Re: 2.6.19-rc2-mm1

[Andrew Morton <akpm@osdl.org>]

On Thu, 19 Oct 2006 00:41:40 +0200 (CEST)
Jiri Kosina <jikos@jikos.cz> wrote:

> On Wed, 18 Oct 2006, Andrew Morton wrote:
> 
> > > The inode->i_mutex should be held every time when calling 
> > > i_size_write(), and the function contains WARN_ON() for that 
> > > condition. page_symlink(), however, does not lock i_mutex. It is 
> > > perfectly OK, as the i_mutex for the directory is held at the time 
> > > page_symlink() is running, so noone is able to change i_size during 
> > > race condition. However, i_size_write() spits out the warning without 
> > > this patch.
> > I suspect it isn't necessary because the symlink's inode hasn't been wired
> > up into the directory tree yet and no other thread can find it and do
> > things to it.
> 
> I completely agree (see my comments to the patch in previous mail). 
> However, the warning emitted by i_size_write() should really go away. I 
> can see the following possibilities:
> 
> - lock the i_mutex, even though it's for sure not necessary. Not nice.
> - remove the warning from i_size_write(). Not nice either, we want to be 
> warned about other calls that are not correct
> - make the warning in i_size_write() conditional on inode->i_dentry not 
> being NULL (?)
> 

I simply dropped the debugging patch.  Which is pretty sad, because it _is_ a
really nasty and subtle-to-show bug.  So I'd be OK with adding sufficient
patches to -mm to make the false positives go away so we can re-add the
check.