Ejemplo n.º 1
0
BIGNUM *BN_mod_sqrt(BIGNUM *in, const BIGNUM *a, const BIGNUM *p, BN_CTX *ctx) 
/* Returns 'ret' such that
 *      ret^2 == a (mod p),
 * using the Tonelli/Shanks algorithm (cf. Henri Cohen, "A Course
 * in Algebraic Computational Number Theory", algorithm 1.5.1).
 * 'p' must be prime!
 */
	{
	BIGNUM *ret = in;
	int err = 1;
	int r;
	BIGNUM *A, *b, *q, *t, *x, *y;
	int e, i, j;
	
	if (!BN_is_odd(p) || BN_abs_is_word(p, 1))
		{
		if (BN_abs_is_word(p, 2))
			{
			if (ret == NULL)
				ret = BN_new();
			if (ret == NULL)
				goto end;
			if (!BN_set_word(ret, BN_is_bit_set(a, 0)))
				{
				if (ret != in)
					BN_free(ret);
				return NULL;
				}
			bn_check_top(ret);
			return ret;
			}

		BNerr(BN_F_BN_MOD_SQRT, BN_R_P_IS_NOT_PRIME);
		return(NULL);
		}

	if (BN_is_zero(a) || BN_is_one(a))
		{
		if (ret == NULL)
			ret = BN_new();
		if (ret == NULL)
			goto end;
		if (!BN_set_word(ret, BN_is_one(a)))
			{
			if (ret != in)
				BN_free(ret);
			return NULL;
			}
		bn_check_top(ret);
		return ret;
		}

	BN_CTX_start(ctx);
	A = BN_CTX_get(ctx);
	b = BN_CTX_get(ctx);
	q = BN_CTX_get(ctx);
	t = BN_CTX_get(ctx);
	x = BN_CTX_get(ctx);
	y = BN_CTX_get(ctx);
	if (y == NULL) goto end;
	
	if (ret == NULL)
		ret = BN_new();
	if (ret == NULL) goto end;

	/* A = a mod p */
	if (!BN_nnmod(A, a, p, ctx)) goto end;

	/* now write  |p| - 1  as  2^e*q  where  q  is odd */
	e = 1;
	while (!BN_is_bit_set(p, e))
		e++;
	/* we'll set  q  later (if needed) */

	if (e == 1)
		{
		/* The easy case:  (|p|-1)/2  is odd, so 2 has an inverse
		 * modulo  (|p|-1)/2,  and square roots can be computed
		 * directly by modular exponentiation.
		 * We have
		 *     2 * (|p|+1)/4 == 1   (mod (|p|-1)/2),
		 * so we can use exponent  (|p|+1)/4,  i.e.  (|p|-3)/4 + 1.
		 */
		if (!BN_rshift(q, p, 2)) goto end;
		q->neg = 0;
		if (!BN_add_word(q, 1)) goto end;
		if (!BN_mod_exp(ret, A, q, p, ctx)) goto end;
		err = 0;
		goto vrfy;
		}
	
	if (e == 2)
		{
		/* |p| == 5  (mod 8)
		 *
		 * In this case  2  is always a non-square since
		 * Legendre(2,p) = (-1)^((p^2-1)/8)  for any odd prime.
		 * So if  a  really is a square, then  2*a  is a non-square.
		 * Thus for
		 *      b := (2*a)^((|p|-5)/8),
		 *      i := (2*a)*b^2
		 * we have
		 *     i^2 = (2*a)^((1 + (|p|-5)/4)*2)
		 *         = (2*a)^((p-1)/2)
		 *         = -1;
		 * so if we set
		 *      x := a*b*(i-1),
		 * then
		 *     x^2 = a^2 * b^2 * (i^2 - 2*i + 1)
		 *         = a^2 * b^2 * (-2*i)
		 *         = a*(-i)*(2*a*b^2)
		 *         = a*(-i)*i
		 *         = a.
		 *
		 * (This is due to A.O.L. Atkin, 
		 * <URL: http://listserv.nodak.edu/scripts/wa.exe?A2=ind9211&L=nmbrthry&O=T&P=562>,
		 * November 1992.)
		 */

		/* t := 2*a */
		if (!BN_mod_lshift1_quick(t, A, p)) goto end;

		/* b := (2*a)^((|p|-5)/8) */
		if (!BN_rshift(q, p, 3)) goto end;
		q->neg = 0;
		if (!BN_mod_exp(b, t, q, p, ctx)) goto end;

		/* y := b^2 */
		if (!BN_mod_sqr(y, b, p, ctx)) goto end;

		/* t := (2*a)*b^2 - 1*/
		if (!BN_mod_mul(t, t, y, p, ctx)) goto end;
		if (!BN_sub_word(t, 1)) goto end;

		/* x = a*b*t */
		if (!BN_mod_mul(x, A, b, p, ctx)) goto end;
		if (!BN_mod_mul(x, x, t, p, ctx)) goto end;

		if (!BN_copy(ret, x)) goto end;
		err = 0;
		goto vrfy;
		}
	
	/* e > 2, so we really have to use the Tonelli/Shanks algorithm.
	 * First, find some  y  that is not a square. */
	if (!BN_copy(q, p)) goto end; /* use 'q' as temp */
	q->neg = 0;
	i = 2;
	do
		{
		/* For efficiency, try small numbers first;
		 * if this fails, try random numbers.
		 */
		if (i < 22)
			{
			if (!BN_set_word(y, i)) goto end;
			}
		else
			{
			if (!BN_pseudo_rand(y, BN_num_bits(p), 0, 0)) goto end;
			if (BN_ucmp(y, p) >= 0)
				{
				if (!(p->neg ? BN_add : BN_sub)(y, y, p)) goto end;
				}
			/* now 0 <= y < |p| */
			if (BN_is_zero(y))
				if (!BN_set_word(y, i)) goto end;
			}
		
		r = BN_kronecker(y, q, ctx); /* here 'q' is |p| */
		if (r < -1) goto end;
		if (r == 0)
			{
			/* m divides p */
			BNerr(BN_F_BN_MOD_SQRT, BN_R_P_IS_NOT_PRIME);
			goto end;
			}
		}
	while (r == 1 && ++i < 82);
	
	if (r != -1)
		{
		/* Many rounds and still no non-square -- this is more likely
		 * a bug than just bad luck.
		 * Even if  p  is not prime, we should have found some  y
		 * such that r == -1.
		 */
		BNerr(BN_F_BN_MOD_SQRT, BN_R_TOO_MANY_ITERATIONS);
		goto end;
		}

	/* Here's our actual 'q': */
	if (!BN_rshift(q, q, e)) goto end;

	/* Now that we have some non-square, we can find an element
	 * of order  2^e  by computing its q'th power. */
	if (!BN_mod_exp(y, y, q, p, ctx)) goto end;
	if (BN_is_one(y))
		{
		BNerr(BN_F_BN_MOD_SQRT, BN_R_P_IS_NOT_PRIME);
		goto end;
		}

	/* Now we know that (if  p  is indeed prime) there is an integer
	 * k,  0 <= k < 2^e,  such that
	 *
	 *      a^q * y^k == 1   (mod p).
	 *
	 * As  a^q  is a square and  y  is not,  k  must be even.
	 * q+1  is even, too, so there is an element
	 *
	 *     X := a^((q+1)/2) * y^(k/2),
	 *
	 * and it satisfies
	 *
	 *     X^2 = a^q * a     * y^k
	 *         = a,
	 *
	 * so it is the square root that we are looking for.
	 */
	
	/* t := (q-1)/2  (note that  q  is odd) */
	if (!BN_rshift1(t, q)) goto end;
	
	/* x := a^((q-1)/2) */
	if (BN_is_zero(t)) /* special case: p = 2^e + 1 */
		{
		if (!BN_nnmod(t, A, p, ctx)) goto end;
		if (BN_is_zero(t))
			{
			/* special case: a == 0  (mod p) */
			BN_zero(ret);
			err = 0;
			goto end;
			}
		else
			if (!BN_one(x)) goto end;
		}
	else
		{
		if (!BN_mod_exp(x, A, t, p, ctx)) goto end;
		if (BN_is_zero(x))
			{
			/* special case: a == 0  (mod p) */
			BN_zero(ret);
			err = 0;
			goto end;
			}
		}

	/* b := a*x^2  (= a^q) */
	if (!BN_mod_sqr(b, x, p, ctx)) goto end;
	if (!BN_mod_mul(b, b, A, p, ctx)) goto end;
	
	/* x := a*x    (= a^((q+1)/2)) */
	if (!BN_mod_mul(x, x, A, p, ctx)) goto end;

	while (1)
		{
		/* Now  b  is  a^q * y^k  for some even  k  (0 <= k < 2^E
		 * where  E  refers to the original value of  e,  which we
		 * don't keep in a variable),  and  x  is  a^((q+1)/2) * y^(k/2).
		 *
		 * We have  a*b = x^2,
		 *    y^2^(e-1) = -1,
		 *    b^2^(e-1) = 1.
		 */

		if (BN_is_one(b))
			{
			if (!BN_copy(ret, x)) goto end;
			err = 0;
			goto vrfy;
			}


		/* find smallest  i  such that  b^(2^i) = 1 */
		i = 1;
		if (!BN_mod_sqr(t, b, p, ctx)) goto end;
		while (!BN_is_one(t))
			{
			i++;
			if (i == e)
				{
				BNerr(BN_F_BN_MOD_SQRT, BN_R_NOT_A_SQUARE);
				goto end;
				}
			if (!BN_mod_mul(t, t, t, p, ctx)) goto end;
			}
		

		/* t := y^2^(e - i - 1) */
		if (!BN_copy(t, y)) goto end;
		for (j = e - i - 1; j > 0; j--)
			{
			if (!BN_mod_sqr(t, t, p, ctx)) goto end;
			}
		if (!BN_mod_mul(y, t, t, p, ctx)) goto end;
		if (!BN_mod_mul(x, x, t, p, ctx)) goto end;
		if (!BN_mod_mul(b, b, y, p, ctx)) goto end;
		e = i;
		}

 vrfy:
	if (!err)
		{
		/* verify the result -- the input might have been not a square
		 * (test added in 0.9.8) */
		
		if (!BN_mod_sqr(x, ret, p, ctx))
			err = 1;
		
		if (!err && 0 != BN_cmp(x, A))
			{
			BNerr(BN_F_BN_MOD_SQRT, BN_R_NOT_A_SQUARE);
			err = 1;
			}
		}

 end:
	if (err)
		{
		if (ret != NULL && ret != in)
			{
			BN_clear_free(ret);
			}
		ret = NULL;
		}
	BN_CTX_end(ctx);
	bn_check_top(ret);
	return ret;
	}
Ejemplo n.º 2
0
int ec_GFp_simple_set_compressed_coordinates(const EC_GROUP *group,
                                             EC_POINT *point, const BIGNUM *x_,
                                             int y_bit, BN_CTX *ctx) {
  BN_CTX *new_ctx = NULL;
  BIGNUM *tmp1, *tmp2, *x, *y;
  int ret = 0;

  ERR_clear_error();

  if (ctx == NULL) {
    ctx = new_ctx = BN_CTX_new();
    if (ctx == NULL) {
      return 0;
    }
  }

  y_bit = (y_bit != 0);

  BN_CTX_start(ctx);
  tmp1 = BN_CTX_get(ctx);
  tmp2 = BN_CTX_get(ctx);
  x = BN_CTX_get(ctx);
  y = BN_CTX_get(ctx);
  if (y == NULL) {
    goto err;
  }

  /* Recover y.  We have a Weierstrass equation
   *     y^2 = x^3 + a*x + b,
   * so  y  is one of the square roots of  x^3 + a*x + b. */

  /* tmp1 := x^3 */
  if (!BN_nnmod(x, x_, &group->field, ctx)) {
    goto err;
  }

  if (group->meth->field_decode == 0) {
    /* field_{sqr,mul} work on standard representation */
    if (!group->meth->field_sqr(group, tmp2, x_, ctx) ||
        !group->meth->field_mul(group, tmp1, tmp2, x_, ctx)) {
      goto err;
    }
  } else {
    if (!BN_mod_sqr(tmp2, x_, &group->field, ctx) ||
        !BN_mod_mul(tmp1, tmp2, x_, &group->field, ctx)) {
      goto err;
    }
  }

  /* tmp1 := tmp1 + a*x */
  if (group->a_is_minus3) {
    if (!BN_mod_lshift1_quick(tmp2, x, &group->field) ||
        !BN_mod_add_quick(tmp2, tmp2, x, &group->field) ||
        !BN_mod_sub_quick(tmp1, tmp1, tmp2, &group->field)) {
      goto err;
    }
  } else {
    if (group->meth->field_decode) {
      if (!group->meth->field_decode(group, tmp2, &group->a, ctx) ||
          !BN_mod_mul(tmp2, tmp2, x, &group->field, ctx)) {
        goto err;
      }
    } else {
      /* field_mul works on standard representation */
      if (!group->meth->field_mul(group, tmp2, &group->a, x, ctx)) {
        goto err;
      }
    }

    if (!BN_mod_add_quick(tmp1, tmp1, tmp2, &group->field)) {
      goto err;
    }
  }

  /* tmp1 := tmp1 + b */
  if (group->meth->field_decode) {
    if (!group->meth->field_decode(group, tmp2, &group->b, ctx) ||
        !BN_mod_add_quick(tmp1, tmp1, tmp2, &group->field)) {
      goto err;
    }
  } else {
    if (!BN_mod_add_quick(tmp1, tmp1, &group->b, &group->field)) {
      goto err;
    }
  }

  if (!BN_mod_sqrt(y, tmp1, &group->field, ctx)) {
    unsigned long err = ERR_peek_last_error();

    if (ERR_GET_LIB(err) == ERR_LIB_BN &&
        ERR_GET_REASON(err) == BN_R_NOT_A_SQUARE) {
      ERR_clear_error();
      OPENSSL_PUT_ERROR(EC, ec_GFp_simple_set_compressed_coordinates, EC_R_INVALID_COMPRESSED_POINT);
    } else {
      OPENSSL_PUT_ERROR(EC, ec_GFp_simple_set_compressed_coordinates, ERR_R_BN_LIB);
    }
    goto err;
  }

  if (y_bit != BN_is_odd(y)) {
    if (BN_is_zero(y)) {
      int kron;

      kron = BN_kronecker(x, &group->field, ctx);
      if (kron == -2) {
        goto err;
      }

      if (kron == 1) {
        OPENSSL_PUT_ERROR(EC, ec_GFp_simple_set_compressed_coordinates,
                          EC_R_INVALID_COMPRESSION_BIT);
      } else {
        /* BN_mod_sqrt() should have cought this error (not a square) */
        OPENSSL_PUT_ERROR(EC, ec_GFp_simple_set_compressed_coordinates,
                          EC_R_INVALID_COMPRESSED_POINT);
      }
      goto err;
    }
    if (!BN_usub(y, &group->field, y)) {
      goto err;
    }
  }
  if (y_bit != BN_is_odd(y)) {
    OPENSSL_PUT_ERROR(EC, ec_GFp_simple_set_compressed_coordinates,
                      ERR_R_INTERNAL_ERROR);
    goto err;
  }

  if (!EC_POINT_set_affine_coordinates_GFp(group, point, x, y, ctx))
    goto err;

  ret = 1;

err:
  BN_CTX_end(ctx);
  if (new_ctx != NULL)
    BN_CTX_free(new_ctx);
  return ret;
}
Ejemplo n.º 3
0
int ec_GFp_simple_dbl(const EC_GROUP *group, EC_POINT *r, const EC_POINT *a,
                      BN_CTX *ctx)
{
    int (*field_mul) (const EC_GROUP *, BIGNUM *, const BIGNUM *,
                      const BIGNUM *, BN_CTX *);
    int (*field_sqr) (const EC_GROUP *, BIGNUM *, const BIGNUM *, BN_CTX *);
    const BIGNUM *p;
    BN_CTX *new_ctx = NULL;
    BIGNUM *n0, *n1, *n2, *n3;
    int ret = 0;

    if (EC_POINT_is_at_infinity(group, a)) {
        BN_zero(r->Z);
        r->Z_is_one = 0;
        return 1;
    }

    field_mul = group->meth->field_mul;
    field_sqr = group->meth->field_sqr;
    p = group->field;

    if (ctx == NULL) {
        ctx = new_ctx = BN_CTX_new();
        if (ctx == NULL)
            return 0;
    }

    BN_CTX_start(ctx);
    n0 = BN_CTX_get(ctx);
    n1 = BN_CTX_get(ctx);
    n2 = BN_CTX_get(ctx);
    n3 = BN_CTX_get(ctx);
    if (n3 == NULL)
        goto err;

    /*
     * Note that in this function we must not read components of 'a' once we
     * have written the corresponding components of 'r'. ('r' might the same
     * as 'a'.)
     */

    /* n1 */
    if (a->Z_is_one) {
        if (!field_sqr(group, n0, a->X, ctx))
            goto err;
        if (!BN_mod_lshift1_quick(n1, n0, p))
            goto err;
        if (!BN_mod_add_quick(n0, n0, n1, p))
            goto err;
        if (!BN_mod_add_quick(n1, n0, group->a, p))
            goto err;
        /* n1 = 3 * X_a^2 + a_curve */
    } else if (group->a_is_minus3) {
        if (!field_sqr(group, n1, a->Z, ctx))
            goto err;
        if (!BN_mod_add_quick(n0, a->X, n1, p))
            goto err;
        if (!BN_mod_sub_quick(n2, a->X, n1, p))
            goto err;
        if (!field_mul(group, n1, n0, n2, ctx))
            goto err;
        if (!BN_mod_lshift1_quick(n0, n1, p))
            goto err;
        if (!BN_mod_add_quick(n1, n0, n1, p))
            goto err;
        /*-
         * n1 = 3 * (X_a + Z_a^2) * (X_a - Z_a^2)
         *    = 3 * X_a^2 - 3 * Z_a^4
         */
    } else {
        if (!field_sqr(group, n0, a->X, ctx))
            goto err;
        if (!BN_mod_lshift1_quick(n1, n0, p))
            goto err;
        if (!BN_mod_add_quick(n0, n0, n1, p))
            goto err;
        if (!field_sqr(group, n1, a->Z, ctx))
            goto err;
        if (!field_sqr(group, n1, n1, ctx))
            goto err;
        if (!field_mul(group, n1, n1, group->a, ctx))
            goto err;
        if (!BN_mod_add_quick(n1, n1, n0, p))
            goto err;
        /* n1 = 3 * X_a^2 + a_curve * Z_a^4 */
    }

    /* Z_r */
    if (a->Z_is_one) {
        if (!BN_copy(n0, a->Y))
            goto err;
    } else {
        if (!field_mul(group, n0, a->Y, a->Z, ctx))
            goto err;
    }
    if (!BN_mod_lshift1_quick(r->Z, n0, p))
        goto err;
    r->Z_is_one = 0;
    /* Z_r = 2 * Y_a * Z_a */

    /* n2 */
    if (!field_sqr(group, n3, a->Y, ctx))
        goto err;
    if (!field_mul(group, n2, a->X, n3, ctx))
        goto err;
    if (!BN_mod_lshift_quick(n2, n2, 2, p))
        goto err;
    /* n2 = 4 * X_a * Y_a^2 */

    /* X_r */
    if (!BN_mod_lshift1_quick(n0, n2, p))
        goto err;
    if (!field_sqr(group, r->X, n1, ctx))
        goto err;
    if (!BN_mod_sub_quick(r->X, r->X, n0, p))
        goto err;
    /* X_r = n1^2 - 2 * n2 */

    /* n3 */
    if (!field_sqr(group, n0, n3, ctx))
        goto err;
    if (!BN_mod_lshift_quick(n3, n0, 3, p))
        goto err;
    /* n3 = 8 * Y_a^4 */

    /* Y_r */
    if (!BN_mod_sub_quick(n0, n2, r->X, p))
        goto err;
    if (!field_mul(group, n0, n1, n0, ctx))
        goto err;
    if (!BN_mod_sub_quick(r->Y, n0, n3, p))
        goto err;
    /* Y_r = n1 * (n2 - X_r) - n3 */

    ret = 1;

 err:
    BN_CTX_end(ctx);
    BN_CTX_free(new_ctx);
    return ret;
}
Ejemplo n.º 4
0
int ec_GFp_simple_is_on_curve(const EC_GROUP *group, const EC_POINT *point,
                              BN_CTX *ctx)
{
    int (*field_mul) (const EC_GROUP *, BIGNUM *, const BIGNUM *,
                      const BIGNUM *, BN_CTX *);
    int (*field_sqr) (const EC_GROUP *, BIGNUM *, const BIGNUM *, BN_CTX *);
    const BIGNUM *p;
    BN_CTX *new_ctx = NULL;
    BIGNUM *rh, *tmp, *Z4, *Z6;
    int ret = -1;

    if (EC_POINT_is_at_infinity(group, point))
        return 1;

    field_mul = group->meth->field_mul;
    field_sqr = group->meth->field_sqr;
    p = group->field;

    if (ctx == NULL) {
        ctx = new_ctx = BN_CTX_new();
        if (ctx == NULL)
            return -1;
    }

    BN_CTX_start(ctx);
    rh = BN_CTX_get(ctx);
    tmp = BN_CTX_get(ctx);
    Z4 = BN_CTX_get(ctx);
    Z6 = BN_CTX_get(ctx);
    if (Z6 == NULL)
        goto err;

    /*-
     * We have a curve defined by a Weierstrass equation
     *      y^2 = x^3 + a*x + b.
     * The point to consider is given in Jacobian projective coordinates
     * where  (X, Y, Z)  represents  (x, y) = (X/Z^2, Y/Z^3).
     * Substituting this and multiplying by  Z^6  transforms the above equation into
     *      Y^2 = X^3 + a*X*Z^4 + b*Z^6.
     * To test this, we add up the right-hand side in 'rh'.
     */

    /* rh := X^2 */
    if (!field_sqr(group, rh, point->X, ctx))
        goto err;

    if (!point->Z_is_one) {
        if (!field_sqr(group, tmp, point->Z, ctx))
            goto err;
        if (!field_sqr(group, Z4, tmp, ctx))
            goto err;
        if (!field_mul(group, Z6, Z4, tmp, ctx))
            goto err;

        /* rh := (rh + a*Z^4)*X */
        if (group->a_is_minus3) {
            if (!BN_mod_lshift1_quick(tmp, Z4, p))
                goto err;
            if (!BN_mod_add_quick(tmp, tmp, Z4, p))
                goto err;
            if (!BN_mod_sub_quick(rh, rh, tmp, p))
                goto err;
            if (!field_mul(group, rh, rh, point->X, ctx))
                goto err;
        } else {
            if (!field_mul(group, tmp, Z4, group->a, ctx))
                goto err;
            if (!BN_mod_add_quick(rh, rh, tmp, p))
                goto err;
            if (!field_mul(group, rh, rh, point->X, ctx))
                goto err;
        }

        /* rh := rh + b*Z^6 */
        if (!field_mul(group, tmp, group->b, Z6, ctx))
            goto err;
        if (!BN_mod_add_quick(rh, rh, tmp, p))
            goto err;
    } else {
        /* point->Z_is_one */

        /* rh := (rh + a)*X */
        if (!BN_mod_add_quick(rh, rh, group->a, p))
            goto err;
        if (!field_mul(group, rh, rh, point->X, ctx))
            goto err;
        /* rh := rh + b */
        if (!BN_mod_add_quick(rh, rh, group->b, p))
            goto err;
    }

    /* 'lh' := Y^2 */
    if (!field_sqr(group, tmp, point->Y, ctx))
        goto err;

    ret = (0 == BN_ucmp(tmp, rh));

 err:
    BN_CTX_end(ctx);
    BN_CTX_free(new_ctx);
    return ret;
}
Ejemplo n.º 5
0
int ec_GFp_simple_add(const EC_GROUP *group, EC_POINT *r, const EC_POINT *a,
                      const EC_POINT *b, BN_CTX *ctx)
{
    int (*field_mul) (const EC_GROUP *, BIGNUM *, const BIGNUM *,
                      const BIGNUM *, BN_CTX *);
    int (*field_sqr) (const EC_GROUP *, BIGNUM *, const BIGNUM *, BN_CTX *);
    const BIGNUM *p;
    BN_CTX *new_ctx = NULL;
    BIGNUM *n0, *n1, *n2, *n3, *n4, *n5, *n6;
    int ret = 0;

    if (a == b)
        return EC_POINT_dbl(group, r, a, ctx);
    if (EC_POINT_is_at_infinity(group, a))
        return EC_POINT_copy(r, b);
    if (EC_POINT_is_at_infinity(group, b))
        return EC_POINT_copy(r, a);

    field_mul = group->meth->field_mul;
    field_sqr = group->meth->field_sqr;
    p = group->field;

    if (ctx == NULL) {
        ctx = new_ctx = BN_CTX_new();
        if (ctx == NULL)
            return 0;
    }

    BN_CTX_start(ctx);
    n0 = BN_CTX_get(ctx);
    n1 = BN_CTX_get(ctx);
    n2 = BN_CTX_get(ctx);
    n3 = BN_CTX_get(ctx);
    n4 = BN_CTX_get(ctx);
    n5 = BN_CTX_get(ctx);
    n6 = BN_CTX_get(ctx);
    if (n6 == NULL)
        goto end;

    /*
     * Note that in this function we must not read components of 'a' or 'b'
     * once we have written the corresponding components of 'r'. ('r' might
     * be one of 'a' or 'b'.)
     */

    /* n1, n2 */
    if (b->Z_is_one) {
        if (!BN_copy(n1, a->X))
            goto end;
        if (!BN_copy(n2, a->Y))
            goto end;
        /* n1 = X_a */
        /* n2 = Y_a */
    } else {
        if (!field_sqr(group, n0, b->Z, ctx))
            goto end;
        if (!field_mul(group, n1, a->X, n0, ctx))
            goto end;
        /* n1 = X_a * Z_b^2 */

        if (!field_mul(group, n0, n0, b->Z, ctx))
            goto end;
        if (!field_mul(group, n2, a->Y, n0, ctx))
            goto end;
        /* n2 = Y_a * Z_b^3 */
    }

    /* n3, n4 */
    if (a->Z_is_one) {
        if (!BN_copy(n3, b->X))
            goto end;
        if (!BN_copy(n4, b->Y))
            goto end;
        /* n3 = X_b */
        /* n4 = Y_b */
    } else {
        if (!field_sqr(group, n0, a->Z, ctx))
            goto end;
        if (!field_mul(group, n3, b->X, n0, ctx))
            goto end;
        /* n3 = X_b * Z_a^2 */

        if (!field_mul(group, n0, n0, a->Z, ctx))
            goto end;
        if (!field_mul(group, n4, b->Y, n0, ctx))
            goto end;
        /* n4 = Y_b * Z_a^3 */
    }

    /* n5, n6 */
    if (!BN_mod_sub_quick(n5, n1, n3, p))
        goto end;
    if (!BN_mod_sub_quick(n6, n2, n4, p))
        goto end;
    /* n5 = n1 - n3 */
    /* n6 = n2 - n4 */

    if (BN_is_zero(n5)) {
        if (BN_is_zero(n6)) {
            /* a is the same point as b */
            BN_CTX_end(ctx);
            ret = EC_POINT_dbl(group, r, a, ctx);
            ctx = NULL;
            goto end;
        } else {
            /* a is the inverse of b */
            BN_zero(r->Z);
            r->Z_is_one = 0;
            ret = 1;
            goto end;
        }
    }

    /* 'n7', 'n8' */
    if (!BN_mod_add_quick(n1, n1, n3, p))
        goto end;
    if (!BN_mod_add_quick(n2, n2, n4, p))
        goto end;
    /* 'n7' = n1 + n3 */
    /* 'n8' = n2 + n4 */

    /* Z_r */
    if (a->Z_is_one && b->Z_is_one) {
        if (!BN_copy(r->Z, n5))
            goto end;
    } else {
        if (a->Z_is_one) {
            if (!BN_copy(n0, b->Z))
                goto end;
        } else if (b->Z_is_one) {
            if (!BN_copy(n0, a->Z))
                goto end;
        } else {
            if (!field_mul(group, n0, a->Z, b->Z, ctx))
                goto end;
        }
        if (!field_mul(group, r->Z, n0, n5, ctx))
            goto end;
    }
    r->Z_is_one = 0;
    /* Z_r = Z_a * Z_b * n5 */

    /* X_r */
    if (!field_sqr(group, n0, n6, ctx))
        goto end;
    if (!field_sqr(group, n4, n5, ctx))
        goto end;
    if (!field_mul(group, n3, n1, n4, ctx))
        goto end;
    if (!BN_mod_sub_quick(r->X, n0, n3, p))
        goto end;
    /* X_r = n6^2 - n5^2 * 'n7' */

    /* 'n9' */
    if (!BN_mod_lshift1_quick(n0, r->X, p))
        goto end;
    if (!BN_mod_sub_quick(n0, n3, n0, p))
        goto end;
    /* n9 = n5^2 * 'n7' - 2 * X_r */

    /* Y_r */
    if (!field_mul(group, n0, n0, n6, ctx))
        goto end;
    if (!field_mul(group, n5, n4, n5, ctx))
        goto end;               /* now n5 is n5^3 */
    if (!field_mul(group, n1, n2, n5, ctx))
        goto end;
    if (!BN_mod_sub_quick(n0, n0, n1, p))
        goto end;
    if (BN_is_odd(n0))
        if (!BN_add(n0, n0, p))
            goto end;
    /* now  0 <= n0 < 2*p,  and n0 is even */
    if (!BN_rshift1(r->Y, n0))
        goto end;
    /* Y_r = (n6 * 'n9' - 'n8' * 'n5^3') / 2 */

    ret = 1;

 end:
    if (ctx)                    /* otherwise we already called BN_CTX_end */
        BN_CTX_end(ctx);
    BN_CTX_free(new_ctx);
    return ret;
}
Ejemplo n.º 6
0
/*-
 * Recovers the y-coordinate of r using Eq. (8) from Brier-Joye, "Weierstrass
 * Elliptic Curves and Side-Channel Attacks", modified to work in projective
 * coordinates and return r in Jacobian projective coordinates.
 *
 * X4 = two*Y1*X2*Z3*Z2*Z1;
 * Y4 = two*b*Z3*SQR(Z2*Z1) + Z3*(a*Z2*Z1+X1*X2)*(X1*Z2+X2*Z1) - X3*SQR(X1*Z2-X2*Z1);
 * Z4 = two*Y1*Z3*SQR(Z2)*Z1;
 *
 * Z4 != 0 because:
 *  - Z1==0 implies p is at infinity, which would have caused an early exit in
 *    the caller;
 *  - Z2==0 implies r is at infinity (handled by the BN_is_zero(r->Z) branch);
 *  - Z3==0 implies s is at infinity (handled by the BN_is_zero(s->Z) branch);
 *  - Y1==0 implies p has order 2, so either r or s are infinity and handled by
 *    one of the BN_is_zero(...) branches.
 */
int ec_GFp_simple_ladder_post(const EC_GROUP *group,
                              EC_POINT *r, EC_POINT *s,
                              EC_POINT *p, BN_CTX *ctx)
{
    int ret = 0;
    BIGNUM *t0, *t1, *t2, *t3, *t4, *t5, *t6 = NULL;

    if (BN_is_zero(r->Z))
        return EC_POINT_set_to_infinity(group, r);

    if (BN_is_zero(s->Z)) {
        /* (X,Y,Z) -> (XZ,YZ**2,Z) */
        if (!group->meth->field_mul(group, r->X, p->X, p->Z, ctx)
            || !group->meth->field_sqr(group, r->Z, p->Z, ctx)
            || !group->meth->field_mul(group, r->Y, p->Y, r->Z, ctx)
            || !BN_copy(r->Z, p->Z)
            || !EC_POINT_invert(group, r, ctx))
            return 0;
        return 1;
    }

    BN_CTX_start(ctx);
    t0 = BN_CTX_get(ctx);
    t1 = BN_CTX_get(ctx);
    t2 = BN_CTX_get(ctx);
    t3 = BN_CTX_get(ctx);
    t4 = BN_CTX_get(ctx);
    t5 = BN_CTX_get(ctx);
    t6 = BN_CTX_get(ctx);

    if (t6 == NULL
        || !BN_mod_lshift1_quick(t0, p->Y, group->field)
        || !group->meth->field_mul(group, t1, r->X, p->Z, ctx)
        || !group->meth->field_mul(group, t2, r->Z, s->Z, ctx)
        || !group->meth->field_mul(group, t2, t1, t2, ctx)
        || !group->meth->field_mul(group, t3, t2, t0, ctx)
        || !group->meth->field_mul(group, t2, r->Z, p->Z, ctx)
        || !group->meth->field_sqr(group, t4, t2, ctx)
        || !BN_mod_lshift1_quick(t5, group->b, group->field)
        || !group->meth->field_mul(group, t4, t4, t5, ctx)
        || !group->meth->field_mul(group, t6, t2, group->a, ctx)
        || !group->meth->field_mul(group, t5, r->X, p->X, ctx)
        || !BN_mod_add_quick(t5, t6, t5, group->field)
        || !group->meth->field_mul(group, t6, r->Z, p->X, ctx)
        || !BN_mod_add_quick(t2, t6, t1, group->field)
        || !group->meth->field_mul(group, t5, t5, t2, ctx)
        || !BN_mod_sub_quick(t6, t6, t1, group->field)
        || !group->meth->field_sqr(group, t6, t6, ctx)
        || !group->meth->field_mul(group, t6, t6, s->X, ctx)
        || !BN_mod_add_quick(t4, t5, t4, group->field)
        || !group->meth->field_mul(group, t4, t4, s->Z, ctx)
        || !BN_mod_sub_quick(t4, t4, t6, group->field)
        || !group->meth->field_sqr(group, t5, r->Z, ctx)
        || !group->meth->field_mul(group, r->Z, p->Z, s->Z, ctx)
        || !group->meth->field_mul(group, r->Z, t5, r->Z, ctx)
        || !group->meth->field_mul(group, r->Z, r->Z, t0, ctx)
        /* t3 := X, t4 := Y */
        /* (X,Y,Z) -> (XZ,YZ**2,Z) */
        || !group->meth->field_mul(group, r->X, t3, r->Z, ctx)
        || !group->meth->field_sqr(group, t3, r->Z, ctx)
        || !group->meth->field_mul(group, r->Y, t4, t3, ctx))
        goto err;

    ret = 1;

 err:
    BN_CTX_end(ctx);
    return ret;
}
Ejemplo n.º 7
0
/*-
 * Differential addition-and-doubling using  Eq. (9) and (10) from Izu-Takagi
 * "A fast parallel elliptic curve multiplication resistant against side channel
 * attacks", as described at
 * https://hyperelliptic.org/EFD/g1p/auto-shortw-xz.html#ladder-ladd-2002-it-4
 */
int ec_GFp_simple_ladder_step(const EC_GROUP *group,
                              EC_POINT *r, EC_POINT *s,
                              EC_POINT *p, BN_CTX *ctx)
{
    int ret = 0;
    BIGNUM *t0, *t1, *t2, *t3, *t4, *t5, *t6, *t7 = NULL;

    BN_CTX_start(ctx);
    t0 = BN_CTX_get(ctx);
    t1 = BN_CTX_get(ctx);
    t2 = BN_CTX_get(ctx);
    t3 = BN_CTX_get(ctx);
    t4 = BN_CTX_get(ctx);
    t5 = BN_CTX_get(ctx);
    t6 = BN_CTX_get(ctx);
    t7 = BN_CTX_get(ctx);

    if (t7 == NULL
        || !group->meth->field_mul(group, t0, r->X, s->X, ctx)
        || !group->meth->field_mul(group, t1, r->Z, s->Z, ctx)
        || !group->meth->field_mul(group, t2, r->X, s->Z, ctx)
        || !group->meth->field_mul(group, t3, r->Z, s->X, ctx)
        || !group->meth->field_mul(group, t4, group->a, t1, ctx)
        || !BN_mod_add_quick(t0, t0, t4, group->field)
        || !BN_mod_add_quick(t4, t3, t2, group->field)
        || !group->meth->field_mul(group, t0, t4, t0, ctx)
        || !group->meth->field_sqr(group, t1, t1, ctx)
        || !BN_mod_lshift_quick(t7, group->b, 2, group->field)
        || !group->meth->field_mul(group, t1, t7, t1, ctx)
        || !BN_mod_lshift1_quick(t0, t0, group->field)
        || !BN_mod_add_quick(t0, t1, t0, group->field)
        || !BN_mod_sub_quick(t1, t2, t3, group->field)
        || !group->meth->field_sqr(group, t1, t1, ctx)
        || !group->meth->field_mul(group, t3, t1, p->X, ctx)
        || !group->meth->field_mul(group, t0, p->Z, t0, ctx)
        /* s->X coord output */
        || !BN_mod_sub_quick(s->X, t0, t3, group->field)
        /* s->Z coord output */
        || !group->meth->field_mul(group, s->Z, p->Z, t1, ctx)
        || !group->meth->field_sqr(group, t3, r->X, ctx)
        || !group->meth->field_sqr(group, t2, r->Z, ctx)
        || !group->meth->field_mul(group, t4, t2, group->a, ctx)
        || !BN_mod_add_quick(t5, r->X, r->Z, group->field)
        || !group->meth->field_sqr(group, t5, t5, ctx)
        || !BN_mod_sub_quick(t5, t5, t3, group->field)
        || !BN_mod_sub_quick(t5, t5, t2, group->field)
        || !BN_mod_sub_quick(t6, t3, t4, group->field)
        || !group->meth->field_sqr(group, t6, t6, ctx)
        || !group->meth->field_mul(group, t0, t2, t5, ctx)
        || !group->meth->field_mul(group, t0, t7, t0, ctx)
        /* r->X coord output */
        || !BN_mod_sub_quick(r->X, t6, t0, group->field)
        || !BN_mod_add_quick(t6, t3, t4, group->field)
        || !group->meth->field_sqr(group, t3, t2, ctx)
        || !group->meth->field_mul(group, t7, t3, t7, ctx)
        || !group->meth->field_mul(group, t5, t5, t6, ctx)
        || !BN_mod_lshift1_quick(t5, t5, group->field)
        /* r->Z coord output */
        || !BN_mod_add_quick(r->Z, t7, t5, group->field))
        goto err;

    ret = 1;

 err:
    BN_CTX_end(ctx);
    return ret;
}
Ejemplo n.º 8
0
int ec_GFp_simple_dbl(const EC_GROUP *group, EC_POINT *r, const EC_POINT *a,
                      BN_CTX *ctx) {
  int (*field_mul)(const EC_GROUP *, BIGNUM *, const BIGNUM *, const BIGNUM *,
                   BN_CTX *);
  int (*field_sqr)(const EC_GROUP *, BIGNUM *, const BIGNUM *, BN_CTX *);
  const BIGNUM *p;
  BN_CTX *new_ctx = NULL;
  BIGNUM *n0, *n1, *n2, *n3;
  int ret = 0;

  if (EC_POINT_is_at_infinity(group, a)) {
    BN_zero(&r->Z);
    return 1;
  }

  field_mul = group->meth->field_mul;
  field_sqr = group->meth->field_sqr;
  p = &group->field;

  if (ctx == NULL) {
    ctx = new_ctx = BN_CTX_new();
    if (ctx == NULL) {
      return 0;
    }
  }

  BN_CTX_start(ctx);
  n0 = BN_CTX_get(ctx);
  n1 = BN_CTX_get(ctx);
  n2 = BN_CTX_get(ctx);
  n3 = BN_CTX_get(ctx);
  if (n3 == NULL) {
    goto err;
  }

  /* Note that in this function we must not read components of 'a'
   * once we have written the corresponding components of 'r'.
   * ('r' might the same as 'a'.)
   */

  /* n1 */
  if (BN_cmp(&a->Z, &group->one) == 0) {
    if (!field_sqr(group, n0, &a->X, ctx) ||
        !BN_mod_lshift1_quick(n1, n0, p) ||
        !BN_mod_add_quick(n0, n0, n1, p) ||
        !BN_mod_add_quick(n1, n0, &group->a, p)) {
      goto err;
    }
    /* n1 = 3 * X_a^2 + a_curve */
  } else {
    /* ring: This assumes a == -3. */
    if (!field_sqr(group, n1, &a->Z, ctx) ||
        !BN_mod_add_quick(n0, &a->X, n1, p) ||
        !BN_mod_sub_quick(n2, &a->X, n1, p) ||
        !field_mul(group, n1, n0, n2, ctx) ||
        !BN_mod_lshift1_quick(n0, n1, p) ||
        !BN_mod_add_quick(n1, n0, n1, p)) {
      goto err;
    }
    /* n1 = 3 * (X_a + Z_a^2) * (X_a - Z_a^2)
     *    = 3 * X_a^2 - 3 * Z_a^4 */
  }

  /* Z_r */
  if (BN_cmp(&a->Z, &group->one) == 0) {
    if (!BN_copy(n0, &a->Y)) {
      goto err;
    }
  } else if (!field_mul(group, n0, &a->Y, &a->Z, ctx)) {
    goto err;
  }
  if (!BN_mod_lshift1_quick(&r->Z, n0, p)) {
    goto err;
  }
  /* Z_r = 2 * Y_a * Z_a */

  /* n2 */
  if (!field_sqr(group, n3, &a->Y, ctx) ||
      !field_mul(group, n2, &a->X, n3, ctx) ||
      !BN_mod_lshift_quick(n2, n2, 2, p)) {
    goto err;
  }
  /* n2 = 4 * X_a * Y_a^2 */

  /* X_r */
  if (!BN_mod_lshift1_quick(n0, n2, p) ||
      !field_sqr(group, &r->X, n1, ctx) ||
      !BN_mod_sub_quick(&r->X, &r->X, n0, p)) {
    goto err;
  }
  /* X_r = n1^2 - 2 * n2 */

  /* n3 */
  if (!field_sqr(group, n0, n3, ctx) ||
      !BN_mod_lshift_quick(n3, n0, 3, p)) {
    goto err;
  }
  /* n3 = 8 * Y_a^4 */

  /* Y_r */
  if (!BN_mod_sub_quick(n0, n2, &r->X, p) ||
      !field_mul(group, n0, n1, n0, ctx) ||
      !BN_mod_sub_quick(&r->Y, n0, n3, p)) {
    goto err;
  }
  /* Y_r = n1 * (n2 - X_r) - n3 */

  ret = 1;

err:
  BN_CTX_end(ctx);
  BN_CTX_free(new_ctx);
  return ret;
}
Ejemplo n.º 9
0
BIGNUM *BN_mod_sqrt(BIGNUM *in, const BIGNUM *a, const BIGNUM *p, BN_CTX *ctx) {
  // Compute a square root of |a| mod |p| using the Tonelli/Shanks algorithm
  // (cf. Henri Cohen, "A Course in Algebraic Computational Number Theory",
  // algorithm 1.5.1). |p| is assumed to be a prime.

  BIGNUM *ret = in;
  int err = 1;
  int r;
  BIGNUM *A, *b, *q, *t, *x, *y;
  int e, i, j;

  if (!BN_is_odd(p) || BN_abs_is_word(p, 1)) {
    if (BN_abs_is_word(p, 2)) {
      if (ret == NULL) {
        ret = BN_new();
      }
      if (ret == NULL) {
        goto end;
      }
      if (!BN_set_word(ret, BN_is_bit_set(a, 0))) {
        if (ret != in) {
          BN_free(ret);
        }
        return NULL;
      }
      return ret;
    }

    OPENSSL_PUT_ERROR(BN, BN_R_P_IS_NOT_PRIME);
    return (NULL);
  }

  if (BN_is_zero(a) || BN_is_one(a)) {
    if (ret == NULL) {
      ret = BN_new();
    }
    if (ret == NULL) {
      goto end;
    }
    if (!BN_set_word(ret, BN_is_one(a))) {
      if (ret != in) {
        BN_free(ret);
      }
      return NULL;
    }
    return ret;
  }

  BN_CTX_start(ctx);
  A = BN_CTX_get(ctx);
  b = BN_CTX_get(ctx);
  q = BN_CTX_get(ctx);
  t = BN_CTX_get(ctx);
  x = BN_CTX_get(ctx);
  y = BN_CTX_get(ctx);
  if (y == NULL) {
    goto end;
  }

  if (ret == NULL) {
    ret = BN_new();
  }
  if (ret == NULL) {
    goto end;
  }

  // A = a mod p
  if (!BN_nnmod(A, a, p, ctx)) {
    goto end;
  }

  // now write  |p| - 1  as  2^e*q  where  q  is odd
  e = 1;
  while (!BN_is_bit_set(p, e)) {
    e++;
  }
  // we'll set  q  later (if needed)

  if (e == 1) {
    // The easy case:  (|p|-1)/2  is odd, so 2 has an inverse
    // modulo  (|p|-1)/2,  and square roots can be computed
    // directly by modular exponentiation.
    // We have
    //     2 * (|p|+1)/4 == 1   (mod (|p|-1)/2),
    // so we can use exponent  (|p|+1)/4,  i.e.  (|p|-3)/4 + 1.
    if (!BN_rshift(q, p, 2)) {
      goto end;
    }
    q->neg = 0;
    if (!BN_add_word(q, 1) ||
        !BN_mod_exp_mont(ret, A, q, p, ctx, NULL)) {
      goto end;
    }
    err = 0;
    goto vrfy;
  }

  if (e == 2) {
    // |p| == 5  (mod 8)
    //
    // In this case  2  is always a non-square since
    // Legendre(2,p) = (-1)^((p^2-1)/8)  for any odd prime.
    // So if  a  really is a square, then  2*a  is a non-square.
    // Thus for
    //      b := (2*a)^((|p|-5)/8),
    //      i := (2*a)*b^2
    // we have
    //     i^2 = (2*a)^((1 + (|p|-5)/4)*2)
    //         = (2*a)^((p-1)/2)
    //         = -1;
    // so if we set
    //      x := a*b*(i-1),
    // then
    //     x^2 = a^2 * b^2 * (i^2 - 2*i + 1)
    //         = a^2 * b^2 * (-2*i)
    //         = a*(-i)*(2*a*b^2)
    //         = a*(-i)*i
    //         = a.
    //
    // (This is due to A.O.L. Atkin,
    // <URL:
    //http://listserv.nodak.edu/scripts/wa.exe?A2=ind9211&L=nmbrthry&O=T&P=562>,
    // November 1992.)

    // t := 2*a
    if (!BN_mod_lshift1_quick(t, A, p)) {
      goto end;
    }

    // b := (2*a)^((|p|-5)/8)
    if (!BN_rshift(q, p, 3)) {
      goto end;
    }
    q->neg = 0;
    if (!BN_mod_exp_mont(b, t, q, p, ctx, NULL)) {
      goto end;
    }

    // y := b^2
    if (!BN_mod_sqr(y, b, p, ctx)) {
      goto end;
    }

    // t := (2*a)*b^2 - 1
    if (!BN_mod_mul(t, t, y, p, ctx) ||
        !BN_sub_word(t, 1)) {
      goto end;
    }

    // x = a*b*t
    if (!BN_mod_mul(x, A, b, p, ctx) ||
        !BN_mod_mul(x, x, t, p, ctx)) {
      goto end;
    }

    if (!BN_copy(ret, x)) {
      goto end;
    }
    err = 0;
    goto vrfy;
  }

  // e > 2, so we really have to use the Tonelli/Shanks algorithm.
  // First, find some  y  that is not a square.
  if (!BN_copy(q, p)) {
    goto end;  // use 'q' as temp
  }
  q->neg = 0;
  i = 2;
  do {
    // For efficiency, try small numbers first;
    // if this fails, try random numbers.
    if (i < 22) {
      if (!BN_set_word(y, i)) {
        goto end;
      }
    } else {
      if (!BN_pseudo_rand(y, BN_num_bits(p), 0, 0)) {
        goto end;
      }
      if (BN_ucmp(y, p) >= 0) {
        if (!(p->neg ? BN_add : BN_sub)(y, y, p)) {
          goto end;
        }
      }
      // now 0 <= y < |p|
      if (BN_is_zero(y)) {
        if (!BN_set_word(y, i)) {
          goto end;
        }
      }
    }

    r = bn_jacobi(y, q, ctx);  // here 'q' is |p|
    if (r < -1) {
      goto end;
    }
    if (r == 0) {
      // m divides p
      OPENSSL_PUT_ERROR(BN, BN_R_P_IS_NOT_PRIME);
      goto end;
    }
  } while (r == 1 && ++i < 82);

  if (r != -1) {
    // Many rounds and still no non-square -- this is more likely
    // a bug than just bad luck.
    // Even if  p  is not prime, we should have found some  y
    // such that r == -1.
    OPENSSL_PUT_ERROR(BN, BN_R_TOO_MANY_ITERATIONS);
    goto end;
  }

  // Here's our actual 'q':
  if (!BN_rshift(q, q, e)) {
    goto end;
  }

  // Now that we have some non-square, we can find an element
  // of order  2^e  by computing its q'th power.
  if (!BN_mod_exp_mont(y, y, q, p, ctx, NULL)) {
    goto end;
  }
  if (BN_is_one(y)) {
    OPENSSL_PUT_ERROR(BN, BN_R_P_IS_NOT_PRIME);
    goto end;
  }

  // Now we know that (if  p  is indeed prime) there is an integer
  // k,  0 <= k < 2^e,  such that
  //
  //      a^q * y^k == 1   (mod p).
  //
  // As  a^q  is a square and  y  is not,  k  must be even.
  // q+1  is even, too, so there is an element
  //
  //     X := a^((q+1)/2) * y^(k/2),
  //
  // and it satisfies
  //
  //     X^2 = a^q * a     * y^k
  //         = a,
  //
  // so it is the square root that we are looking for.

  // t := (q-1)/2  (note that  q  is odd)
  if (!BN_rshift1(t, q)) {
    goto end;
  }

  // x := a^((q-1)/2)
  if (BN_is_zero(t))  // special case: p = 2^e + 1
  {
    if (!BN_nnmod(t, A, p, ctx)) {
      goto end;
    }
    if (BN_is_zero(t)) {
      // special case: a == 0  (mod p)
      BN_zero(ret);
      err = 0;
      goto end;
    } else if (!BN_one(x)) {
      goto end;
    }
  } else {
    if (!BN_mod_exp_mont(x, A, t, p, ctx, NULL)) {
      goto end;
    }
    if (BN_is_zero(x)) {
      // special case: a == 0  (mod p)
      BN_zero(ret);
      err = 0;
      goto end;
    }
  }

  // b := a*x^2  (= a^q)
  if (!BN_mod_sqr(b, x, p, ctx) ||
      !BN_mod_mul(b, b, A, p, ctx)) {
    goto end;
  }

  // x := a*x    (= a^((q+1)/2))
  if (!BN_mod_mul(x, x, A, p, ctx)) {
    goto end;
  }

  while (1) {
    // Now  b  is  a^q * y^k  for some even  k  (0 <= k < 2^E
    // where  E  refers to the original value of  e,  which we
    // don't keep in a variable),  and  x  is  a^((q+1)/2) * y^(k/2).
    //
    // We have  a*b = x^2,
    //    y^2^(e-1) = -1,
    //    b^2^(e-1) = 1.

    if (BN_is_one(b)) {
      if (!BN_copy(ret, x)) {
        goto end;
      }
      err = 0;
      goto vrfy;
    }


    // find smallest  i  such that  b^(2^i) = 1
    i = 1;
    if (!BN_mod_sqr(t, b, p, ctx)) {
      goto end;
    }
    while (!BN_is_one(t)) {
      i++;
      if (i == e) {
        OPENSSL_PUT_ERROR(BN, BN_R_NOT_A_SQUARE);
        goto end;
      }
      if (!BN_mod_mul(t, t, t, p, ctx)) {
        goto end;
      }
    }


    // t := y^2^(e - i - 1)
    if (!BN_copy(t, y)) {
      goto end;
    }
    for (j = e - i - 1; j > 0; j--) {
      if (!BN_mod_sqr(t, t, p, ctx)) {
        goto end;
      }
    }
    if (!BN_mod_mul(y, t, t, p, ctx) ||
        !BN_mod_mul(x, x, t, p, ctx) ||
        !BN_mod_mul(b, b, y, p, ctx)) {
      goto end;
    }
    e = i;
  }

vrfy:
  if (!err) {
    // verify the result -- the input might have been not a square
    // (test added in 0.9.8)

    if (!BN_mod_sqr(x, ret, p, ctx)) {
      err = 1;
    }

    if (!err && 0 != BN_cmp(x, A)) {
      OPENSSL_PUT_ERROR(BN, BN_R_NOT_A_SQUARE);
      err = 1;
    }
  }

end:
  if (err) {
    if (ret != in) {
      BN_clear_free(ret);
    }
    ret = NULL;
  }
  BN_CTX_end(ctx);
  return ret;
}