Log is important to the database system, which is the foundation of maintaining the correctness and consistency. The existing database log mechanism only stores the history of transactions, but can not record the relationship between transactions. Facing the attack, databases with traditional log system can only stop the service of database and recover to the point of attack occurrence. This kind of recovery will abandon all the transactions after the malicious transaction regardless of whether these transactions are related to the malicious transaction. That means the database system is out of service between the fault-point to the end of recovery. By using this vulnerability, the attacker can commit malicious transactions constantly and the database will always be in the state of recovery. In this paper, we present a new model of transaction recovery log and intrusion response. We use ASM to describe the model, give a formal definition of transaction dependency and prove the correctness and categoricalness of the model. Databases with transaction recovery log and intrusion response mechanism roll back only affected transactions rather than all the transactions after malicious attack. This method will not stop the service of the database system, significantly enhancing the performance of recovery for defensive information warfare.