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 1.9 2003/11/09 20:29:59 dillon 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.
124 if (!error && *cnp->cn_pnbuf == '\0')
128 zfree(namei_zone, cnp->cn_pnbuf);
134 if (KTRPOINT(cnp->cn_td, KTR_NAMEI))
135 ktrnamei(cnp->cn_td->td_proc->p_tracep, cnp->cn_pnbuf);
139 * Get starting point for the translation.
141 ndp->ni_rootdir = fdp->fd_rdir;
142 ndp->ni_topdir = fdp->fd_jdir;
148 * Check if root directory should replace current directory.
149 * Done at start of translation and after symbolic link.
151 cnp->cn_nameptr = cnp->cn_pnbuf;
152 if (*(cnp->cn_nameptr) == '/') {
154 while (*(cnp->cn_nameptr) == '/') {
158 dp = ndp->ni_rootdir;
161 ndp->ni_startdir = dp;
164 zfree(namei_zone, cnp->cn_pnbuf);
168 * Check for symbolic link
170 if ((cnp->cn_flags & CNP_ISSYMLINK) == 0) {
171 if ((cnp->cn_flags & (CNP_SAVENAME | CNP_SAVESTART)) == 0)
172 zfree(namei_zone, cnp->cn_pnbuf);
174 cnp->cn_flags |= CNP_HASBUF;
176 if (vn_canvmio(ndp->ni_vp) == TRUE &&
177 (cnp->cn_nameiop != NAMEI_DELETE) &&
178 ((cnp->cn_flags & (CNP_NOOBJ|CNP_LOCKLEAF)) ==
180 vfs_object_create(ndp->ni_vp, ndp->ni_cnd.cn_td);
184 if ((cnp->cn_flags & CNP_LOCKPARENT) && ndp->ni_pathlen == 1)
185 VOP_UNLOCK(ndp->ni_dvp, 0, cnp->cn_td);
186 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
190 if (ndp->ni_pathlen > 1)
191 cp = zalloc(namei_zone);
195 aiov.iov_len = MAXPATHLEN;
196 auio.uio_iov = &aiov;
199 auio.uio_rw = UIO_READ;
200 auio.uio_segflg = UIO_SYSSPACE;
202 auio.uio_resid = MAXPATHLEN;
203 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
205 if (ndp->ni_pathlen > 1)
206 zfree(namei_zone, cp);
209 linklen = MAXPATHLEN - auio.uio_resid;
211 if (ndp->ni_pathlen > 1)
212 zfree(namei_zone, cp);
217 linklen = varsymreplace(cp, linklen, MAXPATHLEN);
219 if (ndp->ni_pathlen > 1)
220 zfree(namei_zone, cp);
221 error = ENAMETOOLONG;
225 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
226 if (ndp->ni_pathlen > 1)
227 zfree(namei_zone, cp);
228 error = ENAMETOOLONG;
231 if (ndp->ni_pathlen > 1) {
232 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
233 zfree(namei_zone, cnp->cn_pnbuf);
236 cnp->cn_pnbuf[linklen] = '\0';
238 ndp->ni_pathlen += linklen;
242 zfree(namei_zone, cnp->cn_pnbuf);
251 * This is a very central and rather complicated routine.
253 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
254 * The starting directory is taken from ni_startdir. The pathname is
255 * descended until done, or a symbolic link is encountered. The variable
256 * ni_more is clear if the path is completed; it is set to one if a
257 * symbolic link needing interpretation is encountered.
259 * The flag argument is NAMEI_LOOKUP, CREATE, RENAME, or DELETE depending on
260 * whether the name is to be looked up, created, renamed, or deleted.
261 * When CREATE, RENAME, or DELETE is specified, information usable in
262 * creating, renaming, or deleting a directory entry may be calculated.
263 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
264 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
265 * returned unlocked. Otherwise the parent directory is not returned. If
266 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
267 * the target is returned locked, otherwise it is returned unlocked.
268 * When creating or renaming and LOCKPARENT is specified, the target may not
269 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
271 * Overall outline of lookup:
274 * identify next component of name at ndp->ni_ptr
275 * handle degenerate case where name is null string
276 * if .. and crossing mount points and on mounted filesys, find parent
277 * call VOP_LOOKUP routine for next component name
278 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
279 * component vnode returned in ni_vp (if it exists), locked.
280 * if result vnode is mounted on and crossing mount points,
281 * find mounted on vnode
282 * if more components of name, do next level at dirloop
283 * return the answer in ni_vp, locked if LOCKLEAF set
284 * if LOCKPARENT set, return locked parent in ni_dvp
285 * if WANTPARENT set, return unlocked parent in ni_dvp
288 lookup(struct nameidata *ndp)
290 char *cp; /* pointer into pathname argument */
291 struct vnode *dp = NULL; /* the directory we are searching */
292 struct vnode *tdp; /* saved dp */
293 struct mount *mp; /* mount table entry */
294 int docache; /* == 0 do not cache last component */
295 int wantparent; /* 1 => wantparent or lockparent flag */
296 int rdonly; /* lookup read-only flag bit */
299 int dpunlocked = 0; /* dp has already been unlocked */
300 struct componentname *cnp = &ndp->ni_cnd;
301 struct thread *td = cnp->cn_td;
304 * Setup: break out flag bits into variables.
306 wantparent = cnp->cn_flags & (CNP_LOCKPARENT | CNP_WANTPARENT);
307 docache = (cnp->cn_flags & CNP_NOCACHE) ^ CNP_NOCACHE;
308 if (cnp->cn_nameiop == NAMEI_DELETE ||
309 (wantparent && cnp->cn_nameiop != NAMEI_CREATE &&
310 cnp->cn_nameiop != NAMEI_LOOKUP))
312 rdonly = cnp->cn_flags & CNP_RDONLY;
314 cnp->cn_flags &= ~CNP_ISSYMLINK;
315 dp = ndp->ni_startdir;
316 ndp->ni_startdir = NULLVP;
317 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, td);
321 * Search a new directory.
323 * The last component of the filename is left accessible via
324 * cnp->cn_nameptr for callers that need the name. Callers needing
325 * the name set the CNP_SAVENAME flag. When done, they assume
326 * responsibility for freeing the pathname buffer.
329 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
331 cnp->cn_namelen = cp - cnp->cn_nameptr;
332 if (cnp->cn_namelen > NAME_MAX) {
333 error = ENAMETOOLONG;
336 #ifdef NAMEI_DIAGNOSTIC
339 printf("{%s}: ", cnp->cn_nameptr);
342 ndp->ni_pathlen -= cnp->cn_namelen;
346 * Replace multiple slashes by a single slash and trailing slashes
347 * by a null. This must be done before VOP_LOOKUP() because some
348 * fs's don't know about trailing slashes. Remember if there were
349 * trailing slashes to handle symlinks, existing non-directories
350 * and non-existing files that won't be directories specially later.
353 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
358 *ndp->ni_next = '\0'; /* XXX for direnter() ... */
363 cnp->cn_flags |= CNP_MAKEENTRY;
364 if (*cp == '\0' && docache == 0)
365 cnp->cn_flags &= ~CNP_MAKEENTRY;
366 if (cnp->cn_namelen == 2 &&
367 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
368 cnp->cn_flags |= CNP_ISDOTDOT;
370 cnp->cn_flags &= ~CNP_ISDOTDOT;
371 if (*ndp->ni_next == 0)
372 cnp->cn_flags |= CNP_ISLASTCN;
374 cnp->cn_flags &= ~CNP_ISLASTCN;
378 * Check for degenerate name (e.g. / or "")
379 * which is a way of talking about a directory,
380 * e.g. like "/." or ".".
382 if (cnp->cn_nameptr[0] == '\0') {
383 if (dp->v_type != VDIR) {
387 if (cnp->cn_nameiop != NAMEI_LOOKUP) {
396 if (!(cnp->cn_flags & (CNP_LOCKPARENT | CNP_LOCKLEAF)))
397 VOP_UNLOCK(dp, 0, cnp->cn_td);
398 /* XXX This should probably move to the top of function. */
399 if (cnp->cn_flags & CNP_SAVESTART)
400 panic("lookup: CNP_SAVESTART");
405 * Handle "..": two special cases.
406 * 1. If at root directory (e.g. after chroot)
407 * or at absolute root directory
408 * then ignore it so can't get out.
409 * 2. If this vnode is the root of a mounted
410 * filesystem, then replace it with the
411 * vnode which was mounted on so we take the
412 * .. in the other file system.
413 * 3. If the vnode is the top directory of
414 * the jail or chroot, don't let them out.
416 if (cnp->cn_flags & CNP_ISDOTDOT) {
418 if (dp == ndp->ni_rootdir ||
419 dp == ndp->ni_topdir ||
426 if ((dp->v_flag & VROOT) == 0 ||
427 (cnp->cn_flags & CNP_NOCROSSMOUNT))
429 if (dp->v_mount == NULL) { /* forced unmount */
434 dp = dp->v_mount->mnt_vnodecovered;
437 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, td);
442 * We now have a segment name to search for, and a directory to search.
447 cnp->cn_flags &= ~CNP_PDIRUNLOCK;
448 ASSERT_VOP_LOCKED(dp, "lookup");
449 if ((error = VOP_LOOKUP(dp, NCPNULL, &ndp->ni_vp, NCPPNULL, cnp)) != 0) {
450 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
451 #ifdef NAMEI_DIAGNOSTIC
452 printf("not found\n");
454 if ((error == ENOENT) &&
455 (dp->v_flag & VROOT) && (dp->v_mount != NULL) &&
456 (dp->v_mount->mnt_flag & MNT_UNION)) {
458 dp = dp->v_mount->mnt_vnodecovered;
459 if (cnp->cn_flags & CNP_PDIRUNLOCK)
464 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, td);
468 if (error != EJUSTRETURN)
471 * If creating and at end of pathname, then can consider
472 * allowing file to be created.
478 if (*cp == '\0' && trailing_slash &&
479 !(cnp->cn_flags & CNP_WILLBEDIR)) {
484 * We return with ni_vp NULL to indicate that the entry
485 * doesn't currently exist, leaving a pointer to the
486 * (possibly locked) directory inode in ndp->ni_dvp.
488 if (cnp->cn_flags & CNP_SAVESTART) {
489 ndp->ni_startdir = ndp->ni_dvp;
490 VREF(ndp->ni_startdir);
494 #ifdef NAMEI_DIAGNOSTIC
498 ASSERT_VOP_LOCKED(ndp->ni_vp, "lookup");
501 * Take into account any additional components consumed by
502 * the underlying filesystem.
504 if (cnp->cn_consume > 0) {
505 cnp->cn_nameptr += cnp->cn_consume;
506 ndp->ni_next += cnp->cn_consume;
507 ndp->ni_pathlen -= cnp->cn_consume;
514 * Check to see if the vnode has been mounted on;
515 * if so find the root of the mounted file system.
517 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
518 (cnp->cn_flags & CNP_NOCROSSMOUNT) == 0) {
519 if (vfs_busy(mp, 0, 0, td))
521 VOP_UNLOCK(dp, 0, td);
522 error = VFS_ROOT(mp, &tdp);
528 cache_mount(dp, tdp);
530 ndp->ni_vp = dp = tdp;
534 * Check for symbolic link
536 if ((dp->v_type == VLNK) &&
537 ((cnp->cn_flags & CNP_FOLLOW) || trailing_slash ||
538 *ndp->ni_next == '/')) {
539 cnp->cn_flags |= CNP_ISSYMLINK;
540 if (dp->v_mount == NULL) {
541 /* We can't know whether the directory was mounted with
542 * NOSYMFOLLOW, so we can't follow safely. */
546 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
554 * Check for bogus trailing slashes.
556 if (trailing_slash && dp->v_type != VDIR) {
563 * Not a symbolic link. If more pathname,
564 * continue at next component, else return.
566 if (*ndp->ni_next == '/') {
567 cnp->cn_nameptr = ndp->ni_next;
568 while (*cnp->cn_nameptr == '/') {
572 if (ndp->ni_dvp != ndp->ni_vp)
573 ASSERT_VOP_UNLOCKED(ndp->ni_dvp, "lookup");
578 * Disallow directory write attempts on read-only file systems.
581 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) {
585 if (cnp->cn_flags & CNP_SAVESTART) {
586 ndp->ni_startdir = ndp->ni_dvp;
587 VREF(ndp->ni_startdir);
592 if ((cnp->cn_flags & CNP_LOCKLEAF) == 0)
593 VOP_UNLOCK(dp, 0, td);
597 if ((cnp->cn_flags & (CNP_LOCKPARENT | CNP_PDIRUNLOCK)) == CNP_LOCKPARENT &&
598 *ndp->ni_next == '\0')
599 VOP_UNLOCK(ndp->ni_dvp, 0, td);
611 * relookup - lookup a path name component
612 * Used by lookup to re-aquire things.
615 relookup(dvp, vpp, cnp)
616 struct vnode *dvp, **vpp;
617 struct componentname *cnp;
619 struct thread *td = cnp->cn_td;
620 struct vnode *dp = 0; /* the directory we are searching */
621 int docache; /* == 0 do not cache last component */
622 int wantparent; /* 1 => wantparent or lockparent flag */
623 int rdonly; /* lookup read-only flag bit */
625 #ifdef NAMEI_DIAGNOSTIC
626 int newhash; /* DEBUG: check name hash */
627 char *cp; /* DEBUG: check name ptr/len */
631 * Setup: break out flag bits into variables.
633 wantparent = cnp->cn_flags & (CNP_LOCKPARENT|CNP_WANTPARENT);
634 docache = (cnp->cn_flags & CNP_NOCACHE) ^ CNP_NOCACHE;
635 if (cnp->cn_nameiop == NAMEI_DELETE ||
636 (wantparent && cnp->cn_nameiop != NAMEI_CREATE))
638 rdonly = cnp->cn_flags & CNP_RDONLY;
639 cnp->cn_flags &= ~CNP_ISSYMLINK;
641 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY, td);
645 * Search a new directory.
647 * The last component of the filename is left accessible via
648 * cnp->cn_nameptr for callers that need the name. Callers needing
649 * the name set the CNP_SAVENAME flag. When done, they assume
650 * responsibility for freeing the pathname buffer.
652 #ifdef NAMEI_DIAGNOSTIC
653 if (cnp->cn_namelen != cp - cnp->cn_nameptr)
654 panic ("relookup: bad len");
656 panic("relookup: not last component");
657 printf("{%s}: ", cnp->cn_nameptr);
661 * Check for degenerate name (e.g. / or "")
662 * which is a way of talking about a directory,
663 * e.g. like "/." or ".".
665 if (cnp->cn_nameptr[0] == '\0') {
666 if (cnp->cn_nameiop != NAMEI_LOOKUP || wantparent) {
670 if (dp->v_type != VDIR) {
674 if (!(cnp->cn_flags & CNP_LOCKLEAF))
675 VOP_UNLOCK(dp, 0, td);
677 /* XXX This should probably move to the top of function. */
678 if (cnp->cn_flags & CNP_SAVESTART)
679 panic("lookup: CNP_SAVESTART");
683 if (cnp->cn_flags & CNP_ISDOTDOT)
684 panic ("relookup: lookup on dot-dot");
687 * We now have a segment name to search for, and a directory to search.
689 if ((error = VOP_LOOKUP(dp, NCPNULL, vpp, NCPPNULL, cnp)) != 0) {
690 KASSERT(*vpp == NULL, ("leaf should be empty"));
691 if (error != EJUSTRETURN)
694 * If creating and at end of pathname, then can consider
695 * allowing file to be created.
701 /* ASSERT(dvp == ndp->ni_startdir) */
702 if (cnp->cn_flags & CNP_SAVESTART)
705 * We return with ni_vp NULL to indicate that the entry
706 * doesn't currently exist, leaving a pointer to the
707 * (possibly locked) directory inode in ndp->ni_dvp.
714 * Check for symbolic link
716 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & CNP_FOLLOW),
717 ("relookup: symlink found.\n"));
720 * Disallow directory write attempts on read-only file systems.
723 (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) {
727 /* ASSERT(dvp == ndp->ni_startdir) */
728 if (cnp->cn_flags & CNP_SAVESTART)
734 if (vn_canvmio(dp) == TRUE &&
735 ((cnp->cn_flags & (CNP_NOOBJ|CNP_LOCKLEAF)) == CNP_LOCKLEAF))
736 vfs_object_create(dp, cnp->cn_td);
738 if ((cnp->cn_flags & CNP_LOCKLEAF) == 0)
739 VOP_UNLOCK(dp, 0, td);
743 if ((cnp->cn_flags & CNP_LOCKPARENT) && (cnp->cn_flags & CNP_ISLASTCN))
744 VOP_UNLOCK(dvp, 0, td);