Got rid of delete_one() since we can (and do) use the updated
[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 compare_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 long block_size; /* "long" because popt can't set an int32. */
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         int32 blength;
140         int s2length;
141
142         if (block_size)
143                 blength = block_size;
144         else if (len <= BLOCK_SIZE * BLOCK_SIZE)
145                 blength = BLOCK_SIZE;
146         else {
147                 int32 c;
148                 uint64 l;
149                 int cnt;
150                 for (c = 1, l = len, cnt = 0; l >>= 2; c <<= 1, cnt++) {}
151                 if (cnt >= 31 || c >= MAX_BLOCK_SIZE)
152                         blength = MAX_BLOCK_SIZE;
153                 else {
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
165         if (protocol_version < 27) {
166                 s2length = csum_length;
167         } else if (csum_length == SUM_LENGTH) {
168                 s2length = SUM_LENGTH;
169         } else {
170                 int32 c;
171                 uint64 l;
172                 int b = BLOCKSUM_BIAS;
173                 for (l = len; l >>= 1; b += 2) {}
174                 for (c = blength; c >>= 1 && b; b--) {}
175                 /* add a bit, subtract rollsum, round up. */
176                 s2length = (b + 1 - 32 + 7) / 8; /* --optimize in compiler-- */
177                 s2length = MAX(s2length, csum_length);
178                 s2length = MIN(s2length, SUM_LENGTH);
179         }
180
181         sum->flength    = len;
182         sum->blength    = blength;
183         sum->s2length   = s2length;
184         sum->count      = (len + (blength - 1)) / blength;
185         sum->remainder  = (len % blength);
186
187         if (sum->count && verbose > 2) {
188                 rprintf(FINFO,
189                         "count=%.0f rem=%ld blength=%ld s2length=%d flength=%.0f\n",
190                         (double)sum->count, (long)sum->remainder, (long)sum->blength,
191                         sum->s2length, (double)sum->flength);
192         }
193 }
194
195
196 /*
197  * Generate and send a stream of signatures/checksums that describe a buffer
198  *
199  * Generate approximately one checksum every block_len bytes.
200  */
201 static void generate_and_send_sums(int fd, OFF_T len, int f_out, int f_copy)
202 {
203         size_t i;
204         struct map_struct *mapbuf;
205         struct sum_struct sum;
206         OFF_T offset = 0;
207
208         sum_sizes_sqroot(&sum, len);
209
210         if (len > 0)
211                 mapbuf = map_file(fd, len, MAX_MAP_SIZE, sum.blength);
212         else
213                 mapbuf = NULL;
214
215         write_sum_head(f_out, &sum);
216
217         for (i = 0; i < sum.count; i++) {
218                 int32 n1 = (int32)MIN(len, (OFF_T)sum.blength);
219                 char *map = map_ptr(mapbuf, offset, n1);
220                 uint32 sum1 = get_checksum1(map, n1);
221                 char sum2[SUM_LENGTH];
222
223                 if (f_copy >= 0)
224                         full_write(f_copy, map, n1);
225
226                 get_checksum2(map, n1, sum2);
227
228                 if (verbose > 3) {
229                         rprintf(FINFO,
230                                 "chunk[%.0f] offset=%.0f len=%ld sum1=%08lx\n",
231                                 (double)i, (double)offset, (long)n1,
232                                 (unsigned long)sum1);
233                 }
234                 write_int(f_out, sum1);
235                 write_buf(f_out, sum2, sum.s2length);
236                 len -= n1;
237                 offset += n1;
238         }
239
240         if (mapbuf)
241                 unmap_file(mapbuf);
242 }
243
244
245
246 /*
247  * Acts on file number @p i from @p flist, whose name is @p fname.
248  *
249  * First fixes up permissions, then generates checksums for the file.
250  *
251  * @note This comment was added later by mbp who was trying to work it
252  * out.  It might be wrong.
253  */
254 static void recv_generator(char *fname, struct file_struct *file, int i,
255                            int f_out, int f_out_name)
256 {
257         int fd = -1, f_copy = -1;
258         STRUCT_STAT st, partial_st;
259         struct file_struct *back_file = NULL;
260         int statret, stat_errno;
261         char *fnamecmp, *partialptr, *backupptr = NULL;
262         char fnamecmpbuf[MAXPATHLEN];
263         uchar fnamecmp_type;
264
265         if (list_only)
266                 return;
267
268         if (verbose > 2)
269                 rprintf(FINFO, "recv_generator(%s,%d)\n", safe_fname(fname), i);
270
271         if (server_exclude_list.head
272             && check_exclude(&server_exclude_list, fname,
273                              S_ISDIR(file->mode)) < 0) {
274                 if (verbose) {
275                         rprintf(FINFO, "skipping server-excluded file \"%s\"\n",
276                                 safe_fname(fname));
277                 }
278                 return;
279         }
280
281         if (dry_run > 1) {
282                 statret = -1;
283                 stat_errno = ENOENT;
284         } else {
285                 statret = link_stat(fname, &st,
286                                     keep_dirlinks && S_ISDIR(file->mode));
287                 stat_errno = errno;
288         }
289
290         if (only_existing && statret == -1 && stat_errno == ENOENT) {
291                 /* we only want to update existing files */
292                 if (verbose > 1) {
293                         rprintf(FINFO, "not creating new file \"%s\"\n",
294                                 safe_fname(fname));
295                 }
296                 return;
297         }
298
299         if (statret == 0 && !preserve_perms
300             && S_ISDIR(st.st_mode) == S_ISDIR(file->mode)) {
301                 /* if the file exists already and we aren't perserving
302                  * permissions then act as though the remote end sent
303                  * us the file permissions we already have */
304                 file->mode = (file->mode & ~CHMOD_BITS)
305                            | (st.st_mode & CHMOD_BITS);
306         }
307
308         if (S_ISDIR(file->mode)) {
309                 /* The file to be received is a directory, so we need
310                  * to prepare appropriately.  If there is already a
311                  * file of that name and it is *not* a directory, then
312                  * we need to delete it.  If it doesn't exist, then
313                  * recursively create it. */
314
315                 if (dry_run)
316                         return; /* TODO: causes inaccuracies -- fix */
317                 if (statret == 0 && !S_ISDIR(st.st_mode)) {
318                         if (robust_unlink(fname) != 0) {
319                                 rsyserr(FERROR, errno,
320                                         "recv_generator: unlink %s to make room for directory",
321                                         full_fname(fname));
322                                 return;
323                         }
324                         statret = -1;
325                 }
326                 if (statret != 0 && do_mkdir(fname,file->mode) != 0 && errno != EEXIST) {
327                         if (!(relative_paths && errno == ENOENT
328                             && create_directory_path(fname, orig_umask) == 0
329                             && do_mkdir(fname, file->mode) == 0)) {
330                                 rsyserr(FERROR, errno,
331                                         "recv_generator: mkdir %s failed",
332                                         full_fname(fname));
333                         }
334                 }
335                 /* f_out is set to -1 when doing final directory-permission
336                  * and modification-time repair. */
337                 if (set_perms(fname, file, statret ? NULL : &st, 0)
338                     && verbose && f_out != -1)
339                         rprintf(FINFO, "%s/\n", safe_fname(fname));
340                 return;
341         } else if (max_size && file->length > max_size) {
342                 if (verbose > 1)
343                         rprintf(FINFO, "%s is over max-size\n", fname);
344                 return;
345         }
346
347         if (preserve_links && S_ISLNK(file->mode)) {
348 #if SUPPORT_LINKS
349                 char lnk[MAXPATHLEN];
350                 int l;
351
352                 if (safe_symlinks && unsafe_symlink(file->u.link, fname)) {
353                         if (verbose) {
354                                 rprintf(FINFO, "ignoring unsafe symlink %s -> \"%s\"\n",
355                                         full_fname(fname), file->u.link);
356                         }
357                         return;
358                 }
359                 if (statret == 0) {
360                         l = readlink(fname,lnk,MAXPATHLEN-1);
361                         if (l > 0) {
362                                 lnk[l] = 0;
363                                 /* A link already pointing to the
364                                  * right place -- no further action
365                                  * required. */
366                                 if (strcmp(lnk,file->u.link) == 0) {
367                                         set_perms(fname, file, &st,
368                                                   PERMS_REPORT);
369                                         return;
370                                 }
371                         }
372                         /* Not a symlink, so delete whatever's
373                          * already there and put a new symlink
374                          * in place. */
375                         delete_file(fname);
376                 }
377                 if (do_symlink(file->u.link,fname) != 0) {
378                         rsyserr(FERROR, errno, "symlink %s -> \"%s\" failed",
379                                 full_fname(fname), safe_fname(file->u.link));
380                 } else {
381                         set_perms(fname,file,NULL,0);
382                         if (verbose) {
383                                 rprintf(FINFO, "%s -> %s\n", safe_fname(fname),
384                                         safe_fname(file->u.link));
385                         }
386                 }
387 #endif
388                 return;
389         }
390
391         if (am_root && preserve_devices && IS_DEVICE(file->mode)) {
392                 if (statret != 0 ||
393                     st.st_mode != file->mode ||
394                     st.st_rdev != file->u.rdev) {
395                         delete_file(fname);
396                         if (verbose > 2) {
397                                 rprintf(FINFO,"mknod(%s,0%o,0x%x)\n",
398                                         safe_fname(fname),
399                                         (int)file->mode, (int)file->u.rdev);
400                         }
401                         if (do_mknod(fname,file->mode,file->u.rdev) != 0) {
402                                 rsyserr(FERROR, errno, "mknod %s failed",
403                                         full_fname(fname));
404                         } else {
405                                 set_perms(fname,file,NULL,0);
406                                 if (verbose) {
407                                         rprintf(FINFO, "%s\n",
408                                                 safe_fname(fname));
409                                 }
410                         }
411                 } else {
412                         set_perms(fname, file, &st, PERMS_REPORT);
413                 }
414                 return;
415         }
416
417         if (preserve_hard_links && hard_link_check(file, HL_CHECK_MASTER))
418                 return;
419
420         if (!S_ISREG(file->mode)) {
421                 rprintf(FINFO, "skipping non-regular file \"%s\"\n",
422                         safe_fname(fname));
423                 return;
424         }
425
426         fnamecmp = fname;
427         fnamecmp_type = FNAMECMP_FNAME;
428
429         if (statret == -1 && basis_dir[0] != NULL) {
430                 int fallback_match = -1;
431                 int match_level = 0;
432                 int i = 0;
433                 do {
434                         pathjoin(fnamecmpbuf, sizeof fnamecmpbuf,
435                                  basis_dir[i], fname);
436                         if (link_stat(fnamecmpbuf, &st, 0) == 0
437                             && S_ISREG(st.st_mode)) {
438                                 statret = 0;
439                                 if (link_dest) {
440                                         if (!match_level) {
441                                                 fallback_match = i;
442                                                 match_level = 1;
443                                         } else if (match_level == 2
444                                             && !unchanged_attrs(file, &st))
445                                                 continue;
446                                         if (!unchanged_file(fnamecmpbuf, file, &st))
447                                                 continue;
448                                         fallback_match = i;
449                                         match_level = 2;
450                                         if (!unchanged_attrs(file, &st))
451                                                 continue;
452                                 }
453                                 match_level = 3;
454                                 break;
455                         }
456                 } while (basis_dir[++i] != NULL);
457                 if (statret == 0) {
458                         if (match_level < 3) {
459                                 i = fallback_match;
460                                 pathjoin(fnamecmpbuf, sizeof fnamecmpbuf,
461                                          basis_dir[i], fname);
462                         }
463 #if HAVE_LINK
464                         if (link_dest && match_level == 3 && !dry_run) {
465                                 if (do_link(fnamecmpbuf, fname) < 0) {
466                                         if (verbose) {
467                                                 rsyserr(FINFO, errno,
468                                                         "link %s => %s",
469                                                         fnamecmpbuf,
470                                                         safe_fname(fname));
471                                         }
472                                         fnamecmp = fnamecmpbuf;
473                                         fnamecmp_type = i;
474                                 }
475                         } else
476 #endif
477                         {
478                                 fnamecmp = fnamecmpbuf;
479                                 fnamecmp_type = i;
480                         }
481                 }
482         }
483
484         if (statret == 0 && !S_ISREG(st.st_mode)) {
485                 if (delete_file(fname) != 0)
486                         return;
487                 statret = -1;
488                 stat_errno = ENOENT;
489         }
490
491         if (partial_dir && (partialptr = partial_dir_fname(fname)) != NULL
492             && link_stat(partialptr, &partial_st, 0) == 0
493             && S_ISREG(partial_st.st_mode)) {
494                 if (statret == -1)
495                         goto prepare_to_open;
496         } else
497                 partialptr = NULL;
498
499         if (statret == -1) {
500                 if (preserve_hard_links && hard_link_check(file, HL_SKIP))
501                         return;
502                 if (stat_errno == ENOENT)
503                         goto notify_others;
504                 if (verbose > 1) {
505                         rsyserr(FERROR, stat_errno,
506                                 "recv_generator: failed to stat %s",
507                                 full_fname(fname));
508                 }
509                 return;
510         }
511
512         if (opt_ignore_existing && fnamecmp_type == FNAMECMP_FNAME) {
513                 if (verbose > 1)
514                         rprintf(FINFO, "%s exists\n", safe_fname(fname));
515                 return;
516         }
517
518         if (update_only && fnamecmp_type == FNAMECMP_FNAME
519             && cmp_modtime(st.st_mtime, file->modtime) > 0) {
520                 if (verbose > 1)
521                         rprintf(FINFO, "%s is newer\n", safe_fname(fname));
522                 return;
523         }
524
525         if (!compare_dest && fnamecmp_type <= FNAMECMP_BASIS_DIR_HIGH)
526                 ;
527         else if (unchanged_file(fnamecmp, file, &st)) {
528                 if (fnamecmp_type == FNAMECMP_FNAME)
529                         set_perms(fname, file, &st, PERMS_REPORT);
530                 return;
531         }
532
533 prepare_to_open:
534         if (partialptr) {
535                 st = partial_st;
536                 fnamecmp = partialptr;
537                 fnamecmp_type = FNAMECMP_PARTIAL_DIR;
538                 statret = 0;
539         }
540
541         if (dry_run || whole_file > 0) {
542                 statret = -1;
543                 goto notify_others;
544         }
545         if (read_batch)
546                 goto notify_others;
547
548         /* open the file */
549         fd = do_open(fnamecmp, O_RDONLY, 0);
550
551         if (fd == -1) {
552                 rsyserr(FERROR, errno, "failed to open %s, continuing",
553                         full_fname(fnamecmp));
554             pretend_missing:
555                 /* pretend the file didn't exist */
556                 if (preserve_hard_links && hard_link_check(file, HL_SKIP))
557                         return;
558                 statret = -1;
559                 goto notify_others;
560         }
561
562         if (inplace && make_backups) {
563                 if (!(backupptr = get_backup_name(fname))) {
564                         close(fd);
565                         return;
566                 }
567                 if (!(back_file = make_file(fname, NULL, NO_EXCLUDES))) {
568                         close(fd);
569                         goto pretend_missing;
570                 }
571                 if (robust_unlink(backupptr) && errno != ENOENT) {
572                         rsyserr(FERROR, errno, "unlink %s",
573                                 full_fname(backupptr));
574                         free(back_file);
575                         close(fd);
576                         return;
577                 }
578                 if ((f_copy = do_open(backupptr,
579                     O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, 0600)) < 0) {
580                         rsyserr(FERROR, errno, "open %s",
581                                 full_fname(backupptr));
582                         free(back_file);
583                         close(fd);
584                         return;
585                 }
586                 fnamecmp_type = FNAMECMP_BACKUP;
587         }
588
589         if (verbose > 3) {
590                 rprintf(FINFO, "gen mapped %s of size %.0f\n",
591                         safe_fname(fnamecmp), (double)st.st_size);
592         }
593
594         if (verbose > 2)
595                 rprintf(FINFO, "generating and sending sums for %d\n", i);
596
597 notify_others:
598         write_int(f_out, i);
599         if (protocol_version >= 29 && inplace && !read_batch)
600                 write_byte(f_out, fnamecmp_type);
601         if (f_out_name >= 0)
602                 write_byte(f_out_name, fnamecmp_type);
603
604         if (dry_run || read_batch)
605                 return;
606
607         if (statret == 0) {
608                 generate_and_send_sums(fd, st.st_size, f_out, f_copy);
609
610                 if (f_copy >= 0) {
611                         close(f_copy);
612                         set_perms(backupptr, back_file, NULL, 0);
613                         if (verbose > 1) {
614                                 rprintf(FINFO, "backed up %s to %s\n",
615                                         fname, backupptr);
616                         }
617                         free(back_file);
618                 }
619
620                 close(fd);
621         } else
622                 write_sum_head(f_out, NULL);
623 }
624
625
626 void generate_files(int f_out, struct file_list *flist, char *local_name,
627                     int f_out_name)
628 {
629         int i;
630         int phase = 0;
631         char fbuf[MAXPATHLEN];
632
633         if (verbose > 2) {
634                 rprintf(FINFO, "generator starting pid=%ld count=%d\n",
635                         (long)getpid(), flist->count);
636         }
637
638         if (verbose >= 2) {
639                 rprintf(FINFO,
640                         whole_file > 0
641                         ? "delta-transmission disabled for local transfer or --whole-file\n"
642                         : "delta transmission enabled\n");
643         }
644
645         /* we expect to just sit around now, so don't exit on a
646            timeout. If we really get a timeout then the other process should
647            exit */
648         io_timeout = 0;
649
650         for (i = 0; i < flist->count; i++) {
651                 struct file_struct *file = flist->files[i];
652                 struct file_struct copy;
653
654                 if (!file->basename)
655                         continue;
656                 /* we need to ensure that any directories we create have writeable
657                    permissions initially so that we can create the files within
658                    them. This is then fixed after the files are transferred */
659                 if (!am_root && S_ISDIR(file->mode) && !(file->mode & S_IWUSR)) {
660                         copy = *file;
661                         /* XXX: Could this be causing a problem on SCO?  Perhaps their
662                          * handling of permissions is strange? */
663                         copy.mode |= S_IWUSR; /* user write */
664                         file = &copy;
665                 }
666
667                 recv_generator(local_name ? local_name : f_name_to(file, fbuf),
668                                file, i, f_out, f_out_name);
669         }
670
671         phase++;
672         csum_length = SUM_LENGTH;
673         ignore_times = 1;
674
675         if (verbose > 2)
676                 rprintf(FINFO,"generate_files phase=%d\n",phase);
677
678         write_int(f_out, -1);
679
680         /* files can cycle through the system more than once
681          * to catch initial checksum errors */
682         while ((i = get_redo_num()) != -1) {
683                 struct file_struct *file = flist->files[i];
684                 recv_generator(local_name ? local_name : f_name_to(file, fbuf),
685                                file, i, f_out, f_out_name);
686         }
687
688         phase++;
689         if (verbose > 2)
690                 rprintf(FINFO,"generate_files phase=%d\n",phase);
691
692         write_int(f_out, -1);
693
694         /* Read post-redo-phase MSG_DONE and any prior messages. */
695         get_redo_num();
696
697         if (preserve_hard_links)
698                 do_hard_links();
699
700         /* now we need to fix any directory permissions that were
701          * modified during the transfer */
702         for (i = 0; i < flist->count; i++) {
703                 struct file_struct *file = flist->files[i];
704                 if (!file->basename || !S_ISDIR(file->mode))
705                         continue;
706                 recv_generator(local_name ? local_name : f_name(file),
707                                file, i, -1, -1);
708         }
709
710         if (verbose > 2)
711                 rprintf(FINFO,"generate_files finished\n");
712 }