This new paper which is about to appear later this month (May, 2006) on the IEEE security and privacy conference describes holes in Linux's random number generator, as well as a clear description of the Linux /dev/random.
The Linux random number generator is part of the kernel of all Linux distributions and is based on generating randomness from entropy of operating system events. The output of this generator is used for almost every security protocol, including TLS/SSL key generation, choosing TCP sequence numbers, and file system and email encryption.
Although the generator is part of an open source project, its source code (about $2500$ lines of code) is poorly documented, and patched with hundreds of code patches.
We used dynamic and static reverse engineering to learn the operation of this generator. This paper presents a description of the underlying algorithms and exposes several security vulnerabilities. In particular, we show an attack on the forward security of the generator which enables an adversary who exposes the state of the generator to compute previous states and outputs. In addition we present a few cryptographic flaws in the design of the generator, as well as measurements of the actual entropy collected by it, and a critical analysis of the use of the generator in Linux distributions on disk-less devices.
Our main results are: 1. Clear description of the Linux /dev/random (Which was far from trivial and very complex).
2. Attack on /dev/random forward security (which is mostly theoretic at the moment because it requires break-in to the computer but is very simple to mount and break an important block in almost any crypt algorithm).
3. Concerning findings about the Linux software engineering process regarding security.
The Linux moderator of /dev/random did not answer our attempts at contacting him.