Fix excessive reseeding in random(4).

[markm]

Summary

Fix braino of mine where the reseeds would happen far too often, making the kernel process way too busy.

PR: 230808
Submitted by: Conrad Meyer <[e-mail]>
Reported by: Danilo Egea Gondolfo <[e-mail]>

Test Plan

Make the PRNG work hard while doing something intensive like "make
world". To make the PRNG work hard, do e.g.
$ dd if=/dev/random of=/dev/null bs=$((1024*1024))

[markm]

Add the differential revision to the commit message.

[jmg]

Please commit these patches separately. They deal w/ different issues.

sys/dev/random/random_harvestq.c

222 ↗

(On Diff #47165)

this still results in excessive seeding. It caps the read rate to pools*8*8, or putting 64 bytes per pool in for every 4 bytes read. As most things will read a minimum of 16 bytes (128 bits), the smallest expected value for read_rate is 5 (the increment isn't a proper ceil function), or 40 bytes per pool for every read. With the number of pools, that's a lot of extra work still.

[markm]

Split into two separate commits as per review request.

[delphij]

LGTM. cem@ or markm@, could you please commit this before code freeze?

[cem]

(Approved — phab won't let me mark it as such because it is "closed.")

@jmg , @markm : Thanks, I also prefer the two independent pieces split into separate patches.

This one looks obviously correct to me. Go ahead and commit it. It fixes a security weakness in our implementation relative to spec Fortuna, so I'd be sure to tag the commit with "Security: yes."

[cem]

In D16866#359249, @markm wrote:

Split into two separate commits as per review request.

Where did the other review end up? Thanks.