Added get_xattr_acl(), set_xattr_acl(), and del_def_xattr_acl().
[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>
ba2133d6 7 * Copyright (C) 2003-2007 Wayne Davison
5cb37436 8 *
0ecfbf27 9 * This program is free software; you can redistribute it and/or modify
8e41b68e
WD
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.
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 19 * You should have received a copy of the GNU General Public License along
4fd842f9 20 * with this program; if not, visit the http://fsf.org website.
0ecfbf27 21 */
c627d613 22
c627d613 23#include "rsync.h"
1b42f628 24#include "ifuncs.h"
c627d613 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;
7c73536c 32extern char *module_dir;
a8167c66 33extern unsigned int module_dirlen;
e9357a2d 34extern mode_t orig_umask;
a7260c40 35extern char *partial_dir;
7842418b 36extern struct filter_list_struct server_filter_list;
c7c11a0d 37
0ecfbf27
MP
38int sanitize_paths = 0;
39
a8167c66
WD
40char curr_dir[MAXPATHLEN];
41unsigned int curr_dir_len;
42int curr_dir_depth; /* This is only set for a sanitizing daemon. */
43
0f78b815 44/* Set a fd into nonblocking mode. */
f0359dd0
AT
45void set_nonblocking(int fd)
46{
47 int val;
48
ff530f04 49 if ((val = fcntl(fd, F_GETFL)) == -1)
f0359dd0
AT
50 return;
51 if (!(val & NONBLOCK_FLAG)) {
52 val |= NONBLOCK_FLAG;
53 fcntl(fd, F_SETFL, val);
54 }
55}
56
0f78b815 57/* Set a fd into blocking mode. */
36349ea0
AT
58void set_blocking(int fd)
59{
60 int val;
61
ff530f04 62 if ((val = fcntl(fd, F_GETFL)) == -1)
36349ea0
AT
63 return;
64 if (val & NONBLOCK_FLAG) {
65 val &= ~NONBLOCK_FLAG;
66 fcntl(fd, F_SETFL, val);
67 }
68}
69
ac13ad10 70/**
0ecfbf27
MP
71 * Create a file descriptor pair - like pipe() but use socketpair if
72 * possible (because of blocking issues on pipes).
5cb37436 73 *
0ecfbf27 74 * Always set non-blocking.
f0359dd0 75 */
08f15335
AT
76int fd_pair(int fd[2])
77{
f0359dd0
AT
78 int ret;
79
4f5b0756 80#ifdef HAVE_SOCKETPAIR
f0359dd0 81 ret = socketpair(AF_UNIX, SOCK_STREAM, 0, fd);
08f15335 82#else
f0359dd0 83 ret = pipe(fd);
08f15335 84#endif
f0359dd0
AT
85
86 if (ret == 0) {
87 set_nonblocking(fd[0]);
88 set_nonblocking(fd[1]);
89 }
0ecfbf27 90
f0359dd0 91 return ret;
08f15335
AT
92}
93
0b515981 94void print_child_argv(const char *prefix, char **cmd)
5ad0e46f 95{
0b515981 96 rprintf(FCLIENT, "%s ", prefix);
5ad0e46f
MP
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)) {
a3f6dbdf 105 rprintf(FCLIENT, "\"%s\" ", *cmd);
5ad0e46f 106 } else {
a3f6dbdf 107 rprintf(FCLIENT, "%s ", *cmd);
5ad0e46f
MP
108 }
109 }
a3f6dbdf 110 rprintf(FCLIENT, "\n");
5ad0e46f
MP
111}
112
4a19c3b2 113NORETURN void out_of_memory(const char *str)
c627d613 114{
2cd421d8 115 rprintf(FERROR, "ERROR: out of memory in %s [%s]\n", str, who_am_i());
c284f34a 116 exit_cleanup(RERR_MALLOC);
575f2fca
AT
117}
118
4a19c3b2 119NORETURN void overflow_exit(const char *str)
575f2fca 120{
2cd421d8 121 rprintf(FERROR, "ERROR: buffer overflow in %s [%s]\n", str, who_am_i());
c284f34a 122 exit_cleanup(RERR_MALLOC);
c627d613
AT
123}
124
4a19c3b2 125int set_modtime(const char *fname, time_t modtime, mode_t mode)
c627d613 126{
25007999
WD
127#if !defined HAVE_LUTIMES || !defined HAVE_UTIMES
128 if (S_ISLNK(mode))
129 return 1;
130#endif
131
404e813c
MP
132 if (verbose > 2) {
133 rprintf(FINFO, "set modtime of %s to (%ld) %s",
785abd48 134 fname, (long)modtime,
404e813c
MP
135 asctime(localtime(&modtime)));
136 }
5cb37436 137
15778afb
WD
138 if (dry_run)
139 return 0;
140
31e12522 141 {
25007999
WD
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
149a78e3
WD
149 if (S_ISLNK(mode)) {
150 lutimes(fname, t);
151 return 0; /* ignore errors */
152 }
25007999
WD
153# endif
154 return utimes(fname, t);
155#elif defined HAVE_UTIMBUF
5cb37436 156 struct utimbuf tbuf;
31e12522
AT
157 tbuf.actime = time(NULL);
158 tbuf.modtime = modtime;
159 return utime(fname,&tbuf);
4f5b0756 160#elif defined HAVE_UTIME
31e12522
AT
161 time_t t[2];
162 t[0] = time(NULL);
163 t[1] = modtime;
164 return utime(fname,t);
c627d613 165#else
25007999 166#error No file-time-modification routine found!
c627d613 167#endif
31e12522 168 }
c627d613 169}
94481d91 170
e9357a2d
WD
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
85c41757
WD
185/* Create any necessary directories in fname. Any missing directories are
186 * created with default permissions. */
e9357a2d 187int create_directory_path(char *fname)
6574b4f7 188{
6574b4f7 189 char *p;
85c41757 190 int ret = 0;
6574b4f7 191
c284f34a
WD
192 while (*fname == '/')
193 fname++;
194 while (strncmp(fname, "./", 2) == 0)
195 fname += 2;
6574b4f7 196
85c41757 197 umask(orig_umask);
6574b4f7 198 p = fname;
c284f34a 199 while ((p = strchr(p,'/')) != NULL) {
85c41757
WD
200 *p = '\0';
201 if (do_mkdir(fname, ACCESSPERMS) < 0 && errno != EEXIST)
202 ret = -1;
203 *p++ = '/';
6574b4f7 204 }
85c41757
WD
205 umask(0);
206
207 return ret;
6574b4f7 208}
950ab32d 209
ac13ad10
MP
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 */
4a19c3b2 220int full_write(int desc, const char *ptr, size_t len)
950ab32d
AT
221{
222 int total_written;
5cb37436 223
950ab32d
AT
224 total_written = 0;
225 while (len > 0) {
5c1b7bfd 226 int written = write(desc, ptr, len);
950ab32d 227 if (written < 0) {
950ab32d
AT
228 if (errno == EINTR)
229 continue;
950ab32d
AT
230 return written;
231 }
232 total_written += written;
233 ptr += written;
234 len -= written;
235 }
236 return total_written;
237}
238
ac13ad10
MP
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. */
9dd891bb 250static int safe_read(int desc, char *ptr, size_t len)
950ab32d
AT
251{
252 int n_chars;
5cb37436 253
9dd891bb 254 if (len == 0)
950ab32d 255 return len;
5cb37436 256
950ab32d
AT
257 do {
258 n_chars = read(desc, ptr, len);
259 } while (n_chars < 0 && errno == EINTR);
5cb37436 260
950ab32d
AT
261 return n_chars;
262}
263
ac13ad10
MP
264/** Copy a file.
265 *
3e13004b
WD
266 * This is used in conjunction with the --temp-dir, --backup, and
267 * --copy-dest options. */
f5b532b1 268int copy_file(const char *source, const char *dest, mode_t mode, int create_bak_dir)
950ab32d
AT
269{
270 int ifd;
271 int ofd;
272 char buf[1024 * 8];
273 int len; /* Number of bytes read into `buf'. */
274
f5b532b1 275 if ((ifd = do_open(source, O_RDONLY, 0)) < 0) {
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
f5b532b1
WD
285 if ((ofd = do_open(dest, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, mode)) < 0
286 && (!create_bak_dir || errno != ENOENT || make_bak_dir(dest) < 0
287 || (ofd = do_open(dest, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, mode)) < 0)) {
d62bcc17 288 rsyserr(FERROR, errno, "open %s", full_fname(dest));
950ab32d
AT
289 close(ifd);
290 return -1;
291 }
292
5cb37436 293 while ((len = safe_read(ifd, buf, sizeof buf)) > 0) {
950ab32d 294 if (full_write(ofd, buf, len) < 0) {
d62bcc17 295 rsyserr(FERROR, errno, "write %s", full_fname(dest));
950ab32d
AT
296 close(ifd);
297 close(ofd);
298 return -1;
299 }
300 }
301
8b602edd 302 if (len < 0) {
d62bcc17 303 rsyserr(FERROR, errno, "read %s", full_fname(source));
8b602edd
WD
304 close(ifd);
305 close(ofd);
306 return -1;
307 }
308
9f27cd8c 309 if (close(ifd) < 0) {
d62bcc17
WD
310 rsyserr(FINFO, errno, "close failed on %s",
311 full_fname(source));
9f27cd8c
WD
312 }
313
314 if (close(ofd) < 0) {
d62bcc17
WD
315 rsyserr(FERROR, errno, "close failed on %s",
316 full_fname(dest));
9f27cd8c
WD
317 return -1;
318 }
950ab32d 319
950ab32d
AT
320 return 0;
321}
feaa89c4 322
c7c11a0d
DD
323/* MAX_RENAMES should be 10**MAX_RENAMES_DIGITS */
324#define MAX_RENAMES_DIGITS 3
325#define MAX_RENAMES 1000
326
ac13ad10 327/**
b4235b31
MP
328 * Robust unlink: some OS'es (HPUX) refuse to unlink busy files, so
329 * rename to <path>/.rsyncNNN instead.
330 *
331 * Note that successive rsync runs will shuffle the filenames around a
332 * bit as long as the file is still busy; this is because this function
333 * does not know if the unlink call is due to a new file coming in, or
334 * --delete trying to remove old .rsyncNNN files, hence it renames it
335 * each time.
336 **/
63cf5ae7 337int robust_unlink(const char *fname)
c7c11a0d
DD
338{
339#ifndef ETXTBSY
340 return do_unlink(fname);
341#else
342 static int counter = 1;
343 int rc, pos, start;
344 char path[MAXPATHLEN];
345
346 rc = do_unlink(fname);
c284f34a 347 if (rc == 0 || errno != ETXTBSY)
c7c11a0d
DD
348 return rc;
349
c284f34a
WD
350 if ((pos = strlcpy(path, fname, MAXPATHLEN)) >= MAXPATHLEN)
351 pos = MAXPATHLEN - 1;
c7c11a0d 352
c284f34a
WD
353 while (pos > 0 && path[pos-1] != '/')
354 pos--;
5cb37436 355 pos += strlcpy(path+pos, ".rsync", MAXPATHLEN-pos);
c7c11a0d
DD
356
357 if (pos > (MAXPATHLEN-MAX_RENAMES_DIGITS-1)) {
358 errno = ETXTBSY;
359 return -1;
360 }
361
362 /* start where the last one left off to reduce chance of clashes */
363 start = counter;
364 do {
434c4098 365 snprintf(&path[pos], MAX_RENAMES_DIGITS+1, "%03d", counter);
c7c11a0d
DD
366 if (++counter >= MAX_RENAMES)
367 counter = 1;
c284f34a 368 } while ((rc = access(path, 0)) == 0 && counter != start);
c7c11a0d 369
4791825d 370 if (verbose > 0) {
c7c11a0d 371 rprintf(FINFO,"renaming %s to %s because of text busy\n",
785abd48 372 fname, path);
4791825d 373 }
c7c11a0d
DD
374
375 /* maybe we should return rename()'s exit status? Nah. */
376 if (do_rename(fname, path) != 0) {
377 errno = ETXTBSY;
378 return -1;
379 }
380 return 0;
381#endif
382}
383
630f548f 384/* Returns 0 on successful rename, 1 if we successfully copied the file
3ed8eafc
WD
385 * across filesystems, -2 if copy_file() failed, and -1 on other errors.
386 * If partialptr is not NULL and we need to do a copy, copy the file into
387 * the active partial-dir instead of over the destination file. */
4a19c3b2 388int robust_rename(const char *from, const char *to, const char *partialptr,
3ed8eafc 389 int mode)
c7c11a0d 390{
62c9e6b3
WD
391 int tries = 4;
392
393 while (tries--) {
394 if (do_rename(from, to) == 0)
395 return 0;
396
397 switch (errno) {
398#ifdef ETXTBSY
399 case ETXTBSY:
400 if (robust_unlink(to) != 0)
401 return -1;
402 break;
c7c11a0d 403#endif
62c9e6b3 404 case EXDEV:
3ed8eafc
WD
405 if (partialptr) {
406 if (!handle_partial_dir(partialptr,PDIR_CREATE))
407 return -1;
408 to = partialptr;
409 }
f5b532b1 410 if (copy_file(from, to, mode, 0) != 0)
62c9e6b3
WD
411 return -2;
412 do_unlink(from);
630f548f 413 return 1;
62c9e6b3
WD
414 default:
415 return -1;
416 }
417 }
418 return -1;
feaa89c4 419}
3ba62a83 420
3ba62a83
AT
421static pid_t all_pids[10];
422static int num_pids;
423
4cf64834 424/** Fork and record the pid of the child. **/
3ba62a83
AT
425pid_t do_fork(void)
426{
427 pid_t newpid = fork();
5cb37436 428
4cf64834 429 if (newpid != 0 && newpid != -1) {
3ba62a83
AT
430 all_pids[num_pids++] = newpid;
431 }
432 return newpid;
433}
434
4cf64834
MP
435/**
436 * Kill all children.
437 *
438 * @todo It would be kind of nice to make sure that they are actually
439 * all our children before we kill them, because their pids may have
440 * been recycled by some other process. Perhaps when we wait for a
441 * child, we should remove it from this array. Alternatively we could
442 * perhaps use process groups, but I think that would not work on
443 * ancient Unix versions that don't support them.
444 **/
3ba62a83
AT
445void kill_all(int sig)
446{
447 int i;
4cf64834
MP
448
449 for (i = 0; i < num_pids; i++) {
450 /* Let's just be a little careful where we
451 * point that gun, hey? See kill(2) for the
452 * magic caused by negative values. */
453 pid_t p = all_pids[i];
454
455 if (p == getpid())
456 continue;
457 if (p <= 0)
458 continue;
459
460 kill(p, sig);
3ba62a83
AT
461 }
462}
9486289c 463
ac13ad10 464/** Turn a user name into a uid */
4a19c3b2 465int name_to_uid(const char *name, uid_t *uid)
8ef4ffd6
AT
466{
467 struct passwd *pass;
b5bd5542
WD
468 if (!name || !*name)
469 return 0;
8ef4ffd6
AT
470 pass = getpwnam(name);
471 if (pass) {
472 *uid = pass->pw_uid;
473 return 1;
474 }
475 return 0;
476}
477
ac13ad10 478/** Turn a group name into a gid */
4a19c3b2 479int name_to_gid(const char *name, gid_t *gid)
8ef4ffd6
AT
480{
481 struct group *grp;
b5bd5542
WD
482 if (!name || !*name)
483 return 0;
8ef4ffd6
AT
484 grp = getgrnam(name);
485 if (grp) {
486 *gid = grp->gr_gid;
487 return 1;
488 }
489 return 0;
490}
491
ac13ad10 492/** Lock a byte range in a open file */
31593dd6 493int lock_range(int fd, int offset, int len)
0c515f17 494{
31593dd6 495 struct flock lock;
0c515f17 496
31593dd6
AT
497 lock.l_type = F_WRLCK;
498 lock.l_whence = SEEK_SET;
499 lock.l_start = offset;
500 lock.l_len = len;
501 lock.l_pid = 0;
5cb37436 502
31593dd6 503 return fcntl(fd,F_SETLK,&lock) == 0;
0c515f17 504}
874895d5 505
7842418b 506static int filter_server_path(char *arg)
4791825d
WD
507{
508 char *s;
4791825d 509
7842418b 510 if (server_filter_list.head) {
4791825d
WD
511 for (s = arg; (s = strchr(s, '/')) != NULL; ) {
512 *s = '\0';
7842418b 513 if (check_filter(&server_filter_list, arg, 1) < 0) {
4791825d
WD
514 /* We must leave arg truncated! */
515 return 1;
516 }
517 *s++ = '/';
518 }
519 }
520 return 0;
521}
874895d5 522
044dc293 523void glob_expand(char *s, char ***argv_ptr, int *argc_ptr, 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)))
044dc293 532 out_of_memory("glob_expand");
b7061c82
WD
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);
044dc293
WD
552 if (!s)
553 out_of_memory("glob_expand");
087bf010 554
5cb37436 555 memset(&globbuf, 0, sizeof globbuf);
7842418b 556 if (!filter_server_path(s))
4791825d 557 glob(s, 0, NULL, &globbuf);
b7061c82
WD
558 if (MAX((int)globbuf.gl_pathc, 1) > maxargs - argc) {
559 maxargs += globbuf.gl_pathc + MAX_ARGS;
560 if (!(argv = realloc_array(argv, char *, maxargs)))
044dc293 561 out_of_memory("glob_expand");
b7061c82
WD
562 *argv_ptr = argv;
563 *maxargs_ptr = maxargs;
564 }
b5bd5542
WD
565 if (globbuf.gl_pathc == 0)
566 argv[argc++] = s;
567 else {
c8933031 568 int i;
b5bd5542 569 free(s);
c8933031 570 for (i = 0; i < (int)globbuf.gl_pathc; i++) {
b5bd5542 571 if (!(argv[argc++] = strdup(globbuf.gl_pathv[i])))
044dc293 572 out_of_memory("glob_expand");
b5bd5542 573 }
874895d5
AT
574 }
575 globfree(&globbuf);
874895d5 576#endif
b5bd5542 577 *argc_ptr = argc;
874895d5 578}
5a96ee05 579
4791825d 580/* This routine is only used in daemon mode. */
044dc293 581void glob_expand_module(char *base1, char *arg, char ***argv_ptr, int *argc_ptr, int *maxargs_ptr)
087bf010 582{
044dc293 583 char *p, *s;
ba5e128d 584 char *base = base1;
4791825d 585 int base_len = strlen(base);
087bf010 586
044dc293 587 if (!arg || !*arg)
b5bd5542 588 return;
087bf010 589
044dc293
WD
590 if (strncmp(arg, base, base_len) == 0)
591 arg += base_len;
e42c9458 592
044dc293
WD
593 if (!(arg = strdup(arg)))
594 out_of_memory("glob_expand_module");
087bf010 595
b5bd5542 596 if (asprintf(&base," %s/", base1) <= 0)
044dc293 597 out_of_memory("glob_expand_module");
4791825d 598 base_len++;
ba5e128d 599
044dc293
WD
600 for (s = arg; *s; s = p + base_len) {
601 if ((p = strstr(s, base)) != NULL)
b5bd5542 602 *p = '\0'; /* split it at this point */
044dc293 603 glob_expand(s, argv_ptr, argc_ptr, maxargs_ptr);
b5bd5542
WD
604 if (!p)
605 break;
087bf010
AT
606 }
607
044dc293 608 free(arg);
ba5e128d 609 free(base);
087bf010 610}
5a96ee05 611
ac13ad10
MP
612/**
613 * Convert a string to lower case
614 **/
5a96ee05
AT
615void strlower(char *s)
616{
617 while (*s) {
2dc7b8bd
WD
618 if (isUpper(s))
619 *s = toLower(s);
5a96ee05
AT
620 s++;
621 }
622}
e42c9458 623
368ad70e
WD
624/* Join strings p1 & p2 into "dest" with a guaranteed '/' between them. (If
625 * p1 ends with a '/', no extra '/' is inserted.) Returns the length of both
a8f7e4b8
WD
626 * strings + 1 (if '/' was inserted), regardless of whether the null-terminated
627 * string fits into destsize. */
368ad70e
WD
628size_t pathjoin(char *dest, size_t destsize, const char *p1, const char *p2)
629{
630 size_t len = strlcpy(dest, p1, destsize);
631 if (len < destsize - 1) {
632 if (!len || dest[len-1] != '/')
633 dest[len++] = '/';
634 if (len < destsize - 1)
635 len += strlcpy(dest + len, p2, destsize - len);
636 else {
637 dest[len] = '\0';
638 len += strlen(p2);
639 }
640 }
641 else
642 len += strlen(p2) + 1; /* Assume we'd insert a '/'. */
643 return len;
644}
645
646/* Join any number of strings together, putting them in "dest". The return
a8f7e4b8
WD
647 * value is the length of all the strings, regardless of whether the null-
648 * terminated whole fits in destsize. Your list of string pointers must end
649 * with a NULL to indicate the end of the list. */
368ad70e
WD
650size_t stringjoin(char *dest, size_t destsize, ...)
651{
5cb37436 652 va_list ap;
368ad70e
WD
653 size_t len, ret = 0;
654 const char *src;
655
656 va_start(ap, destsize);
657 while (1) {
658 if (!(src = va_arg(ap, const char *)))
659 break;
660 len = strlen(src);
661 ret += len;
662 if (destsize > 1) {
663 if (len >= destsize)
664 len = destsize - 1;
665 memcpy(dest, src, len);
666 destsize -= len;
667 dest += len;
668 }
669 }
670 *dest = '\0';
671 va_end(ap);
672
673 return ret;
674}
675
1d6b8f9a
WD
676int count_dir_elements(const char *p)
677{
678 int cnt = 0, new_component = 1;
679 while (*p) {
680 if (*p++ == '/')
a8167c66 681 new_component = (*p != '.' || (p[1] != '/' && p[1] != '\0'));
1d6b8f9a
WD
682 else if (new_component) {
683 new_component = 0;
684 cnt++;
685 }
686 }
687 return cnt;
688}
689
b92693da
WD
690/* Turns multiple adjacent slashes into a single slash, gets rid of "./"
691 * elements (but not a trailing dot dir), removes a trailing slash, and
692 * optionally collapses ".." elements (except for those at the start of the
693 * string). If the resulting name would be empty, change it into a ".". */
694unsigned int clean_fname(char *name, BOOL collapse_dot_dot)
5243c216 695{
e012b94f 696 char *limit = name - 1, *t = name, *f = name;
ebdd24d6 697 int anchored;
5243c216 698
b5bd5542 699 if (!name)
3104620c 700 return 0;
5243c216 701
ebdd24d6
WD
702 if ((anchored = *f == '/') != 0)
703 *t++ = *f++;
704 while (*f) {
705 /* discard extra slashes */
706 if (*f == '/') {
707 f++;
708 continue;
5243c216 709 }
ebdd24d6
WD
710 if (*f == '.') {
711 /* discard "." dirs (but NOT a trailing '.'!) */
712 if (f[1] == '/') {
e012b94f 713 f += 2;
ebdd24d6
WD
714 continue;
715 }
716 /* collapse ".." dirs */
b92693da
WD
717 if (collapse_dot_dot
718 && f[1] == '.' && (f[2] == '/' || !f[2])) {
ebdd24d6
WD
719 char *s = t - 1;
720 if (s == name && anchored) {
721 f += 2;
722 continue;
723 }
724 while (s > limit && *--s != '/') {}
e012b94f 725 if (s != t - 1 && (s < name || *s == '/')) {
ebdd24d6
WD
726 t = s + 1;
727 f += 2;
728 continue;
729 }
f55c2dfc 730 limit = t + 2;
5243c216
AT
731 }
732 }
ebdd24d6 733 while (*f && (*t++ = *f++) != '/') {}
5243c216 734 }
ebdd24d6
WD
735
736 if (t > name+anchored && t[-1] == '/')
737 t--;
738 if (t == name)
739 *t++ = '.';
740 *t = '\0';
3104620c
WD
741
742 return t - name;
5243c216
AT
743}
744
84a63795
WD
745/* Make path appear as if a chroot had occurred. This handles a leading
746 * "/" (either removing it or expanding it) and any leading or embedded
747 * ".." components that attempt to escape past the module's top dir.
b4235b31 748 *
1d6b8f9a
WD
749 * If dest is NULL, a buffer is allocated to hold the result. It is legal
750 * to call with the dest and the path (p) pointing to the same buffer, but
751 * rootdir will be ignored to avoid expansion of the string.
b4235b31 752 *
1d6b8f9a 753 * The rootdir string contains a value to use in place of a leading slash.
7c73536c 754 * Specify NULL to get the default of "module_dir".
ac13ad10 755 *
a8167c66
WD
756 * The depth var is a count of how many '..'s to allow at the start of the
757 * path. If symlink is set, combine its value with the "p" value to get
758 * the target path, and **return NULL if any '..'s try to escape**.
ac13ad10 759 *
b92693da 760 * We also clean the path in a manner similar to clean_fname() but with a
0f78b815 761 * few differences:
b92693da
WD
762 *
763 * Turns multiple adjacent slashes into a single slash, gets rid of "." dir
764 * elements (INCLUDING a trailing dot dir), PRESERVES a trailing slash, and
765 * ALWAYS collapses ".." elements (except for those at the start of the
766 * string up to "depth" deep). If the resulting name would be empty,
767 * change it into a ".". */
a8167c66
WD
768char *sanitize_path(char *dest, const char *p, const char *rootdir, int depth,
769 const char *symlink)
1b8e662a 770{
a8167c66 771 char *start, *sanp, *save_dest = dest;
7e43da81 772 int rlen = 0, leave_one_dotdir = relative_paths;
84a63795 773
a8167c66
WD
774 if (symlink && *symlink == '/') {
775 p = symlink;
776 symlink = "";
777 }
778
84a63795
WD
779 if (dest != p) {
780 int plen = strlen(p);
1d6b8f9a
WD
781 if (*p == '/') {
782 if (!rootdir)
7c73536c 783 rootdir = module_dir;
1d6b8f9a
WD
784 rlen = strlen(rootdir);
785 depth = 0;
84a63795
WD
786 p++;
787 }
788 if (dest) {
789 if (rlen + plen + 1 >= MAXPATHLEN)
790 return NULL;
791 } else if (!(dest = new_array(char, rlen + plen + 1)))
792 out_of_memory("sanitize_path");
793 if (rlen) {
1d6b8f9a 794 memcpy(dest, rootdir, rlen);
84a63795
WD
795 if (rlen > 1)
796 dest[rlen++] = '/';
797 }
798 }
cb13abfe 799
84a63795 800 start = sanp = dest + rlen;
a8167c66
WD
801 while (1) {
802 if (*p == '\0') {
803 if (!symlink || !*symlink)
804 break;
805 while (sanp != start && sanp[-1] != '/') {
806 /* strip last element */
807 sanp--;
808 }
809 /* Append a relative symlink */
810 p = symlink;
811 symlink = "";
812 }
2d41264e
WD
813 /* discard leading or extra slashes */
814 if (*p == '/') {
815 p++;
816 continue;
817 }
b5f9e67d 818 /* this loop iterates once per filename component in p.
44e2e578 819 * both p (and sanp if the original had a slash) should
b5f9e67d
DD
820 * always be left pointing after a slash
821 */
c284f34a 822 if (*p == '.' && (p[1] == '/' || p[1] == '\0')) {
35812ea1 823 if (leave_one_dotdir && p[1])
7e43da81
WD
824 leave_one_dotdir = 0;
825 else {
826 /* skip "." component */
827 p++;
828 continue;
829 }
cb13abfe 830 }
c284f34a 831 if (*p == '.' && p[1] == '.' && (p[2] == '/' || p[2] == '\0')) {
cb13abfe 832 /* ".." component followed by slash or end */
8e5f029e 833 if (depth <= 0 || sanp != start) {
a8167c66
WD
834 if (symlink && sanp == start) {
835 if (!save_dest)
836 free(dest);
837 return NULL;
838 }
8e5f029e
WD
839 p += 2;
840 if (sanp != start) {
841 /* back up sanp one level */
842 --sanp; /* now pointing at slash */
843 while (sanp > start && sanp[-1] != '/') {
844 /* skip back up to slash */
845 sanp--;
846 }
b5f9e67d 847 }
8e5f029e 848 continue;
1b8e662a 849 }
8e5f029e
WD
850 /* allow depth levels of .. at the beginning */
851 depth--;
852 /* move the virtual beginning to leave the .. alone */
853 start = sanp + 3;
1b8e662a 854 }
2d41264e
WD
855 /* copy one component through next slash */
856 while (*p && (*sanp++ = *p++) != '/') {}
1b8e662a 857 }
84a63795 858 if (sanp == dest) {
b5f9e67d 859 /* ended up with nothing, so put in "." component */
44e2e578 860 *sanp++ = '.';
b5f9e67d 861 }
44e2e578 862 *sanp = '\0';
1b8e662a 863
84a63795 864 return dest;
14b61c63 865}
5243c216 866
a8167c66 867/* Like chdir(), but it keeps track of the current directory (in the
a16d8f2b 868 * global "curr_dir"), and ensures that the path size doesn't overflow.
a8167c66 869 * Also cleans the path using the clean_fname() function. */
2c708476 870int push_dir(const char *dir, int set_path_only)
5243c216 871{
5243c216 872 static int initialised;
4af8fe4e 873 unsigned int len;
5243c216
AT
874
875 if (!initialised) {
876 initialised = 1;
5cb37436 877 getcwd(curr_dir, sizeof curr_dir - 1);
4af8fe4e 878 curr_dir_len = strlen(curr_dir);
5243c216
AT
879 }
880
4af8fe4e
WD
881 if (!dir) /* this call was probably just to initialize */
882 return 0;
c226b7c2 883
4af8fe4e
WD
884 len = strlen(dir);
885 if (len == 1 && *dir == '.')
886 return 1;
5243c216 887
7c73536c
WD
888 if ((*dir == '/' ? len : curr_dir_len + 1 + len) >= sizeof curr_dir) {
889 errno = ENAMETOOLONG;
4af8fe4e 890 return 0;
7c73536c 891 }
4af8fe4e 892
ba081be3 893 if (!set_path_only && chdir(dir))
4af8fe4e 894 return 0;
5243c216
AT
895
896 if (*dir == '/') {
4af8fe4e
WD
897 memcpy(curr_dir, dir, len + 1);
898 curr_dir_len = len;
899 } else {
900 curr_dir[curr_dir_len++] = '/';
901 memcpy(curr_dir + curr_dir_len, dir, len + 1);
902 curr_dir_len += len;
5243c216
AT
903 }
904
b92693da 905 curr_dir_len = clean_fname(curr_dir, 1);
a8167c66
WD
906 if (sanitize_paths) {
907 if (module_dirlen > curr_dir_len)
908 module_dirlen = curr_dir_len;
909 curr_dir_depth = count_dir_elements(curr_dir + module_dirlen);
910 }
5243c216 911
4af8fe4e 912 return 1;
5243c216
AT
913}
914
a16d8f2b
WD
915/**
916 * Reverse a push_dir() call. You must pass in an absolute path
917 * that was copied from a prior value of "curr_dir".
918 **/
2c708476 919int pop_dir(const char *dir)
5243c216 920{
4af8fe4e
WD
921 if (chdir(dir))
922 return 0;
5243c216 923
4af8fe4e
WD
924 curr_dir_len = strlcpy(curr_dir, dir, sizeof curr_dir);
925 if (curr_dir_len >= sizeof curr_dir)
926 curr_dir_len = sizeof curr_dir - 1;
a8167c66
WD
927 if (sanitize_paths)
928 curr_dir_depth = count_dir_elements(curr_dir + module_dirlen);
5243c216 929
4af8fe4e 930 return 1;
5243c216 931}
aa9b77a5 932
eb61be19
WD
933/**
934 * Return a quoted string with the full pathname of the indicated filename.
935 * The string " (in MODNAME)" may also be appended. The returned pointer
936 * remains valid until the next time full_fname() is called.
937 **/
9a5ade18 938char *full_fname(const char *fn)
eb61be19 939{
eb61be19
WD
940 static char *result = NULL;
941 char *m1, *m2, *m3;
942 char *p1, *p2;
943
944 if (result)
945 free(result);
946
947 if (*fn == '/')
948 p1 = p2 = "";
949 else {
4cff5fa4 950 p1 = curr_dir + module_dirlen;
bc83274a
WD
951 for (p2 = p1; *p2 == '/'; p2++) {}
952 if (*p2)
953 p2 = "/";
eb61be19
WD
954 }
955 if (module_id >= 0) {
956 m1 = " (in ";
957 m2 = lp_name(module_id);
958 m3 = ")";
eb61be19
WD
959 } else
960 m1 = m2 = m3 = "";
961
c98ad3df
WD
962 if (asprintf(&result, "\"%s%s%s\"%s%s%s", p1, p2, fn, m1, m2, m3) <= 0)
963 out_of_memory("full_fname");
eb61be19
WD
964
965 return result;
966}
967
a7260c40
WD
968static char partial_fname[MAXPATHLEN];
969
970char *partial_dir_fname(const char *fname)
971{
972 char *t = partial_fname;
973 int sz = sizeof partial_fname;
974 const char *fn;
975
976 if ((fn = strrchr(fname, '/')) != NULL) {
977 fn++;
978 if (*partial_dir != '/') {
979 int len = fn - fname;
980 strncpy(t, fname, len); /* safe */
981 t += len;
982 sz -= len;
983 }
984 } else
985 fn = fname;
986 if ((int)pathjoin(t, sz, partial_dir, fn) >= sz)
987 return NULL;
5aa7b20a 988 if (server_filter_list.head) {
89363676
WD
989 t = strrchr(partial_fname, '/');
990 *t = '\0';
5aa7b20a
WD
991 if (check_filter(&server_filter_list, partial_fname, 1) < 0)
992 return NULL;
89363676 993 *t = '/';
5aa7b20a
WD
994 if (check_filter(&server_filter_list, partial_fname, 0) < 0)
995 return NULL;
996 }
a7260c40
WD
997
998 return partial_fname;
999}
1000
1001/* If no --partial-dir option was specified, we don't need to do anything
1002 * (the partial-dir is essentially '.'), so just return success. */
1003int handle_partial_dir(const char *fname, int create)
1004{
1005 char *fn, *dir;
1006
1007 if (fname != partial_fname)
1008 return 1;
1009 if (!create && *partial_dir == '/')
1010 return 1;
1011 if (!(fn = strrchr(partial_fname, '/')))
1012 return 1;
1013
1014 *fn = '\0';
1015 dir = partial_fname;
1016 if (create) {
1017 STRUCT_STAT st;
a7260c40 1018 int statret = do_lstat(dir, &st);
a7260c40
WD
1019 if (statret == 0 && !S_ISDIR(st.st_mode)) {
1020 if (do_unlink(dir) < 0)
1021 return 0;
1022 statret = -1;
1023 }
1024 if (statret < 0 && do_mkdir(dir, 0700) < 0)
1025 return 0;
1026 } else
1027 do_rmdir(dir);
1028 *fn = '/';
1029
1030 return 1;
1031}
1032
ac13ad10
MP
1033/**
1034 * Determine if a symlink points outside the current directory tree.
036e70b0
MP
1035 * This is considered "unsafe" because e.g. when mirroring somebody
1036 * else's machine it might allow them to establish a symlink to
1037 * /etc/passwd, and then read it through a web server.
1038 *
4e5db0ad
MP
1039 * Null symlinks and absolute symlinks are always unsafe.
1040 *
1041 * Basically here we are concerned with symlinks whose target contains
1042 * "..", because this might cause us to walk back up out of the
1043 * transferred directory. We are not allowed to go back up and
1044 * reenter.
1045 *
036e70b0
MP
1046 * @param dest Target of the symlink in question.
1047 *
25d34a5c 1048 * @param src Top source directory currently applicable. Basically this
036e70b0 1049 * is the first parameter to rsync in a simple invocation, but it's
25d34a5c 1050 * modified by flist.c in slightly complex ways.
036e70b0
MP
1051 *
1052 * @retval True if unsafe
1053 * @retval False is unsafe
4e5db0ad
MP
1054 *
1055 * @sa t_unsafe.c
ac13ad10 1056 **/
7afa3a4a 1057int unsafe_symlink(const char *dest, const char *src)
4b957c22 1058{
7afa3a4a 1059 const char *name, *slash;
4b957c22
AT
1060 int depth = 0;
1061
1062 /* all absolute and null symlinks are unsafe */
b5bd5542
WD
1063 if (!dest || !*dest || *dest == '/')
1064 return 1;
4b957c22
AT
1065
1066 /* find out what our safety margin is */
7afa3a4a
WD
1067 for (name = src; (slash = strchr(name, '/')) != 0; name = slash+1) {
1068 if (strncmp(name, "../", 3) == 0) {
c284f34a 1069 depth = 0;
7afa3a4a 1070 } else if (strncmp(name, "./", 2) == 0) {
4b957c22
AT
1071 /* nothing */
1072 } else {
1073 depth++;
1074 }
1075 }
7afa3a4a
WD
1076 if (strcmp(name, "..") == 0)
1077 depth = 0;
4b957c22 1078
7afa3a4a
WD
1079 for (name = dest; (slash = strchr(name, '/')) != 0; name = slash+1) {
1080 if (strncmp(name, "../", 3) == 0) {
1081 /* if at any point we go outside the current directory
1082 then stop - it is unsafe */
1083 if (--depth < 0)
1084 return 1;
1085 } else if (strncmp(name, "./", 2) == 0) {
4b957c22
AT
1086 /* nothing */
1087 } else {
1088 depth++;
1089 }
4b957c22 1090 }
7afa3a4a
WD
1091 if (strcmp(name, "..") == 0)
1092 depth--;
4b957c22 1093
4b957c22
AT
1094 return (depth < 0);
1095}
375a4556 1096
e175fb07
WD
1097/* Return the int64 number as a string. If the --human-readable option was
1098 * specified, we may output the number in K, M, or G units. We can return
1099 * up to 4 buffers at a time. */
1100char *human_num(int64 num)
1101{
1102 static char bufs[4][128]; /* more than enough room */
1103 static unsigned int n;
1104 char *s;
1105
1106 n = (n + 1) % (sizeof bufs / sizeof bufs[0]);
1107
1108 if (human_readable) {
1109 char units = '\0';
7794db7c 1110 int mult = human_readable == 1 ? 1000 : 1024;
e175fb07
WD
1111 double dnum = 0;
1112 if (num > mult*mult*mult) {
1113 dnum = (double)num / (mult*mult*mult);
1114 units = 'G';
1115 } else if (num > mult*mult) {
1116 dnum = (double)num / (mult*mult);
1117 units = 'M';
1118 } else if (num > mult) {
1119 dnum = (double)num / mult;
1120 units = 'K';
1121 }
1122 if (units) {
83078af5 1123 snprintf(bufs[n], sizeof bufs[0], "%.2f%c", dnum, units);
e175fb07
WD
1124 return bufs[n];
1125 }
1126 }
1127
1128 s = bufs[n] + sizeof bufs[0] - 1;
1129 *s = '\0';
1130
1131 if (!num)
1132 *--s = '0';
1133 while (num) {
26404276 1134 *--s = (char)(num % 10) + '0';
e175fb07
WD
1135 num /= 10;
1136 }
1137 return s;
1138}
1139
1140/* Return the double number as a string. If the --human-readable option was
1141 * specified, we may output the number in K, M, or G units. We use a buffer
1142 * from human_num() to return our result. */
1143char *human_dnum(double dnum, int decimal_digits)
1144{
1145 char *buf = human_num(dnum);
1146 int len = strlen(buf);
2dc7b8bd 1147 if (isDigit(buf + len - 1)) {
e175fb07
WD
1148 /* There's extra room in buf prior to the start of the num. */
1149 buf -= decimal_digits + 1;
1150 snprintf(buf, len + decimal_digits + 2, "%.*f", decimal_digits, dnum);
1151 }
1152 return buf;
1153}
f7632fc6 1154
4a19c3b2 1155/* Return the date and time as a string. Some callers tweak returned buf. */
f7632fc6
AT
1156char *timestring(time_t t)
1157{
1158 static char TimeBuf[200];
1159 struct tm *tm = localtime(&t);
afa73c75 1160 char *p;
f7632fc6 1161
4f5b0756 1162#ifdef HAVE_STRFTIME
5cb37436 1163 strftime(TimeBuf, sizeof TimeBuf - 1, "%Y/%m/%d %H:%M:%S", tm);
f7632fc6 1164#else
5cb37436 1165 strlcpy(TimeBuf, asctime(tm), sizeof TimeBuf);
f7632fc6
AT
1166#endif
1167
afa73c75
WD
1168 if ((p = strchr(TimeBuf, '\n')) != NULL)
1169 *p = '\0';
f7632fc6 1170
afa73c75 1171 return TimeBuf;
f7632fc6
AT
1172}
1173
e1bd49d6
MP
1174/**
1175 * Sleep for a specified number of milliseconds.
1176 *
1177 * Always returns TRUE. (In the future it might return FALSE if
1178 * interrupted.)
1179 **/
1180int msleep(int t)
9ec16c83 1181{
c284f34a
WD
1182 int tdiff = 0;
1183 struct timeval tval, t1, t2;
9ec16c83
AT
1184
1185 gettimeofday(&t1, NULL);
5cb37436 1186
9ec16c83
AT
1187 while (tdiff < t) {
1188 tval.tv_sec = (t-tdiff)/1000;
1189 tval.tv_usec = 1000*((t-tdiff)%1000);
5cb37436 1190
9ec16c83
AT
1191 errno = 0;
1192 select(0,NULL,NULL, NULL, &tval);
1193
1194 gettimeofday(&t2, NULL);
5cb37436 1195 tdiff = (t2.tv_sec - t1.tv_sec)*1000 +
9ec16c83
AT
1196 (t2.tv_usec - t1.tv_usec)/1000;
1197 }
e1bd49d6
MP
1198
1199 return True;
9ec16c83
AT
1200}
1201
c2b54076
WD
1202/* Determine if two time_t values are equivalent (either exact, or in
1203 * the modification timestamp window established by --modify-window).
ac13ad10
MP
1204 *
1205 * @retval 0 if the times should be treated as the same
1206 *
1207 * @retval +1 if the first is later
1208 *
1209 * @retval -1 if the 2nd is later
1210 **/
c2b54076 1211int cmp_time(time_t file1, time_t file2)
5b56cc19 1212{
5b56cc19 1213 if (file2 > file1) {
bc6ebcd2
WD
1214 if (file2 - file1 <= modify_window)
1215 return 0;
5b56cc19
AT
1216 return -1;
1217 }
bc6ebcd2
WD
1218 if (file1 - file2 <= modify_window)
1219 return 0;
5b56cc19
AT
1220 return 1;
1221}
1222
1223
1224#ifdef __INSURE__XX
0f8f98c8
AT
1225#include <dlfcn.h>
1226
ac13ad10
MP
1227/**
1228 This routine is a trick to immediately catch errors when debugging
1229 with insure. A xterm with a gdb is popped up when insure catches
1230 a error. It is Linux specific.
1231**/
0f8f98c8
AT
1232int _Insure_trap_error(int a1, int a2, int a3, int a4, int a5, int a6)
1233{
1234 static int (*fn)();
1235 int ret;
8950ac03 1236 char *cmd;
0f8f98c8 1237
5cb37436 1238 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
1239 getpid(), getpid(), getpid());
1240
1241 if (!fn) {
1242 static void *h;
1243 h = dlopen("/usr/local/parasoft/insure++lite/lib.linux2/libinsure.so", RTLD_LAZY);
1244 fn = dlsym(h, "_Insure_trap_error");
1245 }
1246
1247 ret = fn(a1, a2, a3, a4, a5, a6);
1248
1249 system(cmd);
1250
8950ac03
AT
1251 free(cmd);
1252
0f8f98c8
AT
1253 return ret;
1254}
1255#endif
58cadc86 1256
58cadc86
WD
1257#define MALLOC_MAX 0x40000000
1258
a2dc4d68 1259void *_new_array(unsigned long num, unsigned int size, int use_calloc)
58cadc86
WD
1260{
1261 if (num >= MALLOC_MAX/size)
1262 return NULL;
a2dc4d68 1263 return use_calloc ? calloc(num, size) : malloc(num * size);
58cadc86
WD
1264}
1265
1266void *_realloc_array(void *ptr, unsigned int size, unsigned long num)
1267{
1268 if (num >= MALLOC_MAX/size)
1269 return NULL;
58cadc86
WD
1270 if (!ptr)
1271 return malloc(size * num);
1272 return realloc(ptr, size * num);
1273}
e64ae6d7
WD
1274
1275/* Take a filename and filename length and return the most significant
1276 * filename suffix we can find. This ignores suffixes such as "~",
1277 * ".bak", ".orig", ".~1~", etc. */
1278const char *find_filename_suffix(const char *fn, int fn_len, int *len_ptr)
1279{
1280 const char *suf, *s;
1281 BOOL had_tilde;
1282 int s_len;
1283
1284 /* One or more dots at the start aren't a suffix. */
1285 while (fn_len && *fn == '.') fn++, fn_len--;
1286
1287 /* Ignore the ~ in a "foo~" filename. */
1288 if (fn_len > 1 && fn[fn_len-1] == '~')
1289 fn_len--, had_tilde = True;
1290 else
1291 had_tilde = False;
1292
1293 /* Assume we don't find an suffix. */
1294 suf = "";
1295 *len_ptr = 0;
1296
1297 /* Find the last significant suffix. */
1298 for (s = fn + fn_len; fn_len > 1; ) {
1299 while (*--s != '.' && s != fn) {}
1300 if (s == fn)
1301 break;
1302 s_len = fn_len - (s - fn);
1303 fn_len = s - fn;
6012eaa1 1304 if (s_len == 4) {
e64ae6d7
WD
1305 if (strcmp(s+1, "bak") == 0
1306 || strcmp(s+1, "old") == 0)
1307 continue;
6012eaa1 1308 } else if (s_len == 5) {
e64ae6d7
WD
1309 if (strcmp(s+1, "orig") == 0)
1310 continue;
1311 } else if (s_len > 2 && had_tilde
2dc7b8bd 1312 && s[1] == '~' && isDigit(s + 2))
e64ae6d7
WD
1313 continue;
1314 *len_ptr = s_len;
1315 suf = s;
1316 if (s_len == 1)
1317 break;
1318 /* Determine if the suffix is all digits. */
1319 for (s++, s_len--; s_len > 0; s++, s_len--) {
2dc7b8bd 1320 if (!isDigit(s))
e64ae6d7
WD
1321 return suf;
1322 }
1323 /* An all-digit suffix may not be that signficant. */
1324 s = suf;
1325 }
1326
1327 return suf;
1328}
1329
1330/* This is an implementation of the Levenshtein distance algorithm. It
1331 * was implemented to avoid needing a two-dimensional matrix (to save
1332 * memory). It was also tweaked to try to factor in the ASCII distance
1333 * between changed characters as a minor distance quantity. The normal
1334 * Levenshtein units of distance (each signifying a single change between
1335 * the two strings) are defined as a "UNIT". */
1336
1337#define UNIT (1 << 16)
1338
1339uint32 fuzzy_distance(const char *s1, int len1, const char *s2, int len2)
1340{
1341 uint32 a[MAXPATHLEN], diag, above, left, diag_inc, above_inc, left_inc;
1342 int32 cost;
1343 int i1, i2;
1344
1345 if (!len1 || !len2) {
1346 if (!len1) {
1347 s1 = s2;
1348 len1 = len2;
1349 }
1350 for (i1 = 0, cost = 0; i1 < len1; i1++)
1351 cost += s1[i1];
1352 return (int32)len1 * UNIT + cost;
1353 }
1354
1355 for (i2 = 0; i2 < len2; i2++)
1356 a[i2] = (i2+1) * UNIT;
1357
1358 for (i1 = 0; i1 < len1; i1++) {
1359 diag = i1 * UNIT;
1360 above = (i1+1) * UNIT;
1361 for (i2 = 0; i2 < len2; i2++) {
1362 left = a[i2];
1363 if ((cost = *((uchar*)s1+i1) - *((uchar*)s2+i2)) != 0) {
1364 if (cost < 0)
1365 cost = UNIT - cost;
1366 else
1367 cost = UNIT + cost;
1368 }
1369 diag_inc = diag + cost;
1370 left_inc = left + UNIT + *((uchar*)s1+i1);
1371 above_inc = above + UNIT + *((uchar*)s2+i2);
1372 a[i2] = above = left < above
1373 ? (left_inc < diag_inc ? left_inc : diag_inc)
1374 : (above_inc < diag_inc ? above_inc : diag_inc);
1375 diag = left;
1376 }
1377 }
1378
1379 return a[len2-1];
1380}
c2b54076
WD
1381
1382#define BB_SLOT_SIZE (16*1024) /* Desired size in bytes */
1383#define BB_PER_SLOT_BITS (BB_SLOT_SIZE * 8) /* Number of bits per slot */
1384#define BB_PER_SLOT_INTS (BB_SLOT_SIZE / 4) /* Number of int32s per slot */
1385
1386struct bitbag {
1387 uint32 **bits;
1388 int slot_cnt;
1389};
1390
1391struct bitbag *bitbag_create(int max_ndx)
1392{
1393 struct bitbag *bb = new(struct bitbag);
1394 bb->slot_cnt = (max_ndx + BB_PER_SLOT_BITS - 1) / BB_PER_SLOT_BITS;
1395
1396 if (!(bb->bits = (uint32**)calloc(bb->slot_cnt, sizeof (uint32*))))
1397 out_of_memory("bitbag_create");
1398
1399 return bb;
1400}
1401
1402void bitbag_set_bit(struct bitbag *bb, int ndx)
1403{
1404 int slot = ndx / BB_PER_SLOT_BITS;
1405 ndx %= BB_PER_SLOT_BITS;
1406
1407 if (!bb->bits[slot]) {
1408 if (!(bb->bits[slot] = (uint32*)calloc(BB_PER_SLOT_INTS, 4)))
1409 out_of_memory("bitbag_set_bit");
1410 }
1411
1412 bb->bits[slot][ndx/32] |= 1u << (ndx % 32);
1413}
1414
1415#if 0 /* not needed yet */
1416void bitbag_clear_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;
1423
1424 bb->bits[slot][ndx/32] &= ~(1u << (ndx % 32));
1425}
1426
1427int bitbag_check_bit(struct bitbag *bb, int ndx)
1428{
1429 int slot = ndx / BB_PER_SLOT_BITS;
1430 ndx %= BB_PER_SLOT_BITS;
1431
1432 if (!bb->bits[slot])
1433 return 0;
1434
1435 return bb->bits[slot][ndx/32] & (1u << (ndx % 32)) ? 1 : 0;
1436}
1437#endif
1438
1439/* Call this with -1 to start checking from 0. Returns -1 at the end. */
1440int bitbag_next_bit(struct bitbag *bb, int after)
1441{
1442 uint32 bits, mask;
1443 int i, ndx = after + 1;
1444 int slot = ndx / BB_PER_SLOT_BITS;
1445 ndx %= BB_PER_SLOT_BITS;
1446
1447 mask = (1u << (ndx % 32)) - 1;
1448 for (i = ndx / 32; slot < bb->slot_cnt; slot++, i = mask = 0) {
1449 if (!bb->bits[slot])
1450 continue;
1451 for ( ; i < BB_PER_SLOT_INTS; i++, mask = 0) {
1452 if (!(bits = bb->bits[slot][i] & ~mask))
1453 continue;
1454 /* The xor magic figures out the lowest enabled bit in
1455 * bits, and the switch quickly computes log2(bit). */
1456 switch (bits ^ (bits & (bits-1))) {
1457#define LOG2(n) case 1u << n: return slot*BB_PER_SLOT_BITS + i*32 + n
1458 LOG2(0); LOG2(1); LOG2(2); LOG2(3);
1459 LOG2(4); LOG2(5); LOG2(6); LOG2(7);
1460 LOG2(8); LOG2(9); LOG2(10); LOG2(11);
1461 LOG2(12); LOG2(13); LOG2(14); LOG2(15);
1462 LOG2(16); LOG2(17); LOG2(18); LOG2(19);
1463 LOG2(20); LOG2(21); LOG2(22); LOG2(23);
1464 LOG2(24); LOG2(25); LOG2(26); LOG2(27);
1465 LOG2(28); LOG2(29); LOG2(30); LOG2(31);
1466 }
1467 return -1; /* impossible... */
1468 }
1469 }
1470
1471 return -1;
1472}
1c3344a1
WD
1473
1474void *expand_item_list(item_list *lp, size_t item_size,
1475 const char *desc, int incr)
1476{
1477 /* First time through, 0 <= 0, so list is expanded. */
1478 if (lp->malloced <= lp->count) {
1479 void *new_ptr;
1480 size_t new_size = lp->malloced;
1481 if (incr < 0)
0b515981 1482 new_size += -incr; /* increase slowly */
1c3344a1
WD
1483 else if (new_size < (size_t)incr)
1484 new_size += incr;
1485 else
1486 new_size *= 2;
1487 new_ptr = realloc_array(lp->items, char, new_size * item_size);
1488 if (verbose >= 4) {
1489 rprintf(FINFO, "[%s] expand %s to %.0f bytes, did%s move\n",
1490 who_am_i(), desc, (double)new_size * item_size,
1491 new_ptr == lp->items ? " not" : "");
1492 }
1493 if (!new_ptr)
1494 out_of_memory("expand_item_list");
1495
1496 lp->items = new_ptr;
1497 lp->malloced = new_size;
1498 }
1499 return (char*)lp->items + (lp->count++ * item_size);
1500}