Sendmail and secure design
Ancient code base and long vuln history
Comment Recently, Mark Dowd of ISS discovered a signal handling vulnerability in Sendmail. We don't see major bugs in software that's as popular as Sendmail very often (at least, in the Unix world anyways), and that's probably a good thing. According to sendmail.com, Sendmail still handles about 70 per cent of all email on the internet.
As far as software goes, Sendmail is ancient, dating all the way back to 1981. Sendmail 8 itself is well over 10 years-old. To put it nicely, its security track record is less than stellar. However, the last big show stoppers in Sendmail were found about three years ago – Zalewski's prescan() bugs reported in September and March of 2003, and crackaddr(), also in March of 2003. The crackaddr() bug was also discovered by Mark Dowd.
Although the obscurity of these issues varies, none of them are really basic. I'm sure Dowd's most recent discovery took him a while from start to finish. And I think it's great that companies like ISS are willing to support research like this. After all, there really are no guarantees that time spent auditing a piece of software will result in bugs (especially exploitable ones) being discovered.
Thanks to Mark Dowd and ISS, we now have one less vulnerability in Sendmail. At the least, finding the next vulnerability in Sendmail is going to be that much more difficult.
Killing and creating bugs
I've often wondered whether or not there are, in practice, a finite number of vulnerabilities in a given software. Obviously, there is no way to know how many undiscovered (or undisclosed) vulnerabilities lurk in any given piece of software. This means that when a new vulnerability is found and addressed, it's hard to understand what kind of impact it has on the future scrutiny of the product. There is one less bug in the software, and the hurdle standing in the way of the next researcher (or attacker) just gets a little bit taller. There's no way we'll know if this was the last significant bug in the code. It's almost a never-ending cycle of vulnerabilities.
For really simple software, the concept of never-ending vulnerabilities might not seem to have bearing. However, as the complexity of an application increases, this becomes more valid. One can imagine a huge machine with so many inputs and interactions that it's impossible to understand how everything works. When we can't understand something, it's bound to have vulnerabilities. Then, even at a practical level, it becomes impossible to have any sort of assurance about the software's security or its resilience to attacks. So, finding vulnerabilities in or attacking software becomes nothing more than an exercise in time and talent. Given all but the most trivial of programs, someone with enough time and skill can break it.
I am of the opinion that our only hope of eliminating bugs, or at least making further attacks against software impractical, is to keep systems small, simple, and static (by static, I mean that they don't change). And, although there can never be a guarantee of security, we can at the very least have some reasonable level of assurance. As that software continues to be scrutinized by researchers, our level of assurance increases.
On the other side of things, new bugs are being created all the time. With the growing size of software, and the ever-increasing number of "features" (wanted or not) that get introduced into some software, we're guaranteed to be creating new vulnerabilities all the time. This is just one of the reasons that many Unix-based operating systems have fared so much better than Windows: they're smaller, older, and change less. Looking around at the state of some other operating systems, I often wonder if we're actually moving backwards in security instead of forwards.