Cesarano, Carmine (2025) Software Attack Surface Reduction via Security Hardening, Fuzzing, and Runtime Enforcement. [Tesi di dottorato]

Documento PDF cesarano_carmine_38.pdf Visibile a [TBR] Amministratori dell'archivio Download (10MB) | Richiedi una copia

Abstract

Modern software systems integrate open-source and off-the-shelf components across multiple abstraction layers. These layers manage complexity through modular stacks of hardware, hypervisors, kernels, container runtimes, and orchestration frameworks. Each component exposes interfaces through which control or data cross trust boundaries. From Kubernetes APIs and inter-process channels to system calls and hypercalls, these boundaries collectively define the attack surface of modern computing systems. This dissertation addresses how to automatically and precisely reduce attack surfaces across system layers through combined static and dynamic analysis. It introduces novel techniques that infer, enforce, or evaluate least-privilege and minimization principles, either constraining privileges or exposing unsafe behaviors that enlarge the attack surface. Five complementary techniques embody this approach, each targeting a distinct surface. KubeFence enforces runtime Kubernetes API policies at the fine-grained level of resource specification fields, automatically learning allowed orchestration privileges. FuzzBox enables coverage-guided fuzzing of internal communication in closed-source binaries without compiler-time instrumentation, binary rewriting, or hardware tracing. IRIS introduces a record-and-replay method to efficiently explore deep hypervisor control flows, enabling targeted mutations at specific internal states to expose complex logic vulnerabilities. GoSurf presents a Go-specific taxonomy and static analyzer for supply chain attack vectors, guiding code reviewers toward high-risk components that can be exploited for arbitrary code execution. GoLeash prevents import-level supply chain attacks by enforcing allowed privileged capabilities per package rather than application-wide policies. These techniques demonstrate that automated attack surface reduction is effective across abstraction layers, reducing exploitable interfaces without sacrificing functionality in real-world systems. This dissertation advances the state of the art through a cross-layer methodology that automates assessment and enforcement of exposures, reduces manual effort, and provides a process adaptable to diverse software.

Downloads

Actions (login required)

Modifica documento