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"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/fcntl.h>
48 #include <sys/mutex.h>
49 #include <sys/namei.h>
50 #include <sys/vnode.h>
51 #include <sys/mount.h>
52 #include <sys/filedesc.h>
54 #include <sys/syscallsubr.h>
55 #include <sys/sysctl.h>
57 #include <sys/ktrace.h>
60 #include <security/audit/audit.h>
61 #include <security/mac/mac_framework.h>
65 #define NAMEI_DIAGNOSTIC 1
66 #undef NAMEI_DIAGNOSTIC
69 * Allocation zone for namei
71 uma_zone_t namei_zone;
73 * Placeholder vnode for mp traversal
75 static struct vnode *vp_crossmp;
78 nameiinit(void *dummy __unused)
82 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
84 error = getnewvnode("crossmp", NULL, &dead_vnodeops, &vp_crossmp);
86 panic("nameiinit: getnewvnode");
87 VN_LOCK_ASHARE(vp_crossmp);
89 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
91 static int lookup_shared = 0;
92 SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RW, &lookup_shared, 0,
93 "Enables/Disables shared locks for path name translation");
94 TUNABLE_INT("vfs.lookup_shared", &lookup_shared);
97 * Convert a pathname into a pointer to a locked vnode.
99 * The FOLLOW flag is set when symbolic links are to be followed
100 * when they occur at the end of the name translation process.
101 * Symbolic links are always followed for all other pathname
102 * components other than the last.
104 * The segflg defines whether the name is to be copied from user
105 * space or kernel space.
107 * Overall outline of namei:
110 * get starting directory
111 * while (!done && !error) {
112 * call lookup to search path.
113 * if symbolic link, massage name in buffer and continue
117 namei(struct nameidata *ndp)
119 struct filedesc *fdp; /* pointer to file descriptor state */
120 char *cp; /* pointer into pathname argument */
121 struct vnode *dp; /* the directory we are searching */
122 struct iovec aiov; /* uio for reading symbolic links */
125 struct componentname *cnp = &ndp->ni_cnd;
126 struct thread *td = cnp->cn_thread;
127 struct proc *p = td->td_proc;
130 KASSERT((cnp->cn_flags & MPSAFE) != 0 || mtx_owned(&Giant) != 0,
131 ("NOT MPSAFE and Giant not held"));
132 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
133 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
134 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
135 ("namei: nameiop contaminated with flags"));
136 KASSERT((cnp->cn_flags & OPMASK) == 0,
137 ("namei: flags contaminated with nameiops"));
139 cnp->cn_flags &= ~LOCKSHARED;
143 * Get a buffer for the name to be translated, and copy the
144 * name into the buffer.
146 if ((cnp->cn_flags & HASBUF) == 0)
147 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
148 if (ndp->ni_segflg == UIO_SYSSPACE)
149 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
150 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
152 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
153 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
155 /* If we are auditing the kernel pathname, save the user pathname. */
156 if (cnp->cn_flags & AUDITVNODE1)
157 AUDIT_ARG(upath, td, cnp->cn_pnbuf, ARG_UPATH1);
158 if (cnp->cn_flags & AUDITVNODE2)
159 AUDIT_ARG(upath, td, cnp->cn_pnbuf, ARG_UPATH2);
162 * Don't allow empty pathnames.
164 if (!error && *cnp->cn_pnbuf == '\0')
168 uma_zfree(namei_zone, cnp->cn_pnbuf);
170 cnp->cn_pnbuf = NULL;
171 cnp->cn_nameptr = NULL;
178 if (KTRPOINT(td, KTR_NAMEI)) {
179 KASSERT(cnp->cn_thread == curthread,
180 ("namei not using curthread"));
181 ktrnamei(cnp->cn_pnbuf);
186 * Get starting point for the translation.
189 ndp->ni_rootdir = fdp->fd_rdir;
190 ndp->ni_topdir = fdp->fd_jdir;
192 if (cnp->cn_pnbuf[0] != '/' && ndp->ni_dirfd != AT_FDCWD) {
193 error = fgetvp(td, ndp->ni_dirfd, &dp);
194 FILEDESC_SUNLOCK(fdp);
195 if (error == 0 && dp->v_type != VDIR) {
196 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
198 VFS_UNLOCK_GIANT(vfslocked);
202 uma_zfree(namei_zone, cnp->cn_pnbuf);
204 cnp->cn_pnbuf = NULL;
205 cnp->cn_nameptr = NULL;
212 FILEDESC_SUNLOCK(fdp);
214 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
217 * Check if root directory should replace current directory.
218 * Done at start of translation and after symbolic link.
220 cnp->cn_nameptr = cnp->cn_pnbuf;
221 if (*(cnp->cn_nameptr) == '/') {
223 VFS_UNLOCK_GIANT(vfslocked);
224 while (*(cnp->cn_nameptr) == '/') {
228 dp = ndp->ni_rootdir;
229 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
233 ndp->ni_cnd.cn_flags |= GIANTHELD;
234 ndp->ni_startdir = dp;
237 uma_zfree(namei_zone, cnp->cn_pnbuf);
239 cnp->cn_pnbuf = NULL;
240 cnp->cn_nameptr = NULL;
244 vfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
245 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
247 * Check for symbolic link
249 if ((cnp->cn_flags & ISSYMLINK) == 0) {
250 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
251 uma_zfree(namei_zone, cnp->cn_pnbuf);
253 cnp->cn_pnbuf = NULL;
254 cnp->cn_nameptr = NULL;
257 cnp->cn_flags |= HASBUF;
259 if ((cnp->cn_flags & MPSAFE) == 0) {
260 VFS_UNLOCK_GIANT(vfslocked);
261 } else if (vfslocked)
262 ndp->ni_cnd.cn_flags |= GIANTHELD;
265 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
270 if ((cnp->cn_flags & NOMACCHECK) == 0) {
271 error = mac_vnode_check_readlink(td->td_ucred,
277 if (ndp->ni_pathlen > 1)
278 cp = uma_zalloc(namei_zone, M_WAITOK);
282 aiov.iov_len = MAXPATHLEN;
283 auio.uio_iov = &aiov;
286 auio.uio_rw = UIO_READ;
287 auio.uio_segflg = UIO_SYSSPACE;
288 auio.uio_td = (struct thread *)0;
289 auio.uio_resid = MAXPATHLEN;
290 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
292 if (ndp->ni_pathlen > 1)
293 uma_zfree(namei_zone, cp);
296 linklen = MAXPATHLEN - auio.uio_resid;
298 if (ndp->ni_pathlen > 1)
299 uma_zfree(namei_zone, cp);
303 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
304 if (ndp->ni_pathlen > 1)
305 uma_zfree(namei_zone, cp);
306 error = ENAMETOOLONG;
309 if (ndp->ni_pathlen > 1) {
310 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
311 uma_zfree(namei_zone, cnp->cn_pnbuf);
314 cnp->cn_pnbuf[linklen] = '\0';
315 ndp->ni_pathlen += linklen;
319 uma_zfree(namei_zone, cnp->cn_pnbuf);
321 cnp->cn_pnbuf = NULL;
322 cnp->cn_nameptr = NULL;
327 VFS_UNLOCK_GIANT(vfslocked);
332 compute_cn_lkflags(struct mount *mp, int lkflags)
335 ((lkflags & LK_SHARED) && !(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED))) {
336 lkflags &= ~LK_SHARED;
337 lkflags |= LK_EXCLUSIVE;
344 * This is a very central and rather complicated routine.
346 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
347 * The starting directory is taken from ni_startdir. The pathname is
348 * descended until done, or a symbolic link is encountered. The variable
349 * ni_more is clear if the path is completed; it is set to one if a
350 * symbolic link needing interpretation is encountered.
352 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
353 * whether the name is to be looked up, created, renamed, or deleted.
354 * When CREATE, RENAME, or DELETE is specified, information usable in
355 * creating, renaming, or deleting a directory entry may be calculated.
356 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
357 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
358 * returned unlocked. Otherwise the parent directory is not returned. If
359 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
360 * the target is returned locked, otherwise it is returned unlocked.
361 * When creating or renaming and LOCKPARENT is specified, the target may not
362 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
364 * Overall outline of lookup:
367 * identify next component of name at ndp->ni_ptr
368 * handle degenerate case where name is null string
369 * if .. and crossing mount points and on mounted filesys, find parent
370 * call VOP_LOOKUP routine for next component name
371 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
372 * component vnode returned in ni_vp (if it exists), locked.
373 * if result vnode is mounted on and crossing mount points,
374 * find mounted on vnode
375 * if more components of name, do next level at dirloop
376 * return the answer in ni_vp, locked if LOCKLEAF set
377 * if LOCKPARENT set, return locked parent in ni_dvp
378 * if WANTPARENT set, return unlocked parent in ni_dvp
381 lookup(struct nameidata *ndp)
383 char *cp; /* pointer into pathname argument */
384 struct vnode *dp = 0; /* the directory we are searching */
385 struct vnode *tdp; /* saved dp */
386 struct mount *mp; /* mount table entry */
387 int docache; /* == 0 do not cache last component */
388 int wantparent; /* 1 => wantparent or lockparent flag */
389 int rdonly; /* lookup read-only flag bit */
392 int dpunlocked = 0; /* dp has already been unlocked */
393 struct componentname *cnp = &ndp->ni_cnd;
394 struct thread *td = cnp->cn_thread;
395 int vfslocked; /* VFS Giant state for child */
396 int dvfslocked; /* VFS Giant state for parent */
401 * Setup: break out flag bits into variables.
403 dvfslocked = (ndp->ni_cnd.cn_flags & GIANTHELD) != 0;
405 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
406 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
407 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
408 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
409 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
410 if (cnp->cn_nameiop == DELETE ||
411 (wantparent && cnp->cn_nameiop != CREATE &&
412 cnp->cn_nameiop != LOOKUP))
414 rdonly = cnp->cn_flags & RDONLY;
415 cnp->cn_flags &= ~ISSYMLINK;
418 * We use shared locks until we hit the parent of the last cn then
419 * we adjust based on the requesting flags.
422 cnp->cn_lkflags = LK_SHARED;
424 cnp->cn_lkflags = LK_EXCLUSIVE;
425 dp = ndp->ni_startdir;
426 ndp->ni_startdir = NULLVP;
428 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY));
432 * Search a new directory.
434 * The last component of the filename is left accessible via
435 * cnp->cn_nameptr for callers that need the name. Callers needing
436 * the name set the SAVENAME flag. When done, they assume
437 * responsibility for freeing the pathname buffer.
440 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
442 cnp->cn_namelen = cp - cnp->cn_nameptr;
443 if (cnp->cn_namelen > NAME_MAX) {
444 error = ENAMETOOLONG;
447 #ifdef NAMEI_DIAGNOSTIC
450 printf("{%s}: ", cnp->cn_nameptr);
453 ndp->ni_pathlen -= cnp->cn_namelen;
457 * Replace multiple slashes by a single slash and trailing slashes
458 * by a null. This must be done before VOP_LOOKUP() because some
459 * fs's don't know about trailing slashes. Remember if there were
460 * trailing slashes to handle symlinks, existing non-directories
461 * and non-existing files that won't be directories specially later.
464 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
469 *ndp->ni_next = '\0'; /* XXX for direnter() ... */
474 cnp->cn_flags |= MAKEENTRY;
475 if (*cp == '\0' && docache == 0)
476 cnp->cn_flags &= ~MAKEENTRY;
477 if (cnp->cn_namelen == 2 &&
478 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
479 cnp->cn_flags |= ISDOTDOT;
481 cnp->cn_flags &= ~ISDOTDOT;
482 if (*ndp->ni_next == 0)
483 cnp->cn_flags |= ISLASTCN;
485 cnp->cn_flags &= ~ISLASTCN;
489 * Check for degenerate name (e.g. / or "")
490 * which is a way of talking about a directory,
491 * e.g. like "/." or ".".
493 if (cnp->cn_nameptr[0] == '\0') {
494 if (dp->v_type != VDIR) {
498 if (cnp->cn_nameiop != LOOKUP) {
508 if (cnp->cn_flags & AUDITVNODE1)
509 AUDIT_ARG(vnode, dp, ARG_VNODE1);
510 else if (cnp->cn_flags & AUDITVNODE2)
511 AUDIT_ARG(vnode, dp, ARG_VNODE2);
513 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
515 /* XXX This should probably move to the top of function. */
516 if (cnp->cn_flags & SAVESTART)
517 panic("lookup: SAVESTART");
522 * Handle "..": four special cases.
523 * 1. Return an error if this is the last component of
524 * the name and the operation is DELETE or RENAME.
525 * 2. If at root directory (e.g. after chroot)
526 * or at absolute root directory
527 * then ignore it so can't get out.
528 * 3. If this vnode is the root of a mounted
529 * filesystem, then replace it with the
530 * vnode which was mounted on so we take the
531 * .. in the other filesystem.
532 * 4. If the vnode is the top directory of
533 * the jail or chroot, don't let them out.
535 if (cnp->cn_flags & ISDOTDOT) {
536 if ((cnp->cn_flags & ISLASTCN) != 0 &&
537 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
542 if (dp == ndp->ni_rootdir ||
543 dp == ndp->ni_topdir ||
545 ((dp->v_vflag & VV_ROOT) != 0 &&
546 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
549 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
553 if ((dp->v_vflag & VV_ROOT) == 0)
555 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
560 dp = dp->v_mount->mnt_vnodecovered;
561 tvfslocked = dvfslocked;
562 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
565 VFS_UNLOCK_GIANT(tvfslocked);
567 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
573 * We now have a segment name to search for, and a directory to search.
577 if ((cnp->cn_flags & NOMACCHECK) == 0) {
578 error = mac_vnode_check_lookup(td->td_ucred, dp, cnp);
585 ASSERT_VOP_LOCKED(dp, "lookup");
586 VNASSERT(vfslocked == 0, dp, ("lookup: vfslocked %d", vfslocked));
588 * If we have a shared lock we may need to upgrade the lock for the
591 if (dp != vp_crossmp &&
592 VOP_ISLOCKED(dp) == LK_SHARED &&
593 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT))
594 vn_lock(dp, LK_UPGRADE|LK_RETRY);
596 * If we're looking up the last component and we need an exclusive
597 * lock, adjust our lkflags.
599 if ((cnp->cn_flags & (ISLASTCN|LOCKSHARED|LOCKLEAF)) ==
601 cnp->cn_lkflags = LK_EXCLUSIVE;
602 #ifdef NAMEI_DIAGNOSTIC
603 vprint("lookup in", dp);
605 lkflags_save = cnp->cn_lkflags;
606 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags);
607 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
608 cnp->cn_lkflags = lkflags_save;
609 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
610 #ifdef NAMEI_DIAGNOSTIC
611 printf("not found\n");
613 if ((error == ENOENT) &&
614 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
615 (dp->v_mount->mnt_flag & MNT_UNION)) {
617 dp = dp->v_mount->mnt_vnodecovered;
618 tvfslocked = dvfslocked;
619 dvfslocked = VFS_LOCK_GIANT(dp->v_mount);
622 VFS_UNLOCK_GIANT(tvfslocked);
624 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
629 if (error != EJUSTRETURN)
632 * If creating and at end of pathname, then can consider
633 * allowing file to be created.
639 if (*cp == '\0' && trailing_slash &&
640 !(cnp->cn_flags & WILLBEDIR)) {
644 if ((cnp->cn_flags & LOCKPARENT) == 0)
647 * This is a temporary assert to make sure I know what the
650 KASSERT((cnp->cn_flags & (WANTPARENT|LOCKPARENT)) != 0,
651 ("lookup: Unhandled case."));
653 * We return with ni_vp NULL to indicate that the entry
654 * doesn't currently exist, leaving a pointer to the
655 * (possibly locked) directory vnode in ndp->ni_dvp.
657 if (cnp->cn_flags & SAVESTART) {
658 ndp->ni_startdir = ndp->ni_dvp;
659 VREF(ndp->ni_startdir);
663 cnp->cn_lkflags = lkflags_save;
664 #ifdef NAMEI_DIAGNOSTIC
668 * Take into account any additional components consumed by
669 * the underlying filesystem.
671 if (cnp->cn_consume > 0) {
672 cnp->cn_nameptr += cnp->cn_consume;
673 ndp->ni_next += cnp->cn_consume;
674 ndp->ni_pathlen -= cnp->cn_consume;
679 vfslocked = VFS_LOCK_GIANT(dp->v_mount);
682 * Check to see if the vnode has been mounted on;
683 * if so find the root of the mounted filesystem.
685 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
686 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
687 if (vfs_busy(mp, 0, 0))
690 VFS_UNLOCK_GIANT(vfslocked);
691 vfslocked = VFS_LOCK_GIANT(mp);
692 if (dp != ndp->ni_dvp)
696 VFS_UNLOCK_GIANT(dvfslocked);
699 ndp->ni_dvp = vp_crossmp;
700 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags), &tdp, td);
702 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
703 panic("vp_crossmp exclusively locked or reclaimed");
708 ndp->ni_vp = dp = tdp;
712 * Check for symbolic link
714 if ((dp->v_type == VLNK) &&
715 ((cnp->cn_flags & FOLLOW) || trailing_slash ||
716 *ndp->ni_next == '/')) {
717 cnp->cn_flags |= ISSYMLINK;
718 if (dp->v_iflag & VI_DOOMED) {
719 /* We can't know whether the directory was mounted with
720 * NOSYMFOLLOW, so we can't follow safely. */
724 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
729 * Symlink code always expects an unlocked dvp.
731 if (ndp->ni_dvp != ndp->ni_vp)
732 VOP_UNLOCK(ndp->ni_dvp, 0);
737 * Check for bogus trailing slashes.
739 if (trailing_slash && dp->v_type != VDIR) {
746 * Not a symbolic link. If more pathname,
747 * continue at next component, else return.
749 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
750 ("lookup: invalid path state."));
751 if (*ndp->ni_next == '/') {
752 cnp->cn_nameptr = ndp->ni_next;
753 while (*cnp->cn_nameptr == '/') {
757 if (ndp->ni_dvp != dp)
761 VFS_UNLOCK_GIANT(dvfslocked);
762 dvfslocked = vfslocked; /* dp becomes dvp in dirloop */
767 * Disallow directory write attempts on read-only filesystems.
770 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
774 if (cnp->cn_flags & SAVESTART) {
775 ndp->ni_startdir = ndp->ni_dvp;
776 VREF(ndp->ni_startdir);
779 if (ndp->ni_dvp != dp)
783 VFS_UNLOCK_GIANT(dvfslocked);
785 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp)
786 VOP_UNLOCK(ndp->ni_dvp, 0);
788 if (cnp->cn_flags & AUDITVNODE1)
789 AUDIT_ARG(vnode, dp, ARG_VNODE1);
790 else if (cnp->cn_flags & AUDITVNODE2)
791 AUDIT_ARG(vnode, dp, ARG_VNODE2);
793 if ((cnp->cn_flags & LOCKLEAF) == 0)
797 * Because of lookup_shared we may have the vnode shared locked, but
798 * the caller may want it to be exclusively locked.
800 if ((cnp->cn_flags & (ISLASTCN | LOCKSHARED | LOCKLEAF)) ==
801 (ISLASTCN | LOCKLEAF) && VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
802 vn_lock(dp, LK_UPGRADE | LK_RETRY);
804 if (vfslocked && dvfslocked)
805 VFS_UNLOCK_GIANT(dvfslocked); /* Only need one */
806 if (vfslocked || dvfslocked)
807 ndp->ni_cnd.cn_flags |= GIANTHELD;
811 if (dp != ndp->ni_dvp)
818 VFS_UNLOCK_GIANT(vfslocked);
819 VFS_UNLOCK_GIANT(dvfslocked);
820 ndp->ni_cnd.cn_flags &= ~GIANTHELD;
826 * relookup - lookup a path name component
827 * Used by lookup to re-acquire things.
830 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
832 struct vnode *dp = 0; /* the directory we are searching */
833 int wantparent; /* 1 => wantparent or lockparent flag */
834 int rdonly; /* lookup read-only flag bit */
837 KASSERT(cnp->cn_flags & ISLASTCN,
838 ("relookup: Not given last component."));
840 * Setup: break out flag bits into variables.
842 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
843 KASSERT(wantparent, ("relookup: parent not wanted."));
844 rdonly = cnp->cn_flags & RDONLY;
845 cnp->cn_flags &= ~ISSYMLINK;
847 cnp->cn_lkflags = LK_EXCLUSIVE;
848 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
851 * Search a new directory.
853 * The last component of the filename is left accessible via
854 * cnp->cn_nameptr for callers that need the name. Callers needing
855 * the name set the SAVENAME flag. When done, they assume
856 * responsibility for freeing the pathname buffer.
858 #ifdef NAMEI_DIAGNOSTIC
859 printf("{%s}: ", cnp->cn_nameptr);
863 * Check for degenerate name (e.g. / or "")
864 * which is a way of talking about a directory,
865 * e.g. like "/." or ".".
867 if (cnp->cn_nameptr[0] == '\0') {
868 if (cnp->cn_nameiop != LOOKUP || wantparent) {
872 if (dp->v_type != VDIR) {
876 if (!(cnp->cn_flags & LOCKLEAF))
879 /* XXX This should probably move to the top of function. */
880 if (cnp->cn_flags & SAVESTART)
881 panic("lookup: SAVESTART");
885 if (cnp->cn_flags & ISDOTDOT)
886 panic ("relookup: lookup on dot-dot");
889 * We now have a segment name to search for, and a directory to search.
891 #ifdef NAMEI_DIAGNOSTIC
892 vprint("search in:", dp);
894 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
895 KASSERT(*vpp == NULL, ("leaf should be empty"));
896 if (error != EJUSTRETURN)
899 * If creating and at end of pathname, then can consider
900 * allowing file to be created.
906 /* ASSERT(dvp == ndp->ni_startdir) */
907 if (cnp->cn_flags & SAVESTART)
909 if ((cnp->cn_flags & LOCKPARENT) == 0)
912 * This is a temporary assert to make sure I know what the
915 KASSERT((cnp->cn_flags & (WANTPARENT|LOCKPARENT)) != 0,
916 ("relookup: Unhandled case."));
918 * We return with ni_vp NULL to indicate that the entry
919 * doesn't currently exist, leaving a pointer to the
920 * (possibly locked) directory vnode in ndp->ni_dvp.
928 * Disallow directory write attempts on read-only filesystems.
931 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
940 * Set the parent lock/ref state to the requested state.
942 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
947 } else if (!wantparent)
950 * Check for symbolic link
952 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
953 ("relookup: symlink found.\n"));
955 /* ASSERT(dvp == ndp->ni_startdir) */
956 if (cnp->cn_flags & SAVESTART)
959 if ((cnp->cn_flags & LOCKLEAF) == 0)
969 * Free data allocated by namei(); see namei(9) for details.
972 NDFREE(struct nameidata *ndp, const u_int flags)
980 if (!(flags & NDF_NO_FREE_PNBUF) &&
981 (ndp->ni_cnd.cn_flags & HASBUF)) {
982 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
983 ndp->ni_cnd.cn_flags &= ~HASBUF;
985 if (!(flags & NDF_NO_VP_UNLOCK) &&
986 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
988 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
997 VOP_UNLOCK(ndp->ni_vp, 0);
998 if (!(flags & NDF_NO_DVP_UNLOCK) &&
999 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1000 ndp->ni_dvp != ndp->ni_vp)
1002 if (!(flags & NDF_NO_DVP_RELE) &&
1003 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1012 VOP_UNLOCK(ndp->ni_dvp, 0);
1013 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1014 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1015 vrele(ndp->ni_startdir);
1016 ndp->ni_startdir = NULL;
1021 * Determine if there is a suitable alternate filename under the specified
1022 * prefix for the specified path. If the create flag is set, then the
1023 * alternate prefix will be used so long as the parent directory exists.
1024 * This is used by the various compatiblity ABIs so that Linux binaries prefer
1025 * files under /compat/linux for example. The chosen path (whether under
1026 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1027 * to by pathbuf. The caller is responsible for free'ing the buffer from
1028 * the M_TEMP bucket if one is returned.
1031 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1032 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1034 struct nameidata nd, ndroot;
1035 char *ptr, *buf, *cp;
1039 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1042 /* Copy the prefix into the new pathname as a starting point. */
1043 len = strlcpy(buf, prefix, MAXPATHLEN);
1044 if (len >= MAXPATHLEN) {
1049 sz = MAXPATHLEN - len;
1052 /* Append the filename to the prefix. */
1053 if (pathseg == UIO_SYSSPACE)
1054 error = copystr(path, ptr, sz, &len);
1056 error = copyinstr(path, ptr, sz, &len);
1064 /* Only use a prefix with absolute pathnames. */
1070 if (dirfd != AT_FDCWD) {
1072 * We want the original because the "prefix" is
1073 * included in the already opened dirfd.
1075 bcopy(ptr, buf, len);
1080 * We know that there is a / somewhere in this pathname.
1081 * Search backwards for it, to find the file's parent dir
1082 * to see if it exists in the alternate tree. If it does,
1083 * and we want to create a file (cflag is set). We don't
1084 * need to worry about the root comparison in this case.
1088 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1091 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1097 NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, buf, td);
1104 * We now compare the vnode of the prefix to the one
1105 * vnode asked. If they resolve to be the same, then we
1106 * ignore the match so that the real root gets used.
1107 * This avoids the problem of traversing "../.." to find the
1108 * root directory and never finding it, because "/" resolves
1109 * to the emulation root directory. This is expensive :-(
1111 NDINIT(&ndroot, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, prefix,
1114 /* We shouldn't ever get an error from this namei(). */
1115 error = namei(&ndroot);
1117 if (nd.ni_vp == ndroot.ni_vp)
1120 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1121 vrele(ndroot.ni_vp);
1122 VFS_UNLOCK_GIANT(NDHASGIANT(&ndroot));
1126 NDFREE(&nd, NDF_ONLY_PNBUF);
1128 VFS_UNLOCK_GIANT(NDHASGIANT(&nd));
1131 /* If there was an error, use the original path name. */
1133 bcopy(ptr, buf, len);