fts.c revision 1.28
1/*	$NetBSD: fts.c,v 1.28 2005/11/18 16:48:48 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 1.28 2005/11/18 16:48:48 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;
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		saved_errno = fchdir(sp->fts_rfd) ? errno : 0;
269		(void)close(sp->fts_rfd);
270		/* Set errno and return. */
271		if (saved_errno) {
272			errno = saved_errno;
273			return (-1);
274		}
275	}
276
277	/* Free up the stream pointer. */
278	free(sp);
279	/* ISSET() is illegal after this, since the macro touches sp */
280
281	return (0);
282}
283
284/*
285 * Special case of "/" at the end of the path so that slashes aren't
286 * appended which would cause paths to be written as "....//foo".
287 */
288#define	NAPPEND(p)							\
289	(p->fts_path[p->fts_pathlen - 1] == '/'				\
290	    ? p->fts_pathlen - 1 : p->fts_pathlen)
291
292FTSENT *
293fts_read(FTS *sp)
294{
295	FTSENT *p, *tmp;
296	int instr;
297	char *t;
298	int saved_errno;
299
300	_DIAGASSERT(sp != NULL);
301
302	/* If finished or unrecoverable error, return NULL. */
303	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
304		return (NULL);
305
306	/* Set current node pointer. */
307	p = sp->fts_cur;
308
309	/* Save and zero out user instructions. */
310	instr = p->fts_instr;
311	p->fts_instr = FTS_NOINSTR;
312
313	/* Any type of file may be re-visited; re-stat and re-turn. */
314	if (instr == FTS_AGAIN) {
315		p->fts_info = fts_stat(sp, p, 0);
316		return (p);
317	}
318
319	/*
320	 * Following a symlink -- SLNONE test allows application to see
321	 * SLNONE and recover.  If indirecting through a symlink, have
322	 * keep a pointer to current location.  If unable to get that
323	 * pointer, follow fails.
324	 */
325	if (instr == FTS_FOLLOW &&
326	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
327		p->fts_info = fts_stat(sp, p, 1);
328		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
329			if ((p->fts_symfd = open(".", O_RDONLY, 0)) == -1) {
330				p->fts_errno = errno;
331				p->fts_info = FTS_ERR;
332			} else if (fcntl(p->fts_symfd, F_SETFD, FD_CLOEXEC) == -1) {
333				p->fts_errno = errno;
334				p->fts_info = FTS_ERR;
335				close(p->fts_symfd);
336			} else
337				p->fts_flags |= FTS_SYMFOLLOW;
338		}
339		return (p);
340	}
341
342	/* Directory in pre-order. */
343	if (p->fts_info == FTS_D) {
344		/* If skipped or crossed mount point, do post-order visit. */
345		if (instr == FTS_SKIP ||
346		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
347			if (p->fts_flags & FTS_SYMFOLLOW)
348				(void)close(p->fts_symfd);
349			if (sp->fts_child) {
350				fts_lfree(sp->fts_child);
351				sp->fts_child = NULL;
352			}
353			p->fts_info = FTS_DP;
354			return (p);
355		}
356
357		/* Rebuild if only read the names and now traversing. */
358		if (sp->fts_child && ISSET(FTS_NAMEONLY)) {
359			CLR(FTS_NAMEONLY);
360			fts_lfree(sp->fts_child);
361			sp->fts_child = NULL;
362		}
363
364		/*
365		 * Cd to the subdirectory.
366		 *
367		 * If have already read and now fail to chdir, whack the list
368		 * to make the names come out right, and set the parent errno
369		 * so the application will eventually get an error condition.
370		 * Set the FTS_DONTCHDIR flag so that when we logically change
371		 * directories back to the parent we don't do a chdir.
372		 *
373		 * If haven't read do so.  If the read fails, fts_build sets
374		 * FTS_STOP or the fts_info field of the node.
375		 */
376		if (sp->fts_child) {
377			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
378				p->fts_errno = errno;
379				p->fts_flags |= FTS_DONTCHDIR;
380				for (p = sp->fts_child; p; p = p->fts_link)
381					p->fts_accpath =
382					    p->fts_parent->fts_accpath;
383			}
384		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
385			if (ISSET(FTS_STOP))
386				return (NULL);
387			return (p);
388		}
389		p = sp->fts_child;
390		sp->fts_child = NULL;
391		goto name;
392	}
393
394	/* Move to the next node on this level. */
395next:	tmp = p;
396	if ((p = p->fts_link) != NULL) {
397		free(tmp);
398
399		/*
400		 * If reached the top, return to the original directory, and
401		 * load the paths for the next root.
402		 */
403		if (p->fts_level == FTS_ROOTLEVEL) {
404			if (FCHDIR(sp, sp->fts_rfd)) {
405				SET(FTS_STOP);
406				return (NULL);
407			}
408			fts_load(sp, p);
409			return (sp->fts_cur = p);
410		}
411
412		/*
413		 * User may have called fts_set on the node.  If skipped,
414		 * ignore.  If followed, get a file descriptor so we can
415		 * get back if necessary.
416		 */
417		if (p->fts_instr == FTS_SKIP)
418			goto next;
419		if (p->fts_instr == FTS_FOLLOW) {
420			p->fts_info = fts_stat(sp, p, 1);
421			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
422				if ((p->fts_symfd =
423				    open(".", O_RDONLY, 0)) == -1) {
424					p->fts_errno = errno;
425					p->fts_info = FTS_ERR;
426				} else if (fcntl(p->fts_symfd, F_SETFD, FD_CLOEXEC) == -1) {
427					p->fts_errno = errno;
428					p->fts_info = FTS_ERR;
429					close(p->fts_symfd);
430				} else
431					p->fts_flags |= FTS_SYMFOLLOW;
432			}
433			p->fts_instr = FTS_NOINSTR;
434		}
435
436name:		t = sp->fts_path + NAPPEND(p->fts_parent);
437		*t++ = '/';
438		memmove(t, p->fts_name, (size_t)(p->fts_namelen + 1));
439		return (sp->fts_cur = p);
440	}
441
442	/* Move up to the parent node. */
443	p = tmp->fts_parent;
444	free(tmp);
445
446	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
447		/*
448		 * Done; free everything up and set errno to 0 so the user
449		 * can distinguish between error and EOF.
450		 */
451		free(p);
452		errno = 0;
453		return (sp->fts_cur = NULL);
454	}
455
456	/* Nul terminate the pathname. */
457	sp->fts_path[p->fts_pathlen] = '\0';
458
459	/*
460	 * Return to the parent directory.  If at a root node or came through
461	 * a symlink, go back through the file descriptor.  Otherwise, cd up
462	 * one directory.
463	 */
464	if (p->fts_level == FTS_ROOTLEVEL) {
465		if (FCHDIR(sp, sp->fts_rfd)) {
466			SET(FTS_STOP);
467			return (NULL);
468		}
469	} else if (p->fts_flags & FTS_SYMFOLLOW) {
470		if (FCHDIR(sp, p->fts_symfd)) {
471			saved_errno = errno;
472			(void)close(p->fts_symfd);
473			errno = saved_errno;
474			SET(FTS_STOP);
475			return (NULL);
476		}
477		(void)close(p->fts_symfd);
478	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
479	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
480		SET(FTS_STOP);
481		return (NULL);
482	}
483	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
484	return (sp->fts_cur = p);
485}
486
487/*
488 * Fts_set takes the stream as an argument although it's not used in this
489 * implementation; it would be necessary if anyone wanted to add global
490 * semantics to fts using fts_set.  An error return is allowed for similar
491 * reasons.
492 */
493/* ARGSUSED */
494int
495fts_set(FTS *sp, FTSENT *p, int instr)
496{
497
498	_DIAGASSERT(sp != NULL);
499	_DIAGASSERT(p != NULL);
500
501	if (instr && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
502	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
503		errno = EINVAL;
504		return (1);
505	}
506	p->fts_instr = instr;
507	return (0);
508}
509
510FTSENT *
511fts_children(FTS *sp, int instr)
512{
513	FTSENT *p;
514	int fd;
515
516	_DIAGASSERT(sp != NULL);
517
518	if (instr && instr != FTS_NAMEONLY) {
519		errno = EINVAL;
520		return (NULL);
521	}
522
523	/* Set current node pointer. */
524	p = sp->fts_cur;
525
526	/*
527	 * Errno set to 0 so user can distinguish empty directory from
528	 * an error.
529	 */
530	errno = 0;
531
532	/* Fatal errors stop here. */
533	if (ISSET(FTS_STOP))
534		return (NULL);
535
536	/* Return logical hierarchy of user's arguments. */
537	if (p->fts_info == FTS_INIT)
538		return (p->fts_link);
539
540	/*
541	 * If not a directory being visited in pre-order, stop here.  Could
542	 * allow FTS_DNR, assuming the user has fixed the problem, but the
543	 * same effect is available with FTS_AGAIN.
544	 */
545	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
546		return (NULL);
547
548	/* Free up any previous child list. */
549	if (sp->fts_child)
550		fts_lfree(sp->fts_child);
551
552	if (instr == FTS_NAMEONLY) {
553		SET(FTS_NAMEONLY);
554		instr = BNAMES;
555	} else
556		instr = BCHILD;
557
558	/*
559	 * If using chdir on a relative path and called BEFORE fts_read does
560	 * its chdir to the root of a traversal, we can lose -- we need to
561	 * chdir into the subdirectory, and we don't know where the current
562	 * directory is, so we can't get back so that the upcoming chdir by
563	 * fts_read will work.
564	 */
565	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
566	    ISSET(FTS_NOCHDIR))
567		return (sp->fts_child = fts_build(sp, instr));
568
569	if ((fd = open(".", O_RDONLY, 0)) == -1)
570		return (sp->fts_child = NULL);
571	sp->fts_child = fts_build(sp, instr);
572	if (fchdir(fd)) {
573		(void)close(fd);
574		return (NULL);
575	}
576	(void)close(fd);
577	return (sp->fts_child);
578}
579
580/*
581 * This is the tricky part -- do not casually change *anything* in here.  The
582 * idea is to build the linked list of entries that are used by fts_children
583 * and fts_read.  There are lots of special cases.
584 *
585 * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
586 * set and it's a physical walk (so that symbolic links can't be directories),
587 * we can do things quickly.  First, if it's a 4.4BSD file system, the type
588 * of the file is in the directory entry.  Otherwise, we assume that the number
589 * of subdirectories in a node is equal to the number of links to the parent.
590 * The former skips all stat calls.  The latter skips stat calls in any leaf
591 * directories and for any files after the subdirectories in the directory have
592 * been found, cutting the stat calls by about 2/3.
593 */
594static FTSENT *
595fts_build(FTS *sp, int type)
596{
597	struct dirent *dp;
598	FTSENT *p, *head;
599	size_t nitems;
600	FTSENT *cur, *tail;
601	DIR *dirp;
602	void *oldaddr;
603	size_t dnamlen;
604	int cderrno, descend, len, level, nlinks, saved_errno, nostat, doadjust;
605	size_t maxlen;
606#ifdef FTS_WHITEOUT
607	int oflag;
608#endif
609	char *cp = NULL;	/* pacify gcc */
610
611	_DIAGASSERT(sp != NULL);
612
613	/* Set current node pointer. */
614	cur = sp->fts_cur;
615
616	/*
617	 * Open the directory for reading.  If this fails, we're done.
618	 * If being called from fts_read, set the fts_info field.
619	 */
620#ifdef FTS_WHITEOUT
621	if (ISSET(FTS_WHITEOUT))
622		oflag = DTF_NODUP|DTF_REWIND;
623	else
624		oflag = DTF_HIDEW|DTF_NODUP|DTF_REWIND;
625#else
626#define __opendir2(path, flag) opendir(path)
627#endif
628	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
629		if (type == BREAD) {
630			cur->fts_info = FTS_DNR;
631			cur->fts_errno = errno;
632		}
633		return (NULL);
634	}
635
636	/*
637	 * Nlinks is the number of possible entries of type directory in the
638	 * directory if we're cheating on stat calls, 0 if we're not doing
639	 * any stat calls at all, -1 if we're doing stats on everything.
640	 */
641	if (type == BNAMES) {
642		nlinks = 0;
643		nostat = 1;
644	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
645		nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
646		nostat = 1;
647	} else {
648		nlinks = -1;
649		nostat = 0;
650	}
651
652#ifdef notdef
653	(void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
654	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
655	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
656#endif
657	/*
658	 * If we're going to need to stat anything or we want to descend
659	 * and stay in the directory, chdir.  If this fails we keep going,
660	 * but set a flag so we don't chdir after the post-order visit.
661	 * We won't be able to stat anything, but we can still return the
662	 * names themselves.  Note, that since fts_read won't be able to
663	 * chdir into the directory, it will have to return different path
664	 * names than before, i.e. "a/b" instead of "b".  Since the node
665	 * has already been visited in pre-order, have to wait until the
666	 * post-order visit to return the error.  There is a special case
667	 * here, if there was nothing to stat then it's not an error to
668	 * not be able to stat.  This is all fairly nasty.  If a program
669	 * needed sorted entries or stat information, they had better be
670	 * checking FTS_NS on the returned nodes.
671	 */
672	cderrno = 0;
673	if (nlinks || type == BREAD) {
674		if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) {
675			if (nlinks && type == BREAD)
676				cur->fts_errno = errno;
677			cur->fts_flags |= FTS_DONTCHDIR;
678			descend = 0;
679			cderrno = errno;
680		} else
681			descend = 1;
682	} else
683		descend = 0;
684
685	/*
686	 * Figure out the max file name length that can be stored in the
687	 * current path -- the inner loop allocates more path as necessary.
688	 * We really wouldn't have to do the maxlen calculations here, we
689	 * could do them in fts_read before returning the path, but it's a
690	 * lot easier here since the length is part of the dirent structure.
691	 *
692	 * If not changing directories set a pointer so that can just append
693	 * each new name into the path.
694	 */
695	len = NAPPEND(cur);
696	if (ISSET(FTS_NOCHDIR)) {
697		cp = sp->fts_path + len;
698		*cp++ = '/';
699	}
700	len++;
701	maxlen = sp->fts_pathlen - len;
702
703	level = cur->fts_level + 1;
704
705	/* Read the directory, attaching each entry to the `link' pointer. */
706	doadjust = 0;
707	for (head = tail = NULL, nitems = 0; (dp = readdir(dirp)) != NULL;) {
708
709		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
710			continue;
711
712#if HAVE_STRUCT_DIRENT_D_NAMLEN
713		dnamlen = dp->d_namlen;
714#else
715		dnamlen = strlen(dp->d_name);
716#endif
717		if ((p = fts_alloc(sp, dp->d_name, dnamlen)) == NULL)
718			goto mem1;
719		if (dnamlen >= maxlen) {	/* include space for NUL */
720			oldaddr = sp->fts_path;
721			if (fts_palloc(sp, dnamlen + len + 1)) {
722				/*
723				 * No more memory for path or structures.  Save
724				 * errno, free up the current structure and the
725				 * structures already allocated.
726				 */
727mem1:				saved_errno = errno;
728				if (p)
729					free(p);
730				fts_lfree(head);
731				(void)closedir(dirp);
732				errno = saved_errno;
733				cur->fts_info = FTS_ERR;
734				SET(FTS_STOP);
735				return (NULL);
736			}
737			/* Did realloc() change the pointer? */
738			if (oldaddr != sp->fts_path) {
739				doadjust = 1;
740				if (ISSET(FTS_NOCHDIR))
741					cp = sp->fts_path + len;
742			}
743			maxlen = sp->fts_pathlen - len;
744		}
745
746		if (len + dnamlen >= USHRT_MAX) {
747			/*
748			 * In an FTSENT, fts_pathlen is a u_short so it is
749			 * possible to wraparound here.  If we do, free up
750			 * the current structure and the structures already
751			 * allocated, then error out with ENAMETOOLONG.
752			 */
753			free(p);
754			fts_lfree(head);
755			(void)closedir(dirp);
756			cur->fts_info = FTS_ERR;
757			SET(FTS_STOP);
758			errno = ENAMETOOLONG;
759			return (NULL);
760		}
761		p->fts_level = level;
762		p->fts_pathlen = len + dnamlen;
763		p->fts_parent = sp->fts_cur;
764
765#ifdef FTS_WHITEOUT
766		if (dp->d_type == DT_WHT)
767			p->fts_flags |= FTS_ISW;
768#endif
769
770		if (cderrno) {
771			if (nlinks) {
772				p->fts_info = FTS_NS;
773				p->fts_errno = cderrno;
774			} else
775				p->fts_info = FTS_NSOK;
776			p->fts_accpath = cur->fts_accpath;
777		} else if (nlinks == 0
778#ifdef DT_DIR
779		    || (nostat &&
780		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
781#endif
782		    ) {
783			p->fts_accpath =
784			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
785			p->fts_info = FTS_NSOK;
786		} else {
787			/* Build a file name for fts_stat to stat. */
788			if (ISSET(FTS_NOCHDIR)) {
789				p->fts_accpath = p->fts_path;
790				memmove(cp, p->fts_name,
791				        (size_t)(p->fts_namelen + 1));
792			} else
793				p->fts_accpath = p->fts_name;
794			/* Stat it. */
795			p->fts_info = fts_stat(sp, p, 0);
796
797			/* Decrement link count if applicable. */
798			if (nlinks > 0 && (p->fts_info == FTS_D ||
799			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
800				--nlinks;
801		}
802
803		/* We walk in directory order so "ls -f" doesn't get upset. */
804		p->fts_link = NULL;
805		if (head == NULL)
806			head = tail = p;
807		else {
808			tail->fts_link = p;
809			tail = p;
810		}
811		++nitems;
812	}
813	(void)closedir(dirp);
814
815	/*
816	 * If had to realloc the path, adjust the addresses for the rest
817	 * of the tree.
818	 */
819	if (doadjust)
820		fts_padjust(sp, head);
821
822	/*
823	 * If not changing directories, reset the path back to original
824	 * state.
825	 */
826	if (ISSET(FTS_NOCHDIR)) {
827		if (len == sp->fts_pathlen || nitems == 0)
828			--cp;
829		*cp = '\0';
830	}
831
832	/*
833	 * If descended after called from fts_children or after called from
834	 * fts_read and nothing found, get back.  At the root level we use
835	 * the saved fd; if one of fts_open()'s arguments is a relative path
836	 * to an empty directory, we wind up here with no other way back.  If
837	 * can't get back, we're done.
838	 */
839	if (descend && (type == BCHILD || !nitems) &&
840	    (cur->fts_level == FTS_ROOTLEVEL ?
841	    FCHDIR(sp, sp->fts_rfd) :
842	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
843		cur->fts_info = FTS_ERR;
844		SET(FTS_STOP);
845		return (NULL);
846	}
847
848	/* If didn't find anything, return NULL. */
849	if (!nitems) {
850		if (type == BREAD)
851			cur->fts_info = FTS_DP;
852		return (NULL);
853	}
854
855	/* Sort the entries. */
856	if (sp->fts_compar && nitems > 1)
857		head = fts_sort(sp, head, nitems);
858	return (head);
859}
860
861static u_short
862fts_stat(FTS *sp, FTSENT *p, int follow)
863{
864	FTSENT *t;
865	dev_t dev;
866	__fts_ino_t ino;
867	__fts_stat_t *sbp, sb;
868	int saved_errno;
869
870	_DIAGASSERT(sp != NULL);
871	_DIAGASSERT(p != NULL);
872
873	/* If user needs stat info, stat buffer already allocated. */
874	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
875
876#ifdef FTS_WHITEOUT
877	/* check for whiteout */
878	if (p->fts_flags & FTS_ISW) {
879		if (sbp != &sb) {
880			memset(sbp, '\0', sizeof (*sbp));
881			sbp->st_mode = S_IFWHT;
882		}
883		return (FTS_W);
884	}
885#endif
886
887	/*
888	 * If doing a logical walk, or application requested FTS_FOLLOW, do
889	 * a stat(2).  If that fails, check for a non-existent symlink.  If
890	 * fail, set the errno from the stat call.
891	 */
892	if (ISSET(FTS_LOGICAL) || follow) {
893		if (stat(p->fts_accpath, sbp)) {
894			saved_errno = errno;
895			if (!lstat(p->fts_accpath, sbp)) {
896				errno = 0;
897				return (FTS_SLNONE);
898			}
899			p->fts_errno = saved_errno;
900			goto err;
901		}
902	} else if (lstat(p->fts_accpath, sbp)) {
903		p->fts_errno = errno;
904err:		memset(sbp, 0, sizeof(*sbp));
905		return (FTS_NS);
906	}
907
908	if (S_ISDIR(sbp->st_mode)) {
909		/*
910		 * Set the device/inode.  Used to find cycles and check for
911		 * crossing mount points.  Also remember the link count, used
912		 * in fts_build to limit the number of stat calls.  It is
913		 * understood that these fields are only referenced if fts_info
914		 * is set to FTS_D.
915		 */
916		dev = p->fts_dev = sbp->st_dev;
917		ino = p->fts_ino = sbp->st_ino;
918		p->fts_nlink = sbp->st_nlink;
919
920		if (ISDOT(p->fts_name))
921			return (FTS_DOT);
922
923		/*
924		 * Cycle detection is done by brute force when the directory
925		 * is first encountered.  If the tree gets deep enough or the
926		 * number of symbolic links to directories is high enough,
927		 * something faster might be worthwhile.
928		 */
929		for (t = p->fts_parent;
930		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
931			if (ino == t->fts_ino && dev == t->fts_dev) {
932				p->fts_cycle = t;
933				return (FTS_DC);
934			}
935		return (FTS_D);
936	}
937	if (S_ISLNK(sbp->st_mode))
938		return (FTS_SL);
939	if (S_ISREG(sbp->st_mode))
940		return (FTS_F);
941	return (FTS_DEFAULT);
942}
943
944static FTSENT *
945fts_sort(FTS *sp, FTSENT *head, size_t nitems)
946{
947	FTSENT **ap, *p;
948
949	_DIAGASSERT(sp != NULL);
950	_DIAGASSERT(head != NULL);
951
952	/*
953	 * Construct an array of pointers to the structures and call qsort(3).
954	 * Reassemble the array in the order returned by qsort.  If unable to
955	 * sort for memory reasons, return the directory entries in their
956	 * current order.  Allocate enough space for the current needs plus
957	 * 40 so don't realloc one entry at a time.
958	 */
959	if (nitems > sp->fts_nitems) {
960		FTSENT **new;
961
962		new = realloc(sp->fts_array, sizeof(FTSENT *) * (nitems + 40));
963		if (new == 0)
964			return (head);
965		sp->fts_array = new;
966		sp->fts_nitems = nitems + 40;
967	}
968	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
969		*ap++ = p;
970	qsort((void *)sp->fts_array, nitems, sizeof(FTSENT *),
971		(int (*)(const void *, const void *))sp->fts_compar);
972	for (head = *(ap = sp->fts_array); --nitems; ++ap)
973		ap[0]->fts_link = ap[1];
974	ap[0]->fts_link = NULL;
975	return (head);
976}
977
978static FTSENT *
979fts_alloc(FTS *sp, const char *name, size_t namelen)
980{
981	FTSENT *p;
982	size_t len;
983
984	_DIAGASSERT(sp != NULL);
985	_DIAGASSERT(name != NULL);
986
987#if defined(ALIGNBYTES) && defined(ALIGN)
988	/*
989	 * The file name is a variable length array and no stat structure is
990	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
991	 * structure, the file name and the stat structure in one chunk, but
992	 * be careful that the stat structure is reasonably aligned.  Since the
993	 * fts_name field is declared to be of size 1, the fts_name pointer is
994	 * namelen + 2 before the first possible address of the stat structure.
995	 */
996	len = sizeof(FTSENT) + namelen;
997	if (!ISSET(FTS_NOSTAT))
998		len += sizeof(*(p->fts_statp)) + ALIGNBYTES;
999	if ((p = malloc(len)) == NULL)
1000		return (NULL);
1001
1002	if (!ISSET(FTS_NOSTAT))
1003		p->fts_statp =
1004		    (__fts_stat_t *)ALIGN((u_long)(p->fts_name + namelen + 2));
1005#else
1006	if ((p = malloc(sizeof(FTSENT) + namelen)) == NULL)
1007		return (NULL);
1008
1009	if (!ISSET(FTS_NOSTAT))
1010		if ((p->fts_statp = malloc(sizeof(*(p->fts_statp)))) == NULL) {
1011			free(p);
1012			return (NULL);
1013		}
1014#endif
1015
1016	/* Copy the name plus the trailing NULL. */
1017	memmove(p->fts_name, name, namelen + 1);
1018
1019	p->fts_namelen = namelen;
1020	p->fts_path = sp->fts_path;
1021	p->fts_errno = 0;
1022	p->fts_flags = 0;
1023	p->fts_instr = FTS_NOINSTR;
1024	p->fts_number = 0;
1025	p->fts_pointer = NULL;
1026	return (p);
1027}
1028
1029static void
1030fts_lfree(FTSENT *head)
1031{
1032	FTSENT *p;
1033
1034	/* XXX: head may be NULL ? */
1035
1036	/* Free a linked list of structures. */
1037	while ((p = head) != NULL) {
1038		head = head->fts_link;
1039
1040#if !defined(ALIGNBYTES) || !defined(ALIGN)
1041		if (p->fts_statp)
1042			free(p->fts_statp);
1043#endif
1044		free(p);
1045	}
1046}
1047
1048static size_t
1049fts_pow2(size_t x)
1050{
1051
1052	x--;
1053	x |= x>>1;
1054	x |= x>>2;
1055	x |= x>>4;
1056	x |= x>>8;
1057	x |= x>>16;
1058#if LONG_BIT > 32
1059	x |= x>>32;
1060#endif
1061#if LONG_BIT > 64
1062	x |= x>>64;
1063#endif
1064	x++;
1065	return (x);
1066}
1067
1068/*
1069 * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
1070 * Most systems will allow creation of paths much longer than MAXPATHLEN, even
1071 * though the kernel won't resolve them.  Round up the new size to a power of 2,
1072 * so we don't realloc the path 2 bytes at a time.
1073 */
1074static int
1075fts_palloc(FTS *sp, size_t size)
1076{
1077	char *new;
1078
1079	_DIAGASSERT(sp != NULL);
1080
1081#if 1
1082	/* Protect against fts_pathlen overflow. */
1083	if (size > USHRT_MAX + 1) {
1084		errno = ENAMETOOLONG;
1085		return (1);
1086	}
1087#endif
1088	size = fts_pow2(size);
1089	new = realloc(sp->fts_path, size);
1090	if (new == 0)
1091		return (1);
1092	sp->fts_path = new;
1093	sp->fts_pathlen = size;
1094	return (0);
1095}
1096
1097/*
1098 * When the path is realloc'd, have to fix all of the pointers in structures
1099 * already returned.
1100 */
1101static void
1102fts_padjust(FTS *sp, FTSENT *head)
1103{
1104	FTSENT *p;
1105	char *addr;
1106
1107	_DIAGASSERT(sp != NULL);
1108
1109#define	ADJUST(p) do {							\
1110	if ((p)->fts_accpath != (p)->fts_name)				\
1111		(p)->fts_accpath =					\
1112		    addr + ((p)->fts_accpath - (p)->fts_path);		\
1113	(p)->fts_path = addr;						\
1114} while (/*CONSTCOND*/0)
1115
1116	addr = sp->fts_path;
1117
1118	/* Adjust the current set of children. */
1119	for (p = sp->fts_child; p; p = p->fts_link)
1120		ADJUST(p);
1121
1122	/* Adjust the rest of the tree, including the current level. */
1123	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1124		ADJUST(p);
1125		p = p->fts_link ? p->fts_link : p->fts_parent;
1126	}
1127}
1128
1129static size_t
1130fts_maxarglen(char * const *argv)
1131{
1132	size_t len, max;
1133
1134	_DIAGASSERT(argv != NULL);
1135
1136	for (max = 0; *argv; ++argv)
1137		if ((len = strlen(*argv)) > max)
1138			max = len;
1139	return (max + 1);
1140}
1141
1142/*
1143 * Change to dir specified by fd or p->fts_accpath without getting
1144 * tricked by someone changing the world out from underneath us.
1145 * Assumes p->fts_dev and p->fts_ino are filled in.
1146 */
1147static int
1148fts_safe_changedir(const FTS *sp, const FTSENT *p, int fd, const char *path)
1149{
1150	int oldfd = fd, ret = -1;
1151	__fts_stat_t sb;
1152
1153	if (ISSET(FTS_NOCHDIR))
1154		return 0;
1155
1156	if (oldfd < 0 && (fd = open(path, O_RDONLY)) == -1)
1157		return -1;
1158
1159	if (fstat(fd, &sb) == -1)
1160		goto bail;
1161
1162	if (sb.st_ino != p->fts_ino || sb.st_dev != p->fts_dev) {
1163		errno = ENOENT;
1164		goto bail;
1165	}
1166
1167	ret = fchdir(fd);
1168
1169bail:
1170	if (oldfd < 0) {
1171		int save_errno = errno;
1172		(void)close(fd);
1173		errno = save_errno;
1174	}
1175	return ret;
1176}
1177