Make glob_expand() return an indicator if the glob had no matches.
[rsync.git] / util.c
blob99dff8050c88dd07e2f95a60b4adfb4b22e04e8d
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 *= 2))) \
508 out_of_memory("ENSURE_MEMSPACE")
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 if (glob.fbsize < glob.absize) {
526 glob.filt_buf = realloc_array(glob.filt_buf,
527 char, glob.fbsize = glob.absize);
529 memcpy(glob.filt_buf, glob.arg_buf, abpos + 1);
530 fbpos = abpos;
532 ENSURE_MEMSPACE(glob.arg_buf, char, glob.absize, abpos + 2);
533 glob.arg_buf[abpos++] = *arg++;
534 glob.arg_buf[abpos++] = *arg++;
535 glob.arg_buf[abpos] = '\0';
537 if ((slash = strchr(arg, '/')) != NULL) {
538 *slash = '\0';
539 len = slash - arg;
540 } else
541 len = strlen(arg);
542 if (strpbrk(arg, "*?[")) {
543 struct dirent *di;
544 DIR *d;
546 if (!(d = opendir(abpos ? glob.arg_buf : ".")))
547 return;
548 while ((di = readdir(d)) != NULL) {
549 char *dname = d_name(di);
550 if (dname[0] == '.' && (dname[1] == '\0'
551 || (dname[1] == '.' && dname[2] == '\0')))
552 continue;
553 if (!wildmatch(arg, dname))
554 continue;
555 call_glob_match(dname, strlen(dname), 1,
556 slash ? arg + len + 1 : NULL,
557 abpos, fbpos);
559 closedir(d);
560 } else {
561 call_glob_match(arg, len, 0,
562 slash ? arg + len + 1 : NULL,
563 abpos, fbpos);
565 if (slash)
566 *slash = '/';
569 static inline void call_glob_match(const char *name, int len, int from_glob,
570 char *arg, int abpos, int fbpos)
572 char *use_buf;
574 ENSURE_MEMSPACE(glob.arg_buf, char, glob.absize, abpos + len + 2);
575 memcpy(glob.arg_buf + abpos, name, len);
576 abpos += len;
577 glob.arg_buf[abpos] = '\0';
579 if (fbpos >= 0) {
580 ENSURE_MEMSPACE(glob.filt_buf, char, glob.fbsize, fbpos + len + 2);
581 memcpy(glob.filt_buf + fbpos, name, len);
582 fbpos += len;
583 glob.filt_buf[fbpos] = '\0';
584 use_buf = glob.filt_buf;
585 } else
586 use_buf = glob.arg_buf;
588 if (from_glob || arg) {
589 STRUCT_STAT st;
590 int is_dir;
592 if (do_stat(glob.arg_buf, &st) != 0) {
593 if (from_glob)
594 return;
595 is_dir = 0;
596 } else {
597 is_dir = S_ISDIR(st.st_mode) != 0;
598 if (arg && !is_dir)
599 return;
602 if (daemon_filter_list.head
603 && check_filter(&daemon_filter_list, use_buf, is_dir) < 0) {
604 if (from_glob)
605 return;
606 arg = NULL;
610 if (arg) {
611 glob.arg_buf[abpos++] = '/';
612 glob.arg_buf[abpos] = '\0';
613 if (fbpos >= 0) {
614 glob.filt_buf[fbpos++] = '/';
615 glob.filt_buf[fbpos] = '\0';
617 glob_match(arg, abpos, fbpos);
618 } else {
619 ENSURE_MEMSPACE(glob.argv, char *, glob.maxargs, glob.argc + 1);
620 if (!(glob.argv[glob.argc++] = strdup(glob.arg_buf)))
621 out_of_memory("glob_match");
625 /* This routine performs wild-card expansion of the pathname in "arg". Any
626 * daemon-excluded files/dirs will not be matched by the wildcards. Returns 0
627 * if a wild-card string is the only returned item (due to matching nothing). */
628 int glob_expand(const char *arg, char ***argv_p, int *argc_p, int *maxargs_p)
630 int ret, save_argc;
631 char *s;
633 if (!arg) {
634 if (glob.filt_buf)
635 free(glob.filt_buf);
636 free(glob.arg_buf);
637 memset(&glob, 0, sizeof glob);
638 return -1;
641 if (sanitize_paths)
642 s = sanitize_path(NULL, arg, "", 0, SP_KEEP_DOT_DIRS);
643 else {
644 s = strdup(arg);
645 if (!s)
646 out_of_memory("glob_expand");
647 clean_fname(s, CFN_KEEP_DOT_DIRS
648 | CFN_KEEP_TRAILING_SLASH
649 | CFN_COLLAPSE_DOT_DOT_DIRS);
652 if (glob.absize < MAXPATHLEN
653 && !(glob.arg_buf = realloc_array(glob.arg_buf, char, glob.absize = MAXPATHLEN)))
654 out_of_memory("glob_expand");
655 *glob.arg_buf = '\0';
657 glob.argc = save_argc = *argc_p;
658 glob.argv = *argv_p;
659 glob.maxargs = *maxargs_p;
661 if (glob.maxargs < 100
662 && !(glob.argv = realloc_array(glob.argv, char *, glob.maxargs = 100)))
663 out_of_memory("glob_expand");
665 glob_match(s, 0, -1);
667 /* The arg didn't match anything, so add the failed arg to the list. */
668 if (glob.argc == save_argc) {
669 ENSURE_MEMSPACE(glob.argv, char *, glob.maxargs, glob.argc + 1);
670 glob.argv[glob.argc++] = s;
671 ret = 0;
672 } else {
673 free(s);
674 ret = 1;
677 *maxargs_p = glob.maxargs;
678 *argv_p = glob.argv;
679 *argc_p = glob.argc;
681 return ret;
684 /* This routine is only used in daemon mode. */
685 void glob_expand_module(char *base1, char *arg, char ***argv_p, int *argc_p, int *maxargs_p)
687 char *p, *s;
688 char *base = base1;
689 int base_len = strlen(base);
691 if (!arg || !*arg)
692 return;
694 if (strncmp(arg, base, base_len) == 0)
695 arg += base_len;
697 if (!(arg = strdup(arg)))
698 out_of_memory("glob_expand_module");
700 if (asprintf(&base," %s/", base1) <= 0)
701 out_of_memory("glob_expand_module");
702 base_len++;
704 for (s = arg; *s; s = p + base_len) {
705 if ((p = strstr(s, base)) != NULL)
706 *p = '\0'; /* split it at this point */
707 glob_expand(s, argv_p, argc_p, maxargs_p);
708 if (!p)
709 break;
712 free(arg);
713 free(base);
717 * Convert a string to lower case
719 void strlower(char *s)
721 while (*s) {
722 if (isUpper(s))
723 *s = toLower(s);
724 s++;
728 /* Join strings p1 & p2 into "dest" with a guaranteed '/' between them. (If
729 * p1 ends with a '/', no extra '/' is inserted.) Returns the length of both
730 * strings + 1 (if '/' was inserted), regardless of whether the null-terminated
731 * string fits into destsize. */
732 size_t pathjoin(char *dest, size_t destsize, const char *p1, const char *p2)
734 size_t len = strlcpy(dest, p1, destsize);
735 if (len < destsize - 1) {
736 if (!len || dest[len-1] != '/')
737 dest[len++] = '/';
738 if (len < destsize - 1)
739 len += strlcpy(dest + len, p2, destsize - len);
740 else {
741 dest[len] = '\0';
742 len += strlen(p2);
745 else
746 len += strlen(p2) + 1; /* Assume we'd insert a '/'. */
747 return len;
750 /* Join any number of strings together, putting them in "dest". The return
751 * value is the length of all the strings, regardless of whether the null-
752 * terminated whole fits in destsize. Your list of string pointers must end
753 * with a NULL to indicate the end of the list. */
754 size_t stringjoin(char *dest, size_t destsize, ...)
756 va_list ap;
757 size_t len, ret = 0;
758 const char *src;
760 va_start(ap, destsize);
761 while (1) {
762 if (!(src = va_arg(ap, const char *)))
763 break;
764 len = strlen(src);
765 ret += len;
766 if (destsize > 1) {
767 if (len >= destsize)
768 len = destsize - 1;
769 memcpy(dest, src, len);
770 destsize -= len;
771 dest += len;
774 *dest = '\0';
775 va_end(ap);
777 return ret;
780 int count_dir_elements(const char *p)
782 int cnt = 0, new_component = 1;
783 while (*p) {
784 if (*p++ == '/')
785 new_component = (*p != '.' || (p[1] != '/' && p[1] != '\0'));
786 else if (new_component) {
787 new_component = 0;
788 cnt++;
791 return cnt;
794 /* Turns multiple adjacent slashes into a single slash, drops all leading or
795 * interior "." elements unless CFN_KEEP_DOT_DIRS is flagged. Will also drop
796 * a trailing '.' after a '/' if CFN_DROP_TRAILING_DOT_DIR is flagged, removes
797 * a trailing slash (perhaps after removing the aforementioned dot) unless
798 * CFN_KEEP_TRAILING_SLASH is flagged, and will also collapse ".." elements
799 * (except at the start) if CFN_COLLAPSE_DOT_DOT_DIRS is flagged. If the
800 * resulting name would be empty, returns ".". */
801 unsigned int clean_fname(char *name, int flags)
803 char *limit = name - 1, *t = name, *f = name;
804 int anchored;
806 if (!name)
807 return 0;
809 if ((anchored = *f == '/') != 0)
810 *t++ = *f++;
811 else if (flags & CFN_KEEP_DOT_DIRS && *f == '.' && f[1] == '/') {
812 *t++ = *f++;
813 *t++ = *f++;
815 while (*f) {
816 /* discard extra slashes */
817 if (*f == '/') {
818 f++;
819 continue;
821 if (*f == '.') {
822 /* discard interior "." dirs */
823 if (f[1] == '/' && !(flags & CFN_KEEP_DOT_DIRS)) {
824 f += 2;
825 continue;
827 if (f[1] == '\0' && flags & CFN_DROP_TRAILING_DOT_DIR)
828 break;
829 /* collapse ".." dirs */
830 if (flags & CFN_COLLAPSE_DOT_DOT_DIRS
831 && f[1] == '.' && (f[2] == '/' || !f[2])) {
832 char *s = t - 1;
833 if (s == name && anchored) {
834 f += 2;
835 continue;
837 while (s > limit && *--s != '/') {}
838 if (s != t - 1 && (s < name || *s == '/')) {
839 t = s + 1;
840 f += 2;
841 continue;
843 limit = t + 2;
846 while (*f && (*t++ = *f++) != '/') {}
849 if (t > name+anchored && t[-1] == '/' && !(flags & CFN_KEEP_TRAILING_SLASH))
850 t--;
851 if (t == name)
852 *t++ = '.';
853 *t = '\0';
855 return t - name;
858 /* Make path appear as if a chroot had occurred. This handles a leading
859 * "/" (either removing it or expanding it) and any leading or embedded
860 * ".." components that attempt to escape past the module's top dir.
862 * If dest is NULL, a buffer is allocated to hold the result. It is legal
863 * to call with the dest and the path (p) pointing to the same buffer, but
864 * rootdir will be ignored to avoid expansion of the string.
866 * The rootdir string contains a value to use in place of a leading slash.
867 * Specify NULL to get the default of "module_dir".
869 * The depth var is a count of how many '..'s to allow at the start of the
870 * path.
872 * We also clean the path in a manner similar to clean_fname() but with a
873 * few differences:
875 * Turns multiple adjacent slashes into a single slash, gets rid of "." dir
876 * elements (INCLUDING a trailing dot dir), PRESERVES a trailing slash, and
877 * ALWAYS collapses ".." elements (except for those at the start of the
878 * string up to "depth" deep). If the resulting name would be empty,
879 * change it into a ".". */
880 char *sanitize_path(char *dest, const char *p, const char *rootdir, int depth,
881 int flags)
883 char *start, *sanp;
884 int rlen = 0, drop_dot_dirs = !relative_paths || !(flags & SP_KEEP_DOT_DIRS);
886 if (dest != p) {
887 int plen = strlen(p);
888 if (*p == '/') {
889 if (!rootdir)
890 rootdir = module_dir;
891 rlen = strlen(rootdir);
892 depth = 0;
893 p++;
895 if (dest) {
896 if (rlen + plen + 1 >= MAXPATHLEN)
897 return NULL;
898 } else if (!(dest = new_array(char, rlen + plen + 1)))
899 out_of_memory("sanitize_path");
900 if (rlen) {
901 memcpy(dest, rootdir, rlen);
902 if (rlen > 1)
903 dest[rlen++] = '/';
907 if (drop_dot_dirs) {
908 while (*p == '.' && p[1] == '/')
909 p += 2;
912 start = sanp = dest + rlen;
913 /* This loop iterates once per filename component in p, pointing at
914 * the start of the name (past any prior slash) for each iteration. */
915 while (*p) {
916 /* discard leading or extra slashes */
917 if (*p == '/') {
918 p++;
919 continue;
921 if (drop_dot_dirs) {
922 if (*p == '.' && (p[1] == '/' || p[1] == '\0')) {
923 /* skip "." component */
924 p++;
925 continue;
928 if (*p == '.' && p[1] == '.' && (p[2] == '/' || p[2] == '\0')) {
929 /* ".." component followed by slash or end */
930 if (depth <= 0 || sanp != start) {
931 p += 2;
932 if (sanp != start) {
933 /* back up sanp one level */
934 --sanp; /* now pointing at slash */
935 while (sanp > start && sanp[-1] != '/')
936 sanp--;
938 continue;
940 /* allow depth levels of .. at the beginning */
941 depth--;
942 /* move the virtual beginning to leave the .. alone */
943 start = sanp + 3;
945 /* copy one component through next slash */
946 while (*p && (*sanp++ = *p++) != '/') {}
948 if (sanp == dest) {
949 /* ended up with nothing, so put in "." component */
950 *sanp++ = '.';
952 *sanp = '\0';
954 return dest;
957 /* Like chdir(), but it keeps track of the current directory (in the
958 * global "curr_dir"), and ensures that the path size doesn't overflow.
959 * Also cleans the path using the clean_fname() function. */
960 int push_dir(const char *dir, int set_path_only)
962 static int initialised;
963 unsigned int len;
965 if (!initialised) {
966 initialised = 1;
967 getcwd(curr_dir, sizeof curr_dir - 1);
968 curr_dir_len = strlen(curr_dir);
971 if (!dir) /* this call was probably just to initialize */
972 return 0;
974 len = strlen(dir);
975 if (len == 1 && *dir == '.')
976 return 1;
978 if ((*dir == '/' ? len : curr_dir_len + 1 + len) >= sizeof curr_dir) {
979 errno = ENAMETOOLONG;
980 return 0;
983 if (!set_path_only && chdir(dir))
984 return 0;
986 if (*dir == '/') {
987 memcpy(curr_dir, dir, len + 1);
988 curr_dir_len = len;
989 } else {
990 curr_dir[curr_dir_len++] = '/';
991 memcpy(curr_dir + curr_dir_len, dir, len + 1);
992 curr_dir_len += len;
995 curr_dir_len = clean_fname(curr_dir, CFN_COLLAPSE_DOT_DOT_DIRS);
996 if (sanitize_paths) {
997 if (module_dirlen > curr_dir_len)
998 module_dirlen = curr_dir_len;
999 curr_dir_depth = count_dir_elements(curr_dir + module_dirlen);
1002 if (verbose >= 5 && !set_path_only)
1003 rprintf(FINFO, "[%s] push_dir(%s)\n", who_am_i(), curr_dir);
1005 return 1;
1009 * Reverse a push_dir() call. You must pass in an absolute path
1010 * that was copied from a prior value of "curr_dir".
1012 int pop_dir(const char *dir)
1014 if (chdir(dir))
1015 return 0;
1017 curr_dir_len = strlcpy(curr_dir, dir, sizeof curr_dir);
1018 if (curr_dir_len >= sizeof curr_dir)
1019 curr_dir_len = sizeof curr_dir - 1;
1020 if (sanitize_paths)
1021 curr_dir_depth = count_dir_elements(curr_dir + module_dirlen);
1023 if (verbose >= 5)
1024 rprintf(FINFO, "[%s] pop_dir(%s)\n", who_am_i(), curr_dir);
1026 return 1;
1030 * Return a quoted string with the full pathname of the indicated filename.
1031 * The string " (in MODNAME)" may also be appended. The returned pointer
1032 * remains valid until the next time full_fname() is called.
1034 char *full_fname(const char *fn)
1036 static char *result = NULL;
1037 char *m1, *m2, *m3;
1038 char *p1, *p2;
1040 if (result)
1041 free(result);
1043 if (*fn == '/')
1044 p1 = p2 = "";
1045 else {
1046 p1 = curr_dir + module_dirlen;
1047 for (p2 = p1; *p2 == '/'; p2++) {}
1048 if (*p2)
1049 p2 = "/";
1051 if (module_id >= 0) {
1052 m1 = " (in ";
1053 m2 = lp_name(module_id);
1054 m3 = ")";
1055 } else
1056 m1 = m2 = m3 = "";
1058 if (asprintf(&result, "\"%s%s%s\"%s%s%s", p1, p2, fn, m1, m2, m3) <= 0)
1059 out_of_memory("full_fname");
1061 return result;
1064 static char partial_fname[MAXPATHLEN];
1066 char *partial_dir_fname(const char *fname)
1068 char *t = partial_fname;
1069 int sz = sizeof partial_fname;
1070 const char *fn;
1072 if ((fn = strrchr(fname, '/')) != NULL) {
1073 fn++;
1074 if (*partial_dir != '/') {
1075 int len = fn - fname;
1076 strncpy(t, fname, len); /* safe */
1077 t += len;
1078 sz -= len;
1080 } else
1081 fn = fname;
1082 if ((int)pathjoin(t, sz, partial_dir, fn) >= sz)
1083 return NULL;
1084 if (daemon_filter_list.head) {
1085 t = strrchr(partial_fname, '/');
1086 *t = '\0';
1087 if (check_filter(&daemon_filter_list, partial_fname, 1) < 0)
1088 return NULL;
1089 *t = '/';
1090 if (check_filter(&daemon_filter_list, partial_fname, 0) < 0)
1091 return NULL;
1094 return partial_fname;
1097 /* If no --partial-dir option was specified, we don't need to do anything
1098 * (the partial-dir is essentially '.'), so just return success. */
1099 int handle_partial_dir(const char *fname, int create)
1101 char *fn, *dir;
1103 if (fname != partial_fname)
1104 return 1;
1105 if (!create && *partial_dir == '/')
1106 return 1;
1107 if (!(fn = strrchr(partial_fname, '/')))
1108 return 1;
1110 *fn = '\0';
1111 dir = partial_fname;
1112 if (create) {
1113 STRUCT_STAT st;
1114 int statret = do_lstat(dir, &st);
1115 if (statret == 0 && !S_ISDIR(st.st_mode)) {
1116 if (do_unlink(dir) < 0)
1117 return 0;
1118 statret = -1;
1120 if (statret < 0 && do_mkdir(dir, 0700) < 0)
1121 return 0;
1122 } else
1123 do_rmdir(dir);
1124 *fn = '/';
1126 return 1;
1130 * Determine if a symlink points outside the current directory tree.
1131 * This is considered "unsafe" because e.g. when mirroring somebody
1132 * else's machine it might allow them to establish a symlink to
1133 * /etc/passwd, and then read it through a web server.
1135 * Null symlinks and absolute symlinks are always unsafe.
1137 * Basically here we are concerned with symlinks whose target contains
1138 * "..", because this might cause us to walk back up out of the
1139 * transferred directory. We are not allowed to go back up and
1140 * reenter.
1142 * @param dest Target of the symlink in question.
1144 * @param src Top source directory currently applicable. Basically this
1145 * is the first parameter to rsync in a simple invocation, but it's
1146 * modified by flist.c in slightly complex ways.
1148 * @retval True if unsafe
1149 * @retval False is unsafe
1151 * @sa t_unsafe.c
1153 int unsafe_symlink(const char *dest, const char *src)
1155 const char *name, *slash;
1156 int depth = 0;
1158 /* all absolute and null symlinks are unsafe */
1159 if (!dest || !*dest || *dest == '/')
1160 return 1;
1162 /* find out what our safety margin is */
1163 for (name = src; (slash = strchr(name, '/')) != 0; name = slash+1) {
1164 if (strncmp(name, "../", 3) == 0) {
1165 depth = 0;
1166 } else if (strncmp(name, "./", 2) == 0) {
1167 /* nothing */
1168 } else {
1169 depth++;
1172 if (strcmp(name, "..") == 0)
1173 depth = 0;
1175 for (name = dest; (slash = strchr(name, '/')) != 0; name = slash+1) {
1176 if (strncmp(name, "../", 3) == 0) {
1177 /* if at any point we go outside the current directory
1178 then stop - it is unsafe */
1179 if (--depth < 0)
1180 return 1;
1181 } else if (strncmp(name, "./", 2) == 0) {
1182 /* nothing */
1183 } else {
1184 depth++;
1187 if (strcmp(name, "..") == 0)
1188 depth--;
1190 return (depth < 0);
1193 /* Return the int64 number as a string. If the --human-readable option was
1194 * specified, we may output the number in K, M, or G units. We can return
1195 * up to 4 buffers at a time. */
1196 char *human_num(int64 num)
1198 static char bufs[4][128]; /* more than enough room */
1199 static unsigned int n;
1200 char *s;
1202 n = (n + 1) % (sizeof bufs / sizeof bufs[0]);
1204 if (human_readable) {
1205 char units = '\0';
1206 int mult = human_readable == 1 ? 1000 : 1024;
1207 double dnum = 0;
1208 if (num > mult*mult*mult) {
1209 dnum = (double)num / (mult*mult*mult);
1210 units = 'G';
1211 } else if (num > mult*mult) {
1212 dnum = (double)num / (mult*mult);
1213 units = 'M';
1214 } else if (num > mult) {
1215 dnum = (double)num / mult;
1216 units = 'K';
1218 if (units) {
1219 snprintf(bufs[n], sizeof bufs[0], "%.2f%c", dnum, units);
1220 return bufs[n];
1224 s = bufs[n] + sizeof bufs[0] - 1;
1225 *s = '\0';
1227 if (!num)
1228 *--s = '0';
1229 while (num) {
1230 *--s = (char)(num % 10) + '0';
1231 num /= 10;
1233 return s;
1236 /* Return the double number as a string. If the --human-readable option was
1237 * specified, we may output the number in K, M, or G units. We use a buffer
1238 * from human_num() to return our result. */
1239 char *human_dnum(double dnum, int decimal_digits)
1241 char *buf = human_num(dnum);
1242 int len = strlen(buf);
1243 if (isDigit(buf + len - 1)) {
1244 /* There's extra room in buf prior to the start of the num. */
1245 buf -= decimal_digits + 1;
1246 snprintf(buf, len + decimal_digits + 2, "%.*f", decimal_digits, dnum);
1248 return buf;
1251 /* Return the date and time as a string. Some callers tweak returned buf. */
1252 char *timestring(time_t t)
1254 static char TimeBuf[200];
1255 struct tm *tm = localtime(&t);
1256 char *p;
1258 #ifdef HAVE_STRFTIME
1259 strftime(TimeBuf, sizeof TimeBuf - 1, "%Y/%m/%d %H:%M:%S", tm);
1260 #else
1261 strlcpy(TimeBuf, asctime(tm), sizeof TimeBuf);
1262 #endif
1264 if ((p = strchr(TimeBuf, '\n')) != NULL)
1265 *p = '\0';
1267 return TimeBuf;
1271 * Sleep for a specified number of milliseconds.
1273 * Always returns TRUE. (In the future it might return FALSE if
1274 * interrupted.)
1276 int msleep(int t)
1278 int tdiff = 0;
1279 struct timeval tval, t1, t2;
1281 gettimeofday(&t1, NULL);
1283 while (tdiff < t) {
1284 tval.tv_sec = (t-tdiff)/1000;
1285 tval.tv_usec = 1000*((t-tdiff)%1000);
1287 errno = 0;
1288 select(0,NULL,NULL, NULL, &tval);
1290 gettimeofday(&t2, NULL);
1291 tdiff = (t2.tv_sec - t1.tv_sec)*1000 +
1292 (t2.tv_usec - t1.tv_usec)/1000;
1295 return True;
1298 /* Determine if two time_t values are equivalent (either exact, or in
1299 * the modification timestamp window established by --modify-window).
1301 * @retval 0 if the times should be treated as the same
1303 * @retval +1 if the first is later
1305 * @retval -1 if the 2nd is later
1307 int cmp_time(time_t file1, time_t file2)
1309 if (file2 > file1) {
1310 if (file2 - file1 <= modify_window)
1311 return 0;
1312 return -1;
1314 if (file1 - file2 <= modify_window)
1315 return 0;
1316 return 1;
1320 #ifdef __INSURE__XX
1321 #include <dlfcn.h>
1324 This routine is a trick to immediately catch errors when debugging
1325 with insure. A xterm with a gdb is popped up when insure catches
1326 a error. It is Linux specific.
1328 int _Insure_trap_error(int a1, int a2, int a3, int a4, int a5, int a6)
1330 static int (*fn)();
1331 int ret;
1332 char *cmd;
1334 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'",
1335 getpid(), getpid(), getpid());
1337 if (!fn) {
1338 static void *h;
1339 h = dlopen("/usr/local/parasoft/insure++lite/lib.linux2/libinsure.so", RTLD_LAZY);
1340 fn = dlsym(h, "_Insure_trap_error");
1343 ret = fn(a1, a2, a3, a4, a5, a6);
1345 system(cmd);
1347 free(cmd);
1349 return ret;
1351 #endif
1353 #define MALLOC_MAX 0x40000000
1355 void *_new_array(unsigned long num, unsigned int size, int use_calloc)
1357 if (num >= MALLOC_MAX/size)
1358 return NULL;
1359 return use_calloc ? calloc(num, size) : malloc(num * size);
1362 void *_realloc_array(void *ptr, unsigned int size, unsigned long num)
1364 if (num >= MALLOC_MAX/size)
1365 return NULL;
1366 if (!ptr)
1367 return malloc(size * num);
1368 return realloc(ptr, size * num);
1371 /* Take a filename and filename length and return the most significant
1372 * filename suffix we can find. This ignores suffixes such as "~",
1373 * ".bak", ".orig", ".~1~", etc. */
1374 const char *find_filename_suffix(const char *fn, int fn_len, int *len_ptr)
1376 const char *suf, *s;
1377 BOOL had_tilde;
1378 int s_len;
1380 /* One or more dots at the start aren't a suffix. */
1381 while (fn_len && *fn == '.') fn++, fn_len--;
1383 /* Ignore the ~ in a "foo~" filename. */
1384 if (fn_len > 1 && fn[fn_len-1] == '~')
1385 fn_len--, had_tilde = True;
1386 else
1387 had_tilde = False;
1389 /* Assume we don't find an suffix. */
1390 suf = "";
1391 *len_ptr = 0;
1393 /* Find the last significant suffix. */
1394 for (s = fn + fn_len; fn_len > 1; ) {
1395 while (*--s != '.' && s != fn) {}
1396 if (s == fn)
1397 break;
1398 s_len = fn_len - (s - fn);
1399 fn_len = s - fn;
1400 if (s_len == 4) {
1401 if (strcmp(s+1, "bak") == 0
1402 || strcmp(s+1, "old") == 0)
1403 continue;
1404 } else if (s_len == 5) {
1405 if (strcmp(s+1, "orig") == 0)
1406 continue;
1407 } else if (s_len > 2 && had_tilde
1408 && s[1] == '~' && isDigit(s + 2))
1409 continue;
1410 *len_ptr = s_len;
1411 suf = s;
1412 if (s_len == 1)
1413 break;
1414 /* Determine if the suffix is all digits. */
1415 for (s++, s_len--; s_len > 0; s++, s_len--) {
1416 if (!isDigit(s))
1417 return suf;
1419 /* An all-digit suffix may not be that signficant. */
1420 s = suf;
1423 return suf;
1426 /* This is an implementation of the Levenshtein distance algorithm. It
1427 * was implemented to avoid needing a two-dimensional matrix (to save
1428 * memory). It was also tweaked to try to factor in the ASCII distance
1429 * between changed characters as a minor distance quantity. The normal
1430 * Levenshtein units of distance (each signifying a single change between
1431 * the two strings) are defined as a "UNIT". */
1433 #define UNIT (1 << 16)
1435 uint32 fuzzy_distance(const char *s1, int len1, const char *s2, int len2)
1437 uint32 a[MAXPATHLEN], diag, above, left, diag_inc, above_inc, left_inc;
1438 int32 cost;
1439 int i1, i2;
1441 if (!len1 || !len2) {
1442 if (!len1) {
1443 s1 = s2;
1444 len1 = len2;
1446 for (i1 = 0, cost = 0; i1 < len1; i1++)
1447 cost += s1[i1];
1448 return (int32)len1 * UNIT + cost;
1451 for (i2 = 0; i2 < len2; i2++)
1452 a[i2] = (i2+1) * UNIT;
1454 for (i1 = 0; i1 < len1; i1++) {
1455 diag = i1 * UNIT;
1456 above = (i1+1) * UNIT;
1457 for (i2 = 0; i2 < len2; i2++) {
1458 left = a[i2];
1459 if ((cost = *((uchar*)s1+i1) - *((uchar*)s2+i2)) != 0) {
1460 if (cost < 0)
1461 cost = UNIT - cost;
1462 else
1463 cost = UNIT + cost;
1465 diag_inc = diag + cost;
1466 left_inc = left + UNIT + *((uchar*)s1+i1);
1467 above_inc = above + UNIT + *((uchar*)s2+i2);
1468 a[i2] = above = left < above
1469 ? (left_inc < diag_inc ? left_inc : diag_inc)
1470 : (above_inc < diag_inc ? above_inc : diag_inc);
1471 diag = left;
1475 return a[len2-1];
1478 #define BB_SLOT_SIZE (16*1024) /* Desired size in bytes */
1479 #define BB_PER_SLOT_BITS (BB_SLOT_SIZE * 8) /* Number of bits per slot */
1480 #define BB_PER_SLOT_INTS (BB_SLOT_SIZE / 4) /* Number of int32s per slot */
1482 struct bitbag {
1483 uint32 **bits;
1484 int slot_cnt;
1487 struct bitbag *bitbag_create(int max_ndx)
1489 struct bitbag *bb = new(struct bitbag);
1490 bb->slot_cnt = (max_ndx + BB_PER_SLOT_BITS - 1) / BB_PER_SLOT_BITS;
1492 if (!(bb->bits = (uint32**)calloc(bb->slot_cnt, sizeof (uint32*))))
1493 out_of_memory("bitbag_create");
1495 return bb;
1498 void bitbag_set_bit(struct bitbag *bb, int ndx)
1500 int slot = ndx / BB_PER_SLOT_BITS;
1501 ndx %= BB_PER_SLOT_BITS;
1503 if (!bb->bits[slot]) {
1504 if (!(bb->bits[slot] = (uint32*)calloc(BB_PER_SLOT_INTS, 4)))
1505 out_of_memory("bitbag_set_bit");
1508 bb->bits[slot][ndx/32] |= 1u << (ndx % 32);
1511 #if 0 /* not needed yet */
1512 void bitbag_clear_bit(struct bitbag *bb, int ndx)
1514 int slot = ndx / BB_PER_SLOT_BITS;
1515 ndx %= BB_PER_SLOT_BITS;
1517 if (!bb->bits[slot])
1518 return;
1520 bb->bits[slot][ndx/32] &= ~(1u << (ndx % 32));
1523 int bitbag_check_bit(struct bitbag *bb, int ndx)
1525 int slot = ndx / BB_PER_SLOT_BITS;
1526 ndx %= BB_PER_SLOT_BITS;
1528 if (!bb->bits[slot])
1529 return 0;
1531 return bb->bits[slot][ndx/32] & (1u << (ndx % 32)) ? 1 : 0;
1533 #endif
1535 /* Call this with -1 to start checking from 0. Returns -1 at the end. */
1536 int bitbag_next_bit(struct bitbag *bb, int after)
1538 uint32 bits, mask;
1539 int i, ndx = after + 1;
1540 int slot = ndx / BB_PER_SLOT_BITS;
1541 ndx %= BB_PER_SLOT_BITS;
1543 mask = (1u << (ndx % 32)) - 1;
1544 for (i = ndx / 32; slot < bb->slot_cnt; slot++, i = mask = 0) {
1545 if (!bb->bits[slot])
1546 continue;
1547 for ( ; i < BB_PER_SLOT_INTS; i++, mask = 0) {
1548 if (!(bits = bb->bits[slot][i] & ~mask))
1549 continue;
1550 /* The xor magic figures out the lowest enabled bit in
1551 * bits, and the switch quickly computes log2(bit). */
1552 switch (bits ^ (bits & (bits-1))) {
1553 #define LOG2(n) case 1u << n: return slot*BB_PER_SLOT_BITS + i*32 + n
1554 LOG2(0); LOG2(1); LOG2(2); LOG2(3);
1555 LOG2(4); LOG2(5); LOG2(6); LOG2(7);
1556 LOG2(8); LOG2(9); LOG2(10); LOG2(11);
1557 LOG2(12); LOG2(13); LOG2(14); LOG2(15);
1558 LOG2(16); LOG2(17); LOG2(18); LOG2(19);
1559 LOG2(20); LOG2(21); LOG2(22); LOG2(23);
1560 LOG2(24); LOG2(25); LOG2(26); LOG2(27);
1561 LOG2(28); LOG2(29); LOG2(30); LOG2(31);
1563 return -1; /* impossible... */
1567 return -1;
1570 void *expand_item_list(item_list *lp, size_t item_size,
1571 const char *desc, int incr)
1573 /* First time through, 0 <= 0, so list is expanded. */
1574 if (lp->malloced <= lp->count) {
1575 void *new_ptr;
1576 size_t new_size = lp->malloced;
1577 if (incr < 0)
1578 new_size += -incr; /* increase slowly */
1579 else if (new_size < (size_t)incr)
1580 new_size += incr;
1581 else
1582 new_size *= 2;
1583 new_ptr = realloc_array(lp->items, char, new_size * item_size);
1584 if (verbose >= 4) {
1585 rprintf(FINFO, "[%s] expand %s to %.0f bytes, did%s move\n",
1586 who_am_i(), desc, (double)new_size * item_size,
1587 new_ptr == lp->items ? " not" : "");
1589 if (!new_ptr)
1590 out_of_memory("expand_item_list");
1592 lp->items = new_ptr;
1593 lp->malloced = new_size;
1595 return (char*)lp->items + (lp->count++ * item_size);