Swap and Play: Live Updating Hypervisors and Its Application to Xen

Hypervisors provide the means to run multiple isolated virtual machines on the same physical host. Typically, updating hypervisors requires a reboot of the host leading to disruption of services that is highly undesirable, particularly in cloud environments. Nevertheless, security updates have to be applied fast to reduce the risk of attacks, demanding a solution which eliminates the trade-off between availability and security risk. Live updating, in general, is highly challenging and has been investigated for decades. However, all solutions proposed so far require changes to the control flow of the software and/or cause performance degradation. Moreover, currently there are no solutions for live updating of hypervisors and all major products (e.g., Hyper-V, Xen, ESXi) require a reboot for updating. In this paper, we present Swap and Play, the first live update mechanism for hypervisors. Our solution is easy to use, scalable and, in particular, deployable in cloud environments. Our approach leverages the hypervisor’s small memory footprint to swap the hypervisor on-the-fly without affecting the control flow of the (new) hypervisor, or disrupting the guests. In this context, we tackle several technically involved challenges, such as transferring the state of the running hypervisor, updating the configuration of one CPU while reinitializing the configuration of all other CPUs, and passing the control to the new hypervisor, all at run-time. We implemented our approach on the popular Xen hypervisor to show its efficiency and effectiveness