crypto/evp/e_aes_cbc_hmac_sha[1|256].c: fix compiler warnings.
[openssl.git] / crypto / evp / e_aes_cbc_hmac_sha256.c
index 95bdd42..e315603 100644 (file)
@@ -201,13 +201,16 @@ static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
                u32     d[32];
                u8      c[128]; } blocks[8];
        SHA256_MB_CTX   *ctx;
-       unsigned int    frag, last, packlen, i, x4=4*n4x;
+       unsigned int    frag, last, packlen, i, x4=4*n4x, minblocks, processed=0;
        size_t          ret = 0;
        u8              *IVs;
 #if defined(BSWAP8)
        u64             seqnum;
 #endif
 
+       if (RAND_bytes((IVs=blocks[0].c),16*x4)<=0)     /* ask for IVs in bulk */
+               return 0;
+
        ctx = (SHA256_MB_CTX *)(storage+32-((size_t)storage%32));       /* align */
 
        frag = (unsigned int)inp_len>>(1+n4x);
@@ -217,8 +220,21 @@ static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
                last -= x4-1;
        }
 
+       packlen = 5+16+((frag+32+16)&-16);
+
+       /* populate descriptors with pointers and IVs */
        hash_d[0].ptr = inp;
-       for (i=1;i<x4;i++)      hash_d[i].ptr = hash_d[i-1].ptr+frag;
+       ciph_d[0].inp = inp;
+       ciph_d[0].out = out+5+16;       /* 5+16 is place for header and explicit IV */
+       memcpy(ciph_d[0].out-16,IVs,16);
+       memcpy(ciph_d[0].iv,IVs,16);    IVs += 16;
+
+       for (i=1;i<x4;i++) {
+               ciph_d[i].inp = hash_d[i].ptr = hash_d[i-1].ptr+frag;
+               ciph_d[i].out = ciph_d[i-1].out+packlen;
+               memcpy(ciph_d[i].out-16,IVs,16);
+               memcpy(ciph_d[i].iv,IVs,16);    IVs+=16;
+       }
 
 #if defined(BSWAP8)
        memcpy(blocks[0].c,key->md.data,8);
@@ -226,6 +242,9 @@ static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
 #endif
        for (i=0;i<x4;i++) {
                unsigned int len = (i==(x4-1)?last:frag);
+#if !defined(BSWAP8)
+               unsigned int carry, j;
+#endif
 
                ctx->A[i] = key->md.h[0];
                ctx->B[i] = key->md.h[1];
@@ -240,13 +259,9 @@ static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
 #if defined(BSWAP8)
                blocks[i].q[0] = BSWAP8(seqnum+i);
 #else
-               blocks[i].c[7] += ((u8*)key->md.data)[7]+i;
-               if (blocks[i].c[7] < i) {
-                       int j;
-
-                       for (j=6;j>=0;j--) {
-                               if (blocks[i].c[j]=((u8*)key->md.data)[j]+1) break;
-                       }
+               for (carry=i,j=8;j--;) {
+                       blocks[i].c[j] = ((u8*)key->md.data)[j]+carry;
+                       carry = (blocks[i].c[j]-carry)>>(sizeof(carry)*8-1);
                }
 #endif
                blocks[i].c[8] = ((u8*)key->md.data)[8];
@@ -267,6 +282,39 @@ static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
        /* hash 13-byte headers and first 64-13 bytes of inputs */
        sha256_multi_block(ctx,edges,n4x);
        /* hash bulk inputs */
+#define        MAXCHUNKSIZE    2048
+#if    MAXCHUNKSIZE%64
+#error "MAXCHUNKSIZE is not divisible by 64"
+#elif  MAXCHUNKSIZE
+       /* goal is to minimize pressure on L1 cache by moving
+        * in shorter steps, so that hashed data is still in
+        * the cache by the time we encrypt it */
+       minblocks = ((frag<=last ? frag : last)-(64-13))/64;
+       if (minblocks>MAXCHUNKSIZE/64) {
+               for (i=0;i<x4;i++) {
+                       edges[i].ptr     = hash_d[i].ptr;
+                       edges[i].blocks  = MAXCHUNKSIZE/64;
+                       ciph_d[i].blocks = MAXCHUNKSIZE/16;
+               }
+               do {
+                       sha256_multi_block(ctx,edges,n4x);
+                       aesni_multi_cbc_encrypt(ciph_d,&key->ks,n4x);
+
+                       for (i=0;i<x4;i++) {
+                               edges[i].ptr     = hash_d[i].ptr += MAXCHUNKSIZE;
+                               hash_d[i].blocks -= MAXCHUNKSIZE/64;
+                               edges[i].blocks  = MAXCHUNKSIZE/64;
+                               ciph_d[i].inp    += MAXCHUNKSIZE;
+                               ciph_d[i].out    += MAXCHUNKSIZE;
+                               ciph_d[i].blocks = MAXCHUNKSIZE/16;
+                               memcpy(ciph_d[i].iv,ciph_d[i].out-16,16);
+                       }
+                       processed += MAXCHUNKSIZE;
+                       minblocks -= MAXCHUNKSIZE/64;
+               } while (minblocks>MAXCHUNKSIZE/64);
+       }
+#endif
+#undef MAXCHUNKSIZE
        sha256_multi_block(ctx,hash_d,n4x);
 
        memset(blocks,0,sizeof(blocks));
@@ -275,16 +323,24 @@ static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
                                        off = hash_d[i].blocks*64;
                const unsigned char    *ptr = hash_d[i].ptr+off;
 
-               off = len-(64-13)-off;  /* remainder actually */
+               off = (len-processed)-(64-13)-off;      /* remainder actually */
                memcpy(blocks[i].c,ptr,off);
                blocks[i].c[off]=0x80;
                len += 64+13;           /* 64 is HMAC header */
                len *= 8;               /* convert to bits */
                if (off<(64-8)) {
+#ifdef BSWAP4
                        blocks[i].d[15] = BSWAP4(len);
+#else
+                       PUTU32(blocks[i].c+60,len);
+#endif
                        edges[i].blocks = 1;                    
                } else {
+#ifdef BSWAP4
                        blocks[i].d[31] = BSWAP4(len);
+#else
+                       PUTU32(blocks[i].c+124,len);
+#endif
                        edges[i].blocks = 2;
                }
                edges[i].ptr = blocks[i].c;
@@ -295,6 +351,7 @@ static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
 
        memset(blocks,0,sizeof(blocks));
        for (i=0;i<x4;i++) {
+#ifdef BSWAP4
                blocks[i].d[0] = BSWAP4(ctx->A[i]);     ctx->A[i] = key->tail.h[0];
                blocks[i].d[1] = BSWAP4(ctx->B[i]);     ctx->B[i] = key->tail.h[1];
                blocks[i].d[2] = BSWAP4(ctx->C[i]);     ctx->C[i] = key->tail.h[2];
@@ -305,6 +362,18 @@ static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
                blocks[i].d[7] = BSWAP4(ctx->H[i]);     ctx->H[i] = key->tail.h[7];
                blocks[i].c[32] = 0x80;
                blocks[i].d[15] = BSWAP4((64+32)*8);
+#else
+               PUTU32(blocks[i].c+0,ctx->A[i]);        ctx->A[i] = key->tail.h[0];
+               PUTU32(blocks[i].c+4,ctx->B[i]);        ctx->B[i] = key->tail.h[1];
+               PUTU32(blocks[i].c+8,ctx->C[i]);        ctx->C[i] = key->tail.h[2];
+               PUTU32(blocks[i].c+12,ctx->D[i]);       ctx->D[i] = key->tail.h[3];
+               PUTU32(blocks[i].c+16,ctx->E[i]);       ctx->E[i] = key->tail.h[4];
+               PUTU32(blocks[i].c+20,ctx->F[i]);       ctx->F[i] = key->tail.h[5];
+               PUTU32(blocks[i].c+24,ctx->G[i]);       ctx->G[i] = key->tail.h[6];
+               PUTU32(blocks[i].c+28,ctx->H[i]);       ctx->H[i] = key->tail.h[7];
+               blocks[i].c[32] = 0x80;
+               PUTU32(blocks[i].c+60,(64+32)*8);
+#endif
                edges[i].ptr = blocks[i].c;
                edges[i].blocks = 1;
        }
@@ -312,33 +381,24 @@ static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
        /* finalize MACs */
        sha256_multi_block(ctx,edges,n4x);
 
-       packlen = 5+16+((frag+32+16)&-16);
-
-       out += (packlen<<(1+n4x))-packlen;
-       inp += (frag<<(1+n4x))-frag;
-
-       RAND_bytes((IVs=blocks[0].c),16*x4);    /* ask for IVs in bulk */
-
-       for (i=x4-1;;i--) {
+       for (i=0;i<x4;i++) {
                unsigned int len = (i==(x4-1)?last:frag), pad, j;
                unsigned char *out0 = out;
 
-               out += 5+16;            /* place for header and explicit IV */
-               ciph_d[i].inp = out;
-               ciph_d[i].out = out;
+               memcpy(ciph_d[i].out,ciph_d[i].inp,len-processed);
+               ciph_d[i].inp = ciph_d[i].out;
 
-               memmove(out,inp,len);
-               out += len;
+               out += 5+16+len;
 
                /* write MAC */
-               ((u32 *)out)[0] = BSWAP4(ctx->A[i]);
-               ((u32 *)out)[1] = BSWAP4(ctx->B[i]);
-               ((u32 *)out)[2] = BSWAP4(ctx->C[i]);
-               ((u32 *)out)[3] = BSWAP4(ctx->D[i]);
-               ((u32 *)out)[4] = BSWAP4(ctx->E[i]);
-               ((u32 *)out)[5] = BSWAP4(ctx->F[i]);
-               ((u32 *)out)[6] = BSWAP4(ctx->G[i]);
-               ((u32 *)out)[7] = BSWAP4(ctx->H[i]);
+               PUTU32(out+0,ctx->A[i]);
+               PUTU32(out+4,ctx->B[i]);
+               PUTU32(out+8,ctx->C[i]);
+               PUTU32(out+12,ctx->D[i]);
+               PUTU32(out+16,ctx->E[i]);
+               PUTU32(out+20,ctx->F[i]);
+               PUTU32(out+24,ctx->G[i]);
+               PUTU32(out+28,ctx->H[i]);
                out += 32;
                len += 32;
 
@@ -347,7 +407,7 @@ static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
                for (j=0;j<=pad;j++) *(out++) = pad;
                len += pad+1;
 
-               ciph_d[i].blocks = len/16;
+               ciph_d[i].blocks = (len-processed)/16;
                len += 16;      /* account for explicit iv */
 
                /* arrange header */
@@ -357,17 +417,8 @@ static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
                out0[3] = (u8)(len>>8);
                out0[4] = (u8)(len);
 
-               /* explicit iv */
-               memcpy(ciph_d[i].iv, IVs, 16);
-               memcpy(&out0[5],     IVs, 16);
-
                ret += len+5;
-
-               if (i==0) break;
-
-               out = out0-packlen;
-               inp -= frag;
-               IVs += 16;
+               inp += frag;
        }
 
        aesni_multi_cbc_encrypt(ciph_d,&key->ks,n4x);
@@ -674,6 +725,7 @@ static int aesni_cbc_hmac_sha256_cipher(EVP_CIPHER_CTX *ctx, unsigned char *out,
 static int aesni_cbc_hmac_sha256_ctrl(EVP_CIPHER_CTX *ctx, int type, int arg, void *ptr)
        {
        EVP_AES_HMAC_SHA256 *key = data(ctx);
+       unsigned int u_arg = (unsigned int)arg;
 
        switch (type)
                {
@@ -684,7 +736,10 @@ static int aesni_cbc_hmac_sha256_ctrl(EVP_CIPHER_CTX *ctx, int type, int arg, vo
 
                memset (hmac_key,0,sizeof(hmac_key));
 
-               if (arg > (int)sizeof(hmac_key)) {
+               if (arg < 0)
+                       return -1;
+
+               if (u_arg > sizeof(hmac_key)) {
                        SHA256_Init(&key->head);
                        SHA256_Update(&key->head,ptr,arg);
                        SHA256_Final(hmac_key,&key->head);
@@ -744,7 +799,10 @@ static int aesni_cbc_hmac_sha256_ctrl(EVP_CIPHER_CTX *ctx, int type, int arg, vo
                unsigned int n4x=1, x4;
                unsigned int frag, last, packlen, inp_len;
 
-               if (arg<sizeof(EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM)) return -1;
+               if (arg < 0)
+                       return -1;
+
+               if (u_arg < sizeof(EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM)) return -1;
 
                inp_len = param->inp[11]<<8|param->inp[12];