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;
137 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
138 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
139 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
140 ("namei: nameiop contaminated with flags"));
141 KASSERT((cnp->cn_flags & OPMASK) == 0,
142 ("namei: flags contaminated with nameiops"));
144 cnp->cn_flags &= ~LOCKSHARED;
147 /* We will set this ourselves if we need it. */
148 cnp->cn_flags &= ~TRAILINGSLASH;
151 * Get a buffer for the name to be translated, and copy the
152 * name into the buffer.
154 if ((cnp->cn_flags & HASBUF) == 0)
155 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
156 if (ndp->ni_segflg == UIO_SYSSPACE)
157 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
158 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
160 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
161 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
164 * Don't allow empty pathnames.
166 if (!error && *cnp->cn_pnbuf == '\0')
169 #ifdef CAPABILITY_MODE
171 * In capability mode, lookups must be "strictly relative" (i.e.
172 * not an absolute path, and not containing '..' components) to
173 * a real file descriptor, not the pseudo-descriptor AT_FDCWD.
175 if (IN_CAPABILITY_MODE(td) && (cnp->cn_flags & NOCAPCHECK) == 0) {
176 ndp->ni_strictrelative = 1;
177 if (ndp->ni_dirfd == AT_FDCWD) {
179 if (KTRPOINT(td, KTR_CAPFAIL))
180 ktrcapfail(CAPFAIL_LOOKUP, 0, 0);
187 uma_zfree(namei_zone, cnp->cn_pnbuf);
189 cnp->cn_pnbuf = NULL;
190 cnp->cn_nameptr = NULL;
197 if (KTRPOINT(td, KTR_NAMEI)) {
198 KASSERT(cnp->cn_thread == curthread,
199 ("namei not using curthread"));
200 ktrnamei(cnp->cn_pnbuf);
204 * Get starting point for the translation.
207 ndp->ni_rootdir = fdp->fd_rdir;
208 ndp->ni_topdir = fdp->fd_jdir;
211 * If we are auditing the kernel pathname, save the user pathname.
213 if (cnp->cn_flags & AUDITVNODE1)
214 AUDIT_ARG_UPATH1(td, ndp->ni_dirfd, cnp->cn_pnbuf);
215 if (cnp->cn_flags & AUDITVNODE2)
216 AUDIT_ARG_UPATH2(td, ndp->ni_dirfd, cnp->cn_pnbuf);
219 if (cnp->cn_pnbuf[0] != '/') {
220 if (ndp->ni_startdir != NULL) {
221 dp = ndp->ni_startdir;
223 } else if (ndp->ni_dirfd != AT_FDCWD) {
224 if (cnp->cn_flags & AUDITVNODE1)
225 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
226 if (cnp->cn_flags & AUDITVNODE2)
227 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
228 error = fgetvp_rights(td, ndp->ni_dirfd,
229 ndp->ni_rightsneeded | CAP_LOOKUP,
230 &ndp->ni_filecaps, &dp);
233 * If file descriptor doesn't have all rights,
234 * all lookups relative to it must also be
237 if (ndp->ni_filecaps.fc_rights != CAP_ALL ||
238 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
239 ndp->ni_filecaps.fc_nioctls != -1) {
240 ndp->ni_strictrelative = 1;
244 if (error != 0 || dp != NULL) {
245 FILEDESC_SUNLOCK(fdp);
246 if (error == 0 && dp->v_type != VDIR) {
252 uma_zfree(namei_zone, cnp->cn_pnbuf);
254 cnp->cn_pnbuf = NULL;
255 cnp->cn_nameptr = NULL;
263 FILEDESC_SUNLOCK(fdp);
264 if (ndp->ni_startdir != NULL)
265 vrele(ndp->ni_startdir);
267 SDT_PROBE(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
268 cnp->cn_flags, 0, 0);
271 * Check if root directory should replace current directory.
272 * Done at start of translation and after symbolic link.
274 cnp->cn_nameptr = cnp->cn_pnbuf;
275 if (*(cnp->cn_nameptr) == '/') {
277 if (ndp->ni_strictrelative != 0) {
279 if (KTRPOINT(curthread, KTR_CAPFAIL))
280 ktrcapfail(CAPFAIL_LOOKUP, 0, 0);
282 return (ENOTCAPABLE);
284 while (*(cnp->cn_nameptr) == '/') {
288 dp = ndp->ni_rootdir;
291 ndp->ni_startdir = dp;
294 uma_zfree(namei_zone, cnp->cn_pnbuf);
296 cnp->cn_pnbuf = NULL;
297 cnp->cn_nameptr = NULL;
299 SDT_PROBE(vfs, namei, lookup, return, error, NULL, 0,
304 * If not a symbolic link, we're done.
306 if ((cnp->cn_flags & ISSYMLINK) == 0) {
307 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
308 uma_zfree(namei_zone, cnp->cn_pnbuf);
310 cnp->cn_pnbuf = NULL;
311 cnp->cn_nameptr = NULL;
314 cnp->cn_flags |= HASBUF;
316 SDT_PROBE(vfs, namei, lookup, return, 0, ndp->ni_vp,
320 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
325 if ((cnp->cn_flags & NOMACCHECK) == 0) {
326 error = mac_vnode_check_readlink(td->td_ucred,
332 if (ndp->ni_pathlen > 1)
333 cp = uma_zalloc(namei_zone, M_WAITOK);
337 aiov.iov_len = MAXPATHLEN;
338 auio.uio_iov = &aiov;
341 auio.uio_rw = UIO_READ;
342 auio.uio_segflg = UIO_SYSSPACE;
344 auio.uio_resid = MAXPATHLEN;
345 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
347 if (ndp->ni_pathlen > 1)
348 uma_zfree(namei_zone, cp);
351 linklen = MAXPATHLEN - auio.uio_resid;
353 if (ndp->ni_pathlen > 1)
354 uma_zfree(namei_zone, cp);
358 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
359 if (ndp->ni_pathlen > 1)
360 uma_zfree(namei_zone, cp);
361 error = ENAMETOOLONG;
364 if (ndp->ni_pathlen > 1) {
365 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
366 uma_zfree(namei_zone, cnp->cn_pnbuf);
369 cnp->cn_pnbuf[linklen] = '\0';
370 ndp->ni_pathlen += linklen;
374 uma_zfree(namei_zone, cnp->cn_pnbuf);
376 cnp->cn_pnbuf = NULL;
377 cnp->cn_nameptr = NULL;
382 SDT_PROBE(vfs, namei, lookup, return, error, NULL, 0, 0, 0);
387 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
390 if (mp == NULL || ((lkflags & LK_SHARED) &&
391 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
392 ((cnflags & ISDOTDOT) &&
393 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
394 lkflags &= ~LK_SHARED;
395 lkflags |= LK_EXCLUSIVE;
401 needs_exclusive_leaf(struct mount *mp, int flags)
405 * Intermediate nodes can use shared locks, we only need to
406 * force an exclusive lock for leaf nodes.
408 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
411 /* Always use exclusive locks if LOCKSHARED isn't set. */
412 if (!(flags & LOCKSHARED))
416 * For lookups during open(), if the mount point supports
417 * extended shared operations, then use a shared lock for the
418 * leaf node, otherwise use an exclusive lock.
420 if (flags & ISOPEN) {
422 (mp->mnt_kern_flag & MNTK_EXTENDED_SHARED))
429 * Lookup requests outside of open() that specify LOCKSHARED
430 * only need a shared lock on the leaf vnode.
437 * This is a very central and rather complicated routine.
439 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
440 * The starting directory is taken from ni_startdir. The pathname is
441 * descended until done, or a symbolic link is encountered. The variable
442 * ni_more is clear if the path is completed; it is set to one if a
443 * symbolic link needing interpretation is encountered.
445 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
446 * whether the name is to be looked up, created, renamed, or deleted.
447 * When CREATE, RENAME, or DELETE is specified, information usable in
448 * creating, renaming, or deleting a directory entry may be calculated.
449 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
450 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
451 * returned unlocked. Otherwise the parent directory is not returned. If
452 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
453 * the target is returned locked, otherwise it is returned unlocked.
454 * When creating or renaming and LOCKPARENT is specified, the target may not
455 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
457 * Overall outline of lookup:
460 * identify next component of name at ndp->ni_ptr
461 * handle degenerate case where name is null string
462 * if .. and crossing mount points and on mounted filesys, find parent
463 * call VOP_LOOKUP routine for next component name
464 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
465 * component vnode returned in ni_vp (if it exists), locked.
466 * if result vnode is mounted on and crossing mount points,
467 * find mounted on vnode
468 * if more components of name, do next level at dirloop
469 * return the answer in ni_vp, locked if LOCKLEAF set
470 * if LOCKPARENT set, return locked parent in ni_dvp
471 * if WANTPARENT set, return unlocked parent in ni_dvp
474 lookup(struct nameidata *ndp)
476 char *cp; /* pointer into pathname argument */
477 struct vnode *dp = 0; /* the directory we are searching */
478 struct vnode *tdp; /* saved dp */
479 struct mount *mp; /* mount table entry */
481 int docache; /* == 0 do not cache last component */
482 int wantparent; /* 1 => wantparent or lockparent flag */
483 int rdonly; /* lookup read-only flag bit */
485 int dpunlocked = 0; /* dp has already been unlocked */
486 struct componentname *cnp = &ndp->ni_cnd;
491 * Setup: break out flag bits into variables.
494 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
495 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
496 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
497 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
498 if (cnp->cn_nameiop == DELETE ||
499 (wantparent && cnp->cn_nameiop != CREATE &&
500 cnp->cn_nameiop != LOOKUP))
502 rdonly = cnp->cn_flags & RDONLY;
503 cnp->cn_flags &= ~ISSYMLINK;
506 * We use shared locks until we hit the parent of the last cn then
507 * we adjust based on the requesting flags.
510 cnp->cn_lkflags = LK_SHARED;
512 cnp->cn_lkflags = LK_EXCLUSIVE;
513 dp = ndp->ni_startdir;
514 ndp->ni_startdir = NULLVP;
516 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
521 * Search a new directory.
523 * The last component of the filename is left accessible via
524 * cnp->cn_nameptr for callers that need the name. Callers needing
525 * the name set the SAVENAME flag. When done, they assume
526 * responsibility for freeing the pathname buffer.
529 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
531 cnp->cn_namelen = cp - cnp->cn_nameptr;
532 if (cnp->cn_namelen > NAME_MAX) {
533 error = ENAMETOOLONG;
536 #ifdef NAMEI_DIAGNOSTIC
539 printf("{%s}: ", cnp->cn_nameptr);
542 ndp->ni_pathlen -= cnp->cn_namelen;
546 * Replace multiple slashes by a single slash and trailing slashes
547 * by a null. This must be done before VOP_LOOKUP() because some
548 * fs's don't know about trailing slashes. Remember if there were
549 * trailing slashes to handle symlinks, existing non-directories
550 * and non-existing files that won't be directories specially later.
552 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
556 *ndp->ni_next = '\0';
557 cnp->cn_flags |= TRAILINGSLASH;
562 cnp->cn_flags |= MAKEENTRY;
563 if (*cp == '\0' && docache == 0)
564 cnp->cn_flags &= ~MAKEENTRY;
565 if (cnp->cn_namelen == 2 &&
566 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
567 cnp->cn_flags |= ISDOTDOT;
569 cnp->cn_flags &= ~ISDOTDOT;
570 if (*ndp->ni_next == 0)
571 cnp->cn_flags |= ISLASTCN;
573 cnp->cn_flags &= ~ISLASTCN;
575 if ((cnp->cn_flags & ISLASTCN) != 0 &&
576 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
577 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
583 * Check for degenerate name (e.g. / or "")
584 * which is a way of talking about a directory,
585 * e.g. like "/." or ".".
587 if (cnp->cn_nameptr[0] == '\0') {
588 if (dp->v_type != VDIR) {
592 if (cnp->cn_nameiop != LOOKUP) {
602 if (cnp->cn_flags & AUDITVNODE1)
603 AUDIT_ARG_VNODE1(dp);
604 else if (cnp->cn_flags & AUDITVNODE2)
605 AUDIT_ARG_VNODE2(dp);
607 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
609 /* XXX This should probably move to the top of function. */
610 if (cnp->cn_flags & SAVESTART)
611 panic("lookup: SAVESTART");
616 * Handle "..": five special cases.
617 * 0. If doing a capability lookup, return ENOTCAPABLE (this is a
618 * fairly conservative design choice, but it's the only one that we
619 * are satisfied guarantees the property we're looking for).
620 * 1. Return an error if this is the last component of
621 * the name and the operation is DELETE or RENAME.
622 * 2. If at root directory (e.g. after chroot)
623 * or at absolute root directory
624 * then ignore it so can't get out.
625 * 3. If this vnode is the root of a mounted
626 * filesystem, then replace it with the
627 * vnode which was mounted on so we take the
628 * .. in the other filesystem.
629 * 4. If the vnode is the top directory of
630 * the jail or chroot, don't let them out.
632 if (cnp->cn_flags & ISDOTDOT) {
633 if (ndp->ni_strictrelative != 0) {
635 if (KTRPOINT(curthread, KTR_CAPFAIL))
636 ktrcapfail(CAPFAIL_LOOKUP, 0, 0);
641 if ((cnp->cn_flags & ISLASTCN) != 0 &&
642 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
647 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
649 if (dp == pr->pr_root)
651 if (dp == ndp->ni_rootdir ||
652 dp == ndp->ni_topdir ||
655 ((dp->v_vflag & VV_ROOT) != 0 &&
656 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
662 if ((dp->v_vflag & VV_ROOT) == 0)
664 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
669 dp = dp->v_mount->mnt_vnodecovered;
673 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
674 LK_RETRY, ISDOTDOT));
679 * We now have a segment name to search for, and a directory to search.
683 if ((cnp->cn_flags & NOMACCHECK) == 0) {
684 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
692 ASSERT_VOP_LOCKED(dp, "lookup");
694 * If we have a shared lock we may need to upgrade the lock for the
697 if (dp != vp_crossmp &&
698 VOP_ISLOCKED(dp) == LK_SHARED &&
699 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT))
700 vn_lock(dp, LK_UPGRADE|LK_RETRY);
701 if ((dp->v_iflag & VI_DOOMED) != 0) {
706 * If we're looking up the last component and we need an exclusive
707 * lock, adjust our lkflags.
709 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
710 cnp->cn_lkflags = LK_EXCLUSIVE;
711 #ifdef NAMEI_DIAGNOSTIC
712 vprint("lookup in", dp);
714 lkflags_save = cnp->cn_lkflags;
715 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
717 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
718 cnp->cn_lkflags = lkflags_save;
719 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
720 #ifdef NAMEI_DIAGNOSTIC
721 printf("not found\n");
723 if ((error == ENOENT) &&
724 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
725 (dp->v_mount->mnt_flag & MNT_UNION)) {
727 dp = dp->v_mount->mnt_vnodecovered;
731 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
732 LK_RETRY, cnp->cn_flags));
736 if (error != EJUSTRETURN)
739 * At this point, we know we're at the end of the
740 * pathname. If creating / renaming, we can consider
741 * allowing the file or directory to be created / renamed,
742 * provided we're not on a read-only filesystem.
748 /* trailing slash only allowed for directories */
749 if ((cnp->cn_flags & TRAILINGSLASH) &&
750 !(cnp->cn_flags & WILLBEDIR)) {
754 if ((cnp->cn_flags & LOCKPARENT) == 0)
757 * We return with ni_vp NULL to indicate that the entry
758 * doesn't currently exist, leaving a pointer to the
759 * (possibly locked) directory vnode in ndp->ni_dvp.
761 if (cnp->cn_flags & SAVESTART) {
762 ndp->ni_startdir = ndp->ni_dvp;
763 VREF(ndp->ni_startdir);
767 cnp->cn_lkflags = lkflags_save;
768 #ifdef NAMEI_DIAGNOSTIC
772 * Take into account any additional components consumed by
773 * the underlying filesystem.
775 if (cnp->cn_consume > 0) {
776 cnp->cn_nameptr += cnp->cn_consume;
777 ndp->ni_next += cnp->cn_consume;
778 ndp->ni_pathlen -= cnp->cn_consume;
785 * Check to see if the vnode has been mounted on;
786 * if so find the root of the mounted filesystem.
788 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
789 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
793 if (dp != ndp->ni_dvp)
798 ndp->ni_dvp = vp_crossmp;
799 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
800 cnp->cn_flags), &tdp);
802 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
803 panic("vp_crossmp exclusively locked or reclaimed");
808 ndp->ni_vp = dp = tdp;
812 * Check for symbolic link
814 if ((dp->v_type == VLNK) &&
815 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
816 *ndp->ni_next == '/')) {
817 cnp->cn_flags |= ISSYMLINK;
818 if (dp->v_iflag & VI_DOOMED) {
820 * We can't know whether the directory was mounted with
821 * NOSYMFOLLOW, so we can't follow safely.
826 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
831 * Symlink code always expects an unlocked dvp.
833 if (ndp->ni_dvp != ndp->ni_vp) {
834 VOP_UNLOCK(ndp->ni_dvp, 0);
842 * Not a symbolic link that we will follow. Continue with the
843 * next component if there is any; otherwise, we're done.
845 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
846 ("lookup: invalid path state."));
847 if (*ndp->ni_next == '/') {
848 cnp->cn_nameptr = ndp->ni_next;
849 while (*cnp->cn_nameptr == '/') {
853 if (ndp->ni_dvp != dp)
860 * If we're processing a path with a trailing slash,
861 * check that the end result is a directory.
863 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
868 * Disallow directory write attempts on read-only filesystems.
871 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
875 if (cnp->cn_flags & SAVESTART) {
876 ndp->ni_startdir = ndp->ni_dvp;
877 VREF(ndp->ni_startdir);
881 if (ndp->ni_dvp != dp)
885 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
886 VOP_UNLOCK(ndp->ni_dvp, 0);
890 if (cnp->cn_flags & AUDITVNODE1)
891 AUDIT_ARG_VNODE1(dp);
892 else if (cnp->cn_flags & AUDITVNODE2)
893 AUDIT_ARG_VNODE2(dp);
895 if ((cnp->cn_flags & LOCKLEAF) == 0)
899 * Because of lookup_shared we may have the vnode shared locked, but
900 * the caller may want it to be exclusively locked.
902 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
903 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
904 vn_lock(dp, LK_UPGRADE | LK_RETRY);
905 if (dp->v_iflag & VI_DOOMED) {
913 if (ni_dvp_unlocked != 2) {
914 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
927 * relookup - lookup a path name component
928 * Used by lookup to re-acquire things.
931 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
933 struct vnode *dp = 0; /* the directory we are searching */
934 int wantparent; /* 1 => wantparent or lockparent flag */
935 int rdonly; /* lookup read-only flag bit */
938 KASSERT(cnp->cn_flags & ISLASTCN,
939 ("relookup: Not given last component."));
941 * Setup: break out flag bits into variables.
943 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
944 KASSERT(wantparent, ("relookup: parent not wanted."));
945 rdonly = cnp->cn_flags & RDONLY;
946 cnp->cn_flags &= ~ISSYMLINK;
948 cnp->cn_lkflags = LK_EXCLUSIVE;
949 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
952 * Search a new directory.
954 * The last component of the filename is left accessible via
955 * cnp->cn_nameptr for callers that need the name. Callers needing
956 * the name set the SAVENAME flag. When done, they assume
957 * responsibility for freeing the pathname buffer.
959 #ifdef NAMEI_DIAGNOSTIC
960 printf("{%s}: ", cnp->cn_nameptr);
964 * Check for "" which represents the root directory after slash
967 if (cnp->cn_nameptr[0] == '\0') {
969 * Support only LOOKUP for "/" because lookup()
970 * can't succeed for CREATE, DELETE and RENAME.
972 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
973 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
975 if (!(cnp->cn_flags & LOCKLEAF))
978 /* XXX This should probably move to the top of function. */
979 if (cnp->cn_flags & SAVESTART)
980 panic("lookup: SAVESTART");
984 if (cnp->cn_flags & ISDOTDOT)
985 panic ("relookup: lookup on dot-dot");
988 * We now have a segment name to search for, and a directory to search.
990 #ifdef NAMEI_DIAGNOSTIC
991 vprint("search in:", dp);
993 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
994 KASSERT(*vpp == NULL, ("leaf should be empty"));
995 if (error != EJUSTRETURN)
998 * If creating and at end of pathname, then can consider
999 * allowing file to be created.
1005 /* ASSERT(dvp == ndp->ni_startdir) */
1006 if (cnp->cn_flags & SAVESTART)
1008 if ((cnp->cn_flags & LOCKPARENT) == 0)
1011 * We return with ni_vp NULL to indicate that the entry
1012 * doesn't currently exist, leaving a pointer to the
1013 * (possibly locked) directory vnode in ndp->ni_dvp.
1021 * Disallow directory write attempts on read-only filesystems.
1024 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1033 * Set the parent lock/ref state to the requested state.
1035 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1040 } else if (!wantparent)
1043 * Check for symbolic link
1045 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1046 ("relookup: symlink found.\n"));
1048 /* ASSERT(dvp == ndp->ni_startdir) */
1049 if (cnp->cn_flags & SAVESTART)
1052 if ((cnp->cn_flags & LOCKLEAF) == 0)
1062 * Free data allocated by namei(); see namei(9) for details.
1065 NDFREE(struct nameidata *ndp, const u_int flags)
1073 if (!(flags & NDF_NO_FREE_PNBUF) &&
1074 (ndp->ni_cnd.cn_flags & HASBUF)) {
1075 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1076 ndp->ni_cnd.cn_flags &= ~HASBUF;
1078 if (!(flags & NDF_NO_VP_UNLOCK) &&
1079 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1081 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1090 VOP_UNLOCK(ndp->ni_vp, 0);
1091 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1092 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1093 ndp->ni_dvp != ndp->ni_vp)
1095 if (!(flags & NDF_NO_DVP_RELE) &&
1096 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1105 VOP_UNLOCK(ndp->ni_dvp, 0);
1106 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1107 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1108 vrele(ndp->ni_startdir);
1109 ndp->ni_startdir = NULL;
1114 * Determine if there is a suitable alternate filename under the specified
1115 * prefix for the specified path. If the create flag is set, then the
1116 * alternate prefix will be used so long as the parent directory exists.
1117 * This is used by the various compatiblity ABIs so that Linux binaries prefer
1118 * files under /compat/linux for example. The chosen path (whether under
1119 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1120 * to by pathbuf. The caller is responsible for free'ing the buffer from
1121 * the M_TEMP bucket if one is returned.
1124 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1125 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1127 struct nameidata nd, ndroot;
1128 char *ptr, *buf, *cp;
1132 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1135 /* Copy the prefix into the new pathname as a starting point. */
1136 len = strlcpy(buf, prefix, MAXPATHLEN);
1137 if (len >= MAXPATHLEN) {
1142 sz = MAXPATHLEN - len;
1145 /* Append the filename to the prefix. */
1146 if (pathseg == UIO_SYSSPACE)
1147 error = copystr(path, ptr, sz, &len);
1149 error = copyinstr(path, ptr, sz, &len);
1157 /* Only use a prefix with absolute pathnames. */
1163 if (dirfd != AT_FDCWD) {
1165 * We want the original because the "prefix" is
1166 * included in the already opened dirfd.
1168 bcopy(ptr, buf, len);
1173 * We know that there is a / somewhere in this pathname.
1174 * Search backwards for it, to find the file's parent dir
1175 * to see if it exists in the alternate tree. If it does,
1176 * and we want to create a file (cflag is set). We don't
1177 * need to worry about the root comparison in this case.
1181 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1184 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td);
1190 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td);
1197 * We now compare the vnode of the prefix to the one
1198 * vnode asked. If they resolve to be the same, then we
1199 * ignore the match so that the real root gets used.
1200 * This avoids the problem of traversing "../.." to find the
1201 * root directory and never finding it, because "/" resolves
1202 * to the emulation root directory. This is expensive :-(
1204 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1207 /* We shouldn't ever get an error from this namei(). */
1208 error = namei(&ndroot);
1210 if (nd.ni_vp == ndroot.ni_vp)
1213 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1214 vrele(ndroot.ni_vp);
1218 NDFREE(&nd, NDF_ONLY_PNBUF);
1222 /* If there was an error, use the original path name. */
1224 bcopy(ptr, buf, len);