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_cap_dotdot = 1;
164 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot, CTLFLAG_RWTUN,
165 &lookup_cap_dotdot, 0,
166 "enables \"..\" components in path lookup in capability mode");
167 static int lookup_cap_dotdot_nonlocal = 1;
168 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot_nonlocal, CTLFLAG_RWTUN,
169 &lookup_cap_dotdot_nonlocal, 0,
170 "enables \"..\" components in path lookup in capability mode "
171 "on non-local mount");
174 nameicap_tracker_add(struct nameidata *ndp, struct vnode *dp)
176 struct nameicap_tracker *nt;
177 struct componentname *cnp;
179 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp->v_type != VDIR)
182 nt = TAILQ_LAST(&ndp->ni_cap_tracker, nameicap_tracker_head);
183 if (nt != NULL && nt->dp == dp)
185 nt = uma_zalloc(nt_zone, M_WAITOK);
188 TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link);
192 nameicap_cleanup_from(struct nameidata *ndp, struct nameicap_tracker *first)
194 struct nameicap_tracker *nt, *nt1;
197 TAILQ_FOREACH_FROM_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 nameicap_cleanup(struct nameidata *ndp)
207 KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) ||
208 (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative"));
209 nameicap_cleanup_from(ndp, NULL);
213 * For dotdot lookups in capability mode, only allow the component
214 * lookup to succeed if the resulting directory was already traversed
215 * during the operation. This catches situations where already
216 * traversed directory is moved to different parent, and then we walk
217 * over it with dotdots.
219 * Also allow to force failure of dotdot lookups for non-local
220 * filesystems, where external agents might assist local lookups to
221 * escape the compartment.
224 nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp)
226 struct nameicap_tracker *nt;
229 if (dp == NULL || dp->v_type != VDIR || (ndp->ni_lcf &
230 NI_LCF_STRICTRELATIVE) == 0)
232 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0)
233 return (ENOTCAPABLE);
235 if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL &&
236 (mp->mnt_flag & MNT_LOCAL) == 0)
237 return (ENOTCAPABLE);
238 TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head,
241 nt = TAILQ_NEXT(nt, nm_link);
243 nameicap_cleanup_from(ndp, nt);
247 return (ENOTCAPABLE);
251 namei_cleanup_cnp(struct componentname *cnp)
254 uma_zfree(namei_zone, cnp->cn_pnbuf);
256 cnp->cn_pnbuf = NULL;
257 cnp->cn_nameptr = NULL;
262 namei_handle_root(struct nameidata *ndp, struct vnode **dpp)
264 struct componentname *cnp;
267 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0) {
269 if (KTRPOINT(curthread, KTR_CAPFAIL))
270 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
272 return (ENOTCAPABLE);
274 while (*(cnp->cn_nameptr) == '/') {
278 *dpp = ndp->ni_rootdir;
284 * Convert a pathname into a pointer to a locked vnode.
286 * The FOLLOW flag is set when symbolic links are to be followed
287 * when they occur at the end of the name translation process.
288 * Symbolic links are always followed for all other pathname
289 * components other than the last.
291 * The segflg defines whether the name is to be copied from user
292 * space or kernel space.
294 * Overall outline of namei:
297 * get starting directory
298 * while (!done && !error) {
299 * call lookup to search path.
300 * if symbolic link, massage name in buffer and continue
304 namei(struct nameidata *ndp)
306 struct filedesc *fdp; /* pointer to file descriptor state */
307 char *cp; /* pointer into pathname argument */
308 struct vnode *dp; /* the directory we are searching */
309 struct iovec aiov; /* uio for reading symbolic links */
310 struct componentname *cnp;
316 int error, linklen, startdir_used;
321 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
322 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
323 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
324 ("namei: nameiop contaminated with flags"));
325 KASSERT((cnp->cn_flags & OPMASK) == 0,
326 ("namei: flags contaminated with nameiops"));
327 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
328 ndp->ni_startdir->v_type == VBAD);
330 TAILQ_INIT(&ndp->ni_cap_tracker);
333 /* We will set this ourselves if we need it. */
334 cnp->cn_flags &= ~TRAILINGSLASH;
337 * Get a buffer for the name to be translated, and copy the
338 * name into the buffer.
340 if ((cnp->cn_flags & HASBUF) == 0)
341 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
342 if (ndp->ni_segflg == UIO_SYSSPACE)
343 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
346 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
350 * Don't allow empty pathnames.
352 if (error == 0 && *cnp->cn_pnbuf == '\0')
355 #ifdef CAPABILITY_MODE
357 * In capability mode, lookups must be restricted to happen in
358 * the subtree with the root specified by the file descriptor:
359 * - The root must be real file descriptor, not the pseudo-descriptor
361 * - The passed path must be relative and not absolute.
362 * - If lookup_cap_dotdot is disabled, path must not contain the
364 * - If lookup_cap_dotdot is enabled, we verify that all '..'
365 * components lookups result in the directories which were
366 * previously walked by us, which prevents an escape from
369 if (error == 0 && IN_CAPABILITY_MODE(td) &&
370 (cnp->cn_flags & NOCAPCHECK) == 0) {
371 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
372 ndp->ni_resflags |= NIRES_STRICTREL;
373 if (ndp->ni_dirfd == AT_FDCWD) {
375 if (KTRPOINT(td, KTR_CAPFAIL))
376 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
383 namei_cleanup_cnp(cnp);
389 if (KTRPOINT(td, KTR_NAMEI)) {
390 KASSERT(cnp->cn_thread == curthread,
391 ("namei not using curthread"));
392 ktrnamei(cnp->cn_pnbuf);
396 * Get starting point for the translation.
399 ndp->ni_rootdir = fdp->fd_rdir;
400 vrefact(ndp->ni_rootdir);
401 ndp->ni_topdir = fdp->fd_jdir;
404 * If we are auditing the kernel pathname, save the user pathname.
406 if (cnp->cn_flags & AUDITVNODE1)
407 AUDIT_ARG_UPATH1(td, ndp->ni_dirfd, cnp->cn_pnbuf);
408 if (cnp->cn_flags & AUDITVNODE2)
409 AUDIT_ARG_UPATH2(td, ndp->ni_dirfd, cnp->cn_pnbuf);
413 cnp->cn_nameptr = cnp->cn_pnbuf;
414 if (cnp->cn_pnbuf[0] == '/') {
415 ndp->ni_resflags |= NIRES_ABS;
416 error = namei_handle_root(ndp, &dp);
418 if (ndp->ni_startdir != NULL) {
419 dp = ndp->ni_startdir;
421 } else if (ndp->ni_dirfd == AT_FDCWD) {
425 rights = ndp->ni_rightsneeded;
426 cap_rights_set(&rights, CAP_LOOKUP);
428 if (cnp->cn_flags & AUDITVNODE1)
429 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
430 if (cnp->cn_flags & AUDITVNODE2)
431 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
433 * Effectively inlined fgetvp_rights, because we need to
434 * inspect the file as well as grabbing the vnode.
436 error = fget_cap_locked(fdp, ndp->ni_dirfd, &rights,
437 &dfp, &ndp->ni_filecaps);
440 * Preserve the error; it should either be EBADF
441 * or capability-related, both of which can be
442 * safely returned to the caller.
444 } else if (dfp->f_ops == &badfileops) {
446 } else if (dfp->f_vnode == NULL) {
452 if ((dfp->f_flag & FSEARCH) != 0)
453 cnp->cn_flags |= NOEXECCHECK;
457 * If file descriptor doesn't have all rights,
458 * all lookups relative to it must also be
462 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights,
464 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
465 ndp->ni_filecaps.fc_nioctls != -1) {
466 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
467 ndp->ni_resflags |= NIRES_STRICTREL;
471 if (error == 0 && dp->v_type != VDIR)
474 FILEDESC_SUNLOCK(fdp);
476 if (error == 0 && (cnp->cn_flags & RBENEATH) != 0) {
477 if (cnp->cn_pnbuf[0] == '/') {
479 } else if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) == 0) {
480 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE |
485 if (ndp->ni_startdir != NULL && !startdir_used)
486 vrele(ndp->ni_startdir);
492 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 &&
493 lookup_cap_dotdot != 0)
494 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
495 SDT_PROBE3(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
498 ndp->ni_startdir = dp;
504 * If not a symbolic link, we're done.
506 if ((cnp->cn_flags & ISSYMLINK) == 0) {
507 vrele(ndp->ni_rootdir);
508 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
509 namei_cleanup_cnp(cnp);
511 cnp->cn_flags |= HASBUF;
512 nameicap_cleanup(ndp);
513 SDT_PROBE2(vfs, namei, lookup, return, error,
514 (error == 0 ? ndp->ni_vp : NULL));
517 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
522 if ((cnp->cn_flags & NOMACCHECK) == 0) {
523 error = mac_vnode_check_readlink(td->td_ucred,
529 if (ndp->ni_pathlen > 1)
530 cp = uma_zalloc(namei_zone, M_WAITOK);
534 aiov.iov_len = MAXPATHLEN;
535 auio.uio_iov = &aiov;
538 auio.uio_rw = UIO_READ;
539 auio.uio_segflg = UIO_SYSSPACE;
541 auio.uio_resid = MAXPATHLEN;
542 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
544 if (ndp->ni_pathlen > 1)
545 uma_zfree(namei_zone, cp);
548 linklen = MAXPATHLEN - auio.uio_resid;
550 if (ndp->ni_pathlen > 1)
551 uma_zfree(namei_zone, cp);
555 if (linklen + ndp->ni_pathlen > MAXPATHLEN) {
556 if (ndp->ni_pathlen > 1)
557 uma_zfree(namei_zone, cp);
558 error = ENAMETOOLONG;
561 if (ndp->ni_pathlen > 1) {
562 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
563 uma_zfree(namei_zone, cnp->cn_pnbuf);
566 cnp->cn_pnbuf[linklen] = '\0';
567 ndp->ni_pathlen += linklen;
571 * Check if root directory should replace current directory.
573 cnp->cn_nameptr = cnp->cn_pnbuf;
574 if (*(cnp->cn_nameptr) == '/') {
576 error = namei_handle_root(ndp, &dp);
585 vrele(ndp->ni_rootdir);
587 namei_cleanup_cnp(cnp);
588 nameicap_cleanup(ndp);
589 SDT_PROBE2(vfs, namei, lookup, return, error, NULL);
594 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
597 if (mp == NULL || ((lkflags & LK_SHARED) &&
598 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
599 ((cnflags & ISDOTDOT) &&
600 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
601 lkflags &= ~LK_SHARED;
602 lkflags |= LK_EXCLUSIVE;
604 lkflags |= LK_NODDLKTREAT;
609 needs_exclusive_leaf(struct mount *mp, int flags)
613 * Intermediate nodes can use shared locks, we only need to
614 * force an exclusive lock for leaf nodes.
616 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
619 /* Always use exclusive locks if LOCKSHARED isn't set. */
620 if (!(flags & LOCKSHARED))
624 * For lookups during open(), if the mount point supports
625 * extended shared operations, then use a shared lock for the
626 * leaf node, otherwise use an exclusive lock.
628 if ((flags & ISOPEN) != 0)
629 return (!MNT_EXTENDED_SHARED(mp));
632 * Lookup requests outside of open() that specify LOCKSHARED
633 * only need a shared lock on the leaf vnode.
640 * This is a very central and rather complicated routine.
642 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
643 * The starting directory is taken from ni_startdir. The pathname is
644 * descended until done, or a symbolic link is encountered. The variable
645 * ni_more is clear if the path is completed; it is set to one if a
646 * symbolic link needing interpretation is encountered.
648 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
649 * whether the name is to be looked up, created, renamed, or deleted.
650 * When CREATE, RENAME, or DELETE is specified, information usable in
651 * creating, renaming, or deleting a directory entry may be calculated.
652 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
653 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
654 * returned unlocked. Otherwise the parent directory is not returned. If
655 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
656 * the target is returned locked, otherwise it is returned unlocked.
657 * When creating or renaming and LOCKPARENT is specified, the target may not
658 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
660 * Overall outline of lookup:
663 * identify next component of name at ndp->ni_ptr
664 * handle degenerate case where name is null string
665 * if .. and crossing mount points and on mounted filesys, find parent
666 * call VOP_LOOKUP routine for next component name
667 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
668 * component vnode returned in ni_vp (if it exists), locked.
669 * if result vnode is mounted on and crossing mount points,
670 * find mounted on vnode
671 * if more components of name, do next level at dirloop
672 * return the answer in ni_vp, locked if LOCKLEAF set
673 * if LOCKPARENT set, return locked parent in ni_dvp
674 * if WANTPARENT set, return unlocked parent in ni_dvp
677 lookup(struct nameidata *ndp)
679 char *cp; /* pointer into pathname argument */
680 char *prev_ni_next; /* saved ndp->ni_next */
681 struct vnode *dp = NULL; /* the directory we are searching */
682 struct vnode *tdp; /* saved dp */
683 struct mount *mp; /* mount table entry */
685 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
686 int docache; /* == 0 do not cache last component */
687 int wantparent; /* 1 => wantparent or lockparent flag */
688 int rdonly; /* lookup read-only flag bit */
690 int dpunlocked = 0; /* dp has already been unlocked */
691 int relookup = 0; /* do not consume the path component */
692 struct componentname *cnp = &ndp->ni_cnd;
697 * Setup: break out flag bits into variables.
700 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
701 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
702 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
704 * When set to zero, docache causes the last component of the
705 * pathname to be deleted from the cache and the full lookup
706 * of the name to be done (via VOP_CACHEDLOOKUP()). Often
707 * filesystems need some pre-computed values that are made
708 * during the full lookup, for instance UFS sets dp->i_offset.
710 * The docache variable is set to zero when requested by the
711 * NOCACHE flag and for all modifying operations except CREATE.
713 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
714 if (cnp->cn_nameiop == DELETE ||
715 (wantparent && cnp->cn_nameiop != CREATE &&
716 cnp->cn_nameiop != LOOKUP))
718 rdonly = cnp->cn_flags & RDONLY;
719 cnp->cn_flags &= ~ISSYMLINK;
722 * We use shared locks until we hit the parent of the last cn then
723 * we adjust based on the requesting flags.
725 cnp->cn_lkflags = LK_SHARED;
726 dp = ndp->ni_startdir;
727 ndp->ni_startdir = NULLVP;
729 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
734 * Search a new directory.
736 * The last component of the filename is left accessible via
737 * cnp->cn_nameptr for callers that need the name. Callers needing
738 * the name set the SAVENAME flag. When done, they assume
739 * responsibility for freeing the pathname buffer.
741 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
743 cnp->cn_namelen = cp - cnp->cn_nameptr;
744 if (cnp->cn_namelen > NAME_MAX) {
745 error = ENAMETOOLONG;
748 #ifdef NAMEI_DIAGNOSTIC
751 printf("{%s}: ", cnp->cn_nameptr);
754 prev_ni_pathlen = ndp->ni_pathlen;
755 ndp->ni_pathlen -= cnp->cn_namelen;
756 KASSERT(ndp->ni_pathlen <= PATH_MAX,
757 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
758 prev_ni_next = ndp->ni_next;
762 * Replace multiple slashes by a single slash and trailing slashes
763 * by a null. This must be done before VOP_LOOKUP() because some
764 * fs's don't know about trailing slashes. Remember if there were
765 * trailing slashes to handle symlinks, existing non-directories
766 * and non-existing files that won't be directories specially later.
768 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
772 *ndp->ni_next = '\0';
773 cnp->cn_flags |= TRAILINGSLASH;
778 cnp->cn_flags |= MAKEENTRY;
779 if (*cp == '\0' && docache == 0)
780 cnp->cn_flags &= ~MAKEENTRY;
781 if (cnp->cn_namelen == 2 &&
782 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
783 cnp->cn_flags |= ISDOTDOT;
785 cnp->cn_flags &= ~ISDOTDOT;
786 if (*ndp->ni_next == 0)
787 cnp->cn_flags |= ISLASTCN;
789 cnp->cn_flags &= ~ISLASTCN;
791 if ((cnp->cn_flags & ISLASTCN) != 0 &&
792 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
793 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
798 nameicap_tracker_add(ndp, dp);
801 * Check for degenerate name (e.g. / or "")
802 * which is a way of talking about a directory,
803 * e.g. like "/." or ".".
805 if (cnp->cn_nameptr[0] == '\0') {
806 if (dp->v_type != VDIR) {
810 if (cnp->cn_nameiop != LOOKUP) {
820 if (cnp->cn_flags & AUDITVNODE1)
821 AUDIT_ARG_VNODE1(dp);
822 else if (cnp->cn_flags & AUDITVNODE2)
823 AUDIT_ARG_VNODE2(dp);
825 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
827 /* XXX This should probably move to the top of function. */
828 if (cnp->cn_flags & SAVESTART)
829 panic("lookup: SAVESTART");
834 * Handle "..": five special cases.
835 * 0. If doing a capability lookup and lookup_cap_dotdot is
836 * disabled, return ENOTCAPABLE.
837 * 1. Return an error if this is the last component of
838 * the name and the operation is DELETE or RENAME.
839 * 2. If at root directory (e.g. after chroot)
840 * or at absolute root directory
841 * then ignore it so can't get out.
842 * 3. If this vnode is the root of a mounted
843 * filesystem, then replace it with the
844 * vnode which was mounted on so we take the
845 * .. in the other filesystem.
846 * 4. If the vnode is the top directory of
847 * the jail or chroot, don't let them out.
848 * 5. If doing a capability lookup and lookup_cap_dotdot is
849 * enabled, return ENOTCAPABLE if the lookup would escape
850 * from the initial file descriptor directory. Checks are
851 * done by ensuring that namei() already traversed the
852 * result of dotdot lookup.
854 if (cnp->cn_flags & ISDOTDOT) {
855 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT))
856 == NI_LCF_STRICTRELATIVE) {
858 if (KTRPOINT(curthread, KTR_CAPFAIL))
859 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
864 if ((cnp->cn_flags & ISLASTCN) != 0 &&
865 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
870 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
872 if (dp == pr->pr_root)
874 if (dp == ndp->ni_rootdir ||
875 dp == ndp->ni_topdir ||
878 ((dp->v_vflag & VV_ROOT) != 0 &&
879 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
885 if ((dp->v_vflag & VV_ROOT) == 0)
887 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
892 dp = dp->v_mount->mnt_vnodecovered;
896 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
897 LK_RETRY, ISDOTDOT));
898 error = nameicap_check_dotdot(ndp, dp);
901 if (KTRPOINT(curthread, KTR_CAPFAIL))
902 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
910 * We now have a segment name to search for, and a directory to search.
914 if ((cnp->cn_flags & NOMACCHECK) == 0) {
915 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
923 ASSERT_VOP_LOCKED(dp, "lookup");
925 * If we have a shared lock we may need to upgrade the lock for the
928 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
929 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
930 vn_lock(dp, LK_UPGRADE|LK_RETRY);
931 if ((dp->v_iflag & VI_DOOMED) != 0) {
936 * If we're looking up the last component and we need an exclusive
937 * lock, adjust our lkflags.
939 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
940 cnp->cn_lkflags = LK_EXCLUSIVE;
941 #ifdef NAMEI_DIAGNOSTIC
942 vn_printf(dp, "lookup in ");
944 lkflags_save = cnp->cn_lkflags;
945 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
947 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
948 cnp->cn_lkflags = lkflags_save;
950 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
951 #ifdef NAMEI_DIAGNOSTIC
952 printf("not found\n");
954 if ((error == ENOENT) &&
955 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
956 (dp->v_mount->mnt_flag & MNT_UNION)) {
958 dp = dp->v_mount->mnt_vnodecovered;
962 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
963 LK_RETRY, cnp->cn_flags));
964 nameicap_tracker_add(ndp, dp);
968 if (error == ERELOOKUP) {
976 if (error != EJUSTRETURN)
979 * At this point, we know we're at the end of the
980 * pathname. If creating / renaming, we can consider
981 * allowing the file or directory to be created / renamed,
982 * provided we're not on a read-only filesystem.
988 /* trailing slash only allowed for directories */
989 if ((cnp->cn_flags & TRAILINGSLASH) &&
990 !(cnp->cn_flags & WILLBEDIR)) {
994 if ((cnp->cn_flags & LOCKPARENT) == 0)
997 * We return with ni_vp NULL to indicate that the entry
998 * doesn't currently exist, leaving a pointer to the
999 * (possibly locked) directory vnode in ndp->ni_dvp.
1001 if (cnp->cn_flags & SAVESTART) {
1002 ndp->ni_startdir = ndp->ni_dvp;
1003 VREF(ndp->ni_startdir);
1009 #ifdef NAMEI_DIAGNOSTIC
1015 * Check to see if the vnode has been mounted on;
1016 * if so find the root of the mounted filesystem.
1018 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
1019 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
1020 if (vfs_busy(mp, 0))
1023 if (dp != ndp->ni_dvp)
1027 vrefact(vp_crossmp);
1028 ndp->ni_dvp = vp_crossmp;
1029 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
1030 cnp->cn_flags), &tdp);
1032 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
1033 panic("vp_crossmp exclusively locked or reclaimed");
1038 ndp->ni_vp = dp = tdp;
1042 * Check for symbolic link
1044 if ((dp->v_type == VLNK) &&
1045 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
1046 *ndp->ni_next == '/')) {
1047 cnp->cn_flags |= ISSYMLINK;
1048 if (dp->v_iflag & VI_DOOMED) {
1050 * We can't know whether the directory was mounted with
1051 * NOSYMFOLLOW, so we can't follow safely.
1056 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1061 * Symlink code always expects an unlocked dvp.
1063 if (ndp->ni_dvp != ndp->ni_vp) {
1064 VOP_UNLOCK(ndp->ni_dvp, 0);
1065 ni_dvp_unlocked = 1;
1072 * Not a symbolic link that we will follow. Continue with the
1073 * next component if there is any; otherwise, we're done.
1075 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1076 ("lookup: invalid path state."));
1079 ndp->ni_pathlen = prev_ni_pathlen;
1080 ndp->ni_next = prev_ni_next;
1081 if (ndp->ni_dvp != dp)
1087 if (cnp->cn_flags & ISDOTDOT) {
1088 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1091 if (KTRPOINT(curthread, KTR_CAPFAIL))
1092 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1097 if (*ndp->ni_next == '/') {
1098 cnp->cn_nameptr = ndp->ni_next;
1099 while (*cnp->cn_nameptr == '/') {
1103 if (ndp->ni_dvp != dp)
1110 * If we're processing a path with a trailing slash,
1111 * check that the end result is a directory.
1113 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1118 * Disallow directory write attempts on read-only filesystems.
1121 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1125 if (cnp->cn_flags & SAVESTART) {
1126 ndp->ni_startdir = ndp->ni_dvp;
1127 VREF(ndp->ni_startdir);
1130 ni_dvp_unlocked = 2;
1131 if (ndp->ni_dvp != dp)
1135 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1136 VOP_UNLOCK(ndp->ni_dvp, 0);
1137 ni_dvp_unlocked = 1;
1140 if (cnp->cn_flags & AUDITVNODE1)
1141 AUDIT_ARG_VNODE1(dp);
1142 else if (cnp->cn_flags & AUDITVNODE2)
1143 AUDIT_ARG_VNODE2(dp);
1145 if ((cnp->cn_flags & LOCKLEAF) == 0)
1149 * Because of shared lookup we may have the vnode shared locked, but
1150 * the caller may want it to be exclusively locked.
1152 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1153 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1154 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1155 if (dp->v_iflag & VI_DOOMED) {
1160 if (ndp->ni_vp != NULL && ndp->ni_vp->v_type == VDIR) {
1161 if ((cnp->cn_flags & ISDOTDOT) == 0)
1162 nameicap_tracker_add(ndp, ndp->ni_vp);
1167 if (ni_dvp_unlocked != 2) {
1168 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1181 * relookup - lookup a path name component
1182 * Used by lookup to re-acquire things.
1185 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
1187 struct vnode *dp = NULL; /* the directory we are searching */
1188 int wantparent; /* 1 => wantparent or lockparent flag */
1189 int rdonly; /* lookup read-only flag bit */
1192 KASSERT(cnp->cn_flags & ISLASTCN,
1193 ("relookup: Not given last component."));
1195 * Setup: break out flag bits into variables.
1197 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
1198 KASSERT(wantparent, ("relookup: parent not wanted."));
1199 rdonly = cnp->cn_flags & RDONLY;
1200 cnp->cn_flags &= ~ISSYMLINK;
1202 cnp->cn_lkflags = LK_EXCLUSIVE;
1203 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1206 * Search a new directory.
1208 * The last component of the filename is left accessible via
1209 * cnp->cn_nameptr for callers that need the name. Callers needing
1210 * the name set the SAVENAME flag. When done, they assume
1211 * responsibility for freeing the pathname buffer.
1213 #ifdef NAMEI_DIAGNOSTIC
1214 printf("{%s}: ", cnp->cn_nameptr);
1218 * Check for "" which represents the root directory after slash
1221 if (cnp->cn_nameptr[0] == '\0') {
1223 * Support only LOOKUP for "/" because lookup()
1224 * can't succeed for CREATE, DELETE and RENAME.
1226 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1227 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1229 if (!(cnp->cn_flags & LOCKLEAF))
1232 /* XXX This should probably move to the top of function. */
1233 if (cnp->cn_flags & SAVESTART)
1234 panic("lookup: SAVESTART");
1238 if (cnp->cn_flags & ISDOTDOT)
1239 panic ("relookup: lookup on dot-dot");
1242 * We now have a segment name to search for, and a directory to search.
1244 #ifdef NAMEI_DIAGNOSTIC
1245 vn_printf(dp, "search in ");
1247 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1248 KASSERT(*vpp == NULL, ("leaf should be empty"));
1249 if (error != EJUSTRETURN)
1252 * If creating and at end of pathname, then can consider
1253 * allowing file to be created.
1259 /* ASSERT(dvp == ndp->ni_startdir) */
1260 if (cnp->cn_flags & SAVESTART)
1262 if ((cnp->cn_flags & LOCKPARENT) == 0)
1265 * We return with ni_vp NULL to indicate that the entry
1266 * doesn't currently exist, leaving a pointer to the
1267 * (possibly locked) directory vnode in ndp->ni_dvp.
1275 * Disallow directory write attempts on read-only filesystems.
1278 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1287 * Set the parent lock/ref state to the requested state.
1289 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1294 } else if (!wantparent)
1297 * Check for symbolic link
1299 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1300 ("relookup: symlink found.\n"));
1302 /* ASSERT(dvp == ndp->ni_startdir) */
1303 if (cnp->cn_flags & SAVESTART)
1306 if ((cnp->cn_flags & LOCKLEAF) == 0)
1316 NDINIT_ALL(struct nameidata *ndp, u_long op, u_long flags, enum uio_seg segflg,
1317 const char *namep, int dirfd, struct vnode *startdir, cap_rights_t *rightsp,
1321 ndp->ni_cnd.cn_nameiop = op;
1322 ndp->ni_cnd.cn_flags = flags;
1323 ndp->ni_segflg = segflg;
1324 ndp->ni_dirp = namep;
1325 ndp->ni_dirfd = dirfd;
1326 ndp->ni_startdir = startdir;
1327 ndp->ni_resflags = 0;
1328 if (rightsp != NULL)
1329 ndp->ni_rightsneeded = *rightsp;
1331 cap_rights_init(&ndp->ni_rightsneeded);
1332 filecaps_init(&ndp->ni_filecaps);
1333 ndp->ni_cnd.cn_thread = td;
1337 * Free data allocated by namei(); see namei(9) for details.
1340 NDFREE(struct nameidata *ndp, const u_int flags)
1348 if (!(flags & NDF_NO_FREE_PNBUF) &&
1349 (ndp->ni_cnd.cn_flags & HASBUF)) {
1350 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1351 ndp->ni_cnd.cn_flags &= ~HASBUF;
1353 if (!(flags & NDF_NO_VP_UNLOCK) &&
1354 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1356 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1357 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1358 ndp->ni_dvp != ndp->ni_vp)
1360 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1369 VOP_UNLOCK(ndp->ni_vp, 0);
1370 if (!(flags & NDF_NO_DVP_RELE) &&
1371 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1380 VOP_UNLOCK(ndp->ni_dvp, 0);
1381 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1382 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1383 vrele(ndp->ni_startdir);
1384 ndp->ni_startdir = NULL;
1389 * Determine if there is a suitable alternate filename under the specified
1390 * prefix for the specified path. If the create flag is set, then the
1391 * alternate prefix will be used so long as the parent directory exists.
1392 * This is used by the various compatibility ABIs so that Linux binaries prefer
1393 * files under /compat/linux for example. The chosen path (whether under
1394 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1395 * to by pathbuf. The caller is responsible for free'ing the buffer from
1396 * the M_TEMP bucket if one is returned.
1399 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1400 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1402 struct nameidata nd, ndroot;
1403 char *ptr, *buf, *cp;
1407 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1410 /* Copy the prefix into the new pathname as a starting point. */
1411 len = strlcpy(buf, prefix, MAXPATHLEN);
1412 if (len >= MAXPATHLEN) {
1417 sz = MAXPATHLEN - len;
1420 /* Append the filename to the prefix. */
1421 if (pathseg == UIO_SYSSPACE)
1422 error = copystr(path, ptr, sz, &len);
1424 error = copyinstr(path, ptr, sz, &len);
1432 /* Only use a prefix with absolute pathnames. */
1438 if (dirfd != AT_FDCWD) {
1440 * We want the original because the "prefix" is
1441 * included in the already opened dirfd.
1443 bcopy(ptr, buf, len);
1448 * We know that there is a / somewhere in this pathname.
1449 * Search backwards for it, to find the file's parent dir
1450 * to see if it exists in the alternate tree. If it does,
1451 * and we want to create a file (cflag is set). We don't
1452 * need to worry about the root comparison in this case.
1456 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1459 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1465 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1472 * We now compare the vnode of the prefix to the one
1473 * vnode asked. If they resolve to be the same, then we
1474 * ignore the match so that the real root gets used.
1475 * This avoids the problem of traversing "../.." to find the
1476 * root directory and never finding it, because "/" resolves
1477 * to the emulation root directory. This is expensive :-(
1479 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1482 /* We shouldn't ever get an error from this namei(). */
1483 error = namei(&ndroot);
1485 if (nd.ni_vp == ndroot.ni_vp)
1488 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1489 vrele(ndroot.ni_vp);
1493 NDFREE(&nd, NDF_ONLY_PNBUF);
1497 /* If there was an error, use the original path name. */
1499 bcopy(ptr, buf, len);