The Unix Operating System provides this funky system call called 'chroot' whereby root can change the root (or '/') directory to be any other directory on a mounted filesystem.
I use it myself when I have to boot from an install floppy of Linux, from then on working on the hard drive as if I had booted from it.
A chroot jail consists in executing a process after having executed chroot. It therefore does'nt see the rest of the filesystem. It helps secure otherwise faulty daemons such as sendmail or bind with such a bad security track record that you'd better watch your butt as much as possible. Be warned, though, that root can always exit from a chroot jail. All it takes is chroot ../../../../... This is reported as a security defect on linux-kernel every other month. According to Alan Cox, this is the correct behaviour, and would be impossible to fix anyway. The solution is to avoid running processes as root under a chroot, POSIX capabilities make it possible. It's also a good idea to mount the filesystem on which the chroot jail reside with option nosuid.
wu-ftpd, the most widely used FTP daemon, usually runs in a chroot jail for anonymous access.
Note that you usually need to provide a significant subset of the usual system directories in the chroot jail (such as /lib, /etc, /bin ...)