Federated learning, a potent paradigm for collaborative machine learning across multiple parties, offers significant promise for contemporary industries. Nonetheless, its collaborative essence necessitates addressing concerns pertaining to data security and privacy. Sensitive user information, encompassing preferences, behaviors, and identities, remains vulnerable to adversarial analysis, thereby revealing the inadequacies of conventional privacy preservation strategies within federated learning frameworks. To mitigate these challenges, this paper proposes GSFL, an innovative federated learning architecture that amalgamates smart contracts with group signatures. GSFL facilitates secure and reliable distributed machine learning data sharing, while concurrently bolstering privacy protection. Furthermore, its enhanced decentralization fosters greater user participation in federated learning initiatives. Empirical analysis and testing validate GSFL's efficacy in satisfying the prerequisites for data sharing and privacy preservation in federated learning contexts.