- Use an int32 for the each block-size variable.
[rsync/rsync.git] / generator.c
1 /* -*- c-file-style: "linux" -*-
2
3    rsync -- fast file replication program
4
5    Copyright (C) 1996-2000 by Andrew Tridgell
6    Copyright (C) Paul Mackerras 1996
7    Copyright (C) 2002 by Martin Pool <mbp@samba.org>
8
9    This program is free software; you can redistribute it and/or modify
10    it under the terms of the GNU General Public License as published by
11    the Free Software Foundation; either version 2 of the License, or
12    (at your option) any later version.
13
14    This program is distributed in the hope that it will be useful,
15    but WITHOUT ANY WARRANTY; without even the implied warranty of
16    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17    GNU General Public License for more details.
18
19    You should have received a copy of the GNU General Public License
20    along with this program; if not, write to the Free Software
21    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 */
23
24 #include "rsync.h"
25
26 extern int verbose;
27 extern int dry_run;
28 extern int relative_paths;
29 extern int keep_dirlinks;
30 extern int preserve_links;
31 extern int am_root;
32 extern int preserve_devices;
33 extern int preserve_hard_links;
34 extern int preserve_perms;
35 extern int preserve_uid;
36 extern int preserve_gid;
37 extern int update_only;
38 extern int opt_ignore_existing;
39 extern int inplace;
40 extern int make_backups;
41 extern int csum_length;
42 extern int ignore_times;
43 extern int size_only;
44 extern OFF_T max_size;
45 extern int io_timeout;
46 extern int protocol_version;
47 extern int always_checksum;
48 extern char *partial_dir;
49 extern char *basis_dir[];
50 extern int copy_dest;
51 extern int link_dest;
52 extern int whole_file;
53 extern int local_server;
54 extern int list_only;
55 extern int read_batch;
56 extern int only_existing;
57 extern int orig_umask;
58 extern int safe_symlinks;
59 extern unsigned int block_size;
60
61 extern struct exclude_list_struct server_exclude_list;
62
63 static int unchanged_attrs(struct file_struct *file, STRUCT_STAT *st)
64 {
65         if (preserve_perms
66          && (st->st_mode & CHMOD_BITS) != (file->mode & CHMOD_BITS))
67                 return 0;
68
69         if (am_root && preserve_uid && st->st_uid != file->uid)
70                 return 0;
71
72         if (preserve_gid && file->gid != GID_NONE && st->st_gid != file->gid)
73                 return 0;
74
75         return 1;
76 }
77
78 /* Perform our quick-check heuristic for determining if a file is unchanged. */
79 static int unchanged_file(char *fn, struct file_struct *file, STRUCT_STAT *st)
80 {
81         if (st->st_size != file->length)
82                 return 0;
83
84         /* if always checksum is set then we use the checksum instead
85            of the file time to determine whether to sync */
86         if (always_checksum && S_ISREG(st->st_mode)) {
87                 char sum[MD4_SUM_LENGTH];
88                 file_checksum(fn, sum, st->st_size);
89                 return memcmp(sum, file->u.sum, protocol_version < 21 ? 2
90                                                         : MD4_SUM_LENGTH) == 0;
91         }
92
93         if (size_only)
94                 return 1;
95
96         if (ignore_times)
97                 return 0;
98
99         return cmp_modtime(st->st_mtime, file->modtime) == 0;
100 }
101
102
103 /*
104  * NULL sum_struct means we have no checksums
105  */
106 void write_sum_head(int f, struct sum_struct *sum)
107 {
108         static struct sum_struct null_sum;
109
110         if (sum == NULL)
111                 sum = &null_sum;
112
113         write_int(f, sum->count);
114         write_int(f, sum->blength);
115         if (protocol_version >= 27)
116                 write_int(f, sum->s2length);
117         write_int(f, sum->remainder);
118 }
119
120 /*
121  * set (initialize) the size entries in the per-file sum_struct
122  * calculating dynamic block and checksum sizes.
123  *
124  * This is only called from generate_and_send_sums() but is a separate
125  * function to encapsulate the logic.
126  *
127  * The block size is a rounded square root of file length.
128  *
129  * The checksum size is determined according to:
130  *     blocksum_bits = BLOCKSUM_EXP + 2*log2(file_len) - log2(block_len)
131  * provided by Donovan Baarda which gives a probability of rsync
132  * algorithm corrupting data and falling back using the whole md4
133  * checksums.
134  *
135  * This might be made one of several selectable heuristics.
136  */
137 static void sum_sizes_sqroot(struct sum_struct *sum, uint64 len)
138 {
139         unsigned int blength;
140         int s2length;
141         uint32 c;
142         uint64 l;
143
144         if (block_size) {
145                 blength = block_size;
146         } else if (len <= BLOCK_SIZE * BLOCK_SIZE) {
147                 blength = BLOCK_SIZE;
148         } else {
149                 l = len;
150                 c = 1;
151                 while (l >>= 2) {
152                         c <<= 1;
153                 }
154                 blength = 0;
155                 do {
156                         blength |= c;
157                         if (len < (uint64)blength * blength)
158                                 blength &= ~c;
159                         c >>= 1;
160                 } while (c >= 8);       /* round to multiple of 8 */
161                 blength = MAX(blength, BLOCK_SIZE);
162         }
163
164         if (protocol_version < 27) {
165                 s2length = csum_length;
166         } else if (csum_length == SUM_LENGTH) {
167                 s2length = SUM_LENGTH;
168         } else {
169                 int b = BLOCKSUM_BIAS;
170                 l = len;
171                 while (l >>= 1) {
172                         b += 2;
173                 }
174                 c = blength;
175                 while (c >>= 1 && b) {
176                         b--;
177                 }
178                 s2length = (b + 1 - 32 + 7) / 8; /* add a bit,
179                                                   * subtract rollsum,
180                                                   * round up
181                                                   *    --optimize in compiler--
182                                                   */
183                 s2length = MAX(s2length, csum_length);
184                 s2length = MIN(s2length, SUM_LENGTH);
185         }
186
187         sum->flength    = len;
188         sum->blength    = blength;
189         sum->s2length   = s2length;
190         sum->count      = (len + (blength - 1)) / blength;
191         sum->remainder  = (len % blength);
192
193         if (sum->count && verbose > 2) {
194                 rprintf(FINFO, "count=%.0f rem=%u blength=%u s2length=%d flength=%.0f\n",
195                         (double)sum->count, sum->remainder, sum->blength,
196                         sum->s2length, (double)sum->flength);
197         }
198 }
199
200
201 /*
202  * Generate and send a stream of signatures/checksums that describe a buffer
203  *
204  * Generate approximately one checksum every block_len bytes.
205  */
206 static void generate_and_send_sums(int fd, OFF_T len, int f_out, int f_copy)
207 {
208         size_t i;
209         struct map_struct *mapbuf;
210         struct sum_struct sum;
211         OFF_T offset = 0;
212
213         sum_sizes_sqroot(&sum, len);
214
215         if (len > 0)
216                 mapbuf = map_file(fd, len, MAX_MAP_SIZE, sum.blength);
217         else
218                 mapbuf = NULL;
219
220         write_sum_head(f_out, &sum);
221
222         for (i = 0; i < sum.count; i++) {
223                 unsigned int n1 = MIN(len, sum.blength);
224                 char *map = map_ptr(mapbuf, offset, n1);
225                 uint32 sum1 = get_checksum1(map, n1);
226                 char sum2[SUM_LENGTH];
227
228                 if (f_copy >= 0)
229                         full_write(f_copy, map, n1);
230
231                 get_checksum2(map, n1, sum2);
232
233                 if (verbose > 3) {
234                         rprintf(FINFO,
235                                 "chunk[%.0f] offset=%.0f len=%u sum1=%08lx\n",
236                                 (double)i, (double)offset, n1,
237                                 (unsigned long)sum1);
238                 }
239                 write_int(f_out, sum1);
240                 write_buf(f_out, sum2, sum.s2length);
241                 len -= n1;
242                 offset += n1;
243         }
244
245         if (mapbuf)
246                 unmap_file(mapbuf);
247 }
248
249
250
251 /*
252  * Acts on file number @p i from @p flist, whose name is @p fname.
253  *
254  * First fixes up permissions, then generates checksums for the file.
255  *
256  * @note This comment was added later by mbp who was trying to work it
257  * out.  It might be wrong.
258  */
259 static void recv_generator(char *fname, struct file_struct *file, int i,
260                            int f_out, int f_out_name)
261 {
262         int fd = -1, f_copy = -1;
263         STRUCT_STAT st, partial_st;
264         struct file_struct *back_file = NULL;
265         int statret, stat_errno;
266         char *fnamecmp, *partialptr, *backupptr = NULL;
267         char fnamecmpbuf[MAXPATHLEN];
268         uchar fnamecmp_type;
269
270         if (list_only)
271                 return;
272
273         if (verbose > 2)
274                 rprintf(FINFO, "recv_generator(%s,%d)\n", safe_fname(fname), i);
275
276         if (server_exclude_list.head
277             && check_exclude(&server_exclude_list, fname,
278                              S_ISDIR(file->mode)) < 0) {
279                 if (verbose) {
280                         rprintf(FINFO, "skipping server-excluded file \"%s\"\n",
281                                 safe_fname(fname));
282                 }
283                 return;
284         }
285
286         if (dry_run > 1) {
287                 statret = -1;
288                 stat_errno = ENOENT;
289         } else {
290                 statret = link_stat(fname, &st,
291                                     keep_dirlinks && S_ISDIR(file->mode));
292                 stat_errno = errno;
293         }
294
295         if (only_existing && statret == -1 && stat_errno == ENOENT) {
296                 /* we only want to update existing files */
297                 if (verbose > 1) {
298                         rprintf(FINFO, "not creating new file \"%s\"\n",
299                                 safe_fname(fname));
300                 }
301                 return;
302         }
303
304         if (statret == 0 && !preserve_perms
305             && S_ISDIR(st.st_mode) == S_ISDIR(file->mode)) {
306                 /* if the file exists already and we aren't perserving
307                  * permissions then act as though the remote end sent
308                  * us the file permissions we already have */
309                 file->mode = (file->mode & ~CHMOD_BITS)
310                            | (st.st_mode & CHMOD_BITS);
311         }
312
313         if (S_ISDIR(file->mode)) {
314                 /* The file to be received is a directory, so we need
315                  * to prepare appropriately.  If there is already a
316                  * file of that name and it is *not* a directory, then
317                  * we need to delete it.  If it doesn't exist, then
318                  * recursively create it. */
319
320                 if (dry_run)
321                         return; /* TODO: causes inaccuracies -- fix */
322                 if (statret == 0 && !S_ISDIR(st.st_mode)) {
323                         if (robust_unlink(fname) != 0) {
324                                 rsyserr(FERROR, errno,
325                                         "recv_generator: unlink %s to make room for directory",
326                                         full_fname(fname));
327                                 return;
328                         }
329                         statret = -1;
330                 }
331                 if (statret != 0 && do_mkdir(fname,file->mode) != 0 && errno != EEXIST) {
332                         if (!(relative_paths && errno == ENOENT
333                             && create_directory_path(fname, orig_umask) == 0
334                             && do_mkdir(fname, file->mode) == 0)) {
335                                 rsyserr(FERROR, errno,
336                                         "recv_generator: mkdir %s failed",
337                                         full_fname(fname));
338                         }
339                 }
340                 /* f_out is set to -1 when doing final directory-permission
341                  * and modification-time repair. */
342                 if (set_perms(fname, file, statret ? NULL : &st, 0)
343                     && verbose && f_out != -1)
344                         rprintf(FINFO, "%s/\n", safe_fname(fname));
345                 return;
346         } else if (max_size && file->length > max_size) {
347                 if (verbose > 1)
348                         rprintf(FINFO, "%s is over max-size\n", fname);
349                 return;
350         }
351
352         if (preserve_links && S_ISLNK(file->mode)) {
353 #if SUPPORT_LINKS
354                 char lnk[MAXPATHLEN];
355                 int l;
356
357                 if (safe_symlinks && unsafe_symlink(file->u.link, fname)) {
358                         if (verbose) {
359                                 rprintf(FINFO, "ignoring unsafe symlink %s -> \"%s\"\n",
360                                         full_fname(fname), file->u.link);
361                         }
362                         return;
363                 }
364                 if (statret == 0) {
365                         l = readlink(fname,lnk,MAXPATHLEN-1);
366                         if (l > 0) {
367                                 lnk[l] = 0;
368                                 /* A link already pointing to the
369                                  * right place -- no further action
370                                  * required. */
371                                 if (strcmp(lnk,file->u.link) == 0) {
372                                         set_perms(fname, file, &st,
373                                                   PERMS_REPORT);
374                                         return;
375                                 }
376                         }
377                         /* Not a symlink, so delete whatever's
378                          * already there and put a new symlink
379                          * in place. */
380                         delete_file(fname);
381                 }
382                 if (do_symlink(file->u.link,fname) != 0) {
383                         rsyserr(FERROR, errno, "symlink %s -> \"%s\" failed",
384                                 full_fname(fname), safe_fname(file->u.link));
385                 } else {
386                         set_perms(fname,file,NULL,0);
387                         if (verbose) {
388                                 rprintf(FINFO, "%s -> %s\n", safe_fname(fname),
389                                         safe_fname(file->u.link));
390                         }
391                 }
392 #endif
393                 return;
394         }
395
396         if (am_root && preserve_devices && IS_DEVICE(file->mode)) {
397                 if (statret != 0 ||
398                     st.st_mode != file->mode ||
399                     st.st_rdev != file->u.rdev) {
400                         delete_file(fname);
401                         if (verbose > 2) {
402                                 rprintf(FINFO,"mknod(%s,0%o,0x%x)\n",
403                                         safe_fname(fname),
404                                         (int)file->mode, (int)file->u.rdev);
405                         }
406                         if (do_mknod(fname,file->mode,file->u.rdev) != 0) {
407                                 rsyserr(FERROR, errno, "mknod %s failed",
408                                         full_fname(fname));
409                         } else {
410                                 set_perms(fname,file,NULL,0);
411                                 if (verbose) {
412                                         rprintf(FINFO, "%s\n",
413                                                 safe_fname(fname));
414                                 }
415                         }
416                 } else {
417                         set_perms(fname, file, &st, PERMS_REPORT);
418                 }
419                 return;
420         }
421
422         if (preserve_hard_links && hard_link_check(file, HL_CHECK_MASTER))
423                 return;
424
425         if (!S_ISREG(file->mode)) {
426                 rprintf(FINFO, "skipping non-regular file \"%s\"\n",
427                         safe_fname(fname));
428                 return;
429         }
430
431         fnamecmp = fname;
432         fnamecmp_type = FNAMECMP_FNAME;
433
434         if (statret == -1 && basis_dir[0] != NULL) {
435                 int fallback_match = -1;
436                 int match_level = 0;
437                 int i = 0;
438                 do {
439                         pathjoin(fnamecmpbuf, sizeof fnamecmpbuf,
440                                  basis_dir[i], fname);
441                         if (link_stat(fnamecmpbuf, &st, 0) == 0
442                             && S_ISREG(st.st_mode)) {
443                                 statret = 0;
444                                 if (link_dest) {
445                                         if (!match_level) {
446                                                 fallback_match = i;
447                                                 match_level = 1;
448                                         } else if (match_level == 2
449                                             && !unchanged_attrs(file, &st))
450                                                 continue;
451                                         if (!unchanged_file(fnamecmpbuf, file, &st))
452                                                 continue;
453                                         fallback_match = i;
454                                         match_level = 2;
455                                         if (!unchanged_attrs(file, &st))
456                                                 continue;
457                                 }
458                                 match_level = 3;
459                                 break;
460                         }
461                 } while (basis_dir[++i] != NULL);
462                 if (statret == 0) {
463                         if (match_level < 3) {
464                                 i = fallback_match;
465                                 pathjoin(fnamecmpbuf, sizeof fnamecmpbuf,
466                                          basis_dir[i], fname);
467                         }
468 #if HAVE_LINK
469                         if (link_dest && match_level == 3 && !dry_run) {
470                                 if (do_link(fnamecmpbuf, fname) < 0) {
471                                         if (verbose) {
472                                                 rsyserr(FINFO, errno,
473                                                         "link %s => %s",
474                                                         fnamecmpbuf,
475                                                         safe_fname(fname));
476                                         }
477                                         fnamecmp = fnamecmpbuf;
478                                         fnamecmp_type = FNAMECMP_BASIS_DIR + i;
479                                 }
480                         } else
481 #endif
482                         {
483                                 fnamecmp = fnamecmpbuf;
484                                 fnamecmp_type = FNAMECMP_BASIS_DIR + i;
485                         }
486                 }
487         }
488
489         if (statret == 0 && !S_ISREG(st.st_mode)) {
490                 if (delete_file(fname) != 0)
491                         return;
492                 statret = -1;
493                 stat_errno = ENOENT;
494         }
495
496         if (partial_dir && (partialptr = partial_dir_fname(fname))
497             && link_stat(partialptr, &partial_st, 0) == 0
498             && S_ISREG(partial_st.st_mode)) {
499                 if (statret == -1)
500                         goto prepare_to_open;
501         } else
502                 partialptr = NULL;
503
504         if (statret == -1) {
505                 if (preserve_hard_links && hard_link_check(file, HL_SKIP))
506                         return;
507                 if (stat_errno == ENOENT)
508                         goto notify_others;
509                 if (verbose > 1) {
510                         rsyserr(FERROR, stat_errno,
511                                 "recv_generator: failed to stat %s",
512                                 full_fname(fname));
513                 }
514                 return;
515         }
516
517         if (opt_ignore_existing && fnamecmp_type == FNAMECMP_FNAME) {
518                 if (verbose > 1)
519                         rprintf(FINFO, "%s exists\n", safe_fname(fname));
520                 return;
521         }
522
523         if (update_only && fnamecmp_type == FNAMECMP_FNAME
524             && cmp_modtime(st.st_mtime, file->modtime) > 0) {
525                 if (verbose > 1)
526                         rprintf(FINFO, "%s is newer\n", safe_fname(fname));
527                 return;
528         }
529
530         if ((link_dest || copy_dest) && fnamecmp_type != FNAMECMP_FNAME)
531                 ;
532         else if (unchanged_file(fnamecmp, file, &st)) {
533                 if (fnamecmp_type == FNAMECMP_FNAME)
534                         set_perms(fname, file, &st, PERMS_REPORT);
535                 return;
536         }
537
538 prepare_to_open:
539         if (dry_run || whole_file > 0) {
540                 statret = -1;
541                 goto notify_others;
542         }
543         if (read_batch)
544                 goto notify_others;
545
546         if (partialptr) {
547                 st = partial_st;
548                 fnamecmp = partialptr;
549                 fnamecmp_type = FNAMECMP_PARTIAL_DIR;
550         }
551
552         /* open the file */
553         fd = do_open(fnamecmp, O_RDONLY, 0);
554
555         if (fd == -1) {
556                 rsyserr(FERROR, errno, "failed to open %s, continuing",
557                         full_fname(fnamecmp));
558             pretend_missing:
559                 /* pretend the file didn't exist */
560                 if (preserve_hard_links && hard_link_check(file, HL_SKIP))
561                         return;
562                 statret = -1;
563                 goto notify_others;
564         }
565
566         if (inplace && make_backups) {
567                 if (!(backupptr = get_backup_name(fname))) {
568                         close(fd);
569                         return;
570                 }
571                 if (!(back_file = make_file(fname, NULL, NO_EXCLUDES))) {
572                         close(fd);
573                         goto pretend_missing;
574                 }
575                 if (robust_unlink(backupptr) && errno != ENOENT) {
576                         rsyserr(FERROR, errno, "unlink %s",
577                                 full_fname(backupptr));
578                         free(back_file);
579                         close(fd);
580                         return;
581                 }
582                 if ((f_copy = do_open(backupptr,
583                     O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, 0600)) < 0) {
584                         rsyserr(FERROR, errno, "open %s",
585                                 full_fname(backupptr));
586                         free(back_file);
587                         close(fd);
588                         return;
589                 }
590                 fnamecmp_type = FNAMECMP_BACKUP;
591         }
592
593         if (verbose > 3) {
594                 rprintf(FINFO, "gen mapped %s of size %.0f\n",
595                         safe_fname(fnamecmp), (double)st.st_size);
596         }
597
598         if (verbose > 2)
599                 rprintf(FINFO, "generating and sending sums for %d\n", i);
600
601 notify_others:
602         write_int(f_out, i);
603         if (f_out_name >= 0)
604                 write_byte(f_out_name, fnamecmp_type);
605
606         if (dry_run || read_batch)
607                 return;
608
609         if (statret == 0) {
610                 generate_and_send_sums(fd, st.st_size, f_out, f_copy);
611
612                 if (f_copy >= 0) {
613                         close(f_copy);
614                         set_perms(backupptr, back_file, NULL, 0);
615                         if (verbose > 1) {
616                                 rprintf(FINFO, "backed up %s to %s\n",
617                                         fname, backupptr);
618                         }
619                         free(back_file);
620                 }
621
622                 close(fd);
623         } else
624                 write_sum_head(f_out, NULL);
625 }
626
627
628 void generate_files(int f_out, struct file_list *flist, char *local_name,
629                     int f_out_name)
630 {
631         int i;
632         int phase = 0;
633         char fbuf[MAXPATHLEN];
634
635         if (verbose > 2) {
636                 rprintf(FINFO, "generator starting pid=%ld count=%d\n",
637                         (long)getpid(), flist->count);
638         }
639
640         if (verbose >= 2) {
641                 rprintf(FINFO,
642                         whole_file > 0
643                         ? "delta-transmission disabled for local transfer or --whole-file\n"
644                         : "delta transmission enabled\n");
645         }
646
647         /* we expect to just sit around now, so don't exit on a
648            timeout. If we really get a timeout then the other process should
649            exit */
650         io_timeout = 0;
651
652         for (i = 0; i < flist->count; i++) {
653                 struct file_struct *file = flist->files[i];
654                 struct file_struct copy;
655
656                 if (!file->basename)
657                         continue;
658                 /* we need to ensure that any directories we create have writeable
659                    permissions initially so that we can create the files within
660                    them. This is then fixed after the files are transferred */
661                 if (!am_root && S_ISDIR(file->mode) && !(file->mode & S_IWUSR)) {
662                         copy = *file;
663                         /* XXX: Could this be causing a problem on SCO?  Perhaps their
664                          * handling of permissions is strange? */
665                         copy.mode |= S_IWUSR; /* user write */
666                         file = &copy;
667                 }
668
669                 recv_generator(local_name ? local_name : f_name_to(file, fbuf),
670                                file, i, f_out, f_out_name);
671         }
672
673         phase++;
674         csum_length = SUM_LENGTH;
675         ignore_times = 1;
676
677         if (verbose > 2)
678                 rprintf(FINFO,"generate_files phase=%d\n",phase);
679
680         write_int(f_out, -1);
681
682         /* files can cycle through the system more than once
683          * to catch initial checksum errors */
684         while ((i = get_redo_num()) != -1) {
685                 struct file_struct *file = flist->files[i];
686                 recv_generator(local_name ? local_name : f_name_to(file, fbuf),
687                                file, i, f_out, f_out_name);
688         }
689
690         phase++;
691         if (verbose > 2)
692                 rprintf(FINFO,"generate_files phase=%d\n",phase);
693
694         write_int(f_out, -1);
695
696         if (preserve_hard_links)
697                 do_hard_links();
698
699         /* now we need to fix any directory permissions that were
700          * modified during the transfer */
701         for (i = 0; i < flist->count; i++) {
702                 struct file_struct *file = flist->files[i];
703                 if (!file->basename || !S_ISDIR(file->mode))
704                         continue;
705                 recv_generator(local_name ? local_name : f_name(file),
706                                file, i, -1, -1);
707         }
708
709         if (verbose > 2)
710                 rprintf(FINFO,"generate_files finished\n");
711 }