacks via reflected property access.
03 SKILL.md 正文
Problem Pattern
Using hasattr(obj, k) / setattr(obj, k, v) with user-controlled kwargs is insecure. The hasattr check is NOT a security guard — it returns True for ALL exposed properties including dangerous ones.
# INSECURE — do not use
for k, v in kwargs.items():
if hasattr(options, k):
setattr(options, k, v)
Fix: Explicit Allowlist
Define a module-level frozenset of safe attribute names. Raise RuntimeError for known-but-blocked attrs; silently ignore unknown keys.
# Define at module level, before the class
_ALLOWED_SESSION_OPTIONS = frozenset({
"enable_cpu_mem_arena",
"enable_mem_pattern",
# ... only explicitly reviewed safe attrs
})
# In the method
for k, v in kwargs.items():
if k in _ALLOWED_SESSION_OPTIONS:
setattr(options, k, v)
elif hasattr(options, k): # reuse the existing instance, don't create new
raise RuntimeError(
f"SessionOptions attribute '{k}' is not permitted via the backend API. "
f"Allowed attributes: {', '.join(sorted(_ALLOWED_SESSION_OPTIONS))}"
)
# else: silently ignore (may be kwargs for a different config object)
Key Rules
- Use the existing object in
hasattr(options, k)— neverhasattr(ClassName(), k)(creates throwaway C++ objects per iteration) - RuntimeError is the ORT convention for API misuse errors (not ValueError)
- Silent ignore for one path is OK when kwargs are forwarded to both paths:
run_model()passes the same kwargs dict to bothprepare()(validates SessionOptions) andrep.run()(validates RunOptions). A RunOptions kwarg unknown to SessionOptions is silently ignored byprepare()— this is correct becauserep.run()will validate it. Only raise RuntimeError when the attr exists on the target object but is blocked. - Frozenset constant naming:
_ALLOWED_<CLASSNAME>— ALL_CAPS, Google Style - No type annotations on module-level constants (ORT Python convention)
Dangerous SessionOptions Properties (never allowlist)
optimized_model_filepath— triggers Model::Save(), overwrites arbitrary filesprofile_file_prefix+enable_profiling— writes profiling JSON to arbitrary pathregister_custom_ops_library— loads arbitrary shared libraries (method, not property)
Files in ONNX Runtime
onnxruntime/python/backend/backend.py—_ALLOWED_SESSION_OPTIONSonnxruntime/python/backend/backend_rep.py—_ALLOWED_RUN_OPTIONS- Tests:
onnxruntime/test/python/onnxruntime_test_python_backend.py—TestBackendKwargsAllowlist
04 同仓库更多 Skills
workspace-estimation-shared-header
Use when editing a boundary-safe shared header meant to be included by both in-tree and shared-provider/plugin-bridge code (e.g. include/onnxruntime/core/framework/workspace_requirement.h, or any future header supporting OrtKernelWorkspaceEstimateFunc / DeclareWorkspaceRequirements per issue #29775's Phase-A workspace-estimation roadmap), or when a workspace-estimation math helper needs to be callable from a future plugin EP DLL. Covers a Node forward-declaration ODR trap and the math-helper vs. graph-parsing-wrapper reuse boundary discovered while implementing PR #29811 (MatMulNBits pilot).
webgpu-local-testing
Build and run ONNX Runtime WebGPU provider tests on Linux WITHOUT a real GPU, using a software Vulkan adapter (Mesa lavapipe). Use when you need to exercise WebGPU EP kernels off-Mac — the Linux webgpu CI leg is build-only, so software Vulkan is how you actually run WebGPU correctness tests locally. SCOPE - lavapipe only validates host-side enforce/shape bugs and MatMul-free kernels; any graph containing MatMul (including the expanded-Attention node tests) crashes lavapipe and runs ONLY on macOS-arm64 Metal, which is the source of truth for those. Covers install (dnf on Azure Linux), the --use_webgpu build flag, the onnxruntime_provider_test target, VK_ICD_FILENAMES, and the lavapipe MatMul crash gotcha.
ort-transformers-gpu-pytest
Run the ONNX Runtime transformers Python tests (onnxruntime/test/python/transformers) against a GPU wheel, and confirm real cuDNN/flash SDPA dispatch. Use when a transformers pytest fails with ModuleNotFoundError onnxruntime.capi, when torch's bundled CUDA/cuDNN libs shadow the ORT-built ones (wrong-version dispatch or load errors), or when you must PROVE a test exercised the cuDNN SDPA decode tier instead of silently skipping/falling back to MATH.
ort-test
Run ONNX Runtime tests. Use this skill when asked to run tests, debug test failures, or find and execute specific test cases in ONNX Runtime.
ort-release-notes
Draft ONNX Runtime release notes using preset configurations for full ORT or scoped component releases. Use when generating highlights with PR links, compiling human contributor acknowledgments from compile_contributors.py output, and applying preset path filtering.
ort-lint
Lint and format ONNX Runtime code. Use this skill when asked to lint, format, or check code style for C++ or Python files in ONNX Runtime.