/src/openssl30/crypto/bn/bn_mont.c

Source (jump to first uncovered line)
/*
 * Copyright 1995-2023 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 && num <= BN_SOFT_LIMIT && 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;
}

Coverage Report

Created: 2023-09-25 06:45

Line	Count	Source (jump to first uncovered line)
1		/*
2		* Copyright 1995-2023 The OpenSSL Project Authors. All Rights Reserved.
3		*
4		* Licensed under the Apache License 2.0 (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
8		*/
9
10		/*
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
15		*/
16
17		#include "internal/cryptlib.h"
18		#include "bn_local.h"
19
20		#define MONT_WORD /* use the faster word-based algorithm */
21
22		#ifdef MONT_WORD
23		static int bn_from_montgomery_word(BIGNUM ret, BIGNUM r, BN_MONT_CTX *mont);
24		#endif
25
26		int BN_mod_mul_montgomery(BIGNUM r, const BIGNUM a, const BIGNUM *b,
27		BN_MONT_CTX mont, BN_CTX ctx)
28	27.2M	{
29	27.2M	int ret = bn_mul_mont_fixed_top(r, a, b, mont, ctx);
30
31	27.2M	bn_correct_top(r);
32	27.2M	bn_check_top(r);
33
34	27.2M	return ret;
35	27.2M	}
36
37		int bn_mul_mont_fixed_top(BIGNUM r, const BIGNUM a, const BIGNUM *b,
38		BN_MONT_CTX mont, BN_CTX ctx)
39	58.2M	{
40	58.2M	BIGNUM *tmp;
41	58.2M	int ret = 0;
42	58.2M	int num = mont->N.top;
43
44	58.2M	#if defined(OPENSSL_BN_ASM_MONT) && defined(MONT_WORD)
45	58.2M	if (num > 1 && num <= BN_SOFT_LIMIT && a->top == num && b->top == num) {
46	55.3M	if (bn_wexpand(r, num) == NULL)
47	0	return 0;
48	55.3M	if (bn_mul_mont(r->d, a->d, b->d, mont->N.d, mont->n0, num)) {
49	55.3M	r->neg = a->neg ^ b->neg;
50	55.3M	r->top = num;
51	55.3M	r->flags \|= BN_FLG_FIXED_TOP;
52	55.3M	return 1;
53	55.3M	}
54	55.3M	}
55	2.90M	#endif
56
57	2.90M	if ((a->top + b->top) > 2 * num)
58	0	return 0;
59
60	2.90M	BN_CTX_start(ctx);
61	2.90M	tmp = BN_CTX_get(ctx);
62	2.90M	if (tmp == NULL)
63	0	goto err;
64
65	2.90M	bn_check_top(tmp);
66	2.90M	if (a == b) {
67	1.23M	if (!bn_sqr_fixed_top(tmp, a, ctx))
68	0	goto err;
69	1.66M	} else {
70	1.66M	if (!bn_mul_fixed_top(tmp, a, b, ctx))
71	0	goto err;
72	1.66M	}
73		/* reduce from aRR to aR */
74	2.90M	#ifdef MONT_WORD
75	2.90M	if (!bn_from_montgomery_word(r, tmp, mont))
76	0	goto err;
77		#else
78		if (!BN_from_montgomery(r, tmp, mont, ctx))
79		goto err;
80		#endif
81	2.90M	ret = 1;
82	2.90M	err:
83	2.90M	BN_CTX_end(ctx);
84	2.90M	return ret;
85	2.90M	}
86
87		#ifdef MONT_WORD
88		static int bn_from_montgomery_word(BIGNUM ret, BIGNUM r, BN_MONT_CTX *mont)
89	3.34M	{
90	3.34M	BIGNUM *n;
91	3.34M	BN_ULONG ap, np, *rp, n0, v, carry;
92	3.34M	int nl, max, i;
93	3.34M	unsigned int rtop;
94
95	3.34M	n = &(mont->N);
96	3.34M	nl = n->top;
97	3.34M	if (nl == 0) {
98	0	ret->top = 0;
99	0	return 1;
100	0	}
101
102	3.34M	max = (2 * nl); /* carry is stored separately */
103	3.34M	if (bn_wexpand(r, max) == NULL)
104	0	return 0;
105
106	3.34M	r->neg ^= n->neg;
107	3.34M	np = n->d;
108	3.34M	rp = r->d;
109
110		/* clear the top words of T */
111	80.3M	for (rtop = r->top, i = 0; i < max; i++) {
112	77.0M	v = (BN_ULONG)0 - ((i - rtop) >> (8 * sizeof(rtop) - 1));
113	77.0M	rp[i] &= v;
114	77.0M	}
115
116	3.34M	r->top = max;
117	3.34M	r->flags \|= BN_FLG_FIXED_TOP;
118	3.34M	n0 = mont->n0[0];
119
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		*/
125	41.8M	for (carry = 0, i = 0; i < nl; i++, rp++) {
126	38.5M	v = bn_mul_add_words(rp, np, nl, (rp[0] * n0) & BN_MASK2);
127	38.5M	v = (v + carry + rp[nl]) & BN_MASK2;
128	38.5M	carry \|= (v != rp[nl]);
129	38.5M	carry &= (v <= rp[nl]);
130	38.5M	rp[nl] = v;
131	38.5M	}
132
133	3.34M	if (bn_wexpand(ret, nl) == NULL)
134	0	return 0;
135	3.34M	ret->top = nl;
136	3.34M	ret->flags \|= BN_FLG_FIXED_TOP;
137	3.34M	ret->neg = r->neg;
138
139	3.34M	rp = ret->d;
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		*/
145	3.34M	ap = &(r->d[nl]);
146
147	3.34M	carry -= bn_sub_words(rp, ap, np, nl);
148		/*
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.
152		*/
153	41.8M	for (i = 0; i < nl; i++) {
154	38.5M	rp[i] = (carry & ap[i]) \| (~carry & rp[i]);
155	38.5M	ap[i] = 0;
156	38.5M	}
157
158	3.34M	return 1;
159	3.34M	}
160		#endif /* MONT_WORD */
161
162		int BN_from_montgomery(BIGNUM ret, const BIGNUM a, BN_MONT_CTX *mont,
163		BN_CTX *ctx)
164	432k	{
165	432k	int retn;
166
167	432k	retn = bn_from_mont_fixed_top(ret, a, mont, ctx);
168	432k	bn_correct_top(ret);
169	432k	bn_check_top(ret);
170
171	432k	return retn;
172	432k	}
173
174		int bn_from_mont_fixed_top(BIGNUM ret, const BIGNUM a, BN_MONT_CTX *mont,
175		BN_CTX *ctx)
176	442k	{
177	442k	int retn = 0;
178	442k	#ifdef MONT_WORD
179	442k	BIGNUM *t;
180
181	442k	BN_CTX_start(ctx);
182	442k	if ((t = BN_CTX_get(ctx)) && BN_copy(t, a)) {
183	442k	retn = bn_from_montgomery_word(ret, t, mont);
184	442k	}
185	442k	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);
192		if (t2 == NULL)
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);
216		err:
217		BN_CTX_end(ctx);
218		#endif /* MONT_WORD */
219	442k	return retn;
220	442k	}
221
222		int bn_to_mont_fixed_top(BIGNUM r, const BIGNUM a, BN_MONT_CTX *mont,
223		BN_CTX *ctx)
224	260k	{
225	260k	return bn_mul_mont_fixed_top(r, a, &(mont->RR), mont, ctx);
226	260k	}
227
228		BN_MONT_CTX *BN_MONT_CTX_new(void)
229	868k	{
230	868k	BN_MONT_CTX *ret;
231
232	868k	if ((ret = OPENSSL_malloc(sizeof(*ret))) == NULL) {
233	0	ERR_raise(ERR_LIB_BN, ERR_R_MALLOC_FAILURE);
234	0	return NULL;
235	0	}
236
237	868k	BN_MONT_CTX_init(ret);
238	868k	ret->flags = BN_FLG_MALLOCED;
239	868k	return ret;
240	868k	}
241
242		void BN_MONT_CTX_init(BN_MONT_CTX *ctx)
243	868k	{
244	868k	ctx->ri = 0;
245	868k	bn_init(&ctx->RR);
246	868k	bn_init(&ctx->N);
247	868k	bn_init(&ctx->Ni);
248	868k	ctx->n0[0] = ctx->n0[1] = 0;
249	868k	ctx->flags = 0;
250	868k	}
251
252		void BN_MONT_CTX_free(BN_MONT_CTX *mont)
253	2.77M	{
254	2.77M	if (mont == NULL)
255	1.90M	return;
256	868k	BN_clear_free(&mont->RR);
257	868k	BN_clear_free(&mont->N);
258	868k	BN_clear_free(&mont->Ni);
259	868k	if (mont->flags & BN_FLG_MALLOCED)
260	868k	OPENSSL_free(mont);
261	868k	}
262
263		int BN_MONT_CTX_set(BN_MONT_CTX mont, const BIGNUM mod, BN_CTX *ctx)
264	523k	{
265	523k	int i, ret = 0;
266	523k	BIGNUM Ri, R;
267
268	523k	if (BN_is_zero(mod))
269	3	return 0;
270
271	523k	BN_CTX_start(ctx);
272	523k	if ((Ri = BN_CTX_get(ctx)) == NULL)
273	0	goto err;
274	523k	R = &(mont->RR); /* grab RR as a temp */
275	523k	if (!BN_copy(&(mont->N), mod))
276	0	goto err; /* Set N */
277	523k	if (BN_get_flags(mod, BN_FLG_CONSTTIME) != 0)
278	10.4k	BN_set_flags(&(mont->N), BN_FLG_CONSTTIME);
279	523k	mont->N.neg = 0;
280
281	523k	#ifdef MONT_WORD
282	523k	{
283	523k	BIGNUM tmod;
284	523k	BN_ULONG buf[2];
285
286	523k	bn_init(&tmod);
287	523k	tmod.d = buf;
288	523k	tmod.dmax = 2;
289	523k	tmod.neg = 0;
290
291	523k	if (BN_get_flags(mod, BN_FLG_CONSTTIME) != 0)
292	10.4k	BN_set_flags(&tmod, BN_FLG_CONSTTIME);
293
294	523k	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
314		if (BN_is_one(&tmod))
315		BN_zero(Ri);
316		else if ((BN_mod_inverse(Ri, R, &tmod, ctx)) == NULL)
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	523k	BN_zero(R);
342	523k	if (!(BN_set_bit(R, BN_BITS2)))
343	0	goto err; /* R */
344
345	523k	buf[0] = mod->d[0]; /* tmod = N mod word size */
346	523k	buf[1] = 0;
347	523k	tmod.top = buf[0] != 0 ? 1 : 0;
348		/* Ri = R^-1 mod N */
349	523k	if (BN_is_one(&tmod))
350	57.7k	BN_zero(Ri);
351	465k	else if ((BN_mod_inverse(Ri, R, &tmod, ctx)) == NULL)
352	3.00k	goto err;
353	520k	if (!BN_lshift(Ri, Ri, BN_BITS2))
354	0	goto err; /* RRi /
355	520k	if (!BN_is_zero(Ri)) {
356	462k	if (!BN_sub_word(Ri, 1))
357	0	goto err;
358	462k	} else { /* if N mod word size == 1 */
359
360	57.7k	if (!BN_set_word(Ri, BN_MASK2))
361	0	goto err; /* Ri-- (mod word size) */
362	57.7k	}
363	520k	if (!BN_div(Ri, NULL, Ri, &tmod, ctx))
364	0	goto err;
365		/*
366		* Ni = (R*Ri-1)/N, keep only least significant word:
367		*/
368	520k	mont->n0[0] = (Ri->top > 0) ? Ri->d[0] : 0;
369	520k	mont->n0[1] = 0;
370	520k	# endif
371	520k	}
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		}
391		#endif
392
393		/* setup RR for conversions */
394	520k	BN_zero(&(mont->RR));
395	520k	if (!BN_set_bit(&(mont->RR), mont->ri * 2))
396	0	goto err;
397	520k	if (!BN_mod(&(mont->RR), &(mont->RR), &(mont->N), ctx))
398	0	goto err;
399
400	670k	for (i = mont->RR.top, ret = mont->N.top; i < ret; i++)
401	149k	mont->RR.d[i] = 0;
402	520k	mont->RR.top = ret;
403	520k	mont->RR.flags \|= BN_FLG_FIXED_TOP;
404
405	520k	ret = 1;
406	523k	err:
407	523k	BN_CTX_end(ctx);
408	523k	return ret;
409	520k	}
410
411		BN_MONT_CTX BN_MONT_CTX_copy(BN_MONT_CTX to, BN_MONT_CTX *from)
412	344k	{
413	344k	if (to == from)
414	0	return to;
415
416	344k	if (!BN_copy(&(to->RR), &(from->RR)))
417	0	return NULL;
418	344k	if (!BN_copy(&(to->N), &(from->N)))
419	0	return NULL;
420	344k	if (!BN_copy(&(to->Ni), &(from->Ni)))
421	0	return NULL;
422	344k	to->ri = from->ri;
423	344k	to->n0[0] = from->n0[0];
424	344k	to->n0[1] = from->n0[1];
425	344k	return to;
426	344k	}
427
428		BN_MONT_CTX BN_MONT_CTX_set_locked(BN_MONT_CTX pmont, CRYPTO_RWLOCK lock,
429		const BIGNUM mod, BN_CTX ctx)
430	20.0k	{
431	20.0k	BN_MONT_CTX *ret;
432
433	20.0k	if (!CRYPTO_THREAD_read_lock(lock))
434	0	return NULL;
435	20.0k	ret = *pmont;
436	20.0k	CRYPTO_THREAD_unlock(lock);
437	20.0k	if (ret)
438	4.64k	return ret;
439
440		/*
441		* We don't want to serialize globally while doing our lazy-init math in
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	15.3k	ret = BN_MONT_CTX_new();
449	15.3k	if (ret == NULL)
450	0	return NULL;
451	15.3k	if (!BN_MONT_CTX_set(ret, mod, ctx)) {
452	460	BN_MONT_CTX_free(ret);
453	460	return NULL;
454	460	}
455
456		/* The locked compare-and-set, after the local work is done. */
457	14.9k	if (!CRYPTO_THREAD_write_lock(lock)) {
458	0	BN_MONT_CTX_free(ret);
459	0	return NULL;
460	0	}
461
462	14.9k	if (*pmont) {
463	0	BN_MONT_CTX_free(ret);
464	0	ret = *pmont;
465	0	} else
466	14.9k	*pmont = ret;
467	14.9k	CRYPTO_THREAD_unlock(lock);
468	14.9k	return ret;
469	14.9k	}