Software-based methods for Operating system dependability

Abstract

Guaranteeing correct system behaviour in modern computer systems has become essential, in particular for safety-critical computer-based systems. However all modern systems are susceptible to transient faults that can disrupt the intended operation and function of such systems. In order to evaluate the sensitivity of such systems, different methods have been developed, and among them Fault Injection is considered a valid approach widely adopted. This document presents a fault injection tool, called Kernel-based Fault-Injection Tool Open-source (KITO), to analyze the effects of faults in memory elements containing kernel data structures belonging to a Unix-based Operating System and, in particular, elements involved in resources synchronization. This tool was evaluated in different stages of its development with different experimental analyses by performing Faults Injections in the Operating System, while the system was subject to stress from benchmark programs that use different elements of the Linux kernel. The results showed that KITO was capable of generating faults in different elements of the operating systems with limited intrusiveness, and that the data structures belonging to synchronization aspects of the kernel are susceptible to an appreciable set of possible errors ranging from performance degradation to complete system failure, thus preventing benchmark applications to perform their task. Finally, aiming at overcoming the vulnerabilities discovered with KITO, a couple of solutions have been proposed consisting in the implementation of hardening techniques in the source code of the Linux kernel, such as Triple Modular Redundancy and Error Detection And Correction codes. An experimental fault injection analysis has been conducted to evaluate the effectiveness of the proposed solutions. Results have shown that it is possible to successfully detect and correct the noxious effects generated by single faults in the system with a limited performance overhead in kernel data structures of the Linux kernel.