NSF Award
1563393
Software is embedded into our daily activities. Ensuring that the software is trustworthy - does what is intended - and secure - is not vulnerable to attack - is a prime concern. Much attention has been devoted to establishing the correctness of high-level programs. This project is focused on the important task of ensuring that the, often complex and opaque, transformations carried out by a compiler do not degrade the trustworthiness and security guarantees of its input program.<br/><br/>The key innovation pursued in this project is self-certification which guarantees the correctness and security of compilation. A self-certifying compiler creates a tangible, independently-checkable proof, justifying the correctness of the compilation run. By linking in information from external analysis tools certificates can also aid in obtaining better machine code. In particular, they allow for automatic insertion of defensive measures, which protect the program from common security attacks. This work builds on existing theoretical ideas and compiler implementations, while extending them in new directions. The self-certifying compiler is implemented in the popular LLVM framework, making it suitable for immediate adoption by programmers, and its security benefits available to end users in a transparent fashion. Provable program correctness is a true "Grand Challenge" for computing. By developing both theory and implementation of a self-certifying compiler, this project is taking a significant step forward in meeting that challenge.
