Insider threat: memory confidentiality and integrity in the cloud

The advantages of always available services, such as remote device backup or data storage, have helped the widespread adoption of cloud computing. However, cloud computing services challenge the traditional boundary between trusted inside and untrusted outside. A consumer’s data and applications are no longer in premises, fundamentally changing the scope of an insider threat. This thesis looks at the security risks associated with an insider threat. Specifically, we look into the critical challenge of assuring data confidentiality and integrity for the execution of arbitrary software in a consumer’s virtual machine. The problem arises from having multiple virtual machines sharing hardware resources in the same physical host, while an administrator is granted elevated privileges over such host. We used an empirical approach to collect evidence of the existence of this security problem and implemented a prototype of a novel prevention mechanism for such a problem. Finally, we propose a trustworthy cloud architecture which uses the security properties our prevention mechanism guarantees as a building block. To collect the evidence required to demonstrate how an insider threat can become a security problem to a cloud computing infrastructure, we performed a set of attacks targeting the three most commonly used virtualization software solutions. These attacks attempt to compromise data confidentiality and integrity of cloud consumers’ data. The prototype to evaluate our novel prevention mechanism was implemented in the Xen hypervisor and tested against known attacks. The prototype we implemented focuses on applying restrictions to the permissive memory access model currently in use in the most relevant virtualization software solutions. We envision the use of a mandatory memory access control model in the virtualization software. This model enforces the principle of least privilege to memory access, which means cloud administrators are assigned with only enough privileges to successfully perform their administrative tasks. Although the changes we suggest to the virtualization layer make it more restrictive, our solution is versatile enough to port all the functionality available in current virtualization viii solutions. Therefore, our trustworthy cloud architecture guarantees data confidentiality and integrity and achieves a more transparent trustworthy cloud ecosystem while preserving functionality. Our results show that a malicious insider can compromise security sensitive data in the three most important commercial virtualization software solutions. These virtualization solutions are publicly available and the number of cloud servers using these solutions accounts for the majority of the virtualization market. The prevention mechanism prototype we designed and implemented guarantees data confidentiality and integrity against such attacks and reduces the trusted computing base of the virtualization layer. These results indicate how current virtualization solutions need to reconsider their view on insider threats.