Full-duplex (FD) relaying has been considered an effective scheme to increase the spectral efficiency of multiple- input multiple-output (MIMO) relay systems. As well-known, the main concern for the FD system is the cancellation of loop interference (LI). In this paper, we consider the joint source/relay precoding to mitigate the LI problem in FD-MIMO relay systems. In our system, spatial multiplexing is exploited for the signal transmission, and the QR successive-interference-cancellation (SIC) receiver is adopted at the destination. Linear precoders are considered at the source and relay, and the block error rate is used as the criterion for the precoders design. To facilitate the optimization, we propose using the primal decomposition, translating the original problem into a subproblem and a master problem. In the subproblem, the source precoder is first solved with the geometric mean decomposition (GMD) method. Then, the master problem can be formulated as a convex optimization so that the relay precoder can be solved with Karush-Kuhn-Tucker (KKT) conditions. The proposed precoders have closed-form expressions, facilitating real- world implementation. Simulation results show that the proposed method significantly improves the performance of FD-MIMO relay systems.