xref: /netbsd-src/lib/libc/gen/opendir.c (revision 23c8222edbfb0f0932d88a8351d3a0cf817dfb9e)
1 /*	$NetBSD: opendir.c,v 1.25 2004/04/21 01:05:32 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 1.25 2004/04/21 01:05:32 christos Exp $");
38 #endif
39 #endif /* LIBC_SCCS and not lint */
40 
41 #include "namespace.h"
42 #include "reentrant.h"
43 #include <sys/param.h>
44 #include <sys/mount.h>
45 #include <sys/stat.h>
46 
47 #include <assert.h>
48 #include <dirent.h>
49 #include <errno.h>
50 #include <fcntl.h>
51 #include <stdlib.h>
52 #include <string.h>
53 #include <unistd.h>
54 
55 #ifdef __weak_alias
56 __weak_alias(opendir,_opendir)
57 #endif
58 
59 /*
60  * Open a directory.
61  */
62 DIR *
63 opendir(name)
64 	const char *name;
65 {
66 
67 	_DIAGASSERT(name != NULL);
68 
69 	return (__opendir2(name, DTF_HIDEW|DTF_NODUP));
70 }
71 
72 DIR *
73 __opendir2(name, flags)
74 	const char *name;
75 	int flags;
76 {
77 	DIR *dirp = NULL;
78 	int fd;
79 	int serrno;
80 	struct stat sb;
81 	int pagesz;
82 	int incr;
83 	int unionstack, nfsdir;
84 	struct statvfs sfb;
85 
86 	_DIAGASSERT(name != NULL);
87 
88 	if ((fd = open(name, O_RDONLY | O_NONBLOCK)) == -1)
89 		return (NULL);
90 	if (fstat(fd, &sb) || !S_ISDIR(sb.st_mode)) {
91 		close(fd);
92 		errno = ENOTDIR;
93 		return (NULL);
94 	}
95 	if (fcntl(fd, F_SETFD, FD_CLOEXEC) == -1 ||
96 	    (dirp = (DIR *)malloc(sizeof(DIR))) == NULL) {
97 		goto error;
98 	}
99 	dirp->dd_buf = NULL;
100 
101 	/*
102 	 * If the machine's page size is an exact multiple of DIRBLKSIZ,
103 	 * use a buffer that is cluster boundary aligned.
104 	 * Hopefully this can be a big win someday by allowing page trades
105 	 * to user space to be done by getdirentries()
106 	 */
107 	if (((pagesz = getpagesize()) % DIRBLKSIZ) == 0)
108 		incr = pagesz;
109 	else
110 		incr = DIRBLKSIZ;
111 
112 	/*
113 	 * Determine whether this directory is the top of a union stack.
114 	 */
115 
116 	if (fstatvfs1(fd, &sfb, ST_NOWAIT) < 0)
117 		goto error;
118 
119 	if (flags & DTF_NODUP)
120 		unionstack = !(strncmp(sfb.f_fstypename, MOUNT_UNION,
121 		    MFSNAMELEN)) || (sfb.f_flag & MNT_UNION);
122 	else
123 		unionstack = 0;
124 
125 	nfsdir = !(strncmp(sfb.f_fstypename, MOUNT_NFS, MFSNAMELEN));
126 
127 	if (unionstack || nfsdir) {
128 		size_t len;
129 		size_t space;
130 		char *buf, *nbuf;
131 		char *ddptr;
132 		char *ddeptr;
133 		int n;
134 		struct dirent **dpv;
135 
136 		/*
137 		 * The strategy here for directories on top of a union stack
138 		 * is to read all the directory entries into a buffer, sort
139 		 * the buffer, and remove duplicate entries by setting the
140 		 * inode number to zero.
141 		 *
142 		 * For directories on an NFS mounted filesystem, we try
143 	 	 * to get a consistent snapshot by trying until we have
144 		 * successfully read all of the directory without errors
145 		 * (i.e. 'bad cookie' errors from the server because
146 		 * the directory was modified). These errors should not
147 		 * happen often, but need to be dealt with.
148 		 */
149 retry:
150 		len = 0;
151 		space = 0;
152 		buf = 0;
153 		ddptr = 0;
154 
155 		do {
156 			/*
157 			 * Always make at least DIRBLKSIZ bytes
158 			 * available to getdirentries
159 			 */
160 			if (space < DIRBLKSIZ) {
161 				space += incr;
162 				len += incr;
163 				nbuf = realloc(buf, len);
164 				if (nbuf == NULL) {
165 					dirp->dd_buf = buf;
166 					goto error;
167 				}
168 				buf = nbuf;
169 				ddptr = buf + (len - space);
170 			}
171 
172 			dirp->dd_seek = lseek(fd, (off_t)0, SEEK_CUR);
173 			n = getdents(fd, ddptr, space);
174 			/*
175 			 * For NFS: EINVAL means a bad cookie error
176 			 * from the server. Keep trying to get a
177 			 * consistent view, in this case this means
178 			 * starting all over again.
179 			 */
180 			if (n == -1 && errno == EINVAL && nfsdir) {
181 				free(buf);
182 				lseek(fd, (off_t)0, SEEK_SET);
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 				dirp->dd_buf = buf;
205 				goto error;
206 			}
207 		}
208 
209 		/*
210 		 * There is now a buffer full of (possibly) duplicate
211 		 * names.
212 		 */
213 		dirp->dd_buf = buf;
214 
215 		/*
216 		 * Go round this loop twice...
217 		 *
218 		 * Scan through the buffer, counting entries.
219 		 * On the second pass, save pointers to each one.
220 		 * Then sort the pointers and remove duplicate names.
221 		 */
222 		if (!nfsdir) {
223 			for (dpv = 0;;) {
224 				for (n = 0, ddptr = buf; ddptr < ddeptr;) {
225 					struct dirent *dp;
226 
227 					dp = (struct dirent *)(void *)ddptr;
228 					if ((long)dp & 03)
229 						break;
230 					/*
231 					 * d_reclen is unsigned,
232 					 * so no need to compare <= 0
233 					 */
234 					if (dp->d_reclen > (ddeptr + 1 - ddptr))
235 						break;
236 					ddptr += dp->d_reclen;
237 					if (dp->d_fileno) {
238 						if (dpv)
239 							dpv[n] = dp;
240 						n++;
241 					}
242 				}
243 
244 				if (dpv) {
245 					struct dirent *xp;
246 
247 					/*
248 					 * This sort must be stable.
249 					 */
250 					mergesort(dpv, (size_t)n, sizeof(*dpv),
251 					    alphasort);
252 
253 					dpv[n] = NULL;
254 					xp = NULL;
255 
256 					/*
257 					 * Scan through the buffer in sort
258 					 * order, zapping the inode number
259 					 * of any duplicate names.
260 					 */
261 					for (n = 0; dpv[n]; n++) {
262 						struct dirent *dp = dpv[n];
263 
264 						if ((xp == NULL) ||
265 						    strcmp(dp->d_name,
266 						      xp->d_name))
267 							xp = dp;
268 						else
269 							dp->d_fileno = 0;
270 						if (dp->d_type == DT_WHT &&
271 						    (flags & DTF_HIDEW))
272 							dp->d_fileno = 0;
273 					}
274 
275 					free(dpv);
276 					break;
277 				} else {
278 					dpv = malloc((n + 1) *
279 					    sizeof(struct dirent *));
280 					if (dpv == NULL)
281 						break;
282 				}
283 			}
284 		}
285 
286 		dirp->dd_len = len;
287 		dirp->dd_size = ddptr - dirp->dd_buf;
288 	} else {
289 		dirp->dd_len = incr;
290 		dirp->dd_buf = malloc((size_t)dirp->dd_len);
291 		if (dirp->dd_buf == NULL)
292 			goto error;
293 		dirp->dd_seek = 0;
294 		flags &= ~DTF_REWIND;
295 	}
296 
297 	dirp->dd_loc = 0;
298 	dirp->dd_fd = fd;
299 	dirp->dd_flags = flags;
300 
301 	/*
302 	 * Set up seek point for rewinddir.
303 	 */
304 #ifdef _REENTRANT
305 	if (__isthreaded) {
306 		if ((dirp->dd_lock = malloc(sizeof(mutex_t))) == NULL)
307 			goto error;
308 		mutex_init((mutex_t *)dirp->dd_lock, NULL);
309 	}
310 #endif
311 	dirp->dd_rewind = telldir(dirp);
312 	return (dirp);
313 error:
314 	serrno = errno;
315 	if (dirp && dirp->dd_buf)
316 		free(dirp->dd_buf);
317 	if (dirp)
318 		free(dirp);
319 	if (fd != -1)
320 		(void)close(fd);
321 	errno = serrno;
322 	return NULL;
323 }
324