Re: [PATCH net] hsr: broadcast netlink notifications in the device's net namespace

[Jakub Kicinski]

On Wed, 27 May 2026 13:11:57 +0200 Fernando Fernandez Mancera wrote:
> >   	genlmsg_end(skb, msg_head);
> > -	genlmsg_multicast(&hsr_genl_family, skb, 0, 0, GFP_ATOMIC);
> > +	genlmsg_multicast_netns(&hsr_genl_family, dev_net(master->dev),
> > +				skb, 0, 0, GFP_ATOMIC);
> > +	rcu_read_unlock();
> >     
> 
> The patch makes sense to me. Anyway, I think we can reduce the RCU 
> critical section here.
> 
> What about moving rcu_read_unlock() after the if and extract the net 
> before unlocking? The benefit is minimal but I think it is worth it.

Not sure TBH, we'd need to take a ref on the netns and allocate 
a tracker (on DEBUG kernels). One could go either way.

I'm replying because I wanted to question whether this is Fixes+stable@ 
worthy. Sending the notifications to the namespace where the device is
makes sense. But it's as much a behavior changes as it is a fix.
The commit in question was merged to 5.6, real users clearly don't care.