Mapping large-scale underground environments, such as mines and tunnels, is typically a time-consuming and challenging endeavor. Existing methods based on terrestrial laser scanning are inefficient, and commercial mobile mapping systems are not suitable for underground use. Simultaneous Localization and Mapping (SLAM) solutions have the potential to efficiently survey underground environments while continuously moving, without relying on external positioning systems. However, no three-dimensional (3D) SLAM systems have thus far been demonstrated to be capable of mapping underground mines at kilometer scales. We have developed a solution that can accurately estimate, based on laser range and inertial measurements, the six-degrees-of-freedom trajectory of a sensor platform that continuously moves through an environment, as well as a 3D point cloud map of that environment. The key software components of the solution are continuous-time non-rigid registration, scalable place recognition, and robust pose graph optimization. A system consisting of a spinning 2D laser scanner and an industrial-grade inertial measurement unit mounted on a light vehicle was deployed at Northparkes Mine in Australia and used to map over 17 km of mine tunnel in 113 min while traveling at typical mine traffic speeds. Our processing software automatically produces a trajectory and point cloud of the mine in under half of the data acquisition time. The accuracy of the solution exceeds the raw sensor characteristics and matches closely to a surveyed map of the same environment.