Smart contract vulnerability detection has been a hot research topic. One of the popular detecting methods is symbolic execution. However, as reported, existing symbolic tools cost too much time, since they need to execute all paths to detect vulnerabilities. Thus their accuracy is limited by time. To attack this problem, we propose Park, a first general framework of parallel-fork symbolic execution for smart contracts. The main idea is to use multiple processes during symbolic execution, leveraging multiple CPU cores to enhance efficiency. Firstly, we propose a fork-operation based dynamic forking algorithm to achieve parallel symbolic contract execution. Secondly, against the SMT performance loss problem in parallelization, we propose an adaptive processes restriction and adjustment algorithm. Thirdly, we design a shared-memory based global variable reconstruction method to collect and rebuild the global variables from different processes. We implement Park as a plug-in and apply it to two popular symbolic tools of smart contracts: Oyente and Mythril. Experimental results under third-party datasets show that Park-Oyente and Park-Mythril can provide up to 6.84x and 7.49x speedup compared to original tools on 16 CPU cores.