1 /*	$OpenBSD: fts.c,v 1.48 2014/11/20 04:14:15 guenther 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 #include <sys/param.h>
33 #include <sys/stat.h>
34 
35 #include <assert.h>
36 #include <dirent.h>
37 #include <errno.h>
38 #include <fcntl.h>
39 #include <fts.h>
40 #include <limits.h>
41 #include <stdlib.h>
42 #include <string.h>
43 #include <unistd.h>
44 
45 static FTSENT	*fts_alloc(FTS *, char *, size_t);
46 static FTSENT	*fts_build(FTS *, int);
47 static void	 fts_lfree(FTSENT *);
48 static void	 fts_load(FTS *, FTSENT *);
49 static size_t	 fts_maxarglen(char * const *);
50 static void	 fts_padjust(FTS *, FTSENT *);
51 static int	 fts_palloc(FTS *, size_t);
52 static FTSENT	*fts_sort(FTS *, FTSENT *, int);
53 static u_short	 fts_stat(FTS *, FTSENT *, int, int);
54 static int	 fts_safe_changedir(FTS *, FTSENT *, int, char *);
55 
56 #define ALIGNBYTES (sizeof(uintptr_t) - 1)
57 #define ALIGN(p) (((uintptr_t)(p) + ALIGNBYTES) &~ ALIGNBYTES)
58 void* reallocarray(void*, size_t, size_t);
59 
60 #define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
61 
62 #define	CLR(opt)	(sp->fts_options &= ~(opt))
63 #define	ISSET(opt)	(sp->fts_options & (opt))
64 #define	SET(opt)	(sp->fts_options |= (opt))
65 
66 #define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
67 
68 /* fts_build flags */
69 #define	BCHILD		1		/* fts_children */
70 #define	BNAMES		2		/* fts_children, names only */
71 #define	BREAD		3		/* fts_read */
72 
__fts_open(char * const * argv,int options,int (* compar)(const FTSENT **,const FTSENT **))73 FTS* __fts_open(char* const* argv, int options, int (*compar)(const FTSENT**, const FTSENT**)) {
74 	FTS *sp;
75 	FTSENT *p, *root;
76 	int nitems;
77 	FTSENT *parent, *tmp;
78 	size_t len;
79 
80 	/* Allocate/initialize the stream */
81 	if ((sp = calloc(1, sizeof(FTS))) == NULL)
82 		return (NULL);
83 	sp->fts_compar = compar;
84 	sp->fts_options = options;
85 
86 	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
87 	if (ISSET(FTS_LOGICAL))
88 		SET(FTS_NOCHDIR);
89 
90 	/*
91 	 * Start out with 1K of path space, and enough, in any case,
92 	 * to hold the user's paths.
93 	 */
94 	if (fts_palloc(sp, MAX(fts_maxarglen(argv), PATH_MAX)))
95 		goto mem1;
96 
97 	/* Allocate/initialize root's parent. */
98 	if ((parent = fts_alloc(sp, "", 0)) == NULL)
99 		goto mem2;
100 	parent->fts_level = FTS_ROOTPARENTLEVEL;
101 
102 	/* Allocate/initialize root(s). */
103 	for (root = NULL, nitems = 0; *argv; ++argv, ++nitems) {
104 		/* Don't allow zero-length paths. */
105 		if ((len = strlen(*argv)) == 0) {
106 			errno = ENOENT;
107 			goto mem3;
108 		}
109 
110 		if ((p = fts_alloc(sp, *argv, len)) == NULL)
111 			goto mem3;
112 		p->fts_level = FTS_ROOTLEVEL;
113 		p->fts_parent = parent;
114 		p->fts_accpath = p->fts_name;
115 		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW), -1);
116 
117 		// For ftw/nftw we need to fail early: http://b/31152735
118 		if ((options & FTS_FOR_FTW) != 0 && p->fts_info == FTS_NS) goto mem3;
119 
120 		/* Command-line "." and ".." are real directories. */
121 		if (p->fts_info == FTS_DOT)
122 			p->fts_info = FTS_D;
123 
124 		/*
125 		 * If comparison routine supplied, traverse in sorted
126 		 * order; otherwise traverse in the order specified.
127 		 */
128 		if (compar) {
129 			p->fts_link = root;
130 			root = p;
131 		} else {
132 			p->fts_link = NULL;
133 			if (root == NULL)
134 				tmp = root = p;
135 			else {
136 				tmp->fts_link = p;
137 				tmp = p;
138 			}
139 		}
140 	}
141 	if (compar && nitems > 1)
142 		root = fts_sort(sp, root, nitems);
143 
144 	/*
145 	 * Allocate a dummy pointer and make fts_read think that we've just
146 	 * finished the node before the root(s); set p->fts_info to FTS_INIT
147 	 * so that everything about the "current" node is ignored.
148 	 */
149 	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
150 		goto mem3;
151 	sp->fts_cur->fts_link = root;
152 	sp->fts_cur->fts_info = FTS_INIT;
153 
154 	/*
155 	 * If using chdir(2), grab a file descriptor pointing to dot to ensure
156 	 * that we can get back here; this could be avoided for some paths,
157 	 * but almost certainly not worth the effort.  Slashes, symbolic links,
158 	 * and ".." are all fairly nasty problems.  Note, if we can't get the
159 	 * descriptor we run anyway, just more slowly.
160 	 */
161 	if (!ISSET(FTS_NOCHDIR) && (sp->fts_rfd = open(".", O_RDONLY|O_CLOEXEC, 0)) < 0)
162 		SET(FTS_NOCHDIR);
163 
164 	if (nitems == 0)
165 		free(parent);
166 
167 	return (sp);
168 
169 mem3:	fts_lfree(root);
170 	free(parent);
171 mem2:	free(sp->fts_path);
172 mem1:	free(sp);
173 	return (NULL);
174 }
175 
176 static void
fts_load(FTS * sp,FTSENT * p)177 fts_load(FTS *sp, FTSENT *p)
178 {
179 	size_t len;
180 	char *cp;
181 
182 	/*
183 	 * Load the stream structure for the next traversal.  Since we don't
184 	 * actually enter the directory until after the preorder visit, set
185 	 * the fts_accpath field specially so the chdir gets done to the right
186 	 * place and the user can access the first node.  From fts_open it's
187 	 * known that the path will fit.
188 	 */
189 	len = p->fts_pathlen = p->fts_namelen;
190 	memmove(sp->fts_path, p->fts_name, len + 1);
191 	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
192 		len = strlen(++cp);
193 		memmove(p->fts_name, cp, len + 1);
194 		p->fts_namelen = len;
195 	}
196 	p->fts_accpath = p->fts_path = sp->fts_path;
197 	sp->fts_dev = p->fts_dev;
198 }
199 
200 int
fts_close(FTS * sp)201 fts_close(FTS *sp)
202 {
203 	FTSENT *freep, *p;
204 	int rfd, error = 0;
205 
206 	/*
207 	 * This still works if we haven't read anything -- the dummy structure
208 	 * points to the root list, so we step through to the end of the root
209 	 * list which has a valid parent pointer.
210 	 */
211 	if (sp->fts_cur) {
212 		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
213 			freep = p;
214 			p = p->fts_link ? p->fts_link : p->fts_parent;
215 			free(freep);
216 		}
217 		free(p);
218 	}
219 
220 	/* Stash the original directory fd if needed. */
221 	rfd = ISSET(FTS_NOCHDIR) ? -1 : sp->fts_rfd;
222 
223 	/* Free up child linked list, sort array, path buffer, stream ptr.*/
224 	fts_lfree(sp->fts_child);
225 	free(sp->fts_array);
226 	free(sp->fts_path);
227 	free(sp);
228 
229 	/* Return to original directory, checking for error. */
230 	if (rfd != -1) {
231 		int saved_errno;
232 		error = fchdir(rfd);
233 		saved_errno = errno;
234 		(void)close(rfd);
235 		errno = saved_errno;
236 	}
237 
238 	return (error);
239 }
240 
241 /*
242  * Special case of "/" at the end of the path so that slashes aren't
243  * appended which would cause paths to be written as "....//foo".
244  */
245 #define	NAPPEND(p)							\
246 	(p->fts_path[p->fts_pathlen - 1] == '/'				\
247 	    ? p->fts_pathlen - 1 : p->fts_pathlen)
248 
249 FTSENT *
fts_read(FTS * sp)250 fts_read(FTS *sp)
251 {
252 	FTSENT *p, *tmp;
253 	int instr;
254 	char *t;
255 	int saved_errno;
256 
257 	/* If finished or unrecoverable error, return NULL. */
258 	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
259 		return (NULL);
260 
261 	/* Set current node pointer. */
262 	p = sp->fts_cur;
263 
264 	/* Save and zero out user instructions. */
265 	instr = p->fts_instr;
266 	p->fts_instr = FTS_NOINSTR;
267 
268 	/* Any type of file may be re-visited; re-stat and re-turn. */
269 	if (instr == FTS_AGAIN) {
270 		p->fts_info = fts_stat(sp, p, 0, -1);
271 		return (p);
272 	}
273 
274 	/*
275 	 * Following a symlink -- SLNONE test allows application to see
276 	 * SLNONE and recover.  If indirecting through a symlink, have
277 	 * keep a pointer to current location.  If unable to get that
278 	 * pointer, follow fails.
279 	 */
280 	if (instr == FTS_FOLLOW &&
281 	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
282 		p->fts_info = fts_stat(sp, p, 1, -1);
283 		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
284 			if ((p->fts_symfd = open(".", O_RDONLY|O_CLOEXEC, 0)) < 0) {
285 				p->fts_errno = errno;
286 				p->fts_info = FTS_ERR;
287 			} else
288 				p->fts_flags |= FTS_SYMFOLLOW;
289 		}
290 		return (p);
291 	}
292 
293 	/* Directory in pre-order. */
294 	if (p->fts_info == FTS_D) {
295 		/* If skipped or crossed mount point, do post-order visit. */
296 		if (instr == FTS_SKIP ||
297 		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
298 			if (p->fts_flags & FTS_SYMFOLLOW)
299 				(void)close(p->fts_symfd);
300 			if (sp->fts_child) {
301 				fts_lfree(sp->fts_child);
302 				sp->fts_child = NULL;
303 			}
304 			p->fts_info = FTS_DP;
305 			return (p);
306 		}
307 
308 		/* Rebuild if only read the names and now traversing. */
309 		if (sp->fts_child && ISSET(FTS_NAMEONLY)) {
310 			CLR(FTS_NAMEONLY);
311 			fts_lfree(sp->fts_child);
312 			sp->fts_child = NULL;
313 		}
314 
315 		/*
316 		 * Cd to the subdirectory.
317 		 *
318 		 * If have already read and now fail to chdir, whack the list
319 		 * to make the names come out right, and set the parent errno
320 		 * so the application will eventually get an error condition.
321 		 * Set the FTS_DONTCHDIR flag so that when we logically change
322 		 * directories back to the parent we don't do a chdir.
323 		 *
324 		 * If haven't read do so.  If the read fails, fts_build sets
325 		 * FTS_STOP or the fts_info field of the node.
326 		 */
327 		if (sp->fts_child) {
328 			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
329 				p->fts_errno = errno;
330 				p->fts_flags |= FTS_DONTCHDIR;
331 				for (p = sp->fts_child; p; p = p->fts_link)
332 					p->fts_accpath =
333 					    p->fts_parent->fts_accpath;
334 			}
335 		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
336 			if (ISSET(FTS_STOP))
337 				return (NULL);
338 			return (p);
339 		}
340 		p = sp->fts_child;
341 		sp->fts_child = NULL;
342 		goto name;
343 	}
344 
345 	/* Move to the next node on this level. */
346 next:	tmp = p;
347 	if ((p = p->fts_link)) {
348 		free(tmp);
349 
350 		/*
351 		 * If reached the top, return to the original directory (or
352 		 * the root of the tree), and load the paths for the next root.
353 		 */
354 		if (p->fts_level == FTS_ROOTLEVEL) {
355 			if (FCHDIR(sp, sp->fts_rfd)) {
356 				SET(FTS_STOP);
357 				return (NULL);
358 			}
359 			fts_load(sp, p);
360 			return (sp->fts_cur = p);
361 		}
362 
363 		/*
364 		 * User may have called fts_set on the node.  If skipped,
365 		 * ignore.  If followed, get a file descriptor so we can
366 		 * get back if necessary.
367 		 */
368 		if (p->fts_instr == FTS_SKIP)
369 			goto next;
370 		if (p->fts_instr == FTS_FOLLOW) {
371 			p->fts_info = fts_stat(sp, p, 1, -1);
372 			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
373 				if ((p->fts_symfd = open(".", O_RDONLY|O_CLOEXEC, 0)) < 0) {
374 					p->fts_errno = errno;
375 					p->fts_info = FTS_ERR;
376 				} else
377 					p->fts_flags |= FTS_SYMFOLLOW;
378 			}
379 			p->fts_instr = FTS_NOINSTR;
380 		}
381 
382 name:		t = sp->fts_path + NAPPEND(p->fts_parent);
383 		*t++ = '/';
384 		memmove(t, p->fts_name, p->fts_namelen + 1);
385 		return (sp->fts_cur = p);
386 	}
387 
388 	/* Move up to the parent node. */
389 	p = tmp->fts_parent;
390 	free(tmp);
391 
392 	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
393 		/*
394 		 * Done; free everything up and set errno to 0 so the user
395 		 * can distinguish between error and EOF.
396 		 */
397 		free(p);
398 		errno = 0;
399 		return (sp->fts_cur = NULL);
400 	}
401 
402 	/* NUL terminate the pathname. */
403 	sp->fts_path[p->fts_pathlen] = '\0';
404 
405 	/*
406 	 * Return to the parent directory.  If at a root node or came through
407 	 * a symlink, go back through the file descriptor.  Otherwise, cd up
408 	 * one directory.
409 	 */
410 	if (p->fts_level == FTS_ROOTLEVEL) {
411 		if (FCHDIR(sp, sp->fts_rfd)) {
412 			SET(FTS_STOP);
413 			sp->fts_cur = p;
414 			return (NULL);
415 		}
416 	} else if (p->fts_flags & FTS_SYMFOLLOW) {
417 		if (FCHDIR(sp, p->fts_symfd)) {
418 			saved_errno = errno;
419 			(void)close(p->fts_symfd);
420 			errno = saved_errno;
421 			SET(FTS_STOP);
422 			sp->fts_cur = p;
423 			return (NULL);
424 		}
425 		(void)close(p->fts_symfd);
426 	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
427 	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
428 		SET(FTS_STOP);
429 		sp->fts_cur = p;
430 		return (NULL);
431 	}
432 	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
433 	return (sp->fts_cur = p);
434 }
435 
436 /*
437  * Fts_set takes the stream as an argument although it's not used in this
438  * implementation; it would be necessary if anyone wanted to add global
439  * semantics to fts using fts_set.  An error return is allowed for similar
440  * reasons.
441  */
442 /* ARGSUSED */
443 int
fts_set(FTS * sp __unused,FTSENT * p,int instr)444 fts_set(FTS *sp __unused, FTSENT *p, int instr)
445 {
446 	if (instr && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
447 	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
448 		errno = EINVAL;
449 		return (1);
450 	}
451 	p->fts_instr = instr;
452 	return (0);
453 }
454 
455 FTSENT *
fts_children(FTS * sp,int instr)456 fts_children(FTS *sp, int instr)
457 {
458 	FTSENT *p;
459 	int fd;
460 
461 	if (instr && instr != FTS_NAMEONLY) {
462 		errno = EINVAL;
463 		return (NULL);
464 	}
465 
466 	/* Set current node pointer. */
467 	p = sp->fts_cur;
468 
469 	/*
470 	 * Errno set to 0 so user can distinguish empty directory from
471 	 * an error.
472 	 */
473 	errno = 0;
474 
475 	/* Fatal errors stop here. */
476 	if (ISSET(FTS_STOP))
477 		return (NULL);
478 
479 	/* Return logical hierarchy of user's arguments. */
480 	if (p->fts_info == FTS_INIT)
481 		return (p->fts_link);
482 
483 	/*
484 	 * If not a directory being visited in pre-order, stop here.  Could
485 	 * allow FTS_DNR, assuming the user has fixed the problem, but the
486 	 * same effect is available with FTS_AGAIN.
487 	 */
488 	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
489 		return (NULL);
490 
491 	/* Free up any previous child list. */
492 	fts_lfree(sp->fts_child);
493 
494 	if (instr == FTS_NAMEONLY) {
495 		SET(FTS_NAMEONLY);
496 		instr = BNAMES;
497 	} else
498 		instr = BCHILD;
499 
500 	/*
501 	 * If using chdir on a relative path and called BEFORE fts_read does
502 	 * its chdir to the root of a traversal, we can lose -- we need to
503 	 * chdir into the subdirectory, and we don't know where the current
504 	 * directory is, so we can't get back so that the upcoming chdir by
505 	 * fts_read will work.
506 	 */
507 	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
508 	    ISSET(FTS_NOCHDIR))
509 		return (sp->fts_child = fts_build(sp, instr));
510 
511 	if ((fd = open(".", O_RDONLY|O_CLOEXEC, 0)) < 0)
512 		return (NULL);
513 	sp->fts_child = fts_build(sp, instr);
514 	if (fchdir(fd)) {
515 		(void)close(fd);
516 		return (NULL);
517 	}
518 	(void)close(fd);
519 	return (sp->fts_child);
520 }
521 
522 /*
523  * This is the tricky part -- do not casually change *anything* in here.  The
524  * idea is to build the linked list of entries that are used by fts_children
525  * and fts_read.  There are lots of special cases.
526  *
527  * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
528  * set and it's a physical walk (so that symbolic links can't be directories),
529  * we can do things quickly.  First, if it's a 4.4BSD file system, the type
530  * of the file is in the directory entry.  Otherwise, we assume that the number
531  * of subdirectories in a node is equal to the number of links to the parent.
532  * The former skips all stat calls.  The latter skips stat calls in any leaf
533  * directories and for any files after the subdirectories in the directory have
534  * been found, cutting the stat calls by about 2/3.
535  */
536 static FTSENT *
fts_build(FTS * sp,int type)537 fts_build(FTS *sp, int type)
538 {
539 	struct dirent *dp;
540 	FTSENT *p, *head;
541 	FTSENT *cur, *tail;
542 	DIR *dirp;
543 	void *oldaddr;
544 	size_t len, maxlen;
545 	int nitems, cderrno, descend, level, nlinks, nostat = 0, doadjust;
546 	int saved_errno;
547 	char *cp = NULL;
548 
549 	/* Set current node pointer. */
550 	cur = sp->fts_cur;
551 
552 	/*
553 	 * Open the directory for reading.  If this fails, we're done.
554 	 * If being called from fts_read, set the fts_info field.
555 	 */
556 	if ((dirp = opendir(cur->fts_accpath)) == NULL) {
557 		if (type == BREAD) {
558 			cur->fts_info = FTS_DNR;
559 			cur->fts_errno = errno;
560 		}
561 		return (NULL);
562 	}
563 
564 	/*
565 	 * Nlinks is the number of possible entries of type directory in the
566 	 * directory if we're cheating on stat calls, 0 if we're not doing
567 	 * any stat calls at all, -1 if we're doing stats on everything.
568 	 */
569 	if (type == BNAMES)
570 		nlinks = 0;
571 	else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
572 		nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
573 		nostat = 1;
574 	} else {
575 		nlinks = -1;
576 		nostat = 0;
577 	}
578 
579 #ifdef notdef
580 	(void)printf("nlinks == %d (cur: %u)\n", nlinks, cur->fts_nlink);
581 	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
582 	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
583 #endif
584 	/*
585 	 * If we're going to need to stat anything or we want to descend
586 	 * and stay in the directory, chdir.  If this fails we keep going,
587 	 * but set a flag so we don't chdir after the post-order visit.
588 	 * We won't be able to stat anything, but we can still return the
589 	 * names themselves.  Note, that since fts_read won't be able to
590 	 * chdir into the directory, it will have to return different path
591 	 * names than before, i.e. "a/b" instead of "b".  Since the node
592 	 * has already been visited in pre-order, have to wait until the
593 	 * post-order visit to return the error.  There is a special case
594 	 * here, if there was nothing to stat then it's not an error to
595 	 * not be able to stat.  This is all fairly nasty.  If a program
596 	 * needed sorted entries or stat information, they had better be
597 	 * checking FTS_NS on the returned nodes.
598 	 */
599 	cderrno = 0;
600 	if (nlinks || type == BREAD) {
601 		if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) {
602 			if (nlinks && type == BREAD)
603 				cur->fts_errno = errno;
604 			cur->fts_flags |= FTS_DONTCHDIR;
605 			descend = 0;
606 			cderrno = errno;
607 			(void)closedir(dirp);
608 			dirp = NULL;
609 		} else
610 			descend = 1;
611 	} else
612 		descend = 0;
613 
614 	/*
615 	 * Figure out the max file name length that can be stored in the
616 	 * current path -- the inner loop allocates more path as necessary.
617 	 * We really wouldn't have to do the maxlen calculations here, we
618 	 * could do them in fts_read before returning the path, but it's a
619 	 * lot easier here since the length is part of the dirent structure.
620 	 *
621 	 * If not changing directories set a pointer so that can just append
622 	 * each new name into the path.
623 	 */
624 	len = NAPPEND(cur);
625 	if (ISSET(FTS_NOCHDIR)) {
626 		cp = sp->fts_path + len;
627 		*cp++ = '/';
628 	}
629 	len++;
630 	maxlen = sp->fts_pathlen - len;
631 
632 	/*
633 	 * fts_level is signed so we must prevent it from wrapping
634 	 * around to FTS_ROOTLEVEL and FTS_ROOTPARENTLEVEL.
635 	 */
636 	level = cur->fts_level;
637 	if (level < FTS_MAXLEVEL)
638 	    level++;
639 
640 	/* Read the directory, attaching each entry to the `link' pointer. */
641 	doadjust = 0;
642 	for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) {
643 		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
644 			continue;
645 
646 		if (!(p = fts_alloc(sp, dp->d_name, strlen(dp->d_name))))
647 			goto mem1;
648 		if (strlen(dp->d_name) >= maxlen) {	/* include space for NUL */
649 			oldaddr = sp->fts_path;
650 			if (fts_palloc(sp, strlen(dp->d_name) +len + 1)) {
651 				/*
652 				 * No more memory for path or structures.  Save
653 				 * errno, free up the current structure and the
654 				 * structures already allocated.
655 				 */
656 mem1:				saved_errno = errno;
657 				free(p);
658 				fts_lfree(head);
659 				(void)closedir(dirp);
660 				cur->fts_info = FTS_ERR;
661 				SET(FTS_STOP);
662 				errno = saved_errno;
663 				return (NULL);
664 			}
665 			/* Did realloc() change the pointer? */
666 			if (oldaddr != sp->fts_path) {
667 				doadjust = 1;
668 				if (ISSET(FTS_NOCHDIR))
669 					cp = sp->fts_path + len;
670 			}
671 			maxlen = sp->fts_pathlen - len;
672 		}
673 
674 		p->fts_level = level;
675 		p->fts_parent = sp->fts_cur;
676 		p->fts_pathlen = len + strlen(dp->d_name);
677 		if (p->fts_pathlen < len) {
678 			/*
679 			 * If we wrap, free up the current structure and
680 			 * the structures already allocated, then error
681 			 * out with ENAMETOOLONG.
682 			 */
683 			free(p);
684 			fts_lfree(head);
685 			(void)closedir(dirp);
686 			cur->fts_info = FTS_ERR;
687 			SET(FTS_STOP);
688 			errno = ENAMETOOLONG;
689 			return (NULL);
690 		}
691 
692 		if (cderrno) {
693 			if (nlinks) {
694 				p->fts_info = FTS_NS;
695 				p->fts_errno = cderrno;
696 			} else
697 				p->fts_info = FTS_NSOK;
698 			p->fts_accpath = cur->fts_accpath;
699 		} else if (nlinks == 0
700 #ifdef DT_DIR
701 		    || (nostat &&
702 		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
703 #endif
704 		    ) {
705 			p->fts_accpath =
706 			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
707 			p->fts_info = FTS_NSOK;
708 		} else {
709 			/* Build a file name for fts_stat to stat. */
710 			if (ISSET(FTS_NOCHDIR)) {
711 				p->fts_accpath = p->fts_path;
712 				assert(cp && "cp should be non-null if FTS_NOCHDIR is set");
713 				memmove(cp, p->fts_name, p->fts_namelen + 1); // NOLINT
714 				p->fts_info = fts_stat(sp, p, 0, dirfd(dirp));
715 			} else {
716 				p->fts_accpath = p->fts_name;
717 				p->fts_info = fts_stat(sp, p, 0, -1);
718 			}
719 
720 			/* Decrement link count if applicable. */
721 			if (nlinks > 0 && (p->fts_info == FTS_D ||
722 			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
723 				--nlinks;
724 		}
725 
726 		/* We walk in directory order so "ls -f" doesn't get upset. */
727 		p->fts_link = NULL;
728 		if (head == NULL)
729 			head = tail = p;
730 		else {
731 			tail->fts_link = p;
732 			tail = p;
733 		}
734 		++nitems;
735 	}
736 	if (dirp)
737 		(void)closedir(dirp);
738 
739 	/*
740 	 * If realloc() changed the address of the path, adjust the
741 	 * addresses for the rest of the tree and the dir list.
742 	 */
743 	if (doadjust)
744 		fts_padjust(sp, head);
745 
746 	/*
747 	 * If not changing directories, reset the path back to original
748 	 * state.
749 	 */
750 	if (ISSET(FTS_NOCHDIR)) {
751 		if (len == sp->fts_pathlen || nitems == 0)
752 			--cp;
753 		*cp = '\0';
754 	}
755 
756 	/*
757 	 * If descended after called from fts_children or after called from
758 	 * fts_read and nothing found, get back.  At the root level we use
759 	 * the saved fd; if one of fts_open()'s arguments is a relative path
760 	 * to an empty directory, we wind up here with no other way back.  If
761 	 * can't get back, we're done.
762 	 */
763 	if (descend && (type == BCHILD || !nitems) &&
764 	    (cur->fts_level == FTS_ROOTLEVEL ? FCHDIR(sp, sp->fts_rfd) :
765 	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
766 		cur->fts_info = FTS_ERR;
767 		SET(FTS_STOP);
768 		return (NULL);
769 	}
770 
771 	/* If didn't find anything, return NULL. */
772 	if (!nitems) {
773 		if (type == BREAD)
774 			cur->fts_info = FTS_DP;
775 		return (NULL);
776 	}
777 
778 	/* Sort the entries. */
779 	if (sp->fts_compar && nitems > 1)
780 		head = fts_sort(sp, head, nitems);
781 	return (head);
782 }
783 
784 static u_short
fts_stat(FTS * sp,FTSENT * p,int follow,int dfd)785 fts_stat(FTS *sp, FTSENT *p, int follow, int dfd)
786 {
787 	FTSENT *t;
788 	dev_t dev;
789 	ino_t ino;
790 	struct stat *sbp, sb;
791 	int saved_errno;
792 	const char *path;
793 
794 	if (dfd == -1) {
795 		path = p->fts_accpath;
796 		dfd = AT_FDCWD;
797 	} else
798 		path = p->fts_name;
799 
800 	/* If user needs stat info, stat buffer already allocated. */
801 	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
802 
803 	/*
804 	 * If doing a logical walk, or application requested FTS_FOLLOW, do
805 	 * a stat(2).  If that fails, check for a non-existent symlink.  If
806 	 * fail, set the errno from the stat call.
807 	 */
808 	if (ISSET(FTS_LOGICAL) || follow) {
809 		if (fstatat(dfd, path, sbp, 0) == -1) {
810 			saved_errno = errno;
811 			if (fstatat(dfd, path, sbp, AT_SYMLINK_NOFOLLOW) == 0) {
812 				errno = 0;
813 				return (FTS_SLNONE);
814 			}
815 			p->fts_errno = saved_errno;
816 			goto err;
817 		}
818 	} else if (fstatat(dfd, path, sbp, AT_SYMLINK_NOFOLLOW)) {
819 		p->fts_errno = errno;
820 err:		memset(sbp, 0, sizeof(struct stat));
821 		return (FTS_NS);
822 	}
823 
824 	if (S_ISDIR(sbp->st_mode)) {
825 		/*
826 		 * Set the device/inode.  Used to find cycles and check for
827 		 * crossing mount points.  Also remember the link count, used
828 		 * in fts_build to limit the number of stat calls.  It is
829 		 * understood that these fields are only referenced if fts_info
830 		 * is set to FTS_D.
831 		 */
832 		dev = p->fts_dev = sbp->st_dev;
833 		ino = p->fts_ino = sbp->st_ino;
834 		p->fts_nlink = sbp->st_nlink;
835 
836 		if (ISDOT(p->fts_name))
837 			return (FTS_DOT);
838 
839 		/*
840 		 * Cycle detection is done by brute force when the directory
841 		 * is first encountered.  If the tree gets deep enough or the
842 		 * number of symbolic links to directories is high enough,
843 		 * something faster might be worthwhile.
844 		 */
845 		for (t = p->fts_parent;
846 		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
847 			if (ino == t->fts_ino && dev == t->fts_dev) {
848 				p->fts_cycle = t;
849 				return (FTS_DC);
850 			}
851 		return (FTS_D);
852 	}
853 	if (S_ISLNK(sbp->st_mode))
854 		return (FTS_SL);
855 	if (S_ISREG(sbp->st_mode))
856 		return (FTS_F);
857 	return (FTS_DEFAULT);
858 }
859 
860 static FTSENT *
fts_sort(FTS * sp,FTSENT * head,int nitems)861 fts_sort(FTS *sp, FTSENT *head, int nitems)
862 {
863 	FTSENT **ap, *p;
864 
865 	/*
866 	 * Construct an array of pointers to the structures and call qsort(3).
867 	 * Reassemble the array in the order returned by qsort.  If unable to
868 	 * sort for memory reasons, return the directory entries in their
869 	 * current order.  Allocate enough space for the current needs plus
870 	 * 40 so don't realloc one entry at a time.
871 	 */
872 	if (nitems > sp->fts_nitems) {
873 		struct _ftsent **a;
874 
875 		sp->fts_nitems = nitems + 40;
876 		if ((a = reallocarray(sp->fts_array,
877 		    sp->fts_nitems, sizeof(FTSENT *))) == NULL) {
878 			free(sp->fts_array);
879 			sp->fts_array = NULL;
880 			sp->fts_nitems = 0;
881 			return (head);
882 		}
883 		sp->fts_array = a;
884 	}
885 	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
886 		*ap++ = p;
887 	qsort((void *)sp->fts_array, nitems, sizeof(FTSENT *), sp->fts_compar);
888 	for (head = *(ap = sp->fts_array); --nitems; ++ap)
889 		ap[0]->fts_link = ap[1];
890 	ap[0]->fts_link = NULL;
891 	return (head);
892 }
893 
894 static FTSENT *
fts_alloc(FTS * sp,char * name,size_t namelen)895 fts_alloc(FTS *sp, char *name, size_t namelen)
896 {
897 	FTSENT *p;
898 	size_t len;
899 
900 	/*
901 	 * The file name is a variable length array and no stat structure is
902 	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
903 	 * structure, the file name and the stat structure in one chunk, but
904 	 * be careful that the stat structure is reasonably aligned.  Since the
905 	 * fts_name field is declared to be of size 1, the fts_name pointer is
906 	 * namelen + 2 before the first possible address of the stat structure.
907 	 */
908 	len = sizeof(FTSENT) + namelen;
909 	if (!ISSET(FTS_NOSTAT))
910 		len += sizeof(struct stat) + ALIGNBYTES;
911 	if ((p = calloc(1, len)) == NULL)
912 		return (NULL);
913 
914 	p->fts_path = sp->fts_path;
915 	p->fts_namelen = namelen;
916 	p->fts_instr = FTS_NOINSTR;
917 	if (!ISSET(FTS_NOSTAT))
918 		p->fts_statp = (struct stat *)ALIGN(p->fts_name + namelen + 2);
919 	memcpy(p->fts_name, name, namelen);
920 
921 	return (p);
922 }
923 
924 static void
fts_lfree(FTSENT * head)925 fts_lfree(FTSENT *head)
926 {
927 	FTSENT *p;
928 
929 	/* Free a linked list of structures. */
930 	while ((p = head)) {
931 		head = head->fts_link;
932 		free(p);
933 	}
934 }
935 
936 /*
937  * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
938  * Most systems will allow creation of paths much longer than PATH_MAX, even
939  * though the kernel won't resolve them.  Add the size (not just what's needed)
940  * plus 256 bytes so don't realloc the path 2 bytes at a time.
941  */
942 static int
fts_palloc(FTS * sp,size_t more)943 fts_palloc(FTS *sp, size_t more)
944 {
945 	char *p;
946 
947 	/*
948 	 * Check for possible wraparound.
949 	 */
950 	more += 256;
951 	if (sp->fts_pathlen + more < sp->fts_pathlen) {
952 		free(sp->fts_path);
953 		sp->fts_path = NULL;
954 		errno = ENAMETOOLONG;
955 		return (1);
956 	}
957 	sp->fts_pathlen += more;
958 	p = realloc(sp->fts_path, sp->fts_pathlen);
959 	if (p == NULL) {
960 		free(sp->fts_path);
961 		sp->fts_path = NULL;
962 		return (1);
963 	}
964 	sp->fts_path = p;
965 	return (0);
966 }
967 
968 /*
969  * When the path is realloc'd, have to fix all of the pointers in structures
970  * already returned.
971  */
972 static void
fts_padjust(FTS * sp,FTSENT * head)973 fts_padjust(FTS *sp, FTSENT *head)
974 {
975 	FTSENT *p;
976 	char *addr = sp->fts_path;
977 
978 #define	ADJUST(p) {							\
979 	if ((p)->fts_accpath != (p)->fts_name) {			\
980 		(p)->fts_accpath =					\
981 		    (char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
982 	}								\
983 	(p)->fts_path = addr;						\
984 }
985 	/* Adjust the current set of children. */
986 	for (p = sp->fts_child; p; p = p->fts_link)
987 		ADJUST(p);
988 
989 	/* Adjust the rest of the tree, including the current level. */
990 	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
991 		ADJUST(p);
992 		p = p->fts_link ? p->fts_link : p->fts_parent;
993 	}
994 }
995 
996 static size_t
fts_maxarglen(char * const * argv)997 fts_maxarglen(char * const *argv)
998 {
999 	size_t len, max;
1000 
1001 	for (max = 0; *argv; ++argv)
1002 		if ((len = strlen(*argv)) > max)
1003 			max = len;
1004 	return (max + 1);
1005 }
1006 
1007 /*
1008  * Change to dir specified by fd or p->fts_accpath without getting
1009  * tricked by someone changing the world out from underneath us.
1010  * Assumes p->fts_dev and p->fts_ino are filled in.
1011  */
1012 static int
fts_safe_changedir(FTS * sp,FTSENT * p,int fd,char * path)1013 fts_safe_changedir(FTS *sp, FTSENT *p, int fd, char *path)
1014 {
1015 	int ret, oerrno, newfd;
1016 	struct stat sb;
1017 
1018 	newfd = fd;
1019 	if (ISSET(FTS_NOCHDIR))
1020 		return (0);
1021 	if (fd < 0 && (newfd = open(path, O_RDONLY|O_DIRECTORY|O_CLOEXEC, 0)) < 0)
1022 		return (-1);
1023 	if (fstat(newfd, &sb)) {
1024 		ret = -1;
1025 		goto bail;
1026 	}
1027 	if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) {
1028 		errno = ENOENT;		/* disinformation */
1029 		ret = -1;
1030 		goto bail;
1031 	}
1032 	ret = fchdir(newfd);
1033 bail:
1034 	oerrno = errno;
1035 	if (fd < 0)
1036 		(void)close(newfd);
1037 	errno = oerrno;
1038 	return (ret);
1039 }
1040 
fts_open(char * const * argv,int options,int (* compar)(const FTSENT **,const FTSENT **))1041 FTS* fts_open(char* const* argv, int options, int (*compar)(const FTSENT**, const FTSENT**)) {
1042     // Options check.
1043     if ((options & ~FTS_OPTIONMASK) != 0) {
1044         errno = EINVAL;
1045         return NULL;
1046     }
1047     return __fts_open(argv, options, compar);
1048 }
1049