Within the standard 3nu mass-mixing framework, we present an up-to-date global analysis of neutrino oscillation data (as of January 2016), including the latest available results from experiments with atmospheric neutrinos (Super-Kamiokande and IceCube DeepCore), at accelerators (first T2K anti-nu and NOvA nu runs in both appearance and disappearance mode), and at short-baseline reactors (Daya Bay and RENO far/near spectral ratios), as well as a reanalysis of older KamLAND data in the light of the "bump" feature recently observed in reactor spectra. We discuss improved constraints on the five known oscillation parameters (delta m^2, |Delta m^2|, sin^2theta_12, sin^2theta_13, sin^2theta_23), and the status of the three remaining unknown parameters: the mass hierarchy, the theta_23 octant, and the possible CP-violating phase delta. With respect to previous global fits, we find that the reanalysis of KamLAND data induces a slight decrease of both delta m^2 and sin^2theta_12, while the latest accelerator and atmospheric data induce a slight increase of |Delta m^2|. Concerning the unknown parameters, we confirm the previous intriguing preference for negative values of sin(delta) [with best-fit values around sin(delta) ~ -0.9], but we find no statistically significant indication about the theta_23 octant or the mass hierarchy (normal or inverted). Assuming an alternative (so-called LEM) analysis of NOvA data, some delta ranges can be excluded at >3 sigma, and the normal mass hierarchy appears to be slightly favored at 90% C.L. We also describe in detail the covariances of selected pairs of oscillation parameters. Finally, we briefly discuss the implications of the above results on the three non-oscillation observables sensitive to the (unknown) absolute nu mass scale: the sum of nu masses, the effective nu_e mass, and the effective Majorana mass.