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_DEFINE4(vfs, namei, lookup, entry, "struct vnode *", "char *",
75 "unsigned long", "bool");
76 SDT_PROBE_DEFINE3(vfs, namei, lookup, return, "int", "struct vnode *", "bool");
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)
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 namei_setup(struct nameidata *ndp, struct vnode **dpp, struct pwd **pwdp)
286 struct componentname *cnp;
291 struct filecaps dirfd_caps;
298 startdir_used = false;
301 #ifdef CAPABILITY_MODE
303 * In capability mode, lookups must be restricted to happen in
304 * the subtree with the root specified by the file descriptor:
305 * - The root must be real file descriptor, not the pseudo-descriptor
307 * - The passed path must be relative and not absolute.
308 * - If lookup_cap_dotdot is disabled, path must not contain the
310 * - If lookup_cap_dotdot is enabled, we verify that all '..'
311 * components lookups result in the directories which were
312 * previously walked by us, which prevents an escape from
315 if (IN_CAPABILITY_MODE(td) && (cnp->cn_flags & NOCAPCHECK) == 0) {
316 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
317 if (ndp->ni_dirfd == AT_FDCWD) {
319 if (KTRPOINT(td, KTR_CAPFAIL))
320 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
329 * Get starting point for the translation.
333 * The reference on ni_rootdir is acquired in the block below to avoid
334 * back-to-back atomics for absolute lookups.
336 ndp->ni_rootdir = pwd->pwd_rdir;
337 ndp->ni_topdir = pwd->pwd_jdir;
339 if (cnp->cn_pnbuf[0] == '/') {
340 ndp->ni_resflags |= NIRES_ABS;
341 error = namei_handle_root(ndp, dpp);
343 if (ndp->ni_startdir != NULL) {
344 *dpp = ndp->ni_startdir;
345 startdir_used = true;
346 } else if (ndp->ni_dirfd == AT_FDCWD) {
347 *dpp = pwd->pwd_cdir;
350 rights = *ndp->ni_rightsneeded;
351 cap_rights_set_one(&rights, CAP_LOOKUP);
353 if (cnp->cn_flags & AUDITVNODE1)
354 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
355 if (cnp->cn_flags & AUDITVNODE2)
356 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
358 * Effectively inlined fgetvp_rights, because we need to
359 * inspect the file as well as grabbing the vnode.
361 error = fget_cap(td, ndp->ni_dirfd, &rights,
362 &dfp, &ndp->ni_filecaps);
365 * Preserve the error; it should either be EBADF
366 * or capability-related, both of which can be
367 * safely returned to the caller.
370 if (dfp->f_ops == &badfileops) {
372 } else if (dfp->f_vnode == NULL) {
378 if ((dfp->f_flag & FSEARCH) != 0)
379 cnp->cn_flags |= NOEXECCHECK;
385 * If file descriptor doesn't have all rights,
386 * all lookups relative to it must also be
390 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights,
392 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
393 ndp->ni_filecaps.fc_nioctls != -1) {
394 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
398 if (error == 0 && (*dpp)->v_type != VDIR)
401 if (error == 0 && (cnp->cn_flags & BENEATH) != 0) {
402 if (ndp->ni_dirfd == AT_FDCWD) {
403 ndp->ni_beneath_latch = pwd->pwd_cdir;
404 vrefact(ndp->ni_beneath_latch);
406 rights = *ndp->ni_rightsneeded;
407 cap_rights_set_one(&rights, CAP_LOOKUP);
408 error = fgetvp_rights(td, ndp->ni_dirfd, &rights,
409 &dirfd_caps, &ndp->ni_beneath_latch);
410 if (error == 0 && (*dpp)->v_type != VDIR) {
411 vrele(ndp->ni_beneath_latch);
416 ndp->ni_lcf |= NI_LCF_LATCH;
419 * If we are auditing the kernel pathname, save the user pathname.
421 if (cnp->cn_flags & AUDITVNODE1)
422 AUDIT_ARG_UPATH1_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf);
423 if (cnp->cn_flags & AUDITVNODE2)
424 AUDIT_ARG_UPATH2_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf);
425 if (ndp->ni_startdir != NULL && !startdir_used)
426 vrele(ndp->ni_startdir);
432 MPASS((ndp->ni_lcf & (NI_LCF_BENEATH_ABS | NI_LCF_LATCH)) !=
434 if (((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 &&
435 lookup_cap_dotdot != 0) ||
436 ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) == 0 &&
437 (cnp->cn_flags & BENEATH) != 0))
438 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
439 SDT_PROBE4(vfs, namei, lookup, entry, *dpp, cnp->cn_pnbuf,
440 cnp->cn_flags, false);
446 * Convert a pathname into a pointer to a locked vnode.
448 * The FOLLOW flag is set when symbolic links are to be followed
449 * when they occur at the end of the name translation process.
450 * Symbolic links are always followed for all other pathname
451 * components other than the last.
453 * The segflg defines whether the name is to be copied from user
454 * space or kernel space.
456 * Overall outline of namei:
459 * get starting directory
460 * while (!done && !error) {
461 * call lookup to search path.
462 * if symbolic link, massage name in buffer and continue
466 namei(struct nameidata *ndp)
468 char *cp; /* pointer into pathname argument */
469 struct vnode *dp; /* the directory we are searching */
470 struct iovec aiov; /* uio for reading symbolic links */
471 struct componentname *cnp;
477 enum cache_fpl_status status;
482 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
483 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
484 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
485 ("namei: nameiop contaminated with flags"));
486 KASSERT((cnp->cn_flags & OPMASK) == 0,
487 ("namei: flags contaminated with nameiops"));
488 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
489 ndp->ni_startdir->v_type == VBAD);
490 TAILQ_INIT(&ndp->ni_cap_tracker);
495 /* We will set this ourselves if we need it. */
496 cnp->cn_flags &= ~TRAILINGSLASH;
501 * Get a buffer for the name to be translated, and copy the
502 * name into the buffer.
504 if ((cnp->cn_flags & HASBUF) == 0)
505 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
506 if (ndp->ni_segflg == UIO_SYSSPACE)
507 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
510 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
514 namei_cleanup_cnp(cnp);
518 cnp->cn_nameptr = cnp->cn_pnbuf;
521 * Don't allow empty pathnames.
523 if (*cnp->cn_pnbuf == '\0') {
524 namei_cleanup_cnp(cnp);
529 if (KTRPOINT(td, KTR_NAMEI)) {
530 KASSERT(cnp->cn_thread == curthread,
531 ("namei not using curthread"));
532 ktrnamei(cnp->cn_pnbuf);
537 * First try looking up the target without locking any vnodes.
539 * We may need to start from scratch or pick up where it left off.
541 error = cache_fplookup(ndp, &status, &pwd);
543 case CACHE_FPL_STATUS_UNSET:
544 __assert_unreachable();
546 case CACHE_FPL_STATUS_HANDLED:
548 case CACHE_FPL_STATUS_PARTIAL:
549 dp = ndp->ni_startdir;
551 case CACHE_FPL_STATUS_ABORTED:
552 error = namei_setup(ndp, &dp, &pwd);
554 namei_cleanup_cnp(cnp);
564 ndp->ni_startdir = dp;
569 * If not a symbolic link, we're done.
571 if ((cnp->cn_flags & ISSYMLINK) == 0) {
572 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
573 namei_cleanup_cnp(cnp);
575 cnp->cn_flags |= HASBUF;
576 if ((ndp->ni_lcf & (NI_LCF_BENEATH_ABS |
577 NI_LCF_BENEATH_LATCHED)) == NI_LCF_BENEATH_ABS) {
581 nameicap_cleanup(ndp, true);
582 SDT_PROBE3(vfs, namei, lookup, return, error,
583 (error == 0 ? ndp->ni_vp : NULL), false);
587 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
592 if ((cnp->cn_flags & NOMACCHECK) == 0) {
593 error = mac_vnode_check_readlink(td->td_ucred,
599 if (ndp->ni_pathlen > 1)
600 cp = uma_zalloc(namei_zone, M_WAITOK);
604 aiov.iov_len = MAXPATHLEN;
605 auio.uio_iov = &aiov;
608 auio.uio_rw = UIO_READ;
609 auio.uio_segflg = UIO_SYSSPACE;
611 auio.uio_resid = MAXPATHLEN;
612 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
614 if (ndp->ni_pathlen > 1)
615 uma_zfree(namei_zone, cp);
618 linklen = MAXPATHLEN - auio.uio_resid;
620 if (ndp->ni_pathlen > 1)
621 uma_zfree(namei_zone, cp);
625 if (linklen + ndp->ni_pathlen > MAXPATHLEN) {
626 if (ndp->ni_pathlen > 1)
627 uma_zfree(namei_zone, cp);
628 error = ENAMETOOLONG;
631 if (ndp->ni_pathlen > 1) {
632 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
633 uma_zfree(namei_zone, cnp->cn_pnbuf);
636 cnp->cn_pnbuf[linklen] = '\0';
637 ndp->ni_pathlen += linklen;
641 * Check if root directory should replace current directory.
643 cnp->cn_nameptr = cnp->cn_pnbuf;
644 if (*(cnp->cn_nameptr) == '/') {
646 error = namei_handle_root(ndp, &dp);
656 namei_cleanup_cnp(cnp);
657 nameicap_cleanup(ndp, true);
658 SDT_PROBE3(vfs, namei, lookup, return, error, NULL, false);
664 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
667 if (mp == NULL || ((lkflags & LK_SHARED) &&
668 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
669 ((cnflags & ISDOTDOT) &&
670 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
671 lkflags &= ~LK_SHARED;
672 lkflags |= LK_EXCLUSIVE;
674 lkflags |= LK_NODDLKTREAT;
679 needs_exclusive_leaf(struct mount *mp, int flags)
683 * Intermediate nodes can use shared locks, we only need to
684 * force an exclusive lock for leaf nodes.
686 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
689 /* Always use exclusive locks if LOCKSHARED isn't set. */
690 if (!(flags & LOCKSHARED))
694 * For lookups during open(), if the mount point supports
695 * extended shared operations, then use a shared lock for the
696 * leaf node, otherwise use an exclusive lock.
698 if ((flags & ISOPEN) != 0)
699 return (!MNT_EXTENDED_SHARED(mp));
702 * Lookup requests outside of open() that specify LOCKSHARED
703 * only need a shared lock on the leaf vnode.
710 * This is a very central and rather complicated routine.
712 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
713 * The starting directory is taken from ni_startdir. The pathname is
714 * descended until done, or a symbolic link is encountered. The variable
715 * ni_more is clear if the path is completed; it is set to one if a
716 * symbolic link needing interpretation is encountered.
718 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
719 * whether the name is to be looked up, created, renamed, or deleted.
720 * When CREATE, RENAME, or DELETE is specified, information usable in
721 * creating, renaming, or deleting a directory entry may be calculated.
722 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
723 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
724 * returned unlocked. Otherwise the parent directory is not returned. If
725 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
726 * the target is returned locked, otherwise it is returned unlocked.
727 * When creating or renaming and LOCKPARENT is specified, the target may not
728 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
730 * Overall outline of lookup:
733 * identify next component of name at ndp->ni_ptr
734 * handle degenerate case where name is null string
735 * if .. and crossing mount points and on mounted filesys, find parent
736 * call VOP_LOOKUP routine for next component name
737 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
738 * component vnode returned in ni_vp (if it exists), locked.
739 * if result vnode is mounted on and crossing mount points,
740 * find mounted on vnode
741 * if more components of name, do next level at dirloop
742 * return the answer in ni_vp, locked if LOCKLEAF set
743 * if LOCKPARENT set, return locked parent in ni_dvp
744 * if WANTPARENT set, return unlocked parent in ni_dvp
747 lookup(struct nameidata *ndp)
749 char *cp; /* pointer into pathname argument */
750 char *prev_ni_next; /* saved ndp->ni_next */
751 struct vnode *dp = NULL; /* the directory we are searching */
752 struct vnode *tdp; /* saved dp */
753 struct mount *mp; /* mount table entry */
755 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
756 int docache; /* == 0 do not cache last component */
757 int wantparent; /* 1 => wantparent or lockparent flag */
758 int rdonly; /* lookup read-only flag bit */
760 int dpunlocked = 0; /* dp has already been unlocked */
761 int relookup = 0; /* do not consume the path component */
762 struct componentname *cnp = &ndp->ni_cnd;
767 * Setup: break out flag bits into variables.
770 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
771 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
772 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
773 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
774 if (cnp->cn_nameiop == DELETE ||
775 (wantparent && cnp->cn_nameiop != CREATE &&
776 cnp->cn_nameiop != LOOKUP))
778 rdonly = cnp->cn_flags & RDONLY;
779 cnp->cn_flags &= ~ISSYMLINK;
782 * We use shared locks until we hit the parent of the last cn then
783 * we adjust based on the requesting flags.
785 cnp->cn_lkflags = LK_SHARED;
786 dp = ndp->ni_startdir;
787 ndp->ni_startdir = NULLVP;
789 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
794 * Search a new directory.
796 * The last component of the filename is left accessible via
797 * cnp->cn_nameptr for callers that need the name. Callers needing
798 * the name set the SAVENAME flag. When done, they assume
799 * responsibility for freeing the pathname buffer.
801 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
803 cnp->cn_namelen = cp - cnp->cn_nameptr;
804 if (cnp->cn_namelen > NAME_MAX) {
805 error = ENAMETOOLONG;
808 #ifdef NAMEI_DIAGNOSTIC
811 printf("{%s}: ", cnp->cn_nameptr);
814 prev_ni_pathlen = ndp->ni_pathlen;
815 ndp->ni_pathlen -= cnp->cn_namelen;
816 KASSERT(ndp->ni_pathlen <= PATH_MAX,
817 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
818 prev_ni_next = ndp->ni_next;
822 * Replace multiple slashes by a single slash and trailing slashes
823 * by a null. This must be done before VOP_LOOKUP() because some
824 * fs's don't know about trailing slashes. Remember if there were
825 * trailing slashes to handle symlinks, existing non-directories
826 * and non-existing files that won't be directories specially later.
828 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
832 *ndp->ni_next = '\0';
833 cnp->cn_flags |= TRAILINGSLASH;
838 cnp->cn_flags |= MAKEENTRY;
839 if (*cp == '\0' && docache == 0)
840 cnp->cn_flags &= ~MAKEENTRY;
841 if (cnp->cn_namelen == 2 &&
842 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
843 cnp->cn_flags |= ISDOTDOT;
845 cnp->cn_flags &= ~ISDOTDOT;
846 if (*ndp->ni_next == 0)
847 cnp->cn_flags |= ISLASTCN;
849 cnp->cn_flags &= ~ISLASTCN;
851 if ((cnp->cn_flags & ISLASTCN) != 0 &&
852 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
853 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
858 nameicap_tracker_add(ndp, dp);
861 * Check for degenerate name (e.g. / or "")
862 * which is a way of talking about a directory,
863 * e.g. like "/." or ".".
865 if (cnp->cn_nameptr[0] == '\0') {
866 if (dp->v_type != VDIR) {
870 if (cnp->cn_nameiop != LOOKUP) {
880 if (cnp->cn_flags & AUDITVNODE1)
881 AUDIT_ARG_VNODE1(dp);
882 else if (cnp->cn_flags & AUDITVNODE2)
883 AUDIT_ARG_VNODE2(dp);
885 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
887 /* XXX This should probably move to the top of function. */
888 if (cnp->cn_flags & SAVESTART)
889 panic("lookup: SAVESTART");
894 * Handle "..": five special cases.
895 * 0. If doing a capability lookup and lookup_cap_dotdot is
896 * disabled, return ENOTCAPABLE.
897 * 1. Return an error if this is the last component of
898 * the name and the operation is DELETE or RENAME.
899 * 2. If at root directory (e.g. after chroot)
900 * or at absolute root directory
901 * then ignore it so can't get out.
902 * 3. If this vnode is the root of a mounted
903 * filesystem, then replace it with the
904 * vnode which was mounted on so we take the
905 * .. in the other filesystem.
906 * 4. If the vnode is the top directory of
907 * the jail or chroot, don't let them out.
908 * 5. If doing a capability lookup and lookup_cap_dotdot is
909 * enabled, return ENOTCAPABLE if the lookup would escape
910 * from the initial file descriptor directory. Checks are
911 * done by ensuring that namei() already traversed the
912 * result of dotdot lookup.
914 if (cnp->cn_flags & ISDOTDOT) {
915 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT))
916 == NI_LCF_STRICTRELATIVE) {
918 if (KTRPOINT(curthread, KTR_CAPFAIL))
919 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
924 if ((cnp->cn_flags & ISLASTCN) != 0 &&
925 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
930 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
932 if (dp == pr->pr_root)
934 if (dp == ndp->ni_rootdir ||
935 dp == ndp->ni_topdir ||
938 ((dp->v_vflag & VV_ROOT) != 0 &&
939 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
945 if ((dp->v_vflag & VV_ROOT) == 0)
947 if (VN_IS_DOOMED(dp)) { /* forced unmount */
952 dp = dp->v_mount->mnt_vnodecovered;
956 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
957 LK_RETRY, ISDOTDOT));
958 error = nameicap_check_dotdot(ndp, dp);
961 if (KTRPOINT(curthread, KTR_CAPFAIL))
962 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
970 * We now have a segment name to search for, and a directory to search.
974 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp, cnp);
980 ASSERT_VOP_LOCKED(dp, "lookup");
982 * If we have a shared lock we may need to upgrade the lock for the
985 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
986 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
987 vn_lock(dp, LK_UPGRADE|LK_RETRY);
988 if (VN_IS_DOOMED(dp)) {
993 * If we're looking up the last component and we need an exclusive
994 * lock, adjust our lkflags.
996 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
997 cnp->cn_lkflags = LK_EXCLUSIVE;
998 #ifdef NAMEI_DIAGNOSTIC
999 vn_printf(dp, "lookup in ");
1001 lkflags_save = cnp->cn_lkflags;
1002 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
1004 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
1005 cnp->cn_lkflags = lkflags_save;
1007 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
1008 #ifdef NAMEI_DIAGNOSTIC
1009 printf("not found\n");
1011 if ((error == ENOENT) &&
1012 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
1013 (dp->v_mount->mnt_flag & MNT_UNION)) {
1015 dp = dp->v_mount->mnt_vnodecovered;
1019 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
1020 LK_RETRY, cnp->cn_flags));
1021 nameicap_tracker_add(ndp, dp);
1025 if (error == ERELOOKUP) {
1033 if (error != EJUSTRETURN)
1036 * At this point, we know we're at the end of the
1037 * pathname. If creating / renaming, we can consider
1038 * allowing the file or directory to be created / renamed,
1039 * provided we're not on a read-only filesystem.
1045 /* trailing slash only allowed for directories */
1046 if ((cnp->cn_flags & TRAILINGSLASH) &&
1047 !(cnp->cn_flags & WILLBEDIR)) {
1051 if ((cnp->cn_flags & LOCKPARENT) == 0)
1054 * We return with ni_vp NULL to indicate that the entry
1055 * doesn't currently exist, leaving a pointer to the
1056 * (possibly locked) directory vnode in ndp->ni_dvp.
1058 if (cnp->cn_flags & SAVESTART) {
1059 ndp->ni_startdir = ndp->ni_dvp;
1060 VREF(ndp->ni_startdir);
1066 #ifdef NAMEI_DIAGNOSTIC
1072 * Check to see if the vnode has been mounted on;
1073 * if so find the root of the mounted filesystem.
1075 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
1076 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
1077 if (vfs_busy(mp, 0))
1080 if (dp != ndp->ni_dvp)
1084 vrefact(vp_crossmp);
1085 ndp->ni_dvp = vp_crossmp;
1086 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
1087 cnp->cn_flags), &tdp);
1089 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
1090 panic("vp_crossmp exclusively locked or reclaimed");
1095 ndp->ni_vp = dp = tdp;
1099 * Check for symbolic link
1101 if ((dp->v_type == VLNK) &&
1102 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
1103 *ndp->ni_next == '/')) {
1104 cnp->cn_flags |= ISSYMLINK;
1105 if (VN_IS_DOOMED(dp)) {
1107 * We can't know whether the directory was mounted with
1108 * NOSYMFOLLOW, so we can't follow safely.
1113 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1118 * Symlink code always expects an unlocked dvp.
1120 if (ndp->ni_dvp != ndp->ni_vp) {
1121 VOP_UNLOCK(ndp->ni_dvp);
1122 ni_dvp_unlocked = 1;
1129 * Not a symbolic link that we will follow. Continue with the
1130 * next component if there is any; otherwise, we're done.
1132 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1133 ("lookup: invalid path state."));
1136 ndp->ni_pathlen = prev_ni_pathlen;
1137 ndp->ni_next = prev_ni_next;
1138 if (ndp->ni_dvp != dp)
1144 if (cnp->cn_flags & ISDOTDOT) {
1145 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1148 if (KTRPOINT(curthread, KTR_CAPFAIL))
1149 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1154 if (*ndp->ni_next == '/') {
1155 cnp->cn_nameptr = ndp->ni_next;
1156 while (*cnp->cn_nameptr == '/') {
1160 if (ndp->ni_dvp != dp)
1167 * If we're processing a path with a trailing slash,
1168 * check that the end result is a directory.
1170 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1175 * Disallow directory write attempts on read-only filesystems.
1178 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1182 if (cnp->cn_flags & SAVESTART) {
1183 ndp->ni_startdir = ndp->ni_dvp;
1184 VREF(ndp->ni_startdir);
1187 ni_dvp_unlocked = 2;
1188 if (ndp->ni_dvp != dp)
1192 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1193 VOP_UNLOCK(ndp->ni_dvp);
1194 ni_dvp_unlocked = 1;
1197 if (cnp->cn_flags & AUDITVNODE1)
1198 AUDIT_ARG_VNODE1(dp);
1199 else if (cnp->cn_flags & AUDITVNODE2)
1200 AUDIT_ARG_VNODE2(dp);
1202 if ((cnp->cn_flags & LOCKLEAF) == 0)
1206 * Because of shared lookup we may have the vnode shared locked, but
1207 * the caller may want it to be exclusively locked.
1209 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1210 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1211 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1212 if (VN_IS_DOOMED(dp)) {
1220 if (ni_dvp_unlocked != 2) {
1221 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1234 * relookup - lookup a path name component
1235 * Used by lookup to re-acquire things.
1238 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
1240 struct vnode *dp = NULL; /* the directory we are searching */
1241 int wantparent; /* 1 => wantparent or lockparent flag */
1242 int rdonly; /* lookup read-only flag bit */
1245 KASSERT(cnp->cn_flags & ISLASTCN,
1246 ("relookup: Not given last component."));
1248 * Setup: break out flag bits into variables.
1250 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
1251 KASSERT(wantparent, ("relookup: parent not wanted."));
1252 rdonly = cnp->cn_flags & RDONLY;
1253 cnp->cn_flags &= ~ISSYMLINK;
1255 cnp->cn_lkflags = LK_EXCLUSIVE;
1256 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1259 * Search a new directory.
1261 * The last component of the filename is left accessible via
1262 * cnp->cn_nameptr for callers that need the name. Callers needing
1263 * the name set the SAVENAME flag. When done, they assume
1264 * responsibility for freeing the pathname buffer.
1266 #ifdef NAMEI_DIAGNOSTIC
1267 printf("{%s}: ", cnp->cn_nameptr);
1271 * Check for "" which represents the root directory after slash
1274 if (cnp->cn_nameptr[0] == '\0') {
1276 * Support only LOOKUP for "/" because lookup()
1277 * can't succeed for CREATE, DELETE and RENAME.
1279 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1280 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1282 if (!(cnp->cn_flags & LOCKLEAF))
1285 /* XXX This should probably move to the top of function. */
1286 if (cnp->cn_flags & SAVESTART)
1287 panic("lookup: SAVESTART");
1291 if (cnp->cn_flags & ISDOTDOT)
1292 panic ("relookup: lookup on dot-dot");
1295 * We now have a segment name to search for, and a directory to search.
1297 #ifdef NAMEI_DIAGNOSTIC
1298 vn_printf(dp, "search in ");
1300 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1301 KASSERT(*vpp == NULL, ("leaf should be empty"));
1302 if (error != EJUSTRETURN)
1305 * If creating and at end of pathname, then can consider
1306 * allowing file to be created.
1312 /* ASSERT(dvp == ndp->ni_startdir) */
1313 if (cnp->cn_flags & SAVESTART)
1315 if ((cnp->cn_flags & LOCKPARENT) == 0)
1318 * We return with ni_vp NULL to indicate that the entry
1319 * doesn't currently exist, leaving a pointer to the
1320 * (possibly locked) directory vnode in ndp->ni_dvp.
1328 * Disallow directory write attempts on read-only filesystems.
1331 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1340 * Set the parent lock/ref state to the requested state.
1342 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1347 } else if (!wantparent)
1350 * Check for symbolic link
1352 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1353 ("relookup: symlink found.\n"));
1355 /* ASSERT(dvp == ndp->ni_startdir) */
1356 if (cnp->cn_flags & SAVESTART)
1359 if ((cnp->cn_flags & LOCKLEAF) == 0)
1369 * Free data allocated by namei(); see namei(9) for details.
1372 NDFREE_PNBUF(struct nameidata *ndp)
1375 if ((ndp->ni_cnd.cn_flags & HASBUF) != 0) {
1376 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1377 ndp->ni_cnd.cn_flags &= ~HASBUF;
1382 (NDFREE)(struct nameidata *ndp, const u_int flags)
1390 if (!(flags & NDF_NO_FREE_PNBUF)) {
1393 if (!(flags & NDF_NO_VP_UNLOCK) &&
1394 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1396 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1397 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1398 ndp->ni_dvp != ndp->ni_vp)
1400 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1409 VOP_UNLOCK(ndp->ni_vp);
1410 if (!(flags & NDF_NO_DVP_RELE) &&
1411 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1420 VOP_UNLOCK(ndp->ni_dvp);
1421 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1422 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1423 vrele(ndp->ni_startdir);
1424 ndp->ni_startdir = NULL;
1429 * Determine if there is a suitable alternate filename under the specified
1430 * prefix for the specified path. If the create flag is set, then the
1431 * alternate prefix will be used so long as the parent directory exists.
1432 * This is used by the various compatibility ABIs so that Linux binaries prefer
1433 * files under /compat/linux for example. The chosen path (whether under
1434 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1435 * to by pathbuf. The caller is responsible for free'ing the buffer from
1436 * the M_TEMP bucket if one is returned.
1439 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1440 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1442 struct nameidata nd, ndroot;
1443 char *ptr, *buf, *cp;
1447 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1450 /* Copy the prefix into the new pathname as a starting point. */
1451 len = strlcpy(buf, prefix, MAXPATHLEN);
1452 if (len >= MAXPATHLEN) {
1457 sz = MAXPATHLEN - len;
1460 /* Append the filename to the prefix. */
1461 if (pathseg == UIO_SYSSPACE)
1462 error = copystr(path, ptr, sz, &len);
1464 error = copyinstr(path, ptr, sz, &len);
1472 /* Only use a prefix with absolute pathnames. */
1478 if (dirfd != AT_FDCWD) {
1480 * We want the original because the "prefix" is
1481 * included in the already opened dirfd.
1483 bcopy(ptr, buf, len);
1488 * We know that there is a / somewhere in this pathname.
1489 * Search backwards for it, to find the file's parent dir
1490 * to see if it exists in the alternate tree. If it does,
1491 * and we want to create a file (cflag is set). We don't
1492 * need to worry about the root comparison in this case.
1496 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1499 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1505 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1512 * We now compare the vnode of the prefix to the one
1513 * vnode asked. If they resolve to be the same, then we
1514 * ignore the match so that the real root gets used.
1515 * This avoids the problem of traversing "../.." to find the
1516 * root directory and never finding it, because "/" resolves
1517 * to the emulation root directory. This is expensive :-(
1519 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1522 /* We shouldn't ever get an error from this namei(). */
1523 error = namei(&ndroot);
1525 if (nd.ni_vp == ndroot.ni_vp)
1528 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1529 vrele(ndroot.ni_vp);
1533 NDFREE(&nd, NDF_ONLY_PNBUF);
1537 /* If there was an error, use the original path name. */
1539 bcopy(ptr, buf, len);