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;
177 struct componentname *cnp;
179 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp->v_type != VDIR)
182 if ((cnp->cn_flags & BENEATH) != 0 &&
183 (ndp->ni_lcf & NI_LCF_BENEATH_LATCHED) == 0) {
184 MPASS((ndp->ni_lcf & NI_LCF_LATCH) != 0);
185 if (dp != ndp->ni_beneath_latch)
187 ndp->ni_lcf |= NI_LCF_BENEATH_LATCHED;
189 nt = uma_zalloc(nt_zone, M_WAITOK);
192 TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link);
196 nameicap_cleanup(struct nameidata *ndp, bool clean_latch)
198 struct nameicap_tracker *nt, *nt1;
200 KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) ||
201 (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative"));
202 TAILQ_FOREACH_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) {
203 TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link);
205 uma_zfree(nt_zone, nt);
207 if (clean_latch && (ndp->ni_lcf & NI_LCF_LATCH) != 0) {
208 ndp->ni_lcf &= ~NI_LCF_LATCH;
209 vrele(ndp->ni_beneath_latch);
214 * For dotdot lookups in capability mode, only allow the component
215 * lookup to succeed if the resulting directory was already traversed
216 * during the operation. This catches situations where already
217 * traversed directory is moved to different parent, and then we walk
218 * over it with dotdots.
220 * Also allow to force failure of dotdot lookups for non-local
221 * filesystems, where external agents might assist local lookups to
222 * escape the compartment.
225 nameicap_check_dotdot(struct nameidata *ndp, struct vnode *dp)
227 struct nameicap_tracker *nt;
230 if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 || dp == NULL ||
234 if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL &&
235 (mp->mnt_flag & MNT_LOCAL) == 0)
236 return (ENOTCAPABLE);
237 TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head,
239 if ((ndp->ni_lcf & NI_LCF_LATCH) != 0 &&
240 ndp->ni_beneath_latch == nt->dp) {
241 ndp->ni_lcf &= ~NI_LCF_BENEATH_LATCHED;
242 nameicap_cleanup(ndp, false);
248 return (ENOTCAPABLE);
252 namei_cleanup_cnp(struct componentname *cnp)
255 uma_zfree(namei_zone, cnp->cn_pnbuf);
257 cnp->cn_pnbuf = NULL;
258 cnp->cn_nameptr = NULL;
263 namei_handle_root(struct nameidata *ndp, struct vnode **dpp)
265 struct componentname *cnp;
268 if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0) {
270 if (KTRPOINT(curthread, KTR_CAPFAIL))
271 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
273 return (ENOTCAPABLE);
275 if ((cnp->cn_flags & BENEATH) != 0) {
276 ndp->ni_lcf |= NI_LCF_BENEATH_ABS;
277 ndp->ni_lcf &= ~NI_LCF_BENEATH_LATCHED;
278 nameicap_cleanup(ndp, false);
280 while (*(cnp->cn_nameptr) == '/') {
284 *dpp = ndp->ni_rootdir;
290 * Convert a pathname into a pointer to a locked vnode.
292 * The FOLLOW flag is set when symbolic links are to be followed
293 * when they occur at the end of the name translation process.
294 * Symbolic links are always followed for all other pathname
295 * components other than the last.
297 * The segflg defines whether the name is to be copied from user
298 * space or kernel space.
300 * Overall outline of namei:
303 * get starting directory
304 * while (!done && !error) {
305 * call lookup to search path.
306 * if symbolic link, massage name in buffer and continue
310 namei(struct nameidata *ndp)
312 struct filedesc *fdp; /* pointer to file descriptor state */
313 char *cp; /* pointer into pathname argument */
314 struct vnode *dp; /* the directory we are searching */
315 struct iovec aiov; /* uio for reading symbolic links */
316 struct componentname *cnp;
321 struct filecaps dirfd_caps;
323 int error, linklen, startdir_used;
328 ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred;
329 KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc"));
330 KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0,
331 ("namei: nameiop contaminated with flags"));
332 KASSERT((cnp->cn_flags & OPMASK) == 0,
333 ("namei: flags contaminated with nameiops"));
334 MPASS(ndp->ni_startdir == NULL || ndp->ni_startdir->v_type == VDIR ||
335 ndp->ni_startdir->v_type == VBAD);
337 TAILQ_INIT(&ndp->ni_cap_tracker);
340 /* We will set this ourselves if we need it. */
341 cnp->cn_flags &= ~TRAILINGSLASH;
344 * Get a buffer for the name to be translated, and copy the
345 * name into the buffer.
347 if ((cnp->cn_flags & HASBUF) == 0)
348 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
349 if (ndp->ni_segflg == UIO_SYSSPACE)
350 error = copystr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
353 error = copyinstr(ndp->ni_dirp, cnp->cn_pnbuf, MAXPATHLEN,
357 * Don't allow empty pathnames.
359 if (error == 0 && *cnp->cn_pnbuf == '\0')
362 #ifdef CAPABILITY_MODE
364 * In capability mode, lookups must be restricted to happen in
365 * the subtree with the root specified by the file descriptor:
366 * - The root must be real file descriptor, not the pseudo-descriptor
368 * - The passed path must be relative and not absolute.
369 * - If lookup_cap_dotdot is disabled, path must not contain the
371 * - If lookup_cap_dotdot is enabled, we verify that all '..'
372 * components lookups result in the directories which were
373 * previously walked by us, which prevents an escape from
376 if (error == 0 && IN_CAPABILITY_MODE(td) &&
377 (cnp->cn_flags & NOCAPCHECK) == 0) {
378 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
379 ndp->ni_resflags |= NIRES_STRICTREL;
380 if (ndp->ni_dirfd == AT_FDCWD) {
382 if (KTRPOINT(td, KTR_CAPFAIL))
383 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
390 namei_cleanup_cnp(cnp);
396 if (KTRPOINT(td, KTR_NAMEI)) {
397 KASSERT(cnp->cn_thread == curthread,
398 ("namei not using curthread"));
399 ktrnamei(cnp->cn_pnbuf);
403 * Get starting point for the translation.
406 ndp->ni_rootdir = fdp->fd_rdir;
407 vrefact(ndp->ni_rootdir);
408 ndp->ni_topdir = fdp->fd_jdir;
411 * If we are auditing the kernel pathname, save the user pathname.
413 if (cnp->cn_flags & AUDITVNODE1)
414 AUDIT_ARG_UPATH1(td, ndp->ni_dirfd, cnp->cn_pnbuf);
415 if (cnp->cn_flags & AUDITVNODE2)
416 AUDIT_ARG_UPATH2(td, ndp->ni_dirfd, cnp->cn_pnbuf);
420 cnp->cn_nameptr = cnp->cn_pnbuf;
421 if (cnp->cn_pnbuf[0] == '/') {
422 ndp->ni_resflags |= NIRES_ABS;
423 error = namei_handle_root(ndp, &dp);
425 if (ndp->ni_startdir != NULL) {
426 dp = ndp->ni_startdir;
428 } else if (ndp->ni_dirfd == AT_FDCWD) {
432 rights = ndp->ni_rightsneeded;
433 cap_rights_set(&rights, CAP_LOOKUP);
435 if (cnp->cn_flags & AUDITVNODE1)
436 AUDIT_ARG_ATFD1(ndp->ni_dirfd);
437 if (cnp->cn_flags & AUDITVNODE2)
438 AUDIT_ARG_ATFD2(ndp->ni_dirfd);
440 * Effectively inlined fgetvp_rights, because we need to
441 * inspect the file as well as grabbing the vnode.
443 error = fget_cap_locked(fdp, ndp->ni_dirfd, &rights,
444 &dfp, &ndp->ni_filecaps);
447 * Preserve the error; it should either be EBADF
448 * or capability-related, both of which can be
449 * safely returned to the caller.
451 } else if (dfp->f_ops == &badfileops) {
453 } else if (dfp->f_vnode == NULL) {
459 if ((dfp->f_flag & FSEARCH) != 0)
460 cnp->cn_flags |= NOEXECCHECK;
464 * If file descriptor doesn't have all rights,
465 * all lookups relative to it must also be
469 if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights,
471 ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL ||
472 ndp->ni_filecaps.fc_nioctls != -1) {
473 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE;
474 ndp->ni_resflags |= NIRES_STRICTREL;
478 if (error == 0 && dp->v_type != VDIR)
481 if (error == 0 && (cnp->cn_flags & BENEATH) != 0) {
482 if (ndp->ni_dirfd == AT_FDCWD) {
483 ndp->ni_beneath_latch = fdp->fd_cdir;
484 vrefact(ndp->ni_beneath_latch);
486 rights = ndp->ni_rightsneeded;
487 cap_rights_set(&rights, CAP_LOOKUP);
488 error = fgetvp_rights(td, ndp->ni_dirfd, &rights,
489 &dirfd_caps, &ndp->ni_beneath_latch);
490 if (error == 0 && dp->v_type != VDIR) {
491 vrele(ndp->ni_beneath_latch);
496 ndp->ni_lcf |= NI_LCF_LATCH;
498 FILEDESC_SUNLOCK(fdp);
500 if (error == 0 && (cnp->cn_flags & RBENEATH) != 0) {
501 if (cnp->cn_pnbuf[0] == '/' ||
502 (ndp->ni_lcf & NI_LCF_BENEATH_ABS) != 0) {
504 } else if ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) == 0) {
505 ndp->ni_lcf |= NI_LCF_STRICTRELATIVE |
510 if (ndp->ni_startdir != NULL && !startdir_used)
511 vrele(ndp->ni_startdir);
517 MPASS((ndp->ni_lcf & (NI_LCF_BENEATH_ABS | NI_LCF_LATCH)) !=
519 if (((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) != 0 &&
520 lookup_cap_dotdot != 0) ||
521 ((ndp->ni_lcf & NI_LCF_STRICTRELATIVE) == 0 &&
522 (cnp->cn_flags & BENEATH) != 0))
523 ndp->ni_lcf |= NI_LCF_CAP_DOTDOT;
524 SDT_PROBE3(vfs, namei, lookup, entry, dp, cnp->cn_pnbuf,
527 ndp->ni_startdir = dp;
531 * Override an error to not allow user to use
532 * BENEATH as an oracle.
534 if ((ndp->ni_lcf & (NI_LCF_LATCH |
535 NI_LCF_BENEATH_LATCHED)) == NI_LCF_LATCH)
541 * If not a symbolic link, we're done.
543 if ((cnp->cn_flags & ISSYMLINK) == 0) {
544 vrele(ndp->ni_rootdir);
545 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0) {
546 namei_cleanup_cnp(cnp);
548 cnp->cn_flags |= HASBUF;
549 if ((ndp->ni_lcf & (NI_LCF_LATCH |
550 NI_LCF_BENEATH_LATCHED)) == NI_LCF_LATCH) {
554 nameicap_cleanup(ndp, true);
555 SDT_PROBE2(vfs, namei, lookup, return, error,
556 (error == 0 ? ndp->ni_vp : NULL));
559 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
564 if ((cnp->cn_flags & NOMACCHECK) == 0) {
565 error = mac_vnode_check_readlink(td->td_ucred,
571 if (ndp->ni_pathlen > 1)
572 cp = uma_zalloc(namei_zone, M_WAITOK);
576 aiov.iov_len = MAXPATHLEN;
577 auio.uio_iov = &aiov;
580 auio.uio_rw = UIO_READ;
581 auio.uio_segflg = UIO_SYSSPACE;
583 auio.uio_resid = MAXPATHLEN;
584 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
586 if (ndp->ni_pathlen > 1)
587 uma_zfree(namei_zone, cp);
590 linklen = MAXPATHLEN - auio.uio_resid;
592 if (ndp->ni_pathlen > 1)
593 uma_zfree(namei_zone, cp);
597 if (linklen + ndp->ni_pathlen > MAXPATHLEN) {
598 if (ndp->ni_pathlen > 1)
599 uma_zfree(namei_zone, cp);
600 error = ENAMETOOLONG;
603 if (ndp->ni_pathlen > 1) {
604 bcopy(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
605 uma_zfree(namei_zone, cnp->cn_pnbuf);
608 cnp->cn_pnbuf[linklen] = '\0';
609 ndp->ni_pathlen += linklen;
613 * Check if root directory should replace current directory.
615 cnp->cn_nameptr = cnp->cn_pnbuf;
616 if (*(cnp->cn_nameptr) == '/') {
618 error = namei_handle_root(ndp, &dp);
627 vrele(ndp->ni_rootdir);
629 namei_cleanup_cnp(cnp);
630 nameicap_cleanup(ndp, true);
631 SDT_PROBE2(vfs, namei, lookup, return, error, NULL);
636 compute_cn_lkflags(struct mount *mp, int lkflags, int cnflags)
639 if (mp == NULL || ((lkflags & LK_SHARED) &&
640 (!(mp->mnt_kern_flag & MNTK_LOOKUP_SHARED) ||
641 ((cnflags & ISDOTDOT) &&
642 (mp->mnt_kern_flag & MNTK_LOOKUP_EXCL_DOTDOT))))) {
643 lkflags &= ~LK_SHARED;
644 lkflags |= LK_EXCLUSIVE;
646 lkflags |= LK_NODDLKTREAT;
651 needs_exclusive_leaf(struct mount *mp, int flags)
655 * Intermediate nodes can use shared locks, we only need to
656 * force an exclusive lock for leaf nodes.
658 if ((flags & (ISLASTCN | LOCKLEAF)) != (ISLASTCN | LOCKLEAF))
661 /* Always use exclusive locks if LOCKSHARED isn't set. */
662 if (!(flags & LOCKSHARED))
666 * For lookups during open(), if the mount point supports
667 * extended shared operations, then use a shared lock for the
668 * leaf node, otherwise use an exclusive lock.
670 if ((flags & ISOPEN) != 0)
671 return (!MNT_EXTENDED_SHARED(mp));
674 * Lookup requests outside of open() that specify LOCKSHARED
675 * only need a shared lock on the leaf vnode.
682 * This is a very central and rather complicated routine.
684 * The pathname is pointed to by ni_ptr and is of length ni_pathlen.
685 * The starting directory is taken from ni_startdir. The pathname is
686 * descended until done, or a symbolic link is encountered. The variable
687 * ni_more is clear if the path is completed; it is set to one if a
688 * symbolic link needing interpretation is encountered.
690 * The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
691 * whether the name is to be looked up, created, renamed, or deleted.
692 * When CREATE, RENAME, or DELETE is specified, information usable in
693 * creating, renaming, or deleting a directory entry may be calculated.
694 * If flag has LOCKPARENT or'ed into it, the parent directory is returned
695 * locked. If flag has WANTPARENT or'ed into it, the parent directory is
696 * returned unlocked. Otherwise the parent directory is not returned. If
697 * the target of the pathname exists and LOCKLEAF is or'ed into the flag
698 * the target is returned locked, otherwise it is returned unlocked.
699 * When creating or renaming and LOCKPARENT is specified, the target may not
700 * be ".". When deleting and LOCKPARENT is specified, the target may be ".".
702 * Overall outline of lookup:
705 * identify next component of name at ndp->ni_ptr
706 * handle degenerate case where name is null string
707 * if .. and crossing mount points and on mounted filesys, find parent
708 * call VOP_LOOKUP routine for next component name
709 * directory vnode returned in ni_dvp, unlocked unless LOCKPARENT set
710 * component vnode returned in ni_vp (if it exists), locked.
711 * if result vnode is mounted on and crossing mount points,
712 * find mounted on vnode
713 * if more components of name, do next level at dirloop
714 * return the answer in ni_vp, locked if LOCKLEAF set
715 * if LOCKPARENT set, return locked parent in ni_dvp
716 * if WANTPARENT set, return unlocked parent in ni_dvp
719 lookup(struct nameidata *ndp)
721 char *cp; /* pointer into pathname argument */
722 char *prev_ni_next; /* saved ndp->ni_next */
723 struct vnode *dp = NULL; /* the directory we are searching */
724 struct vnode *tdp; /* saved dp */
725 struct mount *mp; /* mount table entry */
727 size_t prev_ni_pathlen; /* saved ndp->ni_pathlen */
728 int docache; /* == 0 do not cache last component */
729 int wantparent; /* 1 => wantparent or lockparent flag */
730 int rdonly; /* lookup read-only flag bit */
732 int dpunlocked = 0; /* dp has already been unlocked */
733 int relookup = 0; /* do not consume the path component */
734 struct componentname *cnp = &ndp->ni_cnd;
739 * Setup: break out flag bits into variables.
742 wantparent = cnp->cn_flags & (LOCKPARENT | WANTPARENT);
743 KASSERT(cnp->cn_nameiop == LOOKUP || wantparent,
744 ("CREATE, DELETE, RENAME require LOCKPARENT or WANTPARENT."));
746 * When set to zero, docache causes the last component of the
747 * pathname to be deleted from the cache and the full lookup
748 * of the name to be done (via VOP_CACHEDLOOKUP()). Often
749 * filesystems need some pre-computed values that are made
750 * during the full lookup, for instance UFS sets dp->i_offset.
752 * The docache variable is set to zero when requested by the
753 * NOCACHE flag and for all modifying operations except CREATE.
755 docache = (cnp->cn_flags & NOCACHE) ^ NOCACHE;
756 if (cnp->cn_nameiop == DELETE ||
757 (wantparent && cnp->cn_nameiop != CREATE &&
758 cnp->cn_nameiop != LOOKUP))
760 rdonly = cnp->cn_flags & RDONLY;
761 cnp->cn_flags &= ~ISSYMLINK;
764 * We use shared locks until we hit the parent of the last cn then
765 * we adjust based on the requesting flags.
767 cnp->cn_lkflags = LK_SHARED;
768 dp = ndp->ni_startdir;
769 ndp->ni_startdir = NULLVP;
771 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags | LK_RETRY,
776 * Search a new directory.
778 * The last component of the filename is left accessible via
779 * cnp->cn_nameptr for callers that need the name. Callers needing
780 * the name set the SAVENAME flag. When done, they assume
781 * responsibility for freeing the pathname buffer.
783 for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++)
785 cnp->cn_namelen = cp - cnp->cn_nameptr;
786 if (cnp->cn_namelen > NAME_MAX) {
787 error = ENAMETOOLONG;
790 #ifdef NAMEI_DIAGNOSTIC
793 printf("{%s}: ", cnp->cn_nameptr);
796 prev_ni_pathlen = ndp->ni_pathlen;
797 ndp->ni_pathlen -= cnp->cn_namelen;
798 KASSERT(ndp->ni_pathlen <= PATH_MAX,
799 ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen));
800 prev_ni_next = ndp->ni_next;
804 * Replace multiple slashes by a single slash and trailing slashes
805 * by a null. This must be done before VOP_LOOKUP() because some
806 * fs's don't know about trailing slashes. Remember if there were
807 * trailing slashes to handle symlinks, existing non-directories
808 * and non-existing files that won't be directories specially later.
810 while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) {
814 *ndp->ni_next = '\0';
815 cnp->cn_flags |= TRAILINGSLASH;
820 cnp->cn_flags |= MAKEENTRY;
821 if (*cp == '\0' && docache == 0)
822 cnp->cn_flags &= ~MAKEENTRY;
823 if (cnp->cn_namelen == 2 &&
824 cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.')
825 cnp->cn_flags |= ISDOTDOT;
827 cnp->cn_flags &= ~ISDOTDOT;
828 if (*ndp->ni_next == 0)
829 cnp->cn_flags |= ISLASTCN;
831 cnp->cn_flags &= ~ISLASTCN;
833 if ((cnp->cn_flags & ISLASTCN) != 0 &&
834 cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.' &&
835 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
840 nameicap_tracker_add(ndp, dp);
843 * Check for degenerate name (e.g. / or "")
844 * which is a way of talking about a directory,
845 * e.g. like "/." or ".".
847 if (cnp->cn_nameptr[0] == '\0') {
848 if (dp->v_type != VDIR) {
852 if (cnp->cn_nameiop != LOOKUP) {
862 if (cnp->cn_flags & AUDITVNODE1)
863 AUDIT_ARG_VNODE1(dp);
864 else if (cnp->cn_flags & AUDITVNODE2)
865 AUDIT_ARG_VNODE2(dp);
867 if (!(cnp->cn_flags & (LOCKPARENT | LOCKLEAF)))
869 /* XXX This should probably move to the top of function. */
870 if (cnp->cn_flags & SAVESTART)
871 panic("lookup: SAVESTART");
876 * Handle "..": five special cases.
877 * 0. If doing a capability lookup and lookup_cap_dotdot is
878 * disabled, return ENOTCAPABLE.
879 * 1. Return an error if this is the last component of
880 * the name and the operation is DELETE or RENAME.
881 * 2. If at root directory (e.g. after chroot)
882 * or at absolute root directory
883 * then ignore it so can't get out.
884 * 3. If this vnode is the root of a mounted
885 * filesystem, then replace it with the
886 * vnode which was mounted on so we take the
887 * .. in the other filesystem.
888 * 4. If the vnode is the top directory of
889 * the jail or chroot, don't let them out.
890 * 5. If doing a capability lookup and lookup_cap_dotdot is
891 * enabled, return ENOTCAPABLE if the lookup would escape
892 * from the initial file descriptor directory. Checks are
893 * done by ensuring that namei() already traversed the
894 * result of dotdot lookup.
896 if (cnp->cn_flags & ISDOTDOT) {
897 if ((ndp->ni_lcf & (NI_LCF_STRICTRELATIVE | NI_LCF_CAP_DOTDOT))
898 == NI_LCF_STRICTRELATIVE) {
900 if (KTRPOINT(curthread, KTR_CAPFAIL))
901 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
906 if ((cnp->cn_flags & ISLASTCN) != 0 &&
907 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
912 for (pr = cnp->cn_cred->cr_prison; pr != NULL;
914 if (dp == pr->pr_root)
916 if (dp == ndp->ni_rootdir ||
917 dp == ndp->ni_topdir ||
920 ((dp->v_vflag & VV_ROOT) != 0 &&
921 (cnp->cn_flags & NOCROSSMOUNT) != 0)) {
927 if ((dp->v_vflag & VV_ROOT) == 0)
929 if (dp->v_iflag & VI_DOOMED) { /* forced unmount */
934 dp = dp->v_mount->mnt_vnodecovered;
938 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
939 LK_RETRY, ISDOTDOT));
940 error = nameicap_check_dotdot(ndp, dp);
943 if (KTRPOINT(curthread, KTR_CAPFAIL))
944 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
952 * We now have a segment name to search for, and a directory to search.
956 if ((cnp->cn_flags & NOMACCHECK) == 0) {
957 error = mac_vnode_check_lookup(cnp->cn_thread->td_ucred, dp,
965 ASSERT_VOP_LOCKED(dp, "lookup");
967 * If we have a shared lock we may need to upgrade the lock for the
970 if ((cnp->cn_flags & LOCKPARENT) && (cnp->cn_flags & ISLASTCN) &&
971 dp != vp_crossmp && VOP_ISLOCKED(dp) == LK_SHARED)
972 vn_lock(dp, LK_UPGRADE|LK_RETRY);
973 if ((dp->v_iflag & VI_DOOMED) != 0) {
978 * If we're looking up the last component and we need an exclusive
979 * lock, adjust our lkflags.
981 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags))
982 cnp->cn_lkflags = LK_EXCLUSIVE;
983 #ifdef NAMEI_DIAGNOSTIC
984 vn_printf(dp, "lookup in ");
986 lkflags_save = cnp->cn_lkflags;
987 cnp->cn_lkflags = compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags,
989 error = VOP_LOOKUP(dp, &ndp->ni_vp, cnp);
990 cnp->cn_lkflags = lkflags_save;
992 KASSERT(ndp->ni_vp == NULL, ("leaf should be empty"));
993 #ifdef NAMEI_DIAGNOSTIC
994 printf("not found\n");
996 if ((error == ENOENT) &&
997 (dp->v_vflag & VV_ROOT) && (dp->v_mount != NULL) &&
998 (dp->v_mount->mnt_flag & MNT_UNION)) {
1000 dp = dp->v_mount->mnt_vnodecovered;
1004 compute_cn_lkflags(dp->v_mount, cnp->cn_lkflags |
1005 LK_RETRY, cnp->cn_flags));
1006 nameicap_tracker_add(ndp, dp);
1010 if (error == ERELOOKUP) {
1018 if (error != EJUSTRETURN)
1021 * At this point, we know we're at the end of the
1022 * pathname. If creating / renaming, we can consider
1023 * allowing the file or directory to be created / renamed,
1024 * provided we're not on a read-only filesystem.
1030 /* trailing slash only allowed for directories */
1031 if ((cnp->cn_flags & TRAILINGSLASH) &&
1032 !(cnp->cn_flags & WILLBEDIR)) {
1036 if ((cnp->cn_flags & LOCKPARENT) == 0)
1039 * We return with ni_vp NULL to indicate that the entry
1040 * doesn't currently exist, leaving a pointer to the
1041 * (possibly locked) directory vnode in ndp->ni_dvp.
1043 if (cnp->cn_flags & SAVESTART) {
1044 ndp->ni_startdir = ndp->ni_dvp;
1045 VREF(ndp->ni_startdir);
1051 #ifdef NAMEI_DIAGNOSTIC
1057 * Check to see if the vnode has been mounted on;
1058 * if so find the root of the mounted filesystem.
1060 while (dp->v_type == VDIR && (mp = dp->v_mountedhere) &&
1061 (cnp->cn_flags & NOCROSSMOUNT) == 0) {
1062 if (vfs_busy(mp, 0))
1065 if (dp != ndp->ni_dvp)
1069 vrefact(vp_crossmp);
1070 ndp->ni_dvp = vp_crossmp;
1071 error = VFS_ROOT(mp, compute_cn_lkflags(mp, cnp->cn_lkflags,
1072 cnp->cn_flags), &tdp);
1074 if (vn_lock(vp_crossmp, LK_SHARED | LK_NOWAIT))
1075 panic("vp_crossmp exclusively locked or reclaimed");
1080 ndp->ni_vp = dp = tdp;
1084 * Check for symbolic link
1086 if ((dp->v_type == VLNK) &&
1087 ((cnp->cn_flags & FOLLOW) || (cnp->cn_flags & TRAILINGSLASH) ||
1088 *ndp->ni_next == '/')) {
1089 cnp->cn_flags |= ISSYMLINK;
1090 if (dp->v_iflag & VI_DOOMED) {
1092 * We can't know whether the directory was mounted with
1093 * NOSYMFOLLOW, so we can't follow safely.
1098 if (dp->v_mount->mnt_flag & MNT_NOSYMFOLLOW) {
1103 * Symlink code always expects an unlocked dvp.
1105 if (ndp->ni_dvp != ndp->ni_vp) {
1106 VOP_UNLOCK(ndp->ni_dvp, 0);
1107 ni_dvp_unlocked = 1;
1114 * Not a symbolic link that we will follow. Continue with the
1115 * next component if there is any; otherwise, we're done.
1117 KASSERT((cnp->cn_flags & ISLASTCN) || *ndp->ni_next == '/',
1118 ("lookup: invalid path state."));
1121 ndp->ni_pathlen = prev_ni_pathlen;
1122 ndp->ni_next = prev_ni_next;
1123 if (ndp->ni_dvp != dp)
1129 if (cnp->cn_flags & ISDOTDOT) {
1130 error = nameicap_check_dotdot(ndp, ndp->ni_vp);
1133 if (KTRPOINT(curthread, KTR_CAPFAIL))
1134 ktrcapfail(CAPFAIL_LOOKUP, NULL, NULL);
1139 if (*ndp->ni_next == '/') {
1140 cnp->cn_nameptr = ndp->ni_next;
1141 while (*cnp->cn_nameptr == '/') {
1145 if (ndp->ni_dvp != dp)
1152 * If we're processing a path with a trailing slash,
1153 * check that the end result is a directory.
1155 if ((cnp->cn_flags & TRAILINGSLASH) && dp->v_type != VDIR) {
1160 * Disallow directory write attempts on read-only filesystems.
1163 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1167 if (cnp->cn_flags & SAVESTART) {
1168 ndp->ni_startdir = ndp->ni_dvp;
1169 VREF(ndp->ni_startdir);
1172 ni_dvp_unlocked = 2;
1173 if (ndp->ni_dvp != dp)
1177 } else if ((cnp->cn_flags & LOCKPARENT) == 0 && ndp->ni_dvp != dp) {
1178 VOP_UNLOCK(ndp->ni_dvp, 0);
1179 ni_dvp_unlocked = 1;
1182 if (cnp->cn_flags & AUDITVNODE1)
1183 AUDIT_ARG_VNODE1(dp);
1184 else if (cnp->cn_flags & AUDITVNODE2)
1185 AUDIT_ARG_VNODE2(dp);
1187 if ((cnp->cn_flags & LOCKLEAF) == 0)
1191 * Because of shared lookup we may have the vnode shared locked, but
1192 * the caller may want it to be exclusively locked.
1194 if (needs_exclusive_leaf(dp->v_mount, cnp->cn_flags) &&
1195 VOP_ISLOCKED(dp) != LK_EXCLUSIVE) {
1196 vn_lock(dp, LK_UPGRADE | LK_RETRY);
1197 if (dp->v_iflag & VI_DOOMED) {
1202 if (ndp->ni_vp != NULL && ndp->ni_vp->v_type == VDIR)
1203 nameicap_tracker_add(ndp, ndp->ni_vp);
1207 if (ni_dvp_unlocked != 2) {
1208 if (dp != ndp->ni_dvp && !ni_dvp_unlocked)
1221 * relookup - lookup a path name component
1222 * Used by lookup to re-acquire things.
1225 relookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
1227 struct vnode *dp = NULL; /* the directory we are searching */
1228 int wantparent; /* 1 => wantparent or lockparent flag */
1229 int rdonly; /* lookup read-only flag bit */
1232 KASSERT(cnp->cn_flags & ISLASTCN,
1233 ("relookup: Not given last component."));
1235 * Setup: break out flag bits into variables.
1237 wantparent = cnp->cn_flags & (LOCKPARENT|WANTPARENT);
1238 KASSERT(wantparent, ("relookup: parent not wanted."));
1239 rdonly = cnp->cn_flags & RDONLY;
1240 cnp->cn_flags &= ~ISSYMLINK;
1242 cnp->cn_lkflags = LK_EXCLUSIVE;
1243 vn_lock(dp, LK_EXCLUSIVE | LK_RETRY);
1246 * Search a new directory.
1248 * The last component of the filename is left accessible via
1249 * cnp->cn_nameptr for callers that need the name. Callers needing
1250 * the name set the SAVENAME flag. When done, they assume
1251 * responsibility for freeing the pathname buffer.
1253 #ifdef NAMEI_DIAGNOSTIC
1254 printf("{%s}: ", cnp->cn_nameptr);
1258 * Check for "" which represents the root directory after slash
1261 if (cnp->cn_nameptr[0] == '\0') {
1263 * Support only LOOKUP for "/" because lookup()
1264 * can't succeed for CREATE, DELETE and RENAME.
1266 KASSERT(cnp->cn_nameiop == LOOKUP, ("nameiop must be LOOKUP"));
1267 KASSERT(dp->v_type == VDIR, ("dp is not a directory"));
1269 if (!(cnp->cn_flags & LOCKLEAF))
1272 /* XXX This should probably move to the top of function. */
1273 if (cnp->cn_flags & SAVESTART)
1274 panic("lookup: SAVESTART");
1278 if (cnp->cn_flags & ISDOTDOT)
1279 panic ("relookup: lookup on dot-dot");
1282 * We now have a segment name to search for, and a directory to search.
1284 #ifdef NAMEI_DIAGNOSTIC
1285 vn_printf(dp, "search in ");
1287 if ((error = VOP_LOOKUP(dp, vpp, cnp)) != 0) {
1288 KASSERT(*vpp == NULL, ("leaf should be empty"));
1289 if (error != EJUSTRETURN)
1292 * If creating and at end of pathname, then can consider
1293 * allowing file to be created.
1299 /* ASSERT(dvp == ndp->ni_startdir) */
1300 if (cnp->cn_flags & SAVESTART)
1302 if ((cnp->cn_flags & LOCKPARENT) == 0)
1305 * We return with ni_vp NULL to indicate that the entry
1306 * doesn't currently exist, leaving a pointer to the
1307 * (possibly locked) directory vnode in ndp->ni_dvp.
1315 * Disallow directory write attempts on read-only filesystems.
1318 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) {
1327 * Set the parent lock/ref state to the requested state.
1329 if ((cnp->cn_flags & LOCKPARENT) == 0 && dvp != dp) {
1334 } else if (!wantparent)
1337 * Check for symbolic link
1339 KASSERT(dp->v_type != VLNK || !(cnp->cn_flags & FOLLOW),
1340 ("relookup: symlink found.\n"));
1342 /* ASSERT(dvp == ndp->ni_startdir) */
1343 if (cnp->cn_flags & SAVESTART)
1346 if ((cnp->cn_flags & LOCKLEAF) == 0)
1356 NDINIT_ALL(struct nameidata *ndp, u_long op, u_long flags, enum uio_seg segflg,
1357 const char *namep, int dirfd, struct vnode *startdir, cap_rights_t *rightsp,
1361 ndp->ni_cnd.cn_nameiop = op;
1362 ndp->ni_cnd.cn_flags = flags;
1363 ndp->ni_segflg = segflg;
1364 ndp->ni_dirp = namep;
1365 ndp->ni_dirfd = dirfd;
1366 ndp->ni_startdir = startdir;
1367 ndp->ni_resflags = 0;
1368 if (rightsp != NULL)
1369 ndp->ni_rightsneeded = *rightsp;
1371 cap_rights_init(&ndp->ni_rightsneeded);
1372 filecaps_init(&ndp->ni_filecaps);
1373 ndp->ni_cnd.cn_thread = td;
1377 * Free data allocated by namei(); see namei(9) for details.
1380 NDFREE(struct nameidata *ndp, const u_int flags)
1388 if (!(flags & NDF_NO_FREE_PNBUF) &&
1389 (ndp->ni_cnd.cn_flags & HASBUF)) {
1390 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
1391 ndp->ni_cnd.cn_flags &= ~HASBUF;
1393 if (!(flags & NDF_NO_VP_UNLOCK) &&
1394 (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
1396 if (!(flags & NDF_NO_DVP_UNLOCK) &&
1397 (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
1398 ndp->ni_dvp != ndp->ni_vp)
1400 if (!(flags & NDF_NO_VP_RELE) && ndp->ni_vp) {
1409 VOP_UNLOCK(ndp->ni_vp, 0);
1410 if (!(flags & NDF_NO_DVP_RELE) &&
1411 (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
1420 VOP_UNLOCK(ndp->ni_dvp, 0);
1421 if (!(flags & NDF_NO_STARTDIR_RELE) &&
1422 (ndp->ni_cnd.cn_flags & SAVESTART)) {
1423 vrele(ndp->ni_startdir);
1424 ndp->ni_startdir = NULL;
1429 * Determine if there is a suitable alternate filename under the specified
1430 * prefix for the specified path. If the create flag is set, then the
1431 * alternate prefix will be used so long as the parent directory exists.
1432 * This is used by the various compatibility ABIs so that Linux binaries prefer
1433 * files under /compat/linux for example. The chosen path (whether under
1434 * the prefix or under /) is returned in a kernel malloc'd buffer pointed
1435 * to by pathbuf. The caller is responsible for free'ing the buffer from
1436 * the M_TEMP bucket if one is returned.
1439 kern_alternate_path(struct thread *td, const char *prefix, const char *path,
1440 enum uio_seg pathseg, char **pathbuf, int create, int dirfd)
1442 struct nameidata nd, ndroot;
1443 char *ptr, *buf, *cp;
1447 buf = (char *) malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
1450 /* Copy the prefix into the new pathname as a starting point. */
1451 len = strlcpy(buf, prefix, MAXPATHLEN);
1452 if (len >= MAXPATHLEN) {
1457 sz = MAXPATHLEN - len;
1460 /* Append the filename to the prefix. */
1461 if (pathseg == UIO_SYSSPACE)
1462 error = copystr(path, ptr, sz, &len);
1464 error = copyinstr(path, ptr, sz, &len);
1472 /* Only use a prefix with absolute pathnames. */
1478 if (dirfd != AT_FDCWD) {
1480 * We want the original because the "prefix" is
1481 * included in the already opened dirfd.
1483 bcopy(ptr, buf, len);
1488 * We know that there is a / somewhere in this pathname.
1489 * Search backwards for it, to find the file's parent dir
1490 * to see if it exists in the alternate tree. If it does,
1491 * and we want to create a file (cflag is set). We don't
1492 * need to worry about the root comparison in this case.
1496 for (cp = &ptr[len] - 1; *cp != '/'; cp--);
1499 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1505 NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_SYSSPACE, buf, td);
1512 * We now compare the vnode of the prefix to the one
1513 * vnode asked. If they resolve to be the same, then we
1514 * ignore the match so that the real root gets used.
1515 * This avoids the problem of traversing "../.." to find the
1516 * root directory and never finding it, because "/" resolves
1517 * to the emulation root directory. This is expensive :-(
1519 NDINIT(&ndroot, LOOKUP, FOLLOW, UIO_SYSSPACE, prefix,
1522 /* We shouldn't ever get an error from this namei(). */
1523 error = namei(&ndroot);
1525 if (nd.ni_vp == ndroot.ni_vp)
1528 NDFREE(&ndroot, NDF_ONLY_PNBUF);
1529 vrele(ndroot.ni_vp);
1533 NDFREE(&nd, NDF_ONLY_PNBUF);
1537 /* If there was an error, use the original path name. */
1539 bcopy(ptr, buf, len);