Re: [PATCH 2/2] mac80211: Support LIVE_ADDRESS_CHANGE feature

[Denis Kenzior <denkenz@gmail.com>]

Hi Johannes,

On 10/7/19 4:16 PM, Johannes Berg wrote:
> Hi,
> 
>>>> If you do care about this being more granular then you should check
>>>> *which* interface is scanning, and then you can still switch the
>>>> MAC
>>>> address for *other* interfaces - but I'd still argue it should be
>>>> independent of interface type.
>>
>> So yes these can scan, but this should be covered by the
>> netif_carrier_ok check which is done first.
> 
> Not sure what you mean by that.
> 
> You could have two interfaces, one which is scanning right now, right?
> And then theoretically you don't care about the other one - it *should*
> be OK to remove/re-add (with new MAC address) the one that *isn't*
> scanning, right?

Actually, I don't think you can?  Unless I'm missing something?  All the 
scan state is stored on struct ieee80211_local, so if that struct is 
allocated per phy as you point out below, then what you suggest is 
currently not possible?

> 
> But we don't have that granularity here for anything - you're just
> checking "sdata->local->something", and by going from sdata to local
> you've now checked the whole NIC, not just a single interface on that
> NIC.

Right.  But that seems to be a limitation of mac80211 actually.  We 
can't run two scans concurrently on different interfaces.  This is 
rather unintuitive given that scan requests require an ifindex/wdev.

Can this be changed / fixed in mac80211 actually?  I would expect that 
if a card supports p2p and station simultaneously, then it can scan / go 
offchannel on two interfaces simultaneously? Or not?  What can iwlwifi 
do for example?

> 
> Which is fine, no complaint from me, just in that case you end up
> failing when really there isn't much need to fail. In fact, in a case
> like this, actually clearing IFF_UP, changing address and setting IFF_UP
> would work, concurrently with another interface scanning.
> 
>>   We can just remove the
>> switch entirely, but the roc_list/scanning check only matters for
>> station/p2p_client so checking for the other interface types is kinda
>> pointless and redundant.
> 
> But it's also completely confusing to do it this way because you go from
> "sdata" to "local", and at that point the data that you're working on is
> no longer specific to that one interface, it's actually for the whole
> NIC.

I agree its confusing, but that seems to be how mac80211 works?

> 
> Basically what I'm saying is this: it's confusing and makes no sense to
> do something like
> 
> 	if (this_is_a_certain_netdev_type)
> 		check_some_global_data();
> 
>> Also I am not sure what you mean by *which* interface. This function is
>> called on a single interface, so checking what other interfaces are
>> doing seems strange...
> 
> My point exactly - but that's what you're doing here in the code. Now I
> think perhaps without even realizing?
> 

Given the above, I'm not sure I see anything wrong?  The switch/case can 
probably be gotten rid of, but it actually makes things clear that only 
station/p2p_device and adhoc are handled specially.

Regards,
-Denis