Protecting location privacy in mobile services has recently received significant consideration as Location-Based Service (LBS) can reveal user locations to attackers. A problem in the existing cloaking schemes is that location vulnerabilities may be exposed when an attacker exploits a street map in their attacks. While both real and synthetic trajectories are based on real street maps, most of previous cloaking schemes assume free space movements to define the distance between users, resulting in the mismatch between privacy models and user movements. In this paper, we present MeshCloak, a novel map-based model for personalized location privacy, which is formulated entirely in map-based setting and resists inference attacks at a minimal performance overhead. The key idea of MeshCloak is to quickly build a sparse constraint graph based on the mutual coverage relationship between queries by pre-computing the distance matrix and applying quadtree search. MeshCloak also takes into account real speed profiles and query frequencies. We evaluate the efficiency and effectiveness of the proposed scheme via a suite of carefully designed experiments on five real maps.