[thirdparty/openssl.git] / crypto / bn / bn_mont.c

/*
 * Copyright 1995-2018 The OpenSSL Project Authors. All Rights Reserved.
 *
 * Licensed under the OpenSSL license (the "License").  You may not use
 * this file except in compliance with the License.  You can obtain a copy
 * in the file LICENSE in the source distribution or at
 * https://www.openssl.org/source/license.html
 */

/*
 * Details about Montgomery multiplication algorithms can be found at
 * http://security.ece.orst.edu/publications.html, e.g.
 * http://security.ece.orst.edu/koc/papers/j37acmon.pdf and
 * sections 3.8 and 4.2 in http://security.ece.orst.edu/koc/papers/r01rsasw.pdf
 */

#include "internal/cryptlib.h"
#include "bn_lcl.h"

#define MONT_WORD               /* use the faster word-based algorithm */

#ifdef MONT_WORD
static int bn_from_montgomery_word(BIGNUM *ret, BIGNUM *r, BN_MONT_CTX *mont);
#endif

int BN_mod_mul_montgomery(BIGNUM *r, const BIGNUM *a, const BIGNUM *b,
                          BN_MONT_CTX *mont, BN_CTX *ctx)
{
    int ret = bn_mul_mont_fixed_top(r, a, b, mont, ctx);

    bn_correct_top(r);
    bn_check_top(r);

    return ret;
}

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;
}

#ifdef MONT_WORD
static int bn_from_montgomery_word(BIGNUM *ret, BIGNUM *r, BN_MONT_CTX *mont)
{
    BIGNUM *n;
    BN_ULONG *ap, *np, *rp, n0, v, carry;
    int nl, max, i;

    n = &(mont->N);
    nl = n->top;
    if (nl == 0) {
        ret->top = 0;
        return 1;
    }

    max = (2 * nl);             /* carry is stored separately */
    if (bn_wexpand(r, max) == NULL)
        return 0;

    r->neg ^= n->neg;
    np = n->d;
    rp = r->d;

    /* clear the top words of T */
    i = max - r->top;
    if (i)
        memset(&rp[r->top], 0, sizeof(*rp) * i);

    r->top = max;
    r->flags |= BN_FLG_FIXED_TOP;
    n0 = mont->n0[0];

    /*
     * Add multiples of |n| to |r| until R = 2^(nl * BN_BITS2) divides it. On
     * input, we had |r| < |n| * R, so now |r| < 2 * |n| * R. Note that |r|
     * includes |carry| which is stored separately.
     */
    for (carry = 0, i = 0; i < nl; i++, rp++) {
        v = bn_mul_add_words(rp, np, nl, (rp[0] * n0) & BN_MASK2);
        v = (v + carry + rp[nl]) & BN_MASK2;
        carry |= (v != rp[nl]);
        carry &= (v <= rp[nl]);
        rp[nl] = v;
    }

    if (bn_wexpand(ret, nl) == NULL)
        return 0;
    ret->top = nl;
    ret->flags |= BN_FLG_FIXED_TOP;
    ret->neg = r->neg;

    rp = ret->d;

    /*
     * Shift |nl| words to divide by R. We have |ap| < 2 * |n|. Note that |ap|
     * includes |carry| which is stored separately.
     */
    ap = &(r->d[nl]);

    carry -= bn_sub_words(rp, ap, np, nl);
    /*
     * |carry| is -1 if |ap| - |np| underflowed or zero if it did not. Note
     * |carry| cannot be 1. That would imply the subtraction did not fit in
     * |nl| words, and we know at most one subtraction is needed.
     */
    for (i = 0; i < nl; i++) {
        rp[i] = (carry & ap[i]) | (~carry & rp[i]);
        ap[i] = 0;
    }

    return 1;
}
#endif                          /* MONT_WORD */

int BN_from_montgomery(BIGNUM *ret, const BIGNUM *a, BN_MONT_CTX *mont,
                       BN_CTX *ctx)
{
    int retn = 0;
#ifdef MONT_WORD
    BIGNUM *t;

    BN_CTX_start(ctx);
    if ((t = BN_CTX_get(ctx)) && BN_copy(t, a)) {
        retn = bn_from_montgomery_word(ret, t, mont);
        bn_correct_top(ret);
        bn_check_top(ret);
    }
    BN_CTX_end(ctx);
#else                           /* !MONT_WORD */
    BIGNUM *t1, *t2;

    BN_CTX_start(ctx);
    t1 = BN_CTX_get(ctx);
    t2 = BN_CTX_get(ctx);
    if (t2 == NULL)
        goto err;

    if (!BN_copy(t1, a))
        goto err;
    BN_mask_bits(t1, mont->ri);

    if (!BN_mul(t2, t1, &mont->Ni, ctx))
        goto err;
    BN_mask_bits(t2, mont->ri);

    if (!BN_mul(t1, t2, &mont->N, ctx))
        goto err;
    if (!BN_add(t2, a, t1))
        goto err;
    if (!BN_rshift(ret, t2, mont->ri))
        goto err;

    if (BN_ucmp(ret, &(mont->N)) >= 0) {
        if (!BN_usub(ret, ret, &(mont->N)))
            goto err;
    }
    retn = 1;
    bn_check_top(ret);
 err:
    BN_CTX_end(ctx);
#endif                          /* MONT_WORD */
    return retn;
}

int bn_to_mont_fixed_top(BIGNUM *r, const BIGNUM *a, BN_MONT_CTX *mont,
                         BN_CTX *ctx)
{
    return bn_mul_mont_fixed_top(r, a, &(mont->RR), mont, ctx);
}

BN_MONT_CTX *BN_MONT_CTX_new(void)
{
    BN_MONT_CTX *ret;

    if ((ret = OPENSSL_malloc(sizeof(*ret))) == NULL) {
        BNerr(BN_F_BN_MONT_CTX_NEW, ERR_R_MALLOC_FAILURE);
        return NULL;
    }

    BN_MONT_CTX_init(ret);
    ret->flags = BN_FLG_MALLOCED;
    return ret;
}

void BN_MONT_CTX_init(BN_MONT_CTX *ctx)
{
    ctx->ri = 0;
    bn_init(&ctx->RR);
    bn_init(&ctx->N);
    bn_init(&ctx->Ni);
    ctx->n0[0] = ctx->n0[1] = 0;
    ctx->flags = 0;
}

void BN_MONT_CTX_free(BN_MONT_CTX *mont)
{
    if (mont == NULL)
        return;
    BN_clear_free(&mont->RR);
    BN_clear_free(&mont->N);
    BN_clear_free(&mont->Ni);
    if (mont->flags & BN_FLG_MALLOCED)
        OPENSSL_free(mont);
}

int BN_MONT_CTX_set(BN_MONT_CTX *mont, const BIGNUM *mod, BN_CTX *ctx)
{
    int i, ret = 0;
    BIGNUM *Ri, *R;

    if (BN_is_zero(mod))
        return 0;

    BN_CTX_start(ctx);
    if ((Ri = BN_CTX_get(ctx)) == NULL)
        goto err;
    R = &(mont->RR);            /* grab RR as a temp */
    if (!BN_copy(&(mont->N), mod))
        goto err;               /* Set N */
    if (BN_get_flags(mod, BN_FLG_CONSTTIME) != 0)
        BN_set_flags(&(mont->N), BN_FLG_CONSTTIME);
    mont->N.neg = 0;

#ifdef MONT_WORD
    {
        BIGNUM tmod;
        BN_ULONG buf[2];

        bn_init(&tmod);
        tmod.d = buf;
        tmod.dmax = 2;
        tmod.neg = 0;

        if (BN_get_flags(mod, BN_FLG_CONSTTIME) != 0)
            BN_set_flags(&tmod, BN_FLG_CONSTTIME);

        mont->ri = (BN_num_bits(mod) + (BN_BITS2 - 1)) / BN_BITS2 * BN_BITS2;

# if defined(OPENSSL_BN_ASM_MONT) && (BN_BITS2<=32)
        /*
         * Only certain BN_BITS2<=32 platforms actually make use of n0[1],
         * and we could use the #else case (with a shorter R value) for the
         * others.  However, currently only the assembler files do know which
         * is which.
         */

        BN_zero(R);
        if (!(BN_set_bit(R, 2 * BN_BITS2)))
            goto err;

        tmod.top = 0;
        if ((buf[0] = mod->d[0]))
            tmod.top = 1;
        if ((buf[1] = mod->top > 1 ? mod->d[1] : 0))
            tmod.top = 2;

        if (BN_is_one(&tmod))
            BN_zero(Ri);
        else if ((BN_mod_inverse(Ri, R, &tmod, ctx)) == NULL)
            goto err;
        if (!BN_lshift(Ri, Ri, 2 * BN_BITS2))
            goto err;           /* R*Ri */
        if (!BN_is_zero(Ri)) {
            if (!BN_sub_word(Ri, 1))
                goto err;
        } else {                /* if N mod word size == 1 */

            if (bn_expand(Ri, (int)sizeof(BN_ULONG) * 2) == NULL)
                goto err;
            /* Ri-- (mod double word size) */
            Ri->neg = 0;
            Ri->d[0] = BN_MASK2;
            Ri->d[1] = BN_MASK2;
            Ri->top = 2;
        }
        if (!BN_div(Ri, NULL, Ri, &tmod, ctx))
            goto err;
        /*
         * Ni = (R*Ri-1)/N, keep only couple of least significant words:
         */
        mont->n0[0] = (Ri->top > 0) ? Ri->d[0] : 0;
        mont->n0[1] = (Ri->top > 1) ? Ri->d[1] : 0;
# else
        BN_zero(R);
        if (!(BN_set_bit(R, BN_BITS2)))
            goto err;           /* R */

        buf[0] = mod->d[0];     /* tmod = N mod word size */
        buf[1] = 0;
        tmod.top = buf[0] != 0 ? 1 : 0;
        /* Ri = R^-1 mod N */
        if (BN_is_one(&tmod))
            BN_zero(Ri);
        else if ((BN_mod_inverse(Ri, R, &tmod, ctx)) == NULL)
            goto err;
        if (!BN_lshift(Ri, Ri, BN_BITS2))
            goto err;           /* R*Ri */
        if (!BN_is_zero(Ri)) {
            if (!BN_sub_word(Ri, 1))
                goto err;
        } else {                /* if N mod word size == 1 */

            if (!BN_set_word(Ri, BN_MASK2))
                goto err;       /* Ri-- (mod word size) */
        }
        if (!BN_div(Ri, NULL, Ri, &tmod, ctx))
            goto err;
        /*
         * Ni = (R*Ri-1)/N, keep only least significant word:
         */
        mont->n0[0] = (Ri->top > 0) ? Ri->d[0] : 0;
        mont->n0[1] = 0;
# endif
    }
#else                           /* !MONT_WORD */
    {                           /* bignum version */
        mont->ri = BN_num_bits(&mont->N);
        BN_zero(R);
        if (!BN_set_bit(R, mont->ri))
            goto err;           /* R = 2^ri */
        /* Ri = R^-1 mod N */
        if ((BN_mod_inverse(Ri, R, &mont->N, ctx)) == NULL)
            goto err;
        if (!BN_lshift(Ri, Ri, mont->ri))
            goto err;           /* R*Ri */
        if (!BN_sub_word(Ri, 1))
            goto err;
        /*
         * Ni = (R*Ri-1) / N
         */
        if (!BN_div(&(mont->Ni), NULL, Ri, &mont->N, ctx))
            goto err;
    }
#endif

    /* setup RR for conversions */
    BN_zero(&(mont->RR));
    if (!BN_set_bit(&(mont->RR), mont->ri * 2))
        goto err;
    if (!BN_mod(&(mont->RR), &(mont->RR), &(mont->N), ctx))
        goto err;

    for (i = mont->RR.top, ret = mont->N.top; i < ret; i++)
        mont->RR.d[i] = 0;
    mont->RR.top = ret;
    mont->RR.flags |= BN_FLG_FIXED_TOP;

    ret = 1;
 err:
    BN_CTX_end(ctx);
    return ret;
}

BN_MONT_CTX *BN_MONT_CTX_copy(BN_MONT_CTX *to, BN_MONT_CTX *from)
{
    if (to == from)
        return to;

    if (!BN_copy(&(to->RR), &(from->RR)))
        return NULL;
    if (!BN_copy(&(to->N), &(from->N)))
        return NULL;
    if (!BN_copy(&(to->Ni), &(from->Ni)))
        return NULL;
    to->ri = from->ri;
    to->n0[0] = from->n0[0];
    to->n0[1] = from->n0[1];
    return to;
}

BN_MONT_CTX *BN_MONT_CTX_set_locked(BN_MONT_CTX **pmont, CRYPTO_RWLOCK *lock,
                                    const BIGNUM *mod, BN_CTX *ctx)
{
    BN_MONT_CTX *ret;

    CRYPTO_THREAD_read_lock(lock);
    ret = *pmont;
    CRYPTO_THREAD_unlock(lock);
    if (ret)
        return ret;

    /*
     * We don't want to serialise globally while doing our lazy-init math in
     * BN_MONT_CTX_set. That punishes threads that are doing independent
     * things. Instead, punish the case where more than one thread tries to
     * lazy-init the same 'pmont', by having each do the lazy-init math work
     * independently and only use the one from the thread that wins the race
     * (the losers throw away the work they've done).
     */
    ret = BN_MONT_CTX_new();
    if (ret == NULL)
        return NULL;
    if (!BN_MONT_CTX_set(ret, mod, ctx)) {
        BN_MONT_CTX_free(ret);
        return NULL;
    }

    /* The locked compare-and-set, after the local work is done. */
    CRYPTO_THREAD_write_lock(lock);
    if (*pmont) {
        BN_MONT_CTX_free(ret);
        ret = *pmont;
    } else
        *pmont = ret;
    CRYPTO_THREAD_unlock(lock);
    return ret;
}
Commit	Line	Data
4f22f405	1	/*
48e5119a	2	* Copyright 1995-2018 The OpenSSL Project Authors. All Rights Reserved.
675f605d	3	*
4f22f405 RS	4	* Licensed under the OpenSSL license (the "License"). You may not use
	5	* this file except in compliance with the License. You can obtain a copy
	6	* in the file LICENSE in the source distribution or at
	7	* https://www.openssl.org/source/license.html
675f605d	8	*/
d02b48c6	9
1b3b0a54	10	/*
b99b1107 BM	11	* Details about Montgomery multiplication algorithms can be found at
	12	* http://security.ece.orst.edu/publications.html, e.g.
	13	* http://security.ece.orst.edu/koc/papers/j37acmon.pdf and
	14	* sections 3.8 and 4.2 in http://security.ece.orst.edu/koc/papers/r01rsasw.pdf
1b3b0a54 RE	15	*/
1b3b0a54 RE	16
b39fc560	17	#include "internal/cryptlib.h"
d02b48c6 RE	18	#include "bn_lcl.h"
d02b48c6 RE	19
0f113f3e	20	#define MONT_WORD /* use the faster word-based algorithm */
6535eb17	21
9b4eab50	22	#ifdef MONT_WORD
71883868	23	static int bn_from_montgomery_word(BIGNUM ret, BIGNUM r, BN_MONT_CTX *mont);
9b4eab50 AP	24	#endif
9b4eab50 AP	25
020fc820	26	int BN_mod_mul_montgomery(BIGNUM r, const BIGNUM a, const BIGNUM *b,
0f113f3e	27	BN_MONT_CTX mont, BN_CTX ctx)
71883868 AP	28	{
	29	int ret = bn_mul_mont_fixed_top(r, a, b, mont, ctx);
	30
	31	bn_correct_top(r);
	32	bn_check_top(r);
	33
	34	return ret;
	35	}
	36
	37	int bn_mul_mont_fixed_top(BIGNUM r, const BIGNUM a, const BIGNUM *b,
	38	BN_MONT_CTX mont, BN_CTX ctx)
0f113f3e MC	39	{
	40	BIGNUM *tmp;
	41	int ret = 0;
0f113f3e MC	42	int num = mont->N.top;
0f113f3e MC	43
3c97e412	44	#if defined(OPENSSL_BN_ASM_MONT) && defined(MONT_WORD)
0f113f3e MC	45	if (num > 1 && a->top == num && b->top == num) {
0f113f3e MC	46	if (bn_wexpand(r, num) == NULL)
26a7d938	47	return 0;
0f113f3e MC	48	if (bn_mul_mont(r->d, a->d, b->d, mont->N.d, mont->n0, num)) {
	49	r->neg = a->neg ^ b->neg;
	50	r->top = num;
71883868	51	r->flags \|= BN_FLG_FIXED_TOP;
208fb891	52	return 1;
0f113f3e MC	53	}
0f113f3e MC	54	}
e7382805	55	#endif
dfeab068	56
3c97e412 AP	57	if ((a->top + b->top) > 2 * num)
	58	return 0;
	59
0f113f3e MC	60	BN_CTX_start(ctx);
	61	tmp = BN_CTX_get(ctx);
	62	if (tmp == NULL)
	63	goto err;
	64
	65	bn_check_top(tmp);
	66	if (a == b) {
	67	if (!BN_sqr(tmp, a, ctx))
	68	goto err;
	69	} else {
	70	if (!BN_mul(tmp, a, b, ctx))
	71	goto err;
	72	}
	73	/* reduce from aRR to aR */
9b4eab50	74	#ifdef MONT_WORD
71883868	75	if (!bn_from_montgomery_word(r, tmp, mont))
0f113f3e	76	goto err;
9b4eab50	77	#else
0f113f3e MC	78	if (!BN_from_montgomery(r, tmp, mont, ctx))
0f113f3e MC	79	goto err;
9b4eab50	80	#endif
0f113f3e MC	81	ret = 1;
	82	err:
	83	BN_CTX_end(ctx);
26a7d938	84	return ret;
0f113f3e	85	}
d02b48c6	86
6535eb17	87	#ifdef MONT_WORD
71883868	88	static int bn_from_montgomery_word(BIGNUM ret, BIGNUM r, BN_MONT_CTX *mont)
0f113f3e MC	89	{
	90	BIGNUM *n;
	91	BN_ULONG ap, np, *rp, n0, v, carry;
	92	int nl, max, i;
	93
	94	n = &(mont->N);
	95	nl = n->top;
	96	if (nl == 0) {
	97	ret->top = 0;
208fb891	98	return 1;
0f113f3e MC	99	}
	100
	101	max = (2 * nl); /* carry is stored separately */
	102	if (bn_wexpand(r, max) == NULL)
26a7d938	103	return 0;
0f113f3e MC	104
	105	r->neg ^= n->neg;
	106	np = n->d;
	107	rp = r->d;
	108
	109	/* clear the top words of T */
9f040d6d RS	110	i = max - r->top;
	111	if (i)
	112	memset(&rp[r->top], 0, sizeof(rp) i);
0f113f3e MC	113
0f113f3e MC	114	r->top = max;
71883868	115	r->flags \|= BN_FLG_FIXED_TOP;
0f113f3e MC	116	n0 = mont->n0[0];
0f113f3e MC	117
f345b1f3 DB	118	/*
	119	* Add multiples of \|n\| to \|r\| until R = 2^(nl * BN_BITS2) divides it. On
	120	* input, we had \|r\| < \|n\| * R, so now \|r\| < 2 * \|n\| * R. Note that \|r\|
	121	* includes \|carry\| which is stored separately.
	122	*/
0f113f3e MC	123	for (carry = 0, i = 0; i < nl; i++, rp++) {
	124	v = bn_mul_add_words(rp, np, nl, (rp[0] * n0) & BN_MASK2);
	125	v = (v + carry + rp[nl]) & BN_MASK2;
	126	carry \|= (v != rp[nl]);
	127	carry &= (v <= rp[nl]);
	128	rp[nl] = v;
	129	}
	130
	131	if (bn_wexpand(ret, nl) == NULL)
26a7d938	132	return 0;
0f113f3e	133	ret->top = nl;
71883868	134	ret->flags \|= BN_FLG_FIXED_TOP;
0f113f3e MC	135	ret->neg = r->neg;
	136
	137	rp = ret->d;
f345b1f3 DB	138
	139	/*
	140	* Shift \|nl\| words to divide by R. We have \|ap\| < 2 * \|n\|. Note that \|ap\|
	141	* includes \|carry\| which is stored separately.
	142	*/
0f113f3e MC	143	ap = &(r->d[nl]);
0f113f3e MC	144
6c90182a	145	carry -= bn_sub_words(rp, ap, np, nl);
f345b1f3	146	/*
6c90182a AP	147	* \|carry\| is -1 if \|ap\| - \|np\| underflowed or zero if it did not. Note
	148	* \|carry\| cannot be 1. That would imply the subtraction did not fit in
	149	* \|nl\| words, and we know at most one subtraction is needed.
f345b1f3	150	*/
f345b1f3	151	for (i = 0; i < nl; i++) {
6c90182a	152	rp[i] = (carry & ap[i]) \| (~carry & rp[i]);
f345b1f3	153	ap[i] = 0;
0f113f3e	154	}
0f113f3e	155
208fb891	156	return 1;
0f113f3e MC	157	}
0f113f3e MC	158	#endif /* MONT_WORD */
9b4eab50 AP	159
9b4eab50 AP	160	int BN_from_montgomery(BIGNUM ret, const BIGNUM a, BN_MONT_CTX *mont,
0f113f3e MC	161	BN_CTX *ctx)
	162	{
	163	int retn = 0;
9b4eab50	164	#ifdef MONT_WORD
0f113f3e MC	165	BIGNUM *t;
	166
	167	BN_CTX_start(ctx);
71883868 AP	168	if ((t = BN_CTX_get(ctx)) && BN_copy(t, a)) {
	169	retn = bn_from_montgomery_word(ret, t, mont);
	170	bn_correct_top(ret);
	171	bn_check_top(ret);
	172	}
0f113f3e MC	173	BN_CTX_end(ctx);
	174	#else /* !MONT_WORD */
	175	BIGNUM t1, t2;
	176
	177	BN_CTX_start(ctx);
	178	t1 = BN_CTX_get(ctx);
	179	t2 = BN_CTX_get(ctx);
edea42c6	180	if (t2 == NULL)
0f113f3e MC	181	goto err;
	182
	183	if (!BN_copy(t1, a))
	184	goto err;
	185	BN_mask_bits(t1, mont->ri);
	186
	187	if (!BN_mul(t2, t1, &mont->Ni, ctx))
	188	goto err;
	189	BN_mask_bits(t2, mont->ri);
	190
	191	if (!BN_mul(t1, t2, &mont->N, ctx))
	192	goto err;
	193	if (!BN_add(t2, a, t1))
	194	goto err;
	195	if (!BN_rshift(ret, t2, mont->ri))
	196	goto err;
	197
	198	if (BN_ucmp(ret, &(mont->N)) >= 0) {
	199	if (!BN_usub(ret, ret, &(mont->N)))
	200	goto err;
	201	}
	202	retn = 1;
	203	bn_check_top(ret);
e93f9a32	204	err:
0f113f3e MC	205	BN_CTX_end(ctx);
0f113f3e MC	206	#endif /* MONT_WORD */
26a7d938	207	return retn;
0f113f3e	208	}
d02b48c6	209
71883868 AP	210	int bn_to_mont_fixed_top(BIGNUM r, const BIGNUM a, BN_MONT_CTX *mont,
	211	BN_CTX *ctx)
	212	{
	213	return bn_mul_mont_fixed_top(r, a, &(mont->RR), mont, ctx);
	214	}
	215
6b691a5c	216	BN_MONT_CTX *BN_MONT_CTX_new(void)
0f113f3e MC	217	{
0f113f3e MC	218	BN_MONT_CTX *ret;
d02b48c6	219
f06080cb F	220	if ((ret = OPENSSL_malloc(sizeof(*ret))) == NULL) {
f06080cb F	221	BNerr(BN_F_BN_MONT_CTX_NEW, ERR_R_MALLOC_FAILURE);
26a7d938	222	return NULL;
f06080cb	223	}
dfeab068	224
0f113f3e MC	225	BN_MONT_CTX_init(ret);
0f113f3e MC	226	ret->flags = BN_FLG_MALLOCED;
26a7d938	227	return ret;
0f113f3e	228	}
d02b48c6	229
6b691a5c	230	void BN_MONT_CTX_init(BN_MONT_CTX *ctx)
0f113f3e MC	231	{
0f113f3e MC	232	ctx->ri = 0;
e6e9170d RS	233	bn_init(&ctx->RR);
	234	bn_init(&ctx->N);
	235	bn_init(&ctx->Ni);
0f113f3e MC	236	ctx->n0[0] = ctx->n0[1] = 0;
	237	ctx->flags = 0;
	238	}
dfeab068	239
6b691a5c	240	void BN_MONT_CTX_free(BN_MONT_CTX *mont)
0f113f3e	241	{
e6e9170d RS	242	if (mont == NULL)
	243	return;
	244	BN_clear_free(&mont->RR);
	245	BN_clear_free(&mont->N);
	246	BN_clear_free(&mont->Ni);
0f113f3e MC	247	if (mont->flags & BN_FLG_MALLOCED)
	248	OPENSSL_free(mont);
	249	}
d02b48c6	250
84c15db5	251	int BN_MONT_CTX_set(BN_MONT_CTX mont, const BIGNUM mod, BN_CTX *ctx)
0f113f3e	252	{
71883868	253	int i, ret = 0;
0f113f3e MC	254	BIGNUM Ri, R;
0f113f3e MC	255
6a009812 MC	256	if (BN_is_zero(mod))
	257	return 0;
	258
0f113f3e MC	259	BN_CTX_start(ctx);
	260	if ((Ri = BN_CTX_get(ctx)) == NULL)
	261	goto err;
	262	R = &(mont->RR); /* grab RR as a temp */
	263	if (!BN_copy(&(mont->N), mod))
	264	goto err; /* Set N */
7d461736 MC	265	if (BN_get_flags(mod, BN_FLG_CONSTTIME) != 0)
7d461736 MC	266	BN_set_flags(&(mont->N), BN_FLG_CONSTTIME);
0f113f3e	267	mont->N.neg = 0;
dfeab068	268
6535eb17	269	#ifdef MONT_WORD
0f113f3e MC	270	{
	271	BIGNUM tmod;
	272	BN_ULONG buf[2];
	273
d59c7c81	274	bn_init(&tmod);
0f113f3e MC	275	tmod.d = buf;
	276	tmod.dmax = 2;
	277	tmod.neg = 0;
	278
3de81a59 SW	279	if (BN_get_flags(mod, BN_FLG_CONSTTIME) != 0)
	280	BN_set_flags(&tmod, BN_FLG_CONSTTIME);
	281
0f113f3e MC	282	mont->ri = (BN_num_bits(mod) + (BN_BITS2 - 1)) / BN_BITS2 * BN_BITS2;
	283
	284	# if defined(OPENSSL_BN_ASM_MONT) && (BN_BITS2<=32)
	285	/*
	286	* Only certain BN_BITS2<=32 platforms actually make use of n0[1],
	287	* and we could use the #else case (with a shorter R value) for the
	288	* others. However, currently only the assembler files do know which
	289	* is which.
	290	*/
	291
	292	BN_zero(R);
	293	if (!(BN_set_bit(R, 2 * BN_BITS2)))
	294	goto err;
	295
	296	tmod.top = 0;
	297	if ((buf[0] = mod->d[0]))
	298	tmod.top = 1;
	299	if ((buf[1] = mod->top > 1 ? mod->d[1] : 0))
	300	tmod.top = 2;
	301
b1860d6c MC	302	if (BN_is_one(&tmod))
	303	BN_zero(Ri);
	304	else if ((BN_mod_inverse(Ri, R, &tmod, ctx)) == NULL)
0f113f3e MC	305	goto err;
	306	if (!BN_lshift(Ri, Ri, 2 * BN_BITS2))
	307	goto err; /* RRi /
	308	if (!BN_is_zero(Ri)) {
	309	if (!BN_sub_word(Ri, 1))
	310	goto err;
	311	} else { /* if N mod word size == 1 */
	312
	313	if (bn_expand(Ri, (int)sizeof(BN_ULONG) * 2) == NULL)
	314	goto err;
	315	/* Ri-- (mod double word size) */
	316	Ri->neg = 0;
	317	Ri->d[0] = BN_MASK2;
	318	Ri->d[1] = BN_MASK2;
	319	Ri->top = 2;
	320	}
	321	if (!BN_div(Ri, NULL, Ri, &tmod, ctx))
	322	goto err;
	323	/*
	324	* Ni = (R*Ri-1)/N, keep only couple of least significant words:
	325	*/
	326	mont->n0[0] = (Ri->top > 0) ? Ri->d[0] : 0;
	327	mont->n0[1] = (Ri->top > 1) ? Ri->d[1] : 0;
	328	# else
	329	BN_zero(R);
	330	if (!(BN_set_bit(R, BN_BITS2)))
	331	goto err; /* R */
	332
	333	buf[0] = mod->d[0]; /* tmod = N mod word size */
	334	buf[1] = 0;
	335	tmod.top = buf[0] != 0 ? 1 : 0;
	336	/* Ri = R^-1 mod N */
b1860d6c MC	337	if (BN_is_one(&tmod))
	338	BN_zero(Ri);
	339	else if ((BN_mod_inverse(Ri, R, &tmod, ctx)) == NULL)
0f113f3e MC	340	goto err;
	341	if (!BN_lshift(Ri, Ri, BN_BITS2))
	342	goto err; /* RRi /
	343	if (!BN_is_zero(Ri)) {
	344	if (!BN_sub_word(Ri, 1))
	345	goto err;
	346	} else { /* if N mod word size == 1 */
	347
	348	if (!BN_set_word(Ri, BN_MASK2))
	349	goto err; /* Ri-- (mod word size) */
	350	}
	351	if (!BN_div(Ri, NULL, Ri, &tmod, ctx))
	352	goto err;
	353	/*
	354	* Ni = (R*Ri-1)/N, keep only least significant word:
	355	*/
	356	mont->n0[0] = (Ri->top > 0) ? Ri->d[0] : 0;
	357	mont->n0[1] = 0;
	358	# endif
	359	}
	360	#else /* !MONT_WORD */
	361	{ /* bignum version */
	362	mont->ri = BN_num_bits(&mont->N);
	363	BN_zero(R);
	364	if (!BN_set_bit(R, mont->ri))
	365	goto err; /* R = 2^ri */
	366	/* Ri = R^-1 mod N */
	367	if ((BN_mod_inverse(Ri, R, &mont->N, ctx)) == NULL)
	368	goto err;
	369	if (!BN_lshift(Ri, Ri, mont->ri))
	370	goto err; /* RRi /
	371	if (!BN_sub_word(Ri, 1))
	372	goto err;
	373	/*
	374	* Ni = (R*Ri-1) / N
	375	*/
	376	if (!BN_div(&(mont->Ni), NULL, Ri, &mont->N, ctx))
	377	goto err;
	378	}
d02b48c6 RE	379	#endif
d02b48c6 RE	380
0f113f3e MC	381	/* setup RR for conversions */
	382	BN_zero(&(mont->RR));
	383	if (!BN_set_bit(&(mont->RR), mont->ri * 2))
	384	goto err;
	385	if (!BN_mod(&(mont->RR), &(mont->RR), &(mont->N), ctx))
	386	goto err;
d02b48c6	387
71883868 AP	388	for (i = mont->RR.top, ret = mont->N.top; i < ret; i++)
	389	mont->RR.d[i] = 0;
	390	mont->RR.top = ret;
	391	mont->RR.flags \|= BN_FLG_FIXED_TOP;
	392
0f113f3e MC	393	ret = 1;
	394	err:
	395	BN_CTX_end(ctx);
	396	return ret;
	397	}
d02b48c6	398
6b691a5c	399	BN_MONT_CTX BN_MONT_CTX_copy(BN_MONT_CTX to, BN_MONT_CTX *from)
0f113f3e MC	400	{
0f113f3e MC	401	if (to == from)
26a7d938	402	return to;
0f113f3e MC	403
	404	if (!BN_copy(&(to->RR), &(from->RR)))
	405	return NULL;
	406	if (!BN_copy(&(to->N), &(from->N)))
	407	return NULL;
	408	if (!BN_copy(&(to->Ni), &(from->Ni)))
	409	return NULL;
	410	to->ri = from->ri;
	411	to->n0[0] = from->n0[0];
	412	to->n0[1] = from->n0[1];
26a7d938	413	return to;
0f113f3e	414	}
dfeab068	415
d188a536	416	BN_MONT_CTX BN_MONT_CTX_set_locked(BN_MONT_CTX pmont, CRYPTO_RWLOCK lock,
0f113f3e MC	417	const BIGNUM mod, BN_CTX ctx)
	418	{
	419	BN_MONT_CTX *ret;
	420
d188a536	421	CRYPTO_THREAD_read_lock(lock);
0f113f3e	422	ret = *pmont;
d188a536	423	CRYPTO_THREAD_unlock(lock);
0f113f3e MC	424	if (ret)
	425	return ret;
	426
	427	/*
	428	* We don't want to serialise globally while doing our lazy-init math in
	429	* BN_MONT_CTX_set. That punishes threads that are doing independent
	430	* things. Instead, punish the case where more than one thread tries to
	431	* lazy-init the same 'pmont', by having each do the lazy-init math work
	432	* independently and only use the one from the thread that wins the race
	433	* (the losers throw away the work they've done).
	434	*/
	435	ret = BN_MONT_CTX_new();
90945fa3	436	if (ret == NULL)
0f113f3e MC	437	return NULL;
	438	if (!BN_MONT_CTX_set(ret, mod, ctx)) {
	439	BN_MONT_CTX_free(ret);
	440	return NULL;
	441	}
	442
	443	/* The locked compare-and-set, after the local work is done. */
d188a536	444	CRYPTO_THREAD_write_lock(lock);
0f113f3e MC	445	if (*pmont) {
	446	BN_MONT_CTX_free(ret);
	447	ret = *pmont;
	448	} else
	449	*pmont = ret;
d188a536	450	CRYPTO_THREAD_unlock(lock);
0f113f3e MC	451	return ret;
0f113f3e MC	452	}