Data Sandboxing for Software Binaries

Data-rich applications, such as web servers, web browsers, and document editors, are inherently vulnerable to non-control data attacks, i.e., attacks that exploit memory corruption vulnerabilities to leak or corrupt sensitive data. We propose Dobby, a tool that statically transforms software binaries to integrate run-time checks ensuring illicit data flows caused by memory corruption errors in the program cannot be used to compromise integrity and confidentiality of the programs data. To minimize the impact of dynamic checking on software runtime, the tool will apply protections heterogeneouslystronger, and thus more expensive, protections will be used to secure the data that is security-sensitive, while low-cost, coarse sandboxing will protect the rest of the data. Additionally, Dobby will detect and isolate private data contexts in data-rich applications (e.g., independent requests handled by a web server or pages visited by a web browser). Dobby will be able to protect COTS and legacy binaries, for which source code is typically not available. In case there is access to buildable source code, the tool will provide the capability for leveraging compile-time information for added accuracy and efficiency.