Novel Techniques of Using Diversity in Software Security and Information Hiding

Diversity is an important and valuable concept that has been adopted in many fields to reduce correlated risks and to increase survivability. In information security, diversity also helps to increase both defense capability and fault tolerance for information systems and communication networks, where diversity can be adopted from many different perspectives. This dissertation, in particular, focuses mainly on two aspects of diversity – the application software diversity and the diversity in data interpretation. Software diversity has many advantages over mono-culture in improving system security. A number of previous researches focused on utilizing existing off the shelf diverse software for network protection and intrusion detection, many of which depend on an important assumption – the diverse software utilized in the system is vulnerable only to different exploits. In the first work of this dissertation, we perform a systematic analysis on more than 6,000 vulnerabilities published in 2007 to evaluate the extent to which this assumption is valid. Our results show that the majority of the vulnerable application software products either do not have the same vulnerability, or cannot be compromised with the same exploit code. Following this work, we then propose an intrusion detection scheme which builds on two diverse programs to detect sophisticated attacks on security-critical data. Our model learns the underlying semantic correlation of the argument values in these programs, and consequently gains more accurate context information compared to existing schemes. Through experiments, we show that such context information is effective in detecting attacks which manipulate erratic arguments with comparable false-positive rates. Software diversity does not only exist on desktop and mainframe computers, it also exists on mobile platforms like smartphone operating systems. In our third work in this dissertation, we propose to investigate applications that run on diverse mobile platforms (e.g., Android and iOS) and to use them as the baseline for comparing their security architectures. Assuming that such applications need the same types of privileges to provide the same functionality on different mobile platforms, our analysis of more than 2,000 applications shows that those executing on iOS consistently ask for more permissions than their counterparts running on Android. We additionally analyze the underlying reasons and find out that part of the permission usage differences is caused by third-party libraries used in these applications. Different from software diversity, the fourth work in this dissertation focuses on the diversity in data interpretation, which helps to defend against coercion attacks. We propose Dummy-Relocatable Steganographic file system (DRSteg) to provide deniability in multi user environments where the adversary may have multiple snapshots of the disk content. The diverse ways of interpreting data in the storage allows a data owner to surrender only some data and attribute the unexplained changes across snapshots to the dummy data which are random bits. The level of deniability offered by our file system is configurable by the users, to balance against the resulting performance overhead. Additionally, our design guarantees the integrity of the protected data, except where users voluntarily overwrite data under duress. This dissertation makes valuable contributions on utilizing diversity in software security and information hiding. The systematic evaluation results obtained for mobile and desktop diverse software are important and useful to both research literature and industrial organizations. The proposed intrusion detection system and steganographic file system have been implemented as prototypes, which are effective in protecting valuable user data against adversaries in various threat scenarios