2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)vfs_lookup.c 8.4 (Berkeley) 2/16/94
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include "opt_capsicum.h"
41 #include "opt_ktrace.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/capsicum.h>
47 #include <sys/fcntl.h>
50 #include <sys/mutex.h>
51 #include <sys/namei.h>
52 #include <sys/vnode.h>
53 #include <sys/mount.h>
54 #include <sys/filedesc.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysctl.h>
60 #include <sys/ktrace.h>
63 #include <security/audit/audit.h>
64 #include <security/mac/mac_framework.h>
68 #define NAMEI_DIAGNOSTIC 1
69 #undef NAMEI_DIAGNOSTIC
71 SDT_PROVIDER_DECLARE(vfs);
72 SDT_PROBE_DEFINE3(vfs, namei, lookup, entry, "struct vnode *", "char *",
74 SDT_PROBE_DEFINE2(vfs, namei, lookup, return, "int", "struct vnode *");
76 /* Allocation zone for namei. */
77 uma_zone_t namei_zone;
79 /* Placeholder vnode for mp traversal. */
80 static struct vnode *vp_crossmp;
83 crossmp_vop_islocked(struct vop_islocked_args *ap)
90 crossmp_vop_lock1(struct vop_lock1_args *ap)
103 if ((flags & LK_SHARED) == 0)
104 panic("invalid lock request for crossmp");
106 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER, file, line,
107 flags & LK_INTERLOCK ? &VI_MTX(vp)->lock_object : NULL);
108 WITNESS_LOCK(&lk->lock_object, 0, file, line);
109 if ((flags & LK_INTERLOCK) != 0)
111 LOCK_LOG_LOCK("SLOCK", &lk->lock_object, 0, 0, ap->a_file, line);
116 crossmp_vop_unlock(struct vop_unlock_args *ap)
126 if ((flags & LK_INTERLOCK) != 0)
128 WITNESS_UNLOCK(&lk->lock_object, 0, LOCK_FILE, LOCK_LINE);
129 LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, LOCK_FILE,
134 static struct vop_vector crossmp_vnodeops = {
135 .vop_default = &default_vnodeops,
136 .vop_islocked = crossmp_vop_islocked,
137 .vop_lock1 = crossmp_vop_lock1,
138 .vop_unlock = crossmp_vop_unlock,
141 struct nameicap_tracker {
143 TAILQ_ENTRY(nameicap_tracker) nm_link;
146 /* Zone for cap mode tracker elements used for dotdot capability checks. */
147 static uma_zone_t nt_zone;
150 nameiinit(void *dummy __unused)
153 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
155 nt_zone = uma_zcreate("rentr", sizeof(struct nameicap_tracker),
156 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
157 getnewvnode("crossmp", NULL, &crossmp_vnodeops, &vp_crossmp);
159 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
161 static int lookup_shared = 1;
162 SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RWTUN, &lookup_shared, 0,
163 "enables shared locks for path name translation");
165 static int lookup_cap_dotdot = 1;
166 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot, CTLFLAG_RWTUN,
167 &lookup_cap_dotdot, 0,
168 "enables \"..\" components in path lookup in capability mode");
169 static int lookup_cap_dotdot_nonlocal = 1;
170 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot_nonlocal, CTLFLAG_RWTUN,
171 &lookup_cap_dotdot_nonlocal, 0,
172 "enables \"..\" components in path lookup in capability mode "
173 "on non-local mount");
176 nameicap_tracker_add(struct nameidata *ndp, struct vnode *dp)
178 struct nameicap_tracker *nt;
180 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp->v_type != VDIR)
182 nt = uma_zalloc(nt_zone, M_WAITOK);
185 TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link);
189 nameicap_cleanup(struct nameidata *ndp)
191 struct nameicap_tracker *nt, *nt1;
193 KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) ||
194 (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative"));
195 TAILQ_FOREACH_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) {
196 TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link);
198 uma_zfree(nt_zone, nt);
203 * For dotdot lookups in capability mode, only allow the component
204 * lookup to succeed if the resulting directory was already traversed
205 * during the operation. Also fail dotdot lookups for non-local
206 * filesystems, where external agents might assist local lookups to
207 * escape the compartment.
210 nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp)
212 struct nameicap_tracker *nt;
215 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp == NULL ||
219 if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL &&
220 (mp->mnt_flag & MNT_LOCAL) == 0)
221 return (ENOTCAPABLE);
222 TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head,
227 return (ENOTCAPABLE);
231 namei_cleanup_cnp(struct componentname *cnp)
234 uma_zfree(namei_zone, cnp->cn_pnbuf);
236 cnp->cn_pnbuf = NULL;
237 cnp->cn_nameptr = NULL;
242 namei_handle_root(struct nameidata *ndp, struct vnode **dpp)
244 struct componentname *cnp;
247 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0) {
249 if (KTRPOINT(curthread, KTR_CAPFAIL))
250 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
252 return (ENOTCAPABLE);
254 while (*(cnp->cn_nameptr) == '/') {
258 *dpp = ndp->ni_rootdir;
264 * Convert a pathname into a pointer to a locked vnode.
266 * The FOLLOW flag is set when symbolic links are to be followed
267 * when they occur at the end of the name translation process.
268 * Symbolic links are always followed for all other pathname
269 * components other than the last.
271 * The segflg defines whether the name is to be copied from user
272 * space or kernel space.
274 * Overall outline of namei:
277 * get starting directory
278 * while (!done && !error) {
279 * call lookup to search path.
280 * if symbolic link, massage name in buffer and continue
284 namei(struct nameidata *ndp)
286 struct filedesc *fdp; /* pointer to file descriptor state */
287 char *cp; /* pointer into pathname argument */
288 struct vnode *dp; /* the directory we are searching */
289 struct iovec aiov; /* uio for reading symbolic links */
290 struct componentname *cnp;
295 int error, linklen, startdir_used;
300 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
301 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
302 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
303 ("namei: nameiop contaminated with flags"));
304 KASSERT((cnp->cn_flags & OPMASK) == 0,
305 ("namei: flags contaminated with nameiops"));
306 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
307 ndp->ni_startdir->v_type == VBAD);
309 cnp->cn_flags &= ~LOCKSHARED;
311 TAILQ_INIT(&ndp->ni_cap_tracker);
314 /* We will set this ourselves if we need it. */
315 cnp->cn_flags &= ~TRAILINGSLASH;
318 * Get a buffer for the name to be translated, and copy the
319 * name into the buffer.
321 if ((cnp->cn_flags & HASBUF) == 0)
322 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
323 if (ndp->ni_segflg == UIO_SYSSPACE)
324 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
327 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
331 * Don't allow empty pathnames.
333 if (error == 0 && *cnp->cn_pnbuf == '\0')
336 #ifdef CAPABILITY_MODE
338 * In capability mode, lookups must be restricted to happen in
339 * the subtree with the root specified by the file descriptor:
340 * - The root must be real file descriptor, not the pseudo-descriptor
342 * - The passed path must be relative and not absolute.
343 * - If lookup_cap_dotdot is disabled, path must not contain the
345 * - If lookup_cap_dotdot is enabled, we verify that all '..'
346 * components lookups result in the directories which were
347 * previously walked by us, which prevents an escape from
350 if (error == 0 && IN_CAPABILITY_MODE(td) &&
351 (cnp->cn_flags & NOCAPCHECK) == 0) {
352 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
353 if (ndp->ni_dirfd == AT_FDCWD) {
355 if (KTRPOINT(td, KTR_CAPFAIL))
356 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
363 namei_cleanup_cnp(cnp);
369 if (KTRPOINT(td, KTR_NAMEI)) {
370 KASSERT(cnp->cn_thread == curthread,
371 ("namei not using curthread"));
372 ktrnamei(cnp->cn_pnbuf);
376 * Get starting point for the translation.
379 ndp->ni_rootdir = fdp->fd_rdir;
380 vrefact(ndp->ni_rootdir);
381 ndp->ni_topdir = fdp->fd_jdir;
384 * If we are auditing the kernel pathname, save the user pathname.
386 if (cnp->cn_flags & AUDITVNODE1)
387 AUDIT_ARG_UPATH1(td, ndp->ni_dirfd, cnp->cn_pnbuf);
388 if (cnp->cn_flags & AUDITVNODE2)
389 AUDIT_ARG_UPATH2(td, ndp->ni_dirfd, cnp->cn_pnbuf);
393 cnp->cn_nameptr = cnp->cn_pnbuf;
394 if (cnp->cn_pnbuf[0] == '/') {
395 error = namei_handle_root(ndp, &dp);
397 if (ndp->ni_startdir != NULL) {
398 dp = ndp->ni_startdir;
400 } else if (ndp->ni_dirfd == AT_FDCWD) {
404 rights = ndp->ni_rightsneeded;
405 cap_rights_set(&rights, CAP_LOOKUP);
407 if (cnp->cn_flags & AUDITVNODE1)
408 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
409 if (cnp->cn_flags & AUDITVNODE2)
410 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
411 error = fgetvp_rights(td, ndp->ni_dirfd,
412 &rights, &ndp->ni_filecaps, &dp);
417 * If file descriptor doesn't have all rights,
418 * all lookups relative to it must also be
422 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights,
424 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
425 ndp->ni_filecaps.fc_nioctls != -1) {
426 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
430 if (error == 0 && dp->v_type != VDIR)
433 FILEDESC_SUNLOCK(fdp);
434 if (ndp->ni_startdir != NULL && !startdir_used)
435 vrele(ndp->ni_startdir);
441 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 &&
442 lookup_cap_dotdot != 0)
443 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
444 SDT_PROBE3(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
447 ndp->ni_startdir = dp;
452 * If not a symbolic link, we're done.
454 if ((cnp->cn_flags & ISSYMLINK) == 0) {
455 vrele(ndp->ni_rootdir);
456 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
457 namei_cleanup_cnp(cnp);
459 cnp->cn_flags |= HASBUF;
460 nameicap_cleanup(ndp);
461 SDT_PROBE2(vfs, namei, lookup, return, 0, ndp->ni_vp);
464 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
469 if ((cnp->cn_flags & NOMACCHECK) == 0) {
470 error = mac_vnode_check_readlink(td->td_ucred,
476 if (ndp->ni_pathlen > 1)
477 cp = uma_zalloc(namei_zone, M_WAITOK);
481 aiov.iov_len = MAXPATHLEN;
482 auio.uio_iov = &aiov;
485 auio.uio_rw = UIO_READ;
486 auio.uio_segflg = UIO_SYSSPACE;
488 auio.uio_resid = MAXPATHLEN;
489 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
491 if (ndp->ni_pathlen > 1)
492 uma_zfree(namei_zone, cp);
495 linklen = MAXPATHLEN - auio.uio_resid;
497 if (ndp->ni_pathlen > 1)
498 uma_zfree(namei_zone, cp);
502 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
503 if (ndp->ni_pathlen > 1)
504 uma_zfree(namei_zone, cp);
505 error = ENAMETOOLONG;
508 if (ndp->ni_pathlen > 1) {
509 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
510 uma_zfree(namei_zone, cnp->cn_pnbuf);
513 cnp->cn_pnbuf[linklen] = '\0';
514 ndp->ni_pathlen += linklen;
518 * Check if root directory should replace current directory.
520 cnp->cn_nameptr = cnp->cn_pnbuf;
521 if (*(cnp->cn_nameptr) == '/') {
523 error = namei_handle_root(ndp, &dp);
532 vrele(ndp->ni_rootdir);
533 namei_cleanup_cnp(cnp);
534 nameicap_cleanup(ndp);
535 SDT_PROBE2(vfs, namei, lookup, return, error, NULL);
540 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
543 if (mp == NULL || ((lkflags & LK_SHARED) &&
544 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
545 ((cnflags & ISDOTDOT) &&
546 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
547 lkflags &= ~LK_SHARED;
548 lkflags |= LK_EXCLUSIVE;
550 lkflags |= LK_NODDLKTREAT;
555 needs_exclusive_leaf(struct mount *mp, int flags)
559 * Intermediate nodes can use shared locks, we only need to
560 * force an exclusive lock for leaf nodes.
562 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
565 /* Always use exclusive locks if LOCKSHARED isn't set. */
566 if (!(flags & LOCKSHARED))
570 * For lookups during open(), if the mount point supports
571 * extended shared operations, then use a shared lock for the
572 * leaf node, otherwise use an exclusive lock.
574 if ((flags & ISOPEN) != 0)
575 return (!MNT_EXTENDED_SHARED(mp));
578 * Lookup requests outside of open() that specify LOCKSHARED
579 * only need a shared lock on the leaf vnode.
586 * This is a very central and rather complicated routine.
588 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
589 * The starting directory is taken from ni_startdir. The pathname is
590 * descended until done, or a symbolic link is encountered. The variable
591 * ni_more is clear if the path is completed; it is set to one if a
592 * symbolic link needing interpretation is encountered.
594 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
595 * whether the name is to be looked up, created, renamed, or deleted.
596 * When CREATE, RENAME, or DELETE is specified, information usable in
597 * creating, renaming, or deleting a directory entry may be calculated.
598 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
599 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
600 * returned unlocked. Otherwise the parent directory is not returned. If
601 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
602 * the target is returned locked, otherwise it is returned unlocked.
603 * When creating or renaming and LOCKPARENT is specified, the target may not
604 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
606 * Overall outline of lookup:
609 * identify next component of name at ndp->ni_ptr
610 * handle degenerate case where name is null string
611 * if .. and crossing mount points and on mounted filesys, find parent
612 * call VOP_LOOKUP routine for next component name
613 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
614 * component vnode returned in ni_vp (if it exists), locked.
615 * if result vnode is mounted on and crossing mount points,
616 * find mounted on vnode
617 * if more components of name, do next level at dirloop
618 * return the answer in ni_vp, locked if LOCKLEAF set
619 * if LOCKPARENT set, return locked parent in ni_dvp
620 * if WANTPARENT set, return unlocked parent in ni_dvp
623 lookup(struct nameidata *ndp)
625 char *cp; /* pointer into pathname argument */
626 char *prev_ni_next; /* saved ndp->ni_next */
627 struct vnode *dp = NULL; /* the directory we are searching */
628 struct vnode *tdp; /* saved dp */
629 struct mount *mp; /* mount table entry */
631 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
632 int docache; /* == 0 do not cache last component */
633 int wantparent; /* 1 => wantparent or lockparent flag */
634 int rdonly; /* lookup read-only flag bit */
636 int dpunlocked = 0; /* dp has already been unlocked */
637 int relookup = 0; /* do not consume the path component */
638 struct componentname *cnp = &ndp->ni_cnd;
643 * Setup: break out flag bits into variables.
646 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
647 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
648 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
649 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
650 if (cnp->cn_nameiop == DELETE ||
651 (wantparent && cnp->cn_nameiop != CREATE &&
652 cnp->cn_nameiop != LOOKUP))
654 rdonly = cnp->cn_flags & RDONLY;
655 cnp->cn_flags &= ~ISSYMLINK;
658 * We use shared locks until we hit the parent of the last cn then
659 * we adjust based on the requesting flags.
662 cnp->cn_lkflags = LK_SHARED;
664 cnp->cn_lkflags = LK_EXCLUSIVE;
665 dp = ndp->ni_startdir;
666 ndp->ni_startdir = NULLVP;
668 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
673 * Search a new directory.
675 * The last component of the filename is left accessible via
676 * cnp->cn_nameptr for callers that need the name. Callers needing
677 * the name set the SAVENAME flag. When done, they assume
678 * responsibility for freeing the pathname buffer.
680 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
682 cnp->cn_namelen = cp - cnp->cn_nameptr;
683 if (cnp->cn_namelen > NAME_MAX) {
684 error = ENAMETOOLONG;
687 #ifdef NAMEI_DIAGNOSTIC
690 printf("{%s}: ", cnp->cn_nameptr);
693 prev_ni_pathlen = ndp->ni_pathlen;
694 ndp->ni_pathlen -= cnp->cn_namelen;
695 KASSERT(ndp->ni_pathlen <= PATH_MAX,
696 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
697 prev_ni_next = ndp->ni_next;
701 * Replace multiple slashes by a single slash and trailing slashes
702 * by a null. This must be done before VOP_LOOKUP() because some
703 * fs's don't know about trailing slashes. Remember if there were
704 * trailing slashes to handle symlinks, existing non-directories
705 * and non-existing files that won't be directories specially later.
707 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
711 *ndp->ni_next = '\0';
712 cnp->cn_flags |= TRAILINGSLASH;
717 cnp->cn_flags |= MAKEENTRY;
718 if (*cp == '\0' && docache == 0)
719 cnp->cn_flags &= ~MAKEENTRY;
720 if (cnp->cn_namelen == 2 &&
721 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
722 cnp->cn_flags |= ISDOTDOT;
724 cnp->cn_flags &= ~ISDOTDOT;
725 if (*ndp->ni_next == 0)
726 cnp->cn_flags |= ISLASTCN;
728 cnp->cn_flags &= ~ISLASTCN;
730 if ((cnp->cn_flags & ISLASTCN) != 0 &&
731 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
732 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
737 nameicap_tracker_add(ndp, dp);
740 * Check for degenerate name (e.g. / or "")
741 * which is a way of talking about a directory,
742 * e.g. like "/." or ".".
744 if (cnp->cn_nameptr[0] == '\0') {
745 if (dp->v_type != VDIR) {
749 if (cnp->cn_nameiop != LOOKUP) {
759 if (cnp->cn_flags & AUDITVNODE1)
760 AUDIT_ARG_VNODE1(dp);
761 else if (cnp->cn_flags & AUDITVNODE2)
762 AUDIT_ARG_VNODE2(dp);
764 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
766 /* XXX This should probably move to the top of function. */
767 if (cnp->cn_flags & SAVESTART)
768 panic("lookup: SAVESTART");
773 * Handle "..": five special cases.
774 * 0. If doing a capability lookup and lookup_cap_dotdot is
775 * disabled, return ENOTCAPABLE.
776 * 1. Return an error if this is the last component of
777 * the name and the operation is DELETE or RENAME.
778 * 2. If at root directory (e.g. after chroot)
779 * or at absolute root directory
780 * then ignore it so can't get out.
781 * 3. If this vnode is the root of a mounted
782 * filesystem, then replace it with the
783 * vnode which was mounted on so we take the
784 * .. in the other filesystem.
785 * 4. If the vnode is the top directory of
786 * the jail or chroot, don't let them out.
787 * 5. If doing a capability lookup and lookup_cap_dotdot is
788 * enabled, return ENOTCAPABLE if the lookup would escape
789 * from the initial file descriptor directory. Checks are
790 * done by ensuring that namei() already traversed the
791 * result of dotdot lookup.
793 if (cnp->cn_flags & ISDOTDOT) {
794 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT))
795 == NI_LCF_STRICTRELATIVE) {
797 if (KTRPOINT(curthread, KTR_CAPFAIL))
798 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
803 if ((cnp->cn_flags & ISLASTCN) != 0 &&
804 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
809 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
811 if (dp == pr->pr_root)
813 if (dp == ndp->ni_rootdir ||
814 dp == ndp->ni_topdir ||
817 ((dp->v_vflag & VV_ROOT) != 0 &&
818 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
824 if ((dp->v_vflag & VV_ROOT) == 0)
826 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
831 dp = dp->v_mount->mnt_vnodecovered;
835 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
836 LK_RETRY, ISDOTDOT));
837 error = nameicap_check_dotdot(ndp, dp);
840 if (KTRPOINT(curthread, KTR_CAPFAIL))
841 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
849 * We now have a segment name to search for, and a directory to search.
853 if ((cnp->cn_flags & NOMACCHECK) == 0) {
854 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
862 ASSERT_VOP_LOCKED(dp, "lookup");
864 * If we have a shared lock we may need to upgrade the lock for the
867 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
868 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
869 vn_lock(dp, LK_UPGRADE|LK_RETRY);
870 if ((dp->v_iflag & VI_DOOMED) != 0) {
875 * If we're looking up the last component and we need an exclusive
876 * lock, adjust our lkflags.
878 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
879 cnp->cn_lkflags = LK_EXCLUSIVE;
880 #ifdef NAMEI_DIAGNOSTIC
881 vn_printf(dp, "lookup in ");
883 lkflags_save = cnp->cn_lkflags;
884 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
886 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
887 cnp->cn_lkflags = lkflags_save;
889 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
890 #ifdef NAMEI_DIAGNOSTIC
891 printf("not found\n");
893 if ((error == ENOENT) &&
894 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
895 (dp->v_mount->mnt_flag & MNT_UNION)) {
897 dp = dp->v_mount->mnt_vnodecovered;
901 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
902 LK_RETRY, cnp->cn_flags));
903 nameicap_tracker_add(ndp, dp);
907 if (error == ERELOOKUP) {
915 if (error != EJUSTRETURN)
918 * At this point, we know we're at the end of the
919 * pathname. If creating / renaming, we can consider
920 * allowing the file or directory to be created / renamed,
921 * provided we're not on a read-only filesystem.
927 /* trailing slash only allowed for directories */
928 if ((cnp->cn_flags & TRAILINGSLASH) &&
929 !(cnp->cn_flags & WILLBEDIR)) {
933 if ((cnp->cn_flags & LOCKPARENT) == 0)
936 * We return with ni_vp NULL to indicate that the entry
937 * doesn't currently exist, leaving a pointer to the
938 * (possibly locked) directory vnode in ndp->ni_dvp.
940 if (cnp->cn_flags & SAVESTART) {
941 ndp->ni_startdir = ndp->ni_dvp;
942 VREF(ndp->ni_startdir);
948 #ifdef NAMEI_DIAGNOSTIC
954 * Check to see if the vnode has been mounted on;
955 * if so find the root of the mounted filesystem.
957 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
958 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
962 if (dp != ndp->ni_dvp)
967 ndp->ni_dvp = vp_crossmp;
968 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
969 cnp->cn_flags), &tdp);
971 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
972 panic("vp_crossmp exclusively locked or reclaimed");
977 ndp->ni_vp = dp = tdp;
981 * Check for symbolic link
983 if ((dp->v_type == VLNK) &&
984 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
985 *ndp->ni_next == '/')) {
986 cnp->cn_flags |= ISSYMLINK;
987 if (dp->v_iflag & VI_DOOMED) {
989 * We can't know whether the directory was mounted with
990 * NOSYMFOLLOW, so we can't follow safely.
995 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1000 * Symlink code always expects an unlocked dvp.
1002 if (ndp->ni_dvp != ndp->ni_vp) {
1003 VOP_UNLOCK(ndp->ni_dvp, 0);
1004 ni_dvp_unlocked = 1;
1011 * Not a symbolic link that we will follow. Continue with the
1012 * next component if there is any; otherwise, we're done.
1014 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1015 ("lookup: invalid path state."));
1018 ndp->ni_pathlen = prev_ni_pathlen;
1019 ndp->ni_next = prev_ni_next;
1020 if (ndp->ni_dvp != dp)
1026 if (cnp->cn_flags & ISDOTDOT) {
1027 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1030 if (KTRPOINT(curthread, KTR_CAPFAIL))
1031 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1036 if (*ndp->ni_next == '/') {
1037 cnp->cn_nameptr = ndp->ni_next;
1038 while (*cnp->cn_nameptr == '/') {
1042 if (ndp->ni_dvp != dp)
1049 * If we're processing a path with a trailing slash,
1050 * check that the end result is a directory.
1052 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1057 * Disallow directory write attempts on read-only filesystems.
1060 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1064 if (cnp->cn_flags & SAVESTART) {
1065 ndp->ni_startdir = ndp->ni_dvp;
1066 VREF(ndp->ni_startdir);
1069 ni_dvp_unlocked = 2;
1070 if (ndp->ni_dvp != dp)
1074 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1075 VOP_UNLOCK(ndp->ni_dvp, 0);
1076 ni_dvp_unlocked = 1;
1079 if (cnp->cn_flags & AUDITVNODE1)
1080 AUDIT_ARG_VNODE1(dp);
1081 else if (cnp->cn_flags & AUDITVNODE2)
1082 AUDIT_ARG_VNODE2(dp);
1084 if ((cnp->cn_flags & LOCKLEAF) == 0)
1088 * Because of lookup_shared we may have the vnode shared locked, but
1089 * the caller may want it to be exclusively locked.
1091 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1092 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1093 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1094 if (dp->v_iflag & VI_DOOMED) {
1102 if (ni_dvp_unlocked != 2) {
1103 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1116 * relookup - lookup a path name component
1117 * Used by lookup to re-acquire things.
1120 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
1122 struct vnode *dp = NULL; /* the directory we are searching */
1123 int wantparent; /* 1 => wantparent or lockparent flag */
1124 int rdonly; /* lookup read-only flag bit */
1127 KASSERT(cnp->cn_flags & ISLASTCN,
1128 ("relookup: Not given last component."));
1130 * Setup: break out flag bits into variables.
1132 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
1133 KASSERT(wantparent, ("relookup: parent not wanted."));
1134 rdonly = cnp->cn_flags & RDONLY;
1135 cnp->cn_flags &= ~ISSYMLINK;
1137 cnp->cn_lkflags = LK_EXCLUSIVE;
1138 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1141 * Search a new directory.
1143 * The last component of the filename is left accessible via
1144 * cnp->cn_nameptr for callers that need the name. Callers needing
1145 * the name set the SAVENAME flag. When done, they assume
1146 * responsibility for freeing the pathname buffer.
1148 #ifdef NAMEI_DIAGNOSTIC
1149 printf("{%s}: ", cnp->cn_nameptr);
1153 * Check for "" which represents the root directory after slash
1156 if (cnp->cn_nameptr[0] == '\0') {
1158 * Support only LOOKUP for "/" because lookup()
1159 * can't succeed for CREATE, DELETE and RENAME.
1161 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1162 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1164 if (!(cnp->cn_flags & LOCKLEAF))
1167 /* XXX This should probably move to the top of function. */
1168 if (cnp->cn_flags & SAVESTART)
1169 panic("lookup: SAVESTART");
1173 if (cnp->cn_flags & ISDOTDOT)
1174 panic ("relookup: lookup on dot-dot");
1177 * We now have a segment name to search for, and a directory to search.
1179 #ifdef NAMEI_DIAGNOSTIC
1180 vn_printf(dp, "search in ");
1182 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1183 KASSERT(*vpp == NULL, ("leaf should be empty"));
1184 if (error != EJUSTRETURN)
1187 * If creating and at end of pathname, then can consider
1188 * allowing file to be created.
1194 /* ASSERT(dvp == ndp->ni_startdir) */
1195 if (cnp->cn_flags & SAVESTART)
1197 if ((cnp->cn_flags & LOCKPARENT) == 0)
1200 * We return with ni_vp NULL to indicate that the entry
1201 * doesn't currently exist, leaving a pointer to the
1202 * (possibly locked) directory vnode in ndp->ni_dvp.
1210 * Disallow directory write attempts on read-only filesystems.
1213 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1222 * Set the parent lock/ref state to the requested state.
1224 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1229 } else if (!wantparent)
1232 * Check for symbolic link
1234 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1235 ("relookup: symlink found.\n"));
1237 /* ASSERT(dvp == ndp->ni_startdir) */
1238 if (cnp->cn_flags & SAVESTART)
1241 if ((cnp->cn_flags & LOCKLEAF) == 0)
1251 NDINIT_ALL(struct nameidata *ndp, u_long op, u_long flags, enum uio_seg segflg,
1252 const char *namep, int dirfd, struct vnode *startdir, cap_rights_t *rightsp,
1256 ndp->ni_cnd.cn_nameiop = op;
1257 ndp->ni_cnd.cn_flags = flags;
1258 ndp->ni_segflg = segflg;
1259 ndp->ni_dirp = namep;
1260 ndp->ni_dirfd = dirfd;
1261 ndp->ni_startdir = startdir;
1262 if (rightsp != NULL)
1263 ndp->ni_rightsneeded = *rightsp;
1265 cap_rights_init(&ndp->ni_rightsneeded);
1266 filecaps_init(&ndp->ni_filecaps);
1267 ndp->ni_cnd.cn_thread = td;
1271 * Free data allocated by namei(); see namei(9) for details.
1274 NDFREE(struct nameidata *ndp, const u_int flags)
1282 if (!(flags & NDF_NO_FREE_PNBUF) &&
1283 (ndp->ni_cnd.cn_flags & HASBUF)) {
1284 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1285 ndp->ni_cnd.cn_flags &= ~HASBUF;
1287 if (!(flags & NDF_NO_VP_UNLOCK) &&
1288 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1290 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1299 VOP_UNLOCK(ndp->ni_vp, 0);
1300 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1301 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1302 ndp->ni_dvp != ndp->ni_vp)
1304 if (!(flags & NDF_NO_DVP_RELE) &&
1305 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1314 VOP_UNLOCK(ndp->ni_dvp, 0);
1315 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1316 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1317 vrele(ndp->ni_startdir);
1318 ndp->ni_startdir = NULL;
1323 * Determine if there is a suitable alternate filename under the specified
1324 * prefix for the specified path. If the create flag is set, then the
1325 * alternate prefix will be used so long as the parent directory exists.
1326 * This is used by the various compatibility ABIs so that Linux binaries prefer
1327 * files under /compat/linux for example. The chosen path (whether under
1328 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1329 * to by pathbuf. The caller is responsible for free'ing the buffer from
1330 * the M_TEMP bucket if one is returned.
1333 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1334 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1336 struct nameidata nd, ndroot;
1337 char *ptr, *buf, *cp;
1341 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1344 /* Copy the prefix into the new pathname as a starting point. */
1345 len = strlcpy(buf, prefix, MAXPATHLEN);
1346 if (len >= MAXPATHLEN) {
1351 sz = MAXPATHLEN - len;
1354 /* Append the filename to the prefix. */
1355 if (pathseg == UIO_SYSSPACE)
1356 error = copystr(path, ptr, sz, &len);
1358 error = copyinstr(path, ptr, sz, &len);
1366 /* Only use a prefix with absolute pathnames. */
1372 if (dirfd != AT_FDCWD) {
1374 * We want the original because the "prefix" is
1375 * included in the already opened dirfd.
1377 bcopy(ptr, buf, len);
1382 * We know that there is a / somewhere in this pathname.
1383 * Search backwards for it, to find the file's parent dir
1384 * to see if it exists in the alternate tree. If it does,
1385 * and we want to create a file (cflag is set). We don't
1386 * need to worry about the root comparison in this case.
1390 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1393 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1399 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1406 * We now compare the vnode of the prefix to the one
1407 * vnode asked. If they resolve to be the same, then we
1408 * ignore the match so that the real root gets used.
1409 * This avoids the problem of traversing "../.." to find the
1410 * root directory and never finding it, because "/" resolves
1411 * to the emulation root directory. This is expensive :-(
1413 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1416 /* We shouldn't ever get an error from this namei(). */
1417 error = namei(&ndroot);
1419 if (nd.ni_vp == ndroot.ni_vp)
1422 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1423 vrele(ndroot.ni_vp);
1427 NDFREE(&nd, NDF_ONLY_PNBUF);
1431 /* If there was an error, use the original path name. */
1433 bcopy(ptr, buf, len);