Re: [PATCH 02/11] vfs: Add better VFS support for page_mkwrite when blocksize < pagesize

[Christoph Hellwig <hch@infradead.org>]

On Fri, Jun 26, 2009 at 10:42:25AM +0200, Nick Piggin wrote:
> Yes well we could get rid of ->truncate and have filesystems do it
> themselves in setattr, but I figure that moving truncate into
> generic setattr is helpful (makes conversions a bit easier too).
> Did you see my patch? What do you think of that basic approach?

I was waiting for a patch series for your for this to appear, but
noticed that you actually had a small proof of concept patch attached,
sorry :)

Looking at your patch I really like that vmtruncate now really just
does what it's name claims to - truncate the VM-information about
the file (well, and the file size).   I'm not so happy about
still keeping the two level setattr/truncate indirection.

But instead of folding truncate into setattr I wonder if we should
just add a new ->setsize (aka new trunacte) methodas a top-level
entry point instead of ->setattr with ATTR_SIZE given that size
changes don't have much in common with the reset of ->setattr.

The only bit shared is updating c/mtime and even that is conditional.
So I'd say take most of your patch, but instead of doing an all at
once migration migrate filesystems to the new ->setsize callback
incrementally and eventually kill off the old code.  This means
we'll need a new name for the new vmtruncate-lite but should otherwise
be pretty easy.

> >  The only problem is the generic aops calling
> > vmtruncate directly.
> 
> What should be done is require that filesystems trim blocks past
> i_size in case of any errors. I actually need to fix up a few
> existing bugs in this area too, so I'll look at this..

Basically we want ->setattr with ATTR_SIZE, execept that we already
have i_sem and possibly other per-inode locks.  Take a look at

	http://oss.sgi.com/archives/xfs/2008-04/msg00542.html

and

	http://oss.sgi.com/archives/xfs/2009-03/msg00214.html