xref: /dflybsd-src/sys/kern/vfs_syscalls.c (revision b14d39bea25ac6b89cc943b26215359832fdb51a)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_syscalls.c	8.13 (Berkeley) 4/15/94
39  * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $
40  */
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/buf.h>
45 #include <sys/conf.h>
46 #include <sys/sysent.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
49 #include <sys/mountctl.h>
50 #include <sys/sysproto.h>
51 #include <sys/filedesc.h>
52 #include <sys/kernel.h>
53 #include <sys/fcntl.h>
54 #include <sys/file.h>
55 #include <sys/linker.h>
56 #include <sys/stat.h>
57 #include <sys/unistd.h>
58 #include <sys/vnode.h>
59 #include <sys/proc.h>
60 #include <sys/priv.h>
61 #include <sys/jail.h>
62 #include <sys/namei.h>
63 #include <sys/nlookup.h>
64 #include <sys/dirent.h>
65 #include <sys/extattr.h>
66 #include <sys/spinlock.h>
67 #include <sys/kern_syscall.h>
68 #include <sys/objcache.h>
69 #include <sys/sysctl.h>
70 
71 #include <sys/buf2.h>
72 #include <sys/file2.h>
73 #include <sys/spinlock2.h>
74 #include <sys/mplock2.h>
75 
76 #include <vm/vm.h>
77 #include <vm/vm_object.h>
78 #include <vm/vm_page.h>
79 
80 #include <machine/limits.h>
81 #include <machine/stdarg.h>
82 
83 #include <vfs/union/union.h>
84 
85 static void mount_warning(struct mount *mp, const char *ctl, ...)
86 		__printflike(2, 3);
87 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb);
88 static int checkvp_chdir (struct vnode *vn, struct thread *td);
89 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch);
90 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
91 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
92 static int getutimes (const struct timeval *, struct timespec *);
93 static int setfown (struct vnode *, uid_t, gid_t);
94 static int setfmode (struct vnode *, int);
95 static int setfflags (struct vnode *, int);
96 static int setutimes (struct vnode *, struct vattr *,
97 			const struct timespec *, int);
98 static int	usermount = 0;	/* if 1, non-root can mount fs. */
99 
100 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *);
101 
102 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "");
103 
104 /*
105  * Virtual File System System Calls
106  */
107 
108 /*
109  * Mount a file system.
110  *
111  * mount_args(char *type, char *path, int flags, caddr_t data)
112  *
113  * MPALMOSTSAFE
114  */
115 int
116 sys_mount(struct mount_args *uap)
117 {
118 	struct thread *td = curthread;
119 	struct vnode *vp;
120 	struct nchandle nch;
121 	struct mount *mp, *nullmp;
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;
129 
130 	get_mplock();
131 	cred = td->td_ucred;
132 	if (jailed(cred)) {
133 		error = EPERM;
134 		goto done;
135 	}
136 	if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
137 		goto done;
138 
139 	/*
140 	 * Do not allow NFS export by non-root users.
141 	 */
142 	if (uap->flags & MNT_EXPORTED) {
143 		error = priv_check(td, PRIV_ROOT);
144 		if (error)
145 			goto done;
146 	}
147 	/*
148 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
149 	 */
150 	if (priv_check(td, PRIV_ROOT))
151 		uap->flags |= MNT_NOSUID | MNT_NODEV;
152 
153 	/*
154 	 * Lookup the requested path and extract the nch and vnode.
155 	 */
156 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
157 	if (error == 0) {
158 		if ((error = nlookup(&nd)) == 0) {
159 			if (nd.nl_nch.ncp->nc_vp == NULL)
160 				error = ENOENT;
161 		}
162 	}
163 	if (error) {
164 		nlookup_done(&nd);
165 		goto done;
166 	}
167 
168 	/*
169 	 * If the target filesystem is resolved via a nullfs mount, then
170 	 * nd.nl_nch.mount will be pointing to the nullfs mount structure
171 	 * instead of the target file system. We need it in case we are
172 	 * doing an update.
173 	 */
174 	nullmp = nd.nl_nch.mount;
175 
176 	/*
177 	 * Extract the locked+refd ncp and cleanup the nd structure
178 	 */
179 	nch = nd.nl_nch;
180 	cache_zero(&nd.nl_nch);
181 	nlookup_done(&nd);
182 
183 	if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) && cache_findmount(&nch))
184 		hasmount = 1;
185 	else
186 		hasmount = 0;
187 
188 
189 	/*
190 	 * now we have the locked ref'd nch and unreferenced vnode.
191 	 */
192 	vp = nch.ncp->nc_vp;
193 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
194 		cache_put(&nch);
195 		goto done;
196 	}
197 	cache_unlock(&nch);
198 
199 	/*
200 	 * Extract the file system type. We need to know this early, to take
201 	 * appropriate actions if we are dealing with a nullfs.
202 	 */
203         if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
204                 cache_drop(&nch);
205                 vput(vp);
206 		goto done;
207         }
208 
209 	/*
210 	 * Now we have an unlocked ref'd nch and a locked ref'd vp
211 	 */
212 	if (uap->flags & MNT_UPDATE) {
213 		if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
214 			cache_drop(&nch);
215 			vput(vp);
216 			error = EINVAL;
217 			goto done;
218 		}
219 
220 		if (strncmp(fstypename, "null", 5) == 0) {
221 			KKASSERT(nullmp);
222 			mp = nullmp;
223 		} else {
224 			mp = vp->v_mount;
225 		}
226 
227 		flag = mp->mnt_flag;
228 		flag2 = mp->mnt_kern_flag;
229 		/*
230 		 * We only allow the filesystem to be reloaded if it
231 		 * is currently mounted read-only.
232 		 */
233 		if ((uap->flags & MNT_RELOAD) &&
234 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
235 			cache_drop(&nch);
236 			vput(vp);
237 			error = EOPNOTSUPP;	/* Needs translation */
238 			goto done;
239 		}
240 		/*
241 		 * Only root, or the user that did the original mount is
242 		 * permitted to update it.
243 		 */
244 		if (mp->mnt_stat.f_owner != cred->cr_uid &&
245 		    (error = priv_check(td, PRIV_ROOT))) {
246 			cache_drop(&nch);
247 			vput(vp);
248 			goto done;
249 		}
250 		if (vfs_busy(mp, LK_NOWAIT)) {
251 			cache_drop(&nch);
252 			vput(vp);
253 			error = EBUSY;
254 			goto done;
255 		}
256 		if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
257 			cache_drop(&nch);
258 			vfs_unbusy(mp);
259 			vput(vp);
260 			error = EBUSY;
261 			goto done;
262 		}
263 		vsetflags(vp, VMOUNT);
264 		mp->mnt_flag |=
265 		    uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
266 		vn_unlock(vp);
267 		goto update;
268 	}
269 	/*
270 	 * If the user is not root, ensure that they own the directory
271 	 * onto which we are attempting to mount.
272 	 */
273 	if ((error = VOP_GETATTR(vp, &va)) ||
274 	    (va.va_uid != cred->cr_uid && (error = priv_check(td, PRIV_ROOT)))) {
275 		cache_drop(&nch);
276 		vput(vp);
277 		goto done;
278 	}
279 	if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
280 		cache_drop(&nch);
281 		vput(vp);
282 		goto done;
283 	}
284 	if (vp->v_type != VDIR) {
285 		cache_drop(&nch);
286 		vput(vp);
287 		error = ENOTDIR;
288 		goto done;
289 	}
290 	if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) {
291 		cache_drop(&nch);
292 		vput(vp);
293 		error = EPERM;
294 		goto done;
295 	}
296 	vfsp = vfsconf_find_by_name(fstypename);
297 	if (vfsp == NULL) {
298 		linker_file_t lf;
299 
300 		/* Only load modules for root (very important!) */
301 		if ((error = priv_check(td, PRIV_ROOT)) != 0) {
302 			cache_drop(&nch);
303 			vput(vp);
304 			goto done;
305 		}
306 		error = linker_load_file(fstypename, &lf);
307 		if (error || lf == NULL) {
308 			cache_drop(&nch);
309 			vput(vp);
310 			if (lf == NULL)
311 				error = ENODEV;
312 			goto done;
313 		}
314 		lf->userrefs++;
315 		/* lookup again, see if the VFS was loaded */
316 		vfsp = vfsconf_find_by_name(fstypename);
317 		if (vfsp == NULL) {
318 			lf->userrefs--;
319 			linker_file_unload(lf);
320 			cache_drop(&nch);
321 			vput(vp);
322 			error = ENODEV;
323 			goto done;
324 		}
325 	}
326 	if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
327 		cache_drop(&nch);
328 		vput(vp);
329 		error = EBUSY;
330 		goto done;
331 	}
332 	vsetflags(vp, VMOUNT);
333 
334 	/*
335 	 * Allocate and initialize the filesystem.
336 	 */
337 	mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
338 	mount_init(mp);
339 	vfs_busy(mp, LK_NOWAIT);
340 	mp->mnt_op = vfsp->vfc_vfsops;
341 	mp->mnt_vfc = vfsp;
342 	vfsp->vfc_refcount++;
343 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
344 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
345 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
346 	mp->mnt_stat.f_owner = cred->cr_uid;
347 	vn_unlock(vp);
348 update:
349 	/*
350 	 * Set the mount level flags.
351 	 */
352 	if (uap->flags & MNT_RDONLY)
353 		mp->mnt_flag |= MNT_RDONLY;
354 	else if (mp->mnt_flag & MNT_RDONLY)
355 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
356 	mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
357 	    MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
358 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
359 	    MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
360 	mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
361 	    MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
362 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
363 	    MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
364 	/*
365 	 * Mount the filesystem.
366 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
367 	 * get.
368 	 */
369 	error = VFS_MOUNT(mp, uap->path, uap->data, cred);
370 	if (mp->mnt_flag & MNT_UPDATE) {
371 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
372 			mp->mnt_flag &= ~MNT_RDONLY;
373 		mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
374 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
375 		if (error) {
376 			mp->mnt_flag = flag;
377 			mp->mnt_kern_flag = flag2;
378 		}
379 		vfs_unbusy(mp);
380 		vclrflags(vp, VMOUNT);
381 		vrele(vp);
382 		cache_drop(&nch);
383 		goto done;
384 	}
385 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
386 	/*
387 	 * Put the new filesystem on the mount list after root.  The mount
388 	 * point gets its own mnt_ncmountpt (unless the VFS already set one
389 	 * up) which represents the root of the mount.  The lookup code
390 	 * detects the mount point going forward and checks the root of
391 	 * the mount going backwards.
392 	 *
393 	 * It is not necessary to invalidate or purge the vnode underneath
394 	 * because elements under the mount will be given their own glue
395 	 * namecache record.
396 	 */
397 	if (!error) {
398 		if (mp->mnt_ncmountpt.ncp == NULL) {
399 			/*
400 			 * allocate, then unlock, but leave the ref intact
401 			 */
402 			cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
403 			cache_unlock(&mp->mnt_ncmountpt);
404 		}
405 		mp->mnt_ncmounton = nch;		/* inherits ref */
406 		nch.ncp->nc_flag |= NCF_ISMOUNTPT;
407 
408 		/* XXX get the root of the fs and cache_setvp(mnt_ncmountpt...) */
409 		vclrflags(vp, VMOUNT);
410 		mountlist_insert(mp, MNTINS_LAST);
411 		vn_unlock(vp);
412 		checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
413 		error = vfs_allocate_syncvnode(mp);
414 		vfs_unbusy(mp);
415 		error = VFS_START(mp, 0);
416 		vrele(vp);
417 	} else {
418 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
419 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
420 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
421 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
422 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
423 		vclrflags(vp, VMOUNT);
424 		mp->mnt_vfc->vfc_refcount--;
425 		vfs_unbusy(mp);
426 		kfree(mp, M_MOUNT);
427 		cache_drop(&nch);
428 		vput(vp);
429 	}
430 done:
431 	rel_mplock();
432 	return (error);
433 }
434 
435 /*
436  * Scan all active processes to see if any of them have a current
437  * or root directory onto which the new filesystem has just been
438  * mounted. If so, replace them with the new mount point.
439  *
440  * The passed ncp is ref'd and locked (from the mount code) and
441  * must be associated with the vnode representing the root of the
442  * mount point.
443  */
444 struct checkdirs_info {
445 	struct nchandle old_nch;
446 	struct nchandle new_nch;
447 	struct vnode *old_vp;
448 	struct vnode *new_vp;
449 };
450 
451 static int checkdirs_callback(struct proc *p, void *data);
452 
453 static void
454 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
455 {
456 	struct checkdirs_info info;
457 	struct vnode *olddp;
458 	struct vnode *newdp;
459 	struct mount *mp;
460 
461 	/*
462 	 * If the old mount point's vnode has a usecount of 1, it is not
463 	 * being held as a descriptor anywhere.
464 	 */
465 	olddp = old_nch->ncp->nc_vp;
466 	if (olddp == NULL || olddp->v_sysref.refcnt == 1)
467 		return;
468 
469 	/*
470 	 * Force the root vnode of the new mount point to be resolved
471 	 * so we can update any matching processes.
472 	 */
473 	mp = new_nch->mount;
474 	if (VFS_ROOT(mp, &newdp))
475 		panic("mount: lost mount");
476 	cache_setunresolved(new_nch);
477 	cache_setvp(new_nch, newdp);
478 
479 	/*
480 	 * Special handling of the root node
481 	 */
482 	if (rootvnode == olddp) {
483 		vref(newdp);
484 		vfs_cache_setroot(newdp, cache_hold(new_nch));
485 	}
486 
487 	/*
488 	 * Pass newdp separately so the callback does not have to access
489 	 * it via new_nch->ncp->nc_vp.
490 	 */
491 	info.old_nch = *old_nch;
492 	info.new_nch = *new_nch;
493 	info.new_vp = newdp;
494 	allproc_scan(checkdirs_callback, &info);
495 	vput(newdp);
496 }
497 
498 /*
499  * NOTE: callback is not MP safe because the scanned process's filedesc
500  * structure can be ripped out from under us, amoung other things.
501  */
502 static int
503 checkdirs_callback(struct proc *p, void *data)
504 {
505 	struct checkdirs_info *info = data;
506 	struct filedesc *fdp;
507 	struct nchandle ncdrop1;
508 	struct nchandle ncdrop2;
509 	struct vnode *vprele1;
510 	struct vnode *vprele2;
511 
512 	if ((fdp = p->p_fd) != NULL) {
513 		cache_zero(&ncdrop1);
514 		cache_zero(&ncdrop2);
515 		vprele1 = NULL;
516 		vprele2 = NULL;
517 
518 		/*
519 		 * MPUNSAFE - XXX fdp can be pulled out from under a
520 		 * foreign process.
521 		 *
522 		 * A shared filedesc is ok, we don't have to copy it
523 		 * because we are making this change globally.
524 		 */
525 		spin_lock(&fdp->fd_spin);
526 		if (fdp->fd_ncdir.mount == info->old_nch.mount &&
527 		    fdp->fd_ncdir.ncp == info->old_nch.ncp) {
528 			vprele1 = fdp->fd_cdir;
529 			vref(info->new_vp);
530 			fdp->fd_cdir = info->new_vp;
531 			ncdrop1 = fdp->fd_ncdir;
532 			cache_copy(&info->new_nch, &fdp->fd_ncdir);
533 		}
534 		if (fdp->fd_nrdir.mount == info->old_nch.mount &&
535 		    fdp->fd_nrdir.ncp == info->old_nch.ncp) {
536 			vprele2 = fdp->fd_rdir;
537 			vref(info->new_vp);
538 			fdp->fd_rdir = info->new_vp;
539 			ncdrop2 = fdp->fd_nrdir;
540 			cache_copy(&info->new_nch, &fdp->fd_nrdir);
541 		}
542 		spin_unlock(&fdp->fd_spin);
543 		if (ncdrop1.ncp)
544 			cache_drop(&ncdrop1);
545 		if (ncdrop2.ncp)
546 			cache_drop(&ncdrop2);
547 		if (vprele1)
548 			vrele(vprele1);
549 		if (vprele2)
550 			vrele(vprele2);
551 	}
552 	return(0);
553 }
554 
555 /*
556  * Unmount a file system.
557  *
558  * Note: unmount takes a path to the vnode mounted on as argument,
559  * not special file (as before).
560  *
561  * umount_args(char *path, int flags)
562  *
563  * MPALMOSTSAFE
564  */
565 int
566 sys_unmount(struct unmount_args *uap)
567 {
568 	struct thread *td = curthread;
569 	struct proc *p __debugvar = td->td_proc;
570 	struct mount *mp = NULL;
571 	struct nlookupdata nd;
572 	int error;
573 
574 	KKASSERT(p);
575 	get_mplock();
576 	if (td->td_ucred->cr_prison != NULL) {
577 		error = EPERM;
578 		goto done;
579 	}
580 	if (usermount == 0 && (error = priv_check(td, PRIV_ROOT)))
581 		goto done;
582 
583 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
584 	if (error == 0)
585 		error = nlookup(&nd);
586 	if (error)
587 		goto out;
588 
589 	mp = nd.nl_nch.mount;
590 
591 	/*
592 	 * Only root, or the user that did the original mount is
593 	 * permitted to unmount this filesystem.
594 	 */
595 	if ((mp->mnt_stat.f_owner != td->td_ucred->cr_uid) &&
596 	    (error = priv_check(td, PRIV_ROOT)))
597 		goto out;
598 
599 	/*
600 	 * Don't allow unmounting the root file system.
601 	 */
602 	if (mp->mnt_flag & MNT_ROOTFS) {
603 		error = EINVAL;
604 		goto out;
605 	}
606 
607 	/*
608 	 * Must be the root of the filesystem
609 	 */
610 	if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
611 		error = EINVAL;
612 		goto out;
613 	}
614 
615 out:
616 	nlookup_done(&nd);
617 	if (error == 0)
618 		error = dounmount(mp, uap->flags);
619 done:
620 	rel_mplock();
621 	return (error);
622 }
623 
624 /*
625  * Do the actual file system unmount.
626  */
627 static int
628 dounmount_interlock(struct mount *mp)
629 {
630 	if (mp->mnt_kern_flag & MNTK_UNMOUNT)
631 		return (EBUSY);
632 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
633 	return(0);
634 }
635 
636 static int
637 unmount_allproc_cb(struct proc *p, void *arg)
638 {
639 	struct mount *mp;
640 
641 	if (p->p_textnch.ncp == NULL)
642 		return 0;
643 
644 	mp = (struct mount *)arg;
645 	if (p->p_textnch.mount == mp)
646 		cache_drop(&p->p_textnch);
647 
648 	return 0;
649 }
650 
651 int
652 dounmount(struct mount *mp, int flags)
653 {
654 	struct namecache *ncp;
655 	struct nchandle nch;
656 	struct vnode *vp;
657 	int error;
658 	int async_flag;
659 	int lflags;
660 	int freeok = 1;
661 
662 	/*
663 	 * Exclusive access for unmounting purposes
664 	 */
665 	if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
666 		return (error);
667 
668 	/*
669 	 * Allow filesystems to detect that a forced unmount is in progress.
670 	 */
671 	if (flags & MNT_FORCE)
672 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
673 	lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
674 	error = lockmgr(&mp->mnt_lock, lflags);
675 	if (error) {
676 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
677 		if (mp->mnt_kern_flag & MNTK_MWAIT)
678 			wakeup(mp);
679 		return (error);
680 	}
681 
682 	if (mp->mnt_flag & MNT_EXPUBLIC)
683 		vfs_setpublicfs(NULL, NULL, NULL);
684 
685 	vfs_msync(mp, MNT_WAIT);
686 	async_flag = mp->mnt_flag & MNT_ASYNC;
687 	mp->mnt_flag &=~ MNT_ASYNC;
688 
689 	/*
690 	 * If this filesystem isn't aliasing other filesystems,
691 	 * try to invalidate any remaining namecache entries and
692 	 * check the count afterwords.
693 	 */
694 	if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
695 		cache_lock(&mp->mnt_ncmountpt);
696 		cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
697 		cache_unlock(&mp->mnt_ncmountpt);
698 
699 		if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
700 		    (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
701 			allproc_scan(&unmount_allproc_cb, mp);
702 		}
703 
704 		if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
705 		    (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
706 
707 			if ((flags & MNT_FORCE) == 0) {
708 				error = EBUSY;
709 				mount_warning(mp, "Cannot unmount: "
710 						  "%d namecache "
711 						  "references still "
712 						  "present",
713 						  ncp->nc_refs - 1);
714 			} else {
715 				mount_warning(mp, "Forced unmount: "
716 						  "%d namecache "
717 						  "references still "
718 						  "present",
719 						  ncp->nc_refs - 1);
720 				freeok = 0;
721 			}
722 		}
723 	}
724 
725 	/*
726 	 * nchandle records ref the mount structure.  Expect a count of 1
727 	 * (our mount->mnt_ncmountpt).
728 	 */
729 	if (mp->mnt_refs != 1) {
730 		if ((flags & MNT_FORCE) == 0) {
731 			mount_warning(mp, "Cannot unmount: "
732 					  "%d process references still "
733 					  "present", mp->mnt_refs);
734 			error = EBUSY;
735 		} else {
736 			mount_warning(mp, "Forced unmount: "
737 					  "%d process references still "
738 					  "present", mp->mnt_refs);
739 			freeok = 0;
740 		}
741 	}
742 
743 	/*
744 	 * Decomission our special mnt_syncer vnode.  This also stops
745 	 * the vnlru code.  If we are unable to unmount we recommission
746 	 * the vnode.
747 	 */
748 	if (error == 0) {
749 		if ((vp = mp->mnt_syncer) != NULL) {
750 			mp->mnt_syncer = NULL;
751 			vrele(vp);
752 		}
753 		if (((mp->mnt_flag & MNT_RDONLY) ||
754 		     (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
755 		    (flags & MNT_FORCE)) {
756 			error = VFS_UNMOUNT(mp, flags);
757 		}
758 	}
759 	if (error) {
760 		if (mp->mnt_syncer == NULL)
761 			vfs_allocate_syncvnode(mp);
762 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
763 		mp->mnt_flag |= async_flag;
764 		lockmgr(&mp->mnt_lock, LK_RELEASE);
765 		if (mp->mnt_kern_flag & MNTK_MWAIT)
766 			wakeup(mp);
767 		return (error);
768 	}
769 	/*
770 	 * Clean up any journals still associated with the mount after
771 	 * filesystem activity has ceased.
772 	 */
773 	journal_remove_all_journals(mp,
774 	    ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
775 
776 	mountlist_remove(mp);
777 
778 	/*
779 	 * Remove any installed vnode ops here so the individual VFSs don't
780 	 * have to.
781 	 */
782 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
783 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
784 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
785 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
786 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
787 
788 	if (mp->mnt_ncmountpt.ncp != NULL) {
789 		nch = mp->mnt_ncmountpt;
790 		cache_zero(&mp->mnt_ncmountpt);
791 		cache_clrmountpt(&nch);
792 		cache_drop(&nch);
793 	}
794 	if (mp->mnt_ncmounton.ncp != NULL) {
795 		nch = mp->mnt_ncmounton;
796 		cache_zero(&mp->mnt_ncmounton);
797 		cache_clrmountpt(&nch);
798 		cache_drop(&nch);
799 	}
800 
801 	mp->mnt_vfc->vfc_refcount--;
802 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
803 		panic("unmount: dangling vnode");
804 	lockmgr(&mp->mnt_lock, LK_RELEASE);
805 	if (mp->mnt_kern_flag & MNTK_MWAIT)
806 		wakeup(mp);
807 	if (freeok)
808 		kfree(mp, M_MOUNT);
809 	return (0);
810 }
811 
812 static
813 void
814 mount_warning(struct mount *mp, const char *ctl, ...)
815 {
816 	char *ptr;
817 	char *buf;
818 	__va_list va;
819 
820 	__va_start(va, ctl);
821 	if (cache_fullpath(NULL, &mp->mnt_ncmounton, &ptr, &buf, 0) == 0) {
822 		kprintf("unmount(%s): ", ptr);
823 		kvprintf(ctl, va);
824 		kprintf("\n");
825 		kfree(buf, M_TEMP);
826 	} else {
827 		kprintf("unmount(%p", mp);
828 		if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
829 			kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
830 		kprintf("): ");
831 		kvprintf(ctl, va);
832 		kprintf("\n");
833 	}
834 	__va_end(va);
835 }
836 
837 /*
838  * Shim cache_fullpath() to handle the case where a process is chrooted into
839  * a subdirectory of a mount.  In this case if the root mount matches the
840  * process root directory's mount we have to specify the process's root
841  * directory instead of the mount point, because the mount point might
842  * be above the root directory.
843  */
844 static
845 int
846 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
847 {
848 	struct nchandle *nch;
849 
850 	if (p && p->p_fd->fd_nrdir.mount == mp)
851 		nch = &p->p_fd->fd_nrdir;
852 	else
853 		nch = &mp->mnt_ncmountpt;
854 	return(cache_fullpath(p, nch, rb, fb, 0));
855 }
856 
857 /*
858  * Sync each mounted filesystem.
859  */
860 
861 #ifdef DEBUG
862 static int syncprt = 0;
863 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
864 #endif /* DEBUG */
865 
866 static int sync_callback(struct mount *mp, void *data);
867 
868 /*
869  * MPALMOSTSAFE
870  */
871 int
872 sys_sync(struct sync_args *uap)
873 {
874 	get_mplock();
875 	mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
876 #ifdef DEBUG
877 	/*
878 	 * print out buffer pool stat information on each sync() call.
879 	 */
880 	if (syncprt)
881 		vfs_bufstats();
882 #endif /* DEBUG */
883 	rel_mplock();
884 	return (0);
885 }
886 
887 static
888 int
889 sync_callback(struct mount *mp, void *data __unused)
890 {
891 	int asyncflag;
892 
893 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
894 		asyncflag = mp->mnt_flag & MNT_ASYNC;
895 		mp->mnt_flag &= ~MNT_ASYNC;
896 		vfs_msync(mp, MNT_NOWAIT);
897 		VFS_SYNC(mp, MNT_NOWAIT);
898 		mp->mnt_flag |= asyncflag;
899 	}
900 	return(0);
901 }
902 
903 /* XXX PRISON: could be per prison flag */
904 static int prison_quotas;
905 #if 0
906 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
907 #endif
908 
909 /*
910  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
911  *
912  * Change filesystem quotas.
913  *
914  * MPALMOSTSAFE
915  */
916 int
917 sys_quotactl(struct quotactl_args *uap)
918 {
919 	struct nlookupdata nd;
920 	struct thread *td;
921 	struct proc *p;
922 	struct mount *mp;
923 	int error;
924 
925 	get_mplock();
926 	td = curthread;
927 	p = td->td_proc;
928 	if (td->td_ucred->cr_prison && !prison_quotas) {
929 		error = EPERM;
930 		goto done;
931 	}
932 
933 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
934 	if (error == 0)
935 		error = nlookup(&nd);
936 	if (error == 0) {
937 		mp = nd.nl_nch.mount;
938 		error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
939 				    uap->arg, nd.nl_cred);
940 	}
941 	nlookup_done(&nd);
942 done:
943 	rel_mplock();
944 	return (error);
945 }
946 
947 /*
948  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
949  *		void *buf, int buflen)
950  *
951  * This function operates on a mount point and executes the specified
952  * operation using the specified control data, and possibly returns data.
953  *
954  * The actual number of bytes stored in the result buffer is returned, 0
955  * if none, otherwise an error is returned.
956  *
957  * MPALMOSTSAFE
958  */
959 int
960 sys_mountctl(struct mountctl_args *uap)
961 {
962 	struct thread *td = curthread;
963 	struct proc *p = td->td_proc;
964 	struct file *fp;
965 	void *ctl = NULL;
966 	void *buf = NULL;
967 	char *path = NULL;
968 	int error;
969 
970 	/*
971 	 * Sanity and permissions checks.  We must be root.
972 	 */
973 	KKASSERT(p);
974 	if (td->td_ucred->cr_prison != NULL)
975 		return (EPERM);
976 	if ((uap->op != MOUNTCTL_MOUNTFLAGS) &&
977 	    (error = priv_check(td, PRIV_ROOT)) != 0)
978 		return (error);
979 
980 	/*
981 	 * Argument length checks
982 	 */
983 	if (uap->ctllen < 0 || uap->ctllen > 1024)
984 		return (EINVAL);
985 	if (uap->buflen < 0 || uap->buflen > 16 * 1024)
986 		return (EINVAL);
987 	if (uap->path == NULL)
988 		return (EINVAL);
989 
990 	/*
991 	 * Allocate the necessary buffers and copyin data
992 	 */
993 	path = objcache_get(namei_oc, M_WAITOK);
994 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
995 	if (error)
996 		goto done;
997 
998 	if (uap->ctllen) {
999 		ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
1000 		error = copyin(uap->ctl, ctl, uap->ctllen);
1001 		if (error)
1002 			goto done;
1003 	}
1004 	if (uap->buflen)
1005 		buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
1006 
1007 	/*
1008 	 * Validate the descriptor
1009 	 */
1010 	if (uap->fd >= 0) {
1011 		fp = holdfp(p->p_fd, uap->fd, -1);
1012 		if (fp == NULL) {
1013 			error = EBADF;
1014 			goto done;
1015 		}
1016 	} else {
1017 		fp = NULL;
1018 	}
1019 
1020 	/*
1021 	 * Execute the internal kernel function and clean up.
1022 	 */
1023 	get_mplock();
1024 	error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
1025 	rel_mplock();
1026 	if (fp)
1027 		fdrop(fp);
1028 	if (error == 0 && uap->sysmsg_result > 0)
1029 		error = copyout(buf, uap->buf, uap->sysmsg_result);
1030 done:
1031 	if (path)
1032 		objcache_put(namei_oc, path);
1033 	if (ctl)
1034 		kfree(ctl, M_TEMP);
1035 	if (buf)
1036 		kfree(buf, M_TEMP);
1037 	return (error);
1038 }
1039 
1040 /*
1041  * Execute a mount control operation by resolving the path to a mount point
1042  * and calling vop_mountctl().
1043  *
1044  * Use the mount point from the nch instead of the vnode so nullfs mounts
1045  * can properly spike the VOP.
1046  */
1047 int
1048 kern_mountctl(const char *path, int op, struct file *fp,
1049 		const void *ctl, int ctllen,
1050 		void *buf, int buflen, int *res)
1051 {
1052 	struct vnode *vp;
1053 	struct mount *mp;
1054 	struct nlookupdata nd;
1055 	int error;
1056 
1057 	*res = 0;
1058 	vp = NULL;
1059 	error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
1060 	if (error == 0)
1061 		error = nlookup(&nd);
1062 	if (error == 0)
1063 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
1064 	mp = nd.nl_nch.mount;
1065 	nlookup_done(&nd);
1066 	if (error)
1067 		return (error);
1068 	vn_unlock(vp);
1069 
1070 	/*
1071 	 * Must be the root of the filesystem
1072 	 */
1073 	if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
1074 		vrele(vp);
1075 		return (EINVAL);
1076 	}
1077 	error = vop_mountctl(mp->mnt_vn_use_ops, vp, op, fp, ctl, ctllen,
1078 			     buf, buflen, res);
1079 	vrele(vp);
1080 	return (error);
1081 }
1082 
1083 int
1084 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1085 {
1086 	struct thread *td = curthread;
1087 	struct proc *p = td->td_proc;
1088 	struct mount *mp;
1089 	struct statfs *sp;
1090 	char *fullpath, *freepath;
1091 	int error;
1092 
1093 	if ((error = nlookup(nd)) != 0)
1094 		return (error);
1095 	mp = nd->nl_nch.mount;
1096 	sp = &mp->mnt_stat;
1097 	if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1098 		return (error);
1099 
1100 	error = mount_path(p, mp, &fullpath, &freepath);
1101 	if (error)
1102 		return(error);
1103 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1104 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1105 	kfree(freepath, M_TEMP);
1106 
1107 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1108 	bcopy(sp, buf, sizeof(*buf));
1109 	/* Only root should have access to the fsid's. */
1110 	if (priv_check(td, PRIV_ROOT))
1111 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1112 	return (0);
1113 }
1114 
1115 /*
1116  * statfs_args(char *path, struct statfs *buf)
1117  *
1118  * Get filesystem statistics.
1119  *
1120  * MPALMOSTSAFE
1121  */
1122 int
1123 sys_statfs(struct statfs_args *uap)
1124 {
1125 	struct nlookupdata nd;
1126 	struct statfs buf;
1127 	int error;
1128 
1129 	get_mplock();
1130 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1131 	if (error == 0)
1132 		error = kern_statfs(&nd, &buf);
1133 	nlookup_done(&nd);
1134 	if (error == 0)
1135 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1136 	rel_mplock();
1137 	return (error);
1138 }
1139 
1140 /*
1141  * MPALMOSTSAFE
1142  */
1143 int
1144 kern_fstatfs(int fd, struct statfs *buf)
1145 {
1146 	struct thread *td = curthread;
1147 	struct proc *p = td->td_proc;
1148 	struct file *fp;
1149 	struct mount *mp;
1150 	struct statfs *sp;
1151 	char *fullpath, *freepath;
1152 	int error;
1153 
1154 	KKASSERT(p);
1155 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1156 		return (error);
1157 	get_mplock();
1158 	mp = ((struct vnode *)fp->f_data)->v_mount;
1159 	if (mp == NULL) {
1160 		error = EBADF;
1161 		goto done;
1162 	}
1163 	if (fp->f_cred == NULL) {
1164 		error = EINVAL;
1165 		goto done;
1166 	}
1167 	sp = &mp->mnt_stat;
1168 	if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1169 		goto done;
1170 
1171 	if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1172 		goto done;
1173 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1174 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1175 	kfree(freepath, M_TEMP);
1176 
1177 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1178 	bcopy(sp, buf, sizeof(*buf));
1179 
1180 	/* Only root should have access to the fsid's. */
1181 	if (priv_check(td, PRIV_ROOT))
1182 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1183 	error = 0;
1184 done:
1185 	rel_mplock();
1186 	fdrop(fp);
1187 	return (error);
1188 }
1189 
1190 /*
1191  * fstatfs_args(int fd, struct statfs *buf)
1192  *
1193  * Get filesystem statistics.
1194  *
1195  * MPSAFE
1196  */
1197 int
1198 sys_fstatfs(struct fstatfs_args *uap)
1199 {
1200 	struct statfs buf;
1201 	int error;
1202 
1203 	error = kern_fstatfs(uap->fd, &buf);
1204 
1205 	if (error == 0)
1206 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1207 	return (error);
1208 }
1209 
1210 int
1211 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1212 {
1213 	struct mount *mp;
1214 	struct statvfs *sp;
1215 	int error;
1216 
1217 	if ((error = nlookup(nd)) != 0)
1218 		return (error);
1219 	mp = nd->nl_nch.mount;
1220 	sp = &mp->mnt_vstat;
1221 	if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1222 		return (error);
1223 
1224 	sp->f_flag = 0;
1225 	if (mp->mnt_flag & MNT_RDONLY)
1226 		sp->f_flag |= ST_RDONLY;
1227 	if (mp->mnt_flag & MNT_NOSUID)
1228 		sp->f_flag |= ST_NOSUID;
1229 	bcopy(sp, buf, sizeof(*buf));
1230 	return (0);
1231 }
1232 
1233 /*
1234  * statfs_args(char *path, struct statfs *buf)
1235  *
1236  * Get filesystem statistics.
1237  *
1238  * MPALMOSTSAFE
1239  */
1240 int
1241 sys_statvfs(struct statvfs_args *uap)
1242 {
1243 	struct nlookupdata nd;
1244 	struct statvfs buf;
1245 	int error;
1246 
1247 	get_mplock();
1248 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1249 	if (error == 0)
1250 		error = kern_statvfs(&nd, &buf);
1251 	nlookup_done(&nd);
1252 	if (error == 0)
1253 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1254 	rel_mplock();
1255 	return (error);
1256 }
1257 
1258 int
1259 kern_fstatvfs(int fd, struct statvfs *buf)
1260 {
1261 	struct thread *td = curthread;
1262 	struct proc *p = td->td_proc;
1263 	struct file *fp;
1264 	struct mount *mp;
1265 	struct statvfs *sp;
1266 	int error;
1267 
1268 	KKASSERT(p);
1269 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1270 		return (error);
1271 	mp = ((struct vnode *)fp->f_data)->v_mount;
1272 	if (mp == NULL) {
1273 		error = EBADF;
1274 		goto done;
1275 	}
1276 	if (fp->f_cred == NULL) {
1277 		error = EINVAL;
1278 		goto done;
1279 	}
1280 	sp = &mp->mnt_vstat;
1281 	if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1282 		goto done;
1283 
1284 	sp->f_flag = 0;
1285 	if (mp->mnt_flag & MNT_RDONLY)
1286 		sp->f_flag |= ST_RDONLY;
1287 	if (mp->mnt_flag & MNT_NOSUID)
1288 		sp->f_flag |= ST_NOSUID;
1289 
1290 	bcopy(sp, buf, sizeof(*buf));
1291 	error = 0;
1292 done:
1293 	fdrop(fp);
1294 	return (error);
1295 }
1296 
1297 /*
1298  * fstatfs_args(int fd, struct statfs *buf)
1299  *
1300  * Get filesystem statistics.
1301  *
1302  * MPALMOSTSAFE
1303  */
1304 int
1305 sys_fstatvfs(struct fstatvfs_args *uap)
1306 {
1307 	struct statvfs buf;
1308 	int error;
1309 
1310 	get_mplock();
1311 	error = kern_fstatvfs(uap->fd, &buf);
1312 	rel_mplock();
1313 
1314 	if (error == 0)
1315 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1316 	return (error);
1317 }
1318 
1319 /*
1320  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1321  *
1322  * Get statistics on all filesystems.
1323  */
1324 
1325 struct getfsstat_info {
1326 	struct statfs *sfsp;
1327 	long count;
1328 	long maxcount;
1329 	int error;
1330 	int flags;
1331 	struct thread *td;
1332 };
1333 
1334 static int getfsstat_callback(struct mount *, void *);
1335 
1336 /*
1337  * MPALMOSTSAFE
1338  */
1339 int
1340 sys_getfsstat(struct getfsstat_args *uap)
1341 {
1342 	struct thread *td = curthread;
1343 	struct getfsstat_info info;
1344 
1345 	bzero(&info, sizeof(info));
1346 
1347 	info.maxcount = uap->bufsize / sizeof(struct statfs);
1348 	info.sfsp = uap->buf;
1349 	info.count = 0;
1350 	info.flags = uap->flags;
1351 	info.td = td;
1352 
1353 	get_mplock();
1354 	mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1355 	rel_mplock();
1356 	if (info.sfsp && info.count > info.maxcount)
1357 		uap->sysmsg_result = info.maxcount;
1358 	else
1359 		uap->sysmsg_result = info.count;
1360 	return (info.error);
1361 }
1362 
1363 static int
1364 getfsstat_callback(struct mount *mp, void *data)
1365 {
1366 	struct getfsstat_info *info = data;
1367 	struct statfs *sp;
1368 	char *freepath;
1369 	char *fullpath;
1370 	int error;
1371 
1372 	if (info->sfsp && info->count < info->maxcount) {
1373 		if (info->td->td_proc &&
1374 		    !chroot_visible_mnt(mp, info->td->td_proc)) {
1375 			return(0);
1376 		}
1377 		sp = &mp->mnt_stat;
1378 
1379 		/*
1380 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1381 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1382 		 * overrides MNT_WAIT.
1383 		 */
1384 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1385 		    (info->flags & MNT_WAIT)) &&
1386 		    (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1387 			return(0);
1388 		}
1389 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1390 
1391 		error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1392 		if (error) {
1393 			info->error = error;
1394 			return(-1);
1395 		}
1396 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1397 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1398 		kfree(freepath, M_TEMP);
1399 
1400 		error = copyout(sp, info->sfsp, sizeof(*sp));
1401 		if (error) {
1402 			info->error = error;
1403 			return (-1);
1404 		}
1405 		++info->sfsp;
1406 	}
1407 	info->count++;
1408 	return(0);
1409 }
1410 
1411 /*
1412  * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1413 		   long bufsize, int flags)
1414  *
1415  * Get statistics on all filesystems.
1416  */
1417 
1418 struct getvfsstat_info {
1419 	struct statfs *sfsp;
1420 	struct statvfs *vsfsp;
1421 	long count;
1422 	long maxcount;
1423 	int error;
1424 	int flags;
1425 	struct thread *td;
1426 };
1427 
1428 static int getvfsstat_callback(struct mount *, void *);
1429 
1430 /*
1431  * MPALMOSTSAFE
1432  */
1433 int
1434 sys_getvfsstat(struct getvfsstat_args *uap)
1435 {
1436 	struct thread *td = curthread;
1437 	struct getvfsstat_info info;
1438 
1439 	bzero(&info, sizeof(info));
1440 
1441 	info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1442 	info.sfsp = uap->buf;
1443 	info.vsfsp = uap->vbuf;
1444 	info.count = 0;
1445 	info.flags = uap->flags;
1446 	info.td = td;
1447 
1448 	get_mplock();
1449 	mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1450 	if (info.vsfsp && info.count > info.maxcount)
1451 		uap->sysmsg_result = info.maxcount;
1452 	else
1453 		uap->sysmsg_result = info.count;
1454 	rel_mplock();
1455 	return (info.error);
1456 }
1457 
1458 static int
1459 getvfsstat_callback(struct mount *mp, void *data)
1460 {
1461 	struct getvfsstat_info *info = data;
1462 	struct statfs *sp;
1463 	struct statvfs *vsp;
1464 	char *freepath;
1465 	char *fullpath;
1466 	int error;
1467 
1468 	if (info->vsfsp && info->count < info->maxcount) {
1469 		if (info->td->td_proc &&
1470 		    !chroot_visible_mnt(mp, info->td->td_proc)) {
1471 			return(0);
1472 		}
1473 		sp = &mp->mnt_stat;
1474 		vsp = &mp->mnt_vstat;
1475 
1476 		/*
1477 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1478 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1479 		 * overrides MNT_WAIT.
1480 		 */
1481 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1482 		    (info->flags & MNT_WAIT)) &&
1483 		    (error = VFS_STATFS(mp, sp, info->td->td_ucred))) {
1484 			return(0);
1485 		}
1486 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1487 
1488 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1489 		    (info->flags & MNT_WAIT)) &&
1490 		    (error = VFS_STATVFS(mp, vsp, info->td->td_ucred))) {
1491 			return(0);
1492 		}
1493 		vsp->f_flag = 0;
1494 		if (mp->mnt_flag & MNT_RDONLY)
1495 			vsp->f_flag |= ST_RDONLY;
1496 		if (mp->mnt_flag & MNT_NOSUID)
1497 			vsp->f_flag |= ST_NOSUID;
1498 
1499 		error = mount_path(info->td->td_proc, mp, &fullpath, &freepath);
1500 		if (error) {
1501 			info->error = error;
1502 			return(-1);
1503 		}
1504 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1505 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1506 		kfree(freepath, M_TEMP);
1507 
1508 		error = copyout(sp, info->sfsp, sizeof(*sp));
1509 		if (error == 0)
1510 			error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1511 		if (error) {
1512 			info->error = error;
1513 			return (-1);
1514 		}
1515 		++info->sfsp;
1516 		++info->vsfsp;
1517 	}
1518 	info->count++;
1519 	return(0);
1520 }
1521 
1522 
1523 /*
1524  * fchdir_args(int fd)
1525  *
1526  * Change current working directory to a given file descriptor.
1527  *
1528  * MPALMOSTSAFE
1529  */
1530 int
1531 sys_fchdir(struct fchdir_args *uap)
1532 {
1533 	struct thread *td = curthread;
1534 	struct proc *p = td->td_proc;
1535 	struct filedesc *fdp = p->p_fd;
1536 	struct vnode *vp, *ovp;
1537 	struct mount *mp;
1538 	struct file *fp;
1539 	struct nchandle nch, onch, tnch;
1540 	int error;
1541 
1542 	if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1543 		return (error);
1544 	get_mplock();
1545 	vp = (struct vnode *)fp->f_data;
1546 	vref(vp);
1547 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1548 	if (fp->f_nchandle.ncp == NULL)
1549 		error = ENOTDIR;
1550 	else
1551 		error = checkvp_chdir(vp, td);
1552 	if (error) {
1553 		vput(vp);
1554 		goto done;
1555 	}
1556 	cache_copy(&fp->f_nchandle, &nch);
1557 
1558 	/*
1559 	 * If the ncp has become a mount point, traverse through
1560 	 * the mount point.
1561 	 */
1562 
1563 	while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1564 	       (mp = cache_findmount(&nch)) != NULL
1565 	) {
1566 		error = nlookup_mp(mp, &tnch);
1567 		if (error == 0) {
1568 			cache_unlock(&tnch);	/* leave ref intact */
1569 			vput(vp);
1570 			vp = tnch.ncp->nc_vp;
1571 			error = vget(vp, LK_SHARED);
1572 			KKASSERT(error == 0);
1573 			cache_drop(&nch);
1574 			nch = tnch;
1575 		}
1576 	}
1577 	if (error == 0) {
1578 		ovp = fdp->fd_cdir;
1579 		onch = fdp->fd_ncdir;
1580 		vn_unlock(vp);		/* leave ref intact */
1581 		fdp->fd_cdir = vp;
1582 		fdp->fd_ncdir = nch;
1583 		cache_drop(&onch);
1584 		vrele(ovp);
1585 	} else {
1586 		cache_drop(&nch);
1587 		vput(vp);
1588 	}
1589 	fdrop(fp);
1590 done:
1591 	rel_mplock();
1592 	return (error);
1593 }
1594 
1595 int
1596 kern_chdir(struct nlookupdata *nd)
1597 {
1598 	struct thread *td = curthread;
1599 	struct proc *p = td->td_proc;
1600 	struct filedesc *fdp = p->p_fd;
1601 	struct vnode *vp, *ovp;
1602 	struct nchandle onch;
1603 	int error;
1604 
1605 	if ((error = nlookup(nd)) != 0)
1606 		return (error);
1607 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1608 		return (ENOENT);
1609 	if ((error = vget(vp, LK_SHARED)) != 0)
1610 		return (error);
1611 
1612 	error = checkvp_chdir(vp, td);
1613 	vn_unlock(vp);
1614 	if (error == 0) {
1615 		ovp = fdp->fd_cdir;
1616 		onch = fdp->fd_ncdir;
1617 		cache_unlock(&nd->nl_nch);	/* leave reference intact */
1618 		fdp->fd_ncdir = nd->nl_nch;
1619 		fdp->fd_cdir = vp;
1620 		cache_drop(&onch);
1621 		vrele(ovp);
1622 		cache_zero(&nd->nl_nch);
1623 	} else {
1624 		vrele(vp);
1625 	}
1626 	return (error);
1627 }
1628 
1629 /*
1630  * chdir_args(char *path)
1631  *
1632  * Change current working directory (``.'').
1633  *
1634  * MPALMOSTSAFE
1635  */
1636 int
1637 sys_chdir(struct chdir_args *uap)
1638 {
1639 	struct nlookupdata nd;
1640 	int error;
1641 
1642 	get_mplock();
1643 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1644 	if (error == 0)
1645 		error = kern_chdir(&nd);
1646 	nlookup_done(&nd);
1647 	rel_mplock();
1648 	return (error);
1649 }
1650 
1651 /*
1652  * Helper function for raised chroot(2) security function:  Refuse if
1653  * any filedescriptors are open directories.
1654  */
1655 static int
1656 chroot_refuse_vdir_fds(struct filedesc *fdp)
1657 {
1658 	struct vnode *vp;
1659 	struct file *fp;
1660 	int error;
1661 	int fd;
1662 
1663 	for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1664 		if ((error = holdvnode(fdp, fd, &fp)) != 0)
1665 			continue;
1666 		vp = (struct vnode *)fp->f_data;
1667 		if (vp->v_type != VDIR) {
1668 			fdrop(fp);
1669 			continue;
1670 		}
1671 		fdrop(fp);
1672 		return(EPERM);
1673 	}
1674 	return (0);
1675 }
1676 
1677 /*
1678  * This sysctl determines if we will allow a process to chroot(2) if it
1679  * has a directory open:
1680  *	0: disallowed for all processes.
1681  *	1: allowed for processes that were not already chroot(2)'ed.
1682  *	2: allowed for all processes.
1683  */
1684 
1685 static int chroot_allow_open_directories = 1;
1686 
1687 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1688      &chroot_allow_open_directories, 0, "");
1689 
1690 /*
1691  * chroot to the specified namecache entry.  We obtain the vp from the
1692  * namecache data.  The passed ncp must be locked and referenced and will
1693  * remain locked and referenced on return.
1694  */
1695 int
1696 kern_chroot(struct nchandle *nch)
1697 {
1698 	struct thread *td = curthread;
1699 	struct proc *p = td->td_proc;
1700 	struct filedesc *fdp = p->p_fd;
1701 	struct vnode *vp;
1702 	int error;
1703 
1704 	/*
1705 	 * Only privileged user can chroot
1706 	 */
1707 	error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1708 	if (error)
1709 		return (error);
1710 
1711 	/*
1712 	 * Disallow open directory descriptors (fchdir() breakouts).
1713 	 */
1714 	if (chroot_allow_open_directories == 0 ||
1715 	   (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1716 		if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1717 			return (error);
1718 	}
1719 	if ((vp = nch->ncp->nc_vp) == NULL)
1720 		return (ENOENT);
1721 
1722 	if ((error = vget(vp, LK_SHARED)) != 0)
1723 		return (error);
1724 
1725 	/*
1726 	 * Check the validity of vp as a directory to change to and
1727 	 * associate it with rdir/jdir.
1728 	 */
1729 	error = checkvp_chdir(vp, td);
1730 	vn_unlock(vp);			/* leave reference intact */
1731 	if (error == 0) {
1732 		vrele(fdp->fd_rdir);
1733 		fdp->fd_rdir = vp;	/* reference inherited by fd_rdir */
1734 		cache_drop(&fdp->fd_nrdir);
1735 		cache_copy(nch, &fdp->fd_nrdir);
1736 		if (fdp->fd_jdir == NULL) {
1737 			fdp->fd_jdir = vp;
1738 			vref(fdp->fd_jdir);
1739 			cache_copy(nch, &fdp->fd_njdir);
1740 		}
1741 	} else {
1742 		vrele(vp);
1743 	}
1744 	return (error);
1745 }
1746 
1747 /*
1748  * chroot_args(char *path)
1749  *
1750  * Change notion of root (``/'') directory.
1751  *
1752  * MPALMOSTSAFE
1753  */
1754 int
1755 sys_chroot(struct chroot_args *uap)
1756 {
1757 	struct thread *td __debugvar = curthread;
1758 	struct nlookupdata nd;
1759 	int error;
1760 
1761 	KKASSERT(td->td_proc);
1762 	get_mplock();
1763 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1764 	if (error == 0) {
1765 		nd.nl_flags |= NLC_EXEC;
1766 		error = nlookup(&nd);
1767 		if (error == 0)
1768 			error = kern_chroot(&nd.nl_nch);
1769 	}
1770 	nlookup_done(&nd);
1771 	rel_mplock();
1772 	return(error);
1773 }
1774 
1775 int
1776 sys_chroot_kernel(struct chroot_kernel_args *uap)
1777 {
1778 	struct thread *td = curthread;
1779 	struct nlookupdata nd;
1780 	struct nchandle *nch;
1781 	struct vnode *vp;
1782 	int error;
1783 
1784 	get_mplock();
1785 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1786 	if (error)
1787 		goto error_nond;
1788 
1789 	error = nlookup(&nd);
1790 	if (error)
1791 		goto error_out;
1792 
1793 	nch = &nd.nl_nch;
1794 
1795 	error = priv_check_cred(td->td_ucred, PRIV_VFS_CHROOT, 0);
1796 	if (error)
1797 		goto error_out;
1798 
1799 	if ((vp = nch->ncp->nc_vp) == NULL) {
1800 		error = ENOENT;
1801 		goto error_out;
1802 	}
1803 
1804 	if ((error = cache_vref(nch, nd.nl_cred, &vp)) != 0)
1805 		goto error_out;
1806 
1807 	kprintf("chroot_kernel: set new rootnch/rootvnode to %s\n", uap->path);
1808 	vfs_cache_setroot(vp, cache_hold(nch));
1809 
1810 error_out:
1811 	nlookup_done(&nd);
1812 error_nond:
1813 	rel_mplock();
1814 	return(error);
1815 }
1816 
1817 /*
1818  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1819  * determine whether it is legal to chdir to the vnode.  The vnode's state
1820  * is not changed by this call.
1821  */
1822 int
1823 checkvp_chdir(struct vnode *vp, struct thread *td)
1824 {
1825 	int error;
1826 
1827 	if (vp->v_type != VDIR)
1828 		error = ENOTDIR;
1829 	else
1830 		error = VOP_EACCESS(vp, VEXEC, td->td_ucred);
1831 	return (error);
1832 }
1833 
1834 /*
1835  * MPSAFE
1836  */
1837 int
1838 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1839 {
1840 	struct thread *td = curthread;
1841 	struct proc *p = td->td_proc;
1842 	struct lwp *lp = td->td_lwp;
1843 	struct filedesc *fdp = p->p_fd;
1844 	int cmode, flags;
1845 	struct file *nfp;
1846 	struct file *fp;
1847 	struct vnode *vp;
1848 	int type, indx, error;
1849 	struct flock lf;
1850 
1851 	if ((oflags & O_ACCMODE) == O_ACCMODE)
1852 		return (EINVAL);
1853 	flags = FFLAGS(oflags);
1854 	error = falloc(lp, &nfp, NULL);
1855 	if (error)
1856 		return (error);
1857 	fp = nfp;
1858 	cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT;
1859 
1860 	/*
1861 	 * XXX p_dupfd is a real mess.  It allows a device to return a
1862 	 * file descriptor to be duplicated rather then doing the open
1863 	 * itself.
1864 	 */
1865 	lp->lwp_dupfd = -1;
1866 
1867 	/*
1868 	 * Call vn_open() to do the lookup and assign the vnode to the
1869 	 * file pointer.  vn_open() does not change the ref count on fp
1870 	 * and the vnode, on success, will be inherited by the file pointer
1871 	 * and unlocked.
1872 	 */
1873 	nd->nl_flags |= NLC_LOCKVP;
1874 	error = vn_open(nd, fp, flags, cmode);
1875 	nlookup_done(nd);
1876 	if (error) {
1877 		/*
1878 		 * handle special fdopen() case.  bleh.  dupfdopen() is
1879 		 * responsible for dropping the old contents of ofiles[indx]
1880 		 * if it succeeds.
1881 		 *
1882 		 * Note that fsetfd() will add a ref to fp which represents
1883 		 * the fd_files[] assignment.  We must still drop our
1884 		 * reference.
1885 		 */
1886 		if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1887 			if (fdalloc(p, 0, &indx) == 0) {
1888 				error = dupfdopen(fdp, indx, lp->lwp_dupfd, flags, error);
1889 				if (error == 0) {
1890 					*res = indx;
1891 					fdrop(fp);	/* our ref */
1892 					return (0);
1893 				}
1894 				fsetfd(fdp, NULL, indx);
1895 			}
1896 		}
1897 		fdrop(fp);	/* our ref */
1898 		if (error == ERESTART)
1899 			error = EINTR;
1900 		return (error);
1901 	}
1902 
1903 	/*
1904 	 * ref the vnode for ourselves so it can't be ripped out from under
1905 	 * is.  XXX need an ND flag to request that the vnode be returned
1906 	 * anyway.
1907 	 *
1908 	 * Reserve a file descriptor but do not assign it until the open
1909 	 * succeeds.
1910 	 */
1911 	vp = (struct vnode *)fp->f_data;
1912 	vref(vp);
1913 	if ((error = fdalloc(p, 0, &indx)) != 0) {
1914 		fdrop(fp);
1915 		vrele(vp);
1916 		return (error);
1917 	}
1918 
1919 	/*
1920 	 * If no error occurs the vp will have been assigned to the file
1921 	 * pointer.
1922 	 */
1923 	lp->lwp_dupfd = 0;
1924 
1925 	if (flags & (O_EXLOCK | O_SHLOCK)) {
1926 		lf.l_whence = SEEK_SET;
1927 		lf.l_start = 0;
1928 		lf.l_len = 0;
1929 		if (flags & O_EXLOCK)
1930 			lf.l_type = F_WRLCK;
1931 		else
1932 			lf.l_type = F_RDLCK;
1933 		if (flags & FNONBLOCK)
1934 			type = 0;
1935 		else
1936 			type = F_WAIT;
1937 
1938 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1939 			/*
1940 			 * lock request failed.  Clean up the reserved
1941 			 * descriptor.
1942 			 */
1943 			vrele(vp);
1944 			fsetfd(fdp, NULL, indx);
1945 			fdrop(fp);
1946 			return (error);
1947 		}
1948 		fp->f_flag |= FHASLOCK;
1949 	}
1950 #if 0
1951 	/*
1952 	 * Assert that all regular file vnodes were created with a object.
1953 	 */
1954 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1955 		("open: regular file has no backing object after vn_open"));
1956 #endif
1957 
1958 	vrele(vp);
1959 
1960 	/*
1961 	 * release our private reference, leaving the one associated with the
1962 	 * descriptor table intact.
1963 	 */
1964 	fsetfd(fdp, fp, indx);
1965 	fdrop(fp);
1966 	*res = indx;
1967 	return (0);
1968 }
1969 
1970 /*
1971  * open_args(char *path, int flags, int mode)
1972  *
1973  * Check permissions, allocate an open file structure,
1974  * and call the device open routine if any.
1975  *
1976  * MPALMOSTSAFE
1977  */
1978 int
1979 sys_open(struct open_args *uap)
1980 {
1981 	CACHE_MPLOCK_DECLARE;
1982 	struct nlookupdata nd;
1983 	int error;
1984 
1985 	CACHE_GETMPLOCK1();
1986 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1987 	if (error == 0) {
1988 		error = kern_open(&nd, uap->flags,
1989 				    uap->mode, &uap->sysmsg_result);
1990 	}
1991 	nlookup_done(&nd);
1992 	CACHE_RELMPLOCK();
1993 	return (error);
1994 }
1995 
1996 /*
1997  * openat_args(int fd, char *path, int flags, int mode)
1998  *
1999  * MPALMOSTSAFE
2000  */
2001 int
2002 sys_openat(struct openat_args *uap)
2003 {
2004 	CACHE_MPLOCK_DECLARE;
2005 	struct nlookupdata nd;
2006 	int error;
2007 	struct file *fp;
2008 
2009 	CACHE_GETMPLOCK1();
2010 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2011 	if (error == 0) {
2012 		error = kern_open(&nd, uap->flags, uap->mode,
2013 					&uap->sysmsg_result);
2014 	}
2015 	nlookup_done_at(&nd, fp);
2016 	CACHE_RELMPLOCK();
2017 	return (error);
2018 }
2019 
2020 int
2021 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
2022 {
2023 	struct thread *td = curthread;
2024 	struct proc *p = td->td_proc;
2025 	struct vnode *vp;
2026 	struct vattr vattr;
2027 	int error;
2028 	int whiteout = 0;
2029 
2030 	KKASSERT(p);
2031 
2032 	VATTR_NULL(&vattr);
2033 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2034 	vattr.va_rmajor = rmajor;
2035 	vattr.va_rminor = rminor;
2036 
2037 	switch (mode & S_IFMT) {
2038 	case S_IFMT:	/* used by badsect to flag bad sectors */
2039 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_BAD, 0);
2040 		vattr.va_type = VBAD;
2041 		break;
2042 	case S_IFCHR:
2043 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2044 		vattr.va_type = VCHR;
2045 		break;
2046 	case S_IFBLK:
2047 		error = priv_check(td, PRIV_VFS_MKNOD_DEV);
2048 		vattr.va_type = VBLK;
2049 		break;
2050 	case S_IFWHT:
2051 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_WHT, 0);
2052 		whiteout = 1;
2053 		break;
2054 	case S_IFDIR:	/* special directories support for HAMMER */
2055 		error = priv_check_cred(td->td_ucred, PRIV_VFS_MKNOD_DIR, 0);
2056 		vattr.va_type = VDIR;
2057 		break;
2058 	default:
2059 		error = EINVAL;
2060 		break;
2061 	}
2062 
2063 	if (error)
2064 		return (error);
2065 
2066 	bwillinode(1);
2067 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2068 	if ((error = nlookup(nd)) != 0)
2069 		return (error);
2070 	if (nd->nl_nch.ncp->nc_vp)
2071 		return (EEXIST);
2072 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2073 		return (error);
2074 
2075 	if (whiteout) {
2076 		error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
2077 				      nd->nl_cred, NAMEI_CREATE);
2078 	} else {
2079 		vp = NULL;
2080 		error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
2081 				   &vp, nd->nl_cred, &vattr);
2082 		if (error == 0)
2083 			vput(vp);
2084 	}
2085 	return (error);
2086 }
2087 
2088 /*
2089  * mknod_args(char *path, int mode, int dev)
2090  *
2091  * Create a special file.
2092  *
2093  * MPALMOSTSAFE
2094  */
2095 int
2096 sys_mknod(struct mknod_args *uap)
2097 {
2098 	struct nlookupdata nd;
2099 	int error;
2100 
2101 	get_mplock();
2102 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2103 	if (error == 0) {
2104 		error = kern_mknod(&nd, uap->mode,
2105 				   umajor(uap->dev), uminor(uap->dev));
2106 	}
2107 	nlookup_done(&nd);
2108 	rel_mplock();
2109 	return (error);
2110 }
2111 
2112 /*
2113  * mknodat_args(int fd, char *path, mode_t mode, dev_t dev)
2114  *
2115  * Create a special file.  The path is relative to the directory associated
2116  * with fd.
2117  *
2118  * MPALMOSTSAFE
2119  */
2120 int
2121 sys_mknodat(struct mknodat_args *uap)
2122 {
2123 	struct nlookupdata nd;
2124 	struct file *fp;
2125 	int error;
2126 
2127 	get_mplock();
2128 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2129 	if (error == 0) {
2130 		error = kern_mknod(&nd, uap->mode,
2131 				   umajor(uap->dev), uminor(uap->dev));
2132 	}
2133 	nlookup_done_at(&nd, fp);
2134 	rel_mplock();
2135 	return (error);
2136 }
2137 
2138 int
2139 kern_mkfifo(struct nlookupdata *nd, int mode)
2140 {
2141 	struct thread *td = curthread;
2142 	struct proc *p = td->td_proc;
2143 	struct vattr vattr;
2144 	struct vnode *vp;
2145 	int error;
2146 
2147 	bwillinode(1);
2148 
2149 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2150 	if ((error = nlookup(nd)) != 0)
2151 		return (error);
2152 	if (nd->nl_nch.ncp->nc_vp)
2153 		return (EEXIST);
2154 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2155 		return (error);
2156 
2157 	VATTR_NULL(&vattr);
2158 	vattr.va_type = VFIFO;
2159 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
2160 	vp = NULL;
2161 	error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
2162 	if (error == 0)
2163 		vput(vp);
2164 	return (error);
2165 }
2166 
2167 /*
2168  * mkfifo_args(char *path, int mode)
2169  *
2170  * Create a named pipe.
2171  *
2172  * MPALMOSTSAFE
2173  */
2174 int
2175 sys_mkfifo(struct mkfifo_args *uap)
2176 {
2177 	struct nlookupdata nd;
2178 	int error;
2179 
2180 	get_mplock();
2181 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2182 	if (error == 0)
2183 		error = kern_mkfifo(&nd, uap->mode);
2184 	nlookup_done(&nd);
2185 	rel_mplock();
2186 	return (error);
2187 }
2188 
2189 /*
2190  * mkfifoat_args(int fd, char *path, mode_t mode)
2191  *
2192  * Create a named pipe.  The path is relative to the directory associated
2193  * with fd.
2194  *
2195  * MPALMOSTSAFE
2196  */
2197 int
2198 sys_mkfifoat(struct mkfifoat_args *uap)
2199 {
2200 	struct nlookupdata nd;
2201 	struct file *fp;
2202 	int error;
2203 
2204 	get_mplock();
2205 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2206 	if (error == 0)
2207 		error = kern_mkfifo(&nd, uap->mode);
2208 	nlookup_done_at(&nd, fp);
2209 	rel_mplock();
2210 	return (error);
2211 }
2212 
2213 static int hardlink_check_uid = 0;
2214 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
2215     &hardlink_check_uid, 0,
2216     "Unprivileged processes cannot create hard links to files owned by other "
2217     "users");
2218 static int hardlink_check_gid = 0;
2219 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
2220     &hardlink_check_gid, 0,
2221     "Unprivileged processes cannot create hard links to files owned by other "
2222     "groups");
2223 
2224 static int
2225 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
2226 {
2227 	struct vattr va;
2228 	int error;
2229 
2230 	/*
2231 	 * Shortcut if disabled
2232 	 */
2233 	if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
2234 		return (0);
2235 
2236 	/*
2237 	 * Privileged user can always hardlink
2238 	 */
2239 	if (priv_check_cred(cred, PRIV_VFS_LINK, 0) == 0)
2240 		return (0);
2241 
2242 	/*
2243 	 * Otherwise only if the originating file is owned by the
2244 	 * same user or group.  Note that any group is allowed if
2245 	 * the file is owned by the caller.
2246 	 */
2247 	error = VOP_GETATTR(vp, &va);
2248 	if (error != 0)
2249 		return (error);
2250 
2251 	if (hardlink_check_uid) {
2252 		if (cred->cr_uid != va.va_uid)
2253 			return (EPERM);
2254 	}
2255 
2256 	if (hardlink_check_gid) {
2257 		if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2258 			return (EPERM);
2259 	}
2260 
2261 	return (0);
2262 }
2263 
2264 int
2265 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2266 {
2267 	struct thread *td = curthread;
2268 	struct vnode *vp;
2269 	int error;
2270 
2271 	/*
2272 	 * Lookup the source and obtained a locked vnode.
2273 	 *
2274 	 * You may only hardlink a file which you have write permission
2275 	 * on or which you own.
2276 	 *
2277 	 * XXX relookup on vget failure / race ?
2278 	 */
2279 	bwillinode(1);
2280 	nd->nl_flags |= NLC_WRITE | NLC_OWN | NLC_HLINK;
2281 	if ((error = nlookup(nd)) != 0)
2282 		return (error);
2283 	vp = nd->nl_nch.ncp->nc_vp;
2284 	KKASSERT(vp != NULL);
2285 	if (vp->v_type == VDIR)
2286 		return (EPERM);		/* POSIX */
2287 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2288 		return (error);
2289 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2290 		return (error);
2291 
2292 	/*
2293 	 * Unlock the source so we can lookup the target without deadlocking
2294 	 * (XXX vp is locked already, possible other deadlock?).  The target
2295 	 * must not exist.
2296 	 */
2297 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2298 	nd->nl_flags &= ~NLC_NCPISLOCKED;
2299 	cache_unlock(&nd->nl_nch);
2300 	vn_unlock(vp);
2301 
2302 	linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2303 	if ((error = nlookup(linknd)) != 0) {
2304 		vrele(vp);
2305 		return (error);
2306 	}
2307 	if (linknd->nl_nch.ncp->nc_vp) {
2308 		vrele(vp);
2309 		return (EEXIST);
2310 	}
2311 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
2312 		vrele(vp);
2313 		return (error);
2314 	}
2315 
2316 	/*
2317 	 * Finally run the new API VOP.
2318 	 */
2319 	error = can_hardlink(vp, td, td->td_ucred);
2320 	if (error == 0) {
2321 		error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2322 				  vp, linknd->nl_cred);
2323 	}
2324 	vput(vp);
2325 	return (error);
2326 }
2327 
2328 /*
2329  * link_args(char *path, char *link)
2330  *
2331  * Make a hard file link.
2332  *
2333  * MPALMOSTSAFE
2334  */
2335 int
2336 sys_link(struct link_args *uap)
2337 {
2338 	struct nlookupdata nd, linknd;
2339 	int error;
2340 
2341 	get_mplock();
2342 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2343 	if (error == 0) {
2344 		error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2345 		if (error == 0)
2346 			error = kern_link(&nd, &linknd);
2347 		nlookup_done(&linknd);
2348 	}
2349 	nlookup_done(&nd);
2350 	rel_mplock();
2351 	return (error);
2352 }
2353 
2354 int
2355 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2356 {
2357 	struct vattr vattr;
2358 	struct vnode *vp;
2359 	struct vnode *dvp;
2360 	int error;
2361 
2362 	bwillinode(1);
2363 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2364 	if ((error = nlookup(nd)) != 0)
2365 		return (error);
2366 	if (nd->nl_nch.ncp->nc_vp)
2367 		return (EEXIST);
2368 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2369 		return (error);
2370 	dvp = nd->nl_dvp;
2371 	VATTR_NULL(&vattr);
2372 	vattr.va_mode = mode;
2373 	error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2374 	if (error == 0)
2375 		vput(vp);
2376 	return (error);
2377 }
2378 
2379 /*
2380  * symlink(char *path, char *link)
2381  *
2382  * Make a symbolic link.
2383  *
2384  * MPALMOSTSAFE
2385  */
2386 int
2387 sys_symlink(struct symlink_args *uap)
2388 {
2389 	struct thread *td = curthread;
2390 	struct nlookupdata nd;
2391 	char *path;
2392 	int error;
2393 	int mode;
2394 
2395 	path = objcache_get(namei_oc, M_WAITOK);
2396 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2397 	if (error == 0) {
2398 		get_mplock();
2399 		error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2400 		if (error == 0) {
2401 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2402 			error = kern_symlink(&nd, path, mode);
2403 		}
2404 		nlookup_done(&nd);
2405 		rel_mplock();
2406 	}
2407 	objcache_put(namei_oc, path);
2408 	return (error);
2409 }
2410 
2411 /*
2412  * symlinkat_args(char *path1, int fd, char *path2)
2413  *
2414  * Make a symbolic link.  The path2 argument is relative to the directory
2415  * associated with fd.
2416  *
2417  * MPALMOSTSAFE
2418  */
2419 int
2420 sys_symlinkat(struct symlinkat_args *uap)
2421 {
2422 	struct thread *td = curthread;
2423 	struct nlookupdata nd;
2424 	struct file *fp;
2425 	char *path1;
2426 	int error;
2427 	int mode;
2428 
2429 	path1 = objcache_get(namei_oc, M_WAITOK);
2430 	error = copyinstr(uap->path1, path1, MAXPATHLEN, NULL);
2431 	if (error == 0) {
2432 		get_mplock();
2433 		error = nlookup_init_at(&nd, &fp, uap->fd, uap->path2,
2434 		    UIO_USERSPACE, 0);
2435 		if (error == 0) {
2436 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2437 			error = kern_symlink(&nd, path1, mode);
2438 		}
2439 		nlookup_done_at(&nd, fp);
2440 		rel_mplock();
2441 	}
2442 	objcache_put(namei_oc, path1);
2443 	return (error);
2444 }
2445 
2446 /*
2447  * undelete_args(char *path)
2448  *
2449  * Delete a whiteout from the filesystem.
2450  *
2451  * MPALMOSTSAFE
2452  */
2453 int
2454 sys_undelete(struct undelete_args *uap)
2455 {
2456 	struct nlookupdata nd;
2457 	int error;
2458 
2459 	get_mplock();
2460 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2461 	bwillinode(1);
2462 	nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2463 	if (error == 0)
2464 		error = nlookup(&nd);
2465 	if (error == 0)
2466 		error = ncp_writechk(&nd.nl_nch);
2467 	if (error == 0) {
2468 		error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2469 				      NAMEI_DELETE);
2470 	}
2471 	nlookup_done(&nd);
2472 	rel_mplock();
2473 	return (error);
2474 }
2475 
2476 int
2477 kern_unlink(struct nlookupdata *nd)
2478 {
2479 	int error;
2480 
2481 	bwillinode(1);
2482 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2483 	if ((error = nlookup(nd)) != 0)
2484 		return (error);
2485 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2486 		return (error);
2487 	error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2488 	return (error);
2489 }
2490 
2491 /*
2492  * unlink_args(char *path)
2493  *
2494  * Delete a name from the filesystem.
2495  *
2496  * MPALMOSTSAFE
2497  */
2498 int
2499 sys_unlink(struct unlink_args *uap)
2500 {
2501 	struct nlookupdata nd;
2502 	int error;
2503 
2504 	get_mplock();
2505 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2506 	if (error == 0)
2507 		error = kern_unlink(&nd);
2508 	nlookup_done(&nd);
2509 	rel_mplock();
2510 	return (error);
2511 }
2512 
2513 
2514 /*
2515  * unlinkat_args(int fd, char *path, int flags)
2516  *
2517  * Delete the file or directory entry pointed to by fd/path.
2518  *
2519  * MPALMOSTSAFE
2520  */
2521 int
2522 sys_unlinkat(struct unlinkat_args *uap)
2523 {
2524 	struct nlookupdata nd;
2525 	struct file *fp;
2526 	int error;
2527 
2528 	if (uap->flags & ~AT_REMOVEDIR)
2529 		return (EINVAL);
2530 
2531 	get_mplock();
2532 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2533 	if (error == 0) {
2534 		if (uap->flags & AT_REMOVEDIR)
2535 			error = kern_rmdir(&nd);
2536 		else
2537 			error = kern_unlink(&nd);
2538 	}
2539 	nlookup_done_at(&nd, fp);
2540 	rel_mplock();
2541 	return (error);
2542 }
2543 
2544 /*
2545  * MPALMOSTSAFE
2546  */
2547 int
2548 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2549 {
2550 	struct thread *td = curthread;
2551 	struct proc *p = td->td_proc;
2552 	struct file *fp;
2553 	struct vnode *vp;
2554 	struct vattr vattr;
2555 	off_t new_offset;
2556 	int error;
2557 
2558 	fp = holdfp(p->p_fd, fd, -1);
2559 	if (fp == NULL)
2560 		return (EBADF);
2561 	if (fp->f_type != DTYPE_VNODE) {
2562 		error = ESPIPE;
2563 		goto done;
2564 	}
2565 	vp = (struct vnode *)fp->f_data;
2566 
2567 	switch (whence) {
2568 	case L_INCR:
2569 		spin_lock(&fp->f_spin);
2570 		new_offset = fp->f_offset + offset;
2571 		error = 0;
2572 		break;
2573 	case L_XTND:
2574 		get_mplock();
2575 		error = VOP_GETATTR(vp, &vattr);
2576 		rel_mplock();
2577 		spin_lock(&fp->f_spin);
2578 		new_offset = offset + vattr.va_size;
2579 		break;
2580 	case L_SET:
2581 		new_offset = offset;
2582 		error = 0;
2583 		spin_lock(&fp->f_spin);
2584 		break;
2585 	default:
2586 		new_offset = 0;
2587 		error = EINVAL;
2588 		spin_lock(&fp->f_spin);
2589 		break;
2590 	}
2591 
2592 	/*
2593 	 * Validate the seek position.  Negative offsets are not allowed
2594 	 * for regular files or directories.
2595 	 *
2596 	 * Normally we would also not want to allow negative offsets for
2597 	 * character and block-special devices.  However kvm addresses
2598 	 * on 64 bit architectures might appear to be negative and must
2599 	 * be allowed.
2600 	 */
2601 	if (error == 0) {
2602 		if (new_offset < 0 &&
2603 		    (vp->v_type == VREG || vp->v_type == VDIR)) {
2604 			error = EINVAL;
2605 		} else {
2606 			fp->f_offset = new_offset;
2607 		}
2608 	}
2609 	*res = fp->f_offset;
2610 	spin_unlock(&fp->f_spin);
2611 done:
2612 	fdrop(fp);
2613 	return (error);
2614 }
2615 
2616 /*
2617  * lseek_args(int fd, int pad, off_t offset, int whence)
2618  *
2619  * Reposition read/write file offset.
2620  *
2621  * MPSAFE
2622  */
2623 int
2624 sys_lseek(struct lseek_args *uap)
2625 {
2626 	int error;
2627 
2628 	error = kern_lseek(uap->fd, uap->offset, uap->whence,
2629 			   &uap->sysmsg_offset);
2630 
2631 	return (error);
2632 }
2633 
2634 /*
2635  * Check if current process can access given file.  amode is a bitmask of *_OK
2636  * access bits.  flags is a bitmask of AT_* flags.
2637  */
2638 int
2639 kern_access(struct nlookupdata *nd, int amode, int flags)
2640 {
2641 	struct vnode *vp;
2642 	int error, mode;
2643 
2644 	if (flags & ~AT_EACCESS)
2645 		return (EINVAL);
2646 	if ((error = nlookup(nd)) != 0)
2647 		return (error);
2648 retry:
2649 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2650 	if (error)
2651 		return (error);
2652 
2653 	/* Flags == 0 means only check for existence. */
2654 	if (amode) {
2655 		mode = 0;
2656 		if (amode & R_OK)
2657 			mode |= VREAD;
2658 		if (amode & W_OK)
2659 			mode |= VWRITE;
2660 		if (amode & X_OK)
2661 			mode |= VEXEC;
2662 		if ((mode & VWRITE) == 0 ||
2663 		    (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2664 			error = VOP_ACCESS_FLAGS(vp, mode, flags, nd->nl_cred);
2665 
2666 		/*
2667 		 * If the file handle is stale we have to re-resolve the
2668 		 * entry.  This is a hack at the moment.
2669 		 */
2670 		if (error == ESTALE) {
2671 			vput(vp);
2672 			cache_setunresolved(&nd->nl_nch);
2673 			error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2674 			if (error == 0) {
2675 				vp = NULL;
2676 				goto retry;
2677 			}
2678 			return(error);
2679 		}
2680 	}
2681 	vput(vp);
2682 	return (error);
2683 }
2684 
2685 /*
2686  * access_args(char *path, int flags)
2687  *
2688  * Check access permissions.
2689  *
2690  * MPALMOSTSAFE
2691  */
2692 int
2693 sys_access(struct access_args *uap)
2694 {
2695 	struct nlookupdata nd;
2696 	int error;
2697 
2698 	get_mplock();
2699 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2700 	if (error == 0)
2701 		error = kern_access(&nd, uap->flags, 0);
2702 	nlookup_done(&nd);
2703 	rel_mplock();
2704 	return (error);
2705 }
2706 
2707 
2708 /*
2709  * faccessat_args(int fd, char *path, int amode, int flags)
2710  *
2711  * Check access permissions.
2712  *
2713  * MPALMOSTSAFE
2714  */
2715 int
2716 sys_faccessat(struct faccessat_args *uap)
2717 {
2718 	struct nlookupdata nd;
2719 	struct file *fp;
2720 	int error;
2721 
2722 	get_mplock();
2723 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE,
2724 				NLC_FOLLOW);
2725 	if (error == 0)
2726 		error = kern_access(&nd, uap->amode, uap->flags);
2727 	nlookup_done_at(&nd, fp);
2728 	rel_mplock();
2729 	return (error);
2730 }
2731 
2732 
2733 /*
2734  * MPSAFE
2735  */
2736 int
2737 kern_stat(struct nlookupdata *nd, struct stat *st)
2738 {
2739 	int error;
2740 	struct vnode *vp;
2741 	thread_t td;
2742 
2743 	if ((error = nlookup(nd)) != 0)
2744 		return (error);
2745 again:
2746 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2747 		return (ENOENT);
2748 
2749 	td = curthread;
2750 	if ((error = vget(vp, LK_SHARED)) != 0)
2751 		return (error);
2752 	error = vn_stat(vp, st, nd->nl_cred);
2753 
2754 	/*
2755 	 * If the file handle is stale we have to re-resolve the entry.  This
2756 	 * is a hack at the moment.
2757 	 */
2758 	if (error == ESTALE) {
2759 		vput(vp);
2760 		cache_setunresolved(&nd->nl_nch);
2761 		error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2762 		if (error == 0)
2763 			goto again;
2764 	} else {
2765 		vput(vp);
2766 	}
2767 	return (error);
2768 }
2769 
2770 /*
2771  * stat_args(char *path, struct stat *ub)
2772  *
2773  * Get file status; this version follows links.
2774  *
2775  * MPSAFE
2776  */
2777 int
2778 sys_stat(struct stat_args *uap)
2779 {
2780 	CACHE_MPLOCK_DECLARE;
2781 	struct nlookupdata nd;
2782 	struct stat st;
2783 	int error;
2784 
2785 	CACHE_GETMPLOCK1();
2786 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2787 	if (error == 0) {
2788 		error = kern_stat(&nd, &st);
2789 		if (error == 0)
2790 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2791 	}
2792 	nlookup_done(&nd);
2793 	CACHE_RELMPLOCK();
2794 	return (error);
2795 }
2796 
2797 /*
2798  * lstat_args(char *path, struct stat *ub)
2799  *
2800  * Get file status; this version does not follow links.
2801  *
2802  * MPALMOSTSAFE
2803  */
2804 int
2805 sys_lstat(struct lstat_args *uap)
2806 {
2807 	CACHE_MPLOCK_DECLARE;
2808 	struct nlookupdata nd;
2809 	struct stat st;
2810 	int error;
2811 
2812 	CACHE_GETMPLOCK1();
2813 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2814 	if (error == 0) {
2815 		error = kern_stat(&nd, &st);
2816 		if (error == 0)
2817 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2818 	}
2819 	nlookup_done(&nd);
2820 	CACHE_RELMPLOCK();
2821 	return (error);
2822 }
2823 
2824 /*
2825  * fstatat_args(int fd, char *path, struct stat *sb, int flags)
2826  *
2827  * Get status of file pointed to by fd/path.
2828  *
2829  * MPALMOSTSAFE
2830  */
2831 int
2832 sys_fstatat(struct fstatat_args *uap)
2833 {
2834 	CACHE_MPLOCK_DECLARE;
2835 	struct nlookupdata nd;
2836 	struct stat st;
2837 	int error;
2838 	int flags;
2839 	struct file *fp;
2840 
2841 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
2842 		return (EINVAL);
2843 
2844 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
2845 
2846 	CACHE_GETMPLOCK1();
2847 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
2848 				UIO_USERSPACE, flags);
2849 	if (error == 0) {
2850 		error = kern_stat(&nd, &st);
2851 		if (error == 0)
2852 			error = copyout(&st, uap->sb, sizeof(*uap->sb));
2853 	}
2854 	nlookup_done_at(&nd, fp);
2855 	CACHE_RELMPLOCK();
2856 	return (error);
2857 }
2858 
2859 /*
2860  * pathconf_Args(char *path, int name)
2861  *
2862  * Get configurable pathname variables.
2863  *
2864  * MPALMOSTSAFE
2865  */
2866 int
2867 sys_pathconf(struct pathconf_args *uap)
2868 {
2869 	struct nlookupdata nd;
2870 	struct vnode *vp;
2871 	int error;
2872 
2873 	vp = NULL;
2874 	get_mplock();
2875 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2876 	if (error == 0)
2877 		error = nlookup(&nd);
2878 	if (error == 0)
2879 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2880 	nlookup_done(&nd);
2881 	if (error == 0) {
2882 		error = VOP_PATHCONF(vp, uap->name, &uap->sysmsg_reg);
2883 		vput(vp);
2884 	}
2885 	rel_mplock();
2886 	return (error);
2887 }
2888 
2889 /*
2890  * XXX: daver
2891  * kern_readlink isn't properly split yet.  There is a copyin burried
2892  * in VOP_READLINK().
2893  */
2894 int
2895 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2896 {
2897 	struct thread *td = curthread;
2898 	struct vnode *vp;
2899 	struct iovec aiov;
2900 	struct uio auio;
2901 	int error;
2902 
2903 	if ((error = nlookup(nd)) != 0)
2904 		return (error);
2905 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2906 	if (error)
2907 		return (error);
2908 	if (vp->v_type != VLNK) {
2909 		error = EINVAL;
2910 	} else {
2911 		aiov.iov_base = buf;
2912 		aiov.iov_len = count;
2913 		auio.uio_iov = &aiov;
2914 		auio.uio_iovcnt = 1;
2915 		auio.uio_offset = 0;
2916 		auio.uio_rw = UIO_READ;
2917 		auio.uio_segflg = UIO_USERSPACE;
2918 		auio.uio_td = td;
2919 		auio.uio_resid = count;
2920 		error = VOP_READLINK(vp, &auio, td->td_ucred);
2921 	}
2922 	vput(vp);
2923 	*res = count - auio.uio_resid;
2924 	return (error);
2925 }
2926 
2927 /*
2928  * readlink_args(char *path, char *buf, int count)
2929  *
2930  * Return target name of a symbolic link.
2931  *
2932  * MPALMOSTSAFE
2933  */
2934 int
2935 sys_readlink(struct readlink_args *uap)
2936 {
2937 	struct nlookupdata nd;
2938 	int error;
2939 
2940 	get_mplock();
2941 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2942 	if (error == 0) {
2943 		error = kern_readlink(&nd, uap->buf, uap->count,
2944 					&uap->sysmsg_result);
2945 	}
2946 	nlookup_done(&nd);
2947 	rel_mplock();
2948 	return (error);
2949 }
2950 
2951 /*
2952  * readlinkat_args(int fd, char *path, char *buf, size_t bufsize)
2953  *
2954  * Return target name of a symbolic link.  The path is relative to the
2955  * directory associated with fd.
2956  *
2957  * MPALMOSTSAFE
2958  */
2959 int
2960 sys_readlinkat(struct readlinkat_args *uap)
2961 {
2962 	struct nlookupdata nd;
2963 	struct file *fp;
2964 	int error;
2965 
2966 	get_mplock();
2967 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
2968 	if (error == 0) {
2969 		error = kern_readlink(&nd, uap->buf, uap->bufsize,
2970 					&uap->sysmsg_result);
2971 	}
2972 	nlookup_done_at(&nd, fp);
2973 	rel_mplock();
2974 	return (error);
2975 }
2976 
2977 static int
2978 setfflags(struct vnode *vp, int flags)
2979 {
2980 	struct thread *td = curthread;
2981 	int error;
2982 	struct vattr vattr;
2983 
2984 	/*
2985 	 * Prevent non-root users from setting flags on devices.  When
2986 	 * a device is reused, users can retain ownership of the device
2987 	 * if they are allowed to set flags and programs assume that
2988 	 * chown can't fail when done as root.
2989 	 */
2990 	if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2991 	    ((error = priv_check_cred(td->td_ucred, PRIV_VFS_CHFLAGS_DEV, 0)) != 0))
2992 		return (error);
2993 
2994 	/*
2995 	 * note: vget is required for any operation that might mod the vnode
2996 	 * so VINACTIVE is properly cleared.
2997 	 */
2998 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2999 		VATTR_NULL(&vattr);
3000 		vattr.va_flags = flags;
3001 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3002 		vput(vp);
3003 	}
3004 	return (error);
3005 }
3006 
3007 /*
3008  * chflags(char *path, int flags)
3009  *
3010  * Change flags of a file given a path name.
3011  *
3012  * MPALMOSTSAFE
3013  */
3014 int
3015 sys_chflags(struct chflags_args *uap)
3016 {
3017 	struct nlookupdata nd;
3018 	struct vnode *vp;
3019 	int error;
3020 
3021 	vp = NULL;
3022 	get_mplock();
3023 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3024 	if (error == 0)
3025 		error = nlookup(&nd);
3026 	if (error == 0)
3027 		error = ncp_writechk(&nd.nl_nch);
3028 	if (error == 0)
3029 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3030 	nlookup_done(&nd);
3031 	if (error == 0) {
3032 		error = setfflags(vp, uap->flags);
3033 		vrele(vp);
3034 	}
3035 	rel_mplock();
3036 	return (error);
3037 }
3038 
3039 /*
3040  * lchflags(char *path, int flags)
3041  *
3042  * Change flags of a file given a path name, but don't follow symlinks.
3043  *
3044  * MPALMOSTSAFE
3045  */
3046 int
3047 sys_lchflags(struct lchflags_args *uap)
3048 {
3049 	struct nlookupdata nd;
3050 	struct vnode *vp;
3051 	int error;
3052 
3053 	vp = NULL;
3054 	get_mplock();
3055 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3056 	if (error == 0)
3057 		error = nlookup(&nd);
3058 	if (error == 0)
3059 		error = ncp_writechk(&nd.nl_nch);
3060 	if (error == 0)
3061 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3062 	nlookup_done(&nd);
3063 	if (error == 0) {
3064 		error = setfflags(vp, uap->flags);
3065 		vrele(vp);
3066 	}
3067 	rel_mplock();
3068 	return (error);
3069 }
3070 
3071 /*
3072  * fchflags_args(int fd, int flags)
3073  *
3074  * Change flags of a file given a file descriptor.
3075  *
3076  * MPALMOSTSAFE
3077  */
3078 int
3079 sys_fchflags(struct fchflags_args *uap)
3080 {
3081 	struct thread *td = curthread;
3082 	struct proc *p = td->td_proc;
3083 	struct file *fp;
3084 	int error;
3085 
3086 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3087 		return (error);
3088 	get_mplock();
3089 	if (fp->f_nchandle.ncp)
3090 		error = ncp_writechk(&fp->f_nchandle);
3091 	if (error == 0)
3092 		error = setfflags((struct vnode *) fp->f_data, uap->flags);
3093 	rel_mplock();
3094 	fdrop(fp);
3095 	return (error);
3096 }
3097 
3098 static int
3099 setfmode(struct vnode *vp, int mode)
3100 {
3101 	struct thread *td = curthread;
3102 	int error;
3103 	struct vattr vattr;
3104 
3105 	/*
3106 	 * note: vget is required for any operation that might mod the vnode
3107 	 * so VINACTIVE is properly cleared.
3108 	 */
3109 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3110 		VATTR_NULL(&vattr);
3111 		vattr.va_mode = mode & ALLPERMS;
3112 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3113 		vput(vp);
3114 	}
3115 	return error;
3116 }
3117 
3118 int
3119 kern_chmod(struct nlookupdata *nd, int mode)
3120 {
3121 	struct vnode *vp;
3122 	int error;
3123 
3124 	if ((error = nlookup(nd)) != 0)
3125 		return (error);
3126 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3127 		return (error);
3128 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3129 		error = setfmode(vp, mode);
3130 	vrele(vp);
3131 	return (error);
3132 }
3133 
3134 /*
3135  * chmod_args(char *path, int mode)
3136  *
3137  * Change mode of a file given path name.
3138  *
3139  * MPALMOSTSAFE
3140  */
3141 int
3142 sys_chmod(struct chmod_args *uap)
3143 {
3144 	struct nlookupdata nd;
3145 	int error;
3146 
3147 	get_mplock();
3148 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3149 	if (error == 0)
3150 		error = kern_chmod(&nd, uap->mode);
3151 	nlookup_done(&nd);
3152 	rel_mplock();
3153 	return (error);
3154 }
3155 
3156 /*
3157  * lchmod_args(char *path, int mode)
3158  *
3159  * Change mode of a file given path name (don't follow links.)
3160  *
3161  * MPALMOSTSAFE
3162  */
3163 int
3164 sys_lchmod(struct lchmod_args *uap)
3165 {
3166 	struct nlookupdata nd;
3167 	int error;
3168 
3169 	get_mplock();
3170 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3171 	if (error == 0)
3172 		error = kern_chmod(&nd, uap->mode);
3173 	nlookup_done(&nd);
3174 	rel_mplock();
3175 	return (error);
3176 }
3177 
3178 /*
3179  * fchmod_args(int fd, int mode)
3180  *
3181  * Change mode of a file given a file descriptor.
3182  *
3183  * MPALMOSTSAFE
3184  */
3185 int
3186 sys_fchmod(struct fchmod_args *uap)
3187 {
3188 	struct thread *td = curthread;
3189 	struct proc *p = td->td_proc;
3190 	struct file *fp;
3191 	int error;
3192 
3193 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3194 		return (error);
3195 	get_mplock();
3196 	if (fp->f_nchandle.ncp)
3197 		error = ncp_writechk(&fp->f_nchandle);
3198 	if (error == 0)
3199 		error = setfmode((struct vnode *)fp->f_data, uap->mode);
3200 	rel_mplock();
3201 	fdrop(fp);
3202 	return (error);
3203 }
3204 
3205 /*
3206  * fchmodat_args(char *path, int mode)
3207  *
3208  * Change mode of a file pointed to by fd/path.
3209  *
3210  * MPALMOSTSAFE
3211  */
3212 int
3213 sys_fchmodat(struct fchmodat_args *uap)
3214 {
3215 	struct nlookupdata nd;
3216 	struct file *fp;
3217 	int error;
3218 	int flags;
3219 
3220 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3221 		return (EINVAL);
3222 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3223 
3224 	get_mplock();
3225 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3226 				UIO_USERSPACE, flags);
3227 	if (error == 0)
3228 		error = kern_chmod(&nd, uap->mode);
3229 	nlookup_done_at(&nd, fp);
3230 	rel_mplock();
3231 	return (error);
3232 }
3233 
3234 static int
3235 setfown(struct vnode *vp, uid_t uid, gid_t gid)
3236 {
3237 	struct thread *td = curthread;
3238 	int error;
3239 	struct vattr vattr;
3240 
3241 	/*
3242 	 * note: vget is required for any operation that might mod the vnode
3243 	 * so VINACTIVE is properly cleared.
3244 	 */
3245 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
3246 		VATTR_NULL(&vattr);
3247 		vattr.va_uid = uid;
3248 		vattr.va_gid = gid;
3249 		error = VOP_SETATTR(vp, &vattr, td->td_ucred);
3250 		vput(vp);
3251 	}
3252 	return error;
3253 }
3254 
3255 int
3256 kern_chown(struct nlookupdata *nd, int uid, int gid)
3257 {
3258 	struct vnode *vp;
3259 	int error;
3260 
3261 	if ((error = nlookup(nd)) != 0)
3262 		return (error);
3263 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3264 		return (error);
3265 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
3266 		error = setfown(vp, uid, gid);
3267 	vrele(vp);
3268 	return (error);
3269 }
3270 
3271 /*
3272  * chown(char *path, int uid, int gid)
3273  *
3274  * Set ownership given a path name.
3275  *
3276  * MPALMOSTSAFE
3277  */
3278 int
3279 sys_chown(struct chown_args *uap)
3280 {
3281 	struct nlookupdata nd;
3282 	int error;
3283 
3284 	get_mplock();
3285 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3286 	if (error == 0)
3287 		error = kern_chown(&nd, uap->uid, uap->gid);
3288 	nlookup_done(&nd);
3289 	rel_mplock();
3290 	return (error);
3291 }
3292 
3293 /*
3294  * lchown_args(char *path, int uid, int gid)
3295  *
3296  * Set ownership given a path name, do not cross symlinks.
3297  *
3298  * MPALMOSTSAFE
3299  */
3300 int
3301 sys_lchown(struct lchown_args *uap)
3302 {
3303 	struct nlookupdata nd;
3304 	int error;
3305 
3306 	get_mplock();
3307 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3308 	if (error == 0)
3309 		error = kern_chown(&nd, uap->uid, uap->gid);
3310 	nlookup_done(&nd);
3311 	rel_mplock();
3312 	return (error);
3313 }
3314 
3315 /*
3316  * fchown_args(int fd, int uid, int gid)
3317  *
3318  * Set ownership given a file descriptor.
3319  *
3320  * MPALMOSTSAFE
3321  */
3322 int
3323 sys_fchown(struct fchown_args *uap)
3324 {
3325 	struct thread *td = curthread;
3326 	struct proc *p = td->td_proc;
3327 	struct file *fp;
3328 	int error;
3329 
3330 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3331 		return (error);
3332 	get_mplock();
3333 	if (fp->f_nchandle.ncp)
3334 		error = ncp_writechk(&fp->f_nchandle);
3335 	if (error == 0)
3336 		error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
3337 	rel_mplock();
3338 	fdrop(fp);
3339 	return (error);
3340 }
3341 
3342 /*
3343  * fchownat(int fd, char *path, int uid, int gid, int flags)
3344  *
3345  * Set ownership of file pointed to by fd/path.
3346  *
3347  * MPALMOSTSAFE
3348  */
3349 int
3350 sys_fchownat(struct fchownat_args *uap)
3351 {
3352 	struct nlookupdata nd;
3353 	struct file *fp;
3354 	int error;
3355 	int flags;
3356 
3357 	if (uap->flags & ~AT_SYMLINK_NOFOLLOW)
3358 		return (EINVAL);
3359 	flags = (uap->flags & AT_SYMLINK_NOFOLLOW) ? 0 : NLC_FOLLOW;
3360 
3361 	get_mplock();
3362 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path,
3363 				UIO_USERSPACE, flags);
3364 	if (error == 0)
3365 		error = kern_chown(&nd, uap->uid, uap->gid);
3366 	nlookup_done_at(&nd, fp);
3367 	rel_mplock();
3368 	return (error);
3369 }
3370 
3371 
3372 static int
3373 getutimes(const struct timeval *tvp, struct timespec *tsp)
3374 {
3375 	struct timeval tv[2];
3376 
3377 	if (tvp == NULL) {
3378 		microtime(&tv[0]);
3379 		TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
3380 		tsp[1] = tsp[0];
3381 	} else {
3382 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
3383 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
3384 	}
3385 	return 0;
3386 }
3387 
3388 static int
3389 setutimes(struct vnode *vp, struct vattr *vattr,
3390 	  const struct timespec *ts, int nullflag)
3391 {
3392 	struct thread *td = curthread;
3393 	int error;
3394 
3395 	VATTR_NULL(vattr);
3396 	vattr->va_atime = ts[0];
3397 	vattr->va_mtime = ts[1];
3398 	if (nullflag)
3399 		vattr->va_vaflags |= VA_UTIMES_NULL;
3400 	error = VOP_SETATTR(vp, vattr, td->td_ucred);
3401 
3402 	return error;
3403 }
3404 
3405 int
3406 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
3407 {
3408 	struct timespec ts[2];
3409 	struct vnode *vp;
3410 	struct vattr vattr;
3411 	int error;
3412 
3413 	if ((error = getutimes(tptr, ts)) != 0)
3414 		return (error);
3415 
3416 	/*
3417 	 * NOTE: utimes() succeeds for the owner even if the file
3418 	 * is not user-writable.
3419 	 */
3420 	nd->nl_flags |= NLC_OWN | NLC_WRITE;
3421 
3422 	if ((error = nlookup(nd)) != 0)
3423 		return (error);
3424 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3425 		return (error);
3426 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3427 		return (error);
3428 
3429 	/*
3430 	 * note: vget is required for any operation that might mod the vnode
3431 	 * so VINACTIVE is properly cleared.
3432 	 */
3433 	if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3434 		error = vget(vp, LK_EXCLUSIVE);
3435 		if (error == 0) {
3436 			error = setutimes(vp, &vattr, ts, (tptr == NULL));
3437 			vput(vp);
3438 		}
3439 	}
3440 	vrele(vp);
3441 	return (error);
3442 }
3443 
3444 /*
3445  * utimes_args(char *path, struct timeval *tptr)
3446  *
3447  * Set the access and modification times of a file.
3448  *
3449  * MPALMOSTSAFE
3450  */
3451 int
3452 sys_utimes(struct utimes_args *uap)
3453 {
3454 	struct timeval tv[2];
3455 	struct nlookupdata nd;
3456 	int error;
3457 
3458 	if (uap->tptr) {
3459  		error = copyin(uap->tptr, tv, sizeof(tv));
3460 		if (error)
3461 			return (error);
3462 	}
3463 	get_mplock();
3464 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3465 	if (error == 0)
3466 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3467 	nlookup_done(&nd);
3468 	rel_mplock();
3469 	return (error);
3470 }
3471 
3472 /*
3473  * lutimes_args(char *path, struct timeval *tptr)
3474  *
3475  * Set the access and modification times of a file.
3476  *
3477  * MPALMOSTSAFE
3478  */
3479 int
3480 sys_lutimes(struct lutimes_args *uap)
3481 {
3482 	struct timeval tv[2];
3483 	struct nlookupdata nd;
3484 	int error;
3485 
3486 	if (uap->tptr) {
3487 		error = copyin(uap->tptr, tv, sizeof(tv));
3488 		if (error)
3489 			return (error);
3490 	}
3491 	get_mplock();
3492 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3493 	if (error == 0)
3494 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
3495 	nlookup_done(&nd);
3496 	rel_mplock();
3497 	return (error);
3498 }
3499 
3500 /*
3501  * Set utimes on a file descriptor.  The creds used to open the
3502  * file are used to determine whether the operation is allowed
3503  * or not.
3504  */
3505 int
3506 kern_futimes(int fd, struct timeval *tptr)
3507 {
3508 	struct thread *td = curthread;
3509 	struct proc *p = td->td_proc;
3510 	struct timespec ts[2];
3511 	struct file *fp;
3512 	struct vnode *vp;
3513 	struct vattr vattr;
3514 	int error;
3515 
3516 	error = getutimes(tptr, ts);
3517 	if (error)
3518 		return (error);
3519 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3520 		return (error);
3521 	if (fp->f_nchandle.ncp)
3522 		error = ncp_writechk(&fp->f_nchandle);
3523 	if (error == 0) {
3524 		vp = fp->f_data;
3525 		error = vget(vp, LK_EXCLUSIVE);
3526 		if (error == 0) {
3527 			error = VOP_GETATTR(vp, &vattr);
3528 			if (error == 0) {
3529 				error = naccess_va(&vattr, NLC_OWN | NLC_WRITE,
3530 						   fp->f_cred);
3531 			}
3532 			if (error == 0) {
3533 				error = setutimes(vp, &vattr, ts,
3534 						  (tptr == NULL));
3535 			}
3536 			vput(vp);
3537 		}
3538 	}
3539 	fdrop(fp);
3540 	return (error);
3541 }
3542 
3543 /*
3544  * futimes_args(int fd, struct timeval *tptr)
3545  *
3546  * Set the access and modification times of a file.
3547  *
3548  * MPALMOSTSAFE
3549  */
3550 int
3551 sys_futimes(struct futimes_args *uap)
3552 {
3553 	struct timeval tv[2];
3554 	int error;
3555 
3556 	if (uap->tptr) {
3557 		error = copyin(uap->tptr, tv, sizeof(tv));
3558 		if (error)
3559 			return (error);
3560 	}
3561 	get_mplock();
3562 	error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
3563 	rel_mplock();
3564 
3565 	return (error);
3566 }
3567 
3568 int
3569 kern_truncate(struct nlookupdata *nd, off_t length)
3570 {
3571 	struct vnode *vp;
3572 	struct vattr vattr;
3573 	int error;
3574 
3575 	if (length < 0)
3576 		return(EINVAL);
3577 	nd->nl_flags |= NLC_WRITE | NLC_TRUNCATE;
3578 	if ((error = nlookup(nd)) != 0)
3579 		return (error);
3580 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3581 		return (error);
3582 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3583 		return (error);
3584 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
3585 		vrele(vp);
3586 		return (error);
3587 	}
3588 	if (vp->v_type == VDIR) {
3589 		error = EISDIR;
3590 	} else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0) {
3591 		VATTR_NULL(&vattr);
3592 		vattr.va_size = length;
3593 		error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
3594 	}
3595 	vput(vp);
3596 	return (error);
3597 }
3598 
3599 /*
3600  * truncate(char *path, int pad, off_t length)
3601  *
3602  * Truncate a file given its path name.
3603  *
3604  * MPALMOSTSAFE
3605  */
3606 int
3607 sys_truncate(struct truncate_args *uap)
3608 {
3609 	struct nlookupdata nd;
3610 	int error;
3611 
3612 	get_mplock();
3613 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3614 	if (error == 0)
3615 		error = kern_truncate(&nd, uap->length);
3616 	nlookup_done(&nd);
3617 	rel_mplock();
3618 	return error;
3619 }
3620 
3621 int
3622 kern_ftruncate(int fd, off_t length)
3623 {
3624 	struct thread *td = curthread;
3625 	struct proc *p = td->td_proc;
3626 	struct vattr vattr;
3627 	struct vnode *vp;
3628 	struct file *fp;
3629 	int error;
3630 
3631 	if (length < 0)
3632 		return(EINVAL);
3633 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3634 		return (error);
3635 	if (fp->f_nchandle.ncp) {
3636 		error = ncp_writechk(&fp->f_nchandle);
3637 		if (error)
3638 			goto done;
3639 	}
3640 	if ((fp->f_flag & FWRITE) == 0) {
3641 		error = EINVAL;
3642 		goto done;
3643 	}
3644 	if (fp->f_flag & FAPPENDONLY) {	/* inode was set s/uapnd */
3645 		error = EINVAL;
3646 		goto done;
3647 	}
3648 	vp = (struct vnode *)fp->f_data;
3649 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3650 	if (vp->v_type == VDIR) {
3651 		error = EISDIR;
3652 	} else if ((error = vn_writechk(vp, NULL)) == 0) {
3653 		VATTR_NULL(&vattr);
3654 		vattr.va_size = length;
3655 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
3656 	}
3657 	vn_unlock(vp);
3658 done:
3659 	fdrop(fp);
3660 	return (error);
3661 }
3662 
3663 /*
3664  * ftruncate_args(int fd, int pad, off_t length)
3665  *
3666  * Truncate a file given a file descriptor.
3667  *
3668  * MPALMOSTSAFE
3669  */
3670 int
3671 sys_ftruncate(struct ftruncate_args *uap)
3672 {
3673 	int error;
3674 
3675 	get_mplock();
3676 	error = kern_ftruncate(uap->fd, uap->length);
3677 	rel_mplock();
3678 
3679 	return (error);
3680 }
3681 
3682 /*
3683  * fsync(int fd)
3684  *
3685  * Sync an open file.
3686  *
3687  * MPALMOSTSAFE
3688  */
3689 int
3690 sys_fsync(struct fsync_args *uap)
3691 {
3692 	struct thread *td = curthread;
3693 	struct proc *p = td->td_proc;
3694 	struct vnode *vp;
3695 	struct file *fp;
3696 	vm_object_t obj;
3697 	int error;
3698 
3699 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3700 		return (error);
3701 	get_mplock();
3702 	vp = (struct vnode *)fp->f_data;
3703 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3704 	if ((obj = vp->v_object) != NULL)
3705 		vm_object_page_clean(obj, 0, 0, 0);
3706 	error = VOP_FSYNC(vp, MNT_WAIT, VOP_FSYNC_SYSCALL);
3707 	if (error == 0 && vp->v_mount)
3708 		error = buf_fsync(vp);
3709 	vn_unlock(vp);
3710 	rel_mplock();
3711 	fdrop(fp);
3712 
3713 	return (error);
3714 }
3715 
3716 int
3717 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
3718 {
3719 	struct nchandle fnchd;
3720 	struct nchandle tnchd;
3721 	struct namecache *ncp;
3722 	struct vnode *fdvp;
3723 	struct vnode *tdvp;
3724 	struct mount *mp;
3725 	int error;
3726 
3727 	bwillinode(1);
3728 	fromnd->nl_flags |= NLC_REFDVP | NLC_RENAME_SRC;
3729 	if ((error = nlookup(fromnd)) != 0)
3730 		return (error);
3731 	if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
3732 		return (ENOENT);
3733 	fnchd.mount = fromnd->nl_nch.mount;
3734 	cache_hold(&fnchd);
3735 
3736 	/*
3737 	 * unlock the source nch so we can lookup the target nch without
3738 	 * deadlocking.  The target may or may not exist so we do not check
3739 	 * for a target vp like kern_mkdir() and other creation functions do.
3740 	 *
3741 	 * The source and target directories are ref'd and rechecked after
3742 	 * everything is relocked to determine if the source or target file
3743 	 * has been renamed.
3744 	 */
3745 	KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
3746 	fromnd->nl_flags &= ~NLC_NCPISLOCKED;
3747 	cache_unlock(&fromnd->nl_nch);
3748 
3749 	tond->nl_flags |= NLC_RENAME_DST | NLC_REFDVP;
3750 	if ((error = nlookup(tond)) != 0) {
3751 		cache_drop(&fnchd);
3752 		return (error);
3753 	}
3754 	if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
3755 		cache_drop(&fnchd);
3756 		return (ENOENT);
3757 	}
3758 	tnchd.mount = tond->nl_nch.mount;
3759 	cache_hold(&tnchd);
3760 
3761 	/*
3762 	 * If the source and target are the same there is nothing to do
3763 	 */
3764 	if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
3765 		cache_drop(&fnchd);
3766 		cache_drop(&tnchd);
3767 		return (0);
3768 	}
3769 
3770 	/*
3771 	 * Mount points cannot be renamed or overwritten
3772 	 */
3773 	if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
3774 	    NCF_ISMOUNTPT
3775 	) {
3776 		cache_drop(&fnchd);
3777 		cache_drop(&tnchd);
3778 		return (EINVAL);
3779 	}
3780 
3781 	/*
3782 	 * Relock the source ncp.  cache_relock() will deal with any
3783 	 * deadlocks against the already-locked tond and will also
3784 	 * make sure both are resolved.
3785 	 *
3786 	 * NOTE AFTER RELOCKING: The source or target ncp may have become
3787 	 * invalid while they were unlocked, nc_vp and nc_mount could
3788 	 * be NULL.
3789 	 */
3790 	cache_relock(&fromnd->nl_nch, fromnd->nl_cred,
3791 		     &tond->nl_nch, tond->nl_cred);
3792 	fromnd->nl_flags |= NLC_NCPISLOCKED;
3793 
3794 	/*
3795 	 * make sure the parent directories linkages are the same
3796 	 */
3797 	if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
3798 	    tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
3799 		cache_drop(&fnchd);
3800 		cache_drop(&tnchd);
3801 		return (ENOENT);
3802 	}
3803 
3804 	/*
3805 	 * Both the source and target must be within the same filesystem and
3806 	 * in the same filesystem as their parent directories within the
3807 	 * namecache topology.
3808 	 *
3809 	 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
3810 	 */
3811 	mp = fnchd.mount;
3812 	if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
3813 	    mp != tond->nl_nch.mount) {
3814 		cache_drop(&fnchd);
3815 		cache_drop(&tnchd);
3816 		return (EXDEV);
3817 	}
3818 
3819 	/*
3820 	 * Make sure the mount point is writable
3821 	 */
3822 	if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
3823 		cache_drop(&fnchd);
3824 		cache_drop(&tnchd);
3825 		return (error);
3826 	}
3827 
3828 	/*
3829 	 * If the target exists and either the source or target is a directory,
3830 	 * then both must be directories.
3831 	 *
3832 	 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
3833 	 * have become NULL.
3834 	 */
3835 	if (tond->nl_nch.ncp->nc_vp) {
3836 		if (fromnd->nl_nch.ncp->nc_vp == NULL) {
3837 			error = ENOENT;
3838 		} else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
3839 			if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
3840 				error = ENOTDIR;
3841 		} else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
3842 			error = EISDIR;
3843 		}
3844 	}
3845 
3846 	/*
3847 	 * You cannot rename a source into itself or a subdirectory of itself.
3848 	 * We check this by travsersing the target directory upwards looking
3849 	 * for a match against the source.
3850 	 *
3851 	 * XXX MPSAFE
3852 	 */
3853 	if (error == 0) {
3854 		for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
3855 			if (fromnd->nl_nch.ncp == ncp) {
3856 				error = EINVAL;
3857 				break;
3858 			}
3859 		}
3860 	}
3861 
3862 	cache_drop(&fnchd);
3863 	cache_drop(&tnchd);
3864 
3865 	/*
3866 	 * Even though the namespaces are different, they may still represent
3867 	 * hardlinks to the same file.  The filesystem might have a hard time
3868 	 * with this so we issue a NREMOVE of the source instead of a NRENAME
3869 	 * when we detect the situation.
3870 	 */
3871 	if (error == 0) {
3872 		fdvp = fromnd->nl_dvp;
3873 		tdvp = tond->nl_dvp;
3874 		if (fdvp == NULL || tdvp == NULL) {
3875 			error = EPERM;
3876 		} else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
3877 			error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
3878 					    fromnd->nl_cred);
3879 		} else {
3880 			error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
3881 					    fdvp, tdvp, tond->nl_cred);
3882 		}
3883 	}
3884 	return (error);
3885 }
3886 
3887 /*
3888  * rename_args(char *from, char *to)
3889  *
3890  * Rename files.  Source and destination must either both be directories,
3891  * or both not be directories.  If target is a directory, it must be empty.
3892  *
3893  * MPALMOSTSAFE
3894  */
3895 int
3896 sys_rename(struct rename_args *uap)
3897 {
3898 	struct nlookupdata fromnd, tond;
3899 	int error;
3900 
3901 	get_mplock();
3902 	error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3903 	if (error == 0) {
3904 		error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3905 		if (error == 0)
3906 			error = kern_rename(&fromnd, &tond);
3907 		nlookup_done(&tond);
3908 	}
3909 	nlookup_done(&fromnd);
3910 	rel_mplock();
3911 	return (error);
3912 }
3913 
3914 /*
3915  * renameat_args(int oldfd, char *old, int newfd, char *new)
3916  *
3917  * Rename files using paths relative to the directories associated with
3918  * oldfd and newfd.  Source and destination must either both be directories,
3919  * or both not be directories.  If target is a directory, it must be empty.
3920  *
3921  * MPALMOSTSAFE
3922  */
3923 int
3924 sys_renameat(struct renameat_args *uap)
3925 {
3926 	struct nlookupdata oldnd, newnd;
3927 	struct file *oldfp, *newfp;
3928 	int error;
3929 
3930 	get_mplock();
3931 	error = nlookup_init_at(&oldnd, &oldfp, uap->oldfd, uap->old,
3932 	    UIO_USERSPACE, 0);
3933 	if (error == 0) {
3934 		error = nlookup_init_at(&newnd, &newfp, uap->newfd, uap->new,
3935 		    UIO_USERSPACE, 0);
3936 		if (error == 0)
3937 			error = kern_rename(&oldnd, &newnd);
3938 		nlookup_done_at(&newnd, newfp);
3939 	}
3940 	nlookup_done_at(&oldnd, oldfp);
3941 	rel_mplock();
3942 	return (error);
3943 }
3944 
3945 int
3946 kern_mkdir(struct nlookupdata *nd, int mode)
3947 {
3948 	struct thread *td = curthread;
3949 	struct proc *p = td->td_proc;
3950 	struct vnode *vp;
3951 	struct vattr vattr;
3952 	int error;
3953 
3954 	bwillinode(1);
3955 	nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
3956 	if ((error = nlookup(nd)) != 0)
3957 		return (error);
3958 
3959 	if (nd->nl_nch.ncp->nc_vp)
3960 		return (EEXIST);
3961 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3962 		return (error);
3963 	VATTR_NULL(&vattr);
3964 	vattr.va_type = VDIR;
3965 	vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3966 
3967 	vp = NULL;
3968 	error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, td->td_ucred, &vattr);
3969 	if (error == 0)
3970 		vput(vp);
3971 	return (error);
3972 }
3973 
3974 /*
3975  * mkdir_args(char *path, int mode)
3976  *
3977  * Make a directory file.
3978  *
3979  * MPALMOSTSAFE
3980  */
3981 int
3982 sys_mkdir(struct mkdir_args *uap)
3983 {
3984 	struct nlookupdata nd;
3985 	int error;
3986 
3987 	get_mplock();
3988 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3989 	if (error == 0)
3990 		error = kern_mkdir(&nd, uap->mode);
3991 	nlookup_done(&nd);
3992 	rel_mplock();
3993 	return (error);
3994 }
3995 
3996 /*
3997  * mkdirat_args(int fd, char *path, mode_t mode)
3998  *
3999  * Make a directory file.  The path is relative to the directory associated
4000  * with fd.
4001  *
4002  * MPALMOSTSAFE
4003  */
4004 int
4005 sys_mkdirat(struct mkdirat_args *uap)
4006 {
4007 	struct nlookupdata nd;
4008 	struct file *fp;
4009 	int error;
4010 
4011 	get_mplock();
4012 	error = nlookup_init_at(&nd, &fp, uap->fd, uap->path, UIO_USERSPACE, 0);
4013 	if (error == 0)
4014 		error = kern_mkdir(&nd, uap->mode);
4015 	nlookup_done_at(&nd, fp);
4016 	rel_mplock();
4017 	return (error);
4018 }
4019 
4020 int
4021 kern_rmdir(struct nlookupdata *nd)
4022 {
4023 	int error;
4024 
4025 	bwillinode(1);
4026 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
4027 	if ((error = nlookup(nd)) != 0)
4028 		return (error);
4029 
4030 	/*
4031 	 * Do not allow directories representing mount points to be
4032 	 * deleted, even if empty.  Check write perms on mount point
4033 	 * in case the vnode is aliased (aka nullfs).
4034 	 */
4035 	if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
4036 		return (EINVAL);
4037 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
4038 		return (error);
4039 	error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
4040 	return (error);
4041 }
4042 
4043 /*
4044  * rmdir_args(char *path)
4045  *
4046  * Remove a directory file.
4047  *
4048  * MPALMOSTSAFE
4049  */
4050 int
4051 sys_rmdir(struct rmdir_args *uap)
4052 {
4053 	struct nlookupdata nd;
4054 	int error;
4055 
4056 	get_mplock();
4057 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
4058 	if (error == 0)
4059 		error = kern_rmdir(&nd);
4060 	nlookup_done(&nd);
4061 	rel_mplock();
4062 	return (error);
4063 }
4064 
4065 int
4066 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
4067     enum uio_seg direction)
4068 {
4069 	struct thread *td = curthread;
4070 	struct proc *p = td->td_proc;
4071 	struct vnode *vp;
4072 	struct file *fp;
4073 	struct uio auio;
4074 	struct iovec aiov;
4075 	off_t loff;
4076 	int error, eofflag;
4077 
4078 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
4079 		return (error);
4080 	if ((fp->f_flag & FREAD) == 0) {
4081 		error = EBADF;
4082 		goto done;
4083 	}
4084 	vp = (struct vnode *)fp->f_data;
4085 unionread:
4086 	if (vp->v_type != VDIR) {
4087 		error = EINVAL;
4088 		goto done;
4089 	}
4090 	aiov.iov_base = buf;
4091 	aiov.iov_len = count;
4092 	auio.uio_iov = &aiov;
4093 	auio.uio_iovcnt = 1;
4094 	auio.uio_rw = UIO_READ;
4095 	auio.uio_segflg = direction;
4096 	auio.uio_td = td;
4097 	auio.uio_resid = count;
4098 	loff = auio.uio_offset = fp->f_offset;
4099 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
4100 	fp->f_offset = auio.uio_offset;
4101 	if (error)
4102 		goto done;
4103 	if (count == auio.uio_resid) {
4104 		if (union_dircheckp) {
4105 			error = union_dircheckp(td, &vp, fp);
4106 			if (error == -1)
4107 				goto unionread;
4108 			if (error)
4109 				goto done;
4110 		}
4111 #if 0
4112 		if ((vp->v_flag & VROOT) &&
4113 		    (vp->v_mount->mnt_flag & MNT_UNION)) {
4114 			struct vnode *tvp = vp;
4115 			vp = vp->v_mount->mnt_vnodecovered;
4116 			vref(vp);
4117 			fp->f_data = vp;
4118 			fp->f_offset = 0;
4119 			vrele(tvp);
4120 			goto unionread;
4121 		}
4122 #endif
4123 	}
4124 
4125 	/*
4126 	 * WARNING!  *basep may not be wide enough to accomodate the
4127 	 * seek offset.   XXX should we hack this to return the upper 32 bits
4128 	 * for offsets greater then 4G?
4129 	 */
4130 	if (basep) {
4131 		*basep = (long)loff;
4132 	}
4133 	*res = count - auio.uio_resid;
4134 done:
4135 	fdrop(fp);
4136 	return (error);
4137 }
4138 
4139 /*
4140  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
4141  *
4142  * Read a block of directory entries in a file system independent format.
4143  *
4144  * MPALMOSTSAFE
4145  */
4146 int
4147 sys_getdirentries(struct getdirentries_args *uap)
4148 {
4149 	long base;
4150 	int error;
4151 
4152 	get_mplock();
4153 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
4154 				   &uap->sysmsg_result, UIO_USERSPACE);
4155 	rel_mplock();
4156 
4157 	if (error == 0 && uap->basep)
4158 		error = copyout(&base, uap->basep, sizeof(*uap->basep));
4159 	return (error);
4160 }
4161 
4162 /*
4163  * getdents_args(int fd, char *buf, size_t count)
4164  *
4165  * MPALMOSTSAFE
4166  */
4167 int
4168 sys_getdents(struct getdents_args *uap)
4169 {
4170 	int error;
4171 
4172 	get_mplock();
4173 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
4174 				   &uap->sysmsg_result, UIO_USERSPACE);
4175 	rel_mplock();
4176 
4177 	return (error);
4178 }
4179 
4180 /*
4181  * Set the mode mask for creation of filesystem nodes.
4182  *
4183  * umask(int newmask)
4184  *
4185  * MPSAFE
4186  */
4187 int
4188 sys_umask(struct umask_args *uap)
4189 {
4190 	struct thread *td = curthread;
4191 	struct proc *p = td->td_proc;
4192 	struct filedesc *fdp;
4193 
4194 	fdp = p->p_fd;
4195 	uap->sysmsg_result = fdp->fd_cmask;
4196 	fdp->fd_cmask = uap->newmask & ALLPERMS;
4197 	return (0);
4198 }
4199 
4200 /*
4201  * revoke(char *path)
4202  *
4203  * Void all references to file by ripping underlying filesystem
4204  * away from vnode.
4205  *
4206  * MPALMOSTSAFE
4207  */
4208 int
4209 sys_revoke(struct revoke_args *uap)
4210 {
4211 	struct nlookupdata nd;
4212 	struct vattr vattr;
4213 	struct vnode *vp;
4214 	struct ucred *cred;
4215 	int error;
4216 
4217 	vp = NULL;
4218 	get_mplock();
4219 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4220 	if (error == 0)
4221 		error = nlookup(&nd);
4222 	if (error == 0)
4223 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4224 	cred = crhold(nd.nl_cred);
4225 	nlookup_done(&nd);
4226 	if (error == 0) {
4227 		if (error == 0)
4228 			error = VOP_GETATTR(vp, &vattr);
4229 		if (error == 0 && cred->cr_uid != vattr.va_uid)
4230 			error = priv_check_cred(cred, PRIV_VFS_REVOKE, 0);
4231 		if (error == 0 && (vp->v_type == VCHR || vp->v_type == VBLK)) {
4232 			if (vcount(vp) > 0)
4233 				error = vrevoke(vp, cred);
4234 		} else if (error == 0) {
4235 			error = vrevoke(vp, cred);
4236 		}
4237 		vrele(vp);
4238 	}
4239 	if (cred)
4240 		crfree(cred);
4241 	rel_mplock();
4242 	return (error);
4243 }
4244 
4245 /*
4246  * getfh_args(char *fname, fhandle_t *fhp)
4247  *
4248  * Get (NFS) file handle
4249  *
4250  * NOTE: We use the fsid of the covering mount, even if it is a nullfs
4251  * mount.  This allows nullfs mounts to be explicitly exported.
4252  *
4253  * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe.
4254  *
4255  * 	    nullfs mounts of subdirectories are not safe.  That is, it will
4256  *	    work, but you do not really have protection against access to
4257  *	    the related parent directories.
4258  *
4259  * MPALMOSTSAFE
4260  */
4261 int
4262 sys_getfh(struct getfh_args *uap)
4263 {
4264 	struct thread *td = curthread;
4265 	struct nlookupdata nd;
4266 	fhandle_t fh;
4267 	struct vnode *vp;
4268 	struct mount *mp;
4269 	int error;
4270 
4271 	/*
4272 	 * Must be super user
4273 	 */
4274 	if ((error = priv_check(td, PRIV_ROOT)) != 0)
4275 		return (error);
4276 
4277 	vp = NULL;
4278 	get_mplock();
4279 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
4280 	if (error == 0)
4281 		error = nlookup(&nd);
4282 	if (error == 0)
4283 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4284 	mp = nd.nl_nch.mount;
4285 	nlookup_done(&nd);
4286 	if (error == 0) {
4287 		bzero(&fh, sizeof(fh));
4288 		fh.fh_fsid = mp->mnt_stat.f_fsid;
4289 		error = VFS_VPTOFH(vp, &fh.fh_fid);
4290 		vput(vp);
4291 		if (error == 0)
4292 			error = copyout(&fh, uap->fhp, sizeof(fh));
4293 	}
4294 	rel_mplock();
4295 	return (error);
4296 }
4297 
4298 /*
4299  * fhopen_args(const struct fhandle *u_fhp, int flags)
4300  *
4301  * syscall for the rpc.lockd to use to translate a NFS file handle into
4302  * an open descriptor.
4303  *
4304  * warning: do not remove the priv_check() call or this becomes one giant
4305  * security hole.
4306  *
4307  * MPALMOSTSAFE
4308  */
4309 int
4310 sys_fhopen(struct fhopen_args *uap)
4311 {
4312 	struct thread *td = curthread;
4313 	struct filedesc *fdp = td->td_proc->p_fd;
4314 	struct mount *mp;
4315 	struct vnode *vp;
4316 	struct fhandle fhp;
4317 	struct vattr vat;
4318 	struct vattr *vap = &vat;
4319 	struct flock lf;
4320 	int fmode, mode, error, type;
4321 	struct file *nfp;
4322 	struct file *fp;
4323 	int indx;
4324 
4325 	/*
4326 	 * Must be super user
4327 	 */
4328 	error = priv_check(td, PRIV_ROOT);
4329 	if (error)
4330 		return (error);
4331 
4332 	fmode = FFLAGS(uap->flags);
4333 
4334 	/*
4335 	 * Why not allow a non-read/write open for our lockd?
4336 	 */
4337 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
4338 		return (EINVAL);
4339 	error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
4340 	if (error)
4341 		return(error);
4342 
4343 	/*
4344 	 * Find the mount point
4345 	 */
4346 	get_mplock();
4347 	mp = vfs_getvfs(&fhp.fh_fsid);
4348 	if (mp == NULL) {
4349 		error = ESTALE;
4350 		goto  done;
4351 	}
4352 	/* now give me my vnode, it gets returned to me locked */
4353 	error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp);
4354 	if (error)
4355 		goto done;
4356  	/*
4357 	 * from now on we have to make sure not
4358 	 * to forget about the vnode
4359 	 * any error that causes an abort must vput(vp)
4360 	 * just set error = err and 'goto bad;'.
4361 	 */
4362 
4363 	/*
4364 	 * from vn_open
4365 	 */
4366 	if (vp->v_type == VLNK) {
4367 		error = EMLINK;
4368 		goto bad;
4369 	}
4370 	if (vp->v_type == VSOCK) {
4371 		error = EOPNOTSUPP;
4372 		goto bad;
4373 	}
4374 	mode = 0;
4375 	if (fmode & (FWRITE | O_TRUNC)) {
4376 		if (vp->v_type == VDIR) {
4377 			error = EISDIR;
4378 			goto bad;
4379 		}
4380 		error = vn_writechk(vp, NULL);
4381 		if (error)
4382 			goto bad;
4383 		mode |= VWRITE;
4384 	}
4385 	if (fmode & FREAD)
4386 		mode |= VREAD;
4387 	if (mode) {
4388 		error = VOP_ACCESS(vp, mode, td->td_ucred);
4389 		if (error)
4390 			goto bad;
4391 	}
4392 	if (fmode & O_TRUNC) {
4393 		vn_unlock(vp);				/* XXX */
4394 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
4395 		VATTR_NULL(vap);
4396 		vap->va_size = 0;
4397 		error = VOP_SETATTR(vp, vap, td->td_ucred);
4398 		if (error)
4399 			goto bad;
4400 	}
4401 
4402 	/*
4403 	 * VOP_OPEN needs the file pointer so it can potentially override
4404 	 * it.
4405 	 *
4406 	 * WARNING! no f_nchandle will be associated when fhopen()ing a
4407 	 * directory.  XXX
4408 	 */
4409 	if ((error = falloc(td->td_lwp, &nfp, &indx)) != 0)
4410 		goto bad;
4411 	fp = nfp;
4412 
4413 	error = VOP_OPEN(vp, fmode, td->td_ucred, fp);
4414 	if (error) {
4415 		/*
4416 		 * setting f_ops this way prevents VOP_CLOSE from being
4417 		 * called or fdrop() releasing the vp from v_data.   Since
4418 		 * the VOP_OPEN failed we don't want to VOP_CLOSE.
4419 		 */
4420 		fp->f_ops = &badfileops;
4421 		fp->f_data = NULL;
4422 		goto bad_drop;
4423 	}
4424 
4425 	/*
4426 	 * The fp is given its own reference, we still have our ref and lock.
4427 	 *
4428 	 * Assert that all regular files must be created with a VM object.
4429 	 */
4430 	if (vp->v_type == VREG && vp->v_object == NULL) {
4431 		kprintf("fhopen: regular file did not have VM object: %p\n", vp);
4432 		goto bad_drop;
4433 	}
4434 
4435 	/*
4436 	 * The open was successful.  Handle any locking requirements.
4437 	 */
4438 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
4439 		lf.l_whence = SEEK_SET;
4440 		lf.l_start = 0;
4441 		lf.l_len = 0;
4442 		if (fmode & O_EXLOCK)
4443 			lf.l_type = F_WRLCK;
4444 		else
4445 			lf.l_type = F_RDLCK;
4446 		if (fmode & FNONBLOCK)
4447 			type = 0;
4448 		else
4449 			type = F_WAIT;
4450 		vn_unlock(vp);
4451 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
4452 			/*
4453 			 * release our private reference.
4454 			 */
4455 			fsetfd(fdp, NULL, indx);
4456 			fdrop(fp);
4457 			vrele(vp);
4458 			goto done;
4459 		}
4460 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4461 		fp->f_flag |= FHASLOCK;
4462 	}
4463 
4464 	/*
4465 	 * Clean up.  Associate the file pointer with the previously
4466 	 * reserved descriptor and return it.
4467 	 */
4468 	vput(vp);
4469 	rel_mplock();
4470 	fsetfd(fdp, fp, indx);
4471 	fdrop(fp);
4472 	uap->sysmsg_result = indx;
4473 	return (0);
4474 
4475 bad_drop:
4476 	fsetfd(fdp, NULL, indx);
4477 	fdrop(fp);
4478 bad:
4479 	vput(vp);
4480 done:
4481 	rel_mplock();
4482 	return (error);
4483 }
4484 
4485 /*
4486  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
4487  *
4488  * MPALMOSTSAFE
4489  */
4490 int
4491 sys_fhstat(struct fhstat_args *uap)
4492 {
4493 	struct thread *td = curthread;
4494 	struct stat sb;
4495 	fhandle_t fh;
4496 	struct mount *mp;
4497 	struct vnode *vp;
4498 	int error;
4499 
4500 	/*
4501 	 * Must be super user
4502 	 */
4503 	error = priv_check(td, PRIV_ROOT);
4504 	if (error)
4505 		return (error);
4506 
4507 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
4508 	if (error)
4509 		return (error);
4510 
4511 	get_mplock();
4512 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
4513 		error = ESTALE;
4514 	if (error == 0) {
4515 		if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) == 0) {
4516 			error = vn_stat(vp, &sb, td->td_ucred);
4517 			vput(vp);
4518 		}
4519 	}
4520 	rel_mplock();
4521 	if (error == 0)
4522 		error = copyout(&sb, uap->sb, sizeof(sb));
4523 	return (error);
4524 }
4525 
4526 /*
4527  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
4528  *
4529  * MPALMOSTSAFE
4530  */
4531 int
4532 sys_fhstatfs(struct fhstatfs_args *uap)
4533 {
4534 	struct thread *td = curthread;
4535 	struct proc *p = td->td_proc;
4536 	struct statfs *sp;
4537 	struct mount *mp;
4538 	struct vnode *vp;
4539 	struct statfs sb;
4540 	char *fullpath, *freepath;
4541 	fhandle_t fh;
4542 	int error;
4543 
4544 	/*
4545 	 * Must be super user
4546 	 */
4547 	if ((error = priv_check(td, PRIV_ROOT)))
4548 		return (error);
4549 
4550 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4551 		return (error);
4552 
4553 	get_mplock();
4554 
4555 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4556 		error = ESTALE;
4557 		goto done;
4558 	}
4559 	if (p != NULL && !chroot_visible_mnt(mp, p)) {
4560 		error = ESTALE;
4561 		goto done;
4562 	}
4563 
4564 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)) != 0)
4565 		goto done;
4566 	mp = vp->v_mount;
4567 	sp = &mp->mnt_stat;
4568 	vput(vp);
4569 	if ((error = VFS_STATFS(mp, sp, td->td_ucred)) != 0)
4570 		goto done;
4571 
4572 	error = mount_path(p, mp, &fullpath, &freepath);
4573 	if (error)
4574 		goto done;
4575 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
4576 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
4577 	kfree(freepath, M_TEMP);
4578 
4579 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
4580 	if (priv_check(td, PRIV_ROOT)) {
4581 		bcopy(sp, &sb, sizeof(sb));
4582 		sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
4583 		sp = &sb;
4584 	}
4585 	error = copyout(sp, uap->buf, sizeof(*sp));
4586 done:
4587 	rel_mplock();
4588 	return (error);
4589 }
4590 
4591 /*
4592  * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
4593  *
4594  * MPALMOSTSAFE
4595  */
4596 int
4597 sys_fhstatvfs(struct fhstatvfs_args *uap)
4598 {
4599 	struct thread *td = curthread;
4600 	struct proc *p = td->td_proc;
4601 	struct statvfs *sp;
4602 	struct mount *mp;
4603 	struct vnode *vp;
4604 	fhandle_t fh;
4605 	int error;
4606 
4607 	/*
4608 	 * Must be super user
4609 	 */
4610 	if ((error = priv_check(td, PRIV_ROOT)))
4611 		return (error);
4612 
4613 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
4614 		return (error);
4615 
4616 	get_mplock();
4617 
4618 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL) {
4619 		error = ESTALE;
4620 		goto done;
4621 	}
4622 	if (p != NULL && !chroot_visible_mnt(mp, p)) {
4623 		error = ESTALE;
4624 		goto done;
4625 	}
4626 
4627 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
4628 		goto done;
4629 	mp = vp->v_mount;
4630 	sp = &mp->mnt_vstat;
4631 	vput(vp);
4632 	if ((error = VFS_STATVFS(mp, sp, td->td_ucred)) != 0)
4633 		goto done;
4634 
4635 	sp->f_flag = 0;
4636 	if (mp->mnt_flag & MNT_RDONLY)
4637 		sp->f_flag |= ST_RDONLY;
4638 	if (mp->mnt_flag & MNT_NOSUID)
4639 		sp->f_flag |= ST_NOSUID;
4640 	error = copyout(sp, uap->buf, sizeof(*sp));
4641 done:
4642 	rel_mplock();
4643 	return (error);
4644 }
4645 
4646 
4647 /*
4648  * Syscall to push extended attribute configuration information into the
4649  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
4650  * a command (int cmd), and attribute name and misc data.  For now, the
4651  * attribute name is left in userspace for consumption by the VFS_op.
4652  * It will probably be changed to be copied into sysspace by the
4653  * syscall in the future, once issues with various consumers of the
4654  * attribute code have raised their hands.
4655  *
4656  * Currently this is used only by UFS Extended Attributes.
4657  *
4658  * MPALMOSTSAFE
4659  */
4660 int
4661 sys_extattrctl(struct extattrctl_args *uap)
4662 {
4663 	struct nlookupdata nd;
4664 	struct vnode *vp;
4665 	char attrname[EXTATTR_MAXNAMELEN];
4666 	int error;
4667 	size_t size;
4668 
4669 	get_mplock();
4670 
4671 	attrname[0] = 0;
4672 	vp = NULL;
4673 	error = 0;
4674 
4675 	if (error == 0 && uap->filename) {
4676 		error = nlookup_init(&nd, uap->filename, UIO_USERSPACE,
4677 				     NLC_FOLLOW);
4678 		if (error == 0)
4679 			error = nlookup(&nd);
4680 		if (error == 0)
4681 			error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
4682 		nlookup_done(&nd);
4683 	}
4684 
4685 	if (error == 0 && uap->attrname) {
4686 		error = copyinstr(uap->attrname, attrname, EXTATTR_MAXNAMELEN,
4687 				  &size);
4688 	}
4689 
4690 	if (error == 0) {
4691 		error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4692 		if (error == 0)
4693 			error = nlookup(&nd);
4694 		if (error == 0)
4695 			error = ncp_writechk(&nd.nl_nch);
4696 		if (error == 0) {
4697 			error = VFS_EXTATTRCTL(nd.nl_nch.mount, uap->cmd, vp,
4698 					       uap->attrnamespace,
4699 					       uap->attrname, nd.nl_cred);
4700 		}
4701 		nlookup_done(&nd);
4702 	}
4703 
4704 	rel_mplock();
4705 
4706 	return (error);
4707 }
4708 
4709 /*
4710  * Syscall to get a named extended attribute on a file or directory.
4711  *
4712  * MPALMOSTSAFE
4713  */
4714 int
4715 sys_extattr_set_file(struct extattr_set_file_args *uap)
4716 {
4717 	char attrname[EXTATTR_MAXNAMELEN];
4718 	struct nlookupdata nd;
4719 	struct vnode *vp;
4720 	struct uio auio;
4721 	struct iovec aiov;
4722 	int error;
4723 
4724 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4725 	if (error)
4726 		return (error);
4727 
4728 	vp = NULL;
4729 	get_mplock();
4730 
4731 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4732 	if (error == 0)
4733 		error = nlookup(&nd);
4734 	if (error == 0)
4735 		error = ncp_writechk(&nd.nl_nch);
4736 	if (error == 0)
4737 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4738 	if (error) {
4739 		nlookup_done(&nd);
4740 		rel_mplock();
4741 		return (error);
4742 	}
4743 
4744 	bzero(&auio, sizeof(auio));
4745 	aiov.iov_base = uap->data;
4746 	aiov.iov_len = uap->nbytes;
4747 	auio.uio_iov = &aiov;
4748 	auio.uio_iovcnt = 1;
4749 	auio.uio_offset = 0;
4750 	auio.uio_resid = uap->nbytes;
4751 	auio.uio_rw = UIO_WRITE;
4752 	auio.uio_td = curthread;
4753 
4754 	error = VOP_SETEXTATTR(vp, uap->attrnamespace, attrname,
4755 			       &auio, nd.nl_cred);
4756 
4757 	vput(vp);
4758 	nlookup_done(&nd);
4759 	rel_mplock();
4760 	return (error);
4761 }
4762 
4763 /*
4764  * Syscall to get a named extended attribute on a file or directory.
4765  *
4766  * MPALMOSTSAFE
4767  */
4768 int
4769 sys_extattr_get_file(struct extattr_get_file_args *uap)
4770 {
4771 	char attrname[EXTATTR_MAXNAMELEN];
4772 	struct nlookupdata nd;
4773 	struct uio auio;
4774 	struct iovec aiov;
4775 	struct vnode *vp;
4776 	int error;
4777 
4778 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4779 	if (error)
4780 		return (error);
4781 
4782 	vp = NULL;
4783 	get_mplock();
4784 
4785 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4786 	if (error == 0)
4787 		error = nlookup(&nd);
4788 	if (error == 0)
4789 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4790 	if (error) {
4791 		nlookup_done(&nd);
4792 		rel_mplock();
4793 		return (error);
4794 	}
4795 
4796 	bzero(&auio, sizeof(auio));
4797 	aiov.iov_base = uap->data;
4798 	aiov.iov_len = uap->nbytes;
4799 	auio.uio_iov = &aiov;
4800 	auio.uio_iovcnt = 1;
4801 	auio.uio_offset = 0;
4802 	auio.uio_resid = uap->nbytes;
4803 	auio.uio_rw = UIO_READ;
4804 	auio.uio_td = curthread;
4805 
4806 	error = VOP_GETEXTATTR(vp, uap->attrnamespace, attrname,
4807 				&auio, nd.nl_cred);
4808 	uap->sysmsg_result = uap->nbytes - auio.uio_resid;
4809 
4810 	vput(vp);
4811 	nlookup_done(&nd);
4812 	rel_mplock();
4813 	return(error);
4814 }
4815 
4816 /*
4817  * Syscall to delete a named extended attribute from a file or directory.
4818  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
4819  *
4820  * MPALMOSTSAFE
4821  */
4822 int
4823 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
4824 {
4825 	char attrname[EXTATTR_MAXNAMELEN];
4826 	struct nlookupdata nd;
4827 	struct vnode *vp;
4828 	int error;
4829 
4830 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4831 	if (error)
4832 		return(error);
4833 
4834 	get_mplock();
4835 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4836 	if (error == 0)
4837 		error = nlookup(&nd);
4838 	if (error == 0)
4839 		error = ncp_writechk(&nd.nl_nch);
4840 	if (error == 0) {
4841 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4842 		if (error == 0) {
4843 			error = VOP_SETEXTATTR(vp, uap->attrnamespace,
4844 					       attrname, NULL, nd.nl_cred);
4845 			vput(vp);
4846 		}
4847 	}
4848 	nlookup_done(&nd);
4849 	rel_mplock();
4850 	return(error);
4851 }
4852 
4853 /*
4854  * Determine if the mount is visible to the process.
4855  */
4856 static int
4857 chroot_visible_mnt(struct mount *mp, struct proc *p)
4858 {
4859 	struct nchandle nch;
4860 
4861 	/*
4862 	 * Traverse from the mount point upwards.  If we hit the process
4863 	 * root then the mount point is visible to the process.
4864 	 */
4865 	nch = mp->mnt_ncmountpt;
4866 	while (nch.ncp) {
4867 		if (nch.mount == p->p_fd->fd_nrdir.mount &&
4868 		    nch.ncp == p->p_fd->fd_nrdir.ncp) {
4869 			return(1);
4870 		}
4871 		if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
4872 			nch = nch.mount->mnt_ncmounton;
4873 		} else {
4874 			nch.ncp = nch.ncp->nc_parent;
4875 		}
4876 	}
4877 
4878 	/*
4879 	 * If the mount point is not visible to the process, but the
4880 	 * process root is in a subdirectory of the mount, return
4881 	 * TRUE anyway.
4882 	 */
4883 	if (p->p_fd->fd_nrdir.mount == mp)
4884 		return(1);
4885 
4886 	return(0);
4887 }
4888 
4889