Use ChannelMojo for NaCl PPAPI channels.

content/renderer/pepper/host_dispatcher_wrapper.cc

Index: content/renderer/pepper/host_dispatcher_wrapper.cc
diff --git a/content/renderer/pepper/host_dispatcher_wrapper.cc b/content/renderer/pepper/host_dispatcher_wrapper.cc
index 88d567a5cdb77d89db20998549ce965ca0e6c96b..034dadc1260f672e9feaae078898c366b18ab56f 100644
--- a/content/renderer/pepper/host_dispatcher_wrapper.cc
+++ b/content/renderer/pepper/host_dispatcher_wrapper.cc
@@ -41,8 +41,9 @@ bool HostDispatcherWrapper::Init(const IPC::ChannelHandle& channel_handle,
     return false;
 
 #if defined(OS_POSIX)
-  DCHECK_NE(-1, channel_handle.socket.fd);
-  if (channel_handle.socket.fd == -1)
+  DCHECK(channel_handle.socket.fd != -1 ||
+         channel_handle.mojo_handle.is_valid());

[Mark Seaborn, 2016/10/12 00:57:24]

Presumably this shouldn't be conditional on OS_POSIX any more?

Maybe IPC::ChannelHandle could provide a method for sanity-checking that
mojo_handle is filled out and the other fields aren't, since this check is
cropping up in multiple places?

[Sam McNally, 2016/10/12 03:15:32]

Allowing non-ChannelMojo handles is still necessary for non-NaCl plugins.
https://codereview.chromium.org/2302053004/ will change that and simplify this
check.

[Mark Seaborn, 2016/10/13 21:27:51]

OK.  My comment still stands for other places where the code is checking
channel_handle.mojo_handle.is_valid() -- it seems like it would be nice if those
could also check that the other fields are not filled out, given that
IPC::ChannelHandle has become fairly complex now.  What do you think?

[Sam McNally, 2016/10/14 02:35:02]

Done. ChannelHandle shouldn't even allow this, but I don't think it's worth
changing at this point.

+  if (channel_handle.socket.fd == -1 && !channel_handle.mojo_handle.is_valid())
     return false;
 #endif