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_DEFINE3(vfs, namei, lookup, entry, "struct vnode *", "char *",
76 SDT_PROBE_DEFINE2(vfs, namei, lookup, return, "int", "struct vnode *");
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;
128 if ((flags & LK_INTERLOCK) != 0)
130 WITNESS_UNLOCK(&lk->lock_object, 0, LOCK_FILE, LOCK_LINE);
131 LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, LOCK_FILE,
136 static struct vop_vector crossmp_vnodeops = {
137 .vop_default = &default_vnodeops,
138 .vop_islocked = crossmp_vop_islocked,
139 .vop_lock1 = crossmp_vop_lock1,
140 .vop_unlock = crossmp_vop_unlock,
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 getnewvnode("crossmp", NULL, &crossmp_vnodeops, &vp_crossmp);
161 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
163 static int lookup_shared = 1;
164 SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RWTUN, &lookup_shared, 0,
165 "enables shared locks for path name translation");
167 static int lookup_cap_dotdot = 1;
168 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot, CTLFLAG_RWTUN,
169 &lookup_cap_dotdot, 0,
170 "enables \"..\" components in path lookup in capability mode");
171 static int lookup_cap_dotdot_nonlocal = 1;
172 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot_nonlocal, CTLFLAG_RWTUN,
173 &lookup_cap_dotdot_nonlocal, 0,
174 "enables \"..\" components in path lookup in capability mode "
175 "on non-local mount");
178 nameicap_tracker_add(struct nameidata *ndp, struct vnode *dp)
180 struct nameicap_tracker *nt;
182 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp->v_type != VDIR)
184 nt = uma_zalloc(nt_zone, M_WAITOK);
187 TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link);
191 nameicap_cleanup(struct nameidata *ndp)
193 struct nameicap_tracker *nt, *nt1;
195 KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) ||
196 (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative"));
197 TAILQ_FOREACH_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) {
198 TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link);
200 uma_zfree(nt_zone, nt);
205 * For dotdot lookups in capability mode, only allow the component
206 * lookup to succeed if the resulting directory was already traversed
207 * during the operation. Also fail dotdot lookups for non-local
208 * filesystems, where external agents might assist local lookups to
209 * escape the compartment.
212 nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp)
214 struct nameicap_tracker *nt;
217 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp == NULL ||
221 if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL &&
222 (mp->mnt_flag & MNT_LOCAL) == 0)
223 return (ENOTCAPABLE);
224 TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head,
229 return (ENOTCAPABLE);
233 namei_cleanup_cnp(struct componentname *cnp)
236 uma_zfree(namei_zone, cnp->cn_pnbuf);
238 cnp->cn_pnbuf = NULL;
239 cnp->cn_nameptr = NULL;
244 namei_handle_root(struct nameidata *ndp, struct vnode **dpp)
246 struct componentname *cnp;
249 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0) {
251 if (KTRPOINT(curthread, KTR_CAPFAIL))
252 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
254 return (ENOTCAPABLE);
256 while (*(cnp->cn_nameptr) == '/') {
260 *dpp = ndp->ni_rootdir;
266 * Convert a pathname into a pointer to a locked vnode.
268 * The FOLLOW flag is set when symbolic links are to be followed
269 * when they occur at the end of the name translation process.
270 * Symbolic links are always followed for all other pathname
271 * components other than the last.
273 * The segflg defines whether the name is to be copied from user
274 * space or kernel space.
276 * Overall outline of namei:
279 * get starting directory
280 * while (!done && !error) {
281 * call lookup to search path.
282 * if symbolic link, massage name in buffer and continue
286 namei(struct nameidata *ndp)
288 struct filedesc *fdp; /* pointer to file descriptor state */
289 char *cp; /* pointer into pathname argument */
290 struct vnode *dp; /* the directory we are searching */
291 struct iovec aiov; /* uio for reading symbolic links */
292 struct componentname *cnp;
297 int error, linklen, startdir_used;
302 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
303 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
304 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
305 ("namei: nameiop contaminated with flags"));
306 KASSERT((cnp->cn_flags & OPMASK) == 0,
307 ("namei: flags contaminated with nameiops"));
308 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
309 ndp->ni_startdir->v_type == VBAD);
311 cnp->cn_flags &= ~LOCKSHARED;
313 TAILQ_INIT(&ndp->ni_cap_tracker);
316 /* We will set this ourselves if we need it. */
317 cnp->cn_flags &= ~TRAILINGSLASH;
320 * Get a buffer for the name to be translated, and copy the
321 * name into the buffer.
323 if ((cnp->cn_flags & HASBUF) == 0)
324 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
325 if (ndp->ni_segflg == UIO_SYSSPACE)
326 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
329 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
333 * Don't allow empty pathnames.
335 if (error == 0 && *cnp->cn_pnbuf == '\0')
338 #ifdef CAPABILITY_MODE
340 * In capability mode, lookups must be restricted to happen in
341 * the subtree with the root specified by the file descriptor:
342 * - The root must be real file descriptor, not the pseudo-descriptor
344 * - The passed path must be relative and not absolute.
345 * - If lookup_cap_dotdot is disabled, path must not contain the
347 * - If lookup_cap_dotdot is enabled, we verify that all '..'
348 * components lookups result in the directories which were
349 * previously walked by us, which prevents an escape from
352 if (error == 0 && IN_CAPABILITY_MODE(td) &&
353 (cnp->cn_flags & NOCAPCHECK) == 0) {
354 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
355 if (ndp->ni_dirfd == AT_FDCWD) {
357 if (KTRPOINT(td, KTR_CAPFAIL))
358 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
365 namei_cleanup_cnp(cnp);
371 if (KTRPOINT(td, KTR_NAMEI)) {
372 KASSERT(cnp->cn_thread == curthread,
373 ("namei not using curthread"));
374 ktrnamei(cnp->cn_pnbuf);
378 * Get starting point for the translation.
381 ndp->ni_rootdir = fdp->fd_rdir;
382 vrefact(ndp->ni_rootdir);
383 ndp->ni_topdir = fdp->fd_jdir;
386 * If we are auditing the kernel pathname, save the user pathname.
388 if (cnp->cn_flags & AUDITVNODE1)
389 AUDIT_ARG_UPATH1(td, ndp->ni_dirfd, cnp->cn_pnbuf);
390 if (cnp->cn_flags & AUDITVNODE2)
391 AUDIT_ARG_UPATH2(td, ndp->ni_dirfd, cnp->cn_pnbuf);
395 cnp->cn_nameptr = cnp->cn_pnbuf;
396 if (cnp->cn_pnbuf[0] == '/') {
397 error = namei_handle_root(ndp, &dp);
399 if (ndp->ni_startdir != NULL) {
400 dp = ndp->ni_startdir;
402 } else if (ndp->ni_dirfd == AT_FDCWD) {
406 rights = ndp->ni_rightsneeded;
407 cap_rights_set(&rights, CAP_LOOKUP);
409 if (cnp->cn_flags & AUDITVNODE1)
410 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
411 if (cnp->cn_flags & AUDITVNODE2)
412 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
413 error = fgetvp_rights(td, ndp->ni_dirfd,
414 &rights, &ndp->ni_filecaps, &dp);
419 * If file descriptor doesn't have all rights,
420 * all lookups relative to it must also be
424 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights,
426 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
427 ndp->ni_filecaps.fc_nioctls != -1) {
428 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
432 if (error == 0 && dp->v_type != VDIR)
435 FILEDESC_SUNLOCK(fdp);
436 if (ndp->ni_startdir != NULL && !startdir_used)
437 vrele(ndp->ni_startdir);
443 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 &&
444 lookup_cap_dotdot != 0)
445 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
446 SDT_PROBE3(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
449 ndp->ni_startdir = dp;
454 * If not a symbolic link, we're done.
456 if ((cnp->cn_flags & ISSYMLINK) == 0) {
457 vrele(ndp->ni_rootdir);
458 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
459 namei_cleanup_cnp(cnp);
461 cnp->cn_flags |= HASBUF;
462 nameicap_cleanup(ndp);
463 SDT_PROBE2(vfs, namei, lookup, return, 0, ndp->ni_vp);
466 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
471 if ((cnp->cn_flags & NOMACCHECK) == 0) {
472 error = mac_vnode_check_readlink(td->td_ucred,
478 if (ndp->ni_pathlen > 1)
479 cp = uma_zalloc(namei_zone, M_WAITOK);
483 aiov.iov_len = MAXPATHLEN;
484 auio.uio_iov = &aiov;
487 auio.uio_rw = UIO_READ;
488 auio.uio_segflg = UIO_SYSSPACE;
490 auio.uio_resid = MAXPATHLEN;
491 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
493 if (ndp->ni_pathlen > 1)
494 uma_zfree(namei_zone, cp);
497 linklen = MAXPATHLEN - auio.uio_resid;
499 if (ndp->ni_pathlen > 1)
500 uma_zfree(namei_zone, cp);
504 if (linklen + ndp->ni_pathlen > MAXPATHLEN) {
505 if (ndp->ni_pathlen > 1)
506 uma_zfree(namei_zone, cp);
507 error = ENAMETOOLONG;
510 if (ndp->ni_pathlen > 1) {
511 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
512 uma_zfree(namei_zone, cnp->cn_pnbuf);
515 cnp->cn_pnbuf[linklen] = '\0';
516 ndp->ni_pathlen += linklen;
520 * Check if root directory should replace current directory.
522 cnp->cn_nameptr = cnp->cn_pnbuf;
523 if (*(cnp->cn_nameptr) == '/') {
525 error = namei_handle_root(ndp, &dp);
534 vrele(ndp->ni_rootdir);
535 namei_cleanup_cnp(cnp);
536 nameicap_cleanup(ndp);
537 SDT_PROBE2(vfs, namei, lookup, return, error, NULL);
542 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
545 if (mp == NULL || ((lkflags & LK_SHARED) &&
546 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
547 ((cnflags & ISDOTDOT) &&
548 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
549 lkflags &= ~LK_SHARED;
550 lkflags |= LK_EXCLUSIVE;
552 lkflags |= LK_NODDLKTREAT;
557 needs_exclusive_leaf(struct mount *mp, int flags)
561 * Intermediate nodes can use shared locks, we only need to
562 * force an exclusive lock for leaf nodes.
564 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
567 /* Always use exclusive locks if LOCKSHARED isn't set. */
568 if (!(flags & LOCKSHARED))
572 * For lookups during open(), if the mount point supports
573 * extended shared operations, then use a shared lock for the
574 * leaf node, otherwise use an exclusive lock.
576 if ((flags & ISOPEN) != 0)
577 return (!MNT_EXTENDED_SHARED(mp));
580 * Lookup requests outside of open() that specify LOCKSHARED
581 * only need a shared lock on the leaf vnode.
588 * This is a very central and rather complicated routine.
590 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
591 * The starting directory is taken from ni_startdir. The pathname is
592 * descended until done, or a symbolic link is encountered. The variable
593 * ni_more is clear if the path is completed; it is set to one if a
594 * symbolic link needing interpretation is encountered.
596 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
597 * whether the name is to be looked up, created, renamed, or deleted.
598 * When CREATE, RENAME, or DELETE is specified, information usable in
599 * creating, renaming, or deleting a directory entry may be calculated.
600 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
601 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
602 * returned unlocked. Otherwise the parent directory is not returned. If
603 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
604 * the target is returned locked, otherwise it is returned unlocked.
605 * When creating or renaming and LOCKPARENT is specified, the target may not
606 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
608 * Overall outline of lookup:
611 * identify next component of name at ndp->ni_ptr
612 * handle degenerate case where name is null string
613 * if .. and crossing mount points and on mounted filesys, find parent
614 * call VOP_LOOKUP routine for next component name
615 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
616 * component vnode returned in ni_vp (if it exists), locked.
617 * if result vnode is mounted on and crossing mount points,
618 * find mounted on vnode
619 * if more components of name, do next level at dirloop
620 * return the answer in ni_vp, locked if LOCKLEAF set
621 * if LOCKPARENT set, return locked parent in ni_dvp
622 * if WANTPARENT set, return unlocked parent in ni_dvp
625 lookup(struct nameidata *ndp)
627 char *cp; /* pointer into pathname argument */
628 char *prev_ni_next; /* saved ndp->ni_next */
629 struct vnode *dp = NULL; /* the directory we are searching */
630 struct vnode *tdp; /* saved dp */
631 struct mount *mp; /* mount table entry */
633 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
634 int docache; /* == 0 do not cache last component */
635 int wantparent; /* 1 => wantparent or lockparent flag */
636 int rdonly; /* lookup read-only flag bit */
638 int dpunlocked = 0; /* dp has already been unlocked */
639 int relookup = 0; /* do not consume the path component */
640 struct componentname *cnp = &ndp->ni_cnd;
645 * Setup: break out flag bits into variables.
648 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
649 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
650 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
651 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
652 if (cnp->cn_nameiop == DELETE ||
653 (wantparent && cnp->cn_nameiop != CREATE &&
654 cnp->cn_nameiop != LOOKUP))
656 rdonly = cnp->cn_flags & RDONLY;
657 cnp->cn_flags &= ~ISSYMLINK;
660 * We use shared locks until we hit the parent of the last cn then
661 * we adjust based on the requesting flags.
664 cnp->cn_lkflags = LK_SHARED;
666 cnp->cn_lkflags = LK_EXCLUSIVE;
667 dp = ndp->ni_startdir;
668 ndp->ni_startdir = NULLVP;
670 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
675 * Search a new directory.
677 * The last component of the filename is left accessible via
678 * cnp->cn_nameptr for callers that need the name. Callers needing
679 * the name set the SAVENAME flag. When done, they assume
680 * responsibility for freeing the pathname buffer.
682 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
684 cnp->cn_namelen = cp - cnp->cn_nameptr;
685 if (cnp->cn_namelen > NAME_MAX) {
686 error = ENAMETOOLONG;
689 #ifdef NAMEI_DIAGNOSTIC
692 printf("{%s}: ", cnp->cn_nameptr);
695 prev_ni_pathlen = ndp->ni_pathlen;
696 ndp->ni_pathlen -= cnp->cn_namelen;
697 KASSERT(ndp->ni_pathlen <= PATH_MAX,
698 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
699 prev_ni_next = ndp->ni_next;
703 * Replace multiple slashes by a single slash and trailing slashes
704 * by a null. This must be done before VOP_LOOKUP() because some
705 * fs's don't know about trailing slashes. Remember if there were
706 * trailing slashes to handle symlinks, existing non-directories
707 * and non-existing files that won't be directories specially later.
709 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
713 *ndp->ni_next = '\0';
714 cnp->cn_flags |= TRAILINGSLASH;
719 cnp->cn_flags |= MAKEENTRY;
720 if (*cp == '\0' && docache == 0)
721 cnp->cn_flags &= ~MAKEENTRY;
722 if (cnp->cn_namelen == 2 &&
723 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
724 cnp->cn_flags |= ISDOTDOT;
726 cnp->cn_flags &= ~ISDOTDOT;
727 if (*ndp->ni_next == 0)
728 cnp->cn_flags |= ISLASTCN;
730 cnp->cn_flags &= ~ISLASTCN;
732 if ((cnp->cn_flags & ISLASTCN) != 0 &&
733 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
734 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
739 nameicap_tracker_add(ndp, dp);
742 * Check for degenerate name (e.g. / or "")
743 * which is a way of talking about a directory,
744 * e.g. like "/." or ".".
746 if (cnp->cn_nameptr[0] == '\0') {
747 if (dp->v_type != VDIR) {
751 if (cnp->cn_nameiop != LOOKUP) {
761 if (cnp->cn_flags & AUDITVNODE1)
762 AUDIT_ARG_VNODE1(dp);
763 else if (cnp->cn_flags & AUDITVNODE2)
764 AUDIT_ARG_VNODE2(dp);
766 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
768 /* XXX This should probably move to the top of function. */
769 if (cnp->cn_flags & SAVESTART)
770 panic("lookup: SAVESTART");
775 * Handle "..": five special cases.
776 * 0. If doing a capability lookup and lookup_cap_dotdot is
777 * disabled, return ENOTCAPABLE.
778 * 1. Return an error if this is the last component of
779 * the name and the operation is DELETE or RENAME.
780 * 2. If at root directory (e.g. after chroot)
781 * or at absolute root directory
782 * then ignore it so can't get out.
783 * 3. If this vnode is the root of a mounted
784 * filesystem, then replace it with the
785 * vnode which was mounted on so we take the
786 * .. in the other filesystem.
787 * 4. If the vnode is the top directory of
788 * the jail or chroot, don't let them out.
789 * 5. If doing a capability lookup and lookup_cap_dotdot is
790 * enabled, return ENOTCAPABLE if the lookup would escape
791 * from the initial file descriptor directory. Checks are
792 * done by ensuring that namei() already traversed the
793 * result of dotdot lookup.
795 if (cnp->cn_flags & ISDOTDOT) {
796 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT))
797 == NI_LCF_STRICTRELATIVE) {
799 if (KTRPOINT(curthread, KTR_CAPFAIL))
800 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
805 if ((cnp->cn_flags & ISLASTCN) != 0 &&
806 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
811 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
813 if (dp == pr->pr_root)
815 if (dp == ndp->ni_rootdir ||
816 dp == ndp->ni_topdir ||
819 ((dp->v_vflag & VV_ROOT) != 0 &&
820 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
826 if ((dp->v_vflag & VV_ROOT) == 0)
828 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
833 dp = dp->v_mount->mnt_vnodecovered;
837 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
838 LK_RETRY, ISDOTDOT));
839 error = nameicap_check_dotdot(ndp, dp);
842 if (KTRPOINT(curthread, KTR_CAPFAIL))
843 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
851 * We now have a segment name to search for, and a directory to search.
855 if ((cnp->cn_flags & NOMACCHECK) == 0) {
856 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
864 ASSERT_VOP_LOCKED(dp, "lookup");
866 * If we have a shared lock we may need to upgrade the lock for the
869 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
870 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
871 vn_lock(dp, LK_UPGRADE|LK_RETRY);
872 if ((dp->v_iflag & VI_DOOMED) != 0) {
877 * If we're looking up the last component and we need an exclusive
878 * lock, adjust our lkflags.
880 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
881 cnp->cn_lkflags = LK_EXCLUSIVE;
882 #ifdef NAMEI_DIAGNOSTIC
883 vn_printf(dp, "lookup in ");
885 lkflags_save = cnp->cn_lkflags;
886 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
888 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
889 cnp->cn_lkflags = lkflags_save;
891 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
892 #ifdef NAMEI_DIAGNOSTIC
893 printf("not found\n");
895 if ((error == ENOENT) &&
896 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
897 (dp->v_mount->mnt_flag & MNT_UNION)) {
899 dp = dp->v_mount->mnt_vnodecovered;
903 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
904 LK_RETRY, cnp->cn_flags));
905 nameicap_tracker_add(ndp, dp);
909 if (error == ERELOOKUP) {
917 if (error != EJUSTRETURN)
920 * At this point, we know we're at the end of the
921 * pathname. If creating / renaming, we can consider
922 * allowing the file or directory to be created / renamed,
923 * provided we're not on a read-only filesystem.
929 /* trailing slash only allowed for directories */
930 if ((cnp->cn_flags & TRAILINGSLASH) &&
931 !(cnp->cn_flags & WILLBEDIR)) {
935 if ((cnp->cn_flags & LOCKPARENT) == 0)
938 * We return with ni_vp NULL to indicate that the entry
939 * doesn't currently exist, leaving a pointer to the
940 * (possibly locked) directory vnode in ndp->ni_dvp.
942 if (cnp->cn_flags & SAVESTART) {
943 ndp->ni_startdir = ndp->ni_dvp;
944 VREF(ndp->ni_startdir);
950 #ifdef NAMEI_DIAGNOSTIC
956 * Check to see if the vnode has been mounted on;
957 * if so find the root of the mounted filesystem.
959 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
960 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
964 if (dp != ndp->ni_dvp)
969 ndp->ni_dvp = vp_crossmp;
970 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
971 cnp->cn_flags), &tdp);
973 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
974 panic("vp_crossmp exclusively locked or reclaimed");
979 ndp->ni_vp = dp = tdp;
983 * Check for symbolic link
985 if ((dp->v_type == VLNK) &&
986 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
987 *ndp->ni_next == '/')) {
988 cnp->cn_flags |= ISSYMLINK;
989 if (dp->v_iflag & VI_DOOMED) {
991 * We can't know whether the directory was mounted with
992 * NOSYMFOLLOW, so we can't follow safely.
997 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1002 * Symlink code always expects an unlocked dvp.
1004 if (ndp->ni_dvp != ndp->ni_vp) {
1005 VOP_UNLOCK(ndp->ni_dvp, 0);
1006 ni_dvp_unlocked = 1;
1013 * Not a symbolic link that we will follow. Continue with the
1014 * next component if there is any; otherwise, we're done.
1016 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1017 ("lookup: invalid path state."));
1020 ndp->ni_pathlen = prev_ni_pathlen;
1021 ndp->ni_next = prev_ni_next;
1022 if (ndp->ni_dvp != dp)
1028 if (cnp->cn_flags & ISDOTDOT) {
1029 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1032 if (KTRPOINT(curthread, KTR_CAPFAIL))
1033 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1038 if (*ndp->ni_next == '/') {
1039 cnp->cn_nameptr = ndp->ni_next;
1040 while (*cnp->cn_nameptr == '/') {
1044 if (ndp->ni_dvp != dp)
1051 * If we're processing a path with a trailing slash,
1052 * check that the end result is a directory.
1054 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1059 * Disallow directory write attempts on read-only filesystems.
1062 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1066 if (cnp->cn_flags & SAVESTART) {
1067 ndp->ni_startdir = ndp->ni_dvp;
1068 VREF(ndp->ni_startdir);
1071 ni_dvp_unlocked = 2;
1072 if (ndp->ni_dvp != dp)
1076 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1077 VOP_UNLOCK(ndp->ni_dvp, 0);
1078 ni_dvp_unlocked = 1;
1081 if (cnp->cn_flags & AUDITVNODE1)
1082 AUDIT_ARG_VNODE1(dp);
1083 else if (cnp->cn_flags & AUDITVNODE2)
1084 AUDIT_ARG_VNODE2(dp);
1086 if ((cnp->cn_flags & LOCKLEAF) == 0)
1090 * Because of lookup_shared we may have the vnode shared locked, but
1091 * the caller may want it to be exclusively locked.
1093 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1094 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1095 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1096 if (dp->v_iflag & VI_DOOMED) {
1104 if (ni_dvp_unlocked != 2) {
1105 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1118 * relookup - lookup a path name component
1119 * Used by lookup to re-acquire things.
1122 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
1124 struct vnode *dp = NULL; /* the directory we are searching */
1125 int wantparent; /* 1 => wantparent or lockparent flag */
1126 int rdonly; /* lookup read-only flag bit */
1129 KASSERT(cnp->cn_flags & ISLASTCN,
1130 ("relookup: Not given last component."));
1132 * Setup: break out flag bits into variables.
1134 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
1135 KASSERT(wantparent, ("relookup: parent not wanted."));
1136 rdonly = cnp->cn_flags & RDONLY;
1137 cnp->cn_flags &= ~ISSYMLINK;
1139 cnp->cn_lkflags = LK_EXCLUSIVE;
1140 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1143 * Search a new directory.
1145 * The last component of the filename is left accessible via
1146 * cnp->cn_nameptr for callers that need the name. Callers needing
1147 * the name set the SAVENAME flag. When done, they assume
1148 * responsibility for freeing the pathname buffer.
1150 #ifdef NAMEI_DIAGNOSTIC
1151 printf("{%s}: ", cnp->cn_nameptr);
1155 * Check for "" which represents the root directory after slash
1158 if (cnp->cn_nameptr[0] == '\0') {
1160 * Support only LOOKUP for "/" because lookup()
1161 * can't succeed for CREATE, DELETE and RENAME.
1163 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1164 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1166 if (!(cnp->cn_flags & LOCKLEAF))
1169 /* XXX This should probably move to the top of function. */
1170 if (cnp->cn_flags & SAVESTART)
1171 panic("lookup: SAVESTART");
1175 if (cnp->cn_flags & ISDOTDOT)
1176 panic ("relookup: lookup on dot-dot");
1179 * We now have a segment name to search for, and a directory to search.
1181 #ifdef NAMEI_DIAGNOSTIC
1182 vn_printf(dp, "search in ");
1184 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1185 KASSERT(*vpp == NULL, ("leaf should be empty"));
1186 if (error != EJUSTRETURN)
1189 * If creating and at end of pathname, then can consider
1190 * allowing file to be created.
1196 /* ASSERT(dvp == ndp->ni_startdir) */
1197 if (cnp->cn_flags & SAVESTART)
1199 if ((cnp->cn_flags & LOCKPARENT) == 0)
1202 * We return with ni_vp NULL to indicate that the entry
1203 * doesn't currently exist, leaving a pointer to the
1204 * (possibly locked) directory vnode in ndp->ni_dvp.
1212 * Disallow directory write attempts on read-only filesystems.
1215 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1224 * Set the parent lock/ref state to the requested state.
1226 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1231 } else if (!wantparent)
1234 * Check for symbolic link
1236 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1237 ("relookup: symlink found.\n"));
1239 /* ASSERT(dvp == ndp->ni_startdir) */
1240 if (cnp->cn_flags & SAVESTART)
1243 if ((cnp->cn_flags & LOCKLEAF) == 0)
1253 NDINIT_ALL(struct nameidata *ndp, u_long op, u_long flags, enum uio_seg segflg,
1254 const char *namep, int dirfd, struct vnode *startdir, cap_rights_t *rightsp,
1258 ndp->ni_cnd.cn_nameiop = op;
1259 ndp->ni_cnd.cn_flags = flags;
1260 ndp->ni_segflg = segflg;
1261 ndp->ni_dirp = namep;
1262 ndp->ni_dirfd = dirfd;
1263 ndp->ni_startdir = startdir;
1264 if (rightsp != NULL)
1265 ndp->ni_rightsneeded = *rightsp;
1267 cap_rights_init(&ndp->ni_rightsneeded);
1268 filecaps_init(&ndp->ni_filecaps);
1269 ndp->ni_cnd.cn_thread = td;
1273 * Free data allocated by namei(); see namei(9) for details.
1276 NDFREE(struct nameidata *ndp, const u_int flags)
1284 if (!(flags & NDF_NO_FREE_PNBUF) &&
1285 (ndp->ni_cnd.cn_flags & HASBUF)) {
1286 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1287 ndp->ni_cnd.cn_flags &= ~HASBUF;
1289 if (!(flags & NDF_NO_VP_UNLOCK) &&
1290 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1292 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1301 VOP_UNLOCK(ndp->ni_vp, 0);
1302 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1303 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1304 ndp->ni_dvp != ndp->ni_vp)
1306 if (!(flags & NDF_NO_DVP_RELE) &&
1307 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1316 VOP_UNLOCK(ndp->ni_dvp, 0);
1317 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1318 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1319 vrele(ndp->ni_startdir);
1320 ndp->ni_startdir = NULL;
1325 * Determine if there is a suitable alternate filename under the specified
1326 * prefix for the specified path. If the create flag is set, then the
1327 * alternate prefix will be used so long as the parent directory exists.
1328 * This is used by the various compatibility ABIs so that Linux binaries prefer
1329 * files under /compat/linux for example. The chosen path (whether under
1330 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1331 * to by pathbuf. The caller is responsible for free'ing the buffer from
1332 * the M_TEMP bucket if one is returned.
1335 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1336 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1338 struct nameidata nd, ndroot;
1339 char *ptr, *buf, *cp;
1343 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1346 /* Copy the prefix into the new pathname as a starting point. */
1347 len = strlcpy(buf, prefix, MAXPATHLEN);
1348 if (len >= MAXPATHLEN) {
1353 sz = MAXPATHLEN - len;
1356 /* Append the filename to the prefix. */
1357 if (pathseg == UIO_SYSSPACE)
1358 error = copystr(path, ptr, sz, &len);
1360 error = copyinstr(path, ptr, sz, &len);
1368 /* Only use a prefix with absolute pathnames. */
1374 if (dirfd != AT_FDCWD) {
1376 * We want the original because the "prefix" is
1377 * included in the already opened dirfd.
1379 bcopy(ptr, buf, len);
1384 * We know that there is a / somewhere in this pathname.
1385 * Search backwards for it, to find the file's parent dir
1386 * to see if it exists in the alternate tree. If it does,
1387 * and we want to create a file (cflag is set). We don't
1388 * need to worry about the root comparison in this case.
1392 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1395 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1401 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1408 * We now compare the vnode of the prefix to the one
1409 * vnode asked. If they resolve to be the same, then we
1410 * ignore the match so that the real root gets used.
1411 * This avoids the problem of traversing "../.." to find the
1412 * root directory and never finding it, because "/" resolves
1413 * to the emulation root directory. This is expensive :-(
1415 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1418 /* We shouldn't ever get an error from this namei(). */
1419 error = namei(&ndroot);
1421 if (nd.ni_vp == ndroot.ni_vp)
1424 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1425 vrele(ndroot.ni_vp);
1429 NDFREE(&nd, NDF_ONLY_PNBUF);
1433 /* If there was an error, use the original path name. */
1435 bcopy(ptr, buf, len);