Modelling and Analysis of Network Security Policies

Nowadays, computers and network communications have a pervasive presence in all our daily activities. Their correct configuration in terms of security is becoming more and more complex due to the growing number and variety of services present in a network. Generally, the security configuration of a computer network is dictated by specifying the policies of the security controls (e.g. firewall, VPN gateway) in the network. This implies that the specification of the network security policies is a crucial step to avoid errors in network configuration (e.g., blocking legitimate traffic, permitting unwanted traffic or sending insecure data). In the literature, an anomaly is an incorrect policy specification that an administrator may introduce in the network. In this thesis, we indicate as policy anomaly any conflict (e.g. two triggered policy rules enforcing contradictory actions), error (e.g. a policy cannot be enforced because it requires a cryptographic algorithm not supported by the security controls) or sub-optimization (e.g. redundant policies) that may arise in the policy specification phase. Security administrators, thus, have to face the hard job of correctly specifying the policies, which requires a high level of competence. Several studies have confirmed, in fact, that many security breaches and breakdowns are attributable to administrators’ responsibilities. Several approaches have been proposed to analyze the presence of anomalies among policy rules, in order to enforce a correct security configuration. However, we have identified two limitations of such approaches. On one hand, current literature identifies only the anomalies among policies of a single security technology (i.e., IPsec, TLS), while a network is generally configured with many technologies. On the other hand, existing approaches work on a single policy type, also named domain (i.e., filtering, communication protection). Unfortunately, the complexity of real systems is not self-contained and each network security control may affect the behavior of other controls in the same network. The objective of this PhD work was to investigate novel approaches for modelling security policies and their anomalies, and formal techniques of anomaly analysis. We present in this dissertation our contributions to the current policy analysis state of the art and the achieved results. A first contribution was the definition of a new class of policy anomalies, i.e. the inter-technology anomalies, which arises in a set of policies of multiple security technologies. We provided also a formal model able to detect these new types of anomalies. One of the results achieved by applying the inter-technology analysis to the communication protection policies was to categorize twelve new types of anomalies. The second result of this activity was derived from an empirical assessment that proved the practical significance of detecting such new anomalies. The second contribution of this thesis was the definition of a newly-defined type of policy analysis, named inter-domain analysis, which identifies any anomaly that may arise among different policy domains. We improved the state of the art by proposing a possible model to detect the inter-domain anomalies, which is a generalization of the aforementioned inter-technology model. In particular, we defined the Unified Model for Policy Analysis (UMPA) to perform the inter-domain analysis by extending the analysis model applied for a single policy domain to comprehensive analysis of anomalies among many policy domains. The result of this last part of our dissertation was to improve the effectiveness of the analysis process. Thanks to the inter-domain analysis, indeed, administrators can detect in a simple and customizable way a greater set of anomalies than the sets they could detect by running individually any other model.