xref: /csrg-svn/sys/kern/vfs_lookup.c (revision 38347)
1 /*
2  * Copyright (c) 1982, 1986, 1989 Regents of the University of California.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms are permitted
6  * provided that the above copyright notice and this paragraph are
7  * duplicated in all such forms and that any documentation,
8  * advertising materials, and other materials related to such
9  * distribution and use acknowledge that the software was developed
10  * by the University of California, Berkeley.  The name of the
11  * University may not be used to endorse or promote products derived
12  * from this software without specific prior written permission.
13  * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
14  * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
15  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
16  *
17  *	@(#)vfs_lookup.c	7.14 (Berkeley) 06/27/89
18  */
19 
20 #include "param.h"
21 #include "time.h"
22 #include "namei.h"
23 #include "vnode.h"
24 #include "mount.h"
25 #include "errno.h"
26 #include "malloc.h"
27 
28 #ifdef KTRACE
29 #include "user.h"
30 #include "proc.h"
31 #include "ktrace.h"
32 #endif
33 
34 /*
35  * Convert a pathname into a pointer to a locked inode.
36  * This is a very central and rather complicated routine.
37  *
38  * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
39  * whether the name is to be looked up, created, renamed, or deleted.
40  * When CREATE, RENAME, or DELETE is specified, information usable in
41  * creating, renaming, or deleting a directory entry may be calculated.
42  * If flag has LOCKPARENT or'ed into it and the target of the pathname
43  * exists, namei returns both the target and its parent directory locked.
44  * When creating or renaming and LOCKPARENT is specified, the target may not
45  * be ".".  When deleting and LOCKPARENT is specified, the target may be ".".
46  *
47  * The FOLLOW flag is set when symbolic links are to be followed
48  * when they occur at the end of the name translation process.
49  * Symbolic links are always followed for all other pathname
50  * components other than the last.
51  *
52  * The segflg defines whether the name is to be copied from user
53  * space or kernel space.
54  *
55  * Overall outline of namei:
56  *
57  *	copy in name
58  *	get starting directory
59  * dirloop:
60  *	copy next component of name to ndp->ni_dent
61  *	handle degenerate case where name is null string
62  *	if .. and on mounted filesys, find parent
63  *	call lookup routine for next component name
64  *	  directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
65  *	  component vnode returned in ni_vp (if it exists), locked.
66  *	if symbolic link, massage name in buffer and continue at dirloop
67  *	if result inode is mounted on, find mounted on vnode
68  *	if more components of name, do next level at dirloop
69  *	return the answer in ni_vp as locked vnode;
70  *	  if LOCKPARENT set, return locked parent in ni_dvp
71  *
72  * NOTE: (LOOKUP | LOCKPARENT) currently returns the parent vnode unlocked.
73  */
74 namei(ndp)
75 	register struct nameidata *ndp;
76 {
77 	register char *cp;		/* pointer into pathname argument */
78 	register struct vnode *dp = 0;	/* the directory we are searching */
79 	register int i;		   	/* Temp counter */
80 	struct vnode *tdp;		/* saved dp */
81 	struct mount *mp;		/* mount table entry */
82 	int docache;			/* == 0 do not cache last component */
83 	int flag;			/* LOOKUP, CREATE, RENAME or DELETE */
84 	int wantparent;			/* 1 => wantparent or lockparent flag */
85 	int lockparent;			/* 1 => lockparent flag */
86 	int error = 0;
87 
88 	flag = ndp->ni_nameiop & OPFLAG;
89 	wantparent = ndp->ni_nameiop & (LOCKPARENT|WANTPARENT);
90 	lockparent = ndp->ni_nameiop & LOCKPARENT;
91 	docache = (ndp->ni_nameiop & NOCACHE) ^ NOCACHE;
92 	if (flag == DELETE || wantparent)
93 		docache = 0;
94 	/*
95 	 * Get a buffer for the name to be translated, and copy the
96 	 * name into the buffer.
97 	 */
98 	MALLOC(ndp->ni_pnbuf, caddr_t, MAXPATHLEN, M_NAMEI, M_WAITOK);
99 	if (ndp->ni_segflg == UIO_SYSSPACE)
100 		error = copystr(ndp->ni_dirp, ndp->ni_pnbuf, MAXPATHLEN,
101 		    &ndp->ni_pathlen);
102 	else
103 		error = copyinstr(ndp->ni_dirp, ndp->ni_pnbuf, MAXPATHLEN,
104 		    &ndp->ni_pathlen);
105 	if (error) {
106 		free(ndp->ni_pnbuf, M_NAMEI);
107 		ndp->ni_vp = NULL;
108 		return (error);
109 	}
110 	ndp->ni_ptr = ndp->ni_pnbuf;
111 	ndp->ni_loopcnt = 0;
112 	dp = ndp->ni_cdir;
113 	VREF(dp);
114 #ifdef KTRACE
115 	if (KTRPOINT(u.u_procp, KTR_NAMEI))
116 		ktrnamei(u.u_procp->p_tracep, ndp->ni_pnbuf);
117 #endif
118 
119 start:
120 	/*
121 	 * Get starting directory.
122 	 * Done at start of translation and after symbolic link.
123 	 */
124 	if (*ndp->ni_ptr == '/') {
125 		vrele(dp);
126 		while (*ndp->ni_ptr == '/') {
127 			ndp->ni_ptr++;
128 			ndp->ni_pathlen--;
129 		}
130 		if ((dp = ndp->ni_rdir) == NULL)
131 			dp = rootdir;
132 		VREF(dp);
133 	}
134 	VOP_LOCK(dp);
135 	ndp->ni_endoff = 0;
136 
137 	/*
138 	 * We come to dirloop to search a new directory.
139 	 */
140 dirloop:
141 	/*
142 	 * Copy next component of name to ndp->ni_dent.
143 	 * XXX kern_exec looks at d_name
144 	 * ??? The ni_hash value may be useful for vfs_cache
145 	 * XXX There must be the last component of the filename left
146 	 * somewhere accessible via. ndp for NFS (and any other stateless file
147 	 * systems) in case they are doing a CREATE. The "Towards a..." noted
148 	 * that ni_ptr would be left pointing to the last component, but since
149 	 * the ni_pnbuf gets free'd, that is not a good idea.
150 	 */
151 #ifdef notdef
152 	for (cp = ndp->ni_ptr; *cp != 0 && *cp != '/'; cp++) {
153 		if ((*cp & 0200) == 0)
154 			continue;
155 		if ((*cp&0377) == ('/'|0200) || flag != DELETE) {
156 			error = EINVAL;
157 			goto bad;
158 		}
159 	}
160 	ndp->ni_namelen = cp - ndp->ni_ptr;
161 	if (ndp->ni_namelen >= MAXNAMLEN) {
162 		error = ENAMETOOLONG;
163 		goto bad;
164 	}
165 	ndp->ni_pathlen -= ndp->ni_namelen;
166 #ifdef NAMEI_DIAGNOSTIC
167 	{ char c = *cp;
168 	*cp = '\0';
169 	printf("{%s}: ", ndp->ni_ptr);
170 	*cp = c; }
171 #endif
172 #else fornow
173 	ndp->ni_hash = 0;
174 	for (cp = ndp->ni_ptr, i = 0; *cp != 0 && *cp != '/'; cp++) {
175 		if (i >= MAXNAMLEN) {
176 			error = ENAMETOOLONG;
177 			goto bad;
178 		}
179 		if (*cp & 0200)
180 			if ((*cp&0377) == ('/'|0200) || flag != DELETE) {
181 				error = EINVAL;
182 				goto bad;
183 			}
184 		ndp->ni_dent.d_name[i++] = *cp;
185 		ndp->ni_hash += (unsigned char)*cp * i;
186 	}
187 	ndp->ni_namelen = i;
188 	ndp->ni_dent.d_namlen = i;
189 	ndp->ni_dent.d_name[i] = '\0';
190 	ndp->ni_pathlen -= i;
191 #ifdef NAMEI_DIAGNOSTIC
192 	printf("{%s}: ", ndp->ni_dent.d_name);
193 #endif
194 #endif fornow
195 	ndp->ni_next = cp;
196 	ndp->ni_makeentry = 1;
197 	if (*cp == '\0' && docache == 0)
198 		ndp->ni_makeentry = 0;
199 	ndp->ni_isdotdot = (ndp->ni_namelen == 2 &&
200 		ndp->ni_dent.d_name[1] == '.' && ndp->ni_dent.d_name[0] == '.');
201 
202 	/*
203 	 * Check for degenerate name (e.g. / or "")
204 	 * which is a way of talking about a directory,
205 	 * e.g. like "/." or ".".
206 	 */
207 	if (ndp->ni_ptr[0] == '\0') {
208 		if (flag != LOOKUP || wantparent) {
209 			error = EISDIR;
210 			goto bad;
211 		}
212 		free(ndp->ni_pnbuf, M_NAMEI);
213 		if (!(ndp->ni_nameiop & LOCKLEAF))
214 			VOP_UNLOCK(dp);
215 		ndp->ni_vp = dp;
216 		return (0);
217 	}
218 
219 	/*
220 	 * Handle "..": two special cases.
221 	 * 1. If at root directory (e.g. after chroot)
222 	 *    then ignore it so can't get out.
223 	 * 2. If this vnode is the root of a mounted
224 	 *    file system, then replace it with the
225 	 *    vnode which was mounted on so we take the
226 	 *    .. in the other file system.
227 	 */
228 	if (ndp->ni_isdotdot) {
229 		for (;;) {
230 			if (dp == ndp->ni_rdir || dp == rootdir) {
231 				ndp->ni_dvp = dp;
232 				VREF(dp);
233 				goto nextname;
234 			}
235 			if ((dp->v_flag & VROOT) == 0)
236 				break;
237 			tdp = dp;
238 			dp = dp->v_mount->m_vnodecovered;
239 			vput(tdp);
240 			VOP_LOCK(dp);
241 			VREF(dp);
242 		}
243 	}
244 
245 	/*
246 	 * We now have a segment name to search for, and a directory to search.
247 	 */
248 	if (error = VOP_LOOKUP(dp, ndp)) {
249 		if (ndp->ni_vp != NULL)
250 			panic("leaf should be empty");
251 #ifdef NAMEI_DIAGNOSTIC
252 		printf("not found\n");
253 #endif
254 		/*
255 		 * If creating and at end of pathname, then can consider
256 		 * allowing file to be created.
257 		 */
258 		if (ndp->ni_dvp->v_mount->m_flag & M_RDONLY)
259 			error = EROFS;
260 		if (flag == LOOKUP || flag == DELETE ||
261 		    error != ENOENT || *cp != 0)
262 			goto bad;
263 		/*
264 		 * We return with ni_vp NULL to indicate that the entry
265 		 * doesn't currently exist, leaving a pointer to the
266 		 * (possibly locked) directory inode in ndp->ni_dvp.
267 		 */
268 		FREE(ndp->ni_pnbuf, M_NAMEI);
269 		return (0);	/* should this be ENOENT? */
270 	}
271 #ifdef NAMEI_DIAGNOSTIC
272 	printf("found\n");
273 #endif
274 
275 	/*
276 	 * Check for symbolic link
277 	 */
278 	dp = ndp->ni_vp;
279 	if ((dp->v_type == VLNK) &&
280 	    ((ndp->ni_nameiop & FOLLOW) || *ndp->ni_next == '/')) {
281 		struct iovec aiov;
282 		struct uio auio;
283 		int linklen;
284 
285 		if (++ndp->ni_loopcnt > MAXSYMLINKS) {
286 			error = ELOOP;
287 			goto bad2;
288 		}
289 		if (ndp->ni_pathlen > 1)
290 			MALLOC(cp, char *, MAXPATHLEN, M_NAMEI, M_WAITOK);
291 		else
292 			cp = ndp->ni_pnbuf;
293 		aiov.iov_base = cp;
294 		aiov.iov_len = MAXPATHLEN;
295 		auio.uio_iov = &aiov;
296 		auio.uio_iovcnt = 1;
297 		auio.uio_offset = 0;
298 		auio.uio_rw = UIO_READ;
299 		auio.uio_segflg = UIO_SYSSPACE;
300 		auio.uio_resid = MAXPATHLEN;
301 		if (error = VOP_READLINK(dp, &auio, ndp->ni_cred)) {
302 			if (ndp->ni_pathlen > 1)
303 				free(cp, M_NAMEI);
304 			goto bad2;
305 		}
306 		linklen = MAXPATHLEN - auio.uio_resid;
307 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
308 			if (ndp->ni_pathlen > 1)
309 				free(cp, M_NAMEI);
310 			error = ENAMETOOLONG;
311 			goto bad2;
312 		}
313 		if (ndp->ni_pathlen > 1) {
314 			bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
315 			FREE(ndp->ni_pnbuf, M_NAMEI);
316 			ndp->ni_pnbuf = cp;
317 		} else
318 			ndp->ni_pnbuf[linklen] = '\0';
319 		ndp->ni_ptr = cp;
320 		ndp->ni_pathlen += linklen;
321 		vput(dp);
322 		dp = ndp->ni_dvp;
323 		if (lockparent && *ndp->ni_next == '\0')
324 			VOP_UNLOCK(dp);
325 		goto start;
326 	}
327 
328 	/*
329 	 * Check to see if the vnode has been mounted on;
330 	 * if so find the root of the mounted file system.
331 	 */
332 mntloop:
333 	while (dp->v_type == VDIR && (mp = dp->v_mountedhere)) {
334 		while(mp->m_flag & M_MLOCK) {
335 			mp->m_flag |= M_MWAIT;
336 			sleep((caddr_t)mp, PVFS);
337 			goto mntloop;
338 		}
339 		error = VFS_ROOT(dp->v_mountedhere, &tdp);
340 		if (error)
341 			goto bad2;
342 		vput(dp);
343 		ndp->ni_vp = dp = tdp;
344 	}
345 
346 nextname:
347 	/*
348 	 * Not a symbolic link.  If more pathname,
349 	 * continue at next component, else return.
350 	 */
351 	ndp->ni_ptr = ndp->ni_next;
352 	if (*ndp->ni_ptr == '/') {
353 		while (*ndp->ni_ptr == '/') {
354 			ndp->ni_ptr++;
355 			ndp->ni_pathlen--;
356 		}
357 		vrele(ndp->ni_dvp);
358 		goto dirloop;
359 	}
360 	/*
361 	 * Check for read-only file systems and executing texts
362 	 */
363 	if (flag != LOOKUP) {
364 		/*
365 		 * Disallow write attempts on read-only file systems;
366 		 * unless the file is a socket or a block or character
367 		 * device resident on the file system.
368 		 */
369 		if ((dp->v_mount->m_flag & M_RDONLY) &&
370 		    dp->v_type != VCHR &&
371 		    dp->v_type != VBLK &&
372 		    dp->v_type != VSOCK) {
373 			error = EROFS;
374 			goto bad2;
375 		}
376 		/*
377 		 * If there's shared text associated with
378 		 * the inode, try to free it up once.  If
379 		 * we fail, we can't allow writing.
380 		 */
381 		if (dp->v_flag & VTEXT)
382 			xrele(dp);
383 		if (dp->v_flag & VTEXT) {
384 			error = ETXTBSY;
385 			goto bad2;
386 		}
387 		if (wantparent && flag != CREATE &&
388 		    (ndp->ni_dvp->v_mount->m_flag & M_RDONLY)) {
389 			error = EROFS;
390 			goto bad2;
391 		}
392 	}
393 	if (!wantparent)
394 		vrele(ndp->ni_dvp);
395 	if ((ndp->ni_nameiop & LOCKLEAF) == 0)
396 		VOP_UNLOCK(dp);
397 	FREE(ndp->ni_pnbuf, M_NAMEI);
398 	return (0);
399 
400 bad2:
401 	if (lockparent && *ndp->ni_next == '\0')
402 		VOP_UNLOCK(ndp->ni_dvp);
403 	vrele(ndp->ni_dvp);
404 bad:
405 	vput(dp);
406 	ndp->ni_vp = NULL;
407 	FREE(ndp->ni_pnbuf, M_NAMEI);
408 	return (error);
409 }
410