My research is about automatically making binary programs more secure (given different attack models). In particular, I have worked on the compiler-level generation of binary code to output programs that are less vulnerable to timing-based side channel attacks, and on rewriting binary code at link-time to make it more resistant against patch-based attacks. I have also worked on other techniques to protect software against different types of attacks: Multi-Variant Execution Environments (MVEEs) to protect software against exploits, anti-debugging techniques to delay some forms of reverse engineering, etc. I am also interested in finding vulnerabilities, and in making operating systems more secure.
Protection against patch-based attacks. The problem of patch-based attacks is that of Patch Tuesday/Exploit Wednesday. Here a security update is released, after which hackers try to find the vulnerability in the unpatched program by analyzing the difference between the original version of the program and the patched version. If this can be done fast enough, attackers can exploit this vulnerability with the mass of users that has not yet applied this patch. Applying software diversity increases the effort for attacker to find the original vulnerability, thus increasing the time window in which a user running unpatched code is safe from such exploitation. I try to mitigate these using software diversity. This is a technique where an application instance is transformed into syntactically different, but semantically equivalent instance. I do this at link time using the binary link time rewriting framework Diablo.
The first step is to evaluate the effectiveness of diversification strategies in slowing down an attacker. Assuming that an attacker will use existing tools to pinpoint the changes in the code, I compared different binary diffing tools and studied different ways in which they can be used to measure the effectiveness of existing diversification strategies.
I then improved the existing diversification techniques in order to deliver a similar slow-down for the attacker, but with a lower overhead in execution time of the patched program. This is done by using the binary diffing tools to compare which code fragments are still matched, and using this information in a feedback loop to iteratively transform only code that is still matched. I worked for the Aspire FP7 project, in which I've extended these techniques.
Protection against side-channel attacks. Differences in the execution time of a program can give an attacker additional information on the internal state of a cryptographic algorithm, potentially leading to the compromise of secret information. I researched mitigating side-channels on modern x86 processors. I worked on a compiler-based toolflow to apply if-conversion to cryptographical code in order to eliminate control-flow related side-channels. Furthermore, using the division instruction as an example of a variable-latency instruction on the x86 architecture, I evaluated different techniques to mitigate data-flow related timing side-channels.