[openib-general] RE: [dat-discussions] socket based connectionmodel for IB proposal - round 3

[Caitlin Bestler]

Roland Dreier wrote:
>     Sean> The current software did not require changes to the CM
>     Sean> protocol, and does provide assurance that the private data
>     Sean> was formatted by a kernel entity.
> 
> How do you prevent a userspace process from using the current
> ucm module to connect to one of the CMA services on a remote
> system?  If you allow that, then the process can use the old
> simple direct IB CM interface to put whatever it wants into
> the REQ private data.
> 
> I don't see any handling of the service ID in the ucm or cm
> code beyond taking what the consumer passes in and formatting
> the CM messages with it.
> 
>  - R.

That's exactly my point. As far as I can see, the only way
to prevent that is to have a bit that a current CM implementation
does not set, meaning it is in the header not the private data.