Re: [PATCH v2] xfs: fix an ABBA deadlock in xfs_rename

[Brian Foster <bfoster@redhat.com>]

On Thu, Jan 07, 2021 at 05:56:48PM -0800, Darrick J. Wong wrote:
> From: Darrick J. Wong <darrick.wong@oracle.com>
> 
> When overlayfs is running on top of xfs and the user unlinks a file in
> the overlay, overlayfs will create a whiteout inode and ask xfs to
> "rename" the whiteout file atop the one being unlinked.  If the file
> being unlinked loses its one nlink, we then have to put the inode on the
> unlinked list.
> 
> This requires us to grab the AGI buffer of the whiteout inode to take it
> off the unlinked list (which is where whiteouts are created) and to grab
> the AGI buffer of the file being deleted.  If the whiteout was created
> in a higher numbered AG than the file being deleted, we'll lock the AGIs
> in the wrong order and deadlock.
> 
> Therefore, grab all the AGI locks we think we'll need ahead of time, and
> in order of increasing AG number per the locking rules.  Set
> t_firstblock so that a subsequent directory block allocation never tries
> to grab a lower-numbered AGF than the AGIs we grabbed.
> 

I don't see this patch do anything with t_firstblock... Even so, is that
necessary? I thought we had to lock AGIs before AGFs and always in agno
order, but not necessarily lock an AGF >= previously locked AGIs. Hm?

> Reduce the likelihood that a directory expansion will ENOSPC by starting
> with AG 0 when allocating whiteout files.
> 
> Reported-by: wenli xie <wlxie7296@gmail.com>
> Fixes: 93597ae8dac0 ("xfs: Fix deadlock between AGI and AGF when target_ip exists in xfs_rename()")
> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> ---
> v2: Make it more obvious that we're grabbing all the AGI locks ahead of
> the AGFs, and hide functions that we don't need to export anymore.
> ---
>  fs/xfs/libxfs/xfs_dir2.h    |    2 --
>  fs/xfs/libxfs/xfs_dir2_sf.c |    2 +-
>  fs/xfs/xfs_inode.c          |   57 ++++++++++++++++++++++++++++++-------------
>  3 files changed, 41 insertions(+), 20 deletions(-)
> 
> diff --git a/fs/xfs/libxfs/xfs_dir2.h b/fs/xfs/libxfs/xfs_dir2.h
> index e55378640b05..d03e6098ded9 100644
> --- a/fs/xfs/libxfs/xfs_dir2.h
> +++ b/fs/xfs/libxfs/xfs_dir2.h
> @@ -47,8 +47,6 @@ extern int xfs_dir_lookup(struct xfs_trans *tp, struct xfs_inode *dp,
>  extern int xfs_dir_removename(struct xfs_trans *tp, struct xfs_inode *dp,
>  				struct xfs_name *name, xfs_ino_t ino,
>  				xfs_extlen_t tot);
> -extern bool xfs_dir2_sf_replace_needblock(struct xfs_inode *dp,
> -				xfs_ino_t inum);
>  extern int xfs_dir_replace(struct xfs_trans *tp, struct xfs_inode *dp,
>  				struct xfs_name *name, xfs_ino_t inum,
>  				xfs_extlen_t tot);
> diff --git a/fs/xfs/libxfs/xfs_dir2_sf.c b/fs/xfs/libxfs/xfs_dir2_sf.c
> index 2463b5d73447..8c4f76bba88b 100644
> --- a/fs/xfs/libxfs/xfs_dir2_sf.c
> +++ b/fs/xfs/libxfs/xfs_dir2_sf.c
> @@ -1018,7 +1018,7 @@ xfs_dir2_sf_removename(
>  /*
>   * Check whether the sf dir replace operation need more blocks.
>   */
> -bool
> +static bool
>  xfs_dir2_sf_replace_needblock(
>  	struct xfs_inode	*dp,
>  	xfs_ino_t		inum)
> diff --git a/fs/xfs/xfs_inode.c b/fs/xfs/xfs_inode.c
> index b7352bc4c815..528152770dc9 100644
> --- a/fs/xfs/xfs_inode.c
> +++ b/fs/xfs/xfs_inode.c
> @@ -3000,6 +3000,24 @@ xfs_rename_alloc_whiteout(
>  	return 0;
>  }
>  
> +/* Decide if we need to lock the target IP's AGI as part of a rename. */
> +static inline bool
> +xfs_rename_lock_target_ip(
> +	struct xfs_inode	*src_ip,
> +	struct xfs_inode	*target_ip)
> +{
> +	unsigned int		tgt_nlink = VFS_I(target_ip)->i_nlink;
> +	bool			src_is_dir = S_ISDIR(VFS_I(src_ip)->i_mode);
> +
> +	/*
> +	 * We only need to lock the AGI if the target ip will end up on the
> +	 * unlinked list.
> +	 */
> +	if (src_is_dir)
> +		return tgt_nlink == 2;

IIUC, we can't rename dirs over nondirs and vice versa. I.e.:

# mv dir file
mv: overwrite 'file'? y
mv: cannot overwrite non-directory 'file' with directory 'dir'
# mv -T file dir
mv: overwrite 'dir'? y
mv: cannot overwrite directory 'dir' with non-directory

That means that if src_is_dir is true, target must either be NULL or an
empty directory for the rename to succeed, right? If so, have we already
enforced that by this point? I'm wondering if we need to check the link
count == 2 in this case or could possibly just reduce the logic to
(tgt_nlink == 1 || src_is_dir) And if so, whether that still warrants an
inline helper (where I suppose we could still assert that nlink == 2).

Brian

> +	return tgt_nlink == 1;
> +}
> +
>  /*
>   * xfs_rename
>   */
> @@ -3017,7 +3035,7 @@ xfs_rename(
>  	struct xfs_trans	*tp;
>  	struct xfs_inode	*wip = NULL;		/* whiteout inode */
>  	struct xfs_inode	*inodes[__XFS_SORT_INODES];
> -	struct xfs_buf		*agibp;
> +	int			i;
>  	int			num_inodes = __XFS_SORT_INODES;
>  	bool			new_parent = (src_dp != target_dp);
>  	bool			src_is_directory = S_ISDIR(VFS_I(src_ip)->i_mode);
> @@ -3130,6 +3148,27 @@ xfs_rename(
>  		}
>  	}
>  
> +	/*
> +	 * Lock the AGI buffers we need to handle bumping the nlink of the
> +	 * whiteout inode off the unlinked list and to handle dropping the
> +	 * nlink of the target inode.  We have to do this in increasing AG
> +	 * order to avoid deadlocks, and before directory block allocation
> +	 * tries to grab AGFs.
> +	 */
> +	for (i = 0; i < num_inodes && inodes[i] != NULL; i++) {
> +		if (inodes[i] == wip ||
> +		    (inodes[i] == target_ip &&
> +		     xfs_rename_lock_target_ip(src_ip, target_ip))) {
> +			struct xfs_buf	*bp;
> +			xfs_agnumber_t	agno;
> +
> +			agno = XFS_INO_TO_AGNO(mp, inodes[i]->i_ino);
> +			error = xfs_read_agi(mp, tp, agno, &bp);
> +			if (error)
> +				goto out_trans_cancel;
> +		}
> +	}
> +
>  	/*
>  	 * Directory entry creation below may acquire the AGF. Remove
>  	 * the whiteout from the unlinked list first to preserve correct
> @@ -3182,22 +3221,6 @@ xfs_rename(
>  		 * In case there is already an entry with the same
>  		 * name at the destination directory, remove it first.
>  		 */
> -
> -		/*
> -		 * Check whether the replace operation will need to allocate
> -		 * blocks.  This happens when the shortform directory lacks
> -		 * space and we have to convert it to a block format directory.
> -		 * When more blocks are necessary, we must lock the AGI first
> -		 * to preserve locking order (AGI -> AGF).
> -		 */
> -		if (xfs_dir2_sf_replace_needblock(target_dp, src_ip->i_ino)) {
> -			error = xfs_read_agi(mp, tp,
> -					XFS_INO_TO_AGNO(mp, target_ip->i_ino),
> -					&agibp);
> -			if (error)
> -				goto out_trans_cancel;
> -		}
> -
>  		error = xfs_dir_replace(tp, target_dp, target_name,
>  					src_ip->i_ino, spaceres);
>  		if (error)
>