We present a unified
approach to different fluctuation relations for classical
nonequilibrium dynamics described by diffusion processes. Such
relations compare the
statistics of fluctuations of the entropy production or work in the
original process to
the similar statistics in the time-reversed process. We trace the
origin of a variety of
fluctuation relations to the use of different time reversals. The
systems considered
include, as special cases, the deterministic dynamics, the Langevin
equation, the linear
equation and the Kraichnan model of hydrodynamic flow. We discuss
briefly a peculiar
one-dimensional Langevin process in which the equilibrium state is
spontanously broken
and replaced by a flux state. The Jarzynski equation and the
fluctuation-dissipation
theorem are adapted to this situation.
The application of our approach to the tangent process describing the
joint evolution of
infinitesimally close trajectories of the original process leads to a
multiplicative
extension of the fluctuation relations.