svn commit: r230230 - head/sys/dev/random

[Andrey Chernov]

On Sun, Jan 22, 2012 at 04:59:55PM +0000, Mark Murray wrote:
> Andrey Chernov writes:
> > > The usual way round this is with a flag. Set a static, volatile
> > > flag, defaulting "off", and set it to "on" when the seeding has
> > > happened. Then arc4random() can do the right thing, depending on
> > > this flag.
> >
> > Ok, what about this version, is it right? libkern/arc4rand.c is not a
> > module but always present in the kernel, so "arc4rand_iniseed_state"
> > will be always accessible.
> >
> > --- dev/random/randomdev_soft.c.old	2011-09-26 07:35:48.000000000 +0400
> > +++ dev/random/randomdev_soft.c	2012-01-21 01:41:37.000000000 +0400
> > @@ -55,6 +55,8 @@ __FBSDID("$FreeBSD: src/sys/dev/random/r
> >  
> >  #define RANDOM_FIFO_MAX	256	/* How many events to queue up */
> >  
> > +extern int arc4rand_iniseed_state;
> > +
> 
> Should be in a header file, nad _possibly_ should be volatile. If it
> works without being volatile, then OK.

It was preliminary patch just to confirm/deny my understanding of your 
idea.

I'll put it into header.
In the final version I also plan to move that lines
+               if (arc4rand_iniseed_state == 1)
+                       arc4rand_iniseed_state = -1;
into arc4_randomstir() where they will be protected with mutex lock, so 
volatile will be not needed. It will be more logical, because other 
reseeding conditions are resetted there too.

> The rest is OK. I've not tested it, so this is not a review, simply an
> "OK" :-)

Thanx for review! I'll send final version to this thread a bit 
later when I'll find more free time.

-- 
http://ache.vniz.net/