In the last few weeks I used a wpa_supplicant with my patches for tx and
rx over nl80211 control port and sometimes observed longer time to
successful connection than without these patches. I think that this is a
result of wpa_supplicant using SOCK_WIFI_STATUS when transmitting EAPOL
frames over AF_PACKET sockets and using a fast retransmit, if the Wi-Fi
ack is not observed timely. So the question here is, should we add an
analog feature for tx over nl80211 control port? I think of a
control_port_tx_status message with a cookie delivered to user-space.
Regards,
Markus
On Mon, 2020-02-03 at 14:50 +0100, Markus Theil wrote:
> In the last few weeks I used a wpa_supplicant with my patches for tx and
> rx over nl80211 control port and sometimes observed longer time to
> successful connection than without these patches. I think that this is a
> result of wpa_supplicant using SOCK_WIFI_STATUS when transmitting EAPOL
> frames over AF_PACKET sockets and using a fast retransmit, if the Wi-Fi
> ack is not observed timely. So the question here is, should we add an
> analog feature for tx over nl80211 control port? I think of a
> control_port_tx_status message with a cookie delivered to user-space.
Somehow I thought we treated this like the mgmt tx and already had a
notification over nl80211 going back, but looks like not?
Then yeah, we should definitely add that.
johannes
On 03.02.20 15:31, Johannes Berg wrote:
> On Mon, 2020-02-03 at 14:50 +0100, Markus Theil wrote:
>> In the last few weeks I used a wpa_supplicant with my patches for tx and
>> rx over nl80211 control port and sometimes observed longer time to
>> successful connection than without these patches. I think that this is a
>> result of wpa_supplicant using SOCK_WIFI_STATUS when transmitting EAPOL
>> frames over AF_PACKET sockets and using a fast retransmit, if the Wi-Fi
>> ack is not observed timely. So the question here is, should we add an
>> analog feature for tx over nl80211 control port? I think of a
>> control_port_tx_status message with a cookie delivered to user-space.
> Somehow I thought we treated this like the mgmt tx and already had a
> notification over nl80211 going back, but looks like not?
>
> Then yeah, we should definitely add that.
I'll take a second look on it and send a patch if necessary, but as far
as I have seen, no such mechanism exists.
Markus
> johannes
>