Implementing DFS.

[Love]

Bjorn Augustsson <august at chips.chalmers.se> writes:

> Hi!
> 
> Some of you might remember me (August) from the MADE 2000 workshop
> earlier this year. We were talking about implementing a DFS* client
> by extending arla.
> 
> We're doing a kind of feasibility study on getting it done. (In 
> terms of how long it will take, and how much will it cost.)
> 
> We're assuming that we have DCE working to begin with. That leaves 
> us with:
> 
> * The RPC layer is obviously different.
> * ACLs on files.

This isn't too hard. In fact the afs3 protocol don't limit ACLs to
directories, it's just a implementation issue. So this shouldn't be a too
hard issue.

> * No write-on-close. This means different client-server protocols, but
>   we don't know to what extent and how hard this can be.
> * Byte-range locking (Don't know if this is strictly neccessary, it ought
>   to work without it in the beginning at least.)

These too are pretty much the same. There might be some changes to the
arla/xfs protocol to accommodate byte-range locks/writing/reading.

> * The tools. 
>    * cm  (Should be farly easy.)
>    * fts
>    * Maybe others that we haven't thought of right now.

This can be done by pretty much anyone.
 
> Comments? Did we leave anything (important) out?
> 
> What are good milestones? 

I think 

* a test-tool that get a file from the fileserver.
* read-only support in arlad[*]
* read-write support in arlad[*]
* tools on the side

I think that some of this things can be does at the same time. The steps
marked with [*] obviously have diffrent level of support, like `working',
`working with test-suite', `working with users'.

> Where can we expect incompatibility problems? Most of the code in 
> DFS and AFS is common, but I guess the ACL stuff might clash a bit
> since they're a DCE-ism in DFS for example.

Well since the syscall is probably diffrent this wont matter. And where
will probably be clean cut between the two diffrent layer so it wont
matter. I don't remeber how diffrent the acl/access stuff in DFS was from
when I read the spec, but I don't think it was.

If you base your code on the arlad you'll have the parts of the cache
manager already implemented, like file-management and volumecache-code.
 
> And most importantly, is anyone else interested in this? Anyone even 
> willing to help out?

We will probably be able to help you out with questions about arlad, but I
don't know how much we will be able to help you write/test code.

Love