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_kdtrace.h"
42 #include "opt_ktrace.h"
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/capability.h>
48 #include <sys/fcntl.h>
51 #include <sys/mutex.h>
52 #include <sys/namei.h>
53 #include <sys/vnode.h>
54 #include <sys/mount.h>
55 #include <sys/filedesc.h>
58 #include <sys/syscallsubr.h>
59 #include <sys/sysctl.h>
61 #include <sys/ktrace.h>
64 #include <security/audit/audit.h>
65 #include <security/mac/mac_framework.h>
69 #define NAMEI_DIAGNOSTIC 1
70 #undef NAMEI_DIAGNOSTIC
72 SDT_PROVIDER_DECLARE(vfs);
73 SDT_PROBE_DEFINE3(vfs, namei, lookup, entry, "struct vnode *", "char *",
75 SDT_PROBE_DEFINE2(vfs, namei, lookup, return, "int", "struct vnode *");
78 * Allocation zone for namei
80 uma_zone_t namei_zone;
82 * Placeholder vnode for mp traversal
84 static struct vnode *vp_crossmp;
87 nameiinit(void *dummy __unused)
90 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
92 getnewvnode("crossmp", NULL, &dead_vnodeops, &vp_crossmp);
93 vn_lock(vp_crossmp, LK_EXCLUSIVE);
94 VN_LOCK_ASHARE(vp_crossmp);
95 VOP_UNLOCK(vp_crossmp, 0);
97 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
99 static int lookup_shared = 1;
100 SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RW, &lookup_shared, 0,
101 "Enables/Disables shared locks for path name translation");
102 TUNABLE_INT("vfs.lookup_shared", &lookup_shared);
105 * Convert a pathname into a pointer to a locked vnode.
107 * The FOLLOW flag is set when symbolic links are to be followed
108 * when they occur at the end of the name translation process.
109 * Symbolic links are always followed for all other pathname
110 * components other than the last.
112 * The segflg defines whether the name is to be copied from user
113 * space or kernel space.
115 * Overall outline of namei:
118 * get starting directory
119 * while (!done && !error) {
120 * call lookup to search path.
121 * if symbolic link, massage name in buffer and continue
125 namei_cleanup_cnp(struct componentname *cnp)
127 uma_zfree(namei_zone, cnp->cn_pnbuf);
129 cnp->cn_pnbuf = NULL;
130 cnp->cn_nameptr = NULL;
135 namei(struct nameidata *ndp)
137 struct filedesc *fdp; /* pointer to file descriptor state */
138 char *cp; /* pointer into pathname argument */
139 struct vnode *dp; /* the directory we are searching */
140 struct iovec aiov; /* uio for reading symbolic links */
143 struct componentname *cnp = &ndp->ni_cnd;
144 struct thread *td = cnp->cn_thread;
145 struct proc *p = td->td_proc;
148 KASSERT((cnp->cn_flags & MPSAFE) != 0 || mtx_owned(&Giant) != 0,
149 ("NOT MPSAFE and Giant not held"));
150 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
151 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
152 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
153 ("namei: nameiop contaminated with flags"));
154 KASSERT((cnp->cn_flags & OPMASK) == 0,
155 ("namei: flags contaminated with nameiops"));
157 cnp->cn_flags &= ~LOCKSHARED;
160 /* We will set this ourselves if we need it. */
161 cnp->cn_flags &= ~TRAILINGSLASH;
164 * Get a buffer for the name to be translated, and copy the
165 * name into the buffer.
167 if ((cnp->cn_flags & HASBUF) == 0)
168 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
169 if (ndp->ni_segflg == UIO_SYSSPACE)
170 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
171 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
173 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
174 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
177 * Don't allow empty pathnames.
179 if (!error && *cnp->cn_pnbuf == '\0')
182 #ifdef CAPABILITY_MODE
184 * In capability mode, lookups must be "strictly relative" (i.e.
185 * not an absolute path, and not containing '..' components) to
186 * a real file descriptor, not the pseudo-descriptor AT_FDCWD.
188 if (error == 0 && IN_CAPABILITY_MODE(td)) {
189 ndp->ni_strictrelative = 1;
190 if (ndp->ni_dirfd == AT_FDCWD)
195 namei_cleanup_cnp(cnp);
201 if (KTRPOINT(td, KTR_NAMEI)) {
202 KASSERT(cnp->cn_thread == curthread,
203 ("namei not using curthread"));
204 ktrnamei(cnp->cn_pnbuf);
208 * Get starting point for the translation.
211 ndp->ni_rootdir = fdp->fd_rdir;
212 ndp->ni_topdir = fdp->fd_jdir;
215 * If we are auditing the kernel pathname, save the user pathname.
217 if (cnp->cn_flags & AUDITVNODE1)
218 AUDIT_ARG_UPATH1(td, ndp->ni_dirfd, cnp->cn_pnbuf);
219 if (cnp->cn_flags & AUDITVNODE2)
220 AUDIT_ARG_UPATH2(td, ndp->ni_dirfd, cnp->cn_pnbuf);
223 if (cnp->cn_pnbuf[0] != '/') {
224 if (ndp->ni_startdir != NULL) {
225 dp = ndp->ni_startdir;
227 } else if (ndp->ni_dirfd != AT_FDCWD) {
228 if (cnp->cn_flags & AUDITVNODE1)
229 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
230 if (cnp->cn_flags & AUDITVNODE2)
231 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
232 error = fgetvp_rights(td, ndp->ni_dirfd,
233 ndp->ni_rightsneeded | CAP_LOOKUP,
234 &(ndp->ni_baserights), &dp);
237 * Lookups relative to a capability must also be
240 * Note that a capability with rights CAP_MASK_VALID
241 * is treated exactly like a regular file descriptor.
243 if (ndp->ni_baserights != CAP_MASK_VALID)
244 ndp->ni_strictrelative = 1;
247 if (error != 0 || dp != NULL) {
248 FILEDESC_SUNLOCK(fdp);
249 if (error == 0 && dp->v_type != VDIR) {
250 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
252 VFS_UNLOCK_GIANT(vfslocked);
257 namei_cleanup_cnp(cnp);
264 FILEDESC_SUNLOCK(fdp);
265 if (ndp->ni_startdir != NULL) {
266 vfslocked = VFS_LOCK_GIANT(ndp->ni_startdir->v_mount);
267 vrele(ndp->ni_startdir);
268 VFS_UNLOCK_GIANT(vfslocked);
271 SDT_PROBE(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
272 cnp->cn_flags, 0, 0);
273 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
276 * Check if root directory should replace current directory.
277 * Done at start of translation and after symbolic link.
279 cnp->cn_nameptr = cnp->cn_pnbuf;
280 if (*(cnp->cn_nameptr) == '/') {
282 VFS_UNLOCK_GIANT(vfslocked);
283 if (ndp->ni_strictrelative != 0) {
284 namei_cleanup_cnp(cnp);
285 return (ENOTCAPABLE);
287 while (*(cnp->cn_nameptr) == '/') {
291 dp = ndp->ni_rootdir;
292 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
296 ndp->ni_cnd.cn_flags |= GIANTHELD;
297 ndp->ni_startdir = dp;
300 namei_cleanup_cnp(cnp);
301 SDT_PROBE(vfs, namei, lookup, return, error, NULL, 0,
305 vfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
306 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
308 * If not a symbolic link, we're done.
310 if ((cnp->cn_flags & ISSYMLINK) == 0) {
311 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
312 namei_cleanup_cnp(cnp);
314 cnp->cn_flags |= HASBUF;
316 if ((cnp->cn_flags & MPSAFE) == 0) {
317 VFS_UNLOCK_GIANT(vfslocked);
318 } else if (vfslocked)
319 ndp->ni_cnd.cn_flags |= GIANTHELD;
320 SDT_PROBE(vfs, namei, lookup, return, 0, ndp->ni_vp,
324 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
329 if ((cnp->cn_flags & NOMACCHECK) == 0) {
330 error = mac_vnode_check_readlink(td->td_ucred,
336 if (ndp->ni_pathlen > 1)
337 cp = uma_zalloc(namei_zone, M_WAITOK);
341 aiov.iov_len = MAXPATHLEN;
342 auio.uio_iov = &aiov;
345 auio.uio_rw = UIO_READ;
346 auio.uio_segflg = UIO_SYSSPACE;
348 auio.uio_resid = MAXPATHLEN;
349 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
351 if (ndp->ni_pathlen > 1)
352 uma_zfree(namei_zone, cp);
355 linklen = MAXPATHLEN - auio.uio_resid;
357 if (ndp->ni_pathlen > 1)
358 uma_zfree(namei_zone, cp);
362 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
363 if (ndp->ni_pathlen > 1)
364 uma_zfree(namei_zone, cp);
365 error = ENAMETOOLONG;
368 if (ndp->ni_pathlen > 1) {
369 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
370 uma_zfree(namei_zone, cnp->cn_pnbuf);
373 cnp->cn_pnbuf[linklen] = '\0';
374 ndp->ni_pathlen += linklen;
378 namei_cleanup_cnp(cnp);
382 VFS_UNLOCK_GIANT(vfslocked);
383 SDT_PROBE(vfs, namei, lookup, return, error, NULL, 0, 0, 0);
388 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
391 if (mp == NULL || ((lkflags & LK_SHARED) &&
392 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
393 ((cnflags & ISDOTDOT) &&
394 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
395 lkflags &= ~LK_SHARED;
396 lkflags |= LK_EXCLUSIVE;
402 needs_exclusive_leaf(struct mount *mp, int flags)
406 * Intermediate nodes can use shared locks, we only need to
407 * force an exclusive lock for leaf nodes.
409 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
412 /* Always use exclusive locks if LOCKSHARED isn't set. */
413 if (!(flags & LOCKSHARED))
417 * For lookups during open(), if the mount point supports
418 * extended shared operations, then use a shared lock for the
419 * leaf node, otherwise use an exclusive lock.
421 if (flags & ISOPEN) {
423 (mp->mnt_kern_flag & MNTK_EXTENDED_SHARED))
430 * Lookup requests outside of open() that specify LOCKSHARED
431 * only need a shared lock on the leaf vnode.
438 * This is a very central and rather complicated routine.
440 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
441 * The starting directory is taken from ni_startdir. The pathname is
442 * descended until done, or a symbolic link is encountered. The variable
443 * ni_more is clear if the path is completed; it is set to one if a
444 * symbolic link needing interpretation is encountered.
446 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
447 * whether the name is to be looked up, created, renamed, or deleted.
448 * When CREATE, RENAME, or DELETE is specified, information usable in
449 * creating, renaming, or deleting a directory entry may be calculated.
450 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
451 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
452 * returned unlocked. Otherwise the parent directory is not returned. If
453 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
454 * the target is returned locked, otherwise it is returned unlocked.
455 * When creating or renaming and LOCKPARENT is specified, the target may not
456 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
458 * Overall outline of lookup:
461 * identify next component of name at ndp->ni_ptr
462 * handle degenerate case where name is null string
463 * if .. and crossing mount points and on mounted filesys, find parent
464 * call VOP_LOOKUP routine for next component name
465 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
466 * component vnode returned in ni_vp (if it exists), locked.
467 * if result vnode is mounted on and crossing mount points,
468 * find mounted on vnode
469 * if more components of name, do next level at dirloop
470 * return the answer in ni_vp, locked if LOCKLEAF set
471 * if LOCKPARENT set, return locked parent in ni_dvp
472 * if WANTPARENT set, return unlocked parent in ni_dvp
475 lookup(struct nameidata *ndp)
477 char *cp; /* pointer into pathname argument */
478 struct vnode *dp = 0; /* the directory we are searching */
479 struct vnode *tdp; /* saved dp */
480 struct mount *mp; /* mount table entry */
482 int docache; /* == 0 do not cache last component */
483 int wantparent; /* 1 => wantparent or lockparent flag */
484 int rdonly; /* lookup read-only flag bit */
486 int dpunlocked = 0; /* dp has already been unlocked */
487 struct componentname *cnp = &ndp->ni_cnd;
488 int vfslocked; /* VFS Giant state for child */
489 int dvfslocked; /* VFS Giant state for parent */
495 * Setup: break out flag bits into variables.
497 dvfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
500 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
501 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
502 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
503 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
504 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
505 if (cnp->cn_nameiop == DELETE ||
506 (wantparent && cnp->cn_nameiop != CREATE &&
507 cnp->cn_nameiop != LOOKUP))
509 rdonly = cnp->cn_flags & RDONLY;
510 cnp->cn_flags &= ~ISSYMLINK;
513 * We use shared locks until we hit the parent of the last cn then
514 * we adjust based on the requesting flags.
517 cnp->cn_lkflags = LK_SHARED;
519 cnp->cn_lkflags = LK_EXCLUSIVE;
520 dp = ndp->ni_startdir;
521 ndp->ni_startdir = NULLVP;
523 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
528 * Search a new directory.
530 * The last component of the filename is left accessible via
531 * cnp->cn_nameptr for callers that need the name. Callers needing
532 * the name set the SAVENAME flag. When done, they assume
533 * responsibility for freeing the pathname buffer.
536 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
538 cnp->cn_namelen = cp - cnp->cn_nameptr;
539 if (cnp->cn_namelen > NAME_MAX) {
540 error = ENAMETOOLONG;
543 #ifdef NAMEI_DIAGNOSTIC
546 printf("{%s}: ", cnp->cn_nameptr);
549 ndp->ni_pathlen -= cnp->cn_namelen;
553 * Replace multiple slashes by a single slash and trailing slashes
554 * by a null. This must be done before VOP_LOOKUP() because some
555 * fs's don't know about trailing slashes. Remember if there were
556 * trailing slashes to handle symlinks, existing non-directories
557 * and non-existing files that won't be directories specially later.
559 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
563 *ndp->ni_next = '\0';
564 cnp->cn_flags |= TRAILINGSLASH;
569 cnp->cn_flags |= MAKEENTRY;
570 if (*cp == '\0' && docache == 0)
571 cnp->cn_flags &= ~MAKEENTRY;
572 if (cnp->cn_namelen == 2 &&
573 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
574 cnp->cn_flags |= ISDOTDOT;
576 cnp->cn_flags &= ~ISDOTDOT;
577 if (*ndp->ni_next == 0)
578 cnp->cn_flags |= ISLASTCN;
580 cnp->cn_flags &= ~ISLASTCN;
582 if ((cnp->cn_flags & ISLASTCN) != 0 &&
583 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
584 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
590 * Check for degenerate name (e.g. / or "")
591 * which is a way of talking about a directory,
592 * e.g. like "/." or ".".
594 if (cnp->cn_nameptr[0] == '\0') {
595 if (dp->v_type != VDIR) {
599 if (cnp->cn_nameiop != LOOKUP) {
609 if (cnp->cn_flags & AUDITVNODE1)
610 AUDIT_ARG_VNODE1(dp);
611 else if (cnp->cn_flags & AUDITVNODE2)
612 AUDIT_ARG_VNODE2(dp);
614 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
616 /* XXX This should probably move to the top of function. */
617 if (cnp->cn_flags & SAVESTART)
618 panic("lookup: SAVESTART");
623 * Handle "..": five special cases.
624 * 0. If doing a capability lookup, return ENOTCAPABLE (this is a
625 * fairly conservative design choice, but it's the only one that we
626 * are satisfied guarantees the property we're looking for).
627 * 1. Return an error if this is the last component of
628 * the name and the operation is DELETE or RENAME.
629 * 2. If at root directory (e.g. after chroot)
630 * or at absolute root directory
631 * then ignore it so can't get out.
632 * 3. If this vnode is the root of a mounted
633 * filesystem, then replace it with the
634 * vnode which was mounted on so we take the
635 * .. in the other filesystem.
636 * 4. If the vnode is the top directory of
637 * the jail or chroot, don't let them out.
639 if (cnp->cn_flags & ISDOTDOT) {
640 if (ndp->ni_strictrelative != 0) {
644 if ((cnp->cn_flags & ISLASTCN) != 0 &&
645 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
650 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
652 if (dp == pr->pr_root)
654 if (dp == ndp->ni_rootdir ||
655 dp == ndp->ni_topdir ||
658 ((dp->v_vflag & VV_ROOT) != 0 &&
659 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
662 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
666 if ((dp->v_vflag & VV_ROOT) == 0)
668 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
673 dp = dp->v_mount->mnt_vnodecovered;
674 tvfslocked = dvfslocked;
675 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
678 VFS_UNLOCK_GIANT(tvfslocked);
680 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
681 LK_RETRY, ISDOTDOT));
686 * We now have a segment name to search for, and a directory to search.
690 if ((cnp->cn_flags & NOMACCHECK) == 0) {
691 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
699 ASSERT_VOP_LOCKED(dp, "lookup");
700 VNASSERT(vfslocked == 0, dp, ("lookup: vfslocked %d", vfslocked));
702 * If we have a shared lock we may need to upgrade the lock for the
705 if (dp != vp_crossmp &&
706 VOP_ISLOCKED(dp) == LK_SHARED &&
707 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT))
708 vn_lock(dp, LK_UPGRADE|LK_RETRY);
709 if ((dp->v_iflag & VI_DOOMED) != 0) {
714 * If we're looking up the last component and we need an exclusive
715 * lock, adjust our lkflags.
717 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
718 cnp->cn_lkflags = LK_EXCLUSIVE;
719 #ifdef NAMEI_DIAGNOSTIC
720 vprint("lookup in", dp);
722 lkflags_save = cnp->cn_lkflags;
723 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
725 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
726 cnp->cn_lkflags = lkflags_save;
727 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
728 #ifdef NAMEI_DIAGNOSTIC
729 printf("not found\n");
731 if ((error == ENOENT) &&
732 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
733 (dp->v_mount->mnt_flag & MNT_UNION)) {
735 dp = dp->v_mount->mnt_vnodecovered;
736 tvfslocked = dvfslocked;
737 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
740 VFS_UNLOCK_GIANT(tvfslocked);
742 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
743 LK_RETRY, cnp->cn_flags));
747 if (error != EJUSTRETURN)
750 * At this point, we know we're at the end of the
751 * pathname. If creating / renaming, we can consider
752 * allowing the file or directory to be created / renamed,
753 * provided we're not on a read-only filesystem.
759 /* trailing slash only allowed for directories */
760 if ((cnp->cn_flags & TRAILINGSLASH) &&
761 !(cnp->cn_flags & WILLBEDIR)) {
765 if ((cnp->cn_flags & LOCKPARENT) == 0)
768 * We return with ni_vp NULL to indicate that the entry
769 * doesn't currently exist, leaving a pointer to the
770 * (possibly locked) directory vnode in ndp->ni_dvp.
772 if (cnp->cn_flags & SAVESTART) {
773 ndp->ni_startdir = ndp->ni_dvp;
774 VREF(ndp->ni_startdir);
778 cnp->cn_lkflags = lkflags_save;
779 #ifdef NAMEI_DIAGNOSTIC
783 * Take into account any additional components consumed by
784 * the underlying filesystem.
786 if (cnp->cn_consume > 0) {
787 cnp->cn_nameptr += cnp->cn_consume;
788 ndp->ni_next += cnp->cn_consume;
789 ndp->ni_pathlen -= cnp->cn_consume;
794 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
797 * Check to see if the vnode has been mounted on;
798 * if so find the root of the mounted filesystem.
800 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
801 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
805 VFS_UNLOCK_GIANT(vfslocked);
806 vfslocked = VFS_LOCK_GIANT(mp);
807 if (dp != ndp->ni_dvp)
811 VFS_UNLOCK_GIANT(dvfslocked);
814 ndp->ni_dvp = vp_crossmp;
815 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
816 cnp->cn_flags), &tdp);
818 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
819 panic("vp_crossmp exclusively locked or reclaimed");
824 ndp->ni_vp = dp = tdp;
828 * Check for symbolic link
830 if ((dp->v_type == VLNK) &&
831 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
832 *ndp->ni_next == '/')) {
833 cnp->cn_flags |= ISSYMLINK;
834 if (dp->v_iflag & VI_DOOMED) {
836 * We can't know whether the directory was mounted with
837 * NOSYMFOLLOW, so we can't follow safely.
842 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
847 * Symlink code always expects an unlocked dvp.
849 if (ndp->ni_dvp != ndp->ni_vp) {
850 VOP_UNLOCK(ndp->ni_dvp, 0);
858 * Not a symbolic link that we will follow. Continue with the
859 * next component if there is any; otherwise, we're done.
861 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
862 ("lookup: invalid path state."));
863 if (*ndp->ni_next == '/') {
864 cnp->cn_nameptr = ndp->ni_next;
865 while (*cnp->cn_nameptr == '/') {
869 if (ndp->ni_dvp != dp)
873 VFS_UNLOCK_GIANT(dvfslocked);
874 dvfslocked = vfslocked; /* dp becomes dvp in dirloop */
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 VFS_UNLOCK_GIANT(dvfslocked);
906 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
907 VOP_UNLOCK(ndp->ni_dvp, 0);
911 if (cnp->cn_flags & AUDITVNODE1)
912 AUDIT_ARG_VNODE1(dp);
913 else if (cnp->cn_flags & AUDITVNODE2)
914 AUDIT_ARG_VNODE2(dp);
916 if ((cnp->cn_flags & LOCKLEAF) == 0)
920 * Because of lookup_shared we may have the vnode shared locked, but
921 * the caller may want it to be exclusively locked.
923 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
924 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
925 vn_lock(dp, LK_UPGRADE | LK_RETRY);
926 if (dp->v_iflag & VI_DOOMED) {
931 if (vfslocked && dvfslocked)
932 VFS_UNLOCK_GIANT(dvfslocked); /* Only need one */
933 if (vfslocked || dvfslocked)
934 ndp->ni_cnd.cn_flags |= GIANTHELD;
938 if (ni_dvp_unlocked != 2) {
939 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
947 VFS_UNLOCK_GIANT(vfslocked);
948 VFS_UNLOCK_GIANT(dvfslocked);
949 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
955 * relookup - lookup a path name component
956 * Used by lookup to re-acquire things.
959 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
961 struct vnode *dp = 0; /* the directory we are searching */
962 int wantparent; /* 1 => wantparent or lockparent flag */
963 int rdonly; /* lookup read-only flag bit */
966 KASSERT(cnp->cn_flags & ISLASTCN,
967 ("relookup: Not given last component."));
969 * Setup: break out flag bits into variables.
971 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
972 KASSERT(wantparent, ("relookup: parent not wanted."));
973 rdonly = cnp->cn_flags & RDONLY;
974 cnp->cn_flags &= ~ISSYMLINK;
976 cnp->cn_lkflags = LK_EXCLUSIVE;
977 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
980 * Search a new directory.
982 * The last component of the filename is left accessible via
983 * cnp->cn_nameptr for callers that need the name. Callers needing
984 * the name set the SAVENAME flag. When done, they assume
985 * responsibility for freeing the pathname buffer.
987 #ifdef NAMEI_DIAGNOSTIC
988 printf("{%s}: ", cnp->cn_nameptr);
992 * Check for "" which represents the root directory after slash
995 if (cnp->cn_nameptr[0] == '\0') {
997 * Support only LOOKUP for "/" because lookup()
998 * can't succeed for CREATE, DELETE and RENAME.
1000 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1001 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1003 if (!(cnp->cn_flags & LOCKLEAF))
1006 /* XXX This should probably move to the top of function. */
1007 if (cnp->cn_flags & SAVESTART)
1008 panic("lookup: SAVESTART");
1012 if (cnp->cn_flags & ISDOTDOT)
1013 panic ("relookup: lookup on dot-dot");
1016 * We now have a segment name to search for, and a directory to search.
1018 #ifdef NAMEI_DIAGNOSTIC
1019 vprint("search in:", dp);
1021 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1022 KASSERT(*vpp == NULL, ("leaf should be empty"));
1023 if (error != EJUSTRETURN)
1026 * If creating and at end of pathname, then can consider
1027 * allowing file to be created.
1033 /* ASSERT(dvp == ndp->ni_startdir) */
1034 if (cnp->cn_flags & SAVESTART)
1036 if ((cnp->cn_flags & LOCKPARENT) == 0)
1039 * We return with ni_vp NULL to indicate that the entry
1040 * doesn't currently exist, leaving a pointer to the
1041 * (possibly locked) directory vnode in ndp->ni_dvp.
1049 * Disallow directory write attempts on read-only filesystems.
1052 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1061 * Set the parent lock/ref state to the requested state.
1063 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1068 } else if (!wantparent)
1071 * Check for symbolic link
1073 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1074 ("relookup: symlink found.\n"));
1076 /* ASSERT(dvp == ndp->ni_startdir) */
1077 if (cnp->cn_flags & SAVESTART)
1080 if ((cnp->cn_flags & LOCKLEAF) == 0)
1090 * Free data allocated by namei(); see namei(9) for details.
1093 NDFREE(struct nameidata *ndp, const u_int flags)
1101 if (!(flags & NDF_NO_FREE_PNBUF) &&
1102 (ndp->ni_cnd.cn_flags & HASBUF)) {
1103 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1104 ndp->ni_cnd.cn_flags &= ~HASBUF;
1106 if (!(flags & NDF_NO_VP_UNLOCK) &&
1107 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1109 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1118 VOP_UNLOCK(ndp->ni_vp, 0);
1119 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1120 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1121 ndp->ni_dvp != ndp->ni_vp)
1123 if (!(flags & NDF_NO_DVP_RELE) &&
1124 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1133 VOP_UNLOCK(ndp->ni_dvp, 0);
1134 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1135 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1136 vrele(ndp->ni_startdir);
1137 ndp->ni_startdir = NULL;
1142 * Determine if there is a suitable alternate filename under the specified
1143 * prefix for the specified path. If the create flag is set, then the
1144 * alternate prefix will be used so long as the parent directory exists.
1145 * This is used by the various compatiblity ABIs so that Linux binaries prefer
1146 * files under /compat/linux for example. The chosen path (whether under
1147 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1148 * to by pathbuf. The caller is responsible for free'ing the buffer from
1149 * the M_TEMP bucket if one is returned.
1152 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1153 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1155 struct nameidata nd, ndroot;
1156 char *ptr, *buf, *cp;
1160 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1163 /* Copy the prefix into the new pathname as a starting point. */
1164 len = strlcpy(buf, prefix, MAXPATHLEN);
1165 if (len >= MAXPATHLEN) {
1170 sz = MAXPATHLEN - len;
1173 /* Append the filename to the prefix. */
1174 if (pathseg == UIO_SYSSPACE)
1175 error = copystr(path, ptr, sz, &len);
1177 error = copyinstr(path, ptr, sz, &len);
1185 /* Only use a prefix with absolute pathnames. */
1191 if (dirfd != AT_FDCWD) {
1193 * We want the original because the "prefix" is
1194 * included in the already opened dirfd.
1196 bcopy(ptr, buf, len);
1201 * We know that there is a / somewhere in this pathname.
1202 * Search backwards for it, to find the file's parent dir
1203 * to see if it exists in the alternate tree. If it does,
1204 * and we want to create a file (cflag is set). We don't
1205 * need to worry about the root comparison in this case.
1209 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1212 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1218 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1225 * We now compare the vnode of the prefix to the one
1226 * vnode asked. If they resolve to be the same, then we
1227 * ignore the match so that the real root gets used.
1228 * This avoids the problem of traversing "../.." to find the
1229 * root directory and never finding it, because "/" resolves
1230 * to the emulation root directory. This is expensive :-(
1232 NDINIT(&ndroot, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, prefix,
1235 /* We shouldn't ever get an error from this namei(). */
1236 error = namei(&ndroot);
1238 if (nd.ni_vp == ndroot.ni_vp)
1241 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1242 vrele(ndroot.ni_vp);
1243 VFS_UNLOCK_GIANT(NDHASGIANT(&ndroot));
1247 NDFREE(&nd, NDF_ONLY_PNBUF);
1249 VFS_UNLOCK_GIANT(NDHASGIANT(&nd));
1252 /* If there was an error, use the original path name. */
1254 bcopy(ptr, buf, len);