How does one implement a universal hash function, and would this be feasible to use, for example, in a filesystem's dentry table?

In Exploiting Unix File-System Races via Algorithmic Complexity Attacks[1], a vulnerability is described which amounts to lengthening the time a call to access(2) or open(2) takes by generating a large number of hash collisions in the dentry table. In personal conversations with Johnson, he told me that if the kernel devs would get their act together and implement a universal hash function in their hashtable implementation, this attack would be effectively neutralized (at least, with high probability). In my own searching, I have only found high-level, mathematical descriptions of universal hash functions. Are they feasible to implement, and do there exist implementations in the wild I can look at? [1]: http://www.cs.sunysb.edu/~rob/papers/races2.pdf
Answer:

Universal hash functions are not hard to implement. Just dot-product with a random vector or evaluate as a polynomial at a random point. However, you need to be careful in using them to fight complexity attacks. An adversary who can make careful timing measurements of your system might find the hash key, and then use it to make just as many collisions. You might try a pseudorandom function instead. These are pretty easy to implement, especially given the AES instructions in modern CPUs. But if you don't have these instructions, they're somewhat slower than UHFs.

Michael Hamburg at Quora Visit the source

Was this solution helpful to you?