"Weird machines" in ELF: a spotlight on the underappreciated metadata

  • Authors:
  • Rebecca Shapiro;Sergey Bratus;Sean W. Smith

  • Affiliations:
  • Dartmouth College;Dartmouth College;Dartmouth College

  • Venue:
  • WOOT'13 Proceedings of the 7th USENIX conference on Offensive Technologies
  • Year:
  • 2013

Quantified Score

Hi-index 0.00

Visualization

Abstract

Although software exploitation historically started as an exercise in coaxing the target's execution into attacker-supplied binary shellcode, it soon became a practical study in pushing the limits of unexpected computation that could be caused by crafted data not containing any native code. We show how the ABI metadata that drives the creation of a process' runtime can also drive arbitrary computation. We introduce our design and implementation of Cobbler, a proof-of-concept toolkit capable of compiling a Turing-complete language into well-formed ELF executable metadata that get "executed" by the runtime loader (RTLD). Our proof-of-concept toolkit highlights how important it is that defenders expand their focus beyond the code and data sections of untrusted binaries, both in static analysis and in the dynamic analysis of the early runtime setup stages as well as any time the RTLD is invoked.