1 /*
2 * Routines to support checksumming of bytes.
3 *
4 * Copyright (C) 1996 Andrew Tridgell
5 * Copyright (C) 1996 Paul Mackerras
6 * Copyright (C) 2004-2020 Wayne Davison
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 3 of the License, or
11 * (at your option) any later version.
12 *
13 * In addition, as a special exception, the copyright holders give
14 * permission to dynamically link rsync with the OpenSSL and xxhash
15 * libraries when those libraries are being distributed in compliance
16 * with their license terms, and to distribute a dynamically linked
17 * combination of rsync and these libraries. This is also considered
18 * to be covered under the GPL's System Libraries exception.
19 *
20 * This program is distributed in the hope that it will be useful,
21 * but WITHOUT ANY WARRANTY; without even the implied warranty of
22 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
23 * GNU General Public License for more details.
24 *
25 * You should have received a copy of the GNU General Public License along
26 * with this program; if not, visit the http://fsf.org website.
27 */
28
29 #include "rsync.h"
30
31 #ifdef SUPPORT_XXHASH
32 #include <xxhash.h>
33 # if XXH_VERSION_NUMBER >= 800
34 # define SUPPORT_XXH3 1
35 # endif
36 #endif
37
38 extern int am_server;
39 extern int whole_file;
40 extern int checksum_seed;
41 extern int protocol_version;
42 extern int proper_seed_order;
43 extern const char *checksum_choice;
44
45 struct name_num_obj valid_checksums = {
46 "checksum", NULL, NULL, 0, 0, {
47 #ifdef SUPPORT_XXH3
48 { CSUM_XXH3_128, "xxh128", NULL },
49 { CSUM_XXH3_64, "xxh3", NULL },
50 #endif
51 #ifdef SUPPORT_XXHASH
52 { CSUM_XXH64, "xxh64", NULL },
53 { CSUM_XXH64, "xxhash", NULL },
54 #endif
55 { CSUM_MD5, "md5", NULL },
56 { CSUM_MD4, "md4", NULL },
57 { CSUM_NONE, "none", NULL },
58 { 0, NULL, NULL }
59 }
60 };
61
62 int xfersum_type = 0; /* used for the file transfer checksums */
63 int checksum_type = 0; /* used for the pre-transfer (--checksum) checksums */
64
parse_csum_name(const char * name,int len)65 int parse_csum_name(const char *name, int len)
66 {
67 struct name_num_item *nni;
68
69 if (len < 0 && name)
70 len = strlen(name);
71
72 if (!name || (len == 4 && strncasecmp(name, "auto", 4) == 0)) {
73 if (protocol_version >= 30)
74 return CSUM_MD5;
75 if (protocol_version >= 27)
76 return CSUM_MD4_OLD;
77 if (protocol_version >= 21)
78 return CSUM_MD4_BUSTED;
79 return CSUM_MD4_ARCHAIC;
80 }
81
82 nni = get_nni_by_name(&valid_checksums, name, len);
83
84 if (!nni) {
85 rprintf(FERROR, "unknown checksum name: %s\n", name);
86 exit_cleanup(RERR_UNSUPPORTED);
87 }
88
89 return nni->num;
90 }
91
checksum_name(int num)92 static const char *checksum_name(int num)
93 {
94 struct name_num_item *nni = get_nni_by_num(&valid_checksums, num);
95
96 return nni ? nni->name : num < CSUM_MD4 ? "md4" : "UNKNOWN";
97 }
98
parse_checksum_choice(int final_call)99 void parse_checksum_choice(int final_call)
100 {
101 if (valid_checksums.negotiated_name)
102 xfersum_type = checksum_type = valid_checksums.negotiated_num;
103 else {
104 char *cp = checksum_choice ? strchr(checksum_choice, ',') : NULL;
105 if (cp) {
106 xfersum_type = parse_csum_name(checksum_choice, cp - checksum_choice);
107 checksum_type = parse_csum_name(cp+1, -1);
108 } else
109 xfersum_type = checksum_type = parse_csum_name(checksum_choice, -1);
110 if (am_server && checksum_choice)
111 validate_choice_vs_env(NSTR_CHECKSUM, xfersum_type, checksum_type);
112 }
113
114 if (xfersum_type == CSUM_NONE)
115 whole_file = 1;
116
117 /* Snag the checksum name for both write_batch's option output & the following debug output. */
118 if (valid_checksums.negotiated_name)
119 checksum_choice = valid_checksums.negotiated_name;
120 else if (checksum_choice == NULL)
121 checksum_choice = checksum_name(xfersum_type);
122
123 if (final_call && DEBUG_GTE(NSTR, am_server ? 3 : 1)) {
124 rprintf(FINFO, "%s%s checksum: %s\n",
125 am_server ? "Server" : "Client",
126 valid_checksums.negotiated_name ? " negotiated" : "",
127 checksum_choice);
128 }
129 }
130
csum_len_for_type(int cst,BOOL flist_csum)131 int csum_len_for_type(int cst, BOOL flist_csum)
132 {
133 switch (cst) {
134 case CSUM_NONE:
135 return 1;
136 case CSUM_MD4_ARCHAIC:
137 /* The oldest checksum code is rather weird: the file-list code only sent
138 * 2-byte checksums, but all other checksums were full MD4 length. */
139 return flist_csum ? 2 : MD4_DIGEST_LEN;
140 case CSUM_MD4:
141 case CSUM_MD4_OLD:
142 case CSUM_MD4_BUSTED:
143 return MD4_DIGEST_LEN;
144 case CSUM_MD5:
145 return MD5_DIGEST_LEN;
146 case CSUM_XXH64:
147 case CSUM_XXH3_64:
148 return 64/8;
149 case CSUM_XXH3_128:
150 return 128/8;
151 default: /* paranoia to prevent missing case values */
152 exit_cleanup(RERR_UNSUPPORTED);
153 }
154 return 0;
155 }
156
157 /* Returns 0 if the checksum is not canonical (i.e. it includes a seed value).
158 * Returns 1 if the public sum order matches our internal sum order.
159 * Returns -1 if the public sum order is the reverse of our internal sum order.
160 */
canonical_checksum(int csum_type)161 int canonical_checksum(int csum_type)
162 {
163 switch (csum_type) {
164 case CSUM_NONE:
165 case CSUM_MD4_ARCHAIC:
166 case CSUM_MD4_OLD:
167 case CSUM_MD4_BUSTED:
168 break;
169 case CSUM_MD4:
170 case CSUM_MD5:
171 return -1;
172 case CSUM_XXH64:
173 case CSUM_XXH3_64:
174 case CSUM_XXH3_128:
175 return 1;
176 default: /* paranoia to prevent missing case values */
177 exit_cleanup(RERR_UNSUPPORTED);
178 }
179 return 0;
180 }
181
182 #ifndef HAVE_SIMD /* See simd-checksum-*.cpp. */
183 /*
184 a simple 32 bit checksum that can be updated from either end
185 (inspired by Mark Adler's Adler-32 checksum)
186 */
get_checksum1(char * buf1,int32 len)187 uint32 get_checksum1(char *buf1, int32 len)
188 {
189 int32 i;
190 uint32 s1, s2;
191 schar *buf = (schar *)buf1;
192
193 s1 = s2 = 0;
194 for (i = 0; i < (len-4); i+=4) {
195 s2 += 4*(s1 + buf[i]) + 3*buf[i+1] + 2*buf[i+2] + buf[i+3] + 10*CHAR_OFFSET;
196 s1 += (buf[i+0] + buf[i+1] + buf[i+2] + buf[i+3] + 4*CHAR_OFFSET);
197 }
198 for (; i < len; i++) {
199 s1 += (buf[i]+CHAR_OFFSET); s2 += s1;
200 }
201 return (s1 & 0xffff) + (s2 << 16);
202 }
203 #endif
204
get_checksum2(char * buf,int32 len,char * sum)205 void get_checksum2(char *buf, int32 len, char *sum)
206 {
207 switch (xfersum_type) {
208 #ifdef SUPPORT_XXHASH
209 case CSUM_XXH64:
210 SIVAL64(sum, 0, XXH64(buf, len, checksum_seed));
211 break;
212 #endif
213 #ifdef SUPPORT_XXH3
214 case CSUM_XXH3_64:
215 SIVAL64(sum, 0, XXH3_64bits_withSeed(buf, len, checksum_seed));
216 break;
217 case CSUM_XXH3_128: {
218 XXH128_hash_t digest = XXH3_128bits_withSeed(buf, len, checksum_seed);
219 SIVAL64(sum, 0, digest.low64);
220 SIVAL64(sum, 8, digest.high64);
221 break;
222 }
223 #endif
224 case CSUM_MD5: {
225 MD5_CTX m5;
226 uchar seedbuf[4];
227 MD5_Init(&m5);
228 if (proper_seed_order) {
229 if (checksum_seed) {
230 SIVALu(seedbuf, 0, checksum_seed);
231 MD5_Update(&m5, seedbuf, 4);
232 }
233 MD5_Update(&m5, (uchar *)buf, len);
234 } else {
235 MD5_Update(&m5, (uchar *)buf, len);
236 if (checksum_seed) {
237 SIVALu(seedbuf, 0, checksum_seed);
238 MD5_Update(&m5, seedbuf, 4);
239 }
240 }
241 MD5_Final((uchar *)sum, &m5);
242 break;
243 }
244 case CSUM_MD4:
245 #ifdef USE_OPENSSL
246 {
247 MD4_CTX m4;
248 MD4_Init(&m4);
249 MD4_Update(&m4, (uchar *)buf, len);
250 if (checksum_seed) {
251 uchar seedbuf[4];
252 SIVALu(seedbuf, 0, checksum_seed);
253 MD4_Update(&m4, seedbuf, 4);
254 }
255 MD4_Final((uchar *)sum, &m4);
256 break;
257 }
258 #endif
259 case CSUM_MD4_OLD:
260 case CSUM_MD4_BUSTED:
261 case CSUM_MD4_ARCHAIC: {
262 md_context m;
263 int32 i;
264 static char *buf1;
265 static int32 len1;
266
267 mdfour_begin(&m);
268
269 if (len > len1) {
270 if (buf1)
271 free(buf1);
272 buf1 = new_array(char, len+4);
273 len1 = len;
274 }
275
276 memcpy(buf1, buf, len);
277 if (checksum_seed) {
278 SIVAL(buf1,len,checksum_seed);
279 len += 4;
280 }
281
282 for (i = 0; i + CSUM_CHUNK <= len; i += CSUM_CHUNK)
283 mdfour_update(&m, (uchar *)(buf1+i), CSUM_CHUNK);
284
285 /*
286 * Prior to version 27 an incorrect MD4 checksum was computed
287 * by failing to call mdfour_tail() for block sizes that
288 * are multiples of 64. This is fixed by calling mdfour_update()
289 * even when there are no more bytes.
290 */
291 if (len - i > 0 || xfersum_type > CSUM_MD4_BUSTED)
292 mdfour_update(&m, (uchar *)(buf1+i), len-i);
293
294 mdfour_result(&m, (uchar *)sum);
295 break;
296 }
297 default: /* paranoia to prevent missing case values */
298 exit_cleanup(RERR_UNSUPPORTED);
299 }
300 }
301
file_checksum(const char * fname,const STRUCT_STAT * st_p,char * sum)302 void file_checksum(const char *fname, const STRUCT_STAT *st_p, char *sum)
303 {
304 struct map_struct *buf;
305 OFF_T i, len = st_p->st_size;
306 int32 remainder;
307 int fd;
308
309 memset(sum, 0, MAX_DIGEST_LEN);
310
311 fd = do_open(fname, O_RDONLY, 0);
312 if (fd == -1)
313 return;
314
315 buf = map_file(fd, len, MAX_MAP_SIZE, CHUNK_SIZE);
316
317 switch (checksum_type) {
318 #ifdef SUPPORT_XXHASH
319 case CSUM_XXH64: {
320 static XXH64_state_t* state = NULL;
321 if (!state && !(state = XXH64_createState()))
322 out_of_memory("file_checksum");
323
324 XXH64_reset(state, 0);
325
326 for (i = 0; i + CHUNK_SIZE <= len; i += CHUNK_SIZE)
327 XXH64_update(state, (uchar *)map_ptr(buf, i, CHUNK_SIZE), CHUNK_SIZE);
328
329 remainder = (int32)(len - i);
330 if (remainder > 0)
331 XXH64_update(state, (uchar *)map_ptr(buf, i, remainder), remainder);
332
333 SIVAL64(sum, 0, XXH64_digest(state));
334 break;
335 }
336 #endif
337 #ifdef SUPPORT_XXH3
338 case CSUM_XXH3_64: {
339 static XXH3_state_t* state = NULL;
340 if (!state && !(state = XXH3_createState()))
341 out_of_memory("file_checksum");
342
343 XXH3_64bits_reset(state);
344
345 for (i = 0; i + CHUNK_SIZE <= len; i += CHUNK_SIZE)
346 XXH3_64bits_update(state, (uchar *)map_ptr(buf, i, CHUNK_SIZE), CHUNK_SIZE);
347
348 remainder = (int32)(len - i);
349 if (remainder > 0)
350 XXH3_64bits_update(state, (uchar *)map_ptr(buf, i, remainder), remainder);
351
352 SIVAL64(sum, 0, XXH3_64bits_digest(state));
353 break;
354 }
355 case CSUM_XXH3_128: {
356 XXH128_hash_t digest;
357 static XXH3_state_t* state = NULL;
358 if (!state && !(state = XXH3_createState()))
359 out_of_memory("file_checksum");
360
361 XXH3_128bits_reset(state);
362
363 for (i = 0; i + CHUNK_SIZE <= len; i += CHUNK_SIZE)
364 XXH3_128bits_update(state, (uchar *)map_ptr(buf, i, CHUNK_SIZE), CHUNK_SIZE);
365
366 remainder = (int32)(len - i);
367 if (remainder > 0)
368 XXH3_128bits_update(state, (uchar *)map_ptr(buf, i, remainder), remainder);
369
370 digest = XXH3_128bits_digest(state);
371 SIVAL64(sum, 0, digest.low64);
372 SIVAL64(sum, 8, digest.high64);
373 break;
374 }
375 #endif
376 case CSUM_MD5: {
377 MD5_CTX m5;
378
379 MD5_Init(&m5);
380
381 for (i = 0; i + CHUNK_SIZE <= len; i += CHUNK_SIZE)
382 MD5_Update(&m5, (uchar *)map_ptr(buf, i, CHUNK_SIZE), CHUNK_SIZE);
383
384 remainder = (int32)(len - i);
385 if (remainder > 0)
386 MD5_Update(&m5, (uchar *)map_ptr(buf, i, remainder), remainder);
387
388 MD5_Final((uchar *)sum, &m5);
389 break;
390 }
391 case CSUM_MD4:
392 #ifdef USE_OPENSSL
393 {
394 MD4_CTX m4;
395
396 MD4_Init(&m4);
397
398 for (i = 0; i + CHUNK_SIZE <= len; i += CHUNK_SIZE)
399 MD4_Update(&m4, (uchar *)map_ptr(buf, i, CHUNK_SIZE), CHUNK_SIZE);
400
401 remainder = (int32)(len - i);
402 if (remainder > 0)
403 MD4_Update(&m4, (uchar *)map_ptr(buf, i, remainder), remainder);
404
405 MD4_Final((uchar *)sum, &m4);
406 break;
407 }
408 #endif
409 case CSUM_MD4_OLD:
410 case CSUM_MD4_BUSTED:
411 case CSUM_MD4_ARCHAIC: {
412 md_context m;
413
414 mdfour_begin(&m);
415
416 for (i = 0; i + CHUNK_SIZE <= len; i += CHUNK_SIZE)
417 mdfour_update(&m, (uchar *)map_ptr(buf, i, CHUNK_SIZE), CHUNK_SIZE);
418
419 /* Prior to version 27 an incorrect MD4 checksum was computed
420 * by failing to call mdfour_tail() for block sizes that
421 * are multiples of 64. This is fixed by calling mdfour_update()
422 * even when there are no more bytes. */
423 remainder = (int32)(len - i);
424 if (remainder > 0 || checksum_type > CSUM_MD4_BUSTED)
425 mdfour_update(&m, (uchar *)map_ptr(buf, i, remainder), remainder);
426
427 mdfour_result(&m, (uchar *)sum);
428 break;
429 }
430 default:
431 rprintf(FERROR, "Invalid checksum-choice for --checksum: %s (%d)\n",
432 checksum_name(checksum_type), checksum_type);
433 exit_cleanup(RERR_UNSUPPORTED);
434 }
435
436 close(fd);
437 unmap_file(buf);
438 }
439
440 static int32 sumresidue;
441 static union {
442 md_context md;
443 #ifdef USE_OPENSSL
444 MD4_CTX m4;
445 #endif
446 MD5_CTX m5;
447 } ctx;
448 #ifdef SUPPORT_XXHASH
449 static XXH64_state_t* xxh64_state;
450 #endif
451 #ifdef SUPPORT_XXH3
452 static XXH3_state_t* xxh3_state;
453 #endif
454 static int cursum_type;
455
sum_init(int csum_type,int seed)456 void sum_init(int csum_type, int seed)
457 {
458 char s[4];
459
460 if (csum_type < 0)
461 csum_type = parse_csum_name(NULL, 0);
462 cursum_type = csum_type;
463
464 switch (csum_type) {
465 #ifdef SUPPORT_XXHASH
466 case CSUM_XXH64:
467 if (!xxh64_state && !(xxh64_state = XXH64_createState()))
468 out_of_memory("sum_init");
469 XXH64_reset(xxh64_state, 0);
470 break;
471 #endif
472 #ifdef SUPPORT_XXH3
473 case CSUM_XXH3_64:
474 if (!xxh3_state && !(xxh3_state = XXH3_createState()))
475 out_of_memory("sum_init");
476 XXH3_64bits_reset(xxh3_state);
477 break;
478 case CSUM_XXH3_128:
479 if (!xxh3_state && !(xxh3_state = XXH3_createState()))
480 out_of_memory("sum_init");
481 XXH3_128bits_reset(xxh3_state);
482 break;
483 #endif
484 case CSUM_MD5:
485 MD5_Init(&ctx.m5);
486 break;
487 case CSUM_MD4:
488 #ifdef USE_OPENSSL
489 MD4_Init(&ctx.m4);
490 #else
491 mdfour_begin(&ctx.md);
492 sumresidue = 0;
493 #endif
494 break;
495 case CSUM_MD4_OLD:
496 case CSUM_MD4_BUSTED:
497 case CSUM_MD4_ARCHAIC:
498 mdfour_begin(&ctx.md);
499 sumresidue = 0;
500 SIVAL(s, 0, seed);
501 sum_update(s, 4);
502 break;
503 case CSUM_NONE:
504 break;
505 default: /* paranoia to prevent missing case values */
506 exit_cleanup(RERR_UNSUPPORTED);
507 }
508 }
509
510 /**
511 * Feed data into an MD4 accumulator, md. The results may be
512 * retrieved using sum_end(). md is used for different purposes at
513 * different points during execution.
514 *
515 * @todo Perhaps get rid of md and just pass in the address each time.
516 * Very slightly clearer and slower.
517 **/
sum_update(const char * p,int32 len)518 void sum_update(const char *p, int32 len)
519 {
520 switch (cursum_type) {
521 #ifdef SUPPORT_XXHASH
522 case CSUM_XXH64:
523 XXH64_update(xxh64_state, p, len);
524 break;
525 #endif
526 #ifdef SUPPORT_XXH3
527 case CSUM_XXH3_64:
528 XXH3_64bits_update(xxh3_state, p, len);
529 break;
530 case CSUM_XXH3_128:
531 XXH3_128bits_update(xxh3_state, p, len);
532 break;
533 #endif
534 case CSUM_MD5:
535 MD5_Update(&ctx.m5, (uchar *)p, len);
536 break;
537 case CSUM_MD4:
538 #ifdef USE_OPENSSL
539 MD4_Update(&ctx.m4, (uchar *)p, len);
540 break;
541 #endif
542 case CSUM_MD4_OLD:
543 case CSUM_MD4_BUSTED:
544 case CSUM_MD4_ARCHAIC:
545 if (len + sumresidue < CSUM_CHUNK) {
546 memcpy(ctx.md.buffer + sumresidue, p, len);
547 sumresidue += len;
548 break;
549 }
550
551 if (sumresidue) {
552 int32 i = CSUM_CHUNK - sumresidue;
553 memcpy(ctx.md.buffer + sumresidue, p, i);
554 mdfour_update(&ctx.md, (uchar *)ctx.md.buffer, CSUM_CHUNK);
555 len -= i;
556 p += i;
557 }
558
559 while (len >= CSUM_CHUNK) {
560 mdfour_update(&ctx.md, (uchar *)p, CSUM_CHUNK);
561 len -= CSUM_CHUNK;
562 p += CSUM_CHUNK;
563 }
564
565 sumresidue = len;
566 if (sumresidue)
567 memcpy(ctx.md.buffer, p, sumresidue);
568 break;
569 case CSUM_NONE:
570 break;
571 default: /* paranoia to prevent missing case values */
572 exit_cleanup(RERR_UNSUPPORTED);
573 }
574 }
575
576 /* NOTE: all the callers of sum_end() pass in a pointer to a buffer that is
577 * MAX_DIGEST_LEN in size, so even if the csum-len is shorter that that (i.e.
578 * CSUM_MD4_ARCHAIC), we don't have to worry about limiting the data we write
579 * into the "sum" buffer. */
sum_end(char * sum)580 int sum_end(char *sum)
581 {
582 switch (cursum_type) {
583 #ifdef SUPPORT_XXHASH
584 case CSUM_XXH64:
585 SIVAL64(sum, 0, XXH64_digest(xxh64_state));
586 break;
587 #endif
588 #ifdef SUPPORT_XXH3
589 case CSUM_XXH3_64:
590 SIVAL64(sum, 0, XXH3_64bits_digest(xxh3_state));
591 break;
592 case CSUM_XXH3_128: {
593 XXH128_hash_t digest = XXH3_128bits_digest(xxh3_state);
594 SIVAL64(sum, 0, digest.low64);
595 SIVAL64(sum, 8, digest.high64);
596 break;
597 }
598 #endif
599 case CSUM_MD5:
600 MD5_Final((uchar *)sum, &ctx.m5);
601 break;
602 case CSUM_MD4:
603 #ifdef USE_OPENSSL
604 MD4_Final((uchar *)sum, &ctx.m4);
605 break;
606 #endif
607 case CSUM_MD4_OLD:
608 mdfour_update(&ctx.md, (uchar *)ctx.md.buffer, sumresidue);
609 mdfour_result(&ctx.md, (uchar *)sum);
610 break;
611 case CSUM_MD4_BUSTED:
612 case CSUM_MD4_ARCHAIC:
613 if (sumresidue)
614 mdfour_update(&ctx.md, (uchar *)ctx.md.buffer, sumresidue);
615 mdfour_result(&ctx.md, (uchar *)sum);
616 break;
617 case CSUM_NONE:
618 *sum = '\0';
619 break;
620 default: /* paranoia to prevent missing case values */
621 exit_cleanup(RERR_UNSUPPORTED);
622 }
623
624 return csum_len_for_type(cursum_type, 0);
625 }
626