#define MOD_TYPE 2

#pragma region Macros

#include <bits/stdc++.h>
using namespace std;

#if 1
#pragma GCC target("avx2")
#pragma GCC optimize("O3")
#pragma GCC optimize("unroll-loops")
#endif

#if 0
#include <ext/pb_ds/assoc_container.hpp>
#include <ext/pb_ds/tree_policy.hpp>
#include <ext/pb_ds/tag_and_trait.hpp>
using namespace __gnu_pbds;
template <typename T>
using extset = tree<T, null_type, less<T>, rb_tree_tag, tree_order_statistics_node_update>;
#endif

#if 0
#include <boost/multiprecision/cpp_int.hpp>
#include <boost/multiprecision/cpp_dec_float.hpp>
using Int = boost::multiprecision::cpp_int;
using lld = boost::multiprecision::cpp_dec_float_100;
#endif

using ll = long long int;
using ld = long double;
using pii = pair<int, int>;
using pll = pair<ll, ll>;
using pld = pair<ld, ld>;
template <typename T>
using smaller_queue = priority_queue<T, vector<T>, greater<T>>;

constexpr ll MOD = (MOD_TYPE == 1 ? (ll)(1e9 + 7) : 998244353);
constexpr int INF = (int)1e9 + 10;
constexpr ll LINF = (ll)4e18;
constexpr ld PI = acos(-1.0);
constexpr ld EPS = 1e-7;
constexpr int Dx[] = {0, 0, -1, 1, -1, 1, -1, 1, 0};
constexpr int Dy[] = {1, -1, 0, 0, -1, -1, 1, 1, 0};

#define REP(i, m, n) for (ll i = m; i < (ll)(n); ++i)
#define rep(i, n) REP(i, 0, n)
#define REPI(i, m, n) for (int i = m; i < (int)(n); ++i)
#define repi(i, n) REPI(i, 0, n)
#define MP make_pair
#define MT make_tuple
#define YES(n) cout << ((n) ? "YES" : "NO") << "\n"
#define Yes(n) cout << ((n) ? "Yes" : "No") << "\n"
#define possible(n) cout << ((n) ? "possible" : "impossible") << "\n"
#define Possible(n) cout << ((n) ? "Possible" : "Impossible") << "\n"
#define all(v) v.begin(), v.end()
#define NP(v) next_permutation(all(v))
#define dbg(x) cerr << #x << ":" << x << "\n";

struct io_init
{
  io_init()
  {
    cin.tie(0);
    ios::sync_with_stdio(false);
    cout << setprecision(30) << setiosflags(ios::fixed);
  };
} io_init;
template <typename T>
inline bool chmin(T &a, T b)
{
  if (a > b)
  {
    a = b;
    return true;
  }
  return false;
}
template <typename T>
inline bool chmax(T &a, T b)
{
  if (a < b)
  {
    a = b;
    return true;
  }
  return false;
}
inline ll CEIL(ll a, ll b)
{
  return (a + b - 1) / b;
}
template <typename A, size_t N, typename T>
inline void Fill(A (&array)[N], const T &val)
{
  fill((T *)array, (T *)(array + N), val);
}
template <typename T, typename U>
constexpr istream &operator>>(istream &is, pair<T, U> &p) noexcept
{
  is >> p.first >> p.second;
  return is;
}
template <typename T, typename U>
constexpr ostream &operator<<(ostream &os, pair<T, U> &p) noexcept
{
  os << p.first << " " << p.second;
  return os;
}
#pragma endregion

// --------------------------------------

struct UnionFind
{
public:
  UnionFind() : _n(0) {}
  UnionFind(int n) : _n(n), parent_or_size(n, -1) {}

  int unite(int a, int b)
  {
    assert(0 <= a && a < _n);
    assert(0 <= b && b < _n);
    int x = root(a), y = root(b);
    if (x == y)
      return x;
    if (-parent_or_size[x] < -parent_or_size[y])
      swap(x, y);
    parent_or_size[x] += parent_or_size[y];
    parent_or_size[y] = x;
    return x;
  }

  bool same(int a, int b)
  {
    assert(0 <= a && a < _n);
    assert(0 <= b && b < _n);
    return root(a) == root(b);
  }

  int root(int a)
  {
    assert(0 <= a && a < _n);
    if (parent_or_size[a] < 0)
      return a;
    return parent_or_size[a] = root(parent_or_size[a]);
  }

  int size(int a)
  {
    assert(0 <= a && a < _n);
    return -parent_or_size[root(a)];
  }

  vector<vector<int>> groups()
  {
    vector<int> leader_buf(_n), group_size(_n);
    for (int i = 0; i < _n; i++)
    {
      leader_buf[i] = root(i);
      group_size[leader_buf[i]]++;
    }
    vector<vector<int>> result(_n);
    for (int i = 0; i < _n; i++)
    {
      result[i].reserve(group_size[i]);
    }
    for (int i = 0; i < _n; i++)
    {
      result[leader_buf[i]].push_back(i);
    }
    result.erase(
        remove_if(result.begin(), result.end(),
                  [&](const vector<int> &v) { return v.empty(); }),
        result.end());
    return result;
  }

private:
  int _n;
  vector<int> parent_or_size;
};

vector<int> E[200010];
int in[200010] = {};
vector<int> v[200010];
bool used[200010] = {};

void solve()
{
  int n, m;
  cin >> n >> m;

  UnionFind uf(n);
  rep(i, m)
  {
    int a, b;
    cin >> a >> b;
    a--, b--;
    E[a].push_back(b);
    in[b]++;
    uf.unite(a, b);
  }

  rep(i, n)
  {
    v[uf.root(i)].push_back(i);
  }

  vector<pii> ans;
  rep(r, n)
  {
    if (v[r].empty())
      continue;
    queue<int> que;
    for (auto t : v[r])
    {
      if (in[t] == 0)
      {
        que.push(t);
        used[t] = true;
      }
    }
    if (que.empty())
    {
      rep(i, v[r].size())
      {
        ans.push_back({v[r][i], v[r][(i + 1) % v[r].size()]});
      }
      continue;
    }
    vector<int> topo;
    while (!que.empty())
    {
      int q = que.front();
      que.pop();
      topo.push_back(q);
      for (auto to : E[q])
      {
        if (used[to]) // 閉路が存在
        {
          while (!que.empty())
            que.pop();
          break;
        }
        in[to]--;
        if (in[to] == 0)
        {
          used[to] = true;
          que.push(to);
        }
      }
    }
    if (topo.size() + 1 < v[r].size())
    {
      rep(i, v[r].size())
      {
        ans.push_back({v[r][i], v[r][(i + 1) % v[r].size()]});
      }
    }
    else
    {
      assert(topo.size() >= 1);
      rep(i, topo.size() - 1)
      {
        ans.push_back({topo[i], topo[i + 1]});
      }
    }
  }
  cout << ans.size() << "\n";
  for (auto [a, b] : ans)
    cout << a + 1 << " " << b + 1 << "\n";
}

int main()
{
  solve();
}