Re: [syzbot] [net?] WARNING in rtmsg_ifinfo_build_skb

[Hangbin Liu]

Hi Sabrina,

Wish you all well.

On Sat, Aug 15, 2026 at 01:14:01AM +0200, Sabrina Dubroca wrote:
> > diff --git a/drivers/net/macsec.c b/drivers/net/macsec.c
> > index ee0e2eb7dbc61c1affe4b0203337f2fdaeaa031c..0844f551d3be696bad72a8322fa4ea1bf9ccf105
> > 100644
> > --- a/drivers/net/macsec.c
> > +++ b/drivers/net/macsec.c
> > @@ -4402,7 +4402,7 @@ static int macsec_fill_info(struct sk_buff *skb,
> >                 csid = secy->xpn ? MACSEC_CIPHER_ID_GCM_AES_XPN_256 :
> > MACSEC_CIPHER_ID_GCM_AES_256;
> >                 break;
> >         default:
> > -               goto nla_put_failure;
> > +               return 0;
> >         }
> 
> It's not possible to create a macsec device with a key length other
> than those 2, but it's true that returning -EMSGSIZE in this case is
> not correct (-EINVAL + DEBUG_NET_WARN_ON_ONCE I guess).
> 
> The real problem is that we register the netdev before private data
> has been fully initialized (macsec_newlink calls register_netdevice,
> then macsec_add_dev), so we end up reaching fill_info before key_len
> has been set.
> 
> So I think we should run macsec_add_dev before registering (and maybe
> move those allocs to ndo_init), and possibly macsec_changelink_common
> too. Setting the sci was intentionally done late, so this bit should
> stay where it is.

IIRC, we talked about this issue some time ago, but I forgot to fix it.

Sorry, I also have to leave it to you to fix it. But feel free to tell me
if you are too busy.

Best Regards
Hangbin