On Fri, 14 Sep 2007 08:41:07 -0600
ebiederm@xmission.com (Eric W. Biederman) wrote:

> Stephen Hemminger <shemminger@linux-foundation.org> writes:
> 
> > On Fri, 14 Sep 2007 11:39:32 +0400
> > Pavel Emelyanov <xemul@openvz.org> wrote:
> >
> >> I proposed introducing a list_for_each_entry_continue_reverse
> >> macro to be used in setup_net() when unrolling the failed
> >> ->init callback.
> >> 
> >> Here is the macro and some more cleanup in the setup_net() itself
> >> to remove one variable from the stack :) Minor, but the code
> >> looks nicer.
> >> 
> >> Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
> >
> > Maybe it is time to just eliminate the init hook from the API.
> > It has very few users, and there is no reason the setup needed
> > could be done before or after registering in most cases.
> 
> I guess only have 5 out of the 29 users I have in my full patchset
> is few.  But that is to be expected because so far only the core
> has been converted.
> 
> I looked again at the initialization to see if you had a point about
> the initialization but in every instance I looked at the function
> was performing work that needed to happen during the creation of
> each network namespace.  So the work very much needs to be done there.
> 
> Ok looking some more I can see why this isn't obvious yet.  copy_net_ns
> hasn't been merged yet, and that is where we create new network namespaces.
> And call setup_net on each new network namespace.
> 
> I will take a look at that patch and see if I can come up with a
> safe version of it to merge to allow for a little more transparency.

Could we just make it so dev->init is not allowed to fail? Then it
can be a void function and the nasty unwind code can go?