P2P: Improve robustness against lost ctrl::ack frames

P2P includes two use cases where one of the devices is going to start a
group and likely change channels immediately after processing a frame.
This operation may be fast enough to make the device leave the current
channel before the peer has completed layer 2 retransmission of the
frame in case the ctrl::ack frame was lost. This can result in the peer
not getting TX status success notification.

For GO Negotiation Confirm frame, p2p_go_neg_conf_cb() has a workaround
that ignores the TX status failure and will continue with the group
formation with the assumption that the peer actually received the frame
even though we did not receive ctrl::ack. For Invitation Response frame
to re-invoke a persistent group, no such workaround is used in
p2p_invitation_resp_cb(). Consequently, TX status failure due to lost
ctrl::ack frame results in one of the peers not starting the group.

Increase the likelihood of layer 2 retransmission getting acknowledged
and ctrl::ack being received by waiting a short duration after having
processed the GO Negotiation Confirm and Invitation Response frames for
the re-invocation case. For the former, use 20 ms wait since this case
has been worked around in deployed devices. For the latter, use 50 ms
wait to get even higher likelihood of getting ctrl::ack through since
deployed devices (and the current wpa_supplicant implementation) do not
have a workaround to ignore TX status failure.

20 ms is long enough to include at least couple of retries and that
should increase likelihood of getting ctrl::ack through quite a bit. The
longer 50 ms wait is likely to include full set of layer 2 retries.

Signed-hostap: Jouni Malinen <jouni@qca.qualcomm.com>

diff --git a/src/p2p/p2p_go_neg.c b/src/p2p/p2p_go_neg.c
index 9137aebe950740631ab20fa95fb61fa8bda7c328..f50618365d4598464a13c679ba069716aaee0a70 100644 (file)
--- a/src/p2p/p2p_go_neg.c
+++ b/src/p2p/p2p_go_neg.c
@@ -1191,5 +1191,17 @@ void p2p_process_go_neg_conf(struct p2p_data *p2p, const u8 *sa,
                return;
        }
 
+       /*
+        * The peer could have missed our ctrl::ack frame for GO Negotiation
+        * Confirm and continue retransmitting the frame. To reduce the
+        * likelihood of the peer not getting successful TX status for the
+        * GO Negotiation Confirm frame, wait a short time here before starting
+        * the group so that we will remain on the current channel to
+        * acknowledge any possible retransmission from the peer.
+        */
+       wpa_msg(p2p->cfg->msg_ctx, MSG_DEBUG, "P2P: 20 ms wait on current "
+               "channel before starting group");
+       os_sleep(0, 20000);
+
        p2p_go_complete(p2p, dev);
 }

diff --git a/wpa_supplicant/p2p_supplicant.c b/wpa_supplicant/p2p_supplicant.c
index 39924b17e2ca76a103a47f1ed8a7417e94238942..9e552c82ee3bef6079d6633fcfe52814db7f4972 100644 (file)
--- a/wpa_supplicant/p2p_supplicant.c
+++ b/wpa_supplicant/p2p_supplicant.c
@@ -2318,6 +2318,18 @@ static void wpas_invitation_result(void *ctx, int status, const u8 *bssid)
                return;
        }
 
+       /*
+        * The peer could have missed our ctrl::ack frame for Invitation
+        * Response and continue retransmitting the frame. To reduce the
+        * likelihood of the peer not getting successful TX status for the
+        * Invitation Response frame, wait a short time here before starting
+        * the persistent group so that we will remain on the current channel to
+        * acknowledge any possible retransmission from the peer.
+        */
+       wpa_dbg(wpa_s, MSG_DEBUG, "P2P: 50 ms wait on current channel before "
+               "starting persistent group");
+       os_sleep(0, 50000);
+
        wpas_p2p_group_add_persistent(wpa_s, ssid,
                                      ssid->mode == WPAS_MODE_P2P_GO,
                                      wpa_s->p2p_persistent_go_freq,