1/*-
2 * Copyright (c) 1990, 1993, 1994
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 *    may be used to endorse or promote products derived from this software
15 *    without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * $OpenBSD: fts.c,v 1.22 1999/10/03 19:22:22 millert Exp $
30 */
31
32#if 0
33#if defined(LIBC_SCCS) && !defined(lint)
34static char sccsid[] = "@(#)fts.c	8.6 (Berkeley) 8/14/94";
35#endif /* LIBC_SCCS and not lint */
36#endif
37
38#include <sys/cdefs.h>
39__FBSDID("$FreeBSD$");
40
41#include "namespace.h"
42#include <sys/param.h>
43#include <sys/mount.h>
44#include <sys/stat.h>
45
46#include <dirent.h>
47#include <errno.h>
48#include <fcntl.h>
49#include <stdlib.h>
50#include <string.h>
51#include <unistd.h>
52#include "fts-compat.h"
53#include "un-namespace.h"
54
55FTSENT	*__fts_children_44bsd(FTS *, int);
56int	 __fts_close_44bsd(FTS *);
57void	*__fts_get_clientptr_44bsd(FTS *);
58FTS	*__fts_get_stream_44bsd(FTSENT *);
59FTS	*__fts_open_44bsd(char * const *, int,
60	    int (*)(const FTSENT * const *, const FTSENT * const *));
61FTSENT	*__fts_read_44bsd(FTS *);
62int	 __fts_set_44bsd(FTS *, FTSENT *, int);
63void	 __fts_set_clientptr_44bsd(FTS *, void *);
64
65static FTSENT	*fts_alloc(FTS *, char *, int);
66static FTSENT	*fts_build(FTS *, int);
67static void	 fts_lfree(FTSENT *);
68static void	 fts_load(FTS *, FTSENT *);
69static size_t	 fts_maxarglen(char * const *);
70static void	 fts_padjust(FTS *, FTSENT *);
71static int	 fts_palloc(FTS *, size_t);
72static FTSENT	*fts_sort(FTS *, FTSENT *, int);
73static u_short	 fts_stat(FTS *, FTSENT *, int);
74static int	 fts_safe_changedir(FTS *, FTSENT *, int, char *);
75static int	 fts_ufslinks(FTS *, const FTSENT *);
76
77#define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
78
79#define	CLR(opt)	(sp->fts_options &= ~(opt))
80#define	ISSET(opt)	(sp->fts_options & (opt))
81#define	SET(opt)	(sp->fts_options |= (opt))
82
83#define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
84
85/* fts_build flags */
86#define	BCHILD		1		/* fts_children */
87#define	BNAMES		2		/* fts_children, names only */
88#define	BREAD		3		/* fts_read */
89
90/*
91 * Internal representation of an FTS, including extra implementation
92 * details.  The FTS returned from fts_open points to this structure's
93 * ftsp_fts member (and can be cast to an _fts_private as required)
94 */
95struct _fts_private {
96	FTS		ftsp_fts;
97	struct statfs	ftsp_statfs;
98	dev_t		ftsp_dev;
99	int		ftsp_linksreliable;
100};
101
102/*
103 * The "FTS_NOSTAT" option can avoid a lot of calls to stat(2) if it
104 * knows that a directory could not possibly have subdirectories.  This
105 * is decided by looking at the link count: a subdirectory would
106 * increment its parent's link count by virtue of its own ".." entry.
107 * This assumption only holds for UFS-like filesystems that implement
108 * links and directories this way, so we must punt for others.
109 */
110
111static const char *ufslike_filesystems[] = {
112	"ufs",
113	"zfs",
114	"nfs",
115	"nfs4",
116	"ext2fs",
117	0
118};
119
120FTS *
121__fts_open_44bsd(argv, options, compar)
122	char * const *argv;
123	int options;
124	int (*compar)(const FTSENT * const *, const FTSENT * const *);
125{
126	struct _fts_private *priv;
127	FTS *sp;
128	FTSENT *p, *root;
129	int nitems;
130	FTSENT *parent, *tmp;
131	int len;
132
133	/* Options check. */
134	if (options & ~FTS_OPTIONMASK) {
135		errno = EINVAL;
136		return (NULL);
137	}
138
139	/* Allocate/initialize the stream. */
140	if ((priv = malloc(sizeof(*priv))) == NULL)
141		return (NULL);
142	memset(priv, 0, sizeof(*priv));
143	sp = &priv->ftsp_fts;
144	sp->fts_compar = compar;
145	sp->fts_options = options;
146
147	/* Shush, GCC. */
148	tmp = NULL;
149
150	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
151	if (ISSET(FTS_LOGICAL))
152		SET(FTS_NOCHDIR);
153
154	/*
155	 * Start out with 1K of path space, and enough, in any case,
156	 * to hold the user's paths.
157	 */
158	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
159		goto mem1;
160
161	/* Allocate/initialize root's parent. */
162	if ((parent = fts_alloc(sp, "", 0)) == NULL)
163		goto mem2;
164	parent->fts_level = FTS_ROOTPARENTLEVEL;
165
166	/* Allocate/initialize root(s). */
167	for (root = NULL, nitems = 0; *argv != NULL; ++argv, ++nitems) {
168		/* Don't allow zero-length paths. */
169		if ((len = strlen(*argv)) == 0) {
170			errno = ENOENT;
171			goto mem3;
172		}
173
174		p = fts_alloc(sp, *argv, len);
175		p->fts_level = FTS_ROOTLEVEL;
176		p->fts_parent = parent;
177		p->fts_accpath = p->fts_name;
178		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW));
179
180		/* Command-line "." and ".." are real directories. */
181		if (p->fts_info == FTS_DOT)
182			p->fts_info = FTS_D;
183
184		/*
185		 * If comparison routine supplied, traverse in sorted
186		 * order; otherwise traverse in the order specified.
187		 */
188		if (compar) {
189			p->fts_link = root;
190			root = p;
191		} else {
192			p->fts_link = NULL;
193			if (root == NULL)
194				tmp = root = p;
195			else {
196				tmp->fts_link = p;
197				tmp = p;
198			}
199		}
200	}
201	if (compar && nitems > 1)
202		root = fts_sort(sp, root, nitems);
203
204	/*
205	 * Allocate a dummy pointer and make fts_read think that we've just
206	 * finished the node before the root(s); set p->fts_info to FTS_INIT
207	 * so that everything about the "current" node is ignored.
208	 */
209	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
210		goto mem3;
211	sp->fts_cur->fts_link = root;
212	sp->fts_cur->fts_info = FTS_INIT;
213
214	/*
215	 * If using chdir(2), grab a file descriptor pointing to dot to ensure
216	 * that we can get back here; this could be avoided for some paths,
217	 * but almost certainly not worth the effort.  Slashes, symbolic links,
218	 * and ".." are all fairly nasty problems.  Note, if we can't get the
219	 * descriptor we run anyway, just more slowly.
220	 */
221	if (!ISSET(FTS_NOCHDIR) &&
222	    (sp->fts_rfd = _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0)
223		SET(FTS_NOCHDIR);
224
225	return (sp);
226
227mem3:	fts_lfree(root);
228	free(parent);
229mem2:	free(sp->fts_path);
230mem1:	free(sp);
231	return (NULL);
232}
233
234static void
235fts_load(sp, p)
236	FTS *sp;
237	FTSENT *p;
238{
239	int len;
240	char *cp;
241
242	/*
243	 * Load the stream structure for the next traversal.  Since we don't
244	 * actually enter the directory until after the preorder visit, set
245	 * the fts_accpath field specially so the chdir gets done to the right
246	 * place and the user can access the first node.  From fts_open it's
247	 * known that the path will fit.
248	 */
249	len = p->fts_pathlen = p->fts_namelen;
250	memmove(sp->fts_path, p->fts_name, len + 1);
251	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
252		len = strlen(++cp);
253		memmove(p->fts_name, cp, len + 1);
254		p->fts_namelen = len;
255	}
256	p->fts_accpath = p->fts_path = sp->fts_path;
257	sp->fts_dev = p->fts_dev;
258}
259
260int
261__fts_close_44bsd(sp)
262	FTS *sp;
263{
264	FTSENT *freep, *p;
265	int saved_errno;
266
267	/*
268	 * This still works if we haven't read anything -- the dummy structure
269	 * points to the root list, so we step through to the end of the root
270	 * list which has a valid parent pointer.
271	 */
272	if (sp->fts_cur) {
273		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
274			freep = p;
275			p = p->fts_link != NULL ? p->fts_link : p->fts_parent;
276			free(freep);
277		}
278		free(p);
279	}
280
281	/* Free up child linked list, sort array, path buffer. */
282	if (sp->fts_child)
283		fts_lfree(sp->fts_child);
284	if (sp->fts_array)
285		free(sp->fts_array);
286	free(sp->fts_path);
287
288	/* Return to original directory, save errno if necessary. */
289	if (!ISSET(FTS_NOCHDIR)) {
290		saved_errno = fchdir(sp->fts_rfd) ? errno : 0;
291		(void)_close(sp->fts_rfd);
292
293		/* Set errno and return. */
294		if (saved_errno != 0) {
295			/* Free up the stream pointer. */
296			free(sp);
297			errno = saved_errno;
298			return (-1);
299		}
300	}
301
302	/* Free up the stream pointer. */
303	free(sp);
304	return (0);
305}
306
307/*
308 * Special case of "/" at the end of the path so that slashes aren't
309 * appended which would cause paths to be written as "....//foo".
310 */
311#define	NAPPEND(p)							\
312	(p->fts_path[p->fts_pathlen - 1] == '/'				\
313	    ? p->fts_pathlen - 1 : p->fts_pathlen)
314
315FTSENT *
316__fts_read_44bsd(sp)
317	FTS *sp;
318{
319	FTSENT *p, *tmp;
320	int instr;
321	char *t;
322	int saved_errno;
323
324	/* If finished or unrecoverable error, return NULL. */
325	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
326		return (NULL);
327
328	/* Set current node pointer. */
329	p = sp->fts_cur;
330
331	/* Save and zero out user instructions. */
332	instr = p->fts_instr;
333	p->fts_instr = FTS_NOINSTR;
334
335	/* Any type of file may be re-visited; re-stat and re-turn. */
336	if (instr == FTS_AGAIN) {
337		p->fts_info = fts_stat(sp, p, 0);
338		return (p);
339	}
340
341	/*
342	 * Following a symlink -- SLNONE test allows application to see
343	 * SLNONE and recover.  If indirecting through a symlink, have
344	 * keep a pointer to current location.  If unable to get that
345	 * pointer, follow fails.
346	 */
347	if (instr == FTS_FOLLOW &&
348	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
349		p->fts_info = fts_stat(sp, p, 1);
350		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
351			if ((p->fts_symfd = _open(".", O_RDONLY | O_CLOEXEC,
352			    0)) < 0) {
353				p->fts_errno = errno;
354				p->fts_info = FTS_ERR;
355			} else
356				p->fts_flags |= FTS_SYMFOLLOW;
357		}
358		return (p);
359	}
360
361	/* Directory in pre-order. */
362	if (p->fts_info == FTS_D) {
363		/* If skipped or crossed mount point, do post-order visit. */
364		if (instr == FTS_SKIP ||
365		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
366			if (p->fts_flags & FTS_SYMFOLLOW)
367				(void)_close(p->fts_symfd);
368			if (sp->fts_child) {
369				fts_lfree(sp->fts_child);
370				sp->fts_child = NULL;
371			}
372			p->fts_info = FTS_DP;
373			return (p);
374		}
375
376		/* Rebuild if only read the names and now traversing. */
377		if (sp->fts_child != NULL && ISSET(FTS_NAMEONLY)) {
378			CLR(FTS_NAMEONLY);
379			fts_lfree(sp->fts_child);
380			sp->fts_child = NULL;
381		}
382
383		/*
384		 * Cd to the subdirectory.
385		 *
386		 * If have already read and now fail to chdir, whack the list
387		 * to make the names come out right, and set the parent errno
388		 * so the application will eventually get an error condition.
389		 * Set the FTS_DONTCHDIR flag so that when we logically change
390		 * directories back to the parent we don't do a chdir.
391		 *
392		 * If haven't read do so.  If the read fails, fts_build sets
393		 * FTS_STOP or the fts_info field of the node.
394		 */
395		if (sp->fts_child != NULL) {
396			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
397				p->fts_errno = errno;
398				p->fts_flags |= FTS_DONTCHDIR;
399				for (p = sp->fts_child; p != NULL;
400				    p = p->fts_link)
401					p->fts_accpath =
402					    p->fts_parent->fts_accpath;
403			}
404		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
405			if (ISSET(FTS_STOP))
406				return (NULL);
407			return (p);
408		}
409		p = sp->fts_child;
410		sp->fts_child = NULL;
411		goto name;
412	}
413
414	/* Move to the next node on this level. */
415next:	tmp = p;
416	if ((p = p->fts_link) != NULL) {
417		free(tmp);
418
419		/*
420		 * If reached the top, return to the original directory (or
421		 * the root of the tree), and load the paths for the next root.
422		 */
423		if (p->fts_level == FTS_ROOTLEVEL) {
424			if (FCHDIR(sp, sp->fts_rfd)) {
425				SET(FTS_STOP);
426				return (NULL);
427			}
428			fts_load(sp, p);
429			return (sp->fts_cur = p);
430		}
431
432		/*
433		 * User may have called fts_set on the node.  If skipped,
434		 * ignore.  If followed, get a file descriptor so we can
435		 * get back if necessary.
436		 */
437		if (p->fts_instr == FTS_SKIP)
438			goto next;
439		if (p->fts_instr == FTS_FOLLOW) {
440			p->fts_info = fts_stat(sp, p, 1);
441			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
442				if ((p->fts_symfd =
443				    _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0) {
444					p->fts_errno = errno;
445					p->fts_info = FTS_ERR;
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, 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
511__fts_set_44bsd(sp, p, instr)
512	FTS *sp;
513	FTSENT *p;
514	int instr;
515{
516	if (instr != 0 && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
517	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
518		errno = EINVAL;
519		return (1);
520	}
521	p->fts_instr = instr;
522	return (0);
523}
524
525FTSENT *
526__fts_children_44bsd(sp, instr)
527	FTS *sp;
528	int instr;
529{
530	FTSENT *p;
531	int fd;
532
533	if (instr != 0 && instr != FTS_NAMEONLY) {
534		errno = EINVAL;
535		return (NULL);
536	}
537
538	/* Set current node pointer. */
539	p = sp->fts_cur;
540
541	/*
542	 * Errno set to 0 so user can distinguish empty directory from
543	 * an error.
544	 */
545	errno = 0;
546
547	/* Fatal errors stop here. */
548	if (ISSET(FTS_STOP))
549		return (NULL);
550
551	/* Return logical hierarchy of user's arguments. */
552	if (p->fts_info == FTS_INIT)
553		return (p->fts_link);
554
555	/*
556	 * If not a directory being visited in pre-order, stop here.  Could
557	 * allow FTS_DNR, assuming the user has fixed the problem, but the
558	 * same effect is available with FTS_AGAIN.
559	 */
560	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
561		return (NULL);
562
563	/* Free up any previous child list. */
564	if (sp->fts_child != NULL)
565		fts_lfree(sp->fts_child);
566
567	if (instr == FTS_NAMEONLY) {
568		SET(FTS_NAMEONLY);
569		instr = BNAMES;
570	} else
571		instr = BCHILD;
572
573	/*
574	 * If using chdir on a relative path and called BEFORE fts_read does
575	 * its chdir to the root of a traversal, we can lose -- we need to
576	 * chdir into the subdirectory, and we don't know where the current
577	 * directory is, so we can't get back so that the upcoming chdir by
578	 * fts_read will work.
579	 */
580	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
581	    ISSET(FTS_NOCHDIR))
582		return (sp->fts_child = fts_build(sp, instr));
583
584	if ((fd = _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0)
585		return (NULL);
586	sp->fts_child = fts_build(sp, instr);
587	if (fchdir(fd))
588		return (NULL);
589	(void)_close(fd);
590	return (sp->fts_child);
591}
592
593#ifndef fts_get_clientptr
594#error "fts_get_clientptr not defined"
595#endif
596
597void *
598(__fts_get_clientptr_44bsd)(FTS *sp)
599{
600
601	return (fts_get_clientptr(sp));
602}
603
604#ifndef fts_get_stream
605#error "fts_get_stream not defined"
606#endif
607
608FTS *
609(__fts_get_stream_44bsd)(FTSENT *p)
610{
611	return (fts_get_stream(p));
612}
613
614void
615__fts_set_clientptr_44bsd(FTS *sp, void *clientptr)
616{
617
618	sp->fts_clientptr = clientptr;
619}
620
621/*
622 * This is the tricky part -- do not casually change *anything* in here.  The
623 * idea is to build the linked list of entries that are used by fts_children
624 * and fts_read.  There are lots of special cases.
625 *
626 * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
627 * set and it's a physical walk (so that symbolic links can't be directories),
628 * we can do things quickly.  First, if it's a 4.4BSD file system, the type
629 * of the file is in the directory entry.  Otherwise, we assume that the number
630 * of subdirectories in a node is equal to the number of links to the parent.
631 * The former skips all stat calls.  The latter skips stat calls in any leaf
632 * directories and for any files after the subdirectories in the directory have
633 * been found, cutting the stat calls by about 2/3.
634 */
635static FTSENT *
636fts_build(sp, type)
637	FTS *sp;
638	int type;
639{
640	struct dirent *dp;
641	FTSENT *p, *head;
642	int nitems;
643	FTSENT *cur, *tail;
644	DIR *dirp;
645	void *oldaddr;
646	size_t dnamlen;
647	int cderrno, descend, len, level, maxlen, nlinks, oflag, saved_errno,
648	    nostat, doadjust;
649	char *cp;
650
651	/* Set current node pointer. */
652	cur = sp->fts_cur;
653
654	/*
655	 * Open the directory for reading.  If this fails, we're done.
656	 * If being called from fts_read, set the fts_info field.
657	 */
658#ifdef FTS_WHITEOUT
659	if (ISSET(FTS_WHITEOUT))
660		oflag = DTF_NODUP | DTF_REWIND;
661	else
662		oflag = DTF_HIDEW | DTF_NODUP | DTF_REWIND;
663#else
664#define __opendir2(path, flag) opendir(path)
665#endif
666	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
667		if (type == BREAD) {
668			cur->fts_info = FTS_DNR;
669			cur->fts_errno = errno;
670		}
671		return (NULL);
672	}
673
674	/*
675	 * Nlinks is the number of possible entries of type directory in the
676	 * directory if we're cheating on stat calls, 0 if we're not doing
677	 * any stat calls at all, -1 if we're doing stats on everything.
678	 */
679	if (type == BNAMES) {
680		nlinks = 0;
681		/* Be quiet about nostat, GCC. */
682		nostat = 0;
683	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
684		if (fts_ufslinks(sp, cur))
685			nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
686		else
687			nlinks = -1;
688		nostat = 1;
689	} else {
690		nlinks = -1;
691		nostat = 0;
692	}
693
694#ifdef notdef
695	(void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
696	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
697	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
698#endif
699	/*
700	 * If we're going to need to stat anything or we want to descend
701	 * and stay in the directory, chdir.  If this fails we keep going,
702	 * but set a flag so we don't chdir after the post-order visit.
703	 * We won't be able to stat anything, but we can still return the
704	 * names themselves.  Note, that since fts_read won't be able to
705	 * chdir into the directory, it will have to return different path
706	 * names than before, i.e. "a/b" instead of "b".  Since the node
707	 * has already been visited in pre-order, have to wait until the
708	 * post-order visit to return the error.  There is a special case
709	 * here, if there was nothing to stat then it's not an error to
710	 * not be able to stat.  This is all fairly nasty.  If a program
711	 * needed sorted entries or stat information, they had better be
712	 * checking FTS_NS on the returned nodes.
713	 */
714	cderrno = 0;
715	if (nlinks || type == BREAD) {
716		if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) {
717			if (nlinks && type == BREAD)
718				cur->fts_errno = errno;
719			cur->fts_flags |= FTS_DONTCHDIR;
720			descend = 0;
721			cderrno = errno;
722		} else
723			descend = 1;
724	} else
725		descend = 0;
726
727	/*
728	 * Figure out the max file name length that can be stored in the
729	 * current path -- the inner loop allocates more path as necessary.
730	 * We really wouldn't have to do the maxlen calculations here, we
731	 * could do them in fts_read before returning the path, but it's a
732	 * lot easier here since the length is part of the dirent structure.
733	 *
734	 * If not changing directories set a pointer so that can just append
735	 * each new name into the path.
736	 */
737	len = NAPPEND(cur);
738	if (ISSET(FTS_NOCHDIR)) {
739		cp = sp->fts_path + len;
740		*cp++ = '/';
741	} else {
742		/* GCC, you're too verbose. */
743		cp = NULL;
744	}
745	len++;
746	maxlen = sp->fts_pathlen - len;
747
748	level = cur->fts_level + 1;
749
750	/* Read the directory, attaching each entry to the `link' pointer. */
751	doadjust = 0;
752	for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) {
753		dnamlen = dp->d_namlen;
754		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
755			continue;
756
757		if ((p = fts_alloc(sp, dp->d_name, (int)dnamlen)) == NULL)
758			goto mem1;
759		if (dnamlen >= maxlen) {	/* include space for NUL */
760			oldaddr = sp->fts_path;
761			if (fts_palloc(sp, dnamlen + len + 1)) {
762				/*
763				 * No more memory for path or structures.  Save
764				 * errno, free up the current structure and the
765				 * structures already allocated.
766				 */
767mem1:				saved_errno = errno;
768				if (p)
769					free(p);
770				fts_lfree(head);
771				(void)closedir(dirp);
772				cur->fts_info = FTS_ERR;
773				SET(FTS_STOP);
774				errno = saved_errno;
775				return (NULL);
776			}
777			/* Did realloc() change the pointer? */
778			if (oldaddr != sp->fts_path) {
779				doadjust = 1;
780				if (ISSET(FTS_NOCHDIR))
781					cp = sp->fts_path + len;
782			}
783			maxlen = sp->fts_pathlen - len;
784		}
785
786		if (len + dnamlen >= USHRT_MAX) {
787			/*
788			 * In an FTSENT, fts_pathlen is a u_short so it is
789			 * possible to wraparound here.  If we do, free up
790			 * the current structure and the structures already
791			 * allocated, then error out with ENAMETOOLONG.
792			 */
793			free(p);
794			fts_lfree(head);
795			(void)closedir(dirp);
796			cur->fts_info = FTS_ERR;
797			SET(FTS_STOP);
798			errno = ENAMETOOLONG;
799			return (NULL);
800		}
801		p->fts_level = level;
802		p->fts_parent = sp->fts_cur;
803		p->fts_pathlen = len + dnamlen;
804
805#ifdef FTS_WHITEOUT
806		if (dp->d_type == DT_WHT)
807			p->fts_flags |= FTS_ISW;
808#endif
809
810		if (cderrno) {
811			if (nlinks) {
812				p->fts_info = FTS_NS;
813				p->fts_errno = cderrno;
814			} else
815				p->fts_info = FTS_NSOK;
816			p->fts_accpath = cur->fts_accpath;
817		} else if (nlinks == 0
818#ifdef DT_DIR
819		    || (nostat &&
820		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
821#endif
822		    ) {
823			p->fts_accpath =
824			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
825			p->fts_info = FTS_NSOK;
826		} else {
827			/* Build a file name for fts_stat to stat. */
828			if (ISSET(FTS_NOCHDIR)) {
829				p->fts_accpath = p->fts_path;
830				memmove(cp, p->fts_name, p->fts_namelen + 1);
831			} else
832				p->fts_accpath = p->fts_name;
833			/* Stat it. */
834			p->fts_info = fts_stat(sp, p, 0);
835
836			/* Decrement link count if applicable. */
837			if (nlinks > 0 && (p->fts_info == FTS_D ||
838			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
839				--nlinks;
840		}
841
842		/* We walk in directory order so "ls -f" doesn't get upset. */
843		p->fts_link = NULL;
844		if (head == NULL)
845			head = tail = p;
846		else {
847			tail->fts_link = p;
848			tail = p;
849		}
850		++nitems;
851	}
852	if (dirp)
853		(void)closedir(dirp);
854
855	/*
856	 * If realloc() changed the address of the path, adjust the
857	 * addresses for the rest of the tree and the dir list.
858	 */
859	if (doadjust)
860		fts_padjust(sp, head);
861
862	/*
863	 * If not changing directories, reset the path back to original
864	 * state.
865	 */
866	if (ISSET(FTS_NOCHDIR)) {
867		if (len == sp->fts_pathlen || nitems == 0)
868			--cp;
869		*cp = '\0';
870	}
871
872	/*
873	 * If descended after called from fts_children or after called from
874	 * fts_read and nothing found, get back.  At the root level we use
875	 * the saved fd; if one of fts_open()'s arguments is a relative path
876	 * to an empty directory, we wind up here with no other way back.  If
877	 * can't get back, we're done.
878	 */
879	if (descend && (type == BCHILD || !nitems) &&
880	    (cur->fts_level == FTS_ROOTLEVEL ?
881	    FCHDIR(sp, sp->fts_rfd) :
882	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
883		cur->fts_info = FTS_ERR;
884		SET(FTS_STOP);
885		return (NULL);
886	}
887
888	/* If didn't find anything, return NULL. */
889	if (!nitems) {
890		if (type == BREAD)
891			cur->fts_info = FTS_DP;
892		return (NULL);
893	}
894
895	/* Sort the entries. */
896	if (sp->fts_compar && nitems > 1)
897		head = fts_sort(sp, head, nitems);
898	return (head);
899}
900
901static u_short
902fts_stat(sp, p, follow)
903	FTS *sp;
904	FTSENT *p;
905	int follow;
906{
907	FTSENT *t;
908	dev_t dev;
909	ino_t ino;
910	struct stat *sbp, sb;
911	int saved_errno;
912
913	/* If user needs stat info, stat buffer already allocated. */
914	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
915
916#ifdef FTS_WHITEOUT
917	/* Check for whiteout. */
918	if (p->fts_flags & FTS_ISW) {
919		if (sbp != &sb) {
920			memset(sbp, '\0', sizeof(*sbp));
921			sbp->st_mode = S_IFWHT;
922		}
923		return (FTS_W);
924	}
925#endif
926
927	/*
928	 * If doing a logical walk, or application requested FTS_FOLLOW, do
929	 * a stat(2).  If that fails, check for a non-existent symlink.  If
930	 * fail, set the errno from the stat call.
931	 */
932	if (ISSET(FTS_LOGICAL) || follow) {
933		if (stat(p->fts_accpath, sbp)) {
934			saved_errno = errno;
935			if (!lstat(p->fts_accpath, sbp)) {
936				errno = 0;
937				return (FTS_SLNONE);
938			}
939			p->fts_errno = saved_errno;
940			goto err;
941		}
942	} else if (lstat(p->fts_accpath, sbp)) {
943		p->fts_errno = errno;
944err:		memset(sbp, 0, sizeof(struct stat));
945		return (FTS_NS);
946	}
947
948	if (S_ISDIR(sbp->st_mode)) {
949		/*
950		 * Set the device/inode.  Used to find cycles and check for
951		 * crossing mount points.  Also remember the link count, used
952		 * in fts_build to limit the number of stat calls.  It is
953		 * understood that these fields are only referenced if fts_info
954		 * is set to FTS_D.
955		 */
956		dev = p->fts_dev = sbp->st_dev;
957		ino = p->fts_ino = sbp->st_ino;
958		p->fts_nlink = sbp->st_nlink;
959
960		if (ISDOT(p->fts_name))
961			return (FTS_DOT);
962
963		/*
964		 * Cycle detection is done by brute force when the directory
965		 * is first encountered.  If the tree gets deep enough or the
966		 * number of symbolic links to directories is high enough,
967		 * something faster might be worthwhile.
968		 */
969		for (t = p->fts_parent;
970		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
971			if (ino == t->fts_ino && dev == t->fts_dev) {
972				p->fts_cycle = t;
973				return (FTS_DC);
974			}
975		return (FTS_D);
976	}
977	if (S_ISLNK(sbp->st_mode))
978		return (FTS_SL);
979	if (S_ISREG(sbp->st_mode))
980		return (FTS_F);
981	return (FTS_DEFAULT);
982}
983
984/*
985 * The comparison function takes pointers to pointers to FTSENT structures.
986 * Qsort wants a comparison function that takes pointers to void.
987 * (Both with appropriate levels of const-poisoning, of course!)
988 * Use a trampoline function to deal with the difference.
989 */
990static int
991fts_compar(const void *a, const void *b)
992{
993	FTS *parent;
994
995	parent = (*(const FTSENT * const *)a)->fts_fts;
996	return (*parent->fts_compar)(a, b);
997}
998
999static FTSENT *
1000fts_sort(sp, head, nitems)
1001	FTS *sp;
1002	FTSENT *head;
1003	int nitems;
1004{
1005	FTSENT **ap, *p;
1006
1007	/*
1008	 * Construct an array of pointers to the structures and call qsort(3).
1009	 * Reassemble the array in the order returned by qsort.  If unable to
1010	 * sort for memory reasons, return the directory entries in their
1011	 * current order.  Allocate enough space for the current needs plus
1012	 * 40 so don't realloc one entry at a time.
1013	 */
1014	if (nitems > sp->fts_nitems) {
1015		sp->fts_nitems = nitems + 40;
1016		if ((sp->fts_array = reallocf(sp->fts_array,
1017		    sp->fts_nitems * sizeof(FTSENT *))) == NULL) {
1018			sp->fts_nitems = 0;
1019			return (head);
1020		}
1021	}
1022	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
1023		*ap++ = p;
1024	qsort(sp->fts_array, nitems, sizeof(FTSENT *), fts_compar);
1025	for (head = *(ap = sp->fts_array); --nitems; ++ap)
1026		ap[0]->fts_link = ap[1];
1027	ap[0]->fts_link = NULL;
1028	return (head);
1029}
1030
1031static FTSENT *
1032fts_alloc(sp, name, namelen)
1033	FTS *sp;
1034	char *name;
1035	int namelen;
1036{
1037	FTSENT *p;
1038	size_t len;
1039
1040	struct ftsent_withstat {
1041		FTSENT	ent;
1042		struct	stat statbuf;
1043	};
1044
1045	/*
1046	 * The file name is a variable length array and no stat structure is
1047	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
1048	 * structure, the file name and the stat structure in one chunk, but
1049	 * be careful that the stat structure is reasonably aligned.
1050	 */
1051	if (ISSET(FTS_NOSTAT))
1052		len = sizeof(FTSENT) + namelen + 1;
1053	else
1054		len = sizeof(struct ftsent_withstat) + namelen + 1;
1055
1056	if ((p = malloc(len)) == NULL)
1057		return (NULL);
1058
1059	if (ISSET(FTS_NOSTAT)) {
1060		p->fts_name = (char *)(p + 1);
1061		p->fts_statp = NULL;
1062	} else {
1063		p->fts_name = (char *)((struct ftsent_withstat *)p + 1);
1064		p->fts_statp = &((struct ftsent_withstat *)p)->statbuf;
1065	}
1066
1067	/* Copy the name and guarantee NUL termination. */
1068	memcpy(p->fts_name, name, namelen);
1069	p->fts_name[namelen] = '\0';
1070	p->fts_namelen = namelen;
1071	p->fts_path = sp->fts_path;
1072	p->fts_errno = 0;
1073	p->fts_flags = 0;
1074	p->fts_instr = FTS_NOINSTR;
1075	p->fts_number = 0;
1076	p->fts_pointer = NULL;
1077	p->fts_fts = sp;
1078	return (p);
1079}
1080
1081static void
1082fts_lfree(head)
1083	FTSENT *head;
1084{
1085	FTSENT *p;
1086
1087	/* Free a linked list of structures. */
1088	while ((p = head)) {
1089		head = head->fts_link;
1090		free(p);
1091	}
1092}
1093
1094/*
1095 * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
1096 * Most systems will allow creation of paths much longer than MAXPATHLEN, even
1097 * though the kernel won't resolve them.  Add the size (not just what's needed)
1098 * plus 256 bytes so don't realloc the path 2 bytes at a time.
1099 */
1100static int
1101fts_palloc(sp, more)
1102	FTS *sp;
1103	size_t more;
1104{
1105
1106	sp->fts_pathlen += more + 256;
1107	/*
1108	 * Check for possible wraparound.  In an FTS, fts_pathlen is
1109	 * a signed int but in an FTSENT it is an unsigned short.
1110	 * We limit fts_pathlen to USHRT_MAX to be safe in both cases.
1111	 */
1112	if (sp->fts_pathlen < 0 || sp->fts_pathlen >= USHRT_MAX) {
1113		if (sp->fts_path)
1114			free(sp->fts_path);
1115		sp->fts_path = NULL;
1116		errno = ENAMETOOLONG;
1117		return (1);
1118	}
1119	sp->fts_path = reallocf(sp->fts_path, sp->fts_pathlen);
1120	return (sp->fts_path == NULL);
1121}
1122
1123/*
1124 * When the path is realloc'd, have to fix all of the pointers in structures
1125 * already returned.
1126 */
1127static void
1128fts_padjust(sp, head)
1129	FTS *sp;
1130	FTSENT *head;
1131{
1132	FTSENT *p;
1133	char *addr = sp->fts_path;
1134
1135#define	ADJUST(p) do {							\
1136	if ((p)->fts_accpath != (p)->fts_name) {			\
1137		(p)->fts_accpath =					\
1138		    (char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
1139	}								\
1140	(p)->fts_path = addr;						\
1141} while (0)
1142	/* Adjust the current set of children. */
1143	for (p = sp->fts_child; p; p = p->fts_link)
1144		ADJUST(p);
1145
1146	/* Adjust the rest of the tree, including the current level. */
1147	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1148		ADJUST(p);
1149		p = p->fts_link ? p->fts_link : p->fts_parent;
1150	}
1151}
1152
1153static size_t
1154fts_maxarglen(argv)
1155	char * const *argv;
1156{
1157	size_t len, max;
1158
1159	for (max = 0; *argv; ++argv)
1160		if ((len = strlen(*argv)) > max)
1161			max = len;
1162	return (max + 1);
1163}
1164
1165/*
1166 * Change to dir specified by fd or p->fts_accpath without getting
1167 * tricked by someone changing the world out from underneath us.
1168 * Assumes p->fts_dev and p->fts_ino are filled in.
1169 */
1170static int
1171fts_safe_changedir(sp, p, fd, path)
1172	FTS *sp;
1173	FTSENT *p;
1174	int fd;
1175	char *path;
1176{
1177	int ret, oerrno, newfd;
1178	struct stat sb;
1179
1180	newfd = fd;
1181	if (ISSET(FTS_NOCHDIR))
1182		return (0);
1183	if (fd < 0 && (newfd = _open(path, O_RDONLY | O_CLOEXEC, 0)) < 0)
1184		return (-1);
1185	if (_fstat(newfd, &sb)) {
1186		ret = -1;
1187		goto bail;
1188	}
1189	if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) {
1190		errno = ENOENT;		/* disinformation */
1191		ret = -1;
1192		goto bail;
1193	}
1194	ret = fchdir(newfd);
1195bail:
1196	oerrno = errno;
1197	if (fd < 0)
1198		(void)_close(newfd);
1199	errno = oerrno;
1200	return (ret);
1201}
1202
1203/*
1204 * Check if the filesystem for "ent" has UFS-style links.
1205 */
1206static int
1207fts_ufslinks(FTS *sp, const FTSENT *ent)
1208{
1209	struct _fts_private *priv;
1210	const char **cpp;
1211
1212	priv = (struct _fts_private *)sp;
1213	/*
1214	 * If this node's device is different from the previous, grab
1215	 * the filesystem information, and decide on the reliability
1216	 * of the link information from this filesystem for stat(2)
1217	 * avoidance.
1218	 */
1219	if (priv->ftsp_dev != ent->fts_dev) {
1220		if (statfs(ent->fts_path, &priv->ftsp_statfs) != -1) {
1221			priv->ftsp_dev = ent->fts_dev;
1222			priv->ftsp_linksreliable = 0;
1223			for (cpp = ufslike_filesystems; *cpp; cpp++) {
1224				if (strcmp(priv->ftsp_statfs.f_fstypename,
1225				    *cpp) == 0) {
1226					priv->ftsp_linksreliable = 1;
1227					break;
1228				}
1229			}
1230		} else {
1231			priv->ftsp_linksreliable = 0;
1232		}
1233	}
1234	return (priv->ftsp_linksreliable);
1235}
1236
1237__sym_compat(fts_open, __fts_open_44bsd, FBSD_1.0);
1238__sym_compat(fts_close, __fts_close_44bsd, FBSD_1.0);
1239__sym_compat(fts_read, __fts_read_44bsd, FBSD_1.0);
1240__sym_compat(fts_set, __fts_set_44bsd, FBSD_1.0);
1241__sym_compat(fts_children, __fts_children_44bsd, FBSD_1.0);
1242__sym_compat(fts_get_clientptr, __fts_get_clientptr_44bsd, FBSD_1.0);
1243__sym_compat(fts_get_stream, __fts_get_stream_44bsd, FBSD_1.0);
1244__sym_compat(fts_set_clientptr, __fts_set_clientptr_44bsd, FBSD_1.0);
1245