Millimeter-wave (mmWave) radars have found applications in a wide range of domains, including human tracking, health monitoring, and autonomous driving, for its unobtrusive nature and high range accuracy. These capabilities, however, if used for malicious purposes, could also lead to serious security and privacy issues. For example, a user's daily life could be secretly monitored by a spy radar. Hence, there is a strong urge to develop systems that can detect and localize such spy radars. In this paper, we propose $Radar^2$, a practical passive spy radar detection and localization system using a single commercial off-the-shelf (COTS) mmWave radar. Specifically, we propose a novel \textit{Frequency Component Detection} method to detect the existence of mmWave signal, distinguish between mmWave radar and WiGig signals using a convolutional neural network (CNN) based waveform classifier, and localize spy radars using the trilateration method based on the detector's observations at multiple anchor points. Not only does $Radar^2$ work for different types of mmWave radar, but it can also detect and localize multiple radars simultaneously. Finally, we perform extensive experiments to evaluate the effectiveness and robustness of $Radar^2$ in various settings. Our evaluation shows that the radar detection rate is constantly above 96$\%$ and the localization error is within 0.3m. The results also reveal that $Radar^2$ is robust against various factors (room layout, human activities, etc.).