[3.8] gh-121285: Remove backtracking when parsing tarfile headers (GH-121286) (#123642)

* Remove backtracking when parsing tarfile headers
* Rewrite PAX header parsing to be stricter
* Optimize parsing of GNU extended sparse headers v0.0

(cherry picked from commit 34ddb64)

Co-authored-by: Seth Michael Larson <[email protected]>
Co-authored-by: Kirill Podoprigora <[email protected]>
Co-authored-by: Gregory P. Smith <[email protected]>

Author: 3 people

Lib/tarfile.py

@@ -840,6 +840,9 @@ def data_filter(member, dest_path):
 # Sentinel for replace() defaults, meaning "don't change the attribute"
 _KEEP = object()
 
+# Header length is digits followed by a space.
+_header_length_prefix_re = re.compile(br"([0-9]{1,20}) ")
+
 class TarInfo(object):
     """Informational class which holds the details about an
        archive member given by a tar header block.
@@ -1390,59 +1393,76 @@ def _proc_pax(self, tarfile):
         else:
             pax_headers = tarfile.pax_headers.copy()
 
-        # Check if the pax header contains a hdrcharset field. This tells us
-        # the encoding of the path, linkpath, uname and gname fields. Normally,
-        # these fields are UTF-8 encoded but since POSIX.1-2008 tar
-        # implementations are allowed to store them as raw binary strings if
-        # the translation to UTF-8 fails.
-        match = re.search(br"\d+ hdrcharset=([^\n]+)\n", buf)
-        if match is not None:
-            pax_headers["hdrcharset"] = match.group(1).decode("utf-8")
-
-        # For the time being, we don't care about anything other than "BINARY".
-        # The only other value that is currently allowed by the standard is
-        # "ISO-IR 10646 2000 UTF-8" in other words UTF-8.
-        hdrcharset = pax_headers.get("hdrcharset")
-        if hdrcharset == "BINARY":
-            encoding = tarfile.encoding
-        else:
-            encoding = "utf-8"
-
         # Parse pax header information. A record looks like that:
         # "%d %s=%s\n" % (length, keyword, value). length is the size
         # of the complete record including the length field itself and
-        # the newline. keyword and value are both UTF-8 encoded strings.
-        regex = re.compile(br"(\d+) ([^=]+)=")
+        # the newline.
         pos = 0
-        while True:
-            match = regex.match(buf, pos)
-            if not match:
-                break
+        encoding = None
+        raw_headers = []
+        while len(buf) > pos and buf[pos] != 0x00:
+            if not (match := _header_length_prefix_re.match(buf, pos)):
+                raise InvalidHeaderError("invalid header")
+            try:
+                length = int(match.group(1))
+            except ValueError:
+                raise InvalidHeaderError("invalid header")
+            # Headers must be at least 5 bytes, shortest being '5 x=\n'.
+            # Value is allowed to be empty.
+            if length < 5:
+                raise InvalidHeaderError("invalid header")
+            if pos + length > len(buf):
+                raise InvalidHeaderError("invalid header")
 
-            length, keyword = match.groups()
-            length = int(length)
-            if length == 0:
+            header_value_end_offset = match.start(1) + length - 1  # Last byte of the header
+            keyword_and_value = buf[match.end(1) + 1:header_value_end_offset]
+            raw_keyword, equals, raw_value = keyword_and_value.partition(b"=")
+
+            # Check the framing of the header. The last character must be '\n' (0x0A)
+            if not raw_keyword or equals != b"=" or buf[header_value_end_offset] != 0x0A:
                 raise InvalidHeaderError("invalid header")
-            value = buf[match.end(2) + 1:match.start(1) + length - 1]
+            raw_headers.append((length, raw_keyword, raw_value))
+
+            # Check if the pax header contains a hdrcharset field. This tells us
+            # the encoding of the path, linkpath, uname and gname fields. Normally,
+            # these fields are UTF-8 encoded but since POSIX.1-2008 tar
+            # implementations are allowed to store them as raw binary strings if
+            # the translation to UTF-8 fails. For the time being, we don't care about
+            # anything other than "BINARY". The only other value that is currently
+            # allowed by the standard is "ISO-IR 10646 2000 UTF-8" in other words UTF-8.
+            # Note that we only follow the initial 'hdrcharset' setting to preserve
+            # the initial behavior of the 'tarfile' module.
+            if raw_keyword == b"hdrcharset" and encoding is None:
+                if raw_value == b"BINARY":
+                    encoding = tarfile.encoding
+                else:  # This branch ensures only the first 'hdrcharset' header is used.
+                    encoding = "utf-8"
+
+            pos += length
 
+        # If no explicit hdrcharset is set, we use UTF-8 as a default.
+        if encoding is None:
+            encoding = "utf-8"
+
+        # After parsing the raw headers we can decode them to text.
+        for length, raw_keyword, raw_value in raw_headers:
             # Normally, we could just use "utf-8" as the encoding and "strict"
             # as the error handler, but we better not take the risk. For
             # example, GNU tar <= 1.23 is known to store filenames it cannot
             # translate to UTF-8 as raw strings (unfortunately without a
             # hdrcharset=BINARY header).
             # We first try the strict standard encoding, and if that fails we
             # fall back on the user's encoding and error handler.
-            keyword = self._decode_pax_field(keyword, "utf-8", "utf-8",
+            keyword = self._decode_pax_field(raw_keyword, "utf-8", "utf-8",
                     tarfile.errors)
             if keyword in PAX_NAME_FIELDS:
-                value = self._decode_pax_field(value, encoding, tarfile.encoding,
+                value = self._decode_pax_field(raw_value, encoding, tarfile.encoding,
                         tarfile.errors)
             else:
-                value = self._decode_pax_field(value, "utf-8", "utf-8",
+                value = self._decode_pax_field(raw_value, "utf-8", "utf-8",
                         tarfile.errors)
 
             pax_headers[keyword] = value
-            pos += length
 
         # Fetch the next header.
         try:
@@ -1457,7 +1477,7 @@ def _proc_pax(self, tarfile):
 
         elif "GNU.sparse.size" in pax_headers:
             # GNU extended sparse format version 0.0.
-            self._proc_gnusparse_00(next, pax_headers, buf)
+            self._proc_gnusparse_00(next, raw_headers)
 
         elif pax_headers.get("GNU.sparse.major") == "1" and pax_headers.get("GNU.sparse.minor") == "0":
             # GNU extended sparse format version 1.0.
@@ -1479,15 +1499,24 @@ def _proc_pax(self, tarfile):
 
         return next
 
-    def _proc_gnusparse_00(self, next, pax_headers, buf):
+    def _proc_gnusparse_00(self, next, raw_headers):
         """Process a GNU tar extended sparse header, version 0.0.
         """
         offsets = []
-        for match in re.finditer(br"\d+ GNU.sparse.offset=(\d+)\n", buf):
-            offsets.append(int(match.group(1)))
         numbytes = []
-        for match in re.finditer(br"\d+ GNU.sparse.numbytes=(\d+)\n", buf):
-            numbytes.append(int(match.group(1)))
+        for _, keyword, value in raw_headers:
+            if keyword == b"GNU.sparse.offset":
+                try:
+                    offsets.append(int(value.decode()))
+                except ValueError:
+                    raise InvalidHeaderError("invalid header")
+
+            elif keyword == b"GNU.sparse.numbytes":
+                try:
+                    numbytes.append(int(value.decode()))
+                except ValueError:
+                    raise InvalidHeaderError("invalid header")
+
         next.sparse = list(zip(offsets, numbytes))
 
     def _proc_gnusparse_01(self, next, pax_headers):

Lib/test/test_tarfile.py

@@ -1047,6 +1047,48 @@ def test_pax_number_fields(self):
         finally:
             tar.close()
 
+    def test_pax_header_bad_formats(self):
+        # The fields from the pax header have priority over the
+        # TarInfo.
+        pax_header_replacements = (
+            b" foo=bar\n",
+            b"0 \n",
+            b"1 \n",
+            b"2 \n",
+            b"3 =\n",
+            b"4 =a\n",
+            b"1000000 foo=bar\n",
+            b"0 foo=bar\n",
+            b"-12 foo=bar\n",
+            b"000000000000000000000000036 foo=bar\n",
+        )
+        pax_headers = {"foo": "bar"}
+
+        for replacement in pax_header_replacements:
+            with self.subTest(header=replacement):
+                tar = tarfile.open(tmpname, "w", format=tarfile.PAX_FORMAT,
+                                   encoding="iso8859-1")
+                try:
+                    t = tarfile.TarInfo()
+                    t.name = "pax"  # non-ASCII
+                    t.uid = 1
+                    t.pax_headers = pax_headers
+                    tar.addfile(t)
+                finally:
+                    tar.close()
+
+                with open(tmpname, "rb") as f:
+                    data = f.read()
+                    self.assertIn(b"11 foo=bar\n", data)
+                    data = data.replace(b"11 foo=bar\n", replacement)
+
+                with open(tmpname, "wb") as f:
+                    f.truncate()
+                    f.write(data)
+
+                with self.assertRaisesRegex(tarfile.ReadError, r"file could not be opened successfully"):
+                    tarfile.open(tmpname, encoding="iso8859-1")
+
 
 class WriteTestBase(TarTest):
     # Put all write tests in here that are supposed to be tested

Misc/NEWS.d/next/Security/2024-07-02-13-39-20.gh-issue-121285.hrl-yI.rst

@@ -0,0 +1,2 @@
+Remove backtracking from tarfile header parsing for ``hdrcharset``, PAX, and
+GNU sparse headers.