PickleScan Security Bypass via Bad CRC in ZIP Archive

@@ -1,6 +1,6 @@
 [metadata]
 name = picklescan
-version = 0.0.30
+version = 0.0.31
 author = Matthieu Maitre
 author_email = mmaitre314@users.noreply.github.com
 description = Security scanner detecting Python Pickle files performing suspicious actions

@@ -1,7 +1,7 @@
 # A more forgiving implementation of zipfile.ZipFile
 # Modified from Python code at
 # https://github.com/python/cpython/blob/edb69578ed74ff04ab78ab953355faa343a7e0ee/Lib/zipfile/__init__.py#L1606
-# Changes: removed flag/password/filename checks to align better with PyTorch's zip decoding
+# Changes: removed flag/password/filename/CRC checks to align better with PyTorch's zip decoding
 
 import struct
 import zipfile
@@ -85,7 +85,12 @@ def open(self, name, mode="r", pwd=None, *, force_zip64=False):
             if fheader[_FH_EXTRA_FIELD_LENGTH]:
                 zef_file.read(fheader[_FH_EXTRA_FIELD_LENGTH])
 
-            return zipfile.ZipExtFile(zef_file, mode, zinfo, pwd, True)
+            zef = zipfile.ZipExtFile(zef_file, mode, zinfo, pwd, True)
+
+            # Disable CRC validation as PyTorch may not use it
+            zef._expected_crc = None
+
+            return zef
         except BaseException:
             zef_file.close()
             raise

@@ -116,7 +116,7 @@ def __str__(self) -> str:
         "open",
         "breakpoint",
     },  # Pickle versions 3, 4 have those function under 'builtins'
-    "aiohttp.client": "*",
+    "aiohttp": "*",
     "asyncio": "*",
     "bdb": "*",
     "commands": "*",  # Python 2 precursor to subprocess
@@ -134,7 +134,6 @@ def __str__(self) -> str:
     "ssl": "*",  # DNS exfiltration via ssl.get_server_certificate()
     "subprocess": "*",
     "sys": "*",
-    "asyncio.unix_events": {"_UnixSubprocessTransport._start"},
     "code": {"InteractiveInterpreter.runcode"},
     "cProfile": {"runctx", "run"},
     "doctest": {"debug_script"},
@@ -257,6 +256,7 @@ def _list_globals(data: IO[bytes], multiple_pickles=True) -> Set[Tuple[str, str]
             for op in pickletools.genops(data):
                 ops.append(op)
         except Exception as e:
+            _log.debug(f"Error parsing pickle: {e}", exc_info=True)
             parsing_pkl_error = str(e)
         last_byte = data.read(1)
         data.seek(-1, 1)
@@ -329,6 +329,11 @@ def _build_scan_result_from_raw_globals(
         g = Global(rg[0], rg[1], SafetyLevel.Dangerous)
         safe_filter = _safe_globals.get(g.module)
         unsafe_filter = _unsafe_globals.get(g.module)
+
+        # If the module as a whole is marked as dangerous, submodules are also dangerous
+        if unsafe_filter is None and "." in g.module and _unsafe_globals.get(g.module.split(".")[0]) == "*":
+            unsafe_filter = "*"
+
         if "unknown" in g.module or "unknown" in g.name:
             g.safety = SafetyLevel.Dangerous
             _log.warning("%s: %s import '%s %s' FOUND", file_id, g.safety.value, g.module, g.name)
@@ -348,11 +353,12 @@ def _build_scan_result_from_raw_globals(
 
 def scan_pickle_bytes(data: IO[bytes], file_id, multiple_pickles=True) -> ScanResult:
     """Disassemble a Pickle stream and report issues"""
+    _log.debug(f"scan_pickle_bytes({file_id})")
 
     try:
         raw_globals = _list_globals(data, multiple_pickles)
     except GenOpsError as e:
-        _log.error(f"ERROR: parsing pickle in {file_id}: {e}")
+        _log.error(f"ERROR: parsing pickle in {file_id}: {e}", exc_info=_log.isEnabledFor(logging.DEBUG))
         if e.globals is not None:
             return _build_scan_result_from_raw_globals(e.globals, file_id, scan_err=True)
         else:
@@ -365,6 +371,8 @@ def scan_pickle_bytes(data: IO[bytes], file_id, multiple_pickles=True) -> ScanRe
 
 # XXX: it appears there is not way to get the byte stream for a given file within the 7z archive and thus forcing us to unzip to disk before scanning
 def scan_7z_bytes(data: IO[bytes], file_id) -> ScanResult:
+    _log.debug(f"scan_7z_bytes({file_id})")
+
     try:
         import py7zr
     except ImportError:
@@ -387,6 +395,8 @@ def scan_7z_bytes(data: IO[bytes], file_id) -> ScanResult:
 
 
 def scan_zip_bytes(data: IO[bytes], file_id) -> ScanResult:
+    _log.debug(f"scan_zip_bytes({file_id})")
+
     result = ScanResult([])
 
     with RelaxedZipFile(data, "r") as zip:
@@ -415,6 +425,8 @@ def scan_zip_bytes(data: IO[bytes], file_id) -> ScanResult:
 
 
 def scan_numpy(data: IO[bytes], file_id) -> ScanResult:
+    _log.debug(f"scan_numpy({file_id})")
+
     # Delay import to avoid dependency on NumPy
     import numpy as np
 
@@ -445,6 +457,8 @@ def scan_numpy(data: IO[bytes], file_id) -> ScanResult:
 
 
 def scan_pytorch(data: IO[bytes], file_id) -> ScanResult:
+    _log.debug(f"scan_pytorch({file_id})")
+
     # new pytorch format
     if _is_zipfile(data):
         return scan_zip_bytes(data, file_id)
@@ -473,26 +487,34 @@ def scan_pytorch(data: IO[bytes], file_id) -> ScanResult:
 
 
 def scan_bytes(data: IO[bytes], file_id, file_ext: Optional[str] = None) -> ScanResult:
+    _log.debug(f"scan_bytes({file_id})")
+
     if file_ext is not None and file_ext in _pytorch_file_extensions:
         try:
             return scan_pytorch(data, file_id)
         except InvalidMagicError as e:
-            _log.error(f"ERROR: Invalid magic number for file {e}")
-            return ScanResult([], scan_err=True)
-    elif file_ext is not None and file_ext in _numpy_file_extensions:
+            _log.warning(
+                f"WARNING: Invalid PyTorch magic number for file {e}. Trying to scan as non-PyTorch file.",
+                exc_info=_log.isEnabledFor(logging.DEBUG),
+            )
+            data.seek(0)
+
+    if file_ext is not None and file_ext in _numpy_file_extensions:
         return scan_numpy(data, file_id)
+
+    is_zip = zipfile.is_zipfile(data)
+    data.seek(0)
+    if is_zip:
+        return scan_zip_bytes(data, file_id)
+    elif _is_7z_file(data):
+        return scan_7z_bytes(data, file_id)
     else:
-        is_zip = zipfile.is_zipfile(data)
-        data.seek(0)
-        if is_zip:
-            return scan_zip_bytes(data, file_id)
-        elif _is_7z_file(data):
-            return scan_7z_bytes(data, file_id)
-        else:
-            return scan_pickle_bytes(data, file_id)
+        return scan_pickle_bytes(data, file_id)
 
 
 def scan_huggingface_model(repo_id):
+    _log.debug(f"scan_huggingface_model({repo_id})")
+
     # List model files
     model = json.loads(_http_get(f"https://huggingface.co/api/models/{repo_id}").decode("utf-8"))
     file_names = [file_name for file_name in (sibling.get("rfilename") for sibling in model["siblings"]) if file_name is not None]
@@ -512,6 +534,8 @@ def scan_huggingface_model(repo_id):
 
 
 def scan_directory_path(path) -> ScanResult:
+    _log.debug(f"scan_directory_path({path})")
+
     scan_result = ScanResult([])
 
     for base_path, _, file_names in os.walk(path):
@@ -532,10 +556,14 @@ def scan_directory_path(path) -> ScanResult:
 
 
 def scan_file_path(path) -> ScanResult:
+    _log.debug(f"scan_file_path({path})")
+
     file_ext = os.path.splitext(path)[1]
     with open(path, "rb") as file:
         return scan_bytes(file, path, file_ext)
 
 
 def scan_url(url) -> ScanResult:
+    _log.debug(f"scan_url({url})")
+
     return scan_bytes(io.BytesIO(_http_get(url)), url)

@@ -438,6 +438,21 @@ def initialize_corrupt_zip_file_central_directory(path: str, file_name: str, dat
             f.write(modified_data)
 
 
+def initialize_corrupt_zip_file_crc(path: str, file_name: str, data: bytes):
+    if not os.path.exists(path):
+        with io.BytesIO() as buffer:
+            with zipfile.ZipFile(buffer, "w") as zip:
+                zip.writestr(file_name, data)
+            data = buffer.getbuffer().tobytes()
+
+        # Corrupt the data, leading to a CRC mismatch
+        modified_data = data.replace(b"print('456')", b"print('123')", 1)
+
+        # Write the corrupted content
+        with open(path, "wb") as f:
+            f.write(modified_data)
+
+
 def initialize_numpy_files():
     import numpy as np
 
@@ -687,6 +702,12 @@ def initialize_pickle_files():
         pickle.dumps(Malicious1(), protocol=4),
     )
 
+    initialize_corrupt_zip_file_crc(
+        f"{_root_path}/data2/malicious1_crc.zip",
+        "data.pkl",
+        pickle.dumps(Malicious1(), protocol=4),
+    )
+
     initialize_zip_file(
         f"{_root_path}/data/malicious1_wrong_ext.zip",
         "data.txt",  # Pickle file with a non-standard extension
@@ -744,6 +765,7 @@ def initialize_pickle_files():
     initialize_pickle_file_from_reduce("GHSA-9w88-8rmg-7g2p.pkl", reduce_GHSA_9w88_8rmg_7g2p)
     initialize_pickle_file_from_reduce("GHSA-49gj-c84q-6qm9.pkl", reduce_GHSA_49gj_c84q_6qm9)
     initialize_pickle_file_from_reduce("GHSA-q77w-mwjj-7mqx.pkl", reduce_GHSA_q77w_mwjj_7mqx)
+    initialize_pickle_file_from_reduce("GHSA-jgw4-cr84-mqxg.bin", reduce_GHSA_q77w_mwjj_7mqx)
 
 
 initialize_pickle_files()
@@ -1022,6 +1044,8 @@ def test_scan_file_path():
     assert_scan("GHSA-9w88-8rmg-7g2p.pkl", [Global("cProfile", "runctx", SafetyLevel.Dangerous)])
     assert_scan("GHSA-49gj-c84q-6qm9.pkl", [Global("cProfile", "run", SafetyLevel.Dangerous)])
     assert_scan("GHSA-q77w-mwjj-7mqx.pkl", [Global("asyncio.unix_events", "_UnixSubprocessTransport._start", SafetyLevel.Dangerous)])
+    assert_scan("GHSA-jgw4-cr84-mqxg.bin", [Global("asyncio.unix_events", "_UnixSubprocessTransport._start", SafetyLevel.Dangerous)])
+    assert_scan("malicious1_crc.zip", [Global("builtins", name="eval", safety=SafetyLevel.Dangerous)])
 
 
 def test_scan_file_path_npz():