fts.c revision 130233
1/*-
2 * Copyright (c) 1990, 1993, 1994
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 *    must display the following acknowledgement:
15 *	This product includes software developed by the University of
16 *	California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 *    may be used to endorse or promote products derived from this software
19 *    without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * $OpenBSD: fts.c,v 1.22 1999/10/03 19:22:22 millert Exp $
34 */
35
36#if 0
37#if defined(LIBC_SCCS) && !defined(lint)
38static char sccsid[] = "@(#)fts.c	8.6 (Berkeley) 8/14/94";
39#endif /* LIBC_SCCS and not lint */
40#endif
41
42#include <sys/cdefs.h>
43__FBSDID("$FreeBSD: head/lib/libc/gen/fts.c 130233 2004-06-08 06:23:23Z das $");
44
45#include "namespace.h"
46#include <sys/param.h>
47#include <sys/mount.h>
48#include <sys/stat.h>
49
50#include <dirent.h>
51#include <errno.h>
52#include <fcntl.h>
53#include <fts.h>
54#include <stdlib.h>
55#include <string.h>
56#include <unistd.h>
57#include "un-namespace.h"
58
59static FTSENT	*fts_alloc(FTS *, char *, int);
60static FTSENT	*fts_build(FTS *, int);
61static void	 fts_lfree(FTSENT *);
62static void	 fts_load(FTS *, FTSENT *);
63static size_t	 fts_maxarglen(char * const *);
64static void	 fts_padjust(FTS *, FTSENT *);
65static int	 fts_palloc(FTS *, size_t);
66static FTSENT	*fts_sort(FTS *, FTSENT *, int);
67static u_short	 fts_stat(FTS *, FTSENT *, int);
68static int	 fts_safe_changedir(FTS *, FTSENT *, int, char *);
69static int	 fts_ufslinks(FTS *, const FTSENT *);
70
71#define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
72
73#define	CLR(opt)	(sp->fts_options &= ~(opt))
74#define	ISSET(opt)	(sp->fts_options & (opt))
75#define	SET(opt)	(sp->fts_options |= (opt))
76
77#define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
78
79/* fts_build flags */
80#define	BCHILD		1		/* fts_children */
81#define	BNAMES		2		/* fts_children, names only */
82#define	BREAD		3		/* fts_read */
83
84/*
85 * Internal representation of an FTS, including extra implementation
86 * details.  The FTS returned from fts_open points to this structure's
87 * ftsp_fts member (and can be cast to an _fts_private as required)
88 */
89struct _fts_private {
90	FTS		ftsp_fts;
91	struct statfs	ftsp_statfs;
92	dev_t		ftsp_dev;
93	int		ftsp_linksreliable;
94};
95
96/*
97 * The "FTS_NOSTAT" option can avoid a lot of calls to stat(2) if it
98 * knows that a directory could not possibly have subdirectories.  This
99 * is decided by looking at the link count: a subdirectory would
100 * increment its parent's link count by virtue of its own ".." entry.
101 * This assumption only holds for UFS-like filesystems that implement
102 * links and directories this way, so we must punt for others.
103 */
104
105static const char *ufslike_filesystems[] = {
106	"ufs",
107	"nfs",
108	"nfs4",
109	"ext2fs",
110	0
111};
112
113FTS *
114fts_open(argv, options, compar)
115	char * const *argv;
116	int options;
117	int (*compar)(const FTSENT * const *, const FTSENT * const *);
118{
119	struct _fts_private *priv;
120	FTS *sp;
121	FTSENT *p, *root;
122	int nitems;
123	FTSENT *parent, *tmp;
124	int len;
125
126	/* Options check. */
127	if (options & ~FTS_OPTIONMASK) {
128		errno = EINVAL;
129		return (NULL);
130	}
131
132	/* Allocate/initialize the stream. */
133	if ((priv = malloc(sizeof(*priv))) == NULL)
134		return (NULL);
135	memset(priv, 0, sizeof(*priv));
136	sp = &priv->ftsp_fts;
137	sp->fts_compar = compar;
138	sp->fts_options = options;
139
140	/* Shush, GCC. */
141	tmp = NULL;
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	/* Allocate/initialize root(s). */
160	for (root = NULL, nitems = 0; *argv != NULL; ++argv, ++nitems) {
161		/* Don't allow zero-length paths. */
162		if ((len = strlen(*argv)) == 0) {
163			errno = ENOENT;
164			goto mem3;
165		}
166
167		p = fts_alloc(sp, *argv, len);
168		p->fts_level = FTS_ROOTLEVEL;
169		p->fts_parent = parent;
170		p->fts_accpath = p->fts_name;
171		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW));
172
173		/* Command-line "." and ".." are real directories. */
174		if (p->fts_info == FTS_DOT)
175			p->fts_info = FTS_D;
176
177		/*
178		 * If comparison routine supplied, traverse in sorted
179		 * order; otherwise traverse in the order specified.
180		 */
181		if (compar) {
182			p->fts_link = root;
183			root = p;
184		} else {
185			p->fts_link = NULL;
186			if (root == NULL)
187				tmp = root = p;
188			else {
189				tmp->fts_link = p;
190				tmp = p;
191			}
192		}
193	}
194	if (compar && nitems > 1)
195		root = fts_sort(sp, root, nitems);
196
197	/*
198	 * Allocate a dummy pointer and make fts_read think that we've just
199	 * finished the node before the root(s); set p->fts_info to FTS_INIT
200	 * so that everything about the "current" node is ignored.
201	 */
202	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
203		goto mem3;
204	sp->fts_cur->fts_link = root;
205	sp->fts_cur->fts_info = FTS_INIT;
206
207	/*
208	 * If using chdir(2), grab a file descriptor pointing to dot to ensure
209	 * that we can get back here; this could be avoided for some paths,
210	 * but almost certainly not worth the effort.  Slashes, symbolic links,
211	 * and ".." are all fairly nasty problems.  Note, if we can't get the
212	 * descriptor we run anyway, just more slowly.
213	 */
214	if (!ISSET(FTS_NOCHDIR) && (sp->fts_rfd = _open(".", O_RDONLY, 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(sp, p)
228	FTS *sp;
229	FTSENT *p;
230{
231	int len;
232	char *cp;
233
234	/*
235	 * Load the stream structure for the next traversal.  Since we don't
236	 * actually enter the directory until after the preorder visit, set
237	 * the fts_accpath field specially so the chdir gets done to the right
238	 * place and the user can access the first node.  From fts_open it's
239	 * known that the path will fit.
240	 */
241	len = p->fts_pathlen = p->fts_namelen;
242	memmove(sp->fts_path, p->fts_name, len + 1);
243	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
244		len = strlen(++cp);
245		memmove(p->fts_name, cp, len + 1);
246		p->fts_namelen = len;
247	}
248	p->fts_accpath = p->fts_path = sp->fts_path;
249	sp->fts_dev = p->fts_dev;
250}
251
252int
253fts_close(sp)
254	FTS *sp;
255{
256	FTSENT *freep, *p;
257	int saved_errno;
258
259	/*
260	 * This still works if we haven't read anything -- the dummy structure
261	 * points to the root list, so we step through to the end of the root
262	 * list which has a valid parent pointer.
263	 */
264	if (sp->fts_cur) {
265		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
266			freep = p;
267			p = p->fts_link != NULL ? p->fts_link : p->fts_parent;
268			free(freep);
269		}
270		free(p);
271	}
272
273	/* Free up child linked list, sort array, path buffer. */
274	if (sp->fts_child)
275		fts_lfree(sp->fts_child);
276	if (sp->fts_array)
277		free(sp->fts_array);
278	free(sp->fts_path);
279
280	/* Return to original directory, save errno if necessary. */
281	if (!ISSET(FTS_NOCHDIR)) {
282		saved_errno = fchdir(sp->fts_rfd) ? errno : 0;
283		(void)_close(sp->fts_rfd);
284
285		/* Set errno and return. */
286		if (saved_errno != 0) {
287			/* Free up the stream pointer. */
288			free(sp);
289			errno = saved_errno;
290			return (-1);
291		}
292	}
293
294	/* Free up the stream pointer. */
295	free(sp);
296	return (0);
297}
298
299/*
300 * Special case of "/" at the end of the path so that slashes aren't
301 * appended which would cause paths to be written as "....//foo".
302 */
303#define	NAPPEND(p)							\
304	(p->fts_path[p->fts_pathlen - 1] == '/'				\
305	    ? p->fts_pathlen - 1 : p->fts_pathlen)
306
307FTSENT *
308fts_read(sp)
309	FTS *sp;
310{
311	FTSENT *p, *tmp;
312	int instr;
313	char *t;
314	int saved_errno;
315
316	/* If finished or unrecoverable error, return NULL. */
317	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
318		return (NULL);
319
320	/* Set current node pointer. */
321	p = sp->fts_cur;
322
323	/* Save and zero out user instructions. */
324	instr = p->fts_instr;
325	p->fts_instr = FTS_NOINSTR;
326
327	/* Any type of file may be re-visited; re-stat and re-turn. */
328	if (instr == FTS_AGAIN) {
329		p->fts_info = fts_stat(sp, p, 0);
330		return (p);
331	}
332
333	/*
334	 * Following a symlink -- SLNONE test allows application to see
335	 * SLNONE and recover.  If indirecting through a symlink, have
336	 * keep a pointer to current location.  If unable to get that
337	 * pointer, follow fails.
338	 */
339	if (instr == FTS_FOLLOW &&
340	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
341		p->fts_info = fts_stat(sp, p, 1);
342		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
343			if ((p->fts_symfd = _open(".", O_RDONLY, 0)) < 0) {
344				p->fts_errno = errno;
345				p->fts_info = FTS_ERR;
346			} else
347				p->fts_flags |= FTS_SYMFOLLOW;
348		}
349		return (p);
350	}
351
352	/* Directory in pre-order. */
353	if (p->fts_info == FTS_D) {
354		/* If skipped or crossed mount point, do post-order visit. */
355		if (instr == FTS_SKIP ||
356		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
357			if (p->fts_flags & FTS_SYMFOLLOW)
358				(void)_close(p->fts_symfd);
359			if (sp->fts_child) {
360				fts_lfree(sp->fts_child);
361				sp->fts_child = NULL;
362			}
363			p->fts_info = FTS_DP;
364			return (p);
365		}
366
367		/* Rebuild if only read the names and now traversing. */
368		if (sp->fts_child != NULL && ISSET(FTS_NAMEONLY)) {
369			CLR(FTS_NAMEONLY);
370			fts_lfree(sp->fts_child);
371			sp->fts_child = NULL;
372		}
373
374		/*
375		 * Cd to the subdirectory.
376		 *
377		 * If have already read and now fail to chdir, whack the list
378		 * to make the names come out right, and set the parent errno
379		 * so the application will eventually get an error condition.
380		 * Set the FTS_DONTCHDIR flag so that when we logically change
381		 * directories back to the parent we don't do a chdir.
382		 *
383		 * If haven't read do so.  If the read fails, fts_build sets
384		 * FTS_STOP or the fts_info field of the node.
385		 */
386		if (sp->fts_child != NULL) {
387			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
388				p->fts_errno = errno;
389				p->fts_flags |= FTS_DONTCHDIR;
390				for (p = sp->fts_child; p != NULL;
391				    p = p->fts_link)
392					p->fts_accpath =
393					    p->fts_parent->fts_accpath;
394			}
395		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
396			if (ISSET(FTS_STOP))
397				return (NULL);
398			return (p);
399		}
400		p = sp->fts_child;
401		sp->fts_child = NULL;
402		goto name;
403	}
404
405	/* Move to the next node on this level. */
406next:	tmp = p;
407	if ((p = p->fts_link) != NULL) {
408		free(tmp);
409
410		/*
411		 * If reached the top, return to the original directory (or
412		 * the root of the tree), and load the paths for the next root.
413		 */
414		if (p->fts_level == FTS_ROOTLEVEL) {
415			if (FCHDIR(sp, sp->fts_rfd)) {
416				SET(FTS_STOP);
417				return (NULL);
418			}
419			fts_load(sp, p);
420			return (sp->fts_cur = p);
421		}
422
423		/*
424		 * User may have called fts_set on the node.  If skipped,
425		 * ignore.  If followed, get a file descriptor so we can
426		 * get back if necessary.
427		 */
428		if (p->fts_instr == FTS_SKIP)
429			goto next;
430		if (p->fts_instr == FTS_FOLLOW) {
431			p->fts_info = fts_stat(sp, p, 1);
432			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
433				if ((p->fts_symfd =
434				    _open(".", O_RDONLY, 0)) < 0) {
435					p->fts_errno = errno;
436					p->fts_info = FTS_ERR;
437				} else
438					p->fts_flags |= FTS_SYMFOLLOW;
439			}
440			p->fts_instr = FTS_NOINSTR;
441		}
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	free(tmp);
452
453	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
454		/*
455		 * Done; free everything up and set errno to 0 so the user
456		 * can distinguish between error and EOF.
457		 */
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	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
491	return (sp->fts_cur = p);
492}
493
494/*
495 * Fts_set takes the stream as an argument although it's not used in this
496 * implementation; it would be necessary if anyone wanted to add global
497 * semantics to fts using fts_set.  An error return is allowed for similar
498 * reasons.
499 */
500/* ARGSUSED */
501int
502fts_set(sp, p, instr)
503	FTS *sp;
504	FTSENT *p;
505	int instr;
506{
507	if (instr != 0 && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
508	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
509		errno = EINVAL;
510		return (1);
511	}
512	p->fts_instr = instr;
513	return (0);
514}
515
516FTSENT *
517fts_children(sp, instr)
518	FTS *sp;
519	int instr;
520{
521	FTSENT *p;
522	int fd;
523
524	if (instr != 0 && instr != FTS_NAMEONLY) {
525		errno = EINVAL;
526		return (NULL);
527	}
528
529	/* Set current node pointer. */
530	p = sp->fts_cur;
531
532	/*
533	 * Errno set to 0 so user can distinguish empty directory from
534	 * an error.
535	 */
536	errno = 0;
537
538	/* Fatal errors stop here. */
539	if (ISSET(FTS_STOP))
540		return (NULL);
541
542	/* Return logical hierarchy of user's arguments. */
543	if (p->fts_info == FTS_INIT)
544		return (p->fts_link);
545
546	/*
547	 * If not a directory being visited in pre-order, stop here.  Could
548	 * allow FTS_DNR, assuming the user has fixed the problem, but the
549	 * same effect is available with FTS_AGAIN.
550	 */
551	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
552		return (NULL);
553
554	/* Free up any previous child list. */
555	if (sp->fts_child != NULL)
556		fts_lfree(sp->fts_child);
557
558	if (instr == FTS_NAMEONLY) {
559		SET(FTS_NAMEONLY);
560		instr = BNAMES;
561	} else
562		instr = BCHILD;
563
564	/*
565	 * If using chdir on a relative path and called BEFORE fts_read does
566	 * its chdir to the root of a traversal, we can lose -- we need to
567	 * chdir into the subdirectory, and we don't know where the current
568	 * directory is, so we can't get back so that the upcoming chdir by
569	 * fts_read will work.
570	 */
571	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
572	    ISSET(FTS_NOCHDIR))
573		return (sp->fts_child = fts_build(sp, instr));
574
575	if ((fd = _open(".", O_RDONLY, 0)) < 0)
576		return (NULL);
577	sp->fts_child = fts_build(sp, instr);
578	if (fchdir(fd))
579		return (NULL);
580	(void)_close(fd);
581	return (sp->fts_child);
582}
583
584#ifndef fts_get_clientptr
585#error "fts_get_clientptr not defined"
586#endif
587
588void *
589(fts_get_clientptr)(FTS *sp)
590{
591
592	return (fts_get_clientptr(sp));
593}
594
595#ifndef fts_get_stream
596#error "fts_get_stream not defined"
597#endif
598
599FTS *
600(fts_get_stream)(FTSENT *p)
601{
602	return (fts_get_stream(p));
603}
604
605void
606fts_set_clientptr(FTS *sp, void *clientptr)
607{
608
609	sp->fts_clientptr = clientptr;
610}
611
612/*
613 * This is the tricky part -- do not casually change *anything* in here.  The
614 * idea is to build the linked list of entries that are used by fts_children
615 * and fts_read.  There are lots of special cases.
616 *
617 * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
618 * set and it's a physical walk (so that symbolic links can't be directories),
619 * we can do things quickly.  First, if it's a 4.4BSD file system, the type
620 * of the file is in the directory entry.  Otherwise, we assume that the number
621 * of subdirectories in a node is equal to the number of links to the parent.
622 * The former skips all stat calls.  The latter skips stat calls in any leaf
623 * directories and for any files after the subdirectories in the directory have
624 * been found, cutting the stat calls by about 2/3.
625 */
626static FTSENT *
627fts_build(sp, type)
628	FTS *sp;
629	int type;
630{
631	struct dirent *dp;
632	FTSENT *p, *head;
633	int nitems;
634	FTSENT *cur, *tail;
635	DIR *dirp;
636	void *oldaddr;
637	size_t dnamlen;
638	int cderrno, descend, len, level, maxlen, nlinks, oflag, saved_errno,
639	    nostat, doadjust;
640	char *cp;
641
642	/* Set current node pointer. */
643	cur = sp->fts_cur;
644
645	/*
646	 * Open the directory for reading.  If this fails, we're done.
647	 * If being called from fts_read, set the fts_info field.
648	 */
649#ifdef FTS_WHITEOUT
650	if (ISSET(FTS_WHITEOUT))
651		oflag = DTF_NODUP | DTF_REWIND;
652	else
653		oflag = DTF_HIDEW | DTF_NODUP | DTF_REWIND;
654#else
655#define __opendir2(path, flag) opendir(path)
656#endif
657	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
658		if (type == BREAD) {
659			cur->fts_info = FTS_DNR;
660			cur->fts_errno = errno;
661		}
662		return (NULL);
663	}
664
665	/*
666	 * Nlinks is the number of possible entries of type directory in the
667	 * directory if we're cheating on stat calls, 0 if we're not doing
668	 * any stat calls at all, -1 if we're doing stats on everything.
669	 */
670	if (type == BNAMES) {
671		nlinks = 0;
672		/* Be quiet about nostat, GCC. */
673		nostat = 0;
674	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
675		if (fts_ufslinks(sp, cur))
676			nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
677		else
678			nlinks = -1;
679		nostat = 1;
680	} else {
681		nlinks = -1;
682		nostat = 0;
683	}
684
685#ifdef notdef
686	(void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
687	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
688	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
689#endif
690	/*
691	 * If we're going to need to stat anything or we want to descend
692	 * and stay in the directory, chdir.  If this fails we keep going,
693	 * but set a flag so we don't chdir after the post-order visit.
694	 * We won't be able to stat anything, but we can still return the
695	 * names themselves.  Note, that since fts_read won't be able to
696	 * chdir into the directory, it will have to return different path
697	 * names than before, i.e. "a/b" instead of "b".  Since the node
698	 * has already been visited in pre-order, have to wait until the
699	 * post-order visit to return the error.  There is a special case
700	 * here, if there was nothing to stat then it's not an error to
701	 * not be able to stat.  This is all fairly nasty.  If a program
702	 * needed sorted entries or stat information, they had better be
703	 * checking FTS_NS on the returned nodes.
704	 */
705	cderrno = 0;
706	if (nlinks || type == BREAD) {
707		if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) {
708			if (nlinks && type == BREAD)
709				cur->fts_errno = errno;
710			cur->fts_flags |= FTS_DONTCHDIR;
711			descend = 0;
712			cderrno = errno;
713		} else
714			descend = 1;
715	} else
716		descend = 0;
717
718	/*
719	 * Figure out the max file name length that can be stored in the
720	 * current path -- the inner loop allocates more path as necessary.
721	 * We really wouldn't have to do the maxlen calculations here, we
722	 * could do them in fts_read before returning the path, but it's a
723	 * lot easier here since the length is part of the dirent structure.
724	 *
725	 * If not changing directories set a pointer so that can just append
726	 * each new name into the path.
727	 */
728	len = NAPPEND(cur);
729	if (ISSET(FTS_NOCHDIR)) {
730		cp = sp->fts_path + len;
731		*cp++ = '/';
732	} else {
733		/* GCC, you're too verbose. */
734		cp = NULL;
735	}
736	len++;
737	maxlen = sp->fts_pathlen - len;
738
739	level = cur->fts_level + 1;
740
741	/* Read the directory, attaching each entry to the `link' pointer. */
742	doadjust = 0;
743	for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) {
744		dnamlen = dp->d_namlen;
745		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
746			continue;
747
748		if ((p = fts_alloc(sp, dp->d_name, (int)dnamlen)) == NULL)
749			goto mem1;
750		if (dnamlen >= maxlen) {	/* include space for NUL */
751			oldaddr = sp->fts_path;
752			if (fts_palloc(sp, dnamlen + len + 1)) {
753				/*
754				 * No more memory for path or structures.  Save
755				 * errno, free up the current structure and the
756				 * structures already allocated.
757				 */
758mem1:				saved_errno = errno;
759				if (p)
760					free(p);
761				fts_lfree(head);
762				(void)closedir(dirp);
763				cur->fts_info = FTS_ERR;
764				SET(FTS_STOP);
765				errno = saved_errno;
766				return (NULL);
767			}
768			/* Did realloc() change the pointer? */
769			if (oldaddr != sp->fts_path) {
770				doadjust = 1;
771				if (ISSET(FTS_NOCHDIR))
772					cp = sp->fts_path + len;
773			}
774			maxlen = sp->fts_pathlen - len;
775		}
776
777		if (len + dnamlen >= USHRT_MAX) {
778			/*
779			 * In an FTSENT, fts_pathlen is a u_short so it is
780			 * possible to wraparound here.  If we do, free up
781			 * the current structure and the structures already
782			 * allocated, then error out with ENAMETOOLONG.
783			 */
784			free(p);
785			fts_lfree(head);
786			(void)closedir(dirp);
787			cur->fts_info = FTS_ERR;
788			SET(FTS_STOP);
789			errno = ENAMETOOLONG;
790			return (NULL);
791		}
792		p->fts_level = level;
793		p->fts_parent = sp->fts_cur;
794		p->fts_pathlen = len + dnamlen;
795
796#ifdef FTS_WHITEOUT
797		if (dp->d_type == DT_WHT)
798			p->fts_flags |= FTS_ISW;
799#endif
800
801		if (cderrno) {
802			if (nlinks) {
803				p->fts_info = FTS_NS;
804				p->fts_errno = cderrno;
805			} else
806				p->fts_info = FTS_NSOK;
807			p->fts_accpath = cur->fts_accpath;
808		} else if (nlinks == 0
809#ifdef DT_DIR
810		    || (nostat &&
811		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
812#endif
813		    ) {
814			p->fts_accpath =
815			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
816			p->fts_info = FTS_NSOK;
817		} else {
818			/* Build a file name for fts_stat to stat. */
819			if (ISSET(FTS_NOCHDIR)) {
820				p->fts_accpath = p->fts_path;
821				memmove(cp, p->fts_name, p->fts_namelen + 1);
822			} else
823				p->fts_accpath = p->fts_name;
824			/* Stat it. */
825			p->fts_info = fts_stat(sp, p, 0);
826
827			/* Decrement link count if applicable. */
828			if (nlinks > 0 && (p->fts_info == FTS_D ||
829			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
830				--nlinks;
831		}
832
833		/* We walk in directory order so "ls -f" doesn't get upset. */
834		p->fts_link = NULL;
835		if (head == NULL)
836			head = tail = p;
837		else {
838			tail->fts_link = p;
839			tail = p;
840		}
841		++nitems;
842	}
843	if (dirp)
844		(void)closedir(dirp);
845
846	/*
847	 * If realloc() changed the address of the path, adjust the
848	 * addresses for the rest of the tree and the dir list.
849	 */
850	if (doadjust)
851		fts_padjust(sp, head);
852
853	/*
854	 * If not changing directories, reset the path back to original
855	 * state.
856	 */
857	if (ISSET(FTS_NOCHDIR)) {
858		if (len == sp->fts_pathlen || nitems == 0)
859			--cp;
860		*cp = '\0';
861	}
862
863	/*
864	 * If descended after called from fts_children or after called from
865	 * fts_read and nothing found, get back.  At the root level we use
866	 * the saved fd; if one of fts_open()'s arguments is a relative path
867	 * to an empty directory, we wind up here with no other way back.  If
868	 * can't get back, we're done.
869	 */
870	if (descend && (type == BCHILD || !nitems) &&
871	    (cur->fts_level == FTS_ROOTLEVEL ?
872	    FCHDIR(sp, sp->fts_rfd) :
873	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
874		cur->fts_info = FTS_ERR;
875		SET(FTS_STOP);
876		return (NULL);
877	}
878
879	/* If didn't find anything, return NULL. */
880	if (!nitems) {
881		if (type == BREAD)
882			cur->fts_info = FTS_DP;
883		return (NULL);
884	}
885
886	/* Sort the entries. */
887	if (sp->fts_compar && nitems > 1)
888		head = fts_sort(sp, head, nitems);
889	return (head);
890}
891
892static u_short
893fts_stat(sp, p, follow)
894	FTS *sp;
895	FTSENT *p;
896	int follow;
897{
898	FTSENT *t;
899	dev_t dev;
900	ino_t ino;
901	struct stat *sbp, sb;
902	int saved_errno;
903
904	/* If user needs stat info, stat buffer already allocated. */
905	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
906
907#ifdef FTS_WHITEOUT
908	/* Check for whiteout. */
909	if (p->fts_flags & FTS_ISW) {
910		if (sbp != &sb) {
911			memset(sbp, '\0', sizeof(*sbp));
912			sbp->st_mode = S_IFWHT;
913		}
914		return (FTS_W);
915	}
916#endif
917
918	/*
919	 * If doing a logical walk, or application requested FTS_FOLLOW, do
920	 * a stat(2).  If that fails, check for a non-existent symlink.  If
921	 * fail, set the errno from the stat call.
922	 */
923	if (ISSET(FTS_LOGICAL) || follow) {
924		if (stat(p->fts_accpath, sbp)) {
925			saved_errno = errno;
926			if (!lstat(p->fts_accpath, sbp)) {
927				errno = 0;
928				return (FTS_SLNONE);
929			}
930			p->fts_errno = saved_errno;
931			goto err;
932		}
933	} else if (lstat(p->fts_accpath, sbp)) {
934		p->fts_errno = errno;
935err:		memset(sbp, 0, sizeof(struct stat));
936		return (FTS_NS);
937	}
938
939	if (S_ISDIR(sbp->st_mode)) {
940		/*
941		 * Set the device/inode.  Used to find cycles and check for
942		 * crossing mount points.  Also remember the link count, used
943		 * in fts_build to limit the number of stat calls.  It is
944		 * understood that these fields are only referenced if fts_info
945		 * is set to FTS_D.
946		 */
947		dev = p->fts_dev = sbp->st_dev;
948		ino = p->fts_ino = sbp->st_ino;
949		p->fts_nlink = sbp->st_nlink;
950
951		if (ISDOT(p->fts_name))
952			return (FTS_DOT);
953
954		/*
955		 * Cycle detection is done by brute force when the directory
956		 * is first encountered.  If the tree gets deep enough or the
957		 * number of symbolic links to directories is high enough,
958		 * something faster might be worthwhile.
959		 */
960		for (t = p->fts_parent;
961		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
962			if (ino == t->fts_ino && dev == t->fts_dev) {
963				p->fts_cycle = t;
964				return (FTS_DC);
965			}
966		return (FTS_D);
967	}
968	if (S_ISLNK(sbp->st_mode))
969		return (FTS_SL);
970	if (S_ISREG(sbp->st_mode))
971		return (FTS_F);
972	return (FTS_DEFAULT);
973}
974
975/*
976 * The comparison function takes pointers to pointers to FTSENT structures.
977 * Qsort wants a comparison function that takes pointers to void.
978 * (Both with appropriate levels of const-poisoning, of course!)
979 * Use a trampoline function to deal with the difference.
980 */
981static int
982fts_compar(const void *a, const void *b)
983{
984	FTS *parent;
985
986	parent = (*(const FTSENT * const *)a)->fts_fts;
987	return (*parent->fts_compar)(a, b);
988}
989
990static FTSENT *
991fts_sort(sp, head, nitems)
992	FTS *sp;
993	FTSENT *head;
994	int nitems;
995{
996	FTSENT **ap, *p;
997
998	/*
999	 * Construct an array of pointers to the structures and call qsort(3).
1000	 * Reassemble the array in the order returned by qsort.  If unable to
1001	 * sort for memory reasons, return the directory entries in their
1002	 * current order.  Allocate enough space for the current needs plus
1003	 * 40 so don't realloc one entry at a time.
1004	 */
1005	if (nitems > sp->fts_nitems) {
1006		sp->fts_nitems = nitems + 40;
1007		if ((sp->fts_array = reallocf(sp->fts_array,
1008		    sp->fts_nitems * sizeof(FTSENT *))) == NULL) {
1009			sp->fts_nitems = 0;
1010			return (head);
1011		}
1012	}
1013	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
1014		*ap++ = p;
1015	qsort(sp->fts_array, nitems, sizeof(FTSENT *), fts_compar);
1016	for (head = *(ap = sp->fts_array); --nitems; ++ap)
1017		ap[0]->fts_link = ap[1];
1018	ap[0]->fts_link = NULL;
1019	return (head);
1020}
1021
1022static FTSENT *
1023fts_alloc(sp, name, namelen)
1024	FTS *sp;
1025	char *name;
1026	int namelen;
1027{
1028	FTSENT *p;
1029	size_t len;
1030
1031	struct ftsent_withstat {
1032		FTSENT	ent;
1033		struct	stat statbuf;
1034	};
1035
1036	/*
1037	 * The file name is a variable length array and no stat structure is
1038	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
1039	 * structure, the file name and the stat structure in one chunk, but
1040	 * be careful that the stat structure is reasonably aligned.
1041	 */
1042	if (ISSET(FTS_NOSTAT))
1043		len = sizeof(FTSENT) + namelen + 1;
1044	else
1045		len = sizeof(struct ftsent_withstat) + namelen + 1;
1046
1047	if ((p = malloc(len)) == NULL)
1048		return (NULL);
1049
1050	if (ISSET(FTS_NOSTAT)) {
1051		p->fts_name = (char *)(p + 1);
1052		p->fts_statp = NULL;
1053	} else {
1054		p->fts_name = (char *)((struct ftsent_withstat *)p + 1);
1055		p->fts_statp = &((struct ftsent_withstat *)p)->statbuf;
1056	}
1057
1058	/* Copy the name and guarantee NUL termination. */
1059	memcpy(p->fts_name, name, namelen);
1060	p->fts_name[namelen] = '\0';
1061	p->fts_namelen = namelen;
1062	p->fts_path = sp->fts_path;
1063	p->fts_errno = 0;
1064	p->fts_flags = 0;
1065	p->fts_instr = FTS_NOINSTR;
1066	p->fts_number = 0;
1067	p->fts_pointer = NULL;
1068	p->fts_fts = sp;
1069	return (p);
1070}
1071
1072static void
1073fts_lfree(head)
1074	FTSENT *head;
1075{
1076	FTSENT *p;
1077
1078	/* Free a linked list of structures. */
1079	while ((p = head)) {
1080		head = head->fts_link;
1081		free(p);
1082	}
1083}
1084
1085/*
1086 * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
1087 * Most systems will allow creation of paths much longer than MAXPATHLEN, even
1088 * though the kernel won't resolve them.  Add the size (not just what's needed)
1089 * plus 256 bytes so don't realloc the path 2 bytes at a time.
1090 */
1091static int
1092fts_palloc(sp, more)
1093	FTS *sp;
1094	size_t more;
1095{
1096
1097	sp->fts_pathlen += more + 256;
1098	/*
1099	 * Check for possible wraparound.  In an FTS, fts_pathlen is
1100	 * a signed int but in an FTSENT it is an unsigned short.
1101	 * We limit fts_pathlen to USHRT_MAX to be safe in both cases.
1102	 */
1103	if (sp->fts_pathlen < 0 || sp->fts_pathlen >= USHRT_MAX) {
1104		if (sp->fts_path)
1105			free(sp->fts_path);
1106		sp->fts_path = NULL;
1107		errno = ENAMETOOLONG;
1108		return (1);
1109	}
1110	sp->fts_path = reallocf(sp->fts_path, sp->fts_pathlen);
1111	return (sp->fts_path == NULL);
1112}
1113
1114/*
1115 * When the path is realloc'd, have to fix all of the pointers in structures
1116 * already returned.
1117 */
1118static void
1119fts_padjust(sp, head)
1120	FTS *sp;
1121	FTSENT *head;
1122{
1123	FTSENT *p;
1124	char *addr = sp->fts_path;
1125
1126#define	ADJUST(p) do {							\
1127	if ((p)->fts_accpath != (p)->fts_name) {			\
1128		(p)->fts_accpath =					\
1129		    (char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
1130	}								\
1131	(p)->fts_path = addr;						\
1132} while (0)
1133	/* Adjust the current set of children. */
1134	for (p = sp->fts_child; p; p = p->fts_link)
1135		ADJUST(p);
1136
1137	/* Adjust the rest of the tree, including the current level. */
1138	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1139		ADJUST(p);
1140		p = p->fts_link ? p->fts_link : p->fts_parent;
1141	}
1142}
1143
1144static size_t
1145fts_maxarglen(argv)
1146	char * const *argv;
1147{
1148	size_t len, max;
1149
1150	for (max = 0; *argv; ++argv)
1151		if ((len = strlen(*argv)) > max)
1152			max = len;
1153	return (max + 1);
1154}
1155
1156/*
1157 * Change to dir specified by fd or p->fts_accpath without getting
1158 * tricked by someone changing the world out from underneath us.
1159 * Assumes p->fts_dev and p->fts_ino are filled in.
1160 */
1161static int
1162fts_safe_changedir(sp, p, fd, path)
1163	FTS *sp;
1164	FTSENT *p;
1165	int fd;
1166	char *path;
1167{
1168	int ret, oerrno, newfd;
1169	struct stat sb;
1170
1171	newfd = fd;
1172	if (ISSET(FTS_NOCHDIR))
1173		return (0);
1174	if (fd < 0 && (newfd = _open(path, O_RDONLY, 0)) < 0)
1175		return (-1);
1176	if (_fstat(newfd, &sb)) {
1177		ret = -1;
1178		goto bail;
1179	}
1180	if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) {
1181		errno = ENOENT;		/* disinformation */
1182		ret = -1;
1183		goto bail;
1184	}
1185	ret = fchdir(newfd);
1186bail:
1187	oerrno = errno;
1188	if (fd < 0)
1189		(void)_close(newfd);
1190	errno = oerrno;
1191	return (ret);
1192}
1193
1194/*
1195 * Check if the filesystem for "ent" has UFS-style links.
1196 */
1197static int
1198fts_ufslinks(FTS *sp, const FTSENT *ent)
1199{
1200	struct _fts_private *priv;
1201	const char **cpp;
1202
1203	priv = (struct _fts_private *)sp;
1204	/*
1205	 * If this node's device is different from the previous, grab
1206	 * the filesystem information, and decide on the reliability
1207	 * of the link information from this filesystem for stat(2)
1208	 * avoidance.
1209	 */
1210	if (priv->ftsp_dev != ent->fts_dev) {
1211		if (statfs(ent->fts_path, &priv->ftsp_statfs) != -1) {
1212			priv->ftsp_dev = ent->fts_dev;
1213			priv->ftsp_linksreliable = 0;
1214			for (cpp = ufslike_filesystems; *cpp; cpp++) {
1215				if (strcmp(priv->ftsp_statfs.f_fstypename,
1216				    *cpp) == 0) {
1217					priv->ftsp_linksreliable = 1;
1218					break;
1219				}
1220			}
1221		} else {
1222			priv->ftsp_linksreliable = 0;
1223		}
1224	}
1225	return (priv->ftsp_linksreliable);
1226}
1227