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;
128 if ((flags & LK_INTERLOCK) != 0)
130 WITNESS_UNLOCK(&lk->lock_object, 0, LOCK_FILE, LOCK_LINE);
131 LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, LOCK_FILE,
136 static struct vop_vector crossmp_vnodeops = {
137 .vop_default = &default_vnodeops,
138 .vop_islocked = crossmp_vop_islocked,
139 .vop_lock1 = crossmp_vop_lock1,
140 .vop_unlock = crossmp_vop_unlock,
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 getnewvnode("crossmp", NULL, &crossmp_vnodeops, &vp_crossmp);
161 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nameiinit, NULL);
163 static int lookup_cap_dotdot = 1;
164 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot, CTLFLAG_RWTUN,
165 &lookup_cap_dotdot, 0,
166 "enables \"..\" components in path lookup in capability mode");
167 static int lookup_cap_dotdot_nonlocal = 1;
168 SYSCTL_INT(_vfs, OID_AUTO, lookup_cap_dotdot_nonlocal, CTLFLAG_RWTUN,
169 &lookup_cap_dotdot_nonlocal, 0,
170 "enables \"..\" components in path lookup in capability mode "
171 "on non-local mount");
174 nameicap_tracker_add(struct nameidata *ndp, struct vnode *dp)
176 struct nameicap_tracker *nt;
178 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp->v_type != VDIR)
180 if ((ndp->ni_lcf & (NI_LCF_BENEATH_ABS | NI_LCF_BENEATH_LATCHED)) ==
181 NI_LCF_BENEATH_ABS) {
182 MPASS((ndp->ni_lcf & NI_LCF_LATCH) != 0);
183 if (dp != ndp->ni_beneath_latch)
185 ndp->ni_lcf |= NI_LCF_BENEATH_LATCHED;
187 nt = uma_zalloc(nt_zone, M_WAITOK);
190 TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link);
194 nameicap_cleanup(struct nameidata *ndp, bool clean_latch)
196 struct nameicap_tracker *nt, *nt1;
198 KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) ||
199 (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative"));
200 TAILQ_FOREACH_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) {
201 TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link);
203 uma_zfree(nt_zone, nt);
205 if (clean_latch && (ndp->ni_lcf & NI_LCF_LATCH) != 0) {
206 ndp->ni_lcf &= ~NI_LCF_LATCH;
207 vrele(ndp->ni_beneath_latch);
212 * For dotdot lookups in capability mode, only allow the component
213 * lookup to succeed if the resulting directory was already traversed
214 * during the operation. Also fail dotdot lookups for non-local
215 * filesystems, where external agents might assist local lookups to
216 * escape the compartment.
219 nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp)
221 struct nameicap_tracker *nt;
224 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp == NULL ||
228 if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL &&
229 (mp->mnt_flag & MNT_LOCAL) == 0)
230 return (ENOTCAPABLE);
231 TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head,
236 if ((ndp->ni_lcf & NI_LCF_BENEATH_ABS) != 0) {
237 ndp->ni_lcf &= ~NI_LCF_BENEATH_LATCHED;
238 nameicap_cleanup(ndp, false);
241 return (ENOTCAPABLE);
245 namei_cleanup_cnp(struct componentname *cnp)
248 uma_zfree(namei_zone, cnp->cn_pnbuf);
250 cnp->cn_pnbuf = NULL;
251 cnp->cn_nameptr = NULL;
256 namei_handle_root(struct nameidata *ndp, struct vnode **dpp)
258 struct componentname *cnp;
261 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0) {
263 if (KTRPOINT(curthread, KTR_CAPFAIL))
264 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
266 return (ENOTCAPABLE);
268 if ((cnp->cn_flags & BENEATH) != 0) {
269 ndp->ni_lcf |= NI_LCF_BENEATH_ABS;
270 ndp->ni_lcf &= ~NI_LCF_BENEATH_LATCHED;
271 nameicap_cleanup(ndp, false);
273 while (*(cnp->cn_nameptr) == '/') {
277 *dpp = ndp->ni_rootdir;
283 * Convert a pathname into a pointer to a locked vnode.
285 * The FOLLOW flag is set when symbolic links are to be followed
286 * when they occur at the end of the name translation process.
287 * Symbolic links are always followed for all other pathname
288 * components other than the last.
290 * The segflg defines whether the name is to be copied from user
291 * space or kernel space.
293 * Overall outline of namei:
296 * get starting directory
297 * while (!done && !error) {
298 * call lookup to search path.
299 * if symbolic link, massage name in buffer and continue
303 namei(struct nameidata *ndp)
305 struct filedesc *fdp; /* pointer to file descriptor state */
306 char *cp; /* pointer into pathname argument */
307 struct vnode *dp; /* the directory we are searching */
308 struct iovec aiov; /* uio for reading symbolic links */
309 struct componentname *cnp;
313 struct filecaps dirfd_caps;
315 int error, linklen, startdir_used;
320 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
321 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
322 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
323 ("namei: nameiop contaminated with flags"));
324 KASSERT((cnp->cn_flags & OPMASK) == 0,
325 ("namei: flags contaminated with nameiops"));
326 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
327 ndp->ni_startdir->v_type == VBAD);
329 TAILQ_INIT(&ndp->ni_cap_tracker);
332 /* We will set this ourselves if we need it. */
333 cnp->cn_flags &= ~TRAILINGSLASH;
336 * Get a buffer for the name to be translated, and copy the
337 * name into the buffer.
339 if ((cnp->cn_flags & HASBUF) == 0)
340 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
341 if (ndp->ni_segflg == UIO_SYSSPACE)
342 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
345 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
349 * Don't allow empty pathnames.
351 if (error == 0 && *cnp->cn_pnbuf == '\0')
354 #ifdef CAPABILITY_MODE
356 * In capability mode, lookups must be restricted to happen in
357 * the subtree with the root specified by the file descriptor:
358 * - The root must be real file descriptor, not the pseudo-descriptor
360 * - The passed path must be relative and not absolute.
361 * - If lookup_cap_dotdot is disabled, path must not contain the
363 * - If lookup_cap_dotdot is enabled, we verify that all '..'
364 * components lookups result in the directories which were
365 * previously walked by us, which prevents an escape from
368 if (error == 0 && IN_CAPABILITY_MODE(td) &&
369 (cnp->cn_flags & NOCAPCHECK) == 0) {
370 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
371 if (ndp->ni_dirfd == AT_FDCWD) {
373 if (KTRPOINT(td, KTR_CAPFAIL))
374 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
381 namei_cleanup_cnp(cnp);
387 if (KTRPOINT(td, KTR_NAMEI)) {
388 KASSERT(cnp->cn_thread == curthread,
389 ("namei not using curthread"));
390 ktrnamei(cnp->cn_pnbuf);
394 * Get starting point for the translation.
397 ndp->ni_rootdir = fdp->fd_rdir;
398 vrefact(ndp->ni_rootdir);
399 ndp->ni_topdir = fdp->fd_jdir;
402 * If we are auditing the kernel pathname, save the user pathname.
404 if (cnp->cn_flags & AUDITVNODE1)
405 AUDIT_ARG_UPATH1(td, ndp->ni_dirfd, cnp->cn_pnbuf);
406 if (cnp->cn_flags & AUDITVNODE2)
407 AUDIT_ARG_UPATH2(td, ndp->ni_dirfd, cnp->cn_pnbuf);
411 cnp->cn_nameptr = cnp->cn_pnbuf;
412 if (cnp->cn_pnbuf[0] == '/') {
413 ndp->ni_resflags |= NIRES_ABS;
414 error = namei_handle_root(ndp, &dp);
416 if (ndp->ni_startdir != NULL) {
417 dp = ndp->ni_startdir;
419 } else if (ndp->ni_dirfd == AT_FDCWD) {
423 rights = ndp->ni_rightsneeded;
424 cap_rights_set(&rights, CAP_LOOKUP);
426 if (cnp->cn_flags & AUDITVNODE1)
427 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
428 if (cnp->cn_flags & AUDITVNODE2)
429 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
430 error = fgetvp_rights(td, ndp->ni_dirfd,
431 &rights, &ndp->ni_filecaps, &dp);
436 * If file descriptor doesn't have all rights,
437 * all lookups relative to it must also be
441 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights,
443 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
444 ndp->ni_filecaps.fc_nioctls != -1) {
445 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
449 if (error == 0 && dp->v_type != VDIR)
452 if (error == 0 && (cnp->cn_flags & BENEATH) != 0) {
453 if (ndp->ni_dirfd == AT_FDCWD) {
454 ndp->ni_beneath_latch = fdp->fd_cdir;
455 vrefact(ndp->ni_beneath_latch);
457 rights = ndp->ni_rightsneeded;
458 cap_rights_set(&rights, CAP_LOOKUP);
459 error = fgetvp_rights(td, ndp->ni_dirfd, &rights,
460 &dirfd_caps, &ndp->ni_beneath_latch);
461 if (error == 0 && dp->v_type != VDIR) {
462 vrele(ndp->ni_beneath_latch);
467 ndp->ni_lcf |= NI_LCF_LATCH;
469 FILEDESC_SUNLOCK(fdp);
470 if (ndp->ni_startdir != NULL && !startdir_used)
471 vrele(ndp->ni_startdir);
477 MPASS((ndp->ni_lcf & (NI_LCF_BENEATH_ABS | NI_LCF_LATCH)) !=
479 if (((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 &&
480 lookup_cap_dotdot != 0) ||
481 ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) == 0 &&
482 (cnp->cn_flags & BENEATH) != 0))
483 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
484 SDT_PROBE3(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
487 ndp->ni_startdir = dp;
492 * If not a symbolic link, we're done.
494 if ((cnp->cn_flags & ISSYMLINK) == 0) {
495 vrele(ndp->ni_rootdir);
496 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
497 namei_cleanup_cnp(cnp);
499 cnp->cn_flags |= HASBUF;
500 if ((ndp->ni_lcf & (NI_LCF_BENEATH_ABS |
501 NI_LCF_BENEATH_LATCHED)) == NI_LCF_BENEATH_ABS) {
505 nameicap_cleanup(ndp, true);
506 SDT_PROBE2(vfs, namei, lookup, return, error,
507 (error == 0 ? ndp->ni_vp : NULL));
510 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
515 if ((cnp->cn_flags & NOMACCHECK) == 0) {
516 error = mac_vnode_check_readlink(td->td_ucred,
522 if (ndp->ni_pathlen > 1)
523 cp = uma_zalloc(namei_zone, M_WAITOK);
527 aiov.iov_len = MAXPATHLEN;
528 auio.uio_iov = &aiov;
531 auio.uio_rw = UIO_READ;
532 auio.uio_segflg = UIO_SYSSPACE;
534 auio.uio_resid = MAXPATHLEN;
535 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
537 if (ndp->ni_pathlen > 1)
538 uma_zfree(namei_zone, cp);
541 linklen = MAXPATHLEN - auio.uio_resid;
543 if (ndp->ni_pathlen > 1)
544 uma_zfree(namei_zone, cp);
548 if (linklen + ndp->ni_pathlen > MAXPATHLEN) {
549 if (ndp->ni_pathlen > 1)
550 uma_zfree(namei_zone, cp);
551 error = ENAMETOOLONG;
554 if (ndp->ni_pathlen > 1) {
555 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
556 uma_zfree(namei_zone, cnp->cn_pnbuf);
559 cnp->cn_pnbuf[linklen] = '\0';
560 ndp->ni_pathlen += linklen;
564 * Check if root directory should replace current directory.
566 cnp->cn_nameptr = cnp->cn_pnbuf;
567 if (*(cnp->cn_nameptr) == '/') {
569 error = namei_handle_root(ndp, &dp);
578 vrele(ndp->ni_rootdir);
580 namei_cleanup_cnp(cnp);
581 nameicap_cleanup(ndp, true);
582 SDT_PROBE2(vfs, namei, lookup, return, error, NULL);
587 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
590 if (mp == NULL || ((lkflags & LK_SHARED) &&
591 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
592 ((cnflags & ISDOTDOT) &&
593 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
594 lkflags &= ~LK_SHARED;
595 lkflags |= LK_EXCLUSIVE;
597 lkflags |= LK_NODDLKTREAT;
602 needs_exclusive_leaf(struct mount *mp, int flags)
606 * Intermediate nodes can use shared locks, we only need to
607 * force an exclusive lock for leaf nodes.
609 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
612 /* Always use exclusive locks if LOCKSHARED isn't set. */
613 if (!(flags & LOCKSHARED))
617 * For lookups during open(), if the mount point supports
618 * extended shared operations, then use a shared lock for the
619 * leaf node, otherwise use an exclusive lock.
621 if ((flags & ISOPEN) != 0)
622 return (!MNT_EXTENDED_SHARED(mp));
625 * Lookup requests outside of open() that specify LOCKSHARED
626 * only need a shared lock on the leaf vnode.
633 * This is a very central and rather complicated routine.
635 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
636 * The starting directory is taken from ni_startdir. The pathname is
637 * descended until done, or a symbolic link is encountered. The variable
638 * ni_more is clear if the path is completed; it is set to one if a
639 * symbolic link needing interpretation is encountered.
641 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
642 * whether the name is to be looked up, created, renamed, or deleted.
643 * When CREATE, RENAME, or DELETE is specified, information usable in
644 * creating, renaming, or deleting a directory entry may be calculated.
645 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
646 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
647 * returned unlocked. Otherwise the parent directory is not returned. If
648 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
649 * the target is returned locked, otherwise it is returned unlocked.
650 * When creating or renaming and LOCKPARENT is specified, the target may not
651 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
653 * Overall outline of lookup:
656 * identify next component of name at ndp->ni_ptr
657 * handle degenerate case where name is null string
658 * if .. and crossing mount points and on mounted filesys, find parent
659 * call VOP_LOOKUP routine for next component name
660 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
661 * component vnode returned in ni_vp (if it exists), locked.
662 * if result vnode is mounted on and crossing mount points,
663 * find mounted on vnode
664 * if more components of name, do next level at dirloop
665 * return the answer in ni_vp, locked if LOCKLEAF set
666 * if LOCKPARENT set, return locked parent in ni_dvp
667 * if WANTPARENT set, return unlocked parent in ni_dvp
670 lookup(struct nameidata *ndp)
672 char *cp; /* pointer into pathname argument */
673 char *prev_ni_next; /* saved ndp->ni_next */
674 struct vnode *dp = NULL; /* the directory we are searching */
675 struct vnode *tdp; /* saved dp */
676 struct mount *mp; /* mount table entry */
678 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
679 int docache; /* == 0 do not cache last component */
680 int wantparent; /* 1 => wantparent or lockparent flag */
681 int rdonly; /* lookup read-only flag bit */
683 int dpunlocked = 0; /* dp has already been unlocked */
684 int relookup = 0; /* do not consume the path component */
685 struct componentname *cnp = &ndp->ni_cnd;
690 * Setup: break out flag bits into variables.
693 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
694 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
695 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
696 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
697 if (cnp->cn_nameiop == DELETE ||
698 (wantparent && cnp->cn_nameiop != CREATE &&
699 cnp->cn_nameiop != LOOKUP))
701 rdonly = cnp->cn_flags & RDONLY;
702 cnp->cn_flags &= ~ISSYMLINK;
705 * We use shared locks until we hit the parent of the last cn then
706 * we adjust based on the requesting flags.
708 cnp->cn_lkflags = LK_SHARED;
709 dp = ndp->ni_startdir;
710 ndp->ni_startdir = NULLVP;
712 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
717 * Search a new directory.
719 * The last component of the filename is left accessible via
720 * cnp->cn_nameptr for callers that need the name. Callers needing
721 * the name set the SAVENAME flag. When done, they assume
722 * responsibility for freeing the pathname buffer.
724 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
726 cnp->cn_namelen = cp - cnp->cn_nameptr;
727 if (cnp->cn_namelen > NAME_MAX) {
728 error = ENAMETOOLONG;
731 #ifdef NAMEI_DIAGNOSTIC
734 printf("{%s}: ", cnp->cn_nameptr);
737 prev_ni_pathlen = ndp->ni_pathlen;
738 ndp->ni_pathlen -= cnp->cn_namelen;
739 KASSERT(ndp->ni_pathlen <= PATH_MAX,
740 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
741 prev_ni_next = ndp->ni_next;
745 * Replace multiple slashes by a single slash and trailing slashes
746 * by a null. This must be done before VOP_LOOKUP() because some
747 * fs's don't know about trailing slashes. Remember if there were
748 * trailing slashes to handle symlinks, existing non-directories
749 * and non-existing files that won't be directories specially later.
751 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
755 *ndp->ni_next = '\0';
756 cnp->cn_flags |= TRAILINGSLASH;
761 cnp->cn_flags |= MAKEENTRY;
762 if (*cp == '\0' && docache == 0)
763 cnp->cn_flags &= ~MAKEENTRY;
764 if (cnp->cn_namelen == 2 &&
765 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
766 cnp->cn_flags |= ISDOTDOT;
768 cnp->cn_flags &= ~ISDOTDOT;
769 if (*ndp->ni_next == 0)
770 cnp->cn_flags |= ISLASTCN;
772 cnp->cn_flags &= ~ISLASTCN;
774 if ((cnp->cn_flags & ISLASTCN) != 0 &&
775 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
776 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
781 nameicap_tracker_add(ndp, dp);
784 * Check for degenerate name (e.g. / or "")
785 * which is a way of talking about a directory,
786 * e.g. like "/." or ".".
788 if (cnp->cn_nameptr[0] == '\0') {
789 if (dp->v_type != VDIR) {
793 if (cnp->cn_nameiop != LOOKUP) {
803 if (cnp->cn_flags & AUDITVNODE1)
804 AUDIT_ARG_VNODE1(dp);
805 else if (cnp->cn_flags & AUDITVNODE2)
806 AUDIT_ARG_VNODE2(dp);
808 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
810 /* XXX This should probably move to the top of function. */
811 if (cnp->cn_flags & SAVESTART)
812 panic("lookup: SAVESTART");
817 * Handle "..": five special cases.
818 * 0. If doing a capability lookup and lookup_cap_dotdot is
819 * disabled, return ENOTCAPABLE.
820 * 1. Return an error if this is the last component of
821 * the name and the operation is DELETE or RENAME.
822 * 2. If at root directory (e.g. after chroot)
823 * or at absolute root directory
824 * then ignore it so can't get out.
825 * 3. If this vnode is the root of a mounted
826 * filesystem, then replace it with the
827 * vnode which was mounted on so we take the
828 * .. in the other filesystem.
829 * 4. If the vnode is the top directory of
830 * the jail or chroot, don't let them out.
831 * 5. If doing a capability lookup and lookup_cap_dotdot is
832 * enabled, return ENOTCAPABLE if the lookup would escape
833 * from the initial file descriptor directory. Checks are
834 * done by ensuring that namei() already traversed the
835 * result of dotdot lookup.
837 if (cnp->cn_flags & ISDOTDOT) {
838 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT))
839 == NI_LCF_STRICTRELATIVE) {
841 if (KTRPOINT(curthread, KTR_CAPFAIL))
842 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
847 if ((cnp->cn_flags & ISLASTCN) != 0 &&
848 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
853 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
855 if (dp == pr->pr_root)
857 if (dp == ndp->ni_rootdir ||
858 dp == ndp->ni_topdir ||
861 ((dp->v_vflag & VV_ROOT) != 0 &&
862 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
868 if ((dp->v_vflag & VV_ROOT) == 0)
870 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
875 dp = dp->v_mount->mnt_vnodecovered;
879 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
880 LK_RETRY, ISDOTDOT));
881 error = nameicap_check_dotdot(ndp, dp);
884 if (KTRPOINT(curthread, KTR_CAPFAIL))
885 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
893 * We now have a segment name to search for, and a directory to search.
897 if ((cnp->cn_flags & NOMACCHECK) == 0) {
898 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
906 ASSERT_VOP_LOCKED(dp, "lookup");
908 * If we have a shared lock we may need to upgrade the lock for the
911 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
912 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
913 vn_lock(dp, LK_UPGRADE|LK_RETRY);
914 if ((dp->v_iflag & VI_DOOMED) != 0) {
919 * If we're looking up the last component and we need an exclusive
920 * lock, adjust our lkflags.
922 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
923 cnp->cn_lkflags = LK_EXCLUSIVE;
924 #ifdef NAMEI_DIAGNOSTIC
925 vn_printf(dp, "lookup in ");
927 lkflags_save = cnp->cn_lkflags;
928 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
930 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
931 cnp->cn_lkflags = lkflags_save;
933 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
934 #ifdef NAMEI_DIAGNOSTIC
935 printf("not found\n");
937 if ((error == ENOENT) &&
938 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
939 (dp->v_mount->mnt_flag & MNT_UNION)) {
941 dp = dp->v_mount->mnt_vnodecovered;
945 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
946 LK_RETRY, cnp->cn_flags));
947 nameicap_tracker_add(ndp, dp);
951 if (error == ERELOOKUP) {
959 if (error != EJUSTRETURN)
962 * At this point, we know we're at the end of the
963 * pathname. If creating / renaming, we can consider
964 * allowing the file or directory to be created / renamed,
965 * provided we're not on a read-only filesystem.
971 /* trailing slash only allowed for directories */
972 if ((cnp->cn_flags & TRAILINGSLASH) &&
973 !(cnp->cn_flags & WILLBEDIR)) {
977 if ((cnp->cn_flags & LOCKPARENT) == 0)
980 * We return with ni_vp NULL to indicate that the entry
981 * doesn't currently exist, leaving a pointer to the
982 * (possibly locked) directory vnode in ndp->ni_dvp.
984 if (cnp->cn_flags & SAVESTART) {
985 ndp->ni_startdir = ndp->ni_dvp;
986 VREF(ndp->ni_startdir);
992 #ifdef NAMEI_DIAGNOSTIC
998 * Check to see if the vnode has been mounted on;
999 * if so find the root of the mounted filesystem.
1001 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
1002 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
1003 if (vfs_busy(mp, 0))
1006 if (dp != ndp->ni_dvp)
1010 vrefact(vp_crossmp);
1011 ndp->ni_dvp = vp_crossmp;
1012 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
1013 cnp->cn_flags), &tdp);
1015 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
1016 panic("vp_crossmp exclusively locked or reclaimed");
1021 ndp->ni_vp = dp = tdp;
1025 * Check for symbolic link
1027 if ((dp->v_type == VLNK) &&
1028 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
1029 *ndp->ni_next == '/')) {
1030 cnp->cn_flags |= ISSYMLINK;
1031 if (dp->v_iflag & VI_DOOMED) {
1033 * We can't know whether the directory was mounted with
1034 * NOSYMFOLLOW, so we can't follow safely.
1039 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1044 * Symlink code always expects an unlocked dvp.
1046 if (ndp->ni_dvp != ndp->ni_vp) {
1047 VOP_UNLOCK(ndp->ni_dvp, 0);
1048 ni_dvp_unlocked = 1;
1055 * Not a symbolic link that we will follow. Continue with the
1056 * next component if there is any; otherwise, we're done.
1058 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1059 ("lookup: invalid path state."));
1062 ndp->ni_pathlen = prev_ni_pathlen;
1063 ndp->ni_next = prev_ni_next;
1064 if (ndp->ni_dvp != dp)
1070 if (cnp->cn_flags & ISDOTDOT) {
1071 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1074 if (KTRPOINT(curthread, KTR_CAPFAIL))
1075 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1080 if (*ndp->ni_next == '/') {
1081 cnp->cn_nameptr = ndp->ni_next;
1082 while (*cnp->cn_nameptr == '/') {
1086 if (ndp->ni_dvp != dp)
1093 * If we're processing a path with a trailing slash,
1094 * check that the end result is a directory.
1096 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1101 * Disallow directory write attempts on read-only filesystems.
1104 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1108 if (cnp->cn_flags & SAVESTART) {
1109 ndp->ni_startdir = ndp->ni_dvp;
1110 VREF(ndp->ni_startdir);
1113 ni_dvp_unlocked = 2;
1114 if (ndp->ni_dvp != dp)
1118 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1119 VOP_UNLOCK(ndp->ni_dvp, 0);
1120 ni_dvp_unlocked = 1;
1123 if (cnp->cn_flags & AUDITVNODE1)
1124 AUDIT_ARG_VNODE1(dp);
1125 else if (cnp->cn_flags & AUDITVNODE2)
1126 AUDIT_ARG_VNODE2(dp);
1128 if ((cnp->cn_flags & LOCKLEAF) == 0)
1132 * Because of shared lookup we may have the vnode shared locked, but
1133 * the caller may want it to be exclusively locked.
1135 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1136 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1137 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1138 if (dp->v_iflag & VI_DOOMED) {
1146 if (ni_dvp_unlocked != 2) {
1147 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1160 * relookup - lookup a path name component
1161 * Used by lookup to re-acquire things.
1164 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
1166 struct vnode *dp = NULL; /* the directory we are searching */
1167 int wantparent; /* 1 => wantparent or lockparent flag */
1168 int rdonly; /* lookup read-only flag bit */
1171 KASSERT(cnp->cn_flags & ISLASTCN,
1172 ("relookup: Not given last component."));
1174 * Setup: break out flag bits into variables.
1176 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
1177 KASSERT(wantparent, ("relookup: parent not wanted."));
1178 rdonly = cnp->cn_flags & RDONLY;
1179 cnp->cn_flags &= ~ISSYMLINK;
1181 cnp->cn_lkflags = LK_EXCLUSIVE;
1182 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1185 * Search a new directory.
1187 * The last component of the filename is left accessible via
1188 * cnp->cn_nameptr for callers that need the name. Callers needing
1189 * the name set the SAVENAME flag. When done, they assume
1190 * responsibility for freeing the pathname buffer.
1192 #ifdef NAMEI_DIAGNOSTIC
1193 printf("{%s}: ", cnp->cn_nameptr);
1197 * Check for "" which represents the root directory after slash
1200 if (cnp->cn_nameptr[0] == '\0') {
1202 * Support only LOOKUP for "/" because lookup()
1203 * can't succeed for CREATE, DELETE and RENAME.
1205 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1206 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1208 if (!(cnp->cn_flags & LOCKLEAF))
1211 /* XXX This should probably move to the top of function. */
1212 if (cnp->cn_flags & SAVESTART)
1213 panic("lookup: SAVESTART");
1217 if (cnp->cn_flags & ISDOTDOT)
1218 panic ("relookup: lookup on dot-dot");
1221 * We now have a segment name to search for, and a directory to search.
1223 #ifdef NAMEI_DIAGNOSTIC
1224 vn_printf(dp, "search in ");
1226 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1227 KASSERT(*vpp == NULL, ("leaf should be empty"));
1228 if (error != EJUSTRETURN)
1231 * If creating and at end of pathname, then can consider
1232 * allowing file to be created.
1238 /* ASSERT(dvp == ndp->ni_startdir) */
1239 if (cnp->cn_flags & SAVESTART)
1241 if ((cnp->cn_flags & LOCKPARENT) == 0)
1244 * We return with ni_vp NULL to indicate that the entry
1245 * doesn't currently exist, leaving a pointer to the
1246 * (possibly locked) directory vnode in ndp->ni_dvp.
1254 * Disallow directory write attempts on read-only filesystems.
1257 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1266 * Set the parent lock/ref state to the requested state.
1268 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1273 } else if (!wantparent)
1276 * Check for symbolic link
1278 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1279 ("relookup: symlink found.\n"));
1281 /* ASSERT(dvp == ndp->ni_startdir) */
1282 if (cnp->cn_flags & SAVESTART)
1285 if ((cnp->cn_flags & LOCKLEAF) == 0)
1295 NDINIT_ALL(struct nameidata *ndp, u_long op, u_long flags, enum uio_seg segflg,
1296 const char *namep, int dirfd, struct vnode *startdir, cap_rights_t *rightsp,
1300 ndp->ni_cnd.cn_nameiop = op;
1301 ndp->ni_cnd.cn_flags = flags;
1302 ndp->ni_segflg = segflg;
1303 ndp->ni_dirp = namep;
1304 ndp->ni_dirfd = dirfd;
1305 ndp->ni_startdir = startdir;
1306 ndp->ni_resflags = 0;
1307 filecaps_init(&ndp->ni_filecaps);
1308 ndp->ni_cnd.cn_thread = td;
1309 if (rightsp != NULL)
1310 ndp->ni_rightsneeded = *rightsp;
1312 cap_rights_init(&ndp->ni_rightsneeded);
1316 * Free data allocated by namei(); see namei(9) for details.
1319 NDFREE(struct nameidata *ndp, const u_int flags)
1327 if (!(flags & NDF_NO_FREE_PNBUF) &&
1328 (ndp->ni_cnd.cn_flags & HASBUF)) {
1329 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1330 ndp->ni_cnd.cn_flags &= ~HASBUF;
1332 if (!(flags & NDF_NO_VP_UNLOCK) &&
1333 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1335 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1344 VOP_UNLOCK(ndp->ni_vp, 0);
1345 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1346 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1347 ndp->ni_dvp != ndp->ni_vp)
1349 if (!(flags & NDF_NO_DVP_RELE) &&
1350 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1359 VOP_UNLOCK(ndp->ni_dvp, 0);
1360 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1361 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1362 vrele(ndp->ni_startdir);
1363 ndp->ni_startdir = NULL;
1368 * Determine if there is a suitable alternate filename under the specified
1369 * prefix for the specified path. If the create flag is set, then the
1370 * alternate prefix will be used so long as the parent directory exists.
1371 * This is used by the various compatibility ABIs so that Linux binaries prefer
1372 * files under /compat/linux for example. The chosen path (whether under
1373 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1374 * to by pathbuf. The caller is responsible for free'ing the buffer from
1375 * the M_TEMP bucket if one is returned.
1378 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1379 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1381 struct nameidata nd, ndroot;
1382 char *ptr, *buf, *cp;
1386 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1389 /* Copy the prefix into the new pathname as a starting point. */
1390 len = strlcpy(buf, prefix, MAXPATHLEN);
1391 if (len >= MAXPATHLEN) {
1396 sz = MAXPATHLEN - len;
1399 /* Append the filename to the prefix. */
1400 if (pathseg == UIO_SYSSPACE)
1401 error = copystr(path, ptr, sz, &len);
1403 error = copyinstr(path, ptr, sz, &len);
1411 /* Only use a prefix with absolute pathnames. */
1417 if (dirfd != AT_FDCWD) {
1419 * We want the original because the "prefix" is
1420 * included in the already opened dirfd.
1422 bcopy(ptr, buf, len);
1427 * We know that there is a / somewhere in this pathname.
1428 * Search backwards for it, to find the file's parent dir
1429 * to see if it exists in the alternate tree. If it does,
1430 * and we want to create a file (cflag is set). We don't
1431 * need to worry about the root comparison in this case.
1435 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1438 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1444 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1451 * We now compare the vnode of the prefix to the one
1452 * vnode asked. If they resolve to be the same, then we
1453 * ignore the match so that the real root gets used.
1454 * This avoids the problem of traversing "../.." to find the
1455 * root directory and never finding it, because "/" resolves
1456 * to the emulation root directory. This is expensive :-(
1458 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1461 /* We shouldn't ever get an error from this namei(). */
1462 error = namei(&ndroot);
1464 if (nd.ni_vp == ndroot.ni_vp)
1467 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1468 vrele(ndroot.ni_vp);
1472 NDFREE(&nd, NDF_ONLY_PNBUF);
1476 /* If there was an error, use the original path name. */
1478 bcopy(ptr, buf, len);