1 /*	$OpenBSD: fts.c,v 1.60 2021/01/08 16:06:30 tb 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>	/* ALIGN */
33 #include <sys/stat.h>
34 
35 #include <dirent.h>
36 #include <errno.h>
37 #include <fcntl.h>
38 #include <fts.h>
39 #include <limits.h>
40 #include <stdlib.h>
41 #include <string.h>
42 #include <unistd.h>
43 
44 #define MAXIMUM(a, b)	(((a) > (b)) ? (a) : (b))
45 
46 static FTSENT	*fts_alloc(FTS *, const char *, size_t);
47 static FTSENT	*fts_build(FTS *, int);
48 static void	 fts_lfree(FTSENT *);
49 static void	 fts_load(FTS *, FTSENT *);
50 static size_t	 fts_maxarglen(char * const *);
51 static void	 fts_padjust(FTS *, FTSENT *);
52 static int	 fts_palloc(FTS *, size_t);
53 static FTSENT	*fts_sort(FTS *, FTSENT *, int);
54 static u_short	 fts_stat(FTS *, FTSENT *, int, int);
55 static int	 fts_safe_changedir(FTS *, FTSENT *, int, const char *);
56 
57 /* Android: OpenBSD source compatibility workarounds. */
58 #include "private/bsd_sys_param.h"
59 #define DEF_WEAK(s) /* nothing */
60 void* recallocarray(void*, size_t, size_t, size_t);
61 
62 #define	ISDOT(a)	(a[0] == '.' && (!a[1] || (a[1] == '.' && !a[2])))
63 
64 #define	CLR(opt)	(sp->fts_options &= ~(opt))
65 #define	ISSET(opt)	(sp->fts_options & (opt))
66 #define	SET(opt)	(sp->fts_options |= (opt))
67 
68 #define	FCHDIR(sp, fd)	(!ISSET(FTS_NOCHDIR) && fchdir(fd))
69 
70 /* fts_build flags */
71 #define	BCHILD		1		/* fts_children */
72 #define	BNAMES		2		/* fts_children, names only */
73 #define	BREAD		3		/* fts_read */
74 
75 FTS *
__fts_open(char * const * argv,int options,int (* compar)(const FTSENT **,const FTSENT **))76 __fts_open(char * const *argv, int options,
77     int (*compar)(const FTSENT **, const FTSENT **))
78 {
79 	FTS *sp;
80 	FTSENT *p, *root;
81 	int nitems;
82 	FTSENT *parent, *prev;
83 
84 	/* Android: options check moved to __fts_open() for ftw(). */
85 
86 	/* At least one path must be specified. */
87 	if (*argv == NULL) {
88 		errno = EINVAL;
89 		return (NULL);
90 	}
91 
92 	/* Allocate/initialize the stream */
93 	if ((sp = calloc(1, sizeof(FTS))) == NULL)
94 		return (NULL);
95 	sp->fts_compar = compar;
96 	sp->fts_options = options;
97 
98 	/* Logical walks turn on NOCHDIR; symbolic links are too hard. */
99 	if (ISSET(FTS_LOGICAL))
100 		SET(FTS_NOCHDIR);
101 
102 	/*
103 	 * Start out with 1K of path space, and enough, in any case,
104 	 * to hold the user's paths.
105 	 */
106 	if (fts_palloc(sp, MAXIMUM(fts_maxarglen(argv), PATH_MAX)))
107 		goto mem1;
108 
109 	/* Allocate/initialize root's parent. */
110 	if ((parent = fts_alloc(sp, "", 0)) == NULL)
111 		goto mem2;
112 	parent->fts_level = FTS_ROOTPARENTLEVEL;
113 
114 	/* Allocate/initialize root(s). */
115 	for (root = prev = NULL, nitems = 0; *argv; ++argv, ++nitems) {
116 		if ((p = fts_alloc(sp, *argv, strlen(*argv))) == NULL)
117 			goto mem3;
118 		p->fts_level = FTS_ROOTLEVEL;
119 		p->fts_parent = parent;
120 		p->fts_accpath = p->fts_name;
121 		p->fts_info = fts_stat(sp, p, ISSET(FTS_COMFOLLOW), -1);
122 
123 		// Android: for ftw/nftw we need to fail early: http://b/31152735
124 		if ((options & FTS_FOR_FTW) != 0 && p->fts_info == FTS_NS) goto mem3;
125 
126 		/* Command-line "." and ".." are real directories. */
127 		if (p->fts_info == FTS_DOT)
128 			p->fts_info = FTS_D;
129 
130 		/*
131 		 * If comparison routine supplied, traverse in sorted
132 		 * order; otherwise traverse in the order specified.
133 		 */
134 		if (compar) {
135 			p->fts_link = root;
136 			root = p;
137 		} else {
138 			p->fts_link = NULL;
139 			if (root == NULL)
140 				root = p;
141 			else
142 				prev->fts_link = p;
143 			prev = p;
144 		}
145 	}
146 	if (compar && nitems > 1)
147 		root = fts_sort(sp, root, nitems);
148 
149 	/*
150 	 * Allocate a dummy pointer and make fts_read think that we've just
151 	 * finished the node before the root(s); set p->fts_info to FTS_INIT
152 	 * so that everything about the "current" node is ignored.
153 	 */
154 	if ((sp->fts_cur = fts_alloc(sp, "", 0)) == NULL)
155 		goto mem3;
156 	sp->fts_cur->fts_link = root;
157 	sp->fts_cur->fts_info = FTS_INIT;
158 
159 	/*
160 	 * If using chdir(2), grab a file descriptor pointing to dot to ensure
161 	 * that we can get back here; this could be avoided for some paths,
162 	 * but almost certainly not worth the effort.  Slashes, symbolic links,
163 	 * and ".." are all fairly nasty problems.  Note, if we can't get the
164 	 * descriptor we run anyway, just more slowly.
165 	 */
166 	if (!ISSET(FTS_NOCHDIR) &&
167 	    (sp->fts_rfd = open(".", O_RDONLY | O_CLOEXEC)) == -1)
168 		SET(FTS_NOCHDIR);
169 
170 	if (nitems == 0)
171 		free(parent);
172 
173 	return (sp);
174 
175 mem3:	fts_lfree(root);
176 	free(parent);
177 mem2:	free(sp->fts_path);
178 mem1:	free(sp);
179 	return (NULL);
180 }
181 DEF_WEAK(fts_open);
182 
183 static void
fts_load(FTS * sp,FTSENT * p)184 fts_load(FTS *sp, FTSENT *p)
185 {
186 	size_t len;
187 	char *cp;
188 
189 	/*
190 	 * Load the stream structure for the next traversal.  Since we don't
191 	 * actually enter the directory until after the preorder visit, set
192 	 * the fts_accpath field specially so the chdir gets done to the right
193 	 * place and the user can access the first node.  From fts_open it's
194 	 * known that the path will fit.
195 	 */
196 	len = p->fts_pathlen = p->fts_namelen;
197 	memmove(sp->fts_path, p->fts_name, len + 1);
198 	if ((cp = strrchr(p->fts_name, '/')) && (cp != p->fts_name || cp[1])) {
199 		len = strlen(++cp);
200 		memmove(p->fts_name, cp, len + 1);
201 		p->fts_namelen = len;
202 	}
203 	p->fts_accpath = p->fts_path = sp->fts_path;
204 	sp->fts_dev = p->fts_dev;
205 }
206 
207 int
fts_close(FTS * sp)208 fts_close(FTS *sp)
209 {
210 	FTSENT *freep, *p;
211 	int rfd, error = 0;
212 
213 	/*
214 	 * This still works if we haven't read anything -- the dummy structure
215 	 * points to the root list, so we step through to the end of the root
216 	 * list which has a valid parent pointer.
217 	 */
218 	if (sp->fts_cur) {
219 		for (p = sp->fts_cur; p->fts_level >= FTS_ROOTLEVEL;) {
220 			freep = p;
221 			p = p->fts_link ? p->fts_link : p->fts_parent;
222 			free(freep);
223 		}
224 		free(p);
225 	}
226 
227 	/* Stash the original directory fd if needed. */
228 	rfd = ISSET(FTS_NOCHDIR) ? -1 : sp->fts_rfd;
229 
230 	/* Free up child linked list, sort array, path buffer, stream ptr.*/
231 	if (sp->fts_child)
232 		fts_lfree(sp->fts_child);
233 	free(sp->fts_array);
234 	free(sp->fts_path);
235 	free(sp);
236 
237 	/* Return to original directory, checking for error. */
238 	if (rfd != -1) {
239 		int saved_errno;
240 		error = fchdir(rfd);
241 		saved_errno = errno;
242 		(void)close(rfd);
243 		errno = saved_errno;
244 	}
245 
246 	return (error);
247 }
248 DEF_WEAK(fts_close);
249 
250 /*
251  * Special case of "/" at the end of the path so that slashes aren't
252  * appended which would cause paths to be written as "....//foo".
253  */
254 #define	NAPPEND(p)							\
255 	(p->fts_path[p->fts_pathlen - 1] == '/'				\
256 	    ? p->fts_pathlen - 1 : p->fts_pathlen)
257 
258 FTSENT *
fts_read(FTS * sp)259 fts_read(FTS *sp)
260 {
261 	FTSENT *p, *tmp;
262 	int instr;
263 	char *t;
264 	int saved_errno;
265 
266 	/* If finished or unrecoverable error, return NULL. */
267 	if (sp->fts_cur == NULL || ISSET(FTS_STOP))
268 		return (NULL);
269 
270 	/* Set current node pointer. */
271 	p = sp->fts_cur;
272 
273 	/* Save and zero out user instructions. */
274 	instr = p->fts_instr;
275 	p->fts_instr = FTS_NOINSTR;
276 
277 	/* Any type of file may be re-visited; re-stat and re-turn. */
278 	if (instr == FTS_AGAIN) {
279 		p->fts_info = fts_stat(sp, p, 0, -1);
280 		return (p);
281 	}
282 
283 	/*
284 	 * Following a symlink -- SLNONE test allows application to see
285 	 * SLNONE and recover.  If indirecting through a symlink, have
286 	 * keep a pointer to current location.  If unable to get that
287 	 * pointer, follow fails.
288 	 */
289 	if (instr == FTS_FOLLOW &&
290 	    (p->fts_info == FTS_SL || p->fts_info == FTS_SLNONE)) {
291 		p->fts_info = fts_stat(sp, p, 1, -1);
292 		if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
293 			if ((p->fts_symfd =
294 			    open(".", O_RDONLY | O_CLOEXEC)) == -1) {
295 				p->fts_errno = errno;
296 				p->fts_info = FTS_ERR;
297 			} else
298 				p->fts_flags |= FTS_SYMFOLLOW;
299 		}
300 		return (p);
301 	}
302 
303 	/* Directory in pre-order. */
304 	if (p->fts_info == FTS_D) {
305 		/* If skipped or crossed mount point, do post-order visit. */
306 		if (instr == FTS_SKIP ||
307 		    (ISSET(FTS_XDEV) && p->fts_dev != sp->fts_dev)) {
308 			if (p->fts_flags & FTS_SYMFOLLOW)
309 				(void)close(p->fts_symfd);
310 			if (sp->fts_child) {
311 				fts_lfree(sp->fts_child);
312 				sp->fts_child = NULL;
313 			}
314 			p->fts_info = FTS_DP;
315 			return (p);
316 		}
317 
318 		/* Rebuild if only read the names and now traversing. */
319 		if (sp->fts_child && ISSET(FTS_NAMEONLY)) {
320 			CLR(FTS_NAMEONLY);
321 			fts_lfree(sp->fts_child);
322 			sp->fts_child = NULL;
323 		}
324 
325 		/*
326 		 * Cd to the subdirectory.
327 		 *
328 		 * If have already read and now fail to chdir, whack the list
329 		 * to make the names come out right, and set the parent errno
330 		 * so the application will eventually get an error condition.
331 		 * Set the FTS_DONTCHDIR flag so that when we logically change
332 		 * directories back to the parent we don't do a chdir.
333 		 *
334 		 * If haven't read do so.  If the read fails, fts_build sets
335 		 * FTS_STOP or the fts_info field of the node.
336 		 */
337 		if (sp->fts_child) {
338 			if (fts_safe_changedir(sp, p, -1, p->fts_accpath)) {
339 				p->fts_errno = errno;
340 				p->fts_flags |= FTS_DONTCHDIR;
341 				for (p = sp->fts_child; p; p = p->fts_link)
342 					p->fts_accpath =
343 					    p->fts_parent->fts_accpath;
344 			}
345 		} else if ((sp->fts_child = fts_build(sp, BREAD)) == NULL) {
346 			if (ISSET(FTS_STOP))
347 				return (NULL);
348 			return (p);
349 		}
350 		p = sp->fts_child;
351 		sp->fts_child = NULL;
352 		goto name;
353 	}
354 
355 	/* Move to the next node on this level. */
356 next:	tmp = p;
357 	if ((p = p->fts_link)) {
358 		free(tmp);
359 
360 		/*
361 		 * If reached the top, return to the original directory (or
362 		 * the root of the tree), and load the paths for the next root.
363 		 */
364 		if (p->fts_level == FTS_ROOTLEVEL) {
365 			if (FCHDIR(sp, sp->fts_rfd)) {
366 				SET(FTS_STOP);
367 				return (NULL);
368 			}
369 			fts_load(sp, p);
370 			return (sp->fts_cur = p);
371 		}
372 
373 		/*
374 		 * User may have called fts_set on the node.  If skipped,
375 		 * ignore.  If followed, get a file descriptor so we can
376 		 * get back if necessary.
377 		 */
378 		if (p->fts_instr == FTS_SKIP)
379 			goto next;
380 		if (p->fts_instr == FTS_FOLLOW) {
381 			p->fts_info = fts_stat(sp, p, 1, -1);
382 			if (p->fts_info == FTS_D && !ISSET(FTS_NOCHDIR)) {
383 				if ((p->fts_symfd =
384 				    open(".", O_RDONLY | O_CLOEXEC)) == -1) {
385 					p->fts_errno = errno;
386 					p->fts_info = FTS_ERR;
387 				} else
388 					p->fts_flags |= FTS_SYMFOLLOW;
389 			}
390 			p->fts_instr = FTS_NOINSTR;
391 		}
392 
393 name:		t = sp->fts_path + NAPPEND(p->fts_parent);
394 		*t++ = '/';
395 		memmove(t, p->fts_name, p->fts_namelen + 1);
396 		return (sp->fts_cur = p);
397 	}
398 
399 	/* Move up to the parent node. */
400 	p = tmp->fts_parent;
401 	free(tmp);
402 
403 	if (p->fts_level == FTS_ROOTPARENTLEVEL) {
404 		/*
405 		 * Done; free everything up and set errno to 0 so the user
406 		 * can distinguish between error and EOF.
407 		 */
408 		free(p);
409 		errno = 0;
410 		return (sp->fts_cur = NULL);
411 	}
412 
413 	/* NUL terminate the pathname. */
414 	sp->fts_path[p->fts_pathlen] = '\0';
415 
416 	/*
417 	 * Return to the parent directory.  If at a root node or came through
418 	 * a symlink, go back through the file descriptor.  Otherwise, cd up
419 	 * one directory.
420 	 */
421 	if (p->fts_level == FTS_ROOTLEVEL) {
422 		if (FCHDIR(sp, sp->fts_rfd)) {
423 			SET(FTS_STOP);
424 			sp->fts_cur = p;
425 			return (NULL);
426 		}
427 	} else if (p->fts_flags & FTS_SYMFOLLOW) {
428 		if (FCHDIR(sp, p->fts_symfd)) {
429 			saved_errno = errno;
430 			(void)close(p->fts_symfd);
431 			errno = saved_errno;
432 			SET(FTS_STOP);
433 			sp->fts_cur = p;
434 			return (NULL);
435 		}
436 		(void)close(p->fts_symfd);
437 	} else if (!(p->fts_flags & FTS_DONTCHDIR) &&
438 	    fts_safe_changedir(sp, p->fts_parent, -1, "..")) {
439 		SET(FTS_STOP);
440 		sp->fts_cur = p;
441 		return (NULL);
442 	}
443 	p->fts_info = p->fts_errno ? FTS_ERR : FTS_DP;
444 	return (sp->fts_cur = p);
445 }
446 DEF_WEAK(fts_read);
447 
448 /*
449  * Fts_set takes the stream as an argument although it's not used in this
450  * implementation; it would be necessary if anyone wanted to add global
451  * semantics to fts using fts_set.  An error return is allowed for similar
452  * reasons.
453  */
454 int
fts_set(FTS * sp __unused,FTSENT * p,int instr)455 fts_set(FTS *sp __unused, FTSENT *p, int instr)
456 {
457 	if (instr && instr != FTS_AGAIN && instr != FTS_FOLLOW &&
458 	    instr != FTS_NOINSTR && instr != FTS_SKIP) {
459 		errno = EINVAL;
460 		return (1);
461 	}
462 	p->fts_instr = instr;
463 	return (0);
464 }
465 DEF_WEAK(fts_set);
466 
467 FTSENT *
fts_children(FTS * sp,int instr)468 fts_children(FTS *sp, int instr)
469 {
470 	FTSENT *p;
471 	int fd;
472 
473 	if (instr && instr != FTS_NAMEONLY) {
474 		errno = EINVAL;
475 		return (NULL);
476 	}
477 
478 	/* Set current node pointer. */
479 	p = sp->fts_cur;
480 
481 	/*
482 	 * Errno set to 0 so user can distinguish empty directory from
483 	 * an error.
484 	 */
485 	errno = 0;
486 
487 	/* Fatal errors stop here. */
488 	if (ISSET(FTS_STOP))
489 		return (NULL);
490 
491 	/* Return logical hierarchy of user's arguments. */
492 	if (p->fts_info == FTS_INIT)
493 		return (p->fts_link);
494 
495 	/*
496 	 * If not a directory being visited in pre-order, stop here.  Could
497 	 * allow FTS_DNR, assuming the user has fixed the problem, but the
498 	 * same effect is available with FTS_AGAIN.
499 	 */
500 	if (p->fts_info != FTS_D /* && p->fts_info != FTS_DNR */)
501 		return (NULL);
502 
503 	/* Free up any previous child list. */
504 	if (sp->fts_child)
505 		fts_lfree(sp->fts_child);
506 
507 	if (instr == FTS_NAMEONLY) {
508 		SET(FTS_NAMEONLY);
509 		instr = BNAMES;
510 	} else
511 		instr = BCHILD;
512 
513 	/*
514 	 * If using chdir on a relative path and called BEFORE fts_read does
515 	 * its chdir to the root of a traversal, we can lose -- we need to
516 	 * chdir into the subdirectory, and we don't know where the current
517 	 * directory is, so we can't get back so that the upcoming chdir by
518 	 * fts_read will work.
519 	 */
520 	if (p->fts_level != FTS_ROOTLEVEL || p->fts_accpath[0] == '/' ||
521 	    ISSET(FTS_NOCHDIR))
522 		return (sp->fts_child = fts_build(sp, instr));
523 
524 	if ((fd = open(".", O_RDONLY | O_CLOEXEC)) == -1)
525 		return (NULL);
526 	sp->fts_child = fts_build(sp, instr);
527 	if (fchdir(fd)) {
528 		(void)close(fd);
529 		return (NULL);
530 	}
531 	(void)close(fd);
532 	return (sp->fts_child);
533 }
534 DEF_WEAK(fts_children);
535 
536 /*
537  * This is the tricky part -- do not casually change *anything* in here.  The
538  * idea is to build the linked list of entries that are used by fts_children
539  * and fts_read.  There are lots of special cases.
540  *
541  * The real slowdown in walking the tree is the stat calls.  If FTS_NOSTAT is
542  * set and it's a physical walk (so that symbolic links can't be directories),
543  * we can do things quickly.  First, if it's a 4.4BSD file system, the type
544  * of the file is in the directory entry.  Otherwise, we assume that the number
545  * of subdirectories in a node is equal to the number of links to the parent.
546  * The former skips all stat calls.  The latter skips stat calls in any leaf
547  * directories and for any files after the subdirectories in the directory have
548  * been found, cutting the stat calls by about 2/3.
549  */
550 static FTSENT *
fts_build(FTS * sp,int type)551 fts_build(FTS *sp, int type)
552 {
553 	struct dirent *dp;
554 	FTSENT *p, *head;
555 	FTSENT *cur, *tail;
556 	DIR *dirp;
557 	void *oldaddr;
558 	size_t len, maxlen;
559 	int nitems, cderrno, descend, level, nlinks, nostat, doadjust;
560 	int saved_errno;
561 	char *cp;
562 
563 	/* Set current node pointer. */
564 	cur = sp->fts_cur;
565 
566 	/*
567 	 * Open the directory for reading.  If this fails, we're done.
568 	 * If being called from fts_read, set the fts_info field.
569 	 */
570 	if ((dirp = opendir(cur->fts_accpath)) == NULL) {
571 		if (type == BREAD) {
572 			cur->fts_info = FTS_DNR;
573 			cur->fts_errno = errno;
574 		}
575 		return (NULL);
576 	}
577 
578 	/*
579 	 * Nlinks is the number of possible entries of type directory in the
580 	 * directory if we're cheating on stat calls, 0 if we're not doing
581 	 * any stat calls at all, -1 if we're doing stats on everything.
582 	 */
583 	if (type == BNAMES)
584 		nlinks = 0;
585 	else if (ISSET(FTS_NOSTAT) && ISSET(FTS_PHYSICAL)) {
586 		nlinks = cur->fts_nlink - (ISSET(FTS_SEEDOT) ? 0 : 2);
587 		nostat = 1;
588 	} else {
589 		nlinks = -1;
590 		nostat = 0;
591 	}
592 
593 #ifdef notdef
594 	(void)printf("nlinks == %d (cur: %u)\n", nlinks, cur->fts_nlink);
595 	(void)printf("NOSTAT %d PHYSICAL %d SEEDOT %d\n",
596 	    ISSET(FTS_NOSTAT), ISSET(FTS_PHYSICAL), ISSET(FTS_SEEDOT));
597 #endif
598 	/*
599 	 * If we're going to need to stat anything or we want to descend
600 	 * and stay in the directory, chdir.  If this fails we keep going,
601 	 * but set a flag so we don't chdir after the post-order visit.
602 	 * We won't be able to stat anything, but we can still return the
603 	 * names themselves.  Note, that since fts_read won't be able to
604 	 * chdir into the directory, it will have to return different path
605 	 * names than before, i.e. "a/b" instead of "b".  Since the node
606 	 * has already been visited in pre-order, have to wait until the
607 	 * post-order visit to return the error.  There is a special case
608 	 * here, if there was nothing to stat then it's not an error to
609 	 * not be able to stat.  This is all fairly nasty.  If a program
610 	 * needed sorted entries or stat information, they had better be
611 	 * checking FTS_NS on the returned nodes.
612 	 */
613 	cderrno = 0;
614 	if (nlinks || type == BREAD) {
615 		if (fts_safe_changedir(sp, cur, dirfd(dirp), NULL)) {
616 			if (nlinks && type == BREAD)
617 				cur->fts_errno = errno;
618 			cur->fts_flags |= FTS_DONTCHDIR;
619 			descend = 0;
620 			cderrno = errno;
621 			(void)closedir(dirp);
622 			dirp = NULL;
623 		} else
624 			descend = 1;
625 	} else
626 		descend = 0;
627 
628 	/*
629 	 * Figure out the max file name length that can be stored in the
630 	 * current path -- the inner loop allocates more path as necessary.
631 	 * We really wouldn't have to do the maxlen calculations here, we
632 	 * could do them in fts_read before returning the path, but it's a
633 	 * lot easier here since the length is part of the dirent structure.
634 	 *
635 	 * If not changing directories set a pointer so that can just append
636 	 * each new name into the path.
637 	 */
638 	len = NAPPEND(cur);
639 	if (ISSET(FTS_NOCHDIR)) {
640 		cp = sp->fts_path + len;
641 		*cp++ = '/';
642 	}
643 	len++;
644 	maxlen = sp->fts_pathlen - len;
645 
646 	/*
647 	 * fts_level is signed so we must prevent it from wrapping
648 	 * around to FTS_ROOTLEVEL and FTS_ROOTPARENTLEVEL.
649 	 */
650 	level = cur->fts_level;
651 	if (level < FTS_MAXLEVEL)
652 	    level++;
653 
654 	/* Read the directory, attaching each entry to the `link' pointer. */
655 	doadjust = 0;
656 	for (head = tail = NULL, nitems = 0; dirp && (dp = readdir(dirp));) {
657 		if (!ISSET(FTS_SEEDOT) && ISDOT(dp->d_name))
658 			continue;
659 
660 		if (!(p = fts_alloc(sp, dp->d_name, strlen(dp->d_name))))
661 			goto mem1;
662 		if (strlen(dp->d_name) >= maxlen) {	/* include space for NUL */
663 			oldaddr = sp->fts_path;
664 			if (fts_palloc(sp, strlen(dp->d_name) +len + 1)) {
665 				/*
666 				 * No more memory for path or structures.  Save
667 				 * errno, free up the current structure and the
668 				 * structures already allocated.
669 				 */
670 mem1:				saved_errno = errno;
671 				free(p);
672 				fts_lfree(head);
673 				(void)closedir(dirp);
674 				cur->fts_info = FTS_ERR;
675 				SET(FTS_STOP);
676 				errno = saved_errno;
677 				return (NULL);
678 			}
679 			/* Did realloc() change the pointer? */
680 			if (oldaddr != sp->fts_path) {
681 				doadjust = 1;
682 				if (ISSET(FTS_NOCHDIR))
683 					cp = sp->fts_path + len;
684 			}
685 			maxlen = sp->fts_pathlen - len;
686 		}
687 
688 		p->fts_level = level;
689 		p->fts_parent = sp->fts_cur;
690 		p->fts_pathlen = len + strlen(dp->d_name);
691 		if (p->fts_pathlen < len) {
692 			/*
693 			 * If we wrap, free up the current structure and
694 			 * the structures already allocated, then error
695 			 * out with ENAMETOOLONG.
696 			 */
697 			free(p);
698 			fts_lfree(head);
699 			(void)closedir(dirp);
700 			cur->fts_info = FTS_ERR;
701 			SET(FTS_STOP);
702 			errno = ENAMETOOLONG;
703 			return (NULL);
704 		}
705 
706 		if (cderrno) {
707 			if (nlinks) {
708 				p->fts_info = FTS_NS;
709 				p->fts_errno = cderrno;
710 			} else
711 				p->fts_info = FTS_NSOK;
712 			p->fts_accpath = cur->fts_accpath;
713 		} else if (nlinks == 0
714 #ifdef DT_DIR
715 		    || (nostat &&
716 		    dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN)
717 #endif
718 		    ) {
719 			p->fts_accpath =
720 			    ISSET(FTS_NOCHDIR) ? p->fts_path : p->fts_name;
721 			p->fts_info = FTS_NSOK;
722 		} else {
723 			/* Build a file name for fts_stat to stat. */
724 			if (ISSET(FTS_NOCHDIR)) {
725 				p->fts_accpath = p->fts_path;
726 				memmove(cp, p->fts_name, p->fts_namelen + 1);
727 				p->fts_info = fts_stat(sp, p, 0, dirfd(dirp));
728 			} else {
729 				p->fts_accpath = p->fts_name;
730 				p->fts_info = fts_stat(sp, p, 0, -1);
731 			}
732 
733 			/* Decrement link count if applicable. */
734 			if (nlinks > 0 && (p->fts_info == FTS_D ||
735 			    p->fts_info == FTS_DC || p->fts_info == FTS_DOT))
736 				--nlinks;
737 		}
738 
739 		/* We walk in directory order so "ls -f" doesn't get upset. */
740 		p->fts_link = NULL;
741 		if (head == NULL)
742 			head = tail = p;
743 		else {
744 			tail->fts_link = p;
745 			tail = p;
746 		}
747 		++nitems;
748 	}
749 	if (dirp)
750 		(void)closedir(dirp);
751 
752 	/*
753 	 * If realloc() changed the address of the path, adjust the
754 	 * addresses for the rest of the tree and the dir list.
755 	 */
756 	if (doadjust)
757 		fts_padjust(sp, head);
758 
759 	/*
760 	 * If not changing directories, reset the path back to original
761 	 * state.
762 	 */
763 	if (ISSET(FTS_NOCHDIR)) {
764 		if (len == sp->fts_pathlen || nitems == 0)
765 			--cp;
766 		*cp = '\0';
767 	}
768 
769 	/*
770 	 * If descended after called from fts_children or after called from
771 	 * fts_read and nothing found, get back.  At the root level we use
772 	 * the saved fd; if one of fts_open()'s arguments is a relative path
773 	 * to an empty directory, we wind up here with no other way back.  If
774 	 * can't get back, we're done.
775 	 */
776 	if (descend && (type == BCHILD || !nitems) &&
777 	    (cur->fts_level == FTS_ROOTLEVEL ? FCHDIR(sp, sp->fts_rfd) :
778 	    fts_safe_changedir(sp, cur->fts_parent, -1, ".."))) {
779 		cur->fts_info = FTS_ERR;
780 		SET(FTS_STOP);
781 		return (NULL);
782 	}
783 
784 	/* If didn't find anything, return NULL. */
785 	if (!nitems) {
786 		if (type == BREAD)
787 			cur->fts_info = FTS_DP;
788 		return (NULL);
789 	}
790 
791 	/* Sort the entries. */
792 	if (sp->fts_compar && nitems > 1)
793 		head = fts_sort(sp, head, nitems);
794 	return (head);
795 }
796 
797 static u_short
fts_stat(FTS * sp,FTSENT * p,int follow,int dfd)798 fts_stat(FTS *sp, FTSENT *p, int follow, int dfd)
799 {
800 	FTSENT *t;
801 	dev_t dev;
802 	ino_t ino;
803 	struct stat *sbp, sb;
804 	int saved_errno;
805 	const char *path;
806 
807 	if (dfd == -1) {
808 		path = p->fts_accpath;
809 		dfd = AT_FDCWD;
810 	} else
811 		path = p->fts_name;
812 
813 	/* If user needs stat info, stat buffer already allocated. */
814 	sbp = ISSET(FTS_NOSTAT) ? &sb : p->fts_statp;
815 
816 	/*
817 	 * If doing a logical walk, or application requested FTS_FOLLOW, do
818 	 * a stat(2).  If that fails, check for a non-existent symlink.  If
819 	 * fail, set the errno from the stat call.
820 	 */
821 	if (ISSET(FTS_LOGICAL) || follow) {
822 		if (fstatat(dfd, path, sbp, 0)) {
823 			saved_errno = errno;
824 			if (!fstatat(dfd, path, sbp, AT_SYMLINK_NOFOLLOW)) {
825 				errno = 0;
826 				return (FTS_SLNONE);
827 			}
828 			p->fts_errno = saved_errno;
829 			goto err;
830 		}
831 	} else if (fstatat(dfd, path, sbp, AT_SYMLINK_NOFOLLOW)) {
832 		p->fts_errno = errno;
833 err:		memset(sbp, 0, sizeof(struct stat));
834 		return (FTS_NS);
835 	}
836 
837 	if (S_ISDIR(sbp->st_mode)) {
838 		/*
839 		 * Set the device/inode.  Used to find cycles and check for
840 		 * crossing mount points.  Also remember the link count, used
841 		 * in fts_build to limit the number of stat calls.  It is
842 		 * understood that these fields are only referenced if fts_info
843 		 * is set to FTS_D.
844 		 */
845 		dev = p->fts_dev = sbp->st_dev;
846 		ino = p->fts_ino = sbp->st_ino;
847 		p->fts_nlink = sbp->st_nlink;
848 
849 		if (ISDOT(p->fts_name))
850 			return (FTS_DOT);
851 
852 		/*
853 		 * Cycle detection is done by brute force when the directory
854 		 * is first encountered.  If the tree gets deep enough or the
855 		 * number of symbolic links to directories is high enough,
856 		 * something faster might be worthwhile.
857 		 */
858 		for (t = p->fts_parent;
859 		    t->fts_level >= FTS_ROOTLEVEL; t = t->fts_parent)
860 			if (ino == t->fts_ino && dev == t->fts_dev) {
861 				p->fts_cycle = t;
862 				return (FTS_DC);
863 			}
864 		return (FTS_D);
865 	}
866 	if (S_ISLNK(sbp->st_mode))
867 		return (FTS_SL);
868 	if (S_ISREG(sbp->st_mode))
869 		return (FTS_F);
870 	return (FTS_DEFAULT);
871 }
872 
873 static FTSENT *
fts_sort(FTS * sp,FTSENT * head,int nitems)874 fts_sort(FTS *sp, FTSENT *head, int nitems)
875 {
876 	FTSENT **ap, *p;
877 
878 	/*
879 	 * Construct an array of pointers to the structures and call qsort(3).
880 	 * Reassemble the array in the order returned by qsort.  If unable to
881 	 * sort for memory reasons, return the directory entries in their
882 	 * current order.  Allocate enough space for the current needs plus
883 	 * 40 so don't realloc one entry at a time.
884 	 */
885 	if (nitems > sp->fts_nitems) {
886 		struct _ftsent **a;
887 
888 		if ((a = reallocarray(sp->fts_array,
889 		    nitems + 40, sizeof(FTSENT *))) == NULL) {
890 			free(sp->fts_array);
891 			sp->fts_array = NULL;
892 			sp->fts_nitems = 0;
893 			return (head);
894 		}
895 		sp->fts_nitems = nitems + 40;
896 		sp->fts_array = a;
897 	}
898 	for (ap = sp->fts_array, p = head; p; p = p->fts_link)
899 		*ap++ = p;
900 	qsort(sp->fts_array, nitems, sizeof(FTSENT *), sp->fts_compar);
901 	for (head = *(ap = sp->fts_array); --nitems; ++ap)
902 		ap[0]->fts_link = ap[1];
903 	ap[0]->fts_link = NULL;
904 	return (head);
905 }
906 
907 static FTSENT *
fts_alloc(FTS * sp,const char * name,size_t namelen)908 fts_alloc(FTS *sp, const char *name, size_t namelen)
909 {
910 	FTSENT *p;
911 	size_t len;
912 
913 	/*
914 	 * The file name is a variable length array and no stat structure is
915 	 * necessary if the user has set the nostat bit.  Allocate the FTSENT
916 	 * structure, the file name and the stat structure in one chunk, but
917 	 * be careful that the stat structure is reasonably aligned.  Since the
918 	 * fts_name field is declared to be of size 1, the fts_name pointer is
919 	 * namelen + 2 before the first possible address of the stat structure.
920 	 */
921 	len = sizeof(FTSENT) + namelen;
922 	if (!ISSET(FTS_NOSTAT))
923 		len += sizeof(struct stat) + ALIGNBYTES;
924 	if ((p = calloc(1, len)) == NULL)
925 		return (NULL);
926 
927 	p->fts_path = sp->fts_path;
928 	p->fts_namelen = namelen;
929 	p->fts_instr = FTS_NOINSTR;
930 	if (!ISSET(FTS_NOSTAT))
931 		p->fts_statp = (struct stat *)ALIGN(p->fts_name + namelen + 2);
932 	memcpy(p->fts_name, name, namelen);
933 
934 	return (p);
935 }
936 
937 static void
fts_lfree(FTSENT * head)938 fts_lfree(FTSENT *head)
939 {
940 	FTSENT *p;
941 
942 	/* Free a linked list of structures. */
943 	while ((p = head)) {
944 		head = head->fts_link;
945 		free(p);
946 	}
947 }
948 
949 /*
950  * Allow essentially unlimited paths; find, rm, ls should all work on any tree.
951  * Most systems will allow creation of paths much longer than PATH_MAX, even
952  * though the kernel won't resolve them.  Add the size (not just what's needed)
953  * plus 256 bytes so don't realloc the path 2 bytes at a time.
954  */
955 static int
fts_palloc(FTS * sp,size_t more)956 fts_palloc(FTS *sp, size_t more)
957 {
958 	char *p;
959 
960 	/*
961 	 * Check for possible wraparound.
962 	 */
963 	more += 256;
964 	if (sp->fts_pathlen + more < sp->fts_pathlen) {
965 		free(sp->fts_path);
966 		sp->fts_path = NULL;
967 		errno = ENAMETOOLONG;
968 		return (1);
969 	}
970 	p = recallocarray(sp->fts_path, sp->fts_pathlen,
971 	    sp->fts_pathlen + more, 1);
972 	if (p == NULL) {
973 		free(sp->fts_path);
974 		sp->fts_path = NULL;
975 		return (1);
976 	}
977 	sp->fts_pathlen += more;
978 	sp->fts_path = p;
979 	return (0);
980 }
981 
982 /*
983  * When the path is realloc'd, have to fix all of the pointers in structures
984  * already returned.
985  */
986 static void
fts_padjust(FTS * sp,FTSENT * head)987 fts_padjust(FTS *sp, FTSENT *head)
988 {
989 	FTSENT *p;
990 	char *addr = sp->fts_path;
991 
992 #define	ADJUST(p) {							\
993 	if ((p)->fts_accpath != (p)->fts_name) {			\
994 		(p)->fts_accpath =					\
995 		    (char *)addr + ((p)->fts_accpath - (p)->fts_path);	\
996 	}								\
997 	(p)->fts_path = addr;						\
998 }
999 	/* Adjust the current set of children. */
1000 	for (p = sp->fts_child; p; p = p->fts_link)
1001 		ADJUST(p);
1002 
1003 	/* Adjust the rest of the tree, including the current level. */
1004 	for (p = head; p->fts_level >= FTS_ROOTLEVEL;) {
1005 		ADJUST(p);
1006 		p = p->fts_link ? p->fts_link : p->fts_parent;
1007 	}
1008 }
1009 
1010 static size_t
fts_maxarglen(char * const * argv)1011 fts_maxarglen(char * const *argv)
1012 {
1013 	size_t len, max;
1014 
1015 	for (max = 0; *argv; ++argv)
1016 		if ((len = strlen(*argv)) > max)
1017 			max = len;
1018 	return (max + 1);
1019 }
1020 
1021 /*
1022  * Change to dir specified by fd or p->fts_accpath without getting
1023  * tricked by someone changing the world out from underneath us.
1024  * Assumes p->fts_dev and p->fts_ino are filled in.
1025  */
1026 static int
fts_safe_changedir(FTS * sp,FTSENT * p,int fd,const char * path)1027 fts_safe_changedir(FTS *sp, FTSENT *p, int fd, const char *path)
1028 {
1029 	int ret, oerrno, newfd;
1030 	struct stat sb;
1031 
1032 	newfd = fd;
1033 	if (ISSET(FTS_NOCHDIR))
1034 		return (0);
1035 	if (fd == -1 && (newfd = open(path, O_RDONLY|O_DIRECTORY|O_CLOEXEC)) == -1)
1036 		return (-1);
1037 	if (fstat(newfd, &sb) == -1) {
1038 		ret = -1;
1039 		goto bail;
1040 	}
1041 	if (p->fts_dev != sb.st_dev || p->fts_ino != sb.st_ino) {
1042 		errno = ENOENT;		/* disinformation */
1043 		ret = -1;
1044 		goto bail;
1045 	}
1046 	ret = fchdir(newfd);
1047 bail:
1048 	oerrno = errno;
1049 	if (fd == -1)
1050 		(void)close(newfd);
1051 	errno = oerrno;
1052 	return (ret);
1053 }
1054 
1055 FTS *
fts_open(char * const * argv,int options,int (* compar)(const FTSENT **,const FTSENT **))1056 fts_open(char * const *argv, int options,
1057     int (*compar)(const FTSENT **, const FTSENT **))
1058 {
1059 	// Android needs to an __fts_open() that doesn't make this check
1060 	// so that FTS_FOR_FTW works.
1061 	if (options & ~FTS_OPTIONMASK) {
1062 		errno = EINVAL;
1063 		return (NULL);
1064 	}
1065 	return __fts_open(argv, options, compar);
1066 }
1067