< Previous by Date Date Index Next by Date >
< Previous in Thread Thread Index  

Re: [reSIProcate] Potential bug in TransportSelector


I don't think this will be baked enough for 1.1.

RjS

On Feb 9, 2007, at 8:20 PM, Byron Campen wrote:

Of course. It'll require dumping a message on the TU's fifo, but shouldn't require anything more complicated. (or, if we want to be extra careful in case the TU might go away, we could do it through the stack) Do we want to try to get this in 1.1, or do we want to give it a release cycle to settle?

Best regards,
Byron Campen

This isn't straightforward since we have to potentially cross thread
boundaries.

On 2/9/07, Byron Campen <bcampen@xxxxxxxxxxxx> wrote:

Ok, so what we need to do here is have DUM (or any TU that acts as an endpoint) register an outbound decorator that inspects the final, filled-out Contact, and remember what it is, so it can keep track of its dialog-state properly (and reuse the fully specified Contact for the rest of the dialog).

Best regards,
Byron Campen


(fixing the list address)

If the TU explicitly specifies a transport, sure.
But if the TU has asked the stack to deal with this for it, the stack needs to populate the right things (think multiple interfaces with different
policies wrt transports).

We may be running into a problem where we're saying the wrong thing by
telling the stack to do this by only leaving the hostname blank.

RjS




On Feb 9, 2007, at 2:22 PM, Byron Campen wrote:

I feel uneasy about touching the transport param at all; perhaps the UAC put one into its Contact for a reason! And if there wasn't one there in the first place, I think it should be safe to assume the UAC doesn't care what protocol stuff comes in on, so we shouldn't go adding one. I assume that the UAC is at least aware of what kinds of transports (TCP, UDP,TLS,...) it is listening on, and if its transport loadout is somehow incomplete, it should be prepared to compensate for its lack of compliance by putting explicit transport params into the Contact. My vote would be to remove the code block
altogether.

Best regards,
Byron Campen


In TransportSelector::transmit we are seeing a bug in the following case:

UAC sends a SUBSCRIBE over transport at TCP:5060 and puts the correct
contact in the message
UAC receives 200/SUBSCRIBE with a Record-Route of UDP:5061

With the following code, we correctly send over UDP to target 5061 but we leave transport=tcp in the Contact which causes a problem for the UAS since it looks like a Contact refresh and subsequent requests from the peer are
sent to TCP:5061 where there is no listener.



      if (target.transport)
      {
// There is a contact header and it contains exactly one entry
         if (msg->exists(h_Contacts) &&
msg->header(h_Contacts).size()==1)
         {
            for (NameAddrs::iterator
i=msg->header(h_Contacts).begin(); i !=
msg->header(h_Contacts).end(); i++)
            {
               NameAddr& contact = *i;
// No host specified, so use the ip address and port of the
               // transport used. Otherwise, leave it as is.
               if (contact.uri().host().empty())
               {
                  contact.uri().host() =
(target.transport->hasSpecificContact() ?

target.transport->interfaceName() :

Tuple::inet_ntop(source) );
                  contact.uri().port() = target.transport->port();

                  if (target.transport->transport() != UDP)
                   {
                      contact.uri().param(p_transport) =
Tuple::toData(target.transport->transport());
                   }


My proposal is to remove the if (target.transport->transport() != UDP). Does
anybody remember why this is here?

_______________________________________________
resiprocate-devel mailing list
resiprocate-devel@xxxxxxxxxxxxxxxxxxxx
https://list.resiprocate.org/mailman/listinfo/resiprocate-devel
_______________________________________________
resiprocate-devel mailing list
resiprocate-devel@xxxxxxxxxxxxxxxxxxxx
https://list.resiprocate.org/mailman/listinfo/resiprocate-devel