ASLR to extreme

I was reading about Artificial Immune Systems (more about that in another post) and in one of the papers the statement was that biological systems increase resiliency by diversity. Furthermore, they give a contra example in computer networks in which Internet Explorer (at the time the paper was written) had 90% market share. It's obvious that when something hits IE, it hits almost the whole Internet. This isn't diversity by any standard.

I think that we have such problems with security in general that we need some new, radical solution. Probably, we are long way from that solution, but it occurred to me that this is exactly what is necessary, diversity that will disallow attackers from influencing single computers and thus large parts of the Internet. Still, it is hard to expect there will be N producers of operating systems, then N of browsers, etc. It's not easy to produce those, it takes long time and huge resources. Now, biological systems are much much older and theoretically it could be that in some distant future there will be such diversity. IMHO, this is questionable, and as I said it's theoretically in some distant future, which is why it is beyond the point. What we need is something that works now.

If you think a bit what we need is a mutation, that will change computer systems, from the bottom up in unpredictable ways. On the bottom I'm thinking about parts of a single application, while on the top I think of the complex systems consisting of computers and networks. Furthermore, this mutation has to be specific to each system so that there are hardly two similar systems in existence. So, for example, the computer you work on isn't similar to any other computer in use, and, as you use it, it evolves and mutates.

Now, why I mentioned Address Space Layout Randomization (ASLR) in the title? Because it seems to me to be a step in the direction of totally mutating everything. Namely, ASLR mutates address space of the process thus making it unpredictable for attackers and making each systems different. This mutation unfortunately, is restricted because it is too coarse grained, i.e. you move whole libraries, but not functions, of even blocks of the code from which functions are built.

Of course there are problems. For a start, similarity is a key to maintenance of systems. Companies having a large number of computers try hard to make them equal, just to lower maintenance costs. Not only that, developers count on similarity to be able to reproduce bugs, and consequently to correct them. So, those requirements should either be kept in a new system (which in part is contradictory) or new ways of achieving the same effect (i.e. maintainability).

Finally, mutation has to be dynamic. Namely, even if attacker gets into one system, or part of the system he needs time to discover other parts of the system. If mutation is quick enough, the knowledge that attacker obtains will be worthless before he manages to use it. Not only that, but potentially what he already achieved will evaporate soon.