What is the correct way to pass 2D ndarrays to C extension and get the new one?

Question

The task is to create a C extention for Python which will take a 2D ndarray from python program and then, after some math operation (written with C) return a new ndarray. Here I wrote an example of code that should work:

#define PY_SSIZE_T_CLEAN
#include <Python.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>

int processing(inarr) {
    printf("Start...");
    /*Here is a place for some C function*/
    return outarr;
}


static PyObject *proc(PyObject *self, PyObject *args) {
    PyObject *argin = NULL, *argout = NULL;
    PyObject *inarr = NULL, *outarr = NULL;
    inarr = PyArray_FROM_OTF(argin, NPY_DOUBLE, NPY_IN_ARRAY);
    outarr = processing(inarr);


    if(!PyArg_ParseTuple(args, "O!:ss", &argin, &argout)) {
        return NULL;
    }

    return PyLong_FromLong(outarr);
}

static PyMethodDef ProcMethods[] = {
    {"proc", proc, METH_VARARGS, "Method description"},
    {NULL,NULL,0,NULL} 
};

static struct PyModuleDef procmodule = 
{
    PyModuleDef_HEAD_INIT,
    "proc",
    "Description of my module",
    -1,
    ProcMethods
};
PyMODINIT_FUNC PyInit_procmodule(void) {
    return PyModule_Create(procmodule)
};

But there are some questions.

1. Should I use PyArg_ParseTuple before if statement?

2. Is the init function correct (are there some special way to initiate it for 2D tuples)?

3. Is this enough to treat an ndarray within C function as array[][] without writing a matrix from a cycle by elements?

4. Will it work from python (after creating a package of course) like

   import proc

   proc(array)

UPD: Added initialisation function