We study the redshift evolution of the luminosity function (LF) and redshift selection effect of long gamma-ray bursts (LGRBs). The method is to fit the observed peak flux and redshift distributions, simultaneously. To account for the complex triggering algorithm of Swift, we use a flux triggering efficiency function. We find evidence supporting an evolving LF, where the break luminosity scales as $L_b∝ (1+z)^{τ}$, with $τ =3.5^{+0.4}_{-0.2}$ and $τ =0.8^{+0.1}_{-0.08}$ for two kind of LGRB rate models. The corresponding local GRB rates are $\dot{R}(0)=0.86^{+0.11}_{-0.08} \yr^{-1}\Gpc^{-3}$ and $\dot{R}(0)= 0.54^{+0.25}_{-0.07} \yr^{-1}\Gpc^{-3}$, respectively. Furthermore, by comparing the redshift distribution between the observed one and our mocked one, we find that the redshift detection efficiency of the flux triggered GRBs decreases with redshift. Especially, a great number of GRBs miss their redshifts in the redshift range of $1