The return of Eratosthenes: Secure Generation of RSA Moduli using Distributed Sieving

Abstract

Secure multiparty generation of an RSA biprime is a challenging task, which increasingly receives attention, due to the numerous privacy-preserving applications that require it. In this work, we construct a new protocol for the RSA biprime generation task, secure against a malicious adversary, who can corrupt any subset of protocol participants. Our protocol is designed with generic multiparty computation (MPC), making it both platform-independent and allowing for weaker security models to be assumed (e.g., honest majority), should the application scenario require it. By carefully "postponing" the check of possible inconsistencies in the shares provided by malicious adversaries, we achieve noteworthy efficiency improvements. Concretely, we are able to produce additive sharings of the prime candidates, from multiplicative sharings via a semi-honest multiplication, without degrading the overall (active) security of our protocol. This is the core of our sieving technique, increasing the probability of our protocol sampling a biprime. Similarly, we perform the first biprimality test, requiring several repetitions, without checking input share consistency, and perform the more costly consistency check only in case of success of the Jacobi symbol based biprimality test. Moreover, we propose a protocol to convert an additive sharing over a ring, into an additive sharing over the integers. Besides being a necessary subprotocol for the RSA biprime generation, this conversion protocol is of independent interest. The cost analysis of our protocol demonstrated that our approach improves the current state-of-the-art (Chen et al.-Crypto 2020), in terms of communication efficiency. Concretely, for the two-party case with malicious security, and primes of 2048 bits, our protocol improves communication by a factor of similar to 37.