Over the past year multiple people have been engaging language maintainers and designers to change their use of CSPRNGs (mainly relying on user-land RNGs like the one from OpenSSL, and sometimes suggesting "adding entropy" by various means from user-land daemons like haveged). In this short presentation we'll survey the struggle of cryptographers, developers and security engineers to change the path various high-profile languages have taken to provide randomness to their userbase. Affected languages include but are not limited to: Ruby, node.js and Erlang. We outline better approaches for language maintainers and implementers as well as coming changes within the Linux kernel crypto subsystem (i.e. /dev/random and /dev/urandom) w.r.t. security and performance. Recently these changes were merged into mainline Linux (4), problems with languages implementations however remain. We'll also discuss operating system provided randomness testing, attacks/mitigation in embedded and virtualized environments.
#Software #Security