In recent years, researchers and practitioners have been studying the impact of test smells on test maintenance. However, there is still limited empirical evidence on why developers remove test smells in software maintenance and the mechanism employed for addressing test smells. In this paper, we conduct an empirical study on 12 real-world open-source systems to study the evolution and maintenance of test smells, and how test smells are related to software quality. Our results show that: 1) Although the number of test smell instances increases, test smell density decreases as systems evolve. 2) However, our qualitative analysis on those removed test smells reveals that most test smell removal (83%) is a by-product of feature maintenance activities. 45% of the removed test smells relocate to other test cases due to refactoring, while developers deliberately address the only 17% of the test smell instances, consisting of largely Exception Catch/Throw and Sleepy Test. 3) Our statistical model shows that test smell metrics can provide additional explanatory power on post-release defects over traditional baseline metrics (an average of 8.25% increase in AUC). However, most types of test smells have a minimal effect on post-release defects. Our study provides insight into how developers resolve test smells and current test maintenance practices. Future studies on test smells may consider focusing on the specific types of test smells that may have a higher correlation with defect-proneness when helping developers with test code maintenance.