[Ocfs2-devel] [PATCH 5/5] ocfs2: Remove bogus link count decrement

[tristan.ye]

On Mon, 2009-01-12 at 23:20 +0100, Jan Kara wrote:
> In ocfs2_unlink() we decremented i_nlink three times for directories (instead
> of just two times). This actually did not matter because we have
> ocfs2_drop_inode() which decides whether the inode should be deleted based on a
> flag but it's at least surprising for a link count to wrap...
> 
> Signed-off-by: Jan Kara <jack at suse.cz>
> ---
>  fs/ocfs2/namei.c |    2 --
>  1 files changed, 0 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/ocfs2/namei.c b/fs/ocfs2/namei.c
> index 084aba8..892d5e7 100644
> --- a/fs/ocfs2/namei.c
> +++ b/fs/ocfs2/namei.c
> @@ -888,8 +888,6 @@ static int ocfs2_unlink(struct inode *dir,
>  	}
>  
>  	dir->i_ctime = dir->i_mtime = CURRENT_TIME;
> -	if (S_ISDIR(inode->i_mode))
> -		drop_nlink(dir);

I thought the logic of original ocfs2_unlink() has no problem, it seems
just drop the link count 2 times to release reference count for  '.' and
'..' in all for the deleting directory inode.

The code you've truncated is the logic to drop link count for its parent
directory, and it's a REQUIRED action, you may see it as
drop_nlink(inode) by mistake:-) ? 

Otherwise, VFS unlink() will hit a failure when deleting a dir which
containes sub_dir
entires.  Like following,

1. mkdir a

2. mkdir -p a/b

3. rm -rf a
rm: cannot remove directory `a/': Directory not empty


Tristan 

>  
>  	status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
>  	if (status < 0) {