A question about NFS and Samba

[Richard Sharpe]

On Mon, 8 Jul 2002 zhangyfh at legend.com.cn wrote:

> hi, guys
> 
> A question here.
> 
>    In former Heterogeneous  network filesystem Environments  , there must
> be a big problem, it is file locking. when I use Samba and NFS at the same
> time, NFS client might can't understand the oplocks of Samba, and Samba
> Client might can't know what NFS client has buffered of data for writing on
> the same file also. Many manufactures has developed their so called
> 'Unified File Locking',for example, Net APP's SecureShare, Veritas' SLFL,
> HP's Unified File Locking.
> 
>    But now, Linux 2.4 kernel immplement mandary lock, Samba released 2.2.5
> version, NFS V4 changed a lot, I wanna know that under Linux 2.4.18, with
> Samba 2.2.5 and NFS 3, is that problem still there? If not, what kind of
> mechanism ensure the data integrity with cross-protocol?

Linux 2.4.x, for some value of x, implements kernel oplocks. Thus, Samba 
gets a signal if it has taken out a kernel oplock on a file, and someone 
else opens the file in an incompatible manner.

It is not clear that signals will support large numbers of oplock breaks 
per second however, and it might be better to use a mechanism like 
kqueue, but for the moment, that is how I have implemented it 
(unpublished) in FreeBSD as well. But then my problem is a little more 
difficult. I had to field callback breaks breaking delegations I have been 
given in a multi-protocol distributed yadda-yadda mumble frotz.

Regards
-----
Richard Sharpe, rsharpe at ns.aus.com, rsharpe at samba.org, 
sharpe at ethereal.com