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);
165 * If we are auditing the kernel pathname, save the user
168 if (cnp->cn_flags & AUDITVNODE1)
169 AUDIT_ARG_UPATH1(td, cnp->cn_pnbuf);
170 if (cnp->cn_flags & AUDITVNODE2)
171 AUDIT_ARG_UPATH2(td, cnp->cn_pnbuf);
175 * Don't allow empty pathnames.
177 if (!error && *cnp->cn_pnbuf == '\0')
180 #ifdef CAPABILITY_MODE
182 * In capability mode, lookups must be "strictly relative" (i.e.
183 * not an absolute path, and not containing '..' components) to
184 * a real file descriptor, not the pseudo-descriptor AT_FDCWD.
186 if (IN_CAPABILITY_MODE(td)) {
187 ndp->ni_strictrelative = 1;
188 if (ndp->ni_dirfd == AT_FDCWD) {
190 if (KTRPOINT(td, KTR_CAPFAIL))
191 ktrcapfail(CAPFAIL_LOOKUP, 0, 0);
198 uma_zfree(namei_zone, cnp->cn_pnbuf);
200 cnp->cn_pnbuf = NULL;
201 cnp->cn_nameptr = NULL;
208 if (KTRPOINT(td, KTR_NAMEI)) {
209 KASSERT(cnp->cn_thread == curthread,
210 ("namei not using curthread"));
211 ktrnamei(cnp->cn_pnbuf);
215 * Get starting point for the translation.
218 ndp->ni_rootdir = fdp->fd_rdir;
219 ndp->ni_topdir = fdp->fd_jdir;
222 if (cnp->cn_pnbuf[0] != '/') {
223 if (ndp->ni_startdir != NULL) {
224 dp = ndp->ni_startdir;
226 } else if (ndp->ni_dirfd != AT_FDCWD) {
227 if (cnp->cn_flags & AUDITVNODE1)
228 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
229 if (cnp->cn_flags & AUDITVNODE2)
230 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
231 error = fgetvp_rights(td, ndp->ni_dirfd,
232 ndp->ni_rightsneeded | CAP_LOOKUP,
233 &(ndp->ni_baserights), &dp);
236 * Lookups relative to a capability must also be
239 * Note that a capability with rights CAP_MASK_VALID
240 * is treated exactly like a regular file descriptor.
242 if (ndp->ni_baserights != CAP_MASK_VALID)
243 ndp->ni_strictrelative = 1;
246 if (error != 0 || dp != NULL) {
247 FILEDESC_SUNLOCK(fdp);
248 if (error == 0 && dp->v_type != VDIR) {
254 uma_zfree(namei_zone, cnp->cn_pnbuf);
256 cnp->cn_pnbuf = NULL;
257 cnp->cn_nameptr = NULL;
265 FILEDESC_SUNLOCK(fdp);
266 if (ndp->ni_startdir != NULL)
267 vrele(ndp->ni_startdir);
269 SDT_PROBE(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
270 cnp->cn_flags, 0, 0);
273 * Check if root directory should replace current directory.
274 * Done at start of translation and after symbolic link.
276 cnp->cn_nameptr = cnp->cn_pnbuf;
277 if (*(cnp->cn_nameptr) == '/') {
279 if (ndp->ni_strictrelative != 0) {
281 if (KTRPOINT(curthread, KTR_CAPFAIL))
282 ktrcapfail(CAPFAIL_LOOKUP, 0, 0);
284 return (ENOTCAPABLE);
286 while (*(cnp->cn_nameptr) == '/') {
290 dp = ndp->ni_rootdir;
293 ndp->ni_startdir = dp;
296 uma_zfree(namei_zone, cnp->cn_pnbuf);
298 cnp->cn_pnbuf = NULL;
299 cnp->cn_nameptr = NULL;
301 SDT_PROBE(vfs, namei, lookup, return, error, NULL, 0,
306 * If not a symbolic link, we're done.
308 if ((cnp->cn_flags & ISSYMLINK) == 0) {
309 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
310 uma_zfree(namei_zone, cnp->cn_pnbuf);
312 cnp->cn_pnbuf = NULL;
313 cnp->cn_nameptr = NULL;
316 cnp->cn_flags |= HASBUF;
318 SDT_PROBE(vfs, namei, lookup, return, 0, ndp->ni_vp,
322 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
327 if ((cnp->cn_flags & NOMACCHECK) == 0) {
328 error = mac_vnode_check_readlink(td->td_ucred,
334 if (ndp->ni_pathlen > 1)
335 cp = uma_zalloc(namei_zone, M_WAITOK);
339 aiov.iov_len = MAXPATHLEN;
340 auio.uio_iov = &aiov;
343 auio.uio_rw = UIO_READ;
344 auio.uio_segflg = UIO_SYSSPACE;
345 auio.uio_td = (struct thread *)0;
346 auio.uio_resid = MAXPATHLEN;
347 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
349 if (ndp->ni_pathlen > 1)
350 uma_zfree(namei_zone, cp);
353 linklen = MAXPATHLEN - auio.uio_resid;
355 if (ndp->ni_pathlen > 1)
356 uma_zfree(namei_zone, cp);
360 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
361 if (ndp->ni_pathlen > 1)
362 uma_zfree(namei_zone, cp);
363 error = ENAMETOOLONG;
366 if (ndp->ni_pathlen > 1) {
367 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
368 uma_zfree(namei_zone, cnp->cn_pnbuf);
371 cnp->cn_pnbuf[linklen] = '\0';
372 ndp->ni_pathlen += linklen;
376 uma_zfree(namei_zone, cnp->cn_pnbuf);
378 cnp->cn_pnbuf = NULL;
379 cnp->cn_nameptr = NULL;
384 SDT_PROBE(vfs, namei, lookup, return, error, NULL, 0, 0, 0);
389 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
392 if (mp == NULL || ((lkflags & LK_SHARED) &&
393 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
394 ((cnflags & ISDOTDOT) &&
395 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
396 lkflags &= ~LK_SHARED;
397 lkflags |= LK_EXCLUSIVE;
403 needs_exclusive_leaf(struct mount *mp, int flags)
407 * Intermediate nodes can use shared locks, we only need to
408 * force an exclusive lock for leaf nodes.
410 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
413 /* Always use exclusive locks if LOCKSHARED isn't set. */
414 if (!(flags & LOCKSHARED))
418 * For lookups during open(), if the mount point supports
419 * extended shared operations, then use a shared lock for the
420 * leaf node, otherwise use an exclusive lock.
422 if (flags & ISOPEN) {
424 (mp->mnt_kern_flag & MNTK_EXTENDED_SHARED))
431 * Lookup requests outside of open() that specify LOCKSHARED
432 * only need a shared lock on the leaf vnode.
439 * This is a very central and rather complicated routine.
441 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
442 * The starting directory is taken from ni_startdir. The pathname is
443 * descended until done, or a symbolic link is encountered. The variable
444 * ni_more is clear if the path is completed; it is set to one if a
445 * symbolic link needing interpretation is encountered.
447 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
448 * whether the name is to be looked up, created, renamed, or deleted.
449 * When CREATE, RENAME, or DELETE is specified, information usable in
450 * creating, renaming, or deleting a directory entry may be calculated.
451 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
452 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
453 * returned unlocked. Otherwise the parent directory is not returned. If
454 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
455 * the target is returned locked, otherwise it is returned unlocked.
456 * When creating or renaming and LOCKPARENT is specified, the target may not
457 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
459 * Overall outline of lookup:
462 * identify next component of name at ndp->ni_ptr
463 * handle degenerate case where name is null string
464 * if .. and crossing mount points and on mounted filesys, find parent
465 * call VOP_LOOKUP routine for next component name
466 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
467 * component vnode returned in ni_vp (if it exists), locked.
468 * if result vnode is mounted on and crossing mount points,
469 * find mounted on vnode
470 * if more components of name, do next level at dirloop
471 * return the answer in ni_vp, locked if LOCKLEAF set
472 * if LOCKPARENT set, return locked parent in ni_dvp
473 * if WANTPARENT set, return unlocked parent in ni_dvp
476 lookup(struct nameidata *ndp)
478 char *cp; /* pointer into pathname argument */
479 struct vnode *dp = 0; /* the directory we are searching */
480 struct vnode *tdp; /* saved dp */
481 struct mount *mp; /* mount table entry */
483 int docache; /* == 0 do not cache last component */
484 int wantparent; /* 1 => wantparent or lockparent flag */
485 int rdonly; /* lookup read-only flag bit */
487 int dpunlocked = 0; /* dp has already been unlocked */
488 struct componentname *cnp = &ndp->ni_cnd;
493 * Setup: break out flag bits into variables.
496 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
497 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
498 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
499 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
500 if (cnp->cn_nameiop == DELETE ||
501 (wantparent && cnp->cn_nameiop != CREATE &&
502 cnp->cn_nameiop != LOOKUP))
504 rdonly = cnp->cn_flags & RDONLY;
505 cnp->cn_flags &= ~ISSYMLINK;
508 * We use shared locks until we hit the parent of the last cn then
509 * we adjust based on the requesting flags.
512 cnp->cn_lkflags = LK_SHARED;
514 cnp->cn_lkflags = LK_EXCLUSIVE;
515 dp = ndp->ni_startdir;
516 ndp->ni_startdir = NULLVP;
518 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
523 * Search a new directory.
525 * The last component of the filename is left accessible via
526 * cnp->cn_nameptr for callers that need the name. Callers needing
527 * the name set the SAVENAME flag. When done, they assume
528 * responsibility for freeing the pathname buffer.
531 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
533 cnp->cn_namelen = cp - cnp->cn_nameptr;
534 if (cnp->cn_namelen > NAME_MAX) {
535 error = ENAMETOOLONG;
538 #ifdef NAMEI_DIAGNOSTIC
541 printf("{%s}: ", cnp->cn_nameptr);
544 ndp->ni_pathlen -= cnp->cn_namelen;
548 * Replace multiple slashes by a single slash and trailing slashes
549 * by a null. This must be done before VOP_LOOKUP() because some
550 * fs's don't know about trailing slashes. Remember if there were
551 * trailing slashes to handle symlinks, existing non-directories
552 * and non-existing files that won't be directories specially later.
554 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
558 *ndp->ni_next = '\0';
559 cnp->cn_flags |= TRAILINGSLASH;
564 cnp->cn_flags |= MAKEENTRY;
565 if (*cp == '\0' && docache == 0)
566 cnp->cn_flags &= ~MAKEENTRY;
567 if (cnp->cn_namelen == 2 &&
568 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
569 cnp->cn_flags |= ISDOTDOT;
571 cnp->cn_flags &= ~ISDOTDOT;
572 if (*ndp->ni_next == 0)
573 cnp->cn_flags |= ISLASTCN;
575 cnp->cn_flags &= ~ISLASTCN;
577 if ((cnp->cn_flags & ISLASTCN) != 0 &&
578 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
579 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
585 * Check for degenerate name (e.g. / or "")
586 * which is a way of talking about a directory,
587 * e.g. like "/." or ".".
589 if (cnp->cn_nameptr[0] == '\0') {
590 if (dp->v_type != VDIR) {
594 if (cnp->cn_nameiop != LOOKUP) {
604 if (cnp->cn_flags & AUDITVNODE1)
605 AUDIT_ARG_VNODE1(dp);
606 else if (cnp->cn_flags & AUDITVNODE2)
607 AUDIT_ARG_VNODE2(dp);
609 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
611 /* XXX This should probably move to the top of function. */
612 if (cnp->cn_flags & SAVESTART)
613 panic("lookup: SAVESTART");
618 * Handle "..": five special cases.
619 * 0. If doing a capability lookup, return ENOTCAPABLE (this is a
620 * fairly conservative design choice, but it's the only one that we
621 * are satisfied guarantees the property we're looking for).
622 * 1. Return an error if this is the last component of
623 * the name and the operation is DELETE or RENAME.
624 * 2. If at root directory (e.g. after chroot)
625 * or at absolute root directory
626 * then ignore it so can't get out.
627 * 3. If this vnode is the root of a mounted
628 * filesystem, then replace it with the
629 * vnode which was mounted on so we take the
630 * .. in the other filesystem.
631 * 4. If the vnode is the top directory of
632 * the jail or chroot, don't let them out.
634 if (cnp->cn_flags & ISDOTDOT) {
635 if (ndp->ni_strictrelative != 0) {
637 if (KTRPOINT(curthread, KTR_CAPFAIL))
638 ktrcapfail(CAPFAIL_LOOKUP, 0, 0);
643 if ((cnp->cn_flags & ISLASTCN) != 0 &&
644 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
649 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
651 if (dp == pr->pr_root)
653 if (dp == ndp->ni_rootdir ||
654 dp == ndp->ni_topdir ||
657 ((dp->v_vflag & VV_ROOT) != 0 &&
658 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
664 if ((dp->v_vflag & VV_ROOT) == 0)
666 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
671 dp = dp->v_mount->mnt_vnodecovered;
675 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
676 LK_RETRY, ISDOTDOT));
681 * We now have a segment name to search for, and a directory to search.
685 if ((cnp->cn_flags & NOMACCHECK) == 0) {
686 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
694 ASSERT_VOP_LOCKED(dp, "lookup");
696 * If we have a shared lock we may need to upgrade the lock for the
699 if (dp != vp_crossmp &&
700 VOP_ISLOCKED(dp) == LK_SHARED &&
701 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT))
702 vn_lock(dp, LK_UPGRADE|LK_RETRY);
704 * If we're looking up the last component and we need an exclusive
705 * lock, adjust our lkflags.
707 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
708 cnp->cn_lkflags = LK_EXCLUSIVE;
709 #ifdef NAMEI_DIAGNOSTIC
710 vprint("lookup in", dp);
712 lkflags_save = cnp->cn_lkflags;
713 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
715 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
716 cnp->cn_lkflags = lkflags_save;
717 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
718 #ifdef NAMEI_DIAGNOSTIC
719 printf("not found\n");
721 if ((error == ENOENT) &&
722 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
723 (dp->v_mount->mnt_flag & MNT_UNION)) {
725 dp = dp->v_mount->mnt_vnodecovered;
729 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
730 LK_RETRY, cnp->cn_flags));
734 if (error != EJUSTRETURN)
737 * At this point, we know we're at the end of the
738 * pathname. If creating / renaming, we can consider
739 * allowing the file or directory to be created / renamed,
740 * provided we're not on a read-only filesystem.
746 /* trailing slash only allowed for directories */
747 if ((cnp->cn_flags & TRAILINGSLASH) &&
748 !(cnp->cn_flags & WILLBEDIR)) {
752 if ((cnp->cn_flags & LOCKPARENT) == 0)
755 * We return with ni_vp NULL to indicate that the entry
756 * doesn't currently exist, leaving a pointer to the
757 * (possibly locked) directory vnode in ndp->ni_dvp.
759 if (cnp->cn_flags & SAVESTART) {
760 ndp->ni_startdir = ndp->ni_dvp;
761 VREF(ndp->ni_startdir);
765 cnp->cn_lkflags = lkflags_save;
766 #ifdef NAMEI_DIAGNOSTIC
770 * Take into account any additional components consumed by
771 * the underlying filesystem.
773 if (cnp->cn_consume > 0) {
774 cnp->cn_nameptr += cnp->cn_consume;
775 ndp->ni_next += cnp->cn_consume;
776 ndp->ni_pathlen -= cnp->cn_consume;
783 * Check to see if the vnode has been mounted on;
784 * if so find the root of the mounted filesystem.
786 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
787 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
791 if (dp != ndp->ni_dvp)
796 ndp->ni_dvp = vp_crossmp;
797 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
798 cnp->cn_flags), &tdp);
800 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
801 panic("vp_crossmp exclusively locked or reclaimed");
806 ndp->ni_vp = dp = tdp;
810 * Check for symbolic link
812 if ((dp->v_type == VLNK) &&
813 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
814 *ndp->ni_next == '/')) {
815 cnp->cn_flags |= ISSYMLINK;
816 if (dp->v_iflag & VI_DOOMED) {
818 * We can't know whether the directory was mounted with
819 * NOSYMFOLLOW, so we can't follow safely.
824 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
829 * Symlink code always expects an unlocked dvp.
831 if (ndp->ni_dvp != ndp->ni_vp) {
832 VOP_UNLOCK(ndp->ni_dvp, 0);
840 * Not a symbolic link that we will follow. Continue with the
841 * next component if there is any; otherwise, we're done.
843 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
844 ("lookup: invalid path state."));
845 if (*ndp->ni_next == '/') {
846 cnp->cn_nameptr = ndp->ni_next;
847 while (*cnp->cn_nameptr == '/') {
851 if (ndp->ni_dvp != dp)
858 * If we're processing a path with a trailing slash,
859 * check that the end result is a directory.
861 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
866 * Disallow directory write attempts on read-only filesystems.
869 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
873 if (cnp->cn_flags & SAVESTART) {
874 ndp->ni_startdir = ndp->ni_dvp;
875 VREF(ndp->ni_startdir);
879 if (ndp->ni_dvp != dp)
883 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
884 VOP_UNLOCK(ndp->ni_dvp, 0);
888 if (cnp->cn_flags & AUDITVNODE1)
889 AUDIT_ARG_VNODE1(dp);
890 else if (cnp->cn_flags & AUDITVNODE2)
891 AUDIT_ARG_VNODE2(dp);
893 if ((cnp->cn_flags & LOCKLEAF) == 0)
897 * Because of lookup_shared we may have the vnode shared locked, but
898 * the caller may want it to be exclusively locked.
900 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
901 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
902 vn_lock(dp, LK_UPGRADE | LK_RETRY);
903 if (dp->v_iflag & VI_DOOMED) {
911 if (ni_dvp_unlocked != 2) {
912 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
925 * relookup - lookup a path name component
926 * Used by lookup to re-acquire things.
929 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
931 struct vnode *dp = 0; /* the directory we are searching */
932 int wantparent; /* 1 => wantparent or lockparent flag */
933 int rdonly; /* lookup read-only flag bit */
936 KASSERT(cnp->cn_flags & ISLASTCN,
937 ("relookup: Not given last component."));
939 * Setup: break out flag bits into variables.
941 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
942 KASSERT(wantparent, ("relookup: parent not wanted."));
943 rdonly = cnp->cn_flags & RDONLY;
944 cnp->cn_flags &= ~ISSYMLINK;
946 cnp->cn_lkflags = LK_EXCLUSIVE;
947 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
950 * Search a new directory.
952 * The last component of the filename is left accessible via
953 * cnp->cn_nameptr for callers that need the name. Callers needing
954 * the name set the SAVENAME flag. When done, they assume
955 * responsibility for freeing the pathname buffer.
957 #ifdef NAMEI_DIAGNOSTIC
958 printf("{%s}: ", cnp->cn_nameptr);
962 * Check for "" which represents the root directory after slash
965 if (cnp->cn_nameptr[0] == '\0') {
967 * Support only LOOKUP for "/" because lookup()
968 * can't succeed for CREATE, DELETE and RENAME.
970 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
971 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
973 if (!(cnp->cn_flags & LOCKLEAF))
976 /* XXX This should probably move to the top of function. */
977 if (cnp->cn_flags & SAVESTART)
978 panic("lookup: SAVESTART");
982 if (cnp->cn_flags & ISDOTDOT)
983 panic ("relookup: lookup on dot-dot");
986 * We now have a segment name to search for, and a directory to search.
988 #ifdef NAMEI_DIAGNOSTIC
989 vprint("search in:", dp);
991 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
992 KASSERT(*vpp == NULL, ("leaf should be empty"));
993 if (error != EJUSTRETURN)
996 * If creating and at end of pathname, then can consider
997 * allowing file to be created.
1003 /* ASSERT(dvp == ndp->ni_startdir) */
1004 if (cnp->cn_flags & SAVESTART)
1006 if ((cnp->cn_flags & LOCKPARENT) == 0)
1009 * We return with ni_vp NULL to indicate that the entry
1010 * doesn't currently exist, leaving a pointer to the
1011 * (possibly locked) directory vnode in ndp->ni_dvp.
1019 * Disallow directory write attempts on read-only filesystems.
1022 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1031 * Set the parent lock/ref state to the requested state.
1033 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1038 } else if (!wantparent)
1041 * Check for symbolic link
1043 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1044 ("relookup: symlink found.\n"));
1046 /* ASSERT(dvp == ndp->ni_startdir) */
1047 if (cnp->cn_flags & SAVESTART)
1050 if ((cnp->cn_flags & LOCKLEAF) == 0)
1060 * Free data allocated by namei(); see namei(9) for details.
1063 NDFREE(struct nameidata *ndp, const u_int flags)
1071 if (!(flags & NDF_NO_FREE_PNBUF) &&
1072 (ndp->ni_cnd.cn_flags & HASBUF)) {
1073 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1074 ndp->ni_cnd.cn_flags &= ~HASBUF;
1076 if (!(flags & NDF_NO_VP_UNLOCK) &&
1077 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1079 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1088 VOP_UNLOCK(ndp->ni_vp, 0);
1089 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1090 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1091 ndp->ni_dvp != ndp->ni_vp)
1093 if (!(flags & NDF_NO_DVP_RELE) &&
1094 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1103 VOP_UNLOCK(ndp->ni_dvp, 0);
1104 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1105 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1106 vrele(ndp->ni_startdir);
1107 ndp->ni_startdir = NULL;
1112 * Determine if there is a suitable alternate filename under the specified
1113 * prefix for the specified path. If the create flag is set, then the
1114 * alternate prefix will be used so long as the parent directory exists.
1115 * This is used by the various compatiblity ABIs so that Linux binaries prefer
1116 * files under /compat/linux for example. The chosen path (whether under
1117 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1118 * to by pathbuf. The caller is responsible for free'ing the buffer from
1119 * the M_TEMP bucket if one is returned.
1122 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1123 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1125 struct nameidata nd, ndroot;
1126 char *ptr, *buf, *cp;
1130 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1133 /* Copy the prefix into the new pathname as a starting point. */
1134 len = strlcpy(buf, prefix, MAXPATHLEN);
1135 if (len >= MAXPATHLEN) {
1140 sz = MAXPATHLEN - len;
1143 /* Append the filename to the prefix. */
1144 if (pathseg == UIO_SYSSPACE)
1145 error = copystr(path, ptr, sz, &len);
1147 error = copyinstr(path, ptr, sz, &len);
1155 /* Only use a prefix with absolute pathnames. */
1161 if (dirfd != AT_FDCWD) {
1163 * We want the original because the "prefix" is
1164 * included in the already opened dirfd.
1166 bcopy(ptr, buf, len);
1171 * We know that there is a / somewhere in this pathname.
1172 * Search backwards for it, to find the file's parent dir
1173 * to see if it exists in the alternate tree. If it does,
1174 * and we want to create a file (cflag is set). We don't
1175 * need to worry about the root comparison in this case.
1179 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1182 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td);
1188 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td);
1195 * We now compare the vnode of the prefix to the one
1196 * vnode asked. If they resolve to be the same, then we
1197 * ignore the match so that the real root gets used.
1198 * This avoids the problem of traversing "../.." to find the
1199 * root directory and never finding it, because "/" resolves
1200 * to the emulation root directory. This is expensive :-(
1202 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1205 /* We shouldn't ever get an error from this namei(). */
1206 error = namei(&ndroot);
1208 if (nd.ni_vp == ndroot.ni_vp)
1211 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1212 vrele(ndroot.ni_vp);
1216 NDFREE(&nd, NDF_ONLY_PNBUF);
1220 /* If there was an error, use the original path name. */
1222 bcopy(ptr, buf, len);