xref: /dflybsd-src/sys/kern/vfs_syscalls.c (revision 201c8c4447cad562e0a54ebbe0e7ee4e8a0be647)
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);
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 ((error = priv_check(td, PRIV_ROOT)) != 0)
912 		return (error);
913 
914 	/*
915 	 * Argument length checks
916 	 */
917 	if (uap->ctllen < 0 || uap->ctllen > 1024)
918 		return (EINVAL);
919 	if (uap->buflen < 0 || uap->buflen > 16 * 1024)
920 		return (EINVAL);
921 	if (uap->path == NULL)
922 		return (EINVAL);
923 
924 	/*
925 	 * Allocate the necessary buffers and copyin data
926 	 */
927 	path = objcache_get(namei_oc, M_WAITOK);
928 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
929 	if (error)
930 		goto done;
931 
932 	if (uap->ctllen) {
933 		ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
934 		error = copyin(uap->ctl, ctl, uap->ctllen);
935 		if (error)
936 			goto done;
937 	}
938 	if (uap->buflen)
939 		buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
940 
941 	/*
942 	 * Validate the descriptor
943 	 */
944 	if (uap->fd >= 0) {
945 		fp = holdfp(p->p_fd, uap->fd, -1);
946 		if (fp == NULL) {
947 			error = EBADF;
948 			goto done;
949 		}
950 	} else {
951 		fp = NULL;
952 	}
953 
954 	/*
955 	 * Execute the internal kernel function and clean up.
956 	 */
957 	error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
958 	if (fp)
959 		fdrop(fp);
960 	if (error == 0 && uap->sysmsg_result > 0)
961 		error = copyout(buf, uap->buf, uap->sysmsg_result);
962 done:
963 	if (path)
964 		objcache_put(namei_oc, path);
965 	if (ctl)
966 		kfree(ctl, M_TEMP);
967 	if (buf)
968 		kfree(buf, M_TEMP);
969 	return (error);
970 }
971 
972 /*
973  * Execute a mount control operation by resolving the path to a mount point
974  * and calling vop_mountctl().
975  *
976  * Use the mount point from the nch instead of the vnode so nullfs mounts
977  * can properly spike the VOP.
978  */
979 int
980 kern_mountctl(const char *path, int op, struct file *fp,
981 		const void *ctl, int ctllen,
982 		void *buf, int buflen, int *res)
983 {
984 	struct vnode *vp;
985 	struct mount *mp;
986 	struct nlookupdata nd;
987 	int error;
988 
989 	*res = 0;
990 	vp = NULL;
991 	error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
992 	if (error == 0)
993 		error = nlookup(&nd);
994 	if (error == 0)
995 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
996 	mp = nd.nl_nch.mount;
997 	nlookup_done(&nd);
998 	if (error)
999 		return (error);
1000 
1001 	/*
1002 	 * Must be the root of the filesystem
1003 	 */
1004 	if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
1005 		vput(vp);
1006 		return (EINVAL);
1007 	}
1008 	error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen,
1009 				buf, buflen, res);
1010 	vput(vp);
1011 	return (error);
1012 }
1013 
1014 int
1015 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1016 {
1017 	struct thread *td = curthread;
1018 	struct proc *p = td->td_proc;
1019 	struct mount *mp;
1020 	struct statfs *sp;
1021 	char *fullpath, *freepath;
1022 	int error;
1023 
1024 	if ((error = nlookup(nd)) != 0)
1025 		return (error);
1026 	mp = nd->nl_nch.mount;
1027 	sp = &mp->mnt_stat;
1028 	if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1029 		return (error);
1030 
1031 	error = mount_path(p, mp, &fullpath, &freepath);
1032 	if (error)
1033 		return(error);
1034 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1035 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1036 	kfree(freepath, M_TEMP);
1037 
1038 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1039 	bcopy(sp, buf, sizeof(*buf));
1040 	/* Only root should have access to the fsid's. */
1041 	if (priv_check(td, PRIV_ROOT))
1042 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1043 	return (0);
1044 }
1045 
1046 /*
1047  * statfs_args(char *path, struct statfs *buf)
1048  *
1049  * Get filesystem statistics.
1050  */
1051 int
1052 sys_statfs(struct statfs_args *uap)
1053 {
1054 	struct nlookupdata nd;
1055 	struct statfs buf;
1056 	int error;
1057 
1058 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1059 	if (error == 0)
1060 		error = kern_statfs(&nd, &buf);
1061 	nlookup_done(&nd);
1062 	if (error == 0)
1063 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1064 	return (error);
1065 }
1066 
1067 int
1068 kern_fstatfs(int fd, struct statfs *buf)
1069 {
1070 	struct thread *td = curthread;
1071 	struct proc *p = td->td_proc;
1072 	struct file *fp;
1073 	struct mount *mp;
1074 	struct statfs *sp;
1075 	char *fullpath, *freepath;
1076 	int error;
1077 
1078 	KKASSERT(p);
1079 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1080 		return (error);
1081 	mp = ((struct vnode *)fp->f_data)->v_mount;
1082 	if (mp == NULL) {
1083 		error = EBADF;
1084 		goto done;
1085 	}
1086 	if (fp->f_cred == NULL) {
1087 		error = EINVAL;
1088 		goto done;
1089 	}
1090 	sp = &mp->mnt_stat;
1091 	if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1092 		goto done;
1093 
1094 	if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1095 		goto done;
1096 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1097 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1098 	kfree(freepath, M_TEMP);
1099 
1100 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1101 	bcopy(sp, buf, sizeof(*buf));
1102 
1103 	/* Only root should have access to the fsid's. */
1104 	if (priv_check(td, PRIV_ROOT))
1105 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1106 	error = 0;
1107 done:
1108 	fdrop(fp);
1109 	return (error);
1110 }
1111 
1112 /*
1113  * fstatfs_args(int fd, struct statfs *buf)
1114  *
1115  * Get filesystem statistics.
1116  */
1117 int
1118 sys_fstatfs(struct fstatfs_args *uap)
1119 {
1120 	struct statfs buf;
1121 	int error;
1122 
1123 	error = kern_fstatfs(uap->fd, &buf);
1124 
1125 	if (error == 0)
1126 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1127 	return (error);
1128 }
1129 
1130 int
1131 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1132 {
1133 	struct mount *mp;
1134 	struct statvfs *sp;
1135 	int error;
1136 
1137 	if ((error = nlookup(nd)) != 0)
1138 		return (error);
1139 	mp = nd->nl_nch.mount;
1140 	sp = &mp->mnt_vstat;
1141 	if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1142 		return (error);
1143 
1144 	sp->f_flag = 0;
1145 	if (mp->mnt_flag & MNT_RDONLY)
1146 		sp->f_flag |= ST_RDONLY;
1147 	if (mp->mnt_flag & MNT_NOSUID)
1148 		sp->f_flag |= ST_NOSUID;
1149 	bcopy(sp, buf, sizeof(*buf));
1150 	return (0);
1151 }
1152 
1153 /*
1154  * statfs_args(char *path, struct statfs *buf)
1155  *
1156  * Get filesystem statistics.
1157  */
1158 int
1159 sys_statvfs(struct statvfs_args *uap)
1160 {
1161 	struct nlookupdata nd;
1162 	struct statvfs buf;
1163 	int error;
1164 
1165 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1166 	if (error == 0)
1167 		error = kern_statvfs(&nd, &buf);
1168 	nlookup_done(&nd);
1169 	if (error == 0)
1170 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1171 	return (error);
1172 }
1173 
1174 int
1175 kern_fstatvfs(int fd, struct statvfs *buf)
1176 {
1177 	struct thread *td = curthread;
1178 	struct proc *p = td->td_proc;
1179 	struct file *fp;
1180 	struct mount *mp;
1181 	struct statvfs *sp;
1182 	int error;
1183 
1184 	KKASSERT(p);
1185 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1186 		return (error);
1187 	mp = ((struct vnode *)fp->f_data)->v_mount;
1188 	if (mp == NULL) {
1189 		error = EBADF;
1190 		goto done;
1191 	}
1192 	if (fp->f_cred == NULL) {
1193 		error = EINVAL;
1194 		goto done;
1195 	}
1196 	sp = &mp->mnt_vstat;
1197 	if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1198 		goto done;
1199 
1200 	sp->f_flag = 0;
1201 	if (mp->mnt_flag & MNT_RDONLY)
1202 		sp->f_flag |= ST_RDONLY;
1203 	if (mp->mnt_flag & MNT_NOSUID)
1204 		sp->f_flag |= ST_NOSUID;
1205 
1206 	bcopy(sp, buf, sizeof(*buf));
1207 	error = 0;
1208 done:
1209 	fdrop(fp);
1210 	return (error);
1211 }
1212 
1213 /*
1214  * fstatfs_args(int fd, struct statfs *buf)
1215  *
1216  * Get filesystem statistics.
1217  */
1218 int
1219 sys_fstatvfs(struct fstatvfs_args *uap)
1220 {
1221 	struct statvfs buf;
1222 	int error;
1223 
1224 	error = kern_fstatvfs(uap->fd, &buf);
1225 
1226 	if (error == 0)
1227 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1228 	return (error);
1229 }
1230 
1231 /*
1232  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1233  *
1234  * Get statistics on all filesystems.
1235  */
1236 
1237 struct getfsstat_info {
1238 	struct statfs *sfsp;
1239 	long count;
1240 	long maxcount;
1241 	int error;
1242 	int flags;
1243 	struct proc *p;
1244 };
1245 
1246 static int getfsstat_callback(struct mount *, void *);
1247 
1248 /* ARGSUSED */
1249 int
1250 sys_getfsstat(struct getfsstat_args *uap)
1251 {
1252 	struct thread *td = curthread;
1253 	struct proc *p = td->td_proc;
1254 	struct getfsstat_info info;
1255 
1256 	bzero(&info, sizeof(info));
1257 
1258 	info.maxcount = uap->bufsize / sizeof(struct statfs);
1259 	info.sfsp = uap->buf;
1260 	info.count = 0;
1261 	info.flags = uap->flags;
1262 	info.p = p;
1263 
1264 	mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1265 	if (info.sfsp && info.count > info.maxcount)
1266 		uap->sysmsg_result = info.maxcount;
1267 	else
1268 		uap->sysmsg_result = info.count;
1269 	return (info.error);
1270 }
1271 
1272 static int
1273 getfsstat_callback(struct mount *mp, void *data)
1274 {
1275 	struct getfsstat_info *info = data;
1276 	struct statfs *sp;
1277 	char *freepath;
1278 	char *fullpath;
1279 	int error;
1280 
1281 	if (info->sfsp && info->count < info->maxcount) {
1282 		if (info->p && !chroot_visible_mnt(mp, info->p))
1283 			return(0);
1284 		sp = &mp->mnt_stat;
1285 
1286 		/*
1287 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1288 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1289 		 * overrides MNT_WAIT.
1290 		 */
1291 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1292 		    (info->flags & MNT_WAIT)) &&
1293 		    (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1294 			return(0);
1295 		}
1296 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1297 
1298 		error = mount_path(info->p, mp, &fullpath, &freepath);
1299 		if (error) {
1300 			info->error = error;
1301 			return(-1);
1302 		}
1303 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1304 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1305 		kfree(freepath, M_TEMP);
1306 
1307 		error = copyout(sp, info->sfsp, sizeof(*sp));
1308 		if (error) {
1309 			info->error = error;
1310 			return (-1);
1311 		}
1312 		++info->sfsp;
1313 	}
1314 	info->count++;
1315 	return(0);
1316 }
1317 
1318 /*
1319  * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1320 		   long bufsize, int flags)
1321  *
1322  * Get statistics on all filesystems.
1323  */
1324 
1325 struct getvfsstat_info {
1326 	struct statfs *sfsp;
1327 	struct statvfs *vsfsp;
1328 	long count;
1329 	long maxcount;
1330 	int error;
1331 	int flags;
1332 	struct proc *p;
1333 };
1334 
1335 static int getvfsstat_callback(struct mount *, void *);
1336 
1337 /* ARGSUSED */
1338 int
1339 sys_getvfsstat(struct getvfsstat_args *uap)
1340 {
1341 	struct thread *td = curthread;
1342 	struct proc *p = td->td_proc;
1343 	struct getvfsstat_info info;
1344 
1345 	bzero(&info, sizeof(info));
1346 
1347 	info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1348 	info.sfsp = uap->buf;
1349 	info.vsfsp = uap->vbuf;
1350 	info.count = 0;
1351 	info.flags = uap->flags;
1352 	info.p = p;
1353 
1354 	mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1355 	if (info.vsfsp && info.count > info.maxcount)
1356 		uap->sysmsg_result = info.maxcount;
1357 	else
1358 		uap->sysmsg_result = info.count;
1359 	return (info.error);
1360 }
1361 
1362 static int
1363 getvfsstat_callback(struct mount *mp, void *data)
1364 {
1365 	struct getvfsstat_info *info = data;
1366 	struct statfs *sp;
1367 	struct statvfs *vsp;
1368 	char *freepath;
1369 	char *fullpath;
1370 	int error;
1371 
1372 	if (info->vsfsp && info->count < info->maxcount) {
1373 		if (info->p && !chroot_visible_mnt(mp, info->p))
1374 			return(0);
1375 		sp = &mp->mnt_stat;
1376 		vsp = &mp->mnt_vstat;
1377 
1378 		/*
1379 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1380 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1381 		 * overrides MNT_WAIT.
1382 		 */
1383 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1384 		    (info->flags & MNT_WAIT)) &&
1385 		    (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1386 			return(0);
1387 		}
1388 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1389 
1390 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1391 		    (info->flags & MNT_WAIT)) &&
1392 		    (error = VFS_STATVFS(mp, vsp, info->p->p_ucred))) {
1393 			return(0);
1394 		}
1395 		vsp->f_flag = 0;
1396 		if (mp->mnt_flag & MNT_RDONLY)
1397 			vsp->f_flag |= ST_RDONLY;
1398 		if (mp->mnt_flag & MNT_NOSUID)
1399 			vsp->f_flag |= ST_NOSUID;
1400 
1401 		error = mount_path(info->p, mp, &fullpath, &freepath);
1402 		if (error) {
1403 			info->error = error;
1404 			return(-1);
1405 		}
1406 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1407 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1408 		kfree(freepath, M_TEMP);
1409 
1410 		error = copyout(sp, info->sfsp, sizeof(*sp));
1411 		if (error == 0)
1412 			error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1413 		if (error) {
1414 			info->error = error;
1415 			return (-1);
1416 		}
1417 		++info->sfsp;
1418 		++info->vsfsp;
1419 	}
1420 	info->count++;
1421 	return(0);
1422 }
1423 
1424 
1425 /*
1426  * fchdir_args(int fd)
1427  *
1428  * Change current working directory to a given file descriptor.
1429  */
1430 /* ARGSUSED */
1431 int
1432 sys_fchdir(struct fchdir_args *uap)
1433 {
1434 	struct thread *td = curthread;
1435 	struct proc *p = td->td_proc;
1436 	struct filedesc *fdp = p->p_fd;
1437 	struct vnode *vp, *ovp;
1438 	struct mount *mp;
1439 	struct file *fp;
1440 	struct nchandle nch, onch, tnch;
1441 	int error;
1442 
1443 	if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1444 		return (error);
1445 	vp = (struct vnode *)fp->f_data;
1446 	vref(vp);
1447 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1448 	if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL)
1449 		error = ENOTDIR;
1450 	else
1451 		error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
1452 	if (error) {
1453 		vput(vp);
1454 		fdrop(fp);
1455 		return (error);
1456 	}
1457 	cache_copy(&fp->f_nchandle, &nch);
1458 
1459 	/*
1460 	 * If the ncp has become a mount point, traverse through
1461 	 * the mount point.
1462 	 */
1463 
1464 	while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1465 	       (mp = cache_findmount(&nch)) != NULL
1466 	) {
1467 		error = nlookup_mp(mp, &tnch);
1468 		if (error == 0) {
1469 			cache_unlock(&tnch);	/* leave ref intact */
1470 			vput(vp);
1471 			vp = tnch.ncp->nc_vp;
1472 			error = vget(vp, LK_SHARED);
1473 			KKASSERT(error == 0);
1474 			cache_drop(&nch);
1475 			nch = tnch;
1476 		}
1477 	}
1478 	if (error == 0) {
1479 		ovp = fdp->fd_cdir;
1480 		onch = fdp->fd_ncdir;
1481 		vn_unlock(vp);		/* leave ref intact */
1482 		fdp->fd_cdir = vp;
1483 		fdp->fd_ncdir = nch;
1484 		cache_drop(&onch);
1485 		vrele(ovp);
1486 	} else {
1487 		cache_drop(&nch);
1488 		vput(vp);
1489 	}
1490 	fdrop(fp);
1491 	return (error);
1492 }
1493 
1494 int
1495 kern_chdir(struct nlookupdata *nd)
1496 {
1497 	struct thread *td = curthread;
1498 	struct proc *p = td->td_proc;
1499 	struct filedesc *fdp = p->p_fd;
1500 	struct vnode *vp, *ovp;
1501 	struct nchandle onch;
1502 	int error;
1503 
1504 	if ((error = nlookup(nd)) != 0)
1505 		return (error);
1506 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1507 		return (ENOENT);
1508 	if ((error = vget(vp, LK_SHARED)) != 0)
1509 		return (error);
1510 
1511 	error = checkvp_chdir(vp, td);
1512 	vn_unlock(vp);
1513 	if (error == 0) {
1514 		ovp = fdp->fd_cdir;
1515 		onch = fdp->fd_ncdir;
1516 		cache_unlock(&nd->nl_nch);	/* leave reference intact */
1517 		fdp->fd_ncdir = nd->nl_nch;
1518 		fdp->fd_cdir = vp;
1519 		cache_drop(&onch);
1520 		vrele(ovp);
1521 		cache_zero(&nd->nl_nch);
1522 	} else {
1523 		vrele(vp);
1524 	}
1525 	return (error);
1526 }
1527 
1528 /*
1529  * chdir_args(char *path)
1530  *
1531  * Change current working directory (``.'').
1532  */
1533 int
1534 sys_chdir(struct chdir_args *uap)
1535 {
1536 	struct nlookupdata nd;
1537 	int error;
1538 
1539 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1540 	if (error == 0)
1541 		error = kern_chdir(&nd);
1542 	nlookup_done(&nd);
1543 	return (error);
1544 }
1545 
1546 /*
1547  * Helper function for raised chroot(2) security function:  Refuse if
1548  * any filedescriptors are open directories.
1549  */
1550 static int
1551 chroot_refuse_vdir_fds(struct filedesc *fdp)
1552 {
1553 	struct vnode *vp;
1554 	struct file *fp;
1555 	int error;
1556 	int fd;
1557 
1558 	for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1559 		if ((error = holdvnode(fdp, fd, &fp)) != 0)
1560 			continue;
1561 		vp = (struct vnode *)fp->f_data;
1562 		if (vp->v_type != VDIR) {
1563 			fdrop(fp);
1564 			continue;
1565 		}
1566 		fdrop(fp);
1567 		return(EPERM);
1568 	}
1569 	return (0);
1570 }
1571 
1572 /*
1573  * This sysctl determines if we will allow a process to chroot(2) if it
1574  * has a directory open:
1575  *	0: disallowed for all processes.
1576  *	1: allowed for processes that were not already chroot(2)'ed.
1577  *	2: allowed for all processes.
1578  */
1579 
1580 static int chroot_allow_open_directories = 1;
1581 
1582 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1583      &chroot_allow_open_directories, 0, "");
1584 
1585 /*
1586  * chroot to the specified namecache entry.  We obtain the vp from the
1587  * namecache data.  The passed ncp must be locked and referenced and will
1588  * remain locked and referenced on return.
1589  */
1590 int
1591 kern_chroot(struct nchandle *nch)
1592 {
1593 	struct thread *td = curthread;
1594 	struct proc *p = td->td_proc;
1595 	struct filedesc *fdp = p->p_fd;
1596 	struct vnode *vp;
1597 	int error;
1598 
1599 	/*
1600 	 * Only privileged user can chroot
1601 	 */
1602 	error = priv_check_cred(p->p_ucred, PRIV_VFS_CHROOT, 0);
1603 	if (error)
1604 		return (error);
1605 
1606 	/*
1607 	 * Disallow open directory descriptors (fchdir() breakouts).
1608 	 */
1609 	if (chroot_allow_open_directories == 0 ||
1610 	   (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1611 		if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1612 			return (error);
1613 	}
1614 	if ((vp = nch->ncp->nc_vp) == NULL)
1615 		return (ENOENT);
1616 
1617 	if ((error = vget(vp, LK_SHARED)) != 0)
1618 		return (error);
1619 
1620 	/*
1621 	 * Check the validity of vp as a directory to change to and
1622 	 * associate it with rdir/jdir.
1623 	 */
1624 	error = checkvp_chdir(vp, td);
1625 	vn_unlock(vp);			/* leave reference intact */
1626 	if (error == 0) {
1627 		vrele(fdp->fd_rdir);
1628 		fdp->fd_rdir = vp;	/* reference inherited by fd_rdir */
1629 		cache_drop(&fdp->fd_nrdir);
1630 		cache_copy(nch, &fdp->fd_nrdir);
1631 		if (fdp->fd_jdir == NULL) {
1632 			fdp->fd_jdir = vp;
1633 			vref(fdp->fd_jdir);
1634 			cache_copy(nch, &fdp->fd_njdir);
1635 		}
1636 	} else {
1637 		vrele(vp);
1638 	}
1639 	return (error);
1640 }
1641 
1642 /*
1643  * chroot_args(char *path)
1644  *
1645  * Change notion of root (``/'') directory.
1646  */
1647 /* ARGSUSED */
1648 int
1649 sys_chroot(struct chroot_args *uap)
1650 {
1651 	struct thread *td = curthread;
1652 	struct nlookupdata nd;
1653 	int error;
1654 
1655 	KKASSERT(td->td_proc);
1656 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1657 	if (error) {
1658 		nlookup_done(&nd);
1659 		return(error);
1660 	}
1661 	nd.nl_flags |= NLC_EXEC;
1662 	error = nlookup(&nd);
1663 	if (error == 0)
1664 		error = kern_chroot(&nd.nl_nch);
1665 	nlookup_done(&nd);
1666 	return(error);
1667 }
1668 
1669 /*
1670  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1671  * determine whether it is legal to chdir to the vnode.  The vnode's state
1672  * is not changed by this call.
1673  */
1674 int
1675 checkvp_chdir(struct vnode *vp, struct thread *td)
1676 {
1677 	int error;
1678 
1679 	if (vp->v_type != VDIR)
1680 		error = ENOTDIR;
1681 	else
1682 		error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred);
1683 	return (error);
1684 }
1685 
1686 int
1687 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1688 {
1689 	struct thread *td = curthread;
1690 	struct proc *p = td->td_proc;
1691 	struct lwp *lp = td->td_lwp;
1692 	struct filedesc *fdp = p->p_fd;
1693 	int cmode, flags;
1694 	struct file *nfp;
1695 	struct file *fp;
1696 	struct vnode *vp;
1697 	int type, indx, error;
1698 	struct flock lf;
1699 
1700 	if ((oflags & O_ACCMODE) == O_ACCMODE)
1701 		return (EINVAL);
1702 	flags = FFLAGS(oflags);
1703 	error = falloc(p, &nfp, NULL);
1704 	if (error)
1705 		return (error);
1706 	fp = nfp;
1707 	cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
1708 
1709 	/*
1710 	 * XXX p_dupfd is a real mess.  It allows a device to return a
1711 	 * file descriptor to be duplicated rather then doing the open
1712 	 * itself.
1713 	 */
1714 	lp->lwp_dupfd = -1;
1715 
1716 	/*
1717 	 * Call vn_open() to do the lookup and assign the vnode to the
1718 	 * file pointer.  vn_open() does not change the ref count on fp
1719 	 * and the vnode, on success, will be inherited by the file pointer
1720 	 * and unlocked.
1721 	 */
1722 	nd->nl_flags |= NLC_LOCKVP;
1723 	error = vn_open(nd, fp, flags, cmode);
1724 	nlookup_done(nd);
1725 	if (error) {
1726 		/*
1727 		 * handle special fdopen() case.  bleh.  dupfdopen() is
1728 		 * responsible for dropping the old contents of ofiles[indx]
1729 		 * if it succeeds.
1730 		 *
1731 		 * Note that fsetfd() will add a ref to fp which represents
1732 		 * the fd_files[] assignment.  We must still drop our
1733 		 * reference.
1734 		 */
1735 		if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1736 			if (fdalloc(p, 0, &indx) == 0) {
1737 				error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error);
1738 				if (error == 0) {
1739 					*res = indx;
1740 					fdrop(fp);	/* our ref */
1741 					return (0);
1742 				}
1743 				fsetfd(p, NULL, indx);
1744 			}
1745 		}
1746 		fdrop(fp);	/* our ref */
1747 		if (error == ERESTART)
1748 			error = EINTR;
1749 		return (error);
1750 	}
1751 
1752 	/*
1753 	 * ref the vnode for ourselves so it can't be ripped out from under
1754 	 * is.  XXX need an ND flag to request that the vnode be returned
1755 	 * anyway.
1756 	 *
1757 	 * Reserve a file descriptor but do not assign it until the open
1758 	 * succeeds.
1759 	 */
1760 	vp = (struct vnode *)fp->f_data;
1761 	vref(vp);
1762 	if ((error = fdalloc(p, 0, &indx)) != 0) {
1763 		fdrop(fp);
1764 		vrele(vp);
1765 		return (error);
1766 	}
1767 
1768 	/*
1769 	 * If no error occurs the vp will have been assigned to the file
1770 	 * pointer.
1771 	 */
1772 	lp->lwp_dupfd = 0;
1773 
1774 	if (flags & (O_EXLOCK | O_SHLOCK)) {
1775 		lf.l_whence = SEEK_SET;
1776 		lf.l_start = 0;
1777 		lf.l_len = 0;
1778 		if (flags & O_EXLOCK)
1779 			lf.l_type = F_WRLCK;
1780 		else
1781 			lf.l_type = F_RDLCK;
1782 		if (flags & FNONBLOCK)
1783 			type = 0;
1784 		else
1785 			type = F_WAIT;
1786 
1787 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1788 			/*
1789 			 * lock request failed.  Clean up the reserved
1790 			 * descriptor.
1791 			 */
1792 			vrele(vp);
1793 			fsetfd(p, NULL, indx);
1794 			fdrop(fp);
1795 			return (error);
1796 		}
1797 		fp->f_flag |= FHASLOCK;
1798 	}
1799 #if 0
1800 	/*
1801 	 * Assert that all regular file vnodes were created with a object.
1802 	 */
1803 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1804 		("open: regular file has no backing object after vn_open"));
1805 #endif
1806 
1807 	vrele(vp);
1808 
1809 	/*
1810 	 * release our private reference, leaving the one associated with the
1811 	 * descriptor table intact.
1812 	 */
1813 	fsetfd(p, fp, indx);
1814 	fdrop(fp);
1815 	*res = indx;
1816 	return (0);
1817 }
1818 
1819 /*
1820  * open_args(char *path, int flags, int mode)
1821  *
1822  * Check permissions, allocate an open file structure,
1823  * and call the device open routine if any.
1824  */
1825 int
1826 sys_open(struct open_args *uap)
1827 {
1828 	struct nlookupdata nd;
1829 	int error;
1830 
1831 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1832 	if (error == 0) {
1833 		error = kern_open(&nd, uap->flags,
1834 				    uap->mode, &uap->sysmsg_result);
1835 	}
1836 	nlookup_done(&nd);
1837 	return (error);
1838 }
1839 
1840 int
1841 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
1842 {
1843 	struct thread *td = curthread;
1844 	struct proc *p = td->td_proc;
1845 	struct vnode *vp;
1846 	struct vattr vattr;
1847 	int error;
1848 	int whiteout = 0;
1849 
1850 	KKASSERT(p);
1851 
1852 	VATTR_NULL(&vattr);
1853 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1854 	vattr.va_rmajor = rmajor;
1855 	vattr.va_rminor = rminor;
1856 
1857 	switch (mode & S_IFMT) {
1858 	case S_IFMT:	/* used by badsect to flag bad sectors */
1859 		error = priv_check_cred(p->p_ucred, PRIV_VFS_MKNOD_BAD, 0);
1860 		vattr.va_type = VBAD;
1861 		break;
1862 	case S_IFCHR:
1863 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
1864 		vattr.va_type = VCHR;
1865 		break;
1866 	case S_IFBLK:
1867 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
1868 		vattr.va_type = VBLK;
1869 		break;
1870 	case S_IFWHT:
1871 		error = priv_check_cred(p->p_ucred, PRIV_VFS_MKNOD_WHT, 0);
1872 		whiteout = 1;
1873 		break;
1874 	case S_IFDIR:	/* special directories support for HAMMER */
1875 		error = priv_check_cred(p->p_ucred, PRIV_VFS_MKNOD_DIR, 0);
1876 		vattr.va_type = VDIR;
1877 		break;
1878 	default:
1879 		error = EINVAL;
1880 		break;
1881 	}
1882 
1883 	if (error)
1884 		return (error);
1885 
1886 	bwillinode(1);
1887 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1888 	if ((error = nlookup(nd)) != 0)
1889 		return (error);
1890 	if (nd->nl_nch.ncp->nc_vp)
1891 		return (EEXIST);
1892 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1893 		return (error);
1894 
1895 	if (whiteout) {
1896 		error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
1897 				      nd->nl_cred, NAMEI_CREATE);
1898 	} else {
1899 		vp = NULL;
1900 		error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
1901 				   &vp, nd->nl_cred, &vattr);
1902 		if (error == 0)
1903 			vput(vp);
1904 	}
1905 	return (error);
1906 }
1907 
1908 /*
1909  * mknod_args(char *path, int mode, int dev)
1910  *
1911  * Create a special file.
1912  */
1913 int
1914 sys_mknod(struct mknod_args *uap)
1915 {
1916 	struct nlookupdata nd;
1917 	int error;
1918 
1919 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1920 	if (error == 0) {
1921 		error = kern_mknod(&nd, uap->mode,
1922 				   umajor(uap->dev), uminor(uap->dev));
1923 	}
1924 	nlookup_done(&nd);
1925 	return (error);
1926 }
1927 
1928 int
1929 kern_mkfifo(struct nlookupdata *nd, int mode)
1930 {
1931 	struct thread *td = curthread;
1932 	struct proc *p = td->td_proc;
1933 	struct vattr vattr;
1934 	struct vnode *vp;
1935 	int error;
1936 
1937 	bwillinode(1);
1938 
1939 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1940 	if ((error = nlookup(nd)) != 0)
1941 		return (error);
1942 	if (nd->nl_nch.ncp->nc_vp)
1943 		return (EEXIST);
1944 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1945 		return (error);
1946 
1947 	VATTR_NULL(&vattr);
1948 	vattr.va_type = VFIFO;
1949 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1950 	vp = NULL;
1951 	error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
1952 	if (error == 0)
1953 		vput(vp);
1954 	return (error);
1955 }
1956 
1957 /*
1958  * mkfifo_args(char *path, int mode)
1959  *
1960  * Create a named pipe.
1961  */
1962 int
1963 sys_mkfifo(struct mkfifo_args *uap)
1964 {
1965 	struct nlookupdata nd;
1966 	int error;
1967 
1968 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1969 	if (error == 0)
1970 		error = kern_mkfifo(&nd, uap->mode);
1971 	nlookup_done(&nd);
1972 	return (error);
1973 }
1974 
1975 static int hardlink_check_uid = 0;
1976 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1977     &hardlink_check_uid, 0,
1978     "Unprivileged processes cannot create hard links to files owned by other "
1979     "users");
1980 static int hardlink_check_gid = 0;
1981 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
1982     &hardlink_check_gid, 0,
1983     "Unprivileged processes cannot create hard links to files owned by other "
1984     "groups");
1985 
1986 static int
1987 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
1988 {
1989 	struct vattr va;
1990 	int error;
1991 
1992 	/*
1993 	 * Shortcut if disabled
1994 	 */
1995 	if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
1996 		return (0);
1997 
1998 	/*
1999 	 * root cred can always hardlink
2000 	 */
2001 	if (priv_check_cred(cred, PRIV_ROOT, PRISON_ROOT) == 0)
2002 		return (0);
2003 
2004 	/*
2005 	 * Otherwise only if the originating file is owned by the
2006 	 * same user or group.  Note that any group is allowed if
2007 	 * the file is owned by the caller.
2008 	 */
2009 	error = VOP_GETATTR(vp, &va);
2010 	if (error != 0)
2011 		return (error);
2012 
2013 	if (hardlink_check_uid) {
2014 		if (cred->cr_uid != va.va_uid)
2015 			return (EPERM);
2016 	}
2017 
2018 	if (hardlink_check_gid) {
2019 		if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2020 			return (EPERM);
2021 	}
2022 
2023 	return (0);
2024 }
2025 
2026 int
2027 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2028 {
2029 	struct thread *td = curthread;
2030 	struct vnode *vp;
2031 	int error;
2032 
2033 	/*
2034 	 * Lookup the source and obtained a locked vnode.
2035 	 *
2036 	 * You may only hardlink a file which you have write permission
2037 	 * on or which you own.
2038 	 *
2039 	 * XXX relookup on vget failure / race ?
2040 	 */
2041 	bwillinode(1);
2042 	nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK;
2043 	if ((error = nlookup(nd)) != 0)
2044 		return (error);
2045 	vp = nd->nl_nch.ncp->nc_vp;
2046 	KKASSERT(vp != NULL);
2047 	if (vp->v_type == VDIR)
2048 		return (EPERM);		/* POSIX */
2049 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2050 		return (error);
2051 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2052 		return (error);
2053 
2054 	/*
2055 	 * Unlock the source so we can lookup the target without deadlocking
2056 	 * (XXX vp is locked already, possible other deadlock?).  The target
2057 	 * must not exist.
2058 	 */
2059 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2060 	nd->nl_flags &= ~NLC_NCPISLOCKED;
2061 	cache_unlock(&nd->nl_nch);
2062 
2063 	linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2064 	if ((error = nlookup(linknd)) != 0) {
2065 		vput(vp);
2066 		return (error);
2067 	}
2068 	if (linknd->nl_nch.ncp->nc_vp) {
2069 		vput(vp);
2070 		return (EEXIST);
2071 	}
2072 
2073 	/*
2074 	 * Finally run the new API VOP.
2075 	 */
2076 	error = can_hardlink(vp, td, td->td_proc->p_ucred);
2077 	if (error == 0) {
2078 		error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2079 				  vp, linknd->nl_cred);
2080 	}
2081 	vput(vp);
2082 	return (error);
2083 }
2084 
2085 /*
2086  * link_args(char *path, char *link)
2087  *
2088  * Make a hard file link.
2089  */
2090 int
2091 sys_link(struct link_args *uap)
2092 {
2093 	struct nlookupdata nd, linknd;
2094 	int error;
2095 
2096 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2097 	if (error == 0) {
2098 		error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2099 		if (error == 0)
2100 			error = kern_link(&nd, &linknd);
2101 		nlookup_done(&linknd);
2102 	}
2103 	nlookup_done(&nd);
2104 	return (error);
2105 }
2106 
2107 int
2108 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2109 {
2110 	struct vattr vattr;
2111 	struct vnode *vp;
2112 	struct vnode *dvp;
2113 	int error;
2114 
2115 	bwillinode(1);
2116 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2117 	if ((error = nlookup(nd)) != 0)
2118 		return (error);
2119 	if (nd->nl_nch.ncp->nc_vp)
2120 		return (EEXIST);
2121 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2122 		return (error);
2123 	dvp = nd->nl_dvp;
2124 	VATTR_NULL(&vattr);
2125 	vattr.va_mode = mode;
2126 	error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2127 	if (error == 0)
2128 		vput(vp);
2129 	return (error);
2130 }
2131 
2132 /*
2133  * symlink(char *path, char *link)
2134  *
2135  * Make a symbolic link.
2136  */
2137 int
2138 sys_symlink(struct symlink_args *uap)
2139 {
2140 	struct thread *td = curthread;
2141 	struct nlookupdata nd;
2142 	char *path;
2143 	int error;
2144 	int mode;
2145 
2146 	path = objcache_get(namei_oc, M_WAITOK);
2147 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2148 	if (error == 0) {
2149 		error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2150 		if (error == 0) {
2151 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2152 			error = kern_symlink(&nd, path, mode);
2153 		}
2154 		nlookup_done(&nd);
2155 	}
2156 	objcache_put(namei_oc, path);
2157 	return (error);
2158 }
2159 
2160 /*
2161  * undelete_args(char *path)
2162  *
2163  * Delete a whiteout from the filesystem.
2164  */
2165 /* ARGSUSED */
2166 int
2167 sys_undelete(struct undelete_args *uap)
2168 {
2169 	struct nlookupdata nd;
2170 	int error;
2171 
2172 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2173 	bwillinode(1);
2174 	nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2175 	if (error == 0)
2176 		error = nlookup(&nd);
2177 	if (error == 0)
2178 		error = ncp_writechk(&nd.nl_nch);
2179 	if (error == 0) {
2180 		error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2181 				      NAMEI_DELETE);
2182 	}
2183 	nlookup_done(&nd);
2184 	return (error);
2185 }
2186 
2187 int
2188 kern_unlink(struct nlookupdata *nd)
2189 {
2190 	int error;
2191 
2192 	bwillinode(1);
2193 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2194 	if ((error = nlookup(nd)) != 0)
2195 		return (error);
2196 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2197 		return (error);
2198 	error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2199 	return (error);
2200 }
2201 
2202 /*
2203  * unlink_args(char *path)
2204  *
2205  * Delete a name from the filesystem.
2206  */
2207 int
2208 sys_unlink(struct unlink_args *uap)
2209 {
2210 	struct nlookupdata nd;
2211 	int error;
2212 
2213 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2214 	if (error == 0)
2215 		error = kern_unlink(&nd);
2216 	nlookup_done(&nd);
2217 	return (error);
2218 }
2219 
2220 int
2221 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2222 {
2223 	struct thread *td = curthread;
2224 	struct proc *p = td->td_proc;
2225 	struct file *fp;
2226 	struct vnode *vp;
2227 	struct vattr vattr;
2228 	off_t new_offset;
2229 	int error;
2230 
2231 	fp = holdfp(p->p_fd, fd, -1);
2232 	if (fp == NULL)
2233 		return (EBADF);
2234 	if (fp->f_type != DTYPE_VNODE) {
2235 		error = ESPIPE;
2236 		goto done;
2237 	}
2238 	vp = (struct vnode *)fp->f_data;
2239 
2240 	switch (whence) {
2241 	case L_INCR:
2242 		new_offset = fp->f_offset + offset;
2243 		error = 0;
2244 		break;
2245 	case L_XTND:
2246 		error = VOP_GETATTR(vp, &vattr);
2247 		new_offset = offset + vattr.va_size;
2248 		break;
2249 	case L_SET:
2250 		new_offset = offset;
2251 		error = 0;
2252 		break;
2253 	default:
2254 		new_offset = 0;
2255 		error = EINVAL;
2256 		break;
2257 	}
2258 
2259 	/*
2260 	 * Validate the seek position.  Negative offsets are not allowed
2261 	 * for regular files or directories.
2262 	 *
2263 	 * Normally we would also not want to allow negative offsets for
2264 	 * character and block-special devices.  However kvm addresses
2265 	 * on 64 bit architectures might appear to be negative and must
2266 	 * be allowed.
2267 	 */
2268 	if (error == 0) {
2269 		if (new_offset < 0 &&
2270 		    (vp->v_type == VREG || vp->v_type == VDIR)) {
2271 			error = EINVAL;
2272 		} else {
2273 			fp->f_offset = new_offset;
2274 		}
2275 	}
2276 	*res = fp->f_offset;
2277 done:
2278 	fdrop(fp);
2279 	return (error);
2280 }
2281 
2282 /*
2283  * lseek_args(int fd, int pad, off_t offset, int whence)
2284  *
2285  * Reposition read/write file offset.
2286  */
2287 int
2288 sys_lseek(struct lseek_args *uap)
2289 {
2290 	int error;
2291 
2292 	error = kern_lseek(uap->fd, uap->offset, uap->whence,
2293 			   &uap->sysmsg_offset);
2294 
2295 	return (error);
2296 }
2297 
2298 int
2299 kern_access(struct nlookupdata *nd, int aflags)
2300 {
2301 	struct vnode *vp;
2302 	int error, flags;
2303 
2304 	if ((error = nlookup(nd)) != 0)
2305 		return (error);
2306 retry:
2307 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2308 	if (error)
2309 		return (error);
2310 
2311 	/* Flags == 0 means only check for existence. */
2312 	if (aflags) {
2313 		flags = 0;
2314 		if (aflags & R_OK)
2315 			flags |= VREAD;
2316 		if (aflags & W_OK)
2317 			flags |= VWRITE;
2318 		if (aflags & X_OK)
2319 			flags |= VEXEC;
2320 		if ((flags & VWRITE) == 0 ||
2321 		    (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2322 			error = VOP_ACCESS(vp, flags, nd->nl_cred);
2323 
2324 		/*
2325 		 * If the file handle is stale we have to re-resolve the
2326 		 * entry.  This is a hack at the moment.
2327 		 */
2328 		if (error == ESTALE) {
2329 			vput(vp);
2330 			cache_setunresolved(&nd->nl_nch);
2331 			error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2332 			if (error == 0) {
2333 				vp = NULL;
2334 				goto retry;
2335 			}
2336 			return(error);
2337 		}
2338 	}
2339 	vput(vp);
2340 	return (error);
2341 }
2342 
2343 /*
2344  * access_args(char *path, int flags)
2345  *
2346  * Check access permissions.
2347  */
2348 int
2349 sys_access(struct access_args *uap)
2350 {
2351 	struct nlookupdata nd;
2352 	int error;
2353 
2354 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2355 	if (error == 0)
2356 		error = kern_access(&nd, uap->flags);
2357 	nlookup_done(&nd);
2358 	return (error);
2359 }
2360 
2361 int
2362 kern_stat(struct nlookupdata *nd, struct stat *st)
2363 {
2364 	int error;
2365 	struct vnode *vp;
2366 	thread_t td;
2367 
2368 	if ((error = nlookup(nd)) != 0)
2369 		return (error);
2370 again:
2371 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2372 		return (ENOENT);
2373 
2374 	td = curthread;
2375 	if ((error = vget(vp, LK_SHARED)) != 0)
2376 		return (error);
2377 	error = vn_stat(vp, st, nd->nl_cred);
2378 
2379 	/*
2380 	 * If the file handle is stale we have to re-resolve the entry.  This
2381 	 * is a hack at the moment.
2382 	 */
2383 	if (error == ESTALE) {
2384 		vput(vp);
2385 		cache_setunresolved(&nd->nl_nch);
2386 		error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2387 		if (error == 0)
2388 			goto again;
2389 	} else {
2390 		vput(vp);
2391 	}
2392 	return (error);
2393 }
2394 
2395 /*
2396  * stat_args(char *path, struct stat *ub)
2397  *
2398  * Get file status; this version follows links.
2399  */
2400 int
2401 sys_stat(struct stat_args *uap)
2402 {
2403 	struct nlookupdata nd;
2404 	struct stat st;
2405 	int error;
2406 
2407 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2408 	if (error == 0) {
2409 		error = kern_stat(&nd, &st);
2410 		if (error == 0)
2411 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2412 	}
2413 	nlookup_done(&nd);
2414 	return (error);
2415 }
2416 
2417 /*
2418  * lstat_args(char *path, struct stat *ub)
2419  *
2420  * Get file status; this version does not follow links.
2421  */
2422 int
2423 sys_lstat(struct lstat_args *uap)
2424 {
2425 	struct nlookupdata nd;
2426 	struct stat st;
2427 	int error;
2428 
2429 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2430 	if (error == 0) {
2431 		error = kern_stat(&nd, &st);
2432 		if (error == 0)
2433 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2434 	}
2435 	nlookup_done(&nd);
2436 	return (error);
2437 }
2438 
2439 /*
2440  * pathconf_Args(char *path, int name)
2441  *
2442  * Get configurable pathname variables.
2443  */
2444 /* ARGSUSED */
2445 int
2446 sys_pathconf(struct pathconf_args *uap)
2447 {
2448 	struct nlookupdata nd;
2449 	struct vnode *vp;
2450 	int error;
2451 
2452 	vp = NULL;
2453 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2454 	if (error == 0)
2455 		error = nlookup(&nd);
2456 	if (error == 0)
2457 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2458 	nlookup_done(&nd);
2459 	if (error == 0) {
2460 		error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
2461 		vput(vp);
2462 	}
2463 	return (error);
2464 }
2465 
2466 /*
2467  * XXX: daver
2468  * kern_readlink isn't properly split yet.  There is a copyin burried
2469  * in VOP_READLINK().
2470  */
2471 int
2472 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2473 {
2474 	struct thread *td = curthread;
2475 	struct proc *p = td->td_proc;
2476 	struct vnode *vp;
2477 	struct iovec aiov;
2478 	struct uio auio;
2479 	int error;
2480 
2481 	if ((error = nlookup(nd)) != 0)
2482 		return (error);
2483 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2484 	if (error)
2485 		return (error);
2486 	if (vp->v_type != VLNK) {
2487 		error = EINVAL;
2488 	} else {
2489 		aiov.iov_base = buf;
2490 		aiov.iov_len = count;
2491 		auio.uio_iov = &aiov;
2492 		auio.uio_iovcnt = 1;
2493 		auio.uio_offset = 0;
2494 		auio.uio_rw = UIO_READ;
2495 		auio.uio_segflg = UIO_USERSPACE;
2496 		auio.uio_td = td;
2497 		auio.uio_resid = count;
2498 		error = VOP_READLINK(vp, &auio, p->p_ucred);
2499 	}
2500 	vput(vp);
2501 	*res = count - auio.uio_resid;
2502 	return (error);
2503 }
2504 
2505 /*
2506  * readlink_args(char *path, char *buf, int count)
2507  *
2508  * Return target name of a symbolic link.
2509  */
2510 int
2511 sys_readlink(struct readlink_args *uap)
2512 {
2513 	struct nlookupdata nd;
2514 	int error;
2515 
2516 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2517 	if (error == 0) {
2518 		error = kern_readlink(&nd, uap->buf, uap->count,
2519 					&uap->sysmsg_result);
2520 	}
2521 	nlookup_done(&nd);
2522 	return (error);
2523 }
2524 
2525 static int
2526 setfflags(struct vnode *vp, int flags)
2527 {
2528 	struct thread *td = curthread;
2529 	struct proc *p = td->td_proc;
2530 	int error;
2531 	struct vattr vattr;
2532 
2533 	/*
2534 	 * Prevent non-root users from setting flags on devices.  When
2535 	 * a device is reused, users can retain ownership of the device
2536 	 * if they are allowed to set flags and programs assume that
2537 	 * chown can't fail when done as root.
2538 	 */
2539 	if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2540 	    ((error = priv_check_cred(p->p_ucred, PRIV_ROOT, PRISON_ROOT)) != 0))
2541 		return (error);
2542 
2543 	/*
2544 	 * note: vget is required for any operation that might mod the vnode
2545 	 * so VINACTIVE is properly cleared.
2546 	 */
2547 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2548 		VATTR_NULL(&vattr);
2549 		vattr.va_flags = flags;
2550 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2551 		vput(vp);
2552 	}
2553 	return (error);
2554 }
2555 
2556 /*
2557  * chflags(char *path, int flags)
2558  *
2559  * Change flags of a file given a path name.
2560  */
2561 /* ARGSUSED */
2562 int
2563 sys_chflags(struct chflags_args *uap)
2564 {
2565 	struct nlookupdata nd;
2566 	struct vnode *vp;
2567 	int error;
2568 
2569 	vp = NULL;
2570 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2571 	if (error == 0)
2572 		error = nlookup(&nd);
2573 	if (error == 0)
2574 		error = ncp_writechk(&nd.nl_nch);
2575 	if (error == 0)
2576 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2577 	nlookup_done(&nd);
2578 	if (error == 0) {
2579 		error = setfflags(vp, uap->flags);
2580 		vrele(vp);
2581 	}
2582 	return (error);
2583 }
2584 
2585 /*
2586  * lchflags(char *path, int flags)
2587  *
2588  * Change flags of a file given a path name, but don't follow symlinks.
2589  */
2590 /* ARGSUSED */
2591 int
2592 sys_lchflags(struct lchflags_args *uap)
2593 {
2594 	struct nlookupdata nd;
2595 	struct vnode *vp;
2596 	int error;
2597 
2598 	vp = NULL;
2599 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2600 	if (error == 0)
2601 		error = nlookup(&nd);
2602 	if (error == 0)
2603 		error = ncp_writechk(&nd.nl_nch);
2604 	if (error == 0)
2605 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2606 	nlookup_done(&nd);
2607 	if (error == 0) {
2608 		error = setfflags(vp, uap->flags);
2609 		vrele(vp);
2610 	}
2611 	return (error);
2612 }
2613 
2614 /*
2615  * fchflags_args(int fd, int flags)
2616  *
2617  * Change flags of a file given a file descriptor.
2618  */
2619 /* ARGSUSED */
2620 int
2621 sys_fchflags(struct fchflags_args *uap)
2622 {
2623 	struct thread *td = curthread;
2624 	struct proc *p = td->td_proc;
2625 	struct file *fp;
2626 	int error;
2627 
2628 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2629 		return (error);
2630 	if (fp->f_nchandle.ncp)
2631 		error = ncp_writechk(&fp->f_nchandle);
2632 	if (error == 0)
2633 		error = setfflags((struct vnode *) fp->f_data, uap->flags);
2634 	fdrop(fp);
2635 	return (error);
2636 }
2637 
2638 static int
2639 setfmode(struct vnode *vp, int mode)
2640 {
2641 	struct thread *td = curthread;
2642 	struct proc *p = td->td_proc;
2643 	int error;
2644 	struct vattr vattr;
2645 
2646 	/*
2647 	 * note: vget is required for any operation that might mod the vnode
2648 	 * so VINACTIVE is properly cleared.
2649 	 */
2650 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2651 		VATTR_NULL(&vattr);
2652 		vattr.va_mode = mode & ALLPERMS;
2653 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2654 		vput(vp);
2655 	}
2656 	return error;
2657 }
2658 
2659 int
2660 kern_chmod(struct nlookupdata *nd, int mode)
2661 {
2662 	struct vnode *vp;
2663 	int error;
2664 
2665 	if ((error = nlookup(nd)) != 0)
2666 		return (error);
2667 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2668 		return (error);
2669 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2670 		error = setfmode(vp, mode);
2671 	vrele(vp);
2672 	return (error);
2673 }
2674 
2675 /*
2676  * chmod_args(char *path, int mode)
2677  *
2678  * Change mode of a file given path name.
2679  */
2680 /* ARGSUSED */
2681 int
2682 sys_chmod(struct chmod_args *uap)
2683 {
2684 	struct nlookupdata nd;
2685 	int error;
2686 
2687 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2688 	if (error == 0)
2689 		error = kern_chmod(&nd, uap->mode);
2690 	nlookup_done(&nd);
2691 	return (error);
2692 }
2693 
2694 /*
2695  * lchmod_args(char *path, int mode)
2696  *
2697  * Change mode of a file given path name (don't follow links.)
2698  */
2699 /* ARGSUSED */
2700 int
2701 sys_lchmod(struct lchmod_args *uap)
2702 {
2703 	struct nlookupdata nd;
2704 	int error;
2705 
2706 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2707 	if (error == 0)
2708 		error = kern_chmod(&nd, uap->mode);
2709 	nlookup_done(&nd);
2710 	return (error);
2711 }
2712 
2713 /*
2714  * fchmod_args(int fd, int mode)
2715  *
2716  * Change mode of a file given a file descriptor.
2717  */
2718 /* ARGSUSED */
2719 int
2720 sys_fchmod(struct fchmod_args *uap)
2721 {
2722 	struct thread *td = curthread;
2723 	struct proc *p = td->td_proc;
2724 	struct file *fp;
2725 	int error;
2726 
2727 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2728 		return (error);
2729 	if (fp->f_nchandle.ncp)
2730 		error = ncp_writechk(&fp->f_nchandle);
2731 	if (error == 0)
2732 		error = setfmode((struct vnode *)fp->f_data, uap->mode);
2733 	fdrop(fp);
2734 	return (error);
2735 }
2736 
2737 static int
2738 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2739 {
2740 	struct thread *td = curthread;
2741 	struct proc *p = td->td_proc;
2742 	int error;
2743 	struct vattr vattr;
2744 
2745 	/*
2746 	 * note: vget is required for any operation that might mod the vnode
2747 	 * so VINACTIVE is properly cleared.
2748 	 */
2749 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2750 		VATTR_NULL(&vattr);
2751 		vattr.va_uid = uid;
2752 		vattr.va_gid = gid;
2753 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2754 		vput(vp);
2755 	}
2756 	return error;
2757 }
2758 
2759 int
2760 kern_chown(struct nlookupdata *nd, int uid, int gid)
2761 {
2762 	struct vnode *vp;
2763 	int error;
2764 
2765 	if ((error = nlookup(nd)) != 0)
2766 		return (error);
2767 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2768 		return (error);
2769 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2770 		error = setfown(vp, uid, gid);
2771 	vrele(vp);
2772 	return (error);
2773 }
2774 
2775 /*
2776  * chown(char *path, int uid, int gid)
2777  *
2778  * Set ownership given a path name.
2779  */
2780 int
2781 sys_chown(struct chown_args *uap)
2782 {
2783 	struct nlookupdata nd;
2784 	int error;
2785 
2786 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2787 	if (error == 0)
2788 		error = kern_chown(&nd, uap->uid, uap->gid);
2789 	nlookup_done(&nd);
2790 	return (error);
2791 }
2792 
2793 /*
2794  * lchown_args(char *path, int uid, int gid)
2795  *
2796  * Set ownership given a path name, do not cross symlinks.
2797  */
2798 int
2799 sys_lchown(struct lchown_args *uap)
2800 {
2801 	struct nlookupdata nd;
2802 	int error;
2803 
2804 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2805 	if (error == 0)
2806 		error = kern_chown(&nd, uap->uid, uap->gid);
2807 	nlookup_done(&nd);
2808 	return (error);
2809 }
2810 
2811 /*
2812  * fchown_args(int fd, int uid, int gid)
2813  *
2814  * Set ownership given a file descriptor.
2815  */
2816 /* ARGSUSED */
2817 int
2818 sys_fchown(struct fchown_args *uap)
2819 {
2820 	struct thread *td = curthread;
2821 	struct proc *p = td->td_proc;
2822 	struct file *fp;
2823 	int error;
2824 
2825 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2826 		return (error);
2827 	if (fp->f_nchandle.ncp)
2828 		error = ncp_writechk(&fp->f_nchandle);
2829 	if (error == 0)
2830 		error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
2831 	fdrop(fp);
2832 	return (error);
2833 }
2834 
2835 static int
2836 getutimes(const struct timeval *tvp, struct timespec *tsp)
2837 {
2838 	struct timeval tv[2];
2839 
2840 	if (tvp == NULL) {
2841 		microtime(&tv[0]);
2842 		TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
2843 		tsp[1] = tsp[0];
2844 	} else {
2845 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
2846 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
2847 	}
2848 	return 0;
2849 }
2850 
2851 static int
2852 setutimes(struct vnode *vp, struct vattr *vattr,
2853 	  const struct timespec *ts, int nullflag)
2854 {
2855 	struct thread *td = curthread;
2856 	struct proc *p = td->td_proc;
2857 	int error;
2858 
2859 	VATTR_NULL(vattr);
2860 	vattr->va_atime = ts[0];
2861 	vattr->va_mtime = ts[1];
2862 	if (nullflag)
2863 		vattr->va_vaflags |= VA_UTIMES_NULL;
2864 	error = VOP_SETATTR(vp, vattr, p->p_ucred);
2865 
2866 	return error;
2867 }
2868 
2869 int
2870 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
2871 {
2872 	struct timespec ts[2];
2873 	struct vnode *vp;
2874 	struct vattr vattr;
2875 	int error;
2876 
2877 	if ((error = getutimes(tptr, ts)) != 0)
2878 		return (error);
2879 
2880 	/*
2881 	 * NOTE: utimes() succeeds for the owner even if the file
2882 	 * is not user-writable.
2883 	 */
2884 	nd->nl_flags |= NLC_OWN | NLC_WRITE;
2885 
2886 	if ((error = nlookup(nd)) != 0)
2887 		return (error);
2888 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2889 		return (error);
2890 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2891 		return (error);
2892 
2893 	/*
2894 	 * note: vget is required for any operation that might mod the vnode
2895 	 * so VINACTIVE is properly cleared.
2896 	 */
2897 	if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
2898 		error = vget(vp, LK_EXCLUSIVE);
2899 		if (error == 0) {
2900 			error = setutimes(vp, &vattr, ts, (tptr == NULL));
2901 			vput(vp);
2902 		}
2903 	}
2904 	vrele(vp);
2905 	return (error);
2906 }
2907 
2908 /*
2909  * utimes_args(char *path, struct timeval *tptr)
2910  *
2911  * Set the access and modification times of a file.
2912  */
2913 int
2914 sys_utimes(struct utimes_args *uap)
2915 {
2916 	struct timeval tv[2];
2917 	struct nlookupdata nd;
2918 	int error;
2919 
2920 	if (uap->tptr) {
2921  		error = copyin(uap->tptr, tv, sizeof(tv));
2922 		if (error)
2923 			return (error);
2924 	}
2925 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2926 	if (error == 0)
2927 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2928 	nlookup_done(&nd);
2929 	return (error);
2930 }
2931 
2932 /*
2933  * lutimes_args(char *path, struct timeval *tptr)
2934  *
2935  * Set the access and modification times of a file.
2936  */
2937 int
2938 sys_lutimes(struct lutimes_args *uap)
2939 {
2940 	struct timeval tv[2];
2941 	struct nlookupdata nd;
2942 	int error;
2943 
2944 	if (uap->tptr) {
2945 		error = copyin(uap->tptr, tv, sizeof(tv));
2946 		if (error)
2947 			return (error);
2948 	}
2949 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2950 	if (error == 0)
2951 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2952 	nlookup_done(&nd);
2953 	return (error);
2954 }
2955 
2956 /*
2957  * Set utimes on a file descriptor.  The creds used to open the
2958  * file are used to determine whether the operation is allowed
2959  * or not.
2960  */
2961 int
2962 kern_futimes(int fd, struct timeval *tptr)
2963 {
2964 	struct thread *td = curthread;
2965 	struct proc *p = td->td_proc;
2966 	struct timespec ts[2];
2967 	struct file *fp;
2968 	struct vnode *vp;
2969 	struct vattr vattr;
2970 	int error;
2971 
2972 	error = getutimes(tptr, ts);
2973 	if (error)
2974 		return (error);
2975 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2976 		return (error);
2977 	if (fp->f_nchandle.ncp)
2978 		error = ncp_writechk(&fp->f_nchandle);
2979 	if (error == 0) {
2980 		vp = fp->f_data;
2981 		error = vget(vp, LK_EXCLUSIVE);
2982 		if (error == 0) {
2983 			error = VOP_GETATTR(vp, &vattr);
2984 			if (error == 0) {
2985 				error = naccess_va(&vattr, NLC_OWN | NLC_WRITE,
2986 						   fp->f_cred);
2987 			}
2988 			if (error == 0) {
2989 				error = setutimes(vp, &vattr, ts,
2990 						  (tptr == NULL));
2991 			}
2992 			vput(vp);
2993 		}
2994 	}
2995 	fdrop(fp);
2996 	return (error);
2997 }
2998 
2999 /*
3000  * futimes_args(int fd, struct timeval *tptr)
3001  *
3002  * Set the access and modification times of a file.
3003  */
3004 int
3005 sys_futimes(struct futimes_args *uap)
3006 {
3007 	struct timeval tv[2];
3008 	int error;
3009 
3010 	if (uap->tptr) {
3011 		error = copyin(uap->tptr, tv, sizeof(tv));
3012 		if (error)
3013 			return (error);
3014 	}
3015 
3016 	error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
3017 
3018 	return (error);
3019 }
3020 
3021 int
3022 kern_truncate(struct nlookupdata *nd, off_t length)
3023 {
3024 	struct vnode *vp;
3025 	struct vattr vattr;
3026 	int error;
3027 
3028 	if (length < 0)
3029 		return(EINVAL);
3030 	nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE;
3031 	if ((error = nlookup(nd)) != 0)
3032 		return (error);
3033 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3034 		return (error);
3035 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3036 		return (error);
3037 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
3038 		vrele(vp);
3039 		return (error);
3040 	}
3041 	if (vp->v_type == VDIR) {
3042 		error = EISDIR;
3043 	} else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3044 		VATTR_NULL(&vattr);
3045 		vattr.va_size = length;
3046 		error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
3047 	}
3048 	vput(vp);
3049 	return (error);
3050 }
3051 
3052 /*
3053  * truncate(char *path, int pad, off_t length)
3054  *
3055  * Truncate a file given its path name.
3056  */
3057 int
3058 sys_truncate(struct truncate_args *uap)
3059 {
3060 	struct nlookupdata nd;
3061 	int error;
3062 
3063 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3064 	if (error == 0)
3065 		error = kern_truncate(&nd, uap->length);
3066 	nlookup_done(&nd);
3067 	return error;
3068 }
3069 
3070 int
3071 kern_ftruncate(int fd, off_t length)
3072 {
3073 	struct thread *td = curthread;
3074 	struct proc *p = td->td_proc;
3075 	struct vattr vattr;
3076 	struct vnode *vp;
3077 	struct file *fp;
3078 	int error;
3079 
3080 	if (length < 0)
3081 		return(EINVAL);
3082 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3083 		return (error);
3084 	if (fp->f_nchandle.ncp) {
3085 		error = ncp_writechk(&fp->f_nchandle);
3086 		if (error)
3087 			goto done;
3088 	}
3089 	if ((fp->f_flag & FWRITE) == 0) {
3090 		error = EINVAL;
3091 		goto done;
3092 	}
3093 	if (fp->f_flag & FAPPENDONLY) {	/* inode was set s/uapnd */
3094 		error = EINVAL;
3095 		goto done;
3096 	}
3097 	vp = (struct vnode *)fp->f_data;
3098 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3099 	if (vp->v_type == VDIR) {
3100 		error = EISDIR;
3101 	} else if ((error = vn_writechk(vp, NULL)) == 0) {
3102 		VATTR_NULL(&vattr);
3103 		vattr.va_size = length;
3104 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
3105 	}
3106 	vn_unlock(vp);
3107 done:
3108 	fdrop(fp);
3109 	return (error);
3110 }
3111 
3112 /*
3113  * ftruncate_args(int fd, int pad, off_t length)
3114  *
3115  * Truncate a file given a file descriptor.
3116  */
3117 int
3118 sys_ftruncate(struct ftruncate_args *uap)
3119 {
3120 	int error;
3121 
3122 	error = kern_ftruncate(uap->fd, uap->length);
3123 
3124 	return (error);
3125 }
3126 
3127 /*
3128  * fsync(int fd)
3129  *
3130  * Sync an open file.
3131  */
3132 /* ARGSUSED */
3133 int
3134 sys_fsync(struct fsync_args *uap)
3135 {
3136 	struct thread *td = curthread;
3137 	struct proc *p = td->td_proc;
3138 	struct vnode *vp;
3139 	struct file *fp;
3140 	vm_object_t obj;
3141 	int error;
3142 
3143 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3144 		return (error);
3145 	vp = (struct vnode *)fp->f_data;
3146 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3147 	if ((obj = vp->v_object) != NULL)
3148 		vm_object_page_clean(obj, 0, 0, 0);
3149 	if ((error = VOP_FSYNC(vp, MNT_WAIT)) == 0 && vp->v_mount)
3150 		error = buf_fsync(vp);
3151 	vn_unlock(vp);
3152 	fdrop(fp);
3153 	return (error);
3154 }
3155 
3156 int
3157 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
3158 {
3159 	struct nchandle fnchd;
3160 	struct nchandle tnchd;
3161 	struct namecache *ncp;
3162 	struct vnode *fdvp;
3163 	struct vnode *tdvp;
3164 	struct mount *mp;
3165 	int error;
3166 
3167 	bwillinode(1);
3168 	fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC;
3169 	if ((error = nlookup(fromnd)) != 0)
3170 		return (error);
3171 	if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
3172 		return (ENOENT);
3173 	fnchd.mount = fromnd->nl_nch.mount;
3174 	cache_hold(&fnchd);
3175 
3176 	/*
3177 	 * unlock the source nch so we can lookup the target nch without
3178 	 * deadlocking.  The target may or may not exist so we do not check
3179 	 * for a target vp like kern_mkdir() and other creation functions do.
3180 	 *
3181 	 * The source and target directories are ref'd and rechecked after
3182 	 * everything is relocked to determine if the source or target file
3183 	 * has been renamed.
3184 	 */
3185 	KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
3186 	fromnd->nl_flags &= ~NLC_NCPISLOCKED;
3187 	cache_unlock(&fromnd->nl_nch);
3188 
3189 	tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP;
3190 	if ((error = nlookup(tond)) != 0) {
3191 		cache_drop(&fnchd);
3192 		return (error);
3193 	}
3194 	if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
3195 		cache_drop(&fnchd);
3196 		return (ENOENT);
3197 	}
3198 	tnchd.mount = tond->nl_nch.mount;
3199 	cache_hold(&tnchd);
3200 
3201 	/*
3202 	 * If the source and target are the same there is nothing to do
3203 	 */
3204 	if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
3205 		cache_drop(&fnchd);
3206 		cache_drop(&tnchd);
3207 		return (0);
3208 	}
3209 
3210 	/*
3211 	 * Mount points cannot be renamed or overwritten
3212 	 */
3213 	if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
3214 	    NCF_ISMOUNTPT
3215 	) {
3216 		cache_drop(&fnchd);
3217 		cache_drop(&tnchd);
3218 		return (EINVAL);
3219 	}
3220 
3221 	/*
3222 	 * relock the source ncp.  NOTE AFTER RELOCKING: the source ncp
3223 	 * may have become invalid while it was unlocked, nc_vp and nc_mount
3224 	 * could be NULL.
3225 	 */
3226 	if (cache_lock_nonblock(&fromnd->nl_nch) == 0) {
3227 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3228 	} else if (fromnd->nl_nch.ncp > tond->nl_nch.ncp) {
3229 		cache_lock(&fromnd->nl_nch);
3230 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3231 	} else {
3232 		cache_unlock(&tond->nl_nch);
3233 		cache_lock(&fromnd->nl_nch);
3234 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3235 		cache_lock(&tond->nl_nch);
3236 		cache_resolve(&tond->nl_nch, tond->nl_cred);
3237 	}
3238 	fromnd->nl_flags |= NLC_NCPISLOCKED;
3239 
3240 	/*
3241 	 * make sure the parent directories linkages are the same
3242 	 */
3243 	if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
3244 	    tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
3245 		cache_drop(&fnchd);
3246 		cache_drop(&tnchd);
3247 		return (ENOENT);
3248 	}
3249 
3250 	/*
3251 	 * Both the source and target must be within the same filesystem and
3252 	 * in the same filesystem as their parent directories within the
3253 	 * namecache topology.
3254 	 *
3255 	 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
3256 	 */
3257 	mp = fnchd.mount;
3258 	if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
3259 	    mp != tond->nl_nch.mount) {
3260 		cache_drop(&fnchd);
3261 		cache_drop(&tnchd);
3262 		return (EXDEV);
3263 	}
3264 
3265 	/*
3266 	 * Make sure the mount point is writable
3267 	 */
3268 	if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
3269 		cache_drop(&fnchd);
3270 		cache_drop(&tnchd);
3271 		return (error);
3272 	}
3273 
3274 	/*
3275 	 * If the target exists and either the source or target is a directory,
3276 	 * then both must be directories.
3277 	 *
3278 	 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
3279 	 * have become NULL.
3280 	 */
3281 	if (tond->nl_nch.ncp->nc_vp) {
3282 		if (fromnd->nl_nch.ncp->nc_vp == NULL) {
3283 			error = ENOENT;
3284 		} else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
3285 			if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
3286 				error = ENOTDIR;
3287 		} else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
3288 			error = EISDIR;
3289 		}
3290 	}
3291 
3292 	/*
3293 	 * You cannot rename a source into itself or a subdirectory of itself.
3294 	 * We check this by travsersing the target directory upwards looking
3295 	 * for a match against the source.
3296 	 */
3297 	if (error == 0) {
3298 		for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
3299 			if (fromnd->nl_nch.ncp == ncp) {
3300 				error = EINVAL;
3301 				break;
3302 			}
3303 		}
3304 	}
3305 
3306 	cache_drop(&fnchd);
3307 	cache_drop(&tnchd);
3308 
3309 	/*
3310 	 * Even though the namespaces are different, they may still represent
3311 	 * hardlinks to the same file.  The filesystem might have a hard time
3312 	 * with this so we issue a NREMOVE of the source instead of a NRENAME
3313 	 * when we detect the situation.
3314 	 */
3315 	if (error == 0) {
3316 		fdvp = fromnd->nl_dvp;
3317 		tdvp = tond->nl_dvp;
3318 		if (fdvp == NULL || tdvp == NULL) {
3319 			error = EPERM;
3320 		} else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
3321 			error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
3322 					    fromnd->nl_cred);
3323 		} else {
3324 			error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
3325 					    fdvp, tdvp, tond->nl_cred);
3326 		}
3327 	}
3328 	return (error);
3329 }
3330 
3331 /*
3332  * rename_args(char *from, char *to)
3333  *
3334  * Rename files.  Source and destination must either both be directories,
3335  * or both not be directories.  If target is a directory, it must be empty.
3336  */
3337 int
3338 sys_rename(struct rename_args *uap)
3339 {
3340 	struct nlookupdata fromnd, tond;
3341 	int error;
3342 
3343 	error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3344 	if (error == 0) {
3345 		error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3346 		if (error == 0)
3347 			error = kern_rename(&fromnd, &tond);
3348 		nlookup_done(&tond);
3349 	}
3350 	nlookup_done(&fromnd);
3351 	return (error);
3352 }
3353 
3354 int
3355 kern_mkdir(struct nlookupdata *nd, int mode)
3356 {
3357 	struct thread *td = curthread;
3358 	struct proc *p = td->td_proc;
3359 	struct vnode *vp;
3360 	struct vattr vattr;
3361 	int error;
3362 
3363 	bwillinode(1);
3364 	nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
3365 	if ((error = nlookup(nd)) != 0)
3366 		return (error);
3367 
3368 	if (nd->nl_nch.ncp->nc_vp)
3369 		return (EEXIST);
3370 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3371 		return (error);
3372 	VATTR_NULL(&vattr);
3373 	vattr.va_type = VDIR;
3374 	vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3375 
3376 	vp = NULL;
3377 	error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, p->p_ucred, &vattr);
3378 	if (error == 0)
3379 		vput(vp);
3380 	return (error);
3381 }
3382 
3383 /*
3384  * mkdir_args(char *path, int mode)
3385  *
3386  * Make a directory file.
3387  */
3388 /* ARGSUSED */
3389 int
3390 sys_mkdir(struct mkdir_args *uap)
3391 {
3392 	struct nlookupdata nd;
3393 	int error;
3394 
3395 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3396 	if (error == 0)
3397 		error = kern_mkdir(&nd, uap->mode);
3398 	nlookup_done(&nd);
3399 	return (error);
3400 }
3401 
3402 int
3403 kern_rmdir(struct nlookupdata *nd)
3404 {
3405 	int error;
3406 
3407 	bwillinode(1);
3408 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
3409 	if ((error = nlookup(nd)) != 0)
3410 		return (error);
3411 
3412 	/*
3413 	 * Do not allow directories representing mount points to be
3414 	 * deleted, even if empty.  Check write perms on mount point
3415 	 * in case the vnode is aliased (aka nullfs).
3416 	 */
3417 	if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3418 		return (EINVAL);
3419 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3420 		return (error);
3421 	error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
3422 	return (error);
3423 }
3424 
3425 /*
3426  * rmdir_args(char *path)
3427  *
3428  * Remove a directory file.
3429  */
3430 /* ARGSUSED */
3431 int
3432 sys_rmdir(struct rmdir_args *uap)
3433 {
3434 	struct nlookupdata nd;
3435 	int error;
3436 
3437 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3438 	if (error == 0)
3439 		error = kern_rmdir(&nd);
3440 	nlookup_done(&nd);
3441 	return (error);
3442 }
3443 
3444 int
3445 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3446     enum uio_seg direction)
3447 {
3448 	struct thread *td = curthread;
3449 	struct proc *p = td->td_proc;
3450 	struct vnode *vp;
3451 	struct file *fp;
3452 	struct uio auio;
3453 	struct iovec aiov;
3454 	off_t loff;
3455 	int error, eofflag;
3456 
3457 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3458 		return (error);
3459 	if ((fp->f_flag & FREAD) == 0) {
3460 		error = EBADF;
3461 		goto done;
3462 	}
3463 	vp = (struct vnode *)fp->f_data;
3464 unionread:
3465 	if (vp->v_type != VDIR) {
3466 		error = EINVAL;
3467 		goto done;
3468 	}
3469 	aiov.iov_base = buf;
3470 	aiov.iov_len = count;
3471 	auio.uio_iov = &aiov;
3472 	auio.uio_iovcnt = 1;
3473 	auio.uio_rw = UIO_READ;
3474 	auio.uio_segflg = direction;
3475 	auio.uio_td = td;
3476 	auio.uio_resid = count;
3477 	loff = auio.uio_offset = fp->f_offset;
3478 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3479 	fp->f_offset = auio.uio_offset;
3480 	if (error)
3481 		goto done;
3482 	if (count == auio.uio_resid) {
3483 		if (union_dircheckp) {
3484 			error = union_dircheckp(td, &vp, fp);
3485 			if (error == -1)
3486 				goto unionread;
3487 			if (error)
3488 				goto done;
3489 		}
3490 #if 0
3491 		if ((vp->v_flag & VROOT) &&
3492 		    (vp->v_mount->mnt_flag & MNT_UNION)) {
3493 			struct vnode *tvp = vp;
3494 			vp = vp->v_mount->mnt_vnodecovered;
3495 			vref(vp);
3496 			fp->f_data = vp;
3497 			fp->f_offset = 0;
3498 			vrele(tvp);
3499 			goto unionread;
3500 		}
3501 #endif
3502 	}
3503 
3504 	/*
3505 	 * WARNING!  *basep may not be wide enough to accomodate the
3506 	 * seek offset.   XXX should we hack this to return the upper 32 bits
3507 	 * for offsets greater then 4G?
3508 	 */
3509 	if (basep) {
3510 		*basep = (long)loff;
3511 	}
3512 	*res = count - auio.uio_resid;
3513 done:
3514 	fdrop(fp);
3515 	return (error);
3516 }
3517 
3518 /*
3519  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3520  *
3521  * Read a block of directory entries in a file system independent format.
3522  */
3523 int
3524 sys_getdirentries(struct getdirentries_args *uap)
3525 {
3526 	long base;
3527 	int error;
3528 
3529 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3530 				   &uap->sysmsg_result, UIO_USERSPACE);
3531 
3532 	if (error == 0 && uap->basep)
3533 		error = copyout(&base, uap->basep, sizeof(*uap->basep));
3534 	return (error);
3535 }
3536 
3537 /*
3538  * getdents_args(int fd, char *buf, size_t count)
3539  */
3540 int
3541 sys_getdents(struct getdents_args *uap)
3542 {
3543 	int error;
3544 
3545 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3546 	    &uap->sysmsg_result, UIO_USERSPACE);
3547 
3548 	return (error);
3549 }
3550 
3551 /*
3552  * umask(int newmask)
3553  *
3554  * Set the mode mask for creation of filesystem nodes.
3555  *
3556  * MP SAFE
3557  */
3558 int
3559 sys_umask(struct umask_args *uap)
3560 {
3561 	struct thread *td = curthread;
3562 	struct proc *p = td->td_proc;
3563 	struct filedesc *fdp;
3564 
3565 	fdp = p->p_fd;
3566 	uap->sysmsg_result = fdp->fd_cmask;
3567 	fdp->fd_cmask = uap->newmask & ALLPERMS;
3568 	return (0);
3569 }
3570 
3571 /*
3572  * revoke(char *path)
3573  *
3574  * Void all references to file by ripping underlying filesystem
3575  * away from vnode.
3576  */
3577 /* ARGSUSED */
3578 int
3579 sys_revoke(struct revoke_args *uap)
3580 {
3581 	struct nlookupdata nd;
3582 	struct vattr vattr;
3583 	struct vnode *vp;
3584 	struct ucred *cred;
3585 	int error;
3586 
3587 	vp = NULL;
3588 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3589 	if (error == 0)
3590 		error = nlookup(&nd);
3591 	if (error == 0)
3592 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3593 	cred = crhold(nd.nl_cred);
3594 	nlookup_done(&nd);
3595 	if (error == 0) {
3596 		if (error == 0)
3597 			error = VOP_GETATTR(vp, &vattr);
3598 		if (error == 0 && cred->cr_uid != vattr.va_uid)
3599 			error = priv_check_cred(cred, PRIV_ROOT, PRISON_ROOT);
3600 		if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) {
3601 			if (count_udev(vp->v_umajor, vp->v_uminor) > 0)
3602 				error = vrevoke(vp, cred);
3603 		} else if (error == 0) {
3604 			error = vrevoke(vp, cred);
3605 		}
3606 		vrele(vp);
3607 	}
3608 	if (cred)
3609 		crfree(cred);
3610 	return (error);
3611 }
3612 
3613 /*
3614  * getfh_args(char *fname, fhandle_t *fhp)
3615  *
3616  * Get (NFS) file handle
3617  *
3618  * NOTE: We use the fsid of the covering mount, even if it is a nullfs
3619  * mount.  This allows nullfs mounts to be explicitly exported.
3620  *
3621  * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe.
3622  *
3623  * 	    nullfs mounts of subdirectories are not safe.  That is, it will
3624  *	    work, but you do not really have protection against access to
3625  *	    the related parent directories.
3626  */
3627 int
3628 sys_getfh(struct getfh_args *uap)
3629 {
3630 	struct thread *td = curthread;
3631 	struct nlookupdata nd;
3632 	fhandle_t fh;
3633 	struct vnode *vp;
3634 	struct mount *mp;
3635 	int error;
3636 
3637 	/*
3638 	 * Must be super user
3639 	 */
3640 	if ((error = priv_check(td, PRIV_ROOT)) != 0)
3641 		return (error);
3642 
3643 	vp = NULL;
3644 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3645 	if (error == 0)
3646 		error = nlookup(&nd);
3647 	if (error == 0)
3648 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3649 	mp = nd.nl_nch.mount;
3650 	nlookup_done(&nd);
3651 	if (error == 0) {
3652 		bzero(&fh, sizeof(fh));
3653 		fh.fh_fsid = mp->mnt_stat.f_fsid;
3654 		error = VFS_VPTOFH(vp, &fh.fh_fid);
3655 		vput(vp);
3656 		if (error == 0)
3657 			error = copyout(&fh, uap->fhp, sizeof(fh));
3658 	}
3659 	return (error);
3660 }
3661 
3662 /*
3663  * fhopen_args(const struct fhandle *u_fhp, int flags)
3664  *
3665  * syscall for the rpc.lockd to use to translate a NFS file handle into
3666  * an open descriptor.
3667  *
3668  * warning: do not remove the priv_check() call or this becomes one giant
3669  * security hole.
3670  */
3671 int
3672 sys_fhopen(struct fhopen_args *uap)
3673 {
3674 	struct thread *td = curthread;
3675 	struct proc *p = td->td_proc;
3676 	struct mount *mp;
3677 	struct vnode *vp;
3678 	struct fhandle fhp;
3679 	struct vattr vat;
3680 	struct vattr *vap = &vat;
3681 	struct flock lf;
3682 	int fmode, mode, error, type;
3683 	struct file *nfp;
3684 	struct file *fp;
3685 	int indx;
3686 
3687 	/*
3688 	 * Must be super user
3689 	 */
3690 	error = priv_check(td, PRIV_ROOT);
3691 	if (error)
3692 		return (error);
3693 
3694 	fmode = FFLAGS(uap->flags);
3695 	/* why not allow a non-read/write open for our lockd? */
3696 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3697 		return (EINVAL);
3698 	error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3699 	if (error)
3700 		return(error);
3701 	/* find the mount point */
3702 	mp = vfs_getvfs(&fhp.fh_fsid);
3703 	if (mp == NULL)
3704 		return (ESTALE);
3705 	/* now give me my vnode, it gets returned to me locked */
3706 	error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp);
3707 	if (error)
3708 		return (error);
3709  	/*
3710 	 * from now on we have to make sure not
3711 	 * to forget about the vnode
3712 	 * any error that causes an abort must vput(vp)
3713 	 * just set error = err and 'goto bad;'.
3714 	 */
3715 
3716 	/*
3717 	 * from vn_open
3718 	 */
3719 	if (vp->v_type == VLNK) {
3720 		error = EMLINK;
3721 		goto bad;
3722 	}
3723 	if (vp->v_type == VSOCK) {
3724 		error = EOPNOTSUPP;
3725 		goto bad;
3726 	}
3727 	mode = 0;
3728 	if (fmode & (FWRITE | O_TRUNC)) {
3729 		if (vp->v_type == VDIR) {
3730 			error = EISDIR;
3731 			goto bad;
3732 		}
3733 		error = vn_writechk(vp, NULL);
3734 		if (error)
3735 			goto bad;
3736 		mode |= VWRITE;
3737 	}
3738 	if (fmode & FREAD)
3739 		mode |= VREAD;
3740 	if (mode) {
3741 		error = VOP_ACCESS(vp, mode, p->p_ucred);
3742 		if (error)
3743 			goto bad;
3744 	}
3745 	if (fmode & O_TRUNC) {
3746 		vn_unlock(vp);				/* XXX */
3747 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
3748 		VATTR_NULL(vap);
3749 		vap->va_size = 0;
3750 		error = VOP_SETATTR(vp, vap, p->p_ucred);
3751 		if (error)
3752 			goto bad;
3753 	}
3754 
3755 	/*
3756 	 * VOP_OPEN needs the file pointer so it can potentially override
3757 	 * it.
3758 	 *
3759 	 * WARNING! no f_nchandle will be associated when fhopen()ing a
3760 	 * directory.  XXX
3761 	 */
3762 	if ((error = falloc(p, &nfp, &indx)) != 0)
3763 		goto bad;
3764 	fp = nfp;
3765 
3766 	error = VOP_OPEN(vp, fmode, p->p_ucred, fp);
3767 	if (error) {
3768 		/*
3769 		 * setting f_ops this way prevents VOP_CLOSE from being
3770 		 * called or fdrop() releasing the vp from v_data.   Since
3771 		 * the VOP_OPEN failed we don't want to VOP_CLOSE.
3772 		 */
3773 		fp->f_ops = &badfileops;
3774 		fp->f_data = NULL;
3775 		goto bad_drop;
3776 	}
3777 
3778 	/*
3779 	 * The fp is given its own reference, we still have our ref and lock.
3780 	 *
3781 	 * Assert that all regular files must be created with a VM object.
3782 	 */
3783 	if (vp->v_type == VREG && vp->v_object == NULL) {
3784 		kprintf("fhopen: regular file did not have VM object: %p\n", vp);
3785 		goto bad_drop;
3786 	}
3787 
3788 	/*
3789 	 * The open was successful.  Handle any locking requirements.
3790 	 */
3791 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
3792 		lf.l_whence = SEEK_SET;
3793 		lf.l_start = 0;
3794 		lf.l_len = 0;
3795 		if (fmode & O_EXLOCK)
3796 			lf.l_type = F_WRLCK;
3797 		else
3798 			lf.l_type = F_RDLCK;
3799 		if (fmode & FNONBLOCK)
3800 			type = 0;
3801 		else
3802 			type = F_WAIT;
3803 		vn_unlock(vp);
3804 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3805 			/*
3806 			 * release our private reference.
3807 			 */
3808 			fsetfd(p, NULL, indx);
3809 			fdrop(fp);
3810 			vrele(vp);
3811 			return (error);
3812 		}
3813 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3814 		fp->f_flag |= FHASLOCK;
3815 	}
3816 
3817 	/*
3818 	 * Clean up.  Associate the file pointer with the previously
3819 	 * reserved descriptor and return it.
3820 	 */
3821 	vput(vp);
3822 	fsetfd(p, fp, indx);
3823 	fdrop(fp);
3824 	uap->sysmsg_result = indx;
3825 	return (0);
3826 
3827 bad_drop:
3828 	fsetfd(p, NULL, indx);
3829 	fdrop(fp);
3830 bad:
3831 	vput(vp);
3832 	return (error);
3833 }
3834 
3835 /*
3836  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3837  */
3838 int
3839 sys_fhstat(struct fhstat_args *uap)
3840 {
3841 	struct thread *td = curthread;
3842 	struct stat sb;
3843 	fhandle_t fh;
3844 	struct mount *mp;
3845 	struct vnode *vp;
3846 	int error;
3847 
3848 	/*
3849 	 * Must be super user
3850 	 */
3851 	error = priv_check(td, PRIV_ROOT);
3852 	if (error)
3853 		return (error);
3854 
3855 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
3856 	if (error)
3857 		return (error);
3858 
3859 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3860 		return (ESTALE);
3861 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
3862 		return (error);
3863 	error = vn_stat(vp, &sb, td->td_proc->p_ucred);
3864 	vput(vp);
3865 	if (error)
3866 		return (error);
3867 	error = copyout(&sb, uap->sb, sizeof(sb));
3868 	return (error);
3869 }
3870 
3871 /*
3872  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
3873  */
3874 int
3875 sys_fhstatfs(struct fhstatfs_args *uap)
3876 {
3877 	struct thread *td = curthread;
3878 	struct proc *p = td->td_proc;
3879 	struct statfs *sp;
3880 	struct mount *mp;
3881 	struct vnode *vp;
3882 	struct statfs sb;
3883 	char *fullpath, *freepath;
3884 	fhandle_t fh;
3885 	int error;
3886 
3887 	/*
3888 	 * Must be super user
3889 	 */
3890 	if ((error = priv_check(td, PRIV_ROOT)))
3891 		return (error);
3892 
3893 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3894 		return (error);
3895 
3896 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3897 		return (ESTALE);
3898 
3899 	if (p != NULL && !chroot_visible_mnt(mp, p))
3900 		return (ESTALE);
3901 
3902 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
3903 		return (error);
3904 	mp = vp->v_mount;
3905 	sp = &mp->mnt_stat;
3906 	vput(vp);
3907 	if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0)
3908 		return (error);
3909 
3910 	error = mount_path(p, mp, &fullpath, &freepath);
3911 	if (error)
3912 		return(error);
3913 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3914 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
3915 	kfree(freepath, M_TEMP);
3916 
3917 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
3918 	if (priv_check(td, PRIV_ROOT)) {
3919 		bcopy(sp, &sb, sizeof(sb));
3920 		sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
3921 		sp = &sb;
3922 	}
3923 	return (copyout(sp, uap->buf, sizeof(*sp)));
3924 }
3925 
3926 /*
3927  * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
3928  */
3929 int
3930 sys_fhstatvfs(struct fhstatvfs_args *uap)
3931 {
3932 	struct thread *td = curthread;
3933 	struct proc *p = td->td_proc;
3934 	struct statvfs *sp;
3935 	struct mount *mp;
3936 	struct vnode *vp;
3937 	fhandle_t fh;
3938 	int error;
3939 
3940 	/*
3941 	 * Must be super user
3942 	 */
3943 	if ((error = priv_check(td, PRIV_ROOT)))
3944 		return (error);
3945 
3946 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3947 		return (error);
3948 
3949 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3950 		return (ESTALE);
3951 
3952 	if (p != NULL && !chroot_visible_mnt(mp, p))
3953 		return (ESTALE);
3954 
3955 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
3956 		return (error);
3957 	mp = vp->v_mount;
3958 	sp = &mp->mnt_vstat;
3959 	vput(vp);
3960 	if ((error = VFS_STATVFS(mp, sp, p->p_ucred)) != 0)
3961 		return (error);
3962 
3963 	sp->f_flag = 0;
3964 	if (mp->mnt_flag & MNT_RDONLY)
3965 		sp->f_flag |= ST_RDONLY;
3966 	if (mp->mnt_flag & MNT_NOSUID)
3967 		sp->f_flag |= ST_NOSUID;
3968 
3969 	return (copyout(sp, uap->buf, sizeof(*sp)));
3970 }
3971 
3972 
3973 /*
3974  * Syscall to push extended attribute configuration information into the
3975  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
3976  * a command (int cmd), and attribute name and misc data.  For now, the
3977  * attribute name is left in userspace for consumption by the VFS_op.
3978  * It will probably be changed to be copied into sysspace by the
3979  * syscall in the future, once issues with various consumers of the
3980  * attribute code have raised their hands.
3981  *
3982  * Currently this is used only by UFS Extended Attributes.
3983  */
3984 int
3985 sys_extattrctl(struct extattrctl_args *uap)
3986 {
3987 	struct nlookupdata nd;
3988 	struct mount *mp;
3989 	struct vnode *vp;
3990 	int error;
3991 
3992 	vp = NULL;
3993 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3994 	if (error == 0)
3995 		error = nlookup(&nd);
3996 	if (error == 0) {
3997 		mp = nd.nl_nch.mount;
3998 		error = VFS_EXTATTRCTL(mp, uap->cmd,
3999 				uap->attrname, uap->arg,
4000 				nd.nl_cred);
4001 	}
4002 	nlookup_done(&nd);
4003 	return (error);
4004 }
4005 
4006 /*
4007  * Syscall to set a named extended attribute on a file or directory.
4008  * Accepts attribute name, and a uio structure pointing to the data to set.
4009  * The uio is consumed in the style of writev().  The real work happens
4010  * in VOP_SETEXTATTR().
4011  */
4012 int
4013 sys_extattr_set_file(struct extattr_set_file_args *uap)
4014 {
4015 	char attrname[EXTATTR_MAXNAMELEN];
4016 	struct iovec aiov[UIO_SMALLIOV];
4017 	struct iovec *needfree;
4018 	struct nlookupdata nd;
4019 	struct iovec *iov;
4020 	struct vnode *vp;
4021 	struct uio auio;
4022 	u_int iovlen;
4023 	u_int cnt;
4024 	int error;
4025 	int i;
4026 
4027 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4028 	if (error)
4029 		return (error);
4030 
4031 	vp = NULL;
4032 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4033 	if (error == 0)
4034 		error = nlookup(&nd);
4035 	if (error == 0)
4036 		error = ncp_writechk(&nd.nl_nch);
4037 	if (error == 0)
4038 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4039 	if (error) {
4040 		nlookup_done(&nd);
4041 		return (error);
4042 	}
4043 
4044 	needfree = NULL;
4045 	iovlen = uap->iovcnt * sizeof(struct iovec);
4046 	if (uap->iovcnt > UIO_SMALLIOV) {
4047 		if (uap->iovcnt > UIO_MAXIOV) {
4048 			error = EINVAL;
4049 			goto done;
4050 		}
4051 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
4052 		needfree = iov;
4053 	} else {
4054 		iov = aiov;
4055 	}
4056 	auio.uio_iov = iov;
4057 	auio.uio_iovcnt = uap->iovcnt;
4058 	auio.uio_rw = UIO_WRITE;
4059 	auio.uio_segflg = UIO_USERSPACE;
4060 	auio.uio_td = nd.nl_td;
4061 	auio.uio_offset = 0;
4062 	if ((error = copyin(uap->iovp, iov, iovlen)))
4063 		goto done;
4064 	auio.uio_resid = 0;
4065 	for (i = 0; i < uap->iovcnt; i++) {
4066 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
4067 			error = EINVAL;
4068 			goto done;
4069 		}
4070 		auio.uio_resid += iov->iov_len;
4071 		iov++;
4072 	}
4073 	cnt = auio.uio_resid;
4074 	error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred);
4075 	cnt -= auio.uio_resid;
4076 	uap->sysmsg_result = cnt;
4077 done:
4078 	vput(vp);
4079 	nlookup_done(&nd);
4080 	if (needfree)
4081 		FREE(needfree, M_IOV);
4082 	return (error);
4083 }
4084 
4085 /*
4086  * Syscall to get a named extended attribute on a file or directory.
4087  * Accepts attribute name, and a uio structure pointing to a buffer for the
4088  * data.  The uio is consumed in the style of readv().  The real work
4089  * happens in VOP_GETEXTATTR();
4090  */
4091 int
4092 sys_extattr_get_file(struct extattr_get_file_args *uap)
4093 {
4094 	char attrname[EXTATTR_MAXNAMELEN];
4095 	struct iovec aiov[UIO_SMALLIOV];
4096 	struct iovec *needfree;
4097 	struct nlookupdata nd;
4098 	struct iovec *iov;
4099 	struct vnode *vp;
4100 	struct uio auio;
4101 	u_int iovlen;
4102 	u_int cnt;
4103 	int error;
4104 	int i;
4105 
4106 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4107 	if (error)
4108 		return (error);
4109 
4110 	vp = NULL;
4111 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4112 	if (error == 0)
4113 		error = nlookup(&nd);
4114 	if (error == 0)
4115 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4116 	if (error) {
4117 		nlookup_done(&nd);
4118 		return (error);
4119 	}
4120 
4121 	iovlen = uap->iovcnt * sizeof (struct iovec);
4122 	needfree = NULL;
4123 	if (uap->iovcnt > UIO_SMALLIOV) {
4124 		if (uap->iovcnt > UIO_MAXIOV) {
4125 			error = EINVAL;
4126 			goto done;
4127 		}
4128 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
4129 		needfree = iov;
4130 	} else {
4131 		iov = aiov;
4132 	}
4133 	auio.uio_iov = iov;
4134 	auio.uio_iovcnt = uap->iovcnt;
4135 	auio.uio_rw = UIO_READ;
4136 	auio.uio_segflg = UIO_USERSPACE;
4137 	auio.uio_td = nd.nl_td;
4138 	auio.uio_offset = 0;
4139 	if ((error = copyin(uap->iovp, iov, iovlen)))
4140 		goto done;
4141 	auio.uio_resid = 0;
4142 	for (i = 0; i < uap->iovcnt; i++) {
4143 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
4144 			error = EINVAL;
4145 			goto done;
4146 		}
4147 		auio.uio_resid += iov->iov_len;
4148 		iov++;
4149 	}
4150 	cnt = auio.uio_resid;
4151 	error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred);
4152 	cnt -= auio.uio_resid;
4153 	uap->sysmsg_result = cnt;
4154 done:
4155 	vput(vp);
4156 	nlookup_done(&nd);
4157 	if (needfree)
4158 		FREE(needfree, M_IOV);
4159 	return(error);
4160 }
4161 
4162 /*
4163  * Syscall to delete a named extended attribute from a file or directory.
4164  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
4165  */
4166 int
4167 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
4168 {
4169 	char attrname[EXTATTR_MAXNAMELEN];
4170 	struct nlookupdata nd;
4171 	struct vnode *vp;
4172 	int error;
4173 
4174 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4175 	if (error)
4176 		return(error);
4177 
4178 	vp = NULL;
4179 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4180 	if (error == 0)
4181 		error = nlookup(&nd);
4182 	if (error == 0)
4183 		error = ncp_writechk(&nd.nl_nch);
4184 	if (error == 0)
4185 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4186 	if (error) {
4187 		nlookup_done(&nd);
4188 		return (error);
4189 	}
4190 
4191 	error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred);
4192 	vput(vp);
4193 	nlookup_done(&nd);
4194 	return(error);
4195 }
4196 
4197 /*
4198  * Determine if the mount is visible to the process.
4199  */
4200 static int
4201 chroot_visible_mnt(struct mount *mp, struct proc *p)
4202 {
4203 	struct nchandle nch;
4204 
4205 	/*
4206 	 * Traverse from the mount point upwards.  If we hit the process
4207 	 * root then the mount point is visible to the process.
4208 	 */
4209 	nch = mp->mnt_ncmountpt;
4210 	while (nch.ncp) {
4211 		if (nch.mount == p->p_fd->fd_nrdir.mount &&
4212 		    nch.ncp == p->p_fd->fd_nrdir.ncp) {
4213 			return(1);
4214 		}
4215 		if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
4216 			nch = nch.mount->mnt_ncmounton;
4217 		} else {
4218 			nch.ncp = nch.ncp->nc_parent;
4219 		}
4220 	}
4221 
4222 	/*
4223 	 * If the mount point is not visible to the process, but the
4224 	 * process root is in a subdirectory of the mount, return
4225 	 * TRUE anyway.
4226 	 */
4227 	if (p->p_fd->fd_nrdir.mount == mp)
4228 		return(1);
4229 
4230 	return(0);
4231 }
4232 
4233