fts.c revision 1.32
1/*	$NetBSD: fts.c,v 1.32 2008/03/10 01:18:44 lukem Exp $	*/
2
3/*-
4 * Copyright (c) 1990, 1993, 1994
5 *	The Regents of the University of California.  All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice, this list of conditions and the following disclaimer in the
14 *    documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 *    may be used to endorse or promote products derived from this software
17 *    without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 */
31
32#if HAVE_NBTOOL_CONFIG_H
33#include "nbtool_config.h"
34#endif
35
36#include <sys/cdefs.h>
37#if defined(LIBC_SCCS) && !defined(lint)
38#if 0
39static char sccsid[] = "@(#)fts.c	8.6 (Berkeley) 8/14/94";
40#else
41__RCSID("$NetBSD: fts.c,v 1.32 2008/03/10 01:18:44 lukem Exp $");
42#endif
43#endif /* LIBC_SCCS and not lint */
44
45#include "namespace.h"
46#include <sys/param.h>
47#include <sys/stat.h>
48
49#include <assert.h>
50#include <dirent.h>
51#include <errno.h>
52#include <fcntl.h>
53#include <fts.h>
54#include <stdlib.h>
55#include <string.h>
56#include <unistd.h>
57
58#if ! HAVE_NBTOOL_CONFIG_H
59#define	HAVE_STRUCT_DIRENT_D_NAMLEN
60#endif
61
62static FTSENT	*fts_alloc(FTS *, const char *, size_t);
63static FTSENT	*fts_build(FTS *, int);
64static void	 fts_lfree(FTSENT *);
65static void	 fts_load(FTS *, FTSENT *);
66static size_t	 fts_maxarglen(char * const *);
67static size_t	 fts_pow2(size_t);
68static int	 fts_palloc(FTS *, size_t);
69static void	 fts_padjust(FTS *, FTSENT *);
70static FTSENT	*fts_sort(FTS *, FTSENT *, size_t);
71static unsigned short fts_stat(FTS *, FTSENT *, int);
72static int	 fts_safe_changedir(const FTS *, const FTSENT *, int,
73    const char *);
74
75#define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
76
77#define	CLR(opt)	(sp->fts_options &= ~(opt))
78#define	ISSET(opt)	(sp->fts_options & (opt))
79#define	SET(opt)	(sp->fts_options |= (opt))
80
81#define	CHDIR(sp, path)	(!ISSET(FTS_NOCHDIR) && chdir(path))
82#define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
83
84/* fts_build flags */
85#define	BCHILD		1		/* fts_children */
86#define	BNAMES		2		/* fts_children, names only */
87#define	BREAD		3		/* fts_read */
88
89#ifndef DTF_HIDEW
90#undef FTS_WHITEOUT
91#endif
92
93FTS *
94fts_open(char * const *argv, int options,
95    int (*compar)(const FTSENT **, const FTSENT **))
96{
97	FTS *sp;
98	FTSENT *p, *root;
99	size_t nitems;
100	FTSENT *parent, *tmp = NULL;	/* pacify gcc */
101	size_t len;
102
103	_DIAGASSERT(argv != NULL);
104
105	/* Options check. */
106	if (options & ~FTS_OPTIONMASK) {
107		errno = EINVAL;
108		return (NULL);
109	}
110
111	/* Allocate/initialize the stream */
112	if ((sp = malloc((unsigned int)sizeof(FTS))) == NULL)
113		return (NULL);
114	memset(sp, 0, sizeof(FTS));
115	sp->fts_compar = compar;
116	sp->fts_options = options;
117
118	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
119	if (ISSET(FTS_LOGICAL))
120		SET(FTS_NOCHDIR);
121
122	/*
123	 * Start out with 1K of path space, and enough, in any case,
124	 * to hold the user's paths.
125	 */
126	if (fts_palloc(sp, MAX(fts_maxarglen(argv), MAXPATHLEN)))
127		goto mem1;
128
129	/* Allocate/initialize root's parent. */
130	if ((parent = fts_alloc(sp, "", 0)) == NULL)
131		goto mem2;
132	parent->fts_level = FTS_ROOTPARENTLEVEL;
133
134	/* Allocate/initialize root(s). */
135	for (root = NULL, nitems = 0; *argv; ++argv, ++nitems) {
136		/* Don't allow zero-length paths. */
137		if ((len = strlen(*argv)) == 0) {
138			errno = ENOENT;
139			goto mem3;
140		}
141
142		if ((p = fts_alloc(sp, *argv, len)) == NULL)
143			goto mem3;
144		p->fts_level = FTS_ROOTLEVEL;
145		p->fts_parent = parent;
146		p->fts_accpath = p->fts_name;
147		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW));
148
149		/* Command-line "." and ".." are real directories. */
150		if (p->fts_info == FTS_DOT)
151			p->fts_info = FTS_D;
152
153		/*
154		 * If comparison routine supplied, traverse in sorted
155		 * order; otherwise traverse in the order specified.
156		 */
157		if (compar) {
158			p->fts_link = root;
159			root = p;
160		} else {
161			p->fts_link = NULL;
162			if (root == NULL)
163				tmp = root = p;
164			else {
165				tmp->fts_link = p;
166				tmp = p;
167			}
168		}
169	}
170	if (compar && nitems > 1)
171		root = fts_sort(sp, root, nitems);
172
173	/*
174	 * Allocate a dummy pointer and make fts_read think that we've just
175	 * finished the node before the root(s); set p->fts_info to FTS_INIT
176	 * so that everything about the "current" node is ignored.
177	 */
178	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
179		goto mem3;
180	sp->fts_cur->fts_link = root;
181	sp->fts_cur->fts_info = FTS_INIT;
182
183	/*
184	 * If using chdir(2), grab a file descriptor pointing to dot to insure
185	 * that we can get back here; this could be avoided for some paths,
186	 * but almost certainly not worth the effort.  Slashes, symbolic links,
187	 * and ".." are all fairly nasty problems.  Note, if we can't get the
188	 * descriptor we run anyway, just more slowly.
189	 */
190	if (!ISSET(FTS_NOCHDIR)) {
191		if ((sp->fts_rfd = open(".", O_RDONLY, 0)) == -1)
192			SET(FTS_NOCHDIR);
193		else if (fcntl(sp->fts_rfd, F_SETFD, FD_CLOEXEC) == -1) {
194			close(sp->fts_rfd);
195			SET(FTS_NOCHDIR);
196		}
197	}
198
199	if (nitems == 0)
200		free(parent);
201
202	return (sp);
203
204mem3:	fts_lfree(root);
205	free(parent);
206mem2:	free(sp->fts_path);
207mem1:	free(sp);
208	return (NULL);
209}
210
211static void
212fts_load(FTS *sp, FTSENT *p)
213{
214	size_t len;
215	char *cp;
216
217	_DIAGASSERT(sp != NULL);
218	_DIAGASSERT(p != NULL);
219
220	/*
221	 * Load the stream structure for the next traversal.  Since we don't
222	 * actually enter the directory until after the preorder visit, set
223	 * the fts_accpath field specially so the chdir gets done to the right
224	 * place and the user can access the first node.  From fts_open it's
225	 * known that the path will fit.
226	 */
227	len = p->fts_pathlen = p->fts_namelen;
228	memmove(sp->fts_path, p->fts_name, len + 1);
229	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
230		len = strlen(++cp);
231		memmove(p->fts_name, cp, len + 1);
232		p->fts_namelen = len;
233	}
234	p->fts_accpath = p->fts_path = sp->fts_path;
235	sp->fts_dev = p->fts_dev;
236}
237
238int
239fts_close(FTS *sp)
240{
241	FTSENT *freep, *p;
242	int saved_errno = 0;
243
244	_DIAGASSERT(sp != NULL);
245
246	/*
247	 * This still works if we haven't read anything -- the dummy structure
248	 * points to the root list, so we step through to the end of the root
249	 * list which has a valid parent pointer.
250	 */
251	if (sp->fts_cur) {
252		if (ISSET(FTS_SYMFOLLOW))
253			(void)close(sp->fts_cur->fts_symfd);
254		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
255			freep = p;
256			p = p->fts_link ? p->fts_link : p->fts_parent;
257			free(freep);
258		}
259		free(p);
260	}
261
262	/* Free up child linked list, sort array, path buffer. */
263	if (sp->fts_child)
264		fts_lfree(sp->fts_child);
265	if (sp->fts_array)
266		free(sp->fts_array);
267	free(sp->fts_path);
268
269	/* Return to original directory, save errno if necessary. */
270	if (!ISSET(FTS_NOCHDIR)) {
271		if (fchdir(sp->fts_rfd) == -1)
272			saved_errno = errno;
273		(void)close(sp->fts_rfd);
274	}
275
276	/* Free up the stream pointer. */
277	free(sp);
278	if (saved_errno) {
279		errno = saved_errno;
280		return -1;
281	}
282
283	return 0;
284}
285
286#if !defined(__FTS_COMPAT_TAILINGSLASH)
287
288/*
289 * Special case of "/" at the end of the path so that slashes aren't
290 * appended which would cause paths to be written as "....//foo".
291 */
292#define	NAPPEND(p)							\
293	(p->fts_path[p->fts_pathlen - 1] == '/'				\
294	    ? p->fts_pathlen - 1 : p->fts_pathlen)
295
296#else /* !defined(__FTS_COMPAT_TAILINGSLASH) */
297
298/*
299 * compatibility with the old behaviour.
300 *
301 * Special case a root of "/" so that slashes aren't appended which would
302 * cause paths to be written as "//foo".
303 */
304
305#define	NAPPEND(p)							\
306	(p->fts_level == FTS_ROOTLEVEL && p->fts_pathlen == 1 &&	\
307	    p->fts_path[0] == '/' ? 0 : p->fts_pathlen)
308
309#endif /* !defined(__FTS_COMPAT_TAILINGSLASH) */
310
311FTSENT *
312fts_read(FTS *sp)
313{
314	FTSENT *p, *tmp;
315	int instr;
316	char *t;
317	int saved_errno;
318
319	_DIAGASSERT(sp != NULL);
320
321	/* If finished or unrecoverable error, return NULL. */
322	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
323		return (NULL);
324
325	/* Set current node pointer. */
326	p = sp->fts_cur;
327
328	/* Save and zero out user instructions. */
329	instr = p->fts_instr;
330	p->fts_instr = FTS_NOINSTR;
331
332	/* Any type of file may be re-visited; re-stat and re-turn. */
333	if (instr == FTS_AGAIN) {
334		p->fts_info = fts_stat(sp, p, 0);
335		return (p);
336	}
337
338	/*
339	 * Following a symlink -- SLNONE test allows application to see
340	 * SLNONE and recover.  If indirecting through a symlink, have
341	 * keep a pointer to current location.  If unable to get that
342	 * pointer, follow fails.
343	 */
344	if (instr == FTS_FOLLOW &&
345	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
346		p->fts_info = fts_stat(sp, p, 1);
347		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
348			if ((p->fts_symfd = open(".", O_RDONLY, 0)) == -1) {
349				p->fts_errno = errno;
350				p->fts_info = FTS_ERR;
351			} else if (fcntl(p->fts_symfd, F_SETFD, FD_CLOEXEC) == -1) {
352				p->fts_errno = errno;
353				p->fts_info = FTS_ERR;
354				close(p->fts_symfd);
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 && 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) {
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; p = p->fts_link)
400					p->fts_accpath =
401					    p->fts_parent->fts_accpath;
402			}
403		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
404			if (ISSET(FTS_STOP))
405				return (NULL);
406			return (p);
407		}
408		p = sp->fts_child;
409		sp->fts_child = NULL;
410		goto name;
411	}
412
413	/* Move to the next node on this level. */
414next:	tmp = p;
415	if ((p = p->fts_link) != NULL) {
416		free(tmp);
417
418		/*
419		 * If reached the top, return to the original directory, and
420		 * load the paths for the next root.
421		 */
422		if (p->fts_level == FTS_ROOTLEVEL) {
423			if (FCHDIR(sp, sp->fts_rfd)) {
424				SET(FTS_STOP);
425				return (NULL);
426			}
427			fts_load(sp, p);
428			return (sp->fts_cur = p);
429		}
430
431		/*
432		 * User may have called fts_set on the node.  If skipped,
433		 * ignore.  If followed, get a file descriptor so we can
434		 * get back if necessary.
435		 */
436		if (p->fts_instr == FTS_SKIP)
437			goto next;
438		if (p->fts_instr == FTS_FOLLOW) {
439			p->fts_info = fts_stat(sp, p, 1);
440			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
441				if ((p->fts_symfd =
442				    open(".", O_RDONLY, 0)) == -1) {
443					p->fts_errno = errno;
444					p->fts_info = FTS_ERR;
445				} else if (fcntl(p->fts_symfd, F_SETFD, FD_CLOEXEC) == -1) {
446					p->fts_errno = errno;
447					p->fts_info = FTS_ERR;
448					close(p->fts_symfd);
449				} else
450					p->fts_flags |= FTS_SYMFOLLOW;
451			}
452			p->fts_instr = FTS_NOINSTR;
453		}
454
455name:		t = sp->fts_path + NAPPEND(p->fts_parent);
456		*t++ = '/';
457		memmove(t, p->fts_name, (size_t)(p->fts_namelen + 1));
458		return (sp->fts_cur = p);
459	}
460
461	/* Move up to the parent node. */
462	p = tmp->fts_parent;
463	free(tmp);
464
465	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
466		/*
467		 * Done; free everything up and set errno to 0 so the user
468		 * can distinguish between error and EOF.
469		 */
470		free(p);
471		errno = 0;
472		return (sp->fts_cur = NULL);
473	}
474
475	/* Nul terminate the pathname. */
476	sp->fts_path[p->fts_pathlen] = '\0';
477
478	/*
479	 * Return to the parent directory.  If at a root node or came through
480	 * a symlink, go back through the file descriptor.  Otherwise, cd up
481	 * one directory.
482	 */
483	if (p->fts_level == FTS_ROOTLEVEL) {
484		if (FCHDIR(sp, sp->fts_rfd)) {
485			SET(FTS_STOP);
486			return (NULL);
487		}
488	} else if (p->fts_flags & FTS_SYMFOLLOW) {
489		if (FCHDIR(sp, p->fts_symfd)) {
490			saved_errno = errno;
491			(void)close(p->fts_symfd);
492			errno = saved_errno;
493			SET(FTS_STOP);
494			return (NULL);
495		}
496		(void)close(p->fts_symfd);
497	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
498	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
499		SET(FTS_STOP);
500		return (NULL);
501	}
502	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
503	return (sp->fts_cur = p);
504}
505
506/*
507 * Fts_set takes the stream as an argument although it's not used in this
508 * implementation; it would be necessary if anyone wanted to add global
509 * semantics to fts using fts_set.  An error return is allowed for similar
510 * reasons.
511 */
512/* ARGSUSED */
513int
514fts_set(FTS *sp, FTSENT *p, int instr)
515{
516
517	_DIAGASSERT(sp != NULL);
518	_DIAGASSERT(p != NULL);
519
520	if (instr && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
521	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
522		errno = EINVAL;
523		return (1);
524	}
525	p->fts_instr = instr;
526	return (0);
527}
528
529FTSENT *
530fts_children(FTS *sp, int instr)
531{
532	FTSENT *p;
533	int fd;
534
535	_DIAGASSERT(sp != NULL);
536
537	if (instr && instr != FTS_NAMEONLY) {
538		errno = EINVAL;
539		return (NULL);
540	}
541
542	/* Set current node pointer. */
543	p = sp->fts_cur;
544
545	/*
546	 * Errno set to 0 so user can distinguish empty directory from
547	 * an error.
548	 */
549	errno = 0;
550
551	/* Fatal errors stop here. */
552	if (ISSET(FTS_STOP))
553		return (NULL);
554
555	/* Return logical hierarchy of user's arguments. */
556	if (p->fts_info == FTS_INIT)
557		return (p->fts_link);
558
559	/*
560	 * If not a directory being visited in pre-order, stop here.  Could
561	 * allow FTS_DNR, assuming the user has fixed the problem, but the
562	 * same effect is available with FTS_AGAIN.
563	 */
564	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
565		return (NULL);
566
567	/* Free up any previous child list. */
568	if (sp->fts_child)
569		fts_lfree(sp->fts_child);
570
571	if (instr == FTS_NAMEONLY) {
572		SET(FTS_NAMEONLY);
573		instr = BNAMES;
574	} else
575		instr = BCHILD;
576
577	/*
578	 * If using chdir on a relative path and called BEFORE fts_read does
579	 * its chdir to the root of a traversal, we can lose -- we need to
580	 * chdir into the subdirectory, and we don't know where the current
581	 * directory is, so we can't get back so that the upcoming chdir by
582	 * fts_read will work.
583	 */
584	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
585	    ISSET(FTS_NOCHDIR))
586		return (sp->fts_child = fts_build(sp, instr));
587
588	if ((fd = open(".", O_RDONLY, 0)) == -1)
589		return (sp->fts_child = NULL);
590	sp->fts_child = fts_build(sp, instr);
591	if (fchdir(fd)) {
592		(void)close(fd);
593		return (NULL);
594	}
595	(void)close(fd);
596	return (sp->fts_child);
597}
598
599/*
600 * This is the tricky part -- do not casually change *anything* in here.  The
601 * idea is to build the linked list of entries that are used by fts_children
602 * and fts_read.  There are lots of special cases.
603 *
604 * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
605 * set and it's a physical walk (so that symbolic links can't be directories),
606 * we can do things quickly.  First, if it's a 4.4BSD file system, the type
607 * of the file is in the directory entry.  Otherwise, we assume that the number
608 * of subdirectories in a node is equal to the number of links to the parent.
609 * The former skips all stat calls.  The latter skips stat calls in any leaf
610 * directories and for any files after the subdirectories in the directory have
611 * been found, cutting the stat calls by about 2/3.
612 */
613static FTSENT *
614fts_build(FTS *sp, int type)
615{
616	struct dirent *dp;
617	FTSENT *p, *head;
618	size_t nitems;
619	FTSENT *cur, *tail;
620	DIR *dirp;
621	void *oldaddr;
622	size_t dnamlen;
623	int cderrno, descend, len, level, nlinks, saved_errno, nostat, doadjust;
624	size_t maxlen;
625#ifdef FTS_WHITEOUT
626	int oflag;
627#endif
628	char *cp = NULL;	/* pacify gcc */
629
630	_DIAGASSERT(sp != NULL);
631
632	/* Set current node pointer. */
633	cur = sp->fts_cur;
634
635	/*
636	 * Open the directory for reading.  If this fails, we're done.
637	 * If being called from fts_read, set the fts_info field.
638	 */
639#ifdef FTS_WHITEOUT
640	if (ISSET(FTS_WHITEOUT))
641		oflag = DTF_NODUP|DTF_REWIND;
642	else
643		oflag = DTF_HIDEW|DTF_NODUP|DTF_REWIND;
644#else
645#define	__opendir2(path, flag) opendir(path)
646#endif
647	if ((dirp = __opendir2(cur->fts_accpath, oflag)) == NULL) {
648		if (type == BREAD) {
649			cur->fts_info = FTS_DNR;
650			cur->fts_errno = errno;
651		}
652		return (NULL);
653	}
654
655	/*
656	 * Nlinks is the number of possible entries of type directory in the
657	 * directory if we're cheating on stat calls, 0 if we're not doing
658	 * any stat calls at all, -1 if we're doing stats on everything.
659	 */
660	if (type == BNAMES) {
661		nlinks = 0;
662		nostat = 1;
663	} else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
664		nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
665		nostat = 1;
666	} else {
667		nlinks = -1;
668		nostat = 0;
669	}
670
671#ifdef notdef
672	(void)printf("nlinks == %d (cur: %d)\n", nlinks, cur->fts_nlink);
673	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
674	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
675#endif
676	/*
677	 * If we're going to need to stat anything or we want to descend
678	 * and stay in the directory, chdir.  If this fails we keep going,
679	 * but set a flag so we don't chdir after the post-order visit.
680	 * We won't be able to stat anything, but we can still return the
681	 * names themselves.  Note, that since fts_read won't be able to
682	 * chdir into the directory, it will have to return different path
683	 * names than before, i.e. "a/b" instead of "b".  Since the node
684	 * has already been visited in pre-order, have to wait until the
685	 * post-order visit to return the error.  There is a special case
686	 * here, if there was nothing to stat then it's not an error to
687	 * not be able to stat.  This is all fairly nasty.  If a program
688	 * needed sorted entries or stat information, they had better be
689	 * checking FTS_NS on the returned nodes.
690	 */
691	cderrno = 0;
692	if (nlinks || type == BREAD) {
693		if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) {
694			if (nlinks && type == BREAD)
695				cur->fts_errno = errno;
696			cur->fts_flags |= FTS_DONTCHDIR;
697			descend = 0;
698			cderrno = errno;
699		} else
700			descend = 1;
701	} else
702		descend = 0;
703
704	/*
705	 * Figure out the max file name length that can be stored in the
706	 * current path -- the inner loop allocates more path as necessary.
707	 * We really wouldn't have to do the maxlen calculations here, we
708	 * could do them in fts_read before returning the path, but it's a
709	 * lot easier here since the length is part of the dirent structure.
710	 *
711	 * If not changing directories set a pointer so that can just append
712	 * each new name into the path.
713	 */
714	len = NAPPEND(cur);
715	if (ISSET(FTS_NOCHDIR)) {
716		cp = sp->fts_path + len;
717		*cp++ = '/';
718	}
719	len++;
720	maxlen = sp->fts_pathlen - len;
721
722	level = cur->fts_level + 1;
723
724	/* Read the directory, attaching each entry to the `link' pointer. */
725	doadjust = 0;
726	for (head = tail = NULL, nitems = 0; (dp = readdir(dirp)) != NULL;) {
727
728		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
729			continue;
730
731#if defined(HAVE_STRUCT_DIRENT_D_NAMLEN)
732		dnamlen = dp->d_namlen;
733#else
734		dnamlen = strlen(dp->d_name);
735#endif
736		if ((p = fts_alloc(sp, dp->d_name, dnamlen)) == NULL)
737			goto mem1;
738		if (dnamlen >= maxlen) {	/* include space for NUL */
739			oldaddr = sp->fts_path;
740			if (fts_palloc(sp, dnamlen + len + 1)) {
741				/*
742				 * No more memory for path or structures.  Save
743				 * errno, free up the current structure and the
744				 * structures already allocated.
745				 */
746mem1:				saved_errno = errno;
747				if (p)
748					free(p);
749				fts_lfree(head);
750				(void)closedir(dirp);
751				errno = saved_errno;
752				cur->fts_info = FTS_ERR;
753				SET(FTS_STOP);
754				return (NULL);
755			}
756			/* Did realloc() change the pointer? */
757			if (oldaddr != sp->fts_path) {
758				doadjust = 1;
759				if (ISSET(FTS_NOCHDIR))
760					cp = sp->fts_path + len;
761			}
762			maxlen = sp->fts_pathlen - len;
763		}
764
765#if defined(__FTS_COMPAT_LENGTH)
766		if (len + dnamlen >= USHRT_MAX) {
767			/*
768			 * In an FTSENT, fts_pathlen is an unsigned short
769			 * so it is possible to wraparound here.
770			 * If we do, free up the current structure and the
771			 * structures already allocated, then error out
772			 * with ENAMETOOLONG.
773			 */
774			free(p);
775			fts_lfree(head);
776			(void)closedir(dirp);
777			cur->fts_info = FTS_ERR;
778			SET(FTS_STOP);
779			errno = ENAMETOOLONG;
780			return (NULL);
781		}
782#endif
783		p->fts_level = level;
784		p->fts_pathlen = len + dnamlen;
785		p->fts_parent = sp->fts_cur;
786
787#ifdef FTS_WHITEOUT
788		if (dp->d_type == DT_WHT)
789			p->fts_flags |= FTS_ISW;
790#endif
791
792		if (cderrno) {
793			if (nlinks) {
794				p->fts_info = FTS_NS;
795				p->fts_errno = cderrno;
796			} else
797				p->fts_info = FTS_NSOK;
798			p->fts_accpath = cur->fts_accpath;
799		} else if (nlinks == 0
800#ifdef DT_DIR
801		    || (nostat &&
802		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
803#endif
804		    ) {
805			p->fts_accpath =
806			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
807			p->fts_info = FTS_NSOK;
808		} else {
809			/* Build a file name for fts_stat to stat. */
810			if (ISSET(FTS_NOCHDIR)) {
811				p->fts_accpath = p->fts_path;
812				memmove(cp, p->fts_name,
813				        (size_t)(p->fts_namelen + 1));
814			} else
815				p->fts_accpath = p->fts_name;
816			/* Stat it. */
817			p->fts_info = fts_stat(sp, p, 0);
818
819			/* Decrement link count if applicable. */
820			if (nlinks > 0 && (p->fts_info == FTS_D ||
821			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
822				--nlinks;
823		}
824
825		/* We walk in directory order so "ls -f" doesn't get upset. */
826		p->fts_link = NULL;
827		if (head == NULL)
828			head = tail = p;
829		else {
830			tail->fts_link = p;
831			tail = p;
832		}
833		++nitems;
834	}
835	(void)closedir(dirp);
836
837	/*
838	 * If had to realloc the path, adjust the addresses for the rest
839	 * of the tree.
840	 */
841	if (doadjust)
842		fts_padjust(sp, head);
843
844	/*
845	 * If not changing directories, reset the path back to original
846	 * state.
847	 */
848	if (ISSET(FTS_NOCHDIR)) {
849		if (len == sp->fts_pathlen || nitems == 0)
850			--cp;
851		*cp = '\0';
852	}
853
854	/*
855	 * If descended after called from fts_children or after called from
856	 * fts_read and nothing found, get back.  At the root level we use
857	 * the saved fd; if one of fts_open()'s arguments is a relative path
858	 * to an empty directory, we wind up here with no other way back.  If
859	 * can't get back, we're done.
860	 */
861	if (descend && (type == BCHILD || !nitems) &&
862	    (cur->fts_level == FTS_ROOTLEVEL ?
863	    FCHDIR(sp, sp->fts_rfd) :
864	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
865		cur->fts_info = FTS_ERR;
866		SET(FTS_STOP);
867		return (NULL);
868	}
869
870	/* If didn't find anything, return NULL. */
871	if (!nitems) {
872		if (type == BREAD)
873			cur->fts_info = FTS_DP;
874		return (NULL);
875	}
876
877	/* Sort the entries. */
878	if (sp->fts_compar && nitems > 1)
879		head = fts_sort(sp, head, nitems);
880	return (head);
881}
882
883static unsigned short
884fts_stat(FTS *sp, FTSENT *p, int follow)
885{
886	FTSENT *t;
887	dev_t dev;
888	__fts_ino_t ino;
889	__fts_stat_t *sbp, sb;
890	int saved_errno;
891
892	_DIAGASSERT(sp != NULL);
893	_DIAGASSERT(p != NULL);
894
895	/* If user needs stat info, stat buffer already allocated. */
896	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
897
898#ifdef FTS_WHITEOUT
899	/* check for whiteout */
900	if (p->fts_flags & FTS_ISW) {
901		if (sbp != &sb) {
902			memset(sbp, '\0', sizeof (*sbp));
903			sbp->st_mode = S_IFWHT;
904		}
905		return (FTS_W);
906	}
907#endif
908
909	/*
910	 * If doing a logical walk, or application requested FTS_FOLLOW, do
911	 * a stat(2).  If that fails, check for a non-existent symlink.  If
912	 * fail, set the errno from the stat call.
913	 */
914	if (ISSET(FTS_LOGICAL) || follow) {
915		if (stat(p->fts_accpath, sbp)) {
916			saved_errno = errno;
917			if (!lstat(p->fts_accpath, sbp)) {
918				errno = 0;
919				return (FTS_SLNONE);
920			}
921			p->fts_errno = saved_errno;
922			goto err;
923		}
924	} else if (lstat(p->fts_accpath, sbp)) {
925		p->fts_errno = errno;
926err:		memset(sbp, 0, sizeof(*sbp));
927		return (FTS_NS);
928	}
929
930	if (S_ISDIR(sbp->st_mode)) {
931		/*
932		 * Set the device/inode.  Used to find cycles and check for
933		 * crossing mount points.  Also remember the link count, used
934		 * in fts_build to limit the number of stat calls.  It is
935		 * understood that these fields are only referenced if fts_info
936		 * is set to FTS_D.
937		 */
938		dev = p->fts_dev = sbp->st_dev;
939		ino = p->fts_ino = sbp->st_ino;
940		p->fts_nlink = sbp->st_nlink;
941
942		if (ISDOT(p->fts_name))
943			return (FTS_DOT);
944
945		/*
946		 * Cycle detection is done by brute force when the directory
947		 * is first encountered.  If the tree gets deep enough or the
948		 * number of symbolic links to directories is high enough,
949		 * something faster might be worthwhile.
950		 */
951		for (t = p->fts_parent;
952		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
953			if (ino == t->fts_ino && dev == t->fts_dev) {
954				p->fts_cycle = t;
955				return (FTS_DC);
956			}
957		return (FTS_D);
958	}
959	if (S_ISLNK(sbp->st_mode))
960		return (FTS_SL);
961	if (S_ISREG(sbp->st_mode))
962		return (FTS_F);
963	return (FTS_DEFAULT);
964}
965
966static FTSENT *
967fts_sort(FTS *sp, FTSENT *head, size_t nitems)
968{
969	FTSENT **ap, *p;
970
971	_DIAGASSERT(sp != NULL);
972	_DIAGASSERT(head != NULL);
973
974	/*
975	 * Construct an array of pointers to the structures and call qsort(3).
976	 * Reassemble the array in the order returned by qsort.  If unable to
977	 * sort for memory reasons, return the directory entries in their
978	 * current order.  Allocate enough space for the current needs plus
979	 * 40 so don't realloc one entry at a time.
980	 */
981	if (nitems > sp->fts_nitems) {
982		FTSENT **new;
983
984		new = realloc(sp->fts_array, sizeof(FTSENT *) * (nitems + 40));
985		if (new == 0)
986			return (head);
987		sp->fts_array = new;
988		sp->fts_nitems = nitems + 40;
989	}
990	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
991		*ap++ = p;
992	qsort((void *)sp->fts_array, nitems, sizeof(FTSENT *),
993		(int (*)(const void *, const void *))sp->fts_compar);
994	for (head = *(ap = sp->fts_array); --nitems; ++ap)
995		ap[0]->fts_link = ap[1];
996	ap[0]->fts_link = NULL;
997	return (head);
998}
999
1000static FTSENT *
1001fts_alloc(FTS *sp, const char *name, size_t namelen)
1002{
1003	FTSENT *p;
1004	size_t len;
1005
1006	_DIAGASSERT(sp != NULL);
1007	_DIAGASSERT(name != NULL);
1008
1009#if defined(ALIGNBYTES) && defined(ALIGN)
1010	/*
1011	 * The file name is a variable length array and no stat structure is
1012	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
1013	 * structure, the file name and the stat structure in one chunk, but
1014	 * be careful that the stat structure is reasonably aligned.  Since the
1015	 * fts_name field is declared to be of size 1, the fts_name pointer is
1016	 * namelen + 2 before the first possible address of the stat structure.
1017	 */
1018	len = sizeof(FTSENT) + namelen;
1019	if (!ISSET(FTS_NOSTAT))
1020		len += sizeof(*(p->fts_statp)) + ALIGNBYTES;
1021	if ((p = malloc(len)) == NULL)
1022		return (NULL);
1023
1024	if (!ISSET(FTS_NOSTAT))
1025		p->fts_statp = (__fts_stat_t *)ALIGN(
1026		    (unsigned long)(p->fts_name + namelen + 2));
1027#else
1028	if ((p = malloc(sizeof(FTSENT) + namelen)) == NULL)
1029		return (NULL);
1030
1031	if (!ISSET(FTS_NOSTAT))
1032		if ((p->fts_statp = malloc(sizeof(*(p->fts_statp)))) == NULL) {
1033			free(p);
1034			return (NULL);
1035		}
1036#endif
1037
1038	/* Copy the name plus the trailing NULL. */
1039	memmove(p->fts_name, name, namelen + 1);
1040
1041	p->fts_namelen = namelen;
1042	p->fts_path = sp->fts_path;
1043	p->fts_errno = 0;
1044	p->fts_flags = 0;
1045	p->fts_instr = FTS_NOINSTR;
1046	p->fts_number = 0;
1047	p->fts_pointer = NULL;
1048	return (p);
1049}
1050
1051static void
1052fts_lfree(FTSENT *head)
1053{
1054	FTSENT *p;
1055
1056	/* XXX: head may be NULL ? */
1057
1058	/* Free a linked list of structures. */
1059	while ((p = head) != NULL) {
1060		head = head->fts_link;
1061
1062#if !defined(ALIGNBYTES) || !defined(ALIGN)
1063		if (p->fts_statp)
1064			free(p->fts_statp);
1065#endif
1066		free(p);
1067	}
1068}
1069
1070static size_t
1071fts_pow2(size_t x)
1072{
1073
1074	x--;
1075	x |= x>>1;
1076	x |= x>>2;
1077	x |= x>>4;
1078	x |= x>>8;
1079	x |= x>>16;
1080#if LONG_BIT > 32
1081	x |= x>>32;
1082#endif
1083#if LONG_BIT > 64
1084	x |= x>>64;
1085#endif
1086	x++;
1087	return (x);
1088}
1089
1090/*
1091 * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
1092 * Most systems will allow creation of paths much longer than MAXPATHLEN, even
1093 * though the kernel won't resolve them.  Round up the new size to a power of 2,
1094 * so we don't realloc the path 2 bytes at a time.
1095 */
1096static int
1097fts_palloc(FTS *sp, size_t size)
1098{
1099	char *new;
1100
1101	_DIAGASSERT(sp != NULL);
1102
1103#ifdef __FTS_COMPAT_LENGTH
1104	/* Protect against fts_pathlen overflow. */
1105	if (size > USHRT_MAX + 1) {
1106		errno = ENAMETOOLONG;
1107		return (1);
1108	}
1109#endif
1110	size = fts_pow2(size);
1111	new = realloc(sp->fts_path, size);
1112	if (new == 0)
1113		return (1);
1114	sp->fts_path = new;
1115	sp->fts_pathlen = size;
1116	return (0);
1117}
1118
1119/*
1120 * When the path is realloc'd, have to fix all of the pointers in structures
1121 * already returned.
1122 */
1123static void
1124fts_padjust(FTS *sp, FTSENT *head)
1125{
1126	FTSENT *p;
1127	char *addr;
1128
1129	_DIAGASSERT(sp != NULL);
1130
1131#define	ADJUST(p) do {							\
1132	if ((p)->fts_accpath != (p)->fts_name)				\
1133		(p)->fts_accpath =					\
1134		    addr + ((p)->fts_accpath - (p)->fts_path);		\
1135	(p)->fts_path = addr;						\
1136} while (/*CONSTCOND*/0)
1137
1138	addr = sp->fts_path;
1139
1140	/* Adjust the current set of children. */
1141	for (p = sp->fts_child; p; p = p->fts_link)
1142		ADJUST(p);
1143
1144	/* Adjust the rest of the tree, including the current level. */
1145	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1146		ADJUST(p);
1147		p = p->fts_link ? p->fts_link : p->fts_parent;
1148	}
1149}
1150
1151static size_t
1152fts_maxarglen(char * const *argv)
1153{
1154	size_t len, max;
1155
1156	_DIAGASSERT(argv != NULL);
1157
1158	for (max = 0; *argv; ++argv)
1159		if ((len = strlen(*argv)) > max)
1160			max = len;
1161	return (max + 1);
1162}
1163
1164/*
1165 * Change to dir specified by fd or p->fts_accpath without getting
1166 * tricked by someone changing the world out from underneath us.
1167 * Assumes p->fts_dev and p->fts_ino are filled in.
1168 */
1169static int
1170fts_safe_changedir(const FTS *sp, const FTSENT *p, int fd, const char *path)
1171{
1172	int oldfd = fd, ret = -1;
1173	__fts_stat_t sb;
1174
1175	if (ISSET(FTS_NOCHDIR))
1176		return 0;
1177
1178	if (oldfd < 0 && (fd = open(path, O_RDONLY)) == -1)
1179		return -1;
1180
1181	if (fstat(fd, &sb) == -1)
1182		goto bail;
1183
1184	if (sb.st_ino != p->fts_ino || sb.st_dev != p->fts_dev) {
1185		errno = ENOENT;
1186		goto bail;
1187	}
1188
1189	ret = fchdir(fd);
1190
1191bail:
1192	if (oldfd < 0) {
1193		int save_errno = errno;
1194		(void)close(fd);
1195		errno = save_errno;
1196	}
1197	return ret;
1198}
1199