[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: Extended Attributes and Access Control Lists



On Fri, Nov 02, 2001 at 04:15:27PM +0000, Stephen C. Tweedie wrote:
> Hi,
> 
> On Fri, Nov 02, 2001 at 09:08:12AM -0700, Peter J. Braam wrote:
> 
> > i_generation != i_version, the latter is really a version
> > counter, the former is an incarnation counter -- completely different
> > viz-a-viz NFS for example.
> 
> i_version may give you better caching by letting you know when the
> object has changed.  Is there anything you'd want i_version for that
> you can't do equally well with monotonically-increasing
> nanosecond-resolution ctime timestamps?

The ctime is changed by generic_file_write (I wonder why in fact). So
it reflects the data version of the file.  The i_version, when
properly used, could version the inode metadata.

A nano second ctime would be great, an independently managed i_version
for the inode changes only would also be very useful.

> 
> > >is held on disk in the inode, 
> > > and ".." is in the
> > > first directory data block
> > 
> > That's not what I want -- it's precisely the elimination of directory
> > data reads for .. that I want. 
> 
> The VFS API problem remains --- ".." lookup never actually gets as far
> as the on-disk filesystem in most cases these days.

The NFS server in fact has to do this (nfsd_find_parent) - when the
dentry corresponding to the file handle was purged from the cache.

_ Peter _

> 
> Cheers,
>  Stephen

-- 





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]