#pragma GCC optimize ("O3")
#pragma GCC target ("sse4")

#include <algorithm>
#include <array>
#include <bitset>
#include <cassert>
#include <chrono>
#include <cmath>
#include <complex>
#include <cstdio>
#include <cstdlib>
#include <cstring>
#include <ctime>
#include <deque>
#include <iostream>
#include <iomanip>
#include <map>
#include <numeric>
#include <queue>
#include <random>
#include <set>
#include <stack>
#include <string>
#include <unordered_map>
#include <vector>
#include <ext/pb_ds/tree_policy.hpp>
#include <ext/pb_ds/assoc_container.hpp>

using namespace std;
using namespace __gnu_pbds;
template <class T> using Tree = tree<T, null_type, less<T>, rb_tree_tag,tree_order_statistics_node_update>;

typedef long long ll;
typedef long double ld;
typedef double db;
typedef string str;

typedef pair<int, int> pi;
typedef pair<ll, ll> pl;
typedef pair<db, db> pd;

typedef vector<int> vi;
typedef vector<bool> vb;
typedef vector<ll> vl;
typedef vector<db> vd;
typedef vector<str> vs;
typedef vector<pi> vpi;
typedef vector<pl> vpl;
typedef vector<pd> vpd;

#define mp make_pair
#define f first
#define s second
#define sz(x) (int) (x).size()
#define all(x) begin(x), end(x)
#define rall(x) (x).rbegin(), (x).rend()
#define sor(x) sort(all(x))
#define rsz resize
#define resz resize
#define ins insert
#define ft front()
#define bk back()
#define pf push_front
#define pb push_back
#define eb emplace_back
#define lb lower_bound
#define ub upper_bound

#define f1r(i, a, b) for (int i = (a); i < (b); ++i)
#define f0r(i, a) f1r(i, 0, a)
#define FOR(i, a, b) for (int i = (a); i < (b); ++i)
#define F0R(i, a) FOR(i,0,a)
#define ROF(i, a, b) for (int i = (b) - 1; i >= (a); --i)
#define R0F(i, a) ROF(i, 0, a)
#define each(a, x) for (auto& a : x)

mt19937 rng((uint32_t) chrono::steady_clock::now().time_since_epoch().count());

template <class T> bool ckmin(T& a, const T& b) { return b < a ? a = b, 1 : 0; }
template <class T> bool ckmax(T& a, const T& b) { return a < b ? a = b, 1 : 0; }
template <class T> using V = vector<T>;

#ifdef LOCAL
#define dbg(...) debug(#__VA_ARGS__, __VA_ARGS__);
#else
#define dbg(...) 17;
#endif

template <typename T, typename S> ostream& operator << (ostream &os, const pair<T, S> &p) { return os << "(" << p.first << ", " << p.second << ")"; }
template <typename C, typename T = decay<decltype(*begin(declval<C>()))>, typename enable_if<!is_same<C, string>::value>::type* = nullptr>
ostream& operator << (ostream &os, const C &c) { bool f = true; os << "{"; for (const auto &x : c) { if (!f) os << ", "; f = false; os << x; } return os << "}"; }
template <typename T> void debug(string s, T x) { cerr << s << " = " << x << "\n"; }
template <typename T, typename... Args> void debug(string s, T x, Args... args) { cerr << s.substr(0, s.find(',')) << " = " << x << " | "; debug(s.substr(s.find(',') + 2), args...); }

constexpr int pct(int x) { return __builtin_popcount(x); }
constexpr int bits(int x) { return 31 - __builtin_clz(x); } // floor(log2(x))

namespace input {
    template <class T> void re(complex<T>& x);
    template <class T1, class T2> void re(pair<T1, T2>& p);
    template <class T> void re(vector<T>& a);
    template <class T, int SZ> void re(array<T, SZ>& a);
    template <class T> void re(T& x) { cin >> x; }
    void re(double& x) { string t; re(t); x = stod(t); }
    void re(ld& x) { string t; re(t); x = stold(t); }
    template <class T, class... Ts> void re(T& t, Ts&... ts) {
        re(t); re(ts...); }
    template <class T> void re(complex<T>& x) { T a, b; re(a, b); x = cd(a, b); }
    template <class T1, class T2> void re(pair<T1, T2>& p) { re(p.f, p.s); }
    template <class T> void re(vector<T>& a) { F0R(i, sz(a)) re(a[i]); }
    template <class T, int SZ> void re(array<T, SZ>& a) { F0R(i, SZ) re(a[i]); }
}

using namespace input;

namespace output {
    void pr(int x) { cout << x; }
    void pr(long x) { cout << x; }
    void pr(ll x) { cout << x; }
    void pr(unsigned x) { cout << x; }
    void pr(unsigned long x) { cout << x; }
    void pr(unsigned long long x) { cout << x; }
    void pr(float x) { cout << x; }
    void pr(double x) { cout << x; }
    void pr(ld x) { cout << x; }
    void pr(char x) { cout << x; }
    void pr(const char* x) { cout << x; }
    void pr(const string& x) { cout << x; }
    void pr(bool x) { pr(x ? "true" : "false"); }
    template <class T> void pr(const complex<T>& x) { cout << x; }
    template <class T1, class T2> void pr(const pair<T1, T2>& x);
    template <class T> void pr(const T& x);
    template <class T, class... Ts> void pr(const T& t, const Ts&... ts) {
        pr(t); pr(ts...); }
    template <class T1, class T2> void pr(const pair<T1,T2>& x) {
        pr("{", x.f, ", ", x.s, "}"); }
    template <class T> void pr(const T& x) {
        pr("{"); // const iterator needed for vector<bool>
        bool fst = 1; for (const auto& a: x) pr(!fst ? ", " : "", a), fst = 0;
        pr("}"); }
    void ps() { pr("\n"); } // print w/ spaces
    template <class T, class... Ts> void ps(const T& t, const Ts&... ts) {
        pr(t); if (sizeof...(ts)) pr(" "); ps(ts...); }
    void pc() { pr("]\n"); } // debug w/ commas
    template <class T, class... Ts> void pc(const T& t, const Ts&... ts) {
        pr(t); if (sizeof...(ts)) pr(", "); pc(ts...); }
}

using namespace output;

namespace io {
    void setIn(string s) { freopen(s.c_str(), "r", stdin); }
    void setOut(string s) { freopen(s.c_str(), "w", stdout); }
    void setIO(string s = "") {
        cin.sync_with_stdio(0); cin.tie(0);
        if (sz(s)) { setIn(s + ".in"), setOut(s + ".out"); }
    }
}

using namespace io;

const int MOD = 1e9 + 7; // 998244353;
const ld PI = acos((ld) -1);

typedef std::decay <decltype(MOD)>::type mod_t;
struct mi {
    mod_t val;
    explicit operator mod_t() const { return val; }
    mi() { val = 0; }
    mi(const long long& v) {
        val = (-MOD <= v && v <= MOD) ? v : v % MOD;
        if (val < 0) val += MOD; }
    friend std::istream& operator >> (std::istream& in, mi& a) { 
        long long x; std::cin >> x; a = mi(x); return in; }
    friend std::ostream& operator << (std::ostream& os, const mi& a) { return os << a.val; }
    friend void pr(const mi& a) { pr(a.val); }
    friend void re(mi& a) { long long x; cin >> x; a = mi(x); }
    friend bool operator == (const mi& a, const mi& b) { return a.val == b.val; }
    friend bool operator != (const mi& a, const mi& b) { return !(a == b); }    
    friend bool operator < (const mi& a, const mi& b) { return a.val < b.val; }
    friend bool operator > (const mi& a, const mi& b) { return a.val > b.val; }
    friend bool operator <= (const mi& a, const mi& b) { return a.val <= b.val; }
    friend bool operator >= (const mi& a, const mi& b) { return a.val >= b.val; }
    mi operator - () const { return mi(-val); }
    mi& operator += (const mi& m) {
        if ((val += m.val) >= MOD) val -= MOD;
        return *this; }
    mi& operator -= (const mi& m) {
        if ((val -= m.val) < 0) val += MOD;
        return *this; }
    mi& operator *= (const mi& m) { val = (long long) val * m.val % MOD;
        return *this; }
    friend mi pow(mi a, long long p) {
        mi ans = 1; assert(p >= 0);
        for (; p; p /= 2, a *= a) if (p & 1) ans *= a;
        return ans; }
    friend mi inv(const mi& a) { assert(a != 0); return pow(a, MOD - 2); }
    mi& operator /= (const mi& m) { return (*this) *= inv(m); }
    friend mi operator + (mi a, const mi& b) { return a += b; }
    friend mi operator - (mi a, const mi& b) { return a -= b; }
    friend mi operator * (mi a, const mi& b) { return a *= b; }
    friend mi operator / (mi a, const mi& b) { return a /= b; }
};
typedef pair<mi, mi> pmi;
typedef vector<mi> vmi;
typedef vector<pmi> vpmi;

struct SCC {
    int n, time, num_comps;
    std::vector<std::vector<int>> adj;
    std::vector<int> disc, id, stk;
    std::vector<std::vector<int>> comps;

    void init(int n_) {
        n = n_;
        time = 0;
        num_comps = 0;
        adj.assign(n, std::vector<int>());
        id.assign(n, -1);
        disc.assign(n, 0);
        comps.clear();
    }

    void ae(int u, int v) {
        adj[u].push_back(v);
    }

    int dfs(int src) {
        int low = disc[src] = ++time;
        stk.push_back(src);
        for (int nxt : adj[src]) 
            if (id[nxt] == -1)
                low = std::min(low, disc[nxt] ? : dfs(nxt));
        if (low == disc[src]) {
            for (int nxt = -1; nxt != src;)
                id[nxt = stk.back()] = num_comps, stk.pop_back();
            num_comps++;
        }
        return low;
    }
    
    void build() {
        // builds in topological order
        for (int i = 0; i < n; i++) 
            if (!disc[i])
                dfs(i);
        for (auto& x : id) 
            x = num_comps - 1 - x;
        comps.resize(num_comps);
        for (int i = 0; i < n; i++)
            comps[id[i]].push_back(i);
    }
};

int main() {
    setIO("");
    int n, m; re(n, m);
    SCC S;
    S.init(n);

    vpi ed;

    f0r(i, m) {
        int u, v; re(u, v);
        u--, v--;
        S.ae(u, v);
        ed.eb(u, v);
    }

    S.build();

    int sz = sz(S.comps);
    if (sz == 1)  {
        ps(0);
        return 0;
    }

    vi lead(sz); // leader

    f0r(i, sz) {
        lead[i] = S.comps[i][0];
    }

    vector<vi> in(sz);
    vector<vi> out(sz);

    each(e, ed) {
        int u = S.id[e.f];
        int v = S.id[e.s];
        if (u == v) continue;
        out[u].pb(v);
        in[v].pb(u);
    }

    vi source;
    vi sink;
    vi alone;

    vi isrc(sz);
    vi isnk(sz);

    f0r(i, sz) {
        if (sz(in[i]) == 0 && sz(out[i]) == 0) {
            alone.pb(i);
        } else if (sz(in[i]) == 0) {
            source.pb(i);
            isrc[i] = 1;
        } else if (sz(out[i]) == 0) {
            sink.pb(i);
            isnk[i] = 1;
        }
    }

    vi mark(sz);

    int ww = -1;

    function<void(int)> search = [&](int x) {
        if (!mark[x]) {
            if (isnk[x] && ww == -1) {
                ww = x;
            }
            mark[x] = 1;
            each(y, out[x]) {
                search(y);
                if (ww != -1) return;
            }
        }
    };

    int ii = 0;

    vi o1(sz(source));
    vi o2(sz(sink));

    int p;

    each(v, source) {
        if (mark[v]) continue;
        ww = -1;
        search(v);
        if (ww != -1) {
            o1[ii] = v;
            o2[ii] = ww;
            ii++;
        }
    }

    p = ii; // [0, p-1] matched

    vi done(n);

    f0r(i, p) {
        done[o1[i]] = 1;
    }
    int cur = p;
    each(v, source) {
        if (!done[v]) {
            o1[cur] = v;
            cur++;
        }
    }

    done = vi(n, 0);
    f0r(i, p) {
        done[o2[i]] = 1;
    }

    cur = p;
    each(v, sink) {
        if (!done[v]) {
            o2[cur] = v;
            cur++;
        }
    }

    vpi res;
    auto ae = [&](int u, int v) {
        res.eb(lead[u], lead[v]);
    };

    f0r(i, p-1) {
        ae(o2[i], o1[i+1]);
    }

    if (sz(alone) == 0) { 
        if (p) {
            ae(o2[p-1], o1[0]);
        }
    } else {
        f1r(i, 0, sz(alone)-1) {
            ae(alone[i], alone[i+1]);
        }
        if (p) {
            ae(o2[p-1], alone[0]);
            ae(alone.back(), o1[0]);
        } else {
            ae(alone.back(), alone[0]);
        }
    }

    if (sz(o1) == p && sz(o2) == p) {
        // do nothing
    } else if (sz(o1) == p) {
        f1r(i, p, sz(o2)) {
            ae(o2[i], o2[0]);
        }
    } else if (sz(o2) == p) {
        f1r(i, p, sz(o1)) {
            ae(o1[0], o1[i]);
        }
    } else {
        if (sz(o1) < sz(o2)) {
            f1r(i, p, sz(o1)) {
                ae(o2[i], o1[i]);
            }
            f1r(i, sz(o1), sz(o2)) {
                ae(o2[i], o2[0]);
            }
        } else {
            f1r(i, p, sz(o2)) {
                ae(o2[i], o1[i]);
            }
            f1r(i, sz(o2), sz(o1)) {
                ae(o1[0], o1[i]);
            }
        }
    }
    // assert(sz(res) == sz(alone) + max(sz(source), sz(sink)));
    // SCC S2;
    // S2.init(n);
    // each(e, ed) {
    //     S2.ae(e.f, e.s);
    // }
    // each(e, res) {
    //     S2.ae(e.f, e.s);
    // }
    // S2.build();
    // assert(sz(S2.comps) == 1);
    ps(sz(res));
    each(e, res) {
        ps(e.f+1, e.s+1);
    }


    return 0;
}