33fcf2abee96b04336e3ef79e6f6dafe64afe3fe
[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 csum_length;
40 extern int ignore_times;
41 extern int size_only;
42 extern int io_timeout;
43 extern int protocol_version;
44 extern int always_checksum;
45 extern char *compare_dest;
46 extern int link_dest;
47 extern int whole_file;
48 extern int local_server;
49 extern int list_only;
50 extern int read_batch;
51 extern int only_existing;
52 extern int orig_umask;
53 extern int safe_symlinks;
54 extern unsigned int block_size;
55
56 extern struct exclude_list_struct server_exclude_list;
57
58
59 /* choose whether to skip a particular file */
60 static int skip_file(char *fname, struct file_struct *file, STRUCT_STAT *st)
61 {
62         if (st->st_size != file->length)
63                 return 0;
64         if (link_dest) {
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
73                     && st->st_gid != file->gid)
74                         return 0;
75         }
76
77         /* if always checksum is set then we use the checksum instead
78            of the file time to determine whether to sync */
79         if (always_checksum && S_ISREG(st->st_mode)) {
80                 char sum[MD4_SUM_LENGTH];
81                 char fnamecmpdest[MAXPATHLEN];
82
83                 if (compare_dest != NULL) {
84                         if (access(fname, 0) != 0) {
85                                 pathjoin(fnamecmpdest, sizeof fnamecmpdest,
86                                          compare_dest, fname);
87                                 fname = fnamecmpdest;
88                         }
89                 }
90                 file_checksum(fname,sum,st->st_size);
91                 return memcmp(sum, file->u.sum, protocol_version < 21 ? 2
92                                                         : MD4_SUM_LENGTH) == 0;
93         }
94
95         if (size_only)
96                 return 1;
97
98         if (ignore_times)
99                 return 0;
100
101         return cmp_modtime(st->st_mtime, file->modtime) == 0;
102 }
103
104
105 /*
106  * NULL sum_struct means we have no checksums
107  */
108 void write_sum_head(int f, struct sum_struct *sum)
109 {
110         static struct sum_struct null_sum;
111
112         if (sum == NULL)
113                 sum = &null_sum;
114
115         write_int(f, sum->count);
116         write_int(f, sum->blength);
117         if (protocol_version >= 27)
118                 write_int(f, sum->s2length);
119         write_int(f, sum->remainder);
120 }
121
122 /*
123  * set (initialize) the size entries in the per-file sum_struct
124  * calculating dynamic block and checksum sizes.
125  *
126  * This is only called from generate_and_send_sums() but is a separate
127  * function to encapsulate the logic.
128  *
129  * The block size is a rounded square root of file length.
130  *
131  * The checksum size is determined according to:
132  *     blocksum_bits = BLOCKSUM_EXP + 2*log2(file_len) - log2(block_len)
133  * provided by Donovan Baarda which gives a probability of rsync
134  * algorithm corrupting data and falling back using the whole md4
135  * checksums.
136  *
137  * This might be made one of several selectable heuristics.
138  */
139
140 static void sum_sizes_sqroot(struct sum_struct *sum, uint64 len)
141 {
142         unsigned int blength;
143         int s2length;
144         uint32 c;
145         uint64 l;
146
147         if (block_size) {
148                 blength = block_size;
149         } else if (len <= BLOCK_SIZE * BLOCK_SIZE) {
150                 blength = BLOCK_SIZE;
151         } else {
152                 l = len;
153                 c = 1;
154                 while (l >>= 2) {
155                         c <<= 1;
156                 }
157                 blength = 0;
158                 do {
159                         blength |= c;
160                         if (len < (uint64)blength * blength)
161                                 blength &= ~c;
162                         c >>= 1;
163                 } while (c >= 8);       /* round to multiple of 8 */
164                 blength = MAX(blength, BLOCK_SIZE);
165                 blength = MIN(blength, MAX_MAP_SIZE);
166         }
167
168         if (protocol_version < 27) {
169                 s2length = csum_length;
170         } else if (csum_length == SUM_LENGTH) {
171                 s2length = SUM_LENGTH;
172         } else {
173                 int b = BLOCKSUM_BIAS;
174                 l = len;
175                 while (l >>= 1) {
176                         b += 2;
177                 }
178                 c = blength;
179                 while (c >>= 1 && b) {
180                         b--;
181                 }
182                 s2length = (b + 1 - 32 + 7) / 8; /* add a bit,
183                                                   * subtract rollsum,
184                                                   * round up
185                                                   *    --optimize in compiler--
186                                                   */
187                 s2length = MAX(s2length, csum_length);
188                 s2length = MIN(s2length, SUM_LENGTH);
189         }
190
191         sum->flength    = len;
192         sum->blength    = blength;
193         sum->s2length   = s2length;
194         sum->count      = (len + (blength - 1)) / blength;
195         sum->remainder  = (len % blength);
196
197         if (sum->count && verbose > 2) {
198                 rprintf(FINFO, "count=%.0f rem=%u blength=%u s2length=%d flength=%.0f\n",
199                         (double)sum->count, sum->remainder, sum->blength,
200                         sum->s2length, (double)sum->flength);
201         }
202 }
203
204
205 /*
206  * Generate and send a stream of signatures/checksums that describe a buffer
207  *
208  * Generate approximately one checksum every block_len bytes.
209  */
210 static void generate_and_send_sums(int fd, OFF_T len, int f_out)
211 {
212         size_t i;
213         struct map_struct *mapbuf;
214         struct sum_struct sum;
215         OFF_T offset = 0;
216
217         sum_sizes_sqroot(&sum, len);
218
219         if (len > 0)
220                 mapbuf = map_file(fd, len, sum.blength);
221         else
222                 mapbuf = NULL;
223
224         write_sum_head(f_out, &sum);
225
226         for (i = 0; i < sum.count; i++) {
227                 unsigned int n1 = MIN(len, sum.blength);
228                 char *map = map_ptr(mapbuf, offset, n1);
229                 uint32 sum1 = get_checksum1(map, n1);
230                 char sum2[SUM_LENGTH];
231
232                 get_checksum2(map, n1, sum2);
233
234                 if (verbose > 3) {
235                         rprintf(FINFO,
236                                 "chunk[%.0f] offset=%.0f len=%u sum1=%08lx\n",
237                                 (double)i, (double)offset, n1,
238                                 (unsigned long)sum1);
239                 }
240                 write_int(f_out, sum1);
241                 write_buf(f_out, sum2, sum.s2length);
242                 len -= n1;
243                 offset += n1;
244         }
245
246         if (mapbuf)
247                 unmap_file(mapbuf);
248 }
249
250
251
252 /*
253  * Acts on file number @p i from @p flist, whose name is @p fname.
254  *
255  * First fixes up permissions, then generates checksums for the file.
256  *
257  * @note This comment was added later by mbp who was trying to work it
258  * out.  It might be wrong.
259  */
260 static void recv_generator(char *fname, struct file_struct *file, int i,
261                            int f_out)
262 {
263         int fd;
264         STRUCT_STAT st;
265         int statret;
266         char *fnamecmp;
267         char fnamecmpbuf[MAXPATHLEN];
268
269         if (list_only)
270                 return;
271
272         if (verbose > 2)
273                 rprintf(FINFO,"recv_generator(%s,%d)\n",fname,i);
274
275         if (server_exclude_list.head
276             && check_exclude(&server_exclude_list, fname,
277                              S_ISDIR(file->mode)) < 0) {
278                 if (verbose) {
279                         rprintf(FINFO, "skipping server-excluded file \"%s\"\n",
280                                 fname);
281                 }
282                 return;
283         }
284
285         statret = link_stat(fname, &st, keep_dirlinks && S_ISDIR(file->mode));
286
287         if (only_existing && statret == -1 && errno == ENOENT) {
288                 /* we only want to update existing files */
289                 if (verbose > 1)
290                         rprintf(FINFO, "not creating new file \"%s\"\n", fname);
291                 return;
292         }
293
294         if (statret == 0 && !preserve_perms
295             && S_ISDIR(st.st_mode) == S_ISDIR(file->mode)) {
296                 /* if the file exists already and we aren't perserving
297                  * permissions then act as though the remote end sent
298                  * us the file permissions we already have */
299                 file->mode = (file->mode & ~CHMOD_BITS)
300                            | (st.st_mode & CHMOD_BITS);
301         }
302
303         if (S_ISDIR(file->mode)) {
304                 /* The file to be received is a directory, so we need
305                  * to prepare appropriately.  If there is already a
306                  * file of that name and it is *not* a directory, then
307                  * we need to delete it.  If it doesn't exist, then
308                  * recursively create it. */
309
310                 if (dry_run)
311                         return; /* TODO: causes inaccuracies -- fix */
312                 if (statret == 0 && !S_ISDIR(st.st_mode)) {
313                         if (robust_unlink(fname) != 0) {
314                                 rsyserr(FERROR, errno,
315                                         "recv_generator: unlink %s to make room for directory",
316                                         full_fname(fname));
317                                 return;
318                         }
319                         statret = -1;
320                 }
321                 if (statret != 0 && do_mkdir(fname,file->mode) != 0 && errno != EEXIST) {
322                         if (!(relative_paths && errno == ENOENT
323                             && create_directory_path(fname, orig_umask) == 0
324                             && do_mkdir(fname, file->mode) == 0)) {
325                                 rsyserr(FERROR, errno,
326                                         "recv_generator: mkdir %s failed",
327                                         full_fname(fname));
328                         }
329                 }
330                 /* f_out is set to -1 when doing final directory-permission
331                  * and modification-time repair. */
332                 if (set_perms(fname, file, statret ? NULL : &st, 0)
333                     && verbose && f_out != -1)
334                         rprintf(FINFO,"%s/\n",fname);
335                 return;
336         }
337
338         if (preserve_links && S_ISLNK(file->mode)) {
339 #if SUPPORT_LINKS
340                 char lnk[MAXPATHLEN];
341                 int l;
342
343                 if (safe_symlinks && unsafe_symlink(file->u.link, fname)) {
344                         if (verbose) {
345                                 rprintf(FINFO, "ignoring unsafe symlink %s -> \"%s\"\n",
346                                         full_fname(fname), file->u.link);
347                         }
348                         return;
349                 }
350                 if (statret == 0) {
351                         l = readlink(fname,lnk,MAXPATHLEN-1);
352                         if (l > 0) {
353                                 lnk[l] = 0;
354                                 /* A link already pointing to the
355                                  * right place -- no further action
356                                  * required. */
357                                 if (strcmp(lnk,file->u.link) == 0) {
358                                         set_perms(fname, file, &st,
359                                                   PERMS_REPORT);
360                                         return;
361                                 }
362                         }
363                         /* Not a symlink, so delete whatever's
364                          * already there and put a new symlink
365                          * in place. */
366                         delete_file(fname);
367                 }
368                 if (do_symlink(file->u.link,fname) != 0) {
369                         rsyserr(FERROR, errno, "symlink %s -> \"%s\" failed",
370                                 full_fname(fname), file->u.link);
371                 } else {
372                         set_perms(fname,file,NULL,0);
373                         if (verbose) {
374                                 rprintf(FINFO,"%s -> %s\n", fname,file->u.link);
375                         }
376                 }
377 #endif
378                 return;
379         }
380
381 #ifdef HAVE_MKNOD
382         if (am_root && preserve_devices && IS_DEVICE(file->mode)) {
383                 if (statret != 0 ||
384                     st.st_mode != file->mode ||
385                     st.st_rdev != file->u.rdev) {
386                         delete_file(fname);
387                         if (verbose > 2) {
388                                 rprintf(FINFO,"mknod(%s,0%o,0x%x)\n",
389                                         fname,(int)file->mode,(int)file->u.rdev);
390                         }
391                         if (do_mknod(fname,file->mode,file->u.rdev) != 0) {
392                                 rsyserr(FERROR, errno, "mknod %s failed",
393                                         full_fname(fname));
394                         } else {
395                                 set_perms(fname,file,NULL,0);
396                                 if (verbose)
397                                         rprintf(FINFO,"%s\n",fname);
398                         }
399                 } else {
400                         set_perms(fname, file, &st, PERMS_REPORT);
401                 }
402                 return;
403         }
404 #endif
405
406         if (preserve_hard_links && hard_link_check(file, HL_CHECK_MASTER))
407                 return;
408
409         if (!S_ISREG(file->mode)) {
410                 rprintf(FINFO, "skipping non-regular file \"%s\"\n",fname);
411                 return;
412         }
413
414         fnamecmp = fname;
415
416         if (statret == -1 && compare_dest != NULL) {
417                 /* try the file at compare_dest instead */
418                 int saveerrno = errno;
419                 pathjoin(fnamecmpbuf, sizeof fnamecmpbuf, compare_dest, fname);
420                 statret = link_stat(fnamecmpbuf, &st, 0);
421                 if (!S_ISREG(st.st_mode))
422                         statret = -1;
423                 if (statret == -1)
424                         errno = saveerrno;
425 #if HAVE_LINK
426                 else if (link_dest && !dry_run) {
427                         if (do_link(fnamecmpbuf, fname) != 0) {
428                                 if (verbose > 0) {
429                                         rsyserr(FINFO, errno, "link %s => %s",
430                                                 fnamecmpbuf, fname);
431                                 }
432                         }
433                         fnamecmp = fnamecmpbuf;
434                 }
435 #endif
436                 else
437                         fnamecmp = fnamecmpbuf;
438         }
439
440         if (statret == -1) {
441                 if (preserve_hard_links && hard_link_check(file, HL_SKIP))
442                         return;
443                 if (errno == ENOENT) {
444                         write_int(f_out,i);
445                         if (!dry_run && !read_batch)
446                                 write_sum_head(f_out, NULL);
447                 } else if (verbose > 1) {
448                         rsyserr(FERROR, errno,
449                                 "recv_generator: failed to open %s",
450                                 full_fname(fname));
451                 }
452                 return;
453         }
454
455         if (!S_ISREG(st.st_mode)) {
456                 if (delete_file(fname) != 0) {
457                         return;
458                 }
459
460                 /* now pretend the file didn't exist */
461                 if (preserve_hard_links && hard_link_check(file, HL_SKIP))
462                         return;
463                 write_int(f_out,i);
464                 if (!dry_run && !read_batch)
465                         write_sum_head(f_out, NULL);
466                 return;
467         }
468
469         if (opt_ignore_existing && fnamecmp == fname) {
470                 if (verbose > 1)
471                         rprintf(FINFO,"%s exists\n",fname);
472                 return;
473         }
474
475         if (update_only && fnamecmp == fname
476             && cmp_modtime(st.st_mtime, file->modtime) > 0) {
477                 if (verbose > 1)
478                         rprintf(FINFO,"%s is newer\n",fname);
479                 return;
480         }
481
482         if (skip_file(fname, file, &st)) {
483                 if (fnamecmp == fname)
484                         set_perms(fname, file, &st, PERMS_REPORT);
485                 return;
486         }
487
488         if (dry_run || read_batch) {
489                 write_int(f_out,i);
490                 return;
491         }
492
493         if (whole_file > 0) {
494                 write_int(f_out,i);
495                 write_sum_head(f_out, NULL);
496                 return;
497         }
498
499         /* open the file */
500         fd = do_open(fnamecmp, O_RDONLY, 0);
501
502         if (fd == -1) {
503                 rsyserr(FERROR, errno, "failed to open %s, continuing",
504                         full_fname(fnamecmp));
505                 /* pretend the file didn't exist */
506                 if (preserve_hard_links && hard_link_check(file, HL_SKIP))
507                         return;
508                 write_int(f_out,i);
509                 write_sum_head(f_out, NULL);
510                 return;
511         }
512
513         if (verbose > 3) {
514                 rprintf(FINFO,"gen mapped %s of size %.0f\n", fnamecmp,
515                         (double)st.st_size);
516         }
517
518         if (verbose > 2)
519                 rprintf(FINFO, "generating and sending sums for %d\n", i);
520
521         write_int(f_out,i);
522         generate_and_send_sums(fd, st.st_size, f_out);
523
524         close(fd);
525 }
526
527
528 void generate_files(int f_out, struct file_list *flist, char *local_name)
529 {
530         int i;
531         int phase = 0;
532         char fbuf[MAXPATHLEN];
533
534         if (verbose > 2) {
535                 rprintf(FINFO, "generator starting pid=%ld count=%d\n",
536                         (long)getpid(), flist->count);
537         }
538
539         if (verbose >= 2) {
540                 rprintf(FINFO,
541                         whole_file > 0
542                         ? "delta-transmission disabled for local transfer or --whole-file\n"
543                         : "delta transmission enabled\n");
544         }
545
546         /* we expect to just sit around now, so don't exit on a
547            timeout. If we really get a timeout then the other process should
548            exit */
549         io_timeout = 0;
550
551         for (i = 0; i < flist->count; i++) {
552                 struct file_struct *file = flist->files[i];
553                 struct file_struct copy;
554
555                 if (!file->basename)
556                         continue;
557                 /* we need to ensure that any directories we create have writeable
558                    permissions initially so that we can create the files within
559                    them. This is then fixed after the files are transferred */
560                 if (!am_root && S_ISDIR(file->mode) && !(file->mode & S_IWUSR)) {
561                         copy = *file;
562                         /* XXX: Could this be causing a problem on SCO?  Perhaps their
563                          * handling of permissions is strange? */
564                         copy.mode |= S_IWUSR; /* user write */
565                         file = &copy;
566                 }
567
568                 recv_generator(local_name ? local_name : f_name_to(file, fbuf),
569                                file, i, f_out);
570         }
571
572         phase++;
573         csum_length = SUM_LENGTH;
574         ignore_times = 1;
575
576         if (verbose > 2)
577                 rprintf(FINFO,"generate_files phase=%d\n",phase);
578
579         write_int(f_out, -1);
580
581         /* files can cycle through the system more than once
582          * to catch initial checksum errors */
583         while ((i = get_redo_num()) != -1) {
584                 struct file_struct *file = flist->files[i];
585                 recv_generator(local_name ? local_name : f_name_to(file, fbuf),
586                                file, i, f_out);
587         }
588
589         phase++;
590         if (verbose > 2)
591                 rprintf(FINFO,"generate_files phase=%d\n",phase);
592
593         write_int(f_out, -1);
594
595         if (preserve_hard_links)
596                 do_hard_links();
597
598         /* now we need to fix any directory permissions that were
599          * modified during the transfer */
600         for (i = 0; i < flist->count; i++) {
601                 struct file_struct *file = flist->files[i];
602                 if (!file->basename || !S_ISDIR(file->mode))
603                         continue;
604                 recv_generator(local_name ? local_name : f_name(file),
605                                file, i, -1);
606         }
607
608         if (verbose > 2)
609                 rprintf(FINFO,"generate_files finished\n");
610 }