NeilBrown wrote:
> Hi Trond,
>  I wonder if you would consider the following two patches.
> 
> The first fixes a real bug that is trivial to demonstrate: script
> below.  The problem is that the fattr which is extracted from
> READDIRPLUS results does not get a ->time_start set.  This can result
> is incorrect attributes being set on files.
> 
> The second is really a work-around for a deeper problem (invalidate_inode_pages2
> racing with do_no_page) but has value in itself I think.  It extents the combination
>  -o nolock,nocto
> to avoid flush and invalidate on locking requests.

This sounds like the behavior enabled by the 'llock' mount option on 
Solaris.  Is that your intent?

IMO it is a clever idea to enable 'llock' behavior on Linux with the 
combination of already existing mount options.