]> git.ipfire.org Git - thirdparty/rsync.git/blob - util1.c
More tweaks for Actions.
[thirdparty/rsync.git] / util1.c
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) 2003-2022 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 #include "itypes.h"
26 #include "inums.h"
27
28 extern int dry_run;
29 extern int module_id;
30 extern int do_fsync;
31 extern int protect_args;
32 extern int modify_window;
33 extern int relative_paths;
34 extern int preserve_xattrs;
35 extern int omit_link_times;
36 extern int preallocate_files;
37 extern char *module_dir;
38 extern unsigned int module_dirlen;
39 extern char *partial_dir;
40 extern filter_rule_list daemon_filter_list;
41
42 int sanitize_paths = 0;
43
44 char curr_dir[MAXPATHLEN];
45 unsigned int curr_dir_len;
46 int curr_dir_depth; /* This is only set for a sanitizing daemon. */
47
48 /* Set a fd into nonblocking mode. */
49 void set_nonblocking(int fd)
50 {
51 int val;
52
53 if ((val = fcntl(fd, F_GETFL)) == -1)
54 return;
55 if (!(val & NONBLOCK_FLAG)) {
56 val |= NONBLOCK_FLAG;
57 fcntl(fd, F_SETFL, val);
58 }
59 }
60
61 /* Set a fd into blocking mode. */
62 void set_blocking(int fd)
63 {
64 int val;
65
66 if ((val = fcntl(fd, F_GETFL)) == -1)
67 return;
68 if (val & NONBLOCK_FLAG) {
69 val &= ~NONBLOCK_FLAG;
70 fcntl(fd, F_SETFL, val);
71 }
72 }
73
74 /**
75 * Create a file descriptor pair - like pipe() but use socketpair if
76 * possible (because of blocking issues on pipes).
77 *
78 * Always set non-blocking.
79 */
80 int fd_pair(int fd[2])
81 {
82 int ret;
83
84 #ifdef HAVE_SOCKETPAIR
85 ret = socketpair(AF_UNIX, SOCK_STREAM, 0, fd);
86 #else
87 ret = pipe(fd);
88 #endif
89
90 if (ret == 0) {
91 set_nonblocking(fd[0]);
92 set_nonblocking(fd[1]);
93 }
94
95 return ret;
96 }
97
98 void print_child_argv(const char *prefix, char **cmd)
99 {
100 int cnt = 0;
101 rprintf(FCLIENT, "%s ", prefix);
102 for (; *cmd; cmd++) {
103 /* Look for characters that ought to be quoted. This
104 * is not a great quoting algorithm, but it's
105 * sufficient for a log message. */
106 if (strspn(*cmd, "abcdefghijklmnopqrstuvwxyz"
107 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
108 "0123456789"
109 ",.-_=+@/") != strlen(*cmd)) {
110 rprintf(FCLIENT, "\"%s\" ", *cmd);
111 } else {
112 rprintf(FCLIENT, "%s ", *cmd);
113 }
114 cnt++;
115 }
116 rprintf(FCLIENT, " (%d args)\n", cnt);
117 }
118
119 /* This returns 0 for success, 1 for a symlink if symlink time-setting
120 * is not possible, or -1 for any other error. */
121 int set_times(const char *fname, STRUCT_STAT *stp)
122 {
123 static int switch_step = 0;
124
125 if (DEBUG_GTE(TIME, 1)) {
126 rprintf(FINFO,
127 "set modtime, atime of %s to (%ld) %s, (%ld) %s\n",
128 fname, (long)stp->st_mtime,
129 timestring(stp->st_mtime), (long)stp->st_atime, timestring(stp->st_atime));
130 }
131
132 switch (switch_step) {
133 #ifdef HAVE_SETATTRLIST
134 #include "case_N.h"
135 if (do_setattrlist_times(fname, stp) == 0)
136 break;
137 if (errno != ENOSYS)
138 return -1;
139 switch_step++;
140 #endif
141
142 #ifdef HAVE_UTIMENSAT
143 #include "case_N.h"
144 if (do_utimensat(fname, stp) == 0)
145 break;
146 if (errno != ENOSYS)
147 return -1;
148 switch_step++;
149 #endif
150
151 #ifdef HAVE_LUTIMES
152 #include "case_N.h"
153 if (do_lutimes(fname, stp) == 0)
154 break;
155 if (errno != ENOSYS)
156 return -1;
157 switch_step++;
158 #endif
159
160 #include "case_N.h"
161 switch_step++;
162 if (!omit_link_times) {
163 omit_link_times = 1;
164 if (S_ISLNK(stp->st_mode))
165 return 1;
166 }
167
168 #include "case_N.h"
169 #ifdef HAVE_UTIMES
170 if (do_utimes(fname, stp) == 0)
171 break;
172 #else
173 if (do_utime(fname, stp) == 0)
174 break;
175 #endif
176
177 return -1;
178 }
179
180 return 0;
181 }
182
183 /* Create any necessary directories in fname. Any missing directories are
184 * created with default permissions. Returns < 0 on error, or the number
185 * of directories created. */
186 int make_path(char *fname, int flags)
187 {
188 char *end, *p;
189 int ret = 0;
190
191 if (flags & MKP_SKIP_SLASH) {
192 while (*fname == '/')
193 fname++;
194 }
195
196 while (*fname == '.' && fname[1] == '/')
197 fname += 2;
198
199 if (flags & MKP_DROP_NAME) {
200 end = strrchr(fname, '/');
201 if (!end || end == fname)
202 return 0;
203 *end = '\0';
204 } else
205 end = fname + strlen(fname);
206
207 /* Try to find an existing dir, starting from the deepest dir. */
208 for (p = end; ; ) {
209 if (dry_run) {
210 STRUCT_STAT st;
211 if (do_stat(fname, &st) == 0) {
212 if (S_ISDIR(st.st_mode))
213 errno = EEXIST;
214 else
215 errno = ENOTDIR;
216 }
217 } else if (do_mkdir(fname, ACCESSPERMS) == 0) {
218 ret++;
219 break;
220 }
221
222 if (errno != ENOENT) {
223 STRUCT_STAT st;
224 if (errno != EEXIST || (do_stat(fname, &st) == 0 && !S_ISDIR(st.st_mode)))
225 ret = -ret - 1;
226 break;
227 }
228 while (1) {
229 if (p == fname) {
230 /* We got a relative path that doesn't exist, so assume that '.'
231 * is there and just break out and create the whole thing. */
232 p = NULL;
233 goto double_break;
234 }
235 if (*--p == '/') {
236 if (p == fname) {
237 /* We reached the "/" dir, which we assume is there. */
238 goto double_break;
239 }
240 *p = '\0';
241 break;
242 }
243 }
244 }
245 double_break:
246
247 /* Make all the dirs that we didn't find on the way here. */
248 while (p != end) {
249 if (p)
250 *p = '/';
251 else
252 p = fname;
253 p += strlen(p);
254 if (ret < 0) /* Skip mkdir on error, but keep restoring the path. */
255 continue;
256 if (do_mkdir(fname, ACCESSPERMS) < 0)
257 ret = -ret - 1;
258 else
259 ret++;
260 }
261
262 if (flags & MKP_DROP_NAME)
263 *end = '/';
264
265 return ret;
266 }
267
268 /**
269 * Write @p len bytes at @p ptr to descriptor @p desc, retrying if
270 * interrupted.
271 *
272 * @retval len upon success
273 *
274 * @retval <0 write's (negative) error code
275 *
276 * Derived from GNU C's cccp.c.
277 */
278 int full_write(int desc, const char *ptr, size_t len)
279 {
280 int total_written;
281
282 total_written = 0;
283 while (len > 0) {
284 int written = write(desc, ptr, len);
285 if (written < 0) {
286 if (errno == EINTR)
287 continue;
288 return written;
289 }
290 total_written += written;
291 ptr += written;
292 len -= written;
293 }
294 return total_written;
295 }
296
297 /**
298 * Read @p len bytes at @p ptr from descriptor @p desc, retrying if
299 * interrupted.
300 *
301 * @retval >0 the actual number of bytes read
302 *
303 * @retval 0 for EOF
304 *
305 * @retval <0 for an error.
306 *
307 * Derived from GNU C's cccp.c. */
308 static int safe_read(int desc, char *ptr, size_t len)
309 {
310 int n_chars;
311
312 if (len == 0)
313 return len;
314
315 do {
316 n_chars = read(desc, ptr, len);
317 } while (n_chars < 0 && errno == EINTR);
318
319 return n_chars;
320 }
321
322 /* Remove existing file @dest and reopen, creating a new file with @mode */
323 static int unlink_and_reopen(const char *dest, mode_t mode)
324 {
325 int ofd;
326
327 if (robust_unlink(dest) && errno != ENOENT) {
328 int save_errno = errno;
329 rsyserr(FERROR_XFER, errno, "unlink %s", full_fname(dest));
330 errno = save_errno;
331 return -1;
332 }
333
334 #ifdef SUPPORT_XATTRS
335 if (preserve_xattrs)
336 mode |= S_IWUSR;
337 #endif
338 mode &= INITACCESSPERMS;
339 if ((ofd = do_open(dest, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, mode)) < 0) {
340 int save_errno = errno;
341 rsyserr(FERROR_XFER, save_errno, "open %s", full_fname(dest));
342 errno = save_errno;
343 return -1;
344 }
345 return ofd;
346 }
347
348 /* Copy contents of file @source to file @dest with mode @mode.
349 *
350 * If @tmpfilefd is < 0, copy_file unlinks @dest and then opens a new
351 * file with name @dest.
352 *
353 * Otherwise, copy_file writes to and closes the provided file
354 * descriptor.
355 *
356 * In either case, if --xattrs are being preserved, the dest file will
357 * have its xattrs set from the source file.
358 *
359 * This is used in conjunction with the --temp-dir, --backup, and
360 * --copy-dest options. */
361 int copy_file(const char *source, const char *dest, int tmpfilefd, mode_t mode)
362 {
363 int ifd, ofd;
364 char buf[1024 * 8];
365 int len; /* Number of bytes read into `buf'. */
366 OFF_T prealloc_len = 0, offset = 0;
367
368 if ((ifd = do_open(source, O_RDONLY, 0)) < 0) {
369 int save_errno = errno;
370 rsyserr(FERROR_XFER, errno, "open %s", full_fname(source));
371 errno = save_errno;
372 return -1;
373 }
374
375 if (tmpfilefd >= 0) {
376 ofd = tmpfilefd;
377 } else {
378 ofd = unlink_and_reopen(dest, mode);
379 if (ofd < 0) {
380 int save_errno = errno;
381 close(ifd);
382 errno = save_errno;
383 return -1;
384 }
385 }
386
387 #ifdef SUPPORT_PREALLOCATION
388 if (preallocate_files) {
389 STRUCT_STAT srcst;
390
391 /* Try to preallocate enough space for file's eventual length. Can
392 * reduce fragmentation on filesystems like ext4, xfs, and NTFS. */
393 if (do_fstat(ifd, &srcst) < 0)
394 rsyserr(FWARNING, errno, "fstat %s", full_fname(source));
395 else if (srcst.st_size > 0) {
396 prealloc_len = do_fallocate(ofd, 0, srcst.st_size);
397 if (prealloc_len < 0)
398 rsyserr(FWARNING, errno, "do_fallocate %s", full_fname(dest));
399 }
400 }
401 #endif
402
403 while ((len = safe_read(ifd, buf, sizeof buf)) > 0) {
404 if (full_write(ofd, buf, len) < 0) {
405 int save_errno = errno;
406 rsyserr(FERROR_XFER, errno, "write %s", full_fname(dest));
407 close(ifd);
408 close(ofd);
409 errno = save_errno;
410 return -1;
411 }
412 offset += len;
413 }
414
415 if (len < 0) {
416 int save_errno = errno;
417 rsyserr(FERROR_XFER, errno, "read %s", full_fname(source));
418 close(ifd);
419 close(ofd);
420 errno = save_errno;
421 return -1;
422 }
423
424 if (close(ifd) < 0) {
425 rsyserr(FWARNING, errno, "close failed on %s",
426 full_fname(source));
427 }
428
429 /* Source file might have shrunk since we fstatted it.
430 * Cut off any extra preallocated zeros from dest file. */
431 if (offset < prealloc_len) {
432 #ifdef HAVE_FTRUNCATE
433 /* If we fail to truncate, the dest file may be wrong, so we
434 * must trigger the "partial transfer" error. */
435 if (do_ftruncate(ofd, offset) < 0)
436 rsyserr(FERROR_XFER, errno, "ftruncate %s", full_fname(dest));
437 #else
438 rprintf(FERROR_XFER, "no ftruncate for over-long pre-alloc: %s", full_fname(dest));
439 #endif
440 }
441
442 if (do_fsync && fsync(ofd) < 0) {
443 int save_errno = errno;
444 rsyserr(FERROR, errno, "fsync failed on %s", full_fname(dest));
445 close(ofd);
446 errno = save_errno;
447 return -1;
448 }
449
450 if (close(ofd) < 0) {
451 int save_errno = errno;
452 rsyserr(FERROR_XFER, errno, "close failed on %s", full_fname(dest));
453 errno = save_errno;
454 return -1;
455 }
456
457 #ifdef SUPPORT_XATTRS
458 if (preserve_xattrs)
459 copy_xattrs(source, dest);
460 #endif
461
462 return 0;
463 }
464
465 /* MAX_RENAMES should be 10**MAX_RENAMES_DIGITS */
466 #define MAX_RENAMES_DIGITS 3
467 #define MAX_RENAMES 1000
468
469 /**
470 * Robust unlink: some OS'es (HPUX) refuse to unlink busy files, so
471 * rename to <path>/.rsyncNNN instead.
472 *
473 * Note that successive rsync runs will shuffle the filenames around a
474 * bit as long as the file is still busy; this is because this function
475 * does not know if the unlink call is due to a new file coming in, or
476 * --delete trying to remove old .rsyncNNN files, hence it renames it
477 * each time.
478 **/
479 int robust_unlink(const char *fname)
480 {
481 #ifndef ETXTBSY
482 return do_unlink(fname);
483 #else
484 static int counter = 1;
485 int rc, pos, start;
486 char path[MAXPATHLEN];
487
488 rc = do_unlink(fname);
489 if (rc == 0 || errno != ETXTBSY)
490 return rc;
491
492 if ((pos = strlcpy(path, fname, MAXPATHLEN)) >= MAXPATHLEN)
493 pos = MAXPATHLEN - 1;
494
495 while (pos > 0 && path[pos-1] != '/')
496 pos--;
497 pos += strlcpy(path+pos, ".rsync", MAXPATHLEN-pos);
498
499 if (pos > (MAXPATHLEN-MAX_RENAMES_DIGITS-1)) {
500 errno = ETXTBSY;
501 return -1;
502 }
503
504 /* start where the last one left off to reduce chance of clashes */
505 start = counter;
506 do {
507 snprintf(&path[pos], MAX_RENAMES_DIGITS+1, "%03d", counter);
508 if (++counter >= MAX_RENAMES)
509 counter = 1;
510 } while ((rc = access(path, 0)) == 0 && counter != start);
511
512 if (INFO_GTE(MISC, 1)) {
513 rprintf(FWARNING, "renaming %s to %s because of text busy\n",
514 fname, path);
515 }
516
517 /* maybe we should return rename()'s exit status? Nah. */
518 if (do_rename(fname, path) != 0) {
519 errno = ETXTBSY;
520 return -1;
521 }
522 return 0;
523 #endif
524 }
525
526 /* Returns 0 on successful rename, 1 if we successfully copied the file
527 * across filesystems, -2 if copy_file() failed, and -1 on other errors.
528 * If partialptr is not NULL and we need to do a copy, copy the file into
529 * the active partial-dir instead of over the destination file. */
530 int robust_rename(const char *from, const char *to, const char *partialptr,
531 int mode)
532 {
533 int tries = 4;
534
535 /* A resumed in-place partial-dir transfer might call us with from and
536 * to pointing to the same buf if the transfer failed yet again. */
537 if (from == to)
538 return 0;
539
540 while (tries--) {
541 if (do_rename(from, to) == 0)
542 return 0;
543
544 switch (errno) {
545 #ifdef ETXTBSY
546 case ETXTBSY:
547 if (robust_unlink(to) != 0) {
548 errno = ETXTBSY;
549 return -1;
550 }
551 errno = ETXTBSY;
552 break;
553 #endif
554 case EXDEV:
555 if (partialptr) {
556 if (!handle_partial_dir(partialptr,PDIR_CREATE))
557 return -2;
558 to = partialptr;
559 }
560 if (copy_file(from, to, -1, mode) != 0)
561 return -2;
562 do_unlink(from);
563 return 1;
564 default:
565 return -1;
566 }
567 }
568 return -1;
569 }
570
571 static pid_t all_pids[10];
572 static int num_pids;
573
574 /** Fork and record the pid of the child. **/
575 pid_t do_fork(void)
576 {
577 pid_t newpid = fork();
578
579 if (newpid != 0 && newpid != -1) {
580 all_pids[num_pids++] = newpid;
581 }
582 return newpid;
583 }
584
585 /**
586 * Kill all children.
587 *
588 * @todo It would be kind of nice to make sure that they are actually
589 * all our children before we kill them, because their pids may have
590 * been recycled by some other process. Perhaps when we wait for a
591 * child, we should remove it from this array. Alternatively we could
592 * perhaps use process groups, but I think that would not work on
593 * ancient Unix versions that don't support them.
594 **/
595 void kill_all(int sig)
596 {
597 int i;
598
599 for (i = 0; i < num_pids; i++) {
600 /* Let's just be a little careful where we
601 * point that gun, hey? See kill(2) for the
602 * magic caused by negative values. */
603 pid_t p = all_pids[i];
604
605 if (p == getpid())
606 continue;
607 if (p <= 0)
608 continue;
609
610 kill(p, sig);
611 }
612 }
613
614 /** Lock a byte range in a open file */
615 int lock_range(int fd, int offset, int len)
616 {
617 struct flock lock;
618
619 lock.l_type = F_WRLCK;
620 lock.l_whence = SEEK_SET;
621 lock.l_start = offset;
622 lock.l_len = len;
623 lock.l_pid = 0;
624
625 return fcntl(fd,F_SETLK,&lock) == 0;
626 }
627
628 #define ENSURE_MEMSPACE(buf, type, sz, req) \
629 do { if ((req) > sz) buf = realloc_array(buf, type, sz = MAX(sz * 2, req)); } while(0)
630
631 static inline void call_glob_match(const char *name, int len, int from_glob,
632 char *arg, int abpos, int fbpos);
633
634 static struct glob_data {
635 char *arg_buf, *filt_buf, **argv;
636 int absize, fbsize, maxargs, argc;
637 } glob;
638
639 static void glob_match(char *arg, int abpos, int fbpos)
640 {
641 int len;
642 char *slash;
643
644 while (*arg == '.' && arg[1] == '/') {
645 if (fbpos < 0) {
646 ENSURE_MEMSPACE(glob.filt_buf, char, glob.fbsize, glob.absize);
647 memcpy(glob.filt_buf, glob.arg_buf, abpos + 1);
648 fbpos = abpos;
649 }
650 ENSURE_MEMSPACE(glob.arg_buf, char, glob.absize, abpos + 3);
651 glob.arg_buf[abpos++] = *arg++;
652 glob.arg_buf[abpos++] = *arg++;
653 glob.arg_buf[abpos] = '\0';
654 }
655 if ((slash = strchr(arg, '/')) != NULL) {
656 *slash = '\0';
657 len = slash - arg;
658 } else
659 len = strlen(arg);
660 if (strpbrk(arg, "*?[")) {
661 struct dirent *di;
662 DIR *d;
663
664 if (!(d = opendir(abpos ? glob.arg_buf : ".")))
665 return;
666 while ((di = readdir(d)) != NULL) {
667 char *dname = d_name(di);
668 if (dname[0] == '.' && (dname[1] == '\0'
669 || (dname[1] == '.' && dname[2] == '\0')))
670 continue;
671 if (!wildmatch(arg, dname))
672 continue;
673 call_glob_match(dname, strlen(dname), 1,
674 slash ? arg + len + 1 : NULL,
675 abpos, fbpos);
676 }
677 closedir(d);
678 } else {
679 call_glob_match(arg, len, 0,
680 slash ? arg + len + 1 : NULL,
681 abpos, fbpos);
682 }
683 if (slash)
684 *slash = '/';
685 }
686
687 static inline void call_glob_match(const char *name, int len, int from_glob,
688 char *arg, int abpos, int fbpos)
689 {
690 char *use_buf;
691
692 ENSURE_MEMSPACE(glob.arg_buf, char, glob.absize, abpos + len + 2);
693 memcpy(glob.arg_buf + abpos, name, len);
694 abpos += len;
695 glob.arg_buf[abpos] = '\0';
696
697 if (fbpos >= 0) {
698 ENSURE_MEMSPACE(glob.filt_buf, char, glob.fbsize, fbpos + len + 2);
699 memcpy(glob.filt_buf + fbpos, name, len);
700 fbpos += len;
701 glob.filt_buf[fbpos] = '\0';
702 use_buf = glob.filt_buf;
703 } else
704 use_buf = glob.arg_buf;
705
706 if (from_glob || (arg && len)) {
707 STRUCT_STAT st;
708 int is_dir;
709
710 if (do_stat(glob.arg_buf, &st) != 0)
711 return;
712 is_dir = S_ISDIR(st.st_mode) != 0;
713 if (arg && !is_dir)
714 return;
715
716 if (daemon_filter_list.head
717 && check_filter(&daemon_filter_list, FLOG, use_buf, is_dir) < 0)
718 return;
719 }
720
721 if (arg) {
722 glob.arg_buf[abpos++] = '/';
723 glob.arg_buf[abpos] = '\0';
724 if (fbpos >= 0) {
725 glob.filt_buf[fbpos++] = '/';
726 glob.filt_buf[fbpos] = '\0';
727 }
728 glob_match(arg, abpos, fbpos);
729 } else {
730 ENSURE_MEMSPACE(glob.argv, char *, glob.maxargs, glob.argc + 1);
731 glob.argv[glob.argc++] = strdup(glob.arg_buf);
732 }
733 }
734
735 /* This routine performs wild-card expansion of the pathname in "arg". Any
736 * daemon-excluded files/dirs will not be matched by the wildcards. Returns 0
737 * if a wild-card string is the only returned item (due to matching nothing). */
738 int glob_expand(const char *arg, char ***argv_p, int *argc_p, int *maxargs_p)
739 {
740 int ret, save_argc;
741 char *s;
742
743 if (!arg) {
744 if (glob.filt_buf)
745 free(glob.filt_buf);
746 free(glob.arg_buf);
747 memset(&glob, 0, sizeof glob);
748 return -1;
749 }
750
751 if (sanitize_paths)
752 s = sanitize_path(NULL, arg, "", 0, SP_KEEP_DOT_DIRS);
753 else {
754 s = strdup(arg);
755 clean_fname(s, CFN_KEEP_DOT_DIRS | CFN_KEEP_TRAILING_SLASH | CFN_COLLAPSE_DOT_DOT_DIRS);
756 }
757
758 ENSURE_MEMSPACE(glob.arg_buf, char, glob.absize, MAXPATHLEN);
759 *glob.arg_buf = '\0';
760
761 glob.argc = save_argc = *argc_p;
762 glob.argv = *argv_p;
763 glob.maxargs = *maxargs_p;
764
765 ENSURE_MEMSPACE(glob.argv, char *, glob.maxargs, 100);
766
767 glob_match(s, 0, -1);
768
769 /* The arg didn't match anything, so add the failed arg to the list. */
770 if (glob.argc == save_argc) {
771 ENSURE_MEMSPACE(glob.argv, char *, glob.maxargs, glob.argc + 1);
772 glob.argv[glob.argc++] = s;
773 ret = 0;
774 } else {
775 free(s);
776 ret = 1;
777 }
778
779 *maxargs_p = glob.maxargs;
780 *argv_p = glob.argv;
781 *argc_p = glob.argc;
782
783 return ret;
784 }
785
786 /* This routine is only used in daemon mode. */
787 void glob_expand_module(char *base1, char *arg, char ***argv_p, int *argc_p, int *maxargs_p)
788 {
789 char *p, *s;
790 char *base = base1;
791 int base_len = strlen(base);
792
793 if (!arg || !*arg)
794 return;
795
796 if (strncmp(arg, base, base_len) == 0)
797 arg += base_len;
798
799 if (protect_args) {
800 glob_expand(arg, argv_p, argc_p, maxargs_p);
801 return;
802 }
803
804 arg = strdup(arg);
805
806 if (asprintf(&base," %s/", base1) < 0)
807 out_of_memory("glob_expand_module");
808 base_len++;
809
810 for (s = arg; *s; s = p + base_len) {
811 if ((p = strstr(s, base)) != NULL)
812 *p = '\0'; /* split it at this point */
813 glob_expand(s, argv_p, argc_p, maxargs_p);
814 if (!p)
815 break;
816 }
817
818 free(arg);
819 free(base);
820 }
821
822 /**
823 * Convert a string to lower case
824 **/
825 void strlower(char *s)
826 {
827 while (*s) {
828 if (isUpper(s))
829 *s = toLower(s);
830 s++;
831 }
832 }
833
834 /**
835 * Split a string into tokens based (usually) on whitespace & commas. If the
836 * string starts with a comma (after skipping any leading whitespace), then
837 * splitting is done only on commas. No empty tokens are ever returned. */
838 char *conf_strtok(char *str)
839 {
840 static int commas_only = 0;
841
842 if (str) {
843 while (isSpace(str)) str++;
844 if (*str == ',') {
845 commas_only = 1;
846 str++;
847 } else
848 commas_only = 0;
849 }
850
851 while (commas_only) {
852 char *end, *tok = strtok(str, ",");
853 if (!tok)
854 return NULL;
855 /* Trim just leading and trailing whitespace. */
856 while (isSpace(tok))
857 tok++;
858 end = tok + strlen(tok);
859 while (end > tok && isSpace(end-1))
860 *--end = '\0';
861 if (*tok)
862 return tok;
863 str = NULL;
864 }
865
866 return strtok(str, " ,\t\r\n");
867 }
868
869 /* Join strings p1 & p2 into "dest" with a guaranteed '/' between them. (If
870 * p1 ends with a '/', no extra '/' is inserted.) Returns the length of both
871 * strings + 1 (if '/' was inserted), regardless of whether the null-terminated
872 * string fits into destsize. */
873 size_t pathjoin(char *dest, size_t destsize, const char *p1, const char *p2)
874 {
875 size_t len = strlcpy(dest, p1, destsize);
876 if (len < destsize - 1) {
877 if (!len || dest[len-1] != '/')
878 dest[len++] = '/';
879 if (len < destsize - 1)
880 len += strlcpy(dest + len, p2, destsize - len);
881 else {
882 dest[len] = '\0';
883 len += strlen(p2);
884 }
885 }
886 else
887 len += strlen(p2) + 1; /* Assume we'd insert a '/'. */
888 return len;
889 }
890
891 /* Join any number of strings together, putting them in "dest". The return
892 * value is the length of all the strings, regardless of whether the null-
893 * terminated whole fits in destsize. Your list of string pointers must end
894 * with a NULL to indicate the end of the list. */
895 size_t stringjoin(char *dest, size_t destsize, ...)
896 {
897 va_list ap;
898 size_t len, ret = 0;
899 const char *src;
900
901 va_start(ap, destsize);
902 while (1) {
903 if (!(src = va_arg(ap, const char *)))
904 break;
905 len = strlen(src);
906 ret += len;
907 if (destsize > 1) {
908 if (len >= destsize)
909 len = destsize - 1;
910 memcpy(dest, src, len);
911 destsize -= len;
912 dest += len;
913 }
914 }
915 *dest = '\0';
916 va_end(ap);
917
918 return ret;
919 }
920
921 int count_dir_elements(const char *p)
922 {
923 int cnt = 0, new_component = 1;
924 while (*p) {
925 if (*p++ == '/')
926 new_component = (*p != '.' || (p[1] != '/' && p[1] != '\0'));
927 else if (new_component) {
928 new_component = 0;
929 cnt++;
930 }
931 }
932 return cnt;
933 }
934
935 /* Turns multiple adjacent slashes into a single slash (possible exception:
936 * the preserving of two leading slashes at the start), drops all leading or
937 * interior "." elements unless CFN_KEEP_DOT_DIRS is flagged. Will also drop
938 * a trailing '.' after a '/' if CFN_DROP_TRAILING_DOT_DIR is flagged, removes
939 * a trailing slash (perhaps after removing the aforementioned dot) unless
940 * CFN_KEEP_TRAILING_SLASH is flagged, and will also collapse ".." elements
941 * (except at the start) if CFN_COLLAPSE_DOT_DOT_DIRS is flagged. If the
942 * resulting name would be empty, returns ".". */
943 int clean_fname(char *name, int flags)
944 {
945 char *limit = name - 1, *t = name, *f = name;
946 int anchored;
947
948 if (!name)
949 return 0;
950
951 #define DOT_IS_DOT_DOT_DIR(bp) (bp[1] == '.' && (bp[2] == '/' || !bp[2]))
952
953 if ((anchored = *f == '/') != 0) {
954 *t++ = *f++;
955 #ifdef __CYGWIN__
956 /* If there are exactly 2 slashes at the start, preserve
957 * them. Would break daemon excludes unless the paths are
958 * really treated differently, so used this sparingly. */
959 if (*f == '/' && f[1] != '/')
960 *t++ = *f++;
961 #endif
962 } else if (flags & CFN_KEEP_DOT_DIRS && *f == '.' && f[1] == '/') {
963 *t++ = *f++;
964 *t++ = *f++;
965 } else if (flags & CFN_REFUSE_DOT_DOT_DIRS && *f == '.' && DOT_IS_DOT_DOT_DIR(f))
966 return -1;
967 while (*f) {
968 /* discard extra slashes */
969 if (*f == '/') {
970 f++;
971 continue;
972 }
973 if (*f == '.') {
974 /* discard interior "." dirs */
975 if (f[1] == '/' && !(flags & CFN_KEEP_DOT_DIRS)) {
976 f += 2;
977 continue;
978 }
979 if (f[1] == '\0' && flags & CFN_DROP_TRAILING_DOT_DIR)
980 break;
981 /* collapse ".." dirs */
982 if (flags & (CFN_COLLAPSE_DOT_DOT_DIRS|CFN_REFUSE_DOT_DOT_DIRS) && DOT_IS_DOT_DOT_DIR(f)) {
983 char *s = t - 1;
984 if (flags & CFN_REFUSE_DOT_DOT_DIRS)
985 return -1;
986 if (s == name && anchored) {
987 f += 2;
988 continue;
989 }
990 while (s > limit && *--s != '/') {}
991 if (s != t - 1 && (s < name || *s == '/')) {
992 t = s + 1;
993 f += 2;
994 continue;
995 }
996 limit = t + 2;
997 }
998 }
999 while (*f && (*t++ = *f++) != '/') {}
1000 }
1001
1002 if (t > name+anchored && t[-1] == '/' && !(flags & CFN_KEEP_TRAILING_SLASH))
1003 t--;
1004 if (t == name)
1005 *t++ = '.';
1006 *t = '\0';
1007
1008 #undef DOT_IS_DOT_DOT_DIR
1009
1010 return t - name;
1011 }
1012
1013 /* Make path appear as if a chroot had occurred. This handles a leading
1014 * "/" (either removing it or expanding it) and any leading or embedded
1015 * ".." components that attempt to escape past the module's top dir.
1016 *
1017 * If dest is NULL, a buffer is allocated to hold the result. It is legal
1018 * to call with the dest and the path (p) pointing to the same buffer, but
1019 * rootdir will be ignored to avoid expansion of the string.
1020 *
1021 * The rootdir string contains a value to use in place of a leading slash.
1022 * Specify NULL to get the default of "module_dir".
1023 *
1024 * The depth var is a count of how many '..'s to allow at the start of the
1025 * path.
1026 *
1027 * We also clean the path in a manner similar to clean_fname() but with a
1028 * few differences:
1029 *
1030 * Turns multiple adjacent slashes into a single slash, gets rid of "." dir
1031 * elements (INCLUDING a trailing dot dir), PRESERVES a trailing slash, and
1032 * ALWAYS collapses ".." elements (except for those at the start of the
1033 * string up to "depth" deep). If the resulting name would be empty,
1034 * change it into a ".". */
1035 char *sanitize_path(char *dest, const char *p, const char *rootdir, int depth, int flags)
1036 {
1037 char *start, *sanp;
1038 int rlen = 0, drop_dot_dirs = !relative_paths || !(flags & SP_KEEP_DOT_DIRS);
1039
1040 if (dest != p) {
1041 int plen = strlen(p); /* the path len INCLUDING any separating slash */
1042 if (*p == '/') {
1043 if (!rootdir)
1044 rootdir = module_dir;
1045 rlen = strlen(rootdir);
1046 depth = 0;
1047 p++;
1048 }
1049 if (!dest)
1050 dest = new_array(char, MAX(rlen + plen + 1, 2));
1051 else if (rlen + plen + 1 >= MAXPATHLEN)
1052 return NULL;
1053 if (rlen) { /* only true if p previously started with a slash */
1054 memcpy(dest, rootdir, rlen);
1055 if (rlen > 1) /* a rootdir of len 1 is "/", so this avoids a 2nd slash */
1056 dest[rlen++] = '/';
1057 }
1058 }
1059
1060 if (drop_dot_dirs) {
1061 while (*p == '.' && p[1] == '/')
1062 p += 2;
1063 }
1064
1065 start = sanp = dest + rlen;
1066 /* This loop iterates once per filename component in p, pointing at
1067 * the start of the name (past any prior slash) for each iteration. */
1068 while (*p) {
1069 /* discard leading or extra slashes */
1070 if (*p == '/') {
1071 p++;
1072 continue;
1073 }
1074 if (drop_dot_dirs) {
1075 if (*p == '.' && (p[1] == '/' || p[1] == '\0')) {
1076 /* skip "." component */
1077 p++;
1078 continue;
1079 }
1080 }
1081 if (*p == '.' && p[1] == '.' && (p[2] == '/' || p[2] == '\0')) {
1082 /* ".." component followed by slash or end */
1083 if (depth <= 0 || sanp != start) {
1084 p += 2;
1085 if (sanp != start) {
1086 /* back up sanp one level */
1087 --sanp; /* now pointing at slash */
1088 while (sanp > start && sanp[-1] != '/')
1089 sanp--;
1090 }
1091 continue;
1092 }
1093 /* allow depth levels of .. at the beginning */
1094 depth--;
1095 /* move the virtual beginning to leave the .. alone */
1096 start = sanp + 3;
1097 }
1098 /* copy one component through next slash */
1099 while (*p && (*sanp++ = *p++) != '/') {}
1100 }
1101 if (sanp == dest) {
1102 /* ended up with nothing, so put in "." component */
1103 *sanp++ = '.';
1104 }
1105 *sanp = '\0';
1106
1107 return dest;
1108 }
1109
1110 /* Like chdir(), but it keeps track of the current directory (in the
1111 * global "curr_dir"), and ensures that the path size doesn't overflow.
1112 * Also cleans the path using the clean_fname() function. */
1113 int change_dir(const char *dir, int set_path_only)
1114 {
1115 static int initialised, skipped_chdir;
1116 unsigned int len;
1117
1118 if (!initialised) {
1119 initialised = 1;
1120 if (getcwd(curr_dir, sizeof curr_dir - 1) == NULL) {
1121 rsyserr(FERROR, errno, "getcwd()");
1122 exit_cleanup(RERR_FILESELECT);
1123 }
1124 curr_dir_len = strlen(curr_dir);
1125 }
1126
1127 if (!dir) /* this call was probably just to initialize */
1128 return 0;
1129
1130 len = strlen(dir);
1131 if (len == 1 && *dir == '.' && (!skipped_chdir || set_path_only))
1132 return 1;
1133
1134 if (*dir == '/') {
1135 if (len >= sizeof curr_dir) {
1136 errno = ENAMETOOLONG;
1137 return 0;
1138 }
1139 if (!set_path_only && chdir(dir))
1140 return 0;
1141 skipped_chdir = set_path_only;
1142 memcpy(curr_dir, dir, len + 1);
1143 } else {
1144 unsigned int save_dir_len = curr_dir_len;
1145 if (curr_dir_len + 1 + len >= sizeof curr_dir) {
1146 errno = ENAMETOOLONG;
1147 return 0;
1148 }
1149 if (!(curr_dir_len && curr_dir[curr_dir_len-1] == '/'))
1150 curr_dir[curr_dir_len++] = '/';
1151 memcpy(curr_dir + curr_dir_len, dir, len + 1);
1152
1153 if (!set_path_only && chdir(curr_dir)) {
1154 curr_dir_len = save_dir_len;
1155 curr_dir[curr_dir_len] = '\0';
1156 return 0;
1157 }
1158 skipped_chdir = set_path_only;
1159 }
1160
1161 curr_dir_len = clean_fname(curr_dir, CFN_COLLAPSE_DOT_DOT_DIRS | CFN_DROP_TRAILING_DOT_DIR);
1162 if (sanitize_paths) {
1163 if (module_dirlen > curr_dir_len)
1164 module_dirlen = curr_dir_len;
1165 curr_dir_depth = count_dir_elements(curr_dir + module_dirlen);
1166 }
1167
1168 if (DEBUG_GTE(CHDIR, 1) && !set_path_only)
1169 rprintf(FINFO, "[%s] change_dir(%s)\n", who_am_i(), curr_dir);
1170
1171 return 1;
1172 }
1173
1174 /* This will make a relative path absolute and clean it up via clean_fname().
1175 * Returns the string, which might be newly allocated, or NULL on error. */
1176 char *normalize_path(char *path, BOOL force_newbuf, unsigned int *len_ptr)
1177 {
1178 unsigned int len;
1179
1180 if (*path != '/') { /* Make path absolute. */
1181 int len = strlen(path);
1182 if (curr_dir_len + 1 + len >= sizeof curr_dir)
1183 return NULL;
1184 curr_dir[curr_dir_len] = '/';
1185 memcpy(curr_dir + curr_dir_len + 1, path, len + 1);
1186 path = strdup(curr_dir);
1187 curr_dir[curr_dir_len] = '\0';
1188 } else if (force_newbuf)
1189 path = strdup(path);
1190
1191 len = clean_fname(path, CFN_COLLAPSE_DOT_DOT_DIRS | CFN_DROP_TRAILING_DOT_DIR);
1192
1193 if (len_ptr)
1194 *len_ptr = len;
1195
1196 return path;
1197 }
1198
1199 /**
1200 * Return a quoted string with the full pathname of the indicated filename.
1201 * The string " (in MODNAME)" may also be appended. The returned pointer
1202 * remains valid until the next time full_fname() is called.
1203 **/
1204 char *full_fname(const char *fn)
1205 {
1206 static char *result = NULL;
1207 char *m1, *m2, *m3;
1208 char *p1, *p2;
1209
1210 if (result)
1211 free(result);
1212
1213 if (*fn == '/')
1214 p1 = p2 = "";
1215 else {
1216 p1 = curr_dir + module_dirlen;
1217 for (p2 = p1; *p2 == '/'; p2++) {}
1218 if (*p2)
1219 p2 = "/";
1220 }
1221 if (module_id >= 0) {
1222 m1 = " (in ";
1223 m2 = lp_name(module_id);
1224 m3 = ")";
1225 } else
1226 m1 = m2 = m3 = "";
1227
1228 if (asprintf(&result, "\"%s%s%s\"%s%s%s", p1, p2, fn, m1, m2, m3) < 0)
1229 out_of_memory("full_fname");
1230
1231 return result;
1232 }
1233
1234 static char partial_fname[MAXPATHLEN];
1235
1236 char *partial_dir_fname(const char *fname)
1237 {
1238 char *t = partial_fname;
1239 int sz = sizeof partial_fname;
1240 const char *fn;
1241
1242 if ((fn = strrchr(fname, '/')) != NULL) {
1243 fn++;
1244 if (*partial_dir != '/') {
1245 int len = fn - fname;
1246 strncpy(t, fname, len); /* safe */
1247 t += len;
1248 sz -= len;
1249 }
1250 } else
1251 fn = fname;
1252 if ((int)pathjoin(t, sz, partial_dir, fn) >= sz)
1253 return NULL;
1254 if (daemon_filter_list.head) {
1255 t = strrchr(partial_fname, '/');
1256 *t = '\0';
1257 if (check_filter(&daemon_filter_list, FLOG, partial_fname, 1) < 0)
1258 return NULL;
1259 *t = '/';
1260 if (check_filter(&daemon_filter_list, FLOG, partial_fname, 0) < 0)
1261 return NULL;
1262 }
1263
1264 return partial_fname;
1265 }
1266
1267 /* If no --partial-dir option was specified, we don't need to do anything
1268 * (the partial-dir is essentially '.'), so just return success. */
1269 int handle_partial_dir(const char *fname, int create)
1270 {
1271 char *fn, *dir;
1272
1273 if (fname != partial_fname)
1274 return 1;
1275 if (!create && *partial_dir == '/')
1276 return 1;
1277 if (!(fn = strrchr(partial_fname, '/')))
1278 return 1;
1279
1280 *fn = '\0';
1281 dir = partial_fname;
1282 if (create) {
1283 STRUCT_STAT st;
1284 int statret = do_lstat(dir, &st);
1285 if (statret == 0 && !S_ISDIR(st.st_mode)) {
1286 if (do_unlink(dir) < 0) {
1287 *fn = '/';
1288 return 0;
1289 }
1290 statret = -1;
1291 }
1292 if (statret < 0 && do_mkdir(dir, 0700) < 0) {
1293 *fn = '/';
1294 return 0;
1295 }
1296 } else
1297 do_rmdir(dir);
1298 *fn = '/';
1299
1300 return 1;
1301 }
1302
1303 /* Determine if a symlink points outside the current directory tree.
1304 * This is considered "unsafe" because e.g. when mirroring somebody
1305 * else's machine it might allow them to establish a symlink to
1306 * /etc/passwd, and then read it through a web server.
1307 *
1308 * Returns 1 if unsafe, 0 if safe.
1309 *
1310 * Null symlinks and absolute symlinks are always unsafe.
1311 *
1312 * Basically here we are concerned with symlinks whose target contains
1313 * "..", because this might cause us to walk back up out of the
1314 * transferred directory. We are not allowed to go back up and
1315 * reenter.
1316 *
1317 * "dest" is the target of the symlink in question.
1318 *
1319 * "src" is the top source directory currently applicable at the level
1320 * of the referenced symlink. This is usually the symlink's full path
1321 * (including its name), as referenced from the root of the transfer. */
1322 int unsafe_symlink(const char *dest, const char *src)
1323 {
1324 const char *name, *slash;
1325 int depth = 0;
1326
1327 /* all absolute and null symlinks are unsafe */
1328 if (!dest || !*dest || *dest == '/')
1329 return 1;
1330
1331 /* find out what our safety margin is */
1332 for (name = src; (slash = strchr(name, '/')) != 0; name = slash+1) {
1333 /* ".." segment starts the count over. "." segment is ignored. */
1334 if (*name == '.' && (name[1] == '/' || (name[1] == '.' && name[2] == '/'))) {
1335 if (name[1] == '.')
1336 depth = 0;
1337 } else
1338 depth++;
1339 while (slash[1] == '/') slash++; /* just in case src isn't clean */
1340 }
1341 if (*name == '.' && name[1] == '.' && name[2] == '\0')
1342 depth = 0;
1343
1344 for (name = dest; (slash = strchr(name, '/')) != 0; name = slash+1) {
1345 if (*name == '.' && (name[1] == '/' || (name[1] == '.' && name[2] == '/'))) {
1346 if (name[1] == '.') {
1347 /* if at any point we go outside the current directory
1348 then stop - it is unsafe */
1349 if (--depth < 0)
1350 return 1;
1351 }
1352 } else
1353 depth++;
1354 while (slash[1] == '/') slash++;
1355 }
1356 if (*name == '.' && name[1] == '.' && name[2] == '\0')
1357 depth--;
1358
1359 return depth < 0;
1360 }
1361
1362 /* Return the date and time as a string. Some callers tweak returned buf. */
1363 char *timestring(time_t t)
1364 {
1365 static int ndx = 0;
1366 static char buffers[4][20]; /* We support 4 simultaneous timestring results. */
1367 char *TimeBuf = buffers[ndx = (ndx + 1) % 4];
1368 struct tm *tm = localtime(&t);
1369 int len = snprintf(TimeBuf, sizeof buffers[0], "%4d/%02d/%02d %02d:%02d:%02d",
1370 (int)tm->tm_year + 1900, (int)tm->tm_mon + 1, (int)tm->tm_mday,
1371 (int)tm->tm_hour, (int)tm->tm_min, (int)tm->tm_sec);
1372 assert(len > 0); /* Silence gcc warning */
1373
1374 return TimeBuf;
1375 }
1376
1377 /* Determine if two time_t values are equivalent (either exact, or in
1378 * the modification timestamp window established by --modify-window).
1379 * Returns 1 if the times the "same", or 0 if they are different. */
1380 int same_time(time_t f1_sec, unsigned long f1_nsec, time_t f2_sec, unsigned long f2_nsec)
1381 {
1382 if (modify_window == 0)
1383 return f1_sec == f2_sec;
1384 if (modify_window < 0)
1385 return f1_sec == f2_sec && f1_nsec == f2_nsec;
1386 /* The nanoseconds do not figure into these checks -- time windows don't care about that. */
1387 if (f2_sec > f1_sec)
1388 return f2_sec - f1_sec <= modify_window;
1389 return f1_sec - f2_sec <= modify_window;
1390 }
1391
1392 #ifdef __INSURE__XX
1393 #include <dlfcn.h>
1394
1395 /**
1396 This routine is a trick to immediately catch errors when debugging
1397 with insure. A xterm with a gdb is popped up when insure catches
1398 a error. It is Linux specific.
1399 **/
1400 int _Insure_trap_error(int a1, int a2, int a3, int a4, int a5, int a6)
1401 {
1402 static int (*fn)();
1403 int ret, pid_int = getpid();
1404 char *cmd;
1405
1406 if (asprintf(&cmd,
1407 "/usr/X11R6/bin/xterm -display :0 -T Panic -n Panic -e /bin/sh -c 'cat /tmp/ierrs.*.%d ; "
1408 "gdb /proc/%d/exe %d'", pid_int, pid_int, pid_int) < 0)
1409 return -1;
1410
1411 if (!fn) {
1412 static void *h;
1413 h = dlopen("/usr/local/parasoft/insure++lite/lib.linux2/libinsure.so", RTLD_LAZY);
1414 fn = dlsym(h, "_Insure_trap_error");
1415 }
1416
1417 ret = fn(a1, a2, a3, a4, a5, a6);
1418
1419 system(cmd);
1420
1421 free(cmd);
1422
1423 return ret;
1424 }
1425 #endif
1426
1427 /* Take a filename and filename length and return the most significant
1428 * filename suffix we can find. This ignores suffixes such as "~",
1429 * ".bak", ".orig", ".~1~", etc. */
1430 const char *find_filename_suffix(const char *fn, int fn_len, int *len_ptr)
1431 {
1432 const char *suf, *s;
1433 BOOL had_tilde;
1434 int s_len;
1435
1436 /* One or more dots at the start aren't a suffix. */
1437 while (fn_len && *fn == '.') fn++, fn_len--;
1438
1439 /* Ignore the ~ in a "foo~" filename. */
1440 if (fn_len > 1 && fn[fn_len-1] == '~')
1441 fn_len--, had_tilde = True;
1442 else
1443 had_tilde = False;
1444
1445 /* Assume we don't find an suffix. */
1446 suf = "";
1447 *len_ptr = 0;
1448
1449 /* Find the last significant suffix. */
1450 for (s = fn + fn_len; fn_len > 1; ) {
1451 while (*--s != '.' && s != fn) {}
1452 if (s == fn)
1453 break;
1454 s_len = fn_len - (s - fn);
1455 fn_len = s - fn;
1456 if (s_len == 4) {
1457 if (strcmp(s+1, "bak") == 0
1458 || strcmp(s+1, "old") == 0)
1459 continue;
1460 } else if (s_len == 5) {
1461 if (strcmp(s+1, "orig") == 0)
1462 continue;
1463 } else if (s_len > 2 && had_tilde && s[1] == '~' && isDigit(s + 2))
1464 continue;
1465 *len_ptr = s_len;
1466 suf = s;
1467 if (s_len == 1)
1468 break;
1469 /* Determine if the suffix is all digits. */
1470 for (s++, s_len--; s_len > 0; s++, s_len--) {
1471 if (!isDigit(s))
1472 return suf;
1473 }
1474 /* An all-digit suffix may not be that significant. */
1475 s = suf;
1476 }
1477
1478 return suf;
1479 }
1480
1481 /* This is an implementation of the Levenshtein distance algorithm. It
1482 * was implemented to avoid needing a two-dimensional matrix (to save
1483 * memory). It was also tweaked to try to factor in the ASCII distance
1484 * between changed characters as a minor distance quantity. The normal
1485 * Levenshtein units of distance (each signifying a single change between
1486 * the two strings) are defined as a "UNIT". */
1487
1488 #define UNIT (1 << 16)
1489
1490 uint32 fuzzy_distance(const char *s1, unsigned len1, const char *s2, unsigned len2, uint32 upperlimit)
1491 {
1492 uint32 a[MAXPATHLEN], diag, above, left, diag_inc, above_inc, left_inc;
1493 int32 cost;
1494 unsigned i1, i2;
1495
1496 /* Check to see if the Levenshtein distance must be greater than the
1497 * upper limit defined by the previously found lowest distance using
1498 * the heuristic that the Levenshtein distance is greater than the
1499 * difference in length of the two strings */
1500 if ((len1 > len2 ? len1 - len2 : len2 - len1) * UNIT > upperlimit)
1501 return 0xFFFFU * UNIT + 1;
1502
1503 if (!len1 || !len2) {
1504 if (!len1) {
1505 s1 = s2;
1506 len1 = len2;
1507 }
1508 for (i1 = 0, cost = 0; i1 < len1; i1++)
1509 cost += s1[i1];
1510 return (int32)len1 * UNIT + cost;
1511 }
1512
1513 for (i2 = 0; i2 < len2; i2++)
1514 a[i2] = (i2+1) * UNIT;
1515
1516 for (i1 = 0; i1 < len1; i1++) {
1517 diag = i1 * UNIT;
1518 above = (i1+1) * UNIT;
1519 for (i2 = 0; i2 < len2; i2++) {
1520 left = a[i2];
1521 if ((cost = *((uchar*)s1+i1) - *((uchar*)s2+i2)) != 0) {
1522 if (cost < 0)
1523 cost = UNIT - cost;
1524 else
1525 cost = UNIT + cost;
1526 }
1527 diag_inc = diag + cost;
1528 left_inc = left + UNIT + *((uchar*)s1+i1);
1529 above_inc = above + UNIT + *((uchar*)s2+i2);
1530 a[i2] = above = left < above
1531 ? (left_inc < diag_inc ? left_inc : diag_inc)
1532 : (above_inc < diag_inc ? above_inc : diag_inc);
1533 diag = left;
1534 }
1535 }
1536
1537 return a[len2-1];
1538 }
1539
1540 #define BB_SLOT_SIZE (16*1024) /* Desired size in bytes */
1541 #define BB_PER_SLOT_BITS (BB_SLOT_SIZE * 8) /* Number of bits per slot */
1542 #define BB_PER_SLOT_INTS (BB_SLOT_SIZE / 4) /* Number of int32s per slot */
1543
1544 struct bitbag {
1545 uint32 **bits;
1546 int slot_cnt;
1547 };
1548
1549 struct bitbag *bitbag_create(int max_ndx)
1550 {
1551 struct bitbag *bb = new(struct bitbag);
1552 bb->slot_cnt = (max_ndx + BB_PER_SLOT_BITS - 1) / BB_PER_SLOT_BITS;
1553
1554 bb->bits = new_array0(uint32*, bb->slot_cnt);
1555
1556 return bb;
1557 }
1558
1559 void bitbag_set_bit(struct bitbag *bb, int ndx)
1560 {
1561 int slot = ndx / BB_PER_SLOT_BITS;
1562 ndx %= BB_PER_SLOT_BITS;
1563
1564 if (!bb->bits[slot])
1565 bb->bits[slot] = new_array0(uint32, BB_PER_SLOT_INTS);
1566
1567 bb->bits[slot][ndx/32] |= 1u << (ndx % 32);
1568 }
1569
1570 #if 0 /* not needed yet */
1571 void bitbag_clear_bit(struct bitbag *bb, int ndx)
1572 {
1573 int slot = ndx / BB_PER_SLOT_BITS;
1574 ndx %= BB_PER_SLOT_BITS;
1575
1576 if (!bb->bits[slot])
1577 return;
1578
1579 bb->bits[slot][ndx/32] &= ~(1u << (ndx % 32));
1580 }
1581
1582 int bitbag_check_bit(struct bitbag *bb, int ndx)
1583 {
1584 int slot = ndx / BB_PER_SLOT_BITS;
1585 ndx %= BB_PER_SLOT_BITS;
1586
1587 if (!bb->bits[slot])
1588 return 0;
1589
1590 return bb->bits[slot][ndx/32] & (1u << (ndx % 32)) ? 1 : 0;
1591 }
1592 #endif
1593
1594 /* Call this with -1 to start checking from 0. Returns -1 at the end. */
1595 int bitbag_next_bit(struct bitbag *bb, int after)
1596 {
1597 uint32 bits, mask;
1598 int i, ndx = after + 1;
1599 int slot = ndx / BB_PER_SLOT_BITS;
1600 ndx %= BB_PER_SLOT_BITS;
1601
1602 mask = (1u << (ndx % 32)) - 1;
1603 for (i = ndx / 32; slot < bb->slot_cnt; slot++, i = mask = 0) {
1604 if (!bb->bits[slot])
1605 continue;
1606 for ( ; i < BB_PER_SLOT_INTS; i++, mask = 0) {
1607 if (!(bits = bb->bits[slot][i] & ~mask))
1608 continue;
1609 /* The xor magic figures out the lowest enabled bit in
1610 * bits, and the switch quickly computes log2(bit). */
1611 switch (bits ^ (bits & (bits-1))) {
1612 #define LOG2(n) case 1u << n: return slot*BB_PER_SLOT_BITS + i*32 + n
1613 LOG2(0); LOG2(1); LOG2(2); LOG2(3);
1614 LOG2(4); LOG2(5); LOG2(6); LOG2(7);
1615 LOG2(8); LOG2(9); LOG2(10); LOG2(11);
1616 LOG2(12); LOG2(13); LOG2(14); LOG2(15);
1617 LOG2(16); LOG2(17); LOG2(18); LOG2(19);
1618 LOG2(20); LOG2(21); LOG2(22); LOG2(23);
1619 LOG2(24); LOG2(25); LOG2(26); LOG2(27);
1620 LOG2(28); LOG2(29); LOG2(30); LOG2(31);
1621 }
1622 return -1; /* impossible... */
1623 }
1624 }
1625
1626 return -1;
1627 }
1628
1629 void flist_ndx_push(flist_ndx_list *lp, int ndx)
1630 {
1631 struct flist_ndx_item *item;
1632
1633 item = new(struct flist_ndx_item);
1634 item->next = NULL;
1635 item->ndx = ndx;
1636 if (lp->tail)
1637 lp->tail->next = item;
1638 else
1639 lp->head = item;
1640 lp->tail = item;
1641 }
1642
1643 int flist_ndx_pop(flist_ndx_list *lp)
1644 {
1645 struct flist_ndx_item *next;
1646 int ndx;
1647
1648 if (!lp->head)
1649 return -1;
1650
1651 ndx = lp->head->ndx;
1652 next = lp->head->next;
1653 free(lp->head);
1654 lp->head = next;
1655 if (!next)
1656 lp->tail = NULL;
1657
1658 return ndx;
1659 }
1660
1661 /* Make sure there is room for one more item in the item list. If there
1662 * is not, expand the list as indicated by the value of "incr":
1663 * - if incr < 0 then increase the malloced size by -1 * incr
1664 * - if incr >= 0 then either make the malloced size equal to "incr"
1665 * or (if that's not large enough) double the malloced size
1666 * After the size check, the list's count is incremented by 1 and a pointer
1667 * to the "new" list item is returned.
1668 */
1669 void *expand_item_list(item_list *lp, size_t item_size, const char *desc, int incr)
1670 {
1671 /* First time through, 0 <= 0, so list is expanded. */
1672 if (lp->malloced <= lp->count) {
1673 void *new_ptr;
1674 size_t expand_size;
1675 if (incr < 0)
1676 expand_size = -incr; /* increase slowly */
1677 else if (lp->malloced < (size_t)incr)
1678 expand_size = incr - lp->malloced;
1679 else if (lp->malloced)
1680 expand_size = lp->malloced; /* double in size */
1681 else
1682 expand_size = 1;
1683 if (SIZE_MAX/item_size - expand_size < lp->malloced)
1684 overflow_exit("expand_item_list");
1685 expand_size += lp->malloced;
1686 new_ptr = realloc_buf(lp->items, expand_size * item_size);
1687 if (DEBUG_GTE(FLIST, 3)) {
1688 rprintf(FINFO, "[%s] expand %s to %s bytes, did%s move\n",
1689 who_am_i(), desc, big_num(expand_size * item_size),
1690 new_ptr == lp->items ? " not" : "");
1691 }
1692
1693 lp->items = new_ptr;
1694 lp->malloced = expand_size;
1695 }
1696 return (char*)lp->items + (lp->count++ * item_size);
1697 }
1698
1699 /* This zeroing of memory won't be optimized away by the compiler. */
1700 void force_memzero(void *buf, size_t len)
1701 {
1702 volatile uchar *z = buf;
1703 while (len-- > 0)
1704 *z++ = '\0';
1705 }