xref: /dflybsd-src/sys/kern/vfs_syscalls.c (revision ee61f228c1c38342493a01fe774b83cffc0f4be9)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_syscalls.c	8.13 (Berkeley) 4/15/94
39  * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $
40  * $DragonFly: src/sys/kern/vfs_syscalls.c,v 1.74 2005/10/09 18:07:55 corecode Exp $
41  */
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/buf.h>
46 #include <sys/conf.h>
47 #include <sys/sysent.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mountctl.h>
51 #include <sys/sysproto.h>
52 #include <sys/filedesc.h>
53 #include <sys/kernel.h>
54 #include <sys/fcntl.h>
55 #include <sys/file.h>
56 #include <sys/linker.h>
57 #include <sys/stat.h>
58 #include <sys/unistd.h>
59 #include <sys/vnode.h>
60 #include <sys/proc.h>
61 #include <sys/namei.h>
62 #include <sys/nlookup.h>
63 #include <sys/dirent.h>
64 #include <sys/extattr.h>
65 #include <sys/kern_syscall.h>
66 
67 #include <machine/limits.h>
68 #include <vfs/union/union.h>
69 #include <sys/sysctl.h>
70 #include <vm/vm.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_zone.h>
73 #include <vm/vm_page.h>
74 
75 #include <sys/file2.h>
76 
77 static int checkvp_chdir (struct vnode *vn, struct thread *td);
78 static void checkdirs (struct vnode *olddp, struct namecache *ncp);
79 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
80 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
81 static int getutimes (const struct timeval *, struct timespec *);
82 static int setfown (struct vnode *, uid_t, gid_t);
83 static int setfmode (struct vnode *, int);
84 static int setfflags (struct vnode *, int);
85 static int setutimes (struct vnode *, const struct timespec *, int);
86 static int	usermount = 0;	/* if 1, non-root can mount fs. */
87 
88 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *);
89 
90 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "");
91 
92 /*
93  * Virtual File System System Calls
94  */
95 
96 /*
97  * Mount a file system.
98  */
99 /*
100  * mount_args(char *type, char *path, int flags, caddr_t data)
101  */
102 /* ARGSUSED */
103 int
104 mount(struct mount_args *uap)
105 {
106 	struct thread *td = curthread;
107 	struct proc *p = td->td_proc;
108 	struct vnode *vp;
109 	struct namecache *ncp;
110 	struct mount *mp;
111 	struct vfsconf *vfsp;
112 	int error, flag = 0, flag2 = 0;
113 	struct vattr va;
114 	struct nlookupdata nd;
115 	char fstypename[MFSNAMELEN];
116 	struct nlcomponent nlc;
117 
118 	KKASSERT(p);
119 	if (p->p_ucred->cr_prison != NULL)
120 		return (EPERM);
121 	if (usermount == 0 && (error = suser(td)))
122 		return (error);
123 	/*
124 	 * Do not allow NFS export by non-root users.
125 	 */
126 	if (uap->flags & MNT_EXPORTED) {
127 		error = suser(td);
128 		if (error)
129 			return (error);
130 	}
131 	/*
132 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
133 	 */
134 	if (suser(td))
135 		uap->flags |= MNT_NOSUID | MNT_NODEV;
136 
137 	/*
138 	 * Lookup the requested path and extract the ncp and vnode.
139 	 */
140 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
141 	if (error == 0) {
142 		if ((error = nlookup(&nd)) == 0) {
143 			if (nd.nl_ncp->nc_vp == NULL)
144 				error = ENOENT;
145 		}
146 	}
147 	if (error) {
148 		nlookup_done(&nd);
149 		return (error);
150 	}
151 
152 	/*
153 	 * Extract the locked+refd ncp and cleanup the nd structure
154 	 */
155 	ncp = nd.nl_ncp;
156 	nd.nl_ncp = NULL;
157 	nlookup_done(&nd);
158 
159 	/*
160 	 * now we have the locked ref'd ncp and unreferenced vnode.
161 	 */
162 	vp = ncp->nc_vp;
163 	if ((error = vget(vp, LK_EXCLUSIVE, td)) != 0) {
164 		cache_put(ncp);
165 		return (error);
166 	}
167 	cache_unlock(ncp);
168 
169 	/*
170 	 * Now we have an unlocked ref'd ncp and a locked ref'd vp
171 	 */
172 	if (uap->flags & MNT_UPDATE) {
173 		if ((vp->v_flag & VROOT) == 0) {
174 			cache_drop(ncp);
175 			vput(vp);
176 			return (EINVAL);
177 		}
178 		mp = vp->v_mount;
179 		flag = mp->mnt_flag;
180 		flag2 = mp->mnt_kern_flag;
181 		/*
182 		 * We only allow the filesystem to be reloaded if it
183 		 * is currently mounted read-only.
184 		 */
185 		if ((uap->flags & MNT_RELOAD) &&
186 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
187 			cache_drop(ncp);
188 			vput(vp);
189 			return (EOPNOTSUPP);	/* Needs translation */
190 		}
191 		/*
192 		 * Only root, or the user that did the original mount is
193 		 * permitted to update it.
194 		 */
195 		if (mp->mnt_stat.f_owner != p->p_ucred->cr_uid &&
196 		    (error = suser(td))) {
197 			cache_drop(ncp);
198 			vput(vp);
199 			return (error);
200 		}
201 		if (vfs_busy(mp, LK_NOWAIT, td)) {
202 			cache_drop(ncp);
203 			vput(vp);
204 			return (EBUSY);
205 		}
206 		if ((vp->v_flag & VMOUNT) != 0 ||
207 		    vp->v_mountedhere != NULL) {
208 			cache_drop(ncp);
209 			vfs_unbusy(mp, td);
210 			vput(vp);
211 			return (EBUSY);
212 		}
213 		vp->v_flag |= VMOUNT;
214 		mp->mnt_flag |=
215 		    uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
216 		VOP_UNLOCK(vp, 0, td);
217 		goto update;
218 	}
219 	/*
220 	 * If the user is not root, ensure that they own the directory
221 	 * onto which we are attempting to mount.
222 	 */
223 	if ((error = VOP_GETATTR(vp, &va, td)) ||
224 	    (va.va_uid != p->p_ucred->cr_uid &&
225 	     (error = suser(td)))) {
226 		cache_drop(ncp);
227 		vput(vp);
228 		return (error);
229 	}
230 	if ((error = vinvalbuf(vp, V_SAVE, td, 0, 0)) != 0) {
231 		cache_drop(ncp);
232 		vput(vp);
233 		return (error);
234 	}
235 	if (vp->v_type != VDIR) {
236 		cache_drop(ncp);
237 		vput(vp);
238 		return (ENOTDIR);
239 	}
240 	if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
241 		cache_drop(ncp);
242 		vput(vp);
243 		return (error);
244 	}
245 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
246 		if (!strcmp(vfsp->vfc_name, fstypename))
247 			break;
248 	}
249 	if (vfsp == NULL) {
250 		linker_file_t lf;
251 
252 		/* Only load modules for root (very important!) */
253 		if ((error = suser(td)) != 0) {
254 			cache_drop(ncp);
255 			vput(vp);
256 			return error;
257 		}
258 		error = linker_load_file(fstypename, &lf);
259 		if (error || lf == NULL) {
260 			cache_drop(ncp);
261 			vput(vp);
262 			if (lf == NULL)
263 				error = ENODEV;
264 			return error;
265 		}
266 		lf->userrefs++;
267 		/* lookup again, see if the VFS was loaded */
268 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
269 			if (!strcmp(vfsp->vfc_name, fstypename))
270 				break;
271 		}
272 		if (vfsp == NULL) {
273 			lf->userrefs--;
274 			linker_file_unload(lf);
275 			cache_drop(ncp);
276 			vput(vp);
277 			return (ENODEV);
278 		}
279 	}
280 	if ((vp->v_flag & VMOUNT) != 0 ||
281 	    vp->v_mountedhere != NULL) {
282 		cache_drop(ncp);
283 		vput(vp);
284 		return (EBUSY);
285 	}
286 	vp->v_flag |= VMOUNT;
287 
288 	/*
289 	 * Allocate and initialize the filesystem.
290 	 */
291 	mp = malloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
292 	TAILQ_INIT(&mp->mnt_nvnodelist);
293 	TAILQ_INIT(&mp->mnt_reservedvnlist);
294 	TAILQ_INIT(&mp->mnt_jlist);
295 	mp->mnt_nvnodelistsize = 0;
296 	lockinit(&mp->mnt_lock, 0, "vfslock", 0, LK_NOPAUSE);
297 	vfs_busy(mp, LK_NOWAIT, td);
298 	mp->mnt_op = vfsp->vfc_vfsops;
299 	mp->mnt_vfc = vfsp;
300 	vfsp->vfc_refcount++;
301 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
302 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
303 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
304 	mp->mnt_vnodecovered = vp;
305 	mp->mnt_stat.f_owner = p->p_ucred->cr_uid;
306 	mp->mnt_iosize_max = DFLTPHYS;
307 	VOP_UNLOCK(vp, 0, td);
308 update:
309 	/*
310 	 * Set the mount level flags.
311 	 */
312 	if (uap->flags & MNT_RDONLY)
313 		mp->mnt_flag |= MNT_RDONLY;
314 	else if (mp->mnt_flag & MNT_RDONLY)
315 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
316 	mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
317 	    MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
318 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
319 	    MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
320 	mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
321 	    MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
322 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
323 	    MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
324 	/*
325 	 * Mount the filesystem.
326 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
327 	 * get.
328 	 */
329 	error = VFS_MOUNT(mp, uap->path, uap->data, td);
330 	if (mp->mnt_flag & MNT_UPDATE) {
331 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
332 			mp->mnt_flag &= ~MNT_RDONLY;
333 		mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
334 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
335 		if (error) {
336 			mp->mnt_flag = flag;
337 			mp->mnt_kern_flag = flag2;
338 		}
339 		vfs_unbusy(mp, td);
340 		vp->v_flag &= ~VMOUNT;
341 		vrele(vp);
342 		cache_drop(ncp);
343 		return (error);
344 	}
345 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
346 	/*
347 	 * Put the new filesystem on the mount list after root.  The mount
348 	 * point gets its own mnt_ncp which is a special ncp linking the
349 	 * vnode-under to the root of the new mount.  The lookup code
350 	 * detects the mount point going forward and detects the special
351 	 * mnt_ncp via NCP_MOUNTPT going backwards.
352 	 *
353 	 * It is not necessary to invalidate or purge the vnode underneath
354 	 * because elements under the mount will be given their own glue
355 	 * namecache record.
356 	 */
357 	if (!error) {
358 		nlc.nlc_nameptr = "";
359 		nlc.nlc_namelen = 0;
360 		mp->mnt_ncp = cache_nlookup(ncp, &nlc);
361 		cache_setunresolved(mp->mnt_ncp);
362 		mp->mnt_ncp->nc_flag |= NCF_MOUNTPT;
363 		mp->mnt_ncp->nc_mount = mp;
364 		cache_drop(ncp);
365 		/* XXX get the root of the fs and cache_setvp(mnt_ncp...) */
366 		vp->v_flag &= ~VMOUNT;
367 		vp->v_mountedhere = mp;
368 		mountlist_insert(mp, MNTINS_LAST);
369 		checkdirs(vp, mp->mnt_ncp);
370 		cache_unlock(mp->mnt_ncp);	/* leave ref intact */
371 		VOP_UNLOCK(vp, 0, td);
372 		error = vfs_allocate_syncvnode(mp);
373 		vfs_unbusy(mp, td);
374 		if ((error = VFS_START(mp, 0, td)) != 0)
375 			vrele(vp);
376 	} else {
377 		vfs_rm_vnodeops(&mp->mnt_vn_coherency_ops);
378 		vfs_rm_vnodeops(&mp->mnt_vn_journal_ops);
379 		vfs_rm_vnodeops(&mp->mnt_vn_norm_ops);
380 		vfs_rm_vnodeops(&mp->mnt_vn_spec_ops);
381 		vfs_rm_vnodeops(&mp->mnt_vn_fifo_ops);
382 		vp->v_flag &= ~VMOUNT;
383 		mp->mnt_vfc->vfc_refcount--;
384 		vfs_unbusy(mp, td);
385 		free(mp, M_MOUNT);
386 		cache_drop(ncp);
387 		vput(vp);
388 	}
389 	return (error);
390 }
391 
392 /*
393  * Scan all active processes to see if any of them have a current
394  * or root directory onto which the new filesystem has just been
395  * mounted. If so, replace them with the new mount point.
396  *
397  * The passed ncp is ref'd and locked (from the mount code) and
398  * must be associated with the vnode representing the root of the
399  * mount point.
400  */
401 static void
402 checkdirs(struct vnode *olddp, struct namecache *ncp)
403 {
404 	struct filedesc *fdp;
405 	struct vnode *newdp;
406 	struct mount *mp;
407 	struct proc *p;
408 
409 	if (olddp->v_usecount == 1)
410 		return;
411 	mp = olddp->v_mountedhere;
412 	if (VFS_ROOT(mp, &newdp))
413 		panic("mount: lost mount");
414 	cache_setvp(ncp, newdp);
415 
416 	if (rootvnode == olddp) {
417 		vref(newdp);
418 		vfs_cache_setroot(newdp, cache_hold(ncp));
419 	}
420 
421 	FOREACH_PROC_IN_SYSTEM(p) {
422 		fdp = p->p_fd;
423 		if (fdp->fd_cdir == olddp) {
424 			vrele(fdp->fd_cdir);
425 			vref(newdp);
426 			fdp->fd_cdir = newdp;
427 			cache_drop(fdp->fd_ncdir);
428 			fdp->fd_ncdir = cache_hold(ncp);
429 		}
430 		if (fdp->fd_rdir == olddp) {
431 			vrele(fdp->fd_rdir);
432 			vref(newdp);
433 			fdp->fd_rdir = newdp;
434 			cache_drop(fdp->fd_nrdir);
435 			fdp->fd_nrdir = cache_hold(ncp);
436 		}
437 	}
438 	vput(newdp);
439 }
440 
441 /*
442  * Unmount a file system.
443  *
444  * Note: unmount takes a path to the vnode mounted on as argument,
445  * not special file (as before).
446  */
447 /*
448  * umount_args(char *path, int flags)
449  */
450 /* ARGSUSED */
451 int
452 unmount(struct unmount_args *uap)
453 {
454 	struct thread *td = curthread;
455 	struct proc *p = td->td_proc;
456 	struct vnode *vp;
457 	struct mount *mp;
458 	int error;
459 	struct nlookupdata nd;
460 
461 	KKASSERT(p);
462 	if (p->p_ucred->cr_prison != NULL)
463 		return (EPERM);
464 	if (usermount == 0 && (error = suser(td)))
465 		return (error);
466 
467 	vp = NULL;
468 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
469 	if (error == 0)
470 		error = nlookup(&nd);
471 	if (error == 0)
472 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
473 	nlookup_done(&nd);
474 	if (error)
475 		return (error);
476 
477 	mp = vp->v_mount;
478 
479 	/*
480 	 * Only root, or the user that did the original mount is
481 	 * permitted to unmount this filesystem.
482 	 */
483 	if ((mp->mnt_stat.f_owner != p->p_ucred->cr_uid) &&
484 	    (error = suser(td))) {
485 		vput(vp);
486 		return (error);
487 	}
488 
489 	/*
490 	 * Don't allow unmounting the root file system.
491 	 */
492 	if (mp->mnt_flag & MNT_ROOTFS) {
493 		vput(vp);
494 		return (EINVAL);
495 	}
496 
497 	/*
498 	 * Must be the root of the filesystem
499 	 */
500 	if ((vp->v_flag & VROOT) == 0) {
501 		vput(vp);
502 		return (EINVAL);
503 	}
504 	vput(vp);
505 	return (dounmount(mp, uap->flags, td));
506 }
507 
508 /*
509  * Do the actual file system unmount.
510  */
511 static int
512 dounmount_interlock(struct mount *mp)
513 {
514 	if (mp->mnt_kern_flag & MNTK_UNMOUNT)
515 		return (EBUSY);
516 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
517 	return(0);
518 }
519 
520 int
521 dounmount(struct mount *mp, int flags, struct thread *td)
522 {
523 	struct vnode *coveredvp;
524 	int error;
525 	int async_flag;
526 
527 	/*
528 	 * Exclusive access for unmounting purposes
529 	 */
530 	if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
531 		return (error);
532 
533 	/*
534 	 * Allow filesystems to detect that a forced unmount is in progress.
535 	 */
536 	if (flags & MNT_FORCE)
537 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
538 	error = lockmgr(&mp->mnt_lock, LK_DRAIN |
539 	    ((flags & MNT_FORCE) ? 0 : LK_NOWAIT), NULL, td);
540 	if (error) {
541 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
542 		if (mp->mnt_kern_flag & MNTK_MWAIT)
543 			wakeup(mp);
544 		return (error);
545 	}
546 
547 	if (mp->mnt_flag & MNT_EXPUBLIC)
548 		vfs_setpublicfs(NULL, NULL, NULL);
549 
550 	vfs_msync(mp, MNT_WAIT);
551 	async_flag = mp->mnt_flag & MNT_ASYNC;
552 	mp->mnt_flag &=~ MNT_ASYNC;
553 	cache_purgevfs(mp);	/* remove cache entries for this file sys */
554 	if (mp->mnt_syncer != NULL)
555 		vrele(mp->mnt_syncer);
556 	if (((mp->mnt_flag & MNT_RDONLY) ||
557 	     (error = VFS_SYNC(mp, MNT_WAIT, td)) == 0) ||
558 	    (flags & MNT_FORCE))
559 		error = VFS_UNMOUNT(mp, flags, td);
560 	if (error) {
561 		if (mp->mnt_syncer == NULL)
562 			vfs_allocate_syncvnode(mp);
563 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
564 		mp->mnt_flag |= async_flag;
565 		lockmgr(&mp->mnt_lock, LK_RELEASE | LK_REENABLE, NULL, td);
566 		if (mp->mnt_kern_flag & MNTK_MWAIT)
567 			wakeup(mp);
568 		return (error);
569 	}
570 	/*
571 	 * Clean up any journals still associated with the mount after
572 	 * filesystem activity has ceased.
573 	 */
574 	journal_remove_all_journals(mp,
575 	    ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
576 
577 	mountlist_remove(mp);
578 
579 	/*
580 	 * Remove any installed vnode ops here so the individual VFSs don't
581 	 * have to.
582 	 */
583 	vfs_rm_vnodeops(&mp->mnt_vn_coherency_ops);
584 	vfs_rm_vnodeops(&mp->mnt_vn_journal_ops);
585 	vfs_rm_vnodeops(&mp->mnt_vn_norm_ops);
586 	vfs_rm_vnodeops(&mp->mnt_vn_spec_ops);
587 	vfs_rm_vnodeops(&mp->mnt_vn_fifo_ops);
588 
589 	if ((coveredvp = mp->mnt_vnodecovered) != NULLVP) {
590 		coveredvp->v_mountedhere = NULL;
591 		vrele(coveredvp);
592 		cache_drop(mp->mnt_ncp);
593 		mp->mnt_ncp = NULL;
594 	}
595 	mp->mnt_vfc->vfc_refcount--;
596 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
597 		panic("unmount: dangling vnode");
598 	lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
599 	if (mp->mnt_kern_flag & MNTK_MWAIT)
600 		wakeup(mp);
601 	free(mp, M_MOUNT);
602 	return (0);
603 }
604 
605 /*
606  * Sync each mounted filesystem.
607  */
608 
609 #ifdef DEBUG
610 static int syncprt = 0;
611 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
612 #endif /* DEBUG */
613 
614 static int sync_callback(struct mount *mp, void *data);
615 
616 /* ARGSUSED */
617 int
618 sync(struct sync_args *uap)
619 {
620 	mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
621 #ifdef DEBUG
622 	/*
623 	 * print out buffer pool stat information on each sync() call.
624 	 */
625 	if (syncprt)
626 		vfs_bufstats();
627 #endif /* DEBUG */
628 	return (0);
629 }
630 
631 static
632 int
633 sync_callback(struct mount *mp, void *data __unused)
634 {
635 	int asyncflag;
636 
637 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
638 		asyncflag = mp->mnt_flag & MNT_ASYNC;
639 		mp->mnt_flag &= ~MNT_ASYNC;
640 		vfs_msync(mp, MNT_NOWAIT);
641 		VFS_SYNC(mp, MNT_NOWAIT, curthread);
642 		mp->mnt_flag |= asyncflag;
643 	}
644 	return(0);
645 }
646 
647 /* XXX PRISON: could be per prison flag */
648 static int prison_quotas;
649 #if 0
650 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
651 #endif
652 
653 /*
654  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
655  *
656  * Change filesystem quotas.
657  */
658 /* ARGSUSED */
659 int
660 quotactl(struct quotactl_args *uap)
661 {
662 	struct nlookupdata nd;
663 	struct thread *td;
664 	struct proc *p;
665 	struct mount *mp;
666 	int error;
667 
668 	td = curthread;
669 	p = td->td_proc;
670 	if (p->p_ucred->cr_prison && !prison_quotas)
671 		return (EPERM);
672 
673 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
674 	if (error == 0)
675 		error = nlookup(&nd);
676 	if (error == 0) {
677 		mp = nd.nl_ncp->nc_mount;
678 		error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
679 				    uap->arg, nd.nl_td);
680 	}
681 	nlookup_done(&nd);
682 	return (error);
683 }
684 
685 /*
686  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
687  *		void *buf, int buflen)
688  *
689  * This function operates on a mount point and executes the specified
690  * operation using the specified control data, and possibly returns data.
691  *
692  * The actual number of bytes stored in the result buffer is returned, 0
693  * if none, otherwise an error is returned.
694  */
695 /* ARGSUSED */
696 int
697 mountctl(struct mountctl_args *uap)
698 {
699 	struct thread *td = curthread;
700 	struct proc *p = td->td_proc;
701 	struct filedesc *fdp = p->p_fd;
702 	struct file *fp;
703 	void *ctl = NULL;
704 	void *buf = NULL;
705 	char *path = NULL;
706 	int error;
707 
708 	/*
709 	 * Sanity and permissions checks.  We must be root.
710 	 */
711 	KKASSERT(p);
712 	if (p->p_ucred->cr_prison != NULL)
713 		return (EPERM);
714 	if ((error = suser(td)) != 0)
715 		return (error);
716 
717 	/*
718 	 * Argument length checks
719 	 */
720 	if (uap->ctllen < 0 || uap->ctllen > 1024)
721 		return (EINVAL);
722 	if (uap->buflen < 0 || uap->buflen > 16 * 1024)
723 		return (EINVAL);
724 	if (uap->path == NULL)
725 		return (EINVAL);
726 
727 	/*
728 	 * Allocate the necessary buffers and copyin data
729 	 */
730 	path = zalloc(namei_zone);
731 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
732 	if (error)
733 		goto done;
734 
735 	if (uap->ctllen) {
736 		ctl = malloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
737 		error = copyin(uap->ctl, ctl, uap->ctllen);
738 		if (error)
739 			goto done;
740 	}
741 	if (uap->buflen)
742 		buf = malloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
743 
744 	/*
745 	 * Validate the descriptor
746 	 */
747 	if (uap->fd == -1) {
748 		fp = NULL;
749 	} else if ((u_int)uap->fd >= fdp->fd_nfiles ||
750 	    (fp = fdp->fd_files[uap->fd].fp) == NULL) {
751 		error = EBADF;
752 		goto done;
753 	}
754 	if (fp)
755 		fhold(fp);
756 
757 	/*
758 	 * Execute the internal kernel function and clean up.
759 	 */
760 	error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
761 	if (fp)
762 		fdrop(fp, td);
763 	if (error == 0 && uap->sysmsg_result > 0)
764 		error = copyout(buf, uap->buf, uap->sysmsg_result);
765 done:
766 	if (path)
767 		zfree(namei_zone, path);
768 	if (ctl)
769 		free(ctl, M_TEMP);
770 	if (buf)
771 		free(buf, M_TEMP);
772 	return (error);
773 }
774 
775 /*
776  * Execute a mount control operation by resolving the path to a mount point
777  * and calling vop_mountctl().
778  */
779 int
780 kern_mountctl(const char *path, int op, struct file *fp,
781 		const void *ctl, int ctllen,
782 		void *buf, int buflen, int *res)
783 {
784 	struct vnode *vp;
785 	struct mount *mp;
786 	struct nlookupdata nd;
787 	int error;
788 
789 	*res = 0;
790 	vp = NULL;
791 	error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
792 	if (error == 0)
793 		error = nlookup(&nd);
794 	if (error == 0)
795 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
796 	nlookup_done(&nd);
797 	if (error)
798 		return (error);
799 
800 	mp = vp->v_mount;
801 
802 	/*
803 	 * Must be the root of the filesystem
804 	 */
805 	if ((vp->v_flag & VROOT) == 0) {
806 		vput(vp);
807 		return (EINVAL);
808 	}
809 	error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen,
810 				buf, buflen, res);
811 	vput(vp);
812 	return (error);
813 }
814 
815 int
816 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
817 {
818 	struct thread *td = curthread;
819 	struct proc *p = td->td_proc;
820 	struct mount *mp;
821 	struct statfs *sp;
822 	char *fullpath, *freepath;
823 	int error;
824 
825 	if ((error = nlookup(nd)) != 0)
826 		return (error);
827 	mp = nd->nl_ncp->nc_mount;
828 	sp = &mp->mnt_stat;
829 	if ((error = VFS_STATFS(mp, sp, td)) != 0)
830 		return (error);
831 
832 	error = cache_fullpath(p, mp->mnt_ncp, &fullpath, &freepath);
833 	if (error)
834 		return(error);
835 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
836 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
837 	free(freepath, M_TEMP);
838 
839 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
840 	bcopy(sp, buf, sizeof(*buf));
841 	/* Only root should have access to the fsid's. */
842 	if (suser(td))
843 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
844 	return (0);
845 }
846 
847 /*
848  * statfs_args(char *path, struct statfs *buf)
849  *
850  * Get filesystem statistics.
851  */
852 int
853 statfs(struct statfs_args *uap)
854 {
855 	struct nlookupdata nd;
856 	struct statfs buf;
857 	int error;
858 
859 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
860 	if (error == 0)
861 		error = kern_statfs(&nd, &buf);
862 	nlookup_done(&nd);
863 	if (error == 0)
864 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
865 	return (error);
866 }
867 
868 int
869 kern_fstatfs(int fd, struct statfs *buf)
870 {
871 	struct thread *td = curthread;
872 	struct proc *p = td->td_proc;
873 	struct file *fp;
874 	struct mount *mp;
875 	struct statfs *sp;
876 	char *fullpath, *freepath;
877 	int error;
878 
879 	KKASSERT(p);
880 	error = getvnode(p->p_fd, fd, &fp);
881 	if (error)
882 		return (error);
883 	mp = ((struct vnode *)fp->f_data)->v_mount;
884 	if (mp == NULL)
885 		return (EBADF);
886 	sp = &mp->mnt_stat;
887 	error = VFS_STATFS(mp, sp, td);
888 	if (error)
889 		return (error);
890 
891 	error = cache_fullpath(p, mp->mnt_ncp, &fullpath, &freepath);
892 	if (error)
893 		return(error);
894 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
895 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
896 	free(freepath, M_TEMP);
897 
898 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
899 	bcopy(sp, buf, sizeof(*buf));
900 
901 	/* Only root should have access to the fsid's. */
902 	if (suser(td))
903 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
904 	return (0);
905 }
906 
907 /*
908  * fstatfs_args(int fd, struct statfs *buf)
909  *
910  * Get filesystem statistics.
911  */
912 int
913 fstatfs(struct fstatfs_args *uap)
914 {
915 	struct statfs buf;
916 	int error;
917 
918 	error = kern_fstatfs(uap->fd, &buf);
919 
920 	if (error == 0)
921 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
922 	return (error);
923 }
924 
925 /*
926  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
927  *
928  * Get statistics on all filesystems.
929  */
930 
931 struct getfsstat_info {
932 	struct statfs *sfsp;
933 	long count;
934 	long maxcount;
935 	int error;
936 	int flags;
937 	int is_chrooted;
938 	struct thread *td;
939 	struct proc *p;
940 };
941 
942 static int getfsstat_callback(struct mount *, void *);
943 
944 /* ARGSUSED */
945 int
946 getfsstat(struct getfsstat_args *uap)
947 {
948 	struct thread *td = curthread;
949 	struct proc *p = td->td_proc;
950 	struct getfsstat_info info;
951 
952 	bzero(&info, sizeof(info));
953 	if (p != NULL && (p->p_fd->fd_nrdir->nc_flag & NCF_ROOT) == 0)
954 		info.is_chrooted = 1;
955 	else
956 		info.is_chrooted = 0;
957 
958 	info.maxcount = uap->bufsize / sizeof(struct statfs);
959 	info.sfsp = uap->buf;
960 	info.count = 0;
961 	info.flags = uap->flags;
962 	info.td = td;
963 	info.p = p;
964 
965 	mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
966 	if (info.sfsp && info.count > info.maxcount)
967 		uap->sysmsg_result = info.maxcount;
968 	else
969 		uap->sysmsg_result = info.count;
970 	return (info.error);
971 }
972 
973 static int
974 getfsstat_callback(struct mount *mp, void *data)
975 {
976 	struct getfsstat_info *info = data;
977 	struct statfs *sp;
978 	char *freepath;
979 	char *fullpath;
980 	int error;
981 
982 	if (info->sfsp && info->count < info->maxcount) {
983 		if (info->is_chrooted && !chroot_visible_mnt(mp, info->p))
984 			return(0);
985 		sp = &mp->mnt_stat;
986 
987 		/*
988 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
989 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
990 		 * overrides MNT_WAIT.
991 		 */
992 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
993 		    (info->flags & MNT_WAIT)) &&
994 		    (error = VFS_STATFS(mp, sp, info->td))) {
995 			return(0);
996 		}
997 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
998 
999 		error = cache_fullpath(info->p, mp->mnt_ncp,
1000 					&fullpath, &freepath);
1001 		if (error) {
1002 			info->error = error;
1003 			return(-1);
1004 		}
1005 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1006 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1007 		free(freepath, M_TEMP);
1008 
1009 		error = copyout(sp, info->sfsp, sizeof(*sp));
1010 		if (error) {
1011 			info->error = error;
1012 			return (-1);
1013 		}
1014 		++info->sfsp;
1015 	}
1016 	info->count++;
1017 	return(0);
1018 }
1019 
1020 /*
1021  * fchdir_args(int fd)
1022  *
1023  * Change current working directory to a given file descriptor.
1024  */
1025 /* ARGSUSED */
1026 int
1027 fchdir(struct fchdir_args *uap)
1028 {
1029 	struct thread *td = curthread;
1030 	struct proc *p = td->td_proc;
1031 	struct filedesc *fdp = p->p_fd;
1032 	struct vnode *vp, *ovp;
1033 	struct mount *mp;
1034 	struct file *fp;
1035 	struct namecache *ncp, *oncp;
1036 	struct namecache *nct;
1037 	int error;
1038 
1039 	if ((error = getvnode(fdp, uap->fd, &fp)) != 0)
1040 		return (error);
1041 	vp = (struct vnode *)fp->f_data;
1042 	vref(vp);
1043 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1044 	if (vp->v_type != VDIR || fp->f_ncp == NULL)
1045 		error = ENOTDIR;
1046 	else
1047 		error = VOP_ACCESS(vp, VEXEC, p->p_ucred, td);
1048 	if (error) {
1049 		vput(vp);
1050 		return (error);
1051 	}
1052 	ncp = cache_hold(fp->f_ncp);
1053 	while (!error && (mp = vp->v_mountedhere) != NULL) {
1054 		error = nlookup_mp(mp, &nct);
1055 		if (error == 0) {
1056 			cache_unlock(nct);	/* leave ref intact */
1057 			vput(vp);
1058 			vp = nct->nc_vp;
1059 			error = vget(vp, LK_SHARED, td);
1060 			KKASSERT(error == 0);
1061 			cache_drop(ncp);
1062 			ncp = nct;
1063 		}
1064 	}
1065 	if (error == 0) {
1066 		ovp = fdp->fd_cdir;
1067 		oncp = fdp->fd_ncdir;
1068 		VOP_UNLOCK(vp, 0, td);	/* leave ref intact */
1069 		fdp->fd_cdir = vp;
1070 		fdp->fd_ncdir = ncp;
1071 		cache_drop(oncp);
1072 		vrele(ovp);
1073 	} else {
1074 		cache_drop(ncp);
1075 		vput(vp);
1076 	}
1077 	return (error);
1078 }
1079 
1080 int
1081 kern_chdir(struct nlookupdata *nd)
1082 {
1083 	struct thread *td = curthread;
1084 	struct proc *p = td->td_proc;
1085 	struct filedesc *fdp = p->p_fd;
1086 	struct vnode *vp, *ovp;
1087 	struct namecache *oncp;
1088 	int error;
1089 
1090 	if ((error = nlookup(nd)) != 0)
1091 		return (error);
1092 	if ((vp = nd->nl_ncp->nc_vp) == NULL)
1093 		return (ENOENT);
1094 	if ((error = vget(vp, LK_SHARED, td)) != 0)
1095 		return (error);
1096 
1097 	error = checkvp_chdir(vp, td);
1098 	VOP_UNLOCK(vp, 0, td);
1099 	if (error == 0) {
1100 		ovp = fdp->fd_cdir;
1101 		oncp = fdp->fd_ncdir;
1102 		cache_unlock(nd->nl_ncp);	/* leave reference intact */
1103 		fdp->fd_ncdir = nd->nl_ncp;
1104 		fdp->fd_cdir = vp;
1105 		cache_drop(oncp);
1106 		vrele(ovp);
1107 		nd->nl_ncp = NULL;
1108 	} else {
1109 		vrele(vp);
1110 	}
1111 	return (error);
1112 }
1113 
1114 /*
1115  * chdir_args(char *path)
1116  *
1117  * Change current working directory (``.'').
1118  */
1119 int
1120 chdir(struct chdir_args *uap)
1121 {
1122 	struct nlookupdata nd;
1123 	int error;
1124 
1125 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1126 	if (error == 0)
1127 		error = kern_chdir(&nd);
1128 	nlookup_done(&nd);
1129 	return (error);
1130 }
1131 
1132 /*
1133  * Helper function for raised chroot(2) security function:  Refuse if
1134  * any filedescriptors are open directories.
1135  */
1136 static int
1137 chroot_refuse_vdir_fds(fdp)
1138 	struct filedesc *fdp;
1139 {
1140 	struct vnode *vp;
1141 	struct file *fp;
1142 	int error;
1143 	int fd;
1144 
1145 	for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1146 		error = getvnode(fdp, fd, &fp);
1147 		if (error)
1148 			continue;
1149 		vp = (struct vnode *)fp->f_data;
1150 		if (vp->v_type != VDIR)
1151 			continue;
1152 		return(EPERM);
1153 	}
1154 	return (0);
1155 }
1156 
1157 /*
1158  * This sysctl determines if we will allow a process to chroot(2) if it
1159  * has a directory open:
1160  *	0: disallowed for all processes.
1161  *	1: allowed for processes that were not already chroot(2)'ed.
1162  *	2: allowed for all processes.
1163  */
1164 
1165 static int chroot_allow_open_directories = 1;
1166 
1167 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1168      &chroot_allow_open_directories, 0, "");
1169 
1170 /*
1171  * chroot to the specified namecache entry.  We obtain the vp from the
1172  * namecache data.  The passed ncp must be locked and referenced and will
1173  * remain locked and referenced on return.
1174  */
1175 int
1176 kern_chroot(struct namecache *ncp)
1177 {
1178 	struct thread *td = curthread;
1179 	struct proc *p = td->td_proc;
1180 	struct filedesc *fdp = p->p_fd;
1181 	struct vnode *vp;
1182 	int error;
1183 
1184 	/*
1185 	 * Only root can chroot
1186 	 */
1187 	if ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0)
1188 		return (error);
1189 
1190 	/*
1191 	 * Disallow open directory descriptors (fchdir() breakouts).
1192 	 */
1193 	if (chroot_allow_open_directories == 0 ||
1194 	   (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1195 		if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1196 			return (error);
1197 	}
1198 	if ((vp = ncp->nc_vp) == NULL)
1199 		return (ENOENT);
1200 
1201 	if ((error = vget(vp, LK_SHARED, td)) != 0)
1202 		return (error);
1203 
1204 	/*
1205 	 * Check the validity of vp as a directory to change to and
1206 	 * associate it with rdir/jdir.
1207 	 */
1208 	error = checkvp_chdir(vp, td);
1209 	VOP_UNLOCK(vp, 0, td);	/* leave reference intact */
1210 	if (error == 0) {
1211 		vrele(fdp->fd_rdir);
1212 		fdp->fd_rdir = vp;	/* reference inherited by fd_rdir */
1213 		cache_drop(fdp->fd_nrdir);
1214 		fdp->fd_nrdir = cache_hold(ncp);
1215 		if (fdp->fd_jdir == NULL) {
1216 			fdp->fd_jdir = vp;
1217 			vref(fdp->fd_jdir);
1218 			fdp->fd_njdir = cache_hold(ncp);
1219 		}
1220 	} else {
1221 		vrele(vp);
1222 	}
1223 	return (error);
1224 }
1225 
1226 /*
1227  * chroot_args(char *path)
1228  *
1229  * Change notion of root (``/'') directory.
1230  */
1231 /* ARGSUSED */
1232 int
1233 chroot(struct chroot_args *uap)
1234 {
1235 	struct thread *td = curthread;
1236 	struct nlookupdata nd;
1237 	int error;
1238 
1239 	KKASSERT(td->td_proc);
1240 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1241 	if (error) {
1242 		nlookup_done(&nd);
1243 		return(error);
1244 	}
1245 	error = nlookup(&nd);
1246 	if (error == 0)
1247 		error = kern_chroot(nd.nl_ncp);
1248 	nlookup_done(&nd);
1249 	return(error);
1250 }
1251 
1252 /*
1253  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1254  * determine whether it is legal to chdir to the vnode.  The vnode's state
1255  * is not changed by this call.
1256  */
1257 int
1258 checkvp_chdir(struct vnode *vp, struct thread *td)
1259 {
1260 	int error;
1261 
1262 	if (vp->v_type != VDIR)
1263 		error = ENOTDIR;
1264 	else
1265 		error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred, td);
1266 	return (error);
1267 }
1268 
1269 int
1270 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1271 {
1272 	struct thread *td = curthread;
1273 	struct proc *p = td->td_proc;
1274 	struct lwp *lp = td->td_lwp;
1275 	struct filedesc *fdp = p->p_fd;
1276 	int cmode, flags;
1277 	struct file *nfp;
1278 	struct file *fp;
1279 	struct vnode *vp;
1280 	int type, indx, error;
1281 	struct flock lf;
1282 
1283 	if ((oflags & O_ACCMODE) == O_ACCMODE)
1284 		return (EINVAL);
1285 	flags = FFLAGS(oflags);
1286 	error = falloc(p, &nfp, NULL);
1287 	if (error)
1288 		return (error);
1289 	fp = nfp;
1290 	cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) &~ S_ISTXT;
1291 
1292 	/*
1293 	 * XXX p_dupfd is a real mess.  It allows a device to return a
1294 	 * file descriptor to be duplicated rather then doing the open
1295 	 * itself.
1296 	 */
1297 	lp->lwp_dupfd = -1;
1298 
1299 	/*
1300 	 * Call vn_open() to do the lookup and assign the vnode to the
1301 	 * file pointer.  vn_open() does not change the ref count on fp
1302 	 * and the vnode, on success, will be inherited by the file pointer
1303 	 * and unlocked.
1304 	 */
1305 	nd->nl_flags |= NLC_LOCKVP;
1306 	error = vn_open(nd, fp, flags, cmode);
1307 	nlookup_done(nd);
1308 	if (error) {
1309 		/*
1310 		 * handle special fdopen() case.  bleh.  dupfdopen() is
1311 		 * responsible for dropping the old contents of ofiles[indx]
1312 		 * if it succeeds.
1313 		 *
1314 		 * Note that if fsetfd() succeeds it will add a ref to fp
1315 		 * which represents the fd_files[] assignment.  We must still
1316 		 * drop our reference.
1317 		 */
1318 		if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1319 			if (fsetfd(p, fp, &indx) == 0) {
1320 				error = dupfdopen(fdp, indx, lp->lwp_dupfd, flags, error);
1321 				if (error == 0) {
1322 					*res = indx;
1323 					fdrop(fp, td);	/* our ref */
1324 					return (0);
1325 				}
1326 				if (fdp->fd_files[indx].fp == fp) {
1327 					funsetfd(fdp, indx);
1328 					fdrop(fp, td);	/* fd_files[] ref */
1329 				}
1330 			}
1331 		}
1332 		fdrop(fp, td);	/* our ref */
1333 		if (error == ERESTART)
1334 			error = EINTR;
1335 		return (error);
1336 	}
1337 
1338 	/*
1339 	 * ref the vnode for ourselves so it can't be ripped out from under
1340 	 * is.  XXX need an ND flag to request that the vnode be returned
1341 	 * anyway.
1342 	 */
1343 	vp = (struct vnode *)fp->f_data;
1344 	vref(vp);
1345 	if ((error = fsetfd(p, fp, &indx)) != 0) {
1346 		fdrop(fp, td);
1347 		vrele(vp);
1348 		return (error);
1349 	}
1350 
1351 	/*
1352 	 * If no error occurs the vp will have been assigned to the file
1353 	 * pointer.
1354 	 */
1355 	lp->lwp_dupfd = 0;
1356 
1357 	/*
1358 	 * There should be 2 references on the file, one from the descriptor
1359 	 * table, and one for us.
1360 	 *
1361 	 * Handle the case where someone closed the file (via its file
1362 	 * descriptor) while we were blocked.  The end result should look
1363 	 * like opening the file succeeded but it was immediately closed.
1364 	 */
1365 	if (fp->f_count == 1) {
1366 		KASSERT(fdp->fd_files[indx].fp != fp,
1367 		    ("Open file descriptor lost all refs"));
1368 		vrele(vp);
1369 		fo_close(fp, td);
1370 		fdrop(fp, td);
1371 		*res = indx;
1372 		return 0;
1373 	}
1374 
1375 	if (flags & (O_EXLOCK | O_SHLOCK)) {
1376 		lf.l_whence = SEEK_SET;
1377 		lf.l_start = 0;
1378 		lf.l_len = 0;
1379 		if (flags & O_EXLOCK)
1380 			lf.l_type = F_WRLCK;
1381 		else
1382 			lf.l_type = F_RDLCK;
1383 		type = F_FLOCK;
1384 		if ((flags & FNONBLOCK) == 0)
1385 			type |= F_WAIT;
1386 
1387 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1388 			/*
1389 			 * lock request failed.  Normally close the descriptor
1390 			 * but handle the case where someone might have dup()d
1391 			 * it when we weren't looking.  One reference is
1392 			 * owned by the descriptor array, the other by us.
1393 			 */
1394 			vrele(vp);
1395 			if (fdp->fd_files[indx].fp == fp) {
1396 				funsetfd(fdp, indx);
1397 				fdrop(fp, td);
1398 			}
1399 			fdrop(fp, td);
1400 			return (error);
1401 		}
1402 		fp->f_flag |= FHASLOCK;
1403 	}
1404 	/* assert that vn_open created a backing object if one is needed */
1405 	KASSERT(!vn_canvmio(vp) || VOP_GETVOBJECT(vp, NULL) == 0,
1406 		("open: vmio vnode has no backing object after vn_open"));
1407 
1408 	vrele(vp);
1409 
1410 	/*
1411 	 * release our private reference, leaving the one associated with the
1412 	 * descriptor table intact.
1413 	 */
1414 	fdrop(fp, td);
1415 	*res = indx;
1416 	return (0);
1417 }
1418 
1419 /*
1420  * open_args(char *path, int flags, int mode)
1421  *
1422  * Check permissions, allocate an open file structure,
1423  * and call the device open routine if any.
1424  */
1425 int
1426 open(struct open_args *uap)
1427 {
1428 	struct nlookupdata nd;
1429 	int error;
1430 
1431 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1432 	if (error == 0) {
1433 		error = kern_open(&nd, uap->flags,
1434 				    uap->mode, &uap->sysmsg_result);
1435 	}
1436 	nlookup_done(&nd);
1437 	return (error);
1438 }
1439 
1440 int
1441 kern_mknod(struct nlookupdata *nd, int mode, int dev)
1442 {
1443 	struct namecache *ncp;
1444 	struct thread *td = curthread;
1445 	struct proc *p = td->td_proc;
1446 	struct vnode *vp;
1447 	struct vattr vattr;
1448 	int error;
1449 	int whiteout = 0;
1450 
1451 	KKASSERT(p);
1452 
1453 	switch (mode & S_IFMT) {
1454 	case S_IFCHR:
1455 	case S_IFBLK:
1456 		error = suser(td);
1457 		break;
1458 	default:
1459 		error = suser_cred(p->p_ucred, PRISON_ROOT);
1460 		break;
1461 	}
1462 	if (error)
1463 		return (error);
1464 
1465 	bwillwrite();
1466 	nd->nl_flags |= NLC_CREATE;
1467 	if ((error = nlookup(nd)) != 0)
1468 		return (error);
1469 	ncp = nd->nl_ncp;
1470 	if (ncp->nc_vp)
1471 		return (EEXIST);
1472 
1473 	VATTR_NULL(&vattr);
1474 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1475 	vattr.va_rdev = dev;
1476 	whiteout = 0;
1477 
1478 	switch (mode & S_IFMT) {
1479 	case S_IFMT:	/* used by badsect to flag bad sectors */
1480 		vattr.va_type = VBAD;
1481 		break;
1482 	case S_IFCHR:
1483 		vattr.va_type = VCHR;
1484 		break;
1485 	case S_IFBLK:
1486 		vattr.va_type = VBLK;
1487 		break;
1488 	case S_IFWHT:
1489 		whiteout = 1;
1490 		break;
1491 	default:
1492 		error = EINVAL;
1493 		break;
1494 	}
1495 	if (error == 0) {
1496 		if (whiteout) {
1497 			error = VOP_NWHITEOUT(ncp, nd->nl_cred, NAMEI_CREATE);
1498 		} else {
1499 			vp = NULL;
1500 			error = VOP_NMKNOD(ncp, &vp, nd->nl_cred, &vattr);
1501 			if (error == 0)
1502 				vput(vp);
1503 		}
1504 	}
1505 	return (error);
1506 }
1507 
1508 /*
1509  * mknod_args(char *path, int mode, int dev)
1510  *
1511  * Create a special file.
1512  */
1513 int
1514 mknod(struct mknod_args *uap)
1515 {
1516 	struct nlookupdata nd;
1517 	int error;
1518 
1519 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1520 	if (error == 0)
1521 		error = kern_mknod(&nd, uap->mode, uap->dev);
1522 	nlookup_done(&nd);
1523 	return (error);
1524 }
1525 
1526 int
1527 kern_mkfifo(struct nlookupdata *nd, int mode)
1528 {
1529 	struct namecache *ncp;
1530 	struct thread *td = curthread;
1531 	struct proc *p = td->td_proc;
1532 	struct vattr vattr;
1533 	struct vnode *vp;
1534 	int error;
1535 
1536 	bwillwrite();
1537 
1538 	nd->nl_flags |= NLC_CREATE;
1539 	if ((error = nlookup(nd)) != 0)
1540 		return (error);
1541 	ncp = nd->nl_ncp;
1542 	if (ncp->nc_vp)
1543 		return (EEXIST);
1544 
1545 	VATTR_NULL(&vattr);
1546 	vattr.va_type = VFIFO;
1547 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1548 	vp = NULL;
1549 	error = VOP_NMKNOD(ncp, &vp, nd->nl_cred, &vattr);
1550 	if (error == 0)
1551 		vput(vp);
1552 	return (error);
1553 }
1554 
1555 /*
1556  * mkfifo_args(char *path, int mode)
1557  *
1558  * Create a named pipe.
1559  */
1560 int
1561 mkfifo(struct mkfifo_args *uap)
1562 {
1563 	struct nlookupdata nd;
1564 	int error;
1565 
1566 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1567 	if (error == 0)
1568 		error = kern_mkfifo(&nd, uap->mode);
1569 	nlookup_done(&nd);
1570 	return (error);
1571 }
1572 
1573 static int hardlink_check_uid = 0;
1574 SYSCTL_INT(_kern, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1575     &hardlink_check_uid, 0,
1576     "Unprivileged processes cannot create hard links to files owned by other "
1577     "users");
1578 static int hardlink_check_gid = 0;
1579 SYSCTL_INT(_kern, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
1580     &hardlink_check_gid, 0,
1581     "Unprivileged processes cannot create hard links to files owned by other "
1582     "groups");
1583 
1584 static int
1585 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
1586 {
1587 	struct vattr va;
1588 	int error;
1589 
1590 	/*
1591 	 * Shortcut if disabled
1592 	 */
1593 	if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
1594 		return (0);
1595 
1596 	/*
1597 	 * root cred can always hardlink
1598 	 */
1599 	if (suser_cred(cred, PRISON_ROOT) == 0)
1600 		return (0);
1601 
1602 	/*
1603 	 * Otherwise only if the originating file is owned by the
1604 	 * same user or group.  Note that any group is allowed if
1605 	 * the file is owned by the caller.
1606 	 */
1607 	error = VOP_GETATTR(vp, &va, td);
1608 	if (error != 0)
1609 		return (error);
1610 
1611 	if (hardlink_check_uid) {
1612 		if (cred->cr_uid != va.va_uid)
1613 			return (EPERM);
1614 	}
1615 
1616 	if (hardlink_check_gid) {
1617 		if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
1618 			return (EPERM);
1619 	}
1620 
1621 	return (0);
1622 }
1623 
1624 int
1625 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
1626 {
1627 	struct thread *td = curthread;
1628 	struct vnode *vp;
1629 	int error;
1630 
1631 	/*
1632 	 * Lookup the source and obtained a locked vnode.
1633 	 *
1634 	 * XXX relookup on vget failure / race ?
1635 	 */
1636 	bwillwrite();
1637 	if ((error = nlookup(nd)) != 0)
1638 		return (error);
1639 	vp = nd->nl_ncp->nc_vp;
1640 	KKASSERT(vp != NULL);
1641 	if (vp->v_type == VDIR)
1642 		return (EPERM);		/* POSIX */
1643 	if ((error = vget(vp, LK_EXCLUSIVE, td)) != 0)
1644 		return (error);
1645 
1646 	/*
1647 	 * Unlock the source so we can lookup the target without deadlocking
1648 	 * (XXX vp is locked already, possible other deadlock?).  The target
1649 	 * must not exist.
1650 	 */
1651 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
1652 	nd->nl_flags &= ~NLC_NCPISLOCKED;
1653 	cache_unlock(nd->nl_ncp);
1654 
1655 	linknd->nl_flags |= NLC_CREATE;
1656 	if ((error = nlookup(linknd)) != 0) {
1657 		vput(vp);
1658 		return (error);
1659 	}
1660 	if (linknd->nl_ncp->nc_vp) {
1661 		vput(vp);
1662 		return (EEXIST);
1663 	}
1664 
1665 	/*
1666 	 * Finally run the new API VOP.
1667 	 */
1668 	error = can_hardlink(vp, td, td->td_proc->p_ucred);
1669 	if (error == 0)
1670 		error = VOP_NLINK(linknd->nl_ncp, vp, linknd->nl_cred);
1671 	vput(vp);
1672 	return (error);
1673 }
1674 
1675 /*
1676  * link_args(char *path, char *link)
1677  *
1678  * Make a hard file link.
1679  */
1680 int
1681 link(struct link_args *uap)
1682 {
1683 	struct nlookupdata nd, linknd;
1684 	int error;
1685 
1686 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1687 	if (error == 0) {
1688 		error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
1689 		if (error == 0)
1690 			error = kern_link(&nd, &linknd);
1691 		nlookup_done(&linknd);
1692 	}
1693 	nlookup_done(&nd);
1694 	return (error);
1695 }
1696 
1697 int
1698 kern_symlink(struct nlookupdata *nd, char *path, int mode)
1699 {
1700 	struct namecache *ncp;
1701 	struct vattr vattr;
1702 	struct vnode *vp;
1703 	int error;
1704 
1705 	bwillwrite();
1706 	nd->nl_flags |= NLC_CREATE;
1707 	if ((error = nlookup(nd)) != 0)
1708 		return (error);
1709 	ncp = nd->nl_ncp;
1710 	if (ncp->nc_vp)
1711 		return (EEXIST);
1712 
1713 	VATTR_NULL(&vattr);
1714 	vattr.va_mode = mode;
1715 	error = VOP_NSYMLINK(ncp, &vp, nd->nl_cred, &vattr, path);
1716 	if (error == 0)
1717 		vput(vp);
1718 	return (error);
1719 }
1720 
1721 /*
1722  * symlink(char *path, char *link)
1723  *
1724  * Make a symbolic link.
1725  */
1726 int
1727 symlink(struct symlink_args *uap)
1728 {
1729 	struct thread *td = curthread;
1730 	struct nlookupdata nd;
1731 	char *path;
1732 	int error;
1733 	int mode;
1734 
1735 	path = zalloc(namei_zone);
1736 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
1737 	if (error == 0) {
1738 		error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
1739 		if (error == 0) {
1740 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
1741 			error = kern_symlink(&nd, path, mode);
1742 		}
1743 		nlookup_done(&nd);
1744 	}
1745 	zfree(namei_zone, path);
1746 	return (error);
1747 }
1748 
1749 /*
1750  * undelete_args(char *path)
1751  *
1752  * Delete a whiteout from the filesystem.
1753  */
1754 /* ARGSUSED */
1755 int
1756 undelete(struct undelete_args *uap)
1757 {
1758 	struct nlookupdata nd;
1759 	int error;
1760 
1761 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1762 	bwillwrite();
1763 	nd.nl_flags |= NLC_DELETE;
1764 	if (error == 0)
1765 		error = nlookup(&nd);
1766 	if (error == 0)
1767 		error = VOP_NWHITEOUT(nd.nl_ncp, nd.nl_cred, NAMEI_DELETE);
1768 	nlookup_done(&nd);
1769 	return (error);
1770 }
1771 
1772 int
1773 kern_unlink(struct nlookupdata *nd)
1774 {
1775 	struct namecache *ncp;
1776 	int error;
1777 
1778 	bwillwrite();
1779 	nd->nl_flags |= NLC_DELETE;
1780 	if ((error = nlookup(nd)) != 0)
1781 		return (error);
1782 	ncp = nd->nl_ncp;
1783 	error = VOP_NREMOVE(ncp, nd->nl_cred);
1784 	return (error);
1785 }
1786 
1787 /*
1788  * unlink_args(char *path)
1789  *
1790  * Delete a name from the filesystem.
1791  */
1792 int
1793 unlink(struct unlink_args *uap)
1794 {
1795 	struct nlookupdata nd;
1796 	int error;
1797 
1798 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1799 	if (error == 0)
1800 		error = kern_unlink(&nd);
1801 	nlookup_done(&nd);
1802 	return (error);
1803 }
1804 
1805 int
1806 kern_lseek(int fd, off_t offset, int whence, off_t *res)
1807 {
1808 	struct thread *td = curthread;
1809 	struct proc *p = td->td_proc;
1810 	struct filedesc *fdp = p->p_fd;
1811 	struct file *fp;
1812 	struct vattr vattr;
1813 	int error;
1814 
1815 	if ((u_int)fd >= fdp->fd_nfiles ||
1816 	    (fp = fdp->fd_files[fd].fp) == NULL)
1817 		return (EBADF);
1818 	if (fp->f_type != DTYPE_VNODE)
1819 		return (ESPIPE);
1820 	switch (whence) {
1821 	case L_INCR:
1822 		fp->f_offset += offset;
1823 		break;
1824 	case L_XTND:
1825 		error=VOP_GETATTR((struct vnode *)fp->f_data, &vattr, td);
1826 		if (error)
1827 			return (error);
1828 		fp->f_offset = offset + vattr.va_size;
1829 		break;
1830 	case L_SET:
1831 		fp->f_offset = offset;
1832 		break;
1833 	default:
1834 		return (EINVAL);
1835 	}
1836 	*res = fp->f_offset;
1837 	return (0);
1838 }
1839 
1840 /*
1841  * lseek_args(int fd, int pad, off_t offset, int whence)
1842  *
1843  * Reposition read/write file offset.
1844  */
1845 int
1846 lseek(struct lseek_args *uap)
1847 {
1848 	int error;
1849 
1850 	error = kern_lseek(uap->fd, uap->offset, uap->whence,
1851 	    &uap->sysmsg_offset);
1852 
1853 	return (error);
1854 }
1855 
1856 int
1857 kern_access(struct nlookupdata *nd, int aflags)
1858 {
1859 	struct thread *td = curthread;
1860 	struct vnode *vp;
1861 	int error, flags;
1862 
1863 	if ((error = nlookup(nd)) != 0)
1864 		return (error);
1865 retry:
1866 	error = cache_vget(nd->nl_ncp, nd->nl_cred, LK_EXCLUSIVE, &vp);
1867 	if (error)
1868 		return (error);
1869 
1870 	/* Flags == 0 means only check for existence. */
1871 	if (aflags) {
1872 		flags = 0;
1873 		if (aflags & R_OK)
1874 			flags |= VREAD;
1875 		if (aflags & W_OK)
1876 			flags |= VWRITE;
1877 		if (aflags & X_OK)
1878 			flags |= VEXEC;
1879 		if ((flags & VWRITE) == 0 || (error = vn_writechk(vp)) == 0)
1880 			error = VOP_ACCESS(vp, flags, nd->nl_cred, td);
1881 
1882 		/*
1883 		 * If the file handle is stale we have to re-resolve the
1884 		 * entry.  This is a hack at the moment.
1885 		 */
1886 		if (error == ESTALE) {
1887 			cache_setunresolved(nd->nl_ncp);
1888 			error = cache_resolve(nd->nl_ncp, nd->nl_cred);
1889 			if (error == 0) {
1890 				vput(vp);
1891 				vp = NULL;
1892 				goto retry;
1893 			}
1894 		}
1895 	}
1896 	vput(vp);
1897 	return (error);
1898 }
1899 
1900 /*
1901  * access_args(char *path, int flags)
1902  *
1903  * Check access permissions.
1904  */
1905 int
1906 access(struct access_args *uap)
1907 {
1908 	struct nlookupdata nd;
1909 	int error;
1910 
1911 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1912 	if (error == 0)
1913 		error = kern_access(&nd, uap->flags);
1914 	nlookup_done(&nd);
1915 	return (error);
1916 }
1917 
1918 int
1919 kern_stat(struct nlookupdata *nd, struct stat *st)
1920 {
1921 	int error;
1922 	struct vnode *vp;
1923 	thread_t td;
1924 
1925 	if ((error = nlookup(nd)) != 0)
1926 		return (error);
1927 again:
1928 	if ((vp = nd->nl_ncp->nc_vp) == NULL)
1929 		return (ENOENT);
1930 
1931 	td = curthread;
1932 	if ((error = vget(vp, LK_SHARED, td)) != 0)
1933 		return (error);
1934 	error = vn_stat(vp, st, td);
1935 
1936 	/*
1937 	 * If the file handle is stale we have to re-resolve the entry.  This
1938 	 * is a hack at the moment.
1939 	 */
1940 	if (error == ESTALE) {
1941 		cache_setunresolved(nd->nl_ncp);
1942 		error = cache_resolve(nd->nl_ncp, nd->nl_cred);
1943 		if (error == 0) {
1944 			vput(vp);
1945 			goto again;
1946 		}
1947 	}
1948 	vput(vp);
1949 	return (error);
1950 }
1951 
1952 /*
1953  * stat_args(char *path, struct stat *ub)
1954  *
1955  * Get file status; this version follows links.
1956  */
1957 int
1958 stat(struct stat_args *uap)
1959 {
1960 	struct nlookupdata nd;
1961 	struct stat st;
1962 	int error;
1963 
1964 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1965 	if (error == 0) {
1966 		error = kern_stat(&nd, &st);
1967 		if (error == 0)
1968 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
1969 	}
1970 	nlookup_done(&nd);
1971 	return (error);
1972 }
1973 
1974 /*
1975  * lstat_args(char *path, struct stat *ub)
1976  *
1977  * Get file status; this version does not follow links.
1978  */
1979 int
1980 lstat(struct lstat_args *uap)
1981 {
1982 	struct nlookupdata nd;
1983 	struct stat st;
1984 	int error;
1985 
1986 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1987 	if (error == 0) {
1988 		error = kern_stat(&nd, &st);
1989 		if (error == 0)
1990 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
1991 	}
1992 	nlookup_done(&nd);
1993 	return (error);
1994 }
1995 
1996 /*
1997  * pathconf_Args(char *path, int name)
1998  *
1999  * Get configurable pathname variables.
2000  */
2001 /* ARGSUSED */
2002 int
2003 pathconf(struct pathconf_args *uap)
2004 {
2005 	struct nlookupdata nd;
2006 	struct vnode *vp;
2007 	int error;
2008 
2009 	vp = NULL;
2010 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2011 	if (error == 0)
2012 		error = nlookup(&nd);
2013 	if (error == 0)
2014 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
2015 	nlookup_done(&nd);
2016 	if (error == 0) {
2017 		error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
2018 		vput(vp);
2019 	}
2020 	return (error);
2021 }
2022 
2023 /*
2024  * XXX: daver
2025  * kern_readlink isn't properly split yet.  There is a copyin burried
2026  * in VOP_READLINK().
2027  */
2028 int
2029 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2030 {
2031 	struct thread *td = curthread;
2032 	struct proc *p = td->td_proc;
2033 	struct vnode *vp;
2034 	struct iovec aiov;
2035 	struct uio auio;
2036 	int error;
2037 
2038 	if ((error = nlookup(nd)) != 0)
2039 		return (error);
2040 	error = cache_vget(nd->nl_ncp, nd->nl_cred, LK_EXCLUSIVE, &vp);
2041 	if (error)
2042 		return (error);
2043 	if (vp->v_type != VLNK) {
2044 		error = EINVAL;
2045 	} else {
2046 		aiov.iov_base = buf;
2047 		aiov.iov_len = count;
2048 		auio.uio_iov = &aiov;
2049 		auio.uio_iovcnt = 1;
2050 		auio.uio_offset = 0;
2051 		auio.uio_rw = UIO_READ;
2052 		auio.uio_segflg = UIO_USERSPACE;
2053 		auio.uio_td = td;
2054 		auio.uio_resid = count;
2055 		error = VOP_READLINK(vp, &auio, p->p_ucred);
2056 	}
2057 	vput(vp);
2058 	*res = count - auio.uio_resid;
2059 	return (error);
2060 }
2061 
2062 /*
2063  * readlink_args(char *path, char *buf, int count)
2064  *
2065  * Return target name of a symbolic link.
2066  */
2067 int
2068 readlink(struct readlink_args *uap)
2069 {
2070 	struct nlookupdata nd;
2071 	int error;
2072 
2073 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2074 	if (error == 0) {
2075 		error = kern_readlink(&nd, uap->buf, uap->count,
2076 					&uap->sysmsg_result);
2077 	}
2078 	nlookup_done(&nd);
2079 	return (error);
2080 }
2081 
2082 static int
2083 setfflags(struct vnode *vp, int flags)
2084 {
2085 	struct thread *td = curthread;
2086 	struct proc *p = td->td_proc;
2087 	int error;
2088 	struct vattr vattr;
2089 
2090 	/*
2091 	 * Prevent non-root users from setting flags on devices.  When
2092 	 * a device is reused, users can retain ownership of the device
2093 	 * if they are allowed to set flags and programs assume that
2094 	 * chown can't fail when done as root.
2095 	 */
2096 	if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2097 	    ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0))
2098 		return (error);
2099 
2100 	/*
2101 	 * note: vget is required for any operation that might mod the vnode
2102 	 * so VINACTIVE is properly cleared.
2103 	 */
2104 	VOP_LEASE(vp, td, p->p_ucred, LEASE_WRITE);
2105 	if ((error = vget(vp, LK_EXCLUSIVE, td)) == 0) {
2106 		VATTR_NULL(&vattr);
2107 		vattr.va_flags = flags;
2108 		error = VOP_SETATTR(vp, &vattr, p->p_ucred, td);
2109 		vput(vp);
2110 	}
2111 	return (error);
2112 }
2113 
2114 /*
2115  * chflags(char *path, int flags)
2116  *
2117  * Change flags of a file given a path name.
2118  */
2119 /* ARGSUSED */
2120 int
2121 chflags(struct chflags_args *uap)
2122 {
2123 	struct nlookupdata nd;
2124 	struct vnode *vp;
2125 	int error;
2126 
2127 	vp = NULL;
2128 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2129 	/* XXX Add NLC flag indicating modifying operation? */
2130 	if (error == 0)
2131 		error = nlookup(&nd);
2132 	if (error == 0)
2133 		error = cache_vref(nd.nl_ncp, nd.nl_cred, &vp);
2134 	nlookup_done(&nd);
2135 	if (error == 0) {
2136 		error = setfflags(vp, uap->flags);
2137 		vrele(vp);
2138 	}
2139 	return (error);
2140 }
2141 
2142 /*
2143  * fchflags_args(int fd, int flags)
2144  *
2145  * Change flags of a file given a file descriptor.
2146  */
2147 /* ARGSUSED */
2148 int
2149 fchflags(struct fchflags_args *uap)
2150 {
2151 	struct thread *td = curthread;
2152 	struct proc *p = td->td_proc;
2153 	struct file *fp;
2154 	int error;
2155 
2156 	if ((error = getvnode(p->p_fd, uap->fd, &fp)) != 0)
2157 		return (error);
2158 	return setfflags((struct vnode *) fp->f_data, uap->flags);
2159 }
2160 
2161 static int
2162 setfmode(struct vnode *vp, int mode)
2163 {
2164 	struct thread *td = curthread;
2165 	struct proc *p = td->td_proc;
2166 	int error;
2167 	struct vattr vattr;
2168 
2169 	/*
2170 	 * note: vget is required for any operation that might mod the vnode
2171 	 * so VINACTIVE is properly cleared.
2172 	 */
2173 	VOP_LEASE(vp, td, p->p_ucred, LEASE_WRITE);
2174 	if ((error = vget(vp, LK_EXCLUSIVE, td)) == 0) {
2175 		VATTR_NULL(&vattr);
2176 		vattr.va_mode = mode & ALLPERMS;
2177 		error = VOP_SETATTR(vp, &vattr, p->p_ucred, td);
2178 		vput(vp);
2179 	}
2180 	return error;
2181 }
2182 
2183 int
2184 kern_chmod(struct nlookupdata *nd, int mode)
2185 {
2186 	struct vnode *vp;
2187 	int error;
2188 
2189 	/* XXX Add NLC flag indicating modifying operation? */
2190 	if ((error = nlookup(nd)) != 0)
2191 		return (error);
2192 	if ((error = cache_vref(nd->nl_ncp, nd->nl_cred, &vp)) != 0)
2193 		return (error);
2194 	error = setfmode(vp, mode);
2195 	vrele(vp);
2196 	return (error);
2197 }
2198 
2199 /*
2200  * chmod_args(char *path, int mode)
2201  *
2202  * Change mode of a file given path name.
2203  */
2204 /* ARGSUSED */
2205 int
2206 chmod(struct chmod_args *uap)
2207 {
2208 	struct nlookupdata nd;
2209 	int error;
2210 
2211 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2212 	if (error == 0)
2213 		error = kern_chmod(&nd, uap->mode);
2214 	nlookup_done(&nd);
2215 	return (error);
2216 }
2217 
2218 /*
2219  * lchmod_args(char *path, int mode)
2220  *
2221  * Change mode of a file given path name (don't follow links.)
2222  */
2223 /* ARGSUSED */
2224 int
2225 lchmod(struct lchmod_args *uap)
2226 {
2227 	struct nlookupdata nd;
2228 	int error;
2229 
2230 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2231 	if (error == 0)
2232 		error = kern_chmod(&nd, uap->mode);
2233 	nlookup_done(&nd);
2234 	return (error);
2235 }
2236 
2237 /*
2238  * fchmod_args(int fd, int mode)
2239  *
2240  * Change mode of a file given a file descriptor.
2241  */
2242 /* ARGSUSED */
2243 int
2244 fchmod(struct fchmod_args *uap)
2245 {
2246 	struct thread *td = curthread;
2247 	struct proc *p = td->td_proc;
2248 	struct file *fp;
2249 	int error;
2250 
2251 	if ((error = getvnode(p->p_fd, uap->fd, &fp)) != 0)
2252 		return (error);
2253 	return setfmode((struct vnode *)fp->f_data, uap->mode);
2254 }
2255 
2256 static int
2257 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2258 {
2259 	struct thread *td = curthread;
2260 	struct proc *p = td->td_proc;
2261 	int error;
2262 	struct vattr vattr;
2263 
2264 	/*
2265 	 * note: vget is required for any operation that might mod the vnode
2266 	 * so VINACTIVE is properly cleared.
2267 	 */
2268 	VOP_LEASE(vp, td, p->p_ucred, LEASE_WRITE);
2269 	if ((error = vget(vp, LK_EXCLUSIVE, td)) == 0) {
2270 		VATTR_NULL(&vattr);
2271 		vattr.va_uid = uid;
2272 		vattr.va_gid = gid;
2273 		error = VOP_SETATTR(vp, &vattr, p->p_ucred, td);
2274 		vput(vp);
2275 	}
2276 	return error;
2277 }
2278 
2279 int
2280 kern_chown(struct nlookupdata *nd, int uid, int gid)
2281 {
2282 	struct vnode *vp;
2283 	int error;
2284 
2285 	/* XXX Add NLC flag indicating modifying operation? */
2286 	if ((error = nlookup(nd)) != 0)
2287 		return (error);
2288 	if ((error = cache_vref(nd->nl_ncp, nd->nl_cred, &vp)) != 0)
2289 		return (error);
2290 	error = setfown(vp, uid, gid);
2291 	vrele(vp);
2292 	return (error);
2293 }
2294 
2295 /*
2296  * chown(char *path, int uid, int gid)
2297  *
2298  * Set ownership given a path name.
2299  */
2300 int
2301 chown(struct chown_args *uap)
2302 {
2303 	struct nlookupdata nd;
2304 	int error;
2305 
2306 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2307 	if (error == 0)
2308 		error = kern_chown(&nd, uap->uid, uap->gid);
2309 	nlookup_done(&nd);
2310 	return (error);
2311 }
2312 
2313 /*
2314  * lchown_args(char *path, int uid, int gid)
2315  *
2316  * Set ownership given a path name, do not cross symlinks.
2317  */
2318 int
2319 lchown(struct lchown_args *uap)
2320 {
2321 	struct nlookupdata nd;
2322 	int error;
2323 
2324 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2325 	if (error == 0)
2326 		error = kern_chown(&nd, uap->uid, uap->gid);
2327 	nlookup_done(&nd);
2328 	return (error);
2329 }
2330 
2331 /*
2332  * fchown_args(int fd, int uid, int gid)
2333  *
2334  * Set ownership given a file descriptor.
2335  */
2336 /* ARGSUSED */
2337 int
2338 fchown(struct fchown_args *uap)
2339 {
2340 	struct thread *td = curthread;
2341 	struct proc *p = td->td_proc;
2342 	struct file *fp;
2343 	int error;
2344 
2345 	if ((error = getvnode(p->p_fd, uap->fd, &fp)) != 0)
2346 		return (error);
2347 	return setfown((struct vnode *)fp->f_data,
2348 		uap->uid, uap->gid);
2349 }
2350 
2351 static int
2352 getutimes(const struct timeval *tvp, struct timespec *tsp)
2353 {
2354 	struct timeval tv[2];
2355 
2356 	if (tvp == NULL) {
2357 		microtime(&tv[0]);
2358 		TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
2359 		tsp[1] = tsp[0];
2360 	} else {
2361 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
2362 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
2363 	}
2364 	return 0;
2365 }
2366 
2367 static int
2368 setutimes(struct vnode *vp, const struct timespec *ts, int nullflag)
2369 {
2370 	struct thread *td = curthread;
2371 	struct proc *p = td->td_proc;
2372 	int error;
2373 	struct vattr vattr;
2374 
2375 	/*
2376 	 * note: vget is required for any operation that might mod the vnode
2377 	 * so VINACTIVE is properly cleared.
2378 	 */
2379 	VOP_LEASE(vp, td, p->p_ucred, LEASE_WRITE);
2380 	if ((error = vget(vp, LK_EXCLUSIVE, td)) == 0) {
2381 		VATTR_NULL(&vattr);
2382 		vattr.va_atime = ts[0];
2383 		vattr.va_mtime = ts[1];
2384 		if (nullflag)
2385 			vattr.va_vaflags |= VA_UTIMES_NULL;
2386 		error = VOP_SETATTR(vp, &vattr, p->p_ucred, td);
2387 		vput(vp);
2388 	}
2389 	return error;
2390 }
2391 
2392 int
2393 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
2394 {
2395 	struct timespec ts[2];
2396 	struct vnode *vp;
2397 	int error;
2398 
2399 	if ((error = getutimes(tptr, ts)) != 0)
2400 		return (error);
2401 	/* XXX Add NLC flag indicating modifying operation? */
2402 	if ((error = nlookup(nd)) != 0)
2403 		return (error);
2404 	if ((error = cache_vref(nd->nl_ncp, nd->nl_cred, &vp)) != 0)
2405 		return (error);
2406 	error = setutimes(vp, ts, tptr == NULL);
2407 	vrele(vp);
2408 	return (error);
2409 }
2410 
2411 /*
2412  * utimes_args(char *path, struct timeval *tptr)
2413  *
2414  * Set the access and modification times of a file.
2415  */
2416 int
2417 utimes(struct utimes_args *uap)
2418 {
2419 	struct timeval tv[2];
2420 	struct nlookupdata nd;
2421 	int error;
2422 
2423 	if (uap->tptr) {
2424  		error = copyin(uap->tptr, tv, sizeof(tv));
2425 		if (error)
2426 			return (error);
2427 	}
2428 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2429 	if (error == 0)
2430 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2431 	nlookup_done(&nd);
2432 	return (error);
2433 }
2434 
2435 /*
2436  * lutimes_args(char *path, struct timeval *tptr)
2437  *
2438  * Set the access and modification times of a file.
2439  */
2440 int
2441 lutimes(struct lutimes_args *uap)
2442 {
2443 	struct timeval tv[2];
2444 	struct nlookupdata nd;
2445 	int error;
2446 
2447 	if (uap->tptr) {
2448 		error = copyin(uap->tptr, tv, sizeof(tv));
2449 		if (error)
2450 			return (error);
2451 	}
2452 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2453 	if (error == 0)
2454 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2455 	nlookup_done(&nd);
2456 	return (error);
2457 }
2458 
2459 int
2460 kern_futimes(int fd, struct timeval *tptr)
2461 {
2462 	struct thread *td = curthread;
2463 	struct proc *p = td->td_proc;
2464 	struct timespec ts[2];
2465 	struct file *fp;
2466 	int error;
2467 
2468 	error = getutimes(tptr, ts);
2469 	if (error)
2470 		return (error);
2471 	error = getvnode(p->p_fd, fd, &fp);
2472 	if (error)
2473 		return (error);
2474 	error =  setutimes((struct vnode *)fp->f_data, ts, tptr == NULL);
2475 	return (error);
2476 }
2477 
2478 /*
2479  * futimes_args(int fd, struct timeval *tptr)
2480  *
2481  * Set the access and modification times of a file.
2482  */
2483 int
2484 futimes(struct futimes_args *uap)
2485 {
2486 	struct timeval tv[2];
2487 	int error;
2488 
2489 	if (uap->tptr) {
2490 		error = copyin(uap->tptr, tv, sizeof(tv));
2491 		if (error)
2492 			return (error);
2493 	}
2494 
2495 	error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
2496 
2497 	return (error);
2498 }
2499 
2500 int
2501 kern_truncate(struct nlookupdata *nd, off_t length)
2502 {
2503 	struct vnode *vp;
2504 	struct vattr vattr;
2505 	int error;
2506 
2507 	if (length < 0)
2508 		return(EINVAL);
2509 	/* XXX Add NLC flag indicating modifying operation? */
2510 	if ((error = nlookup(nd)) != 0)
2511 		return (error);
2512 	if ((error = cache_vref(nd->nl_ncp, nd->nl_cred, &vp)) != 0)
2513 		return (error);
2514 	VOP_LEASE(vp, nd->nl_td, nd->nl_cred, LEASE_WRITE);
2515 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, nd->nl_td)) != 0) {
2516 		vrele(vp);
2517 		return (error);
2518 	}
2519 	if (vp->v_type == VDIR) {
2520 		error = EISDIR;
2521 	} else if ((error = vn_writechk(vp)) == 0 &&
2522 	    (error = VOP_ACCESS(vp, VWRITE, nd->nl_cred, nd->nl_td)) == 0) {
2523 		VATTR_NULL(&vattr);
2524 		vattr.va_size = length;
2525 		error = VOP_SETATTR(vp, &vattr, nd->nl_cred, nd->nl_td);
2526 	}
2527 	vput(vp);
2528 	return (error);
2529 }
2530 
2531 /*
2532  * truncate(char *path, int pad, off_t length)
2533  *
2534  * Truncate a file given its path name.
2535  */
2536 int
2537 truncate(struct truncate_args *uap)
2538 {
2539 	struct nlookupdata nd;
2540 	int error;
2541 
2542 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2543 	if (error == 0)
2544 		error = kern_truncate(&nd, uap->length);
2545 	nlookup_done(&nd);
2546 	return error;
2547 }
2548 
2549 int
2550 kern_ftruncate(int fd, off_t length)
2551 {
2552 	struct thread *td = curthread;
2553 	struct proc *p = td->td_proc;
2554 	struct vattr vattr;
2555 	struct vnode *vp;
2556 	struct file *fp;
2557 	int error;
2558 
2559 	if (length < 0)
2560 		return(EINVAL);
2561 	if ((error = getvnode(p->p_fd, fd, &fp)) != 0)
2562 		return (error);
2563 	if ((fp->f_flag & FWRITE) == 0)
2564 		return (EINVAL);
2565 	vp = (struct vnode *)fp->f_data;
2566 	VOP_LEASE(vp, td, p->p_ucred, LEASE_WRITE);
2567 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
2568 	if (vp->v_type == VDIR)
2569 		error = EISDIR;
2570 	else if ((error = vn_writechk(vp)) == 0) {
2571 		VATTR_NULL(&vattr);
2572 		vattr.va_size = length;
2573 		error = VOP_SETATTR(vp, &vattr, fp->f_cred, td);
2574 	}
2575 	VOP_UNLOCK(vp, 0, td);
2576 	return (error);
2577 }
2578 
2579 /*
2580  * ftruncate_args(int fd, int pad, off_t length)
2581  *
2582  * Truncate a file given a file descriptor.
2583  */
2584 int
2585 ftruncate(struct ftruncate_args *uap)
2586 {
2587 	int error;
2588 
2589 	error = kern_ftruncate(uap->fd, uap->length);
2590 
2591 	return (error);
2592 }
2593 
2594 /*
2595  * fsync(int fd)
2596  *
2597  * Sync an open file.
2598  */
2599 /* ARGSUSED */
2600 int
2601 fsync(struct fsync_args *uap)
2602 {
2603 	struct thread *td = curthread;
2604 	struct proc *p = td->td_proc;
2605 	struct vnode *vp;
2606 	struct file *fp;
2607 	vm_object_t obj;
2608 	int error;
2609 
2610 	if ((error = getvnode(p->p_fd, uap->fd, &fp)) != 0)
2611 		return (error);
2612 	vp = (struct vnode *)fp->f_data;
2613 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
2614 	if (VOP_GETVOBJECT(vp, &obj) == 0)
2615 		vm_object_page_clean(obj, 0, 0, 0);
2616 	if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) == 0 &&
2617 	    vp->v_mount && (vp->v_mount->mnt_flag & MNT_SOFTDEP) &&
2618 	    bioops.io_fsync)
2619 		error = (*bioops.io_fsync)(vp);
2620 	VOP_UNLOCK(vp, 0, td);
2621 	return (error);
2622 }
2623 
2624 int
2625 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
2626 {
2627 	struct namecache *fncpd;
2628 	struct namecache *tncpd;
2629 	struct namecache *ncp;
2630 	struct mount *mp;
2631 	int error;
2632 
2633 	bwillwrite();
2634 	if ((error = nlookup(fromnd)) != 0)
2635 		return (error);
2636 	if ((fncpd = fromnd->nl_ncp->nc_parent) == NULL)
2637 		return (ENOENT);
2638 	cache_hold(fncpd);
2639 
2640 	/*
2641 	 * unlock the source ncp so we can lookup the target ncp without
2642 	 * deadlocking.  The target may or may not exist so we do not check
2643 	 * for a target vp like kern_mkdir() and other creation functions do.
2644 	 *
2645 	 * The source and target directories are ref'd and rechecked after
2646 	 * everything is relocked to determine if the source or target file
2647 	 * has been renamed.
2648 	 */
2649 	KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
2650 	fromnd->nl_flags &= ~NLC_NCPISLOCKED;
2651 	cache_unlock(fromnd->nl_ncp);
2652 
2653 	tond->nl_flags |= NLC_CREATE;
2654 	if ((error = nlookup(tond)) != 0) {
2655 		cache_drop(fncpd);
2656 		return (error);
2657 	}
2658 	if ((tncpd = tond->nl_ncp->nc_parent) == NULL) {
2659 		cache_drop(fncpd);
2660 		return (ENOENT);
2661 	}
2662 	cache_hold(tncpd);
2663 
2664 	/*
2665 	 * If the source and target are the same there is nothing to do
2666 	 */
2667 	if (fromnd->nl_ncp == tond->nl_ncp) {
2668 		cache_drop(fncpd);
2669 		cache_drop(tncpd);
2670 		return (0);
2671 	}
2672 
2673 	/*
2674 	 * relock the source ncp.  NOTE AFTER RELOCKING: the source ncp
2675 	 * may have become invalid while it was unlocked, nc_vp and nc_mount
2676 	 * could be NULL.
2677 	 */
2678 	if (cache_lock_nonblock(fromnd->nl_ncp) == 0) {
2679 		cache_resolve(fromnd->nl_ncp, fromnd->nl_cred);
2680 	} else if (fromnd->nl_ncp > tond->nl_ncp) {
2681 		cache_lock(fromnd->nl_ncp);
2682 		cache_resolve(fromnd->nl_ncp, fromnd->nl_cred);
2683 	} else {
2684 		cache_unlock(tond->nl_ncp);
2685 		cache_lock(fromnd->nl_ncp);
2686 		cache_resolve(fromnd->nl_ncp, fromnd->nl_cred);
2687 		cache_lock(tond->nl_ncp);
2688 		cache_resolve(tond->nl_ncp, tond->nl_cred);
2689 	}
2690 	fromnd->nl_flags |= NLC_NCPISLOCKED;
2691 
2692 	/*
2693 	 * make sure the parent directories linkages are the same
2694 	 */
2695 	if (fncpd != fromnd->nl_ncp->nc_parent ||
2696 	    tncpd != tond->nl_ncp->nc_parent) {
2697 		cache_drop(fncpd);
2698 		cache_drop(tncpd);
2699 		return (ENOENT);
2700 	}
2701 
2702 	/*
2703 	 * Both the source and target must be within the same filesystem and
2704 	 * in the same filesystem as their parent directories within the
2705 	 * namecache topology.
2706 	 *
2707 	 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
2708 	 */
2709 	mp = fncpd->nc_mount;
2710 	if (mp != tncpd->nc_mount || mp != fromnd->nl_ncp->nc_mount ||
2711 	    mp != tond->nl_ncp->nc_mount) {
2712 		cache_drop(fncpd);
2713 		cache_drop(tncpd);
2714 		return (EXDEV);
2715 	}
2716 
2717 	/*
2718 	 * If the target exists and either the source or target is a directory,
2719 	 * then both must be directories.
2720 	 *
2721 	 * Due to relocking of the source, fromnd->nl_ncp->nc_vp might have
2722 	 * become NULL.
2723 	 */
2724 	if (tond->nl_ncp->nc_vp) {
2725 		if (fromnd->nl_ncp->nc_vp == NULL) {
2726 			error = ENOENT;
2727 		} else if (fromnd->nl_ncp->nc_vp->v_type == VDIR) {
2728 			if (tond->nl_ncp->nc_vp->v_type != VDIR)
2729 				error = ENOTDIR;
2730 		} else if (tond->nl_ncp->nc_vp->v_type == VDIR) {
2731 			error = EISDIR;
2732 		}
2733 	}
2734 
2735 	/*
2736 	 * You cannot rename a source into itself or a subdirectory of itself.
2737 	 * We check this by travsersing the target directory upwards looking
2738 	 * for a match against the source.
2739 	 */
2740 	if (error == 0) {
2741 		for (ncp = tncpd; ncp; ncp = ncp->nc_parent) {
2742 			if (fromnd->nl_ncp == ncp) {
2743 				error = EINVAL;
2744 				break;
2745 			}
2746 		}
2747 	}
2748 
2749 	cache_drop(fncpd);
2750 	cache_drop(tncpd);
2751 
2752 	/*
2753 	 * Even though the namespaces are different, they may still represent
2754 	 * hardlinks to the same file.  The filesystem might have a hard time
2755 	 * with this so we issue a NREMOVE of the source instead of a NRENAME
2756 	 * when we detect the situation.
2757 	 */
2758 	if (error == 0) {
2759 		if (fromnd->nl_ncp->nc_vp == tond->nl_ncp->nc_vp) {
2760 			error = VOP_NREMOVE(fromnd->nl_ncp, fromnd->nl_cred);
2761 		} else {
2762 			error = VOP_NRENAME(fromnd->nl_ncp, tond->nl_ncp,
2763 					    tond->nl_cred);
2764 		}
2765 	}
2766 	return (error);
2767 }
2768 
2769 /*
2770  * rename_args(char *from, char *to)
2771  *
2772  * Rename files.  Source and destination must either both be directories,
2773  * or both not be directories.  If target is a directory, it must be empty.
2774  */
2775 int
2776 rename(struct rename_args *uap)
2777 {
2778 	struct nlookupdata fromnd, tond;
2779 	int error;
2780 
2781 	error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
2782 	if (error == 0) {
2783 		error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
2784 		if (error == 0)
2785 			error = kern_rename(&fromnd, &tond);
2786 		nlookup_done(&tond);
2787 	}
2788 	nlookup_done(&fromnd);
2789 	return (error);
2790 }
2791 
2792 int
2793 kern_mkdir(struct nlookupdata *nd, int mode)
2794 {
2795 	struct thread *td = curthread;
2796 	struct proc *p = td->td_proc;
2797 	struct namecache *ncp;
2798 	struct vnode *vp;
2799 	struct vattr vattr;
2800 	int error;
2801 
2802 	bwillwrite();
2803 	nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE;
2804 	if ((error = nlookup(nd)) != 0)
2805 		return (error);
2806 
2807 	ncp = nd->nl_ncp;
2808 	if (ncp->nc_vp)
2809 		return (EEXIST);
2810 
2811 	VATTR_NULL(&vattr);
2812 	vattr.va_type = VDIR;
2813 	vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
2814 
2815 	vp = NULL;
2816 	error = VOP_NMKDIR(ncp, &vp, p->p_ucred, &vattr);
2817 	if (error == 0)
2818 		vput(vp);
2819 	return (error);
2820 }
2821 
2822 /*
2823  * mkdir_args(char *path, int mode)
2824  *
2825  * Make a directory file.
2826  */
2827 /* ARGSUSED */
2828 int
2829 mkdir(struct mkdir_args *uap)
2830 {
2831 	struct nlookupdata nd;
2832 	int error;
2833 
2834 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2835 	if (error == 0)
2836 		error = kern_mkdir(&nd, uap->mode);
2837 	nlookup_done(&nd);
2838 	return (error);
2839 }
2840 
2841 int
2842 kern_rmdir(struct nlookupdata *nd)
2843 {
2844 	struct namecache *ncp;
2845 	int error;
2846 
2847 	bwillwrite();
2848 	nd->nl_flags |= NLC_DELETE;
2849 	if ((error = nlookup(nd)) != 0)
2850 		return (error);
2851 
2852 	ncp = nd->nl_ncp;
2853 	error = VOP_NRMDIR(ncp, nd->nl_cred);
2854 	return (error);
2855 }
2856 
2857 /*
2858  * rmdir_args(char *path)
2859  *
2860  * Remove a directory file.
2861  */
2862 /* ARGSUSED */
2863 int
2864 rmdir(struct rmdir_args *uap)
2865 {
2866 	struct nlookupdata nd;
2867 	int error;
2868 
2869 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2870 	if (error == 0)
2871 		error = kern_rmdir(&nd);
2872 	nlookup_done(&nd);
2873 	return (error);
2874 }
2875 
2876 int
2877 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
2878     enum uio_seg direction)
2879 {
2880 	struct thread *td = curthread;
2881 	struct proc *p = td->td_proc;
2882 	struct vnode *vp;
2883 	struct file *fp;
2884 	struct uio auio;
2885 	struct iovec aiov;
2886 	long loff;
2887 	int error, eofflag;
2888 
2889 	if ((error = getvnode(p->p_fd, fd, &fp)) != 0)
2890 		return (error);
2891 	if ((fp->f_flag & FREAD) == 0)
2892 		return (EBADF);
2893 	vp = (struct vnode *)fp->f_data;
2894 unionread:
2895 	if (vp->v_type != VDIR)
2896 		return (EINVAL);
2897 	aiov.iov_base = buf;
2898 	aiov.iov_len = count;
2899 	auio.uio_iov = &aiov;
2900 	auio.uio_iovcnt = 1;
2901 	auio.uio_rw = UIO_READ;
2902 	auio.uio_segflg = direction;
2903 	auio.uio_td = td;
2904 	auio.uio_resid = count;
2905 	/* vn_lock(vp, LK_SHARED | LK_RETRY, td); */
2906 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
2907 	loff = auio.uio_offset = fp->f_offset;
2908 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
2909 	fp->f_offset = auio.uio_offset;
2910 	VOP_UNLOCK(vp, 0, td);
2911 	if (error)
2912 		return (error);
2913 	if (count == auio.uio_resid) {
2914 		if (union_dircheckp) {
2915 			error = union_dircheckp(td, &vp, fp);
2916 			if (error == -1)
2917 				goto unionread;
2918 			if (error)
2919 				return (error);
2920 		}
2921 		if ((vp->v_flag & VROOT) &&
2922 		    (vp->v_mount->mnt_flag & MNT_UNION)) {
2923 			struct vnode *tvp = vp;
2924 			vp = vp->v_mount->mnt_vnodecovered;
2925 			vref(vp);
2926 			fp->f_data = vp;
2927 			fp->f_offset = 0;
2928 			vrele(tvp);
2929 			goto unionread;
2930 		}
2931 	}
2932 	if (basep) {
2933 		*basep = loff;
2934 	}
2935 	*res = count - auio.uio_resid;
2936 	return (error);
2937 }
2938 
2939 /*
2940  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
2941  *
2942  * Read a block of directory entries in a file system independent format.
2943  */
2944 int
2945 getdirentries(struct getdirentries_args *uap)
2946 {
2947 	long base;
2948 	int error;
2949 
2950 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
2951 	    &uap->sysmsg_result, UIO_USERSPACE);
2952 
2953 	if (error == 0)
2954 		error = copyout(&base, uap->basep, sizeof(*uap->basep));
2955 	return (error);
2956 }
2957 
2958 /*
2959  * getdents_args(int fd, char *buf, size_t count)
2960  */
2961 int
2962 getdents(struct getdents_args *uap)
2963 {
2964 	int error;
2965 
2966 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
2967 	    &uap->sysmsg_result, UIO_USERSPACE);
2968 
2969 	return (error);
2970 }
2971 
2972 /*
2973  * umask(int newmask)
2974  *
2975  * Set the mode mask for creation of filesystem nodes.
2976  *
2977  * MP SAFE
2978  */
2979 int
2980 umask(struct umask_args *uap)
2981 {
2982 	struct thread *td = curthread;
2983 	struct proc *p = td->td_proc;
2984 	struct filedesc *fdp;
2985 
2986 	fdp = p->p_fd;
2987 	uap->sysmsg_result = fdp->fd_cmask;
2988 	fdp->fd_cmask = uap->newmask & ALLPERMS;
2989 	return (0);
2990 }
2991 
2992 /*
2993  * revoke(char *path)
2994  *
2995  * Void all references to file by ripping underlying filesystem
2996  * away from vnode.
2997  */
2998 /* ARGSUSED */
2999 int
3000 revoke(struct revoke_args *uap)
3001 {
3002 	struct thread *td = curthread;
3003 	struct nlookupdata nd;
3004 	struct vattr vattr;
3005 	struct vnode *vp;
3006 	struct ucred *cred;
3007 	int error;
3008 
3009 	vp = NULL;
3010 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3011 	if (error == 0)
3012 		error = nlookup(&nd);
3013 	if (error == 0)
3014 		error = cache_vref(nd.nl_ncp, nd.nl_cred, &vp);
3015 	cred = crhold(nd.nl_cred);
3016 	nlookup_done(&nd);
3017 	if (error == 0) {
3018 		if (vp->v_type != VCHR && vp->v_type != VBLK)
3019 			error = EINVAL;
3020 		if (error == 0)
3021 			error = VOP_GETATTR(vp, &vattr, td);
3022 		if (error == 0 && cred->cr_uid != vattr.va_uid)
3023 			error = suser_cred(cred, PRISON_ROOT);
3024 		if (error == 0 && count_udev(vp->v_udev) > 0) {
3025 			if ((error = vx_lock(vp)) == 0) {
3026 				VOP_REVOKE(vp, REVOKEALL);
3027 				vx_unlock(vp);
3028 			}
3029 		}
3030 		vrele(vp);
3031 	}
3032 	crfree(cred);
3033 	return (error);
3034 }
3035 
3036 /*
3037  * Convert a user file descriptor to a kernel file entry.
3038  */
3039 int
3040 getvnode(struct filedesc *fdp, int fd, struct file **fpp)
3041 {
3042 	struct file *fp;
3043 
3044 	if ((u_int)fd >= fdp->fd_nfiles ||
3045 	    (fp = fdp->fd_files[fd].fp) == NULL)
3046 		return (EBADF);
3047 	if (fp->f_type != DTYPE_VNODE && fp->f_type != DTYPE_FIFO)
3048 		return (EINVAL);
3049 	*fpp = fp;
3050 	return (0);
3051 }
3052 /*
3053  * getfh_args(char *fname, fhandle_t *fhp)
3054  *
3055  * Get (NFS) file handle
3056  */
3057 int
3058 getfh(struct getfh_args *uap)
3059 {
3060 	struct thread *td = curthread;
3061 	struct nlookupdata nd;
3062 	fhandle_t fh;
3063 	struct vnode *vp;
3064 	int error;
3065 
3066 	/*
3067 	 * Must be super user
3068 	 */
3069 	if ((error = suser(td)) != 0)
3070 		return (error);
3071 
3072 	vp = NULL;
3073 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3074 	if (error == 0)
3075 		error = nlookup(&nd);
3076 	if (error == 0)
3077 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
3078 	nlookup_done(&nd);
3079 	if (error == 0) {
3080 		bzero(&fh, sizeof(fh));
3081 		fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
3082 		error = VFS_VPTOFH(vp, &fh.fh_fid);
3083 		vput(vp);
3084 		if (error == 0)
3085 			error = copyout(&fh, uap->fhp, sizeof(fh));
3086 	}
3087 	return (error);
3088 }
3089 
3090 /*
3091  * fhopen_args(const struct fhandle *u_fhp, int flags)
3092  *
3093  * syscall for the rpc.lockd to use to translate a NFS file handle into
3094  * an open descriptor.
3095  *
3096  * warning: do not remove the suser() call or this becomes one giant
3097  * security hole.
3098  */
3099 int
3100 fhopen(struct fhopen_args *uap)
3101 {
3102 	struct thread *td = curthread;
3103 	struct proc *p = td->td_proc;
3104 	struct mount *mp;
3105 	struct vnode *vp;
3106 	struct fhandle fhp;
3107 	struct vattr vat;
3108 	struct vattr *vap = &vat;
3109 	struct flock lf;
3110 	struct filedesc *fdp = p->p_fd;
3111 	int fmode, mode, error, type;
3112 	struct file *nfp;
3113 	struct file *fp;
3114 	int indx;
3115 
3116 	/*
3117 	 * Must be super user
3118 	 */
3119 	error = suser(td);
3120 	if (error)
3121 		return (error);
3122 
3123 	fmode = FFLAGS(uap->flags);
3124 	/* why not allow a non-read/write open for our lockd? */
3125 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3126 		return (EINVAL);
3127 	error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3128 	if (error)
3129 		return(error);
3130 	/* find the mount point */
3131 	mp = vfs_getvfs(&fhp.fh_fsid);
3132 	if (mp == NULL)
3133 		return (ESTALE);
3134 	/* now give me my vnode, it gets returned to me locked */
3135 	error = VFS_FHTOVP(mp, &fhp.fh_fid, &vp);
3136 	if (error)
3137 		return (error);
3138  	/*
3139 	 * from now on we have to make sure not
3140 	 * to forget about the vnode
3141 	 * any error that causes an abort must vput(vp)
3142 	 * just set error = err and 'goto bad;'.
3143 	 */
3144 
3145 	/*
3146 	 * from vn_open
3147 	 */
3148 	if (vp->v_type == VLNK) {
3149 		error = EMLINK;
3150 		goto bad;
3151 	}
3152 	if (vp->v_type == VSOCK) {
3153 		error = EOPNOTSUPP;
3154 		goto bad;
3155 	}
3156 	mode = 0;
3157 	if (fmode & (FWRITE | O_TRUNC)) {
3158 		if (vp->v_type == VDIR) {
3159 			error = EISDIR;
3160 			goto bad;
3161 		}
3162 		error = vn_writechk(vp);
3163 		if (error)
3164 			goto bad;
3165 		mode |= VWRITE;
3166 	}
3167 	if (fmode & FREAD)
3168 		mode |= VREAD;
3169 	if (mode) {
3170 		error = VOP_ACCESS(vp, mode, p->p_ucred, td);
3171 		if (error)
3172 			goto bad;
3173 	}
3174 	if (fmode & O_TRUNC) {
3175 		VOP_UNLOCK(vp, 0, td);			/* XXX */
3176 		VOP_LEASE(vp, td, p->p_ucred, LEASE_WRITE);
3177 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);	/* XXX */
3178 		VATTR_NULL(vap);
3179 		vap->va_size = 0;
3180 		error = VOP_SETATTR(vp, vap, p->p_ucred, td);
3181 		if (error)
3182 			goto bad;
3183 	}
3184 
3185 	/*
3186 	 * VOP_OPEN needs the file pointer so it can potentially override
3187 	 * it.
3188 	 *
3189 	 * WARNING! no f_ncp will be associated when fhopen()ing a directory.
3190 	 * XXX
3191 	 */
3192 	if ((error = falloc(p, &nfp, NULL)) != 0)
3193 		goto bad;
3194 	fp = nfp;
3195 
3196 	fp->f_type = DTYPE_VNODE;
3197 	fp->f_flag = fmode & FMASK;
3198 	fp->f_ops = &vnode_fileops;
3199 	fp->f_data = vp;
3200 
3201 	error = VOP_OPEN(vp, fmode, p->p_ucred, fp, td);
3202 	if (error) {
3203 		/*
3204 		 * setting f_ops this way prevents VOP_CLOSE from being
3205 		 * called or fdrop() releasing the vp from v_data.   Since
3206 		 * the VOP_OPEN failed we don't want to VOP_CLOSE.
3207 		 */
3208 		fp->f_ops = &badfileops;
3209 		fp->f_data = NULL;
3210 		fdrop(fp, td);
3211 		goto bad;
3212 	}
3213 	if (fmode & FWRITE)
3214 		vp->v_writecount++;
3215 
3216 	/*
3217 	 * The fp now owns a reference on the vnode.  We still have our own
3218 	 * ref+lock.
3219 	 */
3220 	vref(vp);
3221 
3222 	/*
3223 	 * Make sure that a VM object is created for VMIO support.  If this
3224 	 * fails just fdrop() normally to clean up.
3225 	 */
3226 	if (vn_canvmio(vp) == TRUE) {
3227 		if ((error = vfs_object_create(vp, td)) != 0) {
3228 			fdrop(fp, td);
3229 			goto bad;
3230 		}
3231 	}
3232 
3233 	/*
3234 	 * The open was successful, associate it with a file descriptor.
3235 	 */
3236 	if ((error = fsetfd(p, fp, &indx)) != 0) {
3237 		if (fmode & FWRITE)
3238 			vp->v_writecount--;
3239 		fdrop(fp, td);
3240 		goto bad;
3241 	}
3242 
3243 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
3244 		lf.l_whence = SEEK_SET;
3245 		lf.l_start = 0;
3246 		lf.l_len = 0;
3247 		if (fmode & O_EXLOCK)
3248 			lf.l_type = F_WRLCK;
3249 		else
3250 			lf.l_type = F_RDLCK;
3251 		type = F_FLOCK;
3252 		if ((fmode & FNONBLOCK) == 0)
3253 			type |= F_WAIT;
3254 		VOP_UNLOCK(vp, 0, td);
3255 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3256 			/*
3257 			 * lock request failed.  Normally close the descriptor
3258 			 * but handle the case where someone might have dup()d
3259 			 * or close()d it when we weren't looking.
3260 			 */
3261 			if (fdp->fd_files[indx].fp == fp) {
3262 				funsetfd(fdp, indx);
3263 				fdrop(fp, td);
3264 			}
3265 
3266 			/*
3267 			 * release our private reference.
3268 			 */
3269 			fdrop(fp, td);
3270 			vrele(vp);
3271 			return (error);
3272 		}
3273 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
3274 		fp->f_flag |= FHASLOCK;
3275 	}
3276 	if ((vp->v_type == VREG) && (VOP_GETVOBJECT(vp, NULL) != 0))
3277 		vfs_object_create(vp, td);
3278 
3279 	vput(vp);
3280 	fdrop(fp, td);
3281 	uap->sysmsg_result = indx;
3282 	return (0);
3283 
3284 bad:
3285 	vput(vp);
3286 	return (error);
3287 }
3288 
3289 /*
3290  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3291  */
3292 int
3293 fhstat(struct fhstat_args *uap)
3294 {
3295 	struct thread *td = curthread;
3296 	struct stat sb;
3297 	fhandle_t fh;
3298 	struct mount *mp;
3299 	struct vnode *vp;
3300 	int error;
3301 
3302 	/*
3303 	 * Must be super user
3304 	 */
3305 	error = suser(td);
3306 	if (error)
3307 		return (error);
3308 
3309 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
3310 	if (error)
3311 		return (error);
3312 
3313 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3314 		return (ESTALE);
3315 	if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3316 		return (error);
3317 	error = vn_stat(vp, &sb, td);
3318 	vput(vp);
3319 	if (error)
3320 		return (error);
3321 	error = copyout(&sb, uap->sb, sizeof(sb));
3322 	return (error);
3323 }
3324 
3325 /*
3326  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
3327  */
3328 int
3329 fhstatfs(struct fhstatfs_args *uap)
3330 {
3331 	struct thread *td = curthread;
3332 	struct proc *p = td->td_proc;
3333 	struct statfs *sp;
3334 	struct mount *mp;
3335 	struct vnode *vp;
3336 	struct statfs sb;
3337 	char *fullpath, *freepath;
3338 	fhandle_t fh;
3339 	int error;
3340 
3341 	/*
3342 	 * Must be super user
3343 	 */
3344 	if ((error = suser(td)))
3345 		return (error);
3346 
3347 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3348 		return (error);
3349 
3350 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3351 		return (ESTALE);
3352 
3353 	if (p != NULL && (p->p_fd->fd_nrdir->nc_flag & NCF_ROOT) == 0 &&
3354 	    !chroot_visible_mnt(mp, p))
3355 		return (ESTALE);
3356 
3357 	if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3358 		return (error);
3359 	mp = vp->v_mount;
3360 	sp = &mp->mnt_stat;
3361 	vput(vp);
3362 	if ((error = VFS_STATFS(mp, sp, td)) != 0)
3363 		return (error);
3364 
3365 	error = cache_fullpath(p, mp->mnt_ncp, &fullpath, &freepath);
3366 	if (error)
3367 		return(error);
3368 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3369 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
3370 	free(freepath, M_TEMP);
3371 
3372 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
3373 	if (suser(td)) {
3374 		bcopy(sp, &sb, sizeof(sb));
3375 		sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
3376 		sp = &sb;
3377 	}
3378 	return (copyout(sp, uap->buf, sizeof(*sp)));
3379 }
3380 
3381 /*
3382  * Syscall to push extended attribute configuration information into the
3383  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
3384  * a command (int cmd), and attribute name and misc data.  For now, the
3385  * attribute name is left in userspace for consumption by the VFS_op.
3386  * It will probably be changed to be copied into sysspace by the
3387  * syscall in the future, once issues with various consumers of the
3388  * attribute code have raised their hands.
3389  *
3390  * Currently this is used only by UFS Extended Attributes.
3391  */
3392 int
3393 extattrctl(struct extattrctl_args *uap)
3394 {
3395 	struct nlookupdata nd;
3396 	struct mount *mp;
3397 	struct vnode *vp;
3398 	int error;
3399 
3400 	vp = NULL;
3401 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3402 	if (error == 0)
3403 		error = nlookup(&nd);
3404 	if (error == 0) {
3405 		mp = nd.nl_ncp->nc_mount;
3406 		error = VFS_EXTATTRCTL(mp, uap->cmd,
3407 				uap->attrname, uap->arg,
3408 				nd.nl_td);
3409 	}
3410 	nlookup_done(&nd);
3411 	return (error);
3412 }
3413 
3414 /*
3415  * Syscall to set a named extended attribute on a file or directory.
3416  * Accepts attribute name, and a uio structure pointing to the data to set.
3417  * The uio is consumed in the style of writev().  The real work happens
3418  * in VOP_SETEXTATTR().
3419  */
3420 int
3421 extattr_set_file(struct extattr_set_file_args *uap)
3422 {
3423 	char attrname[EXTATTR_MAXNAMELEN];
3424 	struct iovec aiov[UIO_SMALLIOV];
3425 	struct iovec *needfree;
3426 	struct nlookupdata nd;
3427 	struct iovec *iov;
3428 	struct vnode *vp;
3429 	struct uio auio;
3430 	u_int iovlen;
3431 	u_int cnt;
3432 	int error;
3433 	int i;
3434 
3435 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3436 	if (error)
3437 		return (error);
3438 
3439 	vp = NULL;
3440 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3441 	if (error == 0)
3442 		error = nlookup(&nd);
3443 	if (error == 0)
3444 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
3445 	if (error) {
3446 		nlookup_done(&nd);
3447 		return (error);
3448 	}
3449 
3450 	needfree = NULL;
3451 	iovlen = uap->iovcnt * sizeof(struct iovec);
3452 	if (uap->iovcnt > UIO_SMALLIOV) {
3453 		if (uap->iovcnt > UIO_MAXIOV) {
3454 			error = EINVAL;
3455 			goto done;
3456 		}
3457 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3458 		needfree = iov;
3459 	} else {
3460 		iov = aiov;
3461 	}
3462 	auio.uio_iov = iov;
3463 	auio.uio_iovcnt = uap->iovcnt;
3464 	auio.uio_rw = UIO_WRITE;
3465 	auio.uio_segflg = UIO_USERSPACE;
3466 	auio.uio_td = nd.nl_td;
3467 	auio.uio_offset = 0;
3468 	if ((error = copyin(uap->iovp, iov, iovlen)))
3469 		goto done;
3470 	auio.uio_resid = 0;
3471 	for (i = 0; i < uap->iovcnt; i++) {
3472 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
3473 			error = EINVAL;
3474 			goto done;
3475 		}
3476 		auio.uio_resid += iov->iov_len;
3477 		iov++;
3478 	}
3479 	cnt = auio.uio_resid;
3480 	error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred, nd.nl_td);
3481 	cnt -= auio.uio_resid;
3482 	uap->sysmsg_result = cnt;
3483 done:
3484 	vput(vp);
3485 	nlookup_done(&nd);
3486 	if (needfree)
3487 		FREE(needfree, M_IOV);
3488 	return (error);
3489 }
3490 
3491 /*
3492  * Syscall to get a named extended attribute on a file or directory.
3493  * Accepts attribute name, and a uio structure pointing to a buffer for the
3494  * data.  The uio is consumed in the style of readv().  The real work
3495  * happens in VOP_GETEXTATTR();
3496  */
3497 int
3498 extattr_get_file(struct extattr_get_file_args *uap)
3499 {
3500 	char attrname[EXTATTR_MAXNAMELEN];
3501 	struct iovec aiov[UIO_SMALLIOV];
3502 	struct iovec *needfree;
3503 	struct nlookupdata nd;
3504 	struct iovec *iov;
3505 	struct vnode *vp;
3506 	struct uio auio;
3507 	u_int iovlen;
3508 	u_int cnt;
3509 	int error;
3510 	int i;
3511 
3512 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3513 	if (error)
3514 		return (error);
3515 
3516 	vp = NULL;
3517 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3518 	if (error == 0)
3519 		error = nlookup(&nd);
3520 	if (error == 0)
3521 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
3522 	if (error) {
3523 		nlookup_done(&nd);
3524 		return (error);
3525 	}
3526 
3527 	iovlen = uap->iovcnt * sizeof (struct iovec);
3528 	needfree = NULL;
3529 	if (uap->iovcnt > UIO_SMALLIOV) {
3530 		if (uap->iovcnt > UIO_MAXIOV) {
3531 			error = EINVAL;
3532 			goto done;
3533 		}
3534 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3535 		needfree = iov;
3536 	} else {
3537 		iov = aiov;
3538 	}
3539 	auio.uio_iov = iov;
3540 	auio.uio_iovcnt = uap->iovcnt;
3541 	auio.uio_rw = UIO_READ;
3542 	auio.uio_segflg = UIO_USERSPACE;
3543 	auio.uio_td = nd.nl_td;
3544 	auio.uio_offset = 0;
3545 	if ((error = copyin(uap->iovp, iov, iovlen)))
3546 		goto done;
3547 	auio.uio_resid = 0;
3548 	for (i = 0; i < uap->iovcnt; i++) {
3549 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
3550 			error = EINVAL;
3551 			goto done;
3552 		}
3553 		auio.uio_resid += iov->iov_len;
3554 		iov++;
3555 	}
3556 	cnt = auio.uio_resid;
3557 	error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred, nd.nl_td);
3558 	cnt -= auio.uio_resid;
3559 	uap->sysmsg_result = cnt;
3560 done:
3561 	vput(vp);
3562 	nlookup_done(&nd);
3563 	if (needfree)
3564 		FREE(needfree, M_IOV);
3565 	return(error);
3566 }
3567 
3568 /*
3569  * Syscall to delete a named extended attribute from a file or directory.
3570  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
3571  */
3572 int
3573 extattr_delete_file(struct extattr_delete_file_args *uap)
3574 {
3575 	char attrname[EXTATTR_MAXNAMELEN];
3576 	struct nlookupdata nd;
3577 	struct vnode *vp;
3578 	int error;
3579 
3580 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3581 	if (error)
3582 		return(error);
3583 
3584 	vp = NULL;
3585 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3586 	if (error == 0)
3587 		error = nlookup(&nd);
3588 	if (error == 0)
3589 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
3590 	if (error) {
3591 		nlookup_done(&nd);
3592 		return (error);
3593 	}
3594 
3595 	error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred, nd.nl_td);
3596 	vput(vp);
3597 	nlookup_done(&nd);
3598 	return(error);
3599 }
3600 
3601 static int
3602 chroot_visible_mnt(struct mount *mp, struct proc *p)
3603 {
3604 	struct namecache *ncp;
3605 	/*
3606 	 * First check if this file system is below
3607 	 * the chroot path.
3608 	 */
3609 	ncp = mp->mnt_ncp;
3610 	while (ncp != NULL && ncp != p->p_fd->fd_nrdir)
3611 		ncp = ncp->nc_parent;
3612 	if (ncp == NULL) {
3613 		/*
3614 		 * This is not below the chroot path.
3615 		 *
3616 		 * Check if the chroot path is on the same filesystem,
3617 		 * by determing if we have to cross a mount point
3618 		 * before reaching mp->mnt_ncp.
3619 		 */
3620 		ncp = p->p_fd->fd_nrdir;
3621 		while (ncp != NULL && ncp != mp->mnt_ncp) {
3622 			if (ncp->nc_flag & NCF_MOUNTPT) {
3623 				ncp = NULL;
3624 				break;
3625 			}
3626 			ncp = ncp->nc_parent;
3627 		}
3628 	}
3629 	return(ncp != NULL);
3630 }
3631