Over the weekend, I’ve been thinking about last week’s disclosure concerning Debian’s OpenSSL package, which in effect stated that all keys and certificates generated by this compromised code have been trivially crackable since late 2006.
The scope of this vulnerability is pretty wide, and the ease with which a weak key can be compromised is significant. Ubuntu packaged up a weak key detector script containing an 8MB data block which, I’m told, included every single possible key value that the Debian OpenSSL package could conceivably create.
The question that kept cropping up for me is: This one-line code change apparently went unnoticed for well over a year. Why is it that crackers and script kiddies never found it and/or exploited it? Numerous exploits on Microsoft Windows would have required far more scrutiny and creativity than this one. Given the rewards involved for 0-day exploits, especially in creating platforms for cross-site scripting attacks, why is it nobody bothered to exploit this?
My hypothesis – sorry, my speculation is this: People at every stage of the production process and everywhere else in the system trusted that the others were doing their job competently. This includes crackers and others with a vested interest in compromising the code. I should exclude from this list those who might have a reasonable motivation to exploit the vulnerability with stealth and to leave no traces. If, however, even they didn’t notice the danger presented by this tiny but fundamental change in the code base, well my point becomes stronger.
The change itself was small, but not really obscure. It was located, after all, in the function that feeds random data into the encryption process. As Ben Laurie states in his blog, if any of the OpenSSL members had actually looked at the final patch, they would almost certainly have noticed immediately that it was non-optimal.
In all this time, apparently, nobody using Debian’s OpenSSL package has actually (or adequately) tested to see whether the Debian flavour of OpenSSL was as strong as it was supposed to be. That level of trust is nothing short of astounding. If in fact malware authors were guilty of investing the same trust in the software, then I’d venture to state that there’s a fundamental lesson to be learned here about human nature, and learning that lesson benefits the attacker far more than the defender:
Probe the most trusted processes first, because if you find vulnerabilities, they will yield the greatest results for the least effort.
P.S. Offhand, there’s one circumstance that I think could undermine the credibility of this speculation, and that’s if there’s any link between this report of an attack that compromised not less than 10,000 servers and the recent discovery of the Debian OpenSSL vulnerability.