Re: [PATCH net-next 2/2] net: reorder dev_addr_sem lock

[Stanislav Fomichev]

On 03/11, Kuniyuki Iwashima wrote:
> From: Stanislav Fomichev <sdf@xxxxxxxxxxx>
> Date: Tue, 11 Mar 2025 01:45:07 -0700
> > diff --git a/net/core/rtnetlink.c b/net/core/rtnetlink.c
> > index 9355058bf996..c9d44dad203d 100644
> > --- a/net/core/rtnetlink.c
> > +++ b/net/core/rtnetlink.c
> > @@ -3080,21 +3080,32 @@ static int do_setlink(const struct sk_buff *skb, struct net_device *dev,
> >  		struct sockaddr *sa;
> >  		int len;
> >  
> > +		netdev_unlock_ops(dev);
> > +
> > +		/* dev_addr_sem is an outer lock, enforce proper ordering */
> > +		down_write(&dev_addr_sem);
> > +		netdev_lock_ops(dev);
> > +
> >  		len = sizeof(sa_family_t) + max_t(size_t, dev->addr_len,
> >  						  sizeof(*sa));
> >  		sa = kmalloc(len, GFP_KERNEL);
> >  		if (!sa) {
> > +			up_write(&dev_addr_sem);
> >  			err = -ENOMEM;
> >  			goto errout;
> >  		}
> >  		sa->sa_family = dev->type;
> >  		memcpy(sa->sa_data, nla_data(tb[IFLA_ADDRESS]),
> >  		       dev->addr_len);
> 
> Can we move down_write() and netdev_lock_ops() here ?

Should be doable, yes, will also remove that up_write from the !sa
error condition. Will do, thanks for the review!