netpfil/ipfw: use arc4random() instead of random()

[pfg]

Summary

This makes better use of the unsigned values improving the randomness.
It should make little, if any, difference from a user view point though.

[cem]

sys/netpfil/ipfw/ip_dn_io.c

408–417

Forwarding my remarks from the earlier review: I don't believe unpredictability matters here.

I also don't believe there is any significant benefit outside of unpredictability for this use. (Although, yikes, random(9) is actually rand(3), not random(3). Perhaps we should replace random(9) with a better PRNG.)

I may be mistaken on either or both counts.

514–515

This is now half as likely to happen, due to the differing ranges of the random(9) and arc4random(9) functions. I believe that violates the intent of the code.

[cem]

sys/netpfil/ipfw/ip_dn_io.c

408–417

I'll add: if we do actually want arc4random-quality entropy, and only need 16 bits, we can waste a few less cycles by using fewer bytes: arc4random_buf(&q->random, 2) (paired with apropriate resizing of dn_queue::random from int to uint16_t).

This is basically insignificant with today's arc4random(9) but would be more significant if this was a significant consumer of arc4random(9) and we adopted the buffered approach arc4random(3) uses today.

[pfg]

Add some developers that are more likely to know the code better.

[pfg]

I am abandoning this line of changes: it doesnt bring any advantage.