Can You Trust Code Copilots? Evaluating LLMs from a Code Security Perspec