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;
126 WITNESS_UNLOCK(&lk->lock_object, 0, LOCK_FILE, LOCK_LINE);
127 LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, LOCK_FILE,
132 static struct vop_vector crossmp_vnodeops = {
133 .vop_default = &default_vnodeops,
134 .vop_islocked = crossmp_vop_islocked,
135 .vop_lock1 = crossmp_vop_lock1,
136 .vop_unlock = crossmp_vop_unlock,
139 * VFS_VOP_VECTOR_REGISTER(crossmp_vnodeops) is not used here since the vnode
140 * gets allocated early. See nameiinit for the direct call below.
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 vfs_vector_op_register(&crossmp_vnodeops);
160 getnewvnode("crossmp", NULL, &crossmp_vnodeops, &vp_crossmp);
162 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
164 static int lookup_cap_dotdot = 1;
165 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot, CTLFLAG_RWTUN,
166 &lookup_cap_dotdot, 0,
167 "enables \"..\" components in path lookup in capability mode");
168 static int lookup_cap_dotdot_nonlocal = 1;
169 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot_nonlocal, CTLFLAG_RWTUN,
170 &lookup_cap_dotdot_nonlocal, 0,
171 "enables \"..\" components in path lookup in capability mode "
172 "on non-local mount");
175 nameicap_tracker_add(struct nameidata *ndp, struct vnode *dp)
177 struct nameicap_tracker *nt;
179 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp->v_type != VDIR)
181 if ((ndp->ni_lcf & (NI_LCF_BENEATH_ABS | NI_LCF_BENEATH_LATCHED)) ==
182 NI_LCF_BENEATH_ABS) {
183 MPASS((ndp->ni_lcf & NI_LCF_LATCH) != 0);
184 if (dp != ndp->ni_beneath_latch)
186 ndp->ni_lcf |= NI_LCF_BENEATH_LATCHED;
188 nt = uma_zalloc(nt_zone, M_WAITOK);
191 TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link);
195 nameicap_cleanup(struct nameidata *ndp, bool clean_latch)
197 struct nameicap_tracker *nt, *nt1;
199 KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) ||
200 (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative"));
201 TAILQ_FOREACH_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) {
202 TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link);
204 uma_zfree(nt_zone, nt);
206 if (clean_latch && (ndp->ni_lcf & NI_LCF_LATCH) != 0) {
207 ndp->ni_lcf &= ~NI_LCF_LATCH;
208 vrele(ndp->ni_beneath_latch);
213 * For dotdot lookups in capability mode, only allow the component
214 * lookup to succeed if the resulting directory was already traversed
215 * during the operation. Also fail dotdot lookups for non-local
216 * filesystems, where external agents might assist local lookups to
217 * escape the compartment.
220 nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp)
222 struct nameicap_tracker *nt;
225 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp == NULL ||
229 if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL &&
230 (mp->mnt_flag & MNT_LOCAL) == 0)
231 return (ENOTCAPABLE);
232 TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head,
237 if ((ndp->ni_lcf & NI_LCF_BENEATH_ABS) != 0) {
238 ndp->ni_lcf &= ~NI_LCF_BENEATH_LATCHED;
239 nameicap_cleanup(ndp, false);
242 return (ENOTCAPABLE);
246 namei_cleanup_cnp(struct componentname *cnp)
249 uma_zfree(namei_zone, cnp->cn_pnbuf);
251 cnp->cn_pnbuf = NULL;
252 cnp->cn_nameptr = NULL;
257 namei_handle_root(struct nameidata *ndp, struct vnode **dpp, u_int n)
259 struct componentname *cnp;
262 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0) {
264 if (KTRPOINT(curthread, KTR_CAPFAIL))
265 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
267 return (ENOTCAPABLE);
269 if ((cnp->cn_flags & BENEATH) != 0) {
270 ndp->ni_lcf |= NI_LCF_BENEATH_ABS;
271 ndp->ni_lcf &= ~NI_LCF_BENEATH_LATCHED;
272 nameicap_cleanup(ndp, false);
274 while (*(cnp->cn_nameptr) == '/') {
278 *dpp = ndp->ni_rootdir;
284 * Convert a pathname into a pointer to a locked vnode.
286 * The FOLLOW flag is set when symbolic links are to be followed
287 * when they occur at the end of the name translation process.
288 * Symbolic links are always followed for all other pathname
289 * components other than the last.
291 * The segflg defines whether the name is to be copied from user
292 * space or kernel space.
294 * Overall outline of namei:
297 * get starting directory
298 * while (!done && !error) {
299 * call lookup to search path.
300 * if symbolic link, massage name in buffer and continue
304 namei(struct nameidata *ndp)
306 struct filedesc *fdp; /* pointer to file descriptor state */
307 char *cp; /* pointer into pathname argument */
308 struct vnode *dp; /* the directory we are searching */
309 struct iovec aiov; /* uio for reading symbolic links */
310 struct componentname *cnp;
315 struct filecaps dirfd_caps;
317 int error, linklen, startdir_used;
322 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
323 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
324 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
325 ("namei: nameiop contaminated with flags"));
326 KASSERT((cnp->cn_flags & OPMASK) == 0,
327 ("namei: flags contaminated with nameiops"));
328 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
329 ndp->ni_startdir->v_type == VBAD);
331 TAILQ_INIT(&ndp->ni_cap_tracker);
334 /* We will set this ourselves if we need it. */
335 cnp->cn_flags &= ~TRAILINGSLASH;
338 * Get a buffer for the name to be translated, and copy the
339 * name into the buffer.
341 if ((cnp->cn_flags & HASBUF) == 0)
342 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
343 if (ndp->ni_segflg == UIO_SYSSPACE)
344 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
347 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
351 * Don't allow empty pathnames.
353 if (error == 0 && *cnp->cn_pnbuf == '\0')
356 #ifdef CAPABILITY_MODE
358 * In capability mode, lookups must be restricted to happen in
359 * the subtree with the root specified by the file descriptor:
360 * - The root must be real file descriptor, not the pseudo-descriptor
362 * - The passed path must be relative and not absolute.
363 * - If lookup_cap_dotdot is disabled, path must not contain the
365 * - If lookup_cap_dotdot is enabled, we verify that all '..'
366 * components lookups result in the directories which were
367 * previously walked by us, which prevents an escape from
370 if (error == 0 && IN_CAPABILITY_MODE(td) &&
371 (cnp->cn_flags & NOCAPCHECK) == 0) {
372 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
373 if (ndp->ni_dirfd == AT_FDCWD) {
375 if (KTRPOINT(td, KTR_CAPFAIL))
376 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
383 namei_cleanup_cnp(cnp);
389 if (KTRPOINT(td, KTR_NAMEI)) {
390 KASSERT(cnp->cn_thread == curthread,
391 ("namei not using curthread"));
392 ktrnamei(cnp->cn_pnbuf);
396 * Get starting point for the translation.
400 * The reference on ni_rootdir is acquired in the block below to avoid
401 * back-to-back atomics for absolute lookups.
403 ndp->ni_rootdir = fdp->fd_rdir;
404 ndp->ni_topdir = fdp->fd_jdir;
408 cnp->cn_nameptr = cnp->cn_pnbuf;
409 if (cnp->cn_pnbuf[0] == '/') {
410 ndp->ni_resflags |= NIRES_ABS;
411 error = namei_handle_root(ndp, &dp, 2);
414 * Simplify error handling, we should almost never be
417 vrefact(ndp->ni_rootdir);
420 if (ndp->ni_startdir != NULL) {
421 vrefact(ndp->ni_rootdir);
422 dp = ndp->ni_startdir;
424 } else if (ndp->ni_dirfd == AT_FDCWD) {
426 if (dp == ndp->ni_rootdir) {
429 vrefact(ndp->ni_rootdir);
433 vrefact(ndp->ni_rootdir);
434 rights = ndp->ni_rightsneeded;
435 cap_rights_set_one(&rights, CAP_LOOKUP);
437 if (cnp->cn_flags & AUDITVNODE1)
438 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
439 if (cnp->cn_flags & AUDITVNODE2)
440 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
442 * Effectively inlined fgetvp_rights, because we need to
443 * inspect the file as well as grabbing the vnode.
445 error = fget_cap_locked(fdp, ndp->ni_dirfd, &rights,
446 &dfp, &ndp->ni_filecaps);
449 * Preserve the error; it should either be EBADF
450 * or capability-related, both of which can be
451 * safely returned to the caller.
453 } else if (dfp->f_ops == &badfileops) {
455 } else if (dfp->f_vnode == NULL) {
461 if ((dfp->f_flag & FSEARCH) != 0)
462 cnp->cn_flags |= NOEXECCHECK;
466 * If file descriptor doesn't have all rights,
467 * all lookups relative to it must also be
471 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights,
473 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
474 ndp->ni_filecaps.fc_nioctls != -1) {
475 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
479 if (error == 0 && dp->v_type != VDIR)
482 if (error == 0 && (cnp->cn_flags & BENEATH) != 0) {
483 if (ndp->ni_dirfd == AT_FDCWD) {
484 ndp->ni_beneath_latch = fdp->fd_cdir;
485 vrefact(ndp->ni_beneath_latch);
487 rights = ndp->ni_rightsneeded;
488 cap_rights_set_one(&rights, CAP_LOOKUP);
489 error = fgetvp_rights(td, ndp->ni_dirfd, &rights,
490 &dirfd_caps, &ndp->ni_beneath_latch);
491 if (error == 0 && dp->v_type != VDIR) {
492 vrele(ndp->ni_beneath_latch);
497 ndp->ni_lcf |= NI_LCF_LATCH;
499 FILEDESC_SUNLOCK(fdp);
501 * If we are auditing the kernel pathname, save the user pathname.
503 if (cnp->cn_flags & AUDITVNODE1)
504 AUDIT_ARG_UPATH1_VP(td, ndp->ni_rootdir, dp, cnp->cn_pnbuf);
505 if (cnp->cn_flags & AUDITVNODE2)
506 AUDIT_ARG_UPATH2_VP(td, ndp->ni_rootdir, dp, cnp->cn_pnbuf);
507 if (ndp->ni_startdir != NULL && !startdir_used)
508 vrele(ndp->ni_startdir);
514 MPASS((ndp->ni_lcf & (NI_LCF_BENEATH_ABS | NI_LCF_LATCH)) !=
516 if (((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 &&
517 lookup_cap_dotdot != 0) ||
518 ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) == 0 &&
519 (cnp->cn_flags & BENEATH) != 0))
520 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
521 SDT_PROBE3(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
524 ndp->ni_startdir = dp;
529 * If not a symbolic link, we're done.
531 if ((cnp->cn_flags & ISSYMLINK) == 0) {
532 vrele(ndp->ni_rootdir);
533 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
534 namei_cleanup_cnp(cnp);
536 cnp->cn_flags |= HASBUF;
537 if ((ndp->ni_lcf & (NI_LCF_BENEATH_ABS |
538 NI_LCF_BENEATH_LATCHED)) == NI_LCF_BENEATH_ABS) {
542 nameicap_cleanup(ndp, true);
543 SDT_PROBE2(vfs, namei, lookup, return, error,
544 (error == 0 ? ndp->ni_vp : NULL));
547 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
552 if ((cnp->cn_flags & NOMACCHECK) == 0) {
553 error = mac_vnode_check_readlink(td->td_ucred,
559 if (ndp->ni_pathlen > 1)
560 cp = uma_zalloc(namei_zone, M_WAITOK);
564 aiov.iov_len = MAXPATHLEN;
565 auio.uio_iov = &aiov;
568 auio.uio_rw = UIO_READ;
569 auio.uio_segflg = UIO_SYSSPACE;
571 auio.uio_resid = MAXPATHLEN;
572 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
574 if (ndp->ni_pathlen > 1)
575 uma_zfree(namei_zone, cp);
578 linklen = MAXPATHLEN - auio.uio_resid;
580 if (ndp->ni_pathlen > 1)
581 uma_zfree(namei_zone, cp);
585 if (linklen + ndp->ni_pathlen > MAXPATHLEN) {
586 if (ndp->ni_pathlen > 1)
587 uma_zfree(namei_zone, cp);
588 error = ENAMETOOLONG;
591 if (ndp->ni_pathlen > 1) {
592 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
593 uma_zfree(namei_zone, cnp->cn_pnbuf);
596 cnp->cn_pnbuf[linklen] = '\0';
597 ndp->ni_pathlen += linklen;
601 * Check if root directory should replace current directory.
603 cnp->cn_nameptr = cnp->cn_pnbuf;
604 if (*(cnp->cn_nameptr) == '/') {
606 error = namei_handle_root(ndp, &dp, 1);
615 vrele(ndp->ni_rootdir);
617 namei_cleanup_cnp(cnp);
618 nameicap_cleanup(ndp, true);
619 SDT_PROBE2(vfs, namei, lookup, return, error, NULL);
624 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
627 if (mp == NULL || ((lkflags & LK_SHARED) &&
628 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
629 ((cnflags & ISDOTDOT) &&
630 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
631 lkflags &= ~LK_SHARED;
632 lkflags |= LK_EXCLUSIVE;
634 lkflags |= LK_NODDLKTREAT;
639 needs_exclusive_leaf(struct mount *mp, int flags)
643 * Intermediate nodes can use shared locks, we only need to
644 * force an exclusive lock for leaf nodes.
646 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
649 /* Always use exclusive locks if LOCKSHARED isn't set. */
650 if (!(flags & LOCKSHARED))
654 * For lookups during open(), if the mount point supports
655 * extended shared operations, then use a shared lock for the
656 * leaf node, otherwise use an exclusive lock.
658 if ((flags & ISOPEN) != 0)
659 return (!MNT_EXTENDED_SHARED(mp));
662 * Lookup requests outside of open() that specify LOCKSHARED
663 * only need a shared lock on the leaf vnode.
670 * This is a very central and rather complicated routine.
672 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
673 * The starting directory is taken from ni_startdir. The pathname is
674 * descended until done, or a symbolic link is encountered. The variable
675 * ni_more is clear if the path is completed; it is set to one if a
676 * symbolic link needing interpretation is encountered.
678 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
679 * whether the name is to be looked up, created, renamed, or deleted.
680 * When CREATE, RENAME, or DELETE is specified, information usable in
681 * creating, renaming, or deleting a directory entry may be calculated.
682 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
683 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
684 * returned unlocked. Otherwise the parent directory is not returned. If
685 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
686 * the target is returned locked, otherwise it is returned unlocked.
687 * When creating or renaming and LOCKPARENT is specified, the target may not
688 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
690 * Overall outline of lookup:
693 * identify next component of name at ndp->ni_ptr
694 * handle degenerate case where name is null string
695 * if .. and crossing mount points and on mounted filesys, find parent
696 * call VOP_LOOKUP routine for next component name
697 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
698 * component vnode returned in ni_vp (if it exists), locked.
699 * if result vnode is mounted on and crossing mount points,
700 * find mounted on vnode
701 * if more components of name, do next level at dirloop
702 * return the answer in ni_vp, locked if LOCKLEAF set
703 * if LOCKPARENT set, return locked parent in ni_dvp
704 * if WANTPARENT set, return unlocked parent in ni_dvp
707 lookup(struct nameidata *ndp)
709 char *cp; /* pointer into pathname argument */
710 char *prev_ni_next; /* saved ndp->ni_next */
711 struct vnode *dp = NULL; /* the directory we are searching */
712 struct vnode *tdp; /* saved dp */
713 struct mount *mp; /* mount table entry */
715 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
716 int docache; /* == 0 do not cache last component */
717 int wantparent; /* 1 => wantparent or lockparent flag */
718 int rdonly; /* lookup read-only flag bit */
720 int dpunlocked = 0; /* dp has already been unlocked */
721 int relookup = 0; /* do not consume the path component */
722 struct componentname *cnp = &ndp->ni_cnd;
727 * Setup: break out flag bits into variables.
730 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
731 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
732 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
733 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
734 if (cnp->cn_nameiop == DELETE ||
735 (wantparent && cnp->cn_nameiop != CREATE &&
736 cnp->cn_nameiop != LOOKUP))
738 rdonly = cnp->cn_flags & RDONLY;
739 cnp->cn_flags &= ~ISSYMLINK;
742 * We use shared locks until we hit the parent of the last cn then
743 * we adjust based on the requesting flags.
745 cnp->cn_lkflags = LK_SHARED;
746 dp = ndp->ni_startdir;
747 ndp->ni_startdir = NULLVP;
749 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
754 * Search a new directory.
756 * The last component of the filename is left accessible via
757 * cnp->cn_nameptr for callers that need the name. Callers needing
758 * the name set the SAVENAME flag. When done, they assume
759 * responsibility for freeing the pathname buffer.
761 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
763 cnp->cn_namelen = cp - cnp->cn_nameptr;
764 if (cnp->cn_namelen > NAME_MAX) {
765 error = ENAMETOOLONG;
768 #ifdef NAMEI_DIAGNOSTIC
771 printf("{%s}: ", cnp->cn_nameptr);
774 prev_ni_pathlen = ndp->ni_pathlen;
775 ndp->ni_pathlen -= cnp->cn_namelen;
776 KASSERT(ndp->ni_pathlen <= PATH_MAX,
777 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
778 prev_ni_next = ndp->ni_next;
782 * Replace multiple slashes by a single slash and trailing slashes
783 * by a null. This must be done before VOP_LOOKUP() because some
784 * fs's don't know about trailing slashes. Remember if there were
785 * trailing slashes to handle symlinks, existing non-directories
786 * and non-existing files that won't be directories specially later.
788 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
792 *ndp->ni_next = '\0';
793 cnp->cn_flags |= TRAILINGSLASH;
798 cnp->cn_flags |= MAKEENTRY;
799 if (*cp == '\0' && docache == 0)
800 cnp->cn_flags &= ~MAKEENTRY;
801 if (cnp->cn_namelen == 2 &&
802 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
803 cnp->cn_flags |= ISDOTDOT;
805 cnp->cn_flags &= ~ISDOTDOT;
806 if (*ndp->ni_next == 0)
807 cnp->cn_flags |= ISLASTCN;
809 cnp->cn_flags &= ~ISLASTCN;
811 if ((cnp->cn_flags & ISLASTCN) != 0 &&
812 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
813 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
818 nameicap_tracker_add(ndp, dp);
821 * Check for degenerate name (e.g. / or "")
822 * which is a way of talking about a directory,
823 * e.g. like "/." or ".".
825 if (cnp->cn_nameptr[0] == '\0') {
826 if (dp->v_type != VDIR) {
830 if (cnp->cn_nameiop != LOOKUP) {
840 if (cnp->cn_flags & AUDITVNODE1)
841 AUDIT_ARG_VNODE1(dp);
842 else if (cnp->cn_flags & AUDITVNODE2)
843 AUDIT_ARG_VNODE2(dp);
845 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
847 /* XXX This should probably move to the top of function. */
848 if (cnp->cn_flags & SAVESTART)
849 panic("lookup: SAVESTART");
854 * Handle "..": five special cases.
855 * 0. If doing a capability lookup and lookup_cap_dotdot is
856 * disabled, return ENOTCAPABLE.
857 * 1. Return an error if this is the last component of
858 * the name and the operation is DELETE or RENAME.
859 * 2. If at root directory (e.g. after chroot)
860 * or at absolute root directory
861 * then ignore it so can't get out.
862 * 3. If this vnode is the root of a mounted
863 * filesystem, then replace it with the
864 * vnode which was mounted on so we take the
865 * .. in the other filesystem.
866 * 4. If the vnode is the top directory of
867 * the jail or chroot, don't let them out.
868 * 5. If doing a capability lookup and lookup_cap_dotdot is
869 * enabled, return ENOTCAPABLE if the lookup would escape
870 * from the initial file descriptor directory. Checks are
871 * done by ensuring that namei() already traversed the
872 * result of dotdot lookup.
874 if (cnp->cn_flags & ISDOTDOT) {
875 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT))
876 == NI_LCF_STRICTRELATIVE) {
878 if (KTRPOINT(curthread, KTR_CAPFAIL))
879 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
884 if ((cnp->cn_flags & ISLASTCN) != 0 &&
885 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
890 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
892 if (dp == pr->pr_root)
894 if (dp == ndp->ni_rootdir ||
895 dp == ndp->ni_topdir ||
898 ((dp->v_vflag & VV_ROOT) != 0 &&
899 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
905 if ((dp->v_vflag & VV_ROOT) == 0)
907 if (VN_IS_DOOMED(dp)) { /* forced unmount */
912 dp = dp->v_mount->mnt_vnodecovered;
916 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
917 LK_RETRY, ISDOTDOT));
918 error = nameicap_check_dotdot(ndp, dp);
921 if (KTRPOINT(curthread, KTR_CAPFAIL))
922 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
930 * We now have a segment name to search for, and a directory to search.
934 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp, cnp);
940 ASSERT_VOP_LOCKED(dp, "lookup");
942 * If we have a shared lock we may need to upgrade the lock for the
945 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
946 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
947 vn_lock(dp, LK_UPGRADE|LK_RETRY);
948 if (VN_IS_DOOMED(dp)) {
953 * If we're looking up the last component and we need an exclusive
954 * lock, adjust our lkflags.
956 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
957 cnp->cn_lkflags = LK_EXCLUSIVE;
958 #ifdef NAMEI_DIAGNOSTIC
959 vn_printf(dp, "lookup in ");
961 lkflags_save = cnp->cn_lkflags;
962 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
964 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
965 cnp->cn_lkflags = lkflags_save;
967 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
968 #ifdef NAMEI_DIAGNOSTIC
969 printf("not found\n");
971 if ((error == ENOENT) &&
972 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
973 (dp->v_mount->mnt_flag & MNT_UNION)) {
975 dp = dp->v_mount->mnt_vnodecovered;
979 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
980 LK_RETRY, cnp->cn_flags));
981 nameicap_tracker_add(ndp, dp);
985 if (error == ERELOOKUP) {
993 if (error != EJUSTRETURN)
996 * At this point, we know we're at the end of the
997 * pathname. If creating / renaming, we can consider
998 * allowing the file or directory to be created / renamed,
999 * provided we're not on a read-only filesystem.
1005 /* trailing slash only allowed for directories */
1006 if ((cnp->cn_flags & TRAILINGSLASH) &&
1007 !(cnp->cn_flags & WILLBEDIR)) {
1011 if ((cnp->cn_flags & LOCKPARENT) == 0)
1014 * We return with ni_vp NULL to indicate that the entry
1015 * doesn't currently exist, leaving a pointer to the
1016 * (possibly locked) directory vnode in ndp->ni_dvp.
1018 if (cnp->cn_flags & SAVESTART) {
1019 ndp->ni_startdir = ndp->ni_dvp;
1020 VREF(ndp->ni_startdir);
1026 #ifdef NAMEI_DIAGNOSTIC
1032 * Check to see if the vnode has been mounted on;
1033 * if so find the root of the mounted filesystem.
1035 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
1036 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
1037 if (vfs_busy(mp, 0))
1040 if (dp != ndp->ni_dvp)
1044 vrefact(vp_crossmp);
1045 ndp->ni_dvp = vp_crossmp;
1046 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
1047 cnp->cn_flags), &tdp);
1049 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
1050 panic("vp_crossmp exclusively locked or reclaimed");
1055 ndp->ni_vp = dp = tdp;
1059 * Check for symbolic link
1061 if ((dp->v_type == VLNK) &&
1062 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
1063 *ndp->ni_next == '/')) {
1064 cnp->cn_flags |= ISSYMLINK;
1065 if (VN_IS_DOOMED(dp)) {
1067 * We can't know whether the directory was mounted with
1068 * NOSYMFOLLOW, so we can't follow safely.
1073 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1078 * Symlink code always expects an unlocked dvp.
1080 if (ndp->ni_dvp != ndp->ni_vp) {
1081 VOP_UNLOCK(ndp->ni_dvp);
1082 ni_dvp_unlocked = 1;
1089 * Not a symbolic link that we will follow. Continue with the
1090 * next component if there is any; otherwise, we're done.
1092 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1093 ("lookup: invalid path state."));
1096 ndp->ni_pathlen = prev_ni_pathlen;
1097 ndp->ni_next = prev_ni_next;
1098 if (ndp->ni_dvp != dp)
1104 if (cnp->cn_flags & ISDOTDOT) {
1105 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1108 if (KTRPOINT(curthread, KTR_CAPFAIL))
1109 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1114 if (*ndp->ni_next == '/') {
1115 cnp->cn_nameptr = ndp->ni_next;
1116 while (*cnp->cn_nameptr == '/') {
1120 if (ndp->ni_dvp != dp)
1127 * If we're processing a path with a trailing slash,
1128 * check that the end result is a directory.
1130 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1135 * Disallow directory write attempts on read-only filesystems.
1138 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1142 if (cnp->cn_flags & SAVESTART) {
1143 ndp->ni_startdir = ndp->ni_dvp;
1144 VREF(ndp->ni_startdir);
1147 ni_dvp_unlocked = 2;
1148 if (ndp->ni_dvp != dp)
1152 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1153 VOP_UNLOCK(ndp->ni_dvp);
1154 ni_dvp_unlocked = 1;
1157 if (cnp->cn_flags & AUDITVNODE1)
1158 AUDIT_ARG_VNODE1(dp);
1159 else if (cnp->cn_flags & AUDITVNODE2)
1160 AUDIT_ARG_VNODE2(dp);
1162 if ((cnp->cn_flags & LOCKLEAF) == 0)
1166 * Because of shared lookup we may have the vnode shared locked, but
1167 * the caller may want it to be exclusively locked.
1169 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1170 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1171 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1172 if (VN_IS_DOOMED(dp)) {
1180 if (ni_dvp_unlocked != 2) {
1181 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1194 * relookup - lookup a path name component
1195 * Used by lookup to re-acquire things.
1198 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
1200 struct vnode *dp = NULL; /* the directory we are searching */
1201 int wantparent; /* 1 => wantparent or lockparent flag */
1202 int rdonly; /* lookup read-only flag bit */
1205 KASSERT(cnp->cn_flags & ISLASTCN,
1206 ("relookup: Not given last component."));
1208 * Setup: break out flag bits into variables.
1210 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
1211 KASSERT(wantparent, ("relookup: parent not wanted."));
1212 rdonly = cnp->cn_flags & RDONLY;
1213 cnp->cn_flags &= ~ISSYMLINK;
1215 cnp->cn_lkflags = LK_EXCLUSIVE;
1216 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1219 * Search a new directory.
1221 * The last component of the filename is left accessible via
1222 * cnp->cn_nameptr for callers that need the name. Callers needing
1223 * the name set the SAVENAME flag. When done, they assume
1224 * responsibility for freeing the pathname buffer.
1226 #ifdef NAMEI_DIAGNOSTIC
1227 printf("{%s}: ", cnp->cn_nameptr);
1231 * Check for "" which represents the root directory after slash
1234 if (cnp->cn_nameptr[0] == '\0') {
1236 * Support only LOOKUP for "/" because lookup()
1237 * can't succeed for CREATE, DELETE and RENAME.
1239 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1240 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1242 if (!(cnp->cn_flags & LOCKLEAF))
1245 /* XXX This should probably move to the top of function. */
1246 if (cnp->cn_flags & SAVESTART)
1247 panic("lookup: SAVESTART");
1251 if (cnp->cn_flags & ISDOTDOT)
1252 panic ("relookup: lookup on dot-dot");
1255 * We now have a segment name to search for, and a directory to search.
1257 #ifdef NAMEI_DIAGNOSTIC
1258 vn_printf(dp, "search in ");
1260 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1261 KASSERT(*vpp == NULL, ("leaf should be empty"));
1262 if (error != EJUSTRETURN)
1265 * If creating and at end of pathname, then can consider
1266 * allowing file to be created.
1272 /* ASSERT(dvp == ndp->ni_startdir) */
1273 if (cnp->cn_flags & SAVESTART)
1275 if ((cnp->cn_flags & LOCKPARENT) == 0)
1278 * We return with ni_vp NULL to indicate that the entry
1279 * doesn't currently exist, leaving a pointer to the
1280 * (possibly locked) directory vnode in ndp->ni_dvp.
1288 * Disallow directory write attempts on read-only filesystems.
1291 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1300 * Set the parent lock/ref state to the requested state.
1302 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1307 } else if (!wantparent)
1310 * Check for symbolic link
1312 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1313 ("relookup: symlink found.\n"));
1315 /* ASSERT(dvp == ndp->ni_startdir) */
1316 if (cnp->cn_flags & SAVESTART)
1319 if ((cnp->cn_flags & LOCKLEAF) == 0)
1329 NDINIT_ALL(struct nameidata *ndp, u_long op, u_long flags, enum uio_seg segflg,
1330 const char *namep, int dirfd, struct vnode *startdir, cap_rights_t *rightsp,
1334 ndp->ni_cnd.cn_nameiop = op;
1335 ndp->ni_cnd.cn_flags = flags;
1336 ndp->ni_segflg = segflg;
1337 ndp->ni_dirp = namep;
1338 ndp->ni_dirfd = dirfd;
1339 ndp->ni_startdir = startdir;
1340 ndp->ni_resflags = 0;
1341 filecaps_init(&ndp->ni_filecaps);
1342 ndp->ni_cnd.cn_thread = td;
1343 if (rightsp != NULL)
1344 ndp->ni_rightsneeded = *rightsp;
1346 cap_rights_init_zero(&ndp->ni_rightsneeded);
1350 * Free data allocated by namei(); see namei(9) for details.
1353 NDFREE(struct nameidata *ndp, const u_int flags)
1361 if (!(flags & NDF_NO_FREE_PNBUF) &&
1362 (ndp->ni_cnd.cn_flags & HASBUF)) {
1363 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1364 ndp->ni_cnd.cn_flags &= ~HASBUF;
1366 if (!(flags & NDF_NO_VP_UNLOCK) &&
1367 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1369 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1370 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1371 ndp->ni_dvp != ndp->ni_vp)
1373 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1382 VOP_UNLOCK(ndp->ni_vp);
1383 if (!(flags & NDF_NO_DVP_RELE) &&
1384 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1393 VOP_UNLOCK(ndp->ni_dvp);
1394 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1395 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1396 vrele(ndp->ni_startdir);
1397 ndp->ni_startdir = NULL;
1402 * Determine if there is a suitable alternate filename under the specified
1403 * prefix for the specified path. If the create flag is set, then the
1404 * alternate prefix will be used so long as the parent directory exists.
1405 * This is used by the various compatibility ABIs so that Linux binaries prefer
1406 * files under /compat/linux for example. The chosen path (whether under
1407 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1408 * to by pathbuf. The caller is responsible for free'ing the buffer from
1409 * the M_TEMP bucket if one is returned.
1412 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1413 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1415 struct nameidata nd, ndroot;
1416 char *ptr, *buf, *cp;
1420 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1423 /* Copy the prefix into the new pathname as a starting point. */
1424 len = strlcpy(buf, prefix, MAXPATHLEN);
1425 if (len >= MAXPATHLEN) {
1430 sz = MAXPATHLEN - len;
1433 /* Append the filename to the prefix. */
1434 if (pathseg == UIO_SYSSPACE)
1435 error = copystr(path, ptr, sz, &len);
1437 error = copyinstr(path, ptr, sz, &len);
1445 /* Only use a prefix with absolute pathnames. */
1451 if (dirfd != AT_FDCWD) {
1453 * We want the original because the "prefix" is
1454 * included in the already opened dirfd.
1456 bcopy(ptr, buf, len);
1461 * We know that there is a / somewhere in this pathname.
1462 * Search backwards for it, to find the file's parent dir
1463 * to see if it exists in the alternate tree. If it does,
1464 * and we want to create a file (cflag is set). We don't
1465 * need to worry about the root comparison in this case.
1469 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1472 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1478 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1485 * We now compare the vnode of the prefix to the one
1486 * vnode asked. If they resolve to be the same, then we
1487 * ignore the match so that the real root gets used.
1488 * This avoids the problem of traversing "../.." to find the
1489 * root directory and never finding it, because "/" resolves
1490 * to the emulation root directory. This is expensive :-(
1492 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1495 /* We shouldn't ever get an error from this namei(). */
1496 error = namei(&ndroot);
1498 if (nd.ni_vp == ndroot.ni_vp)
1501 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1502 vrele(ndroot.ni_vp);
1506 NDFREE(&nd, NDF_ONLY_PNBUF);
1510 /* If there was an error, use the original path name. */
1512 bcopy(ptr, buf, len);