The FreeBSD kernel recently had a issue in the kenv(2) kernel call, and this article describes very well what it is – and why it is bad. The vulnerability itself is not terribly bad, but the problem exposed is a common one and shows how all user data must be vetted before it is used: a programmer must treat all user data as suspect.
In fact, there have been studies done by Professor Barton Miller at the University of Wisconsin showing that both commercial and open source programs (in a variety of operating systems) are vulnerable (to differing extents) to a constant barrage of random data.
If your code is to be secure, you absolutely must treat user data as hostile and unknown: any trust placed in the user will be abused by someone, either accidentally or purposefully. If by accident, the user will think your software broken and unreliable; if by purpose, your system (or someone else’s!) could be compromised.
Two excellent books on this topic (from two different angles) are these: Hacking: the Art of Exploitation (by Jon Erickson) and Secure Coding Principles (by Mark Graff and Kenneth van Wyk). The first will show you how broken code can be taken advantage of; the second will show you how not to write broken code.