xref: /netbsd-src/lib/libc/gen/opendir.c (revision d48f14661dda8638fee055ba15d35bdfb29b9fa8)
1 /*	$NetBSD: opendir.c,v 1.31 2006/05/17 20:36:50 christos 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 1.31 2006/05/17 20:36:50 christos 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 /*
59  * Open a directory.
60  */
61 DIR *
62 opendir(const char *name)
63 {
64 
65 	_DIAGASSERT(name != NULL);
66 
67 	return (__opendir2(name, DTF_HIDEW|DTF_NODUP));
68 }
69 
70 DIR *
71 __opendir2(const char *name, int flags)
72 {
73 	DIR *dirp = NULL;
74 	int fd;
75 	int serrno;
76 	struct stat sb;
77 	int pagesz;
78 	int incr;
79 	int unionstack, nfsdir;
80 	struct statvfs sfb;
81 
82 	_DIAGASSERT(name != NULL);
83 
84 	if ((fd = open(name, O_RDONLY | O_NONBLOCK)) == -1 ||
85 	    fcntl(fd, F_SETFD, FD_CLOEXEC) == -1)
86 		goto error;
87 	if (fstat(fd, &sb) || !S_ISDIR(sb.st_mode)) {
88 		errno = ENOTDIR;
89 		goto error;
90 	}
91 	if ((dirp = (DIR *)malloc(sizeof(DIR))) == NULL)
92 		goto error;
93 	dirp->dd_buf = NULL;
94 
95 	/*
96 	 * If the machine's page size is an exact multiple of DIRBLKSIZ,
97 	 * use a buffer that is cluster boundary aligned.
98 	 * Hopefully this can be a big win someday by allowing page trades
99 	 * to user space to be done by getdirentries()
100 	 */
101 	if (((pagesz = getpagesize()) % DIRBLKSIZ) == 0)
102 		incr = pagesz;
103 	else
104 		incr = DIRBLKSIZ;
105 
106 	/*
107 	 * Determine whether this directory is the top of a union stack.
108 	 */
109 
110 	if (fstatvfs1(fd, &sfb, ST_NOWAIT) < 0)
111 		goto error;
112 
113 	if (flags & DTF_NODUP)
114 		unionstack = !(strncmp(sfb.f_fstypename, MOUNT_UNION,
115 		    MFSNAMELEN)) || (sfb.f_flag & MNT_UNION);
116 	else
117 		unionstack = 0;
118 
119 	nfsdir = !(strncmp(sfb.f_fstypename, MOUNT_NFS, MFSNAMELEN));
120 
121 	if (unionstack || nfsdir) {
122 		size_t len;
123 		size_t space;
124 		char *buf, *nbuf;
125 		char *ddptr;
126 		char *ddeptr;
127 		int n;
128 		struct dirent **dpv;
129 
130 		/*
131 		 * The strategy here for directories on top of a union stack
132 		 * is to read all the directory entries into a buffer, sort
133 		 * the buffer, and remove duplicate entries by setting the
134 		 * inode number to zero.
135 		 *
136 		 * For directories on an NFS mounted filesystem, we try
137 	 	 * to get a consistent snapshot by trying until we have
138 		 * successfully read all of the directory without errors
139 		 * (i.e. 'bad cookie' errors from the server because
140 		 * the directory was modified). These errors should not
141 		 * happen often, but need to be dealt with.
142 		 */
143 retry:
144 		len = 0;
145 		space = 0;
146 		buf = 0;
147 		ddptr = 0;
148 
149 		do {
150 			/*
151 			 * Always make at least DIRBLKSIZ bytes
152 			 * available to getdirentries
153 			 */
154 			if (space < DIRBLKSIZ) {
155 				space += incr;
156 				len += incr;
157 				nbuf = realloc(buf, len);
158 				if (nbuf == NULL) {
159 					dirp->dd_buf = buf;
160 					goto error;
161 				}
162 				buf = nbuf;
163 				ddptr = buf + (len - space);
164 			}
165 
166 			dirp->dd_seek = lseek(fd, (off_t)0, SEEK_CUR);
167 			n = getdents(fd, ddptr, space);
168 			/*
169 			 * For NFS: EINVAL means a bad cookie error
170 			 * from the server. Keep trying to get a
171 			 * consistent view, in this case this means
172 			 * starting all over again.
173 			 */
174 			if (n == -1 && errno == EINVAL && nfsdir) {
175 				free(buf);
176 				lseek(fd, (off_t)0, SEEK_SET);
177 				goto retry;
178 			}
179 			if (n > 0) {
180 				ddptr += n;
181 				space -= n;
182 			}
183 		} while (n > 0);
184 
185 		ddeptr = ddptr;
186 		flags |= __DTF_READALL;
187 
188 		/*
189 		 * Re-open the directory.
190 		 * This has the effect of rewinding back to the
191 		 * top of the union stack and is needed by
192 		 * programs which plan to fchdir to a descriptor
193 		 * which has also been read -- see fts.c.
194 		 */
195 		if (flags & DTF_REWIND) {
196 			(void) close(fd);
197 			if ((fd = open(name, O_RDONLY)) == -1 ||
198 			    fcntl(fd, F_SETFD, FD_CLOEXEC) == -1) {
199 				dirp->dd_buf = buf;
200 				goto error;
201 			}
202 		}
203 
204 		/*
205 		 * There is now a buffer full of (possibly) duplicate
206 		 * names.
207 		 */
208 		dirp->dd_buf = buf;
209 
210 		/*
211 		 * Go round this loop twice...
212 		 *
213 		 * Scan through the buffer, counting entries.
214 		 * On the second pass, save pointers to each one.
215 		 * Then sort the pointers and remove duplicate names.
216 		 */
217 		if (!nfsdir) {
218 			for (dpv = 0;;) {
219 				for (n = 0, ddptr = buf; ddptr < ddeptr;) {
220 					struct dirent *dp;
221 
222 					dp = (struct dirent *)(void *)ddptr;
223 					if ((long)dp & _DIRENT_ALIGN(dp))
224 						break;
225 					/*
226 					 * d_reclen is unsigned,
227 					 * so no need to compare <= 0
228 					 */
229 					if (dp->d_reclen > (ddeptr + 1 - ddptr))
230 						break;
231 					ddptr += dp->d_reclen;
232 					if (dp->d_fileno) {
233 						if (dpv)
234 							dpv[n] = dp;
235 						n++;
236 					}
237 				}
238 
239 				if (dpv) {
240 					struct dirent *xp;
241 
242 					/*
243 					 * This sort must be stable.
244 					 */
245 					mergesort(dpv, (size_t)n, sizeof(*dpv),
246 					    alphasort);
247 
248 					dpv[n] = NULL;
249 					xp = NULL;
250 
251 					/*
252 					 * Scan through the buffer in sort
253 					 * order, zapping the inode number
254 					 * of any duplicate names.
255 					 */
256 					for (n = 0; dpv[n]; n++) {
257 						struct dirent *dp = dpv[n];
258 
259 						if ((xp == NULL) ||
260 						    strcmp(dp->d_name,
261 						      xp->d_name))
262 							xp = dp;
263 						else
264 							dp->d_fileno = 0;
265 						if (dp->d_type == DT_WHT &&
266 						    (flags & DTF_HIDEW))
267 							dp->d_fileno = 0;
268 					}
269 
270 					free(dpv);
271 					break;
272 				} else {
273 					dpv = malloc((n + 1) *
274 					    sizeof(struct dirent *));
275 					if (dpv == NULL)
276 						break;
277 				}
278 			}
279 		}
280 
281 		dirp->dd_len = len;
282 		dirp->dd_size = ddptr - dirp->dd_buf;
283 	} else {
284 		dirp->dd_len = incr;
285 		dirp->dd_buf = malloc((size_t)dirp->dd_len);
286 		if (dirp->dd_buf == NULL)
287 			goto error;
288 		dirp->dd_seek = 0;
289 		flags &= ~DTF_REWIND;
290 	}
291 
292 	dirp->dd_loc = 0;
293 	dirp->dd_fd = fd;
294 	dirp->dd_flags = flags;
295 
296 	/*
297 	 * Set up seek point for rewinddir.
298 	 */
299 #ifdef _REENTRANT
300 	if (__isthreaded) {
301 		if ((dirp->dd_lock = malloc(sizeof(mutex_t))) == NULL)
302 			goto error;
303 		mutex_init((mutex_t *)dirp->dd_lock, NULL);
304 	}
305 #endif
306 	dirp->dd_internal = NULL;
307 	(void)_telldir_unlocked(dirp);
308 	return (dirp);
309 error:
310 	serrno = errno;
311 	if (dirp && dirp->dd_buf)
312 		free(dirp->dd_buf);
313 	if (dirp)
314 		free(dirp);
315 	if (fd != -1)
316 		(void)close(fd);
317 	errno = serrno;
318 	return NULL;
319 }
320