Age at death distributions provide an informative description of the mortality pattern of a population, and they are well suited to study longevity and lifespan inequality. However, they have generally been neglected for modeling and forecasting mortality. In this article, we use age at death distributions to model and forecast the adult age pattern of mortality. In particular, we introduce a segmented linear model that relates a fixed "standard" to a series of observed distributions by a transformation of the age axis uniquely dependent on the modal age at death and the variability of the distributions. This approach is parsimonious and efficient: using only three parameters, it allows capturing the compression and shifting dynamics of mortality, portraying mortality developments and constructing life table functions. In addition, mortality forecasts can be derived from univariate time series models of the three parameters. We illustrate the methodology by estimating and forecasting the distribution and remaining life expectancy at age 30 of four high-longevity countries from 1980 to 2030. We show that the model performs well in terms of goodness-of-fit, and that life expectancy forecasts reflect well the past linear increase and are more optimistic than the Lee-Carter forecasts.