Casting exploit analysis as a Weird Machine reconstruction problem

Abstract

Exploits constitute malware in the form of application inputs. They take advantage of security vulnerabilities inside programs in order to yield execution control to attackers. The root cause of successful exploitation lies in emergent functionality introduced when programs are compiled and loaded in memory for execution, called ‘Weird Machines’ (WMs). Essentially WMs are unexpected virtual machines that execute attackers’ bytecode, complicating malware analysis whenever the bytecode set is unknown. We take the direction that WM bytecode is best understood at the level of the process memory layout attained by exploit execution. Each step building towards this memory layout comprises an exploit primitive, an exploit’s basic building block. This work presents a WM reconstruction algorithm that works by identifying pre-defined exploit primitive-related behaviour during the dynamic analysis of target binaries, associating it with the responsible exploit segment - the WM bytecode. In this manner any analyst familiar with exploit programming will immediately recognise the reconstructed WM bytecode’s semantics. This work is a first attempt at studying the feasibility of this method and focuses on web browsers when targeted by JavaScript exploits.