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_ktrace.h"
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/kernel.h>
47 #include <sys/fcntl.h>
49 #include <sys/mutex.h>
50 #include <sys/namei.h>
51 #include <sys/vnode.h>
52 #include <sys/mount.h>
53 #include <sys/filedesc.h>
55 #include <sys/syscallsubr.h>
56 #include <sys/sysctl.h>
58 #include <sys/ktrace.h>
61 #include <security/audit/audit.h>
62 #include <security/mac/mac_framework.h>
66 #define NAMEI_DIAGNOSTIC 1
67 #undef NAMEI_DIAGNOSTIC
70 * Allocation zone for namei
72 uma_zone_t namei_zone;
74 * Placeholder vnode for mp traversal
76 static struct vnode *vp_crossmp;
79 nameiinit(void *dummy __unused)
83 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
85 error = getnewvnode("crossmp", NULL, &dead_vnodeops, &vp_crossmp);
87 panic("nameiinit: getnewvnode");
88 VN_LOCK_ASHARE(vp_crossmp);
90 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
93 static int lookup_shared = 1;
95 static int lookup_shared = 0;
97 SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RW, &lookup_shared, 0,
98 "Enables/Disables shared locks for path name translation");
101 * Convert a pathname into a pointer to a locked vnode.
103 * The FOLLOW flag is set when symbolic links are to be followed
104 * when they occur at the end of the name translation process.
105 * Symbolic links are always followed for all other pathname
106 * components other than the last.
108 * The segflg defines whether the name is to be copied from user
109 * space or kernel space.
111 * Overall outline of namei:
114 * get starting directory
115 * while (!done && !error) {
116 * call lookup to search path.
117 * if symbolic link, massage name in buffer and continue
121 namei(struct nameidata *ndp)
123 struct filedesc *fdp; /* pointer to file descriptor state */
124 char *cp; /* pointer into pathname argument */
125 struct vnode *dp; /* the directory we are searching */
126 struct iovec aiov; /* uio for reading symbolic links */
129 struct componentname *cnp = &ndp->ni_cnd;
130 struct thread *td = cnp->cn_thread;
131 struct proc *p = td->td_proc;
134 KASSERT((cnp->cn_flags & MPSAFE) != 0 || mtx_owned(&Giant) != 0,
135 ("NOT MPSAFE and Giant not held"));
136 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
137 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
138 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
139 ("namei: nameiop contaminated with flags"));
140 KASSERT((cnp->cn_flags & OPMASK) == 0,
141 ("namei: flags contaminated with nameiops"));
143 cnp->cn_flags &= ~LOCKSHARED;
147 * Get a buffer for the name to be translated, and copy the
148 * name into the buffer.
150 if ((cnp->cn_flags & HASBUF) == 0)
151 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
152 if (ndp->ni_segflg == UIO_SYSSPACE)
153 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
154 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
156 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
157 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
159 /* If we are auditing the kernel pathname, save the user pathname. */
160 if (cnp->cn_flags & AUDITVNODE1)
161 AUDIT_ARG(upath, td, cnp->cn_pnbuf, ARG_UPATH1);
162 if (cnp->cn_flags & AUDITVNODE2)
163 AUDIT_ARG(upath, td, cnp->cn_pnbuf, ARG_UPATH2);
166 * Don't allow empty pathnames.
168 if (!error && *cnp->cn_pnbuf == '\0')
172 uma_zfree(namei_zone, cnp->cn_pnbuf);
174 cnp->cn_pnbuf = NULL;
175 cnp->cn_nameptr = NULL;
182 if (KTRPOINT(td, KTR_NAMEI)) {
183 KASSERT(cnp->cn_thread == curthread,
184 ("namei not using curthread"));
185 ktrnamei(cnp->cn_pnbuf);
190 * Get starting point for the translation.
193 ndp->ni_rootdir = fdp->fd_rdir;
194 ndp->ni_topdir = fdp->fd_jdir;
196 if (cnp->cn_pnbuf[0] != '/' && ndp->ni_dirfd != AT_FDCWD) {
197 error = fgetvp(td, ndp->ni_dirfd, &dp);
198 FILEDESC_SUNLOCK(fdp);
199 if (error == 0 && dp->v_type != VDIR) {
200 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
202 VFS_UNLOCK_GIANT(vfslocked);
206 uma_zfree(namei_zone, cnp->cn_pnbuf);
208 cnp->cn_pnbuf = NULL;
209 cnp->cn_nameptr = NULL;
216 FILEDESC_SUNLOCK(fdp);
218 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
221 * Check if root directory should replace current directory.
222 * Done at start of translation and after symbolic link.
224 cnp->cn_nameptr = cnp->cn_pnbuf;
225 if (*(cnp->cn_nameptr) == '/') {
227 VFS_UNLOCK_GIANT(vfslocked);
228 while (*(cnp->cn_nameptr) == '/') {
232 dp = ndp->ni_rootdir;
233 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
237 ndp->ni_cnd.cn_flags |= GIANTHELD;
238 ndp->ni_startdir = dp;
241 uma_zfree(namei_zone, cnp->cn_pnbuf);
243 cnp->cn_pnbuf = NULL;
244 cnp->cn_nameptr = NULL;
248 vfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
249 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
251 * Check for symbolic link
253 if ((cnp->cn_flags & ISSYMLINK) == 0) {
254 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
255 uma_zfree(namei_zone, cnp->cn_pnbuf);
257 cnp->cn_pnbuf = NULL;
258 cnp->cn_nameptr = NULL;
261 cnp->cn_flags |= HASBUF;
263 if ((cnp->cn_flags & MPSAFE) == 0) {
264 VFS_UNLOCK_GIANT(vfslocked);
265 } else if (vfslocked)
266 ndp->ni_cnd.cn_flags |= GIANTHELD;
269 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
274 if ((cnp->cn_flags & NOMACCHECK) == 0) {
275 error = mac_vnode_check_readlink(td->td_ucred,
281 if (ndp->ni_pathlen > 1)
282 cp = uma_zalloc(namei_zone, M_WAITOK);
286 aiov.iov_len = MAXPATHLEN;
287 auio.uio_iov = &aiov;
290 auio.uio_rw = UIO_READ;
291 auio.uio_segflg = UIO_SYSSPACE;
292 auio.uio_td = (struct thread *)0;
293 auio.uio_resid = MAXPATHLEN;
294 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
296 if (ndp->ni_pathlen > 1)
297 uma_zfree(namei_zone, cp);
300 linklen = MAXPATHLEN - auio.uio_resid;
302 if (ndp->ni_pathlen > 1)
303 uma_zfree(namei_zone, cp);
307 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
308 if (ndp->ni_pathlen > 1)
309 uma_zfree(namei_zone, cp);
310 error = ENAMETOOLONG;
313 if (ndp->ni_pathlen > 1) {
314 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
315 uma_zfree(namei_zone, cnp->cn_pnbuf);
318 cnp->cn_pnbuf[linklen] = '\0';
319 ndp->ni_pathlen += linklen;
323 uma_zfree(namei_zone, cnp->cn_pnbuf);
325 cnp->cn_pnbuf = NULL;
326 cnp->cn_nameptr = NULL;
331 VFS_UNLOCK_GIANT(vfslocked);
336 compute_cn_lkflags(struct mount *mp, int lkflags)
339 ((lkflags & LK_SHARED) && !(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED))) {
340 lkflags &= ~LK_SHARED;
341 lkflags |= LK_EXCLUSIVE;
348 * This is a very central and rather complicated routine.
350 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
351 * The starting directory is taken from ni_startdir. The pathname is
352 * descended until done, or a symbolic link is encountered. The variable
353 * ni_more is clear if the path is completed; it is set to one if a
354 * symbolic link needing interpretation is encountered.
356 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
357 * whether the name is to be looked up, created, renamed, or deleted.
358 * When CREATE, RENAME, or DELETE is specified, information usable in
359 * creating, renaming, or deleting a directory entry may be calculated.
360 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
361 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
362 * returned unlocked. Otherwise the parent directory is not returned. If
363 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
364 * the target is returned locked, otherwise it is returned unlocked.
365 * When creating or renaming and LOCKPARENT is specified, the target may not
366 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
368 * Overall outline of lookup:
371 * identify next component of name at ndp->ni_ptr
372 * handle degenerate case where name is null string
373 * if .. and crossing mount points and on mounted filesys, find parent
374 * call VOP_LOOKUP routine for next component name
375 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
376 * component vnode returned in ni_vp (if it exists), locked.
377 * if result vnode is mounted on and crossing mount points,
378 * find mounted on vnode
379 * if more components of name, do next level at dirloop
380 * return the answer in ni_vp, locked if LOCKLEAF set
381 * if LOCKPARENT set, return locked parent in ni_dvp
382 * if WANTPARENT set, return unlocked parent in ni_dvp
385 lookup(struct nameidata *ndp)
387 char *cp; /* pointer into pathname argument */
388 struct vnode *dp = 0; /* the directory we are searching */
389 struct vnode *tdp; /* saved dp */
390 struct mount *mp; /* mount table entry */
391 int docache; /* == 0 do not cache last component */
392 int wantparent; /* 1 => wantparent or lockparent flag */
393 int rdonly; /* lookup read-only flag bit */
396 int dpunlocked = 0; /* dp has already been unlocked */
397 struct componentname *cnp = &ndp->ni_cnd;
398 struct thread *td = cnp->cn_thread;
399 int vfslocked; /* VFS Giant state for child */
400 int dvfslocked; /* VFS Giant state for parent */
405 * Setup: break out flag bits into variables.
407 dvfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
409 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
410 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
411 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
412 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
413 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
414 if (cnp->cn_nameiop == DELETE ||
415 (wantparent && cnp->cn_nameiop != CREATE &&
416 cnp->cn_nameiop != LOOKUP))
418 rdonly = cnp->cn_flags & RDONLY;
419 cnp->cn_flags &= ~ISSYMLINK;
422 * We use shared locks until we hit the parent of the last cn then
423 * we adjust based on the requesting flags.
426 cnp->cn_lkflags = LK_SHARED;
428 cnp->cn_lkflags = LK_EXCLUSIVE;
429 dp = ndp->ni_startdir;
430 ndp->ni_startdir = NULLVP;
432 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY));
436 * Search a new directory.
438 * The last component of the filename is left accessible via
439 * cnp->cn_nameptr for callers that need the name. Callers needing
440 * the name set the SAVENAME flag. When done, they assume
441 * responsibility for freeing the pathname buffer.
444 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
446 cnp->cn_namelen = cp - cnp->cn_nameptr;
447 if (cnp->cn_namelen > NAME_MAX) {
448 error = ENAMETOOLONG;
451 #ifdef NAMEI_DIAGNOSTIC
454 printf("{%s}: ", cnp->cn_nameptr);
457 ndp->ni_pathlen -= cnp->cn_namelen;
461 * Replace multiple slashes by a single slash and trailing slashes
462 * by a null. This must be done before VOP_LOOKUP() because some
463 * fs's don't know about trailing slashes. Remember if there were
464 * trailing slashes to handle symlinks, existing non-directories
465 * and non-existing files that won't be directories specially later.
468 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
473 *ndp->ni_next = '\0'; /* XXX for direnter() ... */
478 cnp->cn_flags |= MAKEENTRY;
479 if (*cp == '\0' && docache == 0)
480 cnp->cn_flags &= ~MAKEENTRY;
481 if (cnp->cn_namelen == 2 &&
482 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
483 cnp->cn_flags |= ISDOTDOT;
485 cnp->cn_flags &= ~ISDOTDOT;
486 if (*ndp->ni_next == 0)
487 cnp->cn_flags |= ISLASTCN;
489 cnp->cn_flags &= ~ISLASTCN;
493 * Check for degenerate name (e.g. / or "")
494 * which is a way of talking about a directory,
495 * e.g. like "/." or ".".
497 if (cnp->cn_nameptr[0] == '\0') {
498 if (dp->v_type != VDIR) {
502 if (cnp->cn_nameiop != LOOKUP) {
512 if (cnp->cn_flags & AUDITVNODE1)
513 AUDIT_ARG(vnode, dp, ARG_VNODE1);
514 else if (cnp->cn_flags & AUDITVNODE2)
515 AUDIT_ARG(vnode, dp, ARG_VNODE2);
517 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
519 /* XXX This should probably move to the top of function. */
520 if (cnp->cn_flags & SAVESTART)
521 panic("lookup: SAVESTART");
526 * Handle "..": four special cases.
527 * 1. Return an error if this is the last component of
528 * the name and the operation is DELETE or RENAME.
529 * 2. If at root directory (e.g. after chroot)
530 * or at absolute root directory
531 * then ignore it so can't get out.
532 * 3. If this vnode is the root of a mounted
533 * filesystem, then replace it with the
534 * vnode which was mounted on so we take the
535 * .. in the other filesystem.
536 * 4. If the vnode is the top directory of
537 * the jail or chroot, don't let them out.
539 if (cnp->cn_flags & ISDOTDOT) {
540 if ((cnp->cn_flags & ISLASTCN) != 0 &&
541 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
546 if (dp == ndp->ni_rootdir ||
547 dp == ndp->ni_topdir ||
549 ((dp->v_vflag & VV_ROOT) != 0 &&
550 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
553 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
557 if ((dp->v_vflag & VV_ROOT) == 0)
559 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
564 dp = dp->v_mount->mnt_vnodecovered;
565 tvfslocked = dvfslocked;
566 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
569 VFS_UNLOCK_GIANT(tvfslocked);
571 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
577 * We now have a segment name to search for, and a directory to search.
581 if ((cnp->cn_flags & NOMACCHECK) == 0) {
582 error = mac_vnode_check_lookup(td->td_ucred, dp, cnp);
589 ASSERT_VOP_LOCKED(dp, "lookup");
590 VNASSERT(vfslocked == 0, dp, ("lookup: vfslocked %d", vfslocked));
592 * If we have a shared lock we may need to upgrade the lock for the
595 if (dp != vp_crossmp &&
596 VOP_ISLOCKED(dp) == LK_SHARED &&
597 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT))
598 vn_lock(dp, LK_UPGRADE|LK_RETRY);
600 * If we're looking up the last component and we need an exclusive
601 * lock, adjust our lkflags.
603 if ((cnp->cn_flags & (ISLASTCN|LOCKSHARED|LOCKLEAF)) ==
605 cnp->cn_lkflags = LK_EXCLUSIVE;
606 #ifdef NAMEI_DIAGNOSTIC
607 vprint("lookup in", dp);
609 lkflags_save = cnp->cn_lkflags;
610 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags);
611 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
612 cnp->cn_lkflags = lkflags_save;
613 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
614 #ifdef NAMEI_DIAGNOSTIC
615 printf("not found\n");
617 if ((error == ENOENT) &&
618 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
619 (dp->v_mount->mnt_flag & MNT_UNION)) {
621 dp = dp->v_mount->mnt_vnodecovered;
622 tvfslocked = dvfslocked;
623 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
626 VFS_UNLOCK_GIANT(tvfslocked);
628 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
633 if (error != EJUSTRETURN)
636 * If creating and at end of pathname, then can consider
637 * allowing file to be created.
643 if (*cp == '\0' && trailing_slash &&
644 !(cnp->cn_flags & WILLBEDIR)) {
648 if ((cnp->cn_flags & LOCKPARENT) == 0)
651 * This is a temporary assert to make sure I know what the
654 KASSERT((cnp->cn_flags & (WANTPARENT|LOCKPARENT)) != 0,
655 ("lookup: Unhandled case."));
657 * We return with ni_vp NULL to indicate that the entry
658 * doesn't currently exist, leaving a pointer to the
659 * (possibly locked) directory vnode in ndp->ni_dvp.
661 if (cnp->cn_flags & SAVESTART) {
662 ndp->ni_startdir = ndp->ni_dvp;
663 VREF(ndp->ni_startdir);
667 cnp->cn_lkflags = lkflags_save;
668 #ifdef NAMEI_DIAGNOSTIC
672 * Take into account any additional components consumed by
673 * the underlying filesystem.
675 if (cnp->cn_consume > 0) {
676 cnp->cn_nameptr += cnp->cn_consume;
677 ndp->ni_next += cnp->cn_consume;
678 ndp->ni_pathlen -= cnp->cn_consume;
683 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
686 * Check to see if the vnode has been mounted on;
687 * if so find the root of the mounted filesystem.
689 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
690 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
691 if (vfs_busy(mp, 0, 0, td))
694 VFS_UNLOCK_GIANT(vfslocked);
695 vfslocked = VFS_LOCK_GIANT(mp);
696 if (dp != ndp->ni_dvp)
700 VFS_UNLOCK_GIANT(dvfslocked);
703 ndp->ni_dvp = vp_crossmp;
704 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags), &tdp, td);
706 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
707 panic("vp_crossmp exclusively locked or reclaimed");
712 ndp->ni_vp = dp = tdp;
716 * Check for symbolic link
718 if ((dp->v_type == VLNK) &&
719 ((cnp->cn_flags & FOLLOW) || trailing_slash ||
720 *ndp->ni_next == '/')) {
721 cnp->cn_flags |= ISSYMLINK;
722 if (dp->v_iflag & VI_DOOMED) {
723 /* We can't know whether the directory was mounted with
724 * NOSYMFOLLOW, so we can't follow safely. */
728 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
733 * Symlink code always expects an unlocked dvp.
735 if (ndp->ni_dvp != ndp->ni_vp)
736 VOP_UNLOCK(ndp->ni_dvp, 0);
741 * Check for bogus trailing slashes.
743 if (trailing_slash && dp->v_type != VDIR) {
750 * Not a symbolic link. If more pathname,
751 * continue at next component, else return.
753 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
754 ("lookup: invalid path state."));
755 if (*ndp->ni_next == '/') {
756 cnp->cn_nameptr = ndp->ni_next;
757 while (*cnp->cn_nameptr == '/') {
761 if (ndp->ni_dvp != dp)
765 VFS_UNLOCK_GIANT(dvfslocked);
766 dvfslocked = vfslocked; /* dp becomes dvp in dirloop */
771 * Disallow directory write attempts on read-only filesystems.
774 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
778 if (cnp->cn_flags & SAVESTART) {
779 ndp->ni_startdir = ndp->ni_dvp;
780 VREF(ndp->ni_startdir);
783 if (ndp->ni_dvp != dp)
787 VFS_UNLOCK_GIANT(dvfslocked);
789 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp)
790 VOP_UNLOCK(ndp->ni_dvp, 0);
792 if (cnp->cn_flags & AUDITVNODE1)
793 AUDIT_ARG(vnode, dp, ARG_VNODE1);
794 else if (cnp->cn_flags & AUDITVNODE2)
795 AUDIT_ARG(vnode, dp, ARG_VNODE2);
797 if ((cnp->cn_flags & LOCKLEAF) == 0)
801 * Because of lookup_shared we may have the vnode shared locked, but
802 * the caller may want it to be exclusively locked.
804 if ((cnp->cn_flags & (ISLASTCN | LOCKSHARED | LOCKLEAF)) ==
805 (ISLASTCN | LOCKLEAF) && VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
806 vn_lock(dp, LK_UPGRADE | LK_RETRY);
808 if (vfslocked && dvfslocked)
809 VFS_UNLOCK_GIANT(dvfslocked); /* Only need one */
810 if (vfslocked || dvfslocked)
811 ndp->ni_cnd.cn_flags |= GIANTHELD;
815 if (dp != ndp->ni_dvp)
822 VFS_UNLOCK_GIANT(vfslocked);
823 VFS_UNLOCK_GIANT(dvfslocked);
824 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
830 * relookup - lookup a path name component
831 * Used by lookup to re-acquire things.
834 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
836 struct vnode *dp = 0; /* the directory we are searching */
837 int wantparent; /* 1 => wantparent or lockparent flag */
838 int rdonly; /* lookup read-only flag bit */
841 KASSERT(cnp->cn_flags & ISLASTCN,
842 ("relookup: Not given last component."));
844 * Setup: break out flag bits into variables.
846 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
847 KASSERT(wantparent, ("relookup: parent not wanted."));
848 rdonly = cnp->cn_flags & RDONLY;
849 cnp->cn_flags &= ~ISSYMLINK;
851 cnp->cn_lkflags = LK_EXCLUSIVE;
852 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
855 * Search a new directory.
857 * The last component of the filename is left accessible via
858 * cnp->cn_nameptr for callers that need the name. Callers needing
859 * the name set the SAVENAME flag. When done, they assume
860 * responsibility for freeing the pathname buffer.
862 #ifdef NAMEI_DIAGNOSTIC
863 printf("{%s}: ", cnp->cn_nameptr);
867 * Check for degenerate name (e.g. / or "")
868 * which is a way of talking about a directory,
869 * e.g. like "/." or ".".
871 if (cnp->cn_nameptr[0] == '\0') {
872 if (cnp->cn_nameiop != LOOKUP || wantparent) {
876 if (dp->v_type != VDIR) {
880 if (!(cnp->cn_flags & LOCKLEAF))
883 /* XXX This should probably move to the top of function. */
884 if (cnp->cn_flags & SAVESTART)
885 panic("lookup: SAVESTART");
889 if (cnp->cn_flags & ISDOTDOT)
890 panic ("relookup: lookup on dot-dot");
893 * We now have a segment name to search for, and a directory to search.
895 #ifdef NAMEI_DIAGNOSTIC
896 vprint("search in:", dp);
898 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
899 KASSERT(*vpp == NULL, ("leaf should be empty"));
900 if (error != EJUSTRETURN)
903 * If creating and at end of pathname, then can consider
904 * allowing file to be created.
910 /* ASSERT(dvp == ndp->ni_startdir) */
911 if (cnp->cn_flags & SAVESTART)
913 if ((cnp->cn_flags & LOCKPARENT) == 0)
916 * This is a temporary assert to make sure I know what the
919 KASSERT((cnp->cn_flags & (WANTPARENT|LOCKPARENT)) != 0,
920 ("relookup: Unhandled case."));
922 * We return with ni_vp NULL to indicate that the entry
923 * doesn't currently exist, leaving a pointer to the
924 * (possibly locked) directory vnode in ndp->ni_dvp.
932 * Disallow directory write attempts on read-only filesystems.
935 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
944 * Set the parent lock/ref state to the requested state.
946 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
951 } else if (!wantparent)
954 * Check for symbolic link
956 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
957 ("relookup: symlink found.\n"));
959 /* ASSERT(dvp == ndp->ni_startdir) */
960 if (cnp->cn_flags & SAVESTART)
963 if ((cnp->cn_flags & LOCKLEAF) == 0)
973 * Free data allocated by namei(); see namei(9) for details.
976 NDFREE(struct nameidata *ndp, const u_int flags)
984 if (!(flags & NDF_NO_FREE_PNBUF) &&
985 (ndp->ni_cnd.cn_flags & HASBUF)) {
986 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
987 ndp->ni_cnd.cn_flags &= ~HASBUF;
989 if (!(flags & NDF_NO_VP_UNLOCK) &&
990 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
992 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1001 VOP_UNLOCK(ndp->ni_vp, 0);
1002 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1003 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1004 ndp->ni_dvp != ndp->ni_vp)
1006 if (!(flags & NDF_NO_DVP_RELE) &&
1007 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1016 VOP_UNLOCK(ndp->ni_dvp, 0);
1017 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1018 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1019 vrele(ndp->ni_startdir);
1020 ndp->ni_startdir = NULL;
1025 * Determine if there is a suitable alternate filename under the specified
1026 * prefix for the specified path. If the create flag is set, then the
1027 * alternate prefix will be used so long as the parent directory exists.
1028 * This is used by the various compatiblity ABIs so that Linux binaries prefer
1029 * files under /compat/linux for example. The chosen path (whether under
1030 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1031 * to by pathbuf. The caller is responsible for free'ing the buffer from
1032 * the M_TEMP bucket if one is returned.
1035 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1036 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1038 struct nameidata nd, ndroot;
1039 char *ptr, *buf, *cp;
1043 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1046 /* Copy the prefix into the new pathname as a starting point. */
1047 len = strlcpy(buf, prefix, MAXPATHLEN);
1048 if (len >= MAXPATHLEN) {
1053 sz = MAXPATHLEN - len;
1056 /* Append the filename to the prefix. */
1057 if (pathseg == UIO_SYSSPACE)
1058 error = copystr(path, ptr, sz, &len);
1060 error = copyinstr(path, ptr, sz, &len);
1068 /* Only use a prefix with absolute pathnames. */
1074 if (dirfd != AT_FDCWD) {
1076 * We want the original because the "prefix" is
1077 * included in the already opened dirfd.
1079 bcopy(ptr, buf, len);
1084 * We know that there is a / somewhere in this pathname.
1085 * Search backwards for it, to find the file's parent dir
1086 * to see if it exists in the alternate tree. If it does,
1087 * and we want to create a file (cflag is set). We don't
1088 * need to worry about the root comparison in this case.
1092 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1095 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1101 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1108 * We now compare the vnode of the prefix to the one
1109 * vnode asked. If they resolve to be the same, then we
1110 * ignore the match so that the real root gets used.
1111 * This avoids the problem of traversing "../.." to find the
1112 * root directory and never finding it, because "/" resolves
1113 * to the emulation root directory. This is expensive :-(
1115 NDINIT(&ndroot, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, prefix,
1118 /* We shouldn't ever get an error from this namei(). */
1119 error = namei(&ndroot);
1121 if (nd.ni_vp == ndroot.ni_vp)
1124 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1125 vrele(ndroot.ni_vp);
1126 VFS_UNLOCK_GIANT(NDHASGIANT(&ndroot));
1130 NDFREE(&nd, NDF_ONLY_PNBUF);
1132 VFS_UNLOCK_GIANT(NDHASGIANT(&nd));
1135 /* If there was an error, use the original path name. */
1137 bcopy(ptr, buf, len);