fts.c revision 165903
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: head/lib/libc/gen/fts.c 165903 2007-01-09 00:28:16Z imp $");
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 <fts.h>
50#include <stdlib.h>
51#include <string.h>
52#include <unistd.h>
53#include "un-namespace.h"
54
55static FTSENT	*fts_alloc(FTS *, char *, int);
56static FTSENT	*fts_build(FTS *, int);
57static void	 fts_lfree(FTSENT *);
58static void	 fts_load(FTS *, FTSENT *);
59static size_t	 fts_maxarglen(char * const *);
60static void	 fts_padjust(FTS *, FTSENT *);
61static int	 fts_palloc(FTS *, size_t);
62static FTSENT	*fts_sort(FTS *, FTSENT *, int);
63static u_short	 fts_stat(FTS *, FTSENT *, int);
64static int	 fts_safe_changedir(FTS *, FTSENT *, int, char *);
65static int	 fts_ufslinks(FTS *, const FTSENT *);
66
67#define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
68
69#define	CLR(opt)	(sp->fts_options &= ~(opt))
70#define	ISSET(opt)	(sp->fts_options & (opt))
71#define	SET(opt)	(sp->fts_options |= (opt))
72
73#define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
74
75/* fts_build flags */
76#define	BCHILD		1		/* fts_children */
77#define	BNAMES		2		/* fts_children, names only */
78#define	BREAD		3		/* fts_read */
79
80/*
81 * Internal representation of an FTS, including extra implementation
82 * details.  The FTS returned from fts_open points to this structure's
83 * ftsp_fts member (and can be cast to an _fts_private as required)
84 */
85struct _fts_private {
86	FTS		ftsp_fts;
87	struct statfs	ftsp_statfs;
88	dev_t		ftsp_dev;
89	int		ftsp_linksreliable;
90};
91
92/*
93 * The "FTS_NOSTAT" option can avoid a lot of calls to stat(2) if it
94 * knows that a directory could not possibly have subdirectories.  This
95 * is decided by looking at the link count: a subdirectory would
96 * increment its parent's link count by virtue of its own ".." entry.
97 * This assumption only holds for UFS-like filesystems that implement
98 * links and directories this way, so we must punt for others.
99 */
100
101static const char *ufslike_filesystems[] = {
102	"ufs",
103	"nfs",
104	"nfs4",
105	"ext2fs",
106	0
107};
108
109FTS *
110fts_open(argv, options, compar)
111	char * const *argv;
112	int options;
113	int (*compar)(const FTSENT * const *, const FTSENT * const *);
114{
115	struct _fts_private *priv;
116	FTS *sp;
117	FTSENT *p, *root;
118	int nitems;
119	FTSENT *parent, *tmp;
120	int len;
121
122	/* Options check. */
123	if (options & ~FTS_OPTIONMASK) {
124		errno = EINVAL;
125		return (NULL);
126	}
127
128	/* Allocate/initialize the stream. */
129	if ((priv = malloc(sizeof(*priv))) == NULL)
130		return (NULL);
131	memset(priv, 0, sizeof(*priv));
132	sp = &priv->ftsp_fts;
133	sp->fts_compar = compar;
134	sp->fts_options = options;
135
136	/* Shush, GCC. */
137	tmp = NULL;
138
139	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
140	if (ISSET(FTS_LOGICAL))
141		SET(FTS_NOCHDIR);
142
143	/*
144	 * Start out with 1K of path space, and enough, in any case,
145	 * to hold the user's paths.
146	 */
147	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
148		goto mem1;
149
150	/* Allocate/initialize root's parent. */
151	if ((parent = fts_alloc(sp, "", 0)) == NULL)
152		goto mem2;
153	parent->fts_level = FTS_ROOTPARENTLEVEL;
154
155	/* Allocate/initialize root(s). */
156	for (root = NULL, nitems = 0; *argv != NULL; ++argv, ++nitems) {
157		/* Don't allow zero-length paths. */
158		if ((len = strlen(*argv)) == 0) {
159			errno = ENOENT;
160			goto mem3;
161		}
162
163		p = fts_alloc(sp, *argv, len);
164		p->fts_level = FTS_ROOTLEVEL;
165		p->fts_parent = parent;
166		p->fts_accpath = p->fts_name;
167		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW));
168
169		/* Command-line "." and ".." are real directories. */
170		if (p->fts_info == FTS_DOT)
171			p->fts_info = FTS_D;
172
173		/*
174		 * If comparison routine supplied, traverse in sorted
175		 * order; otherwise traverse in the order specified.
176		 */
177		if (compar) {
178			p->fts_link = root;
179			root = p;
180		} else {
181			p->fts_link = NULL;
182			if (root == NULL)
183				tmp = root = p;
184			else {
185				tmp->fts_link = p;
186				tmp = p;
187			}
188		}
189	}
190	if (compar && nitems > 1)
191		root = fts_sort(sp, root, nitems);
192
193	/*
194	 * Allocate a dummy pointer and make fts_read think that we've just
195	 * finished the node before the root(s); set p->fts_info to FTS_INIT
196	 * so that everything about the "current" node is ignored.
197	 */
198	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
199		goto mem3;
200	sp->fts_cur->fts_link = root;
201	sp->fts_cur->fts_info = FTS_INIT;
202
203	/*
204	 * If using chdir(2), grab a file descriptor pointing to dot to ensure
205	 * that we can get back here; this could be avoided for some paths,
206	 * but almost certainly not worth the effort.  Slashes, symbolic links,
207	 * and ".." are all fairly nasty problems.  Note, if we can't get the
208	 * descriptor we run anyway, just more slowly.
209	 */
210	if (!ISSET(FTS_NOCHDIR) && (sp->fts_rfd = _open(".", O_RDONLY, 0)) < 0)
211		SET(FTS_NOCHDIR);
212
213	return (sp);
214
215mem3:	fts_lfree(root);
216	free(parent);
217mem2:	free(sp->fts_path);
218mem1:	free(sp);
219	return (NULL);
220}
221
222static void
223fts_load(sp, p)
224	FTS *sp;
225	FTSENT *p;
226{
227	int len;
228	char *cp;
229
230	/*
231	 * Load the stream structure for the next traversal.  Since we don't
232	 * actually enter the directory until after the preorder visit, set
233	 * the fts_accpath field specially so the chdir gets done to the right
234	 * place and the user can access the first node.  From fts_open it's
235	 * known that the path will fit.
236	 */
237	len = p->fts_pathlen = p->fts_namelen;
238	memmove(sp->fts_path, p->fts_name, len + 1);
239	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
240		len = strlen(++cp);
241		memmove(p->fts_name, cp, len + 1);
242		p->fts_namelen = len;
243	}
244	p->fts_accpath = p->fts_path = sp->fts_path;
245	sp->fts_dev = p->fts_dev;
246}
247
248int
249fts_close(sp)
250	FTS *sp;
251{
252	FTSENT *freep, *p;
253	int saved_errno;
254
255	/*
256	 * This still works if we haven't read anything -- the dummy structure
257	 * points to the root list, so we step through to the end of the root
258	 * list which has a valid parent pointer.
259	 */
260	if (sp->fts_cur) {
261		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
262			freep = p;
263			p = p->fts_link != NULL ? p->fts_link : p->fts_parent;
264			free(freep);
265		}
266		free(p);
267	}
268
269	/* Free up child linked list, sort array, path buffer. */
270	if (sp->fts_child)
271		fts_lfree(sp->fts_child);
272	if (sp->fts_array)
273		free(sp->fts_array);
274	free(sp->fts_path);
275
276	/* Return to original directory, save errno if necessary. */
277	if (!ISSET(FTS_NOCHDIR)) {
278		saved_errno = fchdir(sp->fts_rfd) ? errno : 0;
279		(void)_close(sp->fts_rfd);
280
281		/* Set errno and return. */
282		if (saved_errno != 0) {
283			/* Free up the stream pointer. */
284			free(sp);
285			errno = saved_errno;
286			return (-1);
287		}
288	}
289
290	/* Free up the stream pointer. */
291	free(sp);
292	return (0);
293}
294
295/*
296 * Special case of "/" at the end of the path so that slashes aren't
297 * appended which would cause paths to be written as "....//foo".
298 */
299#define	NAPPEND(p)							\
300	(p->fts_path[p->fts_pathlen - 1] == '/'				\
301	    ? p->fts_pathlen - 1 : p->fts_pathlen)
302
303FTSENT *
304fts_read(sp)
305	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);
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);
338		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
339			if ((p->fts_symfd = _open(".", O_RDONLY, 0)) < 0) {
340				p->fts_errno = errno;
341				p->fts_info = FTS_ERR;
342			} else
343				p->fts_flags |= FTS_SYMFOLLOW;
344		}
345		return (p);
346	}
347
348	/* Directory in pre-order. */
349	if (p->fts_info == FTS_D) {
350		/* If skipped or crossed mount point, do post-order visit. */
351		if (instr == FTS_SKIP ||
352		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
353			if (p->fts_flags & FTS_SYMFOLLOW)
354				(void)_close(p->fts_symfd);
355			if (sp->fts_child) {
356				fts_lfree(sp->fts_child);
357				sp->fts_child = NULL;
358			}
359			p->fts_info = FTS_DP;
360			return (p);
361		}
362
363		/* Rebuild if only read the names and now traversing. */
364		if (sp->fts_child != NULL && ISSET(FTS_NAMEONLY)) {
365			CLR(FTS_NAMEONLY);
366			fts_lfree(sp->fts_child);
367			sp->fts_child = NULL;
368		}
369
370		/*
371		 * Cd to the subdirectory.
372		 *
373		 * If have already read and now fail to chdir, whack the list
374		 * to make the names come out right, and set the parent errno
375		 * so the application will eventually get an error condition.
376		 * Set the FTS_DONTCHDIR flag so that when we logically change
377		 * directories back to the parent we don't do a chdir.
378		 *
379		 * If haven't read do so.  If the read fails, fts_build sets
380		 * FTS_STOP or the fts_info field of the node.
381		 */
382		if (sp->fts_child != NULL) {
383			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
384				p->fts_errno = errno;
385				p->fts_flags |= FTS_DONTCHDIR;
386				for (p = sp->fts_child; p != NULL;
387				    p = p->fts_link)
388					p->fts_accpath =
389					    p->fts_parent->fts_accpath;
390			}
391		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
392			if (ISSET(FTS_STOP))
393				return (NULL);
394			return (p);
395		}
396		p = sp->fts_child;
397		sp->fts_child = NULL;
398		goto name;
399	}
400
401	/* Move to the next node on this level. */
402next:	tmp = p;
403	if ((p = p->fts_link) != NULL) {
404		free(tmp);
405
406		/*
407		 * If reached the top, return to the original directory (or
408		 * the root of the tree), and load the paths for the next root.
409		 */
410		if (p->fts_level == FTS_ROOTLEVEL) {
411			if (FCHDIR(sp, sp->fts_rfd)) {
412				SET(FTS_STOP);
413				return (NULL);
414			}
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			goto next;
426		if (p->fts_instr == FTS_FOLLOW) {
427			p->fts_info = fts_stat(sp, p, 1);
428			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
429				if ((p->fts_symfd =
430				    _open(".", O_RDONLY, 0)) < 0) {
431					p->fts_errno = errno;
432					p->fts_info = FTS_ERR;
433				} else
434					p->fts_flags |= FTS_SYMFOLLOW;
435			}
436			p->fts_instr = FTS_NOINSTR;
437		}
438
439name:		t = sp->fts_path + NAPPEND(p->fts_parent);
440		*t++ = '/';
441		memmove(t, p->fts_name, p->fts_namelen + 1);
442		return (sp->fts_cur = p);
443	}
444
445	/* Move up to the parent node. */
446	p = tmp->fts_parent;
447	free(tmp);
448
449	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
450		/*
451		 * Done; free everything up and set errno to 0 so the user
452		 * can distinguish between error and EOF.
453		 */
454		free(p);
455		errno = 0;
456		return (sp->fts_cur = NULL);
457	}
458
459	/* NUL terminate the pathname. */
460	sp->fts_path[p->fts_pathlen] = '\0';
461
462	/*
463	 * Return to the parent directory.  If at a root node or came through
464	 * a symlink, go back through the file descriptor.  Otherwise, cd up
465	 * one directory.
466	 */
467	if (p->fts_level == FTS_ROOTLEVEL) {
468		if (FCHDIR(sp, sp->fts_rfd)) {
469			SET(FTS_STOP);
470			return (NULL);
471		}
472	} else if (p->fts_flags & FTS_SYMFOLLOW) {
473		if (FCHDIR(sp, p->fts_symfd)) {
474			saved_errno = errno;
475			(void)_close(p->fts_symfd);
476			errno = saved_errno;
477			SET(FTS_STOP);
478			return (NULL);
479		}
480		(void)_close(p->fts_symfd);
481	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
482	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
483		SET(FTS_STOP);
484		return (NULL);
485	}
486	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
487	return (sp->fts_cur = p);
488}
489
490/*
491 * Fts_set takes the stream as an argument although it's not used in this
492 * implementation; it would be necessary if anyone wanted to add global
493 * semantics to fts using fts_set.  An error return is allowed for similar
494 * reasons.
495 */
496/* ARGSUSED */
497int
498fts_set(sp, p, instr)
499	FTS *sp;
500	FTSENT *p;
501	int instr;
502{
503	if (instr != 0 && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
504	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
505		errno = EINVAL;
506		return (1);
507	}
508	p->fts_instr = instr;
509	return (0);
510}
511
512FTSENT *
513fts_children(sp, instr)
514	FTS *sp;
515	int instr;
516{
517	FTSENT *p;
518	int fd;
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, 0)) < 0)
572		return (NULL);
573	sp->fts_child = fts_build(sp, instr);
574	if (fchdir(fd))
575		return (NULL);
576	(void)_close(fd);
577	return (sp->fts_child);
578}
579
580#ifndef fts_get_clientptr
581#error "fts_get_clientptr not defined"
582#endif
583
584void *
585(fts_get_clientptr)(FTS *sp)
586{
587
588	return (fts_get_clientptr(sp));
589}
590
591#ifndef fts_get_stream
592#error "fts_get_stream not defined"
593#endif
594
595FTS *
596(fts_get_stream)(FTSENT *p)
597{
598	return (fts_get_stream(p));
599}
600
601void
602fts_set_clientptr(FTS *sp, void *clientptr)
603{
604
605	sp->fts_clientptr = clientptr;
606}
607
608/*
609 * This is the tricky part -- do not casually change *anything* in here.  The
610 * idea is to build the linked list of entries that are used by fts_children
611 * and fts_read.  There are lots of special cases.
612 *
613 * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
614 * set and it's a physical walk (so that symbolic links can't be directories),
615 * we can do things quickly.  First, if it's a 4.4BSD file system, the type
616 * of the file is in the directory entry.  Otherwise, we assume that the number
617 * of subdirectories in a node is equal to the number of links to the parent.
618 * The former skips all stat calls.  The latter skips stat calls in any leaf
619 * directories and for any files after the subdirectories in the directory have
620 * been found, cutting the stat calls by about 2/3.
621 */
622static FTSENT *
623fts_build(sp, type)
624	FTS *sp;
625	int type;
626{
627	struct dirent *dp;
628	FTSENT *p, *head;
629	int nitems;
630	FTSENT *cur, *tail;
631	DIR *dirp;
632	void *oldaddr;
633	size_t dnamlen;
634	int cderrno, descend, len, level, maxlen, nlinks, oflag, saved_errno,
635	    nostat, doadjust;
636	char *cp;
637
638	/* Set current node pointer. */
639	cur = sp->fts_cur;
640
641	/*
642	 * Open the directory for reading.  If this fails, we're done.
643	 * If being called from fts_read, set the fts_info field.
644	 */
645#ifdef FTS_WHITEOUT
646	if (ISSET(FTS_WHITEOUT))
647		oflag = DTF_NODUP | DTF_REWIND;
648	else
649		oflag = DTF_HIDEW | DTF_NODUP | DTF_REWIND;
650#else
651#define __opendir2(path, flag) opendir(path)
652#endif
653	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
654		if (type == BREAD) {
655			cur->fts_info = FTS_DNR;
656			cur->fts_errno = errno;
657		}
658		return (NULL);
659	}
660
661	/*
662	 * Nlinks is the number of possible entries of type directory in the
663	 * directory if we're cheating on stat calls, 0 if we're not doing
664	 * any stat calls at all, -1 if we're doing stats on everything.
665	 */
666	if (type == BNAMES) {
667		nlinks = 0;
668		/* Be quiet about nostat, GCC. */
669		nostat = 0;
670	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
671		if (fts_ufslinks(sp, cur))
672			nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
673		else
674			nlinks = -1;
675		nostat = 1;
676	} else {
677		nlinks = -1;
678		nostat = 0;
679	}
680
681#ifdef notdef
682	(void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
683	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
684	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
685#endif
686	/*
687	 * If we're going to need to stat anything or we want to descend
688	 * and stay in the directory, chdir.  If this fails we keep going,
689	 * but set a flag so we don't chdir after the post-order visit.
690	 * We won't be able to stat anything, but we can still return the
691	 * names themselves.  Note, that since fts_read won't be able to
692	 * chdir into the directory, it will have to return different path
693	 * names than before, i.e. "a/b" instead of "b".  Since the node
694	 * has already been visited in pre-order, have to wait until the
695	 * post-order visit to return the error.  There is a special case
696	 * here, if there was nothing to stat then it's not an error to
697	 * not be able to stat.  This is all fairly nasty.  If a program
698	 * needed sorted entries or stat information, they had better be
699	 * checking FTS_NS on the returned nodes.
700	 */
701	cderrno = 0;
702	if (nlinks || type == BREAD) {
703		if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) {
704			if (nlinks && type == BREAD)
705				cur->fts_errno = errno;
706			cur->fts_flags |= FTS_DONTCHDIR;
707			descend = 0;
708			cderrno = errno;
709		} else
710			descend = 1;
711	} else
712		descend = 0;
713
714	/*
715	 * Figure out the max file name length that can be stored in the
716	 * current path -- the inner loop allocates more path as necessary.
717	 * We really wouldn't have to do the maxlen calculations here, we
718	 * could do them in fts_read before returning the path, but it's a
719	 * lot easier here since the length is part of the dirent structure.
720	 *
721	 * If not changing directories set a pointer so that can just append
722	 * each new name into the path.
723	 */
724	len = NAPPEND(cur);
725	if (ISSET(FTS_NOCHDIR)) {
726		cp = sp->fts_path + len;
727		*cp++ = '/';
728	} else {
729		/* GCC, you're too verbose. */
730		cp = NULL;
731	}
732	len++;
733	maxlen = sp->fts_pathlen - len;
734
735	level = cur->fts_level + 1;
736
737	/* Read the directory, attaching each entry to the `link' pointer. */
738	doadjust = 0;
739	for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) {
740		dnamlen = dp->d_namlen;
741		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
742			continue;
743
744		if ((p = fts_alloc(sp, dp->d_name, (int)dnamlen)) == NULL)
745			goto mem1;
746		if (dnamlen >= maxlen) {	/* include space for NUL */
747			oldaddr = sp->fts_path;
748			if (fts_palloc(sp, dnamlen + len + 1)) {
749				/*
750				 * No more memory for path or structures.  Save
751				 * errno, free up the current structure and the
752				 * structures already allocated.
753				 */
754mem1:				saved_errno = errno;
755				if (p)
756					free(p);
757				fts_lfree(head);
758				(void)closedir(dirp);
759				cur->fts_info = FTS_ERR;
760				SET(FTS_STOP);
761				errno = saved_errno;
762				return (NULL);
763			}
764			/* Did realloc() change the pointer? */
765			if (oldaddr != sp->fts_path) {
766				doadjust = 1;
767				if (ISSET(FTS_NOCHDIR))
768					cp = sp->fts_path + len;
769			}
770			maxlen = sp->fts_pathlen - len;
771		}
772
773		if (len + dnamlen >= USHRT_MAX) {
774			/*
775			 * In an FTSENT, fts_pathlen is a u_short so it is
776			 * possible to wraparound here.  If we do, free up
777			 * the current structure and the structures already
778			 * allocated, then error out with ENAMETOOLONG.
779			 */
780			free(p);
781			fts_lfree(head);
782			(void)closedir(dirp);
783			cur->fts_info = FTS_ERR;
784			SET(FTS_STOP);
785			errno = ENAMETOOLONG;
786			return (NULL);
787		}
788		p->fts_level = level;
789		p->fts_parent = sp->fts_cur;
790		p->fts_pathlen = len + dnamlen;
791
792#ifdef FTS_WHITEOUT
793		if (dp->d_type == DT_WHT)
794			p->fts_flags |= FTS_ISW;
795#endif
796
797		if (cderrno) {
798			if (nlinks) {
799				p->fts_info = FTS_NS;
800				p->fts_errno = cderrno;
801			} else
802				p->fts_info = FTS_NSOK;
803			p->fts_accpath = cur->fts_accpath;
804		} else if (nlinks == 0
805#ifdef DT_DIR
806		    || (nostat &&
807		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
808#endif
809		    ) {
810			p->fts_accpath =
811			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
812			p->fts_info = FTS_NSOK;
813		} else {
814			/* Build a file name for fts_stat to stat. */
815			if (ISSET(FTS_NOCHDIR)) {
816				p->fts_accpath = p->fts_path;
817				memmove(cp, p->fts_name, p->fts_namelen + 1);
818			} else
819				p->fts_accpath = p->fts_name;
820			/* Stat it. */
821			p->fts_info = fts_stat(sp, p, 0);
822
823			/* Decrement link count if applicable. */
824			if (nlinks > 0 && (p->fts_info == FTS_D ||
825			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
826				--nlinks;
827		}
828
829		/* We walk in directory order so "ls -f" doesn't get upset. */
830		p->fts_link = NULL;
831		if (head == NULL)
832			head = tail = p;
833		else {
834			tail->fts_link = p;
835			tail = p;
836		}
837		++nitems;
838	}
839	if (dirp)
840		(void)closedir(dirp);
841
842	/*
843	 * If realloc() changed the address of the path, adjust the
844	 * addresses for the rest of the tree and the dir list.
845	 */
846	if (doadjust)
847		fts_padjust(sp, head);
848
849	/*
850	 * If not changing directories, reset the path back to original
851	 * state.
852	 */
853	if (ISSET(FTS_NOCHDIR)) {
854		if (len == sp->fts_pathlen || nitems == 0)
855			--cp;
856		*cp = '\0';
857	}
858
859	/*
860	 * If descended after called from fts_children or after called from
861	 * fts_read and nothing found, get back.  At the root level we use
862	 * the saved fd; if one of fts_open()'s arguments is a relative path
863	 * to an empty directory, we wind up here with no other way back.  If
864	 * can't get back, we're done.
865	 */
866	if (descend && (type == BCHILD || !nitems) &&
867	    (cur->fts_level == FTS_ROOTLEVEL ?
868	    FCHDIR(sp, sp->fts_rfd) :
869	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
870		cur->fts_info = FTS_ERR;
871		SET(FTS_STOP);
872		return (NULL);
873	}
874
875	/* If didn't find anything, return NULL. */
876	if (!nitems) {
877		if (type == BREAD)
878			cur->fts_info = FTS_DP;
879		return (NULL);
880	}
881
882	/* Sort the entries. */
883	if (sp->fts_compar && nitems > 1)
884		head = fts_sort(sp, head, nitems);
885	return (head);
886}
887
888static u_short
889fts_stat(sp, p, follow)
890	FTS *sp;
891	FTSENT *p;
892	int follow;
893{
894	FTSENT *t;
895	dev_t dev;
896	ino_t ino;
897	struct stat *sbp, sb;
898	int saved_errno;
899
900	/* If user needs stat info, stat buffer already allocated. */
901	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
902
903#ifdef FTS_WHITEOUT
904	/* Check for whiteout. */
905	if (p->fts_flags & FTS_ISW) {
906		if (sbp != &sb) {
907			memset(sbp, '\0', sizeof(*sbp));
908			sbp->st_mode = S_IFWHT;
909		}
910		return (FTS_W);
911	}
912#endif
913
914	/*
915	 * If doing a logical walk, or application requested FTS_FOLLOW, do
916	 * a stat(2).  If that fails, check for a non-existent symlink.  If
917	 * fail, set the errno from the stat call.
918	 */
919	if (ISSET(FTS_LOGICAL) || follow) {
920		if (stat(p->fts_accpath, sbp)) {
921			saved_errno = errno;
922			if (!lstat(p->fts_accpath, sbp)) {
923				errno = 0;
924				return (FTS_SLNONE);
925			}
926			p->fts_errno = saved_errno;
927			goto err;
928		}
929	} else if (lstat(p->fts_accpath, sbp)) {
930		p->fts_errno = errno;
931err:		memset(sbp, 0, sizeof(struct stat));
932		return (FTS_NS);
933	}
934
935	if (S_ISDIR(sbp->st_mode)) {
936		/*
937		 * Set the device/inode.  Used to find cycles and check for
938		 * crossing mount points.  Also remember the link count, used
939		 * in fts_build to limit the number of stat calls.  It is
940		 * understood that these fields are only referenced if fts_info
941		 * is set to FTS_D.
942		 */
943		dev = p->fts_dev = sbp->st_dev;
944		ino = p->fts_ino = sbp->st_ino;
945		p->fts_nlink = sbp->st_nlink;
946
947		if (ISDOT(p->fts_name))
948			return (FTS_DOT);
949
950		/*
951		 * Cycle detection is done by brute force when the directory
952		 * is first encountered.  If the tree gets deep enough or the
953		 * number of symbolic links to directories is high enough,
954		 * something faster might be worthwhile.
955		 */
956		for (t = p->fts_parent;
957		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
958			if (ino == t->fts_ino && dev == t->fts_dev) {
959				p->fts_cycle = t;
960				return (FTS_DC);
961			}
962		return (FTS_D);
963	}
964	if (S_ISLNK(sbp->st_mode))
965		return (FTS_SL);
966	if (S_ISREG(sbp->st_mode))
967		return (FTS_F);
968	return (FTS_DEFAULT);
969}
970
971/*
972 * The comparison function takes pointers to pointers to FTSENT structures.
973 * Qsort wants a comparison function that takes pointers to void.
974 * (Both with appropriate levels of const-poisoning, of course!)
975 * Use a trampoline function to deal with the difference.
976 */
977static int
978fts_compar(const void *a, const void *b)
979{
980	FTS *parent;
981
982	parent = (*(const FTSENT * const *)a)->fts_fts;
983	return (*parent->fts_compar)(a, b);
984}
985
986static FTSENT *
987fts_sort(sp, head, nitems)
988	FTS *sp;
989	FTSENT *head;
990	int nitems;
991{
992	FTSENT **ap, *p;
993
994	/*
995	 * Construct an array of pointers to the structures and call qsort(3).
996	 * Reassemble the array in the order returned by qsort.  If unable to
997	 * sort for memory reasons, return the directory entries in their
998	 * current order.  Allocate enough space for the current needs plus
999	 * 40 so don't realloc one entry at a time.
1000	 */
1001	if (nitems > sp->fts_nitems) {
1002		sp->fts_nitems = nitems + 40;
1003		if ((sp->fts_array = reallocf(sp->fts_array,
1004		    sp->fts_nitems * sizeof(FTSENT *))) == NULL) {
1005			sp->fts_nitems = 0;
1006			return (head);
1007		}
1008	}
1009	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
1010		*ap++ = p;
1011	qsort(sp->fts_array, nitems, sizeof(FTSENT *), fts_compar);
1012	for (head = *(ap = sp->fts_array); --nitems; ++ap)
1013		ap[0]->fts_link = ap[1];
1014	ap[0]->fts_link = NULL;
1015	return (head);
1016}
1017
1018static FTSENT *
1019fts_alloc(sp, name, namelen)
1020	FTS *sp;
1021	char *name;
1022	int namelen;
1023{
1024	FTSENT *p;
1025	size_t len;
1026
1027	struct ftsent_withstat {
1028		FTSENT	ent;
1029		struct	stat statbuf;
1030	};
1031
1032	/*
1033	 * The file name is a variable length array and no stat structure is
1034	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
1035	 * structure, the file name and the stat structure in one chunk, but
1036	 * be careful that the stat structure is reasonably aligned.
1037	 */
1038	if (ISSET(FTS_NOSTAT))
1039		len = sizeof(FTSENT) + namelen + 1;
1040	else
1041		len = sizeof(struct ftsent_withstat) + namelen + 1;
1042
1043	if ((p = malloc(len)) == NULL)
1044		return (NULL);
1045
1046	if (ISSET(FTS_NOSTAT)) {
1047		p->fts_name = (char *)(p + 1);
1048		p->fts_statp = NULL;
1049	} else {
1050		p->fts_name = (char *)((struct ftsent_withstat *)p + 1);
1051		p->fts_statp = &((struct ftsent_withstat *)p)->statbuf;
1052	}
1053
1054	/* Copy the name and guarantee NUL termination. */
1055	memcpy(p->fts_name, name, namelen);
1056	p->fts_name[namelen] = '\0';
1057	p->fts_namelen = namelen;
1058	p->fts_path = sp->fts_path;
1059	p->fts_errno = 0;
1060	p->fts_flags = 0;
1061	p->fts_instr = FTS_NOINSTR;
1062	p->fts_number = 0;
1063	p->fts_pointer = NULL;
1064	p->fts_fts = sp;
1065	return (p);
1066}
1067
1068static void
1069fts_lfree(head)
1070	FTSENT *head;
1071{
1072	FTSENT *p;
1073
1074	/* Free a linked list of structures. */
1075	while ((p = head)) {
1076		head = head->fts_link;
1077		free(p);
1078	}
1079}
1080
1081/*
1082 * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
1083 * Most systems will allow creation of paths much longer than MAXPATHLEN, even
1084 * though the kernel won't resolve them.  Add the size (not just what's needed)
1085 * plus 256 bytes so don't realloc the path 2 bytes at a time.
1086 */
1087static int
1088fts_palloc(sp, more)
1089	FTS *sp;
1090	size_t more;
1091{
1092
1093	sp->fts_pathlen += more + 256;
1094	/*
1095	 * Check for possible wraparound.  In an FTS, fts_pathlen is
1096	 * a signed int but in an FTSENT it is an unsigned short.
1097	 * We limit fts_pathlen to USHRT_MAX to be safe in both cases.
1098	 */
1099	if (sp->fts_pathlen < 0 || sp->fts_pathlen >= USHRT_MAX) {
1100		if (sp->fts_path)
1101			free(sp->fts_path);
1102		sp->fts_path = NULL;
1103		errno = ENAMETOOLONG;
1104		return (1);
1105	}
1106	sp->fts_path = reallocf(sp->fts_path, sp->fts_pathlen);
1107	return (sp->fts_path == NULL);
1108}
1109
1110/*
1111 * When the path is realloc'd, have to fix all of the pointers in structures
1112 * already returned.
1113 */
1114static void
1115fts_padjust(sp, head)
1116	FTS *sp;
1117	FTSENT *head;
1118{
1119	FTSENT *p;
1120	char *addr = sp->fts_path;
1121
1122#define	ADJUST(p) do {							\
1123	if ((p)->fts_accpath != (p)->fts_name) {			\
1124		(p)->fts_accpath =					\
1125		    (char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
1126	}								\
1127	(p)->fts_path = addr;						\
1128} while (0)
1129	/* Adjust the current set of children. */
1130	for (p = sp->fts_child; p; p = p->fts_link)
1131		ADJUST(p);
1132
1133	/* Adjust the rest of the tree, including the current level. */
1134	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1135		ADJUST(p);
1136		p = p->fts_link ? p->fts_link : p->fts_parent;
1137	}
1138}
1139
1140static size_t
1141fts_maxarglen(argv)
1142	char * const *argv;
1143{
1144	size_t len, max;
1145
1146	for (max = 0; *argv; ++argv)
1147		if ((len = strlen(*argv)) > max)
1148			max = len;
1149	return (max + 1);
1150}
1151
1152/*
1153 * Change to dir specified by fd or p->fts_accpath without getting
1154 * tricked by someone changing the world out from underneath us.
1155 * Assumes p->fts_dev and p->fts_ino are filled in.
1156 */
1157static int
1158fts_safe_changedir(sp, p, fd, path)
1159	FTS *sp;
1160	FTSENT *p;
1161	int fd;
1162	char *path;
1163{
1164	int ret, oerrno, newfd;
1165	struct stat sb;
1166
1167	newfd = fd;
1168	if (ISSET(FTS_NOCHDIR))
1169		return (0);
1170	if (fd < 0 && (newfd = _open(path, O_RDONLY, 0)) < 0)
1171		return (-1);
1172	if (_fstat(newfd, &sb)) {
1173		ret = -1;
1174		goto bail;
1175	}
1176	if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) {
1177		errno = ENOENT;		/* disinformation */
1178		ret = -1;
1179		goto bail;
1180	}
1181	ret = fchdir(newfd);
1182bail:
1183	oerrno = errno;
1184	if (fd < 0)
1185		(void)_close(newfd);
1186	errno = oerrno;
1187	return (ret);
1188}
1189
1190/*
1191 * Check if the filesystem for "ent" has UFS-style links.
1192 */
1193static int
1194fts_ufslinks(FTS *sp, const FTSENT *ent)
1195{
1196	struct _fts_private *priv;
1197	const char **cpp;
1198
1199	priv = (struct _fts_private *)sp;
1200	/*
1201	 * If this node's device is different from the previous, grab
1202	 * the filesystem information, and decide on the reliability
1203	 * of the link information from this filesystem for stat(2)
1204	 * avoidance.
1205	 */
1206	if (priv->ftsp_dev != ent->fts_dev) {
1207		if (statfs(ent->fts_path, &priv->ftsp_statfs) != -1) {
1208			priv->ftsp_dev = ent->fts_dev;
1209			priv->ftsp_linksreliable = 0;
1210			for (cpp = ufslike_filesystems; *cpp; cpp++) {
1211				if (strcmp(priv->ftsp_statfs.f_fstypename,
1212				    *cpp) == 0) {
1213					priv->ftsp_linksreliable = 1;
1214					break;
1215				}
1216			}
1217		} else {
1218			priv->ftsp_linksreliable = 0;
1219		}
1220	}
1221	return (priv->ftsp_linksreliable);
1222}
1223