Re: BUG fs/dcache.c:595 in 2.4.24rc3-git3 during NFS umount

[Trond Myklebust]

On Mon, 2007-12-10 at 19:57 +0100, Andi Kleen wrote:
> > Hi Andi,
> > 
> > Does the attached patch fix the Oops?
> 
> Nope, got that a few hours after boot again:
> 
> -Andi

Hi Andi,

I appear to have misread d_find_alias(). It would seem that the only way
to ensure that a mountpoint won't be found is to remove it altogether
from the inode->i_dentry list. AFAICS that should be largely harmless
since the nfs sb->s_root is never visible to users, and is never part of
a dentry tree.

Cheers
  Trond

> --- Begin Message ---
> Ensure that the dummy 'root dentry' is invisible to d_find_alias(). If not,
> then it may be spliced into the tree if a parent directory from the same
> filesystem gets mounted at a later time.
>
> Signed-off-by: Trond Myklebust <Trond.Myklebust@xxxxxxxxxx>
> ---
>
>  fs/nfs/getroot.c |   11 +++++++++++
>  1 files changed, 11 insertions(+), 0 deletions(-)
>
> diff --git a/fs/nfs/getroot.c b/fs/nfs/getroot.c
> index 0ee4384..e6242cd 100644
> --- a/fs/nfs/getroot.c
> +++ b/fs/nfs/getroot.c
> @@ -57,6 +57,17 @@ static int nfs_superblock_set_dummy_root(struct super_block *sb, struct inode *i
>  		}
>  		/* Circumvent igrab(): we know the inode is not being freed */
>  		atomic_inc(&inode->i_count);
> +		/*
> +		 * Ensure that this dentry is invisible to d_find_alias().
> +		 * Otherwise, it may be spliced into the tree by
> +		 * d_materialise_unique if a parent directory from the same
> +		 * filesystem gets mounted at a later time.
> +		 * This again causes shrink_dcache_for_umount_subtree() to
> +		 * Oops, since the test for IS_ROOT() will fail.
> +		 */
> +		spin_lock(&dcache_lock);
> +		list_del_init(&sb->s_root->d_alias);
> +		spin_unlock(&dcache_lock);
>  	}
>  	return 0;
>  }
>
> --- End Message ---