[openib-general] Re: [PATCH] CMA: allow/require bind before connect

[Tom Tucker]

On Mon, 2006-03-27 at 10:22 -0800, Sean Hefty wrote:
> Michael S. Tsirkin wrote:
> >>>>>Sean, CMA currently bails out if I bind to ANY address before connect.
> >>>>>I think in TCP you bind before connect 
> >>>>
> >>>>It is not typical to use bind on the active side before connect. When it
> >>>>is done, is when the application wants to control which local port
> >>>>and/or interface to use when connecting to a remove peer. bind(2) is
> >>>>typically used on the passive side before listen.
> >>>
> >>>Tom, I have a connection bind on a local port to control which local port
> >>>to use when connecting to a remove peer.
> 
> Couldn't you specify the local port when calling rdma_resolve_addr()?  TCP 
> doesn't require calling bind before connect, so I'd like to avoid requiring that 
> here.
> 
> >>>At the level I work in SDP, Linux net core calls bind before connect if the
> >>>socket source port is 0. I could override this but I don't think its a good
> >>>idea since the point is to emulate TCP: TCP connections always have a source
> >>>port and so should CMA connections.
> 
> The purpose behind the CMA is to establish connections using TCP/IP addressing, 
> versus trying to emulate TCP.  On the passive side, this means that the CMA 
> supports listening on a 16-bit port number on a single IP address, or across all 
> IP addresses.  On the active side, there's no need to even use a port number 
> (when running over IB).  The CMA doesn't need it to establish a connection or 
> use it to match received data with the correct connection.
> 
> > CMA does not allocate ephemeral ports currently.
> > This actually makes some sense since address families/port spaces expect
> > different semantics wrt which port numbers are legal.
> 
> It's correct that the CMA currently does not do this.  I guess I'm still unsure 
> of why it needs to when running over IB.  The CMA should work fine if it 
> assigned every active connection the same port number.  As long as the CMA 
> manages its own port spaces, does it matter if its semantics are defined this way?

Yeah I think it does. Some applications pay attention to the remote port
number. This is what I mean by 'emulating TCP'.  It's important for SDP
in particular because some of these applications will expect getsockname
to return a fully qualified 4-tuple. The same for the 'addr' parameter
to the accept call. 

> 
> - Sean