| 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) 2003-2007 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 | |
| 25 | extern int verbose; |
| 26 | extern int dry_run; |
| 27 | extern int module_id; |
| 28 | extern int modify_window; |
| 29 | extern int relative_paths; |
| 30 | extern int human_readable; |
| 31 | extern char *module_dir; |
| 32 | extern unsigned int module_dirlen; |
| 33 | extern mode_t orig_umask; |
| 34 | extern char *partial_dir; |
| 35 | extern struct filter_list_struct server_filter_list; |
| 36 | |
| 37 | int sanitize_paths = 0; |
| 38 | |
| 39 | char curr_dir[MAXPATHLEN]; |
| 40 | unsigned int curr_dir_len; |
| 41 | int curr_dir_depth; /* This is only set for a sanitizing daemon. */ |
| 42 | |
| 43 | /* Set a fd into nonblocking mode. */ |
| 44 | void set_nonblocking(int fd) |
| 45 | { |
| 46 | int val; |
| 47 | |
| 48 | if ((val = fcntl(fd, F_GETFL)) == -1) |
| 49 | return; |
| 50 | if (!(val & NONBLOCK_FLAG)) { |
| 51 | val |= NONBLOCK_FLAG; |
| 52 | fcntl(fd, F_SETFL, val); |
| 53 | } |
| 54 | } |
| 55 | |
| 56 | /* Set a fd into blocking mode. */ |
| 57 | void set_blocking(int fd) |
| 58 | { |
| 59 | int val; |
| 60 | |
| 61 | if ((val = fcntl(fd, F_GETFL)) == -1) |
| 62 | return; |
| 63 | if (val & NONBLOCK_FLAG) { |
| 64 | val &= ~NONBLOCK_FLAG; |
| 65 | fcntl(fd, F_SETFL, val); |
| 66 | } |
| 67 | } |
| 68 | |
| 69 | /** |
| 70 | * Create a file descriptor pair - like pipe() but use socketpair if |
| 71 | * possible (because of blocking issues on pipes). |
| 72 | * |
| 73 | * Always set non-blocking. |
| 74 | */ |
| 75 | int fd_pair(int fd[2]) |
| 76 | { |
| 77 | int ret; |
| 78 | |
| 79 | #ifdef HAVE_SOCKETPAIR |
| 80 | ret = socketpair(AF_UNIX, SOCK_STREAM, 0, fd); |
| 81 | #else |
| 82 | ret = pipe(fd); |
| 83 | #endif |
| 84 | |
| 85 | if (ret == 0) { |
| 86 | set_nonblocking(fd[0]); |
| 87 | set_nonblocking(fd[1]); |
| 88 | } |
| 89 | |
| 90 | return ret; |
| 91 | } |
| 92 | |
| 93 | void print_child_argv(char **cmd) |
| 94 | { |
| 95 | rprintf(FCLIENT, "opening connection using "); |
| 96 | for (; *cmd; cmd++) { |
| 97 | /* Look for characters that ought to be quoted. This |
| 98 | * is not a great quoting algorithm, but it's |
| 99 | * sufficient for a log message. */ |
| 100 | if (strspn(*cmd, "abcdefghijklmnopqrstuvwxyz" |
| 101 | "ABCDEFGHIJKLMNOPQRSTUVWXYZ" |
| 102 | "0123456789" |
| 103 | ",.-_=+@/") != strlen(*cmd)) { |
| 104 | rprintf(FCLIENT, "\"%s\" ", *cmd); |
| 105 | } else { |
| 106 | rprintf(FCLIENT, "%s ", *cmd); |
| 107 | } |
| 108 | } |
| 109 | rprintf(FCLIENT, "\n"); |
| 110 | } |
| 111 | |
| 112 | NORETURN void out_of_memory(const char *str) |
| 113 | { |
| 114 | rprintf(FERROR, "ERROR: out of memory in %s [%s]\n", str, who_am_i()); |
| 115 | exit_cleanup(RERR_MALLOC); |
| 116 | } |
| 117 | |
| 118 | NORETURN void overflow_exit(const char *str) |
| 119 | { |
| 120 | rprintf(FERROR, "ERROR: buffer overflow in %s [%s]\n", str, who_am_i()); |
| 121 | exit_cleanup(RERR_MALLOC); |
| 122 | } |
| 123 | |
| 124 | int set_modtime(const char *fname, time_t modtime, mode_t mode) |
| 125 | { |
| 126 | #if !defined HAVE_LUTIMES || !defined HAVE_UTIMES |
| 127 | if (S_ISLNK(mode)) |
| 128 | return 1; |
| 129 | #endif |
| 130 | |
| 131 | if (verbose > 2) { |
| 132 | rprintf(FINFO, "set modtime of %s to (%ld) %s", |
| 133 | fname, (long)modtime, |
| 134 | asctime(localtime(&modtime))); |
| 135 | } |
| 136 | |
| 137 | if (dry_run) |
| 138 | return 0; |
| 139 | |
| 140 | { |
| 141 | #ifdef HAVE_UTIMES |
| 142 | struct timeval t[2]; |
| 143 | t[0].tv_sec = time(NULL); |
| 144 | t[0].tv_usec = 0; |
| 145 | t[1].tv_sec = modtime; |
| 146 | t[1].tv_usec = 0; |
| 147 | # ifdef HAVE_LUTIMES |
| 148 | if (S_ISLNK(mode)) { |
| 149 | lutimes(fname, t); |
| 150 | return 0; /* ignore errors */ |
| 151 | } |
| 152 | # endif |
| 153 | return utimes(fname, t); |
| 154 | #elif defined HAVE_UTIMBUF |
| 155 | struct utimbuf tbuf; |
| 156 | tbuf.actime = time(NULL); |
| 157 | tbuf.modtime = modtime; |
| 158 | return utime(fname,&tbuf); |
| 159 | #elif defined HAVE_UTIME |
| 160 | time_t t[2]; |
| 161 | t[0] = time(NULL); |
| 162 | t[1] = modtime; |
| 163 | return utime(fname,t); |
| 164 | #else |
| 165 | #error No file-time-modification routine found! |
| 166 | #endif |
| 167 | } |
| 168 | } |
| 169 | |
| 170 | /* This creates a new directory with default permissions. Since there |
| 171 | * might be some directory-default permissions affecting this, we can't |
| 172 | * force the permissions directly using the original umask and mkdir(). */ |
| 173 | int mkdir_defmode(char *fname) |
| 174 | { |
| 175 | int ret; |
| 176 | |
| 177 | umask(orig_umask); |
| 178 | ret = do_mkdir(fname, ACCESSPERMS); |
| 179 | umask(0); |
| 180 | |
| 181 | return ret; |
| 182 | } |
| 183 | |
| 184 | /* Create any necessary directories in fname. Any missing directories are |
| 185 | * created with default permissions. */ |
| 186 | int create_directory_path(char *fname) |
| 187 | { |
| 188 | char *p; |
| 189 | int ret = 0; |
| 190 | |
| 191 | while (*fname == '/') |
| 192 | fname++; |
| 193 | while (strncmp(fname, "./", 2) == 0) |
| 194 | fname += 2; |
| 195 | |
| 196 | umask(orig_umask); |
| 197 | p = fname; |
| 198 | while ((p = strchr(p,'/')) != NULL) { |
| 199 | *p = '\0'; |
| 200 | if (do_mkdir(fname, ACCESSPERMS) < 0 && errno != EEXIST) |
| 201 | ret = -1; |
| 202 | *p++ = '/'; |
| 203 | } |
| 204 | umask(0); |
| 205 | |
| 206 | return ret; |
| 207 | } |
| 208 | |
| 209 | /** |
| 210 | * Write @p len bytes at @p ptr to descriptor @p desc, retrying if |
| 211 | * interrupted. |
| 212 | * |
| 213 | * @retval len upon success |
| 214 | * |
| 215 | * @retval <0 write's (negative) error code |
| 216 | * |
| 217 | * Derived from GNU C's cccp.c. |
| 218 | */ |
| 219 | int full_write(int desc, const char *ptr, size_t len) |
| 220 | { |
| 221 | int total_written; |
| 222 | |
| 223 | total_written = 0; |
| 224 | while (len > 0) { |
| 225 | int written = write(desc, ptr, len); |
| 226 | if (written < 0) { |
| 227 | if (errno == EINTR) |
| 228 | continue; |
| 229 | return written; |
| 230 | } |
| 231 | total_written += written; |
| 232 | ptr += written; |
| 233 | len -= written; |
| 234 | } |
| 235 | return total_written; |
| 236 | } |
| 237 | |
| 238 | /** |
| 239 | * Read @p len bytes at @p ptr from descriptor @p desc, retrying if |
| 240 | * interrupted. |
| 241 | * |
| 242 | * @retval >0 the actual number of bytes read |
| 243 | * |
| 244 | * @retval 0 for EOF |
| 245 | * |
| 246 | * @retval <0 for an error. |
| 247 | * |
| 248 | * Derived from GNU C's cccp.c. */ |
| 249 | static int safe_read(int desc, char *ptr, size_t len) |
| 250 | { |
| 251 | int n_chars; |
| 252 | |
| 253 | if (len == 0) |
| 254 | return len; |
| 255 | |
| 256 | do { |
| 257 | n_chars = read(desc, ptr, len); |
| 258 | } while (n_chars < 0 && errno == EINTR); |
| 259 | |
| 260 | return n_chars; |
| 261 | } |
| 262 | |
| 263 | /** Copy a file. |
| 264 | * |
| 265 | * This is used in conjunction with the --temp-dir, --backup, and |
| 266 | * --copy-dest options. */ |
| 267 | int copy_file(const char *source, const char *dest, mode_t mode) |
| 268 | { |
| 269 | int ifd; |
| 270 | int ofd; |
| 271 | char buf[1024 * 8]; |
| 272 | int len; /* Number of bytes read into `buf'. */ |
| 273 | |
| 274 | ifd = do_open(source, O_RDONLY, 0); |
| 275 | if (ifd == -1) { |
| 276 | rsyserr(FERROR, errno, "open %s", full_fname(source)); |
| 277 | return -1; |
| 278 | } |
| 279 | |
| 280 | if (robust_unlink(dest) && errno != ENOENT) { |
| 281 | rsyserr(FERROR, errno, "unlink %s", full_fname(dest)); |
| 282 | return -1; |
| 283 | } |
| 284 | |
| 285 | ofd = do_open(dest, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, mode); |
| 286 | if (ofd == -1) { |
| 287 | rsyserr(FERROR, errno, "open %s", full_fname(dest)); |
| 288 | close(ifd); |
| 289 | return -1; |
| 290 | } |
| 291 | |
| 292 | while ((len = safe_read(ifd, buf, sizeof buf)) > 0) { |
| 293 | if (full_write(ofd, buf, len) < 0) { |
| 294 | rsyserr(FERROR, errno, "write %s", full_fname(dest)); |
| 295 | close(ifd); |
| 296 | close(ofd); |
| 297 | return -1; |
| 298 | } |
| 299 | } |
| 300 | |
| 301 | if (len < 0) { |
| 302 | rsyserr(FERROR, errno, "read %s", full_fname(source)); |
| 303 | close(ifd); |
| 304 | close(ofd); |
| 305 | return -1; |
| 306 | } |
| 307 | |
| 308 | if (close(ifd) < 0) { |
| 309 | rsyserr(FINFO, errno, "close failed on %s", |
| 310 | full_fname(source)); |
| 311 | } |
| 312 | |
| 313 | if (close(ofd) < 0) { |
| 314 | rsyserr(FERROR, errno, "close failed on %s", |
| 315 | full_fname(dest)); |
| 316 | return -1; |
| 317 | } |
| 318 | |
| 319 | return 0; |
| 320 | } |
| 321 | |
| 322 | /* MAX_RENAMES should be 10**MAX_RENAMES_DIGITS */ |
| 323 | #define MAX_RENAMES_DIGITS 3 |
| 324 | #define MAX_RENAMES 1000 |
| 325 | |
| 326 | /** |
| 327 | * Robust unlink: some OS'es (HPUX) refuse to unlink busy files, so |
| 328 | * rename to <path>/.rsyncNNN instead. |
| 329 | * |
| 330 | * Note that successive rsync runs will shuffle the filenames around a |
| 331 | * bit as long as the file is still busy; this is because this function |
| 332 | * does not know if the unlink call is due to a new file coming in, or |
| 333 | * --delete trying to remove old .rsyncNNN files, hence it renames it |
| 334 | * each time. |
| 335 | **/ |
| 336 | int robust_unlink(const char *fname) |
| 337 | { |
| 338 | #ifndef ETXTBSY |
| 339 | return do_unlink(fname); |
| 340 | #else |
| 341 | static int counter = 1; |
| 342 | int rc, pos, start; |
| 343 | char path[MAXPATHLEN]; |
| 344 | |
| 345 | rc = do_unlink(fname); |
| 346 | if (rc == 0 || errno != ETXTBSY) |
| 347 | return rc; |
| 348 | |
| 349 | if ((pos = strlcpy(path, fname, MAXPATHLEN)) >= MAXPATHLEN) |
| 350 | pos = MAXPATHLEN - 1; |
| 351 | |
| 352 | while (pos > 0 && path[pos-1] != '/') |
| 353 | pos--; |
| 354 | pos += strlcpy(path+pos, ".rsync", MAXPATHLEN-pos); |
| 355 | |
| 356 | if (pos > (MAXPATHLEN-MAX_RENAMES_DIGITS-1)) { |
| 357 | errno = ETXTBSY; |
| 358 | return -1; |
| 359 | } |
| 360 | |
| 361 | /* start where the last one left off to reduce chance of clashes */ |
| 362 | start = counter; |
| 363 | do { |
| 364 | snprintf(&path[pos], MAX_RENAMES_DIGITS+1, "%03d", counter); |
| 365 | if (++counter >= MAX_RENAMES) |
| 366 | counter = 1; |
| 367 | } while ((rc = access(path, 0)) == 0 && counter != start); |
| 368 | |
| 369 | if (verbose > 0) { |
| 370 | rprintf(FINFO,"renaming %s to %s because of text busy\n", |
| 371 | fname, path); |
| 372 | } |
| 373 | |
| 374 | /* maybe we should return rename()'s exit status? Nah. */ |
| 375 | if (do_rename(fname, path) != 0) { |
| 376 | errno = ETXTBSY; |
| 377 | return -1; |
| 378 | } |
| 379 | return 0; |
| 380 | #endif |
| 381 | } |
| 382 | |
| 383 | /* Returns 0 on successful rename, 1 if we successfully copied the file |
| 384 | * across filesystems, -2 if copy_file() failed, and -1 on other errors. |
| 385 | * If partialptr is not NULL and we need to do a copy, copy the file into |
| 386 | * the active partial-dir instead of over the destination file. */ |
| 387 | int robust_rename(const char *from, const char *to, const char *partialptr, |
| 388 | int mode) |
| 389 | { |
| 390 | int tries = 4; |
| 391 | |
| 392 | while (tries--) { |
| 393 | if (do_rename(from, to) == 0) |
| 394 | return 0; |
| 395 | |
| 396 | switch (errno) { |
| 397 | #ifdef ETXTBSY |
| 398 | case ETXTBSY: |
| 399 | if (robust_unlink(to) != 0) |
| 400 | return -1; |
| 401 | break; |
| 402 | #endif |
| 403 | case EXDEV: |
| 404 | if (partialptr) { |
| 405 | if (!handle_partial_dir(partialptr,PDIR_CREATE)) |
| 406 | return -1; |
| 407 | to = partialptr; |
| 408 | } |
| 409 | if (copy_file(from, to, mode) != 0) |
| 410 | return -2; |
| 411 | do_unlink(from); |
| 412 | return 1; |
| 413 | default: |
| 414 | return -1; |
| 415 | } |
| 416 | } |
| 417 | return -1; |
| 418 | } |
| 419 | |
| 420 | static pid_t all_pids[10]; |
| 421 | static int num_pids; |
| 422 | |
| 423 | /** Fork and record the pid of the child. **/ |
| 424 | pid_t do_fork(void) |
| 425 | { |
| 426 | pid_t newpid = fork(); |
| 427 | |
| 428 | if (newpid != 0 && newpid != -1) { |
| 429 | all_pids[num_pids++] = newpid; |
| 430 | } |
| 431 | return newpid; |
| 432 | } |
| 433 | |
| 434 | /** |
| 435 | * Kill all children. |
| 436 | * |
| 437 | * @todo It would be kind of nice to make sure that they are actually |
| 438 | * all our children before we kill them, because their pids may have |
| 439 | * been recycled by some other process. Perhaps when we wait for a |
| 440 | * child, we should remove it from this array. Alternatively we could |
| 441 | * perhaps use process groups, but I think that would not work on |
| 442 | * ancient Unix versions that don't support them. |
| 443 | **/ |
| 444 | void kill_all(int sig) |
| 445 | { |
| 446 | int i; |
| 447 | |
| 448 | for (i = 0; i < num_pids; i++) { |
| 449 | /* Let's just be a little careful where we |
| 450 | * point that gun, hey? See kill(2) for the |
| 451 | * magic caused by negative values. */ |
| 452 | pid_t p = all_pids[i]; |
| 453 | |
| 454 | if (p == getpid()) |
| 455 | continue; |
| 456 | if (p <= 0) |
| 457 | continue; |
| 458 | |
| 459 | kill(p, sig); |
| 460 | } |
| 461 | } |
| 462 | |
| 463 | /** Turn a user name into a uid */ |
| 464 | int name_to_uid(const char *name, uid_t *uid) |
| 465 | { |
| 466 | struct passwd *pass; |
| 467 | if (!name || !*name) |
| 468 | return 0; |
| 469 | pass = getpwnam(name); |
| 470 | if (pass) { |
| 471 | *uid = pass->pw_uid; |
| 472 | return 1; |
| 473 | } |
| 474 | return 0; |
| 475 | } |
| 476 | |
| 477 | /** Turn a group name into a gid */ |
| 478 | int name_to_gid(const char *name, gid_t *gid) |
| 479 | { |
| 480 | struct group *grp; |
| 481 | if (!name || !*name) |
| 482 | return 0; |
| 483 | grp = getgrnam(name); |
| 484 | if (grp) { |
| 485 | *gid = grp->gr_gid; |
| 486 | return 1; |
| 487 | } |
| 488 | return 0; |
| 489 | } |
| 490 | |
| 491 | /** Lock a byte range in a open file */ |
| 492 | int lock_range(int fd, int offset, int len) |
| 493 | { |
| 494 | struct flock lock; |
| 495 | |
| 496 | lock.l_type = F_WRLCK; |
| 497 | lock.l_whence = SEEK_SET; |
| 498 | lock.l_start = offset; |
| 499 | lock.l_len = len; |
| 500 | lock.l_pid = 0; |
| 501 | |
| 502 | return fcntl(fd,F_SETLK,&lock) == 0; |
| 503 | } |
| 504 | |
| 505 | static int filter_server_path(char *arg) |
| 506 | { |
| 507 | char *s; |
| 508 | |
| 509 | if (server_filter_list.head) { |
| 510 | for (s = arg; (s = strchr(s, '/')) != NULL; ) { |
| 511 | *s = '\0'; |
| 512 | if (check_filter(&server_filter_list, arg, 1) < 0) { |
| 513 | /* We must leave arg truncated! */ |
| 514 | return 1; |
| 515 | } |
| 516 | *s++ = '/'; |
| 517 | } |
| 518 | } |
| 519 | return 0; |
| 520 | } |
| 521 | |
| 522 | static void glob_expand_one(char *s, char ***argv_ptr, int *argc_ptr, |
| 523 | int *maxargs_ptr) |
| 524 | { |
| 525 | char **argv = *argv_ptr; |
| 526 | int argc = *argc_ptr; |
| 527 | int maxargs = *maxargs_ptr; |
| 528 | #if !defined HAVE_GLOB || !defined HAVE_GLOB_H |
| 529 | if (argc == maxargs) { |
| 530 | maxargs += MAX_ARGS; |
| 531 | if (!(argv = realloc_array(argv, char *, maxargs))) |
| 532 | out_of_memory("glob_expand_one"); |
| 533 | *argv_ptr = argv; |
| 534 | *maxargs_ptr = maxargs; |
| 535 | } |
| 536 | if (!*s) |
| 537 | s = "."; |
| 538 | s = argv[argc++] = strdup(s); |
| 539 | filter_server_path(s); |
| 540 | #else |
| 541 | glob_t globbuf; |
| 542 | |
| 543 | if (maxargs <= argc) |
| 544 | return; |
| 545 | if (!*s) |
| 546 | s = "."; |
| 547 | |
| 548 | if (sanitize_paths) |
| 549 | s = sanitize_path(NULL, s, "", 0, NULL); |
| 550 | else |
| 551 | s = strdup(s); |
| 552 | |
| 553 | memset(&globbuf, 0, sizeof globbuf); |
| 554 | if (!filter_server_path(s)) |
| 555 | glob(s, 0, NULL, &globbuf); |
| 556 | if (MAX((int)globbuf.gl_pathc, 1) > maxargs - argc) { |
| 557 | maxargs += globbuf.gl_pathc + MAX_ARGS; |
| 558 | if (!(argv = realloc_array(argv, char *, maxargs))) |
| 559 | out_of_memory("glob_expand_one"); |
| 560 | *argv_ptr = argv; |
| 561 | *maxargs_ptr = maxargs; |
| 562 | } |
| 563 | if (globbuf.gl_pathc == 0) |
| 564 | argv[argc++] = s; |
| 565 | else { |
| 566 | int i; |
| 567 | free(s); |
| 568 | for (i = 0; i < (int)globbuf.gl_pathc; i++) { |
| 569 | if (!(argv[argc++] = strdup(globbuf.gl_pathv[i]))) |
| 570 | out_of_memory("glob_expand_one"); |
| 571 | } |
| 572 | } |
| 573 | globfree(&globbuf); |
| 574 | #endif |
| 575 | *argc_ptr = argc; |
| 576 | } |
| 577 | |
| 578 | /* This routine is only used in daemon mode. */ |
| 579 | void glob_expand(char *base1, char ***argv_ptr, int *argc_ptr, int *maxargs_ptr) |
| 580 | { |
| 581 | char *s = (*argv_ptr)[*argc_ptr]; |
| 582 | char *p, *q; |
| 583 | char *base = base1; |
| 584 | int base_len = strlen(base); |
| 585 | |
| 586 | if (!s || !*s) |
| 587 | return; |
| 588 | |
| 589 | if (strncmp(s, base, base_len) == 0) |
| 590 | s += base_len; |
| 591 | |
| 592 | if (!(s = strdup(s))) |
| 593 | out_of_memory("glob_expand"); |
| 594 | |
| 595 | if (asprintf(&base," %s/", base1) <= 0) |
| 596 | out_of_memory("glob_expand"); |
| 597 | base_len++; |
| 598 | |
| 599 | for (q = s; *q; q = p + base_len) { |
| 600 | if ((p = strstr(q, base)) != NULL) |
| 601 | *p = '\0'; /* split it at this point */ |
| 602 | glob_expand_one(q, argv_ptr, argc_ptr, maxargs_ptr); |
| 603 | if (!p) |
| 604 | break; |
| 605 | } |
| 606 | |
| 607 | free(s); |
| 608 | free(base); |
| 609 | } |
| 610 | |
| 611 | /** |
| 612 | * Convert a string to lower case |
| 613 | **/ |
| 614 | void strlower(char *s) |
| 615 | { |
| 616 | while (*s) { |
| 617 | if (isUpper(s)) |
| 618 | *s = toLower(s); |
| 619 | s++; |
| 620 | } |
| 621 | } |
| 622 | |
| 623 | /* Join strings p1 & p2 into "dest" with a guaranteed '/' between them. (If |
| 624 | * p1 ends with a '/', no extra '/' is inserted.) Returns the length of both |
| 625 | * strings + 1 (if '/' was inserted), regardless of whether the null-terminated |
| 626 | * string fits into destsize. */ |
| 627 | size_t pathjoin(char *dest, size_t destsize, const char *p1, const char *p2) |
| 628 | { |
| 629 | size_t len = strlcpy(dest, p1, destsize); |
| 630 | if (len < destsize - 1) { |
| 631 | if (!len || dest[len-1] != '/') |
| 632 | dest[len++] = '/'; |
| 633 | if (len < destsize - 1) |
| 634 | len += strlcpy(dest + len, p2, destsize - len); |
| 635 | else { |
| 636 | dest[len] = '\0'; |
| 637 | len += strlen(p2); |
| 638 | } |
| 639 | } |
| 640 | else |
| 641 | len += strlen(p2) + 1; /* Assume we'd insert a '/'. */ |
| 642 | return len; |
| 643 | } |
| 644 | |
| 645 | /* Join any number of strings together, putting them in "dest". The return |
| 646 | * value is the length of all the strings, regardless of whether the null- |
| 647 | * terminated whole fits in destsize. Your list of string pointers must end |
| 648 | * with a NULL to indicate the end of the list. */ |
| 649 | size_t stringjoin(char *dest, size_t destsize, ...) |
| 650 | { |
| 651 | va_list ap; |
| 652 | size_t len, ret = 0; |
| 653 | const char *src; |
| 654 | |
| 655 | va_start(ap, destsize); |
| 656 | while (1) { |
| 657 | if (!(src = va_arg(ap, const char *))) |
| 658 | break; |
| 659 | len = strlen(src); |
| 660 | ret += len; |
| 661 | if (destsize > 1) { |
| 662 | if (len >= destsize) |
| 663 | len = destsize - 1; |
| 664 | memcpy(dest, src, len); |
| 665 | destsize -= len; |
| 666 | dest += len; |
| 667 | } |
| 668 | } |
| 669 | *dest = '\0'; |
| 670 | va_end(ap); |
| 671 | |
| 672 | return ret; |
| 673 | } |
| 674 | |
| 675 | int count_dir_elements(const char *p) |
| 676 | { |
| 677 | int cnt = 0, new_component = 1; |
| 678 | while (*p) { |
| 679 | if (*p++ == '/') |
| 680 | new_component = (*p != '.' || (p[1] != '/' && p[1] != '\0')); |
| 681 | else if (new_component) { |
| 682 | new_component = 0; |
| 683 | cnt++; |
| 684 | } |
| 685 | } |
| 686 | return cnt; |
| 687 | } |
| 688 | |
| 689 | /* Turns multiple adjacent slashes into a single slash, gets rid of "./" |
| 690 | * elements (but not a trailing dot dir), removes a trailing slash, and |
| 691 | * optionally collapses ".." elements (except for those at the start of the |
| 692 | * string). If the resulting name would be empty, change it into a ".". */ |
| 693 | unsigned int clean_fname(char *name, BOOL collapse_dot_dot) |
| 694 | { |
| 695 | char *limit = name - 1, *t = name, *f = name; |
| 696 | int anchored; |
| 697 | |
| 698 | if (!name) |
| 699 | return 0; |
| 700 | |
| 701 | if ((anchored = *f == '/') != 0) |
| 702 | *t++ = *f++; |
| 703 | while (*f) { |
| 704 | /* discard extra slashes */ |
| 705 | if (*f == '/') { |
| 706 | f++; |
| 707 | continue; |
| 708 | } |
| 709 | if (*f == '.') { |
| 710 | /* discard "." dirs (but NOT a trailing '.'!) */ |
| 711 | if (f[1] == '/') { |
| 712 | f += 2; |
| 713 | continue; |
| 714 | } |
| 715 | /* collapse ".." dirs */ |
| 716 | if (collapse_dot_dot |
| 717 | && f[1] == '.' && (f[2] == '/' || !f[2])) { |
| 718 | char *s = t - 1; |
| 719 | if (s == name && anchored) { |
| 720 | f += 2; |
| 721 | continue; |
| 722 | } |
| 723 | while (s > limit && *--s != '/') {} |
| 724 | if (s != t - 1 && (s < name || *s == '/')) { |
| 725 | t = s + 1; |
| 726 | f += 2; |
| 727 | continue; |
| 728 | } |
| 729 | limit = t + 2; |
| 730 | } |
| 731 | } |
| 732 | while (*f && (*t++ = *f++) != '/') {} |
| 733 | } |
| 734 | |
| 735 | if (t > name+anchored && t[-1] == '/') |
| 736 | t--; |
| 737 | if (t == name) |
| 738 | *t++ = '.'; |
| 739 | *t = '\0'; |
| 740 | |
| 741 | return t - name; |
| 742 | } |
| 743 | |
| 744 | /* Make path appear as if a chroot had occurred. This handles a leading |
| 745 | * "/" (either removing it or expanding it) and any leading or embedded |
| 746 | * ".." components that attempt to escape past the module's top dir. |
| 747 | * |
| 748 | * If dest is NULL, a buffer is allocated to hold the result. It is legal |
| 749 | * to call with the dest and the path (p) pointing to the same buffer, but |
| 750 | * rootdir will be ignored to avoid expansion of the string. |
| 751 | * |
| 752 | * The rootdir string contains a value to use in place of a leading slash. |
| 753 | * Specify NULL to get the default of "module_dir". |
| 754 | * |
| 755 | * The depth var is a count of how many '..'s to allow at the start of the |
| 756 | * path. If symlink is set, combine its value with the "p" value to get |
| 757 | * the target path, and **return NULL if any '..'s try to escape**. |
| 758 | * |
| 759 | * We also clean the path in a manner similar to clean_fname() but with a |
| 760 | * few differences: |
| 761 | * |
| 762 | * Turns multiple adjacent slashes into a single slash, gets rid of "." dir |
| 763 | * elements (INCLUDING a trailing dot dir), PRESERVES a trailing slash, and |
| 764 | * ALWAYS collapses ".." elements (except for those at the start of the |
| 765 | * string up to "depth" deep). If the resulting name would be empty, |
| 766 | * change it into a ".". */ |
| 767 | char *sanitize_path(char *dest, const char *p, const char *rootdir, int depth, |
| 768 | const char *symlink) |
| 769 | { |
| 770 | char *start, *sanp, *save_dest = dest; |
| 771 | int rlen = 0, leave_one_dotdir = relative_paths; |
| 772 | |
| 773 | if (symlink && *symlink == '/') { |
| 774 | p = symlink; |
| 775 | symlink = ""; |
| 776 | } |
| 777 | |
| 778 | if (dest != p) { |
| 779 | int plen = strlen(p); |
| 780 | if (*p == '/') { |
| 781 | if (!rootdir) |
| 782 | rootdir = module_dir; |
| 783 | rlen = strlen(rootdir); |
| 784 | depth = 0; |
| 785 | p++; |
| 786 | } |
| 787 | if (dest) { |
| 788 | if (rlen + plen + 1 >= MAXPATHLEN) |
| 789 | return NULL; |
| 790 | } else if (!(dest = new_array(char, rlen + plen + 1))) |
| 791 | out_of_memory("sanitize_path"); |
| 792 | if (rlen) { |
| 793 | memcpy(dest, rootdir, rlen); |
| 794 | if (rlen > 1) |
| 795 | dest[rlen++] = '/'; |
| 796 | } |
| 797 | } |
| 798 | |
| 799 | start = sanp = dest + rlen; |
| 800 | while (1) { |
| 801 | if (*p == '\0') { |
| 802 | if (!symlink || !*symlink) |
| 803 | break; |
| 804 | while (sanp != start && sanp[-1] != '/') { |
| 805 | /* strip last element */ |
| 806 | sanp--; |
| 807 | } |
| 808 | /* Append a relative symlink */ |
| 809 | p = symlink; |
| 810 | symlink = ""; |
| 811 | } |
| 812 | /* discard leading or extra slashes */ |
| 813 | if (*p == '/') { |
| 814 | p++; |
| 815 | continue; |
| 816 | } |
| 817 | /* this loop iterates once per filename component in p. |
| 818 | * both p (and sanp if the original had a slash) should |
| 819 | * always be left pointing after a slash |
| 820 | */ |
| 821 | if (*p == '.' && (p[1] == '/' || p[1] == '\0')) { |
| 822 | if (leave_one_dotdir && p[1]) |
| 823 | leave_one_dotdir = 0; |
| 824 | else { |
| 825 | /* skip "." component */ |
| 826 | p++; |
| 827 | continue; |
| 828 | } |
| 829 | } |
| 830 | if (*p == '.' && p[1] == '.' && (p[2] == '/' || p[2] == '\0')) { |
| 831 | /* ".." component followed by slash or end */ |
| 832 | if (depth <= 0 || sanp != start) { |
| 833 | if (symlink && sanp == start) { |
| 834 | if (!save_dest) |
| 835 | free(dest); |
| 836 | return NULL; |
| 837 | } |
| 838 | p += 2; |
| 839 | if (sanp != start) { |
| 840 | /* back up sanp one level */ |
| 841 | --sanp; /* now pointing at slash */ |
| 842 | while (sanp > start && sanp[-1] != '/') { |
| 843 | /* skip back up to slash */ |
| 844 | sanp--; |
| 845 | } |
| 846 | } |
| 847 | continue; |
| 848 | } |
| 849 | /* allow depth levels of .. at the beginning */ |
| 850 | depth--; |
| 851 | /* move the virtual beginning to leave the .. alone */ |
| 852 | start = sanp + 3; |
| 853 | } |
| 854 | /* copy one component through next slash */ |
| 855 | while (*p && (*sanp++ = *p++) != '/') {} |
| 856 | } |
| 857 | if (sanp == dest) { |
| 858 | /* ended up with nothing, so put in "." component */ |
| 859 | *sanp++ = '.'; |
| 860 | } |
| 861 | *sanp = '\0'; |
| 862 | |
| 863 | return dest; |
| 864 | } |
| 865 | |
| 866 | /* Like chdir(), but it keeps track of the current directory (in the |
| 867 | * global "curr_dir"), and ensures that the path size doesn't overflow. |
| 868 | * Also cleans the path using the clean_fname() function. */ |
| 869 | int push_dir(const char *dir, int set_path_only) |
| 870 | { |
| 871 | static int initialised; |
| 872 | unsigned int len; |
| 873 | |
| 874 | if (!initialised) { |
| 875 | initialised = 1; |
| 876 | getcwd(curr_dir, sizeof curr_dir - 1); |
| 877 | curr_dir_len = strlen(curr_dir); |
| 878 | } |
| 879 | |
| 880 | if (!dir) /* this call was probably just to initialize */ |
| 881 | return 0; |
| 882 | |
| 883 | len = strlen(dir); |
| 884 | if (len == 1 && *dir == '.') |
| 885 | return 1; |
| 886 | |
| 887 | if ((*dir == '/' ? len : curr_dir_len + 1 + len) >= sizeof curr_dir) { |
| 888 | errno = ENAMETOOLONG; |
| 889 | return 0; |
| 890 | } |
| 891 | |
| 892 | if (!set_path_only && chdir(dir)) |
| 893 | return 0; |
| 894 | |
| 895 | if (*dir == '/') { |
| 896 | memcpy(curr_dir, dir, len + 1); |
| 897 | curr_dir_len = len; |
| 898 | } else { |
| 899 | curr_dir[curr_dir_len++] = '/'; |
| 900 | memcpy(curr_dir + curr_dir_len, dir, len + 1); |
| 901 | curr_dir_len += len; |
| 902 | } |
| 903 | |
| 904 | curr_dir_len = clean_fname(curr_dir, 1); |
| 905 | if (sanitize_paths) { |
| 906 | if (module_dirlen > curr_dir_len) |
| 907 | module_dirlen = curr_dir_len; |
| 908 | curr_dir_depth = count_dir_elements(curr_dir + module_dirlen); |
| 909 | } |
| 910 | |
| 911 | return 1; |
| 912 | } |
| 913 | |
| 914 | /** |
| 915 | * Reverse a push_dir() call. You must pass in an absolute path |
| 916 | * that was copied from a prior value of "curr_dir". |
| 917 | **/ |
| 918 | int pop_dir(const char *dir) |
| 919 | { |
| 920 | if (chdir(dir)) |
| 921 | return 0; |
| 922 | |
| 923 | curr_dir_len = strlcpy(curr_dir, dir, sizeof curr_dir); |
| 924 | if (curr_dir_len >= sizeof curr_dir) |
| 925 | curr_dir_len = sizeof curr_dir - 1; |
| 926 | if (sanitize_paths) |
| 927 | curr_dir_depth = count_dir_elements(curr_dir + module_dirlen); |
| 928 | |
| 929 | return 1; |
| 930 | } |
| 931 | |
| 932 | /** |
| 933 | * Return a quoted string with the full pathname of the indicated filename. |
| 934 | * The string " (in MODNAME)" may also be appended. The returned pointer |
| 935 | * remains valid until the next time full_fname() is called. |
| 936 | **/ |
| 937 | char *full_fname(const char *fn) |
| 938 | { |
| 939 | static char *result = NULL; |
| 940 | char *m1, *m2, *m3; |
| 941 | char *p1, *p2; |
| 942 | |
| 943 | if (result) |
| 944 | free(result); |
| 945 | |
| 946 | if (*fn == '/') |
| 947 | p1 = p2 = ""; |
| 948 | else { |
| 949 | p1 = curr_dir + module_dirlen; |
| 950 | for (p2 = p1; *p2 == '/'; p2++) {} |
| 951 | if (*p2) |
| 952 | p2 = "/"; |
| 953 | } |
| 954 | if (module_id >= 0) { |
| 955 | m1 = " (in "; |
| 956 | m2 = lp_name(module_id); |
| 957 | m3 = ")"; |
| 958 | } else |
| 959 | m1 = m2 = m3 = ""; |
| 960 | |
| 961 | if (asprintf(&result, "\"%s%s%s\"%s%s%s", p1, p2, fn, m1, m2, m3) <= 0) |
| 962 | out_of_memory("full_fname"); |
| 963 | |
| 964 | return result; |
| 965 | } |
| 966 | |
| 967 | static char partial_fname[MAXPATHLEN]; |
| 968 | |
| 969 | char *partial_dir_fname(const char *fname) |
| 970 | { |
| 971 | char *t = partial_fname; |
| 972 | int sz = sizeof partial_fname; |
| 973 | const char *fn; |
| 974 | |
| 975 | if ((fn = strrchr(fname, '/')) != NULL) { |
| 976 | fn++; |
| 977 | if (*partial_dir != '/') { |
| 978 | int len = fn - fname; |
| 979 | strncpy(t, fname, len); /* safe */ |
| 980 | t += len; |
| 981 | sz -= len; |
| 982 | } |
| 983 | } else |
| 984 | fn = fname; |
| 985 | if ((int)pathjoin(t, sz, partial_dir, fn) >= sz) |
| 986 | return NULL; |
| 987 | if (server_filter_list.head) { |
| 988 | t = strrchr(partial_fname, '/'); |
| 989 | *t = '\0'; |
| 990 | if (check_filter(&server_filter_list, partial_fname, 1) < 0) |
| 991 | return NULL; |
| 992 | *t = '/'; |
| 993 | if (check_filter(&server_filter_list, partial_fname, 0) < 0) |
| 994 | return NULL; |
| 995 | } |
| 996 | |
| 997 | return partial_fname; |
| 998 | } |
| 999 | |
| 1000 | /* If no --partial-dir option was specified, we don't need to do anything |
| 1001 | * (the partial-dir is essentially '.'), so just return success. */ |
| 1002 | int handle_partial_dir(const char *fname, int create) |
| 1003 | { |
| 1004 | char *fn, *dir; |
| 1005 | |
| 1006 | if (fname != partial_fname) |
| 1007 | return 1; |
| 1008 | if (!create && *partial_dir == '/') |
| 1009 | return 1; |
| 1010 | if (!(fn = strrchr(partial_fname, '/'))) |
| 1011 | return 1; |
| 1012 | |
| 1013 | *fn = '\0'; |
| 1014 | dir = partial_fname; |
| 1015 | if (create) { |
| 1016 | STRUCT_STAT st; |
| 1017 | int statret = do_lstat(dir, &st); |
| 1018 | if (statret == 0 && !S_ISDIR(st.st_mode)) { |
| 1019 | if (do_unlink(dir) < 0) |
| 1020 | return 0; |
| 1021 | statret = -1; |
| 1022 | } |
| 1023 | if (statret < 0 && do_mkdir(dir, 0700) < 0) |
| 1024 | return 0; |
| 1025 | } else |
| 1026 | do_rmdir(dir); |
| 1027 | *fn = '/'; |
| 1028 | |
| 1029 | return 1; |
| 1030 | } |
| 1031 | |
| 1032 | /** |
| 1033 | * Determine if a symlink points outside the current directory tree. |
| 1034 | * This is considered "unsafe" because e.g. when mirroring somebody |
| 1035 | * else's machine it might allow them to establish a symlink to |
| 1036 | * /etc/passwd, and then read it through a web server. |
| 1037 | * |
| 1038 | * Null symlinks and absolute symlinks are always unsafe. |
| 1039 | * |
| 1040 | * Basically here we are concerned with symlinks whose target contains |
| 1041 | * "..", because this might cause us to walk back up out of the |
| 1042 | * transferred directory. We are not allowed to go back up and |
| 1043 | * reenter. |
| 1044 | * |
| 1045 | * @param dest Target of the symlink in question. |
| 1046 | * |
| 1047 | * @param src Top source directory currently applicable. Basically this |
| 1048 | * is the first parameter to rsync in a simple invocation, but it's |
| 1049 | * modified by flist.c in slightly complex ways. |
| 1050 | * |
| 1051 | * @retval True if unsafe |
| 1052 | * @retval False is unsafe |
| 1053 | * |
| 1054 | * @sa t_unsafe.c |
| 1055 | **/ |
| 1056 | int unsafe_symlink(const char *dest, const char *src) |
| 1057 | { |
| 1058 | const char *name, *slash; |
| 1059 | int depth = 0; |
| 1060 | |
| 1061 | /* all absolute and null symlinks are unsafe */ |
| 1062 | if (!dest || !*dest || *dest == '/') |
| 1063 | return 1; |
| 1064 | |
| 1065 | /* find out what our safety margin is */ |
| 1066 | for (name = src; (slash = strchr(name, '/')) != 0; name = slash+1) { |
| 1067 | if (strncmp(name, "../", 3) == 0) { |
| 1068 | depth = 0; |
| 1069 | } else if (strncmp(name, "./", 2) == 0) { |
| 1070 | /* nothing */ |
| 1071 | } else { |
| 1072 | depth++; |
| 1073 | } |
| 1074 | } |
| 1075 | if (strcmp(name, "..") == 0) |
| 1076 | depth = 0; |
| 1077 | |
| 1078 | for (name = dest; (slash = strchr(name, '/')) != 0; name = slash+1) { |
| 1079 | if (strncmp(name, "../", 3) == 0) { |
| 1080 | /* if at any point we go outside the current directory |
| 1081 | then stop - it is unsafe */ |
| 1082 | if (--depth < 0) |
| 1083 | return 1; |
| 1084 | } else if (strncmp(name, "./", 2) == 0) { |
| 1085 | /* nothing */ |
| 1086 | } else { |
| 1087 | depth++; |
| 1088 | } |
| 1089 | } |
| 1090 | if (strcmp(name, "..") == 0) |
| 1091 | depth--; |
| 1092 | |
| 1093 | return (depth < 0); |
| 1094 | } |
| 1095 | |
| 1096 | /* Return the int64 number as a string. If the --human-readable option was |
| 1097 | * specified, we may output the number in K, M, or G units. We can return |
| 1098 | * up to 4 buffers at a time. */ |
| 1099 | char *human_num(int64 num) |
| 1100 | { |
| 1101 | static char bufs[4][128]; /* more than enough room */ |
| 1102 | static unsigned int n; |
| 1103 | char *s; |
| 1104 | |
| 1105 | n = (n + 1) % (sizeof bufs / sizeof bufs[0]); |
| 1106 | |
| 1107 | if (human_readable) { |
| 1108 | char units = '\0'; |
| 1109 | int mult = human_readable == 1 ? 1000 : 1024; |
| 1110 | double dnum = 0; |
| 1111 | if (num > mult*mult*mult) { |
| 1112 | dnum = (double)num / (mult*mult*mult); |
| 1113 | units = 'G'; |
| 1114 | } else if (num > mult*mult) { |
| 1115 | dnum = (double)num / (mult*mult); |
| 1116 | units = 'M'; |
| 1117 | } else if (num > mult) { |
| 1118 | dnum = (double)num / mult; |
| 1119 | units = 'K'; |
| 1120 | } |
| 1121 | if (units) { |
| 1122 | snprintf(bufs[n], sizeof bufs[0], "%.2f%c", dnum, units); |
| 1123 | return bufs[n]; |
| 1124 | } |
| 1125 | } |
| 1126 | |
| 1127 | s = bufs[n] + sizeof bufs[0] - 1; |
| 1128 | *s = '\0'; |
| 1129 | |
| 1130 | if (!num) |
| 1131 | *--s = '0'; |
| 1132 | while (num) { |
| 1133 | *--s = (char)(num % 10) + '0'; |
| 1134 | num /= 10; |
| 1135 | } |
| 1136 | return s; |
| 1137 | } |
| 1138 | |
| 1139 | /* Return the double number as a string. If the --human-readable option was |
| 1140 | * specified, we may output the number in K, M, or G units. We use a buffer |
| 1141 | * from human_num() to return our result. */ |
| 1142 | char *human_dnum(double dnum, int decimal_digits) |
| 1143 | { |
| 1144 | char *buf = human_num(dnum); |
| 1145 | int len = strlen(buf); |
| 1146 | if (isDigit(buf + len - 1)) { |
| 1147 | /* There's extra room in buf prior to the start of the num. */ |
| 1148 | buf -= decimal_digits + 1; |
| 1149 | snprintf(buf, len + decimal_digits + 2, "%.*f", decimal_digits, dnum); |
| 1150 | } |
| 1151 | return buf; |
| 1152 | } |
| 1153 | |
| 1154 | /* Return the date and time as a string. Some callers tweak returned buf. */ |
| 1155 | char *timestring(time_t t) |
| 1156 | { |
| 1157 | static char TimeBuf[200]; |
| 1158 | struct tm *tm = localtime(&t); |
| 1159 | char *p; |
| 1160 | |
| 1161 | #ifdef HAVE_STRFTIME |
| 1162 | strftime(TimeBuf, sizeof TimeBuf - 1, "%Y/%m/%d %H:%M:%S", tm); |
| 1163 | #else |
| 1164 | strlcpy(TimeBuf, asctime(tm), sizeof TimeBuf); |
| 1165 | #endif |
| 1166 | |
| 1167 | if ((p = strchr(TimeBuf, '\n')) != NULL) |
| 1168 | *p = '\0'; |
| 1169 | |
| 1170 | return TimeBuf; |
| 1171 | } |
| 1172 | |
| 1173 | /** |
| 1174 | * Sleep for a specified number of milliseconds. |
| 1175 | * |
| 1176 | * Always returns TRUE. (In the future it might return FALSE if |
| 1177 | * interrupted.) |
| 1178 | **/ |
| 1179 | int msleep(int t) |
| 1180 | { |
| 1181 | int tdiff = 0; |
| 1182 | struct timeval tval, t1, t2; |
| 1183 | |
| 1184 | gettimeofday(&t1, NULL); |
| 1185 | |
| 1186 | while (tdiff < t) { |
| 1187 | tval.tv_sec = (t-tdiff)/1000; |
| 1188 | tval.tv_usec = 1000*((t-tdiff)%1000); |
| 1189 | |
| 1190 | errno = 0; |
| 1191 | select(0,NULL,NULL, NULL, &tval); |
| 1192 | |
| 1193 | gettimeofday(&t2, NULL); |
| 1194 | tdiff = (t2.tv_sec - t1.tv_sec)*1000 + |
| 1195 | (t2.tv_usec - t1.tv_usec)/1000; |
| 1196 | } |
| 1197 | |
| 1198 | return True; |
| 1199 | } |
| 1200 | |
| 1201 | /* Determine if two time_t values are equivalent (either exact, or in |
| 1202 | * the modification timestamp window established by --modify-window). |
| 1203 | * |
| 1204 | * @retval 0 if the times should be treated as the same |
| 1205 | * |
| 1206 | * @retval +1 if the first is later |
| 1207 | * |
| 1208 | * @retval -1 if the 2nd is later |
| 1209 | **/ |
| 1210 | int cmp_time(time_t file1, time_t file2) |
| 1211 | { |
| 1212 | if (file2 > file1) { |
| 1213 | if (file2 - file1 <= modify_window) |
| 1214 | return 0; |
| 1215 | return -1; |
| 1216 | } |
| 1217 | if (file1 - file2 <= modify_window) |
| 1218 | return 0; |
| 1219 | return 1; |
| 1220 | } |
| 1221 | |
| 1222 | |
| 1223 | #ifdef __INSURE__XX |
| 1224 | #include <dlfcn.h> |
| 1225 | |
| 1226 | /** |
| 1227 | This routine is a trick to immediately catch errors when debugging |
| 1228 | with insure. A xterm with a gdb is popped up when insure catches |
| 1229 | a error. It is Linux specific. |
| 1230 | **/ |
| 1231 | int _Insure_trap_error(int a1, int a2, int a3, int a4, int a5, int a6) |
| 1232 | { |
| 1233 | static int (*fn)(); |
| 1234 | int ret; |
| 1235 | char *cmd; |
| 1236 | |
| 1237 | 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'", |
| 1238 | getpid(), getpid(), getpid()); |
| 1239 | |
| 1240 | if (!fn) { |
| 1241 | static void *h; |
| 1242 | h = dlopen("/usr/local/parasoft/insure++lite/lib.linux2/libinsure.so", RTLD_LAZY); |
| 1243 | fn = dlsym(h, "_Insure_trap_error"); |
| 1244 | } |
| 1245 | |
| 1246 | ret = fn(a1, a2, a3, a4, a5, a6); |
| 1247 | |
| 1248 | system(cmd); |
| 1249 | |
| 1250 | free(cmd); |
| 1251 | |
| 1252 | return ret; |
| 1253 | } |
| 1254 | #endif |
| 1255 | |
| 1256 | #define MALLOC_MAX 0x40000000 |
| 1257 | |
| 1258 | void *_new_array(unsigned int size, unsigned long num) |
| 1259 | { |
| 1260 | if (num >= MALLOC_MAX/size) |
| 1261 | return NULL; |
| 1262 | return malloc(size * num); |
| 1263 | } |
| 1264 | |
| 1265 | void *_realloc_array(void *ptr, unsigned int size, unsigned long num) |
| 1266 | { |
| 1267 | if (num >= MALLOC_MAX/size) |
| 1268 | return NULL; |
| 1269 | if (!ptr) |
| 1270 | return malloc(size * num); |
| 1271 | return realloc(ptr, size * num); |
| 1272 | } |
| 1273 | |
| 1274 | /* Take a filename and filename length and return the most significant |
| 1275 | * filename suffix we can find. This ignores suffixes such as "~", |
| 1276 | * ".bak", ".orig", ".~1~", etc. */ |
| 1277 | const char *find_filename_suffix(const char *fn, int fn_len, int *len_ptr) |
| 1278 | { |
| 1279 | const char *suf, *s; |
| 1280 | BOOL had_tilde; |
| 1281 | int s_len; |
| 1282 | |
| 1283 | /* One or more dots at the start aren't a suffix. */ |
| 1284 | while (fn_len && *fn == '.') fn++, fn_len--; |
| 1285 | |
| 1286 | /* Ignore the ~ in a "foo~" filename. */ |
| 1287 | if (fn_len > 1 && fn[fn_len-1] == '~') |
| 1288 | fn_len--, had_tilde = True; |
| 1289 | else |
| 1290 | had_tilde = False; |
| 1291 | |
| 1292 | /* Assume we don't find an suffix. */ |
| 1293 | suf = ""; |
| 1294 | *len_ptr = 0; |
| 1295 | |
| 1296 | /* Find the last significant suffix. */ |
| 1297 | for (s = fn + fn_len; fn_len > 1; ) { |
| 1298 | while (*--s != '.' && s != fn) {} |
| 1299 | if (s == fn) |
| 1300 | break; |
| 1301 | s_len = fn_len - (s - fn); |
| 1302 | fn_len = s - fn; |
| 1303 | if (s_len == 4) { |
| 1304 | if (strcmp(s+1, "bak") == 0 |
| 1305 | || strcmp(s+1, "old") == 0) |
| 1306 | continue; |
| 1307 | } else if (s_len == 5) { |
| 1308 | if (strcmp(s+1, "orig") == 0) |
| 1309 | continue; |
| 1310 | } else if (s_len > 2 && had_tilde |
| 1311 | && s[1] == '~' && isDigit(s + 2)) |
| 1312 | continue; |
| 1313 | *len_ptr = s_len; |
| 1314 | suf = s; |
| 1315 | if (s_len == 1) |
| 1316 | break; |
| 1317 | /* Determine if the suffix is all digits. */ |
| 1318 | for (s++, s_len--; s_len > 0; s++, s_len--) { |
| 1319 | if (!isDigit(s)) |
| 1320 | return suf; |
| 1321 | } |
| 1322 | /* An all-digit suffix may not be that signficant. */ |
| 1323 | s = suf; |
| 1324 | } |
| 1325 | |
| 1326 | return suf; |
| 1327 | } |
| 1328 | |
| 1329 | /* This is an implementation of the Levenshtein distance algorithm. It |
| 1330 | * was implemented to avoid needing a two-dimensional matrix (to save |
| 1331 | * memory). It was also tweaked to try to factor in the ASCII distance |
| 1332 | * between changed characters as a minor distance quantity. The normal |
| 1333 | * Levenshtein units of distance (each signifying a single change between |
| 1334 | * the two strings) are defined as a "UNIT". */ |
| 1335 | |
| 1336 | #define UNIT (1 << 16) |
| 1337 | |
| 1338 | uint32 fuzzy_distance(const char *s1, int len1, const char *s2, int len2) |
| 1339 | { |
| 1340 | uint32 a[MAXPATHLEN], diag, above, left, diag_inc, above_inc, left_inc; |
| 1341 | int32 cost; |
| 1342 | int i1, i2; |
| 1343 | |
| 1344 | if (!len1 || !len2) { |
| 1345 | if (!len1) { |
| 1346 | s1 = s2; |
| 1347 | len1 = len2; |
| 1348 | } |
| 1349 | for (i1 = 0, cost = 0; i1 < len1; i1++) |
| 1350 | cost += s1[i1]; |
| 1351 | return (int32)len1 * UNIT + cost; |
| 1352 | } |
| 1353 | |
| 1354 | for (i2 = 0; i2 < len2; i2++) |
| 1355 | a[i2] = (i2+1) * UNIT; |
| 1356 | |
| 1357 | for (i1 = 0; i1 < len1; i1++) { |
| 1358 | diag = i1 * UNIT; |
| 1359 | above = (i1+1) * UNIT; |
| 1360 | for (i2 = 0; i2 < len2; i2++) { |
| 1361 | left = a[i2]; |
| 1362 | if ((cost = *((uchar*)s1+i1) - *((uchar*)s2+i2)) != 0) { |
| 1363 | if (cost < 0) |
| 1364 | cost = UNIT - cost; |
| 1365 | else |
| 1366 | cost = UNIT + cost; |
| 1367 | } |
| 1368 | diag_inc = diag + cost; |
| 1369 | left_inc = left + UNIT + *((uchar*)s1+i1); |
| 1370 | above_inc = above + UNIT + *((uchar*)s2+i2); |
| 1371 | a[i2] = above = left < above |
| 1372 | ? (left_inc < diag_inc ? left_inc : diag_inc) |
| 1373 | : (above_inc < diag_inc ? above_inc : diag_inc); |
| 1374 | diag = left; |
| 1375 | } |
| 1376 | } |
| 1377 | |
| 1378 | return a[len2-1]; |
| 1379 | } |
| 1380 | |
| 1381 | #define BB_SLOT_SIZE (16*1024) /* Desired size in bytes */ |
| 1382 | #define BB_PER_SLOT_BITS (BB_SLOT_SIZE * 8) /* Number of bits per slot */ |
| 1383 | #define BB_PER_SLOT_INTS (BB_SLOT_SIZE / 4) /* Number of int32s per slot */ |
| 1384 | |
| 1385 | struct bitbag { |
| 1386 | uint32 **bits; |
| 1387 | int slot_cnt; |
| 1388 | }; |
| 1389 | |
| 1390 | struct bitbag *bitbag_create(int max_ndx) |
| 1391 | { |
| 1392 | struct bitbag *bb = new(struct bitbag); |
| 1393 | bb->slot_cnt = (max_ndx + BB_PER_SLOT_BITS - 1) / BB_PER_SLOT_BITS; |
| 1394 | |
| 1395 | if (!(bb->bits = (uint32**)calloc(bb->slot_cnt, sizeof (uint32*)))) |
| 1396 | out_of_memory("bitbag_create"); |
| 1397 | |
| 1398 | return bb; |
| 1399 | } |
| 1400 | |
| 1401 | void bitbag_set_bit(struct bitbag *bb, int ndx) |
| 1402 | { |
| 1403 | int slot = ndx / BB_PER_SLOT_BITS; |
| 1404 | ndx %= BB_PER_SLOT_BITS; |
| 1405 | |
| 1406 | if (!bb->bits[slot]) { |
| 1407 | if (!(bb->bits[slot] = (uint32*)calloc(BB_PER_SLOT_INTS, 4))) |
| 1408 | out_of_memory("bitbag_set_bit"); |
| 1409 | } |
| 1410 | |
| 1411 | bb->bits[slot][ndx/32] |= 1u << (ndx % 32); |
| 1412 | } |
| 1413 | |
| 1414 | #if 0 /* not needed yet */ |
| 1415 | void bitbag_clear_bit(struct bitbag *bb, int ndx) |
| 1416 | { |
| 1417 | int slot = ndx / BB_PER_SLOT_BITS; |
| 1418 | ndx %= BB_PER_SLOT_BITS; |
| 1419 | |
| 1420 | if (!bb->bits[slot]) |
| 1421 | return; |
| 1422 | |
| 1423 | bb->bits[slot][ndx/32] &= ~(1u << (ndx % 32)); |
| 1424 | } |
| 1425 | |
| 1426 | int bitbag_check_bit(struct bitbag *bb, int ndx) |
| 1427 | { |
| 1428 | int slot = ndx / BB_PER_SLOT_BITS; |
| 1429 | ndx %= BB_PER_SLOT_BITS; |
| 1430 | |
| 1431 | if (!bb->bits[slot]) |
| 1432 | return 0; |
| 1433 | |
| 1434 | return bb->bits[slot][ndx/32] & (1u << (ndx % 32)) ? 1 : 0; |
| 1435 | } |
| 1436 | #endif |
| 1437 | |
| 1438 | /* Call this with -1 to start checking from 0. Returns -1 at the end. */ |
| 1439 | int bitbag_next_bit(struct bitbag *bb, int after) |
| 1440 | { |
| 1441 | uint32 bits, mask; |
| 1442 | int i, ndx = after + 1; |
| 1443 | int slot = ndx / BB_PER_SLOT_BITS; |
| 1444 | ndx %= BB_PER_SLOT_BITS; |
| 1445 | |
| 1446 | mask = (1u << (ndx % 32)) - 1; |
| 1447 | for (i = ndx / 32; slot < bb->slot_cnt; slot++, i = mask = 0) { |
| 1448 | if (!bb->bits[slot]) |
| 1449 | continue; |
| 1450 | for ( ; i < BB_PER_SLOT_INTS; i++, mask = 0) { |
| 1451 | if (!(bits = bb->bits[slot][i] & ~mask)) |
| 1452 | continue; |
| 1453 | /* The xor magic figures out the lowest enabled bit in |
| 1454 | * bits, and the switch quickly computes log2(bit). */ |
| 1455 | switch (bits ^ (bits & (bits-1))) { |
| 1456 | #define LOG2(n) case 1u << n: return slot*BB_PER_SLOT_BITS + i*32 + n |
| 1457 | LOG2(0); LOG2(1); LOG2(2); LOG2(3); |
| 1458 | LOG2(4); LOG2(5); LOG2(6); LOG2(7); |
| 1459 | LOG2(8); LOG2(9); LOG2(10); LOG2(11); |
| 1460 | LOG2(12); LOG2(13); LOG2(14); LOG2(15); |
| 1461 | LOG2(16); LOG2(17); LOG2(18); LOG2(19); |
| 1462 | LOG2(20); LOG2(21); LOG2(22); LOG2(23); |
| 1463 | LOG2(24); LOG2(25); LOG2(26); LOG2(27); |
| 1464 | LOG2(28); LOG2(29); LOG2(30); LOG2(31); |
| 1465 | } |
| 1466 | return -1; /* impossible... */ |
| 1467 | } |
| 1468 | } |
| 1469 | |
| 1470 | return -1; |
| 1471 | } |
| 1472 | |
| 1473 | void *expand_item_list(item_list *lp, size_t item_size, |
| 1474 | const char *desc, int incr) |
| 1475 | { |
| 1476 | /* First time through, 0 <= 0, so list is expanded. */ |
| 1477 | if (lp->malloced <= lp->count) { |
| 1478 | void *new_ptr; |
| 1479 | size_t new_size = lp->malloced; |
| 1480 | if (incr < 0) |
| 1481 | new_size -= incr; /* increase slowly */ |
| 1482 | else if (new_size < (size_t)incr) |
| 1483 | new_size += incr; |
| 1484 | else |
| 1485 | new_size *= 2; |
| 1486 | new_ptr = realloc_array(lp->items, char, new_size * item_size); |
| 1487 | if (verbose >= 4) { |
| 1488 | rprintf(FINFO, "[%s] expand %s to %.0f bytes, did%s move\n", |
| 1489 | who_am_i(), desc, (double)new_size * item_size, |
| 1490 | new_ptr == lp->items ? " not" : ""); |
| 1491 | } |
| 1492 | if (!new_ptr) |
| 1493 | out_of_memory("expand_item_list"); |
| 1494 | |
| 1495 | lp->items = new_ptr; |
| 1496 | lp->malloced = new_size; |
| 1497 | } |
| 1498 | return (char*)lp->items + (lp->count++ * item_size); |
| 1499 | } |