xref: /netbsd-src/sys/kern/vfs_lookup.c (revision ae1bfcddc410612bc8c58b807e1830becb69a24c)
1 /*
2  * Copyright (c) 1982, 1986, 1989 Regents of the University of California.
3  * All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	from: @(#)vfs_lookup.c	7.32 (Berkeley) 5/21/91
39  *	$Id: vfs_lookup.c,v 1.11 1994/05/18 05:12:44 cgd Exp $
40  */
41 
42 #include <sys/param.h>
43 #include <sys/syslimits.h>
44 #include <sys/time.h>
45 #include <sys/namei.h>
46 #include <sys/vnode.h>
47 #include <sys/mount.h>
48 #include <sys/errno.h>
49 #include <sys/malloc.h>
50 #include <sys/filedesc.h>
51 #include <sys/proc.h>
52 #ifdef KTRACE
53 #include <sys/ktrace.h>
54 #endif
55 
56 /*
57  * Convert a pathname into a pointer to a locked inode.
58  *
59  * The FOLLOW flag is set when symbolic links are to be followed
60  * when they occur at the end of the name translation process.
61  * Symbolic links are always followed for all other pathname
62  * components other than the last.
63  *
64  * The segflg defines whether the name is to be copied from user
65  * space or kernel space.
66  *
67  * Overall outline of namei:
68  *
69  *	copy in name
70  *	get starting directory
71  *	while (!done && !error) {
72  *		call lookup to search path.
73  *		if symbolic link, massage name in buffer and continue
74  *	}
75  */
76 namei(ndp, p)
77 	register struct nameidata *ndp;
78 	struct proc *p;
79 {
80 	register struct filedesc *fdp;	/* pointer to file descriptor state */
81 	register char *cp;		/* pointer into pathname argument */
82 	register struct vnode *dp;	/* the directory we are searching */
83 	struct iovec aiov;		/* uio for reading symbolic links */
84 	struct uio auio;
85 	int error, linklen;
86 
87 	ndp->ni_cred = p->p_ucred;
88 	fdp = p->p_fd;
89 
90 	/*
91 	 * Get a buffer for the name to be translated, and copy the
92 	 * name into the buffer.
93 	 */
94 	if ((ndp->ni_nameiop & HASBUF) == 0)
95 		MALLOC(ndp->ni_pnbuf, caddr_t, MAXPATHLEN, M_NAMEI, M_WAITOK);
96 	if (ndp->ni_segflg == UIO_SYSSPACE)
97 		error = copystr(ndp->ni_dirp, ndp->ni_pnbuf,
98 			    MAXPATHLEN, &ndp->ni_pathlen);
99 	else
100 		error = copyinstr(ndp->ni_dirp, ndp->ni_pnbuf,
101 			    MAXPATHLEN, &ndp->ni_pathlen);
102 	if (error) {
103 		free(ndp->ni_pnbuf, M_NAMEI);
104 		ndp->ni_vp = NULL;
105 		return (error);
106 	}
107 	ndp->ni_loopcnt = 0;
108 #ifdef KTRACE
109 	if (KTRPOINT(p, KTR_NAMEI))
110 		ktrnamei(p->p_tracep, ndp->ni_pnbuf);
111 #endif
112 
113 	/*
114 	 * Get starting point for the translation.
115 	 */
116 	if ((ndp->ni_rootdir = fdp->fd_rdir) == NULL)
117 		ndp->ni_rootdir = rootvnode;
118 	dp = fdp->fd_cdir;
119 	VREF(dp);
120 	for (;;) {
121 		/*
122 		 * Check if root directory should replace current directory.
123 		 * Done at start of translation and after symbolic link.
124 		 */
125 		ndp->ni_ptr = ndp->ni_pnbuf;
126 		if (*ndp->ni_ptr == '/') {
127 			vrele(dp);
128 			while (*ndp->ni_ptr == '/') {
129 				ndp->ni_ptr++;
130 				ndp->ni_pathlen--;
131 			}
132 			dp = ndp->ni_rootdir;
133 			VREF(dp);
134 		}
135 		ndp->ni_startdir = dp;
136 		if (error = lookup(ndp, p)) {
137 			FREE(ndp->ni_pnbuf, M_NAMEI);
138 			return (error);
139 		}
140 		/*
141 		 * Check for symbolic link
142 		 */
143 		if (ndp->ni_more == 0) {
144 			if ((ndp->ni_nameiop & (SAVENAME | SAVESTART)) == 0)
145 				FREE(ndp->ni_pnbuf, M_NAMEI);
146 			else
147 				ndp->ni_nameiop |= HASBUF;
148 			return (0);
149 		}
150 		if ((ndp->ni_nameiop & LOCKPARENT) && ndp->ni_pathlen == 1)
151 			VOP_UNLOCK(ndp->ni_dvp);
152 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
153 			error = ELOOP;
154 			break;
155 		}
156 		if (ndp->ni_pathlen > 1)
157 			MALLOC(cp, char *, MAXPATHLEN, M_NAMEI, M_WAITOK);
158 		else
159 			cp = ndp->ni_pnbuf;
160 		aiov.iov_base = cp;
161 		aiov.iov_len = MAXPATHLEN;
162 		auio.uio_iov = &aiov;
163 		auio.uio_iovcnt = 1;
164 		auio.uio_offset = 0;
165 		auio.uio_rw = UIO_READ;
166 		auio.uio_segflg = UIO_SYSSPACE;
167 		auio.uio_procp = (struct proc *)0;
168 		auio.uio_resid = MAXPATHLEN;
169 		if (error = VOP_READLINK(ndp->ni_vp, &auio, p->p_ucred)) {
170 			if (ndp->ni_pathlen > 1)
171 				free(cp, M_NAMEI);
172 			break;
173 		}
174 		linklen = MAXPATHLEN - auio.uio_resid;
175 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
176 			if (ndp->ni_pathlen > 1)
177 				free(cp, M_NAMEI);
178 			error = ENAMETOOLONG;
179 			break;
180 		}
181 		if (ndp->ni_pathlen > 1) {
182 			bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
183 			FREE(ndp->ni_pnbuf, M_NAMEI);
184 			ndp->ni_pnbuf = cp;
185 		} else
186 			ndp->ni_pnbuf[linklen] = '\0';
187 		ndp->ni_pathlen += linklen;
188 		vput(ndp->ni_vp);
189 		dp = ndp->ni_dvp;
190 	}
191 	FREE(ndp->ni_pnbuf, M_NAMEI);
192 	vrele(ndp->ni_dvp);
193 	vput(ndp->ni_vp);
194 	ndp->ni_vp = NULL;
195 	return (error);
196 }
197 
198 /*
199  * Search a pathname.
200  * This is a very central and rather complicated routine.
201  *
202  * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
203  * The starting directory is taken from ni_startdir. The pathname is
204  * descended until done, or a symbolic link is encountered. The variable
205  * ni_more is clear if the path is completed; it is set to one if a
206  * symbolic link needing interpretation is encountered.
207  *
208  * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
209  * whether the name is to be looked up, created, renamed, or deleted.
210  * When CREATE, RENAME, or DELETE is specified, information usable in
211  * creating, renaming, or deleting a directory entry may be calculated.
212  * If flag has LOCKPARENT or'ed into it, the parent directory is returned
213  * locked. If flag has WANTPARENT or'ed into it, the parent directory is
214  * returned unlocked. Otherwise the parent directory is not returned. If
215  * the target of the pathname exists and LOCKLEAF is or'ed into the flag
216  * the target is returned locked, otherwise it is returned unlocked.
217  * When creating or renaming and LOCKPARENT is specified, the target may not
218  * be ".".  When deleting and LOCKPARENT is specified, the target may be ".".
219  * NOTE: (LOOKUP | LOCKPARENT) currently returns the parent vnode unlocked.
220  *
221  * Overall outline of lookup:
222  *
223  * dirloop:
224  *	identify next component of name at ndp->ni_ptr
225  *	handle degenerate case where name is null string
226  *	if .. and crossing mount points and on mounted filesys, find parent
227  *	call VOP_LOOKUP routine for next component name
228  *	    directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
229  *	    component vnode returned in ni_vp (if it exists), locked.
230  *	if result vnode is mounted on and crossing mount points,
231  *	    find mounted on vnode
232  *	if more components of name, do next level at dirloop
233  *	return the answer in ni_vp, locked if LOCKLEAF set
234  *	    if LOCKPARENT set, return locked parent in ni_dvp
235  *	    if WANTPARENT set, return unlocked parent in ni_dvp
236  */
237 lookup(ndp, p)
238 	register struct nameidata *ndp;
239 	struct proc *p;
240 {
241 	register char *cp;		/* pointer into pathname argument */
242 	register struct vnode *dp = 0;	/* the directory we are searching */
243 	struct vnode *tdp;		/* saved dp */
244 	struct mount *mp;		/* mount table entry */
245 	int docache;			/* == 0 do not cache last component */
246 	int flag;			/* LOOKUP, CREATE, RENAME or DELETE */
247 	int wantparent;			/* 1 => wantparent or lockparent flag */
248 	int rdonly;			/* mounted read-only flag bit(s) */
249 	int error = 0;
250 
251 	/*
252 	 * Setup: break out flag bits into variables.
253 	 */
254 	flag = ndp->ni_nameiop & OPMASK;
255 	wantparent = ndp->ni_nameiop & (LOCKPARENT|WANTPARENT);
256 	docache = (ndp->ni_nameiop & NOCACHE) ^ NOCACHE;
257 	if (flag == DELETE || (wantparent && flag != CREATE))
258 		docache = 0;
259 	rdonly = MNT_RDONLY;
260 	if (ndp->ni_nameiop & REMOTE)
261 		rdonly |= MNT_EXRDONLY;
262 	ndp->ni_dvp = NULL;
263 	ndp->ni_more = 0;
264 	dp = ndp->ni_startdir;
265 	ndp->ni_startdir = NULLVP;
266 	VOP_LOCK(dp);
267 
268 dirloop:
269 	/*
270 	 * Search a new directory.
271 	 *
272 	 * The ni_hash value is for use by vfs_cache.
273 	 * The last component of the filename is left accessible via
274 	 * ndp->ptr for callers that need the name. Callers needing
275 	 * the name set the SAVENAME flag. When done, they assume
276 	 * responsibility for freeing the pathname buffer.
277 	 */
278 	ndp->ni_hash = 0;
279 	for (cp = ndp->ni_ptr; *cp != 0 && *cp != '/'; cp++)
280 		ndp->ni_hash += (unsigned char)*cp;
281 	ndp->ni_hash ^= dp->v_id;
282 	ndp->ni_namelen = cp - ndp->ni_ptr;
283 	if (ndp->ni_namelen >= NAME_MAX) {
284 		error = ENAMETOOLONG;
285 		goto bad;
286 	}
287 #ifdef NAMEI_DIAGNOSTIC
288 	{ char c = *cp;
289 	*cp = '\0';
290 	printf("{%s}: ", ndp->ni_ptr);
291 	*cp = c; }
292 #endif
293 	ndp->ni_pathlen -= ndp->ni_namelen;
294 	ndp->ni_next = cp;
295 	ndp->ni_makeentry = 1;
296 	if (*cp == '\0' && docache == 0)
297 		ndp->ni_makeentry = 0;
298 	ndp->ni_isdotdot = (ndp->ni_namelen == 2 &&
299 		ndp->ni_ptr[1] == '.' && ndp->ni_ptr[0] == '.');
300 
301 	/*
302 	 * Check for degenerate name (e.g. / or "")
303 	 * which is a way of talking about a directory,
304 	 * e.g. like "/." or ".".
305 	 */
306 	if (ndp->ni_ptr[0] == '\0') {
307 		if (flag != LOOKUP || wantparent) {
308 			error = EISDIR;
309 			goto bad;
310 		}
311 		if (dp->v_type != VDIR) {
312 			error = ENOTDIR;
313 			goto bad;
314 		}
315 		if (!(ndp->ni_nameiop & LOCKLEAF))
316 			VOP_UNLOCK(dp);
317 		ndp->ni_vp = dp;
318 		if (ndp->ni_nameiop & SAVESTART)
319 			panic("lookup: SAVESTART");
320 		return (0);
321 	}
322 
323 	/*
324 	 * Handle "..": two special cases.
325 	 * 1. If at root directory (e.g. after chroot)
326 	 *    then ignore it so can't get out.
327 	 * 2. If this vnode is the root of a mounted
328 	 *    filesystem, then replace it with the
329 	 *    vnode which was mounted on so we take the
330 	 *    .. in the other file system.
331 	 */
332 	if (ndp->ni_isdotdot) {
333 		for (;;) {
334 			if ((dp == ndp->ni_rootdir) || (dp == rootvnode)) {
335 				ndp->ni_dvp = dp;
336 				ndp->ni_vp = dp;
337 				VREF(dp);
338 				goto nextname;
339 			}
340 			if ((dp->v_flag & VROOT) == 0 ||
341 			    (ndp->ni_nameiop & NOCROSSMOUNT))
342 				break;
343 			tdp = dp;
344 			dp = dp->v_mount->mnt_vnodecovered;
345 			vput(tdp);
346 			VREF(dp);
347 			VOP_LOCK(dp);
348 		}
349 	}
350 
351 	/*
352 	 * We now have a segment name to search for, and a directory to search.
353 	 * The filesystem will return EJUSTRETURN on the lookup if the file
354 	 * doesn't exist, but can be created.
355 	 */
356 relookup:
357 	if (error = VOP_LOOKUP(dp, ndp, p)) {
358 #ifdef DIAGNOSTIC
359 		if (ndp->ni_vp != NULL)
360 			panic("leaf should be empty");
361 #endif
362 #ifdef NAMEI_DIAGNOSTIC
363 		printf("not found\n");
364 #endif
365 		if (error == ENOENT &&
366 		    (dp->v_flag & VROOT) &&
367 		    (dp->v_mount->mnt_flag & MNT_UNION)) {
368 			tdp = dp;
369 			dp = dp->v_mount->mnt_vnodecovered;
370 			vput(tdp);
371 			VREF(dp);
372 			VOP_LOCK(dp);
373 			goto relookup;
374 		}
375 		if (error != EJUSTRETURN)
376 			goto bad;
377 
378 		/*
379 		 * If creating and at end of pathname, then can consider
380 		 * allowing file to be created.
381 		 */
382 		if (ndp->ni_dvp->v_mount->mnt_flag & rdonly) {
383 			error = EROFS;
384 			goto bad;
385 		}
386 		/*
387 		 * We return with ni_vp NULL to indicate that the entry
388 		 * doesn't currently exist, leaving a pointer to the
389 		 * (possibly locked) directory inode in ndp->ni_dvp.
390 		 */
391 		if (ndp->ni_nameiop & SAVESTART) {
392 			ndp->ni_startdir = ndp->ni_dvp;
393 			VREF(ndp->ni_startdir);
394 		}
395 		return (0);
396 	}
397 #ifdef NAMEI_DIAGNOSTIC
398 	printf("found\n");
399 #endif
400 
401 	dp = ndp->ni_vp;
402 	/*
403 	 * Check for symbolic link
404 	 */
405 	if ((dp->v_type == VLNK) &&
406 	    ((ndp->ni_nameiop & FOLLOW) || *ndp->ni_next == '/')) {
407 		ndp->ni_more = 1;
408 		return (0);
409 	}
410 
411 	/*
412 	 * Check to see if the vnode has been mounted on;
413 	 * if so find the root of the mounted file system.
414 	 */
415 mntloop:
416 	while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
417 	       (ndp->ni_nameiop & NOCROSSMOUNT) == 0) {
418 		while(mp->mnt_flag & MNT_MLOCK) {
419 			mp->mnt_flag |= MNT_MWAIT;
420 			tsleep((caddr_t)mp, PVFS, "lookup", 0);
421 			goto mntloop;
422 		}
423 		if (error = VFS_ROOT(dp->v_mountedhere, &tdp))
424 			goto bad2;
425 		vput(dp);
426 		ndp->ni_vp = dp = tdp;
427 	}
428 
429 nextname:
430 	/*
431 	 * Not a symbolic link.  If more pathname,
432 	 * continue at next component, else return.
433 	 */
434 	if (*ndp->ni_next == '/') {
435 		ndp->ni_ptr = ndp->ni_next;
436 		while (*ndp->ni_ptr == '/') {
437 			ndp->ni_ptr++;
438 			ndp->ni_pathlen--;
439 		}
440 		vrele(ndp->ni_dvp);
441 		goto dirloop;
442 	}
443 	/*
444 	 * Check for read-only file systems.
445 	 */
446 	if (flag == DELETE || flag == RENAME) {
447 		/*
448 		 * Disallow directory write attempts on read-only
449 		 * file systems.
450 		 */
451 		if ((dp->v_mount->mnt_flag & rdonly) ||
452 		    (wantparent && (ndp->ni_dvp->v_mount->mnt_flag & rdonly))) {
453 			error = EROFS;
454 			goto bad2;
455 		}
456 	}
457 	if (ndp->ni_nameiop & SAVESTART) {
458 		ndp->ni_startdir = ndp->ni_dvp;
459 		VREF(ndp->ni_startdir);
460 	}
461 	if (!wantparent)
462 		vrele(ndp->ni_dvp);
463 	if ((ndp->ni_nameiop & LOCKLEAF) == 0)
464 		VOP_UNLOCK(dp);
465 	return (0);
466 
467 bad2:
468 	if ((ndp->ni_nameiop & LOCKPARENT) && *ndp->ni_next == '\0')
469 		VOP_UNLOCK(ndp->ni_dvp);
470 	vrele(ndp->ni_dvp);
471 bad:
472 	vput(dp);
473 	ndp->ni_vp = NULL;
474 	return (error);
475 }
476