Cognitive radio (CR) and full-duplex (FD) have received extensive attention and research due to their high spectrum efficiency, which can effectively solve the problem of low spectrum efficiency in current communication systems. Based on CR and FD techniques, in this paper, a FD spectrum sharing CR networks is considered, in which both secondary users (SU1 and SU2) are each equipped with dual antennas, one antenna is used to transmit signals, and the other antenna is used to receive signals at the same time and frequency. Under peak interference power and peak transmit power constraints, we analysis the ergodic sum capacity and the outage probability based on the FD spectrum sharing CR networks and the conventional spectrum sharing CR networks. Furthermore, under no peak transmit power constrain and perfect self-interference cancellation (SIC), based on the FD spectrum sharing CR networks and the conventional spectrum sharing CR networks, the closed-form expressions of the theoretical performance upper bound of the ergodic sum capacity and the outage probability are derived by two lemmas and four propositions. Accurate mathematical analysis display, under the same bandwidth, the upper bound of ergodic sum capacity for the full-duplex spectrum sharing CR networks is twice as much as the traditional spectrum sharing CR networks, and the FD spectrum sharing CR networks based on SU1, also has better performance upper bound on the outage probability than the traditional spectrum sharing CR networks. Simulations results also validate that, the FD spectrum sharing CR networks obtains better communication performance than the conventional spectrum sharing CR networks, especially when the mean of self-interference channel power gain is small. Finally, we also can see that the simulation performance upper bound is completely consistent with the theoretical analysis performance upper bound, whether in the FD spectrum sharing CR networks or the conventional spectrum sharing CR networks. So also verifies the correctness of the theoretical performance upper bound derivation.