Re: invalidate_inode_pages in 2.5.32/3

[Daniel Phillips <phillips@arcor.de>]

On Tuesday 10 September 2002 00:32, Andrew Morton wrote:
> Daniel Phillips wrote:
> > 
> > > > void invalidate_inode_pages(struct inode *inode)
> > > > {
> > > >         truncate_inode_pages(mapping, 0);
> > > > }
> > > >
> > > > Is it any harder than that?
> > >
> > > Pretty much - need to leave i_size where it was.
> > 
> > This doesn't touch i_size.
> 
> Sorry - I was thinking vmtruncate(). truncate_inode_pages() would
> result in all the mmapped pages becoming out-of-date anonymous
> memory.  NFS needs to take down the pagetables so that processes
> which are mmapping the file which changed on the server will take
> a major fault and read a fresh copy.  I believe.

Oh, um.  Yes, we need the additional pte zapping behaviour of 
vmtruncate_list.  It doesn't look particularly hard to produce a
variant of vmtruncation that does (doesn't do) what you suggest.
Let's see how the discussion goes with the NFS crowd.

-- 
Daniel