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
77 SDT_PROVIDER_DEFINE(vfs);
78 SDT_PROBE_DEFINE4(vfs, namei, lookup, entry, "struct vnode *", "char *",
79 "unsigned long", "bool");
80 SDT_PROBE_DEFINE4(vfs, namei, lookup, return, "int", "struct vnode *", "bool",
83 /* Allocation zone for namei. */
84 uma_zone_t namei_zone;
86 /* Placeholder vnode for mp traversal. */
87 static struct vnode *vp_crossmp;
90 crossmp_vop_islocked(struct vop_islocked_args *ap)
97 crossmp_vop_lock1(struct vop_lock1_args *ap)
100 struct lock *lk __unused;
101 const char *file __unused;
102 int flags, line __unused;
110 if ((flags & LK_SHARED) == 0)
111 panic("invalid lock request for crossmp");
113 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER, file, line,
114 flags & LK_INTERLOCK ? &VI_MTX(vp)->lock_object : NULL);
115 WITNESS_LOCK(&lk->lock_object, 0, file, line);
116 if ((flags & LK_INTERLOCK) != 0)
118 LOCK_LOG_LOCK("SLOCK", &lk->lock_object, 0, 0, ap->a_file, line);
123 crossmp_vop_unlock(struct vop_unlock_args *ap)
126 struct lock *lk __unused;
131 WITNESS_UNLOCK(&lk->lock_object, 0, LOCK_FILE, LOCK_LINE);
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);
165 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
167 static int lookup_cap_dotdot = 1;
168 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot, CTLFLAG_RWTUN,
169 &lookup_cap_dotdot, 0,
170 "enables \"..\" components in path lookup in capability mode");
171 static int lookup_cap_dotdot_nonlocal = 1;
172 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot_nonlocal, CTLFLAG_RWTUN,
173 &lookup_cap_dotdot_nonlocal, 0,
174 "enables \"..\" components in path lookup in capability mode "
175 "on non-local mount");
178 nameicap_tracker_add(struct nameidata *ndp, struct vnode *dp)
180 struct nameicap_tracker *nt;
181 struct componentname *cnp;
183 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);
260 cnp->cn_pnbuf = NULL;
261 cnp->cn_nameptr = NULL;
266 namei_handle_root(struct nameidata *ndp, struct vnode **dpp)
268 struct componentname *cnp;
271 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0) {
273 if (KTRPOINT(curthread, KTR_CAPFAIL))
274 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
276 return (ENOTCAPABLE);
278 while (*(cnp->cn_nameptr) == '/') {
282 *dpp = ndp->ni_rootdir;
288 namei_setup(struct nameidata *ndp, struct vnode **dpp, struct pwd **pwdp)
290 struct componentname *cnp;
301 startdir_used = false;
305 #ifdef CAPABILITY_MODE
307 * In capability mode, lookups must be restricted to happen in
308 * the subtree with the root specified by the file descriptor:
309 * - The root must be real file descriptor, not the pseudo-descriptor
311 * - The passed path must be relative and not absolute.
312 * - If lookup_cap_dotdot is disabled, path must not contain the
314 * - If lookup_cap_dotdot is enabled, we verify that all '..'
315 * components lookups result in the directories which were
316 * previously walked by us, which prevents an escape from
319 if (IN_CAPABILITY_MODE(td) && (cnp->cn_flags & NOCAPCHECK) == 0) {
320 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
321 ndp->ni_resflags |= NIRES_STRICTREL;
322 if (ndp->ni_dirfd == AT_FDCWD) {
324 if (KTRPOINT(td, KTR_CAPFAIL))
325 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
334 * Get starting point for the translation.
338 * The reference on ni_rootdir is acquired in the block below to avoid
339 * back-to-back atomics for absolute lookups.
341 ndp->ni_rootdir = pwd->pwd_rdir;
342 ndp->ni_topdir = pwd->pwd_jdir;
344 if (cnp->cn_pnbuf[0] == '/') {
345 ndp->ni_resflags |= NIRES_ABS;
346 error = namei_handle_root(ndp, dpp);
348 if (ndp->ni_startdir != NULL) {
349 *dpp = ndp->ni_startdir;
350 startdir_used = true;
351 } else if (ndp->ni_dirfd == AT_FDCWD) {
352 *dpp = pwd->pwd_cdir;
355 rights = *ndp->ni_rightsneeded;
356 cap_rights_set_one(&rights, CAP_LOOKUP);
358 if (cnp->cn_flags & AUDITVNODE1)
359 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
360 if (cnp->cn_flags & AUDITVNODE2)
361 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
363 * Effectively inlined fgetvp_rights, because
364 * we need to inspect the file as well as
365 * grabbing the vnode. No check for O_PATH,
366 * files to implement its semantic.
368 error = fget_cap(td, ndp->ni_dirfd, &rights,
369 &dfp, &ndp->ni_filecaps);
372 * Preserve the error; it should either be EBADF
373 * or capability-related, both of which can be
374 * safely returned to the caller.
377 if (dfp->f_ops == &badfileops) {
379 } else if (dfp->f_vnode == NULL) {
385 if ((dfp->f_flag & FSEARCH) != 0)
386 cnp->cn_flags |= NOEXECCHECK;
392 * If file descriptor doesn't have all rights,
393 * all lookups relative to it must also be
397 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights,
399 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
400 ndp->ni_filecaps.fc_nioctls != -1) {
401 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
402 ndp->ni_resflags |= NIRES_STRICTREL;
406 if (error == 0 && (*dpp)->v_type != VDIR &&
407 (cnp->cn_pnbuf[0] != '\0' ||
408 (cnp->cn_flags & EMPTYPATH) == 0))
411 if (error == 0 && (cnp->cn_flags & RBENEATH) != 0) {
412 if (cnp->cn_pnbuf[0] == '/') {
414 } else if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) == 0) {
415 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE |
421 * If we are auditing the kernel pathname, save the user pathname.
423 if (cnp->cn_flags & AUDITVNODE1)
424 AUDIT_ARG_UPATH1_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf);
425 if (cnp->cn_flags & AUDITVNODE2)
426 AUDIT_ARG_UPATH2_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf);
427 if (ndp->ni_startdir != NULL && !startdir_used)
428 vrele(ndp->ni_startdir);
435 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 &&
436 lookup_cap_dotdot != 0)
437 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
438 SDT_PROBE4(vfs, namei, lookup, entry, *dpp, cnp->cn_pnbuf,
439 cnp->cn_flags, false);
445 namei_getpath(struct nameidata *ndp)
447 struct componentname *cnp;
453 * Get a buffer for the name to be translated, and copy the
454 * name into the buffer.
456 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
457 if (ndp->ni_segflg == UIO_SYSSPACE) {
458 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
461 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
465 if (__predict_false(error != 0))
469 * Don't allow empty pathnames unless EMPTYPATH is specified.
470 * Caller checks for ENOENT as an indication for the empty path.
472 if (__predict_false(*cnp->cn_pnbuf == '\0'))
475 cnp->cn_nameptr = cnp->cn_pnbuf;
480 namei_emptypath(struct nameidata *ndp)
482 struct componentname *cnp;
488 MPASS(*cnp->cn_pnbuf == '\0');
489 MPASS((cnp->cn_flags & EMPTYPATH) != 0);
490 MPASS((cnp->cn_flags & (LOCKPARENT | WANTPARENT)) == 0);
492 error = namei_setup(ndp, &dp, &pwd);
494 namei_cleanup_cnp(cnp);
500 namei_cleanup_cnp(cnp);
502 ndp->ni_resflags |= NIRES_EMPTYPATH;
504 if ((cnp->cn_flags & LOCKLEAF) != 0) {
505 VOP_LOCK(dp, (cnp->cn_flags & LOCKSHARED) != 0 ?
506 LK_SHARED : LK_EXCLUSIVE);
507 if (VN_IS_DOOMED(dp)) {
513 SDT_PROBE4(vfs, namei, lookup, return, 0, ndp->ni_vp, false, ndp);
517 SDT_PROBE4(vfs, namei, lookup, return, error, NULL, false, ndp);
522 * Convert a pathname into a pointer to a locked vnode.
524 * The FOLLOW flag is set when symbolic links are to be followed
525 * when they occur at the end of the name translation process.
526 * Symbolic links are always followed for all other pathname
527 * components other than the last.
529 * The segflg defines whether the name is to be copied from user
530 * space or kernel space.
532 * Overall outline of namei:
535 * get starting directory
536 * while (!done && !error) {
537 * call lookup to search path.
538 * if symbolic link, massage name in buffer and continue
542 namei(struct nameidata *ndp)
544 char *cp; /* pointer into pathname argument */
545 struct vnode *dp; /* the directory we are searching */
546 struct iovec aiov; /* uio for reading symbolic links */
547 struct componentname *cnp;
552 enum cache_fpl_status status;
557 KASSERT((ndp->ni_debugflags & NAMEI_DBG_CALLED) == 0,
558 ("%s: repeated call to namei without NDREINIT", __func__));
559 KASSERT(ndp->ni_debugflags == NAMEI_DBG_INITED,
560 ("%s: bad debugflags %d", __func__, ndp->ni_debugflags));
561 ndp->ni_debugflags |= NAMEI_DBG_CALLED;
562 if (ndp->ni_startdir != NULL)
563 ndp->ni_debugflags |= NAMEI_DBG_HADSTARTDIR;
564 if (cnp->cn_flags & FAILIFEXISTS) {
565 KASSERT(cnp->cn_nameiop == CREATE,
566 ("%s: FAILIFEXISTS passed for op %d", __func__, cnp->cn_nameiop));
568 * The limitation below is to restrict hairy corner cases.
570 KASSERT((cnp->cn_flags & (LOCKPARENT | LOCKLEAF)) == LOCKPARENT,
571 ("%s: FAILIFEXISTS must be passed with LOCKPARENT and without LOCKLEAF",
577 * While NDINIT may seem like a more natural place to do it, there are
578 * callers which directly modify flags past invoking init.
580 cnp->cn_origflags = cnp->cn_flags;
582 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->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 if (error == ENOENT && (cnp->cn_flags & EMPTYPATH) != 0)
602 return (namei_emptypath(ndp));
603 namei_cleanup_cnp(cnp);
604 SDT_PROBE4(vfs, namei, lookup, return, error, NULL,
610 if (KTRPOINT(td, KTR_NAMEI)) {
611 KASSERT(cnp->cn_thread == curthread,
612 ("namei not using curthread"));
613 ktrnamei(cnp->cn_pnbuf);
618 * First try looking up the target without locking any vnodes.
620 * We may need to start from scratch or pick up where it left off.
622 error = cache_fplookup(ndp, &status, &pwd);
624 case CACHE_FPL_STATUS_UNSET:
625 __assert_unreachable();
627 case CACHE_FPL_STATUS_HANDLED:
631 case CACHE_FPL_STATUS_PARTIAL:
632 TAILQ_INIT(&ndp->ni_cap_tracker);
633 dp = ndp->ni_startdir;
635 case CACHE_FPL_STATUS_DESTROYED:
637 error = namei_getpath(ndp);
638 if (__predict_false(error != 0)) {
639 namei_cleanup_cnp(cnp);
643 case CACHE_FPL_STATUS_ABORTED:
644 TAILQ_INIT(&ndp->ni_cap_tracker);
645 MPASS(ndp->ni_lcf == 0);
646 error = namei_setup(ndp, &dp, &pwd);
648 namei_cleanup_cnp(cnp);
658 ndp->ni_startdir = dp;
664 * If not a symbolic link, we're done.
666 if ((cnp->cn_flags & ISSYMLINK) == 0) {
667 SDT_PROBE4(vfs, namei, lookup, return, error,
668 (error == 0 ? ndp->ni_vp : NULL), false, ndp);
669 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
670 namei_cleanup_cnp(cnp);
672 cnp->cn_flags |= HASBUF;
673 nameicap_cleanup(ndp);
679 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
684 if ((cnp->cn_flags & NOMACCHECK) == 0) {
685 error = mac_vnode_check_readlink(td->td_ucred,
691 if (ndp->ni_pathlen > 1)
692 cp = uma_zalloc(namei_zone, M_WAITOK);
696 aiov.iov_len = MAXPATHLEN;
697 auio.uio_iov = &aiov;
700 auio.uio_rw = UIO_READ;
701 auio.uio_segflg = UIO_SYSSPACE;
703 auio.uio_resid = MAXPATHLEN;
704 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
706 if (ndp->ni_pathlen > 1)
707 uma_zfree(namei_zone, cp);
710 linklen = MAXPATHLEN - auio.uio_resid;
712 if (ndp->ni_pathlen > 1)
713 uma_zfree(namei_zone, cp);
717 if (linklen + ndp->ni_pathlen > MAXPATHLEN) {
718 if (ndp->ni_pathlen > 1)
719 uma_zfree(namei_zone, cp);
720 error = ENAMETOOLONG;
723 if (ndp->ni_pathlen > 1) {
724 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
725 uma_zfree(namei_zone, cnp->cn_pnbuf);
728 cnp->cn_pnbuf[linklen] = '\0';
729 ndp->ni_pathlen += linklen;
733 * Check if root directory should replace current directory.
735 cnp->cn_nameptr = cnp->cn_pnbuf;
736 if (*(cnp->cn_nameptr) == '/') {
738 error = namei_handle_root(ndp, &dp);
748 SDT_PROBE4(vfs, namei, lookup, return, error, NULL, false, ndp);
749 namei_cleanup_cnp(cnp);
750 nameicap_cleanup(ndp);
756 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
759 if (mp == NULL || ((lkflags & LK_SHARED) &&
760 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
761 ((cnflags & ISDOTDOT) &&
762 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
763 lkflags &= ~LK_SHARED;
764 lkflags |= LK_EXCLUSIVE;
766 lkflags |= LK_NODDLKTREAT;
771 needs_exclusive_leaf(struct mount *mp, int flags)
775 * Intermediate nodes can use shared locks, we only need to
776 * force an exclusive lock for leaf nodes.
778 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
781 /* Always use exclusive locks if LOCKSHARED isn't set. */
782 if (!(flags & LOCKSHARED))
786 * For lookups during open(), if the mount point supports
787 * extended shared operations, then use a shared lock for the
788 * leaf node, otherwise use an exclusive lock.
790 if ((flags & ISOPEN) != 0)
791 return (!MNT_EXTENDED_SHARED(mp));
794 * Lookup requests outside of open() that specify LOCKSHARED
795 * only need a shared lock on the leaf vnode.
801 * Various filesystems expect to be able to copy a name component with length
802 * bounded by NAME_MAX into a directory entry buffer of size MAXNAMLEN. Make
803 * sure that these are the same size.
805 _Static_assert(MAXNAMLEN == NAME_MAX,
806 "MAXNAMLEN and NAME_MAX have different values");
810 * This is a very central and rather complicated routine.
812 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
813 * The starting directory is taken from ni_startdir. The pathname is
814 * descended until done, or a symbolic link is encountered. The variable
815 * ni_more is clear if the path is completed; it is set to one if a
816 * symbolic link needing interpretation is encountered.
818 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
819 * whether the name is to be looked up, created, renamed, or deleted.
820 * When CREATE, RENAME, or DELETE is specified, information usable in
821 * creating, renaming, or deleting a directory entry may be calculated.
822 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
823 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
824 * returned unlocked. Otherwise the parent directory is not returned. If
825 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
826 * the target is returned locked, otherwise it is returned unlocked.
827 * When creating or renaming and LOCKPARENT is specified, the target may not
828 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
830 * Overall outline of lookup:
833 * identify next component of name at ndp->ni_ptr
834 * handle degenerate case where name is null string
835 * if .. and crossing mount points and on mounted filesys, find parent
836 * call VOP_LOOKUP routine for next component name
837 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
838 * component vnode returned in ni_vp (if it exists), locked.
839 * if result vnode is mounted on and crossing mount points,
840 * find mounted on vnode
841 * if more components of name, do next level at dirloop
842 * return the answer in ni_vp, locked if LOCKLEAF set
843 * if LOCKPARENT set, return locked parent in ni_dvp
844 * if WANTPARENT set, return unlocked parent in ni_dvp
847 lookup(struct nameidata *ndp)
849 char *cp; /* pointer into pathname argument */
850 char *prev_ni_next; /* saved ndp->ni_next */
851 struct vnode *dp = NULL; /* the directory we are searching */
852 struct vnode *tdp; /* saved dp */
853 struct mount *mp; /* mount table entry */
855 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
856 int docache; /* == 0 do not cache last component */
857 int wantparent; /* 1 => wantparent or lockparent flag */
858 int rdonly; /* lookup read-only flag bit */
860 int dpunlocked = 0; /* dp has already been unlocked */
861 int relookup = 0; /* do not consume the path component */
862 struct componentname *cnp = &ndp->ni_cnd;
867 * Setup: break out flag bits into variables.
870 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
871 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
872 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
874 * When set to zero, docache causes the last component of the
875 * pathname to be deleted from the cache and the full lookup
876 * of the name to be done (via VOP_CACHEDLOOKUP()). Often
877 * filesystems need some pre-computed values that are made
878 * during the full lookup, for instance UFS sets dp->i_offset.
880 * The docache variable is set to zero when requested by the
881 * NOCACHE flag and for all modifying operations except CREATE.
883 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
884 if (cnp->cn_nameiop == DELETE ||
885 (wantparent && cnp->cn_nameiop != CREATE &&
886 cnp->cn_nameiop != LOOKUP))
888 rdonly = cnp->cn_flags & RDONLY;
889 cnp->cn_flags &= ~ISSYMLINK;
892 * We use shared locks until we hit the parent of the last cn then
893 * we adjust based on the requesting flags.
895 cnp->cn_lkflags = LK_SHARED;
896 dp = ndp->ni_startdir;
897 ndp->ni_startdir = NULLVP;
899 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
904 * Search a new directory.
906 * The last component of the filename is left accessible via
907 * cnp->cn_nameptr for callers that need the name. Callers needing
908 * the name set the SAVENAME flag. When done, they assume
909 * responsibility for freeing the pathname buffer.
911 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
913 cnp->cn_namelen = cp - cnp->cn_nameptr;
914 if (cnp->cn_namelen > NAME_MAX) {
915 error = ENAMETOOLONG;
918 #ifdef NAMEI_DIAGNOSTIC
921 printf("{%s}: ", cnp->cn_nameptr);
924 prev_ni_pathlen = ndp->ni_pathlen;
925 ndp->ni_pathlen -= cnp->cn_namelen;
926 KASSERT(ndp->ni_pathlen <= PATH_MAX,
927 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
928 prev_ni_next = ndp->ni_next;
932 * Replace multiple slashes by a single slash and trailing slashes
933 * by a null. This must be done before VOP_LOOKUP() because some
934 * fs's don't know about trailing slashes. Remember if there were
935 * trailing slashes to handle symlinks, existing non-directories
936 * and non-existing files that won't be directories specially later.
938 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
942 *ndp->ni_next = '\0';
943 cnp->cn_flags |= TRAILINGSLASH;
948 cnp->cn_flags |= MAKEENTRY;
949 if (*cp == '\0' && docache == 0)
950 cnp->cn_flags &= ~MAKEENTRY;
951 if (cnp->cn_namelen == 2 &&
952 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
953 cnp->cn_flags |= ISDOTDOT;
955 cnp->cn_flags &= ~ISDOTDOT;
956 if (*ndp->ni_next == 0)
957 cnp->cn_flags |= ISLASTCN;
959 cnp->cn_flags &= ~ISLASTCN;
961 if ((cnp->cn_flags & ISLASTCN) != 0 &&
962 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
963 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
968 nameicap_tracker_add(ndp, dp);
971 * Check for degenerate name (e.g. / or "")
972 * which is a way of talking about a directory,
973 * e.g. like "/." or ".".
975 if (cnp->cn_nameptr[0] == '\0') {
976 if (dp->v_type != VDIR) {
980 if (cnp->cn_nameiop != LOOKUP) {
990 if (cnp->cn_flags & AUDITVNODE1)
991 AUDIT_ARG_VNODE1(dp);
992 else if (cnp->cn_flags & AUDITVNODE2)
993 AUDIT_ARG_VNODE2(dp);
995 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
997 /* XXX This should probably move to the top of function. */
998 if (cnp->cn_flags & SAVESTART)
999 panic("lookup: SAVESTART");
1004 * Handle "..": five special cases.
1005 * 0. If doing a capability lookup and lookup_cap_dotdot is
1006 * disabled, return ENOTCAPABLE.
1007 * 1. Return an error if this is the last component of
1008 * the name and the operation is DELETE or RENAME.
1009 * 2. If at root directory (e.g. after chroot)
1010 * or at absolute root directory
1011 * then ignore it so can't get out.
1012 * 3. If this vnode is the root of a mounted
1013 * filesystem, then replace it with the
1014 * vnode which was mounted on so we take the
1015 * .. in the other filesystem.
1016 * 4. If the vnode is the top directory of
1017 * the jail or chroot, don't let them out.
1018 * 5. If doing a capability lookup and lookup_cap_dotdot is
1019 * enabled, return ENOTCAPABLE if the lookup would escape
1020 * from the initial file descriptor directory. Checks are
1021 * done by ensuring that namei() already traversed the
1022 * result of dotdot lookup.
1024 if (cnp->cn_flags & ISDOTDOT) {
1025 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT))
1026 == NI_LCF_STRICTRELATIVE) {
1028 if (KTRPOINT(curthread, KTR_CAPFAIL))
1029 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1031 error = ENOTCAPABLE;
1034 if ((cnp->cn_flags & ISLASTCN) != 0 &&
1035 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1040 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
1042 if (dp == pr->pr_root)
1044 if (dp == ndp->ni_rootdir ||
1045 dp == ndp->ni_topdir ||
1048 ((dp->v_vflag & VV_ROOT) != 0 &&
1049 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
1055 if ((dp->v_vflag & VV_ROOT) == 0)
1057 if (VN_IS_DOOMED(dp)) { /* forced unmount */
1062 dp = dp->v_mount->mnt_vnodecovered;
1066 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
1067 LK_RETRY, ISDOTDOT));
1068 error = nameicap_check_dotdot(ndp, dp);
1071 if (KTRPOINT(curthread, KTR_CAPFAIL))
1072 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1080 * We now have a segment name to search for, and a directory to search.
1084 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp, cnp);
1090 ASSERT_VOP_LOCKED(dp, "lookup");
1092 * If we have a shared lock we may need to upgrade the lock for the
1095 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
1096 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
1097 vn_lock(dp, LK_UPGRADE|LK_RETRY);
1098 if (VN_IS_DOOMED(dp)) {
1103 * If we're looking up the last component and we need an exclusive
1104 * lock, adjust our lkflags.
1106 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
1107 cnp->cn_lkflags = LK_EXCLUSIVE;
1108 #ifdef NAMEI_DIAGNOSTIC
1109 vn_printf(dp, "lookup in ");
1111 lkflags_save = cnp->cn_lkflags;
1112 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
1114 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
1115 cnp->cn_lkflags = lkflags_save;
1117 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
1118 #ifdef NAMEI_DIAGNOSTIC
1119 printf("not found\n");
1121 if ((error == ENOENT) &&
1122 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
1123 (dp->v_mount->mnt_flag & MNT_UNION)) {
1125 dp = dp->v_mount->mnt_vnodecovered;
1129 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
1130 LK_RETRY, cnp->cn_flags));
1131 nameicap_tracker_add(ndp, dp);
1135 if (error == ERELOOKUP) {
1143 if (error != EJUSTRETURN)
1146 * At this point, we know we're at the end of the
1147 * pathname. If creating / renaming, we can consider
1148 * allowing the file or directory to be created / renamed,
1149 * provided we're not on a read-only filesystem.
1155 /* trailing slash only allowed for directories */
1156 if ((cnp->cn_flags & TRAILINGSLASH) &&
1157 !(cnp->cn_flags & WILLBEDIR)) {
1161 if ((cnp->cn_flags & LOCKPARENT) == 0)
1164 * We return with ni_vp NULL to indicate that the entry
1165 * doesn't currently exist, leaving a pointer to the
1166 * (possibly locked) directory vnode in ndp->ni_dvp.
1168 if (cnp->cn_flags & SAVESTART) {
1169 ndp->ni_startdir = ndp->ni_dvp;
1170 VREF(ndp->ni_startdir);
1176 #ifdef NAMEI_DIAGNOSTIC
1182 * Check to see if the vnode has been mounted on;
1183 * if so find the root of the mounted filesystem.
1185 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
1186 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
1187 if (vfs_busy(mp, 0))
1190 if (dp != ndp->ni_dvp)
1194 vrefact(vp_crossmp);
1195 ndp->ni_dvp = vp_crossmp;
1196 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
1197 cnp->cn_flags), &tdp);
1199 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
1200 panic("vp_crossmp exclusively locked or reclaimed");
1205 ndp->ni_vp = dp = tdp;
1209 * Check for symbolic link
1211 if ((dp->v_type == VLNK) &&
1212 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
1213 *ndp->ni_next == '/')) {
1214 cnp->cn_flags |= ISSYMLINK;
1215 if (VN_IS_DOOMED(dp)) {
1217 * We can't know whether the directory was mounted with
1218 * NOSYMFOLLOW, so we can't follow safely.
1223 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1228 * Symlink code always expects an unlocked dvp.
1230 if (ndp->ni_dvp != ndp->ni_vp) {
1231 VOP_UNLOCK(ndp->ni_dvp);
1232 ni_dvp_unlocked = 1;
1239 * Not a symbolic link that we will follow. Continue with the
1240 * next component if there is any; otherwise, we're done.
1242 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1243 ("lookup: invalid path state."));
1246 ndp->ni_pathlen = prev_ni_pathlen;
1247 ndp->ni_next = prev_ni_next;
1248 if (ndp->ni_dvp != dp)
1254 if (cnp->cn_flags & ISDOTDOT) {
1255 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1258 if (KTRPOINT(curthread, KTR_CAPFAIL))
1259 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1264 if (*ndp->ni_next == '/') {
1265 cnp->cn_nameptr = ndp->ni_next;
1266 while (*cnp->cn_nameptr == '/') {
1270 if (ndp->ni_dvp != dp)
1277 * If we're processing a path with a trailing slash,
1278 * check that the end result is a directory.
1280 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1285 * Disallow directory write attempts on read-only filesystems.
1288 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1292 if (cnp->cn_flags & SAVESTART) {
1293 ndp->ni_startdir = ndp->ni_dvp;
1294 VREF(ndp->ni_startdir);
1297 ni_dvp_unlocked = 2;
1298 if (ndp->ni_dvp != dp)
1302 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1303 VOP_UNLOCK(ndp->ni_dvp);
1304 ni_dvp_unlocked = 1;
1307 if (cnp->cn_flags & AUDITVNODE1)
1308 AUDIT_ARG_VNODE1(dp);
1309 else if (cnp->cn_flags & AUDITVNODE2)
1310 AUDIT_ARG_VNODE2(dp);
1312 if ((cnp->cn_flags & LOCKLEAF) == 0)
1316 * FIXME: for lookups which only cross a mount point to fetch the
1317 * root vnode, ni_dvp will be set to vp_crossmp. This can be a problem
1318 * if either WANTPARENT or LOCKPARENT is set.
1321 * Because of shared lookup we may have the vnode shared locked, but
1322 * the caller may want it to be exclusively locked.
1324 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1325 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1326 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1327 if (VN_IS_DOOMED(dp)) {
1332 if (ndp->ni_vp != NULL) {
1333 if ((cnp->cn_flags & ISDOTDOT) == 0)
1334 nameicap_tracker_add(ndp, ndp->ni_vp);
1335 if ((cnp->cn_flags & (FAILIFEXISTS | ISSYMLINK)) == FAILIFEXISTS)
1341 if (ni_dvp_unlocked != 2) {
1342 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1354 * FAILIFEXISTS handling.
1356 * XXX namei called with LOCKPARENT but not LOCKLEAF has the strange
1357 * behaviour of leaving the vnode unlocked if the target is the same
1358 * vnode as the parent.
1360 MPASS((cnp->cn_flags & ISSYMLINK) == 0);
1361 if (ndp->ni_vp == ndp->ni_dvp)
1368 NDFREE(ndp, NDF_ONLY_PNBUF);
1373 * relookup - lookup a path name component
1374 * Used by lookup to re-acquire things.
1377 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
1379 struct vnode *dp = NULL; /* the directory we are searching */
1380 int rdonly; /* lookup read-only flag bit */
1383 KASSERT(cnp->cn_flags & ISLASTCN,
1384 ("relookup: Not given last component."));
1386 * Setup: break out flag bits into variables.
1388 KASSERT((cnp->cn_flags & (LOCKPARENT | WANTPARENT)) != 0,
1389 ("relookup: parent not wanted"));
1390 rdonly = cnp->cn_flags & RDONLY;
1391 cnp->cn_flags &= ~ISSYMLINK;
1393 cnp->cn_lkflags = LK_EXCLUSIVE;
1394 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1397 * Search a new directory.
1399 * The last component of the filename is left accessible via
1400 * cnp->cn_nameptr for callers that need the name. Callers needing
1401 * the name set the SAVENAME flag. When done, they assume
1402 * responsibility for freeing the pathname buffer.
1404 #ifdef NAMEI_DIAGNOSTIC
1405 printf("{%s}: ", cnp->cn_nameptr);
1409 * Check for "" which represents the root directory after slash
1412 if (cnp->cn_nameptr[0] == '\0') {
1414 * Support only LOOKUP for "/" because lookup()
1415 * can't succeed for CREATE, DELETE and RENAME.
1417 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1418 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1420 if (!(cnp->cn_flags & LOCKLEAF))
1423 /* XXX This should probably move to the top of function. */
1424 if (cnp->cn_flags & SAVESTART)
1425 panic("lookup: SAVESTART");
1429 if (cnp->cn_flags & ISDOTDOT)
1430 panic ("relookup: lookup on dot-dot");
1433 * We now have a segment name to search for, and a directory to search.
1435 #ifdef NAMEI_DIAGNOSTIC
1436 vn_printf(dp, "search in ");
1438 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1439 KASSERT(*vpp == NULL, ("leaf should be empty"));
1440 if (error != EJUSTRETURN)
1443 * If creating and at end of pathname, then can consider
1444 * allowing file to be created.
1450 /* ASSERT(dvp == ndp->ni_startdir) */
1451 if (cnp->cn_flags & SAVESTART)
1453 if ((cnp->cn_flags & LOCKPARENT) == 0)
1456 * We return with ni_vp NULL to indicate that the entry
1457 * doesn't currently exist, leaving a pointer to the
1458 * (possibly locked) directory vnode in ndp->ni_dvp.
1466 * Disallow directory write attempts on read-only filesystems.
1469 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1478 * Set the parent lock/ref state to the requested state.
1480 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp)
1483 * Check for symbolic link
1485 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1486 ("relookup: symlink found.\n"));
1488 /* ASSERT(dvp == ndp->ni_startdir) */
1489 if (cnp->cn_flags & SAVESTART)
1492 if ((cnp->cn_flags & LOCKLEAF) == 0)
1502 * Free data allocated by namei(); see namei(9) for details.
1505 NDFREE_PNBUF(struct nameidata *ndp)
1508 if ((ndp->ni_cnd.cn_flags & HASBUF) != 0) {
1509 MPASS((ndp->ni_cnd.cn_flags & (SAVENAME | SAVESTART)) != 0);
1510 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1511 ndp->ni_cnd.cn_flags &= ~HASBUF;
1516 * NDFREE_PNBUF replacement for callers that know there is no buffer.
1518 * This is a hack. Preferably the VFS layer would not produce anything more
1519 * than it was asked to do. Unfortunately several non-LOOKUP cases can add the
1520 * HASBUF flag to the result. Even then an interface could be implemented where
1521 * the caller specifies what they expected to see in the result and what they
1522 * are going to take care of.
1524 * In the meantime provide this kludge as a trivial replacement for NDFREE_PNBUF
1525 * calls scattered throughout the kernel where we know for a fact the flag must not
1530 NDFREE_NOTHING(struct nameidata *ndp)
1532 struct componentname *cnp;
1535 KASSERT(cnp->cn_nameiop == LOOKUP, ("%s: got non-LOOKUP op %d\n",
1536 __func__, cnp->cn_nameiop));
1537 KASSERT((cnp->cn_flags & (SAVENAME | HASBUF)) == 0,
1538 ("%s: bad flags \%" PRIx64 "\n", __func__, cnp->cn_flags));
1543 (NDFREE)(struct nameidata *ndp, const u_int flags)
1551 if (!(flags & NDF_NO_FREE_PNBUF)) {
1554 if (!(flags & NDF_NO_VP_UNLOCK) &&
1555 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1557 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1558 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1559 ndp->ni_dvp != ndp->ni_vp)
1561 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1570 VOP_UNLOCK(ndp->ni_vp);
1571 if (!(flags & NDF_NO_DVP_RELE) &&
1572 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1581 VOP_UNLOCK(ndp->ni_dvp);
1582 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1583 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1584 vrele(ndp->ni_startdir);
1585 ndp->ni_startdir = NULL;
1591 * Validate the final state of ndp after the lookup.
1593 * Historically filesystems were allowed to modify cn_flags. Most notably they
1594 * can add SAVENAME to the request, resulting in HASBUF and pushing subsequent
1595 * clean up to the consumer. In practice this seems to only concern != LOOKUP
1598 * As a step towards stricter API contract this routine validates the state to
1599 * clean up. Note validation is a work in progress with the intent of becoming
1600 * stricter over time.
1602 #define NDMODIFYINGFLAGS (LOCKLEAF | LOCKPARENT | WANTPARENT | SAVENAME | SAVESTART | HASBUF)
1604 NDVALIDATE(struct nameidata *ndp)
1606 struct componentname *cnp;
1607 u_int64_t used, orig;
1610 orig = cnp->cn_origflags;
1611 used = cnp->cn_flags;
1612 switch (cnp->cn_nameiop) {
1615 * For plain lookup we require strict conformance -- nothing
1616 * to clean up if it was not requested by the caller.
1618 orig &= NDMODIFYINGFLAGS;
1619 used &= NDMODIFYINGFLAGS;
1620 if ((orig & (SAVENAME | SAVESTART)) != 0)
1630 * Some filesystems set SAVENAME to provoke HASBUF, accomodate
1631 * for it until it gets fixed.
1633 orig &= NDMODIFYINGFLAGS;
1634 orig |= (SAVENAME | HASBUF);
1635 used &= NDMODIFYINGFLAGS;
1636 used |= (SAVENAME | HASBUF);
1644 panic("%s: mismatched flags for op %d: added %" PRIx64 ", "
1645 "removed %" PRIx64" (%" PRIx64" != %" PRIx64"; stored %" PRIx64" != %" PRIx64")",
1646 __func__, cnp->cn_nameiop, used & ~orig, orig &~ used,
1647 orig, used, cnp->cn_origflags, cnp->cn_flags);
1652 * Determine if there is a suitable alternate filename under the specified
1653 * prefix for the specified path. If the create flag is set, then the
1654 * alternate prefix will be used so long as the parent directory exists.
1655 * This is used by the various compatibility ABIs so that Linux binaries prefer
1656 * files under /compat/linux for example. The chosen path (whether under
1657 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1658 * to by pathbuf. The caller is responsible for free'ing the buffer from
1659 * the M_TEMP bucket if one is returned.
1662 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1663 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1665 struct nameidata nd, ndroot;
1666 char *ptr, *buf, *cp;
1670 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1673 /* Copy the prefix into the new pathname as a starting point. */
1674 len = strlcpy(buf, prefix, MAXPATHLEN);
1675 if (len >= MAXPATHLEN) {
1680 sz = MAXPATHLEN - len;
1683 /* Append the filename to the prefix. */
1684 if (pathseg == UIO_SYSSPACE)
1685 error = copystr(path, ptr, sz, &len);
1687 error = copyinstr(path, ptr, sz, &len);
1695 /* Only use a prefix with absolute pathnames. */
1701 if (dirfd != AT_FDCWD) {
1703 * We want the original because the "prefix" is
1704 * included in the already opened dirfd.
1706 bcopy(ptr, buf, len);
1711 * We know that there is a / somewhere in this pathname.
1712 * Search backwards for it, to find the file's parent dir
1713 * to see if it exists in the alternate tree. If it does,
1714 * and we want to create a file (cflag is set). We don't
1715 * need to worry about the root comparison in this case.
1719 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1722 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1728 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1735 * We now compare the vnode of the prefix to the one
1736 * vnode asked. If they resolve to be the same, then we
1737 * ignore the match so that the real root gets used.
1738 * This avoids the problem of traversing "../.." to find the
1739 * root directory and never finding it, because "/" resolves
1740 * to the emulation root directory. This is expensive :-(
1742 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1745 /* We shouldn't ever get an error from this namei(). */
1746 error = namei(&ndroot);
1748 if (nd.ni_vp == ndroot.ni_vp)
1751 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1752 vrele(ndroot.ni_vp);
1756 NDFREE(&nd, NDF_ONLY_PNBUF);
1760 /* If there was an error, use the original path name. */
1762 bcopy(ptr, buf, len);