1 | /*
|
---|
2 | * Copyright 1999-2016 The OpenSSL Project Authors. All Rights Reserved.
|
---|
3 | *
|
---|
4 | * Licensed under the OpenSSL license (the "License"). You may not use
|
---|
5 | * this file except in compliance with the License. You can obtain a copy
|
---|
6 | * in the file LICENSE in the source distribution or at
|
---|
7 | * https://www.openssl.org/source/license.html
|
---|
8 | */
|
---|
9 |
|
---|
10 | /*-
|
---|
11 | * This is a generic 32 bit "collector" for message digest algorithms.
|
---|
12 | * Whenever needed it collects input character stream into chunks of
|
---|
13 | * 32 bit values and invokes a block function that performs actual hash
|
---|
14 | * calculations.
|
---|
15 | *
|
---|
16 | * Porting guide.
|
---|
17 | *
|
---|
18 | * Obligatory macros:
|
---|
19 | *
|
---|
20 | * DATA_ORDER_IS_BIG_ENDIAN or DATA_ORDER_IS_LITTLE_ENDIAN
|
---|
21 | * this macro defines byte order of input stream.
|
---|
22 | * HASH_CBLOCK
|
---|
23 | * size of a unit chunk HASH_BLOCK operates on.
|
---|
24 | * HASH_LONG
|
---|
25 | * has to be at lest 32 bit wide.
|
---|
26 | * HASH_CTX
|
---|
27 | * context structure that at least contains following
|
---|
28 | * members:
|
---|
29 | * typedef struct {
|
---|
30 | * ...
|
---|
31 | * HASH_LONG Nl,Nh;
|
---|
32 | * either {
|
---|
33 | * HASH_LONG data[HASH_LBLOCK];
|
---|
34 | * unsigned char data[HASH_CBLOCK];
|
---|
35 | * };
|
---|
36 | * unsigned int num;
|
---|
37 | * ...
|
---|
38 | * } HASH_CTX;
|
---|
39 | * data[] vector is expected to be zeroed upon first call to
|
---|
40 | * HASH_UPDATE.
|
---|
41 | * HASH_UPDATE
|
---|
42 | * name of "Update" function, implemented here.
|
---|
43 | * HASH_TRANSFORM
|
---|
44 | * name of "Transform" function, implemented here.
|
---|
45 | * HASH_FINAL
|
---|
46 | * name of "Final" function, implemented here.
|
---|
47 | * HASH_BLOCK_DATA_ORDER
|
---|
48 | * name of "block" function capable of treating *unaligned* input
|
---|
49 | * message in original (data) byte order, implemented externally.
|
---|
50 | * HASH_MAKE_STRING
|
---|
51 | * macro convering context variables to an ASCII hash string.
|
---|
52 | *
|
---|
53 | * MD5 example:
|
---|
54 | *
|
---|
55 | * #define DATA_ORDER_IS_LITTLE_ENDIAN
|
---|
56 | *
|
---|
57 | * #define HASH_LONG MD5_LONG
|
---|
58 | * #define HASH_CTX MD5_CTX
|
---|
59 | * #define HASH_CBLOCK MD5_CBLOCK
|
---|
60 | * #define HASH_UPDATE MD5_Update
|
---|
61 | * #define HASH_TRANSFORM MD5_Transform
|
---|
62 | * #define HASH_FINAL MD5_Final
|
---|
63 | * #define HASH_BLOCK_DATA_ORDER md5_block_data_order
|
---|
64 | *
|
---|
65 | * <appro@fy.chalmers.se>
|
---|
66 | */
|
---|
67 |
|
---|
68 | #include <openssl/crypto.h>
|
---|
69 |
|
---|
70 | #if !defined(DATA_ORDER_IS_BIG_ENDIAN) && !defined(DATA_ORDER_IS_LITTLE_ENDIAN)
|
---|
71 | # error "DATA_ORDER must be defined!"
|
---|
72 | #endif
|
---|
73 |
|
---|
74 | #ifndef HASH_CBLOCK
|
---|
75 | # error "HASH_CBLOCK must be defined!"
|
---|
76 | #endif
|
---|
77 | #ifndef HASH_LONG
|
---|
78 | # error "HASH_LONG must be defined!"
|
---|
79 | #endif
|
---|
80 | #ifndef HASH_CTX
|
---|
81 | # error "HASH_CTX must be defined!"
|
---|
82 | #endif
|
---|
83 |
|
---|
84 | #ifndef HASH_UPDATE
|
---|
85 | # error "HASH_UPDATE must be defined!"
|
---|
86 | #endif
|
---|
87 | #ifndef HASH_TRANSFORM
|
---|
88 | # error "HASH_TRANSFORM must be defined!"
|
---|
89 | #endif
|
---|
90 | #ifndef HASH_FINAL
|
---|
91 | # error "HASH_FINAL must be defined!"
|
---|
92 | #endif
|
---|
93 |
|
---|
94 | #ifndef HASH_BLOCK_DATA_ORDER
|
---|
95 | # error "HASH_BLOCK_DATA_ORDER must be defined!"
|
---|
96 | #endif
|
---|
97 |
|
---|
98 | /*
|
---|
99 | * Engage compiler specific rotate intrinsic function if available.
|
---|
100 | */
|
---|
101 | #undef ROTATE
|
---|
102 | #ifndef PEDANTIC
|
---|
103 | # if defined(_MSC_VER)
|
---|
104 | # define ROTATE(a,n) _lrotl(a,n)
|
---|
105 | # elif defined(__ICC)
|
---|
106 | # define ROTATE(a,n) _rotl(a,n)
|
---|
107 | # elif defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
|
---|
108 | /*
|
---|
109 | * Some GNU C inline assembler templates. Note that these are
|
---|
110 | * rotates by *constant* number of bits! But that's exactly
|
---|
111 | * what we need here...
|
---|
112 | * <appro@fy.chalmers.se>
|
---|
113 | */
|
---|
114 | # if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)
|
---|
115 | # define ROTATE(a,n) ({ register unsigned int ret; \
|
---|
116 | asm ( \
|
---|
117 | "roll %1,%0" \
|
---|
118 | : "=r"(ret) \
|
---|
119 | : "I"(n), "0"((unsigned int)(a)) \
|
---|
120 | : "cc"); \
|
---|
121 | ret; \
|
---|
122 | })
|
---|
123 | # elif defined(_ARCH_PPC) || defined(_ARCH_PPC64) || \
|
---|
124 | defined(__powerpc) || defined(__ppc__) || defined(__powerpc64__)
|
---|
125 | # define ROTATE(a,n) ({ register unsigned int ret; \
|
---|
126 | asm ( \
|
---|
127 | "rlwinm %0,%1,%2,0,31" \
|
---|
128 | : "=r"(ret) \
|
---|
129 | : "r"(a), "I"(n)); \
|
---|
130 | ret; \
|
---|
131 | })
|
---|
132 | # elif defined(__s390x__)
|
---|
133 | # define ROTATE(a,n) ({ register unsigned int ret; \
|
---|
134 | asm ("rll %0,%1,%2" \
|
---|
135 | : "=r"(ret) \
|
---|
136 | : "r"(a), "I"(n)); \
|
---|
137 | ret; \
|
---|
138 | })
|
---|
139 | # endif
|
---|
140 | # endif
|
---|
141 | #endif /* PEDANTIC */
|
---|
142 |
|
---|
143 | #ifndef ROTATE
|
---|
144 | # define ROTATE(a,n) (((a)<<(n))|(((a)&0xffffffff)>>(32-(n))))
|
---|
145 | #endif
|
---|
146 |
|
---|
147 | #if defined(DATA_ORDER_IS_BIG_ENDIAN)
|
---|
148 |
|
---|
149 | # ifndef PEDANTIC
|
---|
150 | # if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
|
---|
151 | # if ((defined(__i386) || defined(__i386__)) && !defined(I386_ONLY)) || \
|
---|
152 | (defined(__x86_64) || defined(__x86_64__))
|
---|
153 | # if !defined(B_ENDIAN)
|
---|
154 | /*
|
---|
155 | * This gives ~30-40% performance improvement in SHA-256 compiled
|
---|
156 | * with gcc [on P4]. Well, first macro to be frank. We can pull
|
---|
157 | * this trick on x86* platforms only, because these CPUs can fetch
|
---|
158 | * unaligned data without raising an exception.
|
---|
159 | */
|
---|
160 | # define HOST_c2l(c,l) ({ unsigned int r=*((const unsigned int *)(c)); \
|
---|
161 | asm ("bswapl %0":"=r"(r):"0"(r)); \
|
---|
162 | (c)+=4; (l)=r; })
|
---|
163 | # define HOST_l2c(l,c) ({ unsigned int r=(l); \
|
---|
164 | asm ("bswapl %0":"=r"(r):"0"(r)); \
|
---|
165 | *((unsigned int *)(c))=r; (c)+=4; r; })
|
---|
166 | # endif
|
---|
167 | # elif defined(__aarch64__)
|
---|
168 | # if defined(__BYTE_ORDER__)
|
---|
169 | # if defined(__ORDER_LITTLE_ENDIAN__) && __BYTE_ORDER__==__ORDER_LITTLE_ENDIAN__
|
---|
170 | # define HOST_c2l(c,l) ({ unsigned int r; \
|
---|
171 | asm ("rev %w0,%w1" \
|
---|
172 | :"=r"(r) \
|
---|
173 | :"r"(*((const unsigned int *)(c))));\
|
---|
174 | (c)+=4; (l)=r; })
|
---|
175 | # define HOST_l2c(l,c) ({ unsigned int r; \
|
---|
176 | asm ("rev %w0,%w1" \
|
---|
177 | :"=r"(r) \
|
---|
178 | :"r"((unsigned int)(l)));\
|
---|
179 | *((unsigned int *)(c))=r; (c)+=4; r; })
|
---|
180 | # elif defined(__ORDER_BIG_ENDIAN__) && __BYTE_ORDER__==__ORDER_BIG_ENDIAN__
|
---|
181 | # define HOST_c2l(c,l) ((l)=*((const unsigned int *)(c)), (c)+=4, (l))
|
---|
182 | # define HOST_l2c(l,c) (*((unsigned int *)(c))=(l), (c)+=4, (l))
|
---|
183 | # endif
|
---|
184 | # endif
|
---|
185 | # endif
|
---|
186 | # endif
|
---|
187 | # if defined(__s390__) || defined(__s390x__)
|
---|
188 | # define HOST_c2l(c,l) ((l)=*((const unsigned int *)(c)), (c)+=4, (l))
|
---|
189 | # define HOST_l2c(l,c) (*((unsigned int *)(c))=(l), (c)+=4, (l))
|
---|
190 | # endif
|
---|
191 | # endif
|
---|
192 |
|
---|
193 | # ifndef HOST_c2l
|
---|
194 | # define HOST_c2l(c,l) (l =(((unsigned long)(*((c)++)))<<24), \
|
---|
195 | l|=(((unsigned long)(*((c)++)))<<16), \
|
---|
196 | l|=(((unsigned long)(*((c)++)))<< 8), \
|
---|
197 | l|=(((unsigned long)(*((c)++))) ) )
|
---|
198 | # endif
|
---|
199 | # ifndef HOST_l2c
|
---|
200 | # define HOST_l2c(l,c) (*((c)++)=(unsigned char)(((l)>>24)&0xff), \
|
---|
201 | *((c)++)=(unsigned char)(((l)>>16)&0xff), \
|
---|
202 | *((c)++)=(unsigned char)(((l)>> 8)&0xff), \
|
---|
203 | *((c)++)=(unsigned char)(((l) )&0xff), \
|
---|
204 | l)
|
---|
205 | # endif
|
---|
206 |
|
---|
207 | #elif defined(DATA_ORDER_IS_LITTLE_ENDIAN)
|
---|
208 |
|
---|
209 | # ifndef PEDANTIC
|
---|
210 | # if defined(__GNUC__) && __GNUC__>=2 && !defined(OPENSSL_NO_ASM) && !defined(OPENSSL_NO_INLINE_ASM)
|
---|
211 | # if defined(__s390x__)
|
---|
212 | # define HOST_c2l(c,l) ({ asm ("lrv %0,%1" \
|
---|
213 | :"=d"(l) :"m"(*(const unsigned int *)(c)));\
|
---|
214 | (c)+=4; (l); })
|
---|
215 | # define HOST_l2c(l,c) ({ asm ("strv %1,%0" \
|
---|
216 | :"=m"(*(unsigned int *)(c)) :"d"(l));\
|
---|
217 | (c)+=4; (l); })
|
---|
218 | # endif
|
---|
219 | # endif
|
---|
220 | # if defined(__i386) || defined(__i386__) || defined(__x86_64) || defined(__x86_64__)
|
---|
221 | # ifndef B_ENDIAN
|
---|
222 | /* See comment in DATA_ORDER_IS_BIG_ENDIAN section. */
|
---|
223 | # define HOST_c2l(c,l) ((l)=*((const unsigned int *)(c)), (c)+=4, l)
|
---|
224 | # define HOST_l2c(l,c) (*((unsigned int *)(c))=(l), (c)+=4, l)
|
---|
225 | # endif
|
---|
226 | # endif
|
---|
227 | # endif
|
---|
228 |
|
---|
229 | # ifndef HOST_c2l
|
---|
230 | # define HOST_c2l(c,l) (l =(((unsigned long)(*((c)++))) ), \
|
---|
231 | l|=(((unsigned long)(*((c)++)))<< 8), \
|
---|
232 | l|=(((unsigned long)(*((c)++)))<<16), \
|
---|
233 | l|=(((unsigned long)(*((c)++)))<<24) )
|
---|
234 | # endif
|
---|
235 | # ifndef HOST_l2c
|
---|
236 | # define HOST_l2c(l,c) (*((c)++)=(unsigned char)(((l) )&0xff), \
|
---|
237 | *((c)++)=(unsigned char)(((l)>> 8)&0xff), \
|
---|
238 | *((c)++)=(unsigned char)(((l)>>16)&0xff), \
|
---|
239 | *((c)++)=(unsigned char)(((l)>>24)&0xff), \
|
---|
240 | l)
|
---|
241 | # endif
|
---|
242 |
|
---|
243 | #endif
|
---|
244 |
|
---|
245 | /*
|
---|
246 | * Time for some action:-)
|
---|
247 | */
|
---|
248 |
|
---|
249 | int HASH_UPDATE(HASH_CTX *c, const void *data_, size_t len)
|
---|
250 | {
|
---|
251 | const unsigned char *data = data_;
|
---|
252 | unsigned char *p;
|
---|
253 | HASH_LONG l;
|
---|
254 | size_t n;
|
---|
255 |
|
---|
256 | if (len == 0)
|
---|
257 | return 1;
|
---|
258 |
|
---|
259 | l = (c->Nl + (((HASH_LONG) len) << 3)) & 0xffffffffUL;
|
---|
260 | /*
|
---|
261 | * 95-05-24 eay Fixed a bug with the overflow handling, thanks to Wei Dai
|
---|
262 | * <weidai@eskimo.com> for pointing it out.
|
---|
263 | */
|
---|
264 | if (l < c->Nl) /* overflow */
|
---|
265 | c->Nh++;
|
---|
266 | c->Nh += (HASH_LONG) (len >> 29); /* might cause compiler warning on
|
---|
267 | * 16-bit */
|
---|
268 | c->Nl = l;
|
---|
269 |
|
---|
270 | n = c->num;
|
---|
271 | if (n != 0) {
|
---|
272 | p = (unsigned char *)c->data;
|
---|
273 |
|
---|
274 | if (len >= HASH_CBLOCK || len + n >= HASH_CBLOCK) {
|
---|
275 | memcpy(p + n, data, HASH_CBLOCK - n);
|
---|
276 | HASH_BLOCK_DATA_ORDER(c, p, 1);
|
---|
277 | n = HASH_CBLOCK - n;
|
---|
278 | data += n;
|
---|
279 | len -= n;
|
---|
280 | c->num = 0;
|
---|
281 | /*
|
---|
282 | * We use memset rather than OPENSSL_cleanse() here deliberately.
|
---|
283 | * Using OPENSSL_cleanse() here could be a performance issue. It
|
---|
284 | * will get properly cleansed on finalisation so this isn't a
|
---|
285 | * security problem.
|
---|
286 | */
|
---|
287 | memset(p, 0, HASH_CBLOCK); /* keep it zeroed */
|
---|
288 | } else {
|
---|
289 | memcpy(p + n, data, len);
|
---|
290 | c->num += (unsigned int)len;
|
---|
291 | return 1;
|
---|
292 | }
|
---|
293 | }
|
---|
294 |
|
---|
295 | n = len / HASH_CBLOCK;
|
---|
296 | if (n > 0) {
|
---|
297 | HASH_BLOCK_DATA_ORDER(c, data, n);
|
---|
298 | n *= HASH_CBLOCK;
|
---|
299 | data += n;
|
---|
300 | len -= n;
|
---|
301 | }
|
---|
302 |
|
---|
303 | if (len != 0) {
|
---|
304 | p = (unsigned char *)c->data;
|
---|
305 | c->num = (unsigned int)len;
|
---|
306 | memcpy(p, data, len);
|
---|
307 | }
|
---|
308 | return 1;
|
---|
309 | }
|
---|
310 |
|
---|
311 | void HASH_TRANSFORM(HASH_CTX *c, const unsigned char *data)
|
---|
312 | {
|
---|
313 | HASH_BLOCK_DATA_ORDER(c, data, 1);
|
---|
314 | }
|
---|
315 |
|
---|
316 | int HASH_FINAL(unsigned char *md, HASH_CTX *c)
|
---|
317 | {
|
---|
318 | unsigned char *p = (unsigned char *)c->data;
|
---|
319 | size_t n = c->num;
|
---|
320 |
|
---|
321 | p[n] = 0x80; /* there is always room for one */
|
---|
322 | n++;
|
---|
323 |
|
---|
324 | if (n > (HASH_CBLOCK - 8)) {
|
---|
325 | memset(p + n, 0, HASH_CBLOCK - n);
|
---|
326 | n = 0;
|
---|
327 | HASH_BLOCK_DATA_ORDER(c, p, 1);
|
---|
328 | }
|
---|
329 | memset(p + n, 0, HASH_CBLOCK - 8 - n);
|
---|
330 |
|
---|
331 | p += HASH_CBLOCK - 8;
|
---|
332 | #if defined(DATA_ORDER_IS_BIG_ENDIAN)
|
---|
333 | (void)HOST_l2c(c->Nh, p);
|
---|
334 | (void)HOST_l2c(c->Nl, p);
|
---|
335 | #elif defined(DATA_ORDER_IS_LITTLE_ENDIAN)
|
---|
336 | (void)HOST_l2c(c->Nl, p);
|
---|
337 | (void)HOST_l2c(c->Nh, p);
|
---|
338 | #endif
|
---|
339 | p -= HASH_CBLOCK;
|
---|
340 | HASH_BLOCK_DATA_ORDER(c, p, 1);
|
---|
341 | c->num = 0;
|
---|
342 | OPENSSL_cleanse(p, HASH_CBLOCK);
|
---|
343 |
|
---|
344 | #ifndef HASH_MAKE_STRING
|
---|
345 | # error "HASH_MAKE_STRING must be defined!"
|
---|
346 | #else
|
---|
347 | HASH_MAKE_STRING(c, md);
|
---|
348 | #endif
|
---|
349 |
|
---|
350 | return 1;
|
---|
351 | }
|
---|
352 |
|
---|
353 | #ifndef MD32_REG_T
|
---|
354 | # if defined(__alpha) || defined(__sparcv9) || defined(__mips)
|
---|
355 | # define MD32_REG_T long
|
---|
356 | /*
|
---|
357 | * This comment was originally written for MD5, which is why it
|
---|
358 | * discusses A-D. But it basically applies to all 32-bit digests,
|
---|
359 | * which is why it was moved to common header file.
|
---|
360 | *
|
---|
361 | * In case you wonder why A-D are declared as long and not
|
---|
362 | * as MD5_LONG. Doing so results in slight performance
|
---|
363 | * boost on LP64 architectures. The catch is we don't
|
---|
364 | * really care if 32 MSBs of a 64-bit register get polluted
|
---|
365 | * with eventual overflows as we *save* only 32 LSBs in
|
---|
366 | * *either* case. Now declaring 'em long excuses the compiler
|
---|
367 | * from keeping 32 MSBs zeroed resulting in 13% performance
|
---|
368 | * improvement under SPARC Solaris7/64 and 5% under AlphaLinux.
|
---|
369 | * Well, to be honest it should say that this *prevents*
|
---|
370 | * performance degradation.
|
---|
371 | * <appro@fy.chalmers.se>
|
---|
372 | */
|
---|
373 | # else
|
---|
374 | /*
|
---|
375 | * Above is not absolute and there are LP64 compilers that
|
---|
376 | * generate better code if MD32_REG_T is defined int. The above
|
---|
377 | * pre-processor condition reflects the circumstances under which
|
---|
378 | * the conclusion was made and is subject to further extension.
|
---|
379 | * <appro@fy.chalmers.se>
|
---|
380 | */
|
---|
381 | # define MD32_REG_T int
|
---|
382 | # endif
|
---|
383 | #endif
|
---|