Prevent leaking target info into subsequent policy.check() calls

Due to mutable dictionary being used as the default `target` argument
value the first target calculated from scratch in POLICY_CHECK
function will be used for all subsequent calls to POLICY_CHECK with 2
arguments. The wrong `target` can either lead to a reduced set of
operations on an entity for a given user, or to enlarged one. Due to
independent policy checks at each service's side this doesn't pose a
serious security breach, but can lead to weird UX behaviour.

Change-Id: I744fac28de0fb7060b50c5db689e74631a628c88
Closes-Bug: #1396544

openstack_dashboard/policy.py

@@ -61,7 +61,7 @@ def reset():
     _ENFORCER = None
 
 
-def check(actions, request, target={}):
+def check(actions, request, target=None):
     """Check user permission.
 
     Check if the user has permission to the action according
@@ -98,6 +98,8 @@ def check(actions, request, target={}):
                       {'project_id': object.project_id}
     :returns: boolean if the user has permission or not for the actions.
     """
+    if target is None:
+        target = {}
     user = auth_utils.get_user(request)
 
     # Several service policy engines default to a project id check for