1 /* $NetBSD: fts.c,v 1.38 2009/02/28 14:34:18 pgoyette Exp $ */ 2 3 /*- 4 * Copyright (c) 1990, 1993, 1994 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 #if HAVE_NBTOOL_CONFIG_H 33 #include "nbtool_config.h" 34 #endif 35 36 #include <sys/cdefs.h> 37 #if defined(LIBC_SCCS) && !defined(lint) 38 #if 0 39 static char sccsid[] = "@(#)fts.c 8.6 (Berkeley) 8/14/94"; 40 #else 41 __RCSID("$NetBSD: fts.c,v 1.38 2009/02/28 14:34:18 pgoyette Exp $"); 42 #endif 43 #endif /* LIBC_SCCS and not lint */ 44 45 #include "namespace.h" 46 #include <sys/param.h> 47 #include <sys/stat.h> 48 49 #include <assert.h> 50 #include <dirent.h> 51 #include <errno.h> 52 #include <fcntl.h> 53 #include <fts.h> 54 #include <stdlib.h> 55 #include <string.h> 56 #include <unistd.h> 57 58 #if ! HAVE_NBTOOL_CONFIG_H 59 #define HAVE_STRUCT_DIRENT_D_NAMLEN 60 #endif 61 62 static FTSENT *fts_alloc(FTS *, const char *, size_t); 63 static FTSENT *fts_build(FTS *, int); 64 static void fts_free(FTSENT *); 65 static void fts_lfree(FTSENT *); 66 static void fts_load(FTS *, FTSENT *); 67 static size_t fts_maxarglen(char * const *); 68 static size_t fts_pow2(size_t); 69 static int fts_palloc(FTS *, size_t); 70 static void fts_padjust(FTS *, FTSENT *); 71 static FTSENT *fts_sort(FTS *, FTSENT *, size_t); 72 static unsigned short fts_stat(FTS *, FTSENT *, int); 73 static int fts_safe_changedir(const FTS *, const FTSENT *, int, 74 const char *); 75 76 #if defined(ALIGNBYTES) && defined(ALIGN) 77 #define FTS_ALLOC_ALIGNED 1 78 #else 79 #undef FTS_ALLOC_ALIGNED 80 #endif 81 82 #define ISDOT(a) (a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2]))) 83 84 #define CLR(opt) (sp->fts_options &= ~(opt)) 85 #define ISSET(opt) (sp->fts_options & (opt)) 86 #define SET(opt) (sp->fts_options |= (opt)) 87 88 #define CHDIR(sp, path) (!ISSET(FTS_NOCHDIR) && chdir(path)) 89 #define FCHDIR(sp, fd) (!ISSET(FTS_NOCHDIR) && fchdir(fd)) 90 91 /* fts_build flags */ 92 #define BCHILD 1 /* fts_children */ 93 #define BNAMES 2 /* fts_children, names only */ 94 #define BREAD 3 /* fts_read */ 95 96 #ifndef DTF_HIDEW 97 #undef FTS_WHITEOUT 98 #endif 99 100 FTS * 101 fts_open(char * const *argv, int options, 102 int (*compar)(const FTSENT **, const FTSENT **)) 103 { 104 FTS *sp; 105 FTSENT *p, *root; 106 size_t nitems; 107 FTSENT *parent, *tmp = NULL; /* pacify gcc */ 108 size_t len; 109 110 _DIAGASSERT(argv != NULL); 111 112 /* Options check. */ 113 if (options & ~FTS_OPTIONMASK) { 114 errno = EINVAL; 115 return (NULL); 116 } 117 118 /* Allocate/initialize the stream */ 119 if ((sp = malloc((unsigned int)sizeof(FTS))) == NULL) 120 return (NULL); 121 memset(sp, 0, sizeof(FTS)); 122 sp->fts_compar = compar; 123 sp->fts_options = options; 124 125 /* Logical walks turn on NOCHDIR; symbolic links are too hard. */ 126 if (ISSET(FTS_LOGICAL)) 127 SET(FTS_NOCHDIR); 128 129 /* 130 * Start out with 1K of path space, and enough, in any case, 131 * to hold the user's paths. 132 */ 133 if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN))) 134 goto mem1; 135 136 /* Allocate/initialize root's parent. */ 137 if ((parent = fts_alloc(sp, "", 0)) == NULL) 138 goto mem2; 139 parent->fts_level = FTS_ROOTPARENTLEVEL; 140 141 /* Allocate/initialize root(s). */ 142 for (root = NULL, nitems = 0; *argv; ++argv, ++nitems) { 143 /* Don't allow zero-length paths. */ 144 if ((len = strlen(*argv)) == 0) { 145 errno = ENOENT; 146 goto mem3; 147 } 148 149 if ((p = fts_alloc(sp, *argv, len)) == NULL) 150 goto mem3; 151 p->fts_level = FTS_ROOTLEVEL; 152 p->fts_parent = parent; 153 p->fts_accpath = p->fts_name; 154 p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW)); 155 156 /* Command-line "." and ".." are real directories. */ 157 if (p->fts_info == FTS_DOT) 158 p->fts_info = FTS_D; 159 160 /* 161 * If comparison routine supplied, traverse in sorted 162 * order; otherwise traverse in the order specified. 163 */ 164 if (compar) { 165 p->fts_link = root; 166 root = p; 167 } else { 168 p->fts_link = NULL; 169 if (root == NULL) 170 tmp = root = p; 171 else { 172 tmp->fts_link = p; 173 tmp = p; 174 } 175 } 176 } 177 if (compar && nitems > 1) 178 root = fts_sort(sp, root, nitems); 179 180 /* 181 * Allocate a dummy pointer and make fts_read think that we've just 182 * finished the node before the root(s); set p->fts_info to FTS_INIT 183 * so that everything about the "current" node is ignored. 184 */ 185 if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL) 186 goto mem3; 187 sp->fts_cur->fts_link = root; 188 sp->fts_cur->fts_info = FTS_INIT; 189 190 /* 191 * If using chdir(2), grab a file descriptor pointing to dot to insure 192 * that we can get back here; this could be avoided for some paths, 193 * but almost certainly not worth the effort. Slashes, symbolic links, 194 * and ".." are all fairly nasty problems. Note, if we can't get the 195 * descriptor we run anyway, just more slowly. 196 */ 197 if (!ISSET(FTS_NOCHDIR)) { 198 if ((sp->fts_rfd = open(".", O_RDONLY, 0)) == -1) 199 SET(FTS_NOCHDIR); 200 else if (fcntl(sp->fts_rfd, F_SETFD, FD_CLOEXEC) == -1) { 201 close(sp->fts_rfd); 202 SET(FTS_NOCHDIR); 203 } 204 } 205 206 if (nitems == 0) 207 fts_free(parent); 208 209 return (sp); 210 211 mem3: fts_lfree(root); 212 fts_free(parent); 213 mem2: free(sp->fts_path); 214 mem1: free(sp); 215 return (NULL); 216 } 217 218 static void 219 fts_load(FTS *sp, FTSENT *p) 220 { 221 size_t len; 222 char *cp; 223 224 _DIAGASSERT(sp != NULL); 225 _DIAGASSERT(p != NULL); 226 227 /* 228 * Load the stream structure for the next traversal. Since we don't 229 * actually enter the directory until after the preorder visit, set 230 * the fts_accpath field specially so the chdir gets done to the right 231 * place and the user can access the first node. From fts_open it's 232 * known that the path will fit. 233 */ 234 len = p->fts_pathlen = p->fts_namelen; 235 memmove(sp->fts_path, p->fts_name, len + 1); 236 if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) { 237 len = strlen(++cp); 238 memmove(p->fts_name, cp, len + 1); 239 p->fts_namelen = len; 240 } 241 p->fts_accpath = p->fts_path = sp->fts_path; 242 sp->fts_dev = p->fts_dev; 243 } 244 245 int 246 fts_close(FTS *sp) 247 { 248 FTSENT *freep, *p; 249 int saved_errno = 0; 250 251 _DIAGASSERT(sp != NULL); 252 253 /* 254 * This still works if we haven't read anything -- the dummy structure 255 * points to the root list, so we step through to the end of the root 256 * list which has a valid parent pointer. 257 */ 258 if (sp->fts_cur) { 259 if (sp->fts_cur->fts_flags & FTS_SYMFOLLOW) 260 (void)close(sp->fts_cur->fts_symfd); 261 for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) { 262 freep = p; 263 p = p->fts_link ? p->fts_link : p->fts_parent; 264 fts_free(freep); 265 } 266 fts_free(p); 267 } 268 269 /* Free up child linked list, sort array, path buffer. */ 270 if (sp->fts_child) 271 fts_lfree(sp->fts_child); 272 if (sp->fts_array) 273 free(sp->fts_array); 274 free(sp->fts_path); 275 276 /* Return to original directory, save errno if necessary. */ 277 if (!ISSET(FTS_NOCHDIR)) { 278 if (fchdir(sp->fts_rfd) == -1) 279 saved_errno = errno; 280 (void)close(sp->fts_rfd); 281 } 282 283 /* Free up the stream pointer. */ 284 free(sp); 285 if (saved_errno) { 286 errno = saved_errno; 287 return -1; 288 } 289 290 return 0; 291 } 292 293 #if !defined(__FTS_COMPAT_TAILINGSLASH) 294 295 /* 296 * Special case of "/" at the end of the path so that slashes aren't 297 * appended which would cause paths to be written as "....//foo". 298 */ 299 #define NAPPEND(p) \ 300 (p->fts_path[p->fts_pathlen - 1] == '/' \ 301 ? p->fts_pathlen - 1 : p->fts_pathlen) 302 303 #else /* !defined(__FTS_COMPAT_TAILINGSLASH) */ 304 305 /* 306 * compatibility with the old behaviour. 307 * 308 * Special case a root of "/" so that slashes aren't appended which would 309 * cause paths to be written as "//foo". 310 */ 311 312 #define NAPPEND(p) \ 313 (p->fts_level == FTS_ROOTLEVEL && p->fts_pathlen == 1 && \ 314 p->fts_path[0] == '/' ? 0 : p->fts_pathlen) 315 316 #endif /* !defined(__FTS_COMPAT_TAILINGSLASH) */ 317 318 FTSENT * 319 fts_read(FTS *sp) 320 { 321 FTSENT *p, *tmp; 322 int instr; 323 char *t; 324 int saved_errno; 325 326 _DIAGASSERT(sp != NULL); 327 328 /* If finished or unrecoverable error, return NULL. */ 329 if (sp->fts_cur == NULL || ISSET(FTS_STOP)) 330 return (NULL); 331 332 /* Set current node pointer. */ 333 p = sp->fts_cur; 334 335 /* Save and zero out user instructions. */ 336 instr = p->fts_instr; 337 p->fts_instr = FTS_NOINSTR; 338 339 /* Any type of file may be re-visited; re-stat and re-turn. */ 340 if (instr == FTS_AGAIN) { 341 p->fts_info = fts_stat(sp, p, 0); 342 return (p); 343 } 344 345 /* 346 * Following a symlink -- SLNONE test allows application to see 347 * SLNONE and recover. If indirecting through a symlink, have 348 * keep a pointer to current location. If unable to get that 349 * pointer, follow fails. 350 */ 351 if (instr == FTS_FOLLOW && 352 (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) { 353 p->fts_info = fts_stat(sp, p, 1); 354 if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) { 355 if ((p->fts_symfd = open(".", O_RDONLY, 0)) == -1) { 356 p->fts_errno = errno; 357 p->fts_info = FTS_ERR; 358 } else if (fcntl(p->fts_symfd, F_SETFD, FD_CLOEXEC) == -1) { 359 p->fts_errno = errno; 360 p->fts_info = FTS_ERR; 361 close(p->fts_symfd); 362 } else 363 p->fts_flags |= FTS_SYMFOLLOW; 364 } 365 return (p); 366 } 367 368 /* Directory in pre-order. */ 369 if (p->fts_info == FTS_D) { 370 /* If skipped or crossed mount point, do post-order visit. */ 371 if (instr == FTS_SKIP || 372 (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) { 373 if (p->fts_flags & FTS_SYMFOLLOW) 374 (void)close(p->fts_symfd); 375 if (sp->fts_child) { 376 fts_lfree(sp->fts_child); 377 sp->fts_child = NULL; 378 } 379 p->fts_info = FTS_DP; 380 return (p); 381 } 382 383 /* Rebuild if only read the names and now traversing. */ 384 if (sp->fts_child && ISSET(FTS_NAMEONLY)) { 385 CLR(FTS_NAMEONLY); 386 fts_lfree(sp->fts_child); 387 sp->fts_child = NULL; 388 } 389 390 /* 391 * Cd to the subdirectory. 392 * 393 * If have already read and now fail to chdir, whack the list 394 * to make the names come out right, and set the parent errno 395 * so the application will eventually get an error condition. 396 * Set the FTS_DONTCHDIR flag so that when we logically change 397 * directories back to the parent we don't do a chdir. 398 * 399 * If haven't read do so. If the read fails, fts_build sets 400 * FTS_STOP or the fts_info field of the node. 401 */ 402 if (sp->fts_child) { 403 if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) { 404 p->fts_errno = errno; 405 p->fts_flags |= FTS_DONTCHDIR; 406 for (p = sp->fts_child; p; p = p->fts_link) 407 p->fts_accpath = 408 p->fts_parent->fts_accpath; 409 } 410 } else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) { 411 if (ISSET(FTS_STOP)) 412 return (NULL); 413 return (p); 414 } 415 p = sp->fts_child; 416 sp->fts_child = NULL; 417 goto name; 418 } 419 420 /* Move to the next node on this level. */ 421 next: tmp = p; 422 if ((p = p->fts_link) != NULL) { 423 fts_free(tmp); 424 425 /* 426 * If reached the top, return to the original directory, and 427 * load the paths for the next root. 428 */ 429 if (p->fts_level == FTS_ROOTLEVEL) { 430 if (FCHDIR(sp, sp->fts_rfd)) { 431 SET(FTS_STOP); 432 return (NULL); 433 } 434 fts_load(sp, p); 435 return (sp->fts_cur = p); 436 } 437 438 /* 439 * User may have called fts_set on the node. If skipped, 440 * ignore. If followed, get a file descriptor so we can 441 * get back if necessary. 442 */ 443 if (p->fts_instr == FTS_SKIP) 444 goto next; 445 if (p->fts_instr == FTS_FOLLOW) { 446 p->fts_info = fts_stat(sp, p, 1); 447 if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) { 448 if ((p->fts_symfd = 449 open(".", O_RDONLY, 0)) == -1) { 450 p->fts_errno = errno; 451 p->fts_info = FTS_ERR; 452 } else if (fcntl(p->fts_symfd, F_SETFD, FD_CLOEXEC) == -1) { 453 p->fts_errno = errno; 454 p->fts_info = FTS_ERR; 455 close(p->fts_symfd); 456 } else 457 p->fts_flags |= FTS_SYMFOLLOW; 458 } 459 p->fts_instr = FTS_NOINSTR; 460 } 461 462 name: t = sp->fts_path + NAPPEND(p->fts_parent); 463 *t++ = '/'; 464 memmove(t, p->fts_name, (size_t)(p->fts_namelen + 1)); 465 return (sp->fts_cur = p); 466 } 467 468 /* Move up to the parent node. */ 469 p = tmp->fts_parent; 470 fts_free(tmp); 471 472 if (p->fts_level == FTS_ROOTPARENTLEVEL) { 473 /* 474 * Done; free everything up and set errno to 0 so the user 475 * can distinguish between error and EOF. 476 */ 477 fts_free(p); 478 errno = 0; 479 return (sp->fts_cur = NULL); 480 } 481 482 /* Nul terminate the pathname. */ 483 sp->fts_path[p->fts_pathlen] = '\0'; 484 485 /* 486 * Return to the parent directory. If at a root node or came through 487 * a symlink, go back through the file descriptor. Otherwise, cd up 488 * one directory. 489 */ 490 if (p->fts_level == FTS_ROOTLEVEL) { 491 if (FCHDIR(sp, sp->fts_rfd)) { 492 SET(FTS_STOP); 493 return (NULL); 494 } 495 } else if (p->fts_flags & FTS_SYMFOLLOW) { 496 if (FCHDIR(sp, p->fts_symfd)) { 497 saved_errno = errno; 498 (void)close(p->fts_symfd); 499 errno = saved_errno; 500 SET(FTS_STOP); 501 return (NULL); 502 } 503 (void)close(p->fts_symfd); 504 } else if (!(p->fts_flags & FTS_DONTCHDIR) && 505 fts_safe_changedir(sp, p->fts_parent, -1, "..")) { 506 SET(FTS_STOP); 507 return (NULL); 508 } 509 p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP; 510 return (sp->fts_cur = p); 511 } 512 513 /* 514 * Fts_set takes the stream as an argument although it's not used in this 515 * implementation; it would be necessary if anyone wanted to add global 516 * semantics to fts using fts_set. An error return is allowed for similar 517 * reasons. 518 */ 519 /* ARGSUSED */ 520 int 521 fts_set(FTS *sp, FTSENT *p, int instr) 522 { 523 524 _DIAGASSERT(sp != NULL); 525 _DIAGASSERT(p != NULL); 526 527 if (instr && instr != FTS_AGAIN && instr != FTS_FOLLOW && 528 instr != FTS_NOINSTR && instr != FTS_SKIP) { 529 errno = EINVAL; 530 return (1); 531 } 532 p->fts_instr = instr; 533 return (0); 534 } 535 536 FTSENT * 537 fts_children(FTS *sp, int instr) 538 { 539 FTSENT *p; 540 int fd; 541 542 _DIAGASSERT(sp != NULL); 543 544 if (instr && instr != FTS_NAMEONLY) { 545 errno = EINVAL; 546 return (NULL); 547 } 548 549 /* Set current node pointer. */ 550 p = sp->fts_cur; 551 552 /* 553 * Errno set to 0 so user can distinguish empty directory from 554 * an error. 555 */ 556 errno = 0; 557 558 /* Fatal errors stop here. */ 559 if (ISSET(FTS_STOP)) 560 return (NULL); 561 562 /* Return logical hierarchy of user's arguments. */ 563 if (p->fts_info == FTS_INIT) 564 return (p->fts_link); 565 566 /* 567 * If not a directory being visited in pre-order, stop here. Could 568 * allow FTS_DNR, assuming the user has fixed the problem, but the 569 * same effect is available with FTS_AGAIN. 570 */ 571 if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */) 572 return (NULL); 573 574 /* Free up any previous child list. */ 575 if (sp->fts_child) 576 fts_lfree(sp->fts_child); 577 578 if (instr == FTS_NAMEONLY) { 579 SET(FTS_NAMEONLY); 580 instr = BNAMES; 581 } else 582 instr = BCHILD; 583 584 /* 585 * If using chdir on a relative path and called BEFORE fts_read does 586 * its chdir to the root of a traversal, we can lose -- we need to 587 * chdir into the subdirectory, and we don't know where the current 588 * directory is, so we can't get back so that the upcoming chdir by 589 * fts_read will work. 590 */ 591 if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' || 592 ISSET(FTS_NOCHDIR)) 593 return (sp->fts_child = fts_build(sp, instr)); 594 595 if ((fd = open(".", O_RDONLY, 0)) == -1) 596 return (sp->fts_child = NULL); 597 sp->fts_child = fts_build(sp, instr); 598 if (fchdir(fd)) { 599 (void)close(fd); 600 return (NULL); 601 } 602 (void)close(fd); 603 return (sp->fts_child); 604 } 605 606 /* 607 * This is the tricky part -- do not casually change *anything* in here. The 608 * idea is to build the linked list of entries that are used by fts_children 609 * and fts_read. There are lots of special cases. 610 * 611 * The real slowdown in walking the tree is the stat calls. If FTS_NOSTAT is 612 * set and it's a physical walk (so that symbolic links can't be directories), 613 * we can do things quickly. First, if it's a 4.4BSD file system, the type 614 * of the file is in the directory entry. Otherwise, we assume that the number 615 * of subdirectories in a node is equal to the number of links to the parent. 616 * The former skips all stat calls. The latter skips stat calls in any leaf 617 * directories and for any files after the subdirectories in the directory have 618 * been found, cutting the stat calls by about 2/3. 619 */ 620 static FTSENT * 621 fts_build(FTS *sp, int type) 622 { 623 struct dirent *dp; 624 FTSENT *p, *head; 625 size_t nitems; 626 FTSENT *cur, *tail; 627 DIR *dirp; 628 void *oldaddr; 629 size_t dnamlen; 630 int cderrno, descend, level, nlinks, saved_errno, nostat, doadjust; 631 size_t len, maxlen; 632 #ifdef FTS_WHITEOUT 633 int oflag; 634 #endif 635 char *cp = NULL; /* pacify gcc */ 636 637 _DIAGASSERT(sp != NULL); 638 639 /* Set current node pointer. */ 640 cur = sp->fts_cur; 641 642 /* 643 * Open the directory for reading. If this fails, we're done. 644 * If being called from fts_read, set the fts_info field. 645 */ 646 #ifdef FTS_WHITEOUT 647 if (ISSET(FTS_WHITEOUT)) 648 oflag = DTF_NODUP|DTF_REWIND; 649 else 650 oflag = DTF_HIDEW|DTF_NODUP|DTF_REWIND; 651 #else 652 #define __opendir2(path, flag) opendir(path) 653 #endif 654 if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) { 655 if (type == BREAD) { 656 cur->fts_info = FTS_DNR; 657 cur->fts_errno = errno; 658 } 659 return (NULL); 660 } 661 662 /* 663 * Nlinks is the number of possible entries of type directory in the 664 * directory if we're cheating on stat calls, 0 if we're not doing 665 * any stat calls at all, -1 if we're doing stats on everything. 666 */ 667 if (type == BNAMES) { 668 nlinks = 0; 669 nostat = 1; 670 } else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) { 671 nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2); 672 nostat = 1; 673 } else { 674 nlinks = -1; 675 nostat = 0; 676 } 677 678 #ifdef notdef 679 (void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink); 680 (void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n", 681 ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT)); 682 #endif 683 /* 684 * If we're going to need to stat anything or we want to descend 685 * and stay in the directory, chdir. If this fails we keep going, 686 * but set a flag so we don't chdir after the post-order visit. 687 * We won't be able to stat anything, but we can still return the 688 * names themselves. Note, that since fts_read won't be able to 689 * chdir into the directory, it will have to return different path 690 * names than before, i.e. "a/b" instead of "b". Since the node 691 * has already been visited in pre-order, have to wait until the 692 * post-order visit to return the error. There is a special case 693 * here, if there was nothing to stat then it's not an error to 694 * not be able to stat. This is all fairly nasty. If a program 695 * needed sorted entries or stat information, they had better be 696 * checking FTS_NS on the returned nodes. 697 */ 698 cderrno = 0; 699 if (nlinks || type == BREAD) { 700 if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) { 701 if (nlinks && type == BREAD) 702 cur->fts_errno = errno; 703 cur->fts_flags |= FTS_DONTCHDIR; 704 descend = 0; 705 cderrno = errno; 706 } else 707 descend = 1; 708 } else 709 descend = 0; 710 711 /* 712 * Figure out the max file name length that can be stored in the 713 * current path -- the inner loop allocates more path as necessary. 714 * We really wouldn't have to do the maxlen calculations here, we 715 * could do them in fts_read before returning the path, but it's a 716 * lot easier here since the length is part of the dirent structure. 717 * 718 * If not changing directories set a pointer so that can just append 719 * each new name into the path. 720 */ 721 len = NAPPEND(cur); 722 if (ISSET(FTS_NOCHDIR)) { 723 cp = sp->fts_path + len; 724 *cp++ = '/'; 725 } 726 len++; 727 maxlen = sp->fts_pathlen - len; 728 729 if (cur->fts_level == SHRT_MAX) { 730 (void)closedir(dirp); 731 cur->fts_info = FTS_ERR; 732 SET(FTS_STOP); 733 errno = ENAMETOOLONG; 734 return (NULL); 735 } 736 737 level = cur->fts_level + 1; 738 739 /* Read the directory, attaching each entry to the `link' pointer. */ 740 doadjust = 0; 741 for (head = tail = NULL, nitems = 0; (dp = readdir(dirp)) != NULL;) { 742 743 if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name)) 744 continue; 745 746 #if defined(HAVE_STRUCT_DIRENT_D_NAMLEN) 747 dnamlen = dp->d_namlen; 748 #else 749 dnamlen = strlen(dp->d_name); 750 #endif 751 if ((p = fts_alloc(sp, dp->d_name, dnamlen)) == NULL) 752 goto mem1; 753 if (dnamlen >= maxlen) { /* include space for NUL */ 754 oldaddr = sp->fts_path; 755 if (fts_palloc(sp, dnamlen + len + 1)) { 756 /* 757 * No more memory for path or structures. Save 758 * errno, free up the current structure and the 759 * structures already allocated. 760 */ 761 mem1: saved_errno = errno; 762 if (p) 763 fts_free(p); 764 fts_lfree(head); 765 (void)closedir(dirp); 766 errno = saved_errno; 767 cur->fts_info = FTS_ERR; 768 SET(FTS_STOP); 769 return (NULL); 770 } 771 /* Did realloc() change the pointer? */ 772 if (oldaddr != sp->fts_path) { 773 doadjust = 1; 774 if (ISSET(FTS_NOCHDIR)) 775 cp = sp->fts_path + len; 776 } 777 maxlen = sp->fts_pathlen - len; 778 } 779 780 #if defined(__FTS_COMPAT_LENGTH) 781 if (len + dnamlen >= USHRT_MAX) { 782 /* 783 * In an FTSENT, fts_pathlen is an unsigned short 784 * so it is possible to wraparound here. 785 * If we do, free up the current structure and the 786 * structures already allocated, then error out 787 * with ENAMETOOLONG. 788 */ 789 fts_free(p); 790 fts_lfree(head); 791 (void)closedir(dirp); 792 cur->fts_info = FTS_ERR; 793 SET(FTS_STOP); 794 errno = ENAMETOOLONG; 795 return (NULL); 796 } 797 #endif 798 p->fts_level = level; 799 p->fts_pathlen = len + dnamlen; 800 p->fts_parent = sp->fts_cur; 801 802 #ifdef FTS_WHITEOUT 803 if (dp->d_type == DT_WHT) 804 p->fts_flags |= FTS_ISW; 805 #endif 806 807 if (cderrno) { 808 if (nlinks) { 809 p->fts_info = FTS_NS; 810 p->fts_errno = cderrno; 811 } else 812 p->fts_info = FTS_NSOK; 813 p->fts_accpath = cur->fts_accpath; 814 } else if (nlinks == 0 815 #ifdef DT_DIR 816 || (nostat && 817 dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN) 818 #endif 819 ) { 820 p->fts_accpath = 821 ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name; 822 p->fts_info = FTS_NSOK; 823 } else { 824 /* Build a file name for fts_stat to stat. */ 825 if (ISSET(FTS_NOCHDIR)) { 826 p->fts_accpath = p->fts_path; 827 memmove(cp, p->fts_name, 828 (size_t)(p->fts_namelen + 1)); 829 } else 830 p->fts_accpath = p->fts_name; 831 /* Stat it. */ 832 p->fts_info = fts_stat(sp, p, 0); 833 834 /* Decrement link count if applicable. */ 835 if (nlinks > 0 && (p->fts_info == FTS_D || 836 p->fts_info == FTS_DC || p->fts_info == FTS_DOT)) 837 --nlinks; 838 } 839 840 /* We walk in directory order so "ls -f" doesn't get upset. */ 841 p->fts_link = NULL; 842 if (head == NULL) 843 head = tail = p; 844 else { 845 tail->fts_link = p; 846 tail = p; 847 } 848 ++nitems; 849 } 850 (void)closedir(dirp); 851 852 /* 853 * If had to realloc the path, adjust the addresses for the rest 854 * of the tree. 855 */ 856 if (doadjust) 857 fts_padjust(sp, head); 858 859 /* 860 * If not changing directories, reset the path back to original 861 * state. 862 */ 863 if (ISSET(FTS_NOCHDIR)) { 864 if (len == sp->fts_pathlen || nitems == 0) 865 --cp; 866 *cp = '\0'; 867 } 868 869 /* 870 * If descended after called from fts_children or after called from 871 * fts_read and nothing found, get back. At the root level we use 872 * the saved fd; if one of fts_open()'s arguments is a relative path 873 * to an empty directory, we wind up here with no other way back. If 874 * can't get back, we're done. 875 */ 876 if (descend && (type == BCHILD || !nitems) && 877 (cur->fts_level == FTS_ROOTLEVEL ? 878 FCHDIR(sp, sp->fts_rfd) : 879 fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) { 880 cur->fts_info = FTS_ERR; 881 SET(FTS_STOP); 882 return (NULL); 883 } 884 885 /* If didn't find anything, return NULL. */ 886 if (!nitems) { 887 if (type == BREAD) 888 cur->fts_info = FTS_DP; 889 return (NULL); 890 } 891 892 /* Sort the entries. */ 893 if (sp->fts_compar && nitems > 1) 894 head = fts_sort(sp, head, nitems); 895 return (head); 896 } 897 898 static unsigned short 899 fts_stat(FTS *sp, FTSENT *p, int follow) 900 { 901 FTSENT *t; 902 dev_t dev; 903 __fts_ino_t ino; 904 __fts_stat_t *sbp, sb; 905 int saved_errno; 906 907 _DIAGASSERT(sp != NULL); 908 _DIAGASSERT(p != NULL); 909 910 /* If user needs stat info, stat buffer already allocated. */ 911 sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp; 912 913 #ifdef FTS_WHITEOUT 914 /* check for whiteout */ 915 if (p->fts_flags & FTS_ISW) { 916 if (sbp != &sb) { 917 memset(sbp, '\0', sizeof (*sbp)); 918 sbp->st_mode = S_IFWHT; 919 } 920 return (FTS_W); 921 } 922 #endif 923 924 /* 925 * If doing a logical walk, or application requested FTS_FOLLOW, do 926 * a stat(2). If that fails, check for a non-existent symlink. If 927 * fail, set the errno from the stat call. 928 */ 929 if (ISSET(FTS_LOGICAL) || follow) { 930 if (stat(p->fts_accpath, sbp)) { 931 saved_errno = errno; 932 if (!lstat(p->fts_accpath, sbp)) { 933 errno = 0; 934 return (FTS_SLNONE); 935 } 936 p->fts_errno = saved_errno; 937 goto err; 938 } 939 } else if (lstat(p->fts_accpath, sbp)) { 940 p->fts_errno = errno; 941 err: memset(sbp, 0, sizeof(*sbp)); 942 return (FTS_NS); 943 } 944 945 if (S_ISDIR(sbp->st_mode)) { 946 /* 947 * Set the device/inode. Used to find cycles and check for 948 * crossing mount points. Also remember the link count, used 949 * in fts_build to limit the number of stat calls. It is 950 * understood that these fields are only referenced if fts_info 951 * is set to FTS_D. 952 */ 953 dev = p->fts_dev = sbp->st_dev; 954 ino = p->fts_ino = sbp->st_ino; 955 p->fts_nlink = sbp->st_nlink; 956 957 if (ISDOT(p->fts_name)) 958 return (FTS_DOT); 959 960 /* 961 * Cycle detection is done by brute force when the directory 962 * is first encountered. If the tree gets deep enough or the 963 * number of symbolic links to directories is high enough, 964 * something faster might be worthwhile. 965 */ 966 for (t = p->fts_parent; 967 t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent) 968 if (ino == t->fts_ino && dev == t->fts_dev) { 969 p->fts_cycle = t; 970 return (FTS_DC); 971 } 972 return (FTS_D); 973 } 974 if (S_ISLNK(sbp->st_mode)) 975 return (FTS_SL); 976 if (S_ISREG(sbp->st_mode)) 977 return (FTS_F); 978 return (FTS_DEFAULT); 979 } 980 981 static FTSENT * 982 fts_sort(FTS *sp, FTSENT *head, size_t nitems) 983 { 984 FTSENT **ap, *p; 985 986 _DIAGASSERT(sp != NULL); 987 _DIAGASSERT(head != NULL); 988 989 /* 990 * Construct an array of pointers to the structures and call qsort(3). 991 * Reassemble the array in the order returned by qsort. If unable to 992 * sort for memory reasons, return the directory entries in their 993 * current order. Allocate enough space for the current needs plus 994 * 40 so don't realloc one entry at a time. 995 */ 996 if (nitems > sp->fts_nitems) { 997 FTSENT **new; 998 999 new = realloc(sp->fts_array, sizeof(FTSENT *) * (nitems + 40)); 1000 if (new == 0) 1001 return (head); 1002 sp->fts_array = new; 1003 sp->fts_nitems = nitems + 40; 1004 } 1005 for (ap = sp->fts_array, p = head; p; p = p->fts_link) 1006 *ap++ = p; 1007 qsort((void *)sp->fts_array, nitems, sizeof(FTSENT *), 1008 (int (*)(const void *, const void *))sp->fts_compar); 1009 for (head = *(ap = sp->fts_array); --nitems; ++ap) 1010 ap[0]->fts_link = ap[1]; 1011 ap[0]->fts_link = NULL; 1012 return (head); 1013 } 1014 1015 static FTSENT * 1016 fts_alloc(FTS *sp, const char *name, size_t namelen) 1017 { 1018 FTSENT *p; 1019 #if defined(FTS_ALLOC_ALIGNED) 1020 size_t len; 1021 #endif 1022 1023 _DIAGASSERT(sp != NULL); 1024 _DIAGASSERT(name != NULL); 1025 1026 #if defined(FTS_ALLOC_ALIGNED) 1027 /* 1028 * The file name is a variable length array and no stat structure is 1029 * necessary if the user has set the nostat bit. Allocate the FTSENT 1030 * structure, the file name and the stat structure in one chunk, but 1031 * be careful that the stat structure is reasonably aligned. Since the 1032 * fts_name field is declared to be of size 1, the fts_name pointer is 1033 * namelen + 2 before the first possible address of the stat structure. 1034 */ 1035 len = sizeof(FTSENT) + namelen; 1036 if (!ISSET(FTS_NOSTAT)) 1037 len += sizeof(*(p->fts_statp)) + ALIGNBYTES; 1038 if ((p = malloc(len)) == NULL) 1039 return (NULL); 1040 1041 if (!ISSET(FTS_NOSTAT)) 1042 p->fts_statp = (__fts_stat_t *)ALIGN( 1043 (unsigned long)(p->fts_name + namelen + 2)); 1044 #else 1045 if ((p = malloc(sizeof(FTSENT) + namelen)) == NULL) 1046 return (NULL); 1047 1048 if (!ISSET(FTS_NOSTAT)) 1049 if ((p->fts_statp = malloc(sizeof(*(p->fts_statp)))) == NULL) { 1050 free(p); 1051 return (NULL); 1052 } 1053 #endif 1054 1055 /* Copy the name plus the trailing NULL. */ 1056 memmove(p->fts_name, name, namelen + 1); 1057 1058 p->fts_namelen = namelen; 1059 p->fts_path = sp->fts_path; 1060 p->fts_errno = 0; 1061 p->fts_flags = 0; 1062 p->fts_instr = FTS_NOINSTR; 1063 p->fts_number = 0; 1064 p->fts_pointer = NULL; 1065 return (p); 1066 } 1067 1068 static void 1069 fts_free(FTSENT *p) 1070 { 1071 #if !defined(FTS_ALLOC_ALIGNED) 1072 if (p->fts_statp) 1073 free(p->fts_statp); 1074 #endif 1075 free(p); 1076 } 1077 1078 static void 1079 fts_lfree(FTSENT *head) 1080 { 1081 FTSENT *p; 1082 1083 /* XXX: head may be NULL ? */ 1084 1085 /* Free a linked list of structures. */ 1086 while ((p = head) != NULL) { 1087 head = head->fts_link; 1088 fts_free(p); 1089 } 1090 } 1091 1092 static size_t 1093 fts_pow2(size_t x) 1094 { 1095 1096 x--; 1097 x |= x>>1; 1098 x |= x>>2; 1099 x |= x>>4; 1100 x |= x>>8; 1101 x |= x>>16; 1102 #if LONG_BIT > 32 1103 x |= x>>32; 1104 #endif 1105 #if LONG_BIT > 64 1106 x |= x>>64; 1107 #endif 1108 x++; 1109 return (x); 1110 } 1111 1112 /* 1113 * Allow essentially unlimited paths; find, rm, ls should all work on any tree. 1114 * Most systems will allow creation of paths much longer than MAXPATHLEN, even 1115 * though the kernel won't resolve them. Round up the new size to a power of 2, 1116 * so we don't realloc the path 2 bytes at a time. 1117 */ 1118 static int 1119 fts_palloc(FTS *sp, size_t size) 1120 { 1121 char *new; 1122 1123 _DIAGASSERT(sp != NULL); 1124 1125 #ifdef __FTS_COMPAT_LENGTH 1126 /* Protect against fts_pathlen overflow. */ 1127 if (size > USHRT_MAX + 1) { 1128 errno = ENAMETOOLONG; 1129 return (1); 1130 } 1131 #endif 1132 size = fts_pow2(size); 1133 new = realloc(sp->fts_path, size); 1134 if (new == 0) 1135 return (1); 1136 sp->fts_path = new; 1137 sp->fts_pathlen = size; 1138 return (0); 1139 } 1140 1141 /* 1142 * When the path is realloc'd, have to fix all of the pointers in structures 1143 * already returned. 1144 */ 1145 static void 1146 fts_padjust(FTS *sp, FTSENT *head) 1147 { 1148 FTSENT *p; 1149 char *addr; 1150 1151 _DIAGASSERT(sp != NULL); 1152 1153 #define ADJUST(p) do { \ 1154 if ((p)->fts_accpath != (p)->fts_name) \ 1155 (p)->fts_accpath = \ 1156 addr + ((p)->fts_accpath - (p)->fts_path); \ 1157 (p)->fts_path = addr; \ 1158 } while (/*CONSTCOND*/0) 1159 1160 addr = sp->fts_path; 1161 1162 /* Adjust the current set of children. */ 1163 for (p = sp->fts_child; p; p = p->fts_link) 1164 ADJUST(p); 1165 1166 /* Adjust the rest of the tree, including the current level. */ 1167 for (p = head; p->fts_level >= FTS_ROOTLEVEL;) { 1168 ADJUST(p); 1169 p = p->fts_link ? p->fts_link : p->fts_parent; 1170 } 1171 } 1172 1173 static size_t 1174 fts_maxarglen(char * const *argv) 1175 { 1176 size_t len, max; 1177 1178 _DIAGASSERT(argv != NULL); 1179 1180 for (max = 0; *argv; ++argv) 1181 if ((len = strlen(*argv)) > max) 1182 max = len; 1183 return (max + 1); 1184 } 1185 1186 /* 1187 * Change to dir specified by fd or p->fts_accpath without getting 1188 * tricked by someone changing the world out from underneath us. 1189 * Assumes p->fts_dev and p->fts_ino are filled in. 1190 */ 1191 static int 1192 fts_safe_changedir(const FTS *sp, const FTSENT *p, int fd, const char *path) 1193 { 1194 int oldfd = fd, ret = -1; 1195 __fts_stat_t sb; 1196 1197 if (ISSET(FTS_NOCHDIR)) 1198 return 0; 1199 1200 if (oldfd < 0 && (fd = open(path, O_RDONLY)) == -1) 1201 return -1; 1202 1203 if (fstat(fd, &sb) == -1) 1204 goto bail; 1205 1206 if (sb.st_ino != p->fts_ino || sb.st_dev != p->fts_dev) { 1207 errno = ENOENT; 1208 goto bail; 1209 } 1210 1211 ret = fchdir(fd); 1212 1213 bail: 1214 if (oldfd < 0) { 1215 int save_errno = errno; 1216 (void)close(fd); 1217 errno = save_errno; 1218 } 1219 return ret; 1220 } 1221