Bug 1824

Summary:	L4 protocol sockets should support BindToNetDevice over IPv6
Product:	ns-3	Reporter:	Tommaso Pecorella <tommaso.pecorella>
Component:	internet	Assignee:	Tommaso Pecorella <tommaso.pecorella>
Status:	RESOLVED FIXED
Severity:	enhancement	CC:	ns-bugs, tomh
Priority:	P5
Version:	ns-3-dev
Hardware:	All
OS:	All
Attachments:	patch

Description Tommaso Pecorella 2014-01-05 08:17:53 EST

The problem is in UdpSocketImpl::BindToNetDevice and TcpSocketBase::BindToNetDevice.

In both methods only the v4 methods are initialised, while also the v6 endpoint should be initialised (v6 endpoints have a BindToNetDevice).

Comment 1 Tommaso Pecorella 2014-01-05 12:32:46 EST

Created attachment 1752 [details]
patch

The patch add the required BindToNetDevice calls to the IPv6Endpoint inside TCP and UDP sockets.

The only doubt I have is: if the socket wasn't already bound (normal Bind), BindToNetDevice will do it for the user.

BUT it will open *two* endpoints, v4 and v6.

Since we do have separate Bind and Bind6, and they open just the endpoint they need, maybe it's not a good idea to open another one just because we can.

SO, the BindToNetDevice "logical" thing would be: do nothing or raise an assert if the EndPoint is not already bound by a previous Bind.
This also prevents the idiotic error: call BindToNetDevice() and, after call Bind() (it's causing a a memory leak).

Opinions ?

Comment 2 Tom Henderson 2014-09-12 12:43:31 EDT

I'm OK with this also, as long as there is some documentation added about the expected sequence of Bind() calls supported.

Comment 3 Tommaso Pecorella 2014-09-12 14:52:54 EDT

added in changeset:   10933:7442f5603ef4