Retry when encountering UPnP socket initialization issues.

The UPnP socket can not bind to 0.0.0.0 so needs a stable interface
with a configured IP address to bind to.
If the OS has not initialized the network yet, this will fail, so
add a couple of rounds of retries to make this more robust.

This only showed up as an issue with recent distributions using
systemd. I suspect a bug in debian systemd network-online.target actually
not doing as advertised and claiming readiness when in fact DHCP might
still be ongoing.

Fixes #129

src/upnp_device.c

@@ -412,8 +412,19 @@ static gboolean initialize_device(struct upnp_device_descriptor *device_def,
 	char *buf;
 
 	rc = UpnpInit(ip_address, port);
+	/* There have been situations reported in which UPNP had issues
+	 * initializing right after network came up. #129
+	 */
+	int retries_left = 60;
+	static const int kRetryTimeMs = 1000;
+	while (rc != UPNP_E_SUCCESS && retries_left--) {
+		usleep(kRetryTimeMs * 1000);
+		Log_error("upnp", "UpnpInit(ip=%s, port=%d) Error: %s (%d). Retrying... (%ds)",
+			  ip_address, port, UpnpGetErrorMessage(rc), rc, retries_left);
+		rc = UpnpInit(ip_address, port);
+	}
 	if (UPNP_E_SUCCESS != rc) {
-		Log_error("upnp", "UpnpInit(ip=%s, port=%d) Error: %s (%d)",
+		Log_error("upnp", "UpnpInit(ip=%s, port=%d) Error: %s (%d). Giving up.",
 			  ip_address, port, UpnpGetErrorMessage(rc), rc);
 		return FALSE;
 	}