Date of Award

4-1-2018

Document Type

Thesis (Ph.D.)

Department or Program

Department of Computer Science

First Advisor

Sergey Bratus

Second Advisor

Sean Smith

Abstract

The software chain of trust starts with a chain of loaders. Software is just as reliant on the sequence of loaders that ultimately setup its runtime environment as it is on the libraries with which it shares its address space and offloads tasks onto. Loaders, and especially bootloaders, act as the keystone of trust, and yet their formal security properties -- which should be a part of any solid bootloader design -- are both underappreciated and not well understood. This is especially problematic given the increasing adoption of loader-based code signing and execution enforcement mechanisms. My thesis digs deeply into how loaders have failed to earn our trustworthiness and how they may continue to harbor vulnerabilities even after memory corruption-based vulnerabilities lose their prevalence. In order to address these issues, I propose a memory region-based type system that allows us to better model a loader's intentions and thus mediate its behavior. More specifically, I show how a loader's execution can be broken down into a sequence of typed phases, each semantically classified as either a bookkeeping, loading, or a patching substage, while sections of memory are grouped into semantically related regions and assigned a type, based on their intended use, by which policy access decisions are made. I demonstrate the feasibility of this technique by applying it to Das U-Boot, a well-known and widely-used bootloader, with minimal changes to the bootloader's implementation. In order to do so, I designed and developed an extensive bootloader instrumentation suite to help analyze a bootloader's behaviors, construct a policy, and completely mediate operations, thereby enforcing behaviors governed by the type system's policy.

Comments

Originally posted in the Dartmouth College Computer Science Technical Report Series, number TR2018-863.

Share

COinS