From 698c2fa850bfc8b3bdb768e1c1cd6d57e643811d Mon Sep 17 00:00:00 2001
From: Bruce Merry <bmerry@ska.ac.za>
Date: Tue, 14 Aug 2018 13:30:43 +0200
Subject: [PATCH 2/2] Rework authorization stripping logic as discussed

The exception for http->https upgrade now requires the standard HTTP(S)
ports to be used, either implicitly (no port specified) or explicitly.

Upstream-Status: Backport

Follow-up fix for CVE-2018-18074

Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
---
 requests/sessions.py   | 26 ++++++++++++++++++--------
 tests/test_requests.py | 33 ++++++++++++++++++++++-----------
 2 files changed, 40 insertions(+), 19 deletions(-)

diff --git a/requests/sessions.py b/requests/sessions.py
index 2969d83..c11a3a2 100644
--- a/requests/sessions.py
+++ b/requests/sessions.py
@@ -115,6 +115,22 @@ class SessionRedirectMixin(object):
             return to_native_string(location, 'utf8')
         return None
 
+    def should_strip_auth(self, old_url, new_url):
+        """Decide whether Authorization header should be removed when redirecting"""
+        old_parsed = urlparse(old_url)
+        new_parsed = urlparse(new_url)
+        if old_parsed.hostname != new_parsed.hostname:
+            return True
+        # Special case: allow http -> https redirect when using the standard
+        # ports. This isn't specified by RFC 7235, but is kept to avoid
+        # breaking backwards compatibility with older versions of requests
+        # that allowed any redirects on the same host.
+        if (old_parsed.scheme == 'http' and old_parsed.port in (80, None)
+                and new_parsed.scheme == 'https' and new_parsed.port in (443, None)):
+            return False
+        # Standard case: root URI must match
+        return old_parsed.port != new_parsed.port or old_parsed.scheme != new_parsed.scheme
+
     def resolve_redirects(self, resp, req, stream=False, timeout=None,
                           verify=True, cert=None, proxies=None, yield_requests=False, **adapter_kwargs):
         """Receives a Response. Returns a generator of Responses or Requests."""
@@ -236,16 +252,10 @@ class SessionRedirectMixin(object):
         headers = prepared_request.headers
         url = prepared_request.url
 
-        if 'Authorization' in headers:
+        if 'Authorization' in headers and self.should_strip_auth(response.request.url, url):
             # If we get redirected to a new host, we should strip out any
             # authentication headers.
-            original_parsed = urlparse(response.request.url)
-            redirect_parsed = urlparse(url)
-
-            if (original_parsed.hostname != redirect_parsed.hostname
-                    or original_parsed.port != redirect_parsed.port
-                    or original_parsed.scheme != redirect_parsed.scheme):
-                del headers['Authorization']
+            del headers['Authorization']
 
         # .netrc might have more auth for us on our new host.
         new_auth = get_netrc_auth(url) if self.trust_env else None
diff --git a/tests/test_requests.py b/tests/test_requests.py
index e0e801a..148067b 100644
--- a/tests/test_requests.py
+++ b/tests/test_requests.py
@@ -1567,17 +1567,7 @@ class TestRequests:
             preq = req.prepare()
             assert test_url == preq.url
 
-    @pytest.mark.xfail(raises=ConnectionError)
-    def test_auth_is_stripped_on_redirect_off_host(self, httpbin):
-        r = requests.get(
-            httpbin('redirect-to'),
-            params={'url': 'http://www.google.co.uk'},
-            auth=('user', 'pass'),
-        )
-        assert r.history[0].request.headers['Authorization']
-        assert 'Authorization' not in r.request.headers
-
-    def test_auth_is_stripped_on_scheme_redirect(self, httpbin, httpbin_secure, httpbin_ca_bundle):
+    def test_auth_is_stripped_on_http_downgrade(self, httpbin, httpbin_secure, httpbin_ca_bundle):
         r = requests.get(
             httpbin_secure('redirect-to'),
             params={'url': httpbin('get')},
@@ -1594,6 +1584,27 @@ class TestRequests:
 
         assert h1 == h2
 
+    def test_should_strip_auth_host_change(self):
+        s = requests.Session()
+        assert s.should_strip_auth('http://example.com/foo', 'http://another.example.com/')
+
+    def test_should_strip_auth_http_downgrade(self):
+        s = requests.Session()
+        assert s.should_strip_auth('https://example.com/foo', 'http://example.com/bar')
+
+    def test_should_strip_auth_https_upgrade(self):
+        s = requests.Session()
+        assert not s.should_strip_auth('http://example.com/foo', 'https://example.com/bar')
+        assert not s.should_strip_auth('http://example.com:80/foo', 'https://example.com/bar')
+        assert not s.should_strip_auth('http://example.com/foo', 'https://example.com:443/bar')
+        # Non-standard ports should trigger stripping
+        assert s.should_strip_auth('http://example.com:8080/foo', 'https://example.com/bar')
+        assert s.should_strip_auth('http://example.com/foo', 'https://example.com:8443/bar')
+
+    def test_should_strip_auth_port_change(self):
+        s = requests.Session()
+        assert s.should_strip_auth('http://example.com:1234/foo', 'https://example.com:4321/bar')
+
     def test_manual_redirect_with_partial_body_read(self, httpbin):
         s = requests.Session()
         r1 = s.get(httpbin('redirect/2'), allow_redirects=False, stream=True)
-- 
2.7.4