xref: /dragonfly/lib/libc/gen/fts.c (revision 6b5c5d0d)
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  * $FreeBSD: src/lib/libc/gen/fts.c,v 1.14.2.4 2001/06/01 22:00:34 kris Exp $
36  * $DragonFly: src/lib/libc/gen/fts.c,v 1.7 2005/11/13 00:07:42 swildner Exp $
37  *
38  * @(#)fts.c	8.6 (Berkeley) 8/14/94
39  * $FreeBSD: src/lib/libc/gen/fts.c,v 1.14.2.4 2001/06/01 22:00:34 kris Exp $
40  */
41 
42 #include "namespace.h"
43 #include <sys/param.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 *, const char *, int);
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 *, int);
63 static u_short	 fts_stat(FTS *, FTSENT *, int);
64 static int	 fts_safe_changedir(FTS *, FTSENT *, int, const char *);
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 FTS *
80 fts_open(char * const *argv, int options,
81 	 int (*compar)(const FTSENT **, const FTSENT **))
82 {
83 	FTS *sp;
84 	FTSENT *p, *root;
85 	FTSENT *parent, *tmp;
86 	size_t len, nitems;
87 
88 	/* Options check. */
89 	if (options & ~FTS_OPTIONMASK) {
90 		errno = EINVAL;
91 		return (NULL);
92 	}
93 
94 	/* Allocate/initialize the stream */
95 	if ((sp = malloc((u_int)sizeof(FTS))) == NULL)
96 		return (NULL);
97 	memset(sp, 0, sizeof(FTS));
98 	sp->fts_compar = compar;
99 	sp->fts_options = options;
100 
101 	/* Shush, GCC. */
102 	tmp = NULL;
103 
104 	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
105 	if (ISSET(FTS_LOGICAL))
106 		SET(FTS_NOCHDIR);
107 
108 	/*
109 	 * Start out with 1K of path space, and enough, in any case,
110 	 * to hold the user's paths.
111 	 */
112 	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
113 		goto mem1;
114 
115 	/* Allocate/initialize root's parent. */
116 	if ((parent = fts_alloc(sp, "", 0)) == NULL)
117 		goto mem2;
118 	parent->fts_level = FTS_ROOTPARENTLEVEL;
119 
120 	/* Allocate/initialize root(s). */
121 	for (root = NULL, nitems = 0; *argv != NULL; ++argv, ++nitems) {
122 		/* Don't allow zero-length paths. */
123 		if ((len = strlen(*argv)) == 0) {
124 			errno = ENOENT;
125 			goto mem3;
126 		}
127 
128 		p = fts_alloc(sp, *argv, len);
129 		p->fts_level = FTS_ROOTLEVEL;
130 		p->fts_parent = parent;
131 		p->fts_accpath = p->fts_name;
132 		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW));
133 
134 		/* Command-line "." and ".." are real directories. */
135 		if (p->fts_info == FTS_DOT)
136 			p->fts_info = FTS_D;
137 
138 		/*
139 		 * If comparison routine supplied, traverse in sorted
140 		 * order; otherwise traverse in the order specified.
141 		 */
142 		if (compar) {
143 			p->fts_link = root;
144 			root = p;
145 		} else {
146 			p->fts_link = NULL;
147 			if (root == NULL)
148 				tmp = root = p;
149 			else {
150 				tmp->fts_link = p;
151 				tmp = p;
152 			}
153 		}
154 	}
155 	if (compar && nitems > 1)
156 		root = fts_sort(sp, root, nitems);
157 
158 	/*
159 	 * Allocate a dummy pointer and make fts_read think that we've just
160 	 * finished the node before the root(s); set p->fts_info to FTS_INIT
161 	 * so that everything about the "current" node is ignored.
162 	 */
163 	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
164 		goto mem3;
165 	sp->fts_cur->fts_link = root;
166 	sp->fts_cur->fts_info = FTS_INIT;
167 
168 	/*
169 	 * If using chdir(2), grab a file descriptor pointing to dot to ensure
170 	 * that we can get back here; this could be avoided for some paths,
171 	 * but almost certainly not worth the effort.  Slashes, symbolic links,
172 	 * and ".." are all fairly nasty problems.  Note, if we can't get the
173 	 * descriptor we run anyway, just more slowly.
174 	 */
175 	if (!ISSET(FTS_NOCHDIR) && (sp->fts_rfd = _open(".", O_RDONLY, 0)) < 0)
176 		SET(FTS_NOCHDIR);
177 
178 	return (sp);
179 
180 mem3:	fts_lfree(root);
181 	free(parent);
182 mem2:	free(sp->fts_path);
183 mem1:	free(sp);
184 	return (NULL);
185 }
186 
187 static void
188 fts_load(FTS *sp, FTSENT *p)
189 {
190 	size_t len;
191 	char *cp;
192 
193 	/*
194 	 * Load the stream structure for the next traversal.  Since we don't
195 	 * actually enter the directory until after the preorder visit, set
196 	 * the fts_accpath field specially so the chdir gets done to the right
197 	 * place and the user can access the first node.  From fts_open it's
198 	 * known that the path will fit.
199 	 */
200 	len = p->fts_pathlen = p->fts_namelen;
201 	memmove(sp->fts_path, p->fts_name, len + 1);
202 	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
203 		len = strlen(++cp);
204 		memmove(p->fts_name, cp, len + 1);
205 		p->fts_namelen = len;
206 	}
207 	p->fts_accpath = p->fts_path = sp->fts_path;
208 	sp->fts_dev = p->fts_dev;
209 }
210 
211 int
212 fts_close(FTS *sp)
213 {
214 	FTSENT *freep, *p;
215 	int saved_errno;
216 
217 	/*
218 	 * This still works if we haven't read anything -- the dummy structure
219 	 * points to the root list, so we step through to the end of the root
220 	 * list which has a valid parent pointer.
221 	 */
222 	if (sp->fts_cur) {
223 		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
224 			freep = p;
225 			p = p->fts_link != NULL ? p->fts_link : p->fts_parent;
226 			free(freep);
227 		}
228 		free(p);
229 	}
230 
231 	/* Free up child linked list, sort array, path buffer. */
232 	if (sp->fts_child)
233 		fts_lfree(sp->fts_child);
234 	if (sp->fts_array)
235 		free(sp->fts_array);
236 	free(sp->fts_path);
237 
238 	/* Return to original directory, save errno if necessary. */
239 	if (!ISSET(FTS_NOCHDIR)) {
240 		saved_errno = fchdir(sp->fts_rfd) ? errno : 0;
241 		_close(sp->fts_rfd);
242 
243 		/* Set errno and return. */
244 		if (saved_errno != 0) {
245 			/* Free up the stream pointer. */
246 			free(sp);
247 			errno = saved_errno;
248 			return (-1);
249 		}
250 	}
251 
252 	/* Free up the stream pointer. */
253 	free(sp);
254 	return (0);
255 }
256 
257 /*
258  * Special case of "/" at the end of the path so that slashes aren't
259  * appended which would cause paths to be written as "....//foo".
260  */
261 #define	NAPPEND(p)							\
262 	(p->fts_path[p->fts_pathlen - 1] == '/'				\
263 	    ? p->fts_pathlen - 1 : p->fts_pathlen)
264 
265 FTSENT *
266 fts_read(FTS *sp)
267 {
268 	FTSENT *p, *tmp;
269 	int instr;
270 	char *t;
271 	int saved_errno;
272 
273 	/* If finished or unrecoverable error, return NULL. */
274 	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
275 		return (NULL);
276 
277 	/* Set current node pointer. */
278 	p = sp->fts_cur;
279 
280 	/* Save and zero out user instructions. */
281 	instr = p->fts_instr;
282 	p->fts_instr = FTS_NOINSTR;
283 
284 	/* Any type of file may be re-visited; re-stat and re-turn. */
285 	if (instr == FTS_AGAIN) {
286 		p->fts_info = fts_stat(sp, p, 0);
287 		return (p);
288 	}
289 
290 	/*
291 	 * Following a symlink -- SLNONE test allows application to see
292 	 * SLNONE and recover.  If indirecting through a symlink, have
293 	 * keep a pointer to current location.  If unable to get that
294 	 * pointer, follow fails.
295 	 */
296 	if (instr == FTS_FOLLOW &&
297 	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
298 		p->fts_info = fts_stat(sp, p, 1);
299 		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
300 			if ((p->fts_symfd = _open(".", O_RDONLY, 0)) < 0) {
301 				p->fts_errno = errno;
302 				p->fts_info = FTS_ERR;
303 			} else
304 				p->fts_flags |= FTS_SYMFOLLOW;
305 		}
306 		return (p);
307 	}
308 
309 	/* Directory in pre-order. */
310 	if (p->fts_info == FTS_D) {
311 		/* If skipped or crossed mount point, do post-order visit. */
312 		if (instr == FTS_SKIP ||
313 		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
314 			if (p->fts_flags & FTS_SYMFOLLOW)
315 				_close(p->fts_symfd);
316 			if (sp->fts_child) {
317 				fts_lfree(sp->fts_child);
318 				sp->fts_child = NULL;
319 			}
320 			p->fts_info = FTS_DP;
321 			return (p);
322 		}
323 
324 		/* Rebuild if only read the names and now traversing. */
325 		if (sp->fts_child != NULL && ISSET(FTS_NAMEONLY)) {
326 			CLR(FTS_NAMEONLY);
327 			fts_lfree(sp->fts_child);
328 			sp->fts_child = NULL;
329 		}
330 
331 		/*
332 		 * Cd to the subdirectory.
333 		 *
334 		 * If have already read and now fail to chdir, whack the list
335 		 * to make the names come out right, and set the parent errno
336 		 * so the application will eventually get an error condition.
337 		 * Set the FTS_DONTCHDIR flag so that when we logically change
338 		 * directories back to the parent we don't do a chdir.
339 		 *
340 		 * If haven't read do so.  If the read fails, fts_build sets
341 		 * FTS_STOP or the fts_info field of the node.
342 		 */
343 		if (sp->fts_child != NULL) {
344 			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
345 				p->fts_errno = errno;
346 				p->fts_flags |= FTS_DONTCHDIR;
347 				for (p = sp->fts_child; p != NULL;
348 				    p = p->fts_link)
349 					p->fts_accpath =
350 					    p->fts_parent->fts_accpath;
351 			}
352 		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
353 			if (ISSET(FTS_STOP))
354 				return (NULL);
355 			return (p);
356 		}
357 		p = sp->fts_child;
358 		sp->fts_child = NULL;
359 		goto name;
360 	}
361 
362 	/* Move to the next node on this level. */
363 next:	tmp = p;
364 	if ((p = p->fts_link) != NULL) {
365 		free(tmp);
366 
367 		/*
368 		 * If reached the top, return to the original directory (or
369 		 * the root of the tree), and load the paths for the next root.
370 		 */
371 		if (p->fts_level == FTS_ROOTLEVEL) {
372 			if (FCHDIR(sp, sp->fts_rfd)) {
373 				SET(FTS_STOP);
374 				return (NULL);
375 			}
376 			fts_load(sp, p);
377 			return (sp->fts_cur = p);
378 		}
379 
380 		/*
381 		 * User may have called fts_set on the node.  If skipped,
382 		 * ignore.  If followed, get a file descriptor so we can
383 		 * get back if necessary.
384 		 */
385 		if (p->fts_instr == FTS_SKIP)
386 			goto next;
387 		if (p->fts_instr == FTS_FOLLOW) {
388 			p->fts_info = fts_stat(sp, p, 1);
389 			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
390 				if ((p->fts_symfd =
391 				    _open(".", O_RDONLY, 0)) < 0) {
392 					p->fts_errno = errno;
393 					p->fts_info = FTS_ERR;
394 				} else
395 					p->fts_flags |= FTS_SYMFOLLOW;
396 			}
397 			p->fts_instr = FTS_NOINSTR;
398 		}
399 
400 name:		t = sp->fts_path + NAPPEND(p->fts_parent);
401 		*t++ = '/';
402 		memmove(t, p->fts_name, p->fts_namelen + 1);
403 		return (sp->fts_cur = p);
404 	}
405 
406 	/* Move up to the parent node. */
407 	p = tmp->fts_parent;
408 	free(tmp);
409 
410 	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
411 		/*
412 		 * Done; free everything up and set errno to 0 so the user
413 		 * can distinguish between error and EOF.
414 		 */
415 		free(p);
416 		errno = 0;
417 		return (sp->fts_cur = NULL);
418 	}
419 
420 	/* NUL terminate the pathname. */
421 	sp->fts_path[p->fts_pathlen] = '\0';
422 
423 	/*
424 	 * Return to the parent directory.  If at a root node or came through
425 	 * a symlink, go back through the file descriptor.  Otherwise, cd up
426 	 * one directory.
427 	 */
428 	if (p->fts_level == FTS_ROOTLEVEL) {
429 		if (FCHDIR(sp, sp->fts_rfd)) {
430 			SET(FTS_STOP);
431 			return (NULL);
432 		}
433 	} else if (p->fts_flags & FTS_SYMFOLLOW) {
434 		if (FCHDIR(sp, p->fts_symfd)) {
435 			saved_errno = errno;
436 			_close(p->fts_symfd);
437 			errno = saved_errno;
438 			SET(FTS_STOP);
439 			return (NULL);
440 		}
441 		_close(p->fts_symfd);
442 	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
443 		   fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
444 		SET(FTS_STOP);
445 		return (NULL);
446 	}
447 	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
448 	return (sp->fts_cur = p);
449 }
450 
451 /*
452  * Fts_set takes the stream as an argument although it's not used in this
453  * implementation; it would be necessary if anyone wanted to add global
454  * semantics to fts using fts_set.  An error return is allowed for similar
455  * reasons.
456  */
457 /* ARGSUSED */
458 int
459 fts_set(FTS *sp __unused, FTSENT *p, int instr)
460 {
461 	if (instr != 0 && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
462 	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
463 		errno = EINVAL;
464 		return (1);
465 	}
466 	p->fts_instr = instr;
467 	return (0);
468 }
469 
470 FTSENT *
471 fts_children(FTS *sp, int instr)
472 {
473 	FTSENT *p;
474 	int fd;
475 
476 	if (instr != 0 && instr != FTS_NAMEONLY) {
477 		errno = EINVAL;
478 		return (NULL);
479 	}
480 
481 	/* Set current node pointer. */
482 	p = sp->fts_cur;
483 
484 	/*
485 	 * Errno set to 0 so user can distinguish empty directory from
486 	 * an error.
487 	 */
488 	errno = 0;
489 
490 	/* Fatal errors stop here. */
491 	if (ISSET(FTS_STOP))
492 		return (NULL);
493 
494 	/* Return logical hierarchy of user's arguments. */
495 	if (p->fts_info == FTS_INIT)
496 		return (p->fts_link);
497 
498 	/*
499 	 * If not a directory being visited in pre-order, stop here.  Could
500 	 * allow FTS_DNR, assuming the user has fixed the problem, but the
501 	 * same effect is available with FTS_AGAIN.
502 	 */
503 	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
504 		return (NULL);
505 
506 	/* Free up any previous child list. */
507 	if (sp->fts_child != NULL)
508 		fts_lfree(sp->fts_child);
509 
510 	if (instr == FTS_NAMEONLY) {
511 		SET(FTS_NAMEONLY);
512 		instr = BNAMES;
513 	} else
514 		instr = BCHILD;
515 
516 	/*
517 	 * If using chdir on a relative path and called BEFORE fts_read does
518 	 * its chdir to the root of a traversal, we can lose -- we need to
519 	 * chdir into the subdirectory, and we don't know where the current
520 	 * directory is, so we can't get back so that the upcoming chdir by
521 	 * fts_read will work.
522 	 */
523 	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
524 	    ISSET(FTS_NOCHDIR))
525 		return (sp->fts_child = fts_build(sp, instr));
526 
527 	if ((fd = _open(".", O_RDONLY, 0)) < 0)
528 		return (NULL);
529 	sp->fts_child = fts_build(sp, instr);
530 	if (fchdir(fd))
531 		return (NULL);
532 	_close(fd);
533 	return (sp->fts_child);
534 }
535 
536 /*
537  * This is the tricky part -- do not casually change *anything* in here.  The
538  * idea is to build the linked list of entries that are used by fts_children
539  * and fts_read.  There are lots of special cases.
540  *
541  * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
542  * set and it's a physical walk (so that symbolic links can't be directories),
543  * we can do things quickly.  First, if it's a 4.4BSD file system, the type
544  * of the file is in the directory entry.  Otherwise, we assume that the number
545  * of subdirectories in a node is equal to the number of links to the parent.
546  * The former skips all stat calls.  The latter skips stat calls in any leaf
547  * directories and for any files after the subdirectories in the directory have
548  * been found, cutting the stat calls by about 2/3.
549  */
550 static FTSENT *
551 fts_build(FTS *sp, int type)
552 {
553 	struct dirent *dp;
554 	FTSENT *p, *head;
555 	int nitems;
556 	FTSENT *cur, *tail;
557 	DIR *dirp;
558 	void *oldaddr;
559 	int cderrno, descend, len, level, maxlen, nlinks, oflag, saved_errno,
560 	    nostat, doadjust;
561 	char *cp;
562 
563 	/* Set current node pointer. */
564 	cur = sp->fts_cur;
565 
566 	/*
567 	 * Open the directory for reading.  If this fails, we're done.
568 	 * If being called from fts_read, set the fts_info field.
569 	 */
570 #ifdef FTS_WHITEOUT
571 	if (ISSET(FTS_WHITEOUT))
572 		oflag = DTF_NODUP|DTF_REWIND;
573 	else
574 		oflag = DTF_HIDEW|DTF_NODUP|DTF_REWIND;
575 #else
576 #define __opendir2(path, flag) opendir(path)
577 #endif
578 	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
579 		if (type == BREAD) {
580 			cur->fts_info = FTS_DNR;
581 			cur->fts_errno = errno;
582 		}
583 		return (NULL);
584 	}
585 
586 	/*
587 	 * Nlinks is the number of possible entries of type directory in the
588 	 * directory if we're cheating on stat calls, 0 if we're not doing
589 	 * any stat calls at all, -1 if we're doing stats on everything.
590 	 */
591 	if (type == BNAMES) {
592 		nlinks = 0;
593 		/* Be quiet about nostat, GCC. */
594 		nostat = 0;
595 	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
596 		nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
597 		nostat = 1;
598 	} else {
599 		nlinks = -1;
600 		nostat = 0;
601 	}
602 
603 #ifdef notdef
604 	printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
605 	printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
606 	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
607 #endif
608 	/*
609 	 * If we're going to need to stat anything or we want to descend
610 	 * and stay in the directory, chdir.  If this fails we keep going,
611 	 * but set a flag so we don't chdir after the post-order visit.
612 	 * We won't be able to stat anything, but we can still return the
613 	 * names themselves.  Note, that since fts_read won't be able to
614 	 * chdir into the directory, it will have to return different path
615 	 * names than before, i.e. "a/b" instead of "b".  Since the node
616 	 * has already been visited in pre-order, have to wait until the
617 	 * post-order visit to return the error.  There is a special case
618 	 * here, if there was nothing to stat then it's not an error to
619 	 * not be able to stat.  This is all fairly nasty.  If a program
620 	 * needed sorted entries or stat information, they had better be
621 	 * checking FTS_NS on the returned nodes.
622 	 */
623 	cderrno = 0;
624 	if (nlinks || type == BREAD) {
625 		if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) {
626 			if (nlinks && type == BREAD)
627 				cur->fts_errno = errno;
628 			cur->fts_flags |= FTS_DONTCHDIR;
629 			descend = 0;
630 			cderrno = errno;
631 			closedir(dirp);
632 			dirp = NULL;
633 		} else
634 			descend = 1;
635 	} else
636 		descend = 0;
637 
638 	/*
639 	 * Figure out the max file name length that can be stored in the
640 	 * current path -- the inner loop allocates more path as necessary.
641 	 * We really wouldn't have to do the maxlen calculations here, we
642 	 * could do them in fts_read before returning the path, but it's a
643 	 * lot easier here since the length is part of the dirent structure.
644 	 *
645 	 * If not changing directories set a pointer so that can just append
646 	 * each new name into the path.
647 	 */
648 	len = NAPPEND(cur);
649 	if (ISSET(FTS_NOCHDIR)) {
650 		cp = sp->fts_path + len;
651 		*cp++ = '/';
652 	} else {
653 		/* GCC, you're too verbose. */
654 		cp = NULL;
655 	}
656 	len++;
657 	maxlen = sp->fts_pathlen - len;
658 
659 	level = cur->fts_level + 1;
660 
661 	/* Read the directory, attaching each entry to the `link' pointer. */
662 	doadjust = 0;
663 	for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) {
664 		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
665 			continue;
666 
667 		if ((p = fts_alloc(sp, dp->d_name, (int)dp->d_namlen)) == NULL)
668 			goto mem1;
669 		if (dp->d_namlen >= maxlen) {	/* include space for NUL */
670 			oldaddr = sp->fts_path;
671 			if (fts_palloc(sp, dp->d_namlen + len + 1)) {
672 				/*
673 				 * No more memory for path or structures.  Save
674 				 * errno, free up the current structure and the
675 				 * structures already allocated.
676 				 */
677 mem1:				saved_errno = errno;
678 				if (p)
679 					free(p);
680 				fts_lfree(head);
681 				closedir(dirp);
682 				cur->fts_info = FTS_ERR;
683 				SET(FTS_STOP);
684 				errno = saved_errno;
685 				return (NULL);
686 			}
687 			/* Did realloc() change the pointer? */
688 			if (oldaddr != sp->fts_path) {
689 				doadjust = 1;
690 				if (ISSET(FTS_NOCHDIR))
691 					cp = sp->fts_path + len;
692 			}
693 			maxlen = sp->fts_pathlen - len;
694 		}
695 
696 		if (len + dp->d_namlen >= USHRT_MAX) {
697 			/*
698 			 * In an FTSENT, fts_pathlen is a u_short so it is
699 			 * possible to wraparound here.  If we do, free up
700 			 * the current structure and the structures already
701 			 * allocated, then error out with ENAMETOOLONG.
702 			 */
703 			free(p);
704 			fts_lfree(head);
705 			closedir(dirp);
706 			cur->fts_info = FTS_ERR;
707 			SET(FTS_STOP);
708 			errno = ENAMETOOLONG;
709 			return (NULL);
710 		}
711 		p->fts_level = level;
712 		p->fts_parent = sp->fts_cur;
713 		p->fts_pathlen = len + dp->d_namlen;
714 
715 #ifdef FTS_WHITEOUT
716 		if (dp->d_type == DT_WHT)
717 			p->fts_flags |= FTS_ISW;
718 #endif
719 
720 		if (cderrno) {
721 			if (nlinks) {
722 				p->fts_info = FTS_NS;
723 				p->fts_errno = cderrno;
724 			} else
725 				p->fts_info = FTS_NSOK;
726 			p->fts_accpath = cur->fts_accpath;
727 		} else if (nlinks == 0
728 #ifdef DT_DIR
729 		    || (nostat &&
730 		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
731 #endif
732 		    ) {
733 			p->fts_accpath =
734 			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
735 			p->fts_info = FTS_NSOK;
736 		} else {
737 			/* Build a file name for fts_stat to stat. */
738 			if (ISSET(FTS_NOCHDIR)) {
739 				p->fts_accpath = p->fts_path;
740 				memmove(cp, p->fts_name, p->fts_namelen + 1);
741 			} else
742 				p->fts_accpath = p->fts_name;
743 			/* Stat it. */
744 			p->fts_info = fts_stat(sp, p, 0);
745 
746 			/* Decrement link count if applicable. */
747 			if (nlinks > 0 && (p->fts_info == FTS_D ||
748 			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
749 				--nlinks;
750 		}
751 
752 		/* We walk in directory order so "ls -f" doesn't get upset. */
753 		p->fts_link = NULL;
754 		if (head == NULL)
755 			head = tail = p;
756 		else {
757 			tail->fts_link = p;
758 			tail = p;
759 		}
760 		++nitems;
761 	}
762 	if (dirp)
763 		closedir(dirp);
764 
765 	/*
766 	 * If realloc() changed the address of the path, adjust the
767 	 * addresses for the rest of the tree and the dir list.
768 	 */
769 	if (doadjust)
770 		fts_padjust(sp, head);
771 
772 	/*
773 	 * If not changing directories, reset the path back to original
774 	 * state.
775 	 */
776 	if (ISSET(FTS_NOCHDIR)) {
777 		if (len == sp->fts_pathlen || nitems == 0)
778 			--cp;
779 		*cp = '\0';
780 	}
781 
782 	/*
783 	 * If descended after called from fts_children or after called from
784 	 * fts_read and nothing found, get back.  At the root level we use
785 	 * the saved fd; if one of fts_open()'s arguments is a relative path
786 	 * to an empty directory, we wind up here with no other way back.  If
787 	 * can't get back, we're done.
788 	 */
789 	if (descend && (type == BCHILD || !nitems) &&
790 	    (cur->fts_level == FTS_ROOTLEVEL ?
791 	    FCHDIR(sp, sp->fts_rfd) :
792 	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
793 		cur->fts_info = FTS_ERR;
794 		SET(FTS_STOP);
795 		return (NULL);
796 	}
797 
798 	/* If didn't find anything, return NULL. */
799 	if (!nitems) {
800 		if (type == BREAD)
801 			cur->fts_info = FTS_DP;
802 		return (NULL);
803 	}
804 
805 	/* Sort the entries. */
806 	if (sp->fts_compar && nitems > 1)
807 		head = fts_sort(sp, head, nitems);
808 	return (head);
809 }
810 
811 static u_short
812 fts_stat(FTS *sp, FTSENT *p, int follow)
813 {
814 	FTSENT *t;
815 	dev_t dev;
816 	ino_t ino;
817 	struct stat *sbp, sb;
818 	int saved_errno;
819 
820 	/* If user needs stat info, stat buffer already allocated. */
821 	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
822 
823 #ifdef FTS_WHITEOUT
824 	/* check for whiteout */
825 	if (p->fts_flags & FTS_ISW) {
826 		if (sbp != &sb) {
827 			memset(sbp, '\0', sizeof (*sbp));
828 			sbp->st_mode = S_IFWHT;
829 		}
830 		return (FTS_W);
831 	}
832 #endif
833 
834 	/*
835 	 * If doing a logical walk, or application requested FTS_FOLLOW, do
836 	 * a stat(2).  If that fails, check for a non-existent symlink.  If
837 	 * fail, set the errno from the stat call.
838 	 */
839 	if (ISSET(FTS_LOGICAL) || follow) {
840 		if (stat(p->fts_accpath, sbp)) {
841 			saved_errno = errno;
842 			if (!lstat(p->fts_accpath, sbp)) {
843 				errno = 0;
844 				return (FTS_SLNONE);
845 			}
846 			p->fts_errno = saved_errno;
847 			goto err;
848 		}
849 	} else if (lstat(p->fts_accpath, sbp)) {
850 		p->fts_errno = errno;
851 err:		memset(sbp, 0, sizeof(struct stat));
852 		return (FTS_NS);
853 	}
854 
855 	if (S_ISDIR(sbp->st_mode)) {
856 		/*
857 		 * Set the device/inode.  Used to find cycles and check for
858 		 * crossing mount points.  Also remember the link count, used
859 		 * in fts_build to limit the number of stat calls.  It is
860 		 * understood that these fields are only referenced if fts_info
861 		 * is set to FTS_D.
862 		 */
863 		dev = p->fts_dev = sbp->st_dev;
864 		ino = p->fts_ino = sbp->st_ino;
865 		p->fts_nlink = sbp->st_nlink;
866 
867 		if (ISDOT(p->fts_name))
868 			return (FTS_DOT);
869 
870 		/*
871 		 * Cycle detection is done by brute force when the directory
872 		 * is first encountered.  If the tree gets deep enough or the
873 		 * number of symbolic links to directories is high enough,
874 		 * something faster might be worthwhile.
875 		 */
876 		for (t = p->fts_parent;
877 		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
878 			if (ino == t->fts_ino && dev == t->fts_dev) {
879 				p->fts_cycle = t;
880 				return (FTS_DC);
881 			}
882 		return (FTS_D);
883 	}
884 	if (S_ISLNK(sbp->st_mode))
885 		return (FTS_SL);
886 	if (S_ISREG(sbp->st_mode))
887 		return (FTS_F);
888 	return (FTS_DEFAULT);
889 }
890 
891 static FTSENT *
892 fts_sort(FTS *sp, FTSENT *head, int nitems)
893 {
894 	FTSENT **ap, *p;
895 
896 	/*
897 	 * Construct an array of pointers to the structures and call qsort(3).
898 	 * Reassemble the array in the order returned by qsort.  If unable to
899 	 * sort for memory reasons, return the directory entries in their
900 	 * current order.  Allocate enough space for the current needs plus
901 	 * 40 so don't realloc one entry at a time.
902 	 */
903 	if (nitems > sp->fts_nitems) {
904 		sp->fts_nitems = nitems + 40;
905 		if ((sp->fts_array = reallocf(sp->fts_array,
906 		    sp->fts_nitems * sizeof(FTSENT *))) == NULL) {
907 			sp->fts_nitems = 0;
908 			return (head);
909 		}
910 	}
911 	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
912 		*ap++ = p;
913 	qsort(sp->fts_array, nitems, sizeof(FTSENT *),
914 	      (int (*)(const void *, const void *))sp->fts_compar);
915 	for (head = *(ap = sp->fts_array); --nitems; ++ap)
916 		ap[0]->fts_link = ap[1];
917 	ap[0]->fts_link = NULL;
918 	return (head);
919 }
920 
921 static FTSENT *
922 fts_alloc(FTS *sp, const char *name, int namelen)
923 {
924 	FTSENT *p;
925 	size_t len;
926 
927 	/*
928 	 * The file name is a variable length array and no stat structure is
929 	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
930 	 * structure, the file name and the stat structure in one chunk, but
931 	 * be careful that the stat structure is reasonably aligned.  Since the
932 	 * fts_name field is declared to be of size 1, the fts_name pointer is
933 	 * namelen + 2 before the first possible address of the stat structure.
934 	 */
935 	len = sizeof(FTSENT) + namelen;
936 	if (!ISSET(FTS_NOSTAT))
937 		len += sizeof(struct stat) + ALIGNBYTES;
938 	if ((p = malloc(len)) == NULL)
939 		return (NULL);
940 
941 	/* Copy the name and guarantee NUL termination. */
942 	memmove(p->fts_name, name, namelen);
943 	p->fts_name[namelen] = '\0';
944 
945 	if (!ISSET(FTS_NOSTAT))
946 		p->fts_statp = (struct stat *)ALIGN(p->fts_name + namelen + 2);
947 	p->fts_namelen = namelen;
948 	p->fts_path = sp->fts_path;
949 	p->fts_errno = 0;
950 	p->fts_flags = 0;
951 	p->fts_instr = FTS_NOINSTR;
952 	p->fts_number = 0;
953 	p->fts_pointer = NULL;
954 	return (p);
955 }
956 
957 static void
958 fts_lfree(FTSENT *head)
959 {
960 	FTSENT *p;
961 
962 	/* Free a linked list of structures. */
963 	while ((p = head)) {
964 		head = head->fts_link;
965 		free(p);
966 	}
967 }
968 
969 /*
970  * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
971  * Most systems will allow creation of paths much longer than MAXPATHLEN, even
972  * though the kernel won't resolve them.  Add the size (not just what's needed)
973  * plus 256 bytes so don't realloc the path 2 bytes at a time.
974  */
975 static int
976 fts_palloc(FTS *sp, size_t more)
977 {
978 
979 	sp->fts_pathlen += more + 256;
980 	/*
981 	 * Check for possible wraparound.  In an FTS, fts_pathlen is
982 	 * a signed int but in an FTSENT it is an unsigned short.
983 	 * We limit fts_pathlen to USHRT_MAX to be safe in both cases.
984 	 */
985 	if (sp->fts_pathlen < 0 || sp->fts_pathlen >= USHRT_MAX) {
986 		if (sp->fts_path)
987 			free(sp->fts_path);
988 		sp->fts_path = NULL;
989 		errno = ENAMETOOLONG;
990 		return (1);
991 	}
992 	sp->fts_path = reallocf(sp->fts_path, sp->fts_pathlen);
993 	return (sp->fts_path == NULL);
994 }
995 
996 /*
997  * When the path is realloc'd, have to fix all of the pointers in structures
998  * already returned.
999  */
1000 static void
1001 fts_padjust(FTS *sp, FTSENT *head)
1002 {
1003 	FTSENT *p;
1004 	char *addr = sp->fts_path;
1005 
1006 #define	ADJUST(p) do {							\
1007 	if ((p)->fts_accpath != (p)->fts_name) {			\
1008 		(p)->fts_accpath =					\
1009 		    (char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
1010 	}								\
1011 	(p)->fts_path = addr;						\
1012 } while (0)
1013 	/* Adjust the current set of children. */
1014 	for (p = sp->fts_child; p; p = p->fts_link)
1015 		ADJUST(p);
1016 
1017 	/* Adjust the rest of the tree, including the current level. */
1018 	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1019 		ADJUST(p);
1020 		p = p->fts_link ? p->fts_link : p->fts_parent;
1021 	}
1022 }
1023 
1024 static size_t
1025 fts_maxarglen(char * const *argv)
1026 {
1027 	size_t len, max;
1028 
1029 	for (max = 0; *argv; ++argv)
1030 		if ((len = strlen(*argv)) > max)
1031 			max = len;
1032 	return (max + 1);
1033 }
1034 
1035 /*
1036  * Change to dir specified by fd or p->fts_accpath without getting
1037  * tricked by someone changing the world out from underneath us.
1038  * Assumes p->fts_dev and p->fts_ino are filled in.
1039  */
1040 static int
1041 fts_safe_changedir(FTS *sp, FTSENT *p, int fd, const char *path)
1042 {
1043 	int ret, oerrno, newfd;
1044 	struct stat sb;
1045 
1046 	newfd = fd;
1047 	if (ISSET(FTS_NOCHDIR))
1048 		return (0);
1049 	if (fd < 0 && (newfd = _open(path, O_RDONLY, 0)) < 0)
1050 		return (-1);
1051 	if (_fstat(newfd, &sb)) {
1052 		ret = -1;
1053 		goto bail;
1054 	}
1055 	if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) {
1056 		errno = ENOENT;		/* disinformation */
1057 		ret = -1;
1058 		goto bail;
1059 	}
1060 	ret = fchdir(newfd);
1061 bail:
1062 	oerrno = errno;
1063 	if (fd < 0)
1064 		_close(newfd);
1065 	errno = oerrno;
1066 	return (ret);
1067 }
1068