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, entry, "struct vnode *", "char *",
75 SDT_PROBE_DEFINE2(vfs, namei, lookup, return, 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(struct nameidata *ndp)
127 struct filedesc *fdp; /* pointer to file descriptor state */
128 char *cp; /* pointer into pathname argument */
129 struct vnode *dp; /* the directory we are searching */
130 struct iovec aiov; /* uio for reading symbolic links */
133 struct componentname *cnp = &ndp->ni_cnd;
134 struct thread *td = cnp->cn_thread;
135 struct proc *p = td->td_proc;
138 KASSERT((cnp->cn_flags & MPSAFE) != 0 || mtx_owned(&Giant) != 0,
139 ("NOT MPSAFE and Giant not held"));
140 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
141 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
142 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
143 ("namei: nameiop contaminated with flags"));
144 KASSERT((cnp->cn_flags & OPMASK) == 0,
145 ("namei: flags contaminated with nameiops"));
147 cnp->cn_flags &= ~LOCKSHARED;
150 /* We will set this ourselves if we need it. */
151 cnp->cn_flags &= ~TRAILINGSLASH;
154 * Get a buffer for the name to be translated, and copy the
155 * name into the buffer.
157 if ((cnp->cn_flags & HASBUF) == 0)
158 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
159 if (ndp->ni_segflg == UIO_SYSSPACE)
160 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
161 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
163 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
164 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
168 * If we are auditing the kernel pathname, save the user
171 if (cnp->cn_flags & AUDITVNODE1)
172 AUDIT_ARG_UPATH1(td, cnp->cn_pnbuf);
173 if (cnp->cn_flags & AUDITVNODE2)
174 AUDIT_ARG_UPATH2(td, cnp->cn_pnbuf);
178 * Don't allow empty pathnames.
180 if (!error && *cnp->cn_pnbuf == '\0')
183 #ifdef CAPABILITY_MODE
185 * In capability mode, lookups must be "strictly relative" (i.e.
186 * not an absolute path, and not containing '..' components) to
187 * a real file descriptor, not the pseudo-descriptor AT_FDCWD.
189 if (IN_CAPABILITY_MODE(td)) {
190 ndp->ni_strictrelative = 1;
191 if (ndp->ni_dirfd == AT_FDCWD) {
193 if (KTRPOINT(td, KTR_CAPFAIL))
194 ktrcapfail(CAPFAIL_LOOKUP, 0, 0);
201 uma_zfree(namei_zone, cnp->cn_pnbuf);
203 cnp->cn_pnbuf = NULL;
204 cnp->cn_nameptr = NULL;
211 if (KTRPOINT(td, KTR_NAMEI)) {
212 KASSERT(cnp->cn_thread == curthread,
213 ("namei not using curthread"));
214 ktrnamei(cnp->cn_pnbuf);
218 * Get starting point for the translation.
221 ndp->ni_rootdir = fdp->fd_rdir;
222 ndp->ni_topdir = fdp->fd_jdir;
225 if (cnp->cn_pnbuf[0] != '/') {
226 if (ndp->ni_startdir != NULL) {
227 dp = ndp->ni_startdir;
229 } else if (ndp->ni_dirfd != AT_FDCWD) {
230 if (cnp->cn_flags & AUDITVNODE1)
231 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
232 if (cnp->cn_flags & AUDITVNODE2)
233 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
234 error = fgetvp_rights(td, ndp->ni_dirfd,
235 ndp->ni_rightsneeded | CAP_LOOKUP,
236 &(ndp->ni_baserights), &dp);
239 * Lookups relative to a capability must also be
242 * Note that a capability with rights CAP_MASK_VALID
243 * is treated exactly like a regular file descriptor.
245 if (ndp->ni_baserights != CAP_MASK_VALID)
246 ndp->ni_strictrelative = 1;
249 if (error != 0 || dp != NULL) {
250 FILEDESC_SUNLOCK(fdp);
251 if (error == 0 && dp->v_type != VDIR) {
252 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
254 VFS_UNLOCK_GIANT(vfslocked);
259 uma_zfree(namei_zone, cnp->cn_pnbuf);
261 cnp->cn_pnbuf = NULL;
262 cnp->cn_nameptr = NULL;
270 FILEDESC_SUNLOCK(fdp);
271 if (ndp->ni_startdir != NULL) {
272 vfslocked = VFS_LOCK_GIANT(ndp->ni_startdir->v_mount);
273 vrele(ndp->ni_startdir);
274 VFS_UNLOCK_GIANT(vfslocked);
277 SDT_PROBE(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
278 cnp->cn_flags, 0, 0);
279 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
282 * Check if root directory should replace current directory.
283 * Done at start of translation and after symbolic link.
285 cnp->cn_nameptr = cnp->cn_pnbuf;
286 if (*(cnp->cn_nameptr) == '/') {
288 VFS_UNLOCK_GIANT(vfslocked);
289 if (ndp->ni_strictrelative != 0) {
291 if (KTRPOINT(curthread, KTR_CAPFAIL))
292 ktrcapfail(CAPFAIL_LOOKUP, 0, 0);
294 return (ENOTCAPABLE);
296 while (*(cnp->cn_nameptr) == '/') {
300 dp = ndp->ni_rootdir;
301 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
305 ndp->ni_cnd.cn_flags |= GIANTHELD;
306 ndp->ni_startdir = dp;
309 uma_zfree(namei_zone, cnp->cn_pnbuf);
311 cnp->cn_pnbuf = NULL;
312 cnp->cn_nameptr = NULL;
314 SDT_PROBE(vfs, namei, lookup, return, error, NULL, 0,
318 vfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
319 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
321 * If not a symbolic link, we're done.
323 if ((cnp->cn_flags & ISSYMLINK) == 0) {
324 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
325 uma_zfree(namei_zone, cnp->cn_pnbuf);
327 cnp->cn_pnbuf = NULL;
328 cnp->cn_nameptr = NULL;
331 cnp->cn_flags |= HASBUF;
333 if ((cnp->cn_flags & MPSAFE) == 0) {
334 VFS_UNLOCK_GIANT(vfslocked);
335 } else if (vfslocked)
336 ndp->ni_cnd.cn_flags |= GIANTHELD;
337 SDT_PROBE(vfs, namei, lookup, return, 0, ndp->ni_vp,
341 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
346 if ((cnp->cn_flags & NOMACCHECK) == 0) {
347 error = mac_vnode_check_readlink(td->td_ucred,
353 if (ndp->ni_pathlen > 1)
354 cp = uma_zalloc(namei_zone, M_WAITOK);
358 aiov.iov_len = MAXPATHLEN;
359 auio.uio_iov = &aiov;
362 auio.uio_rw = UIO_READ;
363 auio.uio_segflg = UIO_SYSSPACE;
364 auio.uio_td = (struct thread *)0;
365 auio.uio_resid = MAXPATHLEN;
366 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
368 if (ndp->ni_pathlen > 1)
369 uma_zfree(namei_zone, cp);
372 linklen = MAXPATHLEN - auio.uio_resid;
374 if (ndp->ni_pathlen > 1)
375 uma_zfree(namei_zone, cp);
379 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
380 if (ndp->ni_pathlen > 1)
381 uma_zfree(namei_zone, cp);
382 error = ENAMETOOLONG;
385 if (ndp->ni_pathlen > 1) {
386 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
387 uma_zfree(namei_zone, cnp->cn_pnbuf);
390 cnp->cn_pnbuf[linklen] = '\0';
391 ndp->ni_pathlen += linklen;
395 uma_zfree(namei_zone, cnp->cn_pnbuf);
397 cnp->cn_pnbuf = NULL;
398 cnp->cn_nameptr = NULL;
403 VFS_UNLOCK_GIANT(vfslocked);
404 SDT_PROBE(vfs, namei, lookup, return, error, NULL, 0, 0, 0);
409 compute_cn_lkflags(struct mount *mp, int lkflags)
413 ((lkflags & LK_SHARED) && !(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED))) {
414 lkflags &= ~LK_SHARED;
415 lkflags |= LK_EXCLUSIVE;
421 needs_exclusive_leaf(struct mount *mp, int flags)
425 * Intermediate nodes can use shared locks, we only need to
426 * force an exclusive lock for leaf nodes.
428 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
431 /* Always use exclusive locks if LOCKSHARED isn't set. */
432 if (!(flags & LOCKSHARED))
436 * For lookups during open(), if the mount point supports
437 * extended shared operations, then use a shared lock for the
438 * leaf node, otherwise use an exclusive lock.
440 if (flags & ISOPEN) {
442 (mp->mnt_kern_flag & MNTK_EXTENDED_SHARED))
449 * Lookup requests outside of open() that specify LOCKSHARED
450 * only need a shared lock on the leaf vnode.
457 * This is a very central and rather complicated routine.
459 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
460 * The starting directory is taken from ni_startdir. The pathname is
461 * descended until done, or a symbolic link is encountered. The variable
462 * ni_more is clear if the path is completed; it is set to one if a
463 * symbolic link needing interpretation is encountered.
465 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
466 * whether the name is to be looked up, created, renamed, or deleted.
467 * When CREATE, RENAME, or DELETE is specified, information usable in
468 * creating, renaming, or deleting a directory entry may be calculated.
469 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
470 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
471 * returned unlocked. Otherwise the parent directory is not returned. If
472 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
473 * the target is returned locked, otherwise it is returned unlocked.
474 * When creating or renaming and LOCKPARENT is specified, the target may not
475 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
477 * Overall outline of lookup:
480 * identify next component of name at ndp->ni_ptr
481 * handle degenerate case where name is null string
482 * if .. and crossing mount points and on mounted filesys, find parent
483 * call VOP_LOOKUP routine for next component name
484 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
485 * component vnode returned in ni_vp (if it exists), locked.
486 * if result vnode is mounted on and crossing mount points,
487 * find mounted on vnode
488 * if more components of name, do next level at dirloop
489 * return the answer in ni_vp, locked if LOCKLEAF set
490 * if LOCKPARENT set, return locked parent in ni_dvp
491 * if WANTPARENT set, return unlocked parent in ni_dvp
494 lookup(struct nameidata *ndp)
496 char *cp; /* pointer into pathname argument */
497 struct vnode *dp = 0; /* the directory we are searching */
498 struct vnode *tdp; /* saved dp */
499 struct mount *mp; /* mount table entry */
501 int docache; /* == 0 do not cache last component */
502 int wantparent; /* 1 => wantparent or lockparent flag */
503 int rdonly; /* lookup read-only flag bit */
505 int dpunlocked = 0; /* dp has already been unlocked */
506 struct componentname *cnp = &ndp->ni_cnd;
507 int vfslocked; /* VFS Giant state for child */
508 int dvfslocked; /* VFS Giant state for parent */
514 * Setup: break out flag bits into variables.
516 dvfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
519 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
520 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
521 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
522 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
523 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
524 if (cnp->cn_nameiop == DELETE ||
525 (wantparent && cnp->cn_nameiop != CREATE &&
526 cnp->cn_nameiop != LOOKUP))
528 rdonly = cnp->cn_flags & RDONLY;
529 cnp->cn_flags &= ~ISSYMLINK;
532 * We use shared locks until we hit the parent of the last cn then
533 * we adjust based on the requesting flags.
536 cnp->cn_lkflags = LK_SHARED;
538 cnp->cn_lkflags = LK_EXCLUSIVE;
539 dp = ndp->ni_startdir;
540 ndp->ni_startdir = NULLVP;
542 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY));
546 * Search a new directory.
548 * The last component of the filename is left accessible via
549 * cnp->cn_nameptr for callers that need the name. Callers needing
550 * the name set the SAVENAME flag. When done, they assume
551 * responsibility for freeing the pathname buffer.
554 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
556 cnp->cn_namelen = cp - cnp->cn_nameptr;
557 if (cnp->cn_namelen > NAME_MAX) {
558 error = ENAMETOOLONG;
561 #ifdef NAMEI_DIAGNOSTIC
564 printf("{%s}: ", cnp->cn_nameptr);
567 ndp->ni_pathlen -= cnp->cn_namelen;
571 * Replace multiple slashes by a single slash and trailing slashes
572 * by a null. This must be done before VOP_LOOKUP() because some
573 * fs's don't know about trailing slashes. Remember if there were
574 * trailing slashes to handle symlinks, existing non-directories
575 * and non-existing files that won't be directories specially later.
577 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
581 *ndp->ni_next = '\0';
582 cnp->cn_flags |= TRAILINGSLASH;
587 cnp->cn_flags |= MAKEENTRY;
588 if (*cp == '\0' && docache == 0)
589 cnp->cn_flags &= ~MAKEENTRY;
590 if (cnp->cn_namelen == 2 &&
591 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
592 cnp->cn_flags |= ISDOTDOT;
594 cnp->cn_flags &= ~ISDOTDOT;
595 if (*ndp->ni_next == 0)
596 cnp->cn_flags |= ISLASTCN;
598 cnp->cn_flags &= ~ISLASTCN;
600 if ((cnp->cn_flags & ISLASTCN) != 0 &&
601 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
602 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
608 * Check for degenerate name (e.g. / or "")
609 * which is a way of talking about a directory,
610 * e.g. like "/." or ".".
612 if (cnp->cn_nameptr[0] == '\0') {
613 if (dp->v_type != VDIR) {
617 if (cnp->cn_nameiop != LOOKUP) {
627 if (cnp->cn_flags & AUDITVNODE1)
628 AUDIT_ARG_VNODE1(dp);
629 else if (cnp->cn_flags & AUDITVNODE2)
630 AUDIT_ARG_VNODE2(dp);
632 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
634 /* XXX This should probably move to the top of function. */
635 if (cnp->cn_flags & SAVESTART)
636 panic("lookup: SAVESTART");
641 * Handle "..": five special cases.
642 * 0. If doing a capability lookup, return ENOTCAPABLE (this is a
643 * fairly conservative design choice, but it's the only one that we
644 * are satisfied guarantees the property we're looking for).
645 * 1. Return an error if this is the last component of
646 * the name and the operation is DELETE or RENAME.
647 * 2. If at root directory (e.g. after chroot)
648 * or at absolute root directory
649 * then ignore it so can't get out.
650 * 3. If this vnode is the root of a mounted
651 * filesystem, then replace it with the
652 * vnode which was mounted on so we take the
653 * .. in the other filesystem.
654 * 4. If the vnode is the top directory of
655 * the jail or chroot, don't let them out.
657 if (cnp->cn_flags & ISDOTDOT) {
658 if (ndp->ni_strictrelative != 0) {
660 if (KTRPOINT(curthread, KTR_CAPFAIL))
661 ktrcapfail(CAPFAIL_LOOKUP, 0, 0);
666 if ((cnp->cn_flags & ISLASTCN) != 0 &&
667 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
672 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
674 if (dp == pr->pr_root)
676 if (dp == ndp->ni_rootdir ||
677 dp == ndp->ni_topdir ||
680 ((dp->v_vflag & VV_ROOT) != 0 &&
681 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
684 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
688 if ((dp->v_vflag & VV_ROOT) == 0)
690 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
695 dp = dp->v_mount->mnt_vnodecovered;
696 tvfslocked = dvfslocked;
697 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
700 VFS_UNLOCK_GIANT(tvfslocked);
702 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
708 * We now have a segment name to search for, and a directory to search.
712 if ((cnp->cn_flags & NOMACCHECK) == 0) {
713 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
721 ASSERT_VOP_LOCKED(dp, "lookup");
722 VNASSERT(vfslocked == 0, dp, ("lookup: vfslocked %d", vfslocked));
724 * If we have a shared lock we may need to upgrade the lock for the
727 if (dp != vp_crossmp &&
728 VOP_ISLOCKED(dp) == LK_SHARED &&
729 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT))
730 vn_lock(dp, LK_UPGRADE|LK_RETRY);
732 * If we're looking up the last component and we need an exclusive
733 * lock, adjust our lkflags.
735 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
736 cnp->cn_lkflags = LK_EXCLUSIVE;
737 #ifdef NAMEI_DIAGNOSTIC
738 vprint("lookup in", dp);
740 lkflags_save = cnp->cn_lkflags;
741 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags);
742 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
743 cnp->cn_lkflags = lkflags_save;
744 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
745 #ifdef NAMEI_DIAGNOSTIC
746 printf("not found\n");
748 if ((error == ENOENT) &&
749 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
750 (dp->v_mount->mnt_flag & MNT_UNION)) {
752 dp = dp->v_mount->mnt_vnodecovered;
753 tvfslocked = dvfslocked;
754 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
757 VFS_UNLOCK_GIANT(tvfslocked);
759 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
764 if (error != EJUSTRETURN)
767 * At this point, we know we're at the end of the
768 * pathname. If creating / renaming, we can consider
769 * allowing the file or directory to be created / renamed,
770 * provided we're not on a read-only filesystem.
776 /* trailing slash only allowed for directories */
777 if ((cnp->cn_flags & TRAILINGSLASH) &&
778 !(cnp->cn_flags & WILLBEDIR)) {
782 if ((cnp->cn_flags & LOCKPARENT) == 0)
785 * We return with ni_vp NULL to indicate that the entry
786 * doesn't currently exist, leaving a pointer to the
787 * (possibly locked) directory vnode in ndp->ni_dvp.
789 if (cnp->cn_flags & SAVESTART) {
790 ndp->ni_startdir = ndp->ni_dvp;
791 VREF(ndp->ni_startdir);
795 cnp->cn_lkflags = lkflags_save;
796 #ifdef NAMEI_DIAGNOSTIC
800 * Take into account any additional components consumed by
801 * the underlying filesystem.
803 if (cnp->cn_consume > 0) {
804 cnp->cn_nameptr += cnp->cn_consume;
805 ndp->ni_next += cnp->cn_consume;
806 ndp->ni_pathlen -= cnp->cn_consume;
811 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
814 * Check to see if the vnode has been mounted on;
815 * if so find the root of the mounted filesystem.
817 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
818 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
822 VFS_UNLOCK_GIANT(vfslocked);
823 vfslocked = VFS_LOCK_GIANT(mp);
824 if (dp != ndp->ni_dvp)
828 VFS_UNLOCK_GIANT(dvfslocked);
831 ndp->ni_dvp = vp_crossmp;
832 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags),
835 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
836 panic("vp_crossmp exclusively locked or reclaimed");
841 ndp->ni_vp = dp = tdp;
845 * Check for symbolic link
847 if ((dp->v_type == VLNK) &&
848 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
849 *ndp->ni_next == '/')) {
850 cnp->cn_flags |= ISSYMLINK;
851 if (dp->v_iflag & VI_DOOMED) {
853 * We can't know whether the directory was mounted with
854 * NOSYMFOLLOW, so we can't follow safely.
859 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
864 * Symlink code always expects an unlocked dvp.
866 if (ndp->ni_dvp != ndp->ni_vp) {
867 VOP_UNLOCK(ndp->ni_dvp, 0);
875 * Not a symbolic link that we will follow. Continue with the
876 * next component if there is any; otherwise, we're done.
878 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
879 ("lookup: invalid path state."));
880 if (*ndp->ni_next == '/') {
881 cnp->cn_nameptr = ndp->ni_next;
882 while (*cnp->cn_nameptr == '/') {
886 if (ndp->ni_dvp != dp)
890 VFS_UNLOCK_GIANT(dvfslocked);
891 dvfslocked = vfslocked; /* dp becomes dvp in dirloop */
896 * If we're processing a path with a trailing slash,
897 * check that the end result is a directory.
899 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
904 * Disallow directory write attempts on read-only filesystems.
907 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
911 if (cnp->cn_flags & SAVESTART) {
912 ndp->ni_startdir = ndp->ni_dvp;
913 VREF(ndp->ni_startdir);
917 if (ndp->ni_dvp != dp)
921 VFS_UNLOCK_GIANT(dvfslocked);
923 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
924 VOP_UNLOCK(ndp->ni_dvp, 0);
928 if (cnp->cn_flags & AUDITVNODE1)
929 AUDIT_ARG_VNODE1(dp);
930 else if (cnp->cn_flags & AUDITVNODE2)
931 AUDIT_ARG_VNODE2(dp);
933 if ((cnp->cn_flags & LOCKLEAF) == 0)
937 * Because of lookup_shared we may have the vnode shared locked, but
938 * the caller may want it to be exclusively locked.
940 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
941 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
942 vn_lock(dp, LK_UPGRADE | LK_RETRY);
943 if (dp->v_iflag & VI_DOOMED) {
948 if (vfslocked && dvfslocked)
949 VFS_UNLOCK_GIANT(dvfslocked); /* Only need one */
950 if (vfslocked || dvfslocked)
951 ndp->ni_cnd.cn_flags |= GIANTHELD;
955 if (ni_dvp_unlocked != 2) {
956 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
964 VFS_UNLOCK_GIANT(vfslocked);
965 VFS_UNLOCK_GIANT(dvfslocked);
966 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
972 * relookup - lookup a path name component
973 * Used by lookup to re-acquire things.
976 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
978 struct vnode *dp = 0; /* the directory we are searching */
979 int wantparent; /* 1 => wantparent or lockparent flag */
980 int rdonly; /* lookup read-only flag bit */
983 KASSERT(cnp->cn_flags & ISLASTCN,
984 ("relookup: Not given last component."));
986 * Setup: break out flag bits into variables.
988 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
989 KASSERT(wantparent, ("relookup: parent not wanted."));
990 rdonly = cnp->cn_flags & RDONLY;
991 cnp->cn_flags &= ~ISSYMLINK;
993 cnp->cn_lkflags = LK_EXCLUSIVE;
994 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
997 * Search a new directory.
999 * The last component of the filename is left accessible via
1000 * cnp->cn_nameptr for callers that need the name. Callers needing
1001 * the name set the SAVENAME flag. When done, they assume
1002 * responsibility for freeing the pathname buffer.
1004 #ifdef NAMEI_DIAGNOSTIC
1005 printf("{%s}: ", cnp->cn_nameptr);
1009 * Check for "" which represents the root directory after slash
1012 if (cnp->cn_nameptr[0] == '\0') {
1014 * Support only LOOKUP for "/" because lookup()
1015 * can't succeed for CREATE, DELETE and RENAME.
1017 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1018 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1020 if (!(cnp->cn_flags & LOCKLEAF))
1023 /* XXX This should probably move to the top of function. */
1024 if (cnp->cn_flags & SAVESTART)
1025 panic("lookup: SAVESTART");
1029 if (cnp->cn_flags & ISDOTDOT)
1030 panic ("relookup: lookup on dot-dot");
1033 * We now have a segment name to search for, and a directory to search.
1035 #ifdef NAMEI_DIAGNOSTIC
1036 vprint("search in:", dp);
1038 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1039 KASSERT(*vpp == NULL, ("leaf should be empty"));
1040 if (error != EJUSTRETURN)
1043 * If creating and at end of pathname, then can consider
1044 * allowing file to be created.
1050 /* ASSERT(dvp == ndp->ni_startdir) */
1051 if (cnp->cn_flags & SAVESTART)
1053 if ((cnp->cn_flags & LOCKPARENT) == 0)
1056 * We return with ni_vp NULL to indicate that the entry
1057 * doesn't currently exist, leaving a pointer to the
1058 * (possibly locked) directory vnode in ndp->ni_dvp.
1066 * Disallow directory write attempts on read-only filesystems.
1069 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1078 * Set the parent lock/ref state to the requested state.
1080 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1085 } else if (!wantparent)
1088 * Check for symbolic link
1090 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1091 ("relookup: symlink found.\n"));
1093 /* ASSERT(dvp == ndp->ni_startdir) */
1094 if (cnp->cn_flags & SAVESTART)
1097 if ((cnp->cn_flags & LOCKLEAF) == 0)
1107 * Free data allocated by namei(); see namei(9) for details.
1110 NDFREE(struct nameidata *ndp, const u_int flags)
1118 if (!(flags & NDF_NO_FREE_PNBUF) &&
1119 (ndp->ni_cnd.cn_flags & HASBUF)) {
1120 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1121 ndp->ni_cnd.cn_flags &= ~HASBUF;
1123 if (!(flags & NDF_NO_VP_UNLOCK) &&
1124 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1126 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1135 VOP_UNLOCK(ndp->ni_vp, 0);
1136 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1137 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1138 ndp->ni_dvp != ndp->ni_vp)
1140 if (!(flags & NDF_NO_DVP_RELE) &&
1141 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1150 VOP_UNLOCK(ndp->ni_dvp, 0);
1151 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1152 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1153 vrele(ndp->ni_startdir);
1154 ndp->ni_startdir = NULL;
1159 * Determine if there is a suitable alternate filename under the specified
1160 * prefix for the specified path. If the create flag is set, then the
1161 * alternate prefix will be used so long as the parent directory exists.
1162 * This is used by the various compatiblity ABIs so that Linux binaries prefer
1163 * files under /compat/linux for example. The chosen path (whether under
1164 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1165 * to by pathbuf. The caller is responsible for free'ing the buffer from
1166 * the M_TEMP bucket if one is returned.
1169 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1170 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1172 struct nameidata nd, ndroot;
1173 char *ptr, *buf, *cp;
1177 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1180 /* Copy the prefix into the new pathname as a starting point. */
1181 len = strlcpy(buf, prefix, MAXPATHLEN);
1182 if (len >= MAXPATHLEN) {
1187 sz = MAXPATHLEN - len;
1190 /* Append the filename to the prefix. */
1191 if (pathseg == UIO_SYSSPACE)
1192 error = copystr(path, ptr, sz, &len);
1194 error = copyinstr(path, ptr, sz, &len);
1202 /* Only use a prefix with absolute pathnames. */
1208 if (dirfd != AT_FDCWD) {
1210 * We want the original because the "prefix" is
1211 * included in the already opened dirfd.
1213 bcopy(ptr, buf, len);
1218 * We know that there is a / somewhere in this pathname.
1219 * Search backwards for it, to find the file's parent dir
1220 * to see if it exists in the alternate tree. If it does,
1221 * and we want to create a file (cflag is set). We don't
1222 * need to worry about the root comparison in this case.
1226 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1229 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1235 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1242 * We now compare the vnode of the prefix to the one
1243 * vnode asked. If they resolve to be the same, then we
1244 * ignore the match so that the real root gets used.
1245 * This avoids the problem of traversing "../.." to find the
1246 * root directory and never finding it, because "/" resolves
1247 * to the emulation root directory. This is expensive :-(
1249 NDINIT(&ndroot, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, prefix,
1252 /* We shouldn't ever get an error from this namei(). */
1253 error = namei(&ndroot);
1255 if (nd.ni_vp == ndroot.ni_vp)
1258 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1259 vrele(ndroot.ni_vp);
1260 VFS_UNLOCK_GIANT(NDHASGIANT(&ndroot));
1264 NDFREE(&nd, NDF_ONLY_PNBUF);
1266 VFS_UNLOCK_GIANT(NDHASGIANT(&nd));
1269 /* If there was an error, use the original path name. */
1271 bcopy(ptr, buf, len);