Total effect versus direct effect when you control a mediator

In a causal system, a treatment's total effect on the outcome is $5$. Of that, $3$ flows indirectly through a mediator $m$ (treatment moves $m$, $m$ moves the outcome), and the remaining $2$ is a direct effect that does not pass through $m$.

If you control for $m$ in your regression, what does the treatment coefficient estimate, and how far is it from the total effect you intended to measure?