bfs_iget() reads i_nlink directly from disk without validating that
it is non-zero. A corrupted BFS image with an inode that has
i_nlink == 0 but is still referenced by a directory entry allows that
inode to be loaded. While bfs_unlink() has a recovery guard for this
case, other functions like bfs_rename() calls inode_dec_link_count()
without checking, triggering WARN_ON(inode->i_nlink == 0) in drop_nlink().

Reject inodes with zero link count at load time, consistent with
the approach used by ext4, minix, nilfs2, and other filesystems.

Signed-off-by: Ziyi Guo <n7l8m4@u.northwestern.edu>
---
 fs/bfs/inode.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/fs/bfs/inode.c b/fs/bfs/inode.c
index ce6f83234b67..85d664e169f0 100644
--- a/fs/bfs/inode.c
+++ b/fs/bfs/inode.c
@@ -96,6 +96,10 @@ struct inode *bfs_iget(struct super_block *sb, unsigned long ino)
 	i_uid_write(inode, le32_to_cpu(di->i_uid));
 	i_gid_write(inode,  le32_to_cpu(di->i_gid));
 	set_nlink(inode, le32_to_cpu(di->i_nlink));
+	if (!inode->i_nlink) {
+		brelse(bh);
+		goto error;
+	}
 	inode->i_size = BFS_FILESIZE(di);
 	inode->i_blocks = BFS_FILEBLOCKS(di);
 	inode_set_atime(inode, le32_to_cpu(di->i_atime), 0);
-- 
2.34.1

Hi,

On Sun, 8 Feb 2026 at 18:51, Ziyi Guo <n7l8m4@u.northwestern.edu> wrote:
> diff --git a/fs/bfs/inode.c b/fs/bfs/inode.c
> index ce6f83234b67..85d664e169f0 100644
> --- a/fs/bfs/inode.c
> +++ b/fs/bfs/inode.c
> @@ -96,6 +96,10 @@ struct inode *bfs_iget(struct super_block *sb, unsigned long ino)
>         i_uid_write(inode, le32_to_cpu(di->i_uid));
>         i_gid_write(inode,  le32_to_cpu(di->i_gid));
>         set_nlink(inode, le32_to_cpu(di->i_nlink));
> +       if (!inode->i_nlink) {
> +               brelse(bh);
> +               goto error;
> +       }
>         inode->i_size = BFS_FILESIZE(di);
>         inode->i_blocks = BFS_FILEBLOCKS(di);
>         inode_set_atime(inode, le32_to_cpu(di->i_atime), 0);

Looks fine to me, thank you.

Reviewed-by: Tigran Aivazian <aivazian.tigran@gmail.com>