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_islocked = crossmp_vop_islocked,
136 .vop_lock1 = crossmp_vop_lock1,
137 .vop_unlock = crossmp_vop_unlock,
140 struct nameicap_tracker {
142 TAILQ_ENTRY(nameicap_tracker) nm_link;
145 /* Zone for cap mode tracker elements used for dotdot capability checks. */
146 static uma_zone_t nt_zone;
149 nameiinit(void *dummy __unused)
152 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
154 nt_zone = uma_zcreate("rentr", sizeof(struct nameicap_tracker),
155 NULL, NULL, NULL, NULL, sizeof(void *), 0);
156 getnewvnode("crossmp", NULL, &crossmp_vnodeops, &vp_crossmp);
158 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
160 static int lookup_shared = 1;
161 SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RWTUN, &lookup_shared, 0,
162 "enables shared locks for path name translation");
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 nt = uma_zalloc(nt_zone, M_WAITOK);
184 TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link);
188 nameicap_cleanup(struct nameidata *ndp)
190 struct nameicap_tracker *nt, *nt1;
192 KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) ||
193 (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative"));
194 TAILQ_FOREACH_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) {
195 TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link);
197 uma_zfree(nt_zone, nt);
202 * For dotdot lookups in capability mode, only allow the component
203 * lookup to succeed if the resulting directory was already traversed
204 * during the operation. Also fail dotdot lookups for non-local
205 * filesystems, where external agents might assist local lookups to
206 * escape the compartment.
209 nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp)
211 struct nameicap_tracker *nt;
214 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp == NULL ||
218 if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL &&
219 (mp->mnt_flag & MNT_LOCAL) == 0)
220 return (ENOTCAPABLE);
221 TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head,
226 return (ENOTCAPABLE);
230 namei_cleanup_cnp(struct componentname *cnp)
233 uma_zfree(namei_zone, cnp->cn_pnbuf);
235 cnp->cn_pnbuf = NULL;
236 cnp->cn_nameptr = NULL;
241 namei_handle_root(struct nameidata *ndp, struct vnode **dpp)
243 struct componentname *cnp;
246 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0) {
248 if (KTRPOINT(curthread, KTR_CAPFAIL))
249 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
251 return (ENOTCAPABLE);
253 while (*(cnp->cn_nameptr) == '/') {
257 *dpp = ndp->ni_rootdir;
263 * Convert a pathname into a pointer to a locked vnode.
265 * The FOLLOW flag is set when symbolic links are to be followed
266 * when they occur at the end of the name translation process.
267 * Symbolic links are always followed for all other pathname
268 * components other than the last.
270 * The segflg defines whether the name is to be copied from user
271 * space or kernel space.
273 * Overall outline of namei:
276 * get starting directory
277 * while (!done && !error) {
278 * call lookup to search path.
279 * if symbolic link, massage name in buffer and continue
283 namei(struct nameidata *ndp)
285 struct filedesc *fdp; /* pointer to file descriptor state */
286 char *cp; /* pointer into pathname argument */
287 struct vnode *dp; /* the directory we are searching */
288 struct iovec aiov; /* uio for reading symbolic links */
289 struct componentname *cnp;
294 int error, linklen, startdir_used;
299 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
300 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
301 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
302 ("namei: nameiop contaminated with flags"));
303 KASSERT((cnp->cn_flags & OPMASK) == 0,
304 ("namei: flags contaminated with nameiops"));
305 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
306 ndp->ni_startdir->v_type == VBAD);
308 cnp->cn_flags &= ~LOCKSHARED;
310 TAILQ_INIT(&ndp->ni_cap_tracker);
313 /* We will set this ourselves if we need it. */
314 cnp->cn_flags &= ~TRAILINGSLASH;
317 * Get a buffer for the name to be translated, and copy the
318 * name into the buffer.
320 if ((cnp->cn_flags & HASBUF) == 0)
321 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
322 if (ndp->ni_segflg == UIO_SYSSPACE)
323 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
326 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
330 * Don't allow empty pathnames.
332 if (error == 0 && *cnp->cn_pnbuf == '\0')
335 #ifdef CAPABILITY_MODE
337 * In capability mode, lookups must be restricted to happen in
338 * the subtree with the root specified by the file descriptor:
339 * - The root must be real file descriptor, not the pseudo-descriptor
341 * - The passed path must be relative and not absolute.
342 * - If lookup_cap_dotdot is disabled, path must not contain the
344 * - If lookup_cap_dotdot is enabled, we verify that all '..'
345 * components lookups result in the directories which were
346 * previously walked by us, which prevents an escape from
349 if (error == 0 && IN_CAPABILITY_MODE(td) &&
350 (cnp->cn_flags & NOCAPCHECK) == 0) {
351 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
352 if (ndp->ni_dirfd == AT_FDCWD) {
354 if (KTRPOINT(td, KTR_CAPFAIL))
355 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
362 namei_cleanup_cnp(cnp);
368 if (KTRPOINT(td, KTR_NAMEI)) {
369 KASSERT(cnp->cn_thread == curthread,
370 ("namei not using curthread"));
371 ktrnamei(cnp->cn_pnbuf);
375 * Get starting point for the translation.
378 ndp->ni_rootdir = fdp->fd_rdir;
379 vrefact(ndp->ni_rootdir);
380 ndp->ni_topdir = fdp->fd_jdir;
383 * If we are auditing the kernel pathname, save the user pathname.
385 if (cnp->cn_flags & AUDITVNODE1)
386 AUDIT_ARG_UPATH1(td, ndp->ni_dirfd, cnp->cn_pnbuf);
387 if (cnp->cn_flags & AUDITVNODE2)
388 AUDIT_ARG_UPATH2(td, ndp->ni_dirfd, cnp->cn_pnbuf);
392 cnp->cn_nameptr = cnp->cn_pnbuf;
393 if (cnp->cn_pnbuf[0] == '/') {
394 error = namei_handle_root(ndp, &dp);
396 if (ndp->ni_startdir != NULL) {
397 dp = ndp->ni_startdir;
399 } else if (ndp->ni_dirfd == AT_FDCWD) {
403 rights = ndp->ni_rightsneeded;
404 cap_rights_set(&rights, CAP_LOOKUP);
406 if (cnp->cn_flags & AUDITVNODE1)
407 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
408 if (cnp->cn_flags & AUDITVNODE2)
409 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
410 error = fgetvp_rights(td, ndp->ni_dirfd,
411 &rights, &ndp->ni_filecaps, &dp);
416 * If file descriptor doesn't have all rights,
417 * all lookups relative to it must also be
421 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights,
423 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
424 ndp->ni_filecaps.fc_nioctls != -1) {
425 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
429 if (error == 0 && dp->v_type != VDIR)
432 FILEDESC_SUNLOCK(fdp);
433 if (ndp->ni_startdir != NULL && !startdir_used)
434 vrele(ndp->ni_startdir);
440 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 &&
441 lookup_cap_dotdot != 0)
442 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
443 SDT_PROBE3(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
446 ndp->ni_startdir = dp;
451 * If not a symbolic link, we're done.
453 if ((cnp->cn_flags & ISSYMLINK) == 0) {
454 vrele(ndp->ni_rootdir);
455 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
456 namei_cleanup_cnp(cnp);
458 cnp->cn_flags |= HASBUF;
459 nameicap_cleanup(ndp);
460 SDT_PROBE2(vfs, namei, lookup, return, 0, ndp->ni_vp);
463 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
468 if ((cnp->cn_flags & NOMACCHECK) == 0) {
469 error = mac_vnode_check_readlink(td->td_ucred,
475 if (ndp->ni_pathlen > 1)
476 cp = uma_zalloc(namei_zone, M_WAITOK);
480 aiov.iov_len = MAXPATHLEN;
481 auio.uio_iov = &aiov;
484 auio.uio_rw = UIO_READ;
485 auio.uio_segflg = UIO_SYSSPACE;
487 auio.uio_resid = MAXPATHLEN;
488 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
490 if (ndp->ni_pathlen > 1)
491 uma_zfree(namei_zone, cp);
494 linklen = MAXPATHLEN - auio.uio_resid;
496 if (ndp->ni_pathlen > 1)
497 uma_zfree(namei_zone, cp);
501 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
502 if (ndp->ni_pathlen > 1)
503 uma_zfree(namei_zone, cp);
504 error = ENAMETOOLONG;
507 if (ndp->ni_pathlen > 1) {
508 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
509 uma_zfree(namei_zone, cnp->cn_pnbuf);
512 cnp->cn_pnbuf[linklen] = '\0';
513 ndp->ni_pathlen += linklen;
517 * Check if root directory should replace current directory.
519 cnp->cn_nameptr = cnp->cn_pnbuf;
520 if (*(cnp->cn_nameptr) == '/') {
522 error = namei_handle_root(ndp, &dp);
531 vrele(ndp->ni_rootdir);
532 namei_cleanup_cnp(cnp);
533 nameicap_cleanup(ndp);
534 SDT_PROBE2(vfs, namei, lookup, return, error, NULL);
539 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
542 if (mp == NULL || ((lkflags & LK_SHARED) &&
543 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
544 ((cnflags & ISDOTDOT) &&
545 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
546 lkflags &= ~LK_SHARED;
547 lkflags |= LK_EXCLUSIVE;
549 lkflags |= LK_NODDLKTREAT;
554 needs_exclusive_leaf(struct mount *mp, int flags)
558 * Intermediate nodes can use shared locks, we only need to
559 * force an exclusive lock for leaf nodes.
561 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
564 /* Always use exclusive locks if LOCKSHARED isn't set. */
565 if (!(flags & LOCKSHARED))
569 * For lookups during open(), if the mount point supports
570 * extended shared operations, then use a shared lock for the
571 * leaf node, otherwise use an exclusive lock.
573 if ((flags & ISOPEN) != 0)
574 return (!MNT_EXTENDED_SHARED(mp));
577 * Lookup requests outside of open() that specify LOCKSHARED
578 * only need a shared lock on the leaf vnode.
585 * This is a very central and rather complicated routine.
587 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
588 * The starting directory is taken from ni_startdir. The pathname is
589 * descended until done, or a symbolic link is encountered. The variable
590 * ni_more is clear if the path is completed; it is set to one if a
591 * symbolic link needing interpretation is encountered.
593 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
594 * whether the name is to be looked up, created, renamed, or deleted.
595 * When CREATE, RENAME, or DELETE is specified, information usable in
596 * creating, renaming, or deleting a directory entry may be calculated.
597 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
598 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
599 * returned unlocked. Otherwise the parent directory is not returned. If
600 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
601 * the target is returned locked, otherwise it is returned unlocked.
602 * When creating or renaming and LOCKPARENT is specified, the target may not
603 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
605 * Overall outline of lookup:
608 * identify next component of name at ndp->ni_ptr
609 * handle degenerate case where name is null string
610 * if .. and crossing mount points and on mounted filesys, find parent
611 * call VOP_LOOKUP routine for next component name
612 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
613 * component vnode returned in ni_vp (if it exists), locked.
614 * if result vnode is mounted on and crossing mount points,
615 * find mounted on vnode
616 * if more components of name, do next level at dirloop
617 * return the answer in ni_vp, locked if LOCKLEAF set
618 * if LOCKPARENT set, return locked parent in ni_dvp
619 * if WANTPARENT set, return unlocked parent in ni_dvp
622 lookup(struct nameidata *ndp)
624 char *cp; /* pointer into pathname argument */
625 char *prev_ni_next; /* saved ndp->ni_next */
626 struct vnode *dp = NULL; /* the directory we are searching */
627 struct vnode *tdp; /* saved dp */
628 struct mount *mp; /* mount table entry */
630 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
631 int docache; /* == 0 do not cache last component */
632 int wantparent; /* 1 => wantparent or lockparent flag */
633 int rdonly; /* lookup read-only flag bit */
635 int dpunlocked = 0; /* dp has already been unlocked */
636 int relookup = 0; /* do not consume the path component */
637 struct componentname *cnp = &ndp->ni_cnd;
642 * Setup: break out flag bits into variables.
645 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
646 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
647 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
648 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
649 if (cnp->cn_nameiop == DELETE ||
650 (wantparent && cnp->cn_nameiop != CREATE &&
651 cnp->cn_nameiop != LOOKUP))
653 rdonly = cnp->cn_flags & RDONLY;
654 cnp->cn_flags &= ~ISSYMLINK;
657 * We use shared locks until we hit the parent of the last cn then
658 * we adjust based on the requesting flags.
661 cnp->cn_lkflags = LK_SHARED;
663 cnp->cn_lkflags = LK_EXCLUSIVE;
664 dp = ndp->ni_startdir;
665 ndp->ni_startdir = NULLVP;
667 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
672 * Search a new directory.
674 * The last component of the filename is left accessible via
675 * cnp->cn_nameptr for callers that need the name. Callers needing
676 * the name set the SAVENAME flag. When done, they assume
677 * responsibility for freeing the pathname buffer.
679 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
681 cnp->cn_namelen = cp - cnp->cn_nameptr;
682 if (cnp->cn_namelen > NAME_MAX) {
683 error = ENAMETOOLONG;
686 #ifdef NAMEI_DIAGNOSTIC
689 printf("{%s}: ", cnp->cn_nameptr);
692 prev_ni_pathlen = ndp->ni_pathlen;
693 ndp->ni_pathlen -= cnp->cn_namelen;
694 KASSERT(ndp->ni_pathlen <= PATH_MAX,
695 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
696 prev_ni_next = ndp->ni_next;
700 * Replace multiple slashes by a single slash and trailing slashes
701 * by a null. This must be done before VOP_LOOKUP() because some
702 * fs's don't know about trailing slashes. Remember if there were
703 * trailing slashes to handle symlinks, existing non-directories
704 * and non-existing files that won't be directories specially later.
706 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
710 *ndp->ni_next = '\0';
711 cnp->cn_flags |= TRAILINGSLASH;
716 cnp->cn_flags |= MAKEENTRY;
717 if (*cp == '\0' && docache == 0)
718 cnp->cn_flags &= ~MAKEENTRY;
719 if (cnp->cn_namelen == 2 &&
720 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
721 cnp->cn_flags |= ISDOTDOT;
723 cnp->cn_flags &= ~ISDOTDOT;
724 if (*ndp->ni_next == 0)
725 cnp->cn_flags |= ISLASTCN;
727 cnp->cn_flags &= ~ISLASTCN;
729 if ((cnp->cn_flags & ISLASTCN) != 0 &&
730 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
731 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
736 nameicap_tracker_add(ndp, dp);
739 * Check for degenerate name (e.g. / or "")
740 * which is a way of talking about a directory,
741 * e.g. like "/." or ".".
743 if (cnp->cn_nameptr[0] == '\0') {
744 if (dp->v_type != VDIR) {
748 if (cnp->cn_nameiop != LOOKUP) {
758 if (cnp->cn_flags & AUDITVNODE1)
759 AUDIT_ARG_VNODE1(dp);
760 else if (cnp->cn_flags & AUDITVNODE2)
761 AUDIT_ARG_VNODE2(dp);
763 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
765 /* XXX This should probably move to the top of function. */
766 if (cnp->cn_flags & SAVESTART)
767 panic("lookup: SAVESTART");
772 * Handle "..": five special cases.
773 * 0. If doing a capability lookup and lookup_cap_dotdot is
774 * disabled, return ENOTCAPABLE.
775 * 1. Return an error if this is the last component of
776 * the name and the operation is DELETE or RENAME.
777 * 2. If at root directory (e.g. after chroot)
778 * or at absolute root directory
779 * then ignore it so can't get out.
780 * 3. If this vnode is the root of a mounted
781 * filesystem, then replace it with the
782 * vnode which was mounted on so we take the
783 * .. in the other filesystem.
784 * 4. If the vnode is the top directory of
785 * the jail or chroot, don't let them out.
786 * 5. If doing a capability lookup and lookup_cap_dotdot is
787 * enabled, return ENOTCAPABLE if the lookup would escape
788 * from the initial file descriptor directory. Checks are
789 * done by ensuring that namei() already traversed the
790 * result of dotdot lookup.
792 if (cnp->cn_flags & ISDOTDOT) {
793 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT))
794 == NI_LCF_STRICTRELATIVE) {
796 if (KTRPOINT(curthread, KTR_CAPFAIL))
797 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
802 if ((cnp->cn_flags & ISLASTCN) != 0 &&
803 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
808 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
810 if (dp == pr->pr_root)
812 if (dp == ndp->ni_rootdir ||
813 dp == ndp->ni_topdir ||
816 ((dp->v_vflag & VV_ROOT) != 0 &&
817 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
823 if ((dp->v_vflag & VV_ROOT) == 0)
825 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
830 dp = dp->v_mount->mnt_vnodecovered;
834 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
835 LK_RETRY, ISDOTDOT));
836 error = nameicap_check_dotdot(ndp, dp);
839 if (KTRPOINT(curthread, KTR_CAPFAIL))
840 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
848 * We now have a segment name to search for, and a directory to search.
852 if ((cnp->cn_flags & NOMACCHECK) == 0) {
853 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
861 ASSERT_VOP_LOCKED(dp, "lookup");
863 * If we have a shared lock we may need to upgrade the lock for the
866 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
867 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
868 vn_lock(dp, LK_UPGRADE|LK_RETRY);
869 if ((dp->v_iflag & VI_DOOMED) != 0) {
874 * If we're looking up the last component and we need an exclusive
875 * lock, adjust our lkflags.
877 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
878 cnp->cn_lkflags = LK_EXCLUSIVE;
879 #ifdef NAMEI_DIAGNOSTIC
880 vn_printf(dp, "lookup in ");
882 lkflags_save = cnp->cn_lkflags;
883 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
885 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
886 cnp->cn_lkflags = lkflags_save;
888 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
889 #ifdef NAMEI_DIAGNOSTIC
890 printf("not found\n");
892 if ((error == ENOENT) &&
893 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
894 (dp->v_mount->mnt_flag & MNT_UNION)) {
896 dp = dp->v_mount->mnt_vnodecovered;
900 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
901 LK_RETRY, cnp->cn_flags));
902 nameicap_tracker_add(ndp, dp);
906 if (error == ERELOOKUP) {
914 if (error != EJUSTRETURN)
917 * At this point, we know we're at the end of the
918 * pathname. If creating / renaming, we can consider
919 * allowing the file or directory to be created / renamed,
920 * provided we're not on a read-only filesystem.
926 /* trailing slash only allowed for directories */
927 if ((cnp->cn_flags & TRAILINGSLASH) &&
928 !(cnp->cn_flags & WILLBEDIR)) {
932 if ((cnp->cn_flags & LOCKPARENT) == 0)
935 * We return with ni_vp NULL to indicate that the entry
936 * doesn't currently exist, leaving a pointer to the
937 * (possibly locked) directory vnode in ndp->ni_dvp.
939 if (cnp->cn_flags & SAVESTART) {
940 ndp->ni_startdir = ndp->ni_dvp;
941 VREF(ndp->ni_startdir);
947 #ifdef NAMEI_DIAGNOSTIC
953 * Check to see if the vnode has been mounted on;
954 * if so find the root of the mounted filesystem.
956 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
957 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
961 if (dp != ndp->ni_dvp)
966 ndp->ni_dvp = vp_crossmp;
967 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
968 cnp->cn_flags), &tdp);
970 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
971 panic("vp_crossmp exclusively locked or reclaimed");
976 ndp->ni_vp = dp = tdp;
980 * Check for symbolic link
982 if ((dp->v_type == VLNK) &&
983 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
984 *ndp->ni_next == '/')) {
985 cnp->cn_flags |= ISSYMLINK;
986 if (dp->v_iflag & VI_DOOMED) {
988 * We can't know whether the directory was mounted with
989 * NOSYMFOLLOW, so we can't follow safely.
994 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
999 * Symlink code always expects an unlocked dvp.
1001 if (ndp->ni_dvp != ndp->ni_vp) {
1002 VOP_UNLOCK(ndp->ni_dvp, 0);
1003 ni_dvp_unlocked = 1;
1010 * Not a symbolic link that we will follow. Continue with the
1011 * next component if there is any; otherwise, we're done.
1013 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1014 ("lookup: invalid path state."));
1017 ndp->ni_pathlen = prev_ni_pathlen;
1018 ndp->ni_next = prev_ni_next;
1019 if (ndp->ni_dvp != dp)
1025 if (cnp->cn_flags & ISDOTDOT) {
1026 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1029 if (KTRPOINT(curthread, KTR_CAPFAIL))
1030 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1035 if (*ndp->ni_next == '/') {
1036 cnp->cn_nameptr = ndp->ni_next;
1037 while (*cnp->cn_nameptr == '/') {
1041 if (ndp->ni_dvp != dp)
1048 * If we're processing a path with a trailing slash,
1049 * check that the end result is a directory.
1051 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1056 * Disallow directory write attempts on read-only filesystems.
1059 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1063 if (cnp->cn_flags & SAVESTART) {
1064 ndp->ni_startdir = ndp->ni_dvp;
1065 VREF(ndp->ni_startdir);
1068 ni_dvp_unlocked = 2;
1069 if (ndp->ni_dvp != dp)
1073 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1074 VOP_UNLOCK(ndp->ni_dvp, 0);
1075 ni_dvp_unlocked = 1;
1078 if (cnp->cn_flags & AUDITVNODE1)
1079 AUDIT_ARG_VNODE1(dp);
1080 else if (cnp->cn_flags & AUDITVNODE2)
1081 AUDIT_ARG_VNODE2(dp);
1083 if ((cnp->cn_flags & LOCKLEAF) == 0)
1087 * Because of lookup_shared we may have the vnode shared locked, but
1088 * the caller may want it to be exclusively locked.
1090 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1091 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1092 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1093 if (dp->v_iflag & VI_DOOMED) {
1101 if (ni_dvp_unlocked != 2) {
1102 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1115 * relookup - lookup a path name component
1116 * Used by lookup to re-acquire things.
1119 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
1121 struct vnode *dp = NULL; /* the directory we are searching */
1122 int wantparent; /* 1 => wantparent or lockparent flag */
1123 int rdonly; /* lookup read-only flag bit */
1126 KASSERT(cnp->cn_flags & ISLASTCN,
1127 ("relookup: Not given last component."));
1129 * Setup: break out flag bits into variables.
1131 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
1132 KASSERT(wantparent, ("relookup: parent not wanted."));
1133 rdonly = cnp->cn_flags & RDONLY;
1134 cnp->cn_flags &= ~ISSYMLINK;
1136 cnp->cn_lkflags = LK_EXCLUSIVE;
1137 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1140 * Search a new directory.
1142 * The last component of the filename is left accessible via
1143 * cnp->cn_nameptr for callers that need the name. Callers needing
1144 * the name set the SAVENAME flag. When done, they assume
1145 * responsibility for freeing the pathname buffer.
1147 #ifdef NAMEI_DIAGNOSTIC
1148 printf("{%s}: ", cnp->cn_nameptr);
1152 * Check for "" which represents the root directory after slash
1155 if (cnp->cn_nameptr[0] == '\0') {
1157 * Support only LOOKUP for "/" because lookup()
1158 * can't succeed for CREATE, DELETE and RENAME.
1160 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1161 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1163 if (!(cnp->cn_flags & LOCKLEAF))
1166 /* XXX This should probably move to the top of function. */
1167 if (cnp->cn_flags & SAVESTART)
1168 panic("lookup: SAVESTART");
1172 if (cnp->cn_flags & ISDOTDOT)
1173 panic ("relookup: lookup on dot-dot");
1176 * We now have a segment name to search for, and a directory to search.
1178 #ifdef NAMEI_DIAGNOSTIC
1179 vn_printf(dp, "search in ");
1181 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1182 KASSERT(*vpp == NULL, ("leaf should be empty"));
1183 if (error != EJUSTRETURN)
1186 * If creating and at end of pathname, then can consider
1187 * allowing file to be created.
1193 /* ASSERT(dvp == ndp->ni_startdir) */
1194 if (cnp->cn_flags & SAVESTART)
1196 if ((cnp->cn_flags & LOCKPARENT) == 0)
1199 * We return with ni_vp NULL to indicate that the entry
1200 * doesn't currently exist, leaving a pointer to the
1201 * (possibly locked) directory vnode in ndp->ni_dvp.
1209 * Disallow directory write attempts on read-only filesystems.
1212 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1221 * Set the parent lock/ref state to the requested state.
1223 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1228 } else if (!wantparent)
1231 * Check for symbolic link
1233 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1234 ("relookup: symlink found.\n"));
1236 /* ASSERT(dvp == ndp->ni_startdir) */
1237 if (cnp->cn_flags & SAVESTART)
1240 if ((cnp->cn_flags & LOCKLEAF) == 0)
1250 NDINIT_ALL(struct nameidata *ndp, u_long op, u_long flags, enum uio_seg segflg,
1251 const char *namep, int dirfd, struct vnode *startdir, cap_rights_t *rightsp,
1255 ndp->ni_cnd.cn_nameiop = op;
1256 ndp->ni_cnd.cn_flags = flags;
1257 ndp->ni_segflg = segflg;
1258 ndp->ni_dirp = namep;
1259 ndp->ni_dirfd = dirfd;
1260 ndp->ni_startdir = startdir;
1261 if (rightsp != NULL)
1262 ndp->ni_rightsneeded = *rightsp;
1264 cap_rights_init(&ndp->ni_rightsneeded);
1265 filecaps_init(&ndp->ni_filecaps);
1266 ndp->ni_cnd.cn_thread = td;
1270 * Free data allocated by namei(); see namei(9) for details.
1273 NDFREE(struct nameidata *ndp, const u_int flags)
1281 if (!(flags & NDF_NO_FREE_PNBUF) &&
1282 (ndp->ni_cnd.cn_flags & HASBUF)) {
1283 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1284 ndp->ni_cnd.cn_flags &= ~HASBUF;
1286 if (!(flags & NDF_NO_VP_UNLOCK) &&
1287 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1289 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1298 VOP_UNLOCK(ndp->ni_vp, 0);
1299 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1300 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1301 ndp->ni_dvp != ndp->ni_vp)
1303 if (!(flags & NDF_NO_DVP_RELE) &&
1304 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1313 VOP_UNLOCK(ndp->ni_dvp, 0);
1314 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1315 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1316 vrele(ndp->ni_startdir);
1317 ndp->ni_startdir = NULL;
1322 * Determine if there is a suitable alternate filename under the specified
1323 * prefix for the specified path. If the create flag is set, then the
1324 * alternate prefix will be used so long as the parent directory exists.
1325 * This is used by the various compatibility ABIs so that Linux binaries prefer
1326 * files under /compat/linux for example. The chosen path (whether under
1327 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1328 * to by pathbuf. The caller is responsible for free'ing the buffer from
1329 * the M_TEMP bucket if one is returned.
1332 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1333 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1335 struct nameidata nd, ndroot;
1336 char *ptr, *buf, *cp;
1340 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1343 /* Copy the prefix into the new pathname as a starting point. */
1344 len = strlcpy(buf, prefix, MAXPATHLEN);
1345 if (len >= MAXPATHLEN) {
1350 sz = MAXPATHLEN - len;
1353 /* Append the filename to the prefix. */
1354 if (pathseg == UIO_SYSSPACE)
1355 error = copystr(path, ptr, sz, &len);
1357 error = copyinstr(path, ptr, sz, &len);
1365 /* Only use a prefix with absolute pathnames. */
1371 if (dirfd != AT_FDCWD) {
1373 * We want the original because the "prefix" is
1374 * included in the already opened dirfd.
1376 bcopy(ptr, buf, len);
1381 * We know that there is a / somewhere in this pathname.
1382 * Search backwards for it, to find the file's parent dir
1383 * to see if it exists in the alternate tree. If it does,
1384 * and we want to create a file (cflag is set). We don't
1385 * need to worry about the root comparison in this case.
1389 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1392 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td);
1398 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td);
1405 * We now compare the vnode of the prefix to the one
1406 * vnode asked. If they resolve to be the same, then we
1407 * ignore the match so that the real root gets used.
1408 * This avoids the problem of traversing "../.." to find the
1409 * root directory and never finding it, because "/" resolves
1410 * to the emulation root directory. This is expensive :-(
1412 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1415 /* We shouldn't ever get an error from this namei(). */
1416 error = namei(&ndroot);
1418 if (nd.ni_vp == ndroot.ni_vp)
1421 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1422 vrele(ndroot.ni_vp);
1426 NDFREE(&nd, NDF_ONLY_PNBUF);
1430 /* If there was an error, use the original path name. */
1432 bcopy(ptr, buf, len);