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 (cnp->cn_flags & AUDITVNODE1)
376 AUDIT_ARG_UPATH1_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf);
377 if (cnp->cn_flags & AUDITVNODE2)
378 AUDIT_ARG_UPATH2_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf);
379 if (ndp->ni_startdir != NULL && !startdir_used)
380 vrele(ndp->ni_startdir);
387 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 &&
388 lookup_cap_dotdot != 0)
389 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
390 SDT_PROBE4(vfs, namei, lookup, entry, *dpp, cnp->cn_pnbuf,
391 cnp->cn_flags, false);
397 namei_getpath(struct nameidata *ndp)
399 struct componentname *cnp;
405 * Get a buffer for the name to be translated, and copy the
406 * name into the buffer.
408 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
409 if (ndp->ni_segflg == UIO_SYSSPACE) {
410 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
413 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
421 namei_emptypath(struct nameidata *ndp)
423 struct componentname *cnp;
429 MPASS(*cnp->cn_pnbuf == '\0');
430 MPASS((cnp->cn_flags & EMPTYPATH) != 0);
431 MPASS((cnp->cn_flags & (LOCKPARENT | WANTPARENT)) == 0);
433 ndp->ni_resflags |= NIRES_EMPTYPATH;
434 error = namei_setup(ndp, &dp, &pwd);
440 * Usecount on dp already provided by namei_setup.
445 if ((cnp->cn_flags & LOCKLEAF) != 0) {
446 VOP_LOCK(dp, (cnp->cn_flags & LOCKSHARED) != 0 ?
447 LK_SHARED : LK_EXCLUSIVE);
448 if (VN_IS_DOOMED(dp)) {
454 SDT_PROBE4(vfs, namei, lookup, return, 0, ndp->ni_vp, false, ndp);
458 SDT_PROBE4(vfs, namei, lookup, return, error, NULL, false, ndp);
459 namei_cleanup_cnp(cnp);
463 static int __noinline
464 namei_follow_link(struct nameidata *ndp)
469 struct componentname *cnp;
477 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
482 if ((cnp->cn_flags & NOMACCHECK) == 0) {
483 error = mac_vnode_check_readlink(td->td_ucred, ndp->ni_vp);
488 if (ndp->ni_pathlen > 1)
489 cp = uma_zalloc(namei_zone, M_WAITOK);
493 aiov.iov_len = MAXPATHLEN;
494 auio.uio_iov = &aiov;
497 auio.uio_rw = UIO_READ;
498 auio.uio_segflg = UIO_SYSSPACE;
500 auio.uio_resid = MAXPATHLEN;
501 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
503 if (ndp->ni_pathlen > 1)
504 uma_zfree(namei_zone, cp);
507 linklen = MAXPATHLEN - auio.uio_resid;
509 if (ndp->ni_pathlen > 1)
510 uma_zfree(namei_zone, cp);
514 if (linklen + ndp->ni_pathlen > MAXPATHLEN) {
515 if (ndp->ni_pathlen > 1)
516 uma_zfree(namei_zone, cp);
517 error = ENAMETOOLONG;
520 if (ndp->ni_pathlen > 1) {
521 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
522 uma_zfree(namei_zone, cnp->cn_pnbuf);
525 cnp->cn_pnbuf[linklen] = '\0';
526 ndp->ni_pathlen += linklen;
532 * Convert a pathname into a pointer to a locked vnode.
534 * The FOLLOW flag is set when symbolic links are to be followed
535 * when they occur at the end of the name translation process.
536 * Symbolic links are always followed for all other pathname
537 * components other than the last.
539 * The segflg defines whether the name is to be copied from user
540 * space or kernel space.
542 * Overall outline of namei:
545 * get starting directory
546 * while (!done && !error) {
547 * call lookup to search path.
548 * if symbolic link, massage name in buffer and continue
552 namei(struct nameidata *ndp)
554 struct vnode *dp; /* the directory we are searching */
555 struct componentname *cnp;
559 enum cache_fpl_status status;
564 KASSERT((ndp->ni_debugflags & NAMEI_DBG_CALLED) == 0,
565 ("%s: repeated call to namei without NDREINIT", __func__));
566 KASSERT(ndp->ni_debugflags == NAMEI_DBG_INITED,
567 ("%s: bad debugflags %d", __func__, ndp->ni_debugflags));
568 ndp->ni_debugflags |= NAMEI_DBG_CALLED;
569 if (ndp->ni_startdir != NULL)
570 ndp->ni_debugflags |= NAMEI_DBG_HADSTARTDIR;
571 if (cnp->cn_flags & FAILIFEXISTS) {
572 KASSERT(cnp->cn_nameiop == CREATE,
573 ("%s: FAILIFEXISTS passed for op %d", __func__, cnp->cn_nameiop));
575 * The limitation below is to restrict hairy corner cases.
577 KASSERT((cnp->cn_flags & (LOCKPARENT | LOCKLEAF)) == LOCKPARENT,
578 ("%s: FAILIFEXISTS must be passed with LOCKPARENT and without LOCKLEAF",
582 ndp->ni_cnd.cn_cred = td->td_ucred;
583 KASSERT(ndp->ni_resflags == 0, ("%s: garbage in ni_resflags: %x\n",
584 __func__, ndp->ni_resflags));
585 KASSERT(cnp->cn_cred && td->td_proc, ("namei: bad cred/proc"));
586 KASSERT((cnp->cn_flags & NAMEI_INTERNAL_FLAGS) == 0,
587 ("namei: unexpected flags: %" PRIx64 "\n",
588 cnp->cn_flags & NAMEI_INTERNAL_FLAGS));
589 if (cnp->cn_flags & NOCACHE)
590 KASSERT(cnp->cn_nameiop != LOOKUP,
591 ("%s: NOCACHE passed with LOOKUP", __func__));
592 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
593 ndp->ni_startdir->v_type == VBAD);
599 error = namei_getpath(ndp);
600 if (__predict_false(error != 0)) {
601 namei_cleanup_cnp(cnp);
602 SDT_PROBE4(vfs, namei, lookup, return, error, NULL,
607 cnp->cn_nameptr = cnp->cn_pnbuf;
610 if (KTRPOINT(td, KTR_NAMEI)) {
611 ktrnamei(cnp->cn_pnbuf);
614 TSNAMEI(curthread->td_proc->p_pid, cnp->cn_pnbuf);
617 * First try looking up the target without locking any vnodes.
619 * We may need to start from scratch or pick up where it left off.
621 error = cache_fplookup(ndp, &status, &pwd);
623 case CACHE_FPL_STATUS_UNSET:
624 __assert_unreachable();
626 case CACHE_FPL_STATUS_HANDLED:
630 case CACHE_FPL_STATUS_PARTIAL:
631 TAILQ_INIT(&ndp->ni_cap_tracker);
632 dp = ndp->ni_startdir;
634 case CACHE_FPL_STATUS_DESTROYED:
636 error = namei_getpath(ndp);
637 if (__predict_false(error != 0)) {
638 namei_cleanup_cnp(cnp);
641 cnp->cn_nameptr = cnp->cn_pnbuf;
643 case CACHE_FPL_STATUS_ABORTED:
644 TAILQ_INIT(&ndp->ni_cap_tracker);
645 MPASS(ndp->ni_lcf == 0);
646 if (*cnp->cn_pnbuf == '\0') {
647 if ((cnp->cn_flags & EMPTYPATH) != 0) {
648 return (namei_emptypath(ndp));
650 namei_cleanup_cnp(cnp);
651 SDT_PROBE4(vfs, namei, lookup, return, ENOENT, NULL,
655 error = namei_setup(ndp, &dp, &pwd);
657 namei_cleanup_cnp(cnp);
667 ndp->ni_startdir = dp;
668 error = vfs_lookup(ndp);
673 * If not a symbolic link, we're done.
675 if ((cnp->cn_flags & ISSYMLINK) == 0) {
676 SDT_PROBE4(vfs, namei, lookup, return, error,
677 ndp->ni_vp, false, ndp);
678 nameicap_cleanup(ndp);
683 error = namei_follow_link(ndp);
689 * Check if root directory should replace current directory.
691 cnp->cn_nameptr = cnp->cn_pnbuf;
692 if (*(cnp->cn_nameptr) == '/') {
694 error = namei_handle_root(ndp, &dp);
704 SDT_PROBE4(vfs, namei, lookup, return, error, NULL, false, ndp);
705 namei_cleanup_cnp(cnp);
706 nameicap_cleanup(ndp);
712 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
715 if (mp == NULL || ((lkflags & LK_SHARED) &&
716 !(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED))) {
717 lkflags &= ~LK_SHARED;
718 lkflags |= LK_EXCLUSIVE;
720 lkflags |= LK_NODDLKTREAT;
725 needs_exclusive_leaf(struct mount *mp, int flags)
729 * Intermediate nodes can use shared locks, we only need to
730 * force an exclusive lock for leaf nodes.
732 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
735 /* Always use exclusive locks if LOCKSHARED isn't set. */
736 if (!(flags & LOCKSHARED))
740 * For lookups during open(), if the mount point supports
741 * extended shared operations, then use a shared lock for the
742 * leaf node, otherwise use an exclusive lock.
744 if ((flags & ISOPEN) != 0)
745 return (!MNT_EXTENDED_SHARED(mp));
748 * Lookup requests outside of open() that specify LOCKSHARED
749 * only need a shared lock on the leaf vnode.
755 * Various filesystems expect to be able to copy a name component with length
756 * bounded by NAME_MAX into a directory entry buffer of size MAXNAMLEN. Make
757 * sure that these are the same size.
759 _Static_assert(MAXNAMLEN == NAME_MAX,
760 "MAXNAMLEN and NAME_MAX have different values");
762 static int __noinline
763 vfs_lookup_degenerate(struct nameidata *ndp, struct vnode *dp, int wantparent)
765 struct componentname *cnp;
771 cnp->cn_flags |= ISLASTCN;
773 mp = atomic_load_ptr(&dp->v_mount);
774 if (needs_exclusive_leaf(mp, cnp->cn_flags)) {
775 cnp->cn_lkflags &= ~LK_SHARED;
776 cnp->cn_lkflags |= LK_EXCLUSIVE;
780 compute_cn_lkflags(mp, cnp->cn_lkflags | LK_RETRY,
783 if (dp->v_type != VDIR) {
787 if (cnp->cn_nameiop != LOOKUP) {
798 if (cnp->cn_flags & AUDITVNODE1)
799 AUDIT_ARG_VNODE1(dp);
800 else if (cnp->cn_flags & AUDITVNODE2)
801 AUDIT_ARG_VNODE2(dp);
803 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
812 * FAILIFEXISTS handling.
814 * XXX namei called with LOCKPARENT but not LOCKLEAF has the strange
815 * behaviour of leaving the vnode unlocked if the target is the same
816 * vnode as the parent.
818 static int __noinline
819 vfs_lookup_failifexists(struct nameidata *ndp)
821 struct componentname *cnp __diagused;
825 MPASS((cnp->cn_flags & ISSYMLINK) == 0);
826 if (ndp->ni_vp == ndp->ni_dvp)
839 * This is a very central and rather complicated routine.
841 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
842 * The starting directory is taken from ni_startdir. The pathname is
843 * descended until done, or a symbolic link is encountered. The variable
844 * ni_more is clear if the path is completed; it is set to one if a
845 * symbolic link needing interpretation is encountered.
847 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
848 * whether the name is to be looked up, created, renamed, or deleted.
849 * When CREATE, RENAME, or DELETE is specified, information usable in
850 * creating, renaming, or deleting a directory entry may be calculated.
851 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
852 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
853 * returned unlocked. Otherwise the parent directory is not returned. If
854 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
855 * the target is returned locked, otherwise it is returned unlocked.
856 * When creating or renaming and LOCKPARENT is specified, the target may not
857 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
859 * Overall outline of lookup:
862 * identify next component of name at ndp->ni_ptr
863 * handle degenerate case where name is null string
864 * if .. and crossing mount points and on mounted filesys, find parent
865 * call VOP_LOOKUP routine for next component name
866 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
867 * component vnode returned in ni_vp (if it exists), locked.
868 * if result vnode is mounted on and crossing mount points,
869 * find mounted on vnode
870 * if more components of name, do next level at dirloop
871 * return the answer in ni_vp, locked if LOCKLEAF set
872 * if LOCKPARENT set, return locked parent in ni_dvp
873 * if WANTPARENT set, return unlocked parent in ni_dvp
876 vfs_lookup(struct nameidata *ndp)
878 char *cp; /* pointer into pathname argument */
879 char *prev_ni_next; /* saved ndp->ni_next */
880 char *nulchar; /* location of '\0' in cn_pnbuf */
881 char *lastchar; /* location of the last character */
882 struct vnode *dp = NULL; /* the directory we are searching */
883 struct vnode *tdp; /* saved dp */
884 struct mount *mp; /* mount table entry */
886 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
887 int docache; /* == 0 do not cache last component */
888 int wantparent; /* 1 => wantparent or lockparent flag */
889 int rdonly; /* lookup read-only flag bit */
891 int dpunlocked = 0; /* dp has already been unlocked */
892 int relookup = 0; /* do not consume the path component */
893 struct componentname *cnp = &ndp->ni_cnd;
900 * Setup: break out flag bits into variables.
903 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
904 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
905 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
907 * When set to zero, docache causes the last component of the
908 * pathname to be deleted from the cache and the full lookup
909 * of the name to be done (via VOP_CACHEDLOOKUP()). Often
910 * filesystems need some pre-computed values that are made
911 * during the full lookup, for instance UFS sets dp->i_offset.
913 * The docache variable is set to zero when requested by the
914 * NOCACHE flag and for all modifying operations except CREATE.
916 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
917 if (cnp->cn_nameiop == DELETE ||
918 (wantparent && cnp->cn_nameiop != CREATE &&
919 cnp->cn_nameiop != LOOKUP))
921 rdonly = cnp->cn_flags & RDONLY;
922 cnp->cn_flags &= ~ISSYMLINK;
925 cnp->cn_lkflags = LK_SHARED;
926 dp = ndp->ni_startdir;
927 ndp->ni_startdir = NULLVP;
930 * Leading slashes, if any, are supposed to be skipped by the caller.
932 MPASS(cnp->cn_nameptr[0] != '/');
935 * Check for degenerate name (e.g. / or "") which is a way of talking
936 * about a directory, e.g. like "/." or ".".
938 if (__predict_false(cnp->cn_nameptr[0] == '\0')) {
939 error = vfs_lookup_degenerate(ndp, dp, wantparent);
941 goto success_right_lock;
946 * Nul-out trailing slashes (e.g., "foo///" -> "foo").
948 * This must be done before VOP_LOOKUP() because some fs's don't know
949 * about trailing slashes. Remember if there were trailing slashes to
950 * handle symlinks, existing non-directories and non-existing files
951 * that won't be directories specially later.
953 MPASS(ndp->ni_pathlen >= 2);
954 lastchar = &cnp->cn_nameptr[ndp->ni_pathlen - 2];
955 if (*lastchar == '/') {
956 while (lastchar >= cnp->cn_pnbuf) {
960 if (*lastchar != '/') {
964 cnp->cn_flags |= TRAILINGSLASH;
968 * We use shared locks until we hit the parent of the last cn then
969 * we adjust based on the requesting flags.
972 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
977 * Search a new directory.
979 * The last component of the filename is left accessible via
980 * cnp->cn_nameptr. It has to be freed with a call to NDFREE*.
982 * Store / as a temporary sentinel so that we only have one character
983 * to test for. Pathnames tend to be short so this should not be
984 * resulting in cache misses.
986 nulchar = &cnp->cn_nameptr[ndp->ni_pathlen - 1];
987 KASSERT(*nulchar == '\0',
988 ("%s: expected nul at %p; string [%s]\n", __func__, nulchar,
991 for (cp = cnp->cn_nameptr; *cp != '/'; cp++) {
993 ("%s: encountered unexpected nul; string [%s]\n", __func__,
998 cnp->cn_namelen = cp - cnp->cn_nameptr;
999 if (__predict_false(cnp->cn_namelen > NAME_MAX)) {
1000 error = ENAMETOOLONG;
1003 #ifdef NAMEI_DIAGNOSTIC
1006 printf("{%s}: ", cnp->cn_nameptr);
1009 prev_ni_pathlen = ndp->ni_pathlen;
1010 ndp->ni_pathlen -= cnp->cn_namelen;
1011 KASSERT(ndp->ni_pathlen <= PATH_MAX,
1012 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
1013 prev_ni_next = ndp->ni_next;
1017 * Something else should be clearing this.
1019 cnp->cn_flags &= ~(ISDOTDOT|ISLASTCN);
1021 cnp->cn_flags |= MAKEENTRY;
1022 if (*cp == '\0' && docache == 0)
1023 cnp->cn_flags &= ~MAKEENTRY;
1024 if (cnp->cn_namelen == 2 &&
1025 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
1026 cnp->cn_flags |= ISDOTDOT;
1027 if (*ndp->ni_next == 0) {
1028 cnp->cn_flags |= ISLASTCN;
1030 if (__predict_false(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
1031 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))) {
1037 nameicap_tracker_add(ndp, dp);
1040 * Make sure degenerate names don't get here, their handling was
1041 * previously found in this spot.
1043 MPASS(cnp->cn_nameptr[0] != '\0');
1046 * Handle "..": five special cases.
1047 * 0. If doing a capability lookup and lookup_cap_dotdot is
1048 * disabled, return ENOTCAPABLE.
1049 * 1. Return an error if this is the last component of
1050 * the name and the operation is DELETE or RENAME.
1051 * 2. If at root directory (e.g. after chroot)
1052 * or at absolute root directory
1053 * then ignore it so can't get out.
1054 * 3. If this vnode is the root of a mounted
1055 * filesystem, then replace it with the
1056 * vnode which was mounted on so we take the
1057 * .. in the other filesystem.
1058 * 4. If the vnode is the top directory of
1059 * the jail or chroot, don't let them out.
1060 * 5. If doing a capability lookup and lookup_cap_dotdot is
1061 * enabled, return ENOTCAPABLE if the lookup would escape
1062 * from the initial file descriptor directory. Checks are
1063 * done by ensuring that namei() already traversed the
1064 * result of dotdot lookup.
1066 if (cnp->cn_flags & ISDOTDOT) {
1067 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT))
1068 == NI_LCF_STRICTRELATIVE) {
1070 if (KTRPOINT(curthread, KTR_CAPFAIL))
1071 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1073 error = ENOTCAPABLE;
1076 if ((cnp->cn_flags & ISLASTCN) != 0 &&
1077 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1082 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
1084 if (dp == pr->pr_root)
1086 bool isroot = dp == ndp->ni_rootdir ||
1087 dp == ndp->ni_topdir || dp == rootvnode ||
1089 if (isroot && (ndp->ni_lcf &
1090 NI_LCF_STRICTRELATIVE) != 0) {
1091 error = ENOTCAPABLE;
1094 if (isroot || ((dp->v_vflag & VV_ROOT) != 0 &&
1095 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
1101 if ((dp->v_vflag & VV_ROOT) == 0)
1103 if (VN_IS_DOOMED(dp)) { /* forced unmount */
1108 dp = dp->v_mount->mnt_vnodecovered;
1112 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
1113 LK_RETRY, ISDOTDOT));
1114 error = nameicap_check_dotdot(ndp, dp);
1118 if (KTRPOINT(curthread, KTR_CAPFAIL))
1119 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1127 * We now have a segment name to search for, and a directory to search.
1131 error = mac_vnode_check_lookup(cnp->cn_cred, dp, cnp);
1132 if (__predict_false(error))
1137 ASSERT_VOP_LOCKED(dp, "lookup");
1139 * If we have a shared lock we may need to upgrade the lock for the
1142 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
1143 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
1144 vn_lock(dp, LK_UPGRADE|LK_RETRY);
1145 if (VN_IS_DOOMED(dp)) {
1150 * If we're looking up the last component and we need an exclusive
1151 * lock, adjust our lkflags.
1153 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
1154 cnp->cn_lkflags = LK_EXCLUSIVE;
1155 #ifdef NAMEI_DIAGNOSTIC
1156 vn_printf(dp, "lookup in ");
1158 lkflags_save = cnp->cn_lkflags;
1159 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
1161 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
1162 cnp->cn_lkflags = lkflags_save;
1164 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
1165 #ifdef NAMEI_DIAGNOSTIC
1166 printf("not found\n");
1168 if ((error == ENOENT) &&
1169 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
1170 (dp->v_mount->mnt_flag & MNT_UNION)) {
1172 dp = dp->v_mount->mnt_vnodecovered;
1176 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
1177 LK_RETRY, cnp->cn_flags));
1178 nameicap_tracker_add(ndp, dp);
1182 if (error == ERELOOKUP) {
1190 if (error != EJUSTRETURN)
1193 * At this point, we know we're at the end of the
1194 * pathname. If creating / renaming, we can consider
1195 * allowing the file or directory to be created / renamed,
1196 * provided we're not on a read-only filesystem.
1202 /* trailing slash only allowed for directories */
1203 if ((cnp->cn_flags & TRAILINGSLASH) &&
1204 !(cnp->cn_flags & WILLBEDIR)) {
1208 if ((cnp->cn_flags & LOCKPARENT) == 0)
1211 * We return with ni_vp NULL to indicate that the entry
1212 * doesn't currently exist, leaving a pointer to the
1213 * (possibly locked) directory vnode in ndp->ni_dvp.
1219 #ifdef NAMEI_DIAGNOSTIC
1225 * Check for symbolic link
1227 if ((dp->v_type == VLNK) &&
1228 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
1229 *ndp->ni_next == '/')) {
1230 cnp->cn_flags |= ISSYMLINK;
1231 if (VN_IS_DOOMED(dp)) {
1233 * We can't know whether the directory was mounted with
1234 * NOSYMFOLLOW, so we can't follow safely.
1239 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1244 * Symlink code always expects an unlocked dvp.
1246 if (ndp->ni_dvp != ndp->ni_vp) {
1247 VOP_UNLOCK(ndp->ni_dvp);
1248 ni_dvp_unlocked = 1;
1251 } else if ((vn_irflag_read(dp) & VIRF_MOUNTPOINT) != 0) {
1252 if ((cnp->cn_flags & NOCROSSMOUNT) != 0)
1258 * Check to see if the vnode has been mounted on;
1259 * if so find the root of the mounted filesystem.
1262 mp = dp->v_mountedhere;
1264 ("%s: NULL mountpoint for VIRF_MOUNTPOINT vnode", __func__));
1265 crosslock = (dp->v_vflag & VV_CROSSLOCK) != 0;
1266 crosslkflags = compute_cn_lkflags(mp, cnp->cn_lkflags,
1268 if (__predict_false(crosslock)) {
1270 * We are going to be holding the vnode lock, which
1271 * in this case is shared by the root vnode of the
1272 * filesystem mounted at mp, across the call to
1273 * VFS_ROOT(). Make the situation clear to the
1274 * filesystem by passing LK_CANRECURSE if the
1275 * lock is held exclusive, or by clearinng
1276 * LK_NODDLKTREAT to allow recursion on the shared
1277 * lock in the presence of an exclusive waiter.
1279 if (VOP_ISLOCKED(dp) == LK_EXCLUSIVE) {
1280 crosslkflags &= ~LK_SHARED;
1281 crosslkflags |= LK_EXCLUSIVE | LK_CANRECURSE;
1282 } else if ((crosslkflags & LK_EXCLUSIVE) != 0) {
1283 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1284 if (VN_IS_DOOMED(dp)) {
1288 if (dp->v_mountedhere != mp) {
1292 crosslkflags &= ~LK_NODDLKTREAT;
1294 if (vfs_busy(mp, 0) != 0)
1296 if (__predict_true(!crosslock))
1298 if (dp != ndp->ni_dvp)
1302 vrefact(vp_crossmp);
1303 ndp->ni_dvp = vp_crossmp;
1304 error = VFS_ROOT(mp, crosslkflags, &tdp);
1306 if (__predict_false(crosslock))
1308 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
1309 panic("vp_crossmp exclusively locked or reclaimed");
1314 ndp->ni_vp = dp = tdp;
1315 } while ((vn_irflag_read(dp) & VIRF_MOUNTPOINT) != 0);
1319 * Not a symbolic link that we will follow. Continue with the
1320 * next component if there is any; otherwise, we're done.
1322 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1323 ("lookup: invalid path state."));
1326 ndp->ni_pathlen = prev_ni_pathlen;
1327 ndp->ni_next = prev_ni_next;
1328 if (ndp->ni_dvp != dp)
1334 if (cnp->cn_flags & ISDOTDOT) {
1335 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1338 if (KTRPOINT(curthread, KTR_CAPFAIL))
1339 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1344 if (*ndp->ni_next == '/') {
1345 cnp->cn_nameptr = ndp->ni_next;
1346 while (*cnp->cn_nameptr == '/') {
1350 if (ndp->ni_dvp != dp)
1357 * If we're processing a path with a trailing slash,
1358 * check that the end result is a directory.
1360 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1365 * Disallow directory write attempts on read-only filesystems.
1368 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1373 ni_dvp_unlocked = 2;
1374 if (ndp->ni_dvp != dp)
1378 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1379 VOP_UNLOCK(ndp->ni_dvp);
1380 ni_dvp_unlocked = 1;
1383 if (cnp->cn_flags & AUDITVNODE1)
1384 AUDIT_ARG_VNODE1(dp);
1385 else if (cnp->cn_flags & AUDITVNODE2)
1386 AUDIT_ARG_VNODE2(dp);
1388 if ((cnp->cn_flags & LOCKLEAF) == 0)
1392 * FIXME: for lookups which only cross a mount point to fetch the
1393 * root vnode, ni_dvp will be set to vp_crossmp. This can be a problem
1394 * if either WANTPARENT or LOCKPARENT is set.
1397 * Because of shared lookup we may have the vnode shared locked, but
1398 * the caller may want it to be exclusively locked.
1400 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1401 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1402 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1403 if (VN_IS_DOOMED(dp)) {
1409 if (ndp->ni_vp != NULL) {
1410 if ((cnp->cn_flags & ISDOTDOT) == 0)
1411 nameicap_tracker_add(ndp, ndp->ni_vp);
1412 if ((cnp->cn_flags & (FAILIFEXISTS | ISSYMLINK)) == FAILIFEXISTS)
1413 return (vfs_lookup_failifexists(ndp));
1418 if (ni_dvp_unlocked != 2) {
1419 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1433 * relookup - lookup a path name component
1434 * Used by lookup to re-acquire things.
1437 vfs_relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1440 struct vnode *dp = NULL; /* the directory we are searching */
1441 int rdonly; /* lookup read-only flag bit */
1444 KASSERT(cnp->cn_flags & ISLASTCN,
1445 ("relookup: Not given last component."));
1447 * Setup: break out flag bits into variables.
1449 KASSERT((cnp->cn_flags & (LOCKPARENT | WANTPARENT)) != 0,
1450 ("relookup: parent not wanted"));
1451 rdonly = cnp->cn_flags & RDONLY;
1452 cnp->cn_flags &= ~ISSYMLINK;
1454 cnp->cn_lkflags = LK_EXCLUSIVE;
1455 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1458 * Search a new directory.
1460 * See a comment in vfs_lookup for cnp->cn_nameptr.
1462 #ifdef NAMEI_DIAGNOSTIC
1463 printf("{%s}: ", cnp->cn_nameptr);
1467 * Check for "" which represents the root directory after slash
1470 if (cnp->cn_nameptr[0] == '\0') {
1472 * Support only LOOKUP for "/" because lookup()
1473 * can't succeed for CREATE, DELETE and RENAME.
1475 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1476 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1478 if (!(cnp->cn_flags & LOCKLEAF))
1481 /* XXX This should probably move to the top of function. */
1483 panic("lookup: SAVESTART");
1487 if (cnp->cn_flags & ISDOTDOT)
1488 panic ("relookup: lookup on dot-dot");
1491 * We now have a segment name to search for, and a directory to search.
1493 #ifdef NAMEI_DIAGNOSTIC
1494 vn_printf(dp, "search in ");
1496 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1497 KASSERT(*vpp == NULL, ("leaf should be empty"));
1498 if (error != EJUSTRETURN)
1501 * If creating and at end of pathname, then can consider
1502 * allowing file to be created.
1508 /* ASSERT(dvp == ndp->ni_startdir) */
1511 if ((cnp->cn_flags & LOCKPARENT) == 0)
1514 * We return with ni_vp NULL to indicate that the entry
1515 * doesn't currently exist, leaving a pointer to the
1516 * (possibly locked) directory vnode in ndp->ni_dvp.
1524 * Disallow directory write attempts on read-only filesystems.
1527 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1536 * Set the parent lock/ref state to the requested state.
1538 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp)
1541 * Check for symbolic link
1543 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1544 ("relookup: symlink found.\n"));
1546 /* ASSERT(dvp == ndp->ni_startdir) */
1550 if ((cnp->cn_flags & LOCKLEAF) == 0)
1561 * Validate the final state of ndp after the lookup.
1564 NDVALIDATE_impl(struct nameidata *ndp, int line)
1566 struct componentname *cnp;
1569 if (cnp->cn_pnbuf == NULL)
1570 panic("%s: got no buf! called from %d", __func__, line);
1576 * Determine if there is a suitable alternate filename under the specified
1577 * prefix for the specified path. If the create flag is set, then the
1578 * alternate prefix will be used so long as the parent directory exists.
1579 * This is used by the various compatibility ABIs so that Linux binaries prefer
1580 * files under /compat/linux for example. The chosen path (whether under
1581 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1582 * to by pathbuf. The caller is responsible for free'ing the buffer from
1583 * the M_TEMP bucket if one is returned.
1586 kern_alternate_path(const char *prefix, const char *path, enum uio_seg pathseg,
1587 char **pathbuf, int create, int dirfd)
1589 struct nameidata nd, ndroot;
1590 char *ptr, *buf, *cp;
1594 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1597 /* Copy the prefix into the new pathname as a starting point. */
1598 len = strlcpy(buf, prefix, MAXPATHLEN);
1599 if (len >= MAXPATHLEN) {
1604 sz = MAXPATHLEN - len;
1607 /* Append the filename to the prefix. */
1608 if (pathseg == UIO_SYSSPACE)
1609 error = copystr(path, ptr, sz, &len);
1611 error = copyinstr(path, ptr, sz, &len);
1619 /* Only use a prefix with absolute pathnames. */
1625 if (dirfd != AT_FDCWD) {
1627 * We want the original because the "prefix" is
1628 * included in the already opened dirfd.
1630 bcopy(ptr, buf, len);
1635 * We know that there is a / somewhere in this pathname.
1636 * Search backwards for it, to find the file's parent dir
1637 * to see if it exists in the alternate tree. If it does,
1638 * and we want to create a file (cflag is set). We don't
1639 * need to worry about the root comparison in this case.
1643 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1646 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf);
1652 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf);
1659 * We now compare the vnode of the prefix to the one
1660 * vnode asked. If they resolve to be the same, then we
1661 * ignore the match so that the real root gets used.
1662 * This avoids the problem of traversing "../.." to find the
1663 * root directory and never finding it, because "/" resolves
1664 * to the emulation root directory. This is expensive :-(
1666 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix);
1668 /* We shouldn't ever get an error from this namei(). */
1669 error = namei(&ndroot);
1671 if (nd.ni_vp == ndroot.ni_vp)
1674 NDFREE_PNBUF(&ndroot);
1675 vrele(ndroot.ni_vp);
1683 /* If there was an error, use the original path name. */
1685 bcopy(ptr, buf, len);