Fixed a bug in the truncating of daemon-excluded paths.
[rsync.git] / util.c
blob434c74c547a92f13f88f731d45d7c55dc851ee58
1 /*
2 * Utility routines used in rsync.
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) 2003-2008 Wayne Davison
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 3 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 along
20 * with this program; if not, visit the http://fsf.org website.
23 #include "rsync.h"
24 #include "ifuncs.h"
26 extern int verbose;
27 extern int dry_run;
28 extern int module_id;
29 extern int modify_window;
30 extern int relative_paths;
31 extern int human_readable;
32 extern char *module_dir;
33 extern unsigned int module_dirlen;
34 extern mode_t orig_umask;
35 extern char *partial_dir;
36 extern struct filter_list_struct server_filter_list;
38 int sanitize_paths = 0;
40 char curr_dir[MAXPATHLEN];
41 unsigned int curr_dir_len;
42 int curr_dir_depth; /* This is only set for a sanitizing daemon. */
44 /* Set a fd into nonblocking mode. */
45 void set_nonblocking(int fd)
47 int val;
49 if ((val = fcntl(fd, F_GETFL)) == -1)
50 return;
51 if (!(val & NONBLOCK_FLAG)) {
52 val |= NONBLOCK_FLAG;
53 fcntl(fd, F_SETFL, val);
57 /* Set a fd into blocking mode. */
58 void set_blocking(int fd)
60 int val;
62 if ((val = fcntl(fd, F_GETFL)) == -1)
63 return;
64 if (val & NONBLOCK_FLAG) {
65 val &= ~NONBLOCK_FLAG;
66 fcntl(fd, F_SETFL, val);
70 /**
71 * Create a file descriptor pair - like pipe() but use socketpair if
72 * possible (because of blocking issues on pipes).
74 * Always set non-blocking.
76 int fd_pair(int fd[2])
78 int ret;
80 #ifdef HAVE_SOCKETPAIR
81 ret = socketpair(AF_UNIX, SOCK_STREAM, 0, fd);
82 #else
83 ret = pipe(fd);
84 #endif
86 if (ret == 0) {
87 set_nonblocking(fd[0]);
88 set_nonblocking(fd[1]);
91 return ret;
94 void print_child_argv(const char *prefix, char **cmd)
96 rprintf(FCLIENT, "%s ", prefix);
97 for (; *cmd; cmd++) {
98 /* Look for characters that ought to be quoted. This
99 * is not a great quoting algorithm, but it's
100 * sufficient for a log message. */
101 if (strspn(*cmd, "abcdefghijklmnopqrstuvwxyz"
102 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
103 "0123456789"
104 ",.-_=+@/") != strlen(*cmd)) {
105 rprintf(FCLIENT, "\"%s\" ", *cmd);
106 } else {
107 rprintf(FCLIENT, "%s ", *cmd);
110 rprintf(FCLIENT, "\n");
113 NORETURN void out_of_memory(const char *str)
115 rprintf(FERROR, "ERROR: out of memory in %s [%s]\n", str, who_am_i());
116 exit_cleanup(RERR_MALLOC);
119 NORETURN void overflow_exit(const char *str)
121 rprintf(FERROR, "ERROR: buffer overflow in %s [%s]\n", str, who_am_i());
122 exit_cleanup(RERR_MALLOC);
125 int set_modtime(const char *fname, time_t modtime, mode_t mode)
127 #if !defined HAVE_LUTIMES || !defined HAVE_UTIMES
128 if (S_ISLNK(mode))
129 return 1;
130 #endif
132 if (verbose > 2) {
133 rprintf(FINFO, "set modtime of %s to (%ld) %s",
134 fname, (long)modtime,
135 asctime(localtime(&modtime)));
138 if (dry_run)
139 return 0;
142 #ifdef HAVE_UTIMES
143 struct timeval t[2];
144 t[0].tv_sec = time(NULL);
145 t[0].tv_usec = 0;
146 t[1].tv_sec = modtime;
147 t[1].tv_usec = 0;
148 # ifdef HAVE_LUTIMES
149 if (S_ISLNK(mode)) {
150 if (lutimes(fname, t) < 0)
151 return errno == ENOSYS ? 1 : -1;
152 return 0;
154 # endif
155 return utimes(fname, t);
156 #elif defined HAVE_STRUCT_UTIMBUF
157 struct utimbuf tbuf;
158 tbuf.actime = time(NULL);
159 tbuf.modtime = modtime;
160 return utime(fname,&tbuf);
161 #elif defined HAVE_UTIME
162 time_t t[2];
163 t[0] = time(NULL);
164 t[1] = modtime;
165 return utime(fname,t);
166 #else
167 #error No file-time-modification routine found!
168 #endif
172 /* This creates a new directory with default permissions. Since there
173 * might be some directory-default permissions affecting this, we can't
174 * force the permissions directly using the original umask and mkdir(). */
175 int mkdir_defmode(char *fname)
177 int ret;
179 umask(orig_umask);
180 ret = do_mkdir(fname, ACCESSPERMS);
181 umask(0);
183 return ret;
186 /* Create any necessary directories in fname. Any missing directories are
187 * created with default permissions. */
188 int create_directory_path(char *fname)
190 char *p;
191 int ret = 0;
193 while (*fname == '/')
194 fname++;
195 while (strncmp(fname, "./", 2) == 0)
196 fname += 2;
198 umask(orig_umask);
199 p = fname;
200 while ((p = strchr(p,'/')) != NULL) {
201 *p = '\0';
202 if (do_mkdir(fname, ACCESSPERMS) < 0 && errno != EEXIST)
203 ret = -1;
204 *p++ = '/';
206 umask(0);
208 return ret;
212 * Write @p len bytes at @p ptr to descriptor @p desc, retrying if
213 * interrupted.
215 * @retval len upon success
217 * @retval <0 write's (negative) error code
219 * Derived from GNU C's cccp.c.
221 int full_write(int desc, const char *ptr, size_t len)
223 int total_written;
225 total_written = 0;
226 while (len > 0) {
227 int written = write(desc, ptr, len);
228 if (written < 0) {
229 if (errno == EINTR)
230 continue;
231 return written;
233 total_written += written;
234 ptr += written;
235 len -= written;
237 return total_written;
241 * Read @p len bytes at @p ptr from descriptor @p desc, retrying if
242 * interrupted.
244 * @retval >0 the actual number of bytes read
246 * @retval 0 for EOF
248 * @retval <0 for an error.
250 * Derived from GNU C's cccp.c. */
251 static int safe_read(int desc, char *ptr, size_t len)
253 int n_chars;
255 if (len == 0)
256 return len;
258 do {
259 n_chars = read(desc, ptr, len);
260 } while (n_chars < 0 && errno == EINTR);
262 return n_chars;
265 /* Copy a file. If ofd < 0, copy_file unlinks and opens the "dest" file.
266 * Otherwise, it just writes to and closes the provided file descriptor.
268 * This is used in conjunction with the --temp-dir, --backup, and
269 * --copy-dest options. */
270 int copy_file(const char *source, const char *dest, int ofd,
271 mode_t mode, int create_bak_dir)
273 int ifd;
274 char buf[1024 * 8];
275 int len; /* Number of bytes read into `buf'. */
277 if ((ifd = do_open(source, O_RDONLY, 0)) < 0) {
278 rsyserr(FERROR_XFER, errno, "open %s", full_fname(source));
279 return -1;
282 if (ofd < 0) {
283 if (robust_unlink(dest) && errno != ENOENT) {
284 rsyserr(FERROR_XFER, errno, "unlink %s", full_fname(dest));
285 return -1;
288 if ((ofd = do_open(dest, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, mode)) < 0
289 && (!create_bak_dir || errno != ENOENT || make_bak_dir(dest) < 0
290 || (ofd = do_open(dest, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, mode)) < 0)) {
291 rsyserr(FERROR_XFER, errno, "open %s", full_fname(dest));
292 close(ifd);
293 return -1;
297 while ((len = safe_read(ifd, buf, sizeof buf)) > 0) {
298 if (full_write(ofd, buf, len) < 0) {
299 rsyserr(FERROR_XFER, errno, "write %s", full_fname(dest));
300 close(ifd);
301 close(ofd);
302 return -1;
306 if (len < 0) {
307 rsyserr(FERROR_XFER, errno, "read %s", full_fname(source));
308 close(ifd);
309 close(ofd);
310 return -1;
313 if (close(ifd) < 0) {
314 rsyserr(FWARNING, errno, "close failed on %s",
315 full_fname(source));
318 if (close(ofd) < 0) {
319 rsyserr(FERROR_XFER, errno, "close failed on %s",
320 full_fname(dest));
321 return -1;
324 return 0;
327 /* MAX_RENAMES should be 10**MAX_RENAMES_DIGITS */
328 #define MAX_RENAMES_DIGITS 3
329 #define MAX_RENAMES 1000
332 * Robust unlink: some OS'es (HPUX) refuse to unlink busy files, so
333 * rename to <path>/.rsyncNNN instead.
335 * Note that successive rsync runs will shuffle the filenames around a
336 * bit as long as the file is still busy; this is because this function
337 * does not know if the unlink call is due to a new file coming in, or
338 * --delete trying to remove old .rsyncNNN files, hence it renames it
339 * each time.
341 int robust_unlink(const char *fname)
343 #ifndef ETXTBSY
344 return do_unlink(fname);
345 #else
346 static int counter = 1;
347 int rc, pos, start;
348 char path[MAXPATHLEN];
350 rc = do_unlink(fname);
351 if (rc == 0 || errno != ETXTBSY)
352 return rc;
354 if ((pos = strlcpy(path, fname, MAXPATHLEN)) >= MAXPATHLEN)
355 pos = MAXPATHLEN - 1;
357 while (pos > 0 && path[pos-1] != '/')
358 pos--;
359 pos += strlcpy(path+pos, ".rsync", MAXPATHLEN-pos);
361 if (pos > (MAXPATHLEN-MAX_RENAMES_DIGITS-1)) {
362 errno = ETXTBSY;
363 return -1;
366 /* start where the last one left off to reduce chance of clashes */
367 start = counter;
368 do {
369 snprintf(&path[pos], MAX_RENAMES_DIGITS+1, "%03d", counter);
370 if (++counter >= MAX_RENAMES)
371 counter = 1;
372 } while ((rc = access(path, 0)) == 0 && counter != start);
374 if (verbose > 0) {
375 rprintf(FWARNING, "renaming %s to %s because of text busy\n",
376 fname, path);
379 /* maybe we should return rename()'s exit status? Nah. */
380 if (do_rename(fname, path) != 0) {
381 errno = ETXTBSY;
382 return -1;
384 return 0;
385 #endif
388 /* Returns 0 on successful rename, 1 if we successfully copied the file
389 * across filesystems, -2 if copy_file() failed, and -1 on other errors.
390 * If partialptr is not NULL and we need to do a copy, copy the file into
391 * the active partial-dir instead of over the destination file. */
392 int robust_rename(const char *from, const char *to, const char *partialptr,
393 int mode)
395 int tries = 4;
397 while (tries--) {
398 if (do_rename(from, to) == 0)
399 return 0;
401 switch (errno) {
402 #ifdef ETXTBSY
403 case ETXTBSY:
404 if (robust_unlink(to) != 0)
405 return -1;
406 break;
407 #endif
408 case EXDEV:
409 if (partialptr) {
410 if (!handle_partial_dir(partialptr,PDIR_CREATE))
411 return -1;
412 to = partialptr;
414 if (copy_file(from, to, -1, mode, 0) != 0)
415 return -2;
416 do_unlink(from);
417 return 1;
418 default:
419 return -1;
422 return -1;
425 static pid_t all_pids[10];
426 static int num_pids;
428 /** Fork and record the pid of the child. **/
429 pid_t do_fork(void)
431 pid_t newpid = fork();
433 if (newpid != 0 && newpid != -1) {
434 all_pids[num_pids++] = newpid;
436 return newpid;
440 * Kill all children.
442 * @todo It would be kind of nice to make sure that they are actually
443 * all our children before we kill them, because their pids may have
444 * been recycled by some other process. Perhaps when we wait for a
445 * child, we should remove it from this array. Alternatively we could
446 * perhaps use process groups, but I think that would not work on
447 * ancient Unix versions that don't support them.
449 void kill_all(int sig)
451 int i;
453 for (i = 0; i < num_pids; i++) {
454 /* Let's just be a little careful where we
455 * point that gun, hey? See kill(2) for the
456 * magic caused by negative values. */
457 pid_t p = all_pids[i];
459 if (p == getpid())
460 continue;
461 if (p <= 0)
462 continue;
464 kill(p, sig);
468 /** Turn a user name into a uid */
469 int name_to_uid(const char *name, uid_t *uid_p)
471 struct passwd *pass;
472 if (!name || !*name)
473 return 0;
474 if (!(pass = getpwnam(name)))
475 return 0;
476 *uid_p = pass->pw_uid;
477 return 1;
480 /** Turn a group name into a gid */
481 int name_to_gid(const char *name, gid_t *gid_p)
483 struct group *grp;
484 if (!name || !*name)
485 return 0;
486 if (!(grp = getgrnam(name)))
487 return 0;
488 *gid_p = grp->gr_gid;
489 return 1;
492 /** Lock a byte range in a open file */
493 int lock_range(int fd, int offset, int len)
495 struct flock lock;
497 lock.l_type = F_WRLCK;
498 lock.l_whence = SEEK_SET;
499 lock.l_start = offset;
500 lock.l_len = len;
501 lock.l_pid = 0;
503 return fcntl(fd,F_SETLK,&lock) == 0;
506 static int filter_server_path(char *arg)
508 char *s;
510 if (server_filter_list.head) {
511 for (s = arg; (s = strchr(s, '/')) != NULL; ) {
512 *s = '\0';
513 if (check_filter(&server_filter_list, arg, 1) < 0) {
514 /* We must leave arg truncated! */
515 return 1;
517 *s++ = '/';
520 return 0;
523 void glob_expand(char *s, char ***argv_ptr, int *argc_ptr, int *maxargs_ptr)
525 char **argv = *argv_ptr;
526 int argc = *argc_ptr;
527 int maxargs = *maxargs_ptr;
528 int count, have_glob_results;
530 #if !defined HAVE_GLOB || !defined HAVE_GLOB_H
531 if (argc == maxargs) {
532 maxargs += MAX_ARGS;
533 if (!(argv = realloc_array(argv, char *, maxargs)))
534 out_of_memory("glob_expand");
535 *argv_ptr = argv;
536 *maxargs_ptr = maxargs;
538 if (!*s)
539 s = ".";
540 s = argv[argc++] = strdup(s);
541 filter_server_path(s);
542 #else
543 glob_t globbuf;
545 if (maxargs <= argc)
546 return;
547 if (!*s)
548 s = ".";
550 if (sanitize_paths)
551 s = sanitize_path(NULL, s, "", 0);
552 else
553 s = strdup(s);
554 if (!s)
555 out_of_memory("glob_expand");
557 memset(&globbuf, 0, sizeof globbuf);
558 glob(s, 0, NULL, &globbuf);
559 /* Note: we check the first match against the filter list,
560 * just in case the user specified a wildcard in the path. */
561 if ((count = globbuf.gl_pathc) > 0) {
562 if (filter_server_path(globbuf.gl_pathv[0])) {
563 int slashes = 0;
564 char *cp;
565 /* Truncate original arg at glob's truncation point. */
566 for (cp = globbuf.gl_pathv[0]; *cp; cp++) {
567 if (*cp == '/')
568 slashes++;
570 for (cp = s; *cp; cp++) {
571 if (*cp == '/') {
572 if (slashes-- <= 0) {
573 *cp = '\0';
574 break;
578 have_glob_results = 0;
579 count = 1;
580 } else
581 have_glob_results = 1;
582 } else {
583 /* This truncates "s" at a filtered element, if present. */
584 filter_server_path(s);
585 have_glob_results = 0;
586 count = 1;
588 if (count + argc > maxargs) {
589 maxargs += count + MAX_ARGS;
590 if (!(argv = realloc_array(argv, char *, maxargs)))
591 out_of_memory("glob_expand");
592 *argv_ptr = argv;
593 *maxargs_ptr = maxargs;
595 if (have_glob_results) {
596 int i;
597 free(s);
598 for (i = 0; i < count; i++) {
599 if (!(argv[argc++] = strdup(globbuf.gl_pathv[i])))
600 out_of_memory("glob_expand");
602 } else
603 argv[argc++] = s;
604 globfree(&globbuf);
605 #endif
606 *argc_ptr = argc;
609 /* This routine is only used in daemon mode. */
610 void glob_expand_module(char *base1, char *arg, char ***argv_ptr, int *argc_ptr, int *maxargs_ptr)
612 char *p, *s;
613 char *base = base1;
614 int base_len = strlen(base);
616 if (!arg || !*arg)
617 return;
619 if (strncmp(arg, base, base_len) == 0)
620 arg += base_len;
622 if (!(arg = strdup(arg)))
623 out_of_memory("glob_expand_module");
625 if (asprintf(&base," %s/", base1) <= 0)
626 out_of_memory("glob_expand_module");
627 base_len++;
629 for (s = arg; *s; s = p + base_len) {
630 if ((p = strstr(s, base)) != NULL)
631 *p = '\0'; /* split it at this point */
632 glob_expand(s, argv_ptr, argc_ptr, maxargs_ptr);
633 if (!p)
634 break;
637 free(arg);
638 free(base);
642 * Convert a string to lower case
644 void strlower(char *s)
646 while (*s) {
647 if (isUpper(s))
648 *s = toLower(s);
649 s++;
653 /* Join strings p1 & p2 into "dest" with a guaranteed '/' between them. (If
654 * p1 ends with a '/', no extra '/' is inserted.) Returns the length of both
655 * strings + 1 (if '/' was inserted), regardless of whether the null-terminated
656 * string fits into destsize. */
657 size_t pathjoin(char *dest, size_t destsize, const char *p1, const char *p2)
659 size_t len = strlcpy(dest, p1, destsize);
660 if (len < destsize - 1) {
661 if (!len || dest[len-1] != '/')
662 dest[len++] = '/';
663 if (len < destsize - 1)
664 len += strlcpy(dest + len, p2, destsize - len);
665 else {
666 dest[len] = '\0';
667 len += strlen(p2);
670 else
671 len += strlen(p2) + 1; /* Assume we'd insert a '/'. */
672 return len;
675 /* Join any number of strings together, putting them in "dest". The return
676 * value is the length of all the strings, regardless of whether the null-
677 * terminated whole fits in destsize. Your list of string pointers must end
678 * with a NULL to indicate the end of the list. */
679 size_t stringjoin(char *dest, size_t destsize, ...)
681 va_list ap;
682 size_t len, ret = 0;
683 const char *src;
685 va_start(ap, destsize);
686 while (1) {
687 if (!(src = va_arg(ap, const char *)))
688 break;
689 len = strlen(src);
690 ret += len;
691 if (destsize > 1) {
692 if (len >= destsize)
693 len = destsize - 1;
694 memcpy(dest, src, len);
695 destsize -= len;
696 dest += len;
699 *dest = '\0';
700 va_end(ap);
702 return ret;
705 int count_dir_elements(const char *p)
707 int cnt = 0, new_component = 1;
708 while (*p) {
709 if (*p++ == '/')
710 new_component = (*p != '.' || (p[1] != '/' && p[1] != '\0'));
711 else if (new_component) {
712 new_component = 0;
713 cnt++;
716 return cnt;
719 /* Turns multiple adjacent slashes into a single slash, drops interior "."
720 * elements, drops an intial "./" unless CFN_KEEP_LEADING_DOT_DIR is flagged,
721 * will even drop a trailing '.' after a '/' if CFN_DROP_TRAILING_DOT_DIR is
722 * flagged, removes a trailing slash (perhaps after removing the aforementioned
723 * dot) unless CFN_KEEP_TRAILING_SLASH is flagged, will even collapse ".."
724 * elements (except at the start of the string) if CFN_COLLAPSE_DOT_DOT_DIRS
725 * is flagged. If the resulting name would be empty, we return ".". */
726 unsigned int clean_fname(char *name, int flags)
728 char *limit = name - 1, *t = name, *f = name;
729 int anchored;
731 if (!name)
732 return 0;
734 if ((anchored = *f == '/') != 0)
735 *t++ = *f++;
736 else if (flags & CFN_KEEP_LEADING_DOT_DIR && *f == '.' && f[1] == '/') {
737 *t++ = *f++;
738 *t++ = *f++;
740 while (*f) {
741 /* discard extra slashes */
742 if (*f == '/') {
743 f++;
744 continue;
746 if (*f == '.') {
747 /* discard interior "." dirs */
748 if (f[1] == '/') {
749 f += 2;
750 continue;
752 if (f[1] == '\0' && flags & CFN_DROP_TRAILING_DOT_DIR)
753 break;
754 /* collapse ".." dirs */
755 if (flags & CFN_COLLAPSE_DOT_DOT_DIRS
756 && f[1] == '.' && (f[2] == '/' || !f[2])) {
757 char *s = t - 1;
758 if (s == name && anchored) {
759 f += 2;
760 continue;
762 while (s > limit && *--s != '/') {}
763 if (s != t - 1 && (s < name || *s == '/')) {
764 t = s + 1;
765 f += 2;
766 continue;
768 limit = t + 2;
771 while (*f && (*t++ = *f++) != '/') {}
774 if (t > name+anchored && t[-1] == '/' && !(flags & CFN_KEEP_TRAILING_SLASH))
775 t--;
776 if (t == name)
777 *t++ = '.';
778 *t = '\0';
780 return t - name;
783 /* Make path appear as if a chroot had occurred. This handles a leading
784 * "/" (either removing it or expanding it) and any leading or embedded
785 * ".." components that attempt to escape past the module's top dir.
787 * If dest is NULL, a buffer is allocated to hold the result. It is legal
788 * to call with the dest and the path (p) pointing to the same buffer, but
789 * rootdir will be ignored to avoid expansion of the string.
791 * The rootdir string contains a value to use in place of a leading slash.
792 * Specify NULL to get the default of "module_dir".
794 * The depth var is a count of how many '..'s to allow at the start of the
795 * path.
797 * We also clean the path in a manner similar to clean_fname() but with a
798 * few differences:
800 * Turns multiple adjacent slashes into a single slash, gets rid of "." dir
801 * elements (INCLUDING a trailing dot dir), PRESERVES a trailing slash, and
802 * ALWAYS collapses ".." elements (except for those at the start of the
803 * string up to "depth" deep). If the resulting name would be empty,
804 * change it into a ".". */
805 char *sanitize_path(char *dest, const char *p, const char *rootdir, int depth)
807 char *start, *sanp;
808 int rlen = 0, leave_one_dotdir = relative_paths;
810 if (dest != p) {
811 int plen = strlen(p);
812 if (*p == '/') {
813 if (!rootdir)
814 rootdir = module_dir;
815 rlen = strlen(rootdir);
816 depth = 0;
817 p++;
819 if (dest) {
820 if (rlen + plen + 1 >= MAXPATHLEN)
821 return NULL;
822 } else if (!(dest = new_array(char, rlen + plen + 1)))
823 out_of_memory("sanitize_path");
824 if (rlen) {
825 memcpy(dest, rootdir, rlen);
826 if (rlen > 1)
827 dest[rlen++] = '/';
831 start = sanp = dest + rlen;
832 /* This loop iterates once per filename component in p, pointing at
833 * the start of the name (past any prior slash) for each iteration. */
834 while (*p) {
835 /* discard leading or extra slashes */
836 if (*p == '/') {
837 p++;
838 continue;
840 if (*p == '.' && (p[1] == '/' || p[1] == '\0')) {
841 if (leave_one_dotdir && p[1])
842 leave_one_dotdir = 0;
843 else {
844 /* skip "." component */
845 p++;
846 continue;
849 if (*p == '.' && p[1] == '.' && (p[2] == '/' || p[2] == '\0')) {
850 /* ".." component followed by slash or end */
851 if (depth <= 0 || sanp != start) {
852 p += 2;
853 if (sanp != start) {
854 /* back up sanp one level */
855 --sanp; /* now pointing at slash */
856 while (sanp > start && sanp[-1] != '/') {
857 /* skip back up to slash */
858 sanp--;
861 continue;
863 /* allow depth levels of .. at the beginning */
864 depth--;
865 /* move the virtual beginning to leave the .. alone */
866 start = sanp + 3;
868 /* copy one component through next slash */
869 while (*p && (*sanp++ = *p++) != '/') {}
871 if (sanp == dest) {
872 /* ended up with nothing, so put in "." component */
873 *sanp++ = '.';
875 *sanp = '\0';
877 return dest;
880 /* Like chdir(), but it keeps track of the current directory (in the
881 * global "curr_dir"), and ensures that the path size doesn't overflow.
882 * Also cleans the path using the clean_fname() function. */
883 int push_dir(const char *dir, int set_path_only)
885 static int initialised;
886 unsigned int len;
888 if (!initialised) {
889 initialised = 1;
890 getcwd(curr_dir, sizeof curr_dir - 1);
891 curr_dir_len = strlen(curr_dir);
894 if (!dir) /* this call was probably just to initialize */
895 return 0;
897 len = strlen(dir);
898 if (len == 1 && *dir == '.')
899 return 1;
901 if ((*dir == '/' ? len : curr_dir_len + 1 + len) >= sizeof curr_dir) {
902 errno = ENAMETOOLONG;
903 return 0;
906 if (!set_path_only && chdir(dir))
907 return 0;
909 if (*dir == '/') {
910 memcpy(curr_dir, dir, len + 1);
911 curr_dir_len = len;
912 } else {
913 curr_dir[curr_dir_len++] = '/';
914 memcpy(curr_dir + curr_dir_len, dir, len + 1);
915 curr_dir_len += len;
918 curr_dir_len = clean_fname(curr_dir, CFN_COLLAPSE_DOT_DOT_DIRS);
919 if (sanitize_paths) {
920 if (module_dirlen > curr_dir_len)
921 module_dirlen = curr_dir_len;
922 curr_dir_depth = count_dir_elements(curr_dir + module_dirlen);
925 if (verbose >= 5 && !set_path_only)
926 rprintf(FINFO, "[%s] push_dir(%s)\n", who_am_i(), curr_dir);
928 return 1;
932 * Reverse a push_dir() call. You must pass in an absolute path
933 * that was copied from a prior value of "curr_dir".
935 int pop_dir(const char *dir)
937 if (chdir(dir))
938 return 0;
940 curr_dir_len = strlcpy(curr_dir, dir, sizeof curr_dir);
941 if (curr_dir_len >= sizeof curr_dir)
942 curr_dir_len = sizeof curr_dir - 1;
943 if (sanitize_paths)
944 curr_dir_depth = count_dir_elements(curr_dir + module_dirlen);
946 if (verbose >= 5)
947 rprintf(FINFO, "[%s] pop_dir(%s)\n", who_am_i(), curr_dir);
949 return 1;
953 * Return a quoted string with the full pathname of the indicated filename.
954 * The string " (in MODNAME)" may also be appended. The returned pointer
955 * remains valid until the next time full_fname() is called.
957 char *full_fname(const char *fn)
959 static char *result = NULL;
960 char *m1, *m2, *m3;
961 char *p1, *p2;
963 if (result)
964 free(result);
966 if (*fn == '/')
967 p1 = p2 = "";
968 else {
969 p1 = curr_dir + module_dirlen;
970 for (p2 = p1; *p2 == '/'; p2++) {}
971 if (*p2)
972 p2 = "/";
974 if (module_id >= 0) {
975 m1 = " (in ";
976 m2 = lp_name(module_id);
977 m3 = ")";
978 } else
979 m1 = m2 = m3 = "";
981 if (asprintf(&result, "\"%s%s%s\"%s%s%s", p1, p2, fn, m1, m2, m3) <= 0)
982 out_of_memory("full_fname");
984 return result;
987 static char partial_fname[MAXPATHLEN];
989 char *partial_dir_fname(const char *fname)
991 char *t = partial_fname;
992 int sz = sizeof partial_fname;
993 const char *fn;
995 if ((fn = strrchr(fname, '/')) != NULL) {
996 fn++;
997 if (*partial_dir != '/') {
998 int len = fn - fname;
999 strncpy(t, fname, len); /* safe */
1000 t += len;
1001 sz -= len;
1003 } else
1004 fn = fname;
1005 if ((int)pathjoin(t, sz, partial_dir, fn) >= sz)
1006 return NULL;
1007 if (server_filter_list.head) {
1008 t = strrchr(partial_fname, '/');
1009 *t = '\0';
1010 if (check_filter(&server_filter_list, partial_fname, 1) < 0)
1011 return NULL;
1012 *t = '/';
1013 if (check_filter(&server_filter_list, partial_fname, 0) < 0)
1014 return NULL;
1017 return partial_fname;
1020 /* If no --partial-dir option was specified, we don't need to do anything
1021 * (the partial-dir is essentially '.'), so just return success. */
1022 int handle_partial_dir(const char *fname, int create)
1024 char *fn, *dir;
1026 if (fname != partial_fname)
1027 return 1;
1028 if (!create && *partial_dir == '/')
1029 return 1;
1030 if (!(fn = strrchr(partial_fname, '/')))
1031 return 1;
1033 *fn = '\0';
1034 dir = partial_fname;
1035 if (create) {
1036 STRUCT_STAT st;
1037 int statret = do_lstat(dir, &st);
1038 if (statret == 0 && !S_ISDIR(st.st_mode)) {
1039 if (do_unlink(dir) < 0)
1040 return 0;
1041 statret = -1;
1043 if (statret < 0 && do_mkdir(dir, 0700) < 0)
1044 return 0;
1045 } else
1046 do_rmdir(dir);
1047 *fn = '/';
1049 return 1;
1053 * Determine if a symlink points outside the current directory tree.
1054 * This is considered "unsafe" because e.g. when mirroring somebody
1055 * else's machine it might allow them to establish a symlink to
1056 * /etc/passwd, and then read it through a web server.
1058 * Null symlinks and absolute symlinks are always unsafe.
1060 * Basically here we are concerned with symlinks whose target contains
1061 * "..", because this might cause us to walk back up out of the
1062 * transferred directory. We are not allowed to go back up and
1063 * reenter.
1065 * @param dest Target of the symlink in question.
1067 * @param src Top source directory currently applicable. Basically this
1068 * is the first parameter to rsync in a simple invocation, but it's
1069 * modified by flist.c in slightly complex ways.
1071 * @retval True if unsafe
1072 * @retval False is unsafe
1074 * @sa t_unsafe.c
1076 int unsafe_symlink(const char *dest, const char *src)
1078 const char *name, *slash;
1079 int depth = 0;
1081 /* all absolute and null symlinks are unsafe */
1082 if (!dest || !*dest || *dest == '/')
1083 return 1;
1085 /* find out what our safety margin is */
1086 for (name = src; (slash = strchr(name, '/')) != 0; name = slash+1) {
1087 if (strncmp(name, "../", 3) == 0) {
1088 depth = 0;
1089 } else if (strncmp(name, "./", 2) == 0) {
1090 /* nothing */
1091 } else {
1092 depth++;
1095 if (strcmp(name, "..") == 0)
1096 depth = 0;
1098 for (name = dest; (slash = strchr(name, '/')) != 0; name = slash+1) {
1099 if (strncmp(name, "../", 3) == 0) {
1100 /* if at any point we go outside the current directory
1101 then stop - it is unsafe */
1102 if (--depth < 0)
1103 return 1;
1104 } else if (strncmp(name, "./", 2) == 0) {
1105 /* nothing */
1106 } else {
1107 depth++;
1110 if (strcmp(name, "..") == 0)
1111 depth--;
1113 return (depth < 0);
1116 /* Return the int64 number as a string. If the --human-readable option was
1117 * specified, we may output the number in K, M, or G units. We can return
1118 * up to 4 buffers at a time. */
1119 char *human_num(int64 num)
1121 static char bufs[4][128]; /* more than enough room */
1122 static unsigned int n;
1123 char *s;
1125 n = (n + 1) % (sizeof bufs / sizeof bufs[0]);
1127 if (human_readable) {
1128 char units = '\0';
1129 int mult = human_readable == 1 ? 1000 : 1024;
1130 double dnum = 0;
1131 if (num > mult*mult*mult) {
1132 dnum = (double)num / (mult*mult*mult);
1133 units = 'G';
1134 } else if (num > mult*mult) {
1135 dnum = (double)num / (mult*mult);
1136 units = 'M';
1137 } else if (num > mult) {
1138 dnum = (double)num / mult;
1139 units = 'K';
1141 if (units) {
1142 snprintf(bufs[n], sizeof bufs[0], "%.2f%c", dnum, units);
1143 return bufs[n];
1147 s = bufs[n] + sizeof bufs[0] - 1;
1148 *s = '\0';
1150 if (!num)
1151 *--s = '0';
1152 while (num) {
1153 *--s = (char)(num % 10) + '0';
1154 num /= 10;
1156 return s;
1159 /* Return the double number as a string. If the --human-readable option was
1160 * specified, we may output the number in K, M, or G units. We use a buffer
1161 * from human_num() to return our result. */
1162 char *human_dnum(double dnum, int decimal_digits)
1164 char *buf = human_num(dnum);
1165 int len = strlen(buf);
1166 if (isDigit(buf + len - 1)) {
1167 /* There's extra room in buf prior to the start of the num. */
1168 buf -= decimal_digits + 1;
1169 snprintf(buf, len + decimal_digits + 2, "%.*f", decimal_digits, dnum);
1171 return buf;
1174 /* Return the date and time as a string. Some callers tweak returned buf. */
1175 char *timestring(time_t t)
1177 static char TimeBuf[200];
1178 struct tm *tm = localtime(&t);
1179 char *p;
1181 #ifdef HAVE_STRFTIME
1182 strftime(TimeBuf, sizeof TimeBuf - 1, "%Y/%m/%d %H:%M:%S", tm);
1183 #else
1184 strlcpy(TimeBuf, asctime(tm), sizeof TimeBuf);
1185 #endif
1187 if ((p = strchr(TimeBuf, '\n')) != NULL)
1188 *p = '\0';
1190 return TimeBuf;
1194 * Sleep for a specified number of milliseconds.
1196 * Always returns TRUE. (In the future it might return FALSE if
1197 * interrupted.)
1199 int msleep(int t)
1201 int tdiff = 0;
1202 struct timeval tval, t1, t2;
1204 gettimeofday(&t1, NULL);
1206 while (tdiff < t) {
1207 tval.tv_sec = (t-tdiff)/1000;
1208 tval.tv_usec = 1000*((t-tdiff)%1000);
1210 errno = 0;
1211 select(0,NULL,NULL, NULL, &tval);
1213 gettimeofday(&t2, NULL);
1214 tdiff = (t2.tv_sec - t1.tv_sec)*1000 +
1215 (t2.tv_usec - t1.tv_usec)/1000;
1218 return True;
1221 /* Determine if two time_t values are equivalent (either exact, or in
1222 * the modification timestamp window established by --modify-window).
1224 * @retval 0 if the times should be treated as the same
1226 * @retval +1 if the first is later
1228 * @retval -1 if the 2nd is later
1230 int cmp_time(time_t file1, time_t file2)
1232 if (file2 > file1) {
1233 if (file2 - file1 <= modify_window)
1234 return 0;
1235 return -1;
1237 if (file1 - file2 <= modify_window)
1238 return 0;
1239 return 1;
1243 #ifdef __INSURE__XX
1244 #include <dlfcn.h>
1247 This routine is a trick to immediately catch errors when debugging
1248 with insure. A xterm with a gdb is popped up when insure catches
1249 a error. It is Linux specific.
1251 int _Insure_trap_error(int a1, int a2, int a3, int a4, int a5, int a6)
1253 static int (*fn)();
1254 int ret;
1255 char *cmd;
1257 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'",
1258 getpid(), getpid(), getpid());
1260 if (!fn) {
1261 static void *h;
1262 h = dlopen("/usr/local/parasoft/insure++lite/lib.linux2/libinsure.so", RTLD_LAZY);
1263 fn = dlsym(h, "_Insure_trap_error");
1266 ret = fn(a1, a2, a3, a4, a5, a6);
1268 system(cmd);
1270 free(cmd);
1272 return ret;
1274 #endif
1276 #define MALLOC_MAX 0x40000000
1278 void *_new_array(unsigned long num, unsigned int size, int use_calloc)
1280 if (num >= MALLOC_MAX/size)
1281 return NULL;
1282 return use_calloc ? calloc(num, size) : malloc(num * size);
1285 void *_realloc_array(void *ptr, unsigned int size, unsigned long num)
1287 if (num >= MALLOC_MAX/size)
1288 return NULL;
1289 if (!ptr)
1290 return malloc(size * num);
1291 return realloc(ptr, size * num);
1294 /* Take a filename and filename length and return the most significant
1295 * filename suffix we can find. This ignores suffixes such as "~",
1296 * ".bak", ".orig", ".~1~", etc. */
1297 const char *find_filename_suffix(const char *fn, int fn_len, int *len_ptr)
1299 const char *suf, *s;
1300 BOOL had_tilde;
1301 int s_len;
1303 /* One or more dots at the start aren't a suffix. */
1304 while (fn_len && *fn == '.') fn++, fn_len--;
1306 /* Ignore the ~ in a "foo~" filename. */
1307 if (fn_len > 1 && fn[fn_len-1] == '~')
1308 fn_len--, had_tilde = True;
1309 else
1310 had_tilde = False;
1312 /* Assume we don't find an suffix. */
1313 suf = "";
1314 *len_ptr = 0;
1316 /* Find the last significant suffix. */
1317 for (s = fn + fn_len; fn_len > 1; ) {
1318 while (*--s != '.' && s != fn) {}
1319 if (s == fn)
1320 break;
1321 s_len = fn_len - (s - fn);
1322 fn_len = s - fn;
1323 if (s_len == 4) {
1324 if (strcmp(s+1, "bak") == 0
1325 || strcmp(s+1, "old") == 0)
1326 continue;
1327 } else if (s_len == 5) {
1328 if (strcmp(s+1, "orig") == 0)
1329 continue;
1330 } else if (s_len > 2 && had_tilde
1331 && s[1] == '~' && isDigit(s + 2))
1332 continue;
1333 *len_ptr = s_len;
1334 suf = s;
1335 if (s_len == 1)
1336 break;
1337 /* Determine if the suffix is all digits. */
1338 for (s++, s_len--; s_len > 0; s++, s_len--) {
1339 if (!isDigit(s))
1340 return suf;
1342 /* An all-digit suffix may not be that signficant. */
1343 s = suf;
1346 return suf;
1349 /* This is an implementation of the Levenshtein distance algorithm. It
1350 * was implemented to avoid needing a two-dimensional matrix (to save
1351 * memory). It was also tweaked to try to factor in the ASCII distance
1352 * between changed characters as a minor distance quantity. The normal
1353 * Levenshtein units of distance (each signifying a single change between
1354 * the two strings) are defined as a "UNIT". */
1356 #define UNIT (1 << 16)
1358 uint32 fuzzy_distance(const char *s1, int len1, const char *s2, int len2)
1360 uint32 a[MAXPATHLEN], diag, above, left, diag_inc, above_inc, left_inc;
1361 int32 cost;
1362 int i1, i2;
1364 if (!len1 || !len2) {
1365 if (!len1) {
1366 s1 = s2;
1367 len1 = len2;
1369 for (i1 = 0, cost = 0; i1 < len1; i1++)
1370 cost += s1[i1];
1371 return (int32)len1 * UNIT + cost;
1374 for (i2 = 0; i2 < len2; i2++)
1375 a[i2] = (i2+1) * UNIT;
1377 for (i1 = 0; i1 < len1; i1++) {
1378 diag = i1 * UNIT;
1379 above = (i1+1) * UNIT;
1380 for (i2 = 0; i2 < len2; i2++) {
1381 left = a[i2];
1382 if ((cost = *((uchar*)s1+i1) - *((uchar*)s2+i2)) != 0) {
1383 if (cost < 0)
1384 cost = UNIT - cost;
1385 else
1386 cost = UNIT + cost;
1388 diag_inc = diag + cost;
1389 left_inc = left + UNIT + *((uchar*)s1+i1);
1390 above_inc = above + UNIT + *((uchar*)s2+i2);
1391 a[i2] = above = left < above
1392 ? (left_inc < diag_inc ? left_inc : diag_inc)
1393 : (above_inc < diag_inc ? above_inc : diag_inc);
1394 diag = left;
1398 return a[len2-1];
1401 #define BB_SLOT_SIZE (16*1024) /* Desired size in bytes */
1402 #define BB_PER_SLOT_BITS (BB_SLOT_SIZE * 8) /* Number of bits per slot */
1403 #define BB_PER_SLOT_INTS (BB_SLOT_SIZE / 4) /* Number of int32s per slot */
1405 struct bitbag {
1406 uint32 **bits;
1407 int slot_cnt;
1410 struct bitbag *bitbag_create(int max_ndx)
1412 struct bitbag *bb = new(struct bitbag);
1413 bb->slot_cnt = (max_ndx + BB_PER_SLOT_BITS - 1) / BB_PER_SLOT_BITS;
1415 if (!(bb->bits = (uint32**)calloc(bb->slot_cnt, sizeof (uint32*))))
1416 out_of_memory("bitbag_create");
1418 return bb;
1421 void bitbag_set_bit(struct bitbag *bb, int ndx)
1423 int slot = ndx / BB_PER_SLOT_BITS;
1424 ndx %= BB_PER_SLOT_BITS;
1426 if (!bb->bits[slot]) {
1427 if (!(bb->bits[slot] = (uint32*)calloc(BB_PER_SLOT_INTS, 4)))
1428 out_of_memory("bitbag_set_bit");
1431 bb->bits[slot][ndx/32] |= 1u << (ndx % 32);
1434 #if 0 /* not needed yet */
1435 void bitbag_clear_bit(struct bitbag *bb, int ndx)
1437 int slot = ndx / BB_PER_SLOT_BITS;
1438 ndx %= BB_PER_SLOT_BITS;
1440 if (!bb->bits[slot])
1441 return;
1443 bb->bits[slot][ndx/32] &= ~(1u << (ndx % 32));
1446 int bitbag_check_bit(struct bitbag *bb, int ndx)
1448 int slot = ndx / BB_PER_SLOT_BITS;
1449 ndx %= BB_PER_SLOT_BITS;
1451 if (!bb->bits[slot])
1452 return 0;
1454 return bb->bits[slot][ndx/32] & (1u << (ndx % 32)) ? 1 : 0;
1456 #endif
1458 /* Call this with -1 to start checking from 0. Returns -1 at the end. */
1459 int bitbag_next_bit(struct bitbag *bb, int after)
1461 uint32 bits, mask;
1462 int i, ndx = after + 1;
1463 int slot = ndx / BB_PER_SLOT_BITS;
1464 ndx %= BB_PER_SLOT_BITS;
1466 mask = (1u << (ndx % 32)) - 1;
1467 for (i = ndx / 32; slot < bb->slot_cnt; slot++, i = mask = 0) {
1468 if (!bb->bits[slot])
1469 continue;
1470 for ( ; i < BB_PER_SLOT_INTS; i++, mask = 0) {
1471 if (!(bits = bb->bits[slot][i] & ~mask))
1472 continue;
1473 /* The xor magic figures out the lowest enabled bit in
1474 * bits, and the switch quickly computes log2(bit). */
1475 switch (bits ^ (bits & (bits-1))) {
1476 #define LOG2(n) case 1u << n: return slot*BB_PER_SLOT_BITS + i*32 + n
1477 LOG2(0); LOG2(1); LOG2(2); LOG2(3);
1478 LOG2(4); LOG2(5); LOG2(6); LOG2(7);
1479 LOG2(8); LOG2(9); LOG2(10); LOG2(11);
1480 LOG2(12); LOG2(13); LOG2(14); LOG2(15);
1481 LOG2(16); LOG2(17); LOG2(18); LOG2(19);
1482 LOG2(20); LOG2(21); LOG2(22); LOG2(23);
1483 LOG2(24); LOG2(25); LOG2(26); LOG2(27);
1484 LOG2(28); LOG2(29); LOG2(30); LOG2(31);
1486 return -1; /* impossible... */
1490 return -1;
1493 void *expand_item_list(item_list *lp, size_t item_size,
1494 const char *desc, int incr)
1496 /* First time through, 0 <= 0, so list is expanded. */
1497 if (lp->malloced <= lp->count) {
1498 void *new_ptr;
1499 size_t new_size = lp->malloced;
1500 if (incr < 0)
1501 new_size += -incr; /* increase slowly */
1502 else if (new_size < (size_t)incr)
1503 new_size += incr;
1504 else
1505 new_size *= 2;
1506 new_ptr = realloc_array(lp->items, char, new_size * item_size);
1507 if (verbose >= 4) {
1508 rprintf(FINFO, "[%s] expand %s to %.0f bytes, did%s move\n",
1509 who_am_i(), desc, (double)new_size * item_size,
1510 new_ptr == lp->items ? " not" : "");
1512 if (!new_ptr)
1513 out_of_memory("expand_item_list");
1515 lp->items = new_ptr;
1516 lp->malloced = new_size;
1518 return (char*)lp->items + (lp->count++ * item_size);