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

[Caitlin Bestler]

openib-general-bounces at openib.org wrote:

> 
> 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? 
> 
The 4-tuple must be unique, even across devices. Hence if the
native stack has X:Y to I:J then CMA cannot also create
X:Y to I:J. 

Strictly speaking there is no reason why the CMA cannot
re-use the same active port number as many times as it wants,
as long as it does not violate the uniqueness rule. Back when
64K was considered a large number programmers might have
expected that X:Y was unique on the active side, but I don't
think they can really count on that any more. But it's still
a good idea unless you need more than 64K outbound connections
from the same IP address.