TL;DR
This paper highlights the critical importance of accurate and trustworthy time for a wide range of applications, from financial trading to IoT. It argues that current systems are vulnerable to various timing attacks that can compromise data integrity and system functionality, and proposes requirements for a secure time architecture.
What Problem Does This Paper Address?
In today's interconnected world, many applications rely on precise timing for their correct operation. This includes telecommunications, satellite navigation, banking, and the Internet of Things (IoT). Attackers can exploit vulnerabilities in how systems manage time to gain monetary advantages (e.g., high-frequency trading), steal digital content, disrupt forensic analysis, or even impersonate physical presence. The paper identifies a significant gap: while existing security mechanisms like encryption and trusted execution environments (TEEs) protect data, they often fail to guarantee the integrity of time itself. This leaves systems susceptible to attacks that manipulate time, leading to incorrect operations and system instability.
Methods
The authors analyze the timing capabilities of various Trusted Execution Environments (TEEs) such as Intel SGX, ARM TrustZone, and Trusted Platform Module (TPM). They examine how these TEEs interact with the broader "time stack," which includes hardware timers, system software (like the OS), and network time packets. The paper details various timing attacks, including:
- Hardware timer manipulation: A compromised OS can directly alter hardware timer registers.
- Software attacks: A malicious OS can lie about the current time, signal timeouts prematurely or late, or manipulate timekeeping threads.
- Network delays: Attackers can intercept and delay network time packets (e.g., from NTP servers), making the received time untrustworthy.
Based on this analysis, the paper outlines a set of requirements for a secure time architecture, emphasizing the need to protect all layers of the time stack.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Time is a fundamental security primitive: The paper strongly emphasizes that time is not just a convenience but a critical component for security and correct system operation, akin to data integrity or confidentiality.
- TEEs are not a silver bullet for time security: While TEEs offer some protection for timers, they are not immune to attacks. A privileged OS can still manipulate time by delaying access to TEE-provided time services or by affecting network packets that TEEs rely on.
- The entire time stack must be secured: The paper argues that securing only one layer (e.g., hardware timers) is insufficient. A holistic approach is needed to protect hardware timers, system software, and network time synchronization mechanisms.
- Existing TEEs have limitations for timekeeping: Intel SGX's trusted time service, for instance, offers coarse resolution and lacks a sense of absolute time, making it unsuitable for time-triggered applications that require precise synchronization across distributed systems.
- New attack vectors emerge with TEEs: The paper demonstrates how even with TEEs, a malicious OS can launch "delay attacks" on SGX time packets, distorting the perceived time for applications within the enclave.
Key Results
The paper provides a detailed examination of vulnerabilities in current time services, particularly within TEEs. It demonstrates that SGX's trusted time can be attacked by a malicious OS through delay attacks on time packets, leading to accumulated errors of multiple seconds. The authors also highlight the limitations of TPMs for broader timekeeping needs due to their resolution and access latency. The paper concludes by proposing requirements for a secure time architecture, including a high-resolution trusted timer, a robust scheduling policy, domain-specific solutions for OS delays, and network attack-tolerant synchronization.
Limitations and Future Work
The authors acknowledge that current TEE mechanisms are insufficient to establish timing integrity. They suggest that future work should focus on designing a secure time architecture that addresses the identified vulnerabilities. This includes developing mechanisms to detect and mitigate OS scheduling attacks, delay attacks, and network attacks on time synchronization. The paper implicitly calls for further research into more robust and precise trusted time services within TEEs and improved network time synchronization protocols that are resilient to sophisticated attacks.
Summary updated