2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. 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.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
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.
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
38 * @(#)vfs_lookup.c 8.4 (Berkeley) 2/16/94
39 * $FreeBSD: src/sys/kern/vfs_lookup.c,v 1.38.2.3 2001/08/31 19:36:49 dillon Exp $
40 * $DragonFly: src/sys/kern/vfs_lookup.c,v 1.7 2003/09/28 03:44:02 dillon Exp $
43 #include "opt_ktrace.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/vnode.h>
48 #include <sys/mount.h>
49 #include <sys/filedesc.h>
51 #include <sys/namei.h>
54 #include <sys/ktrace.h>
57 #include <vm/vm_zone.h>
60 * Convert a pathname into a pointer to a locked inode.
62 * The CNP_FOLLOW flag is set when symbolic links are to be followed
63 * when they occur at the end of the name translation process.
64 * Symbolic links are always followed for all other pathname
65 * components other than the last.
67 * The segflg defines whether the name is to be copied from user
68 * space or kernel space.
70 * Overall outline of namei:
73 * get starting directory
74 * while (!done && !error) {
75 * call lookup to search path.
76 * if symbolic link, massage name in buffer and continue
80 namei(struct nameidata *ndp)
82 struct filedesc *fdp; /* pointer to file descriptor state */
83 char *cp; /* pointer into pathname argument */
84 struct vnode *dp; /* the directory we are searching */
85 struct iovec aiov; /* uio for reading symbolic links */
88 struct componentname *cnp = &ndp->ni_cnd;
91 KKASSERT(ndp->ni_cnd.cn_td != NULL);
92 p = cnp->cn_td->td_proc;
94 KASSERT(cnp->cn_cred, ("namei: bad cred/proc"));
95 KKASSERT(cnp->cn_cred == p->p_ucred); /* YYY */
96 KASSERT((cnp->cn_nameiop & (~NAMEI_OPMASK)) == 0,
97 ("namei: nameiop contaminated with flags"));
98 KASSERT((cnp->cn_flags & NAMEI_OPMASK) == 0,
99 ("namei: flags contaminated with nameiops"));
103 * Get a buffer for the name to be translated, and copy the
104 * name into the buffer.
106 if ((cnp->cn_flags & CNP_HASBUF) == 0)
107 cnp->cn_pnbuf = zalloc(namei_zone);
108 if (ndp->ni_segflg == UIO_SYSSPACE)
109 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
110 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
112 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
113 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
116 * Don't allow empty pathnames.
118 if (!error && *cnp->cn_pnbuf == '\0')
122 zfree(namei_zone, cnp->cn_pnbuf);
128 if (KTRPOINT(cnp->cn_td, KTR_NAMEI))
129 ktrnamei(cnp->cn_td->td_proc->p_tracep, cnp->cn_pnbuf);
133 * Get starting point for the translation.
135 ndp->ni_rootdir = fdp->fd_rdir;
136 ndp->ni_topdir = fdp->fd_jdir;
142 * Check if root directory should replace current directory.
143 * Done at start of translation and after symbolic link.
145 cnp->cn_nameptr = cnp->cn_pnbuf;
146 if (*(cnp->cn_nameptr) == '/') {
148 while (*(cnp->cn_nameptr) == '/') {
152 dp = ndp->ni_rootdir;
155 ndp->ni_startdir = dp;
158 zfree(namei_zone, cnp->cn_pnbuf);
162 * Check for symbolic link
164 if ((cnp->cn_flags & CNP_ISSYMLINK) == 0) {
165 if ((cnp->cn_flags & (CNP_SAVENAME | CNP_SAVESTART)) == 0)
166 zfree(namei_zone, cnp->cn_pnbuf);
168 cnp->cn_flags |= CNP_HASBUF;
170 if (vn_canvmio(ndp->ni_vp) == TRUE &&
171 (cnp->cn_nameiop != NAMEI_DELETE) &&
172 ((cnp->cn_flags & (CNP_NOOBJ|CNP_LOCKLEAF)) ==
174 vfs_object_create(ndp->ni_vp, ndp->ni_cnd.cn_td);
178 if ((cnp->cn_flags & CNP_LOCKPARENT) && ndp->ni_pathlen == 1)
179 VOP_UNLOCK(ndp->ni_dvp, 0, cnp->cn_td);
180 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
184 if (ndp->ni_pathlen > 1)
185 cp = zalloc(namei_zone);
189 aiov.iov_len = MAXPATHLEN;
190 auio.uio_iov = &aiov;
193 auio.uio_rw = UIO_READ;
194 auio.uio_segflg = UIO_SYSSPACE;
196 auio.uio_resid = MAXPATHLEN;
197 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
199 if (ndp->ni_pathlen > 1)
200 zfree(namei_zone, cp);
203 linklen = MAXPATHLEN - auio.uio_resid;
205 if (ndp->ni_pathlen > 1)
206 zfree(namei_zone, cp);
210 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
211 if (ndp->ni_pathlen > 1)
212 zfree(namei_zone, cp);
213 error = ENAMETOOLONG;
216 if (ndp->ni_pathlen > 1) {
217 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
218 zfree(namei_zone, cnp->cn_pnbuf);
221 cnp->cn_pnbuf[linklen] = '\0';
222 ndp->ni_pathlen += linklen;
226 zfree(namei_zone, cnp->cn_pnbuf);
235 * This is a very central and rather complicated routine.
237 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
238 * The starting directory is taken from ni_startdir. The pathname is
239 * descended until done, or a symbolic link is encountered. The variable
240 * ni_more is clear if the path is completed; it is set to one if a
241 * symbolic link needing interpretation is encountered.
243 * The flag argument is NAMEI_LOOKUP, CREATE, RENAME, or DELETE depending on
244 * whether the name is to be looked up, created, renamed, or deleted.
245 * When CREATE, RENAME, or DELETE is specified, information usable in
246 * creating, renaming, or deleting a directory entry may be calculated.
247 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
248 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
249 * returned unlocked. Otherwise the parent directory is not returned. If
250 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
251 * the target is returned locked, otherwise it is returned unlocked.
252 * When creating or renaming and LOCKPARENT is specified, the target may not
253 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
255 * Overall outline of lookup:
258 * identify next component of name at ndp->ni_ptr
259 * handle degenerate case where name is null string
260 * if .. and crossing mount points and on mounted filesys, find parent
261 * call VOP_LOOKUP routine for next component name
262 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
263 * component vnode returned in ni_vp (if it exists), locked.
264 * if result vnode is mounted on and crossing mount points,
265 * find mounted on vnode
266 * if more components of name, do next level at dirloop
267 * return the answer in ni_vp, locked if LOCKLEAF set
268 * if LOCKPARENT set, return locked parent in ni_dvp
269 * if WANTPARENT set, return unlocked parent in ni_dvp
273 struct nameidata *ndp;
275 char *cp; /* pointer into pathname argument */
276 struct vnode *dp = 0; /* the directory we are searching */
277 struct vnode *tdp; /* saved dp */
278 struct mount *mp; /* mount table entry */
279 int docache; /* == 0 do not cache last component */
280 int wantparent; /* 1 => wantparent or lockparent flag */
281 int rdonly; /* lookup read-only flag bit */
284 int dpunlocked = 0; /* dp has already been unlocked */
285 struct componentname *cnp = &ndp->ni_cnd;
286 struct thread *td = cnp->cn_td;
289 * Setup: break out flag bits into variables.
291 wantparent = cnp->cn_flags & (CNP_LOCKPARENT | CNP_WANTPARENT);
292 docache = (cnp->cn_flags & CNP_NOCACHE) ^ CNP_NOCACHE;
293 if (cnp->cn_nameiop == NAMEI_DELETE ||
294 (wantparent && cnp->cn_nameiop != NAMEI_CREATE &&
295 cnp->cn_nameiop != NAMEI_LOOKUP))
297 rdonly = cnp->cn_flags & CNP_RDONLY;
299 cnp->cn_flags &= ~CNP_ISSYMLINK;
300 dp = ndp->ni_startdir;
301 ndp->ni_startdir = NULLVP;
302 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, td);
306 * Search a new directory.
308 * The last component of the filename is left accessible via
309 * cnp->cn_nameptr for callers that need the name. Callers needing
310 * the name set the CNP_SAVENAME flag. When done, they assume
311 * responsibility for freeing the pathname buffer.
314 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
316 cnp->cn_namelen = cp - cnp->cn_nameptr;
317 if (cnp->cn_namelen > NAME_MAX) {
318 error = ENAMETOOLONG;
321 #ifdef NAMEI_DIAGNOSTIC
324 printf("{%s}: ", cnp->cn_nameptr);
327 ndp->ni_pathlen -= cnp->cn_namelen;
331 * Replace multiple slashes by a single slash and trailing slashes
332 * by a null. This must be done before VOP_LOOKUP() because some
333 * fs's don't know about trailing slashes. Remember if there were
334 * trailing slashes to handle symlinks, existing non-directories
335 * and non-existing files that won't be directories specially later.
338 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
343 *ndp->ni_next = '\0'; /* XXX for direnter() ... */
348 cnp->cn_flags |= CNP_MAKEENTRY;
349 if (*cp == '\0' && docache == 0)
350 cnp->cn_flags &= ~CNP_MAKEENTRY;
351 if (cnp->cn_namelen == 2 &&
352 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
353 cnp->cn_flags |= CNP_ISDOTDOT;
355 cnp->cn_flags &= ~CNP_ISDOTDOT;
356 if (*ndp->ni_next == 0)
357 cnp->cn_flags |= CNP_ISLASTCN;
359 cnp->cn_flags &= ~CNP_ISLASTCN;
363 * Check for degenerate name (e.g. / or "")
364 * which is a way of talking about a directory,
365 * e.g. like "/." or ".".
367 if (cnp->cn_nameptr[0] == '\0') {
368 if (dp->v_type != VDIR) {
372 if (cnp->cn_nameiop != NAMEI_LOOKUP) {
381 if (!(cnp->cn_flags & (CNP_LOCKPARENT | CNP_LOCKLEAF)))
382 VOP_UNLOCK(dp, 0, cnp->cn_td);
383 /* XXX This should probably move to the top of function. */
384 if (cnp->cn_flags & CNP_SAVESTART)
385 panic("lookup: CNP_SAVESTART");
390 * Handle "..": two special cases.
391 * 1. If at root directory (e.g. after chroot)
392 * or at absolute root directory
393 * then ignore it so can't get out.
394 * 2. If this vnode is the root of a mounted
395 * filesystem, then replace it with the
396 * vnode which was mounted on so we take the
397 * .. in the other file system.
398 * 3. If the vnode is the top directory of
399 * the jail or chroot, don't let them out.
401 if (cnp->cn_flags & CNP_ISDOTDOT) {
403 if (dp == ndp->ni_rootdir ||
404 dp == ndp->ni_topdir ||
411 if ((dp->v_flag & VROOT) == 0 ||
412 (cnp->cn_flags & CNP_NOCROSSMOUNT))
414 if (dp->v_mount == NULL) { /* forced unmount */
419 dp = dp->v_mount->mnt_vnodecovered;
422 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, td);
427 * We now have a segment name to search for, and a directory to search.
432 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
433 ASSERT_VOP_LOCKED(dp, "lookup");
434 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
435 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
436 #ifdef NAMEI_DIAGNOSTIC
437 printf("not found\n");
439 if ((error == ENOENT) &&
440 (dp->v_flag & VROOT) && (dp->v_mount != NULL) &&
441 (dp->v_mount->mnt_flag & MNT_UNION)) {
443 dp = dp->v_mount->mnt_vnodecovered;
444 if (cnp->cn_flags & CNP_PDIRUNLOCK)
449 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, td);
453 if (error != EJUSTRETURN)
456 * If creating and at end of pathname, then can consider
457 * allowing file to be created.
463 if (*cp == '\0' && trailing_slash &&
464 !(cnp->cn_flags & CNP_WILLBEDIR)) {
469 * We return with ni_vp NULL to indicate that the entry
470 * doesn't currently exist, leaving a pointer to the
471 * (possibly locked) directory inode in ndp->ni_dvp.
473 if (cnp->cn_flags & CNP_SAVESTART) {
474 ndp->ni_startdir = ndp->ni_dvp;
475 VREF(ndp->ni_startdir);
479 #ifdef NAMEI_DIAGNOSTIC
483 ASSERT_VOP_LOCKED(ndp->ni_vp, "lookup");
486 * Take into account any additional components consumed by
487 * the underlying filesystem.
489 if (cnp->cn_consume > 0) {
490 cnp->cn_nameptr += cnp->cn_consume;
491 ndp->ni_next += cnp->cn_consume;
492 ndp->ni_pathlen -= cnp->cn_consume;
499 * Check to see if the vnode has been mounted on;
500 * if so find the root of the mounted file system.
502 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
503 (cnp->cn_flags & CNP_NOCROSSMOUNT) == 0) {
504 if (vfs_busy(mp, 0, 0, td))
506 VOP_UNLOCK(dp, 0, td);
507 error = VFS_ROOT(mp, &tdp);
513 cache_mount(dp, tdp);
515 ndp->ni_vp = dp = tdp;
519 * Check for symbolic link
521 if ((dp->v_type == VLNK) &&
522 ((cnp->cn_flags & CNP_FOLLOW) || trailing_slash ||
523 *ndp->ni_next == '/')) {
524 cnp->cn_flags |= CNP_ISSYMLINK;
525 if (dp->v_mount == NULL) {
526 /* We can't know whether the directory was mounted with
527 * NOSYMFOLLOW, so we can't follow safely. */
531 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
539 * Check for bogus trailing slashes.
541 if (trailing_slash && dp->v_type != VDIR) {
548 * Not a symbolic link. If more pathname,
549 * continue at next component, else return.
551 if (*ndp->ni_next == '/') {
552 cnp->cn_nameptr = ndp->ni_next;
553 while (*cnp->cn_nameptr == '/') {
557 if (ndp->ni_dvp != ndp->ni_vp)
558 ASSERT_VOP_UNLOCKED(ndp->ni_dvp, "lookup");
563 * Disallow directory write attempts on read-only file systems.
566 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) {
570 if (cnp->cn_flags & CNP_SAVESTART) {
571 ndp->ni_startdir = ndp->ni_dvp;
572 VREF(ndp->ni_startdir);
577 if ((cnp->cn_flags & CNP_LOCKLEAF) == 0)
578 VOP_UNLOCK(dp, 0, td);
582 if ((cnp->cn_flags & (CNP_LOCKPARENT | CNP_PDIRUNLOCK)) == CNP_LOCKPARENT &&
583 *ndp->ni_next == '\0')
584 VOP_UNLOCK(ndp->ni_dvp, 0, td);
596 * relookup - lookup a path name component
597 * Used by lookup to re-aquire things.
600 relookup(dvp, vpp, cnp)
601 struct vnode *dvp, **vpp;
602 struct componentname *cnp;
604 struct thread *td = cnp->cn_td;
605 struct vnode *dp = 0; /* the directory we are searching */
606 int docache; /* == 0 do not cache last component */
607 int wantparent; /* 1 => wantparent or lockparent flag */
608 int rdonly; /* lookup read-only flag bit */
610 #ifdef NAMEI_DIAGNOSTIC
611 int newhash; /* DEBUG: check name hash */
612 char *cp; /* DEBUG: check name ptr/len */
616 * Setup: break out flag bits into variables.
618 wantparent = cnp->cn_flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
619 docache = (cnp->cn_flags & CNP_NOCACHE) ^ CNP_NOCACHE;
620 if (cnp->cn_nameiop == NAMEI_DELETE ||
621 (wantparent && cnp->cn_nameiop != NAMEI_CREATE))
623 rdonly = cnp->cn_flags & CNP_RDONLY;
624 cnp->cn_flags &= ~CNP_ISSYMLINK;
626 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, td);
630 * Search a new directory.
632 * The last component of the filename is left accessible via
633 * cnp->cn_nameptr for callers that need the name. Callers needing
634 * the name set the CNP_SAVENAME flag. When done, they assume
635 * responsibility for freeing the pathname buffer.
637 #ifdef NAMEI_DIAGNOSTIC
638 if (cnp->cn_namelen != cp - cnp->cn_nameptr)
639 panic ("relookup: bad len");
641 panic("relookup: not last component");
642 printf("{%s}: ", cnp->cn_nameptr);
646 * Check for degenerate name (e.g. / or "")
647 * which is a way of talking about a directory,
648 * e.g. like "/." or ".".
650 if (cnp->cn_nameptr[0] == '\0') {
651 if (cnp->cn_nameiop != NAMEI_LOOKUP || wantparent) {
655 if (dp->v_type != VDIR) {
659 if (!(cnp->cn_flags & CNP_LOCKLEAF))
660 VOP_UNLOCK(dp, 0, td);
662 /* XXX This should probably move to the top of function. */
663 if (cnp->cn_flags & CNP_SAVESTART)
664 panic("lookup: CNP_SAVESTART");
668 if (cnp->cn_flags & CNP_ISDOTDOT)
669 panic ("relookup: lookup on dot-dot");
672 * We now have a segment name to search for, and a directory to search.
674 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
675 KASSERT(*vpp == NULL, ("leaf should be empty"));
676 if (error != EJUSTRETURN)
679 * If creating and at end of pathname, then can consider
680 * allowing file to be created.
686 /* ASSERT(dvp == ndp->ni_startdir) */
687 if (cnp->cn_flags & CNP_SAVESTART)
690 * We return with ni_vp NULL to indicate that the entry
691 * doesn't currently exist, leaving a pointer to the
692 * (possibly locked) directory inode in ndp->ni_dvp.
699 * Check for symbolic link
701 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & CNP_FOLLOW),
702 ("relookup: symlink found.\n"));
705 * Disallow directory write attempts on read-only file systems.
708 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) {
712 /* ASSERT(dvp == ndp->ni_startdir) */
713 if (cnp->cn_flags & CNP_SAVESTART)
719 if (vn_canvmio(dp) == TRUE &&
720 ((cnp->cn_flags & (CNP_NOOBJ|CNP_LOCKLEAF)) == CNP_LOCKLEAF))
721 vfs_object_create(dp, cnp->cn_td);
723 if ((cnp->cn_flags & CNP_LOCKLEAF) == 0)
724 VOP_UNLOCK(dp, 0, td);
728 if ((cnp->cn_flags & CNP_LOCKPARENT) && (cnp->cn_flags & CNP_ISLASTCN))
729 VOP_UNLOCK(dvp, 0, td);