Added a --no-cd option to support/lsh so that the script can be used by
[rsync.git] / util.c
blob2afb63ec8fed36f4ac81bee395e579e123eb385e
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 daemon_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 #define ENSURE_MEMSPACE(buf, type, sz, req) \
507 if ((req) > sz && !(buf = realloc_array(buf, type, sz = MAX(sz * 2, req)))) \
508 out_of_memory("glob_expand")
510 static inline void call_glob_match(const char *name, int len, int from_glob,
511 char *arg, int abpos, int fbpos);
513 static struct glob_data {
514 char *arg_buf, *filt_buf, **argv;
515 int absize, fbsize, maxargs, argc;
516 } glob;
518 static void glob_match(char *arg, int abpos, int fbpos)
520 int len;
521 char *slash;
523 while (*arg == '.' && arg[1] == '/') {
524 if (fbpos < 0) {
525 ENSURE_MEMSPACE(glob.filt_buf, char, glob.fbsize, glob.absize);
526 memcpy(glob.filt_buf, glob.arg_buf, abpos + 1);
527 fbpos = abpos;
529 ENSURE_MEMSPACE(glob.arg_buf, char, glob.absize, abpos + 3);
530 glob.arg_buf[abpos++] = *arg++;
531 glob.arg_buf[abpos++] = *arg++;
532 glob.arg_buf[abpos] = '\0';
534 if ((slash = strchr(arg, '/')) != NULL) {
535 *slash = '\0';
536 len = slash - arg;
537 } else
538 len = strlen(arg);
539 if (strpbrk(arg, "*?[")) {
540 struct dirent *di;
541 DIR *d;
543 if (!(d = opendir(abpos ? glob.arg_buf : ".")))
544 return;
545 while ((di = readdir(d)) != NULL) {
546 char *dname = d_name(di);
547 if (dname[0] == '.' && (dname[1] == '\0'
548 || (dname[1] == '.' && dname[2] == '\0')))
549 continue;
550 if (!wildmatch(arg, dname))
551 continue;
552 call_glob_match(dname, strlen(dname), 1,
553 slash ? arg + len + 1 : NULL,
554 abpos, fbpos);
556 closedir(d);
557 } else {
558 call_glob_match(arg, len, 0,
559 slash ? arg + len + 1 : NULL,
560 abpos, fbpos);
562 if (slash)
563 *slash = '/';
566 static inline void call_glob_match(const char *name, int len, int from_glob,
567 char *arg, int abpos, int fbpos)
569 char *use_buf;
571 ENSURE_MEMSPACE(glob.arg_buf, char, glob.absize, abpos + len + 2);
572 memcpy(glob.arg_buf + abpos, name, len);
573 abpos += len;
574 glob.arg_buf[abpos] = '\0';
576 if (fbpos >= 0) {
577 ENSURE_MEMSPACE(glob.filt_buf, char, glob.fbsize, fbpos + len + 2);
578 memcpy(glob.filt_buf + fbpos, name, len);
579 fbpos += len;
580 glob.filt_buf[fbpos] = '\0';
581 use_buf = glob.filt_buf;
582 } else
583 use_buf = glob.arg_buf;
585 if (from_glob || (arg && len)) {
586 STRUCT_STAT st;
587 int is_dir;
589 if (do_stat(glob.arg_buf, &st) != 0)
590 return;
591 is_dir = S_ISDIR(st.st_mode) != 0;
592 if (arg && !is_dir)
593 return;
595 if (daemon_filter_list.head
596 && check_filter(&daemon_filter_list, FLOG, use_buf, is_dir) < 0)
597 return;
600 if (arg) {
601 glob.arg_buf[abpos++] = '/';
602 glob.arg_buf[abpos] = '\0';
603 if (fbpos >= 0) {
604 glob.filt_buf[fbpos++] = '/';
605 glob.filt_buf[fbpos] = '\0';
607 glob_match(arg, abpos, fbpos);
608 } else {
609 ENSURE_MEMSPACE(glob.argv, char *, glob.maxargs, glob.argc + 1);
610 if (!(glob.argv[glob.argc++] = strdup(glob.arg_buf)))
611 out_of_memory("glob_match");
615 /* This routine performs wild-card expansion of the pathname in "arg". Any
616 * daemon-excluded files/dirs will not be matched by the wildcards. Returns 0
617 * if a wild-card string is the only returned item (due to matching nothing). */
618 int glob_expand(const char *arg, char ***argv_p, int *argc_p, int *maxargs_p)
620 int ret, save_argc;
621 char *s;
623 if (!arg) {
624 if (glob.filt_buf)
625 free(glob.filt_buf);
626 free(glob.arg_buf);
627 memset(&glob, 0, sizeof glob);
628 return -1;
631 if (sanitize_paths)
632 s = sanitize_path(NULL, arg, "", 0, SP_KEEP_DOT_DIRS);
633 else {
634 s = strdup(arg);
635 if (!s)
636 out_of_memory("glob_expand");
637 clean_fname(s, CFN_KEEP_DOT_DIRS
638 | CFN_KEEP_TRAILING_SLASH
639 | CFN_COLLAPSE_DOT_DOT_DIRS);
642 ENSURE_MEMSPACE(glob.arg_buf, char, glob.absize, MAXPATHLEN);
643 *glob.arg_buf = '\0';
645 glob.argc = save_argc = *argc_p;
646 glob.argv = *argv_p;
647 glob.maxargs = *maxargs_p;
649 ENSURE_MEMSPACE(glob.argv, char *, glob.maxargs, 100);
651 glob_match(s, 0, -1);
653 /* The arg didn't match anything, so add the failed arg to the list. */
654 if (glob.argc == save_argc) {
655 ENSURE_MEMSPACE(glob.argv, char *, glob.maxargs, glob.argc + 1);
656 glob.argv[glob.argc++] = s;
657 ret = 0;
658 } else {
659 free(s);
660 ret = 1;
663 *maxargs_p = glob.maxargs;
664 *argv_p = glob.argv;
665 *argc_p = glob.argc;
667 return ret;
670 /* This routine is only used in daemon mode. */
671 void glob_expand_module(char *base1, char *arg, char ***argv_p, int *argc_p, int *maxargs_p)
673 char *p, *s;
674 char *base = base1;
675 int base_len = strlen(base);
677 if (!arg || !*arg)
678 return;
680 if (strncmp(arg, base, base_len) == 0)
681 arg += base_len;
683 if (!(arg = strdup(arg)))
684 out_of_memory("glob_expand_module");
686 if (asprintf(&base," %s/", base1) <= 0)
687 out_of_memory("glob_expand_module");
688 base_len++;
690 for (s = arg; *s; s = p + base_len) {
691 if ((p = strstr(s, base)) != NULL)
692 *p = '\0'; /* split it at this point */
693 glob_expand(s, argv_p, argc_p, maxargs_p);
694 if (!p)
695 break;
698 free(arg);
699 free(base);
703 * Convert a string to lower case
705 void strlower(char *s)
707 while (*s) {
708 if (isUpper(s))
709 *s = toLower(s);
710 s++;
714 /* Join strings p1 & p2 into "dest" with a guaranteed '/' between them. (If
715 * p1 ends with a '/', no extra '/' is inserted.) Returns the length of both
716 * strings + 1 (if '/' was inserted), regardless of whether the null-terminated
717 * string fits into destsize. */
718 size_t pathjoin(char *dest, size_t destsize, const char *p1, const char *p2)
720 size_t len = strlcpy(dest, p1, destsize);
721 if (len < destsize - 1) {
722 if (!len || dest[len-1] != '/')
723 dest[len++] = '/';
724 if (len < destsize - 1)
725 len += strlcpy(dest + len, p2, destsize - len);
726 else {
727 dest[len] = '\0';
728 len += strlen(p2);
731 else
732 len += strlen(p2) + 1; /* Assume we'd insert a '/'. */
733 return len;
736 /* Join any number of strings together, putting them in "dest". The return
737 * value is the length of all the strings, regardless of whether the null-
738 * terminated whole fits in destsize. Your list of string pointers must end
739 * with a NULL to indicate the end of the list. */
740 size_t stringjoin(char *dest, size_t destsize, ...)
742 va_list ap;
743 size_t len, ret = 0;
744 const char *src;
746 va_start(ap, destsize);
747 while (1) {
748 if (!(src = va_arg(ap, const char *)))
749 break;
750 len = strlen(src);
751 ret += len;
752 if (destsize > 1) {
753 if (len >= destsize)
754 len = destsize - 1;
755 memcpy(dest, src, len);
756 destsize -= len;
757 dest += len;
760 *dest = '\0';
761 va_end(ap);
763 return ret;
766 int count_dir_elements(const char *p)
768 int cnt = 0, new_component = 1;
769 while (*p) {
770 if (*p++ == '/')
771 new_component = (*p != '.' || (p[1] != '/' && p[1] != '\0'));
772 else if (new_component) {
773 new_component = 0;
774 cnt++;
777 return cnt;
780 /* Turns multiple adjacent slashes into a single slash, drops all leading or
781 * interior "." elements unless CFN_KEEP_DOT_DIRS is flagged. Will also drop
782 * a trailing '.' after a '/' if CFN_DROP_TRAILING_DOT_DIR is flagged, removes
783 * a trailing slash (perhaps after removing the aforementioned dot) unless
784 * CFN_KEEP_TRAILING_SLASH is flagged, and will also collapse ".." elements
785 * (except at the start) if CFN_COLLAPSE_DOT_DOT_DIRS is flagged. If the
786 * resulting name would be empty, returns ".". */
787 unsigned int clean_fname(char *name, int flags)
789 char *limit = name - 1, *t = name, *f = name;
790 int anchored;
792 if (!name)
793 return 0;
795 if ((anchored = *f == '/') != 0)
796 *t++ = *f++;
797 else if (flags & CFN_KEEP_DOT_DIRS && *f == '.' && f[1] == '/') {
798 *t++ = *f++;
799 *t++ = *f++;
801 while (*f) {
802 /* discard extra slashes */
803 if (*f == '/') {
804 f++;
805 continue;
807 if (*f == '.') {
808 /* discard interior "." dirs */
809 if (f[1] == '/' && !(flags & CFN_KEEP_DOT_DIRS)) {
810 f += 2;
811 continue;
813 if (f[1] == '\0' && flags & CFN_DROP_TRAILING_DOT_DIR)
814 break;
815 /* collapse ".." dirs */
816 if (flags & CFN_COLLAPSE_DOT_DOT_DIRS
817 && f[1] == '.' && (f[2] == '/' || !f[2])) {
818 char *s = t - 1;
819 if (s == name && anchored) {
820 f += 2;
821 continue;
823 while (s > limit && *--s != '/') {}
824 if (s != t - 1 && (s < name || *s == '/')) {
825 t = s + 1;
826 f += 2;
827 continue;
829 limit = t + 2;
832 while (*f && (*t++ = *f++) != '/') {}
835 if (t > name+anchored && t[-1] == '/' && !(flags & CFN_KEEP_TRAILING_SLASH))
836 t--;
837 if (t == name)
838 *t++ = '.';
839 *t = '\0';
841 return t - name;
844 /* Make path appear as if a chroot had occurred. This handles a leading
845 * "/" (either removing it or expanding it) and any leading or embedded
846 * ".." components that attempt to escape past the module's top dir.
848 * If dest is NULL, a buffer is allocated to hold the result. It is legal
849 * to call with the dest and the path (p) pointing to the same buffer, but
850 * rootdir will be ignored to avoid expansion of the string.
852 * The rootdir string contains a value to use in place of a leading slash.
853 * Specify NULL to get the default of "module_dir".
855 * The depth var is a count of how many '..'s to allow at the start of the
856 * path.
858 * We also clean the path in a manner similar to clean_fname() but with a
859 * few differences:
861 * Turns multiple adjacent slashes into a single slash, gets rid of "." dir
862 * elements (INCLUDING a trailing dot dir), PRESERVES a trailing slash, and
863 * ALWAYS collapses ".." elements (except for those at the start of the
864 * string up to "depth" deep). If the resulting name would be empty,
865 * change it into a ".". */
866 char *sanitize_path(char *dest, const char *p, const char *rootdir, int depth,
867 int flags)
869 char *start, *sanp;
870 int rlen = 0, drop_dot_dirs = !relative_paths || !(flags & SP_KEEP_DOT_DIRS);
872 if (dest != p) {
873 int plen = strlen(p);
874 if (*p == '/') {
875 if (!rootdir)
876 rootdir = module_dir;
877 rlen = strlen(rootdir);
878 depth = 0;
879 p++;
881 if (dest) {
882 if (rlen + plen + 1 >= MAXPATHLEN)
883 return NULL;
884 } else if (!(dest = new_array(char, rlen + plen + 1)))
885 out_of_memory("sanitize_path");
886 if (rlen) {
887 memcpy(dest, rootdir, rlen);
888 if (rlen > 1)
889 dest[rlen++] = '/';
893 if (drop_dot_dirs) {
894 while (*p == '.' && p[1] == '/')
895 p += 2;
898 start = sanp = dest + rlen;
899 /* This loop iterates once per filename component in p, pointing at
900 * the start of the name (past any prior slash) for each iteration. */
901 while (*p) {
902 /* discard leading or extra slashes */
903 if (*p == '/') {
904 p++;
905 continue;
907 if (drop_dot_dirs) {
908 if (*p == '.' && (p[1] == '/' || p[1] == '\0')) {
909 /* skip "." component */
910 p++;
911 continue;
914 if (*p == '.' && p[1] == '.' && (p[2] == '/' || p[2] == '\0')) {
915 /* ".." component followed by slash or end */
916 if (depth <= 0 || sanp != start) {
917 p += 2;
918 if (sanp != start) {
919 /* back up sanp one level */
920 --sanp; /* now pointing at slash */
921 while (sanp > start && sanp[-1] != '/')
922 sanp--;
924 continue;
926 /* allow depth levels of .. at the beginning */
927 depth--;
928 /* move the virtual beginning to leave the .. alone */
929 start = sanp + 3;
931 /* copy one component through next slash */
932 while (*p && (*sanp++ = *p++) != '/') {}
934 if (sanp == dest) {
935 /* ended up with nothing, so put in "." component */
936 *sanp++ = '.';
938 *sanp = '\0';
940 return dest;
943 /* Like chdir(), but it keeps track of the current directory (in the
944 * global "curr_dir"), and ensures that the path size doesn't overflow.
945 * Also cleans the path using the clean_fname() function. */
946 int change_dir(const char *dir, int set_path_only)
948 static int initialised;
949 unsigned int len;
951 if (!initialised) {
952 initialised = 1;
953 getcwd(curr_dir, sizeof curr_dir - 1);
954 curr_dir_len = strlen(curr_dir);
957 if (!dir) /* this call was probably just to initialize */
958 return 0;
960 len = strlen(dir);
961 if (len == 1 && *dir == '.')
962 return 1;
964 if (*dir == '/') {
965 if (len >= sizeof curr_dir) {
966 errno = ENAMETOOLONG;
967 return 0;
969 if (!set_path_only && chdir(dir))
970 return 0;
971 memcpy(curr_dir, dir, len + 1);
972 } else {
973 if (curr_dir_len + 1 + len >= sizeof curr_dir) {
974 errno = ENAMETOOLONG;
975 return 0;
977 curr_dir[curr_dir_len] = '/';
978 memcpy(curr_dir + curr_dir_len + 1, dir, len + 1);
980 if (!set_path_only && chdir(curr_dir)) {
981 curr_dir[curr_dir_len] = '\0';
982 return 0;
986 curr_dir_len = clean_fname(curr_dir, CFN_COLLAPSE_DOT_DOT_DIRS);
987 if (sanitize_paths) {
988 if (module_dirlen > curr_dir_len)
989 module_dirlen = curr_dir_len;
990 curr_dir_depth = count_dir_elements(curr_dir + module_dirlen);
993 if (verbose >= 5 && !set_path_only)
994 rprintf(FINFO, "[%s] change_dir(%s)\n", who_am_i(), curr_dir);
996 return 1;
1000 * Return a quoted string with the full pathname of the indicated filename.
1001 * The string " (in MODNAME)" may also be appended. The returned pointer
1002 * remains valid until the next time full_fname() is called.
1004 char *full_fname(const char *fn)
1006 static char *result = NULL;
1007 char *m1, *m2, *m3;
1008 char *p1, *p2;
1010 if (result)
1011 free(result);
1013 if (*fn == '/')
1014 p1 = p2 = "";
1015 else {
1016 p1 = curr_dir + module_dirlen;
1017 for (p2 = p1; *p2 == '/'; p2++) {}
1018 if (*p2)
1019 p2 = "/";
1021 if (module_id >= 0) {
1022 m1 = " (in ";
1023 m2 = lp_name(module_id);
1024 m3 = ")";
1025 } else
1026 m1 = m2 = m3 = "";
1028 if (asprintf(&result, "\"%s%s%s\"%s%s%s", p1, p2, fn, m1, m2, m3) <= 0)
1029 out_of_memory("full_fname");
1031 return result;
1034 static char partial_fname[MAXPATHLEN];
1036 char *partial_dir_fname(const char *fname)
1038 char *t = partial_fname;
1039 int sz = sizeof partial_fname;
1040 const char *fn;
1042 if ((fn = strrchr(fname, '/')) != NULL) {
1043 fn++;
1044 if (*partial_dir != '/') {
1045 int len = fn - fname;
1046 strncpy(t, fname, len); /* safe */
1047 t += len;
1048 sz -= len;
1050 } else
1051 fn = fname;
1052 if ((int)pathjoin(t, sz, partial_dir, fn) >= sz)
1053 return NULL;
1054 if (daemon_filter_list.head) {
1055 t = strrchr(partial_fname, '/');
1056 *t = '\0';
1057 if (check_filter(&daemon_filter_list, FLOG, partial_fname, 1) < 0)
1058 return NULL;
1059 *t = '/';
1060 if (check_filter(&daemon_filter_list, FLOG, partial_fname, 0) < 0)
1061 return NULL;
1064 return partial_fname;
1067 /* If no --partial-dir option was specified, we don't need to do anything
1068 * (the partial-dir is essentially '.'), so just return success. */
1069 int handle_partial_dir(const char *fname, int create)
1071 char *fn, *dir;
1073 if (fname != partial_fname)
1074 return 1;
1075 if (!create && *partial_dir == '/')
1076 return 1;
1077 if (!(fn = strrchr(partial_fname, '/')))
1078 return 1;
1080 *fn = '\0';
1081 dir = partial_fname;
1082 if (create) {
1083 STRUCT_STAT st;
1084 int statret = do_lstat(dir, &st);
1085 if (statret == 0 && !S_ISDIR(st.st_mode)) {
1086 if (do_unlink(dir) < 0)
1087 return 0;
1088 statret = -1;
1090 if (statret < 0 && do_mkdir(dir, 0700) < 0)
1091 return 0;
1092 } else
1093 do_rmdir(dir);
1094 *fn = '/';
1096 return 1;
1100 * Determine if a symlink points outside the current directory tree.
1101 * This is considered "unsafe" because e.g. when mirroring somebody
1102 * else's machine it might allow them to establish a symlink to
1103 * /etc/passwd, and then read it through a web server.
1105 * Null symlinks and absolute symlinks are always unsafe.
1107 * Basically here we are concerned with symlinks whose target contains
1108 * "..", because this might cause us to walk back up out of the
1109 * transferred directory. We are not allowed to go back up and
1110 * reenter.
1112 * @param dest Target of the symlink in question.
1114 * @param src Top source directory currently applicable. Basically this
1115 * is the first parameter to rsync in a simple invocation, but it's
1116 * modified by flist.c in slightly complex ways.
1118 * @retval True if unsafe
1119 * @retval False is unsafe
1121 * @sa t_unsafe.c
1123 int unsafe_symlink(const char *dest, const char *src)
1125 const char *name, *slash;
1126 int depth = 0;
1128 /* all absolute and null symlinks are unsafe */
1129 if (!dest || !*dest || *dest == '/')
1130 return 1;
1132 /* find out what our safety margin is */
1133 for (name = src; (slash = strchr(name, '/')) != 0; name = slash+1) {
1134 if (strncmp(name, "../", 3) == 0) {
1135 depth = 0;
1136 } else if (strncmp(name, "./", 2) == 0) {
1137 /* nothing */
1138 } else {
1139 depth++;
1142 if (strcmp(name, "..") == 0)
1143 depth = 0;
1145 for (name = dest; (slash = strchr(name, '/')) != 0; name = slash+1) {
1146 if (strncmp(name, "../", 3) == 0) {
1147 /* if at any point we go outside the current directory
1148 then stop - it is unsafe */
1149 if (--depth < 0)
1150 return 1;
1151 } else if (strncmp(name, "./", 2) == 0) {
1152 /* nothing */
1153 } else {
1154 depth++;
1157 if (strcmp(name, "..") == 0)
1158 depth--;
1160 return (depth < 0);
1163 /* Return the int64 number as a string. If the --human-readable option was
1164 * specified, we may output the number in K, M, or G units. We can return
1165 * up to 4 buffers at a time. */
1166 char *human_num(int64 num)
1168 static char bufs[4][128]; /* more than enough room */
1169 static unsigned int n;
1170 char *s;
1172 n = (n + 1) % (sizeof bufs / sizeof bufs[0]);
1174 if (human_readable) {
1175 char units = '\0';
1176 int mult = human_readable == 1 ? 1000 : 1024;
1177 double dnum = 0;
1178 if (num > mult*mult*mult) {
1179 dnum = (double)num / (mult*mult*mult);
1180 units = 'G';
1181 } else if (num > mult*mult) {
1182 dnum = (double)num / (mult*mult);
1183 units = 'M';
1184 } else if (num > mult) {
1185 dnum = (double)num / mult;
1186 units = 'K';
1188 if (units) {
1189 snprintf(bufs[n], sizeof bufs[0], "%.2f%c", dnum, units);
1190 return bufs[n];
1194 s = bufs[n] + sizeof bufs[0] - 1;
1195 *s = '\0';
1197 if (!num)
1198 *--s = '0';
1199 while (num) {
1200 *--s = (char)(num % 10) + '0';
1201 num /= 10;
1203 return s;
1206 /* Return the double number as a string. If the --human-readable option was
1207 * specified, we may output the number in K, M, or G units. We use a buffer
1208 * from human_num() to return our result. */
1209 char *human_dnum(double dnum, int decimal_digits)
1211 char *buf = human_num(dnum);
1212 int len = strlen(buf);
1213 if (isDigit(buf + len - 1)) {
1214 /* There's extra room in buf prior to the start of the num. */
1215 buf -= decimal_digits + 1;
1216 snprintf(buf, len + decimal_digits + 2, "%.*f", decimal_digits, dnum);
1218 return buf;
1221 /* Return the date and time as a string. Some callers tweak returned buf. */
1222 char *timestring(time_t t)
1224 static char TimeBuf[200];
1225 struct tm *tm = localtime(&t);
1226 char *p;
1228 #ifdef HAVE_STRFTIME
1229 strftime(TimeBuf, sizeof TimeBuf - 1, "%Y/%m/%d %H:%M:%S", tm);
1230 #else
1231 strlcpy(TimeBuf, asctime(tm), sizeof TimeBuf);
1232 #endif
1234 if ((p = strchr(TimeBuf, '\n')) != NULL)
1235 *p = '\0';
1237 return TimeBuf;
1241 * Sleep for a specified number of milliseconds.
1243 * Always returns TRUE. (In the future it might return FALSE if
1244 * interrupted.)
1246 int msleep(int t)
1248 int tdiff = 0;
1249 struct timeval tval, t1, t2;
1251 gettimeofday(&t1, NULL);
1253 while (tdiff < t) {
1254 tval.tv_sec = (t-tdiff)/1000;
1255 tval.tv_usec = 1000*((t-tdiff)%1000);
1257 errno = 0;
1258 select(0,NULL,NULL, NULL, &tval);
1260 gettimeofday(&t2, NULL);
1261 tdiff = (t2.tv_sec - t1.tv_sec)*1000 +
1262 (t2.tv_usec - t1.tv_usec)/1000;
1265 return True;
1268 /* Determine if two time_t values are equivalent (either exact, or in
1269 * the modification timestamp window established by --modify-window).
1271 * @retval 0 if the times should be treated as the same
1273 * @retval +1 if the first is later
1275 * @retval -1 if the 2nd is later
1277 int cmp_time(time_t file1, time_t file2)
1279 if (file2 > file1) {
1280 if (file2 - file1 <= modify_window)
1281 return 0;
1282 return -1;
1284 if (file1 - file2 <= modify_window)
1285 return 0;
1286 return 1;
1290 #ifdef __INSURE__XX
1291 #include <dlfcn.h>
1294 This routine is a trick to immediately catch errors when debugging
1295 with insure. A xterm with a gdb is popped up when insure catches
1296 a error. It is Linux specific.
1298 int _Insure_trap_error(int a1, int a2, int a3, int a4, int a5, int a6)
1300 static int (*fn)();
1301 int ret;
1302 char *cmd;
1304 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'",
1305 getpid(), getpid(), getpid());
1307 if (!fn) {
1308 static void *h;
1309 h = dlopen("/usr/local/parasoft/insure++lite/lib.linux2/libinsure.so", RTLD_LAZY);
1310 fn = dlsym(h, "_Insure_trap_error");
1313 ret = fn(a1, a2, a3, a4, a5, a6);
1315 system(cmd);
1317 free(cmd);
1319 return ret;
1321 #endif
1323 #define MALLOC_MAX 0x40000000
1325 void *_new_array(unsigned long num, unsigned int size, int use_calloc)
1327 if (num >= MALLOC_MAX/size)
1328 return NULL;
1329 return use_calloc ? calloc(num, size) : malloc(num * size);
1332 void *_realloc_array(void *ptr, unsigned int size, size_t num)
1334 if (num >= MALLOC_MAX/size)
1335 return NULL;
1336 if (!ptr)
1337 return malloc(size * num);
1338 return realloc(ptr, size * num);
1341 /* Take a filename and filename length and return the most significant
1342 * filename suffix we can find. This ignores suffixes such as "~",
1343 * ".bak", ".orig", ".~1~", etc. */
1344 const char *find_filename_suffix(const char *fn, int fn_len, int *len_ptr)
1346 const char *suf, *s;
1347 BOOL had_tilde;
1348 int s_len;
1350 /* One or more dots at the start aren't a suffix. */
1351 while (fn_len && *fn == '.') fn++, fn_len--;
1353 /* Ignore the ~ in a "foo~" filename. */
1354 if (fn_len > 1 && fn[fn_len-1] == '~')
1355 fn_len--, had_tilde = True;
1356 else
1357 had_tilde = False;
1359 /* Assume we don't find an suffix. */
1360 suf = "";
1361 *len_ptr = 0;
1363 /* Find the last significant suffix. */
1364 for (s = fn + fn_len; fn_len > 1; ) {
1365 while (*--s != '.' && s != fn) {}
1366 if (s == fn)
1367 break;
1368 s_len = fn_len - (s - fn);
1369 fn_len = s - fn;
1370 if (s_len == 4) {
1371 if (strcmp(s+1, "bak") == 0
1372 || strcmp(s+1, "old") == 0)
1373 continue;
1374 } else if (s_len == 5) {
1375 if (strcmp(s+1, "orig") == 0)
1376 continue;
1377 } else if (s_len > 2 && had_tilde
1378 && s[1] == '~' && isDigit(s + 2))
1379 continue;
1380 *len_ptr = s_len;
1381 suf = s;
1382 if (s_len == 1)
1383 break;
1384 /* Determine if the suffix is all digits. */
1385 for (s++, s_len--; s_len > 0; s++, s_len--) {
1386 if (!isDigit(s))
1387 return suf;
1389 /* An all-digit suffix may not be that signficant. */
1390 s = suf;
1393 return suf;
1396 /* This is an implementation of the Levenshtein distance algorithm. It
1397 * was implemented to avoid needing a two-dimensional matrix (to save
1398 * memory). It was also tweaked to try to factor in the ASCII distance
1399 * between changed characters as a minor distance quantity. The normal
1400 * Levenshtein units of distance (each signifying a single change between
1401 * the two strings) are defined as a "UNIT". */
1403 #define UNIT (1 << 16)
1405 uint32 fuzzy_distance(const char *s1, int len1, const char *s2, int len2)
1407 uint32 a[MAXPATHLEN], diag, above, left, diag_inc, above_inc, left_inc;
1408 int32 cost;
1409 int i1, i2;
1411 if (!len1 || !len2) {
1412 if (!len1) {
1413 s1 = s2;
1414 len1 = len2;
1416 for (i1 = 0, cost = 0; i1 < len1; i1++)
1417 cost += s1[i1];
1418 return (int32)len1 * UNIT + cost;
1421 for (i2 = 0; i2 < len2; i2++)
1422 a[i2] = (i2+1) * UNIT;
1424 for (i1 = 0; i1 < len1; i1++) {
1425 diag = i1 * UNIT;
1426 above = (i1+1) * UNIT;
1427 for (i2 = 0; i2 < len2; i2++) {
1428 left = a[i2];
1429 if ((cost = *((uchar*)s1+i1) - *((uchar*)s2+i2)) != 0) {
1430 if (cost < 0)
1431 cost = UNIT - cost;
1432 else
1433 cost = UNIT + cost;
1435 diag_inc = diag + cost;
1436 left_inc = left + UNIT + *((uchar*)s1+i1);
1437 above_inc = above + UNIT + *((uchar*)s2+i2);
1438 a[i2] = above = left < above
1439 ? (left_inc < diag_inc ? left_inc : diag_inc)
1440 : (above_inc < diag_inc ? above_inc : diag_inc);
1441 diag = left;
1445 return a[len2-1];
1448 #define BB_SLOT_SIZE (16*1024) /* Desired size in bytes */
1449 #define BB_PER_SLOT_BITS (BB_SLOT_SIZE * 8) /* Number of bits per slot */
1450 #define BB_PER_SLOT_INTS (BB_SLOT_SIZE / 4) /* Number of int32s per slot */
1452 struct bitbag {
1453 uint32 **bits;
1454 int slot_cnt;
1457 struct bitbag *bitbag_create(int max_ndx)
1459 struct bitbag *bb = new(struct bitbag);
1460 bb->slot_cnt = (max_ndx + BB_PER_SLOT_BITS - 1) / BB_PER_SLOT_BITS;
1462 if (!(bb->bits = (uint32**)calloc(bb->slot_cnt, sizeof (uint32*))))
1463 out_of_memory("bitbag_create");
1465 return bb;
1468 void bitbag_set_bit(struct bitbag *bb, int ndx)
1470 int slot = ndx / BB_PER_SLOT_BITS;
1471 ndx %= BB_PER_SLOT_BITS;
1473 if (!bb->bits[slot]) {
1474 if (!(bb->bits[slot] = (uint32*)calloc(BB_PER_SLOT_INTS, 4)))
1475 out_of_memory("bitbag_set_bit");
1478 bb->bits[slot][ndx/32] |= 1u << (ndx % 32);
1481 #if 0 /* not needed yet */
1482 void bitbag_clear_bit(struct bitbag *bb, int ndx)
1484 int slot = ndx / BB_PER_SLOT_BITS;
1485 ndx %= BB_PER_SLOT_BITS;
1487 if (!bb->bits[slot])
1488 return;
1490 bb->bits[slot][ndx/32] &= ~(1u << (ndx % 32));
1493 int bitbag_check_bit(struct bitbag *bb, int ndx)
1495 int slot = ndx / BB_PER_SLOT_BITS;
1496 ndx %= BB_PER_SLOT_BITS;
1498 if (!bb->bits[slot])
1499 return 0;
1501 return bb->bits[slot][ndx/32] & (1u << (ndx % 32)) ? 1 : 0;
1503 #endif
1505 /* Call this with -1 to start checking from 0. Returns -1 at the end. */
1506 int bitbag_next_bit(struct bitbag *bb, int after)
1508 uint32 bits, mask;
1509 int i, ndx = after + 1;
1510 int slot = ndx / BB_PER_SLOT_BITS;
1511 ndx %= BB_PER_SLOT_BITS;
1513 mask = (1u << (ndx % 32)) - 1;
1514 for (i = ndx / 32; slot < bb->slot_cnt; slot++, i = mask = 0) {
1515 if (!bb->bits[slot])
1516 continue;
1517 for ( ; i < BB_PER_SLOT_INTS; i++, mask = 0) {
1518 if (!(bits = bb->bits[slot][i] & ~mask))
1519 continue;
1520 /* The xor magic figures out the lowest enabled bit in
1521 * bits, and the switch quickly computes log2(bit). */
1522 switch (bits ^ (bits & (bits-1))) {
1523 #define LOG2(n) case 1u << n: return slot*BB_PER_SLOT_BITS + i*32 + n
1524 LOG2(0); LOG2(1); LOG2(2); LOG2(3);
1525 LOG2(4); LOG2(5); LOG2(6); LOG2(7);
1526 LOG2(8); LOG2(9); LOG2(10); LOG2(11);
1527 LOG2(12); LOG2(13); LOG2(14); LOG2(15);
1528 LOG2(16); LOG2(17); LOG2(18); LOG2(19);
1529 LOG2(20); LOG2(21); LOG2(22); LOG2(23);
1530 LOG2(24); LOG2(25); LOG2(26); LOG2(27);
1531 LOG2(28); LOG2(29); LOG2(30); LOG2(31);
1533 return -1; /* impossible... */
1537 return -1;
1540 void *expand_item_list(item_list *lp, size_t item_size,
1541 const char *desc, int incr)
1543 /* First time through, 0 <= 0, so list is expanded. */
1544 if (lp->malloced <= lp->count) {
1545 void *new_ptr;
1546 size_t new_size = lp->malloced;
1547 if (incr < 0)
1548 new_size += -incr; /* increase slowly */
1549 else if (new_size < (size_t)incr)
1550 new_size += incr;
1551 else
1552 new_size *= 2;
1553 if (new_size < lp->malloced)
1554 overflow_exit("expand_item_list");
1555 /* Using _realloc_array() lets us pass the size, not a type. */
1556 new_ptr = _realloc_array(lp->items, item_size, new_size);
1557 if (verbose >= 4) {
1558 rprintf(FINFO, "[%s] expand %s to %.0f bytes, did%s move\n",
1559 who_am_i(), desc, (double)new_size * item_size,
1560 new_ptr == lp->items ? " not" : "");
1562 if (!new_ptr)
1563 out_of_memory("expand_item_list");
1565 lp->items = new_ptr;
1566 lp->malloced = new_size;
1568 return (char*)lp->items + (lp->count++ * item_size);