We developed a hydraulic fracturing simulator that implicitly couples fluid flow with the stresses induced by fracture deformation in large, complex, three-dimensional discrete fracture networks. The simulator can describe propagation of hydraulic fractures and opening and shear stimulation of natural fractures. Fracture elements can open or slide, depending on their stress state, fluid pressure, and mechanical properties. Fracture sliding occurs in the direction of maximum resolved shear stress. Nonlinear empirical relations are used to relate normal stress, fracture opening, and fracture sliding to fracture aperture and transmissivity. Fluid leakoff is treated with a semianalytical one-dimensional leakoff model that accounts for changing pressure in the fracture over time. Fracture propagation is treated with linear elastic fracture mechanics. Non-Darcy pressure drop in the fractures due to high flow rate is simulated using Forchheimer's equation. A crossing criterion is implemented that predicts whether propagating hydraulic fractures will cross natural fractures or terminate against them, depending on orientation and stress anisotropy. Height containment of propagating hydraulic fractures between bedding layers can be modeled with a vertically heterogeneous stress field or by explicitly imposing hydraulic fracture height containment as a model assumption. The code is efficient enough to perform field-scale simulations of hydraulic fracturing with a discrete fracture network containing thousands of fractures, using only a single compute node. Limitations of the model are that all fractures must be vertical, the mechanical calculations assume a linearly elastic and homogeneous medium, proppant transport is not included, and the locations of potentially forming hydraulic fractures must be specified in advance. Simulations were performed of a single propagating hydraulic fracture with and without leakoff to validate the code against classical analytical solutions. Field-scale simulations were performed of hydraulic fracturing in a densely naturally fractured formation. The simulations demonstrate how interaction with natural fractures in the formation can help explain the high net pressures, relatively short fracture lengths, and broad regions of microseismicity that are often observed in the field during stimulation in low permeability formations, and which are not predicted by classical hydraulic fracturing models. Depending on input parameters, our simulations predicted a variety of stimulation behaviors, from long hydraulic fractures with minimal leakoff into surrounding fractures to broad regions of dense fracturing with a branching network of many natural and newly formed fractures.