#include <bits/stdc++.h>
#include <atcoder/all>
#include <boost/polygon/point_data.hpp>
#include <boost/polygon/voronoi.hpp>
#define rep(i,a,b) for(int i=(a);i<(b);i++)
#define rrep(i,a,b) for(int i=(b)-1;i>=(a);i--)
#define all(x) begin(x),end(x)
#define sz(x) (int)(x).size()
using namespace std;
using namespace atcoder;
using ll=long long;
using ld=long double;
using vll=vector<ll>;
using vvll=vector<vll>;
using vi=vector<int>;
using vvi=vector<vi>;
using pll=pair<ll,ll>;
using pi=pair<int,int>;

// using mint=modint998244353;

// using BoostPoint=boost::polygon::point_data<int>;
// using Voronoi=boost::polygon::voronoi_diagram<double>;

#line 2 "modint/montgomery-modint.hpp"

template <uint32_t mod>
struct LazyMontgomeryModInt {
  using mint = LazyMontgomeryModInt;
  using i32 = int32_t;
  using u32 = uint32_t;
  using u64 = uint64_t;

  static constexpr u32 get_r() {
    u32 ret = mod;
    for (i32 i = 0; i < 4; ++i) ret *= 2 - mod * ret;
    return ret;
  }

  static constexpr u32 r = get_r();
  static constexpr u32 n2 = -u64(mod) % mod;
  static_assert(mod < (1 << 30), "invalid, mod >= 2 ^ 30");
  static_assert((mod & 1) == 1, "invalid, mod % 2 == 0");
  static_assert(r * mod == 1, "this code has bugs.");

  u32 a;

  constexpr LazyMontgomeryModInt() : a(0) {}
  constexpr LazyMontgomeryModInt(const int64_t &b)
      : a(reduce(u64(b % mod + mod) * n2)){};

  static constexpr u32 reduce(const u64 &b) {
    return (b + u64(u32(b) * u32(-r)) * mod) >> 32;
  }

  constexpr mint &operator+=(const mint &b) {
    if (i32(a += b.a - 2 * mod) < 0) a += 2 * mod;
    return *this;
  }

  constexpr mint &operator-=(const mint &b) {
    if (i32(a -= b.a) < 0) a += 2 * mod;
    return *this;
  }

  constexpr mint &operator*=(const mint &b) {
    a = reduce(u64(a) * b.a);
    return *this;
  }

  constexpr mint &operator/=(const mint &b) {
    *this *= b.inverse();
    return *this;
  }

  constexpr mint operator+(const mint &b) const { return mint(*this) += b; }
  constexpr mint operator-(const mint &b) const { return mint(*this) -= b; }
  constexpr mint operator*(const mint &b) const { return mint(*this) *= b; }
  constexpr mint operator/(const mint &b) const { return mint(*this) /= b; }
  constexpr bool operator==(const mint &b) const {
    return (a >= mod ? a - mod : a) == (b.a >= mod ? b.a - mod : b.a);
  }
  constexpr bool operator!=(const mint &b) const {
    return (a >= mod ? a - mod : a) != (b.a >= mod ? b.a - mod : b.a);
  }
  constexpr mint operator-() const { return mint() - mint(*this); }
  constexpr mint operator+() const { return mint(*this); }

  constexpr mint pow(u64 n) const {
    mint ret(1), mul(*this);
    while (n > 0) {
      if (n & 1) ret *= mul;
      mul *= mul;
      n >>= 1;
    }
    return ret;
  }

  constexpr mint inverse() const {
    int x = get(), y = mod, u = 1, v = 0, t = 0, tmp = 0;
    while (y > 0) {
      t = x / y;
      x -= t * y, u -= t * v;
      tmp = x, x = y, y = tmp;
      tmp = u, u = v, v = tmp;
    }
    return mint{u};
  }

  friend ostream &operator<<(ostream &os, const mint &b) {
    return os << b.get();
  }

  friend istream &operator>>(istream &is, mint &b) {
    int64_t t;
    is >> t;
    b = LazyMontgomeryModInt<mod>(t);
    return (is);
  }

  constexpr u32 get() const {
    u32 ret = reduce(a);
    return ret >= mod ? ret - mod : ret;
  }

  static constexpr u32 get_mod() { return mod; }
};

#line 2 "dp/inversion-counting.hpp"

#line 2 "data-structure/binary-indexed-tree.hpp"

template <typename T>
struct BinaryIndexedTree {
  int N;
  vector<T> data;

  BinaryIndexedTree() = default;

  BinaryIndexedTree(int size) { init(size); }

  void init(int size) {
    N = size + 2;
    data.assign(N + 1, {});
  }

  // get sum of [0,k]
  T sum(int k) const {
    if (k < 0) return T{};  // return 0 if k < 0
    T ret{};
    for (++k; k > 0; k -= k & -k) ret += data[k];
    return ret;
  }

  // getsum of [l,r]
  inline T sum(int l, int r) const { return sum(r) - sum(l - 1); }

  // get value of k
  inline T operator[](int k) const { return sum(k) - sum(k - 1); }

  // data[k] += x
  void add(int k, T x) {
    for (++k; k < N; k += k & -k) data[k] += x;
  }

  // range add x to [l,r]
  void imos(int l, int r, T x) {
    add(l, x);
    add(r + 1, -x);
  }

  // minimize i s.t. sum(i) >= w
  int lower_bound(T w) {
    if (w <= 0) return 0;
    int x = 0;
    for (int k = 1 << __lg(N); k; k >>= 1) {
      if (x + k <= N - 1 && data[x + k] < w) {
        w -= data[x + k];
        x += k;
      }
    }
    return x;
  }

  // minimize i s.t. sum(i) > w
  int upper_bound(T w) {
    if (w < 0) return 0;
    int x = 0;
    for (int k = 1 << __lg(N); k; k >>= 1) {
      if (x + k <= N - 1 && data[x + k] <= w) {
        w -= data[x + k];
        x += k;
      }
    }
    return x;
  }
};

/**
 * @brief Binary Indexed Tree(Fenwick Tree)
 * @docs docs/data-structure/binary-indexed-tree.md
 */
#line 4 "dp/inversion-counting.hpp"

// 転倒数
template <typename T>
long long inversion_counting(const vector<T>& v) {
  vector<T> xs{v};
  sort(begin(xs), end(xs));
  xs.erase(unique(begin(xs), end(xs)), end(xs));
  int s = xs.size();
  BinaryIndexedTree<long long> bit(s + 1);
  long long ans = 0;
  for (auto& x : v) {
    int i = lower_bound(begin(xs), end(xs), x) - begin(xs);
    if (i + 1 != s) ans += bit.sum(i + 1, s - 1);
    bit.add(i, 1);
  }
  return ans;
}

// 隣接 swap によって v を w に変えるのにかかる手数 (不可能 : -1)
template <typename T>
long long swap_distance(const vector<T>& v, const vector<T>& w) {
  if (v.size() != w.size()) return -1;
  int N = v.size();
  vector<pair<T, int>> vv(N), ww(N);
  for (int i = 0; i < N; i++) {
    vv[i] = make_pair(v[i], i);
    ww[i] = make_pair(w[i], i);
  }
  sort(begin(vv), end(vv));
  sort(begin(ww), end(ww));
  for (int i = 0; i < N; i++) {
    if (vv[i].first != ww[i].first) return -1;
  }
  vector<int> order(N);
  for (int i = 0; i < N; i++) {
    order[vv[i].second] = ww[i].second;
  }
  return inversion_counting(order);
}

#line 2 "modulo/binomial.hpp"

#include <cassert>
#include <type_traits>
#include <vector>
using namespace std;

// コンストラクタの MAX に 「C(n, r) や fac(n) でクエリを投げる最大の n 」
// を入れると倍速くらいになる
// mod を超えて前計算して 0 割りを踏むバグは対策済み
template <typename T>
struct Binomial {
  vector<T> f, g, h;
  Binomial(int MAX = 0) {
    assert(T::get_mod() != 0 && "Binomial<mint>()");
    f.resize(1, T{1});
    g.resize(1, T{1});
    h.resize(1, T{1});
    if (MAX > 0) extend(MAX + 1);
  }

  void extend(int m = -1) {
    int n = f.size();
    if (m == -1) m = n * 2;
    m = min<int>(m, T::get_mod());
    if (n >= m) return;
    f.resize(m);
    g.resize(m);
    h.resize(m);
    for (int i = n; i < m; i++) f[i] = f[i - 1] * T(i);
    g[m - 1] = f[m - 1].inverse();
    h[m - 1] = g[m - 1] * f[m - 2];
    for (int i = m - 2; i >= n; i--) {
      g[i] = g[i + 1] * T(i + 1);
      h[i] = g[i] * f[i - 1];
    }
  }

  T fac(int i) {
    if (i < 0) return T(0);
    while (i >= (int)f.size()) extend();
    return f[i];
  }

  T finv(int i) {
    if (i < 0) return T(0);
    while (i >= (int)g.size()) extend();
    return g[i];
  }

  T inv(int i) {
    if (i < 0) return -inv(-i);
    while (i >= (int)h.size()) extend();
    return h[i];
  }

  T C(int n, int r) {
    if (n < 0 || n < r || r < 0) return T(0);
    return fac(n) * finv(n - r) * finv(r);
  }

  inline T operator()(int n, int r) { return C(n, r); }

  template <typename I>
  T multinomial(const vector<I>& r) {
    static_assert(is_integral<I>::value == true);
    int n = 0;
    for (auto& x : r) {
      if (x < 0) return T(0);
      n += x;
    }
    T res = fac(n);
    for (auto& x : r) res *= finv(x);
    return res;
  }

  template <typename I>
  T operator()(const vector<I>& r) {
    return multinomial(r);
  }

  T C_naive(int n, int r) {
    if (n < 0 || n < r || r < 0) return T(0);
    T ret = T(1);
    r = min(r, n - r);
    for (int i = 1; i <= r; ++i) ret *= inv(i) * (n--);
    return ret;
  }

  T P(int n, int r) {
    if (n < 0 || n < r || r < 0) return T(0);
    return fac(n) * finv(n - r);
  }

  // [x^r] 1 / (1-x)^n
  T H(int n, int r) {
    if (n < 0 || r < 0) return T(0);
    return r == 0 ? 1 : C(n + r - 1, r);
  }
};

using mint=LazyMontgomeryModInt<998244353>;

int main()
{
  	ios::sync_with_stdio(false);
	cin.tie(nullptr);
	int N,M;
	cin>>N>>M;
	vector<pi>constraints(M);
	vi fixed(N+1,0),cp(N+1,0),ck(N+1,0);
	rep(i,0,M){
		int p,k;
		cin>>p>>k;
		fixed[k]=p;
		constraints[i]={p,k};
		cp[p]=1; //いもす
		ck[k]=1;
	}

	rep(i,1,N+1){
		cp[i]+=cp[i-1];
		ck[i]+=ck[i-1];
	}

	vi v;
	rep(i,1,N+1){
		if(fixed[i])v.push_back(fixed[i]);
	}
	mint fixed_inv=inversion_counting(v);
	int r=N-M;
	Binomial<mint>comb(r);
	mint ans=comb.fac(r)*fixed_inv;

	if(r>0){
		mint R=r,sum=0;
		for(auto [p,k]:constraints){
			mint a=p-1-cp[p-1],b=k-1-ck[k-1];
			sum+=a*(R-b)+b*(R-a);
		}
		ans+=comb.fac(r-1)*sum;
	}

	cout<<ans+comb.fac(r)*r*(r-1)/4<<'\n';

}