/*
- * Copyright 1995-2017 The OpenSSL Project Authors. All Rights Reserved.
+ * Copyright 1995-2021 The OpenSSL Project Authors. All Rights Reserved.
*
- * Licensed under the OpenSSL license (the "License"). You may not use
+ * 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
*/
#include "internal/cryptlib.h"
-#include "bn_lcl.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);
+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;
-#if defined(OPENSSL_BN_ASM_MONT) && defined(MONT_WORD)
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);
+ 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);
+ 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)
bn_check_top(tmp);
if (a == b) {
- if (!BN_sqr(tmp, a, ctx))
+ if (!bn_sqr_fixed_top(tmp, a, ctx))
goto err;
} else {
- if (!BN_mul(tmp, a, b, ctx))
+ 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))
+ 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);
+ return ret;
}
#ifdef MONT_WORD
-static int BN_from_montgomery_word(BIGNUM *ret, BIGNUM *r, BN_MONT_CTX *mont)
+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);
+ return 1;
}
max = (2 * nl); /* carry is stored separately */
if (bn_wexpand(r, max) == NULL)
- return (0);
+ 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);
+ 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;
}
if (bn_wexpand(ret, nl) == NULL)
- return (0);
+ return 0;
ret->top = nl;
+ ret->flags |= BN_FLG_FIXED_TOP;
ret->neg = r->neg;
rp = ret->d;
- ap = &(r->d[nl]);
-# define BRANCH_FREE 1
-# if BRANCH_FREE
- {
- BN_ULONG *nrp;
- size_t m;
+ /*
+ * 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]);
- v = bn_sub_words(rp, ap, np, nl) - carry;
- /*
- * if subtraction result is real, then trick unconditional memcpy
- * below to perform in-place "refresh" instead of actual copy.
- */
- m = (0 - (size_t)v);
- nrp =
- (BN_ULONG *)(((PTR_SIZE_INT) rp & ~m) | ((PTR_SIZE_INT) ap & m));
-
- for (i = 0, nl -= 4; i < nl; i += 4) {
- BN_ULONG t1, t2, t3, t4;
-
- t1 = nrp[i + 0];
- t2 = nrp[i + 1];
- t3 = nrp[i + 2];
- ap[i + 0] = 0;
- t4 = nrp[i + 3];
- ap[i + 1] = 0;
- rp[i + 0] = t1;
- ap[i + 2] = 0;
- rp[i + 1] = t2;
- ap[i + 3] = 0;
- rp[i + 2] = t3;
- rp[i + 3] = t4;
- }
- for (nl += 4; i < nl; i++)
- rp[i] = nrp[i], ap[i] = 0;
+ 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;
}
-# else
- if (bn_sub_words(rp, ap, np, nl) - carry)
- memcpy(rp, ap, nl * sizeof(BN_ULONG));
-# endif
- bn_correct_top(r);
- bn_correct_top(ret);
- bn_check_top(ret);
- return (1);
+ 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);
+ 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;
err:
BN_CTX_end(ctx);
#endif /* MONT_WORD */
- return (retn);
+ 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)
- return (NULL);
+ 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);
+ 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));
+ bn_init(&ctx->RR);
+ bn_init(&ctx->N);
+ bn_init(&ctx->Ni);
ctx->n0[0] = ctx->n0[1] = 0;
ctx->flags = 0;
}
{
if (mont == NULL)
return;
-
- BN_clear_free(&(mont->RR));
- BN_clear_free(&(mont->N));
- BN_clear_free(&(mont->Ni));
+ 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;
+ int i, ret = 0;
BIGNUM *Ri, *R;
if (BN_is_zero(mod))
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
if ((buf[1] = mod->top > 1 ? mod->d[1] : 0))
tmod.top = 2;
- if ((BN_mod_inverse(Ri, R, &tmod, ctx)) == NULL)
+ 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 */
buf[1] = 0;
tmod.top = buf[0] != 0 ? 1 : 0;
/* Ri = R^-1 mod N */
- if ((BN_mod_inverse(Ri, R, &tmod, ctx)) == NULL)
+ 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_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);
BN_MONT_CTX *BN_MONT_CTX_copy(BN_MONT_CTX *to, BN_MONT_CTX *from)
{
if (to == from)
- return (to);
+ return to;
if (!BN_copy(&(to->RR), &(from->RR)))
return NULL;
to->ri = from->ri;
to->n0[0] = from->n0[0];
to->n0[1] = from->n0[1];
- return (to);
+ return to;
}
BN_MONT_CTX *BN_MONT_CTX_set_locked(BN_MONT_CTX **pmont, CRYPTO_RWLOCK *lock,
{
BN_MONT_CTX *ret;
- CRYPTO_THREAD_read_lock(lock);
+ if (!CRYPTO_THREAD_read_lock(lock))
+ return NULL;
ret = *pmont;
CRYPTO_THREAD_unlock(lock);
if (ret)
return ret;
/*
- * We don't want to serialise globally while doing our lazy-init math in
+ * 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
}
/* The locked compare-and-set, after the local work is done. */
- CRYPTO_THREAD_write_lock(lock);
+ if (!CRYPTO_THREAD_write_lock(lock)) {
+ BN_MONT_CTX_free(ret);
+ return NULL;
+ }
+
if (*pmont) {
BN_MONT_CTX_free(ret);
ret = *pmont;