At the edge of modern cyber-physical systems, Micro-Controller Units (MCUs) are responsible for safety-critical sensing/actuation. However, MCU cost constraints rule out the usual security mechanisms of general-purpose computers. Thus, various low-cost security architectures have been proposed to remotely verify MCU software integrity. Control Flow Attestation (CFA) enables a Verifier (Vrf) to remotely assess the run-time behavior of a prover MCU (Prv), generating an authenticated trace of all of Prv control flow transfers (CFLog). Further, Control Flow Auditing architectures augment CFA by guaranteeing the delivery of evidence to Vrf. Unfortunately, a limitation of existing CFA lies in the cost to store and transmit CFLog, as even simple MCU software may generate large traces. Given these issues, prior work has proposed static (context-insensitive) optimizations. However, they do not support configurable program-specific optimizations. In this work, we note that programs may produce unique predictable control flow sub-paths and argue that program-specific predictability can be leveraged to dynamically optimize CFA while retaining all security guarantees. Therefore, we propose SpecCFA: an approach for dynamic sub-path speculation in CFA. SpecCFA allows Vrf to securely speculate on likely control flow sub-paths for each attested program. At run-time, when a sub-path in CFLog matches a pre-defined speculation, the entire sub-path is replaced by a reserved symbol. SpecCFA can speculate on multiple variable-length control flow sub-paths simultaneously. We implement SpecCFA atop two open-source control flow auditing architectures: one based on a custom hardware design and one based on a commodity Trusted Execution Environment (ARM TrustZone-M). In both cases, SpecCFA significantly lowers storage/performance costs that are critical to resource-constrained MCUs.

翻译：暂无翻译