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