xref: /dragonfly/sys/kern/vfs_syscalls.c (revision fe76c4fb)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_syscalls.c	8.13 (Berkeley) 4/15/94
39  * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $
40  * $DragonFly: src/sys/kern/vfs_syscalls.c,v 1.96 2006/06/05 07:26:10 dillon Exp $
41  */
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/buf.h>
46 #include <sys/conf.h>
47 #include <sys/sysent.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mountctl.h>
51 #include <sys/sysproto.h>
52 #include <sys/filedesc.h>
53 #include <sys/kernel.h>
54 #include <sys/fcntl.h>
55 #include <sys/file.h>
56 #include <sys/linker.h>
57 #include <sys/stat.h>
58 #include <sys/unistd.h>
59 #include <sys/vnode.h>
60 #include <sys/proc.h>
61 #include <sys/namei.h>
62 #include <sys/nlookup.h>
63 #include <sys/dirent.h>
64 #include <sys/extattr.h>
65 #include <sys/spinlock.h>
66 #include <sys/kern_syscall.h>
67 #include <sys/objcache.h>
68 
69 #include <machine/limits.h>
70 #include <vfs/union/union.h>
71 #include <sys/sysctl.h>
72 #include <vm/vm.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_page.h>
75 
76 #include <sys/file2.h>
77 #include <sys/spinlock2.h>
78 
79 static int checkvp_chdir (struct vnode *vn, struct thread *td);
80 static void checkdirs (struct vnode *olddp, struct namecache *ncp);
81 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
82 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
83 static int getutimes (const struct timeval *, struct timespec *);
84 static int setfown (struct vnode *, uid_t, gid_t);
85 static int setfmode (struct vnode *, int);
86 static int setfflags (struct vnode *, int);
87 static int setutimes (struct vnode *, const struct timespec *, int);
88 static int	usermount = 0;	/* if 1, non-root can mount fs. */
89 
90 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *);
91 
92 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "");
93 
94 /*
95  * Virtual File System System Calls
96  */
97 
98 /*
99  * Mount a file system.
100  */
101 /*
102  * mount_args(char *type, char *path, int flags, caddr_t data)
103  */
104 /* ARGSUSED */
105 int
106 sys_mount(struct mount_args *uap)
107 {
108 	struct thread *td = curthread;
109 	struct proc *p = td->td_proc;
110 	struct vnode *vp;
111 	struct namecache *ncp;
112 	struct mount *mp;
113 	struct vfsconf *vfsp;
114 	int error, flag = 0, flag2 = 0;
115 	struct vattr va;
116 	struct nlookupdata nd;
117 	char fstypename[MFSNAMELEN];
118 	struct nlcomponent nlc;
119 	struct ucred *cred = p->p_ucred;
120 
121 	KKASSERT(p);
122 	if (cred->cr_prison != NULL)
123 		return (EPERM);
124 	if (usermount == 0 && (error = suser(td)))
125 		return (error);
126 	/*
127 	 * Do not allow NFS export by non-root users.
128 	 */
129 	if (uap->flags & MNT_EXPORTED) {
130 		error = suser(td);
131 		if (error)
132 			return (error);
133 	}
134 	/*
135 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
136 	 */
137 	if (suser(td))
138 		uap->flags |= MNT_NOSUID | MNT_NODEV;
139 
140 	/*
141 	 * Lookup the requested path and extract the ncp and vnode.
142 	 */
143 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
144 	if (error == 0) {
145 		if ((error = nlookup(&nd)) == 0) {
146 			if (nd.nl_ncp->nc_vp == NULL)
147 				error = ENOENT;
148 		}
149 	}
150 	if (error) {
151 		nlookup_done(&nd);
152 		return (error);
153 	}
154 
155 	/*
156 	 * Extract the locked+refd ncp and cleanup the nd structure
157 	 */
158 	ncp = nd.nl_ncp;
159 	nd.nl_ncp = NULL;
160 	nlookup_done(&nd);
161 
162 	/*
163 	 * now we have the locked ref'd ncp and unreferenced vnode.
164 	 */
165 	vp = ncp->nc_vp;
166 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
167 		cache_put(ncp);
168 		return (error);
169 	}
170 	cache_unlock(ncp);
171 
172 	/*
173 	 * Now we have an unlocked ref'd ncp and a locked ref'd vp
174 	 */
175 	if (uap->flags & MNT_UPDATE) {
176 		if ((vp->v_flag & VROOT) == 0) {
177 			cache_drop(ncp);
178 			vput(vp);
179 			return (EINVAL);
180 		}
181 		mp = vp->v_mount;
182 		flag = mp->mnt_flag;
183 		flag2 = mp->mnt_kern_flag;
184 		/*
185 		 * We only allow the filesystem to be reloaded if it
186 		 * is currently mounted read-only.
187 		 */
188 		if ((uap->flags & MNT_RELOAD) &&
189 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
190 			cache_drop(ncp);
191 			vput(vp);
192 			return (EOPNOTSUPP);	/* Needs translation */
193 		}
194 		/*
195 		 * Only root, or the user that did the original mount is
196 		 * permitted to update it.
197 		 */
198 		if (mp->mnt_stat.f_owner != cred->cr_uid &&
199 		    (error = suser(td))) {
200 			cache_drop(ncp);
201 			vput(vp);
202 			return (error);
203 		}
204 		if (vfs_busy(mp, LK_NOWAIT)) {
205 			cache_drop(ncp);
206 			vput(vp);
207 			return (EBUSY);
208 		}
209 		if ((vp->v_flag & VMOUNT) != 0 ||
210 		    vp->v_mountedhere != NULL) {
211 			cache_drop(ncp);
212 			vfs_unbusy(mp);
213 			vput(vp);
214 			return (EBUSY);
215 		}
216 		vp->v_flag |= VMOUNT;
217 		mp->mnt_flag |=
218 		    uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
219 		VOP_UNLOCK(vp, 0);
220 		goto update;
221 	}
222 	/*
223 	 * If the user is not root, ensure that they own the directory
224 	 * onto which we are attempting to mount.
225 	 */
226 	if ((error = VOP_GETATTR(vp, &va)) ||
227 	    (va.va_uid != cred->cr_uid && (error = suser(td)))) {
228 		cache_drop(ncp);
229 		vput(vp);
230 		return (error);
231 	}
232 	if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
233 		cache_drop(ncp);
234 		vput(vp);
235 		return (error);
236 	}
237 	if (vp->v_type != VDIR) {
238 		cache_drop(ncp);
239 		vput(vp);
240 		return (ENOTDIR);
241 	}
242 	if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
243 		cache_drop(ncp);
244 		vput(vp);
245 		return (error);
246 	}
247 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
248 		if (!strcmp(vfsp->vfc_name, fstypename))
249 			break;
250 	}
251 	if (vfsp == NULL) {
252 		linker_file_t lf;
253 
254 		/* Only load modules for root (very important!) */
255 		if ((error = suser(td)) != 0) {
256 			cache_drop(ncp);
257 			vput(vp);
258 			return error;
259 		}
260 		error = linker_load_file(fstypename, &lf);
261 		if (error || lf == NULL) {
262 			cache_drop(ncp);
263 			vput(vp);
264 			if (lf == NULL)
265 				error = ENODEV;
266 			return error;
267 		}
268 		lf->userrefs++;
269 		/* lookup again, see if the VFS was loaded */
270 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
271 			if (!strcmp(vfsp->vfc_name, fstypename))
272 				break;
273 		}
274 		if (vfsp == NULL) {
275 			lf->userrefs--;
276 			linker_file_unload(lf);
277 			cache_drop(ncp);
278 			vput(vp);
279 			return (ENODEV);
280 		}
281 	}
282 	if ((vp->v_flag & VMOUNT) != 0 ||
283 	    vp->v_mountedhere != NULL) {
284 		cache_drop(ncp);
285 		vput(vp);
286 		return (EBUSY);
287 	}
288 	vp->v_flag |= VMOUNT;
289 
290 	/*
291 	 * Allocate and initialize the filesystem.
292 	 */
293 	mp = malloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
294 	TAILQ_INIT(&mp->mnt_nvnodelist);
295 	TAILQ_INIT(&mp->mnt_reservedvnlist);
296 	TAILQ_INIT(&mp->mnt_jlist);
297 	mp->mnt_nvnodelistsize = 0;
298 	lockinit(&mp->mnt_lock, "vfslock", 0, LK_NOPAUSE);
299 	vfs_busy(mp, LK_NOWAIT);
300 	mp->mnt_op = vfsp->vfc_vfsops;
301 	mp->mnt_vfc = vfsp;
302 	vfsp->vfc_refcount++;
303 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
304 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
305 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
306 	mp->mnt_vnodecovered = vp;
307 	mp->mnt_stat.f_owner = cred->cr_uid;
308 	mp->mnt_iosize_max = DFLTPHYS;
309 	VOP_UNLOCK(vp, 0);
310 update:
311 	/*
312 	 * Set the mount level flags.
313 	 */
314 	if (uap->flags & MNT_RDONLY)
315 		mp->mnt_flag |= MNT_RDONLY;
316 	else if (mp->mnt_flag & MNT_RDONLY)
317 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
318 	mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
319 	    MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
320 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
321 	    MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
322 	mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
323 	    MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
324 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
325 	    MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
326 	/*
327 	 * Mount the filesystem.
328 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
329 	 * get.
330 	 */
331 	error = VFS_MOUNT(mp, uap->path, uap->data, cred);
332 	if (mp->mnt_flag & MNT_UPDATE) {
333 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
334 			mp->mnt_flag &= ~MNT_RDONLY;
335 		mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
336 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
337 		if (error) {
338 			mp->mnt_flag = flag;
339 			mp->mnt_kern_flag = flag2;
340 		}
341 		vfs_unbusy(mp);
342 		vp->v_flag &= ~VMOUNT;
343 		vrele(vp);
344 		cache_drop(ncp);
345 		return (error);
346 	}
347 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
348 	/*
349 	 * Put the new filesystem on the mount list after root.  The mount
350 	 * point gets its own mnt_ncp which is a special ncp linking the
351 	 * vnode-under to the root of the new mount.  The lookup code
352 	 * detects the mount point going forward and detects the special
353 	 * mnt_ncp via NCP_MOUNTPT going backwards.
354 	 *
355 	 * It is not necessary to invalidate or purge the vnode underneath
356 	 * because elements under the mount will be given their own glue
357 	 * namecache record.
358 	 */
359 	if (!error) {
360 		nlc.nlc_nameptr = "";
361 		nlc.nlc_namelen = 0;
362 		mp->mnt_ncp = cache_nlookup(ncp, &nlc);
363 		cache_setunresolved(mp->mnt_ncp);
364 		mp->mnt_ncp->nc_flag |= NCF_MOUNTPT;
365 		mp->mnt_ncp->nc_mount = mp;
366 		cache_drop(ncp);
367 		/* XXX get the root of the fs and cache_setvp(mnt_ncp...) */
368 		vp->v_flag &= ~VMOUNT;
369 		vp->v_mountedhere = mp;
370 		mountlist_insert(mp, MNTINS_LAST);
371 		checkdirs(vp, mp->mnt_ncp);
372 		cache_unlock(mp->mnt_ncp);	/* leave ref intact */
373 		VOP_UNLOCK(vp, 0);
374 		error = vfs_allocate_syncvnode(mp);
375 		vfs_unbusy(mp);
376 		if ((error = VFS_START(mp, 0)) != 0)
377 			vrele(vp);
378 	} else {
379 		vfs_rm_vnodeops(&mp->mnt_vn_coherency_ops);
380 		vfs_rm_vnodeops(&mp->mnt_vn_journal_ops);
381 		vfs_rm_vnodeops(&mp->mnt_vn_norm_ops);
382 		vfs_rm_vnodeops(&mp->mnt_vn_spec_ops);
383 		vfs_rm_vnodeops(&mp->mnt_vn_fifo_ops);
384 		vp->v_flag &= ~VMOUNT;
385 		mp->mnt_vfc->vfc_refcount--;
386 		vfs_unbusy(mp);
387 		free(mp, M_MOUNT);
388 		cache_drop(ncp);
389 		vput(vp);
390 	}
391 	return (error);
392 }
393 
394 /*
395  * Scan all active processes to see if any of them have a current
396  * or root directory onto which the new filesystem has just been
397  * mounted. If so, replace them with the new mount point.
398  *
399  * The passed ncp is ref'd and locked (from the mount code) and
400  * must be associated with the vnode representing the root of the
401  * mount point.
402  */
403 struct checkdirs_info {
404 	struct vnode *olddp;
405 	struct vnode *newdp;
406 	struct namecache *ncp;
407 };
408 
409 static int checkdirs_callback(struct proc *p, void *data);
410 
411 static void
412 checkdirs(struct vnode *olddp, struct namecache *ncp)
413 {
414 	struct checkdirs_info info;
415 	struct vnode *newdp;
416 	struct mount *mp;
417 
418 	if (olddp->v_usecount == 1)
419 		return;
420 	mp = olddp->v_mountedhere;
421 	if (VFS_ROOT(mp, &newdp))
422 		panic("mount: lost mount");
423 	cache_setvp(ncp, newdp);
424 
425 	if (rootvnode == olddp) {
426 		vref(newdp);
427 		vfs_cache_setroot(newdp, cache_hold(ncp));
428 	}
429 
430 	info.olddp = olddp;
431 	info.newdp = newdp;
432 	info.ncp = ncp;
433 	allproc_scan(checkdirs_callback, &info);
434 	vput(newdp);
435 }
436 
437 /*
438  * NOTE: callback is not MP safe because the scanned process's filedesc
439  * structure can be ripped out from under us, amoung other things.
440  */
441 static int
442 checkdirs_callback(struct proc *p, void *data)
443 {
444 	struct checkdirs_info *info = data;
445 	struct filedesc *fdp;
446 	struct namecache *ncdrop1;
447 	struct namecache *ncdrop2;
448 	struct vnode *vprele1;
449 	struct vnode *vprele2;
450 
451 	if ((fdp = p->p_fd) != NULL) {
452 		ncdrop1 = NULL;
453 		ncdrop2 = NULL;
454 		vprele1 = NULL;
455 		vprele2 = NULL;
456 
457 		/*
458 		 * MPUNSAFE - XXX fdp can be pulled out from under a
459 		 * foreign process.
460 		 *
461 		 * A shared filedesc is ok, we don't have to copy it
462 		 * because we are making this change globally.
463 		 */
464 		spin_lock_wr(&fdp->fd_spin);
465 		if (fdp->fd_cdir == info->olddp) {
466 			vprele1 = fdp->fd_cdir;
467 			vref(info->newdp);
468 			fdp->fd_cdir = info->newdp;
469 			ncdrop1 = fdp->fd_ncdir;
470 			fdp->fd_ncdir = cache_hold(info->ncp);
471 		}
472 		if (fdp->fd_rdir == info->olddp) {
473 			vprele2 = fdp->fd_rdir;
474 			vref(info->newdp);
475 			fdp->fd_rdir = info->newdp;
476 			ncdrop2 = fdp->fd_nrdir;
477 			fdp->fd_nrdir = cache_hold(info->ncp);
478 		}
479 		spin_unlock_wr(&fdp->fd_spin);
480 		if (ncdrop1)
481 			cache_drop(ncdrop1);
482 		if (ncdrop2)
483 			cache_drop(ncdrop2);
484 		if (vprele1)
485 			vrele(vprele1);
486 		if (vprele2)
487 			vrele(vprele2);
488 	}
489 	return(0);
490 }
491 
492 /*
493  * Unmount a file system.
494  *
495  * Note: unmount takes a path to the vnode mounted on as argument,
496  * not special file (as before).
497  */
498 /*
499  * umount_args(char *path, int flags)
500  */
501 /* ARGSUSED */
502 int
503 sys_unmount(struct unmount_args *uap)
504 {
505 	struct thread *td = curthread;
506 	struct proc *p = td->td_proc;
507 	struct mount *mp = NULL;
508 	int error;
509 	struct nlookupdata nd;
510 
511 	KKASSERT(p);
512 	if (p->p_ucred->cr_prison != NULL)
513 		return (EPERM);
514 	if (usermount == 0 && (error = suser(td)))
515 		return (error);
516 
517 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
518 	if (error == 0)
519 		error = nlookup(&nd);
520 	if (error)
521 		goto out;
522 
523 	mp = nd.nl_ncp->nc_mount;
524 
525 	/*
526 	 * Only root, or the user that did the original mount is
527 	 * permitted to unmount this filesystem.
528 	 */
529 	if ((mp->mnt_stat.f_owner != p->p_ucred->cr_uid) &&
530 	    (error = suser(td)))
531 		goto out;
532 
533 	/*
534 	 * Don't allow unmounting the root file system.
535 	 */
536 	if (mp->mnt_flag & MNT_ROOTFS) {
537 		error = EINVAL;
538 		goto out;
539 	}
540 
541 	/*
542 	 * Must be the root of the filesystem
543 	 */
544 	if (! (nd.nl_ncp->nc_flag & NCF_MOUNTPT)) {
545 		error = EINVAL;
546 		goto out;
547 	}
548 
549 out:
550 	nlookup_done(&nd);
551 	if (error)
552 		return (error);
553 	return (dounmount(mp, uap->flags));
554 }
555 
556 /*
557  * Do the actual file system unmount.
558  */
559 static int
560 dounmount_interlock(struct mount *mp)
561 {
562 	if (mp->mnt_kern_flag & MNTK_UNMOUNT)
563 		return (EBUSY);
564 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
565 	return(0);
566 }
567 
568 int
569 dounmount(struct mount *mp, int flags)
570 {
571 	struct vnode *coveredvp;
572 	int error;
573 	int async_flag;
574 	int lflags;
575 
576 	/*
577 	 * Exclusive access for unmounting purposes
578 	 */
579 	if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
580 		return (error);
581 
582 	/*
583 	 * Allow filesystems to detect that a forced unmount is in progress.
584 	 */
585 	if (flags & MNT_FORCE)
586 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
587 	lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
588 	error = lockmgr(&mp->mnt_lock, lflags);
589 	if (error) {
590 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
591 		if (mp->mnt_kern_flag & MNTK_MWAIT)
592 			wakeup(mp);
593 		return (error);
594 	}
595 
596 	if (mp->mnt_flag & MNT_EXPUBLIC)
597 		vfs_setpublicfs(NULL, NULL, NULL);
598 
599 	vfs_msync(mp, MNT_WAIT);
600 	async_flag = mp->mnt_flag & MNT_ASYNC;
601 	mp->mnt_flag &=~ MNT_ASYNC;
602 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
603 	if (mp->mnt_syncer != NULL)
604 		vrele(mp->mnt_syncer);
605 	if (((mp->mnt_flag & MNT_RDONLY) ||
606 	     (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
607 	    (flags & MNT_FORCE))
608 		error = VFS_UNMOUNT(mp, flags);
609 	if (error) {
610 		if (mp->mnt_syncer == NULL)
611 			vfs_allocate_syncvnode(mp);
612 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
613 		mp->mnt_flag |= async_flag;
614 		lockmgr(&mp->mnt_lock, LK_RELEASE);
615 		if (mp->mnt_kern_flag & MNTK_MWAIT)
616 			wakeup(mp);
617 		return (error);
618 	}
619 	/*
620 	 * Clean up any journals still associated with the mount after
621 	 * filesystem activity has ceased.
622 	 */
623 	journal_remove_all_journals(mp,
624 	    ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
625 
626 	mountlist_remove(mp);
627 
628 	/*
629 	 * Remove any installed vnode ops here so the individual VFSs don't
630 	 * have to.
631 	 */
632 	vfs_rm_vnodeops(&mp->mnt_vn_coherency_ops);
633 	vfs_rm_vnodeops(&mp->mnt_vn_journal_ops);
634 	vfs_rm_vnodeops(&mp->mnt_vn_norm_ops);
635 	vfs_rm_vnodeops(&mp->mnt_vn_spec_ops);
636 	vfs_rm_vnodeops(&mp->mnt_vn_fifo_ops);
637 
638 	if ((coveredvp = mp->mnt_vnodecovered) != NULLVP) {
639 		coveredvp->v_mountedhere = NULL;
640 		vrele(coveredvp);
641 		cache_drop(mp->mnt_ncp);
642 		mp->mnt_ncp = NULL;
643 	}
644 	mp->mnt_vfc->vfc_refcount--;
645 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
646 		panic("unmount: dangling vnode");
647 	lockmgr(&mp->mnt_lock, LK_RELEASE);
648 	if (mp->mnt_kern_flag & MNTK_MWAIT)
649 		wakeup(mp);
650 	free(mp, M_MOUNT);
651 	return (0);
652 }
653 
654 /*
655  * Sync each mounted filesystem.
656  */
657 
658 #ifdef DEBUG
659 static int syncprt = 0;
660 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
661 #endif /* DEBUG */
662 
663 static int sync_callback(struct mount *mp, void *data);
664 
665 /* ARGSUSED */
666 int
667 sys_sync(struct sync_args *uap)
668 {
669 	mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
670 #ifdef DEBUG
671 	/*
672 	 * print out buffer pool stat information on each sync() call.
673 	 */
674 	if (syncprt)
675 		vfs_bufstats();
676 #endif /* DEBUG */
677 	return (0);
678 }
679 
680 static
681 int
682 sync_callback(struct mount *mp, void *data __unused)
683 {
684 	int asyncflag;
685 
686 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
687 		asyncflag = mp->mnt_flag & MNT_ASYNC;
688 		mp->mnt_flag &= ~MNT_ASYNC;
689 		vfs_msync(mp, MNT_NOWAIT);
690 		VFS_SYNC(mp, MNT_NOWAIT);
691 		mp->mnt_flag |= asyncflag;
692 	}
693 	return(0);
694 }
695 
696 /* XXX PRISON: could be per prison flag */
697 static int prison_quotas;
698 #if 0
699 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
700 #endif
701 
702 /*
703  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
704  *
705  * Change filesystem quotas.
706  */
707 /* ARGSUSED */
708 int
709 sys_quotactl(struct quotactl_args *uap)
710 {
711 	struct nlookupdata nd;
712 	struct thread *td;
713 	struct proc *p;
714 	struct mount *mp;
715 	int error;
716 
717 	td = curthread;
718 	p = td->td_proc;
719 	if (p->p_ucred->cr_prison && !prison_quotas)
720 		return (EPERM);
721 
722 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
723 	if (error == 0)
724 		error = nlookup(&nd);
725 	if (error == 0) {
726 		mp = nd.nl_ncp->nc_mount;
727 		error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
728 				    uap->arg, nd.nl_cred);
729 	}
730 	nlookup_done(&nd);
731 	return (error);
732 }
733 
734 /*
735  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
736  *		void *buf, int buflen)
737  *
738  * This function operates on a mount point and executes the specified
739  * operation using the specified control data, and possibly returns data.
740  *
741  * The actual number of bytes stored in the result buffer is returned, 0
742  * if none, otherwise an error is returned.
743  */
744 /* ARGSUSED */
745 int
746 sys_mountctl(struct mountctl_args *uap)
747 {
748 	struct thread *td = curthread;
749 	struct proc *p = td->td_proc;
750 	struct file *fp;
751 	void *ctl = NULL;
752 	void *buf = NULL;
753 	char *path = NULL;
754 	int error;
755 
756 	/*
757 	 * Sanity and permissions checks.  We must be root.
758 	 */
759 	KKASSERT(p);
760 	if (p->p_ucred->cr_prison != NULL)
761 		return (EPERM);
762 	if ((error = suser(td)) != 0)
763 		return (error);
764 
765 	/*
766 	 * Argument length checks
767 	 */
768 	if (uap->ctllen < 0 || uap->ctllen > 1024)
769 		return (EINVAL);
770 	if (uap->buflen < 0 || uap->buflen > 16 * 1024)
771 		return (EINVAL);
772 	if (uap->path == NULL)
773 		return (EINVAL);
774 
775 	/*
776 	 * Allocate the necessary buffers and copyin data
777 	 */
778 	path = objcache_get(namei_oc, M_WAITOK);
779 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
780 	if (error)
781 		goto done;
782 
783 	if (uap->ctllen) {
784 		ctl = malloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
785 		error = copyin(uap->ctl, ctl, uap->ctllen);
786 		if (error)
787 			goto done;
788 	}
789 	if (uap->buflen)
790 		buf = malloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
791 
792 	/*
793 	 * Validate the descriptor
794 	 */
795 	fp = holdfp(p->p_fd, uap->fd, -1);
796 	if (fp == NULL) {
797 		error = EBADF;
798 		goto done;
799 	}
800 
801 	/*
802 	 * Execute the internal kernel function and clean up.
803 	 */
804 	error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
805 	if (fp)
806 		fdrop(fp);
807 	if (error == 0 && uap->sysmsg_result > 0)
808 		error = copyout(buf, uap->buf, uap->sysmsg_result);
809 done:
810 	if (path)
811 		objcache_put(namei_oc, path);
812 	if (ctl)
813 		free(ctl, M_TEMP);
814 	if (buf)
815 		free(buf, M_TEMP);
816 	return (error);
817 }
818 
819 /*
820  * Execute a mount control operation by resolving the path to a mount point
821  * and calling vop_mountctl().
822  */
823 int
824 kern_mountctl(const char *path, int op, struct file *fp,
825 		const void *ctl, int ctllen,
826 		void *buf, int buflen, int *res)
827 {
828 	struct vnode *vp;
829 	struct mount *mp;
830 	struct nlookupdata nd;
831 	int error;
832 
833 	*res = 0;
834 	vp = NULL;
835 	error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
836 	if (error == 0)
837 		error = nlookup(&nd);
838 	if (error == 0)
839 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
840 	nlookup_done(&nd);
841 	if (error)
842 		return (error);
843 
844 	mp = vp->v_mount;
845 
846 	/*
847 	 * Must be the root of the filesystem
848 	 */
849 	if ((vp->v_flag & VROOT) == 0) {
850 		vput(vp);
851 		return (EINVAL);
852 	}
853 	error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen,
854 				buf, buflen, res);
855 	vput(vp);
856 	return (error);
857 }
858 
859 int
860 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
861 {
862 	struct thread *td = curthread;
863 	struct proc *p = td->td_proc;
864 	struct mount *mp;
865 	struct statfs *sp;
866 	char *fullpath, *freepath;
867 	int error;
868 
869 	if ((error = nlookup(nd)) != 0)
870 		return (error);
871 	mp = nd->nl_ncp->nc_mount;
872 	sp = &mp->mnt_stat;
873 	if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
874 		return (error);
875 
876 	error = cache_fullpath(p, mp->mnt_ncp, &fullpath, &freepath);
877 	if (error)
878 		return(error);
879 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
880 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
881 	free(freepath, M_TEMP);
882 
883 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
884 	bcopy(sp, buf, sizeof(*buf));
885 	/* Only root should have access to the fsid's. */
886 	if (suser(td))
887 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
888 	return (0);
889 }
890 
891 /*
892  * statfs_args(char *path, struct statfs *buf)
893  *
894  * Get filesystem statistics.
895  */
896 int
897 sys_statfs(struct statfs_args *uap)
898 {
899 	struct nlookupdata nd;
900 	struct statfs buf;
901 	int error;
902 
903 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
904 	if (error == 0)
905 		error = kern_statfs(&nd, &buf);
906 	nlookup_done(&nd);
907 	if (error == 0)
908 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
909 	return (error);
910 }
911 
912 int
913 kern_fstatfs(int fd, struct statfs *buf)
914 {
915 	struct thread *td = curthread;
916 	struct proc *p = td->td_proc;
917 	struct file *fp;
918 	struct mount *mp;
919 	struct statfs *sp;
920 	char *fullpath, *freepath;
921 	int error;
922 
923 	KKASSERT(p);
924 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
925 		return (error);
926 	mp = ((struct vnode *)fp->f_data)->v_mount;
927 	if (mp == NULL) {
928 		error = EBADF;
929 		goto done;
930 	}
931 	if (fp->f_cred == NULL) {
932 		error = EINVAL;
933 		goto done;
934 	}
935 	sp = &mp->mnt_stat;
936 	if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
937 		goto done;
938 
939 	if ((error = cache_fullpath(p, mp->mnt_ncp, &fullpath, &freepath)) != 0)
940 		goto done;
941 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
942 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
943 	free(freepath, M_TEMP);
944 
945 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
946 	bcopy(sp, buf, sizeof(*buf));
947 
948 	/* Only root should have access to the fsid's. */
949 	if (suser(td))
950 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
951 	error = 0;
952 done:
953 	fdrop(fp);
954 	return (error);
955 }
956 
957 /*
958  * fstatfs_args(int fd, struct statfs *buf)
959  *
960  * Get filesystem statistics.
961  */
962 int
963 sys_fstatfs(struct fstatfs_args *uap)
964 {
965 	struct statfs buf;
966 	int error;
967 
968 	error = kern_fstatfs(uap->fd, &buf);
969 
970 	if (error == 0)
971 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
972 	return (error);
973 }
974 
975 /*
976  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
977  *
978  * Get statistics on all filesystems.
979  */
980 
981 struct getfsstat_info {
982 	struct statfs *sfsp;
983 	long count;
984 	long maxcount;
985 	int error;
986 	int flags;
987 	int is_chrooted;
988 	struct proc *p;
989 };
990 
991 static int getfsstat_callback(struct mount *, void *);
992 
993 /* ARGSUSED */
994 int
995 sys_getfsstat(struct getfsstat_args *uap)
996 {
997 	struct thread *td = curthread;
998 	struct proc *p = td->td_proc;
999 	struct getfsstat_info info;
1000 
1001 	bzero(&info, sizeof(info));
1002 	if (p != NULL && (p->p_fd->fd_nrdir->nc_flag & NCF_ROOT) == 0)
1003 		info.is_chrooted = 1;
1004 	else
1005 		info.is_chrooted = 0;
1006 
1007 	info.maxcount = uap->bufsize / sizeof(struct statfs);
1008 	info.sfsp = uap->buf;
1009 	info.count = 0;
1010 	info.flags = uap->flags;
1011 	info.p = p;
1012 
1013 	mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1014 	if (info.sfsp && info.count > info.maxcount)
1015 		uap->sysmsg_result = info.maxcount;
1016 	else
1017 		uap->sysmsg_result = info.count;
1018 	return (info.error);
1019 }
1020 
1021 static int
1022 getfsstat_callback(struct mount *mp, void *data)
1023 {
1024 	struct getfsstat_info *info = data;
1025 	struct statfs *sp;
1026 	char *freepath;
1027 	char *fullpath;
1028 	int error;
1029 
1030 	if (info->sfsp && info->count < info->maxcount) {
1031 		if (info->is_chrooted && !chroot_visible_mnt(mp, info->p))
1032 			return(0);
1033 		sp = &mp->mnt_stat;
1034 
1035 		/*
1036 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1037 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1038 		 * overrides MNT_WAIT.
1039 		 */
1040 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1041 		    (info->flags & MNT_WAIT)) &&
1042 		    (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1043 			return(0);
1044 		}
1045 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1046 
1047 		error = cache_fullpath(info->p, mp->mnt_ncp,
1048 					&fullpath, &freepath);
1049 		if (error) {
1050 			info->error = error;
1051 			return(-1);
1052 		}
1053 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1054 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1055 		free(freepath, M_TEMP);
1056 
1057 		error = copyout(sp, info->sfsp, sizeof(*sp));
1058 		if (error) {
1059 			info->error = error;
1060 			return (-1);
1061 		}
1062 		++info->sfsp;
1063 	}
1064 	info->count++;
1065 	return(0);
1066 }
1067 
1068 /*
1069  * fchdir_args(int fd)
1070  *
1071  * Change current working directory to a given file descriptor.
1072  */
1073 /* ARGSUSED */
1074 int
1075 sys_fchdir(struct fchdir_args *uap)
1076 {
1077 	struct thread *td = curthread;
1078 	struct proc *p = td->td_proc;
1079 	struct filedesc *fdp = p->p_fd;
1080 	struct vnode *vp, *ovp;
1081 	struct mount *mp;
1082 	struct file *fp;
1083 	struct namecache *ncp, *oncp;
1084 	struct namecache *nct;
1085 	int error;
1086 
1087 	if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1088 		return (error);
1089 	vp = (struct vnode *)fp->f_data;
1090 	vref(vp);
1091 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1092 	if (vp->v_type != VDIR || fp->f_ncp == NULL)
1093 		error = ENOTDIR;
1094 	else
1095 		error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
1096 	if (error) {
1097 		vput(vp);
1098 		fdrop(fp);
1099 		return (error);
1100 	}
1101 	ncp = cache_hold(fp->f_ncp);
1102 	while (!error && (mp = vp->v_mountedhere) != NULL) {
1103 		error = nlookup_mp(mp, &nct);
1104 		if (error == 0) {
1105 			cache_unlock(nct);	/* leave ref intact */
1106 			vput(vp);
1107 			vp = nct->nc_vp;
1108 			error = vget(vp, LK_SHARED);
1109 			KKASSERT(error == 0);
1110 			cache_drop(ncp);
1111 			ncp = nct;
1112 		}
1113 	}
1114 	if (error == 0) {
1115 		ovp = fdp->fd_cdir;
1116 		oncp = fdp->fd_ncdir;
1117 		VOP_UNLOCK(vp, 0);	/* leave ref intact */
1118 		fdp->fd_cdir = vp;
1119 		fdp->fd_ncdir = ncp;
1120 		cache_drop(oncp);
1121 		vrele(ovp);
1122 	} else {
1123 		cache_drop(ncp);
1124 		vput(vp);
1125 	}
1126 	fdrop(fp);
1127 	return (error);
1128 }
1129 
1130 int
1131 kern_chdir(struct nlookupdata *nd)
1132 {
1133 	struct thread *td = curthread;
1134 	struct proc *p = td->td_proc;
1135 	struct filedesc *fdp = p->p_fd;
1136 	struct vnode *vp, *ovp;
1137 	struct namecache *oncp;
1138 	int error;
1139 
1140 	if ((error = nlookup(nd)) != 0)
1141 		return (error);
1142 	if ((vp = nd->nl_ncp->nc_vp) == NULL)
1143 		return (ENOENT);
1144 	if ((error = vget(vp, LK_SHARED)) != 0)
1145 		return (error);
1146 
1147 	error = checkvp_chdir(vp, td);
1148 	VOP_UNLOCK(vp, 0);
1149 	if (error == 0) {
1150 		ovp = fdp->fd_cdir;
1151 		oncp = fdp->fd_ncdir;
1152 		cache_unlock(nd->nl_ncp);	/* leave reference intact */
1153 		fdp->fd_ncdir = nd->nl_ncp;
1154 		fdp->fd_cdir = vp;
1155 		cache_drop(oncp);
1156 		vrele(ovp);
1157 		nd->nl_ncp = NULL;
1158 	} else {
1159 		vrele(vp);
1160 	}
1161 	return (error);
1162 }
1163 
1164 /*
1165  * chdir_args(char *path)
1166  *
1167  * Change current working directory (``.'').
1168  */
1169 int
1170 sys_chdir(struct chdir_args *uap)
1171 {
1172 	struct nlookupdata nd;
1173 	int error;
1174 
1175 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1176 	if (error == 0)
1177 		error = kern_chdir(&nd);
1178 	nlookup_done(&nd);
1179 	return (error);
1180 }
1181 
1182 /*
1183  * Helper function for raised chroot(2) security function:  Refuse if
1184  * any filedescriptors are open directories.
1185  */
1186 static int
1187 chroot_refuse_vdir_fds(fdp)
1188 	struct filedesc *fdp;
1189 {
1190 	struct vnode *vp;
1191 	struct file *fp;
1192 	int error;
1193 	int fd;
1194 
1195 	for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1196 		if ((error = holdvnode(fdp, fd, &fp)) != 0)
1197 			continue;
1198 		vp = (struct vnode *)fp->f_data;
1199 		if (vp->v_type != VDIR) {
1200 			fdrop(fp);
1201 			continue;
1202 		}
1203 		fdrop(fp);
1204 		return(EPERM);
1205 	}
1206 	return (0);
1207 }
1208 
1209 /*
1210  * This sysctl determines if we will allow a process to chroot(2) if it
1211  * has a directory open:
1212  *	0: disallowed for all processes.
1213  *	1: allowed for processes that were not already chroot(2)'ed.
1214  *	2: allowed for all processes.
1215  */
1216 
1217 static int chroot_allow_open_directories = 1;
1218 
1219 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1220      &chroot_allow_open_directories, 0, "");
1221 
1222 /*
1223  * chroot to the specified namecache entry.  We obtain the vp from the
1224  * namecache data.  The passed ncp must be locked and referenced and will
1225  * remain locked and referenced on return.
1226  */
1227 int
1228 kern_chroot(struct namecache *ncp)
1229 {
1230 	struct thread *td = curthread;
1231 	struct proc *p = td->td_proc;
1232 	struct filedesc *fdp = p->p_fd;
1233 	struct vnode *vp;
1234 	int error;
1235 
1236 	/*
1237 	 * Only root can chroot
1238 	 */
1239 	if ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0)
1240 		return (error);
1241 
1242 	/*
1243 	 * Disallow open directory descriptors (fchdir() breakouts).
1244 	 */
1245 	if (chroot_allow_open_directories == 0 ||
1246 	   (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1247 		if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1248 			return (error);
1249 	}
1250 	if ((vp = ncp->nc_vp) == NULL)
1251 		return (ENOENT);
1252 
1253 	if ((error = vget(vp, LK_SHARED)) != 0)
1254 		return (error);
1255 
1256 	/*
1257 	 * Check the validity of vp as a directory to change to and
1258 	 * associate it with rdir/jdir.
1259 	 */
1260 	error = checkvp_chdir(vp, td);
1261 	VOP_UNLOCK(vp, 0);	/* leave reference intact */
1262 	if (error == 0) {
1263 		vrele(fdp->fd_rdir);
1264 		fdp->fd_rdir = vp;	/* reference inherited by fd_rdir */
1265 		cache_drop(fdp->fd_nrdir);
1266 		fdp->fd_nrdir = cache_hold(ncp);
1267 		if (fdp->fd_jdir == NULL) {
1268 			fdp->fd_jdir = vp;
1269 			vref(fdp->fd_jdir);
1270 			fdp->fd_njdir = cache_hold(ncp);
1271 		}
1272 	} else {
1273 		vrele(vp);
1274 	}
1275 	return (error);
1276 }
1277 
1278 /*
1279  * chroot_args(char *path)
1280  *
1281  * Change notion of root (``/'') directory.
1282  */
1283 /* ARGSUSED */
1284 int
1285 sys_chroot(struct chroot_args *uap)
1286 {
1287 	struct thread *td = curthread;
1288 	struct nlookupdata nd;
1289 	int error;
1290 
1291 	KKASSERT(td->td_proc);
1292 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1293 	if (error) {
1294 		nlookup_done(&nd);
1295 		return(error);
1296 	}
1297 	error = nlookup(&nd);
1298 	if (error == 0)
1299 		error = kern_chroot(nd.nl_ncp);
1300 	nlookup_done(&nd);
1301 	return(error);
1302 }
1303 
1304 /*
1305  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1306  * determine whether it is legal to chdir to the vnode.  The vnode's state
1307  * is not changed by this call.
1308  */
1309 int
1310 checkvp_chdir(struct vnode *vp, struct thread *td)
1311 {
1312 	int error;
1313 
1314 	if (vp->v_type != VDIR)
1315 		error = ENOTDIR;
1316 	else
1317 		error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred);
1318 	return (error);
1319 }
1320 
1321 int
1322 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1323 {
1324 	struct thread *td = curthread;
1325 	struct proc *p = td->td_proc;
1326 	struct lwp *lp = td->td_lwp;
1327 	struct filedesc *fdp = p->p_fd;
1328 	int cmode, flags;
1329 	struct file *nfp;
1330 	struct file *fp;
1331 	struct vnode *vp;
1332 	int type, indx, error;
1333 	struct flock lf;
1334 
1335 	if ((oflags & O_ACCMODE) == O_ACCMODE)
1336 		return (EINVAL);
1337 	flags = FFLAGS(oflags);
1338 	error = falloc(p, &nfp, NULL);
1339 	if (error)
1340 		return (error);
1341 	fp = nfp;
1342 	cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) &~ S_ISTXT;
1343 
1344 	/*
1345 	 * XXX p_dupfd is a real mess.  It allows a device to return a
1346 	 * file descriptor to be duplicated rather then doing the open
1347 	 * itself.
1348 	 */
1349 	lp->lwp_dupfd = -1;
1350 
1351 	/*
1352 	 * Call vn_open() to do the lookup and assign the vnode to the
1353 	 * file pointer.  vn_open() does not change the ref count on fp
1354 	 * and the vnode, on success, will be inherited by the file pointer
1355 	 * and unlocked.
1356 	 */
1357 	nd->nl_flags |= NLC_LOCKVP;
1358 	error = vn_open(nd, fp, flags, cmode);
1359 	nlookup_done(nd);
1360 	if (error) {
1361 		/*
1362 		 * handle special fdopen() case.  bleh.  dupfdopen() is
1363 		 * responsible for dropping the old contents of ofiles[indx]
1364 		 * if it succeeds.
1365 		 *
1366 		 * Note that fsetfd() will add a ref to fp which represents
1367 		 * the fd_files[] assignment.  We must still drop our
1368 		 * reference.
1369 		 */
1370 		if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1371 			if (fdalloc(p, 0, &indx) == 0) {
1372 				error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error);
1373 				if (error == 0) {
1374 					*res = indx;
1375 					fdrop(fp);	/* our ref */
1376 					return (0);
1377 				}
1378 				fsetfd(p, NULL, indx);
1379 			}
1380 		}
1381 		fdrop(fp);	/* our ref */
1382 		if (error == ERESTART)
1383 			error = EINTR;
1384 		return (error);
1385 	}
1386 
1387 	/*
1388 	 * ref the vnode for ourselves so it can't be ripped out from under
1389 	 * is.  XXX need an ND flag to request that the vnode be returned
1390 	 * anyway.
1391 	 *
1392 	 * Reserve a file descriptor but do not assign it until the open
1393 	 * succeeds.
1394 	 */
1395 	vp = (struct vnode *)fp->f_data;
1396 	vref(vp);
1397 	if ((error = fdalloc(p, 0, &indx)) != 0) {
1398 		fdrop(fp);
1399 		vrele(vp);
1400 		return (error);
1401 	}
1402 
1403 	/*
1404 	 * If no error occurs the vp will have been assigned to the file
1405 	 * pointer.
1406 	 */
1407 	lp->lwp_dupfd = 0;
1408 
1409 	if (flags & (O_EXLOCK | O_SHLOCK)) {
1410 		lf.l_whence = SEEK_SET;
1411 		lf.l_start = 0;
1412 		lf.l_len = 0;
1413 		if (flags & O_EXLOCK)
1414 			lf.l_type = F_WRLCK;
1415 		else
1416 			lf.l_type = F_RDLCK;
1417 		if (flags & FNONBLOCK)
1418 			type = 0;
1419 		else
1420 			type = F_WAIT;
1421 
1422 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1423 			/*
1424 			 * lock request failed.  Clean up the reserved
1425 			 * descriptor.
1426 			 */
1427 			vrele(vp);
1428 			fsetfd(p, NULL, indx);
1429 			fdrop(fp);
1430 			return (error);
1431 		}
1432 		fp->f_flag |= FHASLOCK;
1433 	}
1434 #if 0
1435 	/*
1436 	 * Assert that all regular file vnodes were created with a object.
1437 	 */
1438 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1439 		("open: regular file has no backing object after vn_open"));
1440 #endif
1441 
1442 	vrele(vp);
1443 
1444 	/*
1445 	 * release our private reference, leaving the one associated with the
1446 	 * descriptor table intact.
1447 	 */
1448 	fsetfd(p, fp, indx);
1449 	fdrop(fp);
1450 	*res = indx;
1451 	return (0);
1452 }
1453 
1454 /*
1455  * open_args(char *path, int flags, int mode)
1456  *
1457  * Check permissions, allocate an open file structure,
1458  * and call the device open routine if any.
1459  */
1460 int
1461 sys_open(struct open_args *uap)
1462 {
1463 	struct nlookupdata nd;
1464 	int error;
1465 
1466 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1467 	if (error == 0) {
1468 		error = kern_open(&nd, uap->flags,
1469 				    uap->mode, &uap->sysmsg_result);
1470 	}
1471 	nlookup_done(&nd);
1472 	return (error);
1473 }
1474 
1475 int
1476 kern_mknod(struct nlookupdata *nd, int mode, int dev)
1477 {
1478 	struct namecache *ncp;
1479 	struct thread *td = curthread;
1480 	struct proc *p = td->td_proc;
1481 	struct vnode *vp;
1482 	struct vattr vattr;
1483 	int error;
1484 	int whiteout = 0;
1485 
1486 	KKASSERT(p);
1487 
1488 	switch (mode & S_IFMT) {
1489 	case S_IFCHR:
1490 	case S_IFBLK:
1491 		error = suser(td);
1492 		break;
1493 	default:
1494 		error = suser_cred(p->p_ucred, PRISON_ROOT);
1495 		break;
1496 	}
1497 	if (error)
1498 		return (error);
1499 
1500 	bwillwrite();
1501 	nd->nl_flags |= NLC_CREATE;
1502 	if ((error = nlookup(nd)) != 0)
1503 		return (error);
1504 	ncp = nd->nl_ncp;
1505 	if (ncp->nc_vp)
1506 		return (EEXIST);
1507 
1508 	VATTR_NULL(&vattr);
1509 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1510 	vattr.va_rdev = dev;
1511 	whiteout = 0;
1512 
1513 	switch (mode & S_IFMT) {
1514 	case S_IFMT:	/* used by badsect to flag bad sectors */
1515 		vattr.va_type = VBAD;
1516 		break;
1517 	case S_IFCHR:
1518 		vattr.va_type = VCHR;
1519 		break;
1520 	case S_IFBLK:
1521 		vattr.va_type = VBLK;
1522 		break;
1523 	case S_IFWHT:
1524 		whiteout = 1;
1525 		break;
1526 	default:
1527 		error = EINVAL;
1528 		break;
1529 	}
1530 	if (error == 0) {
1531 		if (whiteout) {
1532 			error = VOP_NWHITEOUT(ncp, nd->nl_cred, NAMEI_CREATE);
1533 		} else {
1534 			vp = NULL;
1535 			error = VOP_NMKNOD(ncp, &vp, nd->nl_cred, &vattr);
1536 			if (error == 0)
1537 				vput(vp);
1538 		}
1539 	}
1540 	return (error);
1541 }
1542 
1543 /*
1544  * mknod_args(char *path, int mode, int dev)
1545  *
1546  * Create a special file.
1547  */
1548 int
1549 sys_mknod(struct mknod_args *uap)
1550 {
1551 	struct nlookupdata nd;
1552 	int error;
1553 
1554 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1555 	if (error == 0)
1556 		error = kern_mknod(&nd, uap->mode, uap->dev);
1557 	nlookup_done(&nd);
1558 	return (error);
1559 }
1560 
1561 int
1562 kern_mkfifo(struct nlookupdata *nd, int mode)
1563 {
1564 	struct namecache *ncp;
1565 	struct thread *td = curthread;
1566 	struct proc *p = td->td_proc;
1567 	struct vattr vattr;
1568 	struct vnode *vp;
1569 	int error;
1570 
1571 	bwillwrite();
1572 
1573 	nd->nl_flags |= NLC_CREATE;
1574 	if ((error = nlookup(nd)) != 0)
1575 		return (error);
1576 	ncp = nd->nl_ncp;
1577 	if (ncp->nc_vp)
1578 		return (EEXIST);
1579 
1580 	VATTR_NULL(&vattr);
1581 	vattr.va_type = VFIFO;
1582 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1583 	vp = NULL;
1584 	error = VOP_NMKNOD(ncp, &vp, nd->nl_cred, &vattr);
1585 	if (error == 0)
1586 		vput(vp);
1587 	return (error);
1588 }
1589 
1590 /*
1591  * mkfifo_args(char *path, int mode)
1592  *
1593  * Create a named pipe.
1594  */
1595 int
1596 sys_mkfifo(struct mkfifo_args *uap)
1597 {
1598 	struct nlookupdata nd;
1599 	int error;
1600 
1601 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1602 	if (error == 0)
1603 		error = kern_mkfifo(&nd, uap->mode);
1604 	nlookup_done(&nd);
1605 	return (error);
1606 }
1607 
1608 static int hardlink_check_uid = 0;
1609 SYSCTL_INT(_kern, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1610     &hardlink_check_uid, 0,
1611     "Unprivileged processes cannot create hard links to files owned by other "
1612     "users");
1613 static int hardlink_check_gid = 0;
1614 SYSCTL_INT(_kern, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
1615     &hardlink_check_gid, 0,
1616     "Unprivileged processes cannot create hard links to files owned by other "
1617     "groups");
1618 
1619 static int
1620 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
1621 {
1622 	struct vattr va;
1623 	int error;
1624 
1625 	/*
1626 	 * Shortcut if disabled
1627 	 */
1628 	if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
1629 		return (0);
1630 
1631 	/*
1632 	 * root cred can always hardlink
1633 	 */
1634 	if (suser_cred(cred, PRISON_ROOT) == 0)
1635 		return (0);
1636 
1637 	/*
1638 	 * Otherwise only if the originating file is owned by the
1639 	 * same user or group.  Note that any group is allowed if
1640 	 * the file is owned by the caller.
1641 	 */
1642 	error = VOP_GETATTR(vp, &va);
1643 	if (error != 0)
1644 		return (error);
1645 
1646 	if (hardlink_check_uid) {
1647 		if (cred->cr_uid != va.va_uid)
1648 			return (EPERM);
1649 	}
1650 
1651 	if (hardlink_check_gid) {
1652 		if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
1653 			return (EPERM);
1654 	}
1655 
1656 	return (0);
1657 }
1658 
1659 int
1660 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
1661 {
1662 	struct thread *td = curthread;
1663 	struct vnode *vp;
1664 	int error;
1665 
1666 	/*
1667 	 * Lookup the source and obtained a locked vnode.
1668 	 *
1669 	 * XXX relookup on vget failure / race ?
1670 	 */
1671 	bwillwrite();
1672 	if ((error = nlookup(nd)) != 0)
1673 		return (error);
1674 	vp = nd->nl_ncp->nc_vp;
1675 	KKASSERT(vp != NULL);
1676 	if (vp->v_type == VDIR)
1677 		return (EPERM);		/* POSIX */
1678 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
1679 		return (error);
1680 
1681 	/*
1682 	 * Unlock the source so we can lookup the target without deadlocking
1683 	 * (XXX vp is locked already, possible other deadlock?).  The target
1684 	 * must not exist.
1685 	 */
1686 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
1687 	nd->nl_flags &= ~NLC_NCPISLOCKED;
1688 	cache_unlock(nd->nl_ncp);
1689 
1690 	linknd->nl_flags |= NLC_CREATE;
1691 	if ((error = nlookup(linknd)) != 0) {
1692 		vput(vp);
1693 		return (error);
1694 	}
1695 	if (linknd->nl_ncp->nc_vp) {
1696 		vput(vp);
1697 		return (EEXIST);
1698 	}
1699 
1700 	/*
1701 	 * Finally run the new API VOP.
1702 	 */
1703 	error = can_hardlink(vp, td, td->td_proc->p_ucred);
1704 	if (error == 0)
1705 		error = VOP_NLINK(linknd->nl_ncp, vp, linknd->nl_cred);
1706 	vput(vp);
1707 	return (error);
1708 }
1709 
1710 /*
1711  * link_args(char *path, char *link)
1712  *
1713  * Make a hard file link.
1714  */
1715 int
1716 sys_link(struct link_args *uap)
1717 {
1718 	struct nlookupdata nd, linknd;
1719 	int error;
1720 
1721 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1722 	if (error == 0) {
1723 		error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
1724 		if (error == 0)
1725 			error = kern_link(&nd, &linknd);
1726 		nlookup_done(&linknd);
1727 	}
1728 	nlookup_done(&nd);
1729 	return (error);
1730 }
1731 
1732 int
1733 kern_symlink(struct nlookupdata *nd, char *path, int mode)
1734 {
1735 	struct namecache *ncp;
1736 	struct vattr vattr;
1737 	struct vnode *vp;
1738 	int error;
1739 
1740 	bwillwrite();
1741 	nd->nl_flags |= NLC_CREATE;
1742 	if ((error = nlookup(nd)) != 0)
1743 		return (error);
1744 	ncp = nd->nl_ncp;
1745 	if (ncp->nc_vp)
1746 		return (EEXIST);
1747 
1748 	VATTR_NULL(&vattr);
1749 	vattr.va_mode = mode;
1750 	error = VOP_NSYMLINK(ncp, &vp, nd->nl_cred, &vattr, path);
1751 	if (error == 0)
1752 		vput(vp);
1753 	return (error);
1754 }
1755 
1756 /*
1757  * symlink(char *path, char *link)
1758  *
1759  * Make a symbolic link.
1760  */
1761 int
1762 sys_symlink(struct symlink_args *uap)
1763 {
1764 	struct thread *td = curthread;
1765 	struct nlookupdata nd;
1766 	char *path;
1767 	int error;
1768 	int mode;
1769 
1770 	path = objcache_get(namei_oc, M_WAITOK);
1771 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
1772 	if (error == 0) {
1773 		error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
1774 		if (error == 0) {
1775 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
1776 			error = kern_symlink(&nd, path, mode);
1777 		}
1778 		nlookup_done(&nd);
1779 	}
1780 	objcache_put(namei_oc, path);
1781 	return (error);
1782 }
1783 
1784 /*
1785  * undelete_args(char *path)
1786  *
1787  * Delete a whiteout from the filesystem.
1788  */
1789 /* ARGSUSED */
1790 int
1791 sys_undelete(struct undelete_args *uap)
1792 {
1793 	struct nlookupdata nd;
1794 	int error;
1795 
1796 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1797 	bwillwrite();
1798 	nd.nl_flags |= NLC_DELETE;
1799 	if (error == 0)
1800 		error = nlookup(&nd);
1801 	if (error == 0)
1802 		error = VOP_NWHITEOUT(nd.nl_ncp, nd.nl_cred, NAMEI_DELETE);
1803 	nlookup_done(&nd);
1804 	return (error);
1805 }
1806 
1807 int
1808 kern_unlink(struct nlookupdata *nd)
1809 {
1810 	struct namecache *ncp;
1811 	int error;
1812 
1813 	bwillwrite();
1814 	nd->nl_flags |= NLC_DELETE;
1815 	if ((error = nlookup(nd)) != 0)
1816 		return (error);
1817 	ncp = nd->nl_ncp;
1818 	error = VOP_NREMOVE(ncp, nd->nl_cred);
1819 	return (error);
1820 }
1821 
1822 /*
1823  * unlink_args(char *path)
1824  *
1825  * Delete a name from the filesystem.
1826  */
1827 int
1828 sys_unlink(struct unlink_args *uap)
1829 {
1830 	struct nlookupdata nd;
1831 	int error;
1832 
1833 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1834 	if (error == 0)
1835 		error = kern_unlink(&nd);
1836 	nlookup_done(&nd);
1837 	return (error);
1838 }
1839 
1840 int
1841 kern_lseek(int fd, off_t offset, int whence, off_t *res)
1842 {
1843 	struct thread *td = curthread;
1844 	struct proc *p = td->td_proc;
1845 	struct file *fp;
1846 	struct vattr vattr;
1847 	int error;
1848 
1849 	fp = holdfp(p->p_fd, fd, -1);
1850 	if (fp == NULL)
1851 		return (EBADF);
1852 	if (fp->f_type != DTYPE_VNODE) {
1853 		error = ESPIPE;
1854 		goto done;
1855 	}
1856 
1857 	switch (whence) {
1858 	case L_INCR:
1859 		fp->f_offset += offset;
1860 		error = 0;
1861 		break;
1862 	case L_XTND:
1863 		error = VOP_GETATTR((struct vnode *)fp->f_data, &vattr);
1864 		if (error == 0)
1865 			fp->f_offset = offset + vattr.va_size;
1866 		break;
1867 	case L_SET:
1868 		fp->f_offset = offset;
1869 		error = 0;
1870 		break;
1871 	default:
1872 		error = EINVAL;
1873 		break;
1874 	}
1875 	*res = fp->f_offset;
1876 done:
1877 	fdrop(fp);
1878 	return (error);
1879 }
1880 
1881 /*
1882  * lseek_args(int fd, int pad, off_t offset, int whence)
1883  *
1884  * Reposition read/write file offset.
1885  */
1886 int
1887 sys_lseek(struct lseek_args *uap)
1888 {
1889 	int error;
1890 
1891 	error = kern_lseek(uap->fd, uap->offset, uap->whence,
1892 	    &uap->sysmsg_offset);
1893 
1894 	return (error);
1895 }
1896 
1897 int
1898 kern_access(struct nlookupdata *nd, int aflags)
1899 {
1900 	struct vnode *vp;
1901 	int error, flags;
1902 
1903 	if ((error = nlookup(nd)) != 0)
1904 		return (error);
1905 retry:
1906 	error = cache_vget(nd->nl_ncp, nd->nl_cred, LK_EXCLUSIVE, &vp);
1907 	if (error)
1908 		return (error);
1909 
1910 	/* Flags == 0 means only check for existence. */
1911 	if (aflags) {
1912 		flags = 0;
1913 		if (aflags & R_OK)
1914 			flags |= VREAD;
1915 		if (aflags & W_OK)
1916 			flags |= VWRITE;
1917 		if (aflags & X_OK)
1918 			flags |= VEXEC;
1919 		if ((flags & VWRITE) == 0 || (error = vn_writechk(vp)) == 0)
1920 			error = VOP_ACCESS(vp, flags, nd->nl_cred);
1921 
1922 		/*
1923 		 * If the file handle is stale we have to re-resolve the
1924 		 * entry.  This is a hack at the moment.
1925 		 */
1926 		if (error == ESTALE) {
1927 			cache_setunresolved(nd->nl_ncp);
1928 			error = cache_resolve(nd->nl_ncp, nd->nl_cred);
1929 			if (error == 0) {
1930 				vput(vp);
1931 				vp = NULL;
1932 				goto retry;
1933 			}
1934 		}
1935 	}
1936 	vput(vp);
1937 	return (error);
1938 }
1939 
1940 /*
1941  * access_args(char *path, int flags)
1942  *
1943  * Check access permissions.
1944  */
1945 int
1946 sys_access(struct access_args *uap)
1947 {
1948 	struct nlookupdata nd;
1949 	int error;
1950 
1951 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1952 	if (error == 0)
1953 		error = kern_access(&nd, uap->flags);
1954 	nlookup_done(&nd);
1955 	return (error);
1956 }
1957 
1958 int
1959 kern_stat(struct nlookupdata *nd, struct stat *st)
1960 {
1961 	int error;
1962 	struct vnode *vp;
1963 	thread_t td;
1964 
1965 	if ((error = nlookup(nd)) != 0)
1966 		return (error);
1967 again:
1968 	if ((vp = nd->nl_ncp->nc_vp) == NULL)
1969 		return (ENOENT);
1970 
1971 	td = curthread;
1972 	if ((error = vget(vp, LK_SHARED)) != 0)
1973 		return (error);
1974 	error = vn_stat(vp, st, nd->nl_cred);
1975 
1976 	/*
1977 	 * If the file handle is stale we have to re-resolve the entry.  This
1978 	 * is a hack at the moment.
1979 	 */
1980 	if (error == ESTALE) {
1981 		cache_setunresolved(nd->nl_ncp);
1982 		error = cache_resolve(nd->nl_ncp, nd->nl_cred);
1983 		if (error == 0) {
1984 			vput(vp);
1985 			goto again;
1986 		}
1987 	}
1988 	vput(vp);
1989 	return (error);
1990 }
1991 
1992 /*
1993  * stat_args(char *path, struct stat *ub)
1994  *
1995  * Get file status; this version follows links.
1996  */
1997 int
1998 sys_stat(struct stat_args *uap)
1999 {
2000 	struct nlookupdata nd;
2001 	struct stat st;
2002 	int error;
2003 
2004 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2005 	if (error == 0) {
2006 		error = kern_stat(&nd, &st);
2007 		if (error == 0)
2008 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2009 	}
2010 	nlookup_done(&nd);
2011 	return (error);
2012 }
2013 
2014 /*
2015  * lstat_args(char *path, struct stat *ub)
2016  *
2017  * Get file status; this version does not follow links.
2018  */
2019 int
2020 sys_lstat(struct lstat_args *uap)
2021 {
2022 	struct nlookupdata nd;
2023 	struct stat st;
2024 	int error;
2025 
2026 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2027 	if (error == 0) {
2028 		error = kern_stat(&nd, &st);
2029 		if (error == 0)
2030 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2031 	}
2032 	nlookup_done(&nd);
2033 	return (error);
2034 }
2035 
2036 /*
2037  * pathconf_Args(char *path, int name)
2038  *
2039  * Get configurable pathname variables.
2040  */
2041 /* ARGSUSED */
2042 int
2043 sys_pathconf(struct pathconf_args *uap)
2044 {
2045 	struct nlookupdata nd;
2046 	struct vnode *vp;
2047 	int error;
2048 
2049 	vp = NULL;
2050 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2051 	if (error == 0)
2052 		error = nlookup(&nd);
2053 	if (error == 0)
2054 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
2055 	nlookup_done(&nd);
2056 	if (error == 0) {
2057 		error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
2058 		vput(vp);
2059 	}
2060 	return (error);
2061 }
2062 
2063 /*
2064  * XXX: daver
2065  * kern_readlink isn't properly split yet.  There is a copyin burried
2066  * in VOP_READLINK().
2067  */
2068 int
2069 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2070 {
2071 	struct thread *td = curthread;
2072 	struct proc *p = td->td_proc;
2073 	struct vnode *vp;
2074 	struct iovec aiov;
2075 	struct uio auio;
2076 	int error;
2077 
2078 	if ((error = nlookup(nd)) != 0)
2079 		return (error);
2080 	error = cache_vget(nd->nl_ncp, nd->nl_cred, LK_EXCLUSIVE, &vp);
2081 	if (error)
2082 		return (error);
2083 	if (vp->v_type != VLNK) {
2084 		error = EINVAL;
2085 	} else {
2086 		aiov.iov_base = buf;
2087 		aiov.iov_len = count;
2088 		auio.uio_iov = &aiov;
2089 		auio.uio_iovcnt = 1;
2090 		auio.uio_offset = 0;
2091 		auio.uio_rw = UIO_READ;
2092 		auio.uio_segflg = UIO_USERSPACE;
2093 		auio.uio_td = td;
2094 		auio.uio_resid = count;
2095 		error = VOP_READLINK(vp, &auio, p->p_ucred);
2096 	}
2097 	vput(vp);
2098 	*res = count - auio.uio_resid;
2099 	return (error);
2100 }
2101 
2102 /*
2103  * readlink_args(char *path, char *buf, int count)
2104  *
2105  * Return target name of a symbolic link.
2106  */
2107 int
2108 sys_readlink(struct readlink_args *uap)
2109 {
2110 	struct nlookupdata nd;
2111 	int error;
2112 
2113 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2114 	if (error == 0) {
2115 		error = kern_readlink(&nd, uap->buf, uap->count,
2116 					&uap->sysmsg_result);
2117 	}
2118 	nlookup_done(&nd);
2119 	return (error);
2120 }
2121 
2122 static int
2123 setfflags(struct vnode *vp, int flags)
2124 {
2125 	struct thread *td = curthread;
2126 	struct proc *p = td->td_proc;
2127 	int error;
2128 	struct vattr vattr;
2129 
2130 	/*
2131 	 * Prevent non-root users from setting flags on devices.  When
2132 	 * a device is reused, users can retain ownership of the device
2133 	 * if they are allowed to set flags and programs assume that
2134 	 * chown can't fail when done as root.
2135 	 */
2136 	if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2137 	    ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0))
2138 		return (error);
2139 
2140 	/*
2141 	 * note: vget is required for any operation that might mod the vnode
2142 	 * so VINACTIVE is properly cleared.
2143 	 */
2144 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2145 		VATTR_NULL(&vattr);
2146 		vattr.va_flags = flags;
2147 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2148 		vput(vp);
2149 	}
2150 	return (error);
2151 }
2152 
2153 /*
2154  * chflags(char *path, int flags)
2155  *
2156  * Change flags of a file given a path name.
2157  */
2158 /* ARGSUSED */
2159 int
2160 sys_chflags(struct chflags_args *uap)
2161 {
2162 	struct nlookupdata nd;
2163 	struct vnode *vp;
2164 	int error;
2165 
2166 	vp = NULL;
2167 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2168 	/* XXX Add NLC flag indicating modifying operation? */
2169 	if (error == 0)
2170 		error = nlookup(&nd);
2171 	if (error == 0)
2172 		error = cache_vref(nd.nl_ncp, nd.nl_cred, &vp);
2173 	nlookup_done(&nd);
2174 	if (error == 0) {
2175 		error = setfflags(vp, uap->flags);
2176 		vrele(vp);
2177 	}
2178 	return (error);
2179 }
2180 
2181 /*
2182  * fchflags_args(int fd, int flags)
2183  *
2184  * Change flags of a file given a file descriptor.
2185  */
2186 /* ARGSUSED */
2187 int
2188 sys_fchflags(struct fchflags_args *uap)
2189 {
2190 	struct thread *td = curthread;
2191 	struct proc *p = td->td_proc;
2192 	struct file *fp;
2193 	int error;
2194 
2195 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2196 		return (error);
2197 	error = setfflags((struct vnode *) fp->f_data, uap->flags);
2198 	fdrop(fp);
2199 	return (error);
2200 }
2201 
2202 static int
2203 setfmode(struct vnode *vp, int mode)
2204 {
2205 	struct thread *td = curthread;
2206 	struct proc *p = td->td_proc;
2207 	int error;
2208 	struct vattr vattr;
2209 
2210 	/*
2211 	 * note: vget is required for any operation that might mod the vnode
2212 	 * so VINACTIVE is properly cleared.
2213 	 */
2214 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2215 		VATTR_NULL(&vattr);
2216 		vattr.va_mode = mode & ALLPERMS;
2217 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2218 		vput(vp);
2219 	}
2220 	return error;
2221 }
2222 
2223 int
2224 kern_chmod(struct nlookupdata *nd, int mode)
2225 {
2226 	struct vnode *vp;
2227 	int error;
2228 
2229 	/* XXX Add NLC flag indicating modifying operation? */
2230 	if ((error = nlookup(nd)) != 0)
2231 		return (error);
2232 	if ((error = cache_vref(nd->nl_ncp, nd->nl_cred, &vp)) != 0)
2233 		return (error);
2234 	error = setfmode(vp, mode);
2235 	vrele(vp);
2236 	return (error);
2237 }
2238 
2239 /*
2240  * chmod_args(char *path, int mode)
2241  *
2242  * Change mode of a file given path name.
2243  */
2244 /* ARGSUSED */
2245 int
2246 sys_chmod(struct chmod_args *uap)
2247 {
2248 	struct nlookupdata nd;
2249 	int error;
2250 
2251 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2252 	if (error == 0)
2253 		error = kern_chmod(&nd, uap->mode);
2254 	nlookup_done(&nd);
2255 	return (error);
2256 }
2257 
2258 /*
2259  * lchmod_args(char *path, int mode)
2260  *
2261  * Change mode of a file given path name (don't follow links.)
2262  */
2263 /* ARGSUSED */
2264 int
2265 sys_lchmod(struct lchmod_args *uap)
2266 {
2267 	struct nlookupdata nd;
2268 	int error;
2269 
2270 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2271 	if (error == 0)
2272 		error = kern_chmod(&nd, uap->mode);
2273 	nlookup_done(&nd);
2274 	return (error);
2275 }
2276 
2277 /*
2278  * fchmod_args(int fd, int mode)
2279  *
2280  * Change mode of a file given a file descriptor.
2281  */
2282 /* ARGSUSED */
2283 int
2284 sys_fchmod(struct fchmod_args *uap)
2285 {
2286 	struct thread *td = curthread;
2287 	struct proc *p = td->td_proc;
2288 	struct file *fp;
2289 	int error;
2290 
2291 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2292 		return (error);
2293 	error = setfmode((struct vnode *)fp->f_data, uap->mode);
2294 	fdrop(fp);
2295 	return (error);
2296 }
2297 
2298 static int
2299 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2300 {
2301 	struct thread *td = curthread;
2302 	struct proc *p = td->td_proc;
2303 	int error;
2304 	struct vattr vattr;
2305 
2306 	/*
2307 	 * note: vget is required for any operation that might mod the vnode
2308 	 * so VINACTIVE is properly cleared.
2309 	 */
2310 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2311 		VATTR_NULL(&vattr);
2312 		vattr.va_uid = uid;
2313 		vattr.va_gid = gid;
2314 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2315 		vput(vp);
2316 	}
2317 	return error;
2318 }
2319 
2320 int
2321 kern_chown(struct nlookupdata *nd, int uid, int gid)
2322 {
2323 	struct vnode *vp;
2324 	int error;
2325 
2326 	/* XXX Add NLC flag indicating modifying operation? */
2327 	if ((error = nlookup(nd)) != 0)
2328 		return (error);
2329 	if ((error = cache_vref(nd->nl_ncp, nd->nl_cred, &vp)) != 0)
2330 		return (error);
2331 	error = setfown(vp, uid, gid);
2332 	vrele(vp);
2333 	return (error);
2334 }
2335 
2336 /*
2337  * chown(char *path, int uid, int gid)
2338  *
2339  * Set ownership given a path name.
2340  */
2341 int
2342 sys_chown(struct chown_args *uap)
2343 {
2344 	struct nlookupdata nd;
2345 	int error;
2346 
2347 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2348 	if (error == 0)
2349 		error = kern_chown(&nd, uap->uid, uap->gid);
2350 	nlookup_done(&nd);
2351 	return (error);
2352 }
2353 
2354 /*
2355  * lchown_args(char *path, int uid, int gid)
2356  *
2357  * Set ownership given a path name, do not cross symlinks.
2358  */
2359 int
2360 sys_lchown(struct lchown_args *uap)
2361 {
2362 	struct nlookupdata nd;
2363 	int error;
2364 
2365 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2366 	if (error == 0)
2367 		error = kern_chown(&nd, uap->uid, uap->gid);
2368 	nlookup_done(&nd);
2369 	return (error);
2370 }
2371 
2372 /*
2373  * fchown_args(int fd, int uid, int gid)
2374  *
2375  * Set ownership given a file descriptor.
2376  */
2377 /* ARGSUSED */
2378 int
2379 sys_fchown(struct fchown_args *uap)
2380 {
2381 	struct thread *td = curthread;
2382 	struct proc *p = td->td_proc;
2383 	struct file *fp;
2384 	int error;
2385 
2386 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2387 		return (error);
2388 	error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
2389 	fdrop(fp);
2390 	return (error);
2391 }
2392 
2393 static int
2394 getutimes(const struct timeval *tvp, struct timespec *tsp)
2395 {
2396 	struct timeval tv[2];
2397 
2398 	if (tvp == NULL) {
2399 		microtime(&tv[0]);
2400 		TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
2401 		tsp[1] = tsp[0];
2402 	} else {
2403 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
2404 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
2405 	}
2406 	return 0;
2407 }
2408 
2409 static int
2410 setutimes(struct vnode *vp, const struct timespec *ts, int nullflag)
2411 {
2412 	struct thread *td = curthread;
2413 	struct proc *p = td->td_proc;
2414 	int error;
2415 	struct vattr vattr;
2416 
2417 	/*
2418 	 * note: vget is required for any operation that might mod the vnode
2419 	 * so VINACTIVE is properly cleared.
2420 	 */
2421 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2422 		VATTR_NULL(&vattr);
2423 		vattr.va_atime = ts[0];
2424 		vattr.va_mtime = ts[1];
2425 		if (nullflag)
2426 			vattr.va_vaflags |= VA_UTIMES_NULL;
2427 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2428 		vput(vp);
2429 	}
2430 	return error;
2431 }
2432 
2433 int
2434 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
2435 {
2436 	struct timespec ts[2];
2437 	struct vnode *vp;
2438 	int error;
2439 
2440 	if ((error = getutimes(tptr, ts)) != 0)
2441 		return (error);
2442 	/* XXX Add NLC flag indicating modifying operation? */
2443 	if ((error = nlookup(nd)) != 0)
2444 		return (error);
2445 	if ((error = cache_vref(nd->nl_ncp, nd->nl_cred, &vp)) != 0)
2446 		return (error);
2447 	error = setutimes(vp, ts, tptr == NULL);
2448 	vrele(vp);
2449 	return (error);
2450 }
2451 
2452 /*
2453  * utimes_args(char *path, struct timeval *tptr)
2454  *
2455  * Set the access and modification times of a file.
2456  */
2457 int
2458 sys_utimes(struct utimes_args *uap)
2459 {
2460 	struct timeval tv[2];
2461 	struct nlookupdata nd;
2462 	int error;
2463 
2464 	if (uap->tptr) {
2465  		error = copyin(uap->tptr, tv, sizeof(tv));
2466 		if (error)
2467 			return (error);
2468 	}
2469 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2470 	if (error == 0)
2471 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2472 	nlookup_done(&nd);
2473 	return (error);
2474 }
2475 
2476 /*
2477  * lutimes_args(char *path, struct timeval *tptr)
2478  *
2479  * Set the access and modification times of a file.
2480  */
2481 int
2482 sys_lutimes(struct lutimes_args *uap)
2483 {
2484 	struct timeval tv[2];
2485 	struct nlookupdata nd;
2486 	int error;
2487 
2488 	if (uap->tptr) {
2489 		error = copyin(uap->tptr, tv, sizeof(tv));
2490 		if (error)
2491 			return (error);
2492 	}
2493 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2494 	if (error == 0)
2495 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2496 	nlookup_done(&nd);
2497 	return (error);
2498 }
2499 
2500 int
2501 kern_futimes(int fd, struct timeval *tptr)
2502 {
2503 	struct thread *td = curthread;
2504 	struct proc *p = td->td_proc;
2505 	struct timespec ts[2];
2506 	struct file *fp;
2507 	int error;
2508 
2509 	error = getutimes(tptr, ts);
2510 	if (error)
2511 		return (error);
2512 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2513 		return (error);
2514 	error =  setutimes((struct vnode *)fp->f_data, ts, tptr == NULL);
2515 	fdrop(fp);
2516 	return (error);
2517 }
2518 
2519 /*
2520  * futimes_args(int fd, struct timeval *tptr)
2521  *
2522  * Set the access and modification times of a file.
2523  */
2524 int
2525 sys_futimes(struct futimes_args *uap)
2526 {
2527 	struct timeval tv[2];
2528 	int error;
2529 
2530 	if (uap->tptr) {
2531 		error = copyin(uap->tptr, tv, sizeof(tv));
2532 		if (error)
2533 			return (error);
2534 	}
2535 
2536 	error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
2537 
2538 	return (error);
2539 }
2540 
2541 int
2542 kern_truncate(struct nlookupdata *nd, off_t length)
2543 {
2544 	struct vnode *vp;
2545 	struct vattr vattr;
2546 	int error;
2547 
2548 	if (length < 0)
2549 		return(EINVAL);
2550 	/* XXX Add NLC flag indicating modifying operation? */
2551 	if ((error = nlookup(nd)) != 0)
2552 		return (error);
2553 	if ((error = cache_vref(nd->nl_ncp, nd->nl_cred, &vp)) != 0)
2554 		return (error);
2555 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
2556 		vrele(vp);
2557 		return (error);
2558 	}
2559 	if (vp->v_type == VDIR) {
2560 		error = EISDIR;
2561 	} else if ((error = vn_writechk(vp)) == 0 &&
2562 	    (error = VOP_ACCESS(vp, VWRITE, nd->nl_cred)) == 0) {
2563 		VATTR_NULL(&vattr);
2564 		vattr.va_size = length;
2565 		error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
2566 	}
2567 	vput(vp);
2568 	return (error);
2569 }
2570 
2571 /*
2572  * truncate(char *path, int pad, off_t length)
2573  *
2574  * Truncate a file given its path name.
2575  */
2576 int
2577 sys_truncate(struct truncate_args *uap)
2578 {
2579 	struct nlookupdata nd;
2580 	int error;
2581 
2582 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2583 	if (error == 0)
2584 		error = kern_truncate(&nd, uap->length);
2585 	nlookup_done(&nd);
2586 	return error;
2587 }
2588 
2589 int
2590 kern_ftruncate(int fd, off_t length)
2591 {
2592 	struct thread *td = curthread;
2593 	struct proc *p = td->td_proc;
2594 	struct vattr vattr;
2595 	struct vnode *vp;
2596 	struct file *fp;
2597 	int error;
2598 
2599 	if (length < 0)
2600 		return(EINVAL);
2601 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2602 		return (error);
2603 	if ((fp->f_flag & FWRITE) == 0) {
2604 		error = EINVAL;
2605 		goto done;
2606 	}
2607 	vp = (struct vnode *)fp->f_data;
2608 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2609 	if (vp->v_type == VDIR) {
2610 		error = EISDIR;
2611 	} else if ((error = vn_writechk(vp)) == 0) {
2612 		VATTR_NULL(&vattr);
2613 		vattr.va_size = length;
2614 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
2615 	}
2616 	VOP_UNLOCK(vp, 0);
2617 done:
2618 	fdrop(fp);
2619 	return (error);
2620 }
2621 
2622 /*
2623  * ftruncate_args(int fd, int pad, off_t length)
2624  *
2625  * Truncate a file given a file descriptor.
2626  */
2627 int
2628 sys_ftruncate(struct ftruncate_args *uap)
2629 {
2630 	int error;
2631 
2632 	error = kern_ftruncate(uap->fd, uap->length);
2633 
2634 	return (error);
2635 }
2636 
2637 /*
2638  * fsync(int fd)
2639  *
2640  * Sync an open file.
2641  */
2642 /* ARGSUSED */
2643 int
2644 sys_fsync(struct fsync_args *uap)
2645 {
2646 	struct thread *td = curthread;
2647 	struct proc *p = td->td_proc;
2648 	struct vnode *vp;
2649 	struct file *fp;
2650 	vm_object_t obj;
2651 	int error;
2652 
2653 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2654 		return (error);
2655 	vp = (struct vnode *)fp->f_data;
2656 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2657 	if ((obj = vp->v_object) != NULL)
2658 		vm_object_page_clean(obj, 0, 0, 0);
2659 	if ((error = VOP_FSYNC(vp, MNT_WAIT)) == 0 &&
2660 	    vp->v_mount && (vp->v_mount->mnt_flag & MNT_SOFTDEP) &&
2661 	    bioops.io_fsync) {
2662 		error = (*bioops.io_fsync)(vp);
2663 	}
2664 	VOP_UNLOCK(vp, 0);
2665 	fdrop(fp);
2666 	return (error);
2667 }
2668 
2669 int
2670 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
2671 {
2672 	struct namecache *fncpd;
2673 	struct namecache *tncpd;
2674 	struct namecache *ncp;
2675 	struct mount *mp;
2676 	int error;
2677 
2678 	bwillwrite();
2679 	if ((error = nlookup(fromnd)) != 0)
2680 		return (error);
2681 	if ((fncpd = fromnd->nl_ncp->nc_parent) == NULL)
2682 		return (ENOENT);
2683 	cache_hold(fncpd);
2684 
2685 	/*
2686 	 * unlock the source ncp so we can lookup the target ncp without
2687 	 * deadlocking.  The target may or may not exist so we do not check
2688 	 * for a target vp like kern_mkdir() and other creation functions do.
2689 	 *
2690 	 * The source and target directories are ref'd and rechecked after
2691 	 * everything is relocked to determine if the source or target file
2692 	 * has been renamed.
2693 	 */
2694 	KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
2695 	fromnd->nl_flags &= ~NLC_NCPISLOCKED;
2696 	cache_unlock(fromnd->nl_ncp);
2697 
2698 	tond->nl_flags |= NLC_CREATE;
2699 	if ((error = nlookup(tond)) != 0) {
2700 		cache_drop(fncpd);
2701 		return (error);
2702 	}
2703 	if ((tncpd = tond->nl_ncp->nc_parent) == NULL) {
2704 		cache_drop(fncpd);
2705 		return (ENOENT);
2706 	}
2707 	cache_hold(tncpd);
2708 
2709 	/*
2710 	 * If the source and target are the same there is nothing to do
2711 	 */
2712 	if (fromnd->nl_ncp == tond->nl_ncp) {
2713 		cache_drop(fncpd);
2714 		cache_drop(tncpd);
2715 		return (0);
2716 	}
2717 
2718 	/*
2719 	 * relock the source ncp.  NOTE AFTER RELOCKING: the source ncp
2720 	 * may have become invalid while it was unlocked, nc_vp and nc_mount
2721 	 * could be NULL.
2722 	 */
2723 	if (cache_lock_nonblock(fromnd->nl_ncp) == 0) {
2724 		cache_resolve(fromnd->nl_ncp, fromnd->nl_cred);
2725 	} else if (fromnd->nl_ncp > tond->nl_ncp) {
2726 		cache_lock(fromnd->nl_ncp);
2727 		cache_resolve(fromnd->nl_ncp, fromnd->nl_cred);
2728 	} else {
2729 		cache_unlock(tond->nl_ncp);
2730 		cache_lock(fromnd->nl_ncp);
2731 		cache_resolve(fromnd->nl_ncp, fromnd->nl_cred);
2732 		cache_lock(tond->nl_ncp);
2733 		cache_resolve(tond->nl_ncp, tond->nl_cred);
2734 	}
2735 	fromnd->nl_flags |= NLC_NCPISLOCKED;
2736 
2737 	/*
2738 	 * make sure the parent directories linkages are the same
2739 	 */
2740 	if (fncpd != fromnd->nl_ncp->nc_parent ||
2741 	    tncpd != tond->nl_ncp->nc_parent) {
2742 		cache_drop(fncpd);
2743 		cache_drop(tncpd);
2744 		return (ENOENT);
2745 	}
2746 
2747 	/*
2748 	 * Both the source and target must be within the same filesystem and
2749 	 * in the same filesystem as their parent directories within the
2750 	 * namecache topology.
2751 	 *
2752 	 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
2753 	 */
2754 	mp = fncpd->nc_mount;
2755 	if (mp != tncpd->nc_mount || mp != fromnd->nl_ncp->nc_mount ||
2756 	    mp != tond->nl_ncp->nc_mount) {
2757 		cache_drop(fncpd);
2758 		cache_drop(tncpd);
2759 		return (EXDEV);
2760 	}
2761 
2762 	/*
2763 	 * If the target exists and either the source or target is a directory,
2764 	 * then both must be directories.
2765 	 *
2766 	 * Due to relocking of the source, fromnd->nl_ncp->nc_vp might have
2767 	 * become NULL.
2768 	 */
2769 	if (tond->nl_ncp->nc_vp) {
2770 		if (fromnd->nl_ncp->nc_vp == NULL) {
2771 			error = ENOENT;
2772 		} else if (fromnd->nl_ncp->nc_vp->v_type == VDIR) {
2773 			if (tond->nl_ncp->nc_vp->v_type != VDIR)
2774 				error = ENOTDIR;
2775 		} else if (tond->nl_ncp->nc_vp->v_type == VDIR) {
2776 			error = EISDIR;
2777 		}
2778 	}
2779 
2780 	/*
2781 	 * You cannot rename a source into itself or a subdirectory of itself.
2782 	 * We check this by travsersing the target directory upwards looking
2783 	 * for a match against the source.
2784 	 */
2785 	if (error == 0) {
2786 		for (ncp = tncpd; ncp; ncp = ncp->nc_parent) {
2787 			if (fromnd->nl_ncp == ncp) {
2788 				error = EINVAL;
2789 				break;
2790 			}
2791 		}
2792 	}
2793 
2794 	cache_drop(fncpd);
2795 	cache_drop(tncpd);
2796 
2797 	/*
2798 	 * Even though the namespaces are different, they may still represent
2799 	 * hardlinks to the same file.  The filesystem might have a hard time
2800 	 * with this so we issue a NREMOVE of the source instead of a NRENAME
2801 	 * when we detect the situation.
2802 	 */
2803 	if (error == 0) {
2804 		if (fromnd->nl_ncp->nc_vp == tond->nl_ncp->nc_vp) {
2805 			error = VOP_NREMOVE(fromnd->nl_ncp, fromnd->nl_cred);
2806 		} else {
2807 			error = VOP_NRENAME(fromnd->nl_ncp, tond->nl_ncp,
2808 					    tond->nl_cred);
2809 		}
2810 	}
2811 	return (error);
2812 }
2813 
2814 /*
2815  * rename_args(char *from, char *to)
2816  *
2817  * Rename files.  Source and destination must either both be directories,
2818  * or both not be directories.  If target is a directory, it must be empty.
2819  */
2820 int
2821 sys_rename(struct rename_args *uap)
2822 {
2823 	struct nlookupdata fromnd, tond;
2824 	int error;
2825 
2826 	error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
2827 	if (error == 0) {
2828 		error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
2829 		if (error == 0)
2830 			error = kern_rename(&fromnd, &tond);
2831 		nlookup_done(&tond);
2832 	}
2833 	nlookup_done(&fromnd);
2834 	return (error);
2835 }
2836 
2837 int
2838 kern_mkdir(struct nlookupdata *nd, int mode)
2839 {
2840 	struct thread *td = curthread;
2841 	struct proc *p = td->td_proc;
2842 	struct namecache *ncp;
2843 	struct vnode *vp;
2844 	struct vattr vattr;
2845 	int error;
2846 
2847 	bwillwrite();
2848 	nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE;
2849 	if ((error = nlookup(nd)) != 0)
2850 		return (error);
2851 
2852 	ncp = nd->nl_ncp;
2853 	if (ncp->nc_vp)
2854 		return (EEXIST);
2855 
2856 	VATTR_NULL(&vattr);
2857 	vattr.va_type = VDIR;
2858 	vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
2859 
2860 	vp = NULL;
2861 	error = VOP_NMKDIR(ncp, &vp, p->p_ucred, &vattr);
2862 	if (error == 0)
2863 		vput(vp);
2864 	return (error);
2865 }
2866 
2867 /*
2868  * mkdir_args(char *path, int mode)
2869  *
2870  * Make a directory file.
2871  */
2872 /* ARGSUSED */
2873 int
2874 sys_mkdir(struct mkdir_args *uap)
2875 {
2876 	struct nlookupdata nd;
2877 	int error;
2878 
2879 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2880 	if (error == 0)
2881 		error = kern_mkdir(&nd, uap->mode);
2882 	nlookup_done(&nd);
2883 	return (error);
2884 }
2885 
2886 int
2887 kern_rmdir(struct nlookupdata *nd)
2888 {
2889 	struct namecache *ncp;
2890 	int error;
2891 
2892 	bwillwrite();
2893 	nd->nl_flags |= NLC_DELETE;
2894 	if ((error = nlookup(nd)) != 0)
2895 		return (error);
2896 
2897 	ncp = nd->nl_ncp;
2898 	error = VOP_NRMDIR(ncp, nd->nl_cred);
2899 	return (error);
2900 }
2901 
2902 /*
2903  * rmdir_args(char *path)
2904  *
2905  * Remove a directory file.
2906  */
2907 /* ARGSUSED */
2908 int
2909 sys_rmdir(struct rmdir_args *uap)
2910 {
2911 	struct nlookupdata nd;
2912 	int error;
2913 
2914 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2915 	if (error == 0)
2916 		error = kern_rmdir(&nd);
2917 	nlookup_done(&nd);
2918 	return (error);
2919 }
2920 
2921 int
2922 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
2923     enum uio_seg direction)
2924 {
2925 	struct thread *td = curthread;
2926 	struct proc *p = td->td_proc;
2927 	struct vnode *vp;
2928 	struct file *fp;
2929 	struct uio auio;
2930 	struct iovec aiov;
2931 	long loff;
2932 	int error, eofflag;
2933 
2934 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2935 		return (error);
2936 	if ((fp->f_flag & FREAD) == 0) {
2937 		error = EBADF;
2938 		goto done;
2939 	}
2940 	vp = (struct vnode *)fp->f_data;
2941 unionread:
2942 	if (vp->v_type != VDIR) {
2943 		error = EINVAL;
2944 		goto done;
2945 	}
2946 	aiov.iov_base = buf;
2947 	aiov.iov_len = count;
2948 	auio.uio_iov = &aiov;
2949 	auio.uio_iovcnt = 1;
2950 	auio.uio_rw = UIO_READ;
2951 	auio.uio_segflg = direction;
2952 	auio.uio_td = td;
2953 	auio.uio_resid = count;
2954 	/* vn_lock(vp, LK_SHARED | LK_RETRY); */
2955 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2956 	loff = auio.uio_offset = fp->f_offset;
2957 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
2958 	fp->f_offset = auio.uio_offset;
2959 	VOP_UNLOCK(vp, 0);
2960 	if (error)
2961 		goto done;
2962 	if (count == auio.uio_resid) {
2963 		if (union_dircheckp) {
2964 			error = union_dircheckp(td, &vp, fp);
2965 			if (error == -1)
2966 				goto unionread;
2967 			if (error)
2968 				goto done;
2969 		}
2970 		if ((vp->v_flag & VROOT) &&
2971 		    (vp->v_mount->mnt_flag & MNT_UNION)) {
2972 			struct vnode *tvp = vp;
2973 			vp = vp->v_mount->mnt_vnodecovered;
2974 			vref(vp);
2975 			fp->f_data = vp;
2976 			fp->f_offset = 0;
2977 			vrele(tvp);
2978 			goto unionread;
2979 		}
2980 	}
2981 	if (basep) {
2982 		*basep = loff;
2983 	}
2984 	*res = count - auio.uio_resid;
2985 done:
2986 	fdrop(fp);
2987 	return (error);
2988 }
2989 
2990 /*
2991  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
2992  *
2993  * Read a block of directory entries in a file system independent format.
2994  */
2995 int
2996 sys_getdirentries(struct getdirentries_args *uap)
2997 {
2998 	long base;
2999 	int error;
3000 
3001 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3002 	    &uap->sysmsg_result, UIO_USERSPACE);
3003 
3004 	if (error == 0)
3005 		error = copyout(&base, uap->basep, sizeof(*uap->basep));
3006 	return (error);
3007 }
3008 
3009 /*
3010  * getdents_args(int fd, char *buf, size_t count)
3011  */
3012 int
3013 sys_getdents(struct getdents_args *uap)
3014 {
3015 	int error;
3016 
3017 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3018 	    &uap->sysmsg_result, UIO_USERSPACE);
3019 
3020 	return (error);
3021 }
3022 
3023 /*
3024  * umask(int newmask)
3025  *
3026  * Set the mode mask for creation of filesystem nodes.
3027  *
3028  * MP SAFE
3029  */
3030 int
3031 sys_umask(struct umask_args *uap)
3032 {
3033 	struct thread *td = curthread;
3034 	struct proc *p = td->td_proc;
3035 	struct filedesc *fdp;
3036 
3037 	fdp = p->p_fd;
3038 	uap->sysmsg_result = fdp->fd_cmask;
3039 	fdp->fd_cmask = uap->newmask & ALLPERMS;
3040 	return (0);
3041 }
3042 
3043 /*
3044  * revoke(char *path)
3045  *
3046  * Void all references to file by ripping underlying filesystem
3047  * away from vnode.
3048  */
3049 /* ARGSUSED */
3050 int
3051 sys_revoke(struct revoke_args *uap)
3052 {
3053 	struct nlookupdata nd;
3054 	struct vattr vattr;
3055 	struct vnode *vp;
3056 	struct ucred *cred;
3057 	int error;
3058 
3059 	vp = NULL;
3060 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3061 	if (error == 0)
3062 		error = nlookup(&nd);
3063 	if (error == 0)
3064 		error = cache_vref(nd.nl_ncp, nd.nl_cred, &vp);
3065 	cred = crhold(nd.nl_cred);
3066 	nlookup_done(&nd);
3067 	if (error == 0) {
3068 		if (vp->v_type != VCHR && vp->v_type != VBLK)
3069 			error = EINVAL;
3070 		if (error == 0)
3071 			error = VOP_GETATTR(vp, &vattr);
3072 		if (error == 0 && cred->cr_uid != vattr.va_uid)
3073 			error = suser_cred(cred, PRISON_ROOT);
3074 		if (error == 0 && count_udev(vp->v_udev) > 0) {
3075 			if ((error = vx_lock(vp)) == 0) {
3076 				VOP_REVOKE(vp, REVOKEALL);
3077 				vx_unlock(vp);
3078 			}
3079 		}
3080 		vrele(vp);
3081 	}
3082 	if (cred)
3083 		crfree(cred);
3084 	return (error);
3085 }
3086 
3087 /*
3088  * getfh_args(char *fname, fhandle_t *fhp)
3089  *
3090  * Get (NFS) file handle
3091  */
3092 int
3093 sys_getfh(struct getfh_args *uap)
3094 {
3095 	struct thread *td = curthread;
3096 	struct nlookupdata nd;
3097 	fhandle_t fh;
3098 	struct vnode *vp;
3099 	int error;
3100 
3101 	/*
3102 	 * Must be super user
3103 	 */
3104 	if ((error = suser(td)) != 0)
3105 		return (error);
3106 
3107 	vp = NULL;
3108 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3109 	if (error == 0)
3110 		error = nlookup(&nd);
3111 	if (error == 0)
3112 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
3113 	nlookup_done(&nd);
3114 	if (error == 0) {
3115 		bzero(&fh, sizeof(fh));
3116 		fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
3117 		error = VFS_VPTOFH(vp, &fh.fh_fid);
3118 		vput(vp);
3119 		if (error == 0)
3120 			error = copyout(&fh, uap->fhp, sizeof(fh));
3121 	}
3122 	return (error);
3123 }
3124 
3125 /*
3126  * fhopen_args(const struct fhandle *u_fhp, int flags)
3127  *
3128  * syscall for the rpc.lockd to use to translate a NFS file handle into
3129  * an open descriptor.
3130  *
3131  * warning: do not remove the suser() call or this becomes one giant
3132  * security hole.
3133  */
3134 int
3135 sys_fhopen(struct fhopen_args *uap)
3136 {
3137 	struct thread *td = curthread;
3138 	struct proc *p = td->td_proc;
3139 	struct mount *mp;
3140 	struct vnode *vp;
3141 	struct fhandle fhp;
3142 	struct vattr vat;
3143 	struct vattr *vap = &vat;
3144 	struct flock lf;
3145 	int fmode, mode, error, type;
3146 	struct file *nfp;
3147 	struct file *fp;
3148 	int indx;
3149 
3150 	/*
3151 	 * Must be super user
3152 	 */
3153 	error = suser(td);
3154 	if (error)
3155 		return (error);
3156 
3157 	fmode = FFLAGS(uap->flags);
3158 	/* why not allow a non-read/write open for our lockd? */
3159 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3160 		return (EINVAL);
3161 	error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3162 	if (error)
3163 		return(error);
3164 	/* find the mount point */
3165 	mp = vfs_getvfs(&fhp.fh_fsid);
3166 	if (mp == NULL)
3167 		return (ESTALE);
3168 	/* now give me my vnode, it gets returned to me locked */
3169 	error = VFS_FHTOVP(mp, &fhp.fh_fid, &vp);
3170 	if (error)
3171 		return (error);
3172  	/*
3173 	 * from now on we have to make sure not
3174 	 * to forget about the vnode
3175 	 * any error that causes an abort must vput(vp)
3176 	 * just set error = err and 'goto bad;'.
3177 	 */
3178 
3179 	/*
3180 	 * from vn_open
3181 	 */
3182 	if (vp->v_type == VLNK) {
3183 		error = EMLINK;
3184 		goto bad;
3185 	}
3186 	if (vp->v_type == VSOCK) {
3187 		error = EOPNOTSUPP;
3188 		goto bad;
3189 	}
3190 	mode = 0;
3191 	if (fmode & (FWRITE | O_TRUNC)) {
3192 		if (vp->v_type == VDIR) {
3193 			error = EISDIR;
3194 			goto bad;
3195 		}
3196 		error = vn_writechk(vp);
3197 		if (error)
3198 			goto bad;
3199 		mode |= VWRITE;
3200 	}
3201 	if (fmode & FREAD)
3202 		mode |= VREAD;
3203 	if (mode) {
3204 		error = VOP_ACCESS(vp, mode, p->p_ucred);
3205 		if (error)
3206 			goto bad;
3207 	}
3208 	if (fmode & O_TRUNC) {
3209 		VOP_UNLOCK(vp, 0);			/* XXX */
3210 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
3211 		VATTR_NULL(vap);
3212 		vap->va_size = 0;
3213 		error = VOP_SETATTR(vp, vap, p->p_ucred);
3214 		if (error)
3215 			goto bad;
3216 	}
3217 
3218 	/*
3219 	 * VOP_OPEN needs the file pointer so it can potentially override
3220 	 * it.
3221 	 *
3222 	 * WARNING! no f_ncp will be associated when fhopen()ing a directory.
3223 	 * XXX
3224 	 */
3225 	if ((error = falloc(p, &nfp, &indx)) != 0)
3226 		goto bad;
3227 	fp = nfp;
3228 
3229 	error = VOP_OPEN(vp, fmode, p->p_ucred, fp);
3230 	if (error) {
3231 		/*
3232 		 * setting f_ops this way prevents VOP_CLOSE from being
3233 		 * called or fdrop() releasing the vp from v_data.   Since
3234 		 * the VOP_OPEN failed we don't want to VOP_CLOSE.
3235 		 */
3236 		fp->f_ops = &badfileops;
3237 		fp->f_data = NULL;
3238 		goto bad_drop;
3239 	}
3240 
3241 	/*
3242 	 * The fp is given its own reference, we still have our ref and lock.
3243 	 *
3244 	 * Assert that all regular files must be created with a VM object.
3245 	 */
3246 	if (vp->v_type == VREG && vp->v_object == NULL) {
3247 		printf("fhopen: regular file did not have VM object: %p\n", vp);
3248 		goto bad_drop;
3249 	}
3250 
3251 	/*
3252 	 * The open was successful.  Handle any locking requirements.
3253 	 */
3254 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
3255 		lf.l_whence = SEEK_SET;
3256 		lf.l_start = 0;
3257 		lf.l_len = 0;
3258 		if (fmode & O_EXLOCK)
3259 			lf.l_type = F_WRLCK;
3260 		else
3261 			lf.l_type = F_RDLCK;
3262 		if (fmode & FNONBLOCK)
3263 			type = 0;
3264 		else
3265 			type = F_WAIT;
3266 		VOP_UNLOCK(vp, 0);
3267 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3268 			/*
3269 			 * release our private reference.
3270 			 */
3271 			fsetfd(p, NULL, indx);
3272 			fdrop(fp);
3273 			vrele(vp);
3274 			return (error);
3275 		}
3276 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3277 		fp->f_flag |= FHASLOCK;
3278 	}
3279 
3280 	/*
3281 	 * Clean up.  Associate the file pointer with the previously
3282 	 * reserved descriptor and return it.
3283 	 */
3284 	vput(vp);
3285 	fsetfd(p, fp, indx);
3286 	fdrop(fp);
3287 	uap->sysmsg_result = indx;
3288 	return (0);
3289 
3290 bad_drop:
3291 	fsetfd(p, NULL, indx);
3292 	fdrop(fp);
3293 bad:
3294 	vput(vp);
3295 	return (error);
3296 }
3297 
3298 /*
3299  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3300  */
3301 int
3302 sys_fhstat(struct fhstat_args *uap)
3303 {
3304 	struct thread *td = curthread;
3305 	struct stat sb;
3306 	fhandle_t fh;
3307 	struct mount *mp;
3308 	struct vnode *vp;
3309 	int error;
3310 
3311 	/*
3312 	 * Must be super user
3313 	 */
3314 	error = suser(td);
3315 	if (error)
3316 		return (error);
3317 
3318 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
3319 	if (error)
3320 		return (error);
3321 
3322 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3323 		return (ESTALE);
3324 	if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3325 		return (error);
3326 	error = vn_stat(vp, &sb, td->td_proc->p_ucred);
3327 	vput(vp);
3328 	if (error)
3329 		return (error);
3330 	error = copyout(&sb, uap->sb, sizeof(sb));
3331 	return (error);
3332 }
3333 
3334 /*
3335  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
3336  */
3337 int
3338 sys_fhstatfs(struct fhstatfs_args *uap)
3339 {
3340 	struct thread *td = curthread;
3341 	struct proc *p = td->td_proc;
3342 	struct statfs *sp;
3343 	struct mount *mp;
3344 	struct vnode *vp;
3345 	struct statfs sb;
3346 	char *fullpath, *freepath;
3347 	fhandle_t fh;
3348 	int error;
3349 
3350 	/*
3351 	 * Must be super user
3352 	 */
3353 	if ((error = suser(td)))
3354 		return (error);
3355 
3356 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3357 		return (error);
3358 
3359 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3360 		return (ESTALE);
3361 
3362 	if (p != NULL && (p->p_fd->fd_nrdir->nc_flag & NCF_ROOT) == 0 &&
3363 	    !chroot_visible_mnt(mp, p))
3364 		return (ESTALE);
3365 
3366 	if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3367 		return (error);
3368 	mp = vp->v_mount;
3369 	sp = &mp->mnt_stat;
3370 	vput(vp);
3371 	if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0)
3372 		return (error);
3373 
3374 	error = cache_fullpath(p, mp->mnt_ncp, &fullpath, &freepath);
3375 	if (error)
3376 		return(error);
3377 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3378 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
3379 	free(freepath, M_TEMP);
3380 
3381 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
3382 	if (suser(td)) {
3383 		bcopy(sp, &sb, sizeof(sb));
3384 		sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
3385 		sp = &sb;
3386 	}
3387 	return (copyout(sp, uap->buf, sizeof(*sp)));
3388 }
3389 
3390 /*
3391  * Syscall to push extended attribute configuration information into the
3392  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
3393  * a command (int cmd), and attribute name and misc data.  For now, the
3394  * attribute name is left in userspace for consumption by the VFS_op.
3395  * It will probably be changed to be copied into sysspace by the
3396  * syscall in the future, once issues with various consumers of the
3397  * attribute code have raised their hands.
3398  *
3399  * Currently this is used only by UFS Extended Attributes.
3400  */
3401 int
3402 sys_extattrctl(struct extattrctl_args *uap)
3403 {
3404 	struct nlookupdata nd;
3405 	struct mount *mp;
3406 	struct vnode *vp;
3407 	int error;
3408 
3409 	vp = NULL;
3410 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3411 	if (error == 0)
3412 		error = nlookup(&nd);
3413 	if (error == 0) {
3414 		mp = nd.nl_ncp->nc_mount;
3415 		error = VFS_EXTATTRCTL(mp, uap->cmd,
3416 				uap->attrname, uap->arg,
3417 				nd.nl_cred);
3418 	}
3419 	nlookup_done(&nd);
3420 	return (error);
3421 }
3422 
3423 /*
3424  * Syscall to set a named extended attribute on a file or directory.
3425  * Accepts attribute name, and a uio structure pointing to the data to set.
3426  * The uio is consumed in the style of writev().  The real work happens
3427  * in VOP_SETEXTATTR().
3428  */
3429 int
3430 sys_extattr_set_file(struct extattr_set_file_args *uap)
3431 {
3432 	char attrname[EXTATTR_MAXNAMELEN];
3433 	struct iovec aiov[UIO_SMALLIOV];
3434 	struct iovec *needfree;
3435 	struct nlookupdata nd;
3436 	struct iovec *iov;
3437 	struct vnode *vp;
3438 	struct uio auio;
3439 	u_int iovlen;
3440 	u_int cnt;
3441 	int error;
3442 	int i;
3443 
3444 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3445 	if (error)
3446 		return (error);
3447 
3448 	vp = NULL;
3449 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3450 	if (error == 0)
3451 		error = nlookup(&nd);
3452 	if (error == 0)
3453 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
3454 	if (error) {
3455 		nlookup_done(&nd);
3456 		return (error);
3457 	}
3458 
3459 	needfree = NULL;
3460 	iovlen = uap->iovcnt * sizeof(struct iovec);
3461 	if (uap->iovcnt > UIO_SMALLIOV) {
3462 		if (uap->iovcnt > UIO_MAXIOV) {
3463 			error = EINVAL;
3464 			goto done;
3465 		}
3466 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3467 		needfree = iov;
3468 	} else {
3469 		iov = aiov;
3470 	}
3471 	auio.uio_iov = iov;
3472 	auio.uio_iovcnt = uap->iovcnt;
3473 	auio.uio_rw = UIO_WRITE;
3474 	auio.uio_segflg = UIO_USERSPACE;
3475 	auio.uio_td = nd.nl_td;
3476 	auio.uio_offset = 0;
3477 	if ((error = copyin(uap->iovp, iov, iovlen)))
3478 		goto done;
3479 	auio.uio_resid = 0;
3480 	for (i = 0; i < uap->iovcnt; i++) {
3481 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
3482 			error = EINVAL;
3483 			goto done;
3484 		}
3485 		auio.uio_resid += iov->iov_len;
3486 		iov++;
3487 	}
3488 	cnt = auio.uio_resid;
3489 	error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3490 	cnt -= auio.uio_resid;
3491 	uap->sysmsg_result = cnt;
3492 done:
3493 	vput(vp);
3494 	nlookup_done(&nd);
3495 	if (needfree)
3496 		FREE(needfree, M_IOV);
3497 	return (error);
3498 }
3499 
3500 /*
3501  * Syscall to get a named extended attribute on a file or directory.
3502  * Accepts attribute name, and a uio structure pointing to a buffer for the
3503  * data.  The uio is consumed in the style of readv().  The real work
3504  * happens in VOP_GETEXTATTR();
3505  */
3506 int
3507 sys_extattr_get_file(struct extattr_get_file_args *uap)
3508 {
3509 	char attrname[EXTATTR_MAXNAMELEN];
3510 	struct iovec aiov[UIO_SMALLIOV];
3511 	struct iovec *needfree;
3512 	struct nlookupdata nd;
3513 	struct iovec *iov;
3514 	struct vnode *vp;
3515 	struct uio auio;
3516 	u_int iovlen;
3517 	u_int cnt;
3518 	int error;
3519 	int i;
3520 
3521 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3522 	if (error)
3523 		return (error);
3524 
3525 	vp = NULL;
3526 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3527 	if (error == 0)
3528 		error = nlookup(&nd);
3529 	if (error == 0)
3530 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
3531 	if (error) {
3532 		nlookup_done(&nd);
3533 		return (error);
3534 	}
3535 
3536 	iovlen = uap->iovcnt * sizeof (struct iovec);
3537 	needfree = NULL;
3538 	if (uap->iovcnt > UIO_SMALLIOV) {
3539 		if (uap->iovcnt > UIO_MAXIOV) {
3540 			error = EINVAL;
3541 			goto done;
3542 		}
3543 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3544 		needfree = iov;
3545 	} else {
3546 		iov = aiov;
3547 	}
3548 	auio.uio_iov = iov;
3549 	auio.uio_iovcnt = uap->iovcnt;
3550 	auio.uio_rw = UIO_READ;
3551 	auio.uio_segflg = UIO_USERSPACE;
3552 	auio.uio_td = nd.nl_td;
3553 	auio.uio_offset = 0;
3554 	if ((error = copyin(uap->iovp, iov, iovlen)))
3555 		goto done;
3556 	auio.uio_resid = 0;
3557 	for (i = 0; i < uap->iovcnt; i++) {
3558 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
3559 			error = EINVAL;
3560 			goto done;
3561 		}
3562 		auio.uio_resid += iov->iov_len;
3563 		iov++;
3564 	}
3565 	cnt = auio.uio_resid;
3566 	error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3567 	cnt -= auio.uio_resid;
3568 	uap->sysmsg_result = cnt;
3569 done:
3570 	vput(vp);
3571 	nlookup_done(&nd);
3572 	if (needfree)
3573 		FREE(needfree, M_IOV);
3574 	return(error);
3575 }
3576 
3577 /*
3578  * Syscall to delete a named extended attribute from a file or directory.
3579  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
3580  */
3581 int
3582 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
3583 {
3584 	char attrname[EXTATTR_MAXNAMELEN];
3585 	struct nlookupdata nd;
3586 	struct vnode *vp;
3587 	int error;
3588 
3589 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3590 	if (error)
3591 		return(error);
3592 
3593 	vp = NULL;
3594 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3595 	if (error == 0)
3596 		error = nlookup(&nd);
3597 	if (error == 0)
3598 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
3599 	if (error) {
3600 		nlookup_done(&nd);
3601 		return (error);
3602 	}
3603 
3604 	error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred);
3605 	vput(vp);
3606 	nlookup_done(&nd);
3607 	return(error);
3608 }
3609 
3610 static int
3611 chroot_visible_mnt(struct mount *mp, struct proc *p)
3612 {
3613 	struct namecache *ncp;
3614 	/*
3615 	 * First check if this file system is below
3616 	 * the chroot path.
3617 	 */
3618 	ncp = mp->mnt_ncp;
3619 	while (ncp != NULL && ncp != p->p_fd->fd_nrdir)
3620 		ncp = ncp->nc_parent;
3621 	if (ncp == NULL) {
3622 		/*
3623 		 * This is not below the chroot path.
3624 		 *
3625 		 * Check if the chroot path is on the same filesystem,
3626 		 * by determing if we have to cross a mount point
3627 		 * before reaching mp->mnt_ncp.
3628 		 */
3629 		ncp = p->p_fd->fd_nrdir;
3630 		while (ncp != NULL && ncp != mp->mnt_ncp) {
3631 			if (ncp->nc_flag & NCF_MOUNTPT) {
3632 				ncp = NULL;
3633 				break;
3634 			}
3635 			ncp = ncp->nc_parent;
3636 		}
3637 	}
3638 	return(ncp != NULL);
3639 }
3640