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 * 4. 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 *");
77 * Allocation zone for namei
79 uma_zone_t namei_zone;
81 * Placeholder vnode for mp traversal
83 static struct vnode *vp_crossmp;
86 nameiinit(void *dummy __unused)
89 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
91 getnewvnode("crossmp", NULL, &dead_vnodeops, &vp_crossmp);
92 vn_lock(vp_crossmp, LK_EXCLUSIVE);
93 VN_LOCK_ASHARE(vp_crossmp);
94 VOP_UNLOCK(vp_crossmp, 0);
96 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
98 static int lookup_shared = 1;
99 SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RWTUN, &lookup_shared, 0,
100 "Enables/Disables shared locks for path name translation");
103 namei_cleanup_cnp(struct componentname *cnp)
105 uma_zfree(namei_zone, cnp->cn_pnbuf);
107 cnp->cn_pnbuf = NULL;
108 cnp->cn_nameptr = NULL;
113 namei_handle_root(struct nameidata *ndp, struct vnode **dpp)
115 struct componentname *cnp;
118 if (ndp->ni_strictrelative != 0) {
120 if (KTRPOINT(curthread, KTR_CAPFAIL))
121 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
123 return (ENOTCAPABLE);
125 while (*(cnp->cn_nameptr) == '/') {
129 *dpp = ndp->ni_rootdir;
135 * Convert a pathname into a pointer to a locked vnode.
137 * The FOLLOW flag is set when symbolic links are to be followed
138 * when they occur at the end of the name translation process.
139 * Symbolic links are always followed for all other pathname
140 * components other than the last.
142 * The segflg defines whether the name is to be copied from user
143 * space or kernel space.
145 * Overall outline of namei:
148 * get starting directory
149 * while (!done && !error) {
150 * call lookup to search path.
151 * if symbolic link, massage name in buffer and continue
155 namei(struct nameidata *ndp)
157 struct filedesc *fdp; /* pointer to file descriptor state */
158 char *cp; /* pointer into pathname argument */
159 struct vnode *dp; /* the directory we are searching */
160 struct iovec aiov; /* uio for reading symbolic links */
162 int error, linklen, startdir_used;
163 struct componentname *cnp = &ndp->ni_cnd;
164 struct thread *td = cnp->cn_thread;
165 struct proc *p = td->td_proc;
167 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
168 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
169 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
170 ("namei: nameiop contaminated with flags"));
171 KASSERT((cnp->cn_flags & OPMASK) == 0,
172 ("namei: flags contaminated with nameiops"));
173 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
174 ndp->ni_startdir->v_type == VBAD);
176 cnp->cn_flags &= ~LOCKSHARED;
179 /* We will set this ourselves if we need it. */
180 cnp->cn_flags &= ~TRAILINGSLASH;
183 * Get a buffer for the name to be translated, and copy the
184 * name into the buffer.
186 if ((cnp->cn_flags & HASBUF) == 0)
187 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
188 if (ndp->ni_segflg == UIO_SYSSPACE)
189 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
190 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
192 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
193 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
196 * Don't allow empty pathnames.
198 if (error == 0 && *cnp->cn_pnbuf == '\0')
201 #ifdef CAPABILITY_MODE
203 * In capability mode, lookups must be "strictly relative" (i.e.
204 * not an absolute path, and not containing '..' components) to
205 * a real file descriptor, not the pseudo-descriptor AT_FDCWD.
207 if (error == 0 && IN_CAPABILITY_MODE(td) &&
208 (cnp->cn_flags & NOCAPCHECK) == 0) {
209 ndp->ni_strictrelative = 1;
210 if (ndp->ni_dirfd == AT_FDCWD) {
212 if (KTRPOINT(td, KTR_CAPFAIL))
213 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
220 namei_cleanup_cnp(cnp);
226 if (KTRPOINT(td, KTR_NAMEI)) {
227 KASSERT(cnp->cn_thread == curthread,
228 ("namei not using curthread"));
229 ktrnamei(cnp->cn_pnbuf);
233 * Get starting point for the translation.
236 ndp->ni_rootdir = fdp->fd_rdir;
237 VREF(ndp->ni_rootdir);
238 ndp->ni_topdir = fdp->fd_jdir;
241 * If we are auditing the kernel pathname, save the user pathname.
243 if (cnp->cn_flags & AUDITVNODE1)
244 AUDIT_ARG_UPATH1(td, ndp->ni_dirfd, cnp->cn_pnbuf);
245 if (cnp->cn_flags & AUDITVNODE2)
246 AUDIT_ARG_UPATH2(td, ndp->ni_dirfd, cnp->cn_pnbuf);
250 cnp->cn_nameptr = cnp->cn_pnbuf;
251 if (cnp->cn_pnbuf[0] == '/') {
252 error = namei_handle_root(ndp, &dp);
254 if (ndp->ni_startdir != NULL) {
255 dp = ndp->ni_startdir;
257 } else if (ndp->ni_dirfd == AT_FDCWD) {
263 rights = ndp->ni_rightsneeded;
264 cap_rights_set(&rights, CAP_LOOKUP);
266 if (cnp->cn_flags & AUDITVNODE1)
267 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
268 if (cnp->cn_flags & AUDITVNODE2)
269 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
270 error = fgetvp_rights(td, ndp->ni_dirfd,
271 &rights, &ndp->ni_filecaps, &dp);
276 * If file descriptor doesn't have all rights,
277 * all lookups relative to it must also be
281 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights,
283 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
284 ndp->ni_filecaps.fc_nioctls != -1) {
285 ndp->ni_strictrelative = 1;
289 if (error == 0 && dp->v_type != VDIR)
292 FILEDESC_SUNLOCK(fdp);
293 if (ndp->ni_startdir != NULL && !startdir_used)
294 vrele(ndp->ni_startdir);
298 vrele(ndp->ni_rootdir);
299 namei_cleanup_cnp(cnp);
302 SDT_PROBE3(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
305 ndp->ni_startdir = dp;
308 vrele(ndp->ni_rootdir);
309 namei_cleanup_cnp(cnp);
310 SDT_PROBE2(vfs, namei, lookup, return, error, NULL);
314 * If not a symbolic link, we're done.
316 if ((cnp->cn_flags & ISSYMLINK) == 0) {
317 vrele(ndp->ni_rootdir);
318 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
319 namei_cleanup_cnp(cnp);
321 cnp->cn_flags |= HASBUF;
323 SDT_PROBE2(vfs, namei, lookup, return, 0, ndp->ni_vp);
326 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
331 if ((cnp->cn_flags & NOMACCHECK) == 0) {
332 error = mac_vnode_check_readlink(td->td_ucred,
338 if (ndp->ni_pathlen > 1)
339 cp = uma_zalloc(namei_zone, M_WAITOK);
343 aiov.iov_len = MAXPATHLEN;
344 auio.uio_iov = &aiov;
347 auio.uio_rw = UIO_READ;
348 auio.uio_segflg = UIO_SYSSPACE;
350 auio.uio_resid = MAXPATHLEN;
351 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
353 if (ndp->ni_pathlen > 1)
354 uma_zfree(namei_zone, cp);
357 linklen = MAXPATHLEN - auio.uio_resid;
359 if (ndp->ni_pathlen > 1)
360 uma_zfree(namei_zone, cp);
364 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
365 if (ndp->ni_pathlen > 1)
366 uma_zfree(namei_zone, cp);
367 error = ENAMETOOLONG;
370 if (ndp->ni_pathlen > 1) {
371 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
372 uma_zfree(namei_zone, cnp->cn_pnbuf);
375 cnp->cn_pnbuf[linklen] = '\0';
376 ndp->ni_pathlen += linklen;
380 * Check if root directory should replace current directory.
382 cnp->cn_nameptr = cnp->cn_pnbuf;
383 if (*(cnp->cn_nameptr) == '/') {
385 error = namei_handle_root(ndp, &dp);
387 vrele(ndp->ni_rootdir);
388 namei_cleanup_cnp(cnp);
393 vrele(ndp->ni_rootdir);
394 namei_cleanup_cnp(cnp);
398 SDT_PROBE2(vfs, namei, lookup, return, error, NULL);
403 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
406 if (mp == NULL || ((lkflags & LK_SHARED) &&
407 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
408 ((cnflags & ISDOTDOT) &&
409 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
410 lkflags &= ~LK_SHARED;
411 lkflags |= LK_EXCLUSIVE;
413 lkflags |= LK_NODDLKTREAT;
418 needs_exclusive_leaf(struct mount *mp, int flags)
422 * Intermediate nodes can use shared locks, we only need to
423 * force an exclusive lock for leaf nodes.
425 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
428 /* Always use exclusive locks if LOCKSHARED isn't set. */
429 if (!(flags & LOCKSHARED))
433 * For lookups during open(), if the mount point supports
434 * extended shared operations, then use a shared lock for the
435 * leaf node, otherwise use an exclusive lock.
437 if ((flags & ISOPEN) != 0)
438 return (!MNT_EXTENDED_SHARED(mp));
441 * Lookup requests outside of open() that specify LOCKSHARED
442 * only need a shared lock on the leaf vnode.
449 * This is a very central and rather complicated routine.
451 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
452 * The starting directory is taken from ni_startdir. The pathname is
453 * descended until done, or a symbolic link is encountered. The variable
454 * ni_more is clear if the path is completed; it is set to one if a
455 * symbolic link needing interpretation is encountered.
457 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
458 * whether the name is to be looked up, created, renamed, or deleted.
459 * When CREATE, RENAME, or DELETE is specified, information usable in
460 * creating, renaming, or deleting a directory entry may be calculated.
461 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
462 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
463 * returned unlocked. Otherwise the parent directory is not returned. If
464 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
465 * the target is returned locked, otherwise it is returned unlocked.
466 * When creating or renaming and LOCKPARENT is specified, the target may not
467 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
469 * Overall outline of lookup:
472 * identify next component of name at ndp->ni_ptr
473 * handle degenerate case where name is null string
474 * if .. and crossing mount points and on mounted filesys, find parent
475 * call VOP_LOOKUP routine for next component name
476 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
477 * component vnode returned in ni_vp (if it exists), locked.
478 * if result vnode is mounted on and crossing mount points,
479 * find mounted on vnode
480 * if more components of name, do next level at dirloop
481 * return the answer in ni_vp, locked if LOCKLEAF set
482 * if LOCKPARENT set, return locked parent in ni_dvp
483 * if WANTPARENT set, return unlocked parent in ni_dvp
486 lookup(struct nameidata *ndp)
488 char *cp; /* pointer into pathname argument */
489 struct vnode *dp = NULL; /* the directory we are searching */
490 struct vnode *tdp; /* saved dp */
491 struct mount *mp; /* mount table entry */
493 int docache; /* == 0 do not cache last component */
494 int wantparent; /* 1 => wantparent or lockparent flag */
495 int rdonly; /* lookup read-only flag bit */
497 int dpunlocked = 0; /* dp has already been unlocked */
498 int relookup = 0; /* do not consume the path component */
499 struct componentname *cnp = &ndp->ni_cnd;
504 * Setup: break out flag bits into variables.
507 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
508 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
509 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
510 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
511 if (cnp->cn_nameiop == DELETE ||
512 (wantparent && cnp->cn_nameiop != CREATE &&
513 cnp->cn_nameiop != LOOKUP))
515 rdonly = cnp->cn_flags & RDONLY;
516 cnp->cn_flags &= ~ISSYMLINK;
519 * We use shared locks until we hit the parent of the last cn then
520 * we adjust based on the requesting flags.
523 cnp->cn_lkflags = LK_SHARED;
525 cnp->cn_lkflags = LK_EXCLUSIVE;
526 dp = ndp->ni_startdir;
527 ndp->ni_startdir = NULLVP;
529 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
534 * Search a new directory.
536 * The last component of the filename is left accessible via
537 * cnp->cn_nameptr for callers that need the name. Callers needing
538 * the name set the SAVENAME flag. When done, they assume
539 * responsibility for freeing the pathname buffer.
541 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
543 cnp->cn_namelen = cp - cnp->cn_nameptr;
544 if (cnp->cn_namelen > NAME_MAX) {
545 error = ENAMETOOLONG;
548 #ifdef NAMEI_DIAGNOSTIC
551 printf("{%s}: ", cnp->cn_nameptr);
554 ndp->ni_pathlen -= cnp->cn_namelen;
558 * Replace multiple slashes by a single slash and trailing slashes
559 * by a null. This must be done before VOP_LOOKUP() because some
560 * fs's don't know about trailing slashes. Remember if there were
561 * trailing slashes to handle symlinks, existing non-directories
562 * and non-existing files that won't be directories specially later.
564 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
568 *ndp->ni_next = '\0';
569 cnp->cn_flags |= TRAILINGSLASH;
574 cnp->cn_flags |= MAKEENTRY;
575 if (*cp == '\0' && docache == 0)
576 cnp->cn_flags &= ~MAKEENTRY;
577 if (cnp->cn_namelen == 2 &&
578 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
579 cnp->cn_flags |= ISDOTDOT;
581 cnp->cn_flags &= ~ISDOTDOT;
582 if (*ndp->ni_next == 0)
583 cnp->cn_flags |= ISLASTCN;
585 cnp->cn_flags &= ~ISLASTCN;
587 if ((cnp->cn_flags & ISLASTCN) != 0 &&
588 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
589 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
595 * Check for degenerate name (e.g. / or "")
596 * which is a way of talking about a directory,
597 * e.g. like "/." or ".".
599 if (cnp->cn_nameptr[0] == '\0') {
600 if (dp->v_type != VDIR) {
604 if (cnp->cn_nameiop != LOOKUP) {
614 if (cnp->cn_flags & AUDITVNODE1)
615 AUDIT_ARG_VNODE1(dp);
616 else if (cnp->cn_flags & AUDITVNODE2)
617 AUDIT_ARG_VNODE2(dp);
619 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
621 /* XXX This should probably move to the top of function. */
622 if (cnp->cn_flags & SAVESTART)
623 panic("lookup: SAVESTART");
628 * Handle "..": five special cases.
629 * 0. If doing a capability lookup, return ENOTCAPABLE (this is a
630 * fairly conservative design choice, but it's the only one that we
631 * are satisfied guarantees the property we're looking for).
632 * 1. Return an error if this is the last component of
633 * the name and the operation is DELETE or RENAME.
634 * 2. If at root directory (e.g. after chroot)
635 * or at absolute root directory
636 * then ignore it so can't get out.
637 * 3. If this vnode is the root of a mounted
638 * filesystem, then replace it with the
639 * vnode which was mounted on so we take the
640 * .. in the other filesystem.
641 * 4. If the vnode is the top directory of
642 * the jail or chroot, don't let them out.
644 if (cnp->cn_flags & ISDOTDOT) {
645 if (ndp->ni_strictrelative != 0) {
647 if (KTRPOINT(curthread, KTR_CAPFAIL))
648 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
653 if ((cnp->cn_flags & ISLASTCN) != 0 &&
654 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
659 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
661 if (dp == pr->pr_root)
663 if (dp == ndp->ni_rootdir ||
664 dp == ndp->ni_topdir ||
667 ((dp->v_vflag & VV_ROOT) != 0 &&
668 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
674 if ((dp->v_vflag & VV_ROOT) == 0)
676 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
681 dp = dp->v_mount->mnt_vnodecovered;
685 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
686 LK_RETRY, ISDOTDOT));
691 * We now have a segment name to search for, and a directory to search.
695 if ((cnp->cn_flags & NOMACCHECK) == 0) {
696 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
704 ASSERT_VOP_LOCKED(dp, "lookup");
706 * If we have a shared lock we may need to upgrade the lock for the
709 if (dp != vp_crossmp &&
710 VOP_ISLOCKED(dp) == LK_SHARED &&
711 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT))
712 vn_lock(dp, LK_UPGRADE|LK_RETRY);
713 if ((dp->v_iflag & VI_DOOMED) != 0) {
718 * If we're looking up the last component and we need an exclusive
719 * lock, adjust our lkflags.
721 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
722 cnp->cn_lkflags = LK_EXCLUSIVE;
723 #ifdef NAMEI_DIAGNOSTIC
724 vprint("lookup in", dp);
726 lkflags_save = cnp->cn_lkflags;
727 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
729 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
730 cnp->cn_lkflags = lkflags_save;
732 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
733 #ifdef NAMEI_DIAGNOSTIC
734 printf("not found\n");
736 if ((error == ENOENT) &&
737 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
738 (dp->v_mount->mnt_flag & MNT_UNION)) {
740 dp = dp->v_mount->mnt_vnodecovered;
744 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
745 LK_RETRY, cnp->cn_flags));
749 if (error == ERELOOKUP) {
757 if (error != EJUSTRETURN)
760 * At this point, we know we're at the end of the
761 * pathname. If creating / renaming, we can consider
762 * allowing the file or directory to be created / renamed,
763 * provided we're not on a read-only filesystem.
769 /* trailing slash only allowed for directories */
770 if ((cnp->cn_flags & TRAILINGSLASH) &&
771 !(cnp->cn_flags & WILLBEDIR)) {
775 if ((cnp->cn_flags & LOCKPARENT) == 0)
778 * We return with ni_vp NULL to indicate that the entry
779 * doesn't currently exist, leaving a pointer to the
780 * (possibly locked) directory vnode in ndp->ni_dvp.
782 if (cnp->cn_flags & SAVESTART) {
783 ndp->ni_startdir = ndp->ni_dvp;
784 VREF(ndp->ni_startdir);
790 #ifdef NAMEI_DIAGNOSTIC
796 * Check to see if the vnode has been mounted on;
797 * if so find the root of the mounted filesystem.
799 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
800 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
804 if (dp != ndp->ni_dvp)
809 ndp->ni_dvp = vp_crossmp;
810 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
811 cnp->cn_flags), &tdp);
813 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
814 panic("vp_crossmp exclusively locked or reclaimed");
819 ndp->ni_vp = dp = tdp;
823 * Check for symbolic link
825 if ((dp->v_type == VLNK) &&
826 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
827 *ndp->ni_next == '/')) {
828 cnp->cn_flags |= ISSYMLINK;
829 if (dp->v_iflag & VI_DOOMED) {
831 * We can't know whether the directory was mounted with
832 * NOSYMFOLLOW, so we can't follow safely.
837 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
842 * Symlink code always expects an unlocked dvp.
844 if (ndp->ni_dvp != ndp->ni_vp) {
845 VOP_UNLOCK(ndp->ni_dvp, 0);
853 * Not a symbolic link that we will follow. Continue with the
854 * next component if there is any; otherwise, we're done.
856 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
857 ("lookup: invalid path state."));
860 if (ndp->ni_dvp != dp)
866 if (*ndp->ni_next == '/') {
867 cnp->cn_nameptr = ndp->ni_next;
868 while (*cnp->cn_nameptr == '/') {
872 if (ndp->ni_dvp != dp)
879 * If we're processing a path with a trailing slash,
880 * check that the end result is a directory.
882 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
887 * Disallow directory write attempts on read-only filesystems.
890 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
894 if (cnp->cn_flags & SAVESTART) {
895 ndp->ni_startdir = ndp->ni_dvp;
896 VREF(ndp->ni_startdir);
900 if (ndp->ni_dvp != dp)
904 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
905 VOP_UNLOCK(ndp->ni_dvp, 0);
909 if (cnp->cn_flags & AUDITVNODE1)
910 AUDIT_ARG_VNODE1(dp);
911 else if (cnp->cn_flags & AUDITVNODE2)
912 AUDIT_ARG_VNODE2(dp);
914 if ((cnp->cn_flags & LOCKLEAF) == 0)
918 * Because of lookup_shared we may have the vnode shared locked, but
919 * the caller may want it to be exclusively locked.
921 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
922 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
923 vn_lock(dp, LK_UPGRADE | LK_RETRY);
924 if (dp->v_iflag & VI_DOOMED) {
932 if (ni_dvp_unlocked != 2) {
933 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
946 * relookup - lookup a path name component
947 * Used by lookup to re-acquire things.
950 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
952 struct vnode *dp = NULL; /* the directory we are searching */
953 int wantparent; /* 1 => wantparent or lockparent flag */
954 int rdonly; /* lookup read-only flag bit */
957 KASSERT(cnp->cn_flags & ISLASTCN,
958 ("relookup: Not given last component."));
960 * Setup: break out flag bits into variables.
962 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
963 KASSERT(wantparent, ("relookup: parent not wanted."));
964 rdonly = cnp->cn_flags & RDONLY;
965 cnp->cn_flags &= ~ISSYMLINK;
967 cnp->cn_lkflags = LK_EXCLUSIVE;
968 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
971 * Search a new directory.
973 * The last component of the filename is left accessible via
974 * cnp->cn_nameptr for callers that need the name. Callers needing
975 * the name set the SAVENAME flag. When done, they assume
976 * responsibility for freeing the pathname buffer.
978 #ifdef NAMEI_DIAGNOSTIC
979 printf("{%s}: ", cnp->cn_nameptr);
983 * Check for "" which represents the root directory after slash
986 if (cnp->cn_nameptr[0] == '\0') {
988 * Support only LOOKUP for "/" because lookup()
989 * can't succeed for CREATE, DELETE and RENAME.
991 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
992 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
994 if (!(cnp->cn_flags & LOCKLEAF))
997 /* XXX This should probably move to the top of function. */
998 if (cnp->cn_flags & SAVESTART)
999 panic("lookup: SAVESTART");
1003 if (cnp->cn_flags & ISDOTDOT)
1004 panic ("relookup: lookup on dot-dot");
1007 * We now have a segment name to search for, and a directory to search.
1009 #ifdef NAMEI_DIAGNOSTIC
1010 vprint("search in:", dp);
1012 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1013 KASSERT(*vpp == NULL, ("leaf should be empty"));
1014 if (error != EJUSTRETURN)
1017 * If creating and at end of pathname, then can consider
1018 * allowing file to be created.
1024 /* ASSERT(dvp == ndp->ni_startdir) */
1025 if (cnp->cn_flags & SAVESTART)
1027 if ((cnp->cn_flags & LOCKPARENT) == 0)
1030 * We return with ni_vp NULL to indicate that the entry
1031 * doesn't currently exist, leaving a pointer to the
1032 * (possibly locked) directory vnode in ndp->ni_dvp.
1040 * Disallow directory write attempts on read-only filesystems.
1043 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1052 * Set the parent lock/ref state to the requested state.
1054 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1059 } else if (!wantparent)
1062 * Check for symbolic link
1064 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1065 ("relookup: symlink found.\n"));
1067 /* ASSERT(dvp == ndp->ni_startdir) */
1068 if (cnp->cn_flags & SAVESTART)
1071 if ((cnp->cn_flags & LOCKLEAF) == 0)
1081 NDINIT_ALL(struct nameidata *ndp, u_long op, u_long flags, enum uio_seg segflg,
1082 const char *namep, int dirfd, struct vnode *startdir, cap_rights_t *rightsp,
1086 ndp->ni_cnd.cn_nameiop = op;
1087 ndp->ni_cnd.cn_flags = flags;
1088 ndp->ni_segflg = segflg;
1089 ndp->ni_dirp = namep;
1090 ndp->ni_dirfd = dirfd;
1091 ndp->ni_startdir = startdir;
1092 ndp->ni_strictrelative = 0;
1093 if (rightsp != NULL)
1094 ndp->ni_rightsneeded = *rightsp;
1096 cap_rights_init(&ndp->ni_rightsneeded);
1097 filecaps_init(&ndp->ni_filecaps);
1098 ndp->ni_cnd.cn_thread = td;
1102 * Free data allocated by namei(); see namei(9) for details.
1105 NDFREE(struct nameidata *ndp, const u_int flags)
1113 if (!(flags & NDF_NO_FREE_PNBUF) &&
1114 (ndp->ni_cnd.cn_flags & HASBUF)) {
1115 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1116 ndp->ni_cnd.cn_flags &= ~HASBUF;
1118 if (!(flags & NDF_NO_VP_UNLOCK) &&
1119 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1121 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1130 VOP_UNLOCK(ndp->ni_vp, 0);
1131 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1132 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1133 ndp->ni_dvp != ndp->ni_vp)
1135 if (!(flags & NDF_NO_DVP_RELE) &&
1136 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1145 VOP_UNLOCK(ndp->ni_dvp, 0);
1146 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1147 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1148 vrele(ndp->ni_startdir);
1149 ndp->ni_startdir = NULL;
1154 * Determine if there is a suitable alternate filename under the specified
1155 * prefix for the specified path. If the create flag is set, then the
1156 * alternate prefix will be used so long as the parent directory exists.
1157 * This is used by the various compatibility ABIs so that Linux binaries prefer
1158 * files under /compat/linux for example. The chosen path (whether under
1159 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1160 * to by pathbuf. The caller is responsible for free'ing the buffer from
1161 * the M_TEMP bucket if one is returned.
1164 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1165 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1167 struct nameidata nd, ndroot;
1168 char *ptr, *buf, *cp;
1172 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1175 /* Copy the prefix into the new pathname as a starting point. */
1176 len = strlcpy(buf, prefix, MAXPATHLEN);
1177 if (len >= MAXPATHLEN) {
1182 sz = MAXPATHLEN - len;
1185 /* Append the filename to the prefix. */
1186 if (pathseg == UIO_SYSSPACE)
1187 error = copystr(path, ptr, sz, &len);
1189 error = copyinstr(path, ptr, sz, &len);
1197 /* Only use a prefix with absolute pathnames. */
1203 if (dirfd != AT_FDCWD) {
1205 * We want the original because the "prefix" is
1206 * included in the already opened dirfd.
1208 bcopy(ptr, buf, len);
1213 * We know that there is a / somewhere in this pathname.
1214 * Search backwards for it, to find the file's parent dir
1215 * to see if it exists in the alternate tree. If it does,
1216 * and we want to create a file (cflag is set). We don't
1217 * need to worry about the root comparison in this case.
1221 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1224 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td);
1230 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td);
1237 * We now compare the vnode of the prefix to the one
1238 * vnode asked. If they resolve to be the same, then we
1239 * ignore the match so that the real root gets used.
1240 * This avoids the problem of traversing "../.." to find the
1241 * root directory and never finding it, because "/" resolves
1242 * to the emulation root directory. This is expensive :-(
1244 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1247 /* We shouldn't ever get an error from this namei(). */
1248 error = namei(&ndroot);
1250 if (nd.ni_vp == ndroot.ni_vp)
1253 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1254 vrele(ndroot.ni_vp);
1258 NDFREE(&nd, NDF_ONLY_PNBUF);
1262 /* If there was an error, use the original path name. */
1264 bcopy(ptr, buf, len);