How to work with struct inside struct in Cuda

I have the following (simplified) code in my .cu file

typedef struct
{
    int leg_id;
    int contract_id;
} CudaLeg;

typedef struct
{
    CudaLeg* legs;
    int no_legs;
} CudaPortfolio;

extern "C"
__global__ void kernel(CudaPortfolio* o_portfolios_cuda, const int no_portfolios)
{
//    fill o_portfolios_cuda with data
}

void cudaFunction(CudaPortfolio* o_portfolios, unsigned long long no_portfolios)
{
    CudaPortfolio* o_portfolios_cuda;
    cudaMalloc((void **)& o_portfolios_cuda, sizeof(CudaPortfolio) * no_portfolios);

    kernel<<<32, 32>>>(o_portfolios_cuda, no_portfolios);

    cudaMemcpy(o_portfolios, o_portfolios_cuda, sizeof(CudaPortfolio) * no_portfolios, cudaMemcpyDeviceToHost);

    //printf below works
    printf("CPU no legs strike output portfolio: %dn", o_portfolios[0].no_legs);
    //printf below crashes the program
    printf("CPU Leg 1 multiplier output portfolio: %dn", o_portfolios[0].legs[0].multiplier);

    cudaFree(o_portfolios_cuda);
}

The GPU is a GTX580, sm2.0. The GPU can work fine with o_portfolios_cuda and fill it with data and do calculations with it. The first printf of o_portfolios[0].no_legs gives back the correct function. But when I try to access some of the portfolios legs (o_portfolios[0].legs[0].multiplier) the program crashes. Any ideas how I can fix this? Thank you.

@Robert Crovella I already tried something like that, but it didn't work. I tried it again and added

    CudaLeg* o_portfolios_legs_cuda;
    cudaMalloc((void **)& o_portfolios_legs_cuda, sizeof(CudaLeg));
    cudaMemcpy(o_portfolios_legs_cuda, o_portfolios->legs, sizeof(CudaLeg), cudaMemcpyHostToDevice);
    cudaMemcpy(&(o_portfolios_cuda->legs), &o_portfolios_legs_cuda, sizeof(CudaLeg *), cudaMemcpyHostToDevice);

But now the program crashes on the 3rd line I just added (cudaMemcpy(o_portfolios_legs_cuda, ...)

@MarkoR The CudaLeg objects don't have a fixed number.