fts.c revision 1.43
1/*	$NetBSD: fts.c,v 1.43 2012/03/13 21:13:35 christos Exp $	*/
2
3/*-
4 * Copyright (c) 1990, 1993, 1994
5 *	The Regents of the University of California.  All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice, this list of conditions and the following disclaimer in the
14 *    documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 *    may be used to endorse or promote products derived from this software
17 *    without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 */
31
32#if HAVE_NBTOOL_CONFIG_H
33#include "nbtool_config.h"
34#endif
35
36#include <sys/cdefs.h>
37#if defined(LIBC_SCCS) && !defined(lint)
38#if 0
39static char sccsid[] = "@(#)fts.c	8.6 (Berkeley) 8/14/94";
40#else
41__RCSID("$NetBSD: fts.c,v 1.43 2012/03/13 21:13:35 christos Exp $");
42#endif
43#endif /* LIBC_SCCS and not lint */
44
45#include "namespace.h"
46#include <sys/param.h>
47#include <sys/stat.h>
48
49#include <assert.h>
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
58#if ! HAVE_NBTOOL_CONFIG_H
59#define	HAVE_STRUCT_DIRENT_D_NAMLEN
60#endif
61
62static FTSENT	*fts_alloc(FTS *, const char *, size_t);
63static FTSENT	*fts_build(FTS *, int);
64static void	 fts_free(FTSENT *);
65static void	 fts_lfree(FTSENT *);
66static void	 fts_load(FTS *, FTSENT *);
67static size_t	 fts_maxarglen(char * const *);
68static size_t	 fts_pow2(size_t);
69static int	 fts_palloc(FTS *, size_t);
70static void	 fts_padjust(FTS *, FTSENT *);
71static FTSENT	*fts_sort(FTS *, FTSENT *, size_t);
72static unsigned short fts_stat(FTS *, FTSENT *, int);
73static int	 fts_safe_changedir(const FTS *, const FTSENT *, int,
74    const char *);
75
76#if defined(ALIGNBYTES) && defined(ALIGN)
77#define	FTS_ALLOC_ALIGNED	1
78#else
79#undef	FTS_ALLOC_ALIGNED
80#endif
81
82#ifndef fts_namelen_truncate
83#define fts_namelen_truncate(a)	\
84    ((a) > UINT_MAX ? UINT_MAX : (unsigned int)(a))
85#endif
86#ifndef fts_pathlen_truncate
87#define fts_pathlen_truncate(a)	\
88    ((a) > UINT_MAX ? UINT_MAX : (unsigned int)(a))
89#endif
90#ifndef ftsent_pathlen_truncate
91#define ftsent_pathlen_truncate(a) \
92    ((a) > UINT_MAX ? UINT_MAX : (unsigned int)(a))
93#endif
94#ifndef ftsent_nitems_truncate
95#define ftsent_nitems_truncate(a) \
96    ((a) > UINT_MAX ? UINT_MAX : (unsigned int)(a))
97#endif
98
99#define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
100
101#define	CLR(opt)	(sp->fts_options &= ~(opt))
102#define	ISSET(opt)	(sp->fts_options & (opt))
103#define	SET(opt)	(sp->fts_options |= (opt))
104
105#define	CHDIR(sp, path)	(!ISSET(FTS_NOCHDIR) && chdir(path))
106#define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
107
108/* fts_build flags */
109#define	BCHILD		1		/* fts_children */
110#define	BNAMES		2		/* fts_children, names only */
111#define	BREAD		3		/* fts_read */
112
113#ifndef DTF_HIDEW
114#undef FTS_WHITEOUT
115#endif
116
117FTS *
118fts_open(char * const *argv, int options,
119    int (*compar)(const FTSENT **, const FTSENT **))
120{
121	FTS *sp;
122	FTSENT *p, *root;
123	size_t nitems;
124	FTSENT *parent, *tmp = NULL;	/* pacify gcc */
125	size_t len;
126
127	_DIAGASSERT(argv != NULL);
128
129	/* Options check. */
130	if (options & ~FTS_OPTIONMASK) {
131		errno = EINVAL;
132		return (NULL);
133	}
134
135	/* Allocate/initialize the stream */
136	if ((sp = malloc((unsigned int)sizeof(FTS))) == NULL)
137		return (NULL);
138	memset(sp, 0, sizeof(FTS));
139	sp->fts_compar = compar;
140	sp->fts_options = options;
141
142	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
143	if (ISSET(FTS_LOGICAL))
144		SET(FTS_NOCHDIR);
145
146	/*
147	 * Start out with 1K of path space, and enough, in any case,
148	 * to hold the user's paths.
149	 */
150	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
151		goto mem1;
152
153	/* Allocate/initialize root's parent. */
154	if ((parent = fts_alloc(sp, "", 0)) == NULL)
155		goto mem2;
156	parent->fts_level = FTS_ROOTPARENTLEVEL;
157
158	/* Allocate/initialize root(s). */
159	for (root = NULL, nitems = 0; *argv; ++argv, ++nitems) {
160		/* Don't allow zero-length paths. */
161		if ((len = strlen(*argv)) == 0) {
162			errno = ENOENT;
163			goto mem3;
164		}
165
166		if ((p = fts_alloc(sp, *argv, len)) == NULL)
167			goto mem3;
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 insure
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#ifndef O_CLOEXEC
215#define O_CLOEXEC 0
216#endif
217	if (!ISSET(FTS_NOCHDIR)) {
218		if ((sp->fts_rfd = open(".", O_RDONLY | O_CLOEXEC, 0)) == -1)
219			SET(FTS_NOCHDIR);
220	}
221
222	if (nitems == 0)
223		fts_free(parent);
224
225	return (sp);
226
227mem3:	fts_lfree(root);
228	fts_free(parent);
229mem2:	free(sp->fts_path);
230mem1:	free(sp);
231	return (NULL);
232}
233
234static void
235fts_load(FTS *sp, FTSENT *p)
236{
237	size_t len;
238	char *cp;
239
240	_DIAGASSERT(sp != NULL);
241	_DIAGASSERT(p != NULL);
242
243	/*
244	 * Load the stream structure for the next traversal.  Since we don't
245	 * actually enter the directory until after the preorder visit, set
246	 * the fts_accpath field specially so the chdir gets done to the right
247	 * place and the user can access the first node.  From fts_open it's
248	 * known that the path will fit.
249	 */
250	len = p->fts_pathlen = p->fts_namelen;
251	memmove(sp->fts_path, p->fts_name, len + 1);
252	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
253		len = strlen(++cp);
254		memmove(p->fts_name, cp, len + 1);
255		p->fts_namelen = fts_namelen_truncate(len);
256	}
257	p->fts_accpath = p->fts_path = sp->fts_path;
258	sp->fts_dev = p->fts_dev;
259}
260
261int
262fts_close(FTS *sp)
263{
264	FTSENT *freep, *p;
265	int saved_errno = 0;
266
267	_DIAGASSERT(sp != NULL);
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		if (sp->fts_cur->fts_flags & FTS_SYMFOLLOW)
276			(void)close(sp->fts_cur->fts_symfd);
277		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
278			freep = p;
279			p = p->fts_link ? p->fts_link : p->fts_parent;
280			fts_free(freep);
281		}
282		fts_free(p);
283	}
284
285	/* Free up child linked list, sort array, path buffer. */
286	if (sp->fts_child)
287		fts_lfree(sp->fts_child);
288	if (sp->fts_array)
289		free(sp->fts_array);
290	free(sp->fts_path);
291
292	/* Return to original directory, save errno if necessary. */
293	if (!ISSET(FTS_NOCHDIR)) {
294		if (fchdir(sp->fts_rfd) == -1)
295			saved_errno = errno;
296		(void)close(sp->fts_rfd);
297	}
298
299	/* Free up the stream pointer. */
300	free(sp);
301	if (saved_errno) {
302		errno = saved_errno;
303		return -1;
304	}
305
306	return 0;
307}
308
309#if !defined(__FTS_COMPAT_TAILINGSLASH)
310
311/*
312 * Special case of "/" at the end of the path so that slashes aren't
313 * appended which would cause paths to be written as "....//foo".
314 */
315#define	NAPPEND(p)							\
316	(p->fts_path[p->fts_pathlen - 1] == '/'				\
317	    ? p->fts_pathlen - 1 : p->fts_pathlen)
318
319#else /* !defined(__FTS_COMPAT_TAILINGSLASH) */
320
321/*
322 * compatibility with the old behaviour.
323 *
324 * Special case a root of "/" so that slashes aren't appended which would
325 * cause paths to be written as "//foo".
326 */
327
328#define	NAPPEND(p)							\
329	(p->fts_level == FTS_ROOTLEVEL && p->fts_pathlen == 1 &&	\
330	    p->fts_path[0] == '/' ? 0 : p->fts_pathlen)
331
332#endif /* !defined(__FTS_COMPAT_TAILINGSLASH) */
333
334FTSENT *
335fts_read(FTS *sp)
336{
337	FTSENT *p, *tmp;
338	int instr;
339	char *t;
340	int saved_errno;
341
342	_DIAGASSERT(sp != NULL);
343
344	/* If finished or unrecoverable error, return NULL. */
345	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
346		return (NULL);
347
348	/* Set current node pointer. */
349	p = sp->fts_cur;
350
351	/* Save and zero out user instructions. */
352	instr = p->fts_instr;
353	p->fts_instr = FTS_NOINSTR;
354
355	/* Any type of file may be re-visited; re-stat and re-turn. */
356	if (instr == FTS_AGAIN) {
357		p->fts_info = fts_stat(sp, p, 0);
358		return (p);
359	}
360
361	/*
362	 * Following a symlink -- SLNONE test allows application to see
363	 * SLNONE and recover.  If indirecting through a symlink, have
364	 * keep a pointer to current location.  If unable to get that
365	 * pointer, follow fails.
366	 */
367	if (instr == FTS_FOLLOW &&
368	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
369		p->fts_info = fts_stat(sp, p, 1);
370		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
371			if ((p->fts_symfd = open(".", O_RDONLY | O_CLOEXEC, 0))
372			    == -1) {
373				p->fts_errno = errno;
374				p->fts_info = FTS_ERR;
375			} else
376				p->fts_flags |= FTS_SYMFOLLOW;
377		}
378		return (p);
379	}
380
381	/* Directory in pre-order. */
382	if (p->fts_info == FTS_D) {
383		/* If skipped or crossed mount point, do post-order visit. */
384		if (instr == FTS_SKIP ||
385		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
386			if (p->fts_flags & FTS_SYMFOLLOW)
387				(void)close(p->fts_symfd);
388			if (sp->fts_child) {
389				fts_lfree(sp->fts_child);
390				sp->fts_child = NULL;
391			}
392			p->fts_info = FTS_DP;
393			return (p);
394		}
395
396		/* Rebuild if only read the names and now traversing. */
397		if (sp->fts_child && ISSET(FTS_NAMEONLY)) {
398			CLR(FTS_NAMEONLY);
399			fts_lfree(sp->fts_child);
400			sp->fts_child = NULL;
401		}
402
403		/*
404		 * Cd to the subdirectory.
405		 *
406		 * If have already read and now fail to chdir, whack the list
407		 * to make the names come out right, and set the parent errno
408		 * so the application will eventually get an error condition.
409		 * Set the FTS_DONTCHDIR flag so that when we logically change
410		 * directories back to the parent we don't do a chdir.
411		 *
412		 * If haven't read do so.  If the read fails, fts_build sets
413		 * FTS_STOP or the fts_info field of the node.
414		 */
415		if (sp->fts_child) {
416			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
417				p->fts_errno = errno;
418				p->fts_flags |= FTS_DONTCHDIR;
419				for (p = sp->fts_child; p; p = p->fts_link)
420					p->fts_accpath =
421					    p->fts_parent->fts_accpath;
422			}
423		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
424			if (ISSET(FTS_STOP))
425				return (NULL);
426			return (p);
427		}
428		p = sp->fts_child;
429		sp->fts_child = NULL;
430		goto name;
431	}
432
433	/* Move to the next node on this level. */
434next:	tmp = p;
435	if ((p = p->fts_link) != NULL) {
436		fts_free(tmp);
437
438		/*
439		 * If reached the top, return to the original directory, and
440		 * load the paths for the next root.
441		 */
442		if (p->fts_level == FTS_ROOTLEVEL) {
443			if (FCHDIR(sp, sp->fts_rfd)) {
444				SET(FTS_STOP);
445				return (NULL);
446			}
447			fts_load(sp, p);
448			return (sp->fts_cur = p);
449		}
450
451		/*
452		 * User may have called fts_set on the node.  If skipped,
453		 * ignore.  If followed, get a file descriptor so we can
454		 * get back if necessary.
455		 */
456		if (p->fts_instr == FTS_SKIP)
457			goto next;
458		if (p->fts_instr == FTS_FOLLOW) {
459			p->fts_info = fts_stat(sp, p, 1);
460			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
461				if ((p->fts_symfd =
462				    open(".", O_RDONLY | O_CLOEXEC, 0)) == -1) {
463					p->fts_errno = errno;
464					p->fts_info = FTS_ERR;
465				} else
466					p->fts_flags |= FTS_SYMFOLLOW;
467			}
468			p->fts_instr = FTS_NOINSTR;
469		}
470
471name:		t = sp->fts_path + NAPPEND(p->fts_parent);
472		*t++ = '/';
473		memmove(t, p->fts_name, (size_t)(p->fts_namelen + 1));
474		return (sp->fts_cur = p);
475	}
476
477	/* Move up to the parent node. */
478	p = tmp->fts_parent;
479	fts_free(tmp);
480
481	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
482		/*
483		 * Done; free everything up and set errno to 0 so the user
484		 * can distinguish between error and EOF.
485		 */
486		fts_free(p);
487		errno = 0;
488		return (sp->fts_cur = NULL);
489	}
490
491	/* Nul terminate the pathname. */
492	sp->fts_path[p->fts_pathlen] = '\0';
493
494	/*
495	 * Return to the parent directory.  If at a root node or came through
496	 * a symlink, go back through the file descriptor.  Otherwise, cd up
497	 * one directory.
498	 */
499	if (p->fts_level == FTS_ROOTLEVEL) {
500		if (FCHDIR(sp, sp->fts_rfd)) {
501			SET(FTS_STOP);
502			return (NULL);
503		}
504	} else if (p->fts_flags & FTS_SYMFOLLOW) {
505		if (FCHDIR(sp, p->fts_symfd)) {
506			saved_errno = errno;
507			(void)close(p->fts_symfd);
508			errno = saved_errno;
509			SET(FTS_STOP);
510			return (NULL);
511		}
512		(void)close(p->fts_symfd);
513	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
514	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
515		SET(FTS_STOP);
516		return (NULL);
517	}
518	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
519	return (sp->fts_cur = p);
520}
521
522/*
523 * Fts_set takes the stream as an argument although it's not used in this
524 * implementation; it would be necessary if anyone wanted to add global
525 * semantics to fts using fts_set.  An error return is allowed for similar
526 * reasons.
527 */
528/* ARGSUSED */
529int
530fts_set(FTS *sp, FTSENT *p, int instr)
531{
532
533	_DIAGASSERT(sp != NULL);
534	_DIAGASSERT(p != NULL);
535
536	if (instr && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
537	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
538		errno = EINVAL;
539		return (1);
540	}
541	p->fts_instr = instr;
542	return (0);
543}
544
545FTSENT *
546fts_children(FTS *sp, int instr)
547{
548	FTSENT *p;
549	int fd;
550
551	_DIAGASSERT(sp != NULL);
552
553	if (instr && instr != FTS_NAMEONLY) {
554		errno = EINVAL;
555		return (NULL);
556	}
557
558	/* Set current node pointer. */
559	p = sp->fts_cur;
560
561	/*
562	 * Errno set to 0 so user can distinguish empty directory from
563	 * an error.
564	 */
565	errno = 0;
566
567	/* Fatal errors stop here. */
568	if (ISSET(FTS_STOP))
569		return (NULL);
570
571	/* Return logical hierarchy of user's arguments. */
572	if (p->fts_info == FTS_INIT)
573		return (p->fts_link);
574
575	/*
576	 * If not a directory being visited in pre-order, stop here.  Could
577	 * allow FTS_DNR, assuming the user has fixed the problem, but the
578	 * same effect is available with FTS_AGAIN.
579	 */
580	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
581		return (NULL);
582
583	/* Free up any previous child list. */
584	if (sp->fts_child)
585		fts_lfree(sp->fts_child);
586
587	if (instr == FTS_NAMEONLY) {
588		SET(FTS_NAMEONLY);
589		instr = BNAMES;
590	} else
591		instr = BCHILD;
592
593	/*
594	 * If using chdir on a relative path and called BEFORE fts_read does
595	 * its chdir to the root of a traversal, we can lose -- we need to
596	 * chdir into the subdirectory, and we don't know where the current
597	 * directory is, so we can't get back so that the upcoming chdir by
598	 * fts_read will work.
599	 */
600	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
601	    ISSET(FTS_NOCHDIR))
602		return (sp->fts_child = fts_build(sp, instr));
603
604	if ((fd = open(".", O_RDONLY, 0)) == -1)
605		return (sp->fts_child = NULL);
606	sp->fts_child = fts_build(sp, instr);
607	if (fchdir(fd)) {
608		(void)close(fd);
609		return (NULL);
610	}
611	(void)close(fd);
612	return (sp->fts_child);
613}
614
615/*
616 * This is the tricky part -- do not casually change *anything* in here.  The
617 * idea is to build the linked list of entries that are used by fts_children
618 * and fts_read.  There are lots of special cases.
619 *
620 * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
621 * set and it's a physical walk (so that symbolic links can't be directories),
622 * we can do things quickly.  First, if it's a 4.4BSD file system, the type
623 * of the file is in the directory entry.  Otherwise, we assume that the number
624 * of subdirectories in a node is equal to the number of links to the parent.
625 * The former skips all stat calls.  The latter skips stat calls in any leaf
626 * directories and for any files after the subdirectories in the directory have
627 * been found, cutting the stat calls by about 2/3.
628 */
629static FTSENT *
630fts_build(FTS *sp, int type)
631{
632	struct dirent *dp;
633	FTSENT *p, *head;
634	size_t nitems;
635	FTSENT *cur, *tail;
636	DIR *dirp;
637	void *oldaddr;
638	size_t dnamlen;
639	int cderrno, descend, level, nlinks, saved_errno, nostat, doadjust;
640	size_t len, maxlen;
641#ifdef FTS_WHITEOUT
642	int oflag;
643#endif
644	char *cp = NULL;	/* pacify gcc */
645
646	_DIAGASSERT(sp != NULL);
647
648	/* Set current node pointer. */
649	cur = sp->fts_cur;
650
651	/*
652	 * Open the directory for reading.  If this fails, we're done.
653	 * If being called from fts_read, set the fts_info field.
654	 */
655#ifdef FTS_WHITEOUT
656	if (ISSET(FTS_WHITEOUT))
657		oflag = DTF_NODUP|DTF_REWIND;
658	else
659		oflag = DTF_HIDEW|DTF_NODUP|DTF_REWIND;
660#else
661#define	__opendir2(path, flag) opendir(path)
662#endif
663	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
664		if (type == BREAD) {
665			cur->fts_info = FTS_DNR;
666			cur->fts_errno = errno;
667		}
668		return (NULL);
669	}
670
671	/*
672	 * Nlinks is the number of possible entries of type directory in the
673	 * directory if we're cheating on stat calls, 0 if we're not doing
674	 * any stat calls at all, -1 if we're doing stats on everything.
675	 */
676	if (type == BNAMES) {
677		nlinks = 0;
678		nostat = 1;
679	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
680		nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
681		nostat = 1;
682	} else {
683		nlinks = -1;
684		nostat = 0;
685	}
686
687#ifdef notdef
688	(void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
689	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
690	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
691#endif
692	/*
693	 * If we're going to need to stat anything or we want to descend
694	 * and stay in the directory, chdir.  If this fails we keep going,
695	 * but set a flag so we don't chdir after the post-order visit.
696	 * We won't be able to stat anything, but we can still return the
697	 * names themselves.  Note, that since fts_read won't be able to
698	 * chdir into the directory, it will have to return different path
699	 * names than before, i.e. "a/b" instead of "b".  Since the node
700	 * has already been visited in pre-order, have to wait until the
701	 * post-order visit to return the error.  There is a special case
702	 * here, if there was nothing to stat then it's not an error to
703	 * not be able to stat.  This is all fairly nasty.  If a program
704	 * needed sorted entries or stat information, they had better be
705	 * checking FTS_NS on the returned nodes.
706	 */
707	cderrno = 0;
708	if (nlinks || type == BREAD) {
709		if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) {
710			if (nlinks && type == BREAD)
711				cur->fts_errno = errno;
712			cur->fts_flags |= FTS_DONTCHDIR;
713			descend = 0;
714			cderrno = errno;
715		} else
716			descend = 1;
717	} else
718		descend = 0;
719
720	/*
721	 * Figure out the max file name length that can be stored in the
722	 * current path -- the inner loop allocates more path as necessary.
723	 * We really wouldn't have to do the maxlen calculations here, we
724	 * could do them in fts_read before returning the path, but it's a
725	 * lot easier here since the length is part of the dirent structure.
726	 *
727	 * If not changing directories set a pointer so that can just append
728	 * each new name into the path.
729	 */
730	len = NAPPEND(cur);
731	if (ISSET(FTS_NOCHDIR)) {
732		cp = sp->fts_path + len;
733		*cp++ = '/';
734	}
735	len++;
736	maxlen = sp->fts_pathlen - len;
737
738#if defined(__FTS_COMPAT_LEVEL)
739	if (cur->fts_level == SHRT_MAX) {
740		(void)closedir(dirp);
741		cur->fts_info = FTS_ERR;
742		SET(FTS_STOP);
743		errno = ENAMETOOLONG;
744		return (NULL);
745	}
746#endif
747
748	level = cur->fts_level + 1;
749
750	/* Read the directory, attaching each entry to the `link' pointer. */
751	doadjust = 0;
752	for (head = tail = NULL, nitems = 0; (dp = readdir(dirp)) != NULL;) {
753
754		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
755			continue;
756
757#if defined(HAVE_STRUCT_DIRENT_D_NAMLEN)
758		dnamlen = dp->d_namlen;
759#else
760		dnamlen = strlen(dp->d_name);
761#endif
762		if ((p = fts_alloc(sp, dp->d_name, dnamlen)) == NULL)
763			goto mem1;
764		if (dnamlen >= maxlen) {	/* include space for NUL */
765			oldaddr = sp->fts_path;
766			if (fts_palloc(sp, dnamlen + len + 1)) {
767				/*
768				 * No more memory for path or structures.  Save
769				 * errno, free up the current structure and the
770				 * structures already allocated.
771				 */
772mem1:				saved_errno = errno;
773				if (p)
774					fts_free(p);
775				fts_lfree(head);
776				(void)closedir(dirp);
777				errno = saved_errno;
778				cur->fts_info = FTS_ERR;
779				SET(FTS_STOP);
780				return (NULL);
781			}
782			/* Did realloc() change the pointer? */
783			if (oldaddr != sp->fts_path) {
784				doadjust = 1;
785				if (ISSET(FTS_NOCHDIR))
786					cp = sp->fts_path + len;
787			}
788			maxlen = sp->fts_pathlen - len;
789		}
790
791#if defined(__FTS_COMPAT_LENGTH)
792		if (len + dnamlen >= USHRT_MAX) {
793			/*
794			 * In an FTSENT, fts_pathlen is an unsigned short
795			 * so it is possible to wraparound here.
796			 * If we do, free up the current structure and the
797			 * structures already allocated, then error out
798			 * with ENAMETOOLONG.
799			 */
800			fts_free(p);
801			fts_lfree(head);
802			(void)closedir(dirp);
803			cur->fts_info = FTS_ERR;
804			SET(FTS_STOP);
805			errno = ENAMETOOLONG;
806			return (NULL);
807		}
808#endif
809		p->fts_level = level;
810		p->fts_pathlen = ftsent_pathlen_truncate(len + dnamlen);
811		p->fts_parent = sp->fts_cur;
812
813#ifdef FTS_WHITEOUT
814		if (dp->d_type == DT_WHT)
815			p->fts_flags |= FTS_ISW;
816#endif
817
818		if (cderrno) {
819			if (nlinks) {
820				p->fts_info = FTS_NS;
821				p->fts_errno = cderrno;
822			} else
823				p->fts_info = FTS_NSOK;
824			p->fts_accpath = cur->fts_accpath;
825		} else if (nlinks == 0
826#ifdef DT_DIR
827		    || (nostat &&
828		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
829#endif
830		    ) {
831			p->fts_accpath =
832			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
833			p->fts_info = FTS_NSOK;
834		} else {
835			/* Build a file name for fts_stat to stat. */
836			if (ISSET(FTS_NOCHDIR)) {
837				p->fts_accpath = p->fts_path;
838				memmove(cp, p->fts_name,
839				        (size_t)(p->fts_namelen + 1));
840			} else
841				p->fts_accpath = p->fts_name;
842			/* Stat it. */
843			p->fts_info = fts_stat(sp, p, 0);
844
845			/* Decrement link count if applicable. */
846			if (nlinks > 0 && (p->fts_info == FTS_D ||
847			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
848				--nlinks;
849		}
850
851		/* We walk in directory order so "ls -f" doesn't get upset. */
852		p->fts_link = NULL;
853		if (head == NULL)
854			head = tail = p;
855		else {
856			tail->fts_link = p;
857			tail = p;
858		}
859		++nitems;
860	}
861	(void)closedir(dirp);
862
863	/*
864	 * If had to realloc the path, adjust the addresses for the rest
865	 * of the tree.
866	 */
867	if (doadjust)
868		fts_padjust(sp, head);
869
870	/*
871	 * If not changing directories, reset the path back to original
872	 * state.
873	 */
874	if (ISSET(FTS_NOCHDIR)) {
875		if (len == sp->fts_pathlen || nitems == 0)
876			--cp;
877		*cp = '\0';
878	}
879
880	/*
881	 * If descended after called from fts_children or after called from
882	 * fts_read and nothing found, get back.  At the root level we use
883	 * the saved fd; if one of fts_open()'s arguments is a relative path
884	 * to an empty directory, we wind up here with no other way back.  If
885	 * can't get back, we're done.
886	 */
887	if (descend && (type == BCHILD || !nitems) &&
888	    (cur->fts_level == FTS_ROOTLEVEL ?
889	    FCHDIR(sp, sp->fts_rfd) :
890	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
891		cur->fts_info = FTS_ERR;
892		SET(FTS_STOP);
893		return (NULL);
894	}
895
896	/* If didn't find anything, return NULL. */
897	if (!nitems) {
898		if (type == BREAD)
899			cur->fts_info = FTS_DP;
900		return (NULL);
901	}
902
903	/* Sort the entries. */
904	if (sp->fts_compar && nitems > 1)
905		head = fts_sort(sp, head, nitems);
906	return (head);
907}
908
909static unsigned short
910fts_stat(FTS *sp, FTSENT *p, int follow)
911{
912	FTSENT *t;
913	dev_t dev;
914	__fts_ino_t ino;
915	__fts_stat_t *sbp, sb;
916	int saved_errno;
917
918	_DIAGASSERT(sp != NULL);
919	_DIAGASSERT(p != NULL);
920
921	/* If user needs stat info, stat buffer already allocated. */
922	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
923
924#ifdef FTS_WHITEOUT
925	/* check for whiteout */
926	if (p->fts_flags & FTS_ISW) {
927		if (sbp != &sb) {
928			memset(sbp, '\0', sizeof (*sbp));
929			sbp->st_mode = S_IFWHT;
930		}
931		return (FTS_W);
932	}
933#endif
934
935	/*
936	 * If doing a logical walk, or application requested FTS_FOLLOW, do
937	 * a stat(2).  If that fails, check for a non-existent symlink.  If
938	 * fail, set the errno from the stat call.
939	 */
940	if (ISSET(FTS_LOGICAL) || follow) {
941		if (stat(p->fts_accpath, sbp)) {
942			saved_errno = errno;
943			if (!lstat(p->fts_accpath, sbp)) {
944				errno = 0;
945				return (FTS_SLNONE);
946			}
947			p->fts_errno = saved_errno;
948			goto err;
949		}
950	} else if (lstat(p->fts_accpath, sbp)) {
951		p->fts_errno = errno;
952err:		memset(sbp, 0, sizeof(*sbp));
953		return (FTS_NS);
954	}
955
956	if (S_ISDIR(sbp->st_mode)) {
957		/*
958		 * Set the device/inode.  Used to find cycles and check for
959		 * crossing mount points.  Also remember the link count, used
960		 * in fts_build to limit the number of stat calls.  It is
961		 * understood that these fields are only referenced if fts_info
962		 * is set to FTS_D.
963		 */
964		dev = p->fts_dev = sbp->st_dev;
965		ino = p->fts_ino = sbp->st_ino;
966		p->fts_nlink = sbp->st_nlink;
967
968		if (ISDOT(p->fts_name))
969			return (FTS_DOT);
970
971		/*
972		 * Cycle detection is done by brute force when the directory
973		 * is first encountered.  If the tree gets deep enough or the
974		 * number of symbolic links to directories is high enough,
975		 * something faster might be worthwhile.
976		 */
977		for (t = p->fts_parent;
978		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
979			if (ino == t->fts_ino && dev == t->fts_dev) {
980				p->fts_cycle = t;
981				return (FTS_DC);
982			}
983		return (FTS_D);
984	}
985	if (S_ISLNK(sbp->st_mode))
986		return (FTS_SL);
987	if (S_ISREG(sbp->st_mode))
988		return (FTS_F);
989	return (FTS_DEFAULT);
990}
991
992static FTSENT *
993fts_sort(FTS *sp, FTSENT *head, size_t nitems)
994{
995	FTSENT **ap, *p;
996
997	_DIAGASSERT(sp != NULL);
998	_DIAGASSERT(head != NULL);
999
1000	/*
1001	 * Construct an array of pointers to the structures and call qsort(3).
1002	 * Reassemble the array in the order returned by qsort.  If unable to
1003	 * sort for memory reasons, return the directory entries in their
1004	 * current order.  Allocate enough space for the current needs plus
1005	 * 40 so don't realloc one entry at a time.
1006	 */
1007	if (nitems > sp->fts_nitems) {
1008		FTSENT **new;
1009
1010		new = realloc(sp->fts_array, sizeof(FTSENT *) * (nitems + 40));
1011		if (new == 0)
1012			return (head);
1013		sp->fts_array = new;
1014		sp->fts_nitems = ftsent_nitems_truncate(nitems + 40);
1015	}
1016	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
1017		*ap++ = p;
1018	qsort((void *)sp->fts_array, nitems, sizeof(FTSENT *),
1019		(int (*)(const void *, const void *))sp->fts_compar);
1020	for (head = *(ap = sp->fts_array); --nitems; ++ap)
1021		ap[0]->fts_link = ap[1];
1022	ap[0]->fts_link = NULL;
1023	return (head);
1024}
1025
1026static FTSENT *
1027fts_alloc(FTS *sp, const char *name, size_t namelen)
1028{
1029	FTSENT *p;
1030#if defined(FTS_ALLOC_ALIGNED)
1031	size_t len;
1032#endif
1033
1034	_DIAGASSERT(sp != NULL);
1035	_DIAGASSERT(name != NULL);
1036
1037#if defined(FTS_ALLOC_ALIGNED)
1038	/*
1039	 * The file name is a variable length array and no stat structure is
1040	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
1041	 * structure, the file name and the stat structure in one chunk, but
1042	 * be careful that the stat structure is reasonably aligned.  Since the
1043	 * fts_name field is declared to be of size 1, the fts_name pointer is
1044	 * namelen + 2 before the first possible address of the stat structure.
1045	 */
1046	len = sizeof(FTSENT) + namelen;
1047	if (!ISSET(FTS_NOSTAT))
1048		len += sizeof(*(p->fts_statp)) + ALIGNBYTES;
1049	if ((p = malloc(len)) == NULL)
1050		return (NULL);
1051
1052	if (!ISSET(FTS_NOSTAT))
1053		p->fts_statp = (__fts_stat_t *)ALIGN(
1054		    (unsigned long)(p->fts_name + namelen + 2));
1055#else
1056	if ((p = malloc(sizeof(FTSENT) + namelen)) == NULL)
1057		return (NULL);
1058
1059	if (!ISSET(FTS_NOSTAT))
1060		if ((p->fts_statp = malloc(sizeof(*(p->fts_statp)))) == NULL) {
1061			free(p);
1062			return (NULL);
1063		}
1064#endif
1065
1066        if (ISSET(FTS_NOSTAT))
1067                p->fts_statp = NULL;
1068
1069	/* Copy the name plus the trailing NULL. */
1070	memmove(p->fts_name, name, namelen + 1);
1071
1072	p->fts_namelen = fts_namelen_truncate(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	return (p);
1080}
1081
1082static void
1083fts_free(FTSENT *p)
1084{
1085#if !defined(FTS_ALLOC_ALIGNED)
1086	if (p->fts_statp)
1087		free(p->fts_statp);
1088#endif
1089	free(p);
1090}
1091
1092static void
1093fts_lfree(FTSENT *head)
1094{
1095	FTSENT *p;
1096
1097	/* XXX: head may be NULL ? */
1098
1099	/* Free a linked list of structures. */
1100	while ((p = head) != NULL) {
1101		head = head->fts_link;
1102		fts_free(p);
1103	}
1104}
1105
1106static size_t
1107fts_pow2(size_t x)
1108{
1109
1110	x--;
1111	x |= x>>1;
1112	x |= x>>2;
1113	x |= x>>4;
1114	x |= x>>8;
1115	x |= x>>16;
1116#if LONG_BIT > 32
1117	x |= x>>32;
1118#endif
1119#if LONG_BIT > 64
1120	x |= x>>64;
1121#endif
1122	x++;
1123	return (x);
1124}
1125
1126/*
1127 * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
1128 * Most systems will allow creation of paths much longer than MAXPATHLEN, even
1129 * though the kernel won't resolve them.  Round up the new size to a power of 2,
1130 * so we don't realloc the path 2 bytes at a time.
1131 */
1132static int
1133fts_palloc(FTS *sp, size_t size)
1134{
1135	char *new;
1136
1137	_DIAGASSERT(sp != NULL);
1138
1139#ifdef __FTS_COMPAT_LENGTH
1140	/* Protect against fts_pathlen overflow. */
1141	if (size > USHRT_MAX + 1) {
1142		errno = ENAMETOOLONG;
1143		return (1);
1144	}
1145#endif
1146	size = fts_pow2(size);
1147	new = realloc(sp->fts_path, size);
1148	if (new == 0)
1149		return (1);
1150	sp->fts_path = new;
1151	sp->fts_pathlen = fts_pathlen_truncate(size);
1152	return (0);
1153}
1154
1155/*
1156 * When the path is realloc'd, have to fix all of the pointers in structures
1157 * already returned.
1158 */
1159static void
1160fts_padjust(FTS *sp, FTSENT *head)
1161{
1162	FTSENT *p;
1163	char *addr;
1164
1165	_DIAGASSERT(sp != NULL);
1166
1167#define	ADJUST(p) do {							\
1168	if ((p)->fts_accpath != (p)->fts_name)				\
1169		(p)->fts_accpath =					\
1170		    addr + ((p)->fts_accpath - (p)->fts_path);		\
1171	(p)->fts_path = addr;						\
1172} while (/*CONSTCOND*/0)
1173
1174	addr = sp->fts_path;
1175
1176	/* Adjust the current set of children. */
1177	for (p = sp->fts_child; p; p = p->fts_link)
1178		ADJUST(p);
1179
1180	/* Adjust the rest of the tree, including the current level. */
1181	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1182		ADJUST(p);
1183		p = p->fts_link ? p->fts_link : p->fts_parent;
1184	}
1185}
1186
1187static size_t
1188fts_maxarglen(char * const *argv)
1189{
1190	size_t len, max;
1191
1192	_DIAGASSERT(argv != NULL);
1193
1194	for (max = 0; *argv; ++argv)
1195		if ((len = strlen(*argv)) > max)
1196			max = len;
1197	return (max + 1);
1198}
1199
1200/*
1201 * Change to dir specified by fd or p->fts_accpath without getting
1202 * tricked by someone changing the world out from underneath us.
1203 * Assumes p->fts_dev and p->fts_ino are filled in.
1204 */
1205static int
1206fts_safe_changedir(const FTS *sp, const FTSENT *p, int fd, const char *path)
1207{
1208	int oldfd = fd, ret = -1;
1209	__fts_stat_t sb;
1210
1211	if (ISSET(FTS_NOCHDIR))
1212		return 0;
1213
1214	if (oldfd < 0 && (fd = open(path, O_RDONLY)) == -1)
1215		return -1;
1216
1217	if (fstat(fd, &sb) == -1)
1218		goto bail;
1219
1220	if (sb.st_ino != p->fts_ino || sb.st_dev != p->fts_dev) {
1221		errno = ENOENT;
1222		goto bail;
1223	}
1224
1225	ret = fchdir(fd);
1226
1227bail:
1228	if (oldfd < 0) {
1229		int save_errno = errno;
1230		(void)close(fd);
1231		errno = save_errno;
1232	}
1233	return ret;
1234}
1235