Re: NFS lockdep lock misordering mmap_sem<->i_mutex_key with 2.6.32-git1

[KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>]

> On Tue, Dec 15, 2009 at 10:21:34PM +0000, Al Viro wrote:
> > On Mon, Dec 07, 2009 at 02:20:09PM +0100, Andi Kleen wrote:
> > > >    nfs_readdir
> > > >      nfs_do_filldir
> > > >        filldir
> > > >          copy_to_user
> > > >            [page_fault]			[grab mmap_sem]
> > > > 
> > > >  sys_mmap				[grab mmap_sem]
> > > >    do_mmap_pgoff
> > > >      mmap_region
> > > >        nfs_file_mmap
> > > >          nfs_revalidate_mapping  
> > > >            nfs_invalidate_mapping	[grab i_mutex]
> > > > 
> > > > I guess recent lockdep improvement find old bug.
> > > 
> > > Thanks for the analysis.
> > > 
> > > I guess should never do copy_*_user while holding i_mutex? There might
> > > be lots of cases like that.
> > 
> > No.  mmap_sem inside i_mutex is the normal order; NFS mmap is doing the
> > wrong thing here.  Note that readdir() vs. NFS (file-only, thankfully ;-)
> > mmap() is a non-issue; NFS mmap() vs. write() is much more interesting.
> 
> I see.
> 
> > 
> > Again, a lot of mm/* code expects i_mutex, then mmap_sem order.  It's not
> > just readdir().
> 
> I suppose an easy workaround would be to not revalidate in mmap,
> because open should have already done that?
> 
> Very lightly tested RFC patch attached.
> 
> -Andi
> 
> ---
> 
> NFS: don't revalidate in mmap
> 
> nfs_revalidate_mapping takes i_mutex, but mmap already has mmap_sem
> hold and taking i_mutex inside mmap_sem is not allowed by the VFS.
> 
> So don't revalidate on mmap time and trust it has been already done.
> 
> Signed-off-by: Andi Kleen <ak@linux.intel.com>

afaik, no revalidate mean "I don't mind the file was already removed
at server, I only care client handle".

What's happen if we mmap removed file?