Search the Community

Facebook has announced the availability of Pysa (Python Static Analyzer), an open-source tool designed for the static analysis of Python code. The security-focused tool relies on Pyre, Facebook’s type checker for Python, and allows for the analysis of how data flows through code. It can be used to identify issues related to the protection of user data, as well as flaws such as XSS and SQL injection. In addition to making Pysa available in open source, Facebook released many of the definitions that it leverages when looking for security bugs, making it readily available for others to start analyzing their own Python code. The tool also leverages open source Python server frameworks, including Django and Tornado, and this makes it usable for code analysis right from the start. Furthermore, only few lines of code are needed to use Pysa for additional frameworks, Facebook says. Pysa allows users to define sources of origin for important data and places where that data should not reach, which are called sinks. The tool then identifies functions that return data from a source and those that reach a sink and, if it discovers a connection between a source and a sink, it reports the issue. The tool was designed in such a manner that it avoids false negatives, thus supposedly identifying as many security issues as possible. This, however, results in more false positives, and, to remove these as well, Facebook’s engineers added sanitizers and features into the tool. The social media platform admits that Pysa has its limitations “based on its choice to address security issues related to data flow, together with design decisions that trade off performance for precision and accuracy.” Furthermore, Pysa was designed only for the discovery of data flow–related security issues, meaning that it won’t identify security or privacy issues that cannot be modeled as flows of data. Although nearly half of the results returned in the timeframe were false positives, Facebook was able to tune Pysa up, and says that it eventually returned “100 percent valid issues.” Via securityweek.com