Various little questions

[Richard Sharpe]

On Wed, 1 Oct 2003, Christopher R. Hertel wrote:

> Well folks, I've managed to find myself back in the role of 
> annoying-little-question-asker.  :)
> 
> My annoying little questions:
> 
> 1) Kernel oplocks
> 
>    The smb.conf manpage (for V3.0) does not come right out and say it, but
>    it suggests that the kernel oplock break is an all-or-nothing deal.  
>    In other words, there is no mechanism that allows the kernel to report 
>    that a process has opened an oplocked file in read-only mode, so the 
>    SMB client cannot be downgraded to a level2 oplock.
> 
>    Do I understand that correctly?
> 
>    Here's an example of what I mean.  Suppose client CARTER opens a file 
>    and requests an OpLock.  Samba on server SIMO grants the OpLock.  About 
>    three minutes later process POTTER running on server SIMO opens the 
>    same file for read-only access.
> 
>    Optimally, server SIMO would send an OpLock break to client CARTER 
>    downgrading the OpLock to a level2 OpLock, but that can only happen if 
>    the kernel can report that new open was read-only.
> 
>    Can Linux (or IRIX) kernel oplocks do that?

Well, if Samba does an open(file-name, O_RDONLY, ...), then yes, the 
kernel has the required information, and Samba should convert a read-only 
open into O_RDONLY. There are, however, some interactions between 
DENY_MODE locks and OpLocks that almost require NTCreate&X in the Kernel.

Regards
-----
Richard Sharpe, rsharpe[at]ns.aus.com, rsharpe[at]samba.org, 
sharpe[at]ethereal.com, http://www.richardsharpe.com