const AES_KEY *key, unsigned char iv[16],
SHA_CTX *ctx,const void *in0);
+void aesni256_cbc_sha1_dec (const void *inp, void *out, size_t blocks,
+ const AES_KEY *key, unsigned char iv[16],
+ SHA_CTX *ctx,const void *in0);
+
#define data(ctx) ((EVP_AES_HMAC_SHA1 *)(ctx)->cipher_data)
static int aesni_cbc_hmac_sha1_init_key(EVP_CIPHER_CTX *ctx,
}
#define STITCHED_CALL
+#undef STITCHED_DECRYPT_CALL
#if !defined(STITCHED_CALL)
#define aes_off 0
u32 d[32];
u8 c[128]; } blocks[8];
SHA1_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 = (SHA1_MB_CTX *)(storage+32-((size_t)storage%32)); /* align */
last -= x4-1;
}
+ packlen = 5+16+((frag+20+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);
+ seqnum = BSWAP8(blocks[0].q[0]);
+#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.h0;
ctx->B[i] = key->md.h1;
/* fix seqnum */
#if defined(BSWAP8)
- blocks[i].q[0] = BSWAP8(BSWAP8(*(u64*)key->md.data)+i);
+ 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];
/* hash 13-byte headers and first 64-13 bytes of inputs */
sha1_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 {
+ sha1_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
sha1_multi_block(ctx,hash_d,n4x);
memset(blocks,0,sizeof(blocks));
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;
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.h0;
blocks[i].d[1] = BSWAP4(ctx->B[i]); ctx->B[i] = key->tail.h1;
blocks[i].d[2] = BSWAP4(ctx->C[i]); ctx->C[i] = key->tail.h2;
blocks[i].d[4] = BSWAP4(ctx->E[i]); ctx->E[i] = key->tail.h4;
blocks[i].c[20] = 0x80;
blocks[i].d[15] = BSWAP4((64+20)*8);
+#else
+ PUTU32(blocks[i].c+0,ctx->A[i]); ctx->A[i] = key->tail.h0;
+ PUTU32(blocks[i].c+4,ctx->B[i]); ctx->B[i] = key->tail.h1;
+ PUTU32(blocks[i].c+8,ctx->C[i]); ctx->C[i] = key->tail.h2;
+ PUTU32(blocks[i].c+12,ctx->D[i]); ctx->D[i] = key->tail.h3;
+ PUTU32(blocks[i].c+16,ctx->E[i]); ctx->E[i] = key->tail.h4;
+ blocks[i].c[20] = 0x80;
+ PUTU32(blocks[i].c+60,(64+20)*8);
+#endif
edges[i].ptr = blocks[i].c;
edges[i].blocks = 1;
}
/* finalize MACs */
sha1_multi_block(ctx,edges,n4x);
- packlen = 5+16+((frag+20+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]);
+ 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]);
out += 20;
len += 20;
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 */
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);
/* arrange cache line alignment */
pmac = (void *)(((size_t)mac.c+31)&((size_t)0-32));
- /* decrypt HMAC|padding at once */
- aesni_cbc_encrypt(in,out,len,
- &key->ks,ctx->iv,0);
-
- if (plen) { /* "TLS" mode of operation */
+ if (plen != NO_PAYLOAD_LENGTH) { /* "TLS" mode of operation */
size_t inp_len, mask, j, i;
unsigned int res, maxpad, pad, bitlen;
int ret = 1;
union { unsigned int u[SHA_LBLOCK];
unsigned char c[SHA_CBLOCK]; }
*data = (void *)key->md.data;
+#if defined(STITCHED_DECRYPT_CALL)
+ unsigned char tail_iv[AES_BLOCK_SIZE];
+ int stitch=0;
+#endif
if ((key->aux.tls_aad[plen-4]<<8|key->aux.tls_aad[plen-3])
- >= TLS1_1_VERSION)
- iv = AES_BLOCK_SIZE;
-
- if (len<(iv+SHA_DIGEST_LENGTH+1))
+ >= TLS1_1_VERSION) {
+ if (len<(AES_BLOCK_SIZE+SHA_DIGEST_LENGTH+1))
+ return 0;
+
+ /* omit explicit iv */
+ memcpy(ctx->iv,in,AES_BLOCK_SIZE);
+ in += AES_BLOCK_SIZE;
+ out += AES_BLOCK_SIZE;
+ len -= AES_BLOCK_SIZE;
+ }
+ else if (len<(SHA_DIGEST_LENGTH+1))
return 0;
- /* omit explicit iv */
- out += iv;
- len -= iv;
+#if defined(STITCHED_DECRYPT_CALL)
+ if (len>=1024 && ctx->key_len==32) {
+ /* decrypt last block */
+ memcpy(tail_iv,in+len-2*AES_BLOCK_SIZE,AES_BLOCK_SIZE);
+ aesni_cbc_encrypt(in+len-AES_BLOCK_SIZE,
+ out+len-AES_BLOCK_SIZE,AES_BLOCK_SIZE,
+ &key->ks,tail_iv,0);
+ stitch=1;
+ } else
+#endif
+ /* decrypt HMAC|padding at once */
+ aesni_cbc_encrypt(in,out,len,
+ &key->ks,ctx->iv,0);
/* figure out payload length */
pad = out[len-1];
key->md = key->head;
SHA1_Update(&key->md,key->aux.tls_aad,plen);
+#if defined(STITCHED_DECRYPT_CALL)
+ if (stitch) {
+ blocks = (len-(256+32+SHA_CBLOCK))/SHA_CBLOCK;
+ aes_off = len-AES_BLOCK_SIZE-blocks*SHA_CBLOCK;
+ sha_off = SHA_CBLOCK-plen;
+
+ aesni_cbc_encrypt(in,out,aes_off,
+ &key->ks,ctx->iv,0);
+
+ SHA1_Update(&key->md,out,sha_off);
+ aesni256_cbc_sha1_dec(in+aes_off,
+ out+aes_off,blocks,&key->ks,ctx->iv,
+ &key->md,out+sha_off);
+
+ sha_off += blocks*=SHA_CBLOCK;
+ out += sha_off;
+ len -= sha_off;
+ inp_len -= sha_off;
+
+ key->md.Nl += (blocks<<3); /* at most 18 bits */
+ memcpy(ctx->iv,tail_iv,AES_BLOCK_SIZE);
+ }
+#endif
+
#if 1
len -= SHA_DIGEST_LENGTH; /* amend mac */
if (len>=(256+SHA_CBLOCK)) {
#endif
return ret;
} else {
+#if defined(STITCHED_DECRYPT_CALL)
+ if (len>=1024 && ctx->key_len==32) {
+ if (sha_off%=SHA_CBLOCK)
+ blocks = (len-3*SHA_CBLOCK)/SHA_CBLOCK;
+ else
+ blocks = (len-2*SHA_CBLOCK)/SHA_CBLOCK;
+ aes_off = len-blocks*SHA_CBLOCK;
+
+ aesni_cbc_encrypt(in,out,aes_off,
+ &key->ks,ctx->iv,0);
+ SHA1_Update(&key->md,out,sha_off);
+ aesni256_cbc_sha1_dec(in+aes_off,
+ out+aes_off,blocks,&key->ks,ctx->iv,
+ &key->md,out+sha_off);
+
+ sha_off += blocks*=SHA_CBLOCK;
+ out += sha_off;
+ len -= sha_off;
+
+ key->md.Nh += blocks>>29;
+ key->md.Nl += blocks<<=3;
+ if (key->md.Nl<(unsigned int)blocks) key->md.Nh++;
+ } else
+#endif
+ /* decrypt HMAC|padding at once */
+ aesni_cbc_encrypt(in,out,len,
+ &key->ks,ctx->iv,0);
+
SHA1_Update(&key->md,out,len);
}
}
}
}
#if !defined(OPENSSL_NO_MULTIBLOCK) && EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK
+ case EVP_CTRL_TLS1_1_MULTIBLOCK_MAX_BUFSIZE:
+ return (int)(5+16+((arg+20+16)&-16));
case EVP_CTRL_TLS1_1_MULTIBLOCK_AAD:
{
EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param =
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<(int)sizeof(EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM)) return -1;
inp_len = param->inp[11]<<8|param->inp[12];
if ((param->inp[9]<<8|param->inp[10]) < TLS1_1_VERSION)
return -1;
- if (inp_len<4096) return 0; /* too short */
+ if (inp_len)
+ {
+ if (inp_len<4096) return 0; /* too short */
- if (inp_len>=8192 && OPENSSL_ia32cap_P[2]&(1<<5))
- n4x=2; /* AVX2 */
+ if (inp_len>=8192 && OPENSSL_ia32cap_P[2]&(1<<5))
+ n4x=2; /* AVX2 */
+ }
+ else if ((n4x=param->interleave/4) && n4x<=2)
+ inp_len = param->len;
+ else
+ return -1;
key->md = key->head;
SHA1_Update(&key->md,param->inp,13);