1 /* -*- c-file-style: "linux" -*-
3 rsync -- fast file replication program
5 Copyright (C) 1996-2000 by Andrew Tridgell
6 Copyright (C) Paul Mackerras 1996
7 Copyright (C) 2002 by Martin Pool <mbp@samba.org>
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.
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.
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.
28 extern int relative_paths;
29 extern int keep_dirlinks;
30 extern int preserve_links;
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 preserve_times;
38 extern int omit_dir_times;
39 extern int delete_during;
40 extern int update_only;
41 extern int opt_ignore_existing;
43 extern int make_backups;
44 extern int csum_length;
45 extern int ignore_times;
47 extern OFF_T max_size;
48 extern int io_timeout;
49 extern int protocol_version;
50 extern int fuzzy_basis;
51 extern int always_checksum;
52 extern char *partial_dir;
53 extern char *basis_dir[];
54 extern int compare_dest;
56 extern int whole_file;
57 extern int local_server;
59 extern int read_batch;
60 extern int only_existing;
61 extern int orig_umask;
62 extern int safe_symlinks;
63 extern long block_size; /* "long" because popt can't set an int32. */
65 extern struct filter_list_struct server_filter_list;
67 static int unchanged_attrs(struct file_struct *file, STRUCT_STAT *st)
70 && (st->st_mode & CHMOD_BITS) != (file->mode & CHMOD_BITS))
73 if (am_root && preserve_uid && st->st_uid != file->uid)
76 if (preserve_gid && file->gid != GID_NONE && st->st_gid != file->gid)
83 #define SID_UPDATING ITEM_UPDATING
84 #define SID_REPORT_CHECKSUM ITEM_REPORT_CHECKSUM
85 #define SID_NO_DEST_AND_NO_UPDATE (1<<16)
87 static void itemize(struct file_struct *file, int statret, STRUCT_STAT *st,
88 int32 sflags, int f_out, int ndx)
90 int iflags = sflags & (SID_UPDATING | SID_REPORT_CHECKSUM);
93 if (S_ISREG(file->mode) && file->length != st->st_size)
94 iflags |= ITEM_REPORT_SIZE;
96 iflags |= ITEM_IS_NEW;
97 if (statret >= 0 && !(sflags & SID_NO_DEST_AND_NO_UPDATE)) {
98 int keep_time = !preserve_times ? 0
99 : S_ISDIR(file->mode) ? !omit_dir_times : !S_ISLNK(file->mode);
101 if ((iflags & ITEM_UPDATING && !keep_time)
102 || (keep_time && file->modtime != st->st_mtime))
103 iflags |= ITEM_REPORT_TIME;
104 if (preserve_perms && file->mode != st->st_mode)
105 iflags |= ITEM_REPORT_PERMS;
106 if (preserve_uid && am_root && file->uid != st->st_uid)
107 iflags |= ITEM_REPORT_OWNER;
108 if (preserve_gid && file->gid != GID_NONE && st->st_gid != file->gid)
109 iflags |= ITEM_REPORT_GROUP;
112 if (iflags && !read_batch) {
114 write_int(f_out, ndx);
115 write_byte(f_out, iflags);
116 write_byte(f_out, iflags >> 8);
121 /* Perform our quick-check heuristic for determining if a file is unchanged. */
122 static int unchanged_file(char *fn, struct file_struct *file, STRUCT_STAT *st)
124 if (st->st_size != file->length)
127 /* if always checksum is set then we use the checksum instead
128 of the file time to determine whether to sync */
129 if (always_checksum && S_ISREG(st->st_mode)) {
130 char sum[MD4_SUM_LENGTH];
131 file_checksum(fn, sum, st->st_size);
132 return memcmp(sum, file->u.sum, protocol_version < 21 ? 2
133 : MD4_SUM_LENGTH) == 0;
142 return cmp_modtime(st->st_mtime, file->modtime) == 0;
147 * set (initialize) the size entries in the per-file sum_struct
148 * calculating dynamic block and checksum sizes.
150 * This is only called from generate_and_send_sums() but is a separate
151 * function to encapsulate the logic.
153 * The block size is a rounded square root of file length.
155 * The checksum size is determined according to:
156 * blocksum_bits = BLOCKSUM_EXP + 2*log2(file_len) - log2(block_len)
157 * provided by Donovan Baarda which gives a probability of rsync
158 * algorithm corrupting data and falling back using the whole md4
161 * This might be made one of several selectable heuristics.
163 static void sum_sizes_sqroot(struct sum_struct *sum, int64 len)
169 blength = block_size;
170 else if (len <= BLOCK_SIZE * BLOCK_SIZE)
171 blength = BLOCK_SIZE;
176 for (c = 1, l = len, cnt = 0; l >>= 2; c <<= 1, cnt++) {}
177 if (cnt >= 31 || c >= MAX_BLOCK_SIZE)
178 blength = MAX_BLOCK_SIZE;
183 if (len < (int64)blength * blength)
186 } while (c >= 8); /* round to multiple of 8 */
187 blength = MAX(blength, BLOCK_SIZE);
191 if (protocol_version < 27) {
192 s2length = csum_length;
193 } else if (csum_length == SUM_LENGTH) {
194 s2length = SUM_LENGTH;
198 int b = BLOCKSUM_BIAS;
199 for (l = len; l >>= 1; b += 2) {}
200 for (c = blength; c >>= 1 && b; b--) {}
201 /* add a bit, subtract rollsum, round up. */
202 s2length = (b + 1 - 32 + 7) / 8; /* --optimize in compiler-- */
203 s2length = MAX(s2length, csum_length);
204 s2length = MIN(s2length, SUM_LENGTH);
208 sum->blength = blength;
209 sum->s2length = s2length;
210 sum->count = (len + (blength - 1)) / blength;
211 sum->remainder = (len % blength);
213 if (sum->count && verbose > 2) {
215 "count=%.0f rem=%ld blength=%ld s2length=%d flength=%.0f\n",
216 (double)sum->count, (long)sum->remainder, (long)sum->blength,
217 sum->s2length, (double)sum->flength);
223 * Generate and send a stream of signatures/checksums that describe a buffer
225 * Generate approximately one checksum every block_len bytes.
227 static void generate_and_send_sums(int fd, OFF_T len, int f_out, int f_copy)
230 struct map_struct *mapbuf;
231 struct sum_struct sum;
234 sum_sizes_sqroot(&sum, len);
237 mapbuf = map_file(fd, len, MAX_MAP_SIZE, sum.blength);
241 write_sum_head(f_out, &sum);
243 for (i = 0; i < sum.count; i++) {
244 int32 n1 = (int32)MIN(len, (OFF_T)sum.blength);
245 char *map = map_ptr(mapbuf, offset, n1);
246 uint32 sum1 = get_checksum1(map, n1);
247 char sum2[SUM_LENGTH];
250 full_write(f_copy, map, n1);
252 get_checksum2(map, n1, sum2);
256 "chunk[%.0f] offset=%.0f len=%ld sum1=%08lx\n",
257 (double)i, (double)offset, (long)n1,
258 (unsigned long)sum1);
260 write_int(f_out, sum1);
261 write_buf(f_out, sum2, sum.s2length);
271 /* Try to find a filename in the same dir as "fname" with a similar name. */
272 static int find_fuzzy(struct file_struct *file, struct file_list *dirlist)
274 int fname_len, fname_suf_len;
275 const char *fname_suf, *fname = file->basename;
276 uint32 lowest_dist = 0x7FFFFFFF;
277 int j, lowest_j = -1;
279 fname_len = strlen(fname);
280 fname_suf = find_filename_suffix(fname, fname_len, &fname_suf_len);
282 for (j = 0; j < dirlist->count; j++) {
283 struct file_struct *fp = dirlist->files[j];
284 const char *suf, *name;
288 if (!S_ISREG(fp->mode) || !fp->length
289 || fp->flags & FLAG_NO_FUZZY)
294 if (fp->length == file->length
295 && fp->modtime == file->modtime) {
298 "fuzzy size/modtime match for %s\n",
305 suf = find_filename_suffix(name, len, &suf_len);
307 dist = fuzzy_distance(name, len, fname, fname_len);
308 /* Add some extra weight to how well the suffixes match. */
309 dist += fuzzy_distance(suf, suf_len, fname_suf, fname_suf_len)
312 rprintf(FINFO, "fuzzy distance for %s = %d.%05d\n",
313 name, (int)(dist>>16), (int)(dist&0xFFFF));
315 if (dist <= lowest_dist) {
325 /* Acts on flist->file's ndx'th item, whose name is fname. If a directory,
326 * make sure it exists, and has the right permissions/timestamp info. For
327 * all other non-regular files (symlinks, etc.) we create them here. For
328 * regular files that have changed, we try to find a basis file and then
329 * start sending checksums.
331 * Note that f_out is set to -1 when doing final directory-permission and
332 * modification-time repair. */
333 static void recv_generator(char *fname, struct file_list *flist,
334 struct file_struct *file, int ndx,
335 int f_out, int f_out_name)
337 static int missing_below = -1;
338 static char *fuzzy_dirname = NULL;
339 static struct file_list *fuzzy_dirlist = NULL;
340 struct file_struct *fuzzy_file = NULL;
341 int fd = -1, f_copy = -1;
342 STRUCT_STAT st, partial_st;
343 struct file_struct *back_file = NULL;
344 int statret, stat_errno;
345 char *fnamecmp, *partialptr, *backupptr = NULL;
346 char fnamecmpbuf[MAXPATHLEN];
354 flist_free(fuzzy_dirlist);
355 fuzzy_dirlist = NULL;
356 fuzzy_dirname = NULL;
358 if (missing_below >= 0) {
366 rprintf(FINFO, "recv_generator(%s,%d)\n",
367 safe_fname(fname), ndx);
370 if (server_filter_list.head
371 && check_filter(&server_filter_list, fname,
372 S_ISDIR(file->mode)) < 0) {
374 rprintf(FINFO, "skipping server-excluded file \"%s\"\n",
380 if (missing_below >= 0 && file->dir.depth <= missing_below) {
388 if (fuzzy_basis && S_ISREG(file->mode)) {
389 char *dn = file->dirname ? file->dirname : ".";
390 /* Yes, identical dirnames are guaranteed to have
391 * identical pointers at this point. */
392 if (fuzzy_dirname != dn) {
394 flist_free(fuzzy_dirlist);
396 fuzzy_dirlist = get_dirlist(fuzzy_dirname, 1);
400 statret = link_stat(fname, &st,
401 keep_dirlinks && S_ISDIR(file->mode));
405 if (only_existing && statret == -1 && stat_errno == ENOENT) {
406 /* we only want to update existing files */
408 rprintf(FINFO, "not creating new file \"%s\"\n",
414 if (statret == 0 && !preserve_perms
415 && S_ISDIR(st.st_mode) == S_ISDIR(file->mode)) {
416 /* if the file exists already and we aren't perserving
417 * permissions then act as though the remote end sent
418 * us the file permissions we already have */
419 file->mode = (file->mode & ~CHMOD_BITS)
420 | (st.st_mode & CHMOD_BITS);
423 if (S_ISDIR(file->mode)) {
424 /* The file to be received is a directory, so we need
425 * to prepare appropriately. If there is already a
426 * file of that name and it is *not* a directory, then
427 * we need to delete it. If it doesn't exist, then
428 * (perhaps recursively) create it. */
429 if (statret == 0 && !S_ISDIR(st.st_mode)) {
430 delete_file(fname, DEL_TERSE);
433 if (dry_run && statret != 0 && missing_below < 0) {
434 missing_below = file->dir.depth;
437 if (protocol_version >= 29 && f_out != -1)
438 itemize(file, statret, &st, 0, f_out, ndx);
439 if (statret != 0 && do_mkdir(fname,file->mode) != 0 && errno != EEXIST) {
440 if (!relative_paths || errno != ENOENT
441 || create_directory_path(fname, orig_umask) < 0
442 || do_mkdir(fname, file->mode) < 0) {
443 rsyserr(FERROR, errno,
444 "recv_generator: mkdir %s failed",
448 if (set_perms(fname, file, statret ? NULL : &st, 0)
449 && verbose && protocol_version < 29 && f_out != -1)
450 rprintf(FINFO, "%s/\n", safe_fname(fname));
451 if (delete_during && f_out != -1 && csum_length != SUM_LENGTH
452 && (file->flags & FLAG_DEL_HERE))
453 delete_in_dir(flist, fname, file);
457 if (max_size && file->length > max_size) {
459 rprintf(FINFO, "%s is over max-size\n",
465 if (preserve_links && S_ISLNK(file->mode)) {
467 if (safe_symlinks && unsafe_symlink(file->u.link, fname)) {
470 "ignoring unsafe symlink %s -> \"%s\"\n",
472 safe_fname(file->u.link));
477 int dflag = S_ISDIR(st.st_mode) ? DEL_DIR : 0;
478 char lnk[MAXPATHLEN];
482 && (len = readlink(fname, lnk, MAXPATHLEN-1)) > 0) {
484 /* A link already pointing to the
485 * right place -- no further action
487 if (strcmp(lnk, file->u.link) == 0) {
488 if (protocol_version >= 29) {
489 itemize(file, 0, &st, 0,
492 set_perms(fname, file, &st,
497 /* Not the right symlink (or not a symlink), so
499 delete_file(fname, dflag | DEL_TERSE);
501 if (do_symlink(file->u.link,fname) != 0) {
502 rsyserr(FERROR, errno, "symlink %s -> \"%s\" failed",
503 full_fname(fname), safe_fname(file->u.link));
505 set_perms(fname,file,NULL,0);
506 if (protocol_version >= 29) {
507 itemize(file, statret, &st, SID_UPDATING,
509 } else if (verbose) {
510 rprintf(FINFO, "%s -> %s\n", safe_fname(fname),
511 safe_fname(file->u.link));
518 if (am_root && preserve_devices && IS_DEVICE(file->mode)) {
520 st.st_mode != file->mode ||
521 st.st_rdev != file->u.rdev) {
522 int dflag = S_ISDIR(st.st_mode) ? DEL_DIR : 0;
523 if (protocol_version >= 29) {
524 itemize(file, statret, &st, SID_UPDATING,
527 delete_file(fname, dflag | DEL_TERSE);
529 rprintf(FINFO,"mknod(%s,0%o,0x%x)\n",
531 (int)file->mode, (int)file->u.rdev);
533 if (do_mknod(fname,file->mode,file->u.rdev) != 0) {
534 rsyserr(FERROR, errno, "mknod %s failed",
537 set_perms(fname,file,NULL,0);
538 if (verbose && protocol_version < 29) {
539 rprintf(FINFO, "%s\n",
544 if (protocol_version >= 29) {
545 itemize(file, statret, &st, 0,
548 set_perms(fname, file, &st, PERMS_REPORT);
553 if (preserve_hard_links && hard_link_check(file, HL_CHECK_MASTER))
556 if (!S_ISREG(file->mode)) {
557 rprintf(FINFO, "skipping non-regular file \"%s\"\n",
563 fnamecmp_type = FNAMECMP_FNAME;
565 if (statret != 0 && basis_dir[0] != NULL) {
566 int fallback_match = -1;
570 pathjoin(fnamecmpbuf, sizeof fnamecmpbuf,
571 basis_dir[i], fname);
572 if (link_stat(fnamecmpbuf, &st, 0) == 0
573 && S_ISREG(st.st_mode)) {
579 } else if (match_level == 2
580 && !unchanged_attrs(file, &st))
582 if (!unchanged_file(fnamecmpbuf, file, &st))
586 if (!unchanged_attrs(file, &st))
592 } while (basis_dir[++i] != NULL);
594 if (match_level < 3) {
596 pathjoin(fnamecmpbuf, sizeof fnamecmpbuf,
597 basis_dir[i], fname);
600 if (link_dest && match_level == 3 && !dry_run) {
601 if (do_link(fnamecmpbuf, fname) < 0) {
603 rsyserr(FINFO, errno,
605 full_fname(fnamecmpbuf),
608 fnamecmp = fnamecmpbuf;
614 fnamecmp = fnamecmpbuf;
620 if (statret == 0 && !S_ISREG(st.st_mode)) {
621 int dflag = S_ISDIR(st.st_mode) ? DEL_DIR : 0;
622 if (delete_file(fname, dflag | DEL_TERSE) != 0)
628 if (partial_dir && (partialptr = partial_dir_fname(fname)) != NULL
629 && link_stat(partialptr, &partial_st, 0) == 0
630 && S_ISREG(partial_st.st_mode)) {
632 goto prepare_to_open;
636 if (statret != 0 && fuzzy_basis && dry_run <= 1) {
637 int j = find_fuzzy(file, fuzzy_dirlist);
639 fuzzy_file = fuzzy_dirlist->files[j];
640 f_name_to(fuzzy_file, fnamecmpbuf);
642 rprintf(FINFO, "fuzzy basis selected for %s: %s\n",
643 safe_fname(fname), safe_fname(fnamecmpbuf));
645 st.st_mode = fuzzy_file->mode;
646 st.st_size = fuzzy_file->length;
647 st.st_mtime = fuzzy_file->modtime;
649 fnamecmp = fnamecmpbuf;
650 fnamecmp_type = FNAMECMP_FUZZY;
655 if (preserve_hard_links && hard_link_check(file, HL_SKIP))
657 if (stat_errno == ENOENT)
660 rsyserr(FERROR, stat_errno,
661 "recv_generator: failed to stat %s",
667 if (opt_ignore_existing && fnamecmp_type == FNAMECMP_FNAME) {
669 rprintf(FINFO, "%s exists\n", safe_fname(fname));
673 if (update_only && fnamecmp_type == FNAMECMP_FNAME
674 && cmp_modtime(st.st_mtime, file->modtime) > 0) {
676 rprintf(FINFO, "%s is newer\n", safe_fname(fname));
680 if (!compare_dest && fnamecmp_type <= FNAMECMP_BASIS_DIR_HIGH)
682 else if (fnamecmp_type == FNAMECMP_FUZZY)
684 else if (unchanged_file(fnamecmp, file, &st)) {
685 if (protocol_version >= 29) {
686 itemize(file, statret, &st,
687 fnamecmp_type == FNAMECMP_FNAME
688 ? 0 : SID_NO_DEST_AND_NO_UPDATE,
691 if (fnamecmp_type == FNAMECMP_FNAME)
692 set_perms(fname, file, &st, PERMS_REPORT);
699 fnamecmp = partialptr;
700 fnamecmp_type = FNAMECMP_PARTIAL_DIR;
704 if (dry_run || read_batch)
706 if (whole_file > 0) {
713 int j = flist_find(fuzzy_dirlist, file);
714 if (j >= 0) /* don't use changing file as future fuzzy basis */
715 fuzzy_dirlist->files[j]->flags |= FLAG_NO_FUZZY;
719 fd = do_open(fnamecmp, O_RDONLY, 0);
722 rsyserr(FERROR, errno, "failed to open %s, continuing",
723 full_fname(fnamecmp));
725 /* pretend the file didn't exist */
726 if (preserve_hard_links && hard_link_check(file, HL_SKIP))
732 if (inplace && make_backups) {
733 if (!(backupptr = get_backup_name(fname))) {
737 if (!(back_file = make_file(fname, NULL, NO_FILTERS))) {
739 goto pretend_missing;
741 if (robust_unlink(backupptr) && errno != ENOENT) {
742 rsyserr(FERROR, errno, "unlink %s",
743 full_fname(backupptr));
748 if ((f_copy = do_open(backupptr,
749 O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, 0600)) < 0) {
750 rsyserr(FERROR, errno, "open %s",
751 full_fname(backupptr));
756 fnamecmp_type = FNAMECMP_BACKUP;
760 rprintf(FINFO, "gen mapped %s of size %.0f\n",
761 safe_fname(fnamecmp), (double)st.st_size);
765 rprintf(FINFO, "generating and sending sums for %d\n", ndx);
768 write_int(f_out, ndx);
769 if (protocol_version >= 29) {
770 itemize(file, statret, &st, SID_UPDATING
771 | (always_checksum ? SID_REPORT_CHECKSUM : 0),
773 if (inplace && !read_batch)
774 write_byte(f_out, fnamecmp_type);
776 if (f_out_name >= 0) {
777 write_byte(f_out_name, fnamecmp_type);
778 if (fnamecmp_type == FNAMECMP_FUZZY) {
779 uchar lenbuf[3], *lb = lenbuf;
780 int len = strlen(fuzzy_file->basename);
782 #if MAXPATHLEN > 0x7FFF
783 *lb++ = len / 0x10000 + 0x80;
786 *lb++ = len / 0x100 + 0x80;
790 write_buf(f_out_name, lenbuf, lb - lenbuf + 1);
791 write_buf(f_out_name, fuzzy_file->basename, len);
795 if (dry_run || read_batch)
799 generate_and_send_sums(fd, st.st_size, f_out, f_copy);
803 set_perms(backupptr, back_file, NULL, 0);
805 rprintf(FINFO, "backed up %s to %s\n",
806 safe_fname(fname), safe_fname(backupptr));
813 write_sum_head(f_out, NULL);
817 void generate_files(int f_out, struct file_list *flist, char *local_name,
822 char fbuf[MAXPATHLEN];
823 int need_retouch_dir_times = preserve_times && !omit_dir_times;
824 int need_retouch_dir_perms = 0;
825 int save_only_existing = only_existing;
826 int save_opt_ignore_existing = opt_ignore_existing;
829 rprintf(FINFO, "generator starting pid=%ld count=%d\n",
830 (long)getpid(), flist->count);
836 ? "delta-transmission disabled for local transfer or --whole-file\n"
837 : "delta transmission enabled\n");
840 /* We expect to just sit around now, so don't exit on a timeout.
841 * If we really get a timeout then the other process should exit. */
844 for (i = 0; i < flist->count; i++) {
845 struct file_struct *file = flist->files[i];
846 struct file_struct copy;
851 /* We need to ensure that any dirs we create have writeable
852 * permissions during the time we are putting files within
853 * them. This is then fixed after the transfer is done. */
854 if (!am_root && S_ISDIR(file->mode) && !(file->mode & S_IWUSR)) {
856 copy.mode |= S_IWUSR; /* user write */
858 need_retouch_dir_perms = 1;
861 recv_generator(local_name ? local_name : f_name_to(file, fbuf),
862 flist, file, i, f_out, f_out_name);
864 recv_generator(NULL, NULL, NULL, 0, -1, -1);
866 delete_in_dir(NULL, NULL, NULL);
869 csum_length = SUM_LENGTH;
870 only_existing = max_size = opt_ignore_existing = 0;
871 update_only = always_checksum = size_only = 0;
875 rprintf(FINFO,"generate_files phase=%d\n",phase);
877 write_int(f_out, -1);
879 /* files can cycle through the system more than once
880 * to catch initial checksum errors */
881 while ((i = get_redo_num()) != -1) {
882 struct file_struct *file = flist->files[i];
883 recv_generator(local_name ? local_name : f_name_to(file, fbuf),
884 flist, file, i, f_out, f_out_name);
888 only_existing = save_only_existing;
889 opt_ignore_existing = save_opt_ignore_existing;
892 rprintf(FINFO,"generate_files phase=%d\n",phase);
894 write_int(f_out, -1);
896 /* Read post-redo-phase MSG_DONE and any prior messages. */
899 if (preserve_hard_links)
902 if ((need_retouch_dir_perms || need_retouch_dir_times)
903 && !list_only && !local_name && !dry_run) {
904 /* Now we need to fix any directory permissions that were
905 * modified during the transfer and/or re-set any tweaked
906 * modified-time values. */
907 for (i = 0; i < flist->count; i++) {
908 struct file_struct *file = flist->files[i];
909 if (!file->basename || !S_ISDIR(file->mode))
911 if (!need_retouch_dir_times && file->mode & S_IWUSR)
913 recv_generator(local_name ? local_name : f_name(file),
914 flist, file, i, -1, -1);
917 recv_generator(NULL, NULL, NULL, 0, -1, -1);
920 rprintf(FINFO,"generate_files finished\n");