fts.c revision 1.29
1/*	$NetBSD: fts.c,v 1.29 2005/12/11 04:12:58 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 1.29 2005/12/11 04:12:58 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 1
60#endif
61
62static FTSENT	*fts_alloc(FTS *, const char *, size_t);
63static FTSENT	*fts_build(FTS *, int);
64static void	 fts_lfree(FTSENT *);
65static void	 fts_load(FTS *, FTSENT *);
66static size_t	 fts_maxarglen(char * const *);
67static size_t	 fts_pow2(size_t);
68static int	 fts_palloc(FTS *, size_t);
69static void	 fts_padjust(FTS *, FTSENT *);
70static FTSENT	*fts_sort(FTS *, FTSENT *, size_t);
71static u_short	 fts_stat(FTS *, FTSENT *, int);
72static int	 fts_safe_changedir(const FTS *, const FTSENT *, int,
73    const char *);
74
75#define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
76
77#define	CLR(opt)	(sp->fts_options &= ~(opt))
78#define	ISSET(opt)	(sp->fts_options & (opt))
79#define	SET(opt)	(sp->fts_options |= (opt))
80
81#define	CHDIR(sp, path)	(!ISSET(FTS_NOCHDIR) && chdir(path))
82#define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
83
84/* fts_build flags */
85#define	BCHILD		1		/* fts_children */
86#define	BNAMES		2		/* fts_children, names only */
87#define	BREAD		3		/* fts_read */
88
89#ifndef DTF_HIDEW
90#undef FTS_WHITEOUT
91#endif
92
93FTS *
94fts_open(char * const *argv, int options,
95    int (*compar)(const FTSENT **, const FTSENT **))
96{
97	FTS *sp;
98	FTSENT *p, *root;
99	size_t nitems;
100	FTSENT *parent, *tmp = NULL;	/* pacify gcc */
101	size_t len;
102
103	_DIAGASSERT(argv != NULL);
104
105	/* Options check. */
106	if (options & ~FTS_OPTIONMASK) {
107		errno = EINVAL;
108		return (NULL);
109	}
110
111	/* Allocate/initialize the stream */
112	if ((sp = malloc((u_int)sizeof(FTS))) == NULL)
113		return (NULL);
114	memset(sp, 0, sizeof(FTS));
115	sp->fts_compar = compar;
116	sp->fts_options = options;
117
118	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
119	if (ISSET(FTS_LOGICAL))
120		SET(FTS_NOCHDIR);
121
122	/*
123	 * Start out with 1K of path space, and enough, in any case,
124	 * to hold the user's paths.
125	 */
126	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
127		goto mem1;
128
129	/* Allocate/initialize root's parent. */
130	if ((parent = fts_alloc(sp, "", 0)) == NULL)
131		goto mem2;
132	parent->fts_level = FTS_ROOTPARENTLEVEL;
133
134	/* Allocate/initialize root(s). */
135	for (root = NULL, nitems = 0; *argv; ++argv, ++nitems) {
136		/* Don't allow zero-length paths. */
137		if ((len = strlen(*argv)) == 0) {
138			errno = ENOENT;
139			goto mem3;
140		}
141
142		if ((p = fts_alloc(sp, *argv, len)) == NULL)
143			goto mem3;
144		p->fts_level = FTS_ROOTLEVEL;
145		p->fts_parent = parent;
146		p->fts_accpath = p->fts_name;
147		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW));
148
149		/* Command-line "." and ".." are real directories. */
150		if (p->fts_info == FTS_DOT)
151			p->fts_info = FTS_D;
152
153		/*
154		 * If comparison routine supplied, traverse in sorted
155		 * order; otherwise traverse in the order specified.
156		 */
157		if (compar) {
158			p->fts_link = root;
159			root = p;
160		} else {
161			p->fts_link = NULL;
162			if (root == NULL)
163				tmp = root = p;
164			else {
165				tmp->fts_link = p;
166				tmp = p;
167			}
168		}
169	}
170	if (compar && nitems > 1)
171		root = fts_sort(sp, root, nitems);
172
173	/*
174	 * Allocate a dummy pointer and make fts_read think that we've just
175	 * finished the node before the root(s); set p->fts_info to FTS_INIT
176	 * so that everything about the "current" node is ignored.
177	 */
178	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
179		goto mem3;
180	sp->fts_cur->fts_link = root;
181	sp->fts_cur->fts_info = FTS_INIT;
182
183	/*
184	 * If using chdir(2), grab a file descriptor pointing to dot to insure
185	 * that we can get back here; this could be avoided for some paths,
186	 * but almost certainly not worth the effort.  Slashes, symbolic links,
187	 * and ".." are all fairly nasty problems.  Note, if we can't get the
188	 * descriptor we run anyway, just more slowly.
189	 */
190	if (!ISSET(FTS_NOCHDIR)) {
191		if ((sp->fts_rfd = open(".", O_RDONLY, 0)) == -1)
192			SET(FTS_NOCHDIR);
193		else if (fcntl(sp->fts_rfd, F_SETFD, FD_CLOEXEC) == -1) {
194			close(sp->fts_rfd);
195			SET(FTS_NOCHDIR);
196		}
197	}
198
199	return (sp);
200
201mem3:	fts_lfree(root);
202	free(parent);
203mem2:	free(sp->fts_path);
204mem1:	free(sp);
205	return (NULL);
206}
207
208static void
209fts_load(FTS *sp, FTSENT *p)
210{
211	size_t len;
212	char *cp;
213
214	_DIAGASSERT(sp != NULL);
215	_DIAGASSERT(p != NULL);
216
217	/*
218	 * Load the stream structure for the next traversal.  Since we don't
219	 * actually enter the directory until after the preorder visit, set
220	 * the fts_accpath field specially so the chdir gets done to the right
221	 * place and the user can access the first node.  From fts_open it's
222	 * known that the path will fit.
223	 */
224	len = p->fts_pathlen = p->fts_namelen;
225	memmove(sp->fts_path, p->fts_name, len + 1);
226	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
227		len = strlen(++cp);
228		memmove(p->fts_name, cp, len + 1);
229		p->fts_namelen = len;
230	}
231	p->fts_accpath = p->fts_path = sp->fts_path;
232	sp->fts_dev = p->fts_dev;
233}
234
235int
236fts_close(FTS *sp)
237{
238	FTSENT *freep, *p;
239	int saved_errno = 0;
240
241	_DIAGASSERT(sp != NULL);
242
243	/*
244	 * This still works if we haven't read anything -- the dummy structure
245	 * points to the root list, so we step through to the end of the root
246	 * list which has a valid parent pointer.
247	 */
248	if (sp->fts_cur) {
249		if (ISSET(FTS_SYMFOLLOW))
250			(void)close(sp->fts_cur->fts_symfd);
251		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
252			freep = p;
253			p = p->fts_link ? p->fts_link : p->fts_parent;
254			free(freep);
255		}
256		free(p);
257	}
258
259	/* Free up child linked list, sort array, path buffer. */
260	if (sp->fts_child)
261		fts_lfree(sp->fts_child);
262	if (sp->fts_array)
263		free(sp->fts_array);
264	free(sp->fts_path);
265
266	/* Return to original directory, save errno if necessary. */
267	if (!ISSET(FTS_NOCHDIR)) {
268		if (fchdir(sp->fts_rfd) == -1)
269			saved_errno = errno;
270		(void)close(sp->fts_rfd);
271	}
272
273	/* Free up the stream pointer. */
274	free(sp);
275	if (saved_errno) {
276		errno = saved_errno;
277		return -1;
278	}
279
280	return 0;
281}
282
283#if !defined(__FTS_COMPAT_TAILINGSLASH)
284
285/*
286 * Special case of "/" at the end of the path so that slashes aren't
287 * appended which would cause paths to be written as "....//foo".
288 */
289#define	NAPPEND(p)							\
290	(p->fts_path[p->fts_pathlen - 1] == '/'				\
291	    ? p->fts_pathlen - 1 : p->fts_pathlen)
292
293#else /* !defined(__FTS_COMPAT_TAILINGSLASH) */
294
295/*
296 * compatibility with the old behaviour.
297 *
298 * Special case a root of "/" so that slashes aren't appended which would
299 * cause paths to be written as "//foo".
300 */
301
302#define	NAPPEND(p)							\
303	(p->fts_level == FTS_ROOTLEVEL && p->fts_pathlen == 1 &&	\
304	    p->fts_path[0] == '/' ? 0 : p->fts_pathlen)
305
306#endif /* !defined(__FTS_COMPAT_TAILINGSLASH) */
307
308FTSENT *
309fts_read(FTS *sp)
310{
311	FTSENT *p, *tmp;
312	int instr;
313	char *t;
314	int saved_errno;
315
316	_DIAGASSERT(sp != NULL);
317
318	/* If finished or unrecoverable error, return NULL. */
319	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
320		return (NULL);
321
322	/* Set current node pointer. */
323	p = sp->fts_cur;
324
325	/* Save and zero out user instructions. */
326	instr = p->fts_instr;
327	p->fts_instr = FTS_NOINSTR;
328
329	/* Any type of file may be re-visited; re-stat and re-turn. */
330	if (instr == FTS_AGAIN) {
331		p->fts_info = fts_stat(sp, p, 0);
332		return (p);
333	}
334
335	/*
336	 * Following a symlink -- SLNONE test allows application to see
337	 * SLNONE and recover.  If indirecting through a symlink, have
338	 * keep a pointer to current location.  If unable to get that
339	 * pointer, follow fails.
340	 */
341	if (instr == FTS_FOLLOW &&
342	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
343		p->fts_info = fts_stat(sp, p, 1);
344		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
345			if ((p->fts_symfd = open(".", O_RDONLY, 0)) == -1) {
346				p->fts_errno = errno;
347				p->fts_info = FTS_ERR;
348			} else if (fcntl(p->fts_symfd, F_SETFD, FD_CLOEXEC) == -1) {
349				p->fts_errno = errno;
350				p->fts_info = FTS_ERR;
351				close(p->fts_symfd);
352			} else
353				p->fts_flags |= FTS_SYMFOLLOW;
354		}
355		return (p);
356	}
357
358	/* Directory in pre-order. */
359	if (p->fts_info == FTS_D) {
360		/* If skipped or crossed mount point, do post-order visit. */
361		if (instr == FTS_SKIP ||
362		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
363			if (p->fts_flags & FTS_SYMFOLLOW)
364				(void)close(p->fts_symfd);
365			if (sp->fts_child) {
366				fts_lfree(sp->fts_child);
367				sp->fts_child = NULL;
368			}
369			p->fts_info = FTS_DP;
370			return (p);
371		}
372
373		/* Rebuild if only read the names and now traversing. */
374		if (sp->fts_child && ISSET(FTS_NAMEONLY)) {
375			CLR(FTS_NAMEONLY);
376			fts_lfree(sp->fts_child);
377			sp->fts_child = NULL;
378		}
379
380		/*
381		 * Cd to the subdirectory.
382		 *
383		 * If have already read and now fail to chdir, whack the list
384		 * to make the names come out right, and set the parent errno
385		 * so the application will eventually get an error condition.
386		 * Set the FTS_DONTCHDIR flag so that when we logically change
387		 * directories back to the parent we don't do a chdir.
388		 *
389		 * If haven't read do so.  If the read fails, fts_build sets
390		 * FTS_STOP or the fts_info field of the node.
391		 */
392		if (sp->fts_child) {
393			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
394				p->fts_errno = errno;
395				p->fts_flags |= FTS_DONTCHDIR;
396				for (p = sp->fts_child; p; p = p->fts_link)
397					p->fts_accpath =
398					    p->fts_parent->fts_accpath;
399			}
400		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
401			if (ISSET(FTS_STOP))
402				return (NULL);
403			return (p);
404		}
405		p = sp->fts_child;
406		sp->fts_child = NULL;
407		goto name;
408	}
409
410	/* Move to the next node on this level. */
411next:	tmp = p;
412	if ((p = p->fts_link) != NULL) {
413		free(tmp);
414
415		/*
416		 * If reached the top, return to the original directory, and
417		 * load the paths for the next root.
418		 */
419		if (p->fts_level == FTS_ROOTLEVEL) {
420			if (FCHDIR(sp, sp->fts_rfd)) {
421				SET(FTS_STOP);
422				return (NULL);
423			}
424			fts_load(sp, p);
425			return (sp->fts_cur = p);
426		}
427
428		/*
429		 * User may have called fts_set on the node.  If skipped,
430		 * ignore.  If followed, get a file descriptor so we can
431		 * get back if necessary.
432		 */
433		if (p->fts_instr == FTS_SKIP)
434			goto next;
435		if (p->fts_instr == FTS_FOLLOW) {
436			p->fts_info = fts_stat(sp, p, 1);
437			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
438				if ((p->fts_symfd =
439				    open(".", O_RDONLY, 0)) == -1) {
440					p->fts_errno = errno;
441					p->fts_info = FTS_ERR;
442				} else if (fcntl(p->fts_symfd, F_SETFD, FD_CLOEXEC) == -1) {
443					p->fts_errno = errno;
444					p->fts_info = FTS_ERR;
445					close(p->fts_symfd);
446				} else
447					p->fts_flags |= FTS_SYMFOLLOW;
448			}
449			p->fts_instr = FTS_NOINSTR;
450		}
451
452name:		t = sp->fts_path + NAPPEND(p->fts_parent);
453		*t++ = '/';
454		memmove(t, p->fts_name, (size_t)(p->fts_namelen + 1));
455		return (sp->fts_cur = p);
456	}
457
458	/* Move up to the parent node. */
459	p = tmp->fts_parent;
460	free(tmp);
461
462	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
463		/*
464		 * Done; free everything up and set errno to 0 so the user
465		 * can distinguish between error and EOF.
466		 */
467		free(p);
468		errno = 0;
469		return (sp->fts_cur = NULL);
470	}
471
472	/* Nul terminate the pathname. */
473	sp->fts_path[p->fts_pathlen] = '\0';
474
475	/*
476	 * Return to the parent directory.  If at a root node or came through
477	 * a symlink, go back through the file descriptor.  Otherwise, cd up
478	 * one directory.
479	 */
480	if (p->fts_level == FTS_ROOTLEVEL) {
481		if (FCHDIR(sp, sp->fts_rfd)) {
482			SET(FTS_STOP);
483			return (NULL);
484		}
485	} else if (p->fts_flags & FTS_SYMFOLLOW) {
486		if (FCHDIR(sp, p->fts_symfd)) {
487			saved_errno = errno;
488			(void)close(p->fts_symfd);
489			errno = saved_errno;
490			SET(FTS_STOP);
491			return (NULL);
492		}
493		(void)close(p->fts_symfd);
494	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
495	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
496		SET(FTS_STOP);
497		return (NULL);
498	}
499	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
500	return (sp->fts_cur = p);
501}
502
503/*
504 * Fts_set takes the stream as an argument although it's not used in this
505 * implementation; it would be necessary if anyone wanted to add global
506 * semantics to fts using fts_set.  An error return is allowed for similar
507 * reasons.
508 */
509/* ARGSUSED */
510int
511fts_set(FTS *sp, FTSENT *p, int instr)
512{
513
514	_DIAGASSERT(sp != NULL);
515	_DIAGASSERT(p != NULL);
516
517	if (instr && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
518	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
519		errno = EINVAL;
520		return (1);
521	}
522	p->fts_instr = instr;
523	return (0);
524}
525
526FTSENT *
527fts_children(FTS *sp, int instr)
528{
529	FTSENT *p;
530	int fd;
531
532	_DIAGASSERT(sp != NULL);
533
534	if (instr && instr != FTS_NAMEONLY) {
535		errno = EINVAL;
536		return (NULL);
537	}
538
539	/* Set current node pointer. */
540	p = sp->fts_cur;
541
542	/*
543	 * Errno set to 0 so user can distinguish empty directory from
544	 * an error.
545	 */
546	errno = 0;
547
548	/* Fatal errors stop here. */
549	if (ISSET(FTS_STOP))
550		return (NULL);
551
552	/* Return logical hierarchy of user's arguments. */
553	if (p->fts_info == FTS_INIT)
554		return (p->fts_link);
555
556	/*
557	 * If not a directory being visited in pre-order, stop here.  Could
558	 * allow FTS_DNR, assuming the user has fixed the problem, but the
559	 * same effect is available with FTS_AGAIN.
560	 */
561	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
562		return (NULL);
563
564	/* Free up any previous child list. */
565	if (sp->fts_child)
566		fts_lfree(sp->fts_child);
567
568	if (instr == FTS_NAMEONLY) {
569		SET(FTS_NAMEONLY);
570		instr = BNAMES;
571	} else
572		instr = BCHILD;
573
574	/*
575	 * If using chdir on a relative path and called BEFORE fts_read does
576	 * its chdir to the root of a traversal, we can lose -- we need to
577	 * chdir into the subdirectory, and we don't know where the current
578	 * directory is, so we can't get back so that the upcoming chdir by
579	 * fts_read will work.
580	 */
581	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
582	    ISSET(FTS_NOCHDIR))
583		return (sp->fts_child = fts_build(sp, instr));
584
585	if ((fd = open(".", O_RDONLY, 0)) == -1)
586		return (sp->fts_child = NULL);
587	sp->fts_child = fts_build(sp, instr);
588	if (fchdir(fd)) {
589		(void)close(fd);
590		return (NULL);
591	}
592	(void)close(fd);
593	return (sp->fts_child);
594}
595
596/*
597 * This is the tricky part -- do not casually change *anything* in here.  The
598 * idea is to build the linked list of entries that are used by fts_children
599 * and fts_read.  There are lots of special cases.
600 *
601 * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
602 * set and it's a physical walk (so that symbolic links can't be directories),
603 * we can do things quickly.  First, if it's a 4.4BSD file system, the type
604 * of the file is in the directory entry.  Otherwise, we assume that the number
605 * of subdirectories in a node is equal to the number of links to the parent.
606 * The former skips all stat calls.  The latter skips stat calls in any leaf
607 * directories and for any files after the subdirectories in the directory have
608 * been found, cutting the stat calls by about 2/3.
609 */
610static FTSENT *
611fts_build(FTS *sp, int type)
612{
613	struct dirent *dp;
614	FTSENT *p, *head;
615	size_t nitems;
616	FTSENT *cur, *tail;
617	DIR *dirp;
618	void *oldaddr;
619	size_t dnamlen;
620	int cderrno, descend, len, level, nlinks, saved_errno, nostat, doadjust;
621	size_t maxlen;
622#ifdef FTS_WHITEOUT
623	int oflag;
624#endif
625	char *cp = NULL;	/* pacify gcc */
626
627	_DIAGASSERT(sp != NULL);
628
629	/* Set current node pointer. */
630	cur = sp->fts_cur;
631
632	/*
633	 * Open the directory for reading.  If this fails, we're done.
634	 * If being called from fts_read, set the fts_info field.
635	 */
636#ifdef FTS_WHITEOUT
637	if (ISSET(FTS_WHITEOUT))
638		oflag = DTF_NODUP|DTF_REWIND;
639	else
640		oflag = DTF_HIDEW|DTF_NODUP|DTF_REWIND;
641#else
642#define __opendir2(path, flag) opendir(path)
643#endif
644	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
645		if (type == BREAD) {
646			cur->fts_info = FTS_DNR;
647			cur->fts_errno = errno;
648		}
649		return (NULL);
650	}
651
652	/*
653	 * Nlinks is the number of possible entries of type directory in the
654	 * directory if we're cheating on stat calls, 0 if we're not doing
655	 * any stat calls at all, -1 if we're doing stats on everything.
656	 */
657	if (type == BNAMES) {
658		nlinks = 0;
659		nostat = 1;
660	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
661		nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
662		nostat = 1;
663	} else {
664		nlinks = -1;
665		nostat = 0;
666	}
667
668#ifdef notdef
669	(void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
670	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
671	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
672#endif
673	/*
674	 * If we're going to need to stat anything or we want to descend
675	 * and stay in the directory, chdir.  If this fails we keep going,
676	 * but set a flag so we don't chdir after the post-order visit.
677	 * We won't be able to stat anything, but we can still return the
678	 * names themselves.  Note, that since fts_read won't be able to
679	 * chdir into the directory, it will have to return different path
680	 * names than before, i.e. "a/b" instead of "b".  Since the node
681	 * has already been visited in pre-order, have to wait until the
682	 * post-order visit to return the error.  There is a special case
683	 * here, if there was nothing to stat then it's not an error to
684	 * not be able to stat.  This is all fairly nasty.  If a program
685	 * needed sorted entries or stat information, they had better be
686	 * checking FTS_NS on the returned nodes.
687	 */
688	cderrno = 0;
689	if (nlinks || type == BREAD) {
690		if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) {
691			if (nlinks && type == BREAD)
692				cur->fts_errno = errno;
693			cur->fts_flags |= FTS_DONTCHDIR;
694			descend = 0;
695			cderrno = errno;
696		} else
697			descend = 1;
698	} else
699		descend = 0;
700
701	/*
702	 * Figure out the max file name length that can be stored in the
703	 * current path -- the inner loop allocates more path as necessary.
704	 * We really wouldn't have to do the maxlen calculations here, we
705	 * could do them in fts_read before returning the path, but it's a
706	 * lot easier here since the length is part of the dirent structure.
707	 *
708	 * If not changing directories set a pointer so that can just append
709	 * each new name into the path.
710	 */
711	len = NAPPEND(cur);
712	if (ISSET(FTS_NOCHDIR)) {
713		cp = sp->fts_path + len;
714		*cp++ = '/';
715	}
716	len++;
717	maxlen = sp->fts_pathlen - len;
718
719	level = cur->fts_level + 1;
720
721	/* Read the directory, attaching each entry to the `link' pointer. */
722	doadjust = 0;
723	for (head = tail = NULL, nitems = 0; (dp = readdir(dirp)) != NULL;) {
724
725		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
726			continue;
727
728#if HAVE_STRUCT_DIRENT_D_NAMLEN
729		dnamlen = dp->d_namlen;
730#else
731		dnamlen = strlen(dp->d_name);
732#endif
733		if ((p = fts_alloc(sp, dp->d_name, dnamlen)) == NULL)
734			goto mem1;
735		if (dnamlen >= maxlen) {	/* include space for NUL */
736			oldaddr = sp->fts_path;
737			if (fts_palloc(sp, dnamlen + len + 1)) {
738				/*
739				 * No more memory for path or structures.  Save
740				 * errno, free up the current structure and the
741				 * structures already allocated.
742				 */
743mem1:				saved_errno = errno;
744				if (p)
745					free(p);
746				fts_lfree(head);
747				(void)closedir(dirp);
748				errno = saved_errno;
749				cur->fts_info = FTS_ERR;
750				SET(FTS_STOP);
751				return (NULL);
752			}
753			/* Did realloc() change the pointer? */
754			if (oldaddr != sp->fts_path) {
755				doadjust = 1;
756				if (ISSET(FTS_NOCHDIR))
757					cp = sp->fts_path + len;
758			}
759			maxlen = sp->fts_pathlen - len;
760		}
761
762		if (len + dnamlen >= USHRT_MAX) {
763			/*
764			 * In an FTSENT, fts_pathlen is a u_short so it is
765			 * possible to wraparound here.  If we do, free up
766			 * the current structure and the structures already
767			 * allocated, then error out with ENAMETOOLONG.
768			 */
769			free(p);
770			fts_lfree(head);
771			(void)closedir(dirp);
772			cur->fts_info = FTS_ERR;
773			SET(FTS_STOP);
774			errno = ENAMETOOLONG;
775			return (NULL);
776		}
777		p->fts_level = level;
778		p->fts_pathlen = len + dnamlen;
779		p->fts_parent = sp->fts_cur;
780
781#ifdef FTS_WHITEOUT
782		if (dp->d_type == DT_WHT)
783			p->fts_flags |= FTS_ISW;
784#endif
785
786		if (cderrno) {
787			if (nlinks) {
788				p->fts_info = FTS_NS;
789				p->fts_errno = cderrno;
790			} else
791				p->fts_info = FTS_NSOK;
792			p->fts_accpath = cur->fts_accpath;
793		} else if (nlinks == 0
794#ifdef DT_DIR
795		    || (nostat &&
796		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
797#endif
798		    ) {
799			p->fts_accpath =
800			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
801			p->fts_info = FTS_NSOK;
802		} else {
803			/* Build a file name for fts_stat to stat. */
804			if (ISSET(FTS_NOCHDIR)) {
805				p->fts_accpath = p->fts_path;
806				memmove(cp, p->fts_name,
807				        (size_t)(p->fts_namelen + 1));
808			} else
809				p->fts_accpath = p->fts_name;
810			/* Stat it. */
811			p->fts_info = fts_stat(sp, p, 0);
812
813			/* Decrement link count if applicable. */
814			if (nlinks > 0 && (p->fts_info == FTS_D ||
815			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
816				--nlinks;
817		}
818
819		/* We walk in directory order so "ls -f" doesn't get upset. */
820		p->fts_link = NULL;
821		if (head == NULL)
822			head = tail = p;
823		else {
824			tail->fts_link = p;
825			tail = p;
826		}
827		++nitems;
828	}
829	(void)closedir(dirp);
830
831	/*
832	 * If had to realloc the path, adjust the addresses for the rest
833	 * of the tree.
834	 */
835	if (doadjust)
836		fts_padjust(sp, head);
837
838	/*
839	 * If not changing directories, reset the path back to original
840	 * state.
841	 */
842	if (ISSET(FTS_NOCHDIR)) {
843		if (len == sp->fts_pathlen || nitems == 0)
844			--cp;
845		*cp = '\0';
846	}
847
848	/*
849	 * If descended after called from fts_children or after called from
850	 * fts_read and nothing found, get back.  At the root level we use
851	 * the saved fd; if one of fts_open()'s arguments is a relative path
852	 * to an empty directory, we wind up here with no other way back.  If
853	 * can't get back, we're done.
854	 */
855	if (descend && (type == BCHILD || !nitems) &&
856	    (cur->fts_level == FTS_ROOTLEVEL ?
857	    FCHDIR(sp, sp->fts_rfd) :
858	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
859		cur->fts_info = FTS_ERR;
860		SET(FTS_STOP);
861		return (NULL);
862	}
863
864	/* If didn't find anything, return NULL. */
865	if (!nitems) {
866		if (type == BREAD)
867			cur->fts_info = FTS_DP;
868		return (NULL);
869	}
870
871	/* Sort the entries. */
872	if (sp->fts_compar && nitems > 1)
873		head = fts_sort(sp, head, nitems);
874	return (head);
875}
876
877static u_short
878fts_stat(FTS *sp, FTSENT *p, int follow)
879{
880	FTSENT *t;
881	dev_t dev;
882	__fts_ino_t ino;
883	__fts_stat_t *sbp, sb;
884	int saved_errno;
885
886	_DIAGASSERT(sp != NULL);
887	_DIAGASSERT(p != NULL);
888
889	/* If user needs stat info, stat buffer already allocated. */
890	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
891
892#ifdef FTS_WHITEOUT
893	/* check for whiteout */
894	if (p->fts_flags & FTS_ISW) {
895		if (sbp != &sb) {
896			memset(sbp, '\0', sizeof (*sbp));
897			sbp->st_mode = S_IFWHT;
898		}
899		return (FTS_W);
900	}
901#endif
902
903	/*
904	 * If doing a logical walk, or application requested FTS_FOLLOW, do
905	 * a stat(2).  If that fails, check for a non-existent symlink.  If
906	 * fail, set the errno from the stat call.
907	 */
908	if (ISSET(FTS_LOGICAL) || follow) {
909		if (stat(p->fts_accpath, sbp)) {
910			saved_errno = errno;
911			if (!lstat(p->fts_accpath, sbp)) {
912				errno = 0;
913				return (FTS_SLNONE);
914			}
915			p->fts_errno = saved_errno;
916			goto err;
917		}
918	} else if (lstat(p->fts_accpath, sbp)) {
919		p->fts_errno = errno;
920err:		memset(sbp, 0, sizeof(*sbp));
921		return (FTS_NS);
922	}
923
924	if (S_ISDIR(sbp->st_mode)) {
925		/*
926		 * Set the device/inode.  Used to find cycles and check for
927		 * crossing mount points.  Also remember the link count, used
928		 * in fts_build to limit the number of stat calls.  It is
929		 * understood that these fields are only referenced if fts_info
930		 * is set to FTS_D.
931		 */
932		dev = p->fts_dev = sbp->st_dev;
933		ino = p->fts_ino = sbp->st_ino;
934		p->fts_nlink = sbp->st_nlink;
935
936		if (ISDOT(p->fts_name))
937			return (FTS_DOT);
938
939		/*
940		 * Cycle detection is done by brute force when the directory
941		 * is first encountered.  If the tree gets deep enough or the
942		 * number of symbolic links to directories is high enough,
943		 * something faster might be worthwhile.
944		 */
945		for (t = p->fts_parent;
946		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
947			if (ino == t->fts_ino && dev == t->fts_dev) {
948				p->fts_cycle = t;
949				return (FTS_DC);
950			}
951		return (FTS_D);
952	}
953	if (S_ISLNK(sbp->st_mode))
954		return (FTS_SL);
955	if (S_ISREG(sbp->st_mode))
956		return (FTS_F);
957	return (FTS_DEFAULT);
958}
959
960static FTSENT *
961fts_sort(FTS *sp, FTSENT *head, size_t nitems)
962{
963	FTSENT **ap, *p;
964
965	_DIAGASSERT(sp != NULL);
966	_DIAGASSERT(head != NULL);
967
968	/*
969	 * Construct an array of pointers to the structures and call qsort(3).
970	 * Reassemble the array in the order returned by qsort.  If unable to
971	 * sort for memory reasons, return the directory entries in their
972	 * current order.  Allocate enough space for the current needs plus
973	 * 40 so don't realloc one entry at a time.
974	 */
975	if (nitems > sp->fts_nitems) {
976		FTSENT **new;
977
978		new = realloc(sp->fts_array, sizeof(FTSENT *) * (nitems + 40));
979		if (new == 0)
980			return (head);
981		sp->fts_array = new;
982		sp->fts_nitems = nitems + 40;
983	}
984	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
985		*ap++ = p;
986	qsort((void *)sp->fts_array, nitems, sizeof(FTSENT *),
987		(int (*)(const void *, const void *))sp->fts_compar);
988	for (head = *(ap = sp->fts_array); --nitems; ++ap)
989		ap[0]->fts_link = ap[1];
990	ap[0]->fts_link = NULL;
991	return (head);
992}
993
994static FTSENT *
995fts_alloc(FTS *sp, const char *name, size_t namelen)
996{
997	FTSENT *p;
998	size_t len;
999
1000	_DIAGASSERT(sp != NULL);
1001	_DIAGASSERT(name != NULL);
1002
1003#if defined(ALIGNBYTES) && defined(ALIGN)
1004	/*
1005	 * The file name is a variable length array and no stat structure is
1006	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
1007	 * structure, the file name and the stat structure in one chunk, but
1008	 * be careful that the stat structure is reasonably aligned.  Since the
1009	 * fts_name field is declared to be of size 1, the fts_name pointer is
1010	 * namelen + 2 before the first possible address of the stat structure.
1011	 */
1012	len = sizeof(FTSENT) + namelen;
1013	if (!ISSET(FTS_NOSTAT))
1014		len += sizeof(*(p->fts_statp)) + ALIGNBYTES;
1015	if ((p = malloc(len)) == NULL)
1016		return (NULL);
1017
1018	if (!ISSET(FTS_NOSTAT))
1019		p->fts_statp =
1020		    (__fts_stat_t *)ALIGN((u_long)(p->fts_name + namelen + 2));
1021#else
1022	if ((p = malloc(sizeof(FTSENT) + namelen)) == NULL)
1023		return (NULL);
1024
1025	if (!ISSET(FTS_NOSTAT))
1026		if ((p->fts_statp = malloc(sizeof(*(p->fts_statp)))) == NULL) {
1027			free(p);
1028			return (NULL);
1029		}
1030#endif
1031
1032	/* Copy the name plus the trailing NULL. */
1033	memmove(p->fts_name, name, namelen + 1);
1034
1035	p->fts_namelen = namelen;
1036	p->fts_path = sp->fts_path;
1037	p->fts_errno = 0;
1038	p->fts_flags = 0;
1039	p->fts_instr = FTS_NOINSTR;
1040	p->fts_number = 0;
1041	p->fts_pointer = NULL;
1042	return (p);
1043}
1044
1045static void
1046fts_lfree(FTSENT *head)
1047{
1048	FTSENT *p;
1049
1050	/* XXX: head may be NULL ? */
1051
1052	/* Free a linked list of structures. */
1053	while ((p = head) != NULL) {
1054		head = head->fts_link;
1055
1056#if !defined(ALIGNBYTES) || !defined(ALIGN)
1057		if (p->fts_statp)
1058			free(p->fts_statp);
1059#endif
1060		free(p);
1061	}
1062}
1063
1064static size_t
1065fts_pow2(size_t x)
1066{
1067
1068	x--;
1069	x |= x>>1;
1070	x |= x>>2;
1071	x |= x>>4;
1072	x |= x>>8;
1073	x |= x>>16;
1074#if LONG_BIT > 32
1075	x |= x>>32;
1076#endif
1077#if LONG_BIT > 64
1078	x |= x>>64;
1079#endif
1080	x++;
1081	return (x);
1082}
1083
1084/*
1085 * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
1086 * Most systems will allow creation of paths much longer than MAXPATHLEN, even
1087 * though the kernel won't resolve them.  Round up the new size to a power of 2,
1088 * so we don't realloc the path 2 bytes at a time.
1089 */
1090static int
1091fts_palloc(FTS *sp, size_t size)
1092{
1093	char *new;
1094
1095	_DIAGASSERT(sp != NULL);
1096
1097#if 1
1098	/* Protect against fts_pathlen overflow. */
1099	if (size > USHRT_MAX + 1) {
1100		errno = ENAMETOOLONG;
1101		return (1);
1102	}
1103#endif
1104	size = fts_pow2(size);
1105	new = realloc(sp->fts_path, size);
1106	if (new == 0)
1107		return (1);
1108	sp->fts_path = new;
1109	sp->fts_pathlen = size;
1110	return (0);
1111}
1112
1113/*
1114 * When the path is realloc'd, have to fix all of the pointers in structures
1115 * already returned.
1116 */
1117static void
1118fts_padjust(FTS *sp, FTSENT *head)
1119{
1120	FTSENT *p;
1121	char *addr;
1122
1123	_DIAGASSERT(sp != NULL);
1124
1125#define	ADJUST(p) do {							\
1126	if ((p)->fts_accpath != (p)->fts_name)				\
1127		(p)->fts_accpath =					\
1128		    addr + ((p)->fts_accpath - (p)->fts_path);		\
1129	(p)->fts_path = addr;						\
1130} while (/*CONSTCOND*/0)
1131
1132	addr = sp->fts_path;
1133
1134	/* Adjust the current set of children. */
1135	for (p = sp->fts_child; p; p = p->fts_link)
1136		ADJUST(p);
1137
1138	/* Adjust the rest of the tree, including the current level. */
1139	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1140		ADJUST(p);
1141		p = p->fts_link ? p->fts_link : p->fts_parent;
1142	}
1143}
1144
1145static size_t
1146fts_maxarglen(char * const *argv)
1147{
1148	size_t len, max;
1149
1150	_DIAGASSERT(argv != NULL);
1151
1152	for (max = 0; *argv; ++argv)
1153		if ((len = strlen(*argv)) > max)
1154			max = len;
1155	return (max + 1);
1156}
1157
1158/*
1159 * Change to dir specified by fd or p->fts_accpath without getting
1160 * tricked by someone changing the world out from underneath us.
1161 * Assumes p->fts_dev and p->fts_ino are filled in.
1162 */
1163static int
1164fts_safe_changedir(const FTS *sp, const FTSENT *p, int fd, const char *path)
1165{
1166	int oldfd = fd, ret = -1;
1167	__fts_stat_t sb;
1168
1169	if (ISSET(FTS_NOCHDIR))
1170		return 0;
1171
1172	if (oldfd < 0 && (fd = open(path, O_RDONLY)) == -1)
1173		return -1;
1174
1175	if (fstat(fd, &sb) == -1)
1176		goto bail;
1177
1178	if (sb.st_ino != p->fts_ino || sb.st_dev != p->fts_dev) {
1179		errno = ENOENT;
1180		goto bail;
1181	}
1182
1183	ret = fchdir(fd);
1184
1185bail:
1186	if (oldfd < 0) {
1187		int save_errno = errno;
1188		(void)close(fd);
1189		errno = save_errno;
1190	}
1191	return ret;
1192}
1193