TL;DR
This paper introduces a novel hypervisor-based bus system designed to enforce usage control policies across different layers of a system. By enabling communication between usage control monitors at various abstraction levels, it addresses the challenge of tracking and controlling data usage consistently, even as data transforms into different representations.
What Problem Does This Paper Address?
Data usage control is crucial for protecting sensitive information after access has been granted. However, enforcing these controls becomes complex because data exists in multiple representations across different system layers (e.g., a file, a window's content, a network packet). When data flows between these representations, usage control policies need to be applied consistently. Existing approaches often focus on single layers, making cross-layer enforcement difficult. This paper tackles the problem of how to effectively communicate policies and track data flow across these diverse representations to ensure comprehensive usage control.
Methods
The authors propose a "virtualized usage control bus system" implemented at the hypervisor level. This approach leverages virtualization to create a secure and isolated communication channel. Instead of relying on OS-level inter-process communication (IPC) mechanisms like D-Bus, which can be vulnerable to tampering, their system uses a type-1 hypervisor (NOVA) to host a dedicated bus component.
Key aspects of their method include:
- Hypervisor-based Bus: The bus component runs at the hypervisor level, providing isolation from the guest OS and its potential vulnerabilities.
- Virtual Device Model: The bus is integrated as a virtual device within the hypervisor, minimizing modifications to the guest OS and its applications. This allows usage control monitors (e.g., a Firefox monitor, an OS monitor) to interact with the bus without needing direct OS-level IPC.
- Cross-Layer Communication: The system facilitates communication between monitors residing in different layers of abstraction. For instance, a Firefox monitor can notify an OS monitor about a data flow, enabling the OS monitor to enforce policies on file access.
- Secure Communication Protocol: A defined protocol handles registration of monitors, policy storage and retrieval, and notification messages, ensuring secure and tamper-resistant message exchange.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Hypervisor as a Security Enabler: Implementing the communication infrastructure at the hypervisor level offers a robust security boundary, making it significantly harder for users within the guest OS to tamper with the usage control enforcement mechanisms.
- Layered Abstraction is Key: The paper strongly advocates for a layered approach to usage control, recognizing that data's form and context change, necessitating monitors at each relevant layer to work in concert.
- Virtual Device Model for Integration: Using a virtual device model for the bus component is an elegant way to integrate the communication infrastructure into existing virtualization platforms with minimal guest OS modifications.
- Performance Trade-offs: While virtualization can introduce overhead, the authors demonstrate that their hypervisor-based solution can outperform traditional OS-level IPC mechanisms like D-Bus in certain scenarios, especially when considering security requirements.
Key Results
The authors evaluated their system's performance by comparing it to D-Bus in three scenarios: their UCbus in NOVA, native D-Bus, and D-Bus within NOVA.
- Performance: The NOVA UCbus implementation (Scenario 1) consistently showed the fastest message exchange times across various bundle sizes, outperforming both native D-Bus (Scenario 2) and D-Bus within NOVA (Scenario 3). For example, transmitting 1000 message bundles took approximately 65 seconds for NOVA UCbus, compared to significantly longer times for the D-Bus scenarios. This indicates that their hypervisor-based approach meets the "minimal performance overhead" requirement effectively.
- Security: The system is designed to be secure against attacks aimed at modifying messages or disrupting communication. By operating at the hypervisor level and using a virtual device model, it limits attack vectors available to a user within the guest OS. They also discuss potential attacks and countermeasures, emphasizing the importance of kernel integrity and trusted boot mechanisms.
Limitations and Future Work
- Monitor Authentication: While the system provides a secure communication channel, the authentication of usage control monitors themselves remains an open question.
- Platform Dependence: The current implementation is tied to the NOVA hypervisor and Intel VT-x. Future work includes porting the infrastructure to other operating systems (Windows, macOS) and potentially other hypervisors.
- Cross-VM Communication: Extending the infrastructure to connect monitors across different virtual machines on the same or different platforms is a significant area for future research.
- Information Flow Models: Developing more sophisticated information flow models is needed for automatic generation of notification messages and for specifying policies that handle representations across multiple layers.
- Deployment Strategies: Investigating effective ways to deploy this infrastructure to data consumers is also a future goal.
Summary updated