Assessing The Robustness of Paravirtualized Systems

Abstract

Virtualization is a technology that allows the creation of virtual in- stances of physical devices such as network, storage or processing units. With hardware getting better and cheaper it is possible to have a single physical machine running di erent virtual machines, which reduces hardware and maintenance costs. Virtualized systems are governed by a hypervisor and resources are shared amongst virtual machines. Virtualization provides many bene ts but also introduces new challenges like security, availability and isolation. Paravirtualization di ers from full virtualization because it enables the optimization of the performance of some virtual machine opera- tions. To achieve that, the hypervisor provides a hypercall interface containing the operations that can be used by the kernel of a paravir- tualized guest. As hypercalls are used to execute sensitive operations, their abuse can lead to harmful e ects. Thus, the hypercall interface has to be robust and secure. The goal of this work is to propose a methodology, based on ro- bustness testing, to assess the hypercall interface of paravirtualized systems. Robustness testing is used to characterize the behavior of a system under unexpected input conditions and has long been used to assess applications in multiple domains. It makes it possible to ex- perimentally assess the robustness of the underlying infrastructure, but rst several challenges must be overcome. With this goal in mind, the hypercall interface of Xen was analyzed to understand which are the data types and domains of the func- tion input parameters. A set of mutations rules, which considers the identi ed data types and domains, is proposed in this work. This dissertation also presents the prototype of a tool capable of generat- ing and executing tests in an automated manner, which allows the systematic evaluation of the hypercall interface.