Changed the static file_struct var to match the changes in rsync.h.
[rsync/rsync.git] / util.c
CommitLineData
0f78b815
WD
1/*
2 * Utility routines used in rsync.
5cb37436 3 *
0f78b815
WD
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, 2004, 2005, 2006 Wayne Davison
5cb37436 8 *
0ecfbf27
MP
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 2 of the License, or
12 * (at your option) any later version.
5cb37436 13 *
0ecfbf27
MP
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.
5cb37436 18 *
e7c67065
WD
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write to the Free Software Foundation, Inc.,
21 * 51 Franklin Street - Fifth Floor, Boston, MA 02110-1301, USA.
0ecfbf27 22 */
c627d613 23
c627d613
AT
24#include "rsync.h"
25
c7c11a0d 26extern int verbose;
bc6ebcd2
WD
27extern int dry_run;
28extern int module_id;
29extern int modify_window;
7e43da81 30extern int relative_paths;
e175fb07 31extern int human_readable;
a8167c66 32extern unsigned int module_dirlen;
e9357a2d 33extern mode_t orig_umask;
a7260c40 34extern char *partial_dir;
7842418b 35extern struct filter_list_struct server_filter_list;
c7c11a0d 36
0ecfbf27
MP
37int sanitize_paths = 0;
38
a8167c66
WD
39char curr_dir[MAXPATHLEN];
40unsigned int curr_dir_len;
41int curr_dir_depth; /* This is only set for a sanitizing daemon. */
42
0f78b815 43/* Set a fd into nonblocking mode. */
f0359dd0
AT
44void set_nonblocking(int fd)
45{
46 int val;
47
ff530f04 48 if ((val = fcntl(fd, F_GETFL)) == -1)
f0359dd0
AT
49 return;
50 if (!(val & NONBLOCK_FLAG)) {
51 val |= NONBLOCK_FLAG;
52 fcntl(fd, F_SETFL, val);
53 }
54}
55
0f78b815 56/* Set a fd into blocking mode. */
36349ea0
AT
57void set_blocking(int fd)
58{
59 int val;
60
ff530f04 61 if ((val = fcntl(fd, F_GETFL)) == -1)
36349ea0
AT
62 return;
63 if (val & NONBLOCK_FLAG) {
64 val &= ~NONBLOCK_FLAG;
65 fcntl(fd, F_SETFL, val);
66 }
67}
68
ac13ad10 69/**
0ecfbf27
MP
70 * Create a file descriptor pair - like pipe() but use socketpair if
71 * possible (because of blocking issues on pipes).
5cb37436 72 *
0ecfbf27 73 * Always set non-blocking.
f0359dd0 74 */
08f15335
AT
75int fd_pair(int fd[2])
76{
f0359dd0
AT
77 int ret;
78
4f5b0756 79#ifdef HAVE_SOCKETPAIR
f0359dd0 80 ret = socketpair(AF_UNIX, SOCK_STREAM, 0, fd);
08f15335 81#else
f0359dd0 82 ret = pipe(fd);
08f15335 83#endif
f0359dd0
AT
84
85 if (ret == 0) {
86 set_nonblocking(fd[0]);
87 set_nonblocking(fd[1]);
88 }
0ecfbf27 89
f0359dd0 90 return ret;
08f15335
AT
91}
92
0ecfbf27 93void print_child_argv(char **cmd)
5ad0e46f 94{
a3f6dbdf 95 rprintf(FCLIENT, "opening connection using ");
5ad0e46f
MP
96 for (; *cmd; cmd++) {
97 /* Look for characters that ought to be quoted. This
98 * is not a great quoting algorithm, but it's
99 * sufficient for a log message. */
100 if (strspn(*cmd, "abcdefghijklmnopqrstuvwxyz"
101 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
102 "0123456789"
103 ",.-_=+@/") != strlen(*cmd)) {
a3f6dbdf 104 rprintf(FCLIENT, "\"%s\" ", *cmd);
5ad0e46f 105 } else {
a3f6dbdf 106 rprintf(FCLIENT, "%s ", *cmd);
5ad0e46f
MP
107 }
108 }
a3f6dbdf 109 rprintf(FCLIENT, "\n");
5ad0e46f
MP
110}
111
4a19c3b2 112NORETURN void out_of_memory(const char *str)
c627d613 113{
2cd421d8 114 rprintf(FERROR, "ERROR: out of memory in %s [%s]\n", str, who_am_i());
c284f34a 115 exit_cleanup(RERR_MALLOC);
575f2fca
AT
116}
117
4a19c3b2 118NORETURN void overflow_exit(const char *str)
575f2fca 119{
2cd421d8 120 rprintf(FERROR, "ERROR: buffer overflow in %s [%s]\n", str, who_am_i());
c284f34a 121 exit_cleanup(RERR_MALLOC);
c627d613
AT
122}
123
4a19c3b2 124int set_modtime(const char *fname, time_t modtime, mode_t mode)
c627d613 125{
25007999
WD
126#if !defined HAVE_LUTIMES || !defined HAVE_UTIMES
127 if (S_ISLNK(mode))
128 return 1;
129#endif
130
404e813c
MP
131 if (verbose > 2) {
132 rprintf(FINFO, "set modtime of %s to (%ld) %s",
785abd48 133 fname, (long)modtime,
404e813c
MP
134 asctime(localtime(&modtime)));
135 }
5cb37436 136
15778afb
WD
137 if (dry_run)
138 return 0;
139
31e12522 140 {
25007999
WD
141#ifdef HAVE_UTIMES
142 struct timeval t[2];
143 t[0].tv_sec = time(NULL);
144 t[0].tv_usec = 0;
145 t[1].tv_sec = modtime;
146 t[1].tv_usec = 0;
147# ifdef HAVE_LUTIMES
149a78e3
WD
148 if (S_ISLNK(mode)) {
149 lutimes(fname, t);
150 return 0; /* ignore errors */
151 }
25007999
WD
152# endif
153 return utimes(fname, t);
154#elif defined HAVE_UTIMBUF
5cb37436 155 struct utimbuf tbuf;
31e12522
AT
156 tbuf.actime = time(NULL);
157 tbuf.modtime = modtime;
158 return utime(fname,&tbuf);
4f5b0756 159#elif defined HAVE_UTIME
31e12522
AT
160 time_t t[2];
161 t[0] = time(NULL);
162 t[1] = modtime;
163 return utime(fname,t);
c627d613 164#else
25007999 165#error No file-time-modification routine found!
c627d613 166#endif
31e12522 167 }
c627d613 168}
94481d91 169
e9357a2d
WD
170/* This creates a new directory with default permissions. Since there
171 * might be some directory-default permissions affecting this, we can't
172 * force the permissions directly using the original umask and mkdir(). */
173int mkdir_defmode(char *fname)
174{
175 int ret;
176
177 umask(orig_umask);
178 ret = do_mkdir(fname, ACCESSPERMS);
179 umask(0);
180
181 return ret;
182}
183
85c41757
WD
184/* Create any necessary directories in fname. Any missing directories are
185 * created with default permissions. */
e9357a2d 186int create_directory_path(char *fname)
6574b4f7 187{
6574b4f7 188 char *p;
85c41757 189 int ret = 0;
6574b4f7 190
c284f34a
WD
191 while (*fname == '/')
192 fname++;
193 while (strncmp(fname, "./", 2) == 0)
194 fname += 2;
6574b4f7 195
85c41757 196 umask(orig_umask);
6574b4f7 197 p = fname;
c284f34a 198 while ((p = strchr(p,'/')) != NULL) {
85c41757
WD
199 *p = '\0';
200 if (do_mkdir(fname, ACCESSPERMS) < 0 && errno != EEXIST)
201 ret = -1;
202 *p++ = '/';
6574b4f7 203 }
85c41757
WD
204 umask(0);
205
206 return ret;
6574b4f7 207}
950ab32d 208
ac13ad10
MP
209/**
210 * Write @p len bytes at @p ptr to descriptor @p desc, retrying if
211 * interrupted.
212 *
213 * @retval len upon success
214 *
215 * @retval <0 write's (negative) error code
216 *
217 * Derived from GNU C's cccp.c.
218 */
4a19c3b2 219int full_write(int desc, const char *ptr, size_t len)
950ab32d
AT
220{
221 int total_written;
5cb37436 222
950ab32d
AT
223 total_written = 0;
224 while (len > 0) {
5c1b7bfd 225 int written = write(desc, ptr, len);
950ab32d 226 if (written < 0) {
950ab32d
AT
227 if (errno == EINTR)
228 continue;
950ab32d
AT
229 return written;
230 }
231 total_written += written;
232 ptr += written;
233 len -= written;
234 }
235 return total_written;
236}
237
ac13ad10
MP
238/**
239 * Read @p len bytes at @p ptr from descriptor @p desc, retrying if
240 * interrupted.
241 *
242 * @retval >0 the actual number of bytes read
243 *
244 * @retval 0 for EOF
245 *
246 * @retval <0 for an error.
247 *
248 * Derived from GNU C's cccp.c. */
9dd891bb 249static int safe_read(int desc, char *ptr, size_t len)
950ab32d
AT
250{
251 int n_chars;
5cb37436 252
9dd891bb 253 if (len == 0)
950ab32d 254 return len;
5cb37436 255
950ab32d
AT
256 do {
257 n_chars = read(desc, ptr, len);
258 } while (n_chars < 0 && errno == EINTR);
5cb37436 259
950ab32d
AT
260 return n_chars;
261}
262
ac13ad10
MP
263/** Copy a file.
264 *
3e13004b
WD
265 * This is used in conjunction with the --temp-dir, --backup, and
266 * --copy-dest options. */
63cf5ae7 267int copy_file(const char *source, const char *dest, mode_t mode)
950ab32d
AT
268{
269 int ifd;
270 int ofd;
271 char buf[1024 * 8];
272 int len; /* Number of bytes read into `buf'. */
273
8c9fd200 274 ifd = do_open(source, O_RDONLY, 0);
950ab32d 275 if (ifd == -1) {
d62bcc17 276 rsyserr(FERROR, errno, "open %s", full_fname(source));
950ab32d
AT
277 return -1;
278 }
279
c7c11a0d 280 if (robust_unlink(dest) && errno != ENOENT) {
d62bcc17 281 rsyserr(FERROR, errno, "unlink %s", full_fname(dest));
950ab32d
AT
282 return -1;
283 }
284
31e12522 285 ofd = do_open(dest, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, mode);
c46ded46 286 if (ofd == -1) {
d62bcc17 287 rsyserr(FERROR, errno, "open %s", full_fname(dest));
950ab32d
AT
288 close(ifd);
289 return -1;
290 }
291
5cb37436 292 while ((len = safe_read(ifd, buf, sizeof buf)) > 0) {
950ab32d 293 if (full_write(ofd, buf, len) < 0) {
d62bcc17 294 rsyserr(FERROR, errno, "write %s", full_fname(dest));
950ab32d
AT
295 close(ifd);
296 close(ofd);
297 return -1;
298 }
299 }
300
8b602edd 301 if (len < 0) {
d62bcc17 302 rsyserr(FERROR, errno, "read %s", full_fname(source));
8b602edd
WD
303 close(ifd);
304 close(ofd);
305 return -1;
306 }
307
9f27cd8c 308 if (close(ifd) < 0) {
d62bcc17
WD
309 rsyserr(FINFO, errno, "close failed on %s",
310 full_fname(source));
9f27cd8c
WD
311 }
312
313 if (close(ofd) < 0) {
d62bcc17
WD
314 rsyserr(FERROR, errno, "close failed on %s",
315 full_fname(dest));
9f27cd8c
WD
316 return -1;
317 }
950ab32d 318
950ab32d
AT
319 return 0;
320}
feaa89c4 321
c7c11a0d
DD
322/* MAX_RENAMES should be 10**MAX_RENAMES_DIGITS */
323#define MAX_RENAMES_DIGITS 3
324#define MAX_RENAMES 1000
325
ac13ad10 326/**
b4235b31
MP
327 * Robust unlink: some OS'es (HPUX) refuse to unlink busy files, so
328 * rename to <path>/.rsyncNNN instead.
329 *
330 * Note that successive rsync runs will shuffle the filenames around a
331 * bit as long as the file is still busy; this is because this function
332 * does not know if the unlink call is due to a new file coming in, or
333 * --delete trying to remove old .rsyncNNN files, hence it renames it
334 * each time.
335 **/
63cf5ae7 336int robust_unlink(const char *fname)
c7c11a0d
DD
337{
338#ifndef ETXTBSY
339 return do_unlink(fname);
340#else
341 static int counter = 1;
342 int rc, pos, start;
343 char path[MAXPATHLEN];
344
345 rc = do_unlink(fname);
c284f34a 346 if (rc == 0 || errno != ETXTBSY)
c7c11a0d
DD
347 return rc;
348
c284f34a
WD
349 if ((pos = strlcpy(path, fname, MAXPATHLEN)) >= MAXPATHLEN)
350 pos = MAXPATHLEN - 1;
c7c11a0d 351
c284f34a
WD
352 while (pos > 0 && path[pos-1] != '/')
353 pos--;
5cb37436 354 pos += strlcpy(path+pos, ".rsync", MAXPATHLEN-pos);
c7c11a0d
DD
355
356 if (pos > (MAXPATHLEN-MAX_RENAMES_DIGITS-1)) {
357 errno = ETXTBSY;
358 return -1;
359 }
360
361 /* start where the last one left off to reduce chance of clashes */
362 start = counter;
363 do {
434c4098 364 snprintf(&path[pos], MAX_RENAMES_DIGITS+1, "%03d", counter);
c7c11a0d
DD
365 if (++counter >= MAX_RENAMES)
366 counter = 1;
c284f34a 367 } while ((rc = access(path, 0)) == 0 && counter != start);
c7c11a0d 368
4791825d 369 if (verbose > 0) {
c7c11a0d 370 rprintf(FINFO,"renaming %s to %s because of text busy\n",
785abd48 371 fname, path);
4791825d 372 }
c7c11a0d
DD
373
374 /* maybe we should return rename()'s exit status? Nah. */
375 if (do_rename(fname, path) != 0) {
376 errno = ETXTBSY;
377 return -1;
378 }
379 return 0;
380#endif
381}
382
630f548f 383/* Returns 0 on successful rename, 1 if we successfully copied the file
3ed8eafc
WD
384 * across filesystems, -2 if copy_file() failed, and -1 on other errors.
385 * If partialptr is not NULL and we need to do a copy, copy the file into
386 * the active partial-dir instead of over the destination file. */
4a19c3b2 387int robust_rename(const char *from, const char *to, const char *partialptr,
3ed8eafc 388 int mode)
c7c11a0d 389{
62c9e6b3
WD
390 int tries = 4;
391
392 while (tries--) {
393 if (do_rename(from, to) == 0)
394 return 0;
395
396 switch (errno) {
397#ifdef ETXTBSY
398 case ETXTBSY:
399 if (robust_unlink(to) != 0)
400 return -1;
401 break;
c7c11a0d 402#endif
62c9e6b3 403 case EXDEV:
3ed8eafc
WD
404 if (partialptr) {
405 if (!handle_partial_dir(partialptr,PDIR_CREATE))
406 return -1;
407 to = partialptr;
408 }
62c9e6b3
WD
409 if (copy_file(from, to, mode) != 0)
410 return -2;
411 do_unlink(from);
630f548f 412 return 1;
62c9e6b3
WD
413 default:
414 return -1;
415 }
416 }
417 return -1;
feaa89c4 418}
3ba62a83 419
3ba62a83
AT
420static pid_t all_pids[10];
421static int num_pids;
422
4cf64834 423/** Fork and record the pid of the child. **/
3ba62a83
AT
424pid_t do_fork(void)
425{
426 pid_t newpid = fork();
5cb37436 427
4cf64834 428 if (newpid != 0 && newpid != -1) {
3ba62a83
AT
429 all_pids[num_pids++] = newpid;
430 }
431 return newpid;
432}
433
4cf64834
MP
434/**
435 * Kill all children.
436 *
437 * @todo It would be kind of nice to make sure that they are actually
438 * all our children before we kill them, because their pids may have
439 * been recycled by some other process. Perhaps when we wait for a
440 * child, we should remove it from this array. Alternatively we could
441 * perhaps use process groups, but I think that would not work on
442 * ancient Unix versions that don't support them.
443 **/
3ba62a83
AT
444void kill_all(int sig)
445{
446 int i;
4cf64834
MP
447
448 for (i = 0; i < num_pids; i++) {
449 /* Let's just be a little careful where we
450 * point that gun, hey? See kill(2) for the
451 * magic caused by negative values. */
452 pid_t p = all_pids[i];
453
454 if (p == getpid())
455 continue;
456 if (p <= 0)
457 continue;
458
459 kill(p, sig);
3ba62a83
AT
460 }
461}
9486289c 462
ac13ad10 463/** Turn a user name into a uid */
4a19c3b2 464int name_to_uid(const char *name, uid_t *uid)
8ef4ffd6
AT
465{
466 struct passwd *pass;
b5bd5542
WD
467 if (!name || !*name)
468 return 0;
8ef4ffd6
AT
469 pass = getpwnam(name);
470 if (pass) {
471 *uid = pass->pw_uid;
472 return 1;
473 }
474 return 0;
475}
476
ac13ad10 477/** Turn a group name into a gid */
4a19c3b2 478int name_to_gid(const char *name, gid_t *gid)
8ef4ffd6
AT
479{
480 struct group *grp;
b5bd5542
WD
481 if (!name || !*name)
482 return 0;
8ef4ffd6
AT
483 grp = getgrnam(name);
484 if (grp) {
485 *gid = grp->gr_gid;
486 return 1;
487 }
488 return 0;
489}
490
ac13ad10 491/** Lock a byte range in a open file */
31593dd6 492int lock_range(int fd, int offset, int len)
0c515f17 493{
31593dd6 494 struct flock lock;
0c515f17 495
31593dd6
AT
496 lock.l_type = F_WRLCK;
497 lock.l_whence = SEEK_SET;
498 lock.l_start = offset;
499 lock.l_len = len;
500 lock.l_pid = 0;
5cb37436 501
31593dd6 502 return fcntl(fd,F_SETLK,&lock) == 0;
0c515f17 503}
874895d5 504
7842418b 505static int filter_server_path(char *arg)
4791825d
WD
506{
507 char *s;
4791825d 508
7842418b 509 if (server_filter_list.head) {
4791825d
WD
510 for (s = arg; (s = strchr(s, '/')) != NULL; ) {
511 *s = '\0';
7842418b 512 if (check_filter(&server_filter_list, arg, 1) < 0) {
4791825d
WD
513 /* We must leave arg truncated! */
514 return 1;
515 }
516 *s++ = '/';
517 }
518 }
519 return 0;
520}
874895d5 521
b7061c82
WD
522static void glob_expand_one(char *s, char ***argv_ptr, int *argc_ptr,
523 int *maxargs_ptr)
874895d5 524{
b7061c82 525 char **argv = *argv_ptr;
b5bd5542 526 int argc = *argc_ptr;
b7061c82 527 int maxargs = *maxargs_ptr;
4f5b0756 528#if !defined HAVE_GLOB || !defined HAVE_GLOB_H
b7061c82
WD
529 if (argc == maxargs) {
530 maxargs += MAX_ARGS;
531 if (!(argv = realloc_array(argv, char *, maxargs)))
532 out_of_memory("glob_expand_one");
533 *argv_ptr = argv;
534 *maxargs_ptr = maxargs;
535 }
4135d091
WD
536 if (!*s)
537 s = ".";
b5bd5542 538 s = argv[argc++] = strdup(s);
7842418b 539 filter_server_path(s);
874895d5
AT
540#else
541 glob_t globbuf;
874895d5 542
b5bd5542
WD
543 if (maxargs <= argc)
544 return;
4135d091
WD
545 if (!*s)
546 s = ".";
e42c9458 547
4135d091 548 if (sanitize_paths)
a8167c66 549 s = sanitize_path(NULL, s, "", 0, NULL);
84a63795
WD
550 else
551 s = strdup(s);
087bf010 552
5cb37436 553 memset(&globbuf, 0, sizeof globbuf);
7842418b 554 if (!filter_server_path(s))
4791825d 555 glob(s, 0, NULL, &globbuf);
b7061c82
WD
556 if (MAX((int)globbuf.gl_pathc, 1) > maxargs - argc) {
557 maxargs += globbuf.gl_pathc + MAX_ARGS;
558 if (!(argv = realloc_array(argv, char *, maxargs)))
559 out_of_memory("glob_expand_one");
560 *argv_ptr = argv;
561 *maxargs_ptr = maxargs;
562 }
b5bd5542
WD
563 if (globbuf.gl_pathc == 0)
564 argv[argc++] = s;
565 else {
c8933031 566 int i;
b5bd5542 567 free(s);
c8933031 568 for (i = 0; i < (int)globbuf.gl_pathc; i++) {
b5bd5542
WD
569 if (!(argv[argc++] = strdup(globbuf.gl_pathv[i])))
570 out_of_memory("glob_expand_one");
571 }
874895d5
AT
572 }
573 globfree(&globbuf);
874895d5 574#endif
b5bd5542 575 *argc_ptr = argc;
874895d5 576}
5a96ee05 577
4791825d 578/* This routine is only used in daemon mode. */
b7061c82 579void glob_expand(char *base1, char ***argv_ptr, int *argc_ptr, int *maxargs_ptr)
087bf010 580{
b7061c82 581 char *s = (*argv_ptr)[*argc_ptr];
087bf010 582 char *p, *q;
ba5e128d 583 char *base = base1;
4791825d 584 int base_len = strlen(base);
087bf010 585
b5bd5542
WD
586 if (!s || !*s)
587 return;
087bf010 588
4791825d
WD
589 if (strncmp(s, base, base_len) == 0)
590 s += base_len;
e42c9458 591
b5bd5542
WD
592 if (!(s = strdup(s)))
593 out_of_memory("glob_expand");
087bf010 594
b5bd5542
WD
595 if (asprintf(&base," %s/", base1) <= 0)
596 out_of_memory("glob_expand");
4791825d 597 base_len++;
ba5e128d 598
b5bd5542
WD
599 for (q = s; *q; q = p + base_len) {
600 if ((p = strstr(q, base)) != NULL)
601 *p = '\0'; /* split it at this point */
b7061c82 602 glob_expand_one(q, argv_ptr, argc_ptr, maxargs_ptr);
b5bd5542
WD
603 if (!p)
604 break;
087bf010
AT
605 }
606
087bf010 607 free(s);
ba5e128d 608 free(base);
087bf010 609}
5a96ee05 610
ac13ad10
MP
611/**
612 * Convert a string to lower case
613 **/
5a96ee05
AT
614void strlower(char *s)
615{
616 while (*s) {
2dc7b8bd
WD
617 if (isUpper(s))
618 *s = toLower(s);
5a96ee05
AT
619 s++;
620 }
621}
e42c9458 622
368ad70e
WD
623/* Join strings p1 & p2 into "dest" with a guaranteed '/' between them. (If
624 * p1 ends with a '/', no extra '/' is inserted.) Returns the length of both
a8f7e4b8
WD
625 * strings + 1 (if '/' was inserted), regardless of whether the null-terminated
626 * string fits into destsize. */
368ad70e
WD
627size_t pathjoin(char *dest, size_t destsize, const char *p1, const char *p2)
628{
629 size_t len = strlcpy(dest, p1, destsize);
630 if (len < destsize - 1) {
631 if (!len || dest[len-1] != '/')
632 dest[len++] = '/';
633 if (len < destsize - 1)
634 len += strlcpy(dest + len, p2, destsize - len);
635 else {
636 dest[len] = '\0';
637 len += strlen(p2);
638 }
639 }
640 else
641 len += strlen(p2) + 1; /* Assume we'd insert a '/'. */
642 return len;
643}
644
645/* Join any number of strings together, putting them in "dest". The return
a8f7e4b8
WD
646 * value is the length of all the strings, regardless of whether the null-
647 * terminated whole fits in destsize. Your list of string pointers must end
648 * with a NULL to indicate the end of the list. */
368ad70e
WD
649size_t stringjoin(char *dest, size_t destsize, ...)
650{
5cb37436 651 va_list ap;
368ad70e
WD
652 size_t len, ret = 0;
653 const char *src;
654
655 va_start(ap, destsize);
656 while (1) {
657 if (!(src = va_arg(ap, const char *)))
658 break;
659 len = strlen(src);
660 ret += len;
661 if (destsize > 1) {
662 if (len >= destsize)
663 len = destsize - 1;
664 memcpy(dest, src, len);
665 destsize -= len;
666 dest += len;
667 }
668 }
669 *dest = '\0';
670 va_end(ap);
671
672 return ret;
673}
674
1d6b8f9a
WD
675int count_dir_elements(const char *p)
676{
677 int cnt = 0, new_component = 1;
678 while (*p) {
679 if (*p++ == '/')
a8167c66 680 new_component = (*p != '.' || (p[1] != '/' && p[1] != '\0'));
1d6b8f9a
WD
681 else if (new_component) {
682 new_component = 0;
683 cnt++;
684 }
685 }
686 return cnt;
687}
688
b92693da
WD
689/* Turns multiple adjacent slashes into a single slash, gets rid of "./"
690 * elements (but not a trailing dot dir), removes a trailing slash, and
691 * optionally collapses ".." elements (except for those at the start of the
692 * string). If the resulting name would be empty, change it into a ".". */
693unsigned int clean_fname(char *name, BOOL collapse_dot_dot)
5243c216 694{
e012b94f 695 char *limit = name - 1, *t = name, *f = name;
ebdd24d6 696 int anchored;
5243c216 697
b5bd5542 698 if (!name)
3104620c 699 return 0;
5243c216 700
ebdd24d6
WD
701 if ((anchored = *f == '/') != 0)
702 *t++ = *f++;
703 while (*f) {
704 /* discard extra slashes */
705 if (*f == '/') {
706 f++;
707 continue;
5243c216 708 }
ebdd24d6
WD
709 if (*f == '.') {
710 /* discard "." dirs (but NOT a trailing '.'!) */
711 if (f[1] == '/') {
e012b94f 712 f += 2;
ebdd24d6
WD
713 continue;
714 }
715 /* collapse ".." dirs */
b92693da
WD
716 if (collapse_dot_dot
717 && f[1] == '.' && (f[2] == '/' || !f[2])) {
ebdd24d6
WD
718 char *s = t - 1;
719 if (s == name && anchored) {
720 f += 2;
721 continue;
722 }
723 while (s > limit && *--s != '/') {}
e012b94f 724 if (s != t - 1 && (s < name || *s == '/')) {
ebdd24d6
WD
725 t = s + 1;
726 f += 2;
727 continue;
728 }
f55c2dfc 729 limit = t + 2;
5243c216
AT
730 }
731 }
ebdd24d6 732 while (*f && (*t++ = *f++) != '/') {}
5243c216 733 }
ebdd24d6
WD
734
735 if (t > name+anchored && t[-1] == '/')
736 t--;
737 if (t == name)
738 *t++ = '.';
739 *t = '\0';
3104620c
WD
740
741 return t - name;
5243c216
AT
742}
743
84a63795
WD
744/* Make path appear as if a chroot had occurred. This handles a leading
745 * "/" (either removing it or expanding it) and any leading or embedded
746 * ".." components that attempt to escape past the module's top dir.
b4235b31 747 *
1d6b8f9a
WD
748 * If dest is NULL, a buffer is allocated to hold the result. It is legal
749 * to call with the dest and the path (p) pointing to the same buffer, but
750 * rootdir will be ignored to avoid expansion of the string.
b4235b31 751 *
1d6b8f9a
WD
752 * The rootdir string contains a value to use in place of a leading slash.
753 * Specify NULL to get the default of lp_path(module_id).
ac13ad10 754 *
a8167c66
WD
755 * The depth var is a count of how many '..'s to allow at the start of the
756 * path. If symlink is set, combine its value with the "p" value to get
757 * the target path, and **return NULL if any '..'s try to escape**.
ac13ad10 758 *
b92693da 759 * We also clean the path in a manner similar to clean_fname() but with a
0f78b815 760 * few differences:
b92693da
WD
761 *
762 * Turns multiple adjacent slashes into a single slash, gets rid of "." dir
763 * elements (INCLUDING a trailing dot dir), PRESERVES a trailing slash, and
764 * ALWAYS collapses ".." elements (except for those at the start of the
765 * string up to "depth" deep). If the resulting name would be empty,
766 * change it into a ".". */
a8167c66
WD
767char *sanitize_path(char *dest, const char *p, const char *rootdir, int depth,
768 const char *symlink)
1b8e662a 769{
a8167c66 770 char *start, *sanp, *save_dest = dest;
7e43da81 771 int rlen = 0, leave_one_dotdir = relative_paths;
84a63795 772
a8167c66
WD
773 if (symlink && *symlink == '/') {
774 p = symlink;
775 symlink = "";
776 }
777
84a63795
WD
778 if (dest != p) {
779 int plen = strlen(p);
1d6b8f9a
WD
780 if (*p == '/') {
781 if (!rootdir)
782 rootdir = lp_path(module_id);
783 rlen = strlen(rootdir);
784 depth = 0;
84a63795
WD
785 p++;
786 }
787 if (dest) {
788 if (rlen + plen + 1 >= MAXPATHLEN)
789 return NULL;
790 } else if (!(dest = new_array(char, rlen + plen + 1)))
791 out_of_memory("sanitize_path");
792 if (rlen) {
1d6b8f9a 793 memcpy(dest, rootdir, rlen);
84a63795
WD
794 if (rlen > 1)
795 dest[rlen++] = '/';
796 }
797 }
cb13abfe 798
84a63795 799 start = sanp = dest + rlen;
a8167c66
WD
800 while (1) {
801 if (*p == '\0') {
802 if (!symlink || !*symlink)
803 break;
804 while (sanp != start && sanp[-1] != '/') {
805 /* strip last element */
806 sanp--;
807 }
808 /* Append a relative symlink */
809 p = symlink;
810 symlink = "";
811 }
2d41264e
WD
812 /* discard leading or extra slashes */
813 if (*p == '/') {
814 p++;
815 continue;
816 }
b5f9e67d 817 /* this loop iterates once per filename component in p.
44e2e578 818 * both p (and sanp if the original had a slash) should
b5f9e67d
DD
819 * always be left pointing after a slash
820 */
c284f34a 821 if (*p == '.' && (p[1] == '/' || p[1] == '\0')) {
35812ea1 822 if (leave_one_dotdir && p[1])
7e43da81
WD
823 leave_one_dotdir = 0;
824 else {
825 /* skip "." component */
826 p++;
827 continue;
828 }
cb13abfe 829 }
c284f34a 830 if (*p == '.' && p[1] == '.' && (p[2] == '/' || p[2] == '\0')) {
cb13abfe 831 /* ".." component followed by slash or end */
8e5f029e 832 if (depth <= 0 || sanp != start) {
a8167c66
WD
833 if (symlink && sanp == start) {
834 if (!save_dest)
835 free(dest);
836 return NULL;
837 }
8e5f029e
WD
838 p += 2;
839 if (sanp != start) {
840 /* back up sanp one level */
841 --sanp; /* now pointing at slash */
842 while (sanp > start && sanp[-1] != '/') {
843 /* skip back up to slash */
844 sanp--;
845 }
b5f9e67d 846 }
8e5f029e 847 continue;
1b8e662a 848 }
8e5f029e
WD
849 /* allow depth levels of .. at the beginning */
850 depth--;
851 /* move the virtual beginning to leave the .. alone */
852 start = sanp + 3;
1b8e662a 853 }
2d41264e
WD
854 /* copy one component through next slash */
855 while (*p && (*sanp++ = *p++) != '/') {}
1b8e662a 856 }
84a63795 857 if (sanp == dest) {
b5f9e67d 858 /* ended up with nothing, so put in "." component */
44e2e578 859 *sanp++ = '.';
b5f9e67d 860 }
44e2e578 861 *sanp = '\0';
1b8e662a 862
84a63795 863 return dest;
14b61c63 864}
5243c216 865
a8167c66 866/* Like chdir(), but it keeps track of the current directory (in the
a16d8f2b 867 * global "curr_dir"), and ensures that the path size doesn't overflow.
a8167c66 868 * Also cleans the path using the clean_fname() function. */
2c708476 869int push_dir(const char *dir, int set_path_only)
5243c216 870{
5243c216 871 static int initialised;
4af8fe4e 872 unsigned int len;
5243c216
AT
873
874 if (!initialised) {
875 initialised = 1;
5cb37436 876 getcwd(curr_dir, sizeof curr_dir - 1);
4af8fe4e 877 curr_dir_len = strlen(curr_dir);
5243c216
AT
878 }
879
4af8fe4e
WD
880 if (!dir) /* this call was probably just to initialize */
881 return 0;
c226b7c2 882
4af8fe4e
WD
883 len = strlen(dir);
884 if (len == 1 && *dir == '.')
885 return 1;
5243c216 886
4af8fe4e
WD
887 if ((*dir == '/' ? len : curr_dir_len + 1 + len) >= sizeof curr_dir)
888 return 0;
889
ba081be3 890 if (!set_path_only && chdir(dir))
4af8fe4e 891 return 0;
5243c216
AT
892
893 if (*dir == '/') {
4af8fe4e
WD
894 memcpy(curr_dir, dir, len + 1);
895 curr_dir_len = len;
896 } else {
897 curr_dir[curr_dir_len++] = '/';
898 memcpy(curr_dir + curr_dir_len, dir, len + 1);
899 curr_dir_len += len;
5243c216
AT
900 }
901
b92693da 902 curr_dir_len = clean_fname(curr_dir, 1);
a8167c66
WD
903 if (sanitize_paths) {
904 if (module_dirlen > curr_dir_len)
905 module_dirlen = curr_dir_len;
906 curr_dir_depth = count_dir_elements(curr_dir + module_dirlen);
907 }
5243c216 908
4af8fe4e 909 return 1;
5243c216
AT
910}
911
a16d8f2b
WD
912/**
913 * Reverse a push_dir() call. You must pass in an absolute path
914 * that was copied from a prior value of "curr_dir".
915 **/
2c708476 916int pop_dir(const char *dir)
5243c216 917{
4af8fe4e
WD
918 if (chdir(dir))
919 return 0;
5243c216 920
4af8fe4e
WD
921 curr_dir_len = strlcpy(curr_dir, dir, sizeof curr_dir);
922 if (curr_dir_len >= sizeof curr_dir)
923 curr_dir_len = sizeof curr_dir - 1;
a8167c66
WD
924 if (sanitize_paths)
925 curr_dir_depth = count_dir_elements(curr_dir + module_dirlen);
5243c216 926
4af8fe4e 927 return 1;
5243c216 928}
aa9b77a5 929
eb61be19
WD
930/**
931 * Return a quoted string with the full pathname of the indicated filename.
932 * The string " (in MODNAME)" may also be appended. The returned pointer
933 * remains valid until the next time full_fname() is called.
934 **/
9a5ade18 935char *full_fname(const char *fn)
eb61be19 936{
eb61be19
WD
937 static char *result = NULL;
938 char *m1, *m2, *m3;
939 char *p1, *p2;
940
941 if (result)
942 free(result);
943
944 if (*fn == '/')
945 p1 = p2 = "";
946 else {
4cff5fa4 947 p1 = curr_dir + module_dirlen;
bc83274a
WD
948 for (p2 = p1; *p2 == '/'; p2++) {}
949 if (*p2)
950 p2 = "/";
eb61be19
WD
951 }
952 if (module_id >= 0) {
953 m1 = " (in ";
954 m2 = lp_name(module_id);
955 m3 = ")";
eb61be19
WD
956 } else
957 m1 = m2 = m3 = "";
958
c98ad3df
WD
959 if (asprintf(&result, "\"%s%s%s\"%s%s%s", p1, p2, fn, m1, m2, m3) <= 0)
960 out_of_memory("full_fname");
eb61be19
WD
961
962 return result;
963}
964
a7260c40
WD
965static char partial_fname[MAXPATHLEN];
966
967char *partial_dir_fname(const char *fname)
968{
969 char *t = partial_fname;
970 int sz = sizeof partial_fname;
971 const char *fn;
972
973 if ((fn = strrchr(fname, '/')) != NULL) {
974 fn++;
975 if (*partial_dir != '/') {
976 int len = fn - fname;
977 strncpy(t, fname, len); /* safe */
978 t += len;
979 sz -= len;
980 }
981 } else
982 fn = fname;
983 if ((int)pathjoin(t, sz, partial_dir, fn) >= sz)
984 return NULL;
5aa7b20a 985 if (server_filter_list.head) {
89363676
WD
986 t = strrchr(partial_fname, '/');
987 *t = '\0';
5aa7b20a
WD
988 if (check_filter(&server_filter_list, partial_fname, 1) < 0)
989 return NULL;
89363676 990 *t = '/';
5aa7b20a
WD
991 if (check_filter(&server_filter_list, partial_fname, 0) < 0)
992 return NULL;
993 }
a7260c40
WD
994
995 return partial_fname;
996}
997
998/* If no --partial-dir option was specified, we don't need to do anything
999 * (the partial-dir is essentially '.'), so just return success. */
1000int handle_partial_dir(const char *fname, int create)
1001{
1002 char *fn, *dir;
1003
1004 if (fname != partial_fname)
1005 return 1;
1006 if (!create && *partial_dir == '/')
1007 return 1;
1008 if (!(fn = strrchr(partial_fname, '/')))
1009 return 1;
1010
1011 *fn = '\0';
1012 dir = partial_fname;
1013 if (create) {
1014 STRUCT_STAT st;
a7260c40 1015 int statret = do_lstat(dir, &st);
a7260c40
WD
1016 if (statret == 0 && !S_ISDIR(st.st_mode)) {
1017 if (do_unlink(dir) < 0)
1018 return 0;
1019 statret = -1;
1020 }
1021 if (statret < 0 && do_mkdir(dir, 0700) < 0)
1022 return 0;
1023 } else
1024 do_rmdir(dir);
1025 *fn = '/';
1026
1027 return 1;
1028}
1029
ac13ad10
MP
1030/**
1031 * Determine if a symlink points outside the current directory tree.
036e70b0
MP
1032 * This is considered "unsafe" because e.g. when mirroring somebody
1033 * else's machine it might allow them to establish a symlink to
1034 * /etc/passwd, and then read it through a web server.
1035 *
4e5db0ad
MP
1036 * Null symlinks and absolute symlinks are always unsafe.
1037 *
1038 * Basically here we are concerned with symlinks whose target contains
1039 * "..", because this might cause us to walk back up out of the
1040 * transferred directory. We are not allowed to go back up and
1041 * reenter.
1042 *
036e70b0
MP
1043 * @param dest Target of the symlink in question.
1044 *
25d34a5c 1045 * @param src Top source directory currently applicable. Basically this
036e70b0 1046 * is the first parameter to rsync in a simple invocation, but it's
25d34a5c 1047 * modified by flist.c in slightly complex ways.
036e70b0
MP
1048 *
1049 * @retval True if unsafe
1050 * @retval False is unsafe
4e5db0ad
MP
1051 *
1052 * @sa t_unsafe.c
ac13ad10 1053 **/
7afa3a4a 1054int unsafe_symlink(const char *dest, const char *src)
4b957c22 1055{
7afa3a4a 1056 const char *name, *slash;
4b957c22
AT
1057 int depth = 0;
1058
1059 /* all absolute and null symlinks are unsafe */
b5bd5542
WD
1060 if (!dest || !*dest || *dest == '/')
1061 return 1;
4b957c22
AT
1062
1063 /* find out what our safety margin is */
7afa3a4a
WD
1064 for (name = src; (slash = strchr(name, '/')) != 0; name = slash+1) {
1065 if (strncmp(name, "../", 3) == 0) {
c284f34a 1066 depth = 0;
7afa3a4a 1067 } else if (strncmp(name, "./", 2) == 0) {
4b957c22
AT
1068 /* nothing */
1069 } else {
1070 depth++;
1071 }
1072 }
7afa3a4a
WD
1073 if (strcmp(name, "..") == 0)
1074 depth = 0;
4b957c22 1075
7afa3a4a
WD
1076 for (name = dest; (slash = strchr(name, '/')) != 0; name = slash+1) {
1077 if (strncmp(name, "../", 3) == 0) {
1078 /* if at any point we go outside the current directory
1079 then stop - it is unsafe */
1080 if (--depth < 0)
1081 return 1;
1082 } else if (strncmp(name, "./", 2) == 0) {
4b957c22
AT
1083 /* nothing */
1084 } else {
1085 depth++;
1086 }
4b957c22 1087 }
7afa3a4a
WD
1088 if (strcmp(name, "..") == 0)
1089 depth--;
4b957c22 1090
4b957c22
AT
1091 return (depth < 0);
1092}
375a4556 1093
e175fb07
WD
1094/* Return the int64 number as a string. If the --human-readable option was
1095 * specified, we may output the number in K, M, or G units. We can return
1096 * up to 4 buffers at a time. */
1097char *human_num(int64 num)
1098{
1099 static char bufs[4][128]; /* more than enough room */
1100 static unsigned int n;
1101 char *s;
1102
1103 n = (n + 1) % (sizeof bufs / sizeof bufs[0]);
1104
1105 if (human_readable) {
1106 char units = '\0';
7794db7c 1107 int mult = human_readable == 1 ? 1000 : 1024;
e175fb07
WD
1108 double dnum = 0;
1109 if (num > mult*mult*mult) {
1110 dnum = (double)num / (mult*mult*mult);
1111 units = 'G';
1112 } else if (num > mult*mult) {
1113 dnum = (double)num / (mult*mult);
1114 units = 'M';
1115 } else if (num > mult) {
1116 dnum = (double)num / mult;
1117 units = 'K';
1118 }
1119 if (units) {
83078af5 1120 snprintf(bufs[n], sizeof bufs[0], "%.2f%c", dnum, units);
e175fb07
WD
1121 return bufs[n];
1122 }
1123 }
1124
1125 s = bufs[n] + sizeof bufs[0] - 1;
1126 *s = '\0';
1127
1128 if (!num)
1129 *--s = '0';
1130 while (num) {
26404276 1131 *--s = (char)(num % 10) + '0';
e175fb07
WD
1132 num /= 10;
1133 }
1134 return s;
1135}
1136
1137/* Return the double number as a string. If the --human-readable option was
1138 * specified, we may output the number in K, M, or G units. We use a buffer
1139 * from human_num() to return our result. */
1140char *human_dnum(double dnum, int decimal_digits)
1141{
1142 char *buf = human_num(dnum);
1143 int len = strlen(buf);
2dc7b8bd 1144 if (isDigit(buf + len - 1)) {
e175fb07
WD
1145 /* There's extra room in buf prior to the start of the num. */
1146 buf -= decimal_digits + 1;
1147 snprintf(buf, len + decimal_digits + 2, "%.*f", decimal_digits, dnum);
1148 }
1149 return buf;
1150}
f7632fc6 1151
4a19c3b2 1152/* Return the date and time as a string. Some callers tweak returned buf. */
f7632fc6
AT
1153char *timestring(time_t t)
1154{
1155 static char TimeBuf[200];
1156 struct tm *tm = localtime(&t);
afa73c75 1157 char *p;
f7632fc6 1158
4f5b0756 1159#ifdef HAVE_STRFTIME
5cb37436 1160 strftime(TimeBuf, sizeof TimeBuf - 1, "%Y/%m/%d %H:%M:%S", tm);
f7632fc6 1161#else
5cb37436 1162 strlcpy(TimeBuf, asctime(tm), sizeof TimeBuf);
f7632fc6
AT
1163#endif
1164
afa73c75
WD
1165 if ((p = strchr(TimeBuf, '\n')) != NULL)
1166 *p = '\0';
f7632fc6 1167
afa73c75 1168 return TimeBuf;
f7632fc6
AT
1169}
1170
e1bd49d6
MP
1171/**
1172 * Sleep for a specified number of milliseconds.
1173 *
1174 * Always returns TRUE. (In the future it might return FALSE if
1175 * interrupted.)
1176 **/
1177int msleep(int t)
9ec16c83 1178{
c284f34a
WD
1179 int tdiff = 0;
1180 struct timeval tval, t1, t2;
9ec16c83
AT
1181
1182 gettimeofday(&t1, NULL);
5cb37436 1183
9ec16c83
AT
1184 while (tdiff < t) {
1185 tval.tv_sec = (t-tdiff)/1000;
1186 tval.tv_usec = 1000*((t-tdiff)%1000);
5cb37436 1187
9ec16c83
AT
1188 errno = 0;
1189 select(0,NULL,NULL, NULL, &tval);
1190
1191 gettimeofday(&t2, NULL);
5cb37436 1192 tdiff = (t2.tv_sec - t1.tv_sec)*1000 +
9ec16c83
AT
1193 (t2.tv_usec - t1.tv_usec)/1000;
1194 }
e1bd49d6
MP
1195
1196 return True;
9ec16c83
AT
1197}
1198
c2b54076
WD
1199/* Determine if two time_t values are equivalent (either exact, or in
1200 * the modification timestamp window established by --modify-window).
ac13ad10
MP
1201 *
1202 * @retval 0 if the times should be treated as the same
1203 *
1204 * @retval +1 if the first is later
1205 *
1206 * @retval -1 if the 2nd is later
1207 **/
c2b54076 1208int cmp_time(time_t file1, time_t file2)
5b56cc19 1209{
5b56cc19 1210 if (file2 > file1) {
bc6ebcd2
WD
1211 if (file2 - file1 <= modify_window)
1212 return 0;
5b56cc19
AT
1213 return -1;
1214 }
bc6ebcd2
WD
1215 if (file1 - file2 <= modify_window)
1216 return 0;
5b56cc19
AT
1217 return 1;
1218}
1219
1220
1221#ifdef __INSURE__XX
0f8f98c8
AT
1222#include <dlfcn.h>
1223
ac13ad10
MP
1224/**
1225 This routine is a trick to immediately catch errors when debugging
1226 with insure. A xterm with a gdb is popped up when insure catches
1227 a error. It is Linux specific.
1228**/
0f8f98c8
AT
1229int _Insure_trap_error(int a1, int a2, int a3, int a4, int a5, int a6)
1230{
1231 static int (*fn)();
1232 int ret;
8950ac03 1233 char *cmd;
0f8f98c8 1234
5cb37436 1235 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'",
0f8f98c8
AT
1236 getpid(), getpid(), getpid());
1237
1238 if (!fn) {
1239 static void *h;
1240 h = dlopen("/usr/local/parasoft/insure++lite/lib.linux2/libinsure.so", RTLD_LAZY);
1241 fn = dlsym(h, "_Insure_trap_error");
1242 }
1243
1244 ret = fn(a1, a2, a3, a4, a5, a6);
1245
1246 system(cmd);
1247
8950ac03
AT
1248 free(cmd);
1249
0f8f98c8
AT
1250 return ret;
1251}
1252#endif
58cadc86 1253
58cadc86
WD
1254#define MALLOC_MAX 0x40000000
1255
1256void *_new_array(unsigned int size, unsigned long num)
1257{
1258 if (num >= MALLOC_MAX/size)
1259 return NULL;
1260 return malloc(size * num);
1261}
1262
1263void *_realloc_array(void *ptr, unsigned int size, unsigned long num)
1264{
1265 if (num >= MALLOC_MAX/size)
1266 return NULL;
58cadc86
WD
1267 if (!ptr)
1268 return malloc(size * num);
1269 return realloc(ptr, size * num);
1270}
e64ae6d7
WD
1271
1272/* Take a filename and filename length and return the most significant
1273 * filename suffix we can find. This ignores suffixes such as "~",
1274 * ".bak", ".orig", ".~1~", etc. */
1275const char *find_filename_suffix(const char *fn, int fn_len, int *len_ptr)
1276{
1277 const char *suf, *s;
1278 BOOL had_tilde;
1279 int s_len;
1280
1281 /* One or more dots at the start aren't a suffix. */
1282 while (fn_len && *fn == '.') fn++, fn_len--;
1283
1284 /* Ignore the ~ in a "foo~" filename. */
1285 if (fn_len > 1 && fn[fn_len-1] == '~')
1286 fn_len--, had_tilde = True;
1287 else
1288 had_tilde = False;
1289
1290 /* Assume we don't find an suffix. */
1291 suf = "";
1292 *len_ptr = 0;
1293
1294 /* Find the last significant suffix. */
1295 for (s = fn + fn_len; fn_len > 1; ) {
1296 while (*--s != '.' && s != fn) {}
1297 if (s == fn)
1298 break;
1299 s_len = fn_len - (s - fn);
1300 fn_len = s - fn;
6012eaa1 1301 if (s_len == 4) {
e64ae6d7
WD
1302 if (strcmp(s+1, "bak") == 0
1303 || strcmp(s+1, "old") == 0)
1304 continue;
6012eaa1 1305 } else if (s_len == 5) {
e64ae6d7
WD
1306 if (strcmp(s+1, "orig") == 0)
1307 continue;
1308 } else if (s_len > 2 && had_tilde
2dc7b8bd 1309 && s[1] == '~' && isDigit(s + 2))
e64ae6d7
WD
1310 continue;
1311 *len_ptr = s_len;
1312 suf = s;
1313 if (s_len == 1)
1314 break;
1315 /* Determine if the suffix is all digits. */
1316 for (s++, s_len--; s_len > 0; s++, s_len--) {
2dc7b8bd 1317 if (!isDigit(s))
e64ae6d7
WD
1318 return suf;
1319 }
1320 /* An all-digit suffix may not be that signficant. */
1321 s = suf;
1322 }
1323
1324 return suf;
1325}
1326
1327/* This is an implementation of the Levenshtein distance algorithm. It
1328 * was implemented to avoid needing a two-dimensional matrix (to save
1329 * memory). It was also tweaked to try to factor in the ASCII distance
1330 * between changed characters as a minor distance quantity. The normal
1331 * Levenshtein units of distance (each signifying a single change between
1332 * the two strings) are defined as a "UNIT". */
1333
1334#define UNIT (1 << 16)
1335
1336uint32 fuzzy_distance(const char *s1, int len1, const char *s2, int len2)
1337{
1338 uint32 a[MAXPATHLEN], diag, above, left, diag_inc, above_inc, left_inc;
1339 int32 cost;
1340 int i1, i2;
1341
1342 if (!len1 || !len2) {
1343 if (!len1) {
1344 s1 = s2;
1345 len1 = len2;
1346 }
1347 for (i1 = 0, cost = 0; i1 < len1; i1++)
1348 cost += s1[i1];
1349 return (int32)len1 * UNIT + cost;
1350 }
1351
1352 for (i2 = 0; i2 < len2; i2++)
1353 a[i2] = (i2+1) * UNIT;
1354
1355 for (i1 = 0; i1 < len1; i1++) {
1356 diag = i1 * UNIT;
1357 above = (i1+1) * UNIT;
1358 for (i2 = 0; i2 < len2; i2++) {
1359 left = a[i2];
1360 if ((cost = *((uchar*)s1+i1) - *((uchar*)s2+i2)) != 0) {
1361 if (cost < 0)
1362 cost = UNIT - cost;
1363 else
1364 cost = UNIT + cost;
1365 }
1366 diag_inc = diag + cost;
1367 left_inc = left + UNIT + *((uchar*)s1+i1);
1368 above_inc = above + UNIT + *((uchar*)s2+i2);
1369 a[i2] = above = left < above
1370 ? (left_inc < diag_inc ? left_inc : diag_inc)
1371 : (above_inc < diag_inc ? above_inc : diag_inc);
1372 diag = left;
1373 }
1374 }
1375
1376 return a[len2-1];
1377}
c2b54076
WD
1378
1379#define BB_SLOT_SIZE (16*1024) /* Desired size in bytes */
1380#define BB_PER_SLOT_BITS (BB_SLOT_SIZE * 8) /* Number of bits per slot */
1381#define BB_PER_SLOT_INTS (BB_SLOT_SIZE / 4) /* Number of int32s per slot */
1382
1383struct bitbag {
1384 uint32 **bits;
1385 int slot_cnt;
1386};
1387
1388struct bitbag *bitbag_create(int max_ndx)
1389{
1390 struct bitbag *bb = new(struct bitbag);
1391 bb->slot_cnt = (max_ndx + BB_PER_SLOT_BITS - 1) / BB_PER_SLOT_BITS;
1392
1393 if (!(bb->bits = (uint32**)calloc(bb->slot_cnt, sizeof (uint32*))))
1394 out_of_memory("bitbag_create");
1395
1396 return bb;
1397}
1398
1399void bitbag_set_bit(struct bitbag *bb, int ndx)
1400{
1401 int slot = ndx / BB_PER_SLOT_BITS;
1402 ndx %= BB_PER_SLOT_BITS;
1403
1404 if (!bb->bits[slot]) {
1405 if (!(bb->bits[slot] = (uint32*)calloc(BB_PER_SLOT_INTS, 4)))
1406 out_of_memory("bitbag_set_bit");
1407 }
1408
1409 bb->bits[slot][ndx/32] |= 1u << (ndx % 32);
1410}
1411
1412#if 0 /* not needed yet */
1413void bitbag_clear_bit(struct bitbag *bb, int ndx)
1414{
1415 int slot = ndx / BB_PER_SLOT_BITS;
1416 ndx %= BB_PER_SLOT_BITS;
1417
1418 if (!bb->bits[slot])
1419 return;
1420
1421 bb->bits[slot][ndx/32] &= ~(1u << (ndx % 32));
1422}
1423
1424int bitbag_check_bit(struct bitbag *bb, int ndx)
1425{
1426 int slot = ndx / BB_PER_SLOT_BITS;
1427 ndx %= BB_PER_SLOT_BITS;
1428
1429 if (!bb->bits[slot])
1430 return 0;
1431
1432 return bb->bits[slot][ndx/32] & (1u << (ndx % 32)) ? 1 : 0;
1433}
1434#endif
1435
1436/* Call this with -1 to start checking from 0. Returns -1 at the end. */
1437int bitbag_next_bit(struct bitbag *bb, int after)
1438{
1439 uint32 bits, mask;
1440 int i, ndx = after + 1;
1441 int slot = ndx / BB_PER_SLOT_BITS;
1442 ndx %= BB_PER_SLOT_BITS;
1443
1444 mask = (1u << (ndx % 32)) - 1;
1445 for (i = ndx / 32; slot < bb->slot_cnt; slot++, i = mask = 0) {
1446 if (!bb->bits[slot])
1447 continue;
1448 for ( ; i < BB_PER_SLOT_INTS; i++, mask = 0) {
1449 if (!(bits = bb->bits[slot][i] & ~mask))
1450 continue;
1451 /* The xor magic figures out the lowest enabled bit in
1452 * bits, and the switch quickly computes log2(bit). */
1453 switch (bits ^ (bits & (bits-1))) {
1454#define LOG2(n) case 1u << n: return slot*BB_PER_SLOT_BITS + i*32 + n
1455 LOG2(0); LOG2(1); LOG2(2); LOG2(3);
1456 LOG2(4); LOG2(5); LOG2(6); LOG2(7);
1457 LOG2(8); LOG2(9); LOG2(10); LOG2(11);
1458 LOG2(12); LOG2(13); LOG2(14); LOG2(15);
1459 LOG2(16); LOG2(17); LOG2(18); LOG2(19);
1460 LOG2(20); LOG2(21); LOG2(22); LOG2(23);
1461 LOG2(24); LOG2(25); LOG2(26); LOG2(27);
1462 LOG2(28); LOG2(29); LOG2(30); LOG2(31);
1463 }
1464 return -1; /* impossible... */
1465 }
1466 }
1467
1468 return -1;
1469}