xref: /dflybsd-src/sys/kern/vfs_syscalls.c (revision 1f7ab7c9fc18f47a2f16dc45b13dee254c603ce7)
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 1.106 2006/09/19 18:17:46 dillon 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/spinlock.h>
66 #include <sys/kern_syscall.h>
67 #include <sys/objcache.h>
68 
69 #include <machine/limits.h>
70 #include <vfs/union/union.h>
71 #include <sys/sysctl.h>
72 #include <vm/vm.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_page.h>
75 
76 #include <sys/file2.h>
77 #include <sys/spinlock2.h>
78 
79 static int checkvp_chdir (struct vnode *vn, struct thread *td);
80 static void checkdirs (struct vnode *olddp, struct namecache *ncp);
81 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
82 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
83 static int getutimes (const struct timeval *, struct timespec *);
84 static int setfown (struct vnode *, uid_t, gid_t);
85 static int setfmode (struct vnode *, int);
86 static int setfflags (struct vnode *, int);
87 static int setutimes (struct vnode *, const struct timespec *, int);
88 static int	usermount = 0;	/* if 1, non-root can mount fs. */
89 
90 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *);
91 
92 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "");
93 
94 /*
95  * Virtual File System System Calls
96  */
97 
98 /*
99  * Mount a file system.
100  */
101 /*
102  * mount_args(char *type, char *path, int flags, caddr_t data)
103  */
104 /* ARGSUSED */
105 int
106 sys_mount(struct mount_args *uap)
107 {
108 	struct thread *td = curthread;
109 	struct proc *p = td->td_proc;
110 	struct vnode *vp;
111 	struct namecache *ncp;
112 	struct mount *mp;
113 	struct vfsconf *vfsp;
114 	int error, flag = 0, flag2 = 0;
115 	int hasmount;
116 	struct vattr va;
117 	struct nlookupdata nd;
118 	char fstypename[MFSNAMELEN];
119 	struct nlcomponent nlc;
120 	struct ucred *cred = p->p_ucred;
121 
122 	KKASSERT(p);
123 	if (cred->cr_prison != NULL)
124 		return (EPERM);
125 	if (usermount == 0 && (error = suser(td)))
126 		return (error);
127 	/*
128 	 * Do not allow NFS export by non-root users.
129 	 */
130 	if (uap->flags & MNT_EXPORTED) {
131 		error = suser(td);
132 		if (error)
133 			return (error);
134 	}
135 	/*
136 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
137 	 */
138 	if (suser(td))
139 		uap->flags |= MNT_NOSUID | MNT_NODEV;
140 
141 	/*
142 	 * Lookup the requested path and extract the ncp and vnode.
143 	 */
144 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
145 	if (error == 0) {
146 		if ((error = nlookup(&nd)) == 0) {
147 			if (nd.nl_ncp->nc_vp == NULL)
148 				error = ENOENT;
149 		}
150 	}
151 	if (error) {
152 		nlookup_done(&nd);
153 		return (error);
154 	}
155 
156 	/*
157 	 * Extract the locked+refd ncp and cleanup the nd structure
158 	 */
159 	ncp = nd.nl_ncp;
160 	nd.nl_ncp = NULL;
161 	nlookup_done(&nd);
162 
163 	if ((ncp->nc_flag & NCF_MOUNTEDHERE) && cache_findmount(ncp))
164 		hasmount = 1;
165 	else
166 		hasmount = 0;
167 
168 
169 	/*
170 	 * now we have the locked ref'd ncp and unreferenced vnode.
171 	 */
172 	vp = ncp->nc_vp;
173 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
174 		cache_put(ncp);
175 		return (error);
176 	}
177 	cache_unlock(ncp);
178 
179 	/*
180 	 * Now we have an unlocked ref'd ncp and a locked ref'd vp
181 	 */
182 	if (uap->flags & MNT_UPDATE) {
183 		if ((vp->v_flag & VROOT) == 0) {
184 			cache_drop(ncp);
185 			vput(vp);
186 			return (EINVAL);
187 		}
188 		mp = vp->v_mount;
189 		flag = mp->mnt_flag;
190 		flag2 = mp->mnt_kern_flag;
191 		/*
192 		 * We only allow the filesystem to be reloaded if it
193 		 * is currently mounted read-only.
194 		 */
195 		if ((uap->flags & MNT_RELOAD) &&
196 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
197 			cache_drop(ncp);
198 			vput(vp);
199 			return (EOPNOTSUPP);	/* Needs translation */
200 		}
201 		/*
202 		 * Only root, or the user that did the original mount is
203 		 * permitted to update it.
204 		 */
205 		if (mp->mnt_stat.f_owner != cred->cr_uid &&
206 		    (error = suser(td))) {
207 			cache_drop(ncp);
208 			vput(vp);
209 			return (error);
210 		}
211 		if (vfs_busy(mp, LK_NOWAIT)) {
212 			cache_drop(ncp);
213 			vput(vp);
214 			return (EBUSY);
215 		}
216 		if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
217 			cache_drop(ncp);
218 			vfs_unbusy(mp);
219 			vput(vp);
220 			return (EBUSY);
221 		}
222 		vp->v_flag |= VMOUNT;
223 		mp->mnt_flag |=
224 		    uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
225 		vn_unlock(vp);
226 		goto update;
227 	}
228 	/*
229 	 * If the user is not root, ensure that they own the directory
230 	 * onto which we are attempting to mount.
231 	 */
232 	if ((error = VOP_GETATTR(vp, &va)) ||
233 	    (va.va_uid != cred->cr_uid && (error = suser(td)))) {
234 		cache_drop(ncp);
235 		vput(vp);
236 		return (error);
237 	}
238 	if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
239 		cache_drop(ncp);
240 		vput(vp);
241 		return (error);
242 	}
243 	if (vp->v_type != VDIR) {
244 		cache_drop(ncp);
245 		vput(vp);
246 		return (ENOTDIR);
247 	}
248 	if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
249 		cache_drop(ncp);
250 		vput(vp);
251 		return (error);
252 	}
253 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
254 		if (!strcmp(vfsp->vfc_name, fstypename))
255 			break;
256 	}
257 	if (vfsp == NULL) {
258 		linker_file_t lf;
259 
260 		/* Only load modules for root (very important!) */
261 		if ((error = suser(td)) != 0) {
262 			cache_drop(ncp);
263 			vput(vp);
264 			return error;
265 		}
266 		error = linker_load_file(fstypename, &lf);
267 		if (error || lf == NULL) {
268 			cache_drop(ncp);
269 			vput(vp);
270 			if (lf == NULL)
271 				error = ENODEV;
272 			return error;
273 		}
274 		lf->userrefs++;
275 		/* lookup again, see if the VFS was loaded */
276 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
277 			if (!strcmp(vfsp->vfc_name, fstypename))
278 				break;
279 		}
280 		if (vfsp == NULL) {
281 			lf->userrefs--;
282 			linker_file_unload(lf);
283 			cache_drop(ncp);
284 			vput(vp);
285 			return (ENODEV);
286 		}
287 	}
288 	if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
289 		cache_drop(ncp);
290 		vput(vp);
291 		return (EBUSY);
292 	}
293 	vp->v_flag |= VMOUNT;
294 
295 	/*
296 	 * Allocate and initialize the filesystem.
297 	 */
298 	mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
299 	TAILQ_INIT(&mp->mnt_nvnodelist);
300 	TAILQ_INIT(&mp->mnt_reservedvnlist);
301 	TAILQ_INIT(&mp->mnt_jlist);
302 	mp->mnt_nvnodelistsize = 0;
303 	lockinit(&mp->mnt_lock, "vfslock", 0, 0);
304 	vfs_busy(mp, LK_NOWAIT);
305 	mp->mnt_op = vfsp->vfc_vfsops;
306 	mp->mnt_vfc = vfsp;
307 	vfsp->vfc_refcount++;
308 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
309 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
310 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
311 	mp->mnt_stat.f_owner = cred->cr_uid;
312 	mp->mnt_iosize_max = DFLTPHYS;
313 	vn_unlock(vp);
314 update:
315 	/*
316 	 * Set the mount level flags.
317 	 */
318 	if (uap->flags & MNT_RDONLY)
319 		mp->mnt_flag |= MNT_RDONLY;
320 	else if (mp->mnt_flag & MNT_RDONLY)
321 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
322 	mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
323 	    MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
324 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
325 	    MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
326 	mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
327 	    MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
328 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
329 	    MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
330 	/*
331 	 * Mount the filesystem.
332 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
333 	 * get.
334 	 */
335 	error = VFS_MOUNT(mp, uap->path, uap->data, cred);
336 	if (mp->mnt_flag & MNT_UPDATE) {
337 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
338 			mp->mnt_flag &= ~MNT_RDONLY;
339 		mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
340 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
341 		if (error) {
342 			mp->mnt_flag = flag;
343 			mp->mnt_kern_flag = flag2;
344 		}
345 		vfs_unbusy(mp);
346 		vp->v_flag &= ~VMOUNT;
347 		vrele(vp);
348 		cache_drop(ncp);
349 		return (error);
350 	}
351 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
352 	/*
353 	 * Put the new filesystem on the mount list after root.  The mount
354 	 * point gets its own mnt_ncp which is a special ncp linking the
355 	 * vnode-under to the root of the new mount.  The lookup code
356 	 * detects the mount point going forward and detects the special
357 	 * mnt_ncp via NCP_MOUNTPT going backwards.
358 	 *
359 	 * It is not necessary to invalidate or purge the vnode underneath
360 	 * because elements under the mount will be given their own glue
361 	 * namecache record.
362 	 */
363 	if (!error) {
364 		nlc.nlc_nameptr = "";
365 		nlc.nlc_namelen = 0;
366 		mp->mnt_ncp = cache_nlookup(ncp, &nlc);
367 		cache_setunresolved(mp->mnt_ncp);
368 		cache_setmountpt(mp->mnt_ncp, mp);
369 		cache_drop(ncp);
370 		/* XXX get the root of the fs and cache_setvp(mnt_ncp...) */
371 		vp->v_flag &= ~VMOUNT;
372 		mountlist_insert(mp, MNTINS_LAST);
373 		checkdirs(vp, mp->mnt_ncp);
374 		cache_unlock(mp->mnt_ncp);	/* leave ref intact */
375 		vn_unlock(vp);
376 		error = vfs_allocate_syncvnode(mp);
377 		vfs_unbusy(mp);
378 		error = VFS_START(mp, 0);
379 		vrele(vp);
380 	} else {
381 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
382 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
383 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
384 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
385 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
386 		vp->v_flag &= ~VMOUNT;
387 		mp->mnt_vfc->vfc_refcount--;
388 		vfs_unbusy(mp);
389 		kfree(mp, M_MOUNT);
390 		cache_drop(ncp);
391 		vput(vp);
392 	}
393 	return (error);
394 }
395 
396 /*
397  * Scan all active processes to see if any of them have a current
398  * or root directory onto which the new filesystem has just been
399  * mounted. If so, replace them with the new mount point.
400  *
401  * The passed ncp is ref'd and locked (from the mount code) and
402  * must be associated with the vnode representing the root of the
403  * mount point.
404  */
405 struct checkdirs_info {
406 	struct vnode *olddp;
407 	struct vnode *newdp;
408 	struct namecache *ncp;
409 };
410 
411 static int checkdirs_callback(struct proc *p, void *data);
412 
413 static void
414 checkdirs(struct vnode *olddp, struct namecache *ncp)
415 {
416 	struct checkdirs_info info;
417 	struct vnode *newdp;
418 	struct mount *mp;
419 
420 	if (olddp->v_usecount == 1)
421 		return;
422 	mp = ncp->nc_mount;
423 	if (VFS_ROOT(mp, &newdp))
424 		panic("mount: lost mount");
425 	cache_setvp(ncp, newdp);
426 
427 	if (rootvnode == olddp) {
428 		vref(newdp);
429 		vfs_cache_setroot(newdp, cache_hold(ncp));
430 	}
431 
432 	info.olddp = olddp;
433 	info.newdp = newdp;
434 	info.ncp = ncp;
435 	allproc_scan(checkdirs_callback, &info);
436 	vput(newdp);
437 }
438 
439 /*
440  * NOTE: callback is not MP safe because the scanned process's filedesc
441  * structure can be ripped out from under us, amoung other things.
442  */
443 static int
444 checkdirs_callback(struct proc *p, void *data)
445 {
446 	struct checkdirs_info *info = data;
447 	struct filedesc *fdp;
448 	struct namecache *ncdrop1;
449 	struct namecache *ncdrop2;
450 	struct vnode *vprele1;
451 	struct vnode *vprele2;
452 
453 	if ((fdp = p->p_fd) != NULL) {
454 		ncdrop1 = NULL;
455 		ncdrop2 = NULL;
456 		vprele1 = NULL;
457 		vprele2 = NULL;
458 
459 		/*
460 		 * MPUNSAFE - XXX fdp can be pulled out from under a
461 		 * foreign process.
462 		 *
463 		 * A shared filedesc is ok, we don't have to copy it
464 		 * because we are making this change globally.
465 		 */
466 		spin_lock_wr(&fdp->fd_spin);
467 		if (fdp->fd_cdir == info->olddp) {
468 			vprele1 = fdp->fd_cdir;
469 			vref(info->newdp);
470 			fdp->fd_cdir = info->newdp;
471 			ncdrop1 = fdp->fd_ncdir;
472 			fdp->fd_ncdir = cache_hold(info->ncp);
473 		}
474 		if (fdp->fd_rdir == info->olddp) {
475 			vprele2 = fdp->fd_rdir;
476 			vref(info->newdp);
477 			fdp->fd_rdir = info->newdp;
478 			ncdrop2 = fdp->fd_nrdir;
479 			fdp->fd_nrdir = cache_hold(info->ncp);
480 		}
481 		spin_unlock_wr(&fdp->fd_spin);
482 		if (ncdrop1)
483 			cache_drop(ncdrop1);
484 		if (ncdrop2)
485 			cache_drop(ncdrop2);
486 		if (vprele1)
487 			vrele(vprele1);
488 		if (vprele2)
489 			vrele(vprele2);
490 	}
491 	return(0);
492 }
493 
494 /*
495  * Unmount a file system.
496  *
497  * Note: unmount takes a path to the vnode mounted on as argument,
498  * not special file (as before).
499  */
500 /*
501  * umount_args(char *path, int flags)
502  */
503 /* ARGSUSED */
504 int
505 sys_unmount(struct unmount_args *uap)
506 {
507 	struct thread *td = curthread;
508 	struct proc *p = td->td_proc;
509 	struct mount *mp = NULL;
510 	int error;
511 	struct nlookupdata nd;
512 
513 	KKASSERT(p);
514 	if (p->p_ucred->cr_prison != NULL)
515 		return (EPERM);
516 	if (usermount == 0 && (error = suser(td)))
517 		return (error);
518 
519 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
520 	if (error == 0)
521 		error = nlookup(&nd);
522 	if (error)
523 		goto out;
524 
525 	mp = nd.nl_ncp->nc_mount;
526 
527 	/*
528 	 * Only root, or the user that did the original mount is
529 	 * permitted to unmount this filesystem.
530 	 */
531 	if ((mp->mnt_stat.f_owner != p->p_ucred->cr_uid) &&
532 	    (error = suser(td)))
533 		goto out;
534 
535 	/*
536 	 * Don't allow unmounting the root file system.
537 	 */
538 	if (mp->mnt_flag & MNT_ROOTFS) {
539 		error = EINVAL;
540 		goto out;
541 	}
542 
543 	/*
544 	 * Must be the root of the filesystem
545 	 */
546 	if (!(nd.nl_ncp->nc_flag & NCF_MOUNTPT)) {
547 		error = EINVAL;
548 		goto out;
549 	}
550 
551 out:
552 	nlookup_done(&nd);
553 	if (error)
554 		return (error);
555 	return (dounmount(mp, uap->flags));
556 }
557 
558 /*
559  * Do the actual file system unmount.
560  */
561 static int
562 dounmount_interlock(struct mount *mp)
563 {
564 	if (mp->mnt_kern_flag & MNTK_UNMOUNT)
565 		return (EBUSY);
566 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
567 	return(0);
568 }
569 
570 int
571 dounmount(struct mount *mp, int flags)
572 {
573 	struct namecache *ncp;
574 	int error;
575 	int async_flag;
576 	int lflags;
577 
578 	/*
579 	 * Exclusive access for unmounting purposes
580 	 */
581 	if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
582 		return (error);
583 
584 	/*
585 	 * Allow filesystems to detect that a forced unmount is in progress.
586 	 */
587 	if (flags & MNT_FORCE)
588 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
589 	lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
590 	error = lockmgr(&mp->mnt_lock, lflags);
591 	if (error) {
592 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
593 		if (mp->mnt_kern_flag & MNTK_MWAIT)
594 			wakeup(mp);
595 		return (error);
596 	}
597 
598 	if (mp->mnt_flag & MNT_EXPUBLIC)
599 		vfs_setpublicfs(NULL, NULL, NULL);
600 
601 	vfs_msync(mp, MNT_WAIT);
602 	async_flag = mp->mnt_flag & MNT_ASYNC;
603 	mp->mnt_flag &=~ MNT_ASYNC;
604 
605 	/*
606 	 * remove cache entries for this file sys and determine if anyone
607 	 * other then us is still holding onto any namecache references.
608 	 *
609 	 * XXX need separate ref counter on mount structure to delay
610 	 * kfree()ing it.
611 	 */
612 	cache_purgevfs(mp);
613 	if ((ncp = mp->mnt_ncp) != NULL) {
614 		if (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list)) {
615 			char *ptr;
616 			char *buf;
617 
618 			if ((flags & MNT_FORCE) == 0) {
619 				error = EBUSY;
620 			} else if (cache_fullpath(NULL, ncp, &ptr, &buf)) {
621 				printf("Warning: forced unmount - "
622 				       "namecache references still present\n");
623 			} else {
624 				printf("Warning: forced unmount of %s - "
625 				       "namecache references still present\n",
626 				       ptr
627 				);
628 				kfree(buf, M_TEMP);
629 			}
630 		}
631 	}
632 
633 	if (error == 0) {
634 		if (mp->mnt_syncer != NULL)
635 			vrele(mp->mnt_syncer);
636 		if (((mp->mnt_flag & MNT_RDONLY) ||
637 		     (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
638 		    (flags & MNT_FORCE)) {
639 			error = VFS_UNMOUNT(mp, flags);
640 		}
641 	}
642 	if (error) {
643 		if (mp->mnt_syncer == NULL)
644 			vfs_allocate_syncvnode(mp);
645 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
646 		mp->mnt_flag |= async_flag;
647 		lockmgr(&mp->mnt_lock, LK_RELEASE);
648 		if (mp->mnt_kern_flag & MNTK_MWAIT)
649 			wakeup(mp);
650 		return (error);
651 	}
652 	/*
653 	 * Clean up any journals still associated with the mount after
654 	 * filesystem activity has ceased.
655 	 */
656 	journal_remove_all_journals(mp,
657 	    ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
658 
659 	mountlist_remove(mp);
660 
661 	/*
662 	 * Remove any installed vnode ops here so the individual VFSs don't
663 	 * have to.
664 	 */
665 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
666 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
667 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
668 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
669 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
670 
671 	if ((ncp = mp->mnt_ncp) != NULL) {
672 		cache_clrmountpt(ncp);
673 		cache_drop(ncp);
674 		mp->mnt_ncp = NULL;
675 	}
676 
677 	mp->mnt_vfc->vfc_refcount--;
678 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
679 		panic("unmount: dangling vnode");
680 	lockmgr(&mp->mnt_lock, LK_RELEASE);
681 	if (mp->mnt_kern_flag & MNTK_MWAIT)
682 		wakeup(mp);
683 	kfree(mp, M_MOUNT);
684 	return (0);
685 }
686 
687 /*
688  * Sync each mounted filesystem.
689  */
690 
691 #ifdef DEBUG
692 static int syncprt = 0;
693 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
694 #endif /* DEBUG */
695 
696 static int sync_callback(struct mount *mp, void *data);
697 
698 /* ARGSUSED */
699 int
700 sys_sync(struct sync_args *uap)
701 {
702 	mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
703 #ifdef DEBUG
704 	/*
705 	 * print out buffer pool stat information on each sync() call.
706 	 */
707 	if (syncprt)
708 		vfs_bufstats();
709 #endif /* DEBUG */
710 	return (0);
711 }
712 
713 static
714 int
715 sync_callback(struct mount *mp, void *data __unused)
716 {
717 	int asyncflag;
718 
719 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
720 		asyncflag = mp->mnt_flag & MNT_ASYNC;
721 		mp->mnt_flag &= ~MNT_ASYNC;
722 		vfs_msync(mp, MNT_NOWAIT);
723 		VFS_SYNC(mp, MNT_NOWAIT);
724 		mp->mnt_flag |= asyncflag;
725 	}
726 	return(0);
727 }
728 
729 /* XXX PRISON: could be per prison flag */
730 static int prison_quotas;
731 #if 0
732 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
733 #endif
734 
735 /*
736  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
737  *
738  * Change filesystem quotas.
739  */
740 /* ARGSUSED */
741 int
742 sys_quotactl(struct quotactl_args *uap)
743 {
744 	struct nlookupdata nd;
745 	struct thread *td;
746 	struct proc *p;
747 	struct mount *mp;
748 	int error;
749 
750 	td = curthread;
751 	p = td->td_proc;
752 	if (p->p_ucred->cr_prison && !prison_quotas)
753 		return (EPERM);
754 
755 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
756 	if (error == 0)
757 		error = nlookup(&nd);
758 	if (error == 0) {
759 		mp = nd.nl_ncp->nc_mount;
760 		error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
761 				    uap->arg, nd.nl_cred);
762 	}
763 	nlookup_done(&nd);
764 	return (error);
765 }
766 
767 /*
768  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
769  *		void *buf, int buflen)
770  *
771  * This function operates on a mount point and executes the specified
772  * operation using the specified control data, and possibly returns data.
773  *
774  * The actual number of bytes stored in the result buffer is returned, 0
775  * if none, otherwise an error is returned.
776  */
777 /* ARGSUSED */
778 int
779 sys_mountctl(struct mountctl_args *uap)
780 {
781 	struct thread *td = curthread;
782 	struct proc *p = td->td_proc;
783 	struct file *fp;
784 	void *ctl = NULL;
785 	void *buf = NULL;
786 	char *path = NULL;
787 	int error;
788 
789 	/*
790 	 * Sanity and permissions checks.  We must be root.
791 	 */
792 	KKASSERT(p);
793 	if (p->p_ucred->cr_prison != NULL)
794 		return (EPERM);
795 	if ((error = suser(td)) != 0)
796 		return (error);
797 
798 	/*
799 	 * Argument length checks
800 	 */
801 	if (uap->ctllen < 0 || uap->ctllen > 1024)
802 		return (EINVAL);
803 	if (uap->buflen < 0 || uap->buflen > 16 * 1024)
804 		return (EINVAL);
805 	if (uap->path == NULL)
806 		return (EINVAL);
807 
808 	/*
809 	 * Allocate the necessary buffers and copyin data
810 	 */
811 	path = objcache_get(namei_oc, M_WAITOK);
812 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
813 	if (error)
814 		goto done;
815 
816 	if (uap->ctllen) {
817 		ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
818 		error = copyin(uap->ctl, ctl, uap->ctllen);
819 		if (error)
820 			goto done;
821 	}
822 	if (uap->buflen)
823 		buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
824 
825 	/*
826 	 * Validate the descriptor
827 	 */
828 	fp = holdfp(p->p_fd, uap->fd, -1);
829 	if (fp == NULL) {
830 		error = EBADF;
831 		goto done;
832 	}
833 
834 	/*
835 	 * Execute the internal kernel function and clean up.
836 	 */
837 	error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
838 	if (fp)
839 		fdrop(fp);
840 	if (error == 0 && uap->sysmsg_result > 0)
841 		error = copyout(buf, uap->buf, uap->sysmsg_result);
842 done:
843 	if (path)
844 		objcache_put(namei_oc, path);
845 	if (ctl)
846 		kfree(ctl, M_TEMP);
847 	if (buf)
848 		kfree(buf, M_TEMP);
849 	return (error);
850 }
851 
852 /*
853  * Execute a mount control operation by resolving the path to a mount point
854  * and calling vop_mountctl().
855  */
856 int
857 kern_mountctl(const char *path, int op, struct file *fp,
858 		const void *ctl, int ctllen,
859 		void *buf, int buflen, int *res)
860 {
861 	struct vnode *vp;
862 	struct mount *mp;
863 	struct nlookupdata nd;
864 	int error;
865 
866 	*res = 0;
867 	vp = NULL;
868 	error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
869 	if (error == 0)
870 		error = nlookup(&nd);
871 	if (error == 0)
872 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
873 	nlookup_done(&nd);
874 	if (error)
875 		return (error);
876 
877 	mp = vp->v_mount;
878 
879 	/*
880 	 * Must be the root of the filesystem
881 	 */
882 	if ((vp->v_flag & VROOT) == 0) {
883 		vput(vp);
884 		return (EINVAL);
885 	}
886 	error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen,
887 				buf, buflen, res);
888 	vput(vp);
889 	return (error);
890 }
891 
892 int
893 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
894 {
895 	struct thread *td = curthread;
896 	struct proc *p = td->td_proc;
897 	struct mount *mp;
898 	struct statfs *sp;
899 	char *fullpath, *freepath;
900 	int error;
901 
902 	if ((error = nlookup(nd)) != 0)
903 		return (error);
904 	mp = nd->nl_ncp->nc_mount;
905 	sp = &mp->mnt_stat;
906 	if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
907 		return (error);
908 
909 	error = cache_fullpath(p, mp->mnt_ncp, &fullpath, &freepath);
910 	if (error)
911 		return(error);
912 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
913 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
914 	kfree(freepath, M_TEMP);
915 
916 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
917 	bcopy(sp, buf, sizeof(*buf));
918 	/* Only root should have access to the fsid's. */
919 	if (suser(td))
920 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
921 	return (0);
922 }
923 
924 /*
925  * statfs_args(char *path, struct statfs *buf)
926  *
927  * Get filesystem statistics.
928  */
929 int
930 sys_statfs(struct statfs_args *uap)
931 {
932 	struct nlookupdata nd;
933 	struct statfs buf;
934 	int error;
935 
936 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
937 	if (error == 0)
938 		error = kern_statfs(&nd, &buf);
939 	nlookup_done(&nd);
940 	if (error == 0)
941 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
942 	return (error);
943 }
944 
945 int
946 kern_fstatfs(int fd, struct statfs *buf)
947 {
948 	struct thread *td = curthread;
949 	struct proc *p = td->td_proc;
950 	struct file *fp;
951 	struct mount *mp;
952 	struct statfs *sp;
953 	char *fullpath, *freepath;
954 	int error;
955 
956 	KKASSERT(p);
957 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
958 		return (error);
959 	mp = ((struct vnode *)fp->f_data)->v_mount;
960 	if (mp == NULL) {
961 		error = EBADF;
962 		goto done;
963 	}
964 	if (fp->f_cred == NULL) {
965 		error = EINVAL;
966 		goto done;
967 	}
968 	sp = &mp->mnt_stat;
969 	if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
970 		goto done;
971 
972 	if ((error = cache_fullpath(p, mp->mnt_ncp, &fullpath, &freepath)) != 0)
973 		goto done;
974 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
975 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
976 	kfree(freepath, M_TEMP);
977 
978 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
979 	bcopy(sp, buf, sizeof(*buf));
980 
981 	/* Only root should have access to the fsid's. */
982 	if (suser(td))
983 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
984 	error = 0;
985 done:
986 	fdrop(fp);
987 	return (error);
988 }
989 
990 /*
991  * fstatfs_args(int fd, struct statfs *buf)
992  *
993  * Get filesystem statistics.
994  */
995 int
996 sys_fstatfs(struct fstatfs_args *uap)
997 {
998 	struct statfs buf;
999 	int error;
1000 
1001 	error = kern_fstatfs(uap->fd, &buf);
1002 
1003 	if (error == 0)
1004 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1005 	return (error);
1006 }
1007 
1008 /*
1009  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1010  *
1011  * Get statistics on all filesystems.
1012  */
1013 
1014 struct getfsstat_info {
1015 	struct statfs *sfsp;
1016 	long count;
1017 	long maxcount;
1018 	int error;
1019 	int flags;
1020 	int is_chrooted;
1021 	struct proc *p;
1022 };
1023 
1024 static int getfsstat_callback(struct mount *, void *);
1025 
1026 /* ARGSUSED */
1027 int
1028 sys_getfsstat(struct getfsstat_args *uap)
1029 {
1030 	struct thread *td = curthread;
1031 	struct proc *p = td->td_proc;
1032 	struct getfsstat_info info;
1033 
1034 	bzero(&info, sizeof(info));
1035 	if (p != NULL && (p->p_fd->fd_nrdir->nc_flag & NCF_ROOT) == 0)
1036 		info.is_chrooted = 1;
1037 	else
1038 		info.is_chrooted = 0;
1039 
1040 	info.maxcount = uap->bufsize / sizeof(struct statfs);
1041 	info.sfsp = uap->buf;
1042 	info.count = 0;
1043 	info.flags = uap->flags;
1044 	info.p = p;
1045 
1046 	mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1047 	if (info.sfsp && info.count > info.maxcount)
1048 		uap->sysmsg_result = info.maxcount;
1049 	else
1050 		uap->sysmsg_result = info.count;
1051 	return (info.error);
1052 }
1053 
1054 static int
1055 getfsstat_callback(struct mount *mp, void *data)
1056 {
1057 	struct getfsstat_info *info = data;
1058 	struct statfs *sp;
1059 	char *freepath;
1060 	char *fullpath;
1061 	int error;
1062 
1063 	if (info->sfsp && info->count < info->maxcount) {
1064 		if (info->is_chrooted && !chroot_visible_mnt(mp, info->p))
1065 			return(0);
1066 		sp = &mp->mnt_stat;
1067 
1068 		/*
1069 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1070 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1071 		 * overrides MNT_WAIT.
1072 		 */
1073 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1074 		    (info->flags & MNT_WAIT)) &&
1075 		    (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1076 			return(0);
1077 		}
1078 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1079 
1080 		error = cache_fullpath(info->p, mp->mnt_ncp,
1081 					&fullpath, &freepath);
1082 		if (error) {
1083 			info->error = error;
1084 			return(-1);
1085 		}
1086 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1087 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1088 		kfree(freepath, M_TEMP);
1089 
1090 		error = copyout(sp, info->sfsp, sizeof(*sp));
1091 		if (error) {
1092 			info->error = error;
1093 			return (-1);
1094 		}
1095 		++info->sfsp;
1096 	}
1097 	info->count++;
1098 	return(0);
1099 }
1100 
1101 /*
1102  * fchdir_args(int fd)
1103  *
1104  * Change current working directory to a given file descriptor.
1105  */
1106 /* ARGSUSED */
1107 int
1108 sys_fchdir(struct fchdir_args *uap)
1109 {
1110 	struct thread *td = curthread;
1111 	struct proc *p = td->td_proc;
1112 	struct filedesc *fdp = p->p_fd;
1113 	struct vnode *vp, *ovp;
1114 	struct mount *mp;
1115 	struct file *fp;
1116 	struct namecache *ncp, *oncp;
1117 	struct namecache *nct;
1118 	int error;
1119 
1120 	if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1121 		return (error);
1122 	vp = (struct vnode *)fp->f_data;
1123 	vref(vp);
1124 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1125 	if (vp->v_type != VDIR || fp->f_ncp == NULL)
1126 		error = ENOTDIR;
1127 	else
1128 		error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
1129 	if (error) {
1130 		vput(vp);
1131 		fdrop(fp);
1132 		return (error);
1133 	}
1134 	ncp = cache_hold(fp->f_ncp);
1135 
1136 	/*
1137 	 * If the ncp has become a mount point, traverse through
1138 	 * the mount point.
1139 	 */
1140 
1141 	while (!error && (ncp->nc_flag & NCF_MOUNTEDHERE) &&
1142 	       (mp = cache_findmount(ncp)) != NULL
1143 	) {
1144 		error = nlookup_mp(mp, &nct);
1145 		if (error == 0) {
1146 			cache_unlock(nct);	/* leave ref intact */
1147 			vput(vp);
1148 			vp = nct->nc_vp;
1149 			error = vget(vp, LK_SHARED);
1150 			KKASSERT(error == 0);
1151 			cache_drop(ncp);
1152 			ncp = nct;
1153 		}
1154 	}
1155 	if (error == 0) {
1156 		ovp = fdp->fd_cdir;
1157 		oncp = fdp->fd_ncdir;
1158 		vn_unlock(vp);		/* leave ref intact */
1159 		fdp->fd_cdir = vp;
1160 		fdp->fd_ncdir = ncp;
1161 		cache_drop(oncp);
1162 		vrele(ovp);
1163 	} else {
1164 		cache_drop(ncp);
1165 		vput(vp);
1166 	}
1167 	fdrop(fp);
1168 	return (error);
1169 }
1170 
1171 int
1172 kern_chdir(struct nlookupdata *nd)
1173 {
1174 	struct thread *td = curthread;
1175 	struct proc *p = td->td_proc;
1176 	struct filedesc *fdp = p->p_fd;
1177 	struct vnode *vp, *ovp;
1178 	struct namecache *oncp;
1179 	int error;
1180 
1181 	if ((error = nlookup(nd)) != 0)
1182 		return (error);
1183 	if ((vp = nd->nl_ncp->nc_vp) == NULL)
1184 		return (ENOENT);
1185 	if ((error = vget(vp, LK_SHARED)) != 0)
1186 		return (error);
1187 
1188 	error = checkvp_chdir(vp, td);
1189 	vn_unlock(vp);
1190 	if (error == 0) {
1191 		ovp = fdp->fd_cdir;
1192 		oncp = fdp->fd_ncdir;
1193 		cache_unlock(nd->nl_ncp);	/* leave reference intact */
1194 		fdp->fd_ncdir = nd->nl_ncp;
1195 		fdp->fd_cdir = vp;
1196 		cache_drop(oncp);
1197 		vrele(ovp);
1198 		nd->nl_ncp = NULL;
1199 	} else {
1200 		vrele(vp);
1201 	}
1202 	return (error);
1203 }
1204 
1205 /*
1206  * chdir_args(char *path)
1207  *
1208  * Change current working directory (``.'').
1209  */
1210 int
1211 sys_chdir(struct chdir_args *uap)
1212 {
1213 	struct nlookupdata nd;
1214 	int error;
1215 
1216 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1217 	if (error == 0)
1218 		error = kern_chdir(&nd);
1219 	nlookup_done(&nd);
1220 	return (error);
1221 }
1222 
1223 /*
1224  * Helper function for raised chroot(2) security function:  Refuse if
1225  * any filedescriptors are open directories.
1226  */
1227 static int
1228 chroot_refuse_vdir_fds(fdp)
1229 	struct filedesc *fdp;
1230 {
1231 	struct vnode *vp;
1232 	struct file *fp;
1233 	int error;
1234 	int fd;
1235 
1236 	for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1237 		if ((error = holdvnode(fdp, fd, &fp)) != 0)
1238 			continue;
1239 		vp = (struct vnode *)fp->f_data;
1240 		if (vp->v_type != VDIR) {
1241 			fdrop(fp);
1242 			continue;
1243 		}
1244 		fdrop(fp);
1245 		return(EPERM);
1246 	}
1247 	return (0);
1248 }
1249 
1250 /*
1251  * This sysctl determines if we will allow a process to chroot(2) if it
1252  * has a directory open:
1253  *	0: disallowed for all processes.
1254  *	1: allowed for processes that were not already chroot(2)'ed.
1255  *	2: allowed for all processes.
1256  */
1257 
1258 static int chroot_allow_open_directories = 1;
1259 
1260 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1261      &chroot_allow_open_directories, 0, "");
1262 
1263 /*
1264  * chroot to the specified namecache entry.  We obtain the vp from the
1265  * namecache data.  The passed ncp must be locked and referenced and will
1266  * remain locked and referenced on return.
1267  */
1268 int
1269 kern_chroot(struct namecache *ncp)
1270 {
1271 	struct thread *td = curthread;
1272 	struct proc *p = td->td_proc;
1273 	struct filedesc *fdp = p->p_fd;
1274 	struct vnode *vp;
1275 	int error;
1276 
1277 	/*
1278 	 * Only root can chroot
1279 	 */
1280 	if ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0)
1281 		return (error);
1282 
1283 	/*
1284 	 * Disallow open directory descriptors (fchdir() breakouts).
1285 	 */
1286 	if (chroot_allow_open_directories == 0 ||
1287 	   (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1288 		if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1289 			return (error);
1290 	}
1291 	if ((vp = ncp->nc_vp) == NULL)
1292 		return (ENOENT);
1293 
1294 	if ((error = vget(vp, LK_SHARED)) != 0)
1295 		return (error);
1296 
1297 	/*
1298 	 * Check the validity of vp as a directory to change to and
1299 	 * associate it with rdir/jdir.
1300 	 */
1301 	error = checkvp_chdir(vp, td);
1302 	vn_unlock(vp);			/* leave reference intact */
1303 	if (error == 0) {
1304 		vrele(fdp->fd_rdir);
1305 		fdp->fd_rdir = vp;	/* reference inherited by fd_rdir */
1306 		cache_drop(fdp->fd_nrdir);
1307 		fdp->fd_nrdir = cache_hold(ncp);
1308 		if (fdp->fd_jdir == NULL) {
1309 			fdp->fd_jdir = vp;
1310 			vref(fdp->fd_jdir);
1311 			fdp->fd_njdir = cache_hold(ncp);
1312 		}
1313 	} else {
1314 		vrele(vp);
1315 	}
1316 	return (error);
1317 }
1318 
1319 /*
1320  * chroot_args(char *path)
1321  *
1322  * Change notion of root (``/'') directory.
1323  */
1324 /* ARGSUSED */
1325 int
1326 sys_chroot(struct chroot_args *uap)
1327 {
1328 	struct thread *td = curthread;
1329 	struct nlookupdata nd;
1330 	int error;
1331 
1332 	KKASSERT(td->td_proc);
1333 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1334 	if (error) {
1335 		nlookup_done(&nd);
1336 		return(error);
1337 	}
1338 	error = nlookup(&nd);
1339 	if (error == 0)
1340 		error = kern_chroot(nd.nl_ncp);
1341 	nlookup_done(&nd);
1342 	return(error);
1343 }
1344 
1345 /*
1346  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1347  * determine whether it is legal to chdir to the vnode.  The vnode's state
1348  * is not changed by this call.
1349  */
1350 int
1351 checkvp_chdir(struct vnode *vp, struct thread *td)
1352 {
1353 	int error;
1354 
1355 	if (vp->v_type != VDIR)
1356 		error = ENOTDIR;
1357 	else
1358 		error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred);
1359 	return (error);
1360 }
1361 
1362 int
1363 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1364 {
1365 	struct thread *td = curthread;
1366 	struct proc *p = td->td_proc;
1367 	struct lwp *lp = td->td_lwp;
1368 	struct filedesc *fdp = p->p_fd;
1369 	int cmode, flags;
1370 	struct file *nfp;
1371 	struct file *fp;
1372 	struct vnode *vp;
1373 	int type, indx, error;
1374 	struct flock lf;
1375 
1376 	if ((oflags & O_ACCMODE) == O_ACCMODE)
1377 		return (EINVAL);
1378 	flags = FFLAGS(oflags);
1379 	error = falloc(p, &nfp, NULL);
1380 	if (error)
1381 		return (error);
1382 	fp = nfp;
1383 	cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) &~ S_ISTXT;
1384 
1385 	/*
1386 	 * XXX p_dupfd is a real mess.  It allows a device to return a
1387 	 * file descriptor to be duplicated rather then doing the open
1388 	 * itself.
1389 	 */
1390 	lp->lwp_dupfd = -1;
1391 
1392 	/*
1393 	 * Call vn_open() to do the lookup and assign the vnode to the
1394 	 * file pointer.  vn_open() does not change the ref count on fp
1395 	 * and the vnode, on success, will be inherited by the file pointer
1396 	 * and unlocked.
1397 	 */
1398 	nd->nl_flags |= NLC_LOCKVP;
1399 	error = vn_open(nd, fp, flags, cmode);
1400 	nlookup_done(nd);
1401 	if (error) {
1402 		/*
1403 		 * handle special fdopen() case.  bleh.  dupfdopen() is
1404 		 * responsible for dropping the old contents of ofiles[indx]
1405 		 * if it succeeds.
1406 		 *
1407 		 * Note that fsetfd() will add a ref to fp which represents
1408 		 * the fd_files[] assignment.  We must still drop our
1409 		 * reference.
1410 		 */
1411 		if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1412 			if (fdalloc(p, 0, &indx) == 0) {
1413 				error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error);
1414 				if (error == 0) {
1415 					*res = indx;
1416 					fdrop(fp);	/* our ref */
1417 					return (0);
1418 				}
1419 				fsetfd(p, NULL, indx);
1420 			}
1421 		}
1422 		fdrop(fp);	/* our ref */
1423 		if (error == ERESTART)
1424 			error = EINTR;
1425 		return (error);
1426 	}
1427 
1428 	/*
1429 	 * ref the vnode for ourselves so it can't be ripped out from under
1430 	 * is.  XXX need an ND flag to request that the vnode be returned
1431 	 * anyway.
1432 	 *
1433 	 * Reserve a file descriptor but do not assign it until the open
1434 	 * succeeds.
1435 	 */
1436 	vp = (struct vnode *)fp->f_data;
1437 	vref(vp);
1438 	if ((error = fdalloc(p, 0, &indx)) != 0) {
1439 		fdrop(fp);
1440 		vrele(vp);
1441 		return (error);
1442 	}
1443 
1444 	/*
1445 	 * If no error occurs the vp will have been assigned to the file
1446 	 * pointer.
1447 	 */
1448 	lp->lwp_dupfd = 0;
1449 
1450 	if (flags & (O_EXLOCK | O_SHLOCK)) {
1451 		lf.l_whence = SEEK_SET;
1452 		lf.l_start = 0;
1453 		lf.l_len = 0;
1454 		if (flags & O_EXLOCK)
1455 			lf.l_type = F_WRLCK;
1456 		else
1457 			lf.l_type = F_RDLCK;
1458 		if (flags & FNONBLOCK)
1459 			type = 0;
1460 		else
1461 			type = F_WAIT;
1462 
1463 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1464 			/*
1465 			 * lock request failed.  Clean up the reserved
1466 			 * descriptor.
1467 			 */
1468 			vrele(vp);
1469 			fsetfd(p, NULL, indx);
1470 			fdrop(fp);
1471 			return (error);
1472 		}
1473 		fp->f_flag |= FHASLOCK;
1474 	}
1475 #if 0
1476 	/*
1477 	 * Assert that all regular file vnodes were created with a object.
1478 	 */
1479 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1480 		("open: regular file has no backing object after vn_open"));
1481 #endif
1482 
1483 	vrele(vp);
1484 
1485 	/*
1486 	 * release our private reference, leaving the one associated with the
1487 	 * descriptor table intact.
1488 	 */
1489 	fsetfd(p, fp, indx);
1490 	fdrop(fp);
1491 	*res = indx;
1492 	return (0);
1493 }
1494 
1495 /*
1496  * open_args(char *path, int flags, int mode)
1497  *
1498  * Check permissions, allocate an open file structure,
1499  * and call the device open routine if any.
1500  */
1501 int
1502 sys_open(struct open_args *uap)
1503 {
1504 	struct nlookupdata nd;
1505 	int error;
1506 
1507 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1508 	if (error == 0) {
1509 		error = kern_open(&nd, uap->flags,
1510 				    uap->mode, &uap->sysmsg_result);
1511 	}
1512 	nlookup_done(&nd);
1513 	return (error);
1514 }
1515 
1516 int
1517 kern_mknod(struct nlookupdata *nd, int mode, int dev)
1518 {
1519 	struct namecache *ncp;
1520 	struct thread *td = curthread;
1521 	struct proc *p = td->td_proc;
1522 	struct vnode *vp;
1523 	struct vattr vattr;
1524 	int error;
1525 	int whiteout = 0;
1526 
1527 	KKASSERT(p);
1528 
1529 	switch (mode & S_IFMT) {
1530 	case S_IFCHR:
1531 	case S_IFBLK:
1532 		error = suser(td);
1533 		break;
1534 	default:
1535 		error = suser_cred(p->p_ucred, PRISON_ROOT);
1536 		break;
1537 	}
1538 	if (error)
1539 		return (error);
1540 
1541 	bwillwrite();
1542 	nd->nl_flags |= NLC_CREATE;
1543 	if ((error = nlookup(nd)) != 0)
1544 		return (error);
1545 	ncp = nd->nl_ncp;
1546 	if (ncp->nc_vp)
1547 		return (EEXIST);
1548 	if ((error = ncp_writechk(ncp)) != 0)
1549 		return (error);
1550 
1551 	VATTR_NULL(&vattr);
1552 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1553 	vattr.va_rdev = dev;
1554 	whiteout = 0;
1555 
1556 	switch (mode & S_IFMT) {
1557 	case S_IFMT:	/* used by badsect to flag bad sectors */
1558 		vattr.va_type = VBAD;
1559 		break;
1560 	case S_IFCHR:
1561 		vattr.va_type = VCHR;
1562 		break;
1563 	case S_IFBLK:
1564 		vattr.va_type = VBLK;
1565 		break;
1566 	case S_IFWHT:
1567 		whiteout = 1;
1568 		break;
1569 	default:
1570 		error = EINVAL;
1571 		break;
1572 	}
1573 	if (error == 0) {
1574 		if (whiteout) {
1575 			error = VOP_NWHITEOUT(ncp, nd->nl_cred, NAMEI_CREATE);
1576 		} else {
1577 			vp = NULL;
1578 			error = VOP_NMKNOD(ncp, &vp, nd->nl_cred, &vattr);
1579 			if (error == 0)
1580 				vput(vp);
1581 		}
1582 	}
1583 	return (error);
1584 }
1585 
1586 /*
1587  * mknod_args(char *path, int mode, int dev)
1588  *
1589  * Create a special file.
1590  */
1591 int
1592 sys_mknod(struct mknod_args *uap)
1593 {
1594 	struct nlookupdata nd;
1595 	int error;
1596 
1597 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1598 	if (error == 0)
1599 		error = kern_mknod(&nd, uap->mode, uap->dev);
1600 	nlookup_done(&nd);
1601 	return (error);
1602 }
1603 
1604 int
1605 kern_mkfifo(struct nlookupdata *nd, int mode)
1606 {
1607 	struct namecache *ncp;
1608 	struct thread *td = curthread;
1609 	struct proc *p = td->td_proc;
1610 	struct vattr vattr;
1611 	struct vnode *vp;
1612 	int error;
1613 
1614 	bwillwrite();
1615 
1616 	nd->nl_flags |= NLC_CREATE;
1617 	if ((error = nlookup(nd)) != 0)
1618 		return (error);
1619 	ncp = nd->nl_ncp;
1620 	if (ncp->nc_vp)
1621 		return (EEXIST);
1622 	if ((error = ncp_writechk(ncp)) != 0)
1623 		return (error);
1624 
1625 	VATTR_NULL(&vattr);
1626 	vattr.va_type = VFIFO;
1627 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1628 	vp = NULL;
1629 	error = VOP_NMKNOD(ncp, &vp, nd->nl_cred, &vattr);
1630 	if (error == 0)
1631 		vput(vp);
1632 	return (error);
1633 }
1634 
1635 /*
1636  * mkfifo_args(char *path, int mode)
1637  *
1638  * Create a named pipe.
1639  */
1640 int
1641 sys_mkfifo(struct mkfifo_args *uap)
1642 {
1643 	struct nlookupdata nd;
1644 	int error;
1645 
1646 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1647 	if (error == 0)
1648 		error = kern_mkfifo(&nd, uap->mode);
1649 	nlookup_done(&nd);
1650 	return (error);
1651 }
1652 
1653 static int hardlink_check_uid = 0;
1654 SYSCTL_INT(_kern, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1655     &hardlink_check_uid, 0,
1656     "Unprivileged processes cannot create hard links to files owned by other "
1657     "users");
1658 static int hardlink_check_gid = 0;
1659 SYSCTL_INT(_kern, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
1660     &hardlink_check_gid, 0,
1661     "Unprivileged processes cannot create hard links to files owned by other "
1662     "groups");
1663 
1664 static int
1665 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
1666 {
1667 	struct vattr va;
1668 	int error;
1669 
1670 	/*
1671 	 * Shortcut if disabled
1672 	 */
1673 	if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
1674 		return (0);
1675 
1676 	/*
1677 	 * root cred can always hardlink
1678 	 */
1679 	if (suser_cred(cred, PRISON_ROOT) == 0)
1680 		return (0);
1681 
1682 	/*
1683 	 * Otherwise only if the originating file is owned by the
1684 	 * same user or group.  Note that any group is allowed if
1685 	 * the file is owned by the caller.
1686 	 */
1687 	error = VOP_GETATTR(vp, &va);
1688 	if (error != 0)
1689 		return (error);
1690 
1691 	if (hardlink_check_uid) {
1692 		if (cred->cr_uid != va.va_uid)
1693 			return (EPERM);
1694 	}
1695 
1696 	if (hardlink_check_gid) {
1697 		if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
1698 			return (EPERM);
1699 	}
1700 
1701 	return (0);
1702 }
1703 
1704 int
1705 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
1706 {
1707 	struct thread *td = curthread;
1708 	struct vnode *vp;
1709 	int error;
1710 
1711 	/*
1712 	 * Lookup the source and obtained a locked vnode.
1713 	 *
1714 	 * XXX relookup on vget failure / race ?
1715 	 */
1716 	bwillwrite();
1717 	if ((error = nlookup(nd)) != 0)
1718 		return (error);
1719 	vp = nd->nl_ncp->nc_vp;
1720 	KKASSERT(vp != NULL);
1721 	if (vp->v_type == VDIR)
1722 		return (EPERM);		/* POSIX */
1723 	if ((error = ncp_writechk(nd->nl_ncp)) != 0)
1724 		return (error);
1725 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
1726 		return (error);
1727 
1728 	/*
1729 	 * Unlock the source so we can lookup the target without deadlocking
1730 	 * (XXX vp is locked already, possible other deadlock?).  The target
1731 	 * must not exist.
1732 	 */
1733 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
1734 	nd->nl_flags &= ~NLC_NCPISLOCKED;
1735 	cache_unlock(nd->nl_ncp);
1736 
1737 	linknd->nl_flags |= NLC_CREATE;
1738 	if ((error = nlookup(linknd)) != 0) {
1739 		vput(vp);
1740 		return (error);
1741 	}
1742 	if (linknd->nl_ncp->nc_vp) {
1743 		vput(vp);
1744 		return (EEXIST);
1745 	}
1746 
1747 	/*
1748 	 * Finally run the new API VOP.
1749 	 */
1750 	error = can_hardlink(vp, td, td->td_proc->p_ucred);
1751 	if (error == 0)
1752 		error = VOP_NLINK(linknd->nl_ncp, vp, linknd->nl_cred);
1753 	vput(vp);
1754 	return (error);
1755 }
1756 
1757 /*
1758  * link_args(char *path, char *link)
1759  *
1760  * Make a hard file link.
1761  */
1762 int
1763 sys_link(struct link_args *uap)
1764 {
1765 	struct nlookupdata nd, linknd;
1766 	int error;
1767 
1768 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1769 	if (error == 0) {
1770 		error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
1771 		if (error == 0)
1772 			error = kern_link(&nd, &linknd);
1773 		nlookup_done(&linknd);
1774 	}
1775 	nlookup_done(&nd);
1776 	return (error);
1777 }
1778 
1779 int
1780 kern_symlink(struct nlookupdata *nd, char *path, int mode)
1781 {
1782 	struct namecache *ncp;
1783 	struct vattr vattr;
1784 	struct vnode *vp;
1785 	int error;
1786 
1787 	bwillwrite();
1788 	nd->nl_flags |= NLC_CREATE;
1789 	if ((error = nlookup(nd)) != 0)
1790 		return (error);
1791 	ncp = nd->nl_ncp;
1792 	if (ncp->nc_vp)
1793 		return (EEXIST);
1794 	if ((error = ncp_writechk(ncp)) != 0)
1795 		return (error);
1796 	VATTR_NULL(&vattr);
1797 	vattr.va_mode = mode;
1798 	error = VOP_NSYMLINK(ncp, &vp, nd->nl_cred, &vattr, path);
1799 	if (error == 0)
1800 		vput(vp);
1801 	return (error);
1802 }
1803 
1804 /*
1805  * symlink(char *path, char *link)
1806  *
1807  * Make a symbolic link.
1808  */
1809 int
1810 sys_symlink(struct symlink_args *uap)
1811 {
1812 	struct thread *td = curthread;
1813 	struct nlookupdata nd;
1814 	char *path;
1815 	int error;
1816 	int mode;
1817 
1818 	path = objcache_get(namei_oc, M_WAITOK);
1819 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
1820 	if (error == 0) {
1821 		error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
1822 		if (error == 0) {
1823 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
1824 			error = kern_symlink(&nd, path, mode);
1825 		}
1826 		nlookup_done(&nd);
1827 	}
1828 	objcache_put(namei_oc, path);
1829 	return (error);
1830 }
1831 
1832 /*
1833  * undelete_args(char *path)
1834  *
1835  * Delete a whiteout from the filesystem.
1836  */
1837 /* ARGSUSED */
1838 int
1839 sys_undelete(struct undelete_args *uap)
1840 {
1841 	struct nlookupdata nd;
1842 	int error;
1843 
1844 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1845 	bwillwrite();
1846 	nd.nl_flags |= NLC_DELETE;
1847 	if (error == 0)
1848 		error = nlookup(&nd);
1849 	if (error == 0)
1850 		error = ncp_writechk(nd.nl_ncp);
1851 	if (error == 0)
1852 		error = VOP_NWHITEOUT(nd.nl_ncp, nd.nl_cred, NAMEI_DELETE);
1853 	nlookup_done(&nd);
1854 	return (error);
1855 }
1856 
1857 int
1858 kern_unlink(struct nlookupdata *nd)
1859 {
1860 	struct namecache *ncp;
1861 	int error;
1862 
1863 	bwillwrite();
1864 	nd->nl_flags |= NLC_DELETE;
1865 	if ((error = nlookup(nd)) != 0)
1866 		return (error);
1867 	ncp = nd->nl_ncp;
1868 	if ((error = ncp_writechk(ncp)) != 0)
1869 		return (error);
1870 	error = VOP_NREMOVE(ncp, nd->nl_cred);
1871 	return (error);
1872 }
1873 
1874 /*
1875  * unlink_args(char *path)
1876  *
1877  * Delete a name from the filesystem.
1878  */
1879 int
1880 sys_unlink(struct unlink_args *uap)
1881 {
1882 	struct nlookupdata nd;
1883 	int error;
1884 
1885 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1886 	if (error == 0)
1887 		error = kern_unlink(&nd);
1888 	nlookup_done(&nd);
1889 	return (error);
1890 }
1891 
1892 int
1893 kern_lseek(int fd, off_t offset, int whence, off_t *res)
1894 {
1895 	struct thread *td = curthread;
1896 	struct proc *p = td->td_proc;
1897 	struct file *fp;
1898 	struct vattr vattr;
1899 	int error;
1900 
1901 	fp = holdfp(p->p_fd, fd, -1);
1902 	if (fp == NULL)
1903 		return (EBADF);
1904 	if (fp->f_type != DTYPE_VNODE) {
1905 		error = ESPIPE;
1906 		goto done;
1907 	}
1908 
1909 	switch (whence) {
1910 	case L_INCR:
1911 		fp->f_offset += offset;
1912 		error = 0;
1913 		break;
1914 	case L_XTND:
1915 		error = VOP_GETATTR((struct vnode *)fp->f_data, &vattr);
1916 		if (error == 0)
1917 			fp->f_offset = offset + vattr.va_size;
1918 		break;
1919 	case L_SET:
1920 		fp->f_offset = offset;
1921 		error = 0;
1922 		break;
1923 	default:
1924 		error = EINVAL;
1925 		break;
1926 	}
1927 	*res = fp->f_offset;
1928 done:
1929 	fdrop(fp);
1930 	return (error);
1931 }
1932 
1933 /*
1934  * lseek_args(int fd, int pad, off_t offset, int whence)
1935  *
1936  * Reposition read/write file offset.
1937  */
1938 int
1939 sys_lseek(struct lseek_args *uap)
1940 {
1941 	int error;
1942 
1943 	error = kern_lseek(uap->fd, uap->offset, uap->whence,
1944 	    &uap->sysmsg_offset);
1945 
1946 	return (error);
1947 }
1948 
1949 int
1950 kern_access(struct nlookupdata *nd, int aflags)
1951 {
1952 	struct vnode *vp;
1953 	int error, flags;
1954 
1955 	if ((error = nlookup(nd)) != 0)
1956 		return (error);
1957 retry:
1958 	error = cache_vget(nd->nl_ncp, nd->nl_cred, LK_EXCLUSIVE, &vp);
1959 	if (error)
1960 		return (error);
1961 
1962 	/* Flags == 0 means only check for existence. */
1963 	if (aflags) {
1964 		flags = 0;
1965 		if (aflags & R_OK)
1966 			flags |= VREAD;
1967 		if (aflags & W_OK)
1968 			flags |= VWRITE;
1969 		if (aflags & X_OK)
1970 			flags |= VEXEC;
1971 		if ((flags & VWRITE) == 0 ||
1972 		    (error = vn_writechk(vp, nd->nl_ncp)) == 0)
1973 			error = VOP_ACCESS(vp, flags, nd->nl_cred);
1974 
1975 		/*
1976 		 * If the file handle is stale we have to re-resolve the
1977 		 * entry.  This is a hack at the moment.
1978 		 */
1979 		if (error == ESTALE) {
1980 			cache_setunresolved(nd->nl_ncp);
1981 			error = cache_resolve(nd->nl_ncp, nd->nl_cred);
1982 			if (error == 0) {
1983 				vput(vp);
1984 				vp = NULL;
1985 				goto retry;
1986 			}
1987 		}
1988 	}
1989 	vput(vp);
1990 	return (error);
1991 }
1992 
1993 /*
1994  * access_args(char *path, int flags)
1995  *
1996  * Check access permissions.
1997  */
1998 int
1999 sys_access(struct access_args *uap)
2000 {
2001 	struct nlookupdata nd;
2002 	int error;
2003 
2004 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2005 	if (error == 0)
2006 		error = kern_access(&nd, uap->flags);
2007 	nlookup_done(&nd);
2008 	return (error);
2009 }
2010 
2011 int
2012 kern_stat(struct nlookupdata *nd, struct stat *st)
2013 {
2014 	int error;
2015 	struct vnode *vp;
2016 	thread_t td;
2017 
2018 	if ((error = nlookup(nd)) != 0)
2019 		return (error);
2020 again:
2021 	if ((vp = nd->nl_ncp->nc_vp) == NULL)
2022 		return (ENOENT);
2023 
2024 	td = curthread;
2025 	if ((error = vget(vp, LK_SHARED)) != 0)
2026 		return (error);
2027 	error = vn_stat(vp, st, nd->nl_cred);
2028 
2029 	/*
2030 	 * If the file handle is stale we have to re-resolve the entry.  This
2031 	 * is a hack at the moment.
2032 	 */
2033 	if (error == ESTALE) {
2034 		cache_setunresolved(nd->nl_ncp);
2035 		error = cache_resolve(nd->nl_ncp, nd->nl_cred);
2036 		if (error == 0) {
2037 			vput(vp);
2038 			goto again;
2039 		}
2040 	}
2041 	vput(vp);
2042 	return (error);
2043 }
2044 
2045 /*
2046  * stat_args(char *path, struct stat *ub)
2047  *
2048  * Get file status; this version follows links.
2049  */
2050 int
2051 sys_stat(struct stat_args *uap)
2052 {
2053 	struct nlookupdata nd;
2054 	struct stat st;
2055 	int error;
2056 
2057 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2058 	if (error == 0) {
2059 		error = kern_stat(&nd, &st);
2060 		if (error == 0)
2061 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2062 	}
2063 	nlookup_done(&nd);
2064 	return (error);
2065 }
2066 
2067 /*
2068  * lstat_args(char *path, struct stat *ub)
2069  *
2070  * Get file status; this version does not follow links.
2071  */
2072 int
2073 sys_lstat(struct lstat_args *uap)
2074 {
2075 	struct nlookupdata nd;
2076 	struct stat st;
2077 	int error;
2078 
2079 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2080 	if (error == 0) {
2081 		error = kern_stat(&nd, &st);
2082 		if (error == 0)
2083 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2084 	}
2085 	nlookup_done(&nd);
2086 	return (error);
2087 }
2088 
2089 /*
2090  * pathconf_Args(char *path, int name)
2091  *
2092  * Get configurable pathname variables.
2093  */
2094 /* ARGSUSED */
2095 int
2096 sys_pathconf(struct pathconf_args *uap)
2097 {
2098 	struct nlookupdata nd;
2099 	struct vnode *vp;
2100 	int error;
2101 
2102 	vp = NULL;
2103 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2104 	if (error == 0)
2105 		error = nlookup(&nd);
2106 	if (error == 0)
2107 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
2108 	nlookup_done(&nd);
2109 	if (error == 0) {
2110 		error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
2111 		vput(vp);
2112 	}
2113 	return (error);
2114 }
2115 
2116 /*
2117  * XXX: daver
2118  * kern_readlink isn't properly split yet.  There is a copyin burried
2119  * in VOP_READLINK().
2120  */
2121 int
2122 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2123 {
2124 	struct thread *td = curthread;
2125 	struct proc *p = td->td_proc;
2126 	struct vnode *vp;
2127 	struct iovec aiov;
2128 	struct uio auio;
2129 	int error;
2130 
2131 	if ((error = nlookup(nd)) != 0)
2132 		return (error);
2133 	error = cache_vget(nd->nl_ncp, nd->nl_cred, LK_EXCLUSIVE, &vp);
2134 	if (error)
2135 		return (error);
2136 	if (vp->v_type != VLNK) {
2137 		error = EINVAL;
2138 	} else {
2139 		aiov.iov_base = buf;
2140 		aiov.iov_len = count;
2141 		auio.uio_iov = &aiov;
2142 		auio.uio_iovcnt = 1;
2143 		auio.uio_offset = 0;
2144 		auio.uio_rw = UIO_READ;
2145 		auio.uio_segflg = UIO_USERSPACE;
2146 		auio.uio_td = td;
2147 		auio.uio_resid = count;
2148 		error = VOP_READLINK(vp, &auio, p->p_ucred);
2149 	}
2150 	vput(vp);
2151 	*res = count - auio.uio_resid;
2152 	return (error);
2153 }
2154 
2155 /*
2156  * readlink_args(char *path, char *buf, int count)
2157  *
2158  * Return target name of a symbolic link.
2159  */
2160 int
2161 sys_readlink(struct readlink_args *uap)
2162 {
2163 	struct nlookupdata nd;
2164 	int error;
2165 
2166 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2167 	if (error == 0) {
2168 		error = kern_readlink(&nd, uap->buf, uap->count,
2169 					&uap->sysmsg_result);
2170 	}
2171 	nlookup_done(&nd);
2172 	return (error);
2173 }
2174 
2175 static int
2176 setfflags(struct vnode *vp, int flags)
2177 {
2178 	struct thread *td = curthread;
2179 	struct proc *p = td->td_proc;
2180 	int error;
2181 	struct vattr vattr;
2182 
2183 	/*
2184 	 * Prevent non-root users from setting flags on devices.  When
2185 	 * a device is reused, users can retain ownership of the device
2186 	 * if they are allowed to set flags and programs assume that
2187 	 * chown can't fail when done as root.
2188 	 */
2189 	if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2190 	    ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0))
2191 		return (error);
2192 
2193 	/*
2194 	 * note: vget is required for any operation that might mod the vnode
2195 	 * so VINACTIVE is properly cleared.
2196 	 */
2197 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2198 		VATTR_NULL(&vattr);
2199 		vattr.va_flags = flags;
2200 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2201 		vput(vp);
2202 	}
2203 	return (error);
2204 }
2205 
2206 /*
2207  * chflags(char *path, int flags)
2208  *
2209  * Change flags of a file given a path name.
2210  */
2211 /* ARGSUSED */
2212 int
2213 sys_chflags(struct chflags_args *uap)
2214 {
2215 	struct nlookupdata nd;
2216 	struct vnode *vp;
2217 	int error;
2218 
2219 	vp = NULL;
2220 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2221 	/* XXX Add NLC flag indicating modifying operation? */
2222 	if (error == 0)
2223 		error = nlookup(&nd);
2224 	if (error == 0)
2225 		error = ncp_writechk(nd.nl_ncp);
2226 	if (error == 0)
2227 		error = cache_vref(nd.nl_ncp, nd.nl_cred, &vp);
2228 	nlookup_done(&nd);
2229 	if (error == 0) {
2230 		error = setfflags(vp, uap->flags);
2231 		vrele(vp);
2232 	}
2233 	return (error);
2234 }
2235 
2236 /*
2237  * fchflags_args(int fd, int flags)
2238  *
2239  * Change flags of a file given a file descriptor.
2240  */
2241 /* ARGSUSED */
2242 int
2243 sys_fchflags(struct fchflags_args *uap)
2244 {
2245 	struct thread *td = curthread;
2246 	struct proc *p = td->td_proc;
2247 	struct file *fp;
2248 	int error;
2249 
2250 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2251 		return (error);
2252 	if (fp->f_ncp)
2253 		error = ncp_writechk(fp->f_ncp);
2254 	if (error == 0)
2255 		error = setfflags((struct vnode *) fp->f_data, uap->flags);
2256 	fdrop(fp);
2257 	return (error);
2258 }
2259 
2260 static int
2261 setfmode(struct vnode *vp, int mode)
2262 {
2263 	struct thread *td = curthread;
2264 	struct proc *p = td->td_proc;
2265 	int error;
2266 	struct vattr vattr;
2267 
2268 	/*
2269 	 * note: vget is required for any operation that might mod the vnode
2270 	 * so VINACTIVE is properly cleared.
2271 	 */
2272 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2273 		VATTR_NULL(&vattr);
2274 		vattr.va_mode = mode & ALLPERMS;
2275 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2276 		vput(vp);
2277 	}
2278 	return error;
2279 }
2280 
2281 int
2282 kern_chmod(struct nlookupdata *nd, int mode)
2283 {
2284 	struct vnode *vp;
2285 	int error;
2286 
2287 	/* XXX Add NLC flag indicating modifying operation? */
2288 	if ((error = nlookup(nd)) != 0)
2289 		return (error);
2290 	if ((error = cache_vref(nd->nl_ncp, nd->nl_cred, &vp)) != 0)
2291 		return (error);
2292 	if ((error = ncp_writechk(nd->nl_ncp)) == 0)
2293 		error = setfmode(vp, mode);
2294 	vrele(vp);
2295 	return (error);
2296 }
2297 
2298 /*
2299  * chmod_args(char *path, int mode)
2300  *
2301  * Change mode of a file given path name.
2302  */
2303 /* ARGSUSED */
2304 int
2305 sys_chmod(struct chmod_args *uap)
2306 {
2307 	struct nlookupdata nd;
2308 	int error;
2309 
2310 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2311 	if (error == 0)
2312 		error = kern_chmod(&nd, uap->mode);
2313 	nlookup_done(&nd);
2314 	return (error);
2315 }
2316 
2317 /*
2318  * lchmod_args(char *path, int mode)
2319  *
2320  * Change mode of a file given path name (don't follow links.)
2321  */
2322 /* ARGSUSED */
2323 int
2324 sys_lchmod(struct lchmod_args *uap)
2325 {
2326 	struct nlookupdata nd;
2327 	int error;
2328 
2329 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2330 	if (error == 0)
2331 		error = kern_chmod(&nd, uap->mode);
2332 	nlookup_done(&nd);
2333 	return (error);
2334 }
2335 
2336 /*
2337  * fchmod_args(int fd, int mode)
2338  *
2339  * Change mode of a file given a file descriptor.
2340  */
2341 /* ARGSUSED */
2342 int
2343 sys_fchmod(struct fchmod_args *uap)
2344 {
2345 	struct thread *td = curthread;
2346 	struct proc *p = td->td_proc;
2347 	struct file *fp;
2348 	int error;
2349 
2350 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2351 		return (error);
2352 	if (fp->f_ncp)
2353 		error = ncp_writechk(fp->f_ncp);
2354 	if (error == 0)
2355 		error = setfmode((struct vnode *)fp->f_data, uap->mode);
2356 	fdrop(fp);
2357 	return (error);
2358 }
2359 
2360 static int
2361 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2362 {
2363 	struct thread *td = curthread;
2364 	struct proc *p = td->td_proc;
2365 	int error;
2366 	struct vattr vattr;
2367 
2368 	/*
2369 	 * note: vget is required for any operation that might mod the vnode
2370 	 * so VINACTIVE is properly cleared.
2371 	 */
2372 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2373 		VATTR_NULL(&vattr);
2374 		vattr.va_uid = uid;
2375 		vattr.va_gid = gid;
2376 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2377 		vput(vp);
2378 	}
2379 	return error;
2380 }
2381 
2382 int
2383 kern_chown(struct nlookupdata *nd, int uid, int gid)
2384 {
2385 	struct vnode *vp;
2386 	int error;
2387 
2388 	/* XXX Add NLC flag indicating modifying operation? */
2389 	if ((error = nlookup(nd)) != 0)
2390 		return (error);
2391 	if ((error = cache_vref(nd->nl_ncp, nd->nl_cred, &vp)) != 0)
2392 		return (error);
2393 	if ((error = ncp_writechk(nd->nl_ncp)) == 0)
2394 		error = setfown(vp, uid, gid);
2395 	vrele(vp);
2396 	return (error);
2397 }
2398 
2399 /*
2400  * chown(char *path, int uid, int gid)
2401  *
2402  * Set ownership given a path name.
2403  */
2404 int
2405 sys_chown(struct chown_args *uap)
2406 {
2407 	struct nlookupdata nd;
2408 	int error;
2409 
2410 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2411 	if (error == 0)
2412 		error = kern_chown(&nd, uap->uid, uap->gid);
2413 	nlookup_done(&nd);
2414 	return (error);
2415 }
2416 
2417 /*
2418  * lchown_args(char *path, int uid, int gid)
2419  *
2420  * Set ownership given a path name, do not cross symlinks.
2421  */
2422 int
2423 sys_lchown(struct lchown_args *uap)
2424 {
2425 	struct nlookupdata nd;
2426 	int error;
2427 
2428 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2429 	if (error == 0)
2430 		error = kern_chown(&nd, uap->uid, uap->gid);
2431 	nlookup_done(&nd);
2432 	return (error);
2433 }
2434 
2435 /*
2436  * fchown_args(int fd, int uid, int gid)
2437  *
2438  * Set ownership given a file descriptor.
2439  */
2440 /* ARGSUSED */
2441 int
2442 sys_fchown(struct fchown_args *uap)
2443 {
2444 	struct thread *td = curthread;
2445 	struct proc *p = td->td_proc;
2446 	struct file *fp;
2447 	int error;
2448 
2449 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2450 		return (error);
2451 	if (fp->f_ncp)
2452 		error = ncp_writechk(fp->f_ncp);
2453 	if (error == 0)
2454 		error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
2455 	fdrop(fp);
2456 	return (error);
2457 }
2458 
2459 static int
2460 getutimes(const struct timeval *tvp, struct timespec *tsp)
2461 {
2462 	struct timeval tv[2];
2463 
2464 	if (tvp == NULL) {
2465 		microtime(&tv[0]);
2466 		TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
2467 		tsp[1] = tsp[0];
2468 	} else {
2469 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
2470 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
2471 	}
2472 	return 0;
2473 }
2474 
2475 static int
2476 setutimes(struct vnode *vp, const struct timespec *ts, int nullflag)
2477 {
2478 	struct thread *td = curthread;
2479 	struct proc *p = td->td_proc;
2480 	int error;
2481 	struct vattr vattr;
2482 
2483 	/*
2484 	 * note: vget is required for any operation that might mod the vnode
2485 	 * so VINACTIVE is properly cleared.
2486 	 */
2487 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2488 		VATTR_NULL(&vattr);
2489 		vattr.va_atime = ts[0];
2490 		vattr.va_mtime = ts[1];
2491 		if (nullflag)
2492 			vattr.va_vaflags |= VA_UTIMES_NULL;
2493 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2494 		vput(vp);
2495 	}
2496 	return error;
2497 }
2498 
2499 int
2500 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
2501 {
2502 	struct timespec ts[2];
2503 	struct vnode *vp;
2504 	int error;
2505 
2506 	if ((error = getutimes(tptr, ts)) != 0)
2507 		return (error);
2508 	/* XXX Add NLC flag indicating modifying operation? */
2509 	if ((error = nlookup(nd)) != 0)
2510 		return (error);
2511 	if ((error = ncp_writechk(nd->nl_ncp)) != 0)
2512 		return (error);
2513 	if ((error = cache_vref(nd->nl_ncp, nd->nl_cred, &vp)) != 0)
2514 		return (error);
2515 	error = setutimes(vp, ts, tptr == NULL);
2516 	vrele(vp);
2517 	return (error);
2518 }
2519 
2520 /*
2521  * utimes_args(char *path, struct timeval *tptr)
2522  *
2523  * Set the access and modification times of a file.
2524  */
2525 int
2526 sys_utimes(struct utimes_args *uap)
2527 {
2528 	struct timeval tv[2];
2529 	struct nlookupdata nd;
2530 	int error;
2531 
2532 	if (uap->tptr) {
2533  		error = copyin(uap->tptr, tv, sizeof(tv));
2534 		if (error)
2535 			return (error);
2536 	}
2537 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2538 	if (error == 0)
2539 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2540 	nlookup_done(&nd);
2541 	return (error);
2542 }
2543 
2544 /*
2545  * lutimes_args(char *path, struct timeval *tptr)
2546  *
2547  * Set the access and modification times of a file.
2548  */
2549 int
2550 sys_lutimes(struct lutimes_args *uap)
2551 {
2552 	struct timeval tv[2];
2553 	struct nlookupdata nd;
2554 	int error;
2555 
2556 	if (uap->tptr) {
2557 		error = copyin(uap->tptr, tv, sizeof(tv));
2558 		if (error)
2559 			return (error);
2560 	}
2561 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2562 	if (error == 0)
2563 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2564 	nlookup_done(&nd);
2565 	return (error);
2566 }
2567 
2568 int
2569 kern_futimes(int fd, struct timeval *tptr)
2570 {
2571 	struct thread *td = curthread;
2572 	struct proc *p = td->td_proc;
2573 	struct timespec ts[2];
2574 	struct file *fp;
2575 	int error;
2576 
2577 	error = getutimes(tptr, ts);
2578 	if (error)
2579 		return (error);
2580 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2581 		return (error);
2582 	if (fp->f_ncp)
2583 		error = ncp_writechk(fp->f_ncp);
2584 	if (error == 0)
2585 		error =  setutimes((struct vnode *)fp->f_data, ts, tptr == NULL);
2586 	fdrop(fp);
2587 	return (error);
2588 }
2589 
2590 /*
2591  * futimes_args(int fd, struct timeval *tptr)
2592  *
2593  * Set the access and modification times of a file.
2594  */
2595 int
2596 sys_futimes(struct futimes_args *uap)
2597 {
2598 	struct timeval tv[2];
2599 	int error;
2600 
2601 	if (uap->tptr) {
2602 		error = copyin(uap->tptr, tv, sizeof(tv));
2603 		if (error)
2604 			return (error);
2605 	}
2606 
2607 	error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
2608 
2609 	return (error);
2610 }
2611 
2612 int
2613 kern_truncate(struct nlookupdata *nd, off_t length)
2614 {
2615 	struct vnode *vp;
2616 	struct vattr vattr;
2617 	int error;
2618 
2619 	if (length < 0)
2620 		return(EINVAL);
2621 	/* XXX Add NLC flag indicating modifying operation? */
2622 	if ((error = nlookup(nd)) != 0)
2623 		return (error);
2624 	if ((error = ncp_writechk(nd->nl_ncp)) != 0)
2625 		return (error);
2626 	if ((error = cache_vref(nd->nl_ncp, nd->nl_cred, &vp)) != 0)
2627 		return (error);
2628 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
2629 		vrele(vp);
2630 		return (error);
2631 	}
2632 	if (vp->v_type == VDIR) {
2633 		error = EISDIR;
2634 	} else if ((error = vn_writechk(vp, nd->nl_ncp)) == 0 &&
2635 	    (error = VOP_ACCESS(vp, VWRITE, nd->nl_cred)) == 0) {
2636 		VATTR_NULL(&vattr);
2637 		vattr.va_size = length;
2638 		error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
2639 	}
2640 	vput(vp);
2641 	return (error);
2642 }
2643 
2644 /*
2645  * truncate(char *path, int pad, off_t length)
2646  *
2647  * Truncate a file given its path name.
2648  */
2649 int
2650 sys_truncate(struct truncate_args *uap)
2651 {
2652 	struct nlookupdata nd;
2653 	int error;
2654 
2655 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2656 	if (error == 0)
2657 		error = kern_truncate(&nd, uap->length);
2658 	nlookup_done(&nd);
2659 	return error;
2660 }
2661 
2662 int
2663 kern_ftruncate(int fd, off_t length)
2664 {
2665 	struct thread *td = curthread;
2666 	struct proc *p = td->td_proc;
2667 	struct vattr vattr;
2668 	struct vnode *vp;
2669 	struct file *fp;
2670 	int error;
2671 
2672 	if (length < 0)
2673 		return(EINVAL);
2674 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2675 		return (error);
2676 	if (fp->f_ncp) {
2677 		error = ncp_writechk(fp->f_ncp);
2678 		if (error)
2679 			goto done;
2680 	}
2681 	if ((fp->f_flag & FWRITE) == 0) {
2682 		error = EINVAL;
2683 		goto done;
2684 	}
2685 	vp = (struct vnode *)fp->f_data;
2686 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2687 	if (vp->v_type == VDIR) {
2688 		error = EISDIR;
2689 	} else if ((error = vn_writechk(vp, NULL)) == 0) {
2690 		VATTR_NULL(&vattr);
2691 		vattr.va_size = length;
2692 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
2693 	}
2694 	vn_unlock(vp);
2695 done:
2696 	fdrop(fp);
2697 	return (error);
2698 }
2699 
2700 /*
2701  * ftruncate_args(int fd, int pad, off_t length)
2702  *
2703  * Truncate a file given a file descriptor.
2704  */
2705 int
2706 sys_ftruncate(struct ftruncate_args *uap)
2707 {
2708 	int error;
2709 
2710 	error = kern_ftruncate(uap->fd, uap->length);
2711 
2712 	return (error);
2713 }
2714 
2715 /*
2716  * fsync(int fd)
2717  *
2718  * Sync an open file.
2719  */
2720 /* ARGSUSED */
2721 int
2722 sys_fsync(struct fsync_args *uap)
2723 {
2724 	struct thread *td = curthread;
2725 	struct proc *p = td->td_proc;
2726 	struct vnode *vp;
2727 	struct file *fp;
2728 	vm_object_t obj;
2729 	int error;
2730 
2731 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2732 		return (error);
2733 	vp = (struct vnode *)fp->f_data;
2734 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2735 	if ((obj = vp->v_object) != NULL)
2736 		vm_object_page_clean(obj, 0, 0, 0);
2737 	if ((error = VOP_FSYNC(vp, MNT_WAIT)) == 0 &&
2738 	    vp->v_mount && (vp->v_mount->mnt_flag & MNT_SOFTDEP) &&
2739 	    bioops.io_fsync) {
2740 		error = (*bioops.io_fsync)(vp);
2741 	}
2742 	vn_unlock(vp);
2743 	fdrop(fp);
2744 	return (error);
2745 }
2746 
2747 int
2748 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
2749 {
2750 	struct namecache *fncpd;
2751 	struct namecache *tncpd;
2752 	struct namecache *ncp;
2753 	struct mount *mp;
2754 	int error;
2755 
2756 	bwillwrite();
2757 	if ((error = nlookup(fromnd)) != 0)
2758 		return (error);
2759 	if ((fncpd = fromnd->nl_ncp->nc_parent) == NULL)
2760 		return (ENOENT);
2761 	cache_hold(fncpd);
2762 
2763 	/*
2764 	 * unlock the source ncp so we can lookup the target ncp without
2765 	 * deadlocking.  The target may or may not exist so we do not check
2766 	 * for a target vp like kern_mkdir() and other creation functions do.
2767 	 *
2768 	 * The source and target directories are ref'd and rechecked after
2769 	 * everything is relocked to determine if the source or target file
2770 	 * has been renamed.
2771 	 */
2772 	KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
2773 	fromnd->nl_flags &= ~NLC_NCPISLOCKED;
2774 	cache_unlock(fromnd->nl_ncp);
2775 
2776 	tond->nl_flags |= NLC_CREATE;
2777 	if ((error = nlookup(tond)) != 0) {
2778 		cache_drop(fncpd);
2779 		return (error);
2780 	}
2781 	if ((tncpd = tond->nl_ncp->nc_parent) == NULL) {
2782 		cache_drop(fncpd);
2783 		return (ENOENT);
2784 	}
2785 	cache_hold(tncpd);
2786 
2787 	/*
2788 	 * If the source and target are the same there is nothing to do
2789 	 */
2790 	if (fromnd->nl_ncp == tond->nl_ncp) {
2791 		cache_drop(fncpd);
2792 		cache_drop(tncpd);
2793 		return (0);
2794 	}
2795 
2796 	/*
2797 	 * Mount points cannot be renamed or overwritten
2798 	 */
2799 	if ((fromnd->nl_ncp->nc_flag | tond->nl_ncp->nc_flag) &
2800 	    (NCF_MOUNTPT|NCF_MOUNTEDHERE)
2801 	) {
2802 		cache_drop(fncpd);
2803 		cache_drop(tncpd);
2804 		return (EINVAL);
2805 	}
2806 
2807 	/*
2808 	 * relock the source ncp.  NOTE AFTER RELOCKING: the source ncp
2809 	 * may have become invalid while it was unlocked, nc_vp and nc_mount
2810 	 * could be NULL.
2811 	 */
2812 	if (cache_lock_nonblock(fromnd->nl_ncp) == 0) {
2813 		cache_resolve(fromnd->nl_ncp, fromnd->nl_cred);
2814 	} else if (fromnd->nl_ncp > tond->nl_ncp) {
2815 		cache_lock(fromnd->nl_ncp);
2816 		cache_resolve(fromnd->nl_ncp, fromnd->nl_cred);
2817 	} else {
2818 		cache_unlock(tond->nl_ncp);
2819 		cache_lock(fromnd->nl_ncp);
2820 		cache_resolve(fromnd->nl_ncp, fromnd->nl_cred);
2821 		cache_lock(tond->nl_ncp);
2822 		cache_resolve(tond->nl_ncp, tond->nl_cred);
2823 	}
2824 	fromnd->nl_flags |= NLC_NCPISLOCKED;
2825 
2826 	/*
2827 	 * make sure the parent directories linkages are the same
2828 	 */
2829 	if (fncpd != fromnd->nl_ncp->nc_parent ||
2830 	    tncpd != tond->nl_ncp->nc_parent) {
2831 		cache_drop(fncpd);
2832 		cache_drop(tncpd);
2833 		return (ENOENT);
2834 	}
2835 
2836 	/*
2837 	 * Both the source and target must be within the same filesystem and
2838 	 * in the same filesystem as their parent directories within the
2839 	 * namecache topology.
2840 	 *
2841 	 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
2842 	 */
2843 	mp = fncpd->nc_mount;
2844 	if (mp != tncpd->nc_mount || mp != fromnd->nl_ncp->nc_mount ||
2845 	    mp != tond->nl_ncp->nc_mount) {
2846 		cache_drop(fncpd);
2847 		cache_drop(tncpd);
2848 		return (EXDEV);
2849 	}
2850 
2851 	/*
2852 	 * Make sure the mount point is writable
2853 	 */
2854 	if ((error = ncp_writechk(tond->nl_ncp)) != 0) {
2855 		cache_drop(fncpd);
2856 		cache_drop(tncpd);
2857 		return (error);
2858 	}
2859 
2860 	/*
2861 	 * If the target exists and either the source or target is a directory,
2862 	 * then both must be directories.
2863 	 *
2864 	 * Due to relocking of the source, fromnd->nl_ncp->nc_vp might have
2865 	 * become NULL.
2866 	 */
2867 	if (tond->nl_ncp->nc_vp) {
2868 		if (fromnd->nl_ncp->nc_vp == NULL) {
2869 			error = ENOENT;
2870 		} else if (fromnd->nl_ncp->nc_vp->v_type == VDIR) {
2871 			if (tond->nl_ncp->nc_vp->v_type != VDIR)
2872 				error = ENOTDIR;
2873 		} else if (tond->nl_ncp->nc_vp->v_type == VDIR) {
2874 			error = EISDIR;
2875 		}
2876 	}
2877 
2878 	/*
2879 	 * You cannot rename a source into itself or a subdirectory of itself.
2880 	 * We check this by travsersing the target directory upwards looking
2881 	 * for a match against the source.
2882 	 */
2883 	if (error == 0) {
2884 		for (ncp = tncpd; ncp; ncp = ncp->nc_parent) {
2885 			if (fromnd->nl_ncp == ncp) {
2886 				error = EINVAL;
2887 				break;
2888 			}
2889 		}
2890 	}
2891 
2892 	cache_drop(fncpd);
2893 	cache_drop(tncpd);
2894 
2895 	/*
2896 	 * Even though the namespaces are different, they may still represent
2897 	 * hardlinks to the same file.  The filesystem might have a hard time
2898 	 * with this so we issue a NREMOVE of the source instead of a NRENAME
2899 	 * when we detect the situation.
2900 	 */
2901 	if (error == 0) {
2902 		if (fromnd->nl_ncp->nc_vp == tond->nl_ncp->nc_vp) {
2903 			error = VOP_NREMOVE(fromnd->nl_ncp, fromnd->nl_cred);
2904 		} else {
2905 			error = VOP_NRENAME(fromnd->nl_ncp, tond->nl_ncp,
2906 					    tond->nl_cred);
2907 		}
2908 	}
2909 	return (error);
2910 }
2911 
2912 /*
2913  * rename_args(char *from, char *to)
2914  *
2915  * Rename files.  Source and destination must either both be directories,
2916  * or both not be directories.  If target is a directory, it must be empty.
2917  */
2918 int
2919 sys_rename(struct rename_args *uap)
2920 {
2921 	struct nlookupdata fromnd, tond;
2922 	int error;
2923 
2924 	error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
2925 	if (error == 0) {
2926 		error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
2927 		if (error == 0)
2928 			error = kern_rename(&fromnd, &tond);
2929 		nlookup_done(&tond);
2930 	}
2931 	nlookup_done(&fromnd);
2932 	return (error);
2933 }
2934 
2935 int
2936 kern_mkdir(struct nlookupdata *nd, int mode)
2937 {
2938 	struct thread *td = curthread;
2939 	struct proc *p = td->td_proc;
2940 	struct namecache *ncp;
2941 	struct vnode *vp;
2942 	struct vattr vattr;
2943 	int error;
2944 
2945 	bwillwrite();
2946 	nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE;
2947 	if ((error = nlookup(nd)) != 0)
2948 		return (error);
2949 
2950 	ncp = nd->nl_ncp;
2951 	if (ncp->nc_vp)
2952 		return (EEXIST);
2953 	if ((error = ncp_writechk(ncp)) != 0)
2954 		return (error);
2955 
2956 	VATTR_NULL(&vattr);
2957 	vattr.va_type = VDIR;
2958 	vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
2959 
2960 	vp = NULL;
2961 	error = VOP_NMKDIR(ncp, &vp, p->p_ucred, &vattr);
2962 	if (error == 0)
2963 		vput(vp);
2964 	return (error);
2965 }
2966 
2967 /*
2968  * mkdir_args(char *path, int mode)
2969  *
2970  * Make a directory file.
2971  */
2972 /* ARGSUSED */
2973 int
2974 sys_mkdir(struct mkdir_args *uap)
2975 {
2976 	struct nlookupdata nd;
2977 	int error;
2978 
2979 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2980 	if (error == 0)
2981 		error = kern_mkdir(&nd, uap->mode);
2982 	nlookup_done(&nd);
2983 	return (error);
2984 }
2985 
2986 int
2987 kern_rmdir(struct nlookupdata *nd)
2988 {
2989 	struct namecache *ncp;
2990 	int error;
2991 
2992 	bwillwrite();
2993 	nd->nl_flags |= NLC_DELETE;
2994 	if ((error = nlookup(nd)) != 0)
2995 		return (error);
2996 
2997 	/*
2998 	 * Do not allow directories representing mount points to be
2999 	 * deleted, even if empty.  Check write perms on mount point
3000 	 * in case the vnode is aliased (aka nullfs).
3001 	 */
3002 	if (nd->nl_ncp->nc_flag & (NCF_MOUNTEDHERE|NCF_MOUNTPT))
3003 		return (EINVAL);
3004 	if ((error = ncp_writechk(nd->nl_ncp)) != 0)
3005 		return (error);
3006 
3007 	ncp = nd->nl_ncp;
3008 	error = VOP_NRMDIR(ncp, nd->nl_cred);
3009 	return (error);
3010 }
3011 
3012 /*
3013  * rmdir_args(char *path)
3014  *
3015  * Remove a directory file.
3016  */
3017 /* ARGSUSED */
3018 int
3019 sys_rmdir(struct rmdir_args *uap)
3020 {
3021 	struct nlookupdata nd;
3022 	int error;
3023 
3024 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3025 	if (error == 0)
3026 		error = kern_rmdir(&nd);
3027 	nlookup_done(&nd);
3028 	return (error);
3029 }
3030 
3031 int
3032 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3033     enum uio_seg direction)
3034 {
3035 	struct thread *td = curthread;
3036 	struct proc *p = td->td_proc;
3037 	struct vnode *vp;
3038 	struct file *fp;
3039 	struct uio auio;
3040 	struct iovec aiov;
3041 	long loff;
3042 	int error, eofflag;
3043 
3044 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3045 		return (error);
3046 	if ((fp->f_flag & FREAD) == 0) {
3047 		error = EBADF;
3048 		goto done;
3049 	}
3050 	vp = (struct vnode *)fp->f_data;
3051 unionread:
3052 	if (vp->v_type != VDIR) {
3053 		error = EINVAL;
3054 		goto done;
3055 	}
3056 	aiov.iov_base = buf;
3057 	aiov.iov_len = count;
3058 	auio.uio_iov = &aiov;
3059 	auio.uio_iovcnt = 1;
3060 	auio.uio_rw = UIO_READ;
3061 	auio.uio_segflg = direction;
3062 	auio.uio_td = td;
3063 	auio.uio_resid = count;
3064 	loff = auio.uio_offset = fp->f_offset;
3065 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3066 	fp->f_offset = auio.uio_offset;
3067 	if (error)
3068 		goto done;
3069 	if (count == auio.uio_resid) {
3070 		if (union_dircheckp) {
3071 			error = union_dircheckp(td, &vp, fp);
3072 			if (error == -1)
3073 				goto unionread;
3074 			if (error)
3075 				goto done;
3076 		}
3077 #if 0
3078 		if ((vp->v_flag & VROOT) &&
3079 		    (vp->v_mount->mnt_flag & MNT_UNION)) {
3080 			struct vnode *tvp = vp;
3081 			vp = vp->v_mount->mnt_vnodecovered;
3082 			vref(vp);
3083 			fp->f_data = vp;
3084 			fp->f_offset = 0;
3085 			vrele(tvp);
3086 			goto unionread;
3087 		}
3088 #endif
3089 	}
3090 	if (basep) {
3091 		*basep = loff;
3092 	}
3093 	*res = count - auio.uio_resid;
3094 done:
3095 	fdrop(fp);
3096 	return (error);
3097 }
3098 
3099 /*
3100  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3101  *
3102  * Read a block of directory entries in a file system independent format.
3103  */
3104 int
3105 sys_getdirentries(struct getdirentries_args *uap)
3106 {
3107 	long base;
3108 	int error;
3109 
3110 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3111 	    &uap->sysmsg_result, UIO_USERSPACE);
3112 
3113 	if (error == 0)
3114 		error = copyout(&base, uap->basep, sizeof(*uap->basep));
3115 	return (error);
3116 }
3117 
3118 /*
3119  * getdents_args(int fd, char *buf, size_t count)
3120  */
3121 int
3122 sys_getdents(struct getdents_args *uap)
3123 {
3124 	int error;
3125 
3126 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3127 	    &uap->sysmsg_result, UIO_USERSPACE);
3128 
3129 	return (error);
3130 }
3131 
3132 /*
3133  * umask(int newmask)
3134  *
3135  * Set the mode mask for creation of filesystem nodes.
3136  *
3137  * MP SAFE
3138  */
3139 int
3140 sys_umask(struct umask_args *uap)
3141 {
3142 	struct thread *td = curthread;
3143 	struct proc *p = td->td_proc;
3144 	struct filedesc *fdp;
3145 
3146 	fdp = p->p_fd;
3147 	uap->sysmsg_result = fdp->fd_cmask;
3148 	fdp->fd_cmask = uap->newmask & ALLPERMS;
3149 	return (0);
3150 }
3151 
3152 /*
3153  * revoke(char *path)
3154  *
3155  * Void all references to file by ripping underlying filesystem
3156  * away from vnode.
3157  */
3158 /* ARGSUSED */
3159 int
3160 sys_revoke(struct revoke_args *uap)
3161 {
3162 	struct nlookupdata nd;
3163 	struct vattr vattr;
3164 	struct vnode *vp;
3165 	struct ucred *cred;
3166 	int error;
3167 
3168 	vp = NULL;
3169 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3170 	if (error == 0)
3171 		error = nlookup(&nd);
3172 	if (error == 0)
3173 		error = cache_vref(nd.nl_ncp, nd.nl_cred, &vp);
3174 	cred = crhold(nd.nl_cred);
3175 	nlookup_done(&nd);
3176 	if (error == 0) {
3177 		if (vp->v_type != VCHR && vp->v_type != VBLK)
3178 			error = EINVAL;
3179 		if (error == 0)
3180 			error = VOP_GETATTR(vp, &vattr);
3181 		if (error == 0 && cred->cr_uid != vattr.va_uid)
3182 			error = suser_cred(cred, PRISON_ROOT);
3183 		if (error == 0 && count_udev(vp->v_udev) > 0) {
3184 			error = 0;
3185 			vx_lock(vp);
3186 			VOP_REVOKE(vp, REVOKEALL);
3187 			vx_unlock(vp);
3188 		}
3189 		vrele(vp);
3190 	}
3191 	if (cred)
3192 		crfree(cred);
3193 	return (error);
3194 }
3195 
3196 /*
3197  * getfh_args(char *fname, fhandle_t *fhp)
3198  *
3199  * Get (NFS) file handle
3200  */
3201 int
3202 sys_getfh(struct getfh_args *uap)
3203 {
3204 	struct thread *td = curthread;
3205 	struct nlookupdata nd;
3206 	fhandle_t fh;
3207 	struct vnode *vp;
3208 	int error;
3209 
3210 	/*
3211 	 * Must be super user
3212 	 */
3213 	if ((error = suser(td)) != 0)
3214 		return (error);
3215 
3216 	vp = NULL;
3217 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3218 	if (error == 0)
3219 		error = nlookup(&nd);
3220 	if (error == 0)
3221 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
3222 	nlookup_done(&nd);
3223 	if (error == 0) {
3224 		bzero(&fh, sizeof(fh));
3225 		fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
3226 		error = VFS_VPTOFH(vp, &fh.fh_fid);
3227 		vput(vp);
3228 		if (error == 0)
3229 			error = copyout(&fh, uap->fhp, sizeof(fh));
3230 	}
3231 	return (error);
3232 }
3233 
3234 /*
3235  * fhopen_args(const struct fhandle *u_fhp, int flags)
3236  *
3237  * syscall for the rpc.lockd to use to translate a NFS file handle into
3238  * an open descriptor.
3239  *
3240  * warning: do not remove the suser() call or this becomes one giant
3241  * security hole.
3242  */
3243 int
3244 sys_fhopen(struct fhopen_args *uap)
3245 {
3246 	struct thread *td = curthread;
3247 	struct proc *p = td->td_proc;
3248 	struct mount *mp;
3249 	struct vnode *vp;
3250 	struct fhandle fhp;
3251 	struct vattr vat;
3252 	struct vattr *vap = &vat;
3253 	struct flock lf;
3254 	int fmode, mode, error, type;
3255 	struct file *nfp;
3256 	struct file *fp;
3257 	int indx;
3258 
3259 	/*
3260 	 * Must be super user
3261 	 */
3262 	error = suser(td);
3263 	if (error)
3264 		return (error);
3265 
3266 	fmode = FFLAGS(uap->flags);
3267 	/* why not allow a non-read/write open for our lockd? */
3268 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3269 		return (EINVAL);
3270 	error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3271 	if (error)
3272 		return(error);
3273 	/* find the mount point */
3274 	mp = vfs_getvfs(&fhp.fh_fsid);
3275 	if (mp == NULL)
3276 		return (ESTALE);
3277 	/* now give me my vnode, it gets returned to me locked */
3278 	error = VFS_FHTOVP(mp, &fhp.fh_fid, &vp);
3279 	if (error)
3280 		return (error);
3281  	/*
3282 	 * from now on we have to make sure not
3283 	 * to forget about the vnode
3284 	 * any error that causes an abort must vput(vp)
3285 	 * just set error = err and 'goto bad;'.
3286 	 */
3287 
3288 	/*
3289 	 * from vn_open
3290 	 */
3291 	if (vp->v_type == VLNK) {
3292 		error = EMLINK;
3293 		goto bad;
3294 	}
3295 	if (vp->v_type == VSOCK) {
3296 		error = EOPNOTSUPP;
3297 		goto bad;
3298 	}
3299 	mode = 0;
3300 	if (fmode & (FWRITE | O_TRUNC)) {
3301 		if (vp->v_type == VDIR) {
3302 			error = EISDIR;
3303 			goto bad;
3304 		}
3305 		error = vn_writechk(vp, NULL);
3306 		if (error)
3307 			goto bad;
3308 		mode |= VWRITE;
3309 	}
3310 	if (fmode & FREAD)
3311 		mode |= VREAD;
3312 	if (mode) {
3313 		error = VOP_ACCESS(vp, mode, p->p_ucred);
3314 		if (error)
3315 			goto bad;
3316 	}
3317 	if (fmode & O_TRUNC) {
3318 		vn_unlock(vp);				/* XXX */
3319 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
3320 		VATTR_NULL(vap);
3321 		vap->va_size = 0;
3322 		error = VOP_SETATTR(vp, vap, p->p_ucred);
3323 		if (error)
3324 			goto bad;
3325 	}
3326 
3327 	/*
3328 	 * VOP_OPEN needs the file pointer so it can potentially override
3329 	 * it.
3330 	 *
3331 	 * WARNING! no f_ncp will be associated when fhopen()ing a directory.
3332 	 * XXX
3333 	 */
3334 	if ((error = falloc(p, &nfp, &indx)) != 0)
3335 		goto bad;
3336 	fp = nfp;
3337 
3338 	error = VOP_OPEN(vp, fmode, p->p_ucred, fp);
3339 	if (error) {
3340 		/*
3341 		 * setting f_ops this way prevents VOP_CLOSE from being
3342 		 * called or fdrop() releasing the vp from v_data.   Since
3343 		 * the VOP_OPEN failed we don't want to VOP_CLOSE.
3344 		 */
3345 		fp->f_ops = &badfileops;
3346 		fp->f_data = NULL;
3347 		goto bad_drop;
3348 	}
3349 
3350 	/*
3351 	 * The fp is given its own reference, we still have our ref and lock.
3352 	 *
3353 	 * Assert that all regular files must be created with a VM object.
3354 	 */
3355 	if (vp->v_type == VREG && vp->v_object == NULL) {
3356 		printf("fhopen: regular file did not have VM object: %p\n", vp);
3357 		goto bad_drop;
3358 	}
3359 
3360 	/*
3361 	 * The open was successful.  Handle any locking requirements.
3362 	 */
3363 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
3364 		lf.l_whence = SEEK_SET;
3365 		lf.l_start = 0;
3366 		lf.l_len = 0;
3367 		if (fmode & O_EXLOCK)
3368 			lf.l_type = F_WRLCK;
3369 		else
3370 			lf.l_type = F_RDLCK;
3371 		if (fmode & FNONBLOCK)
3372 			type = 0;
3373 		else
3374 			type = F_WAIT;
3375 		vn_unlock(vp);
3376 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3377 			/*
3378 			 * release our private reference.
3379 			 */
3380 			fsetfd(p, NULL, indx);
3381 			fdrop(fp);
3382 			vrele(vp);
3383 			return (error);
3384 		}
3385 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3386 		fp->f_flag |= FHASLOCK;
3387 	}
3388 
3389 	/*
3390 	 * Clean up.  Associate the file pointer with the previously
3391 	 * reserved descriptor and return it.
3392 	 */
3393 	vput(vp);
3394 	fsetfd(p, fp, indx);
3395 	fdrop(fp);
3396 	uap->sysmsg_result = indx;
3397 	return (0);
3398 
3399 bad_drop:
3400 	fsetfd(p, NULL, indx);
3401 	fdrop(fp);
3402 bad:
3403 	vput(vp);
3404 	return (error);
3405 }
3406 
3407 /*
3408  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3409  */
3410 int
3411 sys_fhstat(struct fhstat_args *uap)
3412 {
3413 	struct thread *td = curthread;
3414 	struct stat sb;
3415 	fhandle_t fh;
3416 	struct mount *mp;
3417 	struct vnode *vp;
3418 	int error;
3419 
3420 	/*
3421 	 * Must be super user
3422 	 */
3423 	error = suser(td);
3424 	if (error)
3425 		return (error);
3426 
3427 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
3428 	if (error)
3429 		return (error);
3430 
3431 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3432 		return (ESTALE);
3433 	if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3434 		return (error);
3435 	error = vn_stat(vp, &sb, td->td_proc->p_ucred);
3436 	vput(vp);
3437 	if (error)
3438 		return (error);
3439 	error = copyout(&sb, uap->sb, sizeof(sb));
3440 	return (error);
3441 }
3442 
3443 /*
3444  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
3445  */
3446 int
3447 sys_fhstatfs(struct fhstatfs_args *uap)
3448 {
3449 	struct thread *td = curthread;
3450 	struct proc *p = td->td_proc;
3451 	struct statfs *sp;
3452 	struct mount *mp;
3453 	struct vnode *vp;
3454 	struct statfs sb;
3455 	char *fullpath, *freepath;
3456 	fhandle_t fh;
3457 	int error;
3458 
3459 	/*
3460 	 * Must be super user
3461 	 */
3462 	if ((error = suser(td)))
3463 		return (error);
3464 
3465 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3466 		return (error);
3467 
3468 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3469 		return (ESTALE);
3470 
3471 	if (p != NULL && (p->p_fd->fd_nrdir->nc_flag & NCF_ROOT) == 0 &&
3472 	    !chroot_visible_mnt(mp, p))
3473 		return (ESTALE);
3474 
3475 	if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3476 		return (error);
3477 	mp = vp->v_mount;
3478 	sp = &mp->mnt_stat;
3479 	vput(vp);
3480 	if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0)
3481 		return (error);
3482 
3483 	error = cache_fullpath(p, mp->mnt_ncp, &fullpath, &freepath);
3484 	if (error)
3485 		return(error);
3486 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3487 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
3488 	kfree(freepath, M_TEMP);
3489 
3490 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
3491 	if (suser(td)) {
3492 		bcopy(sp, &sb, sizeof(sb));
3493 		sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
3494 		sp = &sb;
3495 	}
3496 	return (copyout(sp, uap->buf, sizeof(*sp)));
3497 }
3498 
3499 /*
3500  * Syscall to push extended attribute configuration information into the
3501  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
3502  * a command (int cmd), and attribute name and misc data.  For now, the
3503  * attribute name is left in userspace for consumption by the VFS_op.
3504  * It will probably be changed to be copied into sysspace by the
3505  * syscall in the future, once issues with various consumers of the
3506  * attribute code have raised their hands.
3507  *
3508  * Currently this is used only by UFS Extended Attributes.
3509  */
3510 int
3511 sys_extattrctl(struct extattrctl_args *uap)
3512 {
3513 	struct nlookupdata nd;
3514 	struct mount *mp;
3515 	struct vnode *vp;
3516 	int error;
3517 
3518 	vp = NULL;
3519 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3520 	if (error == 0)
3521 		error = nlookup(&nd);
3522 	if (error == 0) {
3523 		mp = nd.nl_ncp->nc_mount;
3524 		error = VFS_EXTATTRCTL(mp, uap->cmd,
3525 				uap->attrname, uap->arg,
3526 				nd.nl_cred);
3527 	}
3528 	nlookup_done(&nd);
3529 	return (error);
3530 }
3531 
3532 /*
3533  * Syscall to set a named extended attribute on a file or directory.
3534  * Accepts attribute name, and a uio structure pointing to the data to set.
3535  * The uio is consumed in the style of writev().  The real work happens
3536  * in VOP_SETEXTATTR().
3537  */
3538 int
3539 sys_extattr_set_file(struct extattr_set_file_args *uap)
3540 {
3541 	char attrname[EXTATTR_MAXNAMELEN];
3542 	struct iovec aiov[UIO_SMALLIOV];
3543 	struct iovec *needfree;
3544 	struct nlookupdata nd;
3545 	struct iovec *iov;
3546 	struct vnode *vp;
3547 	struct uio auio;
3548 	u_int iovlen;
3549 	u_int cnt;
3550 	int error;
3551 	int i;
3552 
3553 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3554 	if (error)
3555 		return (error);
3556 
3557 	vp = NULL;
3558 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3559 	if (error == 0)
3560 		error = nlookup(&nd);
3561 	if (error == 0)
3562 		error = ncp_writechk(nd.nl_ncp);
3563 	if (error == 0)
3564 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
3565 	if (error) {
3566 		nlookup_done(&nd);
3567 		return (error);
3568 	}
3569 
3570 	needfree = NULL;
3571 	iovlen = uap->iovcnt * sizeof(struct iovec);
3572 	if (uap->iovcnt > UIO_SMALLIOV) {
3573 		if (uap->iovcnt > UIO_MAXIOV) {
3574 			error = EINVAL;
3575 			goto done;
3576 		}
3577 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3578 		needfree = iov;
3579 	} else {
3580 		iov = aiov;
3581 	}
3582 	auio.uio_iov = iov;
3583 	auio.uio_iovcnt = uap->iovcnt;
3584 	auio.uio_rw = UIO_WRITE;
3585 	auio.uio_segflg = UIO_USERSPACE;
3586 	auio.uio_td = nd.nl_td;
3587 	auio.uio_offset = 0;
3588 	if ((error = copyin(uap->iovp, iov, iovlen)))
3589 		goto done;
3590 	auio.uio_resid = 0;
3591 	for (i = 0; i < uap->iovcnt; i++) {
3592 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
3593 			error = EINVAL;
3594 			goto done;
3595 		}
3596 		auio.uio_resid += iov->iov_len;
3597 		iov++;
3598 	}
3599 	cnt = auio.uio_resid;
3600 	error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3601 	cnt -= auio.uio_resid;
3602 	uap->sysmsg_result = cnt;
3603 done:
3604 	vput(vp);
3605 	nlookup_done(&nd);
3606 	if (needfree)
3607 		FREE(needfree, M_IOV);
3608 	return (error);
3609 }
3610 
3611 /*
3612  * Syscall to get a named extended attribute on a file or directory.
3613  * Accepts attribute name, and a uio structure pointing to a buffer for the
3614  * data.  The uio is consumed in the style of readv().  The real work
3615  * happens in VOP_GETEXTATTR();
3616  */
3617 int
3618 sys_extattr_get_file(struct extattr_get_file_args *uap)
3619 {
3620 	char attrname[EXTATTR_MAXNAMELEN];
3621 	struct iovec aiov[UIO_SMALLIOV];
3622 	struct iovec *needfree;
3623 	struct nlookupdata nd;
3624 	struct iovec *iov;
3625 	struct vnode *vp;
3626 	struct uio auio;
3627 	u_int iovlen;
3628 	u_int cnt;
3629 	int error;
3630 	int i;
3631 
3632 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3633 	if (error)
3634 		return (error);
3635 
3636 	vp = NULL;
3637 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3638 	if (error == 0)
3639 		error = nlookup(&nd);
3640 	if (error == 0)
3641 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
3642 	if (error) {
3643 		nlookup_done(&nd);
3644 		return (error);
3645 	}
3646 
3647 	iovlen = uap->iovcnt * sizeof (struct iovec);
3648 	needfree = NULL;
3649 	if (uap->iovcnt > UIO_SMALLIOV) {
3650 		if (uap->iovcnt > UIO_MAXIOV) {
3651 			error = EINVAL;
3652 			goto done;
3653 		}
3654 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3655 		needfree = iov;
3656 	} else {
3657 		iov = aiov;
3658 	}
3659 	auio.uio_iov = iov;
3660 	auio.uio_iovcnt = uap->iovcnt;
3661 	auio.uio_rw = UIO_READ;
3662 	auio.uio_segflg = UIO_USERSPACE;
3663 	auio.uio_td = nd.nl_td;
3664 	auio.uio_offset = 0;
3665 	if ((error = copyin(uap->iovp, iov, iovlen)))
3666 		goto done;
3667 	auio.uio_resid = 0;
3668 	for (i = 0; i < uap->iovcnt; i++) {
3669 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
3670 			error = EINVAL;
3671 			goto done;
3672 		}
3673 		auio.uio_resid += iov->iov_len;
3674 		iov++;
3675 	}
3676 	cnt = auio.uio_resid;
3677 	error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3678 	cnt -= auio.uio_resid;
3679 	uap->sysmsg_result = cnt;
3680 done:
3681 	vput(vp);
3682 	nlookup_done(&nd);
3683 	if (needfree)
3684 		FREE(needfree, M_IOV);
3685 	return(error);
3686 }
3687 
3688 /*
3689  * Syscall to delete a named extended attribute from a file or directory.
3690  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
3691  */
3692 int
3693 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
3694 {
3695 	char attrname[EXTATTR_MAXNAMELEN];
3696 	struct nlookupdata nd;
3697 	struct vnode *vp;
3698 	int error;
3699 
3700 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3701 	if (error)
3702 		return(error);
3703 
3704 	vp = NULL;
3705 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3706 	if (error == 0)
3707 		error = nlookup(&nd);
3708 	if (error == 0)
3709 		error = ncp_writechk(nd.nl_ncp);
3710 	if (error == 0)
3711 		error = cache_vget(nd.nl_ncp, nd.nl_cred, LK_EXCLUSIVE, &vp);
3712 	if (error) {
3713 		nlookup_done(&nd);
3714 		return (error);
3715 	}
3716 
3717 	error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred);
3718 	vput(vp);
3719 	nlookup_done(&nd);
3720 	return(error);
3721 }
3722 
3723 static int
3724 chroot_visible_mnt(struct mount *mp, struct proc *p)
3725 {
3726 	struct namecache *ncp;
3727 	/*
3728 	 * First check if this file system is below
3729 	 * the chroot path.
3730 	 */
3731 	ncp = mp->mnt_ncp;
3732 	while (ncp != NULL && ncp != p->p_fd->fd_nrdir)
3733 		ncp = ncp->nc_parent;
3734 	if (ncp == NULL) {
3735 		/*
3736 		 * This is not below the chroot path.
3737 		 *
3738 		 * Check if the chroot path is on the same filesystem,
3739 		 * by determing if we have to cross a mount point
3740 		 * before reaching mp->mnt_ncp.
3741 		 */
3742 		ncp = p->p_fd->fd_nrdir;
3743 		while (ncp != NULL && ncp != mp->mnt_ncp) {
3744 			if (ncp->nc_flag & NCF_MOUNTPT) {
3745 				ncp = NULL;
3746 				break;
3747 			}
3748 			ncp = ncp->nc_parent;
3749 		}
3750 	}
3751 	return(ncp != NULL);
3752 }
3753