When a regressor is contemporaneously correlated with the disturbance, then classical linear regression assumptions are violated and you should not use ordinary regression to model the data. However, there might exist an instrumental variable, that is, one which is correlated with the independent variable and is contemporaneously uncorrelated with the disturbance. In this case you can proceed with estimation via an instrumental variable estimator which has desirable statistical properties not held by the OLS estimator. 2SLS uses an instrumental variables approach to find parameter estimates in a simultaneous equation model.
