changeset:   69052:b2934d98dac1
branch:      2.7
parent:      69036:1a5aab273332
parent:      69051:9eeda8e3a13f
user:        Guido van Rossum <guido@python.org>
date:        Tue Mar 29 12:53:55 2011 -0700
files:       .hgignore Lib/test/test_urllib.py Lib/test/test_urllib2.py Lib/urllib.py Lib/urllib2.py Misc/NEWS
description:
Merge issue 11662 from 2.6.


diff -r 1a5aab273332 -r b2934d98dac1 Lib/test/test_urllib.py
--- a/Lib/test/test_urllib.py	Tue Mar 29 01:21:48 2011 +0100
+++ b/Lib/test/test_urllib.py	Tue Mar 29 12:53:55 2011 -0700
@@ -161,6 +161,20 @@
         finally:
             self.unfakehttp()
 
+    def test_invalid_redirect(self):
+        # urlopen() should raise IOError for many error codes.
+        self.fakehttp("""HTTP/1.1 302 Found
+Date: Wed, 02 Jan 2008 03:03:54 GMT
+Server: Apache/1.3.33 (Debian GNU/Linux) mod_ssl/2.8.22 OpenSSL/0.9.7e
+Location: file:README
+Connection: close
+Content-Type: text/html; charset=iso-8859-1
+""")
+        try:
+            self.assertRaises(IOError, urllib.urlopen, "http://python.org/")
+        finally:
+            self.unfakehttp()
+
     def test_empty_socket(self):
         # urlopen() raises IOError if the underlying socket does not send any
         # data. (#1680230)
diff -r 1a5aab273332 -r b2934d98dac1 Lib/test/test_urllib2.py
--- a/Lib/test/test_urllib2.py	Tue Mar 29 01:21:48 2011 +0100
+++ b/Lib/test/test_urllib2.py	Tue Mar 29 12:53:55 2011 -0700
@@ -969,6 +969,27 @@
             self.assertEqual(count,
                              urllib2.HTTPRedirectHandler.max_redirections)
 
+    def test_invalid_redirect(self):
+        from_url = "http://example.com/a.html"
+        valid_schemes = ['http', 'https', 'ftp']
+        invalid_schemes = ['file', 'imap', 'ldap']
+        schemeless_url = "example.com/b.html"
+        h = urllib2.HTTPRedirectHandler()
+        o = h.parent = MockOpener()
+        req = Request(from_url)
+
+        for scheme in invalid_schemes:
+            invalid_url = scheme + '://' + schemeless_url
+            self.assertRaises(urllib2.HTTPError, h.http_error_302,
+                              req, MockFile(), 302, "Security Loophole",
+                              MockHeaders({"location": invalid_url}))
+
+        for scheme in valid_schemes:
+            valid_url = scheme + '://' + schemeless_url
+            h.http_error_302(req, MockFile(), 302, "That's fine",
+                MockHeaders({"location": valid_url}))
+            self.assertEqual(o.req.get_full_url(), valid_url)
+
     def test_cookie_redirect(self):
         # cookies shouldn't leak into redirected requests
         from cookielib import CookieJar
diff -r 1a5aab273332 -r b2934d98dac1 Lib/urllib.py
--- a/Lib/urllib.py	Tue Mar 29 01:21:48 2011 +0100
+++ b/Lib/urllib.py	Tue Mar 29 12:53:55 2011 -0700
@@ -644,6 +644,18 @@
         fp.close()
         # In case the server sent a relative URL, join with original:
         newurl = basejoin(self.type + ":" + url, newurl)
+
+        # For security reasons we do not allow redirects to protocols
+        # other than HTTP, HTTPS or FTP.
+        newurl_lower = newurl.lower()
+        if not (newurl_lower.startswith('http://') or
+                newurl_lower.startswith('https://') or
+                newurl_lower.startswith('ftp://')):
+            raise IOError('redirect error', errcode,
+                          errmsg + " - Redirection to url '%s' is not allowed" %
+                          newurl,
+                          headers)
+
         return self.open(newurl)
 
     def http_error_301(self, url, fp, errcode, errmsg, headers, data=None):
diff -r 1a5aab273332 -r b2934d98dac1 Lib/urllib2.py
--- a/Lib/urllib2.py	Tue Mar 29 01:21:48 2011 +0100
+++ b/Lib/urllib2.py	Tue Mar 29 12:53:55 2011 -0700
@@ -578,6 +578,17 @@
 
         newurl = urlparse.urljoin(req.get_full_url(), newurl)
 
+        # For security reasons we do not allow redirects to protocols
+        # other than HTTP, HTTPS or FTP.
+        newurl_lower = newurl.lower()
+        if not (newurl_lower.startswith('http://') or
+                newurl_lower.startswith('https://') or
+                newurl_lower.startswith('ftp://')):
+            raise HTTPError(newurl, code,
+                            msg + " - Redirection to url '%s' is not allowed" %
+                            newurl,
+                            headers, fp)
+
         # XXX Probably want to forget about the state of the current
         # request, although that might interact poorly with other
         # handlers that also use handler-specific request attributes
diff -r 1a5aab273332 -r b2934d98dac1 Misc/NEWS
--- a/Misc/NEWS	Tue Mar 29 01:21:48 2011 +0100
+++ b/Misc/NEWS	Tue Mar 29 12:53:55 2011 -0700
@@ -16,6 +16,9 @@
 - Issue #11450: Don't truncate hg version info in Py_GetBuildInfo() when
   there are many tags (e.g. when using mq).  Patch by Nadeem Vawda.
 
+- Issue #11662: Make urllib and urllib2 ignore redirections if the
+  scheme is not HTTP, HTTPS or FTP (CVE-2011-1521).
+
 - Issue #10451: memoryview objects could allow to mutate a readable buffer.
   Initial patch by Ross Lagerwall.