Complex cloud infrastructure is managed through con guration les that are compiled into atomic deployment instructions as part of a process known as Infrastructure as Code (IaC). Con guration les contain declarations for the resources to be created, their settings, and their connectivity. Unfortunately, the same features that make IaC a convenient and powerful deployment tool| reusability, modularity, and shareability|also threaten the security of the cloud. The vulnerabilities arising from such a practice are subtle and widespread and need to be detected early, at the level of con guration les, before potentiallyvulnerable infrastructure is deployed. To this end, we research the application of knowledge representation formalisms to the modeling and reasoning of IaC les. In particular, description logics allow for a succinct and natural description of these con guration les, and the open-world assumption captures the distributed nature of the cloud, where a newly deployed portion of infrastructure could connect to pre-existing resources not necessarily owned by the same user and whose con guration is only partially known. In previous work, we used the expressive ALCOIQ to model and reason about AWS CloudFormation, Amazon Web Services proprietary Infrastructure as Code framework [6,5]. Here, we suggest a lightweight DL that is speci cally tailored for cloud infrastructure. Core-closed Knowledge Bases We devise an extension of DL-LiteF that allows for combining a core part that is completely de ned (closed-world) and interacts with a partially known environment (open-world). We introduce the so-called \core-closed" knowledge bases, which are DL-LiteF KBs de ned as the tuple K = hT ; A; S; Mi, built from a standard KB hT ; Ai and a core hS; Mi. The set S contains DL-LiteF axioms representing the core structural cloud speci cations for each type of resource that can be deployed, and the set M contains positive concept and role assertions representing the core user con guration. Syntactically, M is similar to an ABox A but, di erently from A, it is assumed to be complete with respect to the speci cations S. As usual, hT ; Ai encodes the incomplete terminological and assertional knowledge that, in our setting, may refer to both the (closed) core and the surrounding (open) world. We consider various reasoning problems over core-closed KBs and study their combined and
Claudia Cauli, Magdalena Ortiz, and Nir Piterman
data complexity [
In our implementation of closed-world reasoning, core-closed KBs resemble safe KBs and are FOL rewritable, but the partial closed-world assumption plays an important role, particularly in the query satis ability problem that arises from the May queries. For future work, we are interested in including more complex knowledge in the Tbox while still keeping (data) complexity tractable. Complex role inclusions would be required to reason about data ow, which is a central aspect of security. Non-monotone extensions would be needed to be considered in order to reason about permissions and access policies.