Fix IPv6 support issue.

webrtc/base/nethelpers.cc

 /*
  *  Copyright 2008 The WebRTC Project Authors. All rights reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
@@ skipping 26 matching lines @@
   ASSERT(false);
   LOG(LS_WARNING) << "ResolveHostname() is not implemented for NaCl";
   return -1;
 #else  // __native_client__
   if (!addresses) {
     return -1;
   }
   addresses->clear();
   struct addrinfo* result = NULL;
   struct addrinfo hints = {0};
-  // TODO(djw): For now this is IPv4 only so existing users remain unaffected.
-  hints.ai_family = AF_INET;
+  // This should return both IPv4 and IPv6 addresses if available.
+  hints.ai_family = AF_UNSPEC;

[pthatcher1, 2016/06/21 05:26:54]

Can you make a separate CL for this in case we have to rollback either half of
this CL?

[andresp, 2016/06/21 11:36:10]

I am also not sure if this AF_UNSPEC is that correct. Documentation on behaviour
is scarce.

I guess since we trigger the resolve from the turnport itself, we do either
AF_INET or AF_INET6 depending on the turnport local address family.

[honghaiz3, 2016/06/21 18:36:06]

Will do it in a separate CL.

[pthatcher1, 2016/06/21 18:50:17]

Would it make sense to use the family already passed into this function?

hints.ai_family = family.

The the caller has control by setting the family of address passed into Start.


As for AF_UNSPEC, it supposedly works for Linux:

http://man7.org/linux/man-pages/man3/getaddrinfo.3.html

And for Android, the Bionic code makes two queries:

https://android.googlesource.com/platform/bionic/+/7e0bfb511e85834d7c6cb96312...

On Windows, it says INET+INET6:

https://msdn.microsoft.com/en-us/library/windows/desktop/ms738520(v=vs.85).aspx

On Mac, it's more complicated:

https://developer.apple.com/legacy/library/documentation/Darwin/Reference/Man...

But it says "    To best support NAT64 networks, it is recommended to resolve
all IP address literals with ai_family set to PF_UNSPEC and ai_flags set to
AI_DEFAULT."


So it seems to be roughly supported.  The only one I'm worried about is Mac/iOS.
 

[honghaiz3, 2016/06/21 20:06:49]

I think it is safer to pass in the family. But the one currently passed is the
address family of the named address. I do not think that will have a family
type.  

[pthatcher1, 2016/06/21 20:55:08]

Yeah, the caller would need to call address.SetFamily, which doesn't exist. 
Perhaps a better option is to require AsyncResolver::Start to pass in an extra
parameter that is the family that should be queried.  

Then the call could choose INET, INET6, or UNSPEC.  But then what do we do with
UNSPEC?  I think for Android, Linux, and Windows, it's safe to use it, according
to the docs.  And even Mac seems to be clear that they support it, even if in a
slightly different way. 

   hints.ai_flags = AI_ADDRCONFIG;
   int ret = getaddrinfo(hostname.c_str(), NULL, &hints, &result);
   if (ret != 0) {
     return ret;
   }
   struct addrinfo* cursor = result;
   for (; cursor; cursor = cursor->ai_next) {
     if (family == AF_UNSPEC || cursor->ai_family == family) {
       IPAddress ip;
       if (IPFromAddrInfo(cursor, &ip)) {
@@ skipping 114 matching lines @@
       break;
     }
   }
   freeifaddrs(ifa);
   return has_ipv6;
 #else
   return true;
 #endif
 }
 }  // namespace rtc