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 1.12 2004/05/21 15:41:23 drhodus Exp $
43 #include "opt_ktrace.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/vnode.h>
49 #include <sys/mount.h>
50 #include <sys/filedesc.h>
52 #include <sys/namei.h>
53 #include <sys/sysctl.h>
56 #include <sys/ktrace.h>
59 #include <vm/vm_zone.h>
61 static int varsym_enable = 0;
62 SYSCTL_INT(_vfs, OID_AUTO, varsym_enable, CTLFLAG_RW, &varsym_enable, 0,
63 "Enable Variant Symlinks");
66 * Convert a pathname into a pointer to a locked inode.
68 * The CNP_FOLLOW flag is set when symbolic links are to be followed
69 * when they occur at the end of the name translation process.
70 * Symbolic links are always followed for all other pathname
71 * components other than the last.
73 * The segflg defines whether the name is to be copied from user
74 * space or kernel space.
76 * Overall outline of namei:
79 * get starting directory
80 * while (!done && !error) {
81 * call lookup to search path.
82 * if symbolic link, massage name in buffer and continue
86 namei(struct nameidata *ndp)
88 struct filedesc *fdp; /* pointer to file descriptor state */
89 char *cp; /* pointer into pathname argument */
90 struct vnode *dp; /* the directory we are searching */
91 struct iovec aiov; /* uio for reading symbolic links */
94 struct componentname *cnp = &ndp->ni_cnd;
97 KKASSERT(ndp->ni_cnd.cn_td != NULL);
98 p = cnp->cn_td->td_proc;
100 KASSERT(cnp->cn_cred, ("namei: bad cred/proc"));
101 KKASSERT(cnp->cn_cred == p->p_ucred); /* YYY */
102 KASSERT((cnp->cn_nameiop & (~NAMEI_OPMASK)) == 0,
103 ("namei: nameiop contaminated with flags"));
104 KASSERT((cnp->cn_flags & NAMEI_OPMASK) == 0,
105 ("namei: flags contaminated with nameiops"));
109 * Get a buffer for the name to be translated, and copy the
110 * name into the buffer.
112 if ((cnp->cn_flags & CNP_HASBUF) == 0)
113 cnp->cn_pnbuf = zalloc(namei_zone);
114 if (ndp->ni_segflg == UIO_SYSSPACE)
115 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
116 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
118 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
119 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
122 * Don't allow empty pathnames.
123 * POSIX.1 requirement: "" is not a vaild file name.
125 if (!error && *cnp->cn_pnbuf == '\0')
129 zfree(namei_zone, cnp->cn_pnbuf);
135 if (KTRPOINT(cnp->cn_td, KTR_NAMEI))
136 ktrnamei(cnp->cn_td->td_proc->p_tracep, cnp->cn_pnbuf);
140 * Get starting point for the translation.
142 ndp->ni_rootdir = fdp->fd_rdir;
143 ndp->ni_topdir = fdp->fd_jdir;
149 * Check if root directory should replace current directory.
150 * Done at start of translation and after symbolic link.
152 cnp->cn_nameptr = cnp->cn_pnbuf;
153 if (*(cnp->cn_nameptr) == '/') {
155 while (*(cnp->cn_nameptr) == '/') {
159 dp = ndp->ni_rootdir;
162 ndp->ni_startdir = dp;
165 zfree(namei_zone, cnp->cn_pnbuf);
169 * Check for symbolic link
171 if ((cnp->cn_flags & CNP_ISSYMLINK) == 0) {
172 if ((cnp->cn_flags & (CNP_SAVENAME | CNP_SAVESTART)) == 0)
173 zfree(namei_zone, cnp->cn_pnbuf);
175 cnp->cn_flags |= CNP_HASBUF;
177 if (vn_canvmio(ndp->ni_vp) == TRUE &&
178 (cnp->cn_nameiop != NAMEI_DELETE) &&
179 ((cnp->cn_flags & (CNP_NOOBJ|CNP_LOCKLEAF)) ==
181 vfs_object_create(ndp->ni_vp, ndp->ni_cnd.cn_td);
185 if ((cnp->cn_flags & CNP_LOCKPARENT) && ndp->ni_pathlen == 1)
186 VOP_UNLOCK(ndp->ni_dvp, NULL, 0, cnp->cn_td);
187 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
191 if (ndp->ni_pathlen > 1)
192 cp = zalloc(namei_zone);
196 aiov.iov_len = MAXPATHLEN;
197 auio.uio_iov = &aiov;
200 auio.uio_rw = UIO_READ;
201 auio.uio_segflg = UIO_SYSSPACE;
203 auio.uio_resid = MAXPATHLEN;
204 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
206 if (ndp->ni_pathlen > 1)
207 zfree(namei_zone, cp);
210 linklen = MAXPATHLEN - auio.uio_resid;
212 if (ndp->ni_pathlen > 1)
213 zfree(namei_zone, cp);
218 linklen = varsymreplace(cp, linklen, MAXPATHLEN);
220 if (ndp->ni_pathlen > 1)
221 zfree(namei_zone, cp);
222 error = ENAMETOOLONG;
226 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
227 if (ndp->ni_pathlen > 1)
228 zfree(namei_zone, cp);
229 error = ENAMETOOLONG;
232 if (ndp->ni_pathlen > 1) {
233 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
234 zfree(namei_zone, cnp->cn_pnbuf);
237 cnp->cn_pnbuf[linklen] = '\0';
239 ndp->ni_pathlen += linklen;
243 zfree(namei_zone, cnp->cn_pnbuf);
252 * This is a very central and rather complicated routine.
254 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
255 * The starting directory is taken from ni_startdir. The pathname is
256 * descended until done, or a symbolic link is encountered. The variable
257 * ni_more is clear if the path is completed; it is set to one if a
258 * symbolic link needing interpretation is encountered.
260 * The flag argument is NAMEI_LOOKUP, CREATE, RENAME, or DELETE depending on
261 * whether the name is to be looked up, created, renamed, or deleted.
262 * When CREATE, RENAME, or DELETE is specified, information usable in
263 * creating, renaming, or deleting a directory entry may be calculated.
264 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
265 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
266 * returned unlocked. Otherwise the parent directory is not returned. If
267 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
268 * the target is returned locked, otherwise it is returned unlocked.
269 * When creating or renaming and LOCKPARENT is specified, the target may not
270 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
272 * Overall outline of lookup:
275 * identify next component of name at ndp->ni_ptr
276 * handle degenerate case where name is null string
277 * if .. and crossing mount points and on mounted filesys, find parent
278 * call VOP_LOOKUP routine for next component name
279 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
280 * component vnode returned in ni_vp (if it exists), locked.
281 * if result vnode is mounted on and crossing mount points,
282 * find mounted on vnode
283 * if more components of name, do next level at dirloop
284 * return the answer in ni_vp, locked if LOCKLEAF set
285 * if LOCKPARENT set, return locked parent in ni_dvp
286 * if WANTPARENT set, return unlocked parent in ni_dvp
289 lookup(struct nameidata *ndp)
291 char *cp; /* pointer into pathname argument */
292 struct vnode *dp = NULL; /* the directory we are searching */
293 struct vnode *tdp; /* saved dp */
294 struct mount *mp; /* mount table entry */
295 int docache; /* == 0 do not cache last component */
296 int wantparent; /* 1 => wantparent or lockparent flag */
297 int rdonly; /* lookup read-only flag bit */
300 int dpunlocked = 0; /* dp has already been unlocked */
301 struct componentname *cnp = &ndp->ni_cnd;
302 struct thread *td = cnp->cn_td;
305 * Setup: break out flag bits into variables.
307 wantparent = cnp->cn_flags & (CNP_LOCKPARENT | CNP_WANTPARENT);
308 docache = (cnp->cn_flags & CNP_NOCACHE) ^ CNP_NOCACHE;
309 if (cnp->cn_nameiop == NAMEI_DELETE ||
310 (wantparent && cnp->cn_nameiop != NAMEI_CREATE &&
311 cnp->cn_nameiop != NAMEI_LOOKUP))
313 rdonly = cnp->cn_flags & CNP_RDONLY;
315 cnp->cn_flags &= ~CNP_ISSYMLINK;
316 dp = ndp->ni_startdir;
317 ndp->ni_startdir = NULLVP;
318 vn_lock(dp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
322 * Search a new directory.
324 * The last component of the filename is left accessible via
325 * cnp->cn_nameptr for callers that need the name. Callers needing
326 * the name set the CNP_SAVENAME flag. When done, they assume
327 * responsibility for freeing the pathname buffer.
330 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
332 cnp->cn_namelen = cp - cnp->cn_nameptr;
333 if (cnp->cn_namelen > NAME_MAX) {
334 error = ENAMETOOLONG;
337 #ifdef NAMEI_DIAGNOSTIC
340 printf("{%s}: ", cnp->cn_nameptr);
343 ndp->ni_pathlen -= cnp->cn_namelen;
347 * Replace multiple slashes by a single slash and trailing slashes
348 * by a null. This must be done before VOP_LOOKUP() because some
349 * fs's don't know about trailing slashes. Remember if there were
350 * trailing slashes to handle symlinks, existing non-directories
351 * and non-existing files that won't be directories specially later.
354 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
359 *ndp->ni_next = '\0'; /* XXX for direnter() ... */
364 cnp->cn_flags |= CNP_MAKEENTRY;
365 if (*cp == '\0' && docache == 0)
366 cnp->cn_flags &= ~CNP_MAKEENTRY;
367 if (cnp->cn_namelen == 2 &&
368 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
369 cnp->cn_flags |= CNP_ISDOTDOT;
371 cnp->cn_flags &= ~CNP_ISDOTDOT;
372 if (*ndp->ni_next == 0)
373 cnp->cn_flags |= CNP_ISLASTCN;
375 cnp->cn_flags &= ~CNP_ISLASTCN;
379 * Check for degenerate name (e.g. / or "")
380 * which is a way of talking about a directory,
381 * e.g. like "/." or ".".
383 if (cnp->cn_nameptr[0] == '\0') {
384 if (dp->v_type != VDIR) {
388 if (cnp->cn_nameiop != NAMEI_LOOKUP) {
397 if (!(cnp->cn_flags & (CNP_LOCKPARENT | CNP_LOCKLEAF)))
398 VOP_UNLOCK(dp, NULL, 0, cnp->cn_td);
399 /* XXX This should probably move to the top of function. */
400 if (cnp->cn_flags & CNP_SAVESTART)
401 panic("lookup: CNP_SAVESTART");
406 * Handle "..": two special cases.
407 * 1. If at root directory (e.g. after chroot)
408 * or at absolute root directory
409 * then ignore it so can't get out.
410 * 2. If this vnode is the root of a mounted
411 * filesystem, then replace it with the
412 * vnode which was mounted on so we take the
413 * .. in the other file system.
414 * 3. If the vnode is the top directory of
415 * the jail or chroot, don't let them out.
417 if (cnp->cn_flags & CNP_ISDOTDOT) {
419 if (dp == ndp->ni_rootdir ||
420 dp == ndp->ni_topdir ||
427 if ((dp->v_flag & VROOT) == 0 ||
428 (cnp->cn_flags & CNP_NOCROSSMOUNT))
430 if (dp->v_mount == NULL) { /* forced unmount */
435 dp = dp->v_mount->mnt_vnodecovered;
438 vn_lock(dp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
443 * We now have a segment name to search for, and a directory to search.
448 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
449 ASSERT_VOP_LOCKED(dp, "lookup");
450 if ((error = VOP_LOOKUP(dp, NCPNULL, &ndp->ni_vp, NCPPNULL, cnp)) != 0) {
451 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
452 #ifdef NAMEI_DIAGNOSTIC
453 printf("not found\n");
455 if ((error == ENOENT) &&
456 (dp->v_flag & VROOT) && (dp->v_mount != NULL) &&
457 (dp->v_mount->mnt_flag & MNT_UNION)) {
459 dp = dp->v_mount->mnt_vnodecovered;
460 if (cnp->cn_flags & CNP_PDIRUNLOCK)
465 vn_lock(dp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
469 if (error != EJUSTRETURN)
472 * If creating and at end of pathname, then can consider
473 * allowing file to be created.
479 if (*cp == '\0' && trailing_slash &&
480 !(cnp->cn_flags & CNP_WILLBEDIR)) {
485 * We return with ni_vp NULL to indicate that the entry
486 * doesn't currently exist, leaving a pointer to the
487 * (possibly locked) directory inode in ndp->ni_dvp.
489 if (cnp->cn_flags & CNP_SAVESTART) {
490 ndp->ni_startdir = ndp->ni_dvp;
491 vref(ndp->ni_startdir);
495 #ifdef NAMEI_DIAGNOSTIC
499 ASSERT_VOP_LOCKED(ndp->ni_vp, "lookup");
502 * Take into account any additional components consumed by
503 * the underlying filesystem.
505 if (cnp->cn_consume > 0) {
506 cnp->cn_nameptr += cnp->cn_consume;
507 ndp->ni_next += cnp->cn_consume;
508 ndp->ni_pathlen -= cnp->cn_consume;
515 * Check to see if the vnode has been mounted on;
516 * if so find the root of the mounted file system.
518 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
519 (cnp->cn_flags & CNP_NOCROSSMOUNT) == 0) {
520 if (vfs_busy(mp, 0, NULL, td))
522 VOP_UNLOCK(dp, NULL, 0, td);
523 error = VFS_ROOT(mp, &tdp);
529 cache_mount(dp, tdp);
531 ndp->ni_vp = dp = tdp;
535 * Check for symbolic link
537 if ((dp->v_type == VLNK) &&
538 ((cnp->cn_flags & CNP_FOLLOW) || trailing_slash ||
539 *ndp->ni_next == '/')) {
540 cnp->cn_flags |= CNP_ISSYMLINK;
541 if (dp->v_mount == NULL) {
542 /* We can't know whether the directory was mounted with
543 * NOSYMFOLLOW, so we can't follow safely. */
547 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
555 * Check for bogus trailing slashes.
557 if (trailing_slash && dp->v_type != VDIR) {
564 * Not a symbolic link. If more pathname,
565 * continue at next component, else return.
567 if (*ndp->ni_next == '/') {
568 cnp->cn_nameptr = ndp->ni_next;
569 while (*cnp->cn_nameptr == '/') {
573 if (ndp->ni_dvp != ndp->ni_vp)
574 ASSERT_VOP_UNLOCKED(ndp->ni_dvp, "lookup");
579 * Disallow directory write attempts on read-only file systems.
582 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) {
586 if (cnp->cn_flags & CNP_SAVESTART) {
587 ndp->ni_startdir = ndp->ni_dvp;
588 vref(ndp->ni_startdir);
593 if ((cnp->cn_flags & CNP_LOCKLEAF) == 0)
594 VOP_UNLOCK(dp, NULL, 0, td);
598 if ((cnp->cn_flags & (CNP_LOCKPARENT | CNP_PDIRUNLOCK)) == CNP_LOCKPARENT &&
599 *ndp->ni_next == '\0')
600 VOP_UNLOCK(ndp->ni_dvp, NULL, 0, td);
612 * relookup - lookup a path name component
613 * Used by lookup to re-aquire things.
616 relookup(dvp, vpp, cnp)
617 struct vnode *dvp, **vpp;
618 struct componentname *cnp;
620 struct thread *td = cnp->cn_td;
621 struct vnode *dp = 0; /* the directory we are searching */
622 int docache; /* == 0 do not cache last component */
623 int wantparent; /* 1 => wantparent or lockparent flag */
624 int rdonly; /* lookup read-only flag bit */
626 #ifdef NAMEI_DIAGNOSTIC
627 int newhash; /* DEBUG: check name hash */
628 char *cp; /* DEBUG: check name ptr/len */
632 * Setup: break out flag bits into variables.
634 wantparent = cnp->cn_flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
635 docache = (cnp->cn_flags & CNP_NOCACHE) ^ CNP_NOCACHE;
636 if (cnp->cn_nameiop == NAMEI_DELETE ||
637 (wantparent && cnp->cn_nameiop != NAMEI_CREATE))
639 rdonly = cnp->cn_flags & CNP_RDONLY;
640 cnp->cn_flags &= ~CNP_ISSYMLINK;
642 vn_lock(dp, NULL, LK_EXCLUSIVE | LK_RETRY, td);
646 * Search a new directory.
648 * The last component of the filename is left accessible via
649 * cnp->cn_nameptr for callers that need the name. Callers needing
650 * the name set the CNP_SAVENAME flag. When done, they assume
651 * responsibility for freeing the pathname buffer.
653 #ifdef NAMEI_DIAGNOSTIC
654 if (cnp->cn_namelen != cp - cnp->cn_nameptr)
655 panic ("relookup: bad len");
657 panic("relookup: not last component");
658 printf("{%s}: ", cnp->cn_nameptr);
662 * Check for degenerate name (e.g. / or "")
663 * which is a way of talking about a directory,
664 * e.g. like "/." or ".".
666 if (cnp->cn_nameptr[0] == '\0') {
667 if (cnp->cn_nameiop != NAMEI_LOOKUP || wantparent) {
671 if (dp->v_type != VDIR) {
675 if (!(cnp->cn_flags & CNP_LOCKLEAF))
676 VOP_UNLOCK(dp, NULL, 0, td);
678 /* XXX This should probably move to the top of function. */
679 if (cnp->cn_flags & CNP_SAVESTART)
680 panic("lookup: CNP_SAVESTART");
684 if (cnp->cn_flags & CNP_ISDOTDOT)
685 panic ("relookup: lookup on dot-dot");
688 * We now have a segment name to search for, and a directory to search.
690 if ((error = VOP_LOOKUP(dp, NCPNULL, vpp, NCPPNULL, cnp)) != 0) {
691 KASSERT(*vpp == NULL, ("leaf should be empty"));
692 if (error != EJUSTRETURN)
695 * If creating and at end of pathname, then can consider
696 * allowing file to be created.
702 /* ASSERT(dvp == ndp->ni_startdir) */
703 if (cnp->cn_flags & CNP_SAVESTART)
706 * We return with ni_vp NULL to indicate that the entry
707 * doesn't currently exist, leaving a pointer to the
708 * (possibly locked) directory inode in ndp->ni_dvp.
715 * Check for symbolic link
717 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & CNP_FOLLOW),
718 ("relookup: symlink found.\n"));
721 * Disallow directory write attempts on read-only file systems.
724 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) {
728 /* ASSERT(dvp == ndp->ni_startdir) */
729 if (cnp->cn_flags & CNP_SAVESTART)
735 if (vn_canvmio(dp) == TRUE &&
736 ((cnp->cn_flags & (CNP_NOOBJ|CNP_LOCKLEAF)) == CNP_LOCKLEAF))
737 vfs_object_create(dp, cnp->cn_td);
739 if ((cnp->cn_flags & CNP_LOCKLEAF) == 0)
740 VOP_UNLOCK(dp, NULL, 0, td);
744 if ((cnp->cn_flags & CNP_LOCKPARENT) && (cnp->cn_flags & CNP_ISLASTCN))
745 VOP_UNLOCK(dvp, NULL, 0, td);