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 * 4. 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 * $OpenBSD: fts.c,v 1.22 1999/10/03 19:22:22 millert Exp $
30 */
31
32#if 0
33#if defined(LIBC_SCCS) && !defined(lint)
34static char sccsid[] = "@(#)fts.c	8.6 (Berkeley) 8/14/94";
35#endif /* LIBC_SCCS and not lint */
36#endif
37
38#include <sys/cdefs.h>
39__FBSDID("$FreeBSD$");
40
41#include "namespace.h"
42#include <sys/param.h>
43#include <sys/mount.h>
44#include <sys/stat.h>
45
46#include <dirent.h>
47#include <errno.h>
48#include <fcntl.h>
49#include <stdlib.h>
50#include <string.h>
51#include <unistd.h>
52#include "fts-compat.h"
53#include "un-namespace.h"
54
55#include "gen-private.h"
56
57FTSENT	*__fts_children_44bsd(FTS *, int);
58int	 __fts_close_44bsd(FTS *);
59void	*__fts_get_clientptr_44bsd(FTS *);
60FTS	*__fts_get_stream_44bsd(FTSENT *);
61FTS	*__fts_open_44bsd(char * const *, int,
62	    int (*)(const FTSENT * const *, const FTSENT * const *));
63FTSENT	*__fts_read_44bsd(FTS *);
64int	 __fts_set_44bsd(FTS *, FTSENT *, int);
65void	 __fts_set_clientptr_44bsd(FTS *, void *);
66
67static FTSENT	*fts_alloc(FTS *, char *, int);
68static FTSENT	*fts_build(FTS *, int);
69static void	 fts_lfree(FTSENT *);
70static void	 fts_load(FTS *, FTSENT *);
71static size_t	 fts_maxarglen(char * const *);
72static void	 fts_padjust(FTS *, FTSENT *);
73static int	 fts_palloc(FTS *, size_t);
74static FTSENT	*fts_sort(FTS *, FTSENT *, int);
75static u_short	 fts_stat(FTS *, FTSENT *, int);
76static int	 fts_safe_changedir(FTS *, FTSENT *, int, char *);
77static int	 fts_ufslinks(FTS *, const FTSENT *);
78
79#define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
80
81#define	CLR(opt)	(sp->fts_options &= ~(opt))
82#define	ISSET(opt)	(sp->fts_options & (opt))
83#define	SET(opt)	(sp->fts_options |= (opt))
84
85#define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
86
87/* fts_build flags */
88#define	BCHILD		1		/* fts_children */
89#define	BNAMES		2		/* fts_children, names only */
90#define	BREAD		3		/* fts_read */
91
92/*
93 * Internal representation of an FTS, including extra implementation
94 * details.  The FTS returned from fts_open points to this structure's
95 * ftsp_fts member (and can be cast to an _fts_private as required)
96 */
97struct _fts_private {
98	FTS		ftsp_fts;
99	struct statfs	ftsp_statfs;
100	dev_t		ftsp_dev;
101	int		ftsp_linksreliable;
102};
103
104/*
105 * The "FTS_NOSTAT" option can avoid a lot of calls to stat(2) if it
106 * knows that a directory could not possibly have subdirectories.  This
107 * is decided by looking at the link count: a subdirectory would
108 * increment its parent's link count by virtue of its own ".." entry.
109 * This assumption only holds for UFS-like filesystems that implement
110 * links and directories this way, so we must punt for others.
111 */
112
113static const char *ufslike_filesystems[] = {
114	"ufs",
115	"zfs",
116	"nfs",
117	"nfs4",
118	"ext2fs",
119	0
120};
121
122FTS *
123__fts_open_44bsd(argv, options, compar)
124	char * const *argv;
125	int options;
126	int (*compar)(const FTSENT * const *, const FTSENT * const *);
127{
128	struct _fts_private *priv;
129	FTS *sp;
130	FTSENT *p, *root;
131	int nitems;
132	FTSENT *parent, *tmp;
133	int len;
134
135	/* Options check. */
136	if (options & ~FTS_OPTIONMASK) {
137		errno = EINVAL;
138		return (NULL);
139	}
140
141	/* Allocate/initialize the stream. */
142	if ((priv = malloc(sizeof(*priv))) == NULL)
143		return (NULL);
144	memset(priv, 0, sizeof(*priv));
145	sp = &priv->ftsp_fts;
146	sp->fts_compar = compar;
147	sp->fts_options = options;
148
149	/* Shush, GCC. */
150	tmp = NULL;
151
152	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
153	if (ISSET(FTS_LOGICAL))
154		SET(FTS_NOCHDIR);
155
156	/*
157	 * Start out with 1K of path space, and enough, in any case,
158	 * to hold the user's paths.
159	 */
160	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
161		goto mem1;
162
163	/* Allocate/initialize root's parent. */
164	if ((parent = fts_alloc(sp, "", 0)) == NULL)
165		goto mem2;
166	parent->fts_level = FTS_ROOTPARENTLEVEL;
167
168	/* Allocate/initialize root(s). */
169	for (root = NULL, nitems = 0; *argv != NULL; ++argv, ++nitems) {
170		/* Don't allow zero-length paths. */
171		if ((len = strlen(*argv)) == 0) {
172			errno = ENOENT;
173			goto mem3;
174		}
175
176		p = fts_alloc(sp, *argv, len);
177		p->fts_level = FTS_ROOTLEVEL;
178		p->fts_parent = parent;
179		p->fts_accpath = p->fts_name;
180		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW));
181
182		/* Command-line "." and ".." are real directories. */
183		if (p->fts_info == FTS_DOT)
184			p->fts_info = FTS_D;
185
186		/*
187		 * If comparison routine supplied, traverse in sorted
188		 * order; otherwise traverse in the order specified.
189		 */
190		if (compar) {
191			p->fts_link = root;
192			root = p;
193		} else {
194			p->fts_link = NULL;
195			if (root == NULL)
196				tmp = root = p;
197			else {
198				tmp->fts_link = p;
199				tmp = p;
200			}
201		}
202	}
203	if (compar && nitems > 1)
204		root = fts_sort(sp, root, nitems);
205
206	/*
207	 * Allocate a dummy pointer and make fts_read think that we've just
208	 * finished the node before the root(s); set p->fts_info to FTS_INIT
209	 * so that everything about the "current" node is ignored.
210	 */
211	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
212		goto mem3;
213	sp->fts_cur->fts_link = root;
214	sp->fts_cur->fts_info = FTS_INIT;
215
216	/*
217	 * If using chdir(2), grab a file descriptor pointing to dot to ensure
218	 * that we can get back here; this could be avoided for some paths,
219	 * but almost certainly not worth the effort.  Slashes, symbolic links,
220	 * and ".." are all fairly nasty problems.  Note, if we can't get the
221	 * descriptor we run anyway, just more slowly.
222	 */
223	if (!ISSET(FTS_NOCHDIR) &&
224	    (sp->fts_rfd = _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0)
225		SET(FTS_NOCHDIR);
226
227	return (sp);
228
229mem3:	fts_lfree(root);
230	free(parent);
231mem2:	free(sp->fts_path);
232mem1:	free(sp);
233	return (NULL);
234}
235
236static void
237fts_load(sp, p)
238	FTS *sp;
239	FTSENT *p;
240{
241	int len;
242	char *cp;
243
244	/*
245	 * Load the stream structure for the next traversal.  Since we don't
246	 * actually enter the directory until after the preorder visit, set
247	 * the fts_accpath field specially so the chdir gets done to the right
248	 * place and the user can access the first node.  From fts_open it's
249	 * known that the path will fit.
250	 */
251	len = p->fts_pathlen = p->fts_namelen;
252	memmove(sp->fts_path, p->fts_name, len + 1);
253	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
254		len = strlen(++cp);
255		memmove(p->fts_name, cp, len + 1);
256		p->fts_namelen = len;
257	}
258	p->fts_accpath = p->fts_path = sp->fts_path;
259	sp->fts_dev = p->fts_dev;
260}
261
262int
263__fts_close_44bsd(sp)
264	FTS *sp;
265{
266	FTSENT *freep, *p;
267	int saved_errno;
268
269	/*
270	 * This still works if we haven't read anything -- the dummy structure
271	 * points to the root list, so we step through to the end of the root
272	 * list which has a valid parent pointer.
273	 */
274	if (sp->fts_cur) {
275		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
276			freep = p;
277			p = p->fts_link != NULL ? p->fts_link : p->fts_parent;
278			free(freep);
279		}
280		free(p);
281	}
282
283	/* Free up child linked list, sort array, path buffer. */
284	if (sp->fts_child)
285		fts_lfree(sp->fts_child);
286	if (sp->fts_array)
287		free(sp->fts_array);
288	free(sp->fts_path);
289
290	/* Return to original directory, save errno if necessary. */
291	if (!ISSET(FTS_NOCHDIR)) {
292		saved_errno = fchdir(sp->fts_rfd) ? errno : 0;
293		(void)_close(sp->fts_rfd);
294
295		/* Set errno and return. */
296		if (saved_errno != 0) {
297			/* Free up the stream pointer. */
298			free(sp);
299			errno = saved_errno;
300			return (-1);
301		}
302	}
303
304	/* Free up the stream pointer. */
305	free(sp);
306	return (0);
307}
308
309/*
310 * Special case of "/" at the end of the path so that slashes aren't
311 * appended which would cause paths to be written as "....//foo".
312 */
313#define	NAPPEND(p)							\
314	(p->fts_path[p->fts_pathlen - 1] == '/'				\
315	    ? p->fts_pathlen - 1 : p->fts_pathlen)
316
317FTSENT *
318__fts_read_44bsd(sp)
319	FTS *sp;
320{
321	FTSENT *p, *tmp;
322	int instr;
323	char *t;
324	int saved_errno;
325
326	/* If finished or unrecoverable error, return NULL. */
327	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
328		return (NULL);
329
330	/* Set current node pointer. */
331	p = sp->fts_cur;
332
333	/* Save and zero out user instructions. */
334	instr = p->fts_instr;
335	p->fts_instr = FTS_NOINSTR;
336
337	/* Any type of file may be re-visited; re-stat and re-turn. */
338	if (instr == FTS_AGAIN) {
339		p->fts_info = fts_stat(sp, p, 0);
340		return (p);
341	}
342
343	/*
344	 * Following a symlink -- SLNONE test allows application to see
345	 * SLNONE and recover.  If indirecting through a symlink, have
346	 * keep a pointer to current location.  If unable to get that
347	 * pointer, follow fails.
348	 */
349	if (instr == FTS_FOLLOW &&
350	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
351		p->fts_info = fts_stat(sp, p, 1);
352		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
353			if ((p->fts_symfd = _open(".", O_RDONLY | O_CLOEXEC,
354			    0)) < 0) {
355				p->fts_errno = errno;
356				p->fts_info = FTS_ERR;
357			} else
358				p->fts_flags |= FTS_SYMFOLLOW;
359		}
360		return (p);
361	}
362
363	/* Directory in pre-order. */
364	if (p->fts_info == FTS_D) {
365		/* If skipped or crossed mount point, do post-order visit. */
366		if (instr == FTS_SKIP ||
367		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
368			if (p->fts_flags & FTS_SYMFOLLOW)
369				(void)_close(p->fts_symfd);
370			if (sp->fts_child) {
371				fts_lfree(sp->fts_child);
372				sp->fts_child = NULL;
373			}
374			p->fts_info = FTS_DP;
375			return (p);
376		}
377
378		/* Rebuild if only read the names and now traversing. */
379		if (sp->fts_child != NULL && ISSET(FTS_NAMEONLY)) {
380			CLR(FTS_NAMEONLY);
381			fts_lfree(sp->fts_child);
382			sp->fts_child = NULL;
383		}
384
385		/*
386		 * Cd to the subdirectory.
387		 *
388		 * If have already read and now fail to chdir, whack the list
389		 * to make the names come out right, and set the parent errno
390		 * so the application will eventually get an error condition.
391		 * Set the FTS_DONTCHDIR flag so that when we logically change
392		 * directories back to the parent we don't do a chdir.
393		 *
394		 * If haven't read do so.  If the read fails, fts_build sets
395		 * FTS_STOP or the fts_info field of the node.
396		 */
397		if (sp->fts_child != NULL) {
398			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
399				p->fts_errno = errno;
400				p->fts_flags |= FTS_DONTCHDIR;
401				for (p = sp->fts_child; p != NULL;
402				    p = p->fts_link)
403					p->fts_accpath =
404					    p->fts_parent->fts_accpath;
405			}
406		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
407			if (ISSET(FTS_STOP))
408				return (NULL);
409			return (p);
410		}
411		p = sp->fts_child;
412		sp->fts_child = NULL;
413		goto name;
414	}
415
416	/* Move to the next node on this level. */
417next:	tmp = p;
418	if ((p = p->fts_link) != NULL) {
419		free(tmp);
420
421		/*
422		 * If reached the top, return to the original directory (or
423		 * the root of the tree), and load the paths for the next root.
424		 */
425		if (p->fts_level == FTS_ROOTLEVEL) {
426			if (FCHDIR(sp, sp->fts_rfd)) {
427				SET(FTS_STOP);
428				return (NULL);
429			}
430			fts_load(sp, p);
431			return (sp->fts_cur = p);
432		}
433
434		/*
435		 * User may have called fts_set on the node.  If skipped,
436		 * ignore.  If followed, get a file descriptor so we can
437		 * get back if necessary.
438		 */
439		if (p->fts_instr == FTS_SKIP)
440			goto next;
441		if (p->fts_instr == FTS_FOLLOW) {
442			p->fts_info = fts_stat(sp, p, 1);
443			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
444				if ((p->fts_symfd =
445				    _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0) {
446					p->fts_errno = errno;
447					p->fts_info = FTS_ERR;
448				} else
449					p->fts_flags |= FTS_SYMFOLLOW;
450			}
451			p->fts_instr = FTS_NOINSTR;
452		}
453
454name:		t = sp->fts_path + NAPPEND(p->fts_parent);
455		*t++ = '/';
456		memmove(t, p->fts_name, p->fts_namelen + 1);
457		return (sp->fts_cur = p);
458	}
459
460	/* Move up to the parent node. */
461	p = tmp->fts_parent;
462	free(tmp);
463
464	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
465		/*
466		 * Done; free everything up and set errno to 0 so the user
467		 * can distinguish between error and EOF.
468		 */
469		free(p);
470		errno = 0;
471		return (sp->fts_cur = NULL);
472	}
473
474	/* NUL terminate the pathname. */
475	sp->fts_path[p->fts_pathlen] = '\0';
476
477	/*
478	 * Return to the parent directory.  If at a root node or came through
479	 * a symlink, go back through the file descriptor.  Otherwise, cd up
480	 * one directory.
481	 */
482	if (p->fts_level == FTS_ROOTLEVEL) {
483		if (FCHDIR(sp, sp->fts_rfd)) {
484			SET(FTS_STOP);
485			return (NULL);
486		}
487	} else if (p->fts_flags & FTS_SYMFOLLOW) {
488		if (FCHDIR(sp, p->fts_symfd)) {
489			saved_errno = errno;
490			(void)_close(p->fts_symfd);
491			errno = saved_errno;
492			SET(FTS_STOP);
493			return (NULL);
494		}
495		(void)_close(p->fts_symfd);
496	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
497	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
498		SET(FTS_STOP);
499		return (NULL);
500	}
501	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
502	return (sp->fts_cur = p);
503}
504
505/*
506 * Fts_set takes the stream as an argument although it's not used in this
507 * implementation; it would be necessary if anyone wanted to add global
508 * semantics to fts using fts_set.  An error return is allowed for similar
509 * reasons.
510 */
511/* ARGSUSED */
512int
513__fts_set_44bsd(sp, p, instr)
514	FTS *sp;
515	FTSENT *p;
516	int instr;
517{
518	if (instr != 0 && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
519	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
520		errno = EINVAL;
521		return (1);
522	}
523	p->fts_instr = instr;
524	return (0);
525}
526
527FTSENT *
528__fts_children_44bsd(sp, instr)
529	FTS *sp;
530	int instr;
531{
532	FTSENT *p;
533	int fd;
534
535	if (instr != 0 && instr != FTS_NAMEONLY) {
536		errno = EINVAL;
537		return (NULL);
538	}
539
540	/* Set current node pointer. */
541	p = sp->fts_cur;
542
543	/*
544	 * Errno set to 0 so user can distinguish empty directory from
545	 * an error.
546	 */
547	errno = 0;
548
549	/* Fatal errors stop here. */
550	if (ISSET(FTS_STOP))
551		return (NULL);
552
553	/* Return logical hierarchy of user's arguments. */
554	if (p->fts_info == FTS_INIT)
555		return (p->fts_link);
556
557	/*
558	 * If not a directory being visited in pre-order, stop here.  Could
559	 * allow FTS_DNR, assuming the user has fixed the problem, but the
560	 * same effect is available with FTS_AGAIN.
561	 */
562	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
563		return (NULL);
564
565	/* Free up any previous child list. */
566	if (sp->fts_child != NULL)
567		fts_lfree(sp->fts_child);
568
569	if (instr == FTS_NAMEONLY) {
570		SET(FTS_NAMEONLY);
571		instr = BNAMES;
572	} else
573		instr = BCHILD;
574
575	/*
576	 * If using chdir on a relative path and called BEFORE fts_read does
577	 * its chdir to the root of a traversal, we can lose -- we need to
578	 * chdir into the subdirectory, and we don't know where the current
579	 * directory is, so we can't get back so that the upcoming chdir by
580	 * fts_read will work.
581	 */
582	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
583	    ISSET(FTS_NOCHDIR))
584		return (sp->fts_child = fts_build(sp, instr));
585
586	if ((fd = _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0)
587		return (NULL);
588	sp->fts_child = fts_build(sp, instr);
589	if (fchdir(fd))
590		return (NULL);
591	(void)_close(fd);
592	return (sp->fts_child);
593}
594
595#ifndef fts_get_clientptr
596#error "fts_get_clientptr not defined"
597#endif
598
599void *
600(__fts_get_clientptr_44bsd)(FTS *sp)
601{
602
603	return (fts_get_clientptr(sp));
604}
605
606#ifndef fts_get_stream
607#error "fts_get_stream not defined"
608#endif
609
610FTS *
611(__fts_get_stream_44bsd)(FTSENT *p)
612{
613	return (fts_get_stream(p));
614}
615
616void
617__fts_set_clientptr_44bsd(FTS *sp, void *clientptr)
618{
619
620	sp->fts_clientptr = clientptr;
621}
622
623/*
624 * This is the tricky part -- do not casually change *anything* in here.  The
625 * idea is to build the linked list of entries that are used by fts_children
626 * and fts_read.  There are lots of special cases.
627 *
628 * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
629 * set and it's a physical walk (so that symbolic links can't be directories),
630 * we can do things quickly.  First, if it's a 4.4BSD file system, the type
631 * of the file is in the directory entry.  Otherwise, we assume that the number
632 * of subdirectories in a node is equal to the number of links to the parent.
633 * The former skips all stat calls.  The latter skips stat calls in any leaf
634 * directories and for any files after the subdirectories in the directory have
635 * been found, cutting the stat calls by about 2/3.
636 */
637static FTSENT *
638fts_build(sp, type)
639	FTS *sp;
640	int type;
641{
642	struct dirent *dp;
643	FTSENT *p, *head;
644	int nitems;
645	FTSENT *cur, *tail;
646	DIR *dirp;
647	void *oldaddr;
648	size_t dnamlen;
649	int cderrno, descend, len, level, maxlen, nlinks, oflag, saved_errno,
650	    nostat, doadjust;
651	char *cp;
652
653	/* Set current node pointer. */
654	cur = sp->fts_cur;
655
656	/*
657	 * Open the directory for reading.  If this fails, we're done.
658	 * If being called from fts_read, set the fts_info field.
659	 */
660#ifdef FTS_WHITEOUT
661	if (ISSET(FTS_WHITEOUT))
662		oflag = DTF_NODUP | DTF_REWIND;
663	else
664		oflag = DTF_HIDEW | DTF_NODUP | DTF_REWIND;
665#else
666#define __opendir2(path, flag) opendir(path)
667#endif
668	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
669		if (type == BREAD) {
670			cur->fts_info = FTS_DNR;
671			cur->fts_errno = errno;
672		}
673		return (NULL);
674	}
675
676	/*
677	 * Nlinks is the number of possible entries of type directory in the
678	 * directory if we're cheating on stat calls, 0 if we're not doing
679	 * any stat calls at all, -1 if we're doing stats on everything.
680	 */
681	if (type == BNAMES) {
682		nlinks = 0;
683		/* Be quiet about nostat, GCC. */
684		nostat = 0;
685	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
686		if (fts_ufslinks(sp, cur))
687			nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
688		else
689			nlinks = -1;
690		nostat = 1;
691	} else {
692		nlinks = -1;
693		nostat = 0;
694	}
695
696#ifdef notdef
697	(void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
698	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
699	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
700#endif
701	/*
702	 * If we're going to need to stat anything or we want to descend
703	 * and stay in the directory, chdir.  If this fails we keep going,
704	 * but set a flag so we don't chdir after the post-order visit.
705	 * We won't be able to stat anything, but we can still return the
706	 * names themselves.  Note, that since fts_read won't be able to
707	 * chdir into the directory, it will have to return different path
708	 * names than before, i.e. "a/b" instead of "b".  Since the node
709	 * has already been visited in pre-order, have to wait until the
710	 * post-order visit to return the error.  There is a special case
711	 * here, if there was nothing to stat then it's not an error to
712	 * not be able to stat.  This is all fairly nasty.  If a program
713	 * needed sorted entries or stat information, they had better be
714	 * checking FTS_NS on the returned nodes.
715	 */
716	cderrno = 0;
717	if (nlinks || type == BREAD) {
718		if (fts_safe_changedir(sp, cur, _dirfd(dirp), NULL)) {
719			if (nlinks && type == BREAD)
720				cur->fts_errno = errno;
721			cur->fts_flags |= FTS_DONTCHDIR;
722			descend = 0;
723			cderrno = errno;
724		} else
725			descend = 1;
726	} else
727		descend = 0;
728
729	/*
730	 * Figure out the max file name length that can be stored in the
731	 * current path -- the inner loop allocates more path as necessary.
732	 * We really wouldn't have to do the maxlen calculations here, we
733	 * could do them in fts_read before returning the path, but it's a
734	 * lot easier here since the length is part of the dirent structure.
735	 *
736	 * If not changing directories set a pointer so that can just append
737	 * each new name into the path.
738	 */
739	len = NAPPEND(cur);
740	if (ISSET(FTS_NOCHDIR)) {
741		cp = sp->fts_path + len;
742		*cp++ = '/';
743	} else {
744		/* GCC, you're too verbose. */
745		cp = NULL;
746	}
747	len++;
748	maxlen = sp->fts_pathlen - len;
749
750	level = cur->fts_level + 1;
751
752	/* Read the directory, attaching each entry to the `link' pointer. */
753	doadjust = 0;
754	for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) {
755		dnamlen = dp->d_namlen;
756		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
757			continue;
758
759		if ((p = fts_alloc(sp, dp->d_name, (int)dnamlen)) == NULL)
760			goto mem1;
761		if (dnamlen >= maxlen) {	/* include space for NUL */
762			oldaddr = sp->fts_path;
763			if (fts_palloc(sp, dnamlen + len + 1)) {
764				/*
765				 * No more memory for path or structures.  Save
766				 * errno, free up the current structure and the
767				 * structures already allocated.
768				 */
769mem1:				saved_errno = errno;
770				if (p)
771					free(p);
772				fts_lfree(head);
773				(void)closedir(dirp);
774				cur->fts_info = FTS_ERR;
775				SET(FTS_STOP);
776				errno = saved_errno;
777				return (NULL);
778			}
779			/* Did realloc() change the pointer? */
780			if (oldaddr != sp->fts_path) {
781				doadjust = 1;
782				if (ISSET(FTS_NOCHDIR))
783					cp = sp->fts_path + len;
784			}
785			maxlen = sp->fts_pathlen - len;
786		}
787
788		if (len + dnamlen >= USHRT_MAX) {
789			/*
790			 * In an FTSENT, fts_pathlen is a u_short so it is
791			 * possible to wraparound here.  If we do, free up
792			 * the current structure and the structures already
793			 * allocated, then error out with ENAMETOOLONG.
794			 */
795			free(p);
796			fts_lfree(head);
797			(void)closedir(dirp);
798			cur->fts_info = FTS_ERR;
799			SET(FTS_STOP);
800			errno = ENAMETOOLONG;
801			return (NULL);
802		}
803		p->fts_level = level;
804		p->fts_parent = sp->fts_cur;
805		p->fts_pathlen = len + dnamlen;
806
807#ifdef FTS_WHITEOUT
808		if (dp->d_type == DT_WHT)
809			p->fts_flags |= FTS_ISW;
810#endif
811
812		if (cderrno) {
813			if (nlinks) {
814				p->fts_info = FTS_NS;
815				p->fts_errno = cderrno;
816			} else
817				p->fts_info = FTS_NSOK;
818			p->fts_accpath = cur->fts_accpath;
819		} else if (nlinks == 0
820#ifdef DT_DIR
821		    || (nostat &&
822		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
823#endif
824		    ) {
825			p->fts_accpath =
826			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
827			p->fts_info = FTS_NSOK;
828		} else {
829			/* Build a file name for fts_stat to stat. */
830			if (ISSET(FTS_NOCHDIR)) {
831				p->fts_accpath = p->fts_path;
832				memmove(cp, p->fts_name, p->fts_namelen + 1);
833			} else
834				p->fts_accpath = p->fts_name;
835			/* Stat it. */
836			p->fts_info = fts_stat(sp, p, 0);
837
838			/* Decrement link count if applicable. */
839			if (nlinks > 0 && (p->fts_info == FTS_D ||
840			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
841				--nlinks;
842		}
843
844		/* We walk in directory order so "ls -f" doesn't get upset. */
845		p->fts_link = NULL;
846		if (head == NULL)
847			head = tail = p;
848		else {
849			tail->fts_link = p;
850			tail = p;
851		}
852		++nitems;
853	}
854	if (dirp)
855		(void)closedir(dirp);
856
857	/*
858	 * If realloc() changed the address of the path, adjust the
859	 * addresses for the rest of the tree and the dir list.
860	 */
861	if (doadjust)
862		fts_padjust(sp, head);
863
864	/*
865	 * If not changing directories, reset the path back to original
866	 * state.
867	 */
868	if (ISSET(FTS_NOCHDIR)) {
869		if (len == sp->fts_pathlen || nitems == 0)
870			--cp;
871		*cp = '\0';
872	}
873
874	/*
875	 * If descended after called from fts_children or after called from
876	 * fts_read and nothing found, get back.  At the root level we use
877	 * the saved fd; if one of fts_open()'s arguments is a relative path
878	 * to an empty directory, we wind up here with no other way back.  If
879	 * can't get back, we're done.
880	 */
881	if (descend && (type == BCHILD || !nitems) &&
882	    (cur->fts_level == FTS_ROOTLEVEL ?
883	    FCHDIR(sp, sp->fts_rfd) :
884	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
885		cur->fts_info = FTS_ERR;
886		SET(FTS_STOP);
887		return (NULL);
888	}
889
890	/* If didn't find anything, return NULL. */
891	if (!nitems) {
892		if (type == BREAD)
893			cur->fts_info = FTS_DP;
894		return (NULL);
895	}
896
897	/* Sort the entries. */
898	if (sp->fts_compar && nitems > 1)
899		head = fts_sort(sp, head, nitems);
900	return (head);
901}
902
903static u_short
904fts_stat(sp, p, follow)
905	FTS *sp;
906	FTSENT *p;
907	int follow;
908{
909	FTSENT *t;
910	dev_t dev;
911	ino_t ino;
912	struct stat *sbp, sb;
913	int saved_errno;
914
915	/* If user needs stat info, stat buffer already allocated. */
916	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
917
918#ifdef FTS_WHITEOUT
919	/* Check for whiteout. */
920	if (p->fts_flags & FTS_ISW) {
921		if (sbp != &sb) {
922			memset(sbp, '\0', sizeof(*sbp));
923			sbp->st_mode = S_IFWHT;
924		}
925		return (FTS_W);
926	}
927#endif
928
929	/*
930	 * If doing a logical walk, or application requested FTS_FOLLOW, do
931	 * a stat(2).  If that fails, check for a non-existent symlink.  If
932	 * fail, set the errno from the stat call.
933	 */
934	if (ISSET(FTS_LOGICAL) || follow) {
935		if (stat(p->fts_accpath, sbp)) {
936			saved_errno = errno;
937			if (!lstat(p->fts_accpath, sbp)) {
938				errno = 0;
939				return (FTS_SLNONE);
940			}
941			p->fts_errno = saved_errno;
942			goto err;
943		}
944	} else if (lstat(p->fts_accpath, sbp)) {
945		p->fts_errno = errno;
946err:		memset(sbp, 0, sizeof(struct stat));
947		return (FTS_NS);
948	}
949
950	if (S_ISDIR(sbp->st_mode)) {
951		/*
952		 * Set the device/inode.  Used to find cycles and check for
953		 * crossing mount points.  Also remember the link count, used
954		 * in fts_build to limit the number of stat calls.  It is
955		 * understood that these fields are only referenced if fts_info
956		 * is set to FTS_D.
957		 */
958		dev = p->fts_dev = sbp->st_dev;
959		ino = p->fts_ino = sbp->st_ino;
960		p->fts_nlink = sbp->st_nlink;
961
962		if (ISDOT(p->fts_name))
963			return (FTS_DOT);
964
965		/*
966		 * Cycle detection is done by brute force when the directory
967		 * is first encountered.  If the tree gets deep enough or the
968		 * number of symbolic links to directories is high enough,
969		 * something faster might be worthwhile.
970		 */
971		for (t = p->fts_parent;
972		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
973			if (ino == t->fts_ino && dev == t->fts_dev) {
974				p->fts_cycle = t;
975				return (FTS_DC);
976			}
977		return (FTS_D);
978	}
979	if (S_ISLNK(sbp->st_mode))
980		return (FTS_SL);
981	if (S_ISREG(sbp->st_mode))
982		return (FTS_F);
983	return (FTS_DEFAULT);
984}
985
986/*
987 * The comparison function takes pointers to pointers to FTSENT structures.
988 * Qsort wants a comparison function that takes pointers to void.
989 * (Both with appropriate levels of const-poisoning, of course!)
990 * Use a trampoline function to deal with the difference.
991 */
992static int
993fts_compar(const void *a, const void *b)
994{
995	FTS *parent;
996
997	parent = (*(const FTSENT * const *)a)->fts_fts;
998	return (*parent->fts_compar)(a, b);
999}
1000
1001static FTSENT *
1002fts_sort(sp, head, nitems)
1003	FTS *sp;
1004	FTSENT *head;
1005	int nitems;
1006{
1007	FTSENT **ap, *p;
1008
1009	/*
1010	 * Construct an array of pointers to the structures and call qsort(3).
1011	 * Reassemble the array in the order returned by qsort.  If unable to
1012	 * sort for memory reasons, return the directory entries in their
1013	 * current order.  Allocate enough space for the current needs plus
1014	 * 40 so don't realloc one entry at a time.
1015	 */
1016	if (nitems > sp->fts_nitems) {
1017		sp->fts_nitems = nitems + 40;
1018		if ((sp->fts_array = reallocf(sp->fts_array,
1019		    sp->fts_nitems * sizeof(FTSENT *))) == NULL) {
1020			sp->fts_nitems = 0;
1021			return (head);
1022		}
1023	}
1024	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
1025		*ap++ = p;
1026	qsort(sp->fts_array, nitems, sizeof(FTSENT *), fts_compar);
1027	for (head = *(ap = sp->fts_array); --nitems; ++ap)
1028		ap[0]->fts_link = ap[1];
1029	ap[0]->fts_link = NULL;
1030	return (head);
1031}
1032
1033static FTSENT *
1034fts_alloc(sp, name, namelen)
1035	FTS *sp;
1036	char *name;
1037	int namelen;
1038{
1039	FTSENT *p;
1040	size_t len;
1041
1042	struct ftsent_withstat {
1043		FTSENT	ent;
1044		struct	stat statbuf;
1045	};
1046
1047	/*
1048	 * The file name is a variable length array and no stat structure is
1049	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
1050	 * structure, the file name and the stat structure in one chunk, but
1051	 * be careful that the stat structure is reasonably aligned.
1052	 */
1053	if (ISSET(FTS_NOSTAT))
1054		len = sizeof(FTSENT) + namelen + 1;
1055	else
1056		len = sizeof(struct ftsent_withstat) + namelen + 1;
1057
1058	if ((p = malloc(len)) == NULL)
1059		return (NULL);
1060
1061	if (ISSET(FTS_NOSTAT)) {
1062		p->fts_name = (char *)(p + 1);
1063		p->fts_statp = NULL;
1064	} else {
1065		p->fts_name = (char *)((struct ftsent_withstat *)p + 1);
1066		p->fts_statp = &((struct ftsent_withstat *)p)->statbuf;
1067	}
1068
1069	/* Copy the name and guarantee NUL termination. */
1070	memcpy(p->fts_name, name, namelen);
1071	p->fts_name[namelen] = '\0';
1072	p->fts_namelen = namelen;
1073	p->fts_path = sp->fts_path;
1074	p->fts_errno = 0;
1075	p->fts_flags = 0;
1076	p->fts_instr = FTS_NOINSTR;
1077	p->fts_number = 0;
1078	p->fts_pointer = NULL;
1079	p->fts_fts = sp;
1080	return (p);
1081}
1082
1083static void
1084fts_lfree(head)
1085	FTSENT *head;
1086{
1087	FTSENT *p;
1088
1089	/* Free a linked list of structures. */
1090	while ((p = head)) {
1091		head = head->fts_link;
1092		free(p);
1093	}
1094}
1095
1096/*
1097 * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
1098 * Most systems will allow creation of paths much longer than MAXPATHLEN, even
1099 * though the kernel won't resolve them.  Add the size (not just what's needed)
1100 * plus 256 bytes so don't realloc the path 2 bytes at a time.
1101 */
1102static int
1103fts_palloc(sp, more)
1104	FTS *sp;
1105	size_t more;
1106{
1107
1108	sp->fts_pathlen += more + 256;
1109	/*
1110	 * Check for possible wraparound.  In an FTS, fts_pathlen is
1111	 * a signed int but in an FTSENT it is an unsigned short.
1112	 * We limit fts_pathlen to USHRT_MAX to be safe in both cases.
1113	 */
1114	if (sp->fts_pathlen < 0 || sp->fts_pathlen >= USHRT_MAX) {
1115		if (sp->fts_path)
1116			free(sp->fts_path);
1117		sp->fts_path = NULL;
1118		errno = ENAMETOOLONG;
1119		return (1);
1120	}
1121	sp->fts_path = reallocf(sp->fts_path, sp->fts_pathlen);
1122	return (sp->fts_path == NULL);
1123}
1124
1125/*
1126 * When the path is realloc'd, have to fix all of the pointers in structures
1127 * already returned.
1128 */
1129static void
1130fts_padjust(sp, head)
1131	FTS *sp;
1132	FTSENT *head;
1133{
1134	FTSENT *p;
1135	char *addr = sp->fts_path;
1136
1137#define	ADJUST(p) do {							\
1138	if ((p)->fts_accpath != (p)->fts_name) {			\
1139		(p)->fts_accpath =					\
1140		    (char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
1141	}								\
1142	(p)->fts_path = addr;						\
1143} while (0)
1144	/* Adjust the current set of children. */
1145	for (p = sp->fts_child; p; p = p->fts_link)
1146		ADJUST(p);
1147
1148	/* Adjust the rest of the tree, including the current level. */
1149	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1150		ADJUST(p);
1151		p = p->fts_link ? p->fts_link : p->fts_parent;
1152	}
1153}
1154
1155static size_t
1156fts_maxarglen(argv)
1157	char * const *argv;
1158{
1159	size_t len, max;
1160
1161	for (max = 0; *argv; ++argv)
1162		if ((len = strlen(*argv)) > max)
1163			max = len;
1164	return (max + 1);
1165}
1166
1167/*
1168 * Change to dir specified by fd or p->fts_accpath without getting
1169 * tricked by someone changing the world out from underneath us.
1170 * Assumes p->fts_dev and p->fts_ino are filled in.
1171 */
1172static int
1173fts_safe_changedir(sp, p, fd, path)
1174	FTS *sp;
1175	FTSENT *p;
1176	int fd;
1177	char *path;
1178{
1179	int ret, oerrno, newfd;
1180	struct stat sb;
1181
1182	newfd = fd;
1183	if (ISSET(FTS_NOCHDIR))
1184		return (0);
1185	if (fd < 0 && (newfd = _open(path, O_RDONLY | O_CLOEXEC, 0)) < 0)
1186		return (-1);
1187	if (_fstat(newfd, &sb)) {
1188		ret = -1;
1189		goto bail;
1190	}
1191	if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) {
1192		errno = ENOENT;		/* disinformation */
1193		ret = -1;
1194		goto bail;
1195	}
1196	ret = fchdir(newfd);
1197bail:
1198	oerrno = errno;
1199	if (fd < 0)
1200		(void)_close(newfd);
1201	errno = oerrno;
1202	return (ret);
1203}
1204
1205/*
1206 * Check if the filesystem for "ent" has UFS-style links.
1207 */
1208static int
1209fts_ufslinks(FTS *sp, const FTSENT *ent)
1210{
1211	struct _fts_private *priv;
1212	const char **cpp;
1213
1214	priv = (struct _fts_private *)sp;
1215	/*
1216	 * If this node's device is different from the previous, grab
1217	 * the filesystem information, and decide on the reliability
1218	 * of the link information from this filesystem for stat(2)
1219	 * avoidance.
1220	 */
1221	if (priv->ftsp_dev != ent->fts_dev) {
1222		if (statfs(ent->fts_path, &priv->ftsp_statfs) != -1) {
1223			priv->ftsp_dev = ent->fts_dev;
1224			priv->ftsp_linksreliable = 0;
1225			for (cpp = ufslike_filesystems; *cpp; cpp++) {
1226				if (strcmp(priv->ftsp_statfs.f_fstypename,
1227				    *cpp) == 0) {
1228					priv->ftsp_linksreliable = 1;
1229					break;
1230				}
1231			}
1232		} else {
1233			priv->ftsp_linksreliable = 0;
1234		}
1235	}
1236	return (priv->ftsp_linksreliable);
1237}
1238
1239__sym_compat(fts_open, __fts_open_44bsd, FBSD_1.0);
1240__sym_compat(fts_close, __fts_close_44bsd, FBSD_1.0);
1241__sym_compat(fts_read, __fts_read_44bsd, FBSD_1.0);
1242__sym_compat(fts_set, __fts_set_44bsd, FBSD_1.0);
1243__sym_compat(fts_children, __fts_children_44bsd, FBSD_1.0);
1244__sym_compat(fts_get_clientptr, __fts_get_clientptr_44bsd, FBSD_1.0);
1245__sym_compat(fts_get_stream, __fts_get_stream_44bsd, FBSD_1.0);
1246__sym_compat(fts_set_clientptr, __fts_set_clientptr_44bsd, FBSD_1.0);
1247