Evaluating the Applicability of Robustness Testing in Virtualized Environments

Virtualization provides many benefits as server consolidation and cost reduction, but it also introduces new challenges like security and isolation. Thus, trust is still one of the roadblocks in their adoption in critical systems. Virtualized systems are governed by a hypervisor and resources are shared amongst virtual machines. Paravirtualization improves the performance of the costly I/ O operations, by providing an hypercall interface to the guests' kernel. Hypercalls must be robust and secure, as their abuse leads to harmful effects. This paper presents an assessment of the applicability of robustness testing to the Xen hypercall interface. For this, we devised a testing campaign by mutating valid hypercall invocations with invalid values. The campaign was then executed from a compromised machine inserted in a representative virtualization environment. The results revealed the compromised machine being crashed frequently, in some cases without notification, and also lead into inconsistent states. Results also show the inadequacy of the approach: new failure mode scales are necessary, as well as new mechanisms for failure detection.