xref: /dragonfly/sys/kern/vfs_syscalls.c (revision cae2835b)
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. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)vfs_syscalls.c	8.13 (Berkeley) 4/15/94
35  * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $
36  */
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/buf.h>
41 #include <sys/conf.h>
42 #include <sys/sysent.h>
43 #include <sys/malloc.h>
44 #include <sys/mount.h>
45 #include <sys/mountctl.h>
46 #include <sys/sysproto.h>
47 #include <sys/filedesc.h>
48 #include <sys/kernel.h>
49 #include <sys/fcntl.h>
50 #include <sys/file.h>
51 #include <sys/linker.h>
52 #include <sys/stat.h>
53 #include <sys/unistd.h>
54 #include <sys/vnode.h>
55 #include <sys/proc.h>
56 #include <sys/priv.h>
57 #include <sys/jail.h>
58 #include <sys/namei.h>
59 #include <sys/nlookup.h>
60 #include <sys/dirent.h>
61 #include <sys/extattr.h>
62 #include <sys/spinlock.h>
63 #include <sys/kern_syscall.h>
64 #include <sys/objcache.h>
65 #include <sys/sysctl.h>
66 
67 #include <sys/buf2.h>
68 #include <sys/file2.h>
69 #include <sys/spinlock2.h>
70 #include <sys/mplock2.h>
71 
72 #include <vm/vm.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_page.h>
75 
76 #include <machine/limits.h>
77 #include <machine/stdarg.h>
78 
79 static void mount_warning(struct mount *mp, const char *ctl, ...)
80 		__printflike(2, 3);
81 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb);
82 static int checkvp_chdir (struct vnode *vn, struct thread *td);
83 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch);
84 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
85 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
86 static int getutimes (struct timeval *, struct timespec *);
87 static int getutimens (const struct timespec *, struct timespec *, int *);
88 static int setfown (struct mount *, struct vnode *, uid_t, gid_t);
89 static int setfmode (struct vnode *, int);
90 static int setfflags (struct vnode *, int);
91 static int setutimes (struct vnode *, struct vattr *,
92 			const struct timespec *, int);
93 static int	usermount = 0;	/* if 1, non-root can mount fs. */
94 
95 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
96     "Allow non-root users to mount filesystems");
97 
98 /*
99  * Virtual File System System Calls
100  */
101 
102 /*
103  * Mount a file system.
104  *
105  * mount_args(char *type, char *path, int flags, caddr_t data)
106  *
107  * MPALMOSTSAFE
108  */
109 int
110 sys_mount(struct mount_args *uap)
111 {
112 	struct thread *td = curthread;
113 	struct vnode *vp;
114 	struct nchandle nch;
115 	struct mount *mp, *nullmp;
116 	struct vfsconf *vfsp;
117 	int error, flag = 0, flag2 = 0;
118 	int hasmount;
119 	struct vattr va;
120 	struct nlookupdata nd;
121 	char fstypename[MFSNAMELEN];
122 	struct ucred *cred;
123 
124 	cred = td->td_ucred;
125 	if (jailed(cred)) {
126 		error = EPERM;
127 		goto done;
128 	}
129 	if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
130 		goto done;
131 
132 	/*
133 	 * Do not allow NFS export by non-root users.
134 	 */
135 	if (uap->flags & MNT_EXPORTED) {
136 		error = priv_check(td, PRIV_ROOT);
137 		if (error)
138 			goto done;
139 	}
140 	/*
141 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
142 	 */
143 	if (priv_check(td, PRIV_ROOT))
144 		uap->flags |= MNT_NOSUID | MNT_NODEV;
145 
146 	/*
147 	 * Lookup the requested path and extract the nch and vnode.
148 	 */
149 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
150 	if (error == 0) {
151 		if ((error = nlookup(&nd)) == 0) {
152 			if (nd.nl_nch.ncp->nc_vp == NULL)
153 				error = ENOENT;
154 		}
155 	}
156 	if (error) {
157 		nlookup_done(&nd);
158 		goto done;
159 	}
160 
161 	/*
162 	 * If the target filesystem is resolved via a nullfs mount, then
163 	 * nd.nl_nch.mount will be pointing to the nullfs mount structure
164 	 * instead of the target file system. We need it in case we are
165 	 * doing an update.
166 	 */
167 	nullmp = nd.nl_nch.mount;
168 
169 	/*
170 	 * Extract the locked+refd ncp and cleanup the nd structure
171 	 */
172 	nch = nd.nl_nch;
173 	cache_zero(&nd.nl_nch);
174 	nlookup_done(&nd);
175 
176 	if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
177 	    (mp = cache_findmount(&nch)) != NULL) {
178 		cache_dropmount(mp);
179 		hasmount = 1;
180 	} else {
181 		hasmount = 0;
182 	}
183 
184 
185 	/*
186 	 * now we have the locked ref'd nch and unreferenced vnode.
187 	 */
188 	vp = nch.ncp->nc_vp;
189 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
190 		cache_put(&nch);
191 		goto done;
192 	}
193 	cache_unlock(&nch);
194 
195 	/*
196 	 * Extract the file system type. We need to know this early, to take
197 	 * appropriate actions if we are dealing with a nullfs.
198 	 */
199         if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
200                 cache_drop(&nch);
201                 vput(vp);
202 		goto done;
203         }
204 
205 	/*
206 	 * Now we have an unlocked ref'd nch and a locked ref'd vp
207 	 */
208 	if (uap->flags & MNT_UPDATE) {
209 		if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
210 			cache_drop(&nch);
211 			vput(vp);
212 			error = EINVAL;
213 			goto done;
214 		}
215 
216 		if (strncmp(fstypename, "null", 5) == 0) {
217 			KKASSERT(nullmp);
218 			mp = nullmp;
219 		} else {
220 			mp = vp->v_mount;
221 		}
222 
223 		flag = mp->mnt_flag;
224 		flag2 = mp->mnt_kern_flag;
225 		/*
226 		 * We only allow the filesystem to be reloaded if it
227 		 * is currently mounted read-only.
228 		 */
229 		if ((uap->flags & MNT_RELOAD) &&
230 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
231 			cache_drop(&nch);
232 			vput(vp);
233 			error = EOPNOTSUPP;	/* Needs translation */
234 			goto done;
235 		}
236 		/*
237 		 * Only root, or the user that did the original mount is
238 		 * permitted to update it.
239 		 */
240 		if (mp->mnt_stat.f_owner != cred->cr_uid &&
241 		    (error = priv_check(td, PRIV_ROOT))) {
242 			cache_drop(&nch);
243 			vput(vp);
244 			goto done;
245 		}
246 		if (vfs_busy(mp, LK_NOWAIT)) {
247 			cache_drop(&nch);
248 			vput(vp);
249 			error = EBUSY;
250 			goto done;
251 		}
252 		if (hasmount) {
253 			cache_drop(&nch);
254 			vfs_unbusy(mp);
255 			vput(vp);
256 			error = EBUSY;
257 			goto done;
258 		}
259 		mp->mnt_flag |=
260 		    uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
261 		lwkt_gettoken(&mp->mnt_token);
262 		vn_unlock(vp);
263 		goto update;
264 	}
265 
266 	/*
267 	 * If the user is not root, ensure that they own the directory
268 	 * onto which we are attempting to mount.
269 	 */
270 	if ((error = VOP_GETATTR(vp, &va)) ||
271 	    (va.va_uid != cred->cr_uid &&
272 	     (error = priv_check(td, PRIV_ROOT)))) {
273 		cache_drop(&nch);
274 		vput(vp);
275 		goto done;
276 	}
277 	if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
278 		cache_drop(&nch);
279 		vput(vp);
280 		goto done;
281 	}
282 	if (vp->v_type != VDIR) {
283 		cache_drop(&nch);
284 		vput(vp);
285 		error = ENOTDIR;
286 		goto done;
287 	}
288 	if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) {
289 		cache_drop(&nch);
290 		vput(vp);
291 		error = EPERM;
292 		goto done;
293 	}
294 	vfsp = vfsconf_find_by_name(fstypename);
295 	if (vfsp == NULL) {
296 		linker_file_t lf;
297 
298 		/* Only load modules for root (very important!) */
299 		if ((error = priv_check(td, PRIV_ROOT)) != 0) {
300 			cache_drop(&nch);
301 			vput(vp);
302 			goto done;
303 		}
304 		error = linker_load_file(fstypename, &lf);
305 		if (error || lf == NULL) {
306 			cache_drop(&nch);
307 			vput(vp);
308 			if (lf == NULL)
309 				error = ENODEV;
310 			goto done;
311 		}
312 		lf->userrefs++;
313 		/* lookup again, see if the VFS was loaded */
314 		vfsp = vfsconf_find_by_name(fstypename);
315 		if (vfsp == NULL) {
316 			lf->userrefs--;
317 			linker_file_unload(lf);
318 			cache_drop(&nch);
319 			vput(vp);
320 			error = ENODEV;
321 			goto done;
322 		}
323 	}
324 	if (hasmount) {
325 		cache_drop(&nch);
326 		vput(vp);
327 		error = EBUSY;
328 		goto done;
329 	}
330 
331 	/*
332 	 * Allocate and initialize the filesystem.
333 	 */
334 	mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
335 	mount_init(mp);
336 	vfs_busy(mp, LK_NOWAIT);
337 	mp->mnt_op = vfsp->vfc_vfsops;
338 	mp->mnt_vfc = vfsp;
339 	vfsp->vfc_refcount++;
340 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
341 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
342 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
343 	mp->mnt_stat.f_owner = cred->cr_uid;
344 	lwkt_gettoken(&mp->mnt_token);
345 	vn_unlock(vp);
346 update:
347 	/*
348 	 * (per-mount token acquired at this point)
349 	 *
350 	 * Set the mount level flags.
351 	 */
352 	if (uap->flags & MNT_RDONLY)
353 		mp->mnt_flag |= MNT_RDONLY;
354 	else if (mp->mnt_flag & MNT_RDONLY)
355 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
356 	mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
357 	    MNT_SYNCHRONOUS | MNT_ASYNC | MNT_NOATIME |
358 	    MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM |
359 	    MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR |
360 	    MNT_AUTOMOUNTED);
361 	mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
362 	    MNT_NODEV | MNT_SYNCHRONOUS | MNT_ASYNC | MNT_FORCE |
363 	    MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM |
364 	    MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR |
365 	    MNT_AUTOMOUNTED);
366 	/*
367 	 * Mount the filesystem.
368 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
369 	 * get.
370 	 */
371 	error = VFS_MOUNT(mp, uap->path, uap->data, cred);
372 	if (mp->mnt_flag & MNT_UPDATE) {
373 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
374 			mp->mnt_flag &= ~MNT_RDONLY;
375 		mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
376 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
377 		if (error) {
378 			mp->mnt_flag = flag;
379 			mp->mnt_kern_flag = flag2;
380 		}
381 		lwkt_reltoken(&mp->mnt_token);
382 		vfs_unbusy(mp);
383 		vrele(vp);
384 		cache_drop(&nch);
385 		goto done;
386 	}
387 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
388 
389 	/*
390 	 * Put the new filesystem on the mount list after root.  The mount
391 	 * point gets its own mnt_ncmountpt (unless the VFS already set one
392 	 * up) which represents the root of the mount.  The lookup code
393 	 * detects the mount point going forward and checks the root of
394 	 * the mount going backwards.
395 	 *
396 	 * It is not necessary to invalidate or purge the vnode underneath
397 	 * because elements under the mount will be given their own glue
398 	 * namecache record.
399 	 */
400 	if (!error) {
401 		if (mp->mnt_ncmountpt.ncp == NULL) {
402 			/*
403 			 * allocate, then unlock, but leave the ref intact
404 			 */
405 			cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
406 			cache_unlock(&mp->mnt_ncmountpt);
407 		}
408 		vn_unlock(vp);
409 		mp->mnt_ncmounton = nch;		/* inherits ref */
410 		cache_lock(&nch);
411 		nch.ncp->nc_flag |= NCF_ISMOUNTPT;
412 		cache_unlock(&nch);
413 		cache_ismounting(mp);
414 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
415 
416 		mountlist_insert(mp, MNTINS_LAST);
417 		vn_unlock(vp);
418 		checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
419 		error = vfs_allocate_syncvnode(mp);
420 		lwkt_reltoken(&mp->mnt_token);
421 		vfs_unbusy(mp);
422 		error = VFS_START(mp, 0);
423 		vrele(vp);
424 		KNOTE(&fs_klist, VQ_MOUNT);
425 	} else {
426 		vn_syncer_thr_stop(mp);
427 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
428 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
429 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
430 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
431 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
432 		mp->mnt_vfc->vfc_refcount--;
433 		lwkt_reltoken(&mp->mnt_token);
434 		vfs_unbusy(mp);
435 		kfree(mp, M_MOUNT);
436 		cache_drop(&nch);
437 		vput(vp);
438 	}
439 done:
440 	return (error);
441 }
442 
443 /*
444  * Scan all active processes to see if any of them have a current
445  * or root directory onto which the new filesystem has just been
446  * mounted. If so, replace them with the new mount point.
447  *
448  * Both old_nch and new_nch are ref'd on call but not locked.
449  * new_nch must be temporarily locked so it can be associated with the
450  * vnode representing the root of the mount point.
451  */
452 struct checkdirs_info {
453 	struct nchandle old_nch;
454 	struct nchandle new_nch;
455 	struct vnode *old_vp;
456 	struct vnode *new_vp;
457 };
458 
459 static int checkdirs_callback(struct proc *p, void *data);
460 
461 static void
462 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
463 {
464 	struct checkdirs_info info;
465 	struct vnode *olddp;
466 	struct vnode *newdp;
467 	struct mount *mp;
468 
469 	/*
470 	 * If the old mount point's vnode has a usecount of 1, it is not
471 	 * being held as a descriptor anywhere.
472 	 */
473 	olddp = old_nch->ncp->nc_vp;
474 	if (olddp == NULL || VREFCNT(olddp) == 1)
475 		return;
476 
477 	/*
478 	 * Force the root vnode of the new mount point to be resolved
479 	 * so we can update any matching processes.
480 	 */
481 	mp = new_nch->mount;
482 	if (VFS_ROOT(mp, &newdp))
483 		panic("mount: lost mount");
484 	vn_unlock(newdp);
485 	cache_lock(new_nch);
486 	vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
487 	cache_setunresolved(new_nch);
488 	cache_setvp(new_nch, newdp);
489 	cache_unlock(new_nch);
490 
491 	/*
492 	 * Special handling of the root node
493 	 */
494 	if (rootvnode == olddp) {
495 		vref(newdp);
496 		vfs_cache_setroot(newdp, cache_hold(new_nch));
497 	}
498 
499 	/*
500 	 * Pass newdp separately so the callback does not have to access
501 	 * it via new_nch->ncp->nc_vp.
502 	 */
503 	info.old_nch = *old_nch;
504 	info.new_nch = *new_nch;
505 	info.new_vp = newdp;
506 	allproc_scan(checkdirs_callback, &info);
507 	vput(newdp);
508 }
509 
510 /*
511  * NOTE: callback is not MP safe because the scanned process's filedesc
512  * structure can be ripped out from under us, amoung other things.
513  */
514 static int
515 checkdirs_callback(struct proc *p, void *data)
516 {
517 	struct checkdirs_info *info = data;
518 	struct filedesc *fdp;
519 	struct nchandle ncdrop1;
520 	struct nchandle ncdrop2;
521 	struct vnode *vprele1;
522 	struct vnode *vprele2;
523 
524 	if ((fdp = p->p_fd) != NULL) {
525 		cache_zero(&ncdrop1);
526 		cache_zero(&ncdrop2);
527 		vprele1 = NULL;
528 		vprele2 = NULL;
529 
530 		/*
531 		 * MPUNSAFE - XXX fdp can be pulled out from under a
532 		 * foreign process.
533 		 *
534 		 * A shared filedesc is ok, we don't have to copy it
535 		 * because we are making this change globally.
536 		 */
537 		spin_lock(&fdp->fd_spin);
538 		if (fdp->fd_ncdir.mount == info->old_nch.mount &&
539 		    fdp->fd_ncdir.ncp == info->old_nch.ncp) {
540 			vprele1 = fdp->fd_cdir;
541 			vref(info->new_vp);
542 			fdp->fd_cdir = info->new_vp;
543 			ncdrop1 = fdp->fd_ncdir;
544 			cache_copy(&info->new_nch, &fdp->fd_ncdir);
545 		}
546 		if (fdp->fd_nrdir.mount == info->old_nch.mount &&
547 		    fdp->fd_nrdir.ncp == info->old_nch.ncp) {
548 			vprele2 = fdp->fd_rdir;
549 			vref(info->new_vp);
550 			fdp->fd_rdir = info->new_vp;
551 			ncdrop2 = fdp->fd_nrdir;
552 			cache_copy(&info->new_nch, &fdp->fd_nrdir);
553 		}
554 		spin_unlock(&fdp->fd_spin);
555 		if (ncdrop1.ncp)
556 			cache_drop(&ncdrop1);
557 		if (ncdrop2.ncp)
558 			cache_drop(&ncdrop2);
559 		if (vprele1)
560 			vrele(vprele1);
561 		if (vprele2)
562 			vrele(vprele2);
563 	}
564 	return(0);
565 }
566 
567 /*
568  * Unmount a file system.
569  *
570  * Note: unmount takes a path to the vnode mounted on as argument,
571  * not special file (as before).
572  *
573  * umount_args(char *path, int flags)
574  *
575  * MPALMOSTSAFE
576  */
577 int
578 sys_unmount(struct unmount_args *uap)
579 {
580 	struct thread *td = curthread;
581 	struct proc *p __debugvar = td->td_proc;
582 	struct mount *mp = NULL;
583 	struct nlookupdata nd;
584 	int error;
585 
586 	KKASSERT(p);
587 	get_mplock();
588 	if (td->td_ucred->cr_prison != NULL) {
589 		error = EPERM;
590 		goto done;
591 	}
592 	if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
593 		goto done;
594 
595 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
596 	if (error == 0)
597 		error = nlookup(&nd);
598 	if (error)
599 		goto out;
600 
601 	mp = nd.nl_nch.mount;
602 
603 	/*
604 	 * Only root, or the user that did the original mount is
605 	 * permitted to unmount this filesystem.
606 	 */
607 	if ((mp->mnt_stat.f_owner != td->td_ucred->cr_uid) &&
608 	    (error = priv_check(td, PRIV_ROOT)))
609 		goto out;
610 
611 	/*
612 	 * Don't allow unmounting the root file system.
613 	 */
614 	if (mp->mnt_flag & MNT_ROOTFS) {
615 		error = EINVAL;
616 		goto out;
617 	}
618 
619 	/*
620 	 * Must be the root of the filesystem
621 	 */
622 	if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
623 		error = EINVAL;
624 		goto out;
625 	}
626 
627 out:
628 	nlookup_done(&nd);
629 	if (error == 0)
630 		error = dounmount(mp, uap->flags);
631 done:
632 	rel_mplock();
633 	return (error);
634 }
635 
636 /*
637  * Do the actual file system unmount.
638  */
639 static int
640 dounmount_interlock(struct mount *mp)
641 {
642 	if (mp->mnt_kern_flag & MNTK_UNMOUNT)
643 		return (EBUSY);
644 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
645 	return(0);
646 }
647 
648 static int
649 unmount_allproc_cb(struct proc *p, void *arg)
650 {
651 	struct mount *mp;
652 
653 	if (p->p_textnch.ncp == NULL)
654 		return 0;
655 
656 	mp = (struct mount *)arg;
657 	if (p->p_textnch.mount == mp)
658 		cache_drop(&p->p_textnch);
659 
660 	return 0;
661 }
662 
663 int
664 dounmount(struct mount *mp, int flags)
665 {
666 	struct namecache *ncp;
667 	struct nchandle nch;
668 	struct vnode *vp;
669 	int error;
670 	int async_flag;
671 	int lflags;
672 	int freeok = 1;
673 	int retry;
674 
675 	lwkt_gettoken(&mp->mnt_token);
676 	/*
677 	 * Exclusive access for unmounting purposes
678 	 */
679 	if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
680 		goto out;
681 
682 	/*
683 	 * Allow filesystems to detect that a forced unmount is in progress.
684 	 */
685 	if (flags & MNT_FORCE)
686 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
687 	lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_TIMELOCK);
688 	error = lockmgr(&mp->mnt_lock, lflags);
689 	if (error) {
690 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
691 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
692 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
693 			wakeup(mp);
694 		}
695 		goto out;
696 	}
697 
698 	if (mp->mnt_flag & MNT_EXPUBLIC)
699 		vfs_setpublicfs(NULL, NULL, NULL);
700 
701 	vfs_msync(mp, MNT_WAIT);
702 	async_flag = mp->mnt_flag & MNT_ASYNC;
703 	mp->mnt_flag &=~ MNT_ASYNC;
704 
705 	/*
706 	 * If this filesystem isn't aliasing other filesystems,
707 	 * try to invalidate any remaining namecache entries and
708 	 * check the count afterwords.
709 	 */
710 	if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
711 		cache_lock(&mp->mnt_ncmountpt);
712 		cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
713 		cache_unlock(&mp->mnt_ncmountpt);
714 
715 		if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
716 		    (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
717 			allproc_scan(&unmount_allproc_cb, mp);
718 		}
719 
720 		if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
721 		    (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
722 
723 			if ((flags & MNT_FORCE) == 0) {
724 				error = EBUSY;
725 				mount_warning(mp, "Cannot unmount: "
726 						  "%d namecache "
727 						  "references still "
728 						  "present",
729 						  ncp->nc_refs - 1);
730 			} else {
731 				mount_warning(mp, "Forced unmount: "
732 						  "%d namecache "
733 						  "references still "
734 						  "present",
735 						  ncp->nc_refs - 1);
736 				freeok = 0;
737 			}
738 		}
739 	}
740 
741 	/*
742 	 * Decomission our special mnt_syncer vnode.  This also stops
743 	 * the vnlru code.  If we are unable to unmount we recommission
744 	 * the vnode.
745 	 *
746 	 * Then sync the filesystem.
747 	 */
748 	if ((vp = mp->mnt_syncer) != NULL) {
749 		mp->mnt_syncer = NULL;
750 		atomic_set_int(&vp->v_refcnt, VREF_FINALIZE);
751 		vrele(vp);
752 	}
753 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
754 		VFS_SYNC(mp, MNT_WAIT);
755 
756 	/*
757 	 * nchandle records ref the mount structure.  Expect a count of 1
758 	 * (our mount->mnt_ncmountpt).
759 	 *
760 	 * Scans can get temporary refs on a mountpoint (thought really
761 	 * heavy duty stuff like cache_findmount() do not).
762 	 */
763 	for (retry = 0; retry < 10 && mp->mnt_refs != 1; ++retry) {
764 		cache_unmounting(mp);
765 		tsleep(&mp->mnt_refs, 0, "mntbsy", hz / 10 + 1);
766 	}
767 	if (mp->mnt_refs != 1) {
768 		if ((flags & MNT_FORCE) == 0) {
769 			mount_warning(mp, "Cannot unmount: "
770 					  "%d mount refs still present",
771 					  mp->mnt_refs);
772 			error = EBUSY;
773 		} else {
774 			mount_warning(mp, "Forced unmount: "
775 					  "%d mount refs still present",
776 					  mp->mnt_refs);
777 			freeok = 0;
778 		}
779 	}
780 
781 	/*
782 	 * So far so good, sync the filesystem once more and
783 	 * call the VFS unmount code if the sync succeeds.
784 	 */
785 	if (error == 0) {
786 		if (mp->mnt_flag & MNT_RDONLY) {
787 			error = VFS_UNMOUNT(mp, flags);
788 		} else {
789 			error = VFS_SYNC(mp, MNT_WAIT);
790 			if ((error == 0) ||
791 			    (error == EOPNOTSUPP) || /* No sync */
792 			    (flags & MNT_FORCE)) {
793 				error = VFS_UNMOUNT(mp, flags);
794 			}
795 		}
796 	}
797 
798 	/*
799 	 * If an error occurred we can still recover, restoring the
800 	 * syncer vnode and misc flags.
801 	 */
802 	if (error) {
803 		if (mp->mnt_syncer == NULL)
804 			vfs_allocate_syncvnode(mp);
805 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
806 		mp->mnt_flag |= async_flag;
807 		lockmgr(&mp->mnt_lock, LK_RELEASE);
808 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
809 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
810 			wakeup(mp);
811 		}
812 		goto out;
813 	}
814 	/*
815 	 * Clean up any journals still associated with the mount after
816 	 * filesystem activity has ceased.
817 	 */
818 	journal_remove_all_journals(mp,
819 	    ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
820 
821 	mountlist_remove(mp);
822 
823 	/*
824 	 * Remove any installed vnode ops here so the individual VFSs don't
825 	 * have to.
826 	 */
827 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
828 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
829 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
830 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
831 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
832 
833 	if (mp->mnt_ncmountpt.ncp != NULL) {
834 		nch = mp->mnt_ncmountpt;
835 		cache_zero(&mp->mnt_ncmountpt);
836 		cache_clrmountpt(&nch);
837 		cache_drop(&nch);
838 	}
839 	if (mp->mnt_ncmounton.ncp != NULL) {
840 		cache_unmounting(mp);
841 		nch = mp->mnt_ncmounton;
842 		cache_zero(&mp->mnt_ncmounton);
843 		cache_clrmountpt(&nch);
844 		cache_drop(&nch);
845 	}
846 
847 	mp->mnt_vfc->vfc_refcount--;
848 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
849 		panic("unmount: dangling vnode");
850 	lockmgr(&mp->mnt_lock, LK_RELEASE);
851 	if (mp->mnt_kern_flag & MNTK_MWAIT) {
852 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
853 		wakeup(mp);
854 	}
855 
856 	/*
857 	 * If we reach here and freeok != 0 we must free the mount.
858 	 * If refs > 1 cycle and wait, just in case someone tried
859 	 * to busy the mount after we decided to do the unmount.
860 	 */
861 	if (freeok) {
862 		while (mp->mnt_refs > 1) {
863 			cache_unmounting(mp);
864 			wakeup(mp);
865 			tsleep(&mp->mnt_refs, 0, "umntrwait", hz / 10 + 1);
866 		}
867 		lwkt_reltoken(&mp->mnt_token);
868 		kfree(mp, M_MOUNT);
869 		mp = NULL;
870 	}
871 	error = 0;
872 	KNOTE(&fs_klist, VQ_UNMOUNT);
873 out:
874 	if (mp)
875 		lwkt_reltoken(&mp->mnt_token);
876 	return (error);
877 }
878 
879 static
880 void
881 mount_warning(struct mount *mp, const char *ctl, ...)
882 {
883 	char *ptr;
884 	char *buf;
885 	__va_list va;
886 
887 	__va_start(va, ctl);
888 	if (cache_fullpath(NULL, &mp->mnt_ncmounton, NULL,
889 			   &ptr, &buf, 0) == 0) {
890 		kprintf("unmount(%s): ", ptr);
891 		kvprintf(ctl, va);
892 		kprintf("\n");
893 		kfree(buf, M_TEMP);
894 	} else {
895 		kprintf("unmount(%p", mp);
896 		if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
897 			kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
898 		kprintf("): ");
899 		kvprintf(ctl, va);
900 		kprintf("\n");
901 	}
902 	__va_end(va);
903 }
904 
905 /*
906  * Shim cache_fullpath() to handle the case where a process is chrooted into
907  * a subdirectory of a mount.  In this case if the root mount matches the
908  * process root directory's mount we have to specify the process's root
909  * directory instead of the mount point, because the mount point might
910  * be above the root directory.
911  */
912 static
913 int
914 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
915 {
916 	struct nchandle *nch;
917 
918 	if (p && p->p_fd->fd_nrdir.mount == mp)
919 		nch = &p->p_fd->fd_nrdir;
920 	else
921 		nch = &mp->mnt_ncmountpt;
922 	return(cache_fullpath(p, nch, NULL, rb, fb, 0));
923 }
924 
925 /*
926  * Sync each mounted filesystem.
927  */
928 
929 #ifdef DEBUG
930 static int syncprt = 0;
931 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
932 #endif /* DEBUG */
933 
934 static int sync_callback(struct mount *mp, void *data);
935 
936 int
937 sys_sync(struct sync_args *uap)
938 {
939 	mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
940 	return (0);
941 }
942 
943 static
944 int
945 sync_callback(struct mount *mp, void *data __unused)
946 {
947 	int asyncflag;
948 
949 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
950 		asyncflag = mp->mnt_flag & MNT_ASYNC;
951 		mp->mnt_flag &= ~MNT_ASYNC;
952 		vfs_msync(mp, MNT_NOWAIT);
953 		VFS_SYNC(mp, MNT_NOWAIT);
954 		mp->mnt_flag |= asyncflag;
955 	}
956 	return(0);
957 }
958 
959 /* XXX PRISON: could be per prison flag */
960 static int prison_quotas;
961 #if 0
962 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
963 #endif
964 
965 /*
966  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
967  *
968  * Change filesystem quotas.
969  *
970  * MPALMOSTSAFE
971  */
972 int
973 sys_quotactl(struct quotactl_args *uap)
974 {
975 	struct nlookupdata nd;
976 	struct thread *td;
977 	struct mount *mp;
978 	int error;
979 
980 	get_mplock();
981 	td = curthread;
982 	if (td->td_ucred->cr_prison && !prison_quotas) {
983 		error = EPERM;
984 		goto done;
985 	}
986 
987 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
988 	if (error == 0)
989 		error = nlookup(&nd);
990 	if (error == 0) {
991 		mp = nd.nl_nch.mount;
992 		error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
993 				    uap->arg, nd.nl_cred);
994 	}
995 	nlookup_done(&nd);
996 done:
997 	rel_mplock();
998 	return (error);
999 }
1000 
1001 /*
1002  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
1003  *		void *buf, int buflen)
1004  *
1005  * This function operates on a mount point and executes the specified
1006  * operation using the specified control data, and possibly returns data.
1007  *
1008  * The actual number of bytes stored in the result buffer is returned, 0
1009  * if none, otherwise an error is returned.
1010  *
1011  * MPALMOSTSAFE
1012  */
1013 int
1014 sys_mountctl(struct mountctl_args *uap)
1015 {
1016 	struct thread *td = curthread;
1017 	struct proc *p = td->td_proc;
1018 	struct file *fp;
1019 	void *ctl = NULL;
1020 	void *buf = NULL;
1021 	char *path = NULL;
1022 	int error;
1023 
1024 	/*
1025 	 * Sanity and permissions checks.  We must be root.
1026 	 */
1027 	KKASSERT(p);
1028 	if (td->td_ucred->cr_prison != NULL)
1029 		return (EPERM);
1030 	if ((uap->op != MOUNTCTL_MOUNTFLAGS) &&
1031 	    (error = priv_check(td, PRIV_ROOT)) != 0)
1032 		return (error);
1033 
1034 	/*
1035 	 * Argument length checks
1036 	 */
1037 	if (uap->ctllen < 0 || uap->ctllen > 1024)
1038 		return (EINVAL);
1039 	if (uap->buflen < 0 || uap->buflen > 16 * 1024)
1040 		return (EINVAL);
1041 	if (uap->path == NULL)
1042 		return (EINVAL);
1043 
1044 	/*
1045 	 * Allocate the necessary buffers and copyin data
1046 	 */
1047 	path = objcache_get(namei_oc, M_WAITOK);
1048 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
1049 	if (error)
1050 		goto done;
1051 
1052 	if (uap->ctllen) {
1053 		ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
1054 		error = copyin(uap->ctl, ctl, uap->ctllen);
1055 		if (error)
1056 			goto done;
1057 	}
1058 	if (uap->buflen)
1059 		buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
1060 
1061 	/*
1062 	 * Validate the descriptor
1063 	 */
1064 	if (uap->fd >= 0) {
1065 		fp = holdfp(p->p_fd, uap->fd, -1);
1066 		if (fp == NULL) {
1067 			error = EBADF;
1068 			goto done;
1069 		}
1070 	} else {
1071 		fp = NULL;
1072 	}
1073 
1074 	/*
1075 	 * Execute the internal kernel function and clean up.
1076 	 */
1077 	get_mplock();
1078 	error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
1079 	rel_mplock();
1080 	if (fp)
1081 		fdrop(fp);
1082 	if (error == 0 && uap->sysmsg_result > 0)
1083 		error = copyout(buf, uap->buf, uap->sysmsg_result);
1084 done:
1085 	if (path)
1086 		objcache_put(namei_oc, path);
1087 	if (ctl)
1088 		kfree(ctl, M_TEMP);
1089 	if (buf)
1090 		kfree(buf, M_TEMP);
1091 	return (error);
1092 }
1093 
1094 /*
1095  * Execute a mount control operation by resolving the path to a mount point
1096  * and calling vop_mountctl().
1097  *
1098  * Use the mount point from the nch instead of the vnode so nullfs mounts
1099  * can properly spike the VOP.
1100  */
1101 int
1102 kern_mountctl(const char *path, int op, struct file *fp,
1103 		const void *ctl, int ctllen,
1104 		void *buf, int buflen, int *res)
1105 {
1106 	struct vnode *vp;
1107 	struct nlookupdata nd;
1108 	struct nchandle nch;
1109 	struct mount *mp;
1110 	int error;
1111 
1112 	*res = 0;
1113 	vp = NULL;
1114 	error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
1115 	if (error)
1116 		return (error);
1117 	error = nlookup(&nd);
1118 	if (error) {
1119 		nlookup_done(&nd);
1120 		return (error);
1121 	}
1122 	error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
1123 	if (error) {
1124 		nlookup_done(&nd);
1125 		return (error);
1126 	}
1127 
1128 	/*
1129 	 * Yes, all this is needed to use the nch.mount below, because
1130 	 * we must maintain a ref on the mount to avoid ripouts (e.g.
1131 	 * due to heavy mount/unmount use by synth or poudriere).
1132 	 */
1133 	nch = nd.nl_nch;
1134 	cache_zero(&nd.nl_nch);
1135 	cache_unlock(&nch);
1136 	nlookup_done(&nd);
1137 	vn_unlock(vp);
1138 
1139 	mp = nch.mount;
1140 
1141 	/*
1142 	 * Must be the root of the filesystem
1143 	 */
1144 	if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
1145 		cache_drop(&nch);
1146 		vrele(vp);
1147 		return (EINVAL);
1148 	}
1149 	if (mp == NULL || mp->mnt_kern_flag & MNTK_UNMOUNT) {
1150 		kprintf("kern_mountctl: Warning, \"%s\" racing unmount\n",
1151 			path);
1152 		cache_drop(&nch);
1153 		vrele(vp);
1154 		return (EINVAL);
1155 	}
1156 	error = vop_mountctl(mp->mnt_vn_use_ops, vp, op, fp, ctl, ctllen,
1157 			     buf, buflen, res);
1158 	vrele(vp);
1159 	cache_drop(&nch);
1160 
1161 	return (error);
1162 }
1163 
1164 int
1165 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1166 {
1167 	struct thread *td = curthread;
1168 	struct proc *p = td->td_proc;
1169 	struct mount *mp;
1170 	struct statfs *sp;
1171 	char *fullpath, *freepath;
1172 	int error;
1173 
1174 	if ((error = nlookup(nd)) != 0)
1175 		return (error);
1176 	mp = nd->nl_nch.mount;
1177 	sp = &mp->mnt_stat;
1178 	if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1179 		return (error);
1180 
1181 	error = mount_path(p, mp, &fullpath, &freepath);
1182 	if (error)
1183 		return(error);
1184 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1185 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1186 	kfree(freepath, M_TEMP);
1187 
1188 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1189 	bcopy(sp, buf, sizeof(*buf));
1190 	/* Only root should have access to the fsid's. */
1191 	if (priv_check(td, PRIV_ROOT))
1192 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1193 	return (0);
1194 }
1195 
1196 /*
1197  * statfs_args(char *path, struct statfs *buf)
1198  *
1199  * Get filesystem statistics.
1200  */
1201 int
1202 sys_statfs(struct statfs_args *uap)
1203 {
1204 	struct nlookupdata nd;
1205 	struct statfs buf;
1206 	int error;
1207 
1208 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1209 	if (error == 0)
1210 		error = kern_statfs(&nd, &buf);
1211 	nlookup_done(&nd);
1212 	if (error == 0)
1213 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1214 	return (error);
1215 }
1216 
1217 int
1218 kern_fstatfs(int fd, struct statfs *buf)
1219 {
1220 	struct thread *td = curthread;
1221 	struct proc *p = td->td_proc;
1222 	struct file *fp;
1223 	struct mount *mp;
1224 	struct statfs *sp;
1225 	char *fullpath, *freepath;
1226 	int error;
1227 
1228 	KKASSERT(p);
1229 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1230 		return (error);
1231 
1232 	/*
1233 	 * Try to use mount info from any overlays rather than the
1234 	 * mount info for the underlying vnode, otherwise we will
1235 	 * fail when operating on null-mounted paths inside a chroot.
1236 	 */
1237 	if ((mp = fp->f_nchandle.mount) == NULL)
1238 		mp = ((struct vnode *)fp->f_data)->v_mount;
1239 	if (mp == NULL) {
1240 		error = EBADF;
1241 		goto done;
1242 	}
1243 	if (fp->f_cred == NULL) {
1244 		error = EINVAL;
1245 		goto done;
1246 	}
1247 	sp = &mp->mnt_stat;
1248 	if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1249 		goto done;
1250 
1251 	if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1252 		goto done;
1253 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1254 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1255 	kfree(freepath, M_TEMP);
1256 
1257 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1258 	bcopy(sp, buf, sizeof(*buf));
1259 
1260 	/* Only root should have access to the fsid's. */
1261 	if (priv_check(td, PRIV_ROOT))
1262 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1263 	error = 0;
1264 done:
1265 	fdrop(fp);
1266 	return (error);
1267 }
1268 
1269 /*
1270  * fstatfs_args(int fd, struct statfs *buf)
1271  *
1272  * Get filesystem statistics.
1273  */
1274 int
1275 sys_fstatfs(struct fstatfs_args *uap)
1276 {
1277 	struct statfs buf;
1278 	int error;
1279 
1280 	error = kern_fstatfs(uap->fd, &buf);
1281 
1282 	if (error == 0)
1283 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1284 	return (error);
1285 }
1286 
1287 int
1288 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1289 {
1290 	struct mount *mp;
1291 	struct statvfs *sp;
1292 	int error;
1293 
1294 	if ((error = nlookup(nd)) != 0)
1295 		return (error);
1296 	mp = nd->nl_nch.mount;
1297 	sp = &mp->mnt_vstat;
1298 	if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1299 		return (error);
1300 
1301 	sp->f_flag = 0;
1302 	if (mp->mnt_flag & MNT_RDONLY)
1303 		sp->f_flag |= ST_RDONLY;
1304 	if (mp->mnt_flag & MNT_NOSUID)
1305 		sp->f_flag |= ST_NOSUID;
1306 	bcopy(sp, buf, sizeof(*buf));
1307 	return (0);
1308 }
1309 
1310 /*
1311  * statfs_args(char *path, struct statfs *buf)
1312  *
1313  * Get filesystem statistics.
1314  */
1315 int
1316 sys_statvfs(struct statvfs_args *uap)
1317 {
1318 	struct nlookupdata nd;
1319 	struct statvfs buf;
1320 	int error;
1321 
1322 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1323 	if (error == 0)
1324 		error = kern_statvfs(&nd, &buf);
1325 	nlookup_done(&nd);
1326 	if (error == 0)
1327 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1328 	return (error);
1329 }
1330 
1331 int
1332 kern_fstatvfs(int fd, struct statvfs *buf)
1333 {
1334 	struct thread *td = curthread;
1335 	struct proc *p = td->td_proc;
1336 	struct file *fp;
1337 	struct mount *mp;
1338 	struct statvfs *sp;
1339 	int error;
1340 
1341 	KKASSERT(p);
1342 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1343 		return (error);
1344 	if ((mp = fp->f_nchandle.mount) == NULL)
1345 		mp = ((struct vnode *)fp->f_data)->v_mount;
1346 	if (mp == NULL) {
1347 		error = EBADF;
1348 		goto done;
1349 	}
1350 	if (fp->f_cred == NULL) {
1351 		error = EINVAL;
1352 		goto done;
1353 	}
1354 	sp = &mp->mnt_vstat;
1355 	if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1356 		goto done;
1357 
1358 	sp->f_flag = 0;
1359 	if (mp->mnt_flag & MNT_RDONLY)
1360 		sp->f_flag |= ST_RDONLY;
1361 	if (mp->mnt_flag & MNT_NOSUID)
1362 		sp->f_flag |= ST_NOSUID;
1363 
1364 	bcopy(sp, buf, sizeof(*buf));
1365 	error = 0;
1366 done:
1367 	fdrop(fp);
1368 	return (error);
1369 }
1370 
1371 /*
1372  * fstatfs_args(int fd, struct statfs *buf)
1373  *
1374  * Get filesystem statistics.
1375  */
1376 int
1377 sys_fstatvfs(struct fstatvfs_args *uap)
1378 {
1379 	struct statvfs buf;
1380 	int error;
1381 
1382 	error = kern_fstatvfs(uap->fd, &buf);
1383 
1384 	if (error == 0)
1385 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1386 	return (error);
1387 }
1388 
1389 /*
1390  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1391  *
1392  * Get statistics on all filesystems.
1393  */
1394 
1395 struct getfsstat_info {
1396 	struct statfs *sfsp;
1397 	long count;
1398 	long maxcount;
1399 	int error;
1400 	int flags;
1401 	struct thread *td;
1402 };
1403 
1404 static int getfsstat_callback(struct mount *, void *);
1405 
1406 int
1407 sys_getfsstat(struct getfsstat_args *uap)
1408 {
1409 	struct thread *td = curthread;
1410 	struct getfsstat_info info;
1411 
1412 	bzero(&info, sizeof(info));
1413 
1414 	info.maxcount = uap->bufsize / sizeof(struct statfs);
1415 	info.sfsp = uap->buf;
1416 	info.count = 0;
1417 	info.flags = uap->flags;
1418 	info.td = td;
1419 
1420 	mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1421 	if (info.sfsp && info.count > info.maxcount)
1422 		uap->sysmsg_result = info.maxcount;
1423 	else
1424 		uap->sysmsg_result = info.count;
1425 	return (info.error);
1426 }
1427 
1428 static int
1429 getfsstat_callback(struct mount *mp, void *data)
1430 {
1431 	struct getfsstat_info *info = data;
1432 	struct statfs *sp;
1433 	char *freepath;
1434 	char *fullpath;
1435 	int error;
1436 
1437 	if (info->sfsp && info->count < info->maxcount) {
1438 		if (info->td->td_proc &&
1439 		    !chroot_visible_mnt(mp, info->td->td_proc)) {
1440 			return(0);
1441 		}
1442 		sp = &mp->mnt_stat;
1443 
1444 		/*
1445 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1446 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1447 		 * overrides MNT_WAIT.
1448 		 */
1449 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1450 		    (info->flags & MNT_WAIT)) &&
1451 		    (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1452 			return(0);
1453 		}
1454 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1455 
1456 		error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1457 		if (error) {
1458 			info->error = error;
1459 			return(-1);
1460 		}
1461 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1462 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1463 		kfree(freepath, M_TEMP);
1464 
1465 		error = copyout(sp, info->sfsp, sizeof(*sp));
1466 		if (error) {
1467 			info->error = error;
1468 			return (-1);
1469 		}
1470 		++info->sfsp;
1471 	}
1472 	info->count++;
1473 	return(0);
1474 }
1475 
1476 /*
1477  * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1478 		   long bufsize, int flags)
1479  *
1480  * Get statistics on all filesystems.
1481  */
1482 
1483 struct getvfsstat_info {
1484 	struct statfs *sfsp;
1485 	struct statvfs *vsfsp;
1486 	long count;
1487 	long maxcount;
1488 	int error;
1489 	int flags;
1490 	struct thread *td;
1491 };
1492 
1493 static int getvfsstat_callback(struct mount *, void *);
1494 
1495 int
1496 sys_getvfsstat(struct getvfsstat_args *uap)
1497 {
1498 	struct thread *td = curthread;
1499 	struct getvfsstat_info info;
1500 
1501 	bzero(&info, sizeof(info));
1502 
1503 	info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1504 	info.sfsp = uap->buf;
1505 	info.vsfsp = uap->vbuf;
1506 	info.count = 0;
1507 	info.flags = uap->flags;
1508 	info.td = td;
1509 
1510 	mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1511 	if (info.vsfsp && info.count > info.maxcount)
1512 		uap->sysmsg_result = info.maxcount;
1513 	else
1514 		uap->sysmsg_result = info.count;
1515 	return (info.error);
1516 }
1517 
1518 static int
1519 getvfsstat_callback(struct mount *mp, void *data)
1520 {
1521 	struct getvfsstat_info *info = data;
1522 	struct statfs *sp;
1523 	struct statvfs *vsp;
1524 	char *freepath;
1525 	char *fullpath;
1526 	int error;
1527 
1528 	if (info->vsfsp && info->count < info->maxcount) {
1529 		if (info->td->td_proc &&
1530 		    !chroot_visible_mnt(mp, info->td->td_proc)) {
1531 			return(0);
1532 		}
1533 		sp = &mp->mnt_stat;
1534 		vsp = &mp->mnt_vstat;
1535 
1536 		/*
1537 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1538 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1539 		 * overrides MNT_WAIT.
1540 		 */
1541 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1542 		    (info->flags & MNT_WAIT)) &&
1543 		    (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1544 			return(0);
1545 		}
1546 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1547 
1548 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1549 		    (info->flags & MNT_WAIT)) &&
1550 		    (error = VFS_STATVFS(mp, vsp, info->td->td_ucred))) {
1551 			return(0);
1552 		}
1553 		vsp->f_flag = 0;
1554 		if (mp->mnt_flag & MNT_RDONLY)
1555 			vsp->f_flag |= ST_RDONLY;
1556 		if (mp->mnt_flag & MNT_NOSUID)
1557 			vsp->f_flag |= ST_NOSUID;
1558 
1559 		error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1560 		if (error) {
1561 			info->error = error;
1562 			return(-1);
1563 		}
1564 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1565 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1566 		kfree(freepath, M_TEMP);
1567 
1568 		error = copyout(sp, info->sfsp, sizeof(*sp));
1569 		if (error == 0)
1570 			error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1571 		if (error) {
1572 			info->error = error;
1573 			return (-1);
1574 		}
1575 		++info->sfsp;
1576 		++info->vsfsp;
1577 	}
1578 	info->count++;
1579 	return(0);
1580 }
1581 
1582 
1583 /*
1584  * fchdir_args(int fd)
1585  *
1586  * Change current working directory to a given file descriptor.
1587  */
1588 int
1589 sys_fchdir(struct fchdir_args *uap)
1590 {
1591 	struct thread *td = curthread;
1592 	struct proc *p = td->td_proc;
1593 	struct filedesc *fdp = p->p_fd;
1594 	struct vnode *vp, *ovp;
1595 	struct mount *mp;
1596 	struct file *fp;
1597 	struct nchandle nch, onch, tnch;
1598 	int error;
1599 
1600 	if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1601 		return (error);
1602 	lwkt_gettoken(&p->p_token);
1603 	vp = (struct vnode *)fp->f_data;
1604 	vref(vp);
1605 	vn_lock(vp, LK_SHARED | LK_RETRY);
1606 	if (fp->f_nchandle.ncp == NULL)
1607 		error = ENOTDIR;
1608 	else
1609 		error = checkvp_chdir(vp, td);
1610 	if (error) {
1611 		vput(vp);
1612 		goto done;
1613 	}
1614 	cache_copy(&fp->f_nchandle, &nch);
1615 
1616 	/*
1617 	 * If the ncp has become a mount point, traverse through
1618 	 * the mount point.
1619 	 */
1620 
1621 	while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1622 	       (mp = cache_findmount(&nch)) != NULL
1623 	) {
1624 		error = nlookup_mp(mp, &tnch);
1625 		if (error == 0) {
1626 			cache_unlock(&tnch);	/* leave ref intact */
1627 			vput(vp);
1628 			vp = tnch.ncp->nc_vp;
1629 			error = vget(vp, LK_SHARED);
1630 			KKASSERT(error == 0);
1631 			cache_drop(&nch);
1632 			nch = tnch;
1633 		}
1634 		cache_dropmount(mp);
1635 	}
1636 	if (error == 0) {
1637 		ovp = fdp->fd_cdir;
1638 		onch = fdp->fd_ncdir;
1639 		vn_unlock(vp);		/* leave ref intact */
1640 		fdp->fd_cdir = vp;
1641 		fdp->fd_ncdir = nch;
1642 		cache_drop(&onch);
1643 		vrele(ovp);
1644 	} else {
1645 		cache_drop(&nch);
1646 		vput(vp);
1647 	}
1648 	fdrop(fp);
1649 done:
1650 	lwkt_reltoken(&p->p_token);
1651 	return (error);
1652 }
1653 
1654 int
1655 kern_chdir(struct nlookupdata *nd)
1656 {
1657 	struct thread *td = curthread;
1658 	struct proc *p = td->td_proc;
1659 	struct filedesc *fdp = p->p_fd;
1660 	struct vnode *vp, *ovp;
1661 	struct nchandle onch;
1662 	int error;
1663 
1664 	nd->nl_flags |= NLC_SHAREDLOCK;
1665 	if ((error = nlookup(nd)) != 0)
1666 		return (error);
1667 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1668 		return (ENOENT);
1669 	if ((error = vget(vp, LK_SHARED)) != 0)
1670 		return (error);
1671 
1672 	lwkt_gettoken(&p->p_token);
1673 	error = checkvp_chdir(vp, td);
1674 	vn_unlock(vp);
1675 	if (error == 0) {
1676 		ovp = fdp->fd_cdir;
1677 		onch = fdp->fd_ncdir;
1678 		cache_unlock(&nd->nl_nch);	/* leave reference intact */
1679 		fdp->fd_ncdir = nd->nl_nch;
1680 		fdp->fd_cdir = vp;
1681 		cache_drop(&onch);
1682 		vrele(ovp);
1683 		cache_zero(&nd->nl_nch);
1684 	} else {
1685 		vrele(vp);
1686 	}
1687 	lwkt_reltoken(&p->p_token);
1688 	return (error);
1689 }
1690 
1691 /*
1692  * chdir_args(char *path)
1693  *
1694  * Change current working directory (``.'').
1695  */
1696 int
1697 sys_chdir(struct chdir_args *uap)
1698 {
1699 	struct nlookupdata nd;
1700 	int error;
1701 
1702 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1703 	if (error == 0)
1704 		error = kern_chdir(&nd);
1705 	nlookup_done(&nd);
1706 	return (error);
1707 }
1708 
1709 /*
1710  * Helper function for raised chroot(2) security function:  Refuse if
1711  * any filedescriptors are open directories.
1712  */
1713 static int
1714 chroot_refuse_vdir_fds(struct filedesc *fdp)
1715 {
1716 	struct vnode *vp;
1717 	struct file *fp;
1718 	int error;
1719 	int fd;
1720 
1721 	for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1722 		if ((error = holdvnode(fdp, fd, &fp)) != 0)
1723 			continue;
1724 		vp = (struct vnode *)fp->f_data;
1725 		if (vp->v_type != VDIR) {
1726 			fdrop(fp);
1727 			continue;
1728 		}
1729 		fdrop(fp);
1730 		return(EPERM);
1731 	}
1732 	return (0);
1733 }
1734 
1735 /*
1736  * This sysctl determines if we will allow a process to chroot(2) if it
1737  * has a directory open:
1738  *	0: disallowed for all processes.
1739  *	1: allowed for processes that were not already chroot(2)'ed.
1740  *	2: allowed for all processes.
1741  */
1742 
1743 static int chroot_allow_open_directories = 1;
1744 
1745 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1746      &chroot_allow_open_directories, 0, "");
1747 
1748 /*
1749  * chroot to the specified namecache entry.  We obtain the vp from the
1750  * namecache data.  The passed ncp must be locked and referenced and will
1751  * remain locked and referenced on return.
1752  */
1753 int
1754 kern_chroot(struct nchandle *nch)
1755 {
1756 	struct thread *td = curthread;
1757 	struct proc *p = td->td_proc;
1758 	struct filedesc *fdp = p->p_fd;
1759 	struct vnode *vp;
1760 	int error;
1761 
1762 	/*
1763 	 * Only privileged user can chroot
1764 	 */
1765 	error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1766 	if (error)
1767 		return (error);
1768 
1769 	/*
1770 	 * Disallow open directory descriptors (fchdir() breakouts).
1771 	 */
1772 	if (chroot_allow_open_directories == 0 ||
1773 	   (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1774 		if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1775 			return (error);
1776 	}
1777 	if ((vp = nch->ncp->nc_vp) == NULL)
1778 		return (ENOENT);
1779 
1780 	if ((error = vget(vp, LK_SHARED)) != 0)
1781 		return (error);
1782 
1783 	/*
1784 	 * Check the validity of vp as a directory to change to and
1785 	 * associate it with rdir/jdir.
1786 	 */
1787 	error = checkvp_chdir(vp, td);
1788 	vn_unlock(vp);			/* leave reference intact */
1789 	if (error == 0) {
1790 		vrele(fdp->fd_rdir);
1791 		fdp->fd_rdir = vp;	/* reference inherited by fd_rdir */
1792 		cache_drop(&fdp->fd_nrdir);
1793 		cache_copy(nch, &fdp->fd_nrdir);
1794 		if (fdp->fd_jdir == NULL) {
1795 			fdp->fd_jdir = vp;
1796 			vref(fdp->fd_jdir);
1797 			cache_copy(nch, &fdp->fd_njdir);
1798 		}
1799 	} else {
1800 		vrele(vp);
1801 	}
1802 	return (error);
1803 }
1804 
1805 /*
1806  * chroot_args(char *path)
1807  *
1808  * Change notion of root (``/'') directory.
1809  */
1810 int
1811 sys_chroot(struct chroot_args *uap)
1812 {
1813 	struct thread *td __debugvar = curthread;
1814 	struct nlookupdata nd;
1815 	int error;
1816 
1817 	KKASSERT(td->td_proc);
1818 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1819 	if (error == 0) {
1820 		nd.nl_flags |= NLC_EXEC;
1821 		error = nlookup(&nd);
1822 		if (error == 0)
1823 			error = kern_chroot(&nd.nl_nch);
1824 	}
1825 	nlookup_done(&nd);
1826 	return(error);
1827 }
1828 
1829 int
1830 sys_chroot_kernel(struct chroot_kernel_args *uap)
1831 {
1832 	struct thread *td = curthread;
1833 	struct nlookupdata nd;
1834 	struct nchandle *nch;
1835 	struct vnode *vp;
1836 	int error;
1837 
1838 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1839 	if (error)
1840 		goto error_nond;
1841 
1842 	error = nlookup(&nd);
1843 	if (error)
1844 		goto error_out;
1845 
1846 	nch = &nd.nl_nch;
1847 
1848 	error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1849 	if (error)
1850 		goto error_out;
1851 
1852 	if ((vp = nch->ncp->nc_vp) == NULL) {
1853 		error = ENOENT;
1854 		goto error_out;
1855 	}
1856 
1857 	if ((error = cache_vref(nch, nd.nl_cred, &vp)) != 0)
1858 		goto error_out;
1859 
1860 	kprintf("chroot_kernel: set new rootnch/rootvnode to %s\n", uap->path);
1861 	get_mplock();
1862 	vfs_cache_setroot(vp, cache_hold(nch));
1863 	rel_mplock();
1864 
1865 error_out:
1866 	nlookup_done(&nd);
1867 error_nond:
1868 	return(error);
1869 }
1870 
1871 /*
1872  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1873  * determine whether it is legal to chdir to the vnode.  The vnode's state
1874  * is not changed by this call.
1875  */
1876 static int
1877 checkvp_chdir(struct vnode *vp, struct thread *td)
1878 {
1879 	int error;
1880 
1881 	if (vp->v_type != VDIR)
1882 		error = ENOTDIR;
1883 	else
1884 		error = VOP_EACCESS(vp, VEXEC, td->td_ucred);
1885 	return (error);
1886 }
1887 
1888 int
1889 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1890 {
1891 	struct thread *td = curthread;
1892 	struct proc *p = td->td_proc;
1893 	struct lwp *lp = td->td_lwp;
1894 	struct filedesc *fdp = p->p_fd;
1895 	int cmode, flags;
1896 	struct file *nfp;
1897 	struct file *fp;
1898 	struct vnode *vp;
1899 	int type, indx, error = 0;
1900 	struct flock lf;
1901 
1902 	if ((oflags & O_ACCMODE) == O_ACCMODE)
1903 		return (EINVAL);
1904 	flags = FFLAGS(oflags);
1905 	error = falloc(lp, &nfp, NULL);
1906 	if (error)
1907 		return (error);
1908 	fp = nfp;
1909 	cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
1910 
1911 	/*
1912 	 * XXX p_dupfd is a real mess.  It allows a device to return a
1913 	 * file descriptor to be duplicated rather then doing the open
1914 	 * itself.
1915 	 */
1916 	lp->lwp_dupfd = -1;
1917 
1918 	/*
1919 	 * Call vn_open() to do the lookup and assign the vnode to the
1920 	 * file pointer.  vn_open() does not change the ref count on fp
1921 	 * and the vnode, on success, will be inherited by the file pointer
1922 	 * and unlocked.
1923 	 *
1924 	 * Request a shared lock on the vnode if possible.
1925 	 */
1926 	nd->nl_flags |= NLC_LOCKVP;
1927 	if ((flags & (O_CREAT|O_TRUNC)) == 0)
1928 		nd->nl_flags |= NLC_SHAREDLOCK;
1929 
1930 	error = vn_open(nd, fp, flags, cmode);
1931 	nlookup_done(nd);
1932 
1933 	if (error) {
1934 		/*
1935 		 * handle special fdopen() case.  bleh.  dupfdopen() is
1936 		 * responsible for dropping the old contents of ofiles[indx]
1937 		 * if it succeeds.
1938 		 *
1939 		 * Note that fsetfd() will add a ref to fp which represents
1940 		 * the fd_files[] assignment.  We must still drop our
1941 		 * reference.
1942 		 */
1943 		if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1944 			if (fdalloc(p, 0, &indx) == 0) {
1945 				error = dupfdopen(fdp, indx, lp->lwp_dupfd, flags, error);
1946 				if (error == 0) {
1947 					*res = indx;
1948 					fdrop(fp);	/* our ref */
1949 					return (0);
1950 				}
1951 				fsetfd(fdp, NULL, indx);
1952 			}
1953 		}
1954 		fdrop(fp);	/* our ref */
1955 		if (error == ERESTART)
1956 			error = EINTR;
1957 		return (error);
1958 	}
1959 
1960 	/*
1961 	 * ref the vnode for ourselves so it can't be ripped out from under
1962 	 * is.  XXX need an ND flag to request that the vnode be returned
1963 	 * anyway.
1964 	 *
1965 	 * Reserve a file descriptor but do not assign it until the open
1966 	 * succeeds.
1967 	 */
1968 	vp = (struct vnode *)fp->f_data;
1969 	vref(vp);
1970 	if ((error = fdalloc(p, 0, &indx)) != 0) {
1971 		fdrop(fp);
1972 		vrele(vp);
1973 		return (error);
1974 	}
1975 
1976 	/*
1977 	 * If no error occurs the vp will have been assigned to the file
1978 	 * pointer.
1979 	 */
1980 	lp->lwp_dupfd = 0;
1981 
1982 	if (flags & (O_EXLOCK | O_SHLOCK)) {
1983 		lf.l_whence = SEEK_SET;
1984 		lf.l_start = 0;
1985 		lf.l_len = 0;
1986 		if (flags & O_EXLOCK)
1987 			lf.l_type = F_WRLCK;
1988 		else
1989 			lf.l_type = F_RDLCK;
1990 		if (flags & FNONBLOCK)
1991 			type = 0;
1992 		else
1993 			type = F_WAIT;
1994 
1995 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1996 			/*
1997 			 * lock request failed.  Clean up the reserved
1998 			 * descriptor.
1999 			 */
2000 			vrele(vp);
2001 			fsetfd(fdp, NULL, indx);
2002 			fdrop(fp);
2003 			return (error);
2004 		}
2005 		atomic_set_int(&fp->f_flag, FHASLOCK); /* race ok */
2006 	}
2007 #if 0
2008 	/*
2009 	 * Assert that all regular file vnodes were created with a object.
2010 	 */
2011 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
2012 		("open: regular file has no backing object after vn_open"));
2013 #endif
2014 
2015 	vrele(vp);
2016 
2017 	/*
2018 	 * release our private reference, leaving the one associated with the
2019 	 * descriptor table intact.
2020 	 */
2021 	if (oflags & O_CLOEXEC)
2022 		fdp->fd_files[indx].fileflags |= UF_EXCLOSE;
2023 	fsetfd(fdp, fp, indx);
2024 	fdrop(fp);
2025 	*res = indx;
2026 	return (error);
2027 }
2028 
2029 /*
2030  * open_args(char *path, int flags, int mode)
2031  *
2032  * Check permissions, allocate an open file structure,
2033  * and call the device open routine if any.
2034  */
2035 int
2036 sys_open(struct open_args *uap)
2037 {
2038 	struct nlookupdata nd;
2039 	int error;
2040 
2041 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2042 	if (error == 0) {
2043 		error = kern_open(&nd, uap->flags,
2044 				    uap->mode, &uap->sysmsg_result);
2045 	}
2046 	nlookup_done(&nd);
2047 	return (error);
2048 }
2049 
2050 /*
2051  * openat_args(int fd, char *path, int flags, int mode)
2052  */
2053 int
2054 sys_openat(struct openat_args *uap)
2055 {
2056 	struct nlookupdata nd;
2057 	int error;
2058 	struct file *fp;
2059 
2060 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2061 	if (error == 0) {
2062 		error = kern_open(&nd, uap->flags, uap->mode,
2063 					&uap->sysmsg_result);
2064 	}
2065 	nlookup_done_at(&nd, fp);
2066 	return (error);
2067 }
2068 
2069 int
2070 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
2071 {
2072 	struct thread *td = curthread;
2073 	struct proc *p = td->td_proc;
2074 	struct vnode *vp;
2075 	struct vattr vattr;
2076 	int error;
2077 	int whiteout = 0;
2078 
2079 	KKASSERT(p);
2080 
2081 	VATTR_NULL(&vattr);
2082 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2083 	vattr.va_rmajor = rmajor;
2084 	vattr.va_rminor = rminor;
2085 
2086 	switch (mode & S_IFMT) {
2087 	case S_IFMT:	/* used by badsect to flag bad sectors */
2088 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_BAD, 0);
2089 		vattr.va_type = VBAD;
2090 		break;
2091 	case S_IFCHR:
2092 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2093 		vattr.va_type = VCHR;
2094 		break;
2095 	case S_IFBLK:
2096 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2097 		vattr.va_type = VBLK;
2098 		break;
2099 	case S_IFWHT:
2100 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_WHT, 0);
2101 		whiteout = 1;
2102 		break;
2103 	case S_IFDIR:	/* special directories support for HAMMER */
2104 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_DIR, 0);
2105 		vattr.va_type = VDIR;
2106 		break;
2107 	default:
2108 		error = EINVAL;
2109 		break;
2110 	}
2111 
2112 	if (error)
2113 		return (error);
2114 
2115 	bwillinode(1);
2116 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2117 	if ((error = nlookup(nd)) != 0)
2118 		return (error);
2119 	if (nd->nl_nch.ncp->nc_vp)
2120 		return (EEXIST);
2121 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2122 		return (error);
2123 
2124 	if (whiteout) {
2125 		error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
2126 				      nd->nl_cred, NAMEI_CREATE);
2127 	} else {
2128 		vp = NULL;
2129 		error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
2130 				   &vp, nd->nl_cred, &vattr);
2131 		if (error == 0)
2132 			vput(vp);
2133 	}
2134 	return (error);
2135 }
2136 
2137 /*
2138  * mknod_args(char *path, int mode, int dev)
2139  *
2140  * Create a special file.
2141  */
2142 int
2143 sys_mknod(struct mknod_args *uap)
2144 {
2145 	struct nlookupdata nd;
2146 	int error;
2147 
2148 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2149 	if (error == 0) {
2150 		error = kern_mknod(&nd, uap->mode,
2151 				   umajor(uap->dev), uminor(uap->dev));
2152 	}
2153 	nlookup_done(&nd);
2154 	return (error);
2155 }
2156 
2157 /*
2158  * mknodat_args(int fd, char *path, mode_t mode, dev_t dev)
2159  *
2160  * Create a special file.  The path is relative to the directory associated
2161  * with fd.
2162  */
2163 int
2164 sys_mknodat(struct mknodat_args *uap)
2165 {
2166 	struct nlookupdata nd;
2167 	struct file *fp;
2168 	int error;
2169 
2170 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2171 	if (error == 0) {
2172 		error = kern_mknod(&nd, uap->mode,
2173 				   umajor(uap->dev), uminor(uap->dev));
2174 	}
2175 	nlookup_done_at(&nd, fp);
2176 	return (error);
2177 }
2178 
2179 int
2180 kern_mkfifo(struct nlookupdata *nd, int mode)
2181 {
2182 	struct thread *td = curthread;
2183 	struct proc *p = td->td_proc;
2184 	struct vattr vattr;
2185 	struct vnode *vp;
2186 	int error;
2187 
2188 	bwillinode(1);
2189 
2190 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2191 	if ((error = nlookup(nd)) != 0)
2192 		return (error);
2193 	if (nd->nl_nch.ncp->nc_vp)
2194 		return (EEXIST);
2195 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2196 		return (error);
2197 
2198 	VATTR_NULL(&vattr);
2199 	vattr.va_type = VFIFO;
2200 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2201 	vp = NULL;
2202 	error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
2203 	if (error == 0)
2204 		vput(vp);
2205 	return (error);
2206 }
2207 
2208 /*
2209  * mkfifo_args(char *path, int mode)
2210  *
2211  * Create a named pipe.
2212  */
2213 int
2214 sys_mkfifo(struct mkfifo_args *uap)
2215 {
2216 	struct nlookupdata nd;
2217 	int error;
2218 
2219 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2220 	if (error == 0)
2221 		error = kern_mkfifo(&nd, uap->mode);
2222 	nlookup_done(&nd);
2223 	return (error);
2224 }
2225 
2226 /*
2227  * mkfifoat_args(int fd, char *path, mode_t mode)
2228  *
2229  * Create a named pipe.  The path is relative to the directory associated
2230  * with fd.
2231  */
2232 int
2233 sys_mkfifoat(struct mkfifoat_args *uap)
2234 {
2235 	struct nlookupdata nd;
2236 	struct file *fp;
2237 	int error;
2238 
2239 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2240 	if (error == 0)
2241 		error = kern_mkfifo(&nd, uap->mode);
2242 	nlookup_done_at(&nd, fp);
2243 	return (error);
2244 }
2245 
2246 static int hardlink_check_uid = 0;
2247 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
2248     &hardlink_check_uid, 0,
2249     "Unprivileged processes cannot create hard links to files owned by other "
2250     "users");
2251 static int hardlink_check_gid = 0;
2252 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
2253     &hardlink_check_gid, 0,
2254     "Unprivileged processes cannot create hard links to files owned by other "
2255     "groups");
2256 
2257 static int
2258 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
2259 {
2260 	struct vattr va;
2261 	int error;
2262 
2263 	/*
2264 	 * Shortcut if disabled
2265 	 */
2266 	if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
2267 		return (0);
2268 
2269 	/*
2270 	 * Privileged user can always hardlink
2271 	 */
2272 	if (priv_check_cred(cred, PRIV_VFS_LINK, 0) == 0)
2273 		return (0);
2274 
2275 	/*
2276 	 * Otherwise only if the originating file is owned by the
2277 	 * same user or group.  Note that any group is allowed if
2278 	 * the file is owned by the caller.
2279 	 */
2280 	error = VOP_GETATTR(vp, &va);
2281 	if (error != 0)
2282 		return (error);
2283 
2284 	if (hardlink_check_uid) {
2285 		if (cred->cr_uid != va.va_uid)
2286 			return (EPERM);
2287 	}
2288 
2289 	if (hardlink_check_gid) {
2290 		if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2291 			return (EPERM);
2292 	}
2293 
2294 	return (0);
2295 }
2296 
2297 int
2298 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2299 {
2300 	struct thread *td = curthread;
2301 	struct vnode *vp;
2302 	int error;
2303 
2304 	/*
2305 	 * Lookup the source and obtained a locked vnode.
2306 	 *
2307 	 * You may only hardlink a file which you have write permission
2308 	 * on or which you own.
2309 	 *
2310 	 * XXX relookup on vget failure / race ?
2311 	 */
2312 	bwillinode(1);
2313 	nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK;
2314 	if ((error = nlookup(nd)) != 0)
2315 		return (error);
2316 	vp = nd->nl_nch.ncp->nc_vp;
2317 	KKASSERT(vp != NULL);
2318 	if (vp->v_type == VDIR)
2319 		return (EPERM);		/* POSIX */
2320 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2321 		return (error);
2322 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2323 		return (error);
2324 
2325 	/*
2326 	 * Unlock the source so we can lookup the target without deadlocking
2327 	 * (XXX vp is locked already, possible other deadlock?).  The target
2328 	 * must not exist.
2329 	 */
2330 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2331 	nd->nl_flags &= ~NLC_NCPISLOCKED;
2332 	cache_unlock(&nd->nl_nch);
2333 	vn_unlock(vp);
2334 
2335 	linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2336 	if ((error = nlookup(linknd)) != 0) {
2337 		vrele(vp);
2338 		return (error);
2339 	}
2340 	if (linknd->nl_nch.ncp->nc_vp) {
2341 		vrele(vp);
2342 		return (EEXIST);
2343 	}
2344 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
2345 	if (error) {
2346 		vrele(vp);
2347 		return (error);
2348 	}
2349 
2350 	/*
2351 	 * Finally run the new API VOP.
2352 	 */
2353 	error = can_hardlink(vp, td, td->td_ucred);
2354 	if (error == 0) {
2355 		error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2356 				  vp, linknd->nl_cred);
2357 	}
2358 	vput(vp);
2359 	return (error);
2360 }
2361 
2362 /*
2363  * link_args(char *path, char *link)
2364  *
2365  * Make a hard file link.
2366  */
2367 int
2368 sys_link(struct link_args *uap)
2369 {
2370 	struct nlookupdata nd, linknd;
2371 	int error;
2372 
2373 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2374 	if (error == 0) {
2375 		error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2376 		if (error == 0)
2377 			error = kern_link(&nd, &linknd);
2378 		nlookup_done(&linknd);
2379 	}
2380 	nlookup_done(&nd);
2381 	return (error);
2382 }
2383 
2384 /*
2385  * linkat_args(int fd1, char *path1, int fd2, char *path2, int flags)
2386  *
2387  * Make a hard file link. The path1 argument is relative to the directory
2388  * associated with fd1, and similarly the path2 argument is relative to
2389  * the directory associated with fd2.
2390  */
2391 int
2392 sys_linkat(struct linkat_args *uap)
2393 {
2394 	struct nlookupdata nd, linknd;
2395 	struct file *fp1, *fp2;
2396 	int error;
2397 
2398 	error = nlookup_init_at(&nd, &fp1, uap->fd1, uap->path1, UIO_USERSPACE,
2399 	    (uap->flags & AT_SYMLINK_FOLLOW) ? NLC_FOLLOW : 0);
2400 	if (error == 0) {
2401 		error = nlookup_init_at(&linknd, &fp2, uap->fd2,
2402 		    uap->path2, UIO_USERSPACE, 0);
2403 		if (error == 0)
2404 			error = kern_link(&nd, &linknd);
2405 		nlookup_done_at(&linknd, fp2);
2406 	}
2407 	nlookup_done_at(&nd, fp1);
2408 	return (error);
2409 }
2410 
2411 int
2412 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2413 {
2414 	struct vattr vattr;
2415 	struct vnode *vp;
2416 	struct vnode *dvp;
2417 	int error;
2418 
2419 	bwillinode(1);
2420 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2421 	if ((error = nlookup(nd)) != 0)
2422 		return (error);
2423 	if (nd->nl_nch.ncp->nc_vp)
2424 		return (EEXIST);
2425 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2426 		return (error);
2427 	dvp = nd->nl_dvp;
2428 	VATTR_NULL(&vattr);
2429 	vattr.va_mode = mode;
2430 	error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2431 	if (error == 0)
2432 		vput(vp);
2433 	return (error);
2434 }
2435 
2436 /*
2437  * symlink(char *path, char *link)
2438  *
2439  * Make a symbolic link.
2440  */
2441 int
2442 sys_symlink(struct symlink_args *uap)
2443 {
2444 	struct thread *td = curthread;
2445 	struct nlookupdata nd;
2446 	char *path;
2447 	int error;
2448 	int mode;
2449 
2450 	path = objcache_get(namei_oc, M_WAITOK);
2451 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2452 	if (error == 0) {
2453 		error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2454 		if (error == 0) {
2455 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2456 			error = kern_symlink(&nd, path, mode);
2457 		}
2458 		nlookup_done(&nd);
2459 	}
2460 	objcache_put(namei_oc, path);
2461 	return (error);
2462 }
2463 
2464 /*
2465  * symlinkat_args(char *path1, int fd, char *path2)
2466  *
2467  * Make a symbolic link.  The path2 argument is relative to the directory
2468  * associated with fd.
2469  */
2470 int
2471 sys_symlinkat(struct symlinkat_args *uap)
2472 {
2473 	struct thread *td = curthread;
2474 	struct nlookupdata nd;
2475 	struct file *fp;
2476 	char *path1;
2477 	int error;
2478 	int mode;
2479 
2480 	path1 = objcache_get(namei_oc, M_WAITOK);
2481 	error = copyinstr(uap->path1, path1, MAXPATHLEN, NULL);
2482 	if (error == 0) {
2483 		error = nlookup_init_at(&nd, &fp, uap->fd, uap->path2,
2484 		    UIO_USERSPACE, 0);
2485 		if (error == 0) {
2486 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2487 			error = kern_symlink(&nd, path1, mode);
2488 		}
2489 		nlookup_done_at(&nd, fp);
2490 	}
2491 	objcache_put(namei_oc, path1);
2492 	return (error);
2493 }
2494 
2495 /*
2496  * undelete_args(char *path)
2497  *
2498  * Delete a whiteout from the filesystem.
2499  */
2500 int
2501 sys_undelete(struct undelete_args *uap)
2502 {
2503 	struct nlookupdata nd;
2504 	int error;
2505 
2506 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2507 	bwillinode(1);
2508 	nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2509 	if (error == 0)
2510 		error = nlookup(&nd);
2511 	if (error == 0)
2512 		error = ncp_writechk(&nd.nl_nch);
2513 	if (error == 0) {
2514 		error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2515 				      NAMEI_DELETE);
2516 	}
2517 	nlookup_done(&nd);
2518 	return (error);
2519 }
2520 
2521 int
2522 kern_unlink(struct nlookupdata *nd)
2523 {
2524 	int error;
2525 
2526 	bwillinode(1);
2527 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2528 	if ((error = nlookup(nd)) != 0)
2529 		return (error);
2530 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2531 		return (error);
2532 	error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2533 	return (error);
2534 }
2535 
2536 /*
2537  * unlink_args(char *path)
2538  *
2539  * Delete a name from the filesystem.
2540  */
2541 int
2542 sys_unlink(struct unlink_args *uap)
2543 {
2544 	struct nlookupdata nd;
2545 	int error;
2546 
2547 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2548 	if (error == 0)
2549 		error = kern_unlink(&nd);
2550 	nlookup_done(&nd);
2551 	return (error);
2552 }
2553 
2554 
2555 /*
2556  * unlinkat_args(int fd, char *path, int flags)
2557  *
2558  * Delete the file or directory entry pointed to by fd/path.
2559  */
2560 int
2561 sys_unlinkat(struct unlinkat_args *uap)
2562 {
2563 	struct nlookupdata nd;
2564 	struct file *fp;
2565 	int error;
2566 
2567 	if (uap->flags & ~AT_REMOVEDIR)
2568 		return (EINVAL);
2569 
2570 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2571 	if (error == 0) {
2572 		if (uap->flags & AT_REMOVEDIR)
2573 			error = kern_rmdir(&nd);
2574 		else
2575 			error = kern_unlink(&nd);
2576 	}
2577 	nlookup_done_at(&nd, fp);
2578 	return (error);
2579 }
2580 
2581 int
2582 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2583 {
2584 	struct thread *td = curthread;
2585 	struct proc *p = td->td_proc;
2586 	struct file *fp;
2587 	struct vnode *vp;
2588 	struct vattr vattr;
2589 	off_t new_offset;
2590 	int error;
2591 
2592 	fp = holdfp(p->p_fd, fd, -1);
2593 	if (fp == NULL)
2594 		return (EBADF);
2595 	if (fp->f_type != DTYPE_VNODE) {
2596 		error = ESPIPE;
2597 		goto done;
2598 	}
2599 	vp = (struct vnode *)fp->f_data;
2600 
2601 	switch (whence) {
2602 	case L_INCR:
2603 		spin_lock(&fp->f_spin);
2604 		new_offset = fp->f_offset + offset;
2605 		error = 0;
2606 		break;
2607 	case L_XTND:
2608 		error = VOP_GETATTR(vp, &vattr);
2609 		spin_lock(&fp->f_spin);
2610 		new_offset = offset + vattr.va_size;
2611 		break;
2612 	case L_SET:
2613 		new_offset = offset;
2614 		error = 0;
2615 		spin_lock(&fp->f_spin);
2616 		break;
2617 	default:
2618 		new_offset = 0;
2619 		error = EINVAL;
2620 		spin_lock(&fp->f_spin);
2621 		break;
2622 	}
2623 
2624 	/*
2625 	 * Validate the seek position.  Negative offsets are not allowed
2626 	 * for regular files or directories.
2627 	 *
2628 	 * Normally we would also not want to allow negative offsets for
2629 	 * character and block-special devices.  However kvm addresses
2630 	 * on 64 bit architectures might appear to be negative and must
2631 	 * be allowed.
2632 	 */
2633 	if (error == 0) {
2634 		if (new_offset < 0 &&
2635 		    (vp->v_type == VREG || vp->v_type == VDIR)) {
2636 			error = EINVAL;
2637 		} else {
2638 			fp->f_offset = new_offset;
2639 		}
2640 	}
2641 	*res = fp->f_offset;
2642 	spin_unlock(&fp->f_spin);
2643 done:
2644 	fdrop(fp);
2645 	return (error);
2646 }
2647 
2648 /*
2649  * lseek_args(int fd, int pad, off_t offset, int whence)
2650  *
2651  * Reposition read/write file offset.
2652  */
2653 int
2654 sys_lseek(struct lseek_args *uap)
2655 {
2656 	int error;
2657 
2658 	error = kern_lseek(uap->fd, uap->offset, uap->whence,
2659 			   &uap->sysmsg_offset);
2660 
2661 	return (error);
2662 }
2663 
2664 /*
2665  * Check if current process can access given file.  amode is a bitmask of *_OK
2666  * access bits.  flags is a bitmask of AT_* flags.
2667  */
2668 int
2669 kern_access(struct nlookupdata *nd, int amode, int flags)
2670 {
2671 	struct vnode *vp;
2672 	int error, mode;
2673 
2674 	if (flags & ~AT_EACCESS)
2675 		return (EINVAL);
2676 	nd->nl_flags |= NLC_SHAREDLOCK;
2677 	if ((error = nlookup(nd)) != 0)
2678 		return (error);
2679 retry:
2680 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &vp);
2681 	if (error)
2682 		return (error);
2683 
2684 	/* Flags == 0 means only check for existence. */
2685 	if (amode) {
2686 		mode = 0;
2687 		if (amode & R_OK)
2688 			mode |= VREAD;
2689 		if (amode & W_OK)
2690 			mode |= VWRITE;
2691 		if (amode & X_OK)
2692 			mode |= VEXEC;
2693 		if ((mode & VWRITE) == 0 ||
2694 		    (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2695 			error = VOP_ACCESS_FLAGS(vp, mode, flags, nd->nl_cred);
2696 
2697 		/*
2698 		 * If the file handle is stale we have to re-resolve the
2699 		 * entry with the ncp held exclusively.  This is a hack
2700 		 * at the moment.
2701 		 */
2702 		if (error == ESTALE) {
2703 			vput(vp);
2704 			cache_unlock(&nd->nl_nch);
2705 			cache_lock(&nd->nl_nch);
2706 			cache_setunresolved(&nd->nl_nch);
2707 			error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2708 			if (error == 0) {
2709 				vp = NULL;
2710 				goto retry;
2711 			}
2712 			return(error);
2713 		}
2714 	}
2715 	vput(vp);
2716 	return (error);
2717 }
2718 
2719 /*
2720  * access_args(char *path, int flags)
2721  *
2722  * Check access permissions.
2723  */
2724 int
2725 sys_access(struct access_args *uap)
2726 {
2727 	struct nlookupdata nd;
2728 	int error;
2729 
2730 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2731 	if (error == 0)
2732 		error = kern_access(&nd, uap->flags, 0);
2733 	nlookup_done(&nd);
2734 	return (error);
2735 }
2736 
2737 
2738 /*
2739  * eaccess_args(char *path, int flags)
2740  *
2741  * Check access permissions.
2742  */
2743 int
2744 sys_eaccess(struct eaccess_args *uap)
2745 {
2746 	struct nlookupdata nd;
2747 	int error;
2748 
2749 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2750 	if (error == 0)
2751 		error = kern_access(&nd, uap->flags, AT_EACCESS);
2752 	nlookup_done(&nd);
2753 	return (error);
2754 }
2755 
2756 
2757 /*
2758  * faccessat_args(int fd, char *path, int amode, int flags)
2759  *
2760  * Check access permissions.
2761  */
2762 int
2763 sys_faccessat(struct faccessat_args *uap)
2764 {
2765 	struct nlookupdata nd;
2766 	struct file *fp;
2767 	int error;
2768 
2769 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE,
2770 				NLC_FOLLOW);
2771 	if (error == 0)
2772 		error = kern_access(&nd, uap->amode, uap->flags);
2773 	nlookup_done_at(&nd, fp);
2774 	return (error);
2775 }
2776 
2777 int
2778 kern_stat(struct nlookupdata *nd, struct stat *st)
2779 {
2780 	int error;
2781 	struct vnode *vp;
2782 
2783 	nd->nl_flags |= NLC_SHAREDLOCK;
2784 	if ((error = nlookup(nd)) != 0)
2785 		return (error);
2786 again:
2787 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2788 		return (ENOENT);
2789 
2790 	if ((error = vget(vp, LK_SHARED)) != 0)
2791 		return (error);
2792 	error = vn_stat(vp, st, nd->nl_cred);
2793 
2794 	/*
2795 	 * If the file handle is stale we have to re-resolve the
2796 	 * entry with the ncp held exclusively.  This is a hack
2797 	 * at the moment.
2798 	 */
2799 	if (error == ESTALE) {
2800 		vput(vp);
2801 		cache_unlock(&nd->nl_nch);
2802 		cache_lock(&nd->nl_nch);
2803 		cache_setunresolved(&nd->nl_nch);
2804 		error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2805 		if (error == 0)
2806 			goto again;
2807 	} else {
2808 		vput(vp);
2809 	}
2810 	return (error);
2811 }
2812 
2813 /*
2814  * stat_args(char *path, struct stat *ub)
2815  *
2816  * Get file status; this version follows links.
2817  */
2818 int
2819 sys_stat(struct stat_args *uap)
2820 {
2821 	struct nlookupdata nd;
2822 	struct stat st;
2823 	int error;
2824 
2825 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2826 	if (error == 0) {
2827 		error = kern_stat(&nd, &st);
2828 		if (error == 0)
2829 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2830 	}
2831 	nlookup_done(&nd);
2832 	return (error);
2833 }
2834 
2835 /*
2836  * lstat_args(char *path, struct stat *ub)
2837  *
2838  * Get file status; this version does not follow links.
2839  */
2840 int
2841 sys_lstat(struct lstat_args *uap)
2842 {
2843 	struct nlookupdata nd;
2844 	struct stat st;
2845 	int error;
2846 
2847 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2848 	if (error == 0) {
2849 		error = kern_stat(&nd, &st);
2850 		if (error == 0)
2851 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2852 	}
2853 	nlookup_done(&nd);
2854 	return (error);
2855 }
2856 
2857 /*
2858  * fstatat_args(int fd, char *path, struct stat *sb, int flags)
2859  *
2860  * Get status of file pointed to by fd/path.
2861  */
2862 int
2863 sys_fstatat(struct fstatat_args *uap)
2864 {
2865 	struct nlookupdata nd;
2866 	struct stat st;
2867 	int error;
2868 	int flags;
2869 	struct file *fp;
2870 
2871 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
2872 		return (EINVAL);
2873 
2874 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
2875 
2876 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
2877 				UIO_USERSPACE, flags);
2878 	if (error == 0) {
2879 		error = kern_stat(&nd, &st);
2880 		if (error == 0)
2881 			error = copyout(&st, uap->sb, sizeof(*uap->sb));
2882 	}
2883 	nlookup_done_at(&nd, fp);
2884 	return (error);
2885 }
2886 
2887 static int
2888 kern_pathconf(char *path, int name, int flags, register_t *sysmsg_regp)
2889 {
2890 	struct nlookupdata nd;
2891 	struct vnode *vp;
2892 	int error;
2893 
2894 	vp = NULL;
2895 	error = nlookup_init(&nd, path, UIO_USERSPACE, flags);
2896 	if (error == 0)
2897 		error = nlookup(&nd);
2898 	if (error == 0)
2899 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2900 	nlookup_done(&nd);
2901 	if (error == 0) {
2902 		error = VOP_PATHCONF(vp, name, sysmsg_regp);
2903 		vput(vp);
2904 	}
2905 	return (error);
2906 }
2907 
2908 /*
2909  * pathconf_Args(char *path, int name)
2910  *
2911  * Get configurable pathname variables.
2912  */
2913 int
2914 sys_pathconf(struct pathconf_args *uap)
2915 {
2916 	return (kern_pathconf(uap->path, uap->name, NLC_FOLLOW,
2917 		&uap->sysmsg_reg));
2918 }
2919 
2920 /*
2921  * lpathconf_Args(char *path, int name)
2922  *
2923  * Get configurable pathname variables, but don't follow symlinks.
2924  */
2925 int
2926 sys_lpathconf(struct lpathconf_args *uap)
2927 {
2928 	return (kern_pathconf(uap->path, uap->name, 0, &uap->sysmsg_reg));
2929 }
2930 
2931 /*
2932  * XXX: daver
2933  * kern_readlink isn't properly split yet.  There is a copyin burried
2934  * in VOP_READLINK().
2935  */
2936 int
2937 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2938 {
2939 	struct thread *td = curthread;
2940 	struct vnode *vp;
2941 	struct iovec aiov;
2942 	struct uio auio;
2943 	int error;
2944 
2945 	nd->nl_flags |= NLC_SHAREDLOCK;
2946 	if ((error = nlookup(nd)) != 0)
2947 		return (error);
2948 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &vp);
2949 	if (error)
2950 		return (error);
2951 	if (vp->v_type != VLNK) {
2952 		error = EINVAL;
2953 	} else {
2954 		aiov.iov_base = buf;
2955 		aiov.iov_len = count;
2956 		auio.uio_iov = &aiov;
2957 		auio.uio_iovcnt = 1;
2958 		auio.uio_offset = 0;
2959 		auio.uio_rw = UIO_READ;
2960 		auio.uio_segflg = UIO_USERSPACE;
2961 		auio.uio_td = td;
2962 		auio.uio_resid = count;
2963 		error = VOP_READLINK(vp, &auio, td->td_ucred);
2964 	}
2965 	vput(vp);
2966 	*res = count - auio.uio_resid;
2967 	return (error);
2968 }
2969 
2970 /*
2971  * readlink_args(char *path, char *buf, int count)
2972  *
2973  * Return target name of a symbolic link.
2974  */
2975 int
2976 sys_readlink(struct readlink_args *uap)
2977 {
2978 	struct nlookupdata nd;
2979 	int error;
2980 
2981 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2982 	if (error == 0) {
2983 		error = kern_readlink(&nd, uap->buf, uap->count,
2984 					&uap->sysmsg_result);
2985 	}
2986 	nlookup_done(&nd);
2987 	return (error);
2988 }
2989 
2990 /*
2991  * readlinkat_args(int fd, char *path, char *buf, size_t bufsize)
2992  *
2993  * Return target name of a symbolic link.  The path is relative to the
2994  * directory associated with fd.
2995  */
2996 int
2997 sys_readlinkat(struct readlinkat_args *uap)
2998 {
2999 	struct nlookupdata nd;
3000 	struct file *fp;
3001 	int error;
3002 
3003 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
3004 	if (error == 0) {
3005 		error = kern_readlink(&nd, uap->buf, uap->bufsize,
3006 					&uap->sysmsg_result);
3007 	}
3008 	nlookup_done_at(&nd, fp);
3009 	return (error);
3010 }
3011 
3012 static int
3013 setfflags(struct vnode *vp, int flags)
3014 {
3015 	struct thread *td = curthread;
3016 	int error;
3017 	struct vattr vattr;
3018 
3019 	/*
3020 	 * Prevent non-root users from setting flags on devices.  When
3021 	 * a device is reused, users can retain ownership of the device
3022 	 * if they are allowed to set flags and programs assume that
3023 	 * chown can't fail when done as root.
3024 	 */
3025 	if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
3026 	    ((error = priv_check_cred(td->td_ucred, PRIV_VFS_CHFLAGS_DEV, 0)) != 0))
3027 		return (error);
3028 
3029 	/*
3030 	 * note: vget is required for any operation that might mod the vnode
3031 	 * so VINACTIVE is properly cleared.
3032 	 */
3033 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3034 		VATTR_NULL(&vattr);
3035 		vattr.va_flags = flags;
3036 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3037 		vput(vp);
3038 	}
3039 	return (error);
3040 }
3041 
3042 /*
3043  * chflags(char *path, int flags)
3044  *
3045  * Change flags of a file given a path name.
3046  */
3047 int
3048 sys_chflags(struct chflags_args *uap)
3049 {
3050 	struct nlookupdata nd;
3051 	struct vnode *vp;
3052 	int error;
3053 
3054 	vp = NULL;
3055 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3056 	if (error == 0)
3057 		error = nlookup(&nd);
3058 	if (error == 0)
3059 		error = ncp_writechk(&nd.nl_nch);
3060 	if (error == 0)
3061 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3062 	nlookup_done(&nd);
3063 	if (error == 0) {
3064 		error = setfflags(vp, uap->flags);
3065 		vrele(vp);
3066 	}
3067 	return (error);
3068 }
3069 
3070 /*
3071  * lchflags(char *path, int flags)
3072  *
3073  * Change flags of a file given a path name, but don't follow symlinks.
3074  */
3075 int
3076 sys_lchflags(struct lchflags_args *uap)
3077 {
3078 	struct nlookupdata nd;
3079 	struct vnode *vp;
3080 	int error;
3081 
3082 	vp = NULL;
3083 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3084 	if (error == 0)
3085 		error = nlookup(&nd);
3086 	if (error == 0)
3087 		error = ncp_writechk(&nd.nl_nch);
3088 	if (error == 0)
3089 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3090 	nlookup_done(&nd);
3091 	if (error == 0) {
3092 		error = setfflags(vp, uap->flags);
3093 		vrele(vp);
3094 	}
3095 	return (error);
3096 }
3097 
3098 /*
3099  * fchflags_args(int fd, int flags)
3100  *
3101  * Change flags of a file given a file descriptor.
3102  */
3103 int
3104 sys_fchflags(struct fchflags_args *uap)
3105 {
3106 	struct thread *td = curthread;
3107 	struct proc *p = td->td_proc;
3108 	struct file *fp;
3109 	int error;
3110 
3111 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3112 		return (error);
3113 	if (fp->f_nchandle.ncp)
3114 		error = ncp_writechk(&fp->f_nchandle);
3115 	if (error == 0)
3116 		error = setfflags((struct vnode *) fp->f_data, uap->flags);
3117 	fdrop(fp);
3118 	return (error);
3119 }
3120 
3121 /*
3122  * chflagsat_args(int fd, const char *path, int flags, int atflags)
3123  * change flags given a pathname relative to a filedescriptor
3124  */
3125 int sys_chflagsat(struct chflagsat_args *uap)
3126 {
3127 	struct nlookupdata nd;
3128 	struct vnode *vp;
3129 	struct file *fp;
3130 	int error;
3131 	int lookupflags;
3132 
3133 	if (uap->atflags & ~AT_SYMLINK_NOFOLLOW)
3134 		return (EINVAL);
3135 
3136 	lookupflags = (uap->atflags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3137 
3138 	vp = NULL;
3139 	error = nlookup_init_at(&nd, &fp, uap->fd,  uap->path, UIO_USERSPACE, lookupflags);
3140 	if (error == 0)
3141 		error = nlookup(&nd);
3142 	if (error == 0)
3143 		error = ncp_writechk(&nd.nl_nch);
3144 	if (error == 0)
3145 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3146 	nlookup_done_at(&nd, fp);
3147 	if (error == 0) {
3148 		error = setfflags(vp, uap->flags);
3149 		vrele(vp);
3150 	}
3151 	return (error);
3152 }
3153 
3154 
3155 static int
3156 setfmode(struct vnode *vp, int mode)
3157 {
3158 	struct thread *td = curthread;
3159 	int error;
3160 	struct vattr vattr;
3161 
3162 	/*
3163 	 * note: vget is required for any operation that might mod the vnode
3164 	 * so VINACTIVE is properly cleared.
3165 	 */
3166 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3167 		VATTR_NULL(&vattr);
3168 		vattr.va_mode = mode & ALLPERMS;
3169 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3170 		vput(vp);
3171 	}
3172 	return error;
3173 }
3174 
3175 int
3176 kern_chmod(struct nlookupdata *nd, int mode)
3177 {
3178 	struct vnode *vp;
3179 	int error;
3180 
3181 	if ((error = nlookup(nd)) != 0)
3182 		return (error);
3183 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3184 		return (error);
3185 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3186 		error = setfmode(vp, mode);
3187 	vrele(vp);
3188 	return (error);
3189 }
3190 
3191 /*
3192  * chmod_args(char *path, int mode)
3193  *
3194  * Change mode of a file given path name.
3195  */
3196 int
3197 sys_chmod(struct chmod_args *uap)
3198 {
3199 	struct nlookupdata nd;
3200 	int error;
3201 
3202 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3203 	if (error == 0)
3204 		error = kern_chmod(&nd, uap->mode);
3205 	nlookup_done(&nd);
3206 	return (error);
3207 }
3208 
3209 /*
3210  * lchmod_args(char *path, int mode)
3211  *
3212  * Change mode of a file given path name (don't follow links.)
3213  */
3214 int
3215 sys_lchmod(struct lchmod_args *uap)
3216 {
3217 	struct nlookupdata nd;
3218 	int error;
3219 
3220 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3221 	if (error == 0)
3222 		error = kern_chmod(&nd, uap->mode);
3223 	nlookup_done(&nd);
3224 	return (error);
3225 }
3226 
3227 /*
3228  * fchmod_args(int fd, int mode)
3229  *
3230  * Change mode of a file given a file descriptor.
3231  */
3232 int
3233 sys_fchmod(struct fchmod_args *uap)
3234 {
3235 	struct thread *td = curthread;
3236 	struct proc *p = td->td_proc;
3237 	struct file *fp;
3238 	int error;
3239 
3240 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3241 		return (error);
3242 	if (fp->f_nchandle.ncp)
3243 		error = ncp_writechk(&fp->f_nchandle);
3244 	if (error == 0)
3245 		error = setfmode((struct vnode *)fp->f_data, uap->mode);
3246 	fdrop(fp);
3247 	return (error);
3248 }
3249 
3250 /*
3251  * fchmodat_args(char *path, int mode)
3252  *
3253  * Change mode of a file pointed to by fd/path.
3254  */
3255 int
3256 sys_fchmodat(struct fchmodat_args *uap)
3257 {
3258 	struct nlookupdata nd;
3259 	struct file *fp;
3260 	int error;
3261 	int flags;
3262 
3263 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3264 		return (EINVAL);
3265 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3266 
3267 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3268 				UIO_USERSPACE, flags);
3269 	if (error == 0)
3270 		error = kern_chmod(&nd, uap->mode);
3271 	nlookup_done_at(&nd, fp);
3272 	return (error);
3273 }
3274 
3275 static int
3276 setfown(struct mount *mp, struct vnode *vp, uid_t uid, gid_t gid)
3277 {
3278 	struct thread *td = curthread;
3279 	int error;
3280 	struct vattr vattr;
3281 	uid_t o_uid;
3282 	gid_t o_gid;
3283 	uint64_t size;
3284 
3285 	/*
3286 	 * note: vget is required for any operation that might mod the vnode
3287 	 * so VINACTIVE is properly cleared.
3288 	 */
3289 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3290 		if ((error = VOP_GETATTR(vp, &vattr)) != 0)
3291 			return error;
3292 		o_uid = vattr.va_uid;
3293 		o_gid = vattr.va_gid;
3294 		size = vattr.va_size;
3295 
3296 		VATTR_NULL(&vattr);
3297 		vattr.va_uid = uid;
3298 		vattr.va_gid = gid;
3299 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3300 		vput(vp);
3301 	}
3302 
3303 	if (error == 0) {
3304 		if (uid == -1)
3305 			uid = o_uid;
3306 		if (gid == -1)
3307 			gid = o_gid;
3308 		VFS_ACCOUNT(mp, o_uid, o_gid, -size);
3309 		VFS_ACCOUNT(mp,   uid,   gid,  size);
3310 	}
3311 
3312 	return error;
3313 }
3314 
3315 int
3316 kern_chown(struct nlookupdata *nd, int uid, int gid)
3317 {
3318 	struct vnode *vp;
3319 	int error;
3320 
3321 	if ((error = nlookup(nd)) != 0)
3322 		return (error);
3323 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3324 		return (error);
3325 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3326 		error = setfown(nd->nl_nch.mount, vp, uid, gid);
3327 	vrele(vp);
3328 	return (error);
3329 }
3330 
3331 /*
3332  * chown(char *path, int uid, int gid)
3333  *
3334  * Set ownership given a path name.
3335  */
3336 int
3337 sys_chown(struct chown_args *uap)
3338 {
3339 	struct nlookupdata nd;
3340 	int error;
3341 
3342 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3343 	if (error == 0)
3344 		error = kern_chown(&nd, uap->uid, uap->gid);
3345 	nlookup_done(&nd);
3346 	return (error);
3347 }
3348 
3349 /*
3350  * lchown_args(char *path, int uid, int gid)
3351  *
3352  * Set ownership given a path name, do not cross symlinks.
3353  */
3354 int
3355 sys_lchown(struct lchown_args *uap)
3356 {
3357 	struct nlookupdata nd;
3358 	int error;
3359 
3360 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3361 	if (error == 0)
3362 		error = kern_chown(&nd, uap->uid, uap->gid);
3363 	nlookup_done(&nd);
3364 	return (error);
3365 }
3366 
3367 /*
3368  * fchown_args(int fd, int uid, int gid)
3369  *
3370  * Set ownership given a file descriptor.
3371  */
3372 int
3373 sys_fchown(struct fchown_args *uap)
3374 {
3375 	struct thread *td = curthread;
3376 	struct proc *p = td->td_proc;
3377 	struct file *fp;
3378 	int error;
3379 
3380 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3381 		return (error);
3382 	if (fp->f_nchandle.ncp)
3383 		error = ncp_writechk(&fp->f_nchandle);
3384 	if (error == 0)
3385 		error = setfown(p->p_fd->fd_ncdir.mount,
3386 			(struct vnode *)fp->f_data, uap->uid, uap->gid);
3387 	fdrop(fp);
3388 	return (error);
3389 }
3390 
3391 /*
3392  * fchownat(int fd, char *path, int uid, int gid, int flags)
3393  *
3394  * Set ownership of file pointed to by fd/path.
3395  */
3396 int
3397 sys_fchownat(struct fchownat_args *uap)
3398 {
3399 	struct nlookupdata nd;
3400 	struct file *fp;
3401 	int error;
3402 	int flags;
3403 
3404 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3405 		return (EINVAL);
3406 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3407 
3408 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3409 				UIO_USERSPACE, flags);
3410 	if (error == 0)
3411 		error = kern_chown(&nd, uap->uid, uap->gid);
3412 	nlookup_done_at(&nd, fp);
3413 	return (error);
3414 }
3415 
3416 
3417 static int
3418 getutimes(struct timeval *tvp, struct timespec *tsp)
3419 {
3420 	struct timeval tv[2];
3421 	int error;
3422 
3423 	if (tvp == NULL) {
3424 		microtime(&tv[0]);
3425 		TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
3426 		tsp[1] = tsp[0];
3427 	} else {
3428 		if ((error = itimerfix(tvp)) != 0)
3429 			return (error);
3430 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
3431 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
3432 	}
3433 	return 0;
3434 }
3435 
3436 static int
3437 getutimens(const struct timespec *ts, struct timespec *newts, int *nullflag)
3438 {
3439 	struct timespec tsnow;
3440 	int error;
3441 
3442 	*nullflag = 0;
3443 	nanotime(&tsnow);
3444 	if (ts == NULL) {
3445 		newts[0] = tsnow;
3446 		newts[1] = tsnow;
3447 		*nullflag = 1;
3448 		return (0);
3449 	}
3450 
3451 	newts[0] = ts[0];
3452 	newts[1] = ts[1];
3453 	if (newts[0].tv_nsec == UTIME_OMIT && newts[1].tv_nsec == UTIME_OMIT)
3454 		return (0);
3455 	if (newts[0].tv_nsec == UTIME_NOW && newts[1].tv_nsec == UTIME_NOW)
3456 		*nullflag = 1;
3457 
3458 	if (newts[0].tv_nsec == UTIME_OMIT)
3459 		newts[0].tv_sec = VNOVAL;
3460 	else if (newts[0].tv_nsec == UTIME_NOW)
3461 		newts[0] = tsnow;
3462 	else if ((error = itimespecfix(&newts[0])) != 0)
3463 		return (error);
3464 
3465 	if (newts[1].tv_nsec == UTIME_OMIT)
3466 		newts[1].tv_sec = VNOVAL;
3467 	else if (newts[1].tv_nsec == UTIME_NOW)
3468 		newts[1] = tsnow;
3469 	else if ((error = itimespecfix(&newts[1])) != 0)
3470 		return (error);
3471 
3472 	return (0);
3473 }
3474 
3475 static int
3476 setutimes(struct vnode *vp, struct vattr *vattr,
3477 	  const struct timespec *ts, int nullflag)
3478 {
3479 	struct thread *td = curthread;
3480 	int error;
3481 
3482 	VATTR_NULL(vattr);
3483 	vattr->va_atime = ts[0];
3484 	vattr->va_mtime = ts[1];
3485 	if (nullflag)
3486 		vattr->va_vaflags |= VA_UTIMES_NULL;
3487 	error = VOP_SETATTR(vp, vattr, td->td_ucred);
3488 
3489 	return error;
3490 }
3491 
3492 int
3493 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
3494 {
3495 	struct timespec ts[2];
3496 	int error;
3497 
3498 	if (tptr) {
3499 		if ((error = getutimes(tptr, ts)) != 0)
3500 			return (error);
3501 	}
3502 	error = kern_utimensat(nd, tptr ? ts : NULL, 0);
3503 	return (error);
3504 }
3505 
3506 /*
3507  * utimes_args(char *path, struct timeval *tptr)
3508  *
3509  * Set the access and modification times of a file.
3510  */
3511 int
3512 sys_utimes(struct utimes_args *uap)
3513 {
3514 	struct timeval tv[2];
3515 	struct nlookupdata nd;
3516 	int error;
3517 
3518 	if (uap->tptr) {
3519  		error = copyin(uap->tptr, tv, sizeof(tv));
3520 		if (error)
3521 			return (error);
3522 	}
3523 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3524 	if (error == 0)
3525 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3526 	nlookup_done(&nd);
3527 	return (error);
3528 }
3529 
3530 /*
3531  * lutimes_args(char *path, struct timeval *tptr)
3532  *
3533  * Set the access and modification times of a file.
3534  */
3535 int
3536 sys_lutimes(struct lutimes_args *uap)
3537 {
3538 	struct timeval tv[2];
3539 	struct nlookupdata nd;
3540 	int error;
3541 
3542 	if (uap->tptr) {
3543 		error = copyin(uap->tptr, tv, sizeof(tv));
3544 		if (error)
3545 			return (error);
3546 	}
3547 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3548 	if (error == 0)
3549 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3550 	nlookup_done(&nd);
3551 	return (error);
3552 }
3553 
3554 /*
3555  * Set utimes on a file descriptor.  The creds used to open the
3556  * file are used to determine whether the operation is allowed
3557  * or not.
3558  */
3559 int
3560 kern_futimens(int fd, struct timespec *ts)
3561 {
3562 	struct thread *td = curthread;
3563 	struct proc *p = td->td_proc;
3564 	struct timespec newts[2];
3565 	struct file *fp;
3566 	struct vnode *vp;
3567 	struct vattr vattr;
3568 	int nullflag;
3569 	int error;
3570 
3571 	error = getutimens(ts, newts, &nullflag);
3572 	if (error)
3573 		return (error);
3574 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3575 		return (error);
3576 	if (fp->f_nchandle.ncp)
3577 		error = ncp_writechk(&fp->f_nchandle);
3578 	if (error == 0) {
3579 		vp = fp->f_data;
3580 		error = vget(vp, LK_EXCLUSIVE);
3581 		if (error == 0) {
3582 			error = VOP_GETATTR(vp, &vattr);
3583 			if (error == 0) {
3584 				error = naccess_va(&vattr, NLC_OWN | NLC_WRITE,
3585 						   fp->f_cred);
3586 			}
3587 			if (error == 0) {
3588 				error = setutimes(vp, &vattr, newts, nullflag);
3589 			}
3590 			vput(vp);
3591 		}
3592 	}
3593 	fdrop(fp);
3594 	return (error);
3595 }
3596 
3597 /*
3598  * futimens_args(int fd, struct timespec *ts)
3599  *
3600  * Set the access and modification times of a file.
3601  */
3602 int
3603 sys_futimens(struct futimens_args *uap)
3604 {
3605 	struct timespec ts[2];
3606 	int error;
3607 
3608 	if (uap->ts) {
3609 		error = copyin(uap->ts, ts, sizeof(ts));
3610 		if (error)
3611 			return (error);
3612 	}
3613 	error = kern_futimens(uap->fd, uap->ts ? ts : NULL);
3614 	return (error);
3615 }
3616 
3617 int
3618 kern_futimes(int fd, struct timeval *tptr)
3619 {
3620 	struct timespec ts[2];
3621 	int error;
3622 
3623 	if (tptr) {
3624 		if ((error = getutimes(tptr, ts)) != 0)
3625 			return (error);
3626 	}
3627 	error = kern_futimens(fd, tptr ? ts : NULL);
3628 	return (error);
3629 }
3630 
3631 /*
3632  * futimes_args(int fd, struct timeval *tptr)
3633  *
3634  * Set the access and modification times of a file.
3635  */
3636 int
3637 sys_futimes(struct futimes_args *uap)
3638 {
3639 	struct timeval tv[2];
3640 	int error;
3641 
3642 	if (uap->tptr) {
3643 		error = copyin(uap->tptr, tv, sizeof(tv));
3644 		if (error)
3645 			return (error);
3646 	}
3647 	error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
3648 	return (error);
3649 }
3650 
3651 int
3652 kern_utimensat(struct nlookupdata *nd, const struct timespec *ts, int flags)
3653 {
3654 	struct timespec newts[2];
3655 	struct vnode *vp;
3656 	struct vattr vattr;
3657 	int nullflag;
3658 	int error;
3659 
3660 	if (flags & ~AT_SYMLINK_NOFOLLOW)
3661 		return (EINVAL);
3662 
3663 	error = getutimens(ts, newts, &nullflag);
3664 	if (error)
3665 		return (error);
3666 
3667 	nd->nl_flags |= NLC_OWN | NLC_WRITE;
3668 	if ((error = nlookup(nd)) != 0)
3669 		return (error);
3670 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3671 		return (error);
3672 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3673 		return (error);
3674 	if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3675 		error = vget(vp, LK_EXCLUSIVE);
3676 		if (error == 0) {
3677 			error = setutimes(vp, &vattr, newts, nullflag);
3678 			vput(vp);
3679 		}
3680 	}
3681 	vrele(vp);
3682 	return (error);
3683 }
3684 
3685 /*
3686  * utimensat_args(int fd, const char *path, const struct timespec *ts, int flags);
3687  *
3688  * Set file access and modification times of a file.
3689  */
3690 int
3691 sys_utimensat(struct utimensat_args *uap)
3692 {
3693 	struct timespec ts[2];
3694 	struct nlookupdata nd;
3695 	struct file *fp;
3696 	int error;
3697 	int flags;
3698 
3699 	if (uap->ts) {
3700 		error = copyin(uap->ts, ts, sizeof(ts));
3701 		if (error)
3702 			return (error);
3703 	}
3704 
3705 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3706 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3707 	                        UIO_USERSPACE, flags);
3708 	if (error == 0)
3709 		error = kern_utimensat(&nd, uap->ts ? ts : NULL, uap->flags);
3710 	nlookup_done_at(&nd, fp);
3711 	return (error);
3712 }
3713 
3714 int
3715 kern_truncate(struct nlookupdata *nd, off_t length)
3716 {
3717 	struct vnode *vp;
3718 	struct vattr vattr;
3719 	int error;
3720 	uid_t uid = 0;
3721 	gid_t gid = 0;
3722 	uint64_t old_size = 0;
3723 
3724 	if (length < 0)
3725 		return(EINVAL);
3726 	nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE;
3727 	if ((error = nlookup(nd)) != 0)
3728 		return (error);
3729 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3730 		return (error);
3731 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3732 		return (error);
3733 	error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
3734 	if (error) {
3735 		vrele(vp);
3736 		return (error);
3737 	}
3738 	if (vp->v_type == VDIR) {
3739 		error = EISDIR;
3740 		goto done;
3741 	}
3742 	if (vfs_quota_enabled) {
3743 		error = VOP_GETATTR(vp, &vattr);
3744 		KASSERT(error == 0, ("kern_truncate(): VOP_GETATTR didn't return 0"));
3745 		uid = vattr.va_uid;
3746 		gid = vattr.va_gid;
3747 		old_size = vattr.va_size;
3748 	}
3749 
3750 	if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3751 		VATTR_NULL(&vattr);
3752 		vattr.va_size = length;
3753 		error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
3754 		VFS_ACCOUNT(nd->nl_nch.mount, uid, gid, length - old_size);
3755 	}
3756 done:
3757 	vput(vp);
3758 	return (error);
3759 }
3760 
3761 /*
3762  * truncate(char *path, int pad, off_t length)
3763  *
3764  * Truncate a file given its path name.
3765  */
3766 int
3767 sys_truncate(struct truncate_args *uap)
3768 {
3769 	struct nlookupdata nd;
3770 	int error;
3771 
3772 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3773 	if (error == 0)
3774 		error = kern_truncate(&nd, uap->length);
3775 	nlookup_done(&nd);
3776 	return error;
3777 }
3778 
3779 int
3780 kern_ftruncate(int fd, off_t length)
3781 {
3782 	struct thread *td = curthread;
3783 	struct proc *p = td->td_proc;
3784 	struct vattr vattr;
3785 	struct vnode *vp;
3786 	struct file *fp;
3787 	int error;
3788 	uid_t uid = 0;
3789 	gid_t gid = 0;
3790 	uint64_t old_size = 0;
3791 	struct mount *mp;
3792 
3793 	if (length < 0)
3794 		return(EINVAL);
3795 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3796 		return (error);
3797 	if (fp->f_nchandle.ncp) {
3798 		error = ncp_writechk(&fp->f_nchandle);
3799 		if (error)
3800 			goto done;
3801 	}
3802 	if ((fp->f_flag & FWRITE) == 0) {
3803 		error = EINVAL;
3804 		goto done;
3805 	}
3806 	if (fp->f_flag & FAPPENDONLY) {	/* inode was set s/uapnd */
3807 		error = EINVAL;
3808 		goto done;
3809 	}
3810 	vp = (struct vnode *)fp->f_data;
3811 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3812 	if (vp->v_type == VDIR) {
3813 		error = EISDIR;
3814 		vn_unlock(vp);
3815 		goto done;
3816 	}
3817 
3818 	if (vfs_quota_enabled) {
3819 		error = VOP_GETATTR(vp, &vattr);
3820 		KASSERT(error == 0, ("kern_ftruncate(): VOP_GETATTR didn't return 0"));
3821 		uid = vattr.va_uid;
3822 		gid = vattr.va_gid;
3823 		old_size = vattr.va_size;
3824 	}
3825 
3826 	if ((error = vn_writechk(vp, NULL)) == 0) {
3827 		VATTR_NULL(&vattr);
3828 		vattr.va_size = length;
3829 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
3830 		mp = vq_vptomp(vp);
3831 		VFS_ACCOUNT(mp, uid, gid, length - old_size);
3832 	}
3833 	vn_unlock(vp);
3834 done:
3835 	fdrop(fp);
3836 	return (error);
3837 }
3838 
3839 /*
3840  * ftruncate_args(int fd, int pad, off_t length)
3841  *
3842  * Truncate a file given a file descriptor.
3843  */
3844 int
3845 sys_ftruncate(struct ftruncate_args *uap)
3846 {
3847 	int error;
3848 
3849 	error = kern_ftruncate(uap->fd, uap->length);
3850 
3851 	return (error);
3852 }
3853 
3854 /*
3855  * fsync(int fd)
3856  *
3857  * Sync an open file.
3858  */
3859 int
3860 sys_fsync(struct fsync_args *uap)
3861 {
3862 	struct thread *td = curthread;
3863 	struct proc *p = td->td_proc;
3864 	struct vnode *vp;
3865 	struct file *fp;
3866 	vm_object_t obj;
3867 	int error;
3868 
3869 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3870 		return (error);
3871 	vp = (struct vnode *)fp->f_data;
3872 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3873 	if ((obj = vp->v_object) != NULL) {
3874 		if (vp->v_mount == NULL ||
3875 		    (vp->v_mount->mnt_kern_flag & MNTK_NOMSYNC) == 0) {
3876 			vm_object_page_clean(obj, 0, 0, 0);
3877 		}
3878 	}
3879 	error = VOP_FSYNC(vp, MNT_WAIT, VOP_FSYNC_SYSCALL);
3880 	if (error == 0 && vp->v_mount)
3881 		error = buf_fsync(vp);
3882 	vn_unlock(vp);
3883 	fdrop(fp);
3884 
3885 	return (error);
3886 }
3887 
3888 int
3889 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
3890 {
3891 	struct nchandle fnchd;
3892 	struct nchandle tnchd;
3893 	struct namecache *ncp;
3894 	struct vnode *fdvp;
3895 	struct vnode *tdvp;
3896 	struct mount *mp;
3897 	int error;
3898 	u_int fncp_gen;
3899 	u_int tncp_gen;
3900 
3901 	bwillinode(1);
3902 	fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC;
3903 	if ((error = nlookup(fromnd)) != 0)
3904 		return (error);
3905 	if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
3906 		return (ENOENT);
3907 	fnchd.mount = fromnd->nl_nch.mount;
3908 	cache_hold(&fnchd);
3909 
3910 	/*
3911 	 * unlock the source nch so we can lookup the target nch without
3912 	 * deadlocking.  The target may or may not exist so we do not check
3913 	 * for a target vp like kern_mkdir() and other creation functions do.
3914 	 *
3915 	 * The source and target directories are ref'd and rechecked after
3916 	 * everything is relocked to determine if the source or target file
3917 	 * has been renamed.
3918 	 */
3919 	KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
3920 	fromnd->nl_flags &= ~NLC_NCPISLOCKED;
3921 
3922 	fncp_gen = fromnd->nl_nch.ncp->nc_generation;
3923 
3924 	cache_unlock(&fromnd->nl_nch);
3925 
3926 	tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP;
3927 	if ((error = nlookup(tond)) != 0) {
3928 		cache_drop(&fnchd);
3929 		return (error);
3930 	}
3931 	tncp_gen = tond->nl_nch.ncp->nc_generation;
3932 
3933 	if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
3934 		cache_drop(&fnchd);
3935 		return (ENOENT);
3936 	}
3937 	tnchd.mount = tond->nl_nch.mount;
3938 	cache_hold(&tnchd);
3939 
3940 	/*
3941 	 * If the source and target are the same there is nothing to do
3942 	 */
3943 	if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
3944 		cache_drop(&fnchd);
3945 		cache_drop(&tnchd);
3946 		return (0);
3947 	}
3948 
3949 	/*
3950 	 * Mount points cannot be renamed or overwritten
3951 	 */
3952 	if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
3953 	    NCF_ISMOUNTPT
3954 	) {
3955 		cache_drop(&fnchd);
3956 		cache_drop(&tnchd);
3957 		return (EINVAL);
3958 	}
3959 
3960 	/*
3961 	 * Relock the source ncp.  cache_relock() will deal with any
3962 	 * deadlocks against the already-locked tond and will also
3963 	 * make sure both are resolved.
3964 	 *
3965 	 * NOTE AFTER RELOCKING: The source or target ncp may have become
3966 	 * invalid while they were unlocked, nc_vp and nc_mount could
3967 	 * be NULL.
3968 	 */
3969 	cache_relock(&fromnd->nl_nch, fromnd->nl_cred,
3970 		     &tond->nl_nch, tond->nl_cred);
3971 	fromnd->nl_flags |= NLC_NCPISLOCKED;
3972 
3973 	/*
3974 	 * If the namecache generation changed for either fromnd or tond,
3975 	 * we must retry.
3976 	 */
3977 	if (fromnd->nl_nch.ncp->nc_generation != fncp_gen ||
3978 	    tond->nl_nch.ncp->nc_generation != tncp_gen) {
3979 		kprintf("kern_rename: retry due to gen on: "
3980 			"\"%s\" -> \"%s\"\n",
3981 			fromnd->nl_nch.ncp->nc_name,
3982 			tond->nl_nch.ncp->nc_name);
3983 		cache_drop(&fnchd);
3984 		cache_drop(&tnchd);
3985 		return (EAGAIN);
3986 	}
3987 
3988 	/*
3989 	 * If either fromnd or tond are marked destroyed a ripout occured
3990 	 * out from under us and we must retry.
3991 	 */
3992 	if ((fromnd->nl_nch.ncp->nc_flag & (NCF_DESTROYED | NCF_UNRESOLVED)) ||
3993 	    fromnd->nl_nch.ncp->nc_vp == NULL ||
3994 	    (tond->nl_nch.ncp->nc_flag & NCF_DESTROYED)) {
3995 		kprintf("kern_rename: retry due to ripout on: "
3996 			"\"%s\" -> \"%s\"\n",
3997 			fromnd->nl_nch.ncp->nc_name,
3998 			tond->nl_nch.ncp->nc_name);
3999 		cache_drop(&fnchd);
4000 		cache_drop(&tnchd);
4001 		return (EAGAIN);
4002 	}
4003 
4004 	/*
4005 	 * Make sure the parent directories linkages are the same.
4006 	 * XXX shouldn't be needed any more w/ generation check above.
4007 	 */
4008 	if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
4009 	    tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
4010 		cache_drop(&fnchd);
4011 		cache_drop(&tnchd);
4012 		return (ENOENT);
4013 	}
4014 
4015 	/*
4016 	 * Both the source and target must be within the same filesystem and
4017 	 * in the same filesystem as their parent directories within the
4018 	 * namecache topology.
4019 	 *
4020 	 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
4021 	 */
4022 	mp = fnchd.mount;
4023 	if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
4024 	    mp != tond->nl_nch.mount) {
4025 		cache_drop(&fnchd);
4026 		cache_drop(&tnchd);
4027 		return (EXDEV);
4028 	}
4029 
4030 	/*
4031 	 * Make sure the mount point is writable
4032 	 */
4033 	if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
4034 		cache_drop(&fnchd);
4035 		cache_drop(&tnchd);
4036 		return (error);
4037 	}
4038 
4039 	/*
4040 	 * If the target exists and either the source or target is a directory,
4041 	 * then both must be directories.
4042 	 *
4043 	 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
4044 	 * have become NULL.
4045 	 */
4046 	if (tond->nl_nch.ncp->nc_vp) {
4047 		if (fromnd->nl_nch.ncp->nc_vp == NULL) {
4048 			error = ENOENT;
4049 		} else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
4050 			if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
4051 				error = ENOTDIR;
4052 		} else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
4053 			error = EISDIR;
4054 		}
4055 	}
4056 
4057 	/*
4058 	 * You cannot rename a source into itself or a subdirectory of itself.
4059 	 * We check this by travsersing the target directory upwards looking
4060 	 * for a match against the source.
4061 	 *
4062 	 * XXX MPSAFE
4063 	 */
4064 	if (error == 0) {
4065 		for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
4066 			if (fromnd->nl_nch.ncp == ncp) {
4067 				error = EINVAL;
4068 				break;
4069 			}
4070 		}
4071 	}
4072 
4073 	cache_drop(&fnchd);
4074 	cache_drop(&tnchd);
4075 
4076 	/*
4077 	 * Even though the namespaces are different, they may still represent
4078 	 * hardlinks to the same file.  The filesystem might have a hard time
4079 	 * with this so we issue a NREMOVE of the source instead of a NRENAME
4080 	 * when we detect the situation.
4081 	 */
4082 	if (error == 0) {
4083 		fdvp = fromnd->nl_dvp;
4084 		tdvp = tond->nl_dvp;
4085 		if (fdvp == NULL || tdvp == NULL) {
4086 			error = EPERM;
4087 		} else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
4088 			error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
4089 					    fromnd->nl_cred);
4090 		} else {
4091 			error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
4092 					    fdvp, tdvp, tond->nl_cred);
4093 		}
4094 	}
4095 	return (error);
4096 }
4097 
4098 /*
4099  * rename_args(char *from, char *to)
4100  *
4101  * Rename files.  Source and destination must either both be directories,
4102  * or both not be directories.  If target is a directory, it must be empty.
4103  */
4104 int
4105 sys_rename(struct rename_args *uap)
4106 {
4107 	struct nlookupdata fromnd, tond;
4108 	int error;
4109 
4110 	do {
4111 		error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
4112 		if (error == 0) {
4113 			error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
4114 			if (error == 0)
4115 				error = kern_rename(&fromnd, &tond);
4116 			nlookup_done(&tond);
4117 		}
4118 		nlookup_done(&fromnd);
4119 	} while (error == EAGAIN);
4120 	return (error);
4121 }
4122 
4123 /*
4124  * renameat_args(int oldfd, char *old, int newfd, char *new)
4125  *
4126  * Rename files using paths relative to the directories associated with
4127  * oldfd and newfd.  Source and destination must either both be directories,
4128  * or both not be directories.  If target is a directory, it must be empty.
4129  */
4130 int
4131 sys_renameat(struct renameat_args *uap)
4132 {
4133 	struct nlookupdata oldnd, newnd;
4134 	struct file *oldfp, *newfp;
4135 	int error;
4136 
4137 	do {
4138 		error = nlookup_init_at(&oldnd, &oldfp,
4139 					uap->oldfd, uap->old,
4140 					UIO_USERSPACE, 0);
4141 		if (error == 0) {
4142 			error = nlookup_init_at(&newnd, &newfp,
4143 						uap->newfd, uap->new,
4144 						UIO_USERSPACE, 0);
4145 			if (error == 0)
4146 				error = kern_rename(&oldnd, &newnd);
4147 			nlookup_done_at(&newnd, newfp);
4148 		}
4149 		nlookup_done_at(&oldnd, oldfp);
4150 	} while (error == EAGAIN);
4151 	return (error);
4152 }
4153 
4154 int
4155 kern_mkdir(struct nlookupdata *nd, int mode)
4156 {
4157 	struct thread *td = curthread;
4158 	struct proc *p = td->td_proc;
4159 	struct vnode *vp;
4160 	struct vattr vattr;
4161 	int error;
4162 
4163 	bwillinode(1);
4164 	nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
4165 	if ((error = nlookup(nd)) != 0)
4166 		return (error);
4167 
4168 	if (nd->nl_nch.ncp->nc_vp)
4169 		return (EEXIST);
4170 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
4171 		return (error);
4172 	VATTR_NULL(&vattr);
4173 	vattr.va_type = VDIR;
4174 	vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
4175 
4176 	vp = NULL;
4177 	error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, td->td_ucred, &vattr);
4178 	if (error == 0)
4179 		vput(vp);
4180 	return (error);
4181 }
4182 
4183 /*
4184  * mkdir_args(char *path, int mode)
4185  *
4186  * Make a directory file.
4187  */
4188 int
4189 sys_mkdir(struct mkdir_args *uap)
4190 {
4191 	struct nlookupdata nd;
4192 	int error;
4193 
4194 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
4195 	if (error == 0)
4196 		error = kern_mkdir(&nd, uap->mode);
4197 	nlookup_done(&nd);
4198 	return (error);
4199 }
4200 
4201 /*
4202  * mkdirat_args(int fd, char *path, mode_t mode)
4203  *
4204  * Make a directory file.  The path is relative to the directory associated
4205  * with fd.
4206  */
4207 int
4208 sys_mkdirat(struct mkdirat_args *uap)
4209 {
4210 	struct nlookupdata nd;
4211 	struct file *fp;
4212 	int error;
4213 
4214 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
4215 	if (error == 0)
4216 		error = kern_mkdir(&nd, uap->mode);
4217 	nlookup_done_at(&nd, fp);
4218 	return (error);
4219 }
4220 
4221 int
4222 kern_rmdir(struct nlookupdata *nd)
4223 {
4224 	int error;
4225 
4226 	bwillinode(1);
4227 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
4228 	if ((error = nlookup(nd)) != 0)
4229 		return (error);
4230 
4231 	/*
4232 	 * Do not allow directories representing mount points to be
4233 	 * deleted, even if empty.  Check write perms on mount point
4234 	 * in case the vnode is aliased (aka nullfs).
4235 	 */
4236 	if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
4237 		return (EBUSY);
4238 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
4239 		return (error);
4240 	error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
4241 	return (error);
4242 }
4243 
4244 /*
4245  * rmdir_args(char *path)
4246  *
4247  * Remove a directory file.
4248  */
4249 int
4250 sys_rmdir(struct rmdir_args *uap)
4251 {
4252 	struct nlookupdata nd;
4253 	int error;
4254 
4255 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
4256 	if (error == 0)
4257 		error = kern_rmdir(&nd);
4258 	nlookup_done(&nd);
4259 	return (error);
4260 }
4261 
4262 int
4263 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
4264 		   enum uio_seg direction)
4265 {
4266 	struct thread *td = curthread;
4267 	struct proc *p = td->td_proc;
4268 	struct vnode *vp;
4269 	struct file *fp;
4270 	struct uio auio;
4271 	struct iovec aiov;
4272 	off_t loff;
4273 	int error, eofflag;
4274 
4275 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
4276 		return (error);
4277 	if ((fp->f_flag & FREAD) == 0) {
4278 		error = EBADF;
4279 		goto done;
4280 	}
4281 	vp = (struct vnode *)fp->f_data;
4282 	if (vp->v_type != VDIR) {
4283 		error = EINVAL;
4284 		goto done;
4285 	}
4286 	aiov.iov_base = buf;
4287 	aiov.iov_len = count;
4288 	auio.uio_iov = &aiov;
4289 	auio.uio_iovcnt = 1;
4290 	auio.uio_rw = UIO_READ;
4291 	auio.uio_segflg = direction;
4292 	auio.uio_td = td;
4293 	auio.uio_resid = count;
4294 	loff = auio.uio_offset = fp->f_offset;
4295 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
4296 	fp->f_offset = auio.uio_offset;
4297 	if (error)
4298 		goto done;
4299 
4300 	/*
4301 	 * WARNING!  *basep may not be wide enough to accomodate the
4302 	 * seek offset.   XXX should we hack this to return the upper 32 bits
4303 	 * for offsets greater then 4G?
4304 	 */
4305 	if (basep) {
4306 		*basep = (long)loff;
4307 	}
4308 	*res = count - auio.uio_resid;
4309 done:
4310 	fdrop(fp);
4311 	return (error);
4312 }
4313 
4314 /*
4315  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
4316  *
4317  * Read a block of directory entries in a file system independent format.
4318  */
4319 int
4320 sys_getdirentries(struct getdirentries_args *uap)
4321 {
4322 	long base;
4323 	int error;
4324 
4325 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
4326 				   &uap->sysmsg_result, UIO_USERSPACE);
4327 
4328 	if (error == 0 && uap->basep)
4329 		error = copyout(&base, uap->basep, sizeof(*uap->basep));
4330 	return (error);
4331 }
4332 
4333 /*
4334  * getdents_args(int fd, char *buf, size_t count)
4335  */
4336 int
4337 sys_getdents(struct getdents_args *uap)
4338 {
4339 	int error;
4340 
4341 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
4342 				   &uap->sysmsg_result, UIO_USERSPACE);
4343 
4344 	return (error);
4345 }
4346 
4347 /*
4348  * Set the mode mask for creation of filesystem nodes.
4349  *
4350  * umask(int newmask)
4351  */
4352 int
4353 sys_umask(struct umask_args *uap)
4354 {
4355 	struct thread *td = curthread;
4356 	struct proc *p = td->td_proc;
4357 	struct filedesc *fdp;
4358 
4359 	fdp = p->p_fd;
4360 	uap->sysmsg_result = fdp->fd_cmask;
4361 	fdp->fd_cmask = uap->newmask & ALLPERMS;
4362 	return (0);
4363 }
4364 
4365 /*
4366  * revoke(char *path)
4367  *
4368  * Void all references to file by ripping underlying filesystem
4369  * away from vnode.
4370  */
4371 int
4372 sys_revoke(struct revoke_args *uap)
4373 {
4374 	struct nlookupdata nd;
4375 	struct vattr vattr;
4376 	struct vnode *vp;
4377 	struct ucred *cred;
4378 	int error;
4379 
4380 	vp = NULL;
4381 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4382 	if (error == 0)
4383 		error = nlookup(&nd);
4384 	if (error == 0)
4385 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4386 	cred = crhold(nd.nl_cred);
4387 	nlookup_done(&nd);
4388 	if (error == 0) {
4389 		if (error == 0)
4390 			error = VOP_GETATTR(vp, &vattr);
4391 		if (error == 0 && cred->cr_uid != vattr.va_uid)
4392 			error = priv_check_cred(cred, PRIV_VFS_REVOKE, 0);
4393 		if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) {
4394 			if (vcount(vp) > 0)
4395 				error = vrevoke(vp, cred);
4396 		} else if (error == 0) {
4397 			error = vrevoke(vp, cred);
4398 		}
4399 		vrele(vp);
4400 	}
4401 	if (cred)
4402 		crfree(cred);
4403 	return (error);
4404 }
4405 
4406 /*
4407  * getfh_args(char *fname, fhandle_t *fhp)
4408  *
4409  * Get (NFS) file handle
4410  *
4411  * NOTE: We use the fsid of the covering mount, even if it is a nullfs
4412  * mount.  This allows nullfs mounts to be explicitly exported.
4413  *
4414  * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe.
4415  *
4416  * 	    nullfs mounts of subdirectories are not safe.  That is, it will
4417  *	    work, but you do not really have protection against access to
4418  *	    the related parent directories.
4419  */
4420 int
4421 sys_getfh(struct getfh_args *uap)
4422 {
4423 	struct thread *td = curthread;
4424 	struct nlookupdata nd;
4425 	fhandle_t fh;
4426 	struct vnode *vp;
4427 	struct mount *mp;
4428 	int error;
4429 
4430 	/*
4431 	 * Must be super user
4432 	 */
4433 	if ((error = priv_check(td, PRIV_ROOT)) != 0)
4434 		return (error);
4435 
4436 	vp = NULL;
4437 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
4438 	if (error == 0)
4439 		error = nlookup(&nd);
4440 	if (error == 0)
4441 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4442 	mp = nd.nl_nch.mount;
4443 	nlookup_done(&nd);
4444 	if (error == 0) {
4445 		bzero(&fh, sizeof(fh));
4446 		fh.fh_fsid = mp->mnt_stat.f_fsid;
4447 		error = VFS_VPTOFH(vp, &fh.fh_fid);
4448 		vput(vp);
4449 		if (error == 0)
4450 			error = copyout(&fh, uap->fhp, sizeof(fh));
4451 	}
4452 	return (error);
4453 }
4454 
4455 /*
4456  * fhopen_args(const struct fhandle *u_fhp, int flags)
4457  *
4458  * syscall for the rpc.lockd to use to translate a NFS file handle into
4459  * an open descriptor.
4460  *
4461  * warning: do not remove the priv_check() call or this becomes one giant
4462  * security hole.
4463  */
4464 int
4465 sys_fhopen(struct fhopen_args *uap)
4466 {
4467 	struct thread *td = curthread;
4468 	struct filedesc *fdp = td->td_proc->p_fd;
4469 	struct mount *mp;
4470 	struct vnode *vp;
4471 	struct fhandle fhp;
4472 	struct vattr vat;
4473 	struct vattr *vap = &vat;
4474 	struct flock lf;
4475 	int fmode, mode, error = 0, type;
4476 	struct file *nfp;
4477 	struct file *fp;
4478 	int indx;
4479 
4480 	/*
4481 	 * Must be super user
4482 	 */
4483 	error = priv_check(td, PRIV_ROOT);
4484 	if (error)
4485 		return (error);
4486 
4487 	fmode = FFLAGS(uap->flags);
4488 
4489 	/*
4490 	 * Why not allow a non-read/write open for our lockd?
4491 	 */
4492 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
4493 		return (EINVAL);
4494 	error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
4495 	if (error)
4496 		return(error);
4497 
4498 	/*
4499 	 * Find the mount point
4500 	 */
4501 	mp = vfs_getvfs(&fhp.fh_fsid);
4502 	if (mp == NULL) {
4503 		error = ESTALE;
4504 		goto  done;
4505 	}
4506 	/* now give me my vnode, it gets returned to me locked */
4507 	error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp);
4508 	if (error)
4509 		goto done;
4510  	/*
4511 	 * from now on we have to make sure not
4512 	 * to forget about the vnode
4513 	 * any error that causes an abort must vput(vp)
4514 	 * just set error = err and 'goto bad;'.
4515 	 */
4516 
4517 	/*
4518 	 * from vn_open
4519 	 */
4520 	if (vp->v_type == VLNK) {
4521 		error = EMLINK;
4522 		goto bad;
4523 	}
4524 	if (vp->v_type == VSOCK) {
4525 		error = EOPNOTSUPP;
4526 		goto bad;
4527 	}
4528 	mode = 0;
4529 	if (fmode & (FWRITE | O_TRUNC)) {
4530 		if (vp->v_type == VDIR) {
4531 			error = EISDIR;
4532 			goto bad;
4533 		}
4534 		error = vn_writechk(vp, NULL);
4535 		if (error)
4536 			goto bad;
4537 		mode |= VWRITE;
4538 	}
4539 	if (fmode & FREAD)
4540 		mode |= VREAD;
4541 	if (mode) {
4542 		error = VOP_ACCESS(vp, mode, td->td_ucred);
4543 		if (error)
4544 			goto bad;
4545 	}
4546 	if (fmode & O_TRUNC) {
4547 		vn_unlock(vp);				/* XXX */
4548 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
4549 		VATTR_NULL(vap);
4550 		vap->va_size = 0;
4551 		error = VOP_SETATTR(vp, vap, td->td_ucred);
4552 		if (error)
4553 			goto bad;
4554 	}
4555 
4556 	/*
4557 	 * VOP_OPEN needs the file pointer so it can potentially override
4558 	 * it.
4559 	 *
4560 	 * WARNING! no f_nchandle will be associated when fhopen()ing a
4561 	 * directory.  XXX
4562 	 */
4563 	if ((error = falloc(td->td_lwp, &nfp, &indx)) != 0)
4564 		goto bad;
4565 	fp = nfp;
4566 
4567 	error = VOP_OPEN(vp, fmode, td->td_ucred, fp);
4568 	if (error) {
4569 		/*
4570 		 * setting f_ops this way prevents VOP_CLOSE from being
4571 		 * called or fdrop() releasing the vp from v_data.   Since
4572 		 * the VOP_OPEN failed we don't want to VOP_CLOSE.
4573 		 */
4574 		fp->f_ops = &badfileops;
4575 		fp->f_data = NULL;
4576 		goto bad_drop;
4577 	}
4578 
4579 	/*
4580 	 * The fp is given its own reference, we still have our ref and lock.
4581 	 *
4582 	 * Assert that all regular files must be created with a VM object.
4583 	 */
4584 	if (vp->v_type == VREG && vp->v_object == NULL) {
4585 		kprintf("fhopen: regular file did not have VM object: %p\n", vp);
4586 		goto bad_drop;
4587 	}
4588 
4589 	/*
4590 	 * The open was successful.  Handle any locking requirements.
4591 	 */
4592 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
4593 		lf.l_whence = SEEK_SET;
4594 		lf.l_start = 0;
4595 		lf.l_len = 0;
4596 		if (fmode & O_EXLOCK)
4597 			lf.l_type = F_WRLCK;
4598 		else
4599 			lf.l_type = F_RDLCK;
4600 		if (fmode & FNONBLOCK)
4601 			type = 0;
4602 		else
4603 			type = F_WAIT;
4604 		vn_unlock(vp);
4605 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
4606 			/*
4607 			 * release our private reference.
4608 			 */
4609 			fsetfd(fdp, NULL, indx);
4610 			fdrop(fp);
4611 			vrele(vp);
4612 			goto done;
4613 		}
4614 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4615 		atomic_set_int(&fp->f_flag, FHASLOCK);	/* race ok */
4616 	}
4617 
4618 	/*
4619 	 * Clean up.  Associate the file pointer with the previously
4620 	 * reserved descriptor and return it.
4621 	 */
4622 	vput(vp);
4623 	if (uap->flags & O_CLOEXEC)
4624 		fdp->fd_files[indx].fileflags |= UF_EXCLOSE;
4625 	fsetfd(fdp, fp, indx);
4626 	fdrop(fp);
4627 	uap->sysmsg_result = indx;
4628 	return (error);
4629 
4630 bad_drop:
4631 	fsetfd(fdp, NULL, indx);
4632 	fdrop(fp);
4633 bad:
4634 	vput(vp);
4635 done:
4636 	return (error);
4637 }
4638 
4639 /*
4640  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
4641  */
4642 int
4643 sys_fhstat(struct fhstat_args *uap)
4644 {
4645 	struct thread *td = curthread;
4646 	struct stat sb;
4647 	fhandle_t fh;
4648 	struct mount *mp;
4649 	struct vnode *vp;
4650 	int error;
4651 
4652 	/*
4653 	 * Must be super user
4654 	 */
4655 	error = priv_check(td, PRIV_ROOT);
4656 	if (error)
4657 		return (error);
4658 
4659 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
4660 	if (error)
4661 		return (error);
4662 
4663 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
4664 		error = ESTALE;
4665 	if (error == 0) {
4666 		if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) == 0) {
4667 			error = vn_stat(vp, &sb, td->td_ucred);
4668 			vput(vp);
4669 		}
4670 	}
4671 	if (error == 0)
4672 		error = copyout(&sb, uap->sb, sizeof(sb));
4673 	return (error);
4674 }
4675 
4676 /*
4677  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
4678  */
4679 int
4680 sys_fhstatfs(struct fhstatfs_args *uap)
4681 {
4682 	struct thread *td = curthread;
4683 	struct proc *p = td->td_proc;
4684 	struct statfs *sp;
4685 	struct mount *mp;
4686 	struct vnode *vp;
4687 	struct statfs sb;
4688 	char *fullpath, *freepath;
4689 	fhandle_t fh;
4690 	int error;
4691 
4692 	/*
4693 	 * Must be super user
4694 	 */
4695 	if ((error = priv_check(td, PRIV_ROOT)))
4696 		return (error);
4697 
4698 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4699 		return (error);
4700 
4701 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4702 		error = ESTALE;
4703 		goto done;
4704 	}
4705 	if (p != NULL && !chroot_visible_mnt(mp, p)) {
4706 		error = ESTALE;
4707 		goto done;
4708 	}
4709 
4710 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) != 0)
4711 		goto done;
4712 	mp = vp->v_mount;
4713 	sp = &mp->mnt_stat;
4714 	vput(vp);
4715 	if ((error = VFS_STATFS(mp, sp, td->td_ucred)) != 0)
4716 		goto done;
4717 
4718 	error = mount_path(p, mp, &fullpath, &freepath);
4719 	if (error)
4720 		goto done;
4721 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4722 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
4723 	kfree(freepath, M_TEMP);
4724 
4725 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
4726 	if (priv_check(td, PRIV_ROOT)) {
4727 		bcopy(sp, &sb, sizeof(sb));
4728 		sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
4729 		sp = &sb;
4730 	}
4731 	error = copyout(sp, uap->buf, sizeof(*sp));
4732 done:
4733 	return (error);
4734 }
4735 
4736 /*
4737  * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
4738  */
4739 int
4740 sys_fhstatvfs(struct fhstatvfs_args *uap)
4741 {
4742 	struct thread *td = curthread;
4743 	struct proc *p = td->td_proc;
4744 	struct statvfs *sp;
4745 	struct mount *mp;
4746 	struct vnode *vp;
4747 	fhandle_t fh;
4748 	int error;
4749 
4750 	/*
4751 	 * Must be super user
4752 	 */
4753 	if ((error = priv_check(td, PRIV_ROOT)))
4754 		return (error);
4755 
4756 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4757 		return (error);
4758 
4759 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4760 		error = ESTALE;
4761 		goto done;
4762 	}
4763 	if (p != NULL && !chroot_visible_mnt(mp, p)) {
4764 		error = ESTALE;
4765 		goto done;
4766 	}
4767 
4768 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
4769 		goto done;
4770 	mp = vp->v_mount;
4771 	sp = &mp->mnt_vstat;
4772 	vput(vp);
4773 	if ((error = VFS_STATVFS(mp, sp, td->td_ucred)) != 0)
4774 		goto done;
4775 
4776 	sp->f_flag = 0;
4777 	if (mp->mnt_flag & MNT_RDONLY)
4778 		sp->f_flag |= ST_RDONLY;
4779 	if (mp->mnt_flag & MNT_NOSUID)
4780 		sp->f_flag |= ST_NOSUID;
4781 	error = copyout(sp, uap->buf, sizeof(*sp));
4782 done:
4783 	return (error);
4784 }
4785 
4786 
4787 /*
4788  * Syscall to push extended attribute configuration information into the
4789  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
4790  * a command (int cmd), and attribute name and misc data.  For now, the
4791  * attribute name is left in userspace for consumption by the VFS_op.
4792  * It will probably be changed to be copied into sysspace by the
4793  * syscall in the future, once issues with various consumers of the
4794  * attribute code have raised their hands.
4795  *
4796  * Currently this is used only by UFS Extended Attributes.
4797  */
4798 int
4799 sys_extattrctl(struct extattrctl_args *uap)
4800 {
4801 	struct nlookupdata nd;
4802 	struct vnode *vp;
4803 	char attrname[EXTATTR_MAXNAMELEN];
4804 	int error;
4805 	size_t size;
4806 
4807 	attrname[0] = 0;
4808 	vp = NULL;
4809 	error = 0;
4810 
4811 	if (error == 0 && uap->filename) {
4812 		error = nlookup_init(&nd, uap->filename, UIO_USERSPACE,
4813 				     NLC_FOLLOW);
4814 		if (error == 0)
4815 			error = nlookup(&nd);
4816 		if (error == 0)
4817 			error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4818 		nlookup_done(&nd);
4819 	}
4820 
4821 	if (error == 0 && uap->attrname) {
4822 		error = copyinstr(uap->attrname, attrname, EXTATTR_MAXNAMELEN,
4823 				  &size);
4824 	}
4825 
4826 	if (error == 0) {
4827 		error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4828 		if (error == 0)
4829 			error = nlookup(&nd);
4830 		if (error == 0)
4831 			error = ncp_writechk(&nd.nl_nch);
4832 		if (error == 0) {
4833 			error = VFS_EXTATTRCTL(nd.nl_nch.mount, uap->cmd, vp,
4834 					       uap->attrnamespace,
4835 					       uap->attrname, nd.nl_cred);
4836 		}
4837 		nlookup_done(&nd);
4838 	}
4839 
4840 	return (error);
4841 }
4842 
4843 /*
4844  * Syscall to get a named extended attribute on a file or directory.
4845  */
4846 int
4847 sys_extattr_set_file(struct extattr_set_file_args *uap)
4848 {
4849 	char attrname[EXTATTR_MAXNAMELEN];
4850 	struct nlookupdata nd;
4851 	struct vnode *vp;
4852 	struct uio auio;
4853 	struct iovec aiov;
4854 	int error;
4855 
4856 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4857 	if (error)
4858 		return (error);
4859 
4860 	vp = NULL;
4861 
4862 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4863 	if (error == 0)
4864 		error = nlookup(&nd);
4865 	if (error == 0)
4866 		error = ncp_writechk(&nd.nl_nch);
4867 	if (error == 0)
4868 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4869 	if (error) {
4870 		nlookup_done(&nd);
4871 		return (error);
4872 	}
4873 
4874 	bzero(&auio, sizeof(auio));
4875 	aiov.iov_base = uap->data;
4876 	aiov.iov_len = uap->nbytes;
4877 	auio.uio_iov = &aiov;
4878 	auio.uio_iovcnt = 1;
4879 	auio.uio_offset = 0;
4880 	auio.uio_resid = uap->nbytes;
4881 	auio.uio_rw = UIO_WRITE;
4882 	auio.uio_td = curthread;
4883 
4884 	error = VOP_SETEXTATTR(vp, uap->attrnamespace, attrname,
4885 			       &auio, nd.nl_cred);
4886 
4887 	vput(vp);
4888 	nlookup_done(&nd);
4889 	return (error);
4890 }
4891 
4892 /*
4893  * Syscall to get a named extended attribute on a file or directory.
4894  */
4895 int
4896 sys_extattr_get_file(struct extattr_get_file_args *uap)
4897 {
4898 	char attrname[EXTATTR_MAXNAMELEN];
4899 	struct nlookupdata nd;
4900 	struct uio auio;
4901 	struct iovec aiov;
4902 	struct vnode *vp;
4903 	int error;
4904 
4905 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4906 	if (error)
4907 		return (error);
4908 
4909 	vp = NULL;
4910 
4911 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4912 	if (error == 0)
4913 		error = nlookup(&nd);
4914 	if (error == 0)
4915 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_SHARED, &vp);
4916 	if (error) {
4917 		nlookup_done(&nd);
4918 		return (error);
4919 	}
4920 
4921 	bzero(&auio, sizeof(auio));
4922 	aiov.iov_base = uap->data;
4923 	aiov.iov_len = uap->nbytes;
4924 	auio.uio_iov = &aiov;
4925 	auio.uio_iovcnt = 1;
4926 	auio.uio_offset = 0;
4927 	auio.uio_resid = uap->nbytes;
4928 	auio.uio_rw = UIO_READ;
4929 	auio.uio_td = curthread;
4930 
4931 	error = VOP_GETEXTATTR(vp, uap->attrnamespace, attrname,
4932 				&auio, nd.nl_cred);
4933 	uap->sysmsg_result = uap->nbytes - auio.uio_resid;
4934 
4935 	vput(vp);
4936 	nlookup_done(&nd);
4937 	return(error);
4938 }
4939 
4940 /*
4941  * Syscall to delete a named extended attribute from a file or directory.
4942  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
4943  */
4944 int
4945 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
4946 {
4947 	char attrname[EXTATTR_MAXNAMELEN];
4948 	struct nlookupdata nd;
4949 	struct vnode *vp;
4950 	int error;
4951 
4952 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4953 	if (error)
4954 		return(error);
4955 
4956 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4957 	if (error == 0)
4958 		error = nlookup(&nd);
4959 	if (error == 0)
4960 		error = ncp_writechk(&nd.nl_nch);
4961 	if (error == 0) {
4962 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4963 		if (error == 0) {
4964 			error = VOP_SETEXTATTR(vp, uap->attrnamespace,
4965 					       attrname, NULL, nd.nl_cred);
4966 			vput(vp);
4967 		}
4968 	}
4969 	nlookup_done(&nd);
4970 	return(error);
4971 }
4972 
4973 /*
4974  * Determine if the mount is visible to the process.
4975  */
4976 static int
4977 chroot_visible_mnt(struct mount *mp, struct proc *p)
4978 {
4979 	struct nchandle nch;
4980 
4981 	/*
4982 	 * Traverse from the mount point upwards.  If we hit the process
4983 	 * root then the mount point is visible to the process.
4984 	 */
4985 	nch = mp->mnt_ncmountpt;
4986 	while (nch.ncp) {
4987 		if (nch.mount == p->p_fd->fd_nrdir.mount &&
4988 		    nch.ncp == p->p_fd->fd_nrdir.ncp) {
4989 			return(1);
4990 		}
4991 		if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
4992 			nch = nch.mount->mnt_ncmounton;
4993 		} else {
4994 			nch.ncp = nch.ncp->nc_parent;
4995 		}
4996 	}
4997 
4998 	/*
4999 	 * If the mount point is not visible to the process, but the
5000 	 * process root is in a subdirectory of the mount, return
5001 	 * TRUE anyway.
5002 	 */
5003 	if (p->p_fd->fd_nrdir.mount == mp)
5004 		return(1);
5005 
5006 	return(0);
5007 }
5008 
5009