BUG/MINOR: quic: Useless probing retransmission in draining or killing state
authorFrédéric Lécaille <flecaille@haproxy.com>
Mon, 24 Apr 2023 09:26:06 +0000 (11:26 +0200)
committerFrédéric Lécaille <flecaille@haproxy.com>
Mon, 24 Apr 2023 09:46:33 +0000 (11:46 +0200)
The timer task responsible of triggering probing retransmission did not inspect
the state of the connection before doing its job. But there is no need to
probe the peer when the connection is in draining or killing state. About the
draining state, this is even forbidden.

Must be backported to 2.7 and 2.6.

src/quic_conn.c

index 007afee..43737ec 100644 (file)
@@ -5352,8 +5352,16 @@ struct task *qc_process_timer(struct task *task, void *ctx, unsigned int state)
        TRACE_ENTER(QUIC_EV_CONN_PTIMER, qc);
        TRACE_PROTO("process timer", QUIC_EV_CONN_PTIMER, qc,
                    NULL, NULL, &qc->path->ifae_pkts);
+
        task->expire = TICK_ETERNITY;
        pktns = quic_loss_pktns(qc);
+
+       if (qc->flags & (QUIC_FL_CONN_DRAINING|QUIC_FL_CONN_TO_KILL)) {
+               TRACE_PROTO("cancelled action (draining state)", QUIC_EV_CONN_PTIMER, qc);
+               task = NULL;
+               goto out;
+       }
+
        if (tick_isset(pktns->tx.loss_time)) {
                struct list lost_pkts = LIST_HEAD_INIT(lost_pkts);