xref: /netbsd-src/lib/libc/gen/opendir.c (revision ce099b40997c43048fb78bd578195f81d2456523)
1 /*	$NetBSD: opendir.c,v 1.33 2008/01/10 09:49:04 elad Exp $	*/
2 
3 /*
4  * Copyright (c) 1983, 1993
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/cdefs.h>
33 #if defined(LIBC_SCCS) && !defined(lint)
34 #if 0
35 static char sccsid[] = "@(#)opendir.c	8.7 (Berkeley) 12/10/94";
36 #else
37 __RCSID("$NetBSD: opendir.c,v 1.33 2008/01/10 09:49:04 elad Exp $");
38 #endif
39 #endif /* LIBC_SCCS and not lint */
40 
41 #include "namespace.h"
42 #include "reentrant.h"
43 #include "extern.h"
44 #include <sys/param.h>
45 #include <sys/mount.h>
46 #include <sys/stat.h>
47 
48 #include <assert.h>
49 #include <dirent.h>
50 #include <errno.h>
51 #include <fcntl.h>
52 #include <stdlib.h>
53 #include <string.h>
54 #include <unistd.h>
55 
56 #include "dirent_private.h"
57 
58 #define	MAXITERATIONS	100
59 
60 /*
61  * Open a directory.
62  */
63 DIR *
64 opendir(const char *name)
65 {
66 
67 	_DIAGASSERT(name != NULL);
68 
69 	return (__opendir2(name, DTF_HIDEW|DTF_NODUP));
70 }
71 
72 DIR *
73 __opendir2(const char *name, int flags)
74 {
75 	DIR *dirp = NULL;
76 	int fd;
77 	int serrno;
78 	struct stat sb;
79 	int pagesz;
80 	int incr;
81 	int unionstack, nfsdir;
82 	struct statvfs sfb;
83 
84 	_DIAGASSERT(name != NULL);
85 
86 	if ((fd = open(name, O_RDONLY | O_NONBLOCK)) == -1 ||
87 	    fcntl(fd, F_SETFD, FD_CLOEXEC) == -1)
88 		goto error;
89 	if (fstat(fd, &sb) || !S_ISDIR(sb.st_mode)) {
90 		errno = ENOTDIR;
91 		goto error;
92 	}
93 	if ((dirp = (DIR *)malloc(sizeof(DIR))) == NULL)
94 		goto error;
95 	dirp->dd_buf = NULL;
96 
97 	/*
98 	 * If the machine's page size is an exact multiple of DIRBLKSIZ,
99 	 * use a buffer that is cluster boundary aligned.
100 	 * Hopefully this can be a big win someday by allowing page trades
101 	 * to user space to be done by getdirentries()
102 	 */
103 	if (((pagesz = getpagesize()) % DIRBLKSIZ) == 0)
104 		incr = pagesz;
105 	else
106 		incr = DIRBLKSIZ;
107 
108 	/*
109 	 * Determine whether this directory is the top of a union stack.
110 	 */
111 
112 	if (fstatvfs1(fd, &sfb, ST_NOWAIT) < 0)
113 		goto error;
114 
115 	if (flags & DTF_NODUP)
116 		unionstack = !(strncmp(sfb.f_fstypename, MOUNT_UNION,
117 		    sizeof(sfb.f_fstypename))) || (sfb.f_flag & MNT_UNION);
118 	else
119 		unionstack = 0;
120 
121 	nfsdir = !(strncmp(sfb.f_fstypename, MOUNT_NFS, sizeof(sfb.f_fstypename)));
122 
123 	if (unionstack || nfsdir) {
124 		size_t len;
125 		size_t space;
126 		char *buf, *nbuf;
127 		char *ddptr;
128 		char *ddeptr;
129 		int n;
130 		struct dirent **dpv;
131 		int i;
132 
133 		/*
134 		 * The strategy here for directories on top of a union stack
135 		 * is to read all the directory entries into a buffer, sort
136 		 * the buffer, and remove duplicate entries by setting the
137 		 * inode number to zero.
138 		 *
139 		 * For directories on an NFS mounted filesystem, we try
140 	 	 * to get a consistent snapshot by trying until we have
141 		 * successfully read all of the directory without errors
142 		 * (i.e. 'bad cookie' errors from the server because
143 		 * the directory was modified). These errors should not
144 		 * happen often, but need to be dealt with.
145 		 */
146 		i = 0;
147 retry:
148 		len = 0;
149 		space = 0;
150 		buf = 0;
151 		ddptr = 0;
152 
153 		do {
154 			/*
155 			 * Always make at least DIRBLKSIZ bytes
156 			 * available to getdirentries
157 			 */
158 			if (space < DIRBLKSIZ) {
159 				space += incr;
160 				len += incr;
161 				nbuf = realloc(buf, len);
162 				if (nbuf == NULL) {
163 					dirp->dd_buf = buf;
164 					goto error;
165 				}
166 				buf = nbuf;
167 				ddptr = buf + (len - space);
168 			}
169 
170 			dirp->dd_seek = lseek(fd, (off_t)0, SEEK_CUR);
171 			n = getdents(fd, ddptr, space);
172 			/*
173 			 * For NFS: EINVAL means a bad cookie error
174 			 * from the server. Keep trying to get a
175 			 * consistent view, in this case this means
176 			 * starting all over again.
177 			 */
178 			if (n == -1 && errno == EINVAL && nfsdir) {
179 				free(buf);
180 				lseek(fd, (off_t)0, SEEK_SET);
181 				if (++i > MAXITERATIONS)
182 					goto error;
183 				goto retry;
184 			}
185 			if (n > 0) {
186 				ddptr += n;
187 				space -= n;
188 			}
189 		} while (n > 0);
190 
191 		ddeptr = ddptr;
192 		flags |= __DTF_READALL;
193 
194 		/*
195 		 * Re-open the directory.
196 		 * This has the effect of rewinding back to the
197 		 * top of the union stack and is needed by
198 		 * programs which plan to fchdir to a descriptor
199 		 * which has also been read -- see fts.c.
200 		 */
201 		if (flags & DTF_REWIND) {
202 			(void) close(fd);
203 			if ((fd = open(name, O_RDONLY)) == -1 ||
204 			    fcntl(fd, F_SETFD, FD_CLOEXEC) == -1) {
205 				dirp->dd_buf = buf;
206 				goto error;
207 			}
208 		}
209 
210 		/*
211 		 * There is now a buffer full of (possibly) duplicate
212 		 * names.
213 		 */
214 		dirp->dd_buf = buf;
215 
216 		/*
217 		 * Go round this loop twice...
218 		 *
219 		 * Scan through the buffer, counting entries.
220 		 * On the second pass, save pointers to each one.
221 		 * Then sort the pointers and remove duplicate names.
222 		 */
223 		if (!nfsdir) {
224 			for (dpv = 0;;) {
225 				for (n = 0, ddptr = buf; ddptr < ddeptr;) {
226 					struct dirent *dp;
227 
228 					dp = (struct dirent *)(void *)ddptr;
229 					if ((long)dp & _DIRENT_ALIGN(dp))
230 						break;
231 					/*
232 					 * d_reclen is unsigned,
233 					 * so no need to compare <= 0
234 					 */
235 					if (dp->d_reclen > (ddeptr + 1 - ddptr))
236 						break;
237 					ddptr += dp->d_reclen;
238 					if (dp->d_fileno) {
239 						if (dpv)
240 							dpv[n] = dp;
241 						n++;
242 					}
243 				}
244 
245 				if (dpv) {
246 					struct dirent *xp;
247 
248 					/*
249 					 * This sort must be stable.
250 					 */
251 					mergesort(dpv, (size_t)n, sizeof(*dpv),
252 					    alphasort);
253 
254 					dpv[n] = NULL;
255 					xp = NULL;
256 
257 					/*
258 					 * Scan through the buffer in sort
259 					 * order, zapping the inode number
260 					 * of any duplicate names.
261 					 */
262 					for (n = 0; dpv[n]; n++) {
263 						struct dirent *dp = dpv[n];
264 
265 						if ((xp == NULL) ||
266 						    strcmp(dp->d_name,
267 						      xp->d_name))
268 							xp = dp;
269 						else
270 							dp->d_fileno = 0;
271 						if (dp->d_type == DT_WHT &&
272 						    (flags & DTF_HIDEW))
273 							dp->d_fileno = 0;
274 					}
275 
276 					free(dpv);
277 					break;
278 				} else {
279 					dpv = malloc((n + 1) *
280 					    sizeof(struct dirent *));
281 					if (dpv == NULL)
282 						break;
283 				}
284 			}
285 		}
286 
287 		dirp->dd_len = len;
288 		dirp->dd_size = ddptr - dirp->dd_buf;
289 	} else {
290 		dirp->dd_len = incr;
291 		dirp->dd_buf = malloc((size_t)dirp->dd_len);
292 		if (dirp->dd_buf == NULL)
293 			goto error;
294 		dirp->dd_seek = 0;
295 		flags &= ~DTF_REWIND;
296 	}
297 
298 	dirp->dd_loc = 0;
299 	dirp->dd_fd = fd;
300 	dirp->dd_flags = flags;
301 
302 	/*
303 	 * Set up seek point for rewinddir.
304 	 */
305 #ifdef _REENTRANT
306 	if (__isthreaded) {
307 		if ((dirp->dd_lock = malloc(sizeof(mutex_t))) == NULL)
308 			goto error;
309 		mutex_init((mutex_t *)dirp->dd_lock, NULL);
310 	}
311 #endif
312 	dirp->dd_internal = NULL;
313 	(void)_telldir_unlocked(dirp);
314 	return (dirp);
315 error:
316 	serrno = errno;
317 	if (dirp && dirp->dd_buf)
318 		free(dirp->dd_buf);
319 	if (dirp)
320 		free(dirp);
321 	if (fd != -1)
322 		(void)close(fd);
323 	errno = serrno;
324 	return NULL;
325 }
326