Tom Tucker wrote:
> On Fri, 2007-09-28 at 14:36 +1000, Neil Brown wrote:
>> On Thursday September 27, tom@opengridcomputing.com wrote:
>>> Move the IP address fields to the svc_xprt structure. Note that this 
>>> assumes that _all_ RPC transports must have IP based 4-tuples. This 
>>> seems reasonable given the tight coupling with the portmapper etc...
>>> Thoughts?
>> I don't think NFSv4 requires portmapper (or rpcbind) ... does it?
>>
>> "Everything uses IP addresses" sounds a lot like "Everything is a
>> socket".  I would have supported the latter strongly until RDMA came
>> along.  Now I'm even less sure about the former.
>>
>> How much cost would there be in leaving the address in the
>> per-transport data?
> 
> Very little. The original patchset had it in the per-transport data.

For comparison, the client keeps the remote endpoint address in the 
generic transport data, stored in a struct sockaddr_storage.

I am also trepidacious about claiming all things use IP addresses, but 
the reality is NFS itself is built assuming IP.  There would be so much 
more to change (like how export rules work) that I think we can safely 
assume that IP addresses are the format for identifying remote endpoints 
until we have clear evidence that a change it needed, imo.