xref: /freebsd/lib/libc/gen/fts.c (revision 941e286383714ef25f1ffe9ba6ae5040afdd7060)
1 /*-
2  * Copyright (c) 1990, 1993, 1994
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  * $OpenBSD: fts.c,v 1.22 1999/10/03 19:22:22 millert Exp $
30  */
31 
32 #if 0
33 #if defined(LIBC_SCCS) && !defined(lint)
34 static char sccsid[] = "@(#)fts.c	8.6 (Berkeley) 8/14/94";
35 #endif /* LIBC_SCCS and not lint */
36 #endif
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include "namespace.h"
42 #include <sys/param.h>
43 #include <sys/mount.h>
44 #include <sys/stat.h>
45 
46 #include <dirent.h>
47 #include <errno.h>
48 #include <fcntl.h>
49 #include <fts.h>
50 #include <stdlib.h>
51 #include <string.h>
52 #include <unistd.h>
53 #include "un-namespace.h"
54 
55 static FTSENT	*fts_alloc(FTS *, char *, size_t);
56 static FTSENT	*fts_build(FTS *, int);
57 static void	 fts_lfree(FTSENT *);
58 static void	 fts_load(FTS *, FTSENT *);
59 static size_t	 fts_maxarglen(char * const *);
60 static void	 fts_padjust(FTS *, FTSENT *);
61 static int	 fts_palloc(FTS *, size_t);
62 static FTSENT	*fts_sort(FTS *, FTSENT *, size_t);
63 static int	 fts_stat(FTS *, FTSENT *, int);
64 static int	 fts_safe_changedir(FTS *, FTSENT *, int, char *);
65 static int	 fts_ufslinks(FTS *, const FTSENT *);
66 
67 #define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
68 
69 #define	CLR(opt)	(sp->fts_options &= ~(opt))
70 #define	ISSET(opt)	(sp->fts_options & (opt))
71 #define	SET(opt)	(sp->fts_options |= (opt))
72 
73 #define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
74 
75 /* fts_build flags */
76 #define	BCHILD		1		/* fts_children */
77 #define	BNAMES		2		/* fts_children, names only */
78 #define	BREAD		3		/* fts_read */
79 
80 /*
81  * Internal representation of an FTS, including extra implementation
82  * details.  The FTS returned from fts_open points to this structure's
83  * ftsp_fts member (and can be cast to an _fts_private as required)
84  */
85 struct _fts_private {
86 	FTS		ftsp_fts;
87 	struct statfs	ftsp_statfs;
88 	dev_t		ftsp_dev;
89 	int		ftsp_linksreliable;
90 };
91 
92 /*
93  * The "FTS_NOSTAT" option can avoid a lot of calls to stat(2) if it
94  * knows that a directory could not possibly have subdirectories.  This
95  * is decided by looking at the link count: a subdirectory would
96  * increment its parent's link count by virtue of its own ".." entry.
97  * This assumption only holds for UFS-like filesystems that implement
98  * links and directories this way, so we must punt for others.
99  */
100 
101 static const char *ufslike_filesystems[] = {
102 	"ufs",
103 	"nfs",
104 	"nfs4",
105 	"ext2fs",
106 	0
107 };
108 
109 FTS *
110 fts_open(argv, options, compar)
111 	char * const *argv;
112 	int options;
113 	int (*compar)(const FTSENT * const *, const FTSENT * const *);
114 {
115 	struct _fts_private *priv;
116 	FTS *sp;
117 	FTSENT *p, *root;
118 	FTSENT *parent, *tmp;
119 	size_t len, nitems;
120 
121 	/* Options check. */
122 	if (options & ~FTS_OPTIONMASK) {
123 		errno = EINVAL;
124 		return (NULL);
125 	}
126 
127 	/* fts_open() requires at least one path */
128 	if (*argv == NULL) {
129 		errno = EINVAL;
130 		return (NULL);
131 	}
132 
133 	/* Allocate/initialize the stream. */
134 	if ((priv = malloc(sizeof(*priv))) == NULL)
135 		return (NULL);
136 	memset(priv, 0, sizeof(*priv));
137 	sp = &priv->ftsp_fts;
138 	sp->fts_compar = compar;
139 	sp->fts_options = options;
140 
141 	/* Shush, GCC. */
142 	tmp = NULL;
143 
144 	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
145 	if (ISSET(FTS_LOGICAL))
146 		SET(FTS_NOCHDIR);
147 
148 	/*
149 	 * Start out with 1K of path space, and enough, in any case,
150 	 * to hold the user's paths.
151 	 */
152 	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
153 		goto mem1;
154 
155 	/* Allocate/initialize root's parent. */
156 	if ((parent = fts_alloc(sp, "", 0)) == NULL)
157 		goto mem2;
158 	parent->fts_level = FTS_ROOTPARENTLEVEL;
159 
160 	/* Allocate/initialize root(s). */
161 	for (root = NULL, nitems = 0; *argv != NULL; ++argv, ++nitems) {
162 		/* Don't allow zero-length paths. */
163 		if ((len = strlen(*argv)) == 0) {
164 			errno = ENOENT;
165 			goto mem3;
166 		}
167 
168 		p = fts_alloc(sp, *argv, len);
169 		p->fts_level = FTS_ROOTLEVEL;
170 		p->fts_parent = parent;
171 		p->fts_accpath = p->fts_name;
172 		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW));
173 
174 		/* Command-line "." and ".." are real directories. */
175 		if (p->fts_info == FTS_DOT)
176 			p->fts_info = FTS_D;
177 
178 		/*
179 		 * If comparison routine supplied, traverse in sorted
180 		 * order; otherwise traverse in the order specified.
181 		 */
182 		if (compar) {
183 			p->fts_link = root;
184 			root = p;
185 		} else {
186 			p->fts_link = NULL;
187 			if (root == NULL)
188 				tmp = root = p;
189 			else {
190 				tmp->fts_link = p;
191 				tmp = p;
192 			}
193 		}
194 	}
195 	if (compar && nitems > 1)
196 		root = fts_sort(sp, root, nitems);
197 
198 	/*
199 	 * Allocate a dummy pointer and make fts_read think that we've just
200 	 * finished the node before the root(s); set p->fts_info to FTS_INIT
201 	 * so that everything about the "current" node is ignored.
202 	 */
203 	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
204 		goto mem3;
205 	sp->fts_cur->fts_link = root;
206 	sp->fts_cur->fts_info = FTS_INIT;
207 
208 	/*
209 	 * If using chdir(2), grab a file descriptor pointing to dot to ensure
210 	 * that we can get back here; this could be avoided for some paths,
211 	 * but almost certainly not worth the effort.  Slashes, symbolic links,
212 	 * and ".." are all fairly nasty problems.  Note, if we can't get the
213 	 * descriptor we run anyway, just more slowly.
214 	 */
215 	if (!ISSET(FTS_NOCHDIR) && (sp->fts_rfd = _open(".", O_RDONLY, 0)) < 0)
216 		SET(FTS_NOCHDIR);
217 
218 	return (sp);
219 
220 mem3:	fts_lfree(root);
221 	free(parent);
222 mem2:	free(sp->fts_path);
223 mem1:	free(sp);
224 	return (NULL);
225 }
226 
227 static void
228 fts_load(sp, p)
229 	FTS *sp;
230 	FTSENT *p;
231 {
232 	size_t len;
233 	char *cp;
234 
235 	/*
236 	 * Load the stream structure for the next traversal.  Since we don't
237 	 * actually enter the directory until after the preorder visit, set
238 	 * the fts_accpath field specially so the chdir gets done to the right
239 	 * place and the user can access the first node.  From fts_open it's
240 	 * known that the path will fit.
241 	 */
242 	len = p->fts_pathlen = p->fts_namelen;
243 	memmove(sp->fts_path, p->fts_name, len + 1);
244 	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
245 		len = strlen(++cp);
246 		memmove(p->fts_name, cp, len + 1);
247 		p->fts_namelen = len;
248 	}
249 	p->fts_accpath = p->fts_path = sp->fts_path;
250 	sp->fts_dev = p->fts_dev;
251 }
252 
253 int
254 fts_close(sp)
255 	FTS *sp;
256 {
257 	FTSENT *freep, *p;
258 	int saved_errno;
259 
260 	/*
261 	 * This still works if we haven't read anything -- the dummy structure
262 	 * points to the root list, so we step through to the end of the root
263 	 * list which has a valid parent pointer.
264 	 */
265 	if (sp->fts_cur) {
266 		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
267 			freep = p;
268 			p = p->fts_link != NULL ? p->fts_link : p->fts_parent;
269 			free(freep);
270 		}
271 		free(p);
272 	}
273 
274 	/* Free up child linked list, sort array, path buffer. */
275 	if (sp->fts_child)
276 		fts_lfree(sp->fts_child);
277 	if (sp->fts_array)
278 		free(sp->fts_array);
279 	free(sp->fts_path);
280 
281 	/* Return to original directory, save errno if necessary. */
282 	if (!ISSET(FTS_NOCHDIR)) {
283 		saved_errno = fchdir(sp->fts_rfd) ? errno : 0;
284 		(void)_close(sp->fts_rfd);
285 
286 		/* Set errno and return. */
287 		if (saved_errno != 0) {
288 			/* Free up the stream pointer. */
289 			free(sp);
290 			errno = saved_errno;
291 			return (-1);
292 		}
293 	}
294 
295 	/* Free up the stream pointer. */
296 	free(sp);
297 	return (0);
298 }
299 
300 /*
301  * Special case of "/" at the end of the path so that slashes aren't
302  * appended which would cause paths to be written as "....//foo".
303  */
304 #define	NAPPEND(p)							\
305 	(p->fts_path[p->fts_pathlen - 1] == '/'				\
306 	    ? p->fts_pathlen - 1 : p->fts_pathlen)
307 
308 FTSENT *
309 fts_read(sp)
310 	FTS *sp;
311 {
312 	FTSENT *p, *tmp;
313 	int instr;
314 	char *t;
315 	int saved_errno;
316 
317 	/* If finished or unrecoverable error, return NULL. */
318 	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
319 		return (NULL);
320 
321 	/* Set current node pointer. */
322 	p = sp->fts_cur;
323 
324 	/* Save and zero out user instructions. */
325 	instr = p->fts_instr;
326 	p->fts_instr = FTS_NOINSTR;
327 
328 	/* Any type of file may be re-visited; re-stat and re-turn. */
329 	if (instr == FTS_AGAIN) {
330 		p->fts_info = fts_stat(sp, p, 0);
331 		return (p);
332 	}
333 
334 	/*
335 	 * Following a symlink -- SLNONE test allows application to see
336 	 * SLNONE and recover.  If indirecting through a symlink, have
337 	 * keep a pointer to current location.  If unable to get that
338 	 * pointer, follow fails.
339 	 */
340 	if (instr == FTS_FOLLOW &&
341 	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
342 		p->fts_info = fts_stat(sp, p, 1);
343 		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
344 			if ((p->fts_symfd = _open(".", O_RDONLY, 0)) < 0) {
345 				p->fts_errno = errno;
346 				p->fts_info = FTS_ERR;
347 			} else
348 				p->fts_flags |= FTS_SYMFOLLOW;
349 		}
350 		return (p);
351 	}
352 
353 	/* Directory in pre-order. */
354 	if (p->fts_info == FTS_D) {
355 		/* If skipped or crossed mount point, do post-order visit. */
356 		if (instr == FTS_SKIP ||
357 		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
358 			if (p->fts_flags & FTS_SYMFOLLOW)
359 				(void)_close(p->fts_symfd);
360 			if (sp->fts_child) {
361 				fts_lfree(sp->fts_child);
362 				sp->fts_child = NULL;
363 			}
364 			p->fts_info = FTS_DP;
365 			return (p);
366 		}
367 
368 		/* Rebuild if only read the names and now traversing. */
369 		if (sp->fts_child != NULL && ISSET(FTS_NAMEONLY)) {
370 			CLR(FTS_NAMEONLY);
371 			fts_lfree(sp->fts_child);
372 			sp->fts_child = NULL;
373 		}
374 
375 		/*
376 		 * Cd to the subdirectory.
377 		 *
378 		 * If have already read and now fail to chdir, whack the list
379 		 * to make the names come out right, and set the parent errno
380 		 * so the application will eventually get an error condition.
381 		 * Set the FTS_DONTCHDIR flag so that when we logically change
382 		 * directories back to the parent we don't do a chdir.
383 		 *
384 		 * If haven't read do so.  If the read fails, fts_build sets
385 		 * FTS_STOP or the fts_info field of the node.
386 		 */
387 		if (sp->fts_child != NULL) {
388 			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
389 				p->fts_errno = errno;
390 				p->fts_flags |= FTS_DONTCHDIR;
391 				for (p = sp->fts_child; p != NULL;
392 				    p = p->fts_link)
393 					p->fts_accpath =
394 					    p->fts_parent->fts_accpath;
395 			}
396 		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
397 			if (ISSET(FTS_STOP))
398 				return (NULL);
399 			return (p);
400 		}
401 		p = sp->fts_child;
402 		sp->fts_child = NULL;
403 		goto name;
404 	}
405 
406 	/* Move to the next node on this level. */
407 next:	tmp = p;
408 	if ((p = p->fts_link) != NULL) {
409 		free(tmp);
410 
411 		/*
412 		 * If reached the top, return to the original directory (or
413 		 * the root of the tree), and load the paths for the next root.
414 		 */
415 		if (p->fts_level == FTS_ROOTLEVEL) {
416 			if (FCHDIR(sp, sp->fts_rfd)) {
417 				SET(FTS_STOP);
418 				return (NULL);
419 			}
420 			fts_load(sp, p);
421 			return (sp->fts_cur = p);
422 		}
423 
424 		/*
425 		 * User may have called fts_set on the node.  If skipped,
426 		 * ignore.  If followed, get a file descriptor so we can
427 		 * get back if necessary.
428 		 */
429 		if (p->fts_instr == FTS_SKIP)
430 			goto next;
431 		if (p->fts_instr == FTS_FOLLOW) {
432 			p->fts_info = fts_stat(sp, p, 1);
433 			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
434 				if ((p->fts_symfd =
435 				    _open(".", O_RDONLY, 0)) < 0) {
436 					p->fts_errno = errno;
437 					p->fts_info = FTS_ERR;
438 				} else
439 					p->fts_flags |= FTS_SYMFOLLOW;
440 			}
441 			p->fts_instr = FTS_NOINSTR;
442 		}
443 
444 name:		t = sp->fts_path + NAPPEND(p->fts_parent);
445 		*t++ = '/';
446 		memmove(t, p->fts_name, p->fts_namelen + 1);
447 		return (sp->fts_cur = p);
448 	}
449 
450 	/* Move up to the parent node. */
451 	p = tmp->fts_parent;
452 	free(tmp);
453 
454 	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
455 		/*
456 		 * Done; free everything up and set errno to 0 so the user
457 		 * can distinguish between error and EOF.
458 		 */
459 		free(p);
460 		errno = 0;
461 		return (sp->fts_cur = NULL);
462 	}
463 
464 	/* NUL terminate the pathname. */
465 	sp->fts_path[p->fts_pathlen] = '\0';
466 
467 	/*
468 	 * Return to the parent directory.  If at a root node or came through
469 	 * a symlink, go back through the file descriptor.  Otherwise, cd up
470 	 * one directory.
471 	 */
472 	if (p->fts_level == FTS_ROOTLEVEL) {
473 		if (FCHDIR(sp, sp->fts_rfd)) {
474 			SET(FTS_STOP);
475 			return (NULL);
476 		}
477 	} else if (p->fts_flags & FTS_SYMFOLLOW) {
478 		if (FCHDIR(sp, p->fts_symfd)) {
479 			saved_errno = errno;
480 			(void)_close(p->fts_symfd);
481 			errno = saved_errno;
482 			SET(FTS_STOP);
483 			return (NULL);
484 		}
485 		(void)_close(p->fts_symfd);
486 	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
487 	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
488 		SET(FTS_STOP);
489 		return (NULL);
490 	}
491 	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
492 	return (sp->fts_cur = p);
493 }
494 
495 /*
496  * Fts_set takes the stream as an argument although it's not used in this
497  * implementation; it would be necessary if anyone wanted to add global
498  * semantics to fts using fts_set.  An error return is allowed for similar
499  * reasons.
500  */
501 /* ARGSUSED */
502 int
503 fts_set(sp, p, instr)
504 	FTS *sp;
505 	FTSENT *p;
506 	int instr;
507 {
508 	if (instr != 0 && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
509 	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
510 		errno = EINVAL;
511 		return (1);
512 	}
513 	p->fts_instr = instr;
514 	return (0);
515 }
516 
517 FTSENT *
518 fts_children(sp, instr)
519 	FTS *sp;
520 	int instr;
521 {
522 	FTSENT *p;
523 	int fd;
524 
525 	if (instr != 0 && instr != FTS_NAMEONLY) {
526 		errno = EINVAL;
527 		return (NULL);
528 	}
529 
530 	/* Set current node pointer. */
531 	p = sp->fts_cur;
532 
533 	/*
534 	 * Errno set to 0 so user can distinguish empty directory from
535 	 * an error.
536 	 */
537 	errno = 0;
538 
539 	/* Fatal errors stop here. */
540 	if (ISSET(FTS_STOP))
541 		return (NULL);
542 
543 	/* Return logical hierarchy of user's arguments. */
544 	if (p->fts_info == FTS_INIT)
545 		return (p->fts_link);
546 
547 	/*
548 	 * If not a directory being visited in pre-order, stop here.  Could
549 	 * allow FTS_DNR, assuming the user has fixed the problem, but the
550 	 * same effect is available with FTS_AGAIN.
551 	 */
552 	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
553 		return (NULL);
554 
555 	/* Free up any previous child list. */
556 	if (sp->fts_child != NULL)
557 		fts_lfree(sp->fts_child);
558 
559 	if (instr == FTS_NAMEONLY) {
560 		SET(FTS_NAMEONLY);
561 		instr = BNAMES;
562 	} else
563 		instr = BCHILD;
564 
565 	/*
566 	 * If using chdir on a relative path and called BEFORE fts_read does
567 	 * its chdir to the root of a traversal, we can lose -- we need to
568 	 * chdir into the subdirectory, and we don't know where the current
569 	 * directory is, so we can't get back so that the upcoming chdir by
570 	 * fts_read will work.
571 	 */
572 	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
573 	    ISSET(FTS_NOCHDIR))
574 		return (sp->fts_child = fts_build(sp, instr));
575 
576 	if ((fd = _open(".", O_RDONLY, 0)) < 0)
577 		return (NULL);
578 	sp->fts_child = fts_build(sp, instr);
579 	if (fchdir(fd)) {
580 		(void)_close(fd);
581 		return (NULL);
582 	}
583 	(void)_close(fd);
584 	return (sp->fts_child);
585 }
586 
587 #ifndef fts_get_clientptr
588 #error "fts_get_clientptr not defined"
589 #endif
590 
591 void *
592 (fts_get_clientptr)(FTS *sp)
593 {
594 
595 	return (fts_get_clientptr(sp));
596 }
597 
598 #ifndef fts_get_stream
599 #error "fts_get_stream not defined"
600 #endif
601 
602 FTS *
603 (fts_get_stream)(FTSENT *p)
604 {
605 	return (fts_get_stream(p));
606 }
607 
608 void
609 fts_set_clientptr(FTS *sp, void *clientptr)
610 {
611 
612 	sp->fts_clientptr = clientptr;
613 }
614 
615 /*
616  * This is the tricky part -- do not casually change *anything* in here.  The
617  * idea is to build the linked list of entries that are used by fts_children
618  * and fts_read.  There are lots of special cases.
619  *
620  * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
621  * set and it's a physical walk (so that symbolic links can't be directories),
622  * we can do things quickly.  First, if it's a 4.4BSD file system, the type
623  * of the file is in the directory entry.  Otherwise, we assume that the number
624  * of subdirectories in a node is equal to the number of links to the parent.
625  * The former skips all stat calls.  The latter skips stat calls in any leaf
626  * directories and for any files after the subdirectories in the directory have
627  * been found, cutting the stat calls by about 2/3.
628  */
629 static FTSENT *
630 fts_build(sp, type)
631 	FTS *sp;
632 	int type;
633 {
634 	struct dirent *dp;
635 	FTSENT *p, *head;
636 	FTSENT *cur, *tail;
637 	DIR *dirp;
638 	void *oldaddr;
639 	char *cp;
640 	int cderrno, descend, oflag, saved_errno, nostat, doadjust;
641 	long level;
642 	long nlinks;	/* has to be signed because -1 is a magic value */
643 	size_t dnamlen, len, maxlen, nitems;
644 
645 	/* Set current node pointer. */
646 	cur = sp->fts_cur;
647 
648 	/*
649 	 * Open the directory for reading.  If this fails, we're done.
650 	 * If being called from fts_read, set the fts_info field.
651 	 */
652 #ifdef FTS_WHITEOUT
653 	if (ISSET(FTS_WHITEOUT))
654 		oflag = DTF_NODUP | DTF_REWIND;
655 	else
656 		oflag = DTF_HIDEW | DTF_NODUP | DTF_REWIND;
657 #else
658 #define __opendir2(path, flag) opendir(path)
659 #endif
660 	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
661 		if (type == BREAD) {
662 			cur->fts_info = FTS_DNR;
663 			cur->fts_errno = errno;
664 		}
665 		return (NULL);
666 	}
667 
668 	/*
669 	 * Nlinks is the number of possible entries of type directory in the
670 	 * directory if we're cheating on stat calls, 0 if we're not doing
671 	 * any stat calls at all, -1 if we're doing stats on everything.
672 	 */
673 	if (type == BNAMES) {
674 		nlinks = 0;
675 		/* Be quiet about nostat, GCC. */
676 		nostat = 0;
677 	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
678 		if (fts_ufslinks(sp, cur))
679 			nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
680 		else
681 			nlinks = -1;
682 		nostat = 1;
683 	} else {
684 		nlinks = -1;
685 		nostat = 0;
686 	}
687 
688 #ifdef notdef
689 	(void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
690 	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
691 	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
692 #endif
693 	/*
694 	 * If we're going to need to stat anything or we want to descend
695 	 * and stay in the directory, chdir.  If this fails we keep going,
696 	 * but set a flag so we don't chdir after the post-order visit.
697 	 * We won't be able to stat anything, but we can still return the
698 	 * names themselves.  Note, that since fts_read won't be able to
699 	 * chdir into the directory, it will have to return different path
700 	 * names than before, i.e. "a/b" instead of "b".  Since the node
701 	 * has already been visited in pre-order, have to wait until the
702 	 * post-order visit to return the error.  There is a special case
703 	 * here, if there was nothing to stat then it's not an error to
704 	 * not be able to stat.  This is all fairly nasty.  If a program
705 	 * needed sorted entries or stat information, they had better be
706 	 * checking FTS_NS on the returned nodes.
707 	 */
708 	cderrno = 0;
709 	if (nlinks || type == BREAD) {
710 		if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) {
711 			if (nlinks && type == BREAD)
712 				cur->fts_errno = errno;
713 			cur->fts_flags |= FTS_DONTCHDIR;
714 			descend = 0;
715 			cderrno = errno;
716 		} else
717 			descend = 1;
718 	} else
719 		descend = 0;
720 
721 	/*
722 	 * Figure out the max file name length that can be stored in the
723 	 * current path -- the inner loop allocates more path as necessary.
724 	 * We really wouldn't have to do the maxlen calculations here, we
725 	 * could do them in fts_read before returning the path, but it's a
726 	 * lot easier here since the length is part of the dirent structure.
727 	 *
728 	 * If not changing directories set a pointer so that can just append
729 	 * each new name into the path.
730 	 */
731 	len = NAPPEND(cur);
732 	if (ISSET(FTS_NOCHDIR)) {
733 		cp = sp->fts_path + len;
734 		*cp++ = '/';
735 	} else {
736 		/* GCC, you're too verbose. */
737 		cp = NULL;
738 	}
739 	len++;
740 	maxlen = sp->fts_pathlen - len;
741 
742 	level = cur->fts_level + 1;
743 
744 	/* Read the directory, attaching each entry to the `link' pointer. */
745 	doadjust = 0;
746 	for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) {
747 		dnamlen = dp->d_namlen;
748 		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
749 			continue;
750 
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 					free(p);
764 				fts_lfree(head);
765 				(void)closedir(dirp);
766 				cur->fts_info = FTS_ERR;
767 				SET(FTS_STOP);
768 				errno = saved_errno;
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 		p->fts_level = level;
781 		p->fts_parent = sp->fts_cur;
782 		p->fts_pathlen = len + dnamlen;
783 
784 #ifdef FTS_WHITEOUT
785 		if (dp->d_type == DT_WHT)
786 			p->fts_flags |= FTS_ISW;
787 #endif
788 
789 		if (cderrno) {
790 			if (nlinks) {
791 				p->fts_info = FTS_NS;
792 				p->fts_errno = cderrno;
793 			} else
794 				p->fts_info = FTS_NSOK;
795 			p->fts_accpath = cur->fts_accpath;
796 		} else if (nlinks == 0
797 #ifdef DT_DIR
798 		    || (nostat &&
799 		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
800 #endif
801 		    ) {
802 			p->fts_accpath =
803 			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
804 			p->fts_info = FTS_NSOK;
805 		} else {
806 			/* Build a file name for fts_stat to stat. */
807 			if (ISSET(FTS_NOCHDIR)) {
808 				p->fts_accpath = p->fts_path;
809 				memmove(cp, p->fts_name, p->fts_namelen + 1);
810 			} else
811 				p->fts_accpath = p->fts_name;
812 			/* Stat it. */
813 			p->fts_info = fts_stat(sp, p, 0);
814 
815 			/* Decrement link count if applicable. */
816 			if (nlinks > 0 && (p->fts_info == FTS_D ||
817 			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
818 				--nlinks;
819 		}
820 
821 		/* We walk in directory order so "ls -f" doesn't get upset. */
822 		p->fts_link = NULL;
823 		if (head == NULL)
824 			head = tail = p;
825 		else {
826 			tail->fts_link = p;
827 			tail = p;
828 		}
829 		++nitems;
830 	}
831 	if (dirp)
832 		(void)closedir(dirp);
833 
834 	/*
835 	 * If realloc() changed the address of the path, adjust the
836 	 * addresses for the rest of the tree and the dir list.
837 	 */
838 	if (doadjust)
839 		fts_padjust(sp, head);
840 
841 	/*
842 	 * If not changing directories, reset the path back to original
843 	 * state.
844 	 */
845 	if (ISSET(FTS_NOCHDIR))
846 		sp->fts_path[cur->fts_pathlen] = '\0';
847 
848 	/*
849 	 * If descended after called from fts_children or after called from
850 	 * fts_read and nothing found, get back.  At the root level we use
851 	 * the saved fd; if one of fts_open()'s arguments is a relative path
852 	 * to an empty directory, we wind up here with no other way back.  If
853 	 * can't get back, we're done.
854 	 */
855 	if (descend && (type == BCHILD || !nitems) &&
856 	    (cur->fts_level == FTS_ROOTLEVEL ?
857 	    FCHDIR(sp, sp->fts_rfd) :
858 	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
859 		cur->fts_info = FTS_ERR;
860 		SET(FTS_STOP);
861 		return (NULL);
862 	}
863 
864 	/* If didn't find anything, return NULL. */
865 	if (!nitems) {
866 		if (type == BREAD)
867 			cur->fts_info = FTS_DP;
868 		return (NULL);
869 	}
870 
871 	/* Sort the entries. */
872 	if (sp->fts_compar && nitems > 1)
873 		head = fts_sort(sp, head, nitems);
874 	return (head);
875 }
876 
877 static int
878 fts_stat(sp, p, follow)
879 	FTS *sp;
880 	FTSENT *p;
881 	int follow;
882 {
883 	FTSENT *t;
884 	dev_t dev;
885 	ino_t ino;
886 	struct stat *sbp, sb;
887 	int saved_errno;
888 
889 	/* If user needs stat info, stat buffer already allocated. */
890 	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
891 
892 #ifdef FTS_WHITEOUT
893 	/* Check for whiteout. */
894 	if (p->fts_flags & FTS_ISW) {
895 		if (sbp != &sb) {
896 			memset(sbp, '\0', sizeof(*sbp));
897 			sbp->st_mode = S_IFWHT;
898 		}
899 		return (FTS_W);
900 	}
901 #endif
902 
903 	/*
904 	 * If doing a logical walk, or application requested FTS_FOLLOW, do
905 	 * a stat(2).  If that fails, check for a non-existent symlink.  If
906 	 * fail, set the errno from the stat call.
907 	 */
908 	if (ISSET(FTS_LOGICAL) || follow) {
909 		if (stat(p->fts_accpath, sbp)) {
910 			saved_errno = errno;
911 			if (!lstat(p->fts_accpath, sbp)) {
912 				errno = 0;
913 				return (FTS_SLNONE);
914 			}
915 			p->fts_errno = saved_errno;
916 			goto err;
917 		}
918 	} else if (lstat(p->fts_accpath, sbp)) {
919 		p->fts_errno = errno;
920 err:		memset(sbp, 0, sizeof(struct stat));
921 		return (FTS_NS);
922 	}
923 
924 	if (S_ISDIR(sbp->st_mode)) {
925 		/*
926 		 * Set the device/inode.  Used to find cycles and check for
927 		 * crossing mount points.  Also remember the link count, used
928 		 * in fts_build to limit the number of stat calls.  It is
929 		 * understood that these fields are only referenced if fts_info
930 		 * is set to FTS_D.
931 		 */
932 		dev = p->fts_dev = sbp->st_dev;
933 		ino = p->fts_ino = sbp->st_ino;
934 		p->fts_nlink = sbp->st_nlink;
935 
936 		if (ISDOT(p->fts_name))
937 			return (FTS_DOT);
938 
939 		/*
940 		 * Cycle detection is done by brute force when the directory
941 		 * is first encountered.  If the tree gets deep enough or the
942 		 * number of symbolic links to directories is high enough,
943 		 * something faster might be worthwhile.
944 		 */
945 		for (t = p->fts_parent;
946 		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
947 			if (ino == t->fts_ino && dev == t->fts_dev) {
948 				p->fts_cycle = t;
949 				return (FTS_DC);
950 			}
951 		return (FTS_D);
952 	}
953 	if (S_ISLNK(sbp->st_mode))
954 		return (FTS_SL);
955 	if (S_ISREG(sbp->st_mode))
956 		return (FTS_F);
957 	return (FTS_DEFAULT);
958 }
959 
960 /*
961  * The comparison function takes pointers to pointers to FTSENT structures.
962  * Qsort wants a comparison function that takes pointers to void.
963  * (Both with appropriate levels of const-poisoning, of course!)
964  * Use a trampoline function to deal with the difference.
965  */
966 static int
967 fts_compar(const void *a, const void *b)
968 {
969 	FTS *parent;
970 
971 	parent = (*(const FTSENT * const *)a)->fts_fts;
972 	return (*parent->fts_compar)(a, b);
973 }
974 
975 static FTSENT *
976 fts_sort(sp, head, nitems)
977 	FTS *sp;
978 	FTSENT *head;
979 	size_t nitems;
980 {
981 	FTSENT **ap, *p;
982 
983 	/*
984 	 * Construct an array of pointers to the structures and call qsort(3).
985 	 * Reassemble the array in the order returned by qsort.  If unable to
986 	 * sort for memory reasons, return the directory entries in their
987 	 * current order.  Allocate enough space for the current needs plus
988 	 * 40 so don't realloc one entry at a time.
989 	 */
990 	if (nitems > sp->fts_nitems) {
991 		sp->fts_nitems = nitems + 40;
992 		if ((sp->fts_array = reallocf(sp->fts_array,
993 		    sp->fts_nitems * sizeof(FTSENT *))) == NULL) {
994 			sp->fts_nitems = 0;
995 			return (head);
996 		}
997 	}
998 	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
999 		*ap++ = p;
1000 	qsort(sp->fts_array, nitems, sizeof(FTSENT *), fts_compar);
1001 	for (head = *(ap = sp->fts_array); --nitems; ++ap)
1002 		ap[0]->fts_link = ap[1];
1003 	ap[0]->fts_link = NULL;
1004 	return (head);
1005 }
1006 
1007 static FTSENT *
1008 fts_alloc(sp, name, namelen)
1009 	FTS *sp;
1010 	char *name;
1011 	size_t namelen;
1012 {
1013 	FTSENT *p;
1014 	size_t len;
1015 
1016 	struct ftsent_withstat {
1017 		FTSENT	ent;
1018 		struct	stat statbuf;
1019 	};
1020 
1021 	/*
1022 	 * The file name is a variable length array and no stat structure is
1023 	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
1024 	 * structure, the file name and the stat structure in one chunk, but
1025 	 * be careful that the stat structure is reasonably aligned.
1026 	 */
1027 	if (ISSET(FTS_NOSTAT))
1028 		len = sizeof(FTSENT) + namelen + 1;
1029 	else
1030 		len = sizeof(struct ftsent_withstat) + namelen + 1;
1031 
1032 	if ((p = malloc(len)) == NULL)
1033 		return (NULL);
1034 
1035 	if (ISSET(FTS_NOSTAT)) {
1036 		p->fts_name = (char *)(p + 1);
1037 		p->fts_statp = NULL;
1038 	} else {
1039 		p->fts_name = (char *)((struct ftsent_withstat *)p + 1);
1040 		p->fts_statp = &((struct ftsent_withstat *)p)->statbuf;
1041 	}
1042 
1043 	/* Copy the name and guarantee NUL termination. */
1044 	memcpy(p->fts_name, name, namelen);
1045 	p->fts_name[namelen] = '\0';
1046 	p->fts_namelen = namelen;
1047 	p->fts_path = sp->fts_path;
1048 	p->fts_errno = 0;
1049 	p->fts_flags = 0;
1050 	p->fts_instr = FTS_NOINSTR;
1051 	p->fts_number = 0;
1052 	p->fts_pointer = NULL;
1053 	p->fts_fts = sp;
1054 	return (p);
1055 }
1056 
1057 static void
1058 fts_lfree(head)
1059 	FTSENT *head;
1060 {
1061 	FTSENT *p;
1062 
1063 	/* Free a linked list of structures. */
1064 	while ((p = head)) {
1065 		head = head->fts_link;
1066 		free(p);
1067 	}
1068 }
1069 
1070 /*
1071  * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
1072  * Most systems will allow creation of paths much longer than MAXPATHLEN, even
1073  * though the kernel won't resolve them.  Add the size (not just what's needed)
1074  * plus 256 bytes so don't realloc the path 2 bytes at a time.
1075  */
1076 static int
1077 fts_palloc(sp, more)
1078 	FTS *sp;
1079 	size_t more;
1080 {
1081 
1082 	sp->fts_pathlen += more + 256;
1083 	sp->fts_path = reallocf(sp->fts_path, sp->fts_pathlen);
1084 	return (sp->fts_path == NULL);
1085 }
1086 
1087 /*
1088  * When the path is realloc'd, have to fix all of the pointers in structures
1089  * already returned.
1090  */
1091 static void
1092 fts_padjust(sp, head)
1093 	FTS *sp;
1094 	FTSENT *head;
1095 {
1096 	FTSENT *p;
1097 	char *addr = sp->fts_path;
1098 
1099 #define	ADJUST(p) do {							\
1100 	if ((p)->fts_accpath != (p)->fts_name) {			\
1101 		(p)->fts_accpath =					\
1102 		    (char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
1103 	}								\
1104 	(p)->fts_path = addr;						\
1105 } while (0)
1106 	/* Adjust the current set of children. */
1107 	for (p = sp->fts_child; p; p = p->fts_link)
1108 		ADJUST(p);
1109 
1110 	/* Adjust the rest of the tree, including the current level. */
1111 	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1112 		ADJUST(p);
1113 		p = p->fts_link ? p->fts_link : p->fts_parent;
1114 	}
1115 }
1116 
1117 static size_t
1118 fts_maxarglen(argv)
1119 	char * const *argv;
1120 {
1121 	size_t len, max;
1122 
1123 	for (max = 0; *argv; ++argv)
1124 		if ((len = strlen(*argv)) > max)
1125 			max = len;
1126 	return (max + 1);
1127 }
1128 
1129 /*
1130  * Change to dir specified by fd or p->fts_accpath without getting
1131  * tricked by someone changing the world out from underneath us.
1132  * Assumes p->fts_dev and p->fts_ino are filled in.
1133  */
1134 static int
1135 fts_safe_changedir(sp, p, fd, path)
1136 	FTS *sp;
1137 	FTSENT *p;
1138 	int fd;
1139 	char *path;
1140 {
1141 	int ret, oerrno, newfd;
1142 	struct stat sb;
1143 
1144 	newfd = fd;
1145 	if (ISSET(FTS_NOCHDIR))
1146 		return (0);
1147 	if (fd < 0 && (newfd = _open(path, O_RDONLY, 0)) < 0)
1148 		return (-1);
1149 	if (_fstat(newfd, &sb)) {
1150 		ret = -1;
1151 		goto bail;
1152 	}
1153 	if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) {
1154 		errno = ENOENT;		/* disinformation */
1155 		ret = -1;
1156 		goto bail;
1157 	}
1158 	ret = fchdir(newfd);
1159 bail:
1160 	oerrno = errno;
1161 	if (fd < 0)
1162 		(void)_close(newfd);
1163 	errno = oerrno;
1164 	return (ret);
1165 }
1166 
1167 /*
1168  * Check if the filesystem for "ent" has UFS-style links.
1169  */
1170 static int
1171 fts_ufslinks(FTS *sp, const FTSENT *ent)
1172 {
1173 	struct _fts_private *priv;
1174 	const char **cpp;
1175 
1176 	priv = (struct _fts_private *)sp;
1177 	/*
1178 	 * If this node's device is different from the previous, grab
1179 	 * the filesystem information, and decide on the reliability
1180 	 * of the link information from this filesystem for stat(2)
1181 	 * avoidance.
1182 	 */
1183 	if (priv->ftsp_dev != ent->fts_dev) {
1184 		if (statfs(ent->fts_path, &priv->ftsp_statfs) != -1) {
1185 			priv->ftsp_dev = ent->fts_dev;
1186 			priv->ftsp_linksreliable = 0;
1187 			for (cpp = ufslike_filesystems; *cpp; cpp++) {
1188 				if (strcmp(priv->ftsp_statfs.f_fstypename,
1189 				    *cpp) == 0) {
1190 					priv->ftsp_linksreliable = 1;
1191 					break;
1192 				}
1193 			}
1194 		} else {
1195 			priv->ftsp_linksreliable = 0;
1196 		}
1197 	}
1198 	return (priv->ftsp_linksreliable);
1199 }
1200