[ovs-discuss] Thread Safety in rconn

[Ben Pfaff]

On Thu, Aug 29, 2013 at 04:35:05PM +0200, Dirk van der Merwe wrote:
> #7  0x000000000043c3ba in connmgr_send_packet_in (mgr=0x213a100,
> pin=0x7f1822424d50)
>     at ofproto/connmgr.c:1437
>         ofconn = 0x2167290
> #8  0x000000000041daec in ofproto_dpif_send_packet_in (ofproto=0x213b940,
> pin=0x7f1822424d50)
>     at ofproto/ofproto-sdn.c:578
> No locals.
> 
> The second thread shown above is the one that generated the packet_in
> message. But in this case, its also the one trying to send it on.
> In other cases, the main thread has tried to send on the packet.

Here is the function ofproto_dpif_send_packet_in() that I see at the
bottom of your additional thread stack trace, quoted above:

    void
    ofproto_dpif_send_packet_in(struct ofproto_dpif *ofproto,
                                struct ofputil_packet_in *pin)
    {
        ovs_mutex_lock(&ofproto->pin_mutex);
        if (ofproto->n_pins > 1024) {
            ovs_mutex_unlock(&ofproto->pin_mutex);
            COVERAGE_INC(packet_in_overflow);
            free(CONST_CAST(void *, pin->packet));
            free(pin);
            return;
        }

        list_push_back(&ofproto->pins, &pin->list_node);
        ofproto->n_pins++;
        ovs_mutex_unlock(&ofproto->pin_mutex);
    }

This function does not call connmgr_send_packet_in(), so I don't see how
your stack trace is possible.