xref: /dflybsd-src/sys/kern/vfs_syscalls.c (revision ca86d83e7d8d6bfef814ef3683c37d99ad62f11c)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_syscalls.c	8.13 (Berkeley) 4/15/94
39  * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $
40  */
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/buf.h>
45 #include <sys/conf.h>
46 #include <sys/sysent.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
49 #include <sys/mountctl.h>
50 #include <sys/sysproto.h>
51 #include <sys/filedesc.h>
52 #include <sys/kernel.h>
53 #include <sys/fcntl.h>
54 #include <sys/file.h>
55 #include <sys/linker.h>
56 #include <sys/stat.h>
57 #include <sys/unistd.h>
58 #include <sys/vnode.h>
59 #include <sys/proc.h>
60 #include <sys/priv.h>
61 #include <sys/jail.h>
62 #include <sys/namei.h>
63 #include <sys/nlookup.h>
64 #include <sys/dirent.h>
65 #include <sys/extattr.h>
66 #include <sys/spinlock.h>
67 #include <sys/kern_syscall.h>
68 #include <sys/objcache.h>
69 #include <sys/sysctl.h>
70 
71 #include <sys/buf2.h>
72 #include <sys/file2.h>
73 #include <sys/spinlock2.h>
74 #include <sys/mplock2.h>
75 
76 #include <vm/vm.h>
77 #include <vm/vm_object.h>
78 #include <vm/vm_page.h>
79 
80 #include <machine/limits.h>
81 #include <machine/stdarg.h>
82 
83 #include <vfs/union/union.h>
84 
85 static void mount_warning(struct mount *mp, const char *ctl, ...)
86 		__printflike(2, 3);
87 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb);
88 static int checkvp_chdir (struct vnode *vn, struct thread *td);
89 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch);
90 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
91 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
92 static int getutimes (const struct timeval *, struct timespec *);
93 static int setfown (struct mount *, struct vnode *, uid_t, gid_t);
94 static int setfmode (struct vnode *, int);
95 static int setfflags (struct vnode *, int);
96 static int setutimes (struct vnode *, struct vattr *,
97 			const struct timespec *, int);
98 static int	usermount = 0;	/* if 1, non-root can mount fs. */
99 
100 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *);
101 
102 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0,
103     "Allow non-root users to mount filesystems");
104 
105 /*
106  * Virtual File System System Calls
107  */
108 
109 /*
110  * Mount a file system.
111  *
112  * mount_args(char *type, char *path, int flags, caddr_t data)
113  *
114  * MPALMOSTSAFE
115  */
116 int
117 sys_mount(struct mount_args *uap)
118 {
119 	struct thread *td = curthread;
120 	struct vnode *vp;
121 	struct nchandle nch;
122 	struct mount *mp, *nullmp;
123 	struct vfsconf *vfsp;
124 	int error, flag = 0, flag2 = 0;
125 	int hasmount;
126 	struct vattr va;
127 	struct nlookupdata nd;
128 	char fstypename[MFSNAMELEN];
129 	struct ucred *cred;
130 
131 	cred = td->td_ucred;
132 	if (jailed(cred)) {
133 		error = EPERM;
134 		goto done;
135 	}
136 	if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
137 		goto done;
138 
139 	/*
140 	 * Do not allow NFS export by non-root users.
141 	 */
142 	if (uap->flags & MNT_EXPORTED) {
143 		error = priv_check(td, PRIV_ROOT);
144 		if (error)
145 			goto done;
146 	}
147 	/*
148 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
149 	 */
150 	if (priv_check(td, PRIV_ROOT))
151 		uap->flags |= MNT_NOSUID | MNT_NODEV;
152 
153 	/*
154 	 * Lookup the requested path and extract the nch and vnode.
155 	 */
156 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
157 	if (error == 0) {
158 		if ((error = nlookup(&nd)) == 0) {
159 			if (nd.nl_nch.ncp->nc_vp == NULL)
160 				error = ENOENT;
161 		}
162 	}
163 	if (error) {
164 		nlookup_done(&nd);
165 		goto done;
166 	}
167 
168 	/*
169 	 * If the target filesystem is resolved via a nullfs mount, then
170 	 * nd.nl_nch.mount will be pointing to the nullfs mount structure
171 	 * instead of the target file system. We need it in case we are
172 	 * doing an update.
173 	 */
174 	nullmp = nd.nl_nch.mount;
175 
176 	/*
177 	 * Extract the locked+refd ncp and cleanup the nd structure
178 	 */
179 	nch = nd.nl_nch;
180 	cache_zero(&nd.nl_nch);
181 	nlookup_done(&nd);
182 
183 	if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
184 	    (mp = cache_findmount(&nch)) != NULL) {
185 		cache_dropmount(mp);
186 		hasmount = 1;
187 	} else {
188 		hasmount = 0;
189 	}
190 
191 
192 	/*
193 	 * now we have the locked ref'd nch and unreferenced vnode.
194 	 */
195 	vp = nch.ncp->nc_vp;
196 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
197 		cache_put(&nch);
198 		goto done;
199 	}
200 	cache_unlock(&nch);
201 
202 	/*
203 	 * Extract the file system type. We need to know this early, to take
204 	 * appropriate actions if we are dealing with a nullfs.
205 	 */
206         if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
207                 cache_drop(&nch);
208                 vput(vp);
209 		goto done;
210         }
211 
212 	/*
213 	 * Now we have an unlocked ref'd nch and a locked ref'd vp
214 	 */
215 	if (uap->flags & MNT_UPDATE) {
216 		if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
217 			cache_drop(&nch);
218 			vput(vp);
219 			error = EINVAL;
220 			goto done;
221 		}
222 
223 		if (strncmp(fstypename, "null", 5) == 0) {
224 			KKASSERT(nullmp);
225 			mp = nullmp;
226 		} else {
227 			mp = vp->v_mount;
228 		}
229 
230 		flag = mp->mnt_flag;
231 		flag2 = mp->mnt_kern_flag;
232 		/*
233 		 * We only allow the filesystem to be reloaded if it
234 		 * is currently mounted read-only.
235 		 */
236 		if ((uap->flags & MNT_RELOAD) &&
237 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
238 			cache_drop(&nch);
239 			vput(vp);
240 			error = EOPNOTSUPP;	/* Needs translation */
241 			goto done;
242 		}
243 		/*
244 		 * Only root, or the user that did the original mount is
245 		 * permitted to update it.
246 		 */
247 		if (mp->mnt_stat.f_owner != cred->cr_uid &&
248 		    (error = priv_check(td, PRIV_ROOT))) {
249 			cache_drop(&nch);
250 			vput(vp);
251 			goto done;
252 		}
253 		if (vfs_busy(mp, LK_NOWAIT)) {
254 			cache_drop(&nch);
255 			vput(vp);
256 			error = EBUSY;
257 			goto done;
258 		}
259 		if (hasmount) {
260 			cache_drop(&nch);
261 			vfs_unbusy(mp);
262 			vput(vp);
263 			error = EBUSY;
264 			goto done;
265 		}
266 		mp->mnt_flag |=
267 		    uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
268 		lwkt_gettoken(&mp->mnt_token);
269 		vn_unlock(vp);
270 		goto update;
271 	}
272 
273 	/*
274 	 * If the user is not root, ensure that they own the directory
275 	 * onto which we are attempting to mount.
276 	 */
277 	if ((error = VOP_GETATTR(vp, &va)) ||
278 	    (va.va_uid != cred->cr_uid &&
279 	     (error = priv_check(td, PRIV_ROOT)))) {
280 		cache_drop(&nch);
281 		vput(vp);
282 		goto done;
283 	}
284 	if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
285 		cache_drop(&nch);
286 		vput(vp);
287 		goto done;
288 	}
289 	if (vp->v_type != VDIR) {
290 		cache_drop(&nch);
291 		vput(vp);
292 		error = ENOTDIR;
293 		goto done;
294 	}
295 	if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) {
296 		cache_drop(&nch);
297 		vput(vp);
298 		error = EPERM;
299 		goto done;
300 	}
301 	vfsp = vfsconf_find_by_name(fstypename);
302 	if (vfsp == NULL) {
303 		linker_file_t lf;
304 
305 		/* Only load modules for root (very important!) */
306 		if ((error = priv_check(td, PRIV_ROOT)) != 0) {
307 			cache_drop(&nch);
308 			vput(vp);
309 			goto done;
310 		}
311 		error = linker_load_file(fstypename, &lf);
312 		if (error || lf == NULL) {
313 			cache_drop(&nch);
314 			vput(vp);
315 			if (lf == NULL)
316 				error = ENODEV;
317 			goto done;
318 		}
319 		lf->userrefs++;
320 		/* lookup again, see if the VFS was loaded */
321 		vfsp = vfsconf_find_by_name(fstypename);
322 		if (vfsp == NULL) {
323 			lf->userrefs--;
324 			linker_file_unload(lf);
325 			cache_drop(&nch);
326 			vput(vp);
327 			error = ENODEV;
328 			goto done;
329 		}
330 	}
331 	if (hasmount) {
332 		cache_drop(&nch);
333 		vput(vp);
334 		error = EBUSY;
335 		goto done;
336 	}
337 
338 	/*
339 	 * Allocate and initialize the filesystem.
340 	 */
341 	mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
342 	mount_init(mp);
343 	vfs_busy(mp, LK_NOWAIT);
344 	mp->mnt_op = vfsp->vfc_vfsops;
345 	mp->mnt_vfc = vfsp;
346 	vfsp->vfc_refcount++;
347 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
348 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
349 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
350 	mp->mnt_stat.f_owner = cred->cr_uid;
351 	lwkt_gettoken(&mp->mnt_token);
352 	vn_unlock(vp);
353 update:
354 	/*
355 	 * (per-mount token acquired at this point)
356 	 *
357 	 * Set the mount level flags.
358 	 */
359 	if (uap->flags & MNT_RDONLY)
360 		mp->mnt_flag |= MNT_RDONLY;
361 	else if (mp->mnt_flag & MNT_RDONLY)
362 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
363 	mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
364 	    MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
365 	    MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM |
366 	    MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
367 	mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
368 	    MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
369 	    MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_TRIM |
370 	    MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
371 	/*
372 	 * Mount the filesystem.
373 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
374 	 * get.
375 	 */
376 	error = VFS_MOUNT(mp, uap->path, uap->data, cred);
377 	if (mp->mnt_flag & MNT_UPDATE) {
378 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
379 			mp->mnt_flag &= ~MNT_RDONLY;
380 		mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
381 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
382 		if (error) {
383 			mp->mnt_flag = flag;
384 			mp->mnt_kern_flag = flag2;
385 		}
386 		lwkt_reltoken(&mp->mnt_token);
387 		vfs_unbusy(mp);
388 		vrele(vp);
389 		cache_drop(&nch);
390 		goto done;
391 	}
392 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
393 
394 	/*
395 	 * Put the new filesystem on the mount list after root.  The mount
396 	 * point gets its own mnt_ncmountpt (unless the VFS already set one
397 	 * up) which represents the root of the mount.  The lookup code
398 	 * detects the mount point going forward and checks the root of
399 	 * the mount going backwards.
400 	 *
401 	 * It is not necessary to invalidate or purge the vnode underneath
402 	 * because elements under the mount will be given their own glue
403 	 * namecache record.
404 	 */
405 	if (!error) {
406 		if (mp->mnt_ncmountpt.ncp == NULL) {
407 			/*
408 			 * allocate, then unlock, but leave the ref intact
409 			 */
410 			cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
411 			cache_unlock(&mp->mnt_ncmountpt);
412 		}
413 		mp->mnt_ncmounton = nch;		/* inherits ref */
414 		nch.ncp->nc_flag |= NCF_ISMOUNTPT;
415 		cache_ismounting(mp);
416 
417 		mountlist_insert(mp, MNTINS_LAST);
418 		vn_unlock(vp);
419 		checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
420 		error = vfs_allocate_syncvnode(mp);
421 		lwkt_reltoken(&mp->mnt_token);
422 		vfs_unbusy(mp);
423 		error = VFS_START(mp, 0);
424 		vrele(vp);
425 	} else {
426 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
427 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
428 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
429 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
430 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
431 		mp->mnt_vfc->vfc_refcount--;
432 		lwkt_reltoken(&mp->mnt_token);
433 		vfs_unbusy(mp);
434 		kfree(mp, M_MOUNT);
435 		cache_drop(&nch);
436 		vput(vp);
437 	}
438 done:
439 	return (error);
440 }
441 
442 /*
443  * Scan all active processes to see if any of them have a current
444  * or root directory onto which the new filesystem has just been
445  * mounted. If so, replace them with the new mount point.
446  *
447  * Both old_nch and new_nch are ref'd on call but not locked.
448  * new_nch must be temporarily locked so it can be associated with the
449  * vnode representing the root of the mount point.
450  */
451 struct checkdirs_info {
452 	struct nchandle old_nch;
453 	struct nchandle new_nch;
454 	struct vnode *old_vp;
455 	struct vnode *new_vp;
456 };
457 
458 static int checkdirs_callback(struct proc *p, void *data);
459 
460 static void
461 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
462 {
463 	struct checkdirs_info info;
464 	struct vnode *olddp;
465 	struct vnode *newdp;
466 	struct mount *mp;
467 
468 	/*
469 	 * If the old mount point's vnode has a usecount of 1, it is not
470 	 * being held as a descriptor anywhere.
471 	 */
472 	olddp = old_nch->ncp->nc_vp;
473 	if (olddp == NULL || olddp->v_sysref.refcnt == 1)
474 		return;
475 
476 	/*
477 	 * Force the root vnode of the new mount point to be resolved
478 	 * so we can update any matching processes.
479 	 */
480 	mp = new_nch->mount;
481 	if (VFS_ROOT(mp, &newdp))
482 		panic("mount: lost mount");
483 	vn_unlock(newdp);
484 	cache_lock(new_nch);
485 	vn_lock(newdp, LK_EXCLUSIVE | LK_RETRY);
486 	cache_setunresolved(new_nch);
487 	cache_setvp(new_nch, newdp);
488 	cache_unlock(new_nch);
489 
490 	/*
491 	 * Special handling of the root node
492 	 */
493 	if (rootvnode == olddp) {
494 		vref(newdp);
495 		vfs_cache_setroot(newdp, cache_hold(new_nch));
496 	}
497 
498 	/*
499 	 * Pass newdp separately so the callback does not have to access
500 	 * it via new_nch->ncp->nc_vp.
501 	 */
502 	info.old_nch = *old_nch;
503 	info.new_nch = *new_nch;
504 	info.new_vp = newdp;
505 	allproc_scan(checkdirs_callback, &info);
506 	vput(newdp);
507 }
508 
509 /*
510  * NOTE: callback is not MP safe because the scanned process's filedesc
511  * structure can be ripped out from under us, amoung other things.
512  */
513 static int
514 checkdirs_callback(struct proc *p, void *data)
515 {
516 	struct checkdirs_info *info = data;
517 	struct filedesc *fdp;
518 	struct nchandle ncdrop1;
519 	struct nchandle ncdrop2;
520 	struct vnode *vprele1;
521 	struct vnode *vprele2;
522 
523 	if ((fdp = p->p_fd) != NULL) {
524 		cache_zero(&ncdrop1);
525 		cache_zero(&ncdrop2);
526 		vprele1 = NULL;
527 		vprele2 = NULL;
528 
529 		/*
530 		 * MPUNSAFE - XXX fdp can be pulled out from under a
531 		 * foreign process.
532 		 *
533 		 * A shared filedesc is ok, we don't have to copy it
534 		 * because we are making this change globally.
535 		 */
536 		spin_lock(&fdp->fd_spin);
537 		if (fdp->fd_ncdir.mount == info->old_nch.mount &&
538 		    fdp->fd_ncdir.ncp == info->old_nch.ncp) {
539 			vprele1 = fdp->fd_cdir;
540 			vref(info->new_vp);
541 			fdp->fd_cdir = info->new_vp;
542 			ncdrop1 = fdp->fd_ncdir;
543 			cache_copy(&info->new_nch, &fdp->fd_ncdir);
544 		}
545 		if (fdp->fd_nrdir.mount == info->old_nch.mount &&
546 		    fdp->fd_nrdir.ncp == info->old_nch.ncp) {
547 			vprele2 = fdp->fd_rdir;
548 			vref(info->new_vp);
549 			fdp->fd_rdir = info->new_vp;
550 			ncdrop2 = fdp->fd_nrdir;
551 			cache_copy(&info->new_nch, &fdp->fd_nrdir);
552 		}
553 		spin_unlock(&fdp->fd_spin);
554 		if (ncdrop1.ncp)
555 			cache_drop(&ncdrop1);
556 		if (ncdrop2.ncp)
557 			cache_drop(&ncdrop2);
558 		if (vprele1)
559 			vrele(vprele1);
560 		if (vprele2)
561 			vrele(vprele2);
562 	}
563 	return(0);
564 }
565 
566 /*
567  * Unmount a file system.
568  *
569  * Note: unmount takes a path to the vnode mounted on as argument,
570  * not special file (as before).
571  *
572  * umount_args(char *path, int flags)
573  *
574  * MPALMOSTSAFE
575  */
576 int
577 sys_unmount(struct unmount_args *uap)
578 {
579 	struct thread *td = curthread;
580 	struct proc *p __debugvar = td->td_proc;
581 	struct mount *mp = NULL;
582 	struct nlookupdata nd;
583 	int error;
584 
585 	KKASSERT(p);
586 	get_mplock();
587 	if (td->td_ucred->cr_prison != NULL) {
588 		error = EPERM;
589 		goto done;
590 	}
591 	if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
592 		goto done;
593 
594 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
595 	if (error == 0)
596 		error = nlookup(&nd);
597 	if (error)
598 		goto out;
599 
600 	mp = nd.nl_nch.mount;
601 
602 	/*
603 	 * Only root, or the user that did the original mount is
604 	 * permitted to unmount this filesystem.
605 	 */
606 	if ((mp->mnt_stat.f_owner != td->td_ucred->cr_uid) &&
607 	    (error = priv_check(td, PRIV_ROOT)))
608 		goto out;
609 
610 	/*
611 	 * Don't allow unmounting the root file system.
612 	 */
613 	if (mp->mnt_flag & MNT_ROOTFS) {
614 		error = EINVAL;
615 		goto out;
616 	}
617 
618 	/*
619 	 * Must be the root of the filesystem
620 	 */
621 	if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
622 		error = EINVAL;
623 		goto out;
624 	}
625 
626 out:
627 	nlookup_done(&nd);
628 	if (error == 0)
629 		error = dounmount(mp, uap->flags);
630 done:
631 	rel_mplock();
632 	return (error);
633 }
634 
635 /*
636  * Do the actual file system unmount.
637  */
638 static int
639 dounmount_interlock(struct mount *mp)
640 {
641 	if (mp->mnt_kern_flag & MNTK_UNMOUNT)
642 		return (EBUSY);
643 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
644 	return(0);
645 }
646 
647 static int
648 unmount_allproc_cb(struct proc *p, void *arg)
649 {
650 	struct mount *mp;
651 
652 	if (p->p_textnch.ncp == NULL)
653 		return 0;
654 
655 	mp = (struct mount *)arg;
656 	if (p->p_textnch.mount == mp)
657 		cache_drop(&p->p_textnch);
658 
659 	return 0;
660 }
661 
662 int
663 dounmount(struct mount *mp, int flags)
664 {
665 	struct namecache *ncp;
666 	struct nchandle nch;
667 	struct vnode *vp;
668 	int error;
669 	int async_flag;
670 	int lflags;
671 	int freeok = 1;
672 	int retry;
673 
674 	lwkt_gettoken(&mp->mnt_token);
675 	/*
676 	 * Exclusive access for unmounting purposes
677 	 */
678 	if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
679 		goto out;
680 
681 	/*
682 	 * Allow filesystems to detect that a forced unmount is in progress.
683 	 */
684 	if (flags & MNT_FORCE)
685 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
686 	lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_TIMELOCK);
687 	error = lockmgr(&mp->mnt_lock, lflags);
688 	if (error) {
689 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
690 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
691 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
692 			wakeup(mp);
693 		}
694 		goto out;
695 	}
696 
697 	if (mp->mnt_flag & MNT_EXPUBLIC)
698 		vfs_setpublicfs(NULL, NULL, NULL);
699 
700 	vfs_msync(mp, MNT_WAIT);
701 	async_flag = mp->mnt_flag & MNT_ASYNC;
702 	mp->mnt_flag &=~ MNT_ASYNC;
703 
704 	/*
705 	 * If this filesystem isn't aliasing other filesystems,
706 	 * try to invalidate any remaining namecache entries and
707 	 * check the count afterwords.
708 	 */
709 	if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
710 		cache_lock(&mp->mnt_ncmountpt);
711 		cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
712 		cache_unlock(&mp->mnt_ncmountpt);
713 
714 		if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
715 		    (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
716 			allproc_scan(&unmount_allproc_cb, mp);
717 		}
718 
719 		if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
720 		    (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
721 
722 			if ((flags & MNT_FORCE) == 0) {
723 				error = EBUSY;
724 				mount_warning(mp, "Cannot unmount: "
725 						  "%d namecache "
726 						  "references still "
727 						  "present",
728 						  ncp->nc_refs - 1);
729 			} else {
730 				mount_warning(mp, "Forced unmount: "
731 						  "%d namecache "
732 						  "references still "
733 						  "present",
734 						  ncp->nc_refs - 1);
735 				freeok = 0;
736 			}
737 		}
738 	}
739 
740 	/*
741 	 * Decomission our special mnt_syncer vnode.  This also stops
742 	 * the vnlru code.  If we are unable to unmount we recommission
743 	 * the vnode.
744 	 *
745 	 * Then sync the filesystem.
746 	 */
747 	if ((vp = mp->mnt_syncer) != NULL) {
748 		mp->mnt_syncer = NULL;
749 		vrele(vp);
750 	}
751 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
752 		VFS_SYNC(mp, MNT_WAIT);
753 
754 	/*
755 	 * nchandle records ref the mount structure.  Expect a count of 1
756 	 * (our mount->mnt_ncmountpt).
757 	 *
758 	 * Scans can get temporary refs on a mountpoint (thought really
759 	 * heavy duty stuff like cache_findmount() do not).
760 	 */
761 	for (retry = 0; retry < 10 && mp->mnt_refs != 1; ++retry) {
762 		cache_unmounting(mp);
763 		tsleep(&mp->mnt_refs, 0, "mntbsy", hz / 10 + 1);
764 	}
765 	if (mp->mnt_refs != 1) {
766 		if ((flags & MNT_FORCE) == 0) {
767 			mount_warning(mp, "Cannot unmount: "
768 					  "%d mount refs still present",
769 					  mp->mnt_refs);
770 			error = EBUSY;
771 		} else {
772 			mount_warning(mp, "Forced unmount: "
773 					  "%d mount refs still present",
774 					  mp->mnt_refs);
775 			freeok = 0;
776 		}
777 	}
778 
779 	/*
780 	 * So far so good, sync the filesystem once more and
781 	 * call the VFS unmount code if the sync succeeds.
782 	 */
783 	if (error == 0) {
784 		if (((mp->mnt_flag & MNT_RDONLY) ||
785 		     (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
786 		    (flags & MNT_FORCE)) {
787 			error = VFS_UNMOUNT(mp, flags);
788 		}
789 	}
790 
791 	/*
792 	 * If an error occurred we can still recover, restoring the
793 	 * syncer vnode and misc flags.
794 	 */
795 	if (error) {
796 		if (mp->mnt_syncer == NULL)
797 			vfs_allocate_syncvnode(mp);
798 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
799 		mp->mnt_flag |= async_flag;
800 		lockmgr(&mp->mnt_lock, LK_RELEASE);
801 		if (mp->mnt_kern_flag & MNTK_MWAIT) {
802 			mp->mnt_kern_flag &= ~MNTK_MWAIT;
803 			wakeup(mp);
804 		}
805 		goto out;
806 	}
807 	/*
808 	 * Clean up any journals still associated with the mount after
809 	 * filesystem activity has ceased.
810 	 */
811 	journal_remove_all_journals(mp,
812 	    ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
813 
814 	mountlist_remove(mp);
815 
816 	/*
817 	 * Remove any installed vnode ops here so the individual VFSs don't
818 	 * have to.
819 	 */
820 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
821 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
822 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
823 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
824 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
825 
826 	if (mp->mnt_ncmountpt.ncp != NULL) {
827 		nch = mp->mnt_ncmountpt;
828 		cache_zero(&mp->mnt_ncmountpt);
829 		cache_clrmountpt(&nch);
830 		cache_drop(&nch);
831 	}
832 	if (mp->mnt_ncmounton.ncp != NULL) {
833 		cache_unmounting(mp);
834 		nch = mp->mnt_ncmounton;
835 		cache_zero(&mp->mnt_ncmounton);
836 		cache_clrmountpt(&nch);
837 		cache_drop(&nch);
838 	}
839 
840 	mp->mnt_vfc->vfc_refcount--;
841 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
842 		panic("unmount: dangling vnode");
843 	lockmgr(&mp->mnt_lock, LK_RELEASE);
844 	if (mp->mnt_kern_flag & MNTK_MWAIT) {
845 		mp->mnt_kern_flag &= ~MNTK_MWAIT;
846 		wakeup(mp);
847 	}
848 
849 	/*
850 	 * If we reach here and freeok != 0 we must free the mount.
851 	 * If refs > 1 cycle and wait, just in case someone tried
852 	 * to busy the mount after we decided to do the unmount.
853 	 */
854 	if (freeok) {
855 		while (mp->mnt_refs > 1) {
856 			cache_unmounting(mp);
857 			wakeup(mp);
858 			tsleep(&mp->mnt_refs, 0, "umntrwait", hz / 10 + 1);
859 		}
860 		lwkt_reltoken(&mp->mnt_token);
861 		kfree(mp, M_MOUNT);
862 		mp = NULL;
863 	}
864 	error = 0;
865 out:
866 	if (mp)
867 		lwkt_reltoken(&mp->mnt_token);
868 	return (error);
869 }
870 
871 static
872 void
873 mount_warning(struct mount *mp, const char *ctl, ...)
874 {
875 	char *ptr;
876 	char *buf;
877 	__va_list va;
878 
879 	__va_start(va, ctl);
880 	if (cache_fullpath(NULL, &mp->mnt_ncmounton, NULL,
881 			   &ptr, &buf, 0) == 0) {
882 		kprintf("unmount(%s): ", ptr);
883 		kvprintf(ctl, va);
884 		kprintf("\n");
885 		kfree(buf, M_TEMP);
886 	} else {
887 		kprintf("unmount(%p", mp);
888 		if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
889 			kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
890 		kprintf("): ");
891 		kvprintf(ctl, va);
892 		kprintf("\n");
893 	}
894 	__va_end(va);
895 }
896 
897 /*
898  * Shim cache_fullpath() to handle the case where a process is chrooted into
899  * a subdirectory of a mount.  In this case if the root mount matches the
900  * process root directory's mount we have to specify the process's root
901  * directory instead of the mount point, because the mount point might
902  * be above the root directory.
903  */
904 static
905 int
906 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
907 {
908 	struct nchandle *nch;
909 
910 	if (p && p->p_fd->fd_nrdir.mount == mp)
911 		nch = &p->p_fd->fd_nrdir;
912 	else
913 		nch = &mp->mnt_ncmountpt;
914 	return(cache_fullpath(p, nch, NULL, rb, fb, 0));
915 }
916 
917 /*
918  * Sync each mounted filesystem.
919  */
920 
921 #ifdef DEBUG
922 static int syncprt = 0;
923 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
924 #endif /* DEBUG */
925 
926 static int sync_callback(struct mount *mp, void *data);
927 
928 int
929 sys_sync(struct sync_args *uap)
930 {
931 	mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
932 #ifdef DEBUG
933 	/*
934 	 * print out buffer pool stat information on each sync() call.
935 	 */
936 	if (syncprt)
937 		vfs_bufstats();
938 #endif /* DEBUG */
939 	return (0);
940 }
941 
942 static
943 int
944 sync_callback(struct mount *mp, void *data __unused)
945 {
946 	int asyncflag;
947 
948 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
949 		asyncflag = mp->mnt_flag & MNT_ASYNC;
950 		mp->mnt_flag &= ~MNT_ASYNC;
951 		vfs_msync(mp, MNT_NOWAIT);
952 		VFS_SYNC(mp, MNT_NOWAIT);
953 		mp->mnt_flag |= asyncflag;
954 	}
955 	return(0);
956 }
957 
958 /* XXX PRISON: could be per prison flag */
959 static int prison_quotas;
960 #if 0
961 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
962 #endif
963 
964 /*
965  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
966  *
967  * Change filesystem quotas.
968  *
969  * MPALMOSTSAFE
970  */
971 int
972 sys_quotactl(struct quotactl_args *uap)
973 {
974 	struct nlookupdata nd;
975 	struct thread *td;
976 	struct mount *mp;
977 	int error;
978 
979 	get_mplock();
980 	td = curthread;
981 	if (td->td_ucred->cr_prison && !prison_quotas) {
982 		error = EPERM;
983 		goto done;
984 	}
985 
986 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
987 	if (error == 0)
988 		error = nlookup(&nd);
989 	if (error == 0) {
990 		mp = nd.nl_nch.mount;
991 		error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
992 				    uap->arg, nd.nl_cred);
993 	}
994 	nlookup_done(&nd);
995 done:
996 	rel_mplock();
997 	return (error);
998 }
999 
1000 /*
1001  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
1002  *		void *buf, int buflen)
1003  *
1004  * This function operates on a mount point and executes the specified
1005  * operation using the specified control data, and possibly returns data.
1006  *
1007  * The actual number of bytes stored in the result buffer is returned, 0
1008  * if none, otherwise an error is returned.
1009  *
1010  * MPALMOSTSAFE
1011  */
1012 int
1013 sys_mountctl(struct mountctl_args *uap)
1014 {
1015 	struct thread *td = curthread;
1016 	struct proc *p = td->td_proc;
1017 	struct file *fp;
1018 	void *ctl = NULL;
1019 	void *buf = NULL;
1020 	char *path = NULL;
1021 	int error;
1022 
1023 	/*
1024 	 * Sanity and permissions checks.  We must be root.
1025 	 */
1026 	KKASSERT(p);
1027 	if (td->td_ucred->cr_prison != NULL)
1028 		return (EPERM);
1029 	if ((uap->op != MOUNTCTL_MOUNTFLAGS) &&
1030 	    (error = priv_check(td, PRIV_ROOT)) != 0)
1031 		return (error);
1032 
1033 	/*
1034 	 * Argument length checks
1035 	 */
1036 	if (uap->ctllen < 0 || uap->ctllen > 1024)
1037 		return (EINVAL);
1038 	if (uap->buflen < 0 || uap->buflen > 16 * 1024)
1039 		return (EINVAL);
1040 	if (uap->path == NULL)
1041 		return (EINVAL);
1042 
1043 	/*
1044 	 * Allocate the necessary buffers and copyin data
1045 	 */
1046 	path = objcache_get(namei_oc, M_WAITOK);
1047 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
1048 	if (error)
1049 		goto done;
1050 
1051 	if (uap->ctllen) {
1052 		ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
1053 		error = copyin(uap->ctl, ctl, uap->ctllen);
1054 		if (error)
1055 			goto done;
1056 	}
1057 	if (uap->buflen)
1058 		buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
1059 
1060 	/*
1061 	 * Validate the descriptor
1062 	 */
1063 	if (uap->fd >= 0) {
1064 		fp = holdfp(p->p_fd, uap->fd, -1);
1065 		if (fp == NULL) {
1066 			error = EBADF;
1067 			goto done;
1068 		}
1069 	} else {
1070 		fp = NULL;
1071 	}
1072 
1073 	/*
1074 	 * Execute the internal kernel function and clean up.
1075 	 */
1076 	get_mplock();
1077 	error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
1078 	rel_mplock();
1079 	if (fp)
1080 		fdrop(fp);
1081 	if (error == 0 && uap->sysmsg_result > 0)
1082 		error = copyout(buf, uap->buf, uap->sysmsg_result);
1083 done:
1084 	if (path)
1085 		objcache_put(namei_oc, path);
1086 	if (ctl)
1087 		kfree(ctl, M_TEMP);
1088 	if (buf)
1089 		kfree(buf, M_TEMP);
1090 	return (error);
1091 }
1092 
1093 /*
1094  * Execute a mount control operation by resolving the path to a mount point
1095  * and calling vop_mountctl().
1096  *
1097  * Use the mount point from the nch instead of the vnode so nullfs mounts
1098  * can properly spike the VOP.
1099  */
1100 int
1101 kern_mountctl(const char *path, int op, struct file *fp,
1102 		const void *ctl, int ctllen,
1103 		void *buf, int buflen, int *res)
1104 {
1105 	struct vnode *vp;
1106 	struct mount *mp;
1107 	struct nlookupdata nd;
1108 	int error;
1109 
1110 	*res = 0;
1111 	vp = NULL;
1112 	error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
1113 	if (error == 0)
1114 		error = nlookup(&nd);
1115 	if (error == 0)
1116 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
1117 	mp = nd.nl_nch.mount;
1118 	nlookup_done(&nd);
1119 	if (error)
1120 		return (error);
1121 	vn_unlock(vp);
1122 
1123 	/*
1124 	 * Must be the root of the filesystem
1125 	 */
1126 	if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
1127 		vrele(vp);
1128 		return (EINVAL);
1129 	}
1130 	error = vop_mountctl(mp->mnt_vn_use_ops, vp, op, fp, ctl, ctllen,
1131 			     buf, buflen, res);
1132 	vrele(vp);
1133 	return (error);
1134 }
1135 
1136 int
1137 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1138 {
1139 	struct thread *td = curthread;
1140 	struct proc *p = td->td_proc;
1141 	struct mount *mp;
1142 	struct statfs *sp;
1143 	char *fullpath, *freepath;
1144 	int error;
1145 
1146 	if ((error = nlookup(nd)) != 0)
1147 		return (error);
1148 	mp = nd->nl_nch.mount;
1149 	sp = &mp->mnt_stat;
1150 	if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1151 		return (error);
1152 
1153 	error = mount_path(p, mp, &fullpath, &freepath);
1154 	if (error)
1155 		return(error);
1156 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1157 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1158 	kfree(freepath, M_TEMP);
1159 
1160 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1161 	bcopy(sp, buf, sizeof(*buf));
1162 	/* Only root should have access to the fsid's. */
1163 	if (priv_check(td, PRIV_ROOT))
1164 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1165 	return (0);
1166 }
1167 
1168 /*
1169  * statfs_args(char *path, struct statfs *buf)
1170  *
1171  * Get filesystem statistics.
1172  */
1173 int
1174 sys_statfs(struct statfs_args *uap)
1175 {
1176 	struct nlookupdata nd;
1177 	struct statfs buf;
1178 	int error;
1179 
1180 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1181 	if (error == 0)
1182 		error = kern_statfs(&nd, &buf);
1183 	nlookup_done(&nd);
1184 	if (error == 0)
1185 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1186 	return (error);
1187 }
1188 
1189 int
1190 kern_fstatfs(int fd, struct statfs *buf)
1191 {
1192 	struct thread *td = curthread;
1193 	struct proc *p = td->td_proc;
1194 	struct file *fp;
1195 	struct mount *mp;
1196 	struct statfs *sp;
1197 	char *fullpath, *freepath;
1198 	int error;
1199 
1200 	KKASSERT(p);
1201 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1202 		return (error);
1203 
1204 	/*
1205 	 * Try to use mount info from any overlays rather than the
1206 	 * mount info for the underlying vnode, otherwise we will
1207 	 * fail when operating on null-mounted paths inside a chroot.
1208 	 */
1209 	if ((mp = fp->f_nchandle.mount) == NULL)
1210 		mp = ((struct vnode *)fp->f_data)->v_mount;
1211 	if (mp == NULL) {
1212 		error = EBADF;
1213 		goto done;
1214 	}
1215 	if (fp->f_cred == NULL) {
1216 		error = EINVAL;
1217 		goto done;
1218 	}
1219 	sp = &mp->mnt_stat;
1220 	if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1221 		goto done;
1222 
1223 	if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1224 		goto done;
1225 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1226 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1227 	kfree(freepath, M_TEMP);
1228 
1229 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1230 	bcopy(sp, buf, sizeof(*buf));
1231 
1232 	/* Only root should have access to the fsid's. */
1233 	if (priv_check(td, PRIV_ROOT))
1234 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1235 	error = 0;
1236 done:
1237 	fdrop(fp);
1238 	return (error);
1239 }
1240 
1241 /*
1242  * fstatfs_args(int fd, struct statfs *buf)
1243  *
1244  * Get filesystem statistics.
1245  */
1246 int
1247 sys_fstatfs(struct fstatfs_args *uap)
1248 {
1249 	struct statfs buf;
1250 	int error;
1251 
1252 	error = kern_fstatfs(uap->fd, &buf);
1253 
1254 	if (error == 0)
1255 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1256 	return (error);
1257 }
1258 
1259 int
1260 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1261 {
1262 	struct mount *mp;
1263 	struct statvfs *sp;
1264 	int error;
1265 
1266 	if ((error = nlookup(nd)) != 0)
1267 		return (error);
1268 	mp = nd->nl_nch.mount;
1269 	sp = &mp->mnt_vstat;
1270 	if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1271 		return (error);
1272 
1273 	sp->f_flag = 0;
1274 	if (mp->mnt_flag & MNT_RDONLY)
1275 		sp->f_flag |= ST_RDONLY;
1276 	if (mp->mnt_flag & MNT_NOSUID)
1277 		sp->f_flag |= ST_NOSUID;
1278 	bcopy(sp, buf, sizeof(*buf));
1279 	return (0);
1280 }
1281 
1282 /*
1283  * statfs_args(char *path, struct statfs *buf)
1284  *
1285  * Get filesystem statistics.
1286  */
1287 int
1288 sys_statvfs(struct statvfs_args *uap)
1289 {
1290 	struct nlookupdata nd;
1291 	struct statvfs buf;
1292 	int error;
1293 
1294 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1295 	if (error == 0)
1296 		error = kern_statvfs(&nd, &buf);
1297 	nlookup_done(&nd);
1298 	if (error == 0)
1299 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1300 	return (error);
1301 }
1302 
1303 int
1304 kern_fstatvfs(int fd, struct statvfs *buf)
1305 {
1306 	struct thread *td = curthread;
1307 	struct proc *p = td->td_proc;
1308 	struct file *fp;
1309 	struct mount *mp;
1310 	struct statvfs *sp;
1311 	int error;
1312 
1313 	KKASSERT(p);
1314 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1315 		return (error);
1316 	if ((mp = fp->f_nchandle.mount) == NULL)
1317 		mp = ((struct vnode *)fp->f_data)->v_mount;
1318 	if (mp == NULL) {
1319 		error = EBADF;
1320 		goto done;
1321 	}
1322 	if (fp->f_cred == NULL) {
1323 		error = EINVAL;
1324 		goto done;
1325 	}
1326 	sp = &mp->mnt_vstat;
1327 	if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1328 		goto done;
1329 
1330 	sp->f_flag = 0;
1331 	if (mp->mnt_flag & MNT_RDONLY)
1332 		sp->f_flag |= ST_RDONLY;
1333 	if (mp->mnt_flag & MNT_NOSUID)
1334 		sp->f_flag |= ST_NOSUID;
1335 
1336 	bcopy(sp, buf, sizeof(*buf));
1337 	error = 0;
1338 done:
1339 	fdrop(fp);
1340 	return (error);
1341 }
1342 
1343 /*
1344  * fstatfs_args(int fd, struct statfs *buf)
1345  *
1346  * Get filesystem statistics.
1347  */
1348 int
1349 sys_fstatvfs(struct fstatvfs_args *uap)
1350 {
1351 	struct statvfs buf;
1352 	int error;
1353 
1354 	error = kern_fstatvfs(uap->fd, &buf);
1355 
1356 	if (error == 0)
1357 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1358 	return (error);
1359 }
1360 
1361 /*
1362  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1363  *
1364  * Get statistics on all filesystems.
1365  */
1366 
1367 struct getfsstat_info {
1368 	struct statfs *sfsp;
1369 	long count;
1370 	long maxcount;
1371 	int error;
1372 	int flags;
1373 	struct thread *td;
1374 };
1375 
1376 static int getfsstat_callback(struct mount *, void *);
1377 
1378 int
1379 sys_getfsstat(struct getfsstat_args *uap)
1380 {
1381 	struct thread *td = curthread;
1382 	struct getfsstat_info info;
1383 
1384 	bzero(&info, sizeof(info));
1385 
1386 	info.maxcount = uap->bufsize / sizeof(struct statfs);
1387 	info.sfsp = uap->buf;
1388 	info.count = 0;
1389 	info.flags = uap->flags;
1390 	info.td = td;
1391 
1392 	mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1393 	if (info.sfsp && info.count > info.maxcount)
1394 		uap->sysmsg_result = info.maxcount;
1395 	else
1396 		uap->sysmsg_result = info.count;
1397 	return (info.error);
1398 }
1399 
1400 static int
1401 getfsstat_callback(struct mount *mp, void *data)
1402 {
1403 	struct getfsstat_info *info = data;
1404 	struct statfs *sp;
1405 	char *freepath;
1406 	char *fullpath;
1407 	int error;
1408 
1409 	if (info->sfsp && info->count < info->maxcount) {
1410 		if (info->td->td_proc &&
1411 		    !chroot_visible_mnt(mp, info->td->td_proc)) {
1412 			return(0);
1413 		}
1414 		sp = &mp->mnt_stat;
1415 
1416 		/*
1417 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1418 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1419 		 * overrides MNT_WAIT.
1420 		 */
1421 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1422 		    (info->flags & MNT_WAIT)) &&
1423 		    (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1424 			return(0);
1425 		}
1426 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1427 
1428 		error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1429 		if (error) {
1430 			info->error = error;
1431 			return(-1);
1432 		}
1433 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1434 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1435 		kfree(freepath, M_TEMP);
1436 
1437 		error = copyout(sp, info->sfsp, sizeof(*sp));
1438 		if (error) {
1439 			info->error = error;
1440 			return (-1);
1441 		}
1442 		++info->sfsp;
1443 	}
1444 	info->count++;
1445 	return(0);
1446 }
1447 
1448 /*
1449  * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1450 		   long bufsize, int flags)
1451  *
1452  * Get statistics on all filesystems.
1453  */
1454 
1455 struct getvfsstat_info {
1456 	struct statfs *sfsp;
1457 	struct statvfs *vsfsp;
1458 	long count;
1459 	long maxcount;
1460 	int error;
1461 	int flags;
1462 	struct thread *td;
1463 };
1464 
1465 static int getvfsstat_callback(struct mount *, void *);
1466 
1467 int
1468 sys_getvfsstat(struct getvfsstat_args *uap)
1469 {
1470 	struct thread *td = curthread;
1471 	struct getvfsstat_info info;
1472 
1473 	bzero(&info, sizeof(info));
1474 
1475 	info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1476 	info.sfsp = uap->buf;
1477 	info.vsfsp = uap->vbuf;
1478 	info.count = 0;
1479 	info.flags = uap->flags;
1480 	info.td = td;
1481 
1482 	mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1483 	if (info.vsfsp && info.count > info.maxcount)
1484 		uap->sysmsg_result = info.maxcount;
1485 	else
1486 		uap->sysmsg_result = info.count;
1487 	return (info.error);
1488 }
1489 
1490 static int
1491 getvfsstat_callback(struct mount *mp, void *data)
1492 {
1493 	struct getvfsstat_info *info = data;
1494 	struct statfs *sp;
1495 	struct statvfs *vsp;
1496 	char *freepath;
1497 	char *fullpath;
1498 	int error;
1499 
1500 	if (info->vsfsp && info->count < info->maxcount) {
1501 		if (info->td->td_proc &&
1502 		    !chroot_visible_mnt(mp, info->td->td_proc)) {
1503 			return(0);
1504 		}
1505 		sp = &mp->mnt_stat;
1506 		vsp = &mp->mnt_vstat;
1507 
1508 		/*
1509 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1510 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1511 		 * overrides MNT_WAIT.
1512 		 */
1513 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1514 		    (info->flags & MNT_WAIT)) &&
1515 		    (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1516 			return(0);
1517 		}
1518 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1519 
1520 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1521 		    (info->flags & MNT_WAIT)) &&
1522 		    (error = VFS_STATVFS(mp, vsp, info->td->td_ucred))) {
1523 			return(0);
1524 		}
1525 		vsp->f_flag = 0;
1526 		if (mp->mnt_flag & MNT_RDONLY)
1527 			vsp->f_flag |= ST_RDONLY;
1528 		if (mp->mnt_flag & MNT_NOSUID)
1529 			vsp->f_flag |= ST_NOSUID;
1530 
1531 		error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1532 		if (error) {
1533 			info->error = error;
1534 			return(-1);
1535 		}
1536 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1537 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1538 		kfree(freepath, M_TEMP);
1539 
1540 		error = copyout(sp, info->sfsp, sizeof(*sp));
1541 		if (error == 0)
1542 			error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1543 		if (error) {
1544 			info->error = error;
1545 			return (-1);
1546 		}
1547 		++info->sfsp;
1548 		++info->vsfsp;
1549 	}
1550 	info->count++;
1551 	return(0);
1552 }
1553 
1554 
1555 /*
1556  * fchdir_args(int fd)
1557  *
1558  * Change current working directory to a given file descriptor.
1559  */
1560 int
1561 sys_fchdir(struct fchdir_args *uap)
1562 {
1563 	struct thread *td = curthread;
1564 	struct proc *p = td->td_proc;
1565 	struct filedesc *fdp = p->p_fd;
1566 	struct vnode *vp, *ovp;
1567 	struct mount *mp;
1568 	struct file *fp;
1569 	struct nchandle nch, onch, tnch;
1570 	int error;
1571 
1572 	if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1573 		return (error);
1574 	lwkt_gettoken(&p->p_token);
1575 	vp = (struct vnode *)fp->f_data;
1576 	vref(vp);
1577 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1578 	if (fp->f_nchandle.ncp == NULL)
1579 		error = ENOTDIR;
1580 	else
1581 		error = checkvp_chdir(vp, td);
1582 	if (error) {
1583 		vput(vp);
1584 		goto done;
1585 	}
1586 	cache_copy(&fp->f_nchandle, &nch);
1587 
1588 	/*
1589 	 * If the ncp has become a mount point, traverse through
1590 	 * the mount point.
1591 	 */
1592 
1593 	while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1594 	       (mp = cache_findmount(&nch)) != NULL
1595 	) {
1596 		error = nlookup_mp(mp, &tnch);
1597 		if (error == 0) {
1598 			cache_unlock(&tnch);	/* leave ref intact */
1599 			vput(vp);
1600 			vp = tnch.ncp->nc_vp;
1601 			error = vget(vp, LK_SHARED);
1602 			KKASSERT(error == 0);
1603 			cache_drop(&nch);
1604 			nch = tnch;
1605 		}
1606 		cache_dropmount(mp);
1607 	}
1608 	if (error == 0) {
1609 		ovp = fdp->fd_cdir;
1610 		onch = fdp->fd_ncdir;
1611 		vn_unlock(vp);		/* leave ref intact */
1612 		fdp->fd_cdir = vp;
1613 		fdp->fd_ncdir = nch;
1614 		cache_drop(&onch);
1615 		vrele(ovp);
1616 	} else {
1617 		cache_drop(&nch);
1618 		vput(vp);
1619 	}
1620 	fdrop(fp);
1621 done:
1622 	lwkt_reltoken(&p->p_token);
1623 	return (error);
1624 }
1625 
1626 int
1627 kern_chdir(struct nlookupdata *nd)
1628 {
1629 	struct thread *td = curthread;
1630 	struct proc *p = td->td_proc;
1631 	struct filedesc *fdp = p->p_fd;
1632 	struct vnode *vp, *ovp;
1633 	struct nchandle onch;
1634 	int error;
1635 
1636 	if ((error = nlookup(nd)) != 0)
1637 		return (error);
1638 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1639 		return (ENOENT);
1640 	if ((error = vget(vp, LK_SHARED)) != 0)
1641 		return (error);
1642 
1643 	lwkt_gettoken(&p->p_token);
1644 	error = checkvp_chdir(vp, td);
1645 	vn_unlock(vp);
1646 	if (error == 0) {
1647 		ovp = fdp->fd_cdir;
1648 		onch = fdp->fd_ncdir;
1649 		cache_unlock(&nd->nl_nch);	/* leave reference intact */
1650 		fdp->fd_ncdir = nd->nl_nch;
1651 		fdp->fd_cdir = vp;
1652 		cache_drop(&onch);
1653 		vrele(ovp);
1654 		cache_zero(&nd->nl_nch);
1655 	} else {
1656 		vrele(vp);
1657 	}
1658 	lwkt_reltoken(&p->p_token);
1659 	return (error);
1660 }
1661 
1662 /*
1663  * chdir_args(char *path)
1664  *
1665  * Change current working directory (``.'').
1666  */
1667 int
1668 sys_chdir(struct chdir_args *uap)
1669 {
1670 	struct nlookupdata nd;
1671 	int error;
1672 
1673 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1674 	if (error == 0)
1675 		error = kern_chdir(&nd);
1676 	nlookup_done(&nd);
1677 	return (error);
1678 }
1679 
1680 /*
1681  * Helper function for raised chroot(2) security function:  Refuse if
1682  * any filedescriptors are open directories.
1683  */
1684 static int
1685 chroot_refuse_vdir_fds(struct filedesc *fdp)
1686 {
1687 	struct vnode *vp;
1688 	struct file *fp;
1689 	int error;
1690 	int fd;
1691 
1692 	for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1693 		if ((error = holdvnode(fdp, fd, &fp)) != 0)
1694 			continue;
1695 		vp = (struct vnode *)fp->f_data;
1696 		if (vp->v_type != VDIR) {
1697 			fdrop(fp);
1698 			continue;
1699 		}
1700 		fdrop(fp);
1701 		return(EPERM);
1702 	}
1703 	return (0);
1704 }
1705 
1706 /*
1707  * This sysctl determines if we will allow a process to chroot(2) if it
1708  * has a directory open:
1709  *	0: disallowed for all processes.
1710  *	1: allowed for processes that were not already chroot(2)'ed.
1711  *	2: allowed for all processes.
1712  */
1713 
1714 static int chroot_allow_open_directories = 1;
1715 
1716 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1717      &chroot_allow_open_directories, 0, "");
1718 
1719 /*
1720  * chroot to the specified namecache entry.  We obtain the vp from the
1721  * namecache data.  The passed ncp must be locked and referenced and will
1722  * remain locked and referenced on return.
1723  */
1724 int
1725 kern_chroot(struct nchandle *nch)
1726 {
1727 	struct thread *td = curthread;
1728 	struct proc *p = td->td_proc;
1729 	struct filedesc *fdp = p->p_fd;
1730 	struct vnode *vp;
1731 	int error;
1732 
1733 	/*
1734 	 * Only privileged user can chroot
1735 	 */
1736 	error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1737 	if (error)
1738 		return (error);
1739 
1740 	/*
1741 	 * Disallow open directory descriptors (fchdir() breakouts).
1742 	 */
1743 	if (chroot_allow_open_directories == 0 ||
1744 	   (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1745 		if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1746 			return (error);
1747 	}
1748 	if ((vp = nch->ncp->nc_vp) == NULL)
1749 		return (ENOENT);
1750 
1751 	if ((error = vget(vp, LK_SHARED)) != 0)
1752 		return (error);
1753 
1754 	/*
1755 	 * Check the validity of vp as a directory to change to and
1756 	 * associate it with rdir/jdir.
1757 	 */
1758 	error = checkvp_chdir(vp, td);
1759 	vn_unlock(vp);			/* leave reference intact */
1760 	if (error == 0) {
1761 		vrele(fdp->fd_rdir);
1762 		fdp->fd_rdir = vp;	/* reference inherited by fd_rdir */
1763 		cache_drop(&fdp->fd_nrdir);
1764 		cache_copy(nch, &fdp->fd_nrdir);
1765 		if (fdp->fd_jdir == NULL) {
1766 			fdp->fd_jdir = vp;
1767 			vref(fdp->fd_jdir);
1768 			cache_copy(nch, &fdp->fd_njdir);
1769 		}
1770 	} else {
1771 		vrele(vp);
1772 	}
1773 	return (error);
1774 }
1775 
1776 /*
1777  * chroot_args(char *path)
1778  *
1779  * Change notion of root (``/'') directory.
1780  */
1781 int
1782 sys_chroot(struct chroot_args *uap)
1783 {
1784 	struct thread *td __debugvar = curthread;
1785 	struct nlookupdata nd;
1786 	int error;
1787 
1788 	KKASSERT(td->td_proc);
1789 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1790 	if (error == 0) {
1791 		nd.nl_flags |= NLC_EXEC;
1792 		error = nlookup(&nd);
1793 		if (error == 0)
1794 			error = kern_chroot(&nd.nl_nch);
1795 	}
1796 	nlookup_done(&nd);
1797 	return(error);
1798 }
1799 
1800 int
1801 sys_chroot_kernel(struct chroot_kernel_args *uap)
1802 {
1803 	struct thread *td = curthread;
1804 	struct nlookupdata nd;
1805 	struct nchandle *nch;
1806 	struct vnode *vp;
1807 	int error;
1808 
1809 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1810 	if (error)
1811 		goto error_nond;
1812 
1813 	error = nlookup(&nd);
1814 	if (error)
1815 		goto error_out;
1816 
1817 	nch = &nd.nl_nch;
1818 
1819 	error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1820 	if (error)
1821 		goto error_out;
1822 
1823 	if ((vp = nch->ncp->nc_vp) == NULL) {
1824 		error = ENOENT;
1825 		goto error_out;
1826 	}
1827 
1828 	if ((error = cache_vref(nch, nd.nl_cred, &vp)) != 0)
1829 		goto error_out;
1830 
1831 	kprintf("chroot_kernel: set new rootnch/rootvnode to %s\n", uap->path);
1832 	get_mplock();
1833 	vfs_cache_setroot(vp, cache_hold(nch));
1834 	rel_mplock();
1835 
1836 error_out:
1837 	nlookup_done(&nd);
1838 error_nond:
1839 	return(error);
1840 }
1841 
1842 /*
1843  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1844  * determine whether it is legal to chdir to the vnode.  The vnode's state
1845  * is not changed by this call.
1846  */
1847 int
1848 checkvp_chdir(struct vnode *vp, struct thread *td)
1849 {
1850 	int error;
1851 
1852 	if (vp->v_type != VDIR)
1853 		error = ENOTDIR;
1854 	else
1855 		error = VOP_EACCESS(vp, VEXEC, td->td_ucred);
1856 	return (error);
1857 }
1858 
1859 int
1860 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1861 {
1862 	struct thread *td = curthread;
1863 	struct proc *p = td->td_proc;
1864 	struct lwp *lp = td->td_lwp;
1865 	struct filedesc *fdp = p->p_fd;
1866 	int cmode, flags;
1867 	struct file *nfp;
1868 	struct file *fp;
1869 	struct vnode *vp;
1870 	int type, indx, error = 0;
1871 	struct flock lf;
1872 
1873 	if ((oflags & O_ACCMODE) == O_ACCMODE)
1874 		return (EINVAL);
1875 	flags = FFLAGS(oflags);
1876 	error = falloc(lp, &nfp, NULL);
1877 	if (error)
1878 		return (error);
1879 	fp = nfp;
1880 	cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
1881 
1882 	/*
1883 	 * XXX p_dupfd is a real mess.  It allows a device to return a
1884 	 * file descriptor to be duplicated rather then doing the open
1885 	 * itself.
1886 	 */
1887 	lp->lwp_dupfd = -1;
1888 
1889 	/*
1890 	 * Call vn_open() to do the lookup and assign the vnode to the
1891 	 * file pointer.  vn_open() does not change the ref count on fp
1892 	 * and the vnode, on success, will be inherited by the file pointer
1893 	 * and unlocked.
1894 	 */
1895 	nd->nl_flags |= NLC_LOCKVP;
1896 	error = vn_open(nd, fp, flags, cmode);
1897 	nlookup_done(nd);
1898 	if (error) {
1899 		/*
1900 		 * handle special fdopen() case.  bleh.  dupfdopen() is
1901 		 * responsible for dropping the old contents of ofiles[indx]
1902 		 * if it succeeds.
1903 		 *
1904 		 * Note that fsetfd() will add a ref to fp which represents
1905 		 * the fd_files[] assignment.  We must still drop our
1906 		 * reference.
1907 		 */
1908 		if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1909 			if (fdalloc(p, 0, &indx) == 0) {
1910 				error = dupfdopen(fdp, indx, lp->lwp_dupfd, flags, error);
1911 				if (error == 0) {
1912 					*res = indx;
1913 					fdrop(fp);	/* our ref */
1914 					return (0);
1915 				}
1916 				fsetfd(fdp, NULL, indx);
1917 			}
1918 		}
1919 		fdrop(fp);	/* our ref */
1920 		if (error == ERESTART)
1921 			error = EINTR;
1922 		return (error);
1923 	}
1924 
1925 	/*
1926 	 * ref the vnode for ourselves so it can't be ripped out from under
1927 	 * is.  XXX need an ND flag to request that the vnode be returned
1928 	 * anyway.
1929 	 *
1930 	 * Reserve a file descriptor but do not assign it until the open
1931 	 * succeeds.
1932 	 */
1933 	vp = (struct vnode *)fp->f_data;
1934 	vref(vp);
1935 	if ((error = fdalloc(p, 0, &indx)) != 0) {
1936 		fdrop(fp);
1937 		vrele(vp);
1938 		return (error);
1939 	}
1940 
1941 	/*
1942 	 * If no error occurs the vp will have been assigned to the file
1943 	 * pointer.
1944 	 */
1945 	lp->lwp_dupfd = 0;
1946 
1947 	if (flags & (O_EXLOCK | O_SHLOCK)) {
1948 		lf.l_whence = SEEK_SET;
1949 		lf.l_start = 0;
1950 		lf.l_len = 0;
1951 		if (flags & O_EXLOCK)
1952 			lf.l_type = F_WRLCK;
1953 		else
1954 			lf.l_type = F_RDLCK;
1955 		if (flags & FNONBLOCK)
1956 			type = 0;
1957 		else
1958 			type = F_WAIT;
1959 
1960 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1961 			/*
1962 			 * lock request failed.  Clean up the reserved
1963 			 * descriptor.
1964 			 */
1965 			vrele(vp);
1966 			fsetfd(fdp, NULL, indx);
1967 			fdrop(fp);
1968 			return (error);
1969 		}
1970 		fp->f_flag |= FHASLOCK;
1971 	}
1972 #if 0
1973 	/*
1974 	 * Assert that all regular file vnodes were created with a object.
1975 	 */
1976 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1977 		("open: regular file has no backing object after vn_open"));
1978 #endif
1979 
1980 	vrele(vp);
1981 
1982 	/*
1983 	 * release our private reference, leaving the one associated with the
1984 	 * descriptor table intact.
1985 	 */
1986 	fsetfd(fdp, fp, indx);
1987 	fdrop(fp);
1988 	*res = indx;
1989 	if (oflags & O_CLOEXEC)
1990 		error = fsetfdflags(fdp, *res, UF_EXCLOSE);
1991 	return (error);
1992 }
1993 
1994 /*
1995  * open_args(char *path, int flags, int mode)
1996  *
1997  * Check permissions, allocate an open file structure,
1998  * and call the device open routine if any.
1999  */
2000 int
2001 sys_open(struct open_args *uap)
2002 {
2003 	struct nlookupdata nd;
2004 	int error;
2005 
2006 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2007 	if (error == 0) {
2008 		error = kern_open(&nd, uap->flags,
2009 				    uap->mode, &uap->sysmsg_result);
2010 	}
2011 	nlookup_done(&nd);
2012 	return (error);
2013 }
2014 
2015 /*
2016  * openat_args(int fd, char *path, int flags, int mode)
2017  */
2018 int
2019 sys_openat(struct openat_args *uap)
2020 {
2021 	struct nlookupdata nd;
2022 	int error;
2023 	struct file *fp;
2024 
2025 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2026 	if (error == 0) {
2027 		error = kern_open(&nd, uap->flags, uap->mode,
2028 					&uap->sysmsg_result);
2029 	}
2030 	nlookup_done_at(&nd, fp);
2031 	return (error);
2032 }
2033 
2034 int
2035 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
2036 {
2037 	struct thread *td = curthread;
2038 	struct proc *p = td->td_proc;
2039 	struct vnode *vp;
2040 	struct vattr vattr;
2041 	int error;
2042 	int whiteout = 0;
2043 
2044 	KKASSERT(p);
2045 
2046 	VATTR_NULL(&vattr);
2047 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2048 	vattr.va_rmajor = rmajor;
2049 	vattr.va_rminor = rminor;
2050 
2051 	switch (mode & S_IFMT) {
2052 	case S_IFMT:	/* used by badsect to flag bad sectors */
2053 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_BAD, 0);
2054 		vattr.va_type = VBAD;
2055 		break;
2056 	case S_IFCHR:
2057 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2058 		vattr.va_type = VCHR;
2059 		break;
2060 	case S_IFBLK:
2061 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2062 		vattr.va_type = VBLK;
2063 		break;
2064 	case S_IFWHT:
2065 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_WHT, 0);
2066 		whiteout = 1;
2067 		break;
2068 	case S_IFDIR:	/* special directories support for HAMMER */
2069 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_DIR, 0);
2070 		vattr.va_type = VDIR;
2071 		break;
2072 	default:
2073 		error = EINVAL;
2074 		break;
2075 	}
2076 
2077 	if (error)
2078 		return (error);
2079 
2080 	bwillinode(1);
2081 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2082 	if ((error = nlookup(nd)) != 0)
2083 		return (error);
2084 	if (nd->nl_nch.ncp->nc_vp)
2085 		return (EEXIST);
2086 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2087 		return (error);
2088 
2089 	if (whiteout) {
2090 		error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
2091 				      nd->nl_cred, NAMEI_CREATE);
2092 	} else {
2093 		vp = NULL;
2094 		error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
2095 				   &vp, nd->nl_cred, &vattr);
2096 		if (error == 0)
2097 			vput(vp);
2098 	}
2099 	return (error);
2100 }
2101 
2102 /*
2103  * mknod_args(char *path, int mode, int dev)
2104  *
2105  * Create a special file.
2106  */
2107 int
2108 sys_mknod(struct mknod_args *uap)
2109 {
2110 	struct nlookupdata nd;
2111 	int error;
2112 
2113 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2114 	if (error == 0) {
2115 		error = kern_mknod(&nd, uap->mode,
2116 				   umajor(uap->dev), uminor(uap->dev));
2117 	}
2118 	nlookup_done(&nd);
2119 	return (error);
2120 }
2121 
2122 /*
2123  * mknodat_args(int fd, char *path, mode_t mode, dev_t dev)
2124  *
2125  * Create a special file.  The path is relative to the directory associated
2126  * with fd.
2127  */
2128 int
2129 sys_mknodat(struct mknodat_args *uap)
2130 {
2131 	struct nlookupdata nd;
2132 	struct file *fp;
2133 	int error;
2134 
2135 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2136 	if (error == 0) {
2137 		error = kern_mknod(&nd, uap->mode,
2138 				   umajor(uap->dev), uminor(uap->dev));
2139 	}
2140 	nlookup_done_at(&nd, fp);
2141 	return (error);
2142 }
2143 
2144 int
2145 kern_mkfifo(struct nlookupdata *nd, int mode)
2146 {
2147 	struct thread *td = curthread;
2148 	struct proc *p = td->td_proc;
2149 	struct vattr vattr;
2150 	struct vnode *vp;
2151 	int error;
2152 
2153 	bwillinode(1);
2154 
2155 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2156 	if ((error = nlookup(nd)) != 0)
2157 		return (error);
2158 	if (nd->nl_nch.ncp->nc_vp)
2159 		return (EEXIST);
2160 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2161 		return (error);
2162 
2163 	VATTR_NULL(&vattr);
2164 	vattr.va_type = VFIFO;
2165 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2166 	vp = NULL;
2167 	error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
2168 	if (error == 0)
2169 		vput(vp);
2170 	return (error);
2171 }
2172 
2173 /*
2174  * mkfifo_args(char *path, int mode)
2175  *
2176  * Create a named pipe.
2177  */
2178 int
2179 sys_mkfifo(struct mkfifo_args *uap)
2180 {
2181 	struct nlookupdata nd;
2182 	int error;
2183 
2184 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2185 	if (error == 0)
2186 		error = kern_mkfifo(&nd, uap->mode);
2187 	nlookup_done(&nd);
2188 	return (error);
2189 }
2190 
2191 /*
2192  * mkfifoat_args(int fd, char *path, mode_t mode)
2193  *
2194  * Create a named pipe.  The path is relative to the directory associated
2195  * with fd.
2196  */
2197 int
2198 sys_mkfifoat(struct mkfifoat_args *uap)
2199 {
2200 	struct nlookupdata nd;
2201 	struct file *fp;
2202 	int error;
2203 
2204 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2205 	if (error == 0)
2206 		error = kern_mkfifo(&nd, uap->mode);
2207 	nlookup_done_at(&nd, fp);
2208 	return (error);
2209 }
2210 
2211 static int hardlink_check_uid = 0;
2212 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
2213     &hardlink_check_uid, 0,
2214     "Unprivileged processes cannot create hard links to files owned by other "
2215     "users");
2216 static int hardlink_check_gid = 0;
2217 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
2218     &hardlink_check_gid, 0,
2219     "Unprivileged processes cannot create hard links to files owned by other "
2220     "groups");
2221 
2222 static int
2223 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
2224 {
2225 	struct vattr va;
2226 	int error;
2227 
2228 	/*
2229 	 * Shortcut if disabled
2230 	 */
2231 	if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
2232 		return (0);
2233 
2234 	/*
2235 	 * Privileged user can always hardlink
2236 	 */
2237 	if (priv_check_cred(cred, PRIV_VFS_LINK, 0) == 0)
2238 		return (0);
2239 
2240 	/*
2241 	 * Otherwise only if the originating file is owned by the
2242 	 * same user or group.  Note that any group is allowed if
2243 	 * the file is owned by the caller.
2244 	 */
2245 	error = VOP_GETATTR(vp, &va);
2246 	if (error != 0)
2247 		return (error);
2248 
2249 	if (hardlink_check_uid) {
2250 		if (cred->cr_uid != va.va_uid)
2251 			return (EPERM);
2252 	}
2253 
2254 	if (hardlink_check_gid) {
2255 		if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2256 			return (EPERM);
2257 	}
2258 
2259 	return (0);
2260 }
2261 
2262 int
2263 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2264 {
2265 	struct thread *td = curthread;
2266 	struct vnode *vp;
2267 	int error;
2268 
2269 	/*
2270 	 * Lookup the source and obtained a locked vnode.
2271 	 *
2272 	 * You may only hardlink a file which you have write permission
2273 	 * on or which you own.
2274 	 *
2275 	 * XXX relookup on vget failure / race ?
2276 	 */
2277 	bwillinode(1);
2278 	nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK;
2279 	if ((error = nlookup(nd)) != 0)
2280 		return (error);
2281 	vp = nd->nl_nch.ncp->nc_vp;
2282 	KKASSERT(vp != NULL);
2283 	if (vp->v_type == VDIR)
2284 		return (EPERM);		/* POSIX */
2285 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2286 		return (error);
2287 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2288 		return (error);
2289 
2290 	/*
2291 	 * Unlock the source so we can lookup the target without deadlocking
2292 	 * (XXX vp is locked already, possible other deadlock?).  The target
2293 	 * must not exist.
2294 	 */
2295 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2296 	nd->nl_flags &= ~NLC_NCPISLOCKED;
2297 	cache_unlock(&nd->nl_nch);
2298 	vn_unlock(vp);
2299 
2300 	linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2301 	if ((error = nlookup(linknd)) != 0) {
2302 		vrele(vp);
2303 		return (error);
2304 	}
2305 	if (linknd->nl_nch.ncp->nc_vp) {
2306 		vrele(vp);
2307 		return (EEXIST);
2308 	}
2309 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
2310 		vrele(vp);
2311 		return (error);
2312 	}
2313 
2314 	/*
2315 	 * Finally run the new API VOP.
2316 	 */
2317 	error = can_hardlink(vp, td, td->td_ucred);
2318 	if (error == 0) {
2319 		error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2320 				  vp, linknd->nl_cred);
2321 	}
2322 	vput(vp);
2323 	return (error);
2324 }
2325 
2326 /*
2327  * link_args(char *path, char *link)
2328  *
2329  * Make a hard file link.
2330  */
2331 int
2332 sys_link(struct link_args *uap)
2333 {
2334 	struct nlookupdata nd, linknd;
2335 	int error;
2336 
2337 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2338 	if (error == 0) {
2339 		error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2340 		if (error == 0)
2341 			error = kern_link(&nd, &linknd);
2342 		nlookup_done(&linknd);
2343 	}
2344 	nlookup_done(&nd);
2345 	return (error);
2346 }
2347 
2348 /*
2349  * linkat_args(int fd1, char *path1, int fd2, char *path2, int flags)
2350  *
2351  * Make a hard file link. The path1 argument is relative to the directory
2352  * associated with fd1, and similarly the path2 argument is relative to
2353  * the directory associated with fd2.
2354  */
2355 int
2356 sys_linkat(struct linkat_args *uap)
2357 {
2358 	struct nlookupdata nd, linknd;
2359 	struct file *fp1, *fp2;
2360 	int error;
2361 
2362 	error = nlookup_init_at(&nd, &fp1, uap->fd1, uap->path1, UIO_USERSPACE,
2363 	    (uap->flags & AT_SYMLINK_FOLLOW) ? NLC_FOLLOW : 0);
2364 	if (error == 0) {
2365 		error = nlookup_init_at(&linknd, &fp2, uap->fd2,
2366 		    uap->path2, UIO_USERSPACE, 0);
2367 		if (error == 0)
2368 			error = kern_link(&nd, &linknd);
2369 		nlookup_done_at(&linknd, fp2);
2370 	}
2371 	nlookup_done_at(&nd, fp1);
2372 	return (error);
2373 }
2374 
2375 int
2376 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2377 {
2378 	struct vattr vattr;
2379 	struct vnode *vp;
2380 	struct vnode *dvp;
2381 	int error;
2382 
2383 	bwillinode(1);
2384 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2385 	if ((error = nlookup(nd)) != 0)
2386 		return (error);
2387 	if (nd->nl_nch.ncp->nc_vp)
2388 		return (EEXIST);
2389 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2390 		return (error);
2391 	dvp = nd->nl_dvp;
2392 	VATTR_NULL(&vattr);
2393 	vattr.va_mode = mode;
2394 	error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2395 	if (error == 0)
2396 		vput(vp);
2397 	return (error);
2398 }
2399 
2400 /*
2401  * symlink(char *path, char *link)
2402  *
2403  * Make a symbolic link.
2404  */
2405 int
2406 sys_symlink(struct symlink_args *uap)
2407 {
2408 	struct thread *td = curthread;
2409 	struct nlookupdata nd;
2410 	char *path;
2411 	int error;
2412 	int mode;
2413 
2414 	path = objcache_get(namei_oc, M_WAITOK);
2415 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2416 	if (error == 0) {
2417 		error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2418 		if (error == 0) {
2419 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2420 			error = kern_symlink(&nd, path, mode);
2421 		}
2422 		nlookup_done(&nd);
2423 	}
2424 	objcache_put(namei_oc, path);
2425 	return (error);
2426 }
2427 
2428 /*
2429  * symlinkat_args(char *path1, int fd, char *path2)
2430  *
2431  * Make a symbolic link.  The path2 argument is relative to the directory
2432  * associated with fd.
2433  */
2434 int
2435 sys_symlinkat(struct symlinkat_args *uap)
2436 {
2437 	struct thread *td = curthread;
2438 	struct nlookupdata nd;
2439 	struct file *fp;
2440 	char *path1;
2441 	int error;
2442 	int mode;
2443 
2444 	path1 = objcache_get(namei_oc, M_WAITOK);
2445 	error = copyinstr(uap->path1, path1, MAXPATHLEN, NULL);
2446 	if (error == 0) {
2447 		error = nlookup_init_at(&nd, &fp, uap->fd, uap->path2,
2448 		    UIO_USERSPACE, 0);
2449 		if (error == 0) {
2450 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2451 			error = kern_symlink(&nd, path1, mode);
2452 		}
2453 		nlookup_done_at(&nd, fp);
2454 	}
2455 	objcache_put(namei_oc, path1);
2456 	return (error);
2457 }
2458 
2459 /*
2460  * undelete_args(char *path)
2461  *
2462  * Delete a whiteout from the filesystem.
2463  */
2464 int
2465 sys_undelete(struct undelete_args *uap)
2466 {
2467 	struct nlookupdata nd;
2468 	int error;
2469 
2470 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2471 	bwillinode(1);
2472 	nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2473 	if (error == 0)
2474 		error = nlookup(&nd);
2475 	if (error == 0)
2476 		error = ncp_writechk(&nd.nl_nch);
2477 	if (error == 0) {
2478 		error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2479 				      NAMEI_DELETE);
2480 	}
2481 	nlookup_done(&nd);
2482 	return (error);
2483 }
2484 
2485 int
2486 kern_unlink(struct nlookupdata *nd)
2487 {
2488 	int error;
2489 
2490 	bwillinode(1);
2491 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2492 	if ((error = nlookup(nd)) != 0)
2493 		return (error);
2494 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2495 		return (error);
2496 	error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2497 	return (error);
2498 }
2499 
2500 /*
2501  * unlink_args(char *path)
2502  *
2503  * Delete a name from the filesystem.
2504  */
2505 int
2506 sys_unlink(struct unlink_args *uap)
2507 {
2508 	struct nlookupdata nd;
2509 	int error;
2510 
2511 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2512 	if (error == 0)
2513 		error = kern_unlink(&nd);
2514 	nlookup_done(&nd);
2515 	return (error);
2516 }
2517 
2518 
2519 /*
2520  * unlinkat_args(int fd, char *path, int flags)
2521  *
2522  * Delete the file or directory entry pointed to by fd/path.
2523  */
2524 int
2525 sys_unlinkat(struct unlinkat_args *uap)
2526 {
2527 	struct nlookupdata nd;
2528 	struct file *fp;
2529 	int error;
2530 
2531 	if (uap->flags & ~AT_REMOVEDIR)
2532 		return (EINVAL);
2533 
2534 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2535 	if (error == 0) {
2536 		if (uap->flags & AT_REMOVEDIR)
2537 			error = kern_rmdir(&nd);
2538 		else
2539 			error = kern_unlink(&nd);
2540 	}
2541 	nlookup_done_at(&nd, fp);
2542 	return (error);
2543 }
2544 
2545 int
2546 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2547 {
2548 	struct thread *td = curthread;
2549 	struct proc *p = td->td_proc;
2550 	struct file *fp;
2551 	struct vnode *vp;
2552 	struct vattr vattr;
2553 	off_t new_offset;
2554 	int error;
2555 
2556 	fp = holdfp(p->p_fd, fd, -1);
2557 	if (fp == NULL)
2558 		return (EBADF);
2559 	if (fp->f_type != DTYPE_VNODE) {
2560 		error = ESPIPE;
2561 		goto done;
2562 	}
2563 	vp = (struct vnode *)fp->f_data;
2564 
2565 	switch (whence) {
2566 	case L_INCR:
2567 		spin_lock(&fp->f_spin);
2568 		new_offset = fp->f_offset + offset;
2569 		error = 0;
2570 		break;
2571 	case L_XTND:
2572 		error = VOP_GETATTR(vp, &vattr);
2573 		spin_lock(&fp->f_spin);
2574 		new_offset = offset + vattr.va_size;
2575 		break;
2576 	case L_SET:
2577 		new_offset = offset;
2578 		error = 0;
2579 		spin_lock(&fp->f_spin);
2580 		break;
2581 	default:
2582 		new_offset = 0;
2583 		error = EINVAL;
2584 		spin_lock(&fp->f_spin);
2585 		break;
2586 	}
2587 
2588 	/*
2589 	 * Validate the seek position.  Negative offsets are not allowed
2590 	 * for regular files or directories.
2591 	 *
2592 	 * Normally we would also not want to allow negative offsets for
2593 	 * character and block-special devices.  However kvm addresses
2594 	 * on 64 bit architectures might appear to be negative and must
2595 	 * be allowed.
2596 	 */
2597 	if (error == 0) {
2598 		if (new_offset < 0 &&
2599 		    (vp->v_type == VREG || vp->v_type == VDIR)) {
2600 			error = EINVAL;
2601 		} else {
2602 			fp->f_offset = new_offset;
2603 		}
2604 	}
2605 	*res = fp->f_offset;
2606 	spin_unlock(&fp->f_spin);
2607 done:
2608 	fdrop(fp);
2609 	return (error);
2610 }
2611 
2612 /*
2613  * lseek_args(int fd, int pad, off_t offset, int whence)
2614  *
2615  * Reposition read/write file offset.
2616  */
2617 int
2618 sys_lseek(struct lseek_args *uap)
2619 {
2620 	int error;
2621 
2622 	error = kern_lseek(uap->fd, uap->offset, uap->whence,
2623 			   &uap->sysmsg_offset);
2624 
2625 	return (error);
2626 }
2627 
2628 /*
2629  * Check if current process can access given file.  amode is a bitmask of *_OK
2630  * access bits.  flags is a bitmask of AT_* flags.
2631  */
2632 int
2633 kern_access(struct nlookupdata *nd, int amode, int flags)
2634 {
2635 	struct vnode *vp;
2636 	int error, mode;
2637 
2638 	if (flags & ~AT_EACCESS)
2639 		return (EINVAL);
2640 	if ((error = nlookup(nd)) != 0)
2641 		return (error);
2642 retry:
2643 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2644 	if (error)
2645 		return (error);
2646 
2647 	/* Flags == 0 means only check for existence. */
2648 	if (amode) {
2649 		mode = 0;
2650 		if (amode & R_OK)
2651 			mode |= VREAD;
2652 		if (amode & W_OK)
2653 			mode |= VWRITE;
2654 		if (amode & X_OK)
2655 			mode |= VEXEC;
2656 		if ((mode & VWRITE) == 0 ||
2657 		    (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2658 			error = VOP_ACCESS_FLAGS(vp, mode, flags, nd->nl_cred);
2659 
2660 		/*
2661 		 * If the file handle is stale we have to re-resolve the
2662 		 * entry.  This is a hack at the moment.
2663 		 */
2664 		if (error == ESTALE) {
2665 			vput(vp);
2666 			cache_setunresolved(&nd->nl_nch);
2667 			error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2668 			if (error == 0) {
2669 				vp = NULL;
2670 				goto retry;
2671 			}
2672 			return(error);
2673 		}
2674 	}
2675 	vput(vp);
2676 	return (error);
2677 }
2678 
2679 /*
2680  * access_args(char *path, int flags)
2681  *
2682  * Check access permissions.
2683  */
2684 int
2685 sys_access(struct access_args *uap)
2686 {
2687 	struct nlookupdata nd;
2688 	int error;
2689 
2690 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2691 	if (error == 0)
2692 		error = kern_access(&nd, uap->flags, 0);
2693 	nlookup_done(&nd);
2694 	return (error);
2695 }
2696 
2697 
2698 /*
2699  * eaccess_args(char *path, int flags)
2700  *
2701  * Check access permissions.
2702  */
2703 int
2704 sys_eaccess(struct eaccess_args *uap)
2705 {
2706 	struct nlookupdata nd;
2707 	int error;
2708 
2709 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2710 	if (error == 0)
2711 		error = kern_access(&nd, uap->flags, AT_EACCESS);
2712 	nlookup_done(&nd);
2713 	return (error);
2714 }
2715 
2716 
2717 /*
2718  * faccessat_args(int fd, char *path, int amode, int flags)
2719  *
2720  * Check access permissions.
2721  */
2722 int
2723 sys_faccessat(struct faccessat_args *uap)
2724 {
2725 	struct nlookupdata nd;
2726 	struct file *fp;
2727 	int error;
2728 
2729 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE,
2730 				NLC_FOLLOW);
2731 	if (error == 0)
2732 		error = kern_access(&nd, uap->amode, uap->flags);
2733 	nlookup_done_at(&nd, fp);
2734 	return (error);
2735 }
2736 
2737 
2738 int
2739 kern_stat(struct nlookupdata *nd, struct stat *st)
2740 {
2741 	int error;
2742 	struct vnode *vp;
2743 
2744 	if ((error = nlookup(nd)) != 0)
2745 		return (error);
2746 again:
2747 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2748 		return (ENOENT);
2749 
2750 	if ((error = vget(vp, LK_SHARED)) != 0)
2751 		return (error);
2752 	error = vn_stat(vp, st, nd->nl_cred);
2753 
2754 	/*
2755 	 * If the file handle is stale we have to re-resolve the entry.  This
2756 	 * is a hack at the moment.
2757 	 */
2758 	if (error == ESTALE) {
2759 		vput(vp);
2760 		cache_setunresolved(&nd->nl_nch);
2761 		error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2762 		if (error == 0)
2763 			goto again;
2764 	} else {
2765 		vput(vp);
2766 	}
2767 	return (error);
2768 }
2769 
2770 /*
2771  * stat_args(char *path, struct stat *ub)
2772  *
2773  * Get file status; this version follows links.
2774  */
2775 int
2776 sys_stat(struct stat_args *uap)
2777 {
2778 	struct nlookupdata nd;
2779 	struct stat st;
2780 	int error;
2781 
2782 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2783 	if (error == 0) {
2784 		error = kern_stat(&nd, &st);
2785 		if (error == 0)
2786 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2787 	}
2788 	nlookup_done(&nd);
2789 	return (error);
2790 }
2791 
2792 /*
2793  * lstat_args(char *path, struct stat *ub)
2794  *
2795  * Get file status; this version does not follow links.
2796  */
2797 int
2798 sys_lstat(struct lstat_args *uap)
2799 {
2800 	struct nlookupdata nd;
2801 	struct stat st;
2802 	int error;
2803 
2804 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2805 	if (error == 0) {
2806 		error = kern_stat(&nd, &st);
2807 		if (error == 0)
2808 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2809 	}
2810 	nlookup_done(&nd);
2811 	return (error);
2812 }
2813 
2814 /*
2815  * fstatat_args(int fd, char *path, struct stat *sb, int flags)
2816  *
2817  * Get status of file pointed to by fd/path.
2818  */
2819 int
2820 sys_fstatat(struct fstatat_args *uap)
2821 {
2822 	struct nlookupdata nd;
2823 	struct stat st;
2824 	int error;
2825 	int flags;
2826 	struct file *fp;
2827 
2828 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
2829 		return (EINVAL);
2830 
2831 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
2832 
2833 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
2834 				UIO_USERSPACE, flags);
2835 	if (error == 0) {
2836 		error = kern_stat(&nd, &st);
2837 		if (error == 0)
2838 			error = copyout(&st, uap->sb, sizeof(*uap->sb));
2839 	}
2840 	nlookup_done_at(&nd, fp);
2841 	return (error);
2842 }
2843 
2844 static int
2845 kern_pathconf(char *path, int name, int flags, register_t *sysmsg_regp)
2846 {
2847 	struct nlookupdata nd;
2848 	struct vnode *vp;
2849 	int error;
2850 
2851 	vp = NULL;
2852 	error = nlookup_init(&nd, path, UIO_USERSPACE, flags);
2853 	if (error == 0)
2854 		error = nlookup(&nd);
2855 	if (error == 0)
2856 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2857 	nlookup_done(&nd);
2858 	if (error == 0) {
2859 		error = VOP_PATHCONF(vp, name, sysmsg_regp);
2860 		vput(vp);
2861 	}
2862 	return (error);
2863 }
2864 
2865 /*
2866  * pathconf_Args(char *path, int name)
2867  *
2868  * Get configurable pathname variables.
2869  */
2870 int
2871 sys_pathconf(struct pathconf_args *uap)
2872 {
2873 	return (kern_pathconf(uap->path, uap->name, NLC_FOLLOW,
2874 		&uap->sysmsg_reg));
2875 }
2876 
2877 /*
2878  * lpathconf_Args(char *path, int name)
2879  *
2880  * Get configurable pathname variables, but don't follow symlinks.
2881  */
2882 int
2883 sys_lpathconf(struct lpathconf_args *uap)
2884 {
2885 	return (kern_pathconf(uap->path, uap->name, 0, &uap->sysmsg_reg));
2886 }
2887 
2888 /*
2889  * XXX: daver
2890  * kern_readlink isn't properly split yet.  There is a copyin burried
2891  * in VOP_READLINK().
2892  */
2893 int
2894 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2895 {
2896 	struct thread *td = curthread;
2897 	struct vnode *vp;
2898 	struct iovec aiov;
2899 	struct uio auio;
2900 	int error;
2901 
2902 	if ((error = nlookup(nd)) != 0)
2903 		return (error);
2904 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2905 	if (error)
2906 		return (error);
2907 	if (vp->v_type != VLNK) {
2908 		error = EINVAL;
2909 	} else {
2910 		aiov.iov_base = buf;
2911 		aiov.iov_len = count;
2912 		auio.uio_iov = &aiov;
2913 		auio.uio_iovcnt = 1;
2914 		auio.uio_offset = 0;
2915 		auio.uio_rw = UIO_READ;
2916 		auio.uio_segflg = UIO_USERSPACE;
2917 		auio.uio_td = td;
2918 		auio.uio_resid = count;
2919 		error = VOP_READLINK(vp, &auio, td->td_ucred);
2920 	}
2921 	vput(vp);
2922 	*res = count - auio.uio_resid;
2923 	return (error);
2924 }
2925 
2926 /*
2927  * readlink_args(char *path, char *buf, int count)
2928  *
2929  * Return target name of a symbolic link.
2930  */
2931 int
2932 sys_readlink(struct readlink_args *uap)
2933 {
2934 	struct nlookupdata nd;
2935 	int error;
2936 
2937 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2938 	if (error == 0) {
2939 		error = kern_readlink(&nd, uap->buf, uap->count,
2940 					&uap->sysmsg_result);
2941 	}
2942 	nlookup_done(&nd);
2943 	return (error);
2944 }
2945 
2946 /*
2947  * readlinkat_args(int fd, char *path, char *buf, size_t bufsize)
2948  *
2949  * Return target name of a symbolic link.  The path is relative to the
2950  * directory associated with fd.
2951  */
2952 int
2953 sys_readlinkat(struct readlinkat_args *uap)
2954 {
2955 	struct nlookupdata nd;
2956 	struct file *fp;
2957 	int error;
2958 
2959 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2960 	if (error == 0) {
2961 		error = kern_readlink(&nd, uap->buf, uap->bufsize,
2962 					&uap->sysmsg_result);
2963 	}
2964 	nlookup_done_at(&nd, fp);
2965 	return (error);
2966 }
2967 
2968 static int
2969 setfflags(struct vnode *vp, int flags)
2970 {
2971 	struct thread *td = curthread;
2972 	int error;
2973 	struct vattr vattr;
2974 
2975 	/*
2976 	 * Prevent non-root users from setting flags on devices.  When
2977 	 * a device is reused, users can retain ownership of the device
2978 	 * if they are allowed to set flags and programs assume that
2979 	 * chown can't fail when done as root.
2980 	 */
2981 	if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2982 	    ((error = priv_check_cred(td->td_ucred, PRIV_VFS_CHFLAGS_DEV, 0)) != 0))
2983 		return (error);
2984 
2985 	/*
2986 	 * note: vget is required for any operation that might mod the vnode
2987 	 * so VINACTIVE is properly cleared.
2988 	 */
2989 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2990 		VATTR_NULL(&vattr);
2991 		vattr.va_flags = flags;
2992 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
2993 		vput(vp);
2994 	}
2995 	return (error);
2996 }
2997 
2998 /*
2999  * chflags(char *path, int flags)
3000  *
3001  * Change flags of a file given a path name.
3002  */
3003 int
3004 sys_chflags(struct chflags_args *uap)
3005 {
3006 	struct nlookupdata nd;
3007 	struct vnode *vp;
3008 	int error;
3009 
3010 	vp = NULL;
3011 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3012 	if (error == 0)
3013 		error = nlookup(&nd);
3014 	if (error == 0)
3015 		error = ncp_writechk(&nd.nl_nch);
3016 	if (error == 0)
3017 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3018 	nlookup_done(&nd);
3019 	if (error == 0) {
3020 		error = setfflags(vp, uap->flags);
3021 		vrele(vp);
3022 	}
3023 	return (error);
3024 }
3025 
3026 /*
3027  * lchflags(char *path, int flags)
3028  *
3029  * Change flags of a file given a path name, but don't follow symlinks.
3030  */
3031 int
3032 sys_lchflags(struct lchflags_args *uap)
3033 {
3034 	struct nlookupdata nd;
3035 	struct vnode *vp;
3036 	int error;
3037 
3038 	vp = NULL;
3039 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3040 	if (error == 0)
3041 		error = nlookup(&nd);
3042 	if (error == 0)
3043 		error = ncp_writechk(&nd.nl_nch);
3044 	if (error == 0)
3045 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3046 	nlookup_done(&nd);
3047 	if (error == 0) {
3048 		error = setfflags(vp, uap->flags);
3049 		vrele(vp);
3050 	}
3051 	return (error);
3052 }
3053 
3054 /*
3055  * fchflags_args(int fd, int flags)
3056  *
3057  * Change flags of a file given a file descriptor.
3058  */
3059 int
3060 sys_fchflags(struct fchflags_args *uap)
3061 {
3062 	struct thread *td = curthread;
3063 	struct proc *p = td->td_proc;
3064 	struct file *fp;
3065 	int error;
3066 
3067 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3068 		return (error);
3069 	if (fp->f_nchandle.ncp)
3070 		error = ncp_writechk(&fp->f_nchandle);
3071 	if (error == 0)
3072 		error = setfflags((struct vnode *) fp->f_data, uap->flags);
3073 	fdrop(fp);
3074 	return (error);
3075 }
3076 
3077 static int
3078 setfmode(struct vnode *vp, int mode)
3079 {
3080 	struct thread *td = curthread;
3081 	int error;
3082 	struct vattr vattr;
3083 
3084 	/*
3085 	 * note: vget is required for any operation that might mod the vnode
3086 	 * so VINACTIVE is properly cleared.
3087 	 */
3088 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3089 		VATTR_NULL(&vattr);
3090 		vattr.va_mode = mode & ALLPERMS;
3091 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3092 		vput(vp);
3093 	}
3094 	return error;
3095 }
3096 
3097 int
3098 kern_chmod(struct nlookupdata *nd, int mode)
3099 {
3100 	struct vnode *vp;
3101 	int error;
3102 
3103 	if ((error = nlookup(nd)) != 0)
3104 		return (error);
3105 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3106 		return (error);
3107 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3108 		error = setfmode(vp, mode);
3109 	vrele(vp);
3110 	return (error);
3111 }
3112 
3113 /*
3114  * chmod_args(char *path, int mode)
3115  *
3116  * Change mode of a file given path name.
3117  */
3118 int
3119 sys_chmod(struct chmod_args *uap)
3120 {
3121 	struct nlookupdata nd;
3122 	int error;
3123 
3124 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3125 	if (error == 0)
3126 		error = kern_chmod(&nd, uap->mode);
3127 	nlookup_done(&nd);
3128 	return (error);
3129 }
3130 
3131 /*
3132  * lchmod_args(char *path, int mode)
3133  *
3134  * Change mode of a file given path name (don't follow links.)
3135  */
3136 int
3137 sys_lchmod(struct lchmod_args *uap)
3138 {
3139 	struct nlookupdata nd;
3140 	int error;
3141 
3142 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3143 	if (error == 0)
3144 		error = kern_chmod(&nd, uap->mode);
3145 	nlookup_done(&nd);
3146 	return (error);
3147 }
3148 
3149 /*
3150  * fchmod_args(int fd, int mode)
3151  *
3152  * Change mode of a file given a file descriptor.
3153  */
3154 int
3155 sys_fchmod(struct fchmod_args *uap)
3156 {
3157 	struct thread *td = curthread;
3158 	struct proc *p = td->td_proc;
3159 	struct file *fp;
3160 	int error;
3161 
3162 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3163 		return (error);
3164 	if (fp->f_nchandle.ncp)
3165 		error = ncp_writechk(&fp->f_nchandle);
3166 	if (error == 0)
3167 		error = setfmode((struct vnode *)fp->f_data, uap->mode);
3168 	fdrop(fp);
3169 	return (error);
3170 }
3171 
3172 /*
3173  * fchmodat_args(char *path, int mode)
3174  *
3175  * Change mode of a file pointed to by fd/path.
3176  */
3177 int
3178 sys_fchmodat(struct fchmodat_args *uap)
3179 {
3180 	struct nlookupdata nd;
3181 	struct file *fp;
3182 	int error;
3183 	int flags;
3184 
3185 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3186 		return (EINVAL);
3187 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3188 
3189 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3190 				UIO_USERSPACE, flags);
3191 	if (error == 0)
3192 		error = kern_chmod(&nd, uap->mode);
3193 	nlookup_done_at(&nd, fp);
3194 	return (error);
3195 }
3196 
3197 static int
3198 setfown(struct mount *mp, struct vnode *vp, uid_t uid, gid_t gid)
3199 {
3200 	struct thread *td = curthread;
3201 	int error;
3202 	struct vattr vattr;
3203 	uid_t o_uid;
3204 	gid_t o_gid;
3205 	uint64_t size;
3206 
3207 	/*
3208 	 * note: vget is required for any operation that might mod the vnode
3209 	 * so VINACTIVE is properly cleared.
3210 	 */
3211 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3212 		if ((error = VOP_GETATTR(vp, &vattr)) != 0)
3213 			return error;
3214 		o_uid = vattr.va_uid;
3215 		o_gid = vattr.va_gid;
3216 		size = vattr.va_size;
3217 
3218 		VATTR_NULL(&vattr);
3219 		vattr.va_uid = uid;
3220 		vattr.va_gid = gid;
3221 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3222 		vput(vp);
3223 	}
3224 
3225 	if (error == 0) {
3226 		if (uid == -1)
3227 			uid = o_uid;
3228 		if (gid == -1)
3229 			gid = o_gid;
3230 		VFS_ACCOUNT(mp, o_uid, o_gid, -size);
3231 		VFS_ACCOUNT(mp,   uid,   gid,  size);
3232 	}
3233 
3234 	return error;
3235 }
3236 
3237 int
3238 kern_chown(struct nlookupdata *nd, int uid, int gid)
3239 {
3240 	struct vnode *vp;
3241 	int error;
3242 
3243 	if ((error = nlookup(nd)) != 0)
3244 		return (error);
3245 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3246 		return (error);
3247 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3248 		error = setfown(nd->nl_nch.mount, vp, uid, gid);
3249 	vrele(vp);
3250 	return (error);
3251 }
3252 
3253 /*
3254  * chown(char *path, int uid, int gid)
3255  *
3256  * Set ownership given a path name.
3257  */
3258 int
3259 sys_chown(struct chown_args *uap)
3260 {
3261 	struct nlookupdata nd;
3262 	int error;
3263 
3264 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3265 	if (error == 0)
3266 		error = kern_chown(&nd, uap->uid, uap->gid);
3267 	nlookup_done(&nd);
3268 	return (error);
3269 }
3270 
3271 /*
3272  * lchown_args(char *path, int uid, int gid)
3273  *
3274  * Set ownership given a path name, do not cross symlinks.
3275  */
3276 int
3277 sys_lchown(struct lchown_args *uap)
3278 {
3279 	struct nlookupdata nd;
3280 	int error;
3281 
3282 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3283 	if (error == 0)
3284 		error = kern_chown(&nd, uap->uid, uap->gid);
3285 	nlookup_done(&nd);
3286 	return (error);
3287 }
3288 
3289 /*
3290  * fchown_args(int fd, int uid, int gid)
3291  *
3292  * Set ownership given a file descriptor.
3293  */
3294 int
3295 sys_fchown(struct fchown_args *uap)
3296 {
3297 	struct thread *td = curthread;
3298 	struct proc *p = td->td_proc;
3299 	struct file *fp;
3300 	int error;
3301 
3302 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3303 		return (error);
3304 	if (fp->f_nchandle.ncp)
3305 		error = ncp_writechk(&fp->f_nchandle);
3306 	if (error == 0)
3307 		error = setfown(p->p_fd->fd_ncdir.mount,
3308 			(struct vnode *)fp->f_data, uap->uid, uap->gid);
3309 	fdrop(fp);
3310 	return (error);
3311 }
3312 
3313 /*
3314  * fchownat(int fd, char *path, int uid, int gid, int flags)
3315  *
3316  * Set ownership of file pointed to by fd/path.
3317  */
3318 int
3319 sys_fchownat(struct fchownat_args *uap)
3320 {
3321 	struct nlookupdata nd;
3322 	struct file *fp;
3323 	int error;
3324 	int flags;
3325 
3326 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3327 		return (EINVAL);
3328 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3329 
3330 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3331 				UIO_USERSPACE, flags);
3332 	if (error == 0)
3333 		error = kern_chown(&nd, uap->uid, uap->gid);
3334 	nlookup_done_at(&nd, fp);
3335 	return (error);
3336 }
3337 
3338 
3339 static int
3340 getutimes(const struct timeval *tvp, struct timespec *tsp)
3341 {
3342 	struct timeval tv[2];
3343 
3344 	if (tvp == NULL) {
3345 		microtime(&tv[0]);
3346 		TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
3347 		tsp[1] = tsp[0];
3348 	} else {
3349 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
3350 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
3351 	}
3352 	return 0;
3353 }
3354 
3355 static int
3356 setutimes(struct vnode *vp, struct vattr *vattr,
3357 	  const struct timespec *ts, int nullflag)
3358 {
3359 	struct thread *td = curthread;
3360 	int error;
3361 
3362 	VATTR_NULL(vattr);
3363 	vattr->va_atime = ts[0];
3364 	vattr->va_mtime = ts[1];
3365 	if (nullflag)
3366 		vattr->va_vaflags |= VA_UTIMES_NULL;
3367 	error = VOP_SETATTR(vp, vattr, td->td_ucred);
3368 
3369 	return error;
3370 }
3371 
3372 int
3373 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
3374 {
3375 	struct timespec ts[2];
3376 	struct vnode *vp;
3377 	struct vattr vattr;
3378 	int error;
3379 
3380 	if ((error = getutimes(tptr, ts)) != 0)
3381 		return (error);
3382 
3383 	/*
3384 	 * NOTE: utimes() succeeds for the owner even if the file
3385 	 * is not user-writable.
3386 	 */
3387 	nd->nl_flags |= NLC_OWN | NLC_WRITE;
3388 
3389 	if ((error = nlookup(nd)) != 0)
3390 		return (error);
3391 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3392 		return (error);
3393 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3394 		return (error);
3395 
3396 	/*
3397 	 * note: vget is required for any operation that might mod the vnode
3398 	 * so VINACTIVE is properly cleared.
3399 	 */
3400 	if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3401 		error = vget(vp, LK_EXCLUSIVE);
3402 		if (error == 0) {
3403 			error = setutimes(vp, &vattr, ts, (tptr == NULL));
3404 			vput(vp);
3405 		}
3406 	}
3407 	vrele(vp);
3408 	return (error);
3409 }
3410 
3411 /*
3412  * utimes_args(char *path, struct timeval *tptr)
3413  *
3414  * Set the access and modification times of a file.
3415  */
3416 int
3417 sys_utimes(struct utimes_args *uap)
3418 {
3419 	struct timeval tv[2];
3420 	struct nlookupdata nd;
3421 	int error;
3422 
3423 	if (uap->tptr) {
3424  		error = copyin(uap->tptr, tv, sizeof(tv));
3425 		if (error)
3426 			return (error);
3427 	}
3428 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3429 	if (error == 0)
3430 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3431 	nlookup_done(&nd);
3432 	return (error);
3433 }
3434 
3435 /*
3436  * lutimes_args(char *path, struct timeval *tptr)
3437  *
3438  * Set the access and modification times of a file.
3439  */
3440 int
3441 sys_lutimes(struct lutimes_args *uap)
3442 {
3443 	struct timeval tv[2];
3444 	struct nlookupdata nd;
3445 	int error;
3446 
3447 	if (uap->tptr) {
3448 		error = copyin(uap->tptr, tv, sizeof(tv));
3449 		if (error)
3450 			return (error);
3451 	}
3452 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3453 	if (error == 0)
3454 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3455 	nlookup_done(&nd);
3456 	return (error);
3457 }
3458 
3459 /*
3460  * Set utimes on a file descriptor.  The creds used to open the
3461  * file are used to determine whether the operation is allowed
3462  * or not.
3463  */
3464 int
3465 kern_futimes(int fd, struct timeval *tptr)
3466 {
3467 	struct thread *td = curthread;
3468 	struct proc *p = td->td_proc;
3469 	struct timespec ts[2];
3470 	struct file *fp;
3471 	struct vnode *vp;
3472 	struct vattr vattr;
3473 	int error;
3474 
3475 	error = getutimes(tptr, ts);
3476 	if (error)
3477 		return (error);
3478 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3479 		return (error);
3480 	if (fp->f_nchandle.ncp)
3481 		error = ncp_writechk(&fp->f_nchandle);
3482 	if (error == 0) {
3483 		vp = fp->f_data;
3484 		error = vget(vp, LK_EXCLUSIVE);
3485 		if (error == 0) {
3486 			error = VOP_GETATTR(vp, &vattr);
3487 			if (error == 0) {
3488 				error = naccess_va(&vattr, NLC_OWN | NLC_WRITE,
3489 						   fp->f_cred);
3490 			}
3491 			if (error == 0) {
3492 				error = setutimes(vp, &vattr, ts,
3493 						  (tptr == NULL));
3494 			}
3495 			vput(vp);
3496 		}
3497 	}
3498 	fdrop(fp);
3499 	return (error);
3500 }
3501 
3502 /*
3503  * futimes_args(int fd, struct timeval *tptr)
3504  *
3505  * Set the access and modification times of a file.
3506  */
3507 int
3508 sys_futimes(struct futimes_args *uap)
3509 {
3510 	struct timeval tv[2];
3511 	int error;
3512 
3513 	if (uap->tptr) {
3514 		error = copyin(uap->tptr, tv, sizeof(tv));
3515 		if (error)
3516 			return (error);
3517 	}
3518 	error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
3519 
3520 	return (error);
3521 }
3522 
3523 int
3524 kern_truncate(struct nlookupdata *nd, off_t length)
3525 {
3526 	struct vnode *vp;
3527 	struct vattr vattr;
3528 	int error;
3529 	uid_t uid = 0;
3530 	gid_t gid = 0;
3531 	uint64_t old_size = 0;
3532 
3533 	if (length < 0)
3534 		return(EINVAL);
3535 	nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE;
3536 	if ((error = nlookup(nd)) != 0)
3537 		return (error);
3538 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3539 		return (error);
3540 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3541 		return (error);
3542 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
3543 		vrele(vp);
3544 		return (error);
3545 	}
3546 	if (vp->v_type == VDIR) {
3547 		error = EISDIR;
3548 		goto done;
3549 	}
3550 	if (vfs_quota_enabled) {
3551 		error = VOP_GETATTR(vp, &vattr);
3552 		KASSERT(error == 0, ("kern_truncate(): VOP_GETATTR didn't return 0"));
3553 		uid = vattr.va_uid;
3554 		gid = vattr.va_gid;
3555 		old_size = vattr.va_size;
3556 	}
3557 
3558 	if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3559 		VATTR_NULL(&vattr);
3560 		vattr.va_size = length;
3561 		error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
3562 		VFS_ACCOUNT(nd->nl_nch.mount, uid, gid, length - old_size);
3563 	}
3564 done:
3565 	vput(vp);
3566 	return (error);
3567 }
3568 
3569 /*
3570  * truncate(char *path, int pad, off_t length)
3571  *
3572  * Truncate a file given its path name.
3573  */
3574 int
3575 sys_truncate(struct truncate_args *uap)
3576 {
3577 	struct nlookupdata nd;
3578 	int error;
3579 
3580 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3581 	if (error == 0)
3582 		error = kern_truncate(&nd, uap->length);
3583 	nlookup_done(&nd);
3584 	return error;
3585 }
3586 
3587 int
3588 kern_ftruncate(int fd, off_t length)
3589 {
3590 	struct thread *td = curthread;
3591 	struct proc *p = td->td_proc;
3592 	struct vattr vattr;
3593 	struct vnode *vp;
3594 	struct file *fp;
3595 	int error;
3596 	uid_t uid = 0;
3597 	gid_t gid = 0;
3598 	uint64_t old_size = 0;
3599 	struct mount *mp;
3600 
3601 	if (length < 0)
3602 		return(EINVAL);
3603 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3604 		return (error);
3605 	if (fp->f_nchandle.ncp) {
3606 		error = ncp_writechk(&fp->f_nchandle);
3607 		if (error)
3608 			goto done;
3609 	}
3610 	if ((fp->f_flag & FWRITE) == 0) {
3611 		error = EINVAL;
3612 		goto done;
3613 	}
3614 	if (fp->f_flag & FAPPENDONLY) {	/* inode was set s/uapnd */
3615 		error = EINVAL;
3616 		goto done;
3617 	}
3618 	vp = (struct vnode *)fp->f_data;
3619 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3620 	if (vp->v_type == VDIR) {
3621 		error = EISDIR;
3622 		goto done;
3623 	}
3624 
3625 	if (vfs_quota_enabled) {
3626 		error = VOP_GETATTR(vp, &vattr);
3627 		KASSERT(error == 0, ("kern_ftruncate(): VOP_GETATTR didn't return 0"));
3628 		uid = vattr.va_uid;
3629 		gid = vattr.va_gid;
3630 		old_size = vattr.va_size;
3631 	}
3632 
3633 	if ((error = vn_writechk(vp, NULL)) == 0) {
3634 		VATTR_NULL(&vattr);
3635 		vattr.va_size = length;
3636 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
3637 		mp = vq_vptomp(vp);
3638 		VFS_ACCOUNT(mp, uid, gid, length - old_size);
3639 	}
3640 	vn_unlock(vp);
3641 done:
3642 	fdrop(fp);
3643 	return (error);
3644 }
3645 
3646 /*
3647  * ftruncate_args(int fd, int pad, off_t length)
3648  *
3649  * Truncate a file given a file descriptor.
3650  */
3651 int
3652 sys_ftruncate(struct ftruncate_args *uap)
3653 {
3654 	int error;
3655 
3656 	error = kern_ftruncate(uap->fd, uap->length);
3657 
3658 	return (error);
3659 }
3660 
3661 /*
3662  * fsync(int fd)
3663  *
3664  * Sync an open file.
3665  */
3666 int
3667 sys_fsync(struct fsync_args *uap)
3668 {
3669 	struct thread *td = curthread;
3670 	struct proc *p = td->td_proc;
3671 	struct vnode *vp;
3672 	struct file *fp;
3673 	vm_object_t obj;
3674 	int error;
3675 
3676 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3677 		return (error);
3678 	vp = (struct vnode *)fp->f_data;
3679 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3680 	if ((obj = vp->v_object) != NULL) {
3681 		if (vp->v_mount == NULL ||
3682 		    (vp->v_mount->mnt_kern_flag & MNTK_NOMSYNC) == 0) {
3683 			vm_object_page_clean(obj, 0, 0, 0);
3684 		}
3685 	}
3686 	error = VOP_FSYNC(vp, MNT_WAIT, VOP_FSYNC_SYSCALL);
3687 	if (error == 0 && vp->v_mount)
3688 		error = buf_fsync(vp);
3689 	vn_unlock(vp);
3690 	fdrop(fp);
3691 
3692 	return (error);
3693 }
3694 
3695 int
3696 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
3697 {
3698 	struct nchandle fnchd;
3699 	struct nchandle tnchd;
3700 	struct namecache *ncp;
3701 	struct vnode *fdvp;
3702 	struct vnode *tdvp;
3703 	struct mount *mp;
3704 	int error;
3705 
3706 	bwillinode(1);
3707 	fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC;
3708 	if ((error = nlookup(fromnd)) != 0)
3709 		return (error);
3710 	if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
3711 		return (ENOENT);
3712 	fnchd.mount = fromnd->nl_nch.mount;
3713 	cache_hold(&fnchd);
3714 
3715 	/*
3716 	 * unlock the source nch so we can lookup the target nch without
3717 	 * deadlocking.  The target may or may not exist so we do not check
3718 	 * for a target vp like kern_mkdir() and other creation functions do.
3719 	 *
3720 	 * The source and target directories are ref'd and rechecked after
3721 	 * everything is relocked to determine if the source or target file
3722 	 * has been renamed.
3723 	 */
3724 	KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
3725 	fromnd->nl_flags &= ~NLC_NCPISLOCKED;
3726 	cache_unlock(&fromnd->nl_nch);
3727 
3728 	tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP;
3729 	if ((error = nlookup(tond)) != 0) {
3730 		cache_drop(&fnchd);
3731 		return (error);
3732 	}
3733 	if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
3734 		cache_drop(&fnchd);
3735 		return (ENOENT);
3736 	}
3737 	tnchd.mount = tond->nl_nch.mount;
3738 	cache_hold(&tnchd);
3739 
3740 	/*
3741 	 * If the source and target are the same there is nothing to do
3742 	 */
3743 	if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
3744 		cache_drop(&fnchd);
3745 		cache_drop(&tnchd);
3746 		return (0);
3747 	}
3748 
3749 	/*
3750 	 * Mount points cannot be renamed or overwritten
3751 	 */
3752 	if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
3753 	    NCF_ISMOUNTPT
3754 	) {
3755 		cache_drop(&fnchd);
3756 		cache_drop(&tnchd);
3757 		return (EINVAL);
3758 	}
3759 
3760 	/*
3761 	 * Relock the source ncp.  cache_relock() will deal with any
3762 	 * deadlocks against the already-locked tond and will also
3763 	 * make sure both are resolved.
3764 	 *
3765 	 * NOTE AFTER RELOCKING: The source or target ncp may have become
3766 	 * invalid while they were unlocked, nc_vp and nc_mount could
3767 	 * be NULL.
3768 	 */
3769 	cache_relock(&fromnd->nl_nch, fromnd->nl_cred,
3770 		     &tond->nl_nch, tond->nl_cred);
3771 	fromnd->nl_flags |= NLC_NCPISLOCKED;
3772 
3773 	/*
3774 	 * If either fromnd or tond are marked destroyed a ripout occured
3775 	 * out from under us and we must retry.
3776 	 */
3777 	if ((fromnd->nl_nch.ncp->nc_flag & (NCF_DESTROYED | NCF_UNRESOLVED)) ||
3778 	    fromnd->nl_nch.ncp->nc_vp == NULL ||
3779 	    (tond->nl_nch.ncp->nc_flag & NCF_DESTROYED)) {
3780 		kprintf("kern_rename: retry due to ripout on: "
3781 			"\"%s\" -> \"%s\"\n",
3782 			fromnd->nl_nch.ncp->nc_name,
3783 			tond->nl_nch.ncp->nc_name);
3784 		cache_drop(&fnchd);
3785 		cache_drop(&tnchd);
3786 		return (EAGAIN);
3787 	}
3788 
3789 	/*
3790 	 * make sure the parent directories linkages are the same
3791 	 */
3792 	if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
3793 	    tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
3794 		cache_drop(&fnchd);
3795 		cache_drop(&tnchd);
3796 		return (ENOENT);
3797 	}
3798 
3799 	/*
3800 	 * Both the source and target must be within the same filesystem and
3801 	 * in the same filesystem as their parent directories within the
3802 	 * namecache topology.
3803 	 *
3804 	 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
3805 	 */
3806 	mp = fnchd.mount;
3807 	if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
3808 	    mp != tond->nl_nch.mount) {
3809 		cache_drop(&fnchd);
3810 		cache_drop(&tnchd);
3811 		return (EXDEV);
3812 	}
3813 
3814 	/*
3815 	 * Make sure the mount point is writable
3816 	 */
3817 	if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
3818 		cache_drop(&fnchd);
3819 		cache_drop(&tnchd);
3820 		return (error);
3821 	}
3822 
3823 	/*
3824 	 * If the target exists and either the source or target is a directory,
3825 	 * then both must be directories.
3826 	 *
3827 	 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
3828 	 * have become NULL.
3829 	 */
3830 	if (tond->nl_nch.ncp->nc_vp) {
3831 		if (fromnd->nl_nch.ncp->nc_vp == NULL) {
3832 			error = ENOENT;
3833 		} else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
3834 			if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
3835 				error = ENOTDIR;
3836 		} else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
3837 			error = EISDIR;
3838 		}
3839 	}
3840 
3841 	/*
3842 	 * You cannot rename a source into itself or a subdirectory of itself.
3843 	 * We check this by travsersing the target directory upwards looking
3844 	 * for a match against the source.
3845 	 *
3846 	 * XXX MPSAFE
3847 	 */
3848 	if (error == 0) {
3849 		for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
3850 			if (fromnd->nl_nch.ncp == ncp) {
3851 				error = EINVAL;
3852 				break;
3853 			}
3854 		}
3855 	}
3856 
3857 	cache_drop(&fnchd);
3858 	cache_drop(&tnchd);
3859 
3860 	/*
3861 	 * Even though the namespaces are different, they may still represent
3862 	 * hardlinks to the same file.  The filesystem might have a hard time
3863 	 * with this so we issue a NREMOVE of the source instead of a NRENAME
3864 	 * when we detect the situation.
3865 	 */
3866 	if (error == 0) {
3867 		fdvp = fromnd->nl_dvp;
3868 		tdvp = tond->nl_dvp;
3869 		if (fdvp == NULL || tdvp == NULL) {
3870 			error = EPERM;
3871 		} else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
3872 			error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
3873 					    fromnd->nl_cred);
3874 		} else {
3875 			error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
3876 					    fdvp, tdvp, tond->nl_cred);
3877 		}
3878 	}
3879 	return (error);
3880 }
3881 
3882 /*
3883  * rename_args(char *from, char *to)
3884  *
3885  * Rename files.  Source and destination must either both be directories,
3886  * or both not be directories.  If target is a directory, it must be empty.
3887  */
3888 int
3889 sys_rename(struct rename_args *uap)
3890 {
3891 	struct nlookupdata fromnd, tond;
3892 	int error;
3893 
3894 	do {
3895 		error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3896 		if (error == 0) {
3897 			error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3898 			if (error == 0)
3899 				error = kern_rename(&fromnd, &tond);
3900 			nlookup_done(&tond);
3901 		}
3902 		nlookup_done(&fromnd);
3903 	} while (error == EAGAIN);
3904 	return (error);
3905 }
3906 
3907 /*
3908  * renameat_args(int oldfd, char *old, int newfd, char *new)
3909  *
3910  * Rename files using paths relative to the directories associated with
3911  * oldfd and newfd.  Source and destination must either both be directories,
3912  * or both not be directories.  If target is a directory, it must be empty.
3913  */
3914 int
3915 sys_renameat(struct renameat_args *uap)
3916 {
3917 	struct nlookupdata oldnd, newnd;
3918 	struct file *oldfp, *newfp;
3919 	int error;
3920 
3921 	do {
3922 		error = nlookup_init_at(&oldnd, &oldfp,
3923 					uap->oldfd, uap->old,
3924 					UIO_USERSPACE, 0);
3925 		if (error == 0) {
3926 			error = nlookup_init_at(&newnd, &newfp,
3927 						uap->newfd, uap->new,
3928 						UIO_USERSPACE, 0);
3929 			if (error == 0)
3930 				error = kern_rename(&oldnd, &newnd);
3931 			nlookup_done_at(&newnd, newfp);
3932 		}
3933 		nlookup_done_at(&oldnd, oldfp);
3934 	} while (error == EAGAIN);
3935 	return (error);
3936 }
3937 
3938 int
3939 kern_mkdir(struct nlookupdata *nd, int mode)
3940 {
3941 	struct thread *td = curthread;
3942 	struct proc *p = td->td_proc;
3943 	struct vnode *vp;
3944 	struct vattr vattr;
3945 	int error;
3946 
3947 	bwillinode(1);
3948 	nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
3949 	if ((error = nlookup(nd)) != 0)
3950 		return (error);
3951 
3952 	if (nd->nl_nch.ncp->nc_vp)
3953 		return (EEXIST);
3954 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3955 		return (error);
3956 	VATTR_NULL(&vattr);
3957 	vattr.va_type = VDIR;
3958 	vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3959 
3960 	vp = NULL;
3961 	error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, td->td_ucred, &vattr);
3962 	if (error == 0)
3963 		vput(vp);
3964 	return (error);
3965 }
3966 
3967 /*
3968  * mkdir_args(char *path, int mode)
3969  *
3970  * Make a directory file.
3971  */
3972 int
3973 sys_mkdir(struct mkdir_args *uap)
3974 {
3975 	struct nlookupdata nd;
3976 	int error;
3977 
3978 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3979 	if (error == 0)
3980 		error = kern_mkdir(&nd, uap->mode);
3981 	nlookup_done(&nd);
3982 	return (error);
3983 }
3984 
3985 /*
3986  * mkdirat_args(int fd, char *path, mode_t mode)
3987  *
3988  * Make a directory file.  The path is relative to the directory associated
3989  * with fd.
3990  */
3991 int
3992 sys_mkdirat(struct mkdirat_args *uap)
3993 {
3994 	struct nlookupdata nd;
3995 	struct file *fp;
3996 	int error;
3997 
3998 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
3999 	if (error == 0)
4000 		error = kern_mkdir(&nd, uap->mode);
4001 	nlookup_done_at(&nd, fp);
4002 	return (error);
4003 }
4004 
4005 int
4006 kern_rmdir(struct nlookupdata *nd)
4007 {
4008 	int error;
4009 
4010 	bwillinode(1);
4011 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
4012 	if ((error = nlookup(nd)) != 0)
4013 		return (error);
4014 
4015 	/*
4016 	 * Do not allow directories representing mount points to be
4017 	 * deleted, even if empty.  Check write perms on mount point
4018 	 * in case the vnode is aliased (aka nullfs).
4019 	 */
4020 	if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
4021 		return (EBUSY);
4022 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
4023 		return (error);
4024 	error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
4025 	return (error);
4026 }
4027 
4028 /*
4029  * rmdir_args(char *path)
4030  *
4031  * Remove a directory file.
4032  */
4033 int
4034 sys_rmdir(struct rmdir_args *uap)
4035 {
4036 	struct nlookupdata nd;
4037 	int error;
4038 
4039 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
4040 	if (error == 0)
4041 		error = kern_rmdir(&nd);
4042 	nlookup_done(&nd);
4043 	return (error);
4044 }
4045 
4046 int
4047 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
4048 		   enum uio_seg direction)
4049 {
4050 	struct thread *td = curthread;
4051 	struct proc *p = td->td_proc;
4052 	struct vnode *vp;
4053 	struct file *fp;
4054 	struct uio auio;
4055 	struct iovec aiov;
4056 	off_t loff;
4057 	int error, eofflag;
4058 
4059 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
4060 		return (error);
4061 	if ((fp->f_flag & FREAD) == 0) {
4062 		error = EBADF;
4063 		goto done;
4064 	}
4065 	vp = (struct vnode *)fp->f_data;
4066 unionread:
4067 	if (vp->v_type != VDIR) {
4068 		error = EINVAL;
4069 		goto done;
4070 	}
4071 	aiov.iov_base = buf;
4072 	aiov.iov_len = count;
4073 	auio.uio_iov = &aiov;
4074 	auio.uio_iovcnt = 1;
4075 	auio.uio_rw = UIO_READ;
4076 	auio.uio_segflg = direction;
4077 	auio.uio_td = td;
4078 	auio.uio_resid = count;
4079 	loff = auio.uio_offset = fp->f_offset;
4080 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
4081 	fp->f_offset = auio.uio_offset;
4082 	if (error)
4083 		goto done;
4084 	if (count == auio.uio_resid) {
4085 		if (union_dircheckp) {
4086 			error = union_dircheckp(td, &vp, fp);
4087 			if (error == -1)
4088 				goto unionread;
4089 			if (error)
4090 				goto done;
4091 		}
4092 #if 0
4093 		if ((vp->v_flag & VROOT) &&
4094 		    (vp->v_mount->mnt_flag & MNT_UNION)) {
4095 			struct vnode *tvp = vp;
4096 			vp = vp->v_mount->mnt_vnodecovered;
4097 			vref(vp);
4098 			fp->f_data = vp;
4099 			fp->f_offset = 0;
4100 			vrele(tvp);
4101 			goto unionread;
4102 		}
4103 #endif
4104 	}
4105 
4106 	/*
4107 	 * WARNING!  *basep may not be wide enough to accomodate the
4108 	 * seek offset.   XXX should we hack this to return the upper 32 bits
4109 	 * for offsets greater then 4G?
4110 	 */
4111 	if (basep) {
4112 		*basep = (long)loff;
4113 	}
4114 	*res = count - auio.uio_resid;
4115 done:
4116 	fdrop(fp);
4117 	return (error);
4118 }
4119 
4120 /*
4121  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
4122  *
4123  * Read a block of directory entries in a file system independent format.
4124  */
4125 int
4126 sys_getdirentries(struct getdirentries_args *uap)
4127 {
4128 	long base;
4129 	int error;
4130 
4131 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
4132 				   &uap->sysmsg_result, UIO_USERSPACE);
4133 
4134 	if (error == 0 && uap->basep)
4135 		error = copyout(&base, uap->basep, sizeof(*uap->basep));
4136 	return (error);
4137 }
4138 
4139 /*
4140  * getdents_args(int fd, char *buf, size_t count)
4141  */
4142 int
4143 sys_getdents(struct getdents_args *uap)
4144 {
4145 	int error;
4146 
4147 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
4148 				   &uap->sysmsg_result, UIO_USERSPACE);
4149 
4150 	return (error);
4151 }
4152 
4153 /*
4154  * Set the mode mask for creation of filesystem nodes.
4155  *
4156  * umask(int newmask)
4157  */
4158 int
4159 sys_umask(struct umask_args *uap)
4160 {
4161 	struct thread *td = curthread;
4162 	struct proc *p = td->td_proc;
4163 	struct filedesc *fdp;
4164 
4165 	fdp = p->p_fd;
4166 	uap->sysmsg_result = fdp->fd_cmask;
4167 	fdp->fd_cmask = uap->newmask & ALLPERMS;
4168 	return (0);
4169 }
4170 
4171 /*
4172  * revoke(char *path)
4173  *
4174  * Void all references to file by ripping underlying filesystem
4175  * away from vnode.
4176  */
4177 int
4178 sys_revoke(struct revoke_args *uap)
4179 {
4180 	struct nlookupdata nd;
4181 	struct vattr vattr;
4182 	struct vnode *vp;
4183 	struct ucred *cred;
4184 	int error;
4185 
4186 	vp = NULL;
4187 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4188 	if (error == 0)
4189 		error = nlookup(&nd);
4190 	if (error == 0)
4191 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4192 	cred = crhold(nd.nl_cred);
4193 	nlookup_done(&nd);
4194 	if (error == 0) {
4195 		if (error == 0)
4196 			error = VOP_GETATTR(vp, &vattr);
4197 		if (error == 0 && cred->cr_uid != vattr.va_uid)
4198 			error = priv_check_cred(cred, PRIV_VFS_REVOKE, 0);
4199 		if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) {
4200 			if (vcount(vp) > 0)
4201 				error = vrevoke(vp, cred);
4202 		} else if (error == 0) {
4203 			error = vrevoke(vp, cred);
4204 		}
4205 		vrele(vp);
4206 	}
4207 	if (cred)
4208 		crfree(cred);
4209 	return (error);
4210 }
4211 
4212 /*
4213  * getfh_args(char *fname, fhandle_t *fhp)
4214  *
4215  * Get (NFS) file handle
4216  *
4217  * NOTE: We use the fsid of the covering mount, even if it is a nullfs
4218  * mount.  This allows nullfs mounts to be explicitly exported.
4219  *
4220  * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe.
4221  *
4222  * 	    nullfs mounts of subdirectories are not safe.  That is, it will
4223  *	    work, but you do not really have protection against access to
4224  *	    the related parent directories.
4225  */
4226 int
4227 sys_getfh(struct getfh_args *uap)
4228 {
4229 	struct thread *td = curthread;
4230 	struct nlookupdata nd;
4231 	fhandle_t fh;
4232 	struct vnode *vp;
4233 	struct mount *mp;
4234 	int error;
4235 
4236 	/*
4237 	 * Must be super user
4238 	 */
4239 	if ((error = priv_check(td, PRIV_ROOT)) != 0)
4240 		return (error);
4241 
4242 	vp = NULL;
4243 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
4244 	if (error == 0)
4245 		error = nlookup(&nd);
4246 	if (error == 0)
4247 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4248 	mp = nd.nl_nch.mount;
4249 	nlookup_done(&nd);
4250 	if (error == 0) {
4251 		bzero(&fh, sizeof(fh));
4252 		fh.fh_fsid = mp->mnt_stat.f_fsid;
4253 		error = VFS_VPTOFH(vp, &fh.fh_fid);
4254 		vput(vp);
4255 		if (error == 0)
4256 			error = copyout(&fh, uap->fhp, sizeof(fh));
4257 	}
4258 	return (error);
4259 }
4260 
4261 /*
4262  * fhopen_args(const struct fhandle *u_fhp, int flags)
4263  *
4264  * syscall for the rpc.lockd to use to translate a NFS file handle into
4265  * an open descriptor.
4266  *
4267  * warning: do not remove the priv_check() call or this becomes one giant
4268  * security hole.
4269  */
4270 int
4271 sys_fhopen(struct fhopen_args *uap)
4272 {
4273 	struct thread *td = curthread;
4274 	struct filedesc *fdp = td->td_proc->p_fd;
4275 	struct mount *mp;
4276 	struct vnode *vp;
4277 	struct fhandle fhp;
4278 	struct vattr vat;
4279 	struct vattr *vap = &vat;
4280 	struct flock lf;
4281 	int fmode, mode, error = 0, type;
4282 	struct file *nfp;
4283 	struct file *fp;
4284 	int indx;
4285 
4286 	/*
4287 	 * Must be super user
4288 	 */
4289 	error = priv_check(td, PRIV_ROOT);
4290 	if (error)
4291 		return (error);
4292 
4293 	fmode = FFLAGS(uap->flags);
4294 
4295 	/*
4296 	 * Why not allow a non-read/write open for our lockd?
4297 	 */
4298 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
4299 		return (EINVAL);
4300 	error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
4301 	if (error)
4302 		return(error);
4303 
4304 	/*
4305 	 * Find the mount point
4306 	 */
4307 	mp = vfs_getvfs(&fhp.fh_fsid);
4308 	if (mp == NULL) {
4309 		error = ESTALE;
4310 		goto  done;
4311 	}
4312 	/* now give me my vnode, it gets returned to me locked */
4313 	error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp);
4314 	if (error)
4315 		goto done;
4316  	/*
4317 	 * from now on we have to make sure not
4318 	 * to forget about the vnode
4319 	 * any error that causes an abort must vput(vp)
4320 	 * just set error = err and 'goto bad;'.
4321 	 */
4322 
4323 	/*
4324 	 * from vn_open
4325 	 */
4326 	if (vp->v_type == VLNK) {
4327 		error = EMLINK;
4328 		goto bad;
4329 	}
4330 	if (vp->v_type == VSOCK) {
4331 		error = EOPNOTSUPP;
4332 		goto bad;
4333 	}
4334 	mode = 0;
4335 	if (fmode & (FWRITE | O_TRUNC)) {
4336 		if (vp->v_type == VDIR) {
4337 			error = EISDIR;
4338 			goto bad;
4339 		}
4340 		error = vn_writechk(vp, NULL);
4341 		if (error)
4342 			goto bad;
4343 		mode |= VWRITE;
4344 	}
4345 	if (fmode & FREAD)
4346 		mode |= VREAD;
4347 	if (mode) {
4348 		error = VOP_ACCESS(vp, mode, td->td_ucred);
4349 		if (error)
4350 			goto bad;
4351 	}
4352 	if (fmode & O_TRUNC) {
4353 		vn_unlock(vp);				/* XXX */
4354 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
4355 		VATTR_NULL(vap);
4356 		vap->va_size = 0;
4357 		error = VOP_SETATTR(vp, vap, td->td_ucred);
4358 		if (error)
4359 			goto bad;
4360 	}
4361 
4362 	/*
4363 	 * VOP_OPEN needs the file pointer so it can potentially override
4364 	 * it.
4365 	 *
4366 	 * WARNING! no f_nchandle will be associated when fhopen()ing a
4367 	 * directory.  XXX
4368 	 */
4369 	if ((error = falloc(td->td_lwp, &nfp, &indx)) != 0)
4370 		goto bad;
4371 	fp = nfp;
4372 
4373 	error = VOP_OPEN(vp, fmode, td->td_ucred, fp);
4374 	if (error) {
4375 		/*
4376 		 * setting f_ops this way prevents VOP_CLOSE from being
4377 		 * called or fdrop() releasing the vp from v_data.   Since
4378 		 * the VOP_OPEN failed we don't want to VOP_CLOSE.
4379 		 */
4380 		fp->f_ops = &badfileops;
4381 		fp->f_data = NULL;
4382 		goto bad_drop;
4383 	}
4384 
4385 	/*
4386 	 * The fp is given its own reference, we still have our ref and lock.
4387 	 *
4388 	 * Assert that all regular files must be created with a VM object.
4389 	 */
4390 	if (vp->v_type == VREG && vp->v_object == NULL) {
4391 		kprintf("fhopen: regular file did not have VM object: %p\n", vp);
4392 		goto bad_drop;
4393 	}
4394 
4395 	/*
4396 	 * The open was successful.  Handle any locking requirements.
4397 	 */
4398 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
4399 		lf.l_whence = SEEK_SET;
4400 		lf.l_start = 0;
4401 		lf.l_len = 0;
4402 		if (fmode & O_EXLOCK)
4403 			lf.l_type = F_WRLCK;
4404 		else
4405 			lf.l_type = F_RDLCK;
4406 		if (fmode & FNONBLOCK)
4407 			type = 0;
4408 		else
4409 			type = F_WAIT;
4410 		vn_unlock(vp);
4411 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
4412 			/*
4413 			 * release our private reference.
4414 			 */
4415 			fsetfd(fdp, NULL, indx);
4416 			fdrop(fp);
4417 			vrele(vp);
4418 			goto done;
4419 		}
4420 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4421 		fp->f_flag |= FHASLOCK;
4422 	}
4423 
4424 	/*
4425 	 * Clean up.  Associate the file pointer with the previously
4426 	 * reserved descriptor and return it.
4427 	 */
4428 	vput(vp);
4429 	fsetfd(fdp, fp, indx);
4430 	fdrop(fp);
4431 	uap->sysmsg_result = indx;
4432 	if (uap->flags & O_CLOEXEC)
4433 		error = fsetfdflags(fdp, indx, UF_EXCLOSE);
4434 	return (error);
4435 
4436 bad_drop:
4437 	fsetfd(fdp, NULL, indx);
4438 	fdrop(fp);
4439 bad:
4440 	vput(vp);
4441 done:
4442 	return (error);
4443 }
4444 
4445 /*
4446  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
4447  */
4448 int
4449 sys_fhstat(struct fhstat_args *uap)
4450 {
4451 	struct thread *td = curthread;
4452 	struct stat sb;
4453 	fhandle_t fh;
4454 	struct mount *mp;
4455 	struct vnode *vp;
4456 	int error;
4457 
4458 	/*
4459 	 * Must be super user
4460 	 */
4461 	error = priv_check(td, PRIV_ROOT);
4462 	if (error)
4463 		return (error);
4464 
4465 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
4466 	if (error)
4467 		return (error);
4468 
4469 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
4470 		error = ESTALE;
4471 	if (error == 0) {
4472 		if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) == 0) {
4473 			error = vn_stat(vp, &sb, td->td_ucred);
4474 			vput(vp);
4475 		}
4476 	}
4477 	if (error == 0)
4478 		error = copyout(&sb, uap->sb, sizeof(sb));
4479 	return (error);
4480 }
4481 
4482 /*
4483  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
4484  */
4485 int
4486 sys_fhstatfs(struct fhstatfs_args *uap)
4487 {
4488 	struct thread *td = curthread;
4489 	struct proc *p = td->td_proc;
4490 	struct statfs *sp;
4491 	struct mount *mp;
4492 	struct vnode *vp;
4493 	struct statfs sb;
4494 	char *fullpath, *freepath;
4495 	fhandle_t fh;
4496 	int error;
4497 
4498 	/*
4499 	 * Must be super user
4500 	 */
4501 	if ((error = priv_check(td, PRIV_ROOT)))
4502 		return (error);
4503 
4504 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4505 		return (error);
4506 
4507 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4508 		error = ESTALE;
4509 		goto done;
4510 	}
4511 	if (p != NULL && !chroot_visible_mnt(mp, p)) {
4512 		error = ESTALE;
4513 		goto done;
4514 	}
4515 
4516 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) != 0)
4517 		goto done;
4518 	mp = vp->v_mount;
4519 	sp = &mp->mnt_stat;
4520 	vput(vp);
4521 	if ((error = VFS_STATFS(mp, sp, td->td_ucred)) != 0)
4522 		goto done;
4523 
4524 	error = mount_path(p, mp, &fullpath, &freepath);
4525 	if (error)
4526 		goto done;
4527 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4528 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
4529 	kfree(freepath, M_TEMP);
4530 
4531 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
4532 	if (priv_check(td, PRIV_ROOT)) {
4533 		bcopy(sp, &sb, sizeof(sb));
4534 		sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
4535 		sp = &sb;
4536 	}
4537 	error = copyout(sp, uap->buf, sizeof(*sp));
4538 done:
4539 	return (error);
4540 }
4541 
4542 /*
4543  * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
4544  */
4545 int
4546 sys_fhstatvfs(struct fhstatvfs_args *uap)
4547 {
4548 	struct thread *td = curthread;
4549 	struct proc *p = td->td_proc;
4550 	struct statvfs *sp;
4551 	struct mount *mp;
4552 	struct vnode *vp;
4553 	fhandle_t fh;
4554 	int error;
4555 
4556 	/*
4557 	 * Must be super user
4558 	 */
4559 	if ((error = priv_check(td, PRIV_ROOT)))
4560 		return (error);
4561 
4562 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4563 		return (error);
4564 
4565 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4566 		error = ESTALE;
4567 		goto done;
4568 	}
4569 	if (p != NULL && !chroot_visible_mnt(mp, p)) {
4570 		error = ESTALE;
4571 		goto done;
4572 	}
4573 
4574 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
4575 		goto done;
4576 	mp = vp->v_mount;
4577 	sp = &mp->mnt_vstat;
4578 	vput(vp);
4579 	if ((error = VFS_STATVFS(mp, sp, td->td_ucred)) != 0)
4580 		goto done;
4581 
4582 	sp->f_flag = 0;
4583 	if (mp->mnt_flag & MNT_RDONLY)
4584 		sp->f_flag |= ST_RDONLY;
4585 	if (mp->mnt_flag & MNT_NOSUID)
4586 		sp->f_flag |= ST_NOSUID;
4587 	error = copyout(sp, uap->buf, sizeof(*sp));
4588 done:
4589 	return (error);
4590 }
4591 
4592 
4593 /*
4594  * Syscall to push extended attribute configuration information into the
4595  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
4596  * a command (int cmd), and attribute name and misc data.  For now, the
4597  * attribute name is left in userspace for consumption by the VFS_op.
4598  * It will probably be changed to be copied into sysspace by the
4599  * syscall in the future, once issues with various consumers of the
4600  * attribute code have raised their hands.
4601  *
4602  * Currently this is used only by UFS Extended Attributes.
4603  */
4604 int
4605 sys_extattrctl(struct extattrctl_args *uap)
4606 {
4607 	struct nlookupdata nd;
4608 	struct vnode *vp;
4609 	char attrname[EXTATTR_MAXNAMELEN];
4610 	int error;
4611 	size_t size;
4612 
4613 	attrname[0] = 0;
4614 	vp = NULL;
4615 	error = 0;
4616 
4617 	if (error == 0 && uap->filename) {
4618 		error = nlookup_init(&nd, uap->filename, UIO_USERSPACE,
4619 				     NLC_FOLLOW);
4620 		if (error == 0)
4621 			error = nlookup(&nd);
4622 		if (error == 0)
4623 			error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4624 		nlookup_done(&nd);
4625 	}
4626 
4627 	if (error == 0 && uap->attrname) {
4628 		error = copyinstr(uap->attrname, attrname, EXTATTR_MAXNAMELEN,
4629 				  &size);
4630 	}
4631 
4632 	if (error == 0) {
4633 		error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4634 		if (error == 0)
4635 			error = nlookup(&nd);
4636 		if (error == 0)
4637 			error = ncp_writechk(&nd.nl_nch);
4638 		if (error == 0) {
4639 			error = VFS_EXTATTRCTL(nd.nl_nch.mount, uap->cmd, vp,
4640 					       uap->attrnamespace,
4641 					       uap->attrname, nd.nl_cred);
4642 		}
4643 		nlookup_done(&nd);
4644 	}
4645 
4646 	return (error);
4647 }
4648 
4649 /*
4650  * Syscall to get a named extended attribute on a file or directory.
4651  */
4652 int
4653 sys_extattr_set_file(struct extattr_set_file_args *uap)
4654 {
4655 	char attrname[EXTATTR_MAXNAMELEN];
4656 	struct nlookupdata nd;
4657 	struct vnode *vp;
4658 	struct uio auio;
4659 	struct iovec aiov;
4660 	int error;
4661 
4662 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4663 	if (error)
4664 		return (error);
4665 
4666 	vp = NULL;
4667 
4668 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4669 	if (error == 0)
4670 		error = nlookup(&nd);
4671 	if (error == 0)
4672 		error = ncp_writechk(&nd.nl_nch);
4673 	if (error == 0)
4674 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4675 	if (error) {
4676 		nlookup_done(&nd);
4677 		return (error);
4678 	}
4679 
4680 	bzero(&auio, sizeof(auio));
4681 	aiov.iov_base = uap->data;
4682 	aiov.iov_len = uap->nbytes;
4683 	auio.uio_iov = &aiov;
4684 	auio.uio_iovcnt = 1;
4685 	auio.uio_offset = 0;
4686 	auio.uio_resid = uap->nbytes;
4687 	auio.uio_rw = UIO_WRITE;
4688 	auio.uio_td = curthread;
4689 
4690 	error = VOP_SETEXTATTR(vp, uap->attrnamespace, attrname,
4691 			       &auio, nd.nl_cred);
4692 
4693 	vput(vp);
4694 	nlookup_done(&nd);
4695 	return (error);
4696 }
4697 
4698 /*
4699  * Syscall to get a named extended attribute on a file or directory.
4700  */
4701 int
4702 sys_extattr_get_file(struct extattr_get_file_args *uap)
4703 {
4704 	char attrname[EXTATTR_MAXNAMELEN];
4705 	struct nlookupdata nd;
4706 	struct uio auio;
4707 	struct iovec aiov;
4708 	struct vnode *vp;
4709 	int error;
4710 
4711 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4712 	if (error)
4713 		return (error);
4714 
4715 	vp = NULL;
4716 
4717 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4718 	if (error == 0)
4719 		error = nlookup(&nd);
4720 	if (error == 0)
4721 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4722 	if (error) {
4723 		nlookup_done(&nd);
4724 		return (error);
4725 	}
4726 
4727 	bzero(&auio, sizeof(auio));
4728 	aiov.iov_base = uap->data;
4729 	aiov.iov_len = uap->nbytes;
4730 	auio.uio_iov = &aiov;
4731 	auio.uio_iovcnt = 1;
4732 	auio.uio_offset = 0;
4733 	auio.uio_resid = uap->nbytes;
4734 	auio.uio_rw = UIO_READ;
4735 	auio.uio_td = curthread;
4736 
4737 	error = VOP_GETEXTATTR(vp, uap->attrnamespace, attrname,
4738 				&auio, nd.nl_cred);
4739 	uap->sysmsg_result = uap->nbytes - auio.uio_resid;
4740 
4741 	vput(vp);
4742 	nlookup_done(&nd);
4743 	return(error);
4744 }
4745 
4746 /*
4747  * Syscall to delete a named extended attribute from a file or directory.
4748  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
4749  */
4750 int
4751 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
4752 {
4753 	char attrname[EXTATTR_MAXNAMELEN];
4754 	struct nlookupdata nd;
4755 	struct vnode *vp;
4756 	int error;
4757 
4758 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4759 	if (error)
4760 		return(error);
4761 
4762 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4763 	if (error == 0)
4764 		error = nlookup(&nd);
4765 	if (error == 0)
4766 		error = ncp_writechk(&nd.nl_nch);
4767 	if (error == 0) {
4768 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4769 		if (error == 0) {
4770 			error = VOP_SETEXTATTR(vp, uap->attrnamespace,
4771 					       attrname, NULL, nd.nl_cred);
4772 			vput(vp);
4773 		}
4774 	}
4775 	nlookup_done(&nd);
4776 	return(error);
4777 }
4778 
4779 /*
4780  * Determine if the mount is visible to the process.
4781  */
4782 static int
4783 chroot_visible_mnt(struct mount *mp, struct proc *p)
4784 {
4785 	struct nchandle nch;
4786 
4787 	/*
4788 	 * Traverse from the mount point upwards.  If we hit the process
4789 	 * root then the mount point is visible to the process.
4790 	 */
4791 	nch = mp->mnt_ncmountpt;
4792 	while (nch.ncp) {
4793 		if (nch.mount == p->p_fd->fd_nrdir.mount &&
4794 		    nch.ncp == p->p_fd->fd_nrdir.ncp) {
4795 			return(1);
4796 		}
4797 		if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
4798 			nch = nch.mount->mnt_ncmounton;
4799 		} else {
4800 			nch.ncp = nch.ncp->nc_parent;
4801 		}
4802 	}
4803 
4804 	/*
4805 	 * If the mount point is not visible to the process, but the
4806 	 * process root is in a subdirectory of the mount, return
4807 	 * TRUE anyway.
4808 	 */
4809 	if (p->p_fd->fd_nrdir.mount == mp)
4810 		return(1);
4811 
4812 	return(0);
4813 }
4814 
4815