xref: /dflybsd-src/sys/kern/vfs_syscalls.c (revision bfc09ba0a4d805c1860f88e64d6ae9a407d3567d)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_syscalls.c	8.13 (Berkeley) 4/15/94
39  * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $
40  * $DragonFly: src/sys/kern/vfs_syscalls.c,v 1.135 2008/11/11 00:55:49 pavalos Exp $
41  */
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/buf.h>
46 #include <sys/conf.h>
47 #include <sys/sysent.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mountctl.h>
51 #include <sys/sysproto.h>
52 #include <sys/filedesc.h>
53 #include <sys/kernel.h>
54 #include <sys/fcntl.h>
55 #include <sys/file.h>
56 #include <sys/linker.h>
57 #include <sys/stat.h>
58 #include <sys/unistd.h>
59 #include <sys/vnode.h>
60 #include <sys/proc.h>
61 #include <sys/priv.h>
62 #include <sys/jail.h>
63 #include <sys/namei.h>
64 #include <sys/nlookup.h>
65 #include <sys/dirent.h>
66 #include <sys/extattr.h>
67 #include <sys/spinlock.h>
68 #include <sys/kern_syscall.h>
69 #include <sys/objcache.h>
70 #include <sys/sysctl.h>
71 
72 #include <sys/buf2.h>
73 #include <sys/file2.h>
74 #include <sys/spinlock2.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 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb);
87 static int checkvp_chdir (struct vnode *vn, struct thread *td);
88 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch);
89 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
90 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
91 static int getutimes (const struct timeval *, struct timespec *);
92 static int setfown (struct vnode *, uid_t, gid_t);
93 static int setfmode (struct vnode *, int);
94 static int setfflags (struct vnode *, int);
95 static int setutimes (struct vnode *, struct vattr *,
96 			const struct timespec *, int);
97 static int	usermount = 0;	/* if 1, non-root can mount fs. */
98 
99 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *);
100 
101 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "");
102 
103 /*
104  * Virtual File System System Calls
105  */
106 
107 /*
108  * Mount a file system.
109  */
110 /*
111  * mount_args(char *type, char *path, int flags, caddr_t data)
112  */
113 /* ARGSUSED */
114 int
115 sys_mount(struct mount_args *uap)
116 {
117 	struct thread *td = curthread;
118 	struct proc *p = td->td_proc;
119 	struct vnode *vp;
120 	struct nchandle nch;
121 	struct mount *mp;
122 	struct vfsconf *vfsp;
123 	int error, flag = 0, flag2 = 0;
124 	int hasmount;
125 	struct vattr va;
126 	struct nlookupdata nd;
127 	char fstypename[MFSNAMELEN];
128 	struct ucred *cred = p->p_ucred;
129 
130 	KKASSERT(p);
131 	if (jailed(cred))
132 		return (EPERM);
133 	if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
134 		return (error);
135 	/*
136 	 * Do not allow NFS export by non-root users.
137 	 */
138 	if (uap->flags & MNT_EXPORTED) {
139 		error = priv_check(td, PRIV_ROOT);
140 		if (error)
141 			return (error);
142 	}
143 	/*
144 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
145 	 */
146 	if (priv_check(td, PRIV_ROOT))
147 		uap->flags |= MNT_NOSUID | MNT_NODEV;
148 
149 	/*
150 	 * Lookup the requested path and extract the nch and vnode.
151 	 */
152 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
153 	if (error == 0) {
154 		if ((error = nlookup(&nd)) == 0) {
155 			if (nd.nl_nch.ncp->nc_vp == NULL)
156 				error = ENOENT;
157 		}
158 	}
159 	if (error) {
160 		nlookup_done(&nd);
161 		return (error);
162 	}
163 
164 	/*
165 	 * Extract the locked+refd ncp and cleanup the nd structure
166 	 */
167 	nch = nd.nl_nch;
168 	cache_zero(&nd.nl_nch);
169 	nlookup_done(&nd);
170 
171 	if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) && cache_findmount(&nch))
172 		hasmount = 1;
173 	else
174 		hasmount = 0;
175 
176 
177 	/*
178 	 * now we have the locked ref'd nch and unreferenced vnode.
179 	 */
180 	vp = nch.ncp->nc_vp;
181 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
182 		cache_put(&nch);
183 		return (error);
184 	}
185 	cache_unlock(&nch);
186 
187 	/*
188 	 * Now we have an unlocked ref'd nch and a locked ref'd vp
189 	 */
190 	if (uap->flags & MNT_UPDATE) {
191 		if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
192 			cache_drop(&nch);
193 			vput(vp);
194 			return (EINVAL);
195 		}
196 		mp = vp->v_mount;
197 		flag = mp->mnt_flag;
198 		flag2 = mp->mnt_kern_flag;
199 		/*
200 		 * We only allow the filesystem to be reloaded if it
201 		 * is currently mounted read-only.
202 		 */
203 		if ((uap->flags & MNT_RELOAD) &&
204 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
205 			cache_drop(&nch);
206 			vput(vp);
207 			return (EOPNOTSUPP);	/* Needs translation */
208 		}
209 		/*
210 		 * Only root, or the user that did the original mount is
211 		 * permitted to update it.
212 		 */
213 		if (mp->mnt_stat.f_owner != cred->cr_uid &&
214 		    (error = priv_check(td, PRIV_ROOT))) {
215 			cache_drop(&nch);
216 			vput(vp);
217 			return (error);
218 		}
219 		if (vfs_busy(mp, LK_NOWAIT)) {
220 			cache_drop(&nch);
221 			vput(vp);
222 			return (EBUSY);
223 		}
224 		if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
225 			cache_drop(&nch);
226 			vfs_unbusy(mp);
227 			vput(vp);
228 			return (EBUSY);
229 		}
230 		vp->v_flag |= VMOUNT;
231 		mp->mnt_flag |=
232 		    uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
233 		vn_unlock(vp);
234 		goto update;
235 	}
236 	/*
237 	 * If the user is not root, ensure that they own the directory
238 	 * onto which we are attempting to mount.
239 	 */
240 	if ((error = VOP_GETATTR(vp, &va)) ||
241 	    (va.va_uid != cred->cr_uid && (error = priv_check(td, PRIV_ROOT)))) {
242 		cache_drop(&nch);
243 		vput(vp);
244 		return (error);
245 	}
246 	if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
247 		cache_drop(&nch);
248 		vput(vp);
249 		return (error);
250 	}
251 	if (vp->v_type != VDIR) {
252 		cache_drop(&nch);
253 		vput(vp);
254 		return (ENOTDIR);
255 	}
256 	if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) {
257 		cache_drop(&nch);
258 		vput(vp);
259 		return (EPERM);
260 	}
261 	if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
262 		cache_drop(&nch);
263 		vput(vp);
264 		return (error);
265 	}
266 	vfsp = vfsconf_find_by_name(fstypename);
267 	if (vfsp == NULL) {
268 		linker_file_t lf;
269 
270 		/* Only load modules for root (very important!) */
271 		if ((error = priv_check(td, PRIV_ROOT)) != 0) {
272 			cache_drop(&nch);
273 			vput(vp);
274 			return error;
275 		}
276 		error = linker_load_file(fstypename, &lf, 0);
277 		if (error || lf == NULL) {
278 			cache_drop(&nch);
279 			vput(vp);
280 			if (lf == NULL)
281 				error = ENODEV;
282 			return error;
283 		}
284 		lf->userrefs++;
285 		/* lookup again, see if the VFS was loaded */
286 		vfsp = vfsconf_find_by_name(fstypename);
287 		if (vfsp == NULL) {
288 			lf->userrefs--;
289 			linker_file_unload(lf);
290 			cache_drop(&nch);
291 			vput(vp);
292 			return (ENODEV);
293 		}
294 	}
295 	if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
296 		cache_drop(&nch);
297 		vput(vp);
298 		return (EBUSY);
299 	}
300 	vp->v_flag |= VMOUNT;
301 
302 	/*
303 	 * Allocate and initialize the filesystem.
304 	 */
305 	mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
306 	TAILQ_INIT(&mp->mnt_nvnodelist);
307 	TAILQ_INIT(&mp->mnt_reservedvnlist);
308 	TAILQ_INIT(&mp->mnt_jlist);
309 	mp->mnt_nvnodelistsize = 0;
310 	lockinit(&mp->mnt_lock, "vfslock", 0, 0);
311 	vfs_busy(mp, LK_NOWAIT);
312 	mp->mnt_op = vfsp->vfc_vfsops;
313 	mp->mnt_vfc = vfsp;
314 	vfsp->vfc_refcount++;
315 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
316 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
317 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
318 	mp->mnt_stat.f_owner = cred->cr_uid;
319 	mp->mnt_iosize_max = DFLTPHYS;
320 	vn_unlock(vp);
321 update:
322 	/*
323 	 * Set the mount level flags.
324 	 */
325 	if (uap->flags & MNT_RDONLY)
326 		mp->mnt_flag |= MNT_RDONLY;
327 	else if (mp->mnt_flag & MNT_RDONLY)
328 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
329 	mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
330 	    MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
331 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
332 	    MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
333 	mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
334 	    MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
335 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
336 	    MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
337 	/*
338 	 * Mount the filesystem.
339 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
340 	 * get.
341 	 */
342 	error = VFS_MOUNT(mp, uap->path, uap->data, cred);
343 	if (mp->mnt_flag & MNT_UPDATE) {
344 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
345 			mp->mnt_flag &= ~MNT_RDONLY;
346 		mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
347 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
348 		if (error) {
349 			mp->mnt_flag = flag;
350 			mp->mnt_kern_flag = flag2;
351 		}
352 		vfs_unbusy(mp);
353 		vp->v_flag &= ~VMOUNT;
354 		vrele(vp);
355 		cache_drop(&nch);
356 		return (error);
357 	}
358 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
359 	/*
360 	 * Put the new filesystem on the mount list after root.  The mount
361 	 * point gets its own mnt_ncmountpt (unless the VFS already set one
362 	 * up) which represents the root of the mount.  The lookup code
363 	 * detects the mount point going forward and checks the root of
364 	 * the mount going backwards.
365 	 *
366 	 * It is not necessary to invalidate or purge the vnode underneath
367 	 * because elements under the mount will be given their own glue
368 	 * namecache record.
369 	 */
370 	if (!error) {
371 		if (mp->mnt_ncmountpt.ncp == NULL) {
372 			/*
373 			 * allocate, then unlock, but leave the ref intact
374 			 */
375 			cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
376 			cache_unlock(&mp->mnt_ncmountpt);
377 		}
378 		mp->mnt_ncmounton = nch;		/* inherits ref */
379 		nch.ncp->nc_flag |= NCF_ISMOUNTPT;
380 
381 		/* XXX get the root of the fs and cache_setvp(mnt_ncmountpt...) */
382 		vp->v_flag &= ~VMOUNT;
383 		mountlist_insert(mp, MNTINS_LAST);
384 		vn_unlock(vp);
385 		checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
386 		error = vfs_allocate_syncvnode(mp);
387 		vfs_unbusy(mp);
388 		error = VFS_START(mp, 0);
389 		vrele(vp);
390 	} else {
391 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
392 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
393 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
394 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
395 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
396 		vp->v_flag &= ~VMOUNT;
397 		mp->mnt_vfc->vfc_refcount--;
398 		vfs_unbusy(mp);
399 		kfree(mp, M_MOUNT);
400 		cache_drop(&nch);
401 		vput(vp);
402 	}
403 	return (error);
404 }
405 
406 /*
407  * Scan all active processes to see if any of them have a current
408  * or root directory onto which the new filesystem has just been
409  * mounted. If so, replace them with the new mount point.
410  *
411  * The passed ncp is ref'd and locked (from the mount code) and
412  * must be associated with the vnode representing the root of the
413  * mount point.
414  */
415 struct checkdirs_info {
416 	struct nchandle old_nch;
417 	struct nchandle new_nch;
418 	struct vnode *old_vp;
419 	struct vnode *new_vp;
420 };
421 
422 static int checkdirs_callback(struct proc *p, void *data);
423 
424 static void
425 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
426 {
427 	struct checkdirs_info info;
428 	struct vnode *olddp;
429 	struct vnode *newdp;
430 	struct mount *mp;
431 
432 	/*
433 	 * If the old mount point's vnode has a usecount of 1, it is not
434 	 * being held as a descriptor anywhere.
435 	 */
436 	olddp = old_nch->ncp->nc_vp;
437 	if (olddp == NULL || olddp->v_sysref.refcnt == 1)
438 		return;
439 
440 	/*
441 	 * Force the root vnode of the new mount point to be resolved
442 	 * so we can update any matching processes.
443 	 */
444 	mp = new_nch->mount;
445 	if (VFS_ROOT(mp, &newdp))
446 		panic("mount: lost mount");
447 	cache_setunresolved(new_nch);
448 	cache_setvp(new_nch, newdp);
449 
450 	/*
451 	 * Special handling of the root node
452 	 */
453 	if (rootvnode == olddp) {
454 		vref(newdp);
455 		vfs_cache_setroot(newdp, cache_hold(new_nch));
456 	}
457 
458 	/*
459 	 * Pass newdp separately so the callback does not have to access
460 	 * it via new_nch->ncp->nc_vp.
461 	 */
462 	info.old_nch = *old_nch;
463 	info.new_nch = *new_nch;
464 	info.new_vp = newdp;
465 	allproc_scan(checkdirs_callback, &info);
466 	vput(newdp);
467 }
468 
469 /*
470  * NOTE: callback is not MP safe because the scanned process's filedesc
471  * structure can be ripped out from under us, amoung other things.
472  */
473 static int
474 checkdirs_callback(struct proc *p, void *data)
475 {
476 	struct checkdirs_info *info = data;
477 	struct filedesc *fdp;
478 	struct nchandle ncdrop1;
479 	struct nchandle ncdrop2;
480 	struct vnode *vprele1;
481 	struct vnode *vprele2;
482 
483 	if ((fdp = p->p_fd) != NULL) {
484 		cache_zero(&ncdrop1);
485 		cache_zero(&ncdrop2);
486 		vprele1 = NULL;
487 		vprele2 = NULL;
488 
489 		/*
490 		 * MPUNSAFE - XXX fdp can be pulled out from under a
491 		 * foreign process.
492 		 *
493 		 * A shared filedesc is ok, we don't have to copy it
494 		 * because we are making this change globally.
495 		 */
496 		spin_lock_wr(&fdp->fd_spin);
497 		if (fdp->fd_ncdir.mount == info->old_nch.mount &&
498 		    fdp->fd_ncdir.ncp == info->old_nch.ncp) {
499 			vprele1 = fdp->fd_cdir;
500 			vref(info->new_vp);
501 			fdp->fd_cdir = info->new_vp;
502 			ncdrop1 = fdp->fd_ncdir;
503 			cache_copy(&info->new_nch, &fdp->fd_ncdir);
504 		}
505 		if (fdp->fd_nrdir.mount == info->old_nch.mount &&
506 		    fdp->fd_nrdir.ncp == info->old_nch.ncp) {
507 			vprele2 = fdp->fd_rdir;
508 			vref(info->new_vp);
509 			fdp->fd_rdir = info->new_vp;
510 			ncdrop2 = fdp->fd_nrdir;
511 			cache_copy(&info->new_nch, &fdp->fd_nrdir);
512 		}
513 		spin_unlock_wr(&fdp->fd_spin);
514 		if (ncdrop1.ncp)
515 			cache_drop(&ncdrop1);
516 		if (ncdrop2.ncp)
517 			cache_drop(&ncdrop2);
518 		if (vprele1)
519 			vrele(vprele1);
520 		if (vprele2)
521 			vrele(vprele2);
522 	}
523 	return(0);
524 }
525 
526 /*
527  * Unmount a file system.
528  *
529  * Note: unmount takes a path to the vnode mounted on as argument,
530  * not special file (as before).
531  */
532 /*
533  * umount_args(char *path, int flags)
534  */
535 /* ARGSUSED */
536 int
537 sys_unmount(struct unmount_args *uap)
538 {
539 	struct thread *td = curthread;
540 	struct proc *p = td->td_proc;
541 	struct mount *mp = NULL;
542 	int error;
543 	struct nlookupdata nd;
544 
545 	KKASSERT(p);
546 	if (p->p_ucred->cr_prison != NULL)
547 		return (EPERM);
548 	if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
549 		return (error);
550 
551 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
552 	if (error == 0)
553 		error = nlookup(&nd);
554 	if (error)
555 		goto out;
556 
557 	mp = nd.nl_nch.mount;
558 
559 	/*
560 	 * Only root, or the user that did the original mount is
561 	 * permitted to unmount this filesystem.
562 	 */
563 	if ((mp->mnt_stat.f_owner != p->p_ucred->cr_uid) &&
564 	    (error = priv_check(td, PRIV_ROOT)))
565 		goto out;
566 
567 	/*
568 	 * Don't allow unmounting the root file system.
569 	 */
570 	if (mp->mnt_flag & MNT_ROOTFS) {
571 		error = EINVAL;
572 		goto out;
573 	}
574 
575 	/*
576 	 * Must be the root of the filesystem
577 	 */
578 	if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
579 		error = EINVAL;
580 		goto out;
581 	}
582 
583 out:
584 	nlookup_done(&nd);
585 	if (error)
586 		return (error);
587 	return (dounmount(mp, uap->flags));
588 }
589 
590 /*
591  * Do the actual file system unmount.
592  */
593 static int
594 dounmount_interlock(struct mount *mp)
595 {
596 	if (mp->mnt_kern_flag & MNTK_UNMOUNT)
597 		return (EBUSY);
598 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
599 	return(0);
600 }
601 
602 int
603 dounmount(struct mount *mp, int flags)
604 {
605 	struct namecache *ncp;
606 	struct nchandle nch;
607 	struct vnode *vp;
608 	int error;
609 	int async_flag;
610 	int lflags;
611 	int freeok = 1;
612 
613 	/*
614 	 * Exclusive access for unmounting purposes
615 	 */
616 	if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
617 		return (error);
618 
619 	/*
620 	 * Allow filesystems to detect that a forced unmount is in progress.
621 	 */
622 	if (flags & MNT_FORCE)
623 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
624 	lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
625 	error = lockmgr(&mp->mnt_lock, lflags);
626 	if (error) {
627 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
628 		if (mp->mnt_kern_flag & MNTK_MWAIT)
629 			wakeup(mp);
630 		return (error);
631 	}
632 
633 	if (mp->mnt_flag & MNT_EXPUBLIC)
634 		vfs_setpublicfs(NULL, NULL, NULL);
635 
636 	vfs_msync(mp, MNT_WAIT);
637 	async_flag = mp->mnt_flag & MNT_ASYNC;
638 	mp->mnt_flag &=~ MNT_ASYNC;
639 
640 	/*
641 	 * If this filesystem isn't aliasing other filesystems,
642 	 * try to invalidate any remaining namecache entries and
643 	 * check the count afterwords.
644 	 */
645 	if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
646 		cache_lock(&mp->mnt_ncmountpt);
647 		cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
648 		cache_unlock(&mp->mnt_ncmountpt);
649 
650 		if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
651 		    (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
652 
653 			if ((flags & MNT_FORCE) == 0) {
654 				error = EBUSY;
655 				mount_warning(mp, "Cannot unmount: "
656 						  "%d namecache "
657 						  "references still "
658 						  "present",
659 						  ncp->nc_refs - 1);
660 			} else {
661 				mount_warning(mp, "Forced unmount: "
662 						  "%d namecache "
663 						  "references still "
664 						  "present",
665 						  ncp->nc_refs - 1);
666 				freeok = 0;
667 			}
668 		}
669 	}
670 
671 	/*
672 	 * nchandle records ref the mount structure.  Expect a count of 1
673 	 * (our mount->mnt_ncmountpt).
674 	 */
675 	if (mp->mnt_refs != 1) {
676 		if ((flags & MNT_FORCE) == 0) {
677 			mount_warning(mp, "Cannot unmount: "
678 					  "%d process references still "
679 					  "present", mp->mnt_refs);
680 			error = EBUSY;
681 		} else {
682 			mount_warning(mp, "Forced unmount: "
683 					  "%d process references still "
684 					  "present", mp->mnt_refs);
685 			freeok = 0;
686 		}
687 	}
688 
689 	/*
690 	 * Decomission our special mnt_syncer vnode.  This also stops
691 	 * the vnlru code.  If we are unable to unmount we recommission
692 	 * the vnode.
693 	 */
694 	if (error == 0) {
695 		if ((vp = mp->mnt_syncer) != NULL) {
696 			mp->mnt_syncer = NULL;
697 			vrele(vp);
698 		}
699 		if (((mp->mnt_flag & MNT_RDONLY) ||
700 		     (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
701 		    (flags & MNT_FORCE)) {
702 			error = VFS_UNMOUNT(mp, flags);
703 		}
704 	}
705 	if (error) {
706 		if (mp->mnt_syncer == NULL)
707 			vfs_allocate_syncvnode(mp);
708 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
709 		mp->mnt_flag |= async_flag;
710 		lockmgr(&mp->mnt_lock, LK_RELEASE);
711 		if (mp->mnt_kern_flag & MNTK_MWAIT)
712 			wakeup(mp);
713 		return (error);
714 	}
715 	/*
716 	 * Clean up any journals still associated with the mount after
717 	 * filesystem activity has ceased.
718 	 */
719 	journal_remove_all_journals(mp,
720 	    ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
721 
722 	mountlist_remove(mp);
723 
724 	/*
725 	 * Remove any installed vnode ops here so the individual VFSs don't
726 	 * have to.
727 	 */
728 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
729 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
730 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
731 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
732 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
733 
734 	if (mp->mnt_ncmountpt.ncp != NULL) {
735 		nch = mp->mnt_ncmountpt;
736 		cache_zero(&mp->mnt_ncmountpt);
737 		cache_clrmountpt(&nch);
738 		cache_drop(&nch);
739 	}
740 	if (mp->mnt_ncmounton.ncp != NULL) {
741 		nch = mp->mnt_ncmounton;
742 		cache_zero(&mp->mnt_ncmounton);
743 		cache_clrmountpt(&nch);
744 		cache_drop(&nch);
745 	}
746 
747 	mp->mnt_vfc->vfc_refcount--;
748 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
749 		panic("unmount: dangling vnode");
750 	lockmgr(&mp->mnt_lock, LK_RELEASE);
751 	if (mp->mnt_kern_flag & MNTK_MWAIT)
752 		wakeup(mp);
753 	if (freeok)
754 		kfree(mp, M_MOUNT);
755 	return (0);
756 }
757 
758 static
759 void
760 mount_warning(struct mount *mp, const char *ctl, ...)
761 {
762 	char *ptr;
763 	char *buf;
764 	__va_list va;
765 
766 	__va_start(va, ctl);
767 	if (cache_fullpath(NULL, &mp->mnt_ncmounton, &ptr, &buf) == 0) {
768 		kprintf("unmount(%s): ", ptr);
769 		kvprintf(ctl, va);
770 		kprintf("\n");
771 		kfree(buf, M_TEMP);
772 	} else {
773 		kprintf("unmount(%p", mp);
774 		if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
775 			kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
776 		kprintf("): ");
777 		kvprintf(ctl, va);
778 		kprintf("\n");
779 	}
780 	__va_end(va);
781 }
782 
783 /*
784  * Shim cache_fullpath() to handle the case where a process is chrooted into
785  * a subdirectory of a mount.  In this case if the root mount matches the
786  * process root directory's mount we have to specify the process's root
787  * directory instead of the mount point, because the mount point might
788  * be above the root directory.
789  */
790 static
791 int
792 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
793 {
794 	struct nchandle *nch;
795 
796 	if (p && p->p_fd->fd_nrdir.mount == mp)
797 		nch = &p->p_fd->fd_nrdir;
798 	else
799 		nch = &mp->mnt_ncmountpt;
800 	return(cache_fullpath(p, nch, rb, fb));
801 }
802 
803 /*
804  * Sync each mounted filesystem.
805  */
806 
807 #ifdef DEBUG
808 static int syncprt = 0;
809 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
810 #endif /* DEBUG */
811 
812 static int sync_callback(struct mount *mp, void *data);
813 
814 /* ARGSUSED */
815 int
816 sys_sync(struct sync_args *uap)
817 {
818 	mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
819 #ifdef DEBUG
820 	/*
821 	 * print out buffer pool stat information on each sync() call.
822 	 */
823 	if (syncprt)
824 		vfs_bufstats();
825 #endif /* DEBUG */
826 	return (0);
827 }
828 
829 static
830 int
831 sync_callback(struct mount *mp, void *data __unused)
832 {
833 	int asyncflag;
834 
835 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
836 		asyncflag = mp->mnt_flag & MNT_ASYNC;
837 		mp->mnt_flag &= ~MNT_ASYNC;
838 		vfs_msync(mp, MNT_NOWAIT);
839 		VFS_SYNC(mp, MNT_NOWAIT);
840 		mp->mnt_flag |= asyncflag;
841 	}
842 	return(0);
843 }
844 
845 /* XXX PRISON: could be per prison flag */
846 static int prison_quotas;
847 #if 0
848 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
849 #endif
850 
851 /*
852  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
853  *
854  * Change filesystem quotas.
855  */
856 /* ARGSUSED */
857 int
858 sys_quotactl(struct quotactl_args *uap)
859 {
860 	struct nlookupdata nd;
861 	struct thread *td;
862 	struct proc *p;
863 	struct mount *mp;
864 	int error;
865 
866 	td = curthread;
867 	p = td->td_proc;
868 	if (p->p_ucred->cr_prison && !prison_quotas)
869 		return (EPERM);
870 
871 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
872 	if (error == 0)
873 		error = nlookup(&nd);
874 	if (error == 0) {
875 		mp = nd.nl_nch.mount;
876 		error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
877 				    uap->arg, nd.nl_cred);
878 	}
879 	nlookup_done(&nd);
880 	return (error);
881 }
882 
883 /*
884  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
885  *		void *buf, int buflen)
886  *
887  * This function operates on a mount point and executes the specified
888  * operation using the specified control data, and possibly returns data.
889  *
890  * The actual number of bytes stored in the result buffer is returned, 0
891  * if none, otherwise an error is returned.
892  */
893 /* ARGSUSED */
894 int
895 sys_mountctl(struct mountctl_args *uap)
896 {
897 	struct thread *td = curthread;
898 	struct proc *p = td->td_proc;
899 	struct file *fp;
900 	void *ctl = NULL;
901 	void *buf = NULL;
902 	char *path = NULL;
903 	int error;
904 
905 	/*
906 	 * Sanity and permissions checks.  We must be root.
907 	 */
908 	KKASSERT(p);
909 	if (p->p_ucred->cr_prison != NULL)
910 		return (EPERM);
911 	if ((uap->op != MOUNTCTL_MOUNTFLAGS) &&
912 	    (error = priv_check(td, PRIV_ROOT)) != 0)
913 		return (error);
914 
915 	/*
916 	 * Argument length checks
917 	 */
918 	if (uap->ctllen < 0 || uap->ctllen > 1024)
919 		return (EINVAL);
920 	if (uap->buflen < 0 || uap->buflen > 16 * 1024)
921 		return (EINVAL);
922 	if (uap->path == NULL)
923 		return (EINVAL);
924 
925 	/*
926 	 * Allocate the necessary buffers and copyin data
927 	 */
928 	path = objcache_get(namei_oc, M_WAITOK);
929 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
930 	if (error)
931 		goto done;
932 
933 	if (uap->ctllen) {
934 		ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
935 		error = copyin(uap->ctl, ctl, uap->ctllen);
936 		if (error)
937 			goto done;
938 	}
939 	if (uap->buflen)
940 		buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
941 
942 	/*
943 	 * Validate the descriptor
944 	 */
945 	if (uap->fd >= 0) {
946 		fp = holdfp(p->p_fd, uap->fd, -1);
947 		if (fp == NULL) {
948 			error = EBADF;
949 			goto done;
950 		}
951 	} else {
952 		fp = NULL;
953 	}
954 
955 	/*
956 	 * Execute the internal kernel function and clean up.
957 	 */
958 	error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
959 	if (fp)
960 		fdrop(fp);
961 	if (error == 0 && uap->sysmsg_result > 0)
962 		error = copyout(buf, uap->buf, uap->sysmsg_result);
963 done:
964 	if (path)
965 		objcache_put(namei_oc, path);
966 	if (ctl)
967 		kfree(ctl, M_TEMP);
968 	if (buf)
969 		kfree(buf, M_TEMP);
970 	return (error);
971 }
972 
973 /*
974  * Execute a mount control operation by resolving the path to a mount point
975  * and calling vop_mountctl().
976  *
977  * Use the mount point from the nch instead of the vnode so nullfs mounts
978  * can properly spike the VOP.
979  */
980 int
981 kern_mountctl(const char *path, int op, struct file *fp,
982 		const void *ctl, int ctllen,
983 		void *buf, int buflen, int *res)
984 {
985 	struct vnode *vp;
986 	struct mount *mp;
987 	struct nlookupdata nd;
988 	int error;
989 
990 	*res = 0;
991 	vp = NULL;
992 	error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
993 	if (error == 0)
994 		error = nlookup(&nd);
995 	if (error == 0)
996 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
997 	mp = nd.nl_nch.mount;
998 	nlookup_done(&nd);
999 	if (error)
1000 		return (error);
1001 
1002 	/*
1003 	 * Must be the root of the filesystem
1004 	 */
1005 	if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
1006 		vput(vp);
1007 		return (EINVAL);
1008 	}
1009 	error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen,
1010 				buf, buflen, res);
1011 	vput(vp);
1012 	return (error);
1013 }
1014 
1015 int
1016 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1017 {
1018 	struct thread *td = curthread;
1019 	struct proc *p = td->td_proc;
1020 	struct mount *mp;
1021 	struct statfs *sp;
1022 	char *fullpath, *freepath;
1023 	int error;
1024 
1025 	if ((error = nlookup(nd)) != 0)
1026 		return (error);
1027 	mp = nd->nl_nch.mount;
1028 	sp = &mp->mnt_stat;
1029 	if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1030 		return (error);
1031 
1032 	error = mount_path(p, mp, &fullpath, &freepath);
1033 	if (error)
1034 		return(error);
1035 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1036 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1037 	kfree(freepath, M_TEMP);
1038 
1039 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1040 	bcopy(sp, buf, sizeof(*buf));
1041 	/* Only root should have access to the fsid's. */
1042 	if (priv_check(td, PRIV_ROOT))
1043 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1044 	return (0);
1045 }
1046 
1047 /*
1048  * statfs_args(char *path, struct statfs *buf)
1049  *
1050  * Get filesystem statistics.
1051  */
1052 int
1053 sys_statfs(struct statfs_args *uap)
1054 {
1055 	struct nlookupdata nd;
1056 	struct statfs buf;
1057 	int error;
1058 
1059 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1060 	if (error == 0)
1061 		error = kern_statfs(&nd, &buf);
1062 	nlookup_done(&nd);
1063 	if (error == 0)
1064 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1065 	return (error);
1066 }
1067 
1068 int
1069 kern_fstatfs(int fd, struct statfs *buf)
1070 {
1071 	struct thread *td = curthread;
1072 	struct proc *p = td->td_proc;
1073 	struct file *fp;
1074 	struct mount *mp;
1075 	struct statfs *sp;
1076 	char *fullpath, *freepath;
1077 	int error;
1078 
1079 	KKASSERT(p);
1080 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1081 		return (error);
1082 	mp = ((struct vnode *)fp->f_data)->v_mount;
1083 	if (mp == NULL) {
1084 		error = EBADF;
1085 		goto done;
1086 	}
1087 	if (fp->f_cred == NULL) {
1088 		error = EINVAL;
1089 		goto done;
1090 	}
1091 	sp = &mp->mnt_stat;
1092 	if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1093 		goto done;
1094 
1095 	if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1096 		goto done;
1097 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1098 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1099 	kfree(freepath, M_TEMP);
1100 
1101 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1102 	bcopy(sp, buf, sizeof(*buf));
1103 
1104 	/* Only root should have access to the fsid's. */
1105 	if (priv_check(td, PRIV_ROOT))
1106 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1107 	error = 0;
1108 done:
1109 	fdrop(fp);
1110 	return (error);
1111 }
1112 
1113 /*
1114  * fstatfs_args(int fd, struct statfs *buf)
1115  *
1116  * Get filesystem statistics.
1117  */
1118 int
1119 sys_fstatfs(struct fstatfs_args *uap)
1120 {
1121 	struct statfs buf;
1122 	int error;
1123 
1124 	error = kern_fstatfs(uap->fd, &buf);
1125 
1126 	if (error == 0)
1127 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1128 	return (error);
1129 }
1130 
1131 int
1132 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1133 {
1134 	struct mount *mp;
1135 	struct statvfs *sp;
1136 	int error;
1137 
1138 	if ((error = nlookup(nd)) != 0)
1139 		return (error);
1140 	mp = nd->nl_nch.mount;
1141 	sp = &mp->mnt_vstat;
1142 	if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1143 		return (error);
1144 
1145 	sp->f_flag = 0;
1146 	if (mp->mnt_flag & MNT_RDONLY)
1147 		sp->f_flag |= ST_RDONLY;
1148 	if (mp->mnt_flag & MNT_NOSUID)
1149 		sp->f_flag |= ST_NOSUID;
1150 	bcopy(sp, buf, sizeof(*buf));
1151 	return (0);
1152 }
1153 
1154 /*
1155  * statfs_args(char *path, struct statfs *buf)
1156  *
1157  * Get filesystem statistics.
1158  */
1159 int
1160 sys_statvfs(struct statvfs_args *uap)
1161 {
1162 	struct nlookupdata nd;
1163 	struct statvfs buf;
1164 	int error;
1165 
1166 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1167 	if (error == 0)
1168 		error = kern_statvfs(&nd, &buf);
1169 	nlookup_done(&nd);
1170 	if (error == 0)
1171 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1172 	return (error);
1173 }
1174 
1175 int
1176 kern_fstatvfs(int fd, struct statvfs *buf)
1177 {
1178 	struct thread *td = curthread;
1179 	struct proc *p = td->td_proc;
1180 	struct file *fp;
1181 	struct mount *mp;
1182 	struct statvfs *sp;
1183 	int error;
1184 
1185 	KKASSERT(p);
1186 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1187 		return (error);
1188 	mp = ((struct vnode *)fp->f_data)->v_mount;
1189 	if (mp == NULL) {
1190 		error = EBADF;
1191 		goto done;
1192 	}
1193 	if (fp->f_cred == NULL) {
1194 		error = EINVAL;
1195 		goto done;
1196 	}
1197 	sp = &mp->mnt_vstat;
1198 	if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1199 		goto done;
1200 
1201 	sp->f_flag = 0;
1202 	if (mp->mnt_flag & MNT_RDONLY)
1203 		sp->f_flag |= ST_RDONLY;
1204 	if (mp->mnt_flag & MNT_NOSUID)
1205 		sp->f_flag |= ST_NOSUID;
1206 
1207 	bcopy(sp, buf, sizeof(*buf));
1208 	error = 0;
1209 done:
1210 	fdrop(fp);
1211 	return (error);
1212 }
1213 
1214 /*
1215  * fstatfs_args(int fd, struct statfs *buf)
1216  *
1217  * Get filesystem statistics.
1218  */
1219 int
1220 sys_fstatvfs(struct fstatvfs_args *uap)
1221 {
1222 	struct statvfs buf;
1223 	int error;
1224 
1225 	error = kern_fstatvfs(uap->fd, &buf);
1226 
1227 	if (error == 0)
1228 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1229 	return (error);
1230 }
1231 
1232 /*
1233  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1234  *
1235  * Get statistics on all filesystems.
1236  */
1237 
1238 struct getfsstat_info {
1239 	struct statfs *sfsp;
1240 	long count;
1241 	long maxcount;
1242 	int error;
1243 	int flags;
1244 	struct proc *p;
1245 };
1246 
1247 static int getfsstat_callback(struct mount *, void *);
1248 
1249 /* ARGSUSED */
1250 int
1251 sys_getfsstat(struct getfsstat_args *uap)
1252 {
1253 	struct thread *td = curthread;
1254 	struct proc *p = td->td_proc;
1255 	struct getfsstat_info info;
1256 
1257 	bzero(&info, sizeof(info));
1258 
1259 	info.maxcount = uap->bufsize / sizeof(struct statfs);
1260 	info.sfsp = uap->buf;
1261 	info.count = 0;
1262 	info.flags = uap->flags;
1263 	info.p = p;
1264 
1265 	mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1266 	if (info.sfsp && info.count > info.maxcount)
1267 		uap->sysmsg_result = info.maxcount;
1268 	else
1269 		uap->sysmsg_result = info.count;
1270 	return (info.error);
1271 }
1272 
1273 static int
1274 getfsstat_callback(struct mount *mp, void *data)
1275 {
1276 	struct getfsstat_info *info = data;
1277 	struct statfs *sp;
1278 	char *freepath;
1279 	char *fullpath;
1280 	int error;
1281 
1282 	if (info->sfsp && info->count < info->maxcount) {
1283 		if (info->p && !chroot_visible_mnt(mp, info->p))
1284 			return(0);
1285 		sp = &mp->mnt_stat;
1286 
1287 		/*
1288 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1289 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1290 		 * overrides MNT_WAIT.
1291 		 */
1292 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1293 		    (info->flags & MNT_WAIT)) &&
1294 		    (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1295 			return(0);
1296 		}
1297 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1298 
1299 		error = mount_path(info->p, mp, &fullpath, &freepath);
1300 		if (error) {
1301 			info->error = error;
1302 			return(-1);
1303 		}
1304 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1305 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1306 		kfree(freepath, M_TEMP);
1307 
1308 		error = copyout(sp, info->sfsp, sizeof(*sp));
1309 		if (error) {
1310 			info->error = error;
1311 			return (-1);
1312 		}
1313 		++info->sfsp;
1314 	}
1315 	info->count++;
1316 	return(0);
1317 }
1318 
1319 /*
1320  * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1321 		   long bufsize, int flags)
1322  *
1323  * Get statistics on all filesystems.
1324  */
1325 
1326 struct getvfsstat_info {
1327 	struct statfs *sfsp;
1328 	struct statvfs *vsfsp;
1329 	long count;
1330 	long maxcount;
1331 	int error;
1332 	int flags;
1333 	struct proc *p;
1334 };
1335 
1336 static int getvfsstat_callback(struct mount *, void *);
1337 
1338 /* ARGSUSED */
1339 int
1340 sys_getvfsstat(struct getvfsstat_args *uap)
1341 {
1342 	struct thread *td = curthread;
1343 	struct proc *p = td->td_proc;
1344 	struct getvfsstat_info info;
1345 
1346 	bzero(&info, sizeof(info));
1347 
1348 	info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1349 	info.sfsp = uap->buf;
1350 	info.vsfsp = uap->vbuf;
1351 	info.count = 0;
1352 	info.flags = uap->flags;
1353 	info.p = p;
1354 
1355 	mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1356 	if (info.vsfsp && info.count > info.maxcount)
1357 		uap->sysmsg_result = info.maxcount;
1358 	else
1359 		uap->sysmsg_result = info.count;
1360 	return (info.error);
1361 }
1362 
1363 static int
1364 getvfsstat_callback(struct mount *mp, void *data)
1365 {
1366 	struct getvfsstat_info *info = data;
1367 	struct statfs *sp;
1368 	struct statvfs *vsp;
1369 	char *freepath;
1370 	char *fullpath;
1371 	int error;
1372 
1373 	if (info->vsfsp && info->count < info->maxcount) {
1374 		if (info->p && !chroot_visible_mnt(mp, info->p))
1375 			return(0);
1376 		sp = &mp->mnt_stat;
1377 		vsp = &mp->mnt_vstat;
1378 
1379 		/*
1380 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1381 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1382 		 * overrides MNT_WAIT.
1383 		 */
1384 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1385 		    (info->flags & MNT_WAIT)) &&
1386 		    (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1387 			return(0);
1388 		}
1389 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1390 
1391 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1392 		    (info->flags & MNT_WAIT)) &&
1393 		    (error = VFS_STATVFS(mp, vsp, info->p->p_ucred))) {
1394 			return(0);
1395 		}
1396 		vsp->f_flag = 0;
1397 		if (mp->mnt_flag & MNT_RDONLY)
1398 			vsp->f_flag |= ST_RDONLY;
1399 		if (mp->mnt_flag & MNT_NOSUID)
1400 			vsp->f_flag |= ST_NOSUID;
1401 
1402 		error = mount_path(info->p, mp, &fullpath, &freepath);
1403 		if (error) {
1404 			info->error = error;
1405 			return(-1);
1406 		}
1407 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1408 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1409 		kfree(freepath, M_TEMP);
1410 
1411 		error = copyout(sp, info->sfsp, sizeof(*sp));
1412 		if (error == 0)
1413 			error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1414 		if (error) {
1415 			info->error = error;
1416 			return (-1);
1417 		}
1418 		++info->sfsp;
1419 		++info->vsfsp;
1420 	}
1421 	info->count++;
1422 	return(0);
1423 }
1424 
1425 
1426 /*
1427  * fchdir_args(int fd)
1428  *
1429  * Change current working directory to a given file descriptor.
1430  */
1431 /* ARGSUSED */
1432 int
1433 sys_fchdir(struct fchdir_args *uap)
1434 {
1435 	struct thread *td = curthread;
1436 	struct proc *p = td->td_proc;
1437 	struct filedesc *fdp = p->p_fd;
1438 	struct vnode *vp, *ovp;
1439 	struct mount *mp;
1440 	struct file *fp;
1441 	struct nchandle nch, onch, tnch;
1442 	int error;
1443 
1444 	if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1445 		return (error);
1446 	vp = (struct vnode *)fp->f_data;
1447 	vref(vp);
1448 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1449 	if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL)
1450 		error = ENOTDIR;
1451 	else
1452 		error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
1453 	if (error) {
1454 		vput(vp);
1455 		fdrop(fp);
1456 		return (error);
1457 	}
1458 	cache_copy(&fp->f_nchandle, &nch);
1459 
1460 	/*
1461 	 * If the ncp has become a mount point, traverse through
1462 	 * the mount point.
1463 	 */
1464 
1465 	while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1466 	       (mp = cache_findmount(&nch)) != NULL
1467 	) {
1468 		error = nlookup_mp(mp, &tnch);
1469 		if (error == 0) {
1470 			cache_unlock(&tnch);	/* leave ref intact */
1471 			vput(vp);
1472 			vp = tnch.ncp->nc_vp;
1473 			error = vget(vp, LK_SHARED);
1474 			KKASSERT(error == 0);
1475 			cache_drop(&nch);
1476 			nch = tnch;
1477 		}
1478 	}
1479 	if (error == 0) {
1480 		ovp = fdp->fd_cdir;
1481 		onch = fdp->fd_ncdir;
1482 		vn_unlock(vp);		/* leave ref intact */
1483 		fdp->fd_cdir = vp;
1484 		fdp->fd_ncdir = nch;
1485 		cache_drop(&onch);
1486 		vrele(ovp);
1487 	} else {
1488 		cache_drop(&nch);
1489 		vput(vp);
1490 	}
1491 	fdrop(fp);
1492 	return (error);
1493 }
1494 
1495 int
1496 kern_chdir(struct nlookupdata *nd)
1497 {
1498 	struct thread *td = curthread;
1499 	struct proc *p = td->td_proc;
1500 	struct filedesc *fdp = p->p_fd;
1501 	struct vnode *vp, *ovp;
1502 	struct nchandle onch;
1503 	int error;
1504 
1505 	if ((error = nlookup(nd)) != 0)
1506 		return (error);
1507 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1508 		return (ENOENT);
1509 	if ((error = vget(vp, LK_SHARED)) != 0)
1510 		return (error);
1511 
1512 	error = checkvp_chdir(vp, td);
1513 	vn_unlock(vp);
1514 	if (error == 0) {
1515 		ovp = fdp->fd_cdir;
1516 		onch = fdp->fd_ncdir;
1517 		cache_unlock(&nd->nl_nch);	/* leave reference intact */
1518 		fdp->fd_ncdir = nd->nl_nch;
1519 		fdp->fd_cdir = vp;
1520 		cache_drop(&onch);
1521 		vrele(ovp);
1522 		cache_zero(&nd->nl_nch);
1523 	} else {
1524 		vrele(vp);
1525 	}
1526 	return (error);
1527 }
1528 
1529 /*
1530  * chdir_args(char *path)
1531  *
1532  * Change current working directory (``.'').
1533  */
1534 int
1535 sys_chdir(struct chdir_args *uap)
1536 {
1537 	struct nlookupdata nd;
1538 	int error;
1539 
1540 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1541 	if (error == 0)
1542 		error = kern_chdir(&nd);
1543 	nlookup_done(&nd);
1544 	return (error);
1545 }
1546 
1547 /*
1548  * Helper function for raised chroot(2) security function:  Refuse if
1549  * any filedescriptors are open directories.
1550  */
1551 static int
1552 chroot_refuse_vdir_fds(struct filedesc *fdp)
1553 {
1554 	struct vnode *vp;
1555 	struct file *fp;
1556 	int error;
1557 	int fd;
1558 
1559 	for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1560 		if ((error = holdvnode(fdp, fd, &fp)) != 0)
1561 			continue;
1562 		vp = (struct vnode *)fp->f_data;
1563 		if (vp->v_type != VDIR) {
1564 			fdrop(fp);
1565 			continue;
1566 		}
1567 		fdrop(fp);
1568 		return(EPERM);
1569 	}
1570 	return (0);
1571 }
1572 
1573 /*
1574  * This sysctl determines if we will allow a process to chroot(2) if it
1575  * has a directory open:
1576  *	0: disallowed for all processes.
1577  *	1: allowed for processes that were not already chroot(2)'ed.
1578  *	2: allowed for all processes.
1579  */
1580 
1581 static int chroot_allow_open_directories = 1;
1582 
1583 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1584      &chroot_allow_open_directories, 0, "");
1585 
1586 /*
1587  * chroot to the specified namecache entry.  We obtain the vp from the
1588  * namecache data.  The passed ncp must be locked and referenced and will
1589  * remain locked and referenced on return.
1590  */
1591 int
1592 kern_chroot(struct nchandle *nch)
1593 {
1594 	struct thread *td = curthread;
1595 	struct proc *p = td->td_proc;
1596 	struct filedesc *fdp = p->p_fd;
1597 	struct vnode *vp;
1598 	int error;
1599 
1600 	/*
1601 	 * Only privileged user can chroot
1602 	 */
1603 	error = priv_check_cred(p->p_ucred, PRIV_VFS_CHROOT, 0);
1604 	if (error)
1605 		return (error);
1606 
1607 	/*
1608 	 * Disallow open directory descriptors (fchdir() breakouts).
1609 	 */
1610 	if (chroot_allow_open_directories == 0 ||
1611 	   (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1612 		if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1613 			return (error);
1614 	}
1615 	if ((vp = nch->ncp->nc_vp) == NULL)
1616 		return (ENOENT);
1617 
1618 	if ((error = vget(vp, LK_SHARED)) != 0)
1619 		return (error);
1620 
1621 	/*
1622 	 * Check the validity of vp as a directory to change to and
1623 	 * associate it with rdir/jdir.
1624 	 */
1625 	error = checkvp_chdir(vp, td);
1626 	vn_unlock(vp);			/* leave reference intact */
1627 	if (error == 0) {
1628 		vrele(fdp->fd_rdir);
1629 		fdp->fd_rdir = vp;	/* reference inherited by fd_rdir */
1630 		cache_drop(&fdp->fd_nrdir);
1631 		cache_copy(nch, &fdp->fd_nrdir);
1632 		if (fdp->fd_jdir == NULL) {
1633 			fdp->fd_jdir = vp;
1634 			vref(fdp->fd_jdir);
1635 			cache_copy(nch, &fdp->fd_njdir);
1636 		}
1637 	} else {
1638 		vrele(vp);
1639 	}
1640 	return (error);
1641 }
1642 
1643 /*
1644  * chroot_args(char *path)
1645  *
1646  * Change notion of root (``/'') directory.
1647  */
1648 /* ARGSUSED */
1649 int
1650 sys_chroot(struct chroot_args *uap)
1651 {
1652 	struct thread *td = curthread;
1653 	struct nlookupdata nd;
1654 	int error;
1655 
1656 	KKASSERT(td->td_proc);
1657 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1658 	if (error) {
1659 		nlookup_done(&nd);
1660 		return(error);
1661 	}
1662 	nd.nl_flags |= NLC_EXEC;
1663 	error = nlookup(&nd);
1664 	if (error == 0)
1665 		error = kern_chroot(&nd.nl_nch);
1666 	nlookup_done(&nd);
1667 	return(error);
1668 }
1669 
1670 /*
1671  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1672  * determine whether it is legal to chdir to the vnode.  The vnode's state
1673  * is not changed by this call.
1674  */
1675 int
1676 checkvp_chdir(struct vnode *vp, struct thread *td)
1677 {
1678 	int error;
1679 
1680 	if (vp->v_type != VDIR)
1681 		error = ENOTDIR;
1682 	else
1683 		error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred);
1684 	return (error);
1685 }
1686 
1687 int
1688 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1689 {
1690 	struct thread *td = curthread;
1691 	struct proc *p = td->td_proc;
1692 	struct lwp *lp = td->td_lwp;
1693 	struct filedesc *fdp = p->p_fd;
1694 	int cmode, flags;
1695 	struct file *nfp;
1696 	struct file *fp;
1697 	struct vnode *vp;
1698 	int type, indx, error;
1699 	struct flock lf;
1700 
1701 	if ((oflags & O_ACCMODE) == O_ACCMODE)
1702 		return (EINVAL);
1703 	flags = FFLAGS(oflags);
1704 	error = falloc(p, &nfp, NULL);
1705 	if (error)
1706 		return (error);
1707 	fp = nfp;
1708 	cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
1709 
1710 	/*
1711 	 * XXX p_dupfd is a real mess.  It allows a device to return a
1712 	 * file descriptor to be duplicated rather then doing the open
1713 	 * itself.
1714 	 */
1715 	lp->lwp_dupfd = -1;
1716 
1717 	/*
1718 	 * Call vn_open() to do the lookup and assign the vnode to the
1719 	 * file pointer.  vn_open() does not change the ref count on fp
1720 	 * and the vnode, on success, will be inherited by the file pointer
1721 	 * and unlocked.
1722 	 */
1723 	nd->nl_flags |= NLC_LOCKVP;
1724 	error = vn_open(nd, fp, flags, cmode);
1725 	nlookup_done(nd);
1726 	if (error) {
1727 		/*
1728 		 * handle special fdopen() case.  bleh.  dupfdopen() is
1729 		 * responsible for dropping the old contents of ofiles[indx]
1730 		 * if it succeeds.
1731 		 *
1732 		 * Note that fsetfd() will add a ref to fp which represents
1733 		 * the fd_files[] assignment.  We must still drop our
1734 		 * reference.
1735 		 */
1736 		if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1737 			if (fdalloc(p, 0, &indx) == 0) {
1738 				error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error);
1739 				if (error == 0) {
1740 					*res = indx;
1741 					fdrop(fp);	/* our ref */
1742 					return (0);
1743 				}
1744 				fsetfd(p, NULL, indx);
1745 			}
1746 		}
1747 		fdrop(fp);	/* our ref */
1748 		if (error == ERESTART)
1749 			error = EINTR;
1750 		return (error);
1751 	}
1752 
1753 	/*
1754 	 * ref the vnode for ourselves so it can't be ripped out from under
1755 	 * is.  XXX need an ND flag to request that the vnode be returned
1756 	 * anyway.
1757 	 *
1758 	 * Reserve a file descriptor but do not assign it until the open
1759 	 * succeeds.
1760 	 */
1761 	vp = (struct vnode *)fp->f_data;
1762 	vref(vp);
1763 	if ((error = fdalloc(p, 0, &indx)) != 0) {
1764 		fdrop(fp);
1765 		vrele(vp);
1766 		return (error);
1767 	}
1768 
1769 	/*
1770 	 * If no error occurs the vp will have been assigned to the file
1771 	 * pointer.
1772 	 */
1773 	lp->lwp_dupfd = 0;
1774 
1775 	if (flags & (O_EXLOCK | O_SHLOCK)) {
1776 		lf.l_whence = SEEK_SET;
1777 		lf.l_start = 0;
1778 		lf.l_len = 0;
1779 		if (flags & O_EXLOCK)
1780 			lf.l_type = F_WRLCK;
1781 		else
1782 			lf.l_type = F_RDLCK;
1783 		if (flags & FNONBLOCK)
1784 			type = 0;
1785 		else
1786 			type = F_WAIT;
1787 
1788 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1789 			/*
1790 			 * lock request failed.  Clean up the reserved
1791 			 * descriptor.
1792 			 */
1793 			vrele(vp);
1794 			fsetfd(p, NULL, indx);
1795 			fdrop(fp);
1796 			return (error);
1797 		}
1798 		fp->f_flag |= FHASLOCK;
1799 	}
1800 #if 0
1801 	/*
1802 	 * Assert that all regular file vnodes were created with a object.
1803 	 */
1804 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1805 		("open: regular file has no backing object after vn_open"));
1806 #endif
1807 
1808 	vrele(vp);
1809 
1810 	/*
1811 	 * release our private reference, leaving the one associated with the
1812 	 * descriptor table intact.
1813 	 */
1814 	fsetfd(p, fp, indx);
1815 	fdrop(fp);
1816 	*res = indx;
1817 	return (0);
1818 }
1819 
1820 /*
1821  * open_args(char *path, int flags, int mode)
1822  *
1823  * Check permissions, allocate an open file structure,
1824  * and call the device open routine if any.
1825  */
1826 int
1827 sys_open(struct open_args *uap)
1828 {
1829 	struct nlookupdata nd;
1830 	int error;
1831 
1832 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1833 	if (error == 0) {
1834 		error = kern_open(&nd, uap->flags,
1835 				    uap->mode, &uap->sysmsg_result);
1836 	}
1837 	nlookup_done(&nd);
1838 	return (error);
1839 }
1840 
1841 /*
1842  * openat_args(int fd, char *path, int flags, int mode)
1843  */
1844 int
1845 sys_openat(struct openat_args *uap)
1846 {
1847 	struct nlookupdata nd;
1848 	int error;
1849 	struct file *fp;
1850 
1851 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
1852 	if (error == 0) {
1853 		error = kern_open(&nd, uap->flags, uap->mode,
1854 					&uap->sysmsg_result);
1855 	}
1856 	nlookup_done_at(&nd, fp);
1857 	return (error);
1858 }
1859 
1860 int
1861 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
1862 {
1863 	struct thread *td = curthread;
1864 	struct proc *p = td->td_proc;
1865 	struct vnode *vp;
1866 	struct vattr vattr;
1867 	int error;
1868 	int whiteout = 0;
1869 
1870 	KKASSERT(p);
1871 
1872 	VATTR_NULL(&vattr);
1873 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1874 	vattr.va_rmajor = rmajor;
1875 	vattr.va_rminor = rminor;
1876 
1877 	switch (mode & S_IFMT) {
1878 	case S_IFMT:	/* used by badsect to flag bad sectors */
1879 		error = priv_check_cred(p->p_ucred, PRIV_VFS_MKNOD_BAD, 0);
1880 		vattr.va_type = VBAD;
1881 		break;
1882 	case S_IFCHR:
1883 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
1884 		vattr.va_type = VCHR;
1885 		break;
1886 	case S_IFBLK:
1887 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
1888 		vattr.va_type = VBLK;
1889 		break;
1890 	case S_IFWHT:
1891 		error = priv_check_cred(p->p_ucred, PRIV_VFS_MKNOD_WHT, 0);
1892 		whiteout = 1;
1893 		break;
1894 	case S_IFDIR:	/* special directories support for HAMMER */
1895 		error = priv_check_cred(p->p_ucred, PRIV_VFS_MKNOD_DIR, 0);
1896 		vattr.va_type = VDIR;
1897 		break;
1898 	default:
1899 		error = EINVAL;
1900 		break;
1901 	}
1902 
1903 	if (error)
1904 		return (error);
1905 
1906 	bwillinode(1);
1907 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1908 	if ((error = nlookup(nd)) != 0)
1909 		return (error);
1910 	if (nd->nl_nch.ncp->nc_vp)
1911 		return (EEXIST);
1912 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1913 		return (error);
1914 
1915 	if (whiteout) {
1916 		error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
1917 				      nd->nl_cred, NAMEI_CREATE);
1918 	} else {
1919 		vp = NULL;
1920 		error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
1921 				   &vp, nd->nl_cred, &vattr);
1922 		if (error == 0)
1923 			vput(vp);
1924 	}
1925 	return (error);
1926 }
1927 
1928 /*
1929  * mknod_args(char *path, int mode, int dev)
1930  *
1931  * Create a special file.
1932  */
1933 int
1934 sys_mknod(struct mknod_args *uap)
1935 {
1936 	struct nlookupdata nd;
1937 	int error;
1938 
1939 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1940 	if (error == 0) {
1941 		error = kern_mknod(&nd, uap->mode,
1942 				   umajor(uap->dev), uminor(uap->dev));
1943 	}
1944 	nlookup_done(&nd);
1945 	return (error);
1946 }
1947 
1948 int
1949 kern_mkfifo(struct nlookupdata *nd, int mode)
1950 {
1951 	struct thread *td = curthread;
1952 	struct proc *p = td->td_proc;
1953 	struct vattr vattr;
1954 	struct vnode *vp;
1955 	int error;
1956 
1957 	bwillinode(1);
1958 
1959 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1960 	if ((error = nlookup(nd)) != 0)
1961 		return (error);
1962 	if (nd->nl_nch.ncp->nc_vp)
1963 		return (EEXIST);
1964 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1965 		return (error);
1966 
1967 	VATTR_NULL(&vattr);
1968 	vattr.va_type = VFIFO;
1969 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1970 	vp = NULL;
1971 	error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
1972 	if (error == 0)
1973 		vput(vp);
1974 	return (error);
1975 }
1976 
1977 /*
1978  * mkfifo_args(char *path, int mode)
1979  *
1980  * Create a named pipe.
1981  */
1982 int
1983 sys_mkfifo(struct mkfifo_args *uap)
1984 {
1985 	struct nlookupdata nd;
1986 	int error;
1987 
1988 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1989 	if (error == 0)
1990 		error = kern_mkfifo(&nd, uap->mode);
1991 	nlookup_done(&nd);
1992 	return (error);
1993 }
1994 
1995 static int hardlink_check_uid = 0;
1996 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1997     &hardlink_check_uid, 0,
1998     "Unprivileged processes cannot create hard links to files owned by other "
1999     "users");
2000 static int hardlink_check_gid = 0;
2001 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
2002     &hardlink_check_gid, 0,
2003     "Unprivileged processes cannot create hard links to files owned by other "
2004     "groups");
2005 
2006 static int
2007 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
2008 {
2009 	struct vattr va;
2010 	int error;
2011 
2012 	/*
2013 	 * Shortcut if disabled
2014 	 */
2015 	if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
2016 		return (0);
2017 
2018 	/*
2019 	 * Privileged user can always hardlink
2020 	 */
2021 	if (priv_check_cred(cred, PRIV_VFS_LINK, 0) == 0)
2022 		return (0);
2023 
2024 	/*
2025 	 * Otherwise only if the originating file is owned by the
2026 	 * same user or group.  Note that any group is allowed if
2027 	 * the file is owned by the caller.
2028 	 */
2029 	error = VOP_GETATTR(vp, &va);
2030 	if (error != 0)
2031 		return (error);
2032 
2033 	if (hardlink_check_uid) {
2034 		if (cred->cr_uid != va.va_uid)
2035 			return (EPERM);
2036 	}
2037 
2038 	if (hardlink_check_gid) {
2039 		if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2040 			return (EPERM);
2041 	}
2042 
2043 	return (0);
2044 }
2045 
2046 int
2047 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2048 {
2049 	struct thread *td = curthread;
2050 	struct vnode *vp;
2051 	int error;
2052 
2053 	/*
2054 	 * Lookup the source and obtained a locked vnode.
2055 	 *
2056 	 * You may only hardlink a file which you have write permission
2057 	 * on or which you own.
2058 	 *
2059 	 * XXX relookup on vget failure / race ?
2060 	 */
2061 	bwillinode(1);
2062 	nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK;
2063 	if ((error = nlookup(nd)) != 0)
2064 		return (error);
2065 	vp = nd->nl_nch.ncp->nc_vp;
2066 	KKASSERT(vp != NULL);
2067 	if (vp->v_type == VDIR)
2068 		return (EPERM);		/* POSIX */
2069 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2070 		return (error);
2071 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2072 		return (error);
2073 
2074 	/*
2075 	 * Unlock the source so we can lookup the target without deadlocking
2076 	 * (XXX vp is locked already, possible other deadlock?).  The target
2077 	 * must not exist.
2078 	 */
2079 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2080 	nd->nl_flags &= ~NLC_NCPISLOCKED;
2081 	cache_unlock(&nd->nl_nch);
2082 
2083 	linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2084 	if ((error = nlookup(linknd)) != 0) {
2085 		vput(vp);
2086 		return (error);
2087 	}
2088 	if (linknd->nl_nch.ncp->nc_vp) {
2089 		vput(vp);
2090 		return (EEXIST);
2091 	}
2092 
2093 	/*
2094 	 * Finally run the new API VOP.
2095 	 */
2096 	error = can_hardlink(vp, td, td->td_proc->p_ucred);
2097 	if (error == 0) {
2098 		error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2099 				  vp, linknd->nl_cred);
2100 	}
2101 	vput(vp);
2102 	return (error);
2103 }
2104 
2105 /*
2106  * link_args(char *path, char *link)
2107  *
2108  * Make a hard file link.
2109  */
2110 int
2111 sys_link(struct link_args *uap)
2112 {
2113 	struct nlookupdata nd, linknd;
2114 	int error;
2115 
2116 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2117 	if (error == 0) {
2118 		error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2119 		if (error == 0)
2120 			error = kern_link(&nd, &linknd);
2121 		nlookup_done(&linknd);
2122 	}
2123 	nlookup_done(&nd);
2124 	return (error);
2125 }
2126 
2127 int
2128 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2129 {
2130 	struct vattr vattr;
2131 	struct vnode *vp;
2132 	struct vnode *dvp;
2133 	int error;
2134 
2135 	bwillinode(1);
2136 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2137 	if ((error = nlookup(nd)) != 0)
2138 		return (error);
2139 	if (nd->nl_nch.ncp->nc_vp)
2140 		return (EEXIST);
2141 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2142 		return (error);
2143 	dvp = nd->nl_dvp;
2144 	VATTR_NULL(&vattr);
2145 	vattr.va_mode = mode;
2146 	error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2147 	if (error == 0)
2148 		vput(vp);
2149 	return (error);
2150 }
2151 
2152 /*
2153  * symlink(char *path, char *link)
2154  *
2155  * Make a symbolic link.
2156  */
2157 int
2158 sys_symlink(struct symlink_args *uap)
2159 {
2160 	struct thread *td = curthread;
2161 	struct nlookupdata nd;
2162 	char *path;
2163 	int error;
2164 	int mode;
2165 
2166 	path = objcache_get(namei_oc, M_WAITOK);
2167 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2168 	if (error == 0) {
2169 		error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2170 		if (error == 0) {
2171 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2172 			error = kern_symlink(&nd, path, mode);
2173 		}
2174 		nlookup_done(&nd);
2175 	}
2176 	objcache_put(namei_oc, path);
2177 	return (error);
2178 }
2179 
2180 /*
2181  * undelete_args(char *path)
2182  *
2183  * Delete a whiteout from the filesystem.
2184  */
2185 /* ARGSUSED */
2186 int
2187 sys_undelete(struct undelete_args *uap)
2188 {
2189 	struct nlookupdata nd;
2190 	int error;
2191 
2192 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2193 	bwillinode(1);
2194 	nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2195 	if (error == 0)
2196 		error = nlookup(&nd);
2197 	if (error == 0)
2198 		error = ncp_writechk(&nd.nl_nch);
2199 	if (error == 0) {
2200 		error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2201 				      NAMEI_DELETE);
2202 	}
2203 	nlookup_done(&nd);
2204 	return (error);
2205 }
2206 
2207 int
2208 kern_unlink(struct nlookupdata *nd)
2209 {
2210 	int error;
2211 
2212 	bwillinode(1);
2213 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2214 	if ((error = nlookup(nd)) != 0)
2215 		return (error);
2216 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2217 		return (error);
2218 	error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2219 	return (error);
2220 }
2221 
2222 /*
2223  * unlink_args(char *path)
2224  *
2225  * Delete a name from the filesystem.
2226  */
2227 int
2228 sys_unlink(struct unlink_args *uap)
2229 {
2230 	struct nlookupdata nd;
2231 	int error;
2232 
2233 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2234 	if (error == 0)
2235 		error = kern_unlink(&nd);
2236 	nlookup_done(&nd);
2237 	return (error);
2238 }
2239 
2240 
2241 /*
2242  * unlinkat_args(int fd, char *path, int flags)
2243  *
2244  * Delete the file or directory entry pointed to by fd/path.
2245  */
2246 int
2247 sys_unlinkat(struct unlinkat_args *uap)
2248 {
2249 	struct nlookupdata nd;
2250 	struct file *fp;
2251 	int error;
2252 
2253 	if (uap->flags & ~AT_REMOVEDIR)
2254 		return (EINVAL);
2255 
2256 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2257 	if (error == 0) {
2258 		if (uap->flags & AT_REMOVEDIR)
2259 			error = kern_rmdir(&nd);
2260 		else
2261 			error = kern_unlink(&nd);
2262 	}
2263 	nlookup_done_at(&nd, fp);
2264 	return (error);
2265 }
2266 
2267 int
2268 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2269 {
2270 	struct thread *td = curthread;
2271 	struct proc *p = td->td_proc;
2272 	struct file *fp;
2273 	struct vnode *vp;
2274 	struct vattr vattr;
2275 	off_t new_offset;
2276 	int error;
2277 
2278 	fp = holdfp(p->p_fd, fd, -1);
2279 	if (fp == NULL)
2280 		return (EBADF);
2281 	if (fp->f_type != DTYPE_VNODE) {
2282 		error = ESPIPE;
2283 		goto done;
2284 	}
2285 	vp = (struct vnode *)fp->f_data;
2286 
2287 	switch (whence) {
2288 	case L_INCR:
2289 		new_offset = fp->f_offset + offset;
2290 		error = 0;
2291 		break;
2292 	case L_XTND:
2293 		error = VOP_GETATTR(vp, &vattr);
2294 		new_offset = offset + vattr.va_size;
2295 		break;
2296 	case L_SET:
2297 		new_offset = offset;
2298 		error = 0;
2299 		break;
2300 	default:
2301 		new_offset = 0;
2302 		error = EINVAL;
2303 		break;
2304 	}
2305 
2306 	/*
2307 	 * Validate the seek position.  Negative offsets are not allowed
2308 	 * for regular files or directories.
2309 	 *
2310 	 * Normally we would also not want to allow negative offsets for
2311 	 * character and block-special devices.  However kvm addresses
2312 	 * on 64 bit architectures might appear to be negative and must
2313 	 * be allowed.
2314 	 */
2315 	if (error == 0) {
2316 		if (new_offset < 0 &&
2317 		    (vp->v_type == VREG || vp->v_type == VDIR)) {
2318 			error = EINVAL;
2319 		} else {
2320 			fp->f_offset = new_offset;
2321 		}
2322 	}
2323 	*res = fp->f_offset;
2324 done:
2325 	fdrop(fp);
2326 	return (error);
2327 }
2328 
2329 /*
2330  * lseek_args(int fd, int pad, off_t offset, int whence)
2331  *
2332  * Reposition read/write file offset.
2333  */
2334 int
2335 sys_lseek(struct lseek_args *uap)
2336 {
2337 	int error;
2338 
2339 	error = kern_lseek(uap->fd, uap->offset, uap->whence,
2340 			   &uap->sysmsg_offset);
2341 
2342 	return (error);
2343 }
2344 
2345 /*
2346  * Check if current process can access given file.  amode is a bitmask of *_OK
2347  * access bits.  flags is a bitmask of AT_* flags.
2348  */
2349 int
2350 kern_access(struct nlookupdata *nd, int amode, int flags)
2351 {
2352 	struct vnode *vp;
2353 	int error, mode;
2354 
2355 	if (flags & ~AT_EACCESS)
2356 		return (EINVAL);
2357 	if ((error = nlookup(nd)) != 0)
2358 		return (error);
2359 retry:
2360 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2361 	if (error)
2362 		return (error);
2363 
2364 	/* Flags == 0 means only check for existence. */
2365 	if (amode) {
2366 		mode = 0;
2367 		if (amode & R_OK)
2368 			mode |= VREAD;
2369 		if (amode & W_OK)
2370 			mode |= VWRITE;
2371 		if (amode & X_OK)
2372 			mode |= VEXEC;
2373 		if ((mode & VWRITE) == 0 ||
2374 		    (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2375 			error = VOP_ACCESS_FLAGS(vp, mode, flags, nd->nl_cred);
2376 
2377 		/*
2378 		 * If the file handle is stale we have to re-resolve the
2379 		 * entry.  This is a hack at the moment.
2380 		 */
2381 		if (error == ESTALE) {
2382 			vput(vp);
2383 			cache_setunresolved(&nd->nl_nch);
2384 			error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2385 			if (error == 0) {
2386 				vp = NULL;
2387 				goto retry;
2388 			}
2389 			return(error);
2390 		}
2391 	}
2392 	vput(vp);
2393 	return (error);
2394 }
2395 
2396 /*
2397  * access_args(char *path, int flags)
2398  *
2399  * Check access permissions.
2400  */
2401 int
2402 sys_access(struct access_args *uap)
2403 {
2404 	struct nlookupdata nd;
2405 	int error;
2406 
2407 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2408 	if (error == 0)
2409 		error = kern_access(&nd, uap->flags, 0);
2410 	nlookup_done(&nd);
2411 	return (error);
2412 }
2413 
2414 
2415 /*
2416  * faccessat_args(int fd, char *path, int amode, int flags)
2417  *
2418  * Check access permissions.
2419  */
2420 int
2421 sys_faccessat(struct faccessat_args *uap)
2422 {
2423 	struct nlookupdata nd;
2424 	struct file *fp;
2425 	int error;
2426 
2427 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE,
2428 				NLC_FOLLOW);
2429 	if (error == 0)
2430 		error = kern_access(&nd, uap->amode, uap->flags);
2431 	nlookup_done_at(&nd, fp);
2432 	return (error);
2433 }
2434 
2435 
2436 int
2437 kern_stat(struct nlookupdata *nd, struct stat *st)
2438 {
2439 	int error;
2440 	struct vnode *vp;
2441 	thread_t td;
2442 
2443 	if ((error = nlookup(nd)) != 0)
2444 		return (error);
2445 again:
2446 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2447 		return (ENOENT);
2448 
2449 	td = curthread;
2450 	if ((error = vget(vp, LK_SHARED)) != 0)
2451 		return (error);
2452 	error = vn_stat(vp, st, nd->nl_cred);
2453 
2454 	/*
2455 	 * If the file handle is stale we have to re-resolve the entry.  This
2456 	 * is a hack at the moment.
2457 	 */
2458 	if (error == ESTALE) {
2459 		vput(vp);
2460 		cache_setunresolved(&nd->nl_nch);
2461 		error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2462 		if (error == 0)
2463 			goto again;
2464 	} else {
2465 		vput(vp);
2466 	}
2467 	return (error);
2468 }
2469 
2470 /*
2471  * stat_args(char *path, struct stat *ub)
2472  *
2473  * Get file status; this version follows links.
2474  */
2475 int
2476 sys_stat(struct stat_args *uap)
2477 {
2478 	struct nlookupdata nd;
2479 	struct stat st;
2480 	int error;
2481 
2482 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2483 	if (error == 0) {
2484 		error = kern_stat(&nd, &st);
2485 		if (error == 0)
2486 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2487 	}
2488 	nlookup_done(&nd);
2489 	return (error);
2490 }
2491 
2492 /*
2493  * lstat_args(char *path, struct stat *ub)
2494  *
2495  * Get file status; this version does not follow links.
2496  */
2497 int
2498 sys_lstat(struct lstat_args *uap)
2499 {
2500 	struct nlookupdata nd;
2501 	struct stat st;
2502 	int error;
2503 
2504 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2505 	if (error == 0) {
2506 		error = kern_stat(&nd, &st);
2507 		if (error == 0)
2508 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2509 	}
2510 	nlookup_done(&nd);
2511 	return (error);
2512 }
2513 
2514 /*
2515  * fstatat_args(int fd, char *path, struct stat *sb, int flags)
2516  *
2517  * Get status of file pointed to by fd/path.
2518  */
2519 int
2520 sys_fstatat(struct fstatat_args *uap)
2521 {
2522 	struct nlookupdata nd;
2523 	struct stat st;
2524 	int error;
2525 	int flags;
2526 	struct file *fp;
2527 
2528 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
2529 		return (EINVAL);
2530 
2531 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
2532 
2533 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
2534 				UIO_USERSPACE, flags);
2535 	if (error == 0) {
2536 		error = kern_stat(&nd, &st);
2537 		if (error == 0)
2538 			error = copyout(&st, uap->sb, sizeof(*uap->sb));
2539 	}
2540 	nlookup_done_at(&nd, fp);
2541 	return (error);
2542 }
2543 
2544 /*
2545  * pathconf_Args(char *path, int name)
2546  *
2547  * Get configurable pathname variables.
2548  */
2549 /* ARGSUSED */
2550 int
2551 sys_pathconf(struct pathconf_args *uap)
2552 {
2553 	struct nlookupdata nd;
2554 	struct vnode *vp;
2555 	int error;
2556 
2557 	vp = NULL;
2558 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2559 	if (error == 0)
2560 		error = nlookup(&nd);
2561 	if (error == 0)
2562 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2563 	nlookup_done(&nd);
2564 	if (error == 0) {
2565 		error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
2566 		vput(vp);
2567 	}
2568 	return (error);
2569 }
2570 
2571 /*
2572  * XXX: daver
2573  * kern_readlink isn't properly split yet.  There is a copyin burried
2574  * in VOP_READLINK().
2575  */
2576 int
2577 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2578 {
2579 	struct thread *td = curthread;
2580 	struct proc *p = td->td_proc;
2581 	struct vnode *vp;
2582 	struct iovec aiov;
2583 	struct uio auio;
2584 	int error;
2585 
2586 	if ((error = nlookup(nd)) != 0)
2587 		return (error);
2588 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2589 	if (error)
2590 		return (error);
2591 	if (vp->v_type != VLNK) {
2592 		error = EINVAL;
2593 	} else {
2594 		aiov.iov_base = buf;
2595 		aiov.iov_len = count;
2596 		auio.uio_iov = &aiov;
2597 		auio.uio_iovcnt = 1;
2598 		auio.uio_offset = 0;
2599 		auio.uio_rw = UIO_READ;
2600 		auio.uio_segflg = UIO_USERSPACE;
2601 		auio.uio_td = td;
2602 		auio.uio_resid = count;
2603 		error = VOP_READLINK(vp, &auio, p->p_ucred);
2604 	}
2605 	vput(vp);
2606 	*res = count - auio.uio_resid;
2607 	return (error);
2608 }
2609 
2610 /*
2611  * readlink_args(char *path, char *buf, int count)
2612  *
2613  * Return target name of a symbolic link.
2614  */
2615 int
2616 sys_readlink(struct readlink_args *uap)
2617 {
2618 	struct nlookupdata nd;
2619 	int error;
2620 
2621 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2622 	if (error == 0) {
2623 		error = kern_readlink(&nd, uap->buf, uap->count,
2624 					&uap->sysmsg_result);
2625 	}
2626 	nlookup_done(&nd);
2627 	return (error);
2628 }
2629 
2630 static int
2631 setfflags(struct vnode *vp, int flags)
2632 {
2633 	struct thread *td = curthread;
2634 	struct proc *p = td->td_proc;
2635 	int error;
2636 	struct vattr vattr;
2637 
2638 	/*
2639 	 * Prevent non-root users from setting flags on devices.  When
2640 	 * a device is reused, users can retain ownership of the device
2641 	 * if they are allowed to set flags and programs assume that
2642 	 * chown can't fail when done as root.
2643 	 */
2644 	if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2645 	    ((error = priv_check_cred(p->p_ucred, PRIV_VFS_CHFLAGS_DEV, 0)) != 0))
2646 		return (error);
2647 
2648 	/*
2649 	 * note: vget is required for any operation that might mod the vnode
2650 	 * so VINACTIVE is properly cleared.
2651 	 */
2652 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2653 		VATTR_NULL(&vattr);
2654 		vattr.va_flags = flags;
2655 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2656 		vput(vp);
2657 	}
2658 	return (error);
2659 }
2660 
2661 /*
2662  * chflags(char *path, int flags)
2663  *
2664  * Change flags of a file given a path name.
2665  */
2666 /* ARGSUSED */
2667 int
2668 sys_chflags(struct chflags_args *uap)
2669 {
2670 	struct nlookupdata nd;
2671 	struct vnode *vp;
2672 	int error;
2673 
2674 	vp = NULL;
2675 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2676 	if (error == 0)
2677 		error = nlookup(&nd);
2678 	if (error == 0)
2679 		error = ncp_writechk(&nd.nl_nch);
2680 	if (error == 0)
2681 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2682 	nlookup_done(&nd);
2683 	if (error == 0) {
2684 		error = setfflags(vp, uap->flags);
2685 		vrele(vp);
2686 	}
2687 	return (error);
2688 }
2689 
2690 /*
2691  * lchflags(char *path, int flags)
2692  *
2693  * Change flags of a file given a path name, but don't follow symlinks.
2694  */
2695 /* ARGSUSED */
2696 int
2697 sys_lchflags(struct lchflags_args *uap)
2698 {
2699 	struct nlookupdata nd;
2700 	struct vnode *vp;
2701 	int error;
2702 
2703 	vp = NULL;
2704 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2705 	if (error == 0)
2706 		error = nlookup(&nd);
2707 	if (error == 0)
2708 		error = ncp_writechk(&nd.nl_nch);
2709 	if (error == 0)
2710 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2711 	nlookup_done(&nd);
2712 	if (error == 0) {
2713 		error = setfflags(vp, uap->flags);
2714 		vrele(vp);
2715 	}
2716 	return (error);
2717 }
2718 
2719 /*
2720  * fchflags_args(int fd, int flags)
2721  *
2722  * Change flags of a file given a file descriptor.
2723  */
2724 /* ARGSUSED */
2725 int
2726 sys_fchflags(struct fchflags_args *uap)
2727 {
2728 	struct thread *td = curthread;
2729 	struct proc *p = td->td_proc;
2730 	struct file *fp;
2731 	int error;
2732 
2733 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2734 		return (error);
2735 	if (fp->f_nchandle.ncp)
2736 		error = ncp_writechk(&fp->f_nchandle);
2737 	if (error == 0)
2738 		error = setfflags((struct vnode *) fp->f_data, uap->flags);
2739 	fdrop(fp);
2740 	return (error);
2741 }
2742 
2743 static int
2744 setfmode(struct vnode *vp, int mode)
2745 {
2746 	struct thread *td = curthread;
2747 	struct proc *p = td->td_proc;
2748 	int error;
2749 	struct vattr vattr;
2750 
2751 	/*
2752 	 * note: vget is required for any operation that might mod the vnode
2753 	 * so VINACTIVE is properly cleared.
2754 	 */
2755 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2756 		VATTR_NULL(&vattr);
2757 		vattr.va_mode = mode & ALLPERMS;
2758 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2759 		vput(vp);
2760 	}
2761 	return error;
2762 }
2763 
2764 int
2765 kern_chmod(struct nlookupdata *nd, int mode)
2766 {
2767 	struct vnode *vp;
2768 	int error;
2769 
2770 	if ((error = nlookup(nd)) != 0)
2771 		return (error);
2772 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2773 		return (error);
2774 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2775 		error = setfmode(vp, mode);
2776 	vrele(vp);
2777 	return (error);
2778 }
2779 
2780 /*
2781  * chmod_args(char *path, int mode)
2782  *
2783  * Change mode of a file given path name.
2784  */
2785 /* ARGSUSED */
2786 int
2787 sys_chmod(struct chmod_args *uap)
2788 {
2789 	struct nlookupdata nd;
2790 	int error;
2791 
2792 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2793 	if (error == 0)
2794 		error = kern_chmod(&nd, uap->mode);
2795 	nlookup_done(&nd);
2796 	return (error);
2797 }
2798 
2799 /*
2800  * lchmod_args(char *path, int mode)
2801  *
2802  * Change mode of a file given path name (don't follow links.)
2803  */
2804 /* ARGSUSED */
2805 int
2806 sys_lchmod(struct lchmod_args *uap)
2807 {
2808 	struct nlookupdata nd;
2809 	int error;
2810 
2811 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2812 	if (error == 0)
2813 		error = kern_chmod(&nd, uap->mode);
2814 	nlookup_done(&nd);
2815 	return (error);
2816 }
2817 
2818 /*
2819  * fchmod_args(int fd, int mode)
2820  *
2821  * Change mode of a file given a file descriptor.
2822  */
2823 /* ARGSUSED */
2824 int
2825 sys_fchmod(struct fchmod_args *uap)
2826 {
2827 	struct thread *td = curthread;
2828 	struct proc *p = td->td_proc;
2829 	struct file *fp;
2830 	int error;
2831 
2832 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2833 		return (error);
2834 	if (fp->f_nchandle.ncp)
2835 		error = ncp_writechk(&fp->f_nchandle);
2836 	if (error == 0)
2837 		error = setfmode((struct vnode *)fp->f_data, uap->mode);
2838 	fdrop(fp);
2839 	return (error);
2840 }
2841 
2842 /*
2843  * fchmodat_args(char *path, int mode)
2844  *
2845  * Change mode of a file pointed to by fd/path.
2846  */
2847 int
2848 sys_fchmodat(struct fchmodat_args *uap)
2849 {
2850 	struct nlookupdata nd;
2851 	struct file *fp;
2852 	int error;
2853 	int flags;
2854 
2855 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
2856 		return (EINVAL);
2857 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
2858 
2859 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
2860 				UIO_USERSPACE, flags);
2861 	if (error == 0)
2862 		error = kern_chmod(&nd, uap->mode);
2863 	nlookup_done_at(&nd, fp);
2864 	return (error);
2865 }
2866 
2867 static int
2868 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2869 {
2870 	struct thread *td = curthread;
2871 	struct proc *p = td->td_proc;
2872 	int error;
2873 	struct vattr vattr;
2874 
2875 	/*
2876 	 * note: vget is required for any operation that might mod the vnode
2877 	 * so VINACTIVE is properly cleared.
2878 	 */
2879 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2880 		VATTR_NULL(&vattr);
2881 		vattr.va_uid = uid;
2882 		vattr.va_gid = gid;
2883 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2884 		vput(vp);
2885 	}
2886 	return error;
2887 }
2888 
2889 int
2890 kern_chown(struct nlookupdata *nd, int uid, int gid)
2891 {
2892 	struct vnode *vp;
2893 	int error;
2894 
2895 	if ((error = nlookup(nd)) != 0)
2896 		return (error);
2897 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2898 		return (error);
2899 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2900 		error = setfown(vp, uid, gid);
2901 	vrele(vp);
2902 	return (error);
2903 }
2904 
2905 /*
2906  * chown(char *path, int uid, int gid)
2907  *
2908  * Set ownership given a path name.
2909  */
2910 int
2911 sys_chown(struct chown_args *uap)
2912 {
2913 	struct nlookupdata nd;
2914 	int error;
2915 
2916 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2917 	if (error == 0)
2918 		error = kern_chown(&nd, uap->uid, uap->gid);
2919 	nlookup_done(&nd);
2920 	return (error);
2921 }
2922 
2923 /*
2924  * lchown_args(char *path, int uid, int gid)
2925  *
2926  * Set ownership given a path name, do not cross symlinks.
2927  */
2928 int
2929 sys_lchown(struct lchown_args *uap)
2930 {
2931 	struct nlookupdata nd;
2932 	int error;
2933 
2934 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2935 	if (error == 0)
2936 		error = kern_chown(&nd, uap->uid, uap->gid);
2937 	nlookup_done(&nd);
2938 	return (error);
2939 }
2940 
2941 /*
2942  * fchown_args(int fd, int uid, int gid)
2943  *
2944  * Set ownership given a file descriptor.
2945  */
2946 /* ARGSUSED */
2947 int
2948 sys_fchown(struct fchown_args *uap)
2949 {
2950 	struct thread *td = curthread;
2951 	struct proc *p = td->td_proc;
2952 	struct file *fp;
2953 	int error;
2954 
2955 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2956 		return (error);
2957 	if (fp->f_nchandle.ncp)
2958 		error = ncp_writechk(&fp->f_nchandle);
2959 	if (error == 0)
2960 		error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
2961 	fdrop(fp);
2962 	return (error);
2963 }
2964 
2965 /*
2966  * fchownat(int fd, char *path, int uid, int gid, int flags)
2967  *
2968  * Set ownership of file pointed to by fd/path.
2969  */
2970 int
2971 sys_fchownat(struct fchownat_args *uap)
2972 {
2973 	struct nlookupdata nd;
2974 	struct file *fp;
2975 	int error;
2976 	int flags;
2977 
2978 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
2979 		return (EINVAL);
2980 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
2981 
2982 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
2983 				UIO_USERSPACE, flags);
2984 	if (error == 0)
2985 		error = kern_chown(&nd, uap->uid, uap->gid);
2986 	nlookup_done_at(&nd, fp);
2987 	return (error);
2988 }
2989 
2990 
2991 static int
2992 getutimes(const struct timeval *tvp, struct timespec *tsp)
2993 {
2994 	struct timeval tv[2];
2995 
2996 	if (tvp == NULL) {
2997 		microtime(&tv[0]);
2998 		TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
2999 		tsp[1] = tsp[0];
3000 	} else {
3001 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
3002 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
3003 	}
3004 	return 0;
3005 }
3006 
3007 static int
3008 setutimes(struct vnode *vp, struct vattr *vattr,
3009 	  const struct timespec *ts, int nullflag)
3010 {
3011 	struct thread *td = curthread;
3012 	struct proc *p = td->td_proc;
3013 	int error;
3014 
3015 	VATTR_NULL(vattr);
3016 	vattr->va_atime = ts[0];
3017 	vattr->va_mtime = ts[1];
3018 	if (nullflag)
3019 		vattr->va_vaflags |= VA_UTIMES_NULL;
3020 	error = VOP_SETATTR(vp, vattr, p->p_ucred);
3021 
3022 	return error;
3023 }
3024 
3025 int
3026 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
3027 {
3028 	struct timespec ts[2];
3029 	struct vnode *vp;
3030 	struct vattr vattr;
3031 	int error;
3032 
3033 	if ((error = getutimes(tptr, ts)) != 0)
3034 		return (error);
3035 
3036 	/*
3037 	 * NOTE: utimes() succeeds for the owner even if the file
3038 	 * is not user-writable.
3039 	 */
3040 	nd->nl_flags |= NLC_OWN | NLC_WRITE;
3041 
3042 	if ((error = nlookup(nd)) != 0)
3043 		return (error);
3044 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3045 		return (error);
3046 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3047 		return (error);
3048 
3049 	/*
3050 	 * note: vget is required for any operation that might mod the vnode
3051 	 * so VINACTIVE is properly cleared.
3052 	 */
3053 	if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3054 		error = vget(vp, LK_EXCLUSIVE);
3055 		if (error == 0) {
3056 			error = setutimes(vp, &vattr, ts, (tptr == NULL));
3057 			vput(vp);
3058 		}
3059 	}
3060 	vrele(vp);
3061 	return (error);
3062 }
3063 
3064 /*
3065  * utimes_args(char *path, struct timeval *tptr)
3066  *
3067  * Set the access and modification times of a file.
3068  */
3069 int
3070 sys_utimes(struct utimes_args *uap)
3071 {
3072 	struct timeval tv[2];
3073 	struct nlookupdata nd;
3074 	int error;
3075 
3076 	if (uap->tptr) {
3077  		error = copyin(uap->tptr, tv, sizeof(tv));
3078 		if (error)
3079 			return (error);
3080 	}
3081 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3082 	if (error == 0)
3083 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3084 	nlookup_done(&nd);
3085 	return (error);
3086 }
3087 
3088 /*
3089  * lutimes_args(char *path, struct timeval *tptr)
3090  *
3091  * Set the access and modification times of a file.
3092  */
3093 int
3094 sys_lutimes(struct lutimes_args *uap)
3095 {
3096 	struct timeval tv[2];
3097 	struct nlookupdata nd;
3098 	int error;
3099 
3100 	if (uap->tptr) {
3101 		error = copyin(uap->tptr, tv, sizeof(tv));
3102 		if (error)
3103 			return (error);
3104 	}
3105 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3106 	if (error == 0)
3107 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3108 	nlookup_done(&nd);
3109 	return (error);
3110 }
3111 
3112 /*
3113  * Set utimes on a file descriptor.  The creds used to open the
3114  * file are used to determine whether the operation is allowed
3115  * or not.
3116  */
3117 int
3118 kern_futimes(int fd, struct timeval *tptr)
3119 {
3120 	struct thread *td = curthread;
3121 	struct proc *p = td->td_proc;
3122 	struct timespec ts[2];
3123 	struct file *fp;
3124 	struct vnode *vp;
3125 	struct vattr vattr;
3126 	int error;
3127 
3128 	error = getutimes(tptr, ts);
3129 	if (error)
3130 		return (error);
3131 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3132 		return (error);
3133 	if (fp->f_nchandle.ncp)
3134 		error = ncp_writechk(&fp->f_nchandle);
3135 	if (error == 0) {
3136 		vp = fp->f_data;
3137 		error = vget(vp, LK_EXCLUSIVE);
3138 		if (error == 0) {
3139 			error = VOP_GETATTR(vp, &vattr);
3140 			if (error == 0) {
3141 				error = naccess_va(&vattr, NLC_OWN | NLC_WRITE,
3142 						   fp->f_cred);
3143 			}
3144 			if (error == 0) {
3145 				error = setutimes(vp, &vattr, ts,
3146 						  (tptr == NULL));
3147 			}
3148 			vput(vp);
3149 		}
3150 	}
3151 	fdrop(fp);
3152 	return (error);
3153 }
3154 
3155 /*
3156  * futimes_args(int fd, struct timeval *tptr)
3157  *
3158  * Set the access and modification times of a file.
3159  */
3160 int
3161 sys_futimes(struct futimes_args *uap)
3162 {
3163 	struct timeval tv[2];
3164 	int error;
3165 
3166 	if (uap->tptr) {
3167 		error = copyin(uap->tptr, tv, sizeof(tv));
3168 		if (error)
3169 			return (error);
3170 	}
3171 
3172 	error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
3173 
3174 	return (error);
3175 }
3176 
3177 int
3178 kern_truncate(struct nlookupdata *nd, off_t length)
3179 {
3180 	struct vnode *vp;
3181 	struct vattr vattr;
3182 	int error;
3183 
3184 	if (length < 0)
3185 		return(EINVAL);
3186 	nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE;
3187 	if ((error = nlookup(nd)) != 0)
3188 		return (error);
3189 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3190 		return (error);
3191 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3192 		return (error);
3193 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
3194 		vrele(vp);
3195 		return (error);
3196 	}
3197 	if (vp->v_type == VDIR) {
3198 		error = EISDIR;
3199 	} else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3200 		VATTR_NULL(&vattr);
3201 		vattr.va_size = length;
3202 		error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
3203 	}
3204 	vput(vp);
3205 	return (error);
3206 }
3207 
3208 /*
3209  * truncate(char *path, int pad, off_t length)
3210  *
3211  * Truncate a file given its path name.
3212  */
3213 int
3214 sys_truncate(struct truncate_args *uap)
3215 {
3216 	struct nlookupdata nd;
3217 	int error;
3218 
3219 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3220 	if (error == 0)
3221 		error = kern_truncate(&nd, uap->length);
3222 	nlookup_done(&nd);
3223 	return error;
3224 }
3225 
3226 int
3227 kern_ftruncate(int fd, off_t length)
3228 {
3229 	struct thread *td = curthread;
3230 	struct proc *p = td->td_proc;
3231 	struct vattr vattr;
3232 	struct vnode *vp;
3233 	struct file *fp;
3234 	int error;
3235 
3236 	if (length < 0)
3237 		return(EINVAL);
3238 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3239 		return (error);
3240 	if (fp->f_nchandle.ncp) {
3241 		error = ncp_writechk(&fp->f_nchandle);
3242 		if (error)
3243 			goto done;
3244 	}
3245 	if ((fp->f_flag & FWRITE) == 0) {
3246 		error = EINVAL;
3247 		goto done;
3248 	}
3249 	if (fp->f_flag & FAPPENDONLY) {	/* inode was set s/uapnd */
3250 		error = EINVAL;
3251 		goto done;
3252 	}
3253 	vp = (struct vnode *)fp->f_data;
3254 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3255 	if (vp->v_type == VDIR) {
3256 		error = EISDIR;
3257 	} else if ((error = vn_writechk(vp, NULL)) == 0) {
3258 		VATTR_NULL(&vattr);
3259 		vattr.va_size = length;
3260 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
3261 	}
3262 	vn_unlock(vp);
3263 done:
3264 	fdrop(fp);
3265 	return (error);
3266 }
3267 
3268 /*
3269  * ftruncate_args(int fd, int pad, off_t length)
3270  *
3271  * Truncate a file given a file descriptor.
3272  */
3273 int
3274 sys_ftruncate(struct ftruncate_args *uap)
3275 {
3276 	int error;
3277 
3278 	error = kern_ftruncate(uap->fd, uap->length);
3279 
3280 	return (error);
3281 }
3282 
3283 /*
3284  * fsync(int fd)
3285  *
3286  * Sync an open file.
3287  */
3288 /* ARGSUSED */
3289 int
3290 sys_fsync(struct fsync_args *uap)
3291 {
3292 	struct thread *td = curthread;
3293 	struct proc *p = td->td_proc;
3294 	struct vnode *vp;
3295 	struct file *fp;
3296 	vm_object_t obj;
3297 	int error;
3298 
3299 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3300 		return (error);
3301 	vp = (struct vnode *)fp->f_data;
3302 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3303 	if ((obj = vp->v_object) != NULL)
3304 		vm_object_page_clean(obj, 0, 0, 0);
3305 	error = VOP_FSYNC(vp, MNT_WAIT, VOP_FSYNC_SYSCALL);
3306 	if (error == 0 && vp->v_mount)
3307 		error = buf_fsync(vp);
3308 	vn_unlock(vp);
3309 	fdrop(fp);
3310 	return (error);
3311 }
3312 
3313 int
3314 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
3315 {
3316 	struct nchandle fnchd;
3317 	struct nchandle tnchd;
3318 	struct namecache *ncp;
3319 	struct vnode *fdvp;
3320 	struct vnode *tdvp;
3321 	struct mount *mp;
3322 	int error;
3323 
3324 	bwillinode(1);
3325 	fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC;
3326 	if ((error = nlookup(fromnd)) != 0)
3327 		return (error);
3328 	if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
3329 		return (ENOENT);
3330 	fnchd.mount = fromnd->nl_nch.mount;
3331 	cache_hold(&fnchd);
3332 
3333 	/*
3334 	 * unlock the source nch so we can lookup the target nch without
3335 	 * deadlocking.  The target may or may not exist so we do not check
3336 	 * for a target vp like kern_mkdir() and other creation functions do.
3337 	 *
3338 	 * The source and target directories are ref'd and rechecked after
3339 	 * everything is relocked to determine if the source or target file
3340 	 * has been renamed.
3341 	 */
3342 	KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
3343 	fromnd->nl_flags &= ~NLC_NCPISLOCKED;
3344 	cache_unlock(&fromnd->nl_nch);
3345 
3346 	tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP;
3347 	if ((error = nlookup(tond)) != 0) {
3348 		cache_drop(&fnchd);
3349 		return (error);
3350 	}
3351 	if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
3352 		cache_drop(&fnchd);
3353 		return (ENOENT);
3354 	}
3355 	tnchd.mount = tond->nl_nch.mount;
3356 	cache_hold(&tnchd);
3357 
3358 	/*
3359 	 * If the source and target are the same there is nothing to do
3360 	 */
3361 	if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
3362 		cache_drop(&fnchd);
3363 		cache_drop(&tnchd);
3364 		return (0);
3365 	}
3366 
3367 	/*
3368 	 * Mount points cannot be renamed or overwritten
3369 	 */
3370 	if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
3371 	    NCF_ISMOUNTPT
3372 	) {
3373 		cache_drop(&fnchd);
3374 		cache_drop(&tnchd);
3375 		return (EINVAL);
3376 	}
3377 
3378 	/*
3379 	 * relock the source ncp.  NOTE AFTER RELOCKING: the source ncp
3380 	 * may have become invalid while it was unlocked, nc_vp and nc_mount
3381 	 * could be NULL.
3382 	 */
3383 	if (cache_lock_nonblock(&fromnd->nl_nch) == 0) {
3384 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3385 	} else if (fromnd->nl_nch.ncp > tond->nl_nch.ncp) {
3386 		cache_lock(&fromnd->nl_nch);
3387 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3388 	} else {
3389 		cache_unlock(&tond->nl_nch);
3390 		cache_lock(&fromnd->nl_nch);
3391 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3392 		cache_lock(&tond->nl_nch);
3393 		cache_resolve(&tond->nl_nch, tond->nl_cred);
3394 	}
3395 	fromnd->nl_flags |= NLC_NCPISLOCKED;
3396 
3397 	/*
3398 	 * make sure the parent directories linkages are the same
3399 	 */
3400 	if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
3401 	    tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
3402 		cache_drop(&fnchd);
3403 		cache_drop(&tnchd);
3404 		return (ENOENT);
3405 	}
3406 
3407 	/*
3408 	 * Both the source and target must be within the same filesystem and
3409 	 * in the same filesystem as their parent directories within the
3410 	 * namecache topology.
3411 	 *
3412 	 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
3413 	 */
3414 	mp = fnchd.mount;
3415 	if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
3416 	    mp != tond->nl_nch.mount) {
3417 		cache_drop(&fnchd);
3418 		cache_drop(&tnchd);
3419 		return (EXDEV);
3420 	}
3421 
3422 	/*
3423 	 * Make sure the mount point is writable
3424 	 */
3425 	if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
3426 		cache_drop(&fnchd);
3427 		cache_drop(&tnchd);
3428 		return (error);
3429 	}
3430 
3431 	/*
3432 	 * If the target exists and either the source or target is a directory,
3433 	 * then both must be directories.
3434 	 *
3435 	 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
3436 	 * have become NULL.
3437 	 */
3438 	if (tond->nl_nch.ncp->nc_vp) {
3439 		if (fromnd->nl_nch.ncp->nc_vp == NULL) {
3440 			error = ENOENT;
3441 		} else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
3442 			if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
3443 				error = ENOTDIR;
3444 		} else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
3445 			error = EISDIR;
3446 		}
3447 	}
3448 
3449 	/*
3450 	 * You cannot rename a source into itself or a subdirectory of itself.
3451 	 * We check this by travsersing the target directory upwards looking
3452 	 * for a match against the source.
3453 	 */
3454 	if (error == 0) {
3455 		for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
3456 			if (fromnd->nl_nch.ncp == ncp) {
3457 				error = EINVAL;
3458 				break;
3459 			}
3460 		}
3461 	}
3462 
3463 	cache_drop(&fnchd);
3464 	cache_drop(&tnchd);
3465 
3466 	/*
3467 	 * Even though the namespaces are different, they may still represent
3468 	 * hardlinks to the same file.  The filesystem might have a hard time
3469 	 * with this so we issue a NREMOVE of the source instead of a NRENAME
3470 	 * when we detect the situation.
3471 	 */
3472 	if (error == 0) {
3473 		fdvp = fromnd->nl_dvp;
3474 		tdvp = tond->nl_dvp;
3475 		if (fdvp == NULL || tdvp == NULL) {
3476 			error = EPERM;
3477 		} else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
3478 			error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
3479 					    fromnd->nl_cred);
3480 		} else {
3481 			error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
3482 					    fdvp, tdvp, tond->nl_cred);
3483 		}
3484 	}
3485 	return (error);
3486 }
3487 
3488 /*
3489  * rename_args(char *from, char *to)
3490  *
3491  * Rename files.  Source and destination must either both be directories,
3492  * or both not be directories.  If target is a directory, it must be empty.
3493  */
3494 int
3495 sys_rename(struct rename_args *uap)
3496 {
3497 	struct nlookupdata fromnd, tond;
3498 	int error;
3499 
3500 	error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3501 	if (error == 0) {
3502 		error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3503 		if (error == 0)
3504 			error = kern_rename(&fromnd, &tond);
3505 		nlookup_done(&tond);
3506 	}
3507 	nlookup_done(&fromnd);
3508 	return (error);
3509 }
3510 
3511 int
3512 kern_mkdir(struct nlookupdata *nd, int mode)
3513 {
3514 	struct thread *td = curthread;
3515 	struct proc *p = td->td_proc;
3516 	struct vnode *vp;
3517 	struct vattr vattr;
3518 	int error;
3519 
3520 	bwillinode(1);
3521 	nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
3522 	if ((error = nlookup(nd)) != 0)
3523 		return (error);
3524 
3525 	if (nd->nl_nch.ncp->nc_vp)
3526 		return (EEXIST);
3527 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3528 		return (error);
3529 	VATTR_NULL(&vattr);
3530 	vattr.va_type = VDIR;
3531 	vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3532 
3533 	vp = NULL;
3534 	error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, p->p_ucred, &vattr);
3535 	if (error == 0)
3536 		vput(vp);
3537 	return (error);
3538 }
3539 
3540 /*
3541  * mkdir_args(char *path, int mode)
3542  *
3543  * Make a directory file.
3544  */
3545 /* ARGSUSED */
3546 int
3547 sys_mkdir(struct mkdir_args *uap)
3548 {
3549 	struct nlookupdata nd;
3550 	int error;
3551 
3552 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3553 	if (error == 0)
3554 		error = kern_mkdir(&nd, uap->mode);
3555 	nlookup_done(&nd);
3556 	return (error);
3557 }
3558 
3559 int
3560 kern_rmdir(struct nlookupdata *nd)
3561 {
3562 	int error;
3563 
3564 	bwillinode(1);
3565 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
3566 	if ((error = nlookup(nd)) != 0)
3567 		return (error);
3568 
3569 	/*
3570 	 * Do not allow directories representing mount points to be
3571 	 * deleted, even if empty.  Check write perms on mount point
3572 	 * in case the vnode is aliased (aka nullfs).
3573 	 */
3574 	if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3575 		return (EINVAL);
3576 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3577 		return (error);
3578 	error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
3579 	return (error);
3580 }
3581 
3582 /*
3583  * rmdir_args(char *path)
3584  *
3585  * Remove a directory file.
3586  */
3587 /* ARGSUSED */
3588 int
3589 sys_rmdir(struct rmdir_args *uap)
3590 {
3591 	struct nlookupdata nd;
3592 	int error;
3593 
3594 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3595 	if (error == 0)
3596 		error = kern_rmdir(&nd);
3597 	nlookup_done(&nd);
3598 	return (error);
3599 }
3600 
3601 int
3602 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3603     enum uio_seg direction)
3604 {
3605 	struct thread *td = curthread;
3606 	struct proc *p = td->td_proc;
3607 	struct vnode *vp;
3608 	struct file *fp;
3609 	struct uio auio;
3610 	struct iovec aiov;
3611 	off_t loff;
3612 	int error, eofflag;
3613 
3614 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3615 		return (error);
3616 	if ((fp->f_flag & FREAD) == 0) {
3617 		error = EBADF;
3618 		goto done;
3619 	}
3620 	vp = (struct vnode *)fp->f_data;
3621 unionread:
3622 	if (vp->v_type != VDIR) {
3623 		error = EINVAL;
3624 		goto done;
3625 	}
3626 	aiov.iov_base = buf;
3627 	aiov.iov_len = count;
3628 	auio.uio_iov = &aiov;
3629 	auio.uio_iovcnt = 1;
3630 	auio.uio_rw = UIO_READ;
3631 	auio.uio_segflg = direction;
3632 	auio.uio_td = td;
3633 	auio.uio_resid = count;
3634 	loff = auio.uio_offset = fp->f_offset;
3635 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3636 	fp->f_offset = auio.uio_offset;
3637 	if (error)
3638 		goto done;
3639 	if (count == auio.uio_resid) {
3640 		if (union_dircheckp) {
3641 			error = union_dircheckp(td, &vp, fp);
3642 			if (error == -1)
3643 				goto unionread;
3644 			if (error)
3645 				goto done;
3646 		}
3647 #if 0
3648 		if ((vp->v_flag & VROOT) &&
3649 		    (vp->v_mount->mnt_flag & MNT_UNION)) {
3650 			struct vnode *tvp = vp;
3651 			vp = vp->v_mount->mnt_vnodecovered;
3652 			vref(vp);
3653 			fp->f_data = vp;
3654 			fp->f_offset = 0;
3655 			vrele(tvp);
3656 			goto unionread;
3657 		}
3658 #endif
3659 	}
3660 
3661 	/*
3662 	 * WARNING!  *basep may not be wide enough to accomodate the
3663 	 * seek offset.   XXX should we hack this to return the upper 32 bits
3664 	 * for offsets greater then 4G?
3665 	 */
3666 	if (basep) {
3667 		*basep = (long)loff;
3668 	}
3669 	*res = count - auio.uio_resid;
3670 done:
3671 	fdrop(fp);
3672 	return (error);
3673 }
3674 
3675 /*
3676  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3677  *
3678  * Read a block of directory entries in a file system independent format.
3679  */
3680 int
3681 sys_getdirentries(struct getdirentries_args *uap)
3682 {
3683 	long base;
3684 	int error;
3685 
3686 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3687 				   &uap->sysmsg_result, UIO_USERSPACE);
3688 
3689 	if (error == 0 && uap->basep)
3690 		error = copyout(&base, uap->basep, sizeof(*uap->basep));
3691 	return (error);
3692 }
3693 
3694 /*
3695  * getdents_args(int fd, char *buf, size_t count)
3696  */
3697 int
3698 sys_getdents(struct getdents_args *uap)
3699 {
3700 	int error;
3701 
3702 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3703 	    &uap->sysmsg_result, UIO_USERSPACE);
3704 
3705 	return (error);
3706 }
3707 
3708 /*
3709  * umask(int newmask)
3710  *
3711  * Set the mode mask for creation of filesystem nodes.
3712  *
3713  * MP SAFE
3714  */
3715 int
3716 sys_umask(struct umask_args *uap)
3717 {
3718 	struct thread *td = curthread;
3719 	struct proc *p = td->td_proc;
3720 	struct filedesc *fdp;
3721 
3722 	fdp = p->p_fd;
3723 	uap->sysmsg_result = fdp->fd_cmask;
3724 	fdp->fd_cmask = uap->newmask & ALLPERMS;
3725 	return (0);
3726 }
3727 
3728 /*
3729  * revoke(char *path)
3730  *
3731  * Void all references to file by ripping underlying filesystem
3732  * away from vnode.
3733  */
3734 /* ARGSUSED */
3735 int
3736 sys_revoke(struct revoke_args *uap)
3737 {
3738 	struct nlookupdata nd;
3739 	struct vattr vattr;
3740 	struct vnode *vp;
3741 	struct ucred *cred;
3742 	int error;
3743 
3744 	vp = NULL;
3745 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3746 	if (error == 0)
3747 		error = nlookup(&nd);
3748 	if (error == 0)
3749 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3750 	cred = crhold(nd.nl_cred);
3751 	nlookup_done(&nd);
3752 	if (error == 0) {
3753 		if (error == 0)
3754 			error = VOP_GETATTR(vp, &vattr);
3755 		if (error == 0 && cred->cr_uid != vattr.va_uid)
3756 			error = priv_check_cred(cred, PRIV_VFS_REVOKE, 0);
3757 		if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) {
3758 			if (vcount(vp) > 0)
3759 				error = vrevoke(vp, cred);
3760 		} else if (error == 0) {
3761 			error = vrevoke(vp, cred);
3762 		}
3763 		vrele(vp);
3764 	}
3765 	if (cred)
3766 		crfree(cred);
3767 	return (error);
3768 }
3769 
3770 /*
3771  * getfh_args(char *fname, fhandle_t *fhp)
3772  *
3773  * Get (NFS) file handle
3774  *
3775  * NOTE: We use the fsid of the covering mount, even if it is a nullfs
3776  * mount.  This allows nullfs mounts to be explicitly exported.
3777  *
3778  * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe.
3779  *
3780  * 	    nullfs mounts of subdirectories are not safe.  That is, it will
3781  *	    work, but you do not really have protection against access to
3782  *	    the related parent directories.
3783  */
3784 int
3785 sys_getfh(struct getfh_args *uap)
3786 {
3787 	struct thread *td = curthread;
3788 	struct nlookupdata nd;
3789 	fhandle_t fh;
3790 	struct vnode *vp;
3791 	struct mount *mp;
3792 	int error;
3793 
3794 	/*
3795 	 * Must be super user
3796 	 */
3797 	if ((error = priv_check(td, PRIV_ROOT)) != 0)
3798 		return (error);
3799 
3800 	vp = NULL;
3801 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3802 	if (error == 0)
3803 		error = nlookup(&nd);
3804 	if (error == 0)
3805 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3806 	mp = nd.nl_nch.mount;
3807 	nlookup_done(&nd);
3808 	if (error == 0) {
3809 		bzero(&fh, sizeof(fh));
3810 		fh.fh_fsid = mp->mnt_stat.f_fsid;
3811 		error = VFS_VPTOFH(vp, &fh.fh_fid);
3812 		vput(vp);
3813 		if (error == 0)
3814 			error = copyout(&fh, uap->fhp, sizeof(fh));
3815 	}
3816 	return (error);
3817 }
3818 
3819 /*
3820  * fhopen_args(const struct fhandle *u_fhp, int flags)
3821  *
3822  * syscall for the rpc.lockd to use to translate a NFS file handle into
3823  * an open descriptor.
3824  *
3825  * warning: do not remove the priv_check() call or this becomes one giant
3826  * security hole.
3827  */
3828 int
3829 sys_fhopen(struct fhopen_args *uap)
3830 {
3831 	struct thread *td = curthread;
3832 	struct proc *p = td->td_proc;
3833 	struct mount *mp;
3834 	struct vnode *vp;
3835 	struct fhandle fhp;
3836 	struct vattr vat;
3837 	struct vattr *vap = &vat;
3838 	struct flock lf;
3839 	int fmode, mode, error, type;
3840 	struct file *nfp;
3841 	struct file *fp;
3842 	int indx;
3843 
3844 	/*
3845 	 * Must be super user
3846 	 */
3847 	error = priv_check(td, PRIV_ROOT);
3848 	if (error)
3849 		return (error);
3850 
3851 	fmode = FFLAGS(uap->flags);
3852 	/* why not allow a non-read/write open for our lockd? */
3853 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3854 		return (EINVAL);
3855 	error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3856 	if (error)
3857 		return(error);
3858 	/* find the mount point */
3859 	mp = vfs_getvfs(&fhp.fh_fsid);
3860 	if (mp == NULL)
3861 		return (ESTALE);
3862 	/* now give me my vnode, it gets returned to me locked */
3863 	error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp);
3864 	if (error)
3865 		return (error);
3866  	/*
3867 	 * from now on we have to make sure not
3868 	 * to forget about the vnode
3869 	 * any error that causes an abort must vput(vp)
3870 	 * just set error = err and 'goto bad;'.
3871 	 */
3872 
3873 	/*
3874 	 * from vn_open
3875 	 */
3876 	if (vp->v_type == VLNK) {
3877 		error = EMLINK;
3878 		goto bad;
3879 	}
3880 	if (vp->v_type == VSOCK) {
3881 		error = EOPNOTSUPP;
3882 		goto bad;
3883 	}
3884 	mode = 0;
3885 	if (fmode & (FWRITE | O_TRUNC)) {
3886 		if (vp->v_type == VDIR) {
3887 			error = EISDIR;
3888 			goto bad;
3889 		}
3890 		error = vn_writechk(vp, NULL);
3891 		if (error)
3892 			goto bad;
3893 		mode |= VWRITE;
3894 	}
3895 	if (fmode & FREAD)
3896 		mode |= VREAD;
3897 	if (mode) {
3898 		error = VOP_ACCESS(vp, mode, p->p_ucred);
3899 		if (error)
3900 			goto bad;
3901 	}
3902 	if (fmode & O_TRUNC) {
3903 		vn_unlock(vp);				/* XXX */
3904 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
3905 		VATTR_NULL(vap);
3906 		vap->va_size = 0;
3907 		error = VOP_SETATTR(vp, vap, p->p_ucred);
3908 		if (error)
3909 			goto bad;
3910 	}
3911 
3912 	/*
3913 	 * VOP_OPEN needs the file pointer so it can potentially override
3914 	 * it.
3915 	 *
3916 	 * WARNING! no f_nchandle will be associated when fhopen()ing a
3917 	 * directory.  XXX
3918 	 */
3919 	if ((error = falloc(p, &nfp, &indx)) != 0)
3920 		goto bad;
3921 	fp = nfp;
3922 
3923 	error = VOP_OPEN(vp, fmode, p->p_ucred, fp);
3924 	if (error) {
3925 		/*
3926 		 * setting f_ops this way prevents VOP_CLOSE from being
3927 		 * called or fdrop() releasing the vp from v_data.   Since
3928 		 * the VOP_OPEN failed we don't want to VOP_CLOSE.
3929 		 */
3930 		fp->f_ops = &badfileops;
3931 		fp->f_data = NULL;
3932 		goto bad_drop;
3933 	}
3934 
3935 	/*
3936 	 * The fp is given its own reference, we still have our ref and lock.
3937 	 *
3938 	 * Assert that all regular files must be created with a VM object.
3939 	 */
3940 	if (vp->v_type == VREG && vp->v_object == NULL) {
3941 		kprintf("fhopen: regular file did not have VM object: %p\n", vp);
3942 		goto bad_drop;
3943 	}
3944 
3945 	/*
3946 	 * The open was successful.  Handle any locking requirements.
3947 	 */
3948 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
3949 		lf.l_whence = SEEK_SET;
3950 		lf.l_start = 0;
3951 		lf.l_len = 0;
3952 		if (fmode & O_EXLOCK)
3953 			lf.l_type = F_WRLCK;
3954 		else
3955 			lf.l_type = F_RDLCK;
3956 		if (fmode & FNONBLOCK)
3957 			type = 0;
3958 		else
3959 			type = F_WAIT;
3960 		vn_unlock(vp);
3961 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3962 			/*
3963 			 * release our private reference.
3964 			 */
3965 			fsetfd(p, NULL, indx);
3966 			fdrop(fp);
3967 			vrele(vp);
3968 			return (error);
3969 		}
3970 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3971 		fp->f_flag |= FHASLOCK;
3972 	}
3973 
3974 	/*
3975 	 * Clean up.  Associate the file pointer with the previously
3976 	 * reserved descriptor and return it.
3977 	 */
3978 	vput(vp);
3979 	fsetfd(p, fp, indx);
3980 	fdrop(fp);
3981 	uap->sysmsg_result = indx;
3982 	return (0);
3983 
3984 bad_drop:
3985 	fsetfd(p, NULL, indx);
3986 	fdrop(fp);
3987 bad:
3988 	vput(vp);
3989 	return (error);
3990 }
3991 
3992 /*
3993  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3994  */
3995 int
3996 sys_fhstat(struct fhstat_args *uap)
3997 {
3998 	struct thread *td = curthread;
3999 	struct stat sb;
4000 	fhandle_t fh;
4001 	struct mount *mp;
4002 	struct vnode *vp;
4003 	int error;
4004 
4005 	/*
4006 	 * Must be super user
4007 	 */
4008 	error = priv_check(td, PRIV_ROOT);
4009 	if (error)
4010 		return (error);
4011 
4012 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
4013 	if (error)
4014 		return (error);
4015 
4016 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
4017 		return (ESTALE);
4018 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
4019 		return (error);
4020 	error = vn_stat(vp, &sb, td->td_proc->p_ucred);
4021 	vput(vp);
4022 	if (error)
4023 		return (error);
4024 	error = copyout(&sb, uap->sb, sizeof(sb));
4025 	return (error);
4026 }
4027 
4028 /*
4029  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
4030  */
4031 int
4032 sys_fhstatfs(struct fhstatfs_args *uap)
4033 {
4034 	struct thread *td = curthread;
4035 	struct proc *p = td->td_proc;
4036 	struct statfs *sp;
4037 	struct mount *mp;
4038 	struct vnode *vp;
4039 	struct statfs sb;
4040 	char *fullpath, *freepath;
4041 	fhandle_t fh;
4042 	int error;
4043 
4044 	/*
4045 	 * Must be super user
4046 	 */
4047 	if ((error = priv_check(td, PRIV_ROOT)))
4048 		return (error);
4049 
4050 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4051 		return (error);
4052 
4053 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
4054 		return (ESTALE);
4055 
4056 	if (p != NULL && !chroot_visible_mnt(mp, p))
4057 		return (ESTALE);
4058 
4059 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
4060 		return (error);
4061 	mp = vp->v_mount;
4062 	sp = &mp->mnt_stat;
4063 	vput(vp);
4064 	if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0)
4065 		return (error);
4066 
4067 	error = mount_path(p, mp, &fullpath, &freepath);
4068 	if (error)
4069 		return(error);
4070 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4071 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
4072 	kfree(freepath, M_TEMP);
4073 
4074 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
4075 	if (priv_check(td, PRIV_ROOT)) {
4076 		bcopy(sp, &sb, sizeof(sb));
4077 		sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
4078 		sp = &sb;
4079 	}
4080 	return (copyout(sp, uap->buf, sizeof(*sp)));
4081 }
4082 
4083 /*
4084  * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
4085  */
4086 int
4087 sys_fhstatvfs(struct fhstatvfs_args *uap)
4088 {
4089 	struct thread *td = curthread;
4090 	struct proc *p = td->td_proc;
4091 	struct statvfs *sp;
4092 	struct mount *mp;
4093 	struct vnode *vp;
4094 	fhandle_t fh;
4095 	int error;
4096 
4097 	/*
4098 	 * Must be super user
4099 	 */
4100 	if ((error = priv_check(td, PRIV_ROOT)))
4101 		return (error);
4102 
4103 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4104 		return (error);
4105 
4106 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
4107 		return (ESTALE);
4108 
4109 	if (p != NULL && !chroot_visible_mnt(mp, p))
4110 		return (ESTALE);
4111 
4112 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
4113 		return (error);
4114 	mp = vp->v_mount;
4115 	sp = &mp->mnt_vstat;
4116 	vput(vp);
4117 	if ((error = VFS_STATVFS(mp, sp, p->p_ucred)) != 0)
4118 		return (error);
4119 
4120 	sp->f_flag = 0;
4121 	if (mp->mnt_flag & MNT_RDONLY)
4122 		sp->f_flag |= ST_RDONLY;
4123 	if (mp->mnt_flag & MNT_NOSUID)
4124 		sp->f_flag |= ST_NOSUID;
4125 
4126 	return (copyout(sp, uap->buf, sizeof(*sp)));
4127 }
4128 
4129 
4130 /*
4131  * Syscall to push extended attribute configuration information into the
4132  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
4133  * a command (int cmd), and attribute name and misc data.  For now, the
4134  * attribute name is left in userspace for consumption by the VFS_op.
4135  * It will probably be changed to be copied into sysspace by the
4136  * syscall in the future, once issues with various consumers of the
4137  * attribute code have raised their hands.
4138  *
4139  * Currently this is used only by UFS Extended Attributes.
4140  */
4141 int
4142 sys_extattrctl(struct extattrctl_args *uap)
4143 {
4144 	struct nlookupdata nd;
4145 	struct mount *mp;
4146 	struct vnode *vp;
4147 	int error;
4148 
4149 	vp = NULL;
4150 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4151 	if (error == 0)
4152 		error = nlookup(&nd);
4153 	if (error == 0) {
4154 		mp = nd.nl_nch.mount;
4155 		error = VFS_EXTATTRCTL(mp, uap->cmd,
4156 				uap->attrname, uap->arg,
4157 				nd.nl_cred);
4158 	}
4159 	nlookup_done(&nd);
4160 	return (error);
4161 }
4162 
4163 /*
4164  * Syscall to set a named extended attribute on a file or directory.
4165  * Accepts attribute name, and a uio structure pointing to the data to set.
4166  * The uio is consumed in the style of writev().  The real work happens
4167  * in VOP_SETEXTATTR().
4168  */
4169 int
4170 sys_extattr_set_file(struct extattr_set_file_args *uap)
4171 {
4172 	char attrname[EXTATTR_MAXNAMELEN];
4173 	struct iovec aiov[UIO_SMALLIOV];
4174 	struct iovec *needfree;
4175 	struct nlookupdata nd;
4176 	struct iovec *iov;
4177 	struct vnode *vp;
4178 	struct uio auio;
4179 	u_int iovlen;
4180 	u_int cnt;
4181 	int error;
4182 	int i;
4183 
4184 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4185 	if (error)
4186 		return (error);
4187 
4188 	vp = NULL;
4189 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4190 	if (error == 0)
4191 		error = nlookup(&nd);
4192 	if (error == 0)
4193 		error = ncp_writechk(&nd.nl_nch);
4194 	if (error == 0)
4195 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4196 	if (error) {
4197 		nlookup_done(&nd);
4198 		return (error);
4199 	}
4200 
4201 	needfree = NULL;
4202 	iovlen = uap->iovcnt * sizeof(struct iovec);
4203 	if (uap->iovcnt > UIO_SMALLIOV) {
4204 		if (uap->iovcnt > UIO_MAXIOV) {
4205 			error = EINVAL;
4206 			goto done;
4207 		}
4208 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
4209 		needfree = iov;
4210 	} else {
4211 		iov = aiov;
4212 	}
4213 	auio.uio_iov = iov;
4214 	auio.uio_iovcnt = uap->iovcnt;
4215 	auio.uio_rw = UIO_WRITE;
4216 	auio.uio_segflg = UIO_USERSPACE;
4217 	auio.uio_td = nd.nl_td;
4218 	auio.uio_offset = 0;
4219 	if ((error = copyin(uap->iovp, iov, iovlen)))
4220 		goto done;
4221 	auio.uio_resid = 0;
4222 	for (i = 0; i < uap->iovcnt; i++) {
4223 		if (iov->iov_len > LONG_MAX - auio.uio_resid) {
4224 			error = EINVAL;
4225 			goto done;
4226 		}
4227 		auio.uio_resid += iov->iov_len;
4228 		iov++;
4229 	}
4230 	cnt = auio.uio_resid;
4231 	error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred);
4232 	cnt -= auio.uio_resid;
4233 	uap->sysmsg_result = cnt;
4234 done:
4235 	vput(vp);
4236 	nlookup_done(&nd);
4237 	if (needfree)
4238 		FREE(needfree, M_IOV);
4239 	return (error);
4240 }
4241 
4242 /*
4243  * Syscall to get a named extended attribute on a file or directory.
4244  * Accepts attribute name, and a uio structure pointing to a buffer for the
4245  * data.  The uio is consumed in the style of readv().  The real work
4246  * happens in VOP_GETEXTATTR();
4247  */
4248 int
4249 sys_extattr_get_file(struct extattr_get_file_args *uap)
4250 {
4251 	char attrname[EXTATTR_MAXNAMELEN];
4252 	struct iovec aiov[UIO_SMALLIOV];
4253 	struct iovec *needfree;
4254 	struct nlookupdata nd;
4255 	struct iovec *iov;
4256 	struct vnode *vp;
4257 	struct uio auio;
4258 	u_int iovlen;
4259 	u_int cnt;
4260 	int error;
4261 	int i;
4262 
4263 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4264 	if (error)
4265 		return (error);
4266 
4267 	vp = NULL;
4268 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4269 	if (error == 0)
4270 		error = nlookup(&nd);
4271 	if (error == 0)
4272 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4273 	if (error) {
4274 		nlookup_done(&nd);
4275 		return (error);
4276 	}
4277 
4278 	iovlen = uap->iovcnt * sizeof (struct iovec);
4279 	needfree = NULL;
4280 	if (uap->iovcnt > UIO_SMALLIOV) {
4281 		if (uap->iovcnt > UIO_MAXIOV) {
4282 			error = EINVAL;
4283 			goto done;
4284 		}
4285 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
4286 		needfree = iov;
4287 	} else {
4288 		iov = aiov;
4289 	}
4290 	auio.uio_iov = iov;
4291 	auio.uio_iovcnt = uap->iovcnt;
4292 	auio.uio_rw = UIO_READ;
4293 	auio.uio_segflg = UIO_USERSPACE;
4294 	auio.uio_td = nd.nl_td;
4295 	auio.uio_offset = 0;
4296 	if ((error = copyin(uap->iovp, iov, iovlen)))
4297 		goto done;
4298 	auio.uio_resid = 0;
4299 	for (i = 0; i < uap->iovcnt; i++) {
4300 		if (iov->iov_len > LONG_MAX - auio.uio_resid) {
4301 			error = EINVAL;
4302 			goto done;
4303 		}
4304 		auio.uio_resid += iov->iov_len;
4305 		iov++;
4306 	}
4307 	cnt = auio.uio_resid;
4308 	error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred);
4309 	cnt -= auio.uio_resid;
4310 	uap->sysmsg_result = cnt;
4311 done:
4312 	vput(vp);
4313 	nlookup_done(&nd);
4314 	if (needfree)
4315 		FREE(needfree, M_IOV);
4316 	return(error);
4317 }
4318 
4319 /*
4320  * Syscall to delete a named extended attribute from a file or directory.
4321  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
4322  */
4323 int
4324 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
4325 {
4326 	char attrname[EXTATTR_MAXNAMELEN];
4327 	struct nlookupdata nd;
4328 	struct vnode *vp;
4329 	int error;
4330 
4331 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4332 	if (error)
4333 		return(error);
4334 
4335 	vp = NULL;
4336 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4337 	if (error == 0)
4338 		error = nlookup(&nd);
4339 	if (error == 0)
4340 		error = ncp_writechk(&nd.nl_nch);
4341 	if (error == 0)
4342 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4343 	if (error) {
4344 		nlookup_done(&nd);
4345 		return (error);
4346 	}
4347 
4348 	error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred);
4349 	vput(vp);
4350 	nlookup_done(&nd);
4351 	return(error);
4352 }
4353 
4354 /*
4355  * Determine if the mount is visible to the process.
4356  */
4357 static int
4358 chroot_visible_mnt(struct mount *mp, struct proc *p)
4359 {
4360 	struct nchandle nch;
4361 
4362 	/*
4363 	 * Traverse from the mount point upwards.  If we hit the process
4364 	 * root then the mount point is visible to the process.
4365 	 */
4366 	nch = mp->mnt_ncmountpt;
4367 	while (nch.ncp) {
4368 		if (nch.mount == p->p_fd->fd_nrdir.mount &&
4369 		    nch.ncp == p->p_fd->fd_nrdir.ncp) {
4370 			return(1);
4371 		}
4372 		if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
4373 			nch = nch.mount->mnt_ncmounton;
4374 		} else {
4375 			nch.ncp = nch.ncp->nc_parent;
4376 		}
4377 	}
4378 
4379 	/*
4380 	 * If the mount point is not visible to the process, but the
4381 	 * process root is in a subdirectory of the mount, return
4382 	 * TRUE anyway.
4383 	 */
4384 	if (p->p_fd->fd_nrdir.mount == mp)
4385 		return(1);
4386 
4387 	return(0);
4388 }
4389 
4390