Speed up lookups of an existing lock context by avoiding the inode->i_lock,
and using RCU instead.
Signed-off-by: NTrond Myklebust <trond.myklebust@hammerspace.com>