xref: /illumos-gate/usr/src/lib/libc/port/gen/fts.c (revision e3ae4b35c024af1196582063ecee3ab79367227d)
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. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/stat.h>
31 
32 #include <dirent.h>
33 #include <errno.h>
34 #include <fcntl.h>
35 #include <fts.h>
36 #include <limits.h>
37 #include <stdlib.h>
38 #include <string.h>
39 #include <unistd.h>
40 
41 #if !defined(_LP64) && _FILE_OFFSET_BITS == 64
42 #define	fts_alloc		fts_alloc64
43 #define	fts_build		fts_build64
44 #define	fts_lfree		fts_lfree64
45 #define	fts_load		fts_load64
46 #define	fts_maxarglen		fts_maxarglen64
47 #define	fts_padjust		fts_padjust64
48 #define	fts_palloc		fts_palloc64
49 #define	fts_sort		fts_sort64
50 #define	fts_stat		fts_stat64
51 #define	fts_safe_changedir	fts_safe_changedir64
52 #endif
53 
54 #define	MAXIMUM(a, b)	(((a) > (b)) ? (a) : (b))
55 
56 #define	ALIGNBYTES	_POINTER_ALIGNMENT
57 #define	ALIGN(p)	(((unsigned long)(p) + ALIGNBYTES) & ~ALIGNBYTES)
58 
59 static FTSENT	*fts_alloc(FTS *, char *, size_t);
60 static FTSENT	*fts_build(FTS *, int);
61 static void	 fts_lfree(FTSENT *);
62 static void	 fts_load(FTS *, FTSENT *);
63 static size_t	 fts_maxarglen(char * const *);
64 static void	 fts_padjust(FTS *, FTSENT *);
65 static int	 fts_palloc(FTS *, size_t);
66 static FTSENT	*fts_sort(FTS *, FTSENT *, int);
67 static ushort_t	 fts_stat(FTS *, FTSENT *, int, int);
68 static int	 fts_safe_changedir(FTS *, FTSENT *, int, char *);
69 
70 #define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
71 
72 #define	CLR(opt)	(sp->fts_options &= ~(opt))
73 #define	ISSET(opt)	(sp->fts_options & (opt))
74 #define	SET(opt)	(sp->fts_options |= (opt))
75 
76 #define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
77 
78 /* fts_build flags */
79 #define	BCHILD		1		/* fts_children */
80 #define	BNAMES		2		/* fts_children, names only */
81 #define	BREAD		3		/* fts_read */
82 
83 FTS *
84 fts_open(char * const *argv, int options,
85     int (*compar)(const FTSENT **, const FTSENT **))
86 {
87 	FTS *sp;
88 	FTSENT *p, *root;
89 	int nitems;
90 	FTSENT *parent, *prev;
91 
92 	/* Options check. */
93 	if (options & ~FTS_OPTIONMASK) {
94 		errno = EINVAL;
95 		return (NULL);
96 	}
97 
98 	/* At least one path must be specified. */
99 	if (*argv == NULL) {
100 		errno = EINVAL;
101 		return (NULL);
102 	}
103 
104 	/* Allocate/initialize the stream */
105 	if ((sp = calloc(1, sizeof (FTS))) == NULL)
106 		return (NULL);
107 	sp->fts_compar = compar;
108 	sp->fts_options = options;
109 
110 	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
111 	if (ISSET(FTS_LOGICAL))
112 		SET(FTS_NOCHDIR);
113 
114 	/*
115 	 * Start out with 1K of path space, and enough, in any case,
116 	 * to hold the user's paths.
117 	 */
118 	if (fts_palloc(sp, MAXIMUM(fts_maxarglen(argv), PATH_MAX)))
119 		goto mem1;
120 
121 	/* Allocate/initialize root's parent. */
122 	if ((parent = fts_alloc(sp, "", 0)) == NULL)
123 		goto mem2;
124 	parent->fts_level = FTS_ROOTPARENTLEVEL;
125 
126 	/* Allocate/initialize root(s). */
127 	for (root = prev = NULL, nitems = 0; *argv; ++argv, ++nitems) {
128 		if ((p = fts_alloc(sp, *argv, strlen(*argv))) == NULL)
129 			goto mem3;
130 		p->fts_level = FTS_ROOTLEVEL;
131 		p->fts_parent = parent;
132 		p->fts_accpath = p->fts_name;
133 		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW), -1);
134 
135 		/* Command-line "." and ".." are real directories. */
136 		if (p->fts_info == FTS_DOT)
137 			p->fts_info = FTS_D;
138 
139 		/*
140 		 * If comparison routine supplied, traverse in sorted
141 		 * order; otherwise traverse in the order specified.
142 		 */
143 		if (compar) {
144 			p->fts_link = root;
145 			root = p;
146 		} else {
147 			p->fts_link = NULL;
148 			if (root == NULL)
149 				root = p;
150 			else
151 				prev->fts_link = p;
152 			prev = p;
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) &&
176 	    (sp->fts_rfd = open(".", O_RDONLY | O_CLOEXEC)) < 0)
177 		SET(FTS_NOCHDIR);
178 
179 	if (nitems == 0)
180 		free(parent);
181 
182 	return (sp);
183 
184 mem3:	fts_lfree(root);
185 	free(parent);
186 mem2:	free(sp->fts_path);
187 mem1:	free(sp);
188 	return (NULL);
189 }
190 
191 static void
192 fts_load(FTS *sp, FTSENT *p)
193 {
194 	size_t len;
195 	char *cp;
196 
197 	/*
198 	 * Load the stream structure for the next traversal.  Since we don't
199 	 * actually enter the directory until after the preorder visit, set
200 	 * the fts_accpath field specially so the chdir gets done to the right
201 	 * place and the user can access the first node.  From fts_open it's
202 	 * known that the path will fit.
203 	 */
204 	len = p->fts_pathlen = p->fts_namelen;
205 	(void) memmove(sp->fts_path, p->fts_name, len + 1);
206 	if (((cp = strrchr(p->fts_name, '/')) != NULL) &&
207 	    (cp != p->fts_name || cp[1])) {
208 		len = strlen(++cp);
209 		(void) memmove(p->fts_name, cp, len + 1);
210 		p->fts_namelen = len;
211 	}
212 	p->fts_accpath = p->fts_path = sp->fts_path;
213 	sp->fts_dev = p->fts_dev;
214 }
215 
216 int
217 fts_close(FTS *sp)
218 {
219 	FTSENT *freep, *p;
220 	int rfd, error = 0;
221 
222 	/*
223 	 * This still works if we haven't read anything -- the dummy structure
224 	 * points to the root list, so we step through to the end of the root
225 	 * list which has a valid parent pointer.
226 	 */
227 	if (sp->fts_cur) {
228 		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL; ) {
229 			freep = p;
230 			p = p->fts_link ? p->fts_link : p->fts_parent;
231 			free(freep);
232 		}
233 		free(p);
234 	}
235 
236 	/* Stash the original directory fd if needed. */
237 	rfd = ISSET(FTS_NOCHDIR) ? -1 : sp->fts_rfd;
238 
239 	/* Free up child linked list, sort array, path buffer, stream ptr. */
240 	if (sp->fts_child)
241 		fts_lfree(sp->fts_child);
242 	free(sp->fts_array);
243 	free(sp->fts_path);
244 	free(sp);
245 
246 	/* Return to original directory, checking for error. */
247 	if (rfd != -1) {
248 		int saved_errno;
249 		error = fchdir(rfd);
250 		saved_errno = errno;
251 		(void) close(rfd);
252 		errno = saved_errno;
253 	}
254 
255 	return (error);
256 }
257 
258 /*
259  * Special case of "/" at the end of the path so that slashes aren't
260  * appended which would cause paths to be written as "....//foo".
261  */
262 #define	NAPPEND(p)							\
263 	(p->fts_path[p->fts_pathlen - 1] == '/'				\
264 	    ? p->fts_pathlen - 1 : p->fts_pathlen)
265 
266 FTSENT *
267 fts_read(FTS *sp)
268 {
269 	FTSENT *p, *tmp;
270 	int instr;
271 	char *t;
272 	int saved_errno;
273 
274 	/* If finished or unrecoverable error, return NULL. */
275 	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
276 		return (NULL);
277 
278 	/* Set current node pointer. */
279 	p = sp->fts_cur;
280 
281 	/* Save and zero out user instructions. */
282 	instr = p->fts_instr;
283 	p->fts_instr = FTS_NOINSTR;
284 
285 	/* Any type of file may be re-visited; re-stat and re-turn. */
286 	if (instr == FTS_AGAIN) {
287 		p->fts_info = fts_stat(sp, p, 0, -1);
288 		return (p);
289 	}
290 
291 	/*
292 	 * Following a symlink -- SLNONE test allows application to see
293 	 * SLNONE and recover.  If indirecting through a symlink, have
294 	 * keep a pointer to current location.  If unable to get that
295 	 * pointer, follow fails.
296 	 */
297 	if (instr == FTS_FOLLOW &&
298 	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
299 		p->fts_info = fts_stat(sp, p, 1, -1);
300 		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
301 			if ((p->fts_symfd =
302 			    open(".", O_RDONLY | O_CLOEXEC)) < 0) {
303 				p->fts_errno = errno;
304 				p->fts_info = FTS_ERR;
305 			} else
306 				p->fts_flags |= FTS_SYMFOLLOW;
307 		}
308 		return (p);
309 	}
310 
311 	/* Directory in pre-order. */
312 	if (p->fts_info == FTS_D) {
313 		/* If skipped or crossed mount point, do post-order visit. */
314 		if (instr == FTS_SKIP ||
315 		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
316 			if (p->fts_flags & FTS_SYMFOLLOW)
317 				(void) close(p->fts_symfd);
318 			if (sp->fts_child) {
319 				fts_lfree(sp->fts_child);
320 				sp->fts_child = NULL;
321 			}
322 			p->fts_info = FTS_DP;
323 			return (p);
324 		}
325 
326 		/* Rebuild if only read the names and now traversing. */
327 		if (sp->fts_child && ISSET(FTS_NAMEONLY)) {
328 			CLR(FTS_NAMEONLY);
329 			fts_lfree(sp->fts_child);
330 			sp->fts_child = NULL;
331 		}
332 
333 		/*
334 		 * Cd to the subdirectory.
335 		 *
336 		 * If have already read and now fail to chdir, whack the list
337 		 * to make the names come out right, and set the parent errno
338 		 * so the application will eventually get an error condition.
339 		 * Set the FTS_DONTCHDIR flag so that when we logically change
340 		 * directories back to the parent we don't do a chdir.
341 		 *
342 		 * If haven't read do so.  If the read fails, fts_build sets
343 		 * FTS_STOP or the fts_info field of the node.
344 		 */
345 		if (sp->fts_child) {
346 			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
347 				p->fts_errno = errno;
348 				p->fts_flags |= FTS_DONTCHDIR;
349 				for (p = sp->fts_child; p; p = p->fts_link)
350 					p->fts_accpath =
351 					    p->fts_parent->fts_accpath;
352 			}
353 		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
354 			if (ISSET(FTS_STOP))
355 				return (NULL);
356 			return (p);
357 		}
358 		p = sp->fts_child;
359 		sp->fts_child = NULL;
360 		goto name;
361 	}
362 
363 	/* Move to the next node on this level. */
364 next:	tmp = p;
365 	if ((p = p->fts_link) != NULL) {
366 		free(tmp);
367 
368 		/*
369 		 * If reached the top, return to the original directory (or
370 		 * the root of the tree), and load the paths for the next root.
371 		 */
372 		if (p->fts_level == FTS_ROOTLEVEL) {
373 			if (FCHDIR(sp, sp->fts_rfd)) {
374 				SET(FTS_STOP);
375 				return (NULL);
376 			}
377 			fts_load(sp, p);
378 			return (sp->fts_cur = p);
379 		}
380 
381 		/*
382 		 * User may have called fts_set on the node.  If skipped,
383 		 * ignore.  If followed, get a file descriptor so we can
384 		 * get back if necessary.
385 		 */
386 		if (p->fts_instr == FTS_SKIP)
387 			goto next;
388 		if (p->fts_instr == FTS_FOLLOW) {
389 			p->fts_info = fts_stat(sp, p, 1, -1);
390 			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
391 				if ((p->fts_symfd =
392 				    open(".", O_RDONLY | O_CLOEXEC)) < 0) {
393 					p->fts_errno = errno;
394 					p->fts_info = FTS_ERR;
395 				} else
396 					p->fts_flags |= FTS_SYMFOLLOW;
397 			}
398 			p->fts_instr = FTS_NOINSTR;
399 		}
400 
401 name:		t = sp->fts_path + NAPPEND(p->fts_parent);
402 		*t++ = '/';
403 		(void) memmove(t, p->fts_name, p->fts_namelen + 1);
404 		return (sp->fts_cur = p);
405 	}
406 
407 	/* Move up to the parent node. */
408 	p = tmp->fts_parent;
409 	free(tmp);
410 
411 	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
412 		/*
413 		 * Done; free everything up and set errno to 0 so the user
414 		 * can distinguish between error and EOF.
415 		 */
416 		free(p);
417 		errno = 0;
418 		return (sp->fts_cur = NULL);
419 	}
420 
421 	/* NUL terminate the pathname. */
422 	sp->fts_path[p->fts_pathlen] = '\0';
423 
424 	/*
425 	 * Return to the parent directory.  If at a root node or came through
426 	 * a symlink, go back through the file descriptor.  Otherwise, cd up
427 	 * one directory.
428 	 */
429 	if (p->fts_level == FTS_ROOTLEVEL) {
430 		if (FCHDIR(sp, sp->fts_rfd)) {
431 			SET(FTS_STOP);
432 			sp->fts_cur = p;
433 			return (NULL);
434 		}
435 	} else if (p->fts_flags & FTS_SYMFOLLOW) {
436 		if (FCHDIR(sp, p->fts_symfd)) {
437 			saved_errno = errno;
438 			(void) close(p->fts_symfd);
439 			errno = saved_errno;
440 			SET(FTS_STOP);
441 			sp->fts_cur = p;
442 			return (NULL);
443 		}
444 		(void) close(p->fts_symfd);
445 	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
446 	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
447 		SET(FTS_STOP);
448 		sp->fts_cur = p;
449 		return (NULL);
450 	}
451 	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
452 	return (sp->fts_cur = p);
453 }
454 
455 /*
456  * Fts_set takes the stream as an argument although it's not used in this
457  * implementation; it would be necessary if anyone wanted to add global
458  * semantics to fts using fts_set.  An error return is allowed for similar
459  * reasons.
460  */
461 int
462 fts_set(FTS *sp __unused, FTSENT *p, int instr)
463 {
464 	if (instr && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
465 	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
466 		errno = EINVAL;
467 		return (1);
468 	}
469 	p->fts_instr = instr;
470 	return (0);
471 }
472 
473 FTSENT *
474 fts_children(FTS *sp, int instr)
475 {
476 	FTSENT *p;
477 	int fd;
478 
479 	if (instr && instr != FTS_NAMEONLY) {
480 		errno = EINVAL;
481 		return (NULL);
482 	}
483 
484 	/* Set current node pointer. */
485 	p = sp->fts_cur;
486 
487 	/*
488 	 * Errno set to 0 so user can distinguish empty directory from
489 	 * an error.
490 	 */
491 	errno = 0;
492 
493 	/* Fatal errors stop here. */
494 	if (ISSET(FTS_STOP))
495 		return (NULL);
496 
497 	/* Return logical hierarchy of user's arguments. */
498 	if (p->fts_info == FTS_INIT)
499 		return (p->fts_link);
500 
501 	/*
502 	 * If not a directory being visited in pre-order, stop here.  Could
503 	 * allow FTS_DNR, assuming the user has fixed the problem, but the
504 	 * same effect is available with FTS_AGAIN.
505 	 */
506 	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
507 		return (NULL);
508 
509 	/* Free up any previous child list. */
510 	if (sp->fts_child)
511 		fts_lfree(sp->fts_child);
512 
513 	if (instr == FTS_NAMEONLY) {
514 		SET(FTS_NAMEONLY);
515 		instr = BNAMES;
516 	} else
517 		instr = BCHILD;
518 
519 	/*
520 	 * If using chdir on a relative path and called BEFORE fts_read does
521 	 * its chdir to the root of a traversal, we can lose -- we need to
522 	 * chdir into the subdirectory, and we don't know where the current
523 	 * directory is, so we can't get back so that the upcoming chdir by
524 	 * fts_read will work.
525 	 */
526 	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
527 	    ISSET(FTS_NOCHDIR))
528 		return (sp->fts_child = fts_build(sp, instr));
529 
530 	if ((fd = open(".", O_RDONLY | O_CLOEXEC)) < 0)
531 		return (NULL);
532 	sp->fts_child = fts_build(sp, instr);
533 	if (fchdir(fd)) {
534 		(void) close(fd);
535 		return (NULL);
536 	}
537 	(void) close(fd);
538 	return (sp->fts_child);
539 }
540 
541 /*
542  * This is the tricky part -- do not casually change *anything* in here.  The
543  * idea is to build the linked list of entries that are used by fts_children
544  * and fts_read.  There are lots of special cases.
545  *
546  * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
547  * set and it's a physical walk (so that symbolic links can't be directories),
548  * we can do things quickly.  First, if it's a 4.4BSD file system, the type
549  * of the file is in the directory entry.  Otherwise, we assume that the number
550  * of subdirectories in a node is equal to the number of links to the parent.
551  * The former skips all stat calls.  The latter skips stat calls in any leaf
552  * directories and for any files after the subdirectories in the directory have
553  * been found, cutting the stat calls by about 2/3.
554  */
555 static FTSENT *
556 fts_build(FTS *sp, int type)
557 {
558 	struct dirent *dp;
559 	FTSENT *p, *head;
560 	FTSENT *cur, *tail;
561 	DIR *dirp;
562 	void *oldaddr;
563 	size_t len, maxlen;
564 	int nitems, cderrno, descend, level, nlinks, doadjust;
565 	int saved_errno;
566 	char *cp = NULL;
567 
568 	/* Set current node pointer. */
569 	cur = sp->fts_cur;
570 
571 	/*
572 	 * Open the directory for reading.  If this fails, we're done.
573 	 * If being called from fts_read, set the fts_info field.
574 	 */
575 	if ((dirp = opendir(cur->fts_accpath)) == NULL) {
576 		if (type == BREAD) {
577 			cur->fts_info = FTS_DNR;
578 			cur->fts_errno = errno;
579 		}
580 		return (NULL);
581 	}
582 
583 	/*
584 	 * Nlinks is the number of possible entries of type directory in the
585 	 * directory if we're cheating on stat calls, 0 if we're not doing
586 	 * any stat calls at all, -1 if we're doing stats on everything.
587 	 */
588 	if (type == BNAMES)
589 		nlinks = 0;
590 	else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL))
591 		nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
592 	else
593 		nlinks = -1;
594 
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, strlen(dp->d_name))))
658 			goto mem1;
659 		if (strlen(dp->d_name) >= maxlen) { /* include space for NUL */
660 			oldaddr = sp->fts_path;
661 			if (fts_palloc(sp, strlen(dp->d_name) +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 + strlen(dp->d_name);
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 			p->fts_accpath =
712 			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
713 			p->fts_info = FTS_NSOK;
714 		} else {
715 			/* Build a file name for fts_stat to stat. */
716 			if (ISSET(FTS_NOCHDIR)) {
717 				p->fts_accpath = p->fts_path;
718 				(void) memmove(cp, p->fts_name,
719 				    p->fts_namelen + 1);
720 				p->fts_info = fts_stat(sp, p, 0, dirfd(dirp));
721 			} else {
722 				p->fts_accpath = p->fts_name;
723 				p->fts_info = fts_stat(sp, p, 0, -1);
724 			}
725 
726 			/* Decrement link count if applicable. */
727 			if (nlinks > 0 && (p->fts_info == FTS_D ||
728 			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
729 				--nlinks;
730 		}
731 
732 		/* We walk in directory order so "ls -f" doesn't get upset. */
733 		p->fts_link = NULL;
734 		if (head == NULL)
735 			head = tail = p;
736 		else {
737 			tail->fts_link = p;
738 			tail = p;
739 		}
740 		++nitems;
741 	}
742 	if (dirp)
743 		(void) closedir(dirp);
744 
745 	/*
746 	 * If realloc() changed the address of the path, adjust the
747 	 * addresses for the rest of the tree and the dir list.
748 	 */
749 	if (doadjust)
750 		fts_padjust(sp, head);
751 
752 	/*
753 	 * If not changing directories, reset the path back to original
754 	 * state.
755 	 */
756 	if (ISSET(FTS_NOCHDIR)) {
757 		if (len == sp->fts_pathlen || nitems == 0)
758 			--cp;
759 		*cp = '\0';
760 	}
761 
762 	/*
763 	 * If descended after called from fts_children or after called from
764 	 * fts_read and nothing found, get back.  At the root level we use
765 	 * the saved fd; if one of fts_open()'s arguments is a relative path
766 	 * to an empty directory, we wind up here with no other way back.  If
767 	 * can't get back, we're done.
768 	 */
769 	if (descend && (type == BCHILD || !nitems) &&
770 	    (cur->fts_level == FTS_ROOTLEVEL ? FCHDIR(sp, sp->fts_rfd) :
771 	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
772 		cur->fts_info = FTS_ERR;
773 		SET(FTS_STOP);
774 		return (NULL);
775 	}
776 
777 	/* If didn't find anything, return NULL. */
778 	if (!nitems) {
779 		if (type == BREAD)
780 			cur->fts_info = FTS_DP;
781 		return (NULL);
782 	}
783 
784 	/* Sort the entries. */
785 	if (sp->fts_compar && nitems > 1)
786 		head = fts_sort(sp, head, nitems);
787 	return (head);
788 }
789 
790 static ushort_t
791 fts_stat(FTS *sp, FTSENT *p, int follow, int dfd)
792 {
793 	FTSENT *t;
794 	dev_t dev;
795 	ino_t ino;
796 	struct stat *sbp, sb;
797 	int saved_errno;
798 	const char *path;
799 
800 	if (dfd == -1) {
801 		path = p->fts_accpath;
802 		dfd = AT_FDCWD;
803 	} else
804 		path = p->fts_name;
805 
806 	/* If user needs stat info, stat buffer already allocated. */
807 	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
808 
809 	/*
810 	 * If doing a logical walk, or application requested FTS_FOLLOW, do
811 	 * a stat(2).  If that fails, check for a non-existent symlink.  If
812 	 * fail, set the errno from the stat call.
813 	 */
814 	if (ISSET(FTS_LOGICAL) || follow) {
815 		if (fstatat(dfd, path, sbp, 0)) {
816 			saved_errno = errno;
817 			if (!fstatat(dfd, path, sbp, AT_SYMLINK_NOFOLLOW)) {
818 				errno = 0;
819 				return (FTS_SLNONE);
820 			}
821 			p->fts_errno = saved_errno;
822 			goto err;
823 		}
824 	} else if (fstatat(dfd, path, sbp, AT_SYMLINK_NOFOLLOW)) {
825 		p->fts_errno = errno;
826 err:		(void) memset(sbp, 0, sizeof (struct stat));
827 		return (FTS_NS);
828 	}
829 
830 	if (S_ISDIR(sbp->st_mode)) {
831 		/*
832 		 * Set the device/inode.  Used to find cycles and check for
833 		 * crossing mount points.  Also remember the link count, used
834 		 * in fts_build to limit the number of stat calls.  It is
835 		 * understood that these fields are only referenced if fts_info
836 		 * is set to FTS_D.
837 		 */
838 		dev = p->fts_dev = sbp->st_dev;
839 		ino = p->fts_ino = sbp->st_ino;
840 		p->fts_nlink = sbp->st_nlink;
841 
842 		if (ISDOT(p->fts_name))
843 			return (FTS_DOT);
844 
845 		/*
846 		 * Cycle detection is done by brute force when the directory
847 		 * is first encountered.  If the tree gets deep enough or the
848 		 * number of symbolic links to directories is high enough,
849 		 * something faster might be worthwhile.
850 		 */
851 		for (t = p->fts_parent;
852 		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
853 			if (ino == t->fts_ino && dev == t->fts_dev) {
854 				p->fts_cycle = t;
855 				return (FTS_DC);
856 			}
857 		return (FTS_D);
858 	}
859 	if (S_ISLNK(sbp->st_mode))
860 		return (FTS_SL);
861 	if (S_ISREG(sbp->st_mode))
862 		return (FTS_F);
863 	return (FTS_DEFAULT);
864 }
865 
866 static FTSENT *
867 fts_sort(FTS *sp, FTSENT *head, int nitems)
868 {
869 	FTSENT **ap, *p;
870 
871 	/*
872 	 * Construct an array of pointers to the structures and call qsort(3).
873 	 * Reassemble the array in the order returned by qsort.  If unable to
874 	 * sort for memory reasons, return the directory entries in their
875 	 * current order.  Allocate enough space for the current needs plus
876 	 * 40 so don't realloc one entry at a time.
877 	 */
878 	if (nitems > sp->fts_nitems) {
879 		struct _ftsent **a;
880 
881 		sp->fts_nitems = nitems + 40;
882 		if ((a = reallocarray(sp->fts_array,
883 		    sp->fts_nitems, sizeof (FTSENT *))) == NULL) {
884 			free(sp->fts_array);
885 			sp->fts_array = NULL;
886 			sp->fts_nitems = 0;
887 			return (head);
888 		}
889 		sp->fts_array = a;
890 	}
891 	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
892 		*ap++ = p;
893 	qsort(sp->fts_array, nitems, sizeof (FTSENT *), sp->fts_compar);
894 	for (head = *(ap = sp->fts_array); --nitems; ++ap)
895 		ap[0]->fts_link = ap[1];
896 	ap[0]->fts_link = NULL;
897 	return (head);
898 }
899 
900 static FTSENT *
901 fts_alloc(FTS *sp, char *name, size_t namelen)
902 {
903 	FTSENT *p;
904 	size_t len;
905 
906 	/*
907 	 * The file name is a variable length array and no stat structure is
908 	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
909 	 * structure, the file name and the stat structure in one chunk, but
910 	 * be careful that the stat structure is reasonably aligned.  Since the
911 	 * fts_name field is declared to be of size 1, the fts_name pointer is
912 	 * namelen + 2 before the first possible address of the stat structure.
913 	 */
914 	len = sizeof (FTSENT) + namelen;
915 	if (!ISSET(FTS_NOSTAT))
916 		len += sizeof (struct stat) + ALIGNBYTES;
917 	if ((p = calloc(1, len)) == NULL)
918 		return (NULL);
919 
920 	p->fts_path = sp->fts_path;
921 	p->fts_namelen = namelen;
922 	p->fts_instr = FTS_NOINSTR;
923 	if (!ISSET(FTS_NOSTAT))
924 		p->fts_statp = (struct stat *)ALIGN(p->fts_name + namelen + 2);
925 	(void) memcpy(p->fts_name, name, namelen);
926 
927 	return (p);
928 }
929 
930 static void
931 fts_lfree(FTSENT *head)
932 {
933 	FTSENT *p;
934 
935 	/* Free a linked list of structures. */
936 	while ((p = head) != NULL) {
937 		head = head->fts_link;
938 		free(p);
939 	}
940 }
941 
942 /*
943  * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
944  * Most systems will allow creation of paths much longer than PATH_MAX, even
945  * though the kernel won't resolve them.  Add the size (not just what's needed)
946  * plus 256 bytes so don't realloc the path 2 bytes at a time.
947  */
948 static int
949 fts_palloc(FTS *sp, size_t more)
950 {
951 	char *p;
952 
953 	/*
954 	 * Check for possible wraparound.
955 	 */
956 	more += 256;
957 	if (sp->fts_pathlen + more < sp->fts_pathlen) {
958 		free(sp->fts_path);
959 		sp->fts_path = NULL;
960 		errno = ENAMETOOLONG;
961 		return (1);
962 	}
963 	sp->fts_pathlen += more;
964 	p = realloc(sp->fts_path, sp->fts_pathlen);
965 	if (p == NULL) {
966 		free(sp->fts_path);
967 		sp->fts_path = NULL;
968 		return (1);
969 	}
970 	sp->fts_path = p;
971 	return (0);
972 }
973 
974 /*
975  * When the path is realloc'd, have to fix all of the pointers in structures
976  * already returned.
977  */
978 static void
979 fts_padjust(FTS *sp, FTSENT *head)
980 {
981 	FTSENT *p;
982 	char *addr = sp->fts_path;
983 
984 #define	ADJUST(p) {							\
985 	if ((p)->fts_accpath != (p)->fts_name) {			\
986 		(p)->fts_accpath =					\
987 		    (char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
988 	}								\
989 	(p)->fts_path = addr;						\
990 }
991 	/* Adjust the current set of children. */
992 	for (p = sp->fts_child; p; p = p->fts_link)
993 		ADJUST(p);
994 
995 	/* Adjust the rest of the tree, including the current level. */
996 	for (p = head; p->fts_level >= FTS_ROOTLEVEL; ) {
997 		ADJUST(p);
998 		p = p->fts_link ? p->fts_link : p->fts_parent;
999 	}
1000 }
1001 
1002 static size_t
1003 fts_maxarglen(char * const *argv)
1004 {
1005 	size_t len, max;
1006 
1007 	for (max = 0; *argv; ++argv)
1008 		if ((len = strlen(*argv)) > max)
1009 			max = len;
1010 	return (max + 1);
1011 }
1012 
1013 /*
1014  * Change to dir specified by fd or p->fts_accpath without getting
1015  * tricked by someone changing the world out from underneath us.
1016  * Assumes p->fts_dev and p->fts_ino are filled in.
1017  */
1018 static int
1019 fts_safe_changedir(FTS *sp, FTSENT *p, int fd, char *path)
1020 {
1021 	int ret, oerrno, newfd;
1022 	struct stat sb;
1023 
1024 	newfd = fd;
1025 	if (ISSET(FTS_NOCHDIR))
1026 		return (0);
1027 	/*
1028 	 * This originally specified O_DIRECTORY as well - restore once/if we
1029 	 * implement the flag.
1030 	 */
1031 	if (fd < 0 && (newfd = open(path, O_RDONLY|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