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 <machine/_inttypes.h>
68 #include <security/audit/audit.h>
69 #include <security/mac/mac_framework.h>
73 #define NAMEI_DIAGNOSTIC 1
74 #undef NAMEI_DIAGNOSTIC
76 SDT_PROVIDER_DEFINE(vfs);
77 SDT_PROBE_DEFINE4(vfs, namei, lookup, entry, "struct vnode *", "char *",
78 "unsigned long", "bool");
79 SDT_PROBE_DEFINE3(vfs, namei, lookup, return, "int", "struct vnode *", "bool");
81 /* Allocation zone for namei. */
82 uma_zone_t namei_zone;
84 /* Placeholder vnode for mp traversal. */
85 static struct vnode *vp_crossmp;
88 crossmp_vop_islocked(struct vop_islocked_args *ap)
95 crossmp_vop_lock1(struct vop_lock1_args *ap)
98 struct lock *lk __unused;
99 const char *file __unused;
100 int flags, line __unused;
108 if ((flags & LK_SHARED) == 0)
109 panic("invalid lock request for crossmp");
111 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER, file, line,
112 flags & LK_INTERLOCK ? &VI_MTX(vp)->lock_object : NULL);
113 WITNESS_LOCK(&lk->lock_object, 0, file, line);
114 if ((flags & LK_INTERLOCK) != 0)
116 LOCK_LOG_LOCK("SLOCK", &lk->lock_object, 0, 0, ap->a_file, line);
121 crossmp_vop_unlock(struct vop_unlock_args *ap)
124 struct lock *lk __unused;
129 WITNESS_UNLOCK(&lk->lock_object, 0, LOCK_FILE, LOCK_LINE);
130 LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, LOCK_FILE,
135 static struct vop_vector crossmp_vnodeops = {
136 .vop_default = &default_vnodeops,
137 .vop_islocked = crossmp_vop_islocked,
138 .vop_lock1 = crossmp_vop_lock1,
139 .vop_unlock = crossmp_vop_unlock,
142 * VFS_VOP_VECTOR_REGISTER(crossmp_vnodeops) is not used here since the vnode
143 * gets allocated early. See nameiinit for the direct call below.
146 struct nameicap_tracker {
148 TAILQ_ENTRY(nameicap_tracker) nm_link;
151 /* Zone for cap mode tracker elements used for dotdot capability checks. */
152 static uma_zone_t nt_zone;
155 nameiinit(void *dummy __unused)
158 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
160 nt_zone = uma_zcreate("rentr", sizeof(struct nameicap_tracker),
161 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
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;
182 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp->v_type != VDIR)
184 if ((ndp->ni_lcf & (NI_LCF_BENEATH_ABS | NI_LCF_BENEATH_LATCHED)) ==
185 NI_LCF_BENEATH_ABS) {
186 MPASS((ndp->ni_lcf & NI_LCF_LATCH) != 0);
187 if (dp != ndp->ni_beneath_latch)
189 ndp->ni_lcf |= NI_LCF_BENEATH_LATCHED;
191 nt = uma_zalloc(nt_zone, M_WAITOK);
194 TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link);
198 nameicap_cleanup(struct nameidata *ndp, bool clean_latch)
200 struct nameicap_tracker *nt, *nt1;
202 KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) ||
203 (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative"));
204 TAILQ_FOREACH_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) {
205 TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link);
207 uma_zfree(nt_zone, nt);
209 if (clean_latch && (ndp->ni_lcf & NI_LCF_LATCH) != 0) {
210 ndp->ni_lcf &= ~NI_LCF_LATCH;
211 vrele(ndp->ni_beneath_latch);
216 * For dotdot lookups in capability mode, only allow the component
217 * lookup to succeed if the resulting directory was already traversed
218 * during the operation. Also fail dotdot lookups for non-local
219 * filesystems, where external agents might assist local lookups to
220 * escape the compartment.
223 nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp)
225 struct nameicap_tracker *nt;
228 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp == NULL ||
232 if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL &&
233 (mp->mnt_flag & MNT_LOCAL) == 0)
234 return (ENOTCAPABLE);
235 TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head,
240 if ((ndp->ni_lcf & NI_LCF_BENEATH_ABS) != 0) {
241 ndp->ni_lcf &= ~NI_LCF_BENEATH_LATCHED;
242 nameicap_cleanup(ndp, false);
245 return (ENOTCAPABLE);
249 namei_cleanup_cnp(struct componentname *cnp)
252 uma_zfree(namei_zone, cnp->cn_pnbuf);
254 cnp->cn_pnbuf = NULL;
255 cnp->cn_nameptr = NULL;
260 namei_handle_root(struct nameidata *ndp, struct vnode **dpp)
262 struct componentname *cnp;
265 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0) {
267 if (KTRPOINT(curthread, KTR_CAPFAIL))
268 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
270 return (ENOTCAPABLE);
272 if ((cnp->cn_flags & BENEATH) != 0) {
273 ndp->ni_lcf |= NI_LCF_BENEATH_ABS;
274 ndp->ni_lcf &= ~NI_LCF_BENEATH_LATCHED;
275 nameicap_cleanup(ndp, false);
277 while (*(cnp->cn_nameptr) == '/') {
281 *dpp = ndp->ni_rootdir;
287 namei_setup(struct nameidata *ndp, struct vnode **dpp, struct pwd **pwdp)
289 struct componentname *cnp;
294 struct filecaps dirfd_caps;
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 if (ndp->ni_dirfd == AT_FDCWD) {
323 if (KTRPOINT(td, KTR_CAPFAIL))
324 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
333 * Get starting point for the translation.
337 * The reference on ni_rootdir is acquired in the block below to avoid
338 * back-to-back atomics for absolute lookups.
340 ndp->ni_rootdir = pwd->pwd_rdir;
341 ndp->ni_topdir = pwd->pwd_jdir;
343 if (cnp->cn_pnbuf[0] == '/') {
344 ndp->ni_resflags |= NIRES_ABS;
345 error = namei_handle_root(ndp, dpp);
347 if (ndp->ni_startdir != NULL) {
348 *dpp = ndp->ni_startdir;
349 startdir_used = true;
350 } else if (ndp->ni_dirfd == AT_FDCWD) {
351 *dpp = pwd->pwd_cdir;
354 rights = *ndp->ni_rightsneeded;
355 cap_rights_set_one(&rights, CAP_LOOKUP);
357 if (cnp->cn_flags & AUDITVNODE1)
358 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
359 if (cnp->cn_flags & AUDITVNODE2)
360 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
362 * Effectively inlined fgetvp_rights, because we need to
363 * inspect the file as well as grabbing the vnode.
365 error = fget_cap(td, ndp->ni_dirfd, &rights,
366 &dfp, &ndp->ni_filecaps);
369 * Preserve the error; it should either be EBADF
370 * or capability-related, both of which can be
371 * safely returned to the caller.
374 if (dfp->f_ops == &badfileops) {
376 } else if (dfp->f_vnode == NULL) {
382 if ((dfp->f_flag & FSEARCH) != 0)
383 cnp->cn_flags |= NOEXECCHECK;
389 * If file descriptor doesn't have all rights,
390 * all lookups relative to it must also be
394 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights,
396 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
397 ndp->ni_filecaps.fc_nioctls != -1) {
398 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
402 if (error == 0 && (*dpp)->v_type != VDIR)
405 if (error == 0 && (cnp->cn_flags & BENEATH) != 0) {
406 if (ndp->ni_dirfd == AT_FDCWD) {
407 ndp->ni_beneath_latch = pwd->pwd_cdir;
408 vrefact(ndp->ni_beneath_latch);
410 rights = *ndp->ni_rightsneeded;
411 cap_rights_set_one(&rights, CAP_LOOKUP);
412 error = fgetvp_rights(td, ndp->ni_dirfd, &rights,
413 &dirfd_caps, &ndp->ni_beneath_latch);
414 if (error == 0 && (*dpp)->v_type != VDIR) {
415 vrele(ndp->ni_beneath_latch);
420 ndp->ni_lcf |= NI_LCF_LATCH;
423 * If we are auditing the kernel pathname, save the user pathname.
425 if (cnp->cn_flags & AUDITVNODE1)
426 AUDIT_ARG_UPATH1_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf);
427 if (cnp->cn_flags & AUDITVNODE2)
428 AUDIT_ARG_UPATH2_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf);
429 if (ndp->ni_startdir != NULL && !startdir_used)
430 vrele(ndp->ni_startdir);
437 MPASS((ndp->ni_lcf & (NI_LCF_BENEATH_ABS | NI_LCF_LATCH)) !=
439 if (((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 &&
440 lookup_cap_dotdot != 0) ||
441 ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) == 0 &&
442 (cnp->cn_flags & BENEATH) != 0))
443 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
444 SDT_PROBE4(vfs, namei, lookup, entry, *dpp, cnp->cn_pnbuf,
445 cnp->cn_flags, false);
451 * Convert a pathname into a pointer to a locked vnode.
453 * The FOLLOW flag is set when symbolic links are to be followed
454 * when they occur at the end of the name translation process.
455 * Symbolic links are always followed for all other pathname
456 * components other than the last.
458 * The segflg defines whether the name is to be copied from user
459 * space or kernel space.
461 * Overall outline of namei:
464 * get starting directory
465 * while (!done && !error) {
466 * call lookup to search path.
467 * if symbolic link, massage name in buffer and continue
471 namei(struct nameidata *ndp)
473 char *cp; /* pointer into pathname argument */
474 struct vnode *dp; /* the directory we are searching */
475 struct iovec aiov; /* uio for reading symbolic links */
476 struct componentname *cnp;
481 enum cache_fpl_status status;
485 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
486 KASSERT(cnp->cn_cred && td->td_proc, ("namei: bad cred/proc"));
487 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
488 ("namei: nameiop contaminated with flags"));
489 KASSERT((cnp->cn_flags & OPMASK) == 0,
490 ("namei: flags contaminated with nameiops"));
491 KASSERT((cnp->cn_flags & NAMEI_INTERNAL_FLAGS) == 0,
492 ("namei: unexpected flags: %" PRIx64 "\n",
493 cnp->cn_flags & NAMEI_INTERNAL_FLAGS));
494 if (cnp->cn_flags & NOCACHE)
495 KASSERT(cnp->cn_nameiop != LOOKUP,
496 ("%s: NOCACHE passed with LOOKUP", __func__));
497 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
498 ndp->ni_startdir->v_type == VBAD);
504 * Get a buffer for the name to be translated, and copy the
505 * name into the buffer.
507 if ((cnp->cn_flags & HASBUF) == 0)
508 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
509 if (ndp->ni_segflg == UIO_SYSSPACE)
510 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
513 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
516 if (__predict_false(error != 0)) {
517 namei_cleanup_cnp(cnp);
522 * Don't allow empty pathnames.
524 if (__predict_false(*cnp->cn_pnbuf == '\0')) {
525 namei_cleanup_cnp(cnp);
530 if (KTRPOINT(td, KTR_NAMEI)) {
531 KASSERT(cnp->cn_thread == curthread,
532 ("namei not using curthread"));
533 ktrnamei(cnp->cn_pnbuf);
537 cnp->cn_nameptr = cnp->cn_pnbuf;
540 * First try looking up the target without locking any vnodes.
542 * We may need to start from scratch or pick up where it left off.
544 error = cache_fplookup(ndp, &status, &pwd);
546 case CACHE_FPL_STATUS_UNSET:
547 __assert_unreachable();
549 case CACHE_FPL_STATUS_HANDLED:
551 case CACHE_FPL_STATUS_PARTIAL:
552 TAILQ_INIT(&ndp->ni_cap_tracker);
553 dp = ndp->ni_startdir;
555 case CACHE_FPL_STATUS_ABORTED:
556 TAILQ_INIT(&ndp->ni_cap_tracker);
557 error = namei_setup(ndp, &dp, &pwd);
559 namei_cleanup_cnp(cnp);
571 ndp->ni_startdir = dp;
576 * If not a symbolic link, we're done.
578 if ((cnp->cn_flags & ISSYMLINK) == 0) {
579 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
580 namei_cleanup_cnp(cnp);
582 cnp->cn_flags |= HASBUF;
583 if ((ndp->ni_lcf & (NI_LCF_BENEATH_ABS |
584 NI_LCF_BENEATH_LATCHED)) == NI_LCF_BENEATH_ABS) {
588 nameicap_cleanup(ndp, true);
589 SDT_PROBE3(vfs, namei, lookup, return, error,
590 (error == 0 ? ndp->ni_vp : NULL), false);
594 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
599 if ((cnp->cn_flags & NOMACCHECK) == 0) {
600 error = mac_vnode_check_readlink(td->td_ucred,
606 if (ndp->ni_pathlen > 1)
607 cp = uma_zalloc(namei_zone, M_WAITOK);
611 aiov.iov_len = MAXPATHLEN;
612 auio.uio_iov = &aiov;
615 auio.uio_rw = UIO_READ;
616 auio.uio_segflg = UIO_SYSSPACE;
618 auio.uio_resid = MAXPATHLEN;
619 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
621 if (ndp->ni_pathlen > 1)
622 uma_zfree(namei_zone, cp);
625 linklen = MAXPATHLEN - auio.uio_resid;
627 if (ndp->ni_pathlen > 1)
628 uma_zfree(namei_zone, cp);
632 if (linklen + ndp->ni_pathlen > MAXPATHLEN) {
633 if (ndp->ni_pathlen > 1)
634 uma_zfree(namei_zone, cp);
635 error = ENAMETOOLONG;
638 if (ndp->ni_pathlen > 1) {
639 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
640 uma_zfree(namei_zone, cnp->cn_pnbuf);
643 cnp->cn_pnbuf[linklen] = '\0';
644 ndp->ni_pathlen += linklen;
648 * Check if root directory should replace current directory.
650 cnp->cn_nameptr = cnp->cn_pnbuf;
651 if (*(cnp->cn_nameptr) == '/') {
653 error = namei_handle_root(ndp, &dp);
663 namei_cleanup_cnp(cnp);
664 nameicap_cleanup(ndp, true);
665 SDT_PROBE3(vfs, namei, lookup, return, error, NULL, false);
671 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
674 if (mp == NULL || ((lkflags & LK_SHARED) &&
675 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
676 ((cnflags & ISDOTDOT) &&
677 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
678 lkflags &= ~LK_SHARED;
679 lkflags |= LK_EXCLUSIVE;
681 lkflags |= LK_NODDLKTREAT;
686 needs_exclusive_leaf(struct mount *mp, int flags)
690 * Intermediate nodes can use shared locks, we only need to
691 * force an exclusive lock for leaf nodes.
693 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
696 /* Always use exclusive locks if LOCKSHARED isn't set. */
697 if (!(flags & LOCKSHARED))
701 * For lookups during open(), if the mount point supports
702 * extended shared operations, then use a shared lock for the
703 * leaf node, otherwise use an exclusive lock.
705 if ((flags & ISOPEN) != 0)
706 return (!MNT_EXTENDED_SHARED(mp));
709 * Lookup requests outside of open() that specify LOCKSHARED
710 * only need a shared lock on the leaf vnode.
717 * This is a very central and rather complicated routine.
719 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
720 * The starting directory is taken from ni_startdir. The pathname is
721 * descended until done, or a symbolic link is encountered. The variable
722 * ni_more is clear if the path is completed; it is set to one if a
723 * symbolic link needing interpretation is encountered.
725 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
726 * whether the name is to be looked up, created, renamed, or deleted.
727 * When CREATE, RENAME, or DELETE is specified, information usable in
728 * creating, renaming, or deleting a directory entry may be calculated.
729 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
730 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
731 * returned unlocked. Otherwise the parent directory is not returned. If
732 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
733 * the target is returned locked, otherwise it is returned unlocked.
734 * When creating or renaming and LOCKPARENT is specified, the target may not
735 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
737 * Overall outline of lookup:
740 * identify next component of name at ndp->ni_ptr
741 * handle degenerate case where name is null string
742 * if .. and crossing mount points and on mounted filesys, find parent
743 * call VOP_LOOKUP routine for next component name
744 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
745 * component vnode returned in ni_vp (if it exists), locked.
746 * if result vnode is mounted on and crossing mount points,
747 * find mounted on vnode
748 * if more components of name, do next level at dirloop
749 * return the answer in ni_vp, locked if LOCKLEAF set
750 * if LOCKPARENT set, return locked parent in ni_dvp
751 * if WANTPARENT set, return unlocked parent in ni_dvp
754 lookup(struct nameidata *ndp)
756 char *cp; /* pointer into pathname argument */
757 char *prev_ni_next; /* saved ndp->ni_next */
758 struct vnode *dp = NULL; /* the directory we are searching */
759 struct vnode *tdp; /* saved dp */
760 struct mount *mp; /* mount table entry */
762 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
763 int docache; /* == 0 do not cache last component */
764 int wantparent; /* 1 => wantparent or lockparent flag */
765 int rdonly; /* lookup read-only flag bit */
767 int dpunlocked = 0; /* dp has already been unlocked */
768 int relookup = 0; /* do not consume the path component */
769 struct componentname *cnp = &ndp->ni_cnd;
774 * Setup: break out flag bits into variables.
777 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
778 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
779 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
780 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
781 if (cnp->cn_nameiop == DELETE ||
782 (wantparent && cnp->cn_nameiop != CREATE &&
783 cnp->cn_nameiop != LOOKUP))
785 rdonly = cnp->cn_flags & RDONLY;
786 cnp->cn_flags &= ~ISSYMLINK;
789 * We use shared locks until we hit the parent of the last cn then
790 * we adjust based on the requesting flags.
792 cnp->cn_lkflags = LK_SHARED;
793 dp = ndp->ni_startdir;
794 ndp->ni_startdir = NULLVP;
796 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
801 * Search a new directory.
803 * The last component of the filename is left accessible via
804 * cnp->cn_nameptr for callers that need the name. Callers needing
805 * the name set the SAVENAME flag. When done, they assume
806 * responsibility for freeing the pathname buffer.
808 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
810 cnp->cn_namelen = cp - cnp->cn_nameptr;
811 if (cnp->cn_namelen > NAME_MAX) {
812 error = ENAMETOOLONG;
815 #ifdef NAMEI_DIAGNOSTIC
818 printf("{%s}: ", cnp->cn_nameptr);
821 prev_ni_pathlen = ndp->ni_pathlen;
822 ndp->ni_pathlen -= cnp->cn_namelen;
823 KASSERT(ndp->ni_pathlen <= PATH_MAX,
824 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
825 prev_ni_next = ndp->ni_next;
829 * Replace multiple slashes by a single slash and trailing slashes
830 * by a null. This must be done before VOP_LOOKUP() because some
831 * fs's don't know about trailing slashes. Remember if there were
832 * trailing slashes to handle symlinks, existing non-directories
833 * and non-existing files that won't be directories specially later.
835 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
839 *ndp->ni_next = '\0';
840 cnp->cn_flags |= TRAILINGSLASH;
845 cnp->cn_flags |= MAKEENTRY;
846 if (*cp == '\0' && docache == 0)
847 cnp->cn_flags &= ~MAKEENTRY;
848 if (cnp->cn_namelen == 2 &&
849 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
850 cnp->cn_flags |= ISDOTDOT;
852 cnp->cn_flags &= ~ISDOTDOT;
853 if (*ndp->ni_next == 0)
854 cnp->cn_flags |= ISLASTCN;
856 cnp->cn_flags &= ~ISLASTCN;
858 if ((cnp->cn_flags & ISLASTCN) != 0 &&
859 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
860 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
865 nameicap_tracker_add(ndp, dp);
868 * Check for degenerate name (e.g. / or "")
869 * which is a way of talking about a directory,
870 * e.g. like "/." or ".".
872 if (cnp->cn_nameptr[0] == '\0') {
873 if (dp->v_type != VDIR) {
877 if (cnp->cn_nameiop != LOOKUP) {
887 if (cnp->cn_flags & AUDITVNODE1)
888 AUDIT_ARG_VNODE1(dp);
889 else if (cnp->cn_flags & AUDITVNODE2)
890 AUDIT_ARG_VNODE2(dp);
892 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
894 /* XXX This should probably move to the top of function. */
895 if (cnp->cn_flags & SAVESTART)
896 panic("lookup: SAVESTART");
901 * Handle "..": five special cases.
902 * 0. If doing a capability lookup and lookup_cap_dotdot is
903 * disabled, return ENOTCAPABLE.
904 * 1. Return an error if this is the last component of
905 * the name and the operation is DELETE or RENAME.
906 * 2. If at root directory (e.g. after chroot)
907 * or at absolute root directory
908 * then ignore it so can't get out.
909 * 3. If this vnode is the root of a mounted
910 * filesystem, then replace it with the
911 * vnode which was mounted on so we take the
912 * .. in the other filesystem.
913 * 4. If the vnode is the top directory of
914 * the jail or chroot, don't let them out.
915 * 5. If doing a capability lookup and lookup_cap_dotdot is
916 * enabled, return ENOTCAPABLE if the lookup would escape
917 * from the initial file descriptor directory. Checks are
918 * done by ensuring that namei() already traversed the
919 * result of dotdot lookup.
921 if (cnp->cn_flags & ISDOTDOT) {
922 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT))
923 == NI_LCF_STRICTRELATIVE) {
925 if (KTRPOINT(curthread, KTR_CAPFAIL))
926 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
931 if ((cnp->cn_flags & ISLASTCN) != 0 &&
932 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
937 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
939 if (dp == pr->pr_root)
941 if (dp == ndp->ni_rootdir ||
942 dp == ndp->ni_topdir ||
945 ((dp->v_vflag & VV_ROOT) != 0 &&
946 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
952 if ((dp->v_vflag & VV_ROOT) == 0)
954 if (VN_IS_DOOMED(dp)) { /* forced unmount */
959 dp = dp->v_mount->mnt_vnodecovered;
963 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
964 LK_RETRY, ISDOTDOT));
965 error = nameicap_check_dotdot(ndp, dp);
968 if (KTRPOINT(curthread, KTR_CAPFAIL))
969 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
977 * We now have a segment name to search for, and a directory to search.
981 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp, cnp);
987 ASSERT_VOP_LOCKED(dp, "lookup");
989 * If we have a shared lock we may need to upgrade the lock for the
992 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
993 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
994 vn_lock(dp, LK_UPGRADE|LK_RETRY);
995 if (VN_IS_DOOMED(dp)) {
1000 * If we're looking up the last component and we need an exclusive
1001 * lock, adjust our lkflags.
1003 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
1004 cnp->cn_lkflags = LK_EXCLUSIVE;
1005 #ifdef NAMEI_DIAGNOSTIC
1006 vn_printf(dp, "lookup in ");
1008 lkflags_save = cnp->cn_lkflags;
1009 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
1011 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
1012 cnp->cn_lkflags = lkflags_save;
1014 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
1015 #ifdef NAMEI_DIAGNOSTIC
1016 printf("not found\n");
1018 if ((error == ENOENT) &&
1019 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
1020 (dp->v_mount->mnt_flag & MNT_UNION)) {
1022 dp = dp->v_mount->mnt_vnodecovered;
1026 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
1027 LK_RETRY, cnp->cn_flags));
1028 nameicap_tracker_add(ndp, dp);
1032 if (error == ERELOOKUP) {
1040 if (error != EJUSTRETURN)
1043 * At this point, we know we're at the end of the
1044 * pathname. If creating / renaming, we can consider
1045 * allowing the file or directory to be created / renamed,
1046 * provided we're not on a read-only filesystem.
1052 /* trailing slash only allowed for directories */
1053 if ((cnp->cn_flags & TRAILINGSLASH) &&
1054 !(cnp->cn_flags & WILLBEDIR)) {
1058 if ((cnp->cn_flags & LOCKPARENT) == 0)
1061 * We return with ni_vp NULL to indicate that the entry
1062 * doesn't currently exist, leaving a pointer to the
1063 * (possibly locked) directory vnode in ndp->ni_dvp.
1065 if (cnp->cn_flags & SAVESTART) {
1066 ndp->ni_startdir = ndp->ni_dvp;
1067 VREF(ndp->ni_startdir);
1073 #ifdef NAMEI_DIAGNOSTIC
1079 * Check to see if the vnode has been mounted on;
1080 * if so find the root of the mounted filesystem.
1082 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
1083 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
1084 if (vfs_busy(mp, 0))
1087 if (dp != ndp->ni_dvp)
1091 vrefact(vp_crossmp);
1092 ndp->ni_dvp = vp_crossmp;
1093 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
1094 cnp->cn_flags), &tdp);
1096 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
1097 panic("vp_crossmp exclusively locked or reclaimed");
1102 ndp->ni_vp = dp = tdp;
1106 * Check for symbolic link
1108 if ((dp->v_type == VLNK) &&
1109 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
1110 *ndp->ni_next == '/')) {
1111 cnp->cn_flags |= ISSYMLINK;
1112 if (VN_IS_DOOMED(dp)) {
1114 * We can't know whether the directory was mounted with
1115 * NOSYMFOLLOW, so we can't follow safely.
1120 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1125 * Symlink code always expects an unlocked dvp.
1127 if (ndp->ni_dvp != ndp->ni_vp) {
1128 VOP_UNLOCK(ndp->ni_dvp);
1129 ni_dvp_unlocked = 1;
1136 * Not a symbolic link that we will follow. Continue with the
1137 * next component if there is any; otherwise, we're done.
1139 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1140 ("lookup: invalid path state."));
1143 ndp->ni_pathlen = prev_ni_pathlen;
1144 ndp->ni_next = prev_ni_next;
1145 if (ndp->ni_dvp != dp)
1151 if (cnp->cn_flags & ISDOTDOT) {
1152 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1155 if (KTRPOINT(curthread, KTR_CAPFAIL))
1156 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1161 if (*ndp->ni_next == '/') {
1162 cnp->cn_nameptr = ndp->ni_next;
1163 while (*cnp->cn_nameptr == '/') {
1167 if (ndp->ni_dvp != dp)
1174 * If we're processing a path with a trailing slash,
1175 * check that the end result is a directory.
1177 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1182 * Disallow directory write attempts on read-only filesystems.
1185 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1189 if (cnp->cn_flags & SAVESTART) {
1190 ndp->ni_startdir = ndp->ni_dvp;
1191 VREF(ndp->ni_startdir);
1194 ni_dvp_unlocked = 2;
1195 if (ndp->ni_dvp != dp)
1199 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1200 VOP_UNLOCK(ndp->ni_dvp);
1201 ni_dvp_unlocked = 1;
1204 if (cnp->cn_flags & AUDITVNODE1)
1205 AUDIT_ARG_VNODE1(dp);
1206 else if (cnp->cn_flags & AUDITVNODE2)
1207 AUDIT_ARG_VNODE2(dp);
1209 if ((cnp->cn_flags & LOCKLEAF) == 0)
1213 * Because of shared lookup we may have the vnode shared locked, but
1214 * the caller may want it to be exclusively locked.
1216 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1217 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1218 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1219 if (VN_IS_DOOMED(dp)) {
1227 if (ni_dvp_unlocked != 2) {
1228 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1241 * relookup - lookup a path name component
1242 * Used by lookup to re-acquire things.
1245 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
1247 struct vnode *dp = NULL; /* the directory we are searching */
1248 int wantparent; /* 1 => wantparent or lockparent flag */
1249 int rdonly; /* lookup read-only flag bit */
1252 KASSERT(cnp->cn_flags & ISLASTCN,
1253 ("relookup: Not given last component."));
1255 * Setup: break out flag bits into variables.
1257 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
1258 KASSERT(wantparent, ("relookup: parent not wanted."));
1259 rdonly = cnp->cn_flags & RDONLY;
1260 cnp->cn_flags &= ~ISSYMLINK;
1262 cnp->cn_lkflags = LK_EXCLUSIVE;
1263 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1266 * Search a new directory.
1268 * The last component of the filename is left accessible via
1269 * cnp->cn_nameptr for callers that need the name. Callers needing
1270 * the name set the SAVENAME flag. When done, they assume
1271 * responsibility for freeing the pathname buffer.
1273 #ifdef NAMEI_DIAGNOSTIC
1274 printf("{%s}: ", cnp->cn_nameptr);
1278 * Check for "" which represents the root directory after slash
1281 if (cnp->cn_nameptr[0] == '\0') {
1283 * Support only LOOKUP for "/" because lookup()
1284 * can't succeed for CREATE, DELETE and RENAME.
1286 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1287 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1289 if (!(cnp->cn_flags & LOCKLEAF))
1292 /* XXX This should probably move to the top of function. */
1293 if (cnp->cn_flags & SAVESTART)
1294 panic("lookup: SAVESTART");
1298 if (cnp->cn_flags & ISDOTDOT)
1299 panic ("relookup: lookup on dot-dot");
1302 * We now have a segment name to search for, and a directory to search.
1304 #ifdef NAMEI_DIAGNOSTIC
1305 vn_printf(dp, "search in ");
1307 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1308 KASSERT(*vpp == NULL, ("leaf should be empty"));
1309 if (error != EJUSTRETURN)
1312 * If creating and at end of pathname, then can consider
1313 * allowing file to be created.
1319 /* ASSERT(dvp == ndp->ni_startdir) */
1320 if (cnp->cn_flags & SAVESTART)
1322 if ((cnp->cn_flags & LOCKPARENT) == 0)
1325 * We return with ni_vp NULL to indicate that the entry
1326 * doesn't currently exist, leaving a pointer to the
1327 * (possibly locked) directory vnode in ndp->ni_dvp.
1335 * Disallow directory write attempts on read-only filesystems.
1338 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1347 * Set the parent lock/ref state to the requested state.
1349 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1354 } else if (!wantparent)
1357 * Check for symbolic link
1359 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1360 ("relookup: symlink found.\n"));
1362 /* ASSERT(dvp == ndp->ni_startdir) */
1363 if (cnp->cn_flags & SAVESTART)
1366 if ((cnp->cn_flags & LOCKLEAF) == 0)
1376 * Free data allocated by namei(); see namei(9) for details.
1379 NDFREE_PNBUF(struct nameidata *ndp)
1382 if ((ndp->ni_cnd.cn_flags & HASBUF) != 0) {
1383 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1384 ndp->ni_cnd.cn_flags &= ~HASBUF;
1389 (NDFREE)(struct nameidata *ndp, const u_int flags)
1397 if (!(flags & NDF_NO_FREE_PNBUF)) {
1400 if (!(flags & NDF_NO_VP_UNLOCK) &&
1401 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1403 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1404 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1405 ndp->ni_dvp != ndp->ni_vp)
1407 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1416 VOP_UNLOCK(ndp->ni_vp);
1417 if (!(flags & NDF_NO_DVP_RELE) &&
1418 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1427 VOP_UNLOCK(ndp->ni_dvp);
1428 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1429 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1430 vrele(ndp->ni_startdir);
1431 ndp->ni_startdir = NULL;
1436 * Determine if there is a suitable alternate filename under the specified
1437 * prefix for the specified path. If the create flag is set, then the
1438 * alternate prefix will be used so long as the parent directory exists.
1439 * This is used by the various compatibility ABIs so that Linux binaries prefer
1440 * files under /compat/linux for example. The chosen path (whether under
1441 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1442 * to by pathbuf. The caller is responsible for free'ing the buffer from
1443 * the M_TEMP bucket if one is returned.
1446 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1447 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1449 struct nameidata nd, ndroot;
1450 char *ptr, *buf, *cp;
1454 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1457 /* Copy the prefix into the new pathname as a starting point. */
1458 len = strlcpy(buf, prefix, MAXPATHLEN);
1459 if (len >= MAXPATHLEN) {
1464 sz = MAXPATHLEN - len;
1467 /* Append the filename to the prefix. */
1468 if (pathseg == UIO_SYSSPACE)
1469 error = copystr(path, ptr, sz, &len);
1471 error = copyinstr(path, ptr, sz, &len);
1479 /* Only use a prefix with absolute pathnames. */
1485 if (dirfd != AT_FDCWD) {
1487 * We want the original because the "prefix" is
1488 * included in the already opened dirfd.
1490 bcopy(ptr, buf, len);
1495 * We know that there is a / somewhere in this pathname.
1496 * Search backwards for it, to find the file's parent dir
1497 * to see if it exists in the alternate tree. If it does,
1498 * and we want to create a file (cflag is set). We don't
1499 * need to worry about the root comparison in this case.
1503 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1506 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1512 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1519 * We now compare the vnode of the prefix to the one
1520 * vnode asked. If they resolve to be the same, then we
1521 * ignore the match so that the real root gets used.
1522 * This avoids the problem of traversing "../.." to find the
1523 * root directory and never finding it, because "/" resolves
1524 * to the emulation root directory. This is expensive :-(
1526 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1529 /* We shouldn't ever get an error from this namei(). */
1530 error = namei(&ndroot);
1532 if (nd.ni_vp == ndroot.ni_vp)
1535 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1536 vrele(ndroot.ni_vp);
1540 NDFREE(&nd, NDF_ONLY_PNBUF);
1544 /* If there was an error, use the original path name. */
1546 bcopy(ptr, buf, len);