Re: [PATCH] ath10k: add dynamic vlan support

[Manikanta Pubbisetty <mpubbise@codeaurora.org>]

On 2018-08-16 13:57, Johannes Berg wrote:
> On Tue, 2018-08-14 at 18:23 +0530, Manikanta Pubbisetty wrote:
> 
>> > I don't think that makes sense. If we split the capability of AP_VLAN
>> > and AP_VLAN_FOR_4ADDR at the "root", then we don't need to play with all
>> > these things. Yes, this is slightly awkward for userspace, and perhaps
>> > with the interface combination checks, but I'd like you to look at that.
> 
>> I was working on splitting the 4-addr functionality from AP/VLAN 
>> iftype;
>> I have introduced a new iftype NL80211_IFTYPE_AP_4ADDR and moved the
>> 4-addr handling from AP/VLAN to this new iftype. But this approach
>> breaks the backward compatibility with older userspace applications.
> 
> Yeah ...
> 
> I'm confused and no longer sure what I was thinking, nor even what 
> we're
> trying to achieve here...
> 

I had introduced a change db3bdcb9c3ff (" mac80211: allow AP_VLAN 
operation on crypto controlled devices ") for supporting VLAN 
functionality on ath10k devices; this commit has broken 4 addr support 
on ath10k devices as I was advertising the AP/VLAN support 
conditionally. Since 4 addr operation is tied to AP/VLAN support, with 
this change, only the chips which support VLAN functionality can support 
4 addr operation but other ath10k chips don't.

 From what I can understand from our previous discussions, we had to 
separate the 4addr support from the AP/VLAN iftype but doing so could 
lead to backward compatibility issues. I have the code which separates 
the 4addr functionality from AP/VLAN but the bigger problem is the 
backward compatibility.

I am hoping that now I have set the context correctly :)

>> Since I am completely moving the 4-addr handling to the new type, 
>> older
>> applications which do not understand this new type will simply fail 
>> and
>> 4-addr mode will be completely broken.
>> 
>> Currently, whenever a 4-addr client attempts a connection, hostapd 
>> just
>> creates a AP/VLAN interface and moves the 4-addr client to the AP/VLAN
>> iface; there are no other checks. I had no other option other than 
>> going
>> with a new iftype for 4-addr handling.
>> 
>> Is there a way we can maintain backward compatibility with this
>> approach? Retaining the 4-addr handling in AP/VLAN and duplicating the
>> same functionality to the new iftype seems will work but I am not sure
>> if this is the right approach.
> 
> I think we have to keep the 4-addr handling in AP_VLAN iftype either
> way, to not break existing hostapd. We could introduce a separate
> AP_VLAN_NO_4ADDR and then require updating hostapd to get non-4addr
> VLAN, but that also seems awkward.
> 
> Since hostapd doesn't currently check anything...
> 
> Ok, no, I'm confused now. If we just clear WIPHY_FLAG_4ADDR_AP, don't 
> we
> get what we want? 4-addr AP_VLAN interfaces would no longer be 
> permitted
> to be created?

As I explained above, the agenda is to provide the driver (in this case, 
ath10k) a better control for advertising the device capabilities; only 
few ath10k chips can support VLAN functionality but all of them can 
support 4 addr operation.

Manikanta