bpo-38576: Disallow control characters in hostnames in http.client (GH-18995)

Add host validation for control characters for more CVE-2019-18348 protection.
(cherry picked from commit 9165addc22)

Co-authored-by: Ashwin Ramaswami <aramaswamis@gmail.com>

Lib/http/client.py

@@ -828,6 +828,8 @@ class HTTPConnection:
 
         (self.host, self.port) = self._get_hostport(host, port)
 
+        self._validate_host(self.host)
+
         # This is stored as an instance variable to allow unit
         # tests to replace it with a suitable mockup
         self._create_connection = socket.create_connection
@@ -1183,6 +1185,14 @@ class HTTPConnection:
             raise InvalidURL(f"URL can't contain control characters. {url!r} "
                              f"(found at least {match.group()!r})")
 
+    def _validate_host(self, host):
+        """Validate a host so it doesn't contain control characters."""
+        # Prevent CVE-2019-18348.
+        match = _contains_disallowed_url_pchar_re.search(host)
+        if match:
+            raise InvalidURL(f"URL can't contain control characters. {host!r} "
+                             f"(found at least {match.group()!r})")
+
     def putheader(self, header, *values):
         """Send a request header line to the server.
 

Lib/test/test_httplib.py

@@ -1155,7 +1155,7 @@ class BasicTest(TestCase):
         thread.join()
         self.assertEqual(result, b"proxied data\n")
 
-    def test_putrequest_override_validation(self):
+    def test_putrequest_override_domain_validation(self):
         """
         It should be possible to override the default validation
         behavior in putrequest (bpo-38216).
@@ -1168,6 +1168,17 @@ class BasicTest(TestCase):
         conn.sock = FakeSocket('')
         conn.putrequest('GET', '/\x00')
 
+    def test_putrequest_override_host_validation(self):
+        class UnsafeHTTPConnection(client.HTTPConnection):
+            def _validate_host(self, url):
+                pass
+
+        conn = UnsafeHTTPConnection('example.com\r\n')
+        conn.sock = FakeSocket('')
+        # set skip_host so a ValueError is not raised upon adding the
+        # invalid URL as the value of the "Host:" header
+        conn.putrequest('GET', '/', skip_host=1)
+
     def test_putrequest_override_encoding(self):
         """
         It should be possible to override the default encoding

Lib/test/test_urllib.py

@@ -361,7 +361,7 @@ class urlopen_HttpTests(unittest.TestCase, FakeHTTPMixin, FakeFTPMixin):
             self.unfakehttp()
 
     @unittest.skipUnless(ssl, "ssl module required")
-    def test_url_with_control_char_rejected(self):
+    def test_url_path_with_control_char_rejected(self):
         for char_no in list(range(0, 0x21)) + [0x7f]:
             char = chr(char_no)
             schemeless_url = f"//localhost:7777/test{char}/"
@@ -388,7 +388,7 @@ class urlopen_HttpTests(unittest.TestCase, FakeHTTPMixin, FakeFTPMixin):
                 self.unfakehttp()
 
     @unittest.skipUnless(ssl, "ssl module required")
-    def test_url_with_newline_header_injection_rejected(self):
+    def test_url_path_with_newline_header_injection_rejected(self):
         self.fakehttp(b"HTTP/1.1 200 OK\r\n\r\nHello.")
         host = "localhost:7777?a=1 HTTP/1.1\r\nX-injected: header\r\nTEST: 123"
         schemeless_url = "//" + host + ":8080/test/?test=a"
@@ -413,6 +413,38 @@ class urlopen_HttpTests(unittest.TestCase, FakeHTTPMixin, FakeFTPMixin):
         finally:
             self.unfakehttp()
 
+    @unittest.skipUnless(ssl, "ssl module required")
+    def test_url_host_with_control_char_rejected(self):
+        for char_no in list(range(0, 0x21)) + [0x7f]:
+            char = chr(char_no)
+            schemeless_url = f"//localhost{char}/test/"
+            self.fakehttp(b"HTTP/1.1 200 OK\r\n\r\nHello.")
+            try:
+                escaped_char_repr = repr(char).replace('\\', r'\\')
+                InvalidURL = http.client.InvalidURL
+                with self.assertRaisesRegex(
+                    InvalidURL, f"contain control.*{escaped_char_repr}"):
+                    urlopen(f"http:{schemeless_url}")
+                with self.assertRaisesRegex(InvalidURL, f"contain control.*{escaped_char_repr}"):
+                    urlopen(f"https:{schemeless_url}")
+            finally:
+                self.unfakehttp()
+
+    @unittest.skipUnless(ssl, "ssl module required")
+    def test_url_host_with_newline_header_injection_rejected(self):
+        self.fakehttp(b"HTTP/1.1 200 OK\r\n\r\nHello.")
+        host = "localhost\r\nX-injected: header\r\n"
+        schemeless_url = "//" + host + ":8080/test/?test=a"
+        try:
+            InvalidURL = http.client.InvalidURL
+            with self.assertRaisesRegex(
+                InvalidURL, r"contain control.*\\r"):
+                urlopen(f"http:{schemeless_url}")
+            with self.assertRaisesRegex(InvalidURL, r"contain control.*\\n"):
+                urlopen(f"https:{schemeless_url}")
+        finally:
+            self.unfakehttp()
+
     def test_read_0_9(self):
         # "0.9" response accepted (but not "simple responses" without
         # a status line)

Misc/NEWS.d/next/Security/2020-03-14-14-57-44.bpo-38576.OowwQn.rst

@@ -0,0 +1 @@
+Disallow control characters in hostnames in http.client, addressing CVE-2019-18348. Such potentially malicious header injection URLs now cause a InvalidURL to be raised.