Switching to GPL 3.
[rsync/rsync.git] / util.c
... / ...
CommitLineData
1/*
2 * Utility routines used in rsync.
3 *
4 * Copyright (C) 1996-2000 Andrew Tridgell
5 * Copyright (C) 1996 Paul Mackerras
6 * Copyright (C) 2001, 2002 Martin Pool <mbp@samba.org>
7 * Copyright (C) 2003-2007 Wayne Davison
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 version 3 as
11 * published by the Free Software Foundation.
12 *
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License along
19 * with this program; if not, visit the http://fsf.org website.
20 */
21
22#include "rsync.h"
23
24extern int verbose;
25extern int dry_run;
26extern int module_id;
27extern int modify_window;
28extern int relative_paths;
29extern int human_readable;
30extern char *module_dir;
31extern unsigned int module_dirlen;
32extern mode_t orig_umask;
33extern char *partial_dir;
34extern struct filter_list_struct server_filter_list;
35
36int sanitize_paths = 0;
37
38char curr_dir[MAXPATHLEN];
39unsigned int curr_dir_len;
40int curr_dir_depth; /* This is only set for a sanitizing daemon. */
41
42/* Set a fd into nonblocking mode. */
43void set_nonblocking(int fd)
44{
45 int val;
46
47 if ((val = fcntl(fd, F_GETFL)) == -1)
48 return;
49 if (!(val & NONBLOCK_FLAG)) {
50 val |= NONBLOCK_FLAG;
51 fcntl(fd, F_SETFL, val);
52 }
53}
54
55/* Set a fd into blocking mode. */
56void set_blocking(int fd)
57{
58 int val;
59
60 if ((val = fcntl(fd, F_GETFL)) == -1)
61 return;
62 if (val & NONBLOCK_FLAG) {
63 val &= ~NONBLOCK_FLAG;
64 fcntl(fd, F_SETFL, val);
65 }
66}
67
68/**
69 * Create a file descriptor pair - like pipe() but use socketpair if
70 * possible (because of blocking issues on pipes).
71 *
72 * Always set non-blocking.
73 */
74int fd_pair(int fd[2])
75{
76 int ret;
77
78#ifdef HAVE_SOCKETPAIR
79 ret = socketpair(AF_UNIX, SOCK_STREAM, 0, fd);
80#else
81 ret = pipe(fd);
82#endif
83
84 if (ret == 0) {
85 set_nonblocking(fd[0]);
86 set_nonblocking(fd[1]);
87 }
88
89 return ret;
90}
91
92void print_child_argv(char **cmd)
93{
94 rprintf(FCLIENT, "opening connection using ");
95 for (; *cmd; cmd++) {
96 /* Look for characters that ought to be quoted. This
97 * is not a great quoting algorithm, but it's
98 * sufficient for a log message. */
99 if (strspn(*cmd, "abcdefghijklmnopqrstuvwxyz"
100 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
101 "0123456789"
102 ",.-_=+@/") != strlen(*cmd)) {
103 rprintf(FCLIENT, "\"%s\" ", *cmd);
104 } else {
105 rprintf(FCLIENT, "%s ", *cmd);
106 }
107 }
108 rprintf(FCLIENT, "\n");
109}
110
111NORETURN void out_of_memory(const char *str)
112{
113 rprintf(FERROR, "ERROR: out of memory in %s [%s]\n", str, who_am_i());
114 exit_cleanup(RERR_MALLOC);
115}
116
117NORETURN void overflow_exit(const char *str)
118{
119 rprintf(FERROR, "ERROR: buffer overflow in %s [%s]\n", str, who_am_i());
120 exit_cleanup(RERR_MALLOC);
121}
122
123int set_modtime(const char *fname, time_t modtime, mode_t mode)
124{
125#if !defined HAVE_LUTIMES || !defined HAVE_UTIMES
126 if (S_ISLNK(mode))
127 return 1;
128#endif
129
130 if (verbose > 2) {
131 rprintf(FINFO, "set modtime of %s to (%ld) %s",
132 fname, (long)modtime,
133 asctime(localtime(&modtime)));
134 }
135
136 if (dry_run)
137 return 0;
138
139 {
140#ifdef HAVE_UTIMES
141 struct timeval t[2];
142 t[0].tv_sec = time(NULL);
143 t[0].tv_usec = 0;
144 t[1].tv_sec = modtime;
145 t[1].tv_usec = 0;
146# ifdef HAVE_LUTIMES
147 if (S_ISLNK(mode)) {
148 lutimes(fname, t);
149 return 0; /* ignore errors */
150 }
151# endif
152 return utimes(fname, t);
153#elif defined HAVE_UTIMBUF
154 struct utimbuf tbuf;
155 tbuf.actime = time(NULL);
156 tbuf.modtime = modtime;
157 return utime(fname,&tbuf);
158#elif defined HAVE_UTIME
159 time_t t[2];
160 t[0] = time(NULL);
161 t[1] = modtime;
162 return utime(fname,t);
163#else
164#error No file-time-modification routine found!
165#endif
166 }
167}
168
169/* This creates a new directory with default permissions. Since there
170 * might be some directory-default permissions affecting this, we can't
171 * force the permissions directly using the original umask and mkdir(). */
172int mkdir_defmode(char *fname)
173{
174 int ret;
175
176 umask(orig_umask);
177 ret = do_mkdir(fname, ACCESSPERMS);
178 umask(0);
179
180 return ret;
181}
182
183/* Create any necessary directories in fname. Any missing directories are
184 * created with default permissions. */
185int create_directory_path(char *fname)
186{
187 char *p;
188 int ret = 0;
189
190 while (*fname == '/')
191 fname++;
192 while (strncmp(fname, "./", 2) == 0)
193 fname += 2;
194
195 umask(orig_umask);
196 p = fname;
197 while ((p = strchr(p,'/')) != NULL) {
198 *p = '\0';
199 if (do_mkdir(fname, ACCESSPERMS) < 0 && errno != EEXIST)
200 ret = -1;
201 *p++ = '/';
202 }
203 umask(0);
204
205 return ret;
206}
207
208/**
209 * Write @p len bytes at @p ptr to descriptor @p desc, retrying if
210 * interrupted.
211 *
212 * @retval len upon success
213 *
214 * @retval <0 write's (negative) error code
215 *
216 * Derived from GNU C's cccp.c.
217 */
218int full_write(int desc, const char *ptr, size_t len)
219{
220 int total_written;
221
222 total_written = 0;
223 while (len > 0) {
224 int written = write(desc, ptr, len);
225 if (written < 0) {
226 if (errno == EINTR)
227 continue;
228 return written;
229 }
230 total_written += written;
231 ptr += written;
232 len -= written;
233 }
234 return total_written;
235}
236
237/**
238 * Read @p len bytes at @p ptr from descriptor @p desc, retrying if
239 * interrupted.
240 *
241 * @retval >0 the actual number of bytes read
242 *
243 * @retval 0 for EOF
244 *
245 * @retval <0 for an error.
246 *
247 * Derived from GNU C's cccp.c. */
248static int safe_read(int desc, char *ptr, size_t len)
249{
250 int n_chars;
251
252 if (len == 0)
253 return len;
254
255 do {
256 n_chars = read(desc, ptr, len);
257 } while (n_chars < 0 && errno == EINTR);
258
259 return n_chars;
260}
261
262/** Copy a file.
263 *
264 * This is used in conjunction with the --temp-dir, --backup, and
265 * --copy-dest options. */
266int copy_file(const char *source, const char *dest, mode_t mode)
267{
268 int ifd;
269 int ofd;
270 char buf[1024 * 8];
271 int len; /* Number of bytes read into `buf'. */
272
273 ifd = do_open(source, O_RDONLY, 0);
274 if (ifd == -1) {
275 rsyserr(FERROR, errno, "open %s", full_fname(source));
276 return -1;
277 }
278
279 if (robust_unlink(dest) && errno != ENOENT) {
280 rsyserr(FERROR, errno, "unlink %s", full_fname(dest));
281 return -1;
282 }
283
284 ofd = do_open(dest, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, mode);
285 if (ofd == -1) {
286 rsyserr(FERROR, errno, "open %s", full_fname(dest));
287 close(ifd);
288 return -1;
289 }
290
291 while ((len = safe_read(ifd, buf, sizeof buf)) > 0) {
292 if (full_write(ofd, buf, len) < 0) {
293 rsyserr(FERROR, errno, "write %s", full_fname(dest));
294 close(ifd);
295 close(ofd);
296 return -1;
297 }
298 }
299
300 if (len < 0) {
301 rsyserr(FERROR, errno, "read %s", full_fname(source));
302 close(ifd);
303 close(ofd);
304 return -1;
305 }
306
307 if (close(ifd) < 0) {
308 rsyserr(FINFO, errno, "close failed on %s",
309 full_fname(source));
310 }
311
312 if (close(ofd) < 0) {
313 rsyserr(FERROR, errno, "close failed on %s",
314 full_fname(dest));
315 return -1;
316 }
317
318 return 0;
319}
320
321/* MAX_RENAMES should be 10**MAX_RENAMES_DIGITS */
322#define MAX_RENAMES_DIGITS 3
323#define MAX_RENAMES 1000
324
325/**
326 * Robust unlink: some OS'es (HPUX) refuse to unlink busy files, so
327 * rename to <path>/.rsyncNNN instead.
328 *
329 * Note that successive rsync runs will shuffle the filenames around a
330 * bit as long as the file is still busy; this is because this function
331 * does not know if the unlink call is due to a new file coming in, or
332 * --delete trying to remove old .rsyncNNN files, hence it renames it
333 * each time.
334 **/
335int robust_unlink(const char *fname)
336{
337#ifndef ETXTBSY
338 return do_unlink(fname);
339#else
340 static int counter = 1;
341 int rc, pos, start;
342 char path[MAXPATHLEN];
343
344 rc = do_unlink(fname);
345 if (rc == 0 || errno != ETXTBSY)
346 return rc;
347
348 if ((pos = strlcpy(path, fname, MAXPATHLEN)) >= MAXPATHLEN)
349 pos = MAXPATHLEN - 1;
350
351 while (pos > 0 && path[pos-1] != '/')
352 pos--;
353 pos += strlcpy(path+pos, ".rsync", MAXPATHLEN-pos);
354
355 if (pos > (MAXPATHLEN-MAX_RENAMES_DIGITS-1)) {
356 errno = ETXTBSY;
357 return -1;
358 }
359
360 /* start where the last one left off to reduce chance of clashes */
361 start = counter;
362 do {
363 snprintf(&path[pos], MAX_RENAMES_DIGITS+1, "%03d", counter);
364 if (++counter >= MAX_RENAMES)
365 counter = 1;
366 } while ((rc = access(path, 0)) == 0 && counter != start);
367
368 if (verbose > 0) {
369 rprintf(FINFO,"renaming %s to %s because of text busy\n",
370 fname, path);
371 }
372
373 /* maybe we should return rename()'s exit status? Nah. */
374 if (do_rename(fname, path) != 0) {
375 errno = ETXTBSY;
376 return -1;
377 }
378 return 0;
379#endif
380}
381
382/* Returns 0 on successful rename, 1 if we successfully copied the file
383 * across filesystems, -2 if copy_file() failed, and -1 on other errors.
384 * If partialptr is not NULL and we need to do a copy, copy the file into
385 * the active partial-dir instead of over the destination file. */
386int robust_rename(const char *from, const char *to, const char *partialptr,
387 int mode)
388{
389 int tries = 4;
390
391 while (tries--) {
392 if (do_rename(from, to) == 0)
393 return 0;
394
395 switch (errno) {
396#ifdef ETXTBSY
397 case ETXTBSY:
398 if (robust_unlink(to) != 0)
399 return -1;
400 break;
401#endif
402 case EXDEV:
403 if (partialptr) {
404 if (!handle_partial_dir(partialptr,PDIR_CREATE))
405 return -1;
406 to = partialptr;
407 }
408 if (copy_file(from, to, mode) != 0)
409 return -2;
410 do_unlink(from);
411 return 1;
412 default:
413 return -1;
414 }
415 }
416 return -1;
417}
418
419static pid_t all_pids[10];
420static int num_pids;
421
422/** Fork and record the pid of the child. **/
423pid_t do_fork(void)
424{
425 pid_t newpid = fork();
426
427 if (newpid != 0 && newpid != -1) {
428 all_pids[num_pids++] = newpid;
429 }
430 return newpid;
431}
432
433/**
434 * Kill all children.
435 *
436 * @todo It would be kind of nice to make sure that they are actually
437 * all our children before we kill them, because their pids may have
438 * been recycled by some other process. Perhaps when we wait for a
439 * child, we should remove it from this array. Alternatively we could
440 * perhaps use process groups, but I think that would not work on
441 * ancient Unix versions that don't support them.
442 **/
443void kill_all(int sig)
444{
445 int i;
446
447 for (i = 0; i < num_pids; i++) {
448 /* Let's just be a little careful where we
449 * point that gun, hey? See kill(2) for the
450 * magic caused by negative values. */
451 pid_t p = all_pids[i];
452
453 if (p == getpid())
454 continue;
455 if (p <= 0)
456 continue;
457
458 kill(p, sig);
459 }
460}
461
462/** Turn a user name into a uid */
463int name_to_uid(const char *name, uid_t *uid)
464{
465 struct passwd *pass;
466 if (!name || !*name)
467 return 0;
468 pass = getpwnam(name);
469 if (pass) {
470 *uid = pass->pw_uid;
471 return 1;
472 }
473 return 0;
474}
475
476/** Turn a group name into a gid */
477int name_to_gid(const char *name, gid_t *gid)
478{
479 struct group *grp;
480 if (!name || !*name)
481 return 0;
482 grp = getgrnam(name);
483 if (grp) {
484 *gid = grp->gr_gid;
485 return 1;
486 }
487 return 0;
488}
489
490/** Lock a byte range in a open file */
491int lock_range(int fd, int offset, int len)
492{
493 struct flock lock;
494
495 lock.l_type = F_WRLCK;
496 lock.l_whence = SEEK_SET;
497 lock.l_start = offset;
498 lock.l_len = len;
499 lock.l_pid = 0;
500
501 return fcntl(fd,F_SETLK,&lock) == 0;
502}
503
504static int filter_server_path(char *arg)
505{
506 char *s;
507
508 if (server_filter_list.head) {
509 for (s = arg; (s = strchr(s, '/')) != NULL; ) {
510 *s = '\0';
511 if (check_filter(&server_filter_list, arg, 1) < 0) {
512 /* We must leave arg truncated! */
513 return 1;
514 }
515 *s++ = '/';
516 }
517 }
518 return 0;
519}
520
521static void glob_expand_one(char *s, char ***argv_ptr, int *argc_ptr,
522 int *maxargs_ptr)
523{
524 char **argv = *argv_ptr;
525 int argc = *argc_ptr;
526 int maxargs = *maxargs_ptr;
527#if !defined HAVE_GLOB || !defined HAVE_GLOB_H
528 if (argc == maxargs) {
529 maxargs += MAX_ARGS;
530 if (!(argv = realloc_array(argv, char *, maxargs)))
531 out_of_memory("glob_expand_one");
532 *argv_ptr = argv;
533 *maxargs_ptr = maxargs;
534 }
535 if (!*s)
536 s = ".";
537 s = argv[argc++] = strdup(s);
538 filter_server_path(s);
539#else
540 glob_t globbuf;
541
542 if (maxargs <= argc)
543 return;
544 if (!*s)
545 s = ".";
546
547 if (sanitize_paths)
548 s = sanitize_path(NULL, s, "", 0, NULL);
549 else
550 s = strdup(s);
551
552 memset(&globbuf, 0, sizeof globbuf);
553 if (!filter_server_path(s))
554 glob(s, 0, NULL, &globbuf);
555 if (MAX((int)globbuf.gl_pathc, 1) > maxargs - argc) {
556 maxargs += globbuf.gl_pathc + MAX_ARGS;
557 if (!(argv = realloc_array(argv, char *, maxargs)))
558 out_of_memory("glob_expand_one");
559 *argv_ptr = argv;
560 *maxargs_ptr = maxargs;
561 }
562 if (globbuf.gl_pathc == 0)
563 argv[argc++] = s;
564 else {
565 int i;
566 free(s);
567 for (i = 0; i < (int)globbuf.gl_pathc; i++) {
568 if (!(argv[argc++] = strdup(globbuf.gl_pathv[i])))
569 out_of_memory("glob_expand_one");
570 }
571 }
572 globfree(&globbuf);
573#endif
574 *argc_ptr = argc;
575}
576
577/* This routine is only used in daemon mode. */
578void glob_expand(char *base1, char ***argv_ptr, int *argc_ptr, int *maxargs_ptr)
579{
580 char *s = (*argv_ptr)[*argc_ptr];
581 char *p, *q;
582 char *base = base1;
583 int base_len = strlen(base);
584
585 if (!s || !*s)
586 return;
587
588 if (strncmp(s, base, base_len) == 0)
589 s += base_len;
590
591 if (!(s = strdup(s)))
592 out_of_memory("glob_expand");
593
594 if (asprintf(&base," %s/", base1) <= 0)
595 out_of_memory("glob_expand");
596 base_len++;
597
598 for (q = s; *q; q = p + base_len) {
599 if ((p = strstr(q, base)) != NULL)
600 *p = '\0'; /* split it at this point */
601 glob_expand_one(q, argv_ptr, argc_ptr, maxargs_ptr);
602 if (!p)
603 break;
604 }
605
606 free(s);
607 free(base);
608}
609
610/**
611 * Convert a string to lower case
612 **/
613void strlower(char *s)
614{
615 while (*s) {
616 if (isUpper(s))
617 *s = toLower(s);
618 s++;
619 }
620}
621
622/* Join strings p1 & p2 into "dest" with a guaranteed '/' between them. (If
623 * p1 ends with a '/', no extra '/' is inserted.) Returns the length of both
624 * strings + 1 (if '/' was inserted), regardless of whether the null-terminated
625 * string fits into destsize. */
626size_t pathjoin(char *dest, size_t destsize, const char *p1, const char *p2)
627{
628 size_t len = strlcpy(dest, p1, destsize);
629 if (len < destsize - 1) {
630 if (!len || dest[len-1] != '/')
631 dest[len++] = '/';
632 if (len < destsize - 1)
633 len += strlcpy(dest + len, p2, destsize - len);
634 else {
635 dest[len] = '\0';
636 len += strlen(p2);
637 }
638 }
639 else
640 len += strlen(p2) + 1; /* Assume we'd insert a '/'. */
641 return len;
642}
643
644/* Join any number of strings together, putting them in "dest". The return
645 * value is the length of all the strings, regardless of whether the null-
646 * terminated whole fits in destsize. Your list of string pointers must end
647 * with a NULL to indicate the end of the list. */
648size_t stringjoin(char *dest, size_t destsize, ...)
649{
650 va_list ap;
651 size_t len, ret = 0;
652 const char *src;
653
654 va_start(ap, destsize);
655 while (1) {
656 if (!(src = va_arg(ap, const char *)))
657 break;
658 len = strlen(src);
659 ret += len;
660 if (destsize > 1) {
661 if (len >= destsize)
662 len = destsize - 1;
663 memcpy(dest, src, len);
664 destsize -= len;
665 dest += len;
666 }
667 }
668 *dest = '\0';
669 va_end(ap);
670
671 return ret;
672}
673
674int count_dir_elements(const char *p)
675{
676 int cnt = 0, new_component = 1;
677 while (*p) {
678 if (*p++ == '/')
679 new_component = (*p != '.' || (p[1] != '/' && p[1] != '\0'));
680 else if (new_component) {
681 new_component = 0;
682 cnt++;
683 }
684 }
685 return cnt;
686}
687
688/* Turns multiple adjacent slashes into a single slash, gets rid of "./"
689 * elements (but not a trailing dot dir), removes a trailing slash, and
690 * optionally collapses ".." elements (except for those at the start of the
691 * string). If the resulting name would be empty, change it into a ".". */
692unsigned int clean_fname(char *name, BOOL collapse_dot_dot)
693{
694 char *limit = name - 1, *t = name, *f = name;
695 int anchored;
696
697 if (!name)
698 return 0;
699
700 if ((anchored = *f == '/') != 0)
701 *t++ = *f++;
702 while (*f) {
703 /* discard extra slashes */
704 if (*f == '/') {
705 f++;
706 continue;
707 }
708 if (*f == '.') {
709 /* discard "." dirs (but NOT a trailing '.'!) */
710 if (f[1] == '/') {
711 f += 2;
712 continue;
713 }
714 /* collapse ".." dirs */
715 if (collapse_dot_dot
716 && f[1] == '.' && (f[2] == '/' || !f[2])) {
717 char *s = t - 1;
718 if (s == name && anchored) {
719 f += 2;
720 continue;
721 }
722 while (s > limit && *--s != '/') {}
723 if (s != t - 1 && (s < name || *s == '/')) {
724 t = s + 1;
725 f += 2;
726 continue;
727 }
728 limit = t + 2;
729 }
730 }
731 while (*f && (*t++ = *f++) != '/') {}
732 }
733
734 if (t > name+anchored && t[-1] == '/')
735 t--;
736 if (t == name)
737 *t++ = '.';
738 *t = '\0';
739
740 return t - name;
741}
742
743/* Make path appear as if a chroot had occurred. This handles a leading
744 * "/" (either removing it or expanding it) and any leading or embedded
745 * ".." components that attempt to escape past the module's top dir.
746 *
747 * If dest is NULL, a buffer is allocated to hold the result. It is legal
748 * to call with the dest and the path (p) pointing to the same buffer, but
749 * rootdir will be ignored to avoid expansion of the string.
750 *
751 * The rootdir string contains a value to use in place of a leading slash.
752 * Specify NULL to get the default of "module_dir".
753 *
754 * The depth var is a count of how many '..'s to allow at the start of the
755 * path. If symlink is set, combine its value with the "p" value to get
756 * the target path, and **return NULL if any '..'s try to escape**.
757 *
758 * We also clean the path in a manner similar to clean_fname() but with a
759 * few differences:
760 *
761 * Turns multiple adjacent slashes into a single slash, gets rid of "." dir
762 * elements (INCLUDING a trailing dot dir), PRESERVES a trailing slash, and
763 * ALWAYS collapses ".." elements (except for those at the start of the
764 * string up to "depth" deep). If the resulting name would be empty,
765 * change it into a ".". */
766char *sanitize_path(char *dest, const char *p, const char *rootdir, int depth,
767 const char *symlink)
768{
769 char *start, *sanp, *save_dest = dest;
770 int rlen = 0, leave_one_dotdir = relative_paths;
771
772 if (symlink && *symlink == '/') {
773 p = symlink;
774 symlink = "";
775 }
776
777 if (dest != p) {
778 int plen = strlen(p);
779 if (*p == '/') {
780 if (!rootdir)
781 rootdir = module_dir;
782 rlen = strlen(rootdir);
783 depth = 0;
784 p++;
785 }
786 if (dest) {
787 if (rlen + plen + 1 >= MAXPATHLEN)
788 return NULL;
789 } else if (!(dest = new_array(char, rlen + plen + 1)))
790 out_of_memory("sanitize_path");
791 if (rlen) {
792 memcpy(dest, rootdir, rlen);
793 if (rlen > 1)
794 dest[rlen++] = '/';
795 }
796 }
797
798 start = sanp = dest + rlen;
799 while (1) {
800 if (*p == '\0') {
801 if (!symlink || !*symlink)
802 break;
803 while (sanp != start && sanp[-1] != '/') {
804 /* strip last element */
805 sanp--;
806 }
807 /* Append a relative symlink */
808 p = symlink;
809 symlink = "";
810 }
811 /* discard leading or extra slashes */
812 if (*p == '/') {
813 p++;
814 continue;
815 }
816 /* this loop iterates once per filename component in p.
817 * both p (and sanp if the original had a slash) should
818 * always be left pointing after a slash
819 */
820 if (*p == '.' && (p[1] == '/' || p[1] == '\0')) {
821 if (leave_one_dotdir && p[1])
822 leave_one_dotdir = 0;
823 else {
824 /* skip "." component */
825 p++;
826 continue;
827 }
828 }
829 if (*p == '.' && p[1] == '.' && (p[2] == '/' || p[2] == '\0')) {
830 /* ".." component followed by slash or end */
831 if (depth <= 0 || sanp != start) {
832 if (symlink && sanp == start) {
833 if (!save_dest)
834 free(dest);
835 return NULL;
836 }
837 p += 2;
838 if (sanp != start) {
839 /* back up sanp one level */
840 --sanp; /* now pointing at slash */
841 while (sanp > start && sanp[-1] != '/') {
842 /* skip back up to slash */
843 sanp--;
844 }
845 }
846 continue;
847 }
848 /* allow depth levels of .. at the beginning */
849 depth--;
850 /* move the virtual beginning to leave the .. alone */
851 start = sanp + 3;
852 }
853 /* copy one component through next slash */
854 while (*p && (*sanp++ = *p++) != '/') {}
855 }
856 if (sanp == dest) {
857 /* ended up with nothing, so put in "." component */
858 *sanp++ = '.';
859 }
860 *sanp = '\0';
861
862 return dest;
863}
864
865/* Like chdir(), but it keeps track of the current directory (in the
866 * global "curr_dir"), and ensures that the path size doesn't overflow.
867 * Also cleans the path using the clean_fname() function. */
868int push_dir(const char *dir, int set_path_only)
869{
870 static int initialised;
871 unsigned int len;
872
873 if (!initialised) {
874 initialised = 1;
875 getcwd(curr_dir, sizeof curr_dir - 1);
876 curr_dir_len = strlen(curr_dir);
877 }
878
879 if (!dir) /* this call was probably just to initialize */
880 return 0;
881
882 len = strlen(dir);
883 if (len == 1 && *dir == '.')
884 return 1;
885
886 if ((*dir == '/' ? len : curr_dir_len + 1 + len) >= sizeof curr_dir) {
887 errno = ENAMETOOLONG;
888 return 0;
889 }
890
891 if (!set_path_only && chdir(dir))
892 return 0;
893
894 if (*dir == '/') {
895 memcpy(curr_dir, dir, len + 1);
896 curr_dir_len = len;
897 } else {
898 curr_dir[curr_dir_len++] = '/';
899 memcpy(curr_dir + curr_dir_len, dir, len + 1);
900 curr_dir_len += len;
901 }
902
903 curr_dir_len = clean_fname(curr_dir, 1);
904 if (sanitize_paths) {
905 if (module_dirlen > curr_dir_len)
906 module_dirlen = curr_dir_len;
907 curr_dir_depth = count_dir_elements(curr_dir + module_dirlen);
908 }
909
910 return 1;
911}
912
913/**
914 * Reverse a push_dir() call. You must pass in an absolute path
915 * that was copied from a prior value of "curr_dir".
916 **/
917int pop_dir(const char *dir)
918{
919 if (chdir(dir))
920 return 0;
921
922 curr_dir_len = strlcpy(curr_dir, dir, sizeof curr_dir);
923 if (curr_dir_len >= sizeof curr_dir)
924 curr_dir_len = sizeof curr_dir - 1;
925 if (sanitize_paths)
926 curr_dir_depth = count_dir_elements(curr_dir + module_dirlen);
927
928 return 1;
929}
930
931/**
932 * Return a quoted string with the full pathname of the indicated filename.
933 * The string " (in MODNAME)" may also be appended. The returned pointer
934 * remains valid until the next time full_fname() is called.
935 **/
936char *full_fname(const char *fn)
937{
938 static char *result = NULL;
939 char *m1, *m2, *m3;
940 char *p1, *p2;
941
942 if (result)
943 free(result);
944
945 if (*fn == '/')
946 p1 = p2 = "";
947 else {
948 p1 = curr_dir + module_dirlen;
949 for (p2 = p1; *p2 == '/'; p2++) {}
950 if (*p2)
951 p2 = "/";
952 }
953 if (module_id >= 0) {
954 m1 = " (in ";
955 m2 = lp_name(module_id);
956 m3 = ")";
957 } else
958 m1 = m2 = m3 = "";
959
960 if (asprintf(&result, "\"%s%s%s\"%s%s%s", p1, p2, fn, m1, m2, m3) <= 0)
961 out_of_memory("full_fname");
962
963 return result;
964}
965
966static char partial_fname[MAXPATHLEN];
967
968char *partial_dir_fname(const char *fname)
969{
970 char *t = partial_fname;
971 int sz = sizeof partial_fname;
972 const char *fn;
973
974 if ((fn = strrchr(fname, '/')) != NULL) {
975 fn++;
976 if (*partial_dir != '/') {
977 int len = fn - fname;
978 strncpy(t, fname, len); /* safe */
979 t += len;
980 sz -= len;
981 }
982 } else
983 fn = fname;
984 if ((int)pathjoin(t, sz, partial_dir, fn) >= sz)
985 return NULL;
986 if (server_filter_list.head) {
987 t = strrchr(partial_fname, '/');
988 *t = '\0';
989 if (check_filter(&server_filter_list, partial_fname, 1) < 0)
990 return NULL;
991 *t = '/';
992 if (check_filter(&server_filter_list, partial_fname, 0) < 0)
993 return NULL;
994 }
995
996 return partial_fname;
997}
998
999/* If no --partial-dir option was specified, we don't need to do anything
1000 * (the partial-dir is essentially '.'), so just return success. */
1001int handle_partial_dir(const char *fname, int create)
1002{
1003 char *fn, *dir;
1004
1005 if (fname != partial_fname)
1006 return 1;
1007 if (!create && *partial_dir == '/')
1008 return 1;
1009 if (!(fn = strrchr(partial_fname, '/')))
1010 return 1;
1011
1012 *fn = '\0';
1013 dir = partial_fname;
1014 if (create) {
1015 STRUCT_STAT st;
1016 int statret = do_lstat(dir, &st);
1017 if (statret == 0 && !S_ISDIR(st.st_mode)) {
1018 if (do_unlink(dir) < 0)
1019 return 0;
1020 statret = -1;
1021 }
1022 if (statret < 0 && do_mkdir(dir, 0700) < 0)
1023 return 0;
1024 } else
1025 do_rmdir(dir);
1026 *fn = '/';
1027
1028 return 1;
1029}
1030
1031/**
1032 * Determine if a symlink points outside the current directory tree.
1033 * This is considered "unsafe" because e.g. when mirroring somebody
1034 * else's machine it might allow them to establish a symlink to
1035 * /etc/passwd, and then read it through a web server.
1036 *
1037 * Null symlinks and absolute symlinks are always unsafe.
1038 *
1039 * Basically here we are concerned with symlinks whose target contains
1040 * "..", because this might cause us to walk back up out of the
1041 * transferred directory. We are not allowed to go back up and
1042 * reenter.
1043 *
1044 * @param dest Target of the symlink in question.
1045 *
1046 * @param src Top source directory currently applicable. Basically this
1047 * is the first parameter to rsync in a simple invocation, but it's
1048 * modified by flist.c in slightly complex ways.
1049 *
1050 * @retval True if unsafe
1051 * @retval False is unsafe
1052 *
1053 * @sa t_unsafe.c
1054 **/
1055int unsafe_symlink(const char *dest, const char *src)
1056{
1057 const char *name, *slash;
1058 int depth = 0;
1059
1060 /* all absolute and null symlinks are unsafe */
1061 if (!dest || !*dest || *dest == '/')
1062 return 1;
1063
1064 /* find out what our safety margin is */
1065 for (name = src; (slash = strchr(name, '/')) != 0; name = slash+1) {
1066 if (strncmp(name, "../", 3) == 0) {
1067 depth = 0;
1068 } else if (strncmp(name, "./", 2) == 0) {
1069 /* nothing */
1070 } else {
1071 depth++;
1072 }
1073 }
1074 if (strcmp(name, "..") == 0)
1075 depth = 0;
1076
1077 for (name = dest; (slash = strchr(name, '/')) != 0; name = slash+1) {
1078 if (strncmp(name, "../", 3) == 0) {
1079 /* if at any point we go outside the current directory
1080 then stop - it is unsafe */
1081 if (--depth < 0)
1082 return 1;
1083 } else if (strncmp(name, "./", 2) == 0) {
1084 /* nothing */
1085 } else {
1086 depth++;
1087 }
1088 }
1089 if (strcmp(name, "..") == 0)
1090 depth--;
1091
1092 return (depth < 0);
1093}
1094
1095/* Return the int64 number as a string. If the --human-readable option was
1096 * specified, we may output the number in K, M, or G units. We can return
1097 * up to 4 buffers at a time. */
1098char *human_num(int64 num)
1099{
1100 static char bufs[4][128]; /* more than enough room */
1101 static unsigned int n;
1102 char *s;
1103
1104 n = (n + 1) % (sizeof bufs / sizeof bufs[0]);
1105
1106 if (human_readable) {
1107 char units = '\0';
1108 int mult = human_readable == 1 ? 1000 : 1024;
1109 double dnum = 0;
1110 if (num > mult*mult*mult) {
1111 dnum = (double)num / (mult*mult*mult);
1112 units = 'G';
1113 } else if (num > mult*mult) {
1114 dnum = (double)num / (mult*mult);
1115 units = 'M';
1116 } else if (num > mult) {
1117 dnum = (double)num / mult;
1118 units = 'K';
1119 }
1120 if (units) {
1121 snprintf(bufs[n], sizeof bufs[0], "%.2f%c", dnum, units);
1122 return bufs[n];
1123 }
1124 }
1125
1126 s = bufs[n] + sizeof bufs[0] - 1;
1127 *s = '\0';
1128
1129 if (!num)
1130 *--s = '0';
1131 while (num) {
1132 *--s = (char)(num % 10) + '0';
1133 num /= 10;
1134 }
1135 return s;
1136}
1137
1138/* Return the double number as a string. If the --human-readable option was
1139 * specified, we may output the number in K, M, or G units. We use a buffer
1140 * from human_num() to return our result. */
1141char *human_dnum(double dnum, int decimal_digits)
1142{
1143 char *buf = human_num(dnum);
1144 int len = strlen(buf);
1145 if (isDigit(buf + len - 1)) {
1146 /* There's extra room in buf prior to the start of the num. */
1147 buf -= decimal_digits + 1;
1148 snprintf(buf, len + decimal_digits + 2, "%.*f", decimal_digits, dnum);
1149 }
1150 return buf;
1151}
1152
1153/* Return the date and time as a string. Some callers tweak returned buf. */
1154char *timestring(time_t t)
1155{
1156 static char TimeBuf[200];
1157 struct tm *tm = localtime(&t);
1158 char *p;
1159
1160#ifdef HAVE_STRFTIME
1161 strftime(TimeBuf, sizeof TimeBuf - 1, "%Y/%m/%d %H:%M:%S", tm);
1162#else
1163 strlcpy(TimeBuf, asctime(tm), sizeof TimeBuf);
1164#endif
1165
1166 if ((p = strchr(TimeBuf, '\n')) != NULL)
1167 *p = '\0';
1168
1169 return TimeBuf;
1170}
1171
1172/**
1173 * Sleep for a specified number of milliseconds.
1174 *
1175 * Always returns TRUE. (In the future it might return FALSE if
1176 * interrupted.)
1177 **/
1178int msleep(int t)
1179{
1180 int tdiff = 0;
1181 struct timeval tval, t1, t2;
1182
1183 gettimeofday(&t1, NULL);
1184
1185 while (tdiff < t) {
1186 tval.tv_sec = (t-tdiff)/1000;
1187 tval.tv_usec = 1000*((t-tdiff)%1000);
1188
1189 errno = 0;
1190 select(0,NULL,NULL, NULL, &tval);
1191
1192 gettimeofday(&t2, NULL);
1193 tdiff = (t2.tv_sec - t1.tv_sec)*1000 +
1194 (t2.tv_usec - t1.tv_usec)/1000;
1195 }
1196
1197 return True;
1198}
1199
1200/* Determine if two time_t values are equivalent (either exact, or in
1201 * the modification timestamp window established by --modify-window).
1202 *
1203 * @retval 0 if the times should be treated as the same
1204 *
1205 * @retval +1 if the first is later
1206 *
1207 * @retval -1 if the 2nd is later
1208 **/
1209int cmp_time(time_t file1, time_t file2)
1210{
1211 if (file2 > file1) {
1212 if (file2 - file1 <= modify_window)
1213 return 0;
1214 return -1;
1215 }
1216 if (file1 - file2 <= modify_window)
1217 return 0;
1218 return 1;
1219}
1220
1221
1222#ifdef __INSURE__XX
1223#include <dlfcn.h>
1224
1225/**
1226 This routine is a trick to immediately catch errors when debugging
1227 with insure. A xterm with a gdb is popped up when insure catches
1228 a error. It is Linux specific.
1229**/
1230int _Insure_trap_error(int a1, int a2, int a3, int a4, int a5, int a6)
1231{
1232 static int (*fn)();
1233 int ret;
1234 char *cmd;
1235
1236 asprintf(&cmd, "/usr/X11R6/bin/xterm -display :0 -T Panic -n Panic -e /bin/sh -c 'cat /tmp/ierrs.*.%d ; gdb /proc/%d/exe %d'",
1237 getpid(), getpid(), getpid());
1238
1239 if (!fn) {
1240 static void *h;
1241 h = dlopen("/usr/local/parasoft/insure++lite/lib.linux2/libinsure.so", RTLD_LAZY);
1242 fn = dlsym(h, "_Insure_trap_error");
1243 }
1244
1245 ret = fn(a1, a2, a3, a4, a5, a6);
1246
1247 system(cmd);
1248
1249 free(cmd);
1250
1251 return ret;
1252}
1253#endif
1254
1255#define MALLOC_MAX 0x40000000
1256
1257void *_new_array(unsigned int size, unsigned long num)
1258{
1259 if (num >= MALLOC_MAX/size)
1260 return NULL;
1261 return malloc(size * num);
1262}
1263
1264void *_realloc_array(void *ptr, unsigned int size, unsigned long num)
1265{
1266 if (num >= MALLOC_MAX/size)
1267 return NULL;
1268 if (!ptr)
1269 return malloc(size * num);
1270 return realloc(ptr, size * num);
1271}
1272
1273/* Take a filename and filename length and return the most significant
1274 * filename suffix we can find. This ignores suffixes such as "~",
1275 * ".bak", ".orig", ".~1~", etc. */
1276const char *find_filename_suffix(const char *fn, int fn_len, int *len_ptr)
1277{
1278 const char *suf, *s;
1279 BOOL had_tilde;
1280 int s_len;
1281
1282 /* One or more dots at the start aren't a suffix. */
1283 while (fn_len && *fn == '.') fn++, fn_len--;
1284
1285 /* Ignore the ~ in a "foo~" filename. */
1286 if (fn_len > 1 && fn[fn_len-1] == '~')
1287 fn_len--, had_tilde = True;
1288 else
1289 had_tilde = False;
1290
1291 /* Assume we don't find an suffix. */
1292 suf = "";
1293 *len_ptr = 0;
1294
1295 /* Find the last significant suffix. */
1296 for (s = fn + fn_len; fn_len > 1; ) {
1297 while (*--s != '.' && s != fn) {}
1298 if (s == fn)
1299 break;
1300 s_len = fn_len - (s - fn);
1301 fn_len = s - fn;
1302 if (s_len == 4) {
1303 if (strcmp(s+1, "bak") == 0
1304 || strcmp(s+1, "old") == 0)
1305 continue;
1306 } else if (s_len == 5) {
1307 if (strcmp(s+1, "orig") == 0)
1308 continue;
1309 } else if (s_len > 2 && had_tilde
1310 && s[1] == '~' && isDigit(s + 2))
1311 continue;
1312 *len_ptr = s_len;
1313 suf = s;
1314 if (s_len == 1)
1315 break;
1316 /* Determine if the suffix is all digits. */
1317 for (s++, s_len--; s_len > 0; s++, s_len--) {
1318 if (!isDigit(s))
1319 return suf;
1320 }
1321 /* An all-digit suffix may not be that signficant. */
1322 s = suf;
1323 }
1324
1325 return suf;
1326}
1327
1328/* This is an implementation of the Levenshtein distance algorithm. It
1329 * was implemented to avoid needing a two-dimensional matrix (to save
1330 * memory). It was also tweaked to try to factor in the ASCII distance
1331 * between changed characters as a minor distance quantity. The normal
1332 * Levenshtein units of distance (each signifying a single change between
1333 * the two strings) are defined as a "UNIT". */
1334
1335#define UNIT (1 << 16)
1336
1337uint32 fuzzy_distance(const char *s1, int len1, const char *s2, int len2)
1338{
1339 uint32 a[MAXPATHLEN], diag, above, left, diag_inc, above_inc, left_inc;
1340 int32 cost;
1341 int i1, i2;
1342
1343 if (!len1 || !len2) {
1344 if (!len1) {
1345 s1 = s2;
1346 len1 = len2;
1347 }
1348 for (i1 = 0, cost = 0; i1 < len1; i1++)
1349 cost += s1[i1];
1350 return (int32)len1 * UNIT + cost;
1351 }
1352
1353 for (i2 = 0; i2 < len2; i2++)
1354 a[i2] = (i2+1) * UNIT;
1355
1356 for (i1 = 0; i1 < len1; i1++) {
1357 diag = i1 * UNIT;
1358 above = (i1+1) * UNIT;
1359 for (i2 = 0; i2 < len2; i2++) {
1360 left = a[i2];
1361 if ((cost = *((uchar*)s1+i1) - *((uchar*)s2+i2)) != 0) {
1362 if (cost < 0)
1363 cost = UNIT - cost;
1364 else
1365 cost = UNIT + cost;
1366 }
1367 diag_inc = diag + cost;
1368 left_inc = left + UNIT + *((uchar*)s1+i1);
1369 above_inc = above + UNIT + *((uchar*)s2+i2);
1370 a[i2] = above = left < above
1371 ? (left_inc < diag_inc ? left_inc : diag_inc)
1372 : (above_inc < diag_inc ? above_inc : diag_inc);
1373 diag = left;
1374 }
1375 }
1376
1377 return a[len2-1];
1378}
1379
1380#define BB_SLOT_SIZE (16*1024) /* Desired size in bytes */
1381#define BB_PER_SLOT_BITS (BB_SLOT_SIZE * 8) /* Number of bits per slot */
1382#define BB_PER_SLOT_INTS (BB_SLOT_SIZE / 4) /* Number of int32s per slot */
1383
1384struct bitbag {
1385 uint32 **bits;
1386 int slot_cnt;
1387};
1388
1389struct bitbag *bitbag_create(int max_ndx)
1390{
1391 struct bitbag *bb = new(struct bitbag);
1392 bb->slot_cnt = (max_ndx + BB_PER_SLOT_BITS - 1) / BB_PER_SLOT_BITS;
1393
1394 if (!(bb->bits = (uint32**)calloc(bb->slot_cnt, sizeof (uint32*))))
1395 out_of_memory("bitbag_create");
1396
1397 return bb;
1398}
1399
1400void bitbag_set_bit(struct bitbag *bb, int ndx)
1401{
1402 int slot = ndx / BB_PER_SLOT_BITS;
1403 ndx %= BB_PER_SLOT_BITS;
1404
1405 if (!bb->bits[slot]) {
1406 if (!(bb->bits[slot] = (uint32*)calloc(BB_PER_SLOT_INTS, 4)))
1407 out_of_memory("bitbag_set_bit");
1408 }
1409
1410 bb->bits[slot][ndx/32] |= 1u << (ndx % 32);
1411}
1412
1413#if 0 /* not needed yet */
1414void bitbag_clear_bit(struct bitbag *bb, int ndx)
1415{
1416 int slot = ndx / BB_PER_SLOT_BITS;
1417 ndx %= BB_PER_SLOT_BITS;
1418
1419 if (!bb->bits[slot])
1420 return;
1421
1422 bb->bits[slot][ndx/32] &= ~(1u << (ndx % 32));
1423}
1424
1425int bitbag_check_bit(struct bitbag *bb, int ndx)
1426{
1427 int slot = ndx / BB_PER_SLOT_BITS;
1428 ndx %= BB_PER_SLOT_BITS;
1429
1430 if (!bb->bits[slot])
1431 return 0;
1432
1433 return bb->bits[slot][ndx/32] & (1u << (ndx % 32)) ? 1 : 0;
1434}
1435#endif
1436
1437/* Call this with -1 to start checking from 0. Returns -1 at the end. */
1438int bitbag_next_bit(struct bitbag *bb, int after)
1439{
1440 uint32 bits, mask;
1441 int i, ndx = after + 1;
1442 int slot = ndx / BB_PER_SLOT_BITS;
1443 ndx %= BB_PER_SLOT_BITS;
1444
1445 mask = (1u << (ndx % 32)) - 1;
1446 for (i = ndx / 32; slot < bb->slot_cnt; slot++, i = mask = 0) {
1447 if (!bb->bits[slot])
1448 continue;
1449 for ( ; i < BB_PER_SLOT_INTS; i++, mask = 0) {
1450 if (!(bits = bb->bits[slot][i] & ~mask))
1451 continue;
1452 /* The xor magic figures out the lowest enabled bit in
1453 * bits, and the switch quickly computes log2(bit). */
1454 switch (bits ^ (bits & (bits-1))) {
1455#define LOG2(n) case 1u << n: return slot*BB_PER_SLOT_BITS + i*32 + n
1456 LOG2(0); LOG2(1); LOG2(2); LOG2(3);
1457 LOG2(4); LOG2(5); LOG2(6); LOG2(7);
1458 LOG2(8); LOG2(9); LOG2(10); LOG2(11);
1459 LOG2(12); LOG2(13); LOG2(14); LOG2(15);
1460 LOG2(16); LOG2(17); LOG2(18); LOG2(19);
1461 LOG2(20); LOG2(21); LOG2(22); LOG2(23);
1462 LOG2(24); LOG2(25); LOG2(26); LOG2(27);
1463 LOG2(28); LOG2(29); LOG2(30); LOG2(31);
1464 }
1465 return -1; /* impossible... */
1466 }
1467 }
1468
1469 return -1;
1470}
1471
1472void *expand_item_list(item_list *lp, size_t item_size,
1473 const char *desc, int incr)
1474{
1475 /* First time through, 0 <= 0, so list is expanded. */
1476 if (lp->malloced <= lp->count) {
1477 void *new_ptr;
1478 size_t new_size = lp->malloced;
1479 if (incr < 0)
1480 new_size -= incr; /* increase slowly */
1481 else if (new_size < (size_t)incr)
1482 new_size += incr;
1483 else
1484 new_size *= 2;
1485 new_ptr = realloc_array(lp->items, char, new_size * item_size);
1486 if (verbose >= 4) {
1487 rprintf(FINFO, "[%s] expand %s to %.0f bytes, did%s move\n",
1488 who_am_i(), desc, (double)new_size * item_size,
1489 new_ptr == lp->items ? " not" : "");
1490 }
1491 if (!new_ptr)
1492 out_of_memory("expand_item_list");
1493
1494 lp->items = new_ptr;
1495 lp->malloced = new_size;
1496 }
1497 return (char*)lp->items + (lp->count++ * item_size);
1498}