Re: HEADSUP: OpenBSD dhclient incoming

On Thu, Jun 16, 2005 at 02:25:07PM -0600, Warner Losh wrote:
> From: Brooks Davis <brooks_at_one-eyed-alien.net>
> Subject: Re: HEADSUP: OpenBSD dhclient incoming
> Date: Thu, 16 Jun 2005 09:47:47 -0700
> 
> > On Wed, Jun 15, 2005 at 05:38:10PM -0700, Darren Pilgrim wrote:
> > > From: Brooks Davis
> > > > There are two issues here.  First, if we're going to keep
> > > > network_interfaces around, /etc/rc.d/dhclient should honor
> > > > it and not start dhclient on interfaces not in either
> > > > network_interfaces or removable interfaces.
> > > 
> > > I think network_interfaces should be gotten rid of entirely for two
> > > reasons:
> > > 
> > > 1: It creates a synchronization issue between it and the ifconfig_*
> > > lines and duplicates functionality.  IIRC, rc.conf being out of sync in
> > > this way has tripped up users in the past.
> > > 
> > > 2: There are real configurations in which some interfaces are not
> > > available when netif is run at boot.  One example is the many newer
> > > mini-PCI wireless NICs that require a firmware upload.  Devd is the
> > > accepted tool for performing such tasks, but rcordering puts devd after
> > > NETWORKING.  The actions taken by devd must therefore include steps
> > > taken by netif.  Calling the rc.d scripts directly from devd avoids
> > > local scripts that duplicate rc.d functionality.  A similar situation
> > > occurs for removable interfaces.
> > 
> > I'm seriously considering removing the following variables:
> > 
> > network_interfaces
> > removable_interfaces
> > pccard_ether
> 
> pccard_ifconfig you mean?
> 
> > If I do that I'll probably add a pair of new start/stop targets to
> > /etc/rc.d/netif to replicate the remaining functions of pccard_ether
> > and nuke /etc/pccard_ether.
> 
> I invented pccard_ifconfig to be done whenever a new device appeared
> on the system.  I test about 20 different drivers these days, and
> having that many different ifconfig_foo= lines was burdonsome.  I'd
> love to see the fallback functionality that this provided retained
> somehow.

Agreed.  There's actually some demand for a default for wired interfaces
as well.  If nothing else I know the EmuLab people would like one
so they didn't have to use their own scripts to configure all the
interfaces.  I've thought about it a bit.  I'm leaning towards a
default_ifconfig variable plus a new magic ifconfig_<if> option NONE to
cause the default option not to be used.

One other issue here is that I'm not sure what happens if you try to run
dhcp on a hardware point-to-point link.  At least with ppp(4) it just
fails and exits.

-- Brooks

-- 
Any statement of the form "X is the one, true Y" is FALSE.
PGP fingerprint 655D 519C 26A7 82E7 2529  9BF0 5D8E 8BE9 F238 1AD4