Re: [PATCH 10/27] wifi: mac80211: isolate driver from inactive links

[Wen Gong <quic_wgong@quicinc.com>]

On 4/18/2023 5:11 PM, Johannes Berg wrote:
> On Tue, 2023-04-18 at 16:59 +0800, Wen Gong wrote:
>> On 4/18/2023 4:15 PM, Johannes Berg wrote:
>>> On Mon, 2023-04-17 at 22:07 +0800, Wen Gong wrote:
>>>> OK. So I will try to put this in mac80211 layer, is it OK?
>>>>
>>> I guess? I'm still not really sure why you even want it, but hey, that's
>>> up to you in a way. I really didn't like the suggestion with
>>> wiphy_iftype_ext_capab (or any other capability for that matter), it
>>> feels like it should be more dynamic, like maybe a new "add link"
>>> callback or something? At least then you can't blame mac80211 for when
>>> it breaks when you have two 5 GHz links ...
>> ok, so I would like to add callback such as
>>
>> "add_link(struct ieee80211_hw *hw, struct ieee80211_vif vif, struct
>> ieee80211_bss_conf *link_conf, unsigned int link_id)"
>>
>> in struct ieee80211_ops, and mac80211 call it in
>> ieee80211_mgd_setup_link()/ieee80211_vif_update_links,
>>
>> then lower-drvier could dynamic set the local addr of assoc
>> link_conf(also for 2nd link_conf), is it OK?
>>
> Seems OK, but I'm not sure that _works_?
>
> After all, we first set the addresses in assoc_data, when we don't have
> a link_conf yet, no? Just what we were discussing in the other thread
> about the leak.
>
> johannes

It should work, I will test it later.

For the 1st assoc link, the data->u.mgd.assoc_data is empty in 
ieee80211_mgd_setup_link(),

because ieee80211_mgd_setup_link() is called from nl80211_authenticate() 
for the 1st assoc link.

So ieee80211_mgd_setup_link() use eth_random_addr() for the 1st assoc link.


For the 2nd link, ieee80211_mgd_setup_link() is called from 
nl80211_associate(),

the sdata->u.mgd.assoc_data is NOT empty,

and the sdata->u.mgd.assoc_data->link[link_id].addr is valid,

it is addr by eth_random_addr(assoc_data->link[i].addr) in 
ieee80211_mgd_assoc().