I am trying to show for the one-loop integral with three propagators with different internal masses $m_1$, $m_2$, $m_3$, and all off-shell external momenta $p_1$, $p_2$, $p_3$ the following formula appearing in 't Hooft(1979), Bardin (1999), Denner (2007): (unfortunate metric $-,+,+,+$)
$$\int d^d q\frac{1}{(q^2+m_1^2)((q+p1)^2+m_2^2)((q+p_1+p_2)^2+m_3^2)} $$ $$=i\pi^2\int_0^1dx\int_0^xdy\frac{1}{ax^2+by^2+cxy+dx+ey+f}$$
where $a$, $b$, $c$, ... are coefficients depending on the momenta in the following way:
$a=-p_2^2$,
$b=-p_1^2$,
$c=-2p_1.p_2$,
$d=m_2^2-m_3^2+p_2^2$,
$e=m_1^2-m_2^2+p_1^2+2(p_1.p_2)$,
$f=m_3^2-i\epsilon$.
I don't really care about factors in fromt like $i\pi^2$. My simple problem is: I am totally unable to reproduce coefficients $d$, $e$ and $f$. The problem is, when I integrate over the third Feynman parameter, $m_3$ appears in all three coefficients $d$, $e$ and $f$. How do I squeeze the denominators to reproduce this formula?