Re: [patch 2.6.8.1] BSD accounting: update chars transferred value

[Guillaume Thouvenin <guillaume.thouvenin@bull.net>]

On Wed, Sep 08, 2004 at 10:17:04AM +0100, Christoph Hellwig wrote:
> On Wed, Sep 08, 2004 at 11:06:57AM +0200, Guillaume Thouvenin wrote:
> > Hello,
> > 
> >   The goal of this patch is to improve BSD accounting by using what 
> > is done in the CSA into BSD accounting. The final goal is to have a
> > uniform accounting structure. 
> > 
> >   This patch updates information given by BSD accounting concerning
> > bytes read and written. A field is already present in the BSD accounting
> > structure but it is never updated. We don't add information about blocks
> > read and written because, as it was discussed in previous email, the 
> > information is inaccurate. Most writes which are flushed delayed would 
> > get accounted to pdflushd. Thus, one solution to get this kind of 
> > information is to add counters when the write back is performed. The 
> > problem is that we don't know how to get information about the IID at 
> > the page level (ie from struct page). So we remove this information for 
> > the moment and it will be provided in another patch.
> > 
> > Changelog:
> >   
> >   - Adds two counters in the task_struct (rchar, wchar)
> >   - Init fields during the creation of the process (during the fork)
> >   - File I/O operations are done through sys_read(), sys_write(), 
> >     sys_readv(), sys_writev() and sys_sendfile(). Thus we increment 
> >     counters into those functions except with sys_sendfile(). For the
> >     latter, the incrementation is done in the do_sendfile() because this
> >     routine be directly called from the return of sys_sendfile().
> 
> I think it should be done in vfs_readv/vfs_write.  Else we'll miss the
> updates from inekrnel consumers like nfsd.

Yes you're right. vfs_readv() and vfs_writev() called do_readv_writev().
Thus I will send a new patch which updates counters in functions
vfs_read(), vfs_write(), do_readv_writev() and sys_sendfile()