“Vulnerability Detection in Device Drivers”

From Navigators

Jump to: navigation, search

Manuel Mendonça (advised by Nuno Ferreira Neves)

Ph.D. dissertation, Departamento de Informática, Faculdade de Ciências, Universidade de Lisboa, Mar. 2017

Abstract: The constant evolution in electronics lets new equipment/devices to be regularly made available on the market, which has led to the situation where common operating systems (OS) include many device drivers (DD) produced by very diverse manufactures. Experience has shown that the development of DD is error prone, as a majority of the OS crashes can be attributed to flaws in their implementation. This thesis addresses the challenge of designing methodologies and tools to facilitate the detection of flaws in DD, contributing to decrease the errors in this kind of software, their impact in the OS stability, and the security threats caused by them. This is especially relevant because it can help developers to improve the quality of drivers during their implementation or when they are integrated into a system. The thesis work started by assessing how DD flaws can impact the correct execution of the Windows OS. The employed approach used a statistical analysis to obtain the list of kernel functions most used by the DD, and then automatically generated synthetic drivers that introduce parameter errors when calling a kernel function, thus mimicking a faulty interaction. The experimental results showed that most targeted functions were ineffective in the defence of the incorrect parameters. A reasonable number of crashes and a small number of hangs were observed suggesting a poor error containment capability of these OS functions. Then, we produced an architecture and a tool that supported the automatic injection of network attacks in mobile equipment (e.g., phone), with the objective of finding security flaws (or vulnerabilities) in Wi-Fi drivers. These DD were selected because they are of easy access to an external adversary, which simply needs to create malicious traffic to exploit them, and therefore the flaws in their implementation could have an important impact. Experiments with the tool uncovered a previously unknown vulnerability that causes OS hangs, when a specific value was assigned to the TIM element in the Beacon frame. The experiments also revealed a potential implementation problem of the TCP-IP stack by the use of disassociation frames when the target device was associated and authenticated with a Wi-Fi access point. Next, we developed a tool capable of registering and instrumenting the interactions between a DD and the OS. The solution used a wrapper DD around the binary of the driver under test, enabling full control over the function calls and parameters involved in the OS-DD interface. This tool can support very diverse testing operations, including the log of system activity and to reverse engineer the driver behaviour. Some experiments were performed with the tool, allowing to record the insights of the behaviour of the interactions between the DD and the OS, the parameter values and return values. Results also showed the ability to identify bugs in drivers, by executing tests based on the knowledge obtained from the driver’s dynamics. Our final contribution is a methodology and framework for the discovery of errors and vulnerabilities in Windows DD by resorting to the execution of the drivers in a fully emulated environment. This approach is capable of testing the drivers without requiring access to the associated hardware or the DD source code, and has a granular control over each machine instruction. Experiments performed with Off the Shelf DD confirmed a high dependency of the correctness of the parameters passed by the OS, identified the precise location and the motive of memory leaks, the existence of dormant and vulnerable code.

Download paper

Download Vulnerability Detection in Device Drivers

Export citation



Research line(s): Fault and Intrusion Tolerance in Open Distributed Systems (FIT)

Personal tools
Navigators toolbox