[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)
{
    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;
            bn_correct_top(r);
            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
    bn_check_top(r);
    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;
    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->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;
    }
    bn_correct_top(r);
    bn_correct_top(ret);
    bn_check_top(ret);

    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_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;
}

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 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;

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