Moved the dir_count increment into an even better spot and make sure that
[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);
4f9b139a 155#elif defined HAVE_STRUCT_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
2cce7545
WD
264/* Copy a file. If ofd < 0, copy_file unlinks and opens the "dest" file.
265 * Otherwise, it just writes to and closes the provided file descriptor.
ac13ad10 266 *
3e13004b
WD
267 * This is used in conjunction with the --temp-dir, --backup, and
268 * --copy-dest options. */
2cce7545
WD
269int copy_file(const char *source, const char *dest, int ofd,
270 mode_t mode, int create_bak_dir)
950ab32d
AT
271{
272 int ifd;
950ab32d
AT
273 char buf[1024 * 8];
274 int len; /* Number of bytes read into `buf'. */
275
f5b532b1 276 if ((ifd = do_open(source, O_RDONLY, 0)) < 0) {
3f0211b6 277 rsyserr(FERROR_XFER, errno, "open %s", full_fname(source));
950ab32d
AT
278 return -1;
279 }
280
2cce7545
WD
281 if (ofd < 0) {
282 if (robust_unlink(dest) && errno != ENOENT) {
3f0211b6 283 rsyserr(FERROR_XFER, errno, "unlink %s", full_fname(dest));
2cce7545
WD
284 return -1;
285 }
950ab32d 286
2cce7545
WD
287 if ((ofd = do_open(dest, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, mode)) < 0
288 && (!create_bak_dir || errno != ENOENT || make_bak_dir(dest) < 0
289 || (ofd = do_open(dest, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, mode)) < 0)) {
3f0211b6 290 rsyserr(FERROR_XFER, errno, "open %s", full_fname(dest));
2cce7545
WD
291 close(ifd);
292 return -1;
293 }
950ab32d
AT
294 }
295
5cb37436 296 while ((len = safe_read(ifd, buf, sizeof buf)) > 0) {
950ab32d 297 if (full_write(ofd, buf, len) < 0) {
3f0211b6 298 rsyserr(FERROR_XFER, errno, "write %s", full_fname(dest));
950ab32d
AT
299 close(ifd);
300 close(ofd);
301 return -1;
302 }
303 }
304
8b602edd 305 if (len < 0) {
3f0211b6 306 rsyserr(FERROR_XFER, errno, "read %s", full_fname(source));
8b602edd
WD
307 close(ifd);
308 close(ofd);
309 return -1;
310 }
311
9f27cd8c 312 if (close(ifd) < 0) {
3f0211b6 313 rsyserr(FWARNING, errno, "close failed on %s",
d62bcc17 314 full_fname(source));
9f27cd8c
WD
315 }
316
317 if (close(ofd) < 0) {
3f0211b6 318 rsyserr(FERROR_XFER, errno, "close failed on %s",
d62bcc17 319 full_fname(dest));
9f27cd8c
WD
320 return -1;
321 }
950ab32d 322
950ab32d
AT
323 return 0;
324}
feaa89c4 325
c7c11a0d
DD
326/* MAX_RENAMES should be 10**MAX_RENAMES_DIGITS */
327#define MAX_RENAMES_DIGITS 3
328#define MAX_RENAMES 1000
329
ac13ad10 330/**
b4235b31
MP
331 * Robust unlink: some OS'es (HPUX) refuse to unlink busy files, so
332 * rename to <path>/.rsyncNNN instead.
333 *
334 * Note that successive rsync runs will shuffle the filenames around a
335 * bit as long as the file is still busy; this is because this function
336 * does not know if the unlink call is due to a new file coming in, or
337 * --delete trying to remove old .rsyncNNN files, hence it renames it
338 * each time.
339 **/
63cf5ae7 340int robust_unlink(const char *fname)
c7c11a0d
DD
341{
342#ifndef ETXTBSY
343 return do_unlink(fname);
344#else
345 static int counter = 1;
346 int rc, pos, start;
347 char path[MAXPATHLEN];
348
349 rc = do_unlink(fname);
c284f34a 350 if (rc == 0 || errno != ETXTBSY)
c7c11a0d
DD
351 return rc;
352
c284f34a
WD
353 if ((pos = strlcpy(path, fname, MAXPATHLEN)) >= MAXPATHLEN)
354 pos = MAXPATHLEN - 1;
c7c11a0d 355
c284f34a
WD
356 while (pos > 0 && path[pos-1] != '/')
357 pos--;
5cb37436 358 pos += strlcpy(path+pos, ".rsync", MAXPATHLEN-pos);
c7c11a0d
DD
359
360 if (pos > (MAXPATHLEN-MAX_RENAMES_DIGITS-1)) {
361 errno = ETXTBSY;
362 return -1;
363 }
364
365 /* start where the last one left off to reduce chance of clashes */
366 start = counter;
367 do {
434c4098 368 snprintf(&path[pos], MAX_RENAMES_DIGITS+1, "%03d", counter);
c7c11a0d
DD
369 if (++counter >= MAX_RENAMES)
370 counter = 1;
c284f34a 371 } while ((rc = access(path, 0)) == 0 && counter != start);
c7c11a0d 372
4791825d 373 if (verbose > 0) {
3f0211b6 374 rprintf(FWARNING, "renaming %s to %s because of text busy\n",
785abd48 375 fname, path);
4791825d 376 }
c7c11a0d
DD
377
378 /* maybe we should return rename()'s exit status? Nah. */
379 if (do_rename(fname, path) != 0) {
380 errno = ETXTBSY;
381 return -1;
382 }
383 return 0;
384#endif
385}
386
630f548f 387/* Returns 0 on successful rename, 1 if we successfully copied the file
3ed8eafc
WD
388 * across filesystems, -2 if copy_file() failed, and -1 on other errors.
389 * If partialptr is not NULL and we need to do a copy, copy the file into
390 * the active partial-dir instead of over the destination file. */
4a19c3b2 391int robust_rename(const char *from, const char *to, const char *partialptr,
3ed8eafc 392 int mode)
c7c11a0d 393{
62c9e6b3
WD
394 int tries = 4;
395
396 while (tries--) {
397 if (do_rename(from, to) == 0)
398 return 0;
399
400 switch (errno) {
401#ifdef ETXTBSY
402 case ETXTBSY:
403 if (robust_unlink(to) != 0)
404 return -1;
405 break;
c7c11a0d 406#endif
62c9e6b3 407 case EXDEV:
3ed8eafc
WD
408 if (partialptr) {
409 if (!handle_partial_dir(partialptr,PDIR_CREATE))
410 return -1;
411 to = partialptr;
412 }
2cce7545 413 if (copy_file(from, to, -1, mode, 0) != 0)
62c9e6b3
WD
414 return -2;
415 do_unlink(from);
630f548f 416 return 1;
62c9e6b3
WD
417 default:
418 return -1;
419 }
420 }
421 return -1;
feaa89c4 422}
3ba62a83 423
3ba62a83
AT
424static pid_t all_pids[10];
425static int num_pids;
426
4cf64834 427/** Fork and record the pid of the child. **/
3ba62a83
AT
428pid_t do_fork(void)
429{
430 pid_t newpid = fork();
5cb37436 431
4cf64834 432 if (newpid != 0 && newpid != -1) {
3ba62a83
AT
433 all_pids[num_pids++] = newpid;
434 }
435 return newpid;
436}
437
4cf64834
MP
438/**
439 * Kill all children.
440 *
441 * @todo It would be kind of nice to make sure that they are actually
442 * all our children before we kill them, because their pids may have
443 * been recycled by some other process. Perhaps when we wait for a
444 * child, we should remove it from this array. Alternatively we could
445 * perhaps use process groups, but I think that would not work on
446 * ancient Unix versions that don't support them.
447 **/
3ba62a83
AT
448void kill_all(int sig)
449{
450 int i;
4cf64834
MP
451
452 for (i = 0; i < num_pids; i++) {
453 /* Let's just be a little careful where we
454 * point that gun, hey? See kill(2) for the
455 * magic caused by negative values. */
456 pid_t p = all_pids[i];
457
458 if (p == getpid())
459 continue;
460 if (p <= 0)
461 continue;
462
463 kill(p, sig);
3ba62a83
AT
464 }
465}
9486289c 466
ac13ad10 467/** Turn a user name into a uid */
4a19c3b2 468int name_to_uid(const char *name, uid_t *uid)
8ef4ffd6
AT
469{
470 struct passwd *pass;
b5bd5542
WD
471 if (!name || !*name)
472 return 0;
8ef4ffd6
AT
473 pass = getpwnam(name);
474 if (pass) {
475 *uid = pass->pw_uid;
476 return 1;
477 }
478 return 0;
479}
480
ac13ad10 481/** Turn a group name into a gid */
4a19c3b2 482int name_to_gid(const char *name, gid_t *gid)
8ef4ffd6
AT
483{
484 struct group *grp;
b5bd5542
WD
485 if (!name || !*name)
486 return 0;
8ef4ffd6
AT
487 grp = getgrnam(name);
488 if (grp) {
489 *gid = grp->gr_gid;
490 return 1;
491 }
492 return 0;
493}
494
ac13ad10 495/** Lock a byte range in a open file */
31593dd6 496int lock_range(int fd, int offset, int len)
0c515f17 497{
31593dd6 498 struct flock lock;
0c515f17 499
31593dd6
AT
500 lock.l_type = F_WRLCK;
501 lock.l_whence = SEEK_SET;
502 lock.l_start = offset;
503 lock.l_len = len;
504 lock.l_pid = 0;
5cb37436 505
31593dd6 506 return fcntl(fd,F_SETLK,&lock) == 0;
0c515f17 507}
874895d5 508
7842418b 509static int filter_server_path(char *arg)
4791825d
WD
510{
511 char *s;
4791825d 512
7842418b 513 if (server_filter_list.head) {
4791825d
WD
514 for (s = arg; (s = strchr(s, '/')) != NULL; ) {
515 *s = '\0';
7842418b 516 if (check_filter(&server_filter_list, arg, 1) < 0) {
4791825d
WD
517 /* We must leave arg truncated! */
518 return 1;
519 }
520 *s++ = '/';
521 }
522 }
523 return 0;
524}
874895d5 525
044dc293 526void glob_expand(char *s, char ***argv_ptr, int *argc_ptr, int *maxargs_ptr)
874895d5 527{
b7061c82 528 char **argv = *argv_ptr;
b5bd5542 529 int argc = *argc_ptr;
b7061c82 530 int maxargs = *maxargs_ptr;
4f5b0756 531#if !defined HAVE_GLOB || !defined HAVE_GLOB_H
b7061c82
WD
532 if (argc == maxargs) {
533 maxargs += MAX_ARGS;
534 if (!(argv = realloc_array(argv, char *, maxargs)))
044dc293 535 out_of_memory("glob_expand");
b7061c82
WD
536 *argv_ptr = argv;
537 *maxargs_ptr = maxargs;
538 }
4135d091
WD
539 if (!*s)
540 s = ".";
b5bd5542 541 s = argv[argc++] = strdup(s);
7842418b 542 filter_server_path(s);
874895d5
AT
543#else
544 glob_t globbuf;
874895d5 545
b5bd5542
WD
546 if (maxargs <= argc)
547 return;
4135d091
WD
548 if (!*s)
549 s = ".";
e42c9458 550
4135d091 551 if (sanitize_paths)
29bca53f 552 s = sanitize_path(NULL, s, "", 0);
84a63795
WD
553 else
554 s = strdup(s);
044dc293
WD
555 if (!s)
556 out_of_memory("glob_expand");
087bf010 557
5cb37436 558 memset(&globbuf, 0, sizeof globbuf);
7842418b 559 if (!filter_server_path(s))
4791825d 560 glob(s, 0, NULL, &globbuf);
b7061c82
WD
561 if (MAX((int)globbuf.gl_pathc, 1) > maxargs - argc) {
562 maxargs += globbuf.gl_pathc + MAX_ARGS;
563 if (!(argv = realloc_array(argv, char *, maxargs)))
044dc293 564 out_of_memory("glob_expand");
b7061c82
WD
565 *argv_ptr = argv;
566 *maxargs_ptr = maxargs;
567 }
b5bd5542
WD
568 if (globbuf.gl_pathc == 0)
569 argv[argc++] = s;
570 else {
c8933031 571 int i;
b5bd5542 572 free(s);
c8933031 573 for (i = 0; i < (int)globbuf.gl_pathc; i++) {
b5bd5542 574 if (!(argv[argc++] = strdup(globbuf.gl_pathv[i])))
044dc293 575 out_of_memory("glob_expand");
b5bd5542 576 }
874895d5
AT
577 }
578 globfree(&globbuf);
874895d5 579#endif
b5bd5542 580 *argc_ptr = argc;
874895d5 581}
5a96ee05 582
4791825d 583/* This routine is only used in daemon mode. */
044dc293 584void glob_expand_module(char *base1, char *arg, char ***argv_ptr, int *argc_ptr, int *maxargs_ptr)
087bf010 585{
044dc293 586 char *p, *s;
ba5e128d 587 char *base = base1;
4791825d 588 int base_len = strlen(base);
087bf010 589
044dc293 590 if (!arg || !*arg)
b5bd5542 591 return;
087bf010 592
044dc293
WD
593 if (strncmp(arg, base, base_len) == 0)
594 arg += base_len;
e42c9458 595
044dc293
WD
596 if (!(arg = strdup(arg)))
597 out_of_memory("glob_expand_module");
087bf010 598
b5bd5542 599 if (asprintf(&base," %s/", base1) <= 0)
044dc293 600 out_of_memory("glob_expand_module");
4791825d 601 base_len++;
ba5e128d 602
044dc293
WD
603 for (s = arg; *s; s = p + base_len) {
604 if ((p = strstr(s, base)) != NULL)
b5bd5542 605 *p = '\0'; /* split it at this point */
044dc293 606 glob_expand(s, argv_ptr, argc_ptr, maxargs_ptr);
b5bd5542
WD
607 if (!p)
608 break;
087bf010
AT
609 }
610
044dc293 611 free(arg);
ba5e128d 612 free(base);
087bf010 613}
5a96ee05 614
ac13ad10
MP
615/**
616 * Convert a string to lower case
617 **/
5a96ee05
AT
618void strlower(char *s)
619{
620 while (*s) {
2dc7b8bd
WD
621 if (isUpper(s))
622 *s = toLower(s);
5a96ee05
AT
623 s++;
624 }
625}
e42c9458 626
368ad70e
WD
627/* Join strings p1 & p2 into "dest" with a guaranteed '/' between them. (If
628 * p1 ends with a '/', no extra '/' is inserted.) Returns the length of both
a8f7e4b8
WD
629 * strings + 1 (if '/' was inserted), regardless of whether the null-terminated
630 * string fits into destsize. */
368ad70e
WD
631size_t pathjoin(char *dest, size_t destsize, const char *p1, const char *p2)
632{
633 size_t len = strlcpy(dest, p1, destsize);
634 if (len < destsize - 1) {
635 if (!len || dest[len-1] != '/')
636 dest[len++] = '/';
637 if (len < destsize - 1)
638 len += strlcpy(dest + len, p2, destsize - len);
639 else {
640 dest[len] = '\0';
641 len += strlen(p2);
642 }
643 }
644 else
645 len += strlen(p2) + 1; /* Assume we'd insert a '/'. */
646 return len;
647}
648
649/* Join any number of strings together, putting them in "dest". The return
a8f7e4b8
WD
650 * value is the length of all the strings, regardless of whether the null-
651 * terminated whole fits in destsize. Your list of string pointers must end
652 * with a NULL to indicate the end of the list. */
368ad70e
WD
653size_t stringjoin(char *dest, size_t destsize, ...)
654{
5cb37436 655 va_list ap;
368ad70e
WD
656 size_t len, ret = 0;
657 const char *src;
658
659 va_start(ap, destsize);
660 while (1) {
661 if (!(src = va_arg(ap, const char *)))
662 break;
663 len = strlen(src);
664 ret += len;
665 if (destsize > 1) {
666 if (len >= destsize)
667 len = destsize - 1;
668 memcpy(dest, src, len);
669 destsize -= len;
670 dest += len;
671 }
672 }
673 *dest = '\0';
674 va_end(ap);
675
676 return ret;
677}
678
1d6b8f9a
WD
679int count_dir_elements(const char *p)
680{
681 int cnt = 0, new_component = 1;
682 while (*p) {
683 if (*p++ == '/')
a8167c66 684 new_component = (*p != '.' || (p[1] != '/' && p[1] != '\0'));
1d6b8f9a
WD
685 else if (new_component) {
686 new_component = 0;
687 cnt++;
688 }
689 }
690 return cnt;
691}
692
6bb82fe0
WD
693/* Turns multiple adjacent slashes into a single slash, drops interior "."
694 * elements, drops an intial "./" unless CFN_KEEP_LEADING_DOT_DIR is flagged,
695 * will even drop a trailing '.' after a '/' if CFN_DROP_TRAILING_DOT_DIR is
696 * flagged, removes a trailing slash (perhaps after removing the aforementioned
697 * dot) unless CFN_KEEP_TRAILING_SLASH is flagged, will even collapse ".."
698 * elements (except at the start of the string) if CFN_COLLAPSE_DOT_DOT_DIRS
699 * is flagged. If the resulting name would be empty, we return ".". */
700unsigned int clean_fname(char *name, int flags)
5243c216 701{
e012b94f 702 char *limit = name - 1, *t = name, *f = name;
ebdd24d6 703 int anchored;
5243c216 704
b5bd5542 705 if (!name)
3104620c 706 return 0;
5243c216 707
ebdd24d6
WD
708 if ((anchored = *f == '/') != 0)
709 *t++ = *f++;
6bb82fe0
WD
710 else if (flags & CFN_KEEP_LEADING_DOT_DIR && *f == '.' && f[1] == '/') {
711 *t++ = *f++;
712 *t++ = *f++;
713 }
ebdd24d6
WD
714 while (*f) {
715 /* discard extra slashes */
716 if (*f == '/') {
717 f++;
718 continue;
5243c216 719 }
ebdd24d6 720 if (*f == '.') {
6bb82fe0 721 /* discard interior "." dirs */
ebdd24d6 722 if (f[1] == '/') {
e012b94f 723 f += 2;
ebdd24d6
WD
724 continue;
725 }
6bb82fe0
WD
726 if (f[1] == '\0' && flags & CFN_DROP_TRAILING_DOT_DIR)
727 break;
ebdd24d6 728 /* collapse ".." dirs */
6bb82fe0
WD
729 if (flags & CFN_COLLAPSE_DOT_DOT_DIRS
730 && f[1] == '.' && (f[2] == '/' || !f[2])) {
ebdd24d6
WD
731 char *s = t - 1;
732 if (s == name && anchored) {
733 f += 2;
734 continue;
735 }
736 while (s > limit && *--s != '/') {}
e012b94f 737 if (s != t - 1 && (s < name || *s == '/')) {
ebdd24d6
WD
738 t = s + 1;
739 f += 2;
740 continue;
741 }
f55c2dfc 742 limit = t + 2;
5243c216
AT
743 }
744 }
ebdd24d6 745 while (*f && (*t++ = *f++) != '/') {}
5243c216 746 }
ebdd24d6 747
6bb82fe0 748 if (t > name+anchored && t[-1] == '/' && !(flags & CFN_KEEP_TRAILING_SLASH))
ebdd24d6
WD
749 t--;
750 if (t == name)
751 *t++ = '.';
752 *t = '\0';
3104620c
WD
753
754 return t - name;
5243c216
AT
755}
756
84a63795
WD
757/* Make path appear as if a chroot had occurred. This handles a leading
758 * "/" (either removing it or expanding it) and any leading or embedded
759 * ".." components that attempt to escape past the module's top dir.
b4235b31 760 *
1d6b8f9a
WD
761 * If dest is NULL, a buffer is allocated to hold the result. It is legal
762 * to call with the dest and the path (p) pointing to the same buffer, but
763 * rootdir will be ignored to avoid expansion of the string.
b4235b31 764 *
1d6b8f9a 765 * The rootdir string contains a value to use in place of a leading slash.
7c73536c 766 * Specify NULL to get the default of "module_dir".
ac13ad10 767 *
a8167c66 768 * The depth var is a count of how many '..'s to allow at the start of the
29bca53f 769 * path.
ac13ad10 770 *
b92693da 771 * We also clean the path in a manner similar to clean_fname() but with a
0f78b815 772 * few differences:
b92693da
WD
773 *
774 * Turns multiple adjacent slashes into a single slash, gets rid of "." dir
775 * elements (INCLUDING a trailing dot dir), PRESERVES a trailing slash, and
776 * ALWAYS collapses ".." elements (except for those at the start of the
777 * string up to "depth" deep). If the resulting name would be empty,
778 * change it into a ".". */
29bca53f 779char *sanitize_path(char *dest, const char *p, const char *rootdir, int depth)
1b8e662a 780{
29bca53f 781 char *start, *sanp;
7e43da81 782 int rlen = 0, leave_one_dotdir = relative_paths;
84a63795
WD
783
784 if (dest != p) {
785 int plen = strlen(p);
1d6b8f9a
WD
786 if (*p == '/') {
787 if (!rootdir)
7c73536c 788 rootdir = module_dir;
1d6b8f9a
WD
789 rlen = strlen(rootdir);
790 depth = 0;
84a63795
WD
791 p++;
792 }
793 if (dest) {
794 if (rlen + plen + 1 >= MAXPATHLEN)
795 return NULL;
796 } else if (!(dest = new_array(char, rlen + plen + 1)))
797 out_of_memory("sanitize_path");
798 if (rlen) {
1d6b8f9a 799 memcpy(dest, rootdir, rlen);
84a63795
WD
800 if (rlen > 1)
801 dest[rlen++] = '/';
802 }
803 }
cb13abfe 804
84a63795 805 start = sanp = dest + rlen;
29bca53f 806 while (*p) {
2d41264e
WD
807 /* discard leading or extra slashes */
808 if (*p == '/') {
809 p++;
810 continue;
811 }
b5f9e67d 812 /* this loop iterates once per filename component in p.
44e2e578 813 * both p (and sanp if the original had a slash) should
b5f9e67d
DD
814 * always be left pointing after a slash
815 */
c284f34a 816 if (*p == '.' && (p[1] == '/' || p[1] == '\0')) {
35812ea1 817 if (leave_one_dotdir && p[1])
7e43da81
WD
818 leave_one_dotdir = 0;
819 else {
820 /* skip "." component */
821 p++;
822 continue;
823 }
cb13abfe 824 }
c284f34a 825 if (*p == '.' && p[1] == '.' && (p[2] == '/' || p[2] == '\0')) {
cb13abfe 826 /* ".." component followed by slash or end */
8e5f029e
WD
827 if (depth <= 0 || sanp != start) {
828 p += 2;
829 if (sanp != start) {
830 /* back up sanp one level */
831 --sanp; /* now pointing at slash */
832 while (sanp > start && sanp[-1] != '/') {
833 /* skip back up to slash */
834 sanp--;
835 }
b5f9e67d 836 }
8e5f029e 837 continue;
1b8e662a 838 }
8e5f029e
WD
839 /* allow depth levels of .. at the beginning */
840 depth--;
841 /* move the virtual beginning to leave the .. alone */
842 start = sanp + 3;
1b8e662a 843 }
2d41264e
WD
844 /* copy one component through next slash */
845 while (*p && (*sanp++ = *p++) != '/') {}
1b8e662a 846 }
84a63795 847 if (sanp == dest) {
b5f9e67d 848 /* ended up with nothing, so put in "." component */
44e2e578 849 *sanp++ = '.';
b5f9e67d 850 }
44e2e578 851 *sanp = '\0';
1b8e662a 852
84a63795 853 return dest;
14b61c63 854}
5243c216 855
a8167c66 856/* Like chdir(), but it keeps track of the current directory (in the
a16d8f2b 857 * global "curr_dir"), and ensures that the path size doesn't overflow.
a8167c66 858 * Also cleans the path using the clean_fname() function. */
2c708476 859int push_dir(const char *dir, int set_path_only)
5243c216 860{
5243c216 861 static int initialised;
4af8fe4e 862 unsigned int len;
5243c216
AT
863
864 if (!initialised) {
865 initialised = 1;
5cb37436 866 getcwd(curr_dir, sizeof curr_dir - 1);
4af8fe4e 867 curr_dir_len = strlen(curr_dir);
5243c216
AT
868 }
869
4af8fe4e
WD
870 if (!dir) /* this call was probably just to initialize */
871 return 0;
c226b7c2 872
4af8fe4e
WD
873 len = strlen(dir);
874 if (len == 1 && *dir == '.')
875 return 1;
5243c216 876
7c73536c
WD
877 if ((*dir == '/' ? len : curr_dir_len + 1 + len) >= sizeof curr_dir) {
878 errno = ENAMETOOLONG;
4af8fe4e 879 return 0;
7c73536c 880 }
4af8fe4e 881
ba081be3 882 if (!set_path_only && chdir(dir))
4af8fe4e 883 return 0;
5243c216
AT
884
885 if (*dir == '/') {
4af8fe4e
WD
886 memcpy(curr_dir, dir, len + 1);
887 curr_dir_len = len;
888 } else {
889 curr_dir[curr_dir_len++] = '/';
890 memcpy(curr_dir + curr_dir_len, dir, len + 1);
891 curr_dir_len += len;
5243c216
AT
892 }
893
6bb82fe0 894 curr_dir_len = clean_fname(curr_dir, CFN_COLLAPSE_DOT_DOT_DIRS);
a8167c66
WD
895 if (sanitize_paths) {
896 if (module_dirlen > curr_dir_len)
897 module_dirlen = curr_dir_len;
898 curr_dir_depth = count_dir_elements(curr_dir + module_dirlen);
899 }
5243c216 900
af5ed0f2
WD
901 if (verbose >= 5)
902 rprintf(FINFO, "[%s] dir is now %s\n", who_am_i(), curr_dir);
903
4af8fe4e 904 return 1;
5243c216
AT
905}
906
a16d8f2b
WD
907/**
908 * Reverse a push_dir() call. You must pass in an absolute path
909 * that was copied from a prior value of "curr_dir".
910 **/
2c708476 911int pop_dir(const char *dir)
5243c216 912{
4af8fe4e
WD
913 if (chdir(dir))
914 return 0;
5243c216 915
4af8fe4e
WD
916 curr_dir_len = strlcpy(curr_dir, dir, sizeof curr_dir);
917 if (curr_dir_len >= sizeof curr_dir)
918 curr_dir_len = sizeof curr_dir - 1;
a8167c66
WD
919 if (sanitize_paths)
920 curr_dir_depth = count_dir_elements(curr_dir + module_dirlen);
5243c216 921
4af8fe4e 922 return 1;
5243c216 923}
aa9b77a5 924
eb61be19
WD
925/**
926 * Return a quoted string with the full pathname of the indicated filename.
927 * The string " (in MODNAME)" may also be appended. The returned pointer
928 * remains valid until the next time full_fname() is called.
929 **/
9a5ade18 930char *full_fname(const char *fn)
eb61be19 931{
eb61be19
WD
932 static char *result = NULL;
933 char *m1, *m2, *m3;
934 char *p1, *p2;
935
936 if (result)
937 free(result);
938
939 if (*fn == '/')
940 p1 = p2 = "";
941 else {
4cff5fa4 942 p1 = curr_dir + module_dirlen;
bc83274a
WD
943 for (p2 = p1; *p2 == '/'; p2++) {}
944 if (*p2)
945 p2 = "/";
eb61be19
WD
946 }
947 if (module_id >= 0) {
948 m1 = " (in ";
949 m2 = lp_name(module_id);
950 m3 = ")";
eb61be19
WD
951 } else
952 m1 = m2 = m3 = "";
953
c98ad3df
WD
954 if (asprintf(&result, "\"%s%s%s\"%s%s%s", p1, p2, fn, m1, m2, m3) <= 0)
955 out_of_memory("full_fname");
eb61be19
WD
956
957 return result;
958}
959
a7260c40
WD
960static char partial_fname[MAXPATHLEN];
961
962char *partial_dir_fname(const char *fname)
963{
964 char *t = partial_fname;
965 int sz = sizeof partial_fname;
966 const char *fn;
967
968 if ((fn = strrchr(fname, '/')) != NULL) {
969 fn++;
970 if (*partial_dir != '/') {
971 int len = fn - fname;
972 strncpy(t, fname, len); /* safe */
973 t += len;
974 sz -= len;
975 }
976 } else
977 fn = fname;
978 if ((int)pathjoin(t, sz, partial_dir, fn) >= sz)
979 return NULL;
5aa7b20a 980 if (server_filter_list.head) {
89363676
WD
981 t = strrchr(partial_fname, '/');
982 *t = '\0';
5aa7b20a
WD
983 if (check_filter(&server_filter_list, partial_fname, 1) < 0)
984 return NULL;
89363676 985 *t = '/';
5aa7b20a
WD
986 if (check_filter(&server_filter_list, partial_fname, 0) < 0)
987 return NULL;
988 }
a7260c40
WD
989
990 return partial_fname;
991}
992
993/* If no --partial-dir option was specified, we don't need to do anything
994 * (the partial-dir is essentially '.'), so just return success. */
995int handle_partial_dir(const char *fname, int create)
996{
997 char *fn, *dir;
998
999 if (fname != partial_fname)
1000 return 1;
1001 if (!create && *partial_dir == '/')
1002 return 1;
1003 if (!(fn = strrchr(partial_fname, '/')))
1004 return 1;
1005
1006 *fn = '\0';
1007 dir = partial_fname;
1008 if (create) {
1009 STRUCT_STAT st;
a7260c40 1010 int statret = do_lstat(dir, &st);
a7260c40
WD
1011 if (statret == 0 && !S_ISDIR(st.st_mode)) {
1012 if (do_unlink(dir) < 0)
1013 return 0;
1014 statret = -1;
1015 }
1016 if (statret < 0 && do_mkdir(dir, 0700) < 0)
1017 return 0;
1018 } else
1019 do_rmdir(dir);
1020 *fn = '/';
1021
1022 return 1;
1023}
1024
ac13ad10
MP
1025/**
1026 * Determine if a symlink points outside the current directory tree.
036e70b0
MP
1027 * This is considered "unsafe" because e.g. when mirroring somebody
1028 * else's machine it might allow them to establish a symlink to
1029 * /etc/passwd, and then read it through a web server.
1030 *
4e5db0ad
MP
1031 * Null symlinks and absolute symlinks are always unsafe.
1032 *
1033 * Basically here we are concerned with symlinks whose target contains
1034 * "..", because this might cause us to walk back up out of the
1035 * transferred directory. We are not allowed to go back up and
1036 * reenter.
1037 *
036e70b0
MP
1038 * @param dest Target of the symlink in question.
1039 *
25d34a5c 1040 * @param src Top source directory currently applicable. Basically this
036e70b0 1041 * is the first parameter to rsync in a simple invocation, but it's
25d34a5c 1042 * modified by flist.c in slightly complex ways.
036e70b0
MP
1043 *
1044 * @retval True if unsafe
1045 * @retval False is unsafe
4e5db0ad
MP
1046 *
1047 * @sa t_unsafe.c
ac13ad10 1048 **/
7afa3a4a 1049int unsafe_symlink(const char *dest, const char *src)
4b957c22 1050{
7afa3a4a 1051 const char *name, *slash;
4b957c22
AT
1052 int depth = 0;
1053
1054 /* all absolute and null symlinks are unsafe */
b5bd5542
WD
1055 if (!dest || !*dest || *dest == '/')
1056 return 1;
4b957c22
AT
1057
1058 /* find out what our safety margin is */
7afa3a4a
WD
1059 for (name = src; (slash = strchr(name, '/')) != 0; name = slash+1) {
1060 if (strncmp(name, "../", 3) == 0) {
c284f34a 1061 depth = 0;
7afa3a4a 1062 } else if (strncmp(name, "./", 2) == 0) {
4b957c22
AT
1063 /* nothing */
1064 } else {
1065 depth++;
1066 }
1067 }
7afa3a4a
WD
1068 if (strcmp(name, "..") == 0)
1069 depth = 0;
4b957c22 1070
7afa3a4a
WD
1071 for (name = dest; (slash = strchr(name, '/')) != 0; name = slash+1) {
1072 if (strncmp(name, "../", 3) == 0) {
1073 /* if at any point we go outside the current directory
1074 then stop - it is unsafe */
1075 if (--depth < 0)
1076 return 1;
1077 } else if (strncmp(name, "./", 2) == 0) {
4b957c22
AT
1078 /* nothing */
1079 } else {
1080 depth++;
1081 }
4b957c22 1082 }
7afa3a4a
WD
1083 if (strcmp(name, "..") == 0)
1084 depth--;
4b957c22 1085
4b957c22
AT
1086 return (depth < 0);
1087}
375a4556 1088
e175fb07
WD
1089/* Return the int64 number as a string. If the --human-readable option was
1090 * specified, we may output the number in K, M, or G units. We can return
1091 * up to 4 buffers at a time. */
1092char *human_num(int64 num)
1093{
1094 static char bufs[4][128]; /* more than enough room */
1095 static unsigned int n;
1096 char *s;
1097
1098 n = (n + 1) % (sizeof bufs / sizeof bufs[0]);
1099
1100 if (human_readable) {
1101 char units = '\0';
7794db7c 1102 int mult = human_readable == 1 ? 1000 : 1024;
e175fb07
WD
1103 double dnum = 0;
1104 if (num > mult*mult*mult) {
1105 dnum = (double)num / (mult*mult*mult);
1106 units = 'G';
1107 } else if (num > mult*mult) {
1108 dnum = (double)num / (mult*mult);
1109 units = 'M';
1110 } else if (num > mult) {
1111 dnum = (double)num / mult;
1112 units = 'K';
1113 }
1114 if (units) {
83078af5 1115 snprintf(bufs[n], sizeof bufs[0], "%.2f%c", dnum, units);
e175fb07
WD
1116 return bufs[n];
1117 }
1118 }
1119
1120 s = bufs[n] + sizeof bufs[0] - 1;
1121 *s = '\0';
1122
1123 if (!num)
1124 *--s = '0';
1125 while (num) {
26404276 1126 *--s = (char)(num % 10) + '0';
e175fb07
WD
1127 num /= 10;
1128 }
1129 return s;
1130}
1131
1132/* Return the double number as a string. If the --human-readable option was
1133 * specified, we may output the number in K, M, or G units. We use a buffer
1134 * from human_num() to return our result. */
1135char *human_dnum(double dnum, int decimal_digits)
1136{
1137 char *buf = human_num(dnum);
1138 int len = strlen(buf);
2dc7b8bd 1139 if (isDigit(buf + len - 1)) {
e175fb07
WD
1140 /* There's extra room in buf prior to the start of the num. */
1141 buf -= decimal_digits + 1;
1142 snprintf(buf, len + decimal_digits + 2, "%.*f", decimal_digits, dnum);
1143 }
1144 return buf;
1145}
f7632fc6 1146
4a19c3b2 1147/* Return the date and time as a string. Some callers tweak returned buf. */
f7632fc6
AT
1148char *timestring(time_t t)
1149{
1150 static char TimeBuf[200];
1151 struct tm *tm = localtime(&t);
afa73c75 1152 char *p;
f7632fc6 1153
4f5b0756 1154#ifdef HAVE_STRFTIME
5cb37436 1155 strftime(TimeBuf, sizeof TimeBuf - 1, "%Y/%m/%d %H:%M:%S", tm);
f7632fc6 1156#else
5cb37436 1157 strlcpy(TimeBuf, asctime(tm), sizeof TimeBuf);
f7632fc6
AT
1158#endif
1159
afa73c75
WD
1160 if ((p = strchr(TimeBuf, '\n')) != NULL)
1161 *p = '\0';
f7632fc6 1162
afa73c75 1163 return TimeBuf;
f7632fc6
AT
1164}
1165
e1bd49d6
MP
1166/**
1167 * Sleep for a specified number of milliseconds.
1168 *
1169 * Always returns TRUE. (In the future it might return FALSE if
1170 * interrupted.)
1171 **/
1172int msleep(int t)
9ec16c83 1173{
c284f34a
WD
1174 int tdiff = 0;
1175 struct timeval tval, t1, t2;
9ec16c83
AT
1176
1177 gettimeofday(&t1, NULL);
5cb37436 1178
9ec16c83
AT
1179 while (tdiff < t) {
1180 tval.tv_sec = (t-tdiff)/1000;
1181 tval.tv_usec = 1000*((t-tdiff)%1000);
5cb37436 1182
9ec16c83
AT
1183 errno = 0;
1184 select(0,NULL,NULL, NULL, &tval);
1185
1186 gettimeofday(&t2, NULL);
5cb37436 1187 tdiff = (t2.tv_sec - t1.tv_sec)*1000 +
9ec16c83
AT
1188 (t2.tv_usec - t1.tv_usec)/1000;
1189 }
e1bd49d6
MP
1190
1191 return True;
9ec16c83
AT
1192}
1193
c2b54076
WD
1194/* Determine if two time_t values are equivalent (either exact, or in
1195 * the modification timestamp window established by --modify-window).
ac13ad10
MP
1196 *
1197 * @retval 0 if the times should be treated as the same
1198 *
1199 * @retval +1 if the first is later
1200 *
1201 * @retval -1 if the 2nd is later
1202 **/
c2b54076 1203int cmp_time(time_t file1, time_t file2)
5b56cc19 1204{
5b56cc19 1205 if (file2 > file1) {
bc6ebcd2
WD
1206 if (file2 - file1 <= modify_window)
1207 return 0;
5b56cc19
AT
1208 return -1;
1209 }
bc6ebcd2
WD
1210 if (file1 - file2 <= modify_window)
1211 return 0;
5b56cc19
AT
1212 return 1;
1213}
1214
1215
1216#ifdef __INSURE__XX
0f8f98c8
AT
1217#include <dlfcn.h>
1218
ac13ad10
MP
1219/**
1220 This routine is a trick to immediately catch errors when debugging
1221 with insure. A xterm with a gdb is popped up when insure catches
1222 a error. It is Linux specific.
1223**/
0f8f98c8
AT
1224int _Insure_trap_error(int a1, int a2, int a3, int a4, int a5, int a6)
1225{
1226 static int (*fn)();
1227 int ret;
8950ac03 1228 char *cmd;
0f8f98c8 1229
5cb37436 1230 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
1231 getpid(), getpid(), getpid());
1232
1233 if (!fn) {
1234 static void *h;
1235 h = dlopen("/usr/local/parasoft/insure++lite/lib.linux2/libinsure.so", RTLD_LAZY);
1236 fn = dlsym(h, "_Insure_trap_error");
1237 }
1238
1239 ret = fn(a1, a2, a3, a4, a5, a6);
1240
1241 system(cmd);
1242
8950ac03
AT
1243 free(cmd);
1244
0f8f98c8
AT
1245 return ret;
1246}
1247#endif
58cadc86 1248
58cadc86
WD
1249#define MALLOC_MAX 0x40000000
1250
a2dc4d68 1251void *_new_array(unsigned long num, unsigned int size, int use_calloc)
58cadc86
WD
1252{
1253 if (num >= MALLOC_MAX/size)
1254 return NULL;
a2dc4d68 1255 return use_calloc ? calloc(num, size) : malloc(num * size);
58cadc86
WD
1256}
1257
1258void *_realloc_array(void *ptr, unsigned int size, unsigned long num)
1259{
1260 if (num >= MALLOC_MAX/size)
1261 return NULL;
58cadc86
WD
1262 if (!ptr)
1263 return malloc(size * num);
1264 return realloc(ptr, size * num);
1265}
e64ae6d7
WD
1266
1267/* Take a filename and filename length and return the most significant
1268 * filename suffix we can find. This ignores suffixes such as "~",
1269 * ".bak", ".orig", ".~1~", etc. */
1270const char *find_filename_suffix(const char *fn, int fn_len, int *len_ptr)
1271{
1272 const char *suf, *s;
1273 BOOL had_tilde;
1274 int s_len;
1275
1276 /* One or more dots at the start aren't a suffix. */
1277 while (fn_len && *fn == '.') fn++, fn_len--;
1278
1279 /* Ignore the ~ in a "foo~" filename. */
1280 if (fn_len > 1 && fn[fn_len-1] == '~')
1281 fn_len--, had_tilde = True;
1282 else
1283 had_tilde = False;
1284
1285 /* Assume we don't find an suffix. */
1286 suf = "";
1287 *len_ptr = 0;
1288
1289 /* Find the last significant suffix. */
1290 for (s = fn + fn_len; fn_len > 1; ) {
1291 while (*--s != '.' && s != fn) {}
1292 if (s == fn)
1293 break;
1294 s_len = fn_len - (s - fn);
1295 fn_len = s - fn;
6012eaa1 1296 if (s_len == 4) {
e64ae6d7
WD
1297 if (strcmp(s+1, "bak") == 0
1298 || strcmp(s+1, "old") == 0)
1299 continue;
6012eaa1 1300 } else if (s_len == 5) {
e64ae6d7
WD
1301 if (strcmp(s+1, "orig") == 0)
1302 continue;
1303 } else if (s_len > 2 && had_tilde
2dc7b8bd 1304 && s[1] == '~' && isDigit(s + 2))
e64ae6d7
WD
1305 continue;
1306 *len_ptr = s_len;
1307 suf = s;
1308 if (s_len == 1)
1309 break;
1310 /* Determine if the suffix is all digits. */
1311 for (s++, s_len--; s_len > 0; s++, s_len--) {
2dc7b8bd 1312 if (!isDigit(s))
e64ae6d7
WD
1313 return suf;
1314 }
1315 /* An all-digit suffix may not be that signficant. */
1316 s = suf;
1317 }
1318
1319 return suf;
1320}
1321
1322/* This is an implementation of the Levenshtein distance algorithm. It
1323 * was implemented to avoid needing a two-dimensional matrix (to save
1324 * memory). It was also tweaked to try to factor in the ASCII distance
1325 * between changed characters as a minor distance quantity. The normal
1326 * Levenshtein units of distance (each signifying a single change between
1327 * the two strings) are defined as a "UNIT". */
1328
1329#define UNIT (1 << 16)
1330
1331uint32 fuzzy_distance(const char *s1, int len1, const char *s2, int len2)
1332{
1333 uint32 a[MAXPATHLEN], diag, above, left, diag_inc, above_inc, left_inc;
1334 int32 cost;
1335 int i1, i2;
1336
1337 if (!len1 || !len2) {
1338 if (!len1) {
1339 s1 = s2;
1340 len1 = len2;
1341 }
1342 for (i1 = 0, cost = 0; i1 < len1; i1++)
1343 cost += s1[i1];
1344 return (int32)len1 * UNIT + cost;
1345 }
1346
1347 for (i2 = 0; i2 < len2; i2++)
1348 a[i2] = (i2+1) * UNIT;
1349
1350 for (i1 = 0; i1 < len1; i1++) {
1351 diag = i1 * UNIT;
1352 above = (i1+1) * UNIT;
1353 for (i2 = 0; i2 < len2; i2++) {
1354 left = a[i2];
1355 if ((cost = *((uchar*)s1+i1) - *((uchar*)s2+i2)) != 0) {
1356 if (cost < 0)
1357 cost = UNIT - cost;
1358 else
1359 cost = UNIT + cost;
1360 }
1361 diag_inc = diag + cost;
1362 left_inc = left + UNIT + *((uchar*)s1+i1);
1363 above_inc = above + UNIT + *((uchar*)s2+i2);
1364 a[i2] = above = left < above
1365 ? (left_inc < diag_inc ? left_inc : diag_inc)
1366 : (above_inc < diag_inc ? above_inc : diag_inc);
1367 diag = left;
1368 }
1369 }
1370
1371 return a[len2-1];
1372}
c2b54076
WD
1373
1374#define BB_SLOT_SIZE (16*1024) /* Desired size in bytes */
1375#define BB_PER_SLOT_BITS (BB_SLOT_SIZE * 8) /* Number of bits per slot */
1376#define BB_PER_SLOT_INTS (BB_SLOT_SIZE / 4) /* Number of int32s per slot */
1377
1378struct bitbag {
1379 uint32 **bits;
1380 int slot_cnt;
1381};
1382
1383struct bitbag *bitbag_create(int max_ndx)
1384{
1385 struct bitbag *bb = new(struct bitbag);
1386 bb->slot_cnt = (max_ndx + BB_PER_SLOT_BITS - 1) / BB_PER_SLOT_BITS;
1387
1388 if (!(bb->bits = (uint32**)calloc(bb->slot_cnt, sizeof (uint32*))))
1389 out_of_memory("bitbag_create");
1390
1391 return bb;
1392}
1393
1394void bitbag_set_bit(struct bitbag *bb, int ndx)
1395{
1396 int slot = ndx / BB_PER_SLOT_BITS;
1397 ndx %= BB_PER_SLOT_BITS;
1398
1399 if (!bb->bits[slot]) {
1400 if (!(bb->bits[slot] = (uint32*)calloc(BB_PER_SLOT_INTS, 4)))
1401 out_of_memory("bitbag_set_bit");
1402 }
1403
1404 bb->bits[slot][ndx/32] |= 1u << (ndx % 32);
1405}
1406
1407#if 0 /* not needed yet */
1408void bitbag_clear_bit(struct bitbag *bb, int ndx)
1409{
1410 int slot = ndx / BB_PER_SLOT_BITS;
1411 ndx %= BB_PER_SLOT_BITS;
1412
1413 if (!bb->bits[slot])
1414 return;
1415
1416 bb->bits[slot][ndx/32] &= ~(1u << (ndx % 32));
1417}
1418
1419int bitbag_check_bit(struct bitbag *bb, int ndx)
1420{
1421 int slot = ndx / BB_PER_SLOT_BITS;
1422 ndx %= BB_PER_SLOT_BITS;
1423
1424 if (!bb->bits[slot])
1425 return 0;
1426
1427 return bb->bits[slot][ndx/32] & (1u << (ndx % 32)) ? 1 : 0;
1428}
1429#endif
1430
1431/* Call this with -1 to start checking from 0. Returns -1 at the end. */
1432int bitbag_next_bit(struct bitbag *bb, int after)
1433{
1434 uint32 bits, mask;
1435 int i, ndx = after + 1;
1436 int slot = ndx / BB_PER_SLOT_BITS;
1437 ndx %= BB_PER_SLOT_BITS;
1438
1439 mask = (1u << (ndx % 32)) - 1;
1440 for (i = ndx / 32; slot < bb->slot_cnt; slot++, i = mask = 0) {
1441 if (!bb->bits[slot])
1442 continue;
1443 for ( ; i < BB_PER_SLOT_INTS; i++, mask = 0) {
1444 if (!(bits = bb->bits[slot][i] & ~mask))
1445 continue;
1446 /* The xor magic figures out the lowest enabled bit in
1447 * bits, and the switch quickly computes log2(bit). */
1448 switch (bits ^ (bits & (bits-1))) {
1449#define LOG2(n) case 1u << n: return slot*BB_PER_SLOT_BITS + i*32 + n
1450 LOG2(0); LOG2(1); LOG2(2); LOG2(3);
1451 LOG2(4); LOG2(5); LOG2(6); LOG2(7);
1452 LOG2(8); LOG2(9); LOG2(10); LOG2(11);
1453 LOG2(12); LOG2(13); LOG2(14); LOG2(15);
1454 LOG2(16); LOG2(17); LOG2(18); LOG2(19);
1455 LOG2(20); LOG2(21); LOG2(22); LOG2(23);
1456 LOG2(24); LOG2(25); LOG2(26); LOG2(27);
1457 LOG2(28); LOG2(29); LOG2(30); LOG2(31);
1458 }
1459 return -1; /* impossible... */
1460 }
1461 }
1462
1463 return -1;
1464}
1c3344a1
WD
1465
1466void *expand_item_list(item_list *lp, size_t item_size,
1467 const char *desc, int incr)
1468{
1469 /* First time through, 0 <= 0, so list is expanded. */
1470 if (lp->malloced <= lp->count) {
1471 void *new_ptr;
1472 size_t new_size = lp->malloced;
1473 if (incr < 0)
0b515981 1474 new_size += -incr; /* increase slowly */
1c3344a1
WD
1475 else if (new_size < (size_t)incr)
1476 new_size += incr;
1477 else
1478 new_size *= 2;
1479 new_ptr = realloc_array(lp->items, char, new_size * item_size);
1480 if (verbose >= 4) {
1481 rprintf(FINFO, "[%s] expand %s to %.0f bytes, did%s move\n",
1482 who_am_i(), desc, (double)new_size * item_size,
1483 new_ptr == lp->items ? " not" : "");
1484 }
1485 if (!new_ptr)
1486 out_of_memory("expand_item_list");
1487
1488 lp->items = new_ptr;
1489 lp->malloced = new_size;
1490 }
1491 return (char*)lp->items + (lp->count++ * item_size);
1492}