提出 #798731: langflow-ai langflow 1.8.4 Command Injection
| タイトル | langflow-ai langflow 1.8.4 Command Injection |
|---|---|
| 説明 | The `CodeParser.parse_callable_details()` method extracts the annotation string via `ast.unparse()` when parsing function return type annotations and directly passes it to `eval()` for execution. The execution environment (`eval_env`) for `eval()` is constructed by `construct_eval_env()`, but `__builtins__ = {}` is never set. The Python specification stipulates: When the globals dict passed to eval() does not contain the __builtins__ key, the Python interpreter automatically adds the full builtins module. As a result, all built-in functions such as __import__, open, exec, and eval can be used in the eval environment. We can embed malicious expressions (such as __import__('os').popen('cmd').read()) in the return type annotations of functions within custom component code. When the CodeParser processes this code, the eval() function will execute the expression directly. |
| ソース | ⚠️ https:/ |
| ユーザー | limshow (UID 92836) |
| 送信 | 2026年04月07日 17:09 (6 月 ago) |
| モデレーション | 2026年05月02日 18:06 (25 days later) |
| ステータス | 承諾済み |
| VulDBエントリ | 360857 [langflow-ai langflow 迄 1.8.4 Full Builtins code_parser.py CodeParser.parse_callable_details 特権昇格] |
| ポイント | 20 |