wiki:IPAddressChange

Version 4 (modified by bennylp, 14 years ago) (diff)

--

IP Address change and Access Point Reconnection Issues

Table of Contents

  1. Problem description
  2. Issues and solution related to IP address change
    1. Approach 1: Restart everything
    2. Approach 2: Selective update
      1. Account Contact URI update
      2. Media addresses update
      3. Call in progress issues
  3. Issues with iPhone/TCP
  4. Symbian specific issues and solution

This article describes some issues and their corresponding solutions related to access point disconnection, reconnection, IP address change, and how to handle these events in your PJSIP applications. The general issues related to the discussion will be explained, along with some specific issues to Symbian applications.


Problem description

IP address change and/or access point disconnection and reconnection are scenarios that need to be handled in mobile applications. Few issues or scenarios related to this for example are:

  • user moves outside the range of a Wi-Fi access point (AP) and lost the connection
  • user moves outside the range of one AP and reconnect to another
  • the handset may get new IP address if user reconnects to different AP

Each of the scenarios above may need different handling in the application.


Issues and solution related to IP address change

When the connection is reconnected, the handset may get different IP address than what it previously got. There are few ramifications of this, for example if PJSUA-LIB is used:

  • the SIP registration needs to be updated with a new Contact URI
  • the account URI also needs to be updated
  • media addresses need to be updated
  • if there is ongoing dialog, the remote party needs to be informed with new Contact URI as well as new media (RTP/RTCP) addresses.

Note that the monitoring of connection/interface status is outside the scope of PJSIP, so the application must implement this itself (for example using connection progress monitor in Symbian). Having said that, PJSIP does have some capability to detect some IP address change scenarios, for example by monitoring the IP address in SIP REGISTER response or in STUN Binding response when ICE transport is used and STUN is enabled.

Once the application has detected that the IP interface address has changed, there are two solutions to inform PJSIP about this.


Approach 1: Restart everything

The most straightforward solution is of course to restart everything, which means in pjsua terms to call pjsua_destroy() and followed by pjsua_create(), pjsua_init(), and so on. While this solution may sound crude, it is the easiest to do and as will be explained later it is not considerably worse then the more refined alternative.


Approach 2: Selective update

Alternatively there may be a way to allow the stack to continue to run, updating the address information when necessary. This approach will require some specific features to be used, as well as some actions by the application when it detects that the IP address has changed.

The specific configuration and tasks will be explained below. Note we assume that the SIP and media sockets are bound to INADDR_ANY (0.0.0.0) and not to a specific interface IP address (this is the default behavior).

Account Contact URI update

Task:
The account URI needs to be updated with the new address, and re-registration is necessary to inform the registrar about the new URI.
Description:
PJSUA-LIB has the capability to detect the (SIP) IP address change based on the response of REGISTER request and automatically update the registration with the correct IP if it detects that the IP/port seen by the server is different than the address specified in the Contact URI. This feature is enabled by default, via the pjsua_acc_config.allow_contact_rewrite setting.

So the solution is simply to trigger the re-registration by calling pjsua_acc_set_registration() function (after the new connection is up of course). The PJSUA-LIB will send re-REGISTER request, check the IP address/port in the response, and re-REGISTER again and update the account URI as necessary.

Media addresses update

Task:
The media (RTP/RTCP) addresses in PJSUA-LIB are normally determined during PJSUA-LIB startup, hence they need to be updated with the new address.
Description:
If ICE media transport is used, and STUN is enabled on the media transport, then the media transport will automatically update its publicly mapped IP address from the STUN Binding response. The transport should send STUN Binding request periodically (approximately every 15 seconds) as NAT keep-alive mechanism, so the address change will be detected by the transport automatically during this operation.

Note that at present there is no API to explicitly request the ICE media transport to initiate STUN Binding request immediately.

If ICE is not used, then at present there is no mechanism to update the IP address of media transport, nor the media transport will update its address even when STUN is used. The only solution would be to recreate the media transports and supply them to PJSUA-LIB with pjsua_media_transports_attach().

Call in progress issues

Task:
Dialog's Contact URI needs to be updated.
Description:
The dialog's Contact URI is set initially when the dialog is created, from the account's Contact URI. While at the PJSIP level the pjsip_inv_reinvite() allows changing of Contact URI via the new_contact argument, currently this feature is not used by PJSUA-LIB, i.e. the pjsua_call_reinvite() does not allow the application to change the Contact URI.

As a workaround, application may #include <pjsua-lib/pjsua_internal.h> and perform the reinvite manually, as the snippet below shows:

pjsip_tx_data *tdata;
pj_str_t new_contact;
pjsip_inv_sessin *inv = pjsua_var.calls[call_id].inv;

new_contact = pjsua_var.acc[account_id].contact;
status = pjsip_inv_reinvite(inv, &new_contact, NULL, &tdata);
if (status==PJ_SUCCESS)
    pjsip_inv_send_msg(inv, tdata);

Note that the usual caveats of including <pjsua_internal.h> apply (i.e. this is not public API and things may change in future releases).

Task:
Changing of RTP/RTCP media addresses of ongoing call
Description:
If ICE is used, then new STUN srflx address will be signaled in updated SDP offer, as long as:
  • ICE media transport has detected that the IP address has changed (via the keep-alive above), and
  • the media was previously inactive, since if media has been active (hence ICE session is active), the SDP will contain only the used candidates and not all the list of candidates.

Alternatively, we may not need to inform the new RTP/RTCP address at all. If the remote media endpoint has the capability to switch its RTP/RTCP transmission to the source address of the RTP/RTCP packets (note: PJMEDIA has this capability), then it should automatically switch its destination address to our new address, provided that the source address of our RTP/RTCP packets (as viewed by the remote peer) have indeed changed.

  • Note: by default we bind transports to INADDRANY/0.0.0.0, so when sending outgoing (UDP) packets, we rely on the OS to select the correct interface for us, based on what interfaces are currently online and the OS's internal routing table. In other words, we just call sendto() and let the OS "do the right thing". In case of IP address change, we are also relying on the OS to switch the interface from one interface to the new one for our UDP transmissions.


Issues with iPhone/TCP

[Update 2011/01/26]

TCP is preferred on iPhone because of the background feature, but it has been reported that simply re-registering after an IP address change is detected may not work, presumably because the TCP socket itself is already in bad state and is unable to communicate anymore. The following steps can be used to perform re-registration with a new TCP transport:

  1. We need to keep track of which transport is being used by the registration, by implementing the on_reg_state2() callback. Add reference counter to it to prevent other from deleting the transport while we're referencing it (it shouldn't happen while the registration is active, but just in case). Sample code:
    static pjsua_acc_id the_acc_id;
    static pjsip_transport *the_transport;
    
    static void on_reg_state2(pjsua_acc_id acc_id, pjsua_reg_info *info)
    {
       struct pjsip_regc_cbparam *rp = info->cbparam;
    
     
        ...
        if (acc_id != the_acc_id)
            return;
    
        if (rp->code/100 == 2 && rp->expiration > 0 && rp->contact_cnt > 0) {
    	/* Registration success */
    	if (the_transport) {
    	    PJ_LOG(3,(THIS_FILE, "xxx: Releasing transport.."));
    	    pjsip_transport_dec_ref(the_transport);
    	    the_transport = NULL;
    	}
    	/* Save transport instance so that we can close it later when
    	 * new IP address is detected.
    	 */
    	PJ_LOG(3,(THIS_FILE, "xxx: Saving transport.."));
    	the_transport = rp->rdata->tp_info.transport;
    	pjsip_transport_add_ref(the_transport);
        } else {
    	if (the_transport) {
    	    PJ_LOG(3,(THIS_FILE, "xxx: Releasing transport.."));
    	    pjsip_transport_dec_ref(the_transport);
    	    the_transport = NULL;
    	}
        }
        ...
    }
    
  1. When IP address change is detected: a) send unregistration, and b) close the TCP transport that we saved in step 1) above. Sample code:
    pj_status_t status;
    
    PJ_LOG(3,(THIS_FILE, "xxx: IP change.."));
    
    status = pjsua_acc_set_registration(the_acc_id, PJ_FALSE);
    if (status != PJ_SUCCESS)
        PJ_PERROR(1,(THIS_FILE, status, "xxx: pjsua_acc_set_registration(0) error"));
    
    if (the_transport) {
        status = pjsip_transport_shutdown(the_transport);
        if (status != PJ_SUCCESS)
    	PJ_PERROR(1,(THIS_FILE, status, "xxx: pjsip_transport_shutdown() error"));
        pjsip_transport_dec_ref(the_transport);
        the_transport = NULL;
    }
    
    
  1. And finally, once unregistration in 2a) above is complete, re-register (with TCP).

Note that ideally the closing the TCP transport is done in step 3 and not in step 2b. The drawback with doing it in 2b is, when the unregistration request is challenged (i.e. step 2a resulted in 401 response being received), a new TCP transport will be created, and the request retry will be sent with this new TCP transport. Your server may not like this, since it will see the unregistration request is coming from different TCP connection than the original request. But having said that, the existing TCP transport may not be in usable state anyway, so I suppose this is not a worse situation than that. But in general, you may need to tweak the timing of this closing the transport part (you may even want to put it before 2a).

Symbian specific issues and solution

Being a mobile operating system, Symbian has good supports in managing access point connection. In Symbian, the RConnection object is used to manage the connection, and each socket handle (RSocket) is created in a context of an RConnection. The RConnection.ProgressNotification() method can be used to register an Active Object to be run when the connection status has changed, so the application has good control over the connection.

However there are still couple of unsolved issues remaining (probably due to lack of knowledge in our part):

  • when the connection in RConnection is down, it seems that the sockets created with that RConnection will detach themselves from the connection, so even though the RConnection is reconnected, this will not automatically make the sockets recover to a "good" state. If the application tries to make use of the socket, for example, to call !SendTo(), it will cause the socket to pop up the access point selection dialog again
  • more over, if user selects different access point in the dialog, this will put the sockets in somewhat worse state. If the application tries to make use of the socket, for example, to call !SendTo(), it will cause the TRequestStatus associated with the operation to block for a long time (about one and half minute). And even worse, a second call to !SendTo() will cause the TRequestStatus to block indefinitely!

At the moment we are not aware of any solutions for the above issues. Lacking this, we created a workaround in PJLIB to prevent it from accessing any Symbian socket API's when the connection has been down and reconnected. The API is pj_symbianos_set_connection_status() and it was added in PJSIP version 1.0.2/1.1 (ticket #733 and #732 respectively).

Below is a sample code, implemented in symbian_ua\ua.cpp sample application, to restart PJSIP and use the pj_symbianos_set_connection_status() function.

class CConnMon : public CActive {
public:
    static CConnMon* NewL(RConnection &conn, RSocketServ &sserver) {
	CConnMon *self = new (ELeave) CConnMon(conn, sserver);
	CleanupStack::PushL(self);
	self->ConstructL();
	CleanupStack::Pop(self);
	return self;
    }
    
    void Start() {
	conn_.ProgressNotification(nif_progress_, iStatus);
	SetActive();
    }
    
    void Stop() {
	Cancel();
    }
    
    ~CConnMon() { Stop(); }
    
private:
    CConnMon(RConnection &conn, RSocketServ &sserver) : 
	CActive(EPriorityHigh), 
	conn_(conn), 
	sserver_(sserver)
    {
	CActiveScheduler::Add(this);
    }
    
    void ConstructL() {}

    void DoCancel() {
	conn_.CancelProgressNotification();
    }

    void RunL() {
	if (nif_progress_().iStage == KLinkLayerClosed) {
	    pj_status_t status;
	    TInt err;

	    // Tell pjlib the connection has been down.
	    pj_symbianos_set_connection_status(PJ_FALSE);
	    
	    PJ_LOG(3, (THIS_FILE, "RConnection closed, restarting PJSUA.."));
	    
	    // Destroy pjsua
	    pjsua_destroy();
	    PJ_LOG(3, (THIS_FILE, "PJSUA destroyed."));

	    // Reopen the connection
	    err = conn_.Open(sserver_);
	    if (err == KErrNone)
		err = conn_.Start();
	    if (err != KErrNone) {
		CActiveScheduler::Stop();
		return;
	    }

	    // Reinit Symbian OS param before pj_init()
	    pj_symbianos_params sym_params;
	    pj_bzero(&sym_params, sizeof(sym_params));
	    sym_params.rsocketserv = &sserver_;
	    sym_params.rconnection = &conn_;
	    pj_symbianos_set_params(&sym_params);

	    // Reinit pjsua
	    status = app_startup();
	    if (status != PJ_SUCCESS) {
		pjsua_perror(THIS_FILE, "app_startup() error", status);
		CActiveScheduler::Stop();
		return;
	    }

	    
	    PJ_LOG(3, (THIS_FILE, "PJSUA restarted."));
	    PrintMenu();
	}
	
	Start();
    }
    
private:
    RConnection& conn_;
    RSocketServ& sserver_;
    TNifProgressBuf nif_progress_;
};

What the code in RunL() above does is it shuts down PJSIP when the connection is down, ask user to reconnect by showing up the access point dialog, and (re)start the application.

Note that the drawback with this approach is that it does not clean up the registration and calls properly, that is no SIP unregistration will be sent and if the application is in the middle of a call while the connection is down then no BYE will be sent either. Currently we can't suggest any other solution, as we can't get rid of the socket gets stuck problem.

More about the socket stuck problem

This is a problem with the socket in general. Below are steps to reproduce with a plain UDP socket:

  1. Create RConnection, call Start() to connect to AP
  2. Create UDP socket, call SendTo() to send a packet
  3. Disconnect the AP (Menu -> Connectivity -> Conn. mgr. -> Active data connections -> (highlight the AP) -> Options (menu) -> Disconnect).
  4. Call udp.SendTo() again
  5. AP selection dialog appears, select different AP
  6. WaitForRequest() to the udp.SendTo() operation now will get stuck for 1-2 minutes before tcpip6_error_NoRoute error (-5105) is returned.
  7. Now if you call udp.SendTo() again, now WaitForRequest() will get stuck indefinitely

The problem above does not occur if:

  • the user selects the same AP. In this case the udp.SendTo() should complete successfully
  • the user cancels the AP selection dialog. In this case the udp.SendTo() will fail immediately with KErrCancel without blocking the application.
  • the socket is closed and re-opened. In this case the udp.SendTo() should complete successfully

As additional info:

  • the problem still persists even if the RConnection is restarted (RConnection.Start() is called to select new AP) in between step 3 and 4 above.

Attachments (1)

Download all attachments as: .zip