Let $G = (V, E)$ be a directed acyclic graph. Let every node $v \in V$ have an additional field $v_d$.
For each vertex $v \in V$, we need to store in $v_d$ the length of the longest path in $G$ that begins at $v$. The length of a path is given by the number of edges on this path.
How can i modify the above algorithm to find the u.dist value correctly for each node u.
I'm not going to tell you the exact modifications you need to do to the pseudo-code, but here is the high-level idea, which also shows why the longest-path problem on DAGs can be solved with just a DFS visit.
Forget about the visit for a second and suppose that you somehow know a topological order of the vertices of the graph. Let this order be $v_1, v_2, \dots, v_n$. Let $\ell(v_i)$ be the length of the longest path starting from $v_i$.
If $v_i$ is a sink we clearly have $\ell(v_i) = 0$. Otherwise: $$ \ell(v_i) = 1 + \max_{v_j \, : \, (v_i, v_j) \in E} \ell(v_j). $$
Notice that, due to our topological order, we necessarily have that $j>i$ in the above formula. This means that we are always able to compute $\ell(v_i)$ if we consider the vertices $v_i$ in reverse topological order.
This is where the above dynamic-programming algorithm connects with the DFS visit: if you consider the order of the vertices induced by the ending times of DFS visit, you obtain a reverse topological order of $G$, which is exactly what you need. This means that you can compute $\ell(v_i)$ just before the DFS visit on node $v_i$ ends.
