It could go both ways. There might be buffered data from the forked
program that ought to be flushed, or buffered data from the original
that ought not to be flushed. I guess I should fflush() stdout and
stderr before calling _exit().
Its kinda a weird situation. I only exec*() if I actually have to run
ssh, and clearly one would want to _exit() after a failed exec*(). It
is less clear what should be done when I decide not to exec*().
-Matt
