NFSv4 server running on RHEL 5.2 -- Server/Client Incompatibilities?

[J. Bruce Fields]

On Mon, Jul 07, 2008 at 03:46:36PM -0700, Paarvai Naai wrote:
> Hi all,
> 
> This somewhat related to the post from Feb 2008:
> 
> http://linux-nfs.org/pipermail/nfsv4/2008-February/007697.html
> 
> Specifically, I am running an NFSv4 server on Centos 5.2 (kernel
> 2.6.18-92.1.6.el5).  I am then using FC8 as an NFSv4 client (kernel
> 2.6.25.6-27.fc8).  I am able to mount the server's NFSv4 export on the
> client and access files with no perceived problems.
> 
> I am using the following krb5-based export:
> 
> /export        gss/krb5(rw,fsid=0,crossmnt,insecure,no_subtree_check,sync)
> 
> and the following mount parameters on the client:
> 
> nfs:/                   /mnt/nfs4               nfs4    noauto,sec=krb5 0 0
> 
> However, when I try to exercise file locking using the command "flock
> -x filename echo" I get an "Input/Output Error" on the client.
> Furthermore, I see the following error in my /var/log/messages on the
> server:
> 
> kernel: NFSD: preprocess_seqid_op: bad seqid (expected 2, got 3)

The two may be unrelated--this will fail anyway because BSD flock locks
aren't really supported over nfsv4.  (Well, it's more complicated than
that: actually the client does attempt to support flock() by translating
its locks into posix locks, but exclusive posix locks aren't permitted
on files open for read, whereas BSD locks are.  And the flock utility
just does a read open.  You won't get this error on v3 since v3 doesn't
have on-the-wire opens....)

If this worked before it may just be because the old client didn't
actually try to do the flock->posix lock translation, hence just
performed the flock locally?

> I am suspecting I am simply out of luck here because I am using wildly
> different versions of the kernel on the server and client machines.
> Is this true?  This brings me to the question of how stable the
> underlying protocol of NFSv4 is.  In other words should I still be
> expecting these sorts of problems for a while or will things stabilize
> as it has for NFSv3?

The NFSv4.0 protocol hasn't been changed for several years, so any
incompatibilities between different versions are implementation bugs.

> Lastly, the following pages seem to be a bit out-of-date.  Are there
> new pages that list the current status/stability of the NFSv4 support
> for Linux?

No.  Unfortunately it seems hard to keep a good up-to-date description.

--b.

> 
> http://client.linux-nfs.org/priorities.html
> http://www.citi.umich.edu/projects/nfsv4/linux/NFSv4_server_priorities.html
> 
> Thank you,
> Paarvai
> _______________________________________________
> NFSv4 mailing list
> NFSv4 at linux-nfs.org
> http://linux-nfs.org/cgi-bin/mailman/listinfo/nfsv4