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;
292 int error, startdir_used;
299 #ifdef CAPABILITY_MODE
301 * In capability mode, lookups must be restricted to happen in
302 * the subtree with the root specified by the file descriptor:
303 * - The root must be real file descriptor, not the pseudo-descriptor
305 * - The passed path must be relative and not absolute.
306 * - If lookup_cap_dotdot is disabled, path must not contain the
308 * - If lookup_cap_dotdot is enabled, we verify that all '..'
309 * components lookups result in the directories which were
310 * previously walked by us, which prevents an escape from
313 if (IN_CAPABILITY_MODE(td) && (cnp->cn_flags & NOCAPCHECK) == 0) {
314 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
315 if (ndp->ni_dirfd == AT_FDCWD) {
317 if (KTRPOINT(td, KTR_CAPFAIL))
318 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
327 * Get starting point for the translation.
331 * The reference on ni_rootdir is acquired in the block below to avoid
332 * back-to-back atomics for absolute lookups.
334 ndp->ni_rootdir = pwd->pwd_rdir;
335 ndp->ni_topdir = pwd->pwd_jdir;
337 if (cnp->cn_pnbuf[0] == '/') {
338 ndp->ni_resflags |= NIRES_ABS;
339 error = namei_handle_root(ndp, dpp);
341 if (ndp->ni_startdir != NULL) {
342 *dpp = ndp->ni_startdir;
344 } else if (ndp->ni_dirfd == AT_FDCWD) {
345 *dpp = pwd->pwd_cdir;
348 rights = ndp->ni_rightsneeded;
349 cap_rights_set_one(&rights, CAP_LOOKUP);
351 if (cnp->cn_flags & AUDITVNODE1)
352 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
353 if (cnp->cn_flags & AUDITVNODE2)
354 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
356 * Effectively inlined fgetvp_rights, because we need to
357 * inspect the file as well as grabbing the vnode.
359 error = fget_cap(td, ndp->ni_dirfd, &rights,
360 &dfp, &ndp->ni_filecaps);
363 * Preserve the error; it should either be EBADF
364 * or capability-related, both of which can be
365 * safely returned to the caller.
368 if (dfp->f_ops == &badfileops) {
370 } else if (dfp->f_vnode == NULL) {
376 if ((dfp->f_flag & FSEARCH) != 0)
377 cnp->cn_flags |= NOEXECCHECK;
383 * If file descriptor doesn't have all rights,
384 * all lookups relative to it must also be
388 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights,
390 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
391 ndp->ni_filecaps.fc_nioctls != -1) {
392 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
396 if (error == 0 && (*dpp)->v_type != VDIR)
399 if (error == 0 && (cnp->cn_flags & BENEATH) != 0) {
400 if (ndp->ni_dirfd == AT_FDCWD) {
401 ndp->ni_beneath_latch = pwd->pwd_cdir;
402 vrefact(ndp->ni_beneath_latch);
404 rights = ndp->ni_rightsneeded;
405 cap_rights_set_one(&rights, CAP_LOOKUP);
406 error = fgetvp_rights(td, ndp->ni_dirfd, &rights,
407 &dirfd_caps, &ndp->ni_beneath_latch);
408 if (error == 0 && (*dpp)->v_type != VDIR) {
409 vrele(ndp->ni_beneath_latch);
414 ndp->ni_lcf |= NI_LCF_LATCH;
417 * If we are auditing the kernel pathname, save the user pathname.
419 if (cnp->cn_flags & AUDITVNODE1)
420 AUDIT_ARG_UPATH1_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf);
421 if (cnp->cn_flags & AUDITVNODE2)
422 AUDIT_ARG_UPATH2_VP(td, ndp->ni_rootdir, *dpp, cnp->cn_pnbuf);
423 if (ndp->ni_startdir != NULL && !startdir_used)
424 vrele(ndp->ni_startdir);
430 MPASS((ndp->ni_lcf & (NI_LCF_BENEATH_ABS | NI_LCF_LATCH)) !=
432 if (((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 &&
433 lookup_cap_dotdot != 0) ||
434 ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) == 0 &&
435 (cnp->cn_flags & BENEATH) != 0))
436 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
437 SDT_PROBE4(vfs, namei, lookup, entry, *dpp, cnp->cn_pnbuf,
438 cnp->cn_flags, false);
444 * Convert a pathname into a pointer to a locked vnode.
446 * The FOLLOW flag is set when symbolic links are to be followed
447 * when they occur at the end of the name translation process.
448 * Symbolic links are always followed for all other pathname
449 * components other than the last.
451 * The segflg defines whether the name is to be copied from user
452 * space or kernel space.
454 * Overall outline of namei:
457 * get starting directory
458 * while (!done && !error) {
459 * call lookup to search path.
460 * if symbolic link, massage name in buffer and continue
464 namei(struct nameidata *ndp)
466 char *cp; /* pointer into pathname argument */
467 struct vnode *dp; /* the directory we are searching */
468 struct iovec aiov; /* uio for reading symbolic links */
469 struct componentname *cnp;
475 enum cache_fpl_status status;
480 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
481 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
482 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
483 ("namei: nameiop contaminated with flags"));
484 KASSERT((cnp->cn_flags & OPMASK) == 0,
485 ("namei: flags contaminated with nameiops"));
486 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
487 ndp->ni_startdir->v_type == VBAD);
488 TAILQ_INIT(&ndp->ni_cap_tracker);
493 /* We will set this ourselves if we need it. */
494 cnp->cn_flags &= ~TRAILINGSLASH;
499 * Get a buffer for the name to be translated, and copy the
500 * name into the buffer.
502 if ((cnp->cn_flags & HASBUF) == 0)
503 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
504 if (ndp->ni_segflg == UIO_SYSSPACE)
505 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
508 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
512 namei_cleanup_cnp(cnp);
516 cnp->cn_nameptr = cnp->cn_pnbuf;
519 * Don't allow empty pathnames.
521 if (*cnp->cn_pnbuf == '\0') {
522 namei_cleanup_cnp(cnp);
527 if (KTRPOINT(td, KTR_NAMEI)) {
528 KASSERT(cnp->cn_thread == curthread,
529 ("namei not using curthread"));
530 ktrnamei(cnp->cn_pnbuf);
535 * First try looking up the target without locking any vnodes.
537 * We may need to start from scratch or pick up where it left off.
539 error = cache_fplookup(ndp, &status, &pwd);
541 case CACHE_FPL_STATUS_UNSET:
542 __assert_unreachable();
544 case CACHE_FPL_STATUS_HANDLED:
546 case CACHE_FPL_STATUS_PARTIAL:
547 dp = ndp->ni_startdir;
549 case CACHE_FPL_STATUS_ABORTED:
550 error = namei_setup(ndp, &dp, &pwd);
552 namei_cleanup_cnp(cnp);
562 ndp->ni_startdir = dp;
567 * If not a symbolic link, we're done.
569 if ((cnp->cn_flags & ISSYMLINK) == 0) {
570 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
571 namei_cleanup_cnp(cnp);
573 cnp->cn_flags |= HASBUF;
574 if ((ndp->ni_lcf & (NI_LCF_BENEATH_ABS |
575 NI_LCF_BENEATH_LATCHED)) == NI_LCF_BENEATH_ABS) {
579 nameicap_cleanup(ndp, true);
580 SDT_PROBE3(vfs, namei, lookup, return, error,
581 (error == 0 ? ndp->ni_vp : NULL), false);
585 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
590 if ((cnp->cn_flags & NOMACCHECK) == 0) {
591 error = mac_vnode_check_readlink(td->td_ucred,
597 if (ndp->ni_pathlen > 1)
598 cp = uma_zalloc(namei_zone, M_WAITOK);
602 aiov.iov_len = MAXPATHLEN;
603 auio.uio_iov = &aiov;
606 auio.uio_rw = UIO_READ;
607 auio.uio_segflg = UIO_SYSSPACE;
609 auio.uio_resid = MAXPATHLEN;
610 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
612 if (ndp->ni_pathlen > 1)
613 uma_zfree(namei_zone, cp);
616 linklen = MAXPATHLEN - auio.uio_resid;
618 if (ndp->ni_pathlen > 1)
619 uma_zfree(namei_zone, cp);
623 if (linklen + ndp->ni_pathlen > MAXPATHLEN) {
624 if (ndp->ni_pathlen > 1)
625 uma_zfree(namei_zone, cp);
626 error = ENAMETOOLONG;
629 if (ndp->ni_pathlen > 1) {
630 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
631 uma_zfree(namei_zone, cnp->cn_pnbuf);
634 cnp->cn_pnbuf[linklen] = '\0';
635 ndp->ni_pathlen += linklen;
639 * Check if root directory should replace current directory.
641 cnp->cn_nameptr = cnp->cn_pnbuf;
642 if (*(cnp->cn_nameptr) == '/') {
644 error = namei_handle_root(ndp, &dp);
654 namei_cleanup_cnp(cnp);
655 nameicap_cleanup(ndp, true);
656 SDT_PROBE3(vfs, namei, lookup, return, error, NULL, false);
662 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
665 if (mp == NULL || ((lkflags & LK_SHARED) &&
666 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
667 ((cnflags & ISDOTDOT) &&
668 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
669 lkflags &= ~LK_SHARED;
670 lkflags |= LK_EXCLUSIVE;
672 lkflags |= LK_NODDLKTREAT;
677 needs_exclusive_leaf(struct mount *mp, int flags)
681 * Intermediate nodes can use shared locks, we only need to
682 * force an exclusive lock for leaf nodes.
684 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
687 /* Always use exclusive locks if LOCKSHARED isn't set. */
688 if (!(flags & LOCKSHARED))
692 * For lookups during open(), if the mount point supports
693 * extended shared operations, then use a shared lock for the
694 * leaf node, otherwise use an exclusive lock.
696 if ((flags & ISOPEN) != 0)
697 return (!MNT_EXTENDED_SHARED(mp));
700 * Lookup requests outside of open() that specify LOCKSHARED
701 * only need a shared lock on the leaf vnode.
708 * This is a very central and rather complicated routine.
710 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
711 * The starting directory is taken from ni_startdir. The pathname is
712 * descended until done, or a symbolic link is encountered. The variable
713 * ni_more is clear if the path is completed; it is set to one if a
714 * symbolic link needing interpretation is encountered.
716 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
717 * whether the name is to be looked up, created, renamed, or deleted.
718 * When CREATE, RENAME, or DELETE is specified, information usable in
719 * creating, renaming, or deleting a directory entry may be calculated.
720 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
721 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
722 * returned unlocked. Otherwise the parent directory is not returned. If
723 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
724 * the target is returned locked, otherwise it is returned unlocked.
725 * When creating or renaming and LOCKPARENT is specified, the target may not
726 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
728 * Overall outline of lookup:
731 * identify next component of name at ndp->ni_ptr
732 * handle degenerate case where name is null string
733 * if .. and crossing mount points and on mounted filesys, find parent
734 * call VOP_LOOKUP routine for next component name
735 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
736 * component vnode returned in ni_vp (if it exists), locked.
737 * if result vnode is mounted on and crossing mount points,
738 * find mounted on vnode
739 * if more components of name, do next level at dirloop
740 * return the answer in ni_vp, locked if LOCKLEAF set
741 * if LOCKPARENT set, return locked parent in ni_dvp
742 * if WANTPARENT set, return unlocked parent in ni_dvp
745 lookup(struct nameidata *ndp)
747 char *cp; /* pointer into pathname argument */
748 char *prev_ni_next; /* saved ndp->ni_next */
749 struct vnode *dp = NULL; /* the directory we are searching */
750 struct vnode *tdp; /* saved dp */
751 struct mount *mp; /* mount table entry */
753 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
754 int docache; /* == 0 do not cache last component */
755 int wantparent; /* 1 => wantparent or lockparent flag */
756 int rdonly; /* lookup read-only flag bit */
758 int dpunlocked = 0; /* dp has already been unlocked */
759 int relookup = 0; /* do not consume the path component */
760 struct componentname *cnp = &ndp->ni_cnd;
765 * Setup: break out flag bits into variables.
768 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
769 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
770 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
771 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
772 if (cnp->cn_nameiop == DELETE ||
773 (wantparent && cnp->cn_nameiop != CREATE &&
774 cnp->cn_nameiop != LOOKUP))
776 rdonly = cnp->cn_flags & RDONLY;
777 cnp->cn_flags &= ~ISSYMLINK;
780 * We use shared locks until we hit the parent of the last cn then
781 * we adjust based on the requesting flags.
783 cnp->cn_lkflags = LK_SHARED;
784 dp = ndp->ni_startdir;
785 ndp->ni_startdir = NULLVP;
787 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
792 * Search a new directory.
794 * The last component of the filename is left accessible via
795 * cnp->cn_nameptr for callers that need the name. Callers needing
796 * the name set the SAVENAME flag. When done, they assume
797 * responsibility for freeing the pathname buffer.
799 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
801 cnp->cn_namelen = cp - cnp->cn_nameptr;
802 if (cnp->cn_namelen > NAME_MAX) {
803 error = ENAMETOOLONG;
806 #ifdef NAMEI_DIAGNOSTIC
809 printf("{%s}: ", cnp->cn_nameptr);
812 prev_ni_pathlen = ndp->ni_pathlen;
813 ndp->ni_pathlen -= cnp->cn_namelen;
814 KASSERT(ndp->ni_pathlen <= PATH_MAX,
815 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
816 prev_ni_next = ndp->ni_next;
820 * Replace multiple slashes by a single slash and trailing slashes
821 * by a null. This must be done before VOP_LOOKUP() because some
822 * fs's don't know about trailing slashes. Remember if there were
823 * trailing slashes to handle symlinks, existing non-directories
824 * and non-existing files that won't be directories specially later.
826 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
830 *ndp->ni_next = '\0';
831 cnp->cn_flags |= TRAILINGSLASH;
836 cnp->cn_flags |= MAKEENTRY;
837 if (*cp == '\0' && docache == 0)
838 cnp->cn_flags &= ~MAKEENTRY;
839 if (cnp->cn_namelen == 2 &&
840 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
841 cnp->cn_flags |= ISDOTDOT;
843 cnp->cn_flags &= ~ISDOTDOT;
844 if (*ndp->ni_next == 0)
845 cnp->cn_flags |= ISLASTCN;
847 cnp->cn_flags &= ~ISLASTCN;
849 if ((cnp->cn_flags & ISLASTCN) != 0 &&
850 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
851 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
856 nameicap_tracker_add(ndp, dp);
859 * Check for degenerate name (e.g. / or "")
860 * which is a way of talking about a directory,
861 * e.g. like "/." or ".".
863 if (cnp->cn_nameptr[0] == '\0') {
864 if (dp->v_type != VDIR) {
868 if (cnp->cn_nameiop != LOOKUP) {
878 if (cnp->cn_flags & AUDITVNODE1)
879 AUDIT_ARG_VNODE1(dp);
880 else if (cnp->cn_flags & AUDITVNODE2)
881 AUDIT_ARG_VNODE2(dp);
883 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
885 /* XXX This should probably move to the top of function. */
886 if (cnp->cn_flags & SAVESTART)
887 panic("lookup: SAVESTART");
892 * Handle "..": five special cases.
893 * 0. If doing a capability lookup and lookup_cap_dotdot is
894 * disabled, return ENOTCAPABLE.
895 * 1. Return an error if this is the last component of
896 * the name and the operation is DELETE or RENAME.
897 * 2. If at root directory (e.g. after chroot)
898 * or at absolute root directory
899 * then ignore it so can't get out.
900 * 3. If this vnode is the root of a mounted
901 * filesystem, then replace it with the
902 * vnode which was mounted on so we take the
903 * .. in the other filesystem.
904 * 4. If the vnode is the top directory of
905 * the jail or chroot, don't let them out.
906 * 5. If doing a capability lookup and lookup_cap_dotdot is
907 * enabled, return ENOTCAPABLE if the lookup would escape
908 * from the initial file descriptor directory. Checks are
909 * done by ensuring that namei() already traversed the
910 * result of dotdot lookup.
912 if (cnp->cn_flags & ISDOTDOT) {
913 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT))
914 == NI_LCF_STRICTRELATIVE) {
916 if (KTRPOINT(curthread, KTR_CAPFAIL))
917 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
922 if ((cnp->cn_flags & ISLASTCN) != 0 &&
923 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
928 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
930 if (dp == pr->pr_root)
932 if (dp == ndp->ni_rootdir ||
933 dp == ndp->ni_topdir ||
936 ((dp->v_vflag & VV_ROOT) != 0 &&
937 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
943 if ((dp->v_vflag & VV_ROOT) == 0)
945 if (VN_IS_DOOMED(dp)) { /* forced unmount */
950 dp = dp->v_mount->mnt_vnodecovered;
954 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
955 LK_RETRY, ISDOTDOT));
956 error = nameicap_check_dotdot(ndp, dp);
959 if (KTRPOINT(curthread, KTR_CAPFAIL))
960 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
968 * We now have a segment name to search for, and a directory to search.
972 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp, cnp);
978 ASSERT_VOP_LOCKED(dp, "lookup");
980 * If we have a shared lock we may need to upgrade the lock for the
983 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
984 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
985 vn_lock(dp, LK_UPGRADE|LK_RETRY);
986 if (VN_IS_DOOMED(dp)) {
991 * If we're looking up the last component and we need an exclusive
992 * lock, adjust our lkflags.
994 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
995 cnp->cn_lkflags = LK_EXCLUSIVE;
996 #ifdef NAMEI_DIAGNOSTIC
997 vn_printf(dp, "lookup in ");
999 lkflags_save = cnp->cn_lkflags;
1000 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
1002 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
1003 cnp->cn_lkflags = lkflags_save;
1005 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
1006 #ifdef NAMEI_DIAGNOSTIC
1007 printf("not found\n");
1009 if ((error == ENOENT) &&
1010 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
1011 (dp->v_mount->mnt_flag & MNT_UNION)) {
1013 dp = dp->v_mount->mnt_vnodecovered;
1017 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
1018 LK_RETRY, cnp->cn_flags));
1019 nameicap_tracker_add(ndp, dp);
1023 if (error == ERELOOKUP) {
1031 if (error != EJUSTRETURN)
1034 * At this point, we know we're at the end of the
1035 * pathname. If creating / renaming, we can consider
1036 * allowing the file or directory to be created / renamed,
1037 * provided we're not on a read-only filesystem.
1043 /* trailing slash only allowed for directories */
1044 if ((cnp->cn_flags & TRAILINGSLASH) &&
1045 !(cnp->cn_flags & WILLBEDIR)) {
1049 if ((cnp->cn_flags & LOCKPARENT) == 0)
1052 * We return with ni_vp NULL to indicate that the entry
1053 * doesn't currently exist, leaving a pointer to the
1054 * (possibly locked) directory vnode in ndp->ni_dvp.
1056 if (cnp->cn_flags & SAVESTART) {
1057 ndp->ni_startdir = ndp->ni_dvp;
1058 VREF(ndp->ni_startdir);
1064 #ifdef NAMEI_DIAGNOSTIC
1070 * Check to see if the vnode has been mounted on;
1071 * if so find the root of the mounted filesystem.
1073 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
1074 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
1075 if (vfs_busy(mp, 0))
1078 if (dp != ndp->ni_dvp)
1082 vrefact(vp_crossmp);
1083 ndp->ni_dvp = vp_crossmp;
1084 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
1085 cnp->cn_flags), &tdp);
1087 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
1088 panic("vp_crossmp exclusively locked or reclaimed");
1093 ndp->ni_vp = dp = tdp;
1097 * Check for symbolic link
1099 if ((dp->v_type == VLNK) &&
1100 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
1101 *ndp->ni_next == '/')) {
1102 cnp->cn_flags |= ISSYMLINK;
1103 if (VN_IS_DOOMED(dp)) {
1105 * We can't know whether the directory was mounted with
1106 * NOSYMFOLLOW, so we can't follow safely.
1111 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1116 * Symlink code always expects an unlocked dvp.
1118 if (ndp->ni_dvp != ndp->ni_vp) {
1119 VOP_UNLOCK(ndp->ni_dvp);
1120 ni_dvp_unlocked = 1;
1127 * Not a symbolic link that we will follow. Continue with the
1128 * next component if there is any; otherwise, we're done.
1130 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1131 ("lookup: invalid path state."));
1134 ndp->ni_pathlen = prev_ni_pathlen;
1135 ndp->ni_next = prev_ni_next;
1136 if (ndp->ni_dvp != dp)
1142 if (cnp->cn_flags & ISDOTDOT) {
1143 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1146 if (KTRPOINT(curthread, KTR_CAPFAIL))
1147 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1152 if (*ndp->ni_next == '/') {
1153 cnp->cn_nameptr = ndp->ni_next;
1154 while (*cnp->cn_nameptr == '/') {
1158 if (ndp->ni_dvp != dp)
1165 * If we're processing a path with a trailing slash,
1166 * check that the end result is a directory.
1168 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1173 * Disallow directory write attempts on read-only filesystems.
1176 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1180 if (cnp->cn_flags & SAVESTART) {
1181 ndp->ni_startdir = ndp->ni_dvp;
1182 VREF(ndp->ni_startdir);
1185 ni_dvp_unlocked = 2;
1186 if (ndp->ni_dvp != dp)
1190 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1191 VOP_UNLOCK(ndp->ni_dvp);
1192 ni_dvp_unlocked = 1;
1195 if (cnp->cn_flags & AUDITVNODE1)
1196 AUDIT_ARG_VNODE1(dp);
1197 else if (cnp->cn_flags & AUDITVNODE2)
1198 AUDIT_ARG_VNODE2(dp);
1200 if ((cnp->cn_flags & LOCKLEAF) == 0)
1204 * Because of shared lookup we may have the vnode shared locked, but
1205 * the caller may want it to be exclusively locked.
1207 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1208 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1209 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1210 if (VN_IS_DOOMED(dp)) {
1218 if (ni_dvp_unlocked != 2) {
1219 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1232 * relookup - lookup a path name component
1233 * Used by lookup to re-acquire things.
1236 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
1238 struct vnode *dp = NULL; /* the directory we are searching */
1239 int wantparent; /* 1 => wantparent or lockparent flag */
1240 int rdonly; /* lookup read-only flag bit */
1243 KASSERT(cnp->cn_flags & ISLASTCN,
1244 ("relookup: Not given last component."));
1246 * Setup: break out flag bits into variables.
1248 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
1249 KASSERT(wantparent, ("relookup: parent not wanted."));
1250 rdonly = cnp->cn_flags & RDONLY;
1251 cnp->cn_flags &= ~ISSYMLINK;
1253 cnp->cn_lkflags = LK_EXCLUSIVE;
1254 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1257 * Search a new directory.
1259 * The last component of the filename is left accessible via
1260 * cnp->cn_nameptr for callers that need the name. Callers needing
1261 * the name set the SAVENAME flag. When done, they assume
1262 * responsibility for freeing the pathname buffer.
1264 #ifdef NAMEI_DIAGNOSTIC
1265 printf("{%s}: ", cnp->cn_nameptr);
1269 * Check for "" which represents the root directory after slash
1272 if (cnp->cn_nameptr[0] == '\0') {
1274 * Support only LOOKUP for "/" because lookup()
1275 * can't succeed for CREATE, DELETE and RENAME.
1277 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1278 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1280 if (!(cnp->cn_flags & LOCKLEAF))
1283 /* XXX This should probably move to the top of function. */
1284 if (cnp->cn_flags & SAVESTART)
1285 panic("lookup: SAVESTART");
1289 if (cnp->cn_flags & ISDOTDOT)
1290 panic ("relookup: lookup on dot-dot");
1293 * We now have a segment name to search for, and a directory to search.
1295 #ifdef NAMEI_DIAGNOSTIC
1296 vn_printf(dp, "search in ");
1298 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1299 KASSERT(*vpp == NULL, ("leaf should be empty"));
1300 if (error != EJUSTRETURN)
1303 * If creating and at end of pathname, then can consider
1304 * allowing file to be created.
1310 /* ASSERT(dvp == ndp->ni_startdir) */
1311 if (cnp->cn_flags & SAVESTART)
1313 if ((cnp->cn_flags & LOCKPARENT) == 0)
1316 * We return with ni_vp NULL to indicate that the entry
1317 * doesn't currently exist, leaving a pointer to the
1318 * (possibly locked) directory vnode in ndp->ni_dvp.
1326 * Disallow directory write attempts on read-only filesystems.
1329 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1338 * Set the parent lock/ref state to the requested state.
1340 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1345 } else if (!wantparent)
1348 * Check for symbolic link
1350 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1351 ("relookup: symlink found.\n"));
1353 /* ASSERT(dvp == ndp->ni_startdir) */
1354 if (cnp->cn_flags & SAVESTART)
1357 if ((cnp->cn_flags & LOCKLEAF) == 0)
1367 NDINIT_ALL(struct nameidata *ndp, u_long op, u_long flags, enum uio_seg segflg,
1368 const char *namep, int dirfd, struct vnode *startdir, cap_rights_t *rightsp,
1372 ndp->ni_cnd.cn_nameiop = op;
1373 ndp->ni_cnd.cn_flags = flags;
1374 ndp->ni_segflg = segflg;
1375 ndp->ni_dirp = namep;
1376 ndp->ni_dirfd = dirfd;
1377 ndp->ni_startdir = startdir;
1378 ndp->ni_resflags = 0;
1379 filecaps_init(&ndp->ni_filecaps);
1380 ndp->ni_cnd.cn_thread = td;
1381 if (rightsp != NULL)
1382 ndp->ni_rightsneeded = *rightsp;
1384 cap_rights_init_zero(&ndp->ni_rightsneeded);
1388 * Free data allocated by namei(); see namei(9) for details.
1391 NDFREE(struct nameidata *ndp, const u_int flags)
1399 if (!(flags & NDF_NO_FREE_PNBUF) &&
1400 (ndp->ni_cnd.cn_flags & HASBUF)) {
1401 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1402 ndp->ni_cnd.cn_flags &= ~HASBUF;
1404 if (!(flags & NDF_NO_VP_UNLOCK) &&
1405 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1407 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1408 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1409 ndp->ni_dvp != ndp->ni_vp)
1411 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1420 VOP_UNLOCK(ndp->ni_vp);
1421 if (!(flags & NDF_NO_DVP_RELE) &&
1422 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1431 VOP_UNLOCK(ndp->ni_dvp);
1432 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1433 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1434 vrele(ndp->ni_startdir);
1435 ndp->ni_startdir = NULL;
1440 * Determine if there is a suitable alternate filename under the specified
1441 * prefix for the specified path. If the create flag is set, then the
1442 * alternate prefix will be used so long as the parent directory exists.
1443 * This is used by the various compatibility ABIs so that Linux binaries prefer
1444 * files under /compat/linux for example. The chosen path (whether under
1445 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1446 * to by pathbuf. The caller is responsible for free'ing the buffer from
1447 * the M_TEMP bucket if one is returned.
1450 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1451 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1453 struct nameidata nd, ndroot;
1454 char *ptr, *buf, *cp;
1458 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1461 /* Copy the prefix into the new pathname as a starting point. */
1462 len = strlcpy(buf, prefix, MAXPATHLEN);
1463 if (len >= MAXPATHLEN) {
1468 sz = MAXPATHLEN - len;
1471 /* Append the filename to the prefix. */
1472 if (pathseg == UIO_SYSSPACE)
1473 error = copystr(path, ptr, sz, &len);
1475 error = copyinstr(path, ptr, sz, &len);
1483 /* Only use a prefix with absolute pathnames. */
1489 if (dirfd != AT_FDCWD) {
1491 * We want the original because the "prefix" is
1492 * included in the already opened dirfd.
1494 bcopy(ptr, buf, len);
1499 * We know that there is a / somewhere in this pathname.
1500 * Search backwards for it, to find the file's parent dir
1501 * to see if it exists in the alternate tree. If it does,
1502 * and we want to create a file (cflag is set). We don't
1503 * need to worry about the root comparison in this case.
1507 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1510 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1516 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1523 * We now compare the vnode of the prefix to the one
1524 * vnode asked. If they resolve to be the same, then we
1525 * ignore the match so that the real root gets used.
1526 * This avoids the problem of traversing "../.." to find the
1527 * root directory and never finding it, because "/" resolves
1528 * to the emulation root directory. This is expensive :-(
1530 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1533 /* We shouldn't ever get an error from this namei(). */
1534 error = namei(&ndroot);
1536 if (nd.ni_vp == ndroot.ni_vp)
1539 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1540 vrele(ndroot.ni_vp);
1544 NDFREE(&nd, NDF_ONLY_PNBUF);
1548 /* If there was an error, use the original path name. */
1550 bcopy(ptr, buf, len);