A team led by Harvard computer scientists, including two undergraduate students, has developed a new tool that could lead to increased security and enhanced performance for commonly used web and mobile applications.
Called RockSalt, the clever bit of code can verify that native computer programming languages comply with a particular security policy.
The use of native code, especially in an online environment, however, opens up the door to hackers who can exploit vulnerabilities and readily gain access to other parts of a computer or device. An initial solution to this problem was offered over a decade ago by computer scientists at the University of California, Berkeley, who developed software fault isolation (SFI).
SFI forces native code to “behave” by rewriting machine code to limit itself to functions that fall within particular parameters. This “sandbox process” sets up a contained environment for running native code. A separate “checker” program can then ensure that the executable code adheres to regulations before running the program.
While considered a major breakthrough, the solution was limited to devices using RISC chips, a processor more common in research than in consumer computing. In 2006, Morrisett developed a way to implement SFI on the more popular CISC-based chips, like the Intel x86 processor. The technique was adopted widely. Google modified the routine for Google Chrome, eventually developing it into Google Native Client (or “NaCl”).
When bugs and vulnerabilities were found in the checker for NaCl, Google sent out a call to arms. Morrissett once again took on the challenge, turning the problem into an opportunity for his students. The result was RockSalt, an improvement over NaCl, built using Coq, a proof development system.
“We built a simple but incredibly powerful system for proving a hypothesis—so powerful that it’s likely to be overlooked. We want to prove that if the checker says ‘yes,’ the code will indeed respect the sandbox security policy,” says undergraduate student Joseph Tassarotti, who built and tested a model of the execution of x86 instructions. “We wanted to get a guarantee that there are no bugs in the checker, so we set out to construct a rigorous, machine-checked proof that the checker is correct.”
Even more impressively, RockSalt comprises a mere 80 lines of code, as compared to the 600 lines of the original Google native code checker. The new checker is also faster, and, to date, no vulnerabilities have been uncovered.
“The biggest benefit may be that users can have more peace of mind that a piece of software works as they want it to,” says Morrisett. “For users, the impact of such a tool is slightly more tangible; it allows users to safely run, for example, games, in a web browser without the painfully slow speeds that translated code traditionally provides.”
Previous efforts to develop a robust, error-free checker have resulted in some success, but RockSalt has the potential to be scaled to software widely used by the general public. The researchers expect that their tool might end up being adopted and integrated into future versions of common web browsers. Morrisett and his team also have plans to adapt the tool for use in a broader variety of processors.