/* $OpenBSD: gcm128.c,v 1.15 2016/11/04 17:30:30 miod Exp $ */ /* ==================================================================== * Copyright (c) 2010 The OpenSSL Project. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in * the documentation and/or other materials provided with the * distribution. * * 3. All advertising materials mentioning features or use of this * software must display the following acknowledgment: * "This product includes software developed by the OpenSSL Project * for use in the OpenSSL Toolkit. (http://www.openssl.org/)" * * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to * endorse or promote products derived from this software without * prior written permission. For written permission, please contact * openssl-core@openssl.org. * * 5. Products derived from this software may not be called "OpenSSL" * nor may "OpenSSL" appear in their names without prior written * permission of the OpenSSL Project. * * 6. Redistributions of any form whatsoever must retain the following * acknowledgment: * "This product includes software developed by the OpenSSL Project * for use in the OpenSSL Toolkit (http://www.openssl.org/)" * * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE OpenSSL PROJECT OR * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED * OF THE POSSIBILITY OF SUCH DAMAGE. * ==================================================================== */ #define OPENSSL_FIPSAPI #include #include "modes_lcl.h" #include #ifndef MODES_DEBUG # ifndef NDEBUG # define NDEBUG # endif #endif #if defined(BSWAP4) && defined(__STRICT_ALIGNMENT) /* redefine, because alignment is ensured */ #undef GETU32 #define GETU32(p) BSWAP4(*(const u32 *)(p)) #undef PUTU32 #define PUTU32(p,v) *(u32 *)(p) = BSWAP4(v) #endif #define PACK(s) ((size_t)(s)<<(sizeof(size_t)*8-16)) #define REDUCE1BIT(V) \ do { \ if (sizeof(size_t)==8) { \ u64 T = U64(0xe100000000000000) & (0-(V.lo&1)); \ V.lo = (V.hi<<63)|(V.lo>>1); \ V.hi = (V.hi>>1 )^T; \ } else { \ u32 T = 0xe1000000U & (0-(u32)(V.lo&1)); \ V.lo = (V.hi<<63)|(V.lo>>1); \ V.hi = (V.hi>>1 )^((u64)T<<32); \ } \ } while(0) /* * Even though permitted values for TABLE_BITS are 8, 4 and 1, it should * never be set to 8. 8 is effectively reserved for testing purposes. * TABLE_BITS>1 are lookup-table-driven implementations referred to as * "Shoup's" in GCM specification. In other words OpenSSL does not cover * whole spectrum of possible table driven implementations. Why? In * non-"Shoup's" case memory access pattern is segmented in such manner, * that it's trivial to see that cache timing information can reveal * fair portion of intermediate hash value. Given that ciphertext is * always available to attacker, it's possible for him to attempt to * deduce secret parameter H and if successful, tamper with messages * [which is nothing but trivial in CTR mode]. In "Shoup's" case it's * not as trivial, but there is no reason to believe that it's resistant * to cache-timing attack. And the thing about "8-bit" implementation is * that it consumes 16 (sixteen) times more memory, 4KB per individual * key + 1KB shared. Well, on pros side it should be twice as fast as * "4-bit" version. And for gcc-generated x86[_64] code, "8-bit" version * was observed to run ~75% faster, closer to 100% for commercial * compilers... Yet "4-bit" procedure is preferred, because it's * believed to provide better security-performance balance and adequate * all-round performance. "All-round" refers to things like: * * - shorter setup time effectively improves overall timing for * handling short messages; * - larger table allocation can become unbearable because of VM * subsystem penalties (for example on Windows large enough free * results in VM working set trimming, meaning that consequent * malloc would immediately incur working set expansion); * - larger table has larger cache footprint, which can affect * performance of other code paths (not necessarily even from same * thread in Hyper-Threading world); * * Value of 1 is not appropriate for performance reasons. */ #if TABLE_BITS==8 static void gcm_init_8bit(u128 Htable[256], u64 H[2]) { int i, j; u128 V; Htable[0].hi = 0; Htable[0].lo = 0; V.hi = H[0]; V.lo = H[1]; for (Htable[128]=V, i=64; i>0; i>>=1) { REDUCE1BIT(V); Htable[i] = V; } for (i=2; i<256; i<<=1) { u128 *Hi = Htable+i, H0 = *Hi; for (j=1; j>8); Z.hi = (Z.hi>>8); if (sizeof(size_t)==8) Z.hi ^= rem_8bit[rem]; else Z.hi ^= (u64)rem_8bit[rem]<<32; } if (BYTE_ORDER == LITTLE_ENDIAN) { #ifdef BSWAP8 Xi[0] = BSWAP8(Z.hi); Xi[1] = BSWAP8(Z.lo); #else u8 *p = (u8 *)Xi; u32 v; v = (u32)(Z.hi>>32); PUTU32(p,v); v = (u32)(Z.hi); PUTU32(p+4,v); v = (u32)(Z.lo>>32); PUTU32(p+8,v); v = (u32)(Z.lo); PUTU32(p+12,v); #endif } else { Xi[0] = Z.hi; Xi[1] = Z.lo; } } #define GCM_MUL(ctx,Xi) gcm_gmult_8bit(ctx->Xi.u,ctx->Htable) #elif TABLE_BITS==4 static void gcm_init_4bit(u128 Htable[16], u64 H[2]) { u128 V; #if defined(OPENSSL_SMALL_FOOTPRINT) int i; #endif Htable[0].hi = 0; Htable[0].lo = 0; V.hi = H[0]; V.lo = H[1]; #if defined(OPENSSL_SMALL_FOOTPRINT) for (Htable[8]=V, i=4; i>0; i>>=1) { REDUCE1BIT(V); Htable[i] = V; } for (i=2; i<16; i<<=1) { u128 *Hi = Htable+i; int j; for (V=*Hi, j=1; j>32; Htable[j].lo = V.hi<<32|V.hi>>32; } } #endif } #ifndef GHASH_ASM static const size_t rem_4bit[16] = { PACK(0x0000), PACK(0x1C20), PACK(0x3840), PACK(0x2460), PACK(0x7080), PACK(0x6CA0), PACK(0x48C0), PACK(0x54E0), PACK(0xE100), PACK(0xFD20), PACK(0xD940), PACK(0xC560), PACK(0x9180), PACK(0x8DA0), PACK(0xA9C0), PACK(0xB5E0) }; static void gcm_gmult_4bit(u64 Xi[2], const u128 Htable[16]) { u128 Z; int cnt = 15; size_t rem, nlo, nhi; nlo = ((const u8 *)Xi)[15]; nhi = nlo>>4; nlo &= 0xf; Z.hi = Htable[nlo].hi; Z.lo = Htable[nlo].lo; while (1) { rem = (size_t)Z.lo&0xf; Z.lo = (Z.hi<<60)|(Z.lo>>4); Z.hi = (Z.hi>>4); if (sizeof(size_t)==8) Z.hi ^= rem_4bit[rem]; else Z.hi ^= (u64)rem_4bit[rem]<<32; Z.hi ^= Htable[nhi].hi; Z.lo ^= Htable[nhi].lo; if (--cnt<0) break; nlo = ((const u8 *)Xi)[cnt]; nhi = nlo>>4; nlo &= 0xf; rem = (size_t)Z.lo&0xf; Z.lo = (Z.hi<<60)|(Z.lo>>4); Z.hi = (Z.hi>>4); if (sizeof(size_t)==8) Z.hi ^= rem_4bit[rem]; else Z.hi ^= (u64)rem_4bit[rem]<<32; Z.hi ^= Htable[nlo].hi; Z.lo ^= Htable[nlo].lo; } if (BYTE_ORDER == LITTLE_ENDIAN) { #ifdef BSWAP8 Xi[0] = BSWAP8(Z.hi); Xi[1] = BSWAP8(Z.lo); #else u8 *p = (u8 *)Xi; u32 v; v = (u32)(Z.hi>>32); PUTU32(p,v); v = (u32)(Z.hi); PUTU32(p+4,v); v = (u32)(Z.lo>>32); PUTU32(p+8,v); v = (u32)(Z.lo); PUTU32(p+12,v); #endif } else { Xi[0] = Z.hi; Xi[1] = Z.lo; } } #if !defined(OPENSSL_SMALL_FOOTPRINT) /* * Streamed gcm_mult_4bit, see CRYPTO_gcm128_[en|de]crypt for * details... Compiler-generated code doesn't seem to give any * performance improvement, at least not on x86[_64]. It's here * mostly as reference and a placeholder for possible future * non-trivial optimization[s]... */ static void gcm_ghash_4bit(u64 Xi[2],const u128 Htable[16], const u8 *inp,size_t len) { u128 Z; int cnt; size_t rem, nlo, nhi; #if 1 do { cnt = 15; nlo = ((const u8 *)Xi)[15]; nlo ^= inp[15]; nhi = nlo>>4; nlo &= 0xf; Z.hi = Htable[nlo].hi; Z.lo = Htable[nlo].lo; while (1) { rem = (size_t)Z.lo&0xf; Z.lo = (Z.hi<<60)|(Z.lo>>4); Z.hi = (Z.hi>>4); if (sizeof(size_t)==8) Z.hi ^= rem_4bit[rem]; else Z.hi ^= (u64)rem_4bit[rem]<<32; Z.hi ^= Htable[nhi].hi; Z.lo ^= Htable[nhi].lo; if (--cnt<0) break; nlo = ((const u8 *)Xi)[cnt]; nlo ^= inp[cnt]; nhi = nlo>>4; nlo &= 0xf; rem = (size_t)Z.lo&0xf; Z.lo = (Z.hi<<60)|(Z.lo>>4); Z.hi = (Z.hi>>4); if (sizeof(size_t)==8) Z.hi ^= rem_4bit[rem]; else Z.hi ^= (u64)rem_4bit[rem]<<32; Z.hi ^= Htable[nlo].hi; Z.lo ^= Htable[nlo].lo; } #else /* * Extra 256+16 bytes per-key plus 512 bytes shared tables * [should] give ~50% improvement... One could have PACK()-ed * the rem_8bit even here, but the priority is to minimize * cache footprint... */ u128 Hshr4[16]; /* Htable shifted right by 4 bits */ u8 Hshl4[16]; /* Htable shifted left by 4 bits */ static const unsigned short rem_8bit[256] = { 0x0000, 0x01C2, 0x0384, 0x0246, 0x0708, 0x06CA, 0x048C, 0x054E, 0x0E10, 0x0FD2, 0x0D94, 0x0C56, 0x0918, 0x08DA, 0x0A9C, 0x0B5E, 0x1C20, 0x1DE2, 0x1FA4, 0x1E66, 0x1B28, 0x1AEA, 0x18AC, 0x196E, 0x1230, 0x13F2, 0x11B4, 0x1076, 0x1538, 0x14FA, 0x16BC, 0x177E, 0x3840, 0x3982, 0x3BC4, 0x3A06, 0x3F48, 0x3E8A, 0x3CCC, 0x3D0E, 0x3650, 0x3792, 0x35D4, 0x3416, 0x3158, 0x309A, 0x32DC, 0x331E, 0x2460, 0x25A2, 0x27E4, 0x2626, 0x2368, 0x22AA, 0x20EC, 0x212E, 0x2A70, 0x2BB2, 0x29F4, 0x2836, 0x2D78, 0x2CBA, 0x2EFC, 0x2F3E, 0x7080, 0x7142, 0x7304, 0x72C6, 0x7788, 0x764A, 0x740C, 0x75CE, 0x7E90, 0x7F52, 0x7D14, 0x7CD6, 0x7998, 0x785A, 0x7A1C, 0x7BDE, 0x6CA0, 0x6D62, 0x6F24, 0x6EE6, 0x6BA8, 0x6A6A, 0x682C, 0x69EE, 0x62B0, 0x6372, 0x6134, 0x60F6, 0x65B8, 0x647A, 0x663C, 0x67FE, 0x48C0, 0x4902, 0x4B44, 0x4A86, 0x4FC8, 0x4E0A, 0x4C4C, 0x4D8E, 0x46D0, 0x4712, 0x4554, 0x4496, 0x41D8, 0x401A, 0x425C, 0x439E, 0x54E0, 0x5522, 0x5764, 0x56A6, 0x53E8, 0x522A, 0x506C, 0x51AE, 0x5AF0, 0x5B32, 0x5974, 0x58B6, 0x5DF8, 0x5C3A, 0x5E7C, 0x5FBE, 0xE100, 0xE0C2, 0xE284, 0xE346, 0xE608, 0xE7CA, 0xE58C, 0xE44E, 0xEF10, 0xEED2, 0xEC94, 0xED56, 0xE818, 0xE9DA, 0xEB9C, 0xEA5E, 0xFD20, 0xFCE2, 0xFEA4, 0xFF66, 0xFA28, 0xFBEA, 0xF9AC, 0xF86E, 0xF330, 0xF2F2, 0xF0B4, 0xF176, 0xF438, 0xF5FA, 0xF7BC, 0xF67E, 0xD940, 0xD882, 0xDAC4, 0xDB06, 0xDE48, 0xDF8A, 0xDDCC, 0xDC0E, 0xD750, 0xD692, 0xD4D4, 0xD516, 0xD058, 0xD19A, 0xD3DC, 0xD21E, 0xC560, 0xC4A2, 0xC6E4, 0xC726, 0xC268, 0xC3AA, 0xC1EC, 0xC02E, 0xCB70, 0xCAB2, 0xC8F4, 0xC936, 0xCC78, 0xCDBA, 0xCFFC, 0xCE3E, 0x9180, 0x9042, 0x9204, 0x93C6, 0x9688, 0x974A, 0x950C, 0x94CE, 0x9F90, 0x9E52, 0x9C14, 0x9DD6, 0x9898, 0x995A, 0x9B1C, 0x9ADE, 0x8DA0, 0x8C62, 0x8E24, 0x8FE6, 0x8AA8, 0x8B6A, 0x892C, 0x88EE, 0x83B0, 0x8272, 0x8034, 0x81F6, 0x84B8, 0x857A, 0x873C, 0x86FE, 0xA9C0, 0xA802, 0xAA44, 0xAB86, 0xAEC8, 0xAF0A, 0xAD4C, 0xAC8E, 0xA7D0, 0xA612, 0xA454, 0xA596, 0xA0D8, 0xA11A, 0xA35C, 0xA29E, 0xB5E0, 0xB422, 0xB664, 0xB7A6, 0xB2E8, 0xB32A, 0xB16C, 0xB0AE, 0xBBF0, 0xBA32, 0xB874, 0xB9B6, 0xBCF8, 0xBD3A, 0xBF7C, 0xBEBE }; /* * This pre-processing phase slows down procedure by approximately * same time as it makes each loop spin faster. In other words * single block performance is approximately same as straightforward * "4-bit" implementation, and then it goes only faster... */ for (cnt=0; cnt<16; ++cnt) { Z.hi = Htable[cnt].hi; Z.lo = Htable[cnt].lo; Hshr4[cnt].lo = (Z.hi<<60)|(Z.lo>>4); Hshr4[cnt].hi = (Z.hi>>4); Hshl4[cnt] = (u8)(Z.lo<<4); } do { for (Z.lo=0, Z.hi=0, cnt=15; cnt; --cnt) { nlo = ((const u8 *)Xi)[cnt]; nlo ^= inp[cnt]; nhi = nlo>>4; nlo &= 0xf; Z.hi ^= Htable[nlo].hi; Z.lo ^= Htable[nlo].lo; rem = (size_t)Z.lo&0xff; Z.lo = (Z.hi<<56)|(Z.lo>>8); Z.hi = (Z.hi>>8); Z.hi ^= Hshr4[nhi].hi; Z.lo ^= Hshr4[nhi].lo; Z.hi ^= (u64)rem_8bit[rem^Hshl4[nhi]]<<48; } nlo = ((const u8 *)Xi)[0]; nlo ^= inp[0]; nhi = nlo>>4; nlo &= 0xf; Z.hi ^= Htable[nlo].hi; Z.lo ^= Htable[nlo].lo; rem = (size_t)Z.lo&0xf; Z.lo = (Z.hi<<60)|(Z.lo>>4); Z.hi = (Z.hi>>4); Z.hi ^= Htable[nhi].hi; Z.lo ^= Htable[nhi].lo; Z.hi ^= ((u64)rem_8bit[rem<<4])<<48; #endif if (BYTE_ORDER == LITTLE_ENDIAN) { #ifdef BSWAP8 Xi[0] = BSWAP8(Z.hi); Xi[1] = BSWAP8(Z.lo); #else u8 *p = (u8 *)Xi; u32 v; v = (u32)(Z.hi>>32); PUTU32(p,v); v = (u32)(Z.hi); PUTU32(p+4,v); v = (u32)(Z.lo>>32); PUTU32(p+8,v); v = (u32)(Z.lo); PUTU32(p+12,v); #endif } else { Xi[0] = Z.hi; Xi[1] = Z.lo; } } while (inp+=16, len-=16); } #endif #else void gcm_gmult_4bit(u64 Xi[2],const u128 Htable[16]); void gcm_ghash_4bit(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len); #endif #define GCM_MUL(ctx,Xi) gcm_gmult_4bit(ctx->Xi.u,ctx->Htable) #if defined(GHASH_ASM) || !defined(OPENSSL_SMALL_FOOTPRINT) #define GHASH(ctx,in,len) gcm_ghash_4bit((ctx)->Xi.u,(ctx)->Htable,in,len) /* GHASH_CHUNK is "stride parameter" missioned to mitigate cache * trashing effect. In other words idea is to hash data while it's * still in L1 cache after encryption pass... */ #define GHASH_CHUNK (3*1024) #endif #else /* TABLE_BITS */ static void gcm_gmult_1bit(u64 Xi[2],const u64 H[2]) { u128 V,Z = { 0,0 }; long X; int i,j; const long *xi = (const long *)Xi; V.hi = H[0]; /* H is in host byte order, no byte swapping */ V.lo = H[1]; for (j=0; j<16/sizeof(long); ++j) { if (BYTE_ORDER == LITTLE_ENDIAN) { if (sizeof(long)==8) { #ifdef BSWAP8 X = (long)(BSWAP8(xi[j])); #else const u8 *p = (const u8 *)(xi+j); X = (long)((u64)GETU32(p)<<32|GETU32(p+4)); #endif } else { const u8 *p = (const u8 *)(xi+j); X = (long)GETU32(p); } } else X = xi[j]; for (i=0; i<8*sizeof(long); ++i, X<<=1) { u64 M = (u64)(X>>(8*sizeof(long)-1)); Z.hi ^= V.hi&M; Z.lo ^= V.lo&M; REDUCE1BIT(V); } } if (BYTE_ORDER == LITTLE_ENDIAN) { #ifdef BSWAP8 Xi[0] = BSWAP8(Z.hi); Xi[1] = BSWAP8(Z.lo); #else u8 *p = (u8 *)Xi; u32 v; v = (u32)(Z.hi>>32); PUTU32(p,v); v = (u32)(Z.hi); PUTU32(p+4,v); v = (u32)(Z.lo>>32); PUTU32(p+8,v); v = (u32)(Z.lo); PUTU32(p+12,v); #endif } else { Xi[0] = Z.hi; Xi[1] = Z.lo; } } #define GCM_MUL(ctx,Xi) gcm_gmult_1bit(ctx->Xi.u,ctx->H.u) #endif #if defined(GHASH_ASM) && \ (defined(__i386) || defined(__i386__) || \ defined(__x86_64) || defined(__x86_64__) || \ defined(_M_IX86) || defined(_M_AMD64) || defined(_M_X64)) #include "x86_arch.h" #endif #if TABLE_BITS==4 && defined(GHASH_ASM) # if (defined(__i386) || defined(__i386__) || \ defined(__x86_64) || defined(__x86_64__) || \ defined(_M_IX86) || defined(_M_AMD64) || defined(_M_X64)) # define GHASH_ASM_X86_OR_64 # define GCM_FUNCREF_4BIT void gcm_init_clmul(u128 Htable[16],const u64 Xi[2]); void gcm_gmult_clmul(u64 Xi[2],const u128 Htable[16]); void gcm_ghash_clmul(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len); # if defined(__i386) || defined(__i386__) || defined(_M_IX86) # define GHASH_ASM_X86 void gcm_gmult_4bit_mmx(u64 Xi[2],const u128 Htable[16]); void gcm_ghash_4bit_mmx(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len); void gcm_gmult_4bit_x86(u64 Xi[2],const u128 Htable[16]); void gcm_ghash_4bit_x86(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len); # endif # elif defined(__arm__) || defined(__arm) # include "arm_arch.h" # if __ARM_ARCH__>=7 # define GHASH_ASM_ARM # define GCM_FUNCREF_4BIT void gcm_gmult_neon(u64 Xi[2],const u128 Htable[16]); void gcm_ghash_neon(u64 Xi[2],const u128 Htable[16],const u8 *inp,size_t len); # endif # endif #endif #ifdef GCM_FUNCREF_4BIT # undef GCM_MUL # define GCM_MUL(ctx,Xi) (*gcm_gmult_p)(ctx->Xi.u,ctx->Htable) # ifdef GHASH # undef GHASH # define GHASH(ctx,in,len) (*gcm_ghash_p)(ctx->Xi.u,ctx->Htable,in,len) # endif #endif void CRYPTO_gcm128_init(GCM128_CONTEXT *ctx,void *key,block128_f block) { memset(ctx,0,sizeof(*ctx)); ctx->block = block; ctx->key = key; (*block)(ctx->H.c,ctx->H.c,key); if (BYTE_ORDER == LITTLE_ENDIAN) { /* H is stored in host byte order */ #ifdef BSWAP8 ctx->H.u[0] = BSWAP8(ctx->H.u[0]); ctx->H.u[1] = BSWAP8(ctx->H.u[1]); #else u8 *p = ctx->H.c; u64 hi,lo; hi = (u64)GETU32(p) <<32|GETU32(p+4); lo = (u64)GETU32(p+8)<<32|GETU32(p+12); ctx->H.u[0] = hi; ctx->H.u[1] = lo; #endif } #if TABLE_BITS==8 gcm_init_8bit(ctx->Htable,ctx->H.u); #elif TABLE_BITS==4 # if defined(GHASH_ASM_X86_OR_64) # if !defined(GHASH_ASM_X86) || defined(OPENSSL_IA32_SSE2) /* check FXSR and PCLMULQDQ bits */ if ((OPENSSL_cpu_caps() & (CPUCAP_MASK_FXSR | CPUCAP_MASK_PCLMUL)) == (CPUCAP_MASK_FXSR | CPUCAP_MASK_PCLMUL)) { gcm_init_clmul(ctx->Htable,ctx->H.u); ctx->gmult = gcm_gmult_clmul; ctx->ghash = gcm_ghash_clmul; return; } # endif gcm_init_4bit(ctx->Htable,ctx->H.u); # if defined(GHASH_ASM_X86) /* x86 only */ # if defined(OPENSSL_IA32_SSE2) if (OPENSSL_cpu_caps() & CPUCAP_MASK_SSE) { /* check SSE bit */ # else if (OPENSSL_cpu_caps() & CPUCAP_MASK_MMX) { /* check MMX bit */ # endif ctx->gmult = gcm_gmult_4bit_mmx; ctx->ghash = gcm_ghash_4bit_mmx; } else { ctx->gmult = gcm_gmult_4bit_x86; ctx->ghash = gcm_ghash_4bit_x86; } # else ctx->gmult = gcm_gmult_4bit; ctx->ghash = gcm_ghash_4bit; # endif # elif defined(GHASH_ASM_ARM) if (OPENSSL_armcap_P & ARMV7_NEON) { ctx->gmult = gcm_gmult_neon; ctx->ghash = gcm_ghash_neon; } else { gcm_init_4bit(ctx->Htable,ctx->H.u); ctx->gmult = gcm_gmult_4bit; ctx->ghash = gcm_ghash_4bit; } # else gcm_init_4bit(ctx->Htable,ctx->H.u); # endif #endif } void CRYPTO_gcm128_setiv(GCM128_CONTEXT *ctx,const unsigned char *iv,size_t len) { unsigned int ctr; #ifdef GCM_FUNCREF_4BIT void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16]) = ctx->gmult; #endif ctx->Yi.u[0] = 0; ctx->Yi.u[1] = 0; ctx->Xi.u[0] = 0; ctx->Xi.u[1] = 0; ctx->len.u[0] = 0; /* AAD length */ ctx->len.u[1] = 0; /* message length */ ctx->ares = 0; ctx->mres = 0; if (len==12) { memcpy(ctx->Yi.c,iv,12); ctx->Yi.c[15]=1; ctr=1; } else { size_t i; u64 len0 = len; while (len>=16) { for (i=0; i<16; ++i) ctx->Yi.c[i] ^= iv[i]; GCM_MUL(ctx,Yi); iv += 16; len -= 16; } if (len) { for (i=0; iYi.c[i] ^= iv[i]; GCM_MUL(ctx,Yi); } len0 <<= 3; if (BYTE_ORDER == LITTLE_ENDIAN) { #ifdef BSWAP8 ctx->Yi.u[1] ^= BSWAP8(len0); #else ctx->Yi.c[8] ^= (u8)(len0>>56); ctx->Yi.c[9] ^= (u8)(len0>>48); ctx->Yi.c[10] ^= (u8)(len0>>40); ctx->Yi.c[11] ^= (u8)(len0>>32); ctx->Yi.c[12] ^= (u8)(len0>>24); ctx->Yi.c[13] ^= (u8)(len0>>16); ctx->Yi.c[14] ^= (u8)(len0>>8); ctx->Yi.c[15] ^= (u8)(len0); #endif } else ctx->Yi.u[1] ^= len0; GCM_MUL(ctx,Yi); if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctr = BSWAP4(ctx->Yi.d[3]); #else ctr = GETU32(ctx->Yi.c+12); #endif else ctr = ctx->Yi.d[3]; } (*ctx->block)(ctx->Yi.c,ctx->EK0.c,ctx->key); ++ctr; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; } int CRYPTO_gcm128_aad(GCM128_CONTEXT *ctx,const unsigned char *aad,size_t len) { size_t i; unsigned int n; u64 alen = ctx->len.u[0]; #ifdef GCM_FUNCREF_4BIT void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16]) = ctx->gmult; # ifdef GHASH void (*gcm_ghash_p)(u64 Xi[2],const u128 Htable[16], const u8 *inp,size_t len) = ctx->ghash; # endif #endif if (ctx->len.u[1]) return -2; alen += len; if (alen>(U64(1)<<61) || (sizeof(len)==8 && alenlen.u[0] = alen; n = ctx->ares; if (n) { while (n && len) { ctx->Xi.c[n] ^= *(aad++); --len; n = (n+1)%16; } if (n==0) GCM_MUL(ctx,Xi); else { ctx->ares = n; return 0; } } #ifdef GHASH if ((i = (len&(size_t)-16))) { GHASH(ctx,aad,i); aad += i; len -= i; } #else while (len>=16) { for (i=0; i<16; ++i) ctx->Xi.c[i] ^= aad[i]; GCM_MUL(ctx,Xi); aad += 16; len -= 16; } #endif if (len) { n = (unsigned int)len; for (i=0; iXi.c[i] ^= aad[i]; } ctx->ares = n; return 0; } int CRYPTO_gcm128_encrypt(GCM128_CONTEXT *ctx, const unsigned char *in, unsigned char *out, size_t len) { unsigned int n, ctr; size_t i; u64 mlen = ctx->len.u[1]; block128_f block = ctx->block; void *key = ctx->key; #ifdef GCM_FUNCREF_4BIT void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16]) = ctx->gmult; # ifdef GHASH void (*gcm_ghash_p)(u64 Xi[2],const u128 Htable[16], const u8 *inp,size_t len) = ctx->ghash; # endif #endif mlen += len; if (mlen>((U64(1)<<36)-32) || (sizeof(len)==8 && mlenlen.u[1] = mlen; if (ctx->ares) { /* First call to encrypt finalizes GHASH(AAD) */ GCM_MUL(ctx,Xi); ctx->ares = 0; } if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctr = BSWAP4(ctx->Yi.d[3]); #else ctr = GETU32(ctx->Yi.c+12); #endif else ctr = ctx->Yi.d[3]; n = ctx->mres; #if !defined(OPENSSL_SMALL_FOOTPRINT) if (16%sizeof(size_t) == 0) do { /* always true actually */ if (n) { while (n && len) { ctx->Xi.c[n] ^= *(out++) = *(in++)^ctx->EKi.c[n]; --len; n = (n+1)%16; } if (n==0) GCM_MUL(ctx,Xi); else { ctx->mres = n; return 0; } } #ifdef __STRICT_ALIGNMENT if (((size_t)in|(size_t)out)%sizeof(size_t) != 0) break; #endif #if defined(GHASH) && defined(GHASH_CHUNK) while (len>=GHASH_CHUNK) { size_t j=GHASH_CHUNK; while (j) { size_t *out_t=(size_t *)out; const size_t *in_t=(const size_t *)in; (*block)(ctx->Yi.c,ctx->EKi.c,key); ++ctr; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; for (i=0; i<16/sizeof(size_t); ++i) out_t[i] = in_t[i] ^ ctx->EKi.t[i]; out += 16; in += 16; j -= 16; } GHASH(ctx,out-GHASH_CHUNK,GHASH_CHUNK); len -= GHASH_CHUNK; } if ((i = (len&(size_t)-16))) { size_t j=i; while (len>=16) { size_t *out_t=(size_t *)out; const size_t *in_t=(const size_t *)in; (*block)(ctx->Yi.c,ctx->EKi.c,key); ++ctr; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; for (i=0; i<16/sizeof(size_t); ++i) out_t[i] = in_t[i] ^ ctx->EKi.t[i]; out += 16; in += 16; len -= 16; } GHASH(ctx,out-j,j); } #else while (len>=16) { size_t *out_t=(size_t *)out; const size_t *in_t=(const size_t *)in; (*block)(ctx->Yi.c,ctx->EKi.c,key); ++ctr; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; for (i=0; i<16/sizeof(size_t); ++i) ctx->Xi.t[i] ^= out_t[i] = in_t[i]^ctx->EKi.t[i]; GCM_MUL(ctx,Xi); out += 16; in += 16; len -= 16; } #endif if (len) { (*block)(ctx->Yi.c,ctx->EKi.c,key); ++ctr; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; while (len--) { ctx->Xi.c[n] ^= out[n] = in[n]^ctx->EKi.c[n]; ++n; } } ctx->mres = n; return 0; } while(0); #endif for (i=0;iYi.c,ctx->EKi.c,key); ++ctr; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; } ctx->Xi.c[n] ^= out[i] = in[i]^ctx->EKi.c[n]; n = (n+1)%16; if (n==0) GCM_MUL(ctx,Xi); } ctx->mres = n; return 0; } int CRYPTO_gcm128_decrypt(GCM128_CONTEXT *ctx, const unsigned char *in, unsigned char *out, size_t len) { unsigned int n, ctr; size_t i; u64 mlen = ctx->len.u[1]; block128_f block = ctx->block; void *key = ctx->key; #ifdef GCM_FUNCREF_4BIT void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16]) = ctx->gmult; # ifdef GHASH void (*gcm_ghash_p)(u64 Xi[2],const u128 Htable[16], const u8 *inp,size_t len) = ctx->ghash; # endif #endif mlen += len; if (mlen>((U64(1)<<36)-32) || (sizeof(len)==8 && mlenlen.u[1] = mlen; if (ctx->ares) { /* First call to decrypt finalizes GHASH(AAD) */ GCM_MUL(ctx,Xi); ctx->ares = 0; } if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctr = BSWAP4(ctx->Yi.d[3]); #else ctr = GETU32(ctx->Yi.c+12); #endif else ctr = ctx->Yi.d[3]; n = ctx->mres; #if !defined(OPENSSL_SMALL_FOOTPRINT) if (16%sizeof(size_t) == 0) do { /* always true actually */ if (n) { while (n && len) { u8 c = *(in++); *(out++) = c^ctx->EKi.c[n]; ctx->Xi.c[n] ^= c; --len; n = (n+1)%16; } if (n==0) GCM_MUL (ctx,Xi); else { ctx->mres = n; return 0; } } #ifdef __STRICT_ALIGNMENT if (((size_t)in|(size_t)out)%sizeof(size_t) != 0) break; #endif #if defined(GHASH) && defined(GHASH_CHUNK) while (len>=GHASH_CHUNK) { size_t j=GHASH_CHUNK; GHASH(ctx,in,GHASH_CHUNK); while (j) { size_t *out_t=(size_t *)out; const size_t *in_t=(const size_t *)in; (*block)(ctx->Yi.c,ctx->EKi.c,key); ++ctr; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; for (i=0; i<16/sizeof(size_t); ++i) out_t[i] = in_t[i]^ctx->EKi.t[i]; out += 16; in += 16; j -= 16; } len -= GHASH_CHUNK; } if ((i = (len&(size_t)-16))) { GHASH(ctx,in,i); while (len>=16) { size_t *out_t=(size_t *)out; const size_t *in_t=(const size_t *)in; (*block)(ctx->Yi.c,ctx->EKi.c,key); ++ctr; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; for (i=0; i<16/sizeof(size_t); ++i) out_t[i] = in_t[i]^ctx->EKi.t[i]; out += 16; in += 16; len -= 16; } } #else while (len>=16) { size_t *out_t=(size_t *)out; const size_t *in_t=(const size_t *)in; (*block)(ctx->Yi.c,ctx->EKi.c,key); ++ctr; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; for (i=0; i<16/sizeof(size_t); ++i) { size_t c = in[i]; out[i] = c^ctx->EKi.t[i]; ctx->Xi.t[i] ^= c; } GCM_MUL(ctx,Xi); out += 16; in += 16; len -= 16; } #endif if (len) { (*block)(ctx->Yi.c,ctx->EKi.c,key); ++ctr; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; while (len--) { u8 c = in[n]; ctx->Xi.c[n] ^= c; out[n] = c^ctx->EKi.c[n]; ++n; } } ctx->mres = n; return 0; } while(0); #endif for (i=0;iYi.c,ctx->EKi.c,key); ++ctr; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; } c = in[i]; out[i] = c^ctx->EKi.c[n]; ctx->Xi.c[n] ^= c; n = (n+1)%16; if (n==0) GCM_MUL(ctx,Xi); } ctx->mres = n; return 0; } int CRYPTO_gcm128_encrypt_ctr32(GCM128_CONTEXT *ctx, const unsigned char *in, unsigned char *out, size_t len, ctr128_f stream) { unsigned int n, ctr; size_t i; u64 mlen = ctx->len.u[1]; void *key = ctx->key; #ifdef GCM_FUNCREF_4BIT void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16]) = ctx->gmult; # ifdef GHASH void (*gcm_ghash_p)(u64 Xi[2],const u128 Htable[16], const u8 *inp,size_t len) = ctx->ghash; # endif #endif mlen += len; if (mlen>((U64(1)<<36)-32) || (sizeof(len)==8 && mlenlen.u[1] = mlen; if (ctx->ares) { /* First call to encrypt finalizes GHASH(AAD) */ GCM_MUL(ctx,Xi); ctx->ares = 0; } if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctr = BSWAP4(ctx->Yi.d[3]); #else ctr = GETU32(ctx->Yi.c+12); #endif else ctr = ctx->Yi.d[3]; n = ctx->mres; if (n) { while (n && len) { ctx->Xi.c[n] ^= *(out++) = *(in++)^ctx->EKi.c[n]; --len; n = (n+1)%16; } if (n==0) GCM_MUL(ctx,Xi); else { ctx->mres = n; return 0; } } #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT) while (len>=GHASH_CHUNK) { (*stream)(in,out,GHASH_CHUNK/16,key,ctx->Yi.c); ctr += GHASH_CHUNK/16; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; GHASH(ctx,out,GHASH_CHUNK); out += GHASH_CHUNK; in += GHASH_CHUNK; len -= GHASH_CHUNK; } #endif if ((i = (len&(size_t)-16))) { size_t j=i/16; (*stream)(in,out,j,key,ctx->Yi.c); ctr += (unsigned int)j; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; in += i; len -= i; #if defined(GHASH) GHASH(ctx,out,i); out += i; #else while (j--) { for (i=0;i<16;++i) ctx->Xi.c[i] ^= out[i]; GCM_MUL(ctx,Xi); out += 16; } #endif } if (len) { (*ctx->block)(ctx->Yi.c,ctx->EKi.c,key); ++ctr; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; while (len--) { ctx->Xi.c[n] ^= out[n] = in[n]^ctx->EKi.c[n]; ++n; } } ctx->mres = n; return 0; } int CRYPTO_gcm128_decrypt_ctr32(GCM128_CONTEXT *ctx, const unsigned char *in, unsigned char *out, size_t len,ctr128_f stream) { unsigned int n, ctr; size_t i; u64 mlen = ctx->len.u[1]; void *key = ctx->key; #ifdef GCM_FUNCREF_4BIT void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16]) = ctx->gmult; # ifdef GHASH void (*gcm_ghash_p)(u64 Xi[2],const u128 Htable[16], const u8 *inp,size_t len) = ctx->ghash; # endif #endif mlen += len; if (mlen>((U64(1)<<36)-32) || (sizeof(len)==8 && mlenlen.u[1] = mlen; if (ctx->ares) { /* First call to decrypt finalizes GHASH(AAD) */ GCM_MUL(ctx,Xi); ctx->ares = 0; } if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctr = BSWAP4(ctx->Yi.d[3]); #else ctr = GETU32(ctx->Yi.c+12); #endif else ctr = ctx->Yi.d[3]; n = ctx->mres; if (n) { while (n && len) { u8 c = *(in++); *(out++) = c^ctx->EKi.c[n]; ctx->Xi.c[n] ^= c; --len; n = (n+1)%16; } if (n==0) GCM_MUL (ctx,Xi); else { ctx->mres = n; return 0; } } #if defined(GHASH) && !defined(OPENSSL_SMALL_FOOTPRINT) while (len>=GHASH_CHUNK) { GHASH(ctx,in,GHASH_CHUNK); (*stream)(in,out,GHASH_CHUNK/16,key,ctx->Yi.c); ctr += GHASH_CHUNK/16; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; out += GHASH_CHUNK; in += GHASH_CHUNK; len -= GHASH_CHUNK; } #endif if ((i = (len&(size_t)-16))) { size_t j=i/16; #if defined(GHASH) GHASH(ctx,in,i); #else while (j--) { size_t k; for (k=0;k<16;++k) ctx->Xi.c[k] ^= in[k]; GCM_MUL(ctx,Xi); in += 16; } j = i/16; in -= i; #endif (*stream)(in,out,j,key,ctx->Yi.c); ctr += (unsigned int)j; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; out += i; in += i; len -= i; } if (len) { (*ctx->block)(ctx->Yi.c,ctx->EKi.c,key); ++ctr; if (BYTE_ORDER == LITTLE_ENDIAN) #ifdef BSWAP4 ctx->Yi.d[3] = BSWAP4(ctr); #else PUTU32(ctx->Yi.c+12,ctr); #endif else ctx->Yi.d[3] = ctr; while (len--) { u8 c = in[n]; ctx->Xi.c[n] ^= c; out[n] = c^ctx->EKi.c[n]; ++n; } } ctx->mres = n; return 0; } int CRYPTO_gcm128_finish(GCM128_CONTEXT *ctx,const unsigned char *tag, size_t len) { u64 alen = ctx->len.u[0]<<3; u64 clen = ctx->len.u[1]<<3; #ifdef GCM_FUNCREF_4BIT void (*gcm_gmult_p)(u64 Xi[2],const u128 Htable[16]) = ctx->gmult; #endif if (ctx->mres || ctx->ares) GCM_MUL(ctx,Xi); if (BYTE_ORDER == LITTLE_ENDIAN) { #ifdef BSWAP8 alen = BSWAP8(alen); clen = BSWAP8(clen); #else u8 *p = ctx->len.c; ctx->len.u[0] = alen; ctx->len.u[1] = clen; alen = (u64)GETU32(p) <<32|GETU32(p+4); clen = (u64)GETU32(p+8)<<32|GETU32(p+12); #endif } ctx->Xi.u[0] ^= alen; ctx->Xi.u[1] ^= clen; GCM_MUL(ctx,Xi); ctx->Xi.u[0] ^= ctx->EK0.u[0]; ctx->Xi.u[1] ^= ctx->EK0.u[1]; if (tag && len<=sizeof(ctx->Xi)) return memcmp(ctx->Xi.c,tag,len); else return -1; } void CRYPTO_gcm128_tag(GCM128_CONTEXT *ctx, unsigned char *tag, size_t len) { CRYPTO_gcm128_finish(ctx, NULL, 0); memcpy(tag, ctx->Xi.c, len<=sizeof(ctx->Xi.c)?len:sizeof(ctx->Xi.c)); } GCM128_CONTEXT *CRYPTO_gcm128_new(void *key, block128_f block) { GCM128_CONTEXT *ret; if ((ret = malloc(sizeof(GCM128_CONTEXT)))) CRYPTO_gcm128_init(ret,key,block); return ret; } void CRYPTO_gcm128_release(GCM128_CONTEXT *ctx) { if (ctx) { explicit_bzero(ctx,sizeof(*ctx)); free(ctx); } }