util.c

   1 /*
   2  * Utility routines used in rsync.
   3  *
   4  * Copyright (C) 1996-2000 Andrew Tridgell
   5  * Copyright (C) 1996 Paul Mackerras
   6  * Copyright (C) 2001, 2002 Martin Pool <mbp@samba.org>
   7  * Copyright (C) 2003-2008 Wayne Davison
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 3 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License along
  20  * with this program; if not, visit the http://fsf.org website.
  21  */
  22
  23 #include "rsync.h"
  24 #include "ifuncs.h"
  25
  26 extern int verbose;
  27 extern int dry_run;
  28 extern int module_id;
  29 extern int modify_window;
  30 extern int relative_paths;
  31 extern int human_readable;
  32 extern int preserve_xattrs;
  33 extern char *module_dir;
  34 extern unsigned int module_dirlen;
  35 extern mode_t orig_umask;
  36 extern char *partial_dir;
  37 extern struct filter_list_struct daemon_filter_list;
  38
  39 int sanitize_paths = 0;
  40
  41 char curr_dir[MAXPATHLEN];
  42 unsigned int curr_dir_len;
  43 int curr_dir_depth; /* This is only set for a sanitizing daemon. */
  44
  45 /* Set a fd into nonblocking mode. */
  46 void set_nonblocking(int fd)
  47 {
  48         int val;
  49
  50         if ((val = fcntl(fd, F_GETFL)) == -1)
  51                 return;
  52         if (!(val & NONBLOCK_FLAG)) {
  53                 val |= NONBLOCK_FLAG;
  54                 fcntl(fd, F_SETFL, val);
  55         }
  56 }
  57
  58 /* Set a fd into blocking mode. */
  59 void set_blocking(int fd)
  60 {
  61         int val;
  62
  63         if ((val = fcntl(fd, F_GETFL)) == -1)
  64                 return;
  65         if (val & NONBLOCK_FLAG) {
  66                 val &= ~NONBLOCK_FLAG;
  67                 fcntl(fd, F_SETFL, val);
  68         }
  69 }
  70
  71 /**
  72  * Create a file descriptor pair - like pipe() but use socketpair if
  73  * possible (because of blocking issues on pipes).
  74  *
  75  * Always set non-blocking.
  76  */
  77 int fd_pair(int fd[2])
  78 {
  79         int ret;
  80
  81 #ifdef HAVE_SOCKETPAIR
  82         ret = socketpair(AF_UNIX, SOCK_STREAM, 0, fd);
  83 #else
  84         ret = pipe(fd);
  85 #endif
  86
  87         if (ret == 0) {
  88                 set_nonblocking(fd[0]);
  89                 set_nonblocking(fd[1]);
  90         }
  91
  92         return ret;
  93 }
  94
  95 void print_child_argv(const char *prefix, char **cmd)
  96 {
  97         rprintf(FCLIENT, "%s ", prefix);
  98         for (; *cmd; cmd++) {
  99                 /* Look for characters that ought to be quoted.  This
 100                 * is not a great quoting algorithm, but it's
 101                 * sufficient for a log message. */
 102                 if (strspn(*cmd, "abcdefghijklmnopqrstuvwxyz"
 103                            "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
 104                            "0123456789"
 105                            ",.-_=+@/") != strlen(*cmd)) {
 106                         rprintf(FCLIENT, "\"%s\" ", *cmd);
 107                 } else {
 108                         rprintf(FCLIENT, "%s ", *cmd);
 109                 }
 110         }
 111         rprintf(FCLIENT, "\n");
 112 }
 113
 114 NORETURN void out_of_memory(const char *str)
 115 {
 116         rprintf(FERROR, "ERROR: out of memory in %s [%s]\n", str, who_am_i());
 117         exit_cleanup(RERR_MALLOC);
 118 }
 119
 120 NORETURN void overflow_exit(const char *str)
 121 {
 122         rprintf(FERROR, "ERROR: buffer overflow in %s [%s]\n", str, who_am_i());
 123         exit_cleanup(RERR_MALLOC);
 124 }
 125
 126 int set_modtime(const char *fname, time_t modtime, mode_t mode)
 127 {
 128 #if !defined HAVE_LUTIMES || !defined HAVE_UTIMES
 129         if (S_ISLNK(mode))
 130                 return 1;
 131 #endif
 132
 133         if (verbose > 2) {
 134                 rprintf(FINFO, "set modtime of %s to (%ld) %s",
 135                         fname, (long)modtime,
 136                         asctime(localtime(&modtime)));
 137         }
 138
 139         if (dry_run)
 140                 return 0;
 141
 142         {
 143 #ifdef HAVE_UTIMES
 144                 struct timeval t[2];
 145                 t[0].tv_sec = time(NULL);
 146                 t[0].tv_usec = 0;
 147                 t[1].tv_sec = modtime;
 148                 t[1].tv_usec = 0;
 149 # ifdef HAVE_LUTIMES
 150                 if (S_ISLNK(mode)) {
 151                         if (lutimes(fname, t) < 0)
 152                                 return errno == ENOSYS ? 1 : -1;
 153                         return 0;
 154                 }
 155 # endif
 156                 return utimes(fname, t);
 157 #elif defined HAVE_STRUCT_UTIMBUF
 158                 struct utimbuf tbuf;
 159                 tbuf.actime = time(NULL);
 160                 tbuf.modtime = modtime;
 161                 return utime(fname,&tbuf);
 162 #elif defined HAVE_UTIME
 163                 time_t t[2];
 164                 t[0] = time(NULL);
 165                 t[1] = modtime;
 166                 return utime(fname,t);
 167 #else
 168 #error No file-time-modification routine found!
 169 #endif
 170         }
 171 }
 172
 173 /* This creates a new directory with default permissions.  Since there
 174  * might be some directory-default permissions affecting this, we can't
 175  * force the permissions directly using the original umask and mkdir(). */
 176 int mkdir_defmode(char *fname)
 177 {
 178         int ret;
 179
 180         umask(orig_umask);
 181         ret = do_mkdir(fname, ACCESSPERMS);
 182         umask(0);
 183
 184         return ret;
 185 }
 186
 187 /* Create any necessary directories in fname.  Any missing directories are
 188  * created with default permissions. */
 189 int create_directory_path(char *fname)
 190 {
 191         char *p;
 192         int ret = 0;
 193
 194         while (*fname == '/')
 195                 fname++;
 196         while (strncmp(fname, "./", 2) == 0)
 197                 fname += 2;
 198
 199         umask(orig_umask);
 200         p = fname;
 201         while ((p = strchr(p,'/')) != NULL) {
 202                 *p = '\0';
 203                 if (do_mkdir(fname, ACCESSPERMS) < 0 && errno != EEXIST)
 204                     ret = -1;
 205                 *p++ = '/';
 206         }
 207         umask(0);
 208
 209         return ret;
 210 }
 211
 212 /**
 213  * Write @p len bytes at @p ptr to descriptor @p desc, retrying if
 214  * interrupted.
 215  *
 216  * @retval len upon success
 217  *
 218  * @retval <0 write's (negative) error code
 219  *
 220  * Derived from GNU C's cccp.c.
 221  */
 222 int full_write(int desc, const char *ptr, size_t len)
 223 {
 224         int total_written;
 225
 226         total_written = 0;
 227         while (len > 0) {
 228                 int written = write(desc, ptr, len);
 229                 if (written < 0)  {
 230                         if (errno == EINTR)
 231                                 continue;
 232                         return written;
 233                 }
 234                 total_written += written;
 235                 ptr += written;
 236                 len -= written;
 237         }
 238         return total_written;
 239 }
 240
 241 /**
 242  * Read @p len bytes at @p ptr from descriptor @p desc, retrying if
 243  * interrupted.
 244  *
 245  * @retval >0 the actual number of bytes read
 246  *
 247  * @retval 0 for EOF
 248  *
 249  * @retval <0 for an error.
 250  *
 251  * Derived from GNU C's cccp.c. */
 252 static int safe_read(int desc, char *ptr, size_t len)
 253 {
 254         int n_chars;
 255
 256         if (len == 0)
 257                 return len;
 258
 259         do {
 260                 n_chars = read(desc, ptr, len);
 261         } while (n_chars < 0 && errno == EINTR);
 262
 263         return n_chars;
 264 }
 265
 266 /* Copy a file.  If ofd < 0, copy_file unlinks and opens the "dest" file.
 267  * Otherwise, it just writes to and closes the provided file descriptor.
 268  * In either case, if --xattrs are being preserved, the dest file will
 269  * have its xattrs set from the source file.
 270  *
 271  * This is used in conjunction with the --temp-dir, --backup, and
 272  * --copy-dest options. */
 273 int copy_file(const char *source, const char *dest, int ofd,
 274               mode_t mode, int create_bak_dir)
 275 {
 276         int ifd;
 277         char buf[1024 * 8];
 278         int len;   /* Number of bytes read into `buf'. */
 279
 280         if ((ifd = do_open(source, O_RDONLY, 0)) < 0) {
 281                 int save_errno = errno;
 282                 rsyserr(FERROR_XFER, errno, "open %s", full_fname(source));
 283                 errno = save_errno;
 284                 return -1;
 285         }
 286
 287         if (ofd < 0) {
 288                 if (robust_unlink(dest) && errno != ENOENT) {
 289                         int save_errno = errno;
 290                         rsyserr(FERROR_XFER, errno, "unlink %s", full_fname(dest));
 291                         errno = save_errno;
 292                         return -1;
 293                 }
 294
 295                 if ((ofd = do_open(dest, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, mode)) < 0) {
 296                         int save_errno = errno ? errno : EINVAL; /* 0 paranoia */
 297                         if (create_bak_dir && errno == ENOENT && make_bak_dir(dest) == 0) {
 298                                 if ((ofd = do_open(dest, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, mode)) < 0)
 299                                         save_errno = errno ? errno : save_errno;
 300                                 else
 301                                         save_errno = 0;
 302                         }
 303                         if (save_errno) {
 304                                 rsyserr(FERROR_XFER, save_errno, "open %s", full_fname(dest));
 305                                 close(ifd);
 306                                 errno = save_errno;
 307                                 return -1;
 308                         }
 309                 }
 310         }
 311
 312         while ((len = safe_read(ifd, buf, sizeof buf)) > 0) {
 313                 if (full_write(ofd, buf, len) < 0) {
 314                         int save_errno = errno;
 315                         rsyserr(FERROR_XFER, errno, "write %s", full_fname(dest));
 316                         close(ifd);
 317                         close(ofd);
 318                         errno = save_errno;
 319                         return -1;
 320                 }
 321         }
 322
 323         if (len < 0) {
 324                 int save_errno = errno;
 325                 rsyserr(FERROR_XFER, errno, "read %s", full_fname(source));
 326                 close(ifd);
 327                 close(ofd);
 328                 errno = save_errno;
 329                 return -1;
 330         }
 331
 332         if (close(ifd) < 0) {
 333                 rsyserr(FWARNING, errno, "close failed on %s",
 334                         full_fname(source));
 335         }
 336
 337         if (close(ofd) < 0) {
 338                 int save_errno = errno;
 339                 rsyserr(FERROR_XFER, errno, "close failed on %s",
 340                         full_fname(dest));
 341                 errno = save_errno;
 342                 return -1;
 343         }
 344
 345 #ifdef SUPPORT_XATTRS
 346         if (preserve_xattrs)
 347                 copy_xattrs(source, dest);
 348 #endif
 349
 350         return 0;
 351 }
 352
 353 /* MAX_RENAMES should be 10**MAX_RENAMES_DIGITS */
 354 #define MAX_RENAMES_DIGITS 3
 355 #define MAX_RENAMES 1000
 356
 357 /**
 358  * Robust unlink: some OS'es (HPUX) refuse to unlink busy files, so
 359  * rename to <path>/.rsyncNNN instead.
 360  *
 361  * Note that successive rsync runs will shuffle the filenames around a
 362  * bit as long as the file is still busy; this is because this function
 363  * does not know if the unlink call is due to a new file coming in, or
 364  * --delete trying to remove old .rsyncNNN files, hence it renames it
 365  * each time.
 366  **/
 367 int robust_unlink(const char *fname)
 368 {
 369 #ifndef ETXTBSY
 370         return do_unlink(fname);
 371 #else
 372         static int counter = 1;
 373         int rc, pos, start;
 374         char path[MAXPATHLEN];
 375
 376         rc = do_unlink(fname);
 377         if (rc == 0 || errno != ETXTBSY)
 378                 return rc;
 379
 380         if ((pos = strlcpy(path, fname, MAXPATHLEN)) >= MAXPATHLEN)
 381                 pos = MAXPATHLEN - 1;
 382
 383         while (pos > 0 && path[pos-1] != '/')
 384                 pos--;
 385         pos += strlcpy(path+pos, ".rsync", MAXPATHLEN-pos);
 386
 387         if (pos > (MAXPATHLEN-MAX_RENAMES_DIGITS-1)) {
 388                 errno = ETXTBSY;
 389                 return -1;
 390         }
 391
 392         /* start where the last one left off to reduce chance of clashes */
 393         start = counter;
 394         do {
 395                 snprintf(&path[pos], MAX_RENAMES_DIGITS+1, "%03d", counter);
 396                 if (++counter >= MAX_RENAMES)
 397                         counter = 1;
 398         } while ((rc = access(path, 0)) == 0 && counter != start);
 399
 400         if (verbose > 0) {
 401                 rprintf(FWARNING, "renaming %s to %s because of text busy\n",
 402                         fname, path);
 403         }
 404
 405         /* maybe we should return rename()'s exit status? Nah. */
 406         if (do_rename(fname, path) != 0) {
 407                 errno = ETXTBSY;
 408                 return -1;
 409         }
 410         return 0;
 411 #endif
 412 }
 413
 414 /* Returns 0 on successful rename, 1 if we successfully copied the file
 415  * across filesystems, -2 if copy_file() failed, and -1 on other errors.
 416  * If partialptr is not NULL and we need to do a copy, copy the file into
 417  * the active partial-dir instead of over the destination file. */
 418 int robust_rename(const char *from, const char *to, const char *partialptr,
 419                   int mode)
 420 {
 421         int tries = 4;
 422
 423         while (tries--) {
 424                 if (do_rename(from, to) == 0)
 425                         return 0;
 426
 427                 switch (errno) {
 428 #ifdef ETXTBSY
 429                 case ETXTBSY:
 430                         if (robust_unlink(to) != 0) {
 431                                 errno = ETXTBSY;
 432                                 return -1;
 433                         }
 434                         errno = ETXTBSY;
 435                         break;
 436 #endif
 437                 case EXDEV:
 438                         if (partialptr) {
 439                                 if (!handle_partial_dir(partialptr,PDIR_CREATE))
 440                                         return -1;
 441                                 to = partialptr;
 442                         }
 443                         if (copy_file(from, to, -1, mode, 0) != 0)
 444                                 return -2;
 445                         do_unlink(from);
 446                         return 1;
 447                 default:
 448                         return -1;
 449                 }
 450         }
 451         return -1;
 452 }
 453
 454 static pid_t all_pids[10];
 455 static int num_pids;
 456
 457 /** Fork and record the pid of the child. **/
 458 pid_t do_fork(void)
 459 {
 460         pid_t newpid = fork();
 461
 462         if (newpid != 0  &&  newpid != -1) {
 463                 all_pids[num_pids++] = newpid;
 464         }
 465         return newpid;
 466 }
 467
 468 /**
 469  * Kill all children.
 470  *
 471  * @todo It would be kind of nice to make sure that they are actually
 472  * all our children before we kill them, because their pids may have
 473  * been recycled by some other process.  Perhaps when we wait for a
 474  * child, we should remove it from this array.  Alternatively we could
 475  * perhaps use process groups, but I think that would not work on
 476  * ancient Unix versions that don't support them.
 477  **/
 478 void kill_all(int sig)
 479 {
 480         int i;
 481
 482         for (i = 0; i < num_pids; i++) {
 483                 /* Let's just be a little careful where we
 484                  * point that gun, hey?  See kill(2) for the
 485                  * magic caused by negative values. */
 486                 pid_t p = all_pids[i];
 487
 488                 if (p == getpid())
 489                         continue;
 490                 if (p <= 0)
 491                         continue;
 492
 493                 kill(p, sig);
 494         }
 495 }
 496
 497 /** Turn a user name into a uid */
 498 int name_to_uid(const char *name, uid_t *uid_p)
 499 {
 500         struct passwd *pass;
 501         if (!name || !*name)
 502                 return 0;
 503         if (!(pass = getpwnam(name)))
 504                 return 0;
 505         *uid_p = pass->pw_uid;
 506         return 1;
 507 }
 508
 509 /** Turn a group name into a gid */
 510 int name_to_gid(const char *name, gid_t *gid_p)
 511 {
 512         struct group *grp;
 513         if (!name || !*name)
 514                 return 0;
 515         if (!(grp = getgrnam(name)))
 516                 return 0;
 517         *gid_p = grp->gr_gid;
 518         return 1;
 519 }
 520
 521 /** Lock a byte range in a open file */
 522 int lock_range(int fd, int offset, int len)
 523 {
 524         struct flock lock;
 525
 526         lock.l_type = F_WRLCK;
 527         lock.l_whence = SEEK_SET;
 528         lock.l_start = offset;
 529         lock.l_len = len;
 530         lock.l_pid = 0;
 531
 532         return fcntl(fd,F_SETLK,&lock) == 0;
 533 }
 534
 535 #define ENSURE_MEMSPACE(buf, type, sz, req) \
 536         if ((req) > sz && !(buf = realloc_array(buf, type, sz = MAX(sz * 2, req)))) \
 537                 out_of_memory("glob_expand")
 538
 539 static inline void call_glob_match(const char *name, int len, int from_glob,
 540                                    char *arg, int abpos, int fbpos);
 541
 542 static struct glob_data {
 543         char *arg_buf, *filt_buf, **argv;
 544         int absize, fbsize, maxargs, argc;
 545 } glob;
 546
 547 static void glob_match(char *arg, int abpos, int fbpos)
 548 {
 549         int len;
 550         char *slash;
 551
 552         while (*arg == '.' && arg[1] == '/') {
 553                 if (fbpos < 0) {
 554                         ENSURE_MEMSPACE(glob.filt_buf, char, glob.fbsize, glob.absize);
 555                         memcpy(glob.filt_buf, glob.arg_buf, abpos + 1);
 556                         fbpos = abpos;
 557                 }
 558                 ENSURE_MEMSPACE(glob.arg_buf, char, glob.absize, abpos + 3);
 559                 glob.arg_buf[abpos++] = *arg++;
 560                 glob.arg_buf[abpos++] = *arg++;
 561                 glob.arg_buf[abpos] = '\0';
 562         }
 563         if ((slash = strchr(arg, '/')) != NULL) {
 564                 *slash = '\0';
 565                 len = slash - arg;
 566         } else
 567                 len = strlen(arg);
 568         if (strpbrk(arg, "*?[")) {
 569                 struct dirent *di;
 570                 DIR *d;
 571
 572                 if (!(d = opendir(abpos ? glob.arg_buf : ".")))
 573                         return;
 574                 while ((di = readdir(d)) != NULL) {
 575                         char *dname = d_name(di);
 576                         if (dname[0] == '.' && (dname[1] == '\0'
 577                           || (dname[1] == '.' && dname[2] == '\0')))
 578                                 continue;
 579                         if (!wildmatch(arg, dname))
 580                                 continue;
 581                         call_glob_match(dname, strlen(dname), 1,
 582                                         slash ? arg + len + 1 : NULL,
 583                                         abpos, fbpos);
 584                 }
 585                 closedir(d);
 586         } else {
 587                 call_glob_match(arg, len, 0,
 588                                 slash ? arg + len + 1 : NULL,
 589                                 abpos, fbpos);
 590         }
 591         if (slash)
 592                 *slash = '/';
 593 }
 594
 595 static inline void call_glob_match(const char *name, int len, int from_glob,
 596                                    char *arg, int abpos, int fbpos)
 597 {
 598         char *use_buf;
 599
 600         ENSURE_MEMSPACE(glob.arg_buf, char, glob.absize, abpos + len + 2);
 601         memcpy(glob.arg_buf + abpos, name, len);
 602         abpos += len;
 603         glob.arg_buf[abpos] = '\0';
 604
 605         if (fbpos >= 0) {
 606                 ENSURE_MEMSPACE(glob.filt_buf, char, glob.fbsize, fbpos + len + 2);
 607                 memcpy(glob.filt_buf + fbpos, name, len);
 608                 fbpos += len;
 609                 glob.filt_buf[fbpos] = '\0';
 610                 use_buf = glob.filt_buf;
 611         } else
 612                 use_buf = glob.arg_buf;
 613
 614         if (from_glob || (arg && len)) {
 615                 STRUCT_STAT st;
 616                 int is_dir;
 617
 618                 if (do_stat(glob.arg_buf, &st) != 0)
 619                         return;
 620                 is_dir = S_ISDIR(st.st_mode) != 0;
 621                 if (arg && !is_dir)
 622                         return;
 623
 624                 if (daemon_filter_list.head
 625                  && check_filter(&daemon_filter_list, FLOG, use_buf, is_dir) < 0)
 626                         return;
 627         }
 628
 629         if (arg) {
 630                 glob.arg_buf[abpos++] = '/';
 631                 glob.arg_buf[abpos] = '\0';
 632                 if (fbpos >= 0) {
 633                         glob.filt_buf[fbpos++] = '/';
 634                         glob.filt_buf[fbpos] = '\0';
 635                 }
 636                 glob_match(arg, abpos, fbpos);
 637         } else {
 638                 ENSURE_MEMSPACE(glob.argv, char *, glob.maxargs, glob.argc + 1);
 639                 if (!(glob.argv[glob.argc++] = strdup(glob.arg_buf)))
 640                         out_of_memory("glob_match");
 641         }
 642 }
 643
 644 /* This routine performs wild-card expansion of the pathname in "arg".  Any
 645  * daemon-excluded files/dirs will not be matched by the wildcards.  Returns 0
 646  * if a wild-card string is the only returned item (due to matching nothing). */
 647 int glob_expand(const char *arg, char ***argv_p, int *argc_p, int *maxargs_p)
 648 {
 649         int ret, save_argc;
 650         char *s;
 651
 652         if (!arg) {
 653                 if (glob.filt_buf)
 654                         free(glob.filt_buf);
 655                 free(glob.arg_buf);
 656                 memset(&glob, 0, sizeof glob);
 657                 return -1;
 658         }
 659
 660         if (sanitize_paths)
 661                 s = sanitize_path(NULL, arg, "", 0, SP_KEEP_DOT_DIRS);
 662         else {
 663                 s = strdup(arg);
 664                 if (!s)
 665                         out_of_memory("glob_expand");
 666                 clean_fname(s, CFN_KEEP_DOT_DIRS
 667                              | CFN_KEEP_TRAILING_SLASH
 668                              | CFN_COLLAPSE_DOT_DOT_DIRS);
 669         }
 670
 671         ENSURE_MEMSPACE(glob.arg_buf, char, glob.absize, MAXPATHLEN);
 672         *glob.arg_buf = '\0';
 673
 674         glob.argc = save_argc = *argc_p;
 675         glob.argv = *argv_p;
 676         glob.maxargs = *maxargs_p;
 677
 678         ENSURE_MEMSPACE(glob.argv, char *, glob.maxargs, 100);
 679
 680         glob_match(s, 0, -1);
 681
 682         /* The arg didn't match anything, so add the failed arg to the list. */
 683         if (glob.argc == save_argc) {
 684                 ENSURE_MEMSPACE(glob.argv, char *, glob.maxargs, glob.argc + 1);
 685                 glob.argv[glob.argc++] = s;
 686                 ret = 0;
 687         } else {
 688                 free(s);
 689                 ret = 1;
 690         }
 691
 692         *maxargs_p = glob.maxargs;
 693         *argv_p = glob.argv;
 694         *argc_p = glob.argc;
 695
 696         return ret;
 697 }
 698
 699 /* This routine is only used in daemon mode. */
 700 void glob_expand_module(char *base1, char *arg, char ***argv_p, int *argc_p, int *maxargs_p)
 701 {
 702         char *p, *s;
 703         char *base = base1;
 704         int base_len = strlen(base);
 705
 706         if (!arg || !*arg)
 707                 return;
 708
 709         if (strncmp(arg, base, base_len) == 0)
 710                 arg += base_len;
 711
 712         if (!(arg = strdup(arg)))
 713                 out_of_memory("glob_expand_module");
 714
 715         if (asprintf(&base," %s/", base1) <= 0)
 716                 out_of_memory("glob_expand_module");
 717         base_len++;
 718
 719         for (s = arg; *s; s = p + base_len) {
 720                 if ((p = strstr(s, base)) != NULL)
 721                         *p = '\0'; /* split it at this point */
 722                 glob_expand(s, argv_p, argc_p, maxargs_p);
 723                 if (!p)
 724                         break;
 725         }
 726
 727         free(arg);
 728         free(base);
 729 }
 730
 731 /**
 732  * Convert a string to lower case
 733  **/
 734 void strlower(char *s)
 735 {
 736         while (*s) {
 737                 if (isUpper(s))
 738                         *s = toLower(s);
 739                 s++;
 740         }
 741 }
 742
 743 /* Join strings p1 & p2 into "dest" with a guaranteed '/' between them.  (If
 744  * p1 ends with a '/', no extra '/' is inserted.)  Returns the length of both
 745  * strings + 1 (if '/' was inserted), regardless of whether the null-terminated
 746  * string fits into destsize. */
 747 size_t pathjoin(char *dest, size_t destsize, const char *p1, const char *p2)
 748 {
 749         size_t len = strlcpy(dest, p1, destsize);
 750         if (len < destsize - 1) {
 751                 if (!len || dest[len-1] != '/')
 752                         dest[len++] = '/';
 753                 if (len < destsize - 1)
 754                         len += strlcpy(dest + len, p2, destsize - len);
 755                 else {
 756                         dest[len] = '\0';
 757                         len += strlen(p2);
 758                 }
 759         }
 760         else
 761                 len += strlen(p2) + 1; /* Assume we'd insert a '/'. */
 762         return len;
 763 }
 764
 765 /* Join any number of strings together, putting them in "dest".  The return
 766  * value is the length of all the strings, regardless of whether the null-
 767  * terminated whole fits in destsize.  Your list of string pointers must end
 768  * with a NULL to indicate the end of the list. */
 769 size_t stringjoin(char *dest, size_t destsize, ...)
 770 {
 771         va_list ap;
 772         size_t len, ret = 0;
 773         const char *src;
 774
 775         va_start(ap, destsize);
 776         while (1) {
 777                 if (!(src = va_arg(ap, const char *)))
 778                         break;
 779                 len = strlen(src);
 780                 ret += len;
 781                 if (destsize > 1) {
 782                         if (len >= destsize)
 783                                 len = destsize - 1;
 784                         memcpy(dest, src, len);
 785                         destsize -= len;
 786                         dest += len;
 787                 }
 788         }
 789         *dest = '\0';
 790         va_end(ap);
 791
 792         return ret;
 793 }
 794
 795 int count_dir_elements(const char *p)
 796 {
 797         int cnt = 0, new_component = 1;
 798         while (*p) {
 799                 if (*p++ == '/')
 800                         new_component = (*p != '.' || (p[1] != '/' && p[1] != '\0'));
 801                 else if (new_component) {
 802                         new_component = 0;
 803                         cnt++;
 804                 }
 805         }
 806         return cnt;
 807 }
 808
 809 /* Turns multiple adjacent slashes into a single slash, drops all leading or
 810  * interior "." elements unless CFN_KEEP_DOT_DIRS is flagged.  Will also drop
 811  * a trailing '.' after a '/' if CFN_DROP_TRAILING_DOT_DIR is flagged, removes
 812  * a trailing slash (perhaps after removing the aforementioned dot) unless
 813  * CFN_KEEP_TRAILING_SLASH is flagged, and will also collapse ".." elements
 814  * (except at the start) if CFN_COLLAPSE_DOT_DOT_DIRS is flagged.  If the
 815  * resulting name would be empty, returns ".". */
 816 unsigned int clean_fname(char *name, int flags)
 817 {
 818         char *limit = name - 1, *t = name, *f = name;
 819         int anchored;
 820
 821         if (!name)
 822                 return 0;
 823
 824         if ((anchored = *f == '/') != 0)
 825                 *t++ = *f++;
 826         else if (flags & CFN_KEEP_DOT_DIRS && *f == '.' && f[1] == '/') {
 827                 *t++ = *f++;
 828                 *t++ = *f++;
 829         }
 830         while (*f) {
 831                 /* discard extra slashes */
 832                 if (*f == '/') {
 833                         f++;
 834                         continue;
 835                 }
 836                 if (*f == '.') {
 837                         /* discard interior "." dirs */
 838                         if (f[1] == '/' && !(flags & CFN_KEEP_DOT_DIRS)) {
 839                                 f += 2;
 840                                 continue;
 841                         }
 842                         if (f[1] == '\0' && flags & CFN_DROP_TRAILING_DOT_DIR)
 843                                 break;
 844                         /* collapse ".." dirs */
 845                         if (flags & CFN_COLLAPSE_DOT_DOT_DIRS
 846                          && f[1] == '.' && (f[2] == '/' || !f[2])) {
 847                                 char *s = t - 1;
 848                                 if (s == name && anchored) {
 849                                         f += 2;
 850                                         continue;
 851                                 }
 852                                 while (s > limit && *--s != '/') {}
 853                                 if (s != t - 1 && (s < name || *s == '/')) {
 854                                         t = s + 1;
 855                                         f += 2;
 856                                         continue;
 857                                 }
 858                                 limit = t + 2;
 859                         }
 860                 }
 861                 while (*f && (*t++ = *f++) != '/') {}
 862         }
 863
 864         if (t > name+anchored && t[-1] == '/' && !(flags & CFN_KEEP_TRAILING_SLASH))
 865                 t--;
 866         if (t == name)
 867                 *t++ = '.';
 868         *t = '\0';
 869
 870         return t - name;
 871 }
 872
 873 /* Make path appear as if a chroot had occurred.  This handles a leading
 874  * "/" (either removing it or expanding it) and any leading or embedded
 875  * ".." components that attempt to escape past the module's top dir.
 876  *
 877  * If dest is NULL, a buffer is allocated to hold the result.  It is legal
 878  * to call with the dest and the path (p) pointing to the same buffer, but
 879  * rootdir will be ignored to avoid expansion of the string.
 880  *
 881  * The rootdir string contains a value to use in place of a leading slash.
 882  * Specify NULL to get the default of "module_dir".
 883  *
 884  * The depth var is a count of how many '..'s to allow at the start of the
 885  * path.
 886  *
 887  * We also clean the path in a manner similar to clean_fname() but with a
 888  * few differences:
 889  *
 890  * Turns multiple adjacent slashes into a single slash, gets rid of "." dir
 891  * elements (INCLUDING a trailing dot dir), PRESERVES a trailing slash, and
 892  * ALWAYS collapses ".." elements (except for those at the start of the
 893  * string up to "depth" deep).  If the resulting name would be empty,
 894  * change it into a ".". */
 895 char *sanitize_path(char *dest, const char *p, const char *rootdir, int depth,
 896                     int flags)
 897 {
 898         char *start, *sanp;
 899         int rlen = 0, drop_dot_dirs = !relative_paths || !(flags & SP_KEEP_DOT_DIRS);
 900
 901         if (dest != p) {
 902                 int plen = strlen(p);
 903                 if (*p == '/') {
 904                         if (!rootdir)
 905                                 rootdir = module_dir;
 906                         rlen = strlen(rootdir);
 907                         depth = 0;
 908                         p++;
 909                 }
 910                 if (dest) {
 911                         if (rlen + plen + 1 >= MAXPATHLEN)
 912                                 return NULL;
 913                 } else if (!(dest = new_array(char, rlen + plen + 1)))
 914                         out_of_memory("sanitize_path");
 915                 if (rlen) {
 916                         memcpy(dest, rootdir, rlen);
 917                         if (rlen > 1)
 918                                 dest[rlen++] = '/';
 919                 }
 920         }
 921
 922         if (drop_dot_dirs) {
 923                 while (*p == '.' && p[1] == '/')
 924                         p += 2;
 925         }
 926
 927         start = sanp = dest + rlen;
 928         /* This loop iterates once per filename component in p, pointing at
 929          * the start of the name (past any prior slash) for each iteration. */
 930         while (*p) {
 931                 /* discard leading or extra slashes */
 932                 if (*p == '/') {
 933                         p++;
 934                         continue;
 935                 }
 936                 if (drop_dot_dirs) {
 937                         if (*p == '.' && (p[1] == '/' || p[1] == '\0')) {
 938                                 /* skip "." component */
 939                                 p++;
 940                                 continue;
 941                         }
 942                 }
 943                 if (*p == '.' && p[1] == '.' && (p[2] == '/' || p[2] == '\0')) {
 944                         /* ".." component followed by slash or end */
 945                         if (depth <= 0 || sanp != start) {
 946                                 p += 2;
 947                                 if (sanp != start) {
 948                                         /* back up sanp one level */
 949                                         --sanp; /* now pointing at slash */
 950                                         while (sanp > start && sanp[-1] != '/')
 951                                                 sanp--;
 952                                 }
 953                                 continue;
 954                         }
 955                         /* allow depth levels of .. at the beginning */
 956                         depth--;
 957                         /* move the virtual beginning to leave the .. alone */
 958                         start = sanp + 3;
 959                 }
 960                 /* copy one component through next slash */
 961                 while (*p && (*sanp++ = *p++) != '/') {}
 962         }
 963         if (sanp == dest) {
 964                 /* ended up with nothing, so put in "." component */
 965                 *sanp++ = '.';
 966         }
 967         *sanp = '\0';
 968
 969         return dest;
 970 }
 971
 972 /* Like chdir(), but it keeps track of the current directory (in the
 973  * global "curr_dir"), and ensures that the path size doesn't overflow.
 974  * Also cleans the path using the clean_fname() function. */
 975 int change_dir(const char *dir, int set_path_only)
 976 {
 977         static int initialised;
 978         unsigned int len;
 979
 980         if (!initialised) {
 981                 initialised = 1;
 982                 getcwd(curr_dir, sizeof curr_dir - 1);
 983                 curr_dir_len = strlen(curr_dir);
 984         }
 985
 986         if (!dir)       /* this call was probably just to initialize */
 987                 return 0;
 988
 989         len = strlen(dir);
 990         if (len == 1 && *dir == '.')
 991                 return 1;
 992
 993         if (*dir == '/') {
 994                 if (len >= sizeof curr_dir) {
 995                         errno = ENAMETOOLONG;
 996                         return 0;
 997                 }
 998                 if (!set_path_only && chdir(dir))
 999                         return 0;
1000                 memcpy(curr_dir, dir, len + 1);
1001         } else {
1002                 if (curr_dir_len + 1 + len >= sizeof curr_dir) {
1003                         errno = ENAMETOOLONG;
1004                         return 0;
1005                 }
1006                 curr_dir[curr_dir_len] = '/';
1007                 memcpy(curr_dir + curr_dir_len + 1, dir, len + 1);
1008
1009                 if (!set_path_only && chdir(curr_dir)) {
1010                         curr_dir[curr_dir_len] = '\0';
1011                         return 0;
1012                 }
1013         }
1014
1015         curr_dir_len = clean_fname(curr_dir, CFN_COLLAPSE_DOT_DOT_DIRS);
1016         if (sanitize_paths) {
1017                 if (module_dirlen > curr_dir_len)
1018                         module_dirlen = curr_dir_len;
1019                 curr_dir_depth = count_dir_elements(curr_dir + module_dirlen);
1020         }
1021
1022         if (verbose >= 5 && !set_path_only)
1023                 rprintf(FINFO, "[%s] change_dir(%s)\n", who_am_i(), curr_dir);
1024
1025         return 1;
1026 }
1027
1028 /**
1029  * Return a quoted string with the full pathname of the indicated filename.
1030  * The string " (in MODNAME)" may also be appended.  The returned pointer
1031  * remains valid until the next time full_fname() is called.
1032  **/
1033 char *full_fname(const char *fn)
1034 {
1035         static char *result = NULL;
1036         char *m1, *m2, *m3;
1037         char *p1, *p2;
1038
1039         if (result)
1040                 free(result);
1041
1042         if (*fn == '/')
1043                 p1 = p2 = "";
1044         else {
1045                 p1 = curr_dir + module_dirlen;
1046                 for (p2 = p1; *p2 == '/'; p2++) {}
1047                 if (*p2)
1048                         p2 = "/";
1049         }
1050         if (module_id >= 0) {
1051                 m1 = " (in ";
1052                 m2 = lp_name(module_id);
1053                 m3 = ")";
1054         } else
1055                 m1 = m2 = m3 = "";
1056
1057         if (asprintf(&result, "\"%s%s%s\"%s%s%s", p1, p2, fn, m1, m2, m3) <= 0)
1058                 out_of_memory("full_fname");
1059
1060         return result;
1061 }
1062
1063 static char partial_fname[MAXPATHLEN];
1064
1065 char *partial_dir_fname(const char *fname)
1066 {
1067         char *t = partial_fname;
1068         int sz = sizeof partial_fname;
1069         const char *fn;
1070
1071         if ((fn = strrchr(fname, '/')) != NULL) {
1072                 fn++;
1073                 if (*partial_dir != '/') {
1074                         int len = fn - fname;
1075                         strncpy(t, fname, len); /* safe */
1076                         t += len;
1077                         sz -= len;
1078                 }
1079         } else
1080                 fn = fname;
1081         if ((int)pathjoin(t, sz, partial_dir, fn) >= sz)
1082                 return NULL;
1083         if (daemon_filter_list.head) {
1084                 t = strrchr(partial_fname, '/');
1085                 *t = '\0';
1086                 if (check_filter(&daemon_filter_list, FLOG, partial_fname, 1) < 0)
1087                         return NULL;
1088                 *t = '/';
1089                 if (check_filter(&daemon_filter_list, FLOG, partial_fname, 0) < 0)
1090                         return NULL;
1091         }
1092
1093         return partial_fname;
1094 }
1095
1096 /* If no --partial-dir option was specified, we don't need to do anything
1097  * (the partial-dir is essentially '.'), so just return success. */
1098 int handle_partial_dir(const char *fname, int create)
1099 {
1100         char *fn, *dir;
1101
1102         if (fname != partial_fname)
1103                 return 1;
1104         if (!create && *partial_dir == '/')
1105                 return 1;
1106         if (!(fn = strrchr(partial_fname, '/')))
1107                 return 1;
1108
1109         *fn = '\0';
1110         dir = partial_fname;
1111         if (create) {
1112                 STRUCT_STAT st;
1113                 int statret = do_lstat(dir, &st);
1114                 if (statret == 0 && !S_ISDIR(st.st_mode)) {
1115                         if (do_unlink(dir) < 0)
1116                                 return 0;
1117                         statret = -1;
1118                 }
1119                 if (statret < 0 && do_mkdir(dir, 0700) < 0)
1120                         return 0;
1121         } else
1122                 do_rmdir(dir);
1123         *fn = '/';
1124
1125         return 1;
1126 }
1127
1128 /**
1129  * Determine if a symlink points outside the current directory tree.
1130  * This is considered "unsafe" because e.g. when mirroring somebody
1131  * else's machine it might allow them to establish a symlink to
1132  * /etc/passwd, and then read it through a web server.
1133  *
1134  * Null symlinks and absolute symlinks are always unsafe.
1135  *
1136  * Basically here we are concerned with symlinks whose target contains
1137  * "..", because this might cause us to walk back up out of the
1138  * transferred directory.  We are not allowed to go back up and
1139  * reenter.
1140  *
1141  * @param dest Target of the symlink in question.
1142  *
1143  * @param src Top source directory currently applicable.  Basically this
1144  * is the first parameter to rsync in a simple invocation, but it's
1145  * modified by flist.c in slightly complex ways.
1146  *
1147  * @retval True if unsafe
1148  * @retval False is unsafe
1149  *
1150  * @sa t_unsafe.c
1151  **/
1152 int unsafe_symlink(const char *dest, const char *src)
1153 {
1154         const char *name, *slash;
1155         int depth = 0;
1156
1157         /* all absolute and null symlinks are unsafe */
1158         if (!dest || !*dest || *dest == '/')
1159                 return 1;
1160
1161         /* find out what our safety margin is */
1162         for (name = src; (slash = strchr(name, '/')) != 0; name = slash+1) {
1163                 if (strncmp(name, "../", 3) == 0) {
1164                         depth = 0;
1165                 } else if (strncmp(name, "./", 2) == 0) {
1166                         /* nothing */
1167                 } else {
1168                         depth++;
1169                 }
1170         }
1171         if (strcmp(name, "..") == 0)
1172                 depth = 0;
1173
1174         for (name = dest; (slash = strchr(name, '/')) != 0; name = slash+1) {
1175                 if (strncmp(name, "../", 3) == 0) {
1176                         /* if at any point we go outside the current directory
1177                            then stop - it is unsafe */
1178                         if (--depth < 0)
1179                                 return 1;
1180                 } else if (strncmp(name, "./", 2) == 0) {
1181                         /* nothing */
1182                 } else {
1183                         depth++;
1184                 }
1185         }
1186         if (strcmp(name, "..") == 0)
1187                 depth--;
1188
1189         return (depth < 0);
1190 }
1191
1192 /* Return the int64 number as a string.  If the --human-readable option was
1193  * specified, we may output the number in K, M, or G units.  We can return
1194  * up to 4 buffers at a time. */
1195 char *human_num(int64 num)
1196 {
1197         static char bufs[4][128]; /* more than enough room */
1198         static unsigned int n;
1199         char *s;
1200
1201         n = (n + 1) % (sizeof bufs / sizeof bufs[0]);
1202
1203         if (human_readable) {
1204                 char units = '\0';
1205                 int mult = human_readable == 1 ? 1000 : 1024;
1206                 double dnum = 0;
1207                 if (num > mult*mult*mult) {
1208                         dnum = (double)num / (mult*mult*mult);
1209                         units = 'G';
1210                 } else if (num > mult*mult) {
1211                         dnum = (double)num / (mult*mult);
1212                         units = 'M';
1213                 } else if (num > mult) {
1214                         dnum = (double)num / mult;
1215                         units = 'K';
1216                 }
1217                 if (units) {
1218                         snprintf(bufs[n], sizeof bufs[0], "%.2f%c", dnum, units);
1219                         return bufs[n];
1220                 }
1221         }
1222
1223         s = bufs[n] + sizeof bufs[0] - 1;
1224         *s = '\0';
1225
1226         if (!num)
1227                 *--s = '0';
1228         while (num) {
1229                 *--s = (char)(num % 10) + '0';
1230                 num /= 10;
1231         }
1232         return s;
1233 }
1234
1235 /* Return the double number as a string.  If the --human-readable option was
1236  * specified, we may output the number in K, M, or G units.  We use a buffer
1237  * from human_num() to return our result. */
1238 char *human_dnum(double dnum, int decimal_digits)
1239 {
1240         char *buf = human_num(dnum);
1241         int len = strlen(buf);
1242         if (isDigit(buf + len - 1)) {
1243                 /* There's extra room in buf prior to the start of the num. */
1244                 buf -= decimal_digits + 1;
1245                 snprintf(buf, len + decimal_digits + 2, "%.*f", decimal_digits, dnum);
1246         }
1247         return buf;
1248 }
1249
1250 /* Return the date and time as a string.  Some callers tweak returned buf. */
1251 char *timestring(time_t t)
1252 {
1253         static char TimeBuf[200];
1254         struct tm *tm = localtime(&t);
1255         char *p;
1256
1257 #ifdef HAVE_STRFTIME
1258         strftime(TimeBuf, sizeof TimeBuf - 1, "%Y/%m/%d %H:%M:%S", tm);
1259 #else
1260         strlcpy(TimeBuf, asctime(tm), sizeof TimeBuf);
1261 #endif
1262
1263         if ((p = strchr(TimeBuf, '\n')) != NULL)
1264                 *p = '\0';
1265
1266         return TimeBuf;
1267 }
1268
1269 /**
1270  * Sleep for a specified number of milliseconds.
1271  *
1272  * Always returns TRUE.  (In the future it might return FALSE if
1273  * interrupted.)
1274  **/
1275 int msleep(int t)
1276 {
1277         int tdiff = 0;
1278         struct timeval tval, t1, t2;
1279
1280         gettimeofday(&t1, NULL);
1281
1282         while (tdiff < t) {
1283                 tval.tv_sec = (t-tdiff)/1000;
1284                 tval.tv_usec = 1000*((t-tdiff)%1000);
1285
1286                 errno = 0;
1287                 select(0,NULL,NULL, NULL, &tval);
1288
1289                 gettimeofday(&t2, NULL);
1290                 tdiff = (t2.tv_sec - t1.tv_sec)*1000 +
1291                         (t2.tv_usec - t1.tv_usec)/1000;
1292         }
1293
1294         return True;
1295 }
1296
1297 /* Determine if two time_t values are equivalent (either exact, or in
1298  * the modification timestamp window established by --modify-window).
1299  *
1300  * @retval 0 if the times should be treated as the same
1301  *
1302  * @retval +1 if the first is later
1303  *
1304  * @retval -1 if the 2nd is later
1305  **/
1306 int cmp_time(time_t file1, time_t file2)
1307 {
1308         if (file2 > file1) {
1309                 if (file2 - file1 <= modify_window)
1310                         return 0;
1311                 return -1;
1312         }
1313         if (file1 - file2 <= modify_window)
1314                 return 0;
1315         return 1;
1316 }
1317
1318
1319 #ifdef __INSURE__XX
1320 #include <dlfcn.h>
1321
1322 /**
1323    This routine is a trick to immediately catch errors when debugging
1324    with insure. A xterm with a gdb is popped up when insure catches
1325    a error. It is Linux specific.
1326 **/
1327 int _Insure_trap_error(int a1, int a2, int a3, int a4, int a5, int a6)
1328 {
1329         static int (*fn)();
1330         int ret;
1331         char *cmd;
1332
1333         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'",
1334                 getpid(), getpid(), getpid());
1335
1336         if (!fn) {
1337                 static void *h;
1338                 h = dlopen("/usr/local/parasoft/insure++lite/lib.linux2/libinsure.so", RTLD_LAZY);
1339                 fn = dlsym(h, "_Insure_trap_error");
1340         }
1341
1342         ret = fn(a1, a2, a3, a4, a5, a6);
1343
1344         system(cmd);
1345
1346         free(cmd);
1347
1348         return ret;
1349 }
1350 #endif
1351
1352 #define MALLOC_MAX 0x40000000
1353
1354 void *_new_array(unsigned long num, unsigned int size, int use_calloc)
1355 {
1356         if (num >= MALLOC_MAX/size)
1357                 return NULL;
1358         return use_calloc ? calloc(num, size) : malloc(num * size);
1359 }
1360
1361 void *_realloc_array(void *ptr, unsigned int size, size_t num)
1362 {
1363         if (num >= MALLOC_MAX/size)
1364                 return NULL;
1365         if (!ptr)
1366                 return malloc(size * num);
1367         return realloc(ptr, size * num);
1368 }
1369
1370 /* Take a filename and filename length and return the most significant
1371  * filename suffix we can find.  This ignores suffixes such as "~",
1372  * ".bak", ".orig", ".~1~", etc. */
1373 const char *find_filename_suffix(const char *fn, int fn_len, int *len_ptr)
1374 {
1375         const char *suf, *s;
1376         BOOL had_tilde;
1377         int s_len;
1378
1379         /* One or more dots at the start aren't a suffix. */
1380         while (fn_len && *fn == '.') fn++, fn_len--;
1381
1382         /* Ignore the ~ in a "foo~" filename. */
1383         if (fn_len > 1 && fn[fn_len-1] == '~')
1384                 fn_len--, had_tilde = True;
1385         else
1386                 had_tilde = False;
1387
1388         /* Assume we don't find an suffix. */
1389         suf = "";
1390         *len_ptr = 0;
1391
1392         /* Find the last significant suffix. */
1393         for (s = fn + fn_len; fn_len > 1; ) {
1394                 while (*--s != '.' && s != fn) {}
1395                 if (s == fn)
1396                         break;
1397                 s_len = fn_len - (s - fn);
1398                 fn_len = s - fn;
1399                 if (s_len == 4) {
1400                         if (strcmp(s+1, "bak") == 0
1401                          || strcmp(s+1, "old") == 0)
1402                                 continue;
1403                 } else if (s_len == 5) {
1404                         if (strcmp(s+1, "orig") == 0)
1405                                 continue;
1406                 } else if (s_len > 2 && had_tilde
1407                     && s[1] == '~' && isDigit(s + 2))
1408                         continue;
1409                 *len_ptr = s_len;
1410                 suf = s;
1411                 if (s_len == 1)
1412                         break;
1413                 /* Determine if the suffix is all digits. */
1414                 for (s++, s_len--; s_len > 0; s++, s_len--) {
1415                         if (!isDigit(s))
1416                                 return suf;
1417                 }
1418                 /* An all-digit suffix may not be that signficant. */
1419                 s = suf;
1420         }
1421
1422         return suf;
1423 }
1424
1425 /* This is an implementation of the Levenshtein distance algorithm.  It
1426  * was implemented to avoid needing a two-dimensional matrix (to save
1427  * memory).  It was also tweaked to try to factor in the ASCII distance
1428  * between changed characters as a minor distance quantity.  The normal
1429  * Levenshtein units of distance (each signifying a single change between
1430  * the two strings) are defined as a "UNIT". */
1431
1432 #define UNIT (1 << 16)
1433
1434 uint32 fuzzy_distance(const char *s1, int len1, const char *s2, int len2)
1435 {
1436         uint32 a[MAXPATHLEN], diag, above, left, diag_inc, above_inc, left_inc;
1437         int32 cost;
1438         int i1, i2;
1439
1440         if (!len1 || !len2) {
1441                 if (!len1) {
1442                         s1 = s2;
1443                         len1 = len2;
1444                 }
1445                 for (i1 = 0, cost = 0; i1 < len1; i1++)
1446                         cost += s1[i1];
1447                 return (int32)len1 * UNIT + cost;
1448         }
1449
1450         for (i2 = 0; i2 < len2; i2++)
1451                 a[i2] = (i2+1) * UNIT;
1452
1453         for (i1 = 0; i1 < len1; i1++) {
1454                 diag = i1 * UNIT;
1455                 above = (i1+1) * UNIT;
1456                 for (i2 = 0; i2 < len2; i2++) {
1457                         left = a[i2];
1458                         if ((cost = *((uchar*)s1+i1) - *((uchar*)s2+i2)) != 0) {
1459                                 if (cost < 0)
1460                                         cost = UNIT - cost;
1461                                 else
1462                                         cost = UNIT + cost;
1463                         }
1464                         diag_inc = diag + cost;
1465                         left_inc = left + UNIT + *((uchar*)s1+i1);
1466                         above_inc = above + UNIT + *((uchar*)s2+i2);
1467                         a[i2] = above = left < above
1468                               ? (left_inc < diag_inc ? left_inc : diag_inc)
1469                               : (above_inc < diag_inc ? above_inc : diag_inc);
1470                         diag = left;
1471                 }
1472         }
1473
1474         return a[len2-1];
1475 }
1476
1477 #define BB_SLOT_SIZE     (16*1024)          /* Desired size in bytes */
1478 #define BB_PER_SLOT_BITS (BB_SLOT_SIZE * 8) /* Number of bits per slot */
1479 #define BB_PER_SLOT_INTS (BB_SLOT_SIZE / 4) /* Number of int32s per slot */
1480
1481 struct bitbag {
1482     uint32 **bits;
1483     int slot_cnt;
1484 };
1485
1486 struct bitbag *bitbag_create(int max_ndx)
1487 {
1488         struct bitbag *bb = new(struct bitbag);
1489         bb->slot_cnt = (max_ndx + BB_PER_SLOT_BITS - 1) / BB_PER_SLOT_BITS;
1490
1491         if (!(bb->bits = (uint32**)calloc(bb->slot_cnt, sizeof (uint32*))))
1492                 out_of_memory("bitbag_create");
1493
1494         return bb;
1495 }
1496
1497 void bitbag_set_bit(struct bitbag *bb, int ndx)
1498 {
1499         int slot = ndx / BB_PER_SLOT_BITS;
1500         ndx %= BB_PER_SLOT_BITS;
1501
1502         if (!bb->bits[slot]) {
1503                 if (!(bb->bits[slot] = (uint32*)calloc(BB_PER_SLOT_INTS, 4)))
1504                         out_of_memory("bitbag_set_bit");
1505         }
1506
1507         bb->bits[slot][ndx/32] |= 1u << (ndx % 32);
1508 }
1509
1510 #if 0 /* not needed yet */
1511 void bitbag_clear_bit(struct bitbag *bb, int ndx)
1512 {
1513         int slot = ndx / BB_PER_SLOT_BITS;
1514         ndx %= BB_PER_SLOT_BITS;
1515
1516         if (!bb->bits[slot])
1517                 return;
1518
1519         bb->bits[slot][ndx/32] &= ~(1u << (ndx % 32));
1520 }
1521
1522 int bitbag_check_bit(struct bitbag *bb, int ndx)
1523 {
1524         int slot = ndx / BB_PER_SLOT_BITS;
1525         ndx %= BB_PER_SLOT_BITS;
1526
1527         if (!bb->bits[slot])
1528                 return 0;
1529
1530         return bb->bits[slot][ndx/32] & (1u << (ndx % 32)) ? 1 : 0;
1531 }
1532 #endif
1533
1534 /* Call this with -1 to start checking from 0.  Returns -1 at the end. */
1535 int bitbag_next_bit(struct bitbag *bb, int after)
1536 {
1537         uint32 bits, mask;
1538         int i, ndx = after + 1;
1539         int slot = ndx / BB_PER_SLOT_BITS;
1540         ndx %= BB_PER_SLOT_BITS;
1541
1542         mask = (1u << (ndx % 32)) - 1;
1543         for (i = ndx / 32; slot < bb->slot_cnt; slot++, i = mask = 0) {
1544                 if (!bb->bits[slot])
1545                         continue;
1546                 for ( ; i < BB_PER_SLOT_INTS; i++, mask = 0) {
1547                         if (!(bits = bb->bits[slot][i] & ~mask))
1548                                 continue;
1549                         /* The xor magic figures out the lowest enabled bit in
1550                          * bits, and the switch quickly computes log2(bit). */
1551                         switch (bits ^ (bits & (bits-1))) {
1552 #define LOG2(n) case 1u << n: return slot*BB_PER_SLOT_BITS + i*32 + n
1553                             LOG2(0);  LOG2(1);  LOG2(2);  LOG2(3);
1554                             LOG2(4);  LOG2(5);  LOG2(6);  LOG2(7);
1555                             LOG2(8);  LOG2(9);  LOG2(10); LOG2(11);
1556                             LOG2(12); LOG2(13); LOG2(14); LOG2(15);
1557                             LOG2(16); LOG2(17); LOG2(18); LOG2(19);
1558                             LOG2(20); LOG2(21); LOG2(22); LOG2(23);
1559                             LOG2(24); LOG2(25); LOG2(26); LOG2(27);
1560                             LOG2(28); LOG2(29); LOG2(30); LOG2(31);
1561                         }
1562                         return -1; /* impossible... */
1563                 }
1564         }
1565
1566         return -1;
1567 }
1568
1569 void *expand_item_list(item_list *lp, size_t item_size,
1570                        const char *desc, int incr)
1571 {
1572         /* First time through, 0 <= 0, so list is expanded. */
1573         if (lp->malloced <= lp->count) {
1574                 void *new_ptr;
1575                 size_t new_size = lp->malloced;
1576                 if (incr < 0)
1577                         new_size += -incr; /* increase slowly */
1578                 else if (new_size < (size_t)incr)
1579                         new_size += incr;
1580                 else
1581                         new_size *= 2;
1582                 if (new_size < lp->malloced)
1583                         overflow_exit("expand_item_list");
1584                 /* Using _realloc_array() lets us pass the size, not a type. */
1585                 new_ptr = _realloc_array(lp->items, item_size, new_size);
1586                 if (verbose >= 4) {
1587                         rprintf(FINFO, "[%s] expand %s to %.0f bytes, did%s move\n",
1588                                 who_am_i(), desc, (double)new_size * item_size,
1589                                 new_ptr == lp->items ? " not" : "");
1590                 }
1591                 if (!new_ptr)
1592                         out_of_memory("expand_item_list");
1593
1594                 lp->items = new_ptr;
1595                 lp->malloced = new_size;
1596         }
1597         return (char*)lp->items + (lp->count++ * item_size);
1598 }