2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)vfs_lookup.c 8.4 (Berkeley) 2/16/94
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
42 #include "opt_capsicum.h"
43 #include "opt_ktrace.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kernel.h>
48 #include <sys/capsicum.h>
49 #include <sys/fcntl.h>
52 #include <sys/mutex.h>
53 #include <sys/namei.h>
54 #include <sys/vnode.h>
55 #include <sys/mount.h>
56 #include <sys/filedesc.h>
59 #include <sys/syscallsubr.h>
60 #include <sys/sysctl.h>
62 #include <sys/ktrace.h>
65 #include <security/audit/audit.h>
66 #include <security/mac/mac_framework.h>
70 #define NAMEI_DIAGNOSTIC 1
71 #undef NAMEI_DIAGNOSTIC
73 SDT_PROVIDER_DECLARE(vfs);
74 SDT_PROBE_DEFINE3(vfs, namei, lookup, entry, "struct vnode *", "char *",
76 SDT_PROBE_DEFINE2(vfs, namei, lookup, return, "int", "struct vnode *");
78 /* Allocation zone for namei. */
79 uma_zone_t namei_zone;
81 /* Placeholder vnode for mp traversal. */
82 static struct vnode *vp_crossmp;
85 crossmp_vop_islocked(struct vop_islocked_args *ap)
92 crossmp_vop_lock1(struct vop_lock1_args *ap)
95 struct lock *lk __unused;
96 const char *file __unused;
97 int flags, line __unused;
105 if ((flags & LK_SHARED) == 0)
106 panic("invalid lock request for crossmp");
108 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER, file, line,
109 flags & LK_INTERLOCK ? &VI_MTX(vp)->lock_object : NULL);
110 WITNESS_LOCK(&lk->lock_object, 0, file, line);
111 if ((flags & LK_INTERLOCK) != 0)
113 LOCK_LOG_LOCK("SLOCK", &lk->lock_object, 0, 0, ap->a_file, line);
118 crossmp_vop_unlock(struct vop_unlock_args *ap)
121 struct lock *lk __unused;
126 WITNESS_UNLOCK(&lk->lock_object, 0, LOCK_FILE, LOCK_LINE);
127 LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, LOCK_FILE,
132 static struct vop_vector crossmp_vnodeops = {
133 .vop_default = &default_vnodeops,
134 .vop_islocked = crossmp_vop_islocked,
135 .vop_lock1 = crossmp_vop_lock1,
136 .vop_unlock = crossmp_vop_unlock,
139 * VFS_VOP_VECTOR_REGISTER(crossmp_vnodeops) is not used here since the vnode
140 * gets allocated early. See nameiinit for the direct call below.
143 struct nameicap_tracker {
145 TAILQ_ENTRY(nameicap_tracker) nm_link;
148 /* Zone for cap mode tracker elements used for dotdot capability checks. */
149 static uma_zone_t nt_zone;
152 nameiinit(void *dummy __unused)
155 namei_zone = uma_zcreate("NAMEI", MAXPATHLEN, NULL, NULL, NULL, NULL,
157 nt_zone = uma_zcreate("rentr", sizeof(struct nameicap_tracker),
158 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
159 vfs_vector_op_register(&crossmp_vnodeops);
160 getnewvnode("crossmp", NULL, &crossmp_vnodeops, &vp_crossmp);
162 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
164 static int lookup_cap_dotdot = 1;
165 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot, CTLFLAG_RWTUN,
166 &lookup_cap_dotdot, 0,
167 "enables \"..\" components in path lookup in capability mode");
168 static int lookup_cap_dotdot_nonlocal = 1;
169 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot_nonlocal, CTLFLAG_RWTUN,
170 &lookup_cap_dotdot_nonlocal, 0,
171 "enables \"..\" components in path lookup in capability mode "
172 "on non-local mount");
175 nameicap_tracker_add(struct nameidata *ndp, struct vnode *dp)
177 struct nameicap_tracker *nt;
179 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp->v_type != VDIR)
181 if ((ndp->ni_lcf & (NI_LCF_BENEATH_ABS | NI_LCF_BENEATH_LATCHED)) ==
182 NI_LCF_BENEATH_ABS) {
183 MPASS((ndp->ni_lcf & NI_LCF_LATCH) != 0);
184 if (dp != ndp->ni_beneath_latch)
186 ndp->ni_lcf |= NI_LCF_BENEATH_LATCHED;
188 nt = uma_zalloc(nt_zone, M_WAITOK);
191 TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link);
195 nameicap_cleanup(struct nameidata *ndp, bool clean_latch)
197 struct nameicap_tracker *nt, *nt1;
199 KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) ||
200 (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative"));
201 TAILQ_FOREACH_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) {
202 TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link);
204 uma_zfree(nt_zone, nt);
206 if (clean_latch && (ndp->ni_lcf & NI_LCF_LATCH) != 0) {
207 ndp->ni_lcf &= ~NI_LCF_LATCH;
208 vrele(ndp->ni_beneath_latch);
213 * For dotdot lookups in capability mode, only allow the component
214 * lookup to succeed if the resulting directory was already traversed
215 * during the operation. Also fail dotdot lookups for non-local
216 * filesystems, where external agents might assist local lookups to
217 * escape the compartment.
220 nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp)
222 struct nameicap_tracker *nt;
225 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp == NULL ||
229 if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL &&
230 (mp->mnt_flag & MNT_LOCAL) == 0)
231 return (ENOTCAPABLE);
232 TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head,
237 if ((ndp->ni_lcf & NI_LCF_BENEATH_ABS) != 0) {
238 ndp->ni_lcf &= ~NI_LCF_BENEATH_LATCHED;
239 nameicap_cleanup(ndp, false);
242 return (ENOTCAPABLE);
246 namei_cleanup_cnp(struct componentname *cnp)
249 uma_zfree(namei_zone, cnp->cn_pnbuf);
251 cnp->cn_pnbuf = NULL;
252 cnp->cn_nameptr = NULL;
257 namei_handle_root(struct nameidata *ndp, struct vnode **dpp, u_int n)
259 struct componentname *cnp;
262 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0) {
264 if (KTRPOINT(curthread, KTR_CAPFAIL))
265 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
267 return (ENOTCAPABLE);
269 if ((cnp->cn_flags & BENEATH) != 0) {
270 ndp->ni_lcf |= NI_LCF_BENEATH_ABS;
271 ndp->ni_lcf &= ~NI_LCF_BENEATH_LATCHED;
272 nameicap_cleanup(ndp, false);
274 while (*(cnp->cn_nameptr) == '/') {
278 *dpp = ndp->ni_rootdir;
284 * Convert a pathname into a pointer to a locked vnode.
286 * The FOLLOW flag is set when symbolic links are to be followed
287 * when they occur at the end of the name translation process.
288 * Symbolic links are always followed for all other pathname
289 * components other than the last.
291 * The segflg defines whether the name is to be copied from user
292 * space or kernel space.
294 * Overall outline of namei:
297 * get starting directory
298 * while (!done && !error) {
299 * call lookup to search path.
300 * if symbolic link, massage name in buffer and continue
304 namei(struct nameidata *ndp)
306 struct filedesc *fdp; /* pointer to file descriptor state */
307 char *cp; /* pointer into pathname argument */
308 struct vnode *dp; /* the directory we are searching */
309 struct iovec aiov; /* uio for reading symbolic links */
310 struct componentname *cnp;
314 struct filecaps dirfd_caps;
316 int error, linklen, startdir_used;
321 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
322 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
323 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
324 ("namei: nameiop contaminated with flags"));
325 KASSERT((cnp->cn_flags & OPMASK) == 0,
326 ("namei: flags contaminated with nameiops"));
327 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
328 ndp->ni_startdir->v_type == VBAD);
330 TAILQ_INIT(&ndp->ni_cap_tracker);
333 /* We will set this ourselves if we need it. */
334 cnp->cn_flags &= ~TRAILINGSLASH;
337 * Get a buffer for the name to be translated, and copy the
338 * name into the buffer.
340 if ((cnp->cn_flags & HASBUF) == 0)
341 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
342 if (ndp->ni_segflg == UIO_SYSSPACE)
343 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
346 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
350 * Don't allow empty pathnames.
352 if (error == 0 && *cnp->cn_pnbuf == '\0')
355 #ifdef CAPABILITY_MODE
357 * In capability mode, lookups must be restricted to happen in
358 * the subtree with the root specified by the file descriptor:
359 * - The root must be real file descriptor, not the pseudo-descriptor
361 * - The passed path must be relative and not absolute.
362 * - If lookup_cap_dotdot is disabled, path must not contain the
364 * - If lookup_cap_dotdot is enabled, we verify that all '..'
365 * components lookups result in the directories which were
366 * previously walked by us, which prevents an escape from
369 if (error == 0 && IN_CAPABILITY_MODE(td) &&
370 (cnp->cn_flags & NOCAPCHECK) == 0) {
371 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
372 if (ndp->ni_dirfd == AT_FDCWD) {
374 if (KTRPOINT(td, KTR_CAPFAIL))
375 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
382 namei_cleanup_cnp(cnp);
388 if (KTRPOINT(td, KTR_NAMEI)) {
389 KASSERT(cnp->cn_thread == curthread,
390 ("namei not using curthread"));
391 ktrnamei(cnp->cn_pnbuf);
395 * Get starting point for the translation.
399 * The reference on ni_rootdir is acquired in the block below to avoid
400 * back-to-back atomics for absolute lookups.
402 ndp->ni_rootdir = fdp->fd_rdir;
403 ndp->ni_topdir = fdp->fd_jdir;
406 * If we are auditing the kernel pathname, save the user pathname.
408 if (cnp->cn_flags & AUDITVNODE1)
409 AUDIT_ARG_UPATH1(td, ndp->ni_dirfd, cnp->cn_pnbuf);
410 if (cnp->cn_flags & AUDITVNODE2)
411 AUDIT_ARG_UPATH2(td, ndp->ni_dirfd, cnp->cn_pnbuf);
415 cnp->cn_nameptr = cnp->cn_pnbuf;
416 if (cnp->cn_pnbuf[0] == '/') {
417 ndp->ni_resflags |= NIRES_ABS;
418 error = namei_handle_root(ndp, &dp, 2);
421 * Simplify error handling, we should almost never be
424 vrefact(ndp->ni_rootdir);
427 if (ndp->ni_startdir != NULL) {
428 vrefact(ndp->ni_rootdir);
429 dp = ndp->ni_startdir;
431 } else if (ndp->ni_dirfd == AT_FDCWD) {
433 if (dp == ndp->ni_rootdir) {
436 vrefact(ndp->ni_rootdir);
440 vrefact(ndp->ni_rootdir);
441 rights = ndp->ni_rightsneeded;
442 cap_rights_set(&rights, CAP_LOOKUP);
444 if (cnp->cn_flags & AUDITVNODE1)
445 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
446 if (cnp->cn_flags & AUDITVNODE2)
447 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
448 error = fgetvp_rights(td, ndp->ni_dirfd,
449 &rights, &ndp->ni_filecaps, &dp);
454 * If file descriptor doesn't have all rights,
455 * all lookups relative to it must also be
459 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights,
461 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
462 ndp->ni_filecaps.fc_nioctls != -1) {
463 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
467 if (error == 0 && dp->v_type != VDIR)
470 if (error == 0 && (cnp->cn_flags & BENEATH) != 0) {
471 if (ndp->ni_dirfd == AT_FDCWD) {
472 ndp->ni_beneath_latch = fdp->fd_cdir;
473 vrefact(ndp->ni_beneath_latch);
475 rights = ndp->ni_rightsneeded;
476 cap_rights_set(&rights, CAP_LOOKUP);
477 error = fgetvp_rights(td, ndp->ni_dirfd, &rights,
478 &dirfd_caps, &ndp->ni_beneath_latch);
479 if (error == 0 && dp->v_type != VDIR) {
480 vrele(ndp->ni_beneath_latch);
485 ndp->ni_lcf |= NI_LCF_LATCH;
487 FILEDESC_SUNLOCK(fdp);
488 if (ndp->ni_startdir != NULL && !startdir_used)
489 vrele(ndp->ni_startdir);
495 MPASS((ndp->ni_lcf & (NI_LCF_BENEATH_ABS | NI_LCF_LATCH)) !=
497 if (((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 &&
498 lookup_cap_dotdot != 0) ||
499 ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) == 0 &&
500 (cnp->cn_flags & BENEATH) != 0))
501 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
502 SDT_PROBE3(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
505 ndp->ni_startdir = dp;
510 * If not a symbolic link, we're done.
512 if ((cnp->cn_flags & ISSYMLINK) == 0) {
513 vrele(ndp->ni_rootdir);
514 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
515 namei_cleanup_cnp(cnp);
517 cnp->cn_flags |= HASBUF;
518 if ((ndp->ni_lcf & (NI_LCF_BENEATH_ABS |
519 NI_LCF_BENEATH_LATCHED)) == NI_LCF_BENEATH_ABS) {
523 nameicap_cleanup(ndp, true);
524 SDT_PROBE2(vfs, namei, lookup, return, error,
525 (error == 0 ? ndp->ni_vp : NULL));
528 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
533 if ((cnp->cn_flags & NOMACCHECK) == 0) {
534 error = mac_vnode_check_readlink(td->td_ucred,
540 if (ndp->ni_pathlen > 1)
541 cp = uma_zalloc(namei_zone, M_WAITOK);
545 aiov.iov_len = MAXPATHLEN;
546 auio.uio_iov = &aiov;
549 auio.uio_rw = UIO_READ;
550 auio.uio_segflg = UIO_SYSSPACE;
552 auio.uio_resid = MAXPATHLEN;
553 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
555 if (ndp->ni_pathlen > 1)
556 uma_zfree(namei_zone, cp);
559 linklen = MAXPATHLEN - auio.uio_resid;
561 if (ndp->ni_pathlen > 1)
562 uma_zfree(namei_zone, cp);
566 if (linklen + ndp->ni_pathlen > MAXPATHLEN) {
567 if (ndp->ni_pathlen > 1)
568 uma_zfree(namei_zone, cp);
569 error = ENAMETOOLONG;
572 if (ndp->ni_pathlen > 1) {
573 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
574 uma_zfree(namei_zone, cnp->cn_pnbuf);
577 cnp->cn_pnbuf[linklen] = '\0';
578 ndp->ni_pathlen += linklen;
582 * Check if root directory should replace current directory.
584 cnp->cn_nameptr = cnp->cn_pnbuf;
585 if (*(cnp->cn_nameptr) == '/') {
587 error = namei_handle_root(ndp, &dp, 1);
596 vrele(ndp->ni_rootdir);
598 namei_cleanup_cnp(cnp);
599 nameicap_cleanup(ndp, true);
600 SDT_PROBE2(vfs, namei, lookup, return, error, NULL);
605 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
608 if (mp == NULL || ((lkflags & LK_SHARED) &&
609 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
610 ((cnflags & ISDOTDOT) &&
611 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
612 lkflags &= ~LK_SHARED;
613 lkflags |= LK_EXCLUSIVE;
615 lkflags |= LK_NODDLKTREAT;
620 needs_exclusive_leaf(struct mount *mp, int flags)
624 * Intermediate nodes can use shared locks, we only need to
625 * force an exclusive lock for leaf nodes.
627 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
630 /* Always use exclusive locks if LOCKSHARED isn't set. */
631 if (!(flags & LOCKSHARED))
635 * For lookups during open(), if the mount point supports
636 * extended shared operations, then use a shared lock for the
637 * leaf node, otherwise use an exclusive lock.
639 if ((flags & ISOPEN) != 0)
640 return (!MNT_EXTENDED_SHARED(mp));
643 * Lookup requests outside of open() that specify LOCKSHARED
644 * only need a shared lock on the leaf vnode.
651 * This is a very central and rather complicated routine.
653 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
654 * The starting directory is taken from ni_startdir. The pathname is
655 * descended until done, or a symbolic link is encountered. The variable
656 * ni_more is clear if the path is completed; it is set to one if a
657 * symbolic link needing interpretation is encountered.
659 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
660 * whether the name is to be looked up, created, renamed, or deleted.
661 * When CREATE, RENAME, or DELETE is specified, information usable in
662 * creating, renaming, or deleting a directory entry may be calculated.
663 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
664 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
665 * returned unlocked. Otherwise the parent directory is not returned. If
666 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
667 * the target is returned locked, otherwise it is returned unlocked.
668 * When creating or renaming and LOCKPARENT is specified, the target may not
669 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
671 * Overall outline of lookup:
674 * identify next component of name at ndp->ni_ptr
675 * handle degenerate case where name is null string
676 * if .. and crossing mount points and on mounted filesys, find parent
677 * call VOP_LOOKUP routine for next component name
678 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
679 * component vnode returned in ni_vp (if it exists), locked.
680 * if result vnode is mounted on and crossing mount points,
681 * find mounted on vnode
682 * if more components of name, do next level at dirloop
683 * return the answer in ni_vp, locked if LOCKLEAF set
684 * if LOCKPARENT set, return locked parent in ni_dvp
685 * if WANTPARENT set, return unlocked parent in ni_dvp
688 lookup(struct nameidata *ndp)
690 char *cp; /* pointer into pathname argument */
691 char *prev_ni_next; /* saved ndp->ni_next */
692 struct vnode *dp = NULL; /* the directory we are searching */
693 struct vnode *tdp; /* saved dp */
694 struct mount *mp; /* mount table entry */
696 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
697 int docache; /* == 0 do not cache last component */
698 int wantparent; /* 1 => wantparent or lockparent flag */
699 int rdonly; /* lookup read-only flag bit */
701 int dpunlocked = 0; /* dp has already been unlocked */
702 int relookup = 0; /* do not consume the path component */
703 struct componentname *cnp = &ndp->ni_cnd;
708 * Setup: break out flag bits into variables.
711 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
712 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
713 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
714 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
715 if (cnp->cn_nameiop == DELETE ||
716 (wantparent && cnp->cn_nameiop != CREATE &&
717 cnp->cn_nameiop != LOOKUP))
719 rdonly = cnp->cn_flags & RDONLY;
720 cnp->cn_flags &= ~ISSYMLINK;
723 * We use shared locks until we hit the parent of the last cn then
724 * we adjust based on the requesting flags.
726 cnp->cn_lkflags = LK_SHARED;
727 dp = ndp->ni_startdir;
728 ndp->ni_startdir = NULLVP;
730 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
735 * Search a new directory.
737 * The last component of the filename is left accessible via
738 * cnp->cn_nameptr for callers that need the name. Callers needing
739 * the name set the SAVENAME flag. When done, they assume
740 * responsibility for freeing the pathname buffer.
742 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
744 cnp->cn_namelen = cp - cnp->cn_nameptr;
745 if (cnp->cn_namelen > NAME_MAX) {
746 error = ENAMETOOLONG;
749 #ifdef NAMEI_DIAGNOSTIC
752 printf("{%s}: ", cnp->cn_nameptr);
755 prev_ni_pathlen = ndp->ni_pathlen;
756 ndp->ni_pathlen -= cnp->cn_namelen;
757 KASSERT(ndp->ni_pathlen <= PATH_MAX,
758 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
759 prev_ni_next = ndp->ni_next;
763 * Replace multiple slashes by a single slash and trailing slashes
764 * by a null. This must be done before VOP_LOOKUP() because some
765 * fs's don't know about trailing slashes. Remember if there were
766 * trailing slashes to handle symlinks, existing non-directories
767 * and non-existing files that won't be directories specially later.
769 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
773 *ndp->ni_next = '\0';
774 cnp->cn_flags |= TRAILINGSLASH;
779 cnp->cn_flags |= MAKEENTRY;
780 if (*cp == '\0' && docache == 0)
781 cnp->cn_flags &= ~MAKEENTRY;
782 if (cnp->cn_namelen == 2 &&
783 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
784 cnp->cn_flags |= ISDOTDOT;
786 cnp->cn_flags &= ~ISDOTDOT;
787 if (*ndp->ni_next == 0)
788 cnp->cn_flags |= ISLASTCN;
790 cnp->cn_flags &= ~ISLASTCN;
792 if ((cnp->cn_flags & ISLASTCN) != 0 &&
793 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
794 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
799 nameicap_tracker_add(ndp, dp);
802 * Check for degenerate name (e.g. / or "")
803 * which is a way of talking about a directory,
804 * e.g. like "/." or ".".
806 if (cnp->cn_nameptr[0] == '\0') {
807 if (dp->v_type != VDIR) {
811 if (cnp->cn_nameiop != LOOKUP) {
821 if (cnp->cn_flags & AUDITVNODE1)
822 AUDIT_ARG_VNODE1(dp);
823 else if (cnp->cn_flags & AUDITVNODE2)
824 AUDIT_ARG_VNODE2(dp);
826 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
828 /* XXX This should probably move to the top of function. */
829 if (cnp->cn_flags & SAVESTART)
830 panic("lookup: SAVESTART");
835 * Handle "..": five special cases.
836 * 0. If doing a capability lookup and lookup_cap_dotdot is
837 * disabled, return ENOTCAPABLE.
838 * 1. Return an error if this is the last component of
839 * the name and the operation is DELETE or RENAME.
840 * 2. If at root directory (e.g. after chroot)
841 * or at absolute root directory
842 * then ignore it so can't get out.
843 * 3. If this vnode is the root of a mounted
844 * filesystem, then replace it with the
845 * vnode which was mounted on so we take the
846 * .. in the other filesystem.
847 * 4. If the vnode is the top directory of
848 * the jail or chroot, don't let them out.
849 * 5. If doing a capability lookup and lookup_cap_dotdot is
850 * enabled, return ENOTCAPABLE if the lookup would escape
851 * from the initial file descriptor directory. Checks are
852 * done by ensuring that namei() already traversed the
853 * result of dotdot lookup.
855 if (cnp->cn_flags & ISDOTDOT) {
856 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT))
857 == NI_LCF_STRICTRELATIVE) {
859 if (KTRPOINT(curthread, KTR_CAPFAIL))
860 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
865 if ((cnp->cn_flags & ISLASTCN) != 0 &&
866 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
871 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
873 if (dp == pr->pr_root)
875 if (dp == ndp->ni_rootdir ||
876 dp == ndp->ni_topdir ||
879 ((dp->v_vflag & VV_ROOT) != 0 &&
880 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
886 if ((dp->v_vflag & VV_ROOT) == 0)
888 if (VN_IS_DOOMED(dp)) { /* forced unmount */
893 dp = dp->v_mount->mnt_vnodecovered;
897 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
898 LK_RETRY, ISDOTDOT));
899 error = nameicap_check_dotdot(ndp, dp);
902 if (KTRPOINT(curthread, KTR_CAPFAIL))
903 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
911 * We now have a segment name to search for, and a directory to search.
915 if ((cnp->cn_flags & NOMACCHECK) == 0) {
916 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
924 ASSERT_VOP_LOCKED(dp, "lookup");
926 * If we have a shared lock we may need to upgrade the lock for the
929 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
930 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
931 vn_lock(dp, LK_UPGRADE|LK_RETRY);
932 if (VN_IS_DOOMED(dp)) {
937 * If we're looking up the last component and we need an exclusive
938 * lock, adjust our lkflags.
940 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
941 cnp->cn_lkflags = LK_EXCLUSIVE;
942 #ifdef NAMEI_DIAGNOSTIC
943 vn_printf(dp, "lookup in ");
945 lkflags_save = cnp->cn_lkflags;
946 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
948 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
949 cnp->cn_lkflags = lkflags_save;
951 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
952 #ifdef NAMEI_DIAGNOSTIC
953 printf("not found\n");
955 if ((error == ENOENT) &&
956 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
957 (dp->v_mount->mnt_flag & MNT_UNION)) {
959 dp = dp->v_mount->mnt_vnodecovered;
963 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
964 LK_RETRY, cnp->cn_flags));
965 nameicap_tracker_add(ndp, dp);
969 if (error == ERELOOKUP) {
977 if (error != EJUSTRETURN)
980 * At this point, we know we're at the end of the
981 * pathname. If creating / renaming, we can consider
982 * allowing the file or directory to be created / renamed,
983 * provided we're not on a read-only filesystem.
989 /* trailing slash only allowed for directories */
990 if ((cnp->cn_flags & TRAILINGSLASH) &&
991 !(cnp->cn_flags & WILLBEDIR)) {
995 if ((cnp->cn_flags & LOCKPARENT) == 0)
998 * We return with ni_vp NULL to indicate that the entry
999 * doesn't currently exist, leaving a pointer to the
1000 * (possibly locked) directory vnode in ndp->ni_dvp.
1002 if (cnp->cn_flags & SAVESTART) {
1003 ndp->ni_startdir = ndp->ni_dvp;
1004 VREF(ndp->ni_startdir);
1010 #ifdef NAMEI_DIAGNOSTIC
1016 * Check to see if the vnode has been mounted on;
1017 * if so find the root of the mounted filesystem.
1019 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
1020 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
1021 if (vfs_busy(mp, 0))
1024 if (dp != ndp->ni_dvp)
1028 vrefact(vp_crossmp);
1029 ndp->ni_dvp = vp_crossmp;
1030 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
1031 cnp->cn_flags), &tdp);
1033 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
1034 panic("vp_crossmp exclusively locked or reclaimed");
1039 ndp->ni_vp = dp = tdp;
1043 * Check for symbolic link
1045 if ((dp->v_type == VLNK) &&
1046 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
1047 *ndp->ni_next == '/')) {
1048 cnp->cn_flags |= ISSYMLINK;
1049 if (VN_IS_DOOMED(dp)) {
1051 * We can't know whether the directory was mounted with
1052 * NOSYMFOLLOW, so we can't follow safely.
1057 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1062 * Symlink code always expects an unlocked dvp.
1064 if (ndp->ni_dvp != ndp->ni_vp) {
1065 VOP_UNLOCK(ndp->ni_dvp);
1066 ni_dvp_unlocked = 1;
1073 * Not a symbolic link that we will follow. Continue with the
1074 * next component if there is any; otherwise, we're done.
1076 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1077 ("lookup: invalid path state."));
1080 ndp->ni_pathlen = prev_ni_pathlen;
1081 ndp->ni_next = prev_ni_next;
1082 if (ndp->ni_dvp != dp)
1088 if (cnp->cn_flags & ISDOTDOT) {
1089 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1092 if (KTRPOINT(curthread, KTR_CAPFAIL))
1093 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1098 if (*ndp->ni_next == '/') {
1099 cnp->cn_nameptr = ndp->ni_next;
1100 while (*cnp->cn_nameptr == '/') {
1104 if (ndp->ni_dvp != dp)
1111 * If we're processing a path with a trailing slash,
1112 * check that the end result is a directory.
1114 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1119 * Disallow directory write attempts on read-only filesystems.
1122 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1126 if (cnp->cn_flags & SAVESTART) {
1127 ndp->ni_startdir = ndp->ni_dvp;
1128 VREF(ndp->ni_startdir);
1131 ni_dvp_unlocked = 2;
1132 if (ndp->ni_dvp != dp)
1136 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1137 VOP_UNLOCK(ndp->ni_dvp);
1138 ni_dvp_unlocked = 1;
1141 if (cnp->cn_flags & AUDITVNODE1)
1142 AUDIT_ARG_VNODE1(dp);
1143 else if (cnp->cn_flags & AUDITVNODE2)
1144 AUDIT_ARG_VNODE2(dp);
1146 if ((cnp->cn_flags & LOCKLEAF) == 0)
1150 * Because of shared lookup we may have the vnode shared locked, but
1151 * the caller may want it to be exclusively locked.
1153 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1154 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1155 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1156 if (VN_IS_DOOMED(dp)) {
1164 if (ni_dvp_unlocked != 2) {
1165 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1178 * relookup - lookup a path name component
1179 * Used by lookup to re-acquire things.
1182 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
1184 struct vnode *dp = NULL; /* the directory we are searching */
1185 int wantparent; /* 1 => wantparent or lockparent flag */
1186 int rdonly; /* lookup read-only flag bit */
1189 KASSERT(cnp->cn_flags & ISLASTCN,
1190 ("relookup: Not given last component."));
1192 * Setup: break out flag bits into variables.
1194 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
1195 KASSERT(wantparent, ("relookup: parent not wanted."));
1196 rdonly = cnp->cn_flags & RDONLY;
1197 cnp->cn_flags &= ~ISSYMLINK;
1199 cnp->cn_lkflags = LK_EXCLUSIVE;
1200 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1203 * Search a new directory.
1205 * The last component of the filename is left accessible via
1206 * cnp->cn_nameptr for callers that need the name. Callers needing
1207 * the name set the SAVENAME flag. When done, they assume
1208 * responsibility for freeing the pathname buffer.
1210 #ifdef NAMEI_DIAGNOSTIC
1211 printf("{%s}: ", cnp->cn_nameptr);
1215 * Check for "" which represents the root directory after slash
1218 if (cnp->cn_nameptr[0] == '\0') {
1220 * Support only LOOKUP for "/" because lookup()
1221 * can't succeed for CREATE, DELETE and RENAME.
1223 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1224 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1226 if (!(cnp->cn_flags & LOCKLEAF))
1229 /* XXX This should probably move to the top of function. */
1230 if (cnp->cn_flags & SAVESTART)
1231 panic("lookup: SAVESTART");
1235 if (cnp->cn_flags & ISDOTDOT)
1236 panic ("relookup: lookup on dot-dot");
1239 * We now have a segment name to search for, and a directory to search.
1241 #ifdef NAMEI_DIAGNOSTIC
1242 vn_printf(dp, "search in ");
1244 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1245 KASSERT(*vpp == NULL, ("leaf should be empty"));
1246 if (error != EJUSTRETURN)
1249 * If creating and at end of pathname, then can consider
1250 * allowing file to be created.
1256 /* ASSERT(dvp == ndp->ni_startdir) */
1257 if (cnp->cn_flags & SAVESTART)
1259 if ((cnp->cn_flags & LOCKPARENT) == 0)
1262 * We return with ni_vp NULL to indicate that the entry
1263 * doesn't currently exist, leaving a pointer to the
1264 * (possibly locked) directory vnode in ndp->ni_dvp.
1272 * Disallow directory write attempts on read-only filesystems.
1275 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1284 * Set the parent lock/ref state to the requested state.
1286 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1291 } else if (!wantparent)
1294 * Check for symbolic link
1296 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1297 ("relookup: symlink found.\n"));
1299 /* ASSERT(dvp == ndp->ni_startdir) */
1300 if (cnp->cn_flags & SAVESTART)
1303 if ((cnp->cn_flags & LOCKLEAF) == 0)
1313 NDINIT_ALL(struct nameidata *ndp, u_long op, u_long flags, enum uio_seg segflg,
1314 const char *namep, int dirfd, struct vnode *startdir, cap_rights_t *rightsp,
1318 ndp->ni_cnd.cn_nameiop = op;
1319 ndp->ni_cnd.cn_flags = flags;
1320 ndp->ni_segflg = segflg;
1321 ndp->ni_dirp = namep;
1322 ndp->ni_dirfd = dirfd;
1323 ndp->ni_startdir = startdir;
1324 ndp->ni_resflags = 0;
1325 filecaps_init(&ndp->ni_filecaps);
1326 ndp->ni_cnd.cn_thread = td;
1327 if (rightsp != NULL)
1328 ndp->ni_rightsneeded = *rightsp;
1330 cap_rights_init(&ndp->ni_rightsneeded);
1334 * Free data allocated by namei(); see namei(9) for details.
1337 NDFREE(struct nameidata *ndp, const u_int flags)
1345 if (!(flags & NDF_NO_FREE_PNBUF) &&
1346 (ndp->ni_cnd.cn_flags & HASBUF)) {
1347 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1348 ndp->ni_cnd.cn_flags &= ~HASBUF;
1350 if (!(flags & NDF_NO_VP_UNLOCK) &&
1351 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1353 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1354 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1355 ndp->ni_dvp != ndp->ni_vp)
1357 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1366 VOP_UNLOCK(ndp->ni_vp);
1367 if (!(flags & NDF_NO_DVP_RELE) &&
1368 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1377 VOP_UNLOCK(ndp->ni_dvp);
1378 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1379 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1380 vrele(ndp->ni_startdir);
1381 ndp->ni_startdir = NULL;
1386 * Determine if there is a suitable alternate filename under the specified
1387 * prefix for the specified path. If the create flag is set, then the
1388 * alternate prefix will be used so long as the parent directory exists.
1389 * This is used by the various compatibility ABIs so that Linux binaries prefer
1390 * files under /compat/linux for example. The chosen path (whether under
1391 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1392 * to by pathbuf. The caller is responsible for free'ing the buffer from
1393 * the M_TEMP bucket if one is returned.
1396 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1397 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1399 struct nameidata nd, ndroot;
1400 char *ptr, *buf, *cp;
1404 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1407 /* Copy the prefix into the new pathname as a starting point. */
1408 len = strlcpy(buf, prefix, MAXPATHLEN);
1409 if (len >= MAXPATHLEN) {
1414 sz = MAXPATHLEN - len;
1417 /* Append the filename to the prefix. */
1418 if (pathseg == UIO_SYSSPACE)
1419 error = copystr(path, ptr, sz, &len);
1421 error = copyinstr(path, ptr, sz, &len);
1429 /* Only use a prefix with absolute pathnames. */
1435 if (dirfd != AT_FDCWD) {
1437 * We want the original because the "prefix" is
1438 * included in the already opened dirfd.
1440 bcopy(ptr, buf, len);
1445 * We know that there is a / somewhere in this pathname.
1446 * Search backwards for it, to find the file's parent dir
1447 * to see if it exists in the alternate tree. If it does,
1448 * and we want to create a file (cflag is set). We don't
1449 * need to worry about the root comparison in this case.
1453 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1456 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1462 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1469 * We now compare the vnode of the prefix to the one
1470 * vnode asked. If they resolve to be the same, then we
1471 * ignore the match so that the real root gets used.
1472 * This avoids the problem of traversing "../.." to find the
1473 * root directory and never finding it, because "/" resolves
1474 * to the emulation root directory. This is expensive :-(
1476 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1479 /* We shouldn't ever get an error from this namei(). */
1480 error = namei(&ndroot);
1482 if (nd.ni_vp == ndroot.ni_vp)
1485 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1486 vrele(ndroot.ni_vp);
1490 NDFREE(&nd, NDF_ONLY_PNBUF);
1494 /* If there was an error, use the original path name. */
1496 bcopy(ptr, buf, len);