Postmortem Intrusion Analysis

1629 Words4 Pages

1 Introduction

Today, postmortem intrusion analysis is an all too familiar problem. Our devices are repeatedly compro- mised while performing seemingly benign activities like browsing the Web [33], interacting on social- networking websites, or by malicious actors that use botnets as platforms for various nefarious activities [12]. Sometimes, the threats can also arise from the inside (e.g., corporate espionage) and often leading to sub- stantial financial losses1. Underscoring each of these security breaches is the need to reconstruct past events to know what happened and to better understand how a particular compromise may have occurred. Sadly, although there has been significant improvements in computer systems over the last few decades, data foren- sics remains a very tedious process; partly because the detailed information we require to reliably reconstruct events is simply not there when we need it the most [11].

Loosely speaking, recent efforts in data forensic research have focused on tracking changes to file system objects by using monitoring code resident in kernel space, or by making changes to the application binary interface. However, without proper isolation these approaches are subject to tampering and therefore can not provide strong guarantees with respect to the integrity of the recorded events. Malicious users can, for in- stance, inject code into either kernel or user space, thereby undermining the integrity of the logs maintained by the tracking mechanism. Virtualization [17] provides a potential avenue for enabling the prerequisite iso- lation criteria by providing a sandbox for operating system code and applications. For example, a hypervisor can mediate disk accesses at the block level by presenting a vir...

... middle of paper ...

...tions on L. We denote these operations as O. Any additional operations (e.g., create or delete) can be modeled as a combination of these base operations. We tie these accesses to the corresponding causal entity that made them, to ensure that a forensic analyst has meaningful semantic information for their exploration [4].

The approach we take to capture these causal relationships is based on an event-based model, where events are defined as accesses, O, on a location L caused by a some entity, i.e., Ei(O, L) → ID. Loosely speaking, an entity is modeled as the set of code pages resident in a process’ address space during an event. The distinct set of code pages belonging to that process is then mapped to a unique identifier. This event- based model also allows us to automatically record events that are causally related to each other, and to chainthesequencesofeventsas

More about Postmortem Intrusion Analysis

Open Document