xref: /openbsd-src/lib/libc/gen/fts.c (revision 850e275390052b330d93020bf619a739a3c277ac)
1 /*	$OpenBSD: fts.c,v 1.40 2007/11/06 10:14:53 chl Exp $	*/
2 
3 /*-
4  * Copyright (c) 1990, 1993, 1994
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 #include <sys/param.h>
33 #include <sys/stat.h>
34 
35 #include <dirent.h>
36 #include <errno.h>
37 #include <fcntl.h>
38 #include <fts.h>
39 #include <stdlib.h>
40 #include <string.h>
41 #include <unistd.h>
42 
43 static FTSENT	*fts_alloc(FTS *, char *, size_t);
44 static FTSENT	*fts_build(FTS *, int);
45 static void	 fts_lfree(FTSENT *);
46 static void	 fts_load(FTS *, FTSENT *);
47 static size_t	 fts_maxarglen(char * const *);
48 static void	 fts_padjust(FTS *, FTSENT *);
49 static int	 fts_palloc(FTS *, size_t);
50 static FTSENT	*fts_sort(FTS *, FTSENT *, int);
51 static u_short	 fts_stat(FTS *, FTSENT *, int);
52 static int	 fts_safe_changedir(FTS *, FTSENT *, int, char *);
53 
54 #define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
55 
56 #define	CLR(opt)	(sp->fts_options &= ~(opt))
57 #define	ISSET(opt)	(sp->fts_options & (opt))
58 #define	SET(opt)	(sp->fts_options |= (opt))
59 
60 #define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
61 
62 /* fts_build flags */
63 #define	BCHILD		1		/* fts_children */
64 #define	BNAMES		2		/* fts_children, names only */
65 #define	BREAD		3		/* fts_read */
66 
67 FTS *
68 fts_open(char * const *argv, int options,
69     int (*compar)(const FTSENT **, const FTSENT **))
70 {
71 	FTS *sp;
72 	FTSENT *p, *root;
73 	int nitems;
74 	FTSENT *parent, *tmp;
75 	size_t len;
76 
77 	/* Options check. */
78 	if (options & ~FTS_OPTIONMASK) {
79 		errno = EINVAL;
80 		return (NULL);
81 	}
82 
83 	/* Allocate/initialize the stream */
84 	if ((sp = calloc(1, sizeof(FTS))) == NULL)
85 		return (NULL);
86 	sp->fts_compar = compar;
87 	sp->fts_options = options;
88 
89 	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
90 	if (ISSET(FTS_LOGICAL))
91 		SET(FTS_NOCHDIR);
92 
93 	/*
94 	 * Start out with 1K of path space, and enough, in any case,
95 	 * to hold the user's paths.
96 	 */
97 	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
98 		goto mem1;
99 
100 	/* Allocate/initialize root's parent. */
101 	if ((parent = fts_alloc(sp, "", 0)) == NULL)
102 		goto mem2;
103 	parent->fts_level = FTS_ROOTPARENTLEVEL;
104 
105 	/* Allocate/initialize root(s). */
106 	for (root = NULL, nitems = 0; *argv; ++argv, ++nitems) {
107 		/* Don't allow zero-length paths. */
108 		if ((len = strlen(*argv)) == 0) {
109 			errno = ENOENT;
110 			goto mem3;
111 		}
112 
113 		if ((p = fts_alloc(sp, *argv, len)) == NULL)
114 			goto mem3;
115 		p->fts_level = FTS_ROOTLEVEL;
116 		p->fts_parent = parent;
117 		p->fts_accpath = p->fts_name;
118 		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW));
119 
120 		/* Command-line "." and ".." are real directories. */
121 		if (p->fts_info == FTS_DOT)
122 			p->fts_info = FTS_D;
123 
124 		/*
125 		 * If comparison routine supplied, traverse in sorted
126 		 * order; otherwise traverse in the order specified.
127 		 */
128 		if (compar) {
129 			p->fts_link = root;
130 			root = p;
131 		} else {
132 			p->fts_link = NULL;
133 			if (root == NULL)
134 				tmp = root = p;
135 			else {
136 				tmp->fts_link = p;
137 				tmp = p;
138 			}
139 		}
140 	}
141 	if (compar && nitems > 1)
142 		root = fts_sort(sp, root, nitems);
143 
144 	/*
145 	 * Allocate a dummy pointer and make fts_read think that we've just
146 	 * finished the node before the root(s); set p->fts_info to FTS_INIT
147 	 * so that everything about the "current" node is ignored.
148 	 */
149 	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
150 		goto mem3;
151 	sp->fts_cur->fts_link = root;
152 	sp->fts_cur->fts_info = FTS_INIT;
153 
154 	/*
155 	 * If using chdir(2), grab a file descriptor pointing to dot to ensure
156 	 * that we can get back here; this could be avoided for some paths,
157 	 * but almost certainly not worth the effort.  Slashes, symbolic links,
158 	 * and ".." are all fairly nasty problems.  Note, if we can't get the
159 	 * descriptor we run anyway, just more slowly.
160 	 */
161 	if (!ISSET(FTS_NOCHDIR) && (sp->fts_rfd = open(".", O_RDONLY, 0)) < 0)
162 		SET(FTS_NOCHDIR);
163 
164 	if (nitems == 0)
165 		free(parent);
166 
167 	return (sp);
168 
169 mem3:	fts_lfree(root);
170 	free(parent);
171 mem2:	free(sp->fts_path);
172 mem1:	free(sp);
173 	return (NULL);
174 }
175 
176 static void
177 fts_load(FTS *sp, FTSENT *p)
178 {
179 	size_t len;
180 	char *cp;
181 
182 	/*
183 	 * Load the stream structure for the next traversal.  Since we don't
184 	 * actually enter the directory until after the preorder visit, set
185 	 * the fts_accpath field specially so the chdir gets done to the right
186 	 * place and the user can access the first node.  From fts_open it's
187 	 * known that the path will fit.
188 	 */
189 	len = p->fts_pathlen = p->fts_namelen;
190 	memmove(sp->fts_path, p->fts_name, len + 1);
191 	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
192 		len = strlen(++cp);
193 		memmove(p->fts_name, cp, len + 1);
194 		p->fts_namelen = len;
195 	}
196 	p->fts_accpath = p->fts_path = sp->fts_path;
197 	sp->fts_dev = p->fts_dev;
198 }
199 
200 int
201 fts_close(FTS *sp)
202 {
203 	FTSENT *freep, *p;
204 	int rfd, error = 0;
205 
206 	/*
207 	 * This still works if we haven't read anything -- the dummy structure
208 	 * points to the root list, so we step through to the end of the root
209 	 * list which has a valid parent pointer.
210 	 */
211 	if (sp->fts_cur) {
212 		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
213 			freep = p;
214 			p = p->fts_link ? p->fts_link : p->fts_parent;
215 			free(freep);
216 		}
217 		free(p);
218 	}
219 
220 	/* Stash the original directory fd if needed. */
221 	rfd = ISSET(FTS_NOCHDIR) ? -1 : sp->fts_rfd;
222 
223 	/* Free up child linked list, sort array, path buffer, stream ptr.*/
224 	if (sp->fts_child)
225 		fts_lfree(sp->fts_child);
226 	if (sp->fts_array)
227 		free(sp->fts_array);
228 	free(sp->fts_path);
229 	free(sp);
230 
231 	/* Return to original directory, checking for error. */
232 	if (rfd != -1) {
233 		int saved_errno;
234 		error = fchdir(rfd);
235 		saved_errno = errno;
236 		(void)close(rfd);
237 		errno = saved_errno;
238 	}
239 
240 	return (error);
241 }
242 
243 /*
244  * Special case of "/" at the end of the path so that slashes aren't
245  * appended which would cause paths to be written as "....//foo".
246  */
247 #define	NAPPEND(p)							\
248 	(p->fts_path[p->fts_pathlen - 1] == '/'				\
249 	    ? p->fts_pathlen - 1 : p->fts_pathlen)
250 
251 FTSENT *
252 fts_read(FTS *sp)
253 {
254 	FTSENT *p, *tmp;
255 	int instr;
256 	char *t;
257 	int saved_errno;
258 
259 	/* If finished or unrecoverable error, return NULL. */
260 	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
261 		return (NULL);
262 
263 	/* Set current node pointer. */
264 	p = sp->fts_cur;
265 
266 	/* Save and zero out user instructions. */
267 	instr = p->fts_instr;
268 	p->fts_instr = FTS_NOINSTR;
269 
270 	/* Any type of file may be re-visited; re-stat and re-turn. */
271 	if (instr == FTS_AGAIN) {
272 		p->fts_info = fts_stat(sp, p, 0);
273 		return (p);
274 	}
275 
276 	/*
277 	 * Following a symlink -- SLNONE test allows application to see
278 	 * SLNONE and recover.  If indirecting through a symlink, have
279 	 * keep a pointer to current location.  If unable to get that
280 	 * pointer, follow fails.
281 	 */
282 	if (instr == FTS_FOLLOW &&
283 	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
284 		p->fts_info = fts_stat(sp, p, 1);
285 		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
286 			if ((p->fts_symfd = open(".", O_RDONLY, 0)) < 0) {
287 				p->fts_errno = errno;
288 				p->fts_info = FTS_ERR;
289 			} else
290 				p->fts_flags |= FTS_SYMFOLLOW;
291 		}
292 		return (p);
293 	}
294 
295 	/* Directory in pre-order. */
296 	if (p->fts_info == FTS_D) {
297 		/* If skipped or crossed mount point, do post-order visit. */
298 		if (instr == FTS_SKIP ||
299 		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
300 			if (p->fts_flags & FTS_SYMFOLLOW)
301 				(void)close(p->fts_symfd);
302 			if (sp->fts_child) {
303 				fts_lfree(sp->fts_child);
304 				sp->fts_child = NULL;
305 			}
306 			p->fts_info = FTS_DP;
307 			return (p);
308 		}
309 
310 		/* Rebuild if only read the names and now traversing. */
311 		if (sp->fts_child && ISSET(FTS_NAMEONLY)) {
312 			CLR(FTS_NAMEONLY);
313 			fts_lfree(sp->fts_child);
314 			sp->fts_child = NULL;
315 		}
316 
317 		/*
318 		 * Cd to the subdirectory.
319 		 *
320 		 * If have already read and now fail to chdir, whack the list
321 		 * to make the names come out right, and set the parent errno
322 		 * so the application will eventually get an error condition.
323 		 * Set the FTS_DONTCHDIR flag so that when we logically change
324 		 * directories back to the parent we don't do a chdir.
325 		 *
326 		 * If haven't read do so.  If the read fails, fts_build sets
327 		 * FTS_STOP or the fts_info field of the node.
328 		 */
329 		if (sp->fts_child) {
330 			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
331 				p->fts_errno = errno;
332 				p->fts_flags |= FTS_DONTCHDIR;
333 				for (p = sp->fts_child; p; p = p->fts_link)
334 					p->fts_accpath =
335 					    p->fts_parent->fts_accpath;
336 			}
337 		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
338 			if (ISSET(FTS_STOP))
339 				return (NULL);
340 			return (p);
341 		}
342 		p = sp->fts_child;
343 		sp->fts_child = NULL;
344 		goto name;
345 	}
346 
347 	/* Move to the next node on this level. */
348 next:	tmp = p;
349 	if ((p = p->fts_link)) {
350 		free(tmp);
351 
352 		/*
353 		 * If reached the top, return to the original directory (or
354 		 * the root of the tree), and load the paths for the next root.
355 		 */
356 		if (p->fts_level == FTS_ROOTLEVEL) {
357 			if (FCHDIR(sp, sp->fts_rfd)) {
358 				SET(FTS_STOP);
359 				return (NULL);
360 			}
361 			fts_load(sp, p);
362 			return (sp->fts_cur = p);
363 		}
364 
365 		/*
366 		 * User may have called fts_set on the node.  If skipped,
367 		 * ignore.  If followed, get a file descriptor so we can
368 		 * get back if necessary.
369 		 */
370 		if (p->fts_instr == FTS_SKIP)
371 			goto next;
372 		if (p->fts_instr == FTS_FOLLOW) {
373 			p->fts_info = fts_stat(sp, p, 1);
374 			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
375 				if ((p->fts_symfd =
376 				    open(".", O_RDONLY, 0)) < 0) {
377 					p->fts_errno = errno;
378 					p->fts_info = FTS_ERR;
379 				} else
380 					p->fts_flags |= FTS_SYMFOLLOW;
381 			}
382 			p->fts_instr = FTS_NOINSTR;
383 		}
384 
385 name:		t = sp->fts_path + NAPPEND(p->fts_parent);
386 		*t++ = '/';
387 		memmove(t, p->fts_name, p->fts_namelen + 1);
388 		return (sp->fts_cur = p);
389 	}
390 
391 	/* Move up to the parent node. */
392 	p = tmp->fts_parent;
393 	free(tmp);
394 
395 	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
396 		/*
397 		 * Done; free everything up and set errno to 0 so the user
398 		 * can distinguish between error and EOF.
399 		 */
400 		free(p);
401 		errno = 0;
402 		return (sp->fts_cur = NULL);
403 	}
404 
405 	/* NUL terminate the pathname. */
406 	sp->fts_path[p->fts_pathlen] = '\0';
407 
408 	/*
409 	 * Return to the parent directory.  If at a root node or came through
410 	 * a symlink, go back through the file descriptor.  Otherwise, cd up
411 	 * one directory.
412 	 */
413 	if (p->fts_level == FTS_ROOTLEVEL) {
414 		if (FCHDIR(sp, sp->fts_rfd)) {
415 			SET(FTS_STOP);
416 			sp->fts_cur = p;
417 			return (NULL);
418 		}
419 	} else if (p->fts_flags & FTS_SYMFOLLOW) {
420 		if (FCHDIR(sp, p->fts_symfd)) {
421 			saved_errno = errno;
422 			(void)close(p->fts_symfd);
423 			errno = saved_errno;
424 			SET(FTS_STOP);
425 			sp->fts_cur = p;
426 			return (NULL);
427 		}
428 		(void)close(p->fts_symfd);
429 	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
430 	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
431 		SET(FTS_STOP);
432 		sp->fts_cur = p;
433 		return (NULL);
434 	}
435 	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
436 	return (sp->fts_cur = p);
437 }
438 
439 /*
440  * Fts_set takes the stream as an argument although it's not used in this
441  * implementation; it would be necessary if anyone wanted to add global
442  * semantics to fts using fts_set.  An error return is allowed for similar
443  * reasons.
444  */
445 /* ARGSUSED */
446 int
447 fts_set(FTS *sp, FTSENT *p, int instr)
448 {
449 	if (instr && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
450 	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
451 		errno = EINVAL;
452 		return (1);
453 	}
454 	p->fts_instr = instr;
455 	return (0);
456 }
457 
458 FTSENT *
459 fts_children(FTS *sp, int instr)
460 {
461 	FTSENT *p;
462 	int fd;
463 
464 	if (instr && instr != FTS_NAMEONLY) {
465 		errno = EINVAL;
466 		return (NULL);
467 	}
468 
469 	/* Set current node pointer. */
470 	p = sp->fts_cur;
471 
472 	/*
473 	 * Errno set to 0 so user can distinguish empty directory from
474 	 * an error.
475 	 */
476 	errno = 0;
477 
478 	/* Fatal errors stop here. */
479 	if (ISSET(FTS_STOP))
480 		return (NULL);
481 
482 	/* Return logical hierarchy of user's arguments. */
483 	if (p->fts_info == FTS_INIT)
484 		return (p->fts_link);
485 
486 	/*
487 	 * If not a directory being visited in pre-order, stop here.  Could
488 	 * allow FTS_DNR, assuming the user has fixed the problem, but the
489 	 * same effect is available with FTS_AGAIN.
490 	 */
491 	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
492 		return (NULL);
493 
494 	/* Free up any previous child list. */
495 	if (sp->fts_child)
496 		fts_lfree(sp->fts_child);
497 
498 	if (instr == FTS_NAMEONLY) {
499 		SET(FTS_NAMEONLY);
500 		instr = BNAMES;
501 	} else
502 		instr = BCHILD;
503 
504 	/*
505 	 * If using chdir on a relative path and called BEFORE fts_read does
506 	 * its chdir to the root of a traversal, we can lose -- we need to
507 	 * chdir into the subdirectory, and we don't know where the current
508 	 * directory is, so we can't get back so that the upcoming chdir by
509 	 * fts_read will work.
510 	 */
511 	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
512 	    ISSET(FTS_NOCHDIR))
513 		return (sp->fts_child = fts_build(sp, instr));
514 
515 	if ((fd = open(".", O_RDONLY, 0)) < 0)
516 		return (NULL);
517 	sp->fts_child = fts_build(sp, instr);
518 	if (fchdir(fd))
519 		return (NULL);
520 	(void)close(fd);
521 	return (sp->fts_child);
522 }
523 
524 /*
525  * This is the tricky part -- do not casually change *anything* in here.  The
526  * idea is to build the linked list of entries that are used by fts_children
527  * and fts_read.  There are lots of special cases.
528  *
529  * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
530  * set and it's a physical walk (so that symbolic links can't be directories),
531  * we can do things quickly.  First, if it's a 4.4BSD file system, the type
532  * of the file is in the directory entry.  Otherwise, we assume that the number
533  * of subdirectories in a node is equal to the number of links to the parent.
534  * The former skips all stat calls.  The latter skips stat calls in any leaf
535  * directories and for any files after the subdirectories in the directory have
536  * been found, cutting the stat calls by about 2/3.
537  */
538 static FTSENT *
539 fts_build(FTS *sp, int type)
540 {
541 	struct dirent *dp;
542 	FTSENT *p, *head;
543 	FTSENT *cur, *tail;
544 	DIR *dirp;
545 	void *oldaddr;
546 	size_t len, maxlen;
547 	int nitems, cderrno, descend, level, nlinks, nostat, doadjust;
548 	int saved_errno;
549 	char *cp;
550 
551 	/* Set current node pointer. */
552 	cur = sp->fts_cur;
553 
554 	/*
555 	 * Open the directory for reading.  If this fails, we're done.
556 	 * If being called from fts_read, set the fts_info field.
557 	 */
558 	if ((dirp = opendir(cur->fts_accpath)) == NULL) {
559 		if (type == BREAD) {
560 			cur->fts_info = FTS_DNR;
561 			cur->fts_errno = errno;
562 		}
563 		return (NULL);
564 	}
565 
566 	/*
567 	 * Nlinks is the number of possible entries of type directory in the
568 	 * directory if we're cheating on stat calls, 0 if we're not doing
569 	 * any stat calls at all, -1 if we're doing stats on everything.
570 	 */
571 	if (type == BNAMES)
572 		nlinks = 0;
573 	else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
574 		nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
575 		nostat = 1;
576 	} else {
577 		nlinks = -1;
578 		nostat = 0;
579 	}
580 
581 #ifdef notdef
582 	(void)printf("nlinks == %d (cur: %u)\n", nlinks, cur->fts_nlink);
583 	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
584 	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
585 #endif
586 	/*
587 	 * If we're going to need to stat anything or we want to descend
588 	 * and stay in the directory, chdir.  If this fails we keep going,
589 	 * but set a flag so we don't chdir after the post-order visit.
590 	 * We won't be able to stat anything, but we can still return the
591 	 * names themselves.  Note, that since fts_read won't be able to
592 	 * chdir into the directory, it will have to return different path
593 	 * names than before, i.e. "a/b" instead of "b".  Since the node
594 	 * has already been visited in pre-order, have to wait until the
595 	 * post-order visit to return the error.  There is a special case
596 	 * here, if there was nothing to stat then it's not an error to
597 	 * not be able to stat.  This is all fairly nasty.  If a program
598 	 * needed sorted entries or stat information, they had better be
599 	 * checking FTS_NS on the returned nodes.
600 	 */
601 	cderrno = 0;
602 	if (nlinks || type == BREAD) {
603 		if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) {
604 			if (nlinks && type == BREAD)
605 				cur->fts_errno = errno;
606 			cur->fts_flags |= FTS_DONTCHDIR;
607 			descend = 0;
608 			cderrno = errno;
609 			(void)closedir(dirp);
610 			dirp = NULL;
611 		} else
612 			descend = 1;
613 	} else
614 		descend = 0;
615 
616 	/*
617 	 * Figure out the max file name length that can be stored in the
618 	 * current path -- the inner loop allocates more path as necessary.
619 	 * We really wouldn't have to do the maxlen calculations here, we
620 	 * could do them in fts_read before returning the path, but it's a
621 	 * lot easier here since the length is part of the dirent structure.
622 	 *
623 	 * If not changing directories set a pointer so that can just append
624 	 * each new name into the path.
625 	 */
626 	len = NAPPEND(cur);
627 	if (ISSET(FTS_NOCHDIR)) {
628 		cp = sp->fts_path + len;
629 		*cp++ = '/';
630 	}
631 	len++;
632 	maxlen = sp->fts_pathlen - len;
633 
634 	level = cur->fts_level + 1;
635 
636 	/* Read the directory, attaching each entry to the `link' pointer. */
637 	doadjust = 0;
638 	for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) {
639 		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
640 			continue;
641 
642 		if (!(p = fts_alloc(sp, dp->d_name, (size_t)dp->d_namlen)))
643 			goto mem1;
644 		if (dp->d_namlen >= maxlen) {	/* include space for NUL */
645 			oldaddr = sp->fts_path;
646 			if (fts_palloc(sp, dp->d_namlen +len + 1)) {
647 				/*
648 				 * No more memory for path or structures.  Save
649 				 * errno, free up the current structure and the
650 				 * structures already allocated.
651 				 */
652 mem1:				saved_errno = errno;
653 				if (p)
654 					free(p);
655 				fts_lfree(head);
656 				(void)closedir(dirp);
657 				cur->fts_info = FTS_ERR;
658 				SET(FTS_STOP);
659 				errno = saved_errno;
660 				return (NULL);
661 			}
662 			/* Did realloc() change the pointer? */
663 			if (oldaddr != sp->fts_path) {
664 				doadjust = 1;
665 				if (ISSET(FTS_NOCHDIR))
666 					cp = sp->fts_path + len;
667 			}
668 			maxlen = sp->fts_pathlen - len;
669 		}
670 
671 		p->fts_level = level;
672 		p->fts_parent = sp->fts_cur;
673 		p->fts_pathlen = len + dp->d_namlen;
674 		if (p->fts_pathlen < len) {
675 			/*
676 			 * If we wrap, free up the current structure and
677 			 * the structures already allocated, then error
678 			 * out with ENAMETOOLONG.
679 			 */
680 			free(p);
681 			fts_lfree(head);
682 			(void)closedir(dirp);
683 			cur->fts_info = FTS_ERR;
684 			SET(FTS_STOP);
685 			errno = ENAMETOOLONG;
686 			return (NULL);
687 		}
688 
689 		if (cderrno) {
690 			if (nlinks) {
691 				p->fts_info = FTS_NS;
692 				p->fts_errno = cderrno;
693 			} else
694 				p->fts_info = FTS_NSOK;
695 			p->fts_accpath = cur->fts_accpath;
696 		} else if (nlinks == 0
697 #ifdef DT_DIR
698 		    || (nostat &&
699 		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
700 #endif
701 		    ) {
702 			p->fts_accpath =
703 			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
704 			p->fts_info = FTS_NSOK;
705 		} else {
706 			/* Build a file name for fts_stat to stat. */
707 			if (ISSET(FTS_NOCHDIR)) {
708 				p->fts_accpath = p->fts_path;
709 				memmove(cp, p->fts_name, p->fts_namelen + 1);
710 			} else
711 				p->fts_accpath = p->fts_name;
712 			/* Stat it. */
713 			p->fts_info = fts_stat(sp, p, 0);
714 
715 			/* Decrement link count if applicable. */
716 			if (nlinks > 0 && (p->fts_info == FTS_D ||
717 			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
718 				--nlinks;
719 		}
720 
721 		/* We walk in directory order so "ls -f" doesn't get upset. */
722 		p->fts_link = NULL;
723 		if (head == NULL)
724 			head = tail = p;
725 		else {
726 			tail->fts_link = p;
727 			tail = p;
728 		}
729 		++nitems;
730 	}
731 	if (dirp)
732 		(void)closedir(dirp);
733 
734 	/*
735 	 * If realloc() changed the address of the path, adjust the
736 	 * addresses for the rest of the tree and the dir list.
737 	 */
738 	if (doadjust)
739 		fts_padjust(sp, head);
740 
741 	/*
742 	 * If not changing directories, reset the path back to original
743 	 * state.
744 	 */
745 	if (ISSET(FTS_NOCHDIR)) {
746 		if (len == sp->fts_pathlen || nitems == 0)
747 			--cp;
748 		*cp = '\0';
749 	}
750 
751 	/*
752 	 * If descended after called from fts_children or after called from
753 	 * fts_read and nothing found, get back.  At the root level we use
754 	 * the saved fd; if one of fts_open()'s arguments is a relative path
755 	 * to an empty directory, we wind up here with no other way back.  If
756 	 * can't get back, we're done.
757 	 */
758 	if (descend && (type == BCHILD || !nitems) &&
759 	    (cur->fts_level == FTS_ROOTLEVEL ? FCHDIR(sp, sp->fts_rfd) :
760 	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
761 		cur->fts_info = FTS_ERR;
762 		SET(FTS_STOP);
763 		return (NULL);
764 	}
765 
766 	/* If didn't find anything, return NULL. */
767 	if (!nitems) {
768 		if (type == BREAD)
769 			cur->fts_info = FTS_DP;
770 		return (NULL);
771 	}
772 
773 	/* Sort the entries. */
774 	if (sp->fts_compar && nitems > 1)
775 		head = fts_sort(sp, head, nitems);
776 	return (head);
777 }
778 
779 static u_short
780 fts_stat(FTS *sp, FTSENT *p, int follow)
781 {
782 	FTSENT *t;
783 	dev_t dev;
784 	ino_t ino;
785 	struct stat *sbp, sb;
786 	int saved_errno;
787 
788 	/* If user needs stat info, stat buffer already allocated. */
789 	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
790 
791 	/*
792 	 * If doing a logical walk, or application requested FTS_FOLLOW, do
793 	 * a stat(2).  If that fails, check for a non-existent symlink.  If
794 	 * fail, set the errno from the stat call.
795 	 */
796 	if (ISSET(FTS_LOGICAL) || follow) {
797 		if (stat(p->fts_accpath, sbp)) {
798 			saved_errno = errno;
799 			if (!lstat(p->fts_accpath, sbp)) {
800 				errno = 0;
801 				return (FTS_SLNONE);
802 			}
803 			p->fts_errno = saved_errno;
804 			goto err;
805 		}
806 	} else if (lstat(p->fts_accpath, sbp)) {
807 		p->fts_errno = errno;
808 err:		memset(sbp, 0, sizeof(struct stat));
809 		return (FTS_NS);
810 	}
811 
812 	if (S_ISDIR(sbp->st_mode)) {
813 		/*
814 		 * Set the device/inode.  Used to find cycles and check for
815 		 * crossing mount points.  Also remember the link count, used
816 		 * in fts_build to limit the number of stat calls.  It is
817 		 * understood that these fields are only referenced if fts_info
818 		 * is set to FTS_D.
819 		 */
820 		dev = p->fts_dev = sbp->st_dev;
821 		ino = p->fts_ino = sbp->st_ino;
822 		p->fts_nlink = sbp->st_nlink;
823 
824 		if (ISDOT(p->fts_name))
825 			return (FTS_DOT);
826 
827 		/*
828 		 * Cycle detection is done by brute force when the directory
829 		 * is first encountered.  If the tree gets deep enough or the
830 		 * number of symbolic links to directories is high enough,
831 		 * something faster might be worthwhile.
832 		 */
833 		for (t = p->fts_parent;
834 		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
835 			if (ino == t->fts_ino && dev == t->fts_dev) {
836 				p->fts_cycle = t;
837 				return (FTS_DC);
838 			}
839 		return (FTS_D);
840 	}
841 	if (S_ISLNK(sbp->st_mode))
842 		return (FTS_SL);
843 	if (S_ISREG(sbp->st_mode))
844 		return (FTS_F);
845 	return (FTS_DEFAULT);
846 }
847 
848 static FTSENT *
849 fts_sort(FTS *sp, FTSENT *head, int nitems)
850 {
851 	FTSENT **ap, *p;
852 
853 	/*
854 	 * Construct an array of pointers to the structures and call qsort(3).
855 	 * Reassemble the array in the order returned by qsort.  If unable to
856 	 * sort for memory reasons, return the directory entries in their
857 	 * current order.  Allocate enough space for the current needs plus
858 	 * 40 so don't realloc one entry at a time.
859 	 */
860 	if (nitems > sp->fts_nitems) {
861 		struct _ftsent **a;
862 
863 		sp->fts_nitems = nitems + 40;
864 		if ((a = realloc(sp->fts_array,
865 		    sp->fts_nitems * sizeof(FTSENT *))) == NULL) {
866 			if (sp->fts_array)
867 				free(sp->fts_array);
868 			sp->fts_array = NULL;
869 			sp->fts_nitems = 0;
870 			return (head);
871 		}
872 		sp->fts_array = a;
873 	}
874 	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
875 		*ap++ = p;
876 	qsort((void *)sp->fts_array, nitems, sizeof(FTSENT *), sp->fts_compar);
877 	for (head = *(ap = sp->fts_array); --nitems; ++ap)
878 		ap[0]->fts_link = ap[1];
879 	ap[0]->fts_link = NULL;
880 	return (head);
881 }
882 
883 static FTSENT *
884 fts_alloc(FTS *sp, char *name, size_t namelen)
885 {
886 	FTSENT *p;
887 	size_t len;
888 
889 	/*
890 	 * The file name is a variable length array and no stat structure is
891 	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
892 	 * structure, the file name and the stat structure in one chunk, but
893 	 * be careful that the stat structure is reasonably aligned.  Since the
894 	 * fts_name field is declared to be of size 1, the fts_name pointer is
895 	 * namelen + 2 before the first possible address of the stat structure.
896 	 */
897 	len = sizeof(FTSENT) + namelen;
898 	if (!ISSET(FTS_NOSTAT))
899 		len += sizeof(struct stat) + ALIGNBYTES;
900 	if ((p = malloc(len)) == NULL)
901 		return (NULL);
902 
903 	memset(p, 0, len);
904 	p->fts_path = sp->fts_path;
905 	p->fts_namelen = namelen;
906 	p->fts_instr = FTS_NOINSTR;
907 	if (!ISSET(FTS_NOSTAT))
908 		p->fts_statp = (struct stat *)ALIGN(p->fts_name + namelen + 2);
909 	memcpy(p->fts_name, name, namelen);
910 
911 	return (p);
912 }
913 
914 static void
915 fts_lfree(FTSENT *head)
916 {
917 	FTSENT *p;
918 
919 	/* Free a linked list of structures. */
920 	while ((p = head)) {
921 		head = head->fts_link;
922 		free(p);
923 	}
924 }
925 
926 /*
927  * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
928  * Most systems will allow creation of paths much longer than MAXPATHLEN, even
929  * though the kernel won't resolve them.  Add the size (not just what's needed)
930  * plus 256 bytes so don't realloc the path 2 bytes at a time.
931  */
932 static int
933 fts_palloc(FTS *sp, size_t more)
934 {
935 	char *p;
936 
937 	/*
938 	 * Check for possible wraparound.
939 	 */
940 	more += 256;
941 	if (sp->fts_pathlen + more < sp->fts_pathlen) {
942 		if (sp->fts_path)
943 			free(sp->fts_path);
944 		sp->fts_path = NULL;
945 		errno = ENAMETOOLONG;
946 		return (1);
947 	}
948 	sp->fts_pathlen += more;
949 	p = realloc(sp->fts_path, sp->fts_pathlen);
950 	if (p == NULL) {
951 		if (sp->fts_path)
952 			free(sp->fts_path);
953 		sp->fts_path = NULL;
954 		return (1);
955 	}
956 	sp->fts_path = p;
957 	return (0);
958 }
959 
960 /*
961  * When the path is realloc'd, have to fix all of the pointers in structures
962  * already returned.
963  */
964 static void
965 fts_padjust(FTS *sp, FTSENT *head)
966 {
967 	FTSENT *p;
968 	char *addr = sp->fts_path;
969 
970 #define	ADJUST(p) {							\
971 	if ((p)->fts_accpath != (p)->fts_name) {			\
972 		(p)->fts_accpath =					\
973 		    (char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
974 	}								\
975 	(p)->fts_path = addr;						\
976 }
977 	/* Adjust the current set of children. */
978 	for (p = sp->fts_child; p; p = p->fts_link)
979 		ADJUST(p);
980 
981 	/* Adjust the rest of the tree, including the current level. */
982 	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
983 		ADJUST(p);
984 		p = p->fts_link ? p->fts_link : p->fts_parent;
985 	}
986 }
987 
988 static size_t
989 fts_maxarglen(char * const *argv)
990 {
991 	size_t len, max;
992 
993 	for (max = 0; *argv; ++argv)
994 		if ((len = strlen(*argv)) > max)
995 			max = len;
996 	return (max + 1);
997 }
998 
999 /*
1000  * Change to dir specified by fd or p->fts_accpath without getting
1001  * tricked by someone changing the world out from underneath us.
1002  * Assumes p->fts_dev and p->fts_ino are filled in.
1003  */
1004 static int
1005 fts_safe_changedir(FTS *sp, FTSENT *p, int fd, char *path)
1006 {
1007 	int ret, oerrno, newfd;
1008 	struct stat sb;
1009 
1010 	newfd = fd;
1011 	if (ISSET(FTS_NOCHDIR))
1012 		return (0);
1013 	if (fd < 0 && (newfd = open(path, O_RDONLY, 0)) < 0)
1014 		return (-1);
1015 	if (fstat(newfd, &sb)) {
1016 		ret = -1;
1017 		goto bail;
1018 	}
1019 	if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) {
1020 		errno = ENOENT;		/* disinformation */
1021 		ret = -1;
1022 		goto bail;
1023 	}
1024 	ret = fchdir(newfd);
1025 bail:
1026 	oerrno = errno;
1027 	if (fd < 0)
1028 		(void)close(newfd);
1029 	errno = oerrno;
1030 	return (ret);
1031 }
1032