xref: /third_party/FreeBSD/lib/libc/gen/fts.c (revision f9f848fa)
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/param.h>
35#include <sys/mount.h>
36#include <sys/stat.h>
37#include <sys/statfs.h>
38
39#include <dirent.h>
40#include <errno.h>
41#include <fcntl.h>
42#include <stdlib.h>
43#include <string.h>
44#include <unistd.h>
45#include <sys/sys/cdefs.h>
46
47#include <linux/magic.h>
48#include "include/fts.h"
49
50static FTSENT	*fts_alloc(FTS *, char *, size_t);
51static FTSENT	*fts_build(FTS *, int);
52static void	 fts_lfree(FTSENT *);
53static void	 fts_load(FTS *, FTSENT *);
54static size_t	 fts_maxarglen(char * const *);
55static void	 fts_padjust(FTS *, FTSENT *);
56static int	 fts_palloc(FTS *, size_t);
57static FTSENT	*fts_sort(FTS *, FTSENT *, size_t);
58static int	 fts_stat(FTS *, FTSENT *, int, int);
59static int	 fts_safe_changedir(FTS *, FTSENT *, int, char *);
60static int	 fts_ufslinks(FTS *, const FTSENT *);
61
62#define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
63
64#define	CLR(opt)	(sp->fts_options &= ~(opt))
65#define	ISSET(opt)	(sp->fts_options & (opt))
66#define	SET(opt)	(sp->fts_options |= (opt))
67
68#define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
69
70/* fts_build flags */
71#define	BCHILD		1		/* fts_children */
72#define	BNAMES		2		/* fts_children, names only */
73#define	BREAD		3		/* fts_read */
74
75/*
76 * Internal representation of an FTS, including extra implementation
77 * details.  The FTS returned from fts_open points to this structure's
78 * ftsp_fts member (and can be cast to an _fts_private as required)
79 */
80struct _fts_private {
81	FTS		ftsp_fts;
82	struct statfs	ftsp_statfs;
83	dev_t		ftsp_dev;
84	int		ftsp_linksreliable;
85};
86
87/*
88 * The "FTS_NOSTAT" option can avoid a lot of calls to stat(2) if it
89 * knows that a directory could not possibly have subdirectories.  This
90 * is decided by looking at the link count: a subdirectory would
91 * increment its parent's link count by virtue of its own ".." entry.
92 * This assumption only holds for UFS-like filesystems that implement
93 * links and directories this way, so we must punt for others.
94 */
95
96//
97// Make it works on Linux compiler
98//
99//static const char *ufslike_filesystems[] = {
100//       "ufs",
101//       "zfs",
102//       "nfs",
103//       "ext2fs",
104//       0
105//};
106//
107// Values from "man 2 statfs". Linux not support zfs.
108static const uint64_t ufslike_filesystems[] = {
109	0x00011954, // ufs
110	0x6969,     // nfs
111	0xef53,     // ext2fs
112};
113
114FTS *
115fts_open(char * const *argv, int options,
116    int (*compar)(const FTSENT * const *, const FTSENT * const *))
117{
118	struct _fts_private *priv;
119	FTS *sp;
120	FTSENT *p, *root;
121	FTSENT *parent, *tmp;
122	size_t len, nitems;
123
124	/* Options check. */
125	if (options & ~FTS_OPTIONMASK) {
126		errno = EINVAL;
127		return (NULL);
128	}
129
130	/* fts_open() requires at least one path */
131	if (*argv == NULL) {
132		errno = EINVAL;
133		return (NULL);
134	}
135
136	/* Allocate/initialize the stream. */
137	if ((priv = calloc(1, sizeof(*priv))) == NULL)
138		return (NULL);
139	sp = &priv->ftsp_fts;
140	sp->fts_compar = compar;
141	sp->fts_options = options;
142
143	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
144	if (ISSET(FTS_LOGICAL))
145		SET(FTS_NOCHDIR);
146
147	/*
148	 * Start out with 1K of path space, and enough, in any case,
149	 * to hold the user's paths.
150	 */
151	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
152		goto mem1;
153
154	/* Allocate/initialize root's parent. */
155	if ((parent = fts_alloc(sp, "", 0)) == NULL)
156		goto mem2;
157	parent->fts_level = FTS_ROOTPARENTLEVEL;
158
159	/* Shush, GCC. */
160	tmp = NULL;
161
162	/* Allocate/initialize root(s). */
163	for (root = NULL, nitems = 0; *argv != NULL; ++argv, ++nitems) {
164		len = strlen(*argv);
165
166		p = fts_alloc(sp, *argv, len);
167		p->fts_level = FTS_ROOTLEVEL;
168		p->fts_parent = parent;
169		p->fts_accpath = p->fts_name;
170		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW), -1);
171
172		/* Command-line "." and ".." are real directories. */
173		if (p->fts_info == FTS_DOT)
174			p->fts_info = FTS_D;
175
176		/*
177		 * If comparison routine supplied, traverse in sorted
178		 * order; otherwise traverse in the order specified.
179		 */
180		if (compar) {
181			p->fts_link = root;
182			root = p;
183		} else {
184			p->fts_link = NULL;
185			if (root == NULL)
186				tmp = root = p;
187			else {
188				tmp->fts_link = p;
189				tmp = p;
190			}
191		}
192	}
193	if (compar && nitems > 1)
194		root = fts_sort(sp, root, nitems);
195
196	/*
197	 * Allocate a dummy pointer and make fts_read think that we've just
198	 * finished the node before the root(s); set p->fts_info to FTS_INIT
199	 * so that everything about the "current" node is ignored.
200	 */
201	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
202		goto mem3;
203	sp->fts_cur->fts_link = root;
204	sp->fts_cur->fts_info = FTS_INIT;
205
206	/*
207	 * If using chdir(2), grab a file descriptor pointing to dot to ensure
208	 * that we can get back here; this could be avoided for some paths,
209	 * but almost certainly not worth the effort.  Slashes, symbolic links,
210	 * and ".." are all fairly nasty problems.  Note, if we can't get the
211	 * descriptor we run anyway, just more slowly.
212	 */
213	if (!ISSET(FTS_NOCHDIR) &&
214	    (sp->fts_rfd = _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0)
215		SET(FTS_NOCHDIR);
216
217	return (sp);
218
219mem3:	fts_lfree(root);
220	free(parent);
221mem2:	free(sp->fts_path);
222mem1:	free(sp);
223	return (NULL);
224}
225
226static void
227fts_load(FTS *sp, FTSENT *p)
228{
229	size_t len;
230	char *cp;
231
232	/*
233	 * Load the stream structure for the next traversal.  Since we don't
234	 * actually enter the directory until after the preorder visit, set
235	 * the fts_accpath field specially so the chdir gets done to the right
236	 * place and the user can access the first node.  From fts_open it's
237	 * known that the path will fit.
238	 */
239	len = p->fts_pathlen = p->fts_namelen;
240	memmove(sp->fts_path, p->fts_name, len + 1);
241	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
242		len = strlen(++cp);
243		memmove(p->fts_name, cp, len + 1);
244		p->fts_namelen = len;
245	}
246	p->fts_accpath = p->fts_path = sp->fts_path;
247	sp->fts_dev = p->fts_dev;
248}
249
250int
251fts_close(FTS *sp)
252{
253	FTSENT *freep, *p;
254	int saved_errno;
255
256	/*
257	 * This still works if we haven't read anything -- the dummy structure
258	 * points to the root list, so we step through to the end of the root
259	 * list which has a valid parent pointer.
260	 */
261	if (sp->fts_cur) {
262		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
263			freep = p;
264			p = p->fts_link != NULL ? p->fts_link : p->fts_parent;
265			free(freep);
266		}
267		free(p);
268	}
269
270	/* Free up child linked list, sort array, path buffer. */
271	if (sp->fts_child)
272		fts_lfree(sp->fts_child);
273	if (sp->fts_array)
274		free(sp->fts_array);
275	free(sp->fts_path);
276
277	/* Return to original directory, save errno if necessary. */
278	if (!ISSET(FTS_NOCHDIR)) {
279		saved_errno = fchdir(sp->fts_rfd) ? errno : 0;
280		(void)_close(sp->fts_rfd);
281
282		/* Set errno and return. */
283		if (saved_errno != 0) {
284			/* Free up the stream pointer. */
285			free(sp);
286			errno = saved_errno;
287			return (-1);
288		}
289	}
290
291	/* Free up the stream pointer. */
292	free(sp);
293	return (0);
294}
295
296/*
297 * Special case of "/" at the end of the path so that slashes aren't
298 * appended which would cause paths to be written as "....//foo".
299 */
300#define	NAPPEND(p)							\
301	(p->fts_path[p->fts_pathlen - 1] == '/'				\
302	    ? p->fts_pathlen - 1 : p->fts_pathlen)
303
304FTSENT *
305fts_read(FTS *sp)
306{
307	FTSENT *p, *tmp;
308	int instr;
309	char *t;
310	int saved_errno;
311
312	/* If finished or unrecoverable error, return NULL. */
313	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
314		return (NULL);
315
316	/* Set current node pointer. */
317	p = sp->fts_cur;
318
319	/* Save and zero out user instructions. */
320	instr = p->fts_instr;
321	p->fts_instr = FTS_NOINSTR;
322
323	/* Any type of file may be re-visited; re-stat and re-turn. */
324	if (instr == FTS_AGAIN) {
325		p->fts_info = fts_stat(sp, p, 0, -1);
326		return (p);
327	}
328
329	/*
330	 * Following a symlink -- SLNONE test allows application to see
331	 * SLNONE and recover.  If indirecting through a symlink, have
332	 * keep a pointer to current location.  If unable to get that
333	 * pointer, follow fails.
334	 */
335	if (instr == FTS_FOLLOW &&
336	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
337		p->fts_info = fts_stat(sp, p, 1, -1);
338		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
339			if ((p->fts_symfd = _open(".", O_RDONLY | O_CLOEXEC,
340			    0)) < 0) {
341				p->fts_errno = errno;
342				p->fts_info = FTS_ERR;
343			} else
344				p->fts_flags |= FTS_SYMFOLLOW;
345		}
346		return (p);
347	}
348
349	/* Directory in pre-order. */
350	if (p->fts_info == FTS_D) {
351		/* If skipped or crossed mount point, do post-order visit. */
352		if (instr == FTS_SKIP ||
353		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
354			if (p->fts_flags & FTS_SYMFOLLOW)
355				(void)_close(p->fts_symfd);
356			if (sp->fts_child) {
357				fts_lfree(sp->fts_child);
358				sp->fts_child = NULL;
359			}
360			p->fts_info = FTS_DP;
361			return (p);
362		}
363
364		/* Rebuild if only read the names and now traversing. */
365		if (sp->fts_child != NULL && ISSET(FTS_NAMEONLY)) {
366			CLR(FTS_NAMEONLY);
367			fts_lfree(sp->fts_child);
368			sp->fts_child = NULL;
369		}
370
371		/*
372		 * Cd to the subdirectory.
373		 *
374		 * If have already read and now fail to chdir, whack the list
375		 * to make the names come out right, and set the parent errno
376		 * so the application will eventually get an error condition.
377		 * Set the FTS_DONTCHDIR flag so that when we logically change
378		 * directories back to the parent we don't do a chdir.
379		 *
380		 * If haven't read do so.  If the read fails, fts_build sets
381		 * FTS_STOP or the fts_info field of the node.
382		 */
383		if (sp->fts_child != NULL) {
384			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
385				p->fts_errno = errno;
386				p->fts_flags |= FTS_DONTCHDIR;
387				for (p = sp->fts_child; p != NULL;
388				    p = p->fts_link)
389					p->fts_accpath =
390					    p->fts_parent->fts_accpath;
391			}
392		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
393			if (ISSET(FTS_STOP))
394				return (NULL);
395			return (p);
396		}
397		p = sp->fts_child;
398		sp->fts_child = NULL;
399		goto name;
400	}
401
402	/* Move to the next node on this level. */
403next:	tmp = p;
404	if ((p = p->fts_link) != NULL) {
405		/*
406		 * If reached the top, return to the original directory (or
407		 * the root of the tree), and load the paths for the next root.
408		 */
409		if (p->fts_level == FTS_ROOTLEVEL) {
410			if (FCHDIR(sp, sp->fts_rfd)) {
411				SET(FTS_STOP);
412				return (NULL);
413			}
414			free(tmp);
415			fts_load(sp, p);
416			return (sp->fts_cur = p);
417		}
418
419		/*
420		 * User may have called fts_set on the node.  If skipped,
421		 * ignore.  If followed, get a file descriptor so we can
422		 * get back if necessary.
423		 */
424		if (p->fts_instr == FTS_SKIP) {
425			free(tmp);
426			goto next;
427		}
428		if (p->fts_instr == FTS_FOLLOW) {
429			p->fts_info = fts_stat(sp, p, 1, -1);
430			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
431				if ((p->fts_symfd =
432				    _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0) {
433					p->fts_errno = errno;
434					p->fts_info = FTS_ERR;
435				} else
436					p->fts_flags |= FTS_SYMFOLLOW;
437			}
438			p->fts_instr = FTS_NOINSTR;
439		}
440
441		free(tmp);
442
443name:		t = sp->fts_path + NAPPEND(p->fts_parent);
444		*t++ = '/';
445		memmove(t, p->fts_name, p->fts_namelen + 1);
446		return (sp->fts_cur = p);
447	}
448
449	/* Move up to the parent node. */
450	p = tmp->fts_parent;
451
452	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
453		/*
454		 * Done; free everything up and set errno to 0 so the user
455		 * can distinguish between error and EOF.
456		 */
457		free(tmp);
458		free(p);
459		errno = 0;
460		return (sp->fts_cur = NULL);
461	}
462
463	/* NUL terminate the pathname. */
464	sp->fts_path[p->fts_pathlen] = '\0';
465
466	/*
467	 * Return to the parent directory.  If at a root node or came through
468	 * a symlink, go back through the file descriptor.  Otherwise, cd up
469	 * one directory.
470	 */
471	if (p->fts_level == FTS_ROOTLEVEL) {
472		if (FCHDIR(sp, sp->fts_rfd)) {
473			SET(FTS_STOP);
474			return (NULL);
475		}
476	} else if (p->fts_flags & FTS_SYMFOLLOW) {
477		if (FCHDIR(sp, p->fts_symfd)) {
478			saved_errno = errno;
479			(void)_close(p->fts_symfd);
480			errno = saved_errno;
481			SET(FTS_STOP);
482			return (NULL);
483		}
484		(void)_close(p->fts_symfd);
485	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
486	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
487		SET(FTS_STOP);
488		return (NULL);
489	}
490	free(tmp);
491	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
492	return (sp->fts_cur = p);
493}
494
495/*
496 * Fts_set takes the stream as an argument although it's not used in this
497 * implementation; it would be necessary if anyone wanted to add global
498 * semantics to fts using fts_set.  An error return is allowed for similar
499 * reasons.
500 */
501/* ARGSUSED */
502int
503fts_set(FTS *sp, FTSENT *p, int instr)
504{
505	if (instr != 0 && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
506	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
507		errno = EINVAL;
508		return (1);
509	}
510	p->fts_instr = instr;
511	return (0);
512}
513
514FTSENT *
515fts_children(FTS *sp, int instr)
516{
517	FTSENT *p;
518	int fd, rc, serrno;
519
520	if (instr != 0 && instr != FTS_NAMEONLY) {
521		errno = EINVAL;
522		return (NULL);
523	}
524
525	/* Set current node pointer. */
526	p = sp->fts_cur;
527
528	/*
529	 * Errno set to 0 so user can distinguish empty directory from
530	 * an error.
531	 */
532	errno = 0;
533
534	/* Fatal errors stop here. */
535	if (ISSET(FTS_STOP))
536		return (NULL);
537
538	/* Return logical hierarchy of user's arguments. */
539	if (p->fts_info == FTS_INIT)
540		return (p->fts_link);
541
542	/*
543	 * If not a directory being visited in pre-order, stop here.  Could
544	 * allow FTS_DNR, assuming the user has fixed the problem, but the
545	 * same effect is available with FTS_AGAIN.
546	 */
547	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
548		return (NULL);
549
550	/* Free up any previous child list. */
551	if (sp->fts_child != NULL)
552		fts_lfree(sp->fts_child);
553
554	if (instr == FTS_NAMEONLY) {
555		SET(FTS_NAMEONLY);
556		instr = BNAMES;
557	} else
558		instr = BCHILD;
559
560	/*
561	 * If using chdir on a relative path and called BEFORE fts_read does
562	 * its chdir to the root of a traversal, we can lose -- we need to
563	 * chdir into the subdirectory, and we don't know where the current
564	 * directory is, so we can't get back so that the upcoming chdir by
565	 * fts_read will work.
566	 */
567	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
568	    ISSET(FTS_NOCHDIR))
569		return (sp->fts_child = fts_build(sp, instr));
570
571	if ((fd = _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0)
572		return (NULL);
573	sp->fts_child = fts_build(sp, instr);
574	serrno = (sp->fts_child == NULL) ? errno : 0;
575	rc = fchdir(fd);
576	if (rc < 0 && serrno == 0)
577		serrno = errno;
578	(void)_close(fd);
579	errno = serrno;
580	if (rc < 0)
581		return (NULL);
582	return (sp->fts_child);
583}
584
585#ifndef fts_get_clientptr
586#error "fts_get_clientptr not defined"
587#endif
588
589void *
590(fts_get_clientptr)(FTS *sp)
591{
592
593	return (fts_get_clientptr(sp));
594}
595
596#ifndef fts_get_stream
597#error "fts_get_stream not defined"
598#endif
599
600FTS *
601(fts_get_stream)(FTSENT *p)
602{
603	return (fts_get_stream(p));
604}
605
606void
607fts_set_clientptr(FTS *sp, void *clientptr)
608{
609
610	sp->fts_clientptr = clientptr;
611}
612
613/*
614 * This is the tricky part -- do not casually change *anything* in here.  The
615 * idea is to build the linked list of entries that are used by fts_children
616 * and fts_read.  There are lots of special cases.
617 *
618 * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
619 * set and it's a physical walk (so that symbolic links can't be directories),
620 * we can do things quickly.  First, if it's a 4.4BSD file system, the type
621 * of the file is in the directory entry.  Otherwise, we assume that the number
622 * of subdirectories in a node is equal to the number of links to the parent.
623 * The former skips all stat calls.  The latter skips stat calls in any leaf
624 * directories and for any files after the subdirectories in the directory have
625 * been found, cutting the stat calls by about 2/3.
626 */
627static FTSENT *
628fts_build(FTS *sp, int type)
629{
630	struct dirent *dp;
631	FTSENT *p, *head;
632	FTSENT *cur, *tail;
633	DIR *dirp;
634	void *oldaddr;
635	char *cp;
636	int cderrno, descend, oflag, saved_errno, nostat, doadjust;
637	long level;
638	long nlinks;	/* has to be signed because -1 is a magic value */
639	size_t dnamlen, len, maxlen, nitems;
640
641	/* Set current node pointer. */
642	cur = sp->fts_cur;
643
644	/*
645	 * Open the directory for reading.  If this fails, we're done.
646	 * If being called from fts_read, set the fts_info field.
647	 */
648#ifdef FTS_WHITEOUT
649	if (ISSET(FTS_WHITEOUT))
650		oflag = DTF_NODUP;
651	else
652		oflag = DTF_HIDEW | DTF_NODUP;
653#else
654#define __opendir2(path, flag) opendir(path)
655#endif
656	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
657		if (type == BREAD) {
658			cur->fts_info = FTS_DNR;
659			cur->fts_errno = errno;
660		}
661		return (NULL);
662	}
663
664	/*
665	 * Nlinks is the number of possible entries of type directory in the
666	 * directory if we're cheating on stat calls, 0 if we're not doing
667	 * any stat calls at all, -1 if we're doing stats on everything.
668	 */
669	if (type == BNAMES) {
670		nlinks = 0;
671		/* Be quiet about nostat, GCC. */
672		nostat = 0;
673	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
674		if (fts_ufslinks(sp, cur))
675			nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
676		else
677			nlinks = -1;
678		nostat = 1;
679	} else {
680		nlinks = -1;
681		nostat = 0;
682	}
683
684#ifdef notdef
685	(void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
686	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
687	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
688#endif
689	/*
690	 * If we're going to need to stat anything or we want to descend
691	 * and stay in the directory, chdir.  If this fails we keep going,
692	 * but set a flag so we don't chdir after the post-order visit.
693	 * We won't be able to stat anything, but we can still return the
694	 * names themselves.  Note, that since fts_read won't be able to
695	 * chdir into the directory, it will have to return different path
696	 * names than before, i.e. "a/b" instead of "b".  Since the node
697	 * has already been visited in pre-order, have to wait until the
698	 * post-order visit to return the error.  There is a special case
699	 * here, if there was nothing to stat then it's not an error to
700	 * not be able to stat.  This is all fairly nasty.  If a program
701	 * needed sorted entries or stat information, they had better be
702	 * checking FTS_NS on the returned nodes.
703	 */
704	cderrno = 0;
705	if (nlinks || type == BREAD) {
706		if (fts_safe_changedir(sp, cur, _dirfd(dirp), NULL)) {
707			if (nlinks && type == BREAD)
708				cur->fts_errno = errno;
709			cur->fts_flags |= FTS_DONTCHDIR;
710			descend = 0;
711			cderrno = errno;
712		} else
713			descend = 1;
714	} else
715		descend = 0;
716
717	/*
718	 * Figure out the max file name length that can be stored in the
719	 * current path -- the inner loop allocates more path as necessary.
720	 * We really wouldn't have to do the maxlen calculations here, we
721	 * could do them in fts_read before returning the path, but it's a
722	 * lot easier here since the length is part of the dirent structure.
723	 *
724	 * If not changing directories set a pointer so that can just append
725	 * each new name into the path.
726	 */
727	len = NAPPEND(cur);
728	if (ISSET(FTS_NOCHDIR)) {
729		cp = sp->fts_path + len;
730		*cp++ = '/';
731	} else {
732		/* GCC, you're too verbose. */
733		cp = NULL;
734	}
735	len++;
736	maxlen = sp->fts_pathlen - len;
737
738	level = cur->fts_level + 1;
739
740	/* Read the directory, attaching each entry to the `link' pointer. */
741	doadjust = 0;
742	for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) {
743		dnamlen = strlen(dp->d_name);
744		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
745			continue;
746
747		if ((p = fts_alloc(sp, dp->d_name, dnamlen)) == NULL)
748			goto mem1;
749		if (dnamlen >= maxlen) {	/* include space for NUL */
750			oldaddr = sp->fts_path;
751			if (fts_palloc(sp, dnamlen + len + 1)) {
752				/*
753				 * No more memory for path or structures.  Save
754				 * errno, free up the current structure and the
755				 * structures already allocated.
756				 */
757mem1:				saved_errno = errno;
758				if (p)
759					free(p);
760				fts_lfree(head);
761				(void)closedir(dirp);
762				cur->fts_info = FTS_ERR;
763				SET(FTS_STOP);
764				errno = saved_errno;
765				return (NULL);
766			}
767			/* Did realloc() change the pointer? */
768			if (oldaddr != sp->fts_path) {
769				doadjust = 1;
770				if (ISSET(FTS_NOCHDIR))
771					cp = sp->fts_path + len;
772			}
773			maxlen = sp->fts_pathlen - len;
774		}
775
776		p->fts_level = level;
777		p->fts_parent = sp->fts_cur;
778		p->fts_pathlen = len + dnamlen;
779
780#ifdef FTS_WHITEOUT
781		if (dp->d_type == DT_WHT)
782			p->fts_flags |= FTS_ISW;
783#endif
784
785		if (cderrno) {
786			if (nlinks) {
787				p->fts_info = FTS_NS;
788				p->fts_errno = cderrno;
789			} else
790				p->fts_info = FTS_NSOK;
791			p->fts_accpath = cur->fts_accpath;
792		} else if (nlinks == 0
793#ifdef DT_DIR
794		    || (nostat &&
795		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
796#endif
797		    ) {
798			p->fts_accpath =
799			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
800			p->fts_info = FTS_NSOK;
801		} else {
802			/* Build a file name for fts_stat to stat. */
803			if (ISSET(FTS_NOCHDIR)) {
804				p->fts_accpath = p->fts_path;
805				memmove(cp, p->fts_name, p->fts_namelen + 1);
806				p->fts_info = fts_stat(sp, p, 0, _dirfd(dirp));
807			} else {
808				p->fts_accpath = p->fts_name;
809				p->fts_info = fts_stat(sp, p, 0, -1);
810			}
811
812			/* Decrement link count if applicable. */
813			if (nlinks > 0 && (p->fts_info == FTS_D ||
814			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
815				--nlinks;
816		}
817
818		/* We walk in directory order so "ls -f" doesn't get upset. */
819		p->fts_link = NULL;
820		if (head == NULL)
821			head = tail = p;
822		else {
823			tail->fts_link = p;
824			tail = p;
825		}
826		++nitems;
827	}
828	if (dirp)
829		(void)closedir(dirp);
830
831	/*
832	 * If realloc() changed the address of the path, adjust the
833	 * addresses for the rest of the tree and the dir list.
834	 */
835	if (doadjust)
836		fts_padjust(sp, head);
837
838	/*
839	 * If not changing directories, reset the path back to original
840	 * state.
841	 */
842	if (ISSET(FTS_NOCHDIR))
843		sp->fts_path[cur->fts_pathlen] = '\0';
844
845	/*
846	 * If descended after called from fts_children or after called from
847	 * fts_read and nothing found, get back.  At the root level we use
848	 * the saved fd; if one of fts_open()'s arguments is a relative path
849	 * to an empty directory, we wind up here with no other way back.  If
850	 * can't get back, we're done.
851	 */
852	if (descend && (type == BCHILD || !nitems) &&
853	    (cur->fts_level == FTS_ROOTLEVEL ?
854	    FCHDIR(sp, sp->fts_rfd) :
855	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
856		fts_lfree(head);
857		cur->fts_info = FTS_ERR;
858		SET(FTS_STOP);
859		return (NULL);
860	}
861
862	/* If didn't find anything, return NULL. */
863	if (!nitems) {
864		if (type == BREAD)
865			cur->fts_info = FTS_DP;
866		return (NULL);
867	}
868
869	/* Sort the entries. */
870	if (sp->fts_compar && nitems > 1)
871		head = fts_sort(sp, head, nitems);
872	return (head);
873}
874
875static int
876fts_stat(FTS *sp, FTSENT *p, int follow, int dfd)
877{
878	FTSENT *t;
879	dev_t dev;
880	ino_t ino;
881	struct stat *sbp, sb;
882	int saved_errno;
883	const char *path;
884
885	if (dfd == -1)
886		path = p->fts_accpath, dfd = AT_FDCWD;
887	else
888		path = p->fts_name;
889
890	/* If user needs stat info, stat buffer already allocated. */
891	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
892
893#ifdef FTS_WHITEOUT
894	/* Check for whiteout. */
895	if (p->fts_flags & FTS_ISW) {
896		if (sbp != &sb) {
897			memset(sbp, '\0', sizeof(*sbp));
898			sbp->st_mode = S_IFWHT;
899		}
900		return (FTS_W);
901	}
902#endif
903
904	/*
905	 * If doing a logical walk, or application requested FTS_FOLLOW, do
906	 * a stat(2).  If that fails, check for a non-existent symlink.  If
907	 * fail, set the errno from the stat call.
908	 */
909	if (ISSET(FTS_LOGICAL) || follow) {
910		if (fstatat(dfd, path, sbp, 0)) {
911			saved_errno = errno;
912			if (fstatat(dfd, path, sbp, AT_SYMLINK_NOFOLLOW)) {
913				p->fts_errno = saved_errno;
914				goto err;
915			}
916			errno = 0;
917			if (S_ISLNK(sbp->st_mode))
918				return (FTS_SLNONE);
919		}
920	} else if (fstatat(dfd, path, sbp, AT_SYMLINK_NOFOLLOW)) {
921		p->fts_errno = errno;
922err:		memset(sbp, 0, sizeof(struct stat));
923		return (FTS_NS);
924	}
925
926	if (S_ISDIR(sbp->st_mode)) {
927		/*
928		 * Set the device/inode.  Used to find cycles and check for
929		 * crossing mount points.  Also remember the link count, used
930		 * in fts_build to limit the number of stat calls.  It is
931		 * understood that these fields are only referenced if fts_info
932		 * is set to FTS_D.
933		 */
934		dev = p->fts_dev = sbp->st_dev;
935		ino = p->fts_ino = sbp->st_ino;
936		p->fts_nlink = sbp->st_nlink;
937
938		if (ISDOT(p->fts_name))
939			return (FTS_DOT);
940
941		/*
942		 * Cycle detection is done by brute force when the directory
943		 * is first encountered.  If the tree gets deep enough or the
944		 * number of symbolic links to directories is high enough,
945		 * something faster might be worthwhile.
946		 */
947		for (t = p->fts_parent;
948		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
949			if (ino == t->fts_ino && dev == t->fts_dev) {
950				p->fts_cycle = t;
951				return (FTS_DC);
952			}
953		return (FTS_D);
954	}
955	if (S_ISLNK(sbp->st_mode))
956		return (FTS_SL);
957	if (S_ISREG(sbp->st_mode))
958		return (FTS_F);
959	return (FTS_DEFAULT);
960}
961
962/*
963 * The comparison function takes pointers to pointers to FTSENT structures.
964 * Qsort wants a comparison function that takes pointers to void.
965 * (Both with appropriate levels of const-poisoning, of course!)
966 * Use a trampoline function to deal with the difference.
967 */
968static int
969fts_compar(const void *a, const void *b)
970{
971	FTS *parent;
972
973	parent = (*(const FTSENT * const *)a)->fts_fts;
974	return (*parent->fts_compar)(a, b);
975}
976
977static FTSENT *
978fts_sort(FTS *sp, FTSENT *head, size_t nitems)
979{
980	FTSENT **ap, *p;
981
982	/*
983	 * Construct an array of pointers to the structures and call qsort(3).
984	 * Reassemble the array in the order returned by qsort.  If unable to
985	 * sort for memory reasons, return the directory entries in their
986	 * current order.  Allocate enough space for the current needs plus
987	 * 40 so don't realloc one entry at a time.
988	 */
989	if (nitems > sp->fts_nitems) {
990		sp->fts_nitems = nitems + 40;
991		if ((sp->fts_array = reallocf(sp->fts_array,
992		    sp->fts_nitems * sizeof(FTSENT *))) == NULL) {
993			sp->fts_nitems = 0;
994			return (head);
995		}
996	}
997	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
998		*ap++ = p;
999	qsort(sp->fts_array, nitems, sizeof(FTSENT *), fts_compar);
1000	for (head = *(ap = sp->fts_array); --nitems; ++ap)
1001		ap[0]->fts_link = ap[1];
1002	ap[0]->fts_link = NULL;
1003	return (head);
1004}
1005
1006static FTSENT *
1007fts_alloc(FTS *sp, char *name, size_t namelen)
1008{
1009	FTSENT *p;
1010	size_t len;
1011
1012	struct ftsent_withstat {
1013		FTSENT	ent;
1014		struct	stat statbuf;
1015	};
1016
1017	/*
1018	 * The file name is a variable length array and no stat structure is
1019	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
1020	 * structure, the file name and the stat structure in one chunk, but
1021	 * be careful that the stat structure is reasonably aligned.
1022	 */
1023	if (ISSET(FTS_NOSTAT))
1024		len = sizeof(FTSENT) + namelen + 1;
1025	else
1026		len = sizeof(struct ftsent_withstat) + namelen + 1;
1027
1028	if ((p = malloc(len)) == NULL)
1029		return (NULL);
1030
1031	if (ISSET(FTS_NOSTAT)) {
1032		p->fts_name = (char *)(p + 1);
1033		p->fts_statp = NULL;
1034	} else {
1035		p->fts_name = (char *)((struct ftsent_withstat *)p + 1);
1036		p->fts_statp = &((struct ftsent_withstat *)p)->statbuf;
1037	}
1038
1039	/* Copy the name and guarantee NUL termination. */
1040	memcpy(p->fts_name, name, namelen);
1041	p->fts_name[namelen] = '\0';
1042	p->fts_namelen = namelen;
1043	p->fts_path = sp->fts_path;
1044	p->fts_errno = 0;
1045	p->fts_flags = 0;
1046	p->fts_instr = FTS_NOINSTR;
1047	p->fts_number = 0;
1048	p->fts_pointer = NULL;
1049	p->fts_fts = sp;
1050	return (p);
1051}
1052
1053static void
1054fts_lfree(FTSENT *head)
1055{
1056	FTSENT *p;
1057
1058	/* Free a linked list of structures. */
1059	while ((p = head)) {
1060		head = head->fts_link;
1061		free(p);
1062	}
1063}
1064
1065/*
1066 * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
1067 * Most systems will allow creation of paths much longer than MAXPATHLEN, even
1068 * though the kernel won't resolve them.  Add the size (not just what's needed)
1069 * plus 256 bytes so don't realloc the path 2 bytes at a time.
1070 */
1071static int
1072fts_palloc(FTS *sp, size_t more)
1073{
1074
1075	sp->fts_pathlen += more + 256;
1076	sp->fts_path = reallocf(sp->fts_path, sp->fts_pathlen);
1077	return (sp->fts_path == NULL);
1078}
1079
1080/*
1081 * When the path is realloc'd, have to fix all of the pointers in structures
1082 * already returned.
1083 */
1084static void
1085fts_padjust(FTS *sp, FTSENT *head)
1086{
1087	FTSENT *p;
1088	char *addr = sp->fts_path;
1089
1090#define	ADJUST(p) do {							\
1091	if ((p)->fts_accpath != (p)->fts_name) {			\
1092		(p)->fts_accpath =					\
1093		    (char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
1094	}								\
1095	(p)->fts_path = addr;						\
1096} while (0)
1097	/* Adjust the current set of children. */
1098	for (p = sp->fts_child; p; p = p->fts_link)
1099		ADJUST(p);
1100
1101	/* Adjust the rest of the tree, including the current level. */
1102	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1103		ADJUST(p);
1104		p = p->fts_link ? p->fts_link : p->fts_parent;
1105	}
1106}
1107
1108static size_t
1109fts_maxarglen(char * const *argv)
1110{
1111	size_t len, max;
1112
1113	for (max = 0; *argv; ++argv)
1114		if ((len = strlen(*argv)) > max)
1115			max = len;
1116	return (max + 1);
1117}
1118
1119/*
1120 * Change to dir specified by fd or p->fts_accpath without getting
1121 * tricked by someone changing the world out from underneath us.
1122 * Assumes p->fts_dev and p->fts_ino are filled in.
1123 */
1124static int
1125fts_safe_changedir(FTS *sp, FTSENT *p, int fd, char *path)
1126{
1127	int ret, oerrno, newfd;
1128	struct stat sb;
1129
1130	newfd = fd;
1131	if (ISSET(FTS_NOCHDIR))
1132		return (0);
1133	if (fd < 0 && (newfd = _open(path, O_RDONLY | O_DIRECTORY |
1134	    O_CLOEXEC, 0)) < 0)
1135		return (-1);
1136	if (_fstat(newfd, &sb)) {
1137		ret = -1;
1138		goto bail;
1139	}
1140	if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) {
1141		errno = ENOENT;		/* disinformation */
1142		ret = -1;
1143		goto bail;
1144	}
1145	ret = fchdir(newfd);
1146bail:
1147	oerrno = errno;
1148	if (fd < 0)
1149		(void)_close(newfd);
1150	errno = oerrno;
1151	return (ret);
1152}
1153
1154/*
1155 * Check if the filesystem for "ent" has UFS-style links.
1156 */
1157static int
1158fts_ufslinks(FTS *sp, const FTSENT *ent)
1159{
1160	struct _fts_private *priv;
1161	size_t fsidx = 0;
1162
1163	priv = (struct _fts_private *)sp;
1164	/*
1165	 * If this node's device is different from the previous, grab
1166	 * the filesystem information, and decide on the reliability
1167	 * of the link information from this filesystem for stat(2)
1168	 * avoidance.
1169	 */
1170	if (priv->ftsp_dev != ent->fts_dev) {
1171		if (statfs(ent->fts_path, &priv->ftsp_statfs) != -1) {
1172			priv->ftsp_dev = ent->fts_dev;
1173			priv->ftsp_linksreliable = 0;
1174			for (fsidx = 0; fsidx < sizeof(ufslike_filesystems) / sizeof(ssize_t);
1175				++fsidx) {
1176				if ((uint64_t)priv->ftsp_statfs.f_type == ufslike_filesystems[fsidx]) {
1177				    priv->ftsp_linksreliable = 1;
1178					break;
1179				}
1180			}
1181		} else {
1182			priv->ftsp_linksreliable = 0;
1183		}
1184	}
1185	return (priv->ftsp_linksreliable);
1186}
1187