If somebody know how to make my variant working faster(if it's possible), please write to nonename@narod.ru.
 I think in the worst case it needs O(n*l), but I also think  it's quite clear to understand. May be using memory will be faster, but haven't tried. Other variant is that algorithm is wrong.



  int i,j,n,l;
  int work[40005];

  cin >> n;

  work[0]=-1;
  for (i=0;i<n;i++){
    int tmp;
    cin >> tmp;
    cin >> work[tmp+1];
  }
  cin >> l;
  int a,b,fl=0;
  for (i=0;i<l;i++){
    fl=0;
    cin >> a >> b;
    int wka=a;
    int wkb=b;
    while (wka+wkb+2){
        wka=work[wka+1];
        wkb=work[wkb+1];
      if (wka==b){
        fl=2;
        break;
      }
      if (wkb==a){
        fl=1;
        break;
      }
    }
   cout << fl << endl;
  }