Context. Massive stars play a vital role in the Universe. However, their evolution even on the main sequence is not yet well understood. Aims. Due to the steep mass-luminosity relation, massive main sequence stars become extremely luminous. This brings their envelopes very close to the Eddington limit. We are analysing stellar evolutionary models in which the Eddington limit is reached and exceeded, and explore the rich diversity of physical phenomena which take place in their envelopes, and we investigate their observational consequences. Methods. We use the published grids of detailed stellar models by Brott et al. (2011) andKöhler et al. (2015), computed with a state-of-the-art one-dimensional hydrodynamic stellar evolution code using LMC composition, to investigate the envelope properties of core hydrogen burning massive stars. Results. We find that at the stellar surface, the Eddington limit is almost never reached, even for stars up to 500 M ⊙ . When we define an appropriate Eddington limit locally in the stellar envelope, we can show that most stars more massive than ∼ 40 M ⊙ actually exceed this limit, in particular in the partial ionization zones of iron, helium or hydrogen. While most models adjust their structure such that the local Eddington limit is exceeded at most by a few per cent, our most extreme models do so by a factor of more than seven. We find that the local violation of the Eddington limit has severe consequences for the envelope structure, as it leads to envelope inflation, convection, density inversions and possibly to pulsations. We find that all models with luminosities higher than 4 × 10 5 L ⊙ , i.e. stars above ∼ 40 M ⊙ show inflation, with a radius increase of up to a factor of about 40. We find that the hot edge of the S Dor variability region coincides with a line beyond which our models are inflated by more than a factor of two, indicating a possible connection between S Dor variability and inflation. Furthermore, our coolest models show highly inflated envelopes with masses of up to several solar masses, and appear to be candidates to produce major LBV eruptions. Conclusions. Our models show that the Eddington limit is expected to be reached in all stars above ∼ 40 M ⊙ in the LMC, and by even lower mass stars in the Galaxy, or in close binaries or rapid rotators. While our results do not support the idea of a direct super-Eddington wind driven by continuum photons, the consequences of the Eddington limit in the form of inflation, pulsations and possibly eruptions may well give rise to a significant enhancement of the the time averaged mass loss rate.