System auditing is a central concern when investigating and responding to security incidents. Unfortunately, attackers regularly engage in anti-forensic activities after a break-in, covering their tracks from the system logs in order to frustrate the efforts of investigators. While a variety of tamper-evident logging solutions have appeared throughout the industry and the literature, these techniques do not meet the operational and scalability requirements of system-layer audit frameworks.

In this work, we introduce Custos, a practical framework for the detection of tampering in system logs. Custos consists of a tamper-evident logging layer and a decentralized auditing protocol. The former enables the verification of log integrity with minimal changes to the underlying logging framework, while the latter enables near real-time detection of log integrity violations within an enterprise-class network. Custos is made practical by the observation that we can decouple the costs of cryptographic log commitments from the act of creating and storing log events, without trading off security, leveraging features of off-the-shelf trusted execution environments. Supporting over one million events per second, we show that Custos' tamper-evident logging protocol is three orders of magnitude (1000×) faster than prior solutions and incurs only between 2% and 7% runtime overhead over insecure logging on intensive workloads. Further, we show that Custos' auditing protocol can detect violations in near real-time even in the presence of a powerful distributed adversary and with minimal (3%) network overhead. Our case study on a real-world APT attack scenario demonstrates that Custos forces anti-forensic attackers into a "lose-lose" situation, where they can either be covert and not tamper with logs (which can be used for forensics), or erase logs but then be detected by Custos.

View More Papers

Metal: A Metadata-Hiding File-Sharing System

Weikeng Chen (UC Berkeley), Raluca Ada Popa (UC Berkeley)

Read More

MACAO: A Maliciously-Secure and Client-Efficient Active ORAM Framework

Thang Hoang (University of South Florida), Jorge Guajardo (Robert Bosch Research and Technology Center), Attila Yavuz (University of South Florida)

Read More

CloudLeak: Large-Scale Deep Learning Models Stealing Through Adversarial Examples

Honggang Yu (University of Florida), Kaichen Yang (University of Florida), Teng Zhang (University of Central Florida), Yun-Yun Tsai (National Tsing...

Read More

Detecting Probe-resistant Proxies

Sergey Frolov (University of Colorado Boulder), Jack Wampler (University of Colorado Boulder), Eric Wustrow (University of Colorado Boulder)

Read More