63c6acda974f1d523a67f9d104912bec9c4c9192
[openssl.git] / crypto / evp / e_aes_cbc_hmac_sha256.c
1 /* ====================================================================
2  * Copyright (c) 2011-2013 The OpenSSL Project.  All rights reserved.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  *
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  *
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in
13  *    the documentation and/or other materials provided with the
14  *    distribution.
15  *
16  * 3. All advertising materials mentioning features or use of this
17  *    software must display the following acknowledgment:
18  *    "This product includes software developed by the OpenSSL Project
19  *    for use in the OpenSSL Toolkit. (http://www.OpenSSL.org/)"
20  *
21  * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to
22  *    endorse or promote products derived from this software without
23  *    prior written permission. For written permission, please contact
24  *    licensing@OpenSSL.org.
25  *
26  * 5. Products derived from this software may not be called "OpenSSL"
27  *    nor may "OpenSSL" appear in their names without prior written
28  *    permission of the OpenSSL Project.
29  *
30  * 6. Redistributions of any form whatsoever must retain the following
31  *    acknowledgment:
32  *    "This product includes software developed by the OpenSSL Project
33  *    for use in the OpenSSL Toolkit (http://www.OpenSSL.org/)"
34  *
35  * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY
36  * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
37  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
38  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE OpenSSL PROJECT OR
39  * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
40  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
41  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
42  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
43  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
44  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
45  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
46  * OF THE POSSIBILITY OF SUCH DAMAGE.
47  * ====================================================================
48  */
49
50 #include <openssl/opensslconf.h>
51
52 #include <stdio.h>
53 #include <string.h>
54
55 #if !defined(OPENSSL_NO_AES) && !defined(OPENSSL_NO_SHA256)
56
57 #include <openssl/evp.h>
58 #include <openssl/objects.h>
59 #include <openssl/aes.h>
60 #include <openssl/sha.h>
61 #include <openssl/rand.h>
62 #include "modes_lcl.h"
63
64 #ifndef EVP_CIPH_FLAG_AEAD_CIPHER
65 #define EVP_CIPH_FLAG_AEAD_CIPHER       0x200000
66 #define EVP_CTRL_AEAD_TLS1_AAD          0x16
67 #define EVP_CTRL_AEAD_SET_MAC_KEY       0x17
68 #endif
69
70 #if !defined(EVP_CIPH_FLAG_DEFAULT_ASN1)
71 #define EVP_CIPH_FLAG_DEFAULT_ASN1 0
72 #endif
73
74 #if !defined(EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK)
75 #define EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK 0
76 #endif
77
78 #define TLS1_1_VERSION 0x0302
79
80 typedef struct
81     {
82     AES_KEY             ks;
83     SHA256_CTX          head,tail,md;
84     size_t              payload_length; /* AAD length in decrypt case */
85     union {
86         unsigned int    tls_ver;
87         unsigned char   tls_aad[16];    /* 13 used */
88     } aux;
89     } EVP_AES_HMAC_SHA256;
90
91 #define NO_PAYLOAD_LENGTH       ((size_t)-1)
92
93 #if     defined(AES_ASM) &&     ( \
94         defined(__x86_64)       || defined(__x86_64__)  || \
95         defined(_M_AMD64)       || defined(_M_X64)      || \
96         defined(__INTEL__)      )
97
98 extern unsigned int OPENSSL_ia32cap_P[3];
99 #define AESNI_CAPABLE   (1<<(57-32))
100
101 int aesni_set_encrypt_key(const unsigned char *userKey, int bits,
102                               AES_KEY *key);
103 int aesni_set_decrypt_key(const unsigned char *userKey, int bits,
104                               AES_KEY *key);
105
106 void aesni_cbc_encrypt(const unsigned char *in,
107                            unsigned char *out,
108                            size_t length,
109                            const AES_KEY *key,
110                            unsigned char *ivec, int enc);
111
112 int aesni_cbc_sha256_enc (const void *inp, void *out, size_t blocks,
113                 const AES_KEY *key, unsigned char iv[16],
114                 SHA256_CTX *ctx,const void *in0);
115
116 #define data(ctx) ((EVP_AES_HMAC_SHA256 *)(ctx)->cipher_data)
117
118 static int aesni_cbc_hmac_sha256_init_key(EVP_CIPHER_CTX *ctx,
119                         const unsigned char *inkey,
120                         const unsigned char *iv, int enc)
121         {
122         EVP_AES_HMAC_SHA256 *key = data(ctx);
123         int ret;
124
125         if (enc)
126                 memset(&key->ks,0,sizeof(key->ks.rd_key)),
127                 ret=aesni_set_encrypt_key(inkey,ctx->key_len*8,&key->ks);
128         else
129                 ret=aesni_set_decrypt_key(inkey,ctx->key_len*8,&key->ks);
130
131         SHA256_Init(&key->head);        /* handy when benchmarking */
132         key->tail = key->head;
133         key->md   = key->head;
134
135         key->payload_length = NO_PAYLOAD_LENGTH;
136
137         return ret<0?0:1;
138         }
139
140 #define STITCHED_CALL
141
142 #if !defined(STITCHED_CALL)
143 #define aes_off 0
144 #endif
145
146 void sha256_block_data_order (void *c,const void *p,size_t len);
147
148 static void sha256_update(SHA256_CTX *c,const void *data,size_t len)
149 {       const unsigned char *ptr = data;
150         size_t res;
151
152         if ((res = c->num)) {
153                 res = SHA256_CBLOCK-res;
154                 if (len<res) res=len;
155                 SHA256_Update (c,ptr,res);
156                 ptr += res;
157                 len -= res;
158         }
159
160         res = len % SHA256_CBLOCK;
161         len -= res;
162
163         if (len) {
164                 sha256_block_data_order(c,ptr,len/SHA256_CBLOCK);
165
166                 ptr += len;
167                 c->Nh += len>>29;
168                 c->Nl += len<<=3;
169                 if (c->Nl<(unsigned int)len) c->Nh++;
170         }
171
172         if (res)
173                 SHA256_Update(c,ptr,res);
174 }
175
176 #ifdef SHA256_Update
177 #undef SHA256_Update
178 #endif
179 #define SHA256_Update sha256_update
180
181 #if !defined(OPENSSL_NO_MULTIBLOCK) && EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK
182
183 typedef struct { unsigned int A[8],B[8],C[8],D[8],E[8],F[8],G[8],H[8]; } SHA256_MB_CTX;
184 typedef struct { const unsigned char *ptr; int blocks;  } HASH_DESC;
185
186 void sha256_multi_block(SHA256_MB_CTX *,const HASH_DESC *,int);
187
188 typedef struct { const unsigned char *inp; unsigned char *out;
189                  int blocks; u64 iv[2]; } CIPH_DESC; 
190
191 void aesni_multi_cbc_encrypt(CIPH_DESC *,void *,int);
192
193 static size_t tls1_1_multi_block_encrypt(EVP_AES_HMAC_SHA256 *key,
194         unsigned char *out, const unsigned char *inp, size_t inp_len,
195         int n4x)        /* n4x is 1 or 2 */
196 {
197         HASH_DESC       hash_d[8], edges[8];
198         CIPH_DESC       ciph_d[8];
199         unsigned char   storage[sizeof(SHA256_MB_CTX)+32];
200         union { u64     q[16];
201                 u32     d[32];
202                 u8      c[128]; } blocks[8];
203         SHA256_MB_CTX   *ctx;
204         unsigned int    frag, last, packlen, i, x4=4*n4x, minblocks, processed=0;
205         size_t          ret = 0;
206         u8              *IVs;
207 #if defined(BSWAP8)
208         u64             seqnum;
209 #else
210         unsigned int    carry,j;
211 #endif
212
213         if (RAND_bytes((IVs=blocks[0].c),16*x4)<=0)     /* ask for IVs in bulk */
214                 return 0;
215
216         ctx = (SHA256_MB_CTX *)(storage+32-((size_t)storage%32));       /* align */
217
218         frag = (unsigned int)inp_len>>(1+n4x);
219         last = (unsigned int)inp_len+frag-(frag<<(1+n4x));
220         if (last>frag && ((last+13+9)%64)<(x4-1)) {
221                 frag++;
222                 last -= x4-1;
223         }
224
225         packlen = 5+16+((frag+32+16)&-16);
226
227         /* populate descriptors with pointers and IVs */
228         hash_d[0].ptr = inp;
229         ciph_d[0].inp = inp;
230         ciph_d[0].out = out+5+16;       /* 5+16 is place for header and explicit IV */
231         memcpy(ciph_d[0].out-16,IVs,16);
232         memcpy(ciph_d[0].iv,IVs,16);    IVs += 16;
233
234         for (i=1;i<x4;i++) {
235                 ciph_d[i].inp = hash_d[i].ptr = hash_d[i-1].ptr+frag;
236                 ciph_d[i].out = ciph_d[i-1].out+packlen;
237                 memcpy(ciph_d[i].out-16,IVs,16);
238                 memcpy(ciph_d[i].iv,IVs,16);    IVs+=16;
239         }
240
241 #if defined(BSWAP8)
242         memcpy(blocks[0].c,key->md.data,8);
243         seqnum = BSWAP8(blocks[0].q[0]);
244 #endif
245         for (i=0;i<x4;i++) {
246                 unsigned int len = (i==(x4-1)?last:frag);
247
248                 ctx->A[i] = key->md.h[0];
249                 ctx->B[i] = key->md.h[1];
250                 ctx->C[i] = key->md.h[2];
251                 ctx->D[i] = key->md.h[3];
252                 ctx->E[i] = key->md.h[4];
253                 ctx->F[i] = key->md.h[5];
254                 ctx->G[i] = key->md.h[6];
255                 ctx->H[i] = key->md.h[7];
256
257                 /* fix seqnum */
258 #if defined(BSWAP8)
259                 blocks[i].q[0] = BSWAP8(seqnum+i);
260 #else
261                 for (carry=i,j=8;j--;) {
262                         blocks[i].c[j] = ((u8*)key->md.data)[j]+carry;
263                         carry = (blocks[i].c[j]-carry)>>(sizeof(carry)*8-1);
264                 }
265 #endif
266                 blocks[i].c[8] = ((u8*)key->md.data)[8];
267                 blocks[i].c[9] = ((u8*)key->md.data)[9];
268                 blocks[i].c[10] = ((u8*)key->md.data)[10];
269                 /* fix length */
270                 blocks[i].c[11] = (u8)(len>>8);
271                 blocks[i].c[12] = (u8)(len);
272
273                 memcpy(blocks[i].c+13,hash_d[i].ptr,64-13);
274                 hash_d[i].ptr += 64-13;
275                 hash_d[i].blocks = (len-(64-13))/64;
276
277                 edges[i].ptr = blocks[i].c;
278                 edges[i].blocks = 1;
279         }
280
281         /* hash 13-byte headers and first 64-13 bytes of inputs */
282         sha256_multi_block(ctx,edges,n4x);
283         /* hash bulk inputs */
284 #define MAXCHUNKSIZE    2048
285 #if     MAXCHUNKSIZE%64
286 #error  "MAXCHUNKSIZE is not divisible by 64"
287 #elif   MAXCHUNKSIZE
288         /* goal is to minimize pressure on L1 cache by moving
289          * in shorter steps, so that hashed data is still in
290          * the cache by the time we encrypt it */
291         minblocks = ((frag<=last ? frag : last)-(64-13))/64;
292         if (minblocks>MAXCHUNKSIZE/64) {
293                 for (i=0;i<x4;i++) {
294                         edges[i].ptr     = hash_d[i].ptr;
295                         edges[i].blocks  = MAXCHUNKSIZE/64;
296                         ciph_d[i].blocks = MAXCHUNKSIZE/16;
297                 }
298                 do {
299                         sha256_multi_block(ctx,edges,n4x);
300                         aesni_multi_cbc_encrypt(ciph_d,&key->ks,n4x);
301
302                         for (i=0;i<x4;i++) {
303                                 edges[i].ptr     = hash_d[i].ptr += MAXCHUNKSIZE;
304                                 hash_d[i].blocks -= MAXCHUNKSIZE/64;
305                                 edges[i].blocks  = MAXCHUNKSIZE/64;
306                                 ciph_d[i].inp    += MAXCHUNKSIZE;
307                                 ciph_d[i].out    += MAXCHUNKSIZE;
308                                 ciph_d[i].blocks = MAXCHUNKSIZE/16;
309                                 memcpy(ciph_d[i].iv,ciph_d[i].out-16,16);
310                         }
311                         processed += MAXCHUNKSIZE;
312                         minblocks -= MAXCHUNKSIZE/64;
313                 } while (minblocks>MAXCHUNKSIZE/64);
314         }
315 #endif
316 #undef  MAXCHUNKSIZE
317         sha256_multi_block(ctx,hash_d,n4x);
318
319         memset(blocks,0,sizeof(blocks));
320         for (i=0;i<x4;i++) {
321                 unsigned int            len = (i==(x4-1)?last:frag),
322                                         off = hash_d[i].blocks*64;
323                 const unsigned char    *ptr = hash_d[i].ptr+off;
324
325                 off = (len-processed)-(64-13)-off;      /* remainder actually */
326                 memcpy(blocks[i].c,ptr,off);
327                 blocks[i].c[off]=0x80;
328                 len += 64+13;           /* 64 is HMAC header */
329                 len *= 8;               /* convert to bits */
330                 if (off<(64-8)) {
331                         PUTU32(blocks[i].c+60,len);
332                         edges[i].blocks = 1;                    
333                 } else {
334                         PUTU32(blocks[i].c+124,len);
335                         edges[i].blocks = 2;
336                 }
337                 edges[i].ptr = blocks[i].c;
338         }
339
340         /* hash input tails and finalize */
341         sha256_multi_block(ctx,edges,n4x);
342
343         memset(blocks,0,sizeof(blocks));
344         for (i=0;i<x4;i++) {
345                 PUTU32(blocks[i].c+0,ctx->A[i]);        ctx->A[i] = key->tail.h[0];
346                 PUTU32(blocks[i].c+4,ctx->B[i]);        ctx->B[i] = key->tail.h[1];
347                 PUTU32(blocks[i].c+8,ctx->C[i]);        ctx->C[i] = key->tail.h[2];
348                 PUTU32(blocks[i].c+12,ctx->D[i]);       ctx->D[i] = key->tail.h[3];
349                 PUTU32(blocks[i].c+16,ctx->E[i]);       ctx->E[i] = key->tail.h[4];
350                 PUTU32(blocks[i].c+20,ctx->F[i]);       ctx->F[i] = key->tail.h[5];
351                 PUTU32(blocks[i].c+24,ctx->G[i]);       ctx->G[i] = key->tail.h[6];
352                 PUTU32(blocks[i].c+28,ctx->H[i]);       ctx->H[i] = key->tail.h[7];
353                 blocks[i].c[32] = 0x80;
354                 PUTU32(blocks[i].c+60,(64+32)*8);
355                 edges[i].ptr = blocks[i].c;
356                 edges[i].blocks = 1;
357         }
358
359         /* finalize MACs */
360         sha256_multi_block(ctx,edges,n4x);
361
362         for (i=0;i<x4;i++) {
363                 unsigned int len = (i==(x4-1)?last:frag), pad, j;
364                 unsigned char *out0 = out;
365
366                 memcpy(ciph_d[i].out,ciph_d[i].inp,len-processed);
367                 ciph_d[i].inp = ciph_d[i].out;
368
369                 out += 5+16+len;
370
371                 /* write MAC */
372                 PUTU32(out+0,ctx->A[i]);
373                 PUTU32(out+4,ctx->B[i]);
374                 PUTU32(out+8,ctx->C[i]);
375                 PUTU32(out+12,ctx->D[i]);
376                 PUTU32(out+16,ctx->E[i]);
377                 PUTU32(out+20,ctx->F[i]);
378                 PUTU32(out+24,ctx->G[i]);
379                 PUTU32(out+28,ctx->H[i]);
380                 out += 32;
381                 len += 32;
382
383                 /* pad */
384                 pad = 15-len%16;
385                 for (j=0;j<=pad;j++) *(out++) = pad;
386                 len += pad+1;
387
388                 ciph_d[i].blocks = (len-processed)/16;
389                 len += 16;      /* account for explicit iv */
390
391                 /* arrange header */
392                 out0[0] = ((u8*)key->md.data)[8];
393                 out0[1] = ((u8*)key->md.data)[9];
394                 out0[2] = ((u8*)key->md.data)[10];
395                 out0[3] = (u8)(len>>8);
396                 out0[4] = (u8)(len);
397
398                 ret += len+5;
399                 inp += frag;
400         }
401
402         aesni_multi_cbc_encrypt(ciph_d,&key->ks,n4x);
403
404         OPENSSL_cleanse(blocks,sizeof(blocks));
405         OPENSSL_cleanse(ctx,sizeof(*ctx));
406
407         return ret;
408 }
409 #endif
410
411 static int aesni_cbc_hmac_sha256_cipher(EVP_CIPHER_CTX *ctx, unsigned char *out,
412                       const unsigned char *in, size_t len)
413         {
414         EVP_AES_HMAC_SHA256 *key = data(ctx);
415         unsigned int l;
416         size_t  plen = key->payload_length,
417                 iv = 0,         /* explicit IV in TLS 1.1 and later */
418                 sha_off = 0;
419 #if defined(STITCHED_CALL)
420         size_t  aes_off = 0,
421                 blocks;
422
423         sha_off = SHA256_CBLOCK-key->md.num;
424 #endif
425
426         key->payload_length = NO_PAYLOAD_LENGTH;
427
428         if (len%AES_BLOCK_SIZE) return 0;
429
430         if (ctx->encrypt) {
431                 if (plen==NO_PAYLOAD_LENGTH)
432                         plen = len;
433                 else if (len!=((plen+SHA256_DIGEST_LENGTH+AES_BLOCK_SIZE)&-AES_BLOCK_SIZE))
434                         return 0;
435                 else if (key->aux.tls_ver >= TLS1_1_VERSION)
436                         iv = AES_BLOCK_SIZE;
437
438 #if defined(STITCHED_CALL)
439                 if (OPENSSL_ia32cap_P[1]&(1<<(60-32)) && /* AVX? */
440                     plen>(sha_off+iv) &&
441                     (blocks=(plen-(sha_off+iv))/SHA256_CBLOCK)) {
442                         SHA256_Update(&key->md,in+iv,sha_off);
443
444                         (void)aesni_cbc_sha256_enc(in,out,blocks,&key->ks,
445                                 ctx->iv,&key->md,in+iv+sha_off);
446                         blocks *= SHA256_CBLOCK;
447                         aes_off += blocks;
448                         sha_off += blocks;
449                         key->md.Nh += blocks>>29;
450                         key->md.Nl += blocks<<=3;
451                         if (key->md.Nl<(unsigned int)blocks) key->md.Nh++;
452                 } else {
453                         sha_off = 0;
454                 }
455 #endif
456                 sha_off += iv;
457                 SHA256_Update(&key->md,in+sha_off,plen-sha_off);
458
459                 if (plen!=len)  {       /* "TLS" mode of operation */
460                         if (in!=out)
461                                 memcpy(out+aes_off,in+aes_off,plen-aes_off);
462
463                         /* calculate HMAC and append it to payload */
464                         SHA256_Final(out+plen,&key->md);
465                         key->md = key->tail;
466                         SHA256_Update(&key->md,out+plen,SHA256_DIGEST_LENGTH);
467                         SHA256_Final(out+plen,&key->md);
468
469                         /* pad the payload|hmac */
470                         plen += SHA256_DIGEST_LENGTH;
471                         for (l=len-plen-1;plen<len;plen++) out[plen]=l;
472                         /* encrypt HMAC|padding at once */
473                         aesni_cbc_encrypt(out+aes_off,out+aes_off,len-aes_off,
474                                         &key->ks,ctx->iv,1);
475                 } else {
476                         aesni_cbc_encrypt(in+aes_off,out+aes_off,len-aes_off,
477                                         &key->ks,ctx->iv,1);
478                 }
479         } else {
480                 union { unsigned int  u[SHA256_DIGEST_LENGTH/sizeof(unsigned int)];
481                         unsigned char c[64+SHA256_DIGEST_LENGTH]; } mac, *pmac;
482
483                 /* arrange cache line alignment */
484                 pmac = (void *)(((size_t)mac.c+63)&((size_t)0-64));
485
486                 /* decrypt HMAC|padding at once */
487                 aesni_cbc_encrypt(in,out,len,
488                                 &key->ks,ctx->iv,0);
489
490                 if (plen != NO_PAYLOAD_LENGTH) {        /* "TLS" mode of operation */
491                         size_t inp_len, mask, j, i;
492                         unsigned int res, maxpad, pad, bitlen;
493                         int ret = 1;
494                         union { unsigned int  u[SHA_LBLOCK];
495                                 unsigned char c[SHA256_CBLOCK]; }
496                                 *data = (void *)key->md.data;
497
498                         if ((key->aux.tls_aad[plen-4]<<8|key->aux.tls_aad[plen-3])
499                             >= TLS1_1_VERSION)
500                                 iv = AES_BLOCK_SIZE;
501
502                         if (len<(iv+SHA256_DIGEST_LENGTH+1))
503                                 return 0;
504
505                         /* omit explicit iv */
506                         out += iv;
507                         len -= iv;
508
509                         /* figure out payload length */
510                         pad = out[len-1];
511                         maxpad = len-(SHA256_DIGEST_LENGTH+1);
512                         maxpad |= (255-maxpad)>>(sizeof(maxpad)*8-8);
513                         maxpad &= 255;
514
515                         inp_len = len - (SHA256_DIGEST_LENGTH+pad+1);
516                         mask = (0-((inp_len-len)>>(sizeof(inp_len)*8-1)));
517                         inp_len &= mask;
518                         ret &= (int)mask;
519
520                         key->aux.tls_aad[plen-2] = inp_len>>8;
521                         key->aux.tls_aad[plen-1] = inp_len;
522
523                         /* calculate HMAC */
524                         key->md = key->head;
525                         SHA256_Update(&key->md,key->aux.tls_aad,plen);
526
527 #if 1
528                         len -= SHA256_DIGEST_LENGTH;            /* amend mac */
529                         if (len>=(256+SHA256_CBLOCK)) {
530                                 j = (len-(256+SHA256_CBLOCK))&(0-SHA256_CBLOCK);
531                                 j += SHA256_CBLOCK-key->md.num;
532                                 SHA256_Update(&key->md,out,j);
533                                 out += j;
534                                 len -= j;
535                                 inp_len -= j;
536                         }
537
538                         /* but pretend as if we hashed padded payload */
539                         bitlen = key->md.Nl+(inp_len<<3);       /* at most 18 bits */
540 #ifdef BSWAP4
541                         bitlen = BSWAP4(bitlen);
542 #else
543                         mac.c[0] = 0;
544                         mac.c[1] = (unsigned char)(bitlen>>16);
545                         mac.c[2] = (unsigned char)(bitlen>>8);
546                         mac.c[3] = (unsigned char)bitlen;
547                         bitlen = mac.u[0];
548 #endif
549
550                         pmac->u[0]=0;
551                         pmac->u[1]=0;
552                         pmac->u[2]=0;
553                         pmac->u[3]=0;
554                         pmac->u[4]=0;
555                         pmac->u[5]=0;
556                         pmac->u[6]=0;
557                         pmac->u[7]=0;
558
559                         for (res=key->md.num, j=0;j<len;j++) {
560                                 size_t c = out[j];
561                                 mask = (j-inp_len)>>(sizeof(j)*8-8);
562                                 c &= mask;
563                                 c |= 0x80&~mask&~((inp_len-j)>>(sizeof(j)*8-8));
564                                 data->c[res++]=(unsigned char)c;
565
566                                 if (res!=SHA256_CBLOCK) continue;
567
568                                 /* j is not incremented yet */
569                                 mask = 0-((inp_len+7-j)>>(sizeof(j)*8-1));
570                                 data->u[SHA_LBLOCK-1] |= bitlen&mask;
571                                 sha256_block_data_order(&key->md,data,1);
572                                 mask &= 0-((j-inp_len-72)>>(sizeof(j)*8-1));
573                                 pmac->u[0] |= key->md.h[0] & mask;
574                                 pmac->u[1] |= key->md.h[1] & mask;
575                                 pmac->u[2] |= key->md.h[2] & mask;
576                                 pmac->u[3] |= key->md.h[3] & mask;
577                                 pmac->u[4] |= key->md.h[4] & mask;
578                                 pmac->u[5] |= key->md.h[5] & mask;
579                                 pmac->u[6] |= key->md.h[6] & mask;
580                                 pmac->u[7] |= key->md.h[7] & mask;
581                                 res=0;
582                         }
583
584                         for(i=res;i<SHA256_CBLOCK;i++,j++) data->c[i]=0;
585
586                         if (res>SHA256_CBLOCK-8) {
587                                 mask = 0-((inp_len+8-j)>>(sizeof(j)*8-1));
588                                 data->u[SHA_LBLOCK-1] |= bitlen&mask;
589                                 sha256_block_data_order(&key->md,data,1);
590                                 mask &= 0-((j-inp_len-73)>>(sizeof(j)*8-1));
591                                 pmac->u[0] |= key->md.h[0] & mask;
592                                 pmac->u[1] |= key->md.h[1] & mask;
593                                 pmac->u[2] |= key->md.h[2] & mask;
594                                 pmac->u[3] |= key->md.h[3] & mask;
595                                 pmac->u[4] |= key->md.h[4] & mask;
596                                 pmac->u[5] |= key->md.h[5] & mask;
597                                 pmac->u[6] |= key->md.h[6] & mask;
598                                 pmac->u[7] |= key->md.h[7] & mask;
599
600                                 memset(data,0,SHA256_CBLOCK);
601                                 j+=64;
602                         }
603                         data->u[SHA_LBLOCK-1] = bitlen;
604                         sha256_block_data_order(&key->md,data,1);
605                         mask = 0-((j-inp_len-73)>>(sizeof(j)*8-1));
606                         pmac->u[0] |= key->md.h[0] & mask;
607                         pmac->u[1] |= key->md.h[1] & mask;
608                         pmac->u[2] |= key->md.h[2] & mask;
609                         pmac->u[3] |= key->md.h[3] & mask;
610                         pmac->u[4] |= key->md.h[4] & mask;
611                         pmac->u[5] |= key->md.h[5] & mask;
612                         pmac->u[6] |= key->md.h[6] & mask;
613                         pmac->u[7] |= key->md.h[7] & mask;
614
615 #ifdef BSWAP4
616                         pmac->u[0] = BSWAP4(pmac->u[0]);
617                         pmac->u[1] = BSWAP4(pmac->u[1]);
618                         pmac->u[2] = BSWAP4(pmac->u[2]);
619                         pmac->u[3] = BSWAP4(pmac->u[3]);
620                         pmac->u[4] = BSWAP4(pmac->u[4]);
621                         pmac->u[5] = BSWAP4(pmac->u[5]);
622                         pmac->u[6] = BSWAP4(pmac->u[6]);
623                         pmac->u[7] = BSWAP4(pmac->u[7]);
624 #else
625                         for (i=0;i<8;i++) {
626                                 res = pmac->u[i];
627                                 pmac->c[4*i+0]=(unsigned char)(res>>24);
628                                 pmac->c[4*i+1]=(unsigned char)(res>>16);
629                                 pmac->c[4*i+2]=(unsigned char)(res>>8);
630                                 pmac->c[4*i+3]=(unsigned char)res;
631                         }
632 #endif
633                         len += SHA256_DIGEST_LENGTH;
634 #else
635                         SHA256_Update(&key->md,out,inp_len);
636                         res = key->md.num;
637                         SHA256_Final(pmac->c,&key->md);
638
639                         {
640                         unsigned int inp_blocks, pad_blocks;
641
642                         /* but pretend as if we hashed padded payload */
643                         inp_blocks = 1+((SHA256_CBLOCK-9-res)>>(sizeof(res)*8-1));
644                         res += (unsigned int)(len-inp_len);
645                         pad_blocks = res / SHA256_CBLOCK;
646                         res %= SHA256_CBLOCK;
647                         pad_blocks += 1+((SHA256_CBLOCK-9-res)>>(sizeof(res)*8-1));
648                         for (;inp_blocks<pad_blocks;inp_blocks++)
649                                 sha1_block_data_order(&key->md,data,1);
650                         }
651 #endif
652                         key->md = key->tail;
653                         SHA256_Update(&key->md,pmac->c,SHA256_DIGEST_LENGTH);
654                         SHA256_Final(pmac->c,&key->md);
655
656                         /* verify HMAC */
657                         out += inp_len;
658                         len -= inp_len;
659 #if 1
660                         {
661                         unsigned char *p = out+len-1-maxpad-SHA256_DIGEST_LENGTH;
662                         size_t off = out-p;
663                         unsigned int c, cmask;
664
665                         maxpad += SHA256_DIGEST_LENGTH;
666                         for (res=0,i=0,j=0;j<maxpad;j++) {
667                                 c = p[j];
668                                 cmask = ((int)(j-off-SHA256_DIGEST_LENGTH))>>(sizeof(int)*8-1);
669                                 res |= (c^pad)&~cmask;  /* ... and padding */
670                                 cmask &= ((int)(off-1-j))>>(sizeof(int)*8-1);
671                                 res |= (c^pmac->c[i])&cmask;
672                                 i += 1&cmask;
673                         }
674                         maxpad -= SHA256_DIGEST_LENGTH;
675
676                         res = 0-((0-res)>>(sizeof(res)*8-1));
677                         ret &= (int)~res;
678                         }
679 #else
680                         for (res=0,i=0;i<SHA256_DIGEST_LENGTH;i++)
681                                 res |= out[i]^pmac->c[i];
682                         res = 0-((0-res)>>(sizeof(res)*8-1));
683                         ret &= (int)~res;
684
685                         /* verify padding */
686                         pad = (pad&~res) | (maxpad&res);
687                         out = out+len-1-pad;
688                         for (res=0,i=0;i<pad;i++)
689                                 res |= out[i]^pad;
690
691                         res = (0-res)>>(sizeof(res)*8-1);
692                         ret &= (int)~res;
693 #endif
694                         return ret;
695                 } else {
696                         SHA256_Update(&key->md,out,len);
697                 }
698         }
699
700         return 1;
701         }
702
703 static int aesni_cbc_hmac_sha256_ctrl(EVP_CIPHER_CTX *ctx, int type, int arg, void *ptr)
704         {
705         EVP_AES_HMAC_SHA256 *key = data(ctx);
706
707         switch (type)
708                 {
709         case EVP_CTRL_AEAD_SET_MAC_KEY:
710                 {
711                 unsigned int  i;
712                 unsigned char hmac_key[64];
713
714                 memset (hmac_key,0,sizeof(hmac_key));
715
716                 if (arg > (int)sizeof(hmac_key)) {
717                         SHA256_Init(&key->head);
718                         SHA256_Update(&key->head,ptr,arg);
719                         SHA256_Final(hmac_key,&key->head);
720                 } else {
721                         memcpy(hmac_key,ptr,arg);
722                 }
723
724                 for (i=0;i<sizeof(hmac_key);i++)
725                         hmac_key[i] ^= 0x36;            /* ipad */
726                 SHA256_Init(&key->head);
727                 SHA256_Update(&key->head,hmac_key,sizeof(hmac_key));
728
729                 for (i=0;i<sizeof(hmac_key);i++)
730                         hmac_key[i] ^= 0x36^0x5c;       /* opad */
731                 SHA256_Init(&key->tail);
732                 SHA256_Update(&key->tail,hmac_key,sizeof(hmac_key));
733
734                 OPENSSL_cleanse(hmac_key,sizeof(hmac_key));
735
736                 return 1;
737                 }
738         case EVP_CTRL_AEAD_TLS1_AAD:
739                 {
740                 unsigned char *p=ptr;
741                 unsigned int   len=p[arg-2]<<8|p[arg-1];
742
743                 if (ctx->encrypt)
744                         {
745                         key->payload_length = len;
746                         if ((key->aux.tls_ver=p[arg-4]<<8|p[arg-3]) >= TLS1_1_VERSION) {
747                                 len -= AES_BLOCK_SIZE;
748                                 p[arg-2] = len>>8;
749                                 p[arg-1] = len;
750                         }
751                         key->md = key->head;
752                         SHA256_Update(&key->md,p,arg);
753
754                         return (int)(((len+SHA256_DIGEST_LENGTH+AES_BLOCK_SIZE)&-AES_BLOCK_SIZE)
755                                 - len);
756                         }
757                 else
758                         {
759                         if (arg>13) arg = 13;
760                         memcpy(key->aux.tls_aad,ptr,arg);
761                         key->payload_length = arg;
762
763                         return SHA256_DIGEST_LENGTH;
764                         }
765                 }
766 #if !defined(OPENSSL_NO_MULTIBLOCK) && EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK
767         case EVP_CTRL_TLS1_1_MULTIBLOCK_MAX_BUFSIZE:
768                 return (int)(5+16+((arg+32+16)&-16));
769         case EVP_CTRL_TLS1_1_MULTIBLOCK_AAD:
770                 {
771                 EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param =
772                         (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *)ptr;
773                 unsigned int n4x=1, x4;
774                 unsigned int frag, last, packlen, inp_len;
775
776                 if (arg<sizeof(EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM)) return -1;
777
778                 inp_len = param->inp[11]<<8|param->inp[12];
779
780                 if (ctx->encrypt)
781                         {
782                         if ((param->inp[9]<<8|param->inp[10]) < TLS1_1_VERSION)
783                                 return -1;
784
785                         if (inp_len)
786                                 {
787                                 if (inp_len<4096) return 0;     /* too short */
788
789                                 if (inp_len>=8192 && OPENSSL_ia32cap_P[2]&(1<<5))
790                                         n4x=2;  /* AVX2 */
791                                 }
792                         else if ((n4x=param->interleave/4) && n4x<=2)
793                                 inp_len = param->len;
794                         else
795                                 return -1;
796
797                         key->md = key->head;
798                         SHA256_Update(&key->md,param->inp,13);
799
800                         x4 = 4*n4x; n4x += 1;
801
802                         frag = inp_len>>n4x;
803                         last = inp_len+frag-(frag<<n4x);
804                         if (last>frag && ((last+13+9)%64<(x4-1))) {
805                                 frag++;
806                                 last -= x4-1;
807                         }
808
809                         packlen = 5+16+((frag+32+16)&-16);
810                         packlen = (packlen<<n4x)-packlen;
811                         packlen += 5+16+((last+32+16)&-16);
812
813                         param->interleave = x4;
814
815                         return (int)packlen;
816                         }
817                 else
818                         return -1;      /* not yet */
819                 }
820         case EVP_CTRL_TLS1_1_MULTIBLOCK_ENCRYPT:
821                 {
822                 EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *param =
823                         (EVP_CTRL_TLS1_1_MULTIBLOCK_PARAM *)ptr;
824
825                 return (int)tls1_1_multi_block_encrypt(key,param->out,param->inp,
826                                                 param->len,param->interleave/4);
827                 }
828         case EVP_CTRL_TLS1_1_MULTIBLOCK_DECRYPT:
829 #endif
830         default:
831                 return -1;
832                 }
833         }
834
835 static EVP_CIPHER aesni_128_cbc_hmac_sha256_cipher =
836         {
837 #ifdef NID_aes_128_cbc_hmac_sha256
838         NID_aes_128_cbc_hmac_sha256,
839 #else
840         NID_undef,
841 #endif
842         16,16,16,
843         EVP_CIPH_CBC_MODE|EVP_CIPH_FLAG_DEFAULT_ASN1|
844         EVP_CIPH_FLAG_AEAD_CIPHER|EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK,
845         aesni_cbc_hmac_sha256_init_key,
846         aesni_cbc_hmac_sha256_cipher,
847         NULL,
848         sizeof(EVP_AES_HMAC_SHA256),
849         EVP_CIPH_FLAG_DEFAULT_ASN1?NULL:EVP_CIPHER_set_asn1_iv,
850         EVP_CIPH_FLAG_DEFAULT_ASN1?NULL:EVP_CIPHER_get_asn1_iv,
851         aesni_cbc_hmac_sha256_ctrl,
852         NULL
853         };
854
855 static EVP_CIPHER aesni_256_cbc_hmac_sha256_cipher =
856         {
857 #ifdef NID_aes_256_cbc_hmac_sha256
858         NID_aes_256_cbc_hmac_sha256,
859 #else
860         NID_undef,
861 #endif
862         16,32,16,
863         EVP_CIPH_CBC_MODE|EVP_CIPH_FLAG_DEFAULT_ASN1|
864         EVP_CIPH_FLAG_AEAD_CIPHER|EVP_CIPH_FLAG_TLS1_1_MULTIBLOCK,
865         aesni_cbc_hmac_sha256_init_key,
866         aesni_cbc_hmac_sha256_cipher,
867         NULL,
868         sizeof(EVP_AES_HMAC_SHA256),
869         EVP_CIPH_FLAG_DEFAULT_ASN1?NULL:EVP_CIPHER_set_asn1_iv,
870         EVP_CIPH_FLAG_DEFAULT_ASN1?NULL:EVP_CIPHER_get_asn1_iv,
871         aesni_cbc_hmac_sha256_ctrl,
872         NULL
873         };
874
875 const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha256(void)
876         {
877         return((OPENSSL_ia32cap_P[1]&AESNI_CAPABLE) &&
878                 aesni_cbc_sha256_enc(NULL,NULL,0,NULL,NULL,NULL,NULL) ?
879                 &aesni_128_cbc_hmac_sha256_cipher:NULL);
880         }
881
882 const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha256(void)
883         {
884         return((OPENSSL_ia32cap_P[1]&AESNI_CAPABLE) &&
885                 aesni_cbc_sha256_enc(NULL,NULL,0,NULL,NULL,NULL,NULL)?
886                 &aesni_256_cbc_hmac_sha256_cipher:NULL);
887         }
888 #else
889 const EVP_CIPHER *EVP_aes_128_cbc_hmac_sha256(void)
890         {
891         return NULL;
892         }
893 const EVP_CIPHER *EVP_aes_256_cbc_hmac_sha256(void)
894         {
895         return NULL;
896         }
897 #endif
898 #endif