Re: NFSv4

Re: NFSv4

[Trond Myklebust]

On Fri, 2010-02-19 at 01:23 +0100, J.A. Magallón wrote: 
> Hi all...
> 
> First of all, kudos for new nfs-utils. Now, nfs4 works automagically and
> like a charm.
> 
> And now the hard part, some questions:
> 
> - I have read that nfs4 includes in the server the locking protocol, no
>   need for separate lockd. But in my servers, it seems it is still running:
> 
> root      2198     2  0 Feb18 ?        00:00:00 [nfsiod]
> root     23501     2  0 01:01 ?        00:00:00 [nfsd4]
> root     23502     2  0 01:01 ?        00:00:00 [nfsd]
> root     23503     2  0 01:01 ?        00:00:00 [nfsd]
> root     23504     2  0 01:01 ?        00:00:00 [nfsd]
> root     23505     2  0 01:01 ?        00:00:00 [nfsd]
> root     23506     2  0 01:01 ?        00:00:00 [nfsd]
> root     23507     2  0 01:01 ?        00:00:00 [nfsd]
> root     23508     2  0 01:01 ?        00:00:00 [nfsd]
> root     23509     2  0 01:01 ?        00:00:00 [nfsd]
> root     23500     2  0 01:01 ?        00:00:00 [lockd]
> 
>   Is there any problem with it ? I suppose (correct me if I'm wrong)
>   that this kernel lockd will only serve for v2 or v3 mounts,
>   that locking for client nfs4 mounts will go through nfsd4.
>   Is that right ?

No. The client NFSv4 traffic goes through the ordinary 'nfsd' daemons.
There is no special locking manager for NFSv4, since POSIX locks are
part of the ordinary protocol.

I don't think that the NFS server will switch off lockd even if you do
specify that you only want to serve NFSv4.

> - Why is there only 1 instance of v4 daemon ?

The 'nfsd4' thread above is actually a workqueue that is used for
garbage-cleaning expired NFSv4 state. It isn't a server thread.

> - Is there any page describing the advantages of v4 ? I will have to
>   convince the department admin to activate v4 in his solaris boxen...;)

One place to start is the NFSv4 design considerations. See
          http://tools.ietf.org/html/rfc2624

Cheers
  Trond