If you trap during one of your update calls the canister will be restored to the state before the call happened (every await is also a commit point), but I think you’re not asking about this case.
Right now, you need to implement custom logic to do rollbacks. There’s some good ideas in this thread. AFAIK we intend to implement canister backup (downloading complete canister state) and restore operations soon™ but no concrete work has started yet.