1169695Skan/*-
2169695Skan * Copyright (c) 1990, 1993, 1994
3169695Skan *	The Regents of the University of California.  All rights reserved.
4169695Skan *
5169695Skan * Redistribution and use in source and binary forms, with or without
6169695Skan * modification, are permitted provided that the following conditions
7169695Skan * are met:
8169695Skan * 1. Redistributions of source code must retain the above copyright
9169695Skan *    notice, this list of conditions and the following disclaimer.
10169695Skan * 2. Redistributions in binary form must reproduce the above copyright
11169695Skan *    notice, this list of conditions and the following disclaimer in the
12169695Skan *    documentation and/or other materials provided with the distribution.
13169695Skan * 4. Neither the name of the University nor the names of its contributors
14169695Skan *    may be used to endorse or promote products derived from this software
15169695Skan *    without specific prior written permission.
16169695Skan *
17169695Skan * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18169695Skan * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19169695Skan * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20169695Skan * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21169695Skan * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22169695Skan * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23169695Skan * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24169695Skan * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25169695Skan * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26169695Skan * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27169695Skan * SUCH DAMAGE.
28169695Skan *
29169695Skan * $OpenBSD: fts.c,v 1.22 1999/10/03 19:22:22 millert Exp $
30169695Skan */
31169695Skan
32169695Skan#if 0
33169695Skan#if defined(LIBC_SCCS) && !defined(lint)
34169695Skanstatic char sccsid[] = "@(#)fts.c	8.6 (Berkeley) 8/14/94";
35169695Skan#endif /* LIBC_SCCS and not lint */
36169695Skan#endif
37169695Skan
38169695Skan#include <sys/cdefs.h>
39169695Skan__FBSDID("$FreeBSD$");
40169695Skan
41169695Skan#include "namespace.h"
42169695Skan#include <sys/param.h>
43169695Skan#include <sys/mount.h>
44169695Skan#include <sys/stat.h>
45169695Skan
46169695Skan#include <dirent.h>
47169695Skan#include <errno.h>
48169695Skan#include <fcntl.h>
49169695Skan#include <fts.h>
50169695Skan#include <stdlib.h>
51169695Skan#include <string.h>
52169695Skan#include <unistd.h>
53169695Skan#include "un-namespace.h"
54169695Skan
55169695Skan#include "gen-private.h"
56169695Skan
57169695Skanstatic FTSENT	*fts_alloc(FTS *, char *, size_t);
58169695Skanstatic FTSENT	*fts_build(FTS *, int);
59169695Skanstatic void	 fts_lfree(FTSENT *);
60169695Skanstatic void	 fts_load(FTS *, FTSENT *);
61169695Skanstatic size_t	 fts_maxarglen(char * const *);
62169695Skanstatic void	 fts_padjust(FTS *, FTSENT *);
63169695Skanstatic int	 fts_palloc(FTS *, size_t);
64169695Skanstatic FTSENT	*fts_sort(FTS *, FTSENT *, size_t);
65169695Skanstatic int	 fts_stat(FTS *, FTSENT *, int, int);
66169695Skanstatic int	 fts_safe_changedir(FTS *, FTSENT *, int, char *);
67169695Skanstatic int	 fts_ufslinks(FTS *, const FTSENT *);
68169695Skan
69169695Skan#define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
70169695Skan
71169695Skan#define	CLR(opt)	(sp->fts_options &= ~(opt))
72169695Skan#define	ISSET(opt)	(sp->fts_options & (opt))
73169695Skan#define	SET(opt)	(sp->fts_options |= (opt))
74169695Skan
75169695Skan#define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
76169695Skan
77169695Skan/* fts_build flags */
78169695Skan#define	BCHILD		1		/* fts_children */
79169695Skan#define	BNAMES		2		/* fts_children, names only */
80169695Skan#define	BREAD		3		/* fts_read */
81169695Skan
82169695Skan/*
83169695Skan * Internal representation of an FTS, including extra implementation
84169695Skan * details.  The FTS returned from fts_open points to this structure's
85169695Skan * ftsp_fts member (and can be cast to an _fts_private as required)
86169695Skan */
87169695Skanstruct _fts_private {
88169695Skan	FTS		ftsp_fts;
89169695Skan	struct statfs	ftsp_statfs;
90169695Skan	dev_t		ftsp_dev;
91169695Skan	int		ftsp_linksreliable;
92169695Skan};
93169695Skan
94169695Skan/*
95169695Skan * The "FTS_NOSTAT" option can avoid a lot of calls to stat(2) if it
96169695Skan * knows that a directory could not possibly have subdirectories.  This
97169695Skan * is decided by looking at the link count: a subdirectory would
98169695Skan * increment its parent's link count by virtue of its own ".." entry.
99169695Skan * This assumption only holds for UFS-like filesystems that implement
100169695Skan * links and directories this way, so we must punt for others.
101169695Skan */
102169695Skan
103169695Skanstatic const char *ufslike_filesystems[] = {
104169695Skan	"ufs",
105169695Skan	"zfs",
106169695Skan	"nfs",
107169695Skan	"ext2fs",
108169695Skan	0
109169695Skan};
110169695Skan
111169695SkanFTS *
112169695Skanfts_open(argv, options, compar)
113169695Skan	char * const *argv;
114169695Skan	int options;
115169695Skan	int (*compar)(const FTSENT * const *, const FTSENT * const *);
116169695Skan{
117169695Skan	struct _fts_private *priv;
118169695Skan	FTS *sp;
119169695Skan	FTSENT *p, *root;
120169695Skan	FTSENT *parent, *tmp;
121169695Skan	size_t len, nitems;
122169695Skan
123169695Skan	/* Options check. */
124169695Skan	if (options & ~FTS_OPTIONMASK) {
125169695Skan		errno = EINVAL;
126169695Skan		return (NULL);
127169695Skan	}
128169695Skan
129169695Skan	/* fts_open() requires at least one path */
130169695Skan	if (*argv == NULL) {
131169695Skan		errno = EINVAL;
132169695Skan		return (NULL);
133169695Skan	}
134169695Skan
135169695Skan	/* Allocate/initialize the stream. */
136169695Skan	if ((priv = calloc(1, sizeof(*priv))) == NULL)
137169695Skan		return (NULL);
138169695Skan	sp = &priv->ftsp_fts;
139169695Skan	sp->fts_compar = compar;
140169695Skan	sp->fts_options = options;
141169695Skan
142169695Skan	/* Shush, GCC. */
143169695Skan	tmp = NULL;
144169695Skan
145169695Skan	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
146169695Skan	if (ISSET(FTS_LOGICAL))
147169695Skan		SET(FTS_NOCHDIR);
148169695Skan
149169695Skan	/*
150169695Skan	 * Start out with 1K of path space, and enough, in any case,
151169695Skan	 * to hold the user's paths.
152169695Skan	 */
153169695Skan	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
154169695Skan		goto mem1;
155169695Skan
156169695Skan	/* Allocate/initialize root's parent. */
157169695Skan	if ((parent = fts_alloc(sp, "", 0)) == NULL)
158169695Skan		goto mem2;
159169695Skan	parent->fts_level = FTS_ROOTPARENTLEVEL;
160169695Skan
161169695Skan	/* Allocate/initialize root(s). */
162169695Skan	for (root = NULL, nitems = 0; *argv != NULL; ++argv, ++nitems) {
163169695Skan		len = strlen(*argv);
164169695Skan
165169695Skan		p = fts_alloc(sp, *argv, len);
166169695Skan		p->fts_level = FTS_ROOTLEVEL;
167169695Skan		p->fts_parent = parent;
168169695Skan		p->fts_accpath = p->fts_name;
169169695Skan		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW), -1);
170169695Skan
171169695Skan		/* Command-line "." and ".." are real directories. */
172169695Skan		if (p->fts_info == FTS_DOT)
173169695Skan			p->fts_info = FTS_D;
174169695Skan
175169695Skan		/*
176169695Skan		 * If comparison routine supplied, traverse in sorted
177169695Skan		 * order; otherwise traverse in the order specified.
178169695Skan		 */
179169695Skan		if (compar) {
180169695Skan			p->fts_link = root;
181169695Skan			root = p;
182169695Skan		} else {
183169695Skan			p->fts_link = NULL;
184169695Skan			if (root == NULL)
185169695Skan				tmp = root = p;
186169695Skan			else {
187169695Skan				tmp->fts_link = p;
188169695Skan				tmp = p;
189169695Skan			}
190169695Skan		}
191169695Skan	}
192169695Skan	if (compar && nitems > 1)
193169695Skan		root = fts_sort(sp, root, nitems);
194169695Skan
195169695Skan	/*
196169695Skan	 * Allocate a dummy pointer and make fts_read think that we've just
197169695Skan	 * finished the node before the root(s); set p->fts_info to FTS_INIT
198169695Skan	 * so that everything about the "current" node is ignored.
199169695Skan	 */
200169695Skan	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
201169695Skan		goto mem3;
202169695Skan	sp->fts_cur->fts_link = root;
203169695Skan	sp->fts_cur->fts_info = FTS_INIT;
204169695Skan
205169695Skan	/*
206169695Skan	 * If using chdir(2), grab a file descriptor pointing to dot to ensure
207169695Skan	 * that we can get back here; this could be avoided for some paths,
208169695Skan	 * but almost certainly not worth the effort.  Slashes, symbolic links,
209169695Skan	 * and ".." are all fairly nasty problems.  Note, if we can't get the
210169695Skan	 * descriptor we run anyway, just more slowly.
211169695Skan	 */
212169695Skan	if (!ISSET(FTS_NOCHDIR) &&
213169695Skan	    (sp->fts_rfd = _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0)
214169695Skan		SET(FTS_NOCHDIR);
215169695Skan
216169695Skan	return (sp);
217169695Skan
218169695Skanmem3:	fts_lfree(root);
219169695Skan	free(parent);
220169695Skanmem2:	free(sp->fts_path);
221169695Skanmem1:	free(sp);
222169695Skan	return (NULL);
223169695Skan}
224169695Skan
225169695Skanstatic void
226169695Skanfts_load(FTS *sp, FTSENT *p)
227169695Skan{
228169695Skan	size_t len;
229169695Skan	char *cp;
230169695Skan
231169695Skan	/*
232169695Skan	 * Load the stream structure for the next traversal.  Since we don't
233169695Skan	 * actually enter the directory until after the preorder visit, set
234169695Skan	 * the fts_accpath field specially so the chdir gets done to the right
235169695Skan	 * place and the user can access the first node.  From fts_open it's
236169695Skan	 * known that the path will fit.
237169695Skan	 */
238169695Skan	len = p->fts_pathlen = p->fts_namelen;
239169695Skan	memmove(sp->fts_path, p->fts_name, len + 1);
240169695Skan	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
241169695Skan		len = strlen(++cp);
242169695Skan		memmove(p->fts_name, cp, len + 1);
243169695Skan		p->fts_namelen = len;
244169695Skan	}
245169695Skan	p->fts_accpath = p->fts_path = sp->fts_path;
246169695Skan	sp->fts_dev = p->fts_dev;
247169695Skan}
248169695Skan
249169695Skanint
250169695Skanfts_close(FTS *sp)
251169695Skan{
252169695Skan	FTSENT *freep, *p;
253169695Skan	int saved_errno;
254169695Skan
255169695Skan	/*
256169695Skan	 * This still works if we haven't read anything -- the dummy structure
257169695Skan	 * points to the root list, so we step through to the end of the root
258169695Skan	 * list which has a valid parent pointer.
259169695Skan	 */
260169695Skan	if (sp->fts_cur) {
261169695Skan		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
262169695Skan			freep = p;
263169695Skan			p = p->fts_link != NULL ? p->fts_link : p->fts_parent;
264169695Skan			free(freep);
265169695Skan		}
266169695Skan		free(p);
267169695Skan	}
268169695Skan
269169695Skan	/* Free up child linked list, sort array, path buffer. */
270169695Skan	if (sp->fts_child)
271169695Skan		fts_lfree(sp->fts_child);
272169695Skan	if (sp->fts_array)
273169695Skan		free(sp->fts_array);
274169695Skan	free(sp->fts_path);
275169695Skan
276169695Skan	/* Return to original directory, save errno if necessary. */
277169695Skan	if (!ISSET(FTS_NOCHDIR)) {
278169695Skan		saved_errno = fchdir(sp->fts_rfd) ? errno : 0;
279169695Skan		(void)_close(sp->fts_rfd);
280169695Skan
281169695Skan		/* Set errno and return. */
282169695Skan		if (saved_errno != 0) {
283169695Skan			/* Free up the stream pointer. */
284169695Skan			free(sp);
285169695Skan			errno = saved_errno;
286169695Skan			return (-1);
287169695Skan		}
288169695Skan	}
289169695Skan
290169695Skan	/* Free up the stream pointer. */
291169695Skan	free(sp);
292169695Skan	return (0);
293169695Skan}
294169695Skan
295169695Skan/*
296169695Skan * Special case of "/" at the end of the path so that slashes aren't
297169695Skan * appended which would cause paths to be written as "....//foo".
298169695Skan */
299169695Skan#define	NAPPEND(p)							\
300169695Skan	(p->fts_path[p->fts_pathlen - 1] == '/'				\
301169695Skan	    ? p->fts_pathlen - 1 : p->fts_pathlen)
302169695Skan
303169695SkanFTSENT *
304169695Skanfts_read(FTS *sp)
305169695Skan{
306169695Skan	FTSENT *p, *tmp;
307169695Skan	int instr;
308169695Skan	char *t;
309169695Skan	int saved_errno;
310169695Skan
311169695Skan	/* If finished or unrecoverable error, return NULL. */
312169695Skan	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
313169695Skan		return (NULL);
314169695Skan
315169695Skan	/* Set current node pointer. */
316169695Skan	p = sp->fts_cur;
317169695Skan
318169695Skan	/* Save and zero out user instructions. */
319169695Skan	instr = p->fts_instr;
320169695Skan	p->fts_instr = FTS_NOINSTR;
321169695Skan
322169695Skan	/* Any type of file may be re-visited; re-stat and re-turn. */
323169695Skan	if (instr == FTS_AGAIN) {
324169695Skan		p->fts_info = fts_stat(sp, p, 0, -1);
325169695Skan		return (p);
326169695Skan	}
327169695Skan
328169695Skan	/*
329169695Skan	 * Following a symlink -- SLNONE test allows application to see
330169695Skan	 * SLNONE and recover.  If indirecting through a symlink, have
331169695Skan	 * keep a pointer to current location.  If unable to get that
332169695Skan	 * pointer, follow fails.
333169695Skan	 */
334169695Skan	if (instr == FTS_FOLLOW &&
335169695Skan	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
336169695Skan		p->fts_info = fts_stat(sp, p, 1, -1);
337169695Skan		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
338169695Skan			if ((p->fts_symfd = _open(".", O_RDONLY | O_CLOEXEC,
339169695Skan			    0)) < 0) {
340169695Skan				p->fts_errno = errno;
341				p->fts_info = FTS_ERR;
342			} else
343				p->fts_flags |= FTS_SYMFOLLOW;
344		}
345		return (p);
346	}
347
348	/* Directory in pre-order. */
349	if (p->fts_info == FTS_D) {
350		/* If skipped or crossed mount point, do post-order visit. */
351		if (instr == FTS_SKIP ||
352		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
353			if (p->fts_flags & FTS_SYMFOLLOW)
354				(void)_close(p->fts_symfd);
355			if (sp->fts_child) {
356				fts_lfree(sp->fts_child);
357				sp->fts_child = NULL;
358			}
359			p->fts_info = FTS_DP;
360			return (p);
361		}
362
363		/* Rebuild if only read the names and now traversing. */
364		if (sp->fts_child != NULL && ISSET(FTS_NAMEONLY)) {
365			CLR(FTS_NAMEONLY);
366			fts_lfree(sp->fts_child);
367			sp->fts_child = NULL;
368		}
369
370		/*
371		 * Cd to the subdirectory.
372		 *
373		 * If have already read and now fail to chdir, whack the list
374		 * to make the names come out right, and set the parent errno
375		 * so the application will eventually get an error condition.
376		 * Set the FTS_DONTCHDIR flag so that when we logically change
377		 * directories back to the parent we don't do a chdir.
378		 *
379		 * If haven't read do so.  If the read fails, fts_build sets
380		 * FTS_STOP or the fts_info field of the node.
381		 */
382		if (sp->fts_child != NULL) {
383			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
384				p->fts_errno = errno;
385				p->fts_flags |= FTS_DONTCHDIR;
386				for (p = sp->fts_child; p != NULL;
387				    p = p->fts_link)
388					p->fts_accpath =
389					    p->fts_parent->fts_accpath;
390			}
391		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
392			if (ISSET(FTS_STOP))
393				return (NULL);
394			return (p);
395		}
396		p = sp->fts_child;
397		sp->fts_child = NULL;
398		goto name;
399	}
400
401	/* Move to the next node on this level. */
402next:	tmp = p;
403	if ((p = p->fts_link) != NULL) {
404		/*
405		 * If reached the top, return to the original directory (or
406		 * the root of the tree), and load the paths for the next root.
407		 */
408		if (p->fts_level == FTS_ROOTLEVEL) {
409			if (FCHDIR(sp, sp->fts_rfd)) {
410				SET(FTS_STOP);
411				return (NULL);
412			}
413			free(tmp);
414			fts_load(sp, p);
415			return (sp->fts_cur = p);
416		}
417
418		/*
419		 * User may have called fts_set on the node.  If skipped,
420		 * ignore.  If followed, get a file descriptor so we can
421		 * get back if necessary.
422		 */
423		if (p->fts_instr == FTS_SKIP) {
424			free(tmp);
425			goto next;
426		}
427		if (p->fts_instr == FTS_FOLLOW) {
428			p->fts_info = fts_stat(sp, p, 1, -1);
429			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
430				if ((p->fts_symfd =
431				    _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0) {
432					p->fts_errno = errno;
433					p->fts_info = FTS_ERR;
434				} else
435					p->fts_flags |= FTS_SYMFOLLOW;
436			}
437			p->fts_instr = FTS_NOINSTR;
438		}
439
440		free(tmp);
441
442name:		t = sp->fts_path + NAPPEND(p->fts_parent);
443		*t++ = '/';
444		memmove(t, p->fts_name, p->fts_namelen + 1);
445		return (sp->fts_cur = p);
446	}
447
448	/* Move up to the parent node. */
449	p = tmp->fts_parent;
450
451	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
452		/*
453		 * Done; free everything up and set errno to 0 so the user
454		 * can distinguish between error and EOF.
455		 */
456		free(tmp);
457		free(p);
458		errno = 0;
459		return (sp->fts_cur = NULL);
460	}
461
462	/* NUL terminate the pathname. */
463	sp->fts_path[p->fts_pathlen] = '\0';
464
465	/*
466	 * Return to the parent directory.  If at a root node or came through
467	 * a symlink, go back through the file descriptor.  Otherwise, cd up
468	 * one directory.
469	 */
470	if (p->fts_level == FTS_ROOTLEVEL) {
471		if (FCHDIR(sp, sp->fts_rfd)) {
472			SET(FTS_STOP);
473			return (NULL);
474		}
475	} else if (p->fts_flags & FTS_SYMFOLLOW) {
476		if (FCHDIR(sp, p->fts_symfd)) {
477			saved_errno = errno;
478			(void)_close(p->fts_symfd);
479			errno = saved_errno;
480			SET(FTS_STOP);
481			return (NULL);
482		}
483		(void)_close(p->fts_symfd);
484	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
485	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
486		SET(FTS_STOP);
487		return (NULL);
488	}
489	free(tmp);
490	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
491	return (sp->fts_cur = p);
492}
493
494/*
495 * Fts_set takes the stream as an argument although it's not used in this
496 * implementation; it would be necessary if anyone wanted to add global
497 * semantics to fts using fts_set.  An error return is allowed for similar
498 * reasons.
499 */
500/* ARGSUSED */
501int
502fts_set(FTS *sp, FTSENT *p, int instr)
503{
504	if (instr != 0 && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
505	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
506		errno = EINVAL;
507		return (1);
508	}
509	p->fts_instr = instr;
510	return (0);
511}
512
513FTSENT *
514fts_children(FTS *sp, int instr)
515{
516	FTSENT *p;
517	int fd, rc, serrno;
518
519	if (instr != 0 && instr != FTS_NAMEONLY) {
520		errno = EINVAL;
521		return (NULL);
522	}
523
524	/* Set current node pointer. */
525	p = sp->fts_cur;
526
527	/*
528	 * Errno set to 0 so user can distinguish empty directory from
529	 * an error.
530	 */
531	errno = 0;
532
533	/* Fatal errors stop here. */
534	if (ISSET(FTS_STOP))
535		return (NULL);
536
537	/* Return logical hierarchy of user's arguments. */
538	if (p->fts_info == FTS_INIT)
539		return (p->fts_link);
540
541	/*
542	 * If not a directory being visited in pre-order, stop here.  Could
543	 * allow FTS_DNR, assuming the user has fixed the problem, but the
544	 * same effect is available with FTS_AGAIN.
545	 */
546	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
547		return (NULL);
548
549	/* Free up any previous child list. */
550	if (sp->fts_child != NULL)
551		fts_lfree(sp->fts_child);
552
553	if (instr == FTS_NAMEONLY) {
554		SET(FTS_NAMEONLY);
555		instr = BNAMES;
556	} else
557		instr = BCHILD;
558
559	/*
560	 * If using chdir on a relative path and called BEFORE fts_read does
561	 * its chdir to the root of a traversal, we can lose -- we need to
562	 * chdir into the subdirectory, and we don't know where the current
563	 * directory is, so we can't get back so that the upcoming chdir by
564	 * fts_read will work.
565	 */
566	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
567	    ISSET(FTS_NOCHDIR))
568		return (sp->fts_child = fts_build(sp, instr));
569
570	if ((fd = _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0)
571		return (NULL);
572	sp->fts_child = fts_build(sp, instr);
573	serrno = (sp->fts_child == NULL) ? errno : 0;
574	rc = fchdir(fd);
575	if (rc < 0 && serrno == 0)
576		serrno = errno;
577	(void)_close(fd);
578	errno = serrno;
579	if (rc < 0)
580		return (NULL);
581	return (sp->fts_child);
582}
583
584#ifndef fts_get_clientptr
585#error "fts_get_clientptr not defined"
586#endif
587
588void *
589(fts_get_clientptr)(FTS *sp)
590{
591
592	return (fts_get_clientptr(sp));
593}
594
595#ifndef fts_get_stream
596#error "fts_get_stream not defined"
597#endif
598
599FTS *
600(fts_get_stream)(FTSENT *p)
601{
602	return (fts_get_stream(p));
603}
604
605void
606fts_set_clientptr(FTS *sp, void *clientptr)
607{
608
609	sp->fts_clientptr = clientptr;
610}
611
612/*
613 * This is the tricky part -- do not casually change *anything* in here.  The
614 * idea is to build the linked list of entries that are used by fts_children
615 * and fts_read.  There are lots of special cases.
616 *
617 * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
618 * set and it's a physical walk (so that symbolic links can't be directories),
619 * we can do things quickly.  First, if it's a 4.4BSD file system, the type
620 * of the file is in the directory entry.  Otherwise, we assume that the number
621 * of subdirectories in a node is equal to the number of links to the parent.
622 * The former skips all stat calls.  The latter skips stat calls in any leaf
623 * directories and for any files after the subdirectories in the directory have
624 * been found, cutting the stat calls by about 2/3.
625 */
626static FTSENT *
627fts_build(FTS *sp, int type)
628{
629	struct dirent *dp;
630	FTSENT *p, *head;
631	FTSENT *cur, *tail;
632	DIR *dirp;
633	void *oldaddr;
634	char *cp;
635	int cderrno, descend, oflag, saved_errno, nostat, doadjust;
636	long level;
637	long nlinks;	/* has to be signed because -1 is a magic value */
638	size_t dnamlen, len, maxlen, nitems;
639
640	/* Set current node pointer. */
641	cur = sp->fts_cur;
642
643	/*
644	 * Open the directory for reading.  If this fails, we're done.
645	 * If being called from fts_read, set the fts_info field.
646	 */
647#ifdef FTS_WHITEOUT
648	if (ISSET(FTS_WHITEOUT))
649		oflag = DTF_NODUP | DTF_REWIND;
650	else
651		oflag = DTF_HIDEW | DTF_NODUP | DTF_REWIND;
652#else
653#define __opendir2(path, flag) opendir(path)
654#endif
655	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
656		if (type == BREAD) {
657			cur->fts_info = FTS_DNR;
658			cur->fts_errno = errno;
659		}
660		return (NULL);
661	}
662
663	/*
664	 * Nlinks is the number of possible entries of type directory in the
665	 * directory if we're cheating on stat calls, 0 if we're not doing
666	 * any stat calls at all, -1 if we're doing stats on everything.
667	 */
668	if (type == BNAMES) {
669		nlinks = 0;
670		/* Be quiet about nostat, GCC. */
671		nostat = 0;
672	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
673		if (fts_ufslinks(sp, cur))
674			nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
675		else
676			nlinks = -1;
677		nostat = 1;
678	} else {
679		nlinks = -1;
680		nostat = 0;
681	}
682
683#ifdef notdef
684	(void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
685	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
686	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
687#endif
688	/*
689	 * If we're going to need to stat anything or we want to descend
690	 * and stay in the directory, chdir.  If this fails we keep going,
691	 * but set a flag so we don't chdir after the post-order visit.
692	 * We won't be able to stat anything, but we can still return the
693	 * names themselves.  Note, that since fts_read won't be able to
694	 * chdir into the directory, it will have to return different path
695	 * names than before, i.e. "a/b" instead of "b".  Since the node
696	 * has already been visited in pre-order, have to wait until the
697	 * post-order visit to return the error.  There is a special case
698	 * here, if there was nothing to stat then it's not an error to
699	 * not be able to stat.  This is all fairly nasty.  If a program
700	 * needed sorted entries or stat information, they had better be
701	 * checking FTS_NS on the returned nodes.
702	 */
703	cderrno = 0;
704	if (nlinks || type == BREAD) {
705		if (fts_safe_changedir(sp, cur, _dirfd(dirp), NULL)) {
706			if (nlinks && type == BREAD)
707				cur->fts_errno = errno;
708			cur->fts_flags |= FTS_DONTCHDIR;
709			descend = 0;
710			cderrno = errno;
711		} else
712			descend = 1;
713	} else
714		descend = 0;
715
716	/*
717	 * Figure out the max file name length that can be stored in the
718	 * current path -- the inner loop allocates more path as necessary.
719	 * We really wouldn't have to do the maxlen calculations here, we
720	 * could do them in fts_read before returning the path, but it's a
721	 * lot easier here since the length is part of the dirent structure.
722	 *
723	 * If not changing directories set a pointer so that can just append
724	 * each new name into the path.
725	 */
726	len = NAPPEND(cur);
727	if (ISSET(FTS_NOCHDIR)) {
728		cp = sp->fts_path + len;
729		*cp++ = '/';
730	} else {
731		/* GCC, you're too verbose. */
732		cp = NULL;
733	}
734	len++;
735	maxlen = sp->fts_pathlen - len;
736
737	level = cur->fts_level + 1;
738
739	/* Read the directory, attaching each entry to the `link' pointer. */
740	doadjust = 0;
741	for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) {
742		dnamlen = dp->d_namlen;
743		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
744			continue;
745
746		if ((p = fts_alloc(sp, dp->d_name, dnamlen)) == NULL)
747			goto mem1;
748		if (dnamlen >= maxlen) {	/* include space for NUL */
749			oldaddr = sp->fts_path;
750			if (fts_palloc(sp, dnamlen + len + 1)) {
751				/*
752				 * No more memory for path or structures.  Save
753				 * errno, free up the current structure and the
754				 * structures already allocated.
755				 */
756mem1:				saved_errno = errno;
757				if (p)
758					free(p);
759				fts_lfree(head);
760				(void)closedir(dirp);
761				cur->fts_info = FTS_ERR;
762				SET(FTS_STOP);
763				errno = saved_errno;
764				return (NULL);
765			}
766			/* Did realloc() change the pointer? */
767			if (oldaddr != sp->fts_path) {
768				doadjust = 1;
769				if (ISSET(FTS_NOCHDIR))
770					cp = sp->fts_path + len;
771			}
772			maxlen = sp->fts_pathlen - len;
773		}
774
775		p->fts_level = level;
776		p->fts_parent = sp->fts_cur;
777		p->fts_pathlen = len + dnamlen;
778
779#ifdef FTS_WHITEOUT
780		if (dp->d_type == DT_WHT)
781			p->fts_flags |= FTS_ISW;
782#endif
783
784		if (cderrno) {
785			if (nlinks) {
786				p->fts_info = FTS_NS;
787				p->fts_errno = cderrno;
788			} else
789				p->fts_info = FTS_NSOK;
790			p->fts_accpath = cur->fts_accpath;
791		} else if (nlinks == 0
792#ifdef DT_DIR
793		    || (nostat &&
794		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
795#endif
796		    ) {
797			p->fts_accpath =
798			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
799			p->fts_info = FTS_NSOK;
800		} else {
801			/* Build a file name for fts_stat to stat. */
802			if (ISSET(FTS_NOCHDIR)) {
803				p->fts_accpath = p->fts_path;
804				memmove(cp, p->fts_name, p->fts_namelen + 1);
805				p->fts_info = fts_stat(sp, p, 0, _dirfd(dirp));
806			} else {
807				p->fts_accpath = p->fts_name;
808				p->fts_info = fts_stat(sp, p, 0, -1);
809			}
810
811			/* Decrement link count if applicable. */
812			if (nlinks > 0 && (p->fts_info == FTS_D ||
813			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
814				--nlinks;
815		}
816
817		/* We walk in directory order so "ls -f" doesn't get upset. */
818		p->fts_link = NULL;
819		if (head == NULL)
820			head = tail = p;
821		else {
822			tail->fts_link = p;
823			tail = p;
824		}
825		++nitems;
826	}
827	if (dirp)
828		(void)closedir(dirp);
829
830	/*
831	 * If realloc() changed the address of the path, adjust the
832	 * addresses for the rest of the tree and the dir list.
833	 */
834	if (doadjust)
835		fts_padjust(sp, head);
836
837	/*
838	 * If not changing directories, reset the path back to original
839	 * state.
840	 */
841	if (ISSET(FTS_NOCHDIR))
842		sp->fts_path[cur->fts_pathlen] = '\0';
843
844	/*
845	 * If descended after called from fts_children or after called from
846	 * fts_read and nothing found, get back.  At the root level we use
847	 * the saved fd; if one of fts_open()'s arguments is a relative path
848	 * to an empty directory, we wind up here with no other way back.  If
849	 * can't get back, we're done.
850	 */
851	if (descend && (type == BCHILD || !nitems) &&
852	    (cur->fts_level == FTS_ROOTLEVEL ?
853	    FCHDIR(sp, sp->fts_rfd) :
854	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
855		cur->fts_info = FTS_ERR;
856		SET(FTS_STOP);
857		return (NULL);
858	}
859
860	/* If didn't find anything, return NULL. */
861	if (!nitems) {
862		if (type == BREAD)
863			cur->fts_info = FTS_DP;
864		return (NULL);
865	}
866
867	/* Sort the entries. */
868	if (sp->fts_compar && nitems > 1)
869		head = fts_sort(sp, head, nitems);
870	return (head);
871}
872
873static int
874fts_stat(FTS *sp, FTSENT *p, int follow, int dfd)
875{
876	FTSENT *t;
877	dev_t dev;
878	ino_t ino;
879	struct stat *sbp, sb;
880	int saved_errno;
881	const char *path;
882
883	if (dfd == -1)
884		path = p->fts_accpath, dfd = AT_FDCWD;
885	else
886		path = p->fts_name;
887
888	/* If user needs stat info, stat buffer already allocated. */
889	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
890
891#ifdef FTS_WHITEOUT
892	/* Check for whiteout. */
893	if (p->fts_flags & FTS_ISW) {
894		if (sbp != &sb) {
895			memset(sbp, '\0', sizeof(*sbp));
896			sbp->st_mode = S_IFWHT;
897		}
898		return (FTS_W);
899	}
900#endif
901
902	/*
903	 * If doing a logical walk, or application requested FTS_FOLLOW, do
904	 * a stat(2).  If that fails, check for a non-existent symlink.  If
905	 * fail, set the errno from the stat call.
906	 */
907	if (ISSET(FTS_LOGICAL) || follow) {
908		if (fstatat(dfd, path, sbp, 0)) {
909			saved_errno = errno;
910			if (fstatat(dfd, path, sbp, AT_SYMLINK_NOFOLLOW)) {
911				p->fts_errno = saved_errno;
912				goto err;
913			}
914			errno = 0;
915			if (S_ISLNK(sbp->st_mode))
916				return (FTS_SLNONE);
917		}
918	} else if (fstatat(dfd, path, sbp, AT_SYMLINK_NOFOLLOW)) {
919		p->fts_errno = errno;
920err:		memset(sbp, 0, sizeof(struct stat));
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
960/*
961 * The comparison function takes pointers to pointers to FTSENT structures.
962 * Qsort wants a comparison function that takes pointers to void.
963 * (Both with appropriate levels of const-poisoning, of course!)
964 * Use a trampoline function to deal with the difference.
965 */
966static int
967fts_compar(const void *a, const void *b)
968{
969	FTS *parent;
970
971	parent = (*(const FTSENT * const *)a)->fts_fts;
972	return (*parent->fts_compar)(a, b);
973}
974
975static FTSENT *
976fts_sort(FTS *sp, FTSENT *head, size_t nitems)
977{
978	FTSENT **ap, *p;
979
980	/*
981	 * Construct an array of pointers to the structures and call qsort(3).
982	 * Reassemble the array in the order returned by qsort.  If unable to
983	 * sort for memory reasons, return the directory entries in their
984	 * current order.  Allocate enough space for the current needs plus
985	 * 40 so don't realloc one entry at a time.
986	 */
987	if (nitems > sp->fts_nitems) {
988		sp->fts_nitems = nitems + 40;
989		if ((sp->fts_array = reallocf(sp->fts_array,
990		    sp->fts_nitems * sizeof(FTSENT *))) == NULL) {
991			sp->fts_nitems = 0;
992			return (head);
993		}
994	}
995	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
996		*ap++ = p;
997	qsort(sp->fts_array, nitems, sizeof(FTSENT *), fts_compar);
998	for (head = *(ap = sp->fts_array); --nitems; ++ap)
999		ap[0]->fts_link = ap[1];
1000	ap[0]->fts_link = NULL;
1001	return (head);
1002}
1003
1004static FTSENT *
1005fts_alloc(FTS *sp, char *name, size_t namelen)
1006{
1007	FTSENT *p;
1008	size_t len;
1009
1010	struct ftsent_withstat {
1011		FTSENT	ent;
1012		struct	stat statbuf;
1013	};
1014
1015	/*
1016	 * The file name is a variable length array and no stat structure is
1017	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
1018	 * structure, the file name and the stat structure in one chunk, but
1019	 * be careful that the stat structure is reasonably aligned.
1020	 */
1021	if (ISSET(FTS_NOSTAT))
1022		len = sizeof(FTSENT) + namelen + 1;
1023	else
1024		len = sizeof(struct ftsent_withstat) + namelen + 1;
1025
1026	if ((p = malloc(len)) == NULL)
1027		return (NULL);
1028
1029	if (ISSET(FTS_NOSTAT)) {
1030		p->fts_name = (char *)(p + 1);
1031		p->fts_statp = NULL;
1032	} else {
1033		p->fts_name = (char *)((struct ftsent_withstat *)p + 1);
1034		p->fts_statp = &((struct ftsent_withstat *)p)->statbuf;
1035	}
1036
1037	/* Copy the name and guarantee NUL termination. */
1038	memcpy(p->fts_name, name, namelen);
1039	p->fts_name[namelen] = '\0';
1040	p->fts_namelen = namelen;
1041	p->fts_path = sp->fts_path;
1042	p->fts_errno = 0;
1043	p->fts_flags = 0;
1044	p->fts_instr = FTS_NOINSTR;
1045	p->fts_number = 0;
1046	p->fts_pointer = NULL;
1047	p->fts_fts = sp;
1048	return (p);
1049}
1050
1051static void
1052fts_lfree(FTSENT *head)
1053{
1054	FTSENT *p;
1055
1056	/* Free a linked list of structures. */
1057	while ((p = head)) {
1058		head = head->fts_link;
1059		free(p);
1060	}
1061}
1062
1063/*
1064 * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
1065 * Most systems will allow creation of paths much longer than MAXPATHLEN, even
1066 * though the kernel won't resolve them.  Add the size (not just what's needed)
1067 * plus 256 bytes so don't realloc the path 2 bytes at a time.
1068 */
1069static int
1070fts_palloc(FTS *sp, size_t more)
1071{
1072
1073	sp->fts_pathlen += more + 256;
1074	sp->fts_path = reallocf(sp->fts_path, sp->fts_pathlen);
1075	return (sp->fts_path == NULL);
1076}
1077
1078/*
1079 * When the path is realloc'd, have to fix all of the pointers in structures
1080 * already returned.
1081 */
1082static void
1083fts_padjust(FTS *sp, FTSENT *head)
1084{
1085	FTSENT *p;
1086	char *addr = sp->fts_path;
1087
1088#define	ADJUST(p) do {							\
1089	if ((p)->fts_accpath != (p)->fts_name) {			\
1090		(p)->fts_accpath =					\
1091		    (char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
1092	}								\
1093	(p)->fts_path = addr;						\
1094} while (0)
1095	/* Adjust the current set of children. */
1096	for (p = sp->fts_child; p; p = p->fts_link)
1097		ADJUST(p);
1098
1099	/* Adjust the rest of the tree, including the current level. */
1100	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1101		ADJUST(p);
1102		p = p->fts_link ? p->fts_link : p->fts_parent;
1103	}
1104}
1105
1106static size_t
1107fts_maxarglen(argv)
1108	char * const *argv;
1109{
1110	size_t len, max;
1111
1112	for (max = 0; *argv; ++argv)
1113		if ((len = strlen(*argv)) > max)
1114			max = len;
1115	return (max + 1);
1116}
1117
1118/*
1119 * Change to dir specified by fd or p->fts_accpath without getting
1120 * tricked by someone changing the world out from underneath us.
1121 * Assumes p->fts_dev and p->fts_ino are filled in.
1122 */
1123static int
1124fts_safe_changedir(FTS *sp, FTSENT *p, int fd, char *path)
1125{
1126	int ret, oerrno, newfd;
1127	struct stat sb;
1128
1129	newfd = fd;
1130	if (ISSET(FTS_NOCHDIR))
1131		return (0);
1132	if (fd < 0 && (newfd = _open(path, O_RDONLY | O_DIRECTORY |
1133	    O_CLOEXEC, 0)) < 0)
1134		return (-1);
1135	if (_fstat(newfd, &sb)) {
1136		ret = -1;
1137		goto bail;
1138	}
1139	if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) {
1140		errno = ENOENT;		/* disinformation */
1141		ret = -1;
1142		goto bail;
1143	}
1144	ret = fchdir(newfd);
1145bail:
1146	oerrno = errno;
1147	if (fd < 0)
1148		(void)_close(newfd);
1149	errno = oerrno;
1150	return (ret);
1151}
1152
1153/*
1154 * Check if the filesystem for "ent" has UFS-style links.
1155 */
1156static int
1157fts_ufslinks(FTS *sp, const FTSENT *ent)
1158{
1159	struct _fts_private *priv;
1160	const char **cpp;
1161
1162	priv = (struct _fts_private *)sp;
1163	/*
1164	 * If this node's device is different from the previous, grab
1165	 * the filesystem information, and decide on the reliability
1166	 * of the link information from this filesystem for stat(2)
1167	 * avoidance.
1168	 */
1169	if (priv->ftsp_dev != ent->fts_dev) {
1170		if (statfs(ent->fts_path, &priv->ftsp_statfs) != -1) {
1171			priv->ftsp_dev = ent->fts_dev;
1172			priv->ftsp_linksreliable = 0;
1173			for (cpp = ufslike_filesystems; *cpp; cpp++) {
1174				if (strcmp(priv->ftsp_statfs.f_fstypename,
1175				    *cpp) == 0) {
1176					priv->ftsp_linksreliable = 1;
1177					break;
1178				}
1179			}
1180		} else {
1181			priv->ftsp_linksreliable = 0;
1182		}
1183	}
1184	return (priv->ftsp_linksreliable);
1185}
1186