xref: /haiku/src/libs/bsd/fts.c (revision e1c4049fed1047bdb957b0529e1921e97ef94770)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
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  * $OpenBSD: fts.c,v 1.22 1999/10/03 19:22:22 millert Exp $
32  */
33 
34 #include <sys/cdefs.h>
35 
36 #ifdef __HAIKU__
37 #include <sys/param.h>
38 #include <sys/stat.h>
39 
40 #include <dirent.h>
41 #include <errno.h>
42 #include <fcntl.h>
43 #include <fts.h>
44 #include <stdlib.h>
45 #include <string.h>
46 #include <unistd.h>
47 #else
48 __SCCSID("@(#)fts.c	8.6 (Berkeley) 8/14/94");
49 __FBSDID("$FreeBSD$");
50 
51 #include "namespace.h"
52 #include <sys/param.h>
53 #include <sys/mount.h>
54 #include <sys/stat.h>
55 
56 #include <dirent.h>
57 #include <errno.h>
58 #include <fcntl.h>
59 #include <fts.h>
60 #include <stdlib.h>
61 #include <string.h>
62 #include <unistd.h>
63 #include "un-namespace.h"
64 
65 #include "gen-private.h"
66 #endif
67 
68 static FTSENT	*fts_alloc(FTS *, char *, size_t);
69 static FTSENT	*fts_build(FTS *, int);
70 static void	 fts_lfree(FTSENT *);
71 static void	 fts_load(FTS *, FTSENT *);
72 static size_t	 fts_maxarglen(char * const *);
73 static void	 fts_padjust(FTS *, FTSENT *);
74 static int	 fts_palloc(FTS *, size_t);
75 static FTSENT	*fts_sort(FTS *, FTSENT *, size_t);
76 static int	 fts_stat(FTS *, FTSENT *, int, int);
77 static int	 fts_safe_changedir(FTS *, FTSENT *, int, char *);
78 static int	 fts_ufslinks(FTS *, const FTSENT *);
79 
80 #define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
81 
82 #define	CLR(opt)	(sp->fts_options &= ~(opt))
83 #define	ISSET(opt)	(sp->fts_options & (opt))
84 #define	SET(opt)	(sp->fts_options |= (opt))
85 
86 #define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
87 
88 /* fts_build flags */
89 #define	BCHILD		1		/* fts_children */
90 #define	BNAMES		2		/* fts_children, names only */
91 #define	BREAD		3		/* fts_read */
92 
93 /*
94  * Internal representation of an FTS, including extra implementation
95  * details.  The FTS returned from fts_open points to this structure's
96  * ftsp_fts member (and can be cast to an _fts_private as required)
97  */
98 struct _fts_private {
99 	FTS		ftsp_fts;
100 #ifndef __HAIKU__
101 	struct statfs	ftsp_statfs;
102 #endif
103 	dev_t		ftsp_dev;
104 	int		ftsp_linksreliable;
105 };
106 
107 #ifndef __HAIKU__
108 /*
109  * The "FTS_NOSTAT" option can avoid a lot of calls to stat(2) if it
110  * knows that a directory could not possibly have subdirectories.  This
111  * is decided by looking at the link count: a subdirectory would
112  * increment its parent's link count by virtue of its own ".." entry.
113  * This assumption only holds for UFS-like filesystems that implement
114  * links and directories this way, so we must punt for others.
115  */
116 
117 static const char *ufslike_filesystems[] = {
118 	"ufs",
119 	"zfs",
120 	"nfs",
121 	"ext2fs",
122 	0
123 };
124 #endif /* !__HAIKU__ */
125 
126 #ifdef __HAIKU__
127 static void *
128 reallocf(void *ptr, size_t size)
129 {
130 	void *nptr;
131 
132 	nptr = realloc(ptr, size);
133 
134 	if (!nptr)
135 		free(ptr);
136 	return (nptr);
137 }
138 #endif
139 
140 FTS *
141 fts_open(char * const *argv, int options,
142 	int (*compar)(const FTSENT * const *, const FTSENT * const *))
143 {
144 	struct _fts_private *priv;
145 	FTS *sp;
146 	FTSENT *p, *root;
147 	FTSENT *parent, *tmp;
148 	size_t len, nitems;
149 
150 	/* Options check. */
151 	if (options & ~FTS_OPTIONMASK) {
152 		errno = EINVAL;
153 		return (NULL);
154 	}
155 
156 	/* fts_open() requires at least one path */
157 	if (*argv == NULL) {
158 		errno = EINVAL;
159 		return (NULL);
160 	}
161 
162 	/* Allocate/initialize the stream. */
163 	if ((priv = calloc(1, sizeof(*priv))) == NULL)
164 		return (NULL);
165 	sp = &priv->ftsp_fts;
166 	sp->fts_compar = compar;
167 	sp->fts_options = options;
168 
169 	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
170 	if (ISSET(FTS_LOGICAL))
171 		SET(FTS_NOCHDIR);
172 
173 	/*
174 	 * Start out with 1K of path space, and enough, in any case,
175 	 * to hold the user's paths.
176 	 */
177 	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
178 		goto mem1;
179 
180 	/* Allocate/initialize root's parent. */
181 	if ((parent = fts_alloc(sp, "", 0)) == NULL)
182 		goto mem2;
183 	parent->fts_level = FTS_ROOTPARENTLEVEL;
184 
185 	/* Shush, GCC. */
186 	tmp = NULL;
187 
188 	/* Allocate/initialize root(s). */
189 	for (root = NULL, nitems = 0; *argv != NULL; ++argv, ++nitems) {
190 		len = strlen(*argv);
191 
192 		p = fts_alloc(sp, *argv, len);
193 		p->fts_level = FTS_ROOTLEVEL;
194 		p->fts_parent = parent;
195 		p->fts_accpath = p->fts_name;
196 		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW), -1);
197 
198 		/* Command-line "." and ".." are real directories. */
199 		if (p->fts_info == FTS_DOT)
200 			p->fts_info = FTS_D;
201 
202 		/*
203 		 * If comparison routine supplied, traverse in sorted
204 		 * order; otherwise traverse in the order specified.
205 		 */
206 		if (compar) {
207 			p->fts_link = root;
208 			root = p;
209 		} else {
210 			p->fts_link = NULL;
211 			if (root == NULL)
212 				tmp = root = p;
213 			else {
214 				tmp->fts_link = p;
215 				tmp = p;
216 			}
217 		}
218 	}
219 	if (compar && nitems > 1)
220 		root = fts_sort(sp, root, nitems);
221 
222 	/*
223 	 * Allocate a dummy pointer and make fts_read think that we've just
224 	 * finished the node before the root(s); set p->fts_info to FTS_INIT
225 	 * so that everything about the "current" node is ignored.
226 	 */
227 	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
228 		goto mem3;
229 	sp->fts_cur->fts_link = root;
230 	sp->fts_cur->fts_info = FTS_INIT;
231 
232 	/*
233 	 * If using chdir(2), grab a file descriptor pointing to dot to ensure
234 	 * that we can get back here; this could be avoided for some paths,
235 	 * but almost certainly not worth the effort.  Slashes, symbolic links,
236 	 * and ".." are all fairly nasty problems.  Note, if we can't get the
237 	 * descriptor we run anyway, just more slowly.
238 	 */
239 	if (!ISSET(FTS_NOCHDIR) &&
240 		(sp->fts_rfd = open(".", O_RDONLY | O_CLOEXEC, 0)) < 0)
241 		SET(FTS_NOCHDIR);
242 
243 	return (sp);
244 
245 mem3:	fts_lfree(root);
246 	free(parent);
247 mem2:	free(sp->fts_path);
248 mem1:	free(sp);
249 	return (NULL);
250 }
251 
252 static void
253 fts_load(FTS *sp, FTSENT *p)
254 {
255 	size_t len;
256 	char *cp;
257 
258 	/*
259 	 * Load the stream structure for the next traversal.  Since we don't
260 	 * actually enter the directory until after the preorder visit, set
261 	 * the fts_accpath field specially so the chdir gets done to the right
262 	 * place and the user can access the first node.  From fts_open it's
263 	 * known that the path will fit.
264 	 */
265 	len = p->fts_pathlen = p->fts_namelen;
266 	memmove(sp->fts_path, p->fts_name, len + 1);
267 	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
268 		len = strlen(++cp);
269 		memmove(p->fts_name, cp, len + 1);
270 		p->fts_namelen = len;
271 	}
272 	p->fts_accpath = p->fts_path = sp->fts_path;
273 	sp->fts_dev = p->fts_dev;
274 }
275 
276 int
277 fts_close(FTS *sp)
278 {
279 	FTSENT *freep, *p;
280 	int saved_errno;
281 
282 	/*
283 	 * This still works if we haven't read anything -- the dummy structure
284 	 * points to the root list, so we step through to the end of the root
285 	 * list which has a valid parent pointer.
286 	 */
287 	if (sp->fts_cur) {
288 		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
289 			freep = p;
290 			p = p->fts_link != NULL ? p->fts_link : p->fts_parent;
291 			free(freep);
292 		}
293 		free(p);
294 	}
295 
296 	/* Free up child linked list, sort array, path buffer. */
297 	if (sp->fts_child)
298 		fts_lfree(sp->fts_child);
299 	if (sp->fts_array)
300 		free(sp->fts_array);
301 	free(sp->fts_path);
302 
303 	/* Return to original directory, save errno if necessary. */
304 	if (!ISSET(FTS_NOCHDIR)) {
305 		saved_errno = fchdir(sp->fts_rfd) ? errno : 0;
306 		(void)close(sp->fts_rfd);
307 
308 		/* Set errno and return. */
309 		if (saved_errno != 0) {
310 			/* Free up the stream pointer. */
311 			free(sp);
312 			errno = saved_errno;
313 			return (-1);
314 		}
315 	}
316 
317 	/* Free up the stream pointer. */
318 	free(sp);
319 	return (0);
320 }
321 
322 /*
323  * Special case of "/" at the end of the path so that slashes aren't
324  * appended which would cause paths to be written as "....//foo".
325  */
326 #define	NAPPEND(p)							\
327 	(p->fts_path[p->fts_pathlen - 1] == '/'				\
328 		? p->fts_pathlen - 1 : p->fts_pathlen)
329 
330 FTSENT *
331 fts_read(FTS *sp)
332 {
333 	FTSENT *p, *tmp;
334 	int instr;
335 	char *t;
336 	int saved_errno;
337 
338 	/* If finished or unrecoverable error, return NULL. */
339 	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
340 		return (NULL);
341 
342 	/* Set current node pointer. */
343 	p = sp->fts_cur;
344 
345 	/* Save and zero out user instructions. */
346 	instr = p->fts_instr;
347 	p->fts_instr = FTS_NOINSTR;
348 
349 	/* Any type of file may be re-visited; re-stat and re-turn. */
350 	if (instr == FTS_AGAIN) {
351 		p->fts_info = fts_stat(sp, p, 0, -1);
352 		return (p);
353 	}
354 
355 	/*
356 	 * Following a symlink -- SLNONE test allows application to see
357 	 * SLNONE and recover.  If indirecting through a symlink, have
358 	 * keep a pointer to current location.  If unable to get that
359 	 * pointer, follow fails.
360 	 */
361 	if (instr == FTS_FOLLOW &&
362 		(p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
363 		p->fts_info = fts_stat(sp, p, 1, -1);
364 		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
365 			if ((p->fts_symfd = open(".", O_RDONLY | O_CLOEXEC,
366 				0)) < 0) {
367 				p->fts_errno = errno;
368 				p->fts_info = FTS_ERR;
369 			} else
370 				p->fts_flags |= FTS_SYMFOLLOW;
371 		}
372 		return (p);
373 	}
374 
375 	/* Directory in pre-order. */
376 	if (p->fts_info == FTS_D) {
377 		/* If skipped or crossed mount point, do post-order visit. */
378 		if (instr == FTS_SKIP ||
379 			(ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
380 			if (p->fts_flags & FTS_SYMFOLLOW)
381 				(void)close(p->fts_symfd);
382 			if (sp->fts_child) {
383 				fts_lfree(sp->fts_child);
384 				sp->fts_child = NULL;
385 			}
386 			p->fts_info = FTS_DP;
387 			return (p);
388 		}
389 
390 		/* Rebuild if only read the names and now traversing. */
391 		if (sp->fts_child != NULL && ISSET(FTS_NAMEONLY)) {
392 			CLR(FTS_NAMEONLY);
393 			fts_lfree(sp->fts_child);
394 			sp->fts_child = NULL;
395 		}
396 
397 		/*
398 		 * Cd to the subdirectory.
399 		 *
400 		 * If have already read and now fail to chdir, whack the list
401 		 * to make the names come out right, and set the parent errno
402 		 * so the application will eventually get an error condition.
403 		 * Set the FTS_DONTCHDIR flag so that when we logically change
404 		 * directories back to the parent we don't do a chdir.
405 		 *
406 		 * If haven't read do so.  If the read fails, fts_build sets
407 		 * FTS_STOP or the fts_info field of the node.
408 		 */
409 		if (sp->fts_child != NULL) {
410 			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
411 				p->fts_errno = errno;
412 				p->fts_flags |= FTS_DONTCHDIR;
413 				for (p = sp->fts_child; p != NULL;
414 					p = p->fts_link)
415 					p->fts_accpath =
416 						p->fts_parent->fts_accpath;
417 			}
418 		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
419 			if (ISSET(FTS_STOP))
420 				return (NULL);
421 			return (p);
422 		}
423 		p = sp->fts_child;
424 		sp->fts_child = NULL;
425 		goto name;
426 	}
427 
428 	/* Move to the next node on this level. */
429 next:	tmp = p;
430 	if ((p = p->fts_link) != NULL) {
431 		/*
432 		 * If reached the top, return to the original directory (or
433 		 * the root of the tree), and load the paths for the next root.
434 		 */
435 		if (p->fts_level == FTS_ROOTLEVEL) {
436 			if (FCHDIR(sp, sp->fts_rfd)) {
437 				SET(FTS_STOP);
438 				return (NULL);
439 			}
440 			free(tmp);
441 			fts_load(sp, p);
442 			return (sp->fts_cur = p);
443 		}
444 
445 		/*
446 		 * User may have called fts_set on the node.  If skipped,
447 		 * ignore.  If followed, get a file descriptor so we can
448 		 * get back if necessary.
449 		 */
450 		if (p->fts_instr == FTS_SKIP) {
451 			free(tmp);
452 			goto next;
453 		}
454 		if (p->fts_instr == FTS_FOLLOW) {
455 			p->fts_info = fts_stat(sp, p, 1, -1);
456 			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
457 				if ((p->fts_symfd =
458 					open(".", O_RDONLY | O_CLOEXEC, 0)) < 0) {
459 					p->fts_errno = errno;
460 					p->fts_info = FTS_ERR;
461 				} else
462 					p->fts_flags |= FTS_SYMFOLLOW;
463 			}
464 			p->fts_instr = FTS_NOINSTR;
465 		}
466 
467 		free(tmp);
468 
469 name:		t = sp->fts_path + NAPPEND(p->fts_parent);
470 		*t++ = '/';
471 		memmove(t, p->fts_name, p->fts_namelen + 1);
472 		return (sp->fts_cur = p);
473 	}
474 
475 	/* Move up to the parent node. */
476 	p = tmp->fts_parent;
477 
478 	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
479 		/*
480 		 * Done; free everything up and set errno to 0 so the user
481 		 * can distinguish between error and EOF.
482 		 */
483 		free(tmp);
484 		free(p);
485 		errno = 0;
486 		return (sp->fts_cur = NULL);
487 	}
488 
489 	/* NUL terminate the pathname. */
490 	sp->fts_path[p->fts_pathlen] = '\0';
491 
492 	/*
493 	 * Return to the parent directory.  If at a root node or came through
494 	 * a symlink, go back through the file descriptor.  Otherwise, cd up
495 	 * one directory.
496 	 */
497 	if (p->fts_level == FTS_ROOTLEVEL) {
498 		if (FCHDIR(sp, sp->fts_rfd)) {
499 			SET(FTS_STOP);
500 			return (NULL);
501 		}
502 	} else if (p->fts_flags & FTS_SYMFOLLOW) {
503 		if (FCHDIR(sp, p->fts_symfd)) {
504 			saved_errno = errno;
505 			(void)close(p->fts_symfd);
506 			errno = saved_errno;
507 			SET(FTS_STOP);
508 			return (NULL);
509 		}
510 		(void)close(p->fts_symfd);
511 	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
512 		fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
513 		SET(FTS_STOP);
514 		return (NULL);
515 	}
516 	free(tmp);
517 	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
518 	return (sp->fts_cur = p);
519 }
520 
521 /*
522  * Fts_set takes the stream as an argument although it's not used in this
523  * implementation; it would be necessary if anyone wanted to add global
524  * semantics to fts using fts_set.  An error return is allowed for similar
525  * reasons.
526  */
527 /* ARGSUSED */
528 int
529 fts_set(FTS *sp, FTSENT *p, int instr)
530 {
531 	if (instr != 0 && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
532 		instr != FTS_NOINSTR && instr != FTS_SKIP) {
533 		errno = EINVAL;
534 		return (1);
535 	}
536 	p->fts_instr = instr;
537 	return (0);
538 }
539 
540 FTSENT *
541 fts_children(FTS *sp, int instr)
542 {
543 	FTSENT *p;
544 	int fd, rc, serrno;
545 
546 	if (instr != 0 && instr != FTS_NAMEONLY) {
547 		errno = EINVAL;
548 		return (NULL);
549 	}
550 
551 	/* Set current node pointer. */
552 	p = sp->fts_cur;
553 
554 	/*
555 	 * Errno set to 0 so user can distinguish empty directory from
556 	 * an error.
557 	 */
558 	errno = 0;
559 
560 	/* Fatal errors stop here. */
561 	if (ISSET(FTS_STOP))
562 		return (NULL);
563 
564 	/* Return logical hierarchy of user's arguments. */
565 	if (p->fts_info == FTS_INIT)
566 		return (p->fts_link);
567 
568 	/*
569 	 * If not a directory being visited in pre-order, stop here.  Could
570 	 * allow FTS_DNR, assuming the user has fixed the problem, but the
571 	 * same effect is available with FTS_AGAIN.
572 	 */
573 	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
574 		return (NULL);
575 
576 	/* Free up any previous child list. */
577 	if (sp->fts_child != NULL)
578 		fts_lfree(sp->fts_child);
579 
580 	if (instr == FTS_NAMEONLY) {
581 		SET(FTS_NAMEONLY);
582 		instr = BNAMES;
583 	} else
584 		instr = BCHILD;
585 
586 	/*
587 	 * If using chdir on a relative path and called BEFORE fts_read does
588 	 * its chdir to the root of a traversal, we can lose -- we need to
589 	 * chdir into the subdirectory, and we don't know where the current
590 	 * directory is, so we can't get back so that the upcoming chdir by
591 	 * fts_read will work.
592 	 */
593 	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
594 		ISSET(FTS_NOCHDIR))
595 		return (sp->fts_child = fts_build(sp, instr));
596 
597 	if ((fd = open(".", O_RDONLY | O_CLOEXEC, 0)) < 0)
598 		return (NULL);
599 	sp->fts_child = fts_build(sp, instr);
600 	serrno = (sp->fts_child == NULL) ? errno : 0;
601 	rc = fchdir(fd);
602 	if (rc < 0 && serrno == 0)
603 		serrno = errno;
604 	(void)close(fd);
605 	errno = serrno;
606 	if (rc < 0)
607 		return (NULL);
608 	return (sp->fts_child);
609 }
610 
611 #ifndef fts_get_clientptr
612 #error "fts_get_clientptr not defined"
613 #endif
614 
615 void *
616 (fts_get_clientptr)(FTS *sp)
617 {
618 
619 	return (fts_get_clientptr(sp));
620 }
621 
622 #ifndef fts_get_stream
623 #error "fts_get_stream not defined"
624 #endif
625 
626 FTS *
627 (fts_get_stream)(FTSENT *p)
628 {
629 	return (fts_get_stream(p));
630 }
631 
632 void
633 fts_set_clientptr(FTS *sp, void *clientptr)
634 {
635 
636 	sp->fts_clientptr = clientptr;
637 }
638 
639 static struct dirent *
640 fts_safe_readdir(DIR *dirp, int *readdir_errno)
641 {
642 	struct dirent *ret;
643 
644 	errno = 0;
645 	if (!dirp)
646 		return (NULL);
647 	ret = readdir(dirp);
648 	*readdir_errno = errno;
649 	return (ret);
650 }
651 
652 /*
653  * This is the tricky part -- do not casually change *anything* in here.  The
654  * idea is to build the linked list of entries that are used by fts_children
655  * and fts_read.  There are lots of special cases.
656  *
657  * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
658  * set and it's a physical walk (so that symbolic links can't be directories),
659  * we can do things quickly.  First, if it's a 4.4BSD file system, the type
660  * of the file is in the directory entry.  Otherwise, we assume that the number
661  * of subdirectories in a node is equal to the number of links to the parent.
662  * The former skips all stat calls.  The latter skips stat calls in any leaf
663  * directories and for any files after the subdirectories in the directory have
664  * been found, cutting the stat calls by about 2/3.
665  */
666 static FTSENT *
667 fts_build(FTS *sp, int type)
668 {
669 	struct dirent *dp;
670 	FTSENT *p, *head;
671 	FTSENT *cur, *tail;
672 	DIR *dirp;
673 	void *oldaddr;
674 	char *cp;
675 	int cderrno, descend, saved_errno, doadjust,
676 		readdir_errno;
677 #ifdef DT_DIR
678 	int nostat;
679 #endif
680 #ifdef FTS_WHITEOUT
681 	int oflag;
682 #endif
683 	long level;
684 	long nlinks;	/* has to be signed because -1 is a magic value */
685 	size_t dnamlen, len, maxlen, nitems;
686 
687 	/* Set current node pointer. */
688 	cur = sp->fts_cur;
689 
690 	/*
691 	 * Open the directory for reading.  If this fails, we're done.
692 	 * If being called from fts_read, set the fts_info field.
693 	 */
694 #ifdef FTS_WHITEOUT
695 	if (ISSET(FTS_WHITEOUT))
696 		oflag = DTF_NODUP;
697 	else
698 		oflag = DTF_HIDEW | DTF_NODUP;
699 #else
700 #define __opendir2(path, flag) opendir(path)
701 #endif
702 	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
703 		if (type == BREAD) {
704 			cur->fts_info = FTS_DNR;
705 			cur->fts_errno = errno;
706 		}
707 		return (NULL);
708 	}
709 
710 	/*
711 	 * Nlinks is the number of possible entries of type directory in the
712 	 * directory if we're cheating on stat calls, 0 if we're not doing
713 	 * any stat calls at all, -1 if we're doing stats on everything.
714 	 */
715 	if (type == BNAMES) {
716 		nlinks = 0;
717 #ifdef DT_DIR
718 		/* Be quiet about nostat, GCC. */
719 		nostat = 0;
720 #endif
721 	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
722 		if (fts_ufslinks(sp, cur))
723 			nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
724 		else
725 			nlinks = -1;
726 #ifdef DT_DIR
727 		nostat = 1;
728 #endif
729 	} else {
730 		nlinks = -1;
731 #ifdef DT_DIR
732 		nostat = 0;
733 #endif
734 	}
735 
736 #ifdef notdef
737 	(void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
738 	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
739 		ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
740 #endif
741 	/*
742 	 * If we're going to need to stat anything or we want to descend
743 	 * and stay in the directory, chdir.  If this fails we keep going,
744 	 * but set a flag so we don't chdir after the post-order visit.
745 	 * We won't be able to stat anything, but we can still return the
746 	 * names themselves.  Note, that since fts_read won't be able to
747 	 * chdir into the directory, it will have to return different path
748 	 * names than before, i.e. "a/b" instead of "b".  Since the node
749 	 * has already been visited in pre-order, have to wait until the
750 	 * post-order visit to return the error.  There is a special case
751 	 * here, if there was nothing to stat then it's not an error to
752 	 * not be able to stat.  This is all fairly nasty.  If a program
753 	 * needed sorted entries or stat information, they had better be
754 	 * checking FTS_NS on the returned nodes.
755 	 */
756 	cderrno = 0;
757 	if (nlinks || type == BREAD) {
758 		if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) {
759 			if (nlinks && type == BREAD)
760 				cur->fts_errno = errno;
761 			cur->fts_flags |= FTS_DONTCHDIR;
762 			descend = 0;
763 			cderrno = errno;
764 		} else
765 			descend = 1;
766 	} else
767 		descend = 0;
768 
769 	/*
770 	 * Figure out the max file name length that can be stored in the
771 	 * current path -- the inner loop allocates more path as necessary.
772 	 * We really wouldn't have to do the maxlen calculations here, we
773 	 * could do them in fts_read before returning the path, but it's a
774 	 * lot easier here since the length is part of the dirent structure.
775 	 *
776 	 * If not changing directories set a pointer so that can just append
777 	 * each new name into the path.
778 	 */
779 	len = NAPPEND(cur);
780 	if (ISSET(FTS_NOCHDIR)) {
781 		cp = sp->fts_path + len;
782 		*cp++ = '/';
783 	} else {
784 		/* GCC, you're too verbose. */
785 		cp = NULL;
786 	}
787 	len++;
788 	maxlen = sp->fts_pathlen - len;
789 
790 	level = cur->fts_level + 1;
791 
792 	/* Read the directory, attaching each entry to the `link' pointer. */
793 	doadjust = 0;
794 	readdir_errno = 0;
795 	for (head = tail = NULL, nitems = 0;
796 		(dp = fts_safe_readdir(dirp, &readdir_errno));) {
797 		dnamlen = strlen(dp->d_name);
798 		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
799 			continue;
800 
801 		if ((p = fts_alloc(sp, dp->d_name, dnamlen)) == NULL)
802 			goto mem1;
803 		if (dnamlen >= maxlen) {	/* include space for NUL */
804 			oldaddr = sp->fts_path;
805 			if (fts_palloc(sp, dnamlen + len + 1)) {
806 				/*
807 				 * No more memory for path or structures.  Save
808 				 * errno, free up the current structure and the
809 				 * structures already allocated.
810 				 */
811 mem1:				saved_errno = errno;
812 				if (p)
813 					free(p);
814 				fts_lfree(head);
815 				(void)closedir(dirp);
816 				cur->fts_info = FTS_ERR;
817 				SET(FTS_STOP);
818 				errno = saved_errno;
819 				return (NULL);
820 			}
821 			/* Did realloc() change the pointer? */
822 			if (oldaddr != sp->fts_path) {
823 				doadjust = 1;
824 				if (ISSET(FTS_NOCHDIR))
825 					cp = sp->fts_path + len;
826 			}
827 			maxlen = sp->fts_pathlen - len;
828 		}
829 
830 		p->fts_level = level;
831 		p->fts_parent = sp->fts_cur;
832 		p->fts_pathlen = len + dnamlen;
833 
834 #ifdef FTS_WHITEOUT
835 		if (dp->d_type == DT_WHT)
836 			p->fts_flags |= FTS_ISW;
837 #endif
838 
839 		if (cderrno) {
840 			if (nlinks) {
841 				p->fts_info = FTS_NS;
842 				p->fts_errno = cderrno;
843 			} else
844 				p->fts_info = FTS_NSOK;
845 			p->fts_accpath = cur->fts_accpath;
846 		} else if (nlinks == 0
847 #ifdef DT_DIR
848 			|| (nostat &&
849 			dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
850 #endif
851 			) {
852 			p->fts_accpath =
853 				ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
854 			p->fts_info = FTS_NSOK;
855 		} else {
856 			/* Build a file name for fts_stat to stat. */
857 			if (ISSET(FTS_NOCHDIR)) {
858 				p->fts_accpath = p->fts_path;
859 				memmove(cp, p->fts_name, p->fts_namelen + 1);
860 				p->fts_info = fts_stat(sp, p, 0, dirfd(dirp));
861 			} else {
862 				p->fts_accpath = p->fts_name;
863 				p->fts_info = fts_stat(sp, p, 0, -1);
864 			}
865 
866 			/* Decrement link count if applicable. */
867 			if (nlinks > 0 && (p->fts_info == FTS_D ||
868 				p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
869 				--nlinks;
870 		}
871 
872 		/* We walk in directory order so "ls -f" doesn't get upset. */
873 		p->fts_link = NULL;
874 		if (head == NULL)
875 			head = tail = p;
876 		else {
877 			tail->fts_link = p;
878 			tail = p;
879 		}
880 		++nitems;
881 	}
882 
883 	if (readdir_errno) {
884 		cur->fts_errno = readdir_errno;
885 		/*
886 		 * If we've not read any items yet, treat
887 		 * the error as if we can't access the dir.
888 		 */
889 		cur->fts_info = nitems ? FTS_ERR : FTS_DNR;
890 	}
891 
892 	if (dirp)
893 		(void)closedir(dirp);
894 
895 	/*
896 	 * If realloc() changed the address of the path, adjust the
897 	 * addresses for the rest of the tree and the dir list.
898 	 */
899 	if (doadjust)
900 		fts_padjust(sp, head);
901 
902 	/*
903 	 * If not changing directories, reset the path back to original
904 	 * state.
905 	 */
906 	if (ISSET(FTS_NOCHDIR))
907 		sp->fts_path[cur->fts_pathlen] = '\0';
908 
909 	/*
910 	 * If descended after called from fts_children or after called from
911 	 * fts_read and nothing found, get back.  At the root level we use
912 	 * the saved fd; if one of fts_open()'s arguments is a relative path
913 	 * to an empty directory, we wind up here with no other way back.  If
914 	 * can't get back, we're done.
915 	 */
916 	if (descend && (type == BCHILD || !nitems) &&
917 		(cur->fts_level == FTS_ROOTLEVEL ?
918 		FCHDIR(sp, sp->fts_rfd) :
919 		fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
920 		fts_lfree(head);
921 		cur->fts_info = FTS_ERR;
922 		SET(FTS_STOP);
923 		return (NULL);
924 	}
925 
926 	/* If didn't find anything, return NULL. */
927 	if (!nitems) {
928 		if (type == BREAD &&
929 			cur->fts_info != FTS_DNR && cur->fts_info != FTS_ERR)
930 			cur->fts_info = FTS_DP;
931 		return (NULL);
932 	}
933 
934 	/* Sort the entries. */
935 	if (sp->fts_compar && nitems > 1)
936 		head = fts_sort(sp, head, nitems);
937 	return (head);
938 }
939 
940 static int
941 fts_stat(FTS *sp, FTSENT *p, int follow, int dfd)
942 {
943 	FTSENT *t;
944 	dev_t dev;
945 	ino_t ino;
946 	struct stat *sbp, sb;
947 	int saved_errno;
948 	const char *path;
949 
950 	if (dfd == -1)
951 		path = p->fts_accpath, dfd = AT_FDCWD;
952 	else
953 		path = p->fts_name;
954 
955 	/* If user needs stat info, stat buffer already allocated. */
956 	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
957 
958 #ifdef FTS_WHITEOUT
959 	/* Check for whiteout. */
960 	if (p->fts_flags & FTS_ISW) {
961 		if (sbp != &sb) {
962 			memset(sbp, '\0', sizeof(*sbp));
963 			sbp->st_mode = S_IFWHT;
964 		}
965 		return (FTS_W);
966 	}
967 #endif
968 
969 	/*
970 	 * If doing a logical walk, or application requested FTS_FOLLOW, do
971 	 * a stat(2).  If that fails, check for a non-existent symlink.  If
972 	 * fail, set the errno from the stat call.
973 	 */
974 	if (ISSET(FTS_LOGICAL) || follow) {
975 		if (fstatat(dfd, path, sbp, 0)) {
976 			saved_errno = errno;
977 			if (fstatat(dfd, path, sbp, AT_SYMLINK_NOFOLLOW)) {
978 				p->fts_errno = saved_errno;
979 				goto err;
980 			}
981 			errno = 0;
982 			if (S_ISLNK(sbp->st_mode))
983 				return (FTS_SLNONE);
984 		}
985 	} else if (fstatat(dfd, path, sbp, AT_SYMLINK_NOFOLLOW)) {
986 		p->fts_errno = errno;
987 err:		memset(sbp, 0, sizeof(struct stat));
988 		return (FTS_NS);
989 	}
990 
991 	if (S_ISDIR(sbp->st_mode)) {
992 		/*
993 		 * Set the device/inode.  Used to find cycles and check for
994 		 * crossing mount points.  Also remember the link count, used
995 		 * in fts_build to limit the number of stat calls.  It is
996 		 * understood that these fields are only referenced if fts_info
997 		 * is set to FTS_D.
998 		 */
999 		dev = p->fts_dev = sbp->st_dev;
1000 		ino = p->fts_ino = sbp->st_ino;
1001 		p->fts_nlink = sbp->st_nlink;
1002 
1003 		if (ISDOT(p->fts_name))
1004 			return (FTS_DOT);
1005 
1006 		/*
1007 		 * Cycle detection is done by brute force when the directory
1008 		 * is first encountered.  If the tree gets deep enough or the
1009 		 * number of symbolic links to directories is high enough,
1010 		 * something faster might be worthwhile.
1011 		 */
1012 		for (t = p->fts_parent;
1013 			t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
1014 			if (ino == t->fts_ino && dev == t->fts_dev) {
1015 				p->fts_cycle = t;
1016 				return (FTS_DC);
1017 			}
1018 		return (FTS_D);
1019 	}
1020 	if (S_ISLNK(sbp->st_mode))
1021 		return (FTS_SL);
1022 	if (S_ISREG(sbp->st_mode))
1023 		return (FTS_F);
1024 	return (FTS_DEFAULT);
1025 }
1026 
1027 /*
1028  * The comparison function takes pointers to pointers to FTSENT structures.
1029  * Qsort wants a comparison function that takes pointers to void.
1030  * (Both with appropriate levels of const-poisoning, of course!)
1031  * Use a trampoline function to deal with the difference.
1032  */
1033 static int
1034 fts_compar(const void *a, const void *b)
1035 {
1036 	FTS *parent;
1037 
1038 	parent = (*(const FTSENT * const *)a)->fts_fts;
1039 	return (*parent->fts_compar)(a, b);
1040 }
1041 
1042 static FTSENT *
1043 fts_sort(FTS *sp, FTSENT *head, size_t nitems)
1044 {
1045 	FTSENT **ap, *p;
1046 
1047 	/*
1048 	 * Construct an array of pointers to the structures and call qsort(3).
1049 	 * Reassemble the array in the order returned by qsort.  If unable to
1050 	 * sort for memory reasons, return the directory entries in their
1051 	 * current order.  Allocate enough space for the current needs plus
1052 	 * 40 so don't realloc one entry at a time.
1053 	 */
1054 	if (nitems > sp->fts_nitems) {
1055 		sp->fts_nitems = nitems + 40;
1056 		if ((sp->fts_array = reallocf(sp->fts_array,
1057 			sp->fts_nitems * sizeof(FTSENT *))) == NULL) {
1058 			sp->fts_nitems = 0;
1059 			return (head);
1060 		}
1061 	}
1062 	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
1063 		*ap++ = p;
1064 	qsort(sp->fts_array, nitems, sizeof(FTSENT *), fts_compar);
1065 	for (head = *(ap = sp->fts_array); --nitems; ++ap)
1066 		ap[0]->fts_link = ap[1];
1067 	ap[0]->fts_link = NULL;
1068 	return (head);
1069 }
1070 
1071 static FTSENT *
1072 fts_alloc(FTS *sp, char *name, size_t namelen)
1073 {
1074 	FTSENT *p;
1075 	size_t len;
1076 
1077 	struct ftsent_withstat {
1078 		FTSENT	ent;
1079 		struct	stat statbuf;
1080 	};
1081 
1082 	/*
1083 	 * The file name is a variable length array and no stat structure is
1084 	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
1085 	 * structure, the file name and the stat structure in one chunk, but
1086 	 * be careful that the stat structure is reasonably aligned.
1087 	 */
1088 	if (ISSET(FTS_NOSTAT))
1089 		len = sizeof(FTSENT) + namelen + 1;
1090 	else
1091 		len = sizeof(struct ftsent_withstat) + namelen + 1;
1092 
1093 	if ((p = malloc(len)) == NULL)
1094 		return (NULL);
1095 
1096 	if (ISSET(FTS_NOSTAT)) {
1097 		p->fts_name = (char *)(p + 1);
1098 		p->fts_statp = NULL;
1099 	} else {
1100 		p->fts_name = (char *)((struct ftsent_withstat *)p + 1);
1101 		p->fts_statp = &((struct ftsent_withstat *)p)->statbuf;
1102 	}
1103 
1104 	/* Copy the name and guarantee NUL termination. */
1105 	memcpy(p->fts_name, name, namelen);
1106 	p->fts_name[namelen] = '\0';
1107 	p->fts_namelen = namelen;
1108 	p->fts_path = sp->fts_path;
1109 	p->fts_errno = 0;
1110 	p->fts_flags = 0;
1111 	p->fts_instr = FTS_NOINSTR;
1112 	p->fts_number = 0;
1113 	p->fts_pointer = NULL;
1114 	p->fts_fts = sp;
1115 	return (p);
1116 }
1117 
1118 static void
1119 fts_lfree(FTSENT *head)
1120 {
1121 	FTSENT *p;
1122 
1123 	/* Free a linked list of structures. */
1124 	while ((p = head)) {
1125 		head = head->fts_link;
1126 		free(p);
1127 	}
1128 }
1129 
1130 /*
1131  * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
1132  * Most systems will allow creation of paths much longer than MAXPATHLEN, even
1133  * though the kernel won't resolve them.  Add the size (not just what's needed)
1134  * plus 256 bytes so don't realloc the path 2 bytes at a time.
1135  */
1136 static int
1137 fts_palloc(FTS *sp, size_t more)
1138 {
1139 
1140 	sp->fts_pathlen += more + 256;
1141 	sp->fts_path = reallocf(sp->fts_path, sp->fts_pathlen);
1142 	return (sp->fts_path == NULL);
1143 }
1144 
1145 /*
1146  * When the path is realloc'd, have to fix all of the pointers in structures
1147  * already returned.
1148  */
1149 static void
1150 fts_padjust(FTS *sp, FTSENT *head)
1151 {
1152 	FTSENT *p;
1153 	char *addr = sp->fts_path;
1154 
1155 #define	ADJUST(p) do {							\
1156 	if ((p)->fts_accpath != (p)->fts_name) {			\
1157 		(p)->fts_accpath =					\
1158 			(char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
1159 	}								\
1160 	(p)->fts_path = addr;						\
1161 } while (0)
1162 	/* Adjust the current set of children. */
1163 	for (p = sp->fts_child; p; p = p->fts_link)
1164 		ADJUST(p);
1165 
1166 	/* Adjust the rest of the tree, including the current level. */
1167 	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1168 		ADJUST(p);
1169 		p = p->fts_link ? p->fts_link : p->fts_parent;
1170 	}
1171 }
1172 
1173 static size_t
1174 fts_maxarglen(char * const *argv)
1175 {
1176 	size_t len, max;
1177 
1178 	for (max = 0; *argv; ++argv)
1179 		if ((len = strlen(*argv)) > max)
1180 			max = len;
1181 	return (max + 1);
1182 }
1183 
1184 /*
1185  * Change to dir specified by fd or p->fts_accpath without getting
1186  * tricked by someone changing the world out from underneath us.
1187  * Assumes p->fts_dev and p->fts_ino are filled in.
1188  */
1189 static int
1190 fts_safe_changedir(FTS *sp, FTSENT *p, int fd, char *path)
1191 {
1192 	int ret, oerrno, newfd;
1193 	struct stat sb;
1194 
1195 	newfd = fd;
1196 	if (ISSET(FTS_NOCHDIR))
1197 		return (0);
1198 	if (fd < 0 && (newfd = open(path, O_RDONLY | O_DIRECTORY |
1199 		O_CLOEXEC, 0)) < 0)
1200 		return (-1);
1201 	if (fstat(newfd, &sb)) {
1202 		ret = -1;
1203 		goto bail;
1204 	}
1205 	if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) {
1206 		errno = ENOENT;		/* disinformation */
1207 		ret = -1;
1208 		goto bail;
1209 	}
1210 	ret = fchdir(newfd);
1211 bail:
1212 	oerrno = errno;
1213 	if (fd < 0)
1214 		(void)close(newfd);
1215 	errno = oerrno;
1216 	return (ret);
1217 }
1218 
1219 /*
1220  * Check if the filesystem for "ent" has UFS-style links.
1221  */
1222 static int
1223 fts_ufslinks(FTS *sp, const FTSENT *ent)
1224 {
1225 	struct _fts_private *priv;
1226 #ifndef __HAIKU__
1227 	const char **cpp;
1228 #endif
1229 
1230 	priv = (struct _fts_private *)sp;
1231 	/*
1232 	 * If this node's device is different from the previous, grab
1233 	 * the filesystem information, and decide on the reliability
1234 	 * of the link information from this filesystem for stat(2)
1235 	 * avoidance.
1236 	 */
1237 	if (priv->ftsp_dev != ent->fts_dev) {
1238 #ifndef __HAIKU__
1239 		if (statfs(ent->fts_path, &priv->ftsp_statfs) != -1) {
1240 			priv->ftsp_dev = ent->fts_dev;
1241 			priv->ftsp_linksreliable = 0;
1242 			for (cpp = ufslike_filesystems; *cpp; cpp++) {
1243 				if (strcmp(priv->ftsp_statfs.f_fstypename,
1244 					*cpp) == 0) {
1245 					priv->ftsp_linksreliable = 1;
1246 					break;
1247 				}
1248 			}
1249 		} else {
1250 			priv->ftsp_linksreliable = 0;
1251 		}
1252 #else
1253 		priv->ftsp_linksreliable = 0;
1254 #endif
1255 	}
1256 	return (priv->ftsp_linksreliable);
1257 }
1258