diff --git a/Lib/httplib.py b/Lib/httplib.py
index da2f346..fc8e895 100644
--- a/Lib/httplib.py
+++ b/Lib/httplib.py
@@ -247,6 +247,15 @@ _MAXHEADERS = 100
_is_legal_header_name = re.compile(r'\A[^:\s][^:\r\n]*\Z').match
_is_illegal_header_value = re.compile(r'\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.
+# Restrict non-ASCII characters above \x7f (0x80-0xff).
+_contains_disallowed_url_pchar_re = re.compile('[\x00-\x20\x7f-\xff]')
+# 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.
class HTTPMessage(mimetools.Message):
@@ -926,6 +935,12 @@ 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)"
+ % (url, match.group()))
hdr = '%s %s %s' % (method, url, self._http_vsn_str)
self._output(hdr)
diff --git a/Lib/test/test_urllib.py b/Lib/test/test_urllib.py
index 3845012..d2da0f8 100644
--- a/Lib/test/test_urllib.py
+++ b/Lib/test/test_urllib.py
@@ -198,6 +198,31 @@ class urlopen_HttpTests(unittest.TestCase, FakeHTTPMixin):
finally:
self.unfakehttp()
+ def test_url_with_control_char_rejected(self):
+ for char_no in range(0, 0x21) + range(0x7f, 0x100):
+ char = chr(char_no)
+ schemeless_url = "//localhost:7777/test%s/" % char
+ self.fakehttp(b"HTTP/1.1 200 OK\r\n\r\nHello.")
+ try:
+ # urllib quotes the URL so there is no injection.
+ resp = urllib.urlopen("http:" + 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:
+ # urllib quotes the URL so there is no injection.
+ resp = urllib.urlopen("http:" + schemeless_url)
+ self.assertNotIn(' ', resp.geturl())
+ self.assertNotIn('\r', resp.geturl())
+ self.assertNotIn('\n', resp.geturl())
+ finally:
+ self.unfakehttp()
+
def test_read_bogus(self):
# urlopen() should raise IOError for many error codes.
self.fakehttp('''HTTP/1.1 401 Authentication Required
@@ -786,6 +811,35 @@ class Pathname_Tests(unittest.TestCase):
class Utility_Tests(unittest.TestCase):
"""Testcase to test the various utility functions in the urllib."""
+ def test_splithost(self):
+ splithost = urllib.splithost
+ self.assertEqual(splithost('//www.example.org:80/foo/bar/baz.html'),
+ ('www.example.org:80', '/foo/bar/baz.html'))
+ self.assertEqual(splithost('//www.example.org:80'),
+ ('www.example.org:80', ''))
+ self.assertEqual(splithost('/foo/bar/baz.html'),
+ (None, '/foo/bar/baz.html'))
+
+ # bpo-30500: # starts a fragment.
+ self.assertEqual(splithost('//127.0.0.1#@host.com'),
+ ('127.0.0.1', '/#@host.com'))
+ self.assertEqual(splithost('//127.0.0.1#@host.com:80'),
+ ('127.0.0.1', '/#@host.com:80'))
+ self.assertEqual(splithost('//127.0.0.1:80#@host.com'),
+ ('127.0.0.1:80', '/#@host.com'))
+
+ # Empty host is returned as empty string.
+ self.assertEqual(splithost("///file"),
+ ('', '/file'))
+
+ # Trailing semicolon, question mark and hash symbol are kept.
+ self.assertEqual(splithost("//example.net/file;"),
+ ('example.net', '/file;'))
+ self.assertEqual(splithost("//example.net/file?"),
+ ('example.net', '/file?'))
+ self.assertEqual(splithost("//example.net/file#"),
+ ('example.net', '/file#'))
+
def test_splitpasswd(self):
"""Some of the password examples are not sensible, but it is added to
confirming to RFC2617 and addressing issue4675.
diff --git a/Lib/test/test_urllib2.py b/Lib/test/test_urllib2.py
index c317b8d..63fefd6 100644
--- a/Lib/test/test_urllib2.py
+++ b/Lib/test/test_urllib2.py
@@ -7,12 +7,16 @@ import StringIO
import urllib2
from urllib2 import Request, OpenerDirector
+import httplib
try:
import ssl
except ImportError:
ssl = None
+from test.test_urllib import FakeHTTPMixin
+
+
# XXX
# Request
# CacheFTPHandler (hard to write)
@@ -1243,7 +1247,7 @@ class HandlerTests(unittest.TestCase):
self.assertEqual(len(http_handler.requests), 1)
self.assertFalse(http_handler.requests[0].has_header(auth_header))
-class MiscTests(unittest.TestCase):
+class MiscTests(unittest.TestCase, FakeHTTPMixin):
def test_build_opener(self):
class MyHTTPHandler(urllib2.HTTPHandler): pass
@@ -1289,6 +1293,53 @@ class MiscTests(unittest.TestCase):
else:
self.assertTrue(False)
+ @unittest.skipUnless(ssl, "ssl module required")
+ def test_url_with_control_char_rejected(self):
+ for char_no in range(0, 0x21) + range(0x7f, 0x100):
+ char = chr(char_no)
+ schemeless_url = "//localhost:7777/test%s/" % 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'\\')
+ InvalidURL = httplib.InvalidURL
+ with self.assertRaisesRegexp(
+ InvalidURL, "contain control.*" + escaped_char_repr):
+ urllib2.urlopen("http:" + schemeless_url)
+ with self.assertRaisesRegexp(
+ InvalidURL, "contain control.*" + escaped_char_repr):
+ urllib2.urlopen("https:" + schemeless_url)
+ finally:
+ self.unfakehttp()
+
+ @unittest.skipUnless(ssl, "ssl module required")
+ 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 urllib2.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.
+ InvalidURL = httplib.InvalidURL
+ with self.assertRaisesRegexp(
+ InvalidURL, r"contain control.*\\r.*(found at least . .)"):
+ urllib2.urlopen("http:" + schemeless_url)
+ with self.assertRaisesRegexp(InvalidURL, r"contain control.*\\n"):
+ urllib2.urlopen("https:" + schemeless_url)
+ finally:
+ self.unfakehttp()
+
+
+
class RequestTests(unittest.TestCase):
def setUp(self):
diff --git a/Lib/test/test_xmlrpc.py b/Lib/test/test_xmlrpc.py
index 79e862a..347b494 100644
--- a/Lib/test/test_xmlrpc.py
+++ b/Lib/test/test_xmlrpc.py
@@ -592,7 +592,13 @@ class SimpleServerTestCase(BaseServerTestCase):
def test_partial_post(self):
# Check that a partial POST doesn't make the server loop: issue #14001.
conn = httplib.HTTPConnection(ADDR, PORT)
- conn.request('POST', '/RPC2 HTTP/1.0\r\nContent-Length: 100\r\n\r\nbye')
+ conn.send('POST /RPC2 HTTP/1.0\r\n'
+ 'Content-Length: 100\r\n\r\n'
+ 'bye HTTP/1.1\r\n'
+ 'Host: %s:%s\r\n'
+ 'Accept-Encoding: identity\r\n'
+ 'Content-Length: 0\r\n\r\n'
+ % (ADDR, PORT))
conn.close()
class MultiPathServerTestCase(BaseServerTestCase):
diff --git a/Lib/urllib.py b/Lib/urllib.py
index 9b31df1..2201e3e 100644
--- a/Lib/urllib.py
+++ b/Lib/urllib.py
@@ -1079,8 +1079,7 @@ def splithost(url):
"""splithost('//host[:port]/path') --> 'host[:port]', '/path'."""
global _hostprog
if _hostprog is None:
- import re
- _hostprog = re.compile('^//([^/?]*)(.*)$')
+ _hostprog = re.compile('//([^/#?]*)(.*)', re.DOTALL)
match = _hostprog.match(url)
if match: