int BN_mod_mul_montgomery(BIGNUM *r, const BIGNUM *a, const BIGNUM *b,
BN_MONT_CTX *mont, BN_CTX *ctx)
{
BIGNUM *tmp;
int ret=0;
BN_CTX_start(ctx);
tmp = BN_CTX_get(ctx);
if (tmp == NULL) goto err;
bn_check_top(tmp);
if (a == b)
{
if (!BN_sqr(tmp,a,ctx)) goto err;
}
else
{
if (!BN_mul(tmp,a,b,ctx)) goto err;
}
/* reduce from aRR to aR */
if (!BN_from_montgomery(r,tmp,mont,ctx)) goto err;
bn_check_top(r);
ret=1;
err:
BN_CTX_end(ctx);
return(ret);
}
// mod_montgomery sets |r| to |I| mod |p|. |I| must already be fully reduced
// modulo |p| times |q|. It returns one on success and zero on error.
static int mod_montgomery(BIGNUM *r, const BIGNUM *I, const BIGNUM *p,
const BN_MONT_CTX *mont_p, const BIGNUM *q,
BN_CTX *ctx) {
// Reducing in constant-time with Montgomery reduction requires I <= p * R. We
// have I < p * q, so this follows if q < R. In particular, this always holds
// if p and q are the same size, which is true for any RSA keys we or anyone
// sane generates. For other keys, we fall back to |BN_mod|.
if (!bn_less_than_montgomery_R(q, mont_p)) {
return BN_mod(r, I, p, ctx);
}
if (// Reduce mod p with Montgomery reduction. This computes I * R^-1 mod p.
!BN_from_montgomery(r, I, mont_p, ctx) ||
// Multiply by R^2 and do another Montgomery reduction to compute
// I * R^-1 * R^2 * R^-1 = I mod p.
!BN_to_montgomery(r, r, mont_p, ctx)) {
return 0;
}
// By precomputing R^3 mod p (normally |BN_MONT_CTX| only uses R^2 mod p) and
// adjusting the API for |BN_mod_exp_mont_consttime|, we could instead compute
// I * R mod p here and save a reduction per prime. But this would require
// changing the RSAZ code and may not be worth it. Note that the RSAZ code
// uses a different radix, so it uses R' = 2^1044. There we'd actually want
// R^2 * R', and would futher benefit from a precomputed R'^2. It currently
// converts |mont_p->RR| to R'^2.
return 1;
}
int ec_GFp_mont_field_decode(const EC_GROUP *group, BIGNUM *r, const BIGNUM *a,
BN_CTX *ctx) {
if (group->field_data1 == NULL) {
OPENSSL_PUT_ERROR(EC, ec_GFp_mont_field_decode, EC_R_NOT_INITIALIZED);
return 0;
}
return BN_from_montgomery(r, a, group->field_data1, ctx);
}
int ec_GFp_mont_field_decode(const EC_GROUP *group, BIGNUM *r, const BIGNUM *a, BN_CTX *ctx)
{
if (group->field_data1 == NULL)
{
ECerr(EC_F_EC_GFP_MONT_FIELD_DECODE, EC_R_NOT_INITIALIZED);
return 0;
}
return BN_from_montgomery(r, a, group->field_data1, ctx);
}
int bn_mul_mont_fixed_top(BIGNUM *r, const BIGNUM *a, const BIGNUM *b,
BN_MONT_CTX *mont, BN_CTX *ctx)
{
BIGNUM *tmp;
int ret = 0;
int num = mont->N.top;
#if defined(OPENSSL_BN_ASM_MONT) && defined(MONT_WORD)
if (num > 1 && a->top == num && b->top == num) {
if (bn_wexpand(r, num) == NULL)
return (0);
if (bn_mul_mont(r->d, a->d, b->d, mont->N.d, mont->n0, num)) {
r->neg = a->neg ^ b->neg;
r->top = num;
r->flags |= BN_FLG_FIXED_TOP;
return (1);
}
}
#endif
if ((a->top + b->top) > 2 * num)
return 0;
BN_CTX_start(ctx);
tmp = BN_CTX_get(ctx);
if (tmp == NULL)
goto err;
bn_check_top(tmp);
if (a == b) {
if (!BN_sqr(tmp, a, ctx))
goto err;
} else {
if (!BN_mul(tmp, a, b, ctx))
goto err;
}
/* reduce from aRR to aR */
#ifdef MONT_WORD
if (!bn_from_montgomery_word(r, tmp, mont))
goto err;
#else
if (!BN_from_montgomery(r, tmp, mont, ctx))
goto err;
#endif
ret = 1;
err:
BN_CTX_end(ctx);
return (ret);
}
static int bn_blinding_create_param(BN_BLINDING *b, const RSA *rsa, BN_CTX *ctx) {
int retry_counter = 32;
do {
if (!BN_rand_range(b->A, rsa->n)) {
OPENSSL_PUT_ERROR(RSA, ERR_R_INTERNAL_ERROR);
return 0;
}
/* `BN_from_montgomery` + `BN_mod_inverse_no_branch` is equivalent to, but
* more efficient than, `BN_mod_inverse_no_branch` + `BN_to_montgomery`. */
if (!BN_from_montgomery(b->Ai, b->A, rsa->mont_n, ctx)) {
return 0;
}
assert(BN_get_flags(b->A, BN_FLG_CONSTTIME));
int no_inverse;
if (BN_mod_inverse_no_branch(b->Ai, &no_inverse, b->Ai, rsa->n, ctx) ==
NULL) {
/* this should almost never happen for good RSA keys */
if (no_inverse) {
if (retry_counter-- == 0) {
OPENSSL_PUT_ERROR(RSA, RSA_R_TOO_MANY_ITERATIONS);
return 0;
}
ERR_clear_error();
} else {
OPENSSL_PUT_ERROR(RSA, ERR_R_INTERNAL_ERROR);
return 0;
}
} else {
break;
}
} while (1);
if (!BN_mod_exp_mont(b->A, b->A, rsa->e, rsa->n, ctx, rsa->mont_n)) {
OPENSSL_PUT_ERROR(RSA, ERR_R_INTERNAL_ERROR);
return 0;
}
if (!BN_to_montgomery(b->A, b->A, rsa->mont_n, ctx)) {
OPENSSL_PUT_ERROR(RSA, ERR_R_INTERNAL_ERROR);
return 0;
}
return 1;
}
int test_mont(BIO *bp, BN_CTX *ctx)
{
BIGNUM a,b,c,d,A,B;
BIGNUM n;
int i;
BN_MONT_CTX *mont;
BN_init(&a);
BN_init(&b);
BN_init(&c);
BN_init(&d);
BN_init(&A);
BN_init(&B);
BN_init(&n);
mont=BN_MONT_CTX_new();
BN_bntest_rand(&a,100,0,0); /**/
BN_bntest_rand(&b,100,0,0); /**/
for (i=0; i<num2; i++)
{
int bits = (200*(i+1))/num2;
if (bits == 0)
continue;
BN_bntest_rand(&n,bits,0,1);
BN_MONT_CTX_set(mont,&n,ctx);
BN_nnmod(&a,&a,&n,ctx);
BN_nnmod(&b,&b,&n,ctx);
BN_to_montgomery(&A,&a,mont,ctx);
BN_to_montgomery(&B,&b,mont,ctx);
BN_mod_mul_montgomery(&c,&A,&B,mont,ctx);/**/
BN_from_montgomery(&A,&c,mont,ctx);/**/
if (bp != NULL)
{
if (!results)
{
#ifdef undef
fprintf(stderr,"%d * %d %% %d\n",
BN_num_bits(&a),
BN_num_bits(&b),
BN_num_bits(mont->N));
#endif
BN_print(bp,&a);
BIO_puts(bp," * ");
BN_print(bp,&b);
BIO_puts(bp," % ");
BN_print(bp,&(mont->N));
BIO_puts(bp," - ");
}
BN_print(bp,&A);
BIO_puts(bp,"\n");
}
BN_mod_mul(&d,&a,&b,&n,ctx);
BN_sub(&d,&d,&A);
if(!BN_is_zero(&d))
{
fprintf(stderr,"Montgomery multiplication test failed!\n");
return 0;
}
}
BN_MONT_CTX_free(mont);
BN_free(&a);
BN_free(&b);
BN_free(&c);
BN_free(&d);
BN_free(&A);
BN_free(&B);
BN_free(&n);
return(1);
}
static int ec_GFp_mont_point_get_affine_coordinates(const EC_GROUP *group,
const EC_POINT *point,
BIGNUM *x, BIGNUM *y,
BN_CTX *ctx) {
if (EC_POINT_is_at_infinity(group, point)) {
OPENSSL_PUT_ERROR(EC, EC_R_POINT_AT_INFINITY);
return 0;
}
BN_CTX *new_ctx = NULL;
if (ctx == NULL) {
ctx = new_ctx = BN_CTX_new();
if (ctx == NULL) {
return 0;
}
}
int ret = 0;
BN_CTX_start(ctx);
if (BN_cmp(&point->Z, &group->one) == 0) {
/* |point| is already affine. */
if (x != NULL && !BN_from_montgomery(x, &point->X, &group->mont, ctx)) {
goto err;
}
if (y != NULL && !BN_from_montgomery(y, &point->Y, &group->mont, ctx)) {
goto err;
}
} else {
/* transform (X, Y, Z) into (x, y) := (X/Z^2, Y/Z^3) */
BIGNUM *Z_1 = BN_CTX_get(ctx);
BIGNUM *Z_2 = BN_CTX_get(ctx);
BIGNUM *Z_3 = BN_CTX_get(ctx);
if (Z_1 == NULL ||
Z_2 == NULL ||
Z_3 == NULL) {
goto err;
}
/* The straightforward way to calculate the inverse of a Montgomery-encoded
* value where the result is Montgomery-encoded is:
*
* |BN_from_montgomery| + |BN_mod_inverse| + |BN_to_montgomery|.
*
* This is equivalent, but more efficient, because |BN_from_montgomery|
* is more efficient (at least in theory) than |BN_to_montgomery|, since it
* doesn't have to do the multiplication before the reduction. */
if (!BN_from_montgomery(Z_1, &point->Z, &group->mont, ctx) ||
!BN_from_montgomery(Z_1, Z_1, &group->mont, ctx) ||
!BN_mod_inverse(Z_1, Z_1, &group->field, ctx)) {
goto err;
}
if (!BN_mod_mul_montgomery(Z_2, Z_1, Z_1, &group->mont, ctx)) {
goto err;
}
/* Instead of using |BN_from_montgomery| to convert the |x| coordinate
* and then calling |BN_from_montgomery| again to convert the |y|
* coordinate below, convert the common factor |Z_2| once now, saving one
* reduction. */
if (!BN_from_montgomery(Z_2, Z_2, &group->mont, ctx)) {
goto err;
}
if (x != NULL) {
if (!BN_mod_mul_montgomery(x, &point->X, Z_2, &group->mont, ctx)) {
goto err;
}
}
if (y != NULL) {
if (!BN_mod_mul_montgomery(Z_3, Z_2, Z_1, &group->mont, ctx) ||
!BN_mod_mul_montgomery(y, &point->Y, Z_3, &group->mont, ctx)) {
goto err;
}
}
}
ret = 1;
err:
BN_CTX_end(ctx);
BN_CTX_free(new_ctx);
return ret;
}
int ec_GFp_mont_field_decode(const EC_GROUP *group, BIGNUM *r, const BIGNUM *a,
BN_CTX *ctx) {
return BN_from_montgomery(r, a, &group->mont, ctx);
}
int BN_mod_exp2_mont(BIGNUM *rr, BIGNUM *a1, BIGNUM *p1, BIGNUM *a2,
BIGNUM *p2, BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *in_mont)
{
int i,j,k,bits,bits1,bits2,ret=0,wstart,wend,window,xvalue,yvalue;
int start=1,ts=0,x,y;
BIGNUM *d,*aa1,*aa2,*r;
BIGNUM val[EXP2_TABLE_SIZE][EXP2_TABLE_SIZE];
BN_MONT_CTX *mont=NULL;
bn_check_top(a1);
bn_check_top(p1);
bn_check_top(a2);
bn_check_top(p2);
bn_check_top(m);
if (!(m->d[0] & 1))
{
BNerr(BN_F_BN_MOD_EXP_MONT,BN_R_CALLED_WITH_EVEN_MODULUS);
return(0);
}
bits1=BN_num_bits(p1);
bits2=BN_num_bits(p2);
if ((bits1 == 0) && (bits2 == 0))
{
BN_one(rr);
return(1);
}
BN_CTX_start(ctx);
d = BN_CTX_get(ctx);
r = BN_CTX_get(ctx);
if (d == NULL || r == NULL) goto err;
bits=(bits1 > bits2)?bits1:bits2;
/* If this is not done, things will break in the montgomery
* part */
if (in_mont != NULL)
mont=in_mont;
else
{
if ((mont=BN_MONT_CTX_new()) == NULL) goto err;
if (!BN_MONT_CTX_set(mont,m,ctx)) goto err;
}
BN_init(&(val[0][0]));
BN_init(&(val[1][1]));
BN_init(&(val[0][1]));
BN_init(&(val[1][0]));
ts=1;
if (BN_ucmp(a1,m) >= 0)
{
BN_mod(&(val[1][0]),a1,m,ctx);
aa1= &(val[1][0]);
}
else
aa1=a1;
if (BN_ucmp(a2,m) >= 0)
{
BN_mod(&(val[0][1]),a2,m,ctx);
aa2= &(val[0][1]);
}
else
aa2=a2;
if (!BN_to_montgomery(&(val[1][0]),aa1,mont,ctx)) goto err;
if (!BN_to_montgomery(&(val[0][1]),aa2,mont,ctx)) goto err;
if (!BN_mod_mul_montgomery(&(val[1][1]),
&(val[1][0]),&(val[0][1]),mont,ctx))
goto err;
#if 0
if (bits <= 20) /* This is probably 3 or 0x10001, so just do singles */
window=1;
else if (bits > 250)
window=5; /* max size of window */
else if (bits >= 120)
window=4;
else
window=3;
#else
window=EXP2_TABLE_BITS;
#endif
k=1<<window;
for (x=0; x<k; x++)
{
if (x >= 2)
{
BN_init(&(val[x][0]));
BN_init(&(val[x][1]));
if (!BN_mod_mul_montgomery(&(val[x][0]),
&(val[1][0]),&(val[x-1][0]),mont,ctx)) goto err;
if (!BN_mod_mul_montgomery(&(val[x][1]),
&(val[1][0]),&(val[x-1][1]),mont,ctx)) goto err;
}
for (y=2; y<k; y++)
{
BN_init(&(val[x][y]));
if (!BN_mod_mul_montgomery(&(val[x][y]),
&(val[x][y-1]),&(val[0][1]),mont,ctx))
goto err;
}
}
ts=k;
start=1; /* This is used to avoid multiplication etc
* when there is only the value '1' in the
* buffer. */
xvalue=0; /* The 'x value' of the window */
yvalue=0; /* The 'y value' of the window */
wstart=bits-1; /* The top bit of the window */
wend=0; /* The bottom bit of the window */
if (!BN_to_montgomery(r,BN_value_one(),mont,ctx)) goto err;
for (;;)
{
xvalue=BN_is_bit_set(p1,wstart);
yvalue=BN_is_bit_set(p2,wstart);
if (!(xvalue || yvalue))
{
if (!start)
{
if (!BN_mod_mul_montgomery(r,r,r,mont,ctx))
goto err;
}
wstart--;
if (wstart < 0) break;
continue;
}
/* We now have wstart on a 'set' bit, we now need to work out
* how bit a window to do. To do this we need to scan
* forward until the last set bit before the end of the
* window */
j=wstart;
/* xvalue=BN_is_bit_set(p1,wstart); already set */
/* yvalue=BN_is_bit_set(p1,wstart); already set */
wend=0;
for (i=1; i<window; i++)
{
if (wstart-i < 0) break;
xvalue+=xvalue;
xvalue|=BN_is_bit_set(p1,wstart-i);
yvalue+=yvalue;
yvalue|=BN_is_bit_set(p2,wstart-i);
}
/* i is the size of the current window */
/* add the 'bytes above' */
if (!start)
for (j=0; j<i; j++)
{
if (!BN_mod_mul_montgomery(r,r,r,mont,ctx))
goto err;
}
/* wvalue will be an odd number < 2^window */
if (xvalue || yvalue)
{
if (!BN_mod_mul_montgomery(r,r,&(val[xvalue][yvalue]),
mont,ctx)) goto err;
}
/* move the 'window' down further */
wstart-=i;
start=0;
if (wstart < 0) break;
}
BN_from_montgomery(rr,r,mont,ctx);
ret=1;
err:
if ((in_mont == NULL) && (mont != NULL)) BN_MONT_CTX_free(mont);
BN_CTX_end(ctx);
for (i=0; i<ts; i++)
{
for (j=0; j<ts; j++)
{
BN_clear_free(&(val[i][j]));
}
}
return(ret);
}
