Re: Re: [B.A.T.M.A.N.] [PATCH] net: fix possible deadlocks in rtnl_trylock/unlock

[Eric Dumazet <eric.dumazet@gmail.com>]

On Sat, 2012-12-01 at 20:04 +0100, Sven Eckelmann wrote:
> On Saturday 01 December 2012 10:36:12 Eric Dumazet wrote:
>  
> > > diff --git a/net/bridge/br_sysfs_br.c b/net/bridge/br_sysfs_br.c
> > > index c5c0593..c122782 100644
> > > --- a/net/bridge/br_sysfs_br.c
> > > +++ b/net/bridge/br_sysfs_br.c
> > > @@ -142,7 +142,7 @@ static ssize_t store_stp_state(struct device *d,
> > > 
> > >  	if (!rtnl_trylock())
> > >  	
> > >  		return restart_syscall();
> > >  	
> > >  	br_stp_set_enabled(br, val);
> > > 
> > > -	rtnl_unlock();
> > > +	__rtnl_unlock();
> > > 
> > >  	return len;
> > >  
> > >  }
> > 
> > I have no idea of why you believe there is a problem here.
> > 
> > Could you explain how net_todo_list could be not empty ?
> > 
> > As long as no device is unregistered between
> > rtnl_trylock()/rtnl_unlock(), there is no possible deadlock.
> 
> I am not sure what "here" means for your. At least batman-adv tries to 
> unregister a device -> problem [1]. I will not make any judgements about the 
> other uses in the kernel/other parts patched by Simon.
> 

I was reacting to the change in net/bridge/br_sysfs_br.c

rtnl_trylock() could set a boolean flag to explicitly WARN_ON()
in case we try to unregister a device.