12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499 |
- /*
- * Copyright 1995-2022 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
- */
- #include "internal/cryptlib.h"
- #include "internal/constant_time.h"
- #include "bn_local.h"
- #include <stdlib.h>
- #ifdef _WIN32
- # include <malloc.h>
- # ifndef alloca
- # define alloca _alloca
- # endif
- #elif defined(__GNUC__)
- # ifndef alloca
- # define alloca(s) __builtin_alloca((s))
- # endif
- #elif defined(__sun)
- # include <alloca.h>
- #endif
- #include "rsaz_exp.h"
- #undef SPARC_T4_MONT
- #if defined(OPENSSL_BN_ASM_MONT) && (defined(__sparc__) || defined(__sparc))
- # include "crypto/sparc_arch.h"
- # define SPARC_T4_MONT
- #endif
- /* maximum precomputation table size for *variable* sliding windows */
- #define TABLE_SIZE 32
- /* this one works - simple but works */
- int BN_exp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, BN_CTX *ctx)
- {
- int i, bits, ret = 0;
- BIGNUM *v, *rr;
- if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0
- || BN_get_flags(a, BN_FLG_CONSTTIME) != 0) {
- /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */
- ERR_raise(ERR_LIB_BN, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED);
- return 0;
- }
- BN_CTX_start(ctx);
- rr = ((r == a) || (r == p)) ? BN_CTX_get(ctx) : r;
- v = BN_CTX_get(ctx);
- if (rr == NULL || v == NULL)
- goto err;
- if (BN_copy(v, a) == NULL)
- goto err;
- bits = BN_num_bits(p);
- if (BN_is_odd(p)) {
- if (BN_copy(rr, a) == NULL)
- goto err;
- } else {
- if (!BN_one(rr))
- goto err;
- }
- for (i = 1; i < bits; i++) {
- if (!BN_sqr(v, v, ctx))
- goto err;
- if (BN_is_bit_set(p, i)) {
- if (!BN_mul(rr, rr, v, ctx))
- goto err;
- }
- }
- if (r != rr && BN_copy(r, rr) == NULL)
- goto err;
- ret = 1;
- err:
- BN_CTX_end(ctx);
- bn_check_top(r);
- return ret;
- }
- int BN_mod_exp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p, const BIGNUM *m,
- BN_CTX *ctx)
- {
- int ret;
- bn_check_top(a);
- bn_check_top(p);
- bn_check_top(m);
- /*-
- * For even modulus m = 2^k*m_odd, it might make sense to compute
- * a^p mod m_odd and a^p mod 2^k separately (with Montgomery
- * exponentiation for the odd part), using appropriate exponent
- * reductions, and combine the results using the CRT.
- *
- * For now, we use Montgomery only if the modulus is odd; otherwise,
- * exponentiation using the reciprocal-based quick remaindering
- * algorithm is used.
- *
- * (Timing obtained with expspeed.c [computations a^p mod m
- * where a, p, m are of the same length: 256, 512, 1024, 2048,
- * 4096, 8192 bits], compared to the running time of the
- * standard algorithm:
- *
- * BN_mod_exp_mont 33 .. 40 % [AMD K6-2, Linux, debug configuration]
- * 55 .. 77 % [UltraSparc processor, but
- * debug-solaris-sparcv8-gcc conf.]
- *
- * BN_mod_exp_recp 50 .. 70 % [AMD K6-2, Linux, debug configuration]
- * 62 .. 118 % [UltraSparc, debug-solaris-sparcv8-gcc]
- *
- * On the Sparc, BN_mod_exp_recp was faster than BN_mod_exp_mont
- * at 2048 and more bits, but at 512 and 1024 bits, it was
- * slower even than the standard algorithm!
- *
- * "Real" timings [linux-elf, solaris-sparcv9-gcc configurations]
- * should be obtained when the new Montgomery reduction code
- * has been integrated into OpenSSL.)
- */
- #define MONT_MUL_MOD
- #define MONT_EXP_WORD
- #define RECP_MUL_MOD
- #ifdef MONT_MUL_MOD
- if (BN_is_odd(m)) {
- # ifdef MONT_EXP_WORD
- if (a->top == 1 && !a->neg
- && (BN_get_flags(p, BN_FLG_CONSTTIME) == 0)
- && (BN_get_flags(a, BN_FLG_CONSTTIME) == 0)
- && (BN_get_flags(m, BN_FLG_CONSTTIME) == 0)) {
- BN_ULONG A = a->d[0];
- ret = BN_mod_exp_mont_word(r, A, p, m, ctx, NULL);
- } else
- # endif
- ret = BN_mod_exp_mont(r, a, p, m, ctx, NULL);
- } else
- #endif
- #ifdef RECP_MUL_MOD
- {
- ret = BN_mod_exp_recp(r, a, p, m, ctx);
- }
- #else
- {
- ret = BN_mod_exp_simple(r, a, p, m, ctx);
- }
- #endif
- bn_check_top(r);
- return ret;
- }
- int BN_mod_exp_recp(BIGNUM *r, const BIGNUM *a, const BIGNUM *p,
- const BIGNUM *m, BN_CTX *ctx)
- {
- int i, j, bits, ret = 0, wstart, wend, window, wvalue;
- int start = 1;
- BIGNUM *aa;
- /* Table of variables obtained from 'ctx' */
- BIGNUM *val[TABLE_SIZE];
- BN_RECP_CTX recp;
- if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0
- || BN_get_flags(a, BN_FLG_CONSTTIME) != 0
- || BN_get_flags(m, BN_FLG_CONSTTIME) != 0) {
- /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */
- ERR_raise(ERR_LIB_BN, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED);
- return 0;
- }
- bits = BN_num_bits(p);
- if (bits == 0) {
- /* x**0 mod 1, or x**0 mod -1 is still zero. */
- if (BN_abs_is_word(m, 1)) {
- ret = 1;
- BN_zero(r);
- } else {
- ret = BN_one(r);
- }
- return ret;
- }
- BN_RECP_CTX_init(&recp);
- BN_CTX_start(ctx);
- aa = BN_CTX_get(ctx);
- val[0] = BN_CTX_get(ctx);
- if (val[0] == NULL)
- goto err;
- if (m->neg) {
- /* ignore sign of 'm' */
- if (!BN_copy(aa, m))
- goto err;
- aa->neg = 0;
- if (BN_RECP_CTX_set(&recp, aa, ctx) <= 0)
- goto err;
- } else {
- if (BN_RECP_CTX_set(&recp, m, ctx) <= 0)
- goto err;
- }
- if (!BN_nnmod(val[0], a, m, ctx))
- goto err; /* 1 */
- if (BN_is_zero(val[0])) {
- BN_zero(r);
- ret = 1;
- goto err;
- }
- window = BN_window_bits_for_exponent_size(bits);
- if (window > 1) {
- if (!BN_mod_mul_reciprocal(aa, val[0], val[0], &recp, ctx))
- goto err; /* 2 */
- j = 1 << (window - 1);
- for (i = 1; i < j; i++) {
- if (((val[i] = BN_CTX_get(ctx)) == NULL) ||
- !BN_mod_mul_reciprocal(val[i], val[i - 1], aa, &recp, ctx))
- goto err;
- }
- }
- start = 1; /* This is used to avoid multiplication etc
- * when there is only the value '1' in the
- * buffer. */
- wvalue = 0; /* The 'value' of the window */
- wstart = bits - 1; /* The top bit of the window */
- wend = 0; /* The bottom bit of the window */
- if (!BN_one(r))
- goto err;
- for (;;) {
- if (BN_is_bit_set(p, wstart) == 0) {
- if (!start)
- if (!BN_mod_mul_reciprocal(r, r, r, &recp, ctx))
- goto err;
- if (wstart == 0)
- break;
- wstart--;
- continue;
- }
- /*
- * We now have wstart on a 'set' bit, we now need to work out how bit
- * a window to do. To do this we need to scan forward until the last
- * set bit before the end of the window
- */
- wvalue = 1;
- wend = 0;
- for (i = 1; i < window; i++) {
- if (wstart - i < 0)
- break;
- if (BN_is_bit_set(p, wstart - i)) {
- wvalue <<= (i - wend);
- wvalue |= 1;
- wend = i;
- }
- }
- /* wend is the size of the current window */
- j = wend + 1;
- /* add the 'bytes above' */
- if (!start)
- for (i = 0; i < j; i++) {
- if (!BN_mod_mul_reciprocal(r, r, r, &recp, ctx))
- goto err;
- }
- /* wvalue will be an odd number < 2^window */
- if (!BN_mod_mul_reciprocal(r, r, val[wvalue >> 1], &recp, ctx))
- goto err;
- /* move the 'window' down further */
- wstart -= wend + 1;
- wvalue = 0;
- start = 0;
- if (wstart < 0)
- break;
- }
- ret = 1;
- err:
- BN_CTX_end(ctx);
- BN_RECP_CTX_free(&recp);
- bn_check_top(r);
- return ret;
- }
- int BN_mod_exp_mont(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p,
- const BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *in_mont)
- {
- int i, j, bits, ret = 0, wstart, wend, window, wvalue;
- int start = 1;
- BIGNUM *d, *r;
- const BIGNUM *aa;
- /* Table of variables obtained from 'ctx' */
- BIGNUM *val[TABLE_SIZE];
- BN_MONT_CTX *mont = NULL;
- if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0
- || BN_get_flags(a, BN_FLG_CONSTTIME) != 0
- || BN_get_flags(m, BN_FLG_CONSTTIME) != 0) {
- return BN_mod_exp_mont_consttime(rr, a, p, m, ctx, in_mont);
- }
- bn_check_top(a);
- bn_check_top(p);
- bn_check_top(m);
- if (!BN_is_odd(m)) {
- ERR_raise(ERR_LIB_BN, BN_R_CALLED_WITH_EVEN_MODULUS);
- return 0;
- }
- bits = BN_num_bits(p);
- if (bits == 0) {
- /* x**0 mod 1, or x**0 mod -1 is still zero. */
- if (BN_abs_is_word(m, 1)) {
- ret = 1;
- BN_zero(rr);
- } else {
- ret = BN_one(rr);
- }
- return ret;
- }
- BN_CTX_start(ctx);
- d = BN_CTX_get(ctx);
- r = BN_CTX_get(ctx);
- val[0] = BN_CTX_get(ctx);
- if (val[0] == NULL)
- goto err;
- /*
- * If this is not done, things will break in the montgomery part
- */
- if (in_mont != NULL)
- mont = in_mont;
- else {
- if ((mont = BN_MONT_CTX_new()) == NULL)
- goto err;
- if (!BN_MONT_CTX_set(mont, m, ctx))
- goto err;
- }
- if (a->neg || BN_ucmp(a, m) >= 0) {
- if (!BN_nnmod(val[0], a, m, ctx))
- goto err;
- aa = val[0];
- } else
- aa = a;
- if (!bn_to_mont_fixed_top(val[0], aa, mont, ctx))
- goto err; /* 1 */
- window = BN_window_bits_for_exponent_size(bits);
- if (window > 1) {
- if (!bn_mul_mont_fixed_top(d, val[0], val[0], mont, ctx))
- goto err; /* 2 */
- j = 1 << (window - 1);
- for (i = 1; i < j; i++) {
- if (((val[i] = BN_CTX_get(ctx)) == NULL) ||
- !bn_mul_mont_fixed_top(val[i], val[i - 1], d, mont, ctx))
- goto err;
- }
- }
- start = 1; /* This is used to avoid multiplication etc
- * when there is only the value '1' in the
- * buffer. */
- wvalue = 0; /* The 'value' of the window */
- wstart = bits - 1; /* The top bit of the window */
- wend = 0; /* The bottom bit of the window */
- #if 1 /* by Shay Gueron's suggestion */
- j = m->top; /* borrow j */
- if (m->d[j - 1] & (((BN_ULONG)1) << (BN_BITS2 - 1))) {
- if (bn_wexpand(r, j) == NULL)
- goto err;
- /* 2^(top*BN_BITS2) - m */
- r->d[0] = (0 - m->d[0]) & BN_MASK2;
- for (i = 1; i < j; i++)
- r->d[i] = (~m->d[i]) & BN_MASK2;
- r->top = j;
- r->flags |= BN_FLG_FIXED_TOP;
- } else
- #endif
- if (!bn_to_mont_fixed_top(r, BN_value_one(), mont, ctx))
- goto err;
- for (;;) {
- if (BN_is_bit_set(p, wstart) == 0) {
- if (!start) {
- if (!bn_mul_mont_fixed_top(r, r, r, mont, ctx))
- goto err;
- }
- if (wstart == 0)
- break;
- wstart--;
- continue;
- }
- /*
- * We now have wstart on a 'set' bit, we now need to work out how bit
- * a window to do. To do this we need to scan forward until the last
- * set bit before the end of the window
- */
- wvalue = 1;
- wend = 0;
- for (i = 1; i < window; i++) {
- if (wstart - i < 0)
- break;
- if (BN_is_bit_set(p, wstart - i)) {
- wvalue <<= (i - wend);
- wvalue |= 1;
- wend = i;
- }
- }
- /* wend is the size of the current window */
- j = wend + 1;
- /* add the 'bytes above' */
- if (!start)
- for (i = 0; i < j; i++) {
- if (!bn_mul_mont_fixed_top(r, r, r, mont, ctx))
- goto err;
- }
- /* wvalue will be an odd number < 2^window */
- if (!bn_mul_mont_fixed_top(r, r, val[wvalue >> 1], mont, ctx))
- goto err;
- /* move the 'window' down further */
- wstart -= wend + 1;
- wvalue = 0;
- start = 0;
- if (wstart < 0)
- break;
- }
- /*
- * Done with zero-padded intermediate BIGNUMs. Final BN_from_montgomery
- * removes padding [if any] and makes return value suitable for public
- * API consumer.
- */
- #if defined(SPARC_T4_MONT)
- if (OPENSSL_sparcv9cap_P[0] & (SPARCV9_VIS3 | SPARCV9_PREFER_FPU)) {
- j = mont->N.top; /* borrow j */
- val[0]->d[0] = 1; /* borrow val[0] */
- for (i = 1; i < j; i++)
- val[0]->d[i] = 0;
- val[0]->top = j;
- if (!BN_mod_mul_montgomery(rr, r, val[0], mont, ctx))
- goto err;
- } else
- #endif
- if (!BN_from_montgomery(rr, r, mont, ctx))
- goto err;
- ret = 1;
- err:
- if (in_mont == NULL)
- BN_MONT_CTX_free(mont);
- BN_CTX_end(ctx);
- bn_check_top(rr);
- return ret;
- }
- static BN_ULONG bn_get_bits(const BIGNUM *a, int bitpos)
- {
- BN_ULONG ret = 0;
- int wordpos;
- wordpos = bitpos / BN_BITS2;
- bitpos %= BN_BITS2;
- if (wordpos >= 0 && wordpos < a->top) {
- ret = a->d[wordpos] & BN_MASK2;
- if (bitpos) {
- ret >>= bitpos;
- if (++wordpos < a->top)
- ret |= a->d[wordpos] << (BN_BITS2 - bitpos);
- }
- }
- return ret & BN_MASK2;
- }
- /*
- * BN_mod_exp_mont_consttime() stores the precomputed powers in a specific
- * layout so that accessing any of these table values shows the same access
- * pattern as far as cache lines are concerned. The following functions are
- * used to transfer a BIGNUM from/to that table.
- */
- static int MOD_EXP_CTIME_COPY_TO_PREBUF(const BIGNUM *b, int top,
- unsigned char *buf, int idx,
- int window)
- {
- int i, j;
- int width = 1 << window;
- BN_ULONG *table = (BN_ULONG *)buf;
- if (top > b->top)
- top = b->top; /* this works because 'buf' is explicitly
- * zeroed */
- for (i = 0, j = idx; i < top; i++, j += width) {
- table[j] = b->d[i];
- }
- return 1;
- }
- static int MOD_EXP_CTIME_COPY_FROM_PREBUF(BIGNUM *b, int top,
- unsigned char *buf, int idx,
- int window)
- {
- int i, j;
- int width = 1 << window;
- /*
- * We declare table 'volatile' in order to discourage compiler
- * from reordering loads from the table. Concern is that if
- * reordered in specific manner loads might give away the
- * information we are trying to conceal. Some would argue that
- * compiler can reorder them anyway, but it can as well be
- * argued that doing so would be violation of standard...
- */
- volatile BN_ULONG *table = (volatile BN_ULONG *)buf;
- if (bn_wexpand(b, top) == NULL)
- return 0;
- if (window <= 3) {
- for (i = 0; i < top; i++, table += width) {
- BN_ULONG acc = 0;
- for (j = 0; j < width; j++) {
- acc |= table[j] &
- ((BN_ULONG)0 - (constant_time_eq_int(j,idx)&1));
- }
- b->d[i] = acc;
- }
- } else {
- int xstride = 1 << (window - 2);
- BN_ULONG y0, y1, y2, y3;
- i = idx >> (window - 2); /* equivalent of idx / xstride */
- idx &= xstride - 1; /* equivalent of idx % xstride */
- y0 = (BN_ULONG)0 - (constant_time_eq_int(i,0)&1);
- y1 = (BN_ULONG)0 - (constant_time_eq_int(i,1)&1);
- y2 = (BN_ULONG)0 - (constant_time_eq_int(i,2)&1);
- y3 = (BN_ULONG)0 - (constant_time_eq_int(i,3)&1);
- for (i = 0; i < top; i++, table += width) {
- BN_ULONG acc = 0;
- for (j = 0; j < xstride; j++) {
- acc |= ( (table[j + 0 * xstride] & y0) |
- (table[j + 1 * xstride] & y1) |
- (table[j + 2 * xstride] & y2) |
- (table[j + 3 * xstride] & y3) )
- & ((BN_ULONG)0 - (constant_time_eq_int(j,idx)&1));
- }
- b->d[i] = acc;
- }
- }
- b->top = top;
- b->flags |= BN_FLG_FIXED_TOP;
- return 1;
- }
- /*
- * Given a pointer value, compute the next address that is a cache line
- * multiple.
- */
- #define MOD_EXP_CTIME_ALIGN(x_) \
- ((unsigned char*)(x_) + (MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH - (((size_t)(x_)) & (MOD_EXP_CTIME_MIN_CACHE_LINE_MASK))))
- /*
- * This variant of BN_mod_exp_mont() uses fixed windows and the special
- * precomputation memory layout to limit data-dependency to a minimum to
- * protect secret exponents (cf. the hyper-threading timing attacks pointed
- * out by Colin Percival,
- * http://www.daemonology.net/hyperthreading-considered-harmful/)
- */
- int BN_mod_exp_mont_consttime(BIGNUM *rr, const BIGNUM *a, const BIGNUM *p,
- const BIGNUM *m, BN_CTX *ctx,
- BN_MONT_CTX *in_mont)
- {
- int i, bits, ret = 0, window, wvalue, wmask, window0;
- int top;
- BN_MONT_CTX *mont = NULL;
- int numPowers;
- unsigned char *powerbufFree = NULL;
- int powerbufLen = 0;
- unsigned char *powerbuf = NULL;
- BIGNUM tmp, am;
- #if defined(SPARC_T4_MONT)
- unsigned int t4 = 0;
- #endif
- bn_check_top(a);
- bn_check_top(p);
- bn_check_top(m);
- if (!BN_is_odd(m)) {
- ERR_raise(ERR_LIB_BN, BN_R_CALLED_WITH_EVEN_MODULUS);
- return 0;
- }
- top = m->top;
- if (in_mont != NULL && BN_is_zero(&in_mont->N)) {
- ERR_raise(ERR_LIB_BN, ERR_R_PASSED_INVALID_ARGUMENT);
- return 0;
- }
- if ((unsigned int)top > INT_MAX / sizeof(m->d[0]) / (1 << 8)) {
- /* Prevent overflowing the powerbufLen computation below */
- ERR_raise(ERR_LIB_BN, BN_R_BIGNUM_TOO_LONG);
- return 0;
- }
- /*
- * Use all bits stored in |p|, rather than |BN_num_bits|, so we do not leak
- * whether the top bits are zero.
- */
- bits = p->top * BN_BITS2;
- if (bits == 0) {
- /* x**0 mod 1, or x**0 mod -1 is still zero. */
- if (BN_abs_is_word(m, 1)) {
- ret = 1;
- BN_zero(rr);
- } else {
- ret = BN_one(rr);
- }
- return ret;
- }
- BN_CTX_start(ctx);
- /*
- * Allocate a montgomery context if it was not supplied by the caller. If
- * this is not done, things will break in the montgomery part.
- */
- if (in_mont != NULL)
- mont = in_mont;
- else {
- if ((mont = BN_MONT_CTX_new()) == NULL)
- goto err;
- if (!BN_MONT_CTX_set(mont, m, ctx))
- goto err;
- }
- if (a->neg || BN_ucmp(a, m) >= 0) {
- BIGNUM *reduced = BN_CTX_get(ctx);
- if (reduced == NULL
- || !BN_nnmod(reduced, a, m, ctx)) {
- goto err;
- }
- a = reduced;
- }
- #ifdef RSAZ_ENABLED
- /*
- * If the size of the operands allow it, perform the optimized
- * RSAZ exponentiation. For further information see
- * crypto/bn/rsaz_exp.c and accompanying assembly modules.
- */
- if ((16 == a->top) && (16 == p->top) && (BN_num_bits(m) == 1024)
- && rsaz_avx2_eligible()) {
- if (NULL == bn_wexpand(rr, 16))
- goto err;
- RSAZ_1024_mod_exp_avx2(rr->d, a->d, p->d, m->d, mont->RR.d,
- mont->n0[0]);
- rr->top = 16;
- rr->neg = 0;
- bn_correct_top(rr);
- ret = 1;
- goto err;
- } else if ((8 == a->top) && (8 == p->top) && (BN_num_bits(m) == 512)) {
- if (NULL == bn_wexpand(rr, 8))
- goto err;
- RSAZ_512_mod_exp(rr->d, a->d, p->d, m->d, mont->n0[0], mont->RR.d);
- rr->top = 8;
- rr->neg = 0;
- bn_correct_top(rr);
- ret = 1;
- goto err;
- }
- #endif
- /* Get the window size to use with size of p. */
- window = BN_window_bits_for_ctime_exponent_size(bits);
- #if defined(SPARC_T4_MONT)
- if (window >= 5 && (top & 15) == 0 && top <= 64 &&
- (OPENSSL_sparcv9cap_P[1] & (CFR_MONTMUL | CFR_MONTSQR)) ==
- (CFR_MONTMUL | CFR_MONTSQR) && (t4 = OPENSSL_sparcv9cap_P[0]))
- window = 5;
- else
- #endif
- #if defined(OPENSSL_BN_ASM_MONT5)
- if (window >= 5) {
- window = 5; /* ~5% improvement for RSA2048 sign, and even
- * for RSA4096 */
- /* reserve space for mont->N.d[] copy */
- powerbufLen += top * sizeof(mont->N.d[0]);
- }
- #endif
- (void)0;
- /*
- * Allocate a buffer large enough to hold all of the pre-computed powers
- * of am, am itself and tmp.
- */
- numPowers = 1 << window;
- powerbufLen += sizeof(m->d[0]) * (top * numPowers +
- ((2 * top) >
- numPowers ? (2 * top) : numPowers));
- #ifdef alloca
- if (powerbufLen < 3072)
- powerbufFree =
- alloca(powerbufLen + MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH);
- else
- #endif
- if ((powerbufFree =
- OPENSSL_malloc(powerbufLen + MOD_EXP_CTIME_MIN_CACHE_LINE_WIDTH))
- == NULL)
- goto err;
- powerbuf = MOD_EXP_CTIME_ALIGN(powerbufFree);
- memset(powerbuf, 0, powerbufLen);
- #ifdef alloca
- if (powerbufLen < 3072)
- powerbufFree = NULL;
- #endif
- /* lay down tmp and am right after powers table */
- tmp.d = (BN_ULONG *)(powerbuf + sizeof(m->d[0]) * top * numPowers);
- am.d = tmp.d + top;
- tmp.top = am.top = 0;
- tmp.dmax = am.dmax = top;
- tmp.neg = am.neg = 0;
- tmp.flags = am.flags = BN_FLG_STATIC_DATA;
- /* prepare a^0 in Montgomery domain */
- #if 1 /* by Shay Gueron's suggestion */
- if (m->d[top - 1] & (((BN_ULONG)1) << (BN_BITS2 - 1))) {
- /* 2^(top*BN_BITS2) - m */
- tmp.d[0] = (0 - m->d[0]) & BN_MASK2;
- for (i = 1; i < top; i++)
- tmp.d[i] = (~m->d[i]) & BN_MASK2;
- tmp.top = top;
- } else
- #endif
- if (!bn_to_mont_fixed_top(&tmp, BN_value_one(), mont, ctx))
- goto err;
- /* prepare a^1 in Montgomery domain */
- if (!bn_to_mont_fixed_top(&am, a, mont, ctx))
- goto err;
- #if defined(SPARC_T4_MONT)
- if (t4) {
- typedef int (*bn_pwr5_mont_f) (BN_ULONG *tp, const BN_ULONG *np,
- const BN_ULONG *n0, const void *table,
- int power, int bits);
- int bn_pwr5_mont_t4_8(BN_ULONG *tp, const BN_ULONG *np,
- const BN_ULONG *n0, const void *table,
- int power, int bits);
- int bn_pwr5_mont_t4_16(BN_ULONG *tp, const BN_ULONG *np,
- const BN_ULONG *n0, const void *table,
- int power, int bits);
- int bn_pwr5_mont_t4_24(BN_ULONG *tp, const BN_ULONG *np,
- const BN_ULONG *n0, const void *table,
- int power, int bits);
- int bn_pwr5_mont_t4_32(BN_ULONG *tp, const BN_ULONG *np,
- const BN_ULONG *n0, const void *table,
- int power, int bits);
- static const bn_pwr5_mont_f pwr5_funcs[4] = {
- bn_pwr5_mont_t4_8, bn_pwr5_mont_t4_16,
- bn_pwr5_mont_t4_24, bn_pwr5_mont_t4_32
- };
- bn_pwr5_mont_f pwr5_worker = pwr5_funcs[top / 16 - 1];
- typedef int (*bn_mul_mont_f) (BN_ULONG *rp, const BN_ULONG *ap,
- const void *bp, const BN_ULONG *np,
- const BN_ULONG *n0);
- int bn_mul_mont_t4_8(BN_ULONG *rp, const BN_ULONG *ap, const void *bp,
- const BN_ULONG *np, const BN_ULONG *n0);
- int bn_mul_mont_t4_16(BN_ULONG *rp, const BN_ULONG *ap,
- const void *bp, const BN_ULONG *np,
- const BN_ULONG *n0);
- int bn_mul_mont_t4_24(BN_ULONG *rp, const BN_ULONG *ap,
- const void *bp, const BN_ULONG *np,
- const BN_ULONG *n0);
- int bn_mul_mont_t4_32(BN_ULONG *rp, const BN_ULONG *ap,
- const void *bp, const BN_ULONG *np,
- const BN_ULONG *n0);
- static const bn_mul_mont_f mul_funcs[4] = {
- bn_mul_mont_t4_8, bn_mul_mont_t4_16,
- bn_mul_mont_t4_24, bn_mul_mont_t4_32
- };
- bn_mul_mont_f mul_worker = mul_funcs[top / 16 - 1];
- void bn_mul_mont_vis3(BN_ULONG *rp, const BN_ULONG *ap,
- const void *bp, const BN_ULONG *np,
- const BN_ULONG *n0, int num);
- void bn_mul_mont_t4(BN_ULONG *rp, const BN_ULONG *ap,
- const void *bp, const BN_ULONG *np,
- const BN_ULONG *n0, int num);
- void bn_mul_mont_gather5_t4(BN_ULONG *rp, const BN_ULONG *ap,
- const void *table, const BN_ULONG *np,
- const BN_ULONG *n0, int num, int power);
- void bn_flip_n_scatter5_t4(const BN_ULONG *inp, size_t num,
- void *table, size_t power);
- void bn_gather5_t4(BN_ULONG *out, size_t num,
- void *table, size_t power);
- void bn_flip_t4(BN_ULONG *dst, BN_ULONG *src, size_t num);
- BN_ULONG *np = mont->N.d, *n0 = mont->n0;
- int stride = 5 * (6 - (top / 16 - 1)); /* multiple of 5, but less
- * than 32 */
- /*
- * BN_to_montgomery can contaminate words above .top [in
- * BN_DEBUG build...
- */
- for (i = am.top; i < top; i++)
- am.d[i] = 0;
- for (i = tmp.top; i < top; i++)
- tmp.d[i] = 0;
- bn_flip_n_scatter5_t4(tmp.d, top, powerbuf, 0);
- bn_flip_n_scatter5_t4(am.d, top, powerbuf, 1);
- if (!(*mul_worker) (tmp.d, am.d, am.d, np, n0) &&
- !(*mul_worker) (tmp.d, am.d, am.d, np, n0))
- bn_mul_mont_vis3(tmp.d, am.d, am.d, np, n0, top);
- bn_flip_n_scatter5_t4(tmp.d, top, powerbuf, 2);
- for (i = 3; i < 32; i++) {
- /* Calculate a^i = a^(i-1) * a */
- if (!(*mul_worker) (tmp.d, tmp.d, am.d, np, n0) &&
- !(*mul_worker) (tmp.d, tmp.d, am.d, np, n0))
- bn_mul_mont_vis3(tmp.d, tmp.d, am.d, np, n0, top);
- bn_flip_n_scatter5_t4(tmp.d, top, powerbuf, i);
- }
- /* switch to 64-bit domain */
- np = alloca(top * sizeof(BN_ULONG));
- top /= 2;
- bn_flip_t4(np, mont->N.d, top);
- /*
- * The exponent may not have a whole number of fixed-size windows.
- * To simplify the main loop, the initial window has between 1 and
- * full-window-size bits such that what remains is always a whole
- * number of windows
- */
- window0 = (bits - 1) % 5 + 1;
- wmask = (1 << window0) - 1;
- bits -= window0;
- wvalue = bn_get_bits(p, bits) & wmask;
- bn_gather5_t4(tmp.d, top, powerbuf, wvalue);
- /*
- * Scan the exponent one window at a time starting from the most
- * significant bits.
- */
- while (bits > 0) {
- if (bits < stride)
- stride = bits;
- bits -= stride;
- wvalue = bn_get_bits(p, bits);
- if ((*pwr5_worker) (tmp.d, np, n0, powerbuf, wvalue, stride))
- continue;
- /* retry once and fall back */
- if ((*pwr5_worker) (tmp.d, np, n0, powerbuf, wvalue, stride))
- continue;
- bits += stride - 5;
- wvalue >>= stride - 5;
- wvalue &= 31;
- bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top);
- bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top);
- bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top);
- bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top);
- bn_mul_mont_t4(tmp.d, tmp.d, tmp.d, np, n0, top);
- bn_mul_mont_gather5_t4(tmp.d, tmp.d, powerbuf, np, n0, top,
- wvalue);
- }
- bn_flip_t4(tmp.d, tmp.d, top);
- top *= 2;
- /* back to 32-bit domain */
- tmp.top = top;
- bn_correct_top(&tmp);
- OPENSSL_cleanse(np, top * sizeof(BN_ULONG));
- } else
- #endif
- #if defined(OPENSSL_BN_ASM_MONT5)
- if (window == 5 && top > 1) {
- /*
- * This optimization uses ideas from https://eprint.iacr.org/2011/239,
- * specifically optimization of cache-timing attack countermeasures,
- * pre-computation optimization, and Almost Montgomery Multiplication.
- *
- * The paper discusses a 4-bit window to optimize 512-bit modular
- * exponentiation, used in RSA-1024 with CRT, but RSA-1024 is no longer
- * important.
- *
- * |bn_mul_mont_gather5| and |bn_power5| implement the "almost"
- * reduction variant, so the values here may not be fully reduced.
- * They are bounded by R (i.e. they fit in |top| words), not |m|.
- * Additionally, we pass these "almost" reduced inputs into
- * |bn_mul_mont|, which implements the normal reduction variant.
- * Given those inputs, |bn_mul_mont| may not give reduced
- * output, but it will still produce "almost" reduced output.
- */
- void bn_mul_mont_gather5(BN_ULONG *rp, const BN_ULONG *ap,
- const void *table, const BN_ULONG *np,
- const BN_ULONG *n0, int num, int power);
- void bn_scatter5(const BN_ULONG *inp, size_t num,
- void *table, size_t power);
- void bn_gather5(BN_ULONG *out, size_t num, void *table, size_t power);
- void bn_power5(BN_ULONG *rp, const BN_ULONG *ap,
- const void *table, const BN_ULONG *np,
- const BN_ULONG *n0, int num, int power);
- int bn_get_bits5(const BN_ULONG *ap, int off);
- BN_ULONG *n0 = mont->n0, *np;
- /*
- * BN_to_montgomery can contaminate words above .top [in
- * BN_DEBUG build...
- */
- for (i = am.top; i < top; i++)
- am.d[i] = 0;
- for (i = tmp.top; i < top; i++)
- tmp.d[i] = 0;
- /*
- * copy mont->N.d[] to improve cache locality
- */
- for (np = am.d + top, i = 0; i < top; i++)
- np[i] = mont->N.d[i];
- bn_scatter5(tmp.d, top, powerbuf, 0);
- bn_scatter5(am.d, am.top, powerbuf, 1);
- bn_mul_mont(tmp.d, am.d, am.d, np, n0, top);
- bn_scatter5(tmp.d, top, powerbuf, 2);
- # if 0
- for (i = 3; i < 32; i++) {
- /* Calculate a^i = a^(i-1) * a */
- bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1);
- bn_scatter5(tmp.d, top, powerbuf, i);
- }
- # else
- /* same as above, but uses squaring for 1/2 of operations */
- for (i = 4; i < 32; i *= 2) {
- bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
- bn_scatter5(tmp.d, top, powerbuf, i);
- }
- for (i = 3; i < 8; i += 2) {
- int j;
- bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1);
- bn_scatter5(tmp.d, top, powerbuf, i);
- for (j = 2 * i; j < 32; j *= 2) {
- bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
- bn_scatter5(tmp.d, top, powerbuf, j);
- }
- }
- for (; i < 16; i += 2) {
- bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1);
- bn_scatter5(tmp.d, top, powerbuf, i);
- bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
- bn_scatter5(tmp.d, top, powerbuf, 2 * i);
- }
- for (; i < 32; i += 2) {
- bn_mul_mont_gather5(tmp.d, am.d, powerbuf, np, n0, top, i - 1);
- bn_scatter5(tmp.d, top, powerbuf, i);
- }
- # endif
- /*
- * The exponent may not have a whole number of fixed-size windows.
- * To simplify the main loop, the initial window has between 1 and
- * full-window-size bits such that what remains is always a whole
- * number of windows
- */
- window0 = (bits - 1) % 5 + 1;
- wmask = (1 << window0) - 1;
- bits -= window0;
- wvalue = bn_get_bits(p, bits) & wmask;
- bn_gather5(tmp.d, top, powerbuf, wvalue);
- /*
- * Scan the exponent one window at a time starting from the most
- * significant bits.
- */
- if (top & 7) {
- while (bits > 0) {
- bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
- bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
- bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
- bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
- bn_mul_mont(tmp.d, tmp.d, tmp.d, np, n0, top);
- bn_mul_mont_gather5(tmp.d, tmp.d, powerbuf, np, n0, top,
- bn_get_bits5(p->d, bits -= 5));
- }
- } else {
- while (bits > 0) {
- bn_power5(tmp.d, tmp.d, powerbuf, np, n0, top,
- bn_get_bits5(p->d, bits -= 5));
- }
- }
- tmp.top = top;
- /*
- * The result is now in |tmp| in Montgomery form, but it may not be
- * fully reduced. This is within bounds for |BN_from_montgomery|
- * (tmp < R <= m*R) so it will, when converting from Montgomery form,
- * produce a fully reduced result.
- *
- * This differs from Figure 2 of the paper, which uses AMM(h, 1) to
- * convert from Montgomery form with unreduced output, followed by an
- * extra reduction step. In the paper's terminology, we replace
- * steps 9 and 10 with MM(h, 1).
- */
- } else
- #endif
- {
- if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, 0, window))
- goto err;
- if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&am, top, powerbuf, 1, window))
- goto err;
- /*
- * If the window size is greater than 1, then calculate
- * val[i=2..2^winsize-1]. Powers are computed as a*a^(i-1) (even
- * powers could instead be computed as (a^(i/2))^2 to use the slight
- * performance advantage of sqr over mul).
- */
- if (window > 1) {
- if (!bn_mul_mont_fixed_top(&tmp, &am, &am, mont, ctx))
- goto err;
- if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, 2,
- window))
- goto err;
- for (i = 3; i < numPowers; i++) {
- /* Calculate a^i = a^(i-1) * a */
- if (!bn_mul_mont_fixed_top(&tmp, &am, &tmp, mont, ctx))
- goto err;
- if (!MOD_EXP_CTIME_COPY_TO_PREBUF(&tmp, top, powerbuf, i,
- window))
- goto err;
- }
- }
- /*
- * The exponent may not have a whole number of fixed-size windows.
- * To simplify the main loop, the initial window has between 1 and
- * full-window-size bits such that what remains is always a whole
- * number of windows
- */
- window0 = (bits - 1) % window + 1;
- wmask = (1 << window0) - 1;
- bits -= window0;
- wvalue = bn_get_bits(p, bits) & wmask;
- if (!MOD_EXP_CTIME_COPY_FROM_PREBUF(&tmp, top, powerbuf, wvalue,
- window))
- goto err;
- wmask = (1 << window) - 1;
- /*
- * Scan the exponent one window at a time starting from the most
- * significant bits.
- */
- while (bits > 0) {
- /* Square the result window-size times */
- for (i = 0; i < window; i++)
- if (!bn_mul_mont_fixed_top(&tmp, &tmp, &tmp, mont, ctx))
- goto err;
- /*
- * Get a window's worth of bits from the exponent
- * This avoids calling BN_is_bit_set for each bit, which
- * is not only slower but also makes each bit vulnerable to
- * EM (and likely other) side-channel attacks like One&Done
- * (for details see "One&Done: A Single-Decryption EM-Based
- * Attack on OpenSSL's Constant-Time Blinded RSA" by M. Alam,
- * H. Khan, M. Dey, N. Sinha, R. Callan, A. Zajic, and
- * M. Prvulovic, in USENIX Security'18)
- */
- bits -= window;
- wvalue = bn_get_bits(p, bits) & wmask;
- /*
- * Fetch the appropriate pre-computed value from the pre-buf
- */
- if (!MOD_EXP_CTIME_COPY_FROM_PREBUF(&am, top, powerbuf, wvalue,
- window))
- goto err;
- /* Multiply the result into the intermediate result */
- if (!bn_mul_mont_fixed_top(&tmp, &tmp, &am, mont, ctx))
- goto err;
- }
- }
- /*
- * Done with zero-padded intermediate BIGNUMs. Final BN_from_montgomery
- * removes padding [if any] and makes return value suitable for public
- * API consumer.
- */
- #if defined(SPARC_T4_MONT)
- if (OPENSSL_sparcv9cap_P[0] & (SPARCV9_VIS3 | SPARCV9_PREFER_FPU)) {
- am.d[0] = 1; /* borrow am */
- for (i = 1; i < top; i++)
- am.d[i] = 0;
- if (!BN_mod_mul_montgomery(rr, &tmp, &am, mont, ctx))
- goto err;
- } else
- #endif
- if (!BN_from_montgomery(rr, &tmp, mont, ctx))
- goto err;
- ret = 1;
- err:
- if (in_mont == NULL)
- BN_MONT_CTX_free(mont);
- if (powerbuf != NULL) {
- OPENSSL_cleanse(powerbuf, powerbufLen);
- OPENSSL_free(powerbufFree);
- }
- BN_CTX_end(ctx);
- return ret;
- }
- int BN_mod_exp_mont_word(BIGNUM *rr, BN_ULONG a, const BIGNUM *p,
- const BIGNUM *m, BN_CTX *ctx, BN_MONT_CTX *in_mont)
- {
- BN_MONT_CTX *mont = NULL;
- int b, bits, ret = 0;
- int r_is_one;
- BN_ULONG w, next_w;
- BIGNUM *r, *t;
- BIGNUM *swap_tmp;
- #define BN_MOD_MUL_WORD(r, w, m) \
- (BN_mul_word(r, (w)) && \
- (/* BN_ucmp(r, (m)) < 0 ? 1 :*/ \
- (BN_mod(t, r, m, ctx) && (swap_tmp = r, r = t, t = swap_tmp, 1))))
- /*
- * BN_MOD_MUL_WORD is only used with 'w' large, so the BN_ucmp test is
- * probably more overhead than always using BN_mod (which uses BN_copy if
- * a similar test returns true).
- */
- /*
- * We can use BN_mod and do not need BN_nnmod because our accumulator is
- * never negative (the result of BN_mod does not depend on the sign of
- * the modulus).
- */
- #define BN_TO_MONTGOMERY_WORD(r, w, mont) \
- (BN_set_word(r, (w)) && BN_to_montgomery(r, r, (mont), ctx))
- if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0
- || BN_get_flags(m, BN_FLG_CONSTTIME) != 0) {
- /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */
- ERR_raise(ERR_LIB_BN, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED);
- return 0;
- }
- bn_check_top(p);
- bn_check_top(m);
- if (!BN_is_odd(m)) {
- ERR_raise(ERR_LIB_BN, BN_R_CALLED_WITH_EVEN_MODULUS);
- return 0;
- }
- if (m->top == 1)
- a %= m->d[0]; /* make sure that 'a' is reduced */
- bits = BN_num_bits(p);
- if (bits == 0) {
- /* x**0 mod 1, or x**0 mod -1 is still zero. */
- if (BN_abs_is_word(m, 1)) {
- ret = 1;
- BN_zero(rr);
- } else {
- ret = BN_one(rr);
- }
- return ret;
- }
- if (a == 0) {
- BN_zero(rr);
- ret = 1;
- return ret;
- }
- BN_CTX_start(ctx);
- r = BN_CTX_get(ctx);
- t = BN_CTX_get(ctx);
- if (t == NULL)
- goto err;
- if (in_mont != NULL)
- mont = in_mont;
- else {
- if ((mont = BN_MONT_CTX_new()) == NULL)
- goto err;
- if (!BN_MONT_CTX_set(mont, m, ctx))
- goto err;
- }
- r_is_one = 1; /* except for Montgomery factor */
- /* bits-1 >= 0 */
- /* The result is accumulated in the product r*w. */
- w = a; /* bit 'bits-1' of 'p' is always set */
- for (b = bits - 2; b >= 0; b--) {
- /* First, square r*w. */
- next_w = w * w;
- if ((next_w / w) != w) { /* overflow */
- if (r_is_one) {
- if (!BN_TO_MONTGOMERY_WORD(r, w, mont))
- goto err;
- r_is_one = 0;
- } else {
- if (!BN_MOD_MUL_WORD(r, w, m))
- goto err;
- }
- next_w = 1;
- }
- w = next_w;
- if (!r_is_one) {
- if (!BN_mod_mul_montgomery(r, r, r, mont, ctx))
- goto err;
- }
- /* Second, multiply r*w by 'a' if exponent bit is set. */
- if (BN_is_bit_set(p, b)) {
- next_w = w * a;
- if ((next_w / a) != w) { /* overflow */
- if (r_is_one) {
- if (!BN_TO_MONTGOMERY_WORD(r, w, mont))
- goto err;
- r_is_one = 0;
- } else {
- if (!BN_MOD_MUL_WORD(r, w, m))
- goto err;
- }
- next_w = a;
- }
- w = next_w;
- }
- }
- /* Finally, set r:=r*w. */
- if (w != 1) {
- if (r_is_one) {
- if (!BN_TO_MONTGOMERY_WORD(r, w, mont))
- goto err;
- r_is_one = 0;
- } else {
- if (!BN_MOD_MUL_WORD(r, w, m))
- goto err;
- }
- }
- if (r_is_one) { /* can happen only if a == 1 */
- if (!BN_one(rr))
- goto err;
- } else {
- if (!BN_from_montgomery(rr, r, mont, ctx))
- goto err;
- }
- ret = 1;
- err:
- if (in_mont == NULL)
- BN_MONT_CTX_free(mont);
- BN_CTX_end(ctx);
- bn_check_top(rr);
- return ret;
- }
- /* The old fallback, simple version :-) */
- int BN_mod_exp_simple(BIGNUM *r, const BIGNUM *a, const BIGNUM *p,
- const BIGNUM *m, BN_CTX *ctx)
- {
- int i, j, bits, ret = 0, wstart, wend, window, wvalue;
- int start = 1;
- BIGNUM *d;
- /* Table of variables obtained from 'ctx' */
- BIGNUM *val[TABLE_SIZE];
- if (BN_get_flags(p, BN_FLG_CONSTTIME) != 0
- || BN_get_flags(a, BN_FLG_CONSTTIME) != 0
- || BN_get_flags(m, BN_FLG_CONSTTIME) != 0) {
- /* BN_FLG_CONSTTIME only supported by BN_mod_exp_mont() */
- ERR_raise(ERR_LIB_BN, ERR_R_SHOULD_NOT_HAVE_BEEN_CALLED);
- return 0;
- }
- bits = BN_num_bits(p);
- if (bits == 0) {
- /* x**0 mod 1, or x**0 mod -1 is still zero. */
- if (BN_abs_is_word(m, 1)) {
- ret = 1;
- BN_zero(r);
- } else {
- ret = BN_one(r);
- }
- return ret;
- }
- BN_CTX_start(ctx);
- d = BN_CTX_get(ctx);
- val[0] = BN_CTX_get(ctx);
- if (val[0] == NULL)
- goto err;
- if (!BN_nnmod(val[0], a, m, ctx))
- goto err; /* 1 */
- if (BN_is_zero(val[0])) {
- BN_zero(r);
- ret = 1;
- goto err;
- }
- window = BN_window_bits_for_exponent_size(bits);
- if (window > 1) {
- if (!BN_mod_mul(d, val[0], val[0], m, ctx))
- goto err; /* 2 */
- j = 1 << (window - 1);
- for (i = 1; i < j; i++) {
- if (((val[i] = BN_CTX_get(ctx)) == NULL) ||
- !BN_mod_mul(val[i], val[i - 1], d, m, ctx))
- goto err;
- }
- }
- start = 1; /* This is used to avoid multiplication etc
- * when there is only the value '1' in the
- * buffer. */
- wvalue = 0; /* The 'value' of the window */
- wstart = bits - 1; /* The top bit of the window */
- wend = 0; /* The bottom bit of the window */
- if (!BN_one(r))
- goto err;
- for (;;) {
- if (BN_is_bit_set(p, wstart) == 0) {
- if (!start)
- if (!BN_mod_mul(r, r, r, m, ctx))
- goto err;
- if (wstart == 0)
- break;
- wstart--;
- continue;
- }
- /*
- * We now have wstart on a 'set' bit, we now need to work out how bit
- * a window to do. To do this we need to scan forward until the last
- * set bit before the end of the window
- */
- wvalue = 1;
- wend = 0;
- for (i = 1; i < window; i++) {
- if (wstart - i < 0)
- break;
- if (BN_is_bit_set(p, wstart - i)) {
- wvalue <<= (i - wend);
- wvalue |= 1;
- wend = i;
- }
- }
- /* wend is the size of the current window */
- j = wend + 1;
- /* add the 'bytes above' */
- if (!start)
- for (i = 0; i < j; i++) {
- if (!BN_mod_mul(r, r, r, m, ctx))
- goto err;
- }
- /* wvalue will be an odd number < 2^window */
- if (!BN_mod_mul(r, r, val[wvalue >> 1], m, ctx))
- goto err;
- /* move the 'window' down further */
- wstart -= wend + 1;
- wvalue = 0;
- start = 0;
- if (wstart < 0)
- break;
- }
- ret = 1;
- err:
- BN_CTX_end(ctx);
- bn_check_top(r);
- return ret;
- }
- /*
- * This is a variant of modular exponentiation optimization that does
- * parallel 2-primes exponentiation using 256-bit (AVX512VL) AVX512_IFMA ISA
- * in 52-bit binary redundant representation.
- * If such instructions are not available, or input data size is not supported,
- * it falls back to two BN_mod_exp_mont_consttime() calls.
- */
- int BN_mod_exp_mont_consttime_x2(BIGNUM *rr1, const BIGNUM *a1, const BIGNUM *p1,
- const BIGNUM *m1, BN_MONT_CTX *in_mont1,
- BIGNUM *rr2, const BIGNUM *a2, const BIGNUM *p2,
- const BIGNUM *m2, BN_MONT_CTX *in_mont2,
- BN_CTX *ctx)
- {
- int ret = 0;
- #ifdef RSAZ_ENABLED
- BN_MONT_CTX *mont1 = NULL;
- BN_MONT_CTX *mont2 = NULL;
- if (ossl_rsaz_avx512ifma_eligible() &&
- (((a1->top == 16) && (p1->top == 16) && (BN_num_bits(m1) == 1024) &&
- (a2->top == 16) && (p2->top == 16) && (BN_num_bits(m2) == 1024)) ||
- ((a1->top == 24) && (p1->top == 24) && (BN_num_bits(m1) == 1536) &&
- (a2->top == 24) && (p2->top == 24) && (BN_num_bits(m2) == 1536)) ||
- ((a1->top == 32) && (p1->top == 32) && (BN_num_bits(m1) == 2048) &&
- (a2->top == 32) && (p2->top == 32) && (BN_num_bits(m2) == 2048)))) {
- int topn = a1->top;
- /* Modulus bits of |m1| and |m2| are equal */
- int mod_bits = BN_num_bits(m1);
- if (bn_wexpand(rr1, topn) == NULL)
- goto err;
- if (bn_wexpand(rr2, topn) == NULL)
- goto err;
- /* Ensure that montgomery contexts are initialized */
- if (in_mont1 != NULL) {
- mont1 = in_mont1;
- } else {
- if ((mont1 = BN_MONT_CTX_new()) == NULL)
- goto err;
- if (!BN_MONT_CTX_set(mont1, m1, ctx))
- goto err;
- }
- if (in_mont2 != NULL) {
- mont2 = in_mont2;
- } else {
- if ((mont2 = BN_MONT_CTX_new()) == NULL)
- goto err;
- if (!BN_MONT_CTX_set(mont2, m2, ctx))
- goto err;
- }
- ret = ossl_rsaz_mod_exp_avx512_x2(rr1->d, a1->d, p1->d, m1->d,
- mont1->RR.d, mont1->n0[0],
- rr2->d, a2->d, p2->d, m2->d,
- mont2->RR.d, mont2->n0[0],
- mod_bits);
- rr1->top = topn;
- rr1->neg = 0;
- bn_correct_top(rr1);
- bn_check_top(rr1);
- rr2->top = topn;
- rr2->neg = 0;
- bn_correct_top(rr2);
- bn_check_top(rr2);
- goto err;
- }
- #endif
- /* rr1 = a1^p1 mod m1 */
- ret = BN_mod_exp_mont_consttime(rr1, a1, p1, m1, ctx, in_mont1);
- /* rr2 = a2^p2 mod m2 */
- ret &= BN_mod_exp_mont_consttime(rr2, a2, p2, m2, ctx, in_mont2);
- #ifdef RSAZ_ENABLED
- err:
- if (in_mont2 == NULL)
- BN_MONT_CTX_free(mont2);
- if (in_mont1 == NULL)
- BN_MONT_CTX_free(mont1);
- #endif
- return ret;
- }
|