// >>> TEMPLATES
#include <bits/stdc++.h>
using namespace std;
using ll = long long;
using ld = long double;
using i32 = int32_t;
using i64 = int64_t;
using u32 = uint32_t;
using u64 = uint64_t;
#define int ll
#define rep(i, n) for (int i = 0; i < (int)(n); i++)
#define rep1(i, n) for (int i = 1; i <= (int)(n); i++)
#define repR(i, n) for (int i = (int)(n)-1; i >= 0; i--)
#define rep1R(i, n) for (int i = (int)(n); i >= 1; i--)
#define loop(i, a, B) for (int i = a; i B; i++)
#define loopR(i, a, B) for (int i = a; i B; i--)
#define all(x) begin(x), end(x)
#define allR(x) rbegin(x), rend(x)
#define rng(x, l, r) begin(x) + (l), begin(x) + (r)
#define pb push_back
#define eb emplace_back
#define fst first
#define snd second
template <class A, class B> constexpr auto mp(A &&a, B &&b) { return make_pair(forward<A>(a), forward<B>(b)); }
template <class... T> constexpr auto mt(T&&... x) { return make_tuple(forward<T>(x)...); }
template <class Int> auto constexpr inf_ = numeric_limits<Int>::max()/2-1;
auto constexpr INF32 = inf_<int32_t>;
auto constexpr INF64 = inf_<int64_t>;
auto constexpr INF   = inf_<int>;
#ifdef LOCAL
#include "debug.hpp"
#else
#define dump(...) (void)(0)
#define say(x) (void)(0)
#define debug if (0)
#endif
template <class T, class Comp> struct pque : priority_queue<T, vector<T>, Comp> { vector<T> &data() { return this->c; } void clear() { this->c.clear(); } };
template <class T> using pque_max = pque<T, less<T>>;
template <class T> using pque_min = pque<T, greater<T>>;
template <class T, class = typename T::iterator, enable_if_t<!is_same<T, string>::value, int> = 0>
ostream& operator<<(ostream& os, T const& a) { bool f = true; for (auto const& x : a) os << (f ? "" : " ") << x, f = false; return os; }
template <class T, size_t N, enable_if_t<!is_same<T, char>::value, int> = 0>
ostream& operator<<(ostream& os, const T (&a)[N]) { bool f = true; for (auto const& x : a) os << (f ? "" : " ") << x, f = false; return os; }
template <class T, class = decltype(begin(declval<T&>())), class = typename enable_if<!is_same<T, string>::value>::type>
istream& operator>>(istream& is, T &a) { for (auto& x : a) is >> x; return is; }
template <class T, class S> ostream& operator<<(ostream& os, pair<T, S> const& p) { return os << p.first << " " << p.second; }
template <class T, class S> istream& operator>>(istream& is, pair<T, S>& p) { return is >> p.first >> p.second; }
struct IOSetup { IOSetup() { cin.tie(nullptr); ios::sync_with_stdio(false); cout << fixed << setprecision(15); } } iosetup;
template <class F> struct FixPoint : private F {
    constexpr FixPoint(F&& f) : F(forward<F>(f)) {}
    template <class... T> constexpr auto operator()(T&&... x) const { return F::operator()(*this, forward<T>(x)...); }
};
struct MakeFixPoint { template <class F> constexpr auto operator|(F&& f) const { return FixPoint<F>(forward<F>(f)); } };
#define MFP MakeFixPoint()|
#define def(name, ...) auto name = MFP [&](auto &&name, __VA_ARGS__)
template <class T, size_t d> struct vec_impl {
    using type = vector<typename vec_impl<T, d-1>::type>;
    template <class... U> static type make_v(size_t n, U&&... x) { return type(n, vec_impl<T, d-1>::make_v(forward<U>(x)...)); }
};
template <class T> struct vec_impl<T, 0> { using type = T; static type make_v(T const& x = {}) { return x; } };
template <class T, size_t d = 1> using vec = typename vec_impl<T, d>::type;
template <class T, size_t d = 1, class... Args> auto make_v(Args&&... args) { return vec_impl<T, d>::make_v(forward<Args>(args)...); }
template <class T> void quit(T const& x) { cout << x << endl; exit(0); }
template <class T, class U> constexpr bool chmin(T& x, U const& y) { if (x > (T)y) { x = (T)y; return true; } return false; }
template <class T, class U> constexpr bool chmax(T& x, U const& y) { if (x < (T)y) { x = (T)y; return true; } return false; }
template <class It> constexpr auto sumof(It b, It e) { return accumulate(b, e, typename iterator_traits<It>::value_type{}); }
template <class T> int sz(T const& x) { return x.size(); }
template <class C, class T> int lbd(C const& v, T const& x) { return lower_bound(begin(v), end(v), x)-begin(v); }
template <class C, class T> int ubd(C const& v, T const& x) { return upper_bound(begin(v), end(v), x)-begin(v); }
constexpr int64_t mod(int64_t x, int64_t m) { assert(m > 0); return (x %= m) < 0 ? x+m : x; }
constexpr int64_t div_floor(int64_t x, int64_t y) { assert(y != 0); return x/y - ((x^y) < 0 and x%y); }
constexpr int64_t div_ceil(int64_t x, int64_t y) { assert(y != 0); return x/y + ((x^y) > 0 and x%y); }
constexpr int dx[] = { 1, 0, -1, 0, 1, -1, -1, 1 };
constexpr int dy[] = { 0, 1, 0, -1, 1, 1, -1, -1 };
constexpr int popcnt(ll x) { return __builtin_popcountll(x); }
mt19937_64 seed_{random_device{}()};
template <class Int> Int rand(Int a, Int b) { return uniform_int_distribution<Int>(a, b)(seed_); }
i64 irand(i64 a, i64 b) { return rand<i64>(a, b); } // [a, b]
u64 urand(u64 a, u64 b) { return rand<u64>(a, b); } //
template <class It> void shuffle(It l, It r) { shuffle(l, r, seed_); }
template <class T> vector<T> &operator--(vector<T> &v) { for (T &x : v) --x; return v; }
template <class T> vector<T> &operator++(vector<T> &v) { for (T &x : v) ++x; return v; }
// <<<
// >>> union find

struct UnionFind {
    int n, sz; // id : 0...n-1
    vector<int32_t> par;

    UnionFind(int n = 0) : n(n), sz(n), par(n, -1) { }
    void clear() {
        rep (i, n) par[i] = -1;
        sz = n;
    }
    int root(int x) {
        assert(0 <= x); assert(x < n);
        return par[x] < 0 ? x : par[x] = root(par[x]);
    }
    bool unite(int x, int y) {
        x = root(x), y = root(y);
        if (x == y) return false;
        sz--;
        if (par[x] < par[y]) swap(x, y);
        par[y] += par[x];
        par[x] = y;
        return true;
    }
    bool same(int x, int y) { return root(x) == root(y); }
    int size(int x) { return -par[root(x)]; }
    int size() const { return sz; }
    vector<vector<int32_t>> groups() {
        vector<vector<int32_t>> g(n);
        rep (i, n) if (par[i] < 0) g[i].reserve(-par[i]);
        rep (i, n) g[root(i)].push_back(i);
        vector<vector<int32_t>> grp; grp.reserve(size());
        rep (i, n) if (g[i].size()) grp.emplace_back(move(g[i]));
        return grp;
    }
};

// <<<
// >>> scc
#ifndef EDGE_INFO
#define EDGE_INFO
constexpr int dest(int v) { return v; }
template <class E, class = decltype(declval<E>().to)>
constexpr int dest(E const& e) { return e.to; }
constexpr int cost(int) { return 1; }
template <class E, class = decltype(declval<E>().cost)>
constexpr auto cost(E const& e) { return e.cost; }
template <class E> using cost_t = decltype(cost(declval<E>()));
#endif
template <class E> struct SCC {
    vector<vector<E>> const& g;
    vector<int32_t> ord, low, st;
    int32_t time = 0, pos = 0, sz = 0;
    UnionFind uf;
    SCC(vector<vector<E>> const& g) : g(g), ord(g.size()), low(g.size()), st(g.size()), uf(g.size()) {
        for (auto v : wcc()) if (ord[v] == 0) dfs(v);
        for (auto &x : ord) x += sz;
    }
    vector<i32> wcc() {
        vector<i32> nxt(g.size()), vs;
        vs.reserve(g.size());
        rep (v, g.size()) nxt[v] = v;
        rep (v, g.size()) {
            for (auto const& e : g[v]) {
                auto const u = dest(e);
                if (uf.unite(v, u)) swap(nxt[v], nxt[u]);
            }
        }
        for (i32 v = (i32)g.size()-1; v >= 0; --v) {
            if (uf.root(v) != v) continue;
            auto u = v;
            do vs.push_back(u), u = nxt[u]; while (u != v);
        }
        return vs;
    }
    void dfs(int32_t v) {
        low[v] = ord[v] = ++time; st[pos++] = v;
        for (auto const& e : g[v]) {
            if (ord[dest(e)] == 0) {
                dfs(dest(e));
                chmin(low[v], low[dest(e)]);
            } else if (ord[dest(e)] > 0) {
                chmin(low[v], ord[dest(e)]);
            }
        }
        if (low[v] == ord[v]) {
            sz++; do ord[st[--pos]] = -sz; while (st[pos] != v);
        }
    }

    // t-sorted id
    int id(int v) const { return ord[v]; }
    int size() const { return sz; }
    vector<vector<int>> groups() const {
        vector<vector<int>> grp(sz);
        rep (v, g.size()) grp[ord[v]].push_back(v);
        return grp;
    }

    // no multiple edges, no self-loops
    // no edge information
    vector<vector<int>> graph() const {
        vector<vector<int>> ret(size());
        vector<int32_t> last(g.size(), -1);
        rep (v, g.size()) {
            for (auto const& e : g[v]) {
                const int32_t x = id(v), y = id(dest(e));
                if (x == y or last[y] == x) continue;
                last[y] = x;
                ret[x].push_back(y);
            }
        }
        return ret;
    }
};
template <class E> auto get_scc(vector<vector<E>> const& g) { return SCC<E>(g); }

// <<<
// >>> bfs
#ifndef EDGE_INFO
#define EDGE_INFO
constexpr int dest(int v) { return v; }
template <class E, class = decltype(declval<E>().to)>
constexpr int dest(E const& e) { return e.to; }
constexpr int cost(int) { return 1; }
template <class E, class = decltype(declval<E>().cost)>
constexpr auto cost(E const& e) { return e.cost; }
template <class E> using cost_t = decltype(cost(declval<E>()));
#endif
template <class E>
vector<int> bfs(vector<vector<E>> const& g, int s, int inf) {
    vector<int> d(g.size(), inf);
    queue<int> q;
    d[s] = 0;
    q.push(s);
    while (q.size()) {
        int x = q.front(); q.pop();
        for (auto const& e : g[x]) {
            const int y = dest(e);
            if (chmin(d[y], d[x]+1)) q.push(y);
        }
    }
    return d;
}
// <<<

int32_t main() {
    int n, m; cin >> n >> m;
    vector<vector<int>> g(n);
    rep (i, m) {
        int x, y; cin >> x >> y; --x, --y;
        g[x].eb(y);
    }

    auto scc = get_scc(g);
    auto grp = scc.groups();
    dump(grp);
    vector<pair<int, int>> es;
    for (auto const& v : grp) {
        if (v.size() == 1) continue;
        rep (i, v.size()) {
            int j = (i+1) % v.size();
            es.eb(v[i], v[j]);
        }
    }
    rep (i, grp.size()-1) {
        int x = grp[i][0];
        int y = grp[i+1][0];
        if (scc.uf.same(x, y)) {
            es.eb(x, y);
        }
    }

    cout << es.size() << '\n';
    for (auto [x, y] : es) cout << x+1 << " " << y+1 << '\n';

    debug {
        vector<vector<int>> h(n);
        for (auto [x, y] : es) h[x].eb(y);

        rep (x, n) {
            for (int y : g[x]) {
                auto d = bfs(h, x, INF);
                assert(d[y] < INF);
            }
        }
    }

}