Re: [patch net-next-2.6] net: allow notifier subscribers to forbid device from closing

[Stephen Hemminger <shemminger@vyatta.com>]

On Wed, 31 Aug 2011 16:25:51 +0100
Ben Hutchings <bhutchings@solarflare.com> wrote:

> On Wed, 2011-08-31 at 17:15 +0200, Jiri Pirko wrote:
> > In some situations, like when the device is used as slave device in
> > bond/br/etc it is not nice if someone closes the device. This allows
> > it's masters to forbid this closure.
> 
> No it doesn't.
> 
> [...]
> > @@ -1269,9 +1282,12 @@ static int dev_close_many(struct list_head *head)
> >  	struct net_device *dev, *tmp;
> >  	LIST_HEAD(tmp_list);
> >  
> > -	list_for_each_entry_safe(dev, tmp, head, unreg_list)
> > +	list_for_each_entry_safe(dev, tmp, head, unreg_list) {
> >  		if (!(dev->flags & IFF_UP))
> >  			list_move(&dev->unreg_list, &tmp_list);
> > +		else
> > +			__dev_pre_close(dev);
> > +	}
> >  
> >  	__dev_close_many(head);
> 
> The return value is ignored here.
> 
> And this is called from dev_close(), where you are adding the
> notification as well.  So the notifier will usually be called twice.
> 
> [...]
> > @@ -1397,6 +1418,7 @@ rollback:
> >  				break;
> >  
> >  			if (dev->flags & IFF_UP) {
> > +				nb->notifier_call(nb, NETDEV_PRE_DOWN, dev);
> >  				nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
> >  				nb->notifier_call(nb, NETDEV_DOWN, dev);
> >  			}
> [...]
> 
> The return value has to be ignored here.  Not sure it makes any sense to
> call the notifier at all.
> 
> Ben.
> 

Also we need to allow rmmod'ing a network device even it is
part of a bridge and that implicitly
calls close.