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/kernel.h>
48 #include <sys/capsicum.h>
49 #include <sys/fcntl.h>
52 #include <sys/mutex.h>
53 #include <sys/namei.h>
54 #include <sys/vnode.h>
55 #include <sys/mount.h>
56 #include <sys/filedesc.h>
59 #include <sys/syscallsubr.h>
60 #include <sys/sysctl.h>
62 #include <sys/ktrace.h>
65 #include <security/audit/audit.h>
66 #include <security/mac/mac_framework.h>
70 #define NAMEI_DIAGNOSTIC 1
71 #undef NAMEI_DIAGNOSTIC
73 SDT_PROVIDER_DECLARE(vfs);
74 SDT_PROBE_DEFINE3(vfs, namei, lookup, entry, "struct vnode *", "char *",
76 SDT_PROBE_DEFINE2(vfs, namei, lookup, return, "int", "struct vnode *");
78 /* Allocation zone for namei. */
79 uma_zone_t namei_zone;
81 /* Placeholder vnode for mp traversal. */
82 static struct vnode *vp_crossmp;
85 crossmp_vop_islocked(struct vop_islocked_args *ap)
92 crossmp_vop_lock1(struct vop_lock1_args *ap)
95 struct lock *lk __unused;
96 const char *file __unused;
97 int flags, line __unused;
105 if ((flags & LK_SHARED) == 0)
106 panic("invalid lock request for crossmp");
108 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER, file, line,
109 flags & LK_INTERLOCK ? &VI_MTX(vp)->lock_object : NULL);
110 WITNESS_LOCK(&lk->lock_object, 0, file, line);
111 if ((flags & LK_INTERLOCK) != 0)
113 LOCK_LOG_LOCK("SLOCK", &lk->lock_object, 0, 0, ap->a_file, line);
118 crossmp_vop_unlock(struct vop_unlock_args *ap)
121 struct lock *lk __unused;
128 if ((flags & LK_INTERLOCK) != 0)
130 WITNESS_UNLOCK(&lk->lock_object, 0, LOCK_FILE, LOCK_LINE);
131 LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, LOCK_FILE,
136 static struct vop_vector crossmp_vnodeops = {
137 .vop_default = &default_vnodeops,
138 .vop_islocked = crossmp_vop_islocked,
139 .vop_lock1 = crossmp_vop_lock1,
140 .vop_unlock = crossmp_vop_unlock,
143 struct nameicap_tracker {
145 TAILQ_ENTRY(nameicap_tracker) nm_link;
148 /* Zone for cap mode tracker elements used for dotdot capability checks. */
149 static uma_zone_t nt_zone;
152 nameiinit(void *dummy __unused)
155 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
157 nt_zone = uma_zcreate("rentr", sizeof(struct nameicap_tracker),
158 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
159 getnewvnode("crossmp", NULL, &crossmp_vnodeops, &vp_crossmp);
161 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
163 static int lookup_cap_dotdot = 1;
164 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot, CTLFLAG_RWTUN,
165 &lookup_cap_dotdot, 0,
166 "enables \"..\" components in path lookup in capability mode");
167 static int lookup_cap_dotdot_nonlocal = 1;
168 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot_nonlocal, CTLFLAG_RWTUN,
169 &lookup_cap_dotdot_nonlocal, 0,
170 "enables \"..\" components in path lookup in capability mode "
171 "on non-local mount");
174 nameicap_tracker_add(struct nameidata *ndp, struct vnode *dp)
176 struct nameicap_tracker *nt;
178 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp->v_type != VDIR)
180 nt = uma_zalloc(nt_zone, M_WAITOK);
183 TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link);
187 nameicap_cleanup(struct nameidata *ndp)
189 struct nameicap_tracker *nt, *nt1;
191 KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) ||
192 (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative"));
193 TAILQ_FOREACH_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) {
194 TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link);
196 uma_zfree(nt_zone, nt);
201 * For dotdot lookups in capability mode, only allow the component
202 * lookup to succeed if the resulting directory was already traversed
203 * during the operation. Also fail dotdot lookups for non-local
204 * filesystems, where external agents might assist local lookups to
205 * escape the compartment.
208 nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp)
210 struct nameicap_tracker *nt;
213 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp == NULL ||
217 if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL &&
218 (mp->mnt_flag & MNT_LOCAL) == 0)
219 return (ENOTCAPABLE);
220 TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head,
225 return (ENOTCAPABLE);
229 namei_cleanup_cnp(struct componentname *cnp)
232 uma_zfree(namei_zone, cnp->cn_pnbuf);
234 cnp->cn_pnbuf = NULL;
235 cnp->cn_nameptr = NULL;
240 namei_handle_root(struct nameidata *ndp, struct vnode **dpp)
242 struct componentname *cnp;
245 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0) {
247 if (KTRPOINT(curthread, KTR_CAPFAIL))
248 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
250 return (ENOTCAPABLE);
252 while (*(cnp->cn_nameptr) == '/') {
256 *dpp = ndp->ni_rootdir;
262 * Convert a pathname into a pointer to a locked vnode.
264 * The FOLLOW flag is set when symbolic links are to be followed
265 * when they occur at the end of the name translation process.
266 * Symbolic links are always followed for all other pathname
267 * components other than the last.
269 * The segflg defines whether the name is to be copied from user
270 * space or kernel space.
272 * Overall outline of namei:
275 * get starting directory
276 * while (!done && !error) {
277 * call lookup to search path.
278 * if symbolic link, massage name in buffer and continue
282 namei(struct nameidata *ndp)
284 struct filedesc *fdp; /* pointer to file descriptor state */
285 char *cp; /* pointer into pathname argument */
286 struct vnode *dp; /* the directory we are searching */
287 struct iovec aiov; /* uio for reading symbolic links */
288 struct componentname *cnp;
294 int error, linklen, startdir_used;
299 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
300 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
301 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
302 ("namei: nameiop contaminated with flags"));
303 KASSERT((cnp->cn_flags & OPMASK) == 0,
304 ("namei: flags contaminated with nameiops"));
305 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
306 ndp->ni_startdir->v_type == VBAD);
308 TAILQ_INIT(&ndp->ni_cap_tracker);
311 /* We will set this ourselves if we need it. */
312 cnp->cn_flags &= ~TRAILINGSLASH;
315 * Get a buffer for the name to be translated, and copy the
316 * name into the buffer.
318 if ((cnp->cn_flags & HASBUF) == 0)
319 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
320 if (ndp->ni_segflg == UIO_SYSSPACE)
321 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
324 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
328 * Don't allow empty pathnames.
330 if (error == 0 && *cnp->cn_pnbuf == '\0')
333 #ifdef CAPABILITY_MODE
335 * In capability mode, lookups must be restricted to happen in
336 * the subtree with the root specified by the file descriptor:
337 * - The root must be real file descriptor, not the pseudo-descriptor
339 * - The passed path must be relative and not absolute.
340 * - If lookup_cap_dotdot is disabled, path must not contain the
342 * - If lookup_cap_dotdot is enabled, we verify that all '..'
343 * components lookups result in the directories which were
344 * previously walked by us, which prevents an escape from
347 if (error == 0 && IN_CAPABILITY_MODE(td) &&
348 (cnp->cn_flags & NOCAPCHECK) == 0) {
349 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
350 if (ndp->ni_dirfd == AT_FDCWD) {
352 if (KTRPOINT(td, KTR_CAPFAIL))
353 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
360 namei_cleanup_cnp(cnp);
366 if (KTRPOINT(td, KTR_NAMEI)) {
367 KASSERT(cnp->cn_thread == curthread,
368 ("namei not using curthread"));
369 ktrnamei(cnp->cn_pnbuf);
373 * Get starting point for the translation.
376 ndp->ni_rootdir = fdp->fd_rdir;
377 vrefact(ndp->ni_rootdir);
378 ndp->ni_topdir = fdp->fd_jdir;
381 * If we are auditing the kernel pathname, save the user pathname.
383 if (cnp->cn_flags & AUDITVNODE1)
384 AUDIT_ARG_UPATH1(td, ndp->ni_dirfd, cnp->cn_pnbuf);
385 if (cnp->cn_flags & AUDITVNODE2)
386 AUDIT_ARG_UPATH2(td, ndp->ni_dirfd, cnp->cn_pnbuf);
390 cnp->cn_nameptr = cnp->cn_pnbuf;
391 if (cnp->cn_pnbuf[0] == '/') {
392 ndp->ni_resflags |= NIRES_ABS;
393 error = namei_handle_root(ndp, &dp);
395 if (ndp->ni_startdir != NULL) {
396 dp = ndp->ni_startdir;
398 } else if (ndp->ni_dirfd == AT_FDCWD) {
402 rights = ndp->ni_rightsneeded;
403 cap_rights_set(&rights, CAP_LOOKUP);
405 if (cnp->cn_flags & AUDITVNODE1)
406 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
407 if (cnp->cn_flags & AUDITVNODE2)
408 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
410 * Effectively inlined fgetvp_rights, because we need to
411 * inspect the file as well as grabbing the vnode.
413 error = fget_cap_locked(fdp, ndp->ni_dirfd, &rights,
414 &dfp, &ndp->ni_filecaps);
417 * Preserve the error; it should either be EBADF
418 * or capability-related, both of which can be
419 * safely returned to the caller.
421 } else if (dfp->f_ops == &badfileops) {
423 } else if (dfp->f_vnode == NULL) {
429 if ((dfp->f_flag & FSEARCH) != 0)
430 cnp->cn_flags |= NOEXECCHECK;
434 * If file descriptor doesn't have all rights,
435 * all lookups relative to it must also be
439 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights,
441 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
442 ndp->ni_filecaps.fc_nioctls != -1) {
443 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
447 if (error == 0 && dp->v_type != VDIR)
450 FILEDESC_SUNLOCK(fdp);
451 if (ndp->ni_startdir != NULL && !startdir_used)
452 vrele(ndp->ni_startdir);
458 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 &&
459 lookup_cap_dotdot != 0)
460 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
461 SDT_PROBE3(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
464 ndp->ni_startdir = dp;
469 * If not a symbolic link, we're done.
471 if ((cnp->cn_flags & ISSYMLINK) == 0) {
472 vrele(ndp->ni_rootdir);
473 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
474 namei_cleanup_cnp(cnp);
476 cnp->cn_flags |= HASBUF;
477 nameicap_cleanup(ndp);
478 SDT_PROBE2(vfs, namei, lookup, return, 0, ndp->ni_vp);
481 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
486 if ((cnp->cn_flags & NOMACCHECK) == 0) {
487 error = mac_vnode_check_readlink(td->td_ucred,
493 if (ndp->ni_pathlen > 1)
494 cp = uma_zalloc(namei_zone, M_WAITOK);
498 aiov.iov_len = MAXPATHLEN;
499 auio.uio_iov = &aiov;
502 auio.uio_rw = UIO_READ;
503 auio.uio_segflg = UIO_SYSSPACE;
505 auio.uio_resid = MAXPATHLEN;
506 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
508 if (ndp->ni_pathlen > 1)
509 uma_zfree(namei_zone, cp);
512 linklen = MAXPATHLEN - auio.uio_resid;
514 if (ndp->ni_pathlen > 1)
515 uma_zfree(namei_zone, cp);
519 if (linklen + ndp->ni_pathlen > MAXPATHLEN) {
520 if (ndp->ni_pathlen > 1)
521 uma_zfree(namei_zone, cp);
522 error = ENAMETOOLONG;
525 if (ndp->ni_pathlen > 1) {
526 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
527 uma_zfree(namei_zone, cnp->cn_pnbuf);
530 cnp->cn_pnbuf[linklen] = '\0';
531 ndp->ni_pathlen += linklen;
535 * Check if root directory should replace current directory.
537 cnp->cn_nameptr = cnp->cn_pnbuf;
538 if (*(cnp->cn_nameptr) == '/') {
540 error = namei_handle_root(ndp, &dp);
549 vrele(ndp->ni_rootdir);
550 namei_cleanup_cnp(cnp);
551 nameicap_cleanup(ndp);
552 SDT_PROBE2(vfs, namei, lookup, return, error, NULL);
557 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
560 if (mp == NULL || ((lkflags & LK_SHARED) &&
561 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
562 ((cnflags & ISDOTDOT) &&
563 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
564 lkflags &= ~LK_SHARED;
565 lkflags |= LK_EXCLUSIVE;
567 lkflags |= LK_NODDLKTREAT;
572 needs_exclusive_leaf(struct mount *mp, int flags)
576 * Intermediate nodes can use shared locks, we only need to
577 * force an exclusive lock for leaf nodes.
579 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
582 /* Always use exclusive locks if LOCKSHARED isn't set. */
583 if (!(flags & LOCKSHARED))
587 * For lookups during open(), if the mount point supports
588 * extended shared operations, then use a shared lock for the
589 * leaf node, otherwise use an exclusive lock.
591 if ((flags & ISOPEN) != 0)
592 return (!MNT_EXTENDED_SHARED(mp));
595 * Lookup requests outside of open() that specify LOCKSHARED
596 * only need a shared lock on the leaf vnode.
603 * This is a very central and rather complicated routine.
605 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
606 * The starting directory is taken from ni_startdir. The pathname is
607 * descended until done, or a symbolic link is encountered. The variable
608 * ni_more is clear if the path is completed; it is set to one if a
609 * symbolic link needing interpretation is encountered.
611 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
612 * whether the name is to be looked up, created, renamed, or deleted.
613 * When CREATE, RENAME, or DELETE is specified, information usable in
614 * creating, renaming, or deleting a directory entry may be calculated.
615 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
616 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
617 * returned unlocked. Otherwise the parent directory is not returned. If
618 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
619 * the target is returned locked, otherwise it is returned unlocked.
620 * When creating or renaming and LOCKPARENT is specified, the target may not
621 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
623 * Overall outline of lookup:
626 * identify next component of name at ndp->ni_ptr
627 * handle degenerate case where name is null string
628 * if .. and crossing mount points and on mounted filesys, find parent
629 * call VOP_LOOKUP routine for next component name
630 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
631 * component vnode returned in ni_vp (if it exists), locked.
632 * if result vnode is mounted on and crossing mount points,
633 * find mounted on vnode
634 * if more components of name, do next level at dirloop
635 * return the answer in ni_vp, locked if LOCKLEAF set
636 * if LOCKPARENT set, return locked parent in ni_dvp
637 * if WANTPARENT set, return unlocked parent in ni_dvp
640 lookup(struct nameidata *ndp)
642 char *cp; /* pointer into pathname argument */
643 char *prev_ni_next; /* saved ndp->ni_next */
644 struct vnode *dp = NULL; /* the directory we are searching */
645 struct vnode *tdp; /* saved dp */
646 struct mount *mp; /* mount table entry */
648 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
649 int docache; /* == 0 do not cache last component */
650 int wantparent; /* 1 => wantparent or lockparent flag */
651 int rdonly; /* lookup read-only flag bit */
653 int dpunlocked = 0; /* dp has already been unlocked */
654 int relookup = 0; /* do not consume the path component */
655 struct componentname *cnp = &ndp->ni_cnd;
660 * Setup: break out flag bits into variables.
663 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
664 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
665 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
666 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
667 if (cnp->cn_nameiop == DELETE ||
668 (wantparent && cnp->cn_nameiop != CREATE &&
669 cnp->cn_nameiop != LOOKUP))
671 rdonly = cnp->cn_flags & RDONLY;
672 cnp->cn_flags &= ~ISSYMLINK;
675 * We use shared locks until we hit the parent of the last cn then
676 * we adjust based on the requesting flags.
678 cnp->cn_lkflags = LK_SHARED;
679 dp = ndp->ni_startdir;
680 ndp->ni_startdir = NULLVP;
682 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
687 * Search a new directory.
689 * The last component of the filename is left accessible via
690 * cnp->cn_nameptr for callers that need the name. Callers needing
691 * the name set the SAVENAME flag. When done, they assume
692 * responsibility for freeing the pathname buffer.
694 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
696 cnp->cn_namelen = cp - cnp->cn_nameptr;
697 if (cnp->cn_namelen > NAME_MAX) {
698 error = ENAMETOOLONG;
701 #ifdef NAMEI_DIAGNOSTIC
704 printf("{%s}: ", cnp->cn_nameptr);
707 prev_ni_pathlen = ndp->ni_pathlen;
708 ndp->ni_pathlen -= cnp->cn_namelen;
709 KASSERT(ndp->ni_pathlen <= PATH_MAX,
710 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
711 prev_ni_next = ndp->ni_next;
715 * Replace multiple slashes by a single slash and trailing slashes
716 * by a null. This must be done before VOP_LOOKUP() because some
717 * fs's don't know about trailing slashes. Remember if there were
718 * trailing slashes to handle symlinks, existing non-directories
719 * and non-existing files that won't be directories specially later.
721 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
725 *ndp->ni_next = '\0';
726 cnp->cn_flags |= TRAILINGSLASH;
731 cnp->cn_flags |= MAKEENTRY;
732 if (*cp == '\0' && docache == 0)
733 cnp->cn_flags &= ~MAKEENTRY;
734 if (cnp->cn_namelen == 2 &&
735 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
736 cnp->cn_flags |= ISDOTDOT;
738 cnp->cn_flags &= ~ISDOTDOT;
739 if (*ndp->ni_next == 0)
740 cnp->cn_flags |= ISLASTCN;
742 cnp->cn_flags &= ~ISLASTCN;
744 if ((cnp->cn_flags & ISLASTCN) != 0 &&
745 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
746 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
751 nameicap_tracker_add(ndp, dp);
754 * Check for degenerate name (e.g. / or "")
755 * which is a way of talking about a directory,
756 * e.g. like "/." or ".".
758 if (cnp->cn_nameptr[0] == '\0') {
759 if (dp->v_type != VDIR) {
763 if (cnp->cn_nameiop != LOOKUP) {
773 if (cnp->cn_flags & AUDITVNODE1)
774 AUDIT_ARG_VNODE1(dp);
775 else if (cnp->cn_flags & AUDITVNODE2)
776 AUDIT_ARG_VNODE2(dp);
778 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
780 /* XXX This should probably move to the top of function. */
781 if (cnp->cn_flags & SAVESTART)
782 panic("lookup: SAVESTART");
787 * Handle "..": five special cases.
788 * 0. If doing a capability lookup and lookup_cap_dotdot is
789 * disabled, return ENOTCAPABLE.
790 * 1. Return an error if this is the last component of
791 * the name and the operation is DELETE or RENAME.
792 * 2. If at root directory (e.g. after chroot)
793 * or at absolute root directory
794 * then ignore it so can't get out.
795 * 3. If this vnode is the root of a mounted
796 * filesystem, then replace it with the
797 * vnode which was mounted on so we take the
798 * .. in the other filesystem.
799 * 4. If the vnode is the top directory of
800 * the jail or chroot, don't let them out.
801 * 5. If doing a capability lookup and lookup_cap_dotdot is
802 * enabled, return ENOTCAPABLE if the lookup would escape
803 * from the initial file descriptor directory. Checks are
804 * done by ensuring that namei() already traversed the
805 * result of dotdot lookup.
807 if (cnp->cn_flags & ISDOTDOT) {
808 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT))
809 == NI_LCF_STRICTRELATIVE) {
811 if (KTRPOINT(curthread, KTR_CAPFAIL))
812 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
817 if ((cnp->cn_flags & ISLASTCN) != 0 &&
818 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
823 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
825 if (dp == pr->pr_root)
827 if (dp == ndp->ni_rootdir ||
828 dp == ndp->ni_topdir ||
831 ((dp->v_vflag & VV_ROOT) != 0 &&
832 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
838 if ((dp->v_vflag & VV_ROOT) == 0)
840 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
845 dp = dp->v_mount->mnt_vnodecovered;
849 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
850 LK_RETRY, ISDOTDOT));
851 error = nameicap_check_dotdot(ndp, dp);
854 if (KTRPOINT(curthread, KTR_CAPFAIL))
855 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
863 * We now have a segment name to search for, and a directory to search.
867 if ((cnp->cn_flags & NOMACCHECK) == 0) {
868 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
876 ASSERT_VOP_LOCKED(dp, "lookup");
878 * If we have a shared lock we may need to upgrade the lock for the
881 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
882 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
883 vn_lock(dp, LK_UPGRADE|LK_RETRY);
884 if ((dp->v_iflag & VI_DOOMED) != 0) {
889 * If we're looking up the last component and we need an exclusive
890 * lock, adjust our lkflags.
892 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
893 cnp->cn_lkflags = LK_EXCLUSIVE;
894 #ifdef NAMEI_DIAGNOSTIC
895 vn_printf(dp, "lookup in ");
897 lkflags_save = cnp->cn_lkflags;
898 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
900 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
901 cnp->cn_lkflags = lkflags_save;
903 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
904 #ifdef NAMEI_DIAGNOSTIC
905 printf("not found\n");
907 if ((error == ENOENT) &&
908 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
909 (dp->v_mount->mnt_flag & MNT_UNION)) {
911 dp = dp->v_mount->mnt_vnodecovered;
915 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
916 LK_RETRY, cnp->cn_flags));
917 nameicap_tracker_add(ndp, dp);
921 if (error == ERELOOKUP) {
929 if (error != EJUSTRETURN)
932 * At this point, we know we're at the end of the
933 * pathname. If creating / renaming, we can consider
934 * allowing the file or directory to be created / renamed,
935 * provided we're not on a read-only filesystem.
941 /* trailing slash only allowed for directories */
942 if ((cnp->cn_flags & TRAILINGSLASH) &&
943 !(cnp->cn_flags & WILLBEDIR)) {
947 if ((cnp->cn_flags & LOCKPARENT) == 0)
950 * We return with ni_vp NULL to indicate that the entry
951 * doesn't currently exist, leaving a pointer to the
952 * (possibly locked) directory vnode in ndp->ni_dvp.
954 if (cnp->cn_flags & SAVESTART) {
955 ndp->ni_startdir = ndp->ni_dvp;
956 VREF(ndp->ni_startdir);
962 #ifdef NAMEI_DIAGNOSTIC
968 * Check to see if the vnode has been mounted on;
969 * if so find the root of the mounted filesystem.
971 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
972 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
976 if (dp != ndp->ni_dvp)
981 ndp->ni_dvp = vp_crossmp;
982 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
983 cnp->cn_flags), &tdp);
985 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
986 panic("vp_crossmp exclusively locked or reclaimed");
991 ndp->ni_vp = dp = tdp;
995 * Check for symbolic link
997 if ((dp->v_type == VLNK) &&
998 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
999 *ndp->ni_next == '/')) {
1000 cnp->cn_flags |= ISSYMLINK;
1001 if (dp->v_iflag & VI_DOOMED) {
1003 * We can't know whether the directory was mounted with
1004 * NOSYMFOLLOW, so we can't follow safely.
1009 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1014 * Symlink code always expects an unlocked dvp.
1016 if (ndp->ni_dvp != ndp->ni_vp) {
1017 VOP_UNLOCK(ndp->ni_dvp, 0);
1018 ni_dvp_unlocked = 1;
1025 * Not a symbolic link that we will follow. Continue with the
1026 * next component if there is any; otherwise, we're done.
1028 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1029 ("lookup: invalid path state."));
1032 ndp->ni_pathlen = prev_ni_pathlen;
1033 ndp->ni_next = prev_ni_next;
1034 if (ndp->ni_dvp != dp)
1040 if (cnp->cn_flags & ISDOTDOT) {
1041 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1044 if (KTRPOINT(curthread, KTR_CAPFAIL))
1045 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1050 if (*ndp->ni_next == '/') {
1051 cnp->cn_nameptr = ndp->ni_next;
1052 while (*cnp->cn_nameptr == '/') {
1056 if (ndp->ni_dvp != dp)
1063 * If we're processing a path with a trailing slash,
1064 * check that the end result is a directory.
1066 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1071 * Disallow directory write attempts on read-only filesystems.
1074 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1078 if (cnp->cn_flags & SAVESTART) {
1079 ndp->ni_startdir = ndp->ni_dvp;
1080 VREF(ndp->ni_startdir);
1083 ni_dvp_unlocked = 2;
1084 if (ndp->ni_dvp != dp)
1088 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1089 VOP_UNLOCK(ndp->ni_dvp, 0);
1090 ni_dvp_unlocked = 1;
1093 if (cnp->cn_flags & AUDITVNODE1)
1094 AUDIT_ARG_VNODE1(dp);
1095 else if (cnp->cn_flags & AUDITVNODE2)
1096 AUDIT_ARG_VNODE2(dp);
1098 if ((cnp->cn_flags & LOCKLEAF) == 0)
1102 * Because of shared lookup we may have the vnode shared locked, but
1103 * the caller may want it to be exclusively locked.
1105 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1106 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1107 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1108 if (dp->v_iflag & VI_DOOMED) {
1116 if (ni_dvp_unlocked != 2) {
1117 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1130 * relookup - lookup a path name component
1131 * Used by lookup to re-acquire things.
1134 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
1136 struct vnode *dp = NULL; /* the directory we are searching */
1137 int wantparent; /* 1 => wantparent or lockparent flag */
1138 int rdonly; /* lookup read-only flag bit */
1141 KASSERT(cnp->cn_flags & ISLASTCN,
1142 ("relookup: Not given last component."));
1144 * Setup: break out flag bits into variables.
1146 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
1147 KASSERT(wantparent, ("relookup: parent not wanted."));
1148 rdonly = cnp->cn_flags & RDONLY;
1149 cnp->cn_flags &= ~ISSYMLINK;
1151 cnp->cn_lkflags = LK_EXCLUSIVE;
1152 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1155 * Search a new directory.
1157 * The last component of the filename is left accessible via
1158 * cnp->cn_nameptr for callers that need the name. Callers needing
1159 * the name set the SAVENAME flag. When done, they assume
1160 * responsibility for freeing the pathname buffer.
1162 #ifdef NAMEI_DIAGNOSTIC
1163 printf("{%s}: ", cnp->cn_nameptr);
1167 * Check for "" which represents the root directory after slash
1170 if (cnp->cn_nameptr[0] == '\0') {
1172 * Support only LOOKUP for "/" because lookup()
1173 * can't succeed for CREATE, DELETE and RENAME.
1175 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1176 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1178 if (!(cnp->cn_flags & LOCKLEAF))
1181 /* XXX This should probably move to the top of function. */
1182 if (cnp->cn_flags & SAVESTART)
1183 panic("lookup: SAVESTART");
1187 if (cnp->cn_flags & ISDOTDOT)
1188 panic ("relookup: lookup on dot-dot");
1191 * We now have a segment name to search for, and a directory to search.
1193 #ifdef NAMEI_DIAGNOSTIC
1194 vn_printf(dp, "search in ");
1196 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1197 KASSERT(*vpp == NULL, ("leaf should be empty"));
1198 if (error != EJUSTRETURN)
1201 * If creating and at end of pathname, then can consider
1202 * allowing file to be created.
1208 /* ASSERT(dvp == ndp->ni_startdir) */
1209 if (cnp->cn_flags & SAVESTART)
1211 if ((cnp->cn_flags & LOCKPARENT) == 0)
1214 * We return with ni_vp NULL to indicate that the entry
1215 * doesn't currently exist, leaving a pointer to the
1216 * (possibly locked) directory vnode in ndp->ni_dvp.
1224 * Disallow directory write attempts on read-only filesystems.
1227 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1236 * Set the parent lock/ref state to the requested state.
1238 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1243 } else if (!wantparent)
1246 * Check for symbolic link
1248 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1249 ("relookup: symlink found.\n"));
1251 /* ASSERT(dvp == ndp->ni_startdir) */
1252 if (cnp->cn_flags & SAVESTART)
1255 if ((cnp->cn_flags & LOCKLEAF) == 0)
1265 NDINIT_ALL(struct nameidata *ndp, u_long op, u_long flags, enum uio_seg segflg,
1266 const char *namep, int dirfd, struct vnode *startdir, cap_rights_t *rightsp,
1270 ndp->ni_cnd.cn_nameiop = op;
1271 ndp->ni_cnd.cn_flags = flags;
1272 ndp->ni_segflg = segflg;
1273 ndp->ni_dirp = namep;
1274 ndp->ni_dirfd = dirfd;
1275 ndp->ni_startdir = startdir;
1276 ndp->ni_resflags = 0;
1277 if (rightsp != NULL)
1278 ndp->ni_rightsneeded = *rightsp;
1280 cap_rights_init(&ndp->ni_rightsneeded);
1281 filecaps_init(&ndp->ni_filecaps);
1282 ndp->ni_cnd.cn_thread = td;
1286 * Free data allocated by namei(); see namei(9) for details.
1289 NDFREE(struct nameidata *ndp, const u_int flags)
1297 if (!(flags & NDF_NO_FREE_PNBUF) &&
1298 (ndp->ni_cnd.cn_flags & HASBUF)) {
1299 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1300 ndp->ni_cnd.cn_flags &= ~HASBUF;
1302 if (!(flags & NDF_NO_VP_UNLOCK) &&
1303 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1305 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1306 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1307 ndp->ni_dvp != ndp->ni_vp)
1309 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1318 VOP_UNLOCK(ndp->ni_vp, 0);
1319 if (!(flags & NDF_NO_DVP_RELE) &&
1320 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1329 VOP_UNLOCK(ndp->ni_dvp, 0);
1330 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1331 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1332 vrele(ndp->ni_startdir);
1333 ndp->ni_startdir = NULL;
1338 * Determine if there is a suitable alternate filename under the specified
1339 * prefix for the specified path. If the create flag is set, then the
1340 * alternate prefix will be used so long as the parent directory exists.
1341 * This is used by the various compatibility ABIs so that Linux binaries prefer
1342 * files under /compat/linux for example. The chosen path (whether under
1343 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1344 * to by pathbuf. The caller is responsible for free'ing the buffer from
1345 * the M_TEMP bucket if one is returned.
1348 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1349 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1351 struct nameidata nd, ndroot;
1352 char *ptr, *buf, *cp;
1356 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1359 /* Copy the prefix into the new pathname as a starting point. */
1360 len = strlcpy(buf, prefix, MAXPATHLEN);
1361 if (len >= MAXPATHLEN) {
1366 sz = MAXPATHLEN - len;
1369 /* Append the filename to the prefix. */
1370 if (pathseg == UIO_SYSSPACE)
1371 error = copystr(path, ptr, sz, &len);
1373 error = copyinstr(path, ptr, sz, &len);
1381 /* Only use a prefix with absolute pathnames. */
1387 if (dirfd != AT_FDCWD) {
1389 * We want the original because the "prefix" is
1390 * included in the already opened dirfd.
1392 bcopy(ptr, buf, len);
1397 * We know that there is a / somewhere in this pathname.
1398 * Search backwards for it, to find the file's parent dir
1399 * to see if it exists in the alternate tree. If it does,
1400 * and we want to create a file (cflag is set). We don't
1401 * need to worry about the root comparison in this case.
1405 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1408 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1414 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1421 * We now compare the vnode of the prefix to the one
1422 * vnode asked. If they resolve to be the same, then we
1423 * ignore the match so that the real root gets used.
1424 * This avoids the problem of traversing "../.." to find the
1425 * root directory and never finding it, because "/" resolves
1426 * to the emulation root directory. This is expensive :-(
1428 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1431 /* We shouldn't ever get an error from this namei(). */
1432 error = namei(&ndroot);
1434 if (nd.ni_vp == ndroot.ni_vp)
1437 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1438 vrele(ndroot.ni_vp);
1442 NDFREE(&nd, NDF_ONLY_PNBUF);
1446 /* If there was an error, use the original path name. */
1448 bcopy(ptr, buf, len);