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

[Caitlin Bestler]

Sean Hefty 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.
> 
> What I what to understand is _why_ it must be unique?  The
> CMA-IB port spaces are separate from the native stack.  And
> the CMA-IB has no issues connecting or routing data among
> multiple connections all having the same 4-tuple.
> 

>From the perspective of any given host, IP addresses are
unique across all interface devices. A given connection
can therefore be identified by just the 4-tuple, with no
need to explicitly state "via this device".

If that were not true, applications would need to include
the device in any refernce to the connection or potential
connection. That is not something that IP develoers expect
to do, or that they want to do.