On 2010/08/16 08:46, Nicholas Marriott wrote:
> cvsps is absolutely invaluable for this sort of stuff.

it is, but our changes to BIND are in areas which have seen quite
a few changes upstream.

> 
> On Mon, Aug 16, 2010 at 08:12:53AM +0200, Denis Fondras wrote:
> > Hello,
> > 
> > Following my previous message from July, 18th, I am back to BIND as
> > my tests with nsd/unbound are not really conclusive (can't make both
> > work with only one IP and they don't support views).

single IP -> you could bind nsd to 127.0.0.1, and let unbound bind
to other addresses on the system. have unbound forward-zone the
relevant zones to the local nsd. or for simple setups, "local-zone"
and "local-data" might be enough.

as always with this configuration, you can end up serving bad
data to resolver clients if a domain moves away before you change
config.

views -> it's a hack, but you can run multiple copies on alternative
ports and use PF to redirect users to the correct instance based on
source address.

> > So I rolled up my sleeves and started to port OpenBSD changes to
> > BIND-9.7.1-P2. Changing str-functions to strl-functions was the easy
> > part :)
> > Unfortunately, I have a hard time with privileges separation and
> > port randomization. In fact I don't know where to place them.
> > I made a diff between OpenBSD version and BIND-9.4.2-P2 and tried to
> > port it to BIND-9.7.1-P2 but it seems there was a huge change in
> > socket and pidfile handling.
> > 
> > Is anyone willing to help understanding these changes ?

unless you really understand the areas that have been modified
in OpenBSD I feel you would be better served by making a straight
port of a more recent bind 9. without full understanding, changes
in this area are more likely to weaken than strengthen things...

Reply via email to