Adversarial Tuning: Defending Against Jailbreak Attacks for LLMs