xref: /openbsd/lib/libc/gen/fts.c (revision d25d28bf)
1 /*	$OpenBSD: fts.c,v 1.56 2016/09/21 04:38:56 guenther 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>	/* ALIGN */
33 #include <sys/stat.h>
34 
35 #include <dirent.h>
36 #include <errno.h>
37 #include <fcntl.h>
38 #include <fts.h>
39 #include <limits.h>
40 #include <stdlib.h>
41 #include <string.h>
42 #include <unistd.h>
43 
44 #define MAXIMUM(a, b)	(((a) > (b)) ? (a) : (b))
45 
46 static FTSENT	*fts_alloc(FTS *, char *, size_t);
47 static FTSENT	*fts_build(FTS *, int);
48 static void	 fts_lfree(FTSENT *);
49 static void	 fts_load(FTS *, FTSENT *);
50 static size_t	 fts_maxarglen(char * const *);
51 static void	 fts_padjust(FTS *, FTSENT *);
52 static int	 fts_palloc(FTS *, size_t);
53 static FTSENT	*fts_sort(FTS *, FTSENT *, int);
54 static u_short	 fts_stat(FTS *, FTSENT *, int, int);
55 static int	 fts_safe_changedir(FTS *, FTSENT *, int, char *);
56 
57 #define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
58 
59 #define	CLR(opt)	(sp->fts_options &= ~(opt))
60 #define	ISSET(opt)	(sp->fts_options & (opt))
61 #define	SET(opt)	(sp->fts_options |= (opt))
62 
63 #define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
64 
65 /* fts_build flags */
66 #define	BCHILD		1		/* fts_children */
67 #define	BNAMES		2		/* fts_children, names only */
68 #define	BREAD		3		/* fts_read */
69 
70 FTS *
71 fts_open(char * const *argv, int options,
72     int (*compar)(const FTSENT **, const FTSENT **))
73 {
74 	FTS *sp;
75 	FTSENT *p, *root;
76 	int nitems;
77 	FTSENT *parent, *tmp;
78 
79 	/* Options check. */
80 	if (options & ~FTS_OPTIONMASK) {
81 		errno = EINVAL;
82 		return (NULL);
83 	}
84 
85 	/* At least one path must be specified. */
86 	if (*argv == NULL) {
87 		errno = EINVAL;
88 		return (NULL);
89 	}
90 
91 	/* Allocate/initialize the stream */
92 	if ((sp = calloc(1, sizeof(FTS))) == NULL)
93 		return (NULL);
94 	sp->fts_compar = compar;
95 	sp->fts_options = options;
96 
97 	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
98 	if (ISSET(FTS_LOGICAL))
99 		SET(FTS_NOCHDIR);
100 
101 	/*
102 	 * Start out with 1K of path space, and enough, in any case,
103 	 * to hold the user's paths.
104 	 */
105 	if (fts_palloc(sp, MAXIMUM(fts_maxarglen(argv), PATH_MAX)))
106 		goto mem1;
107 
108 	/* Allocate/initialize root's parent. */
109 	if ((parent = fts_alloc(sp, "", 0)) == NULL)
110 		goto mem2;
111 	parent->fts_level = FTS_ROOTPARENTLEVEL;
112 
113 	/* Allocate/initialize root(s). */
114 	for (root = NULL, nitems = 0; *argv; ++argv, ++nitems) {
115 		if ((p = fts_alloc(sp, *argv, strlen(*argv))) == NULL)
116 			goto mem3;
117 		p->fts_level = FTS_ROOTLEVEL;
118 		p->fts_parent = parent;
119 		p->fts_accpath = p->fts_name;
120 		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW), -1);
121 
122 		/* Command-line "." and ".." are real directories. */
123 		if (p->fts_info == FTS_DOT)
124 			p->fts_info = FTS_D;
125 
126 		/*
127 		 * If comparison routine supplied, traverse in sorted
128 		 * order; otherwise traverse in the order specified.
129 		 */
130 		if (compar) {
131 			p->fts_link = root;
132 			root = p;
133 		} else {
134 			p->fts_link = NULL;
135 			if (root == NULL)
136 				tmp = root = p;
137 			else {
138 				tmp->fts_link = p;
139 				tmp = p;
140 			}
141 		}
142 	}
143 	if (compar && nitems > 1)
144 		root = fts_sort(sp, root, nitems);
145 
146 	/*
147 	 * Allocate a dummy pointer and make fts_read think that we've just
148 	 * finished the node before the root(s); set p->fts_info to FTS_INIT
149 	 * so that everything about the "current" node is ignored.
150 	 */
151 	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
152 		goto mem3;
153 	sp->fts_cur->fts_link = root;
154 	sp->fts_cur->fts_info = FTS_INIT;
155 
156 	/*
157 	 * If using chdir(2), grab a file descriptor pointing to dot to ensure
158 	 * that we can get back here; this could be avoided for some paths,
159 	 * but almost certainly not worth the effort.  Slashes, symbolic links,
160 	 * and ".." are all fairly nasty problems.  Note, if we can't get the
161 	 * descriptor we run anyway, just more slowly.
162 	 */
163 	if (!ISSET(FTS_NOCHDIR) &&
164 	    (sp->fts_rfd = open(".", O_RDONLY | O_CLOEXEC)) < 0)
165 		SET(FTS_NOCHDIR);
166 
167 	if (nitems == 0)
168 		free(parent);
169 
170 	return (sp);
171 
172 mem3:	fts_lfree(root);
173 	free(parent);
174 mem2:	free(sp->fts_path);
175 mem1:	free(sp);
176 	return (NULL);
177 }
178 DEF_WEAK(fts_open);
179 
180 static void
181 fts_load(FTS *sp, FTSENT *p)
182 {
183 	size_t len;
184 	char *cp;
185 
186 	/*
187 	 * Load the stream structure for the next traversal.  Since we don't
188 	 * actually enter the directory until after the preorder visit, set
189 	 * the fts_accpath field specially so the chdir gets done to the right
190 	 * place and the user can access the first node.  From fts_open it's
191 	 * known that the path will fit.
192 	 */
193 	len = p->fts_pathlen = p->fts_namelen;
194 	memmove(sp->fts_path, p->fts_name, len + 1);
195 	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
196 		len = strlen(++cp);
197 		memmove(p->fts_name, cp, len + 1);
198 		p->fts_namelen = len;
199 	}
200 	p->fts_accpath = p->fts_path = sp->fts_path;
201 	sp->fts_dev = p->fts_dev;
202 }
203 
204 int
205 fts_close(FTS *sp)
206 {
207 	FTSENT *freep, *p;
208 	int rfd, error = 0;
209 
210 	/*
211 	 * This still works if we haven't read anything -- the dummy structure
212 	 * points to the root list, so we step through to the end of the root
213 	 * list which has a valid parent pointer.
214 	 */
215 	if (sp->fts_cur) {
216 		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
217 			freep = p;
218 			p = p->fts_link ? p->fts_link : p->fts_parent;
219 			free(freep);
220 		}
221 		free(p);
222 	}
223 
224 	/* Stash the original directory fd if needed. */
225 	rfd = ISSET(FTS_NOCHDIR) ? -1 : sp->fts_rfd;
226 
227 	/* Free up child linked list, sort array, path buffer, stream ptr.*/
228 	if (sp->fts_child)
229 		fts_lfree(sp->fts_child);
230 	free(sp->fts_array);
231 	free(sp->fts_path);
232 	free(sp);
233 
234 	/* Return to original directory, checking for error. */
235 	if (rfd != -1) {
236 		int saved_errno;
237 		error = fchdir(rfd);
238 		saved_errno = errno;
239 		(void)close(rfd);
240 		errno = saved_errno;
241 	}
242 
243 	return (error);
244 }
245 DEF_WEAK(fts_close);
246 
247 /*
248  * Special case of "/" at the end of the path so that slashes aren't
249  * appended which would cause paths to be written as "....//foo".
250  */
251 #define	NAPPEND(p)							\
252 	(p->fts_path[p->fts_pathlen - 1] == '/'				\
253 	    ? p->fts_pathlen - 1 : p->fts_pathlen)
254 
255 FTSENT *
256 fts_read(FTS *sp)
257 {
258 	FTSENT *p, *tmp;
259 	int instr;
260 	char *t;
261 	int saved_errno;
262 
263 	/* If finished or unrecoverable error, return NULL. */
264 	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
265 		return (NULL);
266 
267 	/* Set current node pointer. */
268 	p = sp->fts_cur;
269 
270 	/* Save and zero out user instructions. */
271 	instr = p->fts_instr;
272 	p->fts_instr = FTS_NOINSTR;
273 
274 	/* Any type of file may be re-visited; re-stat and re-turn. */
275 	if (instr == FTS_AGAIN) {
276 		p->fts_info = fts_stat(sp, p, 0, -1);
277 		return (p);
278 	}
279 
280 	/*
281 	 * Following a symlink -- SLNONE test allows application to see
282 	 * SLNONE and recover.  If indirecting through a symlink, have
283 	 * keep a pointer to current location.  If unable to get that
284 	 * pointer, follow fails.
285 	 */
286 	if (instr == FTS_FOLLOW &&
287 	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
288 		p->fts_info = fts_stat(sp, p, 1, -1);
289 		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
290 			if ((p->fts_symfd =
291 			    open(".", O_RDONLY | O_CLOEXEC)) < 0) {
292 				p->fts_errno = errno;
293 				p->fts_info = FTS_ERR;
294 			} else
295 				p->fts_flags |= FTS_SYMFOLLOW;
296 		}
297 		return (p);
298 	}
299 
300 	/* Directory in pre-order. */
301 	if (p->fts_info == FTS_D) {
302 		/* If skipped or crossed mount point, do post-order visit. */
303 		if (instr == FTS_SKIP ||
304 		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
305 			if (p->fts_flags & FTS_SYMFOLLOW)
306 				(void)close(p->fts_symfd);
307 			if (sp->fts_child) {
308 				fts_lfree(sp->fts_child);
309 				sp->fts_child = NULL;
310 			}
311 			p->fts_info = FTS_DP;
312 			return (p);
313 		}
314 
315 		/* Rebuild if only read the names and now traversing. */
316 		if (sp->fts_child && ISSET(FTS_NAMEONLY)) {
317 			CLR(FTS_NAMEONLY);
318 			fts_lfree(sp->fts_child);
319 			sp->fts_child = NULL;
320 		}
321 
322 		/*
323 		 * Cd to the subdirectory.
324 		 *
325 		 * If have already read and now fail to chdir, whack the list
326 		 * to make the names come out right, and set the parent errno
327 		 * so the application will eventually get an error condition.
328 		 * Set the FTS_DONTCHDIR flag so that when we logically change
329 		 * directories back to the parent we don't do a chdir.
330 		 *
331 		 * If haven't read do so.  If the read fails, fts_build sets
332 		 * FTS_STOP or the fts_info field of the node.
333 		 */
334 		if (sp->fts_child) {
335 			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
336 				p->fts_errno = errno;
337 				p->fts_flags |= FTS_DONTCHDIR;
338 				for (p = sp->fts_child; p; p = p->fts_link)
339 					p->fts_accpath =
340 					    p->fts_parent->fts_accpath;
341 			}
342 		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
343 			if (ISSET(FTS_STOP))
344 				return (NULL);
345 			return (p);
346 		}
347 		p = sp->fts_child;
348 		sp->fts_child = NULL;
349 		goto name;
350 	}
351 
352 	/* Move to the next node on this level. */
353 next:	tmp = p;
354 	if ((p = p->fts_link)) {
355 		free(tmp);
356 
357 		/*
358 		 * If reached the top, return to the original directory (or
359 		 * the root of the tree), and load the paths for the next root.
360 		 */
361 		if (p->fts_level == FTS_ROOTLEVEL) {
362 			if (FCHDIR(sp, sp->fts_rfd)) {
363 				SET(FTS_STOP);
364 				return (NULL);
365 			}
366 			fts_load(sp, p);
367 			return (sp->fts_cur = p);
368 		}
369 
370 		/*
371 		 * User may have called fts_set on the node.  If skipped,
372 		 * ignore.  If followed, get a file descriptor so we can
373 		 * get back if necessary.
374 		 */
375 		if (p->fts_instr == FTS_SKIP)
376 			goto next;
377 		if (p->fts_instr == FTS_FOLLOW) {
378 			p->fts_info = fts_stat(sp, p, 1, -1);
379 			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
380 				if ((p->fts_symfd =
381 				    open(".", O_RDONLY | O_CLOEXEC)) < 0) {
382 					p->fts_errno = errno;
383 					p->fts_info = FTS_ERR;
384 				} else
385 					p->fts_flags |= FTS_SYMFOLLOW;
386 			}
387 			p->fts_instr = FTS_NOINSTR;
388 		}
389 
390 name:		t = sp->fts_path + NAPPEND(p->fts_parent);
391 		*t++ = '/';
392 		memmove(t, p->fts_name, p->fts_namelen + 1);
393 		return (sp->fts_cur = p);
394 	}
395 
396 	/* Move up to the parent node. */
397 	p = tmp->fts_parent;
398 	free(tmp);
399 
400 	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
401 		/*
402 		 * Done; free everything up and set errno to 0 so the user
403 		 * can distinguish between error and EOF.
404 		 */
405 		free(p);
406 		errno = 0;
407 		return (sp->fts_cur = NULL);
408 	}
409 
410 	/* NUL terminate the pathname. */
411 	sp->fts_path[p->fts_pathlen] = '\0';
412 
413 	/*
414 	 * Return to the parent directory.  If at a root node or came through
415 	 * a symlink, go back through the file descriptor.  Otherwise, cd up
416 	 * one directory.
417 	 */
418 	if (p->fts_level == FTS_ROOTLEVEL) {
419 		if (FCHDIR(sp, sp->fts_rfd)) {
420 			SET(FTS_STOP);
421 			sp->fts_cur = p;
422 			return (NULL);
423 		}
424 	} else if (p->fts_flags & FTS_SYMFOLLOW) {
425 		if (FCHDIR(sp, p->fts_symfd)) {
426 			saved_errno = errno;
427 			(void)close(p->fts_symfd);
428 			errno = saved_errno;
429 			SET(FTS_STOP);
430 			sp->fts_cur = p;
431 			return (NULL);
432 		}
433 		(void)close(p->fts_symfd);
434 	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
435 	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
436 		SET(FTS_STOP);
437 		sp->fts_cur = p;
438 		return (NULL);
439 	}
440 	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
441 	return (sp->fts_cur = p);
442 }
443 DEF_WEAK(fts_read);
444 
445 /*
446  * Fts_set takes the stream as an argument although it's not used in this
447  * implementation; it would be necessary if anyone wanted to add global
448  * semantics to fts using fts_set.  An error return is allowed for similar
449  * reasons.
450  */
451 int
452 fts_set(FTS *sp, FTSENT *p, int instr)
453 {
454 	if (instr && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
455 	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
456 		errno = EINVAL;
457 		return (1);
458 	}
459 	p->fts_instr = instr;
460 	return (0);
461 }
462 DEF_WEAK(fts_set);
463 
464 FTSENT *
465 fts_children(FTS *sp, int instr)
466 {
467 	FTSENT *p;
468 	int fd;
469 
470 	if (instr && instr != FTS_NAMEONLY) {
471 		errno = EINVAL;
472 		return (NULL);
473 	}
474 
475 	/* Set current node pointer. */
476 	p = sp->fts_cur;
477 
478 	/*
479 	 * Errno set to 0 so user can distinguish empty directory from
480 	 * an error.
481 	 */
482 	errno = 0;
483 
484 	/* Fatal errors stop here. */
485 	if (ISSET(FTS_STOP))
486 		return (NULL);
487 
488 	/* Return logical hierarchy of user's arguments. */
489 	if (p->fts_info == FTS_INIT)
490 		return (p->fts_link);
491 
492 	/*
493 	 * If not a directory being visited in pre-order, stop here.  Could
494 	 * allow FTS_DNR, assuming the user has fixed the problem, but the
495 	 * same effect is available with FTS_AGAIN.
496 	 */
497 	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
498 		return (NULL);
499 
500 	/* Free up any previous child list. */
501 	if (sp->fts_child)
502 		fts_lfree(sp->fts_child);
503 
504 	if (instr == FTS_NAMEONLY) {
505 		SET(FTS_NAMEONLY);
506 		instr = BNAMES;
507 	} else
508 		instr = BCHILD;
509 
510 	/*
511 	 * If using chdir on a relative path and called BEFORE fts_read does
512 	 * its chdir to the root of a traversal, we can lose -- we need to
513 	 * chdir into the subdirectory, and we don't know where the current
514 	 * directory is, so we can't get back so that the upcoming chdir by
515 	 * fts_read will work.
516 	 */
517 	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
518 	    ISSET(FTS_NOCHDIR))
519 		return (sp->fts_child = fts_build(sp, instr));
520 
521 	if ((fd = open(".", O_RDONLY | O_CLOEXEC)) < 0)
522 		return (NULL);
523 	sp->fts_child = fts_build(sp, instr);
524 	if (fchdir(fd)) {
525 		(void)close(fd);
526 		return (NULL);
527 	}
528 	(void)close(fd);
529 	return (sp->fts_child);
530 }
531 DEF_WEAK(fts_children);
532 
533 /*
534  * This is the tricky part -- do not casually change *anything* in here.  The
535  * idea is to build the linked list of entries that are used by fts_children
536  * and fts_read.  There are lots of special cases.
537  *
538  * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
539  * set and it's a physical walk (so that symbolic links can't be directories),
540  * we can do things quickly.  First, if it's a 4.4BSD file system, the type
541  * of the file is in the directory entry.  Otherwise, we assume that the number
542  * of subdirectories in a node is equal to the number of links to the parent.
543  * The former skips all stat calls.  The latter skips stat calls in any leaf
544  * directories and for any files after the subdirectories in the directory have
545  * been found, cutting the stat calls by about 2/3.
546  */
547 static FTSENT *
548 fts_build(FTS *sp, int type)
549 {
550 	struct dirent *dp;
551 	FTSENT *p, *head;
552 	FTSENT *cur, *tail;
553 	DIR *dirp;
554 	void *oldaddr;
555 	size_t len, maxlen;
556 	int nitems, cderrno, descend, level, nlinks, nostat, doadjust;
557 	int saved_errno;
558 	char *cp;
559 
560 	/* Set current node pointer. */
561 	cur = sp->fts_cur;
562 
563 	/*
564 	 * Open the directory for reading.  If this fails, we're done.
565 	 * If being called from fts_read, set the fts_info field.
566 	 */
567 	if ((dirp = opendir(cur->fts_accpath)) == NULL) {
568 		if (type == BREAD) {
569 			cur->fts_info = FTS_DNR;
570 			cur->fts_errno = errno;
571 		}
572 		return (NULL);
573 	}
574 
575 	/*
576 	 * Nlinks is the number of possible entries of type directory in the
577 	 * directory if we're cheating on stat calls, 0 if we're not doing
578 	 * any stat calls at all, -1 if we're doing stats on everything.
579 	 */
580 	if (type == BNAMES)
581 		nlinks = 0;
582 	else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
583 		nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
584 		nostat = 1;
585 	} else {
586 		nlinks = -1;
587 		nostat = 0;
588 	}
589 
590 #ifdef notdef
591 	(void)printf("nlinks == %d (cur: %u)\n", nlinks, cur->fts_nlink);
592 	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
593 	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
594 #endif
595 	/*
596 	 * If we're going to need to stat anything or we want to descend
597 	 * and stay in the directory, chdir.  If this fails we keep going,
598 	 * but set a flag so we don't chdir after the post-order visit.
599 	 * We won't be able to stat anything, but we can still return the
600 	 * names themselves.  Note, that since fts_read won't be able to
601 	 * chdir into the directory, it will have to return different path
602 	 * names than before, i.e. "a/b" instead of "b".  Since the node
603 	 * has already been visited in pre-order, have to wait until the
604 	 * post-order visit to return the error.  There is a special case
605 	 * here, if there was nothing to stat then it's not an error to
606 	 * not be able to stat.  This is all fairly nasty.  If a program
607 	 * needed sorted entries or stat information, they had better be
608 	 * checking FTS_NS on the returned nodes.
609 	 */
610 	cderrno = 0;
611 	if (nlinks || type == BREAD) {
612 		if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) {
613 			if (nlinks && type == BREAD)
614 				cur->fts_errno = errno;
615 			cur->fts_flags |= FTS_DONTCHDIR;
616 			descend = 0;
617 			cderrno = errno;
618 			(void)closedir(dirp);
619 			dirp = NULL;
620 		} else
621 			descend = 1;
622 	} else
623 		descend = 0;
624 
625 	/*
626 	 * Figure out the max file name length that can be stored in the
627 	 * current path -- the inner loop allocates more path as necessary.
628 	 * We really wouldn't have to do the maxlen calculations here, we
629 	 * could do them in fts_read before returning the path, but it's a
630 	 * lot easier here since the length is part of the dirent structure.
631 	 *
632 	 * If not changing directories set a pointer so that can just append
633 	 * each new name into the path.
634 	 */
635 	len = NAPPEND(cur);
636 	if (ISSET(FTS_NOCHDIR)) {
637 		cp = sp->fts_path + len;
638 		*cp++ = '/';
639 	}
640 	len++;
641 	maxlen = sp->fts_pathlen - len;
642 
643 	/*
644 	 * fts_level is signed so we must prevent it from wrapping
645 	 * around to FTS_ROOTLEVEL and FTS_ROOTPARENTLEVEL.
646 	 */
647 	level = cur->fts_level;
648 	if (level < FTS_MAXLEVEL)
649 	    level++;
650 
651 	/* Read the directory, attaching each entry to the `link' pointer. */
652 	doadjust = 0;
653 	for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) {
654 		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
655 			continue;
656 
657 		if (!(p = fts_alloc(sp, dp->d_name, dp->d_namlen)))
658 			goto mem1;
659 		if (dp->d_namlen >= maxlen) {	/* include space for NUL */
660 			oldaddr = sp->fts_path;
661 			if (fts_palloc(sp, dp->d_namlen +len + 1)) {
662 				/*
663 				 * No more memory for path or structures.  Save
664 				 * errno, free up the current structure and the
665 				 * structures already allocated.
666 				 */
667 mem1:				saved_errno = errno;
668 				free(p);
669 				fts_lfree(head);
670 				(void)closedir(dirp);
671 				cur->fts_info = FTS_ERR;
672 				SET(FTS_STOP);
673 				errno = saved_errno;
674 				return (NULL);
675 			}
676 			/* Did realloc() change the pointer? */
677 			if (oldaddr != sp->fts_path) {
678 				doadjust = 1;
679 				if (ISSET(FTS_NOCHDIR))
680 					cp = sp->fts_path + len;
681 			}
682 			maxlen = sp->fts_pathlen - len;
683 		}
684 
685 		p->fts_level = level;
686 		p->fts_parent = sp->fts_cur;
687 		p->fts_pathlen = len + dp->d_namlen;
688 		if (p->fts_pathlen < len) {
689 			/*
690 			 * If we wrap, free up the current structure and
691 			 * the structures already allocated, then error
692 			 * out with ENAMETOOLONG.
693 			 */
694 			free(p);
695 			fts_lfree(head);
696 			(void)closedir(dirp);
697 			cur->fts_info = FTS_ERR;
698 			SET(FTS_STOP);
699 			errno = ENAMETOOLONG;
700 			return (NULL);
701 		}
702 
703 		if (cderrno) {
704 			if (nlinks) {
705 				p->fts_info = FTS_NS;
706 				p->fts_errno = cderrno;
707 			} else
708 				p->fts_info = FTS_NSOK;
709 			p->fts_accpath = cur->fts_accpath;
710 		} else if (nlinks == 0
711 #ifdef DT_DIR
712 		    || (nostat &&
713 		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
714 #endif
715 		    ) {
716 			p->fts_accpath =
717 			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
718 			p->fts_info = FTS_NSOK;
719 		} else {
720 			/* Build a file name for fts_stat to stat. */
721 			if (ISSET(FTS_NOCHDIR)) {
722 				p->fts_accpath = p->fts_path;
723 				memmove(cp, p->fts_name, p->fts_namelen + 1);
724 				p->fts_info = fts_stat(sp, p, 0, dirfd(dirp));
725 			} else {
726 				p->fts_accpath = p->fts_name;
727 				p->fts_info = fts_stat(sp, p, 0, -1);
728 			}
729 
730 			/* Decrement link count if applicable. */
731 			if (nlinks > 0 && (p->fts_info == FTS_D ||
732 			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
733 				--nlinks;
734 		}
735 
736 		/* We walk in directory order so "ls -f" doesn't get upset. */
737 		p->fts_link = NULL;
738 		if (head == NULL)
739 			head = tail = p;
740 		else {
741 			tail->fts_link = p;
742 			tail = p;
743 		}
744 		++nitems;
745 	}
746 	if (dirp)
747 		(void)closedir(dirp);
748 
749 	/*
750 	 * If realloc() changed the address of the path, adjust the
751 	 * addresses for the rest of the tree and the dir list.
752 	 */
753 	if (doadjust)
754 		fts_padjust(sp, head);
755 
756 	/*
757 	 * If not changing directories, reset the path back to original
758 	 * state.
759 	 */
760 	if (ISSET(FTS_NOCHDIR)) {
761 		if (len == sp->fts_pathlen || nitems == 0)
762 			--cp;
763 		*cp = '\0';
764 	}
765 
766 	/*
767 	 * If descended after called from fts_children or after called from
768 	 * fts_read and nothing found, get back.  At the root level we use
769 	 * the saved fd; if one of fts_open()'s arguments is a relative path
770 	 * to an empty directory, we wind up here with no other way back.  If
771 	 * can't get back, we're done.
772 	 */
773 	if (descend && (type == BCHILD || !nitems) &&
774 	    (cur->fts_level == FTS_ROOTLEVEL ? FCHDIR(sp, sp->fts_rfd) :
775 	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
776 		cur->fts_info = FTS_ERR;
777 		SET(FTS_STOP);
778 		return (NULL);
779 	}
780 
781 	/* If didn't find anything, return NULL. */
782 	if (!nitems) {
783 		if (type == BREAD)
784 			cur->fts_info = FTS_DP;
785 		return (NULL);
786 	}
787 
788 	/* Sort the entries. */
789 	if (sp->fts_compar && nitems > 1)
790 		head = fts_sort(sp, head, nitems);
791 	return (head);
792 }
793 
794 static u_short
795 fts_stat(FTS *sp, FTSENT *p, int follow, int dfd)
796 {
797 	FTSENT *t;
798 	dev_t dev;
799 	ino_t ino;
800 	struct stat *sbp, sb;
801 	int saved_errno;
802 	const char *path;
803 
804 	if (dfd == -1) {
805 		path = p->fts_accpath;
806 		dfd = AT_FDCWD;
807 	} else
808 		path = p->fts_name;
809 
810 	/* If user needs stat info, stat buffer already allocated. */
811 	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
812 
813 	/*
814 	 * If doing a logical walk, or application requested FTS_FOLLOW, do
815 	 * a stat(2).  If that fails, check for a non-existent symlink.  If
816 	 * fail, set the errno from the stat call.
817 	 */
818 	if (ISSET(FTS_LOGICAL) || follow) {
819 		if (fstatat(dfd, path, sbp, 0)) {
820 			saved_errno = errno;
821 			if (!fstatat(dfd, path, sbp, AT_SYMLINK_NOFOLLOW)) {
822 				errno = 0;
823 				return (FTS_SLNONE);
824 			}
825 			p->fts_errno = saved_errno;
826 			goto err;
827 		}
828 	} else if (fstatat(dfd, path, sbp, AT_SYMLINK_NOFOLLOW)) {
829 		p->fts_errno = errno;
830 err:		memset(sbp, 0, sizeof(struct stat));
831 		return (FTS_NS);
832 	}
833 
834 	if (S_ISDIR(sbp->st_mode)) {
835 		/*
836 		 * Set the device/inode.  Used to find cycles and check for
837 		 * crossing mount points.  Also remember the link count, used
838 		 * in fts_build to limit the number of stat calls.  It is
839 		 * understood that these fields are only referenced if fts_info
840 		 * is set to FTS_D.
841 		 */
842 		dev = p->fts_dev = sbp->st_dev;
843 		ino = p->fts_ino = sbp->st_ino;
844 		p->fts_nlink = sbp->st_nlink;
845 
846 		if (ISDOT(p->fts_name))
847 			return (FTS_DOT);
848 
849 		/*
850 		 * Cycle detection is done by brute force when the directory
851 		 * is first encountered.  If the tree gets deep enough or the
852 		 * number of symbolic links to directories is high enough,
853 		 * something faster might be worthwhile.
854 		 */
855 		for (t = p->fts_parent;
856 		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
857 			if (ino == t->fts_ino && dev == t->fts_dev) {
858 				p->fts_cycle = t;
859 				return (FTS_DC);
860 			}
861 		return (FTS_D);
862 	}
863 	if (S_ISLNK(sbp->st_mode))
864 		return (FTS_SL);
865 	if (S_ISREG(sbp->st_mode))
866 		return (FTS_F);
867 	return (FTS_DEFAULT);
868 }
869 
870 static FTSENT *
871 fts_sort(FTS *sp, FTSENT *head, int nitems)
872 {
873 	FTSENT **ap, *p;
874 
875 	/*
876 	 * Construct an array of pointers to the structures and call qsort(3).
877 	 * Reassemble the array in the order returned by qsort.  If unable to
878 	 * sort for memory reasons, return the directory entries in their
879 	 * current order.  Allocate enough space for the current needs plus
880 	 * 40 so don't realloc one entry at a time.
881 	 */
882 	if (nitems > sp->fts_nitems) {
883 		struct _ftsent **a;
884 
885 		sp->fts_nitems = nitems + 40;
886 		if ((a = reallocarray(sp->fts_array,
887 		    sp->fts_nitems, sizeof(FTSENT *))) == NULL) {
888 			free(sp->fts_array);
889 			sp->fts_array = NULL;
890 			sp->fts_nitems = 0;
891 			return (head);
892 		}
893 		sp->fts_array = a;
894 	}
895 	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
896 		*ap++ = p;
897 	qsort(sp->fts_array, nitems, sizeof(FTSENT *), sp->fts_compar);
898 	for (head = *(ap = sp->fts_array); --nitems; ++ap)
899 		ap[0]->fts_link = ap[1];
900 	ap[0]->fts_link = NULL;
901 	return (head);
902 }
903 
904 static FTSENT *
905 fts_alloc(FTS *sp, char *name, size_t namelen)
906 {
907 	FTSENT *p;
908 	size_t len;
909 
910 	/*
911 	 * The file name is a variable length array and no stat structure is
912 	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
913 	 * structure, the file name and the stat structure in one chunk, but
914 	 * be careful that the stat structure is reasonably aligned.  Since the
915 	 * fts_name field is declared to be of size 1, the fts_name pointer is
916 	 * namelen + 2 before the first possible address of the stat structure.
917 	 */
918 	len = sizeof(FTSENT) + namelen;
919 	if (!ISSET(FTS_NOSTAT))
920 		len += sizeof(struct stat) + ALIGNBYTES;
921 	if ((p = calloc(1, len)) == NULL)
922 		return (NULL);
923 
924 	p->fts_path = sp->fts_path;
925 	p->fts_namelen = namelen;
926 	p->fts_instr = FTS_NOINSTR;
927 	if (!ISSET(FTS_NOSTAT))
928 		p->fts_statp = (struct stat *)ALIGN(p->fts_name + namelen + 2);
929 	memcpy(p->fts_name, name, namelen);
930 
931 	return (p);
932 }
933 
934 static void
935 fts_lfree(FTSENT *head)
936 {
937 	FTSENT *p;
938 
939 	/* Free a linked list of structures. */
940 	while ((p = head)) {
941 		head = head->fts_link;
942 		free(p);
943 	}
944 }
945 
946 /*
947  * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
948  * Most systems will allow creation of paths much longer than PATH_MAX, even
949  * though the kernel won't resolve them.  Add the size (not just what's needed)
950  * plus 256 bytes so don't realloc the path 2 bytes at a time.
951  */
952 static int
953 fts_palloc(FTS *sp, size_t more)
954 {
955 	char *p;
956 
957 	/*
958 	 * Check for possible wraparound.
959 	 */
960 	more += 256;
961 	if (sp->fts_pathlen + more < sp->fts_pathlen) {
962 		free(sp->fts_path);
963 		sp->fts_path = NULL;
964 		errno = ENAMETOOLONG;
965 		return (1);
966 	}
967 	sp->fts_pathlen += more;
968 	p = realloc(sp->fts_path, sp->fts_pathlen);
969 	if (p == NULL) {
970 		free(sp->fts_path);
971 		sp->fts_path = NULL;
972 		return (1);
973 	}
974 	sp->fts_path = p;
975 	return (0);
976 }
977 
978 /*
979  * When the path is realloc'd, have to fix all of the pointers in structures
980  * already returned.
981  */
982 static void
983 fts_padjust(FTS *sp, FTSENT *head)
984 {
985 	FTSENT *p;
986 	char *addr = sp->fts_path;
987 
988 #define	ADJUST(p) {							\
989 	if ((p)->fts_accpath != (p)->fts_name) {			\
990 		(p)->fts_accpath =					\
991 		    (char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
992 	}								\
993 	(p)->fts_path = addr;						\
994 }
995 	/* Adjust the current set of children. */
996 	for (p = sp->fts_child; p; p = p->fts_link)
997 		ADJUST(p);
998 
999 	/* Adjust the rest of the tree, including the current level. */
1000 	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1001 		ADJUST(p);
1002 		p = p->fts_link ? p->fts_link : p->fts_parent;
1003 	}
1004 }
1005 
1006 static size_t
1007 fts_maxarglen(char * const *argv)
1008 {
1009 	size_t len, max;
1010 
1011 	for (max = 0; *argv; ++argv)
1012 		if ((len = strlen(*argv)) > max)
1013 			max = len;
1014 	return (max + 1);
1015 }
1016 
1017 /*
1018  * Change to dir specified by fd or p->fts_accpath without getting
1019  * tricked by someone changing the world out from underneath us.
1020  * Assumes p->fts_dev and p->fts_ino are filled in.
1021  */
1022 static int
1023 fts_safe_changedir(FTS *sp, FTSENT *p, int fd, char *path)
1024 {
1025 	int ret, oerrno, newfd;
1026 	struct stat sb;
1027 
1028 	newfd = fd;
1029 	if (ISSET(FTS_NOCHDIR))
1030 		return (0);
1031 	if (fd < 0 && (newfd = open(path, O_RDONLY|O_DIRECTORY|O_CLOEXEC)) < 0)
1032 		return (-1);
1033 	if (fstat(newfd, &sb)) {
1034 		ret = -1;
1035 		goto bail;
1036 	}
1037 	if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) {
1038 		errno = ENOENT;		/* disinformation */
1039 		ret = -1;
1040 		goto bail;
1041 	}
1042 	ret = fchdir(newfd);
1043 bail:
1044 	oerrno = errno;
1045 	if (fd < 0)
1046 		(void)close(newfd);
1047 	errno = oerrno;
1048 	return (ret);
1049 }
1050