Re: [git pull] vfs pile 3

[Al Viro <viro@ZenIV.linux.org.uk>]

On Sat, Oct 13, 2012 at 12:04:55PM -0400, Christoph Hellwig wrote:
> On Sat, Oct 13, 2012 at 05:01:15PM +0100, Al Viro wrote:
> > You know, I'm in the middle of dealing with one such TODO.  Yours, as it
> > were.  From six years ago.  kernel_thread() unexporting.  TODO comments
> > of any form are routinely shat upon and ignored, especially when shuffled
> > away into less read parts of the tree... ;-/
> > 
> > I'd rather see it done fs-by-fs.  Starting with something reasonably easy
> > to test - minixfs would do nicely.  Don't get me wrong - I'm all for
> > burying ->truncate(); what I'm worried about is that we'll end up burying
> > the warning about the reasons why vmtruncate() was a bad idea, leaving the
> > functionality exactly as it used to be...
> 
> As mentioned I agree with the concern in principle.  Let's start by
> taking Marco's patches for filesystems that use vmtruncate but don't 
> actually implement ->truncate.  There's a few I remember offhand, e.g.
> procfs and ufs right now.  Then we can do the actual work required ones
> piece by piece.

Umm... That would be what, procfs?  Frankly, I'm not sure that ATTR_SIZE for
procfs actually should not be silently ignored.  ->i_size there is completely
synthetic - it's not as if truncation would actually change the contents.

And ufs situation is quite different - there vmtruncate() is used only on the
->write_begin() side.  ->setattr() is already vmtruncate-free.  What's needed
there is an analog of e.g. ext2_write_failed().