Reason from Fallacy: Enhancing Large Language Models' Logical Reasoning through Logical Fallacy Understanding