2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)vfs_lookup.c 8.4 (Berkeley) 2/16/94
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include "opt_capsicum.h"
43 #include "opt_ktrace.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/dirent.h>
48 #include <sys/kernel.h>
49 #include <sys/capsicum.h>
50 #include <sys/fcntl.h>
53 #include <sys/mutex.h>
54 #include <sys/namei.h>
55 #include <sys/vnode.h>
56 #include <sys/mount.h>
57 #include <sys/filedesc.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysctl.h>
63 #include <sys/ktrace.h>
66 #include <machine/_inttypes.h>
69 #include <security/audit/audit.h>
70 #include <security/mac/mac_framework.h>
74 #define NAMEI_DIAGNOSTIC 1
75 #undef NAMEI_DIAGNOSTIC
78 static void NDVALIDATE_impl(struct nameidata *, int);
79 #define NDVALIDATE(ndp) NDVALIDATE_impl(ndp, __LINE__)
81 #define NDVALIDATE(ndp)
84 SDT_PROVIDER_DEFINE(vfs);
85 SDT_PROBE_DEFINE4(vfs, namei, lookup, entry, "struct vnode *", "char *",
86 "unsigned long", "bool");
87 SDT_PROBE_DEFINE4(vfs, namei, lookup, return, "int", "struct vnode *", "bool",
90 /* Allocation zone for namei. */
91 uma_zone_t namei_zone;
93 /* Placeholder vnode for mp traversal. */
94 static struct vnode *vp_crossmp;
97 crossmp_vop_islocked(struct vop_islocked_args *ap)
104 crossmp_vop_lock1(struct vop_lock1_args *ap)
107 struct lock *lk __diagused;
114 KASSERT((flags & (LK_SHARED | LK_NOWAIT)) == (LK_SHARED | LK_NOWAIT),
115 ("%s: invalid lock request 0x%x for crossmp", __func__, flags));
117 if ((flags & LK_INTERLOCK) != 0)
119 LOCK_LOG_LOCK("SLOCK", &lk->lock_object, 0, 0, ap->a_file, ap->a_line);
124 crossmp_vop_unlock(struct vop_unlock_args *ap)
127 struct lock *lk __diagused;
132 LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, LOCK_FILE,
137 static struct vop_vector crossmp_vnodeops = {
138 .vop_default = &default_vnodeops,
139 .vop_islocked = crossmp_vop_islocked,
140 .vop_lock1 = crossmp_vop_lock1,
141 .vop_unlock = crossmp_vop_unlock,
144 * VFS_VOP_VECTOR_REGISTER(crossmp_vnodeops) is not used here since the vnode
145 * gets allocated early. See nameiinit for the direct call below.
148 struct nameicap_tracker {
150 TAILQ_ENTRY(nameicap_tracker) nm_link;
153 /* Zone for cap mode tracker elements used for dotdot capability checks. */
154 MALLOC_DEFINE(M_NAMEITRACKER, "namei_tracker", "namei tracking for dotdot");
157 nameiinit(void *dummy __unused)
160 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
162 vfs_vector_op_register(&crossmp_vnodeops);
163 getnewvnode("crossmp", NULL, &crossmp_vnodeops, &vp_crossmp);
164 vp_crossmp->v_state = VSTATE_CONSTRUCTED;
165 vp_crossmp->v_irflag |= VIRF_CROSSMP;
167 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
169 static int lookup_cap_dotdot = 1;
170 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot, CTLFLAG_RWTUN,
171 &lookup_cap_dotdot, 0,
172 "enables \"..\" components in path lookup in capability mode");
173 static int lookup_cap_dotdot_nonlocal = 1;
174 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot_nonlocal, CTLFLAG_RWTUN,
175 &lookup_cap_dotdot_nonlocal, 0,
176 "enables \"..\" components in path lookup in capability mode "
177 "on non-local mount");
180 nameicap_tracker_add(struct nameidata *ndp, struct vnode *dp)
182 struct nameicap_tracker *nt;
184 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp->v_type != VDIR)
186 nt = TAILQ_LAST(&ndp->ni_cap_tracker, nameicap_tracker_head);
187 if (nt != NULL && nt->dp == dp)
189 nt = malloc(sizeof(*nt), M_NAMEITRACKER, M_WAITOK);
192 TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link);
196 nameicap_cleanup_from(struct nameidata *ndp, struct nameicap_tracker *first)
198 struct nameicap_tracker *nt, *nt1;
201 TAILQ_FOREACH_FROM_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) {
202 TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link);
204 free(nt, M_NAMEITRACKER);
209 nameicap_cleanup(struct nameidata *ndp)
211 KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) ||
212 (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative"));
213 nameicap_cleanup_from(ndp, NULL);
217 * For dotdot lookups in capability mode, only allow the component
218 * lookup to succeed if the resulting directory was already traversed
219 * during the operation. This catches situations where already
220 * traversed directory is moved to different parent, and then we walk
221 * over it with dotdots.
223 * Also allow to force failure of dotdot lookups for non-local
224 * filesystems, where external agents might assist local lookups to
225 * escape the compartment.
228 nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp)
230 struct nameicap_tracker *nt;
233 if (dp == NULL || dp->v_type != VDIR || (ndp->ni_lcf &
234 NI_LCF_STRICTRELATIVE) == 0)
236 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0)
237 return (ENOTCAPABLE);
239 if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL &&
240 (mp->mnt_flag & MNT_LOCAL) == 0)
241 return (ENOTCAPABLE);
242 TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head,
245 nt = TAILQ_NEXT(nt, nm_link);
247 nameicap_cleanup_from(ndp, nt);
251 return (ENOTCAPABLE);
255 namei_cleanup_cnp(struct componentname *cnp)
258 uma_zfree(namei_zone, cnp->cn_pnbuf);
259 cnp->cn_pnbuf = NULL;
260 cnp->cn_nameptr = NULL;
264 namei_handle_root(struct nameidata *ndp, struct vnode **dpp)
266 struct componentname *cnp;
269 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0) {
271 if (KTRPOINT(curthread, KTR_CAPFAIL))
272 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
274 return (ENOTCAPABLE);
276 while (*(cnp->cn_nameptr) == '/') {
280 *dpp = ndp->ni_rootdir;
286 namei_setup(struct nameidata *ndp, struct vnode **dpp, struct pwd **pwdp)
288 struct componentname *cnp;
297 startdir_used = false;
301 #ifdef CAPABILITY_MODE
303 * In capability mode, lookups must be restricted to happen in
304 * the subtree with the root specified by the file descriptor:
305 * - The root must be real file descriptor, not the pseudo-descriptor
307 * - The passed path must be relative and not absolute.
308 * - If lookup_cap_dotdot is disabled, path must not contain the
310 * - If lookup_cap_dotdot is enabled, we verify that all '..'
311 * components lookups result in the directories which were
312 * previously walked by us, which prevents an escape from
315 if (IN_CAPABILITY_MODE(td) && (cnp->cn_flags & NOCAPCHECK) == 0) {
316 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
317 ndp->ni_resflags |= NIRES_STRICTREL;
318 if (ndp->ni_dirfd == AT_FDCWD) {
320 if (KTRPOINT(td, KTR_CAPFAIL))
321 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
330 * Get starting point for the translation.
334 * The reference on ni_rootdir is acquired in the block below to avoid
335 * back-to-back atomics for absolute lookups.
337 ndp->ni_rootdir = pwd->pwd_rdir;
338 ndp->ni_topdir = pwd->pwd_jdir;
340 if (cnp->cn_pnbuf[0] == '/') {
341 ndp->ni_resflags |= NIRES_ABS;
342 error = namei_handle_root(ndp, dpp);
344 if (ndp->ni_startdir != NULL) {
345 *dpp = ndp->ni_startdir;
346 startdir_used = true;
347 } else if (ndp->ni_dirfd == AT_FDCWD) {
348 *dpp = pwd->pwd_cdir;
351 if (cnp->cn_flags & AUDITVNODE1)
352 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
353 if (cnp->cn_flags & AUDITVNODE2)
354 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
356 error = fgetvp_lookup(ndp->ni_dirfd, ndp, dpp);
358 if (error == 0 && (*dpp)->v_type != VDIR &&
359 (cnp->cn_pnbuf[0] != '\0' ||
360 (cnp->cn_flags & EMPTYPATH) == 0))
363 if (error == 0 && (cnp->cn_flags & RBENEATH) != 0) {
364 if (cnp->cn_pnbuf[0] == '/') {
366 } else if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) == 0) {
367 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE |
373 * If we are auditing the kernel pathname, save the user pathname.
375 if (AUDITING_TD(td)) {
376 if (cnp->cn_flags & AUDITVNODE1)
377 AUDIT_ARG_UPATH1_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf);
378 if (cnp->cn_flags & AUDITVNODE2)
379 AUDIT_ARG_UPATH2_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf);
381 if (ndp->ni_startdir != NULL && !startdir_used)
382 vrele(ndp->ni_startdir);
389 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 &&
390 lookup_cap_dotdot != 0)
391 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
392 SDT_PROBE4(vfs, namei, lookup, entry, *dpp, cnp->cn_pnbuf,
393 cnp->cn_flags, false);
399 namei_getpath(struct nameidata *ndp)
401 struct componentname *cnp;
407 * Get a buffer for the name to be translated, and copy the
408 * name into the buffer.
410 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
411 if (ndp->ni_segflg == UIO_SYSSPACE) {
412 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
415 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
423 namei_emptypath(struct nameidata *ndp)
425 struct componentname *cnp;
431 MPASS(*cnp->cn_pnbuf == '\0');
432 MPASS((cnp->cn_flags & EMPTYPATH) != 0);
433 MPASS((cnp->cn_flags & (LOCKPARENT | WANTPARENT)) == 0);
435 ndp->ni_resflags |= NIRES_EMPTYPATH;
436 error = namei_setup(ndp, &dp, &pwd);
442 * Usecount on dp already provided by namei_setup.
447 if ((cnp->cn_flags & LOCKLEAF) != 0) {
448 VOP_LOCK(dp, (cnp->cn_flags & LOCKSHARED) != 0 ?
449 LK_SHARED : LK_EXCLUSIVE);
450 if (VN_IS_DOOMED(dp)) {
456 SDT_PROBE4(vfs, namei, lookup, return, 0, ndp->ni_vp, false, ndp);
460 SDT_PROBE4(vfs, namei, lookup, return, error, NULL, false, ndp);
461 namei_cleanup_cnp(cnp);
465 static int __noinline
466 namei_follow_link(struct nameidata *ndp)
471 struct componentname *cnp;
479 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
484 if ((cnp->cn_flags & NOMACCHECK) == 0) {
485 error = mac_vnode_check_readlink(td->td_ucred, ndp->ni_vp);
490 if (ndp->ni_pathlen > 1)
491 cp = uma_zalloc(namei_zone, M_WAITOK);
495 aiov.iov_len = MAXPATHLEN;
496 auio.uio_iov = &aiov;
499 auio.uio_rw = UIO_READ;
500 auio.uio_segflg = UIO_SYSSPACE;
502 auio.uio_resid = MAXPATHLEN;
503 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
505 if (ndp->ni_pathlen > 1)
506 uma_zfree(namei_zone, cp);
509 linklen = MAXPATHLEN - auio.uio_resid;
511 if (ndp->ni_pathlen > 1)
512 uma_zfree(namei_zone, cp);
516 if (linklen + ndp->ni_pathlen > MAXPATHLEN) {
517 if (ndp->ni_pathlen > 1)
518 uma_zfree(namei_zone, cp);
519 error = ENAMETOOLONG;
522 if (ndp->ni_pathlen > 1) {
523 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
524 uma_zfree(namei_zone, cnp->cn_pnbuf);
527 cnp->cn_pnbuf[linklen] = '\0';
528 ndp->ni_pathlen += linklen;
534 * Convert a pathname into a pointer to a locked vnode.
536 * The FOLLOW flag is set when symbolic links are to be followed
537 * when they occur at the end of the name translation process.
538 * Symbolic links are always followed for all other pathname
539 * components other than the last.
541 * The segflg defines whether the name is to be copied from user
542 * space or kernel space.
544 * Overall outline of namei:
547 * get starting directory
548 * while (!done && !error) {
549 * call lookup to search path.
550 * if symbolic link, massage name in buffer and continue
554 namei(struct nameidata *ndp)
556 struct vnode *dp; /* the directory we are searching */
557 struct componentname *cnp;
561 enum cache_fpl_status status;
566 KASSERT((ndp->ni_debugflags & NAMEI_DBG_CALLED) == 0,
567 ("%s: repeated call to namei without NDREINIT", __func__));
568 KASSERT(ndp->ni_debugflags == NAMEI_DBG_INITED,
569 ("%s: bad debugflags %d", __func__, ndp->ni_debugflags));
570 ndp->ni_debugflags |= NAMEI_DBG_CALLED;
571 if (ndp->ni_startdir != NULL)
572 ndp->ni_debugflags |= NAMEI_DBG_HADSTARTDIR;
573 if (cnp->cn_flags & FAILIFEXISTS) {
574 KASSERT(cnp->cn_nameiop == CREATE,
575 ("%s: FAILIFEXISTS passed for op %d", __func__, cnp->cn_nameiop));
577 * The limitation below is to restrict hairy corner cases.
579 KASSERT((cnp->cn_flags & (LOCKPARENT | LOCKLEAF)) == LOCKPARENT,
580 ("%s: FAILIFEXISTS must be passed with LOCKPARENT and without LOCKLEAF",
584 ndp->ni_cnd.cn_cred = td->td_ucred;
585 KASSERT(ndp->ni_resflags == 0, ("%s: garbage in ni_resflags: %x\n",
586 __func__, ndp->ni_resflags));
587 KASSERT(cnp->cn_cred && td->td_proc, ("namei: bad cred/proc"));
588 KASSERT((cnp->cn_flags & NAMEI_INTERNAL_FLAGS) == 0,
589 ("namei: unexpected flags: %" PRIx64 "\n",
590 cnp->cn_flags & NAMEI_INTERNAL_FLAGS));
591 if (cnp->cn_flags & NOCACHE)
592 KASSERT(cnp->cn_nameiop != LOOKUP,
593 ("%s: NOCACHE passed with LOOKUP", __func__));
594 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
595 ndp->ni_startdir->v_type == VBAD);
601 error = namei_getpath(ndp);
602 if (__predict_false(error != 0)) {
603 namei_cleanup_cnp(cnp);
604 SDT_PROBE4(vfs, namei, lookup, return, error, NULL,
609 cnp->cn_nameptr = cnp->cn_pnbuf;
612 if (KTRPOINT(td, KTR_NAMEI)) {
613 ktrnamei(cnp->cn_pnbuf);
616 TSNAMEI(curthread->td_proc->p_pid, cnp->cn_pnbuf);
619 * First try looking up the target without locking any vnodes.
621 * We may need to start from scratch or pick up where it left off.
623 error = cache_fplookup(ndp, &status, &pwd);
625 case CACHE_FPL_STATUS_UNSET:
626 __assert_unreachable();
628 case CACHE_FPL_STATUS_HANDLED:
632 case CACHE_FPL_STATUS_PARTIAL:
633 TAILQ_INIT(&ndp->ni_cap_tracker);
634 dp = ndp->ni_startdir;
636 case CACHE_FPL_STATUS_DESTROYED:
638 error = namei_getpath(ndp);
639 if (__predict_false(error != 0)) {
640 namei_cleanup_cnp(cnp);
643 cnp->cn_nameptr = cnp->cn_pnbuf;
645 case CACHE_FPL_STATUS_ABORTED:
646 TAILQ_INIT(&ndp->ni_cap_tracker);
647 MPASS(ndp->ni_lcf == 0);
648 if (*cnp->cn_pnbuf == '\0') {
649 if ((cnp->cn_flags & EMPTYPATH) != 0) {
650 return (namei_emptypath(ndp));
652 namei_cleanup_cnp(cnp);
653 SDT_PROBE4(vfs, namei, lookup, return, ENOENT, NULL,
657 error = namei_setup(ndp, &dp, &pwd);
659 namei_cleanup_cnp(cnp);
669 ndp->ni_startdir = dp;
670 error = vfs_lookup(ndp);
675 * If not a symbolic link, we're done.
677 if ((cnp->cn_flags & ISSYMLINK) == 0) {
678 SDT_PROBE4(vfs, namei, lookup, return, error,
679 ndp->ni_vp, false, ndp);
680 nameicap_cleanup(ndp);
685 error = namei_follow_link(ndp);
691 * Check if root directory should replace current directory.
693 cnp->cn_nameptr = cnp->cn_pnbuf;
694 if (*(cnp->cn_nameptr) == '/') {
696 error = namei_handle_root(ndp, &dp);
706 SDT_PROBE4(vfs, namei, lookup, return, error, NULL, false, ndp);
707 namei_cleanup_cnp(cnp);
708 nameicap_cleanup(ndp);
714 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
717 if (mp == NULL || ((lkflags & LK_SHARED) &&
718 !(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED))) {
719 lkflags &= ~LK_SHARED;
720 lkflags |= LK_EXCLUSIVE;
722 lkflags |= LK_NODDLKTREAT;
727 needs_exclusive_leaf(struct mount *mp, int flags)
731 * Intermediate nodes can use shared locks, we only need to
732 * force an exclusive lock for leaf nodes.
734 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
737 /* Always use exclusive locks if LOCKSHARED isn't set. */
738 if (!(flags & LOCKSHARED))
742 * For lookups during open(), if the mount point supports
743 * extended shared operations, then use a shared lock for the
744 * leaf node, otherwise use an exclusive lock.
746 if ((flags & ISOPEN) != 0)
747 return (!MNT_EXTENDED_SHARED(mp));
750 * Lookup requests outside of open() that specify LOCKSHARED
751 * only need a shared lock on the leaf vnode.
757 * Various filesystems expect to be able to copy a name component with length
758 * bounded by NAME_MAX into a directory entry buffer of size MAXNAMLEN. Make
759 * sure that these are the same size.
761 _Static_assert(MAXNAMLEN == NAME_MAX,
762 "MAXNAMLEN and NAME_MAX have different values");
764 static int __noinline
765 vfs_lookup_degenerate(struct nameidata *ndp, struct vnode *dp, int wantparent)
767 struct componentname *cnp;
773 cnp->cn_flags |= ISLASTCN;
775 mp = atomic_load_ptr(&dp->v_mount);
776 if (needs_exclusive_leaf(mp, cnp->cn_flags)) {
777 cnp->cn_lkflags &= ~LK_SHARED;
778 cnp->cn_lkflags |= LK_EXCLUSIVE;
782 compute_cn_lkflags(mp, cnp->cn_lkflags | LK_RETRY,
785 if (dp->v_type != VDIR) {
789 if (cnp->cn_nameiop != LOOKUP) {
800 if (cnp->cn_flags & AUDITVNODE1)
801 AUDIT_ARG_VNODE1(dp);
802 else if (cnp->cn_flags & AUDITVNODE2)
803 AUDIT_ARG_VNODE2(dp);
805 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
814 * FAILIFEXISTS handling.
816 * XXX namei called with LOCKPARENT but not LOCKLEAF has the strange
817 * behaviour of leaving the vnode unlocked if the target is the same
818 * vnode as the parent.
820 static int __noinline
821 vfs_lookup_failifexists(struct nameidata *ndp)
823 struct componentname *cnp __diagused;
827 MPASS((cnp->cn_flags & ISSYMLINK) == 0);
828 if (ndp->ni_vp == ndp->ni_dvp)
841 * This is a very central and rather complicated routine.
843 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
844 * The starting directory is taken from ni_startdir. The pathname is
845 * descended until done, or a symbolic link is encountered. The variable
846 * ni_more is clear if the path is completed; it is set to one if a
847 * symbolic link needing interpretation is encountered.
849 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
850 * whether the name is to be looked up, created, renamed, or deleted.
851 * When CREATE, RENAME, or DELETE is specified, information usable in
852 * creating, renaming, or deleting a directory entry may be calculated.
853 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
854 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
855 * returned unlocked. Otherwise the parent directory is not returned. If
856 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
857 * the target is returned locked, otherwise it is returned unlocked.
858 * When creating or renaming and LOCKPARENT is specified, the target may not
859 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
861 * Overall outline of lookup:
864 * identify next component of name at ndp->ni_ptr
865 * handle degenerate case where name is null string
866 * if .. and crossing mount points and on mounted filesys, find parent
867 * call VOP_LOOKUP routine for next component name
868 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
869 * component vnode returned in ni_vp (if it exists), locked.
870 * if result vnode is mounted on and crossing mount points,
871 * find mounted on vnode
872 * if more components of name, do next level at dirloop
873 * return the answer in ni_vp, locked if LOCKLEAF set
874 * if LOCKPARENT set, return locked parent in ni_dvp
875 * if WANTPARENT set, return unlocked parent in ni_dvp
878 vfs_lookup(struct nameidata *ndp)
880 char *cp; /* pointer into pathname argument */
881 char *prev_ni_next; /* saved ndp->ni_next */
882 char *nulchar; /* location of '\0' in cn_pnbuf */
883 char *lastchar; /* location of the last character */
884 struct vnode *dp = NULL; /* the directory we are searching */
885 struct vnode *tdp; /* saved dp */
886 struct mount *mp; /* mount table entry */
888 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
889 int docache; /* == 0 do not cache last component */
890 int wantparent; /* 1 => wantparent or lockparent flag */
891 int rdonly; /* lookup read-only flag bit */
893 int dpunlocked = 0; /* dp has already been unlocked */
894 int relookup = 0; /* do not consume the path component */
895 struct componentname *cnp = &ndp->ni_cnd;
902 * Setup: break out flag bits into variables.
905 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
906 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
907 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
909 * When set to zero, docache causes the last component of the
910 * pathname to be deleted from the cache and the full lookup
911 * of the name to be done (via VOP_CACHEDLOOKUP()). Often
912 * filesystems need some pre-computed values that are made
913 * during the full lookup, for instance UFS sets dp->i_offset.
915 * The docache variable is set to zero when requested by the
916 * NOCACHE flag and for all modifying operations except CREATE.
918 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
919 if (cnp->cn_nameiop == DELETE ||
920 (wantparent && cnp->cn_nameiop != CREATE &&
921 cnp->cn_nameiop != LOOKUP))
923 rdonly = cnp->cn_flags & RDONLY;
924 cnp->cn_flags &= ~ISSYMLINK;
927 cnp->cn_lkflags = LK_SHARED;
928 dp = ndp->ni_startdir;
929 ndp->ni_startdir = NULLVP;
932 * Leading slashes, if any, are supposed to be skipped by the caller.
934 MPASS(cnp->cn_nameptr[0] != '/');
937 * Check for degenerate name (e.g. / or "") which is a way of talking
938 * about a directory, e.g. like "/." or ".".
940 if (__predict_false(cnp->cn_nameptr[0] == '\0')) {
941 error = vfs_lookup_degenerate(ndp, dp, wantparent);
943 goto success_right_lock;
948 * Nul-out trailing slashes (e.g., "foo///" -> "foo").
950 * This must be done before VOP_LOOKUP() because some fs's don't know
951 * about trailing slashes. Remember if there were trailing slashes to
952 * handle symlinks, existing non-directories and non-existing files
953 * that won't be directories specially later.
955 MPASS(ndp->ni_pathlen >= 2);
956 lastchar = &cnp->cn_nameptr[ndp->ni_pathlen - 2];
957 if (*lastchar == '/') {
958 while (lastchar >= cnp->cn_pnbuf) {
962 if (*lastchar != '/') {
966 cnp->cn_flags |= TRAILINGSLASH;
970 * We use shared locks until we hit the parent of the last cn then
971 * we adjust based on the requesting flags.
974 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
979 * Search a new directory.
981 * The last component of the filename is left accessible via
982 * cnp->cn_nameptr. It has to be freed with a call to NDFREE*.
984 * Store / as a temporary sentinel so that we only have one character
985 * to test for. Pathnames tend to be short so this should not be
986 * resulting in cache misses.
988 nulchar = &cnp->cn_nameptr[ndp->ni_pathlen - 1];
989 KASSERT(*nulchar == '\0',
990 ("%s: expected nul at %p; string [%s]\n", __func__, nulchar,
993 for (cp = cnp->cn_nameptr; *cp != '/'; cp++) {
995 ("%s: encountered unexpected nul; string [%s]\n", __func__,
1000 cnp->cn_namelen = cp - cnp->cn_nameptr;
1001 if (__predict_false(cnp->cn_namelen > NAME_MAX)) {
1002 error = ENAMETOOLONG;
1005 #ifdef NAMEI_DIAGNOSTIC
1008 printf("{%s}: ", cnp->cn_nameptr);
1011 prev_ni_pathlen = ndp->ni_pathlen;
1012 ndp->ni_pathlen -= cnp->cn_namelen;
1013 KASSERT(ndp->ni_pathlen <= PATH_MAX,
1014 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
1015 prev_ni_next = ndp->ni_next;
1019 * Something else should be clearing this.
1021 cnp->cn_flags &= ~(ISDOTDOT|ISLASTCN);
1023 cnp->cn_flags |= MAKEENTRY;
1024 if (*cp == '\0' && docache == 0)
1025 cnp->cn_flags &= ~MAKEENTRY;
1026 if (cnp->cn_namelen == 2 &&
1027 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
1028 cnp->cn_flags |= ISDOTDOT;
1029 if (*ndp->ni_next == 0) {
1030 cnp->cn_flags |= ISLASTCN;
1032 if (__predict_false(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
1033 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))) {
1039 nameicap_tracker_add(ndp, dp);
1042 * Make sure degenerate names don't get here, their handling was
1043 * previously found in this spot.
1045 MPASS(cnp->cn_nameptr[0] != '\0');
1048 * Handle "..": five special cases.
1049 * 0. If doing a capability lookup and lookup_cap_dotdot is
1050 * disabled, return ENOTCAPABLE.
1051 * 1. Return an error if this is the last component of
1052 * the name and the operation is DELETE or RENAME.
1053 * 2. If at root directory (e.g. after chroot)
1054 * or at absolute root directory
1055 * then ignore it so can't get out.
1056 * 3. If this vnode is the root of a mounted
1057 * filesystem, then replace it with the
1058 * vnode which was mounted on so we take the
1059 * .. in the other filesystem.
1060 * 4. If the vnode is the top directory of
1061 * the jail or chroot, don't let them out.
1062 * 5. If doing a capability lookup and lookup_cap_dotdot is
1063 * enabled, return ENOTCAPABLE if the lookup would escape
1064 * from the initial file descriptor directory. Checks are
1065 * done by ensuring that namei() already traversed the
1066 * result of dotdot lookup.
1068 if (cnp->cn_flags & ISDOTDOT) {
1069 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT))
1070 == NI_LCF_STRICTRELATIVE) {
1072 if (KTRPOINT(curthread, KTR_CAPFAIL))
1073 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1075 error = ENOTCAPABLE;
1078 if ((cnp->cn_flags & ISLASTCN) != 0 &&
1079 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1084 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
1086 if (dp == pr->pr_root)
1088 bool isroot = dp == ndp->ni_rootdir ||
1089 dp == ndp->ni_topdir || dp == rootvnode ||
1091 if (isroot && (ndp->ni_lcf &
1092 NI_LCF_STRICTRELATIVE) != 0) {
1093 error = ENOTCAPABLE;
1096 if (isroot || ((dp->v_vflag & VV_ROOT) != 0 &&
1097 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
1103 if ((dp->v_vflag & VV_ROOT) == 0)
1105 if (VN_IS_DOOMED(dp)) { /* forced unmount */
1110 dp = dp->v_mount->mnt_vnodecovered;
1114 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
1115 LK_RETRY, ISDOTDOT));
1116 error = nameicap_check_dotdot(ndp, dp);
1120 if (KTRPOINT(curthread, KTR_CAPFAIL))
1121 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1129 * We now have a segment name to search for, and a directory to search.
1133 error = mac_vnode_check_lookup(cnp->cn_cred, dp, cnp);
1134 if (__predict_false(error))
1139 ASSERT_VOP_LOCKED(dp, "lookup");
1141 * If we have a shared lock we may need to upgrade the lock for the
1144 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
1145 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
1146 vn_lock(dp, LK_UPGRADE|LK_RETRY);
1147 if (VN_IS_DOOMED(dp)) {
1152 * If we're looking up the last component and we need an exclusive
1153 * lock, adjust our lkflags.
1155 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
1156 cnp->cn_lkflags = LK_EXCLUSIVE;
1157 #ifdef NAMEI_DIAGNOSTIC
1158 vn_printf(dp, "lookup in ");
1160 lkflags_save = cnp->cn_lkflags;
1161 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
1163 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
1164 cnp->cn_lkflags = lkflags_save;
1166 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
1167 #ifdef NAMEI_DIAGNOSTIC
1168 printf("not found\n");
1170 if ((error == ENOENT) &&
1171 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
1172 (dp->v_mount->mnt_flag & MNT_UNION)) {
1174 dp = dp->v_mount->mnt_vnodecovered;
1178 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
1179 LK_RETRY, cnp->cn_flags));
1180 nameicap_tracker_add(ndp, dp);
1184 if (error == ERELOOKUP) {
1192 if (error != EJUSTRETURN)
1195 * At this point, we know we're at the end of the
1196 * pathname. If creating / renaming, we can consider
1197 * allowing the file or directory to be created / renamed,
1198 * provided we're not on a read-only filesystem.
1204 /* trailing slash only allowed for directories */
1205 if ((cnp->cn_flags & TRAILINGSLASH) &&
1206 !(cnp->cn_flags & WILLBEDIR)) {
1210 if ((cnp->cn_flags & LOCKPARENT) == 0)
1213 * We return with ni_vp NULL to indicate that the entry
1214 * doesn't currently exist, leaving a pointer to the
1215 * (possibly locked) directory vnode in ndp->ni_dvp.
1221 #ifdef NAMEI_DIAGNOSTIC
1227 * Check for symbolic link
1229 if ((dp->v_type == VLNK) &&
1230 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
1231 *ndp->ni_next == '/')) {
1232 cnp->cn_flags |= ISSYMLINK;
1233 if (VN_IS_DOOMED(dp)) {
1235 * We can't know whether the directory was mounted with
1236 * NOSYMFOLLOW, so we can't follow safely.
1241 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1246 * Symlink code always expects an unlocked dvp.
1248 if (ndp->ni_dvp != ndp->ni_vp) {
1249 VOP_UNLOCK(ndp->ni_dvp);
1250 ni_dvp_unlocked = 1;
1253 } else if ((vn_irflag_read(dp) & VIRF_MOUNTPOINT) != 0) {
1254 if ((cnp->cn_flags & NOCROSSMOUNT) != 0)
1260 * Check to see if the vnode has been mounted on;
1261 * if so find the root of the mounted filesystem.
1264 mp = dp->v_mountedhere;
1266 ("%s: NULL mountpoint for VIRF_MOUNTPOINT vnode", __func__));
1267 crosslock = (dp->v_vflag & VV_CROSSLOCK) != 0;
1268 crosslkflags = compute_cn_lkflags(mp, cnp->cn_lkflags,
1270 if (__predict_false(crosslock)) {
1272 * We are going to be holding the vnode lock, which
1273 * in this case is shared by the root vnode of the
1274 * filesystem mounted at mp, across the call to
1275 * VFS_ROOT(). Make the situation clear to the
1276 * filesystem by passing LK_CANRECURSE if the
1277 * lock is held exclusive, or by clearinng
1278 * LK_NODDLKTREAT to allow recursion on the shared
1279 * lock in the presence of an exclusive waiter.
1281 if (VOP_ISLOCKED(dp) == LK_EXCLUSIVE) {
1282 crosslkflags &= ~LK_SHARED;
1283 crosslkflags |= LK_EXCLUSIVE | LK_CANRECURSE;
1284 } else if ((crosslkflags & LK_EXCLUSIVE) != 0) {
1285 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1286 if (VN_IS_DOOMED(dp)) {
1290 if (dp->v_mountedhere != mp) {
1294 crosslkflags &= ~LK_NODDLKTREAT;
1296 if (vfs_busy(mp, 0) != 0)
1298 if (__predict_true(!crosslock))
1300 if (dp != ndp->ni_dvp)
1304 vrefact(vp_crossmp);
1305 ndp->ni_dvp = vp_crossmp;
1306 error = VFS_ROOT(mp, crosslkflags, &tdp);
1308 if (__predict_false(crosslock))
1310 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
1311 panic("vp_crossmp exclusively locked or reclaimed");
1316 ndp->ni_vp = dp = tdp;
1317 } while ((vn_irflag_read(dp) & VIRF_MOUNTPOINT) != 0);
1321 * Not a symbolic link that we will follow. Continue with the
1322 * next component if there is any; otherwise, we're done.
1324 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1325 ("lookup: invalid path state."));
1328 ndp->ni_pathlen = prev_ni_pathlen;
1329 ndp->ni_next = prev_ni_next;
1330 if (ndp->ni_dvp != dp)
1336 if (cnp->cn_flags & ISDOTDOT) {
1337 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1340 if (KTRPOINT(curthread, KTR_CAPFAIL))
1341 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1346 if (*ndp->ni_next == '/') {
1347 cnp->cn_nameptr = ndp->ni_next;
1348 while (*cnp->cn_nameptr == '/') {
1352 if (ndp->ni_dvp != dp)
1359 * If we're processing a path with a trailing slash,
1360 * check that the end result is a directory.
1362 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1367 * Disallow directory write attempts on read-only filesystems.
1370 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1375 ni_dvp_unlocked = 2;
1376 if (ndp->ni_dvp != dp)
1380 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1381 VOP_UNLOCK(ndp->ni_dvp);
1382 ni_dvp_unlocked = 1;
1385 if (cnp->cn_flags & AUDITVNODE1)
1386 AUDIT_ARG_VNODE1(dp);
1387 else if (cnp->cn_flags & AUDITVNODE2)
1388 AUDIT_ARG_VNODE2(dp);
1390 if ((cnp->cn_flags & LOCKLEAF) == 0)
1394 * FIXME: for lookups which only cross a mount point to fetch the
1395 * root vnode, ni_dvp will be set to vp_crossmp. This can be a problem
1396 * if either WANTPARENT or LOCKPARENT is set.
1399 * Because of shared lookup we may have the vnode shared locked, but
1400 * the caller may want it to be exclusively locked.
1402 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1403 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1404 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1405 if (VN_IS_DOOMED(dp)) {
1411 if (ndp->ni_vp != NULL) {
1412 if ((cnp->cn_flags & ISDOTDOT) == 0)
1413 nameicap_tracker_add(ndp, ndp->ni_vp);
1414 if ((cnp->cn_flags & (FAILIFEXISTS | ISSYMLINK)) == FAILIFEXISTS)
1415 return (vfs_lookup_failifexists(ndp));
1420 if (ni_dvp_unlocked != 2) {
1421 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1435 * relookup - lookup a path name component
1436 * Used by lookup to re-acquire things.
1439 vfs_relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1442 struct vnode *dp = NULL; /* the directory we are searching */
1443 int rdonly; /* lookup read-only flag bit */
1446 KASSERT(cnp->cn_flags & ISLASTCN,
1447 ("relookup: Not given last component."));
1449 * Setup: break out flag bits into variables.
1451 KASSERT((cnp->cn_flags & (LOCKPARENT | WANTPARENT)) != 0,
1452 ("relookup: parent not wanted"));
1453 rdonly = cnp->cn_flags & RDONLY;
1454 cnp->cn_flags &= ~ISSYMLINK;
1456 cnp->cn_lkflags = LK_EXCLUSIVE;
1457 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1460 * Search a new directory.
1462 * See a comment in vfs_lookup for cnp->cn_nameptr.
1464 #ifdef NAMEI_DIAGNOSTIC
1465 printf("{%s}: ", cnp->cn_nameptr);
1469 * Check for "" which represents the root directory after slash
1472 if (cnp->cn_nameptr[0] == '\0') {
1474 * Support only LOOKUP for "/" because lookup()
1475 * can't succeed for CREATE, DELETE and RENAME.
1477 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1478 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1480 if (!(cnp->cn_flags & LOCKLEAF))
1483 /* XXX This should probably move to the top of function. */
1485 panic("lookup: SAVESTART");
1489 if (cnp->cn_flags & ISDOTDOT)
1490 panic ("relookup: lookup on dot-dot");
1493 * We now have a segment name to search for, and a directory to search.
1495 #ifdef NAMEI_DIAGNOSTIC
1496 vn_printf(dp, "search in ");
1498 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1499 KASSERT(*vpp == NULL, ("leaf should be empty"));
1500 if (error != EJUSTRETURN)
1503 * If creating and at end of pathname, then can consider
1504 * allowing file to be created.
1510 /* ASSERT(dvp == ndp->ni_startdir) */
1513 if ((cnp->cn_flags & LOCKPARENT) == 0)
1516 * We return with ni_vp NULL to indicate that the entry
1517 * doesn't currently exist, leaving a pointer to the
1518 * (possibly locked) directory vnode in ndp->ni_dvp.
1526 * Disallow directory write attempts on read-only filesystems.
1529 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1538 * Set the parent lock/ref state to the requested state.
1540 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp)
1543 * Check for symbolic link
1545 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1546 ("relookup: symlink found.\n"));
1548 /* ASSERT(dvp == ndp->ni_startdir) */
1552 if ((cnp->cn_flags & LOCKLEAF) == 0)
1563 * Validate the final state of ndp after the lookup.
1566 NDVALIDATE_impl(struct nameidata *ndp, int line)
1568 struct componentname *cnp;
1571 if (cnp->cn_pnbuf == NULL)
1572 panic("%s: got no buf! called from %d", __func__, line);
1578 * Determine if there is a suitable alternate filename under the specified
1579 * prefix for the specified path. If the create flag is set, then the
1580 * alternate prefix will be used so long as the parent directory exists.
1581 * This is used by the various compatibility ABIs so that Linux binaries prefer
1582 * files under /compat/linux for example. The chosen path (whether under
1583 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1584 * to by pathbuf. The caller is responsible for free'ing the buffer from
1585 * the M_TEMP bucket if one is returned.
1588 kern_alternate_path(const char *prefix, const char *path, enum uio_seg pathseg,
1589 char **pathbuf, int create, int dirfd)
1591 struct nameidata nd, ndroot;
1592 char *ptr, *buf, *cp;
1596 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1599 /* Copy the prefix into the new pathname as a starting point. */
1600 len = strlcpy(buf, prefix, MAXPATHLEN);
1601 if (len >= MAXPATHLEN) {
1606 sz = MAXPATHLEN - len;
1609 /* Append the filename to the prefix. */
1610 if (pathseg == UIO_SYSSPACE)
1611 error = copystr(path, ptr, sz, &len);
1613 error = copyinstr(path, ptr, sz, &len);
1621 /* Only use a prefix with absolute pathnames. */
1627 if (dirfd != AT_FDCWD) {
1629 * We want the original because the "prefix" is
1630 * included in the already opened dirfd.
1632 bcopy(ptr, buf, len);
1637 * We know that there is a / somewhere in this pathname.
1638 * Search backwards for it, to find the file's parent dir
1639 * to see if it exists in the alternate tree. If it does,
1640 * and we want to create a file (cflag is set). We don't
1641 * need to worry about the root comparison in this case.
1645 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1648 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf);
1654 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf);
1661 * We now compare the vnode of the prefix to the one
1662 * vnode asked. If they resolve to be the same, then we
1663 * ignore the match so that the real root gets used.
1664 * This avoids the problem of traversing "../.." to find the
1665 * root directory and never finding it, because "/" resolves
1666 * to the emulation root directory. This is expensive :-(
1668 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix);
1670 /* We shouldn't ever get an error from this namei(). */
1671 error = namei(&ndroot);
1673 if (nd.ni_vp == ndroot.ni_vp)
1676 NDFREE_PNBUF(&ndroot);
1677 vrele(ndroot.ni_vp);
1685 /* If there was an error, use the original path name. */
1687 bcopy(ptr, buf, len);