[openib-general] Re: RFC: cma: need rdma_unbind

[Michael S. Tsirkin]

Quoting r. Sean Hefty <mshefty at ichips.intel.com>:
> Subject: Re: RFC: cma: need rdma_unbind
> 
> Michael S. Tsirkin wrote:
> >Hmm. As I see it, keeping the QP alive for a given timeout after user 
> >requested
> >close is a simple way to implement graceful close, not limited to sockets.
> >On the other hand, we need to free the port immediately.
> 
> This seems to imply that there's another connection abstraction on top of 
> the CMA.

Well, that's what ULPs are.

> From the perspective of the CMA, the connection, and hence port, is still 
> in use.

No. A socket is a 5 tuple (proto, local addr, local port, remote addr, remote
port).  unbind just says that you can reuse local addresses, so e.g. a new
connection request will connect to a new socket.

> For example, I don't think that rdma_unbind() could work if the 
> underlying device were iWarp, or something that used the port number when 
> mapping data.

The 5 tuple still must be unique, so no problem I think.

> I also can't think of a better solution to this problem.

Maybe rename it unbind_local?

-- 
MST