Black-, grey-, and white-box side-channel programming for software integrity checking

Abstract

Checking software integrity is a fundamental problem of system security. Many approaches have been proposed trying to enforce that a device runs the original code. Software-based methods such as hypervisors, separation kernels, and control flow integrity checking often rely on processors to provide some form of separation such as operation modes and memory protection. Hardware-based methods such as remote attestation, secure boot, and watchdog coprocessors rely on trusted hardware to execute attestation code such as verifying memory content and examining signatures appearing on buses. However, many embedded systems do not possess such sophisticated capabilities due to prohibitive hardware costs, unacceptably high power consumption, or the inability to update fielded components. Further, security assumption may become invalid as time goes by. For Systems-on-Chip (SoCs), in particular, internal activities cannot be observed directly, while in non-SoCs, sniffing bus traffic between constituent components may suffice for integrity checking. A promising approach to check software integrity for resource-constrained SoCs is through side-channels. Side-channels have been used mostly for attacks, such as eavesdropping from vibration of glass or plant leaves, fingerprinting machines from traffic patterns, or extracting secret key materials of cryptographic routines using power consumption measurements. In this work, side-channels are used to enhance rather than undercut security. First, we study the relationships between the internal states of a target device and side-channel information. We use the uncovered relationships to monitor the internal state of a running device and determine whether the internal state is an expected one. An unexpected state may be a sign of incorrect execution or malicious activity. To further explore the possibilities inherent in side-channel-based software integrity checking, we investigate various hardware platforms, representative of different degrees of knowledge of the hardware from the side-channel profiling point of view. In other words, side-channel information is extracted by black-, grey-, and white-box analysis. Each one involves unique challenges requiring different techniques to successfully derive “side-channel profiles”. We can use these profiles to detect unexpected states with extremely high probability, even when an adversary knows that their code may be subject to side-channel analysis, i.e., the methodology is robust to side-channel-aware adversaries. The research includes: (1) Constructing systematic approaches for black- and grey-box profiling of side channels (and comparing them to white-box analysis); (2) Designing custom measurement instrumentation; and (3) Developing techniques for monitoring and enforcing software integrity utilizing side-channel profiles. We introduce the term “side-channel programming” to refer to techniques we design in which developers explicitly utilize side-channel characteristics of existing hardware to optimize run-time software integrity checking, creating executable code which is more conducive to side-channel-based monitoring. Compared with other software integrity checking techniques, our approach has numerous benefits. Among them are that the measurement process is non-invasive, non-interruptive, and backward-compatible in that it does not require any hardware modification, meaning our approach works with processors that do not include security features. Our method can even be used to augment existing protection mechanism, as it works even when all security mechanisms internal to the device fail.