[lxc-devel] [PATCH 1/6] Move lxcbr0 setup logic into lxc.net script

[Serge Hallyn]

Quoting Michael H. Warfield (mhw at WittsEnd.com):
> On Mon, 2014-08-11 at 18:30 +0000, Serge Hallyn wrote:
> > Quoting Michael H. Warfield (mhw at WittsEnd.com):
> > > On Mon, 2014-08-11 at 15:37 +0000, Serge Hallyn wrote:
> > > > Quoting Michael H. Warfield (mhw at WittsEnd.com):
> > > > > On Thu, 2014-07-31 at 08:53 +0200, Martin Pitt wrote:
> > > > > > Factor this out of the lxc-net.conf upstart job, so that it can be used by
> > > > > > init.d scripts and systemd units, too.
> > > > > 
> > > > > Crap.  Never fails.  I was in Europe when this came out.
> > > > > 
> > > > > Looking at lxc.net, I would say it's going to break some existing setups
> > > > > (notably mine) where lxcbr0 is already setup.  Yes, we can set
> > > > > LXC_BRIDGE to no but we should also include some autodetect logic such
> > > > > that, if lxcbr0 already exists, this doesn't commit random acts of
> > > > > terrorism.
> > > 
> > > > That logic should already be there.  If /sys/class/net/lxcbr0 already
> > > > exists, then start will do nothing;  if /run/lxc/network_up does not
> > > > exist then stop will do nothing.
> > > 
> > > Yeah, I traced through that logic and finally realized that.  I'm a
> > > little nudgey about even calling "stop" from "start" in the case where
> > > the bridge already exists or when USE_LXC_BRIDGE is not true and then
> > > relying on "stop" to rely on the non-existence of network_up to exit
> > > without doing anything to the bridge.  If that's the case, why even call
> > > "stop" in the "start" function at all?  That had me confused.
> > 
> > Yeah, that's an upstart-ism which doesn't belong in the script (and
> > must continue to be done in upstart)
> > 
> > > I don't see the need for this logic at all in "start":
> > > 
> > > -- 
> > >         if [ -d /sys/class/net/${LXC_BRIDGE} ]; then
> > >                 if [ ! -f ${varrun}/network_up ]; then
> > >                         # bridge exists, but we didn't start it
> > >                         stop;
> > >                 fi
> > >                 exit 0;
> > >         fi
> > > -- 
> > > 
> > > If this bridge exists and we didn't start it, why even attempt to stop
> 
> > We don't - the stop is to tell upstart not to continue running the
> > job.  It doesn't belong in this script.
> 
> Interesting.  Many thanks for clarifying that.  If there are no
> objections, then, I'll clean that up with the patch I'm working on now.

That'd be great, thanks.  Indeed I think we want the actual script to
do the minimal start/stop, and as you suggested perhaps make the
iptables rules more configurable (though that may be biting off more
than we can chew for now).

BTW - will you be at the lxc sprint in chicago next week?  That might be
a good place to discuss making the network stuff more configurable.