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_ktrace.h"
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/capsicum.h>
47 #include <sys/fcntl.h>
50 #include <sys/mutex.h>
51 #include <sys/namei.h>
52 #include <sys/vnode.h>
53 #include <sys/mount.h>
54 #include <sys/filedesc.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysctl.h>
60 #include <sys/ktrace.h>
63 #include <security/audit/audit.h>
64 #include <security/mac/mac_framework.h>
68 #define NAMEI_DIAGNOSTIC 1
69 #undef NAMEI_DIAGNOSTIC
71 SDT_PROVIDER_DECLARE(vfs);
72 SDT_PROBE_DEFINE3(vfs, namei, lookup, entry, "struct vnode *", "char *",
74 SDT_PROBE_DEFINE2(vfs, namei, lookup, return, "int", "struct vnode *");
77 * Allocation zone for namei
79 uma_zone_t namei_zone;
81 * Placeholder vnode for mp traversal
83 static struct vnode *vp_crossmp;
86 nameiinit(void *dummy __unused)
89 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
91 getnewvnode("crossmp", NULL, &dead_vnodeops, &vp_crossmp);
92 vn_lock(vp_crossmp, LK_EXCLUSIVE);
93 VN_LOCK_ASHARE(vp_crossmp);
94 VOP_UNLOCK(vp_crossmp, 0);
96 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
98 static int lookup_shared = 1;
99 SYSCTL_INT(_vfs, OID_AUTO, lookup_shared, CTLFLAG_RW, &lookup_shared, 0,
100 "Enables/Disables shared locks for path name translation");
101 TUNABLE_INT("vfs.lookup_shared", &lookup_shared);
104 * Convert a pathname into a pointer to a locked vnode.
106 * The FOLLOW flag is set when symbolic links are to be followed
107 * when they occur at the end of the name translation process.
108 * Symbolic links are always followed for all other pathname
109 * components other than the last.
111 * The segflg defines whether the name is to be copied from user
112 * space or kernel space.
114 * Overall outline of namei:
117 * get starting directory
118 * while (!done && !error) {
119 * call lookup to search path.
120 * if symbolic link, massage name in buffer and continue
124 namei(struct nameidata *ndp)
126 struct filedesc *fdp; /* pointer to file descriptor state */
127 char *cp; /* pointer into pathname argument */
128 struct vnode *dp; /* the directory we are searching */
129 struct iovec aiov; /* uio for reading symbolic links */
132 struct componentname *cnp = &ndp->ni_cnd;
133 struct thread *td = cnp->cn_thread;
134 struct proc *p = td->td_proc;
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;
146 /* We will set this ourselves if we need it. */
147 cnp->cn_flags &= ~TRAILINGSLASH;
150 * Get a buffer for the name to be translated, and copy the
151 * name into the buffer.
153 if ((cnp->cn_flags & HASBUF) == 0)
154 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
155 if (ndp->ni_segflg == UIO_SYSSPACE)
156 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf,
157 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
159 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf,
160 MAXPATHLEN, (size_t *)&ndp->ni_pathlen);
163 * Don't allow empty pathnames.
165 if (!error && *cnp->cn_pnbuf == '\0')
168 #ifdef CAPABILITY_MODE
170 * In capability mode, lookups must be "strictly relative" (i.e.
171 * not an absolute path, and not containing '..' components) to
172 * a real file descriptor, not the pseudo-descriptor AT_FDCWD.
174 if (error == 0 && IN_CAPABILITY_MODE(td) &&
175 (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, NULL, NULL);
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) {
226 rights = ndp->ni_rightsneeded;
227 cap_rights_set(&rights, CAP_LOOKUP);
229 if (cnp->cn_flags & AUDITVNODE1)
230 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
231 if (cnp->cn_flags & AUDITVNODE2)
232 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
233 error = fgetvp_rights(td, ndp->ni_dirfd,
234 &rights, &ndp->ni_filecaps, &dp);
237 * If file descriptor doesn't have all rights,
238 * all lookups relative to it must also be
242 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights,
244 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
245 ndp->ni_filecaps.fc_nioctls != -1) {
246 ndp->ni_strictrelative = 1;
250 if (error != 0 || dp != NULL) {
251 FILEDESC_SUNLOCK(fdp);
252 if (error == 0 && dp->v_type != VDIR) {
258 uma_zfree(namei_zone, cnp->cn_pnbuf);
260 cnp->cn_pnbuf = NULL;
261 cnp->cn_nameptr = NULL;
269 FILEDESC_SUNLOCK(fdp);
270 if (ndp->ni_startdir != NULL)
271 vrele(ndp->ni_startdir);
273 SDT_PROBE(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
274 cnp->cn_flags, 0, 0);
277 * Check if root directory should replace current directory.
278 * Done at start of translation and after symbolic link.
280 cnp->cn_nameptr = cnp->cn_pnbuf;
281 if (*(cnp->cn_nameptr) == '/') {
283 if (ndp->ni_strictrelative != 0) {
285 if (KTRPOINT(curthread, KTR_CAPFAIL))
286 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
288 return (ENOTCAPABLE);
290 while (*(cnp->cn_nameptr) == '/') {
294 dp = ndp->ni_rootdir;
297 ndp->ni_startdir = dp;
300 uma_zfree(namei_zone, cnp->cn_pnbuf);
302 cnp->cn_pnbuf = NULL;
303 cnp->cn_nameptr = NULL;
305 SDT_PROBE(vfs, namei, lookup, return, error, NULL, 0,
310 * If not a symbolic link, we're done.
312 if ((cnp->cn_flags & ISSYMLINK) == 0) {
313 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
314 uma_zfree(namei_zone, cnp->cn_pnbuf);
316 cnp->cn_pnbuf = NULL;
317 cnp->cn_nameptr = NULL;
320 cnp->cn_flags |= HASBUF;
322 SDT_PROBE(vfs, namei, lookup, return, 0, ndp->ni_vp,
326 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
331 if ((cnp->cn_flags & NOMACCHECK) == 0) {
332 error = mac_vnode_check_readlink(td->td_ucred,
338 if (ndp->ni_pathlen > 1)
339 cp = uma_zalloc(namei_zone, M_WAITOK);
343 aiov.iov_len = MAXPATHLEN;
344 auio.uio_iov = &aiov;
347 auio.uio_rw = UIO_READ;
348 auio.uio_segflg = UIO_SYSSPACE;
350 auio.uio_resid = MAXPATHLEN;
351 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
353 if (ndp->ni_pathlen > 1)
354 uma_zfree(namei_zone, cp);
357 linklen = MAXPATHLEN - auio.uio_resid;
359 if (ndp->ni_pathlen > 1)
360 uma_zfree(namei_zone, cp);
364 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
365 if (ndp->ni_pathlen > 1)
366 uma_zfree(namei_zone, cp);
367 error = ENAMETOOLONG;
370 if (ndp->ni_pathlen > 1) {
371 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
372 uma_zfree(namei_zone, cnp->cn_pnbuf);
375 cnp->cn_pnbuf[linklen] = '\0';
376 ndp->ni_pathlen += linklen;
380 uma_zfree(namei_zone, cnp->cn_pnbuf);
382 cnp->cn_pnbuf = NULL;
383 cnp->cn_nameptr = NULL;
388 SDT_PROBE(vfs, namei, lookup, return, error, NULL, 0, 0, 0);
393 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
396 if (mp == NULL || ((lkflags & LK_SHARED) &&
397 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
398 ((cnflags & ISDOTDOT) &&
399 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
400 lkflags &= ~LK_SHARED;
401 lkflags |= LK_EXCLUSIVE;
407 needs_exclusive_leaf(struct mount *mp, int flags)
411 * Intermediate nodes can use shared locks, we only need to
412 * force an exclusive lock for leaf nodes.
414 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
417 /* Always use exclusive locks if LOCKSHARED isn't set. */
418 if (!(flags & LOCKSHARED))
422 * For lookups during open(), if the mount point supports
423 * extended shared operations, then use a shared lock for the
424 * leaf node, otherwise use an exclusive lock.
426 if ((flags & ISOPEN) != 0)
427 return (!MNT_EXTENDED_SHARED(mp));
430 * Lookup requests outside of open() that specify LOCKSHARED
431 * only need a shared lock on the leaf vnode.
438 * This is a very central and rather complicated routine.
440 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
441 * The starting directory is taken from ni_startdir. The pathname is
442 * descended until done, or a symbolic link is encountered. The variable
443 * ni_more is clear if the path is completed; it is set to one if a
444 * symbolic link needing interpretation is encountered.
446 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
447 * whether the name is to be looked up, created, renamed, or deleted.
448 * When CREATE, RENAME, or DELETE is specified, information usable in
449 * creating, renaming, or deleting a directory entry may be calculated.
450 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
451 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
452 * returned unlocked. Otherwise the parent directory is not returned. If
453 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
454 * the target is returned locked, otherwise it is returned unlocked.
455 * When creating or renaming and LOCKPARENT is specified, the target may not
456 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
458 * Overall outline of lookup:
461 * identify next component of name at ndp->ni_ptr
462 * handle degenerate case where name is null string
463 * if .. and crossing mount points and on mounted filesys, find parent
464 * call VOP_LOOKUP routine for next component name
465 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
466 * component vnode returned in ni_vp (if it exists), locked.
467 * if result vnode is mounted on and crossing mount points,
468 * find mounted on vnode
469 * if more components of name, do next level at dirloop
470 * return the answer in ni_vp, locked if LOCKLEAF set
471 * if LOCKPARENT set, return locked parent in ni_dvp
472 * if WANTPARENT set, return unlocked parent in ni_dvp
475 lookup(struct nameidata *ndp)
477 char *cp; /* pointer into pathname argument */
478 struct vnode *dp = 0; /* the directory we are searching */
479 struct vnode *tdp; /* saved dp */
480 struct mount *mp; /* mount table entry */
482 int docache; /* == 0 do not cache last component */
483 int wantparent; /* 1 => wantparent or lockparent flag */
484 int rdonly; /* lookup read-only flag bit */
486 int dpunlocked = 0; /* dp has already been unlocked */
487 struct componentname *cnp = &ndp->ni_cnd;
492 * Setup: break out flag bits into variables.
495 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
496 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
497 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
498 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
499 if (cnp->cn_nameiop == DELETE ||
500 (wantparent && cnp->cn_nameiop != CREATE &&
501 cnp->cn_nameiop != LOOKUP))
503 rdonly = cnp->cn_flags & RDONLY;
504 cnp->cn_flags &= ~ISSYMLINK;
507 * We use shared locks until we hit the parent of the last cn then
508 * we adjust based on the requesting flags.
511 cnp->cn_lkflags = LK_SHARED;
513 cnp->cn_lkflags = LK_EXCLUSIVE;
514 dp = ndp->ni_startdir;
515 ndp->ni_startdir = NULLVP;
517 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
522 * Search a new directory.
524 * The last component of the filename is left accessible via
525 * cnp->cn_nameptr for callers that need the name. Callers needing
526 * the name set the SAVENAME flag. When done, they assume
527 * responsibility for freeing the pathname buffer.
530 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
532 cnp->cn_namelen = cp - cnp->cn_nameptr;
533 if (cnp->cn_namelen > NAME_MAX) {
534 error = ENAMETOOLONG;
537 #ifdef NAMEI_DIAGNOSTIC
540 printf("{%s}: ", cnp->cn_nameptr);
543 ndp->ni_pathlen -= cnp->cn_namelen;
547 * Replace multiple slashes by a single slash and trailing slashes
548 * by a null. This must be done before VOP_LOOKUP() because some
549 * fs's don't know about trailing slashes. Remember if there were
550 * trailing slashes to handle symlinks, existing non-directories
551 * and non-existing files that won't be directories specially later.
553 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
557 *ndp->ni_next = '\0';
558 cnp->cn_flags |= TRAILINGSLASH;
563 cnp->cn_flags |= MAKEENTRY;
564 if (*cp == '\0' && docache == 0)
565 cnp->cn_flags &= ~MAKEENTRY;
566 if (cnp->cn_namelen == 2 &&
567 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
568 cnp->cn_flags |= ISDOTDOT;
570 cnp->cn_flags &= ~ISDOTDOT;
571 if (*ndp->ni_next == 0)
572 cnp->cn_flags |= ISLASTCN;
574 cnp->cn_flags &= ~ISLASTCN;
576 if ((cnp->cn_flags & ISLASTCN) != 0 &&
577 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
578 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
584 * Check for degenerate name (e.g. / or "")
585 * which is a way of talking about a directory,
586 * e.g. like "/." or ".".
588 if (cnp->cn_nameptr[0] == '\0') {
589 if (dp->v_type != VDIR) {
593 if (cnp->cn_nameiop != LOOKUP) {
603 if (cnp->cn_flags & AUDITVNODE1)
604 AUDIT_ARG_VNODE1(dp);
605 else if (cnp->cn_flags & AUDITVNODE2)
606 AUDIT_ARG_VNODE2(dp);
608 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
610 /* XXX This should probably move to the top of function. */
611 if (cnp->cn_flags & SAVESTART)
612 panic("lookup: SAVESTART");
617 * Handle "..": five special cases.
618 * 0. If doing a capability lookup, return ENOTCAPABLE (this is a
619 * fairly conservative design choice, but it's the only one that we
620 * are satisfied guarantees the property we're looking for).
621 * 1. Return an error if this is the last component of
622 * the name and the operation is DELETE or RENAME.
623 * 2. If at root directory (e.g. after chroot)
624 * or at absolute root directory
625 * then ignore it so can't get out.
626 * 3. If this vnode is the root of a mounted
627 * filesystem, then replace it with the
628 * vnode which was mounted on so we take the
629 * .. in the other filesystem.
630 * 4. If the vnode is the top directory of
631 * the jail or chroot, don't let them out.
633 if (cnp->cn_flags & ISDOTDOT) {
634 if (ndp->ni_strictrelative != 0) {
636 if (KTRPOINT(curthread, KTR_CAPFAIL))
637 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
642 if ((cnp->cn_flags & ISLASTCN) != 0 &&
643 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
648 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
650 if (dp == pr->pr_root)
652 if (dp == ndp->ni_rootdir ||
653 dp == ndp->ni_topdir ||
656 ((dp->v_vflag & VV_ROOT) != 0 &&
657 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
663 if ((dp->v_vflag & VV_ROOT) == 0)
665 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
670 dp = dp->v_mount->mnt_vnodecovered;
674 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
675 LK_RETRY, ISDOTDOT));
680 * We now have a segment name to search for, and a directory to search.
684 if ((cnp->cn_flags & NOMACCHECK) == 0) {
685 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
693 ASSERT_VOP_LOCKED(dp, "lookup");
695 * If we have a shared lock we may need to upgrade the lock for the
698 if (dp != vp_crossmp &&
699 VOP_ISLOCKED(dp) == LK_SHARED &&
700 (cnp->cn_flags & ISLASTCN) && (cnp->cn_flags & LOCKPARENT))
701 vn_lock(dp, LK_UPGRADE|LK_RETRY);
702 if ((dp->v_iflag & VI_DOOMED) != 0) {
707 * If we're looking up the last component and we need an exclusive
708 * lock, adjust our lkflags.
710 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
711 cnp->cn_lkflags = LK_EXCLUSIVE;
712 #ifdef NAMEI_DIAGNOSTIC
713 vprint("lookup in", dp);
715 lkflags_save = cnp->cn_lkflags;
716 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
718 if ((error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp)) != 0) {
719 cnp->cn_lkflags = lkflags_save;
720 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
721 #ifdef NAMEI_DIAGNOSTIC
722 printf("not found\n");
724 if ((error == ENOENT) &&
725 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
726 (dp->v_mount->mnt_flag & MNT_UNION)) {
728 dp = dp->v_mount->mnt_vnodecovered;
732 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
733 LK_RETRY, cnp->cn_flags));
737 if (error != EJUSTRETURN)
740 * At this point, we know we're at the end of the
741 * pathname. If creating / renaming, we can consider
742 * allowing the file or directory to be created / renamed,
743 * provided we're not on a read-only filesystem.
749 /* trailing slash only allowed for directories */
750 if ((cnp->cn_flags & TRAILINGSLASH) &&
751 !(cnp->cn_flags & WILLBEDIR)) {
755 if ((cnp->cn_flags & LOCKPARENT) == 0)
758 * We return with ni_vp NULL to indicate that the entry
759 * doesn't currently exist, leaving a pointer to the
760 * (possibly locked) directory vnode in ndp->ni_dvp.
762 if (cnp->cn_flags & SAVESTART) {
763 ndp->ni_startdir = ndp->ni_dvp;
764 VREF(ndp->ni_startdir);
768 cnp->cn_lkflags = lkflags_save;
769 #ifdef NAMEI_DIAGNOSTIC
773 * Take into account any additional components consumed by
774 * the underlying filesystem.
776 if (cnp->cn_consume > 0) {
777 cnp->cn_nameptr += cnp->cn_consume;
778 ndp->ni_next += cnp->cn_consume;
779 ndp->ni_pathlen -= cnp->cn_consume;
786 * Check to see if the vnode has been mounted on;
787 * if so find the root of the mounted filesystem.
789 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
790 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
794 if (dp != ndp->ni_dvp)
799 ndp->ni_dvp = vp_crossmp;
800 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
801 cnp->cn_flags), &tdp);
803 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
804 panic("vp_crossmp exclusively locked or reclaimed");
809 ndp->ni_vp = dp = tdp;
813 * Check for symbolic link
815 if ((dp->v_type == VLNK) &&
816 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
817 *ndp->ni_next == '/')) {
818 cnp->cn_flags |= ISSYMLINK;
819 if (dp->v_iflag & VI_DOOMED) {
821 * We can't know whether the directory was mounted with
822 * NOSYMFOLLOW, so we can't follow safely.
827 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
832 * Symlink code always expects an unlocked dvp.
834 if (ndp->ni_dvp != ndp->ni_vp) {
835 VOP_UNLOCK(ndp->ni_dvp, 0);
843 * Not a symbolic link that we will follow. Continue with the
844 * next component if there is any; otherwise, we're done.
846 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
847 ("lookup: invalid path state."));
848 if (*ndp->ni_next == '/') {
849 cnp->cn_nameptr = ndp->ni_next;
850 while (*cnp->cn_nameptr == '/') {
854 if (ndp->ni_dvp != dp)
861 * If we're processing a path with a trailing slash,
862 * check that the end result is a directory.
864 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
869 * Disallow directory write attempts on read-only filesystems.
872 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
876 if (cnp->cn_flags & SAVESTART) {
877 ndp->ni_startdir = ndp->ni_dvp;
878 VREF(ndp->ni_startdir);
882 if (ndp->ni_dvp != dp)
886 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
887 VOP_UNLOCK(ndp->ni_dvp, 0);
891 if (cnp->cn_flags & AUDITVNODE1)
892 AUDIT_ARG_VNODE1(dp);
893 else if (cnp->cn_flags & AUDITVNODE2)
894 AUDIT_ARG_VNODE2(dp);
896 if ((cnp->cn_flags & LOCKLEAF) == 0)
900 * Because of lookup_shared we may have the vnode shared locked, but
901 * the caller may want it to be exclusively locked.
903 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
904 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
905 vn_lock(dp, LK_UPGRADE | LK_RETRY);
906 if (dp->v_iflag & VI_DOOMED) {
914 if (ni_dvp_unlocked != 2) {
915 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
928 * relookup - lookup a path name component
929 * Used by lookup to re-acquire things.
932 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
934 struct vnode *dp = 0; /* the directory we are searching */
935 int wantparent; /* 1 => wantparent or lockparent flag */
936 int rdonly; /* lookup read-only flag bit */
939 KASSERT(cnp->cn_flags & ISLASTCN,
940 ("relookup: Not given last component."));
942 * Setup: break out flag bits into variables.
944 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
945 KASSERT(wantparent, ("relookup: parent not wanted."));
946 rdonly = cnp->cn_flags & RDONLY;
947 cnp->cn_flags &= ~ISSYMLINK;
949 cnp->cn_lkflags = LK_EXCLUSIVE;
950 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
953 * Search a new directory.
955 * The last component of the filename is left accessible via
956 * cnp->cn_nameptr for callers that need the name. Callers needing
957 * the name set the SAVENAME flag. When done, they assume
958 * responsibility for freeing the pathname buffer.
960 #ifdef NAMEI_DIAGNOSTIC
961 printf("{%s}: ", cnp->cn_nameptr);
965 * Check for "" which represents the root directory after slash
968 if (cnp->cn_nameptr[0] == '\0') {
970 * Support only LOOKUP for "/" because lookup()
971 * can't succeed for CREATE, DELETE and RENAME.
973 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
974 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
976 if (!(cnp->cn_flags & LOCKLEAF))
979 /* XXX This should probably move to the top of function. */
980 if (cnp->cn_flags & SAVESTART)
981 panic("lookup: SAVESTART");
985 if (cnp->cn_flags & ISDOTDOT)
986 panic ("relookup: lookup on dot-dot");
989 * We now have a segment name to search for, and a directory to search.
991 #ifdef NAMEI_DIAGNOSTIC
992 vprint("search in:", dp);
994 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
995 KASSERT(*vpp == NULL, ("leaf should be empty"));
996 if (error != EJUSTRETURN)
999 * If creating and at end of pathname, then can consider
1000 * allowing file to be created.
1006 /* ASSERT(dvp == ndp->ni_startdir) */
1007 if (cnp->cn_flags & SAVESTART)
1009 if ((cnp->cn_flags & LOCKPARENT) == 0)
1012 * We return with ni_vp NULL to indicate that the entry
1013 * doesn't currently exist, leaving a pointer to the
1014 * (possibly locked) directory vnode in ndp->ni_dvp.
1022 * Disallow directory write attempts on read-only filesystems.
1025 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1034 * Set the parent lock/ref state to the requested state.
1036 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1041 } else if (!wantparent)
1044 * Check for symbolic link
1046 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1047 ("relookup: symlink found.\n"));
1049 /* ASSERT(dvp == ndp->ni_startdir) */
1050 if (cnp->cn_flags & SAVESTART)
1053 if ((cnp->cn_flags & LOCKLEAF) == 0)
1063 NDINIT_ALL(struct nameidata *ndp, u_long op, u_long flags, enum uio_seg segflg,
1064 const char *namep, int dirfd, struct vnode *startdir, cap_rights_t *rightsp,
1068 ndp->ni_cnd.cn_nameiop = op;
1069 ndp->ni_cnd.cn_flags = flags;
1070 ndp->ni_segflg = segflg;
1071 ndp->ni_dirp = namep;
1072 ndp->ni_dirfd = dirfd;
1073 ndp->ni_startdir = startdir;
1074 ndp->ni_strictrelative = 0;
1075 if (rightsp != NULL)
1076 ndp->ni_rightsneeded = *rightsp;
1078 cap_rights_init(&ndp->ni_rightsneeded);
1079 filecaps_init(&ndp->ni_filecaps);
1080 ndp->ni_cnd.cn_thread = td;
1084 * Free data allocated by namei(); see namei(9) for details.
1087 NDFREE(struct nameidata *ndp, const u_int flags)
1095 if (!(flags & NDF_NO_FREE_PNBUF) &&
1096 (ndp->ni_cnd.cn_flags & HASBUF)) {
1097 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1098 ndp->ni_cnd.cn_flags &= ~HASBUF;
1100 if (!(flags & NDF_NO_VP_UNLOCK) &&
1101 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1103 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1112 VOP_UNLOCK(ndp->ni_vp, 0);
1113 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1114 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1115 ndp->ni_dvp != ndp->ni_vp)
1117 if (!(flags & NDF_NO_DVP_RELE) &&
1118 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1127 VOP_UNLOCK(ndp->ni_dvp, 0);
1128 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1129 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1130 vrele(ndp->ni_startdir);
1131 ndp->ni_startdir = NULL;
1136 * Determine if there is a suitable alternate filename under the specified
1137 * prefix for the specified path. If the create flag is set, then the
1138 * alternate prefix will be used so long as the parent directory exists.
1139 * This is used by the various compatiblity ABIs so that Linux binaries prefer
1140 * files under /compat/linux for example. The chosen path (whether under
1141 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1142 * to by pathbuf. The caller is responsible for free'ing the buffer from
1143 * the M_TEMP bucket if one is returned.
1146 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1147 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1149 struct nameidata nd, ndroot;
1150 char *ptr, *buf, *cp;
1154 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1157 /* Copy the prefix into the new pathname as a starting point. */
1158 len = strlcpy(buf, prefix, MAXPATHLEN);
1159 if (len >= MAXPATHLEN) {
1164 sz = MAXPATHLEN - len;
1167 /* Append the filename to the prefix. */
1168 if (pathseg == UIO_SYSSPACE)
1169 error = copystr(path, ptr, sz, &len);
1171 error = copyinstr(path, ptr, sz, &len);
1179 /* Only use a prefix with absolute pathnames. */
1185 if (dirfd != AT_FDCWD) {
1187 * We want the original because the "prefix" is
1188 * included in the already opened dirfd.
1190 bcopy(ptr, buf, len);
1195 * We know that there is a / somewhere in this pathname.
1196 * Search backwards for it, to find the file's parent dir
1197 * to see if it exists in the alternate tree. If it does,
1198 * and we want to create a file (cflag is set). We don't
1199 * need to worry about the root comparison in this case.
1203 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1206 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td);
1212 NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, buf, td);
1219 * We now compare the vnode of the prefix to the one
1220 * vnode asked. If they resolve to be the same, then we
1221 * ignore the match so that the real root gets used.
1222 * This avoids the problem of traversing "../.." to find the
1223 * root directory and never finding it, because "/" resolves
1224 * to the emulation root directory. This is expensive :-(
1226 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1229 /* We shouldn't ever get an error from this namei(). */
1230 error = namei(&ndroot);
1232 if (nd.ni_vp == ndroot.ni_vp)
1235 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1236 vrele(ndroot.ni_vp);
1240 NDFREE(&nd, NDF_ONLY_PNBUF);
1244 /* If there was an error, use the original path name. */
1246 bcopy(ptr, buf, len);