🛡️ Sentinel: [CRITICAL] Fix Path Traversal and SafeStaticFiles AttributeError

.jules/sentinel.md

@@ -1,4 +1,8 @@
 ## 2025-02-21 - Path Traversal in Mix Endpoint API Parameter
 **Vulnerability:** The `/projects/{project_id}/mix` API endpoint in `src/audioformation/server/routes.py` accepted a `music` parameter (meant to specify a filename within the `05_MUSIC/generated` directory) but directly passed it to `mix_project` without sanitization. This allowed directory traversal payloads like `../../../etc/passwd` to be used for background music resolution.
 **Learning:** Even internal API inputs that map strictly to filenames inside an expected directory must be sanitized. A simple check for file existence (`if not bg_music_path.exists():`) is insufficient as it confirms existence but allows looking outside the bounded directory.
-**Prevention:** Always use established sanitization helpers (like `sanitize_filename`) or bound checks (like `validate_path_within`) for any user-supplied string that forms part of a filesystem path. Ensure bypass parameters like `FORCE_NO_MUSIC` are handled before and mutually exclusively from sanitization.
+**Prevention:** Always use established sanitization helpers (like `sanitize_filename`) or bound checks (like `validate_path_within`) for any user-supplied string that forms part of a filesystem path. Ensure bypass parameters like `FORCE_NO_MUSIC` are handled before and mutually exclusively from sanitization.
+## 2025-02-28 - Path validation and `Path` object handling
+**Vulnerability:** String-based path validation inside `validate_path_within` was prone to symlink and path traversal bypasses. Also, `.lower()` was incorrectly used on `Path` objects in `SafeStaticFiles` leading to `AttributeError` 500 errors.
+**Learning:** `Path.resolve().is_relative_to()` is the safest and most canonical path traversal defense in Python. Additionally, `Path` objects in `pathlib` do not inherit string methods like `.lower()`.
+**Prevention:** Always use `is_relative_to` after path resolution for boundary checks, and always apply string manipulations to path strings *before* instantiation of a `Path` object.

src/audioformation/server/app.py

@@ -24,7 +24,7 @@ class SafeStaticFiles(StaticFiles):
 
     async def get_response(self, path: str, scope) -> Response:
         # Normalize path for check
-        p = Path(path).lower()
+        p = Path(path.lower())
         if "00_config" in p.parts or p.name.startswith(".env") or ".git" in p.parts:
             raise HTTPException(
                 status_code=403, detail="Access denied to sensitive resource"

src/audioformation/utils/security.py

@@ -68,20 +68,10 @@
     This prevents path traversal and symlink bypass attacks.
     """
     try:
-        # Resolve to absolute paths first
-        abs_path = os.path.abspath(str(path))
-        abs_root = os.path.abspath(str(root))
-
-        # On Windows, abspath can have different casing for the drive letter.
-        # We normalize to lowercase for the preliminary string check.
-        if abs_path.lower().startswith(abs_root.lower()):
-            # String check passed, now do the rigorous resolution check
-            resolved_root = root.resolve()
-            resolved_path = path.resolve()
-            return resolved_path.is_relative_to(resolved_root)
-
-        return False
-    except (ValueError, RuntimeError, OSError):
+        resolved_root = root.resolve()
+        resolved_path = path.resolve()
+        return resolved_path.is_relative_to(resolved_root)
+    except (TypeError, ValueError, RuntimeError, AttributeError, OSError):
         return False