fts-compat.c revision 288351
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: head/lib/libc/gen/fts-compat.c 288351 2015-09-29 04:47:31Z delphij $");
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
55#include "gen-private.h"
56
57FTSENT	*__fts_children_44bsd(FTS *, int);
58int	 __fts_close_44bsd(FTS *);
59void	*__fts_get_clientptr_44bsd(FTS *);
60FTS	*__fts_get_stream_44bsd(FTSENT *);
61FTS	*__fts_open_44bsd(char * const *, int,
62	    int (*)(const FTSENT * const *, const FTSENT * const *));
63FTSENT	*__fts_read_44bsd(FTS *);
64int	 __fts_set_44bsd(FTS *, FTSENT *, int);
65void	 __fts_set_clientptr_44bsd(FTS *, void *);
66
67static FTSENT	*fts_alloc(FTS *, char *, int);
68static FTSENT	*fts_build(FTS *, int);
69static void	 fts_lfree(FTSENT *);
70static void	 fts_load(FTS *, FTSENT *);
71static size_t	 fts_maxarglen(char * const *);
72static void	 fts_padjust(FTS *, FTSENT *);
73static int	 fts_palloc(FTS *, size_t);
74static FTSENT	*fts_sort(FTS *, FTSENT *, int);
75static u_short	 fts_stat(FTS *, FTSENT *, int);
76static int	 fts_safe_changedir(FTS *, FTSENT *, int, char *);
77static int	 fts_ufslinks(FTS *, const FTSENT *);
78
79#define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
80
81#define	CLR(opt)	(sp->fts_options &= ~(opt))
82#define	ISSET(opt)	(sp->fts_options & (opt))
83#define	SET(opt)	(sp->fts_options |= (opt))
84
85#define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
86
87/* fts_build flags */
88#define	BCHILD		1		/* fts_children */
89#define	BNAMES		2		/* fts_children, names only */
90#define	BREAD		3		/* fts_read */
91
92/*
93 * Internal representation of an FTS, including extra implementation
94 * details.  The FTS returned from fts_open points to this structure's
95 * ftsp_fts member (and can be cast to an _fts_private as required)
96 */
97struct _fts_private {
98	FTS		ftsp_fts;
99	struct statfs	ftsp_statfs;
100	dev_t		ftsp_dev;
101	int		ftsp_linksreliable;
102};
103
104/*
105 * The "FTS_NOSTAT" option can avoid a lot of calls to stat(2) if it
106 * knows that a directory could not possibly have subdirectories.  This
107 * is decided by looking at the link count: a subdirectory would
108 * increment its parent's link count by virtue of its own ".." entry.
109 * This assumption only holds for UFS-like filesystems that implement
110 * links and directories this way, so we must punt for others.
111 */
112
113static const char *ufslike_filesystems[] = {
114	"ufs",
115	"zfs",
116	"nfs",
117	"nfs4",
118	"ext2fs",
119	0
120};
121
122FTS *
123__fts_open_44bsd(char * const *argv, 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 = calloc(1, sizeof(*priv))) == NULL)
141		return (NULL);
142	sp = &priv->ftsp_fts;
143	sp->fts_compar = compar;
144	sp->fts_options = options;
145
146	/* Shush, GCC. */
147	tmp = NULL;
148
149	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
150	if (ISSET(FTS_LOGICAL))
151		SET(FTS_NOCHDIR);
152
153	/*
154	 * Start out with 1K of path space, and enough, in any case,
155	 * to hold the user's paths.
156	 */
157	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
158		goto mem1;
159
160	/* Allocate/initialize root's parent. */
161	if ((parent = fts_alloc(sp, "", 0)) == NULL)
162		goto mem2;
163	parent->fts_level = FTS_ROOTPARENTLEVEL;
164
165	/* Allocate/initialize root(s). */
166	for (root = NULL, nitems = 0; *argv != NULL; ++argv, ++nitems) {
167		/* Don't allow zero-length paths. */
168		if ((len = strlen(*argv)) == 0) {
169			errno = ENOENT;
170			goto mem3;
171		}
172
173		p = fts_alloc(sp, *argv, len);
174		p->fts_level = FTS_ROOTLEVEL;
175		p->fts_parent = parent;
176		p->fts_accpath = p->fts_name;
177		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW));
178
179		/* Command-line "." and ".." are real directories. */
180		if (p->fts_info == FTS_DOT)
181			p->fts_info = FTS_D;
182
183		/*
184		 * If comparison routine supplied, traverse in sorted
185		 * order; otherwise traverse in the order specified.
186		 */
187		if (compar) {
188			p->fts_link = root;
189			root = p;
190		} else {
191			p->fts_link = NULL;
192			if (root == NULL)
193				tmp = root = p;
194			else {
195				tmp->fts_link = p;
196				tmp = p;
197			}
198		}
199	}
200	if (compar && nitems > 1)
201		root = fts_sort(sp, root, nitems);
202
203	/*
204	 * Allocate a dummy pointer and make fts_read think that we've just
205	 * finished the node before the root(s); set p->fts_info to FTS_INIT
206	 * so that everything about the "current" node is ignored.
207	 */
208	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
209		goto mem3;
210	sp->fts_cur->fts_link = root;
211	sp->fts_cur->fts_info = FTS_INIT;
212
213	/*
214	 * If using chdir(2), grab a file descriptor pointing to dot to ensure
215	 * that we can get back here; this could be avoided for some paths,
216	 * but almost certainly not worth the effort.  Slashes, symbolic links,
217	 * and ".." are all fairly nasty problems.  Note, if we can't get the
218	 * descriptor we run anyway, just more slowly.
219	 */
220	if (!ISSET(FTS_NOCHDIR) &&
221	    (sp->fts_rfd = _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0)
222		SET(FTS_NOCHDIR);
223
224	return (sp);
225
226mem3:	fts_lfree(root);
227	free(parent);
228mem2:	free(sp->fts_path);
229mem1:	free(sp);
230	return (NULL);
231}
232
233static void
234fts_load(FTS *sp, FTSENT *p)
235{
236	int len;
237	char *cp;
238
239	/*
240	 * Load the stream structure for the next traversal.  Since we don't
241	 * actually enter the directory until after the preorder visit, set
242	 * the fts_accpath field specially so the chdir gets done to the right
243	 * place and the user can access the first node.  From fts_open it's
244	 * known that the path will fit.
245	 */
246	len = p->fts_pathlen = p->fts_namelen;
247	memmove(sp->fts_path, p->fts_name, len + 1);
248	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
249		len = strlen(++cp);
250		memmove(p->fts_name, cp, len + 1);
251		p->fts_namelen = len;
252	}
253	p->fts_accpath = p->fts_path = sp->fts_path;
254	sp->fts_dev = p->fts_dev;
255}
256
257int
258__fts_close_44bsd(FTS *sp)
259{
260	FTSENT *freep, *p;
261	int saved_errno;
262
263	/*
264	 * This still works if we haven't read anything -- the dummy structure
265	 * points to the root list, so we step through to the end of the root
266	 * list which has a valid parent pointer.
267	 */
268	if (sp->fts_cur) {
269		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
270			freep = p;
271			p = p->fts_link != NULL ? p->fts_link : p->fts_parent;
272			free(freep);
273		}
274		free(p);
275	}
276
277	/* Free up child linked list, sort array, path buffer. */
278	if (sp->fts_child)
279		fts_lfree(sp->fts_child);
280	if (sp->fts_array)
281		free(sp->fts_array);
282	free(sp->fts_path);
283
284	/* Return to original directory, save errno if necessary. */
285	if (!ISSET(FTS_NOCHDIR)) {
286		saved_errno = fchdir(sp->fts_rfd) ? errno : 0;
287		(void)_close(sp->fts_rfd);
288
289		/* Set errno and return. */
290		if (saved_errno != 0) {
291			/* Free up the stream pointer. */
292			free(sp);
293			errno = saved_errno;
294			return (-1);
295		}
296	}
297
298	/* Free up the stream pointer. */
299	free(sp);
300	return (0);
301}
302
303/*
304 * Special case of "/" at the end of the path so that slashes aren't
305 * appended which would cause paths to be written as "....//foo".
306 */
307#define	NAPPEND(p)							\
308	(p->fts_path[p->fts_pathlen - 1] == '/'				\
309	    ? p->fts_pathlen - 1 : p->fts_pathlen)
310
311FTSENT *
312__fts_read_44bsd(FTS *sp)
313{
314	FTSENT *p, *tmp;
315	int instr;
316	char *t;
317	int saved_errno;
318
319	/* If finished or unrecoverable error, return NULL. */
320	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
321		return (NULL);
322
323	/* Set current node pointer. */
324	p = sp->fts_cur;
325
326	/* Save and zero out user instructions. */
327	instr = p->fts_instr;
328	p->fts_instr = FTS_NOINSTR;
329
330	/* Any type of file may be re-visited; re-stat and re-turn. */
331	if (instr == FTS_AGAIN) {
332		p->fts_info = fts_stat(sp, p, 0);
333		return (p);
334	}
335
336	/*
337	 * Following a symlink -- SLNONE test allows application to see
338	 * SLNONE and recover.  If indirecting through a symlink, have
339	 * keep a pointer to current location.  If unable to get that
340	 * pointer, follow fails.
341	 */
342	if (instr == FTS_FOLLOW &&
343	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
344		p->fts_info = fts_stat(sp, p, 1);
345		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
346			if ((p->fts_symfd = _open(".", O_RDONLY | O_CLOEXEC,
347			    0)) < 0) {
348				p->fts_errno = errno;
349				p->fts_info = FTS_ERR;
350			} else
351				p->fts_flags |= FTS_SYMFOLLOW;
352		}
353		return (p);
354	}
355
356	/* Directory in pre-order. */
357	if (p->fts_info == FTS_D) {
358		/* If skipped or crossed mount point, do post-order visit. */
359		if (instr == FTS_SKIP ||
360		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
361			if (p->fts_flags & FTS_SYMFOLLOW)
362				(void)_close(p->fts_symfd);
363			if (sp->fts_child) {
364				fts_lfree(sp->fts_child);
365				sp->fts_child = NULL;
366			}
367			p->fts_info = FTS_DP;
368			return (p);
369		}
370
371		/* Rebuild if only read the names and now traversing. */
372		if (sp->fts_child != NULL && ISSET(FTS_NAMEONLY)) {
373			CLR(FTS_NAMEONLY);
374			fts_lfree(sp->fts_child);
375			sp->fts_child = NULL;
376		}
377
378		/*
379		 * Cd to the subdirectory.
380		 *
381		 * If have already read and now fail to chdir, whack the list
382		 * to make the names come out right, and set the parent errno
383		 * so the application will eventually get an error condition.
384		 * Set the FTS_DONTCHDIR flag so that when we logically change
385		 * directories back to the parent we don't do a chdir.
386		 *
387		 * If haven't read do so.  If the read fails, fts_build sets
388		 * FTS_STOP or the fts_info field of the node.
389		 */
390		if (sp->fts_child != NULL) {
391			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
392				p->fts_errno = errno;
393				p->fts_flags |= FTS_DONTCHDIR;
394				for (p = sp->fts_child; p != NULL;
395				    p = p->fts_link)
396					p->fts_accpath =
397					    p->fts_parent->fts_accpath;
398			}
399		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
400			if (ISSET(FTS_STOP))
401				return (NULL);
402			return (p);
403		}
404		p = sp->fts_child;
405		sp->fts_child = NULL;
406		goto name;
407	}
408
409	/* Move to the next node on this level. */
410next:	tmp = p;
411	if ((p = p->fts_link) != NULL) {
412		free(tmp);
413
414		/*
415		 * If reached the top, return to the original directory (or
416		 * the root of the tree), and load the paths for the next root.
417		 */
418		if (p->fts_level == FTS_ROOTLEVEL) {
419			if (FCHDIR(sp, sp->fts_rfd)) {
420				SET(FTS_STOP);
421				return (NULL);
422			}
423			fts_load(sp, p);
424			return (sp->fts_cur = p);
425		}
426
427		/*
428		 * User may have called fts_set on the node.  If skipped,
429		 * ignore.  If followed, get a file descriptor so we can
430		 * get back if necessary.
431		 */
432		if (p->fts_instr == FTS_SKIP)
433			goto next;
434		if (p->fts_instr == FTS_FOLLOW) {
435			p->fts_info = fts_stat(sp, p, 1);
436			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
437				if ((p->fts_symfd =
438				    _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0) {
439					p->fts_errno = errno;
440					p->fts_info = FTS_ERR;
441				} else
442					p->fts_flags |= FTS_SYMFOLLOW;
443			}
444			p->fts_instr = FTS_NOINSTR;
445		}
446
447name:		t = sp->fts_path + NAPPEND(p->fts_parent);
448		*t++ = '/';
449		memmove(t, p->fts_name, p->fts_namelen + 1);
450		return (sp->fts_cur = p);
451	}
452
453	/* Move up to the parent node. */
454	p = tmp->fts_parent;
455	free(tmp);
456
457	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
458		/*
459		 * Done; free everything up and set errno to 0 so the user
460		 * can distinguish between error and EOF.
461		 */
462		free(p);
463		errno = 0;
464		return (sp->fts_cur = NULL);
465	}
466
467	/* NUL terminate the pathname. */
468	sp->fts_path[p->fts_pathlen] = '\0';
469
470	/*
471	 * Return to the parent directory.  If at a root node or came through
472	 * a symlink, go back through the file descriptor.  Otherwise, cd up
473	 * one directory.
474	 */
475	if (p->fts_level == FTS_ROOTLEVEL) {
476		if (FCHDIR(sp, sp->fts_rfd)) {
477			SET(FTS_STOP);
478			return (NULL);
479		}
480	} else if (p->fts_flags & FTS_SYMFOLLOW) {
481		if (FCHDIR(sp, p->fts_symfd)) {
482			saved_errno = errno;
483			(void)_close(p->fts_symfd);
484			errno = saved_errno;
485			SET(FTS_STOP);
486			return (NULL);
487		}
488		(void)_close(p->fts_symfd);
489	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
490	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
491		SET(FTS_STOP);
492		return (NULL);
493	}
494	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
495	return (sp->fts_cur = p);
496}
497
498/*
499 * Fts_set takes the stream as an argument although it's not used in this
500 * implementation; it would be necessary if anyone wanted to add global
501 * semantics to fts using fts_set.  An error return is allowed for similar
502 * reasons.
503 */
504/* ARGSUSED */
505int
506__fts_set_44bsd(FTS *sp, FTSENT *p, int instr)
507{
508	if (instr != 0 && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
509	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
510		errno = EINVAL;
511		return (1);
512	}
513	p->fts_instr = instr;
514	return (0);
515}
516
517FTSENT *
518__fts_children_44bsd(FTS *sp, int instr)
519{
520	FTSENT *p;
521	int fd;
522
523	if (instr != 0 && instr != FTS_NAMEONLY) {
524		errno = EINVAL;
525		return (NULL);
526	}
527
528	/* Set current node pointer. */
529	p = sp->fts_cur;
530
531	/*
532	 * Errno set to 0 so user can distinguish empty directory from
533	 * an error.
534	 */
535	errno = 0;
536
537	/* Fatal errors stop here. */
538	if (ISSET(FTS_STOP))
539		return (NULL);
540
541	/* Return logical hierarchy of user's arguments. */
542	if (p->fts_info == FTS_INIT)
543		return (p->fts_link);
544
545	/*
546	 * If not a directory being visited in pre-order, stop here.  Could
547	 * allow FTS_DNR, assuming the user has fixed the problem, but the
548	 * same effect is available with FTS_AGAIN.
549	 */
550	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
551		return (NULL);
552
553	/* Free up any previous child list. */
554	if (sp->fts_child != NULL)
555		fts_lfree(sp->fts_child);
556
557	if (instr == FTS_NAMEONLY) {
558		SET(FTS_NAMEONLY);
559		instr = BNAMES;
560	} else
561		instr = BCHILD;
562
563	/*
564	 * If using chdir on a relative path and called BEFORE fts_read does
565	 * its chdir to the root of a traversal, we can lose -- we need to
566	 * chdir into the subdirectory, and we don't know where the current
567	 * directory is, so we can't get back so that the upcoming chdir by
568	 * fts_read will work.
569	 */
570	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
571	    ISSET(FTS_NOCHDIR))
572		return (sp->fts_child = fts_build(sp, instr));
573
574	if ((fd = _open(".", O_RDONLY | O_CLOEXEC, 0)) < 0)
575		return (NULL);
576	sp->fts_child = fts_build(sp, instr);
577	if (fchdir(fd))
578		return (NULL);
579	(void)_close(fd);
580	return (sp->fts_child);
581}
582
583#ifndef fts_get_clientptr
584#error "fts_get_clientptr not defined"
585#endif
586
587void *
588(__fts_get_clientptr_44bsd)(FTS *sp)
589{
590
591	return (fts_get_clientptr(sp));
592}
593
594#ifndef fts_get_stream
595#error "fts_get_stream not defined"
596#endif
597
598FTS *
599(__fts_get_stream_44bsd)(FTSENT *p)
600{
601	return (fts_get_stream(p));
602}
603
604void
605__fts_set_clientptr_44bsd(FTS *sp, void *clientptr)
606{
607
608	sp->fts_clientptr = clientptr;
609}
610
611/*
612 * This is the tricky part -- do not casually change *anything* in here.  The
613 * idea is to build the linked list of entries that are used by fts_children
614 * and fts_read.  There are lots of special cases.
615 *
616 * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
617 * set and it's a physical walk (so that symbolic links can't be directories),
618 * we can do things quickly.  First, if it's a 4.4BSD file system, the type
619 * of the file is in the directory entry.  Otherwise, we assume that the number
620 * of subdirectories in a node is equal to the number of links to the parent.
621 * The former skips all stat calls.  The latter skips stat calls in any leaf
622 * directories and for any files after the subdirectories in the directory have
623 * been found, cutting the stat calls by about 2/3.
624 */
625static FTSENT *
626fts_build(FTS *sp, int type)
627{
628	struct dirent *dp;
629	FTSENT *p, *head;
630	int nitems;
631	FTSENT *cur, *tail;
632	DIR *dirp;
633	void *oldaddr;
634	size_t dnamlen;
635	int cderrno, descend, len, level, maxlen, nlinks, oflag, saved_errno,
636	    nostat, doadjust;
637	char *cp;
638
639	/* Set current node pointer. */
640	cur = sp->fts_cur;
641
642	/*
643	 * Open the directory for reading.  If this fails, we're done.
644	 * If being called from fts_read, set the fts_info field.
645	 */
646#ifdef FTS_WHITEOUT
647	if (ISSET(FTS_WHITEOUT))
648		oflag = DTF_NODUP | DTF_REWIND;
649	else
650		oflag = DTF_HIDEW | DTF_NODUP | DTF_REWIND;
651#else
652#define __opendir2(path, flag) opendir(path)
653#endif
654	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
655		if (type == BREAD) {
656			cur->fts_info = FTS_DNR;
657			cur->fts_errno = errno;
658		}
659		return (NULL);
660	}
661
662	/*
663	 * Nlinks is the number of possible entries of type directory in the
664	 * directory if we're cheating on stat calls, 0 if we're not doing
665	 * any stat calls at all, -1 if we're doing stats on everything.
666	 */
667	if (type == BNAMES) {
668		nlinks = 0;
669		/* Be quiet about nostat, GCC. */
670		nostat = 0;
671	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
672		if (fts_ufslinks(sp, cur))
673			nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
674		else
675			nlinks = -1;
676		nostat = 1;
677	} else {
678		nlinks = -1;
679		nostat = 0;
680	}
681
682#ifdef notdef
683	(void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
684	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
685	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
686#endif
687	/*
688	 * If we're going to need to stat anything or we want to descend
689	 * and stay in the directory, chdir.  If this fails we keep going,
690	 * but set a flag so we don't chdir after the post-order visit.
691	 * We won't be able to stat anything, but we can still return the
692	 * names themselves.  Note, that since fts_read won't be able to
693	 * chdir into the directory, it will have to return different path
694	 * names than before, i.e. "a/b" instead of "b".  Since the node
695	 * has already been visited in pre-order, have to wait until the
696	 * post-order visit to return the error.  There is a special case
697	 * here, if there was nothing to stat then it's not an error to
698	 * not be able to stat.  This is all fairly nasty.  If a program
699	 * needed sorted entries or stat information, they had better be
700	 * checking FTS_NS on the returned nodes.
701	 */
702	cderrno = 0;
703	if (nlinks || type == BREAD) {
704		if (fts_safe_changedir(sp, cur, _dirfd(dirp), NULL)) {
705			if (nlinks && type == BREAD)
706				cur->fts_errno = errno;
707			cur->fts_flags |= FTS_DONTCHDIR;
708			descend = 0;
709			cderrno = errno;
710		} else
711			descend = 1;
712	} else
713		descend = 0;
714
715	/*
716	 * Figure out the max file name length that can be stored in the
717	 * current path -- the inner loop allocates more path as necessary.
718	 * We really wouldn't have to do the maxlen calculations here, we
719	 * could do them in fts_read before returning the path, but it's a
720	 * lot easier here since the length is part of the dirent structure.
721	 *
722	 * If not changing directories set a pointer so that can just append
723	 * each new name into the path.
724	 */
725	len = NAPPEND(cur);
726	if (ISSET(FTS_NOCHDIR)) {
727		cp = sp->fts_path + len;
728		*cp++ = '/';
729	} else {
730		/* GCC, you're too verbose. */
731		cp = NULL;
732	}
733	len++;
734	maxlen = sp->fts_pathlen - len;
735
736	level = cur->fts_level + 1;
737
738	/* Read the directory, attaching each entry to the `link' pointer. */
739	doadjust = 0;
740	for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) {
741		dnamlen = dp->d_namlen;
742		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
743			continue;
744
745		if ((p = fts_alloc(sp, dp->d_name, (int)dnamlen)) == NULL)
746			goto mem1;
747		if (dnamlen >= maxlen) {	/* include space for NUL */
748			oldaddr = sp->fts_path;
749			if (fts_palloc(sp, dnamlen + len + 1)) {
750				/*
751				 * No more memory for path or structures.  Save
752				 * errno, free up the current structure and the
753				 * structures already allocated.
754				 */
755mem1:				saved_errno = errno;
756				if (p)
757					free(p);
758				fts_lfree(head);
759				(void)closedir(dirp);
760				cur->fts_info = FTS_ERR;
761				SET(FTS_STOP);
762				errno = saved_errno;
763				return (NULL);
764			}
765			/* Did realloc() change the pointer? */
766			if (oldaddr != sp->fts_path) {
767				doadjust = 1;
768				if (ISSET(FTS_NOCHDIR))
769					cp = sp->fts_path + len;
770			}
771			maxlen = sp->fts_pathlen - len;
772		}
773
774		if (len + dnamlen >= USHRT_MAX) {
775			/*
776			 * In an FTSENT, fts_pathlen is a u_short so it is
777			 * possible to wraparound here.  If we do, free up
778			 * the current structure and the structures already
779			 * allocated, then error out with ENAMETOOLONG.
780			 */
781			free(p);
782			fts_lfree(head);
783			(void)closedir(dirp);
784			cur->fts_info = FTS_ERR;
785			SET(FTS_STOP);
786			errno = ENAMETOOLONG;
787			return (NULL);
788		}
789		p->fts_level = level;
790		p->fts_parent = sp->fts_cur;
791		p->fts_pathlen = len + dnamlen;
792
793#ifdef FTS_WHITEOUT
794		if (dp->d_type == DT_WHT)
795			p->fts_flags |= FTS_ISW;
796#endif
797
798		if (cderrno) {
799			if (nlinks) {
800				p->fts_info = FTS_NS;
801				p->fts_errno = cderrno;
802			} else
803				p->fts_info = FTS_NSOK;
804			p->fts_accpath = cur->fts_accpath;
805		} else if (nlinks == 0
806#ifdef DT_DIR
807		    || (nostat &&
808		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
809#endif
810		    ) {
811			p->fts_accpath =
812			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
813			p->fts_info = FTS_NSOK;
814		} else {
815			/* Build a file name for fts_stat to stat. */
816			if (ISSET(FTS_NOCHDIR)) {
817				p->fts_accpath = p->fts_path;
818				memmove(cp, p->fts_name, p->fts_namelen + 1);
819			} else
820				p->fts_accpath = p->fts_name;
821			/* Stat it. */
822			p->fts_info = fts_stat(sp, p, 0);
823
824			/* Decrement link count if applicable. */
825			if (nlinks > 0 && (p->fts_info == FTS_D ||
826			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
827				--nlinks;
828		}
829
830		/* We walk in directory order so "ls -f" doesn't get upset. */
831		p->fts_link = NULL;
832		if (head == NULL)
833			head = tail = p;
834		else {
835			tail->fts_link = p;
836			tail = p;
837		}
838		++nitems;
839	}
840	if (dirp)
841		(void)closedir(dirp);
842
843	/*
844	 * If realloc() changed the address of the path, adjust the
845	 * addresses for the rest of the tree and the dir list.
846	 */
847	if (doadjust)
848		fts_padjust(sp, head);
849
850	/*
851	 * If not changing directories, reset the path back to original
852	 * state.
853	 */
854	if (ISSET(FTS_NOCHDIR)) {
855		if (len == sp->fts_pathlen || nitems == 0)
856			--cp;
857		*cp = '\0';
858	}
859
860	/*
861	 * If descended after called from fts_children or after called from
862	 * fts_read and nothing found, get back.  At the root level we use
863	 * the saved fd; if one of fts_open()'s arguments is a relative path
864	 * to an empty directory, we wind up here with no other way back.  If
865	 * can't get back, we're done.
866	 */
867	if (descend && (type == BCHILD || !nitems) &&
868	    (cur->fts_level == FTS_ROOTLEVEL ?
869	    FCHDIR(sp, sp->fts_rfd) :
870	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
871		cur->fts_info = FTS_ERR;
872		SET(FTS_STOP);
873		return (NULL);
874	}
875
876	/* If didn't find anything, return NULL. */
877	if (!nitems) {
878		if (type == BREAD)
879			cur->fts_info = FTS_DP;
880		return (NULL);
881	}
882
883	/* Sort the entries. */
884	if (sp->fts_compar && nitems > 1)
885		head = fts_sort(sp, head, nitems);
886	return (head);
887}
888
889static u_short
890fts_stat(FTS *sp, FTSENT *p, int follow)
891{
892	FTSENT *t;
893	dev_t dev;
894	ino_t ino;
895	struct stat *sbp, sb;
896	int saved_errno;
897
898	/* If user needs stat info, stat buffer already allocated. */
899	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
900
901#ifdef FTS_WHITEOUT
902	/* Check for whiteout. */
903	if (p->fts_flags & FTS_ISW) {
904		if (sbp != &sb) {
905			memset(sbp, '\0', sizeof(*sbp));
906			sbp->st_mode = S_IFWHT;
907		}
908		return (FTS_W);
909	}
910#endif
911
912	/*
913	 * If doing a logical walk, or application requested FTS_FOLLOW, do
914	 * a stat(2).  If that fails, check for a non-existent symlink.  If
915	 * fail, set the errno from the stat call.
916	 */
917	if (ISSET(FTS_LOGICAL) || follow) {
918		if (stat(p->fts_accpath, sbp)) {
919			saved_errno = errno;
920			if (!lstat(p->fts_accpath, sbp)) {
921				errno = 0;
922				return (FTS_SLNONE);
923			}
924			p->fts_errno = saved_errno;
925			goto err;
926		}
927	} else if (lstat(p->fts_accpath, sbp)) {
928		p->fts_errno = errno;
929err:		memset(sbp, 0, sizeof(struct stat));
930		return (FTS_NS);
931	}
932
933	if (S_ISDIR(sbp->st_mode)) {
934		/*
935		 * Set the device/inode.  Used to find cycles and check for
936		 * crossing mount points.  Also remember the link count, used
937		 * in fts_build to limit the number of stat calls.  It is
938		 * understood that these fields are only referenced if fts_info
939		 * is set to FTS_D.
940		 */
941		dev = p->fts_dev = sbp->st_dev;
942		ino = p->fts_ino = sbp->st_ino;
943		p->fts_nlink = sbp->st_nlink;
944
945		if (ISDOT(p->fts_name))
946			return (FTS_DOT);
947
948		/*
949		 * Cycle detection is done by brute force when the directory
950		 * is first encountered.  If the tree gets deep enough or the
951		 * number of symbolic links to directories is high enough,
952		 * something faster might be worthwhile.
953		 */
954		for (t = p->fts_parent;
955		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
956			if (ino == t->fts_ino && dev == t->fts_dev) {
957				p->fts_cycle = t;
958				return (FTS_DC);
959			}
960		return (FTS_D);
961	}
962	if (S_ISLNK(sbp->st_mode))
963		return (FTS_SL);
964	if (S_ISREG(sbp->st_mode))
965		return (FTS_F);
966	return (FTS_DEFAULT);
967}
968
969/*
970 * The comparison function takes pointers to pointers to FTSENT structures.
971 * Qsort wants a comparison function that takes pointers to void.
972 * (Both with appropriate levels of const-poisoning, of course!)
973 * Use a trampoline function to deal with the difference.
974 */
975static int
976fts_compar(const void *a, const void *b)
977{
978	FTS *parent;
979
980	parent = (*(const FTSENT * const *)a)->fts_fts;
981	return (*parent->fts_compar)(a, b);
982}
983
984static FTSENT *
985fts_sort(FTS *sp, FTSENT *head, int nitems)
986{
987	FTSENT **ap, *p;
988
989	/*
990	 * Construct an array of pointers to the structures and call qsort(3).
991	 * Reassemble the array in the order returned by qsort.  If unable to
992	 * sort for memory reasons, return the directory entries in their
993	 * current order.  Allocate enough space for the current needs plus
994	 * 40 so don't realloc one entry at a time.
995	 */
996	if (nitems > sp->fts_nitems) {
997		sp->fts_nitems = nitems + 40;
998		if ((sp->fts_array = reallocf(sp->fts_array,
999		    sp->fts_nitems * sizeof(FTSENT *))) == NULL) {
1000			sp->fts_nitems = 0;
1001			return (head);
1002		}
1003	}
1004	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
1005		*ap++ = p;
1006	qsort(sp->fts_array, nitems, sizeof(FTSENT *), fts_compar);
1007	for (head = *(ap = sp->fts_array); --nitems; ++ap)
1008		ap[0]->fts_link = ap[1];
1009	ap[0]->fts_link = NULL;
1010	return (head);
1011}
1012
1013static FTSENT *
1014fts_alloc(FTS *sp, char *name, int namelen)
1015{
1016	FTSENT *p;
1017	size_t len;
1018
1019	struct ftsent_withstat {
1020		FTSENT	ent;
1021		struct	stat statbuf;
1022	};
1023
1024	/*
1025	 * The file name is a variable length array and no stat structure is
1026	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
1027	 * structure, the file name and the stat structure in one chunk, but
1028	 * be careful that the stat structure is reasonably aligned.
1029	 */
1030	if (ISSET(FTS_NOSTAT))
1031		len = sizeof(FTSENT) + namelen + 1;
1032	else
1033		len = sizeof(struct ftsent_withstat) + namelen + 1;
1034
1035	if ((p = malloc(len)) == NULL)
1036		return (NULL);
1037
1038	if (ISSET(FTS_NOSTAT)) {
1039		p->fts_name = (char *)(p + 1);
1040		p->fts_statp = NULL;
1041	} else {
1042		p->fts_name = (char *)((struct ftsent_withstat *)p + 1);
1043		p->fts_statp = &((struct ftsent_withstat *)p)->statbuf;
1044	}
1045
1046	/* Copy the name and guarantee NUL termination. */
1047	memcpy(p->fts_name, name, namelen);
1048	p->fts_name[namelen] = '\0';
1049	p->fts_namelen = namelen;
1050	p->fts_path = sp->fts_path;
1051	p->fts_errno = 0;
1052	p->fts_flags = 0;
1053	p->fts_instr = FTS_NOINSTR;
1054	p->fts_number = 0;
1055	p->fts_pointer = NULL;
1056	p->fts_fts = sp;
1057	return (p);
1058}
1059
1060static void
1061fts_lfree(FTSENT *head)
1062{
1063	FTSENT *p;
1064
1065	/* Free a linked list of structures. */
1066	while ((p = head)) {
1067		head = head->fts_link;
1068		free(p);
1069	}
1070}
1071
1072/*
1073 * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
1074 * Most systems will allow creation of paths much longer than MAXPATHLEN, even
1075 * though the kernel won't resolve them.  Add the size (not just what's needed)
1076 * plus 256 bytes so don't realloc the path 2 bytes at a time.
1077 */
1078static int
1079fts_palloc(FTS *sp, size_t more)
1080{
1081
1082	sp->fts_pathlen += more + 256;
1083	/*
1084	 * Check for possible wraparound.  In an FTS, fts_pathlen is
1085	 * a signed int but in an FTSENT it is an unsigned short.
1086	 * We limit fts_pathlen to USHRT_MAX to be safe in both cases.
1087	 */
1088	if (sp->fts_pathlen < 0 || sp->fts_pathlen >= USHRT_MAX) {
1089		if (sp->fts_path)
1090			free(sp->fts_path);
1091		sp->fts_path = NULL;
1092		errno = ENAMETOOLONG;
1093		return (1);
1094	}
1095	sp->fts_path = reallocf(sp->fts_path, sp->fts_pathlen);
1096	return (sp->fts_path == NULL);
1097}
1098
1099/*
1100 * When the path is realloc'd, have to fix all of the pointers in structures
1101 * already returned.
1102 */
1103static void
1104fts_padjust(FTS *sp, FTSENT *head)
1105{
1106	FTSENT *p;
1107	char *addr = sp->fts_path;
1108
1109#define	ADJUST(p) do {							\
1110	if ((p)->fts_accpath != (p)->fts_name) {			\
1111		(p)->fts_accpath =					\
1112		    (char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
1113	}								\
1114	(p)->fts_path = addr;						\
1115} while (0)
1116	/* Adjust the current set of children. */
1117	for (p = sp->fts_child; p; p = p->fts_link)
1118		ADJUST(p);
1119
1120	/* Adjust the rest of the tree, including the current level. */
1121	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1122		ADJUST(p);
1123		p = p->fts_link ? p->fts_link : p->fts_parent;
1124	}
1125}
1126
1127static size_t
1128fts_maxarglen(char * const *argv)
1129{
1130	size_t len, max;
1131
1132	for (max = 0; *argv; ++argv)
1133		if ((len = strlen(*argv)) > max)
1134			max = len;
1135	return (max + 1);
1136}
1137
1138/*
1139 * Change to dir specified by fd or p->fts_accpath without getting
1140 * tricked by someone changing the world out from underneath us.
1141 * Assumes p->fts_dev and p->fts_ino are filled in.
1142 */
1143static int
1144fts_safe_changedir(FTS *sp, FTSENT *p, int fd, char *path)
1145{
1146	int ret, oerrno, newfd;
1147	struct stat sb;
1148
1149	newfd = fd;
1150	if (ISSET(FTS_NOCHDIR))
1151		return (0);
1152	if (fd < 0 && (newfd = _open(path, O_RDONLY | O_CLOEXEC, 0)) < 0)
1153		return (-1);
1154	if (_fstat(newfd, &sb)) {
1155		ret = -1;
1156		goto bail;
1157	}
1158	if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) {
1159		errno = ENOENT;		/* disinformation */
1160		ret = -1;
1161		goto bail;
1162	}
1163	ret = fchdir(newfd);
1164bail:
1165	oerrno = errno;
1166	if (fd < 0)
1167		(void)_close(newfd);
1168	errno = oerrno;
1169	return (ret);
1170}
1171
1172/*
1173 * Check if the filesystem for "ent" has UFS-style links.
1174 */
1175static int
1176fts_ufslinks(FTS *sp, const FTSENT *ent)
1177{
1178	struct _fts_private *priv;
1179	const char **cpp;
1180
1181	priv = (struct _fts_private *)sp;
1182	/*
1183	 * If this node's device is different from the previous, grab
1184	 * the filesystem information, and decide on the reliability
1185	 * of the link information from this filesystem for stat(2)
1186	 * avoidance.
1187	 */
1188	if (priv->ftsp_dev != ent->fts_dev) {
1189		if (statfs(ent->fts_path, &priv->ftsp_statfs) != -1) {
1190			priv->ftsp_dev = ent->fts_dev;
1191			priv->ftsp_linksreliable = 0;
1192			for (cpp = ufslike_filesystems; *cpp; cpp++) {
1193				if (strcmp(priv->ftsp_statfs.f_fstypename,
1194				    *cpp) == 0) {
1195					priv->ftsp_linksreliable = 1;
1196					break;
1197				}
1198			}
1199		} else {
1200			priv->ftsp_linksreliable = 0;
1201		}
1202	}
1203	return (priv->ftsp_linksreliable);
1204}
1205
1206__sym_compat(fts_open, __fts_open_44bsd, FBSD_1.0);
1207__sym_compat(fts_close, __fts_close_44bsd, FBSD_1.0);
1208__sym_compat(fts_read, __fts_read_44bsd, FBSD_1.0);
1209__sym_compat(fts_set, __fts_set_44bsd, FBSD_1.0);
1210__sym_compat(fts_children, __fts_children_44bsd, FBSD_1.0);
1211__sym_compat(fts_get_clientptr, __fts_get_clientptr_44bsd, FBSD_1.0);
1212__sym_compat(fts_get_stream, __fts_get_stream_44bsd, FBSD_1.0);
1213__sym_compat(fts_set_clientptr, __fts_set_clientptr_44bsd, FBSD_1.0);
1214