In fifth-generation ultradense heterogeneous network, backhaul plays an important role to provide connection to core networks for small base stations (SBSs). With the densification, nonideal backhaul will be deployed extensively, which constrains the throughput improvement due to the limited backhaul capacity. This paper models the SBSs and users as independent Poisson point processes and derives a semi-closed-form expression that analyzes the ergodic throughput of the network where SBSs have limited backhaul capacity. Based on analysis, to accomplish flexible coverage with limited backhaul, an optimization problem is formulated which accounts for cell association and aims at maximizing the number of offloading users while simultaneously minimizing the biasing factors and guaranteeing the received signal-to-interference-and-noise ratio of users when offloading. The optimization problem is shown to be a mixed-integer nonlinear problem, which can obtain a suboptimal solution by convex relaxation and further a distributed solution by problem decomposition via primal-dual method. Considering the complexity of ultradense network and practical implementation, a backhaulaware η-optimal biasing adjustment model is proposed for flexible coverage. This model aims to optimize the coverage for throughput improvement while matching the backhaul capacity. The simulation results show that average user and SBS throughput can be improved significantly by 2× when backhaul is constrained. Due to the transmit power and interference constraints on access link, there is no need to allocate too much bandwidth to backhaul, which provides guidance for self-backhaul resource allocation.