hi Darrick, > Also -- what is the value of i_mtime (ea inode parent pointer) when the > refcount > 1? In the current implementation, xattr inode format is assumed to be in the new format (refcount in i_ctime and l_i_version, hash in i_atime). parent xattr e_hash field incorporates xattr inode hash to achieve strong binding. If hash validation fails, we fallback to old binding validation (xattr inode i_mtime == parent ino && xattr inode i_generation == parent i_generation). The idea is to allow migration from old lustre disks. So, if refcount is valid then value in i_mtime is not defined.
