gh-98433: Fix quadratic time idna decoding. (#99092)

There was an unnecessary quadratic loop in idna decoding. This restores
the behavior to linear.

This also adds an early length check in IDNA decoding to outright reject
huge inputs early on given the ultimate result is defined to be 63 or fewer
characters.

Lib/encodings/idna.py

@@ -39,23 +39,21 @@ def nameprep(label):
 
     # Check bidi
     RandAL = [stringprep.in_table_d1(x) for x in label]
-    for c in RandAL:
-        if c:
-            # There is a RandAL char in the string. Must perform further
-            # tests:
-            # 1) The characters in section 5.8 MUST be prohibited.
-            # This is table C.8, which was already checked
-            # 2) If a string contains any RandALCat character, the string
-            # MUST NOT contain any LCat character.
-            if any(stringprep.in_table_d2(x) for x in label):
-                raise UnicodeError("Violation of BIDI requirement 2")
-
-            # 3) If a string contains any RandALCat character, a
-            # RandALCat character MUST be the first character of the
-            # string, and a RandALCat character MUST be the last
-            # character of the string.
-            if not RandAL[0] or not RandAL[-1]:
-                raise UnicodeError("Violation of BIDI requirement 3")
+    if any(RandAL):
+        # There is a RandAL char in the string. Must perform further
+        # tests:
+        # 1) The characters in section 5.8 MUST be prohibited.
+        # This is table C.8, which was already checked
+        # 2) If a string contains any RandALCat character, the string
+        # MUST NOT contain any LCat character.
+        if any(stringprep.in_table_d2(x) for x in label):
+            raise UnicodeError("Violation of BIDI requirement 2")
+        # 3) If a string contains any RandALCat character, a
+        # RandALCat character MUST be the first character of the
+        # string, and a RandALCat character MUST be the last
+        # character of the string.
+        if not RandAL[0] or not RandAL[-1]:
+            raise UnicodeError("Violation of BIDI requirement 3")
 
     return label
 
@@ -103,6 +101,16 @@ def ToASCII(label):
     raise UnicodeError("label empty or too long")
 
 def ToUnicode(label):
+    if len(label) > 1024:
+        # Protection from https://github.com/python/cpython/issues/98433.
+        # https://datatracker.ietf.org/doc/html/rfc5894#section-6
+        # doesn't specify a label size limit prior to NAMEPREP. But having
+        # one makes practical sense.
+        # This leaves ample room for nameprep() to remove Nothing characters
+        # per https://www.rfc-editor.org/rfc/rfc3454#section-3.1 while still
+        # preventing us from wasting time decoding a big thing that'll just
+        # hit the actual <= 63 length limit in Step 6.
+        raise UnicodeError("label way too long")
     # Step 1: Check for ASCII
     if isinstance(label, bytes):
         pure_ascii = True

Lib/test/test_codecs.py

@@ -1552,6 +1552,12 @@ class IDNACodecTest(unittest.TestCase):
         self.assertEqual("pyth\xf6n.org".encode("idna"), b"xn--pythn-mua.org")
         self.assertEqual("pyth\xf6n.org.".encode("idna"), b"xn--pythn-mua.org.")
 
+    def test_builtin_decode_length_limit(self):
+        with self.assertRaisesRegex(UnicodeError, "way too long"):
+            (b"xn--016c"+b"a"*1100).decode("idna")
+        with self.assertRaisesRegex(UnicodeError, "too long"):
+            (b"xn--016c"+b"a"*70).decode("idna")
+
     def test_stream(self):
         r = codecs.getreader("idna")(io.BytesIO(b"abc"))
         r.read(3)

Misc/NEWS.d/next/Security/2022-11-04-09-29-36.gh-issue-98433.l76c5G.rst

@@ -0,0 +1,14 @@
+The IDNA codec decoder used on DNS hostnames by :mod:`socket` or :mod:`asyncio`
+related name resolution functions no longer involves a quadratic algorithm.
+This prevents a potential CPU denial of service if an out-of-spec excessive
+length hostname involving bidirectional characters were decoded. Some protocols
+such as :mod:`urllib` http ``3xx`` redirects potentially allow for an attacker
+to supply such a name.
+
+Individual labels within an IDNA encoded DNS name will now raise an error early
+during IDNA decoding if they are longer than 1024 unicode characters given that
+each decoded DNS label must be 63 or fewer characters and the entire decoded
+DNS name is limited to 255. Only an application presenting a hostname or label
+consisting primarily of :rfc:`3454` section 3.1 "Nothing" characters to be
+removed would run into of this new limit. See also :rfc:`5894` section 6 and
+:rfc:`3491`.