Linux Kernel Module API

Conceptually, writing a Linux kernel module is as simple as supporting two functions: init_module() and cleanup_module(), and then compiling into object code (with the appropriate defines). However, when considering what is actually necessary in order to initialize KORBit, and the full ramifications of being a Linux kernel module, things become a bit stickier.

In a nutshell, initializing KORBit requires doing enough initialization so that an IOR can be written to /proc, and launching a new kernel thread in order to service requests to the ORB. This first step requires creating and binding a TCP socket, and then performing the gyrations to actually write a file into /proc. This process is best described by the document ``Linux Kernel Module Programming Guide''⁵, by Ori Pomerantz. This document provides some sample code for initializing the requisite data structures, and setting up the callback function necessary for /proc.

While all of the above is simply a matter of driving the Linux kernel, we encounter problems when attempting to support the nuances of kernel modules. Linux kernel modules support the notion of ``reference counting''. But considering the timeframe for our project we will leave as a future enhancement.

When KORBit is loaded, it will set it's reference counter to some non-zero value, and mark its VISITED flag. From this point on, these values will not be changed. The only way to unload KORBit will be manually (via rmmod).