xref: /dflybsd-src/sys/kern/vfs_syscalls.c (revision 744c01d0dc2aa1481a40e5b0988d15691602f5c9)
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 1.111 2006/12/23 23:47:54 swildner 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 #include <sys/sysctl.h>
69 #include <sys/file2.h>
70 #include <sys/spinlock2.h>
71 
72 #include <vm/vm.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_page.h>
75 
76 #include <machine/limits.h>
77 #include <machine/stdarg.h>
78 
79 #include <vfs/union/union.h>
80 
81 static void mount_warning(struct mount *mp, const char *ctl, ...);
82 static int checkvp_chdir (struct vnode *vn, struct thread *td);
83 static void checkdirs (struct vnode *olddp, struct nchandle *nch);
84 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
85 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
86 static int getutimes (const struct timeval *, struct timespec *);
87 static int setfown (struct vnode *, uid_t, gid_t);
88 static int setfmode (struct vnode *, int);
89 static int setfflags (struct vnode *, int);
90 static int setutimes (struct vnode *, const struct timespec *, int);
91 static int	usermount = 0;	/* if 1, non-root can mount fs. */
92 
93 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *);
94 
95 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "");
96 
97 /*
98  * Virtual File System System Calls
99  */
100 
101 /*
102  * Mount a file system.
103  */
104 /*
105  * mount_args(char *type, char *path, int flags, caddr_t data)
106  */
107 /* ARGSUSED */
108 int
109 sys_mount(struct mount_args *uap)
110 {
111 	struct thread *td = curthread;
112 	struct proc *p = td->td_proc;
113 	struct vnode *vp;
114 	struct nchandle nch;
115 	struct mount *mp;
116 	struct vfsconf *vfsp;
117 	int error, flag = 0, flag2 = 0;
118 	int hasmount;
119 	struct vattr va;
120 	struct nlookupdata nd;
121 	char fstypename[MFSNAMELEN];
122 	struct ucred *cred = p->p_ucred;
123 
124 	KKASSERT(p);
125 	if (cred->cr_prison != NULL)
126 		return (EPERM);
127 	if (usermount == 0 && (error = suser(td)))
128 		return (error);
129 	/*
130 	 * Do not allow NFS export by non-root users.
131 	 */
132 	if (uap->flags & MNT_EXPORTED) {
133 		error = suser(td);
134 		if (error)
135 			return (error);
136 	}
137 	/*
138 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
139 	 */
140 	if (suser(td))
141 		uap->flags |= MNT_NOSUID | MNT_NODEV;
142 
143 	/*
144 	 * Lookup the requested path and extract the nch and vnode.
145 	 */
146 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
147 	if (error == 0) {
148 		if ((error = nlookup(&nd)) == 0) {
149 			if (nd.nl_nch.ncp->nc_vp == NULL)
150 				error = ENOENT;
151 		}
152 	}
153 	if (error) {
154 		nlookup_done(&nd);
155 		return (error);
156 	}
157 
158 	/*
159 	 * Extract the locked+refd ncp and cleanup the nd structure
160 	 */
161 	nch = nd.nl_nch;
162 	cache_zero(&nd.nl_nch);
163 	nlookup_done(&nd);
164 
165 	if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) && cache_findmount(&nch))
166 		hasmount = 1;
167 	else
168 		hasmount = 0;
169 
170 
171 	/*
172 	 * now we have the locked ref'd nch and unreferenced vnode.
173 	 */
174 	vp = nch.ncp->nc_vp;
175 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
176 		cache_put(&nch);
177 		return (error);
178 	}
179 	cache_unlock(&nch);
180 
181 	/*
182 	 * Now we have an unlocked ref'd nch and a locked ref'd vp
183 	 */
184 	if (uap->flags & MNT_UPDATE) {
185 		if ((vp->v_flag & VROOT) == 0) {
186 			cache_drop(&nch);
187 			vput(vp);
188 			return (EINVAL);
189 		}
190 		mp = vp->v_mount;
191 		flag = mp->mnt_flag;
192 		flag2 = mp->mnt_kern_flag;
193 		/*
194 		 * We only allow the filesystem to be reloaded if it
195 		 * is currently mounted read-only.
196 		 */
197 		if ((uap->flags & MNT_RELOAD) &&
198 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
199 			cache_drop(&nch);
200 			vput(vp);
201 			return (EOPNOTSUPP);	/* Needs translation */
202 		}
203 		/*
204 		 * Only root, or the user that did the original mount is
205 		 * permitted to update it.
206 		 */
207 		if (mp->mnt_stat.f_owner != cred->cr_uid &&
208 		    (error = suser(td))) {
209 			cache_drop(&nch);
210 			vput(vp);
211 			return (error);
212 		}
213 		if (vfs_busy(mp, LK_NOWAIT)) {
214 			cache_drop(&nch);
215 			vput(vp);
216 			return (EBUSY);
217 		}
218 		if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
219 			cache_drop(&nch);
220 			vfs_unbusy(mp);
221 			vput(vp);
222 			return (EBUSY);
223 		}
224 		vp->v_flag |= VMOUNT;
225 		mp->mnt_flag |=
226 		    uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
227 		vn_unlock(vp);
228 		goto update;
229 	}
230 	/*
231 	 * If the user is not root, ensure that they own the directory
232 	 * onto which we are attempting to mount.
233 	 */
234 	if ((error = VOP_GETATTR(vp, &va)) ||
235 	    (va.va_uid != cred->cr_uid && (error = suser(td)))) {
236 		cache_drop(&nch);
237 		vput(vp);
238 		return (error);
239 	}
240 	if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
241 		cache_drop(&nch);
242 		vput(vp);
243 		return (error);
244 	}
245 	if (vp->v_type != VDIR) {
246 		cache_drop(&nch);
247 		vput(vp);
248 		return (ENOTDIR);
249 	}
250 	if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
251 		cache_drop(&nch);
252 		vput(vp);
253 		return (error);
254 	}
255 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
256 		if (!strcmp(vfsp->vfc_name, fstypename))
257 			break;
258 	}
259 	if (vfsp == NULL) {
260 		linker_file_t lf;
261 
262 		/* Only load modules for root (very important!) */
263 		if ((error = suser(td)) != 0) {
264 			cache_drop(&nch);
265 			vput(vp);
266 			return error;
267 		}
268 		error = linker_load_file(fstypename, &lf);
269 		if (error || lf == NULL) {
270 			cache_drop(&nch);
271 			vput(vp);
272 			if (lf == NULL)
273 				error = ENODEV;
274 			return error;
275 		}
276 		lf->userrefs++;
277 		/* lookup again, see if the VFS was loaded */
278 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
279 			if (!strcmp(vfsp->vfc_name, fstypename))
280 				break;
281 		}
282 		if (vfsp == NULL) {
283 			lf->userrefs--;
284 			linker_file_unload(lf);
285 			cache_drop(&nch);
286 			vput(vp);
287 			return (ENODEV);
288 		}
289 	}
290 	if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
291 		cache_drop(&nch);
292 		vput(vp);
293 		return (EBUSY);
294 	}
295 	vp->v_flag |= VMOUNT;
296 
297 	/*
298 	 * Allocate and initialize the filesystem.
299 	 */
300 	mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
301 	TAILQ_INIT(&mp->mnt_nvnodelist);
302 	TAILQ_INIT(&mp->mnt_reservedvnlist);
303 	TAILQ_INIT(&mp->mnt_jlist);
304 	mp->mnt_nvnodelistsize = 0;
305 	lockinit(&mp->mnt_lock, "vfslock", 0, 0);
306 	vfs_busy(mp, LK_NOWAIT);
307 	mp->mnt_op = vfsp->vfc_vfsops;
308 	mp->mnt_vfc = vfsp;
309 	vfsp->vfc_refcount++;
310 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
311 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
312 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
313 	mp->mnt_stat.f_owner = cred->cr_uid;
314 	mp->mnt_iosize_max = DFLTPHYS;
315 	vn_unlock(vp);
316 update:
317 	/*
318 	 * Set the mount level flags.
319 	 */
320 	if (uap->flags & MNT_RDONLY)
321 		mp->mnt_flag |= MNT_RDONLY;
322 	else if (mp->mnt_flag & MNT_RDONLY)
323 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
324 	mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
325 	    MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
326 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
327 	    MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
328 	mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
329 	    MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
330 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
331 	    MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
332 	/*
333 	 * Mount the filesystem.
334 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
335 	 * get.
336 	 */
337 	error = VFS_MOUNT(mp, uap->path, uap->data, cred);
338 	if (mp->mnt_flag & MNT_UPDATE) {
339 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
340 			mp->mnt_flag &= ~MNT_RDONLY;
341 		mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
342 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
343 		if (error) {
344 			mp->mnt_flag = flag;
345 			mp->mnt_kern_flag = flag2;
346 		}
347 		vfs_unbusy(mp);
348 		vp->v_flag &= ~VMOUNT;
349 		vrele(vp);
350 		cache_drop(&nch);
351 		return (error);
352 	}
353 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
354 	/*
355 	 * Put the new filesystem on the mount list after root.  The mount
356 	 * point gets its own mnt_ncmountpt (unless the VFS already set one
357 	 * up) which represents the root of the mount.  The lookup code
358 	 * detects the mount point going forward and checks the root of
359 	 * the mount going backwards.
360 	 *
361 	 * It is not necessary to invalidate or purge the vnode underneath
362 	 * because elements under the mount will be given their own glue
363 	 * namecache record.
364 	 */
365 	if (!error) {
366 		if (mp->mnt_ncmountpt.ncp == NULL) {
367 			/*
368 			 * allocate, then unlock, but leave the ref intact
369 			 */
370 			cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
371 			cache_unlock(&mp->mnt_ncmountpt);
372 		}
373 		mp->mnt_ncmounton = nch;		/* inherits ref */
374 		nch.ncp->nc_flag |= NCF_ISMOUNTPT;
375 
376 		/* XXX get the root of the fs and cache_setvp(mnt_ncmountpt...) */
377 		vp->v_flag &= ~VMOUNT;
378 		mountlist_insert(mp, MNTINS_LAST);
379 		checkdirs(vp, &mp->mnt_ncmounton);
380 		vn_unlock(vp);
381 		error = vfs_allocate_syncvnode(mp);
382 		vfs_unbusy(mp);
383 		error = VFS_START(mp, 0);
384 		vrele(vp);
385 	} else {
386 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
387 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
388 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
389 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
390 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
391 		vp->v_flag &= ~VMOUNT;
392 		mp->mnt_vfc->vfc_refcount--;
393 		vfs_unbusy(mp);
394 		kfree(mp, M_MOUNT);
395 		cache_drop(&nch);
396 		vput(vp);
397 	}
398 	return (error);
399 }
400 
401 /*
402  * Scan all active processes to see if any of them have a current
403  * or root directory onto which the new filesystem has just been
404  * mounted. If so, replace them with the new mount point.
405  *
406  * The passed ncp is ref'd and locked (from the mount code) and
407  * must be associated with the vnode representing the root of the
408  * mount point.
409  */
410 struct checkdirs_info {
411 	struct vnode *olddp;
412 	struct vnode *newdp;
413 	struct nchandle nch;
414 };
415 
416 static int checkdirs_callback(struct proc *p, void *data);
417 
418 static void
419 checkdirs(struct vnode *olddp, struct nchandle *nch)
420 {
421 	struct checkdirs_info info;
422 	struct vnode *newdp;
423 	struct mount *mp;
424 
425 	if (olddp->v_usecount == 1)
426 		return;
427 	mp = nch->mount;
428 	if (VFS_ROOT(mp, &newdp))
429 		panic("mount: lost mount");
430 	cache_setunresolved(nch);
431 	cache_setvp(nch, newdp);
432 
433 	if (rootvnode == olddp) {
434 		vref(newdp);
435 		vfs_cache_setroot(newdp, cache_hold(nch));
436 	}
437 
438 	info.olddp = olddp;
439 	info.newdp = newdp;
440 	info.nch = *nch;
441 	allproc_scan(checkdirs_callback, &info);
442 	vput(newdp);
443 }
444 
445 /*
446  * NOTE: callback is not MP safe because the scanned process's filedesc
447  * structure can be ripped out from under us, amoung other things.
448  */
449 static int
450 checkdirs_callback(struct proc *p, void *data)
451 {
452 	struct checkdirs_info *info = data;
453 	struct filedesc *fdp;
454 	struct nchandle ncdrop1;
455 	struct nchandle ncdrop2;
456 	struct vnode *vprele1;
457 	struct vnode *vprele2;
458 
459 	if ((fdp = p->p_fd) != NULL) {
460 		cache_zero(&ncdrop1);
461 		cache_zero(&ncdrop2);
462 		vprele1 = NULL;
463 		vprele2 = NULL;
464 
465 		/*
466 		 * MPUNSAFE - XXX fdp can be pulled out from under a
467 		 * foreign process.
468 		 *
469 		 * A shared filedesc is ok, we don't have to copy it
470 		 * because we are making this change globally.
471 		 */
472 		spin_lock_wr(&fdp->fd_spin);
473 		if (fdp->fd_cdir == info->olddp) {
474 			vprele1 = fdp->fd_cdir;
475 			vref(info->newdp);
476 			fdp->fd_cdir = info->newdp;
477 			ncdrop1 = fdp->fd_ncdir;
478 			cache_copy(&info->nch, &fdp->fd_ncdir);
479 		}
480 		if (fdp->fd_rdir == info->olddp) {
481 			vprele2 = fdp->fd_rdir;
482 			vref(info->newdp);
483 			fdp->fd_rdir = info->newdp;
484 			ncdrop2 = fdp->fd_nrdir;
485 			cache_copy(&info->nch, &fdp->fd_nrdir);
486 		}
487 		spin_unlock_wr(&fdp->fd_spin);
488 		if (ncdrop1.ncp)
489 			cache_drop(&ncdrop1);
490 		if (ncdrop2.ncp)
491 			cache_drop(&ncdrop2);
492 		if (vprele1)
493 			vrele(vprele1);
494 		if (vprele2)
495 			vrele(vprele2);
496 	}
497 	return(0);
498 }
499 
500 /*
501  * Unmount a file system.
502  *
503  * Note: unmount takes a path to the vnode mounted on as argument,
504  * not special file (as before).
505  */
506 /*
507  * umount_args(char *path, int flags)
508  */
509 /* ARGSUSED */
510 int
511 sys_unmount(struct unmount_args *uap)
512 {
513 	struct thread *td = curthread;
514 	struct proc *p = td->td_proc;
515 	struct mount *mp = NULL;
516 	int error;
517 	struct nlookupdata nd;
518 
519 	KKASSERT(p);
520 	if (p->p_ucred->cr_prison != NULL)
521 		return (EPERM);
522 	if (usermount == 0 && (error = suser(td)))
523 		return (error);
524 
525 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
526 	if (error == 0)
527 		error = nlookup(&nd);
528 	if (error)
529 		goto out;
530 
531 	mp = nd.nl_nch.mount;
532 
533 	/*
534 	 * Only root, or the user that did the original mount is
535 	 * permitted to unmount this filesystem.
536 	 */
537 	if ((mp->mnt_stat.f_owner != p->p_ucred->cr_uid) &&
538 	    (error = suser(td)))
539 		goto out;
540 
541 	/*
542 	 * Don't allow unmounting the root file system.
543 	 */
544 	if (mp->mnt_flag & MNT_ROOTFS) {
545 		error = EINVAL;
546 		goto out;
547 	}
548 
549 	/*
550 	 * Must be the root of the filesystem
551 	 */
552 	if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
553 		error = EINVAL;
554 		goto out;
555 	}
556 
557 out:
558 	nlookup_done(&nd);
559 	if (error)
560 		return (error);
561 	return (dounmount(mp, uap->flags));
562 }
563 
564 /*
565  * Do the actual file system unmount.
566  */
567 static int
568 dounmount_interlock(struct mount *mp)
569 {
570 	if (mp->mnt_kern_flag & MNTK_UNMOUNT)
571 		return (EBUSY);
572 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
573 	return(0);
574 }
575 
576 int
577 dounmount(struct mount *mp, int flags)
578 {
579 	struct namecache *ncp;
580 	struct nchandle nch;
581 	int error;
582 	int async_flag;
583 	int lflags;
584 	int freeok = 1;
585 
586 	/*
587 	 * Exclusive access for unmounting purposes
588 	 */
589 	if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
590 		return (error);
591 
592 	/*
593 	 * Allow filesystems to detect that a forced unmount is in progress.
594 	 */
595 	if (flags & MNT_FORCE)
596 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
597 	lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
598 	error = lockmgr(&mp->mnt_lock, lflags);
599 	if (error) {
600 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
601 		if (mp->mnt_kern_flag & MNTK_MWAIT)
602 			wakeup(mp);
603 		return (error);
604 	}
605 
606 	if (mp->mnt_flag & MNT_EXPUBLIC)
607 		vfs_setpublicfs(NULL, NULL, NULL);
608 
609 	vfs_msync(mp, MNT_WAIT);
610 	async_flag = mp->mnt_flag & MNT_ASYNC;
611 	mp->mnt_flag &=~ MNT_ASYNC;
612 
613 	/*
614 	 * If this filesystem isn't aliasing other filesystems,
615 	 * try to invalidate any remaining namecache entries and
616 	 * check the count afterwords.
617 	 */
618 	if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
619 		cache_lock(&mp->mnt_ncmountpt);
620 		cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
621 		cache_unlock(&mp->mnt_ncmountpt);
622 
623 		if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
624 		    (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
625 
626 			if ((flags & MNT_FORCE) == 0) {
627 				error = EBUSY;
628 				mount_warning(mp, "Cannot unmount: "
629 						  "%d namecache "
630 						  "references still "
631 						  "present",
632 						  ncp->nc_refs - 1);
633 			} else {
634 				mount_warning(mp, "Forced unmount: "
635 						  "%d namecache "
636 						  "references still "
637 						  "present",
638 						  ncp->nc_refs - 1);
639 				freeok = 0;
640 			}
641 		}
642 	}
643 
644 	/*
645 	 * nchandle records ref the mount structure.  Expect a count of 1
646 	 * (our mount->mnt_ncmountpt).
647 	 */
648 	if (mp->mnt_refs != 1) {
649 		if ((flags & MNT_FORCE) == 0) {
650 			mount_warning(mp, "Cannot unmount: "
651 					  "%d process references still "
652 					  "present", mp->mnt_refs);
653 			error = EBUSY;
654 		} else {
655 			mount_warning(mp, "Forced unmount: "
656 					  "%d process references still "
657 					  "present", mp->mnt_refs);
658 			freeok = 0;
659 		}
660 	}
661 
662 	if (error == 0) {
663 		if (mp->mnt_syncer != NULL)
664 			vrele(mp->mnt_syncer);
665 		if (((mp->mnt_flag & MNT_RDONLY) ||
666 		     (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
667 		    (flags & MNT_FORCE)) {
668 			error = VFS_UNMOUNT(mp, flags);
669 		}
670 	}
671 	if (error) {
672 		if (mp->mnt_syncer == NULL)
673 			vfs_allocate_syncvnode(mp);
674 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
675 		mp->mnt_flag |= async_flag;
676 		lockmgr(&mp->mnt_lock, LK_RELEASE);
677 		if (mp->mnt_kern_flag & MNTK_MWAIT)
678 			wakeup(mp);
679 		return (error);
680 	}
681 	/*
682 	 * Clean up any journals still associated with the mount after
683 	 * filesystem activity has ceased.
684 	 */
685 	journal_remove_all_journals(mp,
686 	    ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
687 
688 	mountlist_remove(mp);
689 
690 	/*
691 	 * Remove any installed vnode ops here so the individual VFSs don't
692 	 * have to.
693 	 */
694 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
695 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
696 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
697 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
698 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
699 
700 	if (mp->mnt_ncmountpt.ncp != NULL) {
701 		nch = mp->mnt_ncmountpt;
702 		cache_zero(&mp->mnt_ncmountpt);
703 		cache_clrmountpt(&nch);
704 		cache_drop(&nch);
705 	}
706 	if (mp->mnt_ncmounton.ncp != NULL) {
707 		nch = mp->mnt_ncmounton;
708 		cache_zero(&mp->mnt_ncmounton);
709 		cache_clrmountpt(&nch);
710 		cache_drop(&nch);
711 	}
712 
713 	mp->mnt_vfc->vfc_refcount--;
714 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
715 		panic("unmount: dangling vnode");
716 	lockmgr(&mp->mnt_lock, LK_RELEASE);
717 	if (mp->mnt_kern_flag & MNTK_MWAIT)
718 		wakeup(mp);
719 	if (freeok)
720 		kfree(mp, M_MOUNT);
721 	return (0);
722 }
723 
724 static
725 void
726 mount_warning(struct mount *mp, const char *ctl, ...)
727 {
728 	char *ptr;
729 	char *buf;
730 	__va_list va;
731 
732 	__va_start(va, ctl);
733 	if (cache_fullpath(NULL, &mp->mnt_ncmounton, &ptr, &buf) == 0) {
734 		kprintf("unmount(%s): ", ptr);
735 		kvprintf(ctl, va);
736 		kprintf("\n");
737 		kfree(buf, M_TEMP);
738 	} else {
739 		kprintf("unmount(%p): ", mp);
740 		kvprintf(ctl, va);
741 		kprintf("\n");
742 	}
743 	__va_end(va);
744 }
745 
746 /*
747  * Sync each mounted filesystem.
748  */
749 
750 #ifdef DEBUG
751 static int syncprt = 0;
752 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
753 #endif /* DEBUG */
754 
755 static int sync_callback(struct mount *mp, void *data);
756 
757 /* ARGSUSED */
758 int
759 sys_sync(struct sync_args *uap)
760 {
761 	mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
762 #ifdef DEBUG
763 	/*
764 	 * print out buffer pool stat information on each sync() call.
765 	 */
766 	if (syncprt)
767 		vfs_bufstats();
768 #endif /* DEBUG */
769 	return (0);
770 }
771 
772 static
773 int
774 sync_callback(struct mount *mp, void *data __unused)
775 {
776 	int asyncflag;
777 
778 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
779 		asyncflag = mp->mnt_flag & MNT_ASYNC;
780 		mp->mnt_flag &= ~MNT_ASYNC;
781 		vfs_msync(mp, MNT_NOWAIT);
782 		VFS_SYNC(mp, MNT_NOWAIT);
783 		mp->mnt_flag |= asyncflag;
784 	}
785 	return(0);
786 }
787 
788 /* XXX PRISON: could be per prison flag */
789 static int prison_quotas;
790 #if 0
791 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
792 #endif
793 
794 /*
795  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
796  *
797  * Change filesystem quotas.
798  */
799 /* ARGSUSED */
800 int
801 sys_quotactl(struct quotactl_args *uap)
802 {
803 	struct nlookupdata nd;
804 	struct thread *td;
805 	struct proc *p;
806 	struct mount *mp;
807 	int error;
808 
809 	td = curthread;
810 	p = td->td_proc;
811 	if (p->p_ucred->cr_prison && !prison_quotas)
812 		return (EPERM);
813 
814 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
815 	if (error == 0)
816 		error = nlookup(&nd);
817 	if (error == 0) {
818 		mp = nd.nl_nch.mount;
819 		error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
820 				    uap->arg, nd.nl_cred);
821 	}
822 	nlookup_done(&nd);
823 	return (error);
824 }
825 
826 /*
827  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
828  *		void *buf, int buflen)
829  *
830  * This function operates on a mount point and executes the specified
831  * operation using the specified control data, and possibly returns data.
832  *
833  * The actual number of bytes stored in the result buffer is returned, 0
834  * if none, otherwise an error is returned.
835  */
836 /* ARGSUSED */
837 int
838 sys_mountctl(struct mountctl_args *uap)
839 {
840 	struct thread *td = curthread;
841 	struct proc *p = td->td_proc;
842 	struct file *fp;
843 	void *ctl = NULL;
844 	void *buf = NULL;
845 	char *path = NULL;
846 	int error;
847 
848 	/*
849 	 * Sanity and permissions checks.  We must be root.
850 	 */
851 	KKASSERT(p);
852 	if (p->p_ucred->cr_prison != NULL)
853 		return (EPERM);
854 	if ((error = suser(td)) != 0)
855 		return (error);
856 
857 	/*
858 	 * Argument length checks
859 	 */
860 	if (uap->ctllen < 0 || uap->ctllen > 1024)
861 		return (EINVAL);
862 	if (uap->buflen < 0 || uap->buflen > 16 * 1024)
863 		return (EINVAL);
864 	if (uap->path == NULL)
865 		return (EINVAL);
866 
867 	/*
868 	 * Allocate the necessary buffers and copyin data
869 	 */
870 	path = objcache_get(namei_oc, M_WAITOK);
871 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
872 	if (error)
873 		goto done;
874 
875 	if (uap->ctllen) {
876 		ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
877 		error = copyin(uap->ctl, ctl, uap->ctllen);
878 		if (error)
879 			goto done;
880 	}
881 	if (uap->buflen)
882 		buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
883 
884 	/*
885 	 * Validate the descriptor
886 	 */
887 	fp = holdfp(p->p_fd, uap->fd, -1);
888 	if (fp == NULL) {
889 		error = EBADF;
890 		goto done;
891 	}
892 
893 	/*
894 	 * Execute the internal kernel function and clean up.
895 	 */
896 	error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
897 	if (fp)
898 		fdrop(fp);
899 	if (error == 0 && uap->sysmsg_result > 0)
900 		error = copyout(buf, uap->buf, uap->sysmsg_result);
901 done:
902 	if (path)
903 		objcache_put(namei_oc, path);
904 	if (ctl)
905 		kfree(ctl, M_TEMP);
906 	if (buf)
907 		kfree(buf, M_TEMP);
908 	return (error);
909 }
910 
911 /*
912  * Execute a mount control operation by resolving the path to a mount point
913  * and calling vop_mountctl().
914  */
915 int
916 kern_mountctl(const char *path, int op, struct file *fp,
917 		const void *ctl, int ctllen,
918 		void *buf, int buflen, int *res)
919 {
920 	struct vnode *vp;
921 	struct mount *mp;
922 	struct nlookupdata nd;
923 	int error;
924 
925 	*res = 0;
926 	vp = NULL;
927 	error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
928 	if (error == 0)
929 		error = nlookup(&nd);
930 	if (error == 0)
931 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
932 	nlookup_done(&nd);
933 	if (error)
934 		return (error);
935 
936 	mp = vp->v_mount;
937 
938 	/*
939 	 * Must be the root of the filesystem
940 	 */
941 	if ((vp->v_flag & VROOT) == 0) {
942 		vput(vp);
943 		return (EINVAL);
944 	}
945 	error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen,
946 				buf, buflen, res);
947 	vput(vp);
948 	return (error);
949 }
950 
951 int
952 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
953 {
954 	struct thread *td = curthread;
955 	struct proc *p = td->td_proc;
956 	struct mount *mp;
957 	struct statfs *sp;
958 	char *fullpath, *freepath;
959 	int error;
960 
961 	if ((error = nlookup(nd)) != 0)
962 		return (error);
963 	mp = nd->nl_nch.mount;
964 	sp = &mp->mnt_stat;
965 	if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
966 		return (error);
967 
968 	error = cache_fullpath(p, &mp->mnt_ncmountpt, &fullpath, &freepath);
969 	if (error)
970 		return(error);
971 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
972 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
973 	kfree(freepath, M_TEMP);
974 
975 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
976 	bcopy(sp, buf, sizeof(*buf));
977 	/* Only root should have access to the fsid's. */
978 	if (suser(td))
979 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
980 	return (0);
981 }
982 
983 /*
984  * statfs_args(char *path, struct statfs *buf)
985  *
986  * Get filesystem statistics.
987  */
988 int
989 sys_statfs(struct statfs_args *uap)
990 {
991 	struct nlookupdata nd;
992 	struct statfs buf;
993 	int error;
994 
995 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
996 	if (error == 0)
997 		error = kern_statfs(&nd, &buf);
998 	nlookup_done(&nd);
999 	if (error == 0)
1000 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1001 	return (error);
1002 }
1003 
1004 int
1005 kern_fstatfs(int fd, struct statfs *buf)
1006 {
1007 	struct thread *td = curthread;
1008 	struct proc *p = td->td_proc;
1009 	struct file *fp;
1010 	struct mount *mp;
1011 	struct statfs *sp;
1012 	char *fullpath, *freepath;
1013 	int error;
1014 
1015 	KKASSERT(p);
1016 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1017 		return (error);
1018 	mp = ((struct vnode *)fp->f_data)->v_mount;
1019 	if (mp == NULL) {
1020 		error = EBADF;
1021 		goto done;
1022 	}
1023 	if (fp->f_cred == NULL) {
1024 		error = EINVAL;
1025 		goto done;
1026 	}
1027 	sp = &mp->mnt_stat;
1028 	if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1029 		goto done;
1030 
1031 	if ((error = cache_fullpath(p, &mp->mnt_ncmountpt, &fullpath, &freepath)) != 0)
1032 		goto done;
1033 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1034 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1035 	kfree(freepath, M_TEMP);
1036 
1037 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1038 	bcopy(sp, buf, sizeof(*buf));
1039 
1040 	/* Only root should have access to the fsid's. */
1041 	if (suser(td))
1042 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1043 	error = 0;
1044 done:
1045 	fdrop(fp);
1046 	return (error);
1047 }
1048 
1049 /*
1050  * fstatfs_args(int fd, struct statfs *buf)
1051  *
1052  * Get filesystem statistics.
1053  */
1054 int
1055 sys_fstatfs(struct fstatfs_args *uap)
1056 {
1057 	struct statfs buf;
1058 	int error;
1059 
1060 	error = kern_fstatfs(uap->fd, &buf);
1061 
1062 	if (error == 0)
1063 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1064 	return (error);
1065 }
1066 
1067 /*
1068  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1069  *
1070  * Get statistics on all filesystems.
1071  */
1072 
1073 struct getfsstat_info {
1074 	struct statfs *sfsp;
1075 	long count;
1076 	long maxcount;
1077 	int error;
1078 	int flags;
1079 	struct proc *p;
1080 };
1081 
1082 static int getfsstat_callback(struct mount *, void *);
1083 
1084 /* ARGSUSED */
1085 int
1086 sys_getfsstat(struct getfsstat_args *uap)
1087 {
1088 	struct thread *td = curthread;
1089 	struct proc *p = td->td_proc;
1090 	struct getfsstat_info info;
1091 
1092 	bzero(&info, sizeof(info));
1093 
1094 	info.maxcount = uap->bufsize / sizeof(struct statfs);
1095 	info.sfsp = uap->buf;
1096 	info.count = 0;
1097 	info.flags = uap->flags;
1098 	info.p = p;
1099 
1100 	mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1101 	if (info.sfsp && info.count > info.maxcount)
1102 		uap->sysmsg_result = info.maxcount;
1103 	else
1104 		uap->sysmsg_result = info.count;
1105 	return (info.error);
1106 }
1107 
1108 static int
1109 getfsstat_callback(struct mount *mp, void *data)
1110 {
1111 	struct getfsstat_info *info = data;
1112 	struct statfs *sp;
1113 	char *freepath;
1114 	char *fullpath;
1115 	int error;
1116 
1117 	if (info->sfsp && info->count < info->maxcount) {
1118 		if (info->p && !chroot_visible_mnt(mp, info->p))
1119 			return(0);
1120 		sp = &mp->mnt_stat;
1121 
1122 		/*
1123 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1124 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1125 		 * overrides MNT_WAIT.
1126 		 */
1127 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1128 		    (info->flags & MNT_WAIT)) &&
1129 		    (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1130 			return(0);
1131 		}
1132 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1133 
1134 		error = cache_fullpath(info->p, &mp->mnt_ncmountpt,
1135 					&fullpath, &freepath);
1136 		if (error) {
1137 			info->error = error;
1138 			return(-1);
1139 		}
1140 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1141 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1142 		kfree(freepath, M_TEMP);
1143 
1144 		error = copyout(sp, info->sfsp, sizeof(*sp));
1145 		if (error) {
1146 			info->error = error;
1147 			return (-1);
1148 		}
1149 		++info->sfsp;
1150 	}
1151 	info->count++;
1152 	return(0);
1153 }
1154 
1155 /*
1156  * fchdir_args(int fd)
1157  *
1158  * Change current working directory to a given file descriptor.
1159  */
1160 /* ARGSUSED */
1161 int
1162 sys_fchdir(struct fchdir_args *uap)
1163 {
1164 	struct thread *td = curthread;
1165 	struct proc *p = td->td_proc;
1166 	struct filedesc *fdp = p->p_fd;
1167 	struct vnode *vp, *ovp;
1168 	struct mount *mp;
1169 	struct file *fp;
1170 	struct nchandle nch, onch, tnch;
1171 	int error;
1172 
1173 	if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1174 		return (error);
1175 	vp = (struct vnode *)fp->f_data;
1176 	vref(vp);
1177 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1178 	if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL)
1179 		error = ENOTDIR;
1180 	else
1181 		error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
1182 	if (error) {
1183 		vput(vp);
1184 		fdrop(fp);
1185 		return (error);
1186 	}
1187 	cache_copy(&fp->f_nchandle, &nch);
1188 
1189 	/*
1190 	 * If the ncp has become a mount point, traverse through
1191 	 * the mount point.
1192 	 */
1193 
1194 	while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1195 	       (mp = cache_findmount(&nch)) != NULL
1196 	) {
1197 		error = nlookup_mp(mp, &tnch);
1198 		if (error == 0) {
1199 			cache_unlock(&tnch);	/* leave ref intact */
1200 			vput(vp);
1201 			vp = tnch.ncp->nc_vp;
1202 			error = vget(vp, LK_SHARED);
1203 			KKASSERT(error == 0);
1204 			cache_drop(&nch);
1205 			nch = tnch;
1206 		}
1207 	}
1208 	if (error == 0) {
1209 		ovp = fdp->fd_cdir;
1210 		onch = fdp->fd_ncdir;
1211 		vn_unlock(vp);		/* leave ref intact */
1212 		fdp->fd_cdir = vp;
1213 		fdp->fd_ncdir = nch;
1214 		cache_drop(&onch);
1215 		vrele(ovp);
1216 	} else {
1217 		cache_drop(&nch);
1218 		vput(vp);
1219 	}
1220 	fdrop(fp);
1221 	return (error);
1222 }
1223 
1224 int
1225 kern_chdir(struct nlookupdata *nd)
1226 {
1227 	struct thread *td = curthread;
1228 	struct proc *p = td->td_proc;
1229 	struct filedesc *fdp = p->p_fd;
1230 	struct vnode *vp, *ovp;
1231 	struct nchandle onch;
1232 	int error;
1233 
1234 	if ((error = nlookup(nd)) != 0)
1235 		return (error);
1236 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1237 		return (ENOENT);
1238 	if ((error = vget(vp, LK_SHARED)) != 0)
1239 		return (error);
1240 
1241 	error = checkvp_chdir(vp, td);
1242 	vn_unlock(vp);
1243 	if (error == 0) {
1244 		ovp = fdp->fd_cdir;
1245 		onch = fdp->fd_ncdir;
1246 		cache_unlock(&nd->nl_nch);	/* leave reference intact */
1247 		fdp->fd_ncdir = nd->nl_nch;
1248 		fdp->fd_cdir = vp;
1249 		cache_drop(&onch);
1250 		vrele(ovp);
1251 		cache_zero(&nd->nl_nch);
1252 	} else {
1253 		vrele(vp);
1254 	}
1255 	return (error);
1256 }
1257 
1258 /*
1259  * chdir_args(char *path)
1260  *
1261  * Change current working directory (``.'').
1262  */
1263 int
1264 sys_chdir(struct chdir_args *uap)
1265 {
1266 	struct nlookupdata nd;
1267 	int error;
1268 
1269 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1270 	if (error == 0)
1271 		error = kern_chdir(&nd);
1272 	nlookup_done(&nd);
1273 	return (error);
1274 }
1275 
1276 /*
1277  * Helper function for raised chroot(2) security function:  Refuse if
1278  * any filedescriptors are open directories.
1279  */
1280 static int
1281 chroot_refuse_vdir_fds(struct filedesc *fdp)
1282 {
1283 	struct vnode *vp;
1284 	struct file *fp;
1285 	int error;
1286 	int fd;
1287 
1288 	for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1289 		if ((error = holdvnode(fdp, fd, &fp)) != 0)
1290 			continue;
1291 		vp = (struct vnode *)fp->f_data;
1292 		if (vp->v_type != VDIR) {
1293 			fdrop(fp);
1294 			continue;
1295 		}
1296 		fdrop(fp);
1297 		return(EPERM);
1298 	}
1299 	return (0);
1300 }
1301 
1302 /*
1303  * This sysctl determines if we will allow a process to chroot(2) if it
1304  * has a directory open:
1305  *	0: disallowed for all processes.
1306  *	1: allowed for processes that were not already chroot(2)'ed.
1307  *	2: allowed for all processes.
1308  */
1309 
1310 static int chroot_allow_open_directories = 1;
1311 
1312 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1313      &chroot_allow_open_directories, 0, "");
1314 
1315 /*
1316  * chroot to the specified namecache entry.  We obtain the vp from the
1317  * namecache data.  The passed ncp must be locked and referenced and will
1318  * remain locked and referenced on return.
1319  */
1320 int
1321 kern_chroot(struct nchandle *nch)
1322 {
1323 	struct thread *td = curthread;
1324 	struct proc *p = td->td_proc;
1325 	struct filedesc *fdp = p->p_fd;
1326 	struct vnode *vp;
1327 	int error;
1328 
1329 	/*
1330 	 * Only root can chroot
1331 	 */
1332 	if ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0)
1333 		return (error);
1334 
1335 	/*
1336 	 * Disallow open directory descriptors (fchdir() breakouts).
1337 	 */
1338 	if (chroot_allow_open_directories == 0 ||
1339 	   (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1340 		if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1341 			return (error);
1342 	}
1343 	if ((vp = nch->ncp->nc_vp) == NULL)
1344 		return (ENOENT);
1345 
1346 	if ((error = vget(vp, LK_SHARED)) != 0)
1347 		return (error);
1348 
1349 	/*
1350 	 * Check the validity of vp as a directory to change to and
1351 	 * associate it with rdir/jdir.
1352 	 */
1353 	error = checkvp_chdir(vp, td);
1354 	vn_unlock(vp);			/* leave reference intact */
1355 	if (error == 0) {
1356 		vrele(fdp->fd_rdir);
1357 		fdp->fd_rdir = vp;	/* reference inherited by fd_rdir */
1358 		cache_drop(&fdp->fd_nrdir);
1359 		cache_copy(nch, &fdp->fd_nrdir);
1360 		if (fdp->fd_jdir == NULL) {
1361 			fdp->fd_jdir = vp;
1362 			vref(fdp->fd_jdir);
1363 			cache_copy(nch, &fdp->fd_njdir);
1364 		}
1365 	} else {
1366 		vrele(vp);
1367 	}
1368 	return (error);
1369 }
1370 
1371 /*
1372  * chroot_args(char *path)
1373  *
1374  * Change notion of root (``/'') directory.
1375  */
1376 /* ARGSUSED */
1377 int
1378 sys_chroot(struct chroot_args *uap)
1379 {
1380 	struct thread *td = curthread;
1381 	struct nlookupdata nd;
1382 	int error;
1383 
1384 	KKASSERT(td->td_proc);
1385 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1386 	if (error) {
1387 		nlookup_done(&nd);
1388 		return(error);
1389 	}
1390 	error = nlookup(&nd);
1391 	if (error == 0)
1392 		error = kern_chroot(&nd.nl_nch);
1393 	nlookup_done(&nd);
1394 	return(error);
1395 }
1396 
1397 /*
1398  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1399  * determine whether it is legal to chdir to the vnode.  The vnode's state
1400  * is not changed by this call.
1401  */
1402 int
1403 checkvp_chdir(struct vnode *vp, struct thread *td)
1404 {
1405 	int error;
1406 
1407 	if (vp->v_type != VDIR)
1408 		error = ENOTDIR;
1409 	else
1410 		error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred);
1411 	return (error);
1412 }
1413 
1414 int
1415 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1416 {
1417 	struct thread *td = curthread;
1418 	struct proc *p = td->td_proc;
1419 	struct lwp *lp = td->td_lwp;
1420 	struct filedesc *fdp = p->p_fd;
1421 	int cmode, flags;
1422 	struct file *nfp;
1423 	struct file *fp;
1424 	struct vnode *vp;
1425 	int type, indx, error;
1426 	struct flock lf;
1427 
1428 	if ((oflags & O_ACCMODE) == O_ACCMODE)
1429 		return (EINVAL);
1430 	flags = FFLAGS(oflags);
1431 	error = falloc(p, &nfp, NULL);
1432 	if (error)
1433 		return (error);
1434 	fp = nfp;
1435 	cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) &~ S_ISTXT;
1436 
1437 	/*
1438 	 * XXX p_dupfd is a real mess.  It allows a device to return a
1439 	 * file descriptor to be duplicated rather then doing the open
1440 	 * itself.
1441 	 */
1442 	lp->lwp_dupfd = -1;
1443 
1444 	/*
1445 	 * Call vn_open() to do the lookup and assign the vnode to the
1446 	 * file pointer.  vn_open() does not change the ref count on fp
1447 	 * and the vnode, on success, will be inherited by the file pointer
1448 	 * and unlocked.
1449 	 */
1450 	nd->nl_flags |= NLC_LOCKVP;
1451 	error = vn_open(nd, fp, flags, cmode);
1452 	nlookup_done(nd);
1453 	if (error) {
1454 		/*
1455 		 * handle special fdopen() case.  bleh.  dupfdopen() is
1456 		 * responsible for dropping the old contents of ofiles[indx]
1457 		 * if it succeeds.
1458 		 *
1459 		 * Note that fsetfd() will add a ref to fp which represents
1460 		 * the fd_files[] assignment.  We must still drop our
1461 		 * reference.
1462 		 */
1463 		if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1464 			if (fdalloc(p, 0, &indx) == 0) {
1465 				error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error);
1466 				if (error == 0) {
1467 					*res = indx;
1468 					fdrop(fp);	/* our ref */
1469 					return (0);
1470 				}
1471 				fsetfd(p, NULL, indx);
1472 			}
1473 		}
1474 		fdrop(fp);	/* our ref */
1475 		if (error == ERESTART)
1476 			error = EINTR;
1477 		return (error);
1478 	}
1479 
1480 	/*
1481 	 * ref the vnode for ourselves so it can't be ripped out from under
1482 	 * is.  XXX need an ND flag to request that the vnode be returned
1483 	 * anyway.
1484 	 *
1485 	 * Reserve a file descriptor but do not assign it until the open
1486 	 * succeeds.
1487 	 */
1488 	vp = (struct vnode *)fp->f_data;
1489 	vref(vp);
1490 	if ((error = fdalloc(p, 0, &indx)) != 0) {
1491 		fdrop(fp);
1492 		vrele(vp);
1493 		return (error);
1494 	}
1495 
1496 	/*
1497 	 * If no error occurs the vp will have been assigned to the file
1498 	 * pointer.
1499 	 */
1500 	lp->lwp_dupfd = 0;
1501 
1502 	if (flags & (O_EXLOCK | O_SHLOCK)) {
1503 		lf.l_whence = SEEK_SET;
1504 		lf.l_start = 0;
1505 		lf.l_len = 0;
1506 		if (flags & O_EXLOCK)
1507 			lf.l_type = F_WRLCK;
1508 		else
1509 			lf.l_type = F_RDLCK;
1510 		if (flags & FNONBLOCK)
1511 			type = 0;
1512 		else
1513 			type = F_WAIT;
1514 
1515 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1516 			/*
1517 			 * lock request failed.  Clean up the reserved
1518 			 * descriptor.
1519 			 */
1520 			vrele(vp);
1521 			fsetfd(p, NULL, indx);
1522 			fdrop(fp);
1523 			return (error);
1524 		}
1525 		fp->f_flag |= FHASLOCK;
1526 	}
1527 #if 0
1528 	/*
1529 	 * Assert that all regular file vnodes were created with a object.
1530 	 */
1531 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1532 		("open: regular file has no backing object after vn_open"));
1533 #endif
1534 
1535 	vrele(vp);
1536 
1537 	/*
1538 	 * release our private reference, leaving the one associated with the
1539 	 * descriptor table intact.
1540 	 */
1541 	fsetfd(p, fp, indx);
1542 	fdrop(fp);
1543 	*res = indx;
1544 	return (0);
1545 }
1546 
1547 /*
1548  * open_args(char *path, int flags, int mode)
1549  *
1550  * Check permissions, allocate an open file structure,
1551  * and call the device open routine if any.
1552  */
1553 int
1554 sys_open(struct open_args *uap)
1555 {
1556 	struct nlookupdata nd;
1557 	int error;
1558 
1559 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1560 	if (error == 0) {
1561 		error = kern_open(&nd, uap->flags,
1562 				    uap->mode, &uap->sysmsg_result);
1563 	}
1564 	nlookup_done(&nd);
1565 	return (error);
1566 }
1567 
1568 int
1569 kern_mknod(struct nlookupdata *nd, int mode, int dev)
1570 {
1571 	struct thread *td = curthread;
1572 	struct proc *p = td->td_proc;
1573 	struct vnode *vp;
1574 	struct vattr vattr;
1575 	int error;
1576 	int whiteout = 0;
1577 
1578 	KKASSERT(p);
1579 
1580 	switch (mode & S_IFMT) {
1581 	case S_IFCHR:
1582 	case S_IFBLK:
1583 		error = suser(td);
1584 		break;
1585 	default:
1586 		error = suser_cred(p->p_ucred, PRISON_ROOT);
1587 		break;
1588 	}
1589 	if (error)
1590 		return (error);
1591 
1592 	bwillwrite();
1593 	nd->nl_flags |= NLC_CREATE;
1594 	if ((error = nlookup(nd)) != 0)
1595 		return (error);
1596 	if (nd->nl_nch.ncp->nc_vp)
1597 		return (EEXIST);
1598 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1599 		return (error);
1600 
1601 	VATTR_NULL(&vattr);
1602 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1603 	vattr.va_rdev = dev;
1604 	whiteout = 0;
1605 
1606 	switch (mode & S_IFMT) {
1607 	case S_IFMT:	/* used by badsect to flag bad sectors */
1608 		vattr.va_type = VBAD;
1609 		break;
1610 	case S_IFCHR:
1611 		vattr.va_type = VCHR;
1612 		break;
1613 	case S_IFBLK:
1614 		vattr.va_type = VBLK;
1615 		break;
1616 	case S_IFWHT:
1617 		whiteout = 1;
1618 		break;
1619 	default:
1620 		error = EINVAL;
1621 		break;
1622 	}
1623 	if (error == 0) {
1624 		if (whiteout) {
1625 			error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_cred, NAMEI_CREATE);
1626 		} else {
1627 			vp = NULL;
1628 			error = VOP_NMKNOD(&nd->nl_nch, &vp, nd->nl_cred, &vattr);
1629 			if (error == 0)
1630 				vput(vp);
1631 		}
1632 	}
1633 	return (error);
1634 }
1635 
1636 /*
1637  * mknod_args(char *path, int mode, int dev)
1638  *
1639  * Create a special file.
1640  */
1641 int
1642 sys_mknod(struct mknod_args *uap)
1643 {
1644 	struct nlookupdata nd;
1645 	int error;
1646 
1647 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1648 	if (error == 0)
1649 		error = kern_mknod(&nd, uap->mode, uap->dev);
1650 	nlookup_done(&nd);
1651 	return (error);
1652 }
1653 
1654 int
1655 kern_mkfifo(struct nlookupdata *nd, int mode)
1656 {
1657 	struct thread *td = curthread;
1658 	struct proc *p = td->td_proc;
1659 	struct vattr vattr;
1660 	struct vnode *vp;
1661 	int error;
1662 
1663 	bwillwrite();
1664 
1665 	nd->nl_flags |= NLC_CREATE;
1666 	if ((error = nlookup(nd)) != 0)
1667 		return (error);
1668 	if (nd->nl_nch.ncp->nc_vp)
1669 		return (EEXIST);
1670 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1671 		return (error);
1672 
1673 	VATTR_NULL(&vattr);
1674 	vattr.va_type = VFIFO;
1675 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1676 	vp = NULL;
1677 	error = VOP_NMKNOD(&nd->nl_nch, &vp, nd->nl_cred, &vattr);
1678 	if (error == 0)
1679 		vput(vp);
1680 	return (error);
1681 }
1682 
1683 /*
1684  * mkfifo_args(char *path, int mode)
1685  *
1686  * Create a named pipe.
1687  */
1688 int
1689 sys_mkfifo(struct mkfifo_args *uap)
1690 {
1691 	struct nlookupdata nd;
1692 	int error;
1693 
1694 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1695 	if (error == 0)
1696 		error = kern_mkfifo(&nd, uap->mode);
1697 	nlookup_done(&nd);
1698 	return (error);
1699 }
1700 
1701 static int hardlink_check_uid = 0;
1702 SYSCTL_INT(_kern, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1703     &hardlink_check_uid, 0,
1704     "Unprivileged processes cannot create hard links to files owned by other "
1705     "users");
1706 static int hardlink_check_gid = 0;
1707 SYSCTL_INT(_kern, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
1708     &hardlink_check_gid, 0,
1709     "Unprivileged processes cannot create hard links to files owned by other "
1710     "groups");
1711 
1712 static int
1713 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
1714 {
1715 	struct vattr va;
1716 	int error;
1717 
1718 	/*
1719 	 * Shortcut if disabled
1720 	 */
1721 	if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
1722 		return (0);
1723 
1724 	/*
1725 	 * root cred can always hardlink
1726 	 */
1727 	if (suser_cred(cred, PRISON_ROOT) == 0)
1728 		return (0);
1729 
1730 	/*
1731 	 * Otherwise only if the originating file is owned by the
1732 	 * same user or group.  Note that any group is allowed if
1733 	 * the file is owned by the caller.
1734 	 */
1735 	error = VOP_GETATTR(vp, &va);
1736 	if (error != 0)
1737 		return (error);
1738 
1739 	if (hardlink_check_uid) {
1740 		if (cred->cr_uid != va.va_uid)
1741 			return (EPERM);
1742 	}
1743 
1744 	if (hardlink_check_gid) {
1745 		if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
1746 			return (EPERM);
1747 	}
1748 
1749 	return (0);
1750 }
1751 
1752 int
1753 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
1754 {
1755 	struct thread *td = curthread;
1756 	struct vnode *vp;
1757 	int error;
1758 
1759 	/*
1760 	 * Lookup the source and obtained a locked vnode.
1761 	 *
1762 	 * XXX relookup on vget failure / race ?
1763 	 */
1764 	bwillwrite();
1765 	if ((error = nlookup(nd)) != 0)
1766 		return (error);
1767 	vp = nd->nl_nch.ncp->nc_vp;
1768 	KKASSERT(vp != NULL);
1769 	if (vp->v_type == VDIR)
1770 		return (EPERM);		/* POSIX */
1771 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1772 		return (error);
1773 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
1774 		return (error);
1775 
1776 	/*
1777 	 * Unlock the source so we can lookup the target without deadlocking
1778 	 * (XXX vp is locked already, possible other deadlock?).  The target
1779 	 * must not exist.
1780 	 */
1781 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
1782 	nd->nl_flags &= ~NLC_NCPISLOCKED;
1783 	cache_unlock(&nd->nl_nch);
1784 
1785 	linknd->nl_flags |= NLC_CREATE;
1786 	if ((error = nlookup(linknd)) != 0) {
1787 		vput(vp);
1788 		return (error);
1789 	}
1790 	if (linknd->nl_nch.ncp->nc_vp) {
1791 		vput(vp);
1792 		return (EEXIST);
1793 	}
1794 
1795 	/*
1796 	 * Finally run the new API VOP.
1797 	 */
1798 	error = can_hardlink(vp, td, td->td_proc->p_ucred);
1799 	if (error == 0)
1800 		error = VOP_NLINK(&linknd->nl_nch, vp, linknd->nl_cred);
1801 	vput(vp);
1802 	return (error);
1803 }
1804 
1805 /*
1806  * link_args(char *path, char *link)
1807  *
1808  * Make a hard file link.
1809  */
1810 int
1811 sys_link(struct link_args *uap)
1812 {
1813 	struct nlookupdata nd, linknd;
1814 	int error;
1815 
1816 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1817 	if (error == 0) {
1818 		error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
1819 		if (error == 0)
1820 			error = kern_link(&nd, &linknd);
1821 		nlookup_done(&linknd);
1822 	}
1823 	nlookup_done(&nd);
1824 	return (error);
1825 }
1826 
1827 int
1828 kern_symlink(struct nlookupdata *nd, char *path, int mode)
1829 {
1830 	struct vattr vattr;
1831 	struct vnode *vp;
1832 	int error;
1833 
1834 	bwillwrite();
1835 	nd->nl_flags |= NLC_CREATE;
1836 	if ((error = nlookup(nd)) != 0)
1837 		return (error);
1838 	if (nd->nl_nch.ncp->nc_vp)
1839 		return (EEXIST);
1840 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1841 		return (error);
1842 	VATTR_NULL(&vattr);
1843 	vattr.va_mode = mode;
1844 	error = VOP_NSYMLINK(&nd->nl_nch, &vp, nd->nl_cred, &vattr, path);
1845 	if (error == 0)
1846 		vput(vp);
1847 	return (error);
1848 }
1849 
1850 /*
1851  * symlink(char *path, char *link)
1852  *
1853  * Make a symbolic link.
1854  */
1855 int
1856 sys_symlink(struct symlink_args *uap)
1857 {
1858 	struct thread *td = curthread;
1859 	struct nlookupdata nd;
1860 	char *path;
1861 	int error;
1862 	int mode;
1863 
1864 	path = objcache_get(namei_oc, M_WAITOK);
1865 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
1866 	if (error == 0) {
1867 		error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
1868 		if (error == 0) {
1869 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
1870 			error = kern_symlink(&nd, path, mode);
1871 		}
1872 		nlookup_done(&nd);
1873 	}
1874 	objcache_put(namei_oc, path);
1875 	return (error);
1876 }
1877 
1878 /*
1879  * undelete_args(char *path)
1880  *
1881  * Delete a whiteout from the filesystem.
1882  */
1883 /* ARGSUSED */
1884 int
1885 sys_undelete(struct undelete_args *uap)
1886 {
1887 	struct nlookupdata nd;
1888 	int error;
1889 
1890 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1891 	bwillwrite();
1892 	nd.nl_flags |= NLC_DELETE;
1893 	if (error == 0)
1894 		error = nlookup(&nd);
1895 	if (error == 0)
1896 		error = ncp_writechk(&nd.nl_nch);
1897 	if (error == 0)
1898 		error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_cred, NAMEI_DELETE);
1899 	nlookup_done(&nd);
1900 	return (error);
1901 }
1902 
1903 int
1904 kern_unlink(struct nlookupdata *nd)
1905 {
1906 	int error;
1907 
1908 	bwillwrite();
1909 	nd->nl_flags |= NLC_DELETE;
1910 	if ((error = nlookup(nd)) != 0)
1911 		return (error);
1912 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1913 		return (error);
1914 	error = VOP_NREMOVE(&nd->nl_nch, nd->nl_cred);
1915 	return (error);
1916 }
1917 
1918 /*
1919  * unlink_args(char *path)
1920  *
1921  * Delete a name from the filesystem.
1922  */
1923 int
1924 sys_unlink(struct unlink_args *uap)
1925 {
1926 	struct nlookupdata nd;
1927 	int error;
1928 
1929 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1930 	if (error == 0)
1931 		error = kern_unlink(&nd);
1932 	nlookup_done(&nd);
1933 	return (error);
1934 }
1935 
1936 int
1937 kern_lseek(int fd, off_t offset, int whence, off_t *res)
1938 {
1939 	struct thread *td = curthread;
1940 	struct proc *p = td->td_proc;
1941 	struct file *fp;
1942 	struct vattr vattr;
1943 	int error;
1944 
1945 	fp = holdfp(p->p_fd, fd, -1);
1946 	if (fp == NULL)
1947 		return (EBADF);
1948 	if (fp->f_type != DTYPE_VNODE) {
1949 		error = ESPIPE;
1950 		goto done;
1951 	}
1952 
1953 	switch (whence) {
1954 	case L_INCR:
1955 		fp->f_offset += offset;
1956 		error = 0;
1957 		break;
1958 	case L_XTND:
1959 		error = VOP_GETATTR((struct vnode *)fp->f_data, &vattr);
1960 		if (error == 0)
1961 			fp->f_offset = offset + vattr.va_size;
1962 		break;
1963 	case L_SET:
1964 		fp->f_offset = offset;
1965 		error = 0;
1966 		break;
1967 	default:
1968 		error = EINVAL;
1969 		break;
1970 	}
1971 	*res = fp->f_offset;
1972 done:
1973 	fdrop(fp);
1974 	return (error);
1975 }
1976 
1977 /*
1978  * lseek_args(int fd, int pad, off_t offset, int whence)
1979  *
1980  * Reposition read/write file offset.
1981  */
1982 int
1983 sys_lseek(struct lseek_args *uap)
1984 {
1985 	int error;
1986 
1987 	error = kern_lseek(uap->fd, uap->offset, uap->whence,
1988 	    &uap->sysmsg_offset);
1989 
1990 	return (error);
1991 }
1992 
1993 int
1994 kern_access(struct nlookupdata *nd, int aflags)
1995 {
1996 	struct vnode *vp;
1997 	int error, flags;
1998 
1999 	if ((error = nlookup(nd)) != 0)
2000 		return (error);
2001 retry:
2002 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2003 	if (error)
2004 		return (error);
2005 
2006 	/* Flags == 0 means only check for existence. */
2007 	if (aflags) {
2008 		flags = 0;
2009 		if (aflags & R_OK)
2010 			flags |= VREAD;
2011 		if (aflags & W_OK)
2012 			flags |= VWRITE;
2013 		if (aflags & X_OK)
2014 			flags |= VEXEC;
2015 		if ((flags & VWRITE) == 0 ||
2016 		    (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2017 			error = VOP_ACCESS(vp, flags, nd->nl_cred);
2018 
2019 		/*
2020 		 * If the file handle is stale we have to re-resolve the
2021 		 * entry.  This is a hack at the moment.
2022 		 */
2023 		if (error == ESTALE) {
2024 			cache_setunresolved(&nd->nl_nch);
2025 			error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2026 			if (error == 0) {
2027 				vput(vp);
2028 				vp = NULL;
2029 				goto retry;
2030 			}
2031 		}
2032 	}
2033 	vput(vp);
2034 	return (error);
2035 }
2036 
2037 /*
2038  * access_args(char *path, int flags)
2039  *
2040  * Check access permissions.
2041  */
2042 int
2043 sys_access(struct access_args *uap)
2044 {
2045 	struct nlookupdata nd;
2046 	int error;
2047 
2048 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2049 	if (error == 0)
2050 		error = kern_access(&nd, uap->flags);
2051 	nlookup_done(&nd);
2052 	return (error);
2053 }
2054 
2055 int
2056 kern_stat(struct nlookupdata *nd, struct stat *st)
2057 {
2058 	int error;
2059 	struct vnode *vp;
2060 	thread_t td;
2061 
2062 	if ((error = nlookup(nd)) != 0)
2063 		return (error);
2064 again:
2065 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2066 		return (ENOENT);
2067 
2068 	td = curthread;
2069 	if ((error = vget(vp, LK_SHARED)) != 0)
2070 		return (error);
2071 	error = vn_stat(vp, st, nd->nl_cred);
2072 
2073 	/*
2074 	 * If the file handle is stale we have to re-resolve the entry.  This
2075 	 * is a hack at the moment.
2076 	 */
2077 	if (error == ESTALE) {
2078 		cache_setunresolved(&nd->nl_nch);
2079 		error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2080 		if (error == 0) {
2081 			vput(vp);
2082 			goto again;
2083 		}
2084 	}
2085 	vput(vp);
2086 	return (error);
2087 }
2088 
2089 /*
2090  * stat_args(char *path, struct stat *ub)
2091  *
2092  * Get file status; this version follows links.
2093  */
2094 int
2095 sys_stat(struct stat_args *uap)
2096 {
2097 	struct nlookupdata nd;
2098 	struct stat st;
2099 	int error;
2100 
2101 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2102 	if (error == 0) {
2103 		error = kern_stat(&nd, &st);
2104 		if (error == 0)
2105 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2106 	}
2107 	nlookup_done(&nd);
2108 	return (error);
2109 }
2110 
2111 /*
2112  * lstat_args(char *path, struct stat *ub)
2113  *
2114  * Get file status; this version does not follow links.
2115  */
2116 int
2117 sys_lstat(struct lstat_args *uap)
2118 {
2119 	struct nlookupdata nd;
2120 	struct stat st;
2121 	int error;
2122 
2123 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2124 	if (error == 0) {
2125 		error = kern_stat(&nd, &st);
2126 		if (error == 0)
2127 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2128 	}
2129 	nlookup_done(&nd);
2130 	return (error);
2131 }
2132 
2133 /*
2134  * pathconf_Args(char *path, int name)
2135  *
2136  * Get configurable pathname variables.
2137  */
2138 /* ARGSUSED */
2139 int
2140 sys_pathconf(struct pathconf_args *uap)
2141 {
2142 	struct nlookupdata nd;
2143 	struct vnode *vp;
2144 	int error;
2145 
2146 	vp = NULL;
2147 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2148 	if (error == 0)
2149 		error = nlookup(&nd);
2150 	if (error == 0)
2151 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2152 	nlookup_done(&nd);
2153 	if (error == 0) {
2154 		error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
2155 		vput(vp);
2156 	}
2157 	return (error);
2158 }
2159 
2160 /*
2161  * XXX: daver
2162  * kern_readlink isn't properly split yet.  There is a copyin burried
2163  * in VOP_READLINK().
2164  */
2165 int
2166 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2167 {
2168 	struct thread *td = curthread;
2169 	struct proc *p = td->td_proc;
2170 	struct vnode *vp;
2171 	struct iovec aiov;
2172 	struct uio auio;
2173 	int error;
2174 
2175 	if ((error = nlookup(nd)) != 0)
2176 		return (error);
2177 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2178 	if (error)
2179 		return (error);
2180 	if (vp->v_type != VLNK) {
2181 		error = EINVAL;
2182 	} else {
2183 		aiov.iov_base = buf;
2184 		aiov.iov_len = count;
2185 		auio.uio_iov = &aiov;
2186 		auio.uio_iovcnt = 1;
2187 		auio.uio_offset = 0;
2188 		auio.uio_rw = UIO_READ;
2189 		auio.uio_segflg = UIO_USERSPACE;
2190 		auio.uio_td = td;
2191 		auio.uio_resid = count;
2192 		error = VOP_READLINK(vp, &auio, p->p_ucred);
2193 	}
2194 	vput(vp);
2195 	*res = count - auio.uio_resid;
2196 	return (error);
2197 }
2198 
2199 /*
2200  * readlink_args(char *path, char *buf, int count)
2201  *
2202  * Return target name of a symbolic link.
2203  */
2204 int
2205 sys_readlink(struct readlink_args *uap)
2206 {
2207 	struct nlookupdata nd;
2208 	int error;
2209 
2210 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2211 	if (error == 0) {
2212 		error = kern_readlink(&nd, uap->buf, uap->count,
2213 					&uap->sysmsg_result);
2214 	}
2215 	nlookup_done(&nd);
2216 	return (error);
2217 }
2218 
2219 static int
2220 setfflags(struct vnode *vp, int flags)
2221 {
2222 	struct thread *td = curthread;
2223 	struct proc *p = td->td_proc;
2224 	int error;
2225 	struct vattr vattr;
2226 
2227 	/*
2228 	 * Prevent non-root users from setting flags on devices.  When
2229 	 * a device is reused, users can retain ownership of the device
2230 	 * if they are allowed to set flags and programs assume that
2231 	 * chown can't fail when done as root.
2232 	 */
2233 	if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2234 	    ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0))
2235 		return (error);
2236 
2237 	/*
2238 	 * note: vget is required for any operation that might mod the vnode
2239 	 * so VINACTIVE is properly cleared.
2240 	 */
2241 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2242 		VATTR_NULL(&vattr);
2243 		vattr.va_flags = flags;
2244 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2245 		vput(vp);
2246 	}
2247 	return (error);
2248 }
2249 
2250 /*
2251  * chflags(char *path, int flags)
2252  *
2253  * Change flags of a file given a path name.
2254  */
2255 /* ARGSUSED */
2256 int
2257 sys_chflags(struct chflags_args *uap)
2258 {
2259 	struct nlookupdata nd;
2260 	struct vnode *vp;
2261 	int error;
2262 
2263 	vp = NULL;
2264 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2265 	/* XXX Add NLC flag indicating modifying operation? */
2266 	if (error == 0)
2267 		error = nlookup(&nd);
2268 	if (error == 0)
2269 		error = ncp_writechk(&nd.nl_nch);
2270 	if (error == 0)
2271 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2272 	nlookup_done(&nd);
2273 	if (error == 0) {
2274 		error = setfflags(vp, uap->flags);
2275 		vrele(vp);
2276 	}
2277 	return (error);
2278 }
2279 
2280 /*
2281  * fchflags_args(int fd, int flags)
2282  *
2283  * Change flags of a file given a file descriptor.
2284  */
2285 /* ARGSUSED */
2286 int
2287 sys_fchflags(struct fchflags_args *uap)
2288 {
2289 	struct thread *td = curthread;
2290 	struct proc *p = td->td_proc;
2291 	struct file *fp;
2292 	int error;
2293 
2294 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2295 		return (error);
2296 	if (fp->f_nchandle.ncp)
2297 		error = ncp_writechk(&fp->f_nchandle);
2298 	if (error == 0)
2299 		error = setfflags((struct vnode *) fp->f_data, uap->flags);
2300 	fdrop(fp);
2301 	return (error);
2302 }
2303 
2304 static int
2305 setfmode(struct vnode *vp, int mode)
2306 {
2307 	struct thread *td = curthread;
2308 	struct proc *p = td->td_proc;
2309 	int error;
2310 	struct vattr vattr;
2311 
2312 	/*
2313 	 * note: vget is required for any operation that might mod the vnode
2314 	 * so VINACTIVE is properly cleared.
2315 	 */
2316 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2317 		VATTR_NULL(&vattr);
2318 		vattr.va_mode = mode & ALLPERMS;
2319 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2320 		vput(vp);
2321 	}
2322 	return error;
2323 }
2324 
2325 int
2326 kern_chmod(struct nlookupdata *nd, int mode)
2327 {
2328 	struct vnode *vp;
2329 	int error;
2330 
2331 	/* XXX Add NLC flag indicating modifying operation? */
2332 	if ((error = nlookup(nd)) != 0)
2333 		return (error);
2334 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2335 		return (error);
2336 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2337 		error = setfmode(vp, mode);
2338 	vrele(vp);
2339 	return (error);
2340 }
2341 
2342 /*
2343  * chmod_args(char *path, int mode)
2344  *
2345  * Change mode of a file given path name.
2346  */
2347 /* ARGSUSED */
2348 int
2349 sys_chmod(struct chmod_args *uap)
2350 {
2351 	struct nlookupdata nd;
2352 	int error;
2353 
2354 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2355 	if (error == 0)
2356 		error = kern_chmod(&nd, uap->mode);
2357 	nlookup_done(&nd);
2358 	return (error);
2359 }
2360 
2361 /*
2362  * lchmod_args(char *path, int mode)
2363  *
2364  * Change mode of a file given path name (don't follow links.)
2365  */
2366 /* ARGSUSED */
2367 int
2368 sys_lchmod(struct lchmod_args *uap)
2369 {
2370 	struct nlookupdata nd;
2371 	int error;
2372 
2373 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2374 	if (error == 0)
2375 		error = kern_chmod(&nd, uap->mode);
2376 	nlookup_done(&nd);
2377 	return (error);
2378 }
2379 
2380 /*
2381  * fchmod_args(int fd, int mode)
2382  *
2383  * Change mode of a file given a file descriptor.
2384  */
2385 /* ARGSUSED */
2386 int
2387 sys_fchmod(struct fchmod_args *uap)
2388 {
2389 	struct thread *td = curthread;
2390 	struct proc *p = td->td_proc;
2391 	struct file *fp;
2392 	int error;
2393 
2394 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2395 		return (error);
2396 	if (fp->f_nchandle.ncp)
2397 		error = ncp_writechk(&fp->f_nchandle);
2398 	if (error == 0)
2399 		error = setfmode((struct vnode *)fp->f_data, uap->mode);
2400 	fdrop(fp);
2401 	return (error);
2402 }
2403 
2404 static int
2405 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2406 {
2407 	struct thread *td = curthread;
2408 	struct proc *p = td->td_proc;
2409 	int error;
2410 	struct vattr vattr;
2411 
2412 	/*
2413 	 * note: vget is required for any operation that might mod the vnode
2414 	 * so VINACTIVE is properly cleared.
2415 	 */
2416 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2417 		VATTR_NULL(&vattr);
2418 		vattr.va_uid = uid;
2419 		vattr.va_gid = gid;
2420 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2421 		vput(vp);
2422 	}
2423 	return error;
2424 }
2425 
2426 int
2427 kern_chown(struct nlookupdata *nd, int uid, int gid)
2428 {
2429 	struct vnode *vp;
2430 	int error;
2431 
2432 	/* XXX Add NLC flag indicating modifying operation? */
2433 	if ((error = nlookup(nd)) != 0)
2434 		return (error);
2435 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2436 		return (error);
2437 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2438 		error = setfown(vp, uid, gid);
2439 	vrele(vp);
2440 	return (error);
2441 }
2442 
2443 /*
2444  * chown(char *path, int uid, int gid)
2445  *
2446  * Set ownership given a path name.
2447  */
2448 int
2449 sys_chown(struct chown_args *uap)
2450 {
2451 	struct nlookupdata nd;
2452 	int error;
2453 
2454 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2455 	if (error == 0)
2456 		error = kern_chown(&nd, uap->uid, uap->gid);
2457 	nlookup_done(&nd);
2458 	return (error);
2459 }
2460 
2461 /*
2462  * lchown_args(char *path, int uid, int gid)
2463  *
2464  * Set ownership given a path name, do not cross symlinks.
2465  */
2466 int
2467 sys_lchown(struct lchown_args *uap)
2468 {
2469 	struct nlookupdata nd;
2470 	int error;
2471 
2472 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2473 	if (error == 0)
2474 		error = kern_chown(&nd, uap->uid, uap->gid);
2475 	nlookup_done(&nd);
2476 	return (error);
2477 }
2478 
2479 /*
2480  * fchown_args(int fd, int uid, int gid)
2481  *
2482  * Set ownership given a file descriptor.
2483  */
2484 /* ARGSUSED */
2485 int
2486 sys_fchown(struct fchown_args *uap)
2487 {
2488 	struct thread *td = curthread;
2489 	struct proc *p = td->td_proc;
2490 	struct file *fp;
2491 	int error;
2492 
2493 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2494 		return (error);
2495 	if (fp->f_nchandle.ncp)
2496 		error = ncp_writechk(&fp->f_nchandle);
2497 	if (error == 0)
2498 		error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
2499 	fdrop(fp);
2500 	return (error);
2501 }
2502 
2503 static int
2504 getutimes(const struct timeval *tvp, struct timespec *tsp)
2505 {
2506 	struct timeval tv[2];
2507 
2508 	if (tvp == NULL) {
2509 		microtime(&tv[0]);
2510 		TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
2511 		tsp[1] = tsp[0];
2512 	} else {
2513 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
2514 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
2515 	}
2516 	return 0;
2517 }
2518 
2519 static int
2520 setutimes(struct vnode *vp, const struct timespec *ts, int nullflag)
2521 {
2522 	struct thread *td = curthread;
2523 	struct proc *p = td->td_proc;
2524 	int error;
2525 	struct vattr vattr;
2526 
2527 	/*
2528 	 * note: vget is required for any operation that might mod the vnode
2529 	 * so VINACTIVE is properly cleared.
2530 	 */
2531 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2532 		VATTR_NULL(&vattr);
2533 		vattr.va_atime = ts[0];
2534 		vattr.va_mtime = ts[1];
2535 		if (nullflag)
2536 			vattr.va_vaflags |= VA_UTIMES_NULL;
2537 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2538 		vput(vp);
2539 	}
2540 	return error;
2541 }
2542 
2543 int
2544 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
2545 {
2546 	struct timespec ts[2];
2547 	struct vnode *vp;
2548 	int error;
2549 
2550 	if ((error = getutimes(tptr, ts)) != 0)
2551 		return (error);
2552 	/* XXX Add NLC flag indicating modifying operation? */
2553 	if ((error = nlookup(nd)) != 0)
2554 		return (error);
2555 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2556 		return (error);
2557 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2558 		return (error);
2559 	error = setutimes(vp, ts, tptr == NULL);
2560 	vrele(vp);
2561 	return (error);
2562 }
2563 
2564 /*
2565  * utimes_args(char *path, struct timeval *tptr)
2566  *
2567  * Set the access and modification times of a file.
2568  */
2569 int
2570 sys_utimes(struct utimes_args *uap)
2571 {
2572 	struct timeval tv[2];
2573 	struct nlookupdata nd;
2574 	int error;
2575 
2576 	if (uap->tptr) {
2577  		error = copyin(uap->tptr, tv, sizeof(tv));
2578 		if (error)
2579 			return (error);
2580 	}
2581 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2582 	if (error == 0)
2583 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2584 	nlookup_done(&nd);
2585 	return (error);
2586 }
2587 
2588 /*
2589  * lutimes_args(char *path, struct timeval *tptr)
2590  *
2591  * Set the access and modification times of a file.
2592  */
2593 int
2594 sys_lutimes(struct lutimes_args *uap)
2595 {
2596 	struct timeval tv[2];
2597 	struct nlookupdata nd;
2598 	int error;
2599 
2600 	if (uap->tptr) {
2601 		error = copyin(uap->tptr, tv, sizeof(tv));
2602 		if (error)
2603 			return (error);
2604 	}
2605 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2606 	if (error == 0)
2607 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2608 	nlookup_done(&nd);
2609 	return (error);
2610 }
2611 
2612 int
2613 kern_futimes(int fd, struct timeval *tptr)
2614 {
2615 	struct thread *td = curthread;
2616 	struct proc *p = td->td_proc;
2617 	struct timespec ts[2];
2618 	struct file *fp;
2619 	int error;
2620 
2621 	error = getutimes(tptr, ts);
2622 	if (error)
2623 		return (error);
2624 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2625 		return (error);
2626 	if (fp->f_nchandle.ncp)
2627 		error = ncp_writechk(&fp->f_nchandle);
2628 	if (error == 0)
2629 		error =  setutimes((struct vnode *)fp->f_data, ts, tptr == NULL);
2630 	fdrop(fp);
2631 	return (error);
2632 }
2633 
2634 /*
2635  * futimes_args(int fd, struct timeval *tptr)
2636  *
2637  * Set the access and modification times of a file.
2638  */
2639 int
2640 sys_futimes(struct futimes_args *uap)
2641 {
2642 	struct timeval tv[2];
2643 	int error;
2644 
2645 	if (uap->tptr) {
2646 		error = copyin(uap->tptr, tv, sizeof(tv));
2647 		if (error)
2648 			return (error);
2649 	}
2650 
2651 	error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
2652 
2653 	return (error);
2654 }
2655 
2656 int
2657 kern_truncate(struct nlookupdata *nd, off_t length)
2658 {
2659 	struct vnode *vp;
2660 	struct vattr vattr;
2661 	int error;
2662 
2663 	if (length < 0)
2664 		return(EINVAL);
2665 	/* XXX Add NLC flag indicating modifying operation? */
2666 	if ((error = nlookup(nd)) != 0)
2667 		return (error);
2668 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2669 		return (error);
2670 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2671 		return (error);
2672 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
2673 		vrele(vp);
2674 		return (error);
2675 	}
2676 	if (vp->v_type == VDIR) {
2677 		error = EISDIR;
2678 	} else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0 &&
2679 	    (error = VOP_ACCESS(vp, VWRITE, nd->nl_cred)) == 0) {
2680 		VATTR_NULL(&vattr);
2681 		vattr.va_size = length;
2682 		error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
2683 	}
2684 	vput(vp);
2685 	return (error);
2686 }
2687 
2688 /*
2689  * truncate(char *path, int pad, off_t length)
2690  *
2691  * Truncate a file given its path name.
2692  */
2693 int
2694 sys_truncate(struct truncate_args *uap)
2695 {
2696 	struct nlookupdata nd;
2697 	int error;
2698 
2699 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2700 	if (error == 0)
2701 		error = kern_truncate(&nd, uap->length);
2702 	nlookup_done(&nd);
2703 	return error;
2704 }
2705 
2706 int
2707 kern_ftruncate(int fd, off_t length)
2708 {
2709 	struct thread *td = curthread;
2710 	struct proc *p = td->td_proc;
2711 	struct vattr vattr;
2712 	struct vnode *vp;
2713 	struct file *fp;
2714 	int error;
2715 
2716 	if (length < 0)
2717 		return(EINVAL);
2718 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2719 		return (error);
2720 	if (fp->f_nchandle.ncp) {
2721 		error = ncp_writechk(&fp->f_nchandle);
2722 		if (error)
2723 			goto done;
2724 	}
2725 	if ((fp->f_flag & FWRITE) == 0) {
2726 		error = EINVAL;
2727 		goto done;
2728 	}
2729 	vp = (struct vnode *)fp->f_data;
2730 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2731 	if (vp->v_type == VDIR) {
2732 		error = EISDIR;
2733 	} else if ((error = vn_writechk(vp, NULL)) == 0) {
2734 		VATTR_NULL(&vattr);
2735 		vattr.va_size = length;
2736 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
2737 	}
2738 	vn_unlock(vp);
2739 done:
2740 	fdrop(fp);
2741 	return (error);
2742 }
2743 
2744 /*
2745  * ftruncate_args(int fd, int pad, off_t length)
2746  *
2747  * Truncate a file given a file descriptor.
2748  */
2749 int
2750 sys_ftruncate(struct ftruncate_args *uap)
2751 {
2752 	int error;
2753 
2754 	error = kern_ftruncate(uap->fd, uap->length);
2755 
2756 	return (error);
2757 }
2758 
2759 /*
2760  * fsync(int fd)
2761  *
2762  * Sync an open file.
2763  */
2764 /* ARGSUSED */
2765 int
2766 sys_fsync(struct fsync_args *uap)
2767 {
2768 	struct thread *td = curthread;
2769 	struct proc *p = td->td_proc;
2770 	struct vnode *vp;
2771 	struct file *fp;
2772 	vm_object_t obj;
2773 	int error;
2774 
2775 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2776 		return (error);
2777 	vp = (struct vnode *)fp->f_data;
2778 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2779 	if ((obj = vp->v_object) != NULL)
2780 		vm_object_page_clean(obj, 0, 0, 0);
2781 	if ((error = VOP_FSYNC(vp, MNT_WAIT)) == 0 &&
2782 	    vp->v_mount && (vp->v_mount->mnt_flag & MNT_SOFTDEP) &&
2783 	    bioops.io_fsync) {
2784 		error = (*bioops.io_fsync)(vp);
2785 	}
2786 	vn_unlock(vp);
2787 	fdrop(fp);
2788 	return (error);
2789 }
2790 
2791 int
2792 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
2793 {
2794 	struct nchandle fnchd;
2795 	struct nchandle tnchd;
2796 	struct namecache *ncp;
2797 	struct mount *mp;
2798 	int error;
2799 
2800 	bwillwrite();
2801 	if ((error = nlookup(fromnd)) != 0)
2802 		return (error);
2803 	if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
2804 		return (ENOENT);
2805 	fnchd.mount = fromnd->nl_nch.mount;
2806 	cache_hold(&fnchd);
2807 
2808 	/*
2809 	 * unlock the source nch so we can lookup the target nch without
2810 	 * deadlocking.  The target may or may not exist so we do not check
2811 	 * for a target vp like kern_mkdir() and other creation functions do.
2812 	 *
2813 	 * The source and target directories are ref'd and rechecked after
2814 	 * everything is relocked to determine if the source or target file
2815 	 * has been renamed.
2816 	 */
2817 	KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
2818 	fromnd->nl_flags &= ~NLC_NCPISLOCKED;
2819 	cache_unlock(&fromnd->nl_nch);
2820 
2821 	tond->nl_flags |= NLC_CREATE;
2822 	if ((error = nlookup(tond)) != 0) {
2823 		cache_drop(&fnchd);
2824 		return (error);
2825 	}
2826 	if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
2827 		cache_drop(&fnchd);
2828 		return (ENOENT);
2829 	}
2830 	tnchd.mount = tond->nl_nch.mount;
2831 	cache_hold(&tnchd);
2832 
2833 	/*
2834 	 * If the source and target are the same there is nothing to do
2835 	 */
2836 	if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
2837 		cache_drop(&fnchd);
2838 		cache_drop(&tnchd);
2839 		return (0);
2840 	}
2841 
2842 	/*
2843 	 * Mount points cannot be renamed or overwritten
2844 	 */
2845 	if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
2846 	    NCF_ISMOUNTPT
2847 	) {
2848 		cache_drop(&fnchd);
2849 		cache_drop(&tnchd);
2850 		return (EINVAL);
2851 	}
2852 
2853 	/*
2854 	 * relock the source ncp.  NOTE AFTER RELOCKING: the source ncp
2855 	 * may have become invalid while it was unlocked, nc_vp and nc_mount
2856 	 * could be NULL.
2857 	 */
2858 	if (cache_lock_nonblock(&fromnd->nl_nch) == 0) {
2859 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2860 	} else if (fromnd->nl_nch.ncp > tond->nl_nch.ncp) {
2861 		cache_lock(&fromnd->nl_nch);
2862 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2863 	} else {
2864 		cache_unlock(&tond->nl_nch);
2865 		cache_lock(&fromnd->nl_nch);
2866 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
2867 		cache_lock(&tond->nl_nch);
2868 		cache_resolve(&tond->nl_nch, tond->nl_cred);
2869 	}
2870 	fromnd->nl_flags |= NLC_NCPISLOCKED;
2871 
2872 	/*
2873 	 * make sure the parent directories linkages are the same
2874 	 */
2875 	if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
2876 	    tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
2877 		cache_drop(&fnchd);
2878 		cache_drop(&tnchd);
2879 		return (ENOENT);
2880 	}
2881 
2882 	/*
2883 	 * Both the source and target must be within the same filesystem and
2884 	 * in the same filesystem as their parent directories within the
2885 	 * namecache topology.
2886 	 *
2887 	 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
2888 	 */
2889 	mp = fnchd.mount;
2890 	if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
2891 	    mp != tond->nl_nch.mount) {
2892 		cache_drop(&fnchd);
2893 		cache_drop(&tnchd);
2894 		return (EXDEV);
2895 	}
2896 
2897 	/*
2898 	 * Make sure the mount point is writable
2899 	 */
2900 	if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
2901 		cache_drop(&fnchd);
2902 		cache_drop(&tnchd);
2903 		return (error);
2904 	}
2905 
2906 	/*
2907 	 * If the target exists and either the source or target is a directory,
2908 	 * then both must be directories.
2909 	 *
2910 	 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
2911 	 * have become NULL.
2912 	 */
2913 	if (tond->nl_nch.ncp->nc_vp) {
2914 		if (fromnd->nl_nch.ncp->nc_vp == NULL) {
2915 			error = ENOENT;
2916 		} else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
2917 			if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
2918 				error = ENOTDIR;
2919 		} else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
2920 			error = EISDIR;
2921 		}
2922 	}
2923 
2924 	/*
2925 	 * You cannot rename a source into itself or a subdirectory of itself.
2926 	 * We check this by travsersing the target directory upwards looking
2927 	 * for a match against the source.
2928 	 */
2929 	if (error == 0) {
2930 		for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
2931 			if (fromnd->nl_nch.ncp == ncp) {
2932 				error = EINVAL;
2933 				break;
2934 			}
2935 		}
2936 	}
2937 
2938 	cache_drop(&fnchd);
2939 	cache_drop(&tnchd);
2940 
2941 	/*
2942 	 * Even though the namespaces are different, they may still represent
2943 	 * hardlinks to the same file.  The filesystem might have a hard time
2944 	 * with this so we issue a NREMOVE of the source instead of a NRENAME
2945 	 * when we detect the situation.
2946 	 */
2947 	if (error == 0) {
2948 		if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
2949 			error = VOP_NREMOVE(&fromnd->nl_nch, fromnd->nl_cred);
2950 		} else {
2951 			error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
2952 					    tond->nl_cred);
2953 		}
2954 	}
2955 	return (error);
2956 }
2957 
2958 /*
2959  * rename_args(char *from, char *to)
2960  *
2961  * Rename files.  Source and destination must either both be directories,
2962  * or both not be directories.  If target is a directory, it must be empty.
2963  */
2964 int
2965 sys_rename(struct rename_args *uap)
2966 {
2967 	struct nlookupdata fromnd, tond;
2968 	int error;
2969 
2970 	error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
2971 	if (error == 0) {
2972 		error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
2973 		if (error == 0)
2974 			error = kern_rename(&fromnd, &tond);
2975 		nlookup_done(&tond);
2976 	}
2977 	nlookup_done(&fromnd);
2978 	return (error);
2979 }
2980 
2981 int
2982 kern_mkdir(struct nlookupdata *nd, int mode)
2983 {
2984 	struct thread *td = curthread;
2985 	struct proc *p = td->td_proc;
2986 	struct vnode *vp;
2987 	struct vattr vattr;
2988 	int error;
2989 
2990 	bwillwrite();
2991 	nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE;
2992 	if ((error = nlookup(nd)) != 0)
2993 		return (error);
2994 
2995 	if (nd->nl_nch.ncp->nc_vp)
2996 		return (EEXIST);
2997 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2998 		return (error);
2999 
3000 	VATTR_NULL(&vattr);
3001 	vattr.va_type = VDIR;
3002 	vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3003 
3004 	vp = NULL;
3005 	error = VOP_NMKDIR(&nd->nl_nch, &vp, p->p_ucred, &vattr);
3006 	if (error == 0)
3007 		vput(vp);
3008 	return (error);
3009 }
3010 
3011 /*
3012  * mkdir_args(char *path, int mode)
3013  *
3014  * Make a directory file.
3015  */
3016 /* ARGSUSED */
3017 int
3018 sys_mkdir(struct mkdir_args *uap)
3019 {
3020 	struct nlookupdata nd;
3021 	int error;
3022 
3023 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3024 	if (error == 0)
3025 		error = kern_mkdir(&nd, uap->mode);
3026 	nlookup_done(&nd);
3027 	return (error);
3028 }
3029 
3030 int
3031 kern_rmdir(struct nlookupdata *nd)
3032 {
3033 	int error;
3034 
3035 	bwillwrite();
3036 	nd->nl_flags |= NLC_DELETE;
3037 	if ((error = nlookup(nd)) != 0)
3038 		return (error);
3039 
3040 	/*
3041 	 * Do not allow directories representing mount points to be
3042 	 * deleted, even if empty.  Check write perms on mount point
3043 	 * in case the vnode is aliased (aka nullfs).
3044 	 */
3045 	if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3046 		return (EINVAL);
3047 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3048 		return (error);
3049 
3050 	error = VOP_NRMDIR(&nd->nl_nch, nd->nl_cred);
3051 	return (error);
3052 }
3053 
3054 /*
3055  * rmdir_args(char *path)
3056  *
3057  * Remove a directory file.
3058  */
3059 /* ARGSUSED */
3060 int
3061 sys_rmdir(struct rmdir_args *uap)
3062 {
3063 	struct nlookupdata nd;
3064 	int error;
3065 
3066 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3067 	if (error == 0)
3068 		error = kern_rmdir(&nd);
3069 	nlookup_done(&nd);
3070 	return (error);
3071 }
3072 
3073 int
3074 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3075     enum uio_seg direction)
3076 {
3077 	struct thread *td = curthread;
3078 	struct proc *p = td->td_proc;
3079 	struct vnode *vp;
3080 	struct file *fp;
3081 	struct uio auio;
3082 	struct iovec aiov;
3083 	long loff;
3084 	int error, eofflag;
3085 
3086 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3087 		return (error);
3088 	if ((fp->f_flag & FREAD) == 0) {
3089 		error = EBADF;
3090 		goto done;
3091 	}
3092 	vp = (struct vnode *)fp->f_data;
3093 unionread:
3094 	if (vp->v_type != VDIR) {
3095 		error = EINVAL;
3096 		goto done;
3097 	}
3098 	aiov.iov_base = buf;
3099 	aiov.iov_len = count;
3100 	auio.uio_iov = &aiov;
3101 	auio.uio_iovcnt = 1;
3102 	auio.uio_rw = UIO_READ;
3103 	auio.uio_segflg = direction;
3104 	auio.uio_td = td;
3105 	auio.uio_resid = count;
3106 	loff = auio.uio_offset = fp->f_offset;
3107 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3108 	fp->f_offset = auio.uio_offset;
3109 	if (error)
3110 		goto done;
3111 	if (count == auio.uio_resid) {
3112 		if (union_dircheckp) {
3113 			error = union_dircheckp(td, &vp, fp);
3114 			if (error == -1)
3115 				goto unionread;
3116 			if (error)
3117 				goto done;
3118 		}
3119 #if 0
3120 		if ((vp->v_flag & VROOT) &&
3121 		    (vp->v_mount->mnt_flag & MNT_UNION)) {
3122 			struct vnode *tvp = vp;
3123 			vp = vp->v_mount->mnt_vnodecovered;
3124 			vref(vp);
3125 			fp->f_data = vp;
3126 			fp->f_offset = 0;
3127 			vrele(tvp);
3128 			goto unionread;
3129 		}
3130 #endif
3131 	}
3132 	if (basep) {
3133 		*basep = loff;
3134 	}
3135 	*res = count - auio.uio_resid;
3136 done:
3137 	fdrop(fp);
3138 	return (error);
3139 }
3140 
3141 /*
3142  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3143  *
3144  * Read a block of directory entries in a file system independent format.
3145  */
3146 int
3147 sys_getdirentries(struct getdirentries_args *uap)
3148 {
3149 	long base;
3150 	int error;
3151 
3152 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3153 	    &uap->sysmsg_result, UIO_USERSPACE);
3154 
3155 	if (error == 0)
3156 		error = copyout(&base, uap->basep, sizeof(*uap->basep));
3157 	return (error);
3158 }
3159 
3160 /*
3161  * getdents_args(int fd, char *buf, size_t count)
3162  */
3163 int
3164 sys_getdents(struct getdents_args *uap)
3165 {
3166 	int error;
3167 
3168 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3169 	    &uap->sysmsg_result, UIO_USERSPACE);
3170 
3171 	return (error);
3172 }
3173 
3174 /*
3175  * umask(int newmask)
3176  *
3177  * Set the mode mask for creation of filesystem nodes.
3178  *
3179  * MP SAFE
3180  */
3181 int
3182 sys_umask(struct umask_args *uap)
3183 {
3184 	struct thread *td = curthread;
3185 	struct proc *p = td->td_proc;
3186 	struct filedesc *fdp;
3187 
3188 	fdp = p->p_fd;
3189 	uap->sysmsg_result = fdp->fd_cmask;
3190 	fdp->fd_cmask = uap->newmask & ALLPERMS;
3191 	return (0);
3192 }
3193 
3194 /*
3195  * revoke(char *path)
3196  *
3197  * Void all references to file by ripping underlying filesystem
3198  * away from vnode.
3199  */
3200 /* ARGSUSED */
3201 int
3202 sys_revoke(struct revoke_args *uap)
3203 {
3204 	struct nlookupdata nd;
3205 	struct vattr vattr;
3206 	struct vnode *vp;
3207 	struct ucred *cred;
3208 	int error;
3209 
3210 	vp = NULL;
3211 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3212 	if (error == 0)
3213 		error = nlookup(&nd);
3214 	if (error == 0)
3215 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3216 	cred = crhold(nd.nl_cred);
3217 	nlookup_done(&nd);
3218 	if (error == 0) {
3219 		if (vp->v_type != VCHR && vp->v_type != VBLK)
3220 			error = EINVAL;
3221 		if (error == 0)
3222 			error = VOP_GETATTR(vp, &vattr);
3223 		if (error == 0 && cred->cr_uid != vattr.va_uid)
3224 			error = suser_cred(cred, PRISON_ROOT);
3225 		if (error == 0 && count_udev(vp->v_udev) > 0) {
3226 			error = 0;
3227 			vx_lock(vp);
3228 			VOP_REVOKE(vp, REVOKEALL);
3229 			vx_unlock(vp);
3230 		}
3231 		vrele(vp);
3232 	}
3233 	if (cred)
3234 		crfree(cred);
3235 	return (error);
3236 }
3237 
3238 /*
3239  * getfh_args(char *fname, fhandle_t *fhp)
3240  *
3241  * Get (NFS) file handle
3242  */
3243 int
3244 sys_getfh(struct getfh_args *uap)
3245 {
3246 	struct thread *td = curthread;
3247 	struct nlookupdata nd;
3248 	fhandle_t fh;
3249 	struct vnode *vp;
3250 	int error;
3251 
3252 	/*
3253 	 * Must be super user
3254 	 */
3255 	if ((error = suser(td)) != 0)
3256 		return (error);
3257 
3258 	vp = NULL;
3259 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3260 	if (error == 0)
3261 		error = nlookup(&nd);
3262 	if (error == 0)
3263 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3264 	nlookup_done(&nd);
3265 	if (error == 0) {
3266 		bzero(&fh, sizeof(fh));
3267 		fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid;
3268 		error = VFS_VPTOFH(vp, &fh.fh_fid);
3269 		vput(vp);
3270 		if (error == 0)
3271 			error = copyout(&fh, uap->fhp, sizeof(fh));
3272 	}
3273 	return (error);
3274 }
3275 
3276 /*
3277  * fhopen_args(const struct fhandle *u_fhp, int flags)
3278  *
3279  * syscall for the rpc.lockd to use to translate a NFS file handle into
3280  * an open descriptor.
3281  *
3282  * warning: do not remove the suser() call or this becomes one giant
3283  * security hole.
3284  */
3285 int
3286 sys_fhopen(struct fhopen_args *uap)
3287 {
3288 	struct thread *td = curthread;
3289 	struct proc *p = td->td_proc;
3290 	struct mount *mp;
3291 	struct vnode *vp;
3292 	struct fhandle fhp;
3293 	struct vattr vat;
3294 	struct vattr *vap = &vat;
3295 	struct flock lf;
3296 	int fmode, mode, error, type;
3297 	struct file *nfp;
3298 	struct file *fp;
3299 	int indx;
3300 
3301 	/*
3302 	 * Must be super user
3303 	 */
3304 	error = suser(td);
3305 	if (error)
3306 		return (error);
3307 
3308 	fmode = FFLAGS(uap->flags);
3309 	/* why not allow a non-read/write open for our lockd? */
3310 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3311 		return (EINVAL);
3312 	error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3313 	if (error)
3314 		return(error);
3315 	/* find the mount point */
3316 	mp = vfs_getvfs(&fhp.fh_fsid);
3317 	if (mp == NULL)
3318 		return (ESTALE);
3319 	/* now give me my vnode, it gets returned to me locked */
3320 	error = VFS_FHTOVP(mp, &fhp.fh_fid, &vp);
3321 	if (error)
3322 		return (error);
3323  	/*
3324 	 * from now on we have to make sure not
3325 	 * to forget about the vnode
3326 	 * any error that causes an abort must vput(vp)
3327 	 * just set error = err and 'goto bad;'.
3328 	 */
3329 
3330 	/*
3331 	 * from vn_open
3332 	 */
3333 	if (vp->v_type == VLNK) {
3334 		error = EMLINK;
3335 		goto bad;
3336 	}
3337 	if (vp->v_type == VSOCK) {
3338 		error = EOPNOTSUPP;
3339 		goto bad;
3340 	}
3341 	mode = 0;
3342 	if (fmode & (FWRITE | O_TRUNC)) {
3343 		if (vp->v_type == VDIR) {
3344 			error = EISDIR;
3345 			goto bad;
3346 		}
3347 		error = vn_writechk(vp, NULL);
3348 		if (error)
3349 			goto bad;
3350 		mode |= VWRITE;
3351 	}
3352 	if (fmode & FREAD)
3353 		mode |= VREAD;
3354 	if (mode) {
3355 		error = VOP_ACCESS(vp, mode, p->p_ucred);
3356 		if (error)
3357 			goto bad;
3358 	}
3359 	if (fmode & O_TRUNC) {
3360 		vn_unlock(vp);				/* XXX */
3361 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
3362 		VATTR_NULL(vap);
3363 		vap->va_size = 0;
3364 		error = VOP_SETATTR(vp, vap, p->p_ucred);
3365 		if (error)
3366 			goto bad;
3367 	}
3368 
3369 	/*
3370 	 * VOP_OPEN needs the file pointer so it can potentially override
3371 	 * it.
3372 	 *
3373 	 * WARNING! no f_nchandle will be associated when fhopen()ing a
3374 	 * directory.  XXX
3375 	 */
3376 	if ((error = falloc(p, &nfp, &indx)) != 0)
3377 		goto bad;
3378 	fp = nfp;
3379 
3380 	error = VOP_OPEN(vp, fmode, p->p_ucred, fp);
3381 	if (error) {
3382 		/*
3383 		 * setting f_ops this way prevents VOP_CLOSE from being
3384 		 * called or fdrop() releasing the vp from v_data.   Since
3385 		 * the VOP_OPEN failed we don't want to VOP_CLOSE.
3386 		 */
3387 		fp->f_ops = &badfileops;
3388 		fp->f_data = NULL;
3389 		goto bad_drop;
3390 	}
3391 
3392 	/*
3393 	 * The fp is given its own reference, we still have our ref and lock.
3394 	 *
3395 	 * Assert that all regular files must be created with a VM object.
3396 	 */
3397 	if (vp->v_type == VREG && vp->v_object == NULL) {
3398 		kprintf("fhopen: regular file did not have VM object: %p\n", vp);
3399 		goto bad_drop;
3400 	}
3401 
3402 	/*
3403 	 * The open was successful.  Handle any locking requirements.
3404 	 */
3405 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
3406 		lf.l_whence = SEEK_SET;
3407 		lf.l_start = 0;
3408 		lf.l_len = 0;
3409 		if (fmode & O_EXLOCK)
3410 			lf.l_type = F_WRLCK;
3411 		else
3412 			lf.l_type = F_RDLCK;
3413 		if (fmode & FNONBLOCK)
3414 			type = 0;
3415 		else
3416 			type = F_WAIT;
3417 		vn_unlock(vp);
3418 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3419 			/*
3420 			 * release our private reference.
3421 			 */
3422 			fsetfd(p, NULL, indx);
3423 			fdrop(fp);
3424 			vrele(vp);
3425 			return (error);
3426 		}
3427 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3428 		fp->f_flag |= FHASLOCK;
3429 	}
3430 
3431 	/*
3432 	 * Clean up.  Associate the file pointer with the previously
3433 	 * reserved descriptor and return it.
3434 	 */
3435 	vput(vp);
3436 	fsetfd(p, fp, indx);
3437 	fdrop(fp);
3438 	uap->sysmsg_result = indx;
3439 	return (0);
3440 
3441 bad_drop:
3442 	fsetfd(p, NULL, indx);
3443 	fdrop(fp);
3444 bad:
3445 	vput(vp);
3446 	return (error);
3447 }
3448 
3449 /*
3450  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3451  */
3452 int
3453 sys_fhstat(struct fhstat_args *uap)
3454 {
3455 	struct thread *td = curthread;
3456 	struct stat sb;
3457 	fhandle_t fh;
3458 	struct mount *mp;
3459 	struct vnode *vp;
3460 	int error;
3461 
3462 	/*
3463 	 * Must be super user
3464 	 */
3465 	error = suser(td);
3466 	if (error)
3467 		return (error);
3468 
3469 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
3470 	if (error)
3471 		return (error);
3472 
3473 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3474 		return (ESTALE);
3475 	if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3476 		return (error);
3477 	error = vn_stat(vp, &sb, td->td_proc->p_ucred);
3478 	vput(vp);
3479 	if (error)
3480 		return (error);
3481 	error = copyout(&sb, uap->sb, sizeof(sb));
3482 	return (error);
3483 }
3484 
3485 /*
3486  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
3487  */
3488 int
3489 sys_fhstatfs(struct fhstatfs_args *uap)
3490 {
3491 	struct thread *td = curthread;
3492 	struct proc *p = td->td_proc;
3493 	struct statfs *sp;
3494 	struct mount *mp;
3495 	struct vnode *vp;
3496 	struct statfs sb;
3497 	char *fullpath, *freepath;
3498 	fhandle_t fh;
3499 	int error;
3500 
3501 	/*
3502 	 * Must be super user
3503 	 */
3504 	if ((error = suser(td)))
3505 		return (error);
3506 
3507 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3508 		return (error);
3509 
3510 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3511 		return (ESTALE);
3512 
3513 	if (p != NULL && !chroot_visible_mnt(mp, p))
3514 		return (ESTALE);
3515 
3516 	if ((error = VFS_FHTOVP(mp, &fh.fh_fid, &vp)))
3517 		return (error);
3518 	mp = vp->v_mount;
3519 	sp = &mp->mnt_stat;
3520 	vput(vp);
3521 	if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0)
3522 		return (error);
3523 
3524 	error = cache_fullpath(p, &mp->mnt_ncmountpt, &fullpath, &freepath);
3525 	if (error)
3526 		return(error);
3527 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3528 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
3529 	kfree(freepath, M_TEMP);
3530 
3531 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
3532 	if (suser(td)) {
3533 		bcopy(sp, &sb, sizeof(sb));
3534 		sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
3535 		sp = &sb;
3536 	}
3537 	return (copyout(sp, uap->buf, sizeof(*sp)));
3538 }
3539 
3540 /*
3541  * Syscall to push extended attribute configuration information into the
3542  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
3543  * a command (int cmd), and attribute name and misc data.  For now, the
3544  * attribute name is left in userspace for consumption by the VFS_op.
3545  * It will probably be changed to be copied into sysspace by the
3546  * syscall in the future, once issues with various consumers of the
3547  * attribute code have raised their hands.
3548  *
3549  * Currently this is used only by UFS Extended Attributes.
3550  */
3551 int
3552 sys_extattrctl(struct extattrctl_args *uap)
3553 {
3554 	struct nlookupdata nd;
3555 	struct mount *mp;
3556 	struct vnode *vp;
3557 	int error;
3558 
3559 	vp = NULL;
3560 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3561 	if (error == 0)
3562 		error = nlookup(&nd);
3563 	if (error == 0) {
3564 		mp = nd.nl_nch.mount;
3565 		error = VFS_EXTATTRCTL(mp, uap->cmd,
3566 				uap->attrname, uap->arg,
3567 				nd.nl_cred);
3568 	}
3569 	nlookup_done(&nd);
3570 	return (error);
3571 }
3572 
3573 /*
3574  * Syscall to set a named extended attribute on a file or directory.
3575  * Accepts attribute name, and a uio structure pointing to the data to set.
3576  * The uio is consumed in the style of writev().  The real work happens
3577  * in VOP_SETEXTATTR().
3578  */
3579 int
3580 sys_extattr_set_file(struct extattr_set_file_args *uap)
3581 {
3582 	char attrname[EXTATTR_MAXNAMELEN];
3583 	struct iovec aiov[UIO_SMALLIOV];
3584 	struct iovec *needfree;
3585 	struct nlookupdata nd;
3586 	struct iovec *iov;
3587 	struct vnode *vp;
3588 	struct uio auio;
3589 	u_int iovlen;
3590 	u_int cnt;
3591 	int error;
3592 	int i;
3593 
3594 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3595 	if (error)
3596 		return (error);
3597 
3598 	vp = NULL;
3599 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3600 	if (error == 0)
3601 		error = nlookup(&nd);
3602 	if (error == 0)
3603 		error = ncp_writechk(&nd.nl_nch);
3604 	if (error == 0)
3605 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3606 	if (error) {
3607 		nlookup_done(&nd);
3608 		return (error);
3609 	}
3610 
3611 	needfree = NULL;
3612 	iovlen = uap->iovcnt * sizeof(struct iovec);
3613 	if (uap->iovcnt > UIO_SMALLIOV) {
3614 		if (uap->iovcnt > UIO_MAXIOV) {
3615 			error = EINVAL;
3616 			goto done;
3617 		}
3618 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3619 		needfree = iov;
3620 	} else {
3621 		iov = aiov;
3622 	}
3623 	auio.uio_iov = iov;
3624 	auio.uio_iovcnt = uap->iovcnt;
3625 	auio.uio_rw = UIO_WRITE;
3626 	auio.uio_segflg = UIO_USERSPACE;
3627 	auio.uio_td = nd.nl_td;
3628 	auio.uio_offset = 0;
3629 	if ((error = copyin(uap->iovp, iov, iovlen)))
3630 		goto done;
3631 	auio.uio_resid = 0;
3632 	for (i = 0; i < uap->iovcnt; i++) {
3633 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
3634 			error = EINVAL;
3635 			goto done;
3636 		}
3637 		auio.uio_resid += iov->iov_len;
3638 		iov++;
3639 	}
3640 	cnt = auio.uio_resid;
3641 	error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3642 	cnt -= auio.uio_resid;
3643 	uap->sysmsg_result = cnt;
3644 done:
3645 	vput(vp);
3646 	nlookup_done(&nd);
3647 	if (needfree)
3648 		FREE(needfree, M_IOV);
3649 	return (error);
3650 }
3651 
3652 /*
3653  * Syscall to get a named extended attribute on a file or directory.
3654  * Accepts attribute name, and a uio structure pointing to a buffer for the
3655  * data.  The uio is consumed in the style of readv().  The real work
3656  * happens in VOP_GETEXTATTR();
3657  */
3658 int
3659 sys_extattr_get_file(struct extattr_get_file_args *uap)
3660 {
3661 	char attrname[EXTATTR_MAXNAMELEN];
3662 	struct iovec aiov[UIO_SMALLIOV];
3663 	struct iovec *needfree;
3664 	struct nlookupdata nd;
3665 	struct iovec *iov;
3666 	struct vnode *vp;
3667 	struct uio auio;
3668 	u_int iovlen;
3669 	u_int cnt;
3670 	int error;
3671 	int i;
3672 
3673 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3674 	if (error)
3675 		return (error);
3676 
3677 	vp = NULL;
3678 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3679 	if (error == 0)
3680 		error = nlookup(&nd);
3681 	if (error == 0)
3682 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3683 	if (error) {
3684 		nlookup_done(&nd);
3685 		return (error);
3686 	}
3687 
3688 	iovlen = uap->iovcnt * sizeof (struct iovec);
3689 	needfree = NULL;
3690 	if (uap->iovcnt > UIO_SMALLIOV) {
3691 		if (uap->iovcnt > UIO_MAXIOV) {
3692 			error = EINVAL;
3693 			goto done;
3694 		}
3695 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
3696 		needfree = iov;
3697 	} else {
3698 		iov = aiov;
3699 	}
3700 	auio.uio_iov = iov;
3701 	auio.uio_iovcnt = uap->iovcnt;
3702 	auio.uio_rw = UIO_READ;
3703 	auio.uio_segflg = UIO_USERSPACE;
3704 	auio.uio_td = nd.nl_td;
3705 	auio.uio_offset = 0;
3706 	if ((error = copyin(uap->iovp, iov, iovlen)))
3707 		goto done;
3708 	auio.uio_resid = 0;
3709 	for (i = 0; i < uap->iovcnt; i++) {
3710 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
3711 			error = EINVAL;
3712 			goto done;
3713 		}
3714 		auio.uio_resid += iov->iov_len;
3715 		iov++;
3716 	}
3717 	cnt = auio.uio_resid;
3718 	error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred);
3719 	cnt -= auio.uio_resid;
3720 	uap->sysmsg_result = cnt;
3721 done:
3722 	vput(vp);
3723 	nlookup_done(&nd);
3724 	if (needfree)
3725 		FREE(needfree, M_IOV);
3726 	return(error);
3727 }
3728 
3729 /*
3730  * Syscall to delete a named extended attribute from a file or directory.
3731  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
3732  */
3733 int
3734 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
3735 {
3736 	char attrname[EXTATTR_MAXNAMELEN];
3737 	struct nlookupdata nd;
3738 	struct vnode *vp;
3739 	int error;
3740 
3741 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3742 	if (error)
3743 		return(error);
3744 
3745 	vp = NULL;
3746 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3747 	if (error == 0)
3748 		error = nlookup(&nd);
3749 	if (error == 0)
3750 		error = ncp_writechk(&nd.nl_nch);
3751 	if (error == 0)
3752 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3753 	if (error) {
3754 		nlookup_done(&nd);
3755 		return (error);
3756 	}
3757 
3758 	error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred);
3759 	vput(vp);
3760 	nlookup_done(&nd);
3761 	return(error);
3762 }
3763 
3764 /*
3765  * Determine if the mount is visible to the process.
3766  */
3767 static int
3768 chroot_visible_mnt(struct mount *mp, struct proc *p)
3769 {
3770 	struct nchandle nch;
3771 
3772 	/*
3773 	 * Traverse from the mount point upwards.  If we hit the process
3774 	 * root then the mount point is visible to the process.
3775 	 */
3776 	nch = mp->mnt_ncmountpt;
3777 	while (nch.ncp) {
3778 		if (nch.mount == p->p_fd->fd_nrdir.mount &&
3779 		    nch.ncp == p->p_fd->fd_nrdir.ncp) {
3780 			return(1);
3781 		}
3782 		if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
3783 			nch = nch.mount->mnt_ncmounton;
3784 		} else {
3785 			nch.ncp = nch.ncp->nc_parent;
3786 		}
3787 	}
3788 	return(0);
3789 }
3790 
3791