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

/*
 * Copyright 1995-2021 The OpenSSL Project Authors. All Rights Reserved.
 *
 * Licensed under the Apache License 2.0 (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_local.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_fixed_top(tmp, a, ctx))
            goto err;
    } else {
        if (!bn_mul_fixed_top(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;
    unsigned int rtop;

    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 */
    for (rtop = r->top, i = 0; i < max; i++) {
        v = (BN_ULONG)0 - ((i - rtop) >> (8 * sizeof(rtop) - 1));
        rp[i] &= v;
    }

    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;

    retn = bn_from_mont_fixed_top(ret, a, mont, ctx);
    bn_correct_top(ret);
    bn_check_top(ret);

    return retn;
}

int bn_from_mont_fixed_top(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;
}

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) {
        ERR_raise(ERR_LIB_BN, 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;

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

    /*
     * We don't want to serialize 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. */
    if (!CRYPTO_THREAD_write_lock(lock)) {
        BN_MONT_CTX_free(ret);
        return NULL;
    }

    if (*pmont) {
        BN_MONT_CTX_free(ret);
        ret = *pmont;
    } else
        *pmont = ret;
    CRYPTO_THREAD_unlock(lock);
    return ret;
}
Commit	Line	Data
4f22f405	1	/*
3c2bdd7d	2	* Copyright 1995-2021 The OpenSSL Project Authors. All Rights Reserved.
675f605d	3	*
367ace68	4	* Licensed under the Apache License 2.0 (the "License"). You may not use
4f22f405 RS	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"
706457b7	18	#include "bn_local.h"
d02b48c6	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) {
fcc4ee09	67	if (!bn_sqr_fixed_top(tmp, a, ctx))
0f113f3e MC	68	goto err;
0f113f3e MC	69	} else {
fcc4ee09	70	if (!bn_mul_fixed_top(tmp, a, b, ctx))
0f113f3e MC	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;
fcc4ee09	93	unsigned int rtop;
0f113f3e MC	94
	95	n = &(mont->N);
	96	nl = n->top;
	97	if (nl == 0) {
	98	ret->top = 0;
208fb891	99	return 1;
0f113f3e MC	100	}
	101
	102	max = (2 * nl); /* carry is stored separately */
	103	if (bn_wexpand(r, max) == NULL)
26a7d938	104	return 0;
0f113f3e MC	105
	106	r->neg ^= n->neg;
	107	np = n->d;
	108	rp = r->d;
	109
	110	/* clear the top words of T */
fcc4ee09 AP	111	for (rtop = r->top, i = 0; i < max; i++) {
	112	v = (BN_ULONG)0 - ((i - rtop) >> (8 * sizeof(rtop) - 1));
	113	rp[i] &= v;
	114	}
0f113f3e MC	115
0f113f3e MC	116	r->top = max;
71883868	117	r->flags \|= BN_FLG_FIXED_TOP;
0f113f3e MC	118	n0 = mont->n0[0];
0f113f3e MC	119
f345b1f3 DB	120	/*
	121	* Add multiples of \|n\| to \|r\| until R = 2^(nl * BN_BITS2) divides it. On
	122	* input, we had \|r\| < \|n\| * R, so now \|r\| < 2 * \|n\| * R. Note that \|r\|
	123	* includes \|carry\| which is stored separately.
	124	*/
0f113f3e MC	125	for (carry = 0, i = 0; i < nl; i++, rp++) {
	126	v = bn_mul_add_words(rp, np, nl, (rp[0] * n0) & BN_MASK2);
	127	v = (v + carry + rp[nl]) & BN_MASK2;
	128	carry \|= (v != rp[nl]);
	129	carry &= (v <= rp[nl]);
	130	rp[nl] = v;
	131	}
	132
	133	if (bn_wexpand(ret, nl) == NULL)
26a7d938	134	return 0;
0f113f3e	135	ret->top = nl;
71883868	136	ret->flags \|= BN_FLG_FIXED_TOP;
0f113f3e MC	137	ret->neg = r->neg;
	138
	139	rp = ret->d;
f345b1f3 DB	140
	141	/*
	142	* Shift \|nl\| words to divide by R. We have \|ap\| < 2 * \|n\|. Note that \|ap\|
	143	* includes \|carry\| which is stored separately.
	144	*/
0f113f3e MC	145	ap = &(r->d[nl]);
0f113f3e MC	146
6c90182a	147	carry -= bn_sub_words(rp, ap, np, nl);
f345b1f3	148	/*
6c90182a AP	149	* \|carry\| is -1 if \|ap\| - \|np\| underflowed or zero if it did not. Note
	150	* \|carry\| cannot be 1. That would imply the subtraction did not fit in
	151	* \|nl\| words, and we know at most one subtraction is needed.
f345b1f3	152	*/
f345b1f3	153	for (i = 0; i < nl; i++) {
6c90182a	154	rp[i] = (carry & ap[i]) \| (~carry & rp[i]);
f345b1f3	155	ap[i] = 0;
0f113f3e	156	}
0f113f3e	157
208fb891	158	return 1;
0f113f3e MC	159	}
0f113f3e MC	160	#endif /* MONT_WORD */
9b4eab50 AP	161
9b4eab50 AP	162	int BN_from_montgomery(BIGNUM ret, const BIGNUM a, BN_MONT_CTX *mont,
0f113f3e	163	BN_CTX *ctx)
fcc4ee09 AP	164	{
	165	int retn;
	166
	167	retn = bn_from_mont_fixed_top(ret, a, mont, ctx);
	168	bn_correct_top(ret);
	169	bn_check_top(ret);
	170
	171	return retn;
	172	}
	173
	174	int bn_from_mont_fixed_top(BIGNUM ret, const BIGNUM a, BN_MONT_CTX *mont,
	175	BN_CTX *ctx)
0f113f3e MC	176	{
0f113f3e MC	177	int retn = 0;
9b4eab50	178	#ifdef MONT_WORD
0f113f3e MC	179	BIGNUM *t;
	180
	181	BN_CTX_start(ctx);
71883868 AP	182	if ((t = BN_CTX_get(ctx)) && BN_copy(t, a)) {
71883868 AP	183	retn = bn_from_montgomery_word(ret, t, mont);
71883868	184	}
0f113f3e MC	185	BN_CTX_end(ctx);
	186	#else /* !MONT_WORD */
	187	BIGNUM t1, t2;
	188
	189	BN_CTX_start(ctx);
	190	t1 = BN_CTX_get(ctx);
	191	t2 = BN_CTX_get(ctx);
edea42c6	192	if (t2 == NULL)
0f113f3e MC	193	goto err;
	194
	195	if (!BN_copy(t1, a))
	196	goto err;
	197	BN_mask_bits(t1, mont->ri);
	198
	199	if (!BN_mul(t2, t1, &mont->Ni, ctx))
	200	goto err;
	201	BN_mask_bits(t2, mont->ri);
	202
	203	if (!BN_mul(t1, t2, &mont->N, ctx))
	204	goto err;
	205	if (!BN_add(t2, a, t1))
	206	goto err;
	207	if (!BN_rshift(ret, t2, mont->ri))
	208	goto err;
	209
	210	if (BN_ucmp(ret, &(mont->N)) >= 0) {
	211	if (!BN_usub(ret, ret, &(mont->N)))
	212	goto err;
	213	}
	214	retn = 1;
	215	bn_check_top(ret);
e93f9a32	216	err:
0f113f3e MC	217	BN_CTX_end(ctx);
0f113f3e MC	218	#endif /* MONT_WORD */
26a7d938	219	return retn;
0f113f3e	220	}
d02b48c6	221
71883868 AP	222	int bn_to_mont_fixed_top(BIGNUM r, const BIGNUM a, BN_MONT_CTX *mont,
	223	BN_CTX *ctx)
	224	{
	225	return bn_mul_mont_fixed_top(r, a, &(mont->RR), mont, ctx);
	226	}
	227
6b691a5c	228	BN_MONT_CTX *BN_MONT_CTX_new(void)
0f113f3e MC	229	{
0f113f3e MC	230	BN_MONT_CTX *ret;
d02b48c6	231
f06080cb	232	if ((ret = OPENSSL_malloc(sizeof(*ret))) == NULL) {
9311d0c4	233	ERR_raise(ERR_LIB_BN, ERR_R_MALLOC_FAILURE);
26a7d938	234	return NULL;
f06080cb	235	}
dfeab068	236
0f113f3e MC	237	BN_MONT_CTX_init(ret);
0f113f3e MC	238	ret->flags = BN_FLG_MALLOCED;
26a7d938	239	return ret;
0f113f3e	240	}
d02b48c6	241
6b691a5c	242	void BN_MONT_CTX_init(BN_MONT_CTX *ctx)
0f113f3e MC	243	{
0f113f3e MC	244	ctx->ri = 0;
e6e9170d RS	245	bn_init(&ctx->RR);
	246	bn_init(&ctx->N);
	247	bn_init(&ctx->Ni);
0f113f3e MC	248	ctx->n0[0] = ctx->n0[1] = 0;
	249	ctx->flags = 0;
	250	}
dfeab068	251
6b691a5c	252	void BN_MONT_CTX_free(BN_MONT_CTX *mont)
0f113f3e	253	{
e6e9170d RS	254	if (mont == NULL)
	255	return;
	256	BN_clear_free(&mont->RR);
	257	BN_clear_free(&mont->N);
	258	BN_clear_free(&mont->Ni);
0f113f3e MC	259	if (mont->flags & BN_FLG_MALLOCED)
	260	OPENSSL_free(mont);
	261	}
d02b48c6	262
84c15db5	263	int BN_MONT_CTX_set(BN_MONT_CTX mont, const BIGNUM mod, BN_CTX *ctx)
0f113f3e	264	{
71883868	265	int i, ret = 0;
0f113f3e MC	266	BIGNUM Ri, R;
0f113f3e MC	267
6a009812 MC	268	if (BN_is_zero(mod))
	269	return 0;
	270
0f113f3e MC	271	BN_CTX_start(ctx);
	272	if ((Ri = BN_CTX_get(ctx)) == NULL)
	273	goto err;
	274	R = &(mont->RR); /* grab RR as a temp */
	275	if (!BN_copy(&(mont->N), mod))
	276	goto err; /* Set N */
7d461736 MC	277	if (BN_get_flags(mod, BN_FLG_CONSTTIME) != 0)
7d461736 MC	278	BN_set_flags(&(mont->N), BN_FLG_CONSTTIME);
0f113f3e	279	mont->N.neg = 0;
dfeab068	280
6535eb17	281	#ifdef MONT_WORD
0f113f3e MC	282	{
	283	BIGNUM tmod;
	284	BN_ULONG buf[2];
	285
d59c7c81	286	bn_init(&tmod);
0f113f3e MC	287	tmod.d = buf;
	288	tmod.dmax = 2;
	289	tmod.neg = 0;
	290
3de81a59 SW	291	if (BN_get_flags(mod, BN_FLG_CONSTTIME) != 0)
	292	BN_set_flags(&tmod, BN_FLG_CONSTTIME);
	293
0f113f3e MC	294	mont->ri = (BN_num_bits(mod) + (BN_BITS2 - 1)) / BN_BITS2 * BN_BITS2;
	295
	296	# if defined(OPENSSL_BN_ASM_MONT) && (BN_BITS2<=32)
	297	/*
	298	* Only certain BN_BITS2<=32 platforms actually make use of n0[1],
	299	* and we could use the #else case (with a shorter R value) for the
	300	* others. However, currently only the assembler files do know which
	301	* is which.
	302	*/
	303
	304	BN_zero(R);
	305	if (!(BN_set_bit(R, 2 * BN_BITS2)))
	306	goto err;
	307
	308	tmod.top = 0;
	309	if ((buf[0] = mod->d[0]))
	310	tmod.top = 1;
	311	if ((buf[1] = mod->top > 1 ? mod->d[1] : 0))
	312	tmod.top = 2;
	313
b1860d6c MC	314	if (BN_is_one(&tmod))
	315	BN_zero(Ri);
	316	else if ((BN_mod_inverse(Ri, R, &tmod, ctx)) == NULL)
0f113f3e MC	317	goto err;
	318	if (!BN_lshift(Ri, Ri, 2 * BN_BITS2))
	319	goto err; /* RRi /
	320	if (!BN_is_zero(Ri)) {
	321	if (!BN_sub_word(Ri, 1))
	322	goto err;
	323	} else { /* if N mod word size == 1 */
	324
	325	if (bn_expand(Ri, (int)sizeof(BN_ULONG) * 2) == NULL)
	326	goto err;
	327	/* Ri-- (mod double word size) */
	328	Ri->neg = 0;
	329	Ri->d[0] = BN_MASK2;
	330	Ri->d[1] = BN_MASK2;
	331	Ri->top = 2;
	332	}
	333	if (!BN_div(Ri, NULL, Ri, &tmod, ctx))
	334	goto err;
	335	/*
	336	* Ni = (R*Ri-1)/N, keep only couple of least significant words:
	337	*/
	338	mont->n0[0] = (Ri->top > 0) ? Ri->d[0] : 0;
	339	mont->n0[1] = (Ri->top > 1) ? Ri->d[1] : 0;
	340	# else
	341	BN_zero(R);
	342	if (!(BN_set_bit(R, BN_BITS2)))
	343	goto err; /* R */
	344
	345	buf[0] = mod->d[0]; /* tmod = N mod word size */
	346	buf[1] = 0;
	347	tmod.top = buf[0] != 0 ? 1 : 0;
	348	/* Ri = R^-1 mod N */
b1860d6c MC	349	if (BN_is_one(&tmod))
	350	BN_zero(Ri);
	351	else if ((BN_mod_inverse(Ri, R, &tmod, ctx)) == NULL)
0f113f3e MC	352	goto err;
	353	if (!BN_lshift(Ri, Ri, BN_BITS2))
	354	goto err; /* RRi /
	355	if (!BN_is_zero(Ri)) {
	356	if (!BN_sub_word(Ri, 1))
	357	goto err;
	358	} else { /* if N mod word size == 1 */
	359
	360	if (!BN_set_word(Ri, BN_MASK2))
	361	goto err; /* Ri-- (mod word size) */
	362	}
	363	if (!BN_div(Ri, NULL, Ri, &tmod, ctx))
	364	goto err;
	365	/*
	366	* Ni = (R*Ri-1)/N, keep only least significant word:
	367	*/
	368	mont->n0[0] = (Ri->top > 0) ? Ri->d[0] : 0;
	369	mont->n0[1] = 0;
	370	# endif
	371	}
	372	#else /* !MONT_WORD */
	373	{ /* bignum version */
	374	mont->ri = BN_num_bits(&mont->N);
	375	BN_zero(R);
	376	if (!BN_set_bit(R, mont->ri))
	377	goto err; /* R = 2^ri */
	378	/* Ri = R^-1 mod N */
	379	if ((BN_mod_inverse(Ri, R, &mont->N, ctx)) == NULL)
	380	goto err;
	381	if (!BN_lshift(Ri, Ri, mont->ri))
	382	goto err; /* RRi /
	383	if (!BN_sub_word(Ri, 1))
	384	goto err;
	385	/*
	386	* Ni = (R*Ri-1) / N
	387	*/
	388	if (!BN_div(&(mont->Ni), NULL, Ri, &mont->N, ctx))
	389	goto err;
	390	}
d02b48c6 RE	391	#endif
d02b48c6 RE	392
0f113f3e MC	393	/* setup RR for conversions */
	394	BN_zero(&(mont->RR));
	395	if (!BN_set_bit(&(mont->RR), mont->ri * 2))
	396	goto err;
	397	if (!BN_mod(&(mont->RR), &(mont->RR), &(mont->N), ctx))
	398	goto err;
d02b48c6	399
71883868 AP	400	for (i = mont->RR.top, ret = mont->N.top; i < ret; i++)
	401	mont->RR.d[i] = 0;
	402	mont->RR.top = ret;
	403	mont->RR.flags \|= BN_FLG_FIXED_TOP;
	404
0f113f3e MC	405	ret = 1;
	406	err:
	407	BN_CTX_end(ctx);
	408	return ret;
	409	}
d02b48c6	410
6b691a5c	411	BN_MONT_CTX BN_MONT_CTX_copy(BN_MONT_CTX to, BN_MONT_CTX *from)
0f113f3e MC	412	{
0f113f3e MC	413	if (to == from)
26a7d938	414	return to;
0f113f3e MC	415
	416	if (!BN_copy(&(to->RR), &(from->RR)))
	417	return NULL;
	418	if (!BN_copy(&(to->N), &(from->N)))
	419	return NULL;
	420	if (!BN_copy(&(to->Ni), &(from->Ni)))
	421	return NULL;
	422	to->ri = from->ri;
	423	to->n0[0] = from->n0[0];
	424	to->n0[1] = from->n0[1];
26a7d938	425	return to;
0f113f3e	426	}
dfeab068	427
d188a536	428	BN_MONT_CTX BN_MONT_CTX_set_locked(BN_MONT_CTX pmont, CRYPTO_RWLOCK lock,
0f113f3e MC	429	const BIGNUM mod, BN_CTX ctx)
	430	{
	431	BN_MONT_CTX *ret;
	432
cd3f8c1b RS	433	if (!CRYPTO_THREAD_read_lock(lock))
cd3f8c1b RS	434	return NULL;
0f113f3e	435	ret = *pmont;
d188a536	436	CRYPTO_THREAD_unlock(lock);
0f113f3e MC	437	if (ret)
	438	return ret;
	439
	440	/*
aa97970c	441	* We don't want to serialize globally while doing our lazy-init math in
0f113f3e MC	442	* BN_MONT_CTX_set. That punishes threads that are doing independent
	443	* things. Instead, punish the case where more than one thread tries to
	444	* lazy-init the same 'pmont', by having each do the lazy-init math work
	445	* independently and only use the one from the thread that wins the race
	446	* (the losers throw away the work they've done).
	447	*/
	448	ret = BN_MONT_CTX_new();
90945fa3	449	if (ret == NULL)
0f113f3e MC	450	return NULL;
	451	if (!BN_MONT_CTX_set(ret, mod, ctx)) {
	452	BN_MONT_CTX_free(ret);
	453	return NULL;
	454	}
	455
	456	/* The locked compare-and-set, after the local work is done. */
cd3f8c1b RS	457	if (!CRYPTO_THREAD_write_lock(lock)) {
	458	BN_MONT_CTX_free(ret);
	459	return NULL;
	460	}
	461
0f113f3e MC	462	if (*pmont) {
	463	BN_MONT_CTX_free(ret);
	464	ret = *pmont;
	465	} else
	466	*pmont = ret;
d188a536	467	CRYPTO_THREAD_unlock(lock);
0f113f3e MC	468	return ret;
0f113f3e MC	469	}