Through this project, Coverity will demonstrate the effectiveness of automated source code analyses in improving the security and reliability of critical software. Coverity will develop and apply source code analysis checks that will automatically detect common causes of security vulnerabilities, such as buffer overflows, as well as application-specific security problems in various critical code bases. Coverity will also detect other types of software defects, including memory and resource leaks, which affect overall reliability. Although the idea of source code analysis is not new, the tools that have come to market are prohibitively difficult to use or fail to scale to the size of commercial applications. Coverity's solutions successfully overcome these barriers with novel, state-of-the-art technology. The prototype for Coverity's patent-pending source code analysis technology was originally developed by a team of researchers in the Computer Systems Lab at Stanford University. Initial applications of this technology in the research setting resulted in the successful detection of over 2000 defects and hundreds of exploitable security holes in the Linux and OpenBSD operating systems. Coverity's current analysis engine uses interprocedural data-flow analyses coupled with sophisticated abstraction techniques that may be used to uncover a wide range of software defects including ú Security holes ú Buffer overruns ú Memory corruption ú Memory/Resource leaks ú Deadlocks and race conditions ú API misuse ú Violations of coding standards Unlike other tools that have surfaced over the years, Coverity's solution scales to millions of lines of code, achieves 100% path coverage, and requires no manual modifications or testing. Coverity's technology can pinpoint hundreds to thousands of critical defects out of the box with orders of magnitude less noise than other source code analysis tools. A unique feature of Coverity's tool is its extensibility. Company specific rules, custom security policies, or errors detected by QA can be easily turned into checks that can be plugged into the analysis platform. Furthermore, statistical learning and data mining techniques can be used to automatically retarget the tool to different code bases with differing coding conventions, styles and idioms.
