xref: /illumos-gate/usr/src/lib/libc/port/gen/nftw.c (revision 7c478bd95313f5f23a4c958a745db2134aa03244)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 /*	Copyright (c) 1988 AT&T	*/
30 /*	  All Rights Reserved  	*/
31 
32 
33 /*
34  *	nftw - new file tree walk
35  *
36  *	int nftw(char *path, int (*fn)(), int depth, int flags);
37  *
38  *	Derived from System V ftw() by David Korn
39  *
40  *	nftw visits each file and directory in the tree starting at
41  *	path. It uses the generic directory reading library so it works
42  *	for any file system type.  The flags field is used to specify:
43  *		FTW_PHYS  Physical walk, does not follow symblolic links
44  *			  Otherwise, nftw will follow links but will not
45  *			  walk down any path the crosses itself.
46  *		FTW_MOUNT The walk will not cross a mount point.
47  *		FTW_DEPTH All subdirectories will be visited before the
48  *			  directory itself.
49  *		FTW_CHDIR The walk will change to each directory before
50  *			  reading it.  This is faster but core dumps
51  *			  may not get generated.
52  *
53  *	The following flags are private, and are used by the find
54  *	utility:
55  *		FTW_ANYERR Call the callback function and return
56  *			   FTW_NS on any stat failure, not just
57  *			   lack of permission.
58  *		FTW_HOPTION Use stat the first time the walk
59  *			    function is called, regardless of
60  *			    whether or not FTW_PHYS is specified.
61  *
62  *	fn is called with four arguments at each file and directory.
63  *	The first argument is the pathname of the object, the second
64  *	is a pointer to the stat buffer and the third is an integer
65  *	giving additional information as follows:
66  *
67  *		FTW_F	The object is a file.
68  *		FTW_D	The object is a directory.
69  *		FTW_DP	The object is a directory and subdirectories
70  *			have been visited.
71  *		FTW_SL	The object is a symbolic link.
72  *		FTW_SLN The object is a symbolic link pointing at a
73  *		        non-existing file.
74  *		FTW_DNR	The object is a directory that cannot be read.
75  *			fn will not be called for any of its descendants.
76  *		FTW_NS	Stat failed on the object because of lack of
77  *			appropriate permission. The stat buffer passed to fn
78  *			is undefined.  Stat failure for any reason is
79  *			considered an error and nftw will return -1.
80  *	The fourth argument is a struct FTW* which contains the depth
81  *	and the offset into pathname to the base name.
82  *	If fn returns nonzero, nftw returns this value to its caller.
83  *
84  *	depth limits the number of open directories that ftw uses
85  *	before it starts recycling file descriptors.  In general,
86  *	a file descriptor is used for each level.
87  *
88  */
89 
90 #include <sys/feature_tests.h>
91 
92 #if !defined(_LP64) && _FILE_OFFSET_BITS == 64
93 #pragma weak nftw64 = _nftw64
94 #define	_nftw		_nftw64
95 #define	fstat64		_fstat64
96 #define	lstat64		_lstat64
97 #define	readdir64	_readdir64
98 #define	stat64		_stat64
99 #else
100 #pragma weak nftw = _nftw
101 #define	fstat		_fstat
102 #define	lstat		_lstat
103 #define	readdir		_readdir
104 #define	stat		_stat
105 #endif /* !_LP64 && _FILE_OFFSET_BITS == 64 */
106 
107 #define	chdir		_chdir
108 #define	closedir	_closedir
109 #define	fchdir		_fchdir
110 #define	fprintf		_fprintf
111 #define	getcwd		_getcwd
112 #define	opendir		_opendir
113 #define	seekdir		_seekdir
114 #define	telldir		_telldir
115 
116 #include "lint.h"
117 #include <mtlib.h>
118 #include <sys/types.h>
119 #include <sys/stat.h>
120 #include <dirent.h>
121 #include <errno.h>
122 #include <limits.h>
123 #include <ftw.h>
124 #include <stdlib.h>
125 #include <string.h>
126 #include <unistd.h>
127 #include <thread.h>
128 #include <synch.h>
129 #include <stdio.h>
130 
131 #ifndef PATH_MAX
132 #define	PATH_MAX	1023
133 #endif
134 
135 /*
136  * Local variables (used to be static local).
137  * Putting them into a structure that is passed
138  * around makes nftw() MT-safe with no locking required.
139  */
140 struct Var {
141 	char	*fullpath;
142 	char	*tmppath;
143 	int	curflags;
144 	dev_t	cur_mount;
145 	struct FTW state;
146 	int	walklevel;
147 	int	(*statf)(const char *, struct stat *);
148 	int	(*savedstatf)(const char *, struct stat *);
149 };
150 
151 struct Save {
152 	struct Save *last;
153 	DIR	*fd;
154 	char	*comp;
155 	long	here;
156 	dev_t	dev;
157 	ino_t	inode;
158 };
159 
160 static int oldclose(struct Save *);
161 
162 /*
163  * This is the recursive walker.
164  */
165 static int
166 walk(char *component,
167     int (*fn)(const char *, const struct stat *, int, struct FTW *),
168     int depth, struct Save *last, struct Var *vp)
169 {
170 	struct stat statb;
171 	char *p;
172 	int type;
173 	char *comp;
174 	struct dirent *dir;
175 	char *q;
176 	int rc = 0;
177 	int val = -1;
178 	int cdval = -1;
179 	int oldbase;
180 	int skip;
181 	struct Save this;
182 
183 	this.last = last;
184 	this.fd = 0;
185 	if ((vp->curflags & FTW_CHDIR) && last)
186 		comp = last->comp;
187 	else
188 		comp = vp->tmppath;
189 
190 	if (vp->savedstatf == NULL)
191 		vp->savedstatf = vp->statf;
192 
193 	if ((vp->walklevel++ == 0) && (vp->curflags & FTW_HOPTION))
194 		vp->statf = stat;
195 	else
196 		vp->statf = vp->savedstatf;
197 
198 	/*
199 	 * Determine the type of the component.
200 	 */
201 	if ((*vp->statf)(comp, &statb) >= 0) {
202 		if ((statb.st_mode & S_IFMT) == S_IFDIR) {
203 			type = FTW_D;
204 			if (depth <= 1)
205 				(void) oldclose(last);
206 			if ((this.fd = opendir(comp)) == 0) {
207 				if (errno == EMFILE && oldclose(last) &&
208 				    (this.fd = opendir(comp)) != 0) {
209 					depth = 1;
210 				} else {
211 					type = FTW_DNR;
212 					goto fail;
213 				}
214 			}
215 			if (statb.st_fstype[0] == 'a' &&
216 			    strcmp(statb.st_fstype, "autofs") == 0) {
217 				/*
218 				 * this dir is on autofs
219 				 */
220 				if (fstat(this.fd->dd_fd, &statb) < 0) {
221 					(void) closedir(this.fd);
222 					type = FTW_NS;
223 					goto fail;
224 				}
225 			}
226 		} else if ((statb.st_mode & S_IFMT) == S_IFLNK) {
227 			type = FTW_SL;
228 		} else {
229 			type = FTW_F;
230 		}
231 	} else if ((vp->curflags & FTW_ANYERR) && errno != ENOENT) {
232 		/*
233 		 * If FTW_ANYERR is specified, then a stat error
234 		 * other than ENOENT automatically results in
235 		 * failure.  This allows the callback function
236 		 * to properly handle ENAMETOOLONG and ELOOP and
237 		 * things of that nature, that would be masked
238 		 * by calling lstat before failing.
239 		 */
240 		type = FTW_NS;
241 		goto fail;
242 	} else {
243 		/*
244 		 * Statf has failed. If stat was used instead of lstat,
245 		 * try using lstat. If lstat doesn't fail, "comp"
246 		 * must be a symbolic link pointing to a non-existent
247 		 * file. Such a symbolic link should be ignored.
248 		 * Also check the file type, if possible, for symbolic
249 		 * link.
250 		 */
251 		if ((vp->statf == stat) && (lstat(comp, &statb) >= 0) &&
252 		    ((statb.st_mode & S_IFMT) == S_IFLNK)) {
253 
254 			/*
255 			 * Ignore bad symbolic link, let "fn"
256 			 * report it.
257 			 */
258 
259 			errno = ENOENT;
260 			type = FTW_SLN;
261 		} else {
262 			type = FTW_NS;
263 	fail:
264 			/*
265 			 * if FTW_ANYERR is set in flags, we call
266 			 * the user function with FTW_NS set, regardless
267 			 * of the reason stat failed.
268 			 */
269 			if (!(vp->curflags & FTW_ANYERR))
270 				if (errno != EACCES)
271 					return (-1);
272 		}
273 	}
274 
275 	/*
276 	 * If the walk is not supposed to cross a mount point,
277 	 * and it did, get ready to return.
278 	 */
279 	if ((vp->curflags & FTW_MOUNT) && type != FTW_NS &&
280 	    statb.st_dev != vp->cur_mount)
281 		goto quit;
282 	vp->state.quit = 0;
283 
284 	/*
285 	 * If current component is not a directory, call user
286 	 * specified function and get ready to return.
287 	 */
288 	if (type != FTW_D || (vp->curflags & FTW_DEPTH) == 0)
289 		rc = (*fn)(vp->tmppath, &statb, type, &vp->state);
290 	if (rc > 0)
291 		val = rc;
292 	skip = (vp->state.quit & FTW_SKD);
293 	if (rc != 0 || type != FTW_D || (vp->state.quit & FTW_PRUNE))
294 		goto quit;
295 
296 	if (vp->tmppath[0] != '\0' && component[-1] != '/')
297 		*component++ = '/';
298 	if (vp->curflags & FTW_CHDIR) {
299 		struct stat statb2;
300 
301 		*component = 0;
302 		/*
303 		 * Security check (there is a window between
304 		 * (*vp->statf)() and opendir() above).
305 		 */
306 		if ((vp->curflags & FTW_PHYS) &&
307 		    (fstat(this.fd->dd_fd, &statb2) < 0 ||
308 		    statb2.st_ino != statb.st_ino ||
309 		    statb2.st_dev != statb.st_dev)) {
310 			errno = EAGAIN;
311 			rc = -1;
312 			goto quit;
313 		}
314 
315 		if ((cdval = fchdir(this.fd->dd_fd)) >= 0) {
316 			this.comp = component;
317 		} else {
318 			type = FTW_DNR;
319 			rc = (*fn)(vp->tmppath, &statb, type, &vp->state);
320 			goto quit;
321 		}
322 	}
323 
324 	/*
325 	 * If the walk has followed a symbolic link, traverse
326 	 * the walk back to make sure there is not a loop.
327 	 *
328 	 * XXX - may need to look at this
329 	 * There's code to check for cycles, but only for FTW_PHYS
330 	 * (find -L flag).  However, all directories should be
331 	 * checked, even if not following links because of hardlinks
332 	 * to directories (not recommended, but can exist).
333 	 *
334 	 * We might have added AVL tree routines here to store and search
335 	 * the inodes and devices, as is done for du/ls/chgrp/chown,
336 	 * but libcmdutils is for for internal use only, so we can't
337 	 * add it to a public libc function (nftw()).
338 	 */
339 	if ((vp->curflags & FTW_PHYS) == 0) {
340 		struct Save *sp = last;
341 		while (sp) {
342 			/*
343 			 * If the same node has already been visited, there
344 			 * is a loop. Get ready to return.
345 			 */
346 			if (sp->dev == statb.st_dev &&
347 			    sp->inode == statb.st_ino)
348 				goto quit;
349 			sp = sp->last;
350 		}
351 	}
352 	this.dev = statb.st_dev;
353 	this.inode = statb.st_ino;
354 	oldbase = vp->state.base;
355 	vp->state.base = (int)(component - vp->tmppath);
356 	while (dir = readdir(this.fd)) {
357 		if (dir->d_ino == 0)
358 			continue;
359 		q = dir->d_name;
360 		if (*q == '.') {
361 			if (q[1] == 0)
362 				continue;
363 			else if (q[1] == '.' && q[2] == 0)
364 				continue;
365 		}
366 		p = component;
367 		while (p < &vp->tmppath[PATH_MAX] && *q != '\0')
368 			*p++ = *q++;
369 		*p = '\0';
370 		vp->state.level++;
371 
372 		/* Call walk() recursively.  */
373 		rc = walk(p, fn, depth-1, &this, vp);
374 		vp->state.level--;
375 		if (this.fd == 0) {
376 			*component = 0;
377 			if (vp->curflags & FTW_CHDIR) {
378 				this.fd = opendir(".");
379 			} else {
380 				this.fd = opendir(comp);
381 			}
382 			if (this.fd == 0) {
383 				rc = -1;
384 				goto quit;
385 			}
386 			seekdir(this.fd, this.here);
387 		}
388 		if (rc != 0) {
389 			if (errno == ENOENT) {
390 				(void) fprintf(stderr, "cannot open %s: %s\n",
391 				    vp->tmppath, strerror(errno));
392 				val = rc;
393 				continue;
394 			}
395 			goto quit;	/* this seems extreme */
396 		}
397 	}
398 	vp->state.base = oldbase;
399 	*--component = 0;
400 	type = FTW_DP;
401 	if ((vp->tmppath[0] != '\0') && (vp->curflags & FTW_DEPTH) && !skip)
402 		rc = (*fn)(vp->tmppath, &statb, type, &vp->state);
403 quit:
404 	if (cdval >= 0 && last) {
405 		/* try to change back to previous directory */
406 		if (last->fd != NULL) {
407 			if (fchdir(last->fd->dd_fd) < 0) {
408 				rc = -1;
409 			}
410 		} else {
411 			if ((cdval = chdir("..")) >= 0) {
412 				if ((*vp->statf)(".", &statb) < 0 ||
413 				    statb.st_ino != last->inode ||
414 				    statb.st_dev != last->dev)
415 					cdval = -1;
416 			}
417 			*comp = 0;
418 			if (cdval < 0) {
419 				if (chdir(vp->fullpath) < 0) {
420 					rc = -1;
421 				} else {
422 					/* Security check */
423 					if ((vp->curflags & FTW_PHYS) &&
424 					    ((*vp->statf)(".", &statb) < 0 ||
425 					    statb.st_ino != last->inode ||
426 					    statb.st_dev != last->dev)) {
427 						errno = EAGAIN;
428 						rc = -1;
429 					}
430 				}
431 			}
432 		}
433 	}
434 	if (this.fd)
435 		(void) closedir(this.fd);
436 	if (val > rc)
437 		return (val);
438 	else
439 		return (rc);
440 }
441 
442 int
443 _nftw(const char *path,
444     int (*fn)(const char *, const struct stat *, int, struct FTW *),
445     int depth, int flags)
446 {
447 	struct Var var;
448 	struct stat statb;
449 	char home[2*(PATH_MAX+1)];
450 	int rc = -1;
451 	char *dp;
452 	char *base;
453 	char *endhome;
454 	const char *savepath = path;
455 	int save_errno;
456 
457 	home[0] = 0;
458 
459 	/*
460 	 * If the walk is going to change directory before
461 	 * reading it, save current woring directory.
462 	 */
463 	if (flags & FTW_CHDIR) {
464 		if (getcwd(home, PATH_MAX+1) == 0)
465 			return (-1);
466 	}
467 	endhome = dp = home + strlen(home);
468 	if (*path == '/')
469 		var.fullpath = dp;
470 	else {
471 		*dp++ = '/';
472 		var.fullpath = home;
473 	}
474 	var.tmppath =  dp;
475 	base = dp-1;
476 	while (*path && dp < &var.tmppath[PATH_MAX]) {
477 		*dp = *path;
478 		if (*dp == '/')
479 			base = dp;
480 		dp++, path++;
481 	}
482 	*dp = 0;
483 	var.state.base = (int)(base + 1 - var.tmppath);
484 	if (*path) {
485 		errno = ENAMETOOLONG;
486 		return (-1);
487 	}
488 	var.curflags = flags;
489 
490 	/*
491 	 * If doing a physical walk (not following symbolic link), set
492 	 * var.statf to lstat(). Otherwise, set var.statf to stat().
493 	 */
494 	if ((flags & FTW_PHYS) == 0)
495 		var.statf = stat;
496 	else
497 		var.statf = lstat;
498 
499 	/*
500 	 * If walk is not going to cross a mount point,
501 	 * save the current mount point.
502 	 */
503 	if (flags & FTW_MOUNT) {
504 		if ((*var.statf)(savepath, &statb) >= 0)
505 			var.cur_mount = statb.st_dev;
506 		else
507 			goto done;
508 	}
509 	var.state.level = 0;
510 
511 	/*
512 	 * Call walk() which does most of the work.
513 	 * walk() uses errno in a rather obtuse way
514 	 * so we shield any incoming errno.
515 	 */
516 	save_errno = errno;
517 	errno = 0;
518 	var.savedstatf = NULL;
519 	var.walklevel = 0;
520 	rc = walk(dp, fn, depth, (struct Save *)0, &var);
521 	if (errno == 0)
522 		errno = save_errno;
523 done:
524 	*endhome = 0;
525 	if (flags & FTW_CHDIR)
526 		(void) chdir(home);
527 	return (rc);
528 }
529 
530 /*
531  * close the oldest directory.  It saves the seek offset.
532  * return value is 0 unless it was unable to close any descriptor
533  */
534 
535 static int
536 oldclose(struct Save *sp)
537 {
538 	struct Save *spnext;
539 	while (sp) {
540 		spnext = sp->last;
541 		if (spnext == 0 || spnext->fd == 0)
542 			break;
543 		sp = spnext;
544 	}
545 	if (sp == 0 || sp->fd == 0)
546 		return (0);
547 	sp->here = telldir(sp->fd);
548 	(void) closedir(sp->fd);
549 	sp->fd = 0;
550 	return (1);
551 }
552