From: Andreas Dilger <adilger@clusterfs.com>
Subject: Re: [RFC] [patch 2/3] change attribute for ext4: ext4 specific code
Date: Wed, 6 Dec 2006 14:49:34 -0700
Message-ID: <20061206214934.GA4551@schatzie.adilger.int>
References: <456DD75A.2010700@bull.net>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: linux-ext4@vger.kernel.org
To: Jean-Noel Cordenner <jean-noel.cordenner@bull.net>
Content-Disposition: inline
In-Reply-To: <456DD75A.2010700@bull.net>
Sender: linux-ext4-owner@vger.kernel.org

On Nov 29, 2006  19:54 +0100, Jean-Noel Cordenner wrote:
> This part of the patch concerns the ext4 code.

I was looking more closely at this code, and wondering two things:
- why not just use the existing inode->i_version field instead of
  adding a new i_change_attribute?  The i_version is not used by
  the VFS at all, and only for detecting directory modifications in
  ext3 (where it has the same semantic as the new i_change_attribute
  anyways).  This avoids bloating the VFS inode more than it already is.
- why not just do an increment of i_version in ext3_do_update_inode()?
  That is ext3_dirty_inode->ext3_mark_inode_dirty->ext3_mark_iloc_dirty()
  and also handles all of the VFS locations that call notify_change().
  This MUST be called anywhere that we make a persistent change to the
  inode in order to flush it to disk.  That would reduce the patch to
  a few lines at most.  I don't think there are any places we need to
  supplement this (even mmap IO or writes to a hole will update mtime).

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.