svn commit: r273958 - head/sys/dev/random

[Mark R V Murray]

> On 2 Nov 2014, at 12:41, Dag-Erling Smørgrav <des at des.no> wrote:
> 
> Mark R V Murray <mark at grondar.org> writes:
>> I’m scared witless of this being on-by-default, for the reason given
>> in the removed comment. I’d much prefer to see it only turned on if a
>> kernel option is set, and the embedded folks /et al/ can use that.
> 
> You didn't seem to mind this code when we introduced it in 10-CURRENT.
> Removing it breaks pretty much everything, not just embedded systems.
> We can add a sysctl to turn it off, but it has to be on by default.

I’ve had a closer look at things, and I’m coming round to your side.

Note that this has NO effect on Fortuna. Fortuna’s self-starting appears
to be more reliable.

> Note that the alternative is to feed more trash into /dev/random at
> boot, as we did before.  It may give us a warm and fuzzy feeling which
> we don't get from automatically seeding, but the reality is that we have
> no idea how good that trash is either.  In fact, most of what we used to
> feed into /dev/random at boot (ps, sysctls etc) was constant or nearly
> so.  I prefer to trust that we get enough entropy from attachtimes and
> I/O in the boot process - and the data I gathered indicates that there
> is more than enough entropy from attachtimes alone, even on SFF systems
> and VMs.

OK, Fair enough. :-)

>> Moving the point of the auto-firstseed to where is good, thanks.
> 
> ...except that I'm not sure it doesn't break root-on-geli etc, but at
> least it doesn't break it more than not having auto-firstseed at all.

M
-- 
Mark R V Murray