[AFS3-std] RPC refresh updates?

Simon Wilkinson simon@sxw.org.uk
Mon, 1 Feb 2010 15:29:20 +0000


On 1 Feb 2010, at 15:09, Jason Edgecombe wrote:

> hi everyone,
>
> I'm working on the February issue of the OpenAFS newsletter.
>
> I found http://www.dementia.org/twiki/bin/view/AFSLore/RpcRefresh
>
> To confirm my understanding, is the RPC refresh the process to  
> convert various 32bit fields into 64bit fields in the AFS protocol?

Essentially the proposals are to create a new set of AFS RPCs which:
   *) Change FID elements from 32bits -> 64bits
   *) Change time from 32bits -> 64bits and standardise on 100ns  
granularity
   *) Make assorted changes to the FetchData structure to accommodate  
rxosd
   *) Make quota and block size information fields 64 bits
   *) Add a field to FetchStatus to store a volume's last update time
   *) Define per-file semantics for ACL information
   *) Extend ACLs to be represented by 32bits on the wire (and reserve  
the new 16bits for standardised use)
   *) Clean up some of the unused fields FetchStatus

These are all purely wire protocol changes. So, for example, the  
fileserver will be unable to actually handle, or store, 32bit ACLs at  
this point in time.

> What is the status of the project? Simon's edits on the wiki page  
> were made in June 2009.
The wiki is from a very early pass at planning the RPC refresh. We  
made much more progress in Edinburgh - there are concrete details of  
the changes we intend on making in the notes from that Hackathon.

I'm supposed to be producing a draft which contains new RPC  
descriptions by merging the results of all of these changes. I have  
text from Jeff and Derrick, but I'm still waiting on details of the  
rxosd changes from Christof. Sadly work has got in the way of me  
driving this forwards as much as I would have liked.

> How are the current protocol proposals like SRV DNS related to the  
> RPC refresh?

They're not. The only intersections with RPC refresh are drafts which  
also want to create new versions of the RPCs that this change touches,  
or drafts which want to do things like time or FIDs in the same way as  
the refresh does, and have to wait until that decision is made.

Cheers,

Simon.