?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
fix typo in manual page
[rsync.git] / token.c
blobc108b3af576e1a8c343b062e3552627ea7aa0373
1 /*
2 * Routines used by the file-transfer code.
3 *
4 * Copyright (C) 1996 Andrew Tridgell
5 * Copyright (C) 1996 Paul Mackerras
6 * Copyright (C) 2003-2022 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 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License along
19 * with this program; if not, visit the http://fsf.org website.
20 */
21
22 #include "rsync.h"
23 #include "itypes.h"
24 #include <zlib.h>
25 #ifdef SUPPORT_ZSTD
26 #include <zstd.h>
27 #endif
28 #ifdef SUPPORT_LZ4
29 #include <lz4.h>
30 #endif
31
32 extern int do_compression;
33 extern int protocol_version;
34 extern int module_id;
35 extern int do_compression_level;
36 extern char *skip_compress;
37
38 #ifndef Z_INSERT_ONLY
39 #define Z_INSERT_ONLY Z_SYNC_FLUSH
40 #endif
41
42 static int skip_compression_level; /* The least possible compressing for handling skip-compress files. */
43 static int per_file_default_level; /* The default level that each new file gets prior to checking its suffix. */
44
45 struct suffix_tree {
46 struct suffix_tree *sibling;
47 struct suffix_tree *child;
48 char letter, word_end;
49 };
50
51 static char *match_list;
52 static struct suffix_tree *suftree;
53
54 void init_compression_level(void)
55 {
56 int min_level, max_level, def_level, off_level;
57
58 switch (do_compression) {
59 case CPRES_NONE:
60 return;
61 case CPRES_ZLIB:
62 case CPRES_ZLIBX:
63 min_level = 1;
64 max_level = Z_BEST_COMPRESSION;
65 def_level = 6; /* Z_DEFAULT_COMPRESSION is -1, so set it to the real default */
66 off_level = skip_compression_level = Z_NO_COMPRESSION;
67 if (do_compression_level == Z_DEFAULT_COMPRESSION)
68 do_compression_level = def_level;
69 break;
70 #ifdef SUPPORT_ZSTD
71 case CPRES_ZSTD:
72 min_level = skip_compression_level = ZSTD_minCLevel();
73 max_level = ZSTD_maxCLevel();
74 def_level = ZSTD_CLEVEL_DEFAULT;
75 off_level = CLVL_NOT_SPECIFIED;
76 if (do_compression_level == 0)
77 do_compression_level = def_level;
78 break;
79 #endif
80 #ifdef SUPPORT_LZ4
81 case CPRES_LZ4:
82 min_level = skip_compression_level = 0;
83 max_level = 0;
84 def_level = 0;
85 off_level = CLVL_NOT_SPECIFIED;
86 break;
87 #endif
88 default: /* paranoia to prevent missing case values */
89 NOISY_DEATH("Unknown do_compression value");
90 }
91
92 if (do_compression_level == CLVL_NOT_SPECIFIED)
93 do_compression_level = def_level;
94 else if (do_compression_level == off_level) {
95 do_compression = CPRES_NONE;
96 return;
97 }
98
99 /* We don't bother with any errors or warnings -- just make sure that the values are valid. */
100 if (do_compression_level < min_level)
101 do_compression_level = min_level;
102 else if (do_compression_level > max_level)
103 do_compression_level = max_level;
104 }
105
106 static void add_suffix(struct suffix_tree **prior, char ltr, const char *str)
107 {
108 struct suffix_tree *node, *newnode;
109
110 if (ltr == '[') {
111 const char *after = strchr(str, ']');
112 /* Treat "[foo" and "[]" as having a literal '['. */
113 if (after && after++ != str+1) {
114 while ((ltr = *str++) != ']')
115 add_suffix(prior, ltr, after);
116 return;
117 }
118 }
119
120 for (node = *prior; node; prior = &node->sibling, node = node->sibling) {
121 if (node->letter == ltr) {
122 if (*str)
123 add_suffix(&node->child, *str, str+1);
124 else
125 node->word_end = 1;
126 return;
127 }
128 if (node->letter > ltr)
129 break;
130 }
131 newnode = new(struct suffix_tree);
132 newnode->sibling = node;
133 newnode->child = NULL;
134 newnode->letter = ltr;
135 *prior = newnode;
136 if (*str) {
137 add_suffix(&newnode->child, *str, str+1);
138 newnode->word_end = 0;
139 } else
140 newnode->word_end = 1;
141 }
142
143 static void add_nocompress_suffixes(const char *str)
144 {
145 char *buf, *t;
146 const char *f = str;
147
148 buf = new_array(char, strlen(f) + 1);
149
150 while (*f) {
151 if (*f == '/') {
152 f++;
153 continue;
154 }
155
156 t = buf;
157 do {
158 if (isUpper(f))
159 *t++ = toLower(f);
160 else
161 *t++ = *f;
162 } while (*++f != '/' && *f);
163 *t++ = '\0';
164
165 add_suffix(&suftree, *buf, buf+1);
166 }
167
168 free(buf);
169 }
170
171 static void init_set_compression(void)
172 {
173 const char *f;
174 char *t, *start;
175
176 if (skip_compress)
177 add_nocompress_suffixes(skip_compress);
178
179 /* A non-daemon transfer skips the default suffix list if the
180 * user specified --skip-compress. */
181 if (skip_compress && module_id < 0)
182 f = "";
183 else
184 f = lp_dont_compress(module_id);
185
186 match_list = t = new_array(char, strlen(f) + 2);
187
188 per_file_default_level = do_compression_level;
189
190 while (*f) {
191 if (*f == ' ') {
192 f++;
193 continue;
194 }
195
196 start = t;
197 do {
198 if (isUpper(f))
199 *t++ = toLower(f);
200 else
201 *t++ = *f;
202 } while (*++f != ' ' && *f);
203 *t++ = '\0';
204
205 if (t - start == 1+1 && *start == '*') {
206 /* Optimize a match-string of "*". */
207 *match_list = '\0';
208 suftree = NULL;
209 per_file_default_level = skip_compression_level;
210 break;
211 }
212
213 /* Move *.foo items into the stuffix tree. */
214 if (*start == '*' && start[1] == '.' && start[2]
215 && !strpbrk(start+2, ".?*")) {
216 add_suffix(&suftree, start[2], start+3);
217 t = start;
218 }
219 }
220 *t++ = '\0';
221 }
222
223 /* determine the compression level based on a wildcard filename list */
224 void set_compression(const char *fname)
225 {
226 #if 0 /* No compression algorithms currently allow mid-stream changing of the level. */
227 const struct suffix_tree *node;
228 const char *s;
229 char ltr;
230 #endif
231
232 if (!do_compression)
233 return;
234
235 if (!match_list)
236 init_set_compression();
237
238 #if 0
239 compression_level = per_file_default_level;
240
241 if (!*match_list && !suftree)
242 return;
243
244 if ((s = strrchr(fname, '/')) != NULL)
245 fname = s + 1;
246
247 for (s = match_list; *s; s += strlen(s) + 1) {
248 if (iwildmatch(s, fname)) {
249 compression_level = skip_compression_level;
250 return;
251 }
252 }
253
254 if (!(node = suftree) || !(s = strrchr(fname, '.'))
255 || s == fname || !(ltr = *++s))
256 return;
257
258 while (1) {
259 if (isUpper(&ltr))
260 ltr = toLower(&ltr);
261 while (node->letter != ltr) {
262 if (node->letter > ltr)
263 return;
264 if (!(node = node->sibling))
265 return;
266 }
267 if ((ltr = *++s) == '\0') {
268 if (node->word_end)
269 compression_level = skip_compression_level;
270 return;
271 }
272 if (!(node = node->child))
273 return;
274 }
275 #else
276 (void)fname;
277 #endif
278 }
279
280 /* non-compressing recv token */
281 static int32 simple_recv_token(int f, char **data)
282 {
283 static int32 residue;
284 static char *buf;
285 int32 n;
286
287 if (!buf)
288 buf = new_array(char, CHUNK_SIZE);
289
290 if (residue == 0) {
291 int32 i = read_int(f);
292 if (i <= 0)
293 return i;
294 residue = i;
295 }
296
297 *data = buf;
298 n = MIN(CHUNK_SIZE,residue);
299 residue -= n;
300 read_buf(f,buf,n);
301 return n;
302 }
303
304 /* non-compressing send token */
305 static void simple_send_token(int f, int32 token, struct map_struct *buf, OFF_T offset, int32 n)
306 {
307 if (n > 0) {
308 int32 len = 0;
309 while (len < n) {
310 int32 n1 = MIN(CHUNK_SIZE, n-len);
311 write_int(f, n1);
312 write_buf(f, map_ptr(buf, offset+len, n1), n1);
313 len += n1;
314 }
315 }
316 /* a -2 token means to send data only and no token */
317 if (token != -2)
318 write_int(f, -(token+1));
319 }
320
321 /* Flag bytes in compressed stream are encoded as follows: */
322 #define END_FLAG 0 /* that's all folks */
323 #define TOKEN_LONG 0x20 /* followed by 32-bit token number */
324 #define TOKENRUN_LONG 0x21 /* ditto with 16-bit run count */
325 #define DEFLATED_DATA 0x40 /* + 6-bit high len, then low len byte */
326 #define TOKEN_REL 0x80 /* + 6-bit relative token number */
327 #define TOKENRUN_REL 0xc0 /* ditto with 16-bit run count */
328
329 #define MAX_DATA_COUNT 16383 /* fit 14 bit count into 2 bytes with flags */
330
331 /* zlib.h says that if we want to be able to compress something in a single
332 * call, avail_out must be at least 0.1% larger than avail_in plus 12 bytes.
333 * We'll add in 0.1%+16, just to be safe (and we'll avoid floating point,
334 * to ensure that this is a compile-time value). */
335 #define AVAIL_OUT_SIZE(avail_in_size) ((avail_in_size)*1001/1000+16)
336
337 /* For coding runs of tokens */
338 static int32 last_token = -1;
339 static int32 run_start;
340 static int32 last_run_end;
341
342 /* Deflation state */
343 static z_stream tx_strm;
344
345 /* Output buffer */
346 static char *obuf;
347
348 /* We want obuf to be able to hold both MAX_DATA_COUNT+2 bytes as well as
349 * AVAIL_OUT_SIZE(CHUNK_SIZE) bytes, so make sure that it's large enough. */
350 #if MAX_DATA_COUNT+2 > AVAIL_OUT_SIZE(CHUNK_SIZE)
351 #define OBUF_SIZE (MAX_DATA_COUNT+2)
352 #else
353 #define OBUF_SIZE AVAIL_OUT_SIZE(CHUNK_SIZE)
354 #endif
355
356 /* Send a deflated token */
357 static void
358 send_deflated_token(int f, int32 token, struct map_struct *buf, OFF_T offset, int32 nb, int32 toklen)
359 {
360 static int init_done, flush_pending;
361 int32 n, r;
362
363 if (last_token == -1) {
364 /* initialization */
365 if (!init_done) {
366 tx_strm.next_in = NULL;
367 tx_strm.zalloc = NULL;
368 tx_strm.zfree = NULL;
369 if (deflateInit2(&tx_strm, per_file_default_level,
370 Z_DEFLATED, -15, 8,
371 Z_DEFAULT_STRATEGY) != Z_OK) {
372 rprintf(FERROR, "compression init failed\n");
373 exit_cleanup(RERR_PROTOCOL);
374 }
375 obuf = new_array(char, OBUF_SIZE);
376 init_done = 1;
377 } else
378 deflateReset(&tx_strm);
379 last_run_end = 0;
380 run_start = token;
381 flush_pending = 0;
382 } else if (last_token == -2) {
383 run_start = token;
384 } else if (nb != 0 || token != last_token + 1 || token >= run_start + 65536) {
385 /* output previous run */
386 r = run_start - last_run_end;
387 n = last_token - run_start;
388 if (r >= 0 && r <= 63) {
389 write_byte(f, (n==0? TOKEN_REL: TOKENRUN_REL) + r);
390 } else {
391 write_byte(f, (n==0? TOKEN_LONG: TOKENRUN_LONG));
392 write_int(f, run_start);
393 }
394 if (n != 0) {
395 write_byte(f, n);
396 write_byte(f, n >> 8);
397 }
398 last_run_end = last_token;
399 run_start = token;
400 }
401
402 last_token = token;
403
404 if (nb != 0 || flush_pending) {
405 /* deflate the data starting at offset */
406 int flush = Z_NO_FLUSH;
407 tx_strm.avail_in = 0;
408 tx_strm.avail_out = 0;
409 do {
410 if (tx_strm.avail_in == 0 && nb != 0) {
411 /* give it some more input */
412 n = MIN(nb, CHUNK_SIZE);
413 tx_strm.next_in = (Bytef *)
414 map_ptr(buf, offset, n);
415 tx_strm.avail_in = n;
416 nb -= n;
417 offset += n;
418 }
419 if (tx_strm.avail_out == 0) {
420 tx_strm.next_out = (Bytef *)(obuf + 2);
421 tx_strm.avail_out = MAX_DATA_COUNT;
422 if (flush != Z_NO_FLUSH) {
423 /*
424 * We left the last 4 bytes in the
425 * buffer, in case they are the
426 * last 4. Move them to the front.
427 */
428 memcpy(tx_strm.next_out, obuf+MAX_DATA_COUNT-2, 4);
429 tx_strm.next_out += 4;
430 tx_strm.avail_out -= 4;
431 }
432 }
433 if (nb == 0 && token != -2)
434 flush = Z_SYNC_FLUSH;
435 r = deflate(&tx_strm, flush);
436 if (r != Z_OK) {
437 rprintf(FERROR, "deflate returned %d\n", r);
438 exit_cleanup(RERR_STREAMIO);
439 }
440 if (nb == 0 || tx_strm.avail_out == 0) {
441 n = MAX_DATA_COUNT - tx_strm.avail_out;
442 if (flush != Z_NO_FLUSH) {
443 /*
444 * We have to trim off the last 4
445 * bytes of output when flushing
446 * (they are just 0, 0, ff, ff).
447 */
448 n -= 4;
449 }
450 if (n > 0) {
451 obuf[0] = DEFLATED_DATA + (n >> 8);
452 obuf[1] = n;
453 write_buf(f, obuf, n+2);
454 }
455 }
456 } while (nb != 0 || tx_strm.avail_out == 0);
457 flush_pending = token == -2;
458 }
459
460 if (token == -1) {
461 /* end of file - clean up */
462 write_byte(f, END_FLAG);
463 } else if (token != -2 && do_compression == CPRES_ZLIB) {
464 /* Add the data in the current block to the compressor's
465 * history and hash table. */
466 do {
467 /* Break up long sections in the same way that
468 * see_deflate_token() does. */
469 int32 n1 = toklen > 0xffff ? 0xffff : toklen;
470 toklen -= n1;
471 tx_strm.next_in = (Bytef *)map_ptr(buf, offset, n1);
472 tx_strm.avail_in = n1;
473 if (protocol_version >= 31) /* Newer protocols avoid a data-duplicating bug */
474 offset += n1;
475 tx_strm.next_out = (Bytef *) obuf;
476 tx_strm.avail_out = AVAIL_OUT_SIZE(CHUNK_SIZE);
477 r = deflate(&tx_strm, Z_INSERT_ONLY);
478 if (r != Z_OK || tx_strm.avail_in != 0) {
479 rprintf(FERROR, "deflate on token returned %d (%d bytes left)\n",
480 r, tx_strm.avail_in);
481 exit_cleanup(RERR_STREAMIO);
482 }
483 } while (toklen > 0);
484 }
485 }
486
487 /* tells us what the receiver is in the middle of doing */
488 static enum { r_init, r_idle, r_running, r_inflating, r_inflated } recv_state;
489
490 /* for inflating stuff */
491 static z_stream rx_strm;
492 static char *cbuf;
493 static char *dbuf;
494
495 /* for decoding runs of tokens */
496 static int32 rx_token;
497 static int32 rx_run;
498
499 /* Receive a deflated token and inflate it */
500 static int32 recv_deflated_token(int f, char **data)
501 {
502 static int init_done;
503 static int32 saved_flag;
504 int32 n, flag;
505 int r;
506
507 for (;;) {
508 switch (recv_state) {
509 case r_init:
510 if (!init_done) {
511 rx_strm.next_out = NULL;
512 rx_strm.zalloc = NULL;
513 rx_strm.zfree = NULL;
514 if (inflateInit2(&rx_strm, -15) != Z_OK) {
515 rprintf(FERROR, "inflate init failed\n");
516 exit_cleanup(RERR_PROTOCOL);
517 }
518 cbuf = new_array(char, MAX_DATA_COUNT);
519 dbuf = new_array(char, AVAIL_OUT_SIZE(CHUNK_SIZE));
520 init_done = 1;
521 } else {
522 inflateReset(&rx_strm);
523 }
524 recv_state = r_idle;
525 rx_token = 0;
526 break;
527
528 case r_idle:
529 case r_inflated:
530 if (saved_flag) {
531 flag = saved_flag & 0xff;
532 saved_flag = 0;
533 } else
534 flag = read_byte(f);
535 if ((flag & 0xC0) == DEFLATED_DATA) {
536 n = ((flag & 0x3f) << 8) + read_byte(f);
537 read_buf(f, cbuf, n);
538 rx_strm.next_in = (Bytef *)cbuf;
539 rx_strm.avail_in = n;
540 recv_state = r_inflating;
541 break;
542 }
543 if (recv_state == r_inflated) {
544 /* check previous inflated stuff ended correctly */
545 rx_strm.avail_in = 0;
546 rx_strm.next_out = (Bytef *)dbuf;
547 rx_strm.avail_out = AVAIL_OUT_SIZE(CHUNK_SIZE);
548 r = inflate(&rx_strm, Z_SYNC_FLUSH);
549 n = AVAIL_OUT_SIZE(CHUNK_SIZE) - rx_strm.avail_out;
550 /*
551 * Z_BUF_ERROR just means no progress was
552 * made, i.e. the decompressor didn't have
553 * any pending output for us.
554 */
555 if (r != Z_OK && r != Z_BUF_ERROR) {
556 rprintf(FERROR, "inflate flush returned %d (%d bytes)\n",
557 r, n);
558 exit_cleanup(RERR_STREAMIO);
559 }
560 if (n != 0 && r != Z_BUF_ERROR) {
561 /* have to return some more data and
562 save the flag for later. */
563 saved_flag = flag + 0x10000;
564 *data = dbuf;
565 return n;
566 }
567 /*
568 * At this point the decompressor should
569 * be expecting to see the 0, 0, ff, ff bytes.
570 */
571 if (!inflateSyncPoint(&rx_strm)) {
572 rprintf(FERROR, "decompressor lost sync!\n");
573 exit_cleanup(RERR_STREAMIO);
574 }
575 rx_strm.avail_in = 4;
576 rx_strm.next_in = (Bytef *)cbuf;
577 cbuf[0] = cbuf[1] = 0;
578 cbuf[2] = cbuf[3] = (char)0xff;
579 inflate(&rx_strm, Z_SYNC_FLUSH);
580 recv_state = r_idle;
581 }
582 if (flag == END_FLAG) {
583 /* that's all folks */
584 recv_state = r_init;
585 return 0;
586 }
587
588 /* here we have a token of some kind */
589 if (flag & TOKEN_REL) {
590 rx_token += flag & 0x3f;
591 flag >>= 6;
592 } else
593 rx_token = read_int(f);
594 if (flag & 1) {
595 rx_run = read_byte(f);
596 rx_run += read_byte(f) << 8;
597 recv_state = r_running;
598 }
599 return -1 - rx_token;
600
601 case r_inflating:
602 rx_strm.next_out = (Bytef *)dbuf;
603 rx_strm.avail_out = AVAIL_OUT_SIZE(CHUNK_SIZE);
604 r = inflate(&rx_strm, Z_NO_FLUSH);
605 n = AVAIL_OUT_SIZE(CHUNK_SIZE) - rx_strm.avail_out;
606 if (r != Z_OK) {
607 rprintf(FERROR, "inflate returned %d (%d bytes)\n", r, n);
608 exit_cleanup(RERR_STREAMIO);
609 }
610 if (rx_strm.avail_in == 0)
611 recv_state = r_inflated;
612 if (n != 0) {
613 *data = dbuf;
614 return n;
615 }
616 break;
617
618 case r_running:
619 ++rx_token;
620 if (--rx_run == 0)
621 recv_state = r_idle;
622 return -1 - rx_token;
623 }
624 }
625 }
626
627 /*
628 * put the data corresponding to a token that we've just returned
629 * from recv_deflated_token into the decompressor's history buffer.
630 */
631 static void see_deflate_token(char *buf, int32 len)
632 {
633 int r;
634 int32 blklen;
635 unsigned char hdr[5];
636
637 rx_strm.avail_in = 0;
638 blklen = 0;
639 hdr[0] = 0;
640 do {
641 if (rx_strm.avail_in == 0 && len != 0) {
642 if (blklen == 0) {
643 /* Give it a fake stored-block header. */
644 rx_strm.next_in = (Bytef *)hdr;
645 rx_strm.avail_in = 5;
646 blklen = len;
647 if (blklen > 0xffff)
648 blklen = 0xffff;
649 hdr[1] = blklen;
650 hdr[2] = blklen >> 8;
651 hdr[3] = ~hdr[1];
652 hdr[4] = ~hdr[2];
653 } else {
654 rx_strm.next_in = (Bytef *)buf;
655 rx_strm.avail_in = blklen;
656 if (protocol_version >= 31) /* Newer protocols avoid a data-duplicating bug */
657 buf += blklen;
658 len -= blklen;
659 blklen = 0;
660 }
661 }
662 rx_strm.next_out = (Bytef *)dbuf;
663 rx_strm.avail_out = AVAIL_OUT_SIZE(CHUNK_SIZE);
664 r = inflate(&rx_strm, Z_SYNC_FLUSH);
665 if (r != Z_OK && r != Z_BUF_ERROR) {
666 rprintf(FERROR, "inflate (token) returned %d\n", r);
667 exit_cleanup(RERR_STREAMIO);
668 }
669 } while (len || rx_strm.avail_out == 0);
670 }
671
672 #ifdef SUPPORT_ZSTD
673
674 static ZSTD_inBuffer zstd_in_buff;
675 static ZSTD_outBuffer zstd_out_buff;
676 static ZSTD_CCtx *zstd_cctx;
677
678 static void send_zstd_token(int f, int32 token, struct map_struct *buf, OFF_T offset, int32 nb)
679 {
680 static int comp_init_done, flush_pending;
681 ZSTD_EndDirective flush = ZSTD_e_continue;
682 int32 n, r;
683
684 /* initialization */
685 if (!comp_init_done) {
686 zstd_cctx = ZSTD_createCCtx();
687 if (!zstd_cctx) {
688 rprintf(FERROR, "compression init failed\n");
689 exit_cleanup(RERR_PROTOCOL);
690 }
691
692 obuf = new_array(char, OBUF_SIZE);
693
694 ZSTD_CCtx_setParameter(zstd_cctx, ZSTD_c_compressionLevel, do_compression_level);
695 zstd_out_buff.dst = obuf + 2;
696
697 comp_init_done = 1;
698 }
699
700 if (last_token == -1) {
701 last_run_end = 0;
702 run_start = token;
703 flush_pending = 0;
704 } else if (last_token == -2) {
705 run_start = token;
706 } else if (nb != 0 || token != last_token + 1 || token >= run_start + 65536) {
707 /* output previous run */
708 r = run_start - last_run_end;
709 n = last_token - run_start;
710
711 if (r >= 0 && r <= 63) {
712 write_byte(f, (n==0? TOKEN_REL: TOKENRUN_REL) + r);
713 } else {
714 write_byte(f, (n==0? TOKEN_LONG: TOKENRUN_LONG));
715 write_int(f, run_start);
716 }
717 if (n != 0) {
718 write_byte(f, n);
719 write_byte(f, n >> 8);
720 }
721 last_run_end = last_token;
722 run_start = token;
723 }
724
725 last_token = token;
726
727 if (nb || flush_pending) {
728
729 zstd_in_buff.src = map_ptr(buf, offset, nb);
730 zstd_in_buff.size = nb;
731 zstd_in_buff.pos = 0;
732
733 do {
734 if (zstd_out_buff.size == 0) {
735 zstd_out_buff.size = MAX_DATA_COUNT;
736 zstd_out_buff.pos = 0;
737 }
738
739 /* File ended, flush */
740 if (token != -2)
741 flush = ZSTD_e_flush;
742
743 r = ZSTD_compressStream2(zstd_cctx, &zstd_out_buff, &zstd_in_buff, flush);
744 if (ZSTD_isError(r)) {
745 rprintf(FERROR, "ZSTD_compressStream returned %d\n", r);
746 exit_cleanup(RERR_STREAMIO);
747 }
748
749 /*
750 * Nothing is sent if the buffer isn't full so avoid smaller
751 * transfers. If a file is finished then we flush the internal
752 * state and send a smaller buffer so that the remote side can
753 * finish the file.
754 */
755 if (zstd_out_buff.pos == zstd_out_buff.size || flush == ZSTD_e_flush) {
756 n = zstd_out_buff.pos;
757
758 obuf[0] = DEFLATED_DATA + (n >> 8);
759 obuf[1] = n;
760 write_buf(f, obuf, n+2);
761
762 zstd_out_buff.size = 0;
763 }
764 /*
765 * Loop while the input buffer isn't full consumed or the
766 * internal state isn't fully flushed.
767 */
768 } while (zstd_in_buff.pos < zstd_in_buff.size || r > 0);
769 flush_pending = token == -2;
770 }
771
772 if (token == -1) {
773 /* end of file - clean up */
774 write_byte(f, END_FLAG);
775 }
776 }
777
778 static ZSTD_DCtx *zstd_dctx;
779
780 static int32 recv_zstd_token(int f, char **data)
781 {
782 static int decomp_init_done;
783 static int out_buffer_size;
784 int32 n, flag;
785 int r;
786
787 if (!decomp_init_done) {
788 zstd_dctx = ZSTD_createDCtx();
789 if (!zstd_dctx) {
790 rprintf(FERROR, "ZSTD_createDStream failed\n");
791 exit_cleanup(RERR_PROTOCOL);
792 }
793
794 /* Output buffer fits two decompressed blocks */
795 out_buffer_size = ZSTD_DStreamOutSize() * 2;
796 cbuf = new_array(char, MAX_DATA_COUNT);
797 dbuf = new_array(char, out_buffer_size);
798
799 zstd_in_buff.src = cbuf;
800 zstd_out_buff.dst = dbuf;
801
802 decomp_init_done = 1;
803 }
804
805 for (;;) {
806 switch (recv_state) {
807 case r_init:
808 recv_state = r_idle;
809 rx_token = 0;
810 break;
811
812 case r_idle:
813 flag = read_byte(f);
814 if ((flag & 0xC0) == DEFLATED_DATA) {
815 n = ((flag & 0x3f) << 8) + read_byte(f);
816 read_buf(f, cbuf, n);
817
818 zstd_in_buff.size = n;
819 zstd_in_buff.pos = 0;
820
821 recv_state = r_inflating;
822 break;
823 }
824
825 if (flag == END_FLAG) {
826 /* that's all folks */
827 recv_state = r_init;
828 return 0;
829 }
830 /* here we have a token of some kind */
831 if (flag & TOKEN_REL) {
832 rx_token += flag & 0x3f;
833 flag >>= 6;
834 } else
835 rx_token = read_int(f);
836 if (flag & 1) {
837 rx_run = read_byte(f);
838 rx_run += read_byte(f) << 8;
839 recv_state = r_running;
840 }
841 return -1 - rx_token;
842
843 case r_inflated: /* zstd doesn't get into this state */
844 break;
845
846 case r_inflating:
847 zstd_out_buff.size = out_buffer_size;
848 zstd_out_buff.pos = 0;
849
850 r = ZSTD_decompressStream(zstd_dctx, &zstd_out_buff, &zstd_in_buff);
851 n = zstd_out_buff.pos;
852 if (ZSTD_isError(r)) {
853 rprintf(FERROR, "ZSTD decomp returned %d (%d bytes)\n", r, n);
854 exit_cleanup(RERR_STREAMIO);
855 }
856
857 /*
858 * If the input buffer is fully consumed and the output
859 * buffer is not full then next step is to read more
860 * data.
861 */
862 if (zstd_in_buff.size == zstd_in_buff.pos && n < out_buffer_size)
863 recv_state = r_idle;
864
865 if (n != 0) {
866 *data = dbuf;
867 return n;
868 }
869 break;
870
871 case r_running:
872 ++rx_token;
873 if (--rx_run == 0)
874 recv_state = r_idle;
875 return -1 - rx_token;
876 }
877 }
878 }
879 #endif /* SUPPORT_ZSTD */
880
881 #ifdef SUPPORT_LZ4
882 static void
883 send_compressed_token(int f, int32 token, struct map_struct *buf, OFF_T offset, int32 nb)
884 {
885 static int init_done, flush_pending;
886 int size = MAX(LZ4_compressBound(CHUNK_SIZE), MAX_DATA_COUNT+2);
887 int32 n, r;
888
889 if (last_token == -1) {
890 if (!init_done) {
891 obuf = new_array(char, size);
892 init_done = 1;
893 }
894 last_run_end = 0;
895 run_start = token;
896 flush_pending = 0;
897 } else if (last_token == -2) {
898 run_start = token;
899 } else if (nb != 0 || token != last_token + 1 || token >= run_start + 65536) {
900 /* output previous run */
901 r = run_start - last_run_end;
902 n = last_token - run_start;
903 if (r >= 0 && r <= 63) {
904 write_byte(f, (n==0? TOKEN_REL: TOKENRUN_REL) + r);
905 } else {
906 write_byte(f, (n==0? TOKEN_LONG: TOKENRUN_LONG));
907 write_int(f, run_start);
908 }
909 if (n != 0) {
910 write_byte(f, n);
911 write_byte(f, n >> 8);
912 }
913 last_run_end = last_token;
914 run_start = token;
915 }
916
917 last_token = token;
918
919 if (nb != 0 || flush_pending) {
920 int available_in, available_out = 0;
921 const char *next_in;
922
923 do {
924 char *next_out = obuf + 2;
925
926 if (available_out == 0) {
927 available_in = MIN(nb, MAX_DATA_COUNT);
928 next_in = map_ptr(buf, offset, available_in);
929 } else
930 available_in /= 2;
931
932 available_out = LZ4_compress_default(next_in, next_out, available_in, size - 2);
933 if (!available_out) {
934 rprintf(FERROR, "compress returned %d\n", available_out);
935 exit_cleanup(RERR_STREAMIO);
936 }
937 if (available_out <= MAX_DATA_COUNT) {
938 obuf[0] = DEFLATED_DATA + (available_out >> 8);
939 obuf[1] = available_out;
940
941 write_buf(f, obuf, available_out + 2);
942
943 available_out = 0;
944 nb -= available_in;
945 offset += available_in;
946 }
947 } while (nb != 0);
948 flush_pending = token == -2;
949 }
950 if (token == -1) {
951 /* end of file - clean up */
952 write_byte(f, END_FLAG);
953 }
954 }
955
956 static int32 recv_compressed_token(int f, char **data)
957 {
958 static int init_done;
959 int32 n, flag;
960 int size = MAX(LZ4_compressBound(CHUNK_SIZE), MAX_DATA_COUNT+2);
961 static const char *next_in;
962 static int avail_in;
963 int avail_out;
964
965 for (;;) {
966 switch (recv_state) {
967 case r_init:
968 if (!init_done) {
969 cbuf = new_array(char, MAX_DATA_COUNT);
970 dbuf = new_array(char, size);
971 init_done = 1;
972 }
973 recv_state = r_idle;
974 rx_token = 0;
975 break;
976
977 case r_idle:
978 flag = read_byte(f);
979 if ((flag & 0xC0) == DEFLATED_DATA) {
980 n = ((flag & 0x3f) << 8) + read_byte(f);
981 read_buf(f, cbuf, n);
982 next_in = (char *)cbuf;
983 avail_in = n;
984 recv_state = r_inflating;
985 break;
986 }
987
988 if (flag == END_FLAG) {
989 /* that's all folks */
990 recv_state = r_init;
991 return 0;
992 }
993
994 /* here we have a token of some kind */
995 if (flag & TOKEN_REL) {
996 rx_token += flag & 0x3f;
997 flag >>= 6;
998 } else
999 rx_token = read_int(f);
1000 if (flag & 1) {
1001 rx_run = read_byte(f);
1002 rx_run += read_byte(f) << 8;
1003 recv_state = r_running;
1004 }
1005 return -1 - rx_token;
1006
1007 case r_inflating:
1008 avail_out = LZ4_decompress_safe(next_in, dbuf, avail_in, size);
1009 if (avail_out < 0) {
1010 rprintf(FERROR, "uncompress failed: %d\n", avail_out);
1011 exit_cleanup(RERR_STREAMIO);
1012 }
1013 recv_state = r_idle;
1014 *data = dbuf;
1015 return avail_out;
1016
1017 case r_inflated: /* lz4 doesn't get into this state */
1018 break;
1019
1020 case r_running:
1021 ++rx_token;
1022 if (--rx_run == 0)
1023 recv_state = r_idle;
1024 return -1 - rx_token;
1025 }
1026 }
1027 }
1028 #endif /* SUPPORT_LZ4 */
1029
1030 /**
1031 * Transmit a verbatim buffer of length @p n followed by a token.
1032 * If token == -1 then we have reached EOF
1033 * If n == 0 then don't send a buffer
1034 */
1035 void send_token(int f, int32 token, struct map_struct *buf, OFF_T offset,
1036 int32 n, int32 toklen)
1037 {
1038 switch (do_compression) {
1039 case CPRES_NONE:
1040 simple_send_token(f, token, buf, offset, n);
1041 break;
1042 case CPRES_ZLIB:
1043 case CPRES_ZLIBX:
1044 send_deflated_token(f, token, buf, offset, n, toklen);
1045 break;
1046 #ifdef SUPPORT_ZSTD
1047 case CPRES_ZSTD:
1048 send_zstd_token(f, token, buf, offset, n);
1049 break;
1050 #endif
1051 #ifdef SUPPORT_LZ4
1052 case CPRES_LZ4:
1053 send_compressed_token(f, token, buf, offset, n);
1054 break;
1055 #endif
1056 default:
1057 NOISY_DEATH("Unknown do_compression value");
1058 }
1059 }
1060
1061 /*
1062 * receive a token or buffer from the other end. If the return value is >0 then
1063 * it is a data buffer of that length, and *data will point at the data.
1064 * if the return value is -i then it represents token i-1
1065 * if the return value is 0 then the end has been reached
1066 */
1067 int32 recv_token(int f, char **data)
1068 {
1069 switch (do_compression) {
1070 case CPRES_NONE:
1071 return simple_recv_token(f,data);
1072 case CPRES_ZLIB:
1073 case CPRES_ZLIBX:
1074 return recv_deflated_token(f, data);
1075 #ifdef SUPPORT_ZSTD
1076 case CPRES_ZSTD:
1077 return recv_zstd_token(f, data);
1078 #endif
1079 #ifdef SUPPORT_LZ4
1080 case CPRES_LZ4:
1081 return recv_compressed_token(f, data);
1082 #endif
1083 default:
1084 NOISY_DEATH("Unknown do_compression value");
1085 }
1086 }
1087
1088 /*
1089 * look at the data corresponding to a token, if necessary
1090 */
1091 void see_token(char *data, int32 toklen)
1092 {
1093 switch (do_compression) {
1094 case CPRES_NONE:
1095 break;
1096 case CPRES_ZLIB:
1097 see_deflate_token(data, toklen);
1098 break;
1099 case CPRES_ZLIBX:
1100 break;
1101 #ifdef SUPPORT_ZSTD
1102 case CPRES_ZSTD:
1103 break;
1104 #endif
1105 #ifdef SUPPORT_LZ4
1106 case CPRES_LZ4:
1107 /*see_uncompressed_token(data, toklen);*/
1108 break;
1109 #endif
1110 default:
1111 NOISY_DEATH("Unknown do_compression value");
1112 }
1113 }
rsync: a fast, versatile, remote (and local) file-copying tool