fts.c revision 1.42
1/*	$NetBSD: fts.c,v 1.42 2011/10/16 05:05:38 mrg 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 1.42 2011/10/16 05:05:38 mrg 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#define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
83
84#define	CLR(opt)	(sp->fts_options &= ~(opt))
85#define	ISSET(opt)	(sp->fts_options & (opt))
86#define	SET(opt)	(sp->fts_options |= (opt))
87
88#define	CHDIR(sp, path)	(!ISSET(FTS_NOCHDIR) && chdir(path))
89#define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
90
91/* fts_build flags */
92#define	BCHILD		1		/* fts_children */
93#define	BNAMES		2		/* fts_children, names only */
94#define	BREAD		3		/* fts_read */
95
96#ifndef DTF_HIDEW
97#undef FTS_WHITEOUT
98#endif
99
100FTS *
101fts_open(char * const *argv, int options,
102    int (*compar)(const FTSENT **, const FTSENT **))
103{
104	FTS *sp;
105	FTSENT *p, *root;
106	size_t nitems;
107	FTSENT *parent, *tmp = NULL;	/* pacify gcc */
108	size_t len;
109
110	_DIAGASSERT(argv != NULL);
111
112	/* Options check. */
113	if (options & ~FTS_OPTIONMASK) {
114		errno = EINVAL;
115		return (NULL);
116	}
117
118	/* Allocate/initialize the stream */
119	if ((sp = malloc((unsigned int)sizeof(FTS))) == NULL)
120		return (NULL);
121	memset(sp, 0, sizeof(FTS));
122	sp->fts_compar = compar;
123	sp->fts_options = options;
124
125	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
126	if (ISSET(FTS_LOGICAL))
127		SET(FTS_NOCHDIR);
128
129	/*
130	 * Start out with 1K of path space, and enough, in any case,
131	 * to hold the user's paths.
132	 */
133	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
134		goto mem1;
135
136	/* Allocate/initialize root's parent. */
137	if ((parent = fts_alloc(sp, "", 0)) == NULL)
138		goto mem2;
139	parent->fts_level = FTS_ROOTPARENTLEVEL;
140
141	/* Allocate/initialize root(s). */
142	for (root = NULL, nitems = 0; *argv; ++argv, ++nitems) {
143		/* Don't allow zero-length paths. */
144		if ((len = strlen(*argv)) == 0) {
145			errno = ENOENT;
146			goto mem3;
147		}
148
149		if ((p = fts_alloc(sp, *argv, len)) == NULL)
150			goto mem3;
151		p->fts_level = FTS_ROOTLEVEL;
152		p->fts_parent = parent;
153		p->fts_accpath = p->fts_name;
154		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW));
155
156		/* Command-line "." and ".." are real directories. */
157		if (p->fts_info == FTS_DOT)
158			p->fts_info = FTS_D;
159
160		/*
161		 * If comparison routine supplied, traverse in sorted
162		 * order; otherwise traverse in the order specified.
163		 */
164		if (compar) {
165			p->fts_link = root;
166			root = p;
167		} else {
168			p->fts_link = NULL;
169			if (root == NULL)
170				tmp = root = p;
171			else {
172				tmp->fts_link = p;
173				tmp = p;
174			}
175		}
176	}
177	if (compar && nitems > 1)
178		root = fts_sort(sp, root, nitems);
179
180	/*
181	 * Allocate a dummy pointer and make fts_read think that we've just
182	 * finished the node before the root(s); set p->fts_info to FTS_INIT
183	 * so that everything about the "current" node is ignored.
184	 */
185	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
186		goto mem3;
187	sp->fts_cur->fts_link = root;
188	sp->fts_cur->fts_info = FTS_INIT;
189
190	/*
191	 * If using chdir(2), grab a file descriptor pointing to dot to insure
192	 * that we can get back here; this could be avoided for some paths,
193	 * but almost certainly not worth the effort.  Slashes, symbolic links,
194	 * and ".." are all fairly nasty problems.  Note, if we can't get the
195	 * descriptor we run anyway, just more slowly.
196	 */
197#ifndef O_CLOEXEC
198#define O_CLOEXEC 0
199#endif
200	if (!ISSET(FTS_NOCHDIR)) {
201		if ((sp->fts_rfd = open(".", O_RDONLY | O_CLOEXEC, 0)) == -1)
202			SET(FTS_NOCHDIR);
203	}
204
205	if (nitems == 0)
206		fts_free(parent);
207
208	return (sp);
209
210mem3:	fts_lfree(root);
211	fts_free(parent);
212mem2:	free(sp->fts_path);
213mem1:	free(sp);
214	return (NULL);
215}
216
217static void
218fts_load(FTS *sp, FTSENT *p)
219{
220	size_t len;
221	char *cp;
222
223	_DIAGASSERT(sp != NULL);
224	_DIAGASSERT(p != NULL);
225
226	/*
227	 * Load the stream structure for the next traversal.  Since we don't
228	 * actually enter the directory until after the preorder visit, set
229	 * the fts_accpath field specially so the chdir gets done to the right
230	 * place and the user can access the first node.  From fts_open it's
231	 * known that the path will fit.
232	 */
233	len = p->fts_pathlen = p->fts_namelen;
234	memmove(sp->fts_path, p->fts_name, len + 1);
235	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
236		len = strlen(++cp);
237		memmove(p->fts_name, cp, len + 1);
238		p->fts_namelen = len;
239	}
240	p->fts_accpath = p->fts_path = sp->fts_path;
241	sp->fts_dev = p->fts_dev;
242}
243
244int
245fts_close(FTS *sp)
246{
247	FTSENT *freep, *p;
248	int saved_errno = 0;
249
250	_DIAGASSERT(sp != NULL);
251
252	/*
253	 * This still works if we haven't read anything -- the dummy structure
254	 * points to the root list, so we step through to the end of the root
255	 * list which has a valid parent pointer.
256	 */
257	if (sp->fts_cur) {
258		if (sp->fts_cur->fts_flags & FTS_SYMFOLLOW)
259			(void)close(sp->fts_cur->fts_symfd);
260		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
261			freep = p;
262			p = p->fts_link ? p->fts_link : p->fts_parent;
263			fts_free(freep);
264		}
265		fts_free(p);
266	}
267
268	/* Free up child linked list, sort array, path buffer. */
269	if (sp->fts_child)
270		fts_lfree(sp->fts_child);
271	if (sp->fts_array)
272		free(sp->fts_array);
273	free(sp->fts_path);
274
275	/* Return to original directory, save errno if necessary. */
276	if (!ISSET(FTS_NOCHDIR)) {
277		if (fchdir(sp->fts_rfd) == -1)
278			saved_errno = errno;
279		(void)close(sp->fts_rfd);
280	}
281
282	/* Free up the stream pointer. */
283	free(sp);
284	if (saved_errno) {
285		errno = saved_errno;
286		return -1;
287	}
288
289	return 0;
290}
291
292#if !defined(__FTS_COMPAT_TAILINGSLASH)
293
294/*
295 * Special case of "/" at the end of the path so that slashes aren't
296 * appended which would cause paths to be written as "....//foo".
297 */
298#define	NAPPEND(p)							\
299	(p->fts_path[p->fts_pathlen - 1] == '/'				\
300	    ? p->fts_pathlen - 1 : p->fts_pathlen)
301
302#else /* !defined(__FTS_COMPAT_TAILINGSLASH) */
303
304/*
305 * compatibility with the old behaviour.
306 *
307 * Special case a root of "/" so that slashes aren't appended which would
308 * cause paths to be written as "//foo".
309 */
310
311#define	NAPPEND(p)							\
312	(p->fts_level == FTS_ROOTLEVEL && p->fts_pathlen == 1 &&	\
313	    p->fts_path[0] == '/' ? 0 : p->fts_pathlen)
314
315#endif /* !defined(__FTS_COMPAT_TAILINGSLASH) */
316
317FTSENT *
318fts_read(FTS *sp)
319{
320	FTSENT *p, *tmp;
321	int instr;
322	char *t;
323	int saved_errno;
324
325	_DIAGASSERT(sp != NULL);
326
327	/* If finished or unrecoverable error, return NULL. */
328	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
329		return (NULL);
330
331	/* Set current node pointer. */
332	p = sp->fts_cur;
333
334	/* Save and zero out user instructions. */
335	instr = p->fts_instr;
336	p->fts_instr = FTS_NOINSTR;
337
338	/* Any type of file may be re-visited; re-stat and re-turn. */
339	if (instr == FTS_AGAIN) {
340		p->fts_info = fts_stat(sp, p, 0);
341		return (p);
342	}
343
344	/*
345	 * Following a symlink -- SLNONE test allows application to see
346	 * SLNONE and recover.  If indirecting through a symlink, have
347	 * keep a pointer to current location.  If unable to get that
348	 * pointer, follow fails.
349	 */
350	if (instr == FTS_FOLLOW &&
351	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
352		p->fts_info = fts_stat(sp, p, 1);
353		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
354			if ((p->fts_symfd = open(".", O_RDONLY | O_CLOEXEC, 0))
355			    == -1) {
356				p->fts_errno = errno;
357				p->fts_info = FTS_ERR;
358			} else
359				p->fts_flags |= FTS_SYMFOLLOW;
360		}
361		return (p);
362	}
363
364	/* Directory in pre-order. */
365	if (p->fts_info == FTS_D) {
366		/* If skipped or crossed mount point, do post-order visit. */
367		if (instr == FTS_SKIP ||
368		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
369			if (p->fts_flags & FTS_SYMFOLLOW)
370				(void)close(p->fts_symfd);
371			if (sp->fts_child) {
372				fts_lfree(sp->fts_child);
373				sp->fts_child = NULL;
374			}
375			p->fts_info = FTS_DP;
376			return (p);
377		}
378
379		/* Rebuild if only read the names and now traversing. */
380		if (sp->fts_child && ISSET(FTS_NAMEONLY)) {
381			CLR(FTS_NAMEONLY);
382			fts_lfree(sp->fts_child);
383			sp->fts_child = NULL;
384		}
385
386		/*
387		 * Cd to the subdirectory.
388		 *
389		 * If have already read and now fail to chdir, whack the list
390		 * to make the names come out right, and set the parent errno
391		 * so the application will eventually get an error condition.
392		 * Set the FTS_DONTCHDIR flag so that when we logically change
393		 * directories back to the parent we don't do a chdir.
394		 *
395		 * If haven't read do so.  If the read fails, fts_build sets
396		 * FTS_STOP or the fts_info field of the node.
397		 */
398		if (sp->fts_child) {
399			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
400				p->fts_errno = errno;
401				p->fts_flags |= FTS_DONTCHDIR;
402				for (p = sp->fts_child; p; p = p->fts_link)
403					p->fts_accpath =
404					    p->fts_parent->fts_accpath;
405			}
406		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
407			if (ISSET(FTS_STOP))
408				return (NULL);
409			return (p);
410		}
411		p = sp->fts_child;
412		sp->fts_child = NULL;
413		goto name;
414	}
415
416	/* Move to the next node on this level. */
417next:	tmp = p;
418	if ((p = p->fts_link) != NULL) {
419		fts_free(tmp);
420
421		/*
422		 * If reached the top, return to the original directory, and
423		 * load the paths for the next root.
424		 */
425		if (p->fts_level == FTS_ROOTLEVEL) {
426			if (FCHDIR(sp, sp->fts_rfd)) {
427				SET(FTS_STOP);
428				return (NULL);
429			}
430			fts_load(sp, p);
431			return (sp->fts_cur = p);
432		}
433
434		/*
435		 * User may have called fts_set on the node.  If skipped,
436		 * ignore.  If followed, get a file descriptor so we can
437		 * get back if necessary.
438		 */
439		if (p->fts_instr == FTS_SKIP)
440			goto next;
441		if (p->fts_instr == FTS_FOLLOW) {
442			p->fts_info = fts_stat(sp, p, 1);
443			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
444				if ((p->fts_symfd =
445				    open(".", O_RDONLY | O_CLOEXEC, 0)) == -1) {
446					p->fts_errno = errno;
447					p->fts_info = FTS_ERR;
448				} else
449					p->fts_flags |= FTS_SYMFOLLOW;
450			}
451			p->fts_instr = FTS_NOINSTR;
452		}
453
454name:		t = sp->fts_path + NAPPEND(p->fts_parent);
455		*t++ = '/';
456		memmove(t, p->fts_name, (size_t)(p->fts_namelen + 1));
457		return (sp->fts_cur = p);
458	}
459
460	/* Move up to the parent node. */
461	p = tmp->fts_parent;
462	fts_free(tmp);
463
464	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
465		/*
466		 * Done; free everything up and set errno to 0 so the user
467		 * can distinguish between error and EOF.
468		 */
469		fts_free(p);
470		errno = 0;
471		return (sp->fts_cur = NULL);
472	}
473
474	/* Nul terminate the pathname. */
475	sp->fts_path[p->fts_pathlen] = '\0';
476
477	/*
478	 * Return to the parent directory.  If at a root node or came through
479	 * a symlink, go back through the file descriptor.  Otherwise, cd up
480	 * one directory.
481	 */
482	if (p->fts_level == FTS_ROOTLEVEL) {
483		if (FCHDIR(sp, sp->fts_rfd)) {
484			SET(FTS_STOP);
485			return (NULL);
486		}
487	} else if (p->fts_flags & FTS_SYMFOLLOW) {
488		if (FCHDIR(sp, p->fts_symfd)) {
489			saved_errno = errno;
490			(void)close(p->fts_symfd);
491			errno = saved_errno;
492			SET(FTS_STOP);
493			return (NULL);
494		}
495		(void)close(p->fts_symfd);
496	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
497	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
498		SET(FTS_STOP);
499		return (NULL);
500	}
501	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
502	return (sp->fts_cur = p);
503}
504
505/*
506 * Fts_set takes the stream as an argument although it's not used in this
507 * implementation; it would be necessary if anyone wanted to add global
508 * semantics to fts using fts_set.  An error return is allowed for similar
509 * reasons.
510 */
511/* ARGSUSED */
512int
513fts_set(FTS *sp, FTSENT *p, int instr)
514{
515
516	_DIAGASSERT(sp != NULL);
517	_DIAGASSERT(p != NULL);
518
519	if (instr && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
520	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
521		errno = EINVAL;
522		return (1);
523	}
524	p->fts_instr = instr;
525	return (0);
526}
527
528FTSENT *
529fts_children(FTS *sp, int instr)
530{
531	FTSENT *p;
532	int fd;
533
534	_DIAGASSERT(sp != NULL);
535
536	if (instr && instr != FTS_NAMEONLY) {
537		errno = EINVAL;
538		return (NULL);
539	}
540
541	/* Set current node pointer. */
542	p = sp->fts_cur;
543
544	/*
545	 * Errno set to 0 so user can distinguish empty directory from
546	 * an error.
547	 */
548	errno = 0;
549
550	/* Fatal errors stop here. */
551	if (ISSET(FTS_STOP))
552		return (NULL);
553
554	/* Return logical hierarchy of user's arguments. */
555	if (p->fts_info == FTS_INIT)
556		return (p->fts_link);
557
558	/*
559	 * If not a directory being visited in pre-order, stop here.  Could
560	 * allow FTS_DNR, assuming the user has fixed the problem, but the
561	 * same effect is available with FTS_AGAIN.
562	 */
563	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
564		return (NULL);
565
566	/* Free up any previous child list. */
567	if (sp->fts_child)
568		fts_lfree(sp->fts_child);
569
570	if (instr == FTS_NAMEONLY) {
571		SET(FTS_NAMEONLY);
572		instr = BNAMES;
573	} else
574		instr = BCHILD;
575
576	/*
577	 * If using chdir on a relative path and called BEFORE fts_read does
578	 * its chdir to the root of a traversal, we can lose -- we need to
579	 * chdir into the subdirectory, and we don't know where the current
580	 * directory is, so we can't get back so that the upcoming chdir by
581	 * fts_read will work.
582	 */
583	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
584	    ISSET(FTS_NOCHDIR))
585		return (sp->fts_child = fts_build(sp, instr));
586
587	if ((fd = open(".", O_RDONLY, 0)) == -1)
588		return (sp->fts_child = NULL);
589	sp->fts_child = fts_build(sp, instr);
590	if (fchdir(fd)) {
591		(void)close(fd);
592		return (NULL);
593	}
594	(void)close(fd);
595	return (sp->fts_child);
596}
597
598/*
599 * This is the tricky part -- do not casually change *anything* in here.  The
600 * idea is to build the linked list of entries that are used by fts_children
601 * and fts_read.  There are lots of special cases.
602 *
603 * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
604 * set and it's a physical walk (so that symbolic links can't be directories),
605 * we can do things quickly.  First, if it's a 4.4BSD file system, the type
606 * of the file is in the directory entry.  Otherwise, we assume that the number
607 * of subdirectories in a node is equal to the number of links to the parent.
608 * The former skips all stat calls.  The latter skips stat calls in any leaf
609 * directories and for any files after the subdirectories in the directory have
610 * been found, cutting the stat calls by about 2/3.
611 */
612static FTSENT *
613fts_build(FTS *sp, int type)
614{
615	struct dirent *dp;
616	FTSENT *p, *head;
617	size_t nitems;
618	FTSENT *cur, *tail;
619	DIR *dirp;
620	void *oldaddr;
621	size_t dnamlen;
622	int cderrno, descend, level, nlinks, saved_errno, nostat, doadjust;
623	size_t len, maxlen;
624#ifdef FTS_WHITEOUT
625	int oflag;
626#endif
627	char *cp = NULL;	/* pacify gcc */
628
629	_DIAGASSERT(sp != NULL);
630
631	/* Set current node pointer. */
632	cur = sp->fts_cur;
633
634	/*
635	 * Open the directory for reading.  If this fails, we're done.
636	 * If being called from fts_read, set the fts_info field.
637	 */
638#ifdef FTS_WHITEOUT
639	if (ISSET(FTS_WHITEOUT))
640		oflag = DTF_NODUP|DTF_REWIND;
641	else
642		oflag = DTF_HIDEW|DTF_NODUP|DTF_REWIND;
643#else
644#define	__opendir2(path, flag) opendir(path)
645#endif
646	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
647		if (type == BREAD) {
648			cur->fts_info = FTS_DNR;
649			cur->fts_errno = errno;
650		}
651		return (NULL);
652	}
653
654	/*
655	 * Nlinks is the number of possible entries of type directory in the
656	 * directory if we're cheating on stat calls, 0 if we're not doing
657	 * any stat calls at all, -1 if we're doing stats on everything.
658	 */
659	if (type == BNAMES) {
660		nlinks = 0;
661		nostat = 1;
662	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
663		nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
664		nostat = 1;
665	} else {
666		nlinks = -1;
667		nostat = 0;
668	}
669
670#ifdef notdef
671	(void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
672	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
673	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
674#endif
675	/*
676	 * If we're going to need to stat anything or we want to descend
677	 * and stay in the directory, chdir.  If this fails we keep going,
678	 * but set a flag so we don't chdir after the post-order visit.
679	 * We won't be able to stat anything, but we can still return the
680	 * names themselves.  Note, that since fts_read won't be able to
681	 * chdir into the directory, it will have to return different path
682	 * names than before, i.e. "a/b" instead of "b".  Since the node
683	 * has already been visited in pre-order, have to wait until the
684	 * post-order visit to return the error.  There is a special case
685	 * here, if there was nothing to stat then it's not an error to
686	 * not be able to stat.  This is all fairly nasty.  If a program
687	 * needed sorted entries or stat information, they had better be
688	 * checking FTS_NS on the returned nodes.
689	 */
690	cderrno = 0;
691	if (nlinks || type == BREAD) {
692		if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) {
693			if (nlinks && type == BREAD)
694				cur->fts_errno = errno;
695			cur->fts_flags |= FTS_DONTCHDIR;
696			descend = 0;
697			cderrno = errno;
698		} else
699			descend = 1;
700	} else
701		descend = 0;
702
703	/*
704	 * Figure out the max file name length that can be stored in the
705	 * current path -- the inner loop allocates more path as necessary.
706	 * We really wouldn't have to do the maxlen calculations here, we
707	 * could do them in fts_read before returning the path, but it's a
708	 * lot easier here since the length is part of the dirent structure.
709	 *
710	 * If not changing directories set a pointer so that can just append
711	 * each new name into the path.
712	 */
713	len = NAPPEND(cur);
714	if (ISSET(FTS_NOCHDIR)) {
715		cp = sp->fts_path + len;
716		*cp++ = '/';
717	}
718	len++;
719	maxlen = sp->fts_pathlen - len;
720
721#if defined(__FTS_COMPAT_LEVEL)
722	if (cur->fts_level == SHRT_MAX) {
723		(void)closedir(dirp);
724		cur->fts_info = FTS_ERR;
725		SET(FTS_STOP);
726		errno = ENAMETOOLONG;
727		return (NULL);
728	}
729#endif
730
731	level = cur->fts_level + 1;
732
733	/* Read the directory, attaching each entry to the `link' pointer. */
734	doadjust = 0;
735	for (head = tail = NULL, nitems = 0; (dp = readdir(dirp)) != NULL;) {
736
737		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
738			continue;
739
740#if defined(HAVE_STRUCT_DIRENT_D_NAMLEN)
741		dnamlen = dp->d_namlen;
742#else
743		dnamlen = strlen(dp->d_name);
744#endif
745		if ((p = fts_alloc(sp, dp->d_name, dnamlen)) == NULL)
746			goto mem1;
747		if (dnamlen >= maxlen) {	/* include space for NUL */
748			oldaddr = sp->fts_path;
749			if (fts_palloc(sp, dnamlen + len + 1)) {
750				/*
751				 * No more memory for path or structures.  Save
752				 * errno, free up the current structure and the
753				 * structures already allocated.
754				 */
755mem1:				saved_errno = errno;
756				if (p)
757					fts_free(p);
758				fts_lfree(head);
759				(void)closedir(dirp);
760				errno = saved_errno;
761				cur->fts_info = FTS_ERR;
762				SET(FTS_STOP);
763				return (NULL);
764			}
765			/* Did realloc() change the pointer? */
766			if (oldaddr != sp->fts_path) {
767				doadjust = 1;
768				if (ISSET(FTS_NOCHDIR))
769					cp = sp->fts_path + len;
770			}
771			maxlen = sp->fts_pathlen - len;
772		}
773
774#if defined(__FTS_COMPAT_LENGTH)
775		if (len + dnamlen >= USHRT_MAX) {
776			/*
777			 * In an FTSENT, fts_pathlen is an unsigned short
778			 * so it is possible to wraparound here.
779			 * If we do, free up the current structure and the
780			 * structures already allocated, then error out
781			 * with ENAMETOOLONG.
782			 */
783			fts_free(p);
784			fts_lfree(head);
785			(void)closedir(dirp);
786			cur->fts_info = FTS_ERR;
787			SET(FTS_STOP);
788			errno = ENAMETOOLONG;
789			return (NULL);
790		}
791#endif
792		p->fts_level = level;
793		p->fts_pathlen = len + dnamlen;
794		p->fts_parent = sp->fts_cur;
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,
822				        (size_t)(p->fts_namelen + 1));
823			} else
824				p->fts_accpath = p->fts_name;
825			/* Stat it. */
826			p->fts_info = fts_stat(sp, p, 0);
827
828			/* Decrement link count if applicable. */
829			if (nlinks > 0 && (p->fts_info == FTS_D ||
830			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
831				--nlinks;
832		}
833
834		/* We walk in directory order so "ls -f" doesn't get upset. */
835		p->fts_link = NULL;
836		if (head == NULL)
837			head = tail = p;
838		else {
839			tail->fts_link = p;
840			tail = p;
841		}
842		++nitems;
843	}
844	(void)closedir(dirp);
845
846	/*
847	 * If had to realloc the path, adjust the addresses for the rest
848	 * of the tree.
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 unsigned short
893fts_stat(FTS *sp, FTSENT *p, int follow)
894{
895	FTSENT *t;
896	dev_t dev;
897	__fts_ino_t ino;
898	__fts_stat_t *sbp, sb;
899	int saved_errno;
900
901	_DIAGASSERT(sp != NULL);
902	_DIAGASSERT(p != NULL);
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(*sbp));
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
975static FTSENT *
976fts_sort(FTS *sp, FTSENT *head, size_t nitems)
977{
978	FTSENT **ap, *p;
979
980	_DIAGASSERT(sp != NULL);
981	_DIAGASSERT(head != NULL);
982
983	/*
984	 * Construct an array of pointers to the structures and call qsort(3).
985	 * Reassemble the array in the order returned by qsort.  If unable to
986	 * sort for memory reasons, return the directory entries in their
987	 * current order.  Allocate enough space for the current needs plus
988	 * 40 so don't realloc one entry at a time.
989	 */
990	if (nitems > sp->fts_nitems) {
991		FTSENT **new;
992
993		new = realloc(sp->fts_array, sizeof(FTSENT *) * (nitems + 40));
994		if (new == 0)
995			return (head);
996		sp->fts_array = new;
997		sp->fts_nitems = nitems + 40;
998	}
999	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
1000		*ap++ = p;
1001	qsort((void *)sp->fts_array, nitems, sizeof(FTSENT *),
1002		(int (*)(const void *, const void *))sp->fts_compar);
1003	for (head = *(ap = sp->fts_array); --nitems; ++ap)
1004		ap[0]->fts_link = ap[1];
1005	ap[0]->fts_link = NULL;
1006	return (head);
1007}
1008
1009static FTSENT *
1010fts_alloc(FTS *sp, const char *name, size_t namelen)
1011{
1012	FTSENT *p;
1013#if defined(FTS_ALLOC_ALIGNED)
1014	size_t len;
1015#endif
1016
1017	_DIAGASSERT(sp != NULL);
1018	_DIAGASSERT(name != NULL);
1019
1020#if defined(FTS_ALLOC_ALIGNED)
1021	/*
1022	 * The file name is a variable length array and no stat structure is
1023	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
1024	 * structure, the file name and the stat structure in one chunk, but
1025	 * be careful that the stat structure is reasonably aligned.  Since the
1026	 * fts_name field is declared to be of size 1, the fts_name pointer is
1027	 * namelen + 2 before the first possible address of the stat structure.
1028	 */
1029	len = sizeof(FTSENT) + namelen;
1030	if (!ISSET(FTS_NOSTAT))
1031		len += sizeof(*(p->fts_statp)) + ALIGNBYTES;
1032	if ((p = malloc(len)) == NULL)
1033		return (NULL);
1034
1035	if (!ISSET(FTS_NOSTAT))
1036		p->fts_statp = (__fts_stat_t *)ALIGN(
1037		    (unsigned long)(p->fts_name + namelen + 2));
1038#else
1039	if ((p = malloc(sizeof(FTSENT) + namelen)) == NULL)
1040		return (NULL);
1041
1042	if (!ISSET(FTS_NOSTAT))
1043		if ((p->fts_statp = malloc(sizeof(*(p->fts_statp)))) == NULL) {
1044			free(p);
1045			return (NULL);
1046		}
1047#endif
1048
1049        if (ISSET(FTS_NOSTAT))
1050                p->fts_statp = NULL;
1051
1052	/* Copy the name plus the trailing NULL. */
1053	memmove(p->fts_name, name, namelen + 1);
1054
1055	p->fts_namelen = namelen;
1056	p->fts_path = sp->fts_path;
1057	p->fts_errno = 0;
1058	p->fts_flags = 0;
1059	p->fts_instr = FTS_NOINSTR;
1060	p->fts_number = 0;
1061	p->fts_pointer = NULL;
1062	return (p);
1063}
1064
1065static void
1066fts_free(FTSENT *p)
1067{
1068#if !defined(FTS_ALLOC_ALIGNED)
1069	if (p->fts_statp)
1070		free(p->fts_statp);
1071#endif
1072	free(p);
1073}
1074
1075static void
1076fts_lfree(FTSENT *head)
1077{
1078	FTSENT *p;
1079
1080	/* XXX: head may be NULL ? */
1081
1082	/* Free a linked list of structures. */
1083	while ((p = head) != NULL) {
1084		head = head->fts_link;
1085		fts_free(p);
1086	}
1087}
1088
1089static size_t
1090fts_pow2(size_t x)
1091{
1092
1093	x--;
1094	x |= x>>1;
1095	x |= x>>2;
1096	x |= x>>4;
1097	x |= x>>8;
1098	x |= x>>16;
1099#if LONG_BIT > 32
1100	x |= x>>32;
1101#endif
1102#if LONG_BIT > 64
1103	x |= x>>64;
1104#endif
1105	x++;
1106	return (x);
1107}
1108
1109/*
1110 * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
1111 * Most systems will allow creation of paths much longer than MAXPATHLEN, even
1112 * though the kernel won't resolve them.  Round up the new size to a power of 2,
1113 * so we don't realloc the path 2 bytes at a time.
1114 */
1115static int
1116fts_palloc(FTS *sp, size_t size)
1117{
1118	char *new;
1119
1120	_DIAGASSERT(sp != NULL);
1121
1122#ifdef __FTS_COMPAT_LENGTH
1123	/* Protect against fts_pathlen overflow. */
1124	if (size > USHRT_MAX + 1) {
1125		errno = ENAMETOOLONG;
1126		return (1);
1127	}
1128#endif
1129	size = fts_pow2(size);
1130	new = realloc(sp->fts_path, size);
1131	if (new == 0)
1132		return (1);
1133	sp->fts_path = new;
1134	sp->fts_pathlen = size;
1135	return (0);
1136}
1137
1138/*
1139 * When the path is realloc'd, have to fix all of the pointers in structures
1140 * already returned.
1141 */
1142static void
1143fts_padjust(FTS *sp, FTSENT *head)
1144{
1145	FTSENT *p;
1146	char *addr;
1147
1148	_DIAGASSERT(sp != NULL);
1149
1150#define	ADJUST(p) do {							\
1151	if ((p)->fts_accpath != (p)->fts_name)				\
1152		(p)->fts_accpath =					\
1153		    addr + ((p)->fts_accpath - (p)->fts_path);		\
1154	(p)->fts_path = addr;						\
1155} while (/*CONSTCOND*/0)
1156
1157	addr = sp->fts_path;
1158
1159	/* Adjust the current set of children. */
1160	for (p = sp->fts_child; p; p = p->fts_link)
1161		ADJUST(p);
1162
1163	/* Adjust the rest of the tree, including the current level. */
1164	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1165		ADJUST(p);
1166		p = p->fts_link ? p->fts_link : p->fts_parent;
1167	}
1168}
1169
1170static size_t
1171fts_maxarglen(char * const *argv)
1172{
1173	size_t len, max;
1174
1175	_DIAGASSERT(argv != NULL);
1176
1177	for (max = 0; *argv; ++argv)
1178		if ((len = strlen(*argv)) > max)
1179			max = len;
1180	return (max + 1);
1181}
1182
1183/*
1184 * Change to dir specified by fd or p->fts_accpath without getting
1185 * tricked by someone changing the world out from underneath us.
1186 * Assumes p->fts_dev and p->fts_ino are filled in.
1187 */
1188static int
1189fts_safe_changedir(const FTS *sp, const FTSENT *p, int fd, const char *path)
1190{
1191	int oldfd = fd, ret = -1;
1192	__fts_stat_t sb;
1193
1194	if (ISSET(FTS_NOCHDIR))
1195		return 0;
1196
1197	if (oldfd < 0 && (fd = open(path, O_RDONLY)) == -1)
1198		return -1;
1199
1200	if (fstat(fd, &sb) == -1)
1201		goto bail;
1202
1203	if (sb.st_ino != p->fts_ino || sb.st_dev != p->fts_dev) {
1204		errno = ENOENT;
1205		goto bail;
1206	}
1207
1208	ret = fchdir(fd);
1209
1210bail:
1211	if (oldfd < 0) {
1212		int save_errno = errno;
1213		(void)close(fd);
1214		errno = save_errno;
1215	}
1216	return ret;
1217}
1218