xref: /dflybsd-src/sys/kern/vfs_syscalls.c (revision 3efc72a71e16af9b5466f187f9e70e52c026d7ae)
1 /*
2  * Copyright (c) 1989, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  * (c) UNIX System Laboratories, Inc.
5  * All or some portions of this file are derived from material licensed
6  * to the University of California by American Telephone and Telegraph
7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8  * the permission of UNIX System Laboratories, Inc.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)vfs_syscalls.c	8.13 (Berkeley) 4/15/94
39  * $FreeBSD: src/sys/kern/vfs_syscalls.c,v 1.151.2.18 2003/04/04 20:35:58 tegge Exp $
40  * $DragonFly: src/sys/kern/vfs_syscalls.c,v 1.135 2008/11/11 00:55:49 pavalos Exp $
41  */
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/buf.h>
46 #include <sys/conf.h>
47 #include <sys/sysent.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mountctl.h>
51 #include <sys/sysproto.h>
52 #include <sys/filedesc.h>
53 #include <sys/kernel.h>
54 #include <sys/fcntl.h>
55 #include <sys/file.h>
56 #include <sys/linker.h>
57 #include <sys/stat.h>
58 #include <sys/unistd.h>
59 #include <sys/vnode.h>
60 #include <sys/proc.h>
61 #include <sys/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 
70 #include <sys/buf2.h>
71 #include <sys/file2.h>
72 #include <sys/spinlock2.h>
73 
74 #include <vm/vm.h>
75 #include <vm/vm_object.h>
76 #include <vm/vm_page.h>
77 
78 #include <machine/limits.h>
79 #include <machine/stdarg.h>
80 
81 #include <vfs/union/union.h>
82 
83 static void mount_warning(struct mount *mp, const char *ctl, ...);
84 static int mount_path(struct proc *p, struct mount *mp, char **rb, char **fb);
85 static int checkvp_chdir (struct vnode *vn, struct thread *td);
86 static void checkdirs (struct nchandle *old_nch, struct nchandle *new_nch);
87 static int chroot_refuse_vdir_fds (struct filedesc *fdp);
88 static int chroot_visible_mnt(struct mount *mp, struct proc *p);
89 static int getutimes (const struct timeval *, struct timespec *);
90 static int setfown (struct vnode *, uid_t, gid_t);
91 static int setfmode (struct vnode *, int);
92 static int setfflags (struct vnode *, int);
93 static int setutimes (struct vnode *, const struct timespec *, int);
94 static int	usermount = 0;	/* if 1, non-root can mount fs. */
95 
96 int (*union_dircheckp) (struct thread *, struct vnode **, struct file *);
97 
98 SYSCTL_INT(_vfs, OID_AUTO, usermount, CTLFLAG_RW, &usermount, 0, "");
99 
100 /*
101  * Virtual File System System Calls
102  */
103 
104 /*
105  * Mount a file system.
106  */
107 /*
108  * mount_args(char *type, char *path, int flags, caddr_t data)
109  */
110 /* ARGSUSED */
111 int
112 sys_mount(struct mount_args *uap)
113 {
114 	struct thread *td = curthread;
115 	struct proc *p = td->td_proc;
116 	struct vnode *vp;
117 	struct nchandle nch;
118 	struct mount *mp;
119 	struct vfsconf *vfsp;
120 	int error, flag = 0, flag2 = 0;
121 	int hasmount;
122 	struct vattr va;
123 	struct nlookupdata nd;
124 	char fstypename[MFSNAMELEN];
125 	struct ucred *cred = p->p_ucred;
126 
127 	KKASSERT(p);
128 	if (cred->cr_prison != NULL)
129 		return (EPERM);
130 	if (usermount == 0 && (error = suser(td)))
131 		return (error);
132 	/*
133 	 * Do not allow NFS export by non-root users.
134 	 */
135 	if (uap->flags & MNT_EXPORTED) {
136 		error = suser(td);
137 		if (error)
138 			return (error);
139 	}
140 	/*
141 	 * Silently enforce MNT_NOSUID and MNT_NODEV for non-root users
142 	 */
143 	if (suser(td))
144 		uap->flags |= MNT_NOSUID | MNT_NODEV;
145 
146 	/*
147 	 * Lookup the requested path and extract the nch and vnode.
148 	 */
149 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
150 	if (error == 0) {
151 		if ((error = nlookup(&nd)) == 0) {
152 			if (nd.nl_nch.ncp->nc_vp == NULL)
153 				error = ENOENT;
154 		}
155 	}
156 	if (error) {
157 		nlookup_done(&nd);
158 		return (error);
159 	}
160 
161 	/*
162 	 * Extract the locked+refd ncp and cleanup the nd structure
163 	 */
164 	nch = nd.nl_nch;
165 	cache_zero(&nd.nl_nch);
166 	nlookup_done(&nd);
167 
168 	if ((nch.ncp->nc_flag & NCF_ISMOUNTPT) && cache_findmount(&nch))
169 		hasmount = 1;
170 	else
171 		hasmount = 0;
172 
173 
174 	/*
175 	 * now we have the locked ref'd nch and unreferenced vnode.
176 	 */
177 	vp = nch.ncp->nc_vp;
178 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0) {
179 		cache_put(&nch);
180 		return (error);
181 	}
182 	cache_unlock(&nch);
183 
184 	/*
185 	 * Now we have an unlocked ref'd nch and a locked ref'd vp
186 	 */
187 	if (uap->flags & MNT_UPDATE) {
188 		if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
189 			cache_drop(&nch);
190 			vput(vp);
191 			return (EINVAL);
192 		}
193 		mp = vp->v_mount;
194 		flag = mp->mnt_flag;
195 		flag2 = mp->mnt_kern_flag;
196 		/*
197 		 * We only allow the filesystem to be reloaded if it
198 		 * is currently mounted read-only.
199 		 */
200 		if ((uap->flags & MNT_RELOAD) &&
201 		    ((mp->mnt_flag & MNT_RDONLY) == 0)) {
202 			cache_drop(&nch);
203 			vput(vp);
204 			return (EOPNOTSUPP);	/* Needs translation */
205 		}
206 		/*
207 		 * Only root, or the user that did the original mount is
208 		 * permitted to update it.
209 		 */
210 		if (mp->mnt_stat.f_owner != cred->cr_uid &&
211 		    (error = suser(td))) {
212 			cache_drop(&nch);
213 			vput(vp);
214 			return (error);
215 		}
216 		if (vfs_busy(mp, LK_NOWAIT)) {
217 			cache_drop(&nch);
218 			vput(vp);
219 			return (EBUSY);
220 		}
221 		if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
222 			cache_drop(&nch);
223 			vfs_unbusy(mp);
224 			vput(vp);
225 			return (EBUSY);
226 		}
227 		vp->v_flag |= VMOUNT;
228 		mp->mnt_flag |=
229 		    uap->flags & (MNT_RELOAD | MNT_FORCE | MNT_UPDATE);
230 		vn_unlock(vp);
231 		goto update;
232 	}
233 	/*
234 	 * If the user is not root, ensure that they own the directory
235 	 * onto which we are attempting to mount.
236 	 */
237 	if ((error = VOP_GETATTR(vp, &va)) ||
238 	    (va.va_uid != cred->cr_uid && (error = suser(td)))) {
239 		cache_drop(&nch);
240 		vput(vp);
241 		return (error);
242 	}
243 	if ((error = vinvalbuf(vp, V_SAVE, 0, 0)) != 0) {
244 		cache_drop(&nch);
245 		vput(vp);
246 		return (error);
247 	}
248 	if (vp->v_type != VDIR) {
249 		cache_drop(&nch);
250 		vput(vp);
251 		return (ENOTDIR);
252 	}
253 	if (vp->v_mount->mnt_kern_flag & MNTK_NOSTKMNT) {
254 		cache_drop(&nch);
255 		vput(vp);
256 		return (EPERM);
257 	}
258 	if ((error = copyinstr(uap->type, fstypename, MFSNAMELEN, NULL)) != 0) {
259 		cache_drop(&nch);
260 		vput(vp);
261 		return (error);
262 	}
263 	for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
264 		if (!strcmp(vfsp->vfc_name, fstypename))
265 			break;
266 	}
267 	if (vfsp == NULL) {
268 		linker_file_t lf;
269 
270 		/* Only load modules for root (very important!) */
271 		if ((error = suser(td)) != 0) {
272 			cache_drop(&nch);
273 			vput(vp);
274 			return error;
275 		}
276 		error = linker_load_file(fstypename, &lf);
277 		if (error || lf == NULL) {
278 			cache_drop(&nch);
279 			vput(vp);
280 			if (lf == NULL)
281 				error = ENODEV;
282 			return error;
283 		}
284 		lf->userrefs++;
285 		/* lookup again, see if the VFS was loaded */
286 		for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
287 			if (!strcmp(vfsp->vfc_name, fstypename))
288 				break;
289 		}
290 		if (vfsp == NULL) {
291 			lf->userrefs--;
292 			linker_file_unload(lf);
293 			cache_drop(&nch);
294 			vput(vp);
295 			return (ENODEV);
296 		}
297 	}
298 	if ((vp->v_flag & VMOUNT) != 0 || hasmount) {
299 		cache_drop(&nch);
300 		vput(vp);
301 		return (EBUSY);
302 	}
303 	vp->v_flag |= VMOUNT;
304 
305 	/*
306 	 * Allocate and initialize the filesystem.
307 	 */
308 	mp = kmalloc(sizeof(struct mount), M_MOUNT, M_ZERO|M_WAITOK);
309 	TAILQ_INIT(&mp->mnt_nvnodelist);
310 	TAILQ_INIT(&mp->mnt_reservedvnlist);
311 	TAILQ_INIT(&mp->mnt_jlist);
312 	mp->mnt_nvnodelistsize = 0;
313 	lockinit(&mp->mnt_lock, "vfslock", 0, 0);
314 	vfs_busy(mp, LK_NOWAIT);
315 	mp->mnt_op = vfsp->vfc_vfsops;
316 	mp->mnt_vfc = vfsp;
317 	vfsp->vfc_refcount++;
318 	mp->mnt_stat.f_type = vfsp->vfc_typenum;
319 	mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
320 	strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
321 	mp->mnt_stat.f_owner = cred->cr_uid;
322 	mp->mnt_iosize_max = DFLTPHYS;
323 	vn_unlock(vp);
324 update:
325 	/*
326 	 * Set the mount level flags.
327 	 */
328 	if (uap->flags & MNT_RDONLY)
329 		mp->mnt_flag |= MNT_RDONLY;
330 	else if (mp->mnt_flag & MNT_RDONLY)
331 		mp->mnt_kern_flag |= MNTK_WANTRDWR;
332 	mp->mnt_flag &=~ (MNT_NOSUID | MNT_NOEXEC | MNT_NODEV |
333 	    MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_NOATIME |
334 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
335 	    MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
336 	mp->mnt_flag |= uap->flags & (MNT_NOSUID | MNT_NOEXEC |
337 	    MNT_NODEV | MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | MNT_FORCE |
338 	    MNT_NOSYMFOLLOW | MNT_IGNORE |
339 	    MNT_NOATIME | MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR);
340 	/*
341 	 * Mount the filesystem.
342 	 * XXX The final recipients of VFS_MOUNT just overwrite the ndp they
343 	 * get.
344 	 */
345 	error = VFS_MOUNT(mp, uap->path, uap->data, cred);
346 	if (mp->mnt_flag & MNT_UPDATE) {
347 		if (mp->mnt_kern_flag & MNTK_WANTRDWR)
348 			mp->mnt_flag &= ~MNT_RDONLY;
349 		mp->mnt_flag &=~ (MNT_UPDATE | MNT_RELOAD | MNT_FORCE);
350 		mp->mnt_kern_flag &=~ MNTK_WANTRDWR;
351 		if (error) {
352 			mp->mnt_flag = flag;
353 			mp->mnt_kern_flag = flag2;
354 		}
355 		vfs_unbusy(mp);
356 		vp->v_flag &= ~VMOUNT;
357 		vrele(vp);
358 		cache_drop(&nch);
359 		return (error);
360 	}
361 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
362 	/*
363 	 * Put the new filesystem on the mount list after root.  The mount
364 	 * point gets its own mnt_ncmountpt (unless the VFS already set one
365 	 * up) which represents the root of the mount.  The lookup code
366 	 * detects the mount point going forward and checks the root of
367 	 * the mount going backwards.
368 	 *
369 	 * It is not necessary to invalidate or purge the vnode underneath
370 	 * because elements under the mount will be given their own glue
371 	 * namecache record.
372 	 */
373 	if (!error) {
374 		if (mp->mnt_ncmountpt.ncp == NULL) {
375 			/*
376 			 * allocate, then unlock, but leave the ref intact
377 			 */
378 			cache_allocroot(&mp->mnt_ncmountpt, mp, NULL);
379 			cache_unlock(&mp->mnt_ncmountpt);
380 		}
381 		mp->mnt_ncmounton = nch;		/* inherits ref */
382 		nch.ncp->nc_flag |= NCF_ISMOUNTPT;
383 
384 		/* XXX get the root of the fs and cache_setvp(mnt_ncmountpt...) */
385 		vp->v_flag &= ~VMOUNT;
386 		mountlist_insert(mp, MNTINS_LAST);
387 		checkdirs(&mp->mnt_ncmounton, &mp->mnt_ncmountpt);
388 		vn_unlock(vp);
389 		error = vfs_allocate_syncvnode(mp);
390 		vfs_unbusy(mp);
391 		error = VFS_START(mp, 0);
392 		vrele(vp);
393 	} else {
394 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
395 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
396 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
397 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
398 		vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
399 		vp->v_flag &= ~VMOUNT;
400 		mp->mnt_vfc->vfc_refcount--;
401 		vfs_unbusy(mp);
402 		kfree(mp, M_MOUNT);
403 		cache_drop(&nch);
404 		vput(vp);
405 	}
406 	return (error);
407 }
408 
409 /*
410  * Scan all active processes to see if any of them have a current
411  * or root directory onto which the new filesystem has just been
412  * mounted. If so, replace them with the new mount point.
413  *
414  * The passed ncp is ref'd and locked (from the mount code) and
415  * must be associated with the vnode representing the root of the
416  * mount point.
417  */
418 struct checkdirs_info {
419 	struct nchandle old_nch;
420 	struct nchandle new_nch;
421 	struct vnode *old_vp;
422 	struct vnode *new_vp;
423 };
424 
425 static int checkdirs_callback(struct proc *p, void *data);
426 
427 static void
428 checkdirs(struct nchandle *old_nch, struct nchandle *new_nch)
429 {
430 	struct checkdirs_info info;
431 	struct vnode *olddp;
432 	struct vnode *newdp;
433 	struct mount *mp;
434 
435 	/*
436 	 * If the old mount point's vnode has a usecount of 1, it is not
437 	 * being held as a descriptor anywhere.
438 	 */
439 	olddp = old_nch->ncp->nc_vp;
440 	if (olddp == NULL || olddp->v_sysref.refcnt == 1)
441 		return;
442 
443 	/*
444 	 * Force the root vnode of the new mount point to be resolved
445 	 * so we can update any matching processes.
446 	 */
447 	mp = new_nch->mount;
448 	if (VFS_ROOT(mp, &newdp))
449 		panic("mount: lost mount");
450 	cache_setunresolved(new_nch);
451 	cache_setvp(new_nch, newdp);
452 
453 	/*
454 	 * Special handling of the root node
455 	 */
456 	if (rootvnode == olddp) {
457 		vref(newdp);
458 		vfs_cache_setroot(newdp, cache_hold(new_nch));
459 	}
460 
461 	/*
462 	 * Pass newdp separately so the callback does not have to access
463 	 * it via new_nch->ncp->nc_vp.
464 	 */
465 	info.old_nch = *old_nch;
466 	info.new_nch = *new_nch;
467 	info.new_vp = newdp;
468 	allproc_scan(checkdirs_callback, &info);
469 	vput(newdp);
470 }
471 
472 /*
473  * NOTE: callback is not MP safe because the scanned process's filedesc
474  * structure can be ripped out from under us, amoung other things.
475  */
476 static int
477 checkdirs_callback(struct proc *p, void *data)
478 {
479 	struct checkdirs_info *info = data;
480 	struct filedesc *fdp;
481 	struct nchandle ncdrop1;
482 	struct nchandle ncdrop2;
483 	struct vnode *vprele1;
484 	struct vnode *vprele2;
485 
486 	if ((fdp = p->p_fd) != NULL) {
487 		cache_zero(&ncdrop1);
488 		cache_zero(&ncdrop2);
489 		vprele1 = NULL;
490 		vprele2 = NULL;
491 
492 		/*
493 		 * MPUNSAFE - XXX fdp can be pulled out from under a
494 		 * foreign process.
495 		 *
496 		 * A shared filedesc is ok, we don't have to copy it
497 		 * because we are making this change globally.
498 		 */
499 		spin_lock_wr(&fdp->fd_spin);
500 		if (fdp->fd_ncdir.mount == info->old_nch.mount &&
501 		    fdp->fd_ncdir.ncp == info->old_nch.ncp) {
502 			vprele1 = fdp->fd_cdir;
503 			vref(info->new_vp);
504 			fdp->fd_cdir = info->new_vp;
505 			ncdrop1 = fdp->fd_ncdir;
506 			cache_copy(&info->new_nch, &fdp->fd_ncdir);
507 		}
508 		if (fdp->fd_nrdir.mount == info->old_nch.mount &&
509 		    fdp->fd_nrdir.ncp == info->old_nch.ncp) {
510 			vprele2 = fdp->fd_rdir;
511 			vref(info->new_vp);
512 			fdp->fd_rdir = info->new_vp;
513 			ncdrop2 = fdp->fd_nrdir;
514 			cache_copy(&info->new_nch, &fdp->fd_nrdir);
515 		}
516 		spin_unlock_wr(&fdp->fd_spin);
517 		if (ncdrop1.ncp)
518 			cache_drop(&ncdrop1);
519 		if (ncdrop2.ncp)
520 			cache_drop(&ncdrop2);
521 		if (vprele1)
522 			vrele(vprele1);
523 		if (vprele2)
524 			vrele(vprele2);
525 	}
526 	return(0);
527 }
528 
529 /*
530  * Unmount a file system.
531  *
532  * Note: unmount takes a path to the vnode mounted on as argument,
533  * not special file (as before).
534  */
535 /*
536  * umount_args(char *path, int flags)
537  */
538 /* ARGSUSED */
539 int
540 sys_unmount(struct unmount_args *uap)
541 {
542 	struct thread *td = curthread;
543 	struct proc *p = td->td_proc;
544 	struct mount *mp = NULL;
545 	int error;
546 	struct nlookupdata nd;
547 
548 	KKASSERT(p);
549 	if (p->p_ucred->cr_prison != NULL)
550 		return (EPERM);
551 	if (usermount == 0 && (error = suser(td)))
552 		return (error);
553 
554 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
555 	if (error == 0)
556 		error = nlookup(&nd);
557 	if (error)
558 		goto out;
559 
560 	mp = nd.nl_nch.mount;
561 
562 	/*
563 	 * Only root, or the user that did the original mount is
564 	 * permitted to unmount this filesystem.
565 	 */
566 	if ((mp->mnt_stat.f_owner != p->p_ucred->cr_uid) &&
567 	    (error = suser(td)))
568 		goto out;
569 
570 	/*
571 	 * Don't allow unmounting the root file system.
572 	 */
573 	if (mp->mnt_flag & MNT_ROOTFS) {
574 		error = EINVAL;
575 		goto out;
576 	}
577 
578 	/*
579 	 * Must be the root of the filesystem
580 	 */
581 	if (nd.nl_nch.ncp != mp->mnt_ncmountpt.ncp) {
582 		error = EINVAL;
583 		goto out;
584 	}
585 
586 out:
587 	nlookup_done(&nd);
588 	if (error)
589 		return (error);
590 	return (dounmount(mp, uap->flags));
591 }
592 
593 /*
594  * Do the actual file system unmount.
595  */
596 static int
597 dounmount_interlock(struct mount *mp)
598 {
599 	if (mp->mnt_kern_flag & MNTK_UNMOUNT)
600 		return (EBUSY);
601 	mp->mnt_kern_flag |= MNTK_UNMOUNT;
602 	return(0);
603 }
604 
605 int
606 dounmount(struct mount *mp, int flags)
607 {
608 	struct namecache *ncp;
609 	struct nchandle nch;
610 	struct vnode *vp;
611 	int error;
612 	int async_flag;
613 	int lflags;
614 	int freeok = 1;
615 
616 	/*
617 	 * Exclusive access for unmounting purposes
618 	 */
619 	if ((error = mountlist_interlock(dounmount_interlock, mp)) != 0)
620 		return (error);
621 
622 	/*
623 	 * Allow filesystems to detect that a forced unmount is in progress.
624 	 */
625 	if (flags & MNT_FORCE)
626 		mp->mnt_kern_flag |= MNTK_UNMOUNTF;
627 	lflags = LK_EXCLUSIVE | ((flags & MNT_FORCE) ? 0 : LK_NOWAIT);
628 	error = lockmgr(&mp->mnt_lock, lflags);
629 	if (error) {
630 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
631 		if (mp->mnt_kern_flag & MNTK_MWAIT)
632 			wakeup(mp);
633 		return (error);
634 	}
635 
636 	if (mp->mnt_flag & MNT_EXPUBLIC)
637 		vfs_setpublicfs(NULL, NULL, NULL);
638 
639 	vfs_msync(mp, MNT_WAIT);
640 	async_flag = mp->mnt_flag & MNT_ASYNC;
641 	mp->mnt_flag &=~ MNT_ASYNC;
642 
643 	/*
644 	 * If this filesystem isn't aliasing other filesystems,
645 	 * try to invalidate any remaining namecache entries and
646 	 * check the count afterwords.
647 	 */
648 	if ((mp->mnt_kern_flag & MNTK_NCALIASED) == 0) {
649 		cache_lock(&mp->mnt_ncmountpt);
650 		cache_inval(&mp->mnt_ncmountpt, CINV_DESTROY|CINV_CHILDREN);
651 		cache_unlock(&mp->mnt_ncmountpt);
652 
653 		if ((ncp = mp->mnt_ncmountpt.ncp) != NULL &&
654 		    (ncp->nc_refs != 1 || TAILQ_FIRST(&ncp->nc_list))) {
655 
656 			if ((flags & MNT_FORCE) == 0) {
657 				error = EBUSY;
658 				mount_warning(mp, "Cannot unmount: "
659 						  "%d namecache "
660 						  "references still "
661 						  "present",
662 						  ncp->nc_refs - 1);
663 			} else {
664 				mount_warning(mp, "Forced unmount: "
665 						  "%d namecache "
666 						  "references still "
667 						  "present",
668 						  ncp->nc_refs - 1);
669 				freeok = 0;
670 			}
671 		}
672 	}
673 
674 	/*
675 	 * nchandle records ref the mount structure.  Expect a count of 1
676 	 * (our mount->mnt_ncmountpt).
677 	 */
678 	if (mp->mnt_refs != 1) {
679 		if ((flags & MNT_FORCE) == 0) {
680 			mount_warning(mp, "Cannot unmount: "
681 					  "%d process references still "
682 					  "present", mp->mnt_refs);
683 			error = EBUSY;
684 		} else {
685 			mount_warning(mp, "Forced unmount: "
686 					  "%d process references still "
687 					  "present", mp->mnt_refs);
688 			freeok = 0;
689 		}
690 	}
691 
692 	/*
693 	 * Decomission our special mnt_syncer vnode.  This also stops
694 	 * the vnlru code.  If we are unable to unmount we recommission
695 	 * the vnode.
696 	 */
697 	if (error == 0) {
698 		if ((vp = mp->mnt_syncer) != NULL) {
699 			mp->mnt_syncer = NULL;
700 			vrele(vp);
701 		}
702 		if (((mp->mnt_flag & MNT_RDONLY) ||
703 		     (error = VFS_SYNC(mp, MNT_WAIT)) == 0) ||
704 		    (flags & MNT_FORCE)) {
705 			error = VFS_UNMOUNT(mp, flags);
706 		}
707 	}
708 	if (error) {
709 		if (mp->mnt_syncer == NULL)
710 			vfs_allocate_syncvnode(mp);
711 		mp->mnt_kern_flag &= ~(MNTK_UNMOUNT | MNTK_UNMOUNTF);
712 		mp->mnt_flag |= async_flag;
713 		lockmgr(&mp->mnt_lock, LK_RELEASE);
714 		if (mp->mnt_kern_flag & MNTK_MWAIT)
715 			wakeup(mp);
716 		return (error);
717 	}
718 	/*
719 	 * Clean up any journals still associated with the mount after
720 	 * filesystem activity has ceased.
721 	 */
722 	journal_remove_all_journals(mp,
723 	    ((flags & MNT_FORCE) ? MC_JOURNAL_STOP_IMM : 0));
724 
725 	mountlist_remove(mp);
726 
727 	/*
728 	 * Remove any installed vnode ops here so the individual VFSs don't
729 	 * have to.
730 	 */
731 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_coherency_ops);
732 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_journal_ops);
733 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_norm_ops);
734 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_spec_ops);
735 	vfs_rm_vnodeops(mp, NULL, &mp->mnt_vn_fifo_ops);
736 
737 	if (mp->mnt_ncmountpt.ncp != NULL) {
738 		nch = mp->mnt_ncmountpt;
739 		cache_zero(&mp->mnt_ncmountpt);
740 		cache_clrmountpt(&nch);
741 		cache_drop(&nch);
742 	}
743 	if (mp->mnt_ncmounton.ncp != NULL) {
744 		nch = mp->mnt_ncmounton;
745 		cache_zero(&mp->mnt_ncmounton);
746 		cache_clrmountpt(&nch);
747 		cache_drop(&nch);
748 	}
749 
750 	mp->mnt_vfc->vfc_refcount--;
751 	if (!TAILQ_EMPTY(&mp->mnt_nvnodelist))
752 		panic("unmount: dangling vnode");
753 	lockmgr(&mp->mnt_lock, LK_RELEASE);
754 	if (mp->mnt_kern_flag & MNTK_MWAIT)
755 		wakeup(mp);
756 	if (freeok)
757 		kfree(mp, M_MOUNT);
758 	return (0);
759 }
760 
761 static
762 void
763 mount_warning(struct mount *mp, const char *ctl, ...)
764 {
765 	char *ptr;
766 	char *buf;
767 	__va_list va;
768 
769 	__va_start(va, ctl);
770 	if (cache_fullpath(NULL, &mp->mnt_ncmounton, &ptr, &buf) == 0) {
771 		kprintf("unmount(%s): ", ptr);
772 		kvprintf(ctl, va);
773 		kprintf("\n");
774 		kfree(buf, M_TEMP);
775 	} else {
776 		kprintf("unmount(%p", mp);
777 		if (mp->mnt_ncmounton.ncp && mp->mnt_ncmounton.ncp->nc_name)
778 			kprintf(",%s", mp->mnt_ncmounton.ncp->nc_name);
779 		kprintf("): ");
780 		kvprintf(ctl, va);
781 		kprintf("\n");
782 	}
783 	__va_end(va);
784 }
785 
786 /*
787  * Shim cache_fullpath() to handle the case where a process is chrooted into
788  * a subdirectory of a mount.  In this case if the root mount matches the
789  * process root directory's mount we have to specify the process's root
790  * directory instead of the mount point, because the mount point might
791  * be above the root directory.
792  */
793 static
794 int
795 mount_path(struct proc *p, struct mount *mp, char **rb, char **fb)
796 {
797 	struct nchandle *nch;
798 
799 	if (p && p->p_fd->fd_nrdir.mount == mp)
800 		nch = &p->p_fd->fd_nrdir;
801 	else
802 		nch = &mp->mnt_ncmountpt;
803 	return(cache_fullpath(p, nch, rb, fb));
804 }
805 
806 /*
807  * Sync each mounted filesystem.
808  */
809 
810 #ifdef DEBUG
811 static int syncprt = 0;
812 SYSCTL_INT(_debug, OID_AUTO, syncprt, CTLFLAG_RW, &syncprt, 0, "");
813 #endif /* DEBUG */
814 
815 static int sync_callback(struct mount *mp, void *data);
816 
817 /* ARGSUSED */
818 int
819 sys_sync(struct sync_args *uap)
820 {
821 	mountlist_scan(sync_callback, NULL, MNTSCAN_FORWARD);
822 #ifdef DEBUG
823 	/*
824 	 * print out buffer pool stat information on each sync() call.
825 	 */
826 	if (syncprt)
827 		vfs_bufstats();
828 #endif /* DEBUG */
829 	return (0);
830 }
831 
832 static
833 int
834 sync_callback(struct mount *mp, void *data __unused)
835 {
836 	int asyncflag;
837 
838 	if ((mp->mnt_flag & MNT_RDONLY) == 0) {
839 		asyncflag = mp->mnt_flag & MNT_ASYNC;
840 		mp->mnt_flag &= ~MNT_ASYNC;
841 		vfs_msync(mp, MNT_NOWAIT);
842 		VFS_SYNC(mp, MNT_NOWAIT);
843 		mp->mnt_flag |= asyncflag;
844 	}
845 	return(0);
846 }
847 
848 /* XXX PRISON: could be per prison flag */
849 static int prison_quotas;
850 #if 0
851 SYSCTL_INT(_kern_prison, OID_AUTO, quotas, CTLFLAG_RW, &prison_quotas, 0, "");
852 #endif
853 
854 /*
855  *  quotactl_args(char *path, int fcmd, int uid, caddr_t arg)
856  *
857  * Change filesystem quotas.
858  */
859 /* ARGSUSED */
860 int
861 sys_quotactl(struct quotactl_args *uap)
862 {
863 	struct nlookupdata nd;
864 	struct thread *td;
865 	struct proc *p;
866 	struct mount *mp;
867 	int error;
868 
869 	td = curthread;
870 	p = td->td_proc;
871 	if (p->p_ucred->cr_prison && !prison_quotas)
872 		return (EPERM);
873 
874 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
875 	if (error == 0)
876 		error = nlookup(&nd);
877 	if (error == 0) {
878 		mp = nd.nl_nch.mount;
879 		error = VFS_QUOTACTL(mp, uap->cmd, uap->uid,
880 				    uap->arg, nd.nl_cred);
881 	}
882 	nlookup_done(&nd);
883 	return (error);
884 }
885 
886 /*
887  * mountctl(char *path, int op, int fd, const void *ctl, int ctllen,
888  *		void *buf, int buflen)
889  *
890  * This function operates on a mount point and executes the specified
891  * operation using the specified control data, and possibly returns data.
892  *
893  * The actual number of bytes stored in the result buffer is returned, 0
894  * if none, otherwise an error is returned.
895  */
896 /* ARGSUSED */
897 int
898 sys_mountctl(struct mountctl_args *uap)
899 {
900 	struct thread *td = curthread;
901 	struct proc *p = td->td_proc;
902 	struct file *fp;
903 	void *ctl = NULL;
904 	void *buf = NULL;
905 	char *path = NULL;
906 	int error;
907 
908 	/*
909 	 * Sanity and permissions checks.  We must be root.
910 	 */
911 	KKASSERT(p);
912 	if (p->p_ucred->cr_prison != NULL)
913 		return (EPERM);
914 	if ((error = suser(td)) != 0)
915 		return (error);
916 
917 	/*
918 	 * Argument length checks
919 	 */
920 	if (uap->ctllen < 0 || uap->ctllen > 1024)
921 		return (EINVAL);
922 	if (uap->buflen < 0 || uap->buflen > 16 * 1024)
923 		return (EINVAL);
924 	if (uap->path == NULL)
925 		return (EINVAL);
926 
927 	/*
928 	 * Allocate the necessary buffers and copyin data
929 	 */
930 	path = objcache_get(namei_oc, M_WAITOK);
931 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
932 	if (error)
933 		goto done;
934 
935 	if (uap->ctllen) {
936 		ctl = kmalloc(uap->ctllen + 1, M_TEMP, M_WAITOK|M_ZERO);
937 		error = copyin(uap->ctl, ctl, uap->ctllen);
938 		if (error)
939 			goto done;
940 	}
941 	if (uap->buflen)
942 		buf = kmalloc(uap->buflen + 1, M_TEMP, M_WAITOK|M_ZERO);
943 
944 	/*
945 	 * Validate the descriptor
946 	 */
947 	if (uap->fd >= 0) {
948 		fp = holdfp(p->p_fd, uap->fd, -1);
949 		if (fp == NULL) {
950 			error = EBADF;
951 			goto done;
952 		}
953 	} else {
954 		fp = NULL;
955 	}
956 
957 	/*
958 	 * Execute the internal kernel function and clean up.
959 	 */
960 	error = kern_mountctl(path, uap->op, fp, ctl, uap->ctllen, buf, uap->buflen, &uap->sysmsg_result);
961 	if (fp)
962 		fdrop(fp);
963 	if (error == 0 && uap->sysmsg_result > 0)
964 		error = copyout(buf, uap->buf, uap->sysmsg_result);
965 done:
966 	if (path)
967 		objcache_put(namei_oc, path);
968 	if (ctl)
969 		kfree(ctl, M_TEMP);
970 	if (buf)
971 		kfree(buf, M_TEMP);
972 	return (error);
973 }
974 
975 /*
976  * Execute a mount control operation by resolving the path to a mount point
977  * and calling vop_mountctl().
978  *
979  * Use the mount point from the nch instead of the vnode so nullfs mounts
980  * can properly spike the VOP.
981  */
982 int
983 kern_mountctl(const char *path, int op, struct file *fp,
984 		const void *ctl, int ctllen,
985 		void *buf, int buflen, int *res)
986 {
987 	struct vnode *vp;
988 	struct mount *mp;
989 	struct nlookupdata nd;
990 	int error;
991 
992 	*res = 0;
993 	vp = NULL;
994 	error = nlookup_init(&nd, path, UIO_SYSSPACE, NLC_FOLLOW);
995 	if (error == 0)
996 		error = nlookup(&nd);
997 	if (error == 0)
998 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
999 	mp = nd.nl_nch.mount;
1000 	nlookup_done(&nd);
1001 	if (error)
1002 		return (error);
1003 
1004 	/*
1005 	 * Must be the root of the filesystem
1006 	 */
1007 	if ((vp->v_flag & (VROOT|VPFSROOT)) == 0) {
1008 		vput(vp);
1009 		return (EINVAL);
1010 	}
1011 	error = vop_mountctl(mp->mnt_vn_use_ops, op, fp, ctl, ctllen,
1012 				buf, buflen, res);
1013 	vput(vp);
1014 	return (error);
1015 }
1016 
1017 int
1018 kern_statfs(struct nlookupdata *nd, struct statfs *buf)
1019 {
1020 	struct thread *td = curthread;
1021 	struct proc *p = td->td_proc;
1022 	struct mount *mp;
1023 	struct statfs *sp;
1024 	char *fullpath, *freepath;
1025 	int error;
1026 
1027 	if ((error = nlookup(nd)) != 0)
1028 		return (error);
1029 	mp = nd->nl_nch.mount;
1030 	sp = &mp->mnt_stat;
1031 	if ((error = VFS_STATFS(mp, sp, nd->nl_cred)) != 0)
1032 		return (error);
1033 
1034 	error = mount_path(p, mp, &fullpath, &freepath);
1035 	if (error)
1036 		return(error);
1037 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1038 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1039 	kfree(freepath, M_TEMP);
1040 
1041 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1042 	bcopy(sp, buf, sizeof(*buf));
1043 	/* Only root should have access to the fsid's. */
1044 	if (suser(td))
1045 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1046 	return (0);
1047 }
1048 
1049 /*
1050  * statfs_args(char *path, struct statfs *buf)
1051  *
1052  * Get filesystem statistics.
1053  */
1054 int
1055 sys_statfs(struct statfs_args *uap)
1056 {
1057 	struct nlookupdata nd;
1058 	struct statfs buf;
1059 	int error;
1060 
1061 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1062 	if (error == 0)
1063 		error = kern_statfs(&nd, &buf);
1064 	nlookup_done(&nd);
1065 	if (error == 0)
1066 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1067 	return (error);
1068 }
1069 
1070 int
1071 kern_fstatfs(int fd, struct statfs *buf)
1072 {
1073 	struct thread *td = curthread;
1074 	struct proc *p = td->td_proc;
1075 	struct file *fp;
1076 	struct mount *mp;
1077 	struct statfs *sp;
1078 	char *fullpath, *freepath;
1079 	int error;
1080 
1081 	KKASSERT(p);
1082 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1083 		return (error);
1084 	mp = ((struct vnode *)fp->f_data)->v_mount;
1085 	if (mp == NULL) {
1086 		error = EBADF;
1087 		goto done;
1088 	}
1089 	if (fp->f_cred == NULL) {
1090 		error = EINVAL;
1091 		goto done;
1092 	}
1093 	sp = &mp->mnt_stat;
1094 	if ((error = VFS_STATFS(mp, sp, fp->f_cred)) != 0)
1095 		goto done;
1096 
1097 	if ((error = mount_path(p, mp, &fullpath, &freepath)) != 0)
1098 		goto done;
1099 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1100 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1101 	kfree(freepath, M_TEMP);
1102 
1103 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1104 	bcopy(sp, buf, sizeof(*buf));
1105 
1106 	/* Only root should have access to the fsid's. */
1107 	if (suser(td))
1108 		buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0;
1109 	error = 0;
1110 done:
1111 	fdrop(fp);
1112 	return (error);
1113 }
1114 
1115 /*
1116  * fstatfs_args(int fd, struct statfs *buf)
1117  *
1118  * Get filesystem statistics.
1119  */
1120 int
1121 sys_fstatfs(struct fstatfs_args *uap)
1122 {
1123 	struct statfs buf;
1124 	int error;
1125 
1126 	error = kern_fstatfs(uap->fd, &buf);
1127 
1128 	if (error == 0)
1129 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1130 	return (error);
1131 }
1132 
1133 int
1134 kern_statvfs(struct nlookupdata *nd, struct statvfs *buf)
1135 {
1136 	struct mount *mp;
1137 	struct statvfs *sp;
1138 	int error;
1139 
1140 	if ((error = nlookup(nd)) != 0)
1141 		return (error);
1142 	mp = nd->nl_nch.mount;
1143 	sp = &mp->mnt_vstat;
1144 	if ((error = VFS_STATVFS(mp, sp, nd->nl_cred)) != 0)
1145 		return (error);
1146 
1147 	sp->f_flag = 0;
1148 	if (mp->mnt_flag & MNT_RDONLY)
1149 		sp->f_flag |= ST_RDONLY;
1150 	if (mp->mnt_flag & MNT_NOSUID)
1151 		sp->f_flag |= ST_NOSUID;
1152 	bcopy(sp, buf, sizeof(*buf));
1153 	return (0);
1154 }
1155 
1156 /*
1157  * statfs_args(char *path, struct statfs *buf)
1158  *
1159  * Get filesystem statistics.
1160  */
1161 int
1162 sys_statvfs(struct statvfs_args *uap)
1163 {
1164 	struct nlookupdata nd;
1165 	struct statvfs buf;
1166 	int error;
1167 
1168 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1169 	if (error == 0)
1170 		error = kern_statvfs(&nd, &buf);
1171 	nlookup_done(&nd);
1172 	if (error == 0)
1173 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1174 	return (error);
1175 }
1176 
1177 int
1178 kern_fstatvfs(int fd, struct statvfs *buf)
1179 {
1180 	struct thread *td = curthread;
1181 	struct proc *p = td->td_proc;
1182 	struct file *fp;
1183 	struct mount *mp;
1184 	struct statvfs *sp;
1185 	int error;
1186 
1187 	KKASSERT(p);
1188 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
1189 		return (error);
1190 	mp = ((struct vnode *)fp->f_data)->v_mount;
1191 	if (mp == NULL) {
1192 		error = EBADF;
1193 		goto done;
1194 	}
1195 	if (fp->f_cred == NULL) {
1196 		error = EINVAL;
1197 		goto done;
1198 	}
1199 	sp = &mp->mnt_vstat;
1200 	if ((error = VFS_STATVFS(mp, sp, fp->f_cred)) != 0)
1201 		goto done;
1202 
1203 	sp->f_flag = 0;
1204 	if (mp->mnt_flag & MNT_RDONLY)
1205 		sp->f_flag |= ST_RDONLY;
1206 	if (mp->mnt_flag & MNT_NOSUID)
1207 		sp->f_flag |= ST_NOSUID;
1208 
1209 	bcopy(sp, buf, sizeof(*buf));
1210 	error = 0;
1211 done:
1212 	fdrop(fp);
1213 	return (error);
1214 }
1215 
1216 /*
1217  * fstatfs_args(int fd, struct statfs *buf)
1218  *
1219  * Get filesystem statistics.
1220  */
1221 int
1222 sys_fstatvfs(struct fstatvfs_args *uap)
1223 {
1224 	struct statvfs buf;
1225 	int error;
1226 
1227 	error = kern_fstatvfs(uap->fd, &buf);
1228 
1229 	if (error == 0)
1230 		error = copyout(&buf, uap->buf, sizeof(*uap->buf));
1231 	return (error);
1232 }
1233 
1234 /*
1235  * getfsstat_args(struct statfs *buf, long bufsize, int flags)
1236  *
1237  * Get statistics on all filesystems.
1238  */
1239 
1240 struct getfsstat_info {
1241 	struct statfs *sfsp;
1242 	long count;
1243 	long maxcount;
1244 	int error;
1245 	int flags;
1246 	struct proc *p;
1247 };
1248 
1249 static int getfsstat_callback(struct mount *, void *);
1250 
1251 /* ARGSUSED */
1252 int
1253 sys_getfsstat(struct getfsstat_args *uap)
1254 {
1255 	struct thread *td = curthread;
1256 	struct proc *p = td->td_proc;
1257 	struct getfsstat_info info;
1258 
1259 	bzero(&info, sizeof(info));
1260 
1261 	info.maxcount = uap->bufsize / sizeof(struct statfs);
1262 	info.sfsp = uap->buf;
1263 	info.count = 0;
1264 	info.flags = uap->flags;
1265 	info.p = p;
1266 
1267 	mountlist_scan(getfsstat_callback, &info, MNTSCAN_FORWARD);
1268 	if (info.sfsp && info.count > info.maxcount)
1269 		uap->sysmsg_result = info.maxcount;
1270 	else
1271 		uap->sysmsg_result = info.count;
1272 	return (info.error);
1273 }
1274 
1275 static int
1276 getfsstat_callback(struct mount *mp, void *data)
1277 {
1278 	struct getfsstat_info *info = data;
1279 	struct statfs *sp;
1280 	char *freepath;
1281 	char *fullpath;
1282 	int error;
1283 
1284 	if (info->sfsp && info->count < info->maxcount) {
1285 		if (info->p && !chroot_visible_mnt(mp, info->p))
1286 			return(0);
1287 		sp = &mp->mnt_stat;
1288 
1289 		/*
1290 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1291 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1292 		 * overrides MNT_WAIT.
1293 		 */
1294 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1295 		    (info->flags & MNT_WAIT)) &&
1296 		    (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1297 			return(0);
1298 		}
1299 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1300 
1301 		error = mount_path(info->p, mp, &fullpath, &freepath);
1302 		if (error) {
1303 			info->error = error;
1304 			return(-1);
1305 		}
1306 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1307 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1308 		kfree(freepath, M_TEMP);
1309 
1310 		error = copyout(sp, info->sfsp, sizeof(*sp));
1311 		if (error) {
1312 			info->error = error;
1313 			return (-1);
1314 		}
1315 		++info->sfsp;
1316 	}
1317 	info->count++;
1318 	return(0);
1319 }
1320 
1321 /*
1322  * getvfsstat_args(struct statfs *buf, struct statvfs *vbuf,
1323 		   long bufsize, int flags)
1324  *
1325  * Get statistics on all filesystems.
1326  */
1327 
1328 struct getvfsstat_info {
1329 	struct statfs *sfsp;
1330 	struct statvfs *vsfsp;
1331 	long count;
1332 	long maxcount;
1333 	int error;
1334 	int flags;
1335 	struct proc *p;
1336 };
1337 
1338 static int getvfsstat_callback(struct mount *, void *);
1339 
1340 /* ARGSUSED */
1341 int
1342 sys_getvfsstat(struct getvfsstat_args *uap)
1343 {
1344 	struct thread *td = curthread;
1345 	struct proc *p = td->td_proc;
1346 	struct getvfsstat_info info;
1347 
1348 	bzero(&info, sizeof(info));
1349 
1350 	info.maxcount = uap->vbufsize / sizeof(struct statvfs);
1351 	info.sfsp = uap->buf;
1352 	info.vsfsp = uap->vbuf;
1353 	info.count = 0;
1354 	info.flags = uap->flags;
1355 	info.p = p;
1356 
1357 	mountlist_scan(getvfsstat_callback, &info, MNTSCAN_FORWARD);
1358 	if (info.vsfsp && info.count > info.maxcount)
1359 		uap->sysmsg_result = info.maxcount;
1360 	else
1361 		uap->sysmsg_result = info.count;
1362 	return (info.error);
1363 }
1364 
1365 static int
1366 getvfsstat_callback(struct mount *mp, void *data)
1367 {
1368 	struct getvfsstat_info *info = data;
1369 	struct statfs *sp;
1370 	struct statvfs *vsp;
1371 	char *freepath;
1372 	char *fullpath;
1373 	int error;
1374 
1375 	if (info->vsfsp && info->count < info->maxcount) {
1376 		if (info->p && !chroot_visible_mnt(mp, info->p))
1377 			return(0);
1378 		sp = &mp->mnt_stat;
1379 		vsp = &mp->mnt_vstat;
1380 
1381 		/*
1382 		 * If MNT_NOWAIT or MNT_LAZY is specified, do not
1383 		 * refresh the fsstat cache. MNT_NOWAIT or MNT_LAZY
1384 		 * overrides MNT_WAIT.
1385 		 */
1386 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1387 		    (info->flags & MNT_WAIT)) &&
1388 		    (error = VFS_STATFS(mp, sp, info->p->p_ucred))) {
1389 			return(0);
1390 		}
1391 		sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
1392 
1393 		if (((info->flags & (MNT_LAZY|MNT_NOWAIT)) == 0 ||
1394 		    (info->flags & MNT_WAIT)) &&
1395 		    (error = VFS_STATVFS(mp, vsp, info->p->p_ucred))) {
1396 			return(0);
1397 		}
1398 		vsp->f_flag = 0;
1399 		if (mp->mnt_flag & MNT_RDONLY)
1400 			vsp->f_flag |= ST_RDONLY;
1401 		if (mp->mnt_flag & MNT_NOSUID)
1402 			vsp->f_flag |= ST_NOSUID;
1403 
1404 		error = mount_path(info->p, mp, &fullpath, &freepath);
1405 		if (error) {
1406 			info->error = error;
1407 			return(-1);
1408 		}
1409 		bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
1410 		strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
1411 		kfree(freepath, M_TEMP);
1412 
1413 		error = copyout(sp, info->sfsp, sizeof(*sp));
1414 		if (error == 0)
1415 			error = copyout(vsp, info->vsfsp, sizeof(*vsp));
1416 		if (error) {
1417 			info->error = error;
1418 			return (-1);
1419 		}
1420 		++info->sfsp;
1421 		++info->vsfsp;
1422 	}
1423 	info->count++;
1424 	return(0);
1425 }
1426 
1427 
1428 /*
1429  * fchdir_args(int fd)
1430  *
1431  * Change current working directory to a given file descriptor.
1432  */
1433 /* ARGSUSED */
1434 int
1435 sys_fchdir(struct fchdir_args *uap)
1436 {
1437 	struct thread *td = curthread;
1438 	struct proc *p = td->td_proc;
1439 	struct filedesc *fdp = p->p_fd;
1440 	struct vnode *vp, *ovp;
1441 	struct mount *mp;
1442 	struct file *fp;
1443 	struct nchandle nch, onch, tnch;
1444 	int error;
1445 
1446 	if ((error = holdvnode(fdp, uap->fd, &fp)) != 0)
1447 		return (error);
1448 	vp = (struct vnode *)fp->f_data;
1449 	vref(vp);
1450 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1451 	if (vp->v_type != VDIR || fp->f_nchandle.ncp == NULL)
1452 		error = ENOTDIR;
1453 	else
1454 		error = VOP_ACCESS(vp, VEXEC, p->p_ucred);
1455 	if (error) {
1456 		vput(vp);
1457 		fdrop(fp);
1458 		return (error);
1459 	}
1460 	cache_copy(&fp->f_nchandle, &nch);
1461 
1462 	/*
1463 	 * If the ncp has become a mount point, traverse through
1464 	 * the mount point.
1465 	 */
1466 
1467 	while (!error && (nch.ncp->nc_flag & NCF_ISMOUNTPT) &&
1468 	       (mp = cache_findmount(&nch)) != NULL
1469 	) {
1470 		error = nlookup_mp(mp, &tnch);
1471 		if (error == 0) {
1472 			cache_unlock(&tnch);	/* leave ref intact */
1473 			vput(vp);
1474 			vp = tnch.ncp->nc_vp;
1475 			error = vget(vp, LK_SHARED);
1476 			KKASSERT(error == 0);
1477 			cache_drop(&nch);
1478 			nch = tnch;
1479 		}
1480 	}
1481 	if (error == 0) {
1482 		ovp = fdp->fd_cdir;
1483 		onch = fdp->fd_ncdir;
1484 		vn_unlock(vp);		/* leave ref intact */
1485 		fdp->fd_cdir = vp;
1486 		fdp->fd_ncdir = nch;
1487 		cache_drop(&onch);
1488 		vrele(ovp);
1489 	} else {
1490 		cache_drop(&nch);
1491 		vput(vp);
1492 	}
1493 	fdrop(fp);
1494 	return (error);
1495 }
1496 
1497 int
1498 kern_chdir(struct nlookupdata *nd)
1499 {
1500 	struct thread *td = curthread;
1501 	struct proc *p = td->td_proc;
1502 	struct filedesc *fdp = p->p_fd;
1503 	struct vnode *vp, *ovp;
1504 	struct nchandle onch;
1505 	int error;
1506 
1507 	if ((error = nlookup(nd)) != 0)
1508 		return (error);
1509 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
1510 		return (ENOENT);
1511 	if ((error = vget(vp, LK_SHARED)) != 0)
1512 		return (error);
1513 
1514 	error = checkvp_chdir(vp, td);
1515 	vn_unlock(vp);
1516 	if (error == 0) {
1517 		ovp = fdp->fd_cdir;
1518 		onch = fdp->fd_ncdir;
1519 		cache_unlock(&nd->nl_nch);	/* leave reference intact */
1520 		fdp->fd_ncdir = nd->nl_nch;
1521 		fdp->fd_cdir = vp;
1522 		cache_drop(&onch);
1523 		vrele(ovp);
1524 		cache_zero(&nd->nl_nch);
1525 	} else {
1526 		vrele(vp);
1527 	}
1528 	return (error);
1529 }
1530 
1531 /*
1532  * chdir_args(char *path)
1533  *
1534  * Change current working directory (``.'').
1535  */
1536 int
1537 sys_chdir(struct chdir_args *uap)
1538 {
1539 	struct nlookupdata nd;
1540 	int error;
1541 
1542 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1543 	if (error == 0)
1544 		error = kern_chdir(&nd);
1545 	nlookup_done(&nd);
1546 	return (error);
1547 }
1548 
1549 /*
1550  * Helper function for raised chroot(2) security function:  Refuse if
1551  * any filedescriptors are open directories.
1552  */
1553 static int
1554 chroot_refuse_vdir_fds(struct filedesc *fdp)
1555 {
1556 	struct vnode *vp;
1557 	struct file *fp;
1558 	int error;
1559 	int fd;
1560 
1561 	for (fd = 0; fd < fdp->fd_nfiles ; fd++) {
1562 		if ((error = holdvnode(fdp, fd, &fp)) != 0)
1563 			continue;
1564 		vp = (struct vnode *)fp->f_data;
1565 		if (vp->v_type != VDIR) {
1566 			fdrop(fp);
1567 			continue;
1568 		}
1569 		fdrop(fp);
1570 		return(EPERM);
1571 	}
1572 	return (0);
1573 }
1574 
1575 /*
1576  * This sysctl determines if we will allow a process to chroot(2) if it
1577  * has a directory open:
1578  *	0: disallowed for all processes.
1579  *	1: allowed for processes that were not already chroot(2)'ed.
1580  *	2: allowed for all processes.
1581  */
1582 
1583 static int chroot_allow_open_directories = 1;
1584 
1585 SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW,
1586      &chroot_allow_open_directories, 0, "");
1587 
1588 /*
1589  * chroot to the specified namecache entry.  We obtain the vp from the
1590  * namecache data.  The passed ncp must be locked and referenced and will
1591  * remain locked and referenced on return.
1592  */
1593 int
1594 kern_chroot(struct nchandle *nch)
1595 {
1596 	struct thread *td = curthread;
1597 	struct proc *p = td->td_proc;
1598 	struct filedesc *fdp = p->p_fd;
1599 	struct vnode *vp;
1600 	int error;
1601 
1602 	/*
1603 	 * Only root can chroot
1604 	 */
1605 	if ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0)
1606 		return (error);
1607 
1608 	/*
1609 	 * Disallow open directory descriptors (fchdir() breakouts).
1610 	 */
1611 	if (chroot_allow_open_directories == 0 ||
1612 	   (chroot_allow_open_directories == 1 && fdp->fd_rdir != rootvnode)) {
1613 		if ((error = chroot_refuse_vdir_fds(fdp)) != 0)
1614 			return (error);
1615 	}
1616 	if ((vp = nch->ncp->nc_vp) == NULL)
1617 		return (ENOENT);
1618 
1619 	if ((error = vget(vp, LK_SHARED)) != 0)
1620 		return (error);
1621 
1622 	/*
1623 	 * Check the validity of vp as a directory to change to and
1624 	 * associate it with rdir/jdir.
1625 	 */
1626 	error = checkvp_chdir(vp, td);
1627 	vn_unlock(vp);			/* leave reference intact */
1628 	if (error == 0) {
1629 		vrele(fdp->fd_rdir);
1630 		fdp->fd_rdir = vp;	/* reference inherited by fd_rdir */
1631 		cache_drop(&fdp->fd_nrdir);
1632 		cache_copy(nch, &fdp->fd_nrdir);
1633 		if (fdp->fd_jdir == NULL) {
1634 			fdp->fd_jdir = vp;
1635 			vref(fdp->fd_jdir);
1636 			cache_copy(nch, &fdp->fd_njdir);
1637 		}
1638 	} else {
1639 		vrele(vp);
1640 	}
1641 	return (error);
1642 }
1643 
1644 /*
1645  * chroot_args(char *path)
1646  *
1647  * Change notion of root (``/'') directory.
1648  */
1649 /* ARGSUSED */
1650 int
1651 sys_chroot(struct chroot_args *uap)
1652 {
1653 	struct thread *td = curthread;
1654 	struct nlookupdata nd;
1655 	int error;
1656 
1657 	KKASSERT(td->td_proc);
1658 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1659 	if (error) {
1660 		nlookup_done(&nd);
1661 		return(error);
1662 	}
1663 	error = nlookup(&nd);
1664 	if (error == 0)
1665 		error = kern_chroot(&nd.nl_nch);
1666 	nlookup_done(&nd);
1667 	return(error);
1668 }
1669 
1670 /*
1671  * Common routine for chroot and chdir.  Given a locked, referenced vnode,
1672  * determine whether it is legal to chdir to the vnode.  The vnode's state
1673  * is not changed by this call.
1674  */
1675 int
1676 checkvp_chdir(struct vnode *vp, struct thread *td)
1677 {
1678 	int error;
1679 
1680 	if (vp->v_type != VDIR)
1681 		error = ENOTDIR;
1682 	else
1683 		error = VOP_ACCESS(vp, VEXEC, td->td_proc->p_ucred);
1684 	return (error);
1685 }
1686 
1687 int
1688 kern_open(struct nlookupdata *nd, int oflags, int mode, int *res)
1689 {
1690 	struct thread *td = curthread;
1691 	struct proc *p = td->td_proc;
1692 	struct lwp *lp = td->td_lwp;
1693 	struct filedesc *fdp = p->p_fd;
1694 	int cmode, flags;
1695 	struct file *nfp;
1696 	struct file *fp;
1697 	struct vnode *vp;
1698 	int type, indx, error;
1699 	struct flock lf;
1700 
1701 	if ((oflags & O_ACCMODE) == O_ACCMODE)
1702 		return (EINVAL);
1703 	flags = FFLAGS(oflags);
1704 	error = falloc(p, &nfp, NULL);
1705 	if (error)
1706 		return (error);
1707 	fp = nfp;
1708 	cmode = ((mode &~ fdp->fd_cmask) & ALLPERMS) &~ S_ISTXT;
1709 
1710 	/*
1711 	 * XXX p_dupfd is a real mess.  It allows a device to return a
1712 	 * file descriptor to be duplicated rather then doing the open
1713 	 * itself.
1714 	 */
1715 	lp->lwp_dupfd = -1;
1716 
1717 	/*
1718 	 * Call vn_open() to do the lookup and assign the vnode to the
1719 	 * file pointer.  vn_open() does not change the ref count on fp
1720 	 * and the vnode, on success, will be inherited by the file pointer
1721 	 * and unlocked.
1722 	 */
1723 	nd->nl_flags |= NLC_LOCKVP;
1724 	error = vn_open(nd, fp, flags, cmode);
1725 	nlookup_done(nd);
1726 	if (error) {
1727 		/*
1728 		 * handle special fdopen() case.  bleh.  dupfdopen() is
1729 		 * responsible for dropping the old contents of ofiles[indx]
1730 		 * if it succeeds.
1731 		 *
1732 		 * Note that fsetfd() will add a ref to fp which represents
1733 		 * the fd_files[] assignment.  We must still drop our
1734 		 * reference.
1735 		 */
1736 		if ((error == ENODEV || error == ENXIO) && lp->lwp_dupfd >= 0) {
1737 			if (fdalloc(p, 0, &indx) == 0) {
1738 				error = dupfdopen(p, indx, lp->lwp_dupfd, flags, error);
1739 				if (error == 0) {
1740 					*res = indx;
1741 					fdrop(fp);	/* our ref */
1742 					return (0);
1743 				}
1744 				fsetfd(p, NULL, indx);
1745 			}
1746 		}
1747 		fdrop(fp);	/* our ref */
1748 		if (error == ERESTART)
1749 			error = EINTR;
1750 		return (error);
1751 	}
1752 
1753 	/*
1754 	 * ref the vnode for ourselves so it can't be ripped out from under
1755 	 * is.  XXX need an ND flag to request that the vnode be returned
1756 	 * anyway.
1757 	 *
1758 	 * Reserve a file descriptor but do not assign it until the open
1759 	 * succeeds.
1760 	 */
1761 	vp = (struct vnode *)fp->f_data;
1762 	vref(vp);
1763 	if ((error = fdalloc(p, 0, &indx)) != 0) {
1764 		fdrop(fp);
1765 		vrele(vp);
1766 		return (error);
1767 	}
1768 
1769 	/*
1770 	 * If no error occurs the vp will have been assigned to the file
1771 	 * pointer.
1772 	 */
1773 	lp->lwp_dupfd = 0;
1774 
1775 	if (flags & (O_EXLOCK | O_SHLOCK)) {
1776 		lf.l_whence = SEEK_SET;
1777 		lf.l_start = 0;
1778 		lf.l_len = 0;
1779 		if (flags & O_EXLOCK)
1780 			lf.l_type = F_WRLCK;
1781 		else
1782 			lf.l_type = F_RDLCK;
1783 		if (flags & FNONBLOCK)
1784 			type = 0;
1785 		else
1786 			type = F_WAIT;
1787 
1788 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
1789 			/*
1790 			 * lock request failed.  Clean up the reserved
1791 			 * descriptor.
1792 			 */
1793 			vrele(vp);
1794 			fsetfd(p, NULL, indx);
1795 			fdrop(fp);
1796 			return (error);
1797 		}
1798 		fp->f_flag |= FHASLOCK;
1799 	}
1800 #if 0
1801 	/*
1802 	 * Assert that all regular file vnodes were created with a object.
1803 	 */
1804 	KASSERT(vp->v_type != VREG || vp->v_object != NULL,
1805 		("open: regular file has no backing object after vn_open"));
1806 #endif
1807 
1808 	vrele(vp);
1809 
1810 	/*
1811 	 * release our private reference, leaving the one associated with the
1812 	 * descriptor table intact.
1813 	 */
1814 	fsetfd(p, fp, indx);
1815 	fdrop(fp);
1816 	*res = indx;
1817 	return (0);
1818 }
1819 
1820 /*
1821  * open_args(char *path, int flags, int mode)
1822  *
1823  * Check permissions, allocate an open file structure,
1824  * and call the device open routine if any.
1825  */
1826 int
1827 sys_open(struct open_args *uap)
1828 {
1829 	struct nlookupdata nd;
1830 	int error;
1831 
1832 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
1833 	if (error == 0) {
1834 		error = kern_open(&nd, uap->flags,
1835 				    uap->mode, &uap->sysmsg_result);
1836 	}
1837 	nlookup_done(&nd);
1838 	return (error);
1839 }
1840 
1841 int
1842 kern_mknod(struct nlookupdata *nd, int mode, int rmajor, int rminor)
1843 {
1844 	struct thread *td = curthread;
1845 	struct proc *p = td->td_proc;
1846 	struct vnode *vp;
1847 	struct vattr vattr;
1848 	int error;
1849 	int whiteout = 0;
1850 
1851 	KKASSERT(p);
1852 
1853 	switch (mode & S_IFMT) {
1854 	case S_IFCHR:
1855 	case S_IFBLK:
1856 		error = suser(td);
1857 		break;
1858 	default:
1859 		error = suser_cred(p->p_ucred, PRISON_ROOT);
1860 		break;
1861 	}
1862 	if (error)
1863 		return (error);
1864 
1865 	bwillinode(1);
1866 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1867 	if ((error = nlookup(nd)) != 0)
1868 		return (error);
1869 	if (nd->nl_nch.ncp->nc_vp)
1870 		return (EEXIST);
1871 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1872 		return (error);
1873 
1874 	VATTR_NULL(&vattr);
1875 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1876 	vattr.va_rmajor = rmajor;
1877 	vattr.va_rminor = rminor;
1878 	whiteout = 0;
1879 
1880 	switch (mode & S_IFMT) {
1881 	case S_IFMT:	/* used by badsect to flag bad sectors */
1882 		vattr.va_type = VBAD;
1883 		break;
1884 	case S_IFCHR:
1885 		vattr.va_type = VCHR;
1886 		break;
1887 	case S_IFBLK:
1888 		vattr.va_type = VBLK;
1889 		break;
1890 	case S_IFWHT:
1891 		whiteout = 1;
1892 		break;
1893 	case S_IFDIR:
1894 		/* special directories support for HAMMER */
1895 		vattr.va_type = VDIR;
1896 		break;
1897 	default:
1898 		error = EINVAL;
1899 		break;
1900 	}
1901 	if (error == 0) {
1902 		if (whiteout) {
1903 			error = VOP_NWHITEOUT(&nd->nl_nch, nd->nl_dvp,
1904 					      nd->nl_cred, NAMEI_CREATE);
1905 		} else {
1906 			vp = NULL;
1907 			error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp,
1908 					   &vp, nd->nl_cred, &vattr);
1909 			if (error == 0)
1910 				vput(vp);
1911 		}
1912 	}
1913 	return (error);
1914 }
1915 
1916 /*
1917  * mknod_args(char *path, int mode, int dev)
1918  *
1919  * Create a special file.
1920  */
1921 int
1922 sys_mknod(struct mknod_args *uap)
1923 {
1924 	struct nlookupdata nd;
1925 	int error;
1926 
1927 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1928 	if (error == 0) {
1929 		error = kern_mknod(&nd, uap->mode,
1930 				   umajor(uap->dev), uminor(uap->dev));
1931 	}
1932 	nlookup_done(&nd);
1933 	return (error);
1934 }
1935 
1936 int
1937 kern_mkfifo(struct nlookupdata *nd, int mode)
1938 {
1939 	struct thread *td = curthread;
1940 	struct proc *p = td->td_proc;
1941 	struct vattr vattr;
1942 	struct vnode *vp;
1943 	int error;
1944 
1945 	bwillinode(1);
1946 
1947 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
1948 	if ((error = nlookup(nd)) != 0)
1949 		return (error);
1950 	if (nd->nl_nch.ncp->nc_vp)
1951 		return (EEXIST);
1952 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
1953 		return (error);
1954 
1955 	VATTR_NULL(&vattr);
1956 	vattr.va_type = VFIFO;
1957 	vattr.va_mode = (mode & ALLPERMS) &~ p->p_fd->fd_cmask;
1958 	vp = NULL;
1959 	error = VOP_NMKNOD(&nd->nl_nch, nd->nl_dvp, &vp, nd->nl_cred, &vattr);
1960 	if (error == 0)
1961 		vput(vp);
1962 	return (error);
1963 }
1964 
1965 /*
1966  * mkfifo_args(char *path, int mode)
1967  *
1968  * Create a named pipe.
1969  */
1970 int
1971 sys_mkfifo(struct mkfifo_args *uap)
1972 {
1973 	struct nlookupdata nd;
1974 	int error;
1975 
1976 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
1977 	if (error == 0)
1978 		error = kern_mkfifo(&nd, uap->mode);
1979 	nlookup_done(&nd);
1980 	return (error);
1981 }
1982 
1983 static int hardlink_check_uid = 0;
1984 SYSCTL_INT(_security, OID_AUTO, hardlink_check_uid, CTLFLAG_RW,
1985     &hardlink_check_uid, 0,
1986     "Unprivileged processes cannot create hard links to files owned by other "
1987     "users");
1988 static int hardlink_check_gid = 0;
1989 SYSCTL_INT(_security, OID_AUTO, hardlink_check_gid, CTLFLAG_RW,
1990     &hardlink_check_gid, 0,
1991     "Unprivileged processes cannot create hard links to files owned by other "
1992     "groups");
1993 
1994 static int
1995 can_hardlink(struct vnode *vp, struct thread *td, struct ucred *cred)
1996 {
1997 	struct vattr va;
1998 	int error;
1999 
2000 	/*
2001 	 * Shortcut if disabled
2002 	 */
2003 	if (hardlink_check_uid == 0 && hardlink_check_gid == 0)
2004 		return (0);
2005 
2006 	/*
2007 	 * root cred can always hardlink
2008 	 */
2009 	if (suser_cred(cred, PRISON_ROOT) == 0)
2010 		return (0);
2011 
2012 	/*
2013 	 * Otherwise only if the originating file is owned by the
2014 	 * same user or group.  Note that any group is allowed if
2015 	 * the file is owned by the caller.
2016 	 */
2017 	error = VOP_GETATTR(vp, &va);
2018 	if (error != 0)
2019 		return (error);
2020 
2021 	if (hardlink_check_uid) {
2022 		if (cred->cr_uid != va.va_uid)
2023 			return (EPERM);
2024 	}
2025 
2026 	if (hardlink_check_gid) {
2027 		if (cred->cr_uid != va.va_uid && !groupmember(va.va_gid, cred))
2028 			return (EPERM);
2029 	}
2030 
2031 	return (0);
2032 }
2033 
2034 int
2035 kern_link(struct nlookupdata *nd, struct nlookupdata *linknd)
2036 {
2037 	struct thread *td = curthread;
2038 	struct vnode *vp;
2039 	int error;
2040 
2041 	/*
2042 	 * Lookup the source and obtained a locked vnode.
2043 	 *
2044 	 * XXX relookup on vget failure / race ?
2045 	 */
2046 	bwillinode(1);
2047 	if ((error = nlookup(nd)) != 0)
2048 		return (error);
2049 	vp = nd->nl_nch.ncp->nc_vp;
2050 	KKASSERT(vp != NULL);
2051 	if (vp->v_type == VDIR)
2052 		return (EPERM);		/* POSIX */
2053 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2054 		return (error);
2055 	if ((error = vget(vp, LK_EXCLUSIVE)) != 0)
2056 		return (error);
2057 
2058 	/*
2059 	 * Unlock the source so we can lookup the target without deadlocking
2060 	 * (XXX vp is locked already, possible other deadlock?).  The target
2061 	 * must not exist.
2062 	 */
2063 	KKASSERT(nd->nl_flags & NLC_NCPISLOCKED);
2064 	nd->nl_flags &= ~NLC_NCPISLOCKED;
2065 	cache_unlock(&nd->nl_nch);
2066 
2067 	linknd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2068 	if ((error = nlookup(linknd)) != 0) {
2069 		vput(vp);
2070 		return (error);
2071 	}
2072 	if (linknd->nl_nch.ncp->nc_vp) {
2073 		vput(vp);
2074 		return (EEXIST);
2075 	}
2076 
2077 	/*
2078 	 * Finally run the new API VOP.
2079 	 */
2080 	error = can_hardlink(vp, td, td->td_proc->p_ucred);
2081 	if (error == 0) {
2082 		error = VOP_NLINK(&linknd->nl_nch, linknd->nl_dvp,
2083 				  vp, linknd->nl_cred);
2084 	}
2085 	vput(vp);
2086 	return (error);
2087 }
2088 
2089 /*
2090  * link_args(char *path, char *link)
2091  *
2092  * Make a hard file link.
2093  */
2094 int
2095 sys_link(struct link_args *uap)
2096 {
2097 	struct nlookupdata nd, linknd;
2098 	int error;
2099 
2100 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2101 	if (error == 0) {
2102 		error = nlookup_init(&linknd, uap->link, UIO_USERSPACE, 0);
2103 		if (error == 0)
2104 			error = kern_link(&nd, &linknd);
2105 		nlookup_done(&linknd);
2106 	}
2107 	nlookup_done(&nd);
2108 	return (error);
2109 }
2110 
2111 int
2112 kern_symlink(struct nlookupdata *nd, char *path, int mode)
2113 {
2114 	struct vattr vattr;
2115 	struct vnode *vp;
2116 	struct vnode *dvp;
2117 	int error;
2118 
2119 	bwillinode(1);
2120 	nd->nl_flags |= NLC_CREATE | NLC_REFDVP;
2121 	if ((error = nlookup(nd)) != 0)
2122 		return (error);
2123 	if (nd->nl_nch.ncp->nc_vp)
2124 		return (EEXIST);
2125 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2126 		return (error);
2127 	dvp = nd->nl_dvp;
2128 	VATTR_NULL(&vattr);
2129 	vattr.va_mode = mode;
2130 	error = VOP_NSYMLINK(&nd->nl_nch, dvp, &vp, nd->nl_cred, &vattr, path);
2131 	if (error == 0)
2132 		vput(vp);
2133 	return (error);
2134 }
2135 
2136 /*
2137  * symlink(char *path, char *link)
2138  *
2139  * Make a symbolic link.
2140  */
2141 int
2142 sys_symlink(struct symlink_args *uap)
2143 {
2144 	struct thread *td = curthread;
2145 	struct nlookupdata nd;
2146 	char *path;
2147 	int error;
2148 	int mode;
2149 
2150 	path = objcache_get(namei_oc, M_WAITOK);
2151 	error = copyinstr(uap->path, path, MAXPATHLEN, NULL);
2152 	if (error == 0) {
2153 		error = nlookup_init(&nd, uap->link, UIO_USERSPACE, 0);
2154 		if (error == 0) {
2155 			mode = ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask;
2156 			error = kern_symlink(&nd, path, mode);
2157 		}
2158 		nlookup_done(&nd);
2159 	}
2160 	objcache_put(namei_oc, path);
2161 	return (error);
2162 }
2163 
2164 /*
2165  * undelete_args(char *path)
2166  *
2167  * Delete a whiteout from the filesystem.
2168  */
2169 /* ARGSUSED */
2170 int
2171 sys_undelete(struct undelete_args *uap)
2172 {
2173 	struct nlookupdata nd;
2174 	int error;
2175 
2176 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2177 	bwillinode(1);
2178 	nd.nl_flags |= NLC_DELETE | NLC_REFDVP;
2179 	if (error == 0)
2180 		error = nlookup(&nd);
2181 	if (error == 0)
2182 		error = ncp_writechk(&nd.nl_nch);
2183 	if (error == 0) {
2184 		error = VOP_NWHITEOUT(&nd.nl_nch, nd.nl_dvp, nd.nl_cred,
2185 				      NAMEI_DELETE);
2186 	}
2187 	nlookup_done(&nd);
2188 	return (error);
2189 }
2190 
2191 int
2192 kern_unlink(struct nlookupdata *nd)
2193 {
2194 	int error;
2195 
2196 	bwillinode(1);
2197 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
2198 	if ((error = nlookup(nd)) != 0)
2199 		return (error);
2200 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2201 		return (error);
2202 	error = VOP_NREMOVE(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
2203 	return (error);
2204 }
2205 
2206 /*
2207  * unlink_args(char *path)
2208  *
2209  * Delete a name from the filesystem.
2210  */
2211 int
2212 sys_unlink(struct unlink_args *uap)
2213 {
2214 	struct nlookupdata nd;
2215 	int error;
2216 
2217 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2218 	if (error == 0)
2219 		error = kern_unlink(&nd);
2220 	nlookup_done(&nd);
2221 	return (error);
2222 }
2223 
2224 int
2225 kern_lseek(int fd, off_t offset, int whence, off_t *res)
2226 {
2227 	struct thread *td = curthread;
2228 	struct proc *p = td->td_proc;
2229 	struct file *fp;
2230 	struct vnode *vp;
2231 	struct vattr vattr;
2232 	off_t new_offset;
2233 	int error;
2234 
2235 	fp = holdfp(p->p_fd, fd, -1);
2236 	if (fp == NULL)
2237 		return (EBADF);
2238 	if (fp->f_type != DTYPE_VNODE) {
2239 		error = ESPIPE;
2240 		goto done;
2241 	}
2242 	vp = (struct vnode *)fp->f_data;
2243 
2244 	switch (whence) {
2245 	case L_INCR:
2246 		new_offset = fp->f_offset + offset;
2247 		error = 0;
2248 		break;
2249 	case L_XTND:
2250 		error = VOP_GETATTR(vp, &vattr);
2251 		new_offset = offset + vattr.va_size;
2252 		break;
2253 	case L_SET:
2254 		new_offset = offset;
2255 		error = 0;
2256 		break;
2257 	default:
2258 		new_offset = 0;
2259 		error = EINVAL;
2260 		break;
2261 	}
2262 
2263 	/*
2264 	 * Validate the seek position.  Negative offsets are not allowed
2265 	 * for regular files, block specials, or directories.
2266 	 */
2267 	if (error == 0) {
2268 		if (new_offset < 0 &&
2269 		    (vp->v_type == VREG || vp->v_type == VDIR ||
2270 		     vp->v_type == VCHR || vp->v_type == VBLK)) {
2271 			error = EINVAL;
2272 		} else {
2273 			fp->f_offset = new_offset;
2274 		}
2275 	}
2276 	*res = fp->f_offset;
2277 done:
2278 	fdrop(fp);
2279 	return (error);
2280 }
2281 
2282 /*
2283  * lseek_args(int fd, int pad, off_t offset, int whence)
2284  *
2285  * Reposition read/write file offset.
2286  */
2287 int
2288 sys_lseek(struct lseek_args *uap)
2289 {
2290 	int error;
2291 
2292 	error = kern_lseek(uap->fd, uap->offset, uap->whence,
2293 			   &uap->sysmsg_offset);
2294 
2295 	return (error);
2296 }
2297 
2298 int
2299 kern_access(struct nlookupdata *nd, int aflags)
2300 {
2301 	struct vnode *vp;
2302 	int error, flags;
2303 
2304 	if ((error = nlookup(nd)) != 0)
2305 		return (error);
2306 retry:
2307 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2308 	if (error)
2309 		return (error);
2310 
2311 	/* Flags == 0 means only check for existence. */
2312 	if (aflags) {
2313 		flags = 0;
2314 		if (aflags & R_OK)
2315 			flags |= VREAD;
2316 		if (aflags & W_OK)
2317 			flags |= VWRITE;
2318 		if (aflags & X_OK)
2319 			flags |= VEXEC;
2320 		if ((flags & VWRITE) == 0 ||
2321 		    (error = vn_writechk(vp, &nd->nl_nch)) == 0)
2322 			error = VOP_ACCESS(vp, flags, nd->nl_cred);
2323 
2324 		/*
2325 		 * If the file handle is stale we have to re-resolve the
2326 		 * entry.  This is a hack at the moment.
2327 		 */
2328 		if (error == ESTALE) {
2329 			vput(vp);
2330 			cache_setunresolved(&nd->nl_nch);
2331 			error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2332 			if (error == 0) {
2333 				vp = NULL;
2334 				goto retry;
2335 			}
2336 			return(error);
2337 		}
2338 	}
2339 	vput(vp);
2340 	return (error);
2341 }
2342 
2343 /*
2344  * access_args(char *path, int flags)
2345  *
2346  * Check access permissions.
2347  */
2348 int
2349 sys_access(struct access_args *uap)
2350 {
2351 	struct nlookupdata nd;
2352 	int error;
2353 
2354 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2355 	if (error == 0)
2356 		error = kern_access(&nd, uap->flags);
2357 	nlookup_done(&nd);
2358 	return (error);
2359 }
2360 
2361 int
2362 kern_stat(struct nlookupdata *nd, struct stat *st)
2363 {
2364 	int error;
2365 	struct vnode *vp;
2366 	thread_t td;
2367 
2368 	if ((error = nlookup(nd)) != 0)
2369 		return (error);
2370 again:
2371 	if ((vp = nd->nl_nch.ncp->nc_vp) == NULL)
2372 		return (ENOENT);
2373 
2374 	td = curthread;
2375 	if ((error = vget(vp, LK_SHARED)) != 0)
2376 		return (error);
2377 	error = vn_stat(vp, st, nd->nl_cred);
2378 
2379 	/*
2380 	 * If the file handle is stale we have to re-resolve the entry.  This
2381 	 * is a hack at the moment.
2382 	 */
2383 	if (error == ESTALE) {
2384 		vput(vp);
2385 		cache_setunresolved(&nd->nl_nch);
2386 		error = cache_resolve(&nd->nl_nch, nd->nl_cred);
2387 		if (error == 0)
2388 			goto again;
2389 	} else {
2390 		vput(vp);
2391 	}
2392 	return (error);
2393 }
2394 
2395 /*
2396  * stat_args(char *path, struct stat *ub)
2397  *
2398  * Get file status; this version follows links.
2399  */
2400 int
2401 sys_stat(struct stat_args *uap)
2402 {
2403 	struct nlookupdata nd;
2404 	struct stat st;
2405 	int error;
2406 
2407 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2408 	if (error == 0) {
2409 		error = kern_stat(&nd, &st);
2410 		if (error == 0)
2411 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2412 	}
2413 	nlookup_done(&nd);
2414 	return (error);
2415 }
2416 
2417 /*
2418  * lstat_args(char *path, struct stat *ub)
2419  *
2420  * Get file status; this version does not follow links.
2421  */
2422 int
2423 sys_lstat(struct lstat_args *uap)
2424 {
2425 	struct nlookupdata nd;
2426 	struct stat st;
2427 	int error;
2428 
2429 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2430 	if (error == 0) {
2431 		error = kern_stat(&nd, &st);
2432 		if (error == 0)
2433 			error = copyout(&st, uap->ub, sizeof(*uap->ub));
2434 	}
2435 	nlookup_done(&nd);
2436 	return (error);
2437 }
2438 
2439 /*
2440  * pathconf_Args(char *path, int name)
2441  *
2442  * Get configurable pathname variables.
2443  */
2444 /* ARGSUSED */
2445 int
2446 sys_pathconf(struct pathconf_args *uap)
2447 {
2448 	struct nlookupdata nd;
2449 	struct vnode *vp;
2450 	int error;
2451 
2452 	vp = NULL;
2453 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2454 	if (error == 0)
2455 		error = nlookup(&nd);
2456 	if (error == 0)
2457 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
2458 	nlookup_done(&nd);
2459 	if (error == 0) {
2460 		error = VOP_PATHCONF(vp, uap->name, uap->sysmsg_fds);
2461 		vput(vp);
2462 	}
2463 	return (error);
2464 }
2465 
2466 /*
2467  * XXX: daver
2468  * kern_readlink isn't properly split yet.  There is a copyin burried
2469  * in VOP_READLINK().
2470  */
2471 int
2472 kern_readlink(struct nlookupdata *nd, char *buf, int count, int *res)
2473 {
2474 	struct thread *td = curthread;
2475 	struct proc *p = td->td_proc;
2476 	struct vnode *vp;
2477 	struct iovec aiov;
2478 	struct uio auio;
2479 	int error;
2480 
2481 	if ((error = nlookup(nd)) != 0)
2482 		return (error);
2483 	error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &vp);
2484 	if (error)
2485 		return (error);
2486 	if (vp->v_type != VLNK) {
2487 		error = EINVAL;
2488 	} else {
2489 		aiov.iov_base = buf;
2490 		aiov.iov_len = count;
2491 		auio.uio_iov = &aiov;
2492 		auio.uio_iovcnt = 1;
2493 		auio.uio_offset = 0;
2494 		auio.uio_rw = UIO_READ;
2495 		auio.uio_segflg = UIO_USERSPACE;
2496 		auio.uio_td = td;
2497 		auio.uio_resid = count;
2498 		error = VOP_READLINK(vp, &auio, p->p_ucred);
2499 	}
2500 	vput(vp);
2501 	*res = count - auio.uio_resid;
2502 	return (error);
2503 }
2504 
2505 /*
2506  * readlink_args(char *path, char *buf, int count)
2507  *
2508  * Return target name of a symbolic link.
2509  */
2510 int
2511 sys_readlink(struct readlink_args *uap)
2512 {
2513 	struct nlookupdata nd;
2514 	int error;
2515 
2516 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2517 	if (error == 0) {
2518 		error = kern_readlink(&nd, uap->buf, uap->count,
2519 					&uap->sysmsg_result);
2520 	}
2521 	nlookup_done(&nd);
2522 	return (error);
2523 }
2524 
2525 static int
2526 setfflags(struct vnode *vp, int flags)
2527 {
2528 	struct thread *td = curthread;
2529 	struct proc *p = td->td_proc;
2530 	int error;
2531 	struct vattr vattr;
2532 
2533 	/*
2534 	 * Prevent non-root users from setting flags on devices.  When
2535 	 * a device is reused, users can retain ownership of the device
2536 	 * if they are allowed to set flags and programs assume that
2537 	 * chown can't fail when done as root.
2538 	 */
2539 	if ((vp->v_type == VCHR || vp->v_type == VBLK) &&
2540 	    ((error = suser_cred(p->p_ucred, PRISON_ROOT)) != 0))
2541 		return (error);
2542 
2543 	/*
2544 	 * note: vget is required for any operation that might mod the vnode
2545 	 * so VINACTIVE is properly cleared.
2546 	 */
2547 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2548 		VATTR_NULL(&vattr);
2549 		vattr.va_flags = flags;
2550 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2551 		vput(vp);
2552 	}
2553 	return (error);
2554 }
2555 
2556 /*
2557  * chflags(char *path, int flags)
2558  *
2559  * Change flags of a file given a path name.
2560  */
2561 /* ARGSUSED */
2562 int
2563 sys_chflags(struct chflags_args *uap)
2564 {
2565 	struct nlookupdata nd;
2566 	struct vnode *vp;
2567 	int error;
2568 
2569 	vp = NULL;
2570 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2571 	/* XXX Add NLC flag indicating modifying operation? */
2572 	if (error == 0)
2573 		error = nlookup(&nd);
2574 	if (error == 0)
2575 		error = ncp_writechk(&nd.nl_nch);
2576 	if (error == 0)
2577 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2578 	nlookup_done(&nd);
2579 	if (error == 0) {
2580 		error = setfflags(vp, uap->flags);
2581 		vrele(vp);
2582 	}
2583 	return (error);
2584 }
2585 
2586 /*
2587  * lchflags(char *path, int flags)
2588  *
2589  * Change flags of a file given a path name, but don't follow symlinks.
2590  */
2591 /* ARGSUSED */
2592 int
2593 sys_lchflags(struct lchflags_args *uap)
2594 {
2595 	struct nlookupdata nd;
2596 	struct vnode *vp;
2597 	int error;
2598 
2599 	vp = NULL;
2600 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2601 	/* XXX Add NLC flag indicating modifying operation? */
2602 	if (error == 0)
2603 		error = nlookup(&nd);
2604 	if (error == 0)
2605 		error = ncp_writechk(&nd.nl_nch);
2606 	if (error == 0)
2607 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
2608 	nlookup_done(&nd);
2609 	if (error == 0) {
2610 		error = setfflags(vp, uap->flags);
2611 		vrele(vp);
2612 	}
2613 	return (error);
2614 }
2615 
2616 /*
2617  * fchflags_args(int fd, int flags)
2618  *
2619  * Change flags of a file given a file descriptor.
2620  */
2621 /* ARGSUSED */
2622 int
2623 sys_fchflags(struct fchflags_args *uap)
2624 {
2625 	struct thread *td = curthread;
2626 	struct proc *p = td->td_proc;
2627 	struct file *fp;
2628 	int error;
2629 
2630 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2631 		return (error);
2632 	if (fp->f_nchandle.ncp)
2633 		error = ncp_writechk(&fp->f_nchandle);
2634 	if (error == 0)
2635 		error = setfflags((struct vnode *) fp->f_data, uap->flags);
2636 	fdrop(fp);
2637 	return (error);
2638 }
2639 
2640 static int
2641 setfmode(struct vnode *vp, int mode)
2642 {
2643 	struct thread *td = curthread;
2644 	struct proc *p = td->td_proc;
2645 	int error;
2646 	struct vattr vattr;
2647 
2648 	/*
2649 	 * note: vget is required for any operation that might mod the vnode
2650 	 * so VINACTIVE is properly cleared.
2651 	 */
2652 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2653 		VATTR_NULL(&vattr);
2654 		vattr.va_mode = mode & ALLPERMS;
2655 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2656 		vput(vp);
2657 	}
2658 	return error;
2659 }
2660 
2661 int
2662 kern_chmod(struct nlookupdata *nd, int mode)
2663 {
2664 	struct vnode *vp;
2665 	int error;
2666 
2667 	/* XXX Add NLC flag indicating modifying operation? */
2668 	if ((error = nlookup(nd)) != 0)
2669 		return (error);
2670 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2671 		return (error);
2672 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2673 		error = setfmode(vp, mode);
2674 	vrele(vp);
2675 	return (error);
2676 }
2677 
2678 /*
2679  * chmod_args(char *path, int mode)
2680  *
2681  * Change mode of a file given path name.
2682  */
2683 /* ARGSUSED */
2684 int
2685 sys_chmod(struct chmod_args *uap)
2686 {
2687 	struct nlookupdata nd;
2688 	int error;
2689 
2690 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2691 	if (error == 0)
2692 		error = kern_chmod(&nd, uap->mode);
2693 	nlookup_done(&nd);
2694 	return (error);
2695 }
2696 
2697 /*
2698  * lchmod_args(char *path, int mode)
2699  *
2700  * Change mode of a file given path name (don't follow links.)
2701  */
2702 /* ARGSUSED */
2703 int
2704 sys_lchmod(struct lchmod_args *uap)
2705 {
2706 	struct nlookupdata nd;
2707 	int error;
2708 
2709 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2710 	if (error == 0)
2711 		error = kern_chmod(&nd, uap->mode);
2712 	nlookup_done(&nd);
2713 	return (error);
2714 }
2715 
2716 /*
2717  * fchmod_args(int fd, int mode)
2718  *
2719  * Change mode of a file given a file descriptor.
2720  */
2721 /* ARGSUSED */
2722 int
2723 sys_fchmod(struct fchmod_args *uap)
2724 {
2725 	struct thread *td = curthread;
2726 	struct proc *p = td->td_proc;
2727 	struct file *fp;
2728 	int error;
2729 
2730 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2731 		return (error);
2732 	if (fp->f_nchandle.ncp)
2733 		error = ncp_writechk(&fp->f_nchandle);
2734 	if (error == 0)
2735 		error = setfmode((struct vnode *)fp->f_data, uap->mode);
2736 	fdrop(fp);
2737 	return (error);
2738 }
2739 
2740 static int
2741 setfown(struct vnode *vp, uid_t uid, gid_t gid)
2742 {
2743 	struct thread *td = curthread;
2744 	struct proc *p = td->td_proc;
2745 	int error;
2746 	struct vattr vattr;
2747 
2748 	/*
2749 	 * note: vget is required for any operation that might mod the vnode
2750 	 * so VINACTIVE is properly cleared.
2751 	 */
2752 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2753 		VATTR_NULL(&vattr);
2754 		vattr.va_uid = uid;
2755 		vattr.va_gid = gid;
2756 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2757 		vput(vp);
2758 	}
2759 	return error;
2760 }
2761 
2762 int
2763 kern_chown(struct nlookupdata *nd, int uid, int gid)
2764 {
2765 	struct vnode *vp;
2766 	int error;
2767 
2768 	/* XXX Add NLC flag indicating modifying operation? */
2769 	if ((error = nlookup(nd)) != 0)
2770 		return (error);
2771 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2772 		return (error);
2773 	if ((error = ncp_writechk(&nd->nl_nch)) == 0)
2774 		error = setfown(vp, uid, gid);
2775 	vrele(vp);
2776 	return (error);
2777 }
2778 
2779 /*
2780  * chown(char *path, int uid, int gid)
2781  *
2782  * Set ownership given a path name.
2783  */
2784 int
2785 sys_chown(struct chown_args *uap)
2786 {
2787 	struct nlookupdata nd;
2788 	int error;
2789 
2790 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2791 	if (error == 0)
2792 		error = kern_chown(&nd, uap->uid, uap->gid);
2793 	nlookup_done(&nd);
2794 	return (error);
2795 }
2796 
2797 /*
2798  * lchown_args(char *path, int uid, int gid)
2799  *
2800  * Set ownership given a path name, do not cross symlinks.
2801  */
2802 int
2803 sys_lchown(struct lchown_args *uap)
2804 {
2805 	struct nlookupdata nd;
2806 	int error;
2807 
2808 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2809 	if (error == 0)
2810 		error = kern_chown(&nd, uap->uid, uap->gid);
2811 	nlookup_done(&nd);
2812 	return (error);
2813 }
2814 
2815 /*
2816  * fchown_args(int fd, int uid, int gid)
2817  *
2818  * Set ownership given a file descriptor.
2819  */
2820 /* ARGSUSED */
2821 int
2822 sys_fchown(struct fchown_args *uap)
2823 {
2824 	struct thread *td = curthread;
2825 	struct proc *p = td->td_proc;
2826 	struct file *fp;
2827 	int error;
2828 
2829 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
2830 		return (error);
2831 	if (fp->f_nchandle.ncp)
2832 		error = ncp_writechk(&fp->f_nchandle);
2833 	if (error == 0)
2834 		error = setfown((struct vnode *)fp->f_data, uap->uid, uap->gid);
2835 	fdrop(fp);
2836 	return (error);
2837 }
2838 
2839 static int
2840 getutimes(const struct timeval *tvp, struct timespec *tsp)
2841 {
2842 	struct timeval tv[2];
2843 
2844 	if (tvp == NULL) {
2845 		microtime(&tv[0]);
2846 		TIMEVAL_TO_TIMESPEC(&tv[0], &tsp[0]);
2847 		tsp[1] = tsp[0];
2848 	} else {
2849 		TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]);
2850 		TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]);
2851 	}
2852 	return 0;
2853 }
2854 
2855 static int
2856 setutimes(struct vnode *vp, const struct timespec *ts, int nullflag)
2857 {
2858 	struct thread *td = curthread;
2859 	struct proc *p = td->td_proc;
2860 	int error;
2861 	struct vattr vattr;
2862 
2863 	/*
2864 	 * note: vget is required for any operation that might mod the vnode
2865 	 * so VINACTIVE is properly cleared.
2866 	 */
2867 	if ((error = vget(vp, LK_EXCLUSIVE)) == 0) {
2868 		VATTR_NULL(&vattr);
2869 		vattr.va_atime = ts[0];
2870 		vattr.va_mtime = ts[1];
2871 		if (nullflag)
2872 			vattr.va_vaflags |= VA_UTIMES_NULL;
2873 		error = VOP_SETATTR(vp, &vattr, p->p_ucred);
2874 		vput(vp);
2875 	}
2876 	return error;
2877 }
2878 
2879 int
2880 kern_utimes(struct nlookupdata *nd, struct timeval *tptr)
2881 {
2882 	struct timespec ts[2];
2883 	struct vnode *vp;
2884 	int error;
2885 
2886 	if ((error = getutimes(tptr, ts)) != 0)
2887 		return (error);
2888 	/* XXX Add NLC flag indicating modifying operation? */
2889 	if ((error = nlookup(nd)) != 0)
2890 		return (error);
2891 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
2892 		return (error);
2893 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
2894 		return (error);
2895 	error = setutimes(vp, ts, tptr == NULL);
2896 	vrele(vp);
2897 	return (error);
2898 }
2899 
2900 /*
2901  * utimes_args(char *path, struct timeval *tptr)
2902  *
2903  * Set the access and modification times of a file.
2904  */
2905 int
2906 sys_utimes(struct utimes_args *uap)
2907 {
2908 	struct timeval tv[2];
2909 	struct nlookupdata nd;
2910 	int error;
2911 
2912 	if (uap->tptr) {
2913  		error = copyin(uap->tptr, tv, sizeof(tv));
2914 		if (error)
2915 			return (error);
2916 	}
2917 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
2918 	if (error == 0)
2919 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2920 	nlookup_done(&nd);
2921 	return (error);
2922 }
2923 
2924 /*
2925  * lutimes_args(char *path, struct timeval *tptr)
2926  *
2927  * Set the access and modification times of a file.
2928  */
2929 int
2930 sys_lutimes(struct lutimes_args *uap)
2931 {
2932 	struct timeval tv[2];
2933 	struct nlookupdata nd;
2934 	int error;
2935 
2936 	if (uap->tptr) {
2937 		error = copyin(uap->tptr, tv, sizeof(tv));
2938 		if (error)
2939 			return (error);
2940 	}
2941 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
2942 	if (error == 0)
2943 		error = kern_utimes(&nd, uap->tptr ? tv : NULL);
2944 	nlookup_done(&nd);
2945 	return (error);
2946 }
2947 
2948 int
2949 kern_futimes(int fd, struct timeval *tptr)
2950 {
2951 	struct thread *td = curthread;
2952 	struct proc *p = td->td_proc;
2953 	struct timespec ts[2];
2954 	struct file *fp;
2955 	int error;
2956 
2957 	error = getutimes(tptr, ts);
2958 	if (error)
2959 		return (error);
2960 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
2961 		return (error);
2962 	if (fp->f_nchandle.ncp)
2963 		error = ncp_writechk(&fp->f_nchandle);
2964 	if (error == 0)
2965 		error =  setutimes((struct vnode *)fp->f_data, ts, tptr == NULL);
2966 	fdrop(fp);
2967 	return (error);
2968 }
2969 
2970 /*
2971  * futimes_args(int fd, struct timeval *tptr)
2972  *
2973  * Set the access and modification times of a file.
2974  */
2975 int
2976 sys_futimes(struct futimes_args *uap)
2977 {
2978 	struct timeval tv[2];
2979 	int error;
2980 
2981 	if (uap->tptr) {
2982 		error = copyin(uap->tptr, tv, sizeof(tv));
2983 		if (error)
2984 			return (error);
2985 	}
2986 
2987 	error = kern_futimes(uap->fd, uap->tptr ? tv : NULL);
2988 
2989 	return (error);
2990 }
2991 
2992 int
2993 kern_truncate(struct nlookupdata *nd, off_t length)
2994 {
2995 	struct vnode *vp;
2996 	struct vattr vattr;
2997 	int error;
2998 
2999 	if (length < 0)
3000 		return(EINVAL);
3001 	/* XXX Add NLC flag indicating modifying operation? */
3002 	if ((error = nlookup(nd)) != 0)
3003 		return (error);
3004 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3005 		return (error);
3006 	if ((error = cache_vref(&nd->nl_nch, nd->nl_cred, &vp)) != 0)
3007 		return (error);
3008 	if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY)) != 0) {
3009 		vrele(vp);
3010 		return (error);
3011 	}
3012 	if (vp->v_type == VDIR) {
3013 		error = EISDIR;
3014 	} else if ((error = vn_writechk(vp, &nd->nl_nch)) == 0 &&
3015 	    (error = VOP_ACCESS(vp, VWRITE, nd->nl_cred)) == 0) {
3016 		VATTR_NULL(&vattr);
3017 		vattr.va_size = length;
3018 		error = VOP_SETATTR(vp, &vattr, nd->nl_cred);
3019 	}
3020 	vput(vp);
3021 	return (error);
3022 }
3023 
3024 /*
3025  * truncate(char *path, int pad, off_t length)
3026  *
3027  * Truncate a file given its path name.
3028  */
3029 int
3030 sys_truncate(struct truncate_args *uap)
3031 {
3032 	struct nlookupdata nd;
3033 	int error;
3034 
3035 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3036 	if (error == 0)
3037 		error = kern_truncate(&nd, uap->length);
3038 	nlookup_done(&nd);
3039 	return error;
3040 }
3041 
3042 int
3043 kern_ftruncate(int fd, off_t length)
3044 {
3045 	struct thread *td = curthread;
3046 	struct proc *p = td->td_proc;
3047 	struct vattr vattr;
3048 	struct vnode *vp;
3049 	struct file *fp;
3050 	int error;
3051 
3052 	if (length < 0)
3053 		return(EINVAL);
3054 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3055 		return (error);
3056 	if (fp->f_nchandle.ncp) {
3057 		error = ncp_writechk(&fp->f_nchandle);
3058 		if (error)
3059 			goto done;
3060 	}
3061 	if ((fp->f_flag & FWRITE) == 0) {
3062 		error = EINVAL;
3063 		goto done;
3064 	}
3065 	vp = (struct vnode *)fp->f_data;
3066 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3067 	if (vp->v_type == VDIR) {
3068 		error = EISDIR;
3069 	} else if ((error = vn_writechk(vp, NULL)) == 0) {
3070 		VATTR_NULL(&vattr);
3071 		vattr.va_size = length;
3072 		error = VOP_SETATTR(vp, &vattr, fp->f_cred);
3073 	}
3074 	vn_unlock(vp);
3075 done:
3076 	fdrop(fp);
3077 	return (error);
3078 }
3079 
3080 /*
3081  * ftruncate_args(int fd, int pad, off_t length)
3082  *
3083  * Truncate a file given a file descriptor.
3084  */
3085 int
3086 sys_ftruncate(struct ftruncate_args *uap)
3087 {
3088 	int error;
3089 
3090 	error = kern_ftruncate(uap->fd, uap->length);
3091 
3092 	return (error);
3093 }
3094 
3095 /*
3096  * fsync(int fd)
3097  *
3098  * Sync an open file.
3099  */
3100 /* ARGSUSED */
3101 int
3102 sys_fsync(struct fsync_args *uap)
3103 {
3104 	struct thread *td = curthread;
3105 	struct proc *p = td->td_proc;
3106 	struct vnode *vp;
3107 	struct file *fp;
3108 	vm_object_t obj;
3109 	int error;
3110 
3111 	if ((error = holdvnode(p->p_fd, uap->fd, &fp)) != 0)
3112 		return (error);
3113 	vp = (struct vnode *)fp->f_data;
3114 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3115 	if ((obj = vp->v_object) != NULL)
3116 		vm_object_page_clean(obj, 0, 0, 0);
3117 	if ((error = VOP_FSYNC(vp, MNT_WAIT)) == 0 && vp->v_mount)
3118 		error = buf_fsync(vp);
3119 	vn_unlock(vp);
3120 	fdrop(fp);
3121 	return (error);
3122 }
3123 
3124 int
3125 kern_rename(struct nlookupdata *fromnd, struct nlookupdata *tond)
3126 {
3127 	struct nchandle fnchd;
3128 	struct nchandle tnchd;
3129 	struct namecache *ncp;
3130 	struct vnode *fdvp;
3131 	struct vnode *tdvp;
3132 	struct mount *mp;
3133 	int error;
3134 
3135 	bwillinode(1);
3136 	fromnd->nl_flags |= NLC_REFDVP;
3137 	if ((error = nlookup(fromnd)) != 0)
3138 		return (error);
3139 	if ((fnchd.ncp = fromnd->nl_nch.ncp->nc_parent) == NULL)
3140 		return (ENOENT);
3141 	fnchd.mount = fromnd->nl_nch.mount;
3142 	cache_hold(&fnchd);
3143 
3144 	/*
3145 	 * unlock the source nch so we can lookup the target nch without
3146 	 * deadlocking.  The target may or may not exist so we do not check
3147 	 * for a target vp like kern_mkdir() and other creation functions do.
3148 	 *
3149 	 * The source and target directories are ref'd and rechecked after
3150 	 * everything is relocked to determine if the source or target file
3151 	 * has been renamed.
3152 	 */
3153 	KKASSERT(fromnd->nl_flags & NLC_NCPISLOCKED);
3154 	fromnd->nl_flags &= ~NLC_NCPISLOCKED;
3155 	cache_unlock(&fromnd->nl_nch);
3156 
3157 	tond->nl_flags |= NLC_CREATE | NLC_REFDVP;
3158 	if ((error = nlookup(tond)) != 0) {
3159 		cache_drop(&fnchd);
3160 		return (error);
3161 	}
3162 	if ((tnchd.ncp = tond->nl_nch.ncp->nc_parent) == NULL) {
3163 		cache_drop(&fnchd);
3164 		return (ENOENT);
3165 	}
3166 	tnchd.mount = tond->nl_nch.mount;
3167 	cache_hold(&tnchd);
3168 
3169 	/*
3170 	 * If the source and target are the same there is nothing to do
3171 	 */
3172 	if (fromnd->nl_nch.ncp == tond->nl_nch.ncp) {
3173 		cache_drop(&fnchd);
3174 		cache_drop(&tnchd);
3175 		return (0);
3176 	}
3177 
3178 	/*
3179 	 * Mount points cannot be renamed or overwritten
3180 	 */
3181 	if ((fromnd->nl_nch.ncp->nc_flag | tond->nl_nch.ncp->nc_flag) &
3182 	    NCF_ISMOUNTPT
3183 	) {
3184 		cache_drop(&fnchd);
3185 		cache_drop(&tnchd);
3186 		return (EINVAL);
3187 	}
3188 
3189 	/*
3190 	 * relock the source ncp.  NOTE AFTER RELOCKING: the source ncp
3191 	 * may have become invalid while it was unlocked, nc_vp and nc_mount
3192 	 * could be NULL.
3193 	 */
3194 	if (cache_lock_nonblock(&fromnd->nl_nch) == 0) {
3195 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3196 	} else if (fromnd->nl_nch.ncp > tond->nl_nch.ncp) {
3197 		cache_lock(&fromnd->nl_nch);
3198 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3199 	} else {
3200 		cache_unlock(&tond->nl_nch);
3201 		cache_lock(&fromnd->nl_nch);
3202 		cache_resolve(&fromnd->nl_nch, fromnd->nl_cred);
3203 		cache_lock(&tond->nl_nch);
3204 		cache_resolve(&tond->nl_nch, tond->nl_cred);
3205 	}
3206 	fromnd->nl_flags |= NLC_NCPISLOCKED;
3207 
3208 	/*
3209 	 * make sure the parent directories linkages are the same
3210 	 */
3211 	if (fnchd.ncp != fromnd->nl_nch.ncp->nc_parent ||
3212 	    tnchd.ncp != tond->nl_nch.ncp->nc_parent) {
3213 		cache_drop(&fnchd);
3214 		cache_drop(&tnchd);
3215 		return (ENOENT);
3216 	}
3217 
3218 	/*
3219 	 * Both the source and target must be within the same filesystem and
3220 	 * in the same filesystem as their parent directories within the
3221 	 * namecache topology.
3222 	 *
3223 	 * NOTE: fromnd's nc_mount or nc_vp could be NULL.
3224 	 */
3225 	mp = fnchd.mount;
3226 	if (mp != tnchd.mount || mp != fromnd->nl_nch.mount ||
3227 	    mp != tond->nl_nch.mount) {
3228 		cache_drop(&fnchd);
3229 		cache_drop(&tnchd);
3230 		return (EXDEV);
3231 	}
3232 
3233 	/*
3234 	 * Make sure the mount point is writable
3235 	 */
3236 	if ((error = ncp_writechk(&tond->nl_nch)) != 0) {
3237 		cache_drop(&fnchd);
3238 		cache_drop(&tnchd);
3239 		return (error);
3240 	}
3241 
3242 	/*
3243 	 * If the target exists and either the source or target is a directory,
3244 	 * then both must be directories.
3245 	 *
3246 	 * Due to relocking of the source, fromnd->nl_nch.ncp->nc_vp might h
3247 	 * have become NULL.
3248 	 */
3249 	if (tond->nl_nch.ncp->nc_vp) {
3250 		if (fromnd->nl_nch.ncp->nc_vp == NULL) {
3251 			error = ENOENT;
3252 		} else if (fromnd->nl_nch.ncp->nc_vp->v_type == VDIR) {
3253 			if (tond->nl_nch.ncp->nc_vp->v_type != VDIR)
3254 				error = ENOTDIR;
3255 		} else if (tond->nl_nch.ncp->nc_vp->v_type == VDIR) {
3256 			error = EISDIR;
3257 		}
3258 	}
3259 
3260 	/*
3261 	 * You cannot rename a source into itself or a subdirectory of itself.
3262 	 * We check this by travsersing the target directory upwards looking
3263 	 * for a match against the source.
3264 	 */
3265 	if (error == 0) {
3266 		for (ncp = tnchd.ncp; ncp; ncp = ncp->nc_parent) {
3267 			if (fromnd->nl_nch.ncp == ncp) {
3268 				error = EINVAL;
3269 				break;
3270 			}
3271 		}
3272 	}
3273 
3274 	cache_drop(&fnchd);
3275 	cache_drop(&tnchd);
3276 
3277 	/*
3278 	 * Even though the namespaces are different, they may still represent
3279 	 * hardlinks to the same file.  The filesystem might have a hard time
3280 	 * with this so we issue a NREMOVE of the source instead of a NRENAME
3281 	 * when we detect the situation.
3282 	 */
3283 	if (error == 0) {
3284 		fdvp = fromnd->nl_dvp;
3285 		tdvp = tond->nl_dvp;
3286 		if (fdvp == NULL || tdvp == NULL) {
3287 			error = EPERM;
3288 		} else if (fromnd->nl_nch.ncp->nc_vp == tond->nl_nch.ncp->nc_vp) {
3289 			error = VOP_NREMOVE(&fromnd->nl_nch, fdvp,
3290 					    fromnd->nl_cred);
3291 		} else {
3292 			error = VOP_NRENAME(&fromnd->nl_nch, &tond->nl_nch,
3293 					    fdvp, tdvp, tond->nl_cred);
3294 		}
3295 	}
3296 	return (error);
3297 }
3298 
3299 /*
3300  * rename_args(char *from, char *to)
3301  *
3302  * Rename files.  Source and destination must either both be directories,
3303  * or both not be directories.  If target is a directory, it must be empty.
3304  */
3305 int
3306 sys_rename(struct rename_args *uap)
3307 {
3308 	struct nlookupdata fromnd, tond;
3309 	int error;
3310 
3311 	error = nlookup_init(&fromnd, uap->from, UIO_USERSPACE, 0);
3312 	if (error == 0) {
3313 		error = nlookup_init(&tond, uap->to, UIO_USERSPACE, 0);
3314 		if (error == 0)
3315 			error = kern_rename(&fromnd, &tond);
3316 		nlookup_done(&tond);
3317 	}
3318 	nlookup_done(&fromnd);
3319 	return (error);
3320 }
3321 
3322 int
3323 kern_mkdir(struct nlookupdata *nd, int mode)
3324 {
3325 	struct thread *td = curthread;
3326 	struct proc *p = td->td_proc;
3327 	struct vnode *vp;
3328 	struct vattr vattr;
3329 	int error;
3330 
3331 	bwillinode(1);
3332 	nd->nl_flags |= NLC_WILLBEDIR | NLC_CREATE | NLC_REFDVP;
3333 	if ((error = nlookup(nd)) != 0)
3334 		return (error);
3335 
3336 	if (nd->nl_nch.ncp->nc_vp)
3337 		return (EEXIST);
3338 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3339 		return (error);
3340 	VATTR_NULL(&vattr);
3341 	vattr.va_type = VDIR;
3342 	vattr.va_mode = (mode & ACCESSPERMS) &~ p->p_fd->fd_cmask;
3343 
3344 	vp = NULL;
3345 	error = VOP_NMKDIR(&nd->nl_nch, nd->nl_dvp, &vp, p->p_ucred, &vattr);
3346 	if (error == 0)
3347 		vput(vp);
3348 	return (error);
3349 }
3350 
3351 /*
3352  * mkdir_args(char *path, int mode)
3353  *
3354  * Make a directory file.
3355  */
3356 /* ARGSUSED */
3357 int
3358 sys_mkdir(struct mkdir_args *uap)
3359 {
3360 	struct nlookupdata nd;
3361 	int error;
3362 
3363 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3364 	if (error == 0)
3365 		error = kern_mkdir(&nd, uap->mode);
3366 	nlookup_done(&nd);
3367 	return (error);
3368 }
3369 
3370 int
3371 kern_rmdir(struct nlookupdata *nd)
3372 {
3373 	int error;
3374 
3375 	bwillinode(1);
3376 	nd->nl_flags |= NLC_DELETE | NLC_REFDVP;
3377 	if ((error = nlookup(nd)) != 0)
3378 		return (error);
3379 
3380 	/*
3381 	 * Do not allow directories representing mount points to be
3382 	 * deleted, even if empty.  Check write perms on mount point
3383 	 * in case the vnode is aliased (aka nullfs).
3384 	 */
3385 	if (nd->nl_nch.ncp->nc_flag & (NCF_ISMOUNTPT))
3386 		return (EINVAL);
3387 	if ((error = ncp_writechk(&nd->nl_nch)) != 0)
3388 		return (error);
3389 	error = VOP_NRMDIR(&nd->nl_nch, nd->nl_dvp, nd->nl_cred);
3390 	return (error);
3391 }
3392 
3393 /*
3394  * rmdir_args(char *path)
3395  *
3396  * Remove a directory file.
3397  */
3398 /* ARGSUSED */
3399 int
3400 sys_rmdir(struct rmdir_args *uap)
3401 {
3402 	struct nlookupdata nd;
3403 	int error;
3404 
3405 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, 0);
3406 	if (error == 0)
3407 		error = kern_rmdir(&nd);
3408 	nlookup_done(&nd);
3409 	return (error);
3410 }
3411 
3412 int
3413 kern_getdirentries(int fd, char *buf, u_int count, long *basep, int *res,
3414     enum uio_seg direction)
3415 {
3416 	struct thread *td = curthread;
3417 	struct proc *p = td->td_proc;
3418 	struct vnode *vp;
3419 	struct file *fp;
3420 	struct uio auio;
3421 	struct iovec aiov;
3422 	off_t loff;
3423 	int error, eofflag;
3424 
3425 	if ((error = holdvnode(p->p_fd, fd, &fp)) != 0)
3426 		return (error);
3427 	if ((fp->f_flag & FREAD) == 0) {
3428 		error = EBADF;
3429 		goto done;
3430 	}
3431 	vp = (struct vnode *)fp->f_data;
3432 unionread:
3433 	if (vp->v_type != VDIR) {
3434 		error = EINVAL;
3435 		goto done;
3436 	}
3437 	aiov.iov_base = buf;
3438 	aiov.iov_len = count;
3439 	auio.uio_iov = &aiov;
3440 	auio.uio_iovcnt = 1;
3441 	auio.uio_rw = UIO_READ;
3442 	auio.uio_segflg = direction;
3443 	auio.uio_td = td;
3444 	auio.uio_resid = count;
3445 	loff = auio.uio_offset = fp->f_offset;
3446 	error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL);
3447 	fp->f_offset = auio.uio_offset;
3448 	if (error)
3449 		goto done;
3450 	if (count == auio.uio_resid) {
3451 		if (union_dircheckp) {
3452 			error = union_dircheckp(td, &vp, fp);
3453 			if (error == -1)
3454 				goto unionread;
3455 			if (error)
3456 				goto done;
3457 		}
3458 #if 0
3459 		if ((vp->v_flag & VROOT) &&
3460 		    (vp->v_mount->mnt_flag & MNT_UNION)) {
3461 			struct vnode *tvp = vp;
3462 			vp = vp->v_mount->mnt_vnodecovered;
3463 			vref(vp);
3464 			fp->f_data = vp;
3465 			fp->f_offset = 0;
3466 			vrele(tvp);
3467 			goto unionread;
3468 		}
3469 #endif
3470 	}
3471 
3472 	/*
3473 	 * WARNING!  *basep may not be wide enough to accomodate the
3474 	 * seek offset.   XXX should we hack this to return the upper 32 bits
3475 	 * for offsets greater then 4G?
3476 	 */
3477 	if (basep) {
3478 		*basep = (long)loff;
3479 	}
3480 	*res = count - auio.uio_resid;
3481 done:
3482 	fdrop(fp);
3483 	return (error);
3484 }
3485 
3486 /*
3487  * getdirentries_args(int fd, char *buf, u_int conut, long *basep)
3488  *
3489  * Read a block of directory entries in a file system independent format.
3490  */
3491 int
3492 sys_getdirentries(struct getdirentries_args *uap)
3493 {
3494 	long base;
3495 	int error;
3496 
3497 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, &base,
3498 				   &uap->sysmsg_result, UIO_USERSPACE);
3499 
3500 	if (error == 0 && uap->basep)
3501 		error = copyout(&base, uap->basep, sizeof(*uap->basep));
3502 	return (error);
3503 }
3504 
3505 /*
3506  * getdents_args(int fd, char *buf, size_t count)
3507  */
3508 int
3509 sys_getdents(struct getdents_args *uap)
3510 {
3511 	int error;
3512 
3513 	error = kern_getdirentries(uap->fd, uap->buf, uap->count, NULL,
3514 	    &uap->sysmsg_result, UIO_USERSPACE);
3515 
3516 	return (error);
3517 }
3518 
3519 /*
3520  * umask(int newmask)
3521  *
3522  * Set the mode mask for creation of filesystem nodes.
3523  *
3524  * MP SAFE
3525  */
3526 int
3527 sys_umask(struct umask_args *uap)
3528 {
3529 	struct thread *td = curthread;
3530 	struct proc *p = td->td_proc;
3531 	struct filedesc *fdp;
3532 
3533 	fdp = p->p_fd;
3534 	uap->sysmsg_result = fdp->fd_cmask;
3535 	fdp->fd_cmask = uap->newmask & ALLPERMS;
3536 	return (0);
3537 }
3538 
3539 /*
3540  * revoke(char *path)
3541  *
3542  * Void all references to file by ripping underlying filesystem
3543  * away from vnode.
3544  */
3545 /* ARGSUSED */
3546 int
3547 sys_revoke(struct revoke_args *uap)
3548 {
3549 	struct nlookupdata nd;
3550 	struct vattr vattr;
3551 	struct vnode *vp;
3552 	struct ucred *cred;
3553 	int error;
3554 
3555 	vp = NULL;
3556 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3557 	if (error == 0)
3558 		error = nlookup(&nd);
3559 	if (error == 0)
3560 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &vp);
3561 	cred = crhold(nd.nl_cred);
3562 	nlookup_done(&nd);
3563 	if (error == 0) {
3564 		if (vp->v_type != VCHR && vp->v_type != VBLK)
3565 			error = EINVAL;
3566 		if (error == 0)
3567 			error = VOP_GETATTR(vp, &vattr);
3568 		if (error == 0 && cred->cr_uid != vattr.va_uid)
3569 			error = suser_cred(cred, PRISON_ROOT);
3570 		if (error == 0 && count_udev(vp->v_umajor, vp->v_uminor) > 0) {
3571 			error = 0;
3572 			vx_lock(vp);
3573 			VOP_REVOKE(vp, REVOKEALL);
3574 			vx_unlock(vp);
3575 		}
3576 		vrele(vp);
3577 	}
3578 	if (cred)
3579 		crfree(cred);
3580 	return (error);
3581 }
3582 
3583 /*
3584  * getfh_args(char *fname, fhandle_t *fhp)
3585  *
3586  * Get (NFS) file handle
3587  *
3588  * NOTE: We use the fsid of the covering mount, even if it is a nullfs
3589  * mount.  This allows nullfs mounts to be explicitly exported.
3590  *
3591  * WARNING: nullfs mounts of HAMMER PFS ROOTs are safe.
3592  *
3593  * 	    nullfs mounts of subdirectories are not safe.  That is, it will
3594  *	    work, but you do not really have protection against access to
3595  *	    the related parent directories.
3596  */
3597 int
3598 sys_getfh(struct getfh_args *uap)
3599 {
3600 	struct thread *td = curthread;
3601 	struct nlookupdata nd;
3602 	fhandle_t fh;
3603 	struct vnode *vp;
3604 	struct mount *mp;
3605 	int error;
3606 
3607 	/*
3608 	 * Must be super user
3609 	 */
3610 	if ((error = suser(td)) != 0)
3611 		return (error);
3612 
3613 	vp = NULL;
3614 	error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW);
3615 	if (error == 0)
3616 		error = nlookup(&nd);
3617 	if (error == 0)
3618 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
3619 	mp = nd.nl_nch.mount;
3620 	nlookup_done(&nd);
3621 	if (error == 0) {
3622 		bzero(&fh, sizeof(fh));
3623 		fh.fh_fsid = mp->mnt_stat.f_fsid;
3624 		error = VFS_VPTOFH(vp, &fh.fh_fid);
3625 		vput(vp);
3626 		if (error == 0)
3627 			error = copyout(&fh, uap->fhp, sizeof(fh));
3628 	}
3629 	return (error);
3630 }
3631 
3632 /*
3633  * fhopen_args(const struct fhandle *u_fhp, int flags)
3634  *
3635  * syscall for the rpc.lockd to use to translate a NFS file handle into
3636  * an open descriptor.
3637  *
3638  * warning: do not remove the suser() call or this becomes one giant
3639  * security hole.
3640  */
3641 int
3642 sys_fhopen(struct fhopen_args *uap)
3643 {
3644 	struct thread *td = curthread;
3645 	struct proc *p = td->td_proc;
3646 	struct mount *mp;
3647 	struct vnode *vp;
3648 	struct fhandle fhp;
3649 	struct vattr vat;
3650 	struct vattr *vap = &vat;
3651 	struct flock lf;
3652 	int fmode, mode, error, type;
3653 	struct file *nfp;
3654 	struct file *fp;
3655 	int indx;
3656 
3657 	/*
3658 	 * Must be super user
3659 	 */
3660 	error = suser(td);
3661 	if (error)
3662 		return (error);
3663 
3664 	fmode = FFLAGS(uap->flags);
3665 	/* why not allow a non-read/write open for our lockd? */
3666 	if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
3667 		return (EINVAL);
3668 	error = copyin(uap->u_fhp, &fhp, sizeof(fhp));
3669 	if (error)
3670 		return(error);
3671 	/* find the mount point */
3672 	mp = vfs_getvfs(&fhp.fh_fsid);
3673 	if (mp == NULL)
3674 		return (ESTALE);
3675 	/* now give me my vnode, it gets returned to me locked */
3676 	error = VFS_FHTOVP(mp, NULL, &fhp.fh_fid, &vp);
3677 	if (error)
3678 		return (error);
3679  	/*
3680 	 * from now on we have to make sure not
3681 	 * to forget about the vnode
3682 	 * any error that causes an abort must vput(vp)
3683 	 * just set error = err and 'goto bad;'.
3684 	 */
3685 
3686 	/*
3687 	 * from vn_open
3688 	 */
3689 	if (vp->v_type == VLNK) {
3690 		error = EMLINK;
3691 		goto bad;
3692 	}
3693 	if (vp->v_type == VSOCK) {
3694 		error = EOPNOTSUPP;
3695 		goto bad;
3696 	}
3697 	mode = 0;
3698 	if (fmode & (FWRITE | O_TRUNC)) {
3699 		if (vp->v_type == VDIR) {
3700 			error = EISDIR;
3701 			goto bad;
3702 		}
3703 		error = vn_writechk(vp, NULL);
3704 		if (error)
3705 			goto bad;
3706 		mode |= VWRITE;
3707 	}
3708 	if (fmode & FREAD)
3709 		mode |= VREAD;
3710 	if (mode) {
3711 		error = VOP_ACCESS(vp, mode, p->p_ucred);
3712 		if (error)
3713 			goto bad;
3714 	}
3715 	if (fmode & O_TRUNC) {
3716 		vn_unlock(vp);				/* XXX */
3717 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);	/* XXX */
3718 		VATTR_NULL(vap);
3719 		vap->va_size = 0;
3720 		error = VOP_SETATTR(vp, vap, p->p_ucred);
3721 		if (error)
3722 			goto bad;
3723 	}
3724 
3725 	/*
3726 	 * VOP_OPEN needs the file pointer so it can potentially override
3727 	 * it.
3728 	 *
3729 	 * WARNING! no f_nchandle will be associated when fhopen()ing a
3730 	 * directory.  XXX
3731 	 */
3732 	if ((error = falloc(p, &nfp, &indx)) != 0)
3733 		goto bad;
3734 	fp = nfp;
3735 
3736 	error = VOP_OPEN(vp, fmode, p->p_ucred, fp);
3737 	if (error) {
3738 		/*
3739 		 * setting f_ops this way prevents VOP_CLOSE from being
3740 		 * called or fdrop() releasing the vp from v_data.   Since
3741 		 * the VOP_OPEN failed we don't want to VOP_CLOSE.
3742 		 */
3743 		fp->f_ops = &badfileops;
3744 		fp->f_data = NULL;
3745 		goto bad_drop;
3746 	}
3747 
3748 	/*
3749 	 * The fp is given its own reference, we still have our ref and lock.
3750 	 *
3751 	 * Assert that all regular files must be created with a VM object.
3752 	 */
3753 	if (vp->v_type == VREG && vp->v_object == NULL) {
3754 		kprintf("fhopen: regular file did not have VM object: %p\n", vp);
3755 		goto bad_drop;
3756 	}
3757 
3758 	/*
3759 	 * The open was successful.  Handle any locking requirements.
3760 	 */
3761 	if (fmode & (O_EXLOCK | O_SHLOCK)) {
3762 		lf.l_whence = SEEK_SET;
3763 		lf.l_start = 0;
3764 		lf.l_len = 0;
3765 		if (fmode & O_EXLOCK)
3766 			lf.l_type = F_WRLCK;
3767 		else
3768 			lf.l_type = F_RDLCK;
3769 		if (fmode & FNONBLOCK)
3770 			type = 0;
3771 		else
3772 			type = F_WAIT;
3773 		vn_unlock(vp);
3774 		if ((error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type)) != 0) {
3775 			/*
3776 			 * release our private reference.
3777 			 */
3778 			fsetfd(p, NULL, indx);
3779 			fdrop(fp);
3780 			vrele(vp);
3781 			return (error);
3782 		}
3783 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3784 		fp->f_flag |= FHASLOCK;
3785 	}
3786 
3787 	/*
3788 	 * Clean up.  Associate the file pointer with the previously
3789 	 * reserved descriptor and return it.
3790 	 */
3791 	vput(vp);
3792 	fsetfd(p, fp, indx);
3793 	fdrop(fp);
3794 	uap->sysmsg_result = indx;
3795 	return (0);
3796 
3797 bad_drop:
3798 	fsetfd(p, NULL, indx);
3799 	fdrop(fp);
3800 bad:
3801 	vput(vp);
3802 	return (error);
3803 }
3804 
3805 /*
3806  * fhstat_args(struct fhandle *u_fhp, struct stat *sb)
3807  */
3808 int
3809 sys_fhstat(struct fhstat_args *uap)
3810 {
3811 	struct thread *td = curthread;
3812 	struct stat sb;
3813 	fhandle_t fh;
3814 	struct mount *mp;
3815 	struct vnode *vp;
3816 	int error;
3817 
3818 	/*
3819 	 * Must be super user
3820 	 */
3821 	error = suser(td);
3822 	if (error)
3823 		return (error);
3824 
3825 	error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t));
3826 	if (error)
3827 		return (error);
3828 
3829 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3830 		return (ESTALE);
3831 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
3832 		return (error);
3833 	error = vn_stat(vp, &sb, td->td_proc->p_ucred);
3834 	vput(vp);
3835 	if (error)
3836 		return (error);
3837 	error = copyout(&sb, uap->sb, sizeof(sb));
3838 	return (error);
3839 }
3840 
3841 /*
3842  * fhstatfs_args(struct fhandle *u_fhp, struct statfs *buf)
3843  */
3844 int
3845 sys_fhstatfs(struct fhstatfs_args *uap)
3846 {
3847 	struct thread *td = curthread;
3848 	struct proc *p = td->td_proc;
3849 	struct statfs *sp;
3850 	struct mount *mp;
3851 	struct vnode *vp;
3852 	struct statfs sb;
3853 	char *fullpath, *freepath;
3854 	fhandle_t fh;
3855 	int error;
3856 
3857 	/*
3858 	 * Must be super user
3859 	 */
3860 	if ((error = suser(td)))
3861 		return (error);
3862 
3863 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3864 		return (error);
3865 
3866 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3867 		return (ESTALE);
3868 
3869 	if (p != NULL && !chroot_visible_mnt(mp, p))
3870 		return (ESTALE);
3871 
3872 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
3873 		return (error);
3874 	mp = vp->v_mount;
3875 	sp = &mp->mnt_stat;
3876 	vput(vp);
3877 	if ((error = VFS_STATFS(mp, sp, p->p_ucred)) != 0)
3878 		return (error);
3879 
3880 	error = mount_path(p, mp, &fullpath, &freepath);
3881 	if (error)
3882 		return(error);
3883 	bzero(sp->f_mntonname, sizeof(sp->f_mntonname));
3884 	strlcpy(sp->f_mntonname, fullpath, sizeof(sp->f_mntonname));
3885 	kfree(freepath, M_TEMP);
3886 
3887 	sp->f_flags = mp->mnt_flag & MNT_VISFLAGMASK;
3888 	if (suser(td)) {
3889 		bcopy(sp, &sb, sizeof(sb));
3890 		sb.f_fsid.val[0] = sb.f_fsid.val[1] = 0;
3891 		sp = &sb;
3892 	}
3893 	return (copyout(sp, uap->buf, sizeof(*sp)));
3894 }
3895 
3896 /*
3897  * fhstatvfs_args(struct fhandle *u_fhp, struct statvfs *buf)
3898  */
3899 int
3900 sys_fhstatvfs(struct fhstatvfs_args *uap)
3901 {
3902 	struct thread *td = curthread;
3903 	struct proc *p = td->td_proc;
3904 	struct statvfs *sp;
3905 	struct mount *mp;
3906 	struct vnode *vp;
3907 	fhandle_t fh;
3908 	int error;
3909 
3910 	/*
3911 	 * Must be super user
3912 	 */
3913 	if ((error = suser(td)))
3914 		return (error);
3915 
3916 	if ((error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t))) != 0)
3917 		return (error);
3918 
3919 	if ((mp = vfs_getvfs(&fh.fh_fsid)) == NULL)
3920 		return (ESTALE);
3921 
3922 	if (p != NULL && !chroot_visible_mnt(mp, p))
3923 		return (ESTALE);
3924 
3925 	if ((error = VFS_FHTOVP(mp, NULL, &fh.fh_fid, &vp)))
3926 		return (error);
3927 	mp = vp->v_mount;
3928 	sp = &mp->mnt_vstat;
3929 	vput(vp);
3930 	if ((error = VFS_STATVFS(mp, sp, p->p_ucred)) != 0)
3931 		return (error);
3932 
3933 	sp->f_flag = 0;
3934 	if (mp->mnt_flag & MNT_RDONLY)
3935 		sp->f_flag |= ST_RDONLY;
3936 	if (mp->mnt_flag & MNT_NOSUID)
3937 		sp->f_flag |= ST_NOSUID;
3938 
3939 	return (copyout(sp, uap->buf, sizeof(*sp)));
3940 }
3941 
3942 
3943 /*
3944  * Syscall to push extended attribute configuration information into the
3945  * VFS.  Accepts a path, which it converts to a mountpoint, as well as
3946  * a command (int cmd), and attribute name and misc data.  For now, the
3947  * attribute name is left in userspace for consumption by the VFS_op.
3948  * It will probably be changed to be copied into sysspace by the
3949  * syscall in the future, once issues with various consumers of the
3950  * attribute code have raised their hands.
3951  *
3952  * Currently this is used only by UFS Extended Attributes.
3953  */
3954 int
3955 sys_extattrctl(struct extattrctl_args *uap)
3956 {
3957 	struct nlookupdata nd;
3958 	struct mount *mp;
3959 	struct vnode *vp;
3960 	int error;
3961 
3962 	vp = NULL;
3963 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
3964 	if (error == 0)
3965 		error = nlookup(&nd);
3966 	if (error == 0) {
3967 		mp = nd.nl_nch.mount;
3968 		error = VFS_EXTATTRCTL(mp, uap->cmd,
3969 				uap->attrname, uap->arg,
3970 				nd.nl_cred);
3971 	}
3972 	nlookup_done(&nd);
3973 	return (error);
3974 }
3975 
3976 /*
3977  * Syscall to set a named extended attribute on a file or directory.
3978  * Accepts attribute name, and a uio structure pointing to the data to set.
3979  * The uio is consumed in the style of writev().  The real work happens
3980  * in VOP_SETEXTATTR().
3981  */
3982 int
3983 sys_extattr_set_file(struct extattr_set_file_args *uap)
3984 {
3985 	char attrname[EXTATTR_MAXNAMELEN];
3986 	struct iovec aiov[UIO_SMALLIOV];
3987 	struct iovec *needfree;
3988 	struct nlookupdata nd;
3989 	struct iovec *iov;
3990 	struct vnode *vp;
3991 	struct uio auio;
3992 	u_int iovlen;
3993 	u_int cnt;
3994 	int error;
3995 	int i;
3996 
3997 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
3998 	if (error)
3999 		return (error);
4000 
4001 	vp = NULL;
4002 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4003 	if (error == 0)
4004 		error = nlookup(&nd);
4005 	if (error == 0)
4006 		error = ncp_writechk(&nd.nl_nch);
4007 	if (error == 0)
4008 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4009 	if (error) {
4010 		nlookup_done(&nd);
4011 		return (error);
4012 	}
4013 
4014 	needfree = NULL;
4015 	iovlen = uap->iovcnt * sizeof(struct iovec);
4016 	if (uap->iovcnt > UIO_SMALLIOV) {
4017 		if (uap->iovcnt > UIO_MAXIOV) {
4018 			error = EINVAL;
4019 			goto done;
4020 		}
4021 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
4022 		needfree = iov;
4023 	} else {
4024 		iov = aiov;
4025 	}
4026 	auio.uio_iov = iov;
4027 	auio.uio_iovcnt = uap->iovcnt;
4028 	auio.uio_rw = UIO_WRITE;
4029 	auio.uio_segflg = UIO_USERSPACE;
4030 	auio.uio_td = nd.nl_td;
4031 	auio.uio_offset = 0;
4032 	if ((error = copyin(uap->iovp, iov, iovlen)))
4033 		goto done;
4034 	auio.uio_resid = 0;
4035 	for (i = 0; i < uap->iovcnt; i++) {
4036 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
4037 			error = EINVAL;
4038 			goto done;
4039 		}
4040 		auio.uio_resid += iov->iov_len;
4041 		iov++;
4042 	}
4043 	cnt = auio.uio_resid;
4044 	error = VOP_SETEXTATTR(vp, attrname, &auio, nd.nl_cred);
4045 	cnt -= auio.uio_resid;
4046 	uap->sysmsg_result = cnt;
4047 done:
4048 	vput(vp);
4049 	nlookup_done(&nd);
4050 	if (needfree)
4051 		FREE(needfree, M_IOV);
4052 	return (error);
4053 }
4054 
4055 /*
4056  * Syscall to get a named extended attribute on a file or directory.
4057  * Accepts attribute name, and a uio structure pointing to a buffer for the
4058  * data.  The uio is consumed in the style of readv().  The real work
4059  * happens in VOP_GETEXTATTR();
4060  */
4061 int
4062 sys_extattr_get_file(struct extattr_get_file_args *uap)
4063 {
4064 	char attrname[EXTATTR_MAXNAMELEN];
4065 	struct iovec aiov[UIO_SMALLIOV];
4066 	struct iovec *needfree;
4067 	struct nlookupdata nd;
4068 	struct iovec *iov;
4069 	struct vnode *vp;
4070 	struct uio auio;
4071 	u_int iovlen;
4072 	u_int cnt;
4073 	int error;
4074 	int i;
4075 
4076 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4077 	if (error)
4078 		return (error);
4079 
4080 	vp = NULL;
4081 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4082 	if (error == 0)
4083 		error = nlookup(&nd);
4084 	if (error == 0)
4085 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4086 	if (error) {
4087 		nlookup_done(&nd);
4088 		return (error);
4089 	}
4090 
4091 	iovlen = uap->iovcnt * sizeof (struct iovec);
4092 	needfree = NULL;
4093 	if (uap->iovcnt > UIO_SMALLIOV) {
4094 		if (uap->iovcnt > UIO_MAXIOV) {
4095 			error = EINVAL;
4096 			goto done;
4097 		}
4098 		MALLOC(iov, struct iovec *, iovlen, M_IOV, M_WAITOK);
4099 		needfree = iov;
4100 	} else {
4101 		iov = aiov;
4102 	}
4103 	auio.uio_iov = iov;
4104 	auio.uio_iovcnt = uap->iovcnt;
4105 	auio.uio_rw = UIO_READ;
4106 	auio.uio_segflg = UIO_USERSPACE;
4107 	auio.uio_td = nd.nl_td;
4108 	auio.uio_offset = 0;
4109 	if ((error = copyin(uap->iovp, iov, iovlen)))
4110 		goto done;
4111 	auio.uio_resid = 0;
4112 	for (i = 0; i < uap->iovcnt; i++) {
4113 		if (iov->iov_len > INT_MAX - auio.uio_resid) {
4114 			error = EINVAL;
4115 			goto done;
4116 		}
4117 		auio.uio_resid += iov->iov_len;
4118 		iov++;
4119 	}
4120 	cnt = auio.uio_resid;
4121 	error = VOP_GETEXTATTR(vp, attrname, &auio, nd.nl_cred);
4122 	cnt -= auio.uio_resid;
4123 	uap->sysmsg_result = cnt;
4124 done:
4125 	vput(vp);
4126 	nlookup_done(&nd);
4127 	if (needfree)
4128 		FREE(needfree, M_IOV);
4129 	return(error);
4130 }
4131 
4132 /*
4133  * Syscall to delete a named extended attribute from a file or directory.
4134  * Accepts attribute name.  The real work happens in VOP_SETEXTATTR().
4135  */
4136 int
4137 sys_extattr_delete_file(struct extattr_delete_file_args *uap)
4138 {
4139 	char attrname[EXTATTR_MAXNAMELEN];
4140 	struct nlookupdata nd;
4141 	struct vnode *vp;
4142 	int error;
4143 
4144 	error = copyin(uap->attrname, attrname, EXTATTR_MAXNAMELEN);
4145 	if (error)
4146 		return(error);
4147 
4148 	vp = NULL;
4149 	error = nlookup_init(&nd, uap->path, UIO_USERSPACE, NLC_FOLLOW);
4150 	if (error == 0)
4151 		error = nlookup(&nd);
4152 	if (error == 0)
4153 		error = ncp_writechk(&nd.nl_nch);
4154 	if (error == 0)
4155 		error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
4156 	if (error) {
4157 		nlookup_done(&nd);
4158 		return (error);
4159 	}
4160 
4161 	error = VOP_SETEXTATTR(vp, attrname, NULL, nd.nl_cred);
4162 	vput(vp);
4163 	nlookup_done(&nd);
4164 	return(error);
4165 }
4166 
4167 /*
4168  * Determine if the mount is visible to the process.
4169  */
4170 static int
4171 chroot_visible_mnt(struct mount *mp, struct proc *p)
4172 {
4173 	struct nchandle nch;
4174 
4175 	/*
4176 	 * Traverse from the mount point upwards.  If we hit the process
4177 	 * root then the mount point is visible to the process.
4178 	 */
4179 	nch = mp->mnt_ncmountpt;
4180 	while (nch.ncp) {
4181 		if (nch.mount == p->p_fd->fd_nrdir.mount &&
4182 		    nch.ncp == p->p_fd->fd_nrdir.ncp) {
4183 			return(1);
4184 		}
4185 		if (nch.ncp == nch.mount->mnt_ncmountpt.ncp) {
4186 			nch = nch.mount->mnt_ncmounton;
4187 		} else {
4188 			nch.ncp = nch.ncp->nc_parent;
4189 		}
4190 	}
4191 
4192 	/*
4193 	 * If the mount point is not visible to the process, but the
4194 	 * process root is in a subdirectory of the mount, return
4195 	 * TRUE anyway.
4196 	 */
4197 	if (p->p_fd->fd_nrdir.mount == mp)
4198 		return(1);
4199 
4200 	return(0);
4201 }
4202 
4203