You have gotten two nice analytical approaches.
I will give a more mechanical and numerical approach that is more general when analytical approaches fail.
We will use the substitution $v = xy$
which by Jacquelin gives $$\frac{d v}{dx} =
\frac{v^2+1} x$$
We will now assume a power series expansion around $0$.
Using some well known facts
- The square of a power series expansion is a convolution of it's coefficients.
- Differentiation and multiplication and division with x are linear operators (- but only for as long as we can ensure we never need to represent reciprocals of x and their powers). Let us call the corresponding operators D and X.
using the notation $v_k$ is the k:th iteration to find a solution we can define a fixed point iteration
$$v_k = D^{-1}X^{-1}(v_{k-1} ^ 2 + 1)$$
Running a loop in Gnu Octave for a power series expansion truncated to a 16 order polynomial gives me a solution with first four terms very close to [0,1,0.5,1/3]. Which I find reasonable as power series expansion of $-log(1-t)$ starts like that and $t\to\tan(t)$ is close to $t\to t$ when $t$ is close to $0$.
The following terms start differing from the $t^{k}/k$ pattern
$\left[\begin{array}{llll}0&1&0.5&0.333333333333333\\0.229166666666667&0.158333333333333&0.109490740740741&0.0757275132275133\\0.0523773354828043&0.0362272728664511&0.0250569590651333&0.0173309023741605\\0.0119870966097813&0.00829099853635585&0.00573455433506161&0.0039663634345758\end{array}\right]
$
Perhaps someone can help me see which C it corresponds to?