We consider the uplink of firefly ultra dense networks which combine the promising features of ultra dense deployment and centralized processing. In these networks, a large number of radio units, also referred to as firefly nodes (FNs), are spatially distributed over an area. The mobile devices (MDs) in the coverage area are simultaneously connected via sub-6 GHz radio frequency links to all FNs. Unlike the cloud radio access networks (C-RAN) architecture, in firefly ultra dense networks, the FNs forward the MDs' data through multi-hop mmWave links to one or multiple root nodes since the coverage radius of each mmWave link is limited. The root nodes forward the data via optical fiber links further to the central unit, where the MDs' signals are decoded. Since the amount of data received at each FN is potentially huge, efficient signal processing is needed at the FNs before the signals are forwarded to the next FNs. Therefore, we investigate locally-designed linear processing schemes, where all received signal streams are linearly filtered at each FN before being forwarded over an available mmWave link. Our simulation results reveal a significant performance improvement compared to a baseline scheme.