File CVE-2019-9947-no-ctrl-char-http.patch of Package python3-base.36306

Overview Repositories Revisions Requests Users Attributes Meta

File CVE-2019-9947-no-ctrl-char-http.patch of Package python3-base.36306

---
 Lib/http/client.py                                                 |   15 +++
 Lib/test/test_urllib.py                                            |   50 ++++++++++
 Lib/test/test_xmlrpc.py                                            |   10 +-
 Misc/NEWS.d/next/Security/2019-04-10-08-53-30.bpo-36276.51E-DA.rst |    1 
 4 files changed, 73 insertions(+), 3 deletions(-)

--- a/Lib/http/client.py
+++ b/Lib/http/client.py
@@ -246,6 +246,16 @@ _MAXHEADERS = 100
 _is_legal_header_name = re.compile(rb'[^:\s][^:\r\n]*').fullmatch
 _is_illegal_header_value = re.compile(rb'\n(?![ \t])|\r(?![ \t\n])').search
 
+# These characters are not allowed within HTTP URL paths.
+#  See https://tools.ietf.org/html/rfc3986#section-3.3 and the
+#  https://tools.ietf.org/html/rfc3986#appendix-A pchar definition.
+# Prevents CVE-2019-9740.  Includes control characters such as \r\n.
+# We don't restrict chars above \x7f as putrequest() limits us to ASCII.
+_contains_disallowed_url_pchar_re = re.compile('[\x00-\x20\x7f]')
+# Arguably only these _should_ allowed:
+#  _is_allowed_url_pchars_re = re.compile(r"^[/!$&'()*+,;=:@%a-zA-Z0-9._~-]+$")
+# We are more lenient for assumed real world compatibility purposes.
+
 # We always set the Content-Length header for these methods because some
 # servers will otherwise respond with a 411
 _METHODS_EXPECTING_BODY = {'PATCH', 'POST', 'PUT'}
@@ -1008,6 +1018,11 @@ class HTTPConnection:
         self._method = method
         if not url:
             url = '/'
+        # Prevent CVE-2019-9740.
+        match = _contains_disallowed_url_pchar_re.search(url)
+        if match:
+            raise InvalidURL(("URL can't contain control characters. {!r} " +
+                             "(found at least {!r})").format(url, match.group))
         request = '%s %s %s' % (method, url, self._http_vsn_str)
 
         # Non-ASCII characters should have been eliminated earlier
--- a/Lib/test/test_urllib.py
+++ b/Lib/test/test_urllib.py
@@ -305,6 +305,56 @@ class urlopen_HttpTests(unittest.TestCas
         finally:
             self.unfakehttp()
 
+    def test_url_with_control_char_rejected(self):
+        for char_no in list(range(0, 0x21)) + [0x7f]:
+            char = chr(char_no)
+            schemeless_url = "//localhost:7777/test{}/".format(char)
+            self.fakehttp(b"HTTP/1.1 200 OK\r\n\r\nHello.")
+            try:
+                # We explicitly test urllib.request.urlopen() instead of
+                # the top level 'def urlopen()' function defined in
+                # this... (quite ugly) test suite.  They use different
+                # url opening codepaths.  Plain urlopen uses
+                # FancyURLOpener which goes via a codepath that calls
+                # urllib.parse.quote() on the URL which makes all of the
+                # above attempts at injection within the url _path_
+                # safe.
+                escaped_char_repr = repr(char).replace('\\', r'\\')
+                with self.assertRaisesRegex(
+                        http.client.InvalidURL, "contain control"):
+                    urllib.request.urlopen("http:{}".format(schemeless_url))
+                with self.assertRaisesRegex(
+                        http.client.InvalidURL, "contain control"):
+                    urllib.request.urlopen("https:{}".format(schemeless_url))
+                # This code path quotes the URL so there is no injection.
+                resp = urlopen("http:{}".format(schemeless_url))
+                self.assertNotIn(char, resp.geturl())
+            finally:
+                self.unfakehttp()
+
+    def test_url_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"
+        try:
+            # We explicitly test urllib.request.urlopen() instead of the top
+            # level 'def urlopen()' function defined in this... (quite ugly)
+            # test suite.  they use different url opening codepaths.  plain
+            # urlopen uses FancyURLOpener which goes via a codepath that
+            # calls urllib.parse.quote() on the URL which makes all of the
+            # above attempts at injection within the url _path_ safe.
+            with self.assertRaisesRegex(http.client.InvalidURL, r"contain control.*\\r"):
+                urllib.request.urlopen("http:{}".format(schemeless_url))
+            with self.assertRaisesRegex(http.client.InvalidURL, r"contain control.*\\n"):
+                urllib.request.urlopen("http:{}".format(schemeless_url))
+            # This code path quotes the URL so there is no injection.
+            resp = urlopen("http:{}".format(schemeless_url))
+            self.assertNotIn(' ', resp.geturl())
+            self.assertNotIn('\r', resp.geturl())
+            self.assertNotIn('\n', resp.geturl())
+        finally:
+            self.unfakehttp()
+
     def test_read_0_9(self):
         # "0.9" response accepted (but not "simple responses" without
         # a status line)
--- a/Lib/test/test_xmlrpc.py
+++ b/Lib/test/test_xmlrpc.py
@@ -709,9 +709,13 @@ class SimpleServerTestCase(BaseServerTes
 
     def test_partial_post(self):
         # Check that a partial POST doesn't make the server loop: issue #14001.
-        conn = http.client.HTTPConnection(ADDR, PORT)
-        conn.request('POST', '/RPC2 HTTP/1.0\r\nContent-Length: 100\r\n\r\nbye')
-        conn.close()
+        with contextlib.closing(socket.create_connection((ADDR, PORT))) as conn:
+            conn.send(('POST /RPC2 HTTP/1.0\r\n' +
+                      'Content-Length: 100\r\n\r\n' +
+                      'bye HTTP/1.1\r\n' +
+                      'Host: {}:{}\r\n'.format(ADDR,PORT) +
+                      'Accept-Encoding: identity\r\n' +
+                      'Content-Length: 0\r\n\r\n').encode('ascii'))
 
 
 class MultiPathServerTestCase(BaseServerTestCase):
--- /dev/null
+++ b/Misc/NEWS.d/next/Security/2019-04-10-08-53-30.bpo-36276.51E-DA.rst
@@ -0,0 +1 @@
+Address CVE-2019-9740 by disallowing URL paths with embedded whitespace or control characters through into the underlying http client request.  Such potentially malicious header injection URLs now cause a InvalidURL to be raised.
\ No newline at end of file

Places

File CVE-2019-9947-no-ctrl-char-http.patch of Package python3-base.36306

Places