Move the concept of multiple route into Network

webrtc/p2p/client/basicportallocator.cc

Index: webrtc/p2p/client/basicportallocator.cc
diff --git a/webrtc/p2p/client/basicportallocator.cc b/webrtc/p2p/client/basicportallocator.cc
index c5cfbd94a154940b6a92eb84337ab19fe02ab7dc..321621b9978bfe7f3fb74860478fc18504204abd 100644
--- a/webrtc/p2p/client/basicportallocator.cc
+++ b/webrtc/p2p/client/basicportallocator.cc
@@ -304,6 +304,17 @@ void BasicPortAllocatorSession::DoAllocate() {
   bool done_signal_needed = false;
   std::vector<rtc::Network*> networks;
 
+  // SignalNetworksChanged should have been fired when we are here.
+  DCHECK(allocator_->network_manager()->network_permission_state() !=
+         rtc::NetworkManager::STATE_UNKNOWN);
+
+  // If the network permission state is BLOCKED, we just act as if the flag has
+  // been passed in.
+  if (allocator_->network_manager()->network_permission_state() ==
+      rtc::NetworkManager::STATE_BLOCKED) {

[pthatcher1, 2015/08/18 18:34:20]

In a release build, we would grant access if in the STATE_UNKNOWN.  I think it
would be safer to test "!= ALLOWED" rather than "== BLOCKED".

+    set_flags(flags() | PORTALLOCATOR_DISABLE_ADAPTER_ENUMERATION);

[pthatcher1, 2015/08/18 18:34:20]

So what if the the permission changes?  Should we have a signal that listens to
that and starts enumeration after it changes?

+  }

[pthatcher1, 2015/08/18 18:47:59]

Can you add a unit test for this?

+
   // If the adapter enumeration is disabled, we'll just bind to any address
   // instead of specific NIC. This is to ensure the same routing for http
   // traffic by OS is also used here to avoid any local or public IP leakage