> I'm confused a little bit about the RNG in OpenBSD. if you read the following > comments in /sys/dev/rnd.c the author explains how to seed the PRNG after > a new boot to prevent predictable randomness: > sed -n '166,206p' /sys/dev/rnd.c | more > > Now, this process is done in /etc/rc the following way: (from -current) > sed -n '262,277p' /usr/src/etc/rc > > Now my question is the following. Since /etc/netstart and pf are started > before this procedure does that mean there is a tiny window upon startup > that randomness is predictable? pf uses a lot of arc4random and I'm sure > in the inet stack there is use of arc4random as well.
No. That seeding is suplimental. There is already some randomness, even if only a little bit, in the system from other sources. I also think that the quality of the entropy does not matter as much as the academics believe. Instead of using one consumer, OpenBSD is using hundreds of simultaneous consumers who are very greedy, and grab bytes all the time. This is cutting up the entropy pool into little slots which will make any other assesment of the data even more difficult.
