[Ocfs2-devel] [PATCH] fs: ocfs2: move_extents.c: Fix to remove null pointer checks that could never happen

[Andrew Morton]

On Thu, 29 May 2014 17:23:08 -0400 Dave Jones <davej at redhat.com> wrote:

> On Thu, May 29, 2014 at 02:03:37PM -0700, Andrew Morton wrote:
>  > On Tue, 27 May 2014 22:23:51 +0200 Rickard Strandqvist <rickard_strandqvist at spectrumdigital.se> wrote:
>  > 
>  > > Removal of null pointer checks that could never happen
>  > 
>  > How do you know it never happens?
>  > 
>  > > --- a/fs/ocfs2/move_extents.c
>  > > +++ b/fs/ocfs2/move_extents.c
>  > > @@ -904,9 +904,6 @@ static int ocfs2_move_extents(struct ocfs2_move_extents_context *context)
>  > >  	struct buffer_head *di_bh = NULL;
>  > >  	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
>  > >  
>  > > -	if (!inode)
>  > > -		return -ENOENT;
>  > > -
>  > 
>  > If it's due to assuming that the previous statement would have oopsed
>  > then that is mistaken.  Is is sometimes the case that gcc will move the
>  > evaluation of inode->i_sb to after the test, so this function can be
>  > passed NULL and it will not oops.
> 
> 'sometimes' ?
> 
> You have a lot more faith in gcc than I do. What happens if we decide to
> switch to llvm one day ? Can we guarantee every compiler will implement
> the same magic ?  This seems fragile as hell to me.
> 

Well yes.  There are two ways to go here:

a) work out if `inode' can legitimately be NULL.  If so, do

	struct ocfs2_super *osb;

	if (!inode)
		return -ENOENT;
	osb = OCFS2_SB(inode->i_sb);

   or

b) if `inode' cannot legitimately be NULL then Rickard's patch is OK.


My point is that we *cannot* assume that `inode' cannot be NULL from
observed runtime results.  Because of the compiler's behaviour.