Re: [PATCH v2] mac80211: reject station addition if AP or MLO link is inactive

On Mon, 2026-08-03 at 15:09 +0200, Slawomir Stepien wrote:
> On lip 31, 2026 14:40, Johannes Berg wrote:
> > On Mon, 2026-07-27 at 15:26 +0200, Slawomir Stepien wrote:
> > > After taking a closer look at this I wonder how AP_VLAN should be han=
dled? Is there a way on
> > > cfg80211 level to be able to check AP_VLAN's main device state?
> >=20
> > Hmm, yeah that might be an issue? I guess in theory it could track it,
> > but it doesn't.
> >=20
> > > My code change would look something like this:
> > >=20
> > > @@ -9404,6 +9404,12 @@ static int nl80211_new_station(struct sk_buff =
*skb, struct genl_info *info)
> > >         case NL80211_IFTYPE_AP:
> > >         case NL80211_IFTYPE_AP_VLAN:
> > >         case NL80211_IFTYPE_P2P_GO:
> > > +               /* Add new station only after the AP and link has bee=
n started */
> > > +               int link =3D params.link_sta_params.link_id >=3D 0 ?
> > > +                          params.link_sta_params.link_id : 0;
> > > +               if (!wdev->links[link].ap.beacon_interval)
> > > +                       return -ENETDOWN;
> > > +
> > >                 /* ignore WME attributes if iface/sta is not capable =
*/
> > >                 if (!(rdev->wiphy.flags & WIPHY_FLAG_AP_UAPSD) ||
> > >                     !(params.sta_flags_set & BIT(NL80211_STA_FLAG_WME=
)))
> > >=20
> > > but the NL80211_IFTYPE_AP_VLAN case would not work here, right? Can I=
 just do the checking only for
> > > NL80211_IFTYPE_AP and NL80211_IFTYPE_P2P_GO and skip NL80211_IFTYPE_A=
P_VLAN?
> >=20
> > Yeah it'd just reject everything, I guess.
> >=20
> > I think in practice stations are added at the AP interface first and
> > then moved to a VLAN, which would argue for actually rejecting
>=20
> But are they moved and then this `case NL80211_IFTYPE_AP_VLAN` should mat=
ch?

I'm not entirely sure, but I believe they're _added_ to the AP first,
and then _modified_ to be in the AP_VLAN via another path
(NL80211_CMD_SET_STATION, NL80211_ATTR_VLAN_ID).

The question is whether or not we can rely on this, i.e. no userspace
ever used the other path.

> In my tests this case doesn't match, even after the STA is assoc with vla=
n interface (e.g.
> wlan0.10).

Right.

> My test is done using qemu with:

I'd probably use UML/hwsim tests, but sure.

> This index 4 is assigned to wlan0, so it will never have iftype =3D=3D NL=
80211_IFTYPE_AP_VLAN.
> Also the print of wdev->identifier will give 0x1 and not 0x2.
>=20
> So, is it even possible that `case NL80211_IFTYPE_AP_VLAN` would match? I=
f it will never match, then
> do I understand that the change I proposed should be correct for other ca=
ses?

Well, it's _possible_, right? It just appears to not be used in
practice, as I described above (guess your test says I'm right).

But then I think we should do two changes:

1) simply refuse adding stations to AP_VLAN, i.e. drop the
   NL80211_IFTYPE_AP_VLAN case entirely from the function

2) make the change you suggested validating the link


Actually only validating the link as you showed would basically achieve
(1) with -ENETDOWN (we've been using -ENOLINK btw) rather than with the
-EOPNOTSUPP error from the default switch case, but it also seems less
explicit...

Or maybe just do both in one commit?

johannes