Cross-VM network attacks & their countermeasures within cloud computing environments

Saeed, Atif and Garraghan, Peter (2020) Cross-VM network attacks & their countermeasures within cloud computing environments. PhD thesis, Lancaster University.

Text (Cross-VM Network Attacks & their Countermeasures within Cloud Computing Environments)
2019AtifPhD.pdf - Published Version
Available under License Creative Commons Attribution-NonCommercial-NoDerivs.
Download (14MB)

Abstract

Cloud computing is a contemporary model in which the computing resources are dynamically scaled-up and scaled-down to customers, hosted within large-scale multi-tenant systems. These resources are delivered as improved, cost-effective and available upon request to customers. As one of the main trends of IT industry in modern ages, cloud computing has extended momentum and started to transform the mode enterprises build and offer IT solutions. The primary motivation in using cloud computing model is cost-effectiveness. These motivations can compel Information and Communication Technologies (ICT) organizations to shift their sensitive data and critical infrastructure on cloud environments. Because of the complex nature of underlying cloud infrastructure, the cloud environments are facing a large number of challenges of misconfigurations, cyber-attacks, root-kits, malware instances etc which manifest themselves as a serious threat to cloud environments. These threats noticeably decline the general trustworthiness, reliability and accessibility of the cloud. Security is the primary concern of a cloud service model. However, a number of significant challenges revealed that cloud environments are not as much secure as one would expect. There is also a limited understanding regarding the offering of secure services in a cloud model that can counter such challenges. This indicates the significance of the fact that what establishes the threat in cloud model. One of the main threats in a cloud model is of cost-effectiveness, normally cloud providers reduce cost by sharing infrastructure between multiple un-trusted VMs. This sharing has also led to several problems including co-location attacks. Cloud providers mitigate co-location attacks by introducing the concept of isolation. Due to this, a guest VM cannot interfere with its host machine, and with other guest VMs running on the same system. Such isolation is one of the prime foundations of cloud security for major public providers. However, such logical boundaries are not impenetrable. A myriad of previous studies have demonstrated how co-resident VMs could be vulnerable to attacks through shared file systems, cache side-channels, or through compromising of hypervisor layer using rootkits. Thus, the threat of cross-VM attacks is still possible because an attacker uses one VM to control or access other VMs on the same hypervisor. Hence, multiple methods are devised for strategic VM placement in order to exploit co-residency. Despite the clear potential for co-location attacks for abusing shared memory and disk, fine grained cross-VM network-channel attacks have not yet been demonstrated. Current network based attacks exploit existing vulnerabilities in networking technologies, such as ARP spoofing and DNS poisoning, which are difficult to use for VM-targeted attacks. The most commonly discussed network-based challenges focus on the fact that cloud providers place more layers of isolation between co-resided VMs than in non-virtualized settings because the attacker and victim are often assigned to separate segmentation of virtual networks. However, it has been demonstrated that this is not necessarily sufficient to prevent manipulation of a victim VM’s traffic. This thesis presents a comprehensive method and empirical analysis on the advancement of co-location attacks in which a malicious VM can negatively affect the security and privacy of other co-located VMs as it breaches the security perimeter of the cloud model. In such a scenario, it is imperative for a cloud provider to be able to appropriately secure access to the data such that it reaches to the appropriate destination. The primary contribution of the work presented in this thesis is to introduce two innovative attack models in leading cloud models, impersonation and privilege escalation, that successfully breach the security perimeter of cloud models and also propose countermeasures that block such types of attacks. The attack model revealed in this thesis, is a combination of impersonation and mirroring. This experimental setting can exploit the network channel of cloud model and successfully redirects the network traffic of other co-located VMs. The main contribution of this attack model is to find a gap in the contemporary network cloud architecture that an attacker can exploit. Prior research has also exploited the network channel using ARP poisoning, spoofing but all such attack schemes have been countered as modern cloud providers place more layers of security features than in preceding settings. Impersonation relies on the already existing regular network devices in order to mislead the security perimeter of the cloud model. The other contribution presented of this thesis is ‘privilege escalation’ attack in which a non-root user can escalate a privilege level by using RoP technique on the network channel and control the management domain through which attacker can manage to control the other co-located VMs which they are not authorized to do so. Finally, a countermeasure solution has been proposed by directly modifying the open source code of cloud model that can inhibit all such attacks.