We have built a policy analysis tool called Gokyo that is designed to identify and enable resolution of conflicting policy specifications [12,13]. The general concept that Gokyo enforces is safety. A policy specification is said to be safe if no subject can obtain an unauthorized permission . If we take policy administration into account, then any permission can be assigned to any subject type by the administrator in a policy such as TE. Therefore, we need an additional specification concept, typically called constraints, to express the safety requirements on the administrators to ensure that our policy meets our goals.
Gokyo implements an approach called access control spaces whereby semantically meaningful permission sets are identified and handling can be associated with each set independently. While there are a variety of semantically meaningful permission sets, the most common to consider are: (1) those assigned to a subject type; (2) those precluded from a subject type by a constraint; and (3) the permissions whose assignment or preclusion status is unknown. The overlapping regions of these sets also form subspaces, so we can consider the set of permissions that are both assigned and precluded. Of course, these sets can be further decomposed to provide more effective control. For example, we may deny all integrity conflicts, except log writes, which we allow, and input from network objects, which we sanitize and audit.
The general idea is that by identifying semantics to the subspaces it may be possible to attach handling semantics with the entire subspace. This would permit administrators to keep the basic, simpler constraints that largely work and specify only the additional handling semantics. We have found that the Apache administrator policy for SELinux 2.4.16 largely adheres to a Biba integrity model, such that 19 conflicting cases can be handled as eight access control spaces .
Gokyo represents policies using a graphical access control model shown by example in Figure 3. Permissions (i.e., object types and the permitted operations), subject types, and subjects are represented by graph nodes. In addition to this information, permission nodes also store the object class (i.e., datatype) and operations permitted by the permission. Note that object types are represented by permissions with no rights. In general, a node represents a set, so it is possible to build set-hierarchies consisting of aggregations of individual permissions, subject types, and subjects.