xref: /freebsd/lib/libc/gen/fts.c (revision 4b2eaea43fec8e8792be611dea204071a10b655a)
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  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  * $OpenBSD: fts.c,v 1.22 1999/10/03 19:22:22 millert Exp $
34  */
35 
36 #if defined(LIBC_SCCS) && !defined(lint)
37 static char sccsid[] = "@(#)fts.c	8.6 (Berkeley) 8/14/94";
38 #endif /* LIBC_SCCS and not lint */
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include "namespace.h"
43 #include <sys/types.h>
44 #include <sys/param.h>
45 #include <sys/stat.h>
46 
47 #include <dirent.h>
48 #include <errno.h>
49 #include <fcntl.h>
50 #include <fts.h>
51 #include <stdlib.h>
52 #include <string.h>
53 #include <unistd.h>
54 #include "un-namespace.h"
55 
56 static FTSENT	*fts_alloc(FTS *, char *, int);
57 static FTSENT	*fts_build(FTS *, int);
58 static void	 fts_lfree(FTSENT *);
59 static void	 fts_load(FTS *, FTSENT *);
60 static size_t	 fts_maxarglen(char * const *);
61 static void	 fts_padjust(FTS *, FTSENT *);
62 static int	 fts_palloc(FTS *, size_t);
63 static FTSENT	*fts_sort(FTS *, FTSENT *, int);
64 static u_short	 fts_stat(FTS *, FTSENT *, int);
65 static int	 fts_safe_changedir(FTS *, FTSENT *, int, char *);
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 FTS *
81 fts_open(argv, options, compar)
82 	char * const *argv;
83 	int options;
84 	int (*compar)(const FTSENT * const *, const FTSENT * const *);
85 {
86 	FTS *sp;
87 	FTSENT *p, *root;
88 	int nitems;
89 	FTSENT *parent, *tmp;
90 	int len;
91 
92 	/* Options check. */
93 	if (options & ~FTS_OPTIONMASK) {
94 		errno = EINVAL;
95 		return (NULL);
96 	}
97 
98 	/* Allocate/initialize the stream */
99 	if ((sp = malloc(sizeof(FTS))) == NULL)
100 		return (NULL);
101 	memset(sp, 0, sizeof(FTS));
102 	sp->fts_compar = compar;
103 	sp->fts_options = options;
104 
105 	/* Shush, GCC. */
106 	tmp = NULL;
107 
108 	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
109 	if (ISSET(FTS_LOGICAL))
110 		SET(FTS_NOCHDIR);
111 
112 	/*
113 	 * Start out with 1K of path space, and enough, in any case,
114 	 * to hold the user's paths.
115 	 */
116 	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
117 		goto mem1;
118 
119 	/* Allocate/initialize root's parent. */
120 	if ((parent = fts_alloc(sp, "", 0)) == NULL)
121 		goto mem2;
122 	parent->fts_level = FTS_ROOTPARENTLEVEL;
123 
124 	/* Allocate/initialize root(s). */
125 	for (root = NULL, nitems = 0; *argv != NULL; ++argv, ++nitems) {
126 		/* Don't allow zero-length paths. */
127 		if ((len = strlen(*argv)) == 0) {
128 			errno = ENOENT;
129 			goto mem3;
130 		}
131 
132 		p = fts_alloc(sp, *argv, len);
133 		p->fts_level = FTS_ROOTLEVEL;
134 		p->fts_parent = parent;
135 		p->fts_accpath = p->fts_name;
136 		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW));
137 
138 		/* Command-line "." and ".." are real directories. */
139 		if (p->fts_info == FTS_DOT)
140 			p->fts_info = FTS_D;
141 
142 		/*
143 		 * If comparison routine supplied, traverse in sorted
144 		 * order; otherwise traverse in the order specified.
145 		 */
146 		if (compar) {
147 			p->fts_link = root;
148 			root = p;
149 		} else {
150 			p->fts_link = NULL;
151 			if (root == NULL)
152 				tmp = root = p;
153 			else {
154 				tmp->fts_link = p;
155 				tmp = p;
156 			}
157 		}
158 	}
159 	if (compar && nitems > 1)
160 		root = fts_sort(sp, root, nitems);
161 
162 	/*
163 	 * Allocate a dummy pointer and make fts_read think that we've just
164 	 * finished the node before the root(s); set p->fts_info to FTS_INIT
165 	 * so that everything about the "current" node is ignored.
166 	 */
167 	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
168 		goto mem3;
169 	sp->fts_cur->fts_link = root;
170 	sp->fts_cur->fts_info = FTS_INIT;
171 
172 	/*
173 	 * If using chdir(2), grab a file descriptor pointing to dot to ensure
174 	 * that we can get back here; this could be avoided for some paths,
175 	 * but almost certainly not worth the effort.  Slashes, symbolic links,
176 	 * and ".." are all fairly nasty problems.  Note, if we can't get the
177 	 * descriptor we run anyway, just more slowly.
178 	 */
179 	if (!ISSET(FTS_NOCHDIR) && (sp->fts_rfd = _open(".", O_RDONLY, 0)) < 0)
180 		SET(FTS_NOCHDIR);
181 
182 	return (sp);
183 
184 mem3:	fts_lfree(root);
185 	free(parent);
186 mem2:	free(sp->fts_path);
187 mem1:	free(sp);
188 	return (NULL);
189 }
190 
191 static void
192 fts_load(sp, p)
193 	FTS *sp;
194 	FTSENT *p;
195 {
196 	int len;
197 	char *cp;
198 
199 	/*
200 	 * Load the stream structure for the next traversal.  Since we don't
201 	 * actually enter the directory until after the preorder visit, set
202 	 * the fts_accpath field specially so the chdir gets done to the right
203 	 * place and the user can access the first node.  From fts_open it's
204 	 * known that the path will fit.
205 	 */
206 	len = p->fts_pathlen = p->fts_namelen;
207 	memmove(sp->fts_path, p->fts_name, len + 1);
208 	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
209 		len = strlen(++cp);
210 		memmove(p->fts_name, cp, len + 1);
211 		p->fts_namelen = len;
212 	}
213 	p->fts_accpath = p->fts_path = sp->fts_path;
214 	sp->fts_dev = p->fts_dev;
215 }
216 
217 int
218 fts_close(sp)
219 	FTS *sp;
220 {
221 	FTSENT *freep, *p;
222 	int saved_errno;
223 
224 	/*
225 	 * This still works if we haven't read anything -- the dummy structure
226 	 * points to the root list, so we step through to the end of the root
227 	 * list which has a valid parent pointer.
228 	 */
229 	if (sp->fts_cur) {
230 		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
231 			freep = p;
232 			p = p->fts_link != NULL ? p->fts_link : p->fts_parent;
233 			free(freep);
234 		}
235 		free(p);
236 	}
237 
238 	/* Free up child linked list, sort array, path buffer. */
239 	if (sp->fts_child)
240 		fts_lfree(sp->fts_child);
241 	if (sp->fts_array)
242 		free(sp->fts_array);
243 	free(sp->fts_path);
244 
245 	/* Return to original directory, save errno if necessary. */
246 	if (!ISSET(FTS_NOCHDIR)) {
247 		saved_errno = fchdir(sp->fts_rfd) ? errno : 0;
248 		(void)_close(sp->fts_rfd);
249 
250 		/* Set errno and return. */
251 		if (saved_errno != 0) {
252 			/* Free up the stream pointer. */
253 			free(sp);
254 			errno = saved_errno;
255 			return (-1);
256 		}
257 	}
258 
259 	/* Free up the stream pointer. */
260 	free(sp);
261 	return (0);
262 }
263 
264 /*
265  * Special case of "/" at the end of the path so that slashes aren't
266  * appended which would cause paths to be written as "....//foo".
267  */
268 #define	NAPPEND(p)							\
269 	(p->fts_path[p->fts_pathlen - 1] == '/'				\
270 	    ? p->fts_pathlen - 1 : p->fts_pathlen)
271 
272 FTSENT *
273 fts_read(sp)
274 	FTS *sp;
275 {
276 	FTSENT *p, *tmp;
277 	int instr;
278 	char *t;
279 	int saved_errno;
280 
281 	/* If finished or unrecoverable error, return NULL. */
282 	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
283 		return (NULL);
284 
285 	/* Set current node pointer. */
286 	p = sp->fts_cur;
287 
288 	/* Save and zero out user instructions. */
289 	instr = p->fts_instr;
290 	p->fts_instr = FTS_NOINSTR;
291 
292 	/* Any type of file may be re-visited; re-stat and re-turn. */
293 	if (instr == FTS_AGAIN) {
294 		p->fts_info = fts_stat(sp, p, 0);
295 		return (p);
296 	}
297 
298 	/*
299 	 * Following a symlink -- SLNONE test allows application to see
300 	 * SLNONE and recover.  If indirecting through a symlink, have
301 	 * keep a pointer to current location.  If unable to get that
302 	 * pointer, follow fails.
303 	 */
304 	if (instr == FTS_FOLLOW &&
305 	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
306 		p->fts_info = fts_stat(sp, p, 1);
307 		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
308 			if ((p->fts_symfd = _open(".", O_RDONLY, 0)) < 0) {
309 				p->fts_errno = errno;
310 				p->fts_info = FTS_ERR;
311 			} else
312 				p->fts_flags |= FTS_SYMFOLLOW;
313 		}
314 		return (p);
315 	}
316 
317 	/* Directory in pre-order. */
318 	if (p->fts_info == FTS_D) {
319 		/* If skipped or crossed mount point, do post-order visit. */
320 		if (instr == FTS_SKIP ||
321 		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
322 			if (p->fts_flags & FTS_SYMFOLLOW)
323 				(void)_close(p->fts_symfd);
324 			if (sp->fts_child) {
325 				fts_lfree(sp->fts_child);
326 				sp->fts_child = NULL;
327 			}
328 			p->fts_info = FTS_DP;
329 			return (p);
330 		}
331 
332 		/* Rebuild if only read the names and now traversing. */
333 		if (sp->fts_child != NULL && ISSET(FTS_NAMEONLY)) {
334 			CLR(FTS_NAMEONLY);
335 			fts_lfree(sp->fts_child);
336 			sp->fts_child = NULL;
337 		}
338 
339 		/*
340 		 * Cd to the subdirectory.
341 		 *
342 		 * If have already read and now fail to chdir, whack the list
343 		 * to make the names come out right, and set the parent errno
344 		 * so the application will eventually get an error condition.
345 		 * Set the FTS_DONTCHDIR flag so that when we logically change
346 		 * directories back to the parent we don't do a chdir.
347 		 *
348 		 * If haven't read do so.  If the read fails, fts_build sets
349 		 * FTS_STOP or the fts_info field of the node.
350 		 */
351 		if (sp->fts_child != NULL) {
352 			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
353 				p->fts_errno = errno;
354 				p->fts_flags |= FTS_DONTCHDIR;
355 				for (p = sp->fts_child; p != NULL;
356 				    p = p->fts_link)
357 					p->fts_accpath =
358 					    p->fts_parent->fts_accpath;
359 			}
360 		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
361 			if (ISSET(FTS_STOP))
362 				return (NULL);
363 			return (p);
364 		}
365 		p = sp->fts_child;
366 		sp->fts_child = NULL;
367 		goto name;
368 	}
369 
370 	/* Move to the next node on this level. */
371 next:	tmp = p;
372 	if ((p = p->fts_link) != NULL) {
373 		free(tmp);
374 
375 		/*
376 		 * If reached the top, return to the original directory (or
377 		 * the root of the tree), and load the paths for the next root.
378 		 */
379 		if (p->fts_level == FTS_ROOTLEVEL) {
380 			if (FCHDIR(sp, sp->fts_rfd)) {
381 				SET(FTS_STOP);
382 				return (NULL);
383 			}
384 			fts_load(sp, p);
385 			return (sp->fts_cur = p);
386 		}
387 
388 		/*
389 		 * User may have called fts_set on the node.  If skipped,
390 		 * ignore.  If followed, get a file descriptor so we can
391 		 * get back if necessary.
392 		 */
393 		if (p->fts_instr == FTS_SKIP)
394 			goto next;
395 		if (p->fts_instr == FTS_FOLLOW) {
396 			p->fts_info = fts_stat(sp, p, 1);
397 			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
398 				if ((p->fts_symfd =
399 				    _open(".", O_RDONLY, 0)) < 0) {
400 					p->fts_errno = errno;
401 					p->fts_info = FTS_ERR;
402 				} else
403 					p->fts_flags |= FTS_SYMFOLLOW;
404 			}
405 			p->fts_instr = FTS_NOINSTR;
406 		}
407 
408 name:		t = sp->fts_path + NAPPEND(p->fts_parent);
409 		*t++ = '/';
410 		memmove(t, p->fts_name, p->fts_namelen + 1);
411 		return (sp->fts_cur = p);
412 	}
413 
414 	/* Move up to the parent node. */
415 	p = tmp->fts_parent;
416 	free(tmp);
417 
418 	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
419 		/*
420 		 * Done; free everything up and set errno to 0 so the user
421 		 * can distinguish between error and EOF.
422 		 */
423 		free(p);
424 		errno = 0;
425 		return (sp->fts_cur = NULL);
426 	}
427 
428 	/* NUL terminate the pathname. */
429 	sp->fts_path[p->fts_pathlen] = '\0';
430 
431 	/*
432 	 * Return to the parent directory.  If at a root node or came through
433 	 * a symlink, go back through the file descriptor.  Otherwise, cd up
434 	 * one directory.
435 	 */
436 	if (p->fts_level == FTS_ROOTLEVEL) {
437 		if (FCHDIR(sp, sp->fts_rfd)) {
438 			SET(FTS_STOP);
439 			return (NULL);
440 		}
441 	} else if (p->fts_flags & FTS_SYMFOLLOW) {
442 		if (FCHDIR(sp, p->fts_symfd)) {
443 			saved_errno = errno;
444 			(void)_close(p->fts_symfd);
445 			errno = saved_errno;
446 			SET(FTS_STOP);
447 			return (NULL);
448 		}
449 		(void)_close(p->fts_symfd);
450 	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
451 		   fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
452 		SET(FTS_STOP);
453 		return (NULL);
454 	}
455 	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
456 	return (sp->fts_cur = p);
457 }
458 
459 /*
460  * Fts_set takes the stream as an argument although it's not used in this
461  * implementation; it would be necessary if anyone wanted to add global
462  * semantics to fts using fts_set.  An error return is allowed for similar
463  * reasons.
464  */
465 /* ARGSUSED */
466 int
467 fts_set(sp, p, instr)
468 	FTS *sp;
469 	FTSENT *p;
470 	int instr;
471 {
472 	if (instr != 0 && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
473 	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
474 		errno = EINVAL;
475 		return (1);
476 	}
477 	p->fts_instr = instr;
478 	return (0);
479 }
480 
481 FTSENT *
482 fts_children(sp, instr)
483 	FTS *sp;
484 	int instr;
485 {
486 	FTSENT *p;
487 	int fd;
488 
489 	if (instr != 0 && instr != FTS_NAMEONLY) {
490 		errno = EINVAL;
491 		return (NULL);
492 	}
493 
494 	/* Set current node pointer. */
495 	p = sp->fts_cur;
496 
497 	/*
498 	 * Errno set to 0 so user can distinguish empty directory from
499 	 * an error.
500 	 */
501 	errno = 0;
502 
503 	/* Fatal errors stop here. */
504 	if (ISSET(FTS_STOP))
505 		return (NULL);
506 
507 	/* Return logical hierarchy of user's arguments. */
508 	if (p->fts_info == FTS_INIT)
509 		return (p->fts_link);
510 
511 	/*
512 	 * If not a directory being visited in pre-order, stop here.  Could
513 	 * allow FTS_DNR, assuming the user has fixed the problem, but the
514 	 * same effect is available with FTS_AGAIN.
515 	 */
516 	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
517 		return (NULL);
518 
519 	/* Free up any previous child list. */
520 	if (sp->fts_child != NULL)
521 		fts_lfree(sp->fts_child);
522 
523 	if (instr == FTS_NAMEONLY) {
524 		SET(FTS_NAMEONLY);
525 		instr = BNAMES;
526 	} else
527 		instr = BCHILD;
528 
529 	/*
530 	 * If using chdir on a relative path and called BEFORE fts_read does
531 	 * its chdir to the root of a traversal, we can lose -- we need to
532 	 * chdir into the subdirectory, and we don't know where the current
533 	 * directory is, so we can't get back so that the upcoming chdir by
534 	 * fts_read will work.
535 	 */
536 	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
537 	    ISSET(FTS_NOCHDIR))
538 		return (sp->fts_child = fts_build(sp, instr));
539 
540 	if ((fd = _open(".", O_RDONLY, 0)) < 0)
541 		return (NULL);
542 	sp->fts_child = fts_build(sp, instr);
543 	if (fchdir(fd))
544 		return (NULL);
545 	(void)_close(fd);
546 	return (sp->fts_child);
547 }
548 
549 #ifndef fts_get_clientptr
550 #error "fts_get_clientptr not defined"
551 #endif
552 
553 void *
554 (fts_get_clientptr)(FTS *sp)
555 {
556 
557 	return (fts_get_clientptr(sp));
558 }
559 
560 #ifndef fts_get_stream
561 #error "fts_get_stream not defined"
562 #endif
563 
564 FTS *
565 (fts_get_stream)(FTSENT *p)
566 {
567 	return (fts_get_stream(p));
568 }
569 
570 void
571 fts_set_clientptr(FTS *sp, void *clientptr)
572 {
573 
574 	sp->fts_clientptr = clientptr;
575 }
576 
577 /*
578  * This is the tricky part -- do not casually change *anything* in here.  The
579  * idea is to build the linked list of entries that are used by fts_children
580  * and fts_read.  There are lots of special cases.
581  *
582  * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
583  * set and it's a physical walk (so that symbolic links can't be directories),
584  * we can do things quickly.  First, if it's a 4.4BSD file system, the type
585  * of the file is in the directory entry.  Otherwise, we assume that the number
586  * of subdirectories in a node is equal to the number of links to the parent.
587  * The former skips all stat calls.  The latter skips stat calls in any leaf
588  * directories and for any files after the subdirectories in the directory have
589  * been found, cutting the stat calls by about 2/3.
590  */
591 static FTSENT *
592 fts_build(sp, type)
593 	FTS *sp;
594 	int type;
595 {
596 	struct dirent *dp;
597 	FTSENT *p, *head;
598 	int nitems;
599 	FTSENT *cur, *tail;
600 	DIR *dirp;
601 	void *oldaddr;
602 	int cderrno, descend, len, level, maxlen, nlinks, oflag, saved_errno,
603 	    nostat, doadjust;
604 	char *cp;
605 
606 	/* Set current node pointer. */
607 	cur = sp->fts_cur;
608 
609 	/*
610 	 * Open the directory for reading.  If this fails, we're done.
611 	 * If being called from fts_read, set the fts_info field.
612 	 */
613 #ifdef FTS_WHITEOUT
614 	if (ISSET(FTS_WHITEOUT))
615 		oflag = DTF_NODUP|DTF_REWIND;
616 	else
617 		oflag = DTF_HIDEW|DTF_NODUP|DTF_REWIND;
618 #else
619 #define __opendir2(path, flag) opendir(path)
620 #endif
621 	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
622 		if (type == BREAD) {
623 			cur->fts_info = FTS_DNR;
624 			cur->fts_errno = errno;
625 		}
626 		return (NULL);
627 	}
628 
629 	/*
630 	 * Nlinks is the number of possible entries of type directory in the
631 	 * directory if we're cheating on stat calls, 0 if we're not doing
632 	 * any stat calls at all, -1 if we're doing stats on everything.
633 	 */
634 	if (type == BNAMES) {
635 		nlinks = 0;
636 		/* Be quiet about nostat, GCC. */
637 		nostat = 0;
638 	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
639 		nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
640 		nostat = 1;
641 	} else {
642 		nlinks = -1;
643 		nostat = 0;
644 	}
645 
646 #ifdef notdef
647 	(void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
648 	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
649 	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
650 #endif
651 	/*
652 	 * If we're going to need to stat anything or we want to descend
653 	 * and stay in the directory, chdir.  If this fails we keep going,
654 	 * but set a flag so we don't chdir after the post-order visit.
655 	 * We won't be able to stat anything, but we can still return the
656 	 * names themselves.  Note, that since fts_read won't be able to
657 	 * chdir into the directory, it will have to return different path
658 	 * names than before, i.e. "a/b" instead of "b".  Since the node
659 	 * has already been visited in pre-order, have to wait until the
660 	 * post-order visit to return the error.  There is a special case
661 	 * here, if there was nothing to stat then it's not an error to
662 	 * not be able to stat.  This is all fairly nasty.  If a program
663 	 * needed sorted entries or stat information, they had better be
664 	 * checking FTS_NS on the returned nodes.
665 	 */
666 	cderrno = 0;
667 	if (nlinks || type == BREAD) {
668 		if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) {
669 			if (nlinks && type == BREAD)
670 				cur->fts_errno = errno;
671 			cur->fts_flags |= FTS_DONTCHDIR;
672 			descend = 0;
673 			cderrno = errno;
674 			(void)closedir(dirp);
675 			dirp = NULL;
676 		} else
677 			descend = 1;
678 	} else
679 		descend = 0;
680 
681 	/*
682 	 * Figure out the max file name length that can be stored in the
683 	 * current path -- the inner loop allocates more path as necessary.
684 	 * We really wouldn't have to do the maxlen calculations here, we
685 	 * could do them in fts_read before returning the path, but it's a
686 	 * lot easier here since the length is part of the dirent structure.
687 	 *
688 	 * If not changing directories set a pointer so that can just append
689 	 * each new name into the path.
690 	 */
691 	len = NAPPEND(cur);
692 	if (ISSET(FTS_NOCHDIR)) {
693 		cp = sp->fts_path + len;
694 		*cp++ = '/';
695 	} else {
696 		/* GCC, you're too verbose. */
697 		cp = NULL;
698 	}
699 	len++;
700 	maxlen = sp->fts_pathlen - len;
701 
702 	level = cur->fts_level + 1;
703 
704 	/* Read the directory, attaching each entry to the `link' pointer. */
705 	doadjust = 0;
706 	for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) {
707 		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
708 			continue;
709 
710 		if ((p = fts_alloc(sp, dp->d_name, (int)dp->d_namlen)) == NULL)
711 			goto mem1;
712 		if (dp->d_namlen >= maxlen) {	/* include space for NUL */
713 			oldaddr = sp->fts_path;
714 			if (fts_palloc(sp, dp->d_namlen + len + 1)) {
715 				/*
716 				 * No more memory for path or structures.  Save
717 				 * errno, free up the current structure and the
718 				 * structures already allocated.
719 				 */
720 mem1:				saved_errno = errno;
721 				if (p)
722 					free(p);
723 				fts_lfree(head);
724 				(void)closedir(dirp);
725 				cur->fts_info = FTS_ERR;
726 				SET(FTS_STOP);
727 				errno = saved_errno;
728 				return (NULL);
729 			}
730 			/* Did realloc() change the pointer? */
731 			if (oldaddr != sp->fts_path) {
732 				doadjust = 1;
733 				if (ISSET(FTS_NOCHDIR))
734 					cp = sp->fts_path + len;
735 			}
736 			maxlen = sp->fts_pathlen - len;
737 		}
738 
739 		if (len + dp->d_namlen >= USHRT_MAX) {
740 			/*
741 			 * In an FTSENT, fts_pathlen is a u_short so it is
742 			 * possible to wraparound here.  If we do, free up
743 			 * the current structure and the structures already
744 			 * allocated, then error out with ENAMETOOLONG.
745 			 */
746 			free(p);
747 			fts_lfree(head);
748 			(void)closedir(dirp);
749 			cur->fts_info = FTS_ERR;
750 			SET(FTS_STOP);
751 			errno = ENAMETOOLONG;
752 			return (NULL);
753 		}
754 		p->fts_level = level;
755 		p->fts_parent = sp->fts_cur;
756 		p->fts_pathlen = len + dp->d_namlen;
757 
758 #ifdef FTS_WHITEOUT
759 		if (dp->d_type == DT_WHT)
760 			p->fts_flags |= FTS_ISW;
761 #endif
762 
763 		if (cderrno) {
764 			if (nlinks) {
765 				p->fts_info = FTS_NS;
766 				p->fts_errno = cderrno;
767 			} else
768 				p->fts_info = FTS_NSOK;
769 			p->fts_accpath = cur->fts_accpath;
770 		} else if (nlinks == 0
771 #ifdef DT_DIR
772 		    || (nostat &&
773 		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
774 #endif
775 		    ) {
776 			p->fts_accpath =
777 			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
778 			p->fts_info = FTS_NSOK;
779 		} else {
780 			/* Build a file name for fts_stat to stat. */
781 			if (ISSET(FTS_NOCHDIR)) {
782 				p->fts_accpath = p->fts_path;
783 				memmove(cp, p->fts_name, p->fts_namelen + 1);
784 			} else
785 				p->fts_accpath = p->fts_name;
786 			/* Stat it. */
787 			p->fts_info = fts_stat(sp, p, 0);
788 
789 			/* Decrement link count if applicable. */
790 			if (nlinks > 0 && (p->fts_info == FTS_D ||
791 			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
792 				--nlinks;
793 		}
794 
795 		/* We walk in directory order so "ls -f" doesn't get upset. */
796 		p->fts_link = NULL;
797 		if (head == NULL)
798 			head = tail = p;
799 		else {
800 			tail->fts_link = p;
801 			tail = p;
802 		}
803 		++nitems;
804 	}
805 	if (dirp)
806 		(void)closedir(dirp);
807 
808 	/*
809 	 * If realloc() changed the address of the path, adjust the
810 	 * addresses for the rest of the tree and the dir list.
811 	 */
812 	if (doadjust)
813 		fts_padjust(sp, head);
814 
815 	/*
816 	 * If not changing directories, reset the path back to original
817 	 * state.
818 	 */
819 	if (ISSET(FTS_NOCHDIR)) {
820 		if (len == sp->fts_pathlen || nitems == 0)
821 			--cp;
822 		*cp = '\0';
823 	}
824 
825 	/*
826 	 * If descended after called from fts_children or after called from
827 	 * fts_read and nothing found, get back.  At the root level we use
828 	 * the saved fd; if one of fts_open()'s arguments is a relative path
829 	 * to an empty directory, we wind up here with no other way back.  If
830 	 * can't get back, we're done.
831 	 */
832 	if (descend && (type == BCHILD || !nitems) &&
833 	    (cur->fts_level == FTS_ROOTLEVEL ?
834 	    FCHDIR(sp, sp->fts_rfd) :
835 	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
836 		cur->fts_info = FTS_ERR;
837 		SET(FTS_STOP);
838 		return (NULL);
839 	}
840 
841 	/* If didn't find anything, return NULL. */
842 	if (!nitems) {
843 		if (type == BREAD)
844 			cur->fts_info = FTS_DP;
845 		return (NULL);
846 	}
847 
848 	/* Sort the entries. */
849 	if (sp->fts_compar && nitems > 1)
850 		head = fts_sort(sp, head, nitems);
851 	return (head);
852 }
853 
854 static u_short
855 fts_stat(sp, p, follow)
856 	FTS *sp;
857 	FTSENT *p;
858 	int follow;
859 {
860 	FTSENT *t;
861 	dev_t dev;
862 	ino_t ino;
863 	struct stat *sbp, sb;
864 	int saved_errno;
865 
866 	/* If user needs stat info, stat buffer already allocated. */
867 	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
868 
869 #ifdef FTS_WHITEOUT
870 	/* check for whiteout */
871 	if (p->fts_flags & FTS_ISW) {
872 		if (sbp != &sb) {
873 			memset(sbp, '\0', sizeof (*sbp));
874 			sbp->st_mode = S_IFWHT;
875 		}
876 		return (FTS_W);
877 	}
878 #endif
879 
880 	/*
881 	 * If doing a logical walk, or application requested FTS_FOLLOW, do
882 	 * a stat(2).  If that fails, check for a non-existent symlink.  If
883 	 * fail, set the errno from the stat call.
884 	 */
885 	if (ISSET(FTS_LOGICAL) || follow) {
886 		if (stat(p->fts_accpath, sbp)) {
887 			saved_errno = errno;
888 			if (!lstat(p->fts_accpath, sbp)) {
889 				errno = 0;
890 				return (FTS_SLNONE);
891 			}
892 			p->fts_errno = saved_errno;
893 			goto err;
894 		}
895 	} else if (lstat(p->fts_accpath, sbp)) {
896 		p->fts_errno = errno;
897 err:		memset(sbp, 0, sizeof(struct stat));
898 		return (FTS_NS);
899 	}
900 
901 	if (S_ISDIR(sbp->st_mode)) {
902 		/*
903 		 * Set the device/inode.  Used to find cycles and check for
904 		 * crossing mount points.  Also remember the link count, used
905 		 * in fts_build to limit the number of stat calls.  It is
906 		 * understood that these fields are only referenced if fts_info
907 		 * is set to FTS_D.
908 		 */
909 		dev = p->fts_dev = sbp->st_dev;
910 		ino = p->fts_ino = sbp->st_ino;
911 		p->fts_nlink = sbp->st_nlink;
912 
913 		if (ISDOT(p->fts_name))
914 			return (FTS_DOT);
915 
916 		/*
917 		 * Cycle detection is done by brute force when the directory
918 		 * is first encountered.  If the tree gets deep enough or the
919 		 * number of symbolic links to directories is high enough,
920 		 * something faster might be worthwhile.
921 		 */
922 		for (t = p->fts_parent;
923 		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
924 			if (ino == t->fts_ino && dev == t->fts_dev) {
925 				p->fts_cycle = t;
926 				return (FTS_DC);
927 			}
928 		return (FTS_D);
929 	}
930 	if (S_ISLNK(sbp->st_mode))
931 		return (FTS_SL);
932 	if (S_ISREG(sbp->st_mode))
933 		return (FTS_F);
934 	return (FTS_DEFAULT);
935 }
936 
937 /*
938  * The comparison function takes pointers to pointers to FTSENT structures.
939  * Qsort wants a comparison function that takes pointers to void.
940  * (Both with appropriate levels of const-poisoning, of course!)
941  * Use a trampoline function to deal with the difference.
942  */
943 static int
944 fts_compar(const void *a, const void *b)
945 {
946 	FTS *parent;
947 
948 	parent = (*(const FTSENT * const *)a)->fts_fts;
949 	return (*parent->fts_compar)(a, b);
950 }
951 
952 static FTSENT *
953 fts_sort(sp, head, nitems)
954 	FTS *sp;
955 	FTSENT *head;
956 	int nitems;
957 {
958 	FTSENT **ap, *p;
959 
960 	/*
961 	 * Construct an array of pointers to the structures and call qsort(3).
962 	 * Reassemble the array in the order returned by qsort.  If unable to
963 	 * sort for memory reasons, return the directory entries in their
964 	 * current order.  Allocate enough space for the current needs plus
965 	 * 40 so don't realloc one entry at a time.
966 	 */
967 	if (nitems > sp->fts_nitems) {
968 		sp->fts_nitems = nitems + 40;
969 		if ((sp->fts_array = reallocf(sp->fts_array,
970 		    sp->fts_nitems * sizeof(FTSENT *))) == NULL) {
971 			sp->fts_nitems = 0;
972 			return (head);
973 		}
974 	}
975 	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
976 		*ap++ = p;
977 	qsort(sp->fts_array, nitems, sizeof(FTSENT *), fts_compar);
978 	for (head = *(ap = sp->fts_array); --nitems; ++ap)
979 		ap[0]->fts_link = ap[1];
980 	ap[0]->fts_link = NULL;
981 	return (head);
982 }
983 
984 static FTSENT *
985 fts_alloc(sp, name, namelen)
986 	FTS *sp;
987 	char *name;
988 	int namelen;
989 {
990 	FTSENT *p;
991 	size_t len;
992 
993 	struct ftsent_withstat {
994 		FTSENT	ent;
995 		struct	stat statbuf;
996 	};
997 
998 	/*
999 	 * The file name is a variable length array and no stat structure is
1000 	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
1001 	 * structure, the file name and the stat structure in one chunk, but
1002 	 * be careful that the stat structure is reasonably aligned.
1003 	 */
1004 	if (ISSET(FTS_NOSTAT))
1005 		len = sizeof(FTSENT) + namelen + 1;
1006 	else
1007 		len = sizeof(struct ftsent_withstat) + namelen + 1;
1008 
1009 	if ((p = malloc(len)) == NULL)
1010 		return (NULL);
1011 
1012 	if (ISSET(FTS_NOSTAT)) {
1013 		p->fts_name = (char *)(p + 1);
1014 		p->fts_statp = NULL;
1015 	} else {
1016 		p->fts_name = (char *)((struct ftsent_withstat *)p + 1);
1017 		p->fts_statp = &((struct ftsent_withstat *)p)->statbuf;
1018 	}
1019 
1020 	/* Copy the name and guarantee NUL termination. */
1021 	memcpy(p->fts_name, name, namelen);
1022 	p->fts_name[namelen] = '\0';
1023 	p->fts_namelen = namelen;
1024 	p->fts_path = sp->fts_path;
1025 	p->fts_errno = 0;
1026 	p->fts_flags = 0;
1027 	p->fts_instr = FTS_NOINSTR;
1028 	p->fts_number = 0;
1029 	p->fts_pointer = NULL;
1030 	p->fts_fts = sp;
1031 	return (p);
1032 }
1033 
1034 static void
1035 fts_lfree(head)
1036 	FTSENT *head;
1037 {
1038 	FTSENT *p;
1039 
1040 	/* Free a linked list of structures. */
1041 	while ((p = head)) {
1042 		head = head->fts_link;
1043 		free(p);
1044 	}
1045 }
1046 
1047 /*
1048  * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
1049  * Most systems will allow creation of paths much longer than MAXPATHLEN, even
1050  * though the kernel won't resolve them.  Add the size (not just what's needed)
1051  * plus 256 bytes so don't realloc the path 2 bytes at a time.
1052  */
1053 static int
1054 fts_palloc(sp, more)
1055 	FTS *sp;
1056 	size_t more;
1057 {
1058 
1059 	sp->fts_pathlen += more + 256;
1060 	/*
1061 	 * Check for possible wraparound.  In an FTS, fts_pathlen is
1062 	 * a signed int but in an FTSENT it is an unsigned short.
1063 	 * We limit fts_pathlen to USHRT_MAX to be safe in both cases.
1064 	 */
1065 	if (sp->fts_pathlen < 0 || sp->fts_pathlen >= USHRT_MAX) {
1066 		if (sp->fts_path)
1067 			free(sp->fts_path);
1068 		sp->fts_path = NULL;
1069 		errno = ENAMETOOLONG;
1070 		return (1);
1071 	}
1072 	sp->fts_path = reallocf(sp->fts_path, sp->fts_pathlen);
1073 	return (sp->fts_path == NULL);
1074 }
1075 
1076 /*
1077  * When the path is realloc'd, have to fix all of the pointers in structures
1078  * already returned.
1079  */
1080 static void
1081 fts_padjust(sp, head)
1082 	FTS *sp;
1083 	FTSENT *head;
1084 {
1085 	FTSENT *p;
1086 	char *addr = sp->fts_path;
1087 
1088 #define	ADJUST(p) do {							\
1089 	if ((p)->fts_accpath != (p)->fts_name) {			\
1090 		(p)->fts_accpath =					\
1091 		    (char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
1092 	}								\
1093 	(p)->fts_path = addr;						\
1094 } while (0)
1095 	/* Adjust the current set of children. */
1096 	for (p = sp->fts_child; p; p = p->fts_link)
1097 		ADJUST(p);
1098 
1099 	/* Adjust the rest of the tree, including the current level. */
1100 	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1101 		ADJUST(p);
1102 		p = p->fts_link ? p->fts_link : p->fts_parent;
1103 	}
1104 }
1105 
1106 static size_t
1107 fts_maxarglen(argv)
1108 	char * const *argv;
1109 {
1110 	size_t len, max;
1111 
1112 	for (max = 0; *argv; ++argv)
1113 		if ((len = strlen(*argv)) > max)
1114 			max = len;
1115 	return (max + 1);
1116 }
1117 
1118 /*
1119  * Change to dir specified by fd or p->fts_accpath without getting
1120  * tricked by someone changing the world out from underneath us.
1121  * Assumes p->fts_dev and p->fts_ino are filled in.
1122  */
1123 static int
1124 fts_safe_changedir(sp, p, fd, path)
1125 	FTS *sp;
1126 	FTSENT *p;
1127 	int fd;
1128 	char *path;
1129 {
1130 	int ret, oerrno, newfd;
1131 	struct stat sb;
1132 
1133 	newfd = fd;
1134 	if (ISSET(FTS_NOCHDIR))
1135 		return (0);
1136 	if (fd < 0 && (newfd = _open(path, O_RDONLY, 0)) < 0)
1137 		return (-1);
1138 	if (_fstat(newfd, &sb)) {
1139 		ret = -1;
1140 		goto bail;
1141 	}
1142 	if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) {
1143 		errno = ENOENT;		/* disinformation */
1144 		ret = -1;
1145 		goto bail;
1146 	}
1147 	ret = fchdir(newfd);
1148 bail:
1149 	oerrno = errno;
1150 	if (fd < 0)
1151 		(void)_close(newfd);
1152 	errno = oerrno;
1153 	return (ret);
1154 }
1155