xref: /freebsd/lib/libc/gen/fts.c (revision 0b57cec536236d46e3dba9bd041533462f33dbb7)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
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  * $OpenBSD: fts.c,v 1.22 1999/10/03 19:22:22 millert Exp $
32  */
33 
34 #include <sys/cdefs.h>
35 __SCCSID("@(#)fts.c	8.6 (Berkeley) 8/14/94");
36 __FBSDID("$FreeBSD$");
37 
38 #include "namespace.h"
39 #include <sys/param.h>
40 #include <sys/mount.h>
41 #include <sys/stat.h>
42 
43 #include <dirent.h>
44 #include <errno.h>
45 #include <fcntl.h>
46 #include <fts.h>
47 #include <stdlib.h>
48 #include <string.h>
49 #include <unistd.h>
50 #include "un-namespace.h"
51 
52 #include "gen-private.h"
53 
54 static FTSENT	*fts_alloc(FTS *, char *, size_t);
55 static FTSENT	*fts_build(FTS *, int);
56 static void	 fts_lfree(FTSENT *);
57 static void	 fts_load(FTS *, FTSENT *);
58 static size_t	 fts_maxarglen(char * const *);
59 static void	 fts_padjust(FTS *, FTSENT *);
60 static int	 fts_palloc(FTS *, size_t);
61 static FTSENT	*fts_sort(FTS *, FTSENT *, size_t);
62 static int	 fts_stat(FTS *, FTSENT *, int, int);
63 static int	 fts_safe_changedir(FTS *, FTSENT *, int, char *);
64 static int	 fts_ufslinks(FTS *, const FTSENT *);
65 
66 #define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
67 
68 #define	CLR(opt)	(sp->fts_options &= ~(opt))
69 #define	ISSET(opt)	(sp->fts_options & (opt))
70 #define	SET(opt)	(sp->fts_options |= (opt))
71 
72 #define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
73 
74 /* fts_build flags */
75 #define	BCHILD		1		/* fts_children */
76 #define	BNAMES		2		/* fts_children, names only */
77 #define	BREAD		3		/* fts_read */
78 
79 /*
80  * Internal representation of an FTS, including extra implementation
81  * details.  The FTS returned from fts_open points to this structure's
82  * ftsp_fts member (and can be cast to an _fts_private as required)
83  */
84 struct _fts_private {
85 	FTS		ftsp_fts;
86 	struct statfs	ftsp_statfs;
87 	dev_t		ftsp_dev;
88 	int		ftsp_linksreliable;
89 };
90 
91 /*
92  * The "FTS_NOSTAT" option can avoid a lot of calls to stat(2) if it
93  * knows that a directory could not possibly have subdirectories.  This
94  * is decided by looking at the link count: a subdirectory would
95  * increment its parent's link count by virtue of its own ".." entry.
96  * This assumption only holds for UFS-like filesystems that implement
97  * links and directories this way, so we must punt for others.
98  */
99 
100 static const char *ufslike_filesystems[] = {
101 	"ufs",
102 	"zfs",
103 	"nfs",
104 	"ext2fs",
105 	0
106 };
107 
108 FTS *
109 fts_open(char * const *argv, int options,
110     int (*compar)(const FTSENT * const *, const FTSENT * const *))
111 {
112 	struct _fts_private *priv;
113 	FTS *sp;
114 	FTSENT *p, *root;
115 	FTSENT *parent, *tmp;
116 	size_t len, nitems;
117 
118 	/* Options check. */
119 	if (options & ~FTS_OPTIONMASK) {
120 		errno = EINVAL;
121 		return (NULL);
122 	}
123 
124 	/* fts_open() requires at least one path */
125 	if (*argv == NULL) {
126 		errno = EINVAL;
127 		return (NULL);
128 	}
129 
130 	/* Allocate/initialize the stream. */
131 	if ((priv = calloc(1, sizeof(*priv))) == NULL)
132 		return (NULL);
133 	sp = &priv->ftsp_fts;
134 	sp->fts_compar = compar;
135 	sp->fts_options = options;
136 
137 	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
138 	if (ISSET(FTS_LOGICAL))
139 		SET(FTS_NOCHDIR);
140 
141 	/*
142 	 * Start out with 1K of path space, and enough, in any case,
143 	 * to hold the user's paths.
144 	 */
145 	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
146 		goto mem1;
147 
148 	/* Allocate/initialize root's parent. */
149 	if ((parent = fts_alloc(sp, "", 0)) == NULL)
150 		goto mem2;
151 	parent->fts_level = FTS_ROOTPARENTLEVEL;
152 
153 	/* Shush, GCC. */
154 	tmp = NULL;
155 
156 	/* Allocate/initialize root(s). */
157 	for (root = NULL, nitems = 0; *argv != NULL; ++argv, ++nitems) {
158 		len = strlen(*argv);
159 
160 		p = fts_alloc(sp, *argv, len);
161 		p->fts_level = FTS_ROOTLEVEL;
162 		p->fts_parent = parent;
163 		p->fts_accpath = p->fts_name;
164 		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW), -1);
165 
166 		/* Command-line "." and ".." are real directories. */
167 		if (p->fts_info == FTS_DOT)
168 			p->fts_info = FTS_D;
169 
170 		/*
171 		 * If comparison routine supplied, traverse in sorted
172 		 * order; otherwise traverse in the order specified.
173 		 */
174 		if (compar) {
175 			p->fts_link = root;
176 			root = p;
177 		} else {
178 			p->fts_link = NULL;
179 			if (root == NULL)
180 				tmp = root = p;
181 			else {
182 				tmp->fts_link = p;
183 				tmp = p;
184 			}
185 		}
186 	}
187 	if (compar && nitems > 1)
188 		root = fts_sort(sp, root, nitems);
189 
190 	/*
191 	 * Allocate a dummy pointer and make fts_read think that we've just
192 	 * finished the node before the root(s); set p->fts_info to FTS_INIT
193 	 * so that everything about the "current" node is ignored.
194 	 */
195 	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
196 		goto mem3;
197 	sp->fts_cur->fts_link = root;
198 	sp->fts_cur->fts_info = FTS_INIT;
199 
200 	/*
201 	 * If using chdir(2), grab a file descriptor pointing to dot to ensure
202 	 * that we can get back here; this could be avoided for some paths,
203 	 * but almost certainly not worth the effort.  Slashes, symbolic links,
204 	 * and ".." are all fairly nasty problems.  Note, if we can't get the
205 	 * descriptor we run anyway, just more slowly.
206 	 */
207 	if (!ISSET(FTS_NOCHDIR) &&
208 	    (sp->fts_rfd = _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0)
209 		SET(FTS_NOCHDIR);
210 
211 	return (sp);
212 
213 mem3:	fts_lfree(root);
214 	free(parent);
215 mem2:	free(sp->fts_path);
216 mem1:	free(sp);
217 	return (NULL);
218 }
219 
220 static void
221 fts_load(FTS *sp, FTSENT *p)
222 {
223 	size_t len;
224 	char *cp;
225 
226 	/*
227 	 * Load the stream structure for the next traversal.  Since we don't
228 	 * actually enter the directory until after the preorder visit, set
229 	 * the fts_accpath field specially so the chdir gets done to the right
230 	 * place and the user can access the first node.  From fts_open it's
231 	 * known that the path will fit.
232 	 */
233 	len = p->fts_pathlen = p->fts_namelen;
234 	memmove(sp->fts_path, p->fts_name, len + 1);
235 	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
236 		len = strlen(++cp);
237 		memmove(p->fts_name, cp, len + 1);
238 		p->fts_namelen = len;
239 	}
240 	p->fts_accpath = p->fts_path = sp->fts_path;
241 	sp->fts_dev = p->fts_dev;
242 }
243 
244 int
245 fts_close(FTS *sp)
246 {
247 	FTSENT *freep, *p;
248 	int saved_errno;
249 
250 	/*
251 	 * This still works if we haven't read anything -- the dummy structure
252 	 * points to the root list, so we step through to the end of the root
253 	 * list which has a valid parent pointer.
254 	 */
255 	if (sp->fts_cur) {
256 		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
257 			freep = p;
258 			p = p->fts_link != NULL ? p->fts_link : p->fts_parent;
259 			free(freep);
260 		}
261 		free(p);
262 	}
263 
264 	/* Free up child linked list, sort array, path buffer. */
265 	if (sp->fts_child)
266 		fts_lfree(sp->fts_child);
267 	if (sp->fts_array)
268 		free(sp->fts_array);
269 	free(sp->fts_path);
270 
271 	/* Return to original directory, save errno if necessary. */
272 	if (!ISSET(FTS_NOCHDIR)) {
273 		saved_errno = fchdir(sp->fts_rfd) ? errno : 0;
274 		(void)_close(sp->fts_rfd);
275 
276 		/* Set errno and return. */
277 		if (saved_errno != 0) {
278 			/* Free up the stream pointer. */
279 			free(sp);
280 			errno = saved_errno;
281 			return (-1);
282 		}
283 	}
284 
285 	/* Free up the stream pointer. */
286 	free(sp);
287 	return (0);
288 }
289 
290 /*
291  * Special case of "/" at the end of the path so that slashes aren't
292  * appended which would cause paths to be written as "....//foo".
293  */
294 #define	NAPPEND(p)							\
295 	(p->fts_path[p->fts_pathlen - 1] == '/'				\
296 	    ? p->fts_pathlen - 1 : p->fts_pathlen)
297 
298 FTSENT *
299 fts_read(FTS *sp)
300 {
301 	FTSENT *p, *tmp;
302 	int instr;
303 	char *t;
304 	int saved_errno;
305 
306 	/* If finished or unrecoverable error, return NULL. */
307 	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
308 		return (NULL);
309 
310 	/* Set current node pointer. */
311 	p = sp->fts_cur;
312 
313 	/* Save and zero out user instructions. */
314 	instr = p->fts_instr;
315 	p->fts_instr = FTS_NOINSTR;
316 
317 	/* Any type of file may be re-visited; re-stat and re-turn. */
318 	if (instr == FTS_AGAIN) {
319 		p->fts_info = fts_stat(sp, p, 0, -1);
320 		return (p);
321 	}
322 
323 	/*
324 	 * Following a symlink -- SLNONE test allows application to see
325 	 * SLNONE and recover.  If indirecting through a symlink, have
326 	 * keep a pointer to current location.  If unable to get that
327 	 * pointer, follow fails.
328 	 */
329 	if (instr == FTS_FOLLOW &&
330 	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
331 		p->fts_info = fts_stat(sp, p, 1, -1);
332 		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
333 			if ((p->fts_symfd = _open(".", O_RDONLY | O_CLOEXEC,
334 			    0)) < 0) {
335 				p->fts_errno = errno;
336 				p->fts_info = FTS_ERR;
337 			} else
338 				p->fts_flags |= FTS_SYMFOLLOW;
339 		}
340 		return (p);
341 	}
342 
343 	/* Directory in pre-order. */
344 	if (p->fts_info == FTS_D) {
345 		/* If skipped or crossed mount point, do post-order visit. */
346 		if (instr == FTS_SKIP ||
347 		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
348 			if (p->fts_flags & FTS_SYMFOLLOW)
349 				(void)_close(p->fts_symfd);
350 			if (sp->fts_child) {
351 				fts_lfree(sp->fts_child);
352 				sp->fts_child = NULL;
353 			}
354 			p->fts_info = FTS_DP;
355 			return (p);
356 		}
357 
358 		/* Rebuild if only read the names and now traversing. */
359 		if (sp->fts_child != NULL && ISSET(FTS_NAMEONLY)) {
360 			CLR(FTS_NAMEONLY);
361 			fts_lfree(sp->fts_child);
362 			sp->fts_child = NULL;
363 		}
364 
365 		/*
366 		 * Cd to the subdirectory.
367 		 *
368 		 * If have already read and now fail to chdir, whack the list
369 		 * to make the names come out right, and set the parent errno
370 		 * so the application will eventually get an error condition.
371 		 * Set the FTS_DONTCHDIR flag so that when we logically change
372 		 * directories back to the parent we don't do a chdir.
373 		 *
374 		 * If haven't read do so.  If the read fails, fts_build sets
375 		 * FTS_STOP or the fts_info field of the node.
376 		 */
377 		if (sp->fts_child != NULL) {
378 			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
379 				p->fts_errno = errno;
380 				p->fts_flags |= FTS_DONTCHDIR;
381 				for (p = sp->fts_child; p != NULL;
382 				    p = p->fts_link)
383 					p->fts_accpath =
384 					    p->fts_parent->fts_accpath;
385 			}
386 		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
387 			if (ISSET(FTS_STOP))
388 				return (NULL);
389 			return (p);
390 		}
391 		p = sp->fts_child;
392 		sp->fts_child = NULL;
393 		goto name;
394 	}
395 
396 	/* Move to the next node on this level. */
397 next:	tmp = p;
398 	if ((p = p->fts_link) != NULL) {
399 		/*
400 		 * If reached the top, return to the original directory (or
401 		 * the root of the tree), and load the paths for the next root.
402 		 */
403 		if (p->fts_level == FTS_ROOTLEVEL) {
404 			if (FCHDIR(sp, sp->fts_rfd)) {
405 				SET(FTS_STOP);
406 				return (NULL);
407 			}
408 			free(tmp);
409 			fts_load(sp, p);
410 			return (sp->fts_cur = p);
411 		}
412 
413 		/*
414 		 * User may have called fts_set on the node.  If skipped,
415 		 * ignore.  If followed, get a file descriptor so we can
416 		 * get back if necessary.
417 		 */
418 		if (p->fts_instr == FTS_SKIP) {
419 			free(tmp);
420 			goto next;
421 		}
422 		if (p->fts_instr == FTS_FOLLOW) {
423 			p->fts_info = fts_stat(sp, p, 1, -1);
424 			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
425 				if ((p->fts_symfd =
426 				    _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0) {
427 					p->fts_errno = errno;
428 					p->fts_info = FTS_ERR;
429 				} else
430 					p->fts_flags |= FTS_SYMFOLLOW;
431 			}
432 			p->fts_instr = FTS_NOINSTR;
433 		}
434 
435 		free(tmp);
436 
437 name:		t = sp->fts_path + NAPPEND(p->fts_parent);
438 		*t++ = '/';
439 		memmove(t, p->fts_name, p->fts_namelen + 1);
440 		return (sp->fts_cur = p);
441 	}
442 
443 	/* Move up to the parent node. */
444 	p = tmp->fts_parent;
445 
446 	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
447 		/*
448 		 * Done; free everything up and set errno to 0 so the user
449 		 * can distinguish between error and EOF.
450 		 */
451 		free(tmp);
452 		free(p);
453 		errno = 0;
454 		return (sp->fts_cur = NULL);
455 	}
456 
457 	/* NUL terminate the pathname. */
458 	sp->fts_path[p->fts_pathlen] = '\0';
459 
460 	/*
461 	 * Return to the parent directory.  If at a root node or came through
462 	 * a symlink, go back through the file descriptor.  Otherwise, cd up
463 	 * one directory.
464 	 */
465 	if (p->fts_level == FTS_ROOTLEVEL) {
466 		if (FCHDIR(sp, sp->fts_rfd)) {
467 			SET(FTS_STOP);
468 			return (NULL);
469 		}
470 	} else if (p->fts_flags & FTS_SYMFOLLOW) {
471 		if (FCHDIR(sp, p->fts_symfd)) {
472 			saved_errno = errno;
473 			(void)_close(p->fts_symfd);
474 			errno = saved_errno;
475 			SET(FTS_STOP);
476 			return (NULL);
477 		}
478 		(void)_close(p->fts_symfd);
479 	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
480 	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
481 		SET(FTS_STOP);
482 		return (NULL);
483 	}
484 	free(tmp);
485 	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
486 	return (sp->fts_cur = p);
487 }
488 
489 /*
490  * Fts_set takes the stream as an argument although it's not used in this
491  * implementation; it would be necessary if anyone wanted to add global
492  * semantics to fts using fts_set.  An error return is allowed for similar
493  * reasons.
494  */
495 /* ARGSUSED */
496 int
497 fts_set(FTS *sp, FTSENT *p, int instr)
498 {
499 	if (instr != 0 && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
500 	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
501 		errno = EINVAL;
502 		return (1);
503 	}
504 	p->fts_instr = instr;
505 	return (0);
506 }
507 
508 FTSENT *
509 fts_children(FTS *sp, int instr)
510 {
511 	FTSENT *p;
512 	int fd, rc, serrno;
513 
514 	if (instr != 0 && instr != FTS_NAMEONLY) {
515 		errno = EINVAL;
516 		return (NULL);
517 	}
518 
519 	/* Set current node pointer. */
520 	p = sp->fts_cur;
521 
522 	/*
523 	 * Errno set to 0 so user can distinguish empty directory from
524 	 * an error.
525 	 */
526 	errno = 0;
527 
528 	/* Fatal errors stop here. */
529 	if (ISSET(FTS_STOP))
530 		return (NULL);
531 
532 	/* Return logical hierarchy of user's arguments. */
533 	if (p->fts_info == FTS_INIT)
534 		return (p->fts_link);
535 
536 	/*
537 	 * If not a directory being visited in pre-order, stop here.  Could
538 	 * allow FTS_DNR, assuming the user has fixed the problem, but the
539 	 * same effect is available with FTS_AGAIN.
540 	 */
541 	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
542 		return (NULL);
543 
544 	/* Free up any previous child list. */
545 	if (sp->fts_child != NULL)
546 		fts_lfree(sp->fts_child);
547 
548 	if (instr == FTS_NAMEONLY) {
549 		SET(FTS_NAMEONLY);
550 		instr = BNAMES;
551 	} else
552 		instr = BCHILD;
553 
554 	/*
555 	 * If using chdir on a relative path and called BEFORE fts_read does
556 	 * its chdir to the root of a traversal, we can lose -- we need to
557 	 * chdir into the subdirectory, and we don't know where the current
558 	 * directory is, so we can't get back so that the upcoming chdir by
559 	 * fts_read will work.
560 	 */
561 	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
562 	    ISSET(FTS_NOCHDIR))
563 		return (sp->fts_child = fts_build(sp, instr));
564 
565 	if ((fd = _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0)
566 		return (NULL);
567 	sp->fts_child = fts_build(sp, instr);
568 	serrno = (sp->fts_child == NULL) ? errno : 0;
569 	rc = fchdir(fd);
570 	if (rc < 0 && serrno == 0)
571 		serrno = errno;
572 	(void)_close(fd);
573 	errno = serrno;
574 	if (rc < 0)
575 		return (NULL);
576 	return (sp->fts_child);
577 }
578 
579 #ifndef fts_get_clientptr
580 #error "fts_get_clientptr not defined"
581 #endif
582 
583 void *
584 (fts_get_clientptr)(FTS *sp)
585 {
586 
587 	return (fts_get_clientptr(sp));
588 }
589 
590 #ifndef fts_get_stream
591 #error "fts_get_stream not defined"
592 #endif
593 
594 FTS *
595 (fts_get_stream)(FTSENT *p)
596 {
597 	return (fts_get_stream(p));
598 }
599 
600 void
601 fts_set_clientptr(FTS *sp, void *clientptr)
602 {
603 
604 	sp->fts_clientptr = clientptr;
605 }
606 
607 /*
608  * This is the tricky part -- do not casually change *anything* in here.  The
609  * idea is to build the linked list of entries that are used by fts_children
610  * and fts_read.  There are lots of special cases.
611  *
612  * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
613  * set and it's a physical walk (so that symbolic links can't be directories),
614  * we can do things quickly.  First, if it's a 4.4BSD file system, the type
615  * of the file is in the directory entry.  Otherwise, we assume that the number
616  * of subdirectories in a node is equal to the number of links to the parent.
617  * The former skips all stat calls.  The latter skips stat calls in any leaf
618  * directories and for any files after the subdirectories in the directory have
619  * been found, cutting the stat calls by about 2/3.
620  */
621 static FTSENT *
622 fts_build(FTS *sp, int type)
623 {
624 	struct dirent *dp;
625 	FTSENT *p, *head;
626 	FTSENT *cur, *tail;
627 	DIR *dirp;
628 	void *oldaddr;
629 	char *cp;
630 	int cderrno, descend, oflag, saved_errno, nostat, doadjust;
631 	long level;
632 	long nlinks;	/* has to be signed because -1 is a magic value */
633 	size_t dnamlen, len, maxlen, nitems;
634 
635 	/* Set current node pointer. */
636 	cur = sp->fts_cur;
637 
638 	/*
639 	 * Open the directory for reading.  If this fails, we're done.
640 	 * If being called from fts_read, set the fts_info field.
641 	 */
642 #ifdef FTS_WHITEOUT
643 	if (ISSET(FTS_WHITEOUT))
644 		oflag = DTF_NODUP;
645 	else
646 		oflag = DTF_HIDEW | DTF_NODUP;
647 #else
648 #define __opendir2(path, flag) opendir(path)
649 #endif
650 	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
651 		if (type == BREAD) {
652 			cur->fts_info = FTS_DNR;
653 			cur->fts_errno = errno;
654 		}
655 		return (NULL);
656 	}
657 
658 	/*
659 	 * Nlinks is the number of possible entries of type directory in the
660 	 * directory if we're cheating on stat calls, 0 if we're not doing
661 	 * any stat calls at all, -1 if we're doing stats on everything.
662 	 */
663 	if (type == BNAMES) {
664 		nlinks = 0;
665 		/* Be quiet about nostat, GCC. */
666 		nostat = 0;
667 	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
668 		if (fts_ufslinks(sp, cur))
669 			nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
670 		else
671 			nlinks = -1;
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 	} else {
726 		/* GCC, you're too verbose. */
727 		cp = NULL;
728 	}
729 	len++;
730 	maxlen = sp->fts_pathlen - len;
731 
732 	level = cur->fts_level + 1;
733 
734 	/* Read the directory, attaching each entry to the `link' pointer. */
735 	doadjust = 0;
736 	for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) {
737 		dnamlen = dp->d_namlen;
738 		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
739 			continue;
740 
741 		if ((p = fts_alloc(sp, dp->d_name, dnamlen)) == NULL)
742 			goto mem1;
743 		if (dnamlen >= maxlen) {	/* include space for NUL */
744 			oldaddr = sp->fts_path;
745 			if (fts_palloc(sp, dnamlen + len + 1)) {
746 				/*
747 				 * No more memory for path or structures.  Save
748 				 * errno, free up the current structure and the
749 				 * structures already allocated.
750 				 */
751 mem1:				saved_errno = errno;
752 				if (p)
753 					free(p);
754 				fts_lfree(head);
755 				(void)closedir(dirp);
756 				cur->fts_info = FTS_ERR;
757 				SET(FTS_STOP);
758 				errno = saved_errno;
759 				return (NULL);
760 			}
761 			/* Did realloc() change the pointer? */
762 			if (oldaddr != sp->fts_path) {
763 				doadjust = 1;
764 				if (ISSET(FTS_NOCHDIR))
765 					cp = sp->fts_path + len;
766 			}
767 			maxlen = sp->fts_pathlen - len;
768 		}
769 
770 		p->fts_level = level;
771 		p->fts_parent = sp->fts_cur;
772 		p->fts_pathlen = len + dnamlen;
773 
774 #ifdef FTS_WHITEOUT
775 		if (dp->d_type == DT_WHT)
776 			p->fts_flags |= FTS_ISW;
777 #endif
778 
779 		if (cderrno) {
780 			if (nlinks) {
781 				p->fts_info = FTS_NS;
782 				p->fts_errno = cderrno;
783 			} else
784 				p->fts_info = FTS_NSOK;
785 			p->fts_accpath = cur->fts_accpath;
786 		} else if (nlinks == 0
787 #ifdef DT_DIR
788 		    || (nostat &&
789 		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
790 #endif
791 		    ) {
792 			p->fts_accpath =
793 			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
794 			p->fts_info = FTS_NSOK;
795 		} else {
796 			/* Build a file name for fts_stat to stat. */
797 			if (ISSET(FTS_NOCHDIR)) {
798 				p->fts_accpath = p->fts_path;
799 				memmove(cp, p->fts_name, p->fts_namelen + 1);
800 				p->fts_info = fts_stat(sp, p, 0, _dirfd(dirp));
801 			} else {
802 				p->fts_accpath = p->fts_name;
803 				p->fts_info = fts_stat(sp, p, 0, -1);
804 			}
805 
806 			/* Decrement link count if applicable. */
807 			if (nlinks > 0 && (p->fts_info == FTS_D ||
808 			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
809 				--nlinks;
810 		}
811 
812 		/* We walk in directory order so "ls -f" doesn't get upset. */
813 		p->fts_link = NULL;
814 		if (head == NULL)
815 			head = tail = p;
816 		else {
817 			tail->fts_link = p;
818 			tail = p;
819 		}
820 		++nitems;
821 	}
822 	if (dirp)
823 		(void)closedir(dirp);
824 
825 	/*
826 	 * If realloc() changed the address of the path, adjust the
827 	 * addresses for the rest of the tree and the dir list.
828 	 */
829 	if (doadjust)
830 		fts_padjust(sp, head);
831 
832 	/*
833 	 * If not changing directories, reset the path back to original
834 	 * state.
835 	 */
836 	if (ISSET(FTS_NOCHDIR))
837 		sp->fts_path[cur->fts_pathlen] = '\0';
838 
839 	/*
840 	 * If descended after called from fts_children or after called from
841 	 * fts_read and nothing found, get back.  At the root level we use
842 	 * the saved fd; if one of fts_open()'s arguments is a relative path
843 	 * to an empty directory, we wind up here with no other way back.  If
844 	 * can't get back, we're done.
845 	 */
846 	if (descend && (type == BCHILD || !nitems) &&
847 	    (cur->fts_level == FTS_ROOTLEVEL ?
848 	    FCHDIR(sp, sp->fts_rfd) :
849 	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
850 		fts_lfree(head);
851 		cur->fts_info = FTS_ERR;
852 		SET(FTS_STOP);
853 		return (NULL);
854 	}
855 
856 	/* If didn't find anything, return NULL. */
857 	if (!nitems) {
858 		if (type == BREAD)
859 			cur->fts_info = FTS_DP;
860 		return (NULL);
861 	}
862 
863 	/* Sort the entries. */
864 	if (sp->fts_compar && nitems > 1)
865 		head = fts_sort(sp, head, nitems);
866 	return (head);
867 }
868 
869 static int
870 fts_stat(FTS *sp, FTSENT *p, int follow, int dfd)
871 {
872 	FTSENT *t;
873 	dev_t dev;
874 	ino_t ino;
875 	struct stat *sbp, sb;
876 	int saved_errno;
877 	const char *path;
878 
879 	if (dfd == -1)
880 		path = p->fts_accpath, dfd = AT_FDCWD;
881 	else
882 		path = p->fts_name;
883 
884 	/* If user needs stat info, stat buffer already allocated. */
885 	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
886 
887 #ifdef FTS_WHITEOUT
888 	/* Check for whiteout. */
889 	if (p->fts_flags & FTS_ISW) {
890 		if (sbp != &sb) {
891 			memset(sbp, '\0', sizeof(*sbp));
892 			sbp->st_mode = S_IFWHT;
893 		}
894 		return (FTS_W);
895 	}
896 #endif
897 
898 	/*
899 	 * If doing a logical walk, or application requested FTS_FOLLOW, do
900 	 * a stat(2).  If that fails, check for a non-existent symlink.  If
901 	 * fail, set the errno from the stat call.
902 	 */
903 	if (ISSET(FTS_LOGICAL) || follow) {
904 		if (fstatat(dfd, path, sbp, 0)) {
905 			saved_errno = errno;
906 			if (fstatat(dfd, path, sbp, AT_SYMLINK_NOFOLLOW)) {
907 				p->fts_errno = saved_errno;
908 				goto err;
909 			}
910 			errno = 0;
911 			if (S_ISLNK(sbp->st_mode))
912 				return (FTS_SLNONE);
913 		}
914 	} else if (fstatat(dfd, path, sbp, AT_SYMLINK_NOFOLLOW)) {
915 		p->fts_errno = errno;
916 err:		memset(sbp, 0, sizeof(struct stat));
917 		return (FTS_NS);
918 	}
919 
920 	if (S_ISDIR(sbp->st_mode)) {
921 		/*
922 		 * Set the device/inode.  Used to find cycles and check for
923 		 * crossing mount points.  Also remember the link count, used
924 		 * in fts_build to limit the number of stat calls.  It is
925 		 * understood that these fields are only referenced if fts_info
926 		 * is set to FTS_D.
927 		 */
928 		dev = p->fts_dev = sbp->st_dev;
929 		ino = p->fts_ino = sbp->st_ino;
930 		p->fts_nlink = sbp->st_nlink;
931 
932 		if (ISDOT(p->fts_name))
933 			return (FTS_DOT);
934 
935 		/*
936 		 * Cycle detection is done by brute force when the directory
937 		 * is first encountered.  If the tree gets deep enough or the
938 		 * number of symbolic links to directories is high enough,
939 		 * something faster might be worthwhile.
940 		 */
941 		for (t = p->fts_parent;
942 		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
943 			if (ino == t->fts_ino && dev == t->fts_dev) {
944 				p->fts_cycle = t;
945 				return (FTS_DC);
946 			}
947 		return (FTS_D);
948 	}
949 	if (S_ISLNK(sbp->st_mode))
950 		return (FTS_SL);
951 	if (S_ISREG(sbp->st_mode))
952 		return (FTS_F);
953 	return (FTS_DEFAULT);
954 }
955 
956 /*
957  * The comparison function takes pointers to pointers to FTSENT structures.
958  * Qsort wants a comparison function that takes pointers to void.
959  * (Both with appropriate levels of const-poisoning, of course!)
960  * Use a trampoline function to deal with the difference.
961  */
962 static int
963 fts_compar(const void *a, const void *b)
964 {
965 	FTS *parent;
966 
967 	parent = (*(const FTSENT * const *)a)->fts_fts;
968 	return (*parent->fts_compar)(a, b);
969 }
970 
971 static FTSENT *
972 fts_sort(FTS *sp, FTSENT *head, size_t nitems)
973 {
974 	FTSENT **ap, *p;
975 
976 	/*
977 	 * Construct an array of pointers to the structures and call qsort(3).
978 	 * Reassemble the array in the order returned by qsort.  If unable to
979 	 * sort for memory reasons, return the directory entries in their
980 	 * current order.  Allocate enough space for the current needs plus
981 	 * 40 so don't realloc one entry at a time.
982 	 */
983 	if (nitems > sp->fts_nitems) {
984 		sp->fts_nitems = nitems + 40;
985 		if ((sp->fts_array = reallocf(sp->fts_array,
986 		    sp->fts_nitems * sizeof(FTSENT *))) == NULL) {
987 			sp->fts_nitems = 0;
988 			return (head);
989 		}
990 	}
991 	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
992 		*ap++ = p;
993 	qsort(sp->fts_array, nitems, sizeof(FTSENT *), fts_compar);
994 	for (head = *(ap = sp->fts_array); --nitems; ++ap)
995 		ap[0]->fts_link = ap[1];
996 	ap[0]->fts_link = NULL;
997 	return (head);
998 }
999 
1000 static FTSENT *
1001 fts_alloc(FTS *sp, char *name, size_t namelen)
1002 {
1003 	FTSENT *p;
1004 	size_t len;
1005 
1006 	struct ftsent_withstat {
1007 		FTSENT	ent;
1008 		struct	stat statbuf;
1009 	};
1010 
1011 	/*
1012 	 * The file name is a variable length array and no stat structure is
1013 	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
1014 	 * structure, the file name and the stat structure in one chunk, but
1015 	 * be careful that the stat structure is reasonably aligned.
1016 	 */
1017 	if (ISSET(FTS_NOSTAT))
1018 		len = sizeof(FTSENT) + namelen + 1;
1019 	else
1020 		len = sizeof(struct ftsent_withstat) + namelen + 1;
1021 
1022 	if ((p = malloc(len)) == NULL)
1023 		return (NULL);
1024 
1025 	if (ISSET(FTS_NOSTAT)) {
1026 		p->fts_name = (char *)(p + 1);
1027 		p->fts_statp = NULL;
1028 	} else {
1029 		p->fts_name = (char *)((struct ftsent_withstat *)p + 1);
1030 		p->fts_statp = &((struct ftsent_withstat *)p)->statbuf;
1031 	}
1032 
1033 	/* Copy the name and guarantee NUL termination. */
1034 	memcpy(p->fts_name, name, namelen);
1035 	p->fts_name[namelen] = '\0';
1036 	p->fts_namelen = namelen;
1037 	p->fts_path = sp->fts_path;
1038 	p->fts_errno = 0;
1039 	p->fts_flags = 0;
1040 	p->fts_instr = FTS_NOINSTR;
1041 	p->fts_number = 0;
1042 	p->fts_pointer = NULL;
1043 	p->fts_fts = sp;
1044 	return (p);
1045 }
1046 
1047 static void
1048 fts_lfree(FTSENT *head)
1049 {
1050 	FTSENT *p;
1051 
1052 	/* Free a linked list of structures. */
1053 	while ((p = head)) {
1054 		head = head->fts_link;
1055 		free(p);
1056 	}
1057 }
1058 
1059 /*
1060  * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
1061  * Most systems will allow creation of paths much longer than MAXPATHLEN, even
1062  * though the kernel won't resolve them.  Add the size (not just what's needed)
1063  * plus 256 bytes so don't realloc the path 2 bytes at a time.
1064  */
1065 static int
1066 fts_palloc(FTS *sp, size_t more)
1067 {
1068 
1069 	sp->fts_pathlen += more + 256;
1070 	sp->fts_path = reallocf(sp->fts_path, sp->fts_pathlen);
1071 	return (sp->fts_path == NULL);
1072 }
1073 
1074 /*
1075  * When the path is realloc'd, have to fix all of the pointers in structures
1076  * already returned.
1077  */
1078 static void
1079 fts_padjust(FTS *sp, FTSENT *head)
1080 {
1081 	FTSENT *p;
1082 	char *addr = sp->fts_path;
1083 
1084 #define	ADJUST(p) do {							\
1085 	if ((p)->fts_accpath != (p)->fts_name) {			\
1086 		(p)->fts_accpath =					\
1087 		    (char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
1088 	}								\
1089 	(p)->fts_path = addr;						\
1090 } while (0)
1091 	/* Adjust the current set of children. */
1092 	for (p = sp->fts_child; p; p = p->fts_link)
1093 		ADJUST(p);
1094 
1095 	/* Adjust the rest of the tree, including the current level. */
1096 	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1097 		ADJUST(p);
1098 		p = p->fts_link ? p->fts_link : p->fts_parent;
1099 	}
1100 }
1101 
1102 static size_t
1103 fts_maxarglen(char * const *argv)
1104 {
1105 	size_t len, max;
1106 
1107 	for (max = 0; *argv; ++argv)
1108 		if ((len = strlen(*argv)) > max)
1109 			max = len;
1110 	return (max + 1);
1111 }
1112 
1113 /*
1114  * Change to dir specified by fd or p->fts_accpath without getting
1115  * tricked by someone changing the world out from underneath us.
1116  * Assumes p->fts_dev and p->fts_ino are filled in.
1117  */
1118 static int
1119 fts_safe_changedir(FTS *sp, FTSENT *p, int fd, char *path)
1120 {
1121 	int ret, oerrno, newfd;
1122 	struct stat sb;
1123 
1124 	newfd = fd;
1125 	if (ISSET(FTS_NOCHDIR))
1126 		return (0);
1127 	if (fd < 0 && (newfd = _open(path, O_RDONLY | O_DIRECTORY |
1128 	    O_CLOEXEC, 0)) < 0)
1129 		return (-1);
1130 	if (_fstat(newfd, &sb)) {
1131 		ret = -1;
1132 		goto bail;
1133 	}
1134 	if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) {
1135 		errno = ENOENT;		/* disinformation */
1136 		ret = -1;
1137 		goto bail;
1138 	}
1139 	ret = fchdir(newfd);
1140 bail:
1141 	oerrno = errno;
1142 	if (fd < 0)
1143 		(void)_close(newfd);
1144 	errno = oerrno;
1145 	return (ret);
1146 }
1147 
1148 /*
1149  * Check if the filesystem for "ent" has UFS-style links.
1150  */
1151 static int
1152 fts_ufslinks(FTS *sp, const FTSENT *ent)
1153 {
1154 	struct _fts_private *priv;
1155 	const char **cpp;
1156 
1157 	priv = (struct _fts_private *)sp;
1158 	/*
1159 	 * If this node's device is different from the previous, grab
1160 	 * the filesystem information, and decide on the reliability
1161 	 * of the link information from this filesystem for stat(2)
1162 	 * avoidance.
1163 	 */
1164 	if (priv->ftsp_dev != ent->fts_dev) {
1165 		if (statfs(ent->fts_path, &priv->ftsp_statfs) != -1) {
1166 			priv->ftsp_dev = ent->fts_dev;
1167 			priv->ftsp_linksreliable = 0;
1168 			for (cpp = ufslike_filesystems; *cpp; cpp++) {
1169 				if (strcmp(priv->ftsp_statfs.f_fstypename,
1170 				    *cpp) == 0) {
1171 					priv->ftsp_linksreliable = 1;
1172 					break;
1173 				}
1174 			}
1175 		} else {
1176 			priv->ftsp_linksreliable = 0;
1177 		}
1178 	}
1179 	return (priv->ftsp_linksreliable);
1180 }
1181