1 /* crypto/md32_common.h */
2 /* ====================================================================
3 * Copyright (c) 1999-2002 The OpenSSL Project. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in
14 * the documentation and/or other materials provided with the
17 * 3. All advertising materials mentioning features or use of this
18 * software must display the following acknowledgment:
19 * "This product includes software developed by the OpenSSL Project
20 * for use in the OpenSSL Toolkit. (http://www.OpenSSL.org/)"
22 * 4. The names "OpenSSL Toolkit" and "OpenSSL Project" must not be used to
23 * endorse or promote products derived from this software without
24 * prior written permission. For written permission, please contact
25 * licensing@OpenSSL.org.
27 * 5. Products derived from this software may not be called "OpenSSL"
28 * nor may "OpenSSL" appear in their names without prior written
29 * permission of the OpenSSL Project.
31 * 6. Redistributions of any form whatsoever must retain the following
33 * "This product includes software developed by the OpenSSL Project
34 * for use in the OpenSSL Toolkit (http://www.OpenSSL.org/)"
36 * THIS SOFTWARE IS PROVIDED BY THE OpenSSL PROJECT ``AS IS'' AND ANY
37 * EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
38 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
39 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE OpenSSL PROJECT OR
40 * ITS CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
41 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
42 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
43 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
45 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
46 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
47 * OF THE POSSIBILITY OF SUCH DAMAGE.
48 * ====================================================================
50 * This product includes cryptographic software written by Eric Young
51 * (eay@cryptsoft.com). This product includes software written by Tim
52 * Hudson (tjh@cryptsoft.com).
57 * This is a generic 32 bit "collector" for message digest algorithms.
58 * Whenever needed it collects input character stream into chunks of
59 * 32 bit values and invokes a block function that performs actual hash
66 * DATA_ORDER_IS_BIG_ENDIAN or DATA_ORDER_IS_LITTLE_ENDIAN
67 * this macro defines byte order of input stream.
69 * size of a unit chunk HASH_BLOCK operates on.
71 * has to be at lest 32 bit wide, if it's wider, then
72 * HASH_LONG_LOG2 *has to* be defined along
74 * context structure that at least contains following
79 * HASH_LONG data[HASH_LBLOCK];
84 * name of "Update" function, implemented here.
86 * name of "Transform" function, implemented here.
88 * name of "Final" function, implemented here.
89 * HASH_BLOCK_HOST_ORDER
90 * name of "block" function treating *aligned* input message
91 * in host byte order, implemented externally.
92 * HASH_BLOCK_DATA_ORDER
93 * name of "block" function treating *unaligned* input message
94 * in original (data) byte order, implemented externally (it
95 * actually is optional if data and host are of the same
98 * macro convering context variables to an ASCII hash string.
102 * B_ENDIAN or L_ENDIAN
103 * defines host byte-order.
105 * defaults to 2 if not states otherwise.
107 * assumed to be HASH_CBLOCK/4 if not stated otherwise.
108 * HASH_BLOCK_DATA_ORDER_ALIGNED
109 * alternative "block" function capable of treating
110 * aligned input message in original (data) order,
111 * implemented externally.
115 * #define DATA_ORDER_IS_LITTLE_ENDIAN
117 * #define HASH_LONG MD5_LONG
118 * #define HASH_LONG_LOG2 MD5_LONG_LOG2
119 * #define HASH_CTX MD5_CTX
120 * #define HASH_CBLOCK MD5_CBLOCK
121 * #define HASH_LBLOCK MD5_LBLOCK
122 * #define HASH_UPDATE MD5_Update
123 * #define HASH_TRANSFORM MD5_Transform
124 * #define HASH_FINAL MD5_Final
125 * #define HASH_BLOCK_HOST_ORDER md5_block_host_order
126 * #define HASH_BLOCK_DATA_ORDER md5_block_data_order
128 * <appro@fy.chalmers.se>
131 #include <openssl/fips.h>
132 #include <openssl/err.h>
134 #if !defined(DATA_ORDER_IS_BIG_ENDIAN) && !defined(DATA_ORDER_IS_LITTLE_ENDIAN)
135 #error "DATA_ORDER must be defined!"
139 #error "HASH_CBLOCK must be defined!"
142 #error "HASH_LONG must be defined!"
145 #error "HASH_CTX must be defined!"
149 #error "HASH_UPDATE must be defined!"
151 #ifndef HASH_TRANSFORM
152 #error "HASH_TRANSFORM must be defined!"
155 #error "HASH_FINAL must be defined!"
158 #ifndef HASH_BLOCK_HOST_ORDER
159 #error "HASH_BLOCK_HOST_ORDER must be defined!"
164 * Moved below as it's required only if HASH_BLOCK_DATA_ORDER_ALIGNED
167 #ifndef HASH_BLOCK_DATA_ORDER
168 #error "HASH_BLOCK_DATA_ORDER must be defined!"
173 #define HASH_LBLOCK (HASH_CBLOCK/4)
176 #ifndef HASH_LONG_LOG2
177 #define HASH_LONG_LOG2 2
181 * Engage compiler specific rotate intrinsic function if available.
185 # if 0 /* defined(_MSC_VER) */
186 # define ROTATE(a,n) _lrotl(a,n)
187 # elif defined(__MWERKS__)
188 # if defined(__POWERPC__)
189 # define ROTATE(a,n) __rlwinm(a,n,0,31)
190 # elif defined(__MC68K__)
191 /* Motorola specific tweak. <appro@fy.chalmers.se> */
192 # define ROTATE(a,n) ( n<24 ? __rol(a,n) : __ror(a,32-n) )
194 # define ROTATE(a,n) __rol(a,n)
196 # elif defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
198 * Some GNU C inline assembler templates. Note that these are
199 * rotates by *constant* number of bits! But that's exactly
200 * what we need here...
202 * <appro@fy.chalmers.se>
204 # if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)
205 # define ROTATE(a,n) ({ register unsigned int ret; \
213 # elif defined(__powerpc) || defined(__ppc)
214 # define ROTATE(a,n) ({ register unsigned int ret; \
216 "rlwinm %0,%1,%2,0,31" \
225 * Engage compiler specific "fetch in reverse byte order"
226 * intrinsic function if available.
228 # if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
229 /* some GNU C inline assembler templates by <appro@fy.chalmers.se> */
230 # if (defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)) && !defined(I386_ONLY)
231 # define BE_FETCH32(a) ({ register unsigned int l=(a);\
234 : "=r"(l) : "0"(l)); \
237 # elif defined(__powerpc)
238 # define LE_FETCH32(a) ({ register unsigned int l; \
246 # elif defined(__sparc) && defined(OPENSSL_SYS_ULTRASPARC)
247 # define LE_FETCH32(a) ({ register unsigned int l; \
249 "lda [%1]#ASI_PRIMARY_LITTLE,%0"\
256 #endif /* PEDANTIC */
258 #if HASH_LONG_LOG2==2 /* Engage only if sizeof(HASH_LONG)== 4 */
259 /* A nice byte order reversal from Wei Dai <weidai@eskimo.com> */
261 /* 5 instructions with rotate instruction, else 9 */
262 #define REVERSE_FETCH32(a,l) ( \
263 l=*(const HASH_LONG *)(a), \
264 ((ROTATE(l,8)&0x00FF00FF)|(ROTATE((l&0x00FF00FF),24))) \
267 /* 6 instructions with rotate instruction, else 8 */
268 #define REVERSE_FETCH32(a,l) ( \
269 l=*(const HASH_LONG *)(a), \
270 l=(((l>>8)&0x00FF00FF)|((l&0x00FF00FF)<<8)), \
274 * Originally the middle line started with l=(((l&0xFF00FF00)>>8)|...
275 * It's rewritten as above for two reasons:
276 * - RISCs aren't good at long constants and have to explicitely
277 * compose 'em with several (well, usually 2) instructions in a
278 * register before performing the actual operation and (as you
279 * already realized:-) having same constant should inspire the
280 * compiler to permanently allocate the only register for it;
281 * - most modern CPUs have two ALUs, but usually only one has
282 * circuitry for shifts:-( this minor tweak inspires compiler
283 * to schedule shift instructions in a better way...
285 * <appro@fy.chalmers.se>
291 #define ROTATE(a,n) (((a)<<(n))|(((a)&0xffffffff)>>(32-(n))))
295 * Make some obvious choices. E.g., HASH_BLOCK_DATA_ORDER_ALIGNED
296 * and HASH_BLOCK_HOST_ORDER ought to be the same if input data
297 * and host are of the same "endianess". It's possible to mask
298 * this with blank #define HASH_BLOCK_DATA_ORDER though...
300 * <appro@fy.chalmers.se>
302 #if defined(B_ENDIAN)
303 # if defined(DATA_ORDER_IS_BIG_ENDIAN)
304 # if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED) && HASH_LONG_LOG2==2
305 # define HASH_BLOCK_DATA_ORDER_ALIGNED HASH_BLOCK_HOST_ORDER
307 # elif defined(DATA_ORDER_IS_LITTLE_ENDIAN)
308 # ifndef HOST_FETCH32
310 # define HOST_FETCH32(p,l) LE_FETCH32(p)
311 # elif defined(REVERSE_FETCH32)
312 # define HOST_FETCH32(p,l) REVERSE_FETCH32(p,l)
316 #elif defined(L_ENDIAN)
317 # if defined(DATA_ORDER_IS_LITTLE_ENDIAN)
318 # if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED) && HASH_LONG_LOG2==2
319 # define HASH_BLOCK_DATA_ORDER_ALIGNED HASH_BLOCK_HOST_ORDER
321 # elif defined(DATA_ORDER_IS_BIG_ENDIAN)
322 # ifndef HOST_FETCH32
324 # define HOST_FETCH32(p,l) BE_FETCH32(p)
325 # elif defined(REVERSE_FETCH32)
326 # define HOST_FETCH32(p,l) REVERSE_FETCH32(p,l)
332 #if !defined(HASH_BLOCK_DATA_ORDER_ALIGNED)
333 #ifndef HASH_BLOCK_DATA_ORDER
334 #error "HASH_BLOCK_DATA_ORDER must be defined!"
338 #if defined(DATA_ORDER_IS_BIG_ENDIAN)
340 #define HOST_c2l(c,l) (l =(((unsigned long)(*((c)++)))<<24), \
341 l|=(((unsigned long)(*((c)++)))<<16), \
342 l|=(((unsigned long)(*((c)++)))<< 8), \
343 l|=(((unsigned long)(*((c)++))) ), \
345 #define HOST_p_c2l(c,l,n) { \
347 case 0: l =((unsigned long)(*((c)++)))<<24; \
348 case 1: l|=((unsigned long)(*((c)++)))<<16; \
349 case 2: l|=((unsigned long)(*((c)++)))<< 8; \
350 case 3: l|=((unsigned long)(*((c)++))); \
352 #define HOST_p_c2l_p(c,l,sc,len) { \
354 case 0: l =((unsigned long)(*((c)++)))<<24; \
355 if (--len == 0) break; \
356 case 1: l|=((unsigned long)(*((c)++)))<<16; \
357 if (--len == 0) break; \
358 case 2: l|=((unsigned long)(*((c)++)))<< 8; \
360 /* NOTE the pointer is not incremented at the end of this */
361 #define HOST_c2l_p(c,l,n) { \
364 case 3: l =((unsigned long)(*(--(c))))<< 8; \
365 case 2: l|=((unsigned long)(*(--(c))))<<16; \
366 case 1: l|=((unsigned long)(*(--(c))))<<24; \
368 #define HOST_l2c(l,c) (*((c)++)=(unsigned char)(((l)>>24)&0xff), \
369 *((c)++)=(unsigned char)(((l)>>16)&0xff), \
370 *((c)++)=(unsigned char)(((l)>> 8)&0xff), \
371 *((c)++)=(unsigned char)(((l) )&0xff), \
374 #elif defined(DATA_ORDER_IS_LITTLE_ENDIAN)
376 #define HOST_c2l(c,l) (l =(((unsigned long)(*((c)++))) ), \
377 l|=(((unsigned long)(*((c)++)))<< 8), \
378 l|=(((unsigned long)(*((c)++)))<<16), \
379 l|=(((unsigned long)(*((c)++)))<<24), \
381 #define HOST_p_c2l(c,l,n) { \
383 case 0: l =((unsigned long)(*((c)++))); \
384 case 1: l|=((unsigned long)(*((c)++)))<< 8; \
385 case 2: l|=((unsigned long)(*((c)++)))<<16; \
386 case 3: l|=((unsigned long)(*((c)++)))<<24; \
388 #define HOST_p_c2l_p(c,l,sc,len) { \
390 case 0: l =((unsigned long)(*((c)++))); \
391 if (--len == 0) break; \
392 case 1: l|=((unsigned long)(*((c)++)))<< 8; \
393 if (--len == 0) break; \
394 case 2: l|=((unsigned long)(*((c)++)))<<16; \
396 /* NOTE the pointer is not incremented at the end of this */
397 #define HOST_c2l_p(c,l,n) { \
400 case 3: l =((unsigned long)(*(--(c))))<<16; \
401 case 2: l|=((unsigned long)(*(--(c))))<< 8; \
402 case 1: l|=((unsigned long)(*(--(c)))); \
404 #define HOST_l2c(l,c) (*((c)++)=(unsigned char)(((l) )&0xff), \
405 *((c)++)=(unsigned char)(((l)>> 8)&0xff), \
406 *((c)++)=(unsigned char)(((l)>>16)&0xff), \
407 *((c)++)=(unsigned char)(((l)>>24)&0xff), \
413 * Time for some action:-)
416 int HASH_UPDATE (HASH_CTX *c, const void *data_, unsigned long len)
418 const unsigned char *data=data_;
419 register HASH_LONG * p;
420 register unsigned long l;
423 if (len==0) return 1;
425 l=(c->Nl+(len<<3))&0xffffffffL;
426 /* 95-05-24 eay Fixed a bug with the overflow handling, thanks to
427 * Wei Dai <weidai@eskimo.com> for pointing it out. */
428 if (l < c->Nl) /* overflow */
439 if ((c->num+len) >= HASH_CBLOCK)
441 l=p[sw]; HOST_p_c2l(data,l,sc); p[sw++]=l;
442 for (; sw<HASH_LBLOCK; sw++)
444 HOST_c2l(data,l); p[sw]=l;
446 HASH_BLOCK_HOST_ORDER (c,p,1);
447 len-=(HASH_CBLOCK-c->num);
449 /* drop through and do the rest */
454 if ((sc+len) < 4) /* ugly, add char's to a word */
456 l=p[sw]; HOST_p_c2l_p(data,l,sc,len); p[sw]=l;
464 HOST_p_c2l(data,l,sc);
466 for (; sw < ew; sw++)
468 HOST_c2l(data,l); p[sw]=l;
472 HOST_c2l_p(data,l,ec); p[sw]=l;
482 #if defined(HASH_BLOCK_DATA_ORDER_ALIGNED)
484 * Note that HASH_BLOCK_DATA_ORDER_ALIGNED gets defined
485 * only if sizeof(HASH_LONG)==4.
487 if ((((unsigned long)data)%4) == 0)
489 /* data is properly aligned so that we can cast it: */
490 HASH_BLOCK_DATA_ORDER_ALIGNED (c,(HASH_LONG *)data,sw);
496 #if !defined(HASH_BLOCK_DATA_ORDER)
499 memcpy (p=c->data,data,HASH_CBLOCK);
500 HASH_BLOCK_DATA_ORDER_ALIGNED(c,p,1);
506 #if defined(HASH_BLOCK_DATA_ORDER)
508 HASH_BLOCK_DATA_ORDER(c,data,sw);
520 ew=len>>2; /* words to copy */
524 HOST_c2l(data,l); *p=l;
526 HOST_c2l_p(data,l,ec);
533 void HASH_TRANSFORM (HASH_CTX *c, const unsigned char *data)
535 #if defined(HASH_BLOCK_DATA_ORDER_ALIGNED)
536 if ((((unsigned long)data)%4) == 0)
537 /* data is properly aligned so that we can cast it: */
538 HASH_BLOCK_DATA_ORDER_ALIGNED (c,(HASH_LONG *)data,1);
540 #if !defined(HASH_BLOCK_DATA_ORDER)
542 memcpy (c->data,data,HASH_CBLOCK);
543 HASH_BLOCK_DATA_ORDER_ALIGNED (c,c->data,1);
547 #if defined(HASH_BLOCK_DATA_ORDER)
548 HASH_BLOCK_DATA_ORDER (c,data,1);
553 int HASH_FINAL (unsigned char *md, HASH_CTX *c)
555 register HASH_LONG *p;
556 register unsigned long l;
558 static const unsigned char end[4]={0x80,0x00,0x00,0x00};
559 const unsigned char *cp=end;
564 FIPSerr(FIPS_F_HASH_FINAL,FIPS_R_NON_FIPS_METHOD);
569 /* c->num should definitly have room for at least one more byte. */
575 /* purify often complains about the following line as an
576 * Uninitialized Memory Read. While this can be true, the
577 * following p_c2l macro will reset l when that case is true.
578 * This is because j&0x03 contains the number of 'valid' bytes
579 * already in p[i]. If and only if j&0x03 == 0, the UMR will
580 * occur but this is also the only time p_c2l will do
581 * l= *(cp++) instead of l|= *(cp++)
582 * Many thanks to Alex Tang <altitude@cic.net> for pickup this
585 if (j==0) p[i]=0; /* Yeah, but that's not the way to fix it:-) */
589 l = (j==0) ? 0 : p[i];
591 HOST_p_c2l(cp,l,j); p[i++]=l; /* i is the next 'undefined word' */
593 if (i>(HASH_LBLOCK-2)) /* save room for Nl and Nh */
595 if (i<HASH_LBLOCK) p[i]=0;
596 HASH_BLOCK_HOST_ORDER (c,p,1);
599 for (; i<(HASH_LBLOCK-2); i++)
602 #if defined(DATA_ORDER_IS_BIG_ENDIAN)
603 p[HASH_LBLOCK-2]=c->Nh;
604 p[HASH_LBLOCK-1]=c->Nl;
605 #elif defined(DATA_ORDER_IS_LITTLE_ENDIAN)
606 p[HASH_LBLOCK-2]=c->Nl;
607 p[HASH_LBLOCK-1]=c->Nh;
609 HASH_BLOCK_HOST_ORDER (c,p,1);
611 #ifndef HASH_MAKE_STRING
612 #error "HASH_MAKE_STRING must be defined!"
614 HASH_MAKE_STRING(c,md);
618 /* clear stuff, HASH_BLOCK may be leaving some stuff on the stack
619 * but I'm not worried :-)
620 OPENSSL_cleanse((void *)c,sizeof(HASH_CTX));
626 #define MD32_REG_T long
628 * This comment was originaly written for MD5, which is why it
629 * discusses A-D. But it basically applies to all 32-bit digests,
630 * which is why it was moved to common header file.
632 * In case you wonder why A-D are declared as long and not
633 * as MD5_LONG. Doing so results in slight performance
634 * boost on LP64 architectures. The catch is we don't
635 * really care if 32 MSBs of a 64-bit register get polluted
636 * with eventual overflows as we *save* only 32 LSBs in
637 * *either* case. Now declaring 'em long excuses the compiler
638 * from keeping 32 MSBs zeroed resulting in 13% performance
639 * improvement under SPARC Solaris7/64 and 5% under AlphaLinux.
640 * Well, to be honest it should say that this *prevents*
641 * performance degradation.
642 * <appro@fy.chalmers.se>
643 * Apparently there're LP64 compilers that generate better
644 * code if A-D are declared int. Most notably GCC-x86_64
645 * generates better code.
646 * <appro@fy.chalmers.se>